Content uploaded by Scott Hawken
Author content
All content in this area was uploaded by Scott Hawken on Feb 22, 2023
Content may be subject to copyright.
Lecture Notes
in Geoinformation and Cartography
Stan Geertman
Qingming Zhan
Andrew Allan
Christopher Pettit Editors
Computational
Urban
Planning and
Management
for Smart Cities
Lecture Notes in Geoinformation
and Cartography
Series Editors
William Cartwright, School of Science, RMIT University,
Melbourne, VIC, Australia
Georg Gartner, Department of Geodesy and Geoinformation,
Vienna University of Technology, Wien, Austria
Liqiu Meng, Department of Civil, Geo and Environmental Engineering,
Technische UniversitätMünchen, München, Germany
Michael P. Peterson, Department of Geography and Geology,
University of Nebraska at Omaha, Omaha, NE, USA
The Lecture Notes in Geoinformation and Cartography series provides a contempo-
rary view of current research and development in Geoinformation and Cartography,
including GIS and Geographic Information Science. Publications with associated
electronic media examine areas of development and current technology. Editors
from multiple continents, in association with national and international organiza-
tions and societies bring together the most comprehensive forum for Geoinformation
and Cartography.
The scope of Lecture Notes in Geoinformation and Cartography spans the range
of interdisciplinary topics in a variety of research and application fields. The type of
material published traditionally includes:
•proceedings that are peer-reviewed and published in association with a
conference;
•post-proceedings consisting of thoroughly revised final papers; and
•research monographs that may be based on individual research projects.
The Lecture Notes in Geoinformation and Cartography series also includes various
other publications, including:
•tutorials or collections of lectures for advanced courses;
•contemporary surveys that offer an objective summary of a current topic of
interest; and
•emerging areas of research directed at a broad community of practitioners.
More information about this series at http://www.springer.com/series/7418
Stan Geertman •Qingming Zhan •
Andrew Allan •Christopher Pettit
Editors
Computational Urban
Planning and Management
for Smart Cities
123
Editors
Stan Geertman
Human Geography and Planning
Utrecht University
Utrecht, The Netherlands
Qingming Zhan
School of Urban Design
Wuhan University
Wuhan, China
Andrew Allan
School of Art, Architecture and Design
University of South Australia
Adelaide, SA, Australia
Christopher Pettit
Faculty of the Built Environment
University of New South Wales
Sydney, NSW, Australia
ISSN 1863-2246 ISSN 1863-2351 (electronic)
Lecture Notes in Geoinformation and Cartography
ISBN 978-3-030-19423-9 ISBN 978-3-030-19424-6 (eBook)
https://doi.org/10.1007/978-3-030-19424-6
©Springer Nature Switzerland AG 2019
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
The international CUPUM conference (Computers in Urban Planning and Urban
Management) has been one of the premier international conferences for the
exchange of ideas and applications of computer technologies to address a range of
social and environmental problems relating to urban areas. The first conference took
place in 1989 in Hong Kong. Since then, this biannual conference has been hosted
in cities across Asia, Australia, Europe, North America and South America
(Table 1). In 2009, Hong Kong once again hosted a CUPUM conference. And now,
in 2019, 10 years after that date and 30 years after the first CUPUM conference,
China once more is the host for a CUPUM conference, now at Wuhan University in
its 16th iteration.
Table 1 Past CUPUM conferences
Number Year Place Country
I 1989 Hong Kong Hong Kong
II 1991 Oxford UK
III 1993 Atlanta USA
IV 1995 Melbourne Australia
V 1997 Mumbai India
VI 1999 Venice Italy
VII 2001 Honolulu USA
VIII 2003 Sendai Japan
IX 2005 London UK
X 2007 Iguazu Falls Brazil
XI 2009 Hong Kong China
XII 2011 Lake Louise (Calgary/Banff) Canada
XIII 2013 Utrecht The Netherlands
XIV 2015 Boston USA
XV 2017 Adelaide Australia
XVI 2019 Wuhan China
v
The CUPUM Board (Tables 2and 3) has promoted the publication of a Springer
CUPUM Book 2019 with a selection of scientific papers that were submitted to the
conference. Those papers went through a competitive review process that resulted
in the selection of what the reviewers deemed to be the best CUPUM papers of
2019. All these papers fit the main overarching central theme of the Wuhan 2019
CUPUM conference: Computational Urban Planning and Management for Smart
Cities. Therein, we acknowledge that the emergent phenomenon of smart cities is in
need of innovative technologies, associated methodologies and their adoption by
the key actors responsible for their planning and management. By gathering at the
conference premises from 8 to 12 July 2019 in Wuhan China and via the publi-
cation of this Springer CUPUM Book 2019, we hope to exchange new innovative
ideas on this theme and bring together science and practice much closer than ever
before.
Organizing the programme of an international conference and editing a volume
of scientific papers requires dedication, time, effort and support. First of all, we
would like to thank all the people closely involved in the organization of the Wuhan
2019 CUPUM conference. Organizing such a conference always turns out to be
much more work and generating much more problems/challenges than envisaged
before.
Table 2 Board of Directors of CUPUM
Name Institute Country
Stan Geertman (Chair of Board) Utrecht University NED
Andrew Allan University of South Australia AUS
Joseph Ferreira Massachusetts Institute of Technology USA
Robert Goodspeed University of Michigan USA
Weifeng Li University of Hong Kong CHN
Christopher Pettit University of New South Wales AUS
Zhan Qingming Wuhan University CHN
Antonio N. Rodrigues da Silva University of Sao Paulo BRA
Renee Sieber McGill University CAN
Atsushi Suzuki Meijo University JPN
Table 3 Advisors to the CUPUM Board
Name Institute Country
Michael Batty (Chair) University College London GBR
Karl Kim University of Hawaii USA
Dick Klosterman University of Akron USA
Kazuaki Miyamoto Tokyo City University JPN
Paola Rizzi Universitàdegli Studi di Sassari ITA
John Stillwell University of Leeds GBR
Anthony G. O. Yeh University of Hong Kong CHN
Ray Wyatt University of Melbourne AUS
vi Preface
Second, as book editors, we would like to thank the authors for their high-quality
contributions. We started with 35 proposals for interesting book chapters and finally
ended up with 26 high-quality full chapters in this book. The double-blind review
process was not an easy task and it is always difficult when potential authors
experience the disappointment of not being selected. By fulfilling the double-blind
review process and demanding at least two reviews per submission, we believe that
the review process has been conducted in a fair and equal way.
Third, we would like to thank our scientific sponsors (Utrecht University,
Wuhan University, University of South Australia, University of New South Wales)
for their contribution in time and resources to this publication. In addition, we
would like to thank Springer Publishers for their willingness to publish these
contributions in their academic series Springer Lecture Notes in Geoinformation
and Cartography. This is already the fourth time that a selection of best papers from
the CUPUM conference has been published by Springer. The first time was in 2013
when we published the book: Planning Support Systems for Sustainable Urban
Development (Stan Geertman, Fred Toppen, John Stillwell (eds.)). The second time
was in 2015 when we published the book: Planning Support Systems and Smart
Cities (Stan Geertman, Joe Ferreira, Robert Goodspeed, John Stillwell (eds.)). And
in 2017, we published the book: Planning Support Science for Smarter Urban
Futures (Stan Geertman, Andrew Allan, Christopher Pettit, John Stillwell (eds.)).
We hope more CUPUM books will follow.
Wuhan, China Stan Geertman
2019 Qingming Zhan
Andrew Allan
Christopher Pettit
Preface vii
Contents
1 Computational Urban Planning and Management for Smart
Cities: An Introduction .................................. 1
Stan Geertman, Andrew Allan, Qingming Zhan and Chris Pettit
Part I Smart City
2 Sejong Smart City: On the Road to Be a City of the Future ..... 17
Yountaik Leem, Hoon Han and Sang Ho Lee
3 Data Protection Law and City Planning: Role of Open Data in
Climate Resilience and Governance of National Capital Territory
of Delhi, India ......................................... 35
Mahak Agrawal
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence
in Chicago and Budapest ................................ 53
Ali Soltani, TamásMátrai, Rosalia Camporeale and Andrew Allan
5 Can Social Media Play a Role in Urban Planning? A Literature
Review .............................................. 69
Yanliu Lin and Stan Geertman
6 Bridging the Information and Physical Space: Measuring Flow
from Geo-Located Social Media Data on the Street Network ..... 85
Alireza Karduni and Eric Sauda
7 Comparing Smart Governance Projects in China: A Contextual
Approach ............................................ 99
Huaxiong Jiang, Stan Geertman and Patrick Witte
ix
Part II Computational Planning
8 A Preliminary Study on Micro-Scale Planning Support
System .............................................. 117
Daosheng Sun, Xiaochun Huang, Lianna He, Tengyun Hu
and Yilong Rong
9 Geodesign—A Tale of Three Cities ......................... 139
Christopher Pettit, Scott Hawken, Carmela Ticzon
and Hitomi Nakanishi
10 Toward a Better Understanding of Urban Sprawl: Linking
Spatial Metrics and Landscape Networks Dynamics ............ 163
Tengyun Hu, Xiaochun Huang, Xuecao Li, Lu Liang and Fei Xue
11 Correlating Household Travel Carbon Emissions, Travel
Behavior and Land Use: Case Study of Wuhan, China ......... 179
Jingnan Huang, Ming Zhang and Ningrui Du
12 A Simulation Platform for Transportation, Land Use and Mobile
Source Emissions ...................................... 205
Liyuan Zhao and Zhong-Ren Peng
13 Hosting a Mega Event, a Drive Towards Sustainable
Development: Dubai’s Expo 2020 .......................... 223
Bashar Taha and Andrew Allan
14 Deep Learning Architect: Classification for Architectural Design
Through the Eye of Artificial Intelligence .................... 249
Yuji Yoshimura, Bill Cai, Zhoutong Wang and Carlo Ratti
15 An Immersive 3D Virtual Environment to Support Collaborative
Learning and Teaching .................................. 267
Aida Afrooz, Lan Ding and Christopher Pettit
16 Spatiotemporal Information System Using Mixed Reality for
Area-Based Learning and Sightseeing ....................... 283
Ryuhei Makino and Kayoko Yamamoto
Part III Mobility
17 Origin-Destination Estimation of Bus Users
by Smart Card Data .................................... 305
Mona Mosallanejad, Sekhar Somenahalli and David Mills
18 The Comparison Between Two Different Algorithms
of Spatio-Temporal Forecasting for Traffic Flow Prediction ...... 321
Haochen Shi, Yufeng Yue and Yunqi Zhou
x Contents
19 Developing a Behavioural Model for Modal Shift
in Commuting ......................................... 347
Ali Soltani, Andrew Allan and Ha Anh Nguyen
20 Planning for Safer Road Facilities for Bicycle Users
at Junctions .......................................... 373
Li Meng, Li Luo, Yanchi Chen and Branko Stazic
21 Method to Evaluate the Location of Aged Care Facilities in
Urban Areas Using Median Share Ratio ..................... 389
Koya Tsukahara and Kayoko Yamamoto
22 Identifying Changes in Critical Locations for Transportation
Networks Using Centrality ............................... 405
Nazli Yonca Aydin, Ylenia Casali, H. Sebnem Duzgun
and Hans R. Heinimann
23 Efficient Regional Travel for Rescue and Relief Activities in a
Disaster .............................................. 425
Toshihiro Osaragi, Masashi Kimura and Takuya Oki
24 A Two-Stage Process for Emergency Evacuation Planning:
Shelter Assignment and Routing ........................... 443
Ali Soltani, Andrew Allan and Mohammad Heydari
25 A Comprehensive Regional Accessibility Model Based on Actual
Routes-of-Travel: A Proposal with Multiple Online Data ........ 463
Yuli Fan, Qingming Zhan, Huizi Zhang and Jiaqi Wu
26 Taxi Behavior Simulation and Improvement with Agent-Based
Modeling ............................................. 483
Saurav Ranjit, Apichon Witayangkurn, Masahiko Nagai
and Ryosuke Shibasaki
Index ...................................................... 505
Contents xi
Chapter 1
Computational Urban Planning
and Management for Smart Cities:
An Introduction
Stan Geertman, Andrew Allan, Qingming Zhan and Chris Pettit
Abstract The world is on the cusp of advancing to a sixth wave of technologically
based innovation, which will significantly impact our rapidly-growing urban envi-
ronments. However, political and social resistance hold us back from the transforma-
tional potential of technological innovation, to empower both urban decision-makers
and citizens to make informed choices about their urban futures. This volume brings
together a collection of chapters that encapsulates the state-of the-art research in data
driven methods for harnessing the potential of computational planning and manage-
ment for smart cities. The scholarly works are organised around three thematic lenses:
(I) smart cities and their data and governance; (II) computational planning of smart
cities; and (III) mobility and transportation modelling. Each chapter addresses the
potential of technologically based innovation in urban environments and provides
innovative methods and a series of research findings to guide cities in harnessing
its potential. This volume provides an important contribution to the body of knowl-
edge interfacing data, technology and smart cities as we strive to plan more liveable,
sustainable and productive cities of the future.
Keywords Computational planning ·Smart cities ·Big data ·Mobility ·
Transport modelling
S. Geertman (B)
Department of Human Geography and Planning, Faculty of Geosciences,
Utrecht University, Princetonlaan 8a, 3584 CB Utrecht, The Netherlands
e-mail: s.c.m.geertman@uu.nl
A. Allan
School of Art, Architecture and Design, University of South Australia,
Adelaide, Australia
e-mail: andrew.allan@unisa.edu.au
Q. Zhan
Wuhan University, Wuhan, Hubei, China
e-mail: qmzhan@whu.edu.cn
C. Pettit
City Futures Research Centre, UNSW, Sydney, Australia
e-mail: c.pettit@unsw.edu.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_1
1
2 S. Geertman et al.
1 Introduction
As this book chapter was prepared, in the southern hemisphere, Adelaide experienced
its hottest day on record (47 °C), with January 2019 becoming Australia’s warmest
month since records began. In the northern hemisphere at the same time, a polar vor-
tex resulted in record breaking low temperatures to the Mid-West region of the United
States (US), with temperatures plunging to −30 °C or lower. Whilst the cold snap is
not intuitively what one would normally expect in a warming world, in the northern
hemisphere, a warming Arctic thrust cold Arctic air and the Jetstream further south.
These unstable weather patterns are symptomatic of an inexorable trend towards
changing climates characterised by a warming world, and rising sea levels. With
more than 55% of the world’s population now urbanised (UN-DESA 2019), and cities
assuming greater economic and social significance in people’s lives globally, cities
will face the challenges of accommodating much larger populations, and balancing
social equity considerations with multiple environmental threats to future well-being
posed by climate change such as rising sea levels, heat stress, declining air quality
and natural resource shortages. Not only are there many more cities in this urbanising
world—551 cities worldwide with a million or more inhabitants in 2016 (31 of which
were mega-cities of 10 million or more)—but cities are projected to increase in size
and number by 2030 and accommodate larger shares of national populations (60%
globally by 2030). For example, the number of mega cities with a population in excess
of 25 million inhabitants is projected to increase from three in 2016 (Tokyo—38M,
Delhi—26M, Shanghai—24M) to eight in 2030 (Tokyo—37M, Delhi—36M,
Shanghai—31M, Mumbai—28M, Beijing—28M, Dhaka—28M, Karachi—25M,
Cairo—25mM) (United Nations 2016). Massive urban planning and logistical chal-
lenges loom, particularly with regard to accommodating the day to day needs and
lifestyle aspirations of these cities’ residents. Cities will need to further embrace data
and technology to become smarter in responding to these challenges to ensure cities
become more liveable, sustainable, productive and resilient (Geertman et al. 2017).
The looming challenges facing cities include the threats posed by anthropogenic
induced climate change, burgeoning urban populations (particularly within the
world’s largest cities), resource depletion, digital innovations leading to societal
change that are redefining human activities, interactions and knowledge acquisi-
tion, and a globally interconnected world. In 2018, it seemed that the world was
on the cusp of introducing autonomous vehicles (AVs), crypto-currencies, sharing
economies and 5G Wi-Fi technology, which would have heralded the “Internet of
Things” (IoT). However, the rate of progress has somewhat slowed because in the
case of AVs, several high profile fatal accidents have highlighted the current tech-
nological fallibility in autonomous vehicles being introduced into mixed road user
environments; crypto-currencies have become shrouded in uncertainty as different
systems jockey for supremacy; shared use business platforms have experienced sig-
nificant commercial challenges in many parts of the work, for example share-bike
business collapses in Australia such as OfO and OBike; and the introduction of 5G
has run into both political and telecommunications business resistance to its uptake.
1 Computational Urban Planning and Management for Smart Cities … 3
Politically, in 2019, the resolve to address the human contribution to climate change
is increasingly uncertain, as manifested by US President Donald Trump’s edict to
withdraw the United States from the 2015 Paris Climate Change Agreement, negoti-
ation difficulties with BREXIT in the United Kingdom, the 2018 fuel riots in France
and many other political and governance challenges.
Technology advocates and futurists posit a concept that the world is entering into
a sixth wave of technologically based innovation. This new wave that emerged in the
mid 2000s includes automation, robotics, digitalisation and sustainability (Rodrigue
2017). Previous waves of technology-based innovation were the Industrial Revolu-
tion (first wave from 1785 to 1840s), the Age of Steam (second wave from 1820s
to 1900s), the Age of Electricity and Internal Combustion Engines (third wave from
1890s to 1950s), the Age of Mass Production, Aviation, Electronics and Petrochem-
icals (fourth wave from 1900s to 1990s) and the Age of Information, Digital Tech-
nologies, the Internet and Biotechnology (fifth wave from 1940s to 2020s) (Silva
and Di Serio 2016). Hence, given the rate of technological progress and the extent
to which societies across both the developed and developing world have embraced
digital technologies and the internet, it seems to be only a matter of time before
the world advances to the promise of the sixth wave of technologically based inno-
vation. Despite impressive strides towards sustainability, particularly with regard to
technologies that facilitate the use of renewable energy, e-mobility, energy, water and
resource efficiency, if one considers sustainability to be an intrinsic part of this new
technology wave, political and social resistance is still holding much of the world
back from progressing successfully towards the crest of the sixth wave. A slowing
world economy has also placed restrictions on the uptake speed of many new inno-
vations, that whilst technically feasible, are too expensive for universal uptake just
yet.
With over 55% the world population living in urban environments, cities today
play a pivotal role in a national economy’s performance, impact on the environment
and the quality of life of their citizens. With the rise of the smart cities movement there
it the opportunity to see cities as living laboratories where new data and technology
can be explored in relation to better planning, design and management of urban
environments. Smart city technology including digital planning tools (Pettit et al.
2018) offer the potential to take a data driven approach to infrastructure services,
mobility flows and networks, building technologies, environmental impacts, human
activities and social needs. What is different about previous waves of innovation in
terms of creating the potential for transformational societal impact, is that nearly
every citizen has a cellular or mobile phone, with the technological capability to
run unlimited personalised applications (or ‘apps’), that facilitates geo-positioned
data flows and digital profiles of each person, to give unprecedented insights into
society’s wants and needs. When personal devices are combined with the rise of
cloud computing, high capacity Wi-Fi and broadband computing infrastructure, every
object and individual on Earth can be connected in an intricate digital network, which
Intel has called the international of everything. A mobile phone today can fulfil
multiple functions far beyond conventional voice communications to facilitating
transactions of any kind, functioning as remote control for automation (such as
4 S. Geertman et al.
in the home), to allowing app developers to harvest digital data flows about their
users. Accurately capitalising on and making sense of these data flows creates many
research opportunities, further expanded with the presence of multiple user profiles
tailored to particular apps and new challenges that arise with the advent of Big Data
(Li et al. 2016).
Social networking apps such as Facebook and Twitter have democratised mass
communication, but the downside of this is that the cornucopia of data collected by
such apps may be passed on to third party users and be used for purposes distant to
what the users of the app thought their data could be used for. As data flows become
increasingly commodified and commercialised (with many services now behind digi-
tal paywalls using cloud computing remote from the user), paradoxically, the initially
very democratic and seemingly open nature of the internet is being challenged. How-
ever, the open data movement is a balance force where many governments around
the world and some companies, such as Uber, are making their data available on
the web and this is in turn powering advanced city analytic toolkits championed
by organisations—such as the Australian Urban Research Infrastructure (AURIN)
(Pettit et al. 2017a,b) and the Urban Big Data Centre in Glasgow (Thakuriah et al.
2016).
Many cities around the world are now embracing the notion of becoming ‘Smart
Cities’, which although a ‘catch-all’ phrase that includes physical planning responses
such as ‘New Urbanism’, is now accepted by civic, political and city administration
leaders to include the features and facilities possible with a digital economy, includ-
ing autonomous transport operation, shared mobility, digital monetary transactions
and activities (including many public services) conducted in cyber-space. This de-
materialising of city activities that has been made possible by digitisation and the
internet has the potential to be as significant to societal change as the rapid decarbon-
isation of our economy during the shift towards renewable energy technologies. The
evidence of change towards ‘smarter living’, is more clearly obvious in cities, and
indeed, in the developed economies of the world, cities are much further advanced
than the political leadership of many countries in embracing the potential offered by
digital technologies as a pathway to greater urban efficiencies, a decarbonised econ-
omy, increased economic growth and enhanced social development. Conservative
national political leaders in North America, Europe and Australia often lag behind
public and political sentiments in cities in the uptake of carbon reduction policies
and in exploring new ways of doing things (such as with shared economy services).
However, national, state, metropolitan and local governments are embracing dig-
ital technologies as rich sources of data to better understand social trends, social
behaviours, the environment and to provide greater efficiency in services. In the 21st
century, data can be seen as the new oil which powers the information age (Castells
2010) and ultimately the smart city.
The definition of smart cities is open to debate when it comes to identifying, rating
and ranking the world’s cities. This is partly a reflection of the diversity of digital
technologies and computer systems available, but it also reflects the increasingly
closed nature of the internet with its limited transparency. Much of the academic
literature on ‘smart cities’ tends to provide a holistic approach that includes social,
1 Computational Urban Planning and Management for Smart Cities … 5
economic, environmental, legal, technological and sustainability aspects (Sujataa
et al. 2016) whereas other authors focus on specific innovations or digital systems
used by particular cities (typified by the use of ‘dashboard systems’ such as IBM’s
open source Intelligent Operations Center, utilised to identify Key Performance Indi-
cators (KPIs)) (Zhuhadar et al. 2017). One useful, albeit business- and world-oriented
schema for assessing the relative ‘smartness’ of the world’s cities is developed by
the Business School of the University of Navarra, and in its fifth iteration is the IESE
Cities in Motion Index 2018 (Berrone et al. 2018). It comprises 83 indicators for
165 cities in 80 countries covering diverse measures, ranging from the number of
Apple stores geographically located in a city to the forecast temperatures resulting
from climate change effects. This system produces a single index based on metrics
that reflect KPIs considered essential for city prosperity in the longer term across
four broad themes: (1) sustainability; (2) social cohesion; (3) connectivity; and (4)
innovation. In the digital technology sphere, the key indicators of interest are: (num-
ber) 38 (Open Data Platform); 39 (E-Government Development Index, which relates
to public sector use of information technology in providing access to its citizens);
73 (Tweet map of Twitter users); 74 (registered users of LinkedIn); 75 (registered
users of Facebook); 76 (number of mobile phones); 77 (mapped Wi-Fi hotspots)
78 (number of Apple stores); 79 (rating of Innovation Cities Programs); 80 (rate
of landline subscriptions); 81 (rate of broadband subscriptions), 82 (rate of Internet
access by households); and 83 (rate of mobile phone use by households). The limi-
tations with an index such as this, are that it is not technology focused, is somewhat
subjective (particularly on social cohesion issues), appears to have a Western world
bias and there are surprising omissions and inclusions when it comes to city selec-
tion. Notwithstanding its limitations, it is an interesting proxy for identifying high
performing ‘smart cities’. The top 15 ‘smart cities’ in the world in 2018 applying
this index are: (1) New York; (2) London; (3) Paris; (4) Tokyo; (5) Reykjavik; (6)
Singapore; (7) Seoul; (8) Toronto; (9) Hong Kong; (10) Amsterdam; (11) Berlin;
(12) Melbourne; (13) Copenhagen; (14) Chicago; and (15) Sydney.
Cities and public agencies from the local to national level are increasingly pro-
viding open access to their data, allowing for example, useful metrics and mapping
of traffic congestion and surface temperature mapping. Notably, a number of trans-
port agencies are at the forefront in releasing vast amounts of open data such as
Transport for London and Transport NSW. Both of these transport agencies have
launched open data portals which are fuelling smart city applications including real-
time city dashboards such as the London Dashboard (Gray et al. 2016) and Sydney
Dashboard (Pettit et al. 2017a,b). With respect to surface temperature mapping, the
South Australian Government in 2019 launched an Urban Heat Mapping Viewer for
Adelaide, Australia, that allows heat island effects to be identified, admittedly from
a single point in time (derived from aerial heat maps taken in March 2018) which
can be combined with mapping of social well-being indices to identify socially vul-
nerable households in heat-wave scenarios (spatialwebapps.environment.sa.gov.au).
The key limitation of this type of static mapping digital tool is that it lacks real-time
interrogation.
6 S. Geertman et al.
Interactive digital tools such as the city dashboards referred to above do allow a
city’s residents to make more informed choices about the development of their cities,
their individual activities (particularly in relation to travel), emergency responses
(such as action plans and real-time evacuation way-finding in the event of natural
disasters), mobility choices and other useful metrics that relate to their day to day
lives. Planners, urban operational managers and city administrators can use city
dashboards to provide much more informed decision making and planning, reducing
the risk of unintended consequences. Most significantly, scenario planning can now
become a two-way dialogue that responds to citizens’ interests and concerns whilst
advancing a holistic metropolitan scale vision that exploits the power of Big Data and
that is sophisticated enough to yield quantifiable solutions for KPIs when exploring
‘what-if’ planning and city operational scenarios.
‘Smart cities’ should have a circular metabolism in their use of resources and
operate within resource limits determined by the bioregion within which they are
located (Newman et al. 2017). Smart city technologies can provide the capabilities to
monitor and control use of resources, activities and minimise waste dynamic flows
in real-time within a city’s Bio-region. Providing knowledge of these data flows
to urban planners, urban managers, political decision-makers and citizens—for
everything from water and energy consumption to carbon emissions and solid
wastes—empowers to city communities, in built environments that can seem to be
overpowering to the individual person. Smart cities can potentially empower all con-
cerned stakeholders to make informed choices about their urban futures in a timely
manner. As more people choose to reside in cities, it becomes imperative that cities
become smarter to meet people’s needs and lifestyle aspirations without diminishing
the environment and ensuring a high quality of life for all residents and visitors.
It is in this context that this volume contributes to the theory, practice and adoption
of computational planning and management for smart cities. The remainder of this
chapter examines three distinct themes relating to particular aspects, and summarises
the effort of our colleagues, in tackling the challenges and opportunities in harnessing
the potential of computational planning and management for smart cities.
2 Introduction to This Volume
2.1 Part 1: Smart City
In the first part of the book we bring together a collection of chapters from colleagues
which focus on the Smart City; the use of data including Big Data, open data, spatial
data and social media data in the context of a Smart City; and the governance of a
Smart City.
In Chap. 2‘Sejong Smart City: On the Road to be a City of the Future’, the
authors Yountaik Leem, Hoon Han and Sang Ho Lee present Korea’s ICT-driven
smart city concept, called Ubiquitous Cities (U-City). One of the strategic adopters
1 Computational Urban Planning and Management for Smart Cities … 7
of this concept is Sejong City, which is being developed in central South Korea as
a new administrative centre. Sejong Smart City is one of the full-scale greenfield
development models of a future city armed with cutting-edge ICT. In their chapter,
the authors present Korean smart cities in general and Sejong Smart City in particular
from the viewpoint of industry-mix, infrastructure, technology and services, followed
by a discussion on the future of smart cities in Korea.
In Chap. 3‘Data Protection Law and City Planning: Role of Open Data in Climate
Resilience and Governance of National Capital Territory of Delhi, India’ the author
Mahak Agrawal emphasises the importance of open data for urban planners and
administrators. In that, he starts with concern over data availability and coverage,
an important factor that guides decision making and public policies, due to costs
in expenses and time associated with data collection. According to the author these
costs can be drastically reduced if the data is available online through government
regulated portals, either free of cost or at affordable rates. For an example, the chapter
highlights a case in Delhi in which the role of data in climate resilient development
is worked out with the help of open data, available on national and international
geoportals.
In Chap. 4‘Exploring Shared-Bike Travel Patterns Using Big Data: Evidence in
Chicago and Budapest’ the authors Ali Soltani, Tamás Mátrai and Rosalia Campo-
reale examine the travel patterns of bike-share users in two metropolitan areas. For
each location they possess approximately two million transaction data associated
with bike trips made over a three-month period. These include several aspects of
user travel behaviour, such as day and time of travel, frequency of usage, duration
of usage, seasonal and peak/off-peak variations and major origin/destinations. The
results show that in both cities the bike-sharing option is a male-dominated alterna-
tive, particularly welcomed by younger generations, who make the largest share of
trips in the afternoon. From this study the authors conclude that a proper usage of
open-source Big Data can be of big help to gain more insight into the usage of these
kinds of vehicle-sharing systems.
In Chap. 5‘Can Social Media Play a Role in Urban Planning? A Literature
Review’ the authors Yanliu Lin and Stan Geertman conduct a systematic review of
the extent to which social media can be usefully applied in urban planning. In their
chapter they arrive at two main findings. On the one hand, they identify that social
media data are increasingly used for urban analysis and modelling. The domains of
application include research after individual activity patterns, urban land use, trans-
portation behavior, and landscape research. On the other hand, they identify that
social media provide new platforms for participation, communication and collabo-
ration. This offers new opportunities for cities to hear the voices of distinctive social
groups. The authors end their chapter by discussing some pressing issues of using
social media data in urban planning, including population and spatial biases and
difficulties in extracting useful information out of the social media data.
In Chap. 6‘Bridging the Information and Physical Space: Measuring Flow from
Geo-Located Social Media Data on the Street Network’ the authors Alireza Karduni
and Eric Sauda investigate the relationship between urban space and human behavior
with the help of social media data. In that, he develops a new method to extrapolate
8 S. Geertman et al.
flows of geolocated social media data on a street network. By applying this method
to a corpus of geolocated tweets collected from the Los Angeles metropolitan area
the author is able to compare the results to betweenness centrality of the streets
as a measurement of connectivity and density of businesses and as a measurement
of public activity. It is found that the flows calculated from Twitter have a high
correlation with public activities hinting towards the relationship between geolocated
social media usage and businesses and public spaces.
In Chap. 7‘Comparing Smart Governance Projects in China: A Contextual
Approach’ the authors Huaxiong Jiang, Stan Geertman and Patrick Witte investi-
gate the impact of so-called urban contextual factors on the governance of smart
cities. For that, they first conceptually elaborate on the notion of smart governance.
Thereafter, they analyze a range of distinctive smart governance projects in different
Chinese cities to identify the impact of urban contextual factors on smart governance
practices. Their comparative exploration of four Chinese projects representing four
types of smart governance show that the urban contextual factors clearly affect the
interaction of technology and urban actors. From this they conclude that more spe-
cific research and knowledge on these urban contextual factors is of vital importance
to better predict the expected outcomes of intended smart governance policies.
2.2 Part 2: Computational Planning
In the second part of this book we bring together a collection of chapters from
colleagues who are focused on the computational planning of (smart) cities and its
associated methodology.
In their Chap. 8‘A Preliminary Study on Micro-Scale Planning Support System’
the authors Sun Daosheng, Huang Xiaochun, He Lianna, Hu Tengyun, and Rong
Yi-Long stress the need for a new Planning Support System given the transformation
in China from the traditional macro-scale planning towards a micro-scale planning.
According to the authors, given the increased human focus in micro-scale planning,
this system should be able to consider human’s subjective feelings and needs. In
that, they propose it should be organized into three categories, namely the sub-
system of natural environment and micro-ecology, the sub-system of urban design
and spatial layout, and the sub-system of human behavior and community life. The
authors consider these sub-systems to play an essential role in the future of micro-
scale planning in its focus on central-city-planning, urban-physical-examination and
livable-city-construction.
The Chap. 9‘Geodesign—A Tale of Three Cities’ by Christopher Pettit, Scott
Hawken, Carmela Ticzon and Hitomi Nakanishi the authors discuss the pros and
cons of applying Steinitz’s (2012) Geodesign Framework in the context of three
Australian cities including (i) South East Sydney, (ii) the emerging Western City of
Sydney and (iii) the City of Canberra. In all three case studies the Geodesign Hub
platform is applied to develop a series of future city scenarios, driven through the
common themes of population growth and integrated land use transport planning.
1 Computational Urban Planning and Management for Smart Cities … 9
The research conducted is reflective and undertaken in the context of studio work
by three different Geodesign classes run across two universities. The chapter reflects
on the strengths and opportunities of the Geodesign Framework in supporting the
planning and design of future cities in the context of (i) data and technology, (ii)
process, and (iii) outputs.
In Chap. 10 ‘Toward a Better Understanding of Urban Sprawl: Linking Spatial
Metrics and Landscape Networks Dynamics’ the authors Tengyun Hu, Xiaochun
Huang, Xuecao Li, Lu Liang and Fei Xue explore the urban sprawl process in Beijing
over the past three decades (i.e. 1984–2013). This is done on an annual basis by
linking spatial metrics and landscape networks to trace the dynamics of urban patches.
The authors were able to identify six main growth periods of urban expansion with
distinctive patterns and to explain the spatiotemporal dynamics of urban patches,
with a linkage to policies behind each hotspot of urban expansion. Based on that, it
was identified that the major trajectories of urban growth in Beijing started from the
northern and southern parts of the main built-up region to its southeast side, which
has developed a bit differently than the planned two axes (i.e. horizontal and vertical)
along the core area of the city.
In Chap. 11 ‘Correlating Household Travel Carbon Emissions, Travel Behavior,
and Land Use: Case Study of Wuhan, China’ the authors Jingnan Huang, Ming
Zhang, and Ningrui Du aim at unraveling the factors that contribute to change in
family carbon emissions. The authors perform an empirical study with a sample
of 1194 families from Wuhan, China. Alongside socioeconomic characteristics, the
study pays particular attention to the role of the spatial context in family living
and travelling. A regression analysis shows that urban spatial structure and land use
context offer additional explanatory power to variations in travel carbon emission
after controlling for socio-economic factors. It turns out that emission hot spots and
high-emission families most likely concentrate in newly developed suburban areas.
From this finding a range of both place-based and people-based planning and policy
measurements are proposed to reduce carbon emissions.
In Chap. 12 ‘A Simulation Platform for Transportation, Land Use and Mobile
Source Emissions’ the authors Liyuan Zhao and Zhong-Ren Peng describe an inte-
grated model platform for assessing the interaction among land-use, transportation,
and mobile source emissions. The authors aim at identifying the added value of
an integrated model platform above one in which distinctive standalone models are
applied. In the integrated framework a land use model produces land use change
over the space and time dimensions, allocates land use forecast results in terms of
household and employment at the traffic analysis zone (TAZ) level, and feeds these
socioeconomic data into a travel demand model. Then, the travel time and accessibil-
ity index produced by the demand model are fed back into the land-use model which
then quantifies the emissions. The results show significant differences in emission
outcomes of the integrated platform and standalone models.
In Chap. 13 ‘Hosting a Mega Event, a Drive Towards Sustainable Development:
Dubai’s Expo 2020’ the authors Bashar Taha and Andrew Allan explore the strategic
elements of planning for a mega event. The hosting of a mega event creates enormous
demand for new buildings and facilities and requires the development of new urban
10 S. Geertman et al.
areas and infrastructures. The downside of hosting a mega event is its relatively short
period and the financing pressures on the host country. In cases where integration
with strategic planning is absent, the potential risk is losing the benefits of the massive
investments and being left with redundant infrastructures and facilities. This chapter
explores the critical elements adopted by the Dubai Government to deliver effective
sustainable planning and metro route planning in the development of Dubai’s Expo
2020 to create a ‘smart city’ legacy.
In Chap. 14 ‘Deep Learning Architect: Classification for Architectural Design
Through the Eye of Artificial Intelligence’ the authors Yuji Yoshimura, Bill Cai,
Zhoutong Wang and Carlo Ratti make use of a neural network model to measure
similarities between architectural designs. In that, they apply state-of-the-art tech-
niques in deep learning and computer visioning to measure the visual similarities
between architectural designs of different architects. Using a dataset consisting of
web scraped images and an original collection of images of architectural works, the
authors were able to train their deep convolutional neural network (DCNN) model
to achieve 73% accuracy in classifying works belonging to 34 different architects.
Finally, through examining the weights in the trained DCNN model, the authors
were able to quantitatively measure the visual similarities between architects that
were implicitly learned by the neural network model and accomplish a high level of
similarity.
In Chap. 15 ‘An Immersive 3D Virtual Environment to Support Collaborative
Learning and Teaching’ the authors Aida Afrooz, Lan Ding, and Christopher Pettit
reflect on a Virtual Learning Environment (VLE) in the context of architecture, urban
planning and design. The research aims to critically assess the ability of virtual
environments to support experiential online learning. It concentrates on a 3D virtual
platform to support collaboration among students in Built Environment courses.
Feedback on the usage and functionality of this 3D virtual platform is collected from
students through post evaluation surveys. The chapter discusses the strengths and
limitations of the 3D virtual environment to support collaborative learning.
In Chap. 16 ‘Spatiotemporal Information System Using Mixed Reality for Area-
Based Learning and Sightseeing’ the authors Ryuhei Makino and Kayoko Yamamoto
develop a system that visualises spatiotemporal information in both real and virtual
spaces, by integrating Social Networking Services (SNS), Web-GIS, Mixed Reality
(MR), and gallery system as well as Wikitude, all connected to external Social Media.
From the evaluation results, it shows that all functions in the system are evaluated
highly, and the majority of functions for area-based learning are most popular. Con-
sequently, the chapter proves the possibility of the system to support both area-based
learning and sightseeing by making use of Virtual Reality (VR), Augmented Reality
(AR) and Mixed Reality (MR).
1 Computational Urban Planning and Management for Smart Cities … 11
2.3 Part 3: Mobility
In the third part of the book we bring together a collection of chapters from colleagues
concerning transportation, mobility, route planning and travel modelling.
In Chap. 17 ‘Origin-Destination Estimation of Bus Users by Smart Card Data’ the
authors Mona Mosallanejad, Sekhar Somenahalli and David Mills optimise public
transport routes and their schedule with the help of smart card data. The smart
cards offer transit planners access to a tremendous source of spatial-temporal data
which can be used for the optimisation of public transport routes and schedules.
The authors developed a new approach using a trip chain model to estimate public
transport commuter’s trajectories in a multi-legged journey. New algorithms have
been developed to link the passenger’s journeys involving the mode transfers using
assumptions relating to the passenger paths in between their successive boardings
and their acceptable walking distances. Ultimately this optimisation will lead to a
higher patronage in the public transport system.
In Chap. 18 ‘The Comparison Between Two Different Algorithms of Spatio-
Temporal Forecasting for Traffic Flow Prediction’ the authors Haochen Shi, Yufeng
Yue and Yunqi Zhou aim to predict traffic flows with diverse methods and com-
pare their differences within the forecasting process. First, two of the most com-
monly adopted methods, Space-Time Autoregressive Integrated Moving Average
(STARIMA) and the Elman Recurrent Neural Network (ERNN), an Artificial Neural
Network, have been harnessed to establish the space-time predicting models. Sec-
ondly, according to the successfully trained models a multi-dimensional comparison
has been performed based on four aspects: interpretability; ease of implementation;
running time and instability. Based on this, the authors conclude with some possi-
ble improvements in the light of their forecasting performance which also indirectly
reflects their unique features and application environments.
In Chap. 19 ‘Developing a Behavioural Model for Modal Shift in Commuting’
the authors Ali Soltani, Andrew Allan and Ha Anh Nguyen explore the determinants
of people’s willingness to transition to more sustainable modes of transport. Using
a discrete choice model, based on outcomes of an online questionnaire survey held
in Adelaide, Australia, the authors determined that home relocation and job changes
were strongly associated with peoples’ modal shift. It appears that car dominance
can be reduced since there is a willingness to opt for non-motorised transport options
and shared mobility services. The chapter concludes with a varied set of transport
policies and strategies addressing different socio-economic groups to increase their
share of sustainable mobility.
In Chap. 20 ‘Planning for Safer Road Facilities for Bicycle Users at Junctions’
the authors Li Meng, Li Luo, Yanchi Chen and Branko Stazic focus on safety as a
major factor in promoting bicycle travel. Junctions in particular appear to be hotspots
of unsafety. This study reviews junction design and traffic flow conditions at an
upgraded junction in Adelaide, Australia which contains bicycle signals and a storage
zone. It is found that the bicycle lane can be designed into two sections to separate
left and right turns. Also, the provision of blue bicycle crossing lanes has potential
12 S. Geertman et al.
to improve cyclist safety by warning pedestrians and motorists of possible cyclist
presence. Furthermore, the study recommends smarter data collection and better
traffic modelling to help test improved infrastructures and policies regarding the
safety of cyclists.
In Chap. 21 ‘Method to Evaluate the Location of Aged Care Facilities in
Urban Areas Using Median Share Ratio’ the authors Koya Tsukahara and Kayoko
Yamamoto develop a method to evaluate the location of public facilities according to
a measure of equity. By evaluating nursing facilities and using the improved Median
Share Ratio (MSR), the authors extract the districts which are short of nursing facil-
ities. The evaluation method was applied to Chofu City in Tokyo Metropolis, Japan.
Therein, a distinction has been made in the evaluation with and without weighting
the MSR by elderly population. It turns out that the evaluation method that includes
weighting makes it possible to adequately identify the districts where new nursing
facilities are most needed.
In Chap. 22 ‘Identifying Changes in Critical Locations for Transportation Net-
works Using Centrality’ the authors Nazli Yonca Aydin, Ylenia Casali, H. Sebnem
Duzgun, and Hans R. Heinimann develop a new method to secure that critical loca-
tions like hospitals will still be accessible in case of a disruption. Crucial in this is
that critical locations change when people move towards a specific service inside its
catchment area. Therefore, the authors developed a modified betweenness centrality
index to identify critical locations when moving towards a single service like a hos-
pital. The index has been applied on a case study from Kathmandu, Nepal. Random
disruptions with increasing magnitude were simulated to understand the networks’
behaviour and to identify the changes in those critical locations under extreme con-
ditions. The results show that the origin-destination betweenness centrality is an
effective index for this purpose.
In Chap. 23 ‘Efficient Regional Travel for Rescue and Relief Activities in a Dis-
aster’ the authors Toshihiro Osaragi, Masashi Kimura and Takuya Oki develop an
optimization method for rescue activities in the immediate aftermath of a large-scale
disaster. Regularly, the locations of demanders (those requiring special care or assis-
tance) and responders (those supporting or assisting the demanders) are often widely
separated. To overcome this distance the authors propose a new method for supporting
efficient travel and navigation for rescue activities by making use of fuzzy c-means
clustering and a genetic algorithm. In that, the differences in workload required by
demanders, the compatibility between responders and demanders, and the urgency
of demanders are also taken into consideration. The chapter concludes with a demon-
stration of the efficiency of the proposed method based on numerical simulations and
field experiments using a web application that incorporates the method.
In Chap. 24 ‘A Two-Stage Process for Emergency Evacuation Planning: Shelter
Assignment and Routing’ the authors Ali Soltani, Andrew Allan and Mohammad
Heydari develop a method for efficient evacuation planning. Urban centers have
become more vulnerable to terrorism attacks and responses at community and indi-
vidual level in the form of evacuation and shelter are needed. Evacuation planning is
a key component of emergency preparedness and requires an integrated analysis of
heterogeneous spatial datasets including population, the road network and facilities.
1 Computational Urban Planning and Management for Smart Cities … 13
Decisions surrounding evacuation focuses on the availability of shelters and the time
required to reach these shelters by the optimal route through an urban area. The
authors examine the process and results of identifying appropriate shelter locations
and efficient routings through the road network for the new town of Sadra, Iran.
In Chap. 25 ‘A Comprehensive Regional Accessibility Model Based on Actual
Routes-of-Travel: A Proposal with Multiple Online Data’ the authors Yuli Fan, Qing-
ming Zhan, Huizi Zhang and Jiaqi Wu develop a model to improve regional acces-
sibility by increasing the accuracy of estimated travel costs. Their objection against
existing accessibility models is that these are mostly focused on either cost, oppor-
tunity, network complexity, or other individual criteria—and fail to cover important
factors including road toll, road condition, actual time table, etc. The model pro-
posed increases the accuracy of estimating the travel cost by making use of actual
time tables and road trip recommendations provided by digital map providers. It
colligates cost criteria, opportunity criteria and network complexity criteria in one
accessibility index by accumulating the value of different actual routes, and in that
potentially providing much more accurate descriptions of accessibility.
In Chap. 26 ‘Taxi Behavior Simulation and Improvement with Agent-Based Mod-
elling’ the authors Saurav Ranjit, Apichon Witayangkurn, Masahiko Nagai and
Ryosuke Shibasaki make use of an agent-based model to gear to one another the
perspectives of the taxi driver and the customer. According to the authors, the driver’s
perspective is that they are working long hours while the income generated does not
justify these hours. On the contrary, the customer’s perspective is that passengers are
often rejected or denied by the taxi service. By developing a taxi behaviour simu-
lation model, the authors try to optimize and thereby improve taxi operations. The
evaluation of the model shows that it generates improvement for both taxi driver and
passenger.
3 Conclusion
This latest volume ‘Computational Planning and Management for Smart Cities’ is the
fourth book in the Springer Lecture Notes in Geoinformation and Geography series
and accompanies the 16th International Conference on Computers in Urban Plan-
ning and Urban Management hosted by Wuhan University, China, 2019. It follows
on from the previous books in the CUPUM conference series including: Planning
Support Science for Smarter Urban Futures (from the 15th International Conference
CUPUM in 2015, University of Adelaide in Australia), ‘Planning Support Systems
and Smart Cities’ (from the 14th International CUPUM in 2014 at Boston’s Mas-
sachusetts Institute of Technology in the United States) and ‘Planning Support Sys-
tems for Sustainable Development’ (from the 13th International CUPUM in 2013 at
Utrecht University in the Netherlands). As noted by Goodspeed et al. (2018) data and
technology have become ubiquitous in urban management and planning scholarship.
Yet we still face challenges in the adoption of technologies such as planning support
systems (Russo et al. 2018) as we continue to plan for ever expanding population
14 S. Geertman et al.
residing in cities and megacities. We hope this volume of scholarly works will pro-
vide an important contribution to the body of knowledge interfacing data, technology
and smart cities as we strive to plan more liveable, sustainable and productive cities
of the future.
References
Berrone P, Ricart JE, Carrasco C, Duch A (2018) IESE cities in motion index. Business School,
University of Navarra. https://doi.org/10.15581/018.st-471
Castells M (2010) The information age: economy, society and culture. In: The rise of the network
society, 2nd edn, vol 1. Wiley Blackwell, Oxford, pp 99–101 (Webster, 2002)
Geertman S, Allan A, Pettit C, Stillwell J (2017) Introduction to ‘planning support science for
smarter urban futures’. In: Planning support science for smarter urban futures. Springer Interna-
tional Publishing, pp 1–19
Goodspeed R, Pelzer P, Pettit C (2018) Planning our future cities: the role computer technologies can
play. In: Sanchez TW (ed) For urban planning knowledge and research. Routledge, pp 210–225
Government of South Australia (Department for Environment and Water) (2019) Urban heat map-
ping of Adelaide metropolitan area. See spatialwebapps.environment.sa.gov.au
Gray S, O’Brien O, Hügel S (2016) Collecting and visualizing real-time urban data through city
dashboards. Built Environ 42(3):498–509
Li S, Dragicevic S, Castro FA, Sester M, Winter S, Coltekin A, Pettit CJ, Jiang B, Haworth J,
Stein A, Cheng T (2016) Geospatial big data handling theory and methods: a review and research
challenges. ISPRS J Photogram Remote Sens 115:119–133
Newman P, Beatley T, Boyer H (2017) Resilient cities-overcoming fossil fuel dependence, 2nd edn.
Island Press, Washington, USA
Pettit C, Lieske SN, Jamal M (2017a) CityDash: visualising a changing city using open data. In:
International conference on computers in urban planning and urban management, Springer, Cham
(July)
Pettit CJ, Tanton R, Hunter J (2017b) An online platform for conducting spatial-statistical analyses
of national census data. Comput Environ Urban Syst 63:68–79
Pettit C, Bakelmun A, Lieske SN, Glackin S, Thomson G, Shearer H, Dia H, Newman P (2018)
Planning support systems for smart cities. City Cult Soc 12:13–24
Rodrigue JP (2017) The geography of transport systems. Routledge, New York. ISBN 978-
1138669574
Russo P, Lanzilotti R, Costabile MF, Pettit CJ (2018) Towards satisfying practitioners in using
Planning Support Systems. Comput Environ Urban Syst 67:9–20
Silva G, Di Serio LC (2016) The sixth wave of innovation: are we ready? J Innov Manage Rev (RAI
Rev Adm) 13(2016):128–134. https://doi.org/10.1016/j.rai.2016.03.005
Sujataa J, Sakshamb S, Tanvic G, Shreyad (2016) In: 6th international conference on advances on
computing & communications, ICACC 2016, Cochin, India, 6–8 Sept 2016
Thakuriah PV, Tilahun N, Zellner M (eds) (2016) Seeing cities through big data: research, methods
and applications in urban informatics. Springer, Cham
UN-DESA (2019) United Nations, News: 68% of the world population projected to live in urban
areas by 2050, un.org. Accessed 10 Feb 2019
United Nations (UN) (Economic and Social Affairs) (2016) The world’s cities in 2016 data booklet,
un.org. Accessed 10 Feb 2019
Zhuhadara L, Thrasher E, Marklin S, Pablos PO (2016) The next wave of innovation—review of
smart cities intelligent operation systems. Comput Hum Behav Elsevier 66:273–281 (Jan 2017)
Part I
Smart City
Chapter 2
Sejong Smart City: On the Road to Be
a City of the Future
Yountaik Leem, Hoon Han and Sang Ho Lee
Abstract Based on its advanced information and communication technologies
(ICTs) and construction industry, Korea has developed an ICT-driven smart city
called Ubiquitous Cities (U-City). One of the strategic adopters of this concept is
Sejong City, which is being developed in central South Korea as a new administra-
tive city. Besides being different from the European model of a smart city, Smart
Sejong City is one of the full-scale greenfield development models of a future city
armed with cutting-edge ICTs. From the beginning stages, high levels of ICT infras-
tructures were facilitated together with urban integrated information centres (UIICs)
and devices for service provision. In addition to transportation and public safety ser-
vices, smart community design that is derived from citizens’ needs and a zero-energy
community strategy for environmental contribution are under development. In this
chapter, the background of Korean smart cities and contents of the Sejong Smart
City are presented in terms of viewpoint of industry-mix, infrastructure, technology
and services, followed by discussion on the future of the smart city.
Keywords Smart city ·Ubiquitous City ·Sejong City ·Korea ·Greenfield
development ·Industry-mix
Y. Leem ·S. H. Lee
Department of Urban Engineering, Hanbat National University, Daejeon, Korea
e-mail: ytleem@hanbat.ac.kr
S. H. Lee
e-mail: lshsw@hanbat.ac.kr
H. Han (B)
University of New South Wales, Sydney, Australia
e-mail: h.han@unsw.edu.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_2
17
18 Y. Leem et al.
1 Introduction
Hundreds of new smart cities have been, or are being, developed around the world.
Numerous sensors and networks are deployed across countries, cities and precincts.
These networked communities are generally accepted as our modern city’s lifestyle,
which is not well regarded as a cutting-edge technology in smart cities but is rather
a basic requirement of global cities. Leading smart technologies such as Internet
of Things (IoTs), Big Data analytics and Artificial Intelligent (AI) are one of the
most prominent themes in smart cities (BMW Guggenheim Lab 2012). However,
limited studies seek to understand how these technologies are implemented to a
greenfield development requiring the segment of smart city services and how people
form networked communities in smart cities. The concept of smart city planning is
to embed information and communication technologies (ICTs) into physical spaces
to enhance the quality of citizens’ lives (Smart Kalasatama 2018) and often aims to
protect the city from natural disasters and the consequences of global warming (e.g.
a resilient city).
The Republic of Korea (hereafter Korea) is one of the leading countries in smart
city planning and, with a high-tech driven planning approach, the Korean government
applied world-class ICTs to the country’s smart city development. Songdo is the
flagship of the Korean smart cities and commonly cited as a best practice in the world’s
smart cities. More recently, Korea planned a new administrative multifunctional
smart city, which will accommodate almost all central government administrative
services in Seoul and Sejong has joined Songdo as a model of the Korean smart city
(Yigitcanlar 2015). Unlike other smart cities in Korea, Sejong City aims to reduce
a regional imbalance by relocating a higher concentration of population and jobs in
the Seoul Metropolitan Area (SMA) and developing a global standard of the future
smart city model.
Han and Hawken (2018, p. 3) argued that ‘fostering the distinctive digital cultures
that are increasingly evident in smart cities worldwide will allow cities to become
more sustainable and resilient’. This chapter will show how Sejong City contributes
to creating high-quality, multifunctional and liveable places, rather than a single city
service to reach the top-end of digital technology solutions. Further, this chapter will
contribute to international smart city planning as there is a limited smart city case
in the world for developing a master-planned, greenfield-based and multifunctional
smart city. This unique example will provide a guideline for comprehensive and large-
scale smart city planning that considers urban design and tailored ICT development,
as well as their implications for service integration and infrastructure provision. The
chapter will first review the smart city planning strategy in Sejong with detailed dis-
trict design planning and introduce key innovative smart technologies implemented
in the city.
2 Sejong Smart City: On the Road to Be a City of the Future 19
1.1 Pathway to Smart Cities in Korea (2000–2029)
After experiencing serious destruction during the civil war (Korean War, 1950–1953),
Korea has achieved one of the most successful economic growth periods in the world,
known as a ‘miracle of Han River’. In 2016, Korea placed 11th in global GDP
rankings, having achieved remarkable growth in the IT and digital sectors (World
Bank 2017). Digital devices, digital networks and Internet of Things (IoTs) sectors
in Korea have received heavy investment led by the Korean tech giants—Samsung,
LG and SKT. This allows Korea to test numerous new digital solutions and efficient
system integration in cities. The 2000s saw the arrival of the first generation Korean
smart city, the so-called ‘U-City’ (Ubiquitous City), focusing on the ubiquity of
urban services and infrastructure, which citizens can access anytime and anywhere
(Lee et al. 2008; Yigitcanlar and Han 2010a). The second generation smart city
in Korea focuses on urban ecology and environmental conservation (Yigitcanlar
and Han 2010b). The U-Eco City aims to overcome the climate change and energy
crises by adopting Eco-technologies (EcoTs) for zero-energy community planning.
These EcoTs help citizens to reduce daily energy use through a real-time energy
consumption monitoring and warning system.
Generally, Korean planning aims to address unbalanced economic growth by
correcting regional disparities. However, a large proportion of greenfield-based smart
cities (i.e., Songdo, Dongtan) in Korea are sited in the SMA, where more than 50%
of the nation’s economic share is located. In 2002, a presidential candidate suggested
a new capital city external to the SMA to decentralise the governance of Seoul. This
idea was intended to reduce regionally imbalanced development but encountered
increasing public opposition as a new capital city could potentially lead an economic
downgrade of Seoul. After extensive discussions and public scrutiny, Seoul remains
the capital city of Korea and Sejong is planned as a new administrative multifunctional
city for the Korean government located in the centre of South Korea. The land size is
72.91 km2and over 80% of the central governance will be relocated to this master-
planned smart city with dedicated residential, commercial, cultural and research
precincts.
1.2 Strategic Planning of Sejong City
Sejong is planned on an area of 72.91 km2within 2 h driving time from Seoul (see
Fig. 1). The target population is 300,000 people, with government officials and their
families as well as employees of related companies and institutes moved from Seoul
and its adjacent cities. The massive-scale greenfield development project was led by
the Korean central government in a top-down approach; various ideas were collected
and adapted during the planning process through international design competitions,
with consulting committees and public participation regarding a range of issues from
urban structure to community design as a bottom-up approach.
20 Y. Leem et al.
Fig. 1 Location and facts of Sejong City (source NAACC 2017)
The primary goals of Sejong are to enhance regionally balanced development by
reducing spatial disparities between the Seoul Metropolitan Area (SMA) and the
outer regions of Korea, and to increase national competitiveness by developing a
model smart city (NAACC 2008). In the initial stage of the development, a large
amount of greenfield development was planned to relocate central government func-
tions to the city. To enhance national competitiveness, accumulative cluster planning
was adopted to allocate R&D facilities and high-tech industries to the Sejong master
plan. A subsequent plan for redistribution of the city’s functions to adjacent cities
was also considered in strategic planning. The location of Sejong is proximate to
the Daejeon Metropolitan City, which has a population of over 1.5 million and is
the location of the Daedok research precinct, a world-renowned R&D cluster. This
location allows a synergistic effect in R&D, entrepreneurship and administration for
future growth potential.
Through an international design competition for the new city, a ring-shaped urban
design was selected and assigned major urban functions—central and local admin-
istration, culture and international exchange, high-tech industries and medical, wel-
fare and research facilities with universities. Offices and amenities are located within
walking distance and major functions are connected by a double-layered ring road.
All these major functions are assigned based on the ICT infrastructure and the resi-
dential and commercial area required (see Fig. 2). The ring-city design with a service
clustering strategy is a unique characteristic of the new city, which differs from a
focus in other cities on mixed-use development. The Sejong city masterplan con-
siders both positive and negative externalities of the industry clustering strategy.
Hi-tech industry cluster like Silicon Valley, US, synergises other similar industries
by a spill-over effect, knowledge sharing and labour pooling but the homogenous
industry (e.g. ICT) within a high-tech industry cluster is relatively vulnerable to a
2 Sejong Smart City: On the Road to Be a City of the Future 21
Fig. 2 Urban structure and main function allocations of Sejong City (source NAACC 2017)
global recession. In this regard a mixed land use and industry mix strategy becomes
increasingly popular in the contemporary city planning. Hawken and Han (2017)
also pointed out the importance of industry mix strategy for Sydney downtown, Aus-
tralia. The potential negative effects of clustering high-tech industries in Sejong can
be mitigated by the double layered ring design with a series of service clusters. This
unique city design could provide not only an even service accessibility to each city
services/facilities but also assure service quality and improved choices by clustering
the same service function. This allows same industry clustering within the cluster
but different industry mix between the clusters as a Win-Win strategy.
Although there are other new administrative cities, including Canberra in Aus-
tralia, Ankara in Turkey and Putrajaya in Malaysia, Sejong has the more definite
goal of balancing national growth to overcome regional–economic disparities. The
dynamic clustering (ring-pattern) in Sejong could result in further uneven growth,
with skilled labor relocating to Sejong and unskilled labor being priced out of the
city. This is a major planning challenge in Sejong at this early stage of develop-
ment and should be addressed in the next stage of the city plan. Further, innovative
smart technologies are planned for application in this city. Beyond new urban design
technologies like transit-oriented development, Sejong is also the testbed for spatial
information research (National R&D for Geo-database) and future ICT-based cities
(National R&D for U-Eco City). Planning and design technologies were adapted
together with devices and control platforms, which will be operated behind offices
and embedded into physical spaces. In particular, the U-Eco City is one of the most
ambitious research and development topics by which Korea intends to improve the
quality of life of its citizens with the development of ICTs industry.
22 Y. Leem et al.
1.3 Planning of Sejong Smart City: A Pathway from U-Cities
During the development stage, one of the main planning goals was to create a high-
tech digital city to enhance citizens’ quality of life. Thus, the U-City (Ubiquitous
City) Strategic Plan was prepared by the National Agency for Administrative City
Construction (NAACC), the major public development authority for the Sejong smart
city.
In the first stage of U-City planning for Sejong, Uvolution: A Futuristic Evolutive
City through U-network was chosen as a vision of the future city (NAACC 2007).
Three planning goals were set up: (1) creation of new urban space; (2) gaining a
new competitive advantage; (3) establishing urban value capture with 10 municipal
services provided to the city. In 2006, Korea Land and Housing Corporation (LH)
completed the ‘Implementation plan and basic design for U-City of multifunctional
administrative city’ (KLHC 2006). Based on this, a specific design for physical
installation of digital networks, devices and control centres was planned over the
development stages.
For the implementation of smart city principles across the development stages of
Sejong, designs for IT infrastructure, control centres and urban facilities are required
and must be linked with seamless and ubiquitous sensors and networks. As such,
detailed designs for wired and wireless network will be deployed to integrate infor-
mation operation centres and intelligent building and facilities. Public services such
as transport management and public safety are also integrated with smart environ-
ment services for carbon emission reduction.
2 Features of Sejong Smart City
2.1 ICT Infrastructure for Sejong City
The installation of wired and wireless networks is one of the most important factors
to provide ICT services in a city. For the smart city service, a 395 km-long public
communication network was facilitated along with urban development (see Fig. 3).
Further, private networks are also available for specific services such as transporta-
tion or for commercial purposes. In smart cities, moving objects (humans, cars, etc.)
are very important information sources and are simultaneously targets of smart city
services. Connected to a wired network, many access points for wireless communica-
tion were installed for mobile devices with consideration of the serviceable distance.
One interesting aspect is that there are ‘digital-free’ areas where smart city services
are not possible (UCRC 2016).
As an urban operation platform, the urban integrated information centre (UIIC)
was facilitated at an early stage of the development of Sejong City. UIIC plays a
core role and operates ICT infrastructure and controls real-time situations in the
city. Its main function is to collect, analyse, process and integrate various data from
2 Sejong Smart City: On the Road to Be a City of the Future 23
Fig. 3 Wired networks and their conduits for Sejong Smart City (source www.sejong.go.kr/
smartcity/sub01_03.do)
urban facilities and devices to provide efficient management of, and services to, the
city. The UIIC accommodates a traffic and crime information control room, cyber
infringement response centre, local administration information system, underground
utility pipe conduit control room, urban energy integrated operation centre and a
system equipment room in an exclusive building of 2977 m2spread across three
stories.
2.2 Open Data and Smart Services of Sejong City
Sejong has a system integration that is unique among the administration systems
implemented in Korea. While the country’s administration system is divided into a
regional level and a local level, Sejong has a hierarchical system with both levels.
For instance, during the development of a local administration system, one of the
most important considerations is the linkage of a large amount of open data to spatial
information collected by local sensors. Such local-level open data already exists but
is used and stored in isolation. For instance, open data such as traffic volume data and
pollution monitoring data should be integrated to managing transport services. Local
level open data is transferred to the data control centre simultaneously and immediate
action should be taken on a case-by-case basis. This requires an automatic decision
system with complex algorithm that is scalable from a local to a regional data level
so that the system may provide a range of municipal services to local governments
as well as the private sector.
A new smart city requires various urban facilities. In Sejong, almost all service
utilities are integrated in underground utility conduits, even the household waste
treatment system (see Fig. 4). These urban facilities are connected and share open
24 Y. Leem et al.
Fig. 4 Smart urban facility management based on GIS and the Urban Information System (source
NAACC 2017)
data. Most importantly, urban facilities are controlled based on integrated databases
that include their history. All locations and situations are presented on digital maps in
real-time and underground facilities, in particular, are managed very carefully. The
IT network is strictly protected and the water network is also monitored for quantity
and quality.
Some of the most important facilities are the roads and various devices that service
them. The condition of tunnels and bridges is continuously monitored and the data
are sent to the UIIC. Multifunctional street lights with sensors and other devices are
remotely controlled (Fig. 4).
Sejong smart city services are focused on safety and transportation. Based on
the intelligent transport system concept, a data collection and manipulation system
was implemented in Sejong. Well-managed transportation systems increase the qual-
ity of life for citizens by reducing the time lost and decrease the overall effect on
global warming by promoting public transportation. The UIIC integrated transport
management system collects and manages transport data for operation of public
transportation, public bike rental and management of demand, traffic flow and the
parking system (see Fig. 5). Big data on road condition and public transport situa-
tions are provided to each user and the collective data becomes the basic material for
future transportation policies.
Public safety is one of the most important factors for quality of life in a city.
The Sejong smart city plan indicates that ICTs should play a major role in crime
prevention to improve the lifestyle quality for citizens. As a newly developing city,
it adopted cutting-edge technologies for public safety linked with police precincts.
Although numerous CCTV systems are largely ubiquitous in major cities, video
image data are not effectively shared with police, bankers, fire fighters and doctors
2 Sejong Smart City: On the Road to Be a City of the Future 25
Fig. 5 Smart transportation services in Sejong City (source NAACC 2017)
(Han et al. 2015). In Sejong, CCTV data is managed by multiple departments related
to transportation and forest management, and waste disposal has been linked with
the police, which not only improved the crime detection rate but also helped to
alleviate security concerns. Much of crime and accident-related data flow to a fire
and rescue department and a hospital. Accident and road collision data are sent to
the UIIC and simultaneously shared with rescue agencies and hospitals (see Fig. 6).
In emergencies, personal information including headshot images and locations are
transmitted to the authorised agencies responsible for responding.
3 Efforts for the Future of Sejong Smart City
3.1 A Smart City for the Environment
During the development of Sejong, most of the experts engaged in the smart urban
service recognised the importance of environmentally friendly urban planning for
a future-oriented city as well as the role of data from daily human life. Samsung
SDS (2011), the contractor of new electronic administration systems development
and delivery, acknowledged this and resolved to include a pilot system for a Sejong
carbon emission monitoring system (CEM). This system is linked to administrative
and spatial data managed by the local government as well as census data (see Fig. 7).
These data allow local government officers to analyse energy consumption behaviour
26 Y. Leem et al.
Fig. 6 Expansion of CCTV linked with police and rescue systems (source NAACC 2017)
Fig. 7 Sejong CEMS system structure (source Leem et al. 2013)
and build policies to reduce greenhouse gas emissions. This is expected to reorient
citizens’ behavioural patterns towards a more efficient, energy-effective urban life.
The Ubiquitous City Research Cluster (UCRC) at Hanbat National University,
Korea provided a structure for the Sejong CEM. This will directly collect and manip-
ulate the everyday energy usage data of each household (electricity, city water, gas
and heat). In combination with urban administration data such as household profiles
and building information, the CEM analyses the energy consumption pattern of each
family compared to others with the same house type and similar family structure.
2 Sejong Smart City: On the Road to Be a City of the Future 27
Aggregated carbon emission data calculated for each type of energy consumption
are presented to public officers while the energy consumption behaviour of each
household is given to the residents. The most prominent feature of this system is that
it is designed as a two-way system: energy consumption log data can be transferred
upwards (to local government) for policy preparation and downwards (to citizens) for
reducing energy consumption, contributing to the economy and challenging global
warming.
In CEMS socio-economic information such as the area of a building or the resi-
dent’s information is inputted as basic data. In addition to analysing cross-sectional
information related to energy consumption by building, household, day or season,
CEMS can analyse trends by accumulating time series and distribution data for each
analysis unit. Overall, the system can analyse energy consumption in each period
according to the natural conditions of Korea, which has four seasons, and apply this
to the policy immediately.
3.2 Smart Zero-Energy Community Planning
Despite the many varying visions of a smart city, one of the most important goals is to
improve energy efficiency and reduce carbon emissions to combat global warming
(Han et al. 2018; Thorpe et al. 2012). Gilijamse (1995) defined a ‘zero-energy’
community as one that balances its electricity consumption and production while
not consuming any fossil fuel. The Korean central government promotes a smart
zero-energy community as a future city model in Sejong City. The Sejong smart city
plan is to create a smart zero-energy community rather than a partial plan such as the
promotion of public transportation. To this end, the Smart Zero-energy Community
aims to make Zero Energy City, Zero Emission City and Zen Emotion City (ZEC) a
place where citizens can pursue a healthy and enjoyable life.
Sejong City already has well-established infrastructure and significant related
services. Therefore, if Smart ZEC in Sejong City is presented, it is expected to
improve the quality of life of residents and expand the interest and investment of
international smart city proponents. One of the important reasons for promoting the
smart city concept is to nurture it as a next-generation industry. In addition, the Korean
government aims to promote the concept to export it and Smart ZEC is expected to
enable the development of related industries and smart city exports.
The Sejong Smart ZEC plan has three major parts. First is the building of smart
infrastructure to support future transportation systems, energy saving systems and
safe spaces such as unmanned vehicles. Second is the promotion of resource cir-
culation, carbon neutrality and water quality management through smart eco-plans.
The final goal is to construct a physical space that enhances citizens’ quality of life
through ICT-based healthcare, education and cultural services.
Smart City planning can be realised in combination with urban planning by estab-
lishing a service placement plan in conjunction with urban design to include smart
city services in the space. This design technology has already been secured through
28 Y. Leem et al.
Fig. 8 Sejong City zero Energy city comprehensive plan (source NA ACC an d L.H. 2017)
U-City planning and design, and a strategy to expand the unit building-community-
urban space agenda was applied. Figure 8shows the three themes of Sejong ZEC:
‘Smart Infra’ for energy, transportation and safety; ‘Smart Eco’ for carbon emission,
resource circulation and water management; ‘Smart Life’ for citizens’ health and
welfare, culture and education.
3.3 Smart Community
The final goal of the Sejong smart city plan is to enrich the future lifestyle into
citizens’ everyday lives. As part of the National R&D Project, the UCRC planned
a community unit smart city where everyday life takes place in Sejong. Using the
persona method, normal behaviours of citizens were analysed (An et al. 2016) and
three themes were selected for smart community planning as follows: circulation (the
way to walk without stopping), the moment (the place to experience the moment)
and the liveliness (a pleasant space for everyone).
For this purpose, multiple features are included in planning: a safe walking envi-
ronment where vehicles and pedestrians communicate with each other; a media
façade by which the exterior walls of the building become human-computer inter-
faces; a smart work centre that saves commuting time and a parking space where an
office and adjacent residential area share a parking space (thus saving 48% of the
physical space). Based on ICTs networks and space-embedded devices, these smart
city services are intended to improve spatial efficiency and citizens’ quality of life.
2 Sejong Smart City: On the Road to Be a City of the Future 29
Fig. 9 Structure of COS for Sejong Smart City media Façade services (source UCRC 2016)
Together with the urban design and service plan, the Community Operating Sys-
tem (COS) was developed to operate as a platform to manage these services for
citizens (see Fig. 9). To increase the quality of physical space, some devices were
designed to fit the urban context with ICT functionality.
4 Discussion: Barriers and Benefits
Despite the wave of future spaces based on informatisation, the early U-City (which
mainly applied to new cities) in Korea did not attract much interest in the market. This
is because they focused on building infrastructure related to IT and could not offer
tangible consumer services. Further, there was insufficient participation of private
entrepreneurs such as corporations and citizens (The Dong-A Ilbo 2017). Both the
U-City and its business model failed to be profitable as a commodity because it is a
government-led business. However, there is enough evidence that the Sejong City is
innovative in terms of community participation, statuary planning and data security.
4.1 Open Data Analytics for Community Participation
The Sejong smart city is substantial example of spaces with embedded ICTs and
EcoTs where citizens can access open data and eco-services anywhere, anytime
30 Y. Leem et al.
like a ubiquitous city. It could thus answer the world’s question of how to build a
next generation smart city. The Korean case of Sejong provides innovative solutions
to urban problems through Public Private People Participation (PPPP or 4Ps) and
cooperation. Public Private Partnership (PPP) has been often adopted in the planning
and development project. However, the PPP overlooks the significance of a bottom-
up approach and planning decision is often led by a top-down approach. One missing
piece of the puzzle is the People in the planning decision process. Ahmed and Ali
(2006) pointed out citizens significantly contribute to service delivery and play a
moderating role in enhancing accountability and service quality of both public and
private sector.
The city uses the collective intelligence of a democratic society, which leads the
traditional city towards low cost and high efficiency through smart facility manage-
ment based on a geographic information system (GIS) and its Urban Information
System, Local Administration System, Real-Time Traffic Control System and UIIC.
Recently, Sejong opened big data analytic platforms such as the UIIC, where many
diverse open data are collected and linked to tailor many services according to pub-
lic and private demand. GIS data sometimes meets the sensing data from intelligent
infrastructures and citizens both give and take data as consumers and producers
through the UIIC. Real-time and historical structured and unstructured data are gath-
ered and analysed to be supplied to anyone, anytime, anywhere.
Using the cutting-edge technologies Sejong has evolved the eco city into the eco
intelligence city to store. The city uses big data for analysing climate change and
energy use. For instance, Sejong CEMS is the platform for zero-energy community
planning with citizens, who can save energy by following energy consumption warn-
ings. In addition, ICTs and/or EcoTs embedded in eco intelligence provide citizens
with green accessibility such as clean water and fresh air through automatic water/air
pollution monitoring. The city was a pioneer of IoT in the field of ICTs and/or EcoTs
embedded intelligent infrastructure. Intelligent infrastructure provides citizens with
a traffic, public bus and smart parking services. Foreigners can access these services
through a QR code. The Sejong complex community centre has built a media façade
and canvas for both cyber and face-to-face communication.
Most importantly, Smart Sejong has made smart citizens to improve quality of
life and to increase job diversity. It showed the possibility of a smart and sustainable
society. What does the future hold? The city is working towards a fourth industrial
revolution by creating a new industrial ecosystem for the completion of Smart Sejong.
4.2 Legal and Social Barriers
Korea has legal and social restrictions on the acceptance of ICTs in each sector. For
example, despite the highly developed U-Health technology, doctors cannot provide
ICT-based health services to citizens living in areas where there is a lack of healthcare
due to laws stating that medical treatment can only be given face-to-face. Similarly,
although technical preparations for the implementation of distance education have
2 Sejong Smart City: On the Road to Be a City of the Future 31
been completed, U-Education has not been fully realised due to insufficient social
conditions. This means that a smart city can be operated not only by the development
of technology, but also by various social systems such as administration, culture and
economy. It is important to ensure that smart cities improve citizens’ daily lives
while meeting social demands, rather than simply following a hasty policy of pilot
smart city establishment and exportation. For example, in Kalasatama Smart City,
Finland, the slogan ‘One more hour a day’ is presented and offers citizens an extra
hour through the reduction of commuting time (by eliminating traffic congestion) and
decreasing waiting time through computerisation of administrative services (Smart
Kalasatama 2018). Therefore, it is more important for Korean smart cities to provide
services matching the needs of their citizens and to build a social system that can
apply smart city principles to the local situation of each city or region.
In Korea, the six vulnerable groups (the handicapped, elderly, low-income earners,
farmers, defectors from North Korea and marriage immigrants) have been consid-
ered. Recently, the gaps of these vulnerable groups have been narrowed compared to
the past, but they now show different patterns. As ‘smartisation’ has progressed, the
access gaps are widening in qualitative achievements (such as capacity and utilisa-
tion) rather than in simple accessibility, gaps within the group rather than between
groups and the mobile or multiple access to a PC. The demand for information access
rights grows as universal citizenship increases (NIA 2016). If smartness becomes an
index that distinguishes between informatised and non-informatised spaces, it can
be regarded as a side-effect of smart cities and planners should prepare accordingly.
4.3 Data Security and Privacy
Data security and privacy are another issue of informatisation. Recently, the instal-
lation of CCTV and dashboard cameras has increased, but the situation is unclear.
These devices are an important requirement for the acquisition, connection and shar-
ing of information in a smart city. However, while the CCTV system is established to
expand the social safety net for crime and disaster prevention, it may infringe on an
unspecified number of rights (right of self-determination for personal information,
privacy and freedom of privacy, etc.). If it is linked with a GIS, there are problems in
that electronic surveillance can expose personal location through facial recognition
functions. In addition, due to the emergence of network CCTV and the increase of
CCTV-integrated control centres, there is a need for CCTV administration and techni-
cal measures to protect against the personal data leakage through hacking and viruses.
5 Conclusion
Korea experienced rapid development of its ICT sectors and has prepared for these
industries to be a locomotive of the national economy. Most of industries in Korea
tried to reflect this trend into their own field, for example, automation in the auto-
32 Y. Leem et al.
mobile industry and ‘fin-tech’ in financial sectors. One of the trial but nevertheless
fast growing sectors involves embedding ICTs in people and places. This was first
accepted by the real estate industry to advertise property sales with a location service.
However, urban planners and scientists remain uncertain about the role of ICTs in
urban spaces and in future cities. Lee (2007) argued that the role of ICTs in future
urban space is to improve the quality of life by solving current urban problems and
leading a future lifestyle in advance. It proposed a sustainable society as a reason to
integrate ICTs in spaces (Curwell and Hamilton 2003).
Despite various practical constraints and challenges, the Sejong smart city in
Korea has a nationally and socially supported role that it is gradually carrying out. As
urban development achieves its physical targets, the city is also completing its func-
tions. Sejong should answer not only to building physical infrastructures, devices and
services, but also to the social responsibilities of future cities (Spinak and Casalegno
2012). This is why cities can only become smart when they are able to balance tech-
nology and its benefits (BMW Guggenheim Lab 2012). It is expected that Sejong will
present a smart model of greenfield development to the world because it pursues a
technology solution first at the beginning of masterplan, avoiding the legacy problem
of retrofitting in many cities where the system and service integration is difficult.
References
Ahmed S, Ali S (2006) People as partners: Facilitating people’s participation in public–private
partnerships for solid waste management. Habitat Int 30:781–796
An S, Leem Y, Kim S, Lee SH (2016) A study of media façade service design for promotion of local
community. In: Chien S, Choo S, Schnabel MA, Nakapan W, Kim MJ, Roudavski S (eds) Living
systems and micro-utopias: towards continuous designing. Proceedings of the 21st international
conference on computer-aided architectural design research in Asia, pp 55–62
BMW Guggenheim Lab (2012) 100 urban trends. Retrieved from http://www.bmwguggenheimlab.
org/press/press-release-archive/2012/342-bmw-guggenheim-lab-publishes-100-urban-trends
Curwell S, Hamilton A (2003) IntelCity Roadmap ver. 4, IST-2001-7373, EU
Gilijamse W (1995) Zero-energy houses in the Netherlands. In: Proceedings of building simula-
tion ’95, the fourth conference at the international building performance simulation association.
I.B.P.S.A, Madison, Wisconsin, pp 276–283
Han H, Hawken S (2018) Introduction: innovation and identity in next-generation smart cities. City
Cult Soc 12:1–4
Han JH, Hawken S, Williams A (2015) Smart CCTV and the management of urban space. In:
Harrison D (ed) Handbook of research on digital media and creative technologies. IGI Global,
Hershey, PA, pp 430–447
Han JH, Kim S, Kim J, Lee S (2018) A review of zero energy housing regulations for low-income
households. Int J Knowl Based Dev. Advance online publication. https://doi.org/10.1504/ijkbd.
2018.10012269
Hawken S, Han JH (2017) Innovation districts and urban heterogeneity: 3D mapping of industry
mix in downtown Sydney. J Urban Des 22(5):568–590
Korea Land and Housing Corporation (2006) Implementation plan and basic design for U-city of
multi-functional administrative city (in Korean)
Lee YH (2007) U-city strategy for multi-functional administrative city. Plann Policy 5:32–41
(Korean Institute for Human Settlement (in Korean))
2 Sejong Smart City: On the Road to Be a City of the Future 33
Lee S, Han H, Leem Y, Yigitcanlar T (2008) Towards ubiquitous city: concept, planning and
experiences in the Republic of Korea. In: Yigitcanlar T, Velibeyoglu K, Baum S (eds) Knowledge
based urban development: planning and applications in the information era. Information Science
Reference, London, pp 148–170
Leem Y, Lee SH, Kim M (2013) Households’ characteristics in energy consumption—data from
carbon emission monitoring system (CEMS) in Sejong City, Korea. In: Proceedings of 6th knowl-
edge cities world summit (KCWS 2013). Lookus Scientific, Istanbul, Turkey
NAACC (2007) U-City strategic plan for multi-functional administration city (in Korean)
NAACC (2008) Comprehensive plan for multi-functional administration city (in Korean)
NAACC (2017) Environment-friendly smart city: administrative city (not published)
NAACC and Korea Land and Housing Corporation (LH) (2017) Happy city: administration city (in
Korean)
NIA (2016) 2016 The Report on the Digital Divide, NIA IV-RER-C 15038, Seoul: National Infor-
mation Society Agency of Korea (in Korean)
Samsung SDS Consortium (2011) Final report of Local Administration System Development for
Sejong City
Smart Kalasatama (2018) One more hour a day. Retrieved from https://fiksukalasatama.fi/en/
Spinak A, Casalegno F (2012) Sustainable and equitable urbanism: the role of ICT in ecological
culture change and poverty alleviation. In: Ercoskun OZ (ed) Green and ecological technologies
for urban planning: creating smart cities. IGI Global, Hershey, PA, pp 37–57
Thorpe G, ArziFard A, Kashuk S (2012) Sustainable cities and the convergence of knowledge. Int
J Knowledge-Based Dev 3(4):388
UCRC (Ubiquitous City Research Cluster) (2016) A research report for the advancement of U-City:
Daejeon, Korea (in Korean)
Yigitcanlar T (2015) Smart cities: an effective urban development and management model? Aust
Planners 52(1):27–34
Yigitcanlar T, Han JH (2010a) Urban telecommunications network: technology convergence and
urban infrastructure. In: Yigitcanlar T (ed) Wireless technologies: concepts, methodologies, tools
and applications. IGI Global, Hershey, PA, pp 77–90
Yigitcanlar T, Han JH (2010b) Ubiquitous eco cities: telecommunication infrastructure, technology
convergence and urban management. Int J Adv Pervasive Ubiquit Comput 2:1–17
Web Sources
City of Kalasatama, Smart Kalasatama Homepage. https://fiksukalasatama.fi/en/building-blocks/
one-more-hour-a-day/. Accessed 16 Nov 2017
Sejong City Homepage. www.sejong.go.kr/smartcitysub01_03.do
The Dong-a Ilbo Homepage (2017) http://news.donga.com/Main/3/all/20171030/87013115/1
World Bank (2017) https://databank.worldbank.org/data/download/GDP.pdf
Chapter 3
Data Protection Law and City Planning:
Role of Open Data in Climate Resilience
and Governance of National Capital
Territory of Delhi, India
Mahak Agrawal
Abstract In planning profession, every step of the planning process is guided by
data-either in the form of primary data or secondary data, collected by various insti-
tutions and agencies, but rarely shared in public forum. Each project allocates a
significant portion of the project timeline and cost for data collection, but these
costs and challenges can be drastically reduced if the data is open. This chapter
highlights a case example on the role of open data in climate resilient development
for Delhi. The example indicates the use of open data, available on the geoportal
of international agencies, national agencies and local government, in conducting
spatial-temporal analysis of spatial development and climate change environmental
degradation, for three decades within a time span of 5 months. The study is grounded
in remote sensing based on Landsat imageries made available from the online portal
of United States Geological Survey (USGS) and a digital elevation model available
from the USGS as well as national geoportal called Bhuvan. The models built from
these aerial imageries are empirically analyzed in a correlation model supported by
data inputs through national, state and local government data available on various
government portals of Delhi, including Geospatial Delhi Limited. The crux of the
research highlighted in the chapter, is to emphasize on the importance of open data
for urban planners and administrators of India in ever changing dynamics of city and
its region—and the need for climate resilient urban strategies for a sustainable urban
future.
Keywords Open data ·Climate resiliency ·Urban development ·Climate
variability ·Climate-sensitive planning ·Informed urbanization
M. Agrawal (B)
ISOCARP Review, The International Society of City and Regional Planners,
The Hague, The Netherlands
e-mail: mahakagrawal2016@gmail.com
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_3
35
36 M. Agrawal
1 Introduction
Climate Change is a global phenomenon and variability of climate over decades
is attributed, directly or indirectly, to human activity which alters the composition
of atmosphere, contributing to natural climate variability observed over comparable
time periods. Emerging trends of climate change indicate a global rise in human
induced warming, higher than the natural warming of Earth; which will continue to
increase at a much rapid rate. The IPCC 2014 report highlights that although the
population grew from 4 billion to 7 billion (which is a 75% increase post 1970), the
greenhouse gas emissions increased by 82%.
The recently ratified Paris Agreement, adopted in 2015, addresses climate change
from global scale issues to urban scale development. The environmental impacts of
urban development is not a new finding. In 2008 urban development became central
to the international discourse on climate change, when the global urban population
increase to the 50% mark coincided with the finding that 70% of total GHG emissions
resulted from cities (UN-Habitat 2011). Four cases for this situation were identified
including land use and land cover change, transportation, building construction and
pollution problems related to industry. These four categories of activities coincide
with areas where the role of urban planner and urban planning fits in—in terms
of planned development, adapting to climate impacts and mitigating future risks.
Essentially, cities and climate change are intertwined, and urban planning plays a
vital role in this equation.
2007–08 also saw the publication of a tremendous amount of literature and
research works on cities and climate change. In India a plethora of research has
been produced which studies the impacts of climate change, the spatial distribution
of greenhouse gas emissions, and urban drainage in relation to changing intensities
of precipitation. But only a few try to think beyond these topics and assess climate
change in terms of urban development patterns. The need for studying climate change
and the impact of urban development on climate change becomes more important
today since cities cover less than three per cent of the Earth’s surface but contribute
over 70% of GHG emissions and account for 75% of global energy consumption
(UN-Habitat 2011). Moreover, this trend of urban growth is forecast to continue as
rural populations decline.
This chapter emphasizes the utility of open data in climate resilient urban future
through spatial-temporal analysis of inter-relationship between climate variability
and urban development in Delhi for the time period 1986–2016. It is structured
into three broad sections. In the first section, urban development trends exhibited
by Delhi are described in terms of the increase in built up areas, loss of heat sinks
and loss of flood plains. In the second section, climate change is assessed in terms of
natural climate variability and human induced climate variability. Climate variability
is examined and reflected through temperature and precipitation variables. In the last
section, this climate variability is examined in light of urban development and a
relationship is developed between the two. The section also highlights the vicious
3 Data Protection Law and City Planning: Role of Open Data … 37
cycles Delhi has entered into and concludes by identifying the dire need for climate
resilient urban strategies for a sustainable urban future of the city.
2 Urban Development Pattern of Delhi, 1986–2016
Delhi, a cosmopolitan administrative center of India, is testimony to numerous
changes and cumulative challenges. This section is a documentation of the changes
alone and underscores the trend of urban development in the city. The trend is assessed
using the Landsat imagery and digital elevation models. With raw imagery obtained
from the online portal of the United States Geological Survey (USGS), the land cover
and land use pattern of Delhi has been classified with the help of GIS software, for
1986, 1996, 2006 and 2016. The spatial-temporal analysis based on digital elevation
models is used to assess built up on natural drainage and thereby the loss of flood
plains.
2.1 Change in Built Up Area
The change in land cover of Delhi indicates that the city spread post 1986 around
its core with infill developments. Also, developments in 1986–1996 were 1.3 times
the development of the preceding two decades, which may be attributed to the real
estate growth that emerged post-1980s Asian Games and economic liberalization of
the country’s economy in 1990. Through GIS analysis, it is analyzed that Delhi’s
developed area increased from 39.2% of city total in 1986 to 58.2% in 2016.
The increases in population and the built-up areas in the city, resulted in the
conversion of agricultural fields into non-agricultural use like residential, commercial
and such other non-permeable concrete jungles. The city has witnessed a rise in
its density as well. Empirical and spatial-temporal analysis, supported by GIS and
correlation tests on Microsoft Excel, note that for a population increase of 12.4
million during the three decades, there is a corresponding increase in developed area
from 581.45 km2in 1986 to 86,350 km2in 2016. This is accompanied by an increase
of urban density by 2.6 times, an increase of developed area density by 1.8 times and
an increase in gross residential density by 2.14 times.
2.2 Depletion of Heat Sinks
These developments have come up by engulfing the natural green area (as indicated
in Fig. 1) and flood plains of the city, thus disturbing the city’s microclimate and
ecological balance. Both of these natural areas serve as heat sinks. Empirical analysis
indicates that the city had 57.5% of its area under heat sinks in 1986 which came
38 M. Agrawal
Fig. 1 Pattern of urban development and associated loss of heat sinks in Delhi, 1986–2016.
Extracted by Author (2017) from USGS (1986, 1996, 2006 and 2016)
down to 37% in 2016, with an annual rate of depletion equivalent to 1.4%. Moreover,
it is observed that heat sinks are depleting at a much faster rate (equivalent to 1.4%)
than the rate at which built up area is increasing (equivalent to 1.3%). Also, it is
found that with for every 100 ha increase in built up area (total 28,205 ha), there is a
consequent loss of 94 ha of vegetative heat sinks and 6-ha loss of water-based heat
sinks.
2.3 Loss of Flood Plains
The city comprises of 24,840 ha of flood plains of which 68% forms a part of the
river Yamuna floodplains. The city has three drainage basins based on the watershed
that includes the North Basin with a basin area of 26,694 ha; the West basin with an
area of 75,633 ha, and the South and East Basin spread over an area of 45,973 ha.
Assessing the development pattern of Delhi, it is observed that city has lost over
41% of its flood plains and the loss has increased by 1.4 times since 1986 (as indicated
in Fig. 2). Moreover, the city’s flood plains have reduced in width from 800 m in
3 Data Protection Law and City Planning: Role of Open Data … 39
Fig. 2 Loss of flood plains in Delhi, 1986–2016. Extracted by Author (2017) from USGS (1986,
1996, 2006 and 2016)
1986 to 300 m in 2016 as a result of construction and developments that located in
flood plains.
Summing up, the National Capital Territory (NCT) of Delhi in the past three
decades witnessed a significant change—derogatory for the environment and urban
future of the capital city.
3 Climate Change in Delhi
The climate of NCT of Delhi is categorized into four seasons by the Indian Mete-
orological Department-winters, summers, monsoon and post monsoon. The winter
season extends from the month of December to the month of February. Summers
include the months of March, April and May while monsoon extends from June to
September. The post-monsoon season includes the months of October and November.
For the research study, change in climate of Delhi is assessed only for the tempera-
ture and precipitation variables using 115 years of meteorological data (from 1901
to 2016), for better examination of natural and human induced climate change.
40 M. Agrawal
3.1 Temperature Variability
The variability of annual temperature for Delhi is assessed in terms of its annual
average temperature, annual average maximum temperature and annualaverage min-
imum temperature. Assessing the average annual temperature of the city from 1901
to 2016 (as indicated in Fig. 3), it is inferred that the city has experienced a 0.95 °C
rise in temperature, of which 0.2 °C was experienced post-1986, which marks an era
of economic liberalization and increased construction activities.
The seasonal temperature variability for Delhi is explored in terms of its annual
average temperature variation post 1901 for the four seasons of Delhi, which includes
winters, summers, monsoon and post-monsoon (as indicated in Fig. 4).
Assessing the average annual temperature for the four seasons from 1901 to 2016,
it is observed that the temperature for winter, summer, monsoon and post monsoon
seasons have increased by 1.1, 1.5, 0.8 and 1.3 °C, respectively. Also, the seasonal
annual average temperature trend indicates that the seasonal temperatures are rising
but summer temperature increase is twice that of monsoon increase.
3.2 Precipitation Variability
The annual precipitation variability is assessed in terms of annual rainfall and annual
number of rainy days for a time frame of 115 years, from 1901 to 2016. The trend
of annual precipitation post-1901 (as indicated in Fig. 5) indicates that the average
rainfall has increased by 210 mm and that the periods of drought have become longer
than periods of heavy rain.
Fig. 3 Change in Delhi’s annual average temperature, 1901–2016. Source IMD (2016)
3 Data Protection Law and City Planning: Role of Open Data … 41
Fig. 4 Seasonal temperature variation for Delhi, 1901–2016. Source IMD (2016)
Fig. 5 Change in annual precipitation for Delhi, 1901–2016. Source IMD (2016)
Assessing the trend in the number of rainy days for Delhi (as indicated in Fig. 6),
in the same time period, indicates that the average number of annual rainy days has
increase by 9 rainy days while the average precipitation per rainy day has increased
by 2.5%. Since the annual precipitation and number of rainy days are increasing and
given that the actual duration of precipitation has reduced, this results in a sharp rise
in rainfall intensity from 13.2 mm/h in 1986 to 22.9 mm/h in 2016 (the latter leading
to inundation of over 50% of city in 2016 in three hours).
The seasonal precipitation variability is assessed in terms of seasonal share of
annual precipitation and rainy days for the timeframe 1901–2016. Analysis of the
seasonal share of annual precipitation (as indicated in Fig. 7) indicates a trend of
wetter summers and drier post monsoons. Rainfall and rainy days are increasing but
the actual duration of precipitation is reducing leading to increase in rainfall intensity
42 M. Agrawal
Fig. 6 Change in annual number of rainy days for Delhi, 1901–2016. Source IMD (2016)
Fig. 7 Seasonal share of annual precipitation for Delhi, 1901–2016. Source IMD (2016)
from 13.2 mm/h in 1986 to 22.9 mm/h in 2016. In 2016, three hours of rainfall at this
intensity flooded over 50% of the city, breaking down city’s mobility and livelihoods.
4 Impact of Urban Development on Climate Variability
and Drainage of Delhi
4.1 Impact of Built Environment on GHG Emissions
The increase in GHG emissions for Delhi have been assessed at two levels. First,
a spatial distribution of GHG emitters has been identified which included built up
3 Data Protection Law and City Planning: Role of Open Data … 43
area as well as wasteland. Second, the sectoral contribution of GHG emissions from
the sectors of waste, transport, domestic and industries is estimated using Tier II
methodology formulated by the Intergovernmental panel on Climate Change (IPCC
2006).
In the first case, imageries indicate an increase in the total area of greenhouse gas
emitters, which has a direct correlation to the developed area densities. Empirical
analysis of the same indicates that the city had 42.5% of its area under greenhouse
gas emitters in 1986 which increased to 56.8% in 2016. That is at an annual rate of
increase equivalent to 1.3%, with the result that the city’s emissions are increasing
rapidly. The increase is related to increases in densities of developed area.
In the second case, GHG emissions from the sectors of waste, domestic, industries
and transportation was calculated using the Tier II methodology (IPCC 2006). The
method utilizes emission factors for energy consumption in each sector. Based on
this, emissions for NCT of Delhi have been estimated as shown in Table 1. This table
shows that the city’s GHG emissions have increased 4.5 times since 1986. Moreover,
the increase has been over 12 times in case of transportation sector, 3.5 times for
domestic sector, three times for waste sector and 2.9 times for industrial sector.
Empirically, for every 100 ha (total 28,205 ha) increase in built-up area between
1986 and 2016, the area under vegetative heat sinks reduces by 94 ha and water
bodies deplete by 6 ha, leading to an increase of GHG emissions by 0.078 million
metric tonnes of CO2equivalent.
4.2 Impact of Built Environment on Surface Temperature
For NCT of Delhi, the land surface temperature has been modelled for each of the
four time-periods (as illustrated in Fig. 8). Empirical analysis of the land surface
temperature of the city indicates that the average city level surface temperature has
increased from 32.8 °C in 1986 to 35.9 °C in 2016. This change is equivalent to an
annual increase in surface temperature by 0.31%, which is 1.6 times the increase
in air temperature. Also, it is inferred that more area is getting affected by higher
temperature ranges while areas with lower temperatures, particularly in the city’s
periphery are gaining temperature, primarily due to conversion of heat sinks into
wasteland and barren land.
Summing up, the period 1986–2016 witnessed a loss of heat sinks by 292 km2
and a rise in surface temperature by 3.1 °C and air temperature by 0.2 °C. That is,
for every 100 ha of heat sinks lost to development, surface temperature of the city
increases by 0.01 °C which is 1.6 times the rise in air temperature of the city. The
figures and values are a result of GIS analysis of Landsat imageries available for the
period and subject to changes in the resolution of the raw data.
44 M. Agrawal
Tabl e 1 Sectoral contribution of GHG emissions in Delhi, 1986–2016
Year Waste sector Domestic sector Industrial Transportation NET GHG
EMIS-
SIONS (in
MMT)
Waste
generation
(in million
kg)
GHG
EMIS-
SIONS (in
MMT) (EF
=0.13)
Area (in
hectare)
Population
(in million)
GHG
EMIS-
SIONS (in
MMT) (EF
=0.06)
Energy con-
sumption
(in million
units)
GHG
EMIS-
SIONS (in
MMT) (EF
=0.15)
Fuel con-
sumption
(Petrol,
Diesel,
CNG) (in
‘000 MT)
GHG
EMIS-
SIONS (in
MMT) (EF
=0.09)
1986 2.8 0.36 54,600 5.8 3.51 1085 0.16 948 0.57 4.60
1996 4.3 0.56 65,640 8.6 5.20 1537 0.23 1718 1.03 7.02
2006 6.8 0.88 67,780 11.9 10.74 2518 0.38 2028 1.83 13.82
2016 8.4 1.09 70,520 13.6 12.25 3135 0.47 7693 6.92 20.73
Estimated by Author (2017) from GNCTD (2016) and IPCC (2007)
3 Data Protection Law and City Planning: Role of Open Data … 45
Fig. 8 Change in surface temperature of Delhi, 1986–2016. Extracted by Author (2017) from
USGS (1986, 1996, 2006 and 2016)
4.3 Impact of Built Environment on Surface Run off
The increasing development in drainage basins and the resulting loss of flood plains,
coupled with increase in impermeable surface, has led to an increase in surface run-
off from the city. Due to an interplay of urban development and natural climate
variability, in terms of rainfall intensity, the city’s surface run-off has increased from
211 million litres per day in 1986 to 622 in 2016 (as indicated in Table 2), which
is 2.9 times increase in the last 30 years. Also, it is observed that with the loss of
every 10 ha of green cover, the surface run off increases by 0.014 million litres per
day (MLD). Annually the surface run-off is increasing at 3.7% while the loss of heat
sinks is 1.4%. That is, surface run-off is increasing at a much faster rate than the loss
of permeable surfaces in the city.
Increasing surface run-off and impermeable surface along with increasing inten-
sity of rainfall has led to increases in the area inundated by precipitation in Delhi.
While the period witnessed an annual growth of surface run-off by 3.8%, it led to
increase in inundation by 2.5%. Moreover, for every increase in surface run-off by
1 MLD, the inundation increased by 85 ha, while the road length affected increased
by 68 m and vector borne diseases increased by 7.8%. The problem is aggravated by
extraction and increasing reliance on groundwater to meet the demand supply gap.
46 M. Agrawal
Tabl e 2 Surface run-off from Delhi, 1986–2016
Land cover 1986 1996 2006 2016
Area (in
km2)
Run off
Coeff
Surface run
off
(422 mm)
(in MLD)
Area (in
km2)
Surface run
off
(733 mm)
(in MLD)
Area (in
km2)
Surface run
off
(757 mm)
(in MLD)
Area (in
km2)
Surface run
off
(718 mm)
(in MLD)
Total built
up
499.45 0.6 126.461 614.0 270.037 783.6 415.230 869.5 437.01
Forests 176.8 0.2 11.191 178.1 19.582 172.1 26.056 176.2 18.98
Other greens 48.6 0.2 3.076 66.8 7.345 70.2 10.621 72.8 7.84
Wat e r
bodies
41.9 0.0 0.000 34.4 0.000 29.6 0.000 19.5 0.00
Agriculture
land
586.0 0.1 24.729 432.5 31.702 368.2 27.873 287.4 30.95
Transportation 82.0 0.9 29.413 96.0 59.813 118.0 75.927 148.0 90.32
Waste land 48.3 0.8 16.306 60.8 35.647 59.3 38.157 57.6 37.22
Tot a l 1483 211 1483 424 1601 594 1631 622
Extracted by Author (2017) from USGS (1986, 1996, 2006 and 2016)
3 Data Protection Law and City Planning: Role of Open Data … 47
5 Discussion
The paper highlights the relevance of open data sharing and their multiple uses
for municipalities, development authorities and other agencies and organizations
working in the field of public policy and decision making.
Modelling and forecasting for 2041 as the horizon year, the research emphasizes
on propagating a carrying capacity guided development, as a wake-up call to the
business as usual scenario. In the business as usual scenario, the assumption is that
status quo would be maintained till 2041 in terms of development pattern and growth
rate and its impacts on environment and climate. The analysis and correlation model
indicate that for a population of 28.7 million and the rate at which current urban
development exists, if continued unabated till 2041 will lead to a further rise in
surface temperature by 2.8 °C, subjecting Delhi to an average surface temperature
of 38.71 °C. Also, over 80% of the city would inundate and GHG emissions would
increase by 84.6 million metric tons (MMT) while heat sinks would be further lost by
15,600 ha. To sum it up, over 80% of the city would be covered with built impermeable
surface.
The research presented here highlights the interdependence of urban development,
climate change and the natural environment, as well as a multiplicity of implications
(as indicated in Fig. 9) arising from these interdependent phenomena. The need for
planners and cities to deal with them in the planning system becomes critical, with
cities being guzzlers of over 3/4 of overall resources and generators of about 3/4 of
waste and pollution, while accommodating a little over half of the global population.
Innovative thinking, planning principles and design within an appropriate framework
to set strategies and priorities will be of the essence.
The study emphasizes a need for a climate resilient urban development for Delhi
which means to start envisioning and planning the city according to its carrying
capacity. The city’s expanse and political as well as socio-economic importance has
led to its relentless growth in area, population, vehicles as well as pollution and
degradation of natural resources. For that reason, there is a need to put a break
on the increasing trend of city development. This requires strong mobilization of
political support for fruitful planning strategies and policies. An example is the recent
initiative of Clean India Mission, famously known as Swachh Bharat Abhiyan by
Prime Minister Narendra Modi. It has brought about a wave of behavioral change at
every level of governance across India and given sanitation a political priority at the
center. Thus, political nexus and push, as well as bureaucratic support play a binding
role in ensuring success of planners’ efforts.
One of the priority proposals to be rolled out with central support is to initiate
decongestion of the city, which can be supplemented by the upcoming Regional
Rail Transit System connecting Delhi to surrounding towns of the National Capital
Region. Delhi would continue to exist as an employment hub. However, a pressing
current need is to start containing the development of the city, create heat sinks at an
accelerated rate and redistribute population along more ecological principles. This
could take the shape of land use-transport integration, redistribution of population
48 M. Agrawal
Fig. 9 Relationship between urban development, environment and climate variability for National
Capital Territory of Delhi, 1986–2016. Model by Author (2017)
densities and opening up public space, earmarking aquifer and recharging zones for
no development, among others.
Enhancing the climate resilience of population and infrastructure becomes indis-
pensable to counteract the impacts which have arisen from years of past develop-
ments. Moreover, urban planning needs to widen its scope beyond the administrative
boundaries of NCT of Delhi and start working at the level of Delhi Metropolitan
region. This is particularly important for a climate resilient urban future. This would
encompass making it a mandatory provision for all spatial plans to have a chapter on
climate change and its implications on urban development. In particular, it should
become a statutory requirement for the urban planning processes and planning doc-
uments to have a chapter with explicit mention of, and focus on climate change and
its relation to urban development in Delhi. It is also proposed that any spatial plan
shall have a chapter on climate change and policies for climate resilience, before it
can be approved or notified in the official Gazette. The master plan document would
have to elaborate climate strategies at city level and provide details at spatial level
as well.
It is long overdue that planners start looking beyond the jurisdiction of the National
Capital Territory of Delhi and start working and assessing climate and its relation to
urban development for a region beyond the state boundary. That is, the urban planning
3 Data Protection Law and City Planning: Role of Open Data … 49
jurisdiction should extend to the Delhi Metropolitan region. This recommendation is
further supported by the fact that the predominant climate of the city is determined
within 60 km in radius of the city.
Apart from spatial development strategies including transit-oriented development,
redistribution of population and densities, protection and conservation of the city’s
drainage pattern, recharging the ground water aquifer and enhancing the green infras-
tructure, certain other spatial development and planning strategies should be com-
pulsory as well.
First and foremost, the master plan of Delhi needs to include a comprehensive and
clear non-disputable policy for relocation and rehabilitation of climate vulnerable
population. Unambiguous provisions for the resettlement of population at risk of
climate change have to be included in writing in all spatial plans. Resettlements
within the same planning area have to be given priority. In case this is not possible
due to space constraints, the resettlement location must not exceed 5 km from the
original stay.
Another strategy of paramount importance relates to enhancing the climate
resilience for existing immovable infrastructure. There are three approaches to ensure
that. First, roads could be aligned according to high flood risk level, or put out of use
during the monsoon season. The second approach relates to the ‘asset management
approach’, whereby planners, engineers and professionals from other disciplines
would move from road design to planning and maintenance. That is, this approach is
a departure from a reactive patch-and-mend approach to a preventive management
approach. Lastly, it is necessary to opt for ‘user behavior management’, whereby sig-
nage will guide users to alternative routes which are less or not affected by climate
risk.
In this paper strategies for climate resilient urban development have been proposed
for the National Capital Territory of Delhi. New guidelines for climate resilient urban
development are also envisaged more generally for any megacity in India with similar
attributes and evidence to that of Delhi. They include: land use and urban planning
measures; planning for drainage including floods and solid waste management; man-
agement of water demand and conservation systems; building and enhancing resilient
housing and transport systems; and strengthening of ecosystem services. These five
categories of guidelines are directly related to spatial planning and development
strategies, that need to be included and comprehensively detailed in spatial planning
documents. Beyond that, another five categories of guidelines are proposed which
are more related to institutional capacities and multiple sectors, affected by climate
change and induced risks. They include: diversification and protection of livelihoods;
encouraging institutional coordination mechanisms; establishment and strengthening
of emergency and warning systems, improved technology and information systems,
and enhancing education and capacity building of citizens.
50 M. Agrawal
6 Scope for Further Research
In the ongoing research, correlations between urban development and climate vari-
ability at city, spatial and planning division level are explored. It also examines the
impact of climate variability on vulnerable populations and forecasts multidimen-
sional risks for horizon year of 2041. With the alternative scenarios, it is made clear
that Delhi needs to start implementing climate resilient urban development strategies
for a liveable urban future, at an accelerated pace.
However, there is much scope for further research in areas of wind and humidity
variables of climate. Also, we can do back casting of the correlation model and
strategies and assess how the city would have developed had we come to terms with
the fact that climate change is ubiquitous, a challenge to urban development and
requires climate resilient urban development strategies (for instance, in 1990s before
the second master plan was being revised or in early 2000s when the third plan
document was being drafted).
Thus, there exists a few important areas of research to further investigate. Research
is currently exploring few more dimensions through relationship models. Simulta-
neously, current research is studying relationships for other megacities, which apart
from population sizes and areas, exhibit completely different contexts to study the
relationships between climate resiliency and urban development.
References
Alankar (2015) India’s megacities and climate change: explorations from Delhi and Mumbai.
STEPS, Brighton
Albrecht J, Marcotullio P, Sarzynski A (2011) The geography of greenhouse gas emissions from
within urban areas of India. Resour Energy Dev 8(1):11–35
Asian Development Bank (2014) Urban climate change resilience—a synopsis. Asian Development
Bank, Metro Manila
Calthorpe P (2011) Urbanism in the age of climate change, 1st edn. Island Press, Washington DC
Cannon T, Mahn DM (2010) Vulnerability, resilience and development discourse in context of
climate change. Nat Hazards 55(3):621–635
Dodman D (2009) Blaming cities for climate change? An analysis of urban greenhouse gas emissions
inventories. Environ Urbanisation 21(1):185–201
GNCTD (2009a) Climate change agenda for Delhi, 2009–12. Government of National Capital
Territory of Delhi, New Delhi
GNCTD (2009b) Delhi climate change action plan. Government of National Capital Territory of
Delhi, New Delhi
Intergovernmental Panel on Climate Change (IPCC) (2003) In: Penman J, Gytarsky M, Hiraishi
T, Krug T, Kruger D, Pipatti R, Buendia L, Miwa K, Ngara T, Tanabe K, Wagner F (eds) Good
practice guidance for land use, land-use change and forestry. IPCC/IGES, Hayama, Japan
Intergovernmental Panel on Climate Change (IPCC) (2006) In: Eggleston S, Buendia L, Miwa
K, Ngara T, Tanabe K (eds) 2006 IPCC guidelines for national greenhouse gas inventories.
IPCC/IGES, Hayama, Japan
3 Data Protection Law and City Planning: Role of Open Data … 51
IPCC 2014: Climate Change (2014): synthesis report. contribution of working groups I, II and III
to the fifth assessment report of the intergovernmental panel on climate change [Core Writing
Team, Pachauri RK, Meyer LA (eds)]. IPCC, Geneva, Switzerland
Jabareen Y (2013) Planning the resilient city: concepts and strategies for coping with climate change
and environmental risk. Cities 31:220–229
Lisa E, Schipper F, Langston L (2015) A comparative overview of resilience measurement frame-
works—analyzing indicators and approaches. Overseas Development Institute, London
Padma T (2016) Will India’s ‘smart cities’ be climate-resilient? The Wire, 5 Feb
Phansalkar A (2016) Climate change and cities: it’s not too late to bring green buildings in policy
making. FirstPost, 29 Oct
Pratt SE (2012) Carbon and the city: tracking emissions from megacities. http://www.earthmagazine.
org/article/carbon-and-city-tracking-emissions-megacities. Accessed 16 Dec 2016 (Online)
Satterthwaite D, Bicknell J, Dodman D (2009) Adapting cities to climate change: understanding
and addressing the development challenges. Earthscan, London
The World Bank (2009) Climate resilient cities—a primer on reducing vulnerabilities to disasters.
The World Bank, Washington DC
UN-Habitat (2011) Global report on human settlements: cities and climate change. UN-Habitat,
Nairobi, Kenya
United Nations Framework Conventionon Climate Change (2015) Adoption of the Paris agreement,
21st conference of the parties. United Nations, Paris
Chapter 4
Exploring Shared-Bike Travel Patterns
Using Big Data: Evidence in Chicago
and Budapest
Ali Soltani, Tamás Mátrai, Rosalia Camporeale and Andrew Allan
Abstract Bike-sharing systems are an emerging form of sharing-mobility in many
cities worldwide. The travel patterns of users that take advantage of smart devices to
ride a shared-bicycle in two large cities (Chicago and Budapest) have been investi-
gated, with analysis of approximately two million transaction data records associated
with bike trips made over a three-month period in each location. Several aspects of
user travel behavior—such as day and time of travel, frequency of usage, duration of
usage, seasonal and peak/off-peak variations, major origin/destinations—have been
included in this analysis. The results show that in both cities the bike-sharing option
is a male-dominated alternative, particularly welcomed by younger groups, with the
largest share of trips occurring in the afternoon peak. Appropriate usage of open-
source big-data provides important lessons for successful vehicle sharing models,
allowing the application of the findings to other cities and mobility options where
these systems are still developing.
Keywords Bike-sharing systems ·Big data ·User travel behavior ·Mobility
A. Soltani (B)
School of Art, Architecture and Design City West Campus, University of South Australia, P.O.
Box 2471, Adelaide, SA, Australia
e-mail: Ali.Soltani@unisa.edu.au
Shiraz University, Shiraz, Iran
T. Má trai
Department of Transport Technology and Economics, Budapest University of Technology and
Economics, M˝uegyetem rkp, 3, Budapest 1111, Hungary
e-mail: tamas.matrai@mail.bme.hu
R. Camporeale
Department of Technology and Society, Lund University, P.O. Box 118, 211 00 Lund, Sweden
e-mail: rosalia.camporeale@tft.lth.se
A. Allan
Urban and Regional Planning, School of Art, Architecture and Design City West Campus,
University of South Australia, P.O. Box 2471, Adelaide, SA, Australia
e-mail: Andrew.allan@unisa.edu.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3-030-19424- 6_4
53
54 A.Soltanietal.
1 Introduction
Being a type of non-motorized transport, a Bike-Sharing System (BSS) is a travel
alternative that brings significant benefits to its users and society: overall, it has poten-
tial impacts in reduction of car use, positive effects on the environment, and health
benefits (Shaheen et al. 2010). As the urban population of the world is increasing,
the congestion in cities both on its roads and its public transport system will increase
(Rudolph and Mátrai 2018; Saif et al. 2018), thereby providing an opportunity for
BSS to emerge as a viable, sensible, healthy and environmentally sustainable trans-
port option.
In essence, a BSS can be defined as ‘shared use of a bicycle fleet’ (Shaheen et al.
2010). It provides shared bicycles that allow cyclists to travel from their origin to
their destination. Users are able to pick-up and drop-off bikes between different
self-service docking stations within a short-term rental time Fishman (2016). This
definition could be extended to the newly emerging type of dockless BSS, where the
bikes can be picked up and dropped off at any location within a predefined service
area, which is similar in concept to a free-floating car sharing system.
Several distinct features are needed to provide a seamless BSS experience which
includes: suitable provision of bicycles between places (i.e. via docking stations);
a technology application used for system management; appropriate and practical
rental duration; and an efficient, affordable payments method. Technological appli-
cations allow BSS operators to keep track both of the docking station status (i.e.
available bicycles and racks) and users’ movements through the network (Fishman
et al. 2013). In terms of the pricing and payment method, the pricing models can
vary considerably. In general, the aim is to maximize the utilization rate of vehicles,
with services normally free for the first 30 or 60 min (DeMaio 2009; Vogel 2016).
Users can use credit cards to pay additional fees that arise for further time usage of
the system. Although BSS operation principles might seem quite simple, they have
passed through a long development process before reaching their current state as a
comprehensive system.
BSS performances usually can be assessed by different indicators. One of them is
the usage rate, probably the most common metric used for evaluating performances
of BSS in different cities. User frequency is another equally important indicator,
often mentioned in studies by different scholars such as Rojas-Rueda et al. (2011),
Buck et al. (2013), Fishman et al. (2014), Médard de Chardon and Caruso (2015),
Fishman (2016), Saif et al. (2018) and Soltani et al. (2019).
The scope of this study is limited to exploring the travel pattern characteristics
of bike-sharing users. The BSSs investigated in this chapter—Divvy, in Chicago
and MOL Bubi, in Budapest—belong to the 4th generation of bike-sharing (Mátrai
and Tóth 2016). The demographic features (i.e. gender and age), of users and their
membership type have been also discussed.
The main objectives of this research can be summarized as follows:
•Showing the capability of open-source big data in discovering and analyzing travel
behavior at a large-scale;
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence … 55
•Discovering the overall pattern of bike-sharing usage in two cities (Chicago and
Budapest) where those systems have proven to be successful;
•Learning from the experience of well-established systems to improve and develop
sharing-mobility schemes in developing cities.
In the following sections, the two selected case studies have been described and
compared. The analyzed user travel behaviors include trip frequency and duration;
day and time of traveling; and seasonal and peak/off-peak time variations of trips.
Ordinary Least Square (OLS) regression has been used to explore the casual associ-
ations between trip duration and several different factors (such as age, gender, time
of travel, etc.). Conclusions and recommendations for practitioners conclude this
chapter.
2 Dataset and Preprocessing
This section contains a short introduction of the two selected BSS; furthermore, it
provides information about the data cleaning process that we have performed before
carrying out our analysis.
2.1 Case Study Selection
As only a limited number of systems publish their data frequently, the possible
choice set for comparison is quite limited. The main selection criteria are usually
the data availability, the authors’ familiarity of the selected systems and their main
characteristics (e.g. number of stations, number of bikes, average usage). On the other
hand, it can also be meaningful to compare two systems with some distinct features,
or within a different context. Due to the above-mentioned reasons, more specifically,
in this study we have selected a European bike-sharing system, in Budapest, Hungary
and an American case study of bike sharing in Chicago.
Political circumstances have strongly affected the street patterns of American
cities and towns. Morris (1994) explained that the law required that the territory to be
laid out with a rectangular grid within which townships were arranged in advance of
land being sold, with the result that the orthogonal section boundaries of land parcels
provided naturally straight road alignments. Chicago is a clear example of this: it has
a special orthogonal grid pattern, with streets that are straight and continuous.
In comparison, European countries have different forms of urban settlements: each
city has its own characteristic pattern, developed over the years with their history,
culture, population, and so on. Budapest has a radial street layout pattern, divided
by the Danube that enters the city from the north. Rode et al. (2017) provides a
comprehensive explanation of the close links between transport and urban form in
the US context.
56 A.Soltanietal.
In comparing the bike-sharing programs of such widely different cultural settings,
we expect to highlight similarities and differences, whilst at the same time setting
both European and American benchmarks for future studies that will follow our line
of research.
2.2 Description of the Divvy Bike-Sharing Program
(Chicago)
Chicago has a successful story to tell in running its bike-sharing program, ‘Divvy’.
Chicago covers an area of 600 km2and sits 176 m above sea level, on the southwestern
shore of Lake Michigan. The city is traversed by the Chicago and Calumet Rivers.
Chicago’s extensive parklands (about 3000 ha) attract an estimated 86 million visitors
annually. Chicago is also recognized across the United States as a very passionate
sports town.
‘Divvy’ is a BSS serving the City of Chicago and two adjacent suburbs and
is operated by Motivate for the Chicago Department of Transportation. The name
“Divvy” is a playful reference to sharing (“divvy it up”). Divvy’s light-blue color
palette and four stars evoke the Chicago flag.
Currently, using a bike for a single trip costs $US3, while if the user takes it for
a day, the price will be $US15. The fare for annual membership is $US99. The first
30 min of each ride are included in the membership or pass price. However, if the
user keeps a bike out of any docking station for longer, extra fees are payable. The
payment is by credit card, and no up-front payment is required to rent a bike.
Once a user joins the system for the first time, the actual use of the bicycle is
straightforward: after paying the fees as a new customer, he/she is given a ride code
or can use the member key to unlock the bike. Users are allowed to use the bicycle
for as many short rides as they want, within the time window they have paid for.
The bicycle can be dropped off at any Divvy station with empty racks; the system
map, displayed on the Divvy app, shows in real-time the station locations, with the
available bikes and racks for each of them. The service provider expects the users
to respect and obey any common traffic rules, such as riding with vehicular traffic
while keeping a safe distance, obeying traffic signals, not riding on sidewalks and
giving way to pedestrians.
2.3 Description of the MOL Bubi Bike-Sharing System
(Budapest)
MOL Bubi is the bike-sharing system operating in Budapest. Budapest has a popu-
lation of 1.75 million inhabitants; however around 3 million citizens live within the
larger metropolitan area. Approximately 800,000 people commute into Budapest on
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence … 57
a daily basis. Budapest is the political, cultural and commercial center of the country
and the city is a popular tourist destination (4.2 million tourists stayed in Budapest in
2017). It is situated on both sides of the Danube River, covering an area of 525 km2.
One side of the city is flat and cyclist-friendly, while the other is hilly.
MOL Bubi is the public bike sharing system of Budapest, owned and operated by
the city. The name MOL relates to sponsorship by a local Hungarian petrol company,
while Bubi refers to ‘Budapest Bike’. It launched in 2014 as a new form of ‘public’
transport. The main objective at the time was to encourage more and more passengers
to opt for cycling when reaching their destinations within a short distance in Budapest.
The system was not designed to provide an alternative solution for mass public
transport, but rather to provide an extension to it.
The MOL Bubi system guarantees access for bikes via different tickets and passes,
which are valid for periods from $US2.5 for 24 h up to $US58 for 365 days. Three
short-time access ticket options are available to hirers: 24 h; 72 h; or 7 days. When
buying a ticket, a refundable deposit of $US120 per bike is charged against the
hirer’s bank account. Tickets can be purchased by credit card from the touchscreen
terminals at the docking stations, or alternatively on the molbubi.bkk.hu website, or
via a mobile phone application.
There are also long-term access options, with the MOL Bubi available with options
of 3 months, 6 months or 12 months. A pass can be purchased at any Budapesti
Közlekedési Központ (BKK) customer service center, on the website, or via the
mobile app. Although it does not require a deposit, after the purchase, a customer
service center needs to be visited in person in order to finalize the procedure.
For hirers in possession of a valid ticket or pass, the system can be used free of
charge for up to 30 min per trip. If the bike is not parked in any docking station within
30 min, additional usage fees are applied, with a progressive increase according to
the actual minutes of usage.
One ticket allows a hirer to unlock one bike, but a single user can buy up to
four tickets at one time, with the deposit amount charged for each bike. The same
arrangement also applies to MOL Bubi passes, where up to four bikes connected to
a registrant (i.e. purchased pass) can be used by up to four users at the same time.
2.4 Available Data and Cleaning Process
The (secondary) data that was used to conduct our travel behavior analysis, with
regard to Chicago, were collected between the 1st of April 2018 and the 30th of June
2018, (i.e. over a period of three months). This was possible since the website of the
Divvy company provides access to open-source big-data of all of its user transactions.
Although these big data records are immediately available on the website, before
using them appropriately, two cleaning actions were applied. Firstly, those trips with
a duration of only two minutes or less (roughly one-third of a mile), were removed.
The assumption was made that these short rides may include records where Divvy’s
bikes are updated, or where a user reserved a bike and then changed his/her mind,
58 A.Soltanietal.
deciding to cancel the booking (customers are allowed a 30 min gap between booking
and beginning a trip). Secondly, those bike trips with a very long-time duration (over
24 h) have been removed from the data set, because these cases are most likely due
to either bike maintenance, cleaning, or replacing.
The data cleaning process over the chosen time window resulted in a 0.5% decrease
in the number of selected cases. The final number of trips that have been considered
for the Divvy BSS is equal to 1,048,574 in total.
The MOL Bubi data used in this study was provided directly by BKK for the
sole purpose of this analysis, during the same three months (April to June 2018).
BKK follows a general open data policy, where most of their data are available for
research purposes. This dataset contains the same information as the one from Divvy,
in Chicago. The only main difference relates to demographic characteristics, which
are those of the registrant, not of the actual user. Since several bikes can be connected
to a single pass, up to four individuals can be associated with a single registration.
This makes the assessment related to gender and age irrelevant as the real users
cannot be identified.
The MOL Bubi data was cleaned as well, in order to provide appropriate results.
Even in this case, the cleaning process has been based on the trip duration. All trips
shorter than two minutes and longer than 120 min were excluded. Very short trips are
usually due to some problems with the bike, indicating that the bike may have been
returned almost immediately. Excessively long periods of hire are usually linked to
problems with the docking procedure. This process resulted in the sample size being
reduced from 190,059 cases to 180,726, over the same 3-month time window, which
corresponds roughly to a 5% decrease. The difference between the upper limit arises
from the difference between the data collection and data structure.
3 Comparison of the Two Selected Systems
3.1 Descriptive Statistics
Divvy owns 6500 bikes registered for its sharing program, and data from 6400 cases
were recorded on its data file. The shared bicycles are distributed across 577 docking
stations (see Fig. 1); on average, according to the collected travel behavior data,
each station had been used 606 times in one month, with an average riding time of
24.2 min. The average usage of bikes was 1.8 hirings/day/bike.
The MOL Bubi system in November 2018, operated with 126 docking stations
(see Fig. 1) and 1600 bikes. All stations were functional during the analyzed
3-month period, which means that on average each station had been used 478 times
in a month, with an average trip duration of 14.6 min. The average usage of bikes
was 1.3 hirings/day/bike. This proves that not only is Divvy a considerably larger
system, but that it is more heavily utilized than MOL Bubi.
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence … 59
Fig. 1 Location of divvy stations in Chicago (left) and MOL-Bubi stations in Budapest (right)
[data source Divvy (2018), BKK (2018)]
In both cities, most of the trips were undertaken during weekdays (76.4% in
Chicago versus 74.1% in Budapest), while the remaining one-fourth of trips occurred
on weekends. The start and end time of each ride shows that shared bikes were mainly
used for social trips, and less for regular commuting to work. Based on that, in both
cities, the PM trips have resulted in twice as frequent use when compared with the
AM trips.
According to Fig. 2, June was the most popular month for bike users in Chicago;
an increasing trend can be noticed from April to June, most probably explained by
better weather conditions.
According to historical weather data, in May and June very similar meteorological
conditions suffered a slight decline in June, suggesting that in this city the main user
group is less likely to be tourists, but with regular users—since the holiday season
in Budapest starts in June.
In both cities, the start time distribution (Fig. 3) shows an AM peak hour between
8:00 and 9:00 AM, probably for work purposes. The afternoon peak hour is between
17:00 and 18:00 PM. However, the share of trips during the PM peak hour is about
1.5 times higher, showing the capability of shared bicycles in servicing the afternoon
and evening trip demand, mostly linked to social/recreation and exercise activities.
Furthermore, the evening usage of the bikes is as high as the morning peak. This
indicates that the main purpose of the trips is recreational.
In the Divvy system, two types of usage were reported: subscribers, as defined
by those who had already joined as a member (834,295); and customers (214,282),
60 A.Soltanietal.
Fig. 2 Monthly distribution of trips [data source Divvy (2018), BKK (2018)]
who are defined as tourists or casual users without an interest in joining as a member.
This implies that one-fifth of the users were non-subscribers.
The MOL Bubi database provides the registered user IDs in an anonymized way:
3782 passes and 5089 tickets were used during the selected study period; and 82%
of the trips were made with a subscription. It can be stated that the average usage
among ticket users declined over the 3-month period. On average, 5 trips per day
were completed using a 24 h ticket, while only 1.9 trips were done with a weekly
ticket.
In Chicago, the gender composition of BSS users corresponded to 20.9% females,
60.8% males and 18.3% not declaring their gender. Females preferred to be a sub-
scriber rather than become short term users: for example, over 42% of subscribers
were female, but only 6% of the total number of customers were female.
In Budapest, as anticipated previously, the gender and age distributions for pass
users are not relevant; furthermore, the gender and age are not compulsory infor-
mation to fill in for ticket users. Consequently, only 1824 ticket users provided this
information out of 3288.
It can be stated that the main age group for short-term bike sharing usage in
Budapest was between 19 and 39 years old. In Chicago, the youngest bike customer
was born in 2003 and the average age of users was 34.5 years, while the two main
age groups using shared bikes were 25–29 years old (27.7%) and 30–34 years old
(22.5%), accounting for approximately 50% of the total number of bike users. They
were followed by the 35–39 years old age group (13.5%) and the 20–24 years old
age cohort (10.8%). Two very minor groups were those aged 17–19 years old (1%)
and those aged 70 years old and over (1.3%).
62 A.Soltanietal.
Fig. 4 Daily variation in shared bike usage [data source Divvy (2018), BKK (2018)]
Regarding the distribution of bike trips in Chicago across weekdays and weekends,
Wednesdays had the highest frequency, followed by Thursdays. The lowest share
was on Sundays, showing a lower overall demand for bikes during the weekends.
However, this pattern can vary according to the considered month, as shown in Fig. 4.
For example, in April 2018, Monday was the most popular day, while Saturday was
the least popular day. On May 2018, Tuesday became the most popular day for bike
users, whilst Sunday was the least common. On June 2018, a relatively equal share
for all the days of the week had been registered. This daily/monthly variation can be
mainly explained by variations in weather conditions relating to rainfall, temperature,
humidity, and wind.
The distribution of bike trips across different days in Budapest shows a generally
equal split pattern. However, the weekends have a somewhat lower share, which
appears to reflect the fact that BSS was mainly used by locals.
Looking at Fig. 5, the highest shares of trips in both cities coincide basically
with two time-windows: the PM peak (39% in Budapest and 34% in Chicago); and
midday (30% in Budapest and 32% in Chicago). In both cases, the lowest share was
in the early morning; however, data showed that the MOL Bubi system was better
integrated with the nightlife in Budapest, considering that more than 30% of trips
had been registered in this period of the day. Since more than one bike can be rented
by the same user at the same time in Budapest, this suggests that the usage of BSS is
a social activity. In the 3-month analyzed period, 27% of the trips were made with
somebody else using the same account (i.e. more than one rental in the same time),
26% originated from the same station using the same account, while 22% have the
same origin and destination using the same account, thereby indicating a common
trip.
Another interesting finding from the data in relates to the most popular origin
and destination stations. The 10 major stations (for both origin and destination) were
detailed in Tables 1and 2. These stations alone account for 12.8% of origins and
12.4% of destinations and crucial to trip generation of BSS users in Chicago.
The Budapest system is smaller therefore the top 10 stations produce 20.2%
of origins and 20.4% of destinations. Further investigation about the function and
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence … 63
Fig. 5 Peak and off-peak variation in bike usage [data source Divvy (2018), BKK (2018)]
character of the adjacent areas of these stops may show more clearly the reasons
for their popularity, together with the effects of built environment factors on the trip
generation processes.
3.2 Regression Analysis
Different statistical tests and regression analysis have been applied to the Chicago
dataset. The age group proved to be an effective factor in describing the duration of
abiketrip.
The results of the Analysis of Variance (ANOVA) show that there is a significant
difference among age groups: F =269.577, p≤0.000 in terms of duration of a bike
ride.
Furthermore, the correlation between age and trip duration is negative (−0.0019),
demonstrating that younger age groups make longer trips (n =856,314; p< 0.10).
While the usage of shared bikes is significantly different among male and female
(t =518.217; sig =0.000) BBS users, there is a statistical difference among these
two groups in terms of usage duration, as shown by the t-test (t =6.861, sig =
0.000). The average trip time for males and females determined to be 836 and 1053 s
respectively. This has been confirmed by conducting a correlation test, which pro-
vides the correlation value (rho =−0.0074) between being male (gender dummy)
and trip duration. Therefore, it is possible to conclude that females take longer trips
than males (n =856,313; p< 0.05).
Aiming at determining the factors that affect trip duration, the ordinary least
square (OLS) regression test has been applied. The results have shown that the most
significant factors are: age; gender; user type; day of travel; and time of travel. The
model is linear based on the ANOVA result (F =28,052.9; p< 0.000), with a goodness
of fit measure of 27.8% (adjusted R-Square). The gender dummy variable (male) has
64 A.Soltanietal.
Tabl e 1 10 main stations for both origin and destination in Budapest
No. Station
address
Frequency
of origin
Frequency
of
destination
Rank of
origin
Rank
of
desti-
nation
1301 Jászai Mari
tér
PT
connection
to the
recreational
area
4952 4645 1 2
1304 Margitsziget Main
recreation
area
4669 4753 2 1
1101 Szent
Gellért tér
University 4267 4136 3 3
0905 Kálvin tér University,
downtown
3737 3792 4 4
0515 F˝ovám tér University,
downtown
3640 3735 5 5
0101 Batthyány
tér
Main PT
hub
3242 3621 6 6
0508 Erzsébet tér Downtown 3159 3177 7 8
0517 Városháza
Park
Downtown 3026 2804 813
0607 Oktogon Downtown 2949 2886 910
0611 Nyugati tér Shopping
center main
PT hub,
main train
station
2928 3236 10 7
0802 Astoria Downtown 2771 2833 11 12
0103 Clark Ádám
tér
Touristic
attraction
2653 2891 12 9
Data source BKK (2018)
shown a negative association (t =−61.136, p< 0.000), confirming that females on
average make longer trips, as discussed earlier.
The age factor plays an interesting role, in that for the age group of 25–29-year-old
BSS users, the trip duration increases if the age value increases. By contrast, for the
age group of 30–34 years old (t =5.185, p< 0.000), there was a negative association
between the age and trip duration (t =−1.852, p< 0.10). Furthermore, the user type
‘subscriber’ showed a negative association (t =−295.921, p< 0.000), confirming
that casual users make longer trips than members.
With regard to the day and time of travel, trips over the weekend were positively
associated with trip duration (t =12.736, p< 0.000), but those trips made during peak
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence … 65
Tabl e 2 10 main stations for both origin and destination in Chicago
No. Station
address
Frequency
of origin
Frequency of
destination
Rank of
origin
Rank
of
desti-
nation
35 Streeter Dr
and Grand
Ave
Park,
shopping,
restaurants,
waterfront
25,587 21,108 1 1
192 Canal St and
Adams St
Metro,
restaurants,
banks
17,158 17,724 2 2
91 Clinton St
and
Washington
Blvd
Metro,
working area
12,767 13,836 4 3
77 Clinton St
and Madison
St
Restaurants,
banking
14,014 13,200 3 4
76 Lake Shore
Dr and
Monroe St
Waterfront,
park
10,212 12,779 9 5
43 Michigan
Ave and
Washington
St
Park,
cultural,
sport
11,064 10,788 8 6
90 Millennium
Park
Park,
waterfront
11,457 10,424 7 7
177 Theatre on
the Lake
Park,
waterfront,
cultural
11,474 10,276 6 9
81 Daley
Center Plaza
Shopping 10,182 9996 10 10
133 Kingsbury
St and
Kinzie St
Wor k o ffic e
area,
warehouse
9888 9864 11 11
Data source DIIVY (2018)
PM hours have a shorter trip duration (t =−3.755, p< 0.000). The last factor that
affects trip duration is the selection of the same station as the origin and destination
of the trips (t =45.114, p< 0.000) (Tables 3and 4).
66 A.Soltanietal.
Tabl e 3 Descriptive statistics of the regression model parameters
NMinimum Maximum Mean Std. dev
Age 1,048,573 0119 28.97 16.816
Trip duration 1,048,573 61 13,453,200 1452.50 32,884.238
Start time 1,048,573 0:00 23:59 14:19 4:40
End time 1,048,573 0:00 23:59 14:34 4:44
PM (dummy) 1,048,573 0 1 0.68 0.465
Early (dummy) 32,724 111.00 0.000
AM peak (dummy) 1,048,573 0 1 0.16 0.363
Midday (dummy) 849,346 0 1 0.40 0.489
PM peak (dummy) 511,276 0 1 0.70 0.460
Late (dummy) 155,445 111.00 0.000
Data source DIIVY (2018)
Tabl e 4 Coefficients of the regression model parameters
Unstandardized
coefficients
Standardized
coeffi-
cients
t-statistics Significance
(p-value)
BStd. error Beta
Constant 7.478 0.003 2520.964 0.000
Gender male – 0.157 0.003 −0.089 −61.136 0.000
Age (dummy) =2.0 0.013 0.002 0.007 5.185 0.000
Age (dummy) =3.0 −0.006 0.003 −0.002 −1.852 0.064
Subscriber (dummy) −0.989 0.003 −0.458 −295.921 0.000
Weekend 0.033 0.003 0.015 12.736 0.000
PM peak (dummy) −0.008 0.002 −0.004 −3.755 0.000
Same_Origin_Destination 0.244 0.005 0.056 45.114 0.000
Data source DIIVY (2018)
aDependent variable: trip duration Ln
4 Conclusion and Recommendations
This chapter examined two bike-sharing systems of Chicago and Budapest. It aimed
at demonstrating how big-data collected by smart-device transactions can be used to
profile the usage of such systems.
Although shared bicycles may be a minor modal alternative in the overall trans-
portation portfolio, they can offer a significant efficiency in providing supply to
job-related, social and recreational short trips. BSSs can be an attractive option not
only for those individuals residing in a certain city but also for regular commuters
and tourists.
4 Exploring Shared-Bike Travel Patterns Using Big Data: Evidence … 67
Their importance is also crucial when dealing with the connections to/from major
trip generators (i.e. public transport hubs), which in turn can reduce the demand for
motorized transport with its adverse traffic and environmental consequences.
However, some BSS operations such as OfO and OBike have experienced ‘market
failure’, and this has been due to a variety of reasons (i.e. bike vandalism, inappro-
priate bike maintenance, insufficient availability in certain locations or at certain
times, personal concern of users about sharing their credit card details, availabil-
ity of mandatory safety helmets, etc.). The usage of relevant open-source data for
research purposes has been demonstrated to be a feasible strategy to gain a deeper
understanding of the functioning of those BSSs that have actually proved to be suc-
cessful.
This chapter is the first step in this research work. Our key recommendation is that
the stories of Divvy and MOL Bubi narrated by the data that we have collected and
interpreted can provide planners and practitioners with increased knowledge, which
would be useful to prevent future bike share failures, whilst building on the positive
outcomes to facilitate successes in other cities elsewhere in the world, particularly
where BSSs are now at a nascent or developing stage. The apparent male gender bias
favoring bikeshare in the case study results are perhaps explained by imbalances in
roadcraft experience, an aggressive road environment, greater familiarity in using
bikes and a tendency to cycle faster. However, this hypothesis can only be validated
when individual GPS based tracks are available. Further research is required to dis-
cover additional factors (i.e. related to land-use patterns, existing public transport
networks, bike-related facilities and bicycle infrastructure), that may affect the suc-
cess of shared bicycle systems. Moreover, a further avenue for research, is to examine
how BSS compares with rapidly emerging dockless bike systems.
References
BKK (2018) Budapesti Közlekedési Központ (Centre for Budapest Transport). https://bkk.hu/en/
welcome-to-bkk/. Accessed 30 Nov 2018
Buck D, Buehler R, Happ P et al (2013) Are bikeshare users different from regular cyclists? Transp
Res Rec J Transp Res Board 2387:112–119. https://doi.org/10.3141/2387-13
DeMaio P (2009) Bike-sharing: history, impacts, models of provision, and future. J Public Transp
12:41–56. https://doi.org/10.1016/0965-8564(93)90040-R
Divvy (2018) https://www.divvybikes.com/. Accessed 30 Nov 2018
Fishman E (2016) Bikeshare: a review of recent literature. Transp Rev 36:92–113. https://doi.org/
10.1080/01441647.2015.1033036
Fishman E, Washington S, Haworth N (2013) Bike share: a synthesis of the literature. Transp Rev
33:148–165. https://doi.org/10.1080/01441647.2013.775612
Fishman E, Washington S, Haworth N (2014) Bike share’s impact on car use: evidence from the
United States, Great Britain, and Australia. Transp Res Part D Transp Environ 31:13–20. https://
doi.org/10.1016/j.trd.2014.05.013
Mátrai T,Tóth J (2016) Comparative assessment of public bike sharing systems. TranspRes Procedia
14:2344–2351. https://doi.org/10.1016/j.trpro.2016.05.261
Médard de Chardon C, Caruso G (2015) Estimating bike-share trips using station level data. Transp
Res Part B Methodol 78:260–279. https://doi.org/10.1016/j.trb.2015.05.003
68 A.Soltanietal.
Morris AEJ (1994) History of urban form before the industrial revolutions, 3rd edn. Longman
Scientific & Technical, England
Rode P, Floater G, Thomopoulos N, Docherty J, Schwinger P, Mahendra A, Fang W (2017) Acces-
sibility in cities: transport and urban form. In: Disrupting mobility. Springer, Cham, pp. 239–273
Rojas-Rueda D, de Nazelle A, Tainio M, Nieuwenhuijsen MJ (2011) The health risks and benefits
of cycling in urban environments compared with car use: health impact assessment study. BMJ
343:d4521. https://doi.org/10.1136/bmj.d4521
Rudolph F, Mátrai T (2018) Congestion from a multimodal perspective. Period Polytech Transp
Eng 46:215–221. https://doi.org/10.3311/pptr.12048
Saif MA, Zefreh MM, Torok A (2018) Public transport accessibility: a literature review. Period
Polytech Transp Eng. https://doi.org/10.3311/pptr.12072
Shaheen S, Guzman S, Zhang H (2010) Bikesharing in Europe, the Americas, and Asia: past,
present, and future. In: TRB annual meeting. pp 1–20
Soltani A, Andrew A, Nguyen HA, Berry S (2019) Bikesharing experience in the city of Adelaide:
insight from a preliminary study. In: Case Studies on Transport Policy. https://doi.org/10.1016/j.
cstp.2019.01.001
Vogel P (2016) Service network design of bike sharing systems. Springer International Publishing
Switzerland. https://doi.org/10.1007/978-3-319-27735-6
Chapter 5
Can Social Media Play a Role in Urban
Planning? A Literature Review
Yanliu Lin and Stan Geertman
Abstract In recent years, the widespread use of social media has generated new and
big datasets and provided new platforms for urban planning. However, existing stud-
ies have often been case-specific or focused on a specific planning domain, leaving
the role of social media in urban planning generally questioned. This study conducts
a systematic review of to which extent social media can be used in urban planning.
There are two main findings. On the one hand, social media data have been increas-
ingly used for urban analysis and modelling, often combined with conventional and
new datasets. The domains of application include individual activity patterns, urban
land use, transportation behavior, and landscape. On the other hand, social media
have provided a new platform for participation, communication and collaboration.
They provide new opportunities for cities to hear the voices of distinctive social
groups, even those who do not formally participate in planning processes. In recent
years, citizens have used social media to initiate and organize themselves collective
actions in planning practices. Issues of using social media data in urban planning
include population and spatial biases, privacy issues, and difficulties in extracting
useful information out of the social media data. It is necessary to pay more attention
to the proper dealing with these issues during the collection and methodological
handling of social media data.
Keywords Social media ·Urban planning ·Data ·Urban analysis ·Participation
Y. L i n ( B)·S. Geertman
Department of Human Geography and Planning, Utrecht University, Vening Meinesz Building A,
Princetonlaan 8 A, 3584 CB Utrecht, The Netherlands
e-mail: y.lin@uu.nl
S. Geertman
e-mail: S.C.M.Geertman@uu.nl
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_5
69
70 Y. Lin and S. Geertman
1 Introduction
Social media have dramatically changed social relationships, offering opportuni-
ties for individuals to communicate and interact with a diverse group of people in
global and local networks (Lewis et al. 2010). The “Web 2.0” features make social
media enable information sharing, networking, and wider participation. Social media
usage is growing due to the improvement of internet accessibility and an increase
in smartphone users all over the world. As a result, a large number of volunteered
data have been generated by users, including the posting of comments, observations,
and the uploading of photos to social networking sites such as Facebook or Twitter
(Kitchin 2014). These data often possess location information which can be valuable
information for urban planning. They provide an addition to conventional data and
can be used for urban analysis that leads to a real-time understanding of processes
in urban space. Therefore, social media data can be considered as a resource for
evidence-based decision making and strategic management in urban planning.
In recent years, scholars have increasingly debated on the role of social media in
urban planning. Batty et al. (2012) argue that social media provide new and unconven-
tional dataset for planning smart cities. They explore the way in which community
networks can be generated through using social media data minded form mobile
device data bases and websites, and how these data can be linked with data on hous-
ing and labor markets. Jendryke et al. (2017) argue that social media can provide
information for understanding many urban topics, such as healthy issues, emergency
locations, and social activity hotspots. Silva et al. (2014) indicate that social media
data can be used for analyzing and clustering selected groups and activities in the
city, which eventually makes it possible to characterize urban areas distinctively. The
study of Kleinhans et al. (2015) shows that social media and mobile applications can
increase public participation, engagement, and communication in urban planning.
In practice, many local governments in Europe have actively participated in online
conversations by using Facebook and Twitter (CIVITAS Policy Note 2015), while
Chinese governments have used Chinese social media such as Weibo and WeChat to
interact and communicate with citizens (Lin 2018).
However, existing studies have often been case-specific or focused on a specific
planning domain, leaving the role of social media in urban planning generally ques-
tioned. To fill this gap, this study conducts a systematic review of the usefulness of
social media in urban planning. It finds that social media data have been increasingly
used for urban analysis and modelling in the domains of individual activity patterns,
urban land uses, transportation behaviours, ecosystems, and landscapes. Besides,
social media have provided a new platform for public participation, communication,
and collective actions. With the widespread use of social media, citizens now easily
establish large-scale online social networks and initiate collective actions themselves.
More research is required to understand these emerging forms of bottom-up plan-
ning. However, this study also identifies several issues regarding the use of social
media in urban planning, including population and spatial biases, privacy issues, and
5 Can Social Media Play a Role in Urban Planning? A Literature … 71
difficulties in extracting useful information. It is necessary to deal with these issues
during the collection and methodological handling of these social media data.
2 Social Media: Definition and Typology
According to Kaplan and Haenlein (2010, p. 61), “Social Media is a group of Inter-
net-based applications that build on the ideological and technological foundations
of Web 2.0, and that allow the creation and exchange of User Generated Content”.
Web 2.0 refers to a collection of electronic, Web-based applications and technologies
that facilitate interactive information sharing, user-centered design and collaboration.
Kaplan and Haenlein (2010) further provide a classification of social media: collabo-
rative projects (e.g., Wikipedia), content communities (e.g., YouTube, Flickr), social
networking sites (e.g., Twitter and Facebook), blogs, and virtual social and game
worlds.
Among them, social networking sites are applications that enable users to con-
nect by creating personal information profiles, which include the information of
photos, video, and blogs (Kaplan and Haenlein 2010). Users can invite friends and
colleagues to have access to those profiles and send instant messages between each
other. According to Statista (2018), the most popular social networking sites in the
world include Facebook, Youtube, WhatsApp, Facebook Messenger, WeChat, Insta-
gram, QQ, Qzone, Douyin, Sina Weibo and Twitter (Fig. 1). These leading social
networking sites are usually available in multiple languages and enable users to con-
nect with people across geographical borders. As the first social networking site,
Facebook surpasses 1 billion registered accounts and currently have 2.23 billion
monthly active users. About 2 billion internet users are using social networking sites
and these figures are still expected to grow, with an increase in mobile device usage
and mobile social networks. The 11th popular social networking site is Twitter on
which users post and interact with messages known as “tweets”. The data of Twitter
has been widely extracted and used in urban analysis.
3 Literature Review: Method and Resulting Corpus
The literature review performed consisted of three phases. Being aware of the mul-
tidisciplinary nature of the topic, the first phase sought to retrieve a broad set of
papers. To achieve this aim, an advance search query was performed on the ISI Web
of Knowledge, Scopus, ScienceDirect, and other databases. In each database, we
entered the two key words “social media” and “urban planning” and searched for
related papers. Besides, we collected relevant policy documents and other online
materials about the topic. After reviewing all the received papers and documents, we
selected 65 key papers related to social media and urban planning.
72 Y. Lin and S. Geertman
Fig. 1 Global social networking sites ranked by number of users in 2018 (in millions). https://
www.statista.com/statistics/272014/global-social-networks-ranked-by-number-of-users/
In the second phase, we imported the 65 articles into Nvivo for content analysis.
Nvivo is a data analysis computer software package to analyze text documents and
other resources. It allows users to classify and sort information, identify patterns,
and examine relations in the data. We used the function of the word cloud to ana-
lyze the word frequency in all the papers. In Fig. 2, we show the most important
terms: “social”, “media”, “data”, and “urban”. The other important terms include
“information”, “new”, “city”, “planning”, “people”, “public”, “users”, “analysis”
and “time”.
In the third phase, we looked in particular at how the mentioned key terms were
used in the received papers. We identified two main ways of using social media in
urban planning in the literature. On the one hand, social media data were used in
urban modeling and analysis, in the domains of individual activity patterns, urban
land uses, ecosystems and landscape. On the other hand, social media provided new
platforms for public participation, communication, and citizens’ activism in planning
practices. Most of the studies in this domain mainly focus on how social media is used
as a tool in supporting participation and communication in planning processes, while
some scholars have recently extracted social media data for analyzing information
transmission and actor relationships. In the following sections, we firstly introduce
the contents, characteristics, and types of social media data that can be used in urban
planning. We then discuss the use of social media data in urban analysis and modelling
in several planning domains. After that, we illustrate the application of social media
5 Can Social Media Play a Role in Urban Planning? A Literature … 73
Fig. 2 The word cloud of the keywords of publications (source authors 2018)
data in participation, communication and citizens’ activism. Finally, we discuss the
opportunities and challenges of using social media in urban planning.
4 Social Media Data
Social media data contain valuable information such as geo-location, time, and texts
of users. One feature of many social networking tools is that they allow users to create
microblogs and post short contents such as comments, images and video (Lewis et al.
2010). Researchers analyze the content or the semantic meaning of a message text
or a photo and use the geo-location of a message or a user for urban analysis. In most
cases, data from social media platforms can be retrieved through their public APIs at
low costs. “Any smart device or computer that is able to access the Twitter domain
can enable a geo-coding facility that locates where the message is sent from and
74 Y. Lin and S. Geertman
there is considerable research into how social networks as well as spatial networks
might be fashioned from such data” (Batty et al. 2012, p. 499).
Location-based services (LBS) in online social networks have provided an
unprecedented amount of public-generated data on human movements and activities.
They enable people to share their activity related choices (check-in) in their virtual
social networks. Through location-based services, users can share their activity-
locations when they visit restaurants, shopping malls, and so on (Rashidi et al. 2017).
Thereby, the human activity pattern can be revealed from the “check in” information
of places produced by users. While social media data is tremendously beneficial
in modeling individual activity patterns, it is also greatly useful in inferring plan-
ning related variables such as urban land use characteristics (Zhan et al. 2014). For
instance, Twitter and Flickr record the interactions between people and their sur-
rounding environment, especially the information about the behaviors of people in
geographic spaces. Geotagged Flickr photos possess a high suitability for exploring
urban areas of interest, because they reflect the interest of people towards locations
(Hu et al. 2015).
Previous research illustrates the broadness and depth of information that can be
extracted from social media. Social media analytics make it possible to measure
public sentiment and understand public opinions with real-time data mined from
Twitter, blogs and other social media platforms (CIVITAS Policy Note 2015). Text
analytics use natural language processing to spot key words and to gauge sentiment.
Besides, social networking sites such as Weibo can also show the followers and
forwarding messages. This information can be extracted and used for analyzing the
relations and communication of online participants (Zhao et al. 2018).
5 Urban Analysis and Modelling
5.1 Individual Activity Pattern
Social media check-in data, which contain users’ location information, have recently
been used to understand individual activity patterns in urban spaces. Zhi et al. (2016)
use the data from about 15 million social media check-in records during a year-long
period in Shanghai to identify a series of latent spatio-temporal activity structures
of the city. Using geotagged Twitter data, Shelton et al. (2015) analyze the everyday
activity spaces of different groups of Louisvillians in the United States, reflecting
that those neighborhoods were fluid and porous rather than rigid and static.
Nevertheless, social media data often lack the information of socioeconomic status
of individuals and detailed spatial information. As a consequence, social media data
are often combined with other datasets such as maps, images and other resources
in urban analysis and modeling. For instance, Huang and Wong (2016) combine
Twitter data with the American Community Survey (ACS) data to analyze the activity
patterns of Twitter users with different socioeconomic status. Their research shows
5 Can Social Media Play a Role in Urban Planning? A Literature … 75
that, while socioeconomic status is highly important, the urban spatial structure and
the geographical layout of the region plays a critical role in affecting the variation in
activity patterns between users from different communities. Jendryke et al. (2017)
link social media data with remote sensing imagery to enhance contextual urban
information. In their approach, remote sensing images are used to identify urban
built-up areas and changes within those areas, while geocoded mobile social media
messages deliver valuable information about human activity and the vitality found
in these areas.
5.2 Urban Land Use
Social media data are used also to reflect urban land use. Tu et al. (2017) uncoverurban
functions by aggregating human activities inferred from mobile phone positioning
and social media data. In their approach, the homes and workplaces of travelers are
estimated from mobile phone positioning data to annotate the activities conducted
at these locations. Chen et al. (2017) delineate urban functional areas based on
building-level social media data. They assess the diversity of urban functions at the
community level and identify several potential “central places” based on hot spot
analysis. The analysis provides an alternative way of characterizing intra-city urban
spatial structures and could inform future planning and policy evaluation.
Online point-of-interest (POI) data are also used to identify and estimate urban
land uses. There are currently hundreds of voluntarily generated POI directories
on the Web, such as Yahoo! and Facebook places. The POI data is a type of online
volunteered geographic information, e.g., a specific point location that a considerable
group of people find useful or interesting. Jiang et al. (2015) extract and classify the
POI data from a user-content platform (i.e. Yahoo!) and combine these new data with
census data, GIS data and proprietary business establishment data to disaggregate
the aggregate data to a finer level (Fig. 3). They also use infoUSA POI data, which
contains detailed information of business establishments in the United States, to
evaluate this newly developed method. They find that the disaggregated employment
estimations using these two different POI data sources are very similar. They argue
that this new approach using POI data from social media provides opportunities for
cities to estimate land use at high resolution with low cost while ensuring its quality
with a certain accuracy threshold.
5.3 Transportation
There is a growing body of literature on the application of social media data on
transport analysis and planning. To fully realize the idea of an urban mobility atlas
for the smart city, there is a need to integrate increasingly richer sources of mobility
data, including the data from public transportation, road sensors, surveys and official
76 Y. Lin and S. Geertman
Fig. 3 Disaggregated retail employment density at block level, using infoUSA (left) POIs and
Yahoo! (right) POIs (source Jiang et al. 2015)
statistics, participatory sensing and social media into coherently integrated databases
(Batty et al. 2012). Rashidi et al. (2017) conduct an overview of transport related
studies which used social media data for transportation planning and management.
Their study reflects how social media data from different sources can be used to indi-
rectly extract: (1) travel attributes, such as trip purpose, mode of transport, activity
duration and destination choice, (2) land use variables, such as home, job and school
location, and (3) socio-demographic attributes, such as gender, age and income.
They argue that social media data have been used to develop models for estimating
travel demand, managing operation, and long-term planning purpose. However, their
research also finds that though the cost of obtaining social media data is low, pro-
cessing such massive databases to extract travel information is a challenging task,
especially for attributes such as trip purpose.
5.4 Ecosystem and Landscape
Social media data including geo-tagged photos are becoming an increasingly attrac-
tive source of information about ecosystem and landscape. Landscape photographs
could tell us about the significance of human relationships with landscapes, human
5 Can Social Media Play a Role in Urban Planning? A Literature … 77
practices in landscapes and landscape features. Oteros-Rozas et al. (2018) develop a
methodological approach suitable for eliciting the importance of cultural ecosystem
services and the landscape features underpinning their provision across five different
countries in Europe. They perform a content analysis of 1404 photos uploaded in
Flickr and Panoramio platforms that can reflect cultural ecosystem services. They
find a positive though weak relationship between landscape diversity and cultural
ecosystem services. Tieskens et al. (2018) use social media photos from Flickr and
Panoramio to estimate the correlation between landscape attributes and landscape
preferences in the Netherlands. They indicate that social media data can be incor-
porated as evidence of what elements of landscape are valued, where people are
interacting with the landscape, and how these interactions characterize a landscape.
Zhang and Zhou (2018) use social media check-in data in Beijing to develop multi-
ple linear regressions to investigate how park attributes, locations, and contexts, and
public transportation affect the number of park check-in visits. Their study shows
that there are two effective measures for improving park use: (1) improving park
accessibility through public transportation, and (2) planning small and accessible
green spaces in residential areas.
6 Participation, Communication and Citizens’ Activism
6.1 Augmenting Public Participation
According to CIVITAS Policy Note (2015), social media has a decisive role to play
in motivating and empowering citizens, as well as in increasing engagement with
the voluntary groups and NGOs. Many scholars argue that social media and mobile
applications can increase public participation and community engagement in urban
planning (Kleinhans et al. 2015). Citizens are now keen on using social media tools to
interact with local governments in urban planning. The use of social media can reduce
a top-down information dissemination channel and can open up citizens’ activism.
Furthermore, social media appeals to younger generations of citizens (Schroeter
and Houghton 2011). Social media also provides opportunities for cities to hear the
voice of social groups who do not formally participate in urban planning and decision
processes. Therefore, Fredericks and Foth (2013) advocate the term of “augmenting
public participation”, i.e. capturing a wider audience of participants through the
use of social media and web 2.0 applications. Social media provides a new way of
supplementing traditional methods of public participation that are often face-to-face
and engage small groups of participants.
Resch et al. (2016) argue that public participation in urban planning can acquire
citizens’ ideas and feedbacks in participatory sensing approaches like “People as
Sensors”. They indicate that citizen-centric planning can be achieved by analyzing
Volunteered Geographic Information (VGI) data such as Twitter tweets and posts
from other social media channels. They analyze tweets in three dimensions (space,
78 Y. Lin and S. Geertman
time, and linguistics), and use a graph-based semi-supervised learning algorithm to
classify the data into discrete emotions. Their study shows that comments concerning
problems of urban environments such as traffic jams and pollution can be detected
in tweets. However, the limitations of using social media for participation are also
identified by scholars. Kleinhans et al. (2015) argue that wider engagement only
“materializes”, if virtual connections also manifest themselves in real space through
concrete actions, by using both online and offline engagement tools.
6.2 Communication and Collaboration
As Castells (2009) observes, new communication mechanisms become the main
source of signals leading to the construction of meaning in people’s minds, refram-
ing power. He points out that “the communication process decisively mediates the
way in which power relationships are constructed and challenged in every domain of
social practice, including political practice”. Brkovic and Stetovic (2013) argue that
social media provides opportunities for communication, community empowerment
and collaboration. They indicate that social networking sites can be used to expand
the outreach capabilities of governments and planners, and to broaden the abilities
to interact with citizens, through sharing information, making announcements, ask-
ing and answering questions. For instance, in 2013 the Victorian government used
Facebook, Twitter and YouTube to share ideas and engage citizens in a consultation
process for the preparation of the new Metropolitan Planning Strategy for Melbourne
(ibid). In China, the communication and interaction between governments and citi-
zens has increased in recent years due to the development of smart cities (Lin 2018).
Many city governments have used Weibo or WeChat to publicize information such
as policies, plans, and regulations.
Social media can be used as a tool to support the interaction and collaboration
between different groups of people who share a common interest (Sui and Good-
child 2011). Rice and Hancock (2018, p. 96) argue that “collaboration needs to
include new forms of social participation (development of virtual networks and e-
government tools) and social media by employing effective tools that facilitate citizen
decision-making, thereby improving governance processes through empowerment”.
They suggest that such a people-centered participatory and collaborative approach
can promote sustainability and social equity.
6.3 Collective Actions and Citizens’ Activism
Social media has grown beyond the pure “social” realm and is now increasingly
used to cause real impact including community activism (Foth et al. 2011). Social
networking sites make it easy for citizens not only to maintain a number of weak ties,
but also to create large-scale social networks that can perform powerful collective
5 Can Social Media Play a Role in Urban Planning? A Literature … 79
actions (Gordon and Manosevitch 2011). Several cases have been identified in the
literature regarding how citizens have used social media to initiate and organize
collective actions in planning practices.
The case of Puerto Ayora in Ecuador illustrates that social media is an empowering
tool for collective action claiming and proposing a better city (Pinzon 2013). Local
residents used social media such as YouTube to oppose the construction of a suspi-
ciously big building, communicate with planners and local authorities, and organize
social protests. The collective action can be intersected with urban planning through
semi-formal and mixed discursive spheres, driving the changes of urban planning
from the traditional planning approach to the co-produced approach. Pinzon (2013)
studies how social media, as a tool for collective organization and information shar-
ing, affects the power relationships in urban transformation. She analyzes key driving
factors including the social features of new technologies, the tensions between global
and local implications of digital connectivity, the different ways how social media
support social movements, and the limitations and challenges of digital tools.
The case of the installation of the new airport in Mexico City shows the role of
Twitter in activist movements in urban planning (López-Ornelasa et al. 2017). They
find that the knowledge of social media participation can be used to discover the wills
of citizens and be a valid support for design, analysis and decision-making in urban
planning. These new planning practices are thus characterized by citizens’ activism,
different from the communicative or consensus-building approach. It is important
to critically evaluate the democratic potential of social media and recognize the
potential power of local knowledge in shaping urban development (Pinzon 2013).
In China, social media have recently been used by citizens, civil society and
experts to organize collective actions in planning practices. For instance, the resi-
dents in Shifang City of Sichuan Province used Weibo to organize a collective action
against the local government’s decision to build a molybdenum-copper plant (Cheng
2013). Although merely around 20 posts about the plan were publicized on the local
government’s Weibo, they were quickly read and forwarded by thousands of people.
The people then gathered and were in conflict with the local police. As a consequence,
the local government stopped the project permanently. Another case was that citizens
and experts used Weibo to oppose local governments for the cancellation of the num-
ber 55 bus route in a planning practice in Shanghai (Zhao et al. 2018). Citizens and
experts posted their comments and forwarded the information through the Chinese
social media platform of Weibo, creating an open and decentralized network (Fig. 4).
This bottom-up participation has led to the adjustment of the initial plan.
The mentioned emerging planning practices are related to “agonistic planning”,
which is a more radical action rather than consensus-seeking negotiations. The con-
cept of “agonistic planning” entails that the planning process as a concrete activity
supports the encounter between different conceptions of reality (Bäcklund and Män-
tysalo 2010). For instance, the residents and local associations in the area of Vuores
in Finland chose a more radical mode of influence involving the active use of social
media outside the formal channels of planning participation (ibid). However, as an
emerging field of urban planning this form of “agonistic planning” still lacks stud-
ies. More in general, with the increasing use of social media, citizens will have
80 Y. Lin and S. Geertman
Fig. 4 Information transmission between online participants (source Zhao et al. 2018)
more opportunities to establish new networks and relationships and organize their
own collective actions in urban planning, even outside the formal participatory plan-
ning processes. More research is required to understand the role of social media in
bottom-up approaches and collective actions in planning practices.
7 Conclusion and Discussion
The literature review shows that social media is used in two main fields of urban
planning (Fig. 5). On the one hand, social media data are extracted and used in urban
analysis and modelling in the domains of individual activity patterns, urban land use,
transportation and landscape. These data provide the information of geo-locations,
time, texts, and photos of events happening. They are often combined with other
conventional and new datasets and resources such as census data, GIS data, and
remote sensing images for urban analysis and modelling. As pointed out by Batty
et al. (2012), linking social media data with other datasets provides new and open
sources of data essential to a better understanding how smart cities will function.
On the other hand, social media provides a new platform for citizen participation,
communication, collaboration and citizens’ activism. Local governments in many
countries have increasingly used social media to communicate with citizens for urban
policies, regulations and plans. With the rapid development of ICT, citizens have
increasingly participated via websites etcetera about urban planning that affects their
quality of life. Citizens can now easily initiate and organize themselves collective
actions with the support of social media. In that, government is not in charge of the
application of (social) media for participation, but the citizens themselves will decide
upon the format, frequency, intensity, content, etcetera of their involvement in urban
5 Can Social Media Play a Role in Urban Planning? A Literature … 81
Social
Media
Data Platform
Participation
Communication
Citizens’ activism
Activity pattern
Land use
Transportation
Landscape
Challenges: population and spatial bias;
difficulties in extracting useful information
Solutions: advanced techniques for data processing;
combine with other data and conventional methods
Analysis & Modelling
Top-down & Bottom-up
Fig. 5 The role of social media in urban planning (source authors 2018)
planning. These are emerging bottom-up forms of influencing urban planning, which
are very different from conventionally participative or communicative planning that
relies on rational communication of affected stakeholders (Bäcklund and Mäntysalo
2010). Participation through social media can also transcend the physical boundary
of a local community and incorporate non-local actors into the discussion (Zhao et al.
2018). More attentions should be paid to the influence of social media on bottom-up
participation and other emerging forms of citizen involvement in urban planning.
This study also shows several challenges associated with using social media in
urban planning. First, substantial population biases exist across different social media
platforms (Ruths and Pfeffer 2014). Social media is often used by young people, lead-
ing older generations and those lacking internet access out of important discussions.
Second, there are inequalities of internet accessibility and social media usages in dif-
ferent countries, regions, cities and villages. As a consequence, spatial biases exist
among different areas. For instance, the spatial distribution of social media check-in
data is highly heterogeneous, i.e. data is mostly concentrated with big cities and
small cities and rural areas have very few data (Zhan et al. 2014). Third, there are
issues of privacy, since social media data contain the information of locations, texts
and even photos of users. Any analysis conducted on personalized social media data
requires careful attention to aggregate the geotagged information of people that is
not identifiable (Rashidi et al. 2017). Attention should also pay to recent policies on
data privacy and protection, such as General Data Protection Regulation in EU, and
Information Technology Security of Personal Information Security Specifications in
China. Fourth, the most challenging issue in front of using social media data pertains
to complications associated with extracting useful information from the content of
the data (Rashidi et al. 2017).
Several strategies can be applied to deal with the mentioned issues. Using social
media in urban planning needs to be supplemented by other communication mech-
82 Y. Lin and S. Geertman
anisms in order to include people that are disconnected from digital networks. The
limitations of social media lie within the flawed nature of social media data, so
it is necessary to reduce biases during the collection and methodological handling
(Jendryke et al. 2017). Comparing different networks or the same network at differ-
ent times might mitigate these biases (Ruths and Pfeffer 2014). The users may not
constitute a representative sample of the entire population, but social media data are
generated by millions of people from different countries throughout the world (Hu
et al. 2015). This bias is becoming less severe as social media users are growing,
which will make the sample a close representative of the population. Furthermore,
advanced text and data mining techniques, such as linguistic and text mining tech-
niques, can be employed or developed to extract the useful information from social
media data (Rashidi et al. 2017). Besides, the problems can be reduced by inte-
grating social media data with other conventional and new datasets (Kovacs-Gyori
et al. 2018). Finally, the government should support various forms of participation
including bottom-up approaches and pay attention to emerging forms of agonistic
planning. Different types of participation tools such as web-based planning support
systems can be developed to support citizen participation and the cooperation of
various online actors in the planning process (Lin and Geertman 2015). Conven-
tional methods such as in-depth fieldwork and face-to-face meetings are still needed
to understand the actual situation facing a group of citizens and communicate with
affected stakeholders. With the widespread use of social media, it is expected that
there will be an increasing impact on urban planning. More research is needed to
explore this impact of the use of social media on urban planning. In that, it is also
still necessary to pay more attention to the proper dealing with the mentioned issues
during the collection and methodological handling of the social media data. Taken
all that into consideration social media can change urban planning considerable.
References
Bäcklund P, Mäntysalo R (2010) Agonism and institutional ambiguity: ideas on democracy and the
role of participation in the development of planning theory and practice—the case of Finland.
Plann Theor 9(4):333–350
Batty M, Axhausen KW, Giannotti F, Pozdnoukhov A, Bazzani A (2012) Smart cities of the future.
Eur Phys J Spec Top 214:481–518
Brkovic M, Stetovic BV (2013) Harnessing social media for urban planning: an overview. In:
Proceedings of the 49th international society of city and regional planners congress, Brisbane,
Australia. 1–4 Oct 2013
Castells M (2009) Communication power. Oxford University Press, Oxford
Chen Y (2013) Collaborative planning in the network: consensus seeking in urban planning issues
on the internet—the case of China. Plann Theor 12(4):351–368
Chen YM, Liu XP, Li X, Liu XJ, Yao Y, Hu GH, Xu XC (2017) Delineating urban functional areas
with building-level social media data: a dynamic time warping (DTW) distance based k-medoids
method. Landscape Urban Plann 160:48–60
5 Can Social Media Play a Role in Urban Planning? A Literature … 83
CIVITAS Policy Note (2015) The use of social media to involve citizens in urban mobil-
ity projects and city planning. http://civitas.eu/content/civitas-policy-notethe-use-social-media-
involve-citizens-urban-mobility-projects-and-city
Forth M, Forlano L, Satchell C, Gibbs M (eds) (2011) From social butterfly to engaged citizen:
urban informatics, social media, ubiquitous computing, and mobile technology to support citizen
engagement. MIT Press, Cambridge
Fredericks J, Foth M (2013) Augmenting public participation: enhancing planning outcomes through
the use of social media and web 2.0. Aust Planner 50(3):244–256
Gordon E, Manosevitch E (2011) Augmented deliberation: merging physical and virtual interaction
to engage communities in urban planning. New Media Soc 13(1):75–95
Hu YJ, Gao S, Janowicz K, Yu BL, Li WW, Prasad S (2015) Extracting and understanding urban
areas of interest using geotagged photos. Comput Environ Urban Syst 54:240–254
Huang QY, Wong DWS (2016) Activity patterns, socioeconomic status and urban spatial structure:
what can social media data tell us? Int J Geogr Inf Sci 30(9):1873–1898
Jendryke M, Balz T, McClure SC, Liao MS (2017) Putting people in the picture: combining big
location-based social media data and remote sensing imagery for enhanced contextual urban
information in Shanghai. Comput Environ Urban Syst 62:99–122
Jiang S, Alves A, Rodrigues F, Ferreira J Jr, Pereira FC (2015) Mining point-of-interest data from
social networks for urban land use classification and disaggregation. Comput Environ Urban Syst
53:36–46
Kaplan AM, Haenlein M (2010) Users of the world, unite! the challenges and opportunities of social
media. Bus Horiz 53(1):59–68
Kitchin R (2014) The real-time city? big data and smart urbanism. GeoJournal 79:1–14
Kleinhans R, Van Ham M, Evans-Cowley J (2015) Using social media and mobile technologies
to foster engagement and self-organization in participatory urban planning and neighborhood
governance. Plann Pract Res 30(3):237–247
Kovacs-Gyori A, Ristea A, Havas C, Resch B, Cabrera-Barona P (2018) #London2012: towards
citizen-contributed urban planning through sentiment analysis of Twitter data. Urban Plann
3(1):75–99
Lewis S, Pea R, Rosen J (2010) Beyond participation to co-creation of meaning: mobile social
media in generative learning communities. Soc Sci Inf 49:351–369
Lin YL (2018) A comparison of selected western and Chinese smart governance: the applica-
tion of ICT in governmental management, participation and collaboration. Telecommun Policy
42(10):800–809
Lin YL, Geertman S (2015) Smart governance, collaborativeplanning and planning support systems:
a fruitful triangle? In: Geertman S, Stillwell J, Ferreira J, Stillwell J (eds) Planning support systems
and smart cities. Springer, Dordrecht
López-Ornelasa E, Abascal-Mena R, Zepeda-Hernández S (2017) Social media participation in
urban planning: a new way to interact and take decision. In: The international archives of the
photogrammetry, remote sensing and spatial information sciences, XLII-4/W3
Oteros-Rozas E, Martin-Lopez B, Fagerholm N, Bieling C, Plieninger T (2018) Using social media
photos to explore the relation between cultural ecosystem services and landscape features across
five European sites. Ecol Ind 94(2):74–86
Pinzon L (2013) Beyond connectivity: the impacts of social media in urban development in Puerto
Ayora, Ecuador. DPU workingpaper No 160. www.bartlett.ucl.ac.uk/dpu/latest/publications/dpu-
papers
Rashidi TH, Abbasi A, Maghrebi M, Hasan S, Waller TS (2017) Exploring the capacity of social
media data for modelling travel behavior: opportunities and challenges. Transp Res Part C Emerg
Technol 75:197–211
Resch B, Summa A, Zeile P, Strube M (2016) Citizen-centric urban planning through extracting
emotion information from Twitter in an interdisciplinary space-time-linguistics algorithm. Urban
Plann 1(2):114–127
84 Y. Lin and S. Geertman
Rice M, Hancock T (2018) Equity, sustainability and governance in urban settings. Glob Health
Promot 23(1):1757–9759
Ruths D, Pfeffer J (2014) Social media for large studies of behavior. Science 346(6213):1063–1064
Schroeter R, Houghton K (2011) Neo-planning: location-based social media to engage Australia’s
new digital locals. Aust Planner 48(3):191–202
Shelton T, Poorthuis A, Zook M (2015) Social media and the city: rethinking urban socio-spatial
inequality using user-generated geographic information. Landscape Urban Plann 142:198–211
Silva T, Vaz De Melo P, Almeida J, Loureiro A (2014) Large-scale study of city dynamics and urban
social behavior using participatory sensing. Wirel Commun IEEE 21(1):42–51
Statista (2018) Most popular social networks worldwide as of October 2018, ranked by number
of active users (in millions). https://www.statista.com/statistics/272014/global-social-networks-
ranked-by-number-of-users/
Sui D, Goodchild MF (2011) The convergence of GIS and social media: challenges for GIS science.
Int J Geogr Inf Sci 25(11):1737–1748
Tieskens K, Van Zanten BT, Schulp GJE, Verburg PH (2018) Aesthetic appreciation of the cultural
landscape through social media: an analysis of revealed preference in the Dutch river landscape.
Landscape Urban Plann 177:128–137
Tu W, Cao JZ, Yue Y, Shaw SL, Zhou M, Wang ZS (2017) Coupling mobile phone and social media
data: a new approach to understanding urban functions and diurnal patterns. Int J Geogr Inf Sci
31(12):2331–2358
Zhan XY, Ukkusuri SV, Zhu F (2014) Inferring urban land use using large-scale social media
check-in data. Netw Spat Econ 14:647–667
Zhang S, Zhou WQ (2018) Recreational visits to urban parks and factors affecting park visits:
evidence from geotagged social media. Landscape Urban Plann 180:27–35
Zhao MX, Lin YL, Derudder B (2018) Demonstration of public participation and communication
through social media in the network society within Shanghai. Environ Plann B Urban Anal City
Sci 45(3):529–547
Zhi Y, Li HF, Wang DS, Deng M, Wang SW, Gao J, Duan ZY, Liu Y (2016) Latent spatio-temporal
activity structures: a new approach to inferring intra-urban functional regions via social media
check-in data. Geo-spatial Inf Sci 19(2):94–105
Chapter 6
Bridging the Information and Physical
Space: Measuring Flow
from Geo-Located Social Media Data
on the Street Network
Alireza Karduni and Eric Sauda
Abstract Social Media usage is becoming more and more interwoven with activities
in urban space. Understanding the flows of information in cities can open new doors
for us to understand how urban space relates to human behavior. In this chapter,
we introduce a method to extrapolate flow of geolocated social media data for a
street network. We then apply this method to a corpus of geolocated tweets collected
from the Los Angeles metropolitan area. We compared the results to betweenness
centrality of the streets as a measurement of connectivity and density of businesses
as a measurement of public activity. We find that the flows calculated from Twitter
have a higher correlation with public activities hinting that there is a relationship
between geolocated social media usage and businesses and public space.
Keywords Social media ·Street network ·Information space ·Physical space
1 Introduction
I could tell you how many steps make up the streets rising like stairways, and the degree of
the arcades’ curves, and what kind of zinc scales over the roofs; but I already knowthis would
be the same as telling you nothing. The city does not consist of this, but of relationships
between the measurements of its space and the events of its past: the height of a lamppost
and the distance from the ground of a hanged usurper’s swaying feet; the line strung from
the lamppost of the railing opposite and the festoons that decorate the course of the queen’s
nuptial procession; the height of that railing and the leap of the adulterer who climbed over
it at dawn (Calvino 1978)
We cannot comprehend cities by studying space, time or occupation in isolation.
An understanding of the tangled flow of people, information, and goods over time
A. Karduni (B)
College of Computing and Informatics, UNC-Charlotte, Charlotte, USA
e-mail: akarduni@uncc.edu
E. Sauda
School of Architecture, UNC-Charlotte, Charlotte, USA
e-mail: ejsauda@uncc.edu
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_6
85
86 A. Karduni and E. Sauda
within architecture, public space and infrastructure would more accurately capture
the life of a city. It is through the study of such flows that we can understand how a
city lives, thrives, and changes. Indeed, we might be better able to understand and
plan for issues such as crime, innovation, accessibility and public gatherings.
The rise of smartphone technology presents unique opportunities for studying
these issues. As of 2016, the number of smartphone users globally has exceeded
3 billion; this is projected to exceed 6 billion by 2021 (Ericson.com 2017). Every
day, more information is mobile, used by people while moving through the city and
conducting their everyday tasks, leaving behind a trail of data from their activities on
social media. Data from social media has been used to understand various phenomena
such as situational awareness (Yin et al. 2015) and misinformation (Karduni et al.
2018), thereby transforming the way that we understand cities (Wessel et al. 2018).
One particular class of applications, mobile social media (Twitter, Foursquare,
Yelp), contains a mix of geospatial, temporal and topic information that has the
potential to be particularly useful in understanding the flows of ideas in a city (Sauda
et al. 2018). In this chapter, we investigate opportunities and constraints using one
of these applications (Twitter) to understand the city. Specifically, we identify three
critical issues critical for urban social media and propose innovative methods for
studying and understanding their impacts and usage.
Firstly, data from social media is usually partial data. For example, the data from
Twitter, while widely available for research, contains geolocations for only a small
percentage of the total. Interpreting the geolocation requires new methods of map-
ping and visualization. In this chapter, we present a unique method of visualization
and understanding of this flow based on the Dijkstra shortest path algorithm (Dijkstra
1959). We then compare our method with betweenness centrality as a measurement
of physical connectivity and Point of Interest (PoI) density as a measure of concen-
trations of activity.
Secondly, the empirical nature of the data available from these sources is both a
challenge and an opportunity. The primary opportunity is the ability to study how
people use the city; a “bottom-up” approach offers an opportunity to develop a
predictive model to supplement and test the mainly normative models in use for urban
design. A significant challenge is the scale of the data, which is inevitably very large
(typically millions of records) and present considerable problems of cognition and
interpretation. Understanding these large data sets requires the use of methods from
computer science and statistics. We present visual analytic methods that facilitate
the understanding of such data sets.
Thirdly, the mix of data types available can lead to unique interpretations of the
data based on sets and subsets of the total corpus. The data available from mobile
social media sites all contain content (text, photos), a spatial location and a time. By
understanding and exploring these parallel data types, a new understanding of the
city of flows emerges. We present one example of this type of exploration based on
different patterns of French and Italian language tweets.
Our primary objective is to study and understand the flow of Twitter users within
urban space. We do so by introducing a new method in studying the movement of
Twitter users using the underlying road system as the network of movement (from
6 Bridging the Information and Physical Space: Measuring Flow … 87
now on we call this process “flow” analysis). With this method, we connect the
locations of every user who tweets more than once during a day when using the
shortest path that they would normally travel on the street network. We applied this
method to a large dataset of geolocated tweets in Los Angeles, California during the
period from September and October 2015.
Our secondary objective is to start the process of understanding the nature of this
flow analysis. We do so by testing the flow of Twitter users with two other elements
for Los Angeles. Namely, we compare the flow analysis created with our method with
betweenness centrality of the street network of Los Angeles as a model that has been
used to study flows within street networks (Borgatti 2005). This comparison allows
us to understand whether our flow model can be predicted by a similar predictive
method that uses shortest paths on street networks or not. Next, we compare the
flow analysis with the locations of Points of Interest (PoIs) in Los Angeles. This
comparison allows us to understand further whether the movement of Twitter users
occurs in areas with a higher concentration of commercial activity or not.
This chapter is organized as follows: In Sect. 2of this chapter, we first visit the
previous related work that has been conducted on studying flows and movement using
social media data and then thoroughly discuss our methodology and tools utilized
to conduct this research. Next, we describe all of the datasets and tools utilized to
perform our analysis. In Sect. 4, we describe the benefits of using Twitter data as
a proxy for studying flow. Next, we discuss the limitations and considerations of
our datasets and methods, and finally, we conclude with remarks concerning future
research directions and potential improvement of our methodology.
2 Related Work
In a sense, humans can be considered “sensors” for their own everyday actions (Ratti
and Claudel 2016). As a major trend in human society with clear spatial implications,
this phenomenon is worthy of careful study. It has created new ways for researchers
studying the movement and behavior of humans within cities.
This task has always been an important yet difficult task within the realm of
urbanism. Many researchers have investigated and emphasized the importance of
these studies from various perspectives. Researchers at the Space Syntax try to define
the network structure of urban space as the main dictator of natural movement within
cities (Hillier et al. 1993). Similarly, Salingaros abstracts urban systems to nodes of
human activities and their connections and emphasizes that this abstraction can be
a useful axiom for urban planning and design (Salingaros 1998). Researchers at
University College of London have created various simulation software packages to
model pedestrian movement using Agent Based Models (Kerridge et al. 2001). Many
of these models, however, rely heavily on theoretical methods to predict or simulate
human movement.
The abundance of data created from social media and mobile devices has created
a wave of new and diverse studies and methods for the study of human activity and
88 A. Karduni and E. Sauda
movement within urban space in real time. The Real-time Rome project uses an
aggregate of various data sources to observe spatio-temporal activities within Rome
which showed real-time spatial response to the events of World Cup 2006 Final
(Rojas et al. 2008). Wessel et al. have utilized location, time, and content of Twitter
data related to food trucks in Charlotte NC to show how food trucks move around
and communicate with their customers using social media (Wessel et al. 2016).
Other research has been conducted on accurately identifying city centers using flows
observed from social media (Sun et al. 2016), real-time extreme event detection
(Sakaki et al. 2010), real-time extreme event detection (Sakaki et al. 2010), and
geographic distribution of art and cultural industries of metropolitan areas (Currid
and Williams 2009). These examples show the possibilities that study of real-time
social media bring to the study of urban systems.
3 Materials and Methods
Our analysis process consists of three major steps:
•Modeling flow from tweet geolocation data.
•Calculating edge betweenness centrality of the streets.
•Calculating the density of PoIs for the same area.
•Comparing the results of the flow analysis with betweenness centrality and PoIs
density.
3.1 Calculating Twitter Flow from Geo-Located Tweets
Before describing the details of how flow is modeled, we need to define flow. Within
the context of our research, flow is the movement of every tweet user who tweets more
than two times during a day. With this definition, the initial way to both visualize
and analyze flow would be to connect a straight line between every tweet point for
every user for a day. However, this method does not consider the context and space
of where the users move around. More specifically, we can safely assume that almost
all of the users move within the boundaries of streets of a city especially if that city
has no public transportation or mostly surface public transportation modes.
In order to consider the space of the city in our analysis, we created a model that
would connect every consecutive tweet point of a user using the shortest path between
those two points calculated with the well-known Dijkstra’s algorithm (Dijkstra 1959).
We would then continue this task for every user during a single day that has tweeted
more than three times. And finally, we would repeat the same task for every day we
have collected tweet data and then merge the final results.
In summary, our flow pseudocode is as follows:
6 Bridging the Information and Physical Space: Measuring Flow … 89
1. Build a graph network of the streets that covers the exact area that covers all
tweet locations (in our case, our graph consists of street intersections as nodes
and street segments between them as edges).
2. Calculate the flow:
a. Get all the tweets for the first day of the dataset.
b. Get all the unique user names who have tweeted more than 2 times.
c. For every user:
i. Sort the tweets based on the time tweet was posted.
ii. For every point:
1. Find the closest point on the closest edge on the graph.
2. Find the closest point on the closest edge for the next tweet point.
3. Calculate the shortest path of those points on the graph.
4. Save the result.
d. If there is a next day in the dataset:
i. Get all the data for that day.
ii. Repeat task 2.
e. Then: complete the program.
This method results in a separate polyline file for each unique user on each day
with each polyline showing the result of the shortest paths between consecutive tweet
locations of that user. It is important to note that this method is beneficial for any
episodic dataset where we can identify different users’ locations throughout a single
time frame.
3.2 Betweenness Centrality as a Measure of Flow
To compare our model of flow using Twitter data with another mode of flow that
is based on the concept of the structural importance of elements within a network,
we calculated the edge betweenness centrality of every street segment for the same
graph network created for our previous step. Betweenness centrality of an edge or
node is a measure of how likely that edge is to be used to link any two pairs of nodes
within the network (Freeman 1978). A high betweenness centrality for an edge or
node shows that there are shortest paths going through that edge and it can be a
measurement of flow volume and robustness of that edge (Duan and Lu 2013).
Mathematically, betweenness centrality is defined as:
CBn=
s=n=t
σst(n)
σst
Where σst is the total number of shortest paths from node s to node t that pass-
through node n. Edge betweenness centrality is also the total number shortest paths
that go through each edge.
90 A. Karduni and E. Sauda
3.3 Comparing Flow, Centrality and Density
Our goal for comparison is to answer the question “does the tweet flow follow
betweenness centrality of the streets or does it follow where PoIs are concentrated?”
The answer to this question will effectively show us whether our flow analysis can
be predicted by a static street-based method that the same shortest path method on
the street network produces.
Similar to the research conducted by Porta et al. (2009) that studied the relation-
ships of street centralities and business locations using kernel density as a method of
converting different data sources to the same analysis unit, we will calculate the ker-
nel density of edge-betweenness centrality of street networks, Twitter flow analysis
results and PoI locations. In order to calculate the kernel densities, we will use a fixed
bandwidth (search radius) of 0.5 miles. For calculating the kernel density of the edge
betweenness centrality of street networks, we will use the betweenness value of each
street as the population value. For the flow analysis results and PoIs, no population
value exists and every feature in both datasets will have an equal population of 1.
To compare the two density calculations, we will use Spearman’s rank correlation.
The reason for choosing this correlation method is that the nature of our data is
nonparametric, and the distribution of our data is not normal which makes Spearman’s
rank correlation a more valid method. Moreover, the numerical values of our analysis
are not entirely meaningful by themselves, rather than the rankings of how high or
low these numbers are in comparison to each other. This method will rank the areas
based on betweenness centrality density, flow density, and PoIs density from highest
to lowest and then calculate Spearman’s correlation.
Spearman’s rank-order correlation is calculated as:
ρ=6dd2
i
n(n2−1)
where diis the difference in paired ranks and n is the number of cases.
The result of the Spearman’s correlation will be a coefficient number and a P-Value
which shows the significance of the correlation between the two ranked datasets.
3.4 Data Sources
As the main focus of our study is to analyze the flow of Twitter tweets, we assessed a
dataset of 3,117,738 tweets in the Los Angeles Metropolitan area. The Twitter dataset
consists of two types of tweet data: tweets with a place label (e.g. Santa Monica and
a rectangular area for where the tweet occurred) and tweets with a spatial location
that includes latitude and longitude information. For our analysis, we only utilized
tweets that had latitude and longitude information. The final number of tweets used
6 Bridging the Information and Physical Space: Measuring Flow … 91
Fig. 1 Data assessment bounding box
within our analysis was 968,424 from the commencement of September to the end
of October 2015 (Fig. 1).
Our dataset was assessed by GNIP of the UNC-Charlotte data science initia-
tive, a platform that offers historical and real-time tweets. The data was received in
November 2015 in a Javascript Object Notation (JSON) format and was stored in a
MongoDB No-SQL database. All of the preprocessing and queries were made using
PyMongo, the MongoDB interface for python. Each tweet object within our database
includes many metadata along with the text body of the tweet (See https://developer.
twitter.com/en/docs for more information). For our flow analysis, we only utilized
a portion of the metadata for each tweet including Date, Latitude and Longitude,
Time, and the User Name.
For our flow analysis and betweenness centrality, we utilized street centerline data
from OpenStreetMaps (OSM) for the same bounding box that the tweet data were
received in. The data were accessed from GeoFabrik in May 2016. In order to create
a network graph file from the street in both ESRI Shapefile and Edge List, we used
GISF2E which creates a network of nodes and edges shapefile, and an Edge List
graph that shows the start node and end node of every street segment as well as the
length of the street segment in meters (Karduni et al. 2016). After the network edge
list was created, we used python Igraph to clean up the graph by first obtaining the
92 A. Karduni and E. Sauda
giant component (i.e. the largest connected portion of the graph) and then iteratively
removing all of the nodes with a degree of 1 to reduce the size of the graph without
affecting the analysis results.
Similar to our street centerline dataset, we used PoI data from OpenStreetMaps.
OSM has a very loose definition of PoIs which includes point location for many
different places from bars and restaurants to water fountains. The dataset includes
tags for the type of business which is not standardized. Our team cleaned the dataset
for the same bounding box to include only points that convey regular human activity
(businesses, tourist attractions, offices, grocery stores, etc.).
Our analysis process consists of three major steps. Preprocessing and cleaning
the data, analyzing the data, and visualizing the data.
For data cleanup and preprocessing, we used python and MongoDB. Our data
cleanup process only consisted of filtering the tweets that included latitude and
longitude location and storing each tweet as a document in a MongoDB.
Data analysis was the most complex portion of our process. We used MongoDB
and python to access the tweet data. We used an arcpy and ArcGIS Geodatabase to
store and access OSM streets and POI data.
Our analysis consists of calculating the shortest paths between every location for
every user that tweets more than two times during every single day for two months for
which we used the shortest path using ArcGIS for Network Analyst Routes Analysis
layer using arcpy. Another version of the model that produces shortest paths between
tweets is developed using Python and Networkx available for usage.
For betweenness centrality calculation, we used Python Igraph which is a library
for graph analysis. We used an equivalent network edge list of the same street dataset
for this purpose.
For calculating Kernel Densities of our datasets, we used ArcGIS 10.2 and to
calculate Spearman’s correlation we used SciPy 0.17.0 for python.
For static data visualization, we used ArcMap, QGIS, and Excel. For interactive
visualization of the data we used JavaScript, D3, and Leaflet.
4 Results
In this section we will first present some raw results from our flow analysis to better
understand the nature of the resulting dataset and next we will move on with com-
paring the flow analysis result with betweenness centrality of the street network and
PoIs.3.1. Subsection.
4.1 Flow Analysis Results
We applied the flow analysis method described in the previous section to a dataset
of geolocated tweets for Los Angeles in the months of September and October of
6 Bridging the Information and Physical Space: Measuring Flow … 93
Fig. 2 Number of days unique users have movement in the dataset
Fig. 3 Count of tweet flow polylines per day for the span of the dataset
2015. For the total span of the dataset, there were more than 39,000 flow polylines
created which were generated by more than 13,000 unique users. Out of the unique
users who have tweeted multiple times on a single day, the majority were one-time
only users. This result shows that geolocated tweets do not come from only a few
users with high activity rather they are generated from a diverse set of users (Fig. 2).
Figure 3shows the timeline of the flow analysis. This chart shows that we can
observe a repeating pattern of Saturdays with the highest flow activity and Tuesdays
with lowest. Geographically, the flow polylines are spread out across the metropolitan
region. In the next section, we will study whether the densities of these flow polylines
are more reflective of street connectivity or business activities. Figure 4shows a
snapshot of the flow lines created.
94 A. Karduni and E. Sauda
Fig. 4 All flow polylines for a portion of the Los Angeles metropolitan area
Fig. 5 Juxtaposing these maps shows that Tweet flow has a higher spatial correlation with concen-
trations of businesses. Densities of flow, betweenness, and PoIs
4.2 Comparing Densities of Twitter Flows with Business
Density and Street Connectivity
As described in the methodology section, to compare the results from our flow
analysis with betweenness centrality of the street networks and concentrations of the
points of interests, we will calculate kernel density for each of these datasets and
then conduct Spearman’s rank correlation between the pixel values of these density
layers. Figure 5juxtaposes the results of the kernel densities of the flow analysis to
both betweenness centrality and PoIs. Table 1shows the results of the Spearman’s
rank correlations.
The correlation analysis shows that the flow analysis density has a correlation
coefficient of 0.54 with a P-Value of 0.00. The same correlation analysis between
flow and betweenness shows a correlation coefficient of 0.37 with a P-Value of 0.01.
These two correlation coefficient results show that the areas in Los Angeles where
6 Bridging the Information and Physical Space: Measuring Flow … 95
Tabl e 1 Spearman’s
correlation between flow and
betweenness, and between
flow and PoI density
Spearman’s correlation
for 33,074 raster pixels
Correlation coefficient Pvalue
Flow and PoIs 0.54 <0.01
Flow and betweenness
centrality
0.37 <0.01
Betweenness and PoIs 0.23 <0.01
we see a higher density of tweet flow happen to coincide more with areas with high
concentrations of points of interests in comparison to the areas with higher between-
ness centrality. This is of course partially caused by the higher correlation between
geolocated tweets and commercial areas. The lower correlation coefficient of 0.23
between betweenness centrality and PoIs shows that for Los Angeles, our tweet anal-
ysis could be a better method for studying the flows in commercial areas, especially,
because social media data comes with text and other semantics that inexpensively
open doors for spatiotemporal and semantic analysis of flows in commercial areas.
Using tweet data to conduct urban flow analysis has its limitations. However,
there are some unique features that our method can bring about for researchers. The
first obvious benefit is the fact that accessing Twitter data is relatively inexpensive.
Moreover, the fact that Twitter is widely used in most parts of the world makes
such a method easily applicable to many other cities around the world where other
technologies are not as easily accessible. Moreover, even though this research was
conducted with historical data from 2015, similar methods could easily be turned
into real-time monitoring of activity within cities.
Twitter data has been used to enhance trip prediction in urban networks (Poure-
brahim et al. 2018). Besides the accessibility and abundance of Twitter data, one other
feature makes this analysis unique: each tweet comes with 280 characters that could
be mined for more contextual information regarding the flow of people. For exam-
ple, while generating the shortest paths between different users, we also recorded
the language each user tweets in as well as the start and end time of each flow poly-
line. These metadata, along with location information, makes the flow analysis very
unique and useful. Figure 6shows two maps created with flows in French and Italian
languages. Language is the simplest form of data which can be extracted from the
content of tweets and turned into flows. Using various topic modeling algorithms
such as LDA, it is possible to study the movement of similar topics across urban
areas (Karduni et al. 2017b).
Finally, we would want to mention that this type of flow data is very accessible
to visualize and analyze with almost all popular platforms including ArcGIS, QGIS,
Leaflet, etc. As an example, our research group has developed a Visual Analytics
System that combines this flow analysis, with other types of visualizations and anal-
yses derived from the same tweet dataset with the goal of enabling other researchers
to create new forms of knowledge about cities from this type of analysis by exploring
content, space, and time of tweets at the same time (Karduni et al. 2017a).
96 A. Karduni and E. Sauda
Fig. 6 Example of how the flow analysis method can be expanded to explore other aspects of social
media within urban space. Densities of French and Italian tweet flows
5 Conclusion
The most important point to consider is the nature of tweet data and the fact that
only a miniscule 1.6% of tweets have latitude and longitude geolocations (Leetaru
et al. 2013). It is also important to note that Twitter has a bias towards the younger
population of those aged under 35 years of age (Sloan et al. 2015). Any demographic
interpretation from Twitter data should consider this bias. Given these facts, future
research should be dedicated to understanding how much the movement of tweet
users can be generalizable to the wider population.
Another problem with this type of data derived from Application Programming
Interfaces (APIs) that tap into public social media is that these datasets are not
reproducible. What this means is that if you make the same call for the same period
for the same geographic area, it will result in a different dataset than what we have
now. However, our basic analysis of the number of shortest paths from the results
section is that we can expect similar behavior from the dataset if new calls were to
be made for new datasets.
Moreover, it is also safe to say that not all tweets are made by humans. Many
tweets are made by bots, and it is also possible for them to have a fake geolocation.
There have been methods developed for identifying bots from the text of the tweet or
the metadata of the user (Ferrara et al. 2016). Accurately identifying bot generated
tweets was beyond the scope and interest of our research focus. However, based on
the results of our tweet user per day analysis (Fig. 2) and the fact that the majority of
unique users appeared in less than 10 days of our datasets, our datasets are unlikely
to have been afflicted by bots. We can assume that a significant portion of our flow
analysis is generated by human users, as bots are more likely to tweet in a more
intense repeating pattern. We plan to incorporate bot identifying algorithms in our
future analysis.
Finally, it is crucial to mention that this analysis was applied to only a two-month
period for Los Angeles. To truly benefit from this analytical method, a larger dataset
6 Bridging the Information and Physical Space: Measuring Flow … 97
for Los Angeles and other cities with substantial spatial and cultural differences from
Los Angeles could bring to light many other features about the nature of our method
and social media activity within different urban areas. We plan to collect data and
conduct a comprehensive comparative analysis using our method for various cities
around the world in the future.
We believe that social media data, such as Twitter, has tremendous potential for
transforming our understanding of cities and humans within them. The method and
the tool provided that we developed in this chapter has only scratched the surface on
the wealth of information we can derive from such datasets. We call for researchers
in urbanism and urban planning to use methods and tools that represent a genuine
advance on traditional methods of understanding spaces, and we hope that such
methods will find applications within the practice of future urban planning and design.
References
Borgatti SP (2005) Centrality and network flow. Soc Netw 27(1):55–71
Calvino I (1978) Invisible cities. Houghton Mifflin Harcourt, Boston
Currid E, Williams S (2009) The geography of buzz: art, culture and the social milieu in Los Angeles
and New York. J Econ Geogr 10(3):423–451
Dijkstra EW (1959) A note on two problems in connexion with graphs. Numer Math 1(1):269–271
Duan Y, Lu F (2013) Structural robustness of city road networks based on community. Comput
Environ Urban Syst 41:75–87
Ericson.com (2017) Mobility report. https://www.ericsson.com/en/press-releases/2017/6/over-one-
million-new-mobile-internet-users-added-per-day. Accessed 2018
Ferrara E, Varol O, Davis C, Menczer F, Flammini A (2016) The rise of social bots. Commun ACM
59(7):96–104
Freeman LC (1978) Centrality in social networks conceptual clarification. Soc Netw 1(3):215–239
GNIP.com GNIP Twitter Data Format. http://support.gnip.com/sources/twitter/data_format.html.
Accessed
Hillier B, Penn A, Hanson J, Grajewski T, Xu J (1993) Natural movement: or, configuration and
attraction in urban pedestrian movement. Environ Plann B Plann Des 20(1):29–66
Karduni A, Kermanshah A, Derrible S (2016) A protocol to convert spatial polyline data to network
formats and applications to world urban road networks. Sci Data 3:160046
Karduni A, Cho I, Wessel G, Dou W, Ribarsky W, Sauda E (2017a) Urban activity explorer: visual
analytics and planning support systems. In: International conference on computers in urban
planning and urban management. Springer, pp 65–76
Karduni A, Cho I, Wessel G, Ribarsky W, Sauda E, Dou W (2017b) Urban space explorer: a visual
analytics system for urban planning. IEEE Comput Graph Appl 37(5):50–60
Karduni A, Wesslen R, Santhanam S, Cho I, Volkova S, Arendt D, Shaikh S, Dou W (2018, June)
Can you verifi this? studying uncertainty and decision-making about misinformation using visual
analytics. In: Twelfth international AAAI conference on web and social media
Kerridge J, Hine J, Wigan M (2001) Agent-based modelling of pedestrian movements: the questions
that need to be asked and answered. Environ Plann B Plann Des 28(3):327–341
Leetaru K, Wang S, Cao G, Padmanabhan A, Shook E (2013) Mapping the global Twitter heartbeat:
the geography of Twitter. First Monday 18(5)
Porta S, Strano E, Iacoviello V, Messora R, Latora V, Cardillo A et al (2009) Street centrality and
densities of retail and services in Bologna, Italy. Environ Plann B Plann Des 36(3):450–465
98 A. Karduni and E. Sauda
Pourebrahim N, Sultana S, Thill JC, Mohanty S (2018, Nov) Enhancing trip distribution prediction
with Twitter data: comparison of neural network and gravity models. In: Proceedings of the 2nd
ACM SIGSPATIAL international workshop on AI for geographic knowledge discovery, ACM.
pp 5–8
Ratti C, Claudel M (2016) The city of tomorrow: sensors, networks, hackers, and the future of urban
life. Yale University Press, New Haven
Rojas F,Calabrese F, Krishnan S, Ratti C (2008) Real time rome. Netw Commun Stud 20(3):247–258
Sakaki T, Okazaki M, Matsuo Y (2010) Earthquake shakes Twitter users: real-time event detection
by social sensors. In: Proceedings of the 19th international conference on world wide web, ACM.
pp 851–860
Salingaros NA (1998) Theory of the urban web. J Urban Des 3(1):53–71
Sauda E, Wessel G, Karduni A (2018) Data and design: using knowledge generation/visual analytic
paradigms to understand mobile social media in urban design. In: Humanizing digital reality,
Springer. pp 537–545
Sloan L, Morgan J, Burnap P, Williams M (2015) Who tweets? Deriving the demographic character-
istics of age, occupation and social class from Twitter user meta-data. PLoS ONE 10(3):e0115545
Sun Y, Fan H, Li M, Zipf A (2016) Identifying the city center using human travel flows generated
from location-based social networking data. Environ Plann B Plann Des 43(3):480–498
Wessel G, Ziemkiewicz C, Sauda E (2016) Revaluating urban space through tweets: an analysis of
Twitter-based mobile food vendors and online communication. New Media Soc 18(8):1636–1656
Wessel G, Karduni A, Sauda E (2018) The image of the digital city: revisiting Lynch’s principles
of urban legibility. J Am Plann Assoc 84(3–4):280–283
Yin J, Karimi S, Lampert A, Cameron M, Robinson B, Power R (2015, June) Using social media
to enhance emergency situation awareness. In: Twenty-fourth international joint conference on
artificial intelligence
Chapter 7
Comparing Smart Governance Projects
in China: A Contextual Approach
Huaxiong Jiang, Stan Geertman and Patrick Witte
Abstract Smart governance has been increasingly gaining momentum to deal with
the challenges of fast urbanization in China. However, only limited English liter-
ature is available to enhance our understanding of Chinese smart governance. In
this chapter, special emphasis is put on our transformative understanding of smart
governance in China by identifying the impacts of urban context on smart gover-
nance arrangements from an urban planning perspective. As such, we concentrate
on smart governance projects in different Chinese cities, foremost based on Chinese
literature. We aim to investigate: (1) what smart governance means in Chinese urban
planning settings, and (2) how urban context influences the smart governance projects
in different Chinese cities. On the basis of an intensive review of Chinese literature
on smart governance, we find that smart governance in China varies significantly.
Therein, we can identify four types of smart governance in China, including: (1)
constructing a pilot area, (2) improving government performance, (3) building trust
in government, and (4) encouraging innovation. A comparative exploration of four
Chinese projects representing these four types of smart governance shows that the
urban context affects the interaction between technology and urban actors. Moreover,
it shows that this interaction has a feedback effect on the urban context itself. From
this we can conclude that knowledge on the urban context is vital to understand the
expected outcomes of intended smart governance arrangements.
Keywords Smart governance ·China ·Urban context ·Technology ·Urban actors
H. Jiang (B)·S. Geertman ·P. Witte
Department of Human Geography and Planning, Faculty of Geosciences, Utrecht University,
3584 CS Utrecht, The Netherlands
e-mail: h.jiang@uu.nl
S. Geertman
e-mail: s.c.m.geertman@uu.nl
P. Wi t te
e-mail: p.a.witte@uu.nl
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_7
99
100 H. Jiang et al.
1 Introduction
An estimated one billion people in China will live in urbanized areas by 2025 (United
Nations 2018). As a way to deal with these imminent challenges of fast urbanization,
the concept of smart city and its governance has increasingly gained momentum in
China. As of August 2013, 193 national smart city pilot projects have been announced
by China’s Ministry of Housing and Urban and Rural Development (MOHURD).
After five years of construction, about 500 smart city pilots have been implemented
in 2018, outnumbering all other countries combined.1In China, the development of
Information and Communication Technology (ICT) and national policies have been
key drivers of smart city initiatives (Hao et al. 2012). The use of ICT has enabled
governments to promote public policy and provide better services to its citizens (Lin
2018). By reconsidering the role of governments in a knowledge-based smart city,
traditional governing approaches are transformed into what has been called as ‘smart
governance’ (Bolívar and Meijer 2016).
According to Batty et al. (2012), smart governance means using digital innovations
to optimize the efficiency of city operations and services. Giffinger et al. (2007)
claim that smart governance is a broader concept and includes political participation,
services for citizens and the functioning of the administration. Based on an intensive
literature review, Meijer and Bolívar (2016) define smart governance as a new form
of human collaboration by using ICTs to gain more open governance processes and
better outcomes. In the limited English literature dealing with the Chinese context,
smart governance in China is mainly understood as the use of technology to integrate
public services, community activities and city management (Lin 2018; Shi 2018). By
utilizing the strengths of some top technology companies such as Alibaba, Tencent,
and Huawei, a technology-centric approach is deemed as the main way to govern
smart city initiatives.2In the next section we will elaborate more on the content of
the smart governance concept.
Ruhlandt (2018:9) highlights that contextual factors should be assumed to impact
the governance of smart cities. However, current research on smart governance mostly
neglects the importance of urban contextual factors in shaping smart governance.
Context refers to the circumstances and situations that form the setting for one spe-
cific smart governance approach, and in terms of which it can be fully understood
and assessed. According to Nielsen and Pedersen (2014:412), ‘decision-makers have
different personal preferences […] and a range of contextual factors encourage dif-
ferent behaviors’. Factors such as ‘political or demographic factors, administrative
cultures, and technological factors’ are assumed to affect smart governance (Bolí-
var and Meijer 2016). Thus, it is vital to take context-specificity into account in the
decision-making process. Accordingly, Janowski (2015) presents a four-stage ICT-
enabled governance model and argues that the success of ICT-enabled governance
1http://www.xinhuanet.com/english/2018-02/20/c_136987058.htm.
2https://www.businesswire.com/news/home/20180821005297/en/Ping-Showcases-Innovative-
Solutions-Cities-China-Smart.
7 Comparing Smart Governance Projects in China: A Contextual … 101
should evolve towards the complexity and greater contextualization and specializa-
tion. This means that the initiation of new forms of smart governance needs to be
contextualized, since approaches that work in one city may fail in another (Meijer
2016). From literature (Meijer et al. 2016) it shows that the importance of context
in influencing smart governance arrangements is increasingly stressed, but a more
systematic analysis of the role of context is lacking.
In this chapter, special emphasis is put on a transformative understanding of smart
governance in China by identifying the impacts of urban contexts on smart gover-
nance arrangements from an urban planning perspective. As such, we concentrate
on smart governance projects in different Chinese cities, foremost based on Chinese
literature. This brings us to the research questions we will address in this chapter: (1)
what does smart governance mean in Chinese urban planning settings? and (2) how
do urban contexts influence smart governance projects in different Chinese cities?
To answer these questions, this chapter is structured as follows. Section 2focuses on
the theoretical debates on smart governance and its relationship with urban contexts.
A conceptual model is presented to study smart governance projects in different
Chinese cities. Section 3outlines our methodology. Section 4presents the empirical
findings based on an explorative analysis of four distinctive types of smart gover-
nance projects described in the Chinese literature. Section 5provides a discussion
of the findings, along with the final conclusion.
2 Integrating Context into the Smart Governance Debate
2.1 Defining Smart Governance
Smart governance suggests an integration of new technologies into traditional urban
governance processes. In this view, smart governance deals with the utilization of
ICTs to promote a much stronger intelligence function for smartening the many dif-
ferent components of a city (Batty et al. 2012). Doing so, technology is identified as
the defining characteristic of smart governance (Scholl and AlAwadhi 2016). ICTs
such as big data, sensors, social media and monitoring tools provide the informa-
tion basis. Information is extracted from multiple sources and organized by ICTs,
which not only helps urban government and planners to analyze the urban problems,
evaluate alternatives, and forecast future scenarios, but also facilitates engagement
of distinctive stakeholders and produces greater transparency on the part of gov-
ernment (Bertot et al. 2010). The increased transparency further speeds up the pro-
cess of democratizing decision-making. For instance, the advent of web-connected
collaboration platforms allows individuals and communities to become more effec-
tively self-organized in participatory urban planning and neighborhood governance
(Kleinhans et al. 2015). The utilization of various ICTs in government organiza-
tions supports a transformation of the ‘government-to-government’ relationship to a
102 H. Jiang et al.
‘government-to-citizen’, ‘citizen-to-government’, or ‘citizen-to-citizen’ relationship
(Linders 2012).
However, smart governance, as a sub-domain of governance theory, cannot avoid
scrutinizing power and authority that is distributed over different actors (Gil-Garcia
2012), since governance is closely related to the steering and co-ordination of interde-
pendent (usually collective) actors (Treib et al. 2007). From this perspective, smart
governance highlights the importance of investments in human and social capital
(Caragliu et al. 2011) and calls for pro-active and open-minded governancestructures,
with all actors involved, to improve urban performance (Kourtit et al. 2012). Inter-
actions between different urban actors will help contribute to differentiated knowl-
edge, ideas and opinions to enhancing the process of mutual learning and improve
the effectiveness of decision-making. Besides, interactive power relations among
different actors further impact the design, implementation and use of technologies in
ICT-enabled governance, because technology as social artifact has various meanings
to different urban stakeholders (Yang 2003).
Nevertheless, Gil-Garcia (2012) suggests a double-sided interaction between tech-
nology and social actors in ICT-enabled governance. In the same vein Johnston and
Hansen (2011) claim that the governance of smart cities should not separate tech-
nology from human-based capital; instead, smart governance ought to bring smarter
e-participation devices, government and society closer together for collaborating
over proposed issues. According to Hammad and Ludlow (2016), the principle of
integrating ICT with participatory decision-making is crucial to defining smart gov-
ernance. As a result, the issue of socio-techno synergy has been put at the heart of the
smart governance debate (Meijer and Bolívar 2013). The interaction between differ-
ent social actors leads to technology change, which in turn transforms actor relations.
Thus, conceptualizing smart governance as an emergent socio-techno practice will
help to develop a better theoretical understanding of the concept of smart gover-
nance (Meijer and Bolívar 2016). Despite, Meijer (2016:75) argues that ‘studying
the effects of smart (city) governance is complicated since the relations between
governance arrangements, use of technologies, and effects on the quality of urban
life are contextual’. Hence, smart governance should place itself at a higher level
of transformation because it not only demands a timely transformation of internal
and external structures, but also requires the ability to express its context-sensitivity
(Janowski 2015).
2.2 Why Context Matters
Several articles have theorized and investigated the importance of contextual factors
in affecting smart governance. Bolívar and Meijer (2016) claim that factors such
as ‘political or demographic factors, administrative cultures, and technological
factors’ are assumed to affect smart governance. Meijer (2016) highlights the hidden
role of ‘the local cooperative knowledge potential’ and ‘the nature of the problem
domain’ in configuring smart city governance. For instance, according to Kalathil
7 Comparing Smart Governance Projects in China: A Contextual … 103
Fig. 1 A contextual
approach towards smart
governance
Smart governance
arrangements
Local contexts
(Economic,
political,
cultural,
technological, &
the urban issue
itself) Urban actors
Technology
and Boas (2010), the level of economic development will impact how technology
is organized and used in ICT-enabled governance. A recent comparative study by
Lin (2018) indicates that different institutional and technological contextual factors
have made smart governance in some Western countries relate more to e-governance
and e-democracy, while smart governance in China is focusing more on smart
management and services. Furthermore, it shows that the nature of the urban
issue itself like social, economic and environmental challenges associated with
urbanization also comprises an important contextual factor (Lin 2018; Meijer 2016).
In summary, five contextual factors can be identified: economic, political, cultural,
technological, and the urban issue itself. Nevertheless, Ruhlandt (2018) claims ‘the
influence of contextual factors on smart governance still remains unclear’ and a lack
of empirical studies weakens this connection. Thus, more detailed analyses of smart
governance in different contexts are strongly needed (Meijer et al. 2016).
2.3 Integrating Context into Smart Governance
To examine the potential role of contextual factors in influencing smart governance,
a contextual approach towards smart governance is proposed. Figure 1suggests: (1)
a potential relation between urban context and smart governance arrangements, (2)
that the potential effect of the urban context on smart governance arrangements relies
on the interaction between technology and urban actors, and (3) a potential feedback
effect of smart governance arrangements on the urban context.
3 Methodology
This study is based on a systematic review of the Chinese literature on smart gover-
nance. The search for published journal articles was based on the China Academic
Journals Full-text Database (CJFD)3—one of the most important online academic
3http://gb.oversea.cnki.net/kns55/support/gb/products.aspx.
104 H. Jiang et al.
Fig. 2 Number of articles
by year
1
2
0
4
5
6
9
7
0
2
4
6
8
10
2011 2012 2013 2014 2015 2016 2017 2018
Number
Yea
r
databases in China—covering more than 7672 notable and significant Chinese jour-
nals, across 10 disciplines, from 1996 to the present. Because of its rigorous selection,
the CJFD provides access to journals that have strict academic standards and these
journals are often deemed as the leading journals of natural and social science in
China.
Then, selected keywords—smart governance, smart city governance, smart city
and governance—were used to identify the most relevant articles relating to smart
governance. The search included all works between January 2010 and September
2018, since the rapid growth of the smart city and its governance research started
from 2010 onwards (Dameri and Cocchia 2013). A three-step review process based
on Tofthagen and Fagerström (2010) produced 34 Chinese articles concerning smart
governance. The selected articles span over 8 years, ranging from 2011 to 2018 and
the year 2017 saw the greatest increase of articles (Fig. 2). Among the 34 identified
articles, 9 smart governance projects can be identified. Based on the five identi-
fied urban contextual factors as summarized in Sect. 2, these 9 smart governance
projects can be further categorized into 4 groups (Table 1). These projects have
been researched more in-depth with the help of additional literature, official reports,
research documents, and index systems. For illustration reasons from each group just
one project will be presented below.
The first group is featured with the urban issue itself in defining the vision and
strategies for ensuring the smartness of these cities. Smart governance projects in
these cities (mainly Tianjin, Guangzhou and Chengdu) are supposed to build pilot
areas for future urban development. The second group consists of those cities (mainly
Shenyang and Qiqihar) that once have been important industrial bases under the cen-
trally planned economy but are currently facing restraints from local government’s
political-culture conservatism (e.g. excessive government intervention, centraliza-
tion, and resistance to change) to transform its development. The third group includes
smart governance projects (mainly in Beijing and Shanghai) that are influenced
strongly by political intentions and pressures from the Chinese Central Government.
Smart governance projects in these cities on the one hand are purposed to remove
redundant functions within their organization and improve the delivery of services
to key stakeholders by using ICTs. On the other hand, by facilitating citizens’ trust
towards governments and enhancing the capacity of governments to manage crises
7 Comparing Smart Governance Projects in China: A Contextual … 105
Tabl e 1 Identified smart governance projects in China
Groups/variables Cities Urban contextual
difference
Smart governance
projects
Group 1: Constructing
pilot areas
Tianjin, Guangzhou,
Chengdu
The urban issue itself
Rapid urbanization
and environmental
pollution
Large-scale pilots
and experiments
Tianjin Smar t
Eco-city Project
Guangzhou Smart
Urban Management
Chengdu Wenjiang
Smart Planning and
Management
Group 2: Improving
government
performance
Shenyang, Qiqihar Political-cultural
conservatism
Excessive
government
intervention
Centralization
routineness and
resistance to change
Shenyang Smart
Social Governance
Qiqihar Smart
Eco-city
Management
Group 3: Building
trust in government
Beijing, Shanghai National political
intentions
Facilitating
service-oriented
government
Political dominance
Beijing Changyang
Smart Community
Shanghai Lujiazui
Smart Community
Group 4: Encouraging
innovation
Hangzhou,
Shenzhen,
Innovation economy
and technology
Innovation culture
Technological basis
Human-centric
development
Hangzhou
Shanghang Smart
Governance Project
Shenzhen Nanshan
Smart City
Governance Projects
and to implement plans, it intends to consolidate government’s political dominance.
The fourth group is featured with its innovation culture and strong technological
basis. Home to some of China’s most renowned tech giants from Huawei to Alibaba,
smart governance projects in these cities (e.g. Shenzhen, Hangzhou and Foshan)
heavily rely on their innovation economy and technologies that foster open systems
and platforms through communication and information sharing.
4 Comparing Smart Governance Projects in Different
Chinese Urban Contexts
In this section we focus on four empirical smart governance projects and strive to
investigate how urban context influences smart governance in these cases. These
four projects are taken from the four identified groups and contain Tianjin Smart
106 H. Jiang et al.
Eco-city Project (Wang et al. 2017), Shenyang Smart Social Governance (Jia and Li
2017), Beijing Changyang Smart Community Project (Wang 2015), and Hangzhou
Shanghang Smart Community (Wang 2014).
4.1 Constructing a Pilot Area: Tianjin Smart Eco-City
Project
With Tianjin’s rapid urbanization, a large territory of coastal land in South Tianjin has
been polluted and influenced by industrial wastewater and salinization. This harsh
situation has driven the Chinese government to find new ways to govern these urban
problems. The construction of Tianjin Smart Eco-city is such a response made by the
Chinese government in cooperation with the Singapore government aimed at building
a pilot area—a place under experimentation to provide innovative solutions for urban
development and ecological protection. Covering an area of thirty square kilometers,
the key challenge of this project is to repair a degraded ecosystem and build an eco-
city simultaneously. Specifically, the project focuses on repairing a large territory of
degraded land, constructing affordable housing, building sustainable transportation,
creating good quality jobs, and providing public services and social amenities within
walking distance of residential estates. However, the scope of work and the large-scale
of construction require a large amount of financial investment and strong political
support, which has put government actors at the center of the governance of this
project.
Collaboration between China and Singapore in the Tianjin Smart Eco-city Project
occurs at two levels. At the national government level, a Joint Steering Council and
aJoint Working Committee have established eight working-level sub-committees to
govern the eco-city. At the private level, the Sino-Singapore Tianjin Eco-city Invest-
ment and Development Co., Ltd. is erected by a Singapore consortium and a Chinese
consortium to develop this project. The inclusion of only the governments of China
and Singapore has further determined the design, implementation, and use of tech-
nology. Drawing on the experience in Singapore, an ICT-enabled ‘Systematic City
Management’ model has been developed to realize the governance of this eco-city at
three eco-scales. At the eco-city scale, a comprehensive online platform—the Intel-
ligent Urban Service Platform—is built to realize online declaration of forty-nine
items relating to industrial and commercial registration, project establishment, envi-
ronmental assessment, and other important approval matters. At the eco-community
level, various e-citizen centers are put into use, providing residents with thirty online
service items concerning medicine, food, housing, travel, music and education. And
at the eco-cell level, different management and service teams—Social Work Sta-
tion—are built to master the public sentiment and maintain social stability.
In this project, the political context (i.e. cooperation between the governments of
China and Singapore) in combination with the extensiveness of the urban planning
issue (e.g. degraded land restoration, housing construction, transportation building,
7 Comparing Smart Governance Projects in China: A Contextual … 107
(a) 2007 (b) 2016
Fig. 3 A pilot Tianjin Smart Eco-city Project (source https://www.tianjinecocity.gov.sg/gal.htm)
jobs creation, and so on) constitutes the main urban contexts facing the smart gover-
nance of this eco-city. The effect of these urban contexts has produced a government-
to-government relationship and closed technologies to implement this project. Then,
closed technologies enhance government’s ability in governing these urgent urban
issues. As a result, the direction and activities of this project are much easier to be
carried out and substantive outcomes such as better ecosystem, infrastructures, and
economic vibrancy have been achieved at a prescribed time. However, in Tianjin
Smart Eco-city Project, to restore such a large territory of degraded land and build
a wholly new eco-city simultaneously makes the smart governance process more
rely on national government’s support, which distinguishes itself from other eco-city
projects (Fig. 3).
4.2 Improving Government Performance: Shenyang Smart
Social Governance
As one of the most important old industrial bases under the state-led development
in China, Shenyang government’s excessive market intervention (e.g. encouraging
the survival of inefficient firms, government-led investment, insufficient public ser-
vice, and unfair treatment of private companies) has caused a system of relatively
antiquated values and beliefs within the government system. These antiquated values
and beliefs contribute to the forming of a unique conservatism in political system and
organization culture that Shenyang government has been slow to respond to the con-
sistently changing environment. This conservatism strongly influences the process
of Shenyang Smart Social Governance—a project aimed at improving government
performance and spurring Shenyang’s market vitality.
In this project, a public-private partnership is established between Shenyang gov-
ernment and a local agent company called NEUNN. However, the role of this local
company is only restricted to developing relevant platforms that government requires.
For instance, dedicated to departments interoperability, the Smart Shenyang Col-
laborative Office Platform is developed to allow for data exchange and information
108 H. Jiang et al.
sharing between different governmental components. To enhance government’s abil-
ity of social control, the Digital Smart Management System is designed to combine
transportation monitoring and public safety management with environmental pro-
tection. This platform divides Shenyang into different unit grids so that different
urban management officers can specialize in their own territories. Then, by initiating
the Shenyang Public Service Portal (or Shenyang 12345), Shenyang government
aims to improve its business environment by receiving complaints (e.g. lack of laws
and regulations, traffic noise, environmental pollution) from private companies and
citizens.
Identified as the key urban contextual factor in this project, the political-cultural
conservatism of Shenyang government has made itself prefer to use technologies that
promote a government-to-consumer relationship rather than build two-way commu-
nications between the government and non-government actors. Various government-
steered platforms are only used to strengthen the ability of Shenyang government for
policy-making, service provisions and social control. Although those technologies to
some degree streamline processes and enable certain degree of participation in urban
issues, private sectors and citizens are only allowed to post their ideas, comments,
and requests as what government expects. Instead of promoting an inclusive gov-
ernment, this ICT-driven social governance actually has enhanced what government
considers traditional values or behaviors and the conservatism of Shenyang govern-
ment—a hesitance to share information and power with local citizens to reach its
goal. Attentively, Shenyang Smart Social Governance is a typical smart governance
model representing those old-industry cities.
4.3 Building Trust in Government: Changyang Smart
Community Project, Beijing
As a suburban town located in the Southwest Beijing, Changyang has a 200,000
floating population while the household population is only around 50,000. A large
number of migrants in Changyang have impeded its achievement of a livable town.
More importantly, political pressures and intentions from the Chinese Central Gov-
ernment have largely affected the behavior and role of Changyang government since
central government has a strong ambition to influence the development of Beijing’s
local member districts. They expect that increased satisfaction through better service
delivery can improve citizens’ trust in government on the one hand; and government’s
political authority and dominance can be enhanced through innovative governance
approaches on the other.
Thus, Changyang Smart Community Project is proposed to build a “service-
oriented and facilitating government”, showing its determination of a human-centric
development. Two aspects have been paid special attention. To smarten government’s
ability of urban management and social control, the Public Information Platform of
7 Comparing Smart Governance Projects in China: A Contextual … 109
Smart Changyang has been introduced to break the segmentation between differ-
ent government divisions and provide a one-stop management experience for city
managers. To transform its external organization and build satisfaction and trust in
government, Changyang government has directed the establishment of the Commu-
nity Service Management Platform to create a community governance model for
local Neighborhood Committees. This platform integrates administrative manage-
ment, public affairs and daily services and improves the responsiveness of local
governments and their efficiency and effectiveness of service provision. In addition,
several open-source systems have also been formed to enable e-participation from
civil society in managing their ‘niggling’ daily issues. For instance, Elderly Care
System aims to mobilize the resources distributed among community volunteers,
private companies, and non-government organizations to provide services to elderly
people (e.g. psychological assessment, health monitoring, rehabilitation guidance,
and financial services).
Although it is impossible for the central government to directly intervene in the
governing affairs of Changyang, its political intentions and pressures have been
strongly transmitted to Changyang government, constituting the unique urban context
in Changyang Smart Community Project. To satisfy central government’s require-
ment, a strategy of human-centric development is advanced to transform its internal
and external organization. On the one hand, Changyang government invites techno-
logical companies to establish platforms to improve its ability of urban management
and the quality of services. On the other, they create open technologies for govern-
ment, private sectors and citizens to work together for their daily issues. Improved
services, a more transparent government and allowed participation with the help of
a combined use of open and closed technologies have not only increased the sat-
isfaction and trust of local citizens in Changyang government, but also enhanced
government credibility and authority. Conversely, the increased citizen’s trust and
government credibility and authority have enhanced central government’s political
intentions. Compared to other cities, the dual meaning of smart governance in Bei-
jing (i.e. building trust in government with enhanced government authority) is more
prominent due to central government’s influence.
4.4 Encouraging Innovation: Shangcheng Smart
Governance Project, Hangzhou
As a core urban district of Hangzhou, Shangcheng is among the first batch of Smart
City Pilots initiated by MOHURD. The reason why Shangcheng can become a smart
city pilot is mainly due to Hangzhou’s technology strength. As one of the most
influential hi-tech innovation and hi-tech industry bases in China, Hangzhou is the
headquarters of Alibaba, one of the world’s largest companies specializing in e-
commerce, internet, retail, and artificial intelligence. In addition, major international
tech companies such as Siemens, Motorola and Nokia have also established their
110 H. Jiang et al.
research and development centers in this city. The very active innovation economy
and strong technological basis have escalated Shangcheng to the frontier of exploring
new modes of smart city governance.
Aimed at enhancing its urban competitiveness through innovation and improving
the quality of life for local people, the Shangcheng local government has adopted a
strategy of “Smart Government, Smart Governance, and Smart living” to reach its
goal. Smart government is about using big data and Internet of Things as important
instruments to upgrade infrastructures and improve the efficiency and effectiveness
of decision-making and service delivery in the public sector. Alibaba and Cisco
have become the main partners to control the selection of hardware and software
and facilitate the ICT strategy of Shangcheng government. For instance, City Brain
initiated by Alibaba has replaced conventional road signages in Shangcheng, which
allow digital signages and AR-based messaging systems to monitor the condition of
roads and trigger alerts for any immediate maintenance work.
Smart governance here is about building open-source systems to enable citizen
participation and engagement for concerned issues. For instance, all community
service agencies in Shangcheng district have provided free WIFI, allowing citizens
have access to internet and make comments on service delivery. Shangcheng Safety
365 Platform builds an effective interaction channel and local people can report their
safety concerns. Officers are required to make a response and find an appropriate
solution. Meanwhile, ICT has enabled different stakeholders to collaborate with
each other and build various collaboration innovation spaces. An example of this is
the Wangjiang Youth Innovation Street, where various resources such as financing,
tutor, social, and legal affairs are integrated by an ICT-enabled collaboration network.
This network provides differentiated online and offline services for entrepreneurial
teams and enterprises at their different stages of development. The mass use of social
media, websites and online platforms has created a culture of innovation in this area,
which has incubated more than 200 startup companies.
In this project, the rise of the innovation economy and innovation culture in
Shangcheng has constituted its specific urban context. The effect of these contex-
tual factors has first enabled administrative officers to treat citizens not as observed
subjects but as a source of creation. Different urban actors are allowed to organize
their own governance networks to concurrently draft new connections, ideas and
creations. Then, these contextual factors have led to a multiple use of technolo-
gies, (e.g. closed or open, informing or communicating, and single or complex),
since technology innovation is a major advantage of Hangzhou over other cities.
The various enacted technologies in Hangzhou have speeded up the forming of a
collaborative society between different actors to deal with various urban issues such
as urban innovation, technological development, living conditions, traffic congests,
health and education services, water pollution, and so on. Reversely, these ICT-
enabled collaboration networks produce a positive feedback to Hangzhou’s innova-
tion settings, which has enhanced its overall innovation capability and technological
strength. However, attention should be paid that due to the existence of a large num-
ber of state-owned companies operating in market, government still has a role in the
smart governance process of innovation economy. This differs from what has been
7 Comparing Smart Governance Projects in China: A Contextual … 111
Fig. 4 ET city brain in Shangcheng, Hangzhou (source https://www.alibabacloud.com/et/city)
observed in some Western countries that innovation economy is mainly incubated
among non-government actors [e.g. Helsinki’s City-as-a-platform (Anttiroiko 2016)]
(Fig. 4).
5 Discussion and Conclusion
In this chapter, special emphasis is put on a transformative understanding of smart
governance in China by identifying the impacts of urban contexts on smart gover-
nance arrangement from an urban planning perspective. According to Lin (2018),
smart governance in China is mainly linked with smart management and services.
However, an extension of Lin’s study through an intensive exploration of Chinese
literature on smart governance projects has revealed that smart governance in China
varies significantly. Four types of smart governance—constructing a pilot area,
improving government performance, building trust in government, and encourag-
ing innovation—have been identified through a systematic review of the Chinese
literature on smart governance.
Based on our analysis of four Chinese smart governance projects, we verify that
the effects of urban contexts on Chinese smart governance is mainly through the
interaction between technology and urban actors. In Tianjin Smart Eco-city Project,
the political context along with the urban issue itself have put government and its
agent companies at the center, which further leads to a top-down organization of
technology. In Shenyang Smart Social Governance, the relatively political-cultural
conservatism makes Shenyang government resist opening up the governing process
to outsiders and technologies are preferred to connect their internal organizations
112 H. Jiang et al.
and sectors. In Changyang Smart Community Project, national political pressures
and intentions have encouraged a combined use of open and closed technologi-
cal platforms to improve the connectivity and trust between government and non-
government actors on the one hand, and enhance government’s political authority
and dominance on the other. Finally, strong innovation economy and technological
basis have suggested the existence of a double-sided interaction between urban actors
and technology in Shangcheng Smart Community Project.
Besides, our analysis also indicates that the interaction between technology and
social actors can either reduce or enhance the effects of urban contexts on smart gov-
ernance. For instance, Tianjin Smart Eco-city Project provides comprehensive expe-
riences to fields of fundamental transformation including degraded land restoration,
housing, employment, and so on, which to a large degree eliminates the negative
impacts brought by the challenges of fast urbanization. In Changyang Smart Com-
munity Project, by promoting a service-oriented and facilitating government with
the help of ICTs, an increase in citizen’s trust and government’s political authority
and credibility has enhanced central government’s political intentions. In Shenyang
Smart Social Governance, ICT-driven social governance reinforces the conservatism
of Shenyang government. In Shangcheng Smart Community Project, various ICT-
enabled collaboration networks have optimized Hangzhou’s innovation settings and
enhanced its overall innovation capability and technological strength.
Based on our findings, we conclude that knowledge on the urban context is vital to
understand the expected outcomes of intended smart governance arrangements. An
in-depth understanding of the specificity of urban context will help develop an adap-
tive smart governance arrangement for that situation. However, due to geographical
particularities and differences, a risk can exist when other contextual factors relating
to smart governance have been neglected but may be significant (Ruhlandt 2018).
Thus, more comparative empirical research between different places or cities on
smart governance in the future is needed and will help to identify the diversity of
smart governance modes and their usefulness in dealing with harsh urban issues.
References
Anttiroiko A (2016) City-as-a-platform: towards citizen-centre platform governance. In: Proceed-
ings of the RSA winter conference
Batty M, Axhausen KW, Giannotti F, Pozdnoukhov A, Bazzani A, Wachowicz M, Ouzounis G,
Portugali Y (2012) Smart cities of the future. Eur Phys J Spec Top 214(1):481–518
Bertot JC, Jaeger PT, Grimes JM (2010) Using ICTs to create a culture of transparency: e-
government and social media as openness and anti-corruption tools for societies. Gov Inf Q
27(3):264–271
Bolívar MPR, Meijer AJ (2016) Smart governance: using a literature review and empirical analysis
to build a research model. Soc Sci Comput Rev 34(6):673–692
Caragliu A, Del Bo C, Nijkamp P (2011) Smart cities in Europe. J Urban Technol 18(2):65–82
Dameri RP, Cocchia A (2013, Dec) Smart city and digital city: twenty years of terminology evolu-
tion. In: X Conference of the Italian chapter of AIS, ITAIS, pp 1–8
7 Comparing Smart Governance Projects in China: A Contextual … 113
Giffinger R, Fertner C, Kramar H, Meijers E (2007) City-ranking of European medium-sized cities.
Cent Reg Sci Vienna UT, 1–12
Gil-Garcia JR (2012) Enacting electronic government success: an integrative study of government-
wide websites, organizational capabilities, and institutions, vol 31. Springer Science & Business
Media, Berlin
Hammad R, Ludlow D (2016, Dec) Towards a smart learning environment for smart city governance.
In: Proceedings of the 9th international conference on utility and cloud computing, ACM, pp
185–190
Hao L, Lei X, Yan Z, ChunLi Y (2012, June) The application and implementation research of smart
city in China. In: 2012 international conference on system science and engineering (ICSSE),
IEEE, pp 288–292
Janowski T (2015) Digital government evolution: from transformation to contextualization. Gov
Inf Q 32(3):221–236
Jia L, Li Z (2017) Social governance in the process of smart city construction: a case study of
Shenyang. J Northeast Univ 19(6):608–615
Johnston EW, Hansen DL (2011) Design lessons for smart governance infrastructures. In: Trans-
forming American governance: rebooting the public square, pp 197–212
Kalathil S, Boas TC (2010) Open networks, closed regimes: the impact of the internet on authori-
tarian rule. Carnegie Endowment
Kleinhans R, Van Ham M, Evans-Cowley J (2015) Using social media and mobile technologies
to foster engagement and self-organization in participatory urban planning and neighbourhood
governance
Kourtit K, Nijkamp P, Arribas D (2012) Smart cities in perspective—a comparative European study
by means of self-organizing maps. Innov Eur J Soc Sci Res 25(2):229–246
Lin Y (2018) A comparison of selected Western and Chinese smart governance: the application
of ICT in governmental management, participation and collaboration. In: Telecommunications
policy
Linders D (2012) From e-government to we-government: defining a typology for citizen coproduc-
tion in the age of social media. Gov Inf Q 29(4):446–454
Meijer A (2016) Smart city governance: a local emergent perspective. In: Smarter as the new urban
agenda. Springer, Cham, pp 73–85
Meijer AJ, Gil-Garcia JR, Bolívar MPR (2016) Smart city research: contextual conditions, gover-
nance models, and public value assessment. Soc Sci Comput Rev 34(6):647–656
Meijer A, Bolivar M (2013, Sept) Governing the smart city: scaling-up the search for socio-techno
synergy. In: 2013 EGPA conference proceedings
Meijer A, Bolívar MPR (2016) Governing the smart city: a review of the literature on smart urban
governance. Int Rev Admin Sci 82(2):392–408
Nielsen JA, Pedersen K (2014) IT portfolio decision-making in local governments: rationality,
politics, intuition and coincidences. Gov Inf Q 31(3):411–420
Ruhlandt RWS (2018) The governance of smart cities: a systematic literature review. Cities 81:1–23
Scholl HJ, AlAwadhi S (2016) Smart governance as key to multi-jurisdictional smart city initiatives:
the case of the eCityGov alliance. Soc Sci Inf 55(2):255–277
Shi Y (2018) The model of smart governance of Lujiazui community in Shanghai. In: Lin YL, Wang
SF, Geertman S, Deng ZH (eds) Smart governance and new forms of collaborative planning. The
Commercial Press, Beijing (in press)
Tofthagen R, Fagerstrøm LM (2010) Rodgers’ evolutionary concept analysis—a valid method for
developing knowledge in nursing science. Scand J Caring Sci 24:21–31
Treib O, Bähr H, Falkner G (2007) Modes of governance: towards a conceptual clarification. J Eur
Public Policy 14(1):1–20
United Nations (2018) World urbanization prospects: the 2018 revision
Wang J (2014) Smart city governance to integrate ICT, government functions, and sociality: an anal-
ysis of Hangzhou Shangcheng smart community [in Chinese]. Observation Ponderation 7:50–54
114 H. Jiang et al.
Wang D (2015) Case analysis of smart community: characteristics and contents of smart changyang
[in Chinese]. Geomatics World 22(4):34–38
Wang D, Cui S, Yao X, Chi T (2017) Practice and research on intelligent information platform
of eco-city construction based on integration with sky, space and earth [in Chinese]. Geomatics
World 4:19–23
Yang K (2003) Neoinstitutionalism and e-government: beyond Jane Fountain. Soc Sci Comput Rev
21(4):432–442
Part II
Computational Planning
Chapter 8
A Preliminary Study on Micro-Scale
Planning Support System
Daosheng Sun, Xiaochun Huang, Lianna He, Tengyun Hu and Yilong Rong
Abstract Along with the transformation of urban development and planning in
China, the traditional macro-scale urban planning which takes function delimitation
as the main working mode and land exploitation as the guiding principle has begun
to stretch towards the micro-scale which focuses on the human. This new research
tendency requires a new planning support system. Since there are existing decision
support systems for both meso and macro scale urban planning, this research puts
forward that a micro-scale support system for planning should be constructed. This
system should considerate human’s subjective feelings and needs as the core, and
should be organized in three categories: ecological elements in micro living environ-
ment, urban design elements in local space, and individual behavior elements. Based
on this, three sub-systems are proposed. For the first, the natural environment and
ecological factors involved in micro-scale urban planning are the research objects,
and this sub-system is established according to the logic of “cause-effect-response”
model. Secondly, by integrating quantitative research and evaluation methods for
urban design, the sub-system of urban design and spatial layout provides support for
micro urban design simulation and local area design optimization, from the dimen-
sions of urban perception, spatial morphology and function optimization. Lastly, the
D. Sun (B)·X. Huang ·L. He ·T. Hu
Beijing Municipal Institute of City Planning and Design, Beijing 100045, China
e-mail: sundaosheng@foxmail.com
X. Huang
e-mail: huanxiaochun@bmicpd.com.cn
L. He
e-mail: helianna@bmicpd.com.cn
T. Hu
e-mail: hutengyun88@163.com
D. Sun
School of Architecture, Tsinghua University, Beijing 100084, China
Y. Rong
Beijing City Interface Technology Co. Ltd, Beijing, China
e-mail: 380014813@qq.com
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3-030-19424- 6_8
117
118 D. Sun et al.
sub-system of human behavior and community life is based on the urban space-time
behavioral theory and is applied to urban space-time analysis residential area plan-
ning, functional area planning, traffic planning, etc. These sub-systems will play an
essential role in the future vital work such as central-city-planning, urban-physical-
examination and livable-city-construction.
Keywords Micro-scale planning ·Planning support system ·Human-oriented ·
Planning framework ·Micro-ecology ·Urban design
1 Introduction
Scale is the first problem that must be faced in any type of urban planning. At dif-
ferent scales, the planning tools chosen should also be different. For instance, urban
design applies to small-scale contexts, while some social planning measures such
as tax incentive policies apply to larger-scale contexts (Park and Rogers 2015). The
applicability of planning scale is determined by the planning object and the devel-
opment requirements. Conversely, the planning scale also makes a big difference in
the planning elements and work content.
As Jane Jacobs discussed in her book, various micro-urban spaces such as side-
walks, neighborhood parks and city neighborhoods are related to a series of advan-
tages such as the safety and vitality of the city (Jacobs 1961). Now in China, along
with the transformation of urban development and planning management modes,
the incremental planning is switching into stock planning, the urban exploitation
is switching into urban renewal, and the extensive urban management is switching
into refined urban governance mode. When it comes to the planning scale, former
urban planning with macro-function delimitation as the main job, which is oriented
by large-scale land exploitation, has gradually begun to focus on the micro-scale of
“human” and “daily life” (Zhang and Lu 2003).
In order to adapt to this trend, a new Planning Support System (PSS) needs to be
introduced. It must address complex issues that were seldomly addressed by tradi-
tional macro-analysis methods, such as urban micro-climate, environment control,
local spatial design, and urban behavior simulation analysis. However, traditional
location theory and spatial planning methods often lose their effectiveness on this
scale. At present, the development of multi-disciplinary integration for urban micro
analysis, the ubiquitous development of urban quantitative analysis technology and a
large number of micro-scale urban planning practices have laid a theoretical, method-
ological and technical foundation for planning support of micro-scale planning. The
moment is right to establish a micro-scale urban planning support system to cope
with the change of planning scale (Fig. 1).
8 A Preliminary Study on Micro-Scale Planning Support System 119
Fig. 1 Changes in the theory and methodology of planning support system
2 The Theoretical Basis for Micro-Scale Planning Support
Systems
2.1 Former Work on Planning Support Systems
PSS were first proposed by Harris (1989) in his discussion of use of technology in
planning. Although some discussion of PSS is about how to encourage planners to
participate in planning tasks collaboratively to reach a consensus (Simonovic and
Bender 1996), most PSS researchers focus on how to effectively assist planning
schemes, mainly in land use planning or urban planning (Wang et al. 2014). A large
number of models have been put forward during the development of PSS, some even
tracing back far before the concept came into being, such as the spatial interaction
model founded by Hansen (1959) and developed by Lowry (1964), the discrete-
choice model based on utility theory (Timmermans and Golledge 1990), and the
bid-rent model which derived UrbanSim (Borning et al. 2008); in addition, other
models based on particular methods such as micro-simulation (Landis 1995) and
programming models (Brotchie et al. 2013) have been proposed. From a technical
point of view, GIS as a tool for planning has become one of the most powerful systems
for PSS (Sui 1998). Currently, with the rapid boom of big data and algorithmic
geographies, planners are capable of recognizing and dealing with the city more
efficiently (Kwan 2016), and hence planning support now has a more solid theoretical
and methodological foundation.
Long and Mao (2010) introduced the concept, the aim and the framework of PSS
into China. In recent years, in order to alleviate and optimize urban functions, China
has entered a critical period of several major planning and construction tasks, such
120 D. Sun et al.
Fig. 2 Contents of planning decision support system for meso scale
as planning of the Beijing Urban Vice-Center and the Xiongan New District - which
is not only a huge project but also a test for urbanization reform (Lin 2012,2017).
Meanwhile, massive planning data as well as a huge number of planning-decision
tasks are to be addressed in these projects. Therefore, a scientific decision-making
system for drawing up urban and rural planning scenarios, and for planning manage-
ment, needs to be carried out urgently. To meet the applicative demand in urban and
rural planning, Long et al. (2015) explored the theoretical basis, system construction
and application practice of urban and rural planning decision-making support at a
meso scale, and they considered it to be a useful reference for planners in applying
new techniques in urban and rural planning systems in China. In their theoretical
framework, they included the application support of the delimitation of restricted
construction areas, the comprehensive analysis of planning status, the analysis of
urban spatial development, the analysis of traffic carrying capacity, the coordinated
development of land use and traffic, and the analysis of municipal capacity. It was
one of the first application researches on the PSS in China, and also the first attempt
to summarize the theory, design and overall framework of the system (Fig. 2).
In addition to the support of urban and rural planning at the meso scale level,
based on the application of regional big data, the research on support systems for
regional planning was carried out successively by planners (Rong et al. 2017). In
their opinion, at a macro scale, the new type of data, mainly referring to remotely
sensed data, location-based service (LBS) data and points of interest (POI) data, was
helpful to depict the activities between the cities, which can describe the land use
structure and reflect spatial boundaries. Besides, they thought both new types of data
and traditional data were essential as two parts of the data base in regional research
and planning. By constructing a series of regional research based on big data and
other census data, effective support for regional issues came into being (Fig. 3).
8 A Preliminary Study on Micro-Scale Planning Support System 121
Fig. 3 Contents of planning decision support system for macro scale
2.2 The Planning Elements of Human Orientation
With the aim to construct a micro scale planning support system, how to organize the
framework with micro-elements becomes the primary task. Human beings are the
target of urban planning, especially when it comes to the micro-scale; that is, micro
living space is what has the most direct spatial relationship with human beings,
and subjective feelings and demand satisfaction of people should be included in the
criteria for scientific planning decisions.
To categorize, planning elements can be divided into objective elements and sub-
jective elements, while objective elements can be divided into natural and artificial
ones. Firstly, the micro-climate shaped by the ecological elements in a micro living
environment (natural elements), can affect human physical and mental health, and
thus, activity patterns. Secondly, the artificial elements, which can be interpreted
to be urban design elements in local space, form the places for human activities,
and generate objective opportunities as well as constraints on behavior. Thirdly, the
subjective individual behavior elements (the behavior-choice and the rules behind it)
are the media and mechanism of interaction between people as the subject and the
environment as the object (Golledge and Stimson 1997).
2.2.1 Ecological Factors in Micro Living Environment
In recent years, there is a common view among the studies of human living envi-
ronments that the micro human living environment acts as the basic unit in urban
space (Luo 2004). And ecological elements, as the primary factor in the human living
environment, are becoming more and more significant. They play important roles
as the basis of urban developments and are gradually becoming the core of urban
planning.
122 D. Sun et al.
2.2.2 Micro Urban Design Elements
Livability is the criterion for judging high-quality urban design (Zheng 2017). For a
long time, there has been a lack of definitions and contents of urban design. However,
a series of documents issued by the central government of China along with pilot
urban design work in many places promoted by the Ministry of Housing and Urban-
Rural Development of the People’s Republic of China, have clarified the mission
of urban design and put forward its working requirements. It is pointed out that
the object of urban design is “detailed design of urban pattern, spatial environment,
architectural scale and style” (Song 2016).
2.2.3 Micro Individual Human Behavior Elements
Space-time behavior research based on time geography and behavioral geography
has become an important research method in urban geography, urban planning and
traffic planning. Its research perspective adopts a non-aggregate approach, paying
attention to individual differences, and it has strong applicability for micro-scale
urban planning. In recent years, the study of urban space-time behavior has begun
to combine with Chinese studies, carried out around many urban and social issues
such as daily life, quality of life, social justice, low-carbon society, smart cities and
so on. Practical work has begun in the fields of urban transportation, tourism and
urban planning (Chai and Tana 2013).
3 Research Methods
Based on the above analysis of urban micro-planning elements, this study intends
to divide the micro urban planning support system into three sub-systems: natural
environment and micro-ecology, urban design and spatial layout, individual behavior
and living space.
In each sub-system, there are both fully explored research points with ample case
studies to depend on, and underexplored research points with limited literature basis
but with practical value. Differentiated research methods are adopted for the two
kinds of research points.
For the former, the main method is to classify and summarize the directions to
compose the system, based on existing work or literature. While for the later, new
research directions will be created for there is no complete and modeled research
paradigm, and limited case studies can be referred to, but these directions are essential
for practical work.
The aim of the study is to build a four-leveled hierarchical framework system to
explain the details of the micro planning support system. To build such a framework,
we mainly think about what elements are included or what we can do in local urban
spaces such as the neighborhood and the street, and maybe also spaces of scales
8 A Preliminary Study on Micro-Scale Planning Support System 123
Fig. 4 Urban micro planning support system researching method
a slightly larger than that. The first-level is defined by the classification of three
types of elements. The second-level is classified according to the logic of natural
environment elements, urban design dimensions and behavior planning domains.
The third-level classification is constructed according to the features of each subject
separately, including a Pressure-State-Response (P-S-R) model, urban design key
points and planning workflow supported by behavioral research. Lastly the fourth
level depicts the concrete research key points (Fig. 4).
4 Sub-system of Natural Environment and Micro-Ecology
The urban climatic condition differs from normal rural areas and the magnitude of
difference could be quite large (Taha 1997). To model the micro-climate character-
istics for urban planning and building design, scholars used to simulate the climate
change at a micro-scale spatial resolution (5 m) and a short-time temporal resolution
(hourly) (Loibl et al. 2011). It is necessary to understand the urban climate patterns
at a micro scale and to help decision-makers make decisions to facilitate the living
conditions of the inhabitants (Jha et al. 2015).
In this sub-system, the related theories of urban ecology and sustainable devel-
opment are synthetically applied, that is, we not only cover the natural elements at
a micro scale in this sub-system, but also organize or structure them, aiming for a
positive environmental effect. This sub-system should combine spatial information
124 D. Sun et al.
technology and computer digital simulation experiments. Taking the natural environ-
ment and ecological factors involved in micro-scale urban planning as the research
object, it provides a framework for the harmonious development of human, society,
economy and natural environment in urban planning.
4.1 The Catalogues of Sub-system of Natural Environment
and Micro-Ecology
There are so many factors in the micro-climate analysis that it is a tough job to clar-
ify the relationship of each factor. A Pressure-State-Response (P-S-R) logical model
makes the job much easier. The P-S-R model was first proposed by Organization
for Economic Cooperation and Development (FAO 1997). It is indeed a concep-
tual framework of evaluation index system, but it also shows a very clear causal
relationship between natural elements (Zuo et al. 2003).
According to the categories of natural elements, the second level catalogues of geo-
logical environment, geomorphological environment, climate environment, hydro-
logical environment and acoustic environment are formed. Then following the logic
of the P-S-R model, describing the cause mechanism, negative effect and regulatory
response strategy of micro-environmental problems, the third level catalogues are
set up. The fourth level catalogues demonstrate the detailed contents of the research.
Taking climatic environment factors as an example, in the step of “pressure” analysis,
the effects of micro-wind direction, temperature and humidity, sunshine conditions
impact on spatial pattern, building layout, building lighting, building thermal per-
formance and aging are discussed, and then in the next step of “state” analysis, the
harmful effects of the above-mentioned effects on human living environment are
discussed. Finally, as a responsive measure, we can put forward planning measures
from the perspective of energy structure and wind environment design (Table 1).
4.2 Key Models of Sub-system of Natural Environment
and Micro-Ecology
Since McHarg’s influential book Design with Nature (McHarg 1969), the analysis
of natural environment and micro-ecology has been an important part of preliminary
research of urban planning. Nowadays, urban design and planning based on ecology
is widely recognized (Frederick et al. 2016). In order to improve the sustainability
of urban micro-ecology, it is necessary to focus on the relationship between local
population and urban function and micro-environmental capacity, especially for the
recent work in Beijing and other major cities in China such as population alleviation,
functional decay, enriching white space and green space (Lu 2017). That is what the
key models should focus on.
8 A Preliminary Study on Micro-Scale Planning Support System 125
Tabl e 1 Catalogues of sub-system of natural environment and micro-ecology
First-level catalogue: sub-system of natural environment and micro-ecology
Second level
catalogues
Third level
catalogues
Fourth level catalogues
Geomorphologic
environment
Environmental
status assessment
Impact of surface morphology on urban layout
Impact of surface morphology on solar radiation
Impact of surface morphology on temperature and
humidity
Impact of geomorphologic process on location
selection
Environmental
hazards
Ground hardening
Environmental
maintenance
measures
Effects of plant photosynthesis and transpiration on
thermal environment
Effects of greening on temperature and ventilation
Effects of greening on air purification and noise
control
Relationship between real estate and urban green
environment
Benefits of green space for residents’ leisure and
recreation
Climatic
environment
Environmental
status assessment
Impact of wind conditions on urban spatial pattern
Assessment of the impact of wind conditions on
building morphology
Suitability evaluation of temperature
Evaluation of the influence of sunshine on building
lighting
Impact of air humidity on thermal performance and
aging of buildings
Effect of air humidity on human comfort
Impact of precipitation conditions on urban
distribution
Environmental
hazards
Temporal variation characteristics of air pollutants and
pollution sources
Natural factors of air pollution
Anthropogenic factors of air pollution
Urban light pollution
Urban heat island effect
Environmental
maintenance
measures
Urban energy structure
Design of urban ventilation environment
(continued)
126 D. Sun et al.
Tabl e 1 (continued)
First-level catalogue: sub-system of natural environment and micro-ecology
Second level
catalogues
Third level
catalogues
Fourth level catalogues
Hydrological
environment
Environmental
status assessment
Surface runoff estimation
Estimation of water environmental capacity
Environmental
hazards
Water pollution source
Risk of urban rainstorm and waterlogging disasters
Environmental
maintenance
measures
Preventive measures for urban waterlogging disasters
Acoustic
environment
Environmental
hazards
Traffic noise
Regional environmental noise
4.2.1 Micro Models for Environmental Capacity Assessment
The main function of the models is to evaluate the possible impact of current micro-
ecological conditions on the construction or renewal of local urban areas. It analy-
ses the impact of groundwater quantity, quality, outflow location and depth on the
exploitation of urban land and the stability of buildings, the impact of the angle and
intensity of solar radiation affected by surface morphological factors such as slope
and aspect on the layout and engineering of buildings in cities, and the impact of
wind speed, wind direction and sunshine conditions on the orientation, spacing of
buildings and the road, etc. A series of assessing methods including net primary pro-
ductivity, ecological footprint, supply and demand balance, or comprehensive index
evaluation could be applied (Chen et al 2018).
4.2.2 Micro Models for Environmental Effect Analysis of Construction
Activities
The purpose of the models is to evaluate the negative environmental effects of local
construction mode, intensity and speed. For example, land subsidence, soil erosion,
vegetation destruction and destruction of ecological environment continuity caused
by the construction of local areas; air pollution, carbon emissions, light pollution,
heat island effect caused by human activities in local areas; increase of waterlogging
points in local areas caused by the construction of transportation system and ground
elevation design, etc. A regression method or system dynamics method could be
used in analyzing or simulating the impact of construction activities on urban micro
ecology.
8 A Preliminary Study on Micro-Scale Planning Support System 127
4.2.3 Micro Models for Improvement of Ecological Environment
These models are mainly aimed at supporting relevant maintenance measures to
promote positive effects on the environment—for example, the analysis of the eco-
logical service function of green space such as the improvement of air conditions,
or the layout of streets and building groups (according to different climatic con-
ditions) to improve ventilation and sewage disposal. A data envelopment analysis
(DEA) method can be used to evaluate the performance of urban micro ecological
construction (Charnes et al. 1978).
5 Sub-system of Urban Design and Spatial Layout
Urban design is divided into different scales according to the different objects of
study, and the micro urban design is focused on a section of an area or a street (Yang
2009). The micro scale urban design takes more people’s spatial experience into
account, for instance through urban legibility and design for a barrier-free environ-
ment (Barman 2004). Yet, the discussion on the different scales in urban design is
not just about the physical space itself, it is also about the change of the concept, the
characteristic and the social relevance of the space (Jiang 2003).
Under this theme, the micro-urban space, such as streets, residential areas, square
parks, important buildings and so on, is taken as the target object, and the comfort of
the space is taken as a standard. By means of morphological analysis, spatial analysis,
image analysis and quantitative evaluation methods of urban design, the sub-system
provides effective support for micro urban design simulation and evaluation and
optimization of regional urban design in urban renewal.
5.1 The Catalogues of Sub-system of Urban Design
and Spatial Layout
Traditionally, urban design mainly paid attention to the urban form (Lan 2004).
But when it comes to the micro scale, obviously more aspects should be covered.
This leads to the question—how to organize the structure of the sub-system of urban
design and spatial layout? The six dimensions of urban design, namely visual dimen-
sion, perceptual dimension, social dimension, functional dimension, morphological
dimension, and temporal dimension, have been fully supported and widely validated
by scholars (Carmona et al. 2012), which can be used as a framework to cover the
research contents of urban design. Through the reorganization and merging of repeti-
tive research elements, three directions of quantitative research on micro-scale urban
design are determined: urban perception, spatial form and function optimization,
128 D. Sun et al.
which are regarded as the second level catalogues. Then, the third level catalogues
are subdivided according to the elements of urban design (Table 2).
5.2 Key Models of Sub-system of Urban Design and Spatial
Layout
Although urban design has a long history of research, quantitative urban design has
been for long at a very lagging stage, when some indicators invented by Lynch (1960)
for quantitative description of urban space were chronically followed by scholars.
Not until recently are the advanced analytical tools such as semantic differential (SD),
GIS, public open space tool (POST) and so forth are applied (Niu et al. 2017). So, the
key models of urban design and spatial layout should focus more on the synthesis of
the quantitative methods, through which the basic paradigm of micro-urban design
can be quickly confirmed. Therefore, from the three dimensions of urban perception,
spatial form and function optimization, the model framework that can cover most of
the micro-scale urban design elements can be determined.
5.2.1 Micro Models of Urban Perception
These models are composed of a series of analyses based on the human’s perception
of urban design elements. For example, the perception analysis of social factors
such as urban public opinion, urban cultural atmosphere, the perception analysis of
urban skyline, architectural outline, visual path and other urban interface elements,
the perception analysis of landscape design elements such as street space quality,
micro built environment and so on. Generally, the perception of urban design can
be divided into the subjective and the objective (Tang et al. 2016). For the former,
revealed preference (RP) and stated preference (SP) surveys (Teddy et al. 2018)
can help to find the subjective preference of people, while for the later, the image
segmentation and recognition makes it possible to assess the environment objectively
(Rundle et al. 2011).
5.2.2 Micro Models of Spatial Form
Models of this dimension carry out analyses on the spatial form involved in urban
design for quantitative design support. For example, to make judgements on spatial
form development trends from the local space, to analyze the urban internal form
from the two-dimensional perspective of land use or three-dimensional perspective of
buildings, to analyze the street spatial form from the welting rate, integration degree,
greening rate and other quantitative indicators, and to support the architectural design
8 A Preliminary Study on Micro-Scale Planning Support System 129
Tabl e 2 Catalogues of sub-system of urban design and spatial layout
First-level catalogue: sub-system of urban design and spatial layout
Second level
catalogues
Third level
catalogues
Fourth level catalogues
Urban
perception
Social
perception
Analysis of urban public opinion
Analysis of urban regional characteristics
Urban
interface
perception
Quantitative analysis of skyline
Analysis of architectural contour line
City height analysis
Visual path analysis
Analysis of urban openness
Analysis of urban night lighting
Quantitative control of main tone
Landscape
design
perception
Street space quality analysis
Quantitative analysis of street landscape vision
Visual entropy analysis of pedestrian street
Quantitative analysis of urban design quality
Quantitative analysis of environmental elements of scenic
spots
Assessment and analysis of built environment
Spatial form Spatial
development
form
Analysis of urban development trend
Analysis of urban development evolution
Prediction and analysis of urban development boundary
Urban interior
form
Evaluation of land use form
Analysis of three-dimensional urban form
Spatial distribution analysis of urban parks
Street space
form
Analysis of street wiring rate
Analysis of street space integration degree
Quantitative analysis of street greening
Morphological analysis of network space
Architectural
space form
Analysis of integration degree of horizon in buildings
Quantitative evaluation of architectural design
Function
optimization
Life function
optimization
Urban business analysis
Functional analysis of urban green space
Analysis of urban leisure function
Analysis of urban river index
Analysis of environmental attraction of residential areas
(continued)
130 D. Sun et al.
Tabl e 2 (continued)
First-level catalogue: sub-system of urban design and spatial layout
Second level
catalogues
Third level
catalogues
Fourth level catalogues
Traffic
function
optimization
Analysis of urban traffic nodes
Analysis of influencing range of traffic station
Traffic path analysis
Street vitality analysis
Assessment of street walking environment
Accessibility analysis of public space
Street accessibility analysis
Security
function
optimization
Functional evaluation of historic blocks
Evaluation of social sharing of public resources
Evaluation of street crime prevention and control
from the analysis of internal space form. As for the methodology, the space-syntax
proves useful for reading the space and depicting the spatial form objectively (Önder
and Gigi 2010).
5.2.3 Micro Models of Function Optimization
In this dimension, the models take the functionality of space as the design object,
aiming at analyzing and optimizing the various functional elements. For example,
the optimization and promotion of local business, leisure, living and other life func-
tions, the optimization and promotion of sites, routes, accessibility on the aspect of
traffic functions; optimization and promotion of public resources sharing and crime
prevention in safety functions. GIS combined with a data base composed of land use
data or some other urban micro activity data plays as the foundation of such a model.
6 Sub-system of Individual Behavior and Living Space
The broad sense of behavioral geography used to mean the trends of rebellion on
classical location theory which neglected the subjective mechanism of economic
production (Dennison 1939). But it was a macroscopic location theory rather than a
microscopic one for no individual person was considered then. In 60–70s of the last
century, the introduction of behavioral science helped to focus on individual behavior
(Liu 1992). By the concept of time geography and space-time-prism, Hägerstraand
(1970) elaborated how an individual person carried out his daily behavior under
the chances and constraints of the urban environment. A series of spatial analysis
8 A Preliminary Study on Micro-Scale Planning Support System 131
methods (Sherman et al. 2005) as well as spatial choice models (Timmermans 1991)
have enriched the application research of individual behavior in living space planning.
Nowadays, a lot of studies have shown that the behavior mechanism can explain the
relationship between micro environment and individuals’ spatial choice (Kim et al.
2014), by which feedback on planning measures could be drawn.
Based on the theory of time geography and behavioral geography, focusing on
the micro-daily behavior of urban residents, this sub-system uses the methods of
visualization and restriction analysis, descriptive statistics, activity space depiction,
spatial discrete choice and micro-simulation to provide support for urban space-time
analysis, residential planning, functional area planning and traffic planning.
6.1 The Catalogues of Sub-system of Individual Behavior
and Living Space
The framework of this sub-system is organized according to the domain division of
space-time behavior research, which can be reflected in the second level catalogues,
sorted from basic feature analysis to practical support of specific planning tasks.
Among them, the basic feature analysis models can cover individual or group behav-
ior patterns, ranging from space analysis to time analysis, while the work support
model in specific fields is organized according to the workflow of planning. Thus,
the sub structures including the third level catalogues and fourth level catalogues of
the model system are formed (Table 3).
6.2 Key Models of Sub-system of Individual Behavior
and Living Space
Although a rich methodology to combine behavioral analysis and urban planning
has been founded by the pioneer researchers, the application case studies are quite
limited until now. It is a primary task to establish a model system to provide a new
perspective for microscopic urban space research and stock planning. Firstly, basic
analysis of space-time behavior should be conducted; secondly, various planning
areas as land use planning, residential planning, industry zone planning and public
facility planning are exploited with behavioral methods.
6.2.1 Micro Models for Urban Space-Time Analysis
These models mainly carry on the characteristic analysis by analyzing the resident
space-time data, to comprehensively evaluate the city from various aspects. For
example, by applying methods of activity space, anchor structure, space-time path,
132 D. Sun et al.
Tabl e 3 Catalogues of sub-system of individual behavior and living space
First-level catalogue: sub-system of individual behavior and living space
Second level catalogues Third level catalogues Fourth level catalogues
Urban space-time
analysis
Space-time analysis of
residents’ behavior
Activity space definition
Space-time path simulation
Comprehensive space-time
evaluation of urban
Assessment of health level
Assessment of walkability
Assessment of pollution/risk
level
Assessment of facilities
opportunity
Land use planning
support
Analysis of land capacity Assessment of land capacity
Determination of spatial pattern
of land exploitation
Spatial distribution simulation
of land exploitation
Zoning control of exploitation
intensity
Determination of land
exploitation intensity
Analysis of land use proportion
Determination of plot ratio
Residential planning
support
Site selection and planning area
delimitation
Residential site selection based
on residential choice
Residential area delimitation
based on daily living sphere
Determination of spatial range
and size
Determination of the size of
residential quarters
Industrial zone planning
support
Site selection and planning area
delimitation
Industrial zone site selection
based on employment choice
Facilities supplying in industrial
zones
Estimation and optimization of
facility demand in industrial
zones
Public service facility
planning support
Commercial facility planning
support based on shopping
behavior
Opening time management
based on shopping time decision
Site selection of commercial
facilities based on shopping
destination choice
Leisure facility planning support
based on leisure behavior
Determination of the scale of
leisure facilities
Determination of the type of
leisure facilities
Green space system
planning support
Park planning support Pedestrian flow simulation of
parks
Spatial organization of parks
Green space system planning
support
Forecast of green space using
demand
(continued)
8 A Preliminary Study on Micro-Scale Planning Support System 133
Tabl e 3 (continued)
First-level catalogue: sub-system of individual behavior and living space
Second level catalogues Third level catalogues Fourth level catalogues
Spatial organization of green
space system
Disaster prevention
planning support
Disaster prevention planning
support for public space
Assessment of disasters based
on daily behavior
Pedestrian flow prediction
simulation in disasters
Emergency shelter planning Site selection of emergency
shelter
Spatial organization of
emergency shelters
Traffic planning support Simulation of comprehensive
planning of transportation
facilities and land use
Traffic demand management
based on travel behavior
Site selection of traffic stations
Evaluation of traffic planning
basedontravelbehavior
Transport policy evaluation
Traffic facilities evaluation
living sphere and time rhythm can be depicted, to evaluate the daily life range, behav-
ior ability and accessibility for individuals or groups. They also help to extract the
urban occupational-residential pattern, or the specific functional space and the time
character, and to make assessments on the city health level, pedestrian level, pollu-
tion or risk level, carbon emission level or facility opportunities based on behavior
characteristics. Individual mobility is the core of urban space-time analysis, and GIS
technology together with new types of big data (GPS, cell-phone, smart card or WIFI)
offers the capability for insight into the space-time regularity of the city (Pappalardo
et al. 2015).
6.2.2 Micro Models for Land Use Planning
Activity-based models provide references to formulate a land exploitation strategy
from a disaggregate perspective. For example, to clarify the conversion relationship
between urban activity intensity and land capacity and to evaluate the land exploita-
tion potential according to residents’ activity; to determine the plot ratio scheme by
considering the avoidance of traffic overload according to the road network and the
travel demand. On one hand, a spatial choice model can analyze the influence of land
use on the travel behavior, which can provide feedback for land use planning (Zhang
2004). On another hand, micro simulation method, taking multi-agent system (MAS)
for example, is a good way to understanding future land use patterns from individual
perspectives (Xue and Yang 2003). Models down below are indeed derivatives of the
land use planning model, so the same methods can be also suitable for them.
134 D. Sun et al.
6.2.3 Micro Models for Residential Planning
Based on the residential behavior of community residents, the site selection, spa-
tial design and facilities supporting residential areas are determined. For example,
community site selection can be based on residential choice behavior, and planning
area can be based on the daily living sphere; in addition, ancillary service facilities
planning and space optimization can be carried out based on the measurement of the
usage demand on public service facilities.
6.2.4 Micro Models for Industry Zone Planning
Based on the work behavior and commuting patterns of urban residents, the spatial
layout of urban industrial functions, service facilities and transportation planning
are supported. For example, to provide reference for the site selection of industrial
zones according to people’s preferences in choosing places of work; to measure the
demand for facilities in the places of work by investigating people’s activities using
them and provide reference for facilities matching; to arrange transport facilities
such as parking lots, public transport facilities regulations, bus routes and stations
reasonably based on people’s travel preferences.
6.2.5 Micro Models for Public Service Planning Support
It includes the planning support for commercial facilities based on shopping behavior
and leisure facilities based on leisure behavior. For instance, customizing the dynamic
operation time of facilities based on the difference of people’s decisions on the time of
shopping; guiding the spatial layout planning and design of commercial space such
as shopping blocks and commercial complexes according to people’s wandering
behavior in shopping; and determining the proportion of leisure facilities according
to leisure types.
7 Summary
Based on literature review of previous studies, this article discusses how PSS could
be applied at the micro scale. The research is with both theoretical basis and appli-
cation orientation. On the aspect of theory, it clarifies the significance of planning
support systems on the micro scale for current planning transformation, and it forms
a theoretical framework based on the integration of micro ecological theory, micro
urban design theory and micro space-time behavior theory, which extends and sup-
plements planning support system to the micro-scale. Regarding application, the
research of the micro models aims to form a technical reference framework, to help
planners dealing with problems in local urban space, especially when concurrently
8 A Preliminary Study on Micro-Scale Planning Support System 135
faced with missions of central city planning, urban physical examination, livable city
construction and other key work. Besides, representative micro models are proposed
in each sub-system, which could be pilot points in the construction of this system
in the future. Still, there is a long way to go to move this framework into a tangible
and operational system, with a lot of algorithm development and programming. And
many empirical researches will be essential for that.
References
Barman J (2004) Improving urban legibility and design for barriers free environment: an approach
towards micro scale urban design. In: Conference on accessibility, Kolkata
Borning A, Waddell P, Förster R (2008) Urbansim: using simulation to inform public deliberation
and decision-making. In: Digital government. Springer, Boston, pp 439–464
Brotchie JF, Dickey JW, Sharpe R (2013) TOPAZ: general planning technique and its applications
at the regional, urban, and facility planning levels, vol 180. Springer Science & Business Media,
New York
Carmona M, Heath T, Oc T, Tiesdell S (2012) Public places—urban spaces. Routledge, London
Chai Y, Na T (2013) Progress in space-time behavior research in China. Prog Geogr
32(9):1362–1373
Charnes A, Cooper WW, Rhodes E (1978) Measuring the efficiency of decision making units. Eur
J Oper Res 2(6):429–444
Chen T, Yang M, Xu J, Xie Q, Fang Z, Ye H et al (2018) A summary of ecological carrying capacity
evaluation model. J Green Sci Technol 8:7–9
Dennison SR (1939) The location of industry and the depressed areas. Oxford University Press,
Oxford
FAO (1997) Land quality indicators and their use in sustainable agriculture and rural development.
In: Proceedings of the workshop organized by the land and water development division FAO
agriculture department, 2, p 5
Frederick RS, George FT, Armando C (2016) Nature and cities: the ecological imperative in urban
design and planning. Lincoln Institute of Land Policy, Cambridge
Golledge RG, Stimson RJ (1997) Spatial behavior: a geographic perspective. Econ Geog
74(1):83–85
Hägerstraand T (1970) What about people in regional science? Pap Reg Sci 24(1):7–24
Harris B (1989) Beyond geographic information systems. J Am Plann Assoc 55(1):85–90
Hansen WG (1959) How accessibility shapes land use. Am Inst Planners 25:73–76
Jacobs J (1961) The death and life of great American cities: the failure of town planning. Random
House Inc, New York
Jha M, Marpu PR, Chau CK, Armstrong P (2015) Design of sensor network for urban micro-climate
monitoring. In: Smart cities conference
Jiang Y (2003) On scale in city design. Planners 19(5):53–54, 63
Kim S, Park S, Lee JS (2014) Meso- or micro-scale? environmental factors influencing pedestrian
satisfaction. Transp Res Part D Transp Environ 30:10–20
Kwan MP (2016) Algorithmic geographies: big data, algorithmic uncertainty, and the production
of geographic knowledge. Ann Assoc Am Geogr 106(2):274–282
Lan B (2004) Urban traffic and urban form–an unavoidable topic in urban design. Eng J Wuhan
Univ (Eng Ed) 37(2):173–175
Landis JD (1995) Imagining land use futures: applying the California urban futures model. J Am
Plann Assoc 61(4):438–457
136 D. Sun et al.
Lin X (2012) The connotation, orientation and construction pattern of Tongzhou district as vice-
capital center. Urban Rural Dev 11:10, 33–35
Liu J (1992) A preliminary study of behavioral location theory. Hum Geog 3:40–47
Liu R (2017) Understanding the fundamental strategic intention of building Xiongan new district.
People’s Tribune 12:44–45
Loibl W, Tötzer T, Köstl M, Züger H, Knoflacher M (2011) Modelling micro-climate characteristics
for urban planning and building design. In: International symposium on environmental software
systems. Springer, Berlin, pp 605–617
Long Y, Mao Q (2010) Definition, objectives and framework of urban planning support system. J
Tsinghua Univ Sci Tech 50(3):335–337
Long Y, Huang X, He L, Cheng H, Yu W (2015) Construction and application of planning support
system. Beijing Plann Rev 2:24–27
Lowry IS (1964) A model of metropolis. RAND Corporation, California
Lu M (2017) Multi measures simultaneously, Beijing will be more livable in the future. Urban Rural
Dev 24:18
Luo Z (2004) A preliminary study on hierarchy evolvement law in microform of residential envi-
ronment system—differentiation and reframe of basic unit. Planners 20(10):95–99
Lynch K (1960) The image of the city. MIT press, Cambridge
McHarg IL (1969) Design with nature. American Museum of Natural History, New York
Niu Q, Yan J, Xia Y (2017) An overview of the quantitative analysis methods in urban design.
Urban Plann Int 32(6):61–68
Önder DE, Gigi Y (2010) Reading urban spaces by the space-syntax method: a proposal for the
South Haliç Region. Cities 27(4):260–271
Pappalardo L, Pedreschi D, Smoreda Z, Giannotti F (2015) Using big data to study the link between
human mobility and socio-economic development. In: 2015 IEEE international conference on
big data, pp 871–878
Park Y, Rogers GO (2015) Neighborhood planning theory, guidelines, and research: can area,
population, and boundary guide conceptual framing? J Plann Lit 30(1):18–36
Rong Y, Huang X, Wang B (2017) Quantitative identification of urban functional areas based on
big data. Beijing Plann Rev 5:95–101
Rundle AG, Bader MD, Richards CA, Neckerman KM, Teitler JO (2011) Using Google street view
to audit neighborhood environments. Am J Prev Med 40(1):94–100
Sherman JE, Spencer J, Preisser JS, Gesler WM, Arcury TA (2005) A suite of methods for repre-
senting activity space in a healthcare accessibility study. Int J Health Geogr 4(1):24
Simonovic SP, Bender MJ (1996) Collaborative planning-support system: an approach for deter-
mining evaluation criteria. J Hydrol 177(3–4):237–251
Song Z (2016) On suggestions for improvement of management over urban planning construction.
Shanxi Archit 43(27):222–223
Sui D (1998) Gis-based urban modelling: practices, problems, and prospects. Int J Geogr Inf Sci
12(7):651–671
Taha H (1997) Urban climates and heat islands: albedo, evapotranspiration, and anthropogenic heat.
Energy Build 25(2):99–103
Tang J, Long Y, Zhai W, Ma Y (2016) Measuring quality of street space, its temporal variation and
impact factors: an analysis based on massive street view pictures. New Archit 5:110–115
Teddy L, Siva S, Eric M, Amer S (2018) Subway user behaviour when affected by incidents in
Toronto (subwait) survey: a joint revealed preference and stated preference survey with a trip
planner tool. Can J Civ Eng 45(8):623–633
Timmermans H (1991) Decision-making processes, choice behaviour, and environmental design:
conceptual issues and problems of application. In: Environment, cognition, and action: an inte-
grated approach, pp 63–77
Timmermans H, Golledge RG (1990) Applications of behavioural research on spatial problems II:
preference and choice. Prog Hum Geogr 14(3):311–354
8 A Preliminary Study on Micro-Scale Planning Support System 137
Wang H, Shen Q, Tang BS (2014) A review of planning support systems for urban land use planning.
In: Proceedings of the 17th international symposium on advancement of construction management
and real estate, pp 233–248
Xue L, Yang K (2003) Research on urban evolution using agent-based simulation. Syst Eng Theor
Pract 23(12):1–9, 17
Yang M (2009) Ecological layout and design strategies at the micro-level of urban design: tak-
ing the urban area of Baiyun international conference center in Guangzhou as an example. In:
International forum on urban development and planning, pp 39–45
Zhang M (2004) The role of land use in travel mode choice: evidence from Boston and Hong Kong.
J Am Plann Assoc 70(3):344–360
Zhang J, Lu J (2003) From mega-urban project design towards daily living space. City Plann Rev
27(9):40–45
Zheng S (2017) Micro urban design should aim at livability. Beijing Obs 2:62
Zuo W, Zhou H, Wang Q (2003) Conceptual framework for selection of an indicator system for
assessment of regional ecological safety. Soils 35(1):2–7
Chapter 9
Geodesign—A Tale of Three Cities
Christopher Pettit, Scott Hawken, Carmela Ticzon and Hitomi Nakanishi
Abstract In this paper we discuss the application of the Steinitz (A framework
for geodesign: changing geography by design. ESRI, Redlands, CA, 2012) Geode-
sign Framework in the context of three cities including (i) South East Sydney, (ii)
the emerging Western City of Sydney and (iii) Canberra. In all three of these case
studies we have used the Geodesign Hub platform to develop a series of future city
scenarios. A common theme with each of these cities is they are all experiencing
population growth. Another common theme is that each city required integrated land
use transport planning given new transformational infrastructure including light rail,
mass transit and in the case of the Western City of Sydney a new airport being built.
The research conducted is reflective and based on case studies in the context of studio
work undertaken by three different Geodesign classes run across two universities.
The research reflects on the strengths and opportunities of the Geodesign Framework
in supporting the planning and design of future cities in the context of (i) data and
technology, (ii) process, and (iii) outputs. Future work will examine the pedagogical
experiences of students in working with Geodesign methods and software as we train
the next generation of city planners and designers.
Keywords Geodesign ·Sustainable urban development ·Urbanization
C. Pettit (B)·S. Hawken ·C. Ticzon ·H. Nakanishi
Built Environment, University of New South Wales, Sydney, NSW 2052, Australia
e-mail: c.pettit@unsw.edu.au
S. Hawken
e-mail: s.hawken@unsw.edu.au
C. Ticzon
e-mail: c.ticzon@unsw.edu.au
H. Nakanishi
e-mail: hitomi.nakanishi@canberra.edu.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_9
139
140 C. Pettit et al.
1 Introduction
We are experiencing significant population growth in many parts of the world
and many of our cities continue to rapidly urbanise to accommodate this growth
(Geertman et al. 2017).
Australia provides a microcosm of this phenomena as it is a highly urbanised
country with a population over 24 million people and 85% living in cities. Australia’s
populate is projected to significantly increase over the coming decades and hence,
as in many places in the world, there is a critical need to explore future urban growth
scenarios to assist planners and decision-makers in supporting sustainable urban
development.
Digital planning tools known as planning support systems (PSS) provide the
opportunity to collect, analyse and synthesise city data and importantly engage across
government, industry, and the community to collaboratively design and plan sustain-
able urban futures (Pettit et al. 2018). In this chapter we focus on the application
and reflection of the Geodesign framework (Steinitz 2012) (Fig. 1). The framework
supports a systems view approach for understanding place and space. The framework
as iterative allowing a series of WHAT,WHERE and WHEN critical questions to be
asked with data and metrics to be collected and run through a series of models (rep-
resentation, process, evaluation, change, impact and decision) to create and evaluate
sustainable urban futures.
Fig. 1 Geodesign Framework with 6 questions and 6 corresponding models (Source Steinitz 2012)
9 Geodesign—A Tale of Three Cities 141
In the context of this research we have used the supporting Geodesign Hub plat-
form (Ervin 2011) to create end evaluate a number of sustainable urban development
scenarios across three cities in Australia; (i) South East Sydney, Western Sydney and
Canberra. In the following sections we describe a tale of three cities—specifically,
how Geodesign has been applied in the context of studio work undertaken across two
universities in Australia. A reflection on these three case studies from the perspective
of data and technology, process, and outputs is then provided. The chapter concludes
with recommendations for future research into the emerging field of Geodesign.
2 Methodology—Case Study Approach
The Geodesign approach is unique in that it enables rapid, model-based and con-
tinuous feedback on multiple aspects of performance to improve design-in-progress
scenario proposals (Albert and Vargas-Moreno 2012; Flaxman 2010). This chapter
is based on the learnings developed from three cases of applying Geodesign Hub
to urban design and infrastructure planning scenarios in Australia. Geodesign Hub
enables users to rapidly create and compare many design alternatives, which is found
to be effective in the early stages of a complex project (Campagna et al. 2016). It also
provides users with opportunities to collaborate, by negotiating in a mutual learning
and consensus-building process (Campagna et al. 2016). Although these advantages
are attractive to urban planners and designers, practice using the Geodesign approach
requires advanced knowledge and time to pre-test models (Albert and Vargas-Moreno
2012). It requires a clear understanding of the framework and GIS mapping, along
with careful preparation that includes developing, testing and constantly monitoring
the tool (Rivero et al. 2015).
The application of Geodesign Hub in this study is exploratory in nature—the
scenarios in each case study were developed by postgraduate students (master’s
degree) at the University of New South Wales (UNSW) Sydney and University of
Canberra, Australia, as part of studio-based learning. A master’s level is appropriate
in using Geodesign education (Steinitz 2012). The case study approach is particularly
suitable where the observer (in this case the educator) has access to an unexplained
phenomenon (in this case, application of Geodesign Hub for students’ learning) (Yin
2003). The methodology taken here is not about the comparison of data or each case
or generalizing the outcome. Rather, it is reflective, focused on developing novel
ideas from cases—enabling a deeper understanding of the Geodesign process and
the narratives it generates (Flyvbjerg 2006). This fits with the purpose of this study,
which aims to contribute to an enhanced understanding of the Geodesign framework,
and its application to the development of future city scenarios.
142 C. Pettit et al.
Tabl e 1 Data inputs and sources for the three Geodesign case studies
Input data Data source Case study
Electricity transmission lines
Major pipelines and canals
Digital elevation model
Very high resolution imagery
Water bodies
Creek lines
Road centerlines
Road easements
Schools and institutions
Railway stations
Railway lines
Land and Property Information (now Land
Registry Information) https://www.records.nsw.
gov.au/agency/2020,http://www.nswlrs.com.au/
1&2
Land zoning
Possible maximum flood
100 year flood line
Proposed biodiversity zones
NSW Dept Planning and Environment (DPE),
https://www.planningportal.nsw.gov.au/
opendata/
1&2
Heritage register
Vegetation types
Generalized canopy coverage
High value ecological zones
Acid sulphate soils
Soils
Office of Environment and Heritage (OEH),
https://www.seed.nsw.gov.au/
1&2
New and proposed infrastructure
light rail line
Manually digitised from PDFs available from
NSW Dept Planning and Environment (DPE),
https://www.planningportal.nsw.gov.au/
opendata/
Western Sydney Airport https://
westernsydneyairport.gov.au/
Greater Sydney Commission https://www.
greater.sydney/
2
Territory zones and overlays
Commerce
– Parks and recreation
– Non-urban/hills–ridges
– Transport and services
– Mixed use commercial
– High density Residential
– Medium density residential
– Low density residential
– Community facilities
ACT Government’s ACT map i pla tfo rm
http://app.actmapi.act.gov.au/actmapi/index.
html?viewer=tp
ACT Planning Strategy (2012)
Transport for Canberra Strategy (2012)
3
Cycle and pedestrian paths ACT cycling and walking maps
http://files.transport.act.gov.au/cyclingmap/
index.html
3
Bus network ACTION Bus network service maps
https://www.transport.act.gov.au/getting-around/
find-a-stop-or-map
3
Roads ACT Government’s ACT map i pla tfo rm
http://app.actmapi.act.gov.au/actmapi/index.
html?viewer=tp
3
New and proposed infrastructure
Light rail line
Manually digitised based on Light Rail Update
(Transport Canberra August 2017)
3
9 Geodesign—A Tale of Three Cities 143
Data on each case were collected from class activities and student designs using
Geodesign Hub. The data was sourced from a number of open government data repos-
itories with a number of data inputs manually digitised as summarised in Table 1.The
data were juxtaposed in a context of studio-based learning from three aspects—data
and technology, process and outputs. Overall patterns/observations are synthesised
with the reflection of lecturers from each of these aspects.
2.1 Case Study 1—South East Sydney
The South East Sydney catchment (Fig. 2) covers an area of approximately 60 km2
and overlaps three local government areas; City of Sydney, Randwick City and Bay-
side City councils. There are several sites within the catchment that are significant
to both local and regional socio-economic status, including: Port Botany, UNSW
and Prince of Wales Hospital, Randwick Race Course, and several golf courses. The
port’s industrial zone and the presence of a growing education and health precinct
centred around UNSW and Prince of Wales Hospital are the major economic drivers
within the study area. In addition, Coogee and Maroubra beaches and La Perouse and
Malabar Headlands—sites of high tourist and environmental value—are also within
the catchment.
Furthermore, the study area is bounded by Kingsford Smith Sydney International
Airport to the South-west and by Sydney’s Central Business District to the North.
Proximity to these locations influence the study area’s attractiveness as a place of
residence.
Geodesign students were assigned to plan for the future of the study area in 2050
given the following assumptions: (i) the population in the study area would double
and an additional 175,000 dwellings would be needed; (ii) 25% of the population
would be people over 65 years of age, while 16–18% will be composed of children
0–14 years of age; (iii) communities are expected to increase in cultural diversity
as the university and hospital’s education-health precinct continue to attract global
talent; (iv) sea level rise is expected to impact low-lying places in the study area.
Students, mainly from UNSW’s Master of City Analytics and Land-
scapeArchitecture programs, were grouped into five teams of three to four members.
Each team was assigned to take on the role of system experts as they characterised the
study area’s strengths and weaknesses and proposed interventions to address future
needs. The Geodesign course was run in intensive mode, with students attending
full-day classes in three weekends spaced over the course of the semester (July–Oct
2018). For this class students had access to the City Analytics Lab which includes
6 interactive touch screen displays, connected via Cruiser Technology (Fig. 3). The
students used Geodesign Hub in tandem with a number of GISsoftware packages
including ArcGIS and QGIS, and online-based collaboration and project manage-
ment tools to complete their assessment tasks.
144 C. Pettit et al.
Fig. 2 South East Sydney on Geodesign Hub
Fig. 3 Students in City Analytics Lab presenting Geodesign scenarios
9 Geodesign—A Tale of Three Cities 145
In setting up the Geodesign exercise, six ‘systems’ where chosen for this case
study. This was a subset of a more sustainable implementation of Geodesign for the
South East Sydney area, where nine systems were selected (Pettit et al. 2019). The
six systems are described below:
1. Transport—transport infrastructure and policies that would promote active trans-
port in the study area.
2. Industry—as opposed to more traditional, heavy industries present in the study
area, this system focuses more on providing infrastructure and a policy envi-
ronment that would promote the growth of creative and high-tech industries
that could complement the presence of UNSW’s and Prince of Wales Hospital’s
research facilities.
3. Green Infrastructure—this system focuses on infrastructure projects and policies
that would encourage water sensitive urban design in the study area.
4. Education—this system focuses on the provision of additional educational facil-
ities in the study area to meet future enrolment forecasts.
5. Mixed-Use (High Density Housing +Traditional Commercial)—this system
focuses on the delivery of high density housing (130–1000 dwellings per hectare)
in the study area to accommodate population growth in the study area.
6. Medium Density Housing—this system focuses on delivering housing options
for residents in the study area. “Medium Density” has been defined as 70–130
dwellings per hectare.
The students were tasked to produce evaluation maps for their respective sys-
tem as part of their initial desktop research on the study area. These are thematic
maps expressing where in the study area each group has deemed that an interven-
tion/development associated with their system would be most feasible and where
interventions would be inappropriate. These ‘feasibility categories’ were informed
by available spatial data from the NSW Open Data portal. The students were given
in-class ArcGIS tutorials to assist them in creating these maps.
While the intention behind the evaluation maps is to guide Geodesign participants
in locating their proposed projects and policies, the exercise of creating them is
perhaps more valuable in that if forces students to recognise the limits of spatial
information upon which strategic plans would be based upon.
A total of 79 individual projects and policies were proposed, averaging 13 pro-
posals per system. The students used Geodesign Hub to draw the spatial extent of
each proposal in the study area (Fig. 4). To facilitate documentation, each group also
kept their own ‘project register’; a simple spreadsheet listing each of the projects and
policies under the system they are responsible for, their respective diagram number
on the Geodesign Hub platform, and the URL to their respective project description.
This practice allowed the other groups to glean more information about each pro-
posal put forth by the other groups. This included information such as objectives of
the proposal, implementation timeframe, and approximate costing.
In general, the projects and policies that were proposed for each system were aimed
at accommodating population increase and demographic changes forecasted for the
study area, as well as responding to the effects of climate change. The students worked
146 C. Pettit et al.
Fig. 4 Screenshot of projects and policies in Geodesign Hub—DNA of South East Sydney
in five groups each developing a future scenario for South East Sydney. Students expe-
rience the negotiation process of presenting their scenarios then collectively working
towards one final agreed-upon scenario for the study area (Fig. 5)—Scenario B.
2.2 Case Study 2—Western Sydney Aerotropolis
In Sydney’s South-west a new model of urban development has been proposed
whereby an airport, a network of new motorways and train lines are used as a cat-
alytic device to stimulate economic growth and support the development of a new
city (Fig. 6). The government has proposed that the new city will consist of a mix
9 Geodesign—A Tale of Three Cities 147
Fig. 5 Final two negotiated scenarios—a(left) and b(right)—for South East Sydney
of industries to complement what has been described in infrastructural terms as an
‘Aerotropolis’ after the populist term associated with the entrepreneurial academic
Kasarda (2014,2015). The scale of the vision has varied in government documents.
In the recent versions, the small rural hamlet of Bringelly is envisaged as developing
as a future new large urban centre to rival neighbouring centres such as Penrith and
Liverpool.
The development is a key part of the plan to position Western Sydney as a distinct
city, 1 million in population, to complement the intense development in the demo-
graphic centre of Sydney around Parramatta, and in Sydney’s global economic centre
situated on the harbour (Fig. 7). Current objectives for the development include an
ambitious planned 200,000 jobs across a range of industrial and knowledge based
industries (Australian Government and NSW Government 2018). Although future
residential figures are not clear from government documents it is clear that the pro-
posed land releases could accommodate a city of the same size as Penrith or Liverpool
which have almost 200,000 residents each (Rezek et al. 2015).
The construction of an Aerotropolis in the context of sensitive greenfield land-
scapes was the objective of a Geodesign studio held in late 2018 in UNSW’s Master
of Urban Development and Design Program. The Aerotropolis is to be situated in
the hottest part of Sydney and an area that suffers from extended heatwaves. The
question “how to design a climate sensitive city” was posed within the studio and
explored through complex scenario-based development that emulated the various
scenarios posed in the Paris Agreement (Schleussner et al. 2016). Five Geodesign
teams of four to five students worked on developing three scenarios each for the new
urban district. Each scenario was considered in terms of three time periods: 2025,
2035 and 2050. The intent was to evaluate the varying impacts of each scenario on
local urban climate at micro (1 km ×1 km) and meso (20 km ×20 km) scales.
Ten systems were used for the Geodesign case study, listed below. These were
evaluated within the 20 km by 20 km square boundary area which captured the basic
extents of the urban development planned for western Aerotropolis (Fig. 6).
148 C. Pettit et al.
Fig. 6 Western Sydney and its planned development (Source Australian Government and NSW
Government 2018). The study area, outlined in red, consisted of a 20 km by 20 km2area of interest
centred on the new airport
1. Green Infrastructure (biodiversity and conservation)
2. Blue (Water) Infrastructure (supply, disposal, storage and recycling)
3. Grey Infrastructure (transportation, communication)
4. Energy Infrastructure (production, distribution)
5. Agriculture
6. Industry (e.g. manufacturing, distribution commerce, mining, etc.)
7. Low density housing
8. Mixed use (high density housing plus services-commerce)
9. Institutional (schools, hospitals, civic)
10. Terrain and Landform
A wide range of projects was put forward, and blue infrastructure, landform, grey
infrastructure and green infrastructure were considered as the four most influential
existing systems in influencing future projects.
Each team was then asked to ‘generate’ or ‘borrow’ three projects for each system.
Geodesign facilitates the ‘borrowing’ or replication of projects within a studio as part
of its collaborative capacity. Thirty projects in total were drawn on tracing paper by
each team as a reflective analogue preparation for inputting the projects digitally into
the PSS. The five teams were then asked to enter the projects in Geodesign Hub. Each
9 Geodesign—A Tale of Three Cities 149
Fig. 7 Western Sydney in relation to Parramatta and Sydney CBD (Source Greater Sydney Com-
mission 2018)
team developed a unique and large variety of projects. These were often whimsical
rather than data driven and many lacked a strong relationship with the underlying
data layers (Fig. 8).
The most convincing scenarios focused on using a single system to guide the
design of the vision. For example, low density housing or mixed use housing were
considered in different configurations with varying consequences for urban climate
(Fig. 9). Likewise some teams developed strong green infrastructure projects centred
on South Creek which is the largest blue infrastructure and geographic feature in the
district.
The two teams that proposed the most compelling plans had differing approaches
to their investigation of climate impacts. One focused on the climatic benefits of a
strong green infrastructure network and located strategic mixed use centres in relation
to existing infrastructure. The second team focused on preserving the agricultural
landscape and demonstrating the contrasting climate impacts of extensive low density
development versus an intensive linear mixed use development.A negotiation process
was held in one of the later Geodesign classes and a preferred scheme was elected
through a process using a sociogram (Steinitz 2012). Students presented their final
scenarios back to the class through the use of posters (Fig. 10).
150 C. Pettit et al.
Fig. 8 A diverse range of projects were generated in the studio—although these were often not
closely linked with the underlying data layers
Fig. 9 This figure shows the possibility of land for further low density development. The majority
of existing green and agricultural space is likely to be lost to future housing development
9 Geodesign—A Tale of Three Cities 151
Fig. 10 Analogue presentation of scenarios for South West Sydney facilitated collaboration and
feedback in the studio context
2.3 Case Study 3—Canberra
Canberra is the national capital of Australia, situated in the Australian Capital Ter-
ritory (ACT) with a population of 397,397 (Australian Bureau of Statistics Census
2016). ACT recorded the largest population growth rate of all States and Territories
across Australia in the 2016 Australian Census results and an 11.2% increase from
2011. It is expected that ACT’s population will reach 421,000 by 2020 according to
the latest population projection (ACT Government 2017).
Canberra was originally designed as a garden city (Commonwealth of Australia
1913, in Hall 2002) by Walter Burley Griffin, the winner of the international compe-
tition of the design of Canberra held in 1911. It is a poly-centric structure comprised
with self-contained towns with a small CBD area, making its city shape unique
and different from other Australian capital cities. Car is a dominant travel mode for
Canberrans. Mode share of car transport for travel to work (as a driver) was 75%
on the 2016 census day. Cycling is increasingly being promoted as a key transport
strategy, and the share of cycling in Canberra (8.4%) is top among Australian capital
cities. Canberra has a stable population of young families because of the employ-
ment opportunities in national government. The city is rapidly expanding with the
development of new neighbourhoods in north and west. A new light rail corridor
is under construction on the north side, expected to open in early 2019. Increasing
amounts of medium and high density housing are being constructed to capture the
value along this transit corridor.
Students of the Master of Urban and Regional Planning, University of Can-
berra participated in the case study using the Geodesign Hub platform to propose a
North Canberra Infrastructure Plan 2050. The study boundary is shown in Fig. 11;
it excludes areas which are owned by federal government. Geodesign Hub was first
used in class in 2017 and was continued in 2018. As groups of three or four, stu-
152 C. Pettit et al.
Fig. 11 North Canberra study boundary on Geodesign Hub
dents were advised to create two or three plans, then proceed to create a final version
through a negotiation process. The systems that were applied to this case study were:
1. Parks and public spaces (PPS)
2. Medium density housing (MDH)
3. High density housing (HDH)
4 Mixed use (MXD)
5. Bus/public transport (BUS)
6. Tram transport (TRAM)
7. Active transport (ATRANS)
9 Geodesign—A Tale of Three Cities 153
Fig. 12 A portion of the 242 proposed projects and policies students developed with the Geodesign
Hub platform
Each group was asked to set a ‘vision’ for their Canberra 2050 strategy first and
present at the class. After receiving questions and feedback, each group updated their
vision and created projects and policy proposals. Similar to the experience from the
second case study (Aerotropolis), borrowing and sharing of common projects did
not happen. Students preferred to draw projects and policies for their own team. A
total of 242 projects and policies were created, averaging 34.6 proposals per system
(Fig. 12).
Because the class was focused on infrastructure planning and common under-
standing of the need for integrated land use and transport planning, the main policy
that was suggested was twofold: transit oriented development and active transport.
Accordingly, students put greater focus on TRAM and ATRANS systems.
For example, one group created a design based on the principle of ‘to enhance
inclusive levels of modes of transport service for people, providing alterations to
public transport which are constant with ACT Government policies’ and ‘to extend
and grow more sustainable urban development and redevelopment alongside the rail
corridor, with occupation, commercial and other profits potentially extending to the
154 C. Pettit et al.
Fig. 13 Students presenting their scenario focusing on improving public transport network
whole community of North Canberra, with rail infrastructure expansion’ (students
present this in Fig. 13).
For active transport, students focused on cycling paths, because Canberra has the
highest proportion of cycling (travel to work journey) among Australian capital cities.
The main principles of a cycling precinct are one where cyclists are protected from
traffic. To promote walking, sidewalks are designed beside natural landscape, CBD
and high-density corridors in suggested plans. Three of the active design proposals
were then compared to show the impacts of different design interventions (Fig. 14).
The ‘negotiation’ module was used to develop students’ consultation skills. While
one group was presenting their design, students in the audience asked questions and
provided comments as stakeholders. Each group then considered the comments and
reflected on their final designs. This worked well for students to understand which
design features were missing or redundant. In addition, students could enhance their
presentation/communication skills as future planners. The most challenging task
was to consider the timeline and cost. Students who had not worked in planning or
the construction industry did not know the budget cycle and realistic timeframe for
managing projects. Yet, it was still a useful exercise in guiding students to greater
awareness of their designs’ implications to cost and time management.
9 Geodesign—A Tale of Three Cities 155
Fig. 14 Comparison of designs of three student groups
156 C. Pettit et al.
Tabl e 2 Descriptive statistics comparing the three case studies
Case study # students #systems # scenarios #
diagrams
#class
length
# number
of classes
South East
Sydney
19 6 5 79 6h 6
Western
Sydney
23 10 15 1106 3h 14
Canberra 10 7 1 242 3h 4
3 Discussion
This section reflects on the experience and observations of the educators in running
the three Geodesign studios for the respective case studies. This has been done by
reflecting on: (i) data and technology; (ii) process; and (iii) outputs. Descriptive
statistics comparing the three scenarios are provided in Table 2. This highlights the
significant differences in the experience of participants—in this case students—in
the Geodesign exercises.
3.1 Data and Technology
A paucity of good quality data remains a challenge for the application of digital
planning tools such as Geodesign Hub. Even though there was significant open data
resources available from NSW and ACT Government NSW data portals, it was
still an unresolved challenge to construct comprehensive, complete property level
baseline datasets to support the development and evaluation of Geodesign scenarios
for the three case studies reported in this chapter. Also, acquiring project costings
for the construction of buildings and infrastructure was a challenging experience in
three different Geodesign exercises. The costings of policies was a subject of much
discussion, particularly with participants in the South East Sydney study.
The support technology platform used for all three case studies was Geodesign
Hub (Ervin 2011). The strengths of the Geodesign Hub platform include its user-
friendly interface and easy online access. Drawing projects and policies with it’s
simple sketch planning functionality was found relatively easy for students to mas-
ter. For example, in the Canberra study area, students easily mapped the unique shape
of neighbourhoods (e.g. round shape blocks and roundabouts) comprising the city.
Another strength of Geodesign Hub is that it is aligned perfectly to the Steinitz Frame-
work (Steinitz 2012) and provides both an administrative interface and participant
interface to sequentially and iteratively undertake the Steinitz approach. However,
the researchers also identified some challenges to using Geodesign Hub. Firstly,
the Hub is not open source and thus new models and functionality cannot directly
be developed and implemented by researchers using the platform. Secondly, it was
9 Geodesign—A Tale of Three Cities 157
observed in all three Geodesign exercises that synchronising projects and policies
with multiple end users was sometimes challenging when students were creating
multiple (five or more) projects at the same time, when the data layer functionality
was used by multiple students, or when too many projects and policies were entered
into the software. The ability to edit projects and policies and make detailed notes on
projects within Geodesign Hub was also observed as a challenge for all three case
study applications.
3.2 Process
The Geodesign process undertaken for all three case studies was based on the Steinitz
Framework (Steinitz 2012) which is logical, systematic and lends itself to a data
assisted approach to urban planning, infrastructure planning and design. However,
while the framework is comprehensive it is also complicated and takes participants
significant time to learn. This in itself is not a problem given that Geodesign was
taught over the duration of master’s level course, which provided students sufficient
timetolearntheframework.
One of the challenges for participants in undertaking the Geodesign approach is
the negotiation in and between Geodesign teams. Negotiation skills are essential for
meaningful dialogue and decisions on which projects and policies to select through a
consensus approach. This was observed across all three Geodesign exercises under-
taken by the students. Negotiation skills have been identified as critical ingredient in
a successful Geodesign exercise (Steinitz 2012). In the context of our three case stud-
ies, the negotiation between students was observed as somewhat limited. This could
be attributed to the fact that communication skills are critical for robust negotiation
and across all three case studies there were students with various levels of communi-
cation skills. We believe this challenge is not limited to just students. However, this
requires further investigation with built environment practitioners and community
participants to understand if this is in fact a challenge for Geodesign when applied
in the real world.
As indicated in Table 2, there were significant differences in the number of future
city scenarios created by the students Across the three case studies students where
given the opportunity to develop their own future growth scenarios informed through
their interpretation of available key planning government documents. In the Western
Sydney case study, students were asked to created nine scenarios per group, resulting
in forty-five scenarios in total. However, when students were left more to their own
devices, as in the Canberra case study, they did not explore multiple designs and
scenario options. Students felt comfortable building their systems and then coming
together to formulate a scenario but were reluctant to take the opportunity to construct
and explore multiple scenarios. Another possible explanation is that students had less
time to create scenarios in the Canberra case study.
158 C. Pettit et al.
Finally, it was a challenge for some students to develop a deep knowledge of the
relevant study area. This is potentially similar to the real world, as built environment
practitioners might not necessarily have a deep knowledge of the geography and
place where Geodesign future scenarios are being developed. Students participating
in the South East Sydney Geodesign case study did not take a field trip and some
found it challenging to grasp the key drivers of change for the study area (as was
demonstrated in some of their earlier projects and policies). Students in the UNSW
Urban Development and Design studio did a field trip, experienced high level brief-
ings and spent five weeks doing GIS analysis. Nevertheless, when it came to linking
this knowledge to the act of designing many teams were challenged and found this
very difficult. A recommendation to address this is to ensure students and practi-
tioners undertake extensive field work investigation and desktop analysis including
community profiling, economic analysis and sustainability analysis to develop a deep
understanding of the study area.
3.3 Outputs
The Geodesign Hub platform provide the opportunity for planners to create real-
time living, breathing plans for a city (Pettit et al. 2019). Geodesign Hub supports
the creation of real-time plans which support infrastructure sequencing, and the
timeline module of Geodesign Hub is a useful feature to support this. However,
it is noted that the costings of projects and policies which can be entered into the
platform are aggregate and independencies between projects and policies cannot be
linked. From the experience of the teaching studios, the strength of the Geodesign
Hub software is in its ability to interrogate process and social interactions rather
than the perfection and development of high quality final scenarios. For product and
aesthetically motivated students this served as a frustration.
Another insight is that the ability to generate metrics from the software is mostly
related to projects rather than policies. For this reason projects rather than policies
were encouraged. Policies may help the communication of a vision but do not con-
tribute to the metrics generated by Geodesign Hub, limiting their functionality in
assessing the studio visions.
Students were asked to limit the number of projects they designed and entered
in Geodesign Hub. This was due to previous difficulties with the handling time of
Geodesign Hub when too many projects were created. Although students drafted
the projects on tracing paper before committing them to the Geodesign PSS, they
often then disregarded these hardcopies when switching to the digital PSS. This
rejection of the pedagogical scaffolding can be attributed to several potential factors.
Digital systems encourage redundancy and superfluity. Whilst a hardcopy project
takes considerable time to generate and be pinned up for discussion by the team
and the class, digital projects can be generated much more rapidly and without the
same level of conscious creation, much like a facebook or social media post for
example. This created challenges in the studio: although projects were initially rapid
9 Geodesign—A Tale of Three Cities 159
and easy to produce in Geodesign Hub, when hundreds of projects are created the
PSS becomes slower, limiting the ease and manageability of the software. At this
stage in the software’s development large amounts of data are not able to be handled
by the platform.
Although borrowing and sharing of common projects was included in the studio
workflow and scaffolding, this collaborative potential of the software was not fully
taken advantage of. This may be because of the educational background and expe-
rience of the students, who throughout their master’s level education at Australian
institutions, are generally trained to think of replication or copying as a negative act.
Collaboration and discussion and sharing of ideas within class is therefore much
more challenging than the simple introduction of a collaborative software.
An important observation when undertaking all three Geodesign exercises was
that students did not extensively use the evaluation maps created in the Geodesign
process, specifically the land feasibility-suitability maps generated through a simple
GIS multiple criteria approach. This can potentially be attributed to the previously
reported data concern that a paucity of good quality fine scale data exists to sup-
port land suitability-feasibility analysis across large urban areas in Australia. Land
suitability is also a subjective process relying on expert judgement ion determining
which land use is developable or not developable and thus making such judgement
calls can be challenging and problematic for non-experts.
4 Conclusions
In this chapter we have presented and reflected on three Geodesign studios under-
taken in Australia with students in 2018. The three case studies highlight how dif-
ferent groups engage in the Geodesign process. For example, in the Canberra case
study students were hesitant to explore formulating multiple scenarios and devel-
oped one shared scenario. Students involved in all three studios learned important
skills in negotiation and collaboration which are essential when undertaking a sys-
tems approach to planning and design. However, the skill of workshop and project
facilitation was taught; the Geodesign facilitation was undertaken by the educators
of the respective courses.
The capacity of Geodesign Hub and Geodesign approaches pushed the students
creativity and collaborative abilities. The complexity of the adopted framework was a
very useful pedagogical approach with the students gaining new skills and conceptual
insights. However, the complexity of the process limited the clarity of the final
outputs. In the context of the Western Sydney case study one reason for this might
be that climate sensitive approaches are an emerging topic in urban development and
design and clear approaches and examples are difficult for students to access—and
need to be generated first from design principles.
The WHAT, WHERE and WHEN questions are asked in the stage of implementing
the study (i.e. through models of representation, process, evaluation, change, impact
and decision). With the likelihood of overload of data and more methodological
160 C. Pettit et al.
options available in the future for students, practitioners and decision makers, balance
of activities and collaboration may shift. However, the WHAT, WHERE and WHEN
questions should be still used (Steinitz 2012).
With an increasing number of courses of Geodesign be offered in Australia and
indeed internationally, the signals are promising that the next generation of city plan-
ners and designers will be skilled to undertake data assisted approaches in creating
strategies for realising sustainable urban futures. As identified by Russo et al. (2018),
addressing the education and training gap should lead to a greater level of adoption
of PSS tools such as Geodesign in planning and design practice. This is absolutely
critical in a time when globally we are faced with the grand challenges of population
growth, rapid urbanisation and climate change.
References
ACT Government (2017) ACT population projections 2017 to 2020. Government document. https://
apps.treasury.act.gov.au/__data/assets/pdf_file/0016/1041280/ACT-Population-Projections-
2017-2020.pdf. Accessed 13 Dec 2018
Albert C, Vargas-Moreno JC (2012) Testing geodesign in landscape planning–first results. In: Digital
landscape architecture conference, vol 31, May–June 2012
Australian Bureau of Statistics Census (2016) 2016 Census QuickStats. http://quickstats.censusdata.
abs.gov.au/census_services/getproduct/census/2016/quickstat/8?opendocument. Accessed 13
Dec 2018
Australian Government & NSW Government (2018) Western Sydney aerotropolis investor guide.
Government document. www.industry.nsw.gov.au/invest-western-sydney. Accessed 13 Dec 2018
Campagna M, Moura ACM, Borges J, Cocco C (2016) Future scenarios for the Pampulha region:
a geodesign workshop. J Digit Landsc Arch 1:292–301
Ervin S (2011) A system for GeoDesign. In: Buhmann E, Ervin S, Pietsch M (eds) Proceed-
ings digital landscape architecture 2011. Wichmann Anhalt University of Applied Sciences,
Berlin/Offenbach, pp 145–154
Flaxman M (2010) Fundamentals of geodesign. In: Proceedings of digital landscape architecture,
vol 2. Anhalt University of Applied Science, pp 69–72
Flyvbjerg B (2006) Five misunderstandings about case-study research. Qual Inq 12(2):219–245
Geertman S, Allan A, Pettit C, Stillwell J (2017) Introduction to ‘planning support science for
smarter urban futures’. In: Geertman S, Allan A, Pettit C, Stillwell J (eds) planning support
science for smarter urban futures. Springer International Publishing, Berlin, pp 1–19
Greater Sydney Commission (2018) Western Sydney district plan. NSW Government, Parramatta
Hall P (2002) Cities of tomorrow: an intellectual history of urban planning and design in the
twentieth century, 3rd edn. Blackwell, Oxford
Kasarda JD (2015) Welcome to aerotropolis, the city of the future. New Perspect Q 32(3):43–45
Kasarda JD, Appold SJ (2014) Planning a competitive aerotropolis. In: The economics of interna-
tional airline transport. Emerald Group Publishing Limited, Bingley, pp 281–308
Pettit C, Hawken C, Ticzon C, Steinitz C, Ballal H, Cranfield T, Leao S, Afrooz A, Lieske C (2019)
Breaking down the silos through geodesign—envisioning Sydney’s urban future. Environ Plan
B (in Press)
Pettit C, Bakelmun A, Lieske SN, Glackin S, Thomson G, Shearer H, Dia H, Newman P (2018)
Planning support systems for smart cities. City, Cult Soc 12:13–24
Rezek D, Psychogios T, Artup B (2015) Shaping future cities, designing Western Sydney. Deloitte,
Sydney
9 Geodesign—A Tale of Three Cities 161
Rivero R, Smith A, Ballal H, Steinitz C (2015) Promoting collaborative Geodesign in a multidisci-
plinary and multiscale environment: Coastal Georgia 2050, USA. Digit Landsc Arch, 42–58
Russo P, Lanzilotti R, Costabile MF, Pettit C (2018) Towards satisfying practitioners in using
planning support systems. Comput Environ Urban Syst 67:9–20
Schleussner C-F, Rogelj J, Schaeffer M, Lissner T, Licker R, Fischer EM, Knutti R, Levermann A,
Frieler K, Hare W (2016) Science and policy characteristics of the Paris Agreement temperature
goal. Nat Clim Chang 6(9):827–835
Steinitz C (2012) A framework for geodesign: changing geography by design. ESRI, Redlands, CA
Yin RK (2003) Case study research: design and methods, third edition, applied social research
methods series, vol 5. Sage, Beverly Hills (Incorporated)
Chapter 10
Toward a Better Understanding of Urban
Sprawl: Linking Spatial Metrics
and Landscape Networks Dynamics
Tengyun Hu, Xiaochun Huang, Xuecao Li, Lu Liang and Fei Xue
Abstract Although many studies have been made on urban land use change, most of
them focus on a specific aspect (e.g., landscape or policy) with fewer considerations
of spatiotemporal dynamics of urban sprawl. In this study, we explored the urban
sprawl process in Beijing over the past three decades (1984–2013) on an annual
basis by linking spatial metrics and landscape networks to trace the dynamics of urban
patches. First, we identified six main growth periods of urban expansion with different
patterns, through a relatively comprehensive change analysis of different landscape
matrix. Second, we used the urban landscape network to explain the spatiotemporal
dynamics of urban patches, with a linkage to policies (or planning) behind each
hotspot of urban expansion. We found the major trajectory of urban growth in Beijing
started from the northern and southern parts of the central built-up region to its
southeast side, which is different from the planned two axes (i.e. horizontal and
vertical) along the core area of the city. Our results indicated that urban expansion
is highly uneven both in time and space, which relies on more satellite observations
for a better understanding of this process. Also, the identified urban growth periods
and patterns are valuable to the development of urban sprawl model.
Keywords Urban morphology ·Spatial dynamics ·Land use planning ·Urban
sprawl ·Urbanization
T. Hu ( B)·X. Huang
Beijing Municipal Institute of City Planning and Design, Beijing 100045, China
e-mail: hutengyun88@163.com
X. Li
Department of Geological & Atmospheric Science, Iowa State University,
Ames, IA 50011, USA
L. Liang
Department of Geography and the Environment, University of North Texas,
Denton, TX 76203, USA
F. Xu e
Department of City Planning, School of Architecture and Urban Planning, Beijing
University of Technology, Beijing 100124, China
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_10
163
164 T. Hu et al.
1 Introduction
Our planet is experiencing rapid global urbanization due to the growth of urban
population, which is the direct driver to the physical expansion of urban areas as
well as other relevant issues such as climate change (Georgescu et al. 2014), public
health (Gong et al. 2012), air pollution (Zhang et al. 2012), urban heat island (Clinton
and Gong 2013), and shortage of water and/or energy (McDonald et al. 2014; Zhang
and Seto 2013). More people are living in cities than those living in rural areas in
the world, and the gap between urban and rural population will continue to increase
in the future (United Nations 2018). Correspondingly, the trend of urban sprawl will
also continue (Liu et al. 2011; Seto et al. 2000). Therefore, it is of great importance
to monitor the dynamics of urban areas and to depict their characteristics to allow
for better urban planning and management (Li and Gong 2016b).
Urban sprawl has different stages and different growth types or patterns over
spaces (Li et al. 2015; Liu et al. 2014; Sexton et al. 2013). Dynamics of urban mor-
phology reflect the inherent heterogeneity of urban development due to a variety of
factors, such as socioeconomic status, the biophysical environment, and particularly
planning policies (Irwin and Bockstael 2007). Therefore, understanding the reasons
behind the urbanization and acquiring the pathways of urban sprawl over different
stages, is of great importance to urban growth modeling (Brown et al. 2005). Urban
sprawl is always accompanied with two processes: coalescence and diffusion, which
may coexist in most times with varying intensities (Yu and Ng 2006). Unfortunately,
few attempts have been made to explore various driving factors and resultant patterns
in rapidly urbanized areas where monitoring land use change requires observations
with finer spatiotemporal resolutions, to reconstruct an overall storyline of the urban
sprawl process (Li and Gong 2016a; Sexton et al. 2013).
Remote sensing is the primary source to detect urban extent dynamics over a large
area. A variety of remotely sensed imagery can be used to map urban land from space
over regions or continents, such as Moderate Resolution Imaging Spectroradiome-
ter (MODIS) (Schneider et al. 2010) or Defense Meteorological Satellite Program
Operational Line Scanner (DMSP/OLS) datasets (Liu et al. 2012; Zhou et al. 2014).
However, these coarse resolution datasets fail to capture details of urban shape or
patch (Sohl et al. 2007), which is a direct reflection of urban expansion by inherent
socioeconomic activities. For instance, at city or metropolis scales, urban landscape
can be used to explore a variety of relative topics such as energy consumption (Chen
et al. 2011), carbon estimation (Ou et al. 2013), and other socioeconomic activities
(e.g. population and gross domestic product, GDP) (Irwin and Bockstael 2007;Li
et al. 2014; Schneider et al. 2005; Zhou and Sun 2010). Therefore, Landsat images
are a priority for urban mapping and further to support urban sprawl modeling due
to the finer spatial resolution (i.e., 30 m) (Gong 2012; Wang et al. 2012), particularly
when multiple phrases of urban expansion need to be revealed.
Urban landscapes can provide a relatively comprehensive description of urban
forms, such as patch size, shape, distribution or overall patterns, compared with
traditional approaches that identify urban growth periods based on size increment or
10 Toward a Better Understanding of Urban Sprawl: Linking … 165
growth rate of the economy (McGarigal et al. 2012). With continuous observations
over multiple times based on Landsat images, it is feasible to capture the spatial
dimension, structure or pattern of urban sprawl through analyzing urban landscapes
dynamics (Irwin and Bockstael 2007; Taubenböck et al. 2014; Yu and Ng 2006),
which is helpful to identify different urban growth phrases (or temporal contexts)
since urban sprawl is a spatially heterogeneous process both in space and time (Li et al.
2014; Seto and Fragkias 2005). However, most existing studies on urban landscape
metrics often cover a long temporal range (e.g., decades) with wide intervals (e.g.,
5 or 10 years or even longer). For example, Ji et al. (2006) adopted six images to
analyze changes from 1972 to 2001. Taubenböck et al. (2014) quantified the spatial
characteristics of a mega-region (i.e. Pearl River Delta) from 1975 to 2011 with only
four intervals. Also, there are other relevant cases such as Schneider et al. (2005)
and Yu and Ng (2006). Coarse temporal resolution with relatively equal intervals
will bias the derived conclusions since (1) change of urban form within a short
period will be overwhelmed in a coarse temporal frequency and (2) urban sprawl
is temporally uneven (Li et al. 2015; Sexton et al. 2013). Both these two factors
will hinder our understanding of urbanization, e.g., fail to capture the rhythm of
urbanization (acceleration or deceleration due to specific policies or planning) (Du
et al. 2014).
Therefore, in this study, we focus on the dynamics of the urban landscape in
Beijing over the past three decades. Annual records of urban land in Beijing were
produced with Landsat images as we documented in previous work (Li et al. 2015).
In addition, the landscape network was adopted in our study to explore the dynamics
of detailed urban patches, which can be linked to particular policies (Foltête et al.
2012). We hope to capture the main phrases of urban sprawl in Beijing and identify
some hotspots of development associated with policies or planning behind.
2 Methodology
In this study, we proposed an analysis framework (Fig. 1) to understand the process
of urbanization in Beijing (China) with a continuous sequence from 1984 to 2013. As
the capital of China, Beijing experienced a considerable urbanization process during
the past three decades (Chen et al. 2002;Wuetal.2006). Our research starts from
urban spatial metrics, derived from different growth periods of urban expansion.
After that, landscape network analysis was then applied to uncover the interactions
of urban patches at a local scale. Details of each component are presented below.
2.1 Generation of Long Historical Records
We employed an annual urban land sequence of Beijing over a 30 years. More than
100 Landsat images have been collected and mapped with a supervised classification
166 T. Hu et al.
Fig. 1 The proposed analysis framework to understand the urbanization process
scheme. After the initial classification, a temporal consistency check was carried
out, which contained temporal filtering for noise moving and a logical assumption
that it is unlikely for an urbanized area to revert to non-urban land (Li et al. 2015;
Mertes et al. 2015). The average accuracy of urban lands in this dataset is above
90% through independent validation; meanwhile, for change detection in rapidly
expanding regions, the mean accuracy is higher than 80%. More details about this
product and the method used can be referred to in Li et al. (2015).
2.2 Spatial Metrics for Identifying Growth Periods
2.2.1 Spatial Metrics
In this study, we only considered metrics derived from class-level with a focus on “ur-
ban” specifically. These metrics were obtained with FRAGSTATS software (McGari-
gal et al. 2012; McGarigal and Marks 1995). The calculation of each adopted metric
is shown in Table 1. The spatial metrics used in our study can be divided into two
categories (McGarigal et al. 2012): Class Area (CA), Largest Patch Index (LPI) and
AREA are metrics used for the area edge category while Number of Patches (NP),
Patch Density (PD), and Euclidean Nearest Neighbor Distance (ENN) belong to the
class of aggregation-metric. Also, the modified form of LPI (MLPI) was used in this
study that considers the total area of urban areas not all kinds of patches (Tauben-
böck et al. 2014). These indices are relatively stable along with urban expansion (Wu
2004).
10 Toward a Better Understanding of Urban Sprawl: Linking … 167
Tabl e 1 Mathematical details of adopted spatial metrics
Metrics Definition Description
AREA aij1
10,000 Patch area
CA n
j=1aij1
10,000 Total class area
MLPI max j=1aij
CA ×100 Percentage of landscape comprised by the
largest patch
NP niNumber of patches of the corresponding type i
PD ni
A(10,000)(100)Number of patches within 100 hectares
ENN hij Distance to nearest neighboring patch
aij is the area (m2)ofpatchjfor class i,Ais the total landscape area (m2), niis the patch number
of i,andhij is the distance (m) from patch ij to the nearest neighboring patch based on patch
edge-to-edge distance
2.2.2 Heuristic Reasoning of Urban Sprawl Types
From a landscape perspective, morphological changes caused by urban sprawl can
be summarized as two processes: coalescence or diffusion (Liu et al. 2014; Seto et al.
2012) (Fig. 2). Coalescence is the growth and merging of small urban patches into
a large unit, while diffusion represents the emergence of new urban patches in the
nonurban areas around a relatively larger urban patch. During urban sprawl, these
two processes can alternatively occur.
Fig. 2 Illustration of
coalescence and diffusion
processes
168 T. Hu et al.
Tabl e 2 The metric-based
reasoning of urban growth
pattern
Growth
pattern
MLPI AREA_MN PD ENN_MN
Coalescence + + − −
Diffusion − − + +
‘+’and‘−’ indicate the value of metric increasing and decreasing
within a specific period
Based on the conceptual framework of Fig. 2, we designed an approach to heuristic
reasoning to identify different growth periods based on multiple spatial metrics.
We selected four metrics to serve this purpose, namely MLPI, PD, ENN_MN and
AREA_MN (both ENN_MN and AREA_MN are mean values of metrics ENN and
AREA for all urban patches) since they characterize urban patches from multiple
aspects (i.e., size, quantity, and distribution). The changes of these metrics within a
specific period, as well as reasoned growth types, are briefly expressed in Table 2.
For coalescence, the metrics of MLPI and AREA_MN increase while those of
PD and ENN_MN decrease. However, for diffusion, the urban landscape is accom-
panied by a lot of isolated patches, whereby PD and ENN_MN increase accordingly
(Taubenböck et al. 2014). Given that these metrics changed with different trajectories
over the past three decades, we divided each period of urban growth according to
relatively consistent patterns of changes in these metrics. Thus, we calculated the
relative growth rate γfor each metric during period ias Eq. (1).
γi=mn−m1
n
j=1mj
∗100 (1)
where nis the number of metric we considered during period i, and mjis the metric
value (i.e., m1is the earliest one and mnis the latest). Therefore, γidenotes the
relative change of a specific metric, which has been normalized to eliminate the
magnitude effect, and it can be used for comparison among different metrics.
In addition, we used an assembled metric DI (dispersion index) (Taubenböck
et al. 2014), to quantitatively depict the dynamics of the urban landscape. The DI is
calculated based on NP and MLPI, both of which have been normalized in advance
through Eqs. (2–4).
NPn=NP −1
CA −1×100 (2)
MLPIn=MLPI −1
CA
100 −1
CA
×100 (3)
DI =NP
n+(100 −MLPIn)
2(4)
where NPnand MLPInare two normalized metrics of NP and MLPI since the total
urban area CA has been considered as a modification.
10 Toward a Better Understanding of Urban Sprawl: Linking … 169
The DI indicates a clear meaning of urban landscape change. Theoretically,
if the whole region is occupied by one single urban patch (i.e., the largest urban
patch), then it belongs to an idealistically homogenous landscape with a low DI
value (0). Otherwise, if the study region is filled with randomly distributed non-
coalescent patches, it is an idealistically dispersed landscape with a high DI value
(approaching 1).
2.3 Interaction via a Landscape Network
The overall landscape of urban land is comprised of many urban patches, with dif-
ferent shapes, sizes or distributions. Thus, the evolution of urban landscape can be
regarded as patch migration or interaction in both space and time. Here, we simplified
the entity of a patch as a node, whose location is its geometric center. Thereby, these
nodes connect with each other forming a landscape network (Fig. 3).
As presented in Fig. 3, each urban patch is abstracted as a node (green nodes), and
its connective paths (gray lines) leads to ambient ones. A larger urban patch always
has more nodes for connection (i.e., node degree). Therefore, we assume it has a
higher chance to interact with other patches and would likely to be expanded. In the
landscape network analysis, the determinative size of a patch is neglected while its
spatial location and interactions are enhanced. The detection of urban patches and
their connectivity were calculated by software Graphab 1.0 (Foltête et al. 2012). The
minimum patch size considered for analysis is 1 km2. Based on the obtained urban
landscape networks, we visualized their spatiotemporal changes and used them to
understand the sprawl pattern according to localized knowledge (i.e., event or policy).
Fig. 3 Illustration of urban
landscape networks
170 T. Hu et al.
3 Results and Analysis
3.1 Continuous Change of Urban Landscape Metrics
The dynamics of urban landscape metrics is shown in Fig. 4. The entire time-sequence
was divided into 6 periods according to trajectories of multiple metrics, i.e., if the
adopted metric shows a different trend (e.g., increasing) compared to other metrics
(e.g., decreasing) within neighboring years. For periods P1 and P2, the change of
MLPI in P1 increased but decreased in P2, while for ENN_MN, the opposite tendency
was observed. The metrics for PD and AREA_MN both slowly increased from P1
to P2. Hence, we chose the year 1988 as the division for these the two neighboring
periods. Similar processing was carried out for all periods, e.g., there is a significant
turning point between periods P2–P3 and P3–P4 regarding ENN_MN. However, it
should be noted that these manually defined periods are not rigorously consistent. For
instance, change of spatial matrix for the latest short period P6 is more complicated
than others. The temporal intervals of these periods are different, ranging unequally
from 3 to 10. Details of these identified periods as well as the calculated relative
growth rate γfor metrics are shown in Table 3.
Several basic findings can be drawn from Fig. 4and Table 3. In the first period
(P1: 1984–1988), the growth pattern was dominated by coalescence since both MLPI
Fig. 4 Continuous change of landscape metrics in Beijing from 1984 to 2013. aMLPI, bPD,
cENN_MN and dAREA_MN
10 Toward a Better Understanding of Urban Sprawl: Linking … 171
Tabl e 3 Metric-based relative growth rate γfor each period
Periods MLPI PD ENN_MN AREA_MN
P1: 1984–1988 3.92 2.29 −1.28 2.44
P2: 1988–1993 −1.32 0.01 0.58 3.56
P3: 1993–2002 0.25 1.14 −0.61 1.33
P4: 2002–2006 1.95 5.10 −2.84 −1.46
P5: 2006–2011 1.61 2.21 −1.18 1.02
P6: 2011–2013 6.02 −3.47 2.64 9.08
‘−‘ indicates a negative relative growth rate
(3.92) and AREA_MN (2.44) considerably increased. There were also some new
urban patches emerging as PD (2.29) increased also. In the second period (P2:
1988–1993), the MLPI (−1.32) decreased slightly, but AREA_MN (3.56) signif-
icantly increased. Meanwhile, changes in PD and ENN_MN were small (i.e., less
than 1). Therefore, a plausible growth pattern could be small urban patches growing
independently. For P3 (1993–2002), changes in these four metrics were relatively
stable (i.e., the absolute rates γare lower). The sprawl pattern in this period followed
the pathway of P2, but some new urban patches emerged during P3. After that, a
dramatic shift in the growth pattern occurred at the fourth period (P4: 2002–2006),
during which period the PD increased significantly (5.31), indicating that a large
number of new urban patches were emerging. At the same time, in the urban core
region, a coalescence pattern was formed as shown by the increase of MLPI (1.95)
or decrease of ENN_MN (−2.84). The decreased AREA_MN (−1.46) may largely
be attributed to the increased patch number. A similar growth pattern was inherited
in the fifth period (P5: 2006–2011). Nevertheless, its intensity of dispersed sprawl
was reduced (i.e., PD decreased to 2.21 and AREA_MN increased). Therefore, in the
last period (P6: 2011–2013), a significant coalescence pattern was observed. That
is, small patches grew and merged considerably. During this period, increments of
MLPI and AREA_MN were 6.02 and 9.08, respectively. Accordingly, the PD fell
(−3.47). In addition, many new patches occurred far away from the urban core region
with increased ENN_MN (2.64).
3.2 Change of DI in Different Years
Based on the identified six periods, we selected 7 years (i.e., 1984, 1988, 1993, 2002,
2006. 2011 and 2013) to understand the sprawl process quantitatively with metric
DI and two indices of NP and MLPI (Fig. 5). Overall, from 1984 to 2013, the urban
landscape in Beijing became more compact (i.e., the DI decreased from 50.17 to
40.64), with correspondingly decreased NP and increased MLPI. Two processes were
identified within the period 1984–1988 when DI decreased by 2.92% and 2011–2013
172 T. Hu et al.
Fig. 5 Temporal dynamics
of urban growth pattern with
quantitative metric DI. The
bold texts are DI values
captured at different years
when DI decreased by 3.55%. The dynamics of DI reveal that urban sprawl is an
uneven process in time, i.e., the overall pattern of urban land use did not change
considerably during the period from 1993 to 2011 (Fig. 5).
3.3 Dynamics of Urban Landscape Network
We visualized the spatiotemporal change of urban landscape network (Fig. 6)in
Beijing for a better understanding of their interactions. The Node Degree is the
connectivity of a central node to its neighboring nodes. The largest urban patch (i.e.,
urban core extent) has expanded considerably (i.e., the yellow patch in Fig. 6). Before
2002, the dominating pathways of urban expansion were in southwest and northwest
(i.e., dotted ellipses in Fig. 6). Thereafter, two new booming areas were formed at
the southeast and northeast of the urban core region (i.e., solid ellipses in Fig. 6).
Also, patches with a relatively large node degree, temporally emerged since they
can be easily aggregated into an even larger patch or absorb others. Therefore, we
can identify the shift of patch nodes in space among different years. The network
can be approximately regarded as interactions among different patches, i.e., nodes
with more connected pathways are likely to expand in time. Some patches, although
represented as nodes, have the potential to influence ambient patches or generate
new patches nearby to change the pathway of urban sprawl.
A zoomed-in view of the urban landscape network is shown in Fig. 7with the same
legend as in Fig. 6except for the red color representing additional urban patches.
Typically, there are three evolution categories of urban patches. The first is that small
10 Toward a Better Understanding of Urban Sprawl: Linking … 173
Fig. 6 Dynamics of urban landscape networks
Fig. 7 Zoomed-in views of urban landscape network
urban patches distributed separately join together due to their natural growths and
form a new one (Fig. 7, region A). For example, three nodes in Fig. 7-V assembled
to become one node in Fig. 7-VI, and the newly emerged patch among these nodes
played a key role (Fig. 7-II, blue ellipse) such as corridors. The second (Fig. 7,region
B) is a process whereby the dominating patch absorbs small ambient ones to form
a unified entity. For instance, the urban core patch absorbed the small but highly
connected patch (Fig. 7-VII, blue solid node). Thirdly, there are also several newly
emerged patches (Fig. 7, regions of C and D), which are mainly driven by policies
or other events (e.g., construction of the Liangxiang University Town since the year
of 2003–2006). And their emergence will influence those urban patches initially far
away from each other, by either serving as a corridor or through self-growth, to
become a new urban core region.
174 T. Hu et al.
3.4 Linking Landscape Networks with Policy and Planning
The landscape network reveals that the interaction or exchange of urban patches has
different patterns. Linking them with policies or planning would be interesting. We
try to explain this linkage in Fig. 8. Based on the analysis of landscape networks, we
identified six growth hotspots over Beijing. The first is Yan Qing (Fig. 8-1), where its
sprawl is almost linear along the valley and river. The second is Fang Shan. There is an
apparent shift of nodes from the original west side (i.e., before 1988) to the east side
(i.e., after 2002) (Fig. 8-2). Liang Xiang university town was planned and began to
operate in 2003 (Fig. 8-2, red ellipse). A new trend has emerged since 2011 due to the
recently proposed policy of “Beijing-Tianjin-Hebei Integration” (Fig. 8-3). Transfor-
mational infrastructure project has also played a significant role in the urbanization
process, particularly to the land use nearby, such as Beijing International Capital Air-
port (Fig. 8-4). Many new urban patches were developed after 2004 when the airport
was extended. In the northwest part of the urban core region (Fig. 8-5), many high-
technique enterprises and universities emerged. Such effects have radiated to Chang
Ping with well-known examples such as “Shahe University Park”. On the other hand,
along with the Chaobai River (Fig. 8-6), some new urban patches have emerged.
Master Plan (2016–2035). The nodes with size greater than 7 degrees in Fig. 7
were kept and reasoned here. Different colors correspond to different years, for both
nodes and paths.
Beijing City Master Plan (2016–2035) (Beijing Government, 2016) (Fig. 8b) has
laid out a core area, a central urban area, a sub-center, horizontal and vertical axes,
multiple new towns and an ecological protection zone to guide the urban spatial
development. The two main axes were interpreted as two arrows in northern (i.e.,
Chang Ping) and southern (i.e., Fang Shan) side, respectively (Fig. 8a). Moreover, the
Fig. 8 a Linking landscape network with policy and planning and bBeijing City
10 Toward a Better Understanding of Urban Sprawl: Linking … 175
sprawling directions (i.e., north to southeast) are much stronger than others which are
planned as polycentric spatial structure. Also, the sub-center in Tong Zhou district
has been released as the high urban patch level in our analysis.
4 Discussions
4.1 Urban Growth Is Uneven in Time
Although the fact that urban growth (or evolution) is uneven in time has been widely
accepted, quantitative approaches to measuring it are fully agreed by scholars. Meth-
ods adopted traditionally with the analysis of socioeconomic variables (e.g., popu-
lation or GDP) or based on subjectively selected datasets derived from remote sens-
ing may introduce some uncertainties, since urban landscapes are changing rapidly
within a short time but a strong intensity. In this study, we focused on an urban land-
scape with a 30-year record to identify different growth periods in Beijing, using both
quantitative and qualitative approaches. Urban patterns with coalescence or diffusion
or their co-existed form have been detected. Indeed, urban sprawl is entirely different
over time, e.g. the development intensity within the recent short term (2011–2013)
is much higher than in previous periods (from 1993–2002).
The uneven urban growth periods indicate different urban policies or infrastruc-
ture projects constructed in the city. For example, the coalescence growth we iden-
tified in the period of 2002–2006 confirmed the high intensity municipal construc-
tions, i.e., industry and residential constructed projects, which served the Beijing
Olympic Games in the year 2008. During the period around the Beijing Olympic
year (2006–2011), most constructions were stopped. At that time, the priority of the
entire city was focused on the Olympic Games ceremony and sports events. That is
the reason for the lowest rate of compaction change in the period of 2002–2006 in
Fig. 5.
Here we proposed a conceptual framework for the heuristic reasoning of urban
growth patterns based on multiple landscape metrics. They are capable of providing
different aspects of urban sprawl patterns, which go beyond traditional approaches
that are merely based on urban areal statistics. The observable landscape can help
us understand the status of urbanization, behind which more detailed incentives of
policies (i.e., change of socio-economic conditions) are associated (Seto et al. 2012).
4.2 Landscape Network to Understand Their Interactions
Landscape network analysis without full consideration of the physical extent but
simplified with nodes is innovative in the urban study. Small urban patches may
play a crucial role in the overall change of urban landscape patterns at a later time.
176 T. Hu et al.
Landscape network analysis can give a new perspective to exploring this relationship
spatially. For instance, a small urban patch developed in the middle of two existing
patches can link them together as a newly merged area, or an independently grown
patch can stimulate many small patches emerging around it. The network visualizes
the connections among different patches in a graph-based representation, which
can be linked to landscape patterns with more specific indicators that reflect both
coalescence and diffusion at various locations within a specific year. The patch based
spatial interaction to form a complicated urban form is also inspiring to urban growth
modeling, with more detailed consideration of growth types or patterns (Li and
Gong 2016b;Lietal.2014). For example, a patch-based urban growth model which
accord with theories of urban planning and administration can be developed through
considering its growth type of urban patches rather than the traditional grid based
model.
Urban sprawl is a complex process, which is related to many measurable (e.g.,
terrains) or immeasurable (e.g., policy) factors (Batty and Torrens 2005; Seto et al.
2012). Merely considering the relationship between urban area and socio-economic
variables is not sufficient for explaining the underlying causal-relationship, particu-
larly on the heterogeneity of urban land use. The network-based approach can be used
in identifying hotspots of urban patches and is helpful for understanding the under-
lying reasons as illustrated in Sect. 3.4. More efforts should be made in the network
analysis area as the current analysis is purely based on a mathematical graph-based
approach that ignores the influence of terrain and transportation. Many other factors
or features should be considered for a more comprehensive understanding, such as
terrain (e.g., mountains, river) and traffic networks (e.g., highway or subway) (Hu
et al. 2016).
5 Conclusions
The complexity of the urban system, as well as its multi-disciplinary nature, brings a
significant challenge to urban planners and managers. Remote sensing can be used to
monitor the change of urban land, while the connection of urban land expansion with
socioeconomic variables, policies, and planning are rarely systematically explored.
In this study, we used a 30-year record of urban land derived from remotely sensed
images and linked the dynamics of urban land use with policies through a landscape
network analysis. There are six primary growth periods from 1984 to 2013 in Bei-
jing, and they are uneven in time. A dramatically intensive urbanization process has
recently occurred. The analysis of urban landscape networks reveals trajectories of
new development hotspots, which allowed for interpreting possible reasons behind
(i.e., policies). Our results are consistent with present urban policies in Beijing, such
as the establishment of sub-center in Tong Zhou District. This paradigm is help-
ful to understand the urbanization process, as well it is meaningful to model more
complicated urban expansion.
10 Toward a Better Understanding of Urban Sprawl: Linking … 177
References
Batty M, Torrens PM (2005) Modelling and prediction in a complex world. Futures 37(7):745–766
Brown DG, Page S, Riolo R, Zellner M, Rand W (2005) Path dependence and the validation of
agent-based spatial models of land use. Int J Geogr Inf Sci 19(2):153–174
Chen J, Gong P, He C, Luo W, Tamura M, Shi P (2002) Assessment of the urban development
plan of Beijing by using a CA-based urban growth model. Photogramm Eng Remote Sens
68(10):1063–1072
Chen Y, Li X, Zheng Y, Guan Y, Liu X (2011) Estimating the relationship between urban forms
and energy consumption: a case study in the Pearl River Delta, 2005–2008. Landsc Urban Plan
102(1):33–42
Clinton N, Gong P (2013) MODIS detected surface urban heat islands and sinks: global locations
and controls. Remote Sens Environ 134:294–304
Du J, Thill J-C, Peiser RB, Feng C (2014) Urban land market and land-use changes in post-reform
China: a case study of Beijing. Landsc Urban Plan 124:118–128
Foltête J-C, Clauzel C, Vuidel G (2012) A software tool dedicated to the modelling of landscape
networks. Environ Model Softw 38:316–327
Georgescu M, Morefield PE, Bierwagen BG, Weaver CP (2014) Urban adaptation can roll back
warming of emerging megapolitan regions. Proc Natl Acad Sci 111(8):2909–2914
Gong P (2012) Remote sensing of environmental change over China: a review. Chin Sci Bull
57(22):2793–2801
Gong P, Liang S, Carlton EJ, Jiang Q, Wu J, Wang L, Remais JV (2012) Urbanisation and health
in China. The Lancet 379(9818):843–852
Hu T, Yang J, Li X, Gong P (2016) Mapping urban land use by using landsat images and open
social data. Remote Sens 8(2):151
Irwin EG, Bockstael NE (2007) The evolution of urban sprawl: evidence of spatial heterogeneity
and increasing land fragmentation. Proc Natl Acad Sci 104(52):20672–20677
Ji W, Ma J, Twibell RW, Underhill K (2006) Characterizing urban sprawl using multi-stage remote
sensing images and landscape metrics. Comput Environ Urban Syst 30(6):861–879
Li X, Gong P (2016a) An “exclusion-inclusion” framework for extracting human settlements in
rapidly developing regions of China from Landsat images. Remote Sens Environ 186:286–296
Li X, Gong P (2016b) Urban growth models: progress and perspective. Sci Bull 61(21):1637–1650
Li X, Gong P, Liang L (2015) A 30-year (1984–2013) record of annual urban dynamics of Beijing
City derived from Landsat data. Remote Sens Environ 166:78–90
Li X, Liu X, Yu L (2014) A systematic sensitivity analysis of constrained cellular automata model for
urban growth simulation based on different transition rules. Int J Geogr Inf Sci 28(7):1317–1335
Liu X, Li X, Tan Z, Chen Y (2011) Zoning farmland protection under spatial constraints by inte-
grating remote sensing, GIS and artificial immune systems. Int J Geogr Inf Sci 25(11):1829–1848
Liu X, Ma L, Li X, Ai B, Li S, He Z (2014) Simulating urban growth by integrating landscape
expansion index (LEI) and cellular automata. Int J Geogr Inf Sci 28(1):148–163
Liu Z, He C, Zhang Q, Huang Q, Yang Y (2012) Extracting the dynamics of urban expansion in China
using DMSP-OLS nighttime light data from 1992 to 2008. Landsc Urban Plan 106(1):62–72
McDonald RI, Weber K, Padowski J, Flörke M, Schneider C, Green PA, Gleeson T, Eckman
S, Lehner B, Balk D, Boucher T, Grill G, Montgomery M (2014) Water on an urban planet:
urbanization and the reach of urban water infrastructure. Glob Environ Change 27:96–105
McGarigal K, Cushman SA, Ene E (2012) FRAGSTATS v4: spatial pattern analysis program for
categorical and continuous maps
McGarigal K, Marks BJ (1995) Spatial pattern analysis program for quantifying landscape structure.
In: General Technical Reports, PNW-GTR-351. US Department of Agriculture, Forest Service,
Pacific Northwest Research Station
Mertes CM, Schneider A, Sulla-Menashe D, Tatem AJ, Tan B (2015) Detecting change in urban
areas at continental scales with MODIS data. Remote Sens Environ 158:331–347
178 T. Hu et al.
Ou J, Liu X, Li X, Chen Y (2013) Quantifying the relationship between urban forms and carbon
emissions using panel data analysis. Landscape Ecol 28(10):1889–1907
Schneider A, Friedl MA, Potere D (2010) Mapping global urban areas using MODIS 500-m data:
new methods and datasets based on ‘urban ecoregions’. Remote Sens Environ 114(8):1733–1746
Schneider A, Seto KC, Webster DR (2005) Urban growth in Chengdu, Western China: application
of remote sensing to assess planning and policy outcomes. Environ Plan 32(3):323–345
Seto KC, Fragkias M (2005) Quantifying spatiotemporal patterns of urban land-use change in four
cities of China with time series landscape metrics. Landscape Ecol 20(7):871–888
Seto KC, Kaufmann RK, Woodcock CE (2000) Landsat reveals China’s farmland reserves, but
they’re vanishing fast. Nature 406(6792):121
Seto KC, Reenberg A, Boone CG, Fragkias M, Haase D, Langanke T, Marcotullio P, Munroe
DK, Olah B, Simon D (2012) Urban land teleconnections and sustainability. Proc Natl Acad Sci
109(20):7687–7692
Sexton JO, Song X-P, Huang C, Channan S, Baker ME, Townshend JR (2013) Urban growth of the
Washington, D.C.–Baltimore, MD metropolitan region from 1984 to 2010 by annual, Landsat-
based estimates of impervious cover. Remote Sens Environ 129(0):42–53
Sohl TL, Sayler KL, Drummond MA, Loveland TR (2007) The FORE-SCE model: a practical
approach for projecting land cover change using scenario-based modeling. J Land Use Sci
2(2):103–126
Taubenböck H, Wiesner M, Felbier A, Marconcini M, Esch T, Dech S (2014) New dimensions of
urban landscapes: the spatio-temporal evolution from a polynuclei area to a mega-region based
on remote sensing data. Appl Geogr 47:137–153
United Nations (2018) World urbanization prospects: the 2018 revision, UN
Wang L, Li CC, Ying Q, Cheng X, Wang XY, Li XY, Hu LY, Liang L, Yu L, Huang HB, Gong
P (2012) China’s urban expansion from 1990 to 2010 determined with satellite remote sensing.
Chin Sci Bull 57(22):2802–2812
Wu J (2004) Effects of changing scale on landscape pattern analysis: scaling relations. Landscape
Ecol 19(2):125–138
Wu Q, Li H-Q, Wang R-S, Paulussen J, He Y, Wang M, Wang B-H, Wang Z (2006) Monitoring
and predicting land use change in Beijing using remote sensing and GIS. Landsc Urban Plan
78(4):322–333
Yu X, Ng C (2006) An integrated evaluation of landscape change using remote sensing and landscape
metrics: a case study of Panyu. Int J Remote Sens (Guangzhou) 27(6):1075–1092
Zhang Q, He K, Huo H (2012) Policy: cleaning China’s air. Nature 484(7393):161–162
Zhang Q, Seto K (2013) Can night-time light data identify typologies of urbanization? A global
assessment of successes and failures. Remote Sens 5(7):3476
Zhou QM, Sun B (2010) Analysis of spatio-temporal pattern and driving force of land cover change
using multi-temporal remote sensing images. Sci China Technol Sci 53:111–119
Zhou Y, Smith SJ, Elvidge CD, Zhao K, Thomson A, Imhoff M (2014) A cluster-based method to
map urban area from DMSP/OLS nightlights. Remote Sens Environ 147:173–185
Chapter 11
Correlating Household Travel Carbon
Emissions, Travel Behavior and Land
Use: Case Study of Wuhan, China
Jingnan Huang, Ming Zhang and Ningrui Du
Abstract Domesticity contributes a significant portion to the amount of total CO2
emissions. Understanding the influence factors of household travel carbon emissions
helps develop effective policy to minimize carbon emissions. This paper presents an
empirical study of household travel carbon emissions with 1194 samples in Wuhan,
China. Besides looking at the household socioeconomic characteristics, the study
pays attention to the role of the spatial context in household living and travel and
how it affects travel emission outcomes. A regression analysis shows urban spatial
structure and land use context offer additional explanatory power to variation of travel
carbon emissions under the effects of socio-economic factors controlled. Emission
hot spots and high-emission households most likely appear in newly developed sub-
urban areas. The paper concludes by suggesting both place-based and people-based
policies to achieve the goal of reducing carbon emissions.
Keywords Household travel carbon emissions ·Spatial correlation analysis ·
Travel modes ·Mixed land use ·Wuhan China
1 Introduction
Reducing energy consumption and carbon dioxide (CO2) emissions has topped the
national agenda of many countries around the world since the early 21st century
(International Energy Agency or IEA 2011;Defra2008). China’s central govern-
J. Huang ·N. Du (B)
School of Urban Design, Wuhan University, 8 Donghu Road South, Wuchang, Wuhan
430072, Hubei Province, China
e-mail: 112598782@qq.com
J. Huang
e-mail: huangjn73@qq.com
M. Zhang
School of Architecture, The University of Texas at Austin, Austin, USA
e-mail: zhangm@austin.utexas.edu
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_11
179
180 J. Huang et al.
ment has set a goal of reducing its per capita CO2emissions to 40–50% of the
2005 emission rate by 2020. However, China is fighting an uphill battle in achiev-
ing its goal because of at least three developing trends. First, the country’s rapid
urbanization will continue over the next twenty years. United Nation Development
Programme (UNDP) (2013) estimated that by 2030, there will be one billion urban
residents in China and the country’s urbanization level will increase from its cur-
rent 50–70%. High energy consumption and high carbon emissions are one of the
prominent features of this development stage (Un-Habitat 2012). Second, the aver-
age household income continues to rise. This rising income will lead to changes
in household consumption. Families will move from being clothing-and-food-based
to being housing-and-travel-based (Zhang 2010), which means that there will be a
growing demand for home electricity and other forms of energy consumption. Third,
personal mobility continues to grow. This is evident in the rapidly increasing rate of
private car ownership and motorcycle ownership. The central and local governments
cannot avoid these pressing trends. Instead, they must seek effective strategies to
abate carbon emissions.
Household carbon emissions make up a significant portion of China’s total emis-
sions. Studies show that 40% of total urban carbon emissions in developed countries
come from household living and travel activities (Kerkhof et al. 2009). In China, the
figure is slightly lower, but it has been increasing (Glaeser and Kahn 2010). From
1995 to 2004, the household share of urban emissions in China rose from 19 to 30%
(Yan and Minjun 2009). International experience indicates that, as urbanization con-
tinues, household carbon emissions will likely increase (IEA 2011; Qiong 2009).
Hence, it is important for policymakers to understand factors driving the changes
in household carbon emissions so as to develop effective strategies that minimize
household carbon emissions and achieve the national emission reduction goal in
China.
This study aims to contribute to the knowledge base by presenting an empirical
investigation of household travel emissions in China. Aside from looking at house-
hold socioeconomic characteristics, the study pays attention to the role of the spatial
context in household living and travel in order to evaluate how it affects travel emis-
sion outcomes. Studying the land use features (or urban form or the built environ-
ment—all three terms are used interchangeably in the paper) will offer suggestions
for lowering urban household carbon emissions (Qin and Shao 2011; Liu et al. 2017).
There have been innovative planning efforts internationally, such as New Urbanism,
Smart Growth, and Compact City, whose direct planning goals target lowering the
impacts of human activities on the environment and energy consumption (Cervero
2002; Chen 2010; Khattak and Rodriguez 2005).
The paper consists of six parts. Following this introduction, the paper reviews
relevant literature on household travel carbon emissions. Thirdly it is followed by
the study methodology. Fourthly, study results are presented. Fifthly, discussions of
the findings and their implications are offered. Lastly, implications are drawn for
policy making for the purpose of reducing household travel emissions.
11 Correlating Household Travel Carbon … 181
2 Literature Review
2.1 Household Travel Carbon Emissions
Household carbon emissions can be divided into two types: household living carbon
emissions and household travel carbon emissions. Household travel carbon emis-
sions are generated by the energy consumption related to a household member’s
travel, including daily commuting to work and school and travel for shopping by
various motor vehicles such as plane, cars, buses, and underground subway. House-
hold daily commuting carbon emissions are a vital component of total household
carbon emissions. Most research is focused on studying the relationship between
travel carbon emissions and the socioeconomic attributes. Their studies found that
household income is the largest influence on household travel carbon emissions.
According to Ko et al. (2011), the top 10% of high-carbon residents caused 63%
of the travel carbon emissions in Seoul, South Korea. Similar result were found in
Germany that the highest income household emit 4.25 times as much CO2than the
lowest (Miehe et al. 2016).
Spatial factors also matter concerning household travel carbon emissions. Zahabi
et al.’s (2012) study of Montreal, Canada discovered that the urban pattern and
transportation accessibility strongly influence urban travel carbon emissions. They
found that for every 10% increase in traffic accessibility, the travel carbon emissions
were reduced 5.7% (Zahabi et al. 2012). The study of Wang et al. (2017)inXi’an
and Bangalore indicates that higher density, more compact urban pattern, shorter
commuting distances, higher transit shares and more clean energy vehicles bears a
positive impact on lower carbon emissions. Based on a survey of 2045 families in
33 Guangzhou communities, Jiang et al. (2013) found that the larger the scale that a
single-function residential community is, the larger the household carbon emissions
amount is. The researchers also discovered that traffic carbon emissions rose with
the increase of distance to the city center (Jiang et al. 2013).
2.2 Residents’ Travel Behavior and Household Travel
Carbon Emissions
In China the amount of carbon emissions caused by travel is gradually increasing, as
showed by Chai et al.’s (2012) study of household travel carbon emissions in Beijing.
They found that the emissions amount was influenced primarily by household travel
distance and travel modes (Chai et al. 2012). The study of Yang et al.’s (2018)
found that residents’ mode choice is closely related to their characteristics and public
transport improvements can help reduce carbon emissions. Generally, long-distance
travel brings about more pressure on urban transportation systems. Basic residents’
travel features, such as total travel amount and travel mode choice can affect urban
traffic carbon emissions.
182 J. Huang et al.
Travel means bear close relationship to travel carbon emissions because the mas-
sive use of high-carbon-emitting vehicles, such as private cars and taxis, quickly
increase the amount of carbon emissions. Some scholars have attempted to analyze
the relationship between residents’ travel behavior and carbon emissions in the effort
to control traffic carbon emissions. In their study in Beijing, Wang and Liu (2015)
examined the features and driving factors of CO2emissions from household daily
travel, and pointed out that transportation intensity and mode share were the impor-
tant factors for household daily travel CO2emissions. In seven Chinese provinces,
the changes in residents’ living habits, particularly the habit of private traffic expen-
diture, was a critical factor that resulted in the increase of carbon emissions (Yan and
Minjun 2009). After exploring the relationship of transportation and carbon emis-
sions, Banister (2011) held that changes in people’s transportation travel behavior
can influence travel carbon emissions over the long term. The author argued that
providing easily accessible public transportation can help in efforts to reduce carbon
emissions (Banister 2011).
2.3 Land Use and Residents’ Travel Features and Household
Travel Carbon Emissions
Land use mix refers to locating different types of land close together that their
functions are closely bound with residents’ life in a specific urban block. In accor-
dance with influencing residents’ travel behavior, land use can be classified into
housing, office space, commercial land, retailing land, education land, and partial
industrial land. In urban area, density, regional accessibility, land use mixture and
roadway connectivity all affect travel behavior (Litman and Steele 2017). It is gener-
ally acknowledged that mixed land use impacts the distribution of residents’ working,
living, shopping, and entertainment destinations and the distribution of transporta-
tion facilities to reach these destinations, which then affects urban residents’ travel
needs (Mindali et al. 2004). The study showed that residents are more inclined to use
non-motor vehicles and public transportation in blocks with more public facilities
especially with office and commercial land use types (Wu 2010).
Mixed land use can affect travel energy consumption by travel distance and travel
structure. Zhang and Zhao (2017) presented in their research in Beijing that high land-
use diversity and good jobs-housing balance significantly reduces commuting travel
and generates less travel energy consumption. Xiao et al. (2017) pointed out that
population density, land use mix and access to metro stations have negative impact
on emissions and household travel emission increased along with the residential
distance to the city center. In their examination of the relation of community spatial
features and residents’ travel carbon emissions, Chai et al. (2012) stated that physical
spatial organization and reorganization means such as mixed land use and service
facilities should be adopted to regulate and optimize urban spatial arrangement so as
to construct a low-carbon urban spatial structure.
11 Correlating Household Travel Carbon … 183
There is a great quantity of research on household carbon emissions, yet there
remain gaps in the knowledge base. Current research mainly focuses on the aggre-
gate socioeconomic factors influencing carbon emissions. Few have explored the
spatial dimension of emissions at the disaggregate level. Understanding the spatial
characteristics of carbon emissions is important as it offers great potential for emis-
sion reduction through spatial decision-making in land use planning, transportation
network operation and management.
3 Methodology
3.1 Study Case
The selected case for this study is Wuhan, the capital of Hubei Province. It is one of
the China’s most important industrial bases, transportation hubs, and education and
research centers. The municipality of Wuhan includes seven urban districts and six
suburban districts, covering a total area of 8494 km2. The Yangtze and Han Rivers
cross through the city center. At the end of 2017, Wuhan had a total population
of 10.89 million (8.7 million in the urban districts), occupying 678 km2(Wuhan
Municipal Bureau of Statistics 2018). In recent years, Wuhan’s new development
has revolved around three national development zones: Wuhan Economic and Tech-
nological Development Zone (locally known as Zhuankou) in Hanyang, East Lake
New Technological Development Zone in Wuchang (locally known as Donghu),
and Wujiashan Economic and Technological Development Zone in Hankou (locally
known as Wujiashan). These three national development zones and the Wuhan Iron
and Steel Cooperation (locally known as Qingshan District) constitute the growth
poles of the Wuhan economy (Fig. 1).
Wuhan’s public transportation system covers most of the inner city. The main
traffic mode is bus; in the survey, which was conducted in 2011, there was only one
light rail as urban rail transit in operation. The enlarged urban area and increased liv-
ing standards have spurred Wuhan’s motorization. Its motorization level has soared
rapidly over the past thirty years. In 1983, Wuhan had 62,000 registered motor vehi-
cles, while in 2012 increased to 1.84 million (Wuhan Statistics Bureau 1978–2012).
The substantial growth in the number of motor vehicles has led to a rapid rise in
traffic carbon emissions. From 2005 to 2009, the number of privately owned cars
and CO2emissions in Wuhan have both doubled, and traffic-related CO2emissions
have risen by approximately 53% (Huang and Wu 2012). Controlling or lowering
travel carbon emissions, thus, becomes a key point of focus for Wuhan in its effort
to build a low-carbon-emitting sustainable city.
184 J. Huang et al.
Fig. 1 Wuhan city map
3.2 Data Sources
A structured survey was designed and carried out in Wuhan from September to
December in 2010. The survey distributed 1504 questionnaires and obtained 1194
valid samples after incomplete responses were removed. The factors of sample selec-
tion points are Wuhan’s urban spatial structure, the distribution of residential blocks,
and population density (Fig. 2). The survey questionnaire includes five parts: house-
hold living environment, household daily commuting, household living energy con-
sumption, basic household information, and residents’ awareness of energy con-
servation and emission reduction. Household daily travel data involves all major
household members’ commuting characteristics, such as travel modes, travel time,
11 Correlating Household Travel Carbon … 185
Fig. 2 Distribution of sample points
and travel distance, which are the basis for calculating household commuting carbon
emissions. Table 1reports the sample’s descriptive statistics.
3.3 Calculation of Household Commuting Emissions
Household commuting emissions were calculated based on travel modes used, travel
distances, fuel consumption, and rates of emissions from secondary sources. Consid-
ering local user and transportation operating characteristics as well as the availability
of data, there were two base equations for emission calculation:
Emissions =the amount of fuel consumed
∗the fuel carbon emission factors (1)
Emissions =vehicle mileage
∗carbon emission factors of different vehicle types (2)
For calculation of emissions by private cars, Eq. (1) was applied. For travel by
public transit or other modes, the study utilized travel distance and other information
186 J. Huang et al.
Tabl e 1 Sample descriptive characteristics
N Min Max Mean Std. Dev.
Household member #1 age
(years)
1194 21.00 82.00 41.77 7.33
Household member #2 age
(years)
1146 6.00 75.00 39.83 7.09
Household member #3 age
(years)
1080 1.00 61.00 14.93 6.56
Year home built 1194 1.00 4.00 3.21 0.90
Building area of home (m2)1194 4.00 500.00 95.35 43.82
Household size (persons) 1194 1.00 7.00 3.27 0.91
Household member #1
educational level
1194 1.00 6.00 2.88 1.24
Household member #2
educational level
1145 1.00 6.00 2.58 1.13
Monthly household income
(category)
1194 1.00 7.00 3.29 1.05
With private cars 1194 0 1 0.256 0.436
Notes (a) Household Members: #1: Household Head, #2: Spouse of Household Head, #3: Others
in household (e.g., child or grandparents)
(b) Monthly Household Income (Yuan): 1: ≤1000, 2: 1001–3000, 3: 3001–5000, 4: 5001–10,000,
5: 10,001–20,000, 6:20,001–40,000, 7: >40,000
(c) Educational Level: 1: 9 years (intermediate school) or less, 2: High-school or professional school,
3: Associate Degree, 4: Bachelor’s, 5: Master’s, 6: Ph.D.
(d) Years Home Built: 1: before 1980, 2: 1980–1990, 3: 1990–2000, 4: 2000–2010
obtained from the survey and applied Eq. (2) to estimate emissions. Carbon emissions
factors of different types of fuels and vehicles referred to the database of Green House
Gas (GHG) emissions factor released by Intergovernmental Panel on Climate Change
(IPCC) and GHG Protocol (GHG Protocol 2005;IPCC2006).
Specific emission factors are shown below.
Bus
Estimating bus emissions takes into consideration bus vehicle type, peak load, and
bus fuel efficiency. It was calculated with Eq. (3):
C(bus)=α1∗β1∗V1 ∗ρ1
M(3)
CCO
2emission factor, kgCO2/person-km
α1dieselCO
2emission coefficient; it takes a value of 74 per IPCC (2006) guides
β1 net caloric value of diesel, 43 TJ/Gg, 100 kgCO2/TJ
V1 diesel fuel consumption per km for buses in Wuhan
ρ1 diesel fuel density, kg/L
M average bus load during peak hours in Wuhan.
11 Correlating Household Travel Carbon … 187
In Wuhan, public buses come from three companies: Jinglong, Yangtze, and
Yutong. Their fuel efficiency ranges from 0.3 to 0.4 L per vehicle km. This study
used the average of 0.35. Field studies reported an average bus load of 55 persons
per bus. By the end of the survey, Wuhan buses used diesel fuel. The grade #0 diesel
has a density of 0.835 kg/L. With the above input data and references, the emission
factor for public buses in Wuhan was estimated at 0.0169 kg CO2/person-km.
Chartered buses and employer bus services are common in Chinese cities. The
study used Jinglong, the dominant bus brand, for emission factor calculation. Jinglong
has a fuel efficiency of 0.364 L/km. Furthermore, chartered buses and employer buses
tend to have lower bus loads. This study took a value of 45 people per bus. As a result,
the emission factor was estimated at 0.0215 kg CO2/person-km for chartered buses
and employer bus services.
Total commuting bus emissions for one year were given by the following Eq. (4):
G(Bus)=T
60 ∗2∗5∗52 ∗γ∗0.0169 (4)
T reported commuting time in minutes in one direction
γthe bus’s average speed.
Private Cars and Motorcycles
C(private cars)=α2∗β2∗ρ2(5)
CCO
2emission factor, kgCO2/Liter
α2CO
2gasoline emission coefficient, 69
β2 net heat value, 44.3 TJ/Gg, 300 kgCO2/TJ
ρ2 gasoline density, for grade #93: 0.725 kg/L.
For Wuhan, the calculated value of C was 2.226 kgCO2/L. Total commuting
emissions for private cars and motorcycles for one year were given by monthly
consumption of gasoline multiplied by C for 12 months.
Taxi
C(Taxi)=B(1−μ)P(6)
B gasoline consumption per km, 0.088 L
μTaxi vacancy rate, 36% for Wuhan (WLRUL 2011)
P 0.1253 kgCO2/person-km.
Total commuting emissions by taxi for one year were given by the following
Eq. (7):
G (Tax) =S
n∗0.1253 ∗12 (7)
188 J. Huang et al.
S represents the monthly expense of taking taxis
n is the taxi fare.
In Wuhan, the base fare is 6 yuan for distance of 3 km. After the first 3 km, the
fare is 1.5 yuan/km. The fare goes up to 2.2 yuan/km beyond 8 km.
Electric Scooters
C (Electric Scooters)=A∗D(8)
CCO
2emission factor, kgCO2/km
A electricity consumption/km
D Emission coefficient of electricity.
Most two-wheel electric scooters have a battery of 0.576 kw-h. A new battery, on
average, lasts 50 km. Electricity consumption is 11.52 wh/km. The emission coeffi-
cient is 0.92 kgCO2per kw-h. This gave an emission factor of 0.01 kg CO2/person-km
for electric scooters. Total commuting emissions by electric scooters for one year
were estimated in the same way as they are for buses except for the use of the
mode-specific emission coefficient.
Light Rail
At the time of the survey, Wuhan had one elevated light rail system partially in
operation. Not much emission data can be obtained. Therefore, this study borrowed
data from Beijing and derived an emission rate of 0.0091 kgCO2/person-km. Total
commuting emissions by light rail for one year were estimated in the same way as
for buses except for the use of the light rail emission coefficient.
3.4 Spatial Autocorrelation Analysis
In this study, spatial autocorrelation was used to analyze the spatial distribution fea-
tures of household carbon emissions in Wuhan. According to Goovaerts et al. (2005),
there are global and local indicators to measure spatial autocorrelation. Global auto-
correlation indicators reflect the degree of similarity of the attributes of neighboring
geospatial units (Zhang and Zhang 2007). Local autocorrelation, or adopting local
indicators of spatial association (LISA), describes the degree of correlation, such
as clustering relationships and spatial heterogeneity, on a similar attribute shared
between a specific geographic area unit and its surrounding units (Ord and Getis
1995; Zhang and Zhang 2007).
Of the various global autocorrelation indicators, the researchers selected the most
commonly used indicators—the Moran’s I index. Varying from 1 to −1, Moran’s I
index reflects the transition from adjacent and similar positive correlations to dis-
adjacent and dissimilar negative correlations. When I =0, there is no correlation in
the observed value of the studied adjacent spaces.
11 Correlating Household Travel Carbon … 189
LISA analysis essentially converts Moran’s I index into the area units. Z-Score
value was used to measure the relationship between the unit and its surrounding
units. The greater the Z-Score is, the stronger its correlation with the surrounding
units will be and vice versa.
This study also conducted hot spot analysis to examine the household’s spatial
distribution patterns at the disaggregate scale. We calculated Getis-Ord Gi statistics
of every data-concentrated element and obtained a Z-Score as well as pvalues to
identify the clustering high- or low-value locations. For a statistically significant
positive Z-Score, the higher the Z-Score is the more closely the hot spots will cluster.
3.5 Characterization of Travel Behavior
The analysis of household residents’ traffic travel behavior characteristics and carbon
emissions was carried out on two levels: the household and individual household
members. The researchers first calculated the household members’ travel factors,
such as travel carbon emissions, travel distance, and travel time, represented the
differences in members’ travel features with graphics, and, finally, discussed their
correlation. When studying travel features, it is of vital importance to calculate travel
time and travel distance. This study designed two methods for the estimation of time
and distance. Initially, the interviewees provided travel distances on their question-
naire. However, these travel distances were estimated, which were a certain devi-
ation from the actual travel distances. As a result, they were treated as references.
The researchers chose the travel distances calculated by inputting residence places,
workplaces, or schools listed on the questionnaire with the distance measurement
tools of Google Earth. The traffic distances acquired from Google Earth are not
merely Euclidean distances between two places but an actual one that follows the
most frequently-used traffic routes, which can be trusted as a more accurate calcu-
lation. The confirmation of travel time was gained by combining the set travel time
estimations in the questionnaires and the time estimation of selected routes according
to Google Earth.
3.6 Land Use Context
There are many ways to quantify land use context in which travel and the conse-
quent emissions take place (Boarnet and Crane 2001; Ewing and Clemente 2013).
In this study, six indicators were developed based on the surveyed information and
secondary GIS data. They were: land use mixture within 500-m and 1-km distances
from home locations, index of public services available near home locations, popula-
tion density in the neighborhood, distance from home to the nearest bus stop, number
of bus routes within 500 m to homes, and number of street intersections.
190 J. Huang et al.
Tabl e 2 Principle
component analysis of land
use context indicators
Component
F1 F2
Land use mixture near homes 0.492 −0.129
Public services availability near homes 0.115 0.114
Number of bus routes hear homes 0.383 0.010
Distance from home to nearest bus
stop
−0.442 0.094
Number of street intersections −0.071 0.550
Population density −0.134 0.581
Note Numbers in bold correspond to the land use indicators that
are the principle components of composite factor F
Given the wide use of information entropy in calculating degree of mixed land
use, this article referred to the definition of entropy provided by Frank and Pivo
(1994) to calculate degree of mixed land use. The formula is as follows:
Land Use Mix =(−1)[(b1/a)ln(b1/a)+(b2/a)ln(b2/a)
+···+ (bi /a)ln(bi /a)](9)
a total acreage of land use types influencing residents’ travel in the buffer areas;
bi acreage of one type of land use that affects residents’ travel in the buffer areas;
i number of land types that affect residents’ travel in the buffer areas.
Correlation analysis showed that these six indicators were correlated. A principle
component analysis (PCA) was then carried out. Table 2reports the PCA output. Two
composite factors were generated. Factor 1 (F1), a functional factor, was derived from
land use mixture, number of bus routes, and number of bus stops. Factor 2 (F2), a
density factor, was derived from population density and intersection density.
4 Results
4.1 Socio-demographic Analysis of Household Travel
Carbon Emissions
Figure 3shows a cross-tabulation of household members by shares of travel modes.
Household member #1, the householder, had the largest share (22.1%) of using
personal travel modes (cars or motorcycles). The personal modal shares for household
member #2 (the householder’s spouse) and #3 (the couple’s child or grandparents)
were 14.0 and 7.2%, respectively. Household member #3 used public transit and
non-motorized modes more than other household members. The kind of modal split
11 Correlating Household Travel Carbon … 191
Fig. 3 Commuting modes of household members
distributions among household members indicate the linkage between the household
members’ roles and their levels of personal mobility. The householder, who is the
person with the most significant household responsibilities, tends to have the highest
personal mobility (Fig. 3).
The householder’s high personal mobility is revealed by his/her travel distances
and times compared to other household members. The average travel distance of
household member #1 was 6.32 km, longer than that of household members #2 and
#3, which were 5.46 and 3.53 km, respectively. The average travel time of household
member #1 was 23.52 min, which was about one minute longer than household
member #2 and five minutes longer than household member #3 (Table 3).
With regard to the household members’ travel distances, 45% were shorter than
2 km, while 72% (45% +27%) within 5 km, and nearly 90% within 10 km. Household
member #1 had a more balanced travel range with 38% of travel within 0–2 km, 25%
within 2–5 km, and 34% within 5–30 km. Household member #2 made mostly local
commutes, with 70% within 0–5 km and over 25% within 5–30 km. Household
member #3 had the most centralized travel distance. Over 50% of their travel was
within 0–2 km, which is the highest proportion in the household, and 30% within
2–5 km (Table 3).
192 J. Huang et al.
Tabl e 3 Commuting and emission characteristics of household members
Household
member
Commuting
characteristics
N Min Max Mean Std. Dev.
#1 Distance (km) 1107 0.00 80.00 6.32 8.52
Time (min) 1107 0.00 180.00 23.52 21.47
Carbon Emission
(kg/year)
1107 0.00 13,500.00 671.33 1542.33
#2 Distance (km) 959 0.00 80.00 5.46 7.47
Time (min) 959 0.00 180.00 22.57 21.12
Carbon Emission
(kg/year)
959 0.00 13,500.00 162.66 753.78
#3 Distance (km) 874 0.10 60.00 3.53 5.30
Time (min) 874 0.00 120.00 18.06 16.07
Carbon Emission
(kg/year)
874 0.00 5778.00 30.53 225.15
Under the international context, Chinese urban commuters have the travel distance
30–50% less than that of American and British. As income continues to rise, the gap
in commuting time between China and the developed countries will likely reduce.
Given the differences in personal mobility among household members, it is no
surprise to see similar differences among them in travel carbon emissions. The travel
carbon emissions of household member #1 were 671.33 kg per year, accounting for
77% of the household total. Household member #2 generated 162.66 kg per year,
accounting for 19%; and household member #3 made the least, creating 30.53 kg
per year, which is a 4% share (Table 3).
The average value of each household’s travel carbon emissions of the 1194 sur-
veyed samples is 811 kg per year (note that the survey did not have information on
carpooling. The estimate of household travel carbon emissions by summing up indi-
vidual household members’ emission involved possible double-counting of house-
hold members who shared vehicles for travel). High-carbon-emission travel means
that vehicles such as cars or motorcycles are the major sources of household travel
carbon emissions, accounting for 82% of the total travel carbon emissions. Taxis
were second place with 12%, and low-carbon-emission traffic tools such as electric
motors (scooters), buses, and light rails accounted for 6%.
4.2 Spatial Analysis
4.2.1 Spatial Clustering of Emissions
The calculated Moran’s I index is 0.262 (>0), with a Z-score of 3.19 (>2.58). This
suggests that the commuting carbon emissions of Wuhan families present a significant
11 Correlating Household Travel Carbon … 193
Fig. 4 Spatial clusters of household travel carbon emissions
spatial autocorrelation on the global level. A local spatial autocorrelation analysis
of household commuting carbon emissions displayed that the spatial clustering of
most areas, especially in the urban center, was insignificant. But there were some
additional cases, such as a cluster of high values in Zhuankou, South Lake, Guanggu,
and Changqinghuayuan (Fig. 4).
Household commuting carbon emissions’ hot spot analysis illustrated that the hot
spot appeared in Zhuankou, while the cold spot areas were mainly distributed in
Wangjiadun, Wujiashan and Qingshan (Fig. 5). A closer look at the emission hot and
cold spots gives a more detailed picture of spatial distributions of emissions in the
Wuhan subareas.
By comparison of household travel distances, the shortest one of the hot spots in
the Zhuankou Development Zone was 2.3 km, higher than that of the cold spots in
Qingshan, Wangjiadun and Wujiashan, which were 1.2, 0.5, and 2 km, respectively.
At the same time, the maximum value in these cold spot areas were respectively
72.6, 47.2, and 97.5 km. All values are higher than the maximum value in Zhuankou,
which is 42.9 km. The average value of travel distance showed the same results,
indicating that cold spot areas are more diversified with a larger range of residents’
activities than that in hot spot areas (Table 4).
By comparing household travel modes, 9.5% of residents in the hot spots of
Zhuankou used public transport, which is much lower than the cold spots Qing-
shan (42.1%), Wangjiadun (22.5%), and Wujiashan (32.7%). Comparing the pro-
194 J. Huang et al.
Fig. 5 Emission hot and cold spots
Tabl e 4 Commuting distances of carbon emission hot spots and cold spots
Hot/cold spot Min Max Mean Std. Dev. N
Hot spot Zhuankou 2.30 42.9 9.80 10.21 19
Cold spots Wuji as han 297.5 16.98 22.15 55
Wangjiadun 0.5 47.2 10.23 7.91 117
Qunigshan 1.2 72.6 20.28 18.33 28
portions of residents choosing cars as traffic tools, Zhuankou had the highest pro-
portion (<28.6%), while Qiangshan, Wangjiadun and Wujiashan reached 12.5, 9.0,
and 26.9%. In the ratio of choosing motorcycles as traffic tools, cold spot areas were
relatively average, with Qingshan of 13.6%, Wangjiadun of 9.0%, and Wujiashan of
9.6%. However, the ratio of hot spot Zhuankou was 23.8%, which is considerably
higher than the average in cold spots. Overall, most Zhuankou families pick motor
vehicles (cars and motorcycles) as travel modes. Meanwhile, there are no obvious
differences between cold spots and hot spots with regard to walking or biking. Hot
spot Zhuankou (38.1%) was higher than Qingshan (31.8%) and Wujiashan (30.8%).
Many Wangjiadun residents (59.5%) chose to travel by foot or bike (Fig. 6).
11 Correlating Household Travel Carbon … 195
Fig. 6 Commuting mode of carbon emission hot and cold spots
4.2.2 Correlation Analysis of Land Use Mix and Household Travel
Carbon Emissions
The urban central areas typically had high levels of mixed land use. This was evident
in Wuhan’s traditional commercial areas of Wuchang, Hankou, and Hanyang. It was
also observed in the strip commercial corridors, especially on the two sides of the
internal ring roads that link Wuchang, Hankou, and Hanyang. Furthermore, there was
a higher degree of mixed land use in the urban sub-centers being built such as Optics
Valley (Guanggu) and Wujiashan. The degree of mixed land use was generally low in
the urban periphery, single-function, large residential areas and old industrial areas.
These areas include Jinyinhu, Zhuankou, Baibuting, Nanhu, and Qingshan (Fig. 7).
The correlation analysis between land use mix and residents’ traffic carbon emis-
sions showed that household monthly income and land use mix correlated nega-
tively with travel carbon emission at the 0.05 level of statistical significance (corre-
lation coefficient is −0.063). It means that the higher the mixed degree of land use,
the lower level of travel carbon emissions, and vice versa. In contrast, household
monthly income bore an obvious relationship to household travel carbon emissions
and degree of mixed land use, showing a statistical significance of 0.01. This means
that the higher the household monthly income, the higher the families’ travel carbon
emissions and the higher the degree of mixed land use (Table 5).
4.2.3 Influence Factor Analysis of Household Travel Carbon Emissions
Multiple regression analyses were carried out to examine the joint effects of sociode-
mographic and spatial factors on household travel carbon emissions. Table 6reports
the final model. The regression coefficients allowed us to exam the partial effects of
one independent variable on emissions when the influence of other variables was con-
196 J. Huang et al.
Fig. 7 Land use mix of area 1 km from household
Tabl e 5 Correlation analysis of land use mix, travel carbon emission, and household monthly
income
Commuting carbon
emission
Household monthly
income
Household monthly
income
Pearson correlation 0.377**
Sig. (2-tailed) 0.000
Land use mix Pearson correlation −0.063* 0.091**
Sig. (2-tailed) 0.029 0.002
**. Significant correlation at 0.01 level (2-tailed)
*. Significant correlation at 0.05 level (2-tailed)
N=1194
trolled. The standardized coefficient enabled us to compare the relative importance
of the independent variables in explaining the variation in travel carbon emissions.
Some observations can be made from Table 6.
•Income (beta coefficient: 0.355) was the most influential factor in explaining
household travel emissions.
•Emission amount was positively correlated with the age and educational levels of
both household members #1 and #2.
•Those employed by the private sector or self-employed tended to generate more
travel emissions than those working in the governmental or public sectors.
•The composite measures of land use context had additional effects on travel emis-
sions after the effects of income, age, education, and occupational factors were
11 Correlating Household Travel Carbon … 197
Tabl e 6 Multiple regression analysis of influence factors of Household commuting carbon emis-
siona
Non
standardized
coefficient
Standardized
coefficient
TSig.
BStd. Err. Beta
Constant 647.150 72.060 8.981 0.000
Household
monthly income
602.376 54.606 0.355 11.031 0.000
Age group of
household
member #2
122.911 52.657 0.070 2.334 0.020
Educational level
of household
member #1
125.950 56.338 0.074 2.236 0.026
Educational level
of household
member #2
216.756 62.078 0.116 3.492 0.001
Work U n i t of
household
member #1b
181.944 53.065 0.107 3.429 0.001
Work U n i t of
Household
member #2
190.395 68.835 0.085 2.766 0.006
Factor 1
(functional)
−127.388 50.724 −0.075 −2.511 0.012
Factor 2 (density) −129.784 50.619 −0.076 −2.564 0.011
R2=0.205
N=1194
aDependent variable: Household travel carbon emissions
bWork Unit refers to the type of employer. 1 if private sector; 0 all other sectors, including govern-
ments, state or public-owned enterprises
controlled. Higher level of land use mixture and more bus routes and bus stops,
which are loaded into Factor 1, are associated with lower emission amounts. Simi-
lar effects are found from Factor 2: increased population and intersection densities
could lead to decreased travel emissions.
5 Discussions
The spatial autocorrelation analysis of household commuting carbon emissions
shows that the high values mainly correspond with areas in the urban suburbs, such as
198 J. Huang et al.
South Lake (Nanhu), Optics Valley-Guanshan (Guanggu), Jinyinhu, and Zhuankou.
These four areas share several common features. They are all newly developed urban
districts. The planned public service facilities, such as hospitals and schools, are not
fully operational. Other supporting facilities have not yet opened. Therefore, many
residents in these areas must travel to the central urban area for working, schooling,
shopping, and seeing doctor. Furthermore, the percentage of household travel by car
is high because of these areas’ underdeveloped public transportation system and their
long distance from services and shops.
The hot spots and cold spots appear in areas with contrasting spatial patterns.
Figure 8provides a closer look at cold spot Qingshan’s development patterns and
hot spot Zhuankou’s districts. Qingshan is an urban area founded in 1950s following
the construction of Wuhan Iron and Steel Corporation. Hence, it is characterized
by a relatively long developmental history and an excellent spatial balance of jobs,
housing, and services. Following many years of development, basic support and
public service facilities have been completed, making convenient for Qingshan resi-
dents in all aspects. However, as the economic and technological development zone,
Zhuankou presents a totally different picture. Founded in the early 1990s, Zhuankou,
which has shortage of residential land, currently functions as an industrial base in
Wuhan. Its traditional residential areas and commercial centers, such as Zhongjiacun
and Wangjiawan, are located in the region’s southwest portion. These factors have
led to a low degree of mixed land use and the relative separation of workplace and
residence. Furthermore, due to its short developmental history, public service facili-
ties in this region are not completed, so residents must leave the area to access daily
necessities such as banking and shopping.
Fig. 8 Land use pattern of cold (left: Qingshan) and hot (right: Zhuankou) spots of household
commuting carbon emission
11 Correlating Household Travel Carbon … 199
Fig. 9 Street pattern and public transit of cold (left: Qingshan) and hot (right: Zhuankou) spots of
household commuting carbon emission
Secondly, we also found that, in terms of household travel carbon emissions,
road network density and public transportation station density are relatively high
in cold spots while low in hot spots. In general, the denser road network, the more
public transport stations, and the shorter the distance between stations, the more
residents are inclined to choose public transit. This can be seen in the cases of cold
spot Qingshan and hot spot Zhuankou. Qingshan’s road network density is about
6.5 km/km2. Its corresponding distance between main roads is around 300 meters,
and there are approximately 70 public transport stations. While the road network
density in Zhuankou is about 4.1 km/km2, its corresponding distance between main
roads is about 500 m, and the area only has 5–6 public transit stations (Fig. 9).
Therefore, it is more convenient for Qingshan residents to take public transit. As a
result, the household travel carbon emissions drop, making the district a cold spot.
Zhuankou residents, on the other hand, rely more on private cars, which makes the
district into a hot spot.
Furthermore, residents’ socioeconomic status has a significant influence on travel
carbon emissions. The correlation analysis on land use mix and residents’ travel
carbon emissions indicates that they are closely related: the higher the degree of
mixed land use is, the lower the carbon emissions are. It is also noted that household
income bears a significant positive correlation with land use mix; they tend to locate
in rather centralized areas.
This spatial pattern of household location may be attributed to unique urban geo-
graphical features in Chinese cities. First, high-income families tend to locate in
central areas while low-income households tend to live in fringe areas. With several
hundred in-town lakes and two major rivers going through the city center, Wuhan has
a unique multi-center urban spatial structure. Such a naturally multi-centric structure
offers plenty of spatial opportunities for residents to live in proximity to services that
200 J. Huang et al.
satisfy daily needs. Second, the feature pertains to the unique institutional geography
typical in Chinese cities. The work unit presents a quality balance of jobs, housing,
and services with external travel minimized. Examples of work units include govern-
mental compounds, public agencies, factories, universities, and schools. They pro-
vide on-site housing and retail services. Large work units such as universities could
function self-sufficiently with nearly all urban services available on the campuses. In
Wuhan, there are more than 80 universities and research institutes. They house more
than one million students and several hundred thousand faculty and staff. The work
unit settings help lower travel and related emissions. By the regression analysis, the
evidence shows that public sector employees tend to emit less than others.
The variation among household members in travel and emission intensity reflected
the traditional structural characteristics of Chinese families. The householder (house-
hold member #1 in the survey) typically takes on major responsibilities in the house-
hold. They make trips for work and for household business, travelling the longest in
terms of time and distance, as we saw in the survey. Furthermore, the householder
usually has within-household privilege to access the household’s cars or motorcycle.
Therefore, they emitted the most among all household members. Household member
#2, who is the householder’s spouse, travelled and emitted the second, reflecting his
or her supporting role in the household.
Interestingly, household member #2’s travel carbon emissions are positively corre-
lated with the member’s educational level. Since there is no member-specific income
data available from the survey, one may speculate that household member #2’s higher
educational level indicates more opportunities to be employed or to have a better pay-
ing job. It may also mean more social activities. The varying role of the spouse in
the household leads to varying amounts of household travel and emissions.
Household member #3 is either the child in the household or the grandparent in
childless families. They are dependent and, therefore, travelled and emitted the least.
Most of this household member’s travel related to either schooling or household
errands. In Wuhan and other Chinese cities, school-age children are required to go
to school within their residential district. Many go to the schools affiliated with
the parents’ work unit. The ‘going to the nearest school’ policy keeps the school
trips relatively short. However, the changing trend demands attention as increasingly
families live off the job sites (i.e., work units); children end up commuting with their
parents to a work-unit-affiliated school. Some affluent families send their children
to their ‘desired schools’ even though they have to pay extra fees for cross-district
schooling.
6 Conclusion
This paper presents a disaggregate study of travel carbon emissions in Wuhan, China.
The study extends the analysis of the correlation between socioeconomic factors and
the spatial characteristics of household travel emissions. The main conclusion is that
urban spatial structure and land use context matter. Both offer additional explanatory
11 Correlating Household Travel Carbon … 201
power to carbon emissions after the effects of income and other sociodemographic
factors are controlled. Emission hot spots and high-emission families mostly appear
in newly developed suburban areas where jobs and housing provisions appear unbal-
anced, and service establishments are not yet in place. Land use planning can play
an active role in cutting back household travel carbon emissions and reducing the
effects of hot spots. Practices such as mixed land use development, job-housing-
service balancing, and transit provisions can work jointly to lower carbon emissions.
The traditional work unit in Chinese cities offers such a desired attribute of urban
spatial structure. However, these attributes have been disappearing over the past three
decades of urbanization. How the spatial features can be retained in the presence of
market provisioning of housing, jobs, and services is a major challenge to planning
and to public policy making as well.
Household members’ travel emission patterns match their status in the household.
Householder is most mobilized and, thus, the highest emitter. In the forthcoming
two to three decades, China is expected to continue the rapid urbanization process.
Household income will grow and car ownership is likely to increase as well. Unless
there are effective policy interventions, additional household members aside from the
householder will likely join the driving force, which will lead to increase in household
travel carbon emissions. Now is a critical moment when Chinese household’s private
motorization is about to take off. Strategies should be developed to restrain the
growth of private vehicle ownership and to improve public transportation services
for likely growth of household travel carbon emissions in the foreseeable future.
The study reinforces that both place-based spatial policy and people-based policy
are important and urgently needed for reducing carbon emissions for Chinese cities
under the urbanization and economic development in a critical transition point.
Acknowledgements The study was supported by Ministry of Science and Technology, China
(2015BAJ05B00), Australian Research Council (DP1094801), China National Science Foundation
under Grant (No. 51278385), and the Snell Grant of Center for Sustainable Development at the
University of Texas at Austin. USDOT University Transportation Center Cooperative Mobility for
Competitive Megaregions (CM2) provided in-kind support.
References
Banister D (2011) Cities, mobility and climate change. J Transp Geogr 19(6):1538–1546
Boarnet MG, Crane R (2001) Travel by design: the influence of urban form on travel. Oxford
University Press on Demand
Cervero R (2002) Built environments and mode choice: toward a normative framework. Transp Res
Part D: Transp Environ 7(4):265–284
Chai YW, Xiao ZP, Liu ZL (2012) Low-carbon optimization strategies based on CO2emission
mechanism of household daily travels: a case study of Beijing. Geogr Res 2
Chen F (2010) Research on strategy of low carbon city: Shanghai empirical analysis. China Archi-
tecture & Building Press, Beijing, China
Defra (2008) Carbon trust & energy saving trust. Low carbon cities. Carbon Trust, London
202 J. Huang et al.
Eggelston HS, Buendia L, Miwa K, Ngara T, Tanabe K (2006) Guidelines for national greenhouse
gas inventories. IPCC National Greenhouse Gas Inventories Programme, Hayama, Japan
Ewing R, Clemente O (2013) Measuring urban design: metrics for livable places. Island Press
Frank LD, Pivo G (1994) Impacts of mixed use and density on utilization of three modes of travel:
single-occupant vehicle, transit, and walking. Transp Res Rec 1466:44–52
Glaeser EL, Kahn ME (2010) The greenness of cities: carbon dioxide emissions and urban devel-
opment. J Urban Econ 67(3):404–418
Goovaerts P, Jacquez GM, Marcus A (2005) Geostatistical and local cluster analysis of
high-resolution hyperspectral imagery for detection of anomal ies. Remote Sens Environ
95(3):351–367
GHG Protocol (2005) Calculating CO2emissions from mobile sources. Greenhouse Gas Protocol,
Publishing city (GHG Protocol)
Huang Y, Wu X (2012) CO2emission from road transport and the emission reduction potential in
Wuhan City. Environ Prot Sci 38:9–21
IEA (2011) CO2emissions from fuel combustion (2010)–highlights. International Energy Agency,
Publisher city
Jiang H, Xiao R, Wu J (2013) The impact pattern of urban family carbon emission and implications
for the low-carbon resident community planning: a case study of Guangzhou. Mod Urban Res
pp 100–106
Kerkhof AC, Benders RM, Moll HC (2009) Determinants of variation in household CO2emissions
between and within countries. Energy Policy 37(4):1509–1517
Khattak AJ, Rodriguez D (2005) Travel behavior in neo-traditional neighborhood developments: a
case study in USA. Transp Res Part A: Policy Pract 39(6):481–500
Ko J, Park D, Lim H, Hwang IC (2011) Who produces the most CO2emissions for trips in the
Seoul metropolis area? Transp Res Part D: Transp Environ 16(5):358–364
Litman T,Steele R (2017) Land use impacts on transport. Victoria Transport PolicyInstitute, Canada
Liu Z, Ma J, Chai Y (2017) Neighborhood-scale urban form, travel behavior, and CO2emissions
in Beijing: implications for low-carbon urban planning. Urban Geogr 38(3):381–400
Miehe R, Scheumann R, Jones CM, Kammen DM, Finkbeiner M (2016) Regional carbon footprints
of households: a German case study. Environ Dev Sustain 18(2):577–591
Mindali O, Raveh A, Salomon I (2004) Urban density and energy consumption: a new look at old
statistics. Transp Res Part A: Policy Pract 38(2):143–162
Ord JK, Getis A (1995) Local spatial autocorrelation statistics: distributional issues and an appli-
cation. Geogr Anal 27(4):286–306
Qin BS, Shao R (2011) Low-carbon city and spatial structure optimization: connotation, evidences
and practices. Urban Plan Int 2011:72–77
Qiong B (2009) Transforming economic development mode, constructing low carbon cities. The
Central People’s Government of the People’s Republic of China, Beijing
Un-Habitat (2012) State of the world’s cities 2008/9: Harmonious Cities. Routledge
United Nations Development Programme (UNDP) (2013) China human development report (2013).
UNDP, Beijing
Wang Z, Liu W (2015) Determinants of CO2emissions from household daily travel in Beijing,
China: individual travel characteristic perspectives. Appl Energy 158:292–299
Wang Y, Yang L, Han S, Li C, Ramachandra T (2017) Urban CO2emissions in Xi’an and Bangalore
by commuters: implications for controlling urban transportation carbon dioxide emissions in
developing countries. Mitig Adapt Strat Glob Change 22(7):993–1019
Wu J (2010) The research on block characteristics impact on residents travel. Master, Tongji
Wuhan Municipal Bureau of Statistics (2018) Wuhan statistical yearbook (2017). Wuhan Municipal
Bureau of Statistics, Wuhan
Wuhan Statistics Bureau (1978–2012) Wuhan Statistical Yearbook (1978–2012). Statistical Pub-
lishing House, Beijing, China
Xiao Z, Lenzer JH Jr, Chai Y (2017) Examining the uneven distribution of household travel carbon
emissions within and across neighborhoods: the case of Beijing. J Reg Sci 57(3):487–506
11 Correlating Household Travel Carbon … 203
Yan W, Minjun S (2009) CO2emission induced by urban household consumption in China. Chin J
Popul Resour Environ 7(3):11–19
Yang Y, Wang C, Liu W, Zhou P (2018) Understanding the determinants of travel mode choice of
residents and its carbon mitigation potential. Energy Policy 115:486–493
Zahabi SAH, Miranda-Moreno L, Patterson Z, Barla P, Harding C (2012) Transportation greenhouse
gas emissions and its relationship with urban form, transit accessibility and emerging green
technologies: a Montreal case study. Procedia -Soc Behav Sci 54:966–978
Zhang S (2010) Study on urban-life energy use and impact of human behaviors through investiga-
tions in seven Chinese cities. M.A. Thesis. Tsinghua University, Beijing
Zhang SL, Zhang K (2007) Comparison between general Moran’s Index and Getis-Ord general G
of spatial autocorrelation. Acta Sci Nat Univ Sunyatseni 4:022
Zhang M, Zhao P (2017) The impact of land-use mix on residents’ travel energy consumption: new
evidence from Beijing. Transp Res Part D: Transp Environ 57:224–236
Chapter 12
A Simulation Platform
for Transportation, Land Use and Mobile
Source Emissions
Liyuan Zhao and Zhong-Ren Peng
Abstract This study describes an integrated platform model for assessing the
interaction among land use, transportation, and mobile source emissions. In the pro-
posed integrated framework, LandSys, a home grown land use model, produces land
use change over the dimensions of space and time, allocates land use forecast results
in terms of household and employment at the traffic analysis zone (TAZ) level, and
feeds these socioeconomic data into a travel demand model, the Florida Standard
Urban Transportation Model Structure (FSUTMS). Then, the produced travel time
and accessibility index by FSUTMS are fed back into LandSys to quantify the emis-
sions. Finally, the emissions from standalone FSUTMS and integrated framework
are compared to quantify the air quality benefits of the land use development from
the integrated land use and transportation model. In the case application of Orange
County, Florida in the United States in 2000, 2012 and 2025, five major indicators
of transportation networks were used: link saturation in the transportation network,
overall vehicle miles traveled (VMT), vehicle hours traveled (VHT), mobile source
greenhouse gas emissions and fuel consumption. The results show that the values
of the five indicators were lower when utilizing the integrated platform than was
predicted by the standalone FSUTMS models, which demonstrates that the inte-
grated platform achieved greater effectiveness in environmental improvements by
considering the interactions between land use and transportation.
Keywords LandSys ·Mobile sourced emissions ·Cellular automata ·
Multi-Agent
L. Zhao (B)
School of Architecture and Urban Planning, Huazhong University of Science and Technology,
Wuhan, China
e-mail: liyuanzhao@hust.edu.cn
Z.-R. Peng
Department of Urban and Regional Planning College of Design, Construction and Planning,
University of Florida, Gainesville, USA
State Key Laboratory of Ocean Engineering, School of Naval Architecture, Ocean and Civil
Engineering, Center for ITS and UAV Applications Research, Shanghai Jiao Tong University,
Shanghai, China
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_12
205
206 L. Zhao and Z.-R. Peng
1 Introduction
Globally, the challenge of reducing greenhouse gas emissions, improving air quality,
and achieving more sustainable development has motivated renewed development
of integrated land use and transportation planning. The relationship among land use,
transportation and air quality are interrelated (Ang-Olson et al. 2000; Zahabi et al.
2012). First, the transportation system is an important factor of land location choice
through accessibility and travel cost. While conversely, changes in land-use directly
affect transportation demand patterns. The spatial pattern of activities resulting from
various land uses are essential inputs to predicting transportation system perfor-
mance. Furthermore, the emissions from transportation have been proved to be a
major source of air quality degradation (Ang-Olson et al. 2000). Improved and better
integrated land use and transportation models have been strongly recommended by
states and MPOs (Metropolitan Planning Organizations) for state growth manage-
ment programs, air quality control and sustainable development that promotes closer
and more numerous linkages between land use and transportation planning. Based
on forecasts of future land use and transportation situations, planners can simulate
what measures can be taken to enhance positive impacts and avoid negatives effects
based on forecasts being realized.
Many integrated land use and transportation models have been developed and
reported in the literature with various approaches and structures. The related the-
ories for modeling land use change can be generalized them into eight principles,
including economic driven, spatial interaction, cellular automata, statistical analy-
sis, optimization techniques, rule-based simulation, multi-agent models, and micro
simulation (Koomen and Stillwell 2007;Suetal.2014). Different models special-
ize in different aspects of interaction between land use and transportation by taking
into account different driving forces at specified land sizes (Newman and Kenwor-
thy 1996). DRAM/EMPAL, MEPLAN, PECAS and MUSSA models are based on
top-down allocation of households and employment to each of the Forecast Analysis
Zones (FAZs) or Traffic Analysis Zones (TAZs) (Abraham and Hunt 2003; Con-
cha 2018; Kockelman et al. 2005; Martinez 1996,2007; Putman 1995,1996). The
analysis zones usually cover a large area and are hard to capture the land use/cover
changes of small areas. UrbanSim fills these gaps by simulating land use changes of
grid cells and parcels. It takes various driving forces into consideration by weaving
these forces into the transition rules of each cell or parcels (Waddell et al. 2003,
2010; Waddell and Nourzad 2002).
Although the model features enable some existing models to simulate small areas
and be sensitive to various land use and transportation policies, it has difficulties in
providing all land use and socioeconomic inputs for the specific FSUTMS (Florida
Standard Urban Transportation Model Structure) (FSUTMS 2016). FSUTMS, the
transportation model developed in Florida, in the United States (U.S.), is required
by the Florida State Government to be integrated with land use planning to facilitate
assessment of the interaction between land use and transportation. To bridge this
gap, Zhao and Peng proposed a CA-Agent land use model, LandSys, to become
12 A Simulation Platform for Transportation … 207
integrated with FSUTMS for the purposes of encapsulating the interaction between
land use change and transportation (Zhao and Peng 2010,2012). Meanwhile, the
FDOT (Florida Department of Transportation) with respect to recent developments
in particular, have highlighted the need for a better understanding of the manner that
air quality is attributed to the impacts of land use and transportation. Based on this
background, this study describes the development of an integrated platform model for
assessing the interactions of land use, transportation, and mobile sourced emissions.
With the improvements to the study of land use and transportation, there is a rich
and rapidly growing body of literature on simulating travel and emissions’ impacts
of different land use scenarios. Qualification of the emissions’ reduction are usually
taken as being one of the important measures for better accounting of the air qual-
ity benefits of sustainable land use (Ang-Olson et al. 2000; Tayarani et al. 2018).
Hatzopoulou and Miller (2010) extends an activity-based transportation model with
capabilities of quantifying vehicle emissions, meteorology and air dispersion, for
the purpose of refining the modeling of transport-induced air pollution. Some intel-
ligent methods have been used to look at the relationship between air pollution and
transportation. Pai et al. (2007) conducted a study of the grey relational grade of air
pollutants from ambient and roadside air pollution monitoring stations using the inte-
grated travel demand and emissions model. Rodier and Johnston (2002) conducted
a sensitivity analysis of plausible errors in population, employment, fuel price, and
income projections. Significant efforts have been invested into developing a trans-
portation and emissions model. However, indeed there are few examples where these
models have been studied, in terms of how and to what extent the planning of inte-
grated land use and transportation can decrease emissions and improve air quality.
In this study, the emissions from a standalone and integrated transportation model
are compared to quantify the air quality benefits of land use development utilizing
an integrated land use and transportation model.
This study begins with a description of the Landsys model, followed by detail of
the module structure and its construction. Then the development of an integrated plat-
form model of land use, transportation and mobile source emissions is described,
including model components, integration, emissions quantification, as well as the
integrated platform analysis. In the model application and results discussion, we first
show the validation and land use forecasting performance of households and employ-
ment in LandSys. Secondly, we predict the transportation system changes in response
to different land use scenarios by comparing the saturation of the transportation net-
work in Orange County utilizing a FSUTMS model, with and without integrating
this with LandSys. Moreover, the effects of transportation on land use development
have been analyzed by investigating the differences between the spatial distributions
of land use simulated with/without FSUTMS integrated. Finally, the greenhouse gas
emissions and fuel consumption are analyzed to quantify the air quality benefits of
land use development derived from an integrated land use and transportation model.
208 L. Zhao and Z.-R. Peng
2 Integration of LandSys and Transportation Model
2.1 LandSys Model
The LandSys integrates Cellular Automata (CA), Agents, and bid-rent market
mechanisms to model the dynamic land use change at a manageable cell level (50 m
×50 m) (Zhao and Peng 2012). LandSys is firstly used in Orange County, FL.
In this land use model, CA sub-model aims to capture the spatial relationships of
land development and agglomeration factors, while multi-agent models represent the
behavior of individual agents, policy factors, and economic factors. A bid-rent based
land market equilibrium sub-model is deployed to represent the interaction among
agents with endogenous land price under equilibrium conditions.
Figure 1shows the detail model framework, including model components and
function (Zhao and Peng 2012). CA captures the spatial attributions of land use
dynamics. A Multinomial Logit (MNL) model is integrated into the CA model to
Fig. 1 Model framework of cellular automata-agents land use model
12 A Simulation Platform for Transportation … 209
process multiple land use change. MNL-based CA model determines the spatial
suitability of each vacant land cell to convert to the land use type related with travel
demand, including residential, industry, commercial and services, and institutions.
Spatial properties of each land cell include physical attributes (e.g., soil quality,
slope, and topographic), neighborhoods characteristics, accessibility, and distance
to other locations (e.g., the CBD, shopping centers, education institutes, and other
main public facilities).
Agent models represent the individual’s decision-making behavior, including
demand side and supply side choices. This study investigated the following agents:
household, employment, developer and government agents. The government agent is
an external agent to account for land use policies. The government agent includes the
policy quantification model, which translate land polices (e.g., zoning, urban growth
boundary, conservation area) into constraints or preference and integrates them into
the predicted land use change for next time period.
Household agents capture the residential mobility, location choice, and residen-
tial type choices for the households. To generate uniform input data for FSUTMS,
employment agents consider three sectors: industrial, commercial, and retail ser-
vice employment. Each employment sector then uses two sub-models: employment
mobility and location choice. Developers and land owners decide whether to demol-
ish or construct houses, apartments, or buildings of each employment sector on a
specific land cell. Table 1shows the detail model input/output for the household,
employment, and developer agents. In the land market, the behaviors of developers
constitute the market supply, whereas the household and employment agents form
the market demand. Bid-rent theory can represent the interaction between demand
and supply sides. The land price is generated under equilibrium conditions. From
the interaction between demand and supply, land use prices are updated and provide
feedback to the land use forecasting model for next period.
To generate required inputs for FSUTMS, the LandSys considers the following
factors: (1) population changes; (2) the spatial suitability for land use change (e.g.,
topographic, slope, neighborhoods attribution); and (3) the decision-making behavior
of stakeholders (e.g., households, employers, developers). The forecasting results
of land use development are produced in the results of combined CA and agent
models. From the bid-rent based interaction among agents, the specific household
characteristics (e.g., population, residential type, car ownership) in the residential
cells and the number of firms by sector in the non-residential cells are produced,
aggregated to TAZ level thereafter to update those inputs of FSUTMS. The new
travel cost and accessibility taken from FSUTMS are fed back into the data store,
updating the inputs for the land use model in the next time period.
2.2 Land Use Development Transition Rules
The transition rules for land use (cover) development consist of two parts, one for
developed lands and the other for vacant lands. For the developed lands, the transition
210 L. Zhao and Z.-R. Peng
Tabl e 1 Detail model input/output for household, employment and developer agent
Model Agent
Household/employment agent
Mobility
model
Input The utility of the mobility:
umov
ik =w1,mov Imov +w2,movOi+w3,mov Ai
• the household/employment’s characteristics (for household
agent: including household size, workers, children between 5
and 17 years of age, persons above 50 years of age, and average
income; for employment agent: firm size);
• the attributes of current cell i(including vacant dwelling units
and owner occupancy) and accessibility;
• the household/employment’s accessibility;
k: land type, k=1 denotes residential land and k=2, 3, 4 denotes
industrial, commercial services, and institutional employment
respectively
Output The mobility probability:
Prmov
ik =exp(ξ uh,mov
ik )/[1+exp(ξ uh,mov
ik )]
Location
choice
Input Utility of choosing location i:ucho
ik =Bik −rik
• the willingness-to-pay function of household/employment agent
for location iwith type kland;
• rent for location iwith type kland
Output The choice probability, that an alternative location yields the
highest utility can be given by the Logit model:
Prcho
ik =exp(θ (Bik −rik))/ ixi1exp(θ ( Bh
i1−rik))
Model Agent
Developer agent
Demolition
model
Input The profit of demolition:
• demolished costs;
• maintenance costs;
• the expected utility of existing locator’s mobility
Output The probability of demolition for existing developed cell:
Prdem
i=exp(σdemudem
i)/[1+exp(σdemudem
i)]
Construction
model
Input ucon
ik =w1,conrik −w2,conlik −w3,con ccon
ik −w4,convik
• land price extracted from census data
• construction costs
• maintenance costs
Output The probability of the developer developing the land into type
kbuilding. Prcon
ik =exp(σconucon
ik )/ kexp(σconucon
ik )
12 A Simulation Platform for Transportation … 211
rules imply whether it is to be demolished and transitioned into vacant lands, which
are produced by an optimization problem that maximizes the household/employment
agents’ mobility preference and demolition behaviors of the developer agent. For the
vacant lands, the transition rules determine whether it is to be developed and what
type it will be if developed into, including residential, industrial, commercial and
services, or institutional land. They are derived from a combinatorial optimization
problem, including the probability of spatial suitability from a CA model, construc-
tion probability from the developer agent construction model, the location choice
probability from a household/employment agents model, and compliance with the
government policies.
2.3 Households/Employments Allocation Equilibrium
The land use supply-demand market achieves equilibrium when all agents are located
in the developed land cells. For each new developed land cell location in land type k,
the developer chooses the highest bidder among all consumers that will be allocated in
this cell. According to the auction-type process with stochastic bids at each location,
the number of household/employment agents allocated in each developed cell i for
land use type k is created as follows:
Nik =Sik
xik exp(θ Bik)
ixik exp(θ Bik)(1)
where xik denotes the characteristic variable representing if the cell iis developed into
type kland which can be generated from the results of land use cover development.
Sik denotes the total household/employment accommodated by building supply of
type kin the cell i. The optimal number of building supply of land type kin cell i
can be evaluated according to maximizing developer’s profit.
2.4 LandSys Simulation Module Structure
As shown in Fig. 2, LandSys simulation platform includes three modules: the basic
module, calibration module, and bounded validation and application module. Within
the three modules, the basic module is for early-stage preparation; the calibration
module is used for system learning and determination of model parameters; after
plugging in the parameters, LandSys can simulate land use changes and generate
input data for FSUTMS transportation models.
The basic module contains three sub modules: land use categorization; data pro-
cessing; and data storage. Land use categorization classifies initial land use types
via a proposed “quantitative change in-out” method. The input data are the land
use/cover data of case study areas in the past two periods (to avoid misclassifica-
212 L. Zhao and Z.-R. Peng
Fig. 2 LandSys module structure
tion, the selected two periods are always 10 years apart). The output data include
reclassification of land use types that take into account land use and transportation
integration of the case study area.
The input data of data processing and the storage module consist of three groups:
spatial related data (named by data i1), including digital elevation model (DEM)
and land use data, transportation network data (named by data i2) from FSUTMS
concerning travel time and accessibility, and households/employment data (named
by data i3). The data processing and storage can be divided into two steps. Firstly,
converting of data i1 and data i2 to 50 m ×50 m raster data by ArcGIS Spatial Anal-
ysis tools, saving the data as an ACSII matrix and then reading the data in Matlab
software. Secondly, based on maps of household/employment and land use classifi-
cation, assigning household data to the spatially corresponding residential cells and
12 A Simulation Platform for Transportation … 213
employment data for three sectors to industrial/commercial and services/institutional
land cells respectively. The output data of data processing and storage module include
cell based physical and agglomerations attributes (data o1), household attributes (data
o2) and employment attributes (data o3).
The calibration module includes two sub-modules: CA Calibration Module and
the Agent Calibration Module. Then data o1 of year Tand T+1 are plugged into
the CA model calibration to calculate the parameters of the MNL model through to
the least-squares method based on Monte Carlo random sampling. In a similar way,
data o2 and data o3 are plugged into the Agent Calibration Module to calibrate the
parameters for household/employment and developer agent model. All the calibrated
parameters are important inputs of model validation and application.
The Validation and Application Module includes a land use change sub-module
and household/employment allocation sub-module. The land use change sub-module
includes two parts: conversion rules based on external factors, such as land use and
transportation policies; the estimated land use of the next year based on maximizing
the development suitability according to land development balance. The allocation
module generates a TAZ-based household and employment data of the next time
period based on the solutions of bid-rent theory and the balance model of the supply
and demand of the land market.
3 Integrated Transportation, Land Use and Mobile Source
Emissions
3.1 Mobile Source Emissions Calculation
Along with the rapid process of urban development, deteriorating air quality has
become one of the most serious city problems that concern public health. Traffic
emissions have become one of the major sources of air pollutants and greenhouse
gas. The U.S. Environmental Protection Agency (EPA) estimates that mobile (car,
truck, and bus) emissions account for as much as half of outdoor toxic air chapter
proposes an integrated model platform for assessing the interaction among land
use, transportation, and mobile source emissions. The mobile source emissions are
calculated by the following function:
E=
l,mv
Ql,mvKl·rmv(2)
where ldenotes the road link; mand vare road hierarchy and the related traffic speed.
Ql,mvmeans the traffic volume of link l, which is produced by traffic assignment
module of FSUTMS. Klrepresents the length of link land rmvdenotes the emission
factor of specific mobile source (CO, HC or NO) with road level mand travel speed
v. The emission factors of CO, HC and NO for road with different hierarchy are
214 L. Zhao and Z.-R. Peng
extracted from FSUTMS. The emission simulator in the proposed integrated platform
can be developed according to Eq. (2),
3.2 Model Components Integration
FSUTMS, powered by Cube Voyager, is currently adopted and used by 26 MPOs,
FDOT (Florida Department of Transportation) districts, and other planning agencies
in Florida. Integrated FSUTMS/Land Use has been identified by Florida Model Task
Force members and leaders as the top priority for future development. The FDOT
Central Office conducted a survey of FSUTMS model users to identify the most
pressing needs for further development. The need to interface an existing transporta-
tion model FSUTMS with a new land use forecasting model was identified as having
the highest priority to improve evaluation of interrelated impacts and the effect of
various polices on land use and transportation system.
FSUTMS utilizes socio-economic data (e.g., spatial allocation of households and
employments) as the inputs of transportation systems and then generates outputs (e.g.,
network travel time, accessibility) for transportation policy making and planning.
Unfortunately, the spatial allocation data in the current FSUTMS is static and never
changes with new transportation conditions in long-range planning. In the proposed
integrated platform, the spatial allocation data could in turn, respond to changes in
travel time and accessibility. The integration aims to explore this complex interactive
relationship between transportation and land use by combining LandSys with the
FSUTMS.
The interactive procedure of LandSys and FSUTMS are implemented in the inte-
gration framework (see Fig. 3). The integration of LandSys and transportation is
a bi-level modeling structure. The household/employment allocation results from
LandSys are plugged into the lower-level transportation models to update new travel
demand. And then the travel cost and accessibility from the transport model are
generated through trip generation, distribution, mode choice and traffic assignment,
and fed back to LandSys database that reflects the results of CA and agents model.
Moreover, the traffic emissions can be calculated by traffic flow distribution on the
road network to evaluate the environmental impact of transportation.
An effective integrated land use and transportation model can well represent the
interaction among land use, transportation and environment. One important charac-
teristic of the proposed integrated framework is that it is freely transferable to other
cases studies since LandSys can calibrate its parameters by self-learning. Hence, the
integrated framework is intended to be capable of simulating and analyzing govern-
mental policies and planning decision-making from various dimensions. Firstly, the
demographics and regional economic data, and specific land policies such as zoning
preference can be analyzed through LandSys. Secondly, transportation system infor-
mation, such as network expansion and the impact of new roads can simultaneously
be added as an exogenous input into FSUTMS. Moreover, various land development
policies, transportation policies, transportation planning projects and other related
12 A Simulation Platform for Transportation … 215
Fig. 3 The model integration structure
issues can be tested by scenario planning using the integrated framework to check
future effects and performances of the two systems. Finally, statewide future envi-
ronment evaluation, benefit-cost analysis, as well as air quality can also be generated
as outputs from the integrated model.
4 Model Application and Results Discussion
4.1 Forecasting the Allocation of Household
and Employment
In LandSys, households and employment (firms) are allocated at the cell level
under bid-rent equilibrium. To generate inputs for transportation models (such as
FSUTMS), the allocation results of households and employment at the cell level,
are further aggregated into TAZs. In the FSUTMS model, Orange County, Florida is
divided into 662 TAZs. The LandSys is validated with GIS data of Orange County
using a three-step process: (1) calibrate data from 1990, (2) predict land use devel-
opment (household/employment) in 2000, and (3) compare the predictions with real
data from 2000. The prediction results of the households and employment (firms) at
the TAZ level are shown in Fig. 4.
216 L. Zhao and Z.-R. Peng
Fig. 4 Forecasting the allocation of household/employment: aHousehold; bEmployment
Fig. 5 Network link saturation: awithout integration; bwith integration
4.2 Impacts of Land Use on Transportation
To evaluate how land use change affects the transportation network, the link satu-
ration, calculated by the division of simulated traffic flow by the link capacity, is
quantified and visualized. Figure 5compares the saturation of the transportation
network in Orange County from the FSUTMS model with and without integrating
with LandSys. The saturation of links is categorized into four groups: 0–0.5 (blue),
0.5–0.8 (yellow), 0.8–1 (purple), and higher than 1.2 (red). According to Fig. 5,the
road paths with high saturation (shown in purple and red) concentrate in the center of
study area. This may because the urban center tends to have a high intensity of land
development. On the edge of the study area, the saturation level is lower, as these
places generate less travel demand.
12 A Simulation Platform for Transportation … 217
Tabl e 2 The distribution of saturation with and without the integrated LandSys model
Total links Year 2000 Year 2012 Year 2025
13,862 14,816 15,372
Saturation Without With Without With Without Wit h
[0, 0.5) 6797 6910 6909 7015 5847 6029
[0.5, 0.8) 2306 2394 2134 2514 1909 2101
[0.8, 1.2) 3075 2943 3466 3392 3655 3815
>=1.2 1684 1615 2310 1895 3961 3427
Table 2lists the saturation distribution across different intervals from FSUTMS
without and with integration with LandSys in 2000, 2012 and 2025. The total number
of links increases from the year 2000 to the year 2025 because the transportation
network expands in unison with the increase in population. Among the simulation
results of the years 2000, 2012 and 2025, the FSUTMS standalone model produces
more links with higher saturation levels (more than 0.8), and less links with lower
saturation levels (less than 0.8) than those in the “after” integration scenario. It shows
that the land use development arising from the integrated model which responds to
expected demands for transportation accessibility can reduce network saturation and
relieve traffic congestion.
4.3 Land Spatial Distribution Results
To test the effects of transportation on land use development, differences between
the spatial distributions of land use simulated with/without FSUTMS integrated are
analyzed. Figure 6illustrates the spatial distribution of households and employment
generated by LandSys without and with integration with FSUTMS in 2000. The
numbers shown in each TAZ (Fig. 6) denote the difference that the allocation of
household/employment generated by integrated LandSys and FSUTMS minus the
allocation results generated by standalone LandSys.
The household number of some TAZs in the center generated by the integration
of LandSys and FSUTMS models is less than the one without FSUTMS integrated
(the original land use spatial distribution). This occurs because the integrated models
tend to avoid high travel costs (because of a high congestion condition). Basically,
the number of households simulated by the integrated models in the outer TAZs
is more than that without the transportation model. In the integrated land use and
transportation running results, more households are allocated to the outer TAZs, to
avoid the high congestion levels in the inner region.
The allocation of employment shows similar characteristics to the allocation of
households. Generally, LandSys-FSUTMS models allocate greater employment on
the edge of the study area than FSUTMS models, and less employment in the center.
218 L. Zhao and Z.-R. Peng
Fig. 6 The differences in household/employment allocations using LandSys without/with integra-
tion with FSUTMS: aHousehold; and bEmployment
It implies that the integrated models assign less employment to the already congested
areas. This is so because travel costs and accessibility are updated and plugged back
in land use models. Therefore, LandSys-FSUTMS models can adjust the spatial
distribution of households and employments to mitigate traffic congestion.
4.4 Transportation Environment Results
Comparing the estimated emissions of transportation without and with integration
can shed light on the environmental impact of the predicted transportation system. It
must be noted that the current FSUTMS takes only road hierarchy and travel speed
into consideration and ignores types of vehicles. The estimations of emissions can
be greatly improved if the types of vehicles can be taken into account, such as cars,
buses, trucks, motorcycles, etc.
Table 3suggests that there is an increase of emissions, fuel consumption, vehi-
cle miles traveled (VMT) and vehicle hours travelled (VHT) from the year 2000 to
2025. This is caused by the increase of population, car ownership, travel demand
and the sprawl of Orange County. The integrated LandSys-FSUT MS model esti-
mates that the emissions of CO would decrease by 1.53% in 2000 and by 5.62%
in 2025. The land use development from integrated models can help change the
spatial pattern of traffic demand and finally reduce the traffic emissions in the long
term. Also, for LandSys-FSUTMS models, fuel consumption decreases by 0.33%
and VHT decreases by 5.72% from the year 2000 to 2012, while VMT increases by
0.69%. All these values become larger in the next time period (from the year 2012
to year 2025): fuel consumption decreases by 3.32% and VHT decreases by 6.45%,
while VMT increases by 3.52%. As time increases, the percentages of each indicator
12 A Simulation Platform for Transportation … 219
Tabl e 3 Mobile source emissions from LandSys-FSUTMS models (with) and FSUTMS (without) models of Orange County, FL
2000 2012 2025
Without With Decrease Without With Decrease Without With Decrease
CO (kg) 334,823 329,703 1.53 434,007 416,242 4.09 645,049 608,824 5.62
HC (kg) 40,748 40,180 1.39 52,471 50,753 3.27 76,124 72,426 4.86
NO (kg) 37,263 37,179 0.23 47,561 46,578 2.07 64,174 62,572 2.50
Fuel (L) 8,975,106 8,945,106 0.33 11,440,219 11,171,170 2.35 15,626,855 15,108,465 3.32
VMT 27,595 27,404 0.69 35,335 34,547 2.23 48,811 47,091 3.52
VHT 1,044,983 978,818 6.33 1,334,495 1,258,126 5.72 2,169,366 2,029,379 6.45
220 L. Zhao and Z.-R. Peng
increase which suggests the importance of LandSys-FSUTMS models in improving
the transportation environment.
Compared to FSUTMS models, the integrated models estimate less pollutant
emissions, fuel usage, VMT and VHT. The percentage of this decrease is not obvious
in the year 2000, but is apparent in 2025. One possible reason is that LandSys-
FSUTMS models focus on Orange County, while FSUTMS models cover all of
central Florida. In other words, the integrated models are not concerned about land
use changes in places other than Orange County. This can impact the results of
models. In conclusion, comparisons show that the land use strategy produced from
the integrated models can optimize the spatial pattern of travel demand and reduce
pollutant emissions and overall VMT and VHT of transportation system. Therefore,
the provided results of the integrated land use and transportation model can help
improving air quality and the transportation environment.
5 Conclusion
By integrating LandSys, with a transportation model and a mobile source emission
model, a new platform has been developed to provide a comprehensive approach for
simulating the dynamic process of land use and transportation changes, that includes
accounting for mobile source emissions. Data from the Orange County, Florida, was
used as a case study. The LandSys simulates land use change at multiple spatial and
temporal dimensions, as well as representing decision making behaviors of house-
holds, employment, and developers. Future land use patterns and socioeconomic
data (e.g., household, firms, and population) can be produced to update those inputs
of a transportation model. The traffic emissions are evaluated by the road network
outputs from a transportation model.
The simulation results show that in the integrated model, the values of five indi-
cators are lower than those predicted by standalone FSUTMS models, indicating the
importance and effectiveness of integrated planning of land use and transportation.
In addition, the standalone LandSys model produces fewer households and employ-
ments in the center of the study area, and more at the edge of the city. The land
use allocation results of the integrated model can also mitigate traffic congestion
by adjusting the spatial distribution of households and employments. Moreover, the
land use strategy produced from the integrated platform can reduce mobile source
emissions, VMT and VHT over a longer period of time.
Acknowledgements This research was supported by the National Natural Social Science Founda-
tion of China (Grant Number: 18BGL270).
12 A Simulation Platform for Transportation … 221
References
Abraham JE, Hunt JD (2003) Design and application of the PECAS land use modeling system.
University of California, Davis
Ang-Olson J, Fischer M, Dulla R (2000) Land use as an air quality control measure: review of
current practice and examination of policy options. Transp Res Rec 1738:33–38
Concha FJM (2018) The land-use and transportation system. In: Concha FJM (ed) Microeconomic
modeling in urban science. Academic, London
FSUTMS (2016) Florida standard Urban transportation model structure 2016, http://www.
fsutmsonline.net/index.php. Accessed 22 July 2016
Hatzopoulou M, Miller EJ (2010) Linking an activity-based travel demand model with traffic emis-
sion and dispersion models: transport’s contribution to air pollution in Toronto. Transp Res Part
D: Transp Environ 15(6):315–325
Kockelman KM et al (2005) Tracking land use, transport, and industrial production using random-
utility-based multiregional input-output models: applications for texas trade. J Transp Geogr
13(3):275–286
Koomen E, Stillwell J, (2007) Modeling land use change: theories and methods. In: Stillwell J (ed)
Modeling land use change: progress and application. Springer, Netherlands
Martinez F (1996) MUSSA: land use model for Santiago city. Transp Res Rec 1552:126–134
Martinez F (2007) MUSSA II: a land use equilibrium model based on constrained idiosyncratic
behavior of all agents in an auction market, Transportation Research Board 86th Annual Meeting,
Washington, DC
Newman PWG, Kenworthy JR (1996) The land use-transport connection: an overview. Land use
policy 13(1):1–22
Pai TY et al (2007) Using grey system theory to evaluate transportation effects on air quality trends
in Japan. Transp Res Part D: Transp Environ 12(3):158–166
Putman S (1995) EMPAL and DRAM location and Land Use models: a technical overview. Pro-
ceedings of the land use modeling conference proceedings, Dallas, TX
Putman S (1996) Extending DRAM Model: theory-practice nexus. Transp Res Rec 1552:112–119
Rodier CJ, Johnston RA (2002) Uncertain socioeconomic projections used in travel demand and
emissions models: could plausible errors result in air quality nonconformity? Transp Res Part A:
Policy Pract 36(7):613–631
Su H et al (2014) A land use and transportation integration method for land use allocation and
transportation strategies in China. Transp Res Part A: Policy Pract 69:329–353
Tayarani M et al (2018) Can regional transportation and land-use planning achieve deep reductions
in GHG emissions from vehicles? Transp Res Part D: Transp Environ 63:222–235
Waddell P, Nourzad F (2002) UrbanSim: modeling Urban development for land use, transportation
and environmental planning. J Am Plan Assoc 68(3):297–314
Waddell P et al (2003) Microsimulation of Urban development and location choices: design and
implementation of UrbanSim. Netw Spat Econ 3(1):43–67
WaddellP et al (2010) Micro simulating parcel-level land use and activity-based travel: development
of a prototype application in San Francisco. J Transp Land Use 3(2):65–84
Zahabi SAH et al (2012) Transportation greenhouse gas emissions and its relationship with urban
form, transit accessibility and emerging green technologies: a montreal case study. Procedia -
Soc Behav Sci 54:966–978
Zhao L, Peng ZR (2010) Integrated bi-level model to explore interaction between land use allocation
and transportation. Transp Res Rec 2176:14–25
Zhao L, Peng ZR (2012) LandSys: an agent-based cellular automata model of land use change
developed for transportation analysis. J Transp Geogr 25:35–49
Chapter 13
Hosting a Mega Event, a Drive Towards
Sustainable Development: Dubai’s Expo
2020
Bashar Taha and Andrew Allan
Abstract The hosting of a mega event creates enormous demand for new buildings
and facilities and requires the development of new urban areas and infrastructure.
Therefore, it needs considerable investment and planning by government to finance
the projects to build these new facilities. Notwithstanding this, the hosting of a mega
event is an opportunity for the host country to attract investment for the construction
of new buildings and facilities required for the event and it expedites the upgrading of
its infrastructure and transportation network. The downside of hosting a mega event
is its relatively short period and the financing pressures on the host country. One of
the critical elements of hosting mega event is that the planning, during and post event
must be integrated and in harmony with the overall strategic planning of the host city
in order to deliver a sustainable and lasting legacy, to avoid the risk of redundant
investment in new facilities, buildings and infrastructure. This research work exam-
ines and reviews the consideration given during the planning phase by Dubai, host
city for Expo 2020, in its efforts to deliver a sustainable urban environment and cre-
ate a lasting legacy beyond the event. Interviews with the event organizing authority,
public transportation authority and major developers were carried out to collect data,
understand the conceptual element of planning and examine the transport models
employed and analysis that the RTA applied to determine the effectiveness of the
planning tools used by the project authorities of Expo 2020.
Keywords Expo 2020 ·Dubai ·Mega event ·Post-event planning ·Transport
planning ·Legacy plans
B. Taha (B)·A. Allan
School of Art, Architecture and Design, University of South Australia (UniSA),
Adelaide, Australia
e-mail: bashar.almosuli@mymail.unisa.edu.au
A. Allan
e-mail: Andrew.Allan@unisa.edu.au
B. Taha
Rail Right of Way Department, NOC Roads and Transport Authority, Dubai, UAE
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_13
223
224 B. Taha and A. Allan
1 Introduction
The term ‘mega-event’ appeared fairly recently in academic studies and its first
use can be traced to the 37th Congress of the Association Internationale d’Experts
Scientifiques du Tourisme in Calgary in 1987 with the theme ‘The Role and Impact
of Mega-Events and Attractions on Regional and National Tourism Development’
(Müller 2015). A mega event is usually typified by its mass popular appeal and
global significance and it provides a potential opportunity to catalyze major urban
redevelopments that are more closely attuned to achieving a sustainable environment
and economic development of the hosting city.
Mega events such as the Olympic Games, FIFA World Cup or World Fair Expo are
large-scale events that draw the interest of massive numbers of people from all walks
of life. They are intended to encourage local and regional economic development by
attracting investment, tourism and media attention for the host city.
Many researchers (Grix 2013; Hiller 2000a,b; Smith 2012) stated that host cities of
a mega event aim to make strategic use of mega events and attach great importance
to factors such as the event’s economic implications, event-related income, urban
development and regeneration, building and upgrading infrastructure, providing a
transportation system capable of transporting the expected numerous number of
visitors.
The multi-billion-dollar spending on mega events has an immediate impact on host
cities and regions, on both population and the built environment. Some researchers
adopted systematic approaches to defining the term mega event and distinguish
between different events in term of size based on a multi-dimensional, point-based
classification model to distinguish between different event scales. The model con-
sists of four key dimensions namely: visitor attractiveness; mediated reach; cost; and
transformative impact (Müller 2015).
The Bureau International des Expositions (BIE) defines Expo as a global event that
aims to educate the public, share innovation, promote progress and foster cooperation.
It can be considered to be an international dialogue platform, as the Expo brings the
whole world together to find solutions to the fundamental challenges of humanity.
Therefore, hosting a mega event is a great opportunity for the host city to foster
development, generate employment, improve the global business and investment
environment and develop tourism often through leveraging funds that would not
otherwise be available for a city to capitalize on (BIE 2017).
By contrast, the hosting of a mega event may become a threat that could overwhelm
the economic position of the host city, particularly if the event is not planned as a
contribution to a sustainable urban environment with a lasting legacy of infrastructure
and facilities.
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 225
2 Research Significance
This chapter explores the conceptual elements adopted by the various project author-
ities of the Dubai government during the event’s planning phase to minimize the risk
associated with not creating a lasting urban development legacy. The spatial con-
text includes post-event direct impacts associated with the planning of facilities and
buildings for the Expo event site, and indirect impacts such as the new autonomous
rail metro route extension which will form a spine of new transit oriented urban
development.
In the planning for the new metro route, the most important factor considered was
that the service shall continue beyond the event period to ensure the sustainability
and feasibility of the capital investment in building the new metro railway line.
In this regard, the Roads and Transport Authority (RTA) of Dubai examined dif-
ferent alignment options to achieve this goal. The decision criterion for the optimum
alignment was that the selected extension should best serve the event, in addition
to serving other new and existing developments and contributing towards a last-
ing legacy. Interviews were carried out with project authorities and developers, and
important secondary sources such as RTA documents and reports were critically
reviewed. This chapter critically examines the investigation and planning of an exten-
sion of the autonomous commuter rail Dubai Metro (Route 2020 Metro Line) to serve
the transport needs of Dubai’s Expo 2020, including the new urban development that
will emerge from Expo 2020. Expo 2020 presented a sustainable plan for reusing
80% of buildings, facilities and infrastructures constructed for the event of Expo for
legacy uses. The required budget for public and private transportation and people
mobility for this legacy was considered during the transportation planning phase,
and long-term utilities and infrastructure needs were integrated into the planning
phase of the event. Ultimately, however, post-project completion, further research
work will need to be carried out to examine the implementation of the transition
from event to legacy to measure and validate the concepts adopted by the planning
authorities in Dubai to deliver a sustainable event with a lasting legacy.
3 The Impact of Hosting a Mega Event: Previous Examples
Previous research (Surborg et al. 2008; Short 2008) findings determined that hosting
a mega event provides an incentive and opportunity for city elites to restructure
their cities in an increasingly competitive environment. Mega events have often been
described as a lucrative tool for place promotion and marketing—and as a key link
between the local and the global. For instance, South Korea is an example of utilizing
the hosting of a mega event to provide effective pathways to facilitate epic large scale
urban redevelopment, such as in developing the new national capital Songdo within
a tight timeframe (Surborg et al. 2008; Short 2008).
226 B. Taha and A. Allan
In another example of a hosting city capitalizing on its Olympic legacy, Barcelona
was able to boost its economic growth, enhance its image and transform itself into
a globally competitive city. Barcelona’s success indicates the significance that the
Olympic Games can have for urban development practices and urban policy in host
cities, and, equally, the importance of understanding the Olympic Games from an
urban development perspective (Chen and Spaans 2009).
When the legacy concept is not properly addressed in the planning phase for a
mega event, the host city will invest capital in under-utilized developments and infras-
tructures, such as was the case of Montreal’s 1976 Summer Olympic Games, where
the mega event saddled the government with massive debts and created potential
urban blight with under-utilized assets and “white elephant” projects.
In Sydney, host of the 2000 Summer Olympics, the Games were a significant
catalyst for urban infrastructure development around the region. Besides the direct
investments made for the Games, the indirect investments after the Olympic Games
were expedited. These improvements included better transport connectivity and a
major capacity expansion scheme to its airport, Kingsford Smith International, as
well as capacity improvements at its main rail hub, Central Station (Richter 2012).
The 2000 Summer Olympics Games reinvigorated and rehabilitated a part of
Sydney that was largely brownfield land with limited appeal to the community;
hence, transforming this setting into prime real estate is a lasting legacy of Sydney’s
hosting of the Olympic Games. However, it is worth noting that the Homebush Bay
site is not quite what the State Government wanted it to become, as it has been
unable to capitalize on tourism and only achieved partial success in transitioning
itself to an office park precinct. In addition, many of the sporting facilities were
eventually deemed to be surplus to sport and recreation requirements post Olympics,
and the Sydney Olympic Stadium was downscaled immediately after the Games.
Interestingly, up until recently, the Stadium faced the prospect of demolition, although
this was rescinded by the Berejiklian Liberal New South Wales Government in late
2018 in response to community protest at the profligate waste involved and concerns
around securing government in a state election looming in early 2019.
Athens, host of the 2004 Summer Olympics, had transport issues that were signif-
icantly different from Sydney’s. Athens is an ancient city with a dense urban form.
It also did not have much of the tertiary structure that is necessary to handle the
increased demands of an Olympic Games. Due to the city’s urban form and a lack of
large parcels of available public land, Athens had to spread out its Olympic venues
across the Attica Plain. This was problematic due to the notorious traffic conges-
tion facing Athens and the limited existing public transport infrastructure within the
city. Thus, by agreeing to host the Olympic Games, Athens embarked on large-scale
transport investments. The direct and indirect investments in transport infrastructure
included a new international airport, two metro lines, a tram system, and a suburban
railway. All of these infrastructure improvements were built with the goal of making
transport more efficient during the Olympics (Richter 2012).
Chalkley and Essex (Chalkley and Essex 1999) stated that different cities have
shown increased interest in the idea of promoting urban development and change
through the hosting of major events. This approach offers host cities the possibility
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 227
of ‘fast track’ urban regeneration, a stimulus to economic growth, improved transport
and cultural facilities, and enhanced global recognition and prestige.
The Olympic Park site at Stratford, London is one of best examples of an urban
regeneration initiative and a sustainable development that resulted from hosting the
2012 Olympic Games in London (Richter 2012). The land of the Olympic Park was
used as landfill after the 2nd World War bombing of London, and was compromised
by poor drainage issues, with utility and transport infrastructure criss-crossing the
site resulting in its functional fragmentation. The objective of using the Stratford site
was to provide quality infrastructure: the value of the site and its surrounding areas
was to be improved socially, physically and economically.
London’s model for urban development was similar to Sydney. It had an area
ripe for regeneration at Stratford. London also had transport connections near the
site of the Olympic Park but needed significant investment in public transportation
infrastructure to make the site accessible. The Olympic Village was also adjacent
to the Olympic Park, in a similar arrangement to that found in Sydney. However,
the similarities between the two cities end there. London had a much more complex
set of existing transport infrastructure already in place when the Olympic Park was
developed. The success that was the key for London’s 2012 Plan was to arrange and
maximize the efficiencies of its transport infrastructure to serve the Games and assist
in regenerating the area around the Olympic Park afterwards (Richter 2012).
4 Dubai Expo 2020
Dubai Expo 2020 will attract around 180 nations worldwide and is expected to
receive millions of visitors. Following the six-month event period of Expo 2020, the
Dubai Expo site, when completed, is expected to push the boundaries of architecture,
smart technology and sustainability. Easy access and availability of transportation
are important factors in successfully hosting the mega event. The Expo 2020 Dubai
location was chosen for optimal operational and logistical efficiency. Figures 1and
2detail the geographic location of the Expo 2020 site within Dubai and the nearest
airports and metro railway network to the location.
The Expo site is located 6 km away to the new Al Maktoum International Airport
(DWC) and halfway between Abu Dhabi and Dubai and within an hour drive from two
other major international airports (Dubai and Abu Dhabi International). The proposed
location of the new Dubai World Central within Expo 2020 site and its integration
with the DWC’s larger aerotropolis concept enables the Expo to leverage transport
connectivity to Al Maktoum International Airport, in terms of logistics, aviation,
residential and commercial facilities. The Airport has been operational for air-freight
since 2010, but because of financial constraints its eventual opening as a full service
airport to include passengers has been delayed until 2027. Whilst the experience
with mega-events elsewhere provide the nucleus to provide the longer term growth
spread effects across the immediate hinterland, as characterized in Perroux’s growth
pole theory (Lasuen 1969), it could be argued that Dubai Expo is part of a much
228 B. Taha and A. Allan
Fig. 1 Location of Dubai Expo 2020 in Dubai (UAE) and nearby airports (Reproduced with the
permission of the publisher Dubai EXPO) (Dubai_Expo 2020; Dubai_Expo_Annual_Report 2016)
Fig. 2 Location of Dubai Expo 2020 in Dubai (Reproduced with the permission of the publisher
Dubai EXPO) (Dubai_Expo 2020; Dubai_Expo_Annual_Report 2016)
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 229
grander metropolitan vision for Dubai by the Dubai authorities, because so many
large scale projects appear to be occurring in tandem, and not necessarily because of
Expo 2020. For example, the new Dubai Creek Tower, scheduled for completion in
2021, was originally intended to eclipse the Burj Khalif in height to secure the title of
the world’s tallest, although the Jeddah Tower in Jeddah has overtaken this ambition.
Nevertheless, the scale of Expo 2020s construction undertaking is considerably larger
than Dubai’s other recent urban development initiatives such as Dubai’s Palm Island.
4.1 Expo 2020 Master Plan
A key objective of the Masterplan is to facilitate a global dialogue during Expo 2020,
bringing to life the main theme, ‘Connecting Minds, Creating the Future’, and seam-
lessly integrating the three sub-themes of Opportunity, Mobility and Sustainability.
The Expo 2020 site covers 4.38 km2that includes 180 country pavilions. Al Wasal
Plaza (‘The Connection’) is a huge dome in the heart of Expo 2020, bringing the
subthemes together in a single large space that is a physical manifestation of the main
theme of Expo 2020 (expo2020dubai.ae 2016).
Three pavilions are designed to represent the mobility, sustainability and oppor-
tunity sub-themes. The mobility pavilion is designed to enable smarter and more
productive movement of people, goods and ideas and allowing individuals and com-
munities to reach their potential. While, the sustainability pavilion champions ways
to live in balance within the boundaries of the environment. The opportunity pavilion
is planned to represent Expo 2020 Dubai’s commitments to unlocking the potential
of individuals to create a better future.
At least 50% of the Expo site’s energy needs will be supplied from renewable
resources. A power supply of 100 MW will be provided from Mohammed Bin Rashid
Al Maktoum solar park. The Expo site is designed to provide accommodation for
staff and participants within the Expo village. The Expo 2020 Dubai’s onsite support
facilities include a retail mall and the Expo 2020 Dubai village that will have 2100
residential units for the participants, along with food and beverage outlets and shops.
In theory, Masterplanning is a holistic planning approach, and indeed in the case
of Expo 2020, the guiding principles in the design embrace contemporary planning
concerns, except with regard to the preferences of a future community. While the
vacant nature of the site and the Metro extension corridor prior to development
provide some understanding of the difficulty in including community input into
the design process, and instead leaving this in the hands of experts, nevertheless,
it does pose a significant question concerning the successful use of mega events
as a major vehicle in legacy planning. The stakeholders interviewed accepted the
design concepts developed for Expo 2020, and viewed the design and planning
process as a technocratic undertaking, determined by key performance milestones
set in place by the Masterplan and the contracting authority. There was never any
questioning of Expo 2020s Masterplan concept or its design principles, and there
was strong confidence in the design solutions and planning process. The Masterplan
230 B. Taha and A. Allan
Fig. 3 Dubai Expo master plan, provided by the organizing authority (Reproduced from www.
expo2020.ae) (Dubai_EXPO2020 2020)
guiding Expo 2020s development (Fig. 3) was based on a number of guiding planning
principles (expo2020dubai.ae 2016) and these are:
1. Manifest the theme and sub-themes.
2. Build a lasting and viable legacy.
3. Maximise visibility for all participants.
4. Facilitate collaboration and flexibility among participants.
5. Reflect local culture and heritage, authentically representing the UAE.
6. Raise standards in visitor experience.
7. Achieve excellence in logistics for participants’ experience.
8. Become a reference in sustainability for future World Expos.
9. Design activity and accessibility for all audiences.
10. Prioritise safety in delivery and operation of the Expo site.
11. Catalyse and showcase innovation.
5 Expo Buildings and Facilities During Event and Legacy
As mentioned earlier, legacyplanning is embedded in mega event planning; therefore,
Expo 2020 buildings and facilities are designed to meet the requirements of a mega
event and become key elements that create a lasting legacy. The data collected from
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 231
interviews and reports revealed that the ideas surrounding the concept of sustainable
planning in creating a lasting legacy are largely explained in the design and function
of the event’s major buildings.
The Expo 2020 site and its infrastructure are designed to be the platform for this
legacy. As per Expo 2020 authorities, over 80% of the site including the buildings and
facilities will be reused, providing a lasting asset and legacy to Dubai. The concept
of the legacy is defined as a transition from an incredible world expo to a community
unlike any other where people can work and where future business happens, living
within a culturally rich and vibrant community amongst breathtaking architecture
and exemplary design (Dubai_Expo 2020; Dubai_Expo_Annual_Report 2016). The
function of the major buildings of Expo 2020 during event and lasting legacy are
explained in the next section.
5.1 Conference and Exhibition Facilities (CoEX)
CoEx will have over 35,000 m2of floor area for the Expo 2020 event. The legacy
of the CoEx area is that it will provide a major event and exhibition venue in Dubai
that will be owned and operated by Dubai World Trade Centre.
5.2 Al Wasl Plaza
Al Wasl Plaza is at the heart of the Expo site in geographical location, orientation
and the visitor experience. Al Wasl Plaza will be within 15 minutes walk from any
part of the site. It is designed as a central open public space, functioning as the main
point of orientation for visitors, as well as an event and entertainment hub, hosting
large-scale events, performances and concerts. Al Wasl Plaza will also be the key
public space for the National Day celebrations of participating countries, as well as
forming an integral part of the Opening and Closing Ceremonies.
5.3 Sustainability Pavilion
The Sustainability Pavilion will be built on an area of 29,000 m2. The pavilion will
target net zero energy by harvesting its energy from solar panels. Humidity harvesting
systems will also be used to meet substantial water usage targets. The legacy of the
Sustainability Pavilion is that it is intended to become a Science Exploratorium with
an emphasis on inspiring and empowering youth to become standard bearers for the
better stewardship of our planet.
232 B. Taha and A. Allan
5.4 Mobility Pavilion
The Mobility Pavilion will be built on a 12,000 m2plot. This building is also being
developed and delivered by local developer (Emaar) and its legacy will be in providing
premium future office space.
5.5 Opportunity Pavilion
The Opportunity Pavilion will be built on a 12,000 m2plot. The Opportunity Pavil-
ion is designed to be influenced by a belief that contemporary urban life is shaped
by a confluence of cultural exchange, global economic trends and communication
technology.
5.6 UAE Pavilion
The UAE Pavilion will be built on a 15,000 m2site. The opportunities for its use in
legacy as a media centre are being assessed by the National Media Council.
At this stage of the research, the fulfilment of the planning principles outlined
earlier in Sect. 4is yet to be determined as the project is still in its construction
phase. The event will start on Oct 2020 and will last for 6 months until April 2021.
As per Expo 2020, the implementation of legacy planning for District 2020 will
continue in phases in accordance with Dubai’s strategic plans. Figure 4shows the
project phasing including planning for its legacy beyond the life of Expo 2020. The
authors prepared the project lifecycle using information available from the District
2020 website.
6 New Railway Line for Expo 2020—Route 2020
One of the most important elements considered by the authorities while planning
for the new railway route to serve 2020 was the sustainability and continuity of
the service that can be translated to increasing ridership and revenue. The RTA
planned to achieve ridership levels that will secure a return on investment. In this
regard, the RTA explored different options for route alignments to address this key
element—development of a new metro line in more densely populated urban areas,
aligning with and further supporting additions to the Expo 2020 site.
The surrounding area affected by the potential alignment of new metro Route
2020 is a mixture of business, residential, industrial, and mixed-use developments
that encompass a number of emerging developments, such as Nakheel developments
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 233
Fig. 4 Project phasing including legacy planning for the site (Reproduced from www.district2020.
ae) and (Dubai_EXPO2020 2020)
(The Gardens, Discovery Gardens and Al Furjan), Jumeirah Golf Estate (JGE), Dubai
Investment Park (DIP), Expo 2020 and DWC. The detailed key performance indica-
tors (i.e. population, jobs and tourists) for developers are expected to improve with
the completion of the new railway line Route 2020, as shown in Table 1.
7 Factors Affecting Trip Generators for Route 2020
The analysis of the data related to trip generation and expected ridership of the
new Metro line is based on information obtained in Dubai and the study area for
population, jobs and tourism. The modelling and analysis was carried out using
Visum software provided by the PTV Group.
7.1 Population
The strategic plan of Dubai for 2020 projects the population will reach over 3 million
and increase this to almost 6 million by 2030. The corresponding population within
the Expo 2020 area is expected to reach approximately 430,000, in 2020, expanding
to 890,000 by 2030. The estimated population for the whole of Dubai and Expo 2020
in 2020 and 2030 are presented in Fig. 5. The resulting forecast population spatial
distribution obtained from the analysis of data in 2020 and 2030 is shown in Fig. 6.
234 B. Taha and A. Allan
Tabl e 1 Key planning data for main zones in study area (Transport_Model_Report 2015)
Zone Population
2020
Population
2030
Jobs
2020
Jobs
2030
Tourism 2020 Tourism 2030
Dubai World
Central
23,806 83,954 18,359 118,906 4666 10,857
EXPO 2020 –6249 –38,106 – –
Dubai
Investment
Park
58,576 152,855 28,417 71,040 4902 13,033
Golf Estates 18,786 52,214 26,482 67,286 16 46
Me’aisem 1 8824 25,628 10,402 27,739 26 72
Nakheel 141,679 257,666 111,465 239,056 1842 5440
Emaar 24,805 27,762 8119 8124 221 268
Jebel Ali
Vil la ge
6659 18,036 1361 3401 – –
Limitless 3663 5722 17,497 34,993 1129 1800
Jebel Ali
Free Zone
75,703 95,639 17,809 28,318 – –
Dubai Sport
City
26,793 67,155 16,503 38,082 1574 3858
Golf City 5654 39,670 275 1789 329 2274
Others 34,882 93,324 19,408 72,996 317 2520
Total (Study
Area)
429,830 925,874 276,097 749,836 15,022 40,168
30,73,699
58,96,596
4,29,830
9,25,874
0 20,00,000 40,00,000 60,00,000 80,00,000
2020
2030
Study Area Dubai
Fig. 5 Population in 2020 and 2030 (Transport_Model_Report 2015)
236 B. Taha and A. Allan
7.2 Jobs
The projected number of jobs in Dubai in 2020 will reach 2 million and by 2030,
increase to over 4 million. There will be over 276,000 jobs in the study area in 2020
and around 750,000 in 2030 as presented in Figs. 7and 8.
7.3 Tourism
In 2020, the daily average population of tourists in Dubai provided by Expo and
tourism authorities is predicted to reach 100,000 and by 2030 this is expected to
increase to over 200,000. The daily number of tourists residing in hotels in the study
area is expected to increase from approximately 15,000 in 2020 to around 40,200 by
2030 as shown in Fig. 9. Figure 10 compares the change in the spatial distribution
of tourists from 2020 to 2030 as a result of the legacy effects of Expo 2020 and the
associated investment in Route 2020.
8 Transportation Modelling and Analyzing Planning Data
The planning of Route 2020 has played a key part in identifying potential oppor-
tunities to serve major developments along its route. The current transportation
model adopted by the RTA Dubai (Transport_Model_Report 2015) using the soft-
ware Visum of the PTV Group was used to estimate reliable passenger ridership
forecasts for defining key system parameters and station sizing.
The model has been updated incorporating the most recent forecasts for Expo 2020
rail passengers, including planning data from the local land authority and the major
20,72,442
42,77,078
2,76,097
7,49,836
0 10,00,000 20,00,000 30,00,000 40,00,000 50,00,000
2020
2030
Study Area Dubai
Fig. 7 Employment forecast in 2020 and 2030 (Dubai vs. Study Area)
238 B. Taha and A. Allan
99,347
2,03,846
15,022
40,168
0 50,000 1,00,000 1,50,000 2,00,000 2,50,000
2020
2030
Study Area Dubai
Fig. 9 Tourist figures in 2020 and 2030 (Transport_Model_Report 2015)
developers within the study area. In order to obtain the most sustainable alignment
among different options, many documents and information were collected during
stakeholder interviews to assist in selecting a coordinated route. The various com-
ponents are as follows:
1. RTA Transportation Models for 2020 and 2030.
2. Expo 2020 Transport Model.
3. Legacy Transportation Model.
4. Observed Dubai Metro passengers’ ridership.
5. The Masterplan for key developments within the study area.
6. Existing and future bus routes within the study area.
According to RTA reports and data, the new extension route will transport 46,000
passenger per hour to and from the Expo site with a frequency of every 16 min from
Dubai Marina (Transport_Model_Report 2015; RTA_Feasibility_Report 2015). As
determined by the transportation and crowd management modelling, the expected
share of transportation of the Route 2020 Metro line to the event is 19% of the total
transportation budget as shown in Fig. 11.
9 Route 2020 Alignment Options and Evaluation Criteria
Route 2020 was conceived as a transport solution to connect Dubai’s Metro Red
Line to the Expo 2020 site and provide a catalyst for further urban development in
this part of Dubai’s underdeveloped metropolitan region. A key objective of the RTA
in developing the Route 2020 extension of the Dubai Metro was that in light of the
existing and proposed developments associated with Expo 2020, transit ridership on
the Metro would be maximized. Route 2020 would be also be an ideal opportunity
to shift Dubai’s modal share from cars to public transit and create densely populated
transit corridors. The alignment options are shown in Fig. 12.
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 241
Tabl e 2 Summary of alignment options for the Route 2020 extension of the Dubai Metro
Option 1 Option 2 Option 3 Option 4
Length Stations Length Stations Length
(km)
Stations Length
(km)
Stations
Elevated 9.0 km 412.86 km 2+
Airport
8.1 311.5 5
Tunnel – – – – 6.3 23.5 2
Tot a l 9.0 km 412.86 km 2+
Airport
14.4 515.0 7
Summary Including a length
of about 300 m
required to
connect with
Existing Red line
At grade for a
length of 500 m at
the beginning of
the Alignment
Including a length
of about 375 m
required to connect
with Exist. Red
line
Includes length of
about 940 m
required to connect
with Exist. Red
line
Key data for the four alignment options obtained from RTA Dubai are listed in
Table 2with a summary noted against each option.
The principles of the evaluative approach applied by the RTA in this project is akin
to Planning Balance Sheet (or Matrix) evaluative techniques (Lichfield et al. 1975).
The advantage of the applied approach over traditional Cost-Benefit Analysis (CBA)
approaches that are traditionally used in project feasibility studies is that it can take
into account stakeholder concerns where CBA is deficient because all components
of a project are reduced to a single financial value, that is not necessarily objectively
determined.
To identify the best route alignment, the options were examined against predefined
measuring criteria developed by RTA, in order to produce an aggregated score that
would simply determine the preferred alignment option. The defined criteria by RTA
and their weight are:
1. Transportation and future development (Weighting of 55%).
2. Route feasibility (Weighting of 15%).
3. Constructability (Weighting of 20%).
4. Sustainability (Weighting of 10%).
The analysis method was based on qualitative assessment in the form of the Likert
rating scale in order to equally evaluate all alignment options and choose the best
alignment.
Each of the proposed alternative alignments were assessed and scored from 1 to
5 based upon each sub-criteria defined and weighted by RTA with regard to their
respective technical importance. A score of 5 is the highest rating achievable whilst
a score of 1 is the lowest. The value of these weightings were determined through
consultation with project stakeholders and developers.
These themes were weighted to reflect the RTA’s key considerations in determining
Route 2020s long term viability, based on maximized patronage which also related
to likely development opportunity.
242 B. Taha and A. Allan
Tabl e 3 Alignment evaluation matrix of transportation and future developments
#Criteria Weight of sub criteria (%) Options
1 2 3 4
1Passenger ridership projections 30 1 3 4 5
2Travel time to expo/Dubai
world center
20 3 1 4 5
3Multi-modal Integration 10 1 2 3 5
4 Station catchment
area—existing development
10
5 Station catchment area—future
development (TOD)
30 4 1 5 5
Score (Weighted) 2.5 1.7 4.2 5
Score (Out of 100) 50 34 82 100
9.1 Transportation and Future Development
The evaluation criteria related to transportation and its capacity to facilitate future
development were measured against the criteria listed in Table 3.RTAratedthe
potential of future development highly, due to its subsequent indication of continuity
of service after the Expo 2020 event. The area evaluated was the catchment area
around railway stations and the expected ridership and subsequent revenue generated.
9.2 Route Feasibility
The availability of the easement to build the railway line with minimal interruption
or impact on other infrastructure was an important element in delivering the project
on time. The potential route was assessed according to its Route Feasibility, defined
as the difficulty or ease of the route insertion along an optimal route alignment based
on the criteria listed in Table 4.
9.3 Constructability
This evaluation criterion measured the ease of construction and delivery of the project
in a timely manner without extra cost, whilst adhering to the health, safety and
environmental requirements. The constructability of the optimal route alignment
was assessed against five criteria listed in Table 5.
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 243
Tabl e 4 Alignment evaluation matrix of route insertion
#Criteria Weight of sub criteria (%) Options
1 2 3 4
1Availability of ROW 15 4 3 2 3
2Route quality 20 5 4 3 3
3 Station location consideration 5 4 2 3 3
4Conflicts with
roads/interchanges
30 3 3 4 3
5 Impact of utility/other
Infrastructure
30 4 1 3 3
Score (Weighted) 3.9 2.6 3.2 3.0
Score (Out of 100) 78 51 63 60
Tabl e 5 Route alignment evaluation matrix of constructability
#Criteria Weight of sub criteria (%) Options
1 2 3 4
1Site access 10 5 1 4 3
2Ease of construction 45 4 3 2 2
3Equip. and materials laydown
areas
10 4 3 5 4
4 Demolition of existing features 10 4 2 3 3
5 Health, safety and
environmental
25 3 3 2 1
Score (Weighted) 3.9 2.7 2.6 2.2
Score (Out of 100) 77 54 52 43
9.4 Sustainability
Sustainability is a major component in the planning of a project of this nature. The
assessment of this set of criteria evaluated critically significant issues that could
prevent the project from proceeding. This evaluation factor was measured against
the following evaluation criteria (Table 6).
9.5 Alignment Options Overall Scouring
Based on the technical assessment and the evaluation criteria of the four route align-
ment options, the overall scoring is presented in Table 7.
Figure 13 reveals the final selected route for Route 2020 (Option 4 with a score
of 78.6) and indicates the main developers within the alignment of Route 2020,
244 B. Taha and A. Allan
Tabl e 6 Alignment evaluation matrix of sustainability
#Criteria Weight of sub criteria (%) Options
1 2 3 4
1Ambient environment impact 25 4 3 3 2
2Conserve resources 25 4 3 1 2
3 Contamination discovery
potential
35 1 1 3 4
4Conserve green fields and
landscaping
15 1 2 4 4
Score (Weighted) 2.5 2.2 2.7 3.0
Score (Out of 100) 50 43 53 60
Tabl e 7 The technical overall score of all four alignment options
#Criteria Weight of sub criteria (%) Options
1 2 3 4
1 Transportation and its
capacity to facilitate future
development
55 2.5 1.7 4.2 5.0
2Route insertion 15 3.9 2.6 3.2 3.0
3Constructability 20 3.9 2.7 2.6 2.2
4Sustainability 10 2.5 2.2 2.7 3.0
Score (Weighted) 3.0 2.1 3.6 3.9
Score (Out of 100) 59.6 41.5 71.4 78.6
along with 400 m diameter Transit Oriented Development catchments around sta-
tions. The approach taken in evaluating the various route options appears to have
successfully blended the technique of transport modelling outputs with a Planning
Balance Sheet to determine a future development scenario that will extend Dubai’s
urban development in an environmentally sustainable manner with its electric com-
muter rail Metro and environmentally sustainable development (both in terms or
urban densities and energy efficient buildings). However, the absence of community
consultation in advance of planning urban development within the Route 2020 Metro
corridor is potentially a limitation in the planning and design process. Overcoming
this limitation could still be addressed through the use of focus community groups
at the planning and design stages of future developments that will be serviced by
Route 2020. The challenge in determining who is consulted can partly be solved by
including participants that are representative of the preferred or likely profile of the
community expected to reside in these developments (Bickerstaff et al. 2004).
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 245
Fig. 13 Route 2020 alignment of 400 m diameter around station
10 Conclusion
It can be concluded from the data and information obtained from the provided reports
and interviews undertaken with project stakeholders, such as RTA Dubai, Expo 2020
and major developers in the study area, that the main focus of RTA while planning the
new railway route is focused primarily on linking existing areas to new development
areas to generate maximized ridership and return on investment, which provides a
strong basis to creating an enduring legacy from Expo 2020. Notwithstanding this,
the initial driver for development of the new railway line is in connecting people to
the Expo 2020 site.
The approach adopted by RTA in terms of identifying weighted criteria and sub-
criteria to measure the alignment options against is in principle sound and effective
in equipping decision-makers with a framework to choose the best alignment option.
The framework could potentially be challenged by opponents due to its lack of
input from communities likely to be affected by the Masterplan process. The use
of technical experts encourages objectivity in the evaluation process, but as in any
qualitative rating scheme, subjectivity can taint the apparent technical objectivity of
the process.
The principles of the evaluative approach applied by the RTA in this project is
akin to Planning Balance Sheet (or Multi-Criteria Analysis) evaluative method. The
advantage of the applied approach over traditional Cost-Benefit Analysis (CBA)
approaches that are traditionally used in project feasibility studies is that it can take
into account stakeholder concerns, which CBA is deficient in because all components
of a project are reduced to a financial value.
246 B. Taha and A. Allan
The chosen alignment of Option 4 will potentially provide a fantastic opportunity
for the surrounding areas to further develop as it is expected that land and property
values will be uplifted upon completion of the new metro line.
However, an area of criticism in the study is the unforeseen impact that could
come from the global financial situation that can directly influence the employment
and population figures of Dubai—and other parts of the world. This factor can lead
to a recession in the housing market due to an increase in supply with less demand.
From the provided data and analysis, the new railway line will serve a population
of 270,000 people, operating over a total length of 15 km, with 7 stations. The
capacity of Route 2020 is estimated at about 46,000 passengers per hour in both
directions—23,000 passengers in each direction per hour. The number of users of
Route 2020 is expected to reach 125,000 passengers per day by 2020 and will rise
to about 275,000 passengers per day by 2030. It is anticipated that about 35,000 of
Expo visitors per day will use route 2020 and this number will rise to 47,000 visitors
a day during the weekend. These figures represent about 20% of the total number of
daily visitors expected to visit the Expo during the event.
The alignment of Route 2020 is part of a metropolitan Metro network that will
eventually integrate Expo 2020 and its environs and Al Maktoum International Air-
port through an additional 3.4 km extension.
It is believed that the main driver behind the selection of the multi-criteria analysis
(MCA) approach over cost-benefit analysis (CBA) in the evaluation process carried
out by RTA is the wide spectrum and diversity of project stakeholders, as the align-
ment passes through the lands of developers that follow different zoning and regula-
tory authorities. Therefore, due to different regulatory frameworks and requirements
of these developers and zoning authorities, the MCA evaluation approach became
more effective and flexible in addressing the needs of each stakeholder, with greater
capacity to incorporate environmental sustainability assessments, rather than CBA
which tends to be far too reductionist.
Expo 2020 presented a sustainable plan to deliver the event and a lasting legacy
(the District 2020), with plans to reuse 80% of buildings, facilities and infrastructures
constructed for the Expo. The required budget for both public and private transporta-
tion and people mobility for the legacy was considered during the planning phase, in
addition to the upgrade of other infrastructure, such as the networks of power supply,
potable water, storm water and sewerage. However, further research work is recom-
mended to explore the implementation of a transition from the mega event to a lasting
legacy through examining the extent to which the design, development, planning and
operational concepts will deliver a sustainable mega-event with an enduring legacy.
In addition, further studies are recommended to evaluate how hosting a Mega Event
will assist Dubai in meeting the UN Sustainable Development Goals of 2030.
Finally, it can be concluded that the effect and benefit of hosting a major event such
as Expo 2020 has been to accelerate Dubai’s expansion and bring forward expansion
of its Metro network, which will be instrumental in transitioning Dubai to a smarter,
more environmentally sustainable metropolis.
13 Hosting a Mega Event, a Drive Towards Sustainable Development … 247
Acknowledgements The researchers would like to express their sincere gratitude to Roads &
Transport Authority (RTA) Dubai and Expo 2020 Dubai for the provided support and information
and graphical presentation required to carry out this research work. All data pertaining to the route
analysis was kindly provided by the RTA.
References
Bickerstaff K, Simmons P, Pidgeon P (2004) Public perceptions of risk, science and governance:
main findings of a qualitative study of six risk cases. Centre for Environmental Risk, University
of East Anglia, Norwich
BIdE (BIE) (2017) Bureau international des expositions. Available at: http://www.bie-paris.org
[Accessed 12 Oct 2018]
Chalkley B, Essex S (1999) Urban development through hosting international events: a history of
the Olympic Games. Plan Perspect 1(14):369–394
Chen Y, Spaans M (2009) Mega-event strategy as a tool of Urban transformation: Sydney’s expe-
rience. The New Urban Question—Urbanism beyond Neo-Liberalism, Amsterdam, Delft
Dubai_Expo (2020) Dubai_Expo_2020 [Online]. Available at: expo2020dubai.ae, 2016 [Accessed
10 Nov 2018]
Dubai_EXPO2020 (2020) EXPO 2020 Dubai [Online]. Available at: http://www.expo2020dubai.
ae [Accessed 10 Oct 2017]
Dubai_Expo_Annual_Report (2016) Planning an Expentional Expo- Expo 2020 Dubai 2016 Annual
Report. Dubai Expo, Dubai.
expo2020dubai.ae (2016) Planning and exertional expo. Expo 2020 Dubai—2016 Annual Review.
EXPO 2020, Dubai
Grix J (2013) Sport politics and the olympics. Polit Stud Rev 11(1):15–25
Hiller H (2000a) Mega-events, urban boosterism and growth strategies: An analysis of the legiti-
mations of the Cape Town 2004 Olympic bid. Int J Urban Reg Res 24(1):439–458
Hiller H (2000b) Toward an urban sociology of mega-events. Res Urban Sociol 5(1):181–205
Lasuen, JR (1969) On growth poles. Urban Stud 6(2):137–161
Lichfield N, Kettle P, Whitbread M (1975) Evaluation in the Planning Process. Urban Reg Plan Ser
10(1):344
Müller M (2015) What makes an event a mega-event? Definitions and sizes by: Leisure Studies, pp
10–26
Richter R (2012) The olympic games and Urban development. TRANSPORT NEXUS, Chicago
RTA_Feasibility_Report (2015) Red line connection to EXPO 2020 site: feasibility report. Rail
Agency Roads & Transport Authority, Dubai
Short J (2008) Globalization, cities and the summer olympics. City 12(3):321–340
Smith A (2012) Events and urban regeneration: the strategic use of events to revitalise cities, 1st
edn. Routledge, London
Surborg B, van Wynsberhe R, Wyly E (2008) Mapping the olympic growth machine: transnational
urbanism and the growth machine diaspora. City 12(3):341–355
Traffic_Impact_Report (2017) Traffic impact study. RTA, Dubai
Transport_Model_Report (2015) Red line connection to EXPO 2020 site: transport model review
report. Rail Agency RTA Dubai, Dubai
Chapter 14
Deep Learning Architect: Classification
for Architectural Design Through
the Eye of Artificial Intelligence
Yuji Yoshimura, Bill Cai, Zhoutong Wang and Carlo Ratti
Abstract This paper applies state-of-the-art techniques in deep learning and
computer vision to measure visual similarities between architectural designs by dif-
ferent architects. Using a dataset consisting of web-scraped images and an original
collection of images of architectural works, we first train a deep convolutional neural
network (DCNN) model capable of achieving 73% accuracy in classifying works
belonging to 34 different architects. By examining the weights in the trained DCNN
model, we are able to quantitatively measure the visual similarities between architects
that are implicitly learned by our model. Using this measure, we cluster architects that
are identified to be similar and compare our findings to conventional classification
made by architectural historians and theorists. Our clustering of architectural designs
remarkably corroborates conventional views in architectural history, and the learned
architectural features also cohere with the traditional understanding of architectural
designs.
Keywords Architecture ·Design classification ·Deep learning ·Computer vision
Y. Yo s h i m u r a ( B)·B. Cai ·C. Ratti
SENSEable City Laboratory, Massachusetts Institute of Technology,
77 Massachusetts Avenue, Cambridge, MA 02139, USA
e-mail: yyoshi@mit.edu
B. Cai
e-mail: billcai@mit.edu
C. Ratti
e-mail: ratti@mit.edu
Z. Wang
Department of Architecture, Harvard GSD, 48 Quincy St. Cambridge, Cambridge,
MA 02138, USA
e-mail: zwang1@gsd.harvard.edu
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_14
249
250 Y. Yoshimura
1 Introduction
This paper proposes to classify architectural designs through computer vision tech-
niques, purely based on their visual appearances. The question asked is whether or
not state-of-the-art deep learning techniques can identify the distinguishing design
features of each architect and cluster them in a similar way to architectural historians
and theorists. Our hypothesis is that computer vision, or “the machine’s eye,” could
provide different views and insights than traditional architectural theory, and it could
explain, quantitatively, the difference in designs between architects.
Architectural history and theory classify architectural styles and types from vari-
ous perspectives (see Forty 2000, pp. 304–311). A style of architecture (e.g., Renais-
sance, Baroque) provides a basic format for the design of an individual building in a
geographical region during a specific epoch. The ornaments that pertain to the spe-
cific style are considered as each age’s expression of beauty; these features convert
ordinary buildings into structures of architectural significance. Thus, visual elements
such as windows, pillars, or architectural orders (Onians 1988) can provide clues for
identifying and classifying a building’s architecture into a specific style.
Conversely, compared with the element-based classification for the historical
types of architecture, most classifications for modern and contemporary architec-
ture are largely either function-based or building type-based. This is largely due to
changes in design concepts. The international style (Hitchcock and Johnson 1932)
aims to express the function of the building through a “machine aesthetic,” resulting
in the shaping of modern architecture into a white cube. The historical ornament and
decoration were rejected, and “the machine” became the model for modern archi-
tecture. Thus, the modernists tend to reduce any forms to abstraction (Frampton
1992, p. 210). In addition, space and its experience have become some of the most
important topics in the design of modern and contemporary architecture. This fur-
ther complicates the classification because space cannot be described by elements;
rather, it appears when it is enclosed by the combination of several spatial elements
together with light. Consequently, the classification comes to rely more on abstract
and dematerialized concepts rather than being based on elements, as it did in previous
periods.
In order to fill this gap, this paper proposes a computational approach to clas-
sify designs of modern and contemporary architecture. The objective of this paper
is twofold: (1) present our analytical framework as the research methodology and
(2) show the preliminary result of our current research. For this purpose, we employ
recently developed deep learning techniques in image processing to classify the given
datasets through the training samples. The obtained results are clustered depending
on the visual similarities measured by the algorithm. The final results are com-
pared with the classifications made by architectural historians and theorists. Thus,
we demonstrate that artificial intelligence is capable of developing an aesthetic clas-
sification of modern and contemporary architecture and can help us to enhance our
understanding of architectural design through the machine’s eye.
14 Deep Learning Architect: Classification for Architectural … 251
The paper is structured as follows: Sect. 2provides a literature review and
describes the analytical methodology for this paper. Section 3describes the dataset
used in our study. Section 4presents our study and the preliminary results. We con-
clude in Sect. 5, suggesting future work.
2 Related Works and Analytical Methodology
2.1 Related Works
The analytical methodology of this paper relies on deep convolutional neural net-
works (DCNNs), which have recently achieved remarkable performance in the fields
of image classification (Krizhevsky et al. 2012), scene recognition (Zhou et al. 2014),
speech recognition (Abdel-Hamid et al. 2012), and machine translation (Bahdanau
et al. 2015). The main advantage of DCNN methods over traditional computer vision
and machine learning techniques is their ability to identify and generalize important
features and employ these learned features to classify objects according to their appro-
priate labels. The visual features are engineered and extracted, assigning high-level
and semantic features to the input images without human intervention.
Table 1presents a summary of previous literature on classification of images in art,
architecture, and urban studies. Many studies deal with classification using low-level
features-based approaches (Li et al. 2012; Llamas et al. 2017; Obeso et al. 2017). For
example, Li et al. (2012) propose edge detection and clustering-based segmentation
to extract the characteristics of van Gogh’s brushstrokes and distinguish the artist
from others. Similarly, architecture studies propose to classify historical architecture
into styles based on historical architectural elements such as windows or pillars
(Llamas et al. 2017; Obeso et al. 2017). Conversely, urban studies focus on ordinary
buildings dispersed throughout a city to identify the urban elements that are the
determinant factors of each city (Doersch et al. 2012; Lee et al. 2015). Doersch
et al. (2012) explore the urban elements that appear frequently in a geographically
determined location but do not appear in other areas, while Lee et al. (2015) attempt
to identify the visual features that specify the architectural styles of each period and
the evolution of architectural elements over time.
Although the classification of historical architecture, including buildings, mon-
uments, and cultural heritage, is well researched, there have been few attempts to
classify modern and contemporary architecture or architects. In terms of techniques,
most previous literature employs clustering and learning of local features (Shalunts
et al. 2011) but not deep learning (Llamas et al. 2017).
This paper attempts to classify designs of modern and contemporary architecture
using a deep convolutional neural network. We try to capture spatial design features
rather than recognize specific visual features of buildings (e.g., windows, domes,
pillars). Our approach is similar to artistic style classification, in which recognizing
an artistic style is a different topic from identifying elements (Elgammal et al. 2018;
252 Y. Yoshimura
Tabl e 1 Summary of previous literature on classification of visual elements in art, architecture,
and urban studies
Objective Model Dataset
Elgammal et al.
(2018)
Characteristics of
style in art and
patterns of style
changes
AlexNet, VGGnet,
ResNet, and variants
76,921 paintings
from 1119 artists
with 20 classes from
WikiArt. 1485
images of paintings
from Artchive dataset
with 60 artists for
visualization and
analysis
Obeso et al. (2017)Classification of
Mexican heritage
buildings’
architectural styles
GoogLeNet and
AlexNet for a
Saliency-Based and a
Center-Based
approach
16,000 labeled
images in four
categories, out of
which three are
Mexican buildings
(pre-Hispanic,
colonial, modern)
and one is “other”
Llamas et al. (2017)Classification of
architectural heritage
elements
AlexNet and
InceptionV3for
CNN, ResNet and
Inception-resNet-v2
for Residual
Networks
More than 10,000
images classified into
10 types of
architectural
elements, mostly
churches and temples
Zhang et al. (2018)Prediction of urban
elements that cause
human perceptions
DCNN, PSPNet 1,169,078 images
from MIT Place
Pulse for training a
DCNN model.
245,388 images from
Google Street View
from Shanghai and
135,175 from Beijing
to predict human
perception
Cai et al. (2018) Quantification of
street-level urban
greenery
PSPNet (Pyramid
Scene Parsing
Network) and ResNet
for DCNN semantic
segmentation
500 street images
from Google Street
View and 500 images
of cityscapes from
vehicle-mounted
cameras
14 Deep Learning Architect: Classification for Architectural … 253
Saleh et al. 2016) because style can be considered independent from the content of
a drawing (Gatys et al. 2016). Thus, we explore the learned internal discriminative
factors to explain modern and contemporary architecture and its space.
2.2 Deep Convolutional Neural Network
The deep convolutional neural network (DCNN) is a class of deep, feed-forward
artificial neural networks mainly applied in the analysis of imagery. Figure 1ashows
a simple diagram of the reproduction of the neuron’s system in DCNN. Input data x
are multiplied by weight w, to which bias bis also added, for function f, producing y
as the output. Several neurons are combined to create the neural network (see Fig. 1b
for a diagram of the neural network). In order to be effective, a neural network
has to discover the optimal weights for all the connections in the network. Like
the human brain, which changes the strength of connections between synapses, the
neural network adjusts the weights through the learning process and seeks the best
combination of weights that minimizes the error between the correct classification
of an input and the output of the network at the last layer.
DCNNs stack many convolutional layers into a single network. Convolutional lay-
ers allow for dimensional reduction in high-dimensional problems and have driven
recent success in object detection, classification, and segmentation (Krizhevsky et al.
2012). Multiple stacked convolutional layers allow DCNNs to learn feature hierar-
chies, beginning with simple edges and shapes in the early layers and ending with
complex semantic features such as windows and roofs (Girshick et al. 2014).
In our experiment, we utilized NASNet, a novel program that achieves state-of-
the-art accuracy while halving the computational cost of the best reported results
(Zoph and Shlens 2018). NASNet is composed of two types of layers: a normal layer
and a reduction layer (Fig. 1c), both designed by auto machine learning, which is the
automated process for constructing models (Zoph and Shlens 2018).
2.3 Visual Explanation of DCNN
In a stacked convolutional neural network model, each layer contains increasingly
complex features and is optimized to identify architect’s distinguishing traits. Conse-
quently, the numerical matrix representing the weights in the last activation layer of
DCNN models represents high-level visual concepts that help to distinguish between
architects.
We employed gradient-weighted Class Activation Mapping (Grad-CAM) (Sel-
varaju et al. 2017) to examine NASNet’s visual explanations. It clarifies the influen-
tial gradients and their regions with respect to NASNet’s output. Unlike other pop-
ular visualization techniques, such as Class Activation Mapping (Zhou et al. 2014),
Grad-CAM combines feature maps using a gradient signal that does not require any
254 Y. Yoshimura
Fig. 1 a Diagram of the neuron. bDiagram of a fully connected neural network with N layers. The
input layer (zero layer) has three neurons, and the hidden layers have four neurons. cTwo repeated
motifs, termed “normal cell” and “reduction cell”, discovered as the best convolutional cells in the
CIFAR-10 dataset Adapted from Fig. 4 in Zoph and Shlens (2018)
modification in the network architecture, thus making it possible to apply to NAS-
Net. To compute Grad-CAM, we used the following formulas proposed by Selvaraju
et al. (2017):
Lc
Grad−CAM ∈Ru×v(1)
αc
k=1
Z
i
j
∂yc
∂Ak
ij
(2)
14 Deep Learning Architect: Classification for Architectural … 255
Fig. 2 Grad-CAM applied to an image taken from Alvaro Siza’s work and Grad-CAM’s heat map
by the model optimized for each architect
Lc
Grad−CAM =ReLU
k
αc
kAk(3)
The objective function in this task is defined as (1), where the uis the width and
vis the height for any class. First, we compute αc
k, the global average pooling, by (2),
in which Ak
ij indicates the element in matrix ij of the kth feature map, Ak
ij is the output
of the feature map A, and Z is the normalized item. Second, we compute Lc
Grad−CAM,
the heat map of Grad-CAM, by summing up the feature maps Akweighted by αc
k.
ReLU computes the pixel to increase the output of yc(see Selvaraju et al. 2017,for
a more detailed technical description).
Figure 2presents an example of Grad-CAM applied to an image taken from
Alvaro Siza’s work. The left picture is the original, and the right pictures are the
results of Grad-CAM for each architect. The red color indicates the location where
the machine’s eye focuses in order to identify the similarity of the design with the
original picture. For example, in Gehry’s picture, it focuses on the acute triangle
form, while the machine’s eye reacts to the form of the slope for Koolhass’s picture.
Thus, this technique enables us to understand the focus of the machine’s eye for the
classification of objects.
2.4 Dimension Reduction and Clustering
Using the outputs of the last layer, which is the product of the weights and the
outputs of the second-to-last layer (deep feature), we are able to cluster and measure
the similarities between the visual signatures used to distinguish different architects
by using linear principal component analysis (PCA) and kernel PCA (Jolliffe 2002).
256 Y. Yoshimura
To carry out PCA, we take and normalize the last layer before softmax for nimages
in DCNN. Then, we can construct D =[d1,d2,...,dn]matrix D ∈Rd×n, where
ddenotes the number of categories and ndenotes the number of images. Let the
first kprincipal components of Dbe B =[b1,b2,...,bk]. In our case, k =2. The
objective function is
max
b
n
i=1
BTdi2
2=BTB(4)
with constraint BTB=I. We use least-squares estimation to minimize the objective
function:
Epca(B)=
n
i=1
epca(ei)=
n
i=1
di−BBTdi2
2(5)
After dimension deduction, we use k-means to find clusters among different archi-
tects.
3 Data Collection and Sampling
We used a combination of a private collection of photographs and a public collection
of images found via the internet. We chose works of 34 architects, most of whom are
past Pritzker Prize recipients, which are considered to have specific and distinguished
architectural design features. We also included categories of typical residential houses
to allow the trained model to better differentiate between general buildings and works
from renowned architects. Moreover, this sample class enabled us to measure how
easily distinguishable well-known architectural designs are from typical designs.
For more detailed analysis, we classify all images into two categories: outdoor and
indoor. This classification enables us to examine whether the exterior forms are more
significant factors in differentiating architects’ designs or vice versa.
To collect images from the internet, we created several combinations of keywords
relevant to a specific architect (the architect’s name, type of architecture, etc.). After
collecting the raw dataset, we manually cleaned it by eliminating the mislabeled and
unclear images from the samples. We also added photographs of specific works by
some of the architects which were personally taken by the authors of this paper. As
a result, the total number of collected samples is 19,568 (see Table 2for architects
and the corresponding sample sizes). All images were annotated with the ID of the
architect who was responsible for the design.
We implemented our DCNN model using the Google TensorFlow library, and the
algorithms were implemented in Python. The computer had a Linux system (Ubuntu
16.04) with an Intel Core i5 CPU, 16 GB memory, and two parallel GeForce GTX
1070Ti. The training was completed in 8 h.
14 Deep Learning Architect: Classification for Architectural … 257
Tabl e 2 Architects and
sample sizes of collected
photographs
Architect Sample
Size
Architect Sample
Size
Alvar Aalto 460 Oscar Niemeyer 437
Alvaro Siza 1289 Peter Eisenman 331
Bernard Tschumi 288 Rafael Moneo 278
Coop Himmelblau 390 Rem Koolhaas 373
Le Corbusier 527 Renzo Piano 542
Daniel Libeskind 406 Richard Meier 464
Dominique Perrault 234 Richard Rogers 406
E. Souto de Moura 559 SANNA 393
Enric Miralles 518 Shigeru Ban 216
Frank Gehry 669 Steven Holl 498
Frank Lloyd Wright 1177 Tadao Ando 730
Fumihiko Maki 457 Kenzo Tange 454
I.M. Pei 419 Thom Mayne 723
Jean Nouvel 358 To y o I to 672
Louis Kahn 1442 Yoshio Taniguchi 528
Mies van der Rohe 881 Zaha Hadid 635
MVRDV 253 Normal house 1305
Norman Foster 256 Tot a l 19,568
4 Results
This section presents the results of our proposed methodology. First, we examine the
overall model accuracy and comparisons between different architects and between
different types of images. Second, we present the Grad-CAM–generated heat maps,
which were used to analyze the point where the model focused in each picture during
the identification process. Finally, we apply a principal component analysis (PCA)
and k-means to the weighted matrix of the convolutional deep network to find clusters
among architects.
4.1 Model Accuracy
For the DCNN classification task, top-kerror rates are important indicators in eval-
uating the model’s performance. Top-1 accuracy indicates the probability that the
image correctly matches the target label. Conversely, top-5 accuracy represents the
probability that the correct image appears with the target label among five pictures
ordered according to their highest probability. The model was trained in 30 epochs,
and the learning rate was set to 0.1 for the first 20 epochs and 0.01 for the last 10
258 Y. Yoshimura
Tabl e 3 Model accuracy for all categories
Architect Top1
accuracy
Top5
accuracy
Architect Top1
accuracy
Top5
accuracy
A. Aalto 65.07 82.53 O. Niemeyer 72.41 86.20
A. Siza 78.97 97.15 P. E is e nm an 77.50 92.50
B. Tschumi 90.47 92.85 R. Moneo 82.85 91.42
C.
Himmelblau
70.28 82.41 R. Koolhaas 32.60 63.04
L. Corbusier 67.12 82.19 R. Piano 66.23 89.61
D. Libeskind 76.36 85.45 R. Meier 79.03 90.32
D. Perrault 40.62 75.00 R. Rogers 62.50 92.85
E. S. de
Moura
71.62 87.83 SANNA 77.08 87.50
E. Miralles 69.11 86.76 S. Ban 82.14 85.71
F. Ge h r y 80.80 95.95 S. Holl 48.48 69.69
F. Ll o y d
Wright
87.73 98.15 T. Ando 73.23 89.23
F. Maki 68.85 77.04 K. Tange 73.21 83.92
I.M. Pei 65.45 76.36 T. Mayne 77.77 80.00
J. Nouvel 71.15 90.38 T. Ito 56.79 92.59
L. Kahn 87.67 99.05 Y. Taniguchi 72.13 95.08
M. van der
Rohe
84.42 95.90 Z. Hadid 65.51 83.90
MVRDV 60.00 88.57 House 79.78 93.14
N. Foster 77.41 90.32 Tota l 73.17 87.07
epochs. Batch size was set to 16 images. The overall top-1 and top-5 training accuracy
reached 99.7 and 100%, respectively.
Table 3shows the results of computing our model’s top-1 accuracy and top-5
accuracy for the architects. The average of the top-1 accuracy rate on the testing
set is 73.2%, meaning that our model can predict the architect with this probability.
The highest probabilities for top-1 accuracy were attained for Tschumi (90.4%),
Lloyd Wright (87.7%), Kahn (87.6%), van der Rohe (84.4%), and Moneo (82.8%).
Conversely, the lowest probabilities for top-1 accuracy were achieved for Koolhaas
(32.6%), Perrault (40.6%), Holl (48.4%), Ito (56.7%), and MVRDV (60.0%).
We can interpret these results as follows: The computer’s eye tends to be able to
capture design features for the former group, which enables it to distinguish their
architecture from others’, but is likely to detect similar features for the latter group.
Thus, the machine’s eye tends to confuse Koolhaas, Holl, Perrault with other archi-
tects, but it correctly distinguishes Kahn, Siza, and van der Rohe from others. This
tendency does not change if we focus on top-5 accuracy: Kahn (99.0%), Lloyd Wright
(98.1%), Siza (97.1%), van der Rohe (95.9%), and Gehry (95.9%) yielded the highest
14 Deep Learning Architect: Classification for Architectural … 259
Tabl e 4 Accuracy of
different image types Image source Self-taken Images Internet images
Image perspective Indoor Outdoor Indoor Outdoor
Top1 accuracy 70.72 66.19 74.72 73.84
Top5 accuracy 85.24 81.25 89.90 88.13
probabilities, and Koolhaas (63.0%), Holl (69.6%), Perrault (75.0%), Pei (76.3%),
and Maki (77.0%), the lowest ones. On average, almost 70% of architects can be
distinguished with more than 80% probability if we focus on top-5 accuracy, and
45% of architects can be distinguished with more than 90% probability. In Kahn’s
case, this rises to 99.0%.
The result is intriguing because we tend to consider that the characteristics of
Koolhaas’s and Holl’s architecture lie in its unique material usage and form. On the
other hand, Kahn’s, Siza’s, and van der Rohe’s works are known as basic geometry-
based designs, suggesting it would be easier to find more similarities between these
architects. For example, van der Rohe is frequently considered to have established the
design model for the office building, which is a rectangular appearance with multiple
layers surrounded by a glass-curtain wall, and which makes up the landscape of our
contemporary cities.
We also examine whether or not there are significant differences that the com-
puter’s vision captures between the indoor and outdoor images. Our result indicates
that the indoor scenes are much more distinguishable to the machine’s eye than the
outdoor ones (see Table 4). Although there may exist similar objects and features
in the outdoor images (e.g., trees, pavements), it seems that the machine’s eye can
capture the characteristics of the interior spaces better than external design features,
such as the form itself. This preliminary result provides us the possibility and poten-
tiality that the machine’s eye finds the characteristics of modern and contemporary
architecture in the spatial design rather than the mass forms.
4.2 Grad-CAM
Figure 3shows an example of how the machine’s eye works by presenting Grad-
CAM outputs. An exterior photo of Alvaro Siza’s Porto School of Architecture was
fed into the trained model. The prediction of the top four categories is as follows:
Siza, Tschumi, Hadid, and Pei. By using Grad-CAM, we were able to observe the
evidence of the machine’s eye’s focus in each image and the reason the computer
vision made these decisions, with the probability for each choice. In this example,
we can observe that the building form was the main reason for the model’s picking
Siza as its top choice. However, Pei’s designs often have similar geometries; thus,
the model predicted Pei as its fourth choice and highlighted the similar area in the
examined image.
260 Y. Yoshimura
Siza
0.512
Tsch u m i
0.293
Hadid
0.119
Pei
0.102
Fig. 3 Image of Alvaro Siza’s Porto School of Architecture (left) and top four predictions (right)
4.3 Clustering by Principal Component Analysis
Based on the results outlined in Sect. 4.1, a clustering analysis was carried out using
linear principal component analysis (see Fig. 4). We measured the distance to judge
similarities in architectural design between architects. Next, we reduced them using
k-means to find clusters. Finally, we visualized the obtained results and observed four
clusters of architects grouped by the machine. The following is our interpretation of
the results.
The first cluster consists of Norman Forster, Richard Rogers, and Renzo Piano.
They are frequently labeled in terms of “high-tech design” (Kron and Slesin 1984),
which pursues the expression of technology (i.e., structure and facilities) as a design
elements. They tend to borrow established technologies and materials from other
fields (e.g., the automobile or aircraft industry) and apply them to the construction
process. Thus, their approach enables them to push the boundaries of architectural
design. The development of the high-tech style is oriented toward eco-tech design
or sustainable design architecture, which tries to reduce the environmental burdens.
The second cluster consists of Frank Lloyd Wright and “normal house.” Most of
his works are individual homes, although he designed more than 400 built works
(800 if we include the unbuilt works). Wright established the “prairie style” at the
beginning of his career, indicating the emphasis on horizontalness by lowering the
height of the roof and using continuous windows and walls surrounding the building.
This became the model for the middle-class suburban house in the U.S., and it spread
rapidly through the entire country, resulting in the formation of urban and suburban
landscapes.
The third cluster is made up of Frank Gehry and Thom Mayne (Morphosis). Both
are based in Los Angeles, where digital technology and industrial materials pro-
vide their architectural characteristics. To generate the form of a building, they start
from the materials themselves and assemble those materials. Gehry’s architectural
14 Deep Learning Architect: Classification for Architectural … 261
Fig. 4 Linear PCA and clusters
method is “to transform ordinary raw materials–unadorned chain link, sheet metal,
glass, stucco and plywood–into essential formal elements of an intriguing archi-
tecture” (Stern 1993, p. 8), while Mayne overlaps several elements and expresses
incompleteness through his architecture. The fourth cluster consists of Enric Miralles,
Peter Eisenman, and Tadao Ando. Although Miralles was not originally classified
as a “deconstructivist” (Johnson and Wigley 1988), the characteristics of his archi-
tecture can be described as fragmented, inclined roofs and walls, and are seemingly
under construction, which is similar to the architectural characteristics of Eisenman,
who is classified as a “deconstructivist.” Conversely, Ando’s distinguishing features
lie in severe geometric composition, together with exposed concrete and glass as
materials, which seems to create a contrast with the other two architects. However,
the Grad-CAM analysis gave us the insight that the machine’s eye captured curves
and circles as features of Ando’s architecture, thus establishing a similarity with
Miralles and Eisenman.
262 Y. Yoshimura
5 Discussion
This paper discusses the classification of architectural designs using the computer
vision technique. We employed a deep convolutional neural network (DCNN) on a
large-scale sample dataset of 34 architects and their architectural works. Our prelim-
inary result provides an alternative view to the conventional classification method-
ology, i.e., that of architectural historians and theorists. Although it does not replace
conventional classifications, we show that our proposed methodology works rather
as a complement to them and can shed light on unknown aspects of modern and
contemporary architecture. The current study suggests the following contributions
to the classification of an architect’s design:
•Our algorithm enables us to identify an individual architect with 73% validity.
We examined 34 architects from different geographical areas and eras, along with
normal houses. This indicates that the trained neural network correctly captures
the characteristics of an architect’s design and differentiates them.
•The analysis of the model’s acccuracy provides us with the difference between the
machine’s eye’s classification and those of architectural historians and theorists.
While the computer‘s eye can correctly classify Kahn, Siza, and van der Rohe with
high probabilities, it confuses Koolhaas and Holl with other architects. This indi-
cates that the latter architects’ design features cannot be detected by the computer,
which is almost contrary to our intuition. Also, the computer vision’s prediction
is more accurate for indoor scenes than for outdoor ones.
•Our analysis of the Grad-CAM of each architect identifies the design elements
that differentiate architectural works. The visualization of this process enables us
to uncover the significant areas that the machine‘s eye captures for the purpose
of classification. For example, in the case of the Porto School of Architecture by
Alvaro Siza, the trained neural network identified the building‘s form as Siza’s
design feature, resulting in a correct classification, but it also picked up Pei due to
the similar geometries.
•Most of our clustering analysis coincides with the conventional description of
architectural historians and theorists, indicating the validity of our methodology.
The result shows that, for example, Forster, Rogers and Piano are successfully
clustered as high-tech design. We also found that Wright is correctly clustered
with the U.S. suburban house.
The proposed method provides clear value and novel perspectives to the existing
research, but it also has limitations. First, our sample size is small and varies greatly
for each architect. Although the maximum number is more than 1400, the minimum
is only around 200. These imbalanced categories may cause bias in the analysis.
Exploring the adequate sample size and the model’s accuracy is one of the challenges
in the computer vision community, and they largely depend on several factors, such as
the subject of the detection or the model to be applied. Second, the current analysis is
based on NASNet, but not on other models such as AlexNet (Krizhevsky et al. 2012),
VGGNet (Simonyan and Zisserman 2014), ResNet (He et al. 2016), or their variants.
14 Deep Learning Architect: Classification for Architectural … 263
This makes it difficult for us to perform comparative studies and, consequently,
to discern if the obtained results derive from the intrinsic properties of the design
differences or just from the properties of the data or algorithms. Finally, the current
analysis does not consider the temporal factor, indicating that we do not distinguish
an architect’s work by his or her era. A specific architect’s design is not necessarily
consistent during his or her entire professional career; rather, it changes due to new
available technologies or social requirements. For example, Le Corbusier’s earlier
work is significantly different from his later work (e.g., Savoi, Ronchamp). A dataset
considering the temporal factor would provide us with more insights on how some
architects “grow together” or “grow apart” as time goes by.
Considering these limitations, future work should be planned in several directions.
First, we are interested in performing comparative studies between different models.
This would enable us to uncover the models and parameters that are more appropriate
for distinguishing architects’ designs. Second, we are also interested in performing
experiments to explore the number of samples necessary for robust analysis and its
relationship with the model’s accuracy. Finally, we would like to apply the style
transfer algorithm (Gatys et al. 2016) to our dataset. In the field of art, there have
been several attempts to separate the artistic style from the content in a picture
using a neural network algorithm. Its application could provide us more insight into
identifying the design features of each architect and his or her architectural style,
which should be independent of the spatial elements.
The application of a deep convolutional neural network (DCNN) in the context of
architecture and urban planning would allow researchers to analyze visual similari-
ties between types of architecture and create typologies and classifications of their
design features. Although the methodology presented herein gives us preliminary
results rather than complete ones, the method offers an effective means to analyze
visual similarities and extract the features of an architect’s design. In this way, mea-
suring visual similarities using a machine’s eye provides us with insights without
considering any prior knowledge or any other human sensory information, which
can be different from an analysis by a human being. Thus, the current analysis can
complement Kant (1952) and Wolfflin (1950), who analyze the aesthetics of spaces
in terms of perception and discuss the cognitive process of architecture. This is a
piece of critical information that was not obtainable prior to this study.
Acknowledgements The authors thank Cisco, Teck,Dover Corporation, Lab Campus, Anas, SNCF
Gares and Connexions, Brose, Allianz, UBER, Austrian Institute of Technology, Fraunhofer Insti-
tute, Kuwait-MIT Center for Natural Resources, SMART-Singapore-MIT Alliance for Research and
Technology, AMS Institute, Shenzhen, Amsterdam, Victoria State Government and all the members
of the MIT Senseable City Lab Consortium for supporting this research.
264 Y. Yoshimura
References
Abdel-Hamid O, Mohamed AR, Jiang H, Penn G (2012) Applying convolutional neural networks
concepts to hybrid NN-HMM model for speech recognition. In: 2012 IEEE international con-
ference on acoustics, speech and signal processing (ICASSP), Kyoto international conference
center, Kyoto, 25–30 March 2012
Bahdanau D, Cho K, Bengio Y (2015) Neural machine translation by jointly learning to align and
translate. ArXiv Preprint ArXiv:1409.0473
Cai BY, Li X, Seiferling I, Ratti C (2018) Treepedia 2.0: applying deep learning for large-scale
quantification of urban tree cover. In: 2018 IEEE international congress on big data (BigData
Congress), Seattle, 25–30 June 2018
Doersch C, Singh S, Gupta A, Sivic J Efros AA (2012) What makes paris look like paris? In: ACM
transactions on graphics (SIGGRAPH 2012), vol 31(4). ACM Press, New York
Elgammal A, Mazzone M, Liu B, Kim D (2018) The shape of art history in the eyes of the machine.
ArXiv Preprint, ArXiv:1801.07729
Forty A (2000) Words and buildings: a vocabulary of modern architecture. Thames & Hudson, New
Yor k
Frampton K (1992) Modern architecture: A critical history (3rd edn, Revised and Enlarged). London:
Thames and Hudson
Gatys LA, Ecker AS, Bethge M (2016) Image style transfer using convolutional neural networks.
In: 2016 IEEE conference on computer vision and pattern recognition (CVPR), Las Vegas, 26
June to 1 July 2016
Girshick R, Donahue J, Darrell T, Berkeley UC Malik J (2014) Rich feature hierarchies for accurate
object detection and semantic segmentation. In: 2014 IEEE conference on computer vision and
pattern recognition, Columbus, 24–27 June 2014
He K, Zhang X, Ren S, Jian S (2016) Deep residual learning for image recognition. In: 2016 IEEE
conference on computer vision and pattern recognition (CVPR), Las Vegas, 26 June to 1 July
2016
Hitchcock HR, Johnson P (1932) The international style: architecture since 1922. W.W. Norton &
Company, New York
Johnson P, Wigley M (1988) Deconstructivist architecture: the Museum of Modern Art. Museum
of Modern Art, New York
Jolliffe IT (2002) Principal component analysis, 2nd edn. Springer-Verlag, New York
Kant I (1952) The critique of judgment (1790). (trans: Meredith JC). Clarendon Press, Oxford
Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional
neural networks. Advances In Neural Information Processing Systems, pp. 1097–1105
Kron J, Slesin S (1984) High-tech: the industrial style and source book for the home. Clarkson
Potter, New York
Lee S, Maisonneuve N, Crandall D, Efros AA Sivic J (2015) Linking past to present: discovering
style in two centuries of architecture. In: 2015 IEEE international conference on computational
photography (ICCP), Houston, 24–26 April 2015
Li J, Yao L, Hendriks E, Wang JZ (2012) Rhythmic brushstrokes distinguish van gogh from his
contemporaries: findings via automated brushstroke extraction. IEEE Trans Pattern Anal Mach
Intell 34(6):1159–1176
Llamas J, Lerones PM, Medina R, Zalama E, Gómez-García-Bermejo J (2017) Classification of
architectural heritage images using deep learning techniques. Appl Sci 7(10):992
Obeso AM, Vázquez GMS, Acosta AAR, Benois-Pineau J (2017) Connoisseur: classification of
styles of Mexican architectural heritage with deep learning and visual attention prediction. In:
15th international workshop on content-based multimedia indexing (CBMI), Florence, 19–21
June 2017
Onians J (1988) Bearers of meaning: the classical orders in antiquity, the middle ages, and the
renaissance. Princeton University Press, Princeton
14 Deep Learning Architect: Classification for Architectural … 265
Saleh B, Abe K, Arora RS, Elgammal A (2016) Toward automated discovery of artistic influence.
Multimed Tools Appl 75(7):3565–3591
Selvaraju RR, Cogswell M, Das A, Vedantam R, Parikh D, Batra D (2017) Grad-CAM: visual
explanations from deep networks via gradient-based localization. In: 2017 IEEE international
conference on computer vision (ICCV), Venice, 22–29 October 2017
Shalunts G, Haxhimusa Y, Sablatnig R (2011) Architectural style classification of building facade
windows. In: Bebis G, Boyle R, Parvin B, Koracin D, Fowlkes C, Wang S, Choi M-H, Mantler
S, Schulze J, Acevedo D, Mueller K, Michael P (eds) Advances in visual computing, ISVC
2011. Lecture Notes in Computer Science, vol 6939. Springer, Berlin, Heidelberg, New York, pp
280–289
Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recog-
nition. ArXiv preprint ArXiv:1409.1556
Stern RAM (1993) Frank O. Gehry: architecture with a serious smile. In: Futagawa Y (ed) Frank
O. Gehry. GA Architect 10. A.D.A. Edita, Tokyo, pp 8–9
Wolfflin H (1950) Principles of art history: the problem of the development of style in later art
(1915). (7th Edition Trans. Hottinger MD) Dover, New York
Zhang F, Zhou B, Liu L, Liu Y, Fung HH, Lin H, Ratti C (2018) Measuring human perceptions of
a large-scale urban region using machine learning. Landsc Urban Plan 180:148–160
Zhou B, Lapedriza A, Xiao J, Torralba A, Oliva A (2014) Learning deep features for scene recogni-
tion using places database. In: Ghahramani Z, Welling M, Cortes C, Lawrence ND, Weinberger
KQ (eds) Advances in neural information processing systems 27 (NIPS 2014), Curran Associates,
Inc, pp 487–495
Zoph B, Shlens J (2018) Learning transferable architectures for scalable image recognition. ArXiv
Preprint ArXiv:1707.07012
Chapter 15
An Immersive 3D Virtual Environment
to Support Collaborative Learning
and Teaching
Aida Afrooz, Lan Ding and Christopher Pettit
Abstract This paper reflects on a Virtual Learning Environment (VLE) in the
context of architecture, urban planning and design. The paper aims to critically
assess the ability of virtual environments to support experiential online learning. It
contributes to the literature by providing the experimental results of implementing
the TERF virtual world for undergraduate and graduate Built Environment courses.
TERF virtual world provides videos and other communication tools to support col-
laboration among students. Feedback on the usage and functionality of this 3D virtual
platform was collected from students through post evaluation surveys. This experi-
ment provided opportunities to facilitate team communication and a route to more
collaborative leaning. It is also discussed the strengths and limitations of the 3D
collaborative virtual environment to support deeper learning environment.
Keywords 3D virtual environment ·Collaborative learning ·TERF immersive 3D
virtual world
1 Introduction
By increasing the class sizes and rising the number of students, traditional learning
approaches are stretched in supporting active learning (Cohn 2016). It is increasingly
important to access learning opportunities that support and extend traditional lectures
A. Afrooz (B)·C. Pettit
School of Architecture and Built Environment, Faculty of Science Engineering & Built
Environment, Deakin University, 1 Gheringhap Street, Locked Bag 20001, Geelong, VIC 3220,
Australia
e-mail: aida.afrooz@deakin.edu.au
C. Pettit
e-mail: c.pettit@unsw.edu.au
L. Ding
Faculty of Built Environment, University of New South Wales, Sydney, Australia
e-mail: lan.ding@unsw.edu.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_15
267
268 A. Afrooz et al.
to a greater collaboration class environment. It is also important to maximize deep
learning opportunities. On the one hand, in architecture, urban planning and design
which involve actions and plans, it is the planner’s ability to clearly communicate
with professionals and residents and convey ideas to the clients. Therefore, a planning
curriculum should ensure developing such communication and presentation skills
for its graduates to be competent in their professional practice. Hence, information
and communication technologies (ICT) are being used in the curricula of many
undergraduate and masters’ degree such as architecture and urban design (Fonseca
et al. 2017a).
On the other hand, the benefits of 3D models and virtual reality for education have
been explored extensively (Yin 2010). For instance, Kalyuga (2007) found that virtual
worlds are highly interactive and can provide dynamic learning environment. In the
context of urban planning and architecture, virtual worlds can contribute to better
understanding of the context by providing abilities to ‘build and experience’ buildings
and cities, and to support virtual interactions and collaborations. Virtual worlds can
provide other benefits such as enhancing creativity and students’ motivation and
encouragement, provide opportunities for social interactions and communications
(Soukup 2004), facilitating collaboration, and reducing social anxiety and stress
(Pioggia et al. 2010). However, the use of 3D immersive environments for teaching
built environment courses is in its infancy. This research aims to contribute to this
important area through the evaluation of a 3D immersive environment known as
TERF specifically for City Planning and Design.
There are some online learning platforms to support built environment courses
such as Moodle, and blackboard available at some universities such as UNSW Syd-
ney. Course convenors can upload the lectures and create links to assignments’
submissions where students can submit their assignments and download recorded
lectures and PowerPoint slides. However, such online learning platforms have limi-
tations in supporting digital geographical artefacts display, ability to engage students
and support collaboration.
The authors believe that collaboration is vitally important within any learning
environment, in particular for the field of City Planning and Design. In turn, the
authors anticipate that a 3D Virtual Learning Environment can promote virtual col-
laboration as a pedagogical tool. Accordingly, the primary purpose of this paper
is to examine the strength of a specific 3D Virtual Learning Environment- named
“TERF” developed by 3D Immersive Collaboration (ICC) (3D Immersive Collabo-
ration Consulting 2011)—to support online learning and enhance the understanding
of students’ learning. This paper will provide insights on virtual collaboration for
a better understanding on 3D virtual learning environment’s requirements for deep
learning. The TERF 3D immersive virtual environment has been utilised in this study
because of its unique collaboration functionality (Zhang et al. 2016).
An experiment was conducted for the undergraduate and graduate Built Environ-
ment courses; in city planning and design, using a project-based learning approach
where students worked on a project during the course to solve a real-world problem,
question or challenge (NSW Department of Education 2018).
15 An Immersive 3D Virtual Environment to Support … 269
Both courses had some restrictions that are discussed later in the methods section.
To balance such restrictions on collaborative learning and limitations in experiencing
3D precinct and building models, the course convenors implemented some changes in
their course outlines to use a more immersive approach by applying a 3D collaborative
learning environment. The changes in the course outlines were to get use of the TERF
immersive collaborative virtual environment to benefit from group interactions in
order to enrich the learning process.
2 Literature Review
This section provides innovative exemplars where 3D virtual environments have suc-
cessfully been used as learning and teaching platforms. These exemplars are reviewed
in the following two areas: (a) collaborative learning, and (b) 3D collaborative virtual
learning environments.
2.1 Collaborative Learning
Collaborative learning is defined as a “student-centred approach” where students
work in a group on a specific task (Lee 2009). The Virtual European Schools project
is an early example of merging the collaborative tools and 3D models for learning
(Bouras et al. 1999). They simulate a classroom using 3D graphics. Students were
able to navigate within the environment simultaneously and communicate with each
other by sending short messages.
Another example of collaborative learning environment is CLEV-R (Collaborative
Learning Environment with Virtual Reality), a web-based, multi-user 3D virtual
environment (Monahan et al. 2008). CLEV-R offers more interactions for students
and tutors. Students can upload their files onto the designated boards within the
virtual environment. The environment is like a real university with different classes
and meeting rooms and enables social interaction between students in informal areas.
In another study, Hai-Jew (2011) described 2 studies using CLEVR interface
assessing the usability of this program for e-learning and determining the factors
affecting users’ performances in the Virtual Reality environment. Using a self-
evaluation questionnaire, he reported that 95% of participants mentioned that they
could easily follow the lecture. He also found that some factors such as age and
previous experience in playing virtual reality games influence students’ success.
Taken together from the above-mentioned examples, the educational virtual envi-
ronments found to be very appealing for both students and lecturers. The evaluations
of such virtual environments had positive feedbacks (95% satisfaction in the study
done by Hai-Jew) in delivering learning materials (e.g. Hai-Jew 2011; Bouras and
Tsiatsos 2006). This literature review has revealed that there is a paucity of appli-
270 A. Afrooz et al.
cations of virtual collaborative immersive platforms for teaching 3D modelling and
collaboration, particularly in the field of city planning and design.
2.2 Collaborative Virtual Learning Environments
3D collaborative virtual environments are platforms where avatars can be used to
represent the users’ real presence (DeNoyelles and Seo 2012). There are a number
of 3D virtual environments available which were first developed for gaming and are
currently being used for educational purposes. These include: SecondLife, Active
Worlds, TERF, Open Simulator, and Adobe Atmosphere (Wang et al. 2012). Such
platforms are called 3D virtual learning environments (Zuiker 2012). Students are
able to communicate and interact with peers in 3D virtual worlds using live voice
and video, text chat, presentation tools, etc. available in the platform (Dickey 2005).
3D collaborative virtual learning environments simulate the real world and allow
users to navigate, interact and communicate via avatars (Wang et al. 2017) during
the learning and teaching process. Emotional skills such as presence, satisfaction,
and enjoyment and communication skills including engagement, and language learn-
ing were the most frequent achievements in using 3D collaborative virtual learning
environments (Reiso˘glu et al. 2017).
A Google Scholar search of most recent (i.e. since 2014) academic articles on
“3D collaborative virtual environment for learning and teaching” yielded to 17,400
academic papers which mainly focused on education and distant learning and ranged
from psychology (e.g. self-esteem, motivation) and health to geography and archi-
tecture (conducted November 2018). A cursory refined search to specific fields of
Architecture and City Planning and Design yielded to only three that are more akin to
the scenario of the present study and are described below. This showed the widespread
use of 3D virtual environments in education but less published articles in the field
of architecture and city planning and design. However, in 2013, Freita and Ruschel
(2013) reviewed 200 papers on virtual environments applied to architecture but focus-
ing on research areas and technological development stages rather than teaching and
learning.
Fonseca et al. (2017a) argued that while ICTs have revolutionized the society,
education has failed to accept many changes. The authors investigated the degree
of which students as “digital native” (Bennett et al. 2008) can adapt to high density
of technological contents in educational environment. They found that the stress
generated from using high density of technological contents- known as technological
stress- resulted in loss of motivation. Secondly, they found that new and advanced
visualization technologies such as complex 3D modelling can improve students’
motivation and satisfaction.
In another study, Fonseca et al. (2017b) aimed to incorporate gaming strategies
in an urban collaborative environment. They analyzed the impact of visualization
systems as educational tools in Architecture. The aim of their project was to recreate
urban areas in the City of Barcelona by allowing students and citizens interact with the
15 An Immersive 3D Virtual Environment to Support … 271
environment. They found that although the initial motivation was considered “low”
and “medium”, the students’ motivation increased significantly after the completion
of the learning task.
Bower et al. (2017) undertook an experiment among undergraduate students in
Macquarie university, Australia using a blended reality collaborative environment of
students who were in the face-to-face classroom students with other students who
were using avatar in 3D virtual world and participated remotely. Disregarding tech-
nical issue, they reported that blended learning can support effective communication
and collaboration. They found that blended reality collaborative learning environ-
ments enhance engagement by providing students with an embodied presence in
face to face classes. The authors suggested a number of pedagogical, technological
and logistic factors that support and constrain learning. For instance, they note that
working in a group with face-to-face peers, being able to see the student avatar and
his/her name, are factors that can support face-to-face learning. Moreover, factors
such as technical issues, repetition of instruction, lack of opportunities to interact
with peers are examples of factors that can restrict face-to-face learning and remote
learning. Such factors were investigated in this paper and explained in the discussion
section in the context of City Planning and design.
Cho and Lim (2017) investigated the collaborative problem solving and collab-
orative observation using a teacher-controlled avatar in virtual worlds among sec-
ondary school students. To investigate the collaborative problem solving, students
were worked in groups to solve topographical problems. In investigating the collab-
orative observation condition, students were observed teacher-controlled avatar and
were discussed to solve problems. Cho and Lim (2017) found one of the implica-
tions of virtual worlds in problem solving and learning in the field of geography. They
suggested that collaborative action with peers can enhance learning in virtual world.
They also found that collaboration skills can be developed through collaborative
virtual environments.
Many tools have been developed in addressing collaborative learning environ-
ment. TERF has been recently used in collaboration-based projects and plans in
support of global teamwork. Fruchter (2014) utilized the TERF as a virtual office
to facilitate global collaboration in the AEC Global Teamwork course at Stanford.
developed a tool based on TERF named Urban Redesign TERF (UR Terf)—a digi-
tal toolkit for urban engagement- with the aim of increasing designers’ freedom in
design process for public participation. They indicated some advantages of using
immersive environments as being suitable for: younger generation, experts and non-
experts, collaboration and public participation, and stakeholders.
In summary, the potential benefits of learning in 3D collaborative virtual envi-
ronments for tertiary students are that they will be able to interact with each other
in the 3D virtual environment, which enhances the students’ learning experience,
motivations, and satisfaction.
272 A. Afrooz et al.
3 Methods
3.1 Course Context
In the undergraduate course (i.e. Urban Modelling), the experiment required stu-
dents producing 3D precinct models (i.e. a city block), displaying the sustainability
information of the precinct and presenting the work for discussions in the 3D vir-
tual environment as part of their assessment. Previously, the course was conducted
using standard 3D modelling and GIS software packages. Urban Modelling course
was conducted through lectures and laboratory practical sessions using 3D software
available in the market (e.g. ESRI CityEngine) to model buildings and precincts as
well as conducting solar access analysis. However, the main challenge for the stu-
dents was to work with other students in group projects. Some of the students were
working part-time, some were living far from the university campus and were unable
to commute frequently to the university campus for team meetings and gatherings.
This issue made it difficult for students to conduct an effective group discussions and
team work. Another problem in running the Urban Modelling course traditionally
was that students were not able to discuss the 3D precinct models which their peers
and potential improvements and changes of models in real-time in a 3D virtual envi-
ronment. This made it very difficult for students to collaborate, engage and comment
on each other’s models, hence, could potentially affect the quality of students’ group
work. Accordingly, TERF was selected for this course to examine if it can provide
collaborative learning environment.
The assignment for Digital Cities Graduate course was using about visualisation
tools and techniques, exploring and communicating the ‘Smart City’ through a virtual
collaborative environment.
Similarly, students who enrolled in the postgraduate course (i.e. Digital Cities)
had difficulty in collaborating and fulfilling group assignments for similar reasons
outlined above. This course was conducted in an intensive mode where students
attended lectures, discussions and laboratory sessions for a full two days every month.
Therefore, students had less chance to collaborate effectively. Similar to the under-
graduate course, the second issue in running the course traditionally was the inability
to experience the 3D model, previously students had lectures and demonstrations on
3D modelling but no hands-on exercises. Students were asked to critically explore
social media and digital technologies, review and create quality of life indicators,
and finally, develop a business case for implementing a planning support system.
The immersive 3D modelling assessment was designed to teach students how to
build basic 3D models and communicate these models in an immersive multi-user
environment. This was possible by using the TERF platform.
15 An Immersive 3D Virtual Environment to Support … 273
3.2 Procedure
An experiment was conducted in two phases prior and after using TERF immer-
sive 3D virtual platform which consisted of navigating in a virtual environment and
presenting the assignments. This platform was used in the Built Environment under-
graduate course (i.e. Urban Modelling) and Built Environment graduate course (i.e.
Digital Cities).
Students in Urban Modelling course were asked to model a small city block in
a selected suburban area consisting of a number of city elements such as: buildings
with texture, open spaces, streets and city furniture. They modelled a precinct using
SketchUp then imported their models into the TERF platform for virtual experience
of potential problems and opportunities and further discussions on urban interven-
tion options. The TERF immersive 3D virtual environment consisted of both a shared
reception space for all students and individual virtual worlds for students to demon-
strate their precinct models. Students can enter individual virtual worlds, ‘walked’ or
‘flew’ through each other’s models, experienced live changes of their models, shared
sustainability information of buildings and precincts, exchanges issues identified and
opportunities for improvements. The live voice, video, web camera, text chat, tabbed
walls, and virtual presenting tools in TERF were used to support real-time commu-
nication between student teams and tutors. Accordingly, students and tutors were
presented by avatars and communicated with each other in the immersive 3D virtual
environment in the process of developing an assignment project. The feedback from
the students on TERF platform was collected through an online survey before and
after the experiment.
Each student was allocated an individual virtual world to present his/her work
(Fig. 1). Such virtual worlds were provided by the TERF technical team for the
purpose of this course. While students were able to enter each individual virtual
world, there was a reception area where students and tutors could meet and wait
for the presentations to begin (Fig. 2). During the assessment, tutors and students
watched and listened to a single student’s presentation and discussed and provided
comments and feedback to the student in real-time in the virtual world (Fig. 3).
As for the Digital Cities course, students were instructed to explore and com-
municate the ‘Smart City’ through a virtual collaborative environment. They were
investigating how virtual worlds tools and techniques may assist in city planning.
To do so, they were asked to prepare a digital presentation with a number of digi-
tal artefacts such as PowerPoint presentation, images and so on, and to present the
assignment using the TERF by utilising as many features as appropriate within the
TERF environment. Similar to Urban Modelling assignment, TERF was used for dis-
cussion of students’ assignments, learning from peers, and presentation and marking
the assignment.
274 A. Afrooz et al.
Fig. 1 The shared virtual reception space with doors for access to students’ individual virtual
worlds where their 3D precinct models were demonstrated
Fig. 2 Students and tutors discussed the assignment project using live voice, web camera, text chat,
etc.
15 An Immersive 3D Virtual Environment to Support … 275
Fig. 3 Assessments of the assignments. During the assessments one student used the “lead meeting”
function available in TERF while other students and tutors were attending the presentation
3.3 Participants
Students in both courses (i.e. Urban Modeling and Digital Cities) were asked to vol-
unteer to participate in an online questionnaire about their experiences of using TERF
immersive 3D virtual world. Call for participation was posted through Moodle– an
open-source learning management system to support blended learning throughout
UNSW—to students. Forty students from both courses were participated in this sur-
vey before the experiment (pre-questionnaire survey) and at the end of the experiment
(post-questionnaire).
3.4 Questionnaire Survey
The questionnaire aimed to identify the strength and weaknesses of TERF in
teaching and learning. There were two surveys named pre-questionnaire and post-
questionnaire. The questions were structured into three sections: (a) the 3D-related
skills’ questions including participants’ skill levels in utilizing the presentation soft-
ware such as power point and audio/video file preparation as well as navigation in
3D virtual environment (pre-questionnaire); (b) effectiveness, accuracy, and usability
of TERF (post-questionnaire survey). These questions had multiple objectives such
as understanding how TERF was used in preparing their assignment, the accuracy
and completeness of goals, and the usability and success of using TERF; and, (c)
speculative open-ended questions around expectations and improvements for TERF.
These questions were included in both pre-questionnaire and post-questionnaire sur-
276 A. Afrooz et al.
veys to assess the satisfaction level of using the 3D virtual environment by providing
comments and feedbacks.
4 Results
4.1 Analysis of Pre-questionnaire
Likert questions covered the aspects related to different levels of 3D modelling skills
and presentation skills of students. Figure 4illustrates the percentage of the responses
to each skill described by a 5-point Likert scale where 5 is the highest rating. Although
the majority of the respondents developed skills in using word documentations,
PowerPoint presentation, excel spreadsheets, and video/audio files (80, 72.5, 47.5,
and 47.5% rated above 4, respectively), they developed less skills in interacting with
others in VR, navigating in a 3D environment, and using 3D models, and they were
not very familiar to 3D virtual worlds such as TERF and Second Life (77.5, 65, 73,
and 77.5% of students rated below 3, respectively).
12345
Fig. 4 Percentage of students’ responses to different categories of presentation and 3D-related
skills
15 An Immersive 3D Virtual Environment to Support … 277
Students provided some comments for the potential benefits of using a virtual
world such as TERF for city planning, urban design, and urban modelling. A selection
of students’ comments is presented below.
Respondent 1. “Virtual Worlds or other 3D visualisation tools have the potential to provide
planners,et al. with the real-feel of how it would be to predict, forecast or identify areas
in cities that need solutions or improvement. This encourages and influences policy-makers
towards development or creation of cities not just in a faster way, but smarter as well.”
Respondent 6. “I can see that TERF could be used to bring stakeholders together for a
project when collaboration and information sharing is necessary.”
Students considered using a virtual world such as TERF could reduce costs for
providing a visually stimulating model (physical vs. digital models). It can be utilized
as a collaboration tool, to support communication between individuals, architects and
urban designers, especially in larger groups who are in different cities. In addition,
it can support a more efficient way of designing and communicating between the
designer and their clients. It allows a user to gain a better understanding of a model
and its surroundings, especially when one can navigate around the model and make
changes through real-time collaborations. Respondents believed that virtual worlds
allow sharing large information in real-time by being immersed in a full multi-sensory
experience. Finally, the students commented that using TERF could potentially let
planners to predict and improve the built environment by getting a real-feeling of
how a plan would change and improve the existing conditions.
4.2 Analysis of Post-questionnaire
Students who participated in the survey have rated their learning experiences in using
the TERF. Table 1provides the responses to the post-questionnaire for both Urban
Modelling and Digital Cities courses. The average score is based on the frequency of
responses with weights ranging from 5 (highest rate) to 1 (lowest rate). Any rating
over 4 is considered as above average, ratings between 2 and 4 are considered average,
any rating below 2.0 is considered as problematic. Applying these weighted scores
to the students’ ratings demonstrate that students rated their learning experiences in
using TERF 3D virtual environment as average.
Tabl e 1 Average Likert
score for student ratings of the
TERF 3D virtual environment
of urban modelling and
digital cities courses offering
in semester 2 2016 (standard
deviation in parentheses)
Learning experiences
Preparing the
assignment
Score Presenting the
assignment
Score
Usability 2.5 (±1.3) Usability 2.5 (±1.2)
Effectiveness 2.5 (±1.2) Effectiveness 2.8 (±1.3)
Efficiency 2.7 (±1.1) Efficiency 2.6 (±1.3)
Satisfaction 2.5 (±1.1) Satisfaction 2.5 (±1.3)
278 A. Afrooz et al.
In terms of usability of TERF in preparing and presenting assignments, students
rated average (Table 1). As for effectiveness of the TERF 3D virtual world the stu-
dents ranked the accuracy and completeness in achieving their specified goals for
both preparing and presenting the assignments as average. They noted real-time com-
munication concept, effective feature to navigate around models, and the excitement
in familiarising with the software as items they enjoyed working with TERF. How-
ever, the interface was not very friendly and easy to use. In addition, the restricted file
sizes for videos and models had dropped their ratings. In contrary to preparing the
assignment, students found TERF very effective in presenting their work. They noted
opportunities to meet and communicate as a very effective component in present-
ing their works. They enjoyed walking around the model in real-time which created
unique experiences. However, some technical issues such as the time to import large
3D building and precinct models into TERF negatively affected the ratings.
Students rated the efficiency of TERF in preparing and presenting the assignment
as 2.7 and 2.6, respectively. They remarked the high capabilities of the software,
acknowledged the capacity of TERF for a greater use than just presentation, and
ease of collaboration in a large group. Nevertheless, preparing the work in TERF
was time consuming for them because of difficulty in understanding how TERF
works and slow operation of the software during presentations. This identified some
issues in the step-by-step instructions which could be edited and updated.
Students rated their satisfaction of using TERF in terms of comfort and accept-
ability of the work system to its users as average (i.e. 2.5) for both presenting and
preparing the assignments. Some considered TERF as a fast and easy to use software
while others mentioned that the interface was not intuitive and loading times took
too long. They were agreed that TERF was worthy for presentation as it engaged
the audience effectively although some preferred the traditional means of presenta-
tion. Even when they were not sure about the efficiency of TERF, they mentioned
opportunities for use in future. Some emphasized the satisfaction of using TERF
after overcoming the abovementioned technical issues and challenges.
In summary, these empirical findings suggest the possibility and potential of using
TERF 3D virtual world technology for students in classroom-based courses that
require team collaboration and communication in assignments.
5 Discussion and Conclusion
This paper presents the design of two courses in City Design and City Planning with
essential use of advance collaboration approaches. Both courses had a duration of
one semester. The TERF 3D virtual world is chosen to examine students’ engage-
ment and deep learning in the two above-mentioned planning curricula to support
collaborative learning environment. The live lectures and collaborative discussions
and assessments were conducted in the TERF immersive 3D virtual environment. A
survey questionnaire was conducted before and after the experiment. Results suggest
that a 3D virtual learning environment such as TERF enables students to explore col-
15 An Immersive 3D Virtual Environment to Support … 279
laborative city planning and design and provide an interactive virtual environment
for students and tutors to work together, forging through a deeper understanding of
the project work and social interaction among students.
Despite some shortcomings of using TERF in Urban Modelling and Digital Cities
courses, TERF performed well in engaging students as a pedagogical tool in urban
planning and architecture fields. The results show that 3D virtual learning envi-
ronment promotes collaborative learning and teaching. Furthermore, TERF virtual
worlds provided innovative learning opportunities and platforms for the students:
•Individual 3D virtual worlds enabling each student to demonstrate their project
work and collect feedback from peers in real-time;
•Virtual experience of student project works by both tutors and students creating
a collaborative learning and teaching environment and a new way for assessing
student project works;
•Data interoperability techniques enabling to import 3D building and/or precinct
models into virtual worlds;
•Gaming interaction and meeting control functions enabling to navigate through
the imported 3D building and/or precinct models and conduct real-time commu-
nications and collaborations;
•Live voice and video, web camera, text chat, virtual presentation tools, etc. prov-
ing innovative ways for project presentations and communications with potential
stakeholders in the city planning and design process.
The 3D immersive collaborative virtual environments enabled the students to
meet and discuss in a dynamic and non-linear environment from viewing the 3D
models into experiencing and discussing the precinct environment by navigating and
interacting with it. This reveals that TERF can be useful in supporting interactive
and collaborative city planning and design and associated deep learning outcomes.
Despite the many advantages, TERF had some limitations and drawbacks such as
more amount of time which is required to prepare and produce the assignment. Other
limitations were related to the size of a 3D model, and it took long for projects to be
loaded. Central challenges for instructors and students were the slow performance
and low speed of loading projects for assessment. All the factors that have been
identified from the experiment was classified into the pedagogical and technological
and are presented in Table 2.
Accordingly, the pedagogical effectiveness of TERF, results in a number of rec-
ommendations informing future development of the software: some implementation
issues, particularly bugs related to importing models, presentations, and videos can
be addressed; the limit of file size can be increased; the user interface can avoid ambi-
guity and become more user-friendly; the step-by-step instruction should simply be
updated to enhance efficiency of the software.
The obtained results suggest that the improvement of technical issues in 3D virtual
environments will facilitate the future application of such platform in collaborative
learning and teaching. An immersive 3D virtual environment has the potential to
provide digital education innovation to support future collaborative city planning
and design.
280 A. Afrooz et al.
Tabl e 2 Factors supporting and restricting 3D collaborative learning environment in 3D TERF
immersive environment
Factors supporting collaborative
learning
Factors restricting collaborative
learning
Pedagogical • Active learning (learning by doing)
• Communicating with a group,
particularly large groups
• Effective in presenting work
• Assessing assignments
• Real-time communication
• Students had fun and were excited
while working with TERF
• End product was very satisfying
• Students found it very efficient
when moving through the 3D
models and present it
simultaneously
• Engage audience effectively
• A useful collaboration tool
• Ambiguity in limited parts of the
step-by-step instruction
Technological • Being able to see avatar names
• Being able to share the screen
• Help was easily provided
• Easytonavigateinthevirtual
environment
• Using whiteboards
• Being able to record voice and
video
•Textchat
• Web camera
• Limited size for videos and models
• Students had difficulty in using
TERF interface
• Issues in importing models, videos,
and presentations
• Slow performance and low-speed
of loading projects
• Compatibility issues with texture
• Difficulty in positioning and
rotating models
• Crashes during importing
Acknowledgements The authors would like to thank 3D immersive Collaboration Consulting for
providing free licenses for the TERF platform and technical support and training so that the authors
could conduct this evaluation.
References
3D Immersive Collaboration Consulting (2011) TERF. Available online at http://www.3dicc.com/.
Accessed on 27 Nov 2017
Bennett S, Maton K, Kervin L (2008) The ‘digital natives’ debate. A critical review of the evidence.
Br J Educ Technol 39(5):775–786. https://doi.org/10.1111/j.1467-8535.2007.00793.x
Bouras C, Tsiatsos T (2006) Educational virtual environments. Design rationale and architecture.
In: Multimedia tools and applications. Kluwer Academic Publishers, vol 29(2), pp 153–173
Bouras C, Fotakis D, Kapoulas V, Koubek A, Mayer H, Rehatschek H (1999) Virtual European
school-VES. In: Proceedings IEEE international conference on multimedia computing and sys-
tems. ICMCS99: IEEE multimedia systems ‘99: international conference on multimedia com-
puting and systems. Florence, Italy, 7–11 June 1999: IEEE Computer Society, pp 1055–1057
15 An Immersive 3D Virtual Environment to Support … 281
Bower M, Lee MJW, Dalgarno B (2017) Collaborative learning across physical and virtual worlds.
Factors supporting and constraining learners in a blended reality environment. Br J Educ Technol
48(2):407–430. https://doi.org/10.1111/bjet.12435
Cho YH, Lim KTY (2017) Effectiveness of collaborative learning with 3D virtual worlds. Br J Edu
Technol 48(1):202–211. https://doi.org/10.1111/bjet.12356
Cohn D (2016) Active learning. In: Sammut C (ed) Encyclopedia of machine learning and data min-
ing. Springer, Berlin, New York, Heidelberg, pp 9–14. Available online at: https://link.springer.
com/content/pdf/10.1007/978-1-4899-7687-1_916.pdf
DeNoyelles A, Seo KKJ (2012) Inspiring equal contribution and opportunity in a 3d multi-user
virtual environment. Bringing together men gamers and women non-gamers in Second Life®.
Comput Educ 58(1):21–29. https://doi.org/10.1016/j.compedu.2011.07.007
Dickey MD (2005) Three-dimensional virtual worlds and distance learning. Two case studies of
Active Worlds as a medium for distance education. Br J Educ Technol 36(3):439–451. https://
doi.org/10.1111/j.1467-8535.2005.00477.x
Fonseca D, Redondo E, Valls F, Villagrasa S (2017a) Technological adaptation of the student to
the educational density of the course. A case study. 3D architectural visualization. Comput Hum
Behav 72:599–611. https://doi.org/10.1016/j.chb.2016.05.048
Fonseca D, Villagrasa S, Navarro I, Redondo E, Valls F, Llorca J et al (2017b) Student motivation
assessment using and learning virtual and gamified urban environments. In: Dodero JM, Sáiz MSI,
Rube IR (eds) The 5th international conference proceedings of the 5th international conference
on technological ecosystems for enhancing multiculturality—TEEM 2017. Spain. ACM Press,
New York, USA, pp 1–7
Freita MR, de Ruschel RC (2013) What is happening to virtual and augmentes reality applied to
architecture? In: Stouffs R, Janssen P, Roudavski S, Tunçer B (eds) Proceedings of the 18th
international conference on computer-aided architectural design research in Asia (CAADRIA
2013). The Association for Computer-Aided Architectural Design Research in Asia (CAADRIA),
Hong Kong
Fruchter R (2014) Transformative 3D immersive collaboration environment in support of AEC
global teamwork. In: Issa RI, Flood I (eds) Computing in civil and building engineering (2014).
2014 international conference on computing in civil and building engineering. Orlando, Florida,
United States, 23–25 June 2014. American Society of Civil Engineers, Reston, pp 1425–1432
Hai-Jew S (ed) (2011) Virtual immersive and 3D learning spaces: IGI global
Kalyuga S (2007) Enhancing instructional efficiency of interactiveE-learning environments. A Cogn
Load Perspect Educ Psychol Rev 19(3):387–399. https://doi.org/10.1007/s10648-007-9051-6
Lee M (2009) How can 3d virtual worlds be used to support collaborative learning? An analysis of
cases from the literature. J e-Learn Knowl Soc 5(1):149–158
Monahan T, McArdle G, Bertolotto M (2008) Virtual reality for collaborative e-learning. Comput
Educ 50(4):1339–1353. https://doi.org/10.1016/j.compedu.2006.12.008
NSW Department of Education (2018) Learning for the future. NSW Department of Education.
Australia. Available online at https://education.nsw.gov.au/teaching-and-learning/curriculum/
learning-for-the-future/future-focused-learning-and-teaching/project-based-learning-resource-
guide/introducing-project-based-learning, checked on 10/10/2018
Pioggia G, Carbonaro N, Anania G, Tognetti A, Tartarisco G, Ferro M et al (2010) Interreality.
The use of advanced technologies in the assessment and treatment of psychological stress. In:
2010 10th international conference on intelligent systems design and applications (ISDA). Cairo,
Egypt, 29/11/2010–1/12/2010: IEEE, pp 1047–1051
Reiso˘glu I, Topu B, Yılmaz R, Karaku¸s Yılmaz T, Gökta¸s Y (2017) 3D virtual learning environ-
ments in education. A meta-review. Asia Pacific Educ Rev 18(1):81–100. https://doi.org/10.1007/
s12564-016-9467-0
Soukup C (2004) Multimedia performance in a computer-mediated community. Communication as
a virtual drama. J Comput-Mediated Commun 9(4):0. https://doi.org/10.1111/j.1083-6101.2004.
tb00294.x
282 A. Afrooz et al.
Wang CX, Calandra B, Hibbard ST, McDowell Lefaiver ML (2012) Learning effects of an exper-
imental EFL program in second life. Educat Tech Res Dev 60(5):943–961. https://doi.org/10.
1007/s11423-012-9259-0
Wang X, Laffey J, Xing W, Galyen K, Stichter J (2017) Fostering verbal and non-verbal social
interactions in a 3D collaborative virtual learning environment.A case study of youth with Autism
spectrum disorders learning social competence in iSocial. Educat TechRes Dev 65(4):1015–1039.
https://doi.org/10.1007/s11423-017-9512-7
Yin L (2010) Integrating 3D visualization and GIS in planning education. J Geogr High Educ
34(3):419–438. https://doi.org/10.1080/03098260903556030
Zhang B, Benton S, Pearson W, LeMoine J, Herbertson N, Williams H, Goodman L (2016) Playing
3D. Digital technologies and novel 3d virtual environments to support the needs of Chinese
learners in western education: cross-cultural collaboration, gamification, well-being and social
inclusion. In: 2016 22nd international conference on virtual system & multimedia (VSMM).
Kuala Lumpur, Malaysia, 17/10/2016–21/10/2016: IEEE, pp 1–9
Zuiker SJ (2012) Educational virtual environments as a lens for understanding both precise repeata-
bility and specific variation in learning ecologies. Br J Edu Technol 43(6):981–992. https://doi.
org/10.1111/j.1467-8535.2011.01266.x
Chapter 16
Spatiotemporal Information System
Using Mixed Reality for Area-Based
Learning and Sightseeing
Ryuhei Makino and Kayoko Yamamoto
Abstract The study aims to develop a system that visualizes spatiotemporal
information in both real and virtual spaces, integrating SNS, Web-GIS, MR and
the gallery system as well as Wikitude, and connecting external social media. Using
these systems and technologies, the system has four functions for area-based learn-
ing, three functions for sightseeing, four functions specialized for VR, AR and MR.
The system was applied for five weeks, and the total number of users was 66. From
the evaluation results, it was clear that all of the functions in the system were highly
evaluated, and most of the functions for area-based learning are more popular than
other two kinds of functions. Though all users used the functions specialized for VR,
AR and MR, they were more negatively evaluated than other two kinds of functions.
Consequently, the present study showed the possibility that the system will support
both area-based learning and sightseeing using VR, AR and MR.
Keywords Spatiotemporal information ·Mixed reality (MR) ·Virtual reality
(VR) ·Augmented reality (AR) ·Area-based learning ·Sightseeing
1 Introduction
In recent years, spatiotemporal information, which includes past spatial information
in addition to present spatial information, is utilized in a wide range of fields such as
geography, history, archaeology, sightseeing and culture besides spatial information
science. In this background, as geographical information systems (GIS) and global
positioning systems (GPS) are widely spreading, it is possible for not only scientists
and technicians but also the general public to easily utilize spatiotemporal informa-
R. Makino ·K. Yamamoto (B)
Graduate School of Informatics and Engineering, University of Electro-Communications,
1-5-1 Chofugaoka, Chofu-shi, Tokyo 182-8585, Japan
e-mail: kayoko.yamamoto@uec.ac.jp
R. Makino
e-mail: m1730096@edu.cc.uec.ac.jp
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_16
283
284 R. Makino and K. Yamamoto
tion in various fields. It is expected that spatiotemporal information will be more
important as a point of contact to arrange and relate other information. In response
to the above social and academic trend, in Japan, “comprehensive geography” will
become a compulsory subject, and problem-solving learning which responds to the
present needs will be emphasized in the social studies of high schools starting in the
2022 academic year. GIS are positioned as one of the most important elements in the
social studies of high schools, and they are already utilized in a wide range of our
daily life. Therefore, area-based learning using GIS is conducted in the schools, and
it is also adopted as a method of lifelong learning for a wide range of age groups. In
this way, area-based learning is conducted for not only students but also the general
public to understand geography and history of specific regions.
On the other hand, more recently, virtual reality (VR) and augmented reality (AR)
have been widely applied and widespread in our everyday life rather than academic
fields. VR can be defined as a synthetic or virtual environment which gives a person
a sense of reality, and AR can be defined to supplement the real space with virtual
(computer-generated) objects that appear to coexist in the same space as the real
space. Furthermore, both VR and AR are integrated into mixed reality (MR) which
is used in the systems for entertainment, e-sports and sightseeing.
Therefore, it is expected to effectively utilize spatiotemporal information using
GIS, VR and AR in various fields. Based on the above social and academic back-
grounds, the present study aims to develop a system that visualizes spatiotemporal
information in both real and virtual spaces, integrating social networking services
(SNS), Web-GIS, MR and the gallery system as well as Wikitude, and connecting
external social media. The system has various functions specialized for area-based
learning, sightseeing, and VR, AR and MR for area-based learning and sightseeing
in particular. Because spatiotemporal information is the most useful in area-based
learning and sightseeing as described above, the above functions should be inte-
grated into a single system in the present study. Therefore, all users can use all of the
functions for area-based learning and sightseeing as they like during the application
period of the system, and they can evaluate the use of the system and all of the func-
tions. Specifically, for area-based learning, the system is useful for a wide range of
users to learn geography and history using Web-GIS and VR. For sightseeing, the
system accumulates, shares and provides the spatiotemporal information concern-
ing sightseeing using Web-GIS, VR and external social media. Thus, the system is
expected to be used by a wide range of users while enjoying especially for area-based
learning and sightseeing.
2 Related Work
The present study is related to four study fields, namely, (1) studies that regenerated
landscapes using three-dimensional (3D) GIS (Yano et al. 2006,2008; Yamamura
et al. 2012), (2) studies that developed a system to support area-based learning using
old paper maps and information and communication technology (ICT) (Kudo et al.
16 Spatiotemporal Information System Using Mixed Reality … 285
2009; Maekawa 2012; Tsukamoto et al. 2013), (3) studies that developed a system to
support sightseeing integrating Web-GIS and social media (Yamada and Yamamoto
2013; Okuma and Yamamoto 2013; Murakoshi and Yamamoto 2014; Ikeda and
Yamamoto 2014; Yamamoto and Fujita 2015; Mizutani and Yamamoto 2017), and
(4) studies that developed a system to support sightseeing integrating Web-GIS,
social media and AR (Jang and Hudson-Smith 2012; Fujita and Yamamoto 2016;
Zhou and Yamamoto 2016; Ito et al. 2018).
Referring to the results of the preceding studies in related fields as listed above, the
present study demonstrates the originality to develop a system by efficiently accu-
mulating various kind of spatiotemporal information in virtual space using 3D GIS
and VR, and provide the information related to the target area using 3D GIS and AR.
Additionally, the present study shows the usefulness to design and develop various
functions of the system by optimizing for area-based learning and sightseeing. Thus,
as mentioned in the previous section, for both area-based learning and sightseeing,
the system is expected to support a wide range of users including the general public
in addition to students to utilize spatiotemporal information in both real and virtual
spaces.
3 System Design
3.1 System Configuration
In order to implement several unique functions, as described in detail in the next
section, in response to the aim of the present study, the system is made up of SNS,
Web-GIS, MR and the gallery system as well as Wikitude, and connected to external
social media as shown in Fig. 1. Web-GIS is provided by the Environmental Systems
Research Institute, Inc. (ESRI). The gallery system is originally developed to display
digital text, images using slideshow, tests in geographical and history, and references
on the users’ PC screens and mobile information terminal screens, accumulating them
in the database of the system. Wikitude is the Software Development Kit (SDK)
for AR development for mobile information terminals which is provided by the
GrapeCity Inc.
The system can support a wide range of users to learn geography and history, by
visualizing a variety of past and present information on the digital maps of Web-
GIS. Especially for area-based learning, the system can also support users to learn
in detail while enjoying, viewing digital text and the related images using slideshow,
and challenging the test related to the targets area in the gallery system. Additionally,
especially for sightseeing, the system can support users to effectively obtain various
kinds of information and knowledge using 3D GIS, AR and Wikitude. The system
can provide the information all over the world obtained from external social media.
The system applies using the Web server, database server and the GIS server.
The Web server and database server both use the Heroku, which is a Platform as
286 R. Makino and K. Yamamoto
Fig. 1 System design of spatiotemporal information system
a Service (PaaS) provided by the Salesforce Company. For the GIS server, ArcGIS
Online provided by the ESRI was used. The Web application software developed with
the system was implemented using Hypertext Preprocessor (PHP) and JavaScript.
3.2 Target Information Terminals
Though the system is set with the assumption that it will be used from PCs or
mobile information terminals (smartphones and tablet PCs), there is no difference in
functions on different information terminals, and the same functions can be used from
any information terminal as shown in Fig. 1. PCs are assumed to be basically used
indoors for all of the functions in the system. Though mobile information terminals
are assumed to be used indoors and outdoors for all of the functions in the system,
users may have some difficulty in viewing the scenes in high graphic way displayed
on the 3D digital maps of Web-GIS using VR. When users use all of the functions
specialized for VR, AR and MR, they have to use some kind of mobile information
terminals. The operating systems are Windows and Mac for PCs, and Android for
mobile information terminals.
16 Spatiotemporal Information System Using Mixed Reality … 287
4 System Development
4.1 System Frontend
The system will implement unique functions for users, which will be mentioned
below, in response to the aim of the present study as mentioned in Sect. 1. The target
users are a wide range, including the general public in addition to students who are
limited to over 18 years old. In order to implement these several unique functions,
the system was developed by integrating plural systems into a single system, and is
also connected with external social media. Additionally, using SNS, the system will
implement basic functions for registration, and submission and view of comments.
4.1.1 Functions for Area-Based Learning
The system has the following four functions for area-based learning whose com-
ponents are respectively depicted in Fig. 1. The functions for area-based learning
except for the function of statistical geography target the area surrounding Tokyo
Station in the central Tokyo Metropolis. The first reason for this is that there are a lot
of spots related to historically important events in a narrow range, and there are also
multiple famous places have been located since the Edo era in this area. The second
reason is that we can confirm distinct differences by comparing the land and space
uses in the modern age and the Edo era. However, as described below in detail, the
function of statistical geography expands the range to target all over the world.
(1) Function of Edo and Tokyo reproduction using VR
The function of Edo and Tokyo reproduction using VR was implemented by the
combination of 3D GIS and VR. Figures 2and 3show the pages for the function of
Edo and Tokyo reproduction using VR. Using the function, users can compare the 3D
digital maps of the Edo Era and present day to confirm the changes of landscape and
places of interest, and the damaged areas of the Great Meireki Fire (Furisode Fire)
and the expansion of reclaimed land in the Edo Era. Regarding the Great Meireki
Fire and the expansion of reclaimed land, using the function, as clearly shown in
Fig. 3, users can visually understand the width of such special historic areas on the
present 2D digital map of Web-GIS.
(2) Function for history learning
The function for history learning was implemented by the combination of Web-GIS
and the gallery system. Figure 4shows the digital text in the gallery system, and it
links to the 2D digital maps of Web-GIS. Summing up the information related to
the target area and the historical backgrounds during the Edo Era in digital text, the
system provides them to users in easy-to-understand manner. For example, as shown
in Fig. 4, users can refer to the explanation of all 15 shoguns (generals), and view the
special historic events in their reigns using the original 2D digital maps of Web-GIS
288 R. Makino and K. Yamamoto
Fig. 2 Pages for the function of Edo and Tokyo reproduction using VR (Left side: Edo, Right side:
Tokyo)
Fig. 3 Page for the damaged areas of the Great Meireki Fire and the expansion of reclaimed land on
the present map (Red area: 1st day of the fire, Orange area: 2nd day of the fire, Blue area: reclaimed
land in the Edo Era)
16 Spatiotemporal Information System Using Mixed Reality … 289
Fig. 4 Page for the function for history learning (Left side: explanation of the fourth shogun, Right
side: special historic event (the Great Meireki Fire) in his reign)
which were created by adopting various references. The digital text also links to such
references, and users can immediately access to their websites. Figure 4shows the
explanation of the fourth shogun and the special historic event in his reign shown in
Fig. 3.
(3) Function of test in geography and history
The test function of test in geography and history was implemented using the gallery
system. Users can view 60 questions, and download their own results after finishing
the test. For this, it is possible for users to examine and deepen their own understand-
ing of geography and history related to the Edo Era. Additionally, users can create
and submit new questions using the submission form of the system. As a result, using
the function, users can conduct area-based learning on their own initiative.
(4) Function of statistical geography
The function of statistical geography was implemented using 3D GIS. Specifically,
using the function, the screen displays the distributions and productions of main
mineral resources such as natural gas, iron ore, silver ore, gold ore, copper ore,
coal, diamond and crude oil all over the world on the 3D digital maps of Web-GIS.
Referring to the statistical data related to main mineral resources accumulated in the
database of the system, users can learn geography of respective areas all over the
world. Because Japan is an island country with few natural resources and greatly
depends on imported goods from other countries, it is significant for users to learn
290 R. Makino and K. Yamamoto
the distributions and productions of main mineral resources as part of area-based
learning.
4.1.2 Functions for Sightseeing
The system has the following three functions for sightseeing whose components are
respectively depicted in Fig. 1. Though all of the functions of sightseeing target all
over the world, it is possible just for users of the system to use them.
(1) Function of social media mapping
The function of social media mapping was implemented by the combination of
Web-GIS and external social media. All of the information with location informa-
tion submitted from social media such as Twitter, Flickr, Instagram, YouTube and
Webcamera can be gathered and displayed on the 2D digital map of Web-GIS. For
example, Fig. 5shows the page for the function of social media mapping, focusing
on Tokyo Station. Using the function, users can discover new sightseeing spots.
(2) Function of world natural heritages using VR
The function of world natural heritages using VR was implemented by the combina-
tion of 3D GIS and VR. As there are a lot of huge natural heritages especially in the
United States (U.S.) and Australia, the function mainly targets these two countries.
For example, Fig. 6shows the landscape of the Grand Canyon in the western part
of the U.S. The reason for this is that users can appreciate the effects of the function
using VR, taking up this example. Comparing with Google Earth, utilizing the func-
tion using VR, the virtual space reproduces the landscape of world natural heritages
equivalent to real ones to enable users’ experience of feelings the same as those in
the real space.
Fig. 5 Page for the function of social media mapping
16 Spatiotemporal Information System Using Mixed Reality … 291
Fig. 6 Page for the function of world natural heritages using VR
(3) Function for display of Edo ukiyo-e painting story maps
The function for display of Edo ukiyo-e painting story maps was implemented by
the combination of Web-GIS and the gallery system. Specifically, the images of Edo
ukiyo-e paintings in the gallery system were linked to the present 2D digital map
to create Edo ukiyo-e story maps. The images are related to “Fugaku Sanjurokkei
(Thirty-six Sceneries of Mt. Fuji, 46 places in total)” (Hokusai Katsushika), “Meisho
Edo Hyakkei (One hundred good sceneries in Edo, 99 places in total)” (Hiroshige
Utagawa), and “Tokaido Gojusan Tsugi (53 stages of the Tokaido Road)” (Hiroshige
Utagawa). For example, Fig. 7shows the Edo ukiyo-e story map related to the offshore
of present Kanagawa Prefecture in “Fugaku Sanjurokkei”. In this way, users can view
the landscape which cannot be seen in the present time referring to ukiyo-e paintings.
4.1.3 Functions Specialized for VR, AR and MR
As the functions partly using VR for area-based leaning and sightseeing were intro-
duced in Sects. 4.1.1 and 4.1.2, this section introduces the following four functions
specialized for VR, AR and MR whose components are respectively depicted in
Fig. 1. The function for area-based learning using AR is implemented by location-
based AR, and other three functions are implemented by the combination of the
functions for area-based leaning or sightseeing and image processing-based AR.
Regarding the former, users have to install a specific application software which was
originally developed using Wikitude in the present study into their mobile informa-
tion terminals, and permit the authorizations of cameras and GPS of their mobile
information terminals. Regarding the latter, users have to install the application soft-
292 R. Makino and K. Yamamoto
Fig. 7 Page for the function for display of Edo ukiyo-e painting story maps
wares of Wikitude and quick response code (QR code) reader into their mobile
information terminals.
(1) Function of Edo and Tokyo reproduction using MR
Figure 8shows the situation for the function of Edo and Tokyo reproduction using
MR. The function was implemented by the combination of the function of Edo and
Tokyo reproduction using VR and image processing-based AR. By overlaying a
mobile information terminal on the page for the function of Edo and Tokyo repro-
duction using VR (Fig. 2) in real space, and using Wikitude and QR code reader, the
images and movies of the present time are displayed in virtual space of the mobile
information terminal screen.
(2) Function of world natural heritages using MR
Figure 9shows the situation for the function of world natural heritages using MR. The
function is implemented by the combination of the function of world natural heritages
using VR and image processing-based AR. By overlaying a mobile information
terminal on the page for the function of world natural heritages using VR (Fig. 6)
in real space, and using Wikitude and QR code reader, the explanation, images and
movies related to the selected world natural heritages are displayed in virtual space
of the mobile information terminal screen.
(3) Function for history learning using AR
Figure 10 shows the situation for the function for history learning using AR. The
function is implemented by the combination of the function of history learning and
image processing-based AR. By overlaying a mobile information terminal on the
16 Spatiotemporal Information System Using Mixed Reality … 293
Fig. 8 Situation for the function of Edo and Tokyo reproduction using MR
Fig. 9 Situation for the function of world natural heritages using MR
294 R. Makino and K. Yamamoto
Fig. 10 Situation for the function for history learning using AR
page for the function for history learning (Fig. 4) in real space, and using Wikitude
and QR code reader, the lesson video which explains the related page of the digital
text in the gallery system is displayed in virtual space of the mobile information
terminal screen.
(4) Function for area-based learning using AR
Users can view the explanation of all bridges which span the Sumida River and the
Edo ukiyo-e paintings related to the target area (the area surrounding the Sumida
River) in virtual space of their mobile information terminal screen, by using the
specific application software developed in the present study, and the cameras and
GPS of their mobile information terminals. However, users have to visit the target
area in order to use the function. For example, Fig. 11 shows the situation for the
function for area-based learning using AR related to the Azuma-bashi as one of the
bridges which span the Sumida River.
16 Spatiotemporal Information System Using Mixed Reality … 295
Fig. 11 Situation for the function for area-based learning using AR
4.2 System Backend
(1) Creation of database to accumulate test in geography and history
When users create and submit new questions in geography and history on the page
for function of test in geography and history, the related files are accumulated and
listed in the database of administrators’ PC.
(2) Process concerning information obtainment from social media
In the backend of the system, in order to obtain all of the information with location
information submitted from social media such as Twitter, Flickr, Instagram, YouTube
and Webcamera, an API of each social media was authenticated using OAuth. This
enables all of the information with location information related to the points on
the 2D digital map of Web-GIS. Using the retrieval function of each social media
platform, users can search their favorite information on this digital map.
(3) Process concerning image processing-based AR
Using Target Manager which is a service provided based on the website of Wikitude,
administrators can upload images in the system, and display comments, other images
296 R. Makino and K. Yamamoto
and movies on them. After finishing this procedure, QR code will be created. Using
Wikitude and QR code reader, users can utilize the functions using image processing-
based AR.
(4) Process concerning location-based AR
At first, administrators investigate the location information of the special historic
spots of the target area, and insert it into the specific application software which
was originally developed using Wikitude in the present study. Next, they also insert
the explanations and images of the above spots into the application software. Using
the cameras and GPS of users’ mobile information terminals, the above spots are
displayed as icons on the screens. Tapping these icons, the explanations, images and
distance to the spots are displayed in virtual space of the mobile information terminal
screen.
4.3 System Interface
The interface is optimized according to the users’ PC screens and mobile informa-
tion terminal screens, and the administrators’ PC screens. However, the functions
using 3D GIS (functions of Edo & Tokyo reproduction using VR and MR, functions
of world natural heritages using VR and MR) should be used mainly with PC con-
sidering their high graphic performance. In the latter, the “ID”, “name”, “age” and
“gender” of all users can be checked on a list. Because information is displayed in
a list form on the administrators’ screen, and inappropriate submissions are deleted
using the Graphical User Interface (GUI) application, the system is designed so that
ites management is possible regardless of the administrators’ IT literacy.
5 Application
5.1 Application Overview
The application of the system was conducted over a period of five weeks (October
22–November 25, 2018) with the general public in addition to students (over 18 years
old). The application of the system was advertised using the website of the authors’
lab as well as Twitter and Facebook. Users register when using the system for the
first time. Users’ registration can be done by registering an ID and password. After
completing the registration, users will automatically go to the top page, and the use
of functions within the system will be made available.
16 Spatiotemporal Information System Using Mixed Reality … 297
5.2 Application Results
Table 1shows an outline of users of the system. The system has a total of 66 users
with 52 male and 14 female users. Regarding age groups, there are many male and
female users in their 20s making up 70% of the total. Subsequently, those in their
10s were 14% and those in their 50s were 9%. All users are Japanese, and their
places of residence mainly concentrate in the Tokyo metropolitan area. As a result
of the application, most of the users were young people who are familiar with new
technologies such as VR, AR and MR.
First, the access log analysis of users during the application period of the system
was conducted. The study incorporated an API of Google Analytics into the devel-
oped program, and then the access analysis will be conducted. The total number
of sessions was 257, and regarding the information terminals used as the method
for accessing to the system, PCs were 43%, smartphones were 44% and tablet PCs
were 13%. The total number of accesses to the system was 682, and Table 2shows
the accesses to the page of each function for area-based learning and sightseeing.
As it is clear from Table 2, the most accessed was the “function of Edo and Tokyo
reproduction using VR (39%)”, followed by the “function for display of Edo ukiyo-e
painting story maps (15%)”. Additionally, all users of the system used the functions
specialized for VR, AR and MR, despite that they can only use these functions to
access their related pages.
From these application results, it can be said that mobile information terminals
were used rather than PCs as access methods to the system, and the most popular
function is the function of Edo and Tokyo reproduction using VR as one of the func-
Tabl e 1 Outlines of users and online questionnaire survey respondents
10–19 20–29 30–39 40–49 50–59 60– To tal
Number of users 946 3 1 6 1 66
Number of questionnaire
respondents
332 1 1 3 1 41
Valid response rate (%) 33.3 69.6 33.3 100.0 50.0 100.0 62.1
Tabl e 2 Accesses to the page of each function for area-based learning and sightseeing
Function Percentage of accesses (%)
Function of Edo and Tokyo reproduction using VR 39.2
Function of history learning 10.1
Function of statistical geography 8.6
Function of test in geography and history 7.4
Function of social media mapping 9.5
Function of world natural heritages using VR 10.4
Function for display of Edo ukiyo-e painting story maps 14.8
298 R. Makino and K. Yamamoto
tions for area-based learning among users. Furthermore, the functions specialized
for VR, AR and MR as the most original functions of the system in the present study
were used by all users.
6 Evaluation
6.1 Overview of the Web Questionnaire Survey
Along with the purpose of the present study, a web questionnaire survey was imple-
mented in order to conduct an (1) evaluation concerning the use of the system and an
(2) evaluation concerning the functions of the system. The web questionnaire survey
was conducted for one week after the start of the application. Table 1also shows an
outline of the web questionnaire survey. Though there were no incentives for users, as
it is clear from Table 1, 41 out of 66 users submitted their web questionnaire survey,
and the valid response rate was 62%. The questionnaire respondents can evaluate the
use of the system and all of the functions.
6.2 Evaluation Concerning the Use of the System
Regarding the viewing frequency of the websites, 95% answered “every day”. On
the other hand, for the experience to use any application softwares using VR, 54%
answered “I have used it”, and for the experience to use any application softwares
using AR, 66% answered “I have used it”. From these results, it is evident that most
of the users were used to utilizing the websites, and more than half of them have used
the application softwares using VR and AR. Therefore, it is possible for most of the
users to easily utilize the system, by developing the web system integrated VR, AR
and MR in the present study.
6.3 Evaluation Concerning the Functions of the System
Figure 12 describes the evaluation results for the functions of the system. Specifically,
evaluation of the usefulness for all of the functions implemented in the system for
area-based learning, sightseeing, and VR, AR and MR was conducted.
(1) Evaluations for each function for area-based learning
Regarding the function of Edo and Tokyo reproduction using VR, all respondents
answered “useful” or “somewhat useful”. Regarding the function for history learning
and the function of statistical geography, 95% answered “useful” or “somewhat use-
16 Spatiotemporal Information System Using Mixed Reality … 299
Fig. 12 Evaluation results for the functions of the system
ful”. However, regarding the function of test in geography and history, 81% answered
“useful” or “somewhat useful”, and 19% answered “neither”. Accordingly, the func-
tion of test in geography and history were more negatively evaluated than other
three functions. It is likely because the questions prepared beforehand in the system
were not so easy, and it is difficult for users who were not interested in geography
and history to answer them. However, as users learn using other functions for area-
based learning by the continuous application of the system, the usefulness of the test
function may improve.
(2) Evaluations for each function for sightseeing
Regarding the function of social media mapping and the function for display of
Edo ukiyo-e painting story maps, 95% answered “useful” or “somewhat useful”, and
only 5% answered “neither”. Regarding the function of world natural heritages using
VR, 88% answered “useful” or “somewhat useful”, 10% answered “neither”, and
only 2% answered “not so useful”. The reason for this is that the function of world
natural heritages using VR can be used mainly with PC considering the high graphic
performance.
(3) Evaluation specialized for VR, AR and MR
Regarding the function of Edo and Tokyo reproduction using MR, though 93%
answered “useful” or “somewhat useful”, 5% answered “neither”, and only 2%
answered “not so useful”. Regarding the function for history learning using AR,
though 95% answered “useful” or “somewhat useful”, 3% answered “neither”, and
only 2% answered “not so useful”. Regarding the function for area-based learning
using AR, 98% answered “useful” or “somewhat useful”, and 2% answered “nei-
ther”. However, regarding the function of world natural heritages using MR which
300 R. Makino and K. Yamamoto
can be used mainly with PC, though 85% answered “useful” or “somewhat useful”,
13% answered “neither”, and only 2% answered “not so useful”.
(4) Discussion about the evaluation concerning the functions of the system
Thus, all of the functions for area-based learning and sightseeing were highly eval-
uated by the same users of the system. Comparing these functions, the functions
specialized for VR, AR and MR were negatively evaluated by some users. The rea-
son for this is that some users had not used the application softwares using AR or
VR yet, and they needed a little more time to gain familiarity with these functions
implemented in the system. Therefore, these functions may be used more often by
the continuous application of the system, and this may advance their usefulness.
Additionally, as descried in Sect. 4.1.3, regarding the function for area-based learn-
ing using AR, users had to install specific application software into their mobile
information terminals, permit the authorizations of cameras and GPS of their mobile
information terminals, and visit the target area. Regarding the other three functions
using image processing-based AR, users have to install the application softwares of
Wikitude and QR code reader into their mobile information terminals, and overlay
their mobile information terminals on the PC screens.
7 Conclusion
Thus, in order to support a wide range of users while enjoying especially for area-
based learning and sightseeing, the present study developed a unique system that
visualizes spatiotemporal information in both real and virtual spaces, integrating
SNS, Web-GIS, MR and the gallery system as well as Wikitude, and connecting
external social media. Using these systems and technologies, the system has four
functions for area-based learning, three functions for sightseeing, and four functions
specialized for VR, AR and MR for area-based learning and sightseeing.
From the evaluation results based on the questionnaire survey to users after the
application, it was clear that all of the functions in the system were highly evaluated,
and most of the functions for area-based learning are more popular than other two
kinds of functions. Based on the results of the access log analysis of users during
the application period of the system, the most accessed was the function of Edo
and Tokyo reproduction using VR as one of the functions for area-based learning.
However, among three kinds of the functions, the functions specialized for VR, AR
and MR were more negatively evaluated than other two kinds of functions. Though all
users used the functions, some of them were unfamiliar with the functions during the
application period of the system. Additionally, users can not use the functions only
accessing to the related pages. Therefore, these functions may be used more often
by the continuous application of the system, and this may advance the usefulness.
Consequently, the present study showed the possibility that the system will support
both area-based learning and sightseeing using VR, AR and MR.
16 Spatiotemporal Information System Using Mixed Reality … 301
From the evaluation results in the previous section, the two points of improvement
for the system can be summarized as shown below.
(1) Regarding the functions for area-based learning, there still remains a lot of past
information related to the target area that has not been adopted into the system
yet. By accumulating such information in the system, it will be possible for
users to understand more detailed changes of the times.
(2) After completing the subjects described in (1), it is necessary to apply the system
for a longer period, and evaluate it in detail to advance the usefulness, referring
to Bishop et al. (2013), Russo et al. (2015,2018). Additionally, by targeting
other areas, it is necessary to enhance the significance of the use of the system.
Acknowledgements In the application of the spatiotemporal information system and the online
questionnaire surveys of the present study, enormous cooperation was received from those in Japan.
We would like to take this opportunity to gratefully acknowledge them.
References
Bishop I, Pettit CJ, Sheth F, Sharma S (2013) Evaluation of data visualisation options for land use
policy and decision making in response to climate change. Environ Plann B 40(2):213–233
Fujita S, Yamamoto K (2016) Development of dynamic real-time navigation system. Int J Adv
Comput Sci Appl 7(11):116–130
Ikeda T, Yamamoto K (2014) Development of social recommendation GIS for tourist spots. Int J
Adv Comput Sci Appl 5(12):8–21
Ito H, Ohyanagi T, Ishida T, Yonekura T (2018) Proposal of a regional knowledge inheritance
system using location-based AR and historical maps. IMIS 2018: innovative mobile and internet
services in ubiquitous computing. Springer, Berlin, 320–328
Jang SH, Hudson-Smith A (2012) Exploring mobile augmented reality navigation system for pedes-
trians. In: Proceedings of the GIS research UK, 6p
Kudo A, Kumota S, Ichikawa H, Abe A (2009) Development of regional study support system
for residents considering changes of area. In: Proceedings of the 71th national convention of the
information processing Society of Japan, pp 707–708
Maekawa M (2012) Interchange learning model between regions: interchange learning between
Japan and the Silk Road. In: Proceedings of the 28 the national convention of Japan Society of
Educational Information, pp 142–145
Mizutani Y, Yamamoto K (2017) A sightseeing spot recommendation system that takes into account
the change in circumstances of users. Int J Geo-Inf 6(19):303. https://doi.org/10.3390/ijgi6100303
Murakoshi T, Yamamoto K (2014) Study on a social media GIS to support the utilization of disaster
information: for disaster reduction measures from normal times to disaster outbreak times. Socio-
Informatics 3(2):17–31
Okuma T, Yamamoto K (2013) Study on a social media GIS to accumulate urban disaster informa-
tion: accumulation of disaster information during normal times for disaster reduction measures.
Socio-Informatics 2(2):49–65
Russo P, Costabile FM, Lanzilotti R, Pettit CJ (2015) Usability of planning support systems: an
evaluation framework. In: Geertman S, Stillwell J, Ferreira J, Goodspeed J (eds) Smart cities and
planning support systems. Springer, Berlin, pp 337–353
Russo P, Lanzilotti R, Costabile MF, Pettit CJ (2018) Towards satisfying practitioners in using
planning support systems. Comput Environ Urban Syst 67:9–20
302 R. Makino and K. Yamamoto
Tsukamoto A, Shibata Y, Kurumida H, Takahashi T, Narumi K (2013) Development of a support
application for the local study based on historical maps and mobile devices. In: Proceedings of
the general meeting of the Association of Japanese Geographers. Spring, p 113
Ya mada S, Yamamoto K (2013) Development of social media GIS for information exchange between
regions. Int J Adv Comput Sci s 4(8):62–73
Yamamoto K, Fujita S (2015) Development of social media GIS to support information utilization
from normal times to disaster outbreak times. Int J Adv Comput Sci Appl 6(9):1–14
Yamamura H, Ando M, Manpuku K, Hirabe T, Tsukamoto A, Isoda Y, Nakata S, Tanaka S, Yano
K (2012) Development of a tool generating 3D urban model for supporting reconstruction of
tsunami-hit town. In: Proceedings of the 74th national convention of the information processing
Society of Japan, pp 253–254
Yano K, Isoda I, Nakaya T, Kawasumi T, Matsuoka K, Takase Y, Kawahara D, Kawahara N, Inoue
M, Tsukamoto A, Kirimura T (2006) Creating virtual time-space of Kyoto. E-J Geo 1:12–21
Ya no K, Nakaya T, Isoda I, Takase Y, Kawasumi T, Matsuoka K, Seto T, Kawahara D, Tsukamoto A,
Inoue M, Kirimura T (2008) Virtual Kyoto: 4D GIS comprising spatial and temporal dimensions.
J Geogr 117(2):464–478
Zhou J, Yamamoto K (2016) Development of the system to support tourists’ excursion behavior
using augmented reality. Int J Adv Comput Sci Appl 7(7):197–209
Part III
Mobility
Chapter 17
Origin-Destination Estimation of Bus
Users by Smart Card Data
Mona Mosallanejad, Sekhar Somenahalli and David Mills
Abstract The public transport smart cards offer transit planners access to a tremen-
dous source of spatial-temporal data, offering opportunities to infer a passenger’s
mobility pattern and path choices. It is essential to accurately estimate the origin and
destination (OD) matrix to understand the travel demand. This research has devel-
oped a new approach using a trip chain model to estimate public transport commuter’s
trajectories in a multi-legged journey. This research has proposed new algorithms to
link the passenger’s journeys involving the mode transfers using assumptions relating
to the passenger paths in between their successive boarding’s and their acceptable
walking distances. The study also developed assumptions to distinguish “transfer’
from ‘activity’ to accurately predict the passenger destination. This study results will
enable the public transport agencies to optimise the public transport routes and their
schedule; which will ultimately lead to the public transport system improvements
resulting in higher patronage.
Keywords Origin-destination matrix ·Public transport ·Trip chain model ·Smart
card
1 Introduction
Transport planners attempt to design transit facilities that will encourage people to
use public transport instead of private vehicles. As public transport agencies increas-
ingly adopt the use of automatic data collection systems, a significant amount of
M. Mosallanejad (B)·S. Somenahalli
University of South Australia, Adelaide, Australia
e-mail: mosmy007@mymail.unisa.edu.au
S. Somenahalli
e-mail: Sekhar.Somenahalli@unisa.edu.au
D. Mills
Department of Planning Transport and Infrastructure, Adelaide, Australia
e-mail: David.mills@sa.gov.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_17
305
306 M. Mosallanejad et al.
boarding data becomes available, providing an excellent opportunity for transit plan-
ners to access spatial-temporal data (Rahbar et al. 2017;Tao2018) which can be
used for a better understanding of human mobility and the performance of a transit
system (El Mahrsi et al. 2017). In comparison with traditional surveys, which are
usually time-consuming, expensive and of the ‘snapshot’ type, smart card data can be
used to examine a whole network regularly and to make realistic estimates of pas-
senger origin-destination (OD) patterns.
Developing approaches for estimating accurate OD matrices from smart card
data is critical for transit planners (Alsger et al. 2015). Having knowledge of travel
demand will facilitate the design of appropriate public transport routes, and lead to
the optimisation of schedules. In turn, this will enhance public transport patronage,
with the potential of improving the public transport system’s performance.
In this paper, a one-month (May 2017) dataset was used. The data was provided
by the Department of Planning Transport and Infrastructure (DPTI) in Adelaide,
South Australia. A new methodology was developed, using the trip chain model, to
estimate an OD matrix for Adelaide’s bus users. Adelaide was chosen for this study
because unlike in other cities, commuters scan their smart card upon boarding but
not on alighting. This allows the algorithm to be generic and therefore applicable
elsewhere.
2 Origin-Destination Estimation Methods
Demand for public transport depends on factors such as time of travel, weather, and
service reliability (Morency et al. 2007). Many procedures have been used to make
such predictions and estimates of OD matrices based on smart card data have been
carried out since the 20th century. These methodologies and their accuracy vary,
depending on the availability of data and the time they cover, which can vary from
a week to a year. Before the evolution of new technologies for collecting data, most
studies were based on household and on-board survey data, used in a variety of
methods to estimate an OD matrix. These methods included non-iterative algorithms
(Tsygalnitsky 1977), Fluid mechanics (Tsygalnitsky 1977), passenger on-off counts
and checker records at each stop (Simon and Furth 1985), constrained least squares
and the Fratar model (Gur and Ben-Shabat 1997), and fuzzy theory (Friedrich et al.
2000).
The introduction of the automatic fare collection system made it possible to
develop different methods for estimating an OD matrix. Initially a new method-
ology was proposed to compare OD trips versus the number of passengers (Barry
et al. 2002); since then, researchers have explored the potential of smart card data
to infer trip rates, turnover rates, and travel behaviour to improve planning aims
(Bagchi and White 2005; Utsunomiya et al. 2006). Methods based on automatic data
collection systems for OD matrix estimation include the Furness model (Lianfu et al.
2007), fusion approaches (Kusakabe and Asakura 2011), multiple linear regression
(Kalaanidhi and Gunasekaran 2013), iterative proportional fitting (Cui 2006; Gordon
17 Origin-Destination Estimation of Bus Users by Smart Card Data 307
et al. 2013; Horváth et al. 2014; Li and Cassidy 2007), maximum likelihood esti-
mation (Cui 2006; Ickowicz and Sparks 2015; Li and Cassidy 2007), Inferring the
alighting station via the straightforward algorithm and iterative method (Chapleau
et al. 2008; Seaborn et al. 2009; Zhao 2004; Zhao et al. 2007), and the trip chain
model (Ali et al. 2015; Alsger et al. 2018; Munizaga and Palma 2012; Nassir et al.
2011; Wang 2010).
The time-dependent OD matrix is estimated from passenger counts at both board-
ing and alighting stations and is based on the forecasting method linking boarding
and alighting data (Horváth 2012). This method included transfer time, and its vali-
dation is based on an application in the Hungarian capital city. Yang and Jun (2018)
develop a new methodology to visualise the travel patterns of transit commuters
in Seoul, South Korea, by calculating trajectories and using Carto to create a map.
The moth-flamed optimisation (MFO) algorithm is a new population-based meta-
heuristic algorithm that investigates the celestial navigation of moths to estimate
the OD matrix (Heidari et al. 2017). Li et al. (2018) compare different studies using
smart card information, to examine passengers’ travel behaviours and provide a com-
prehensive review of them. The trip chain model is a recently devised method for
determining travel patterns and travel behaviours, first utilised by Barry et al. (2002)
to estimate destinations (Li et al. 2018). Although there is no exact definition for a
trip chain, a basic description is that each chain consists of one or more stops to the
next destination, and a trip chain is specified according to the number of stops. The
algorithm which will be used here to estimate the alighting stop is based on the trip
chain model (Alsger et al. 2016; Langlois et al. 2016;Lietal.2018).
3 Data Structure
The smart card must be tapped, swiped or waved at the station, stop or vehi-
cle. Flat fare policy and some zonal fare policies require commuters to tap once
before boarding and records only this single transaction. However, in some cities
where an exit reader is available as well, and the fare policy is based on distance or
zone, for each trip, two records are available, for boarding and alighting (Kurauchi
and Schmöcker 2016).
The data used in this paper is based on the ‘MetroCard’ database used in Adelaide
and is collected by the DPTI for a specific period: May 2017. Each MetroCard
contains spatial and temporal information. In Adelaide, where a flat fare policy
operates, commuters validate their cards when they board a public vehicle but not on
alighting. Three modes of transport are available: bus, train and tram. The information
for each smart card transaction contains card identification, fare type, transport mode
used, time, date, stop code, route code and direction for each boarding (see Table 1).
When passengers swipe their card and pay an initial transaction, the fare is valid
for two hours, and passengers can use any public transport within this time without
incurring further costs.
308 M. Mosallanejad et al.
Tabl e 1 Individual MetroCard information
Media code Fare type Transport
mode
Date and
time
Stop code Route code Direction
807***CB SV 42017-05-01
09:49:35
8089 Tram 1
94E***FB TICKETS 1 2017-05-01
10:39:15
3351 251 1
11C***89 28DAY 12017-05-05
10:46:32
3285 271 1
707***27 OTHE R 12017-05-01
11:04:05
2072 H22 1
584***97 SV 52017-05-08
11:06:36
1852 GWC 1
Note Transport mode: 1 =Bus, 3 =Station, 4 =Tram, 5 =Train
There are some deviations from the one-swipe rule: railway stations in Adelaide
operate under a closed system, and swiping is required for both boarding and alight-
ing, and various systemic and user issues mean that transfers between the train and
other modes cannot be estimated directly from the MetroCard. Also, there is a free
tram zone in Adelaide where passengers do not need to swipe their cards; this means
that the tram boarding point is not available. Given these limitations, this study
focuses on bus users.
4 Methodology
Knowledge of transit demand plays a decisive role in public transport plans to improve
the performance of the system. One common method for estimating the destination
is the trip chain model. As mentioned previously, each smart card can provide the
boarding location and time of each bus trip but not the alighting location. This study
used various assumptions (as listed below) to estimate the passenger’s ultimate des-
tination. In the case of transfer trips, the trip chain model assumes the alighting stop
is located within an acceptable walking distance of the next stop and for calculating
the walking distance, the Euclidian distance was utilised.
Some assumptions considered in this algorithm are:
•The initial boarding location of a trip leg is the ‘origin’.
•A passenger’s alighting point is assumed to be within walking distance of the next
boarding stop in the case of transfer trips.
•Passengers return to the place where they first boarded that day, or to some other
nearby station.
•Commuters take the first available service after arriving at a boarding place.
17 Origin-Destination Estimation of Bus Users by Smart Card Data 309
•Each smart card is used by a single commuter and cannot be used by multiple
passengers.
•Commuters who use the public transport system do not use any other mode of
transport on that same day.
Here is the explanation for some of the terms used in this study:
•Media code: the unique identifier for each MetroCard in Adelaide.
•Time threshold: the waiting time between two consecutive transactions.
•Trip leg: the trip for an individual commuter between boarding and alighting stops.
•Walking distance: the maximum distance between two consecutive trip legs that
commuters walk to transfer to another public transport service.
•Trip ID: identifies an ID for each trip, which is unique for every service.
•Route ID: identifies a unique ID for each route.
•Stop ID: identifies a unique ID for an individual stop or station entrance; a multiple
route ID may use the same stop.
•Service ID: contains a unique ID of the available service for one or more routes.
•Block ID: identifies the block to which a specific trip belongs. A block can consist
of a single trip or more for the same vehicle.
4.1 Estimating the Alighting Stop
A new heuristic algorithm is used to estimate stop-level origins and destinations,
based on the boarding transactions in the MetroCard datasets. The algorithm used
to estimate the alighting stop is shown in Fig. 1. This flowchart was used for finding
the alighting stop and not the destination because not all alighting stops are the
destination of a trip leg, as some of the alighting stops may be used for transferring to
other modes or other buses. For OD estimation, some criteria like trip ID and service
ID were extracted from the Google Transit Feed Specification (GTFS) dataset. In
the database provided by DPTI, the stop ID for each MetroCard is different from the
stop code in GTFS data, and these need to be matched. Once that was done, the data
based on the transaction time was sorted, and a MetroCard ID was selected. Based
on the trip chain model, the subsequent transaction in each trip leg is a key point for
inferring the alighting stop. By considering the following transaction of a MetroCard
(the next boarding), the alighting stop was estimated by calculating the minimum
Euclidian distance. Based on the algorithm, for each transaction, the trip ID, service
ID and block ID from ‘stop_times.txt’ in GTFS data were selected. These criteria
are unique for each service for various modes of public transport: for example, a bus
which departs at a specific time from its origin has its own trip ID, service ID and
block ID, which may be different from the subsequent bus. By matching the time of
each transaction in MetroCard data with the arrival and departure time in GTFS data,
and by considering the day that the commuter swiped the card, a trip ID is chosen. If
there is no trip ID relevant to the MetroCard data, as an interval of five minutes was
310 M. Mosallanejad et al.
No
Yes
Read Metro Card ID
Match the stop ID from Metro card
Sort it based on the time
Is it the first transaction of a day?
Label as
"Origin "
Sort the following stops based on distance by using stop code and route ID, Label
them as X0,Y0(If there is a thru route, then select stops for both route number)
Read the following transaction Is it the last transaction of a day?
Read the latitude and longitude for the following
transaction and label them as X1,Y1
Label as
"Destination"
Calculate the Euclidean distance
((X1-X0)^2+(Y1-Y0)^2))^0.5
Find the stop with minimum Euclidean distance
Label as
"Alighting"
Is the distance less
than walking distance?
Label as "Cannot be inferred"
Yes
No
No
Yes
Fig. 1 Estimation of alighting stop
considered for selecting the trip ID. If in this period no trip ID was selected, then the
next available trip ID was chosen for the algorithm by considering a delay.
In Adelaide, some buses change their route ID in the middle of the route for some
specific hours, especially before entering the central business district (CBD). This is
known as a thru-linking route. The first stage is to infer the stop at which the route
ID changed to another one: in other words, by identifying the last stop for the current
route ID, the changing location can be inferred. To find the last stop, the data were
sorted based on arrival time. Then, based on the trip ID which was selected for the
transaction and the existing route ID, the last stop and its arrival time were chosen.
By entering the chosen stop and relevant time in the timetable database, the available
route could be selected. Routes with the same service ID and block ID could be
chosen and labelled as thru-link routes.
17 Origin-Destination Estimation of Bus Users by Smart Card Data 311
In the next step, the Euclidian distance was calculated between all stops along the
current route and the following transaction (next boarding). By using the stop code
and route ID, subsequent stops based on distance could be identified. The latitude
and longitude of these stops were labelled X0, Y0, and the latitude and longitude
of the successive transaction (next boarding) were labelled X1, Y1. Based on the
formula in the algorithm, the Euclidean distance could be calculated. For the next
stage, the stop ID with minimum Euclidean distance was selected. The distance was
compared with the maximum acceptable walking distance of 1000 m, derived for
Adelaide through sensitivity analysis; this distance will vary from city to city. If the
distance to the selected stop is less than the walking distance, then it was labelled
‘alighting stop’; otherwise, the alighting stop was labelled ‘cannot be inferred’.
Figure 2depicts an example of a trip chain model for inferring a passenger’s
alighting stop. If a commuter starts the trip at stop i on route 1 and the next transaction
is at stop j on route 3; then the alighting point can be estimated. As mentioned earlier,
some routes in Adelaide change their route ID, but passengers are not required to
revalidate their cards. For example, if route 1 changes to route 2 as shown in Fig. 2
(a thru-linking route), the Euclidian distance is used to find the alighting stop; all
distances from stops in route 1 and route 2 to stop j, ED1, ED2, ED3 and ED4, should
be calculated (see Fig. 2) and the stop with the minimum Euclidian distance selected
as the alighting stop: this should be less than the acceptable walking distance. For
instance, if the first boarding is at stop i and the second boarding at stop j, then the
commuter alighted at stop m in route 2 (the thru-linking route for route 1). Also, stop
i is the origin of the first trip leg because it is the first transaction of a day. If the next
transaction will be k, this is the last transaction of a day and based on the assumptions
the destination should be near the origin of a day i. By using the minimum Euclidian
distance from stop k to i by route 4, the alighting stop will be i which is the last
destination of a day, and there is no other transaction afterwards (Mosallanejad et al.
2018).
In some cases, the alighting stop could not be inferred if the distance to the next
boarding was higher than the acceptable walking distance. Manual analysis showed
that the GPS incorrectly selected stops in certain situations due to their proximity to
a stop on the other side of the road. If the alighting stop could not be inferred, then
the opposite stop was considered in the algorithm to check whether the alighting
stop could be estimated or not. An earlier study in Chile (Munizaga and Palma 2012)
for estimating the alighting stop considered a trajectory time to minimise the time
distance with the next boarding position time, for bus routes that utilise the same
street for both direction. They estimated this variable by adding the time associated
with position i to walking time from position i to next boarding by multiplying a
penalization factor. However, in this research, a new algorithm is developed for non-
inferred OD pairs due to observed GPS data errors in some boarding locations. The
new improved trip chain model algorithm developed in this research helped us to
accurately locate an additional 5% of alighting stops. An additional algorithm was
developed for locating the opposite stop (see Fig. 3).
If any commuter in special circumstances used different mode on his return trip
(for example occasional use of a friend’s car), it is difficult to track those trips. In
312 M. Mosallanejad et al.
Fig. 2 An example of a trip chain for inferring the alighting stop
Select the media code with no
alighting info in the previous step
Read latitude and longitude and label it as
X0, Y0
Read latitude and longitude for other stops in
Stop Reference and label it as X1, Y1
Calculate the Euclidean distance
((X1-X0) ^2+(Y1-Y0)^2))^0.5
Choose the minimum Euclidian Distance
For the selected stop check if the
route id is available or not
Select the next minimum Euclidean distance
Lebel as
"Opposite stop"
Yes
No
Fig. 3 Estimation of the opposite stop
17 Origin-Destination Estimation of Bus Users by Smart Card Data 313
such situations, the study tracked the travel pattern of the commuters over a week,
and then their alighting stops were accurately derived. Such an approach further
improved the accuracy of OD pair estimation by about 3%.
4.2 Estimating the Alighting Time
To estimate the alighting time, first trip ID, which is unique for each service, has
to be identified. The trip ID is selected using route ID and stop ID for boarding
transactions, based on the boarding time and date. If the trip ID is in 5-min intervals,
then alighting time is selected based on the alighting stop and trip ID. If this was
later than the boarding time, then it is labelled as the alighting time.
4.3 Destination Estimation
After estimating the alighting stop, four categories were considered to infer the
destination: First, the data was checked to see if it was the last transaction of a day;
if yes, the inferred alighting stop was labelled as the destination. If the alighting stop
for the last transaction of a day could not be inferred, the destination could not be
estimated. Next, it was checked to see if a commuter used the same route twice, or
used a parallel route, to reach a destination in a single day; if so, this was an ‘activity’,
since no-one alights from a direct route and takes the same or a parallel route again.
Thus, the alighting stop was taken as the destination point. This approach of using
parallel route information is an improvement of a standard trip chain model. The
third criterion to infer the destination of each trip leg was the time threshold between
two consecutive transactions. If the time threshold was less than 20 min, then the
commuter was assumed to have transferred to another bus, and the inferred stop was
also the alighting point. For time threshold of more than 20 min but less than an
hour, the label ‘short activity’ was used; if the time threshold was more than 1 h
the label ‘long activity’ was used. Both short and long activities were labelled as
the destination. The fourth criterion for investigating the destination stop was the
distance between the boarding stop and the subsequent alighting stop. If this value
was less than 400 m, then the alighting stop was labelled as the destination (see
Fig. 4).
5 Origin-Destination Analysis
One of the critical considerations when planning transit services is estimating the
demand for each route, to determine the frequency and capacity of the vehicles
(Tamblay et al. 2018). An OD matrix provides critical information for transit planners
314 M. Mosallanejad et al.
Read the
Alighting stop
Check if for the
subsequent trans-
action, the same
route or parallel
route is selected
Check if the distance between
boarding and subsequent alighting
transaction is less than 400 m Check if it is the
last transaction
of a day?
Check if the time threshold
between 2 transactions is
more than 1 hour
If it is less than 20 min,
label it as transfer
If more than 20
min and less than 1
hour, label it as short
activity
Label it as
Activity
Label as “Destination”
No
Yes
Fig. 4 Distinguishing transfer from activity
by estimating the number of journeys between different zones, information which
can be used in transportation planning, design and management. After analysing
the data based on the trip chain model, bus users’ origins and destination counts
during the morning peak were derived for each suburb (Fig. 5). Most trips originated
from Paradise, Modbury, Adelaide, and Klemzig suburbs. Three of them, Modbury,
Paradise and Klemzig are major interchanges for O-Bahn busway. Adelaide, Bedford
and Modbury are suburbs which destined most journeys during the day.
5.1 Discussion of Origin-Destination Analysis
The origin-destination analysis showed that bus movements were radial, and most
trips during the morning peak ended in the CBD. These movements were further
explored to rationalise the existing routes. The information below came from an OD
analysis that was used to identify specific routes. Suburbs with the highest origins and
destinations were shortlisted and analysed further; Fig. 6shows movement patterns
from these suburbs in terms of percentages, shown as the thickness of the desire
lines, of trips originated or attracted.
•Modbury–Bedford Park: the OD analysis showed high demand from Modbury
to Flinders University during the morning peak, but just one route (G40) runs
between the suburbs, going through the CBD. The results indicate that a direct
route is required from Modbury to Bedford Park.
•Paradise–Bedford Park: there are two bus routes between these two suburbs (W90
and G40), and both pass through the CBD, which is heavily congested during the
morning peak. It is worth exploring the option of a direct route from Paradise to
Flinders University that avoids congested city links.
17 Origin-Destination Estimation of Bus Users by Smart Card Data 315
Fig. 5 Origin and destination counts for each suburb (bus system)
Fig. 6 Percentage of trip movements between suburbs with high origins and destinations
•Modbury–North Adelaide: bus routes between these two suburbs run through the
CBD. Given the high demand on this route, it would be better to explore another
direct route and divert some buses.
316 M. Mosallanejad et al.
6 Validation
The best way to examine a model’s accuracy is to validate its results differently.
This was done through a survey in which fifteen volunteers were recruited randomly,
and their 407 transactions were analysed. This differed from the approach of earlier
studies, which undertook a household survey or utilised data from a closed system
where both boarding and alighting statistics are available. For example, Barry et al.
(2009) validated their assumptions by taking passenger counts at the exit and entrance
of the subway station in New York. Later, Devillaine et al. (2012) validated their
findings of the smart card by undertaking a travel survey in which users’ smart card
IDs were recorded. Munizaga et al. (2014) validated the assumptions they used in
the trip chain model by taking a travel survey of a small group of volunteers, which
returned 90% confirmation. In Brisbane where the ‘tap on, tap off’ system records
data for both boarding and alighting, the trip chain model assumptions were validated
against the go card dataset (Alsger 2016;Heetal.2015).
6.1 Estimating the Sample Size for a Survey
Estimating the sample size is critical for obtaining accurate results, and it is necessary
to investigate how much an increase in the sample size will lead to proper results with
fewer errors. In the context of survey objectives, two rationales can be considered:
the first is estimating the specific population parameters, and the second is testing the
statistical hypotheses. In this paper, the objective of the survey is related to population
parameters, and in such case factors that should be taken into account (Richardson
et al. 1995) include the variability of parameters across the population; the required
degree of precision; and population size.
Some approaches that consider estimating the sample size, such as that of Ceder
(2016), employ a procedure involving a survey for OD matrix, by taking into account
the percentage of passengers who travel between specific origins and destinations, the
population of each suburb, and the accuracy of each cell in the OD matrix. Previous
studies’ sample sizes vary as follows: 37 volunteers (Ebadi and Kang 2016), 53
(Munizaga et al. 2014), 306 (Lee and Hickman 2014) and 8000 households (Seaborn
et al. 2009).
This paper takes a different approach, using the discrete variable and based on a
random sample method. In this dataset which includes discrete variables, the standard
error for estimating a proportion p is given in Eq. 1(Richardson et al. 1995).
s.e.(p)=N−n
n(p(1−p)
n)(1)
17 Origin-Destination Estimation of Bus Users by Smart Card Data 317
where
n: sample size
N: population
In this study, the sample size was based on the population of the whole dataset
and assumed there would be a 95% correlation with the results. In the present study,
only the number of commuters who used buses was considered, and N =the number
of transactions per day by these passengers: 139,187. To calculate the population of
the whole dataset, the first week of May 2017 was considered, and Wednesday’s data
were selected as showing the most transactions. As per the equation, the minimum
sample size (with 95% confidence) for the transactions is estimated as 105. However,
this study analysed 407 transactions.
6.2 Survey
In this study, a survey was conducted by recruiting volunteers who usually used bus
services. Fifteen volunteers were randomly identified, and their smart card details
were collected after obtaining their written consent and ethics approval. The Depart-
ment of Planning Transport and Infrastructure provided the media code (unique
identifier in the dataset) for the smart card numbers, and two sources of data can
be matched by using the relevant ID. For fifteen participants over five months, 1686
transactions were collected, in which 1177 were related to the bus system. This inter-
view data helped in validating the estimated OD pair information derived for the trip
chain model developed in this study. Out of the 1177 transactions collected from
the interview survey, only 944 OD pair information was considered as error-free
data. The reported errors which are insignificant are due to the reporting of unusual
walking distance and also due to trip id errors. So only 944 OD pair information was
further used for validation purposes. When this information was compared with the
reported OD pair information derived from the interview survey, as many as 926 OD
pair information was tallied with the model results which amounts to 98% accuracy
(refer Table 2).
Tabl e 2 Survey data
information Number of volunteers 15
Number of transactions (5 months) 1686
Number of transactions for the bus system 1177
Number of inferred OD pairs 944
Number of accurate OD based on an interview 926
Accuracy level 98.09
318 M. Mosallanejad et al.
7 Conclusions
The transit OD matrix is a useful prerequisite for planners to optimise public trans-
port systems. The reliability of the system is an important criterion to encourage
people to leave their vehicles at home and take public transport instead. The primary
aim of this paper is in estimating an accurate OD matrix. A new methodology has
been developed, using SQL software and based on the trip chain model, to create an
OD matrix for Adelaide’s bus users and, as a result, to estimate the demand on the
transit system. The methodology assumes that passengers’ alighting points can be
determined using the Euclidian distance to the next boarding stop and considering
a minimum walking distance. This approach used various improvements over tra-
ditional methods for improving the estimated OD pair accuracy. These include (i)
minimising the GPS errors by using the stops on the opposite side of the road (ii)
increasing the OD estimation accuracy by observing commuter travel pattern over
a week period and (iii) improving the estimated OD accuracy by using the parallel
routes.
This study presents an overview of ridership patterns using one-month estimate
more accurate matrix. MetroCard data in Adelaide. The survey indicates that the
method used in this paper is 98% accurate and can be utilised elsewhere. An accu-
rate estimation of public transport OD will be a significant help to public agencies
involved in route rationalisation, which will lead to higher public transport patronage.
In further studies, census data could be used to validate this algorithm, and sensitivity
analysis could also be considered for various assumptions. It may also be possible
to estimate the purposes of various trips, based on smart card information if access
to such information is made available.
References
Ali A, Kim J, Lee S (2015) Travel behavior analysis using smart card data. KSCE J Civil Eng: 1–8
Alsger AA (2016) Estimation of transit origin destination matrices using smart card fare data.
School of Civil Engineering, The University of Queensland
Alsger A, Mesbah M, Ferreira L, Safi H (2015) Public transport origin-destination estimation using
smart card fare data. In: Transportation research board 94th annual meeting
Alsger A, Assemi B, Mesbah M, Ferreira L (2016) Validating and improving public transport
origin–destination estimation algorithm using smart card fare data. Transp Res Part C: Emerg
Technol 68:490–506
Alsger A, Tavassoli A, Mesbah M, Ferreira L, Hickman M (2018) Public transport trip purpose
inference using smart card fare data. Transp Res Part C: Emerg Technol 87:123–137
Bagchi M, White P (2005) The potential of public transport smart card data. Transp Policy
12(5):464–474
Barry J, Newhouser R, Rahbee A, Sayeda S (2002) Origin and destination estimation in New York
City with automated fare system data. Transp Res Record: J Transp Res Board 1817:183–187
Barry J, Freimer R, Slavin H (2009) Use of entry-only automatic fare collection data to estimate
linked transit trips in New York City. Transp Res Record: J Transp Res Board 2112:53–61
17 Origin-Destination Estimation of Bus Users by Smart Card Data 319
Devillaine F, Munizaga M, Trépanier M (2012) Detection of activities of public transport users by
analyzing smart card data. Transport Res Rec: J Transport Res Board 2276: 48–55
Ceder A (2016) Public transit planning and operation: modeling, practice and behavior. CRC Press
Chapleau R, Trépanier M, Chu KK (2008) The ultimate survey for transit planning: complete
information with smart card data and GIS. In: Proceedings of the 8th international conference on
survey methods in transport: harmonisation and data comparability, pp 25–31
Cui A (2006) Bus passenger origin-destination matrix estimation using automated data collection
systems. Massachusetts Institute of Technology
Ebadi N, Kang JE (2016) Constructing activity-mobility patterns of university at buffalo students
based on UB card transactions
El Mahrsi M, Come E, Oukhellou L, Verleysen M (2017) Clustering smart card data for urban
mobility analysis. IEEE Trans. Intell Transp Syst 18(3):712–728
Friedrich M, Mott P, Noekel K (2000) Keeping passenger surveys up to date: A fuzzy approach.
Transp Res Record: J Transp Res Board 1735:35–42
Gordon J, Koutsopoulos H, Wilson N, Attanucci J (2013) Automated inference of linked transit
journeys in London using fare-transaction and vehicle location data. Transp Res Record: J Transp
Res Board 2343:17–24
Gur Y, Ben-Shabat E (1997) Estimating bus boarding matrix using boarding counts in individual
vehicles. Transp Res Record: J Transp Res Board 1607:81–86
He L, Nassir N, Trépanier M, Hickman M (2015) Validating and calibrating a destination estimation
algorithm for public transport smart card fare collection systems. CIRRELT
Heidari A, Moayedi A, Abbaspour RA (2017) Estimating origin-destination matrices using an
efiicient moth flame- based spatial clustering approach. Int Arch Photogrammetry Remote Sens
Spat Inf Sci 42
Horváth B (2012) A simple method to forecast travel demand in urban public transport. Acta
Polytech Hung 9(4):165–176
Horváth B, Horváth R, Gaál B (2014) A new iterative method to estimate origin-destination matrix
in urban public transport. Transport Research Arena Europe, pp 14–17
Ickowicz A, Sparks R (2015) Estimation of an origin/destination matrix: application to a ferry
transport data. Public Transport 7(2):235–258
Kalaanidhi S, Gunasekaran K (2013) Estimation of bus transport ridership accounting accessibility.
Procedia-Soc Behav Sci 104:885–893
Kurauchi F, Schmöcker J-D (2016) Public transport planning with smart card data. CRC Press
Kusakabe T, Asakura Y (2011) Behavioural data mining for railway travellers with smart card data.
Behavioural data mining for railway travellers with smart card data
Langlois GG, Koutsopoulos HN, Zhao J (2016) Inferring patterns in the multi-week activity
sequences of public transport users. Transp Res Part C: Emerg Technol 64:1–16
Lee S, Hickman M (2014) Trip purpose inference using automated fare collection data. Public
Transport 6(1–2):1–20
Li Y, Cassidy MJ (2007) A generalized and efficient algorithm for estimating transit route ODs
from passenger counts. Transp Res Part B: Methodol 41(1):114–125
Li T, Sun D, Jing P, Yang K (2018) Smart card data mining of public transport destination: a literature
review. Information 9(1):18
Lianfu Z, ShuzhiZ, Yonggang Z, Ziyin Z (2007) Study on the method of constructing bus stops
OD matrix based on IC card data. In: International conference on wireless communications,
networking and mobile computing, 2007. WiCom 2007. IEEE, pp 3147–3150
Morency C, Trepanier M, Agard B (2007) Measuring transit use variability with smart-card data.
Transp Policy 14(3):193–203
Mosallanejad M, Somenahalli S, Vij A, Mills D (2018) Estimation of the Origin-Destination matrix
for bus system using MetroCard data, HKSTS conference
Munizaga MA, Palma C (2012) Estimation of a disaggregate multimodal public transport orig-
in–destination matrix from passive smartcard data from Santiago, Chile. Transp Res Part C:
Emerg Technol 24:9–18
320 M. Mosallanejad et al.
Munizaga M, Devillaine F, Navarrete C Silva D (2014) Validating travel behavior estimated from
smartcard data. Transp Res Part C: Emerg Technol 44:70–79
Nassir N, Khani A, Lee S, Noh H, Hickman M (2011) Transit stop-levelorigin-destination estimation
through use of transit schedule and automated data collection system. Transp Res Record: J Transp
Res Board 2263:140–150
Rahbar M, Mesbah M, Hickman M, Tavassoli A (2017) Determining route-choice behaviour of
public transport passengers using Bayesian statistical inference. Road Trans Res: J Aust NZ Res
Pract 26(1):64
Richardson AJ, Ampt ES, Meyburg AH (1995) Survey methods for transport planning. Eucalyptus
Press Melbourne
Seaborn C, Attanucci J, Wilson N (2009) Analyzing multimodal public transport journeys in London
with smart card fare payment data. Transp Res Record: J Transp Res Board 2121:55–62
Simon J, Furth PG (1985) Generating a bus route OD matrix from on-off data. J Transp Eng
111(6):583–593
Tamblay S, Muñoz JC, Ortúzar JdD (2018) Extended methodology for the estimation of a zonal
origin–destination matrix: a planning software application based on smartcard trip data
Tao S (2018) Public transport planning with smart card data. In: Kurauchi F, Schmöcker JD (eds)
Boca Raton: CRC Press. ISBN 9781498726580, Elsevier
Tsygalnitsky S (1977) Simplified methods for transportation planning. Master’s thesis, Mas-
sachusetts Institute of Technology Cambridge
Utsunomiya M, Attanucci J, Wilson N (2006) Potential uses of transit smart card registration
and transaction data to improve transit planning. Transp Res Record: J Transp Res Board
1971:119–126
Wang W (2010) Bus passenger origin-destination estimation and travel behavior using automated
data collection systems in London. Massachusetts Institute of Technology, UK
Yang H, Jun C (2018) Visualization of public bus passenger travel for travel pattern analysis. In:
Adjunct proceedings of the 14th international conference on location based services, pp 121–126
Zhao J (2004) The planning and analysis implications of automated data collection systems: rail
transit OD matrix inference and path choice modeling examples. Massachusetts Institute of Tech-
nology
Zhao J, Rahbee A, Wilson NH (2007) Estimating a rail passenger trip origin-destination matrix
using automatic data collection systems. Comput-Aided Civil Infrastruct Eng 22(5):376–387
Chapter 18
The Comparison Between Two Different
Algorithms of Spatio-Temporal
Forecasting for Traffic Flow Prediction
Haochen Shi, Yufeng Yue and Yunqi Zhou
Abstract Nowadays, there is an extensive body of literature that demonstrates
the methods of forecasting traffic flows, which includes artificial neural networks,
Kalman filtering, support vector regression, (seasonal) ARIMA models. However,
seldom articles use two or more than two methods to predict the traffic flows and
compare their difference within the forecasting process, which might be gradually
recognized as a potentially important research area in the future. Two of the most
commonly adopted methods, Space-Time Autoregressive Integrated Moving Aver-
age (STARIMA) and the Elman Recurrent Neural Network (ERNN), an Artificial
Neural Network, have been firstly harnessed to establish the space-time predicting
models. Secondly, according to the successfully trained models, the dissertation con-
ducts the multi-dimensional comparison based on four aspects: interpretability; ease
of implementation; running time and instability. Finally, some possible improve-
ments are put forward according to their forecasting performance which also indi-
rectly reflects their unique features and application environments.
Keywords STARIMA ·ERNN ·Traffic forecasting ·Traffic flows ·Space-time
predicting models ·Spatio-temporal
H. Shi (B)
Guangzhou Urban Planning & Design Survey Research Institute, Guangzhou
510641, China
e-mail: 329081772@qq.com
Y. Yu e
Department of Architect and Urban Planning, Tongji University,
Shanghai, China
e-mail: yufeng_yue@tongji.edu.cn
Y. Zhou
School of Geographical Sciences, University of Bristol, Bristol, UK
e-mail: ql18400@bristol.ac.uk
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_18
321
322 H.Shietal.
1 Background Introduction
1.1 Introduction and Related Research
Nowadays, traffic congestion, one of the most significant issues in cities, has gradu-
ally attracted more attention from both authority and scholars. A number of experts
attempt to figure out possible solutions for dealing with this challenging problem
while others tend to analyse the cause of its formation. On the one hand, both of
these research directions may involve the spatio-temporal analysis of traffic road
flows because it may be conducive for constructing, improving or even reforming
the cognition of entire traffic flow patterns. On the other hand, our current data-rich
environment makes specific traffic road data, including time and location informa-
tion, possible to be collected and handled (Cheng et al. 2012). Therefore, such a topic
has gained significant attention from many scholars.
From the aforementioned discussion, it can be said that spatio-temporal analysis of
traffic flows is possible and academically worthwhile to conduct and that is the reason
why there is a large amount of research focusing on this topic. Thus, many regres-
sion and forecasting methods, containing both linear and non-linear algorithms, are
introduced or created for conducting such research. Such methods include artificial
neural networks (Dougherty and Cobbett 1997; van Lint et al. 2005), Kalman filter-
ing (Liu et al. 2006), support vector regression (Wu et al. 2004), (seasonal) ARIMA
(Williams and Hoel 2003) models, etc. In this chapter, non-linear methods STARIMA
and ERNN will be used to establish the space-time predicting model. As a simple
but effective method, STARIMA has been widely introduced into different spatio-
temporal research since the early 1980s. It has been applied in physical, social and
environmental sciences, such as river flow (Perry and Aroian 1979), spatial econo-
metrics (Giacomini and Granger 2004), traffic flow (Kamarianakis and Prastacos),
spread of disease (Martin and Oeppen 1975), population diffusion (Bennett 1975),
criminal justice (Deutsch and Pfeifer 1981), innovation diffusion (Tinline 1971), etc.
To conclude, a large number of studies demonstrate the success and high accuracy
of STATIMA when building a non-linear forecasting model (Lin et al. 2009). As for
ERNN, it is also widely involved in many spatial and temporal analyses of different
fields, such as, the spatial part of spatio-temporal prediction model for forecasting
forest fires in Canada (Cheng and Wang 2008); short-term traffic flows forecasting
of main roads in Beijing (Dong et al. 2009); a case study of forecasting high-speed
network traffic (Feng et al. 2006) and a prediction of daily foreign exchange rates
(Giles et al. 2001). Similar to STARIMA, the ERNN also successfully solves many
time or space forecasting issues.
This chapter consists of three parts. The first part is defined as model building,
which mainly focuses on identifying and adjusting the parameters of the forecasting
model based on the results of accuracy evaluation. The second part is the model
comparisons, which initially illustrate and subsequently compare both the strengths
and weaknesses of the two forecasting models. As for the third part, the research
18 The Comparison Between Two Different Algorithms … 323
will attempt to put forward some suggestions for improving each model and provide
some critical comments on each model.
1.2 Data Description and Selection
There are two main datasets utilized; one is the traffic flow data captured by cameras
and the other is the London Congestion Analysis Project (LCAP) adjacency matrix
data. In addition, there are two other kinds of datasets, including a shapefile of LCAP
with all links and dates of data collection (Fig. 1). There are in total 256 road links
within 30 days from January 1 to January 30 in 2011. Data was collected between
6 am and 9 am every date across 5 min time intervals. Thereby there exist 180
observations per day. The travel times data are collected utilizing automatic number
plate recognition (ANPR) cameras by Transport for London’s (TfL’s) LCAP (Cheng
et al. 2012).
In terms of selecting road links, the Borough Southwark is selected for analysis
due to its location. More specifically, the Southwark is located at the south bank of
the Thames River and next to the City of London where a large number of citizens
work. Therefore, on one hand, the main roads in Southwark are possibly suffering
traffic congestion during both AM peak and PM peak period every weekday. On the
Fig. 1 Spatial extent of LCAP all links network in London (Base Map: Open Street Map)
324 H.Shietal.
Fig. 2 Spatial extent of selected road network in London (Base Map: Open street map)
other hand, their traffic flows follow cyclical patterns, which is not only simple to
predict but also significant for further research on traffic congestion. As for the data,
there are in total 24 road links in this Borough, but four of them are separate and do
not have any connections with other roads. Therefore, these four road links are not
selected and only 20 road links are selected as components of the final road network
(Fig. 2).
2 Exploratory Spatio-Temporal Data Analysis
2.1 Data Aggregation and Division
2.1.1 Data Aggregation—Average Hour Data
As for traffic flow data, it comprises 180 observations per day that collected from 6 am
to 9 pm. As for analysis and later prediction, it is not convenient to plot ST-ACF (the
space–time autocorrelation function) or ST-PACF (space-time partial autocorrelation
function) and observed results. Therefore, data is aggregated to each hour. After
aggregation, there are 15 observations per day and each observation is the average
traffic data of whole hour. As for each road link, this equate to y 450 records of data.
18 The Comparison Between Two Different Algorithms … 325
2.1.2 Data Division
It is obvious that weekend travel patterns are different from weekday patterns. Overall
peak time for weekends is different from weekdays and the peak values in week-
ends are smaller to the peak values in weekdays. Thus the data are divided into
weekdays and weekends for analysis and prediction. In order to observe and analyze
the autocorrelation structure varying whole day, weekdays’ traffic flowing data are
divided into three periods, which are AM peak, interpeak and PM peak according
to Transport for London (TfL). AM peak is from 7:00 to 10:00, PM peak is from
16:00 to 19:00 and interpeak is time period between two peaks which is from 10:00
to 16:00. This division method is widely accepted in London transportation studies.
It is accepted that traffic situation is distinct in each time period (Cheng et al. 2012).
In the meanwhile, as for weekends, the peak value and peak time is not obvious thus
it is not necessary to divide data into three periods.
2.2 STACF and STPACF and Corresponding Pattern
2.2.1 Network Adjacency Matrix
As for road network, it is clear to regard network as graph G =(N, E). There are
N nodes and E edges among network. The network could be expressed as an N * N
binary adjacency matrix, given LCAP adjacency matrix includes first order relation-
ships among nodes. Except defining the structure of the network, the directions of
each road link are also included in LCAP adjacency matrix.
2.2.2 STACP and STPACF Plot and Analysis
When all lags are larger than zero, the corresponding ST-ACF that are insignificant
space time data is stationarity and significantly positive ST-ACF values. Analysis of
this data indicates there are space-time autocorrelation (Cheng et al. 2012). A cyclic
pattern in ST-ACF points out that there exists seasonal pattern in spatial-temporal
traffic flow data. A ST-PACF plot is able to reveal whether there exists seasonal
pattern as well. In the meanwhile, through ST-PACF plot, it is also able to point out
which lag has significant influence compared with others (Fig. 3).
Weekdays AM Peak As for Weekday AM peak, there is an obvious periodic and
seasonal pattern with lag equal to three. Three lag means three hours, which is the
length of peak period, which reveals that in AM peak time traffic flow data increases
and would then decrease in the Interpeak period. The cyclic pattern is obvious and
repeated each peak time with significant autocorrelation. In terms of the ST-PACF
plot, it does not decay and indicates that there exits seasonal pattern of traffic flow
data in AM peak period on weekdays (Fig. 4).
326 H.Shietal.
Fig. 3 Plot of three selected roads
Fig. 4 ST-ACF plot and ST-PACF of weekdays’ AM peak
Weekdays PM Peak In terms of PM peak, the strength of periodic component is
much less than in AM peak period. This indicates that the PM peak period is not
stable and fixable from 16:00 to 19:00. In addition, the peak time of each day also
varies among this month. The peak time may begin earlier in Interpeak time or later,
the corresponding end time may also change. As for ST-PACF plot, there does not
exist significant cyclic pattern and it is not able to find any lag which has significant
influence on need-predicted date (Fig. 5).
Weekdays InterPeak As for ST-ACF plot of interpeak period, it displays a similar
pattern to AM peak with obvious seasonal and positive autocorrelation at spatial
order one. In addition, the lag is seven, which is the Interpeak period for each day.
Whereas, the amplitude of periodic component is much lower compared with AM
peak or PM peak. However as for Interpeak period, according to Cheng et al. 2012,
there should not exist obvious seasonal pattern due to Interpeak; time should be free
flowing situation without obvious cyclic pattern. Therefore, the division of each day
data into three periods is not sufficient to isolate data into different states. As for the
ST-PACF plot, to a certain extent, first eight lag has influence, which indicates that
18 The Comparison Between Two Different Algorithms … 327
Fig. 5 ST-ACF and ST-PACF plot of weekdays’ PM Peak
Fig. 6 ST-ACF and ST-PACF plot of weekdays’ inter peak
yesterday traffic flow data would have influence on need-predicted date. In addition,
seasonal pattern could be revealed through the ST-PACF plot (Fig. 6).
Weekends In terms of ST-ACF for weekends, there also exists a strongly seasonal
pattern and significant autocorrelation at the first spatial order. The lag equals to
15, which is the whole period time for each day. In addition, to certain extent, first
15 lags have effect on need-predicted data through ST-PACF plot. Simultaneously,
ST-PACF plot is also able to indicate seasonal patterns of weekend traffic flow data
(Fig. 7).
According to the analyzed results of ST-ACF and ST-PACF plots of weekdays
and weekends, it is found that lag is defined to 15, thus it is not necessary to divide
peak period and Interpeak period in the following forecasting. However, the patterns
between weekdays are so different from weekends, therefore it is needed to divide
traffic flow data into weekday and weekend and subsequently predict them separately.
328 H.Shietal.
Fig. 7 ST-ACF and ST-PACF plot of weekends
3 Methodology
3.1 Methodology 1: Space-Time Autoregressive Integrated
Moving Average Model (STARIMA)
3.1.1 The Introduction of STARIMA
Extending univariate ARIMA time series models into spatial domain generates
Space-Time Autoregressive Integrated Moving Average Model (STARIMA) (Lin
et al. 2009). The STARIMA model could be applied to single random variable obser-
vations at N immovable locations or site at distinct time points (Lin et al. 2009).
The STARIMA model displays Zi(t), which are observations of random variables
at location i (i equals to 1 to N) and time t as the weighed linear combination of
previous observations and errors (Pfeifer and Deutsch 1980). The STARIMA model
could be expressed as:
Zi(t)=
p
k=1
mk
h=0
ϕkhW(h)Zi(t−k)
−
q
l=1
nl
h=0
θlhW(h)εi(t−l)+εi(t)(1)
where Zi(t)is the space-time series variable at location i and time t;p is the autore-
gressive order; q is the moving average order; k is the time lag; h is the space lag; W is
the spatial weight matrix; mkis the spatial order of the kth autoregressive term; nlis
the spatial order of the lth moving average term; ϕkh is the autoregressive parameter
at temporal lag k and spatial lag h; θlh is the moving average parameter at temporal
lag l and spatial lag h; εi(t)is the random normally distributed error at location i and
time t (Cheng et al. 2012). The STARIMA model is incredibly valuable for forecast-
18 The Comparison Between Two Different Algorithms … 329
ing data when there exits spatial autocorrelation among observed system (Lin et al.
2009). In addition, the STARIMA models could be regarded as particular circum-
stances of the Vector Autoregressive Moving Average (VARMA) models (Lutkerpohl
1987). The VARMA model harnesses moving average parameter matrices and N×N
autoregressive to display cross-correlations and autocorrelations within N time series
(Kamarianakis and Prastacos 2005). Compared to VARMA model, the STARIMA
model provides better reduction results due to estimating number of parameters and
therefore facilitates the modelling performance in larger spatial scale (Kamarianakis
and Prastacos 2005).
3.1.2 The Procedure of STARIMA
The entire STARIMA model-building process mainly consists of five steps, which
will be described and illustrated in detail in the following section.
Building spatial weight matrix Binary is the simplest weighting method and has
been utilized many times on transportation analysis (Kamarianakis and Prastacos
2005). The first order of spatial weight matrix is utilized. In addition, row standard-
ization is harnessed into simple binary weighting matrix to obtain the final spatial
matrix for analysis (Cheng et al. 2012).
Space-time Autocorrelation (STACF) and partial autocorrelation analysis
(STPACF) Analysis and Model identification In this step, space-time autocor-
relation and partial autocorrelation analysis is applied to discover space-time auto-
correlation structure. Space-time autoregressive order p and moving average order
q would be determined. In addition, by utilizing seasonal differencing, it is able to
remove cyclic patterns and therefore d could be decided, which achieves the step to
transferring data into stationarity.
According to spatial temporal data analysis of Road 2090, the traffic patterns in
weekends are obviously different from patterns in weekdays. Thus all datasets are
divided into weekdays and weekends to conduct forecasting separately.
Weekday
It is obvious that distinct and stable cyclic autocorrelation is displayed in Fig. 8a due to
the annually periodic pattern in hour average traffic flow data in weekdays. Therefore,
seasonal differencing is necessary to remove seasonal patterns. The parameter d
determines the lag of the differencing and the chosen lag is 15 according to Fig. 8a.
According to Fig. 8b, it is obvious that cyclic patterns have been removed with the
lag equaling to 15. Figure 9a, b represents STPACF plot of weekdays’ real traffic
data original and after seasonal differencing.
Model identification to determine STARIMA order (p, d, q) of weekdays During
this step, the space-time autoregressive order p and space-time moving average order
q in STARIMA are determined. After seasonal differencing, the STACF plot is able
to determine moving average order q in STARIMA. According to Fig. 8b, the series
330 H.Shietal.
Fig. 8 STACF plot of hour average traffic data on Weekday. aBefore differencing. bAfter differ-
encing
Fig. 9 STPACF plot of hour average traffic data on weekday. aBefore differencing. bAfter dif-
ferencing
of data are stationary confirmed by the decreasing amplitude of STACF plot. First
five lags represent more significant correlation, suggesting moving average order q
could be chosen to five. The STPACF plot after differencing determines the moving
average order p. According to Fig. 9b it is discovered that STPACF plot breaks off
after lag one, which recommending that number I is the autoregressive order p. The
differencing is determined as 15 in previous steps.
Weekend
Similar to weekdays, it is obvious that strong cyclic autocorrelation is shown in
Fig. 10a due to the annually periodic pattern in weekends. Therefore, seasonal dif-
ferencing is necessary to remove seasonal patterns. The chosen lag is 15 according to
Fig. 10a. According to Fig. 10b, it is obvious that cyclic patterns have been removed
with the lag equaling to 15. Figure 11a, b represents STPACF plot of weekdays’ real
traffic data originally and after seasonal differencing.
Model identification to determine STARIMA order (p, d, q) of weekends
Similarly as for weekends, according to Fig. 13, the series of data are stationary
confirmed by the decreasing amplitude of STACF plot. First three lags represent sig-
nificant correlation, suggesting moving average order q should be three. According
18 The Comparison Between Two Different Algorithms … 331
Fig. 10 STACF plot of hour average traffic data on weekend. aBefore differencing. bAfter dif-
ferencing
Fig. 11 STACF plot of hour average traffic data on weekend. aBefore differencing. bAfter dif-
ferencing
to Fig. 3.8 it is discovered that STPACF plot breaks off after lag one, which recom-
mending that the one (1) is the autoregressive order p. The differencing is determined
as 15 in previous steps.
Parameter Estimation and Fitting of STARIMA, identifying parameters
of model by solving the equation Once the order of STARIMA (p, d, q) are deter-
mined, the parameters can be estimated. The requirement is to utilize 23 days’ data
to predict seven days’ data. Therefore 15 days of weekdays’ data are harnessed as
training data to predict the last five weekdays, and eight days of weekends’ data are
utilized as training data to forecast the last two days of weekends.
Diagnostic Check
The STACF is utilized to check whether residual is random. In addition, a histogram
graph was also harnessed to analyze the statistical residuals distribution of Road
2090 to present results more obviously and specifically. The diagnostic checking is
conducted separately for weekdays and weekends. As for weekdays, Fig. 12 reveals
that STACF of residual in weekdays is random and display statistical residuals distri-
332 H.Shietal.
Fig. 12 Diagnosis graphs of STARIMA model of Road 2090 on Weekday
Fig. 13 Diagnosis graphs of STARIMA model of Road2090 on Weekend
bution of Road 2090 in weekdays. Similarly, Fig. 13 reveals that STACF of residual
in weekends is random and display statistical residuals distribution of Road 2090 on
weekends.
Prediction with the STARIMA model: applying the model to the traffic data
forecasting In this step, function is applied to forecast weekday and weekend
traffic data separately and the results are shown in Fig. 14.
3.2 Methodology 2: Elman Recurrent Neural Networks
(ERNN)
3.2.1 The Introduction of ERNN
As for the compared method, the essay will harness Elman Recurrent Neural Net-
works (ERNN), one of Dynamic Recurrent Neural Networks (DRNNs) to build the
regression model and prediction model. Unlike ANN (Artificial Neural Networks)
which is undertaken only through modifying a set of weights to establish the corre-
18 The Comparison Between Two Different Algorithms … 333
Fig. 14 Comparison between forecasting value and actual value of weekdays and weekends
lation between dependent variables and independent variables, the ERNN’s connec-
tions between units form a directed cycle that allow previous states to engage in new
iterations. More specifically, a state layer is updated with the combination of both the
external input and the previous forward propagation during each iteration. This spe-
cial operating mechanism is able to store short-term memory as a future modelling
reference and therefore promote ERNN as one of the neural network models for
conducting both spatial and temporal forecasting research (McDonnell and Waagen
1994; Cheng and Wang 2008).
yk(t)=g(netk(t))(2.a)
netk(t)=
m
j
yj(t)wkj +θk(2.b)
yi(t)=fnetj(t)(2.c)
334 H.Shietal.
Fig. 15 The structure of
ERNN
netj(t)=n
i
xi(t)vij +
m
h
yh(t−1)ujh+θj(2.d)
As the structure showing in Fig. 15, a simple Elman recurrent neural network is
usually divided into four layers, input layer, hidden layer, undertaken layer and output
layer (Dong et al. 2009). Each layer represents different meanings and is linked by
two algorithms, output and input algorithms. The output algorithms are shown as
Eqs. (2.a) and (2.b), which propagates the result of hidden layers through weight w
while the input algorithms is defined as Eqs. (2.c) and (2.d), aiming at combining
input vector and the previous state activation together after multiplying weight v and
u respectively. In these equations, mrepresents the number of ‘state’ nodes and nis
the number of inputs; θjis a bias while fand gare output functions of two algorithms
respectively.
3.2.2 ERNN Spatial-Temporal Model Building
The entire model-building process will consist of three steps which will be illustrated
in the following part. In addition, as the reason demonstrated previously, the week-
days and weekend forecasting model, aiming at predicting last five weekdays and
last two weekends, will be separately built based on weekdays and weekend training
data respectively.
Time-series research, temporal anticipation In this step, an ERNN is established
to capture the temporal impacts of all previous n hours on the target hour. In the
scenario undertaken in this research, n is defined as 15 since 15 is the time lag
in the previous STARIMA model, which contains all the traffic flows data in the
previous day. In other words, the ERNN model we built is to use traffic flow data
of previous day for forecasting the traffic scenario of the following day. In addition,
few parameters should also be input and then adjusted when training the ERNN
model, which includes the number of hidden nodes, the number of iterations and the
18 The Comparison Between Two Different Algorithms … 335
learning rate. In Table 1, the essay lists corresponding temporal forecasting results
of the different parameters in both weekdays and weekend data and it is clear that the
forecasting model of both weekdays and weekend data is considerably accurate under
the conditions that the number of hidden nodes (n) is 15; the number of iterations
(maxit) is equal to 200 while the learning rate (r) is 0.9.
Spatial anticipation In this stage, based on the spatial matrix which is built for
demonstrating spatial relationship (adjacency) of road traffic flows, an ERNN is
established to calculate the spatial impacts of adjacent roads on the target road.
According to Cheng et al. (2012), the traffic flows of a road may be influenced by not
only first order roads (direct up or down stream roads) flows but also second order
roads (the up or down stream roads of first order roads) flows. Thus, instead of the
provided first order spatial matrix, the essay tends to use the combined spatial matrix
of both first order and second order spatial matrix. In this scenario, the combined
spatial matrix of target roads is built and shown in Table 2, which indicates that road
“2097”, “1614”, “1518”, “1402”, “2358”, “434” and “2416” are involved in spatial
ERNN model building as inputs. According to Cheng et al. (2012), when ERNN is
employed for spatial forecasting, the number of hidden nodes had better to be equal
to the input nodes. Except for that, similar with the temporal forecasting process, the
essay also tests different parameters in both weekdays and weekend data in spatial
forecasting and find that under the condition that the number of iterations is equal to
200 and the learning rate is 0.6, the model performs best.
Overall space-time anticipation—multiple linear regression This is the last stage
of forecasting model building, which is to produce the final spatio-temporal fore-
casting model through combining spatial and temporal forecasting model together
by a simple and easily understandable multiple linear regression. The equation of
multiple linear regression is as follows,
ffinal =aftime +bfspace +constantc(3)
where ftime and fspace represents ERNN temporal and spatial model respectively;
a and bare the regression coefficients while constant_cis the regression constant.
All of these coefficients need to be estimated beforehand. In the scenario, before
conducting the regression analysis, the essay calculates the correlation between both
observation and variables of testing data and plots it in Fig. 16. From that figure, it
can be concluded that both spatial and temporal variables are somewhat related to
observations (Fig. 17).
After that, the research establishes two final multiple linear regression equations
for weekdays and weekend forecasting respectively, which are shown as follows:
weekdays :fweek day =a1ftime +b1fspace +constantc1
weekend s :fweeken d =a2ftime +b2fspace +constantc2(4)
336 H.Shietal.
Tabl e 1 Temporal model information of ERNN (Average value of 100 times running)
Indicators n=10, maxit =200, r =0.6 n=15, maxit =200, r =0.6 n=20, maxit =200, r =0.6 n=30, maxit =200, r =0.6
Weekdays Weekends Weekdays Weekends Weekdays Weekends Weekdays Weekends
RMSE 0.01197 0.02513 0.01089 0.01941 0.01195 0.02129 0.01189 0.02721
R20.39101 0.11254 0.46234 0.22639 0.38977 0.18791 0.38388 0.16599
Indicators n=15, maxit =200, r =0.1 n=15, maxit =200, r =0.3 n=15, maxit =200, r =0.6 n=15, maxit =200, r =0.9
Weekdays Weekends Weekdays Weekends Weekdays Weekends Weekdays Weekends
RMSE 0.01663 0.02716 0.01855 0.03559 0.01144 0.02065 0.01059 0.02003
R20.16523 0.10201 0.19672 0.15397 0.43589 0.19927 0.48897 0.25073
18 The Comparison Between Two Different Algorithms … 337
Tabl e 2 Spatial model information of ERNN (average value of 100 times running)
Indicators n=7, maxit =200, r =0.1 n=7, maxit =200, r =0.3 n=7, maxit =200, r =0.6 n=7, maxit =200, r =0.9
Weekdays Weekends Weekdays Weekends Weekdays Weekends Weekdays Weekends
RMSE 0.01264 0.04741 0.01145 0.03810 0.01047 0.04869 0.01469 0.04859
R20.18490 0.43478 0.38469 0.53793 0.42513 0.58909 0.42926 0.48480
338 H.Shietal.
Fig. 16 Comparison between predict and actual of ERNN (Weekdays)
Fig. 17 Comparison between predict and actual of ERNN (Weekend)
Based on training data, a1,b1and constantc1are calculated as equal to 0.461,
0.719 and −0.03 respectively, while a2,b2and constantc2are equal to 0.343, 0.470
and 0.008 respectively. Here, the forecasting model has been established and their
predicting results are shown in Fig. 18.
3.3 Model Validation and Accuracy Evaluation
In the previous part, even if the research established two different spatio-temporal
models for forecasting traffic flow of a road based on the training data, the accuracy of
these models has never been assessed. Thus, it is necessary to validate the developed
model by testing data, which is known as model validation. In the research, three
indicators, R Square, the Root Mean Square Error (RMSE) and the Normalised Mean
Square Error (NMSE), will be introduced into validation process for assessing the
18 The Comparison Between Two Different Algorithms … 339
Fig. 18 Comparison between ERNN forecasting value and actual value of weekdays and weekends
accuracy of the models and their equations are as follows (Chatfield and Weigend
1994):
NMSE =1
N·1
σ2·
N
t=1yt−ˆyt2(5)
RMSE =N
t=1ˆyt−yt2
N(6)
R2=N
t=1ˆyt−¯y2
N
t=1(yt−¯y)2(7)
In the Eqs. (5)–(7)Nis defined as the number of pairs of both predicted values (ˆy)
and actual values (y) while yt,ˆyt,¯y,σ2represent each individual actual value, each
individual predicted value, the mean of all actual values and the estimated variance
of the data respectively. If the model performs well, both NMSE and RMSE will be
340 H.Shietal.
considerably low while R square will be relatively high. On the contrary, if the model
has high value of both NMSE and RMSE but low R square value, it means that the
model needs to be modified and rebuilt.
4 Results and Discussions
4.1 Results
According to previous part, both the STARIMA and ERNN model have been estab-
lished using training data and the majority of their model information is shown in
the Table 3.
Three previously introduced indicators of both STARIMA and ERNN models,
which are used for assessing prediction models, are listed in Table 4. Meanwhile,
Fig. 19 shows the performance of two different models in forecasting both target
weekdays’ and weekends’ traffic flow. From these tables and figures, it is explicit
that both weekdays and weekends model of STRIMA and ERNN are to some extent
acceptable as all of their R squares are around 0.5–0.6, while NMSE and RMSE are
less than 0.05.
Tabl e 3 The information of two models
Model types STARIMA ERNN
Space model Time model
Weekdays model p=1, d =15, q =5 N =7
Iteration times =200
Learning rate =0.6
N=15
Iteration times =200
Learning rate =0.9
weekdays :fweekday =
0.461ftime +0.719fspace −0.03
Weekend model p=1, d =15, q =3 N =7
Iteration times =200
Learning rate =0.6
N=15
Iteration times =200
Learning rate =0.9
weekends :fweekend =
0.343ftime +0.470fspace +0.008
Tabl e 4 The comparison of two models
Indicators Weekdays Weekends
STARIMA ERNN STARIMA ERNN
R square 0.5416 0.5818 0.5128 0.5887
RMSE 0.0106 0.0088 0.0138 0.0113
NMSE 0.0229 0.0195 0.0323 0.0110
18 The Comparison Between Two Different Algorithms … 341
Fig. 19 Comparison between forecasting and actual values of weekdays and weekends (both
STARIMA and ERNN)
4.2 Discussions
4.2.1 Performance Evaluation of Prediction Results and Possible
Reasons
Due to division of predicting weekday and weekend separately, the performance
evaluation of two models are also separate. As for weekday prediction, the modelling
fitting performances are similar. The ERNN has a slightly better prediction result.
Simultaneously, in terms of weekend prediction, ERNN model performs much better
according to all three indicators, including R square, RMSE and NMSE.
There are mainly three possible reasons. Firstly, STARIMA only utilizes first order
adjacency matrix in spatial anticipation while ERNN combines first order and second
order adjacency matrix. According to Cheng et al. (2012),thetrafficflowsofaroad
may not only be influenced by first order roads flows, but also by second order roads
flows. Therefore, the combined adjacency matrix perhaps leads to better prediction
342 H.Shietal.
results both on weekdays and weekends. In addition, compared with weekday traffic
flow data, there does not exist such obvious seasonal patterns among weekend traffic
flow data. As for the STARIMA model, it utilizes ARIMA model to predict in time
serious, which is a linear model while ERNN is non-linear model in time serious.
ARIMA model is more suitable for obvious seasonal pattern analysis. Thirdly, as for
data quantity of weekday and weekend, weekday has 300 records while weekend
only has 150 records. The data quantity of weekend is much smaller. According
to Button et al. (2013), small sample size data would undermine the reliability of
results. Thus, as for predicting weekends, the sample size is not too large and it is
more difficult to predict compared with weekday, and prediction performance may
have more differences.
4.2.2 Relative Merits of STARIMA and ERNN in Different Fields
Interpretability In terms of interpretability we divide it into two parts. Firstly,
according to compared results of the three indicators, both for weekday and week-
end prediction, the ERNN model has better prediction performances. In addition,
according to the coefficients of multiple linear regression, it is found that space has
more influence compared with time due to a larger coefficient of time. However, in
the STARIMA model, there is no method to find which aspect has more influence
on overall prediction result.
Ease of implementation The STARIMA model is easier in the field of imple-
mentation, which only has limited steps. However, the determining of parameters
(p, d, q) needs attention, which requires experience and deep understanding of the
STARIMA model to avoid mistakes. Whereas, as for ERNN model, it first needs to
conduct temporal anticipation, then conducts spatial anticipation and finally builds
the multiple linear regression combining temporal anticipation and spatial anticipa-
tion. If the regression results are not satisfied, it is necessary to restart from first step
and conduct whole process again.
Running time In terms of running time, the STARIMA model is less time consum-
ing due to maturely existing functions and packages in R studio. By utilizing these
tools, both trains and validations of model can be conveniently approached in a few
steps. In addition, the whole procedure of training and validations do not need to
repeat, which is different from the Artificial Neural Networks. However, as for the
ERNN model, it is necessary to run the programming code many times to obtain the
required parameters for temporal and spatial anticipation, which takes much more
time compared with the STARIMA model.
Instability As for the ERNN model, each time would output different forecasting
model. The final selected model has parameters which are approximate average
values after many trials. In terms of STARIMA model, parameter (p, d, q) are fixed
when the time of differencing is confirmed, which means there is only one unique
model for prediction.
18 The Comparison Between Two Different Algorithms … 343
Tabl e 5 The prediction of
other roads Indicators Weekdays Weekends
STARIMA ERNN STARIMA ERNN
Road 2090
R square 0.5416 0.5818 0.5127 0.5887
RMSE 0.0106 0.0088 0.0138 0.0113
NMSE 0.0229 0.0195 0.0323 0.1306
Road 1614
R square 0.6297 0.6446 0.2756 0.6024
RMSE 0.0212 0.0195 0.0260 0.0139
NMSE 0.0217 0.0339 0.0654 0.0352
Road 2415
R square 0.0414 0.3605 0.5040 0.6230
RMSE 0.1504 0.0960 0.0946 0.0729
NMSE 0.0730 0.0965 0.0662 0.0389
4.2.3 Other Roads Forecasting
As for different road links, forecasting performances of both two models are distinctly
different. For example, both models for Road 2415 are not able to predict very well
due to seasonal patterns of traffic data which is not easily observed. When the study
area varies, it may lead to certain changes to forecasting performances for each
model. Each road link has its own pattern along time series, therefore, the time
division method (weekday and weekend) for Road 2090 may not suitable for others.
However, Road 2415 is the boundary of road network, which may have other
connections with other roads which are not included in the selected 20 road links
network. Therefore the spatial adjacency matrix for Road 2415 is not completed,
thus forecasting results are not satisfactory for this type of road (Table 5).
4.2.4 Improvement
Although these two forecasting models to some extent perform acceptably, there are
still many improvements needed to be applied to their algorithms. For the STARIMA
model, the research puts forward two possible enhancement suggestions. The first
one is about the spatial matrix which is able to dominate spatial impact on forecasting
model. Our research suggestion is that compared with first-order adjacency matrix
input in STARIMA, second-order adjacency matrix might perform better, neverthe-
less seldom research publications about building the STARIMA model use it (Cheng
et al. 2012). In addition, instead of adjacency matrix, is there any possibility to intro-
duce distance matrix into STARIMA? As for the second suggestion, although the
majority of given traffic data is temporally stationary or near stationary, which is fit
344 H.Shietal.
for STARIMA application, it is necessary to consider some scenarios that the data
have some obvious time-series trends, such as increase or decrease. According to
Cheng et al. (2011), STARIMA is difficult to tackle with non-stationary temporal
data unless it is combined with other methods. Therefore, integrated methods, for
example ANN +STARIMA, may be the solution for these scenarios.
Similar with STARIMA, ERNN also has many improvements can to be accom-
plished. First of all, although the integrated spatial matrix (including both first and
second spatial matrix) has been introduced into the spatial model building process,
the “position” where it is involved in the model can be debated. In the ERNN spatial
model, only the first order roads (direct up or down stream roads) and second order
roads (the up or down stream roads of first order roads) are considered as variables
input to the ERNN models. In other words, the spatial matrix acts as a threshold to
label whether the variable should be engaged in the model. Thus, the first suggestion
is that it is better to introduce spatial matrix into each iteration process of ERNN
based on the method Cheng and Wang (2008) used. Secondly, although ERNN is
labelled as a common method in temporal forecasting, it is better to be used for deal-
ing with short-term time-series prediction, due mainly to the limitation that it does
not store well using long-term “memory”. Thus, for improving temporal forecast-
ing accuracy, the research suggests that an updated recurrent neural network, long
short term memory network (LSTM) which have capability to storage both long and
short-term memory, should be employed in time forecasting model.
5 Conclusions
To conclude the chapter, out research initially applied a spatio-temporal analysis of
traffic flow data in Southwark, which gave the results that there was a significant sea-
sonal pattern of AM peak in weekdays while PM peak is to some extent not explicit.
Secondly, two different model-constructing methods, STAMIRA and ERNN, were
introduced into the model building process to underpin the formulation of two fore-
casting models. Finally, the research compared these two different models and put
forth some suggestions to improve both models. In conclusion, although the research
provided an understandable but relatively insightful comparison of the STARIMA
and ERNN models, there is still a significant number of further analyses waiting
to be accomplished, such as the deeper internal principle comparison of STARIMA
and ERNN; the appropriately applicable scenarios of STRARIMA and ERNN; the
improvement of STARIMA and ERNN in forecast, etc.
References
Bennett RJ (1975) The representation and identification of spatio-temporal systems: an example of
population diffusion in North-West England. Trans Inst Br Geograph, 73–94
18 The Comparison Between Two Different Algorithms … 345
Button KS, Ioannidis JP, Mokrysz C, Nosek BA, Flint J, Robinson ES, Munafò MR (2013) Power
failure: why small sample size undermines the reliability of neuroscience. Nat Rev Neurosci
14(5):365–376
Chatfield C, Weigend A (1994) Time series prediction: forecasting the future and understanding
the past. Int J Forecast 10(1): 161–163
Cheng T, Wang J (2008) Integrated spatio-temporal data mining for forest fire prediction. Trans
GIS 12(5):591–611
Cheng T, Wang J, Li X (2011) A hybrid framework for space-time modeling of environmental data.
环境数据时空建模的混合框架. Geograph Anal 43(2):188–210
Cheng T, Haworth J, Wang J (2012) Spatio-temporal autocorrelation of road network data. J Geogr
Syst 14(4):389–413
Deutsch SJ, Pfeifer PE (1981) Space-time ARMA Modeling with contemporaneously correlated
innovations. Technometrics 23(4):401–409
Dong C, Shao C, Xiong Z, Li J (2009) Short-term traffic flow forecasting of road network based on
Elman neural network. J Transp Syst Eng Inf Technol, 146–149
Dougherty M, Cobbett M (1997) Short-term inter-urban traffic forecasts using neural networks. Int
J Forecast 13(1):21–31
Feng H, Chen D, Lin Q, Chen C (2006) Multi-scale network traffic prediction using a two-stage
neural network combined model. In: 2006 international conference on wireless communications,
networking and mobile computing, pp 1–5. IEEE
Giacomini R, Granger CW (2004) Aggregation of space-time processes. J Econometr 118(1):7–26
Giles C, Lawrence S, Tsoi A (2001) Noisy time series prediction using recurrent neural networks
and grammatical inference. Mach Learn 44(1):161–183
Kamarianakis Y, Prastacos P (2005) Space–time modeling of traffic flow. Comput Geosci
31(2):119–133
Lin SL, Huang HQ, Zhu DQ, Wang TZ (2009) The application of space-time ARIMA model on
traffic flow forecasting. In: 2009 international conference on machine learning and cybernetics,
vol 6, pp 3408–3412. IEEE
Liu H, van Zuylen HJ, van Lint H, Salomons M (2006) Predicting urban arterial traveltime with state-
space neural networks and kalman filters. Transp Res Rec J Transp Res Board 1968(1):99–108
Lutkerpohl H (1987) Forecasting aggregated vector ARMA processes. Springer, Berlin, p 323
Martin RL, Oeppen JE (1975) The identification of regional forecasting models using space: time
correlation functions. Trans Inst Br Geograph, 95–118
Mcdonnell J, Waagen D (1994) Evolving recurrent perceptrons for time-series modeling. IEEE
Trans on Neur Net 5(1): 24–38
Perry R, Aroian LA (1979) ‘Of time and the river: time series in M dimensions, the one-dimensional
autoregressive model. In: Proceedings of the American Statistical Association, Statistical Com-
puting Section
Pfeifer PE, Bodily SE (1990) A test of space-time arma modelling and forecasting of hotel data. J
Forecast 9(3):255–272
Pfeifer PE, Deutsch SJ (1980) A STARIMA model-building procedure with application to descrip-
tion and regional forecasting. Trans Inst Br Geograph, 330–349
Tinline R (1971) Linear operators in diffusion research. In: Chisolm MDI et al (eds) Regional
forecasting. Butterworths, London
van Lint JWC, Hoogendoorn SP, van Zuylen HJ (2005) Accurate freeway travel time predic-
tion with state-space neural networks under missing data. Transp Res Part C Emerg Technol
13(5–6):347–369
Wang J, Cheng T, Heydecker BG, Howarth J (2010) STARIMA for journey time prediction in
London. In: Proceedings of the 5th IMA conference on mathematics in transport. IMA
Williams B, Hoel L (2003) Modeling and forecasting vehicular traffic flow as a seasonal ARIMA
process: theoretical basis and empirical results. J Transp Eng ASCE 129(6):664–672
Wu C, Ho J, Lee D (2004) Travel-time prediction with support vector regression. IEEE Trans Intell
Transp Sys 5(4):276–281
Chapter 19
Developing a Behavioural Model
for Modal Shift in Commuting
Ali Soltani, Andrew Allan and Ha Anh Nguyen
Abstract Travel patterns of people across Australian cities have been dominated by
private cars. As noted by transport researchers, a sustainable transportation system
encourages people to make the shift towards non-motorised transport (i.e. public
and active transport) and emerging types of transport (i.e. ride-hailing and shared
bikes). Using an online questionnaire survey (n =410), this research reports on
the determinants of people’s transition to more sustainable modes of transport in
Adelaide, Australia. Further analysis undertaken using a discrete choice model, found
that home relocation and job changes were strongly associated with the modal shift
of respondents. Younger cohorts were likely to shift away from car usage despite the
significant influence in the change in participants’ family composition (i.e. birth of
a child), level of education, driving license, dwelling tenure, perceived safety and
costs. The significance of this study is that it determined that car dominance can
be reduced since there is a willingness of people to opt for non-motorised transport
options and other new shared mobility services. The chapter concludes with a varied
set of transport policies and strategies addressing different socio-economic groups
to increase the share of sustainable mobility, a critical step in moving towards a
‘smarter’ city.
Keywords Travel behaviour ·Modal shift ·Life-cycle events ·Sustainable
transport ·Discrete choice modelling ·Adelaide
A. Soltani
Shiraz University, Shiraz, Iran
e-mail: Soltani@shirazu.ac.ir
A. Allan (B)
School of Art, Architecture and Design, University of South Australia, Adelaide, Australia
e-mail: Andrew.Allan@unisa.edu.au
H. A. Nguyen
Faculty of Transport—Economics, University of Transport and Communications,
Hanoi, Vietnam
e-mail: nguyenhaanh@utc.edu.vn
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_19
347
348 A.Soltanietal.
1 Introduction
Currently 14% of global carbon emissions are generated by transport, predominantly
produced by road transport, both passenger and freight (Hensher 2008). In Australia,
between 1990 and 2004, transport was the second highest contributor of greenhouse
emissions with an increase of 23.4% (ABS 2007). The Australian Bureau of Statistics
has stated that the leading cause for the increase of transport emissions in this period
was due to greater usage of private vehicles with cities continuing to sprawl and
people having less access to public transport. This has occurred in response to a
growing population, increased low density urban sprawl and increased reliance on
personal fossil fuel powered private vehicles (many being SUVs, the most popular
segment of vehicles in Australia), resulting in a 71% share of transport emissions
(Adelaide City Council 2017).
Journey to work trips compromise over 40% of total travel within Australia’s cities
and regions. According to the 2016 Census data, driving is the dominant method of
journey to work in Australia, which accounted for 70% of the working population
(more than 6.5 million persons) commuting by car (ABS 2017). On the other hand,
the share using public transport for commuting increased significantly in Melbourne
and Sydney but decreased significantly in Perth and Brisbane, and remained nearly
constant in Adelaide between 2011 and 2016. A partial explanation for these modal
changes is due to job decentralisation and the changing distribution of jobs within
the metropolitan areas (Loader 2018).
The central question examined in this chapter is the extent that a transport modal
shift can be achieved towards more sustainable modes of transport such as walk-
ing/cycling, public transit and shared-mobility options, which would reduce the
share of car usage and its consequent carbon emissions. Other important questions
examined in this research included: the determinants of modal shift; how different
socio-economic and physical factors affect modal change in metropolitan Adelaide;
and the policy implications that can be extracted from the analysis of commuter
attitudes to changing their modal shift. Adelaide, the capital city of South Australia,
is the chosen case study due mainly to the recent planning strategies that have been
regulated by the local government to achieve sustainable development by promoting
Adelaide as the world’s first carbon neutral city (City of Adelaide 2016). In this
regard, transport policy that mitigates car dependency by shifting Adelaideans to
other alternative modes of transport has been placed among State Government’s top
priorities.
Although in the short term, it is nearly impossible for Adelaide and its suburbs to
reach zero carbon emissions with its current modes of transport (excluding the use of
carbon offsets), the State Government has endeavoured to increase the patronage of
Adelaide’s public transport systems in an attempt to reduce carbon emissions from
private vehicles. If public transit can cater for a large amount of people within a
close distance to the city, then there will be less private vehicle usage and therefore
less carbon emissions (Yang et al. 2012). To be able to increase the use of public
transport, more compact developments such as the Bowden redevelopment need to
19 Developing a Behavioural Model for Modal Shift in Commuting 349
occur (Rafat et al. 2018). Compact developments results in more people living in a
smaller area, and also generally closer to the city. Many of these developments also
limit the availability of car parking for each dwelling.
The rest of this chapter is structured as follows. The next section reviews the
background studies and draws a conclusion on what we already know about the
topic of commuting modal change. In the subsequent section, we then describe the
general patterns of journey to work and changes for commuting modes in metropoli-
tan Adelaide in addition to a description on the data collected through a questionnaire
survey. The fourth section presents the modelling, analysis and results. This chapter
concludes with a discussion of the findings in view of previous research and issues
for future investigation.
2 Background
Modal shift is defined as the shift from private cars to more sustainable modes of
transport such as public transport, or walking and cycling (Graham-Rowe et al.
2011). Previous studies (Buehler and Hamre 2014; Martin and Shaheen 2014;Rail
and Nazelle 2012; Scheepers et al. 2014; Steinbach et al. 2011) can be divided into
three groups based on identifying the main drivers of modal change: (a) reloca-
tions of job or housing place; (b) changes in transport infrastructure and physical
(built) environment; and (c) changes in personal/family characteristics or his/her
attitudes/habits.
2.1 Relocations of Job or Housing Place
The relocation of workplace or residence is one of the major causes of modal change
(Oakiletal.2011). It was recorded that about 18% of commuters in the UK changed
their commuting mode between years (Dargay and Hanly 2007). This figure was 28%
for those who moved house, 33% for those that had a workplace change and 45%
for those that relocated residence and workplace both. Changes in commuter modal
choices over a three month period were affected by job-related characteristics, access
to mobility resources, satisfaction with current commuting, awareness of sustainable
transport measures and changed life circumstances (Chatterjee et al. 2016). Moving
to a new house and retirement were two major factors affecting modal choice as
stated by interviewees (TfL 2010).
Clark et al. (2016), by using panel data from the UK Household Longitudinal
Study on commuting modes, found that about one fifth of workers have changed
their commuting mode from one year to the next. The modal shift is more likely
where journey to work distance changed dramatically especially in cases of job
or home displacement (Oakil et al. 2011). In a similar finding (but with the focus
on modal shift associated with residential change), a study of 295 respondents in
350 A.Soltanietal.
Halifax used a retrospective survey with the support of a random parameters logit
model (RPL) that clearly identified the modal shift of those who had relocated their
house. Over half (57.09%) of surveyed participants stated that they switched to new
travel modes along with their residential change. It is consistent with the fact that
residential displacement leads to changes in the built environment which motivates
people’s use of public and active transport, thereby decreasing their car usage (Cao
et al. 2009). Most recently, Klinger and Lanzendorf (2016) explored changes towards
car use, rail-based transit and bicycle participation of 1450 sampled respondents who
changed their residential locations in five German cities. The findings revealed that
the changes of the modal choice of surveyed respondents were for car use, rail-based
transit and bicycle participation. Most importantly, a clear association between the
mobility cultures of each city defined as ‘travel-related socio-physical context’ and
the likely shift of people towards three travel modes was clearly justified.
2.2 Changes in Transport Infrastructure and Physical (Built)
Environments
The modal shift of people as a result of the operation of new transport infrastructure
has been widely discussed among transport researchers. Babakan et al.’s (2015)
study on the simulation of modal shift after adding a highway and a new BRT line
in Tehran (Iran) showed that the new highway led to an increase in private car use
in commuting of households while adding a BRT line changed their commuting
mode from the private car to public transit. However, these changes are more evident
among low-income households without a private car. By contrast, households with
high incomes and higher rates of car ownership were less likely to change their
commuting mode from the private car to BRT.
Heinen et al. (2015) studied the commuting behaviour of 470 employees in Cam-
bridge after establishing a guided busway with a designated walking/cycling route in
2011, which showed that although net changes in modal transition were minor, the
new infrastructure promoted an increase in the share of commuting trips involving
active travel and a decrease in the share made entirely by car (the usage of public
transit remained constant). In relation to new tram services, studies of Pradono et al.
(2015), Termida et al. (2016), and Nguyen and Allan (2017) revealed individuals’
preferences of shifting away from car to tram transit.
In recent years, the emergence of new transportation modes, especially shared
mobility options such as bike sharing and car sharing highlight the potential of these
new mobility trends to shift people away from motorised transport. Fishman et al.
(2014) compared the shift from cars to shared bikes in five different cities of the
United States (US), United Kingdom (UK) and Australia and found that bike share
programs are associated with the reduction of motor vehicle use in Melbourne, and
Minneapolis where private cars are the dominant mode of transport. By contrast, an
opposite trend was revealed in cases of Washington D.C. and London where the car
19 Developing a Behavioural Model for Modal Shift in Commuting 351
substitution rates were quite low (at 2 and 7% respectively), which is in line with the
low car usage in these cities. The authors concluded that the modal shift from car to
bikeshare was higher in cities with greater car usage. Sharing the same objectives,
Shaheen et al. (2013) noted the benefits of bike sharing in modal shift in Toronto.
This is consistent with the research results of Fuller et al. (2013), who stated that the
percentage of car users that shifted to the new Montreal BIXI bike sharing program
ranged from 7.9 to 10.1% between 2009 and 2010. A later study by Shaheen and
Martin (2015) also confirmed the potential of shared bike schemes in shifting people
in North American cities away from private cars.
The cross-sectional study by Heinen et al. (2017) in Cambridge made an in-
depth exploration of travel behaviour change in five categories: (1) no changes;
(2) a complete modal shift; (3) a partial modal shift; (4) non-stable; and (5) random
patterns. The study found no specific evidence that introduction of changes in physical
environment was correlated with definite modal shifts, or with fitting in with any of
the categories of change patterns.
Panter et al. (2013) selected 655 commuters in Cambridge, UK and asked them
to report their personal and household information, psychological audits relating to
car usage and environmental settings on the way to workplace. They then tested
for any statistically significant association between these characteristics and their
willingness to change their mode of commuting by applying multivariate logistic
regression. The results showed that a combination of practices in parallel would be
effective in switching from car use to non-motorised and public transport. These
include improving the quality of cycling routes and walking paths in addition to
making restrictions on workplace parking. By contrast, Goodman et al. (2014) found
that this intervention in the built environment had not brought a significant effect
after one year, but after two years, exposure to the intervention predicted changes
in travel behaviour. Therefore, future studies collecting follow-up data for a longer
period after the intervention may result in additional insights.
2.3 Changes in Personal/Family Characteristics or Gender
Attitudes/Habits
Mode change is significantly associated with changes in family status (Oakil et al.
2011). Households with newborn babies are likely to shift to car usage as the needs
of baby-related maintenance activities such as regular health checks, kindergarten, or
playgrounds. This is also confirmed by study of Lanzendorf (2010), who conducted
qualitative retrospective interviews with 20 young parents of small children in Leipzig
(Germany). Almost all sampled mothers stated that the private car was preferred
because of increased convenience. However, the authors also acknowledged other
attributes that determined the shift towards other modes of transport among sampled
mothers as their maternal leave, income reduction, biographic reasons after the birth,
and their ‘strong emotional ties’ to other transportation options.
352 A.Soltanietal.
Redman et al. (2013), using a qualitative systematic review, concluded that while
public transport service frequency and reliability are crucial, the features that are most
influential in car usage are mostly linked with personal perceptions, background and
inspirations. Some modes of travel such as walking and cycling complement other
modes, because they act as access and egress modes. Therefore, it is expected that a
car user who infrequently cycles may be more likely to shift to public transport than
someone who just uses the car (Perone and Volinski 2003).
The study by Ramadurai and Srinivasan (2006) proved an inherent rigidity (inertia)
among people in changing their mode of travel. This effect was especially strong for
those used to cycling or walking. Furthermore, a transitional state-dependence was
observed between public transport and car usage. In fact, those who used a car for the
previous trip were less likely to select public transport or cycling. Similarly, those
who selected a bicycle in the former trip are more likely to choose walking among
the set of alternatives within the present journey.
Diana and Mokhtarian (2009) noted that single-mode travellers generally have
dissimilar expectations and attitudes toward different choices than multi-modal trav-
ellers. For instance, car drivers normally have biased judgments toward the cost and
time imposed by public transport. According to Diana (2010), multi-modal travel
habits affects modal switching. In fact, those who are aware of multiple choices are
more likely to engage in modal shift.
According to Idris et al. (2015) car drivers stated that on average they used a
car for 87.4% of their non-commute activities (73.6% as car driver and 13.7% as
car passenger), in comparison with only 3.8% who used public transport and 9.3%
that used walking/cycling options. This level of strong habit formation towards car
driving makes it hard to change the mode they are already habituated to. The study
concluded that while improving transit service performance is crucial in surging
modal shift, transport planning policies should also emphasis breaking habitual car
usage.
Based on the findings from the literature in this field, a discrete choice model based
on the micro-economic theory of utility maximisation philosophy was developed to
examine the complex impacts of socio-demographic, physical characteristics and the
personal habit/psychological factors likely to change commuting behaviour and the
likelihood of modal shift.
3 Data and Analysis
3.1 General Trend of Commuting in Adelaide
Before exploring the collected data, a descriptive analysis of journey to work patterns
in metropolitan Adelaide was conducted. According to job distribution data extracted
from ABS (2016), it is evident that Adelaide is relatively mono-centric city where
over a third of jobs (34%) are located a maximum of 4 km from Adelaide’s central
19 Developing a Behavioural Model for Modal Shift in Commuting 353
business district (CBD). Furthermore, comparing the share of outer jobs (66%) with
2011 (65%) shows that a minor decentralisation of employment occurred between
2011 and 2016 (ABS 2011,2016). The distribution of all jobs versus distance from
Adelaide’s CBD is detailed in Fig. 1(note, active transport has not been included
due to the total number of commutes being too low: less than 5% across non-CBD
areas).
From Fig. 1it is clear that private modal share is lower in areas closer to the CBD,
with about 80% of modal share dominating in Adelaide as close as 2 km from the city
centre. Public transit usage decreased dramatically 2 km beyond the city centre. The
time-series ABS data (in Fig. 2) illustrates the changes in commuting mode since
1976, with very little change in modal share patterns since 2005.
The ABS data on the method of travel to work for 7560 workers living in central
Adelaide (as the main hub of employment) where 61% worked full-time and 37%
part-time and using one method for going to work shows that using a car as driver
0
20
40
60
80
100
120
0.0 -
0.5
0.5 -
1.0
1.0 -
2.0
2.0 -
5.0
5.0 -
8.0
8.0 -
12.0
12.0 -
16.0
16.0 -
21.0
21.0 -
26.0
26.0 -
32.0
32.0 -
38.0
38.0 -
45.0
45.0
and
above
Per cent (private vehicle) Per cent (public transport) Share of jobs (percent)
Fig. 1 Distribution of jobs and the mode of commuting in Adelaide (adapted from ABS 2016)
0
10
20
30
40
50
60
70
80
90
1976 1981 1986 1991 1996 2001 2006 2011 2016
Car only Public transit Walk only Bicycle only Other
Fig. 2 Modes of commuting in Adelaide, 1976–2016 (adapted from ABS 2016)
354 A.Soltanietal.
(43.0%) is the most preferable mode of going to work (ABS 2016). This is followed
by walking (32.9%) and using the bus (9.99%). The figure for bike usage is also
significant (4.6%). The relatively high usage of non-motorised modes (37.5%) and
public transit (13.1%) illustrates the suitability of the urban environment for walk-
ing/cycling and the closeness of working places to residential areas for those living
in central Adelaide.
3.2 Data Collection
This paper is based on the analysis of primary data collected via a questionnaire
survey carried out in February and March 2018 in central Adelaide (n =410). The
geographical distribution of respondents is presented in Fig. 3. The aspects of com-
muting that influence informants’ modal shift for journey to work were recorded.
They were asked to report other travel-related factors for a typical day, including
travel mode, travel time and distance. Additionally, respondents were given ques-
tions on travelling which partly measured respondents’ travel attitudes (e.g. attitudes
towards the environment; safety; cost; independence; image and status, etc.), what
personal (e.g. health issue with using bikes) or external (e.g. about the quality of
infrastructure) barriers they perceived towards different modes of travel (the ques-
tionnaire form can be found on Appendix 1). The purpose of data analysis was to
find out that how the binary choice of modal shift was affected by different factors
including changes in job and housing distribution.
3.3 Model Specification
The dependent variable examined here was whether or not the respondent changed
his/her commuting mode within the last three years. Since this variable has a binary
value of one (for those who changed the mode) and zero (for those who did not
change the mode), the behavioural theory of random utility was employed to explain
the behaviour of individuals. Discrete choice theory was developed in the 1970s by
Nobel economist Daniel McFadden based on the traditional microeconomic theory
of consumer behaviour (Train 2009). However, while in theory the goods per se gen-
erate utility, in discrete choice modelling, the properties of the goods generate the
utility. The logit function is regarded as the main essence of discrete choice models.
Logit models are inherently able to represent complex characteristics of travel deci-
sions of individuals by including important socio-demographic and policy-sensitive
explanatory factors (Anwar and Yang 2017). The outputs of discrete choice models
are frequently utilised as an input for cost benefit analyses (CBA) of transportation
projects. The other advantage of logit to conventional regression is that it does not
assume that independent and dependent variables are correlated linearly, therefore it
does not entail that the variables be normally distributed. Rather, the logistic regres-
19 Developing a Behavioural Model for Modal Shift in Commuting 355
Fig. 3 Distribution of respondents based on residential suburb
356 A.Soltanietal.
sion function estimates the likelihood that a certain event would happen based on the
independent variables.
A discrete choice model is a mathematical function which forecasts an individ-
ual’s personal choice based on the utility or comparative benefit (Ben-Akiva et al.
1985). According to the purpose of this chapter, the binary logit model is used as
an analytically convenient modelling method for discovering the causal relationship
between modal shift and explanatory factors. Mathematically, for the n-th individual,
let i and j be the two alternatives in the binary choice set of each individual:
Uin =Vin +εin (1)
Ujn =Vjn +εjn (2)
where: Uin—the true utility of the alternative ito the n-th individual; Vin—the deter-
ministic or observable portion of the utility estimated by the analyst; εin—the error
of the portion of the utility unknown to the analyst.
Vin =f(Xi,Sn)(3)
where: Xi—the portion of utility associated with the attributes of alternative i; Sn—the
portion of utility associated with characteristics of the n-th individual.
The deterministic component of utility can be written as below for the model:
VModeshif ted(MS)=β0+β1_ MS ∗retired +β2MS ∗age +β3MS
∗house strucure +β4MS ∗level of ed ucat i on +β5MS
∗house strucure +β6MS ∗level of ed ucat i on +β7MS
∗house relocat ion +β8∗jobchange +β9
∗havingdriver license (4)
where β0is the constant, β1,β2,β3,β4,β5,β6,β7,β8,β9are the coefficients of vari-
ables.
The probability that the n-th individual choose alternative (Pin) as proposed by
Ben-Akiva and Lerman is presented as follows:
Pin =1
1+e−vn
=evin
evin +evjn (5)
The probability that an individual will choose mode shifted can be written as:
PMS =evin
evin +evjn
=evMS
evMS +evMNS (6)
The binary logit model employed in model estimation has the following form:
19 Developing a Behavioural Model for Modal Shift in Commuting 357
Modal shift =f(x):f(x)=1
1+e−βX(7)
ln f(x)
1−f(x)=βX=β0+β1x1+β2x2+β3x3+β4x4
+β5x5+β6x6+β7x7+β8x8+β9x9(8)
f(x)
1−f(x)=eβX(9)
where: xis a vector of selected explanatory variables, β0is the constant and βis a
vector of estimated coefficients.
Where: PMS is the probability that the n-th individual makes a switch to the other
modes. A binary logit model for commuting was developed for two choices, namely,
mode shifted (MS) and mode not shifted (MNS), in order to compare the utility of
these two alternatives and identify those factors which would affect an individual
to move from travelling by one mode to choosing another mode. In this model, the
dependent variable was “1” if the commuter made a change in his/her mode within
a certain period (last three years) and “0” for not changing the mode.
The coefficients are estimated by fitting the data to the model. The maximum like-
lihood (MLL) estimation method is a frequently used fitting method. This technique
comprises choosing values for the coefficients to maximise the probability (or like-
lihood) that the model predicts the same choices made by the observed individuals.
The method yields highly accurate estimates.
The Omnibus Tests of Model Coefficients is used to check that the new model (with
explanatory variables included) is an improvement over the baseline model. It uses
chi-square tests to see if there is a significant difference between the Log-likelihoods
(specifically the −2LLs) of the baseline model and the new model. If the new model
has a significantly reduced −2LL compared to the baseline then it suggests that the
new model is explaining more of the variance in the outcome and is an improvement.
Here the chi-square is highly significant (chi-square =91.271, df =10, p< 0.000)
so our new model is significantly better.
The pseudo R squared (Nagelkerke R2) value of 0.30 (compared to the model with
no coefficients) for the individual’s modal change model show an appropriate fit for
the model developed for entire metropolitan area (−2 Log likelihood =360.885)
(Table 1). In fact, the explanatory power of this model is modest, even though not
oddly low for modal choice models. The t-statistics of the constant and the coefficients
of variables in the model are all above the threshold values of ±1.96 (95% level of
confidence) showing the coefficient estimates of attributes are all significant.
This showed that overall 84.2% of prediction by the model was true.
The classification table gives the overall percent of cases that are correctly pre-
dicted by the model (in this case, the full model that we specified). This percentage
has increased from 79.2% for the null model to 88.6% for the full model. The model
coefficients show the importance and strengths of urban factors and their ability to
improve the explanatory power of behavioural models. All analysis was done by
358 A.Soltanietal.
Tabl e 1 Variables in the equation
Var i a b l e s BS.E. Wal d df Sig. Exp(B)
Retired/lost job (dummy) 4.267 1.342 10.111 10.001 71.310
Age_16_29 (dummy) 3.875 1.369 8.018 10.005 48.195
Flat, apartment or unit (dummy) 4.433 1.390 10.176 10.001 84.218
Postgraduate_study (dummy) −1.585 0.371 18.294 10.000 0.205
Safety and/or personal security 0.349 0.149 5.460 10.019 1.418
Cost savings 0.601 0.172 12.219 10.000 0.548
CarUsers_Movedhouse 0.969 0.490 3.915 10.048 0.379
CarUsers_Changedjob 1.912 0.475 16.225 10.000 0.148
Possessing_DrivingLicense 2.417 0.655 13.638 10.000 0.089
Non_Motorised −2.925 0.733 15.939 10.000 0.054
Constant 1.935 1.494 2.677 10.095 6.927
SPSS ver. 22.0, produced by IBM. The model, and the values of attribute coeffi-
cients, their significance and the Wald values and Exp(B) as the measure of elasticity
are detailed in Table 1.
4 Results and Discussion
This model shows that two reasons were highly significant in making a shift in
modal choice. One was the change in home location (wald =3.915, p< 0.048)
and the other one was job change (wald =16.225, p< 0.000). These important
findings are in line with previous research as noted in the Background section of
this chapter. A brief study of modal shift for journey to work in Australian cities
found that between 2011 and 2016, journey to work public transport modal shares
went up significantly in Melbourne and Sydney but dropped significantly in Perth,
Brisbane and Adelaide. Private transport modal shifts did the opposite. The main
drivers of change included the changing distribution of jobs within cities; changes
in transport costs; increases in workplace density; (negative) growth in the cost of
“private motoring” (including vehicles, fuel and maintenance); changes in car parking
costs and changes in population distribution (Loader 2018).
Similarly, Song et al. (2017) found that change in employment status could affect
modal change from private car to non-motorised choices if the distance to job was
reduced. Changes in home location could affect car usage as found by Bamberg
(2006). Santos et al. (2010) noted that people who experience a substantial life
change are increasingly expected to respond to changes in the relative utility of
different travel modes. When people change their work or residence places, in fact,
they change their travel behaviour to adapt to the new conditions (Song et al. 2017).
19 Developing a Behavioural Model for Modal Shift in Commuting 359
Changes in commuting mode are a function of changes in life, changes in cost and
marketing (Clark et al. 2016; Kroesen 2014).
Not all socio-demographic variables are associated with the probability of making
a transition in modal choice for commuting. Gender, income, household size, ethnic
background, home-ownership and type of job as the main socio-economic charac-
teristics in our study, did not appear to be associated with increased or decreased
probability of making a modal shift. Kroesen (2014) in a similar German study
found that gender is not an important factor affecting modal shift. He argued that
males and females have become more equal regarding employment conditions thus
making their travel activity patterns similar. However, the findings here are partially
contradictory with similar studies in rest of the world.
Results of data analysis indicated that young adults in Adelaide lean towards car
commuting in their early stages in the labour force. This finding is contradictory to
some studies which found older adults (aged 50–59) have higher tendency to change
the mode of commuting (Chatterjee et al. 2016). However, our finding is in line with
Clark et al. (2016) confirming that the younger generation is more likely than other
age groups to switch towards car commuting. In a European study (in Netherlands),
it was found that younger people are also more likely to switch from car usage to the
bicycle or public transport (Kroesen 2014).
Another notable finding from the survey are that holders of a driving licence were
more likely to change their mode of travel to cars. This can be explained by the fact
that those being certified as driving licence holders are more likely to get access to
a motor vehicle and change their commuting mode to a vehicular option.
Surveyed participants who were residents of apartments/flats/units tended to
switch from one particular mode to another compared to those who live in other
dwelling types. This can be explained by parking space limitations associated nor-
mally with apartment living especially in the inner suburbs or central Adelaide area.
Those retired or who lost their jobs within last three years are more likely to
move to non-car and cheaper modes (perhaps explained by these individuals wanting
the flexibility in choosing public transit or walk/cycling to fulfil lesser activity and
lifestyle needs). Interestingly, Clark et al. (2016) discussed that these groups have
less obligation to commute at certain times, instead, they are more flexible to choose
other modes included non-car choices.
Data analysis using the questionnaire survey also pointed out that those who had
higher education (a postgraduate degree) were less likely than other educational
groups to switch to other modes for commuting, which can be explained by the
fact that a highly educated group normally have more fixed jobs and residential
locations that do not require them to change their mode of commuting. One reason
is that highly-educated people are likely to have higher incomes and thus travel
more by private vehicles (Brand et al. 2013; Thornton et al. 2011). In contrast, some
argue that having higher academic qualifications may change their personal attitudes
towards the environment and lead to reducing car usage (Van Denderand and Clever
2013). Interestingly, some former studies have found that highly-educated adults
are inclined to have more pro-environmental attitudes but choose less sustainable
transport options (Anable et al. 2006).
360 A.Soltanietal.
Two travel-related personal factors were found to impact modal change: one is
attitudes to safety and another one is attitudes towards the cost. The former has
positive effect on modal shift showing that higher expectation from car use leads
to higher likelihood of modal change. On the other hand, a greater attention to the
cost is associated with lower probability of modal shift. This finding confirmed the
role of perceived factors in affecting behaviour. According to Prochaska’s models
(Prochaska and DiClemente 1986; Prochaska et al. 1994), the change of behaviour
is a deliberate procedure which needs constant consideration.
A positive attitude to safety and a negative attitude to the cost when choosing
a mode appeared across respondents and is at least an important starting point for
behaviour change if more reliable, safer and cheaper options are provided.
Relocation of home or job is the main determinant of commuting modal shift as
discussed in several studies (Chatterjee et al. 2016; Santos et al. 2010;TfL2010; Song
et al. 2017). Our study found this significant only for those who currently use a car
and experienced a modal shift (from non-car to car commuting). Clark et al. (2016)
suggested that presenting travel information packs explaining accessible transport
options within the neighbourhood area would be an appropriate strategy for those
who have recently moved to an area.
The use of discrete choice modelling in this study assists in pointing out the
determinants of modal shift among surveyed participants in Adelaide. Indeed, one
interesting finding of our model is that the value of Exp(B) parameter as the index
of elasticity for home relocation (0.379) is 2.6 times larger than the elasticity for job
relocation (0.148) confirming the stronger impact of home location on the mode of
commuting. In this study, the correlation between positive attitudes to the environ-
ment and job changes was also examined but there was no statistically significant
association.
5 Conclusion
5.1 Significance and Policy Implications
The significant (albeit different in quantity) impacts of moving house and changing
jobs showed that these actions are crucial in defining the patterns of activity, therefore,
any policy in jobs and housing distribution as exogenous factors would significantly
affect travel patterns.
One important aspect which was neglected in our study is distinguishing between
car usage as the driver or passenger. In fact, those who wish to decline driving but
increase car usage as passengers, are potential customers of car-sharing or ridesharing
options. Most of the findings of this study support previous research.
This study also presented some new visions which may be beneficial when trying
to encourage more people to take shared mobility options (more often). Most notably,
it was shown that neither all car users nor all non-car users are the same which
19 Developing a Behavioural Model for Modal Shift in Commuting 361
may have significant implications for sustainable mobility policies. The positive
association between being in a young category (17–29 years old) with the likelihood
engaging in modal shift, opens up an opportunity to consider this group as a preferred
target. As many from the younger generation, including students, cannot afford to
buy or use a car and are less willing to use infrequent bus services, for many of them
sharing mobility services can be a reliable and flexible option.
Our survey found that about 28.8% choose multimodal options rather than a
single mode to reach their destination. A trip-maker who takes multiple modes can
be viewed as a thoughtful choice maker, while an individual who solely chooses
a single mode is more probable to be a habitual travel maker (Kroesen 2014). By
contrast, single-modal persons are more likely to be stable commuters and less likely
to respond to behavioural change measures/actions.
A varied set of transport policies and strategies addressing different socio-
economic groups is required to be adopted to increase the share of sustainable modes
in the short and the long term.
Short term strategies include identifying and supporting those who have already
used non-motorised or public transport infrequently. In the longer term, bringing
jobs closer to homes and encouraging job concentration in centres and physical
improvements such as increasing the coverage of safe cycling routes within central
Adelaide area are suggested. Increasing public knowledge of the carbon footprint of
their travel through general campaigns and media are essential.
Former studies advocating for smarter planning through Transit-Oriented Devel-
opment (TOD) or Traditional Neighbourhood Development (TND) approaches claim
that the local built environment has a significant facilitating role to play in encour-
aging commuting by non-motorised modes when people relocate their job or home.
This is apparent from our study, since central Adelaide comprises 20% of jobs and the
local government jurisdiction in Adelaide has the highest rate of walking and cycling
(37%). In addition to ensuring reasonable accessibility to workplaces, the quality of
the built environment, especially with regard to the presence of non-residential land
uses and having safer footpaths and cycling routes, qualifies the Adelaide CBD as
having amongst the best areas to walk/cycle for employees (comparing with the rest
of metropolitan area).
Furthermore, more reliable and efficient public transport, where the journey time
to workplaces can be reduced, can be a good option for those commuters wish to
change their mode. In fact, having the right mix of urban planning and transport
strategies which target a mixed-development, that is well serviced by public trans-
port, can be effective in achieving a modal shift from cars to more environmentally
sustainable travel options.
Acknowledgements The survey for this research was funded by CRC Low Carbon Living research
node and received the ethics approval from the Research Department of the University of South
Australia (Ethics protocol: Major Trip Generators Survey, ID #200525). The authors acknowledge
the support from CRC for Low Carbon Living.
362 A.Soltanietal.
Appendix 1: UniSA Travel Behaviours Questionnaire
Section A: Attitudes to Mobility Options
A1 [ASK ALL] How important are environmental issues (for instance, CO2emis-
sions) for you when it comes to selecting a mode of transport within Adelaide
CBD area? SR
Code Response Routing
1Very important Continue
2Important
3Moderately important
4 Slightly important
5Not important
A2 [ASK ALL] How do you rate the following criteria when choosing a transport
mode? (Please rate on a scale of 1 to 5 where 1 =not at all important and 5 =
very important).
Code Response (1)
Not
important
(2)
Slightly
important
(3)
Moderately
important
(4)
important
(5)
Very
important
Routing
1Comfort Continue
2Convenience
and/or
practicality
3Safety and/or
personal
security
4 Cost savings
5Speed
6 Time savings
7Health
8 Exercise
9Travel
distance
10 Independence
11 Status/image
19 Developing a Behavioural Model for Modal Shift in Commuting 363
The following questions are asked for those who indicated code 1,11,12 at A2vi.
For respondents who did not use a car to get the destination skip to A7.
A3 [ASK IF CODE 1,11,12 AT A2vi] Thinking about your most recent trip to
____________ [USE CODE AT A2] that you made by CAR, what are the main
reasons you used your car to get to/from this destination (choose up to 3)? MR
Code Response Routing
1 Saving in time Continue
2Convenience and/or comfort
3Flexibility and/or Reliability
4 Safety and/or Personal security
5 Easy to find a park
6Habit
7 Health/physical condition
8Independence/status
9 Children and/or family issue
10 Lack of alternative
11 Don’t like other modes e.g. public
transport/walking/cycling
12 Other (please specify)
A4 [ASK IF CODE 1,11,12 AT A2vi] And still thinking about your most recent
trip to _______________ [USE CODE AT A2], where did you park your CAR
at that time?
Code Response Routing
1 Private parking area Continue
2Off-street public parking area
3On-street public parking area
4Public parking garage (car parking
structure)
5 Other (please specify)
364 A.Soltanietal.
A5 [ASK IF CODE 1,11,12 AT A2vi] And how did you get from your CAR PARK
to your final destination for that most recent trip?
Code Response Routing
1Wal k ed Continue
2 Used public transit
3 Taxi/Uber
4Cycled
5None, I parked at the
destination itself
A6 [ASK IF CODE 1,11,12 AT A2vi] And on a scale of 1–10 where 1 is very
unlikely and 10 is very likely how likely would you be to use any of the following
options to access this location in the future rather than taking a personal car?
Code Type 1=Ver y
unlikely
2 3 4 5 =Ver y
likely
Routing
1Shared bike
(OfO)
Continue
2Shared bike
(O’Bike)
3Shared bike
(Adelaide
CityBike)
4Your own
bike
5UBER
6GoGet
sharing car
7 Driverless
autonomous
car
8Free City
bus
9Free tram
10 Other
(please
specify)
19 Developing a Behavioural Model for Modal Shift in Commuting 365
A7 [ASK ALL] Have you changed the primary mode of transport by which you
travel to work in the last two years at all? SR
Code Response Routing
1Yes Continue
2No Skip to B11
3Unsure Skip to B11
A8 [Ask if YES at A7] What was the main reason for changing your mode of
transport? SR
Code Response Routing
1 Relocation of job Continue
2 Relocation of home Continue
3 Other (please specify) Continue
Section B: Personal and Household Information
B1 What is your gender SR
Code Response Routing
1Male Continue
2Female
3 Prefer not to say
B2 What category of age are you in? SR
Code Response Routing
117–19 Continue
220–24
(continued)
366 A.Soltanietal.
(continued)
Code Response Routing
325–29
430–34
535–39
640–44
745–49
850–54
955–59
10 60–64
11 65–69
12 70–74
13 75–79
14 80–84
15 85 and over
16 Prefer not to say
B3 What is your employment status? SR
Code Response Routing
1Working full time (35 +hours per
week)
Continue
2 Working part time (less than 35 h per
week)
3 Casual worker
4Working from home
5Not working (e.g., stay at home parent)
6Seeking for job
7Student (and not working)
8Retired
9 Other (please specify)
B4 What is your highest level of education? SR
19 Developing a Behavioural Model for Modal Shift in Commuting 367
Code Response Routing
1None Continue
2Primary School level
3High School Certificate
4Undergraduate University degree
5Postgraduate University degree
6 Other (please specify) …….
B5 What is your residency status? SR
Code Response Routing
1 Australian born Continue
2 Australian resident or citizen (born
overseas)
3Short term non-Australian resident (on a
student visa)
4 Visiting (tourist)
5 Other (please specify)
B6 How would you describe your home? SR
Code Response Routing
1Separate house Continue
2Semi-detached, row or terrace house,
townhouse etc.
3 Flat or apartment
4 Other (please specify)
B7 Which of the following categories best describes your weekly personal pre-tax
income? SR
Code Response Routing
1Nil income Continue
2$1–$199
(continued)
368 A.Soltanietal.
(continued)
Code Response Routing
3$200–$299
4$300–$399
5$400–$599
6$600–$799
7$800–$999
8$1000–$1249
9$1250–$1499
10 $1500–$1999
11 $2000 or more
B8 What is the size of your household? SR
Code Response Routing
11, just me Continue
2 2
3 3
4 4
55ormore
B9 How many registered cars are available at your household? SR
Code Response Routing
1 0 Continue
2 1
3 2
4 3
54ormore
19 Developing a Behavioural Model for Modal Shift in Commuting 369
B10 Do you use a smartphone or tablet (e.g., an iPad) for transport purpose (pub-
lic transit application; bike share application, check up on a bus/tram/train
timetable or route map)? SR
Code Response Routing
1Yes Continue
2No
3Unsure
B11 What is the name of the suburb and Street where you live? OE
Suburb __________________________________________________
Street ________________________________________________
References
Adelaide City Council (2017) City of Adelaide community GHG emissions inventory. Accessed 28
May 2018 from http://dmzweb.adelaidecitycouncil.com/agendasminutes/files08/Attachments/
Council_27_March_2018_Item_12.4_Link_1.pdf
Anable J, Lane B, Kelay T (2006) An evidence base review of public attitudes to climate change
and transport behaviour. The Department
Anwar AM, Yang J (2017) Examining the effects of transport policy on modal shift from private
car to public bus
Australian Bureau of Statistics (ABS) (2007) Greenhouse gas and energy usage 2010. Canberra
Australian Bureau of Statistics (ABS) (2011) Employment and unemployment. Accessed 28 May
2018 from http://www.abs.gov.au/Employment-and-Unemployment
Australian Bureau of Statistics (ABS) (2016) Employment and unemployment. Accessed 28 May
2018 from http://www.abs.gov.au/Employment-and-Unemployment
Australian Bureau of Statistics (ABS) (2017) 2016 Census QuickStats. Accessed 28 May 2018 from
http://quickstats.censusdata.abs.gov.au/census_services/getproduct/census/2016/quickstat/036
Bamberg S (2006) Is a residential relocation a good opportunity to change people’s travel behavior?
Results from a theory-driven intervention study. Environ Behav 38(6):820–840
Ben-Akiva ME, Lerman SR, Lerman SR (1985) Discrete choice analysis: theory and application to
travel demand (Vol. 9): MIT press
Brand C, Anable J, Tran M (2013) Accelerating the transformation to a low carbon passenger
transport system: the role of car purchase taxes, feebates, road taxes and scrappage incentives in
the UK. Transp Res Part A Policy Pract 49:132–148
Buehler R, Hamre A (2014) National trends in multimodal individual travel behavior in the USA
Cao XJ, Mokhtarian PL, Handy SL (2009) The relationship between the built environment and
nonwork travel: a case study of Northern California. Transportation Research Part A: Policy
Pract 43(5):548–559
Chatterjee K, Clark B, Bartle C (2016) Commute mode choice dynamics: accounting for day-to-day
variability in longer term change. Eur J Transp Infrastruct Res 16(4):713–734
City of Adelaide (2016) Carbon neutral Adelaide. Accessed 20 Oct 2018 from https://www.
cityofadelaide.com.au/city-living/sustainable-adelaide/carbon-neutral-adelaide/
370 A.Soltanietal.
Clark B, Chatterjee K, Melia S (2016) Changes to commute mode: the role of life events, spatial
context and environmental attitude. Transp Res Part A Policy Pract 89:89–105
Dargay J, Hanly M (2007) Volatility of car ownership, commuting mode and time in the UK. Transp
Res Part A Policy Pract 41(10):934–948
Diana M (2010) From mode choice to modal diversion: a new behavioural paradigm and an appli-
cation to the study of the demand for innovative transport services. Technol Forecast Soc Chang
77(3):429–441
Diana M, Mokhtarian PL (2009) Desire to change one’s multimodality and its relationship to the
use of different transport means. Transp Res Part F Traffic Psychol Behav 12(2):107–119
Fishman E, Washington S, Haworth N (2014) Bike share’s impact on car use: evidence from the
United States, Great Britain, and Australia. Transp Res Part D Transp Environ 31:13–20
Fuller D, Gauvin L, Kestens Y, Morency P, Drouin L (2013) The potential modal shift and health
benefits of implementing a public bicycle share program in Montreal, Canada. Int J Behav Nutr
Phys Act 10(1):66
Goodman A, Sahlqvist S, Ogilvie D, iConnect Consortium (2014) New walking and cycling routes
and increased physical activity: one-and 2-year findings from the UK iConnect study. Am J Public
Health 104(9):e38–e46
Graham-Rowe E, Skippon S, Gardner B, Abraham C (2011) Can we reduce car use and if so, how?
A review of available evidence. Transp Res Part A Policy Pract 45(5):401–418
Heinen E, Panter J, Mackett R, Ogilvie D (2015) Changes in mode of travel to work: a natural
experimental study of new transport infrastructure. Int J Behav Nutr Phys Act 12(1):81
Heinen E, Harshfield A, Panter J, Mackett R, Ogilvie D (2017) Does exposure to new transport
infrastructure result in modal shifts? Patterns of change in commute mode choices in a four-year
quasi-experimental cohort study. J Transp Health 6:396–410
Hensher DA (2008) Climate change, enhanced greenhouse gas emissions and passenger trans-
port–What can we do to make a difference? Transp Res Part D Transp Environ 13(2):95–111
Idris AO, Habib KMN, Shalaby A (2015) An investigation on the performances of mode shift
models in transit ridership forecasting. Transp Res Part A Policy Pract 78:551–565
Klinger T, Lanzendorf M (2016) Moving between mobility cultures: what affects t he travel behavior
of new residents? Transportation 43(2):243–271
Kroesen M (2014) Modeling the behavioral determinants of travel behavior: an application of latent
transition analysis. Transp Res Part A Policy Pract 65:56–67
Lanzendorf M (2010) Key events and their effect on mobility biographies: the case of childbirth.
Int J Sustain Transp 4(5):272–292
Loader C (2018) Suburban employment clusters and the journey to work in Australian cities. Chart-
ing transport. Accessed 20 Oct 2018 from https://chartingtransport.com/2018/07/08/suburban-
employment-clusters-and-the-journey-to-work-in-australian-cities/
Martin EW, Shaheen SA (2014) Evaluating public transit modal shift dynamics in response to
bikesharing: a tale of two US cities. J Transp Geogr 41:315–324
Nguyen HA, Allan A (2017) The extension of Adelaide’s tram line to the City’s East End and its
implications for building a sustainable urban transport system in Adelaide’sCBD. In: International
conference on computers in urban planning and urban management, Adelaide, Australia
Oakil A, Ettema D, Arentze T, Timmermans H (2011) A longitudinal analysis of the dependence of
the commute mode switching decision on mobility decisions and life cycle events. Paperpresented
at the Transportdynamics: Proceedings of 16th international conference of Hong Kong Society
for Transportation Studies (HKSTS). Hong Kong Society for Transportation Studies, Hong Kong
Panter J, Griffin S, Dalton AM, Ogilvie D (2013) Patterns and predictors of changes in active
commuting over 12 months. Prev Med 57(6):776–784
Perone JS, Volinski JM (2003) Fare, free, or something in between?
Pradono P, Kusumantoro IP, Retapradana A (2015) Towards carbon dioxide emission reduction
through public transport revitalization: case of proposed surabaya tram transit project. J Eastern
Asia Soc Transp Stud 11:967–978
19 Developing a Behavioural Model for Modal Shift in Commuting 371
Prochaska JO, DiClemente CC (1986) Toward a comprehensive model of change. Treatingaddictive
behaviors. Springer, Heidelberg, pp 3–27
Prochaska JO, Velicer WF, Rossi JS, Goldstein MG, Marcus BH, Rakowski W, Rosenbloom D
(1994) Stages of change and decisional balance for 12 problem behaviors. Health Psychol 13(1):39
Rabl A, De Nazelle A (2012) Benefits of shift from car to active transport. Transp Policy
19(1):121–131
Rafat R, Mirhadi S, Sharifi E, Soltani A (2018) The evaluation of urban development project by
approach of landscape urbanism tenets: the case study of Bowden, Adelaide. Paper presented at
the State Of Australian Cities, Adelaide, Australia
Ramadurai G, Srinivasan K (2006) Dynamics and variability in within-day mode choice decisions:
role of state dependence, habit persistence, and unobserved heterogeneity. Transp Res Rec J
Transp Res Board 1977(1):43–52
Redman L, Friman M, Gärling T, Hartig T (2013) Quality attributes of public transport that attract
car users: a research review. Transp Policy 25:119–127
Santos G, Behrendt H, Teytelboym A (2010) Part II: policy instruments for sustainable road trans-
port. Res Transp Econ 28(1):46–91
Scheepers C, Wendel-Vos G, Den Broeder J, Van Kempen E, Van Wesemael P, Schuit A (2014)
Shifting from car to active transport: a systematic review of the effectiveness of interventions.
Transp Res Part A Policy Pract 70:264–280
Shaheen S, Martin E (2015) Unraveling the modal impacts of bikesharing. Access 47:8–15
Shaheen S, Martin E, Cohen A (2013) Public bikesharing and modal shift behavior: a comparative
study of early bikesharing systems in North America. Int J Transp 1(1):35–54
Shirzadi Babakan A, Alimohammadi A, Taleai M (2015) An agent-based evaluation of impacts of
transport developments on the modal shift in Tehran, Iran. J Dev Effectiveness 7(2):230–251
Song Y, Preston J, Ogilvie D, Consortium. I (2017) New walking and cycling infrastructure
and modal shift in the UK: a quasi-experimental panel study. Transp Res Part A Policy Pract
95:320–333
Steinbach R, Green J, Datta J, Edwards P (2011) Cycling and the city: a case study of how gendered,
ethnic and class identities can shape healthy transport choices. Soc Sci Med 72(7):1123–1130
Termida NA, Susilo YO, Franklin JP (2016) Observing dynamic behavioural responses due to the
extension of a tram line by using panel survey. Transp Res Part A Policy Pract 86:78–95
Thornton LK, Pearce JR, Kavanagh AM (2011) Using geographic information systems (GIS) to
assess the role of the built environment in influencing obesity: a glossary. Int J Behav Nutr Phys
Act 8(71):1–9. http://HDR.handle.net/10536/DRO/DU:30044447
Train KE (2009a) Discrete choice methods with simulation. Cambridge UniversityPress, Cambridge
Train KE (2009b) Discrete choice methods with simulation. Cambridge University Press, Cam-
bridge
Transport for London (TfL) (2010) Modal shift triggers in Richmond. Accessed 28 May 2018 from
http://content.tfl.gov.uk/modal-shift-triggers-in-richmond-presentation.pdf (TfL 2010)
Van Dender K, Clever M (2013) Recent trends in car usage in advanced economies: slower growth
ahead?; summary and conclusions
Yang J, Shen Q, Shen J, He C (2012) Transport impacts of clustered development in Beijing:
compact development versus overconcentration. Urban Stud 49(6):1315–1331
Chapter 20
Planning for Safer Road Facilities
for Bicycle Users at Junctions
Li Meng, Li Luo, Yanchi Chen and Branko Stazic
Abstract Promoting bicycle travel in a city where people are used to driving private
cars is a difficult task. Safety is a major factor discouraging people from cycling
and it needs to be addressed in order to achieve any significant mode shifts away
from private cars. One suggestion that has been put forward to improve the safety
of cyclists is to provide a separate area for bikes at junctions. This study reviewed
junction design and traffic flow conditions at an upgraded junction in Adelaide,
Australia that contained bicycle signals and a storage zone. It was found that the
bicycle lane could be designed in such way that would separate left and right turning
bikes into two sections. Also, the provision of blue bicycle crossing lanes has a
potential to improve cyclist safety by warning pedestrians and motorists of possible
cyclist presence. The study also recommends smarter data collection and better traffic
modelling to help test improved infrastructure and enable the development of better
policies regarding the safety of cyclists.
Keywords Bicycle ride safety ·Intersection design ·Bicycle policy ·Bicycle
infrastructure
L. Meng (B)·L. Luo ·Y. C h e n
School of Natural and Built Environments, University of South Australia,
Adelaide, SA, Australia
e-mail: li.meng@unisa.edu.au
L. Luo
e-mail: luoly010@mymail.unisa.edu.au
Y. C h e n
e-mail: cheyy188@mymail.unisa.edu.au
B. Stazic
College of Science and Engineering, Flinders University, Adelaide, Australia
e-mail: branko.stazic@flinders.edu.au
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_20
373
374 L. Meng et al.
1 Introduction
Bicycle rider safety has been claimed as the most significant barrier to encourage
people to ride a bike (Loidl and Hochmair 2018). This is especially the case in low
density and car-oriented cities. When people are asked how they could be made to
feel safer, some suggest that there should be separate bicycle routes provided. Many
on-street cycling routes are constructed, but the majority of bicycle lanes end before
the intersection. Intersections as a part of the cycling route network are a high-risk
area for accidents. For example, in Copenhagen, it was found the construction of
cycling routes has contributed to a drop in the total number of accidents and injuries
on the road by 10 and 4% respectively, while incidents have increased significantly
at junctions by 18% (Jensen et al. 2007). Junctions as shared zones enable left/right
turning for both cars and bicycles and also present areas of multiple conflict points
between motorized traffic and cyclists.
There are many papers discussing bicycle safety in shared spaces at roundabouts
and intersections (e.g. noted in the review of Reynolds et al. 2009), while the junction
type of intersection shared zone is often overlooked. Junction structure design should
arouse awareness of the safety risk and help adapt the behavior that considers all other
traffic users. The situation is severe in low density cities, such as Australia’s cities,
where roads are originally designed primarily for cars, and the promotion of bicycle
usage is hampered by travelers’ perceptions of cycling safety and real safety.
This research analyses junction structure and utilization for safe cycling and aims
to improve intersection design and planning for a bicycle friendly environment.
The chapter starts with a literature review on shared space zone issues at inter-
sections and examines the World’s best designs and practices. It follows with a case
study to illustrate real problems at intersections for bicycle riders. The chapter then
discusses solutions to help improve the safety of intersections for cyclists and ends
with a conclusion and future directions.
2 Literature Review
In order to improve planning and design at intersections to make cycling safer, there
is a need to review related issues and find possible solutions to handle the desired
increase in bicycle usage while providing more safer shared spaces on road (Chong
et al. 2010).
Shared space presents a new approach for improving street design, road safety
and traffic flow (Hamilton-Baillie 2008), which is a design feature that serves the
purpose of encouraging cyclists, pedestrians and motor vehicle users to share the
same deregulated space (Hammond and Musselwhite 2013).
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 375
2.1 Intersection for Bicycles
Road intersections represent configurations encountered while changing directions
or traversing road networks (Ibanez-Guzman et al. 2010). Intersection accidents fre-
quently happen and cause severe consequences (Sander 2017). Collisions between
bicycle and motor vehicles in these locations have resulted in severe mortality and
property losses globally with the most of the bicycle-motor vehicle (BMV) collisions
occurring at intersections (Wang and Nihan 2004). The casualty factors of BMV col-
lisions currently in use are various. Schepers et al. (2011) classified these factors into
two types addressing the shortcomings of human drivers, which are bicycle related
collisions and vehicle related collisions. An educational method to reduce bicycle
related collisions has been proposed by Kosaka and Noda (2015), who considered a
new approach to ride bicycles safely by improving the behavior of cyclists to increase
adherence to regulations and decrease the bicycle-related intersection accidents.
Cyclists have a high level of traffic stress (Wang et al. 2016). Studies suggest that
one of the causes of bicycle traffic stress or travel stress is perceived safety issues
(Pucher et al. 1999). Potential safety hazards at intersections can cause travel stress
for bicycle.
users who are vulnerable road users, and account for a significant number of
traffic-related injuries (Chong et al. 2010). These injuries can deter people from
undertaking cycling activities as a means of both recreation and transport. Different
treatment at intersections can affect the behavior of bicycle users at these intersections
(Alexander 2015).
In the past thirty years, the development of autonomous vehicles for transport
has seen rapid growth (Campbell et al. 2010). Autonomous vehicles have a positive
impact on intersection capacity and level-of-service (Le Vine et al. 2015), which can
benefit cyclists in shared spaces. These types of vehicles have a potential to increase
the perceived cyclist safety due to reduction in vehicle driver error.
2.2 Better Intersection Design for Safe Cycling
Intersection signal phasing design is an important step that can influence the final
intersection geometric design and traffic flow control. Traffic congestion occurs fre-
quently at intersections and it is particularly pronounced when traffic flow exceeds
the capacity of an intersection which often happens in peak hours. This congestion
can result in additional conflicts in shared spaces, which can adversely affect the
safety of an intersection. Further, Sun et al. (2015) proposes some traditional miti-
gation strategies to increase intersection capacity by merely adjusting signal control
parameters including well-recognized CFI (continuous flow intersection) designs.
Road and intersection design can affect both vehicle drivers and cyclists. There are
many studies and guidelines for appropriate intersection design for elderly drivers
(Oxley et al. 2006), and young and senior adult drivers (Shechtman et al. 2007),
376 L. Meng et al.
which help drivers to improve their driving manners and also benefit bicycle users
in shared space. Intersection line marking in combination with intersection paving
material design is one intersection design aspect that can affect the traffic safety.
Pavement markings can be used to make the bicycle users more visible, illustrate
the bicycle travel area at the intersection (Yan et al. 2007) and enhance the safety of
bicycle users.
Evaluation of the safety effectiveness of intersections is important. One approach
involves a well-designed before-and-after evaluation to improve intersection design,
which uses data including geometric design, traffic control, traffic volume and traffic
accident data (Harwood et al. 2003). Another study employed operational character-
istics to allow for a preliminary evaluation of a wider range of design possibility (Kirk
et al. 2011). An effective evaluation approach can contribute to better intersection
design and improve overall intersection safety.
2.3 Transport Policy
Transport policies and policy related aspects of intersections can be classified into
three broad categories, which are physical policies with a physical infrastructure
element (e.g. public transport, cycling, intersection construction), soft policies (e.g.
behavioral change by informing road users about the potential risk of their transport
choices) and knowledge policies (Santos et al. 2010). In Santos et al.’s study, physical
policy ‘Walking and cycling’ includes crime reduction to make streets safer, clean
pavements and clear marking, safe crossings at intersections with shorter waiting
times and lower speed limits of motor vehicle at intersections aiming to improve
bicycle and other road users’ safety. Further, this study also addresses soft policies.
Within these policies, ‘Information and education’ policies are advocated as tools of
behavioral change, and ‘Advertising and marketing’ may be sufficient in changing
people’s behavior (Santos et al. 2010). Cyclists’ behavior directly contributes to their
safety condition.
Transport policy in different areas depends on local governments, transport depart-
ments or some other authorities. When introducing or conducting transport policy,
there is a need to consider safety issues. On the one hand, city authorities make deci-
sions regarding intersection control involving new technology, managing systems
and intelligent strategies to achieve safer and more effective intersection environ-
ment (Ahmane et al. 2013), which is advantageous to bicycle users. On the other
hand, in early 2007, the Taiwanese government introduced an idling stop policy at
intersections for vehicles. The main purpose of this policy is to reduce carbon emis-
sions, but turning off engines while stopped at the intersection could be dangerous
to intersection users (Jou et al. 2011).
Some cities in Europe, particularly in the Netherlands, Denmark and Germany,
have the highest bicycle usage level in the world, and they have advanced experience
in policy-making to promote cycling and ensure cyclists’ safety to get around cities of
virtually any size (Pucher and Buelher 2007). To elaborate, cycling-supported policy
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 377
in these countries includes intersection treatments and traffic priority for cyclists. For
example, the city of Groningen has applied four-way green lights for bicycle users,
allowing safer and more rapid crossings at intersections for cyclists.
2.4 Traffic Management Software/Systems
Current traffic management software and systems mainly focus on vehicle users and
seldom pay attention to cyclists and pedestrians. Intersection decision support (IDS)
system aims to assist drivers to make a suitable and safer gap acceptance decision at
suburban intersections (Laberge et al. 2006). However, suitable traffic management
software and systems can also benefit all road users including cyclists. Since the mid-
1980s, agent or multi-agent systems have been introduced and evolved rapidly and
widely employed in controlling traffic management systems, which achieve effec-
tive real-time traffic applications (Fei-Yue 2005). Among these systems, the most
well-known control systems are CRONOS, OPAC, SCOOT, SCATS, PRODYN and
RHODES. All of these systems can be considered to be a part of the intelligent trans-
portation systems. A further example of an advanced transportation management
system is Georgia NaviGAtor (Guin et al. 2007). Effective control or management
systems play a crucial role in incident management, and managing traffic flow and
congestion in intersections (Guin et al. 2007), which can improve the overall safety
condition of cyclists at intersections. Intelligent traffic management systems pro-
vide policy makers a tool for policy management in improving cyclists and other
road users’ safety level (Gregory et al. 2004). With the development of modern-
day technologies, there are many other traffic management software and systems
assisting cyclists to cross the intersection. Wireless sensor networks (WSN) using
self-powered sensing tools are interconnecting by wireless ad hoc technologies (Pas-
cale et al. 2012). This increasing availability of sensing networks in urban districts
now provides the opportunity to conduct continuous evaluations of transport sys-
tems (Lathia et al. 2012). Cycling and shared bicycle schemes (Shaheen et al. 2010)
equipped with sensors are also a part of an intelligent traffic management system,
which can in turn improve traffic management system (Lathia et al. 2012).
Intersections are identified as locations where two or more roads meet or cross
at the same level. There is a significant safety gap at intersections between bicycle
users and other road users such as pedestrians and motor vehicle users (Reynolds
et al. 2009). Such significant differences in safety make intersections design more
complex. Many studies show that cyclists have a relatively high risk of injury and
high level of traffic stress when crossing intersections. Possible solutions to this
safety issue could be better design of intersections, appropriate transport policy and
the application of suitable traffic management software and systems.
While it has been recognized that the vulnerable road user’s safety is an issue
at roundabouts the intersection studies have mainly concentrated on roundabouts
but not enough on the other type of junctions. The safe ride for cyclists has not yet
been achieved as a range of facilities lack of controlled risk exposure (Reynolds et al.
378 L. Meng et al.
2009). Transport policy is influenced by local governments’ decisions and local traffic
conditions, which need to be completely assessed and reconsidered if the reduction
in number of crashes and their severity is to be reduced. Further opportunity exists in
the advancement of traffic management software and systems that grow and evolve
rapidly. Another area of opportunity that exists but it is yet to be fully explored is the
Artificial Intelligent applications in traffic management systems.
3 Methodology
This study uses SIDRA Intersection, a software package, which was designed for
analyzing intersection (junction) and network capacity, level of service and perfor-
mance analysis, and signalized intersection and network timing calculations in order
to assist traffic road design and operations (Sidra Solutions 2019).
Cycle length is to demonstrate the traffic capacity of the intersection and to mini-
mize the overall delay (Webster 1958). Cycle time is described as the minimum time
in which a complete succession of signals occurs (Cantarella and Improta 1988). In
SIDRA, cycle time is calculated as the sum of green and lost times (the green-amber
period that is not used for departures) (Akcelik 1994).
The performance factor, delay, is the difference between travel time and the free
flow travel time (the time required by an unimpeded vehicle to survey section),
which is calculated as the total time taken to travel through the network section by
subtracting a ‘free’ travel time from the actual travel time (Taylor and Young 1988).
This study uses cycle time designs to estimate traffic delays that caused by bicycle
volume, intersection designs and influence at the intersection. The study discuses
traffic delays volume, intersection design, approach speeds and data issues.
4 A Case Study in Adelaide
Transportation infrastructure and cyclist safety require further investigation into a
range of facilities and methods to control risk exposure (Reynolds et al. 2009). The
shared space zone at the junction functions as a link node and redirection point in the
cycling network where all the road users pay high attention to each other’s intention
and travel direction. It is important to make the junction a comfortable, minimum
delay and safe shared space zone.
The study selected the intersection of Pulteney Street and Pirie Street in the Ade-
laide CBD as a case study area (see Fig. 1). This intersection is an important cross
of north-south and east-west cycling corridors, and in addition it is equipped with
bicycle count and signal control for western bound on Pirie St. The analysis of this
intersection that included the intersection lane design, traffic volume, turning maneu-
vers, stopped time and delays, has revealed how the junction can be better designed
in order to promote safe cycling.
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 379
Fig. 1 Pulteney St and Pirie St intersection (Google image)
The Pulteney Street and Pirie Street intersection has been upgraded with the
installation of bicycle signals and bike lane inductive loop detectors for Pirie Street
West approach. Figure 1shows the intersection layout and lane discipline following
the upgrade. The bicycle lanes were introduced on Pirie Street and the bicycles have
been given priority over cars by using the bicycle storage area for eastbound Pirie
Street traffic. It should be noted that there are no bicycle lanes provided on Pulteney
Street yet.
In order to understand better how bicycle volume impacts the intersection oper-
ation and performance, this study collected SCATS (Sydney Coordinated Adaptive
Traffic System) data. The data collected included both, vehicle counts and signal
phasing data. Figure 2shows the signal phasing diagram and the loop detector loca-
tions and numbering which provided counts for all cars and bicycles at the interchange
of the Pulteney and Pire Street intersection. In addition to SCATS data, some addi-
tional on-site surveys were undertaken that included bicycle counts, number of stops
and turning maneuvers in order to evaluate the accuracy of the SCATS loop detector
counts.
SCATS data has showed bicycle volume at the peak hour as 115 from 4:45 to
5:45 pm in day 1, while the onsite counting data revealed 142 cyclists (see Fig. 3).
There was an even bigger difference in day 2 at the same period, a lower number
of counts as 88 because the sensor system, compared with the observed number of
138. However, if looking the counting figure from 4 to 6 pm, the difference becomes
smaller. It can be concluded that the average difference has been over 30%.
380 L. Meng et al.
Fig. 2 Pulteney and Pirie Streets intersection SCATS screen dump
0
50
100
150
200
250
4:45-5:45pm 4:45-5:45pm 4-6pm 4-6pm
Day 1 Day2 Day 1 Day2
SCATS vs Observation
SCATS Observation
Fig. 3 The differences between SCATS data and observation data at Pirie Street West
Bicycle numbers only consisted of a very small proportion of overall traffic and
varied between 2 and 10% (see Fig. 4). Turning vehicles and bicycles account for
nearly a quarter to 30% of the overall traffic volume, which represents a potential for
an increased safety risks at the intersection.
For bicycle stopped times at the intersection, Fig. 5shows that more than half of
bicycles are stopped for less than 30 s, 16% proceed without stopping, 21% wait for
30 s–1 min and only a very small amount of people are waiting for 1–1 min 30 s.
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 381
0
100
200
300
400
500
600
700
800
900
Bicycle Car Bicycle Car Bicycle Car
Turn Left Straight Forward Turn Right
Bicycles and Car Volume at 4:45-5:45pm
Pulteney St South Pirie St East Pulteney St North Pirie St West
Fig. 4 Volume and movement of bicycles and cars at the intersection
Fig. 5 Bicycle stop time
duration at the Pirie Street
West intersection from 4 to
6pm 1%
21%
62%
16%
Bicycle Waiting Time at the Intersection
Wait for 1min -
1min30s
Wait for 30s-
1mins
Wait for 0-30s
Using SIDRA intersection design and evaluation software together with SCATS
data, the delays at the intersections were estimated. Since different cycle times lead
to different delay times, the existing cycle time of 100 s was used in order to keep
the existing traffic signal synchronization (green wave). Nevertheless, the effects of
different cycle lengths were evaluated see Fig. 6.
When adding observed bicycle data at Pirie Street West to the SIDRA model,
it is found that the optimized cycle time is 130 s. This optimized cycle time was
determined by the SIDRA software based on the principal of equalizing the degrees
of saturation for all the turning movement. Also, there are obvious changes in the
delays on all intersection approaches (Fig. 7).
The Pulteney Street and Pirie Street junction only provides limited design of
terminate bicycle lane and bicycle storage (a safety feature at an intersection to
allow bicycle riders to be more visible to drivers) for bicycle safety as the bike
lanes terminate before the intersection. Also, the traffic signals for bicycles are only
382 L. Meng et al.
0
20
40
60
80
100
120
80 100 120 140 160 180 200 220 240
Delay (Sec)
Cycle Time (Sec)
Pulteney St South
Pirie St East
Pulteney St North
Pirie St West
Fig. 6 Average SIDRA control delay for Pulteney/Pirie streets on a weekday PM peak
(4:45–5:45 pm) using SCATS traffic volumes
0
50
100
150
200
80 100 120 140 160 180 200 220 240
Delay (Sec)
Cycle Time (Sec)
Pulteney St South
Pirie St East
Pulteney St North
Pirie St West
Fig. 7 SIDRA Average control delay for Pulteney/Pirie streets on weekday PM peak
(4:45–5:45 pm) using observed bicycle data
mounted on one intersection approach. The bicycle storage provided at Pirie Street
West will enable a clear view of the intersection for cyclists and may bring some
advantage to reduce stopped time. Bicycle turning is perceived as a big safety risk for
cyclists (Jensen et al. 2007). There is also a lack of adequate bicycle number count
information to input into a traffic modelling software in order to conduct proper
intersection evaluation.
In addition to bicycle counts, there is a lack of other bicycle data, such as speeds,
travel times, acceleration/deceleration, queue lengths etc. that would be very useful in
model calibration and validation. It is found that the SCATS bicycle detector counts
are not accurate enough, with up to 30% mismatch with the observed bicycle counts.
5 Discussion
Improved bicycle safety at the shared space zone at junctions could encourage more
cyclists on the road. There are some solutions that can be applied through under-
standing local cycling issues, improving infrastructure, making suitable policies and
applying smart road management systems.
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 383
Fig. 8 Separate cyclist lanes for turning into two steps. Source T. Asperges, D. Dufour
Although midblock bicycle lanes may have accidents and injuries associated with
them, the complex and busy junction is still more dangerous and causes bicycle riders
significant stress and discomfort. Visible and clear road signs can help cyclists feel
safer. There needs to be more bicycle crossing lights, and they should be lowered on
the post to fit cyclists’ convenience. Cycling and the Law defines the shared zone
for road users and special attention is needed to each other to make travel safer
(Government of South Australia 2017).
Some intersections have no bicycle lane provided in which people could ride
freely in between or in front of cars for priority, which may be less safe. If bicycle
lanes had some additional structures for example: a bicycle storage area (bicycle
boxes) in front of cars, two steps for turning (both shown in Fig. 8, (PRESTO 2010))
and blue bicycle crossing (see Figs. 9and 10) cyclist safety could be improved. A
bicycle storage area is a safety feature at an intersection with traffic lights to allow
bicycle riders to be more visible to other road users. These facilities would improve
motorist vision of cyclists and reduce turning safety risks. Two step turning aims to
separate cyclists from road traffic and to do the left/right turn in two steps (PRESTO
2010). Blue bicycle crossings support hook turns which improve intersection safety.
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 385
Cycling law in South Australia requires motorists when turning or entering an
intersection to look for bicycle riders and give way as you would for any other
vehicle (Government of South Australia 2017). If a driver is turning left or right, the
driver must give way to any pedestrian or cyclist near the intersection on the road,
or part of the road the driver is entering.
The intersection design guide has included necessary design requirements (Aus-
troads 2017):
•Consider the need for bicycle lanes at intersections
•Avoid termination of bicycle lanes
•If provided, carry bicycle lanes through intersections on major roads at signalized
intersections
•Assess the need for bicycle storage
•Consider storage space for ‘hook turns’.
However, these are only recommendations and not requirements using language
such as ‘consider, if provided, and assess’. Implementation has then been subjected
to many other issues such as a lack of funds or alternative policy preferences.
One additional issue is that there is a lack of appropriate modelling software and
sufficient data to test the impact of intersection design changes. The SIDRA software
considers bicycle as equivalent to the car type, and there is no detailed data about
cyclist behavior of headway, acceleration and speed at the intersection.
Junctions should be designed to enable cyclists to turn left or right safely, speedily
and comfortably. Any new design can be modelled to test the impact on the road via
professional software. Future study can test the cycling safety and friendly designs
at the intersections and modelling the impact at the network level.
6 Conclusion
Road and policy designers should support a safer cycling network and provide the
most appropriate junction design solution. This study has found that intersections
would be safer if designed with blue bicycle crossings or if separate cyclist lanes were
provided for left and right turning cyclists. Also, the hook turning bicycle maneuvers
for right turners could be considered at some intersections. The design guide for
cycling has been well defined. However, in low density cities, some policy makers
still prefer to provide more convenience to cars rather than protect cyclists. On the
other hand, policy makers need better modelling tools to test new bicycle designs,
while there is a need to upgrade data collection systems and develop suitable bicycle
software. The road should provide consistent and safe environment at any single
section in the network, and then bicycle riding can be well promoted.
386 L. Meng et al.
References
Ahmane M, Abbas-Turki A, Perronnet F, Wu J, Moudni AE, Buisson J, Zeo R (2013) Modeling
and controlling an isolated urban intersection based on cooperative vehicles. Transp Res Part C
Emerg Technol 28:44–62
Akcelik R (1994) Estimation of green times and cycle for vehicle-actuated. Transp Res Rec
(1457):63–72
Alexander M (2015) Analyzing the behavior of cyclists at intersections to improve behavior vari-
ability within micro-simulation traffic models. UWSpace. http://hdl.handle.net/10012/9655
Austroads (2017) Guide to road design
Campbell M, Egerstdt M, How JP, Murray RM (2010) Autonomous driving in urban envi-
ronments: approaches, lessons and challenges. Philos Trans R Soc A Math Phys Eng
Sci 368(1928):4649–4672. http://rsta.royalsocietypublishing.org/content/roypta/368/1928/4649.
full.pdf
Cantarella GE, Improta G (1988) Capacity factor or cycle time optimization for signalized junctions:
a graph theory approach. Transp Res Part B Methodological 22(1):1–23
Chong S, Poulos R, Olivier J, Watson WL, Grzebieta R (2010) Relative injury severity among
vulnerable non-motorised road users: comparative analysis of injury arising from bicycle–motor
vehicle and bicycle–pedestrian collisions. Accid Anal Prev 42(1):290–296
Fei-Yue W (2005) Agent-based control for networked traffic management systems. IEEE Intell Syst
20(5):92–96
Government of South Australia (2017) Cycling and the law, Adelaide
Government of South Australia (2018) The driver’s handbook, Adelaide
Gregory R, Rao PS, Aseem V (2004) Directory enabled policy management tool for intelligent
traffic management, Patent
Guin A, Porter C, Smith B (2000) Holmes C (2007) Benefits analysis for incident management
program integrated with intelligent transportation systems operations: case study. Transp Res
Rec 1:78–87
Hamilton-Baillie B (2008) Shared space: reconciling people, places and traffic. Built Environ
34(2):161–181
Hammond V, Musselwhite C (2013) The attitudes, perceptions and concerns of pedestrians and
vulnerable road users to shared space: a case study from the UK. J Urban Des 18(1):78–97
Harwood D, Bauer K, Potts I, Torbic D, Richard K, Rabbani E, Hauer E, Elefteriadou L (1840)
Griffith M (2003) Safety effectiveness of intersection left- and right-turn lanes. Transp Res Rec
J Transp Res Board 1:131–139
Ibanez-Guzman J, Lefevre S, Mokkadem A, Rodhaim S (2010, September) Vehicle to vehicle
communications applied to road intersection safety, field results. In: 13th International IEEE
conference on intelligent transportation systems. IEEE, pp 192–197
Jensen SU, Rosenkilde C, Jensen N (2007) Road safety and perceived risk of cycle facilities in
Copenhagen. Presentation to AGM of European Cyclists Federation
Jou R-C, Wu Y-C, Chen K-HJT (2011) Analysis of the environmental benefits of a motorcycle
idling stop policy at urban intersections. Transportation 38(6):1017–1033. https://doi.org/10.
1007/s11116-010-9318-5
Kirk A, Jones C, Stamatiadis N (2011) Improving intersection design practices. Transp Res Rec
2223(1):1–8. https://doi.org/10.3141/2223-01
Kosaka H, Noda M (2015, August) Education method for safe bicycle riding to evaluate actual
cycling behaviors when entering an intersection. In: International conference on human interface
and the management of information. Springer, Cham, pp 225–232
Laberge JC, Creaser JI, Rakauskas ME, Ward NJ (2006) Design of an intersection decision support
(IDS) interface to reduce crashes at rural stop-controlled intersections. Transp Res Part C Emerg
Technol 14(1):39–56
Lathia N, Ahmed S, Capra L (2012) Measuring the impact of opening the London shared bicycle
scheme to casual users. Transp Res Part C Emerg Technol 22:88–102
20 Planning for Safer Road Facilities for Bicycle Users at Junctions 387
Le Vine S, Zolfaghari A, Polak J (2015) Autonomous cars: the tension between occupant experience
and intersection capacity. Transp Res Part C Emerg Technol 52(C):1–14
Loidl M, Hochmair HH (2018) Do online bicycle routing portals adequately address prevalent safety
concerns? Safety 4(1):9
Oxley J, Fildes B, Corben B, Langford J (2006) Intersection design for older drivers. Transp Res
Part F Traffic Psychol Behav 9(5):335–346
Pascale A, Nicoli M, Deflorio F, Chiara BD, Spagnolini U (2012) Wireless sensor networks for
traffic management and road safety. IET Intell Transp Syst 6(1):67–77
PRESTO (2010) PRESTO cycling policy guide. www.presto-cycling.eu
Pucher J, Buelher R (2007) At the frontiers of cycling: policy innovations in the Netherlands,
Denmark and Germany. J World Transp Policy Pract 13(3):8–56
Pucher J, Komanoff C, Schimek P (1999) Bicycling renaissance in North America?: Recent trends
and alternative policies to promote bicycling. Transp Res Part A Policy Pract 33(7–8):625–654
Reynolds CC, Harris MA, Teschke K, Cripton PA, Winters M (2009) The impact of transportation
infrastructure on bicycling injuries and crashes: a review of the literature. Environ Health 8(1):47.
https://doi.org/10.1186/1476-069X-8-47
Sander U (2017) Opportunities and limitations for intersection collision intervention—a study of
real world ‘left turn across path’ accidents. Accid Anal Prev 99(Pt A):342–355
Santos G, Behrendt H, Teytelboym A (2010) Part II: policy instruments for sustainable road trans-
port. Res Transp Econ 28(1):46–91
Schepers JP, Kroeze PA, Sweers W, Wüst JC (2011) Road factors and bicycle–motor vehicle crashes
at unsignalized priority intersections. Accid Anal Prev 43(3):853–861
Shaheen S, Guzman S, Zhang H (2010) Bikesharing in Europe, the Americas, and Asia Past. Present
Future 2143(2143):159–167
Shechtman O, Classen S, Stephens B, Bendixen R, Belchior P, Sandhu M, McCarthy D, Mann W,
Davis E (2007) The impact of intersection design on simulated driving performance of young
and senior adults. Traffic Inj Prev 8(1):78–86. https://doi.org/10.1080/15389580600994321
Sidra Solutions (2019) Introduction. Akcelik & Associates Pty Ltd. http://www.sidrasolutions.com/
Software/INTERSECTION/Introduction, viewed 08/09
Sun W, Wu X, Wang Y, Yu G (2015) A continuous-flow-intersection-lite design and traffic control
for oversaturated bottleneck intersections. Transp Res Part C Emerg Technol 56:18–33
Taylor APM, Young W (1988) Traffic analysis new technology & new solutions. Hargreen Publish-
ing Company, Melbourne
Wang Y, Nihan NL (2004) Estimating the risk of collisions between bicycles and motor vehicles at
signalized intersections. Accid Anal Prev 36(3):313–321
Wang H, Palm M, Chen C, Vogt R, Wang Y (2016) Does bicycle network level of traffic stress
(LTS) explain bicycle travel behavior? Mixed results from an Oregon case study. J Transp Geogr
57:8–18
Webster FV (1958) Traffic signal settings. H.M.S.O, London
Yan X, Radwan E, Guo D (2007) Effect of a pavement marking countermeasure on improving
signalized intersection safety. ITE J 77(8):30–39
Chapter 21
Method to Evaluate the Location of Aged
Care Facilities in Urban Areas Using
Median Share Ratio
Koya Tsukahara and Kayoko Yamamoto
Abstract The study aims to develop and test a method to evaluate the location of
aged care facilities from the viewpoint of whether they are equitably located for
users, using the improved Median Share Ratio (MSR). By evaluating the current
location of aged care facilities, it is possible to extract the districts which are short
of facilities. The evaluation method was applied to Chofu City in Tokyo Metropolis,
Japan, and the evaluation result of weighting and that of not weighting by elderly
population were compared and discussed. Consequently, adopting the evaluation
method with weighting by elderly population, it is possible to adequately examine
the districts where new aged care facilities should be constructed. From this evidence,
it is significant to evaluate the location of aged care facilities, using the improved
MSR with weighting by elderly population in the study.
Keywords Aged care facility ·Equitable access ·Median Share Ratio (MSR) ·
Geographical information system (GIS) ·Applied statistical method ·Public open
data
1 Introduction
While the aging populations in many advanced countries around the world are
increasing, the increase is tremendously rapid and the lack of aged care facilities
has become a serious social issue, especially in Japan. According to the surveys con-
ducted by the Ministry of Health, Labour and Welfare (MHLW) in Japan, in 2016a,
b, though the admission capacity of special aged care facilities is approximately
498,000, the number of people who desire to enter such facilities was 524,000. This
K. Tsukahara ·K. Yamamoto (B)
Graduate School of Informatics and Engineering, University of Electro-Communications,
1-5-1 Chofugaoka Chofu-Shi, Tokyo 182-8585, Japan
e-mail: kayoko.yamamoto@uec.ac.jp
K. Tsukahara
e-mail: t1730063@edu.cc.uec.ac.jp
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_21
389
390 K. Tsukahara and K. Yamamoto
grave situation highlights the severity of the lack of aged care facilities currently in
Japan.
Additionally, though the number of aged care facilities and their capacities are
increasing, the utilization rates of such facilities remain at the same level and the
lack of facilities has not been addressed. The reason for this is that Japan is aging at
a pace unparalleled in other countries. Though a financial subsidy is provided by the
national and local governments for the construction of aged care facilities, the amount
of such a subsidy cannot be greatly increased due to the need for more childcare and
medical facilities as well. Accordingly, it is not expected that the number of aged
care facilities will greatly increase in the future. Therefore, the construction of new
aged care facilities should be prioritized in districts with greater needs. In order to
make this possible, first of all, it is necessary to accurately grasp the districts which
are short of aged care facilities.
Based on the background mentioned above, using Geographic Information Sys-
tems (GIS), an applied statistical method and public open data related to aging pop-
ulation and aged care facility, the present study aims to develop and test a method
to quantitatively evaluate the locations of aged care facilities in urban areas within
Japan. By evaluating the current location of aged care facilities, it is possible to
extract the districts which are short of facilities, and it will assist the policy and
decision makers in planning new aged care facilities.
2 Related Work
The present study develops and tests a new method to evaluate the location of aged
care facilities. Therefore, the present study is related to three study fields, namely,
(1) studies related to facility location problem, (2) studies related to facility location
problem adopting economic methods, and (3) studies related to the siting of health
care facilities. The following will introduce the major preceding studies in the above
three study areas, and demonstrate the originality of the present study in comparison
with the others.
In (1) studies related to facility location problem, Voogd (1982) developed a
multi-criteria evaluation (MCE) for urban and regional planning. Referring to this,
Pettit and Pullar (1999) developed an integrated planning tool of spatial information
in local government for parcel mapping and asset management using MCE and
GIS. Based on these studies, there were a lot of studies related to facility location
problem using MCE and GIS. For example, in most recent years, Uddin et al. (2018)
and Asborno and Hernandez (2018) developed methods to evaluate the locations
of transportation facilities. In addition to these studies, there are some preceding
studies adopting p-median problems using GIS. For example, Satoh et al. (2018) and
Tsukahara and Yamamoto (2018) respectively developed methods to evaluate the
locations of childcare facilities and aged care facilities.
In (2) studies related to facility location problem adopting economic methods,
Tanaka and Furuta (2015) applied the Quintile Share Ratio (QSR), which is an
21 Method to Evaluate the Location of Aged Care Facilities … 391
indicator showing the degree of bias in income, to the facility locational analysis for
linear cities which are theoretical line-formed cities. The QSR is an inequity measure
of income distribution defined as the ratio of total income received by the 20% of the
population with the highest income (top quintile) to that received by the 20% of the
population with the lowest income (lowest quintile). Additionally, with the QSR as a
reference, Tanaka and Furuta (2016) modified the well-known QSR inequity measure,
and newly defined the Median Share Ratio (MSR) as the ratio of the average distance
among the top half of the population with longer distance to a facility to that among
the bottom half of the population with shorter distance to a facility. Furthermore,
Tanaka and Furuta (2016) used the MSR, which is an equity measure, to develop a
facility location evaluation model in a linear city with one or two facilities, as well
as a uniformly distributed population. Furuta and Tanaka (2017) used a method that
generalized the QSR and proposed a solution to optimize multiple facility locations in
cases where the demand and candidate facility locations are discrete. Other than the
above studies, there are some preceding studies adopting Gini coefficient. Drezner
et al. (2009) investigated the location of facilities, considering that perfect equity is
achieved when distances to the closest facility are the same for all customers. Taylor
and Pettit (2018) analyzed the equity of current health services distribution and the
scale of future demand using GIS.
In (3) studies related to the siting of health care facilities, Kondo et al. (2010)
developed a planning model for medical facilities in addition to road network consid-
ering accessibility and connectivity in large cities. Gu et al. (2010) identified optimal
locations for preventive health care facilities so as to maximize participation. Shariff
et al. (2012) used maximal covering location problem (MCLP) to determine good
locations for the health care facilities such that the population coverage is maxi-
mized. Beheshtifar and Alimoahmmadi (2015) applied a multi-objective model that
combined GIS analysis with a multi-objective genetic algorism (GA) to determine
the optimal number and locations of new health care facilities. Zhang et al. (2016)
examined the problem of where health care facilities should be located to raise the
total accessibility for the entire population in highly developed cities, using a multi-
objective GA. Oppio et al. (2016) proposed an evaluation system to be applied for
the site selection of new hospitals. Segall et al. (2017) applied data envelopment
analysis (DEA) to select the candidate town for location of a health care facility.
Regarding studies related to (1), there are a lot of preceding studies using MCE to
evaluate the locations of various facilities, and Tsukahara and Yamamoto (2018)just
targeted aged care facility. Regarding studies related to (2), though Tanaka and Furuta
(2016) and Furuta and Tanaka (2017) focused on the equity to propose the evaluation
method for facility location, it has been applied to theoretical city modelling exercises
and not to any real cities. In the above studies, equity was an indicator to show
the degree of the accessibility to facilities for users. Additionally, regarding studies
related to (3), though most of the studies target medical facilities, there are very few
preceding studies related to aged care facilities. Therefore, with the results of the
preceding studies mentioned above as a reference, the present study will demonstrate
the originality and significance by considering the lack of aged care facilities, which
has become a serious social issue as mentioned in the previous section, and developing
392 K. Tsukahara and K. Yamamoto
a new method to quantitatively evaluating current location of aged care facilities
focusing on the equity in terms of the accessibility to facilities for users in a real
city. Additionally, based on the evaluation results, the present study will extract the
districts which are short of facilities and propose the places where new facilities
should be constructed.
3 Evaluation Method
3.1 Previous Method
In order to develop a new method to evaluate the location of aged care facilities in a
real city, and focus on equity indicators which show the degree of facility accessibility
to users, the present study improves upon the MSR. Figures 1and 2respectively show
the examples of a linear city and a real city. According to Tanaka and Furuta (2015,
2016), the MSR in the case where one facility exists in a linear city as shown in Fig. 1
is derived by the following procedure:
(i) Deduce the cumulative distribution function FX(x)of the distance Xto the
facility;
(ii) Specify the median mof the distance Xat which FX(m)=0.5;
(iii) Find the probability density function fX(x)of the distance Xto the facility;
(iv) Calculate the mean value vHof the interval where the distance is longer than
the median m, and the mean value vLof the interval where the distance is
shorter than the median m;
(v) The MSR can be calculated by dividing vHby vL.
Fig. 1 Example of a linear
city
Fig. 2 Example of a real
city
21 Method to Evaluate the Location of Aged Care Facilities … 393
The cumulative distribution function FX(x)and the probability density function
fX(x)have a relationship expressed by Eq. (1). The MSR can be calculated using
probability density function fX(x)as shown in Eq. (2). From Eq. (2), it can be judged
that facilities are fairly located as the MSR value is lower. Incidentally, the constant
lin Eq. (2) is the distance from the facility to the furthest district.
fX(x)=d
dx FX(x)(1)
MSR =vH
vL
=∫l
mxfX(x)dx
∫m
0xfX(x)dx (2)
3.2 Expanded Method
There are some differences between a linear city and a real city, comparing Figs. 1
and 2. Therefore, it is necessary to improve the MSR to be suitable for a real city in
the present study. At first, distance Xmust be considered as area. In a linear city, as
showninFig.1, the distance was treated as a line segment. However, in a real city,
as shown in Fig. 2, since the aggregate of line segments is a distance, the distance X
can be regarded as the area S(x)with a radius of distance x. Therefore, FX(x)and
fX(x)are respectively expressed by Eqs. (3) and (4).
FX(x)=∫x
0S(t)dt
∫l
0S(s)ds (3)
fX(x)=S(x)
∫l
0S(s)ds (4)
Second, the number of facilities increases from 1 to n. In this case, the area S(x)
is the sum of the areas which are away from each facility by distance x. Assuming
that the user uses the nearest facility, Voronoi tessellation is performed with each
facility as a kernel point, and the sum of area of each Voronoi polygon is the area
S(x)[Eq. (5)].
S(x)=S1(x)+···+ Sn(x)(5)
Finally, it is necessary to reflect the deviation of the distance from the facility by
the population in Voronoi domain. When weighting by the population, the product
of the number of users pi(x)and the distance xfrom the facility is the area of the
Voronoi polygon iS
i(x)[Eq. (6)].
Si(x)=pi(x)x(6)
394 K. Tsukahara and K. Yamamoto
In Sect. 5, the evaluation results in the cases of weighting and not weighting by the
population will be compared. Based on the comparison result, in Sect. 6, the validity
of the evaluation method using the improved MSR with weighting by the population
will be verified.
The present study used ArcGIS Pro Ver. 2.0 provided by the Environmental Sys-
tems Research Institute, Inc. (ESRI) as GIS. Utilizing specific functions of this GIS,
the following sections will process the data related to aging population and aged care
facility in the digital map format, evaluate the location of aged care facilities, and
visualize the evaluation results on the digital maps.
3.3 Application of Expanded Method
As the situations are different every region, it is impossible to compare the equity
of the facility locations among multiple regions using the previous MSR which was
introduced in Sect. 3.1. Because area, population and number of facilities are different
in every region, these differences have an enormous influence on the evaluation
results. Therefore, in order to eliminate these influences and compare the equity of
the facility locations using the MSR values among multiple regions, the present study
propose to adopt the ratio of the evaluation results of current and reference locations
focusing on the MSR value.
The former is the present location, and the latter is the location with the best MSR
value (the lowest MSR value) and the best equity for users in the region. The area,
population and number of facilities of the latter are same with the former, and there is
the difference of the facility locations between these two locations. Therefore, based
on the reference location, it is possible to evaluate the current location from the
viewpoint of whether the facilities are equitably located for users. Additionally, as
only the locations are different, it is also possible to compare the equity of the facility
locations among multiple regions, ignoring the influences caused by the differences
of area, population and number of facilities.
A reference location is determined to have the best MSR value, based on the
simulation results in which the same number of facilities as the current location
are randomly placed in the center of each district in the evaluation target area. The
5000 simulations are conducted just in the case of not weighting by the popula-
tion. Referring the preceding studies related to facility location problem, the number
of simulations is generally 1000–5000 times. The same reference location is also
adopted in the case of weighting by the population.
21 Method to Evaluate the Location of Aged Care Facilities … 395
4 Selection of Evaluation Target Area and Data Processing
4.1 Selection of Evaluation Target Area
For the evaluation target area in the present study, Chofu City in Tokyo Metropolis,
Japan is selected. Chofu City is located in the suburban area of Tokyo Metropolis
as shown in Fig. 3. In Chofu city, the aging population (age 65 or over) has already
exceeded the youth population (under age 15), and the former population is expected
to increase in the future. According to Hashimoto (2015), a healthy life expectancy
is 71.19 years old for men and 74.21 years old for women in the present Japan.
Because elderly people over 75 years old are called as “the latter-stage elderly peo-
ple” according to Japanese regulations in social welfare services and have a high
possibility to become the users of aged care facilities, the present study targets this
age group. According to the survey on the aging population in Chofu City, 23,545
people, equivalent to approximately 10% of the total population, fit in the current
age range of the present study subject which is 75 and over. In the present study,
evaluation will be conducted in the unit of 105 districts within Chofu City.
The present study targets 36 aged care facilities at which elderly people in need
usually stay located in Chofu City and excludes the ones in neighboring other cities.
According to Japanese regulations in social welfare services, it is necessary for
elderly people to enter the aged care facilities in the cities where they live.
Fig. 3 Location of Chofu City in Tokyo Metropolis
396 K. Tsukahara and K. Yamamoto
Tabl e 1 List of utilized data
Utilized data Utilization method of data
Population by age
(National Census 2010 by the Statistics
Bureau)
URL: https://www.e-stat.go.jp/
Creation of the distribution map of the aging
population in the unit of 105 districts
Local care resources
(Regional figures—Chofu City, Tokyo
Metropolis by the Japan Medical Association)
URL: http://jmap.jp/pages/guide
Creation of the distribution map of aged care
facilities
4.2 Data Processing
4.2.1 Data Overview
The utilized data and the utilization method of the data in the present study are shown
in Table 1.
4.2.2 Distribution Maps of Aging Population and Aged Care Facilities
Figure 4shows the distribution of the elderly population over 75 years old in Chofu
City. Figure 5shows the distribution of aged care facilities in Chofu City and the
Euclidean distance between the center of district and the nearest aged care in the
case of current location. As shown in Fig. 4, the southern and eastern parts have high
aging populations and there are old housing complexes in most of these areas. As
indicated in Fig. 5, aged care facilities are distributed throughout the entire Chofu
City excluding the central part. Additionally, in response to the evaluation method
in the present study in Sect. 3.2, Euclidean distance is adopted for buffer analysis in
Fig. 5.
5 Evaluation
5.1 Evaluation for the Case of Not Weighting by Elderly
Population
First, this section will introduce the evaluation result without weighting by the elderly
population over 75 years old, using Eq. (5) in Sect. 3.2. In the case of current location
asshowninFig.5, calculating the area every 0.1 km from the nearest aged care facility
based on this figure, the graphs of the cumulative distribution function FX(x)and
the probability density function fX(x)are obtained as shown in Fig. 6. Additionally,
21 Method to Evaluate the Location of Aged Care Facilities … 397
Fig. 4 Distribution of the elderly population over 75 years old (Average population per district is
approximately 180)
Fig. 5 Distribution of aged care facilities and distance between the center of district and the nearest
aged care facilities (km) in the case of current location
398 K. Tsukahara and K. Yamamoto
Fig. 6 Cumulative distribution function FX(x)and probability density function fX(x)without
weighting by elderly population
Fig. 7shows the distribution of aged care facilities and distance between the center
of district and the nearest aged care in the case of reference location as described in
Sect. 3.3. In the case of reference location as shown in Fig. 7,bythesamewayas
the case of current location, the graphs of the cumulative distribution function FX(x)
and the probability density function fX(x)are also obtained as shown in Fig. 6.
The results of calculating the median m, the mean value vH, the mean value vL,
and the MSR value by using the data in Fig. 6are shown in Table 2. As is clear from
Table 2, all values of reference location are lower than those of current location.
21 Method to Evaluate the Location of Aged Care Facilities … 399
Fig. 7 Distribution of aged care facilities and distance between the center of district and the nearest
aged care facilities (km) in the case of reference location
Tabl e 2 Evaluation result of the MSR without weighting by elderly population
MSR m(km) vHvL
Current location 2.72809 0.47822 0.41787 0.15317
Reference location 1.88373 0.44307 0.28690 0.15230
Current location/Reference location 1.44824 1.07933 1.45650 1.00571
5.2 Evaluation for the Case of Weighting by Elderly
Population
Next, this section will introduce the evaluation result with weighting by the elderly
population over 75 years old, using Eq. (6) in Sect. 3.2. In the case of current location
as shown in Fig. 5, calculating the area every 0.1 km from the nearest aged care
facility based on this figure, and weighting the distance by the elderly population
over 75 years old, the graphs of the cumulative distribution function FX(x)and the
probability density function fX(x)are obtained as shown in Fig. 8. Additionally,
in the case of reference location shown in Fig. 7, by the same way as the case of
current location, the graphs of the cumulative distribution function FX(x)and the
probability density function fX(x)are also obtained as shown in Fig. 8.
The results of calculating the median m, the mean value vH, the mean value vL,
and the MSR value by using the data in Fig. 8are shown in the Table 3. As is clear
400 K. Tsukahara and K. Yamamoto
Fig. 8 Cumulative distributionfunction FX(x)and probability density function fX(x)with weight-
ing by elderly population
Tabl e 3 Evaluation result of the MSR with weighting by elderly population
MSR m(km) vHvL
Current location 2.30765 0.37581 0.30257 0.13112
Reference location 1.75077 0.41032 0.25144 0.14362
Current location/Reference location 1.31808 0.91589 1.20335 0.91296
from Table 3, all values of reference location are lower than those of current location.
Comparing Tables 2and 3, all values in the case of weighting by elderly population
are lower than those in the case of not weighting by elderly population introduced
in the previous section.
21 Method to Evaluate the Location of Aged Care Facilities … 401
6 Discussion
In this section, the evaluation results in the previous section will be compared and
discussed, and the validity of the evaluation method using the improved MSR with
weighting by elderly population will be verified. First, in the case of not weighting
by the elderly population over 75 years old, according to Table 2, the MSR value and
the mean value vHof current location are approximately 1.4 times higher than those
of reference location. On the other hand, median mand the mean value vLof current
location are almost same as those of reference location. From these, in Chofu City,
in the case of reference location, it is clear that the height of the MSR value is due
to the height of the mean value vH. According to Fig. 6, it can be said that the small
peak of probability density function fX(x)between the distance of 1.5 and 2.0 km
is the cause of the height of the mean value vH. Therefore, it is evident that the MSR
will become lower by constructing new aged care facilities in the districts which are
1.5–2.0 km away from the nearest facilities in the northwestern and southern parts
of Chofu City.
Then, based on Tables 2and 3, the evaluation results of weighting and that of
not weighting by the elderly population over 75 years old are compared. It is clear
that all values are lower and the mean value vHof current location is tremendously
lower in the latter case than the former case. From these, all of the ratios of the
evaluation results of current and reference locations are also lower in the latter case
than the former case. Additionally, comparing Figs. 4and 5, in the northwestern
and southern parts of Chofu City, there are the districts which are 1.5–2.0 km away
from the nearest facilities and have very few elderly populations over 75 years old.
According to Fig. 8, the peak of probability density function fX(x)between the
distance of 1.5 and 2.0 km is not seen in the case of weighting by elderly population.
Therefore, in the case of weighting by elderly population, it is not necessary to
construct new facilities in these parts. Consequently, comparing the evaluation result
of weighting and that of not weighting by elderly population, it is evident that the
evaluation method using the improved MSR with weighting by the elderly population
over 75 years old appropriately reflected the influence of the distribution of such an
elderly population.
However, even in the case of weighting by elderly population, since the ratio of
the MSR value of current and reference locations is approximately 1.3, it is essential
to examine the current location from the viewpoint of whether they are equitably
located for users. As mentioned in Sect. 2, in the present study, equity is an indicator
to show the degree of the accessibility to facilities for users. In Chofu City, there is a
district which is more than 0.9 km away from the nearest aged care facility and has
an elderly population of more than 280 in the eastern part. Additionally, there are no
aged care facilities in the districts which have an elderly population of less than 70.
Therefore, in order to pull down the ratio of the MSR value of current and reference
locations, it is necessary to examine the construction of new aged care facilities in
the above districts. Though it is desirable to consider the capacity of existing and
402 K. Tsukahara and K. Yamamoto
planned aged care facilities in the evaluation method developed in the present study,
such data is not accessible to the public in the present Japan.
On the other hand, referring to Tables 2and 3, the MSR values of reference
location are lower than those of current location in both cases of weighting and not
weighting by the elderly population over 75 years old. From this, it is evident that
the reference location is reasonable in these two cases assumed in the evaluation
method of the present study. In the case of reference location, though the MSR value
is weighted by elderly population, a few of the aged care facilities are located in
districts where elderly people over 75 years old do not live. Additionally, though the
parallel distributed processing was conducted using plural high-efficiency computers,
it takes a huge amount of calculation time to detect the location with the best MSR
value. Therefore, it is necessary to devise and improve the method to determine a
reference location.
Thus, the present study evaluated the location of aged care facilities, by applying
the improved MSR to Chofu City, and using the public open data related to the elderly
population over 75 years old and aged care facilities, and the evaluation method
has high spatial reproducibility. Moreover, the evaluation method is based on the
above public open data and public information as described in Sect. 4. Therefore, by
obtaining population data and geospatial data similar to those of the present study,
the evaluation method can also be applied to other areas in the past and future.
Accordingly, it can be said that the evaluation method has high time reproducibility
as well as spatial reproducibility. However, if the data related to the capacity of
existing aged care facilities is obtained, it is possible to raise the accuracy of the
evaluation method in the present study.
7 Conclusion
The present study developed and tested a method to evaluate the location of aged care
facilities from the viewpoint of whether they are equitably located for users. In order
to improve the MSR, which has been applied only to a linear city in the preceding
studies, to apply it to a real city, the present study changed a distance to an appropriate
shape, targeted multiple facilities, and reflected the deviation of the distance from the
facility by the population. Moreover, as the ratio of the evaluation results of current
and reference locations focusing on the MSR value can be calculated, it makes it
possible to compare the equity of the facility locations among multiple regions.
As a model case of evaluation in the present study, the improved MSR with and
without weighting by the elder population over 75 years old is applied to Chofu
City in Tokyo Metropolis, Japan. Based on the evaluation results, it is possible to
adequately examine the districts where new aged care facilities should be constructed
in the case of weighting by elderly population. From this evidence, it is significant
to weight the distance by elderly population to evaluate the location of aged care
facilities using the improved MSR in the present study. As the evaluation method is
based on public open data and public information, by obtaining population data and
21 Method to Evaluate the Location of Aged Care Facilities … 403
geospatial data similar to those of the present study, evaluations can be conducted in
other areas as well as for the past and future. Therefore, the evaluation method in the
present study has a high temporal reproducibility as well as spatial reproducibility.
The future research issue is to raise the accuracy of the evaluation method in
the present study, and reduce the amount of calculation time to detect the reference
location with the best MSR value, as mentioned in Sect. 6. For these, it is necessary to
devise and improve the method to determine a reference location in case of weighting
by elderly population, and ameliorate the method of parallel distributed processing.
After that, it is essential to apply the improved evaluation method to other areas to
verify the validity.
References
Asborno MI, Hernandez S (2018) Using data from a state travel demand model to fevelop a multi-
criteria framework for transload facility location planning. Transp Res Rec J Transp Res Board
2672(9):12–23. https://doi.org/10.1177/0361198118772699
Beheshtifar S, Alimoahmmadi A (2015) A multi-objective optimization approach for location-
allocation of clinics. Int Trans Oper Res 22(2):313–328
Chofu City: A Survey on the aging population in Chofu City [Internet]. 2018. Available from http://
www.city.chofu.tokyo.jp/www/contents/1515569448760/index.html. Accessed 23 Aug 2018
Dreznera T, Dreznera Z, Guyseb J (2009) Equitable service by a facility: minimizing the Gini
coefficient. Comput Oper Res 36(12):3240–3246
Furuta T, Tanaka K (2017) Minimizing quantiles share ratio in multiple facility location problem
with total distance constraint. Trans Oper Res Soc Jpn 60:36–49
Gu W, Wang X, McGregor SE (2010) Optimization of preventive health care facility locations. Int
J Health Geographics 9(1):17. http://www.ij-healthgeographics.com/content/9/1/17
Hashimoto S (2015) A study on the Indicator of healthy life expectancy: an examination of healthy
life expectancy for promotion of national health promotion exercise (Health Japan 21) in the
21st Century. Report for grants-in-aid for scientific research expenses provided by the Ministry
of Health, Labour and Welfare (A study on the promotion of national health promotion exercise
(Health Japan 21) in the 21st Century), pp 26–38
Kondo R, Shiomi Y, Uno N (2010) Road network and medical facilities planning model considering
accessibility and connectivity. Infrastruct Planning Rev 27:579–588
Ministry of Health, Labour and Welfare (2016a) A survey on aged care facilities and office, 2 p
Ministry of Health, Labour and Welfare (2016b) Situation of the applicants for entrance to aged
care facilities, 3 p
Oppio A, Buffoli M, Dell’Ovo M, Capolongo S (2016) Addressing decisions about new hospi-
tals’ siting: a multidimensional evaluation approach. Annali dell’Istituto Superiore di Sanità
52(1):78–87
Pettit C, Pullar D (1999) An integrated planning tool based upon multiple criteria evaluation of
spatial information. Comput Environ Urban Syst 23(5):339–357
Satoh K, Tsukahara K, Yamamoto K (2018) Location evaluation of childcare facilities focusing on
transportation in Japanese urban areas. J Geogr Inf Syst 10(6):521–538
Segall M, Lall LT, Abel VM (2017) Health care facility location: a DEA approach. Am J Manage
17(6):54–65
Shariff SSR, Moin MH, Omar M (2012) Location allocation modeling for health care facility
planning in Malaysia. Comput Ind Eng 62(4):1000–1010
Tanaka K, Furuta T (2015) Quintile share ratio in a linear city with one and two facilities. J City
Planning Inst Jpn 50(3):628–635
404 K. Tsukahara and K. Yamamoto
Tanaka K, Furuta T (2016) Evaluating inequity of facility location in a linear city using median
share ratio. J City Planning Inst Jpn 51(3):894–900
Taylor A, Pettit C (2018) Are health services in New South Wales available where they are needed?
Using AURIN to understand the equity of service distribution and future demand. In: Applied
spatial analysis and policy, pp 1–25. https://doi.org/10.1007/s12061-018-9282-6
Tsukahara K, Yamamoto K (2018) Evaluation of nursing facility locations using the specializa-
tion coefficient of the population aging rate. In: Rocha J (ed) Geographic information systems,
IntechOpen. http://dx.doi.org/10.5772/intechopen.81364
Uddin S, Chakravorty S, Ray A, Sherpa KS (2018) Optimal location of sub-station using GIS and
multi-criteria decision making approach. Int J Decis Support Syst Technol https://doi.org/10.
4018/ijdsst.2018040104
Voogd H (1982) Multicriteria evaluation for urban and regional planning. Pion, London, p 367
Zhang W, Cao K, Liu S, Huang B (2016) A multi-objective optimization approach for health care
facility location-allocation problems in highly developed cities such as Hong Kong. Comput
Environ Urban Syst 59:220–230
Chapter 22
Identifying Changes in Critical Locations
for Transportation Networks Using
Centrality
Nazli Yonca Aydin, Ylenia Casali, H. Sebnem Duzgun
and Hans R. Heinimann
Abstract Identifying critical locations in road networks assists in reducing the risks
of intermittent services and increases the quality of life. Complex network applica-
tions are used in transportation networks to identify critical locations from a topo-
logical point of view. However, critical locations change when there are disruptions
and people move towards a specific service inside its catchment area. In this chapter,
a modified betweenness centrality is used to identify critical locations when moving
towards a single service. This index, the origin-destination betweenness central-
ity, is used to identify important locations in the baseline scenario for a case study
from Kathmandu, Nepal. Furthermore, random disruptions with increasing magni-
tude are simulated to understand a network’s behavior and to identify the changes in
those critical locations under extreme conditions. The results demonstrated that the
origin-destination betweenness centrality is an effective index. Furthermore, random
disruption simulations can assist decision-makers in preparing recovery plans.
Keywords Transportation network ·Centrality ·Critical locations
N. Y. Aydin (B)·Y. C a s a l i ·H. R. Heinimann
ETH Zurich, Future Resilient Systems, Singapore-ETH Centre, 1 CREATE Way
CREATE Tower, Singapore 138602, Singapore
e-mail: nazli.aydin@frs.ethz.ch
Y. C a s a l i
e-mail: ylenia.casali@frs.ethz.ch
H. R. Heinimann
e-mail: hans.heinimann@env.ethz.ch
H. Sebnem Duzgun
Department of Mining Engineering, Colorado School of Mines, 1500 Illinois Str.,
Golden, CO 80401, USA
e-mail: duzgun@mines.edu
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_22
405
406 N. Y. Aydin et al.
1 Introduction
Transport networks are mainly designed to deliver services to societies continuously
in business as usual conditions. Quite recently, the behavior of transport networks sub-
ject to major disruptions has gained attention, calling for more fault tolerant systems.
Resilience is an emerging concept, enabling systems with rebounding capabilities
(i.e. restitution to its original performance), which can be framed within four generic
functions: (1) resisting within acceptable limits of degradation; (2) re-stabilizing
critical functionality; (3) rebuilding up to a sufficient level of functionality; and (4)
re-configuring the flows of services and the underlying infrastructure. Resilience is
defined as “the capacity of a system to absorb disturbance and to reorganize so as
to retain essentially its structures, functions and feed-back loops.” (Singapore-ETH
Center 2014). Although human behavior responses to disruptions can be unpre-
dictable and difficult to characterize (Batty 2001), identifying critical locations and
functions will assist in improving the robustness, re-building capabilities, emergency
response performance and preparedness.
Critical locations in spatial networks have been studied by using centrality indices.
Specifically, betweenness centrality has been used to identify important locations in
spatial networks as defined by the number of times a node was passed by while travel-
ing throughout a network (Freeman 1977). The information regarding the importance
of a node is crucial, as its removal will impact on a network’s behavior, lengthen the
paths between node pairs and provide an overall indication of a network’s resilience
(Barthelemy 2004). There is a wide range of literature using the betweenness central-
ity to evaluate the importance of nodes such as analyzing the importance of individual
channels in a multi-threaded channel system in Bangladesh (Marra et al. 2014) and
observing the evolution of road networks in Milan, Italy (Strano et al. 2012). Addi-
tional examples for critical nodes and/or edges evaluations in large physical graphs
can be found in Aydin et al. (2018b), Porta et al. (2006a,2011), and Aydin (2018).
In addition, there are examples in the literature that utilize the graph-based indices
to investigate disaster management and recovery after disruptions for transport infras-
tructures. One previous example on this subject was proposed by Aydin et al. (2018a)
who investigated rural road recovery strategies in order to increase the resilience
and dynamic recovery performance after earthquakes and earthquake-triggered land-
slides in Sindhupalchok District, Nepal. The resilience enhancement strategies were
evaluated using a connectivity metric called Giant Connected Component (GCC).
Similarly, changes in graph diameter and average path length were used as indicators
for the failure simulations in transportation networks by Schintler et al. (2007).
In some cases, the commonly used centrality indices were modified to provide
case specific solutions. For example, the authors identified critical locations in the
city of Kathmandu using the origin-destination betweenness centrality (Aydin et al.
2017) in order to identify important locations when traveling to a hospital. While this
previous research also utilized origin-destination betweenness centrality, it adopted
22 Identifying Changes in Critical Locations for Transportation … 407
a deterministic approach by exclusively focusing on a single disruption type. The
important locations were only identified based on the existing road closure data set.
There is still a lack of understanding of the way that the origin-destination between-
ness centrality performs in transportation networks with increasing magnitude of
disruption randomly. Based on the magnitude of the disruption, the critical locations
that are identified by the origin-destination betweenness centrality would change.
In addition, most of the literature utilizes betweenness centrality applied to the
entire road network, while accessibility to critical services, such as emergency head-
quarters, shopping centers, or business hubs is considered to be an adequate repre-
sentation of critical transport functionality. However, important nodes in a sub-graph
level (i.e. catchment areas of those critical services) while traveling to a specific
location has limited examples.
In this current research, our first aim was to close this gap and improve our
understanding of a network’s topological properties subjected to randomly occurring
yet an increasing magnitude of disruptions when traveling to a specific service.
We will investigate the limit of degradation through a randomized process. Our
subsequent goal is to investigate the changes in node importance of catchment areas
(i.e. sub-graphs) using origin-destination betweenness centrality when the network
is subject to those random disruptions.
2 Methodology
In this study, we focused on the following conditions: (1) identifying critical loca-
tions in the baseline and (2) changes in those important locations when the network
is subject to random disruptions. In both conditions, rather than focusing on the
entire network using the betweenness centrality, we used the modified betweenness
centrality to investigate the important nodes when traveling to a specific location or a
service such as a hospital(s), shopping center(s) or a housing area. The modification
in the betweenness centrality enables specifying a set of origins and a destination
in a graph and/or a sub-graph. In this study, the network behavior and the spatial
changes in importance were observed for the baseline conditions and under random
disruptions by examining the origin-destination betweenness centrality distributions
and the spatial distribution of the difference index, respectively.
The study of random disruptions requires a disturbance generating mechanism
that mimics events as close as possible to reality. Here, we set up an experimental
layout that systematically increases disruption intensities and accordingly alters the
network topology. This simulation analysis also reveals critical disruption intensities,
at which loss of transport robustness goes beyond the acceptable limits of degradation.
408 N. Y. Aydin et al.
2.1 Origin-Destination Betweenness Centrality
Centrality metrics are essential to complex network theory and applications to
evaluate networks structure and a group of important nodes within a graph. There are
different metrics to evaluate centrality depending on the purpose (i.e. degree central-
ity, closeness centrality etc.). Although it is an effective way to evaluate the influence
or ranking of nodes in networks in most cases, some case studies require modifications
to these commonly used metrics to solve more network-specific research questions.
For example, Giustolisi et al. (2017) was inspired by degree centrality and expanded
it to a new metric called neighborhood nodal degree centrality to classify water net-
works. Similarly, Wang et al. (2017) studied the ranking nodal importance based on
the position of a node as well as its neighborhood in artificial and real-world networks.
This multi-attribute ranking method showed good performance when compared to
previous methods.
The betweenness centrality has been modified to study different aspects of impor-
tance in transportation networks. Kazerani and Winter (2009) modified betweenness
centrality to predict travel demand when considering the dynamic properties of trans-
portation networks. Similarly, Ye et al. (2016) used a modified betweenness metric to
study transportation network and traffic flow using the location-based taxi data in the
USA and China. Zhao et al. (2017) used the previously proposed betweenness cen-
trality and investigated if there is a correlation between metric and flow information
in transportation networks.
The commonly used betweenness centrality is evaluated for a single node by
counting the number of shortest paths passing through that node while routing
between all other node pairs in the entire network (Newman 2010). In this paper,
we propose a modification to the standard betweenness centrality that will enable
calculation of the node importance when there is a specific target or a group of target
nodes that all the other nodes direct travel towards. Road networks provide services
to different uses in the urban area. In a city, people would travel from a housing
area to job centers or to service locations which have their own catchment area (i.e.,
service area). This metric would allow focusing on specific use and their service
areas. The representation is given as follows.
Let G=(N,M) be an undirected and weighted graph with Nnumber of nodes and
Mnumber of edges. Weights are the length of an edge. We modelled the network as
an undirected graph and ignored the traffic rules and flows restrictions since those can
be omitted in emergency situations specifically after natural disasters. An edge, ei,j,
connects nodes vito vjand is categorized by the road hierarchy. Edges have variable
speed limits associated with their hierarchical level. Each specific service in a graph
(i.e. road network) has its own service area. The service area can be determined by
different techniques. In this study, we use an isochrone-based methodology to identify
a service area (i.e. catchment area) which is a widely used technique (Ertugay et al.
2016).
Ideally, all functions in cities have their own service zone that serves communities
that reside within their respective zones. An example of an isochrone-based map
22 Identifying Changes in Critical Locations for Transportation … 409
which is essentially a network-based buffer analysis, can be found in QGIS (2018).
For example, a hospital, vh, has a service area which is determined using the network
distances and the amount of time to reach vh. While evaluating the service area, travel
time using the network distances is calculated by dividing the length of an edge (i.e.
edge weight) by its speed limit. A service zone, Z, is essentially a sub-graph Z=(N,
M), such that a destination node and origin nodes are a subset of the graph, N⊆N
and corresponding edges are the subset of the graph, M⊆M. The origin-destination
betweenness centrality of a node located in the sub-graph, vi, is calculated as follows:
Bx,OD =
n
d=1
k
o=1
no,d(x)
no,d
(1)
where Bx,OD is the origin–destination (OD) betweenness degree of node x, n is the
total number of destinations in a catchment area, kis the total number of nodes
excluding target(s) in a service area, no,d(x)is the number of times node xis used
while traveling to the destination, and no,dis the total number of shortest paths
between origin, o, and destination, d. As each hospital had its own catchment area,
there is a single destination, n=1, however, the origin–destination betweenness
centrality can be evaluated in cases in which there are multiple destinations in the
same sub-graph or in a graph.
Figure 1illustrates the difference between the most commonly used graph metrics
and the proposed origin–destination betweenness degree on an example network. In
this figure, node identification numbers (node IDs) are presented on the nodes and
the sizes of the nodes signify the importance levels (i.e. the size of the node is
proportional to the importance of the node).
Here, Fig. 1a–c illustrate the degree, closeness and betweenness centralities. The
most important nodes in Fig. 1a are node IDs 9, 10, 11, and 12 and the most impor-
tant nodes in Fig. 1b, c are node IDs 9 and 12. The difference between the widely
used metrics and our proposed metric is two-fold; (1) using an origin-destination
betweenness centrality allows the use of a specific target node (see Fig. 1d, e) instead
of calculating the shortest paths between all node pairs (see Fig. 1c). (2) We evalu-
ated the importance of nodes on a sub-graph which represents the catchment areas
of critical services (see Fig. 1e where yellow edges and navy nodes represent the
sub-graph). While our aim was to find important locations when traveling to a “Tar-
get” node within its own service area as marked in Fig. 1e, it can also be calculated
for the entire study area as marked in Fig. 1d. Regardless, there is a clear shift in
the node importance when we apply the origin-destination betweenness centrality,
as the most critical is node ID 6 which could not be captured using other methods.
2.2 Baseline Scenario
In order to thoroughly understand the network’s behavior under disruptions, baseline
conditions need to be identified as a first step. Depending on the decision-makers’
410 N. Y. Aydin et al.
Fig. 1 The images display the results of four centrality metrics on a graph, which are calculated
by using: adegree centrality, bcloseness centrality, cbetweenness centrality, dorigin-destination
betweenness centrality results for the entire study area, eorigin-destination betweenness centrality
results on a sub-graph where edges are marked in yellow color
preference, critical services might change. For example, from a disaster management
perspective, the target nodes in a transportation network can be hospitals, emergency
management centers, and/or shelters. These services are attraction points for com-
munities and create the human flow or other services (e.g. business and trade centers,
recreational areas etc.). Each service has its own catchment area in the baseline condi-
tion depending on the functional and locational requirements. For example, hospitals
should be accessible within 5 min at all times which is determined as the critical time
for saving lives (Ertugay and Duzgun 2011). In this study, we defined accessibility
to hospitals, health care centers as critical services (i.e. targets) for applying the
proposed methodology. Ideally, service areas of hospitals in a city should cover the
entire network.
In the next step, catchment areas for the selected services were identified. Method-
ologies to determine catchment areas range from a more sophisticated [i.e. two-step
floating catchment area method by Delamater (2013)] to less complicated options
[i.e. isochrone-based approaches by O’Sullivan et al. (2000), and Albacete et al.
(2017)]. Each method has its own advantages and disadvantages. As mentioned in
Sect. 2.1, the isochrone-based method is used in this study for identifying a catch-
ment area of a hospital as it has a low computation time. The threshold time to reach
a hospital is determined as 5 min. In other words, all the nodes within that catchment
area must reach the target hospital within 5 min in a graph. The travel time for each
road segment was calculated as dividing the road segment distance by the speed lim-
its applying to that particular road segment which were assigned based on the road
hierarchy level with speed limits of 70, 50, and 30 km/h being assigned to primary,
secondary and local road segments, respectively.
22 Identifying Changes in Critical Locations for Transportation … 411
Finally, the origin–destination betweenness of origin nodes within this catchment
area was evaluated. This measure emphasizes the importance while traveling to a
specific hospital in its own service area which is essentially a sub-graph, while the
standard betweenness centrality was only used to measure the importance of nodes
(i.e. the number of times a node was visited) when traveling throughout the network.
2.3 Random Disruption Scenario
Random disruptions were applied to investigate the consequences of disturbances on
the spatial distribution of important nodes in networks. Here, we would like to analyze
the impacts of disruptions with increasing magnitudes in relation to accessibility
to hospitals. Random disruptions are modelled such that each randomly selected
node within a hospital catchment area is removed together with its adjacent edges
and a new sub-graph was created with the remaining nodes. The origin–destination
betweenness was calculated for each node within this new sub-graph. The random
disruptions were modeled using a list of fractions/probabilities that represents the
number of nodes to be selected randomly.
In order to get as close real-world representation as possible within a reasonable
computation time, the random disruption simulations for each fraction were repeated
a number of times, which we refer to here as a sample size. We determined the sample
size using another graph-based metric called Giant Connected Component (GCC)
which represents the size of the largest connected component in a network. Essen-
tially, the sample size was determined by increasing the number of simulations sys-
tematically until the variation of the normalized mean GCC is reasonable. The GCC
is normalized by the original size of the sub-graph (i.e. GCCsimulated/GCCsub-graph ).
After determining the sample size, s, the following procedure was followed:
1: Create a sub-graph, Z=(N', M'), to represent the catchment area of a
hospital
2: Define a list of fraction f
3: For each fraction, f:
4: For the number of times, s;
5: Randomly select f*N' number of nodes from a sub-graph
6: Remove the selected nodes and adjacent road segments
7: Create a new sub-graph with the remaining nodes
8: Evaluate the origin-destination betweenness centrality for
each node
9: Return mean and standard deviation of the origin-destination
betweenness centrality for each node for the fraction, f.
412 N. Y. Aydin et al.
2.4 Evaluating the Changes in Importance
In this study, the changes in a node’s importance were evaluated spatially by the
following metric called a difference index:
Diffx=Bx,OD,baseline −Bx,OD,mean random disruption (2)
where Diff xis the difference in the origin-destination betweennesses centrality
indices, Bx,OD,baseline is the origin-destination betweennesses for each node that
belongs to the service area of a hospital (i.e. sub-graph) within the baseline network,
and Bx,OD,mean random disruption is the mean value of the origin-destination between-
nesses for each node calculated by step 9 in the random disruption scenario procedure.
This difference index facilitated the identification and interpretation of changes in the
node importance spatially. The difference index has been used to evaluate changes
in importance before by Aydin et al. (2018b).
2.5 Evaluating the Origin-Destination Betweenness
Centrality Distributions
In the distribution analysis, we analyzed the variability of the origin-destination
betweenness centrality for the baseline and random disruption scenarios to observe
the changes in a network’s behavior as well as to observe the limit of degradation for
the network. We normalized the origin-destination betweenness centrality values for
the baseline scenario and the mean origin-destination betweenness centrality values
for random disruption scenarios to compare the results. We utilized the metric as used
by Crucitti et al. (2006). In this study, we calculated the normalized origin-destination
betweenness centrality, NB, of a node xas in 3.
NB
x,OD =1
(N−1)(N−2)
n
d=1
k
o=1
no,d(x)
no,d
(3)
where Nis the total number of nodes. The origin-destination betweenness cen-
trality is normalized to the maximum number of possible pairs of nodes that is
(N−1)(N−2)/2 (Freeman 1977). Then, for each scenario, we calculated the com-
plementary distributions, with the exceedance probability 1 −P(NB
OD)being a
function of the normalized origin-destination betweenness centrality values. All cal-
culations were developed by using R packages. Here, tails of the distributions were
of interest since they referred to the variability of the nodes with the highest values
in the distributions. Those nodes are the most critical since they control the highest
number of shortest paths inside the sub-graph.
22 Identifying Changes in Critical Locations for Transportation … 413
3 Results
The methodology was tested on a case study from a part of Kathmandu, Nepal
in order to understand the changes in important locations affected by disruptions
when traveling to a single health care service. Different centrality indices have been
investigated in order to identify structural properties of physical networks previously
(e.g. Zhong et al. 2016). The novelty of this study is that the methodology allows us
to focus on a single service in its own catchment area (i.e. service area) which can be
considered as a sub-graph. This provides more realistic results as in physical networks
such as transportation or even water distribution systems, supply and demand points
that are determinant in the traveling behaviors of communities.
3.1 Baseline Scenario for Kathmandu, Nepal
In this study, the catchment area of a hospital was created by the isochrone-based
approach using ArcGIS 10.4 Network Analyst tool which required assigning speed
limits to each edge. Then, the network was modeled as an undirected and weighted
network. The entire road network of Kathmandu, Nepal, is composed of 4379 nodes
and 5552 edges. Although large-scale health care facilities might serve for the entire
city, ambulances and emergency response services are required to reach the nearest
hospital within a 5-min timeframe as it is considered as a critical time limit for saving
lives. Therefore, in this study, in determining a catchment zone, we only considered
locations where a hospital could be reached within a 5-min timeframe. This means
that all nodes in the catchment area will be within a maximum of 5 min travel time
from their nearest hospital via the most practical route in the road network. Although
there are a number of hospitals and health care services in the study area which
should ideally cover the entire network, the methodology is demonstrated here for
a single hospital which is marked in Fig. 2. In order to identify, critical points in
other parts of the network or over the entire network, the proposed analysis must be
applied to all of the remaining hospitals.
Figure 2illustrates the important locations whilst accessing the selected hospital
which is located to the north of the study area. The selected hospital and its service
area are composed of 1982 nodes and 2400 edges which also reduces the computation
time for evaluating criticalities. Here, thematic maps display the origin-destination
betweenness centrality results. We classified the results by using the natural breaks
(i.e. using a Jenks algorithm) method, which maximizes the difference between
each cluster of results (ESRI 2018). As the betweenness centrality values are not
evenly distributed this visualization/clustering method is the most suitable one for
the case study. The results are displayed in 4 clusters to represents high, medium,
medium-low, and low criticalities. The most important nodes are illustrated as black
circles (i.e. high criticality) while the least critical nodes are smaller in size and
marked in a dark navy color (i.e. low criticality). Yellow circles illustrate a medium
414 N. Y. Aydin et al.
Fig. 2 The maps compare the betweenness centrality results in a hospital service area of Kath-
mandu. ashows the origin-destination betweenness centrality results, bis the betweenness centrality
results
level importance while green circles represent medium-low importance in the origin-
destination betweenness (Fig. 2a) or in the standard betweenness centrality (Fig. 2b)
results.
Figure 2a shows the most important nodes identified by the origin-destination
betweenness centrality results for the selected hospital as a destination node, while
Fig. 2b demonstrates the most important nodes of standard betweenness centrality
analysis in the same sub-graph. These results illustrate the difference between the
spatial distribution of importance values when traveling throughout the network and
traveling to a single location or a service in physical networks. Figure 2b results
obtained using a standard betweenness centrality which indicates the nodes that are
located within the hinterland of the selected hospital are deemed as low and medium-
low levels of criticality. The reason for this is because the standard betweenness
centrality is evaluated by considering all shortest paths between all node pairs. On the
other hand, the most critical locations in Fig. 2a are distributed around the selected
hospital that is estimated by using the origin-destination betweenness centrality.
Specifically, it is meaningful since cities are composed of different zones such as
housing areas, job centers and industrial locations that influence the criticalities in
transportation networks, because origin-destination pairs and flows depend on those
districts.
A random disruption scenario requires the identification of the sample sizes which
is based on removing the nodes and adjacent edges and evaluating the normalized
GCC metric in a hospital’s service area (i.e. in the sub-graph). This step is applied
22 Identifying Changes in Critical Locations for Transportation … 415
Fig. 3 Determining sample size; aSample size simulation for the fractions from 0.01 to 0.45,
bsample size simulations for the fractions from 0.5 to 0.9
to decrease the variations in the randomization process and to present the near real-
world representation in the study area. Figure 3shows the results of the connectivity
of the Kathmandu sub-graph (i.e. the selected hospital service area).
For a detailed investigation, 19 different fractions were used to calculate the
sample size. Random removal simulations were applied initially with a sample size
of 5 that was later increased to 150. Note that the larger sample sizes increase the
computation time specifically when evaluating the origin-destination betweenness
centrality. Based on Fig. 3.a, the lines that represent the normalized mean GCC
becomes steadier after the simulations were repeated 125 times, as seen in Fig. 3b
where some of the normalized mean GCC, specifically the fractions 0.6 and larger,
converge at this point. This means that variations decrease at this level and increasing
the number of simulations would not change the results drastically, while it would
increase the computation time. Therefore, all simulations were applied with a sample
size of 125 while evaluating the origin-destination betweenness centrality.
3.2 Random Disruption Scenario for Kathmandu, Nepal
Disruptions in physical networks that are vulnerable to geohazards, can have detri-
mental consequences. The Gorkha earthquake in 2015 with a magnitude of 7.8 (Mw)
and severe aftershocks left the Kathmandu city and the surrounding valley with sig-
nificant damages to both lives (8800 people were killed) and properties (estimated to
be around 798000 houses) (Shakya and Kawan 2016;Lamaetal.2017). In particular,
during these critical times, it is crucial to provide accessibility to health care centers
to save lives and prevent further injuries.
416 N. Y. Aydin et al.
In this scenario, the network was put under increasing stress to identify the changes
in important locations when traveling to the hospital. For this purpose, the fraction
of network nodes and the adjacent road segments were removed from the network
and the origin-destination betweenness centrality were evaluated for the remaining
sub-graph. Hence, decision-makers can have an insight into the important locations
and enhance the capacity of those road segments while recovering the damaged ones
in order to sustain a satisfactory level of service.
The simulations were applied for 19 different fractions, but only 3 of them are
illustrated here. Figures 4,5, and 6corresponds to factions 0.01, 0.05 and 0.1. We
selected those fractions as they also represent critical results when we investigated
the origin-destination betweenness centrality distributions which will be discussed
in Sect. 3.3. In each fraction, the number of randomly selected nodes was removed
125 times (i.e. the sample size determined as explained in Sect. 3.1) and the mean
origin-destination betweenness centrality was evaluated. Mean values for the origin-
destination betweenness centrality as well as the difference index using Eq. 2.2 are
presented in Figs. 4,5and 6.
In Figs. 4a, 5a, and 6a, the point data is displayed in 4 clusters (i.e. high importance
in black, medium importance in yellow, medium-low importance in green, and low
importance in navy). A negative value of difference index indicates that the node
importance increased, which was also displayed as red points, while the positive
Fig. 4 Fraction 0.01 nodes removed. aOrigin-destination betweenness centrality: There are 15
nodes represented in black, 30 nodes in orange, 111 nodes in green color. bThe difference index:
There are 35 nodes represented in red, 26 nodes in grey color
22 Identifying Changes in Critical Locations for Transportation … 417
Fig. 5 Fraction 0.05 nodes removed. aOrigin-destination betweenness centrality: There are 15
nodes represented in black, 38 nodes in orange, 163 nodes in green color. bThe difference index:
There are 69 nodes represented in red, 36 nodes in grey color
Fig. 6 Fraction 0.1 nodes removed. aOrigin-destination betweenness centrality: There are 17
nodes represented in black, 45 nodes represented in orange, 212 nodes represented in green color.
bThe difference index: There are 109 nodes represented in red, 42 nodes represented in grey color
418 N. Y. Aydin et al.
difference index means that the node importance reduced which is displayed as grey
points in Figs. 4b, 5b, and 6b.
While the disruption intensity increases, the number of nodes that is removed
increases as well. Decision makers can be aware of the expected changes in criticali-
ties when the stress level is at its peak (e.g. 10% of the entire network is disrupted) or
when there is a low stress (i.e., 1%) in the network. As can be seen from the Figs. 4b,
5b and 6b, the number of red points is increasing meaning that more nodes become
progressively important (i.e. used more) when the disruption magnitude increases.
While the red points accumulated on specific routes towards the selected hospital in
fraction 0.01 (see Fig. 4b), they extend further and surround the selected hospital as
a result of an increase in disruption intensity. This can be interpreted as the group
of nodes which were not on the route when traveling to the selected hospital in the
baseline scenario, now providing alternative routes that allows the selected hospital
to be accessed during the disruption. This information can be used in the pre-disaster
recovery planning phase, thereby allowing the capacity of these roads to be improved
to tolerate altered traffic loadings due to disasters.
3.3 Origin-Destination Betweenness Centrality Distributions
for Kathmandu, Nepal
In this section, we investigate the distribution of origin-destination betweenness cen-
trality for all random distribution scenarios as well as the baseline condition. Firstly,
we analyze the maximum values of the normalized origin-destination betweenness
centrality. Those nodes are the most important nodes since they refer to the locations
with the greatest number of shortest paths. The maximum values decrease gradually
with the increase of the disruption fractions, from 1.6 ×10−4at the baseline scenario
to 1.3 ×10−7at the scenario with a 0.7 disruption fraction. These findings show that
the normalized origin-destination betweenness values are reduced by the increase of
the disruption magnitude. Secondly, we observe the entire complementary distribu-
tion functions, with the exceedance probability 1 −P(NB
OD)being a function of
the normalized origin-destination betweenness values. This allows us to observe the
impacts of random disruptions on the sub-graph and provides an easy comparison.
Figure 7shows the tails of the distributions for all scenarios. We are specifically
interested in the tail as it illustrates the changes that occur due to random disruptions
on nodes that have the highest origin-destination betweenness value.
The distribution function for the baseline scenario (marked as black in Fig. 7)
shows lower exceedance probabilities than the scenarios with disruption fractions
from 0.01 to 0.2 up to a normalized origin-destination betweenness value of around
1.4 ×10−5. This is also the indication that some of the nodes’ origin-destination
betweenness values in these scenarios are increasing. Later, around 1.4 ×10−5,the
curves switch positions and the distribution of the baseline scenario is higher than
all the other distributions. The distributions of the scenarios (i.e. fractions from 0.25
22 Identifying Changes in Critical Locations for Transportation … 419
Fig. 7 Distributions of nodes’ normalized origin-destination betweenness centrality for different
scenarios. Disruption scenarios are characterized by different disruption fractions, which range here
from 0.01 to 0.7
to 0.7) are always lower than the baseline distribution and the scenarios with low
disruption fractions (i.e. up to 0.25). Their distribution functions decrease gradually
accordingly with the increasing magnitude of disruption. These findings suggest
that for high probabilities of disruption (fractions from 0.25 to 0.7) the road system
changes significantly compared to when it is under lower probabilities of disruption
(fractions from 0.01 to 0.2). Under high probability of disruption, the majority of
origin-destination betweenness centralities of nodes were reduced to 0 and very
few nodes are critical. This is due to the lack of availability of shortest paths in
the sub-graph. Overall, these findings show that the sub-graph collapses with high
magnitudes of disruption (see the distribution fractions from 0.25 and above which
is marked from green to purple color code in Fig. 7).
420 N. Y. Aydin et al.
4 Conclusion
Transportation networks provide access to service sectors that are vital to societies
and rural communities. Therefore, it is essential that transportation networks absorb
and rebound from disruptions caused by natural disasters, malicious attacks or any
unexpected events. Identifying critical locations and estimating the possible conse-
quences of disruptions and changes in criticalities are some of the challenges that
decision-makers are facing today. Complex network theory and applications can
provide essential tools to deal with these problems and to improve the resilience of
infrastructure systems, specifically, transportation networks.
In this study, the main goal was to characterize the networks’ topological prop-
erties in the baseline condition as well as under random disruption scenarios when
traveling to a specific service. This also allowed us to observe the limit of degradation
in a catchment area for a hospital. The second goal was to investigate the changes of
the node importance in catchment areas by using the origin-destination betweenness
centrality when the network is subject to those random disruptions.
Here, instead of using a traditional betweenness centrality method which has
been previously used for evaluating the critical nodes in transportation networks
(Porta et al. 2006b,2011), the origin-destination betweenness centrality is used by
modifying it to reflect the changes in criticalities when traveling to a specific function
or a location. Results were investigated in terms of the spatial distribution of critical
locations as shown in figures from 2 to 6 as well as the probability distribution of the
normalized origin-betweenness centrality as given in Fig. 7. Overall, the standard
betweenness centrality (i.e. the number of times a node was passed by while traveling
throughout the network) indeed does not capture the essential information when
decision-makers seek to investigate scenarios such as what happens after a disaster
when people need to reach to the nearest hospital or an emergency management
center.
In addition, physical networks such as transportation networks have intrinsic prop-
erties and functions that are planned to serve the public with close proximity in cities,
especially hospitals. In these cases, it is not representative to look at the topology
holistically, but it is necessary to investigate the sub-networks that developed organ-
ically over time or were created by decision makers. This study closes the gap by
proposing the application of the powerful spatial analytical tool of origin-destination
betweenness centrality.
Furthermore, the distribution analysis shows that the normalized origin-
destination centrality is distributed differently according to the level of the disruption
magnitude. Overall, this method can assist decision-makers and stakeholders in two
ways. Firstly, the results assist in identifying the disruption fraction of a system fac-
ing collapse using distribution analysis. They can test what is the maximum limit
of degradation for a particular road system and also its efficiency in serving hospi-
tals or emergency centers. Secondly, below this failure point, decision-makers can
identify the locations of those changes (i.e. increasing or decreasing values of origin-
22 Identifying Changes in Critical Locations for Transportation … 421
destination betweenness) and take adaptation measures, through enhancing resilience
by designing alternative paths.
There are several points in this study that would benefit from improvement in the
future. The first point is that the accessibility evaluation and/or creation of catch-
ment areas for services are a data intensive process. As the proposed betweenness
evaluation is applied to the sub-graph that is created based on the catchment area,
the boundaries directly affect the origin-destination betweenness centrality results.
Therefore, the size of the catchment area should be identified carefully.
The second point is that a node can belong to multiple service areas. As a result,
it might have a multiple origin-destination centrality value. In addition, a single
node can be associated with multiple catchment areas for accommodating different
functions. These two possibilities create a variability in the origin-destination cen-
trality values associated with a single node. This study does not provide a universal
value of importance but rather one value for each particular case. In other words,
a node can be most important in one sub-graph but not necessarily important in
another hospital’s catchment area. Future research in this field could investigate how
the variability of importance of one node changes when using multiple functions
and/or service areas in Kathmandu. One solution for this could be evaluating the
origin-destination betweenness centrality by combining catchment areas for a single
function and assigning a group of target nodes. This would provide a solution for
how criticalities would change when traveling to hospitals in an urban context. In
case a node is included in service areas for different functions (i.e. hospitals, subway
stations, etc.), a weighting factor can be assigned.
This study was concerned with the topology of networks and does not consider
the capacity or hierarchy levels of road segments. These are important properties for
designing networks and managing flows and later could be included into our analysis
by using penalty functions or weighting approaches on the edges. Our structural
analysis provides a basis for any traffic flow management. Without understanding the
functional properties of road networks, analyzing or managing traffic flows would
not be possible, as the foundation should be substantial enough to determine the
flows with a high level of confidence. Transportation resilience can be feasible with
a functional topology and traffic management. However, this will be the next step
to evaluate the flow resilience of networks to provide an insight into the speed of
recovery for transportation networks.
Even though this study focuses on a single hospital and its catchment area, it could
be applied to any other services (i.e. business hubs, housing area, shopping centers)
in other cities as well. For example, for a metropolitan city in a developed country
that is not subject to natural disasters, a random disruption scenario may represent
the intensity of road closures due to congestion, traffic accidents or malicious attacks.
The results would assist in identifying the changes in critical roads when people are
traveling to a housing area from job centers.
Finally, we conclude that the consequences of extreme events also depend on the
topological characteristics of cities, which are the results of the particular growth
process and geography of a place. Previous studies have investigated the standard
betweenness centrality characteristics of different cities (Kirkley et al. 2018; Crucitti
422 N. Y. Aydin et al.
et al. 2006) and found that some cities are comparable to each other and not unique
as they share the same topological properties in their standard betweenness central-
ity distributions. We expect that using the origin-destination betweenness centrality
on different urban road networks would add value and reveal the complexity and
differences of cities that the standard betweenness centrality analysis cannot reveal.
Overall, origin-destination betweenness centrality is a useful tool for identifying
the changing importance in the degrees of nodes. Simulating random disruptions
can assist decision-makers in terms of identifying the consequences of disruptions
with different magnitude levels and ultimately assist in enhancing resilience for
communities.
Acknowledgements This work was funded by a grant to N.Y.A., Y.C., and H.R.H. from the
National Research Foundation of Singapore (NRF) under its Campus for Research Excellence
and Technological Enterprise (CREATE) program (FI 370074011) for the Future Resilient Systems
project at the Singapore-ETH Centre (SEC) and by an Alexander von Humboldt Foundation Georg
Forster Experienced Researcher Fellowship Grant to H.S.D.
References
Albacete X, Olaru D, Paül V, Biermann S (2017) Measuring the accessibility of public transport: a
critical comparison between methods in Helsinki. Appl Spat Anal Policy 10(2):161–188. https://
doi.org/10.1007/s12061-015-9177-8
Aydin NY (2018) Measuring topological and operational resilience and recovery of water networks
for planning and management. In: World environmental and water resources congress 2018:
hydraulics and waterways, water distribution systems analysis, and smart water, pp 370–379.
https://doi.org/10.1061/9780784481424.039
Aydin NY, Duzgun HS, Heinimann HR (2017) Resilience evaluation for transportation network
accessibility under seismic hazards. Paper presented at the 15th international conference on
computers in urban planning and urban management (CUPUM), Adelaide, Australia, July 11–14
Aydin NY, Duzgun HS, Heinimann HR, Wenzel F, Gnyawali KR (2018a) Framework for improving
the resilience and recovery of transportation networks under geohazard risks. Int J Disaster Risk
Reduction 31:832–843. https://doi.org/10.1016/j.ijdrr.2018.07.022
Aydin NY, Duzgun HS, Wenzel F, Heinimann HR (2018b) Integration of stress testing with graph
theory to assess the resilience of urban road networks under seismic hazards. Nat Hazards
91:37–68. https://doi.org/10.1007/s11069-017-3112-z
Barthelemy M (2004) Betweenness centrality in large complex networks. Eur Phys J B 38:163–168.
https://doi.org/10.1140/epjb/e2004-00111-4
Batty M (2001) Exploring isovist fields: space and shape in architectural and urban morphology.
Environ Plann B Plann Des 28:123–150
Crucitti P, Latora V, Porta S (2006) Centrality in networks of urban streets. Chaos 16(1):015113.
https://doi.org/10.1063/1.2150162
Delamater PL (2013) Spatial accessibility in suboptimally configured health care systems: a modi-
fied two-step floating catchment area (M2SFCA) metric. Health Place 24:30–43. https://doi.org/
10.1016/j.healthplace.2013.07.012
Ertugay K, Duzgun S (2011) GIS-based stochastic modeling of physical accessibility using GPS-
based floating car data and Monte Carlo simulation. Int J Geogr Inf Sci 25(9):1491–1506. https://
doi.org/10.1080/13658816.2010.528419
22 Identifying Changes in Critical Locations for Transportation … 423
Ertugay K, Argyroudis S, Düzgün H ¸S (2016) Accessibility modeling in earthquake case considering
road closure probabilities: a case study of health and shelter service accessibility in Thessaloniki,
Greece. Int J Disaster Risk Reduction 17:49–66. https://doi.org/10.1016/j.ijdrr.2016.03.005
ESRI (2018) Data classification methods. http://pro.arcgis.com/en/pro-app/help/mapping/layer-
properties/data-classification-methods.htm. Accessed 30 Aug 2018
Freeman LC (1977) A set of measures of centrality based on betweenness. Sociometry 40(1):35–41
Giustolisi O, Simone A, Ridolfi L (2017) Network structure classification and features of water
distribution systems. Water Resour Res 53(4):3407–3423. https://doi.org/10.1002/2016wr020071
Kazerani A, Winter S (2009) Can betweenness centrality explain traffic flow. In 12th AGILE inter-
national conference on geographic information science, Hanover, Germany
Kirkley A, Barbosa H, Barthelemy M, Ghoshal G (2018) From the betweenness centrality in street
networks to structural invariants in random planar graphs. Nat Commun 9(1):2501. https://doi.
org/10.1038/s41467-018-04978-z
Lama PD, Becker P, Bergström J (2017) Scrutinizing the relationship between adaptation and
resilience: longitudinal comparative case studies across shocks in two Nepalese villages. Int J
Disaster Risk Reduction 23:193–203. https://doi.org/10.1016/j.ijdrr.2017.04.010
Marra WA, Kleinhans MG, Addink EA (2014) Network concepts to describe channel importance
and change in multichannel systems: test results for the Jamuna River, Bangladesh. Earth Surf
Process Land 39(6):766–778. https://doi.org/10.1002/esp.3482
Newman MEJ (2010) Networks: an introduction. Oxford University Press, Oxford, UK
O’Sullivan D, Morrison A, Shearer J (2000) Using desktop GIS for the investigation of accessibility
by public transport: an isochrone approach. Int J Geogr Inf Sci 14(1):85–104. https://doi.org/10.
1080/136588100240976
Porta S, Crucitti P, Latora V (2006a) The network analysis of urban streets: a dual approach. Phys
A 369(2):853–866. https://doi.org/10.1016/j.physa.2005.12.063
Porta S, Crucitti P, Latora V (2006b) The network analysis of urban streets: a primal approach.
Environ Plan 33(5):705–725. https://doi.org/10.1068/b32045
Porta S, Latora V, Wang F, Rueda S, Strano E, Scellato S et al (2011) Street centrality and the
location of economic activities in Barcelona. Urban Stud 49(7):1471–1488. https://doi.org/10.
1177/0042098011422570
QGIS (2018) QGIS network analysis toolbox 3. https://root676.github.io/. Accessed 28 Nov 2018
Schintler LA, Kulkarni R, Gorman S, Stough R (2007) Using raster-based GIS and graph theory to
analyze complex networks. Netw Spat Econ 7(4):301–313. https://doi.org/10.1007/s11067-007-
9029-4
Shakya M, Kawan CK (2016) Reconnaissance based damage survey of buildings in Kathmandu val-
ley: an aftermath of 7.8 Mw,25 April 2015 Gorkha (Nepal) earthquake. Eng Fail Anal 59:161–184.
https://doi.org/10.1016/j.engfailanal.2015.10.003
Singapore-ETH Center (2014) Future Resilient Systems programme booklet. https://www.ethz.ch/
content/dam/ethz/special-interest/dual/frs-dam/documents/FRS-Booklet.pdf. Accessed 23 Feb
2019
Strano E, Nicosia V, Latora V, Porta S, Barthelemy M (2012) Elementary processes governing the
evolution of road networks. Sci Rep 2:296. https://doi.org/10.1038/srep00296
WangZ, Du C, Fan J, Xing Y (2017) Ranking influential nodes in social networks based on node posi-
tion and neighborhood. Neurocomputing 260:466–477. https://doi.org/10.1016/j.neucom.2017.
04.064
Ye P, Wu B, Fan W (2016) Modified betweenness-based measure for traffic flow prediction of urban
road. Paper presented at the Transportation Research Board 95th annual meeting
Zhao S, Zhao P, Cui Y (2017) A network centrality measure framework for analyzing urban traffic
flow: a case study of Wuhan, China. Phys A 478:143–157. https://doi.org/10.1016/j.physa.2017.
02.069
Zhong C, Schläpfer M, Müller Arisona S, Batty M, Ratti C, Schmitt G (2016) Revealing centrality
in the spatial structure of cities from human activity patterns. Urban Stud 54(2):437–455. https://
doi.org/10.1177/0042098015601599
Chapter 23
Efficient Regional Travel for Rescue
and Relief Activities in a Disaster
Toshihiro Osaragi, Masashi Kimura and Takuya Oki
Abstract Efficient and rapid rescue activities are vital in the immediate aftermath
of a large-scale disaster. However, the locations of demanders (those requiring spe-
cial care or assistance) and responders (those supporting or assisting the demanders)
are often widely separated. In this paper, we propose a method of supporting effi-
cient travel and navigation for rescue activities using fuzzy c-means clustering and
a genetic algorithm. We also propose an optimization method that takes into con-
sideration the difference in workload required by demanders, compatibility between
responders and demanders, and the urgency of demanders. We then demonstrate the
efficiency of our proposed method based on numerical simulations and field experi-
ments using a web application that incorporates the method.
Keywords Travel ·Rescue activity ·Fuzzy c-means clustering ·Simulation ·
Field experiment
1 Introduction
1.1 Research Background and Purpose
In the immediate aftermath of a disaster, rapid rescue/relief/assistance responses are
demanded, but it is difficult to accurately predict where and when these demands
will arise. Additionally, the people responsible for responding to these demands
T. Osaragi (B)·M. Kimura ·T. Oki
School of Environment and Society, Tokyo Institute of Technology, 2-12-1-M1-25 Ookayama,
Meguro-Ku, Tokyo 152-8550, Japan
e-mail: osaragi.t.aa@m.titech.ac.jp
M. Kimura
e-mail: kimura.m.an@m.titech.ac.jp
T. Ok i
e-mail: oki.t.ab@m.titech.ac.jp
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_23
425
426 T. Osaragi et al.
may be unable to respond immediately because they are also affected by the disas-
ter, or because they are located far from the disaster scene, etc. In other words, in
post-disaster responses, referring to the person requiring rescue/relief/assistance as
a “demander” and the person responding to the demand as a “responder,” we are
faced with the problem of deciding which responders to assign to which demanders.
Furthermore, responders often have to respond sequentially to multiple demanders
and it is necessary to ensure they travel efficiently to their required locations.
Demanders are assumed to be vulnerable people such as the elderly, young chil-
dren, people with disabilities, as well as people whose safety needs to be confirmed.
Evacuees seeking relief supplies, injured people, people who are trapped, and peo-
ple requesting firefighting assistance can also be thought of as demanders. Taking
a longer-term perspective, citizens requesting help to sort through and clear away
debris can also be regarded as demanders. Responders are assumed to be employees
of public organizations such as the Japan Self-Defense Forces, the police and fire
departments, as well as welfare commissioners, residents’ association representa-
tives, volunteer firefighters, and possibly student volunteers. Previous disasters have
seen imbalances and delays in the assignment of responders to demanders, and this
is regarded as an issue to be addressed (Tanaka 2018; SankeiBiz 2018).
Specifically, in the chaotic situation immediately following a disaster, it is nec-
essary to match demanders and responders who are spatially scattered at the time
of the disaster and to promptly determine which responder should travel to which
demander and in which order to respond to demands, so that maximum use of limited
time/personnel/supplies is made and damage is minimized.
This study assumes a situation in which information on the locations and numbers
of demanders and responders can be obtained by some method (Osaragi and Niwa
2018), and examines the problem of deriving efficient travel routes (hereafter referred
to as the “regional travel problem”) by rationally matching responders and demanders
in scattered locations after the disaster.
1.2 Relevant Past Research
The problem of multiple responders traveling to multiple demanders closely resem-
bles the Multiple Traveling Salesman Problem (hereafter, mTSP). Strictly speaking,
these problems are different in that the mTSP is a problem of minimizing the time
taken for the responders to complete their travel and return to their point of departure.
In contrast, the regional travel problem in this study is a problem of minimizing the
time taken for the responders to finish responding to all of the demanders. The mTSP
is known to belong to a class of problems called non-deterministic polynomial-time
hardness (NP-hard), which means that when the number of demanders and respon-
ders increases, it becomes difficult to find an exact solution in a finite amount of
time. Therefore, a large amount of research has been done on fast computational
algorithms and heuristics for approximate solutions.
23 Efficient Regional Travel for Rescue and Relief Activities … 427
For example, Ogawa and Inoue (2014) showed a method of finding an exact
solution to the mTSP by applying the Simpath algorithm (Knuth 2006), which allows
fast enumeration of all routes between two points, while Imada et al. (2016) proposed
a method of preferentially searching the vicinity of exact solutions by improving the
“local clustering organization” method. Additionally, Ono et al. (2004) examined
a method of deriving an approximate solution using the fuzzy c-means clustering
algorithm.
In addition to theoretical research on algorithm development, real-world applied
research is also being performed. For example, taking supply delivery as their sub-
ject, Nakayama et al. (2004) proposed an optimal delivery scheduling method using
genetic algorithms (GA) and fuzzy logic that considers conditions such as arrival
time specifications.
An example of research that addresses the problem of traveling to demanders
during a disaster is a study by Okabayashi et al. (2011), which considers the task of
delivering relief supplies as quickly as possible to multiple evacuation centers that
have run out of supplies. Additionally, Suto and Tokunaga (2002) used a simulation
in a virtual city to analyze the effect of possible vehicle travel speeds, the shape of
the emergency route network, distribution center layouts, and the number of delivery
vehicles provided, etc. on the efficiency of supply delivery to evacuation centers.
1.3 Structure of This Study
In Sect. 2of this paper, we examine a method of formulating the regional travel
problem in disasters and solving it efficiently using a solution to the mTSP. Here,
we show a method that takes the perspectives required in regional travel (differ-
ent workload and order of priority for each demander) into consideration. Next, in
Sect. 3, we attempt to evaluate the proposed method using a simulation Experiment
that simulates regional travel. In Sect. 4, we incorporate the proposed method into
a web application capable of capturing information such as the locations of deman-
ders/responders in real-time (Osaragi and Niwa 2018) and attempt to evaluate the
proposed method through field experiments using this web application.
2 Derivation of Methods of Efficient Regional Travel
in Disasters
2.1 Formulation and Solution of Regional Travel Problem
For the regional travel problem, not only is it necessary to derive a solution rapidly,
but it is also necessary to respond to various conditions and demands specific to
disasters. In this paper, we examine a solution to the regional travel problem with
428 T. Osaragi et al.
reference to the method proposed by Ono et al. (2004). This method is quite simple
and easy to incorporate into the system, and the calculation load time is relatively
small as far as authors’ experience.
Figure 1shows an overview of the solution to the regional travel problem proposed
in this paper. First, based on location information, a responder is assigned to each of
demanders. At this point, using a typical non-hierarchical clustering method (such
as the k-means method) means that every responder will definitely be assigned to
specific demanders, which risks the development of a localized solution.
Therefore, responders here are assigned stochastically using a fuzzy c-means
method. Specifically, the strength of affiliation uij (defined as the distance between
demander jand the center of gravity of the group of demanders that responder iis
responsible for) is calculated, and if the value of uij is at least a certain threshold (0.5
in this study), responder iis assigned to demander j. If the strength of affiliation uij to
all responders is below the threshold, the demander is not assigned and is regarded
as undecided. Next, a GA (Ni 1997) is used to search for responders to undecided
demanders. Specifically, the assignment of responders to demanders and travel routes
are explored in such a way that the time taken to respond to all of the demanders
(hereafter, travel completion time) is minimized.
2.2 Accounting for Workload Differences
The above discussion assumes that the workload (response time) required by each
demander is the same. However, this is not necessarily the case. For instance, the
simplest workload is to just confirm the safety of demanders, who are registered
beforehand in the system as persons who need special cares. If they are confirmed to
be safe and no damage, it doesn’t take much time to complete this task. In contrast,
if a demander is being injured or locked inside building, a responder should take
care of him/her and make an emergency call for additional helps. Therefore, in this
section, we examine a method that considers workload differences.
Assuming that multiple responders cooperate to respond to demanders requiring
large workloads, as shown in Fig. 2, and then using the smallest workload as a basic
unit, all workloads are divided into multiple workloads, which are then distributed in
the vicinity of the respective demander. At this point, the problem can be solved by
regarding the divided/distributed workloads as hypothetical demanders and apply-
ing the solution shown in Fig. 1. However, this calculation assumes that the travel
time between the demanders requiring large workloads and the demanders placed
hypothetically by dividing that workload is regarded as 0.
23 Efficient Regional Travel for Rescue and Relief Activities … 429
Fig. 1 Efficient travel for local rescue activity
430 T. Osaragi et al.
Fig. 2 Method to consider the difference of workload
2.3 Considerations for Compatibility and Urgency
In addition to the travel distance and travel time, the compatibility of responders to
demanders (the relationship between the type of work demanded and the abilities of
the responder, etc.) and the urgency level, which varies depending on the demander,
are considered to be important elements in regional travel. Therefore, in this section,
we examine regional travel that considers both compatibility and urgency.
First, we define the regional travel evaluation indices. The index of distance from
responder ito demander jis defined as Dij , the index of compatibility of responder
23 Efficient Regional Travel for Rescue and Relief Activities … 431
ito demander jas Rij, the index of urgency of demander jas Ej, and the product
of these indices as the index Yij (Fig. 3a). At this point, it is difficult to uniquely
determine how to set the values for the indices Dij,Rij , and Ejbecause they depend
on the situations, objectives, etc. of the responders and demanders. However, the
problem can be solved based on the distance index Dij by setting the compatibility
index Rij to a higher value to increase the compatibility between the responder and
demander; setting the urgency index Ejto a high value when the urgency of the
demander is high; and regarding index Yij as din Fig. 1. Figure 3b, c show examples
that consider compatibility and urgency, respectively. This method makes it possible
to achieve travel that reflects compatibility and urgency, in contrast to travel based
solely on the distance index.
Fig. 3 Method to consider the adaptability of a responder to a demander and the emergency
432 T. Osaragi et al.
Fig. 4 Study areas
In the travel assistance app, demanders and responders can register themselves
with their attributes including special demands or capabilities respectively. After the
event occurs, an area manager (a super user of the system) activates the system, and
responders are automatically allocated to demanders in the system. Any changes in
demanders or responders, the system can recalculate the optimum result in real time.
3 Evaluation of Regional Travel Using Simulation
3.1 Study Areas
Area A (approx. 1.6 km ×1.2 km) around Okusawa Station in Setagaya Ward, Tokyo,
was taken as the study area (Fig. 4, left). The walking speed of responders was set
with reference to values obtained from a preliminary experiment that envisioned
confirming people’s safety immediately after a disaster (Kimura et al. 2016).
3.2 Verification of Method that Considers Differences
in Workload
The effectiveness of regional travel that considers differences in workload was ver-
ified using a simulation. Here, it was assumed that 10 responders travel around to
50 demanders (300 s workload ×10 people, 10 s workload ×40 people) (Fig. 5a).
The locations of the demanders and the travel start points of the responders were set
randomly, and the simulation was run for a total of 50 cases.
23 Efficient Regional Travel for Rescue and Relief Activities … 433
Fig. 5 Effect of considering the difference of workload
Figure 5b shows the simulation results. In all 50 cases, the results show that taking
differences in workload into account makes it possible to shorten the travel comple-
tion time compared to cases when differences in workload are not considered for the
optimization process. In particular, the results confirm that when multiple deman-
ders requiring workloads of 300 s are located close together, multiple responders can
respond efficiently by sharing the work (Fig. 5c).
434 T. Osaragi et al.
3.3 Verification of Method that Considers Changes
in Numbers of Responders/Demanders
Here, it is presumed that new demanders and responders appear partway through
the workload after regional travel has begun. In this situation, recalculation is nec-
essary, but if the interval between recalculations is too short, there will be frequent
changes to responder assignment and travel routes, which may cause efficiency to
decline. Conversely, if the interval between recalculations is too long, responder
reassignments will be slow, which may also cause efficiency to decline. Therefore,
in this section, we examine the relationship between recalculation timing and travel
completion time when the number of responders and demanders changes over time.
As a numerical example, we examined a situation in which, after 10 responders
started traveling to 40 demanders, (a total of 10) new demanders appeared (in random
locations) at a rate of one demander every two minutes. Specifically, the simulation
was run for 1000 cases with different travel responder starting points and different
demander locations, while changing the recalculation interval a(min) from 2 min
(=directly after a new demander appeared) to 4, 10, and 20 min (=when all of the
new demanders had appeared) (Fig. 6a).
The results show that, contrary to our initial expectations, the smaller the value of a,
the shorter the travel completion time. In other words, the travel time is shortest when
recalculation was performed immediately after a new demander appeared (Fig. 6c,
left). Looking closely at the calculation results, it can be seen that in many cases only
the travel route of the responder in the vicinity of the new demander was changed,
and that there were few situations in which the other responders were affected.
Next, we examined a situation in which, after seven responders started traveling
to 50 demanders, (a total of three) new responders appeared (in random locations)
at a rate of one responder every five minutes. Specifically, the simulation was run
for 1000 cases with different responder travel starting points and different demander
locations, assuming two recalculation intervals b(min) (5 and 15 min) (Fig. 6b). The
results show that, similar to the situation of an increasing number of demanders, the
travel completion time was shortest when recalculation was done immediately after
a new responder appeared (Fig. 6c, right).
From the above, within the range of demander/responder ratio and density (area
of the study space) assumed here, we found that immediate recalculation in response
to the appearance of a new demander/responder is effective in terms of shortening the
travel completion time. However, when the number of demanders and responders is
large, the computation load increases, so further investigations into the relationship
with calculation time will be necessary.
23 Efficient Regional Travel for Rescue and Relief Activities … 435
Fig. 6 Influence of increase of the number of demanders or responders
4 Evaluation of Regional Travel Using Field Experiments
4.1 Field Experiment Method
We conducted four field experiments (1–4) to evaluate the solution to the regional
travel problem proposed in this paper. In Experiments 1 and 2, four responders
cooperated with each other to travel to and confirm the safety of 25 demanders in
the study area defined in Sect. 3, Area A (approx. 1.6 km ×1.2 km) (Fig. 7a). In
Experiment 1, the four responders worked to minimize the travel completion time
by communicating with each other using the text function in LINE, an existing
social networking service (SNS). They were provided with maps (a paper map and
436 T. Osaragi et al.
Google Maps) in order to determine the locations of the demanders. Additionally,
the responders conferred with each other every time they finished responding to a
demander by sending the demander’s name via LINE. The experiment was deemed
to have finished when the four responders decided/reported that they had finished
traveling to all of the demanders.
To summarize, in Experiment 1, responders had to determine which demanders
they would respond to and determine their own travel route, while constantly con-
firming the state of their work progress via LINE. In contrast, in Experiment 2, the
same four responders traveled within the region according to travel routes derived
using the proposed solution. Specifically, we incorporated the proposed solution into
a web application that we developed previously (Osaragi and Niwa 2018), and we
developed a new web application that allows the results to be displayed on a mobile
device (smartphone), which is hereafter referred to as the travel assistance app. The
locations of other responders and their work progress can be shared constantly via
the travel assistance app. The experiment was deemed to have finished when all of
the demanders had been responded to. The initial locations of responders and the
locations of demanders were identical in Experiments 1 and 2, but the initial locations
of individual responders were swapped to eliminate the learning effect.
In Experiments 3 and 4, 10 responders cooperated with each other to travel to and
confirm the safety of 50 demanders in Area B (approx. 0.5 km ×0.5 km) (Fig. 4,
right) (Fig. 8a). In Experiment 3, the responders traveled within the region using
LINE in the same manner as in Experiment 1; while in Experiment 4, they traveled
using the travel assistance app in the same manner as in Experiment 2. This time, as
discussed in Sect. 3.2, it was assumed that the workload required by each demander
differed (10 s workload ×40 people, 300 s workload ×10 people). Here again, the
initial locations of responders and the locations of demanders in Experiments 3 and 4
were identical. However, similarly, the initial locations of responders were swapped
so that each responder started traveling from a different initial location.
4.2 Field Experiment Results
Figure 7c, d show the responder trajectories in Experiments 1 and 2, respectively.
Here, it can be seen that in Experiment 1 using LINE (Fig. 7c), multiple responders’
lines of travel intersect; whereas, in Experiment 2 using the travel assistance app
(Fig. 7d), the responders’ lines of travel do not intersect and the travel routes are
more efficient. Looking at the relationship between time elapsed since the start of
the experiment and the number of demanders responded to, we find that in Experiment
1 it took more than 60 min to travel to 25 demanders. In contrast, in Experiment 2
the travel was completed in less than half the time, approximately 30 min, and it is
clear that more efficient regional travel was achieved (Fig. 7e).
Figure 8c, d show the results of Experiments 3 and 4, which took workload dif-
ferences into account. The travel completion time was approximately 20% shorter
in Experiment 4 (using the travel assistance app) compared to Experiment 3 (using
23 Efficient Regional Travel for Rescue and Relief Activities … 437
Fig. 7 Field experiment of the proposed method for local rescue activities
438 T. Osaragi et al.
Fig. 8 Field experiment considering the difference of workload
23 Efficient Regional Travel for Rescue and Relief Activities … 439
Tabl e 1 Results summary of field experiment and simulation
Number of
demanders
Number of
responders
Area Method Completion
time
Experiment 1 425 ALINE 62 min
Experiment 2 425 ATravel
assistance
app
28 min
Experiment 3 10 40 (10s)
10 (300s)
BLINE 20 min (not
completed)
Experiment 4 10 40 (10s)
10 (300s)
BTravel
assistance
app
16 min
Simulation 10 40 (10s)
10 (300s)
BSimulation 14 min
LINE). In particular, when LINE was used, efficiency declined rapidly after approx-
imately half of the travel was completed because it was difficult to determine the
assignment of demanders rationally while taking workload differences into account
(Fig. 8c). Furthermore, due to the difficulty of sharing information between respon-
ders, there were two omissions and four duplications of demander responses. Similar
omissions and duplications are said to have occurred during the task of confirming
the safety of people requiring special help after the Great Hanshin-Awaji Earthquake
(1995) and the Great East Japan Earthquake (2011) (Usui et al. 2013; Takamura and
Yamada 2018).
In contrast, in the experiment using the travel assistance app, the responders man-
aged to complete their travels without a decline in efficiency and neither omissions
nor duplications occurred. Additionally, the results largely approximate the results
of the multi-agent simulation that reflected the walking speeds of responders and
work processing times assumed in the travel assistance app. Hence, it is clear that
travel in accordance with the optimization calculation was achieved.
The above series of results demonstrates the validity of our solution to the regional
travel problem proposed in this paper and the travel assistance app incorporating this
solution.
Table 1summarizes the results and enables comparison between both the field
experiments and the simulation. Obviously, the travel assistance app is superior to
an existing social networking services. The simulation result shows a little better
performance than the field Experiment 4. We can compare each responder’s activity
with the corresponding agent’s activity, and can improve the system to reflect the
individual characteristic of each responder such as walking speed and performance.
The computational load time for increases in the number of demanders and respon-
ders is not so serious. However, the authors consider that there is the adequate number
of users and the size of area for this system. Also, we should consider the case that
Internet connection by smartphones is not available after the event occurs. We would
like to address these issues in our future work.
440 T. Osaragi et al.
5 Summary and Conclusions
In this paper, we formulated the problem of a limited number of responders trav-
eling efficiently to a large number of demanders in the aftermath of a disaster for
rescue/relief as a regional travel problem. We also examined a solution that solves
this problem efficiently and attempted to verify this solution using simulations and
field experiments.
First, we proposed a solution to the regional travel problem with reference to
past research on the Multiple Traveling Salesman Problem (mTSP). Specifically, we
constructed a method of determining the assignment of responders to demanders and
the travel routes of responders from demander and responder location information
using fuzzy c-means clustering and a genetic algorithm (GA). We also proposed
a method that considers workload differences (response times) for each demander
and a method that considers the compatibility of responders to demanders and the
urgency of demanders.
Next, we evaluated the proposed regional travel solution through simulations,
the results of which showed that (1) in a case in which 10 responders travel to 50
demanders (300 s workload ×10 people, 10 s workload ×40 people), the solution
is particularly effective when demanders requiring large workloads are distributed
unevenly, and (2) when the numbers of demanders and responders change, immediate
recalculation is effective in shortening the travel completion time.
We also carried out field experiments using a Web application incorporating the
proposed solution (travel assistance app), which showed that (1) using the app can
significantly shorten travel completion time compared to using an existing SNS
(LINE), and (2) neither omissions nor duplications occur, thereby indicating that
efficient regional travel can be achieved when the app is used.
Acknowledgements A portion of this work is supported by Cross-ministerial Strategic Innovation
Promotion Program (SIP). The authors wish to express their sincere thanks to Japan Science and
Technology Agency (JST).
References
Imada S, Nakajima J, Furukawa M (2016) Practical solution for n-TSP by use of LCO. The Japan
Society for Precision Engineering Spring Meeting, H19
Kimura M, Osaragi T, Oki T (2016) Efficiency indices and traveling method for safety confir-
mation after a large earthquake. Papers and proceedings of the geographic information systems
association (CD-ROM), C-4-4
Knuth DE (2006) The art of computer programming, 14, Fascicle 1
Nakayama T, Maeda M, Nakanishi S (2004) Transport planning of supplies using genetic algorithms
under consideration for roads of plural vehicles. In: The 21st fuzzy system symposium, vol 20,
pp 310–313
Ni Y (1997) A genetic algorithm for the traveling salesman problem. J Fac Int Stud Utsunomiya
Univ 3:31–40
23 Efficient Regional Travel for Rescue and Relief Activities … 441
Ogawa M, Inoue M (2014) Enumerating all the optimal solutions of multiple travelling salesman
problem by using Simpath Algorithm. IEICE Technical Report IBISML 2014-78, Information
Processing Society of Japan, vol 114(306), pp 321–328
Okabayashi K, Nakamura A, Ando N, Yamada T, Taniguchi E (2011) Distribution model for
relief supply considering the priority of refuses in the aftermath of a disaster. Proc JSCE D1,
67(5):887–897
Ono T, Kanagawa A, Takahashi H (2004) Heuristic solution for multiple traveling salesman prob-
lems with multiple depots using fuzzy C-means clustering method. IEICE, A, J87-A(7):938–948
Osaragi T, Niwa I (2018) Development of system for real-time collection, sharing, and use of dis-
aster information. In: The 21st AGILE conference on geographic information science, geospatial
technologies for all, Lecture notes in geoinformation and cartography. Springer, pp 211–229
SankeiBiz (2018) West Japan heavy rain—shortage and spatial bias of volunteers, https://www.
sankeibiz.jp/econome/news/180723/ecc1807230901004-n1.html. Accessed 7 Aug 2018
Suto A, Tokunaga Y (2002) A study on relief goods distribution planning based on suffering
situations and countermeasures. Proc JSCE 695(IV-54):67–75
Takamura H, Yamada R (2018) A study on construction of supporting activities to monitor the
elderly in disasters in the large cities -Through the activities in Sumida Ward in the Great East
Japan Earthquake, vol 5, pp 36–42. Studies on Social Welfare, Toyo University
Tanaka T (2018) Support for disaster prevention volunteers, review and recommendation database
of 10 years restoration. Website of Hyogo Prefecture, 167. https://web.pref.hyogo.lg.jp/kk41/
documents/000039304.pdf. Accessed 7 Aug 2018 (in Japanese)
Usui M, Hatayama M, Fukuyama K (2013) A study on safety confirmation with information system
in local community. J Soc Saf Sci 16:1–10
Chapter 24
A Two-Stage Process for Emergency
Evacuation Planning: Shelter
Assignment and Routing
Ali Soltani, Andrew Allan and Mohammad Heydari
Abstract With the rapid growth of population and volume of urban flows, cities
have become more vulnerable to uneven natural and man-made disasters. In this
chapter, we applied a two-stage approach to first find the most appropriate sites
for shelters based on the multi-criteria decision-making (MCDM) technique, then an
algorithm for determining the best routes of evacuation under an emergency situation
was examined. Two well-known scenarios Capacity-Aware Shortest Path Evacuation
Routing (CASPER) and the Shortest Path (SP) were applied, then the results were
compared together. The CASPER scenario, based on the navigation time and traffic
volume of the network, required a longer navigation distance than the SP scenario,
although it considered road capacity and the volume of traffic, in conjunction with
the minimization of total evacuation time. The case study of research was the City
of Sadra, a new town in southern Iran, in the Middle East.
Keywords Evacuation planning ·Emergency management ·Spatial allocation ·
Passive defense ·Route optimization ·Shelter
1 Introduction
With the growth of population and activity levels, cities have become more vulnerable
to various disasters. A natural response from individuals and public society against
the emergence of hazardous events is population evacuation in order to decrease
A. Soltani
Shiraz University, Shiraz, Iran
e-mail: Soltani@shirazu.ac.ir
A. Allan (B)
Urban and Regional Planning, School of Art, Architecture and Design,
University of South Australia, Adelaide, Australia
e-mail: Andrew.Allan@unisa.edu.au
M. Heydari
University of Tarbiat Modares, Tehran, Iran
e-mail: mohammad.heydari@modares.ac.ir
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_24
443
444 A.Soltanietal.
the risk and limit exposure to dangers (Harris et al. 2015). Due to this issue, there
is an increasing demand for evacuation planning in emergency management litera-
ture (Goerigk et al. 2014). Evacuation planning is a key component of emergency
preparedness as one of the four major phases of the emergency management cycle
(FEMA 1990;Cova1999). To support evacuation planning an integrated analysis of
heterogeneous spatial datasets including population, road network and facilities is
required (Liu and Lim 2015). Evacuation is a process in which threatened people are
displaced from dangerous places to safer places.
The decision to evacuate is primarily based on concerning factors such as the
available shelters and evacuation time estimates (ETE) (Sorensen et al. 2004). Evac-
uation time estimates are strongly dependent on the capacity of the evacuation route
system being capable of handling the resulting traffic demand from an event (Lindell
and Prater 2007). Urban transportation networks are poorly designed to cope with
sudden increases in traffic flows (Shahabi and Wilson 2014). It is obvious that under
emergency situations, conditions are different because a large volume of vehicles
enter into the network within a short time period which leads to traffic congestion.
In this chapter, we present the process and results of a practiced location/routing
problem to facilitate evacuation planning in an emergency scenario. This research in
particular is aimed at determining the location of shelters, and the best routes that
are available for evacuees to reach a shelter. In this context, the research focuses on
providing results to two questions: (i) where should the population of a study area
evacuate to under an emergency situation, and (ii) which route is the best one?
This chapter is presented in two steps. Firstly, we have used the Fuzzy Analyt-
ical Hierarchy Process (FAHP) technique to determine the most suitable site for
emergency evacuation shelters; secondly, two routing algorithms available within
the custom network analyst routing tool in ArcGIS (Arc CASPER) were used and
evaluated for two scenarios: (i) a basic Shortest Path (SP) Algorithm; and (ii) a
Capacity-Aware Shortest Path Evacuation Routing (CASPER) Algorithm. GIS was
used as the source of data inputting, storing and retrieving georeferenced data, data
analyzing and then representing the results. The case study area of this research was
the new town of Sadra which is located 15 km northwest of Shiraz in southern Iran
(Fig. 1).
This chapter reviews the theoretical and methodological background by referring
to relevant studies. The next section then describes the research methodology and
the characteristics of spatial and non-spatial data collected from secondary sources.
It then describes the analysis undertaken to select the most suitable sites for shelter
establishment and the best routes for evacuation in a hazardous condition. The final
section provides the chapter conclusion and discussion on the main findings and
suggested directions for further research.
Research Significance and Contribution
As a prototype of an optimisation model, crowd dynamics are shown in a realistic
way. While most current models in the literature are based on theoretical data, this
research benefits from a practical project undertaken by the authors in Sadra City.
24 A Two-Stage Process for Emergency Evacuation Planning … 445
Fig. 1 Location of Sadra City, Iran
The other key contribution of this chapter is the comparative analysis of two distinct
optimization techniques (i.e. the Shortest Path Algorithm and the Capacity Aware
Shortest Path Algorithm). This in turn, makes it possible to evaluate two well-known
algorithms based on empirical outcomes.
2 Literature
2.1 Key Concepts
Many countries in contemporary times are at risk of terrorist attack. Cities are con-
sidered the main target for terrorism due to crowding of population and concentration
of infrastructure and services.
Development of highly resistant shelters for protection of a population against
terrorism and public evacuation of highly populated areas are some of the most
important measures in emergency management and passive defence in many coun-
tries. One key aspect of an emergency response is that evacuation can be defined as
the moving of residents from a given area that is a danger zone to safety as quickly
446 A.Soltanietal.
as possible and with the greatest reliability (Saeed Osman and Ram 2012). Evacu-
ation is defined as a significant component of emergency management as it entails
transporting people or valuable assets at risk to safer zones (Lim et al. 2013). Two
different approaches can be followed to facilitate the evacuation of a crowded popu-
lation in an emergency situation: (i) to establish a “shelter in place” which requires
people to stay in secure shelter buildings; or (ii) to “evacuate” which instructs peo-
ple to leave their present risky locations and move to secured spaces. Evacuation
planning consists of five distinct phases: (i) decision to evacuate; (ii) warning; (iii)
evacuation; (iv) shelter; and (v) return (Hamilton 2008). Accordingly, this research
is on the evacuation planning process. Two main issues in the decision phase of the
evacuation planning process include: (i) determining the suitable site of emergency
evacuation shelters; and (ii) generating evacuation routes. Emergency shelters have
an important role in protecting people against terrorist attack which can be regarded
as an effective security measure. Sufficient service capacity is necessary for deter-
mining suitable sites of emergency shelters (Chang and Liao 2015). Therefore, the
main factors that need to be considered in determining the location of emergency
shelters include: accessibility to major roads; accessibility to emergency facilities
(hospital, fire station, etc.); and their distance from vulnerable zones.
On the other hand, the level of service of the road network, trip generation rate
and average traffic speeds are critical priorities in evacuation planning (Zhang et al.
2015). Any transportation problem can deteriorate during an evacuation. For exam-
ple, notifying evacuees may be difficult, since traffic delays are common, and trans-
port arterials are often compromised by a hazard (Cova and Johnson 2003). Thus,
transport network analysis is one of the most critical steps of evacuation planning. An
evacuation network, which is generally represented by nodes and arcs, is a structure
of regions accessible to individuals during an evacuation (Zong et al. 2014).
In dealing with extreme disaster, many of the critical problems that arise are
inherently spatial, such as assessing the potential impact of a hazard, or an emer-
gency manager identifying the best evacuation routes during a disaster, or a civil
engineer planning a rebuilding effort following a disaster (Cova 1999). Most of the
data required for evacuation planning has a spatial component representing a signif-
icant opportunity to utilise GIS (Mansourian et al. 2006). In recent years, the use of
GIS in evacuation planning in response to natural disasters, accidents, or intentional
attacks have been increasingly considered (Saadatseresht et al. 2009; Widener and
Horner 2011). GIS is regarded as a priceless tool due to its wide capability to man-
age, analyze and visualize a network dataset for handling traffic flows, determine
appropriate evacuation routes, and identify safer spots within adjacent areas to a
hazardous site. In this research, GIS was used as the source of input data, to store
and retrieve geo-referenced data, for data analysis and to present the results of the
spatial analysis.
24 A Two-Stage Process for Emergency Evacuation Planning … 447
2.2 Background Studies
With the growth of new technologies, research interest in evacuation planning
emerged as a key challenge in emergency management and it has attracted increased
interest of governments and academia around the world (Xu et al. 2018). Recent
studies for evacuation planning can be divided into the two different categories: the
behavioral responses of evacuees during an evacuation (Trainor et al. 2012; Hou et al.
2014;Maetal.2016); and evacuation transportation modelling (Wang et al. 2014;
Tanachawengsakul et al. 2016; Shahabi and Wilson 2018).
The second category of evacuation studies with an emphasis on evacuation trans-
portation modelling includes: traffic demand modelling (Wilmot and Meduri 2005;
Hasan et al. 2013); and route choice during an evacuation (Zografos and Androut-
sopoulos 2008; Stepanov and Smith 2009; Chu and Su 2012; Guo et al. 2012).
Several methods and models have been used in evacuation routes planning. These
methods can be divided into simulation methods: Cellular automata modelling and
Agent-based modelling; and optimization methods (which includes a maximum-
flow/minimum-cut algorithm, an artificial intelligence algorithm and ArcCASPER)
(Santos and Aguirre 2004; Shahabi and Wilson 2014). Simulation methods are close
to reality and allow control of agent behavior to simulate “real life” situations. Disad-
vantages of these methods include: difficulty with validating and reproducing results
from the model; an overwhelming amount of data required to influence agent behav-
ior; and difficulties in disaggregating models (Harris et al. 2015). Recently, the usage
of behavioral models is suggested as being either cellular automata (CA) (Lu et al.
2017) or agent-based (Tan et al. 2015). Optimisation methods typically utilize mathe-
matical techniques to process data and suggest routes and the objective is to minimize
total evacuation time and reduce traffic congestion (Shahabi and Wilson 2014). By
considering our perspective goal in the practical project, the research reported in this
chapter is based on an optimization approach.
In determining the suitable location for shelter using quantitative models, there are
several precedents. Zhao et al. (2015) considered the effect of different earthquake
scenarios on evacuees when solving the problems of shelter location and evacuee
allocation for an earthquake shelter (Zhao et al. 2015). Multi-Criteria Decision Mak-
ing (MCDM) techniques have been preferred and commonly used by researchers
for the purpose of finding a suitable shelter location. Some include: AHP and GIS
combination (Choi et al. 2012), Fuzzy-AHP (Trivedi and Singh 2017; Hernández
et al. 2015)
448 A.Soltanietal.
3 Methodology and Data
3.1 Case Study Area
This study is focused on the new town of Sadra in Fars province, in Iran’s south. The
case study city has a population of 72,000 within an area of 1800 ha and is located
15 km northwest of metropolitan Shiraz.
The development of Sadra was planned to provide affordable housing for blue-
collar workers in Shiraz. However, Sadra is today regarded as an independent and
self-contained town, with its own facilities and infrastructure (Fig. 1). In this research,
a large-scale evacuation plan (for all residents) for Sadra city was investigated as a
location/routing problem. This plan aims to initially determine the location of shelters
(phase 1) and then determine the routes that evacuees should take to reach a shelter
(phase 2). Figure 2illustrates the detailed stages of the research.
We first identified a number of effective criteria for the location selection of an
emergency shelter, drawing upon crisis literature. We then applied the FAHP method
to determine the most suitable site for emergency shelters. The CASPER Algorithm
was chosen to determine an ordinary Shortest Path Algorithm and Dijkstra’s Algo-
rithm (this algorithm was used for finding the shortest path from an origin point to
a destination in a graph, for example, road networks). In this research, the aim of
selecting the CASPER Algorithm is its novel capacity-aware approach that can be
used in crisis situation.
3.2 Data
The three main datasets required for evacuation planning in this research included:
the location of shelters; the location of evacuee points; and the road network dataset.
The location of shelters is itself secondary data which was produced in phase 1 of
this research using MCDM techniques (FAHP). The location of evacuee points was
extracted from the Census block map. Census blocks are areas bounded on all sides
by several roads and represent the smallest level of a geographical district with basic
demographic data (such as total population by age, sex and etc.). This data was
collected from the latest Census data collected in 2017 by the Statistics Center of
Iran (SCI). The road network with attributes (e.g. number of lanes, road length) was
imported from Open Street Map (OSM map) and was refined by the authors based
on field observations. Then a network dataset was created using ArcGIS10. All of
these data were then imported into ArcGIS for further analysis. In this research, the
assumption is that each household will use its own vehicle for evacuation because
it was assumed that no collective transport service would be provided by the city’s
authorities. Accordingly, shelters are required to provide sufficient parking spaces
and evacuee capacity. The number of households in each block is assumed to be the
number of evacuees (or cars) in that block.
24 A Two-Stage Process for Emergency Evacuation Planning … 449
Fig. 2 Flowchart of the evacuation plan
450 A.Soltanietal.
4 Analysis and Results
4.1 Location Finding of Shelters
The steps of this phase include: (i) extraction of criteria for shelter site selection; (ii)
assigning weights to the criteria; (iii) generating the criteria and fuzzy criteria maps;
(iv) combining the criterion maps; and (v) determining a set of locations for shelters.
4.1.1 Selection Criteria
Several aspects should be considered when choosing the criteria for multi-criteria
analysis. A set of criteria should be measurable, comprehensive, without redundancy
and minimal (Malczewski 1999). Using different criteria, many studies have been
reported in the context of shelter site selection based on natural disaster types and
case studies (Kılcı et al. 2015). Through reviewing the literature and background
studies and personal knowledge and experience of the authors, a set of criteria were
developed (Table 1).
Tabl e 1 Selection criteria for shelter location
Criteria Var i a b l e n am e Definition Proposed by
Population density PopDEN Population by area
(person/ha)
Zhao et al.
(2017)
Distance to main roads DisToRD Network distance to nearest
main access road (m)
Trivedi and
Singh (2017)
Distance to police
stations
DisToPLC Network distance to nearest
police station (m)
Soltani and
Marandi (2011)
Distance to fire stations
(m)
DisToFireST Network distance to nearest
fire station (m)
Trivedi and
Singh (2017)
Distance to healthcare
facility (m)
DisToHPL Network distance to nearest
healthcare (m)
Liu et al. (2011)
Slope of urban land SLP Slope interpolation of lands in
the Sadra city district
Chen et al.
(2018)
Distance to vulnerable
facility
DisToVNF Network distance to
vulnerable facility (Gas
station and pipelines, power
and electrical infrastructure)
(m)
Trivedi and
Singh (2017)
Suitability of land-use LU Feasibility of land-use
conversion from current
status to shelter (open spaces
and parks are more suitable)
Chen et al.
(2018)
24 A Two-Stage Process for Emergency Evacuation Planning … 451
4.1.2 Assigning Weights to the Criteria
Generally, four different techniques are used for assigning weights including rank-
ing, rating, pairwise comparison, and trade of analysis method (Malczewski 1999). In
this research, pairwise comparisons using experts’ opinions, in the framework of the
FAHP technique, were used to assign the weight to each criterion. Initially, experts
(10 persons) completed the pairwise matrix using linguistic judgement individually.
After collecting each expert’s opinion, the validity of comparisons for each pairwise
comparison matrices were examined by applying an inconsistency ratio (to ensure this
was less than 0.1). The final matrix was obtained from the combination of elements
corresponding to each pairwise comparison matrices using fuzzy geometric means.
Next, the defuzzification of fuzzy numbers corresponding to each criterion was con-
ducted. For this purpose, the fuzzy number of each criterion,
Ci=(li,mi,ui),was
calculated using geometric means of each row (Eq. 1); then normalization of Ciwas
done using Eq. 2.
Ci=⎡
⎣
n
j=1
˜vij⎤
⎦
1
n
,∀i(1)
where
Ciis the fuzzy number of criterion i and ˜vij is the relative priority of criterion
i in respect to criterion j based on the geometric mean of experts’ opinions,
Ni=li
n
i=1ui
,mi
n
i=1mi
,ui
n
i=1li(2)
where
Miis the normalized fuzzy number of criterion i. As a result
Ni=(Li,Mi,Ui)
At the end of this step, the defuzzificated number of each criterion was calculated
using the centre of gravity method (Calabrese et al. 2016)(Eq.3).
Crisp
Ni=(Li+(2×Mi)+Ui)
4(3)
where Crisp
Niis the final weight of criterion i. The results are presented in Table 2.
4.1.3 Generate the Fuzzy Criteria Maps
The fuzzy criteria maps were generated for each of the eight criteria, then normalized
by classifying this into five levels, ranging from ‘very low’ to ‘very high’. It is obvious
that using such linguistic variables adds a certain level of uncertainty to the analysis,
as it may not always be easy to make a clear distinction between two different levels
of class (Ghajari et al. 2018). For this problem, based on the nature of criteria, the
linear fuzzy membership function in ArcGIS was used (ESRI 2018)(Eq.4).
452 A.Soltanietal.
Tabl e 2 Fuzzy pairwise matrix and criteria weights
Criteria DisToHPL …DisToFireST Normalized
geometric
means
Criteria fuzzy
weights
Final
criteria
weights
DisToHPL (1, 1, 1) …(1.2, 1.17,
1.39)
(0.66, 0.79,
0.97)
(0.063, 0.091,
0.135)
0.095
DisToRD (2.4, 3.2, 4.1) …(1.49, 1.86,
2.31)
(1.67, 2.11,
2.48)
(0.16, 0.243,
0.347)
0.248
DisToPLC (0.51, 0.61,
0.82)
…(0.59, 0.75,
0.88)
(0.43, 0.51,
0.64)
(0.041, 0.059,
0.09)
0.062
PopDEN (1.1, 1.4, 1.8) …(1.08, 1.37,
1.69)
(0.93, 1.15,
1.44)
(0.089, 0.133,
0.201)
0.139
LU (1.2, 1.8, 2.4) …(1.14, 1.42,
1.91)
(1.21, 1.48,
1.75)
(0.117, 0.17,
0.245)
0.175
DisToVNF (1.02, 1.12,
1.28)
…(1.15, 1.45,
1.84)
(0.79, 0.93,
1.1)
(0.076, 0.107,
0.154)
0.111
SLP (1.07, 1.29,
1.45)
…(1.07, 1.18,
1.24)
(0.74, 0.87,
1.06)
(0.071, 0.1,
0.149)
0.105
DisToFireST (0.71, 0.85,
0.83)
…(1, 1, 1) (0.68, 0.8,
0.94)
(0.065, 0.092,
0.131)
0.095
µ(x)=0if x <min,µ
(x)=1if x >max,
otherwise µ(x)=(x−min)
max −min
(4)
where minimum and maximum correspond to the ‘very low’ and ‘very high’ class in
the criteria maps. The values of the final fuzzy criteria maps are within the interval
[0, 1] and these maps are presented in Fig. 3.
4.1.4 Combination of Criterion Maps
Using fuzzy overlay operators, the eight criteria layers were aggregated. To determine
the suitable locations for shelter, the results have been compared (Fig. 4). Fuzzy
Gamma establishes the relationships between the multiple input criteria and does not
simply return the value of a single membership set as does “fuzzy OR” and “fuzzy
AND” (ESRI 2018). In this study the default value of the gamma operator (0.9) was
used, because this value presented better results in comparing it with other values.
The results of GAMMA and PRODUCT operators shows the lowest scoring area in
the “excellent” class in the overlaid maps (Table 3). The results obtained from fuzzy
operators suggest two suitable sites for the establishment of an emergency shelter in
Sadra City. According to the research assumptions, the complete evacuation of the
population was based on the use of private vehicles and in providing sufficient car-
parking capacity. Thus, the alternative location 1 was selected to fulfil this purpose.
In the following, the locations of these two sites are shown (Fig. 5).
24 A Two-Stage Process for Emergency Evacuation Planning … 453
Fig. 3 Fuzzy criterion maps
Tabl e 3 The suitability amount of operator classes (area)
Very poor Poor Moderate Good Very good Excellent
SUM 2.696 31.403 29.438 18.898 13.299 4.265
OR 0.661 44.737 2.151 26.624 12.875 12.951
AND 87.349 2.039 9.311 0.209 0.776 0.317
GAMMA 87.349 0.762 4.975 4.647 1.969 0.299
PRODUCT 99.074 0.652 0.217 0.036 0.019 0.002
454 A.Soltanietal.
Fig. 4 The overlay of criterion maps by fuzzy operators: aSUM; bOR; cAND; dGAMMA (0.9);
ePRODUCT
Fig. 5 Selected locations for shelter establishment
24 A Two-Stage Process for Emergency Evacuation Planning … 455
4.2 Best Evacuation Routes
In this chapter, two evacuation routing scenarios were proposed. The “Shortest Path
Algorithm” (Dijkstra’s Algorithm) is the first scenario in which it assumes selection
of the evacuation shortest path for each evacuee is based on there being negligible
traffic congestion. In other words, all routes are generated according to the shortest
distance to the shelter and the distance is assigned to each of road segments as the
cost factor. The second scenario involves the use of a Capacity-Aware Algorithm
in which, network capacity and the cost of the network segments grow during an
evacuation process, and the objective is to minimise the total time of evacuation whilst
minimising the distance travelled by evacuees. This algorithm determines realistic
traversal speeds and times for each road segment intelligently and dynamically. The
ArcCASPER tool in ArGIS was used to run both scenarios and then compare the
outputs.
4.2.1 Arc CASPER
The ArcCASPER tool is a custom Network Analyst tool (an evacuation routing
extension to ArcGIS Network Analyst) that uses a state-of-art routing algorithm
to produce evacuation routes to the nearest safe area for each evacuee or group
of evacuees based on the road capacity and the number of evacuees (Shahabi and
Wilson 2014). The CASPER tool generates an evacuation routing model as a graph
with four inputs: the graph (road network), the traffic model, the origin point of
evacuees, and the destination (shelters); therefore, the preparation of the three data
sets required for utilizing this tool included: creating the network dataset; specifying
the evacuation and safe zones; and selecting the traffic model for evacuation. The
CASPER Algorithm is embedded in this tool and the main purpose of using it is to
reduce the total evacuation time, minimize traffic congestion, and minimize exposure
to the risk. When evacuation routes to a shelter are generated, each of evacuees will
generate only one path to one of the shelter points. This evacuation path, produced
by the CASPER tool, can be described as an ordered set of edges that will direct
evacuees to safety. Thus, the total flow on an edge represents the sum of all flows
from all paths that pass through the same edge (Alabdouli 2015). A choice of 5
different traffic models are consolidated in the CASPER tool.
One of these traffic models is the Power traffic model which is the result of
fitting an empirical curve that is considered as fixable (Shahabi and Wilson 2014).
The predicted traffic congestion and estimation of evacuation time can be enhanced
using the power model and the results using this model is similar to simulation models
(Alabdouli 2015). In this chapter, the power traffic model was used to calculate the
evacuation time with congestion in both scenarios (number of evacuees).
456 A.Soltanietal.
4.2.2 Generated Routes
The results show that use of the CASPER Algorithm in evacuation routing led to
higher efficiency in utilizing all of the road network capacity and a more desirable
equilibrium in the distribution of through traffic. Outcomes included:
1. A decrease in traffic congestion: the average of traffic congestion (number
of vehicles per minute was reduced applying the CASPER Algorithm (45.5)
versus the other scenario (61.2). On the other hand, the maximum total through
traffic per road segment was decreased by using the CASPER Algorithm (2102
vehicles compared with 2437 vehicles) because of the full utilization of local
road capacities in the traffic assignment process (Fig. 6).
2. Reduction in the total evacuation time: The total evacuation time in the first
scenario (SP) was 73 min and in the second scenario (CASPER) 49 min. The
24 min difference in the total evacuation time, indicates better efficiency in using
the CASPER Algorithm in a large-scale evacuation routing. As shown in Fig. 7,
we compared the assigned route to the evacuation shelter based on two scenarios.
The results of this comparison show that traveling time in normal conditions
without taking the road capacity and through traffic congestion into account in
the CASPER Algorithm was 1.5 min more than for the SP Algorithm; but it was
reduced by 8 min when including the capacity and through traffic congestion in
a critical condition (Fig. 7).
As shown in Table 4, by using the CASPER Algorithm in the first half hour of
evacuation, 49% of population were evacuated to the shelter; however, during the
same time, with using SP Algorithm, only 13% of population were able to reach the
shelter.
The results of running two scenarios clearly indicate that the CASPER Algorithm
achieves better evacuation time and traffic congestion predictions in comparison to
the SP Algorithm.
Based on the through traffic congestion, the road segments were sorted and the
main roads in emergent situations were identified (Table 5). The results show that
the streets numbered 5 (Danesh); 6 (Pasdaran); 3 (Iran); 12 (Molana) and 1 (Sadra
Road), respectively, were the key streets used for the routes for evacuation. Therefore,
Tabl e 4 Comparison of the
cumulative frequency of the
evacuated population during
the evacuation process
CASPER SP
First 15 min 5.60 5.97
Second 15 min 49.02 13.13
Third15min 91.47 68.37
Fourth 15 min 100 98.16
Fifth15min 100 100
24 A Two-Stage Process for Emergency Evacuation Planning … 457
Fig. 6 Number of vehicles (evacuees) per road segment
458 A.Soltanietal.
Fig. 7 comparison of assigned routes with in two scenarios
Tabl e 5 Sadra’s Main Street in crisis situation
Street
No.
Street name Congestion (maximum
vehicle per min)
Priority
5 Danesh St. 45.5 A
6 Pasdaran St. 31.2 B
3Iran St. 20.1 C
12 Molana St. 18.5 D
1Sadra Rd. 14.8 E
for preparation before the disaster, a set of actions should be considered to reduce
vulnerability of the network while increasing the efficiency of these roads (Fig. 8).
24 A Two-Stage Process for Emergency Evacuation Planning … 459
Fig. 8 Sadra’s Main Street
in a crisis situation
5 Summary and Conclusion
In this chapter, we proposed a model for evacuation planning in two phases. In
the first phase, using the FAHP technique and GIS capabilities, two suitable sites
were proposed for emergency evacuation shelters. In the second phase, we chose
the ArcCASPER tool for generating optimum evacuation routes to each shelter. Two
different scenarios or routing algorithms were used for this purpose: the Shortest
Path Algorithm and the Capacity-Aware Algorithm. We concluded that:
The first scenario with routing based on the Shortest Path (SP) Algorithm and
which minimized the navigation distance regardless of considerations of roadway
capacity and the volume of through traffic.
The next scenario, using CASPER conducted a route optimisation based on nav-
igation time and the volume of through traffic. The routes assigned by this scenario,
although requiring a longer navigation distance than the previous scenario, took into
account the capacity and volume of transit traffic, and has the capacity to trans-
port evacuees in a shorter time in an emergency situation. Furthermore, CASPER is
recommended in solving a route-finding problem in a large-scale context.
While the usage of CASPER was helpful for the evacuation planning of Sadra
City, it suffers from several limitations. One of the limitations experienced was in
considering private vehicles as the only available means of transport, thus neglecting
460 A.Soltanietal.
the capabilities of public transport and pedestrian networks. Furthermore, the role of
some natural and built environment factors such as terrain topography, the geometric
layout of the road network and intersections were not examined. This study can be
improved by extending the list of decision criteria and considering more influential
factors in spatial location and allocation as discussed earlier. Also, in this research, we
used census blocks since this is the typical unit of resident population analysis used in
the census. However, the block size, as the unit of analysis used, may adversely affect
the model results and this impact can be investigated in the future. Furthermore, the
benefit of this research is that the usage of behavioral simulation models enables us
to model the evacuees’ transport behaviour at the micro scale.
References
Alabdouli K (2015) Tsunami evacuation: using GIS to integrate behavioral and vulnerability data
with transportation modelling. The Florida State University
Calabrese A, Costa R, Levialdi N, Menichini T (2016) A fuzzy analytic hierarchy process method
to support materiality assessment in sustainability reporting. J Clean Prod 121:248–264
Chen W, Zhai G, Ren C, Shi Y, Zhang J (2018) Urban resources selection and allocation for
emergency shelters: in a multi-hazard environment. Int J Environ Res Public Health 15(6):1261
Chang H-S, Liao C-H (2015) Planning emergency shelter locations based on evacuation behavior.
Nat Hazards 76(3):1551–1571
Choi J-H, Kang I-J, King S-S (2012) Study on the selection of the tsunami shelter using AHP and
GIS analysis. J Korean Soc Surv Geodesy Photogram Cartogr 30(6_1):503–509
Chu J, Su Y (2012) The application of TOPSIS method in selecting fixed seismic shelter for
evacuation in cities. Syst Eng Proc 3:391–397
Cova TJ (1999) GIS in emergency management. Geogr Inf Syst 2:845–858
Cova TJ, Johnson JP (2003) A network flow model for lane-based evacuation routing. Transp Res
Part A: Policy Pract 37(7):579–604
ESRI (2018) ArcGIS for desktop
FEMA (1990) Definitions of rerms. FEMA. Federal Emergency Management Agency. Instruction
Number 5000.2
Ghajari YE, Alesheikh AA, Modiri M, Hosnavi R, Abbasi M, Sharifi A (2018) Urban vulnerability
under various blast loading scenarios: analysis using GIS-based multi-criteria decision analysis
techniques. Cities 72:102–114
Goerigk M, Deghdak K, Heßler P (2014) A comprehensive evacuation planning model and genetic
solution algorithm. Trans Res Part E: Logistics Transp Rev 71:82–97
Guo R-Y, Huang H-J, Wong S (2012) Route choice in pedestrian evacuation under conditions
of good and zero visibility: experimental and simulation results. Transp Res Part B: Methodol
46(6):669–686
Hamilton J (2008) Mass evacuation planning. Director’s Guidelines for Civil Defence Emergency
Management (CDEM) Groups [DGL 07/08], Ministry of Civil Defence & Emergency Manage-
ment
Harris JB, Gregg C, Joyner TA, Luffman I (2015) Preparing a small town for a hazardous materials
incident: an examination of evacuation routing algorithms and plume models. Int J Geospatial
Environ Res 2(1):5
Hasan S, Mesa-Arango R, Ukkusuri S (2013) A random-parameter hazard-based model to under-
stand household evacuation timing behavior. Transp Res Part C: Emerg Technol 27:108–116
24 A Two-Stage Process for Emergency Evacuation Planning … 461
Hernández R, José G, García G, María J, Hernández G, Gilberto J (2015) Shelter selection with
AHP making use of the ideal alternative. Encyclopedia of information science and technology,
3rd edn. IGI Global, Hershey, pp 2003–2015
Hou L, Liu J-G, Pan X, Wang B-H (2014) A social force evacuation model with the leadership
effect. Physica A 400:93–99
Kılcı F, Kara BY, Bozkaya B (2015) Locating temporary shelter areas after an earthquake: a case
for Turkey. Eur J Oper Res 243(1):323–332
Lim H Jr, Lim MB, Piantanakulchai M (2013) A review of recent studies on flood evacuation
planning. J Eastern Asia Soc Transp Stud 10:147–162
Lindell MK, Prater CS (2007) Critical behavioral assumptions in evacuation time estimate anal-
ysis for private vehicles: Examples from hurricane research and planning. J Urban Plan Dev
133(1):18–29
Liu Q, Ruan X, Shi P (2011) Selection of emergency shelter sites for seismic disasters in moun-
tainous regions: lessons from the 2008 Wenchuan Ms 8.0 Earthquake, China. J Asian Earth Sci
40(4):926–934
Liu X, Lim S (2015) A spatial analysis approach to evacuation management: shelter assignment
and routing. Research@ Locate, Brisbane, Australia
Lu L, Chan C-Y, Wang J, Wang W (2017) A study of pedestrian group behaviors in crowd evacuation
based on an extended floor field cellular automaton model. Transp Res Part C: Emerg Technol
81:317–329
Ma Y, Yuen RKK, Lee EWM (2016) Effective leadership for crowd evacuation. Phys A 450:333–341
Malczewski J (1999) GIS and multicriteria decision analysis. Wiley, New York
Mansourian A, Rajabifard A, Zoej MV, Williamson I (2006) Using SDI and web-based system to
facilitate disaster management. Comput Geosci 32(3):303–315
Saadatseresht M, Mansourian A, Taleai M (2009) Evacuation planning using multiobjective evolu-
tionary optimization approach. Eur J Oper Res 198(1):305–314
Saeed Osman M, Ram B (2012) Application of static uncapacitated multi-commodity flow models to
dynamic capacity-constrained evacuation path networks.In: Proceedings of the 2012 International
Conference on Industrial Engineering and Operations Management Istanbul, Turkey, July 3–6,
2012
Santos G, Aguirre BE (2004) A critical review of emergency evacuation simulation models. Pre-
liminary Paper #339, Disaster Research Center, University of Delaware. http://udspace.udel.edu/
handle/19716/299
Shahabi K, Wilson JP (2014) CASPER: intelligent capacity-aware evacuation routing. Comput
Environ Urban Syst 46:12–24
Shahabi K, Wilson JP (2018) Scalable evacuation routing in a dynamic environment. Comput
Environ Urban Syst 67:29–40
Soltani A, Marandi EZ (2011) Hospital site selection using two-stage fuzzy multi-criteria decision
making process. J Urban Environ Eng 5(1):32–43
Sorensen JH, Shumpert BL, Vogt BM (2004) Planning for protective action decision making:
evacuate or shelter-in-place. J Hazard Mater 109(1–3):1–11
Stepanov A, Smith JM (2009) Multi-objective evacuation routing in transportation networks. Eur J
Oper Res 198(2):435–446
Tan L, Hu M, Lin H (2015) Agent-based simulation of building evacuation: combining human
behavior with predictable spatial accessibility in a fire emergency. Inf Sci 295:53–66
Tanachawengsakul T, Mishima N, Fuchikami T (2016) A simulation study on fire evacuation routes
in primary stage for a historic canal residential area. Proc-Soc Behav Sci 216:492–502
Trainor JE, Murray-Tuite P, Edara P, Fallah-Fini S, Triantis K (2012) Interdisciplinary approach to
evacuation modeling. Nat Hazards Rev 14(3):151–162
Trivedi A, Singh A (2017) Prioritizing emergency shelter areas using hybrid multi-criteria decision
approach: a case study. J Multi-Criteria Decis Anal 24(3–4):133–145
Wang Z, Zlatanova S, Moreno A, Van Oosterom P, Toro C (2014) A data model for route planning
in the case of forest fires. Comput Geosci 68:1–10
462 A.Soltanietal.
Widener MJ, Horner MW (2011) A hierarchical approach to modelling hurricane disaster relief
goods distribution. J Transp Geogr 19(4):821–828
Wilmot CG, Meduri N (2005) Methodology to establish hurricane evacuation zones. Transp Res
Rec 1922(1):129–137
Xu X, Zhang L, Sotiriadis S, Asimakopoulou E, Li M, Bessis N (2018) CLOTHO: a large-scale
internet of things based crowd evacuationplanning system for disaster management. IEEE Internet
of Things Journal
Zhang Z, Parr SA, Jiang H, Wolshon B (2015) Optimization model for regional evacuation
transportation system using macroscopic productivity function. Transp Res Part B: Methodol
81:616–630
Zhao X, Xu W, Ma Y, Hu F (2015) Scenario-based multi-objective optimum allocation model for
earthquake emergency shelters using a modified particle swarm optimization algorithm: a case
study in Chaoyang District, Beijing, China. PloS one 10(12):e0144455
Zhao X, Xu W, Ma Y, Qin L, Zhang J, Wang Y (2017) Relationships between evacuation population
size, earthquake emergency shelter capacity, and evacuation. Int J Disaster Risk Sci 8(4):457–470
Zografos KG, Androutsopoulos KN (2008) A decision support system for integrated hazardous
materials and emergency response decisions. Transp Res Part C: Emerg Technol 16(6):684–703
Zong X, Xiong S, Fang Z (2014) A conflict-congestion model for pedestrian-vehicle mixed evacu-
ation based on discrete particle swarm optimization algorithm. Comput Oper Res 44:1–12
Chapter 25
A Comprehensive Regional Accessibility
Model Based on Actual Routes-of-Travel:
A Proposal with Multiple Online Data
Yuli Fan, Qingming Zhan, Huizi Zhang and Jiaqi Wu
Abstract Accessibility models are important tools in evaluating economic potential
and estimating inter-city transport connections. However, existing models are mostly
based on infrastructure networks, where the actual arrangement of time-tables, road
condition etc. are not considered. Internet booking platforms and online digital maps
now provide detailed train and flight time-tables and accurate road trip recommen-
dations, which can be synthesized into travel routes that are very close to predicting
what occurs in reality. On this basis we propose a new structure of accessibility
model, where the accessibility between cities are represented as the accumulation of
all feasible travel routes, and the travel routes are weighted by their actual time and
financial cost. By validating the model with economic data and actual traffic volume
acquired through location-based services data, the model proves more effective than
traditional accessibility model and network indicators.
Keywords Regional accessibility ·Accessibility models ·Routes-of-travel ·
Multiple online data
1 Introduction
1.1 Quantification of Accessibility
Assessing the accessibility of city-pairs and cities is the goal of this study; Accessi-
bility, as an important concept in urban and regional planning, describes the general
cost of goods or people travelling from one place to another, or to all others, in
a given region. Previous research in economic geography has proved that there is
Y. F a n ·Q. Zhan (B)·H. Zhang ·J. Wu
School of Urban Design, Wuhan University, Wuhan, China
e-mail: qmzhan@whu.edu.cn
Y. F a n
e-mail: whu_fanyuli@foxmail.com
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_25
463
464 Y. F a n e t a l .
a strong relationship between accessibility and the potential of economic develop-
ment, encouraging many subsequent researchers to formulate accessibility models
or propose an accessibility-based planning framework (Jin 2010;Rich1978).
Traditionally, the quantification of accessibility is built on the basis of a network,
either as a network of roads, railways, ferry lines or airlines (Smith et al. 1970;
Straatemeier 2008; Zhang and Lu 2007). The connection between two neighboring
cities is regarded as a link, with simple properties such as length, road level, etc.,
and three criteria are employed to assess the accessibility between two given cities:
minimal cost, in terms of either total time or total money; complexity, stressing the
number of transfers needed; and accumulated opportunities, stressing the number
of potential routes in the network. However, some researchers believe that such
abstraction does not provide sufficient information on the actual convenience of
travelling between two cities (Rich 1978).
Nowadays, with the emergence of digital maps, online navigation services and
ticketing platforms, new possibilities are available to facilitate more in-depth studies
on this topic. Foremost, the connection between two cities is no longer regarded
as simple links, but can be deconstructed into numerous potential routes, provid-
ing a new basis for modelling city-pair accessibility; secondly, some very detailed
information on these routes can be introduced so that empirical studies can be more
pertinent to realistic topics. This study aims to take advantage of such possibilities,
and build an accessibility model for regional studies based on extracting, describing,
evaluating, and synthesizing actual travel routes.
1.2 Choice of Travelling Routes
The setting of preferabilities for travel routes weights all possible routes in calculating
city-pair accessibility in this study. With reference to how policies affect the overall
distribution of traffic volume among different routes, travel prediction models are
built by policy researchers based on surveys and other traffic network investigations
(Ettema et al. 2004).
In this previous research, three categories of variables are of importance in deter-
mining how people design their trips: properties of the traveler; properties of the trip;
and properties of the vehicle (Ettema et al. 2004; Eluru et al. 2012). Among these
categories, the properties of the trip and of the vehicle is applicable to the descrip-
tion of a travel route, mainly including the objective of the trip, financial cost, time
of departure, length and number of transfers and intervals, total time spent in mak-
ing the trip, etc. In this study, we formulate properties to describe the travel routes
accordingly, and weight different routes by these properties.
25 A Comprehensive Regional Accessibility Model … 465
1.3 Network Centrality
Network centrality is applied to synthesize city-pair accessibility into node acces-
sibility. Historically, this method applied graph theory to evaluate the importance
of vertices in a network (Sabidussi 1966), and was applied to identify influential
persons within a social network (Freeman 1979; Borgatti 2005;Katz1953), super-
spreaders of diseases and key nodes in communications networks (Mackenzie 1966)
etc. Network centrality tools including Ucinet, Gephi and Pajek, were developed
for such analysis and its visualization, thereby significantly lowering the barriers to
network analysis and centrality evaluation.
Various indicators were developed to describe the importance of vertices from
different aspects. Degree centrality is defined as the number of links connected to a
node; closeness centrality is the average length of the shortest distance between the
given node and all other nodes; harmonic centrality reverses the sum and recipro-
cal operations in the definition of closeness centrality (Marchiori and Latora 2000);
betweenness centrality quantifies the number of times a node acts as a bridge along
the shortest path between two other nodes; and eigenvector centrality assigns rel-
ative scores to all nodes in the network based on the concept that connections to
high-scoring nodes contribute more to the score of the node in question than equal
connections to low-scoring nodes.
As a typical example of network, transportation systems and its nodes can most
certainly be described by such indicators. However, lack of applicability does exist:
in the original model the distance between two nodes are defined as the minimum
number of nodes on the path connecting them (corresponding to number of transfers),
regardless of the different cost to travel between neighboring nodes (corresponding
to the distance of travelling). Modifying network centrality indicators by introducing
geographical distance or time cost can make it more applicable for describing regional
transportation networks. In this particular study, harmonic centrality will be modified
and applied to adapt to the evaluation of city accessibility.
1.4 Online Travel Route Information
Online route data ensures the sufficiency and stability of the data source in this study.
Two kinds of internet platforms are now providing profoundly better information on
transportation infrastructure than was previously possible, namely: ticket booking
platforms, covering trains, flights and buses; and digital map and navigation services,
provided by, for example, Google, Baidu, and AMAP (Auto Navi Map).
Internet ticket booking platforms provide relatively consistent, detailed and gen-
uine information on public transport services. In China, for example, 12306.cn serves
as an official platform for booking train tickets, while each airline has its own booking
website; internet travel agencies such as Ctrip and Qu’nar provide comprehensive
466 Y. F a n e t a l .
travelling solutions involving air, train and inter-city buses. All these sources cover
real-time information including travel routes, time-of-stops, ticket price, etc.
Digital map and navigation services provide real-time road trip route recommen-
dations based on vast digital repositories of information, including road networks,
historic and predicted traffic congestion and traffic management etc. gained from its
users, providing much better accuracy than simple calculations based on road net-
works. Moreover, in some regions (including most large cities in China), in recent
years, certain specific road vehicles including taxis, long-distance buses and heavy
trucks have been tracked using compulsory GPS devices. Ferries, flights and cargo
ships around the world are now also constantly tracked, and their position can be
acquired from a network platform.
Most of this type of data can be bought from the provider or acquired through
batch processing using the API (Application Programming Interface) provided by the
map platform or the ticket booking platform on a regular basis, ensuring a stable data
source. In this study, we used train time-tables from the website location 12306.cn,
flight schedules from ctrip.com and road trip recommendations through AMAP API
to acquire data for potential routes.
1.5 This Study
Generally, this study aims to colligate different criteria of traveler’s accessibility in
an improved harmonic centrality model by using actual routes-of-travel acquired
from internet data sources. The basic starting point of the study is that the properties
of a travel route describes how costly and how complicated the trip is, and compre-
hensively how preferable this route is; the summary of the values of all travel routes
between a given city pair, on the other hand, describes the abundancy of traveling
opportunities. A weighted sum of all travel routes between two given cities can thus
take all three basic criteria of regional accessibility into account in describing how
convenient it is to travel between them.
To achieve this end, the model consists of three steps. Firstly, it picks out all feasi-
ble travel routes from one city to another; secondly, a weighting mechanism synthe-
sizes all travel routes between a given city-pair and returns the city-pair accessibility;
and thirdly, for any given city, all related city-pair accessibilities are synthesized by
an improved harmonic centrality model that describes a city’s accessibility perfor-
mance. Travel routes are considered as a combination of train trips, flight trips and
road trips that connect two places, including transfer trips that involve airports or rail
stations in the same city.
This chapter explains the model in four sections. The next section describes the
three-step structure of the model, and then it explains the algorithms and methods used
in the model. The third section tests the model with actual data and verifies the results
with traffic flow data and economic data. The final section discusses the advantages
and shortcomings of the model and potential improvements and applications.
25 A Comprehensive Regional Accessibility Model … 467
2 Model Structure and Methods
2.1 Model Structure
The model consists of three layers: (1) travel route extraction; (2) calculation of
city-pair accessibility; and (3) calculation of node accessibility. The overall model
framework is demonstrated in Fig. 1.
(1) In travel route extraction, blank data tables are prepared for every directed city
pair, and the information of all feasible routes between the two cities are found
and listed as data fields in the table. The more routes there are from one city
to another, and the more convenient and low-cost these routes are, the more
accessible the city-pair is. But then the issue arises, what routes are feasible and
what routes are not?
When there are non-stop trains, flights and highways connecting two cities, a traveler
might choose any of the three—flights are faster, trains are relatively cheaper and
arguably more comfortable, whilst road trips are flexible. For a road trip traveler,
a travel route would be either the shortest, the cheapest, or the fastest route, given
that he/she has no predetermined pass points; for a train or air traveler, he/she might
choose any of the non-stop train or flights by personal preferences, as most non-stop
flights or trains are similar in length and cost. Thus, if you can take a train or flight
directly from city M to city N, that then makes this choice a feasible route; if you
can drive there, then that being the lowest cost route also makes it a feasible route.
As for transfer routes, investigative studies have found that travelers prefer routes
with two or more transfers much less than routes with one or no transfers; hence for
Fig. 1 Research framework
468 Y. F a n e t a l .
city pairs that can be reached with one or no transfers, routes that require multiple
transfers are ignored; and for all routes using the same transport mode and connecting
the same city pair, then those costing more than twice that of the cheapest route and
those that take twice as long as the fastest route are ignored. Same-city, trans-station
transfers are considered to be connected with in-city road trips. Details are given in
Sect. 2.2.
(2) In calculating city-pair accessibility, the travel route tables for a city pair were
synthesized by giving every route a weight that values how much the route
contributes to the accessibility of the city pair, and the city pair accessibility is
then defined as the summed weight.
First, we view any given city as a destination city. By collecting all the routes from all
other cities to this city, we can see the distribution of total travel time, total monetary
cost, time of departure, time of arrival and number of transfers. The weighting share
between these factors are determined by an entropy-based model that quantifies the
dispersity of each factor. The preferability of each property of a travel route is decided
by a utility function, and the total contribution of this route is the weighted sum of
the preferability of each property.
Then we move on from destination cities to city pairs. The previous steps have
listed all routes to the corresponding city pairs and have calculated all their weights;
now we simply add up the preferability of all routes serving the same city-pair and
consider it to represent the accessibility of this city pair. Details are given in Sect. 2.3.
(3) In the calculation of node accessibility, a modified harmonic centrality model
accumulates all the city-pair accessibility that is linked to a city. The result is
stored as the node accessibility of this city. Details are given in Sect. 2.4.
2.2 Travel Route Extraction
Travel route extraction determines non-stop travel routes, filters feasible transfer
routes, and generates property fields for all routes.
(1) Preparation
For N cities, create N×(N−1)tables with flexible length to store the index of travel
routes toward each city. Create another N groups of matrix tables to store the cost to
travel between different train stations and airports within the same city (Fig. 2).
(2) Find non-stop train routes
A python program traverses every row in the reorganized train table (Fig. 3). In a
single row for train T, any station pair (M →N) satisfying that M lies to the left
of N means that T is a feasible non-stop train route from M to N. Both stations are
searched in the station-city table to see whether they serve the same city. If not, find
the corresponding city pair by searching the two-level indices, and record total travel
time, time of departure, time of arrival and ticket prices into the travel route table.
25 A Comprehensive Regional Accessibility Model … 469
Fig. 2 Preparing uniform tables to store inter-city and in-city routes information
Fig. 3 Reorganization of train tables
(3) Finding non-stop flight routes
Another python program traverses the flight calendar table. For every row in the
table, the destination field and the origin field are extracted, and the corresponding
city pair is determined the same way as in finding train routes. Write the total travel
time, time of departure, and time of arrival into the travel route table. The ticket
price of flights is highly unsteady and requires long-term monitoring, hence are not
considered in this paper and the distance of the trip is recorded instead.
(4) Finding road trip routes
A JavaScript program is fed with string matrixes containing the default POI of all
prefectures, train stations and airports as its parameters. It then acquires the distance
of travel, time of travel and total cost of recommended road travel routes through
AMAP API, and stores them in route tables.
470 Y. F a n e t a l .
Fig. 4 Demonstration of a complicated transfer route
(5) Finding transferring routes
Transferring routes that do not involve trans-station transfers (in other words, the
second part of the trip starts at the station where the first ends) are simply identified
by traversing train and flight tables for all possible one-transfer routes. The optimal
route between a given city-pair is updated simultaneously, so that unfeasible routes
can be efficiently ignored.
As for transfers that involve travelling between stations, the route information
between the corresponding stations stored in in-city route tables is extracted and
added to cost estimation and time tags on the first part of the trips. Possibilities of a
feasible trans-station transfer are then searched on this basis (Fig. 4).
2.3 City-Pair Accessibility
City-pair accessibility calculation determines the weight share of different route
properties for each destination city, calculate how costly and complicated every
route is, and evaluate the overall accessibility of every city pair.
(1) Preparation
Create route information tables for each possible city pair. Properties of different
kinds of routes are categorized into financial cost C f, travel time Ctand number of
transfers Ctr.
For any city N, find all the routes that ends in this city. Record the following infor-
mation in relation to these routes: minimum financial cost C fmin(→N)in Renminbi
yuan; minimum travel time Ctmin(→N)in minutes; and the minimum number of
transfers Ctrmin(→N). Then normalize the properties of all routes to this city by
(Fig. 5)
25 A Comprehensive Regional Accessibility Model … 471
Fig. 5 Organize routes leading to a given destination
Cnorm =C
Cmin(N),C∈Cf,Ct,Ctr(1)
(2) Weighting route properties
Use an entropy-based approach (Ma et al. 1999) to allocate weights to each property.
For each route from M to N Routei(M→N), the contribution of property j of the
route to the uncertainty among all mroutes towards city N is
Pij =xij/
m
i=1
xij (2)
The total contribution of all mroutes to property j is
Ej=−k
m
i=1
Pij ln Pij (3)
where k =1/ln m, and the weight of property j is
Wj=1−Ej/
3
j=1
1−Ej(4)
W1,W
2and W3is the weight of financial cost, travel time and number of transfers
respectively.
(3) Calculating city-pair accessibility
The cost of each route is represented as
CRoutei(M→N)=W1CRoutei
fnorm +W2CRoutei
tnorm +W3CRoutei
trnorm (5)
472 Y. F a n e t a l .
Then the accessibility value of city pair (M→N)is calculated by accumulating
all feasible routes:
Accessibility(M→N)=
m
i=1
1
CRoutei(M→N)
(6)
And all city-pair accessibility values are stored in a matrix table for each possible
city pair.
2.4 Node Accessibility
The method of node accessibility calculation is generated from harmonic central-
ity in network analysis. As a derivative of closeness centrality, harmonic centrality
characterizes the importance of vertices by the sum of reversed costs rather than the
reverse of summed costs, as in
H(x)=
y=x
1
d(y,x)(7)
Similarly, we define the node accessibility of a city as
Accessibility(M)=
N
i=1
Accessibility(M→N)(8)
which is essentially the sum of reversed cost given the definition of
Accessibility(M→N)in Eq. (6).
3 Empirical Research
3.1 Study Area and Spatial Unit
3.1.1 The Study Area
Mainland China is our study area, where there is a centralized train ticket booking sys-
tem and flight administration institute that ensures complete national data coverage.
Up until September 2018, the area consisted of 334 prefecture-level administrative
areas, 2851 county-level administrative areas, and 39,888 town-level administrative
areas, hereinafter referred to as prefectures,counties and towns. Prefectures generally
consist of a core city and several counties.
25 A Comprehensive Regional Accessibility Model … 473
Fig. 6 The study area
Up until September 2017, there were 2878 passenger train stations and stop points
in the study area. Most prefecture cores own one train station, some own several;
other stations serve counties and towns on connecting lines and branch lines. 219
airports serve prefectures that are important big cities, popular tourist destinations,
or in isolated mountainous areas (Fig. 6).
3.1.2 Spatial Units
Previous studies have emphasized the importance of choosing spatial units under
two criteria: (1) all spatial units should be basically comparable to each other in
terms of size and scale; (2) statistical areas are preferred so that further analysis and
comparison is feasible. Considering that statistical reports are issued by prefecture
governments and most prefectures are similar in land area and total population, we
choose prefectures as the basic spatial units. This implies that: (1) train and flight
transfers between different stations in the same prefecture are considered; (2) the
default Point of Interest (POI) of a prefecture, mostly its government office, is chosen
as the origin or the destination for a travel route.
474 Y. F a n e t a l .
3.2 Data Collection
3.2.1 Travel Route Data
Three kinds of transportation data were collected for this study: train schedule, flight
calendars and road trip route recommendations. Ferries are hardly used for travelling
between cities in China (except for particular cases like Dalian and Yantai), hence
they were not considered (Fig. 7).
The train schedule data is a .txt table extracted from 12306.cn on September 30,
2017 by udparty.com, a spatial data provider. It contains 60,567 data rows, each row
representing one stop of a scheduled train. The table contains the following property
fields: train number; station ID; station coordinates; time of arrival; and time of
departure.
The flight calendar data is a .csv table acquired from internet booking platforms
during the week of October 9 to October 15, 2017, also by udparty.com. It contains
53,423 rows, with each row representing one flight. The table contains the following
fields: flight ID; time of departure; time of arrival; origin airport; origin airport
coordinates; destination airports; destination coordinates; and the week.
The road trip route recommendation data is a .txt table acquired from AMAP
Location Based Services (lbs.amap.com) by our research team. AMAP exploits
crowd-sourced information and GPS records of its users to calculate toll fees and
estimate real-time travel time, and make route recommendations accordingly. This
information can also be acquired through its official API. By uploading an origin’s
POI/coordinates, the destination’s POI/coordinates and a route preference (minimum
time, minimum distance, or minimum toll fees), the AMAP API returns a detailed
route including every intermediate point and the estimated time cost for every part
of the trip.
3.2.2 Spatial Reference Data
A polygon shapefile showing the administrative boundaries of more than 300
prefecture-level cities in CGS_WGS_1984 geographic coordination system was used
as the spatial reference data, covering 33 provinces in China but missing the admin-
istrative boundary information in Taiwan Province. Considering that neither the train
system nor highway network in Taiwan Province is connected to other parts of China,
such an omission hardly influences the outcome of the model.
3.2.3 Validation Data
Two datasets were utilized to validate the model: Regional traffic flow (annual total),
to characterize inter-city connections and thus validate the results of each city-pair
accessibility calculation; and Gross Domestic Product (GDP), to characterize the
25 A Comprehensive Regional Accessibility Model … 475
Fig. 7 Left: train table. Central: Flight calendar. Right: AMAP route planning
476 Y. F a n e t a l .
state of economic development of a city and thus validate the results of each node
accessibility calculation.
The traffic flow data is based on the record of positioning requests to Tencent,
the major mobile terminal SNS (Social Networking Services) provider in China. We
bought the aggregated traffic flows to and from Wuhan for train and air services in
2018 from wayhe.com which preprocessed the original data, involving more than
120 prefecture-level cities.
The GDP data of 322 prefecture-level cities were acquired through the statistical
yearbooks published on the official website of the Bureau of Statistics of each city
government and were processed into .xls format.
3.3 Results of City-Pair Accessibility Calculations
Wuhan was chosen as an example to demonstrate the result of city-pair accessi-
bility calculation, and the results are shown in Fig. 8. Four kinds of cities demon-
strate high accessibility with Wuhan: cities within and around Hubei Province,
which have convenient highway access and frequent inter-city trains from and to
Wuhan; cities on major highspeed railways connected to or close to Wuhan, including
those on Beijing-Guangzhou Railway, Shanghai-Chengdu Railway, Wuhan-Fuzhou
Fig. 8 City-pair accessibility values for Wuhan
25 A Comprehensive Regional Accessibility Model … 477
Tabl e 1 Performance of
comparable city-pair
accessibility indicators
Model R2p-value Variance
This approach 0.45 0.05 6.2×106
Minimal cost 0.36 0.02 7.8×106
Railway, Shanghai-Kunming Railway and Xi’an-Zhengzhou Railway; major
national or regional centers that have frequent airlines to Wuhan, including Qingdao,
Urumqi, Harbin, Nanning, Kunming and Chengdu; and favorite tourist destinations
like Lijiang and Sanya.
These are also cities that are closely connected to Wuhan in terms of regional
traffic flow. As is shown in Table 1, the city-pair accessibility acquired through this
approach is significantly more strongly correlated with annual traffic flow between
Wuhan and other cities than simply using the reverse of minimal time cost.
3.4 Results of Node Accessibility Calculations
The comprehensive accessibility values of 318 prefecture-level cities and 4
provincial-level cities are calculated and linearly normalized to (0,1], together with
three other measures of regional accessibility that are commonly used: accumulated
minimal cost, harmonic centrality and closeness centrality. To compare the effec-
tiveness of each method, linear regression is applied to these indicators and the total
GDP of each city in 2017. The results are shown in Figs. 9,10 and Table 2.
Obviously, the accessibility value calculated by a route-based approach demon-
strates better coherence with actual economic size than the others. We can also see
that the distribution of route-based accessibility fits quite well with Zipf’s Law, as
the number of cities and accessibility value generally follows power law distribu-
tion, which we would expect for the size distribution of cities in a region (Alperovich
1984; Gabaix 1999). Among the other approaches, a simple minimal cost that does
not consider transfer routes has the poorest performance, smoothing much of the gap
between different tiers of cities as neither the number of shifts nor the convenience of
transfer hubs is reflected; closeness centrality performs better, as such network-based
methods takes transfers into account; and harmonic centrality, which the node acces-
sibility calculation in our approach is based on, demonstrates the closest performance
to our approach.
Still, there are many data points with a huge residual value. To look into these
abnormities, we put the ratio of the predicted GDP by the regression value in our
model and the actual GDP on to the map in Fig. 11. We can see several regions of
interest, where the estimated GDP is significantly higher or lower than the actual
value. Among them, Region A and Region C are major railway and highway corri-
dors that extend into the relatively remote north-western and north-eastern regions.
The cities in this corridor enjoy easy land access to the more developed regions
of Central and Eastern China but are in less active regions in general. Region B is
478 Y. F a n e t a l .
Fig. 9 Node accessibility acquired by this approach
Tabl e 2 Performance of
comparable accessibility
models
Model R2p-value Variance
This approach 0.70 0.05 3.3×104
Minimal cost 0.42 0.05 6.5×104
Harmonic
centrality
0.62 0.05 4.2×104
Closeness
centrality
0.50 0.05 5.6×104
Shanxi Province, which depended heavily on coal mining industries in earlier years
and is experiencing an economic stagnation since China began promoting an eco-
friendlier energy composition (Zhao and Zhao 2011). Regions D and H are tourist
destinations that enjoys high levels of flight accessibility due to airlines serving the
tourist market, but do not have a high overall GDP. Regions F and G are cities near
major urban agglomerations but are not connected to trunk routes in transportation
networks, resulting in low accessibility. Region E is a mountainous area with a high-
speed railway connection but is generally sparsely populated. In general, specialized
industrial structure, overly developed infrastructure in under-developed regions and
the spillover of larger comparatively well-developed urban agglomerations are the
main causes of such inconsistencies. Details are given in Table 3.
25 A Comprehensive Regional Accessibility Model … 479
Fig. 10 The performance of this approach and other accessibility indicators
480 Y. F a n e t a l .
Fig. 11 Estimated GDP/actual GDP
Tabl e 3 Regions
Region Regression
result
Name Details
AOverestimated Lanzhou-Urumqi Corridor Railway and highway trunk
route in less developed
region
BOverestimated Shanxi Province Economic stagnation as
major industries decline
COverestimated Qinhuangdao-Harbin
Corridor
Railway and highway
backbone in less developed
region
DOverestimated Lijiang Well-connected airlines for
tourist attraction
EOverestimated Southern mountains Endeavored high-speed
railway in mountainous areas
FUnderestimated South Jiangsu Benefits from major urban
agglomeration
GUnderestimated West Guangdong Benefits from major urban
agglomeration
HOverestimated Hainan Province Well-connected airlines for
tourist attraction
25 A Comprehensive Regional Accessibility Model … 481
4 Conclusion
As regional and urban planners start to access and have the opportunity to analyze
massive volumes of reliable and accurate data, it becomes imperative for us to acquire
the ability not only to see the whole picture but also to investigate, describe and
utilize details. Those interested in regional transportation have progressed macro level
analysis such as measuring purely geographical distances and analyzing network
structures, to our work where we have progressed to micro level analysis where we
determine the actual trip volumes and characteristics on specific routes. This work
has achieved success in this regard, but there is still a great deal more future work
to be done. For example, numerous empirically based topics can be investigated by
utilizing detailed transportation data from different perspectives; the objective of the
research can be expanded with increased access to more comprehensive data; the
model itself can be improved; and more reference data can be introduced.
Most importantly, additional pertinent subjects can and should be discussed under
this model framework given the depth of information on travel routes that we can now
obtain. For instance, long range trains and buses tend to stop at more preferable times
in important cities and major stations, resulting in untimely stops at intermediate
stations. This can influence the degree of convenience conceived by actual passen-
gers, whilst different passengers—commuters, tourists, business traveler, migrants,
etc.—have different preferences for timings. For air travelers, as flights are basically
arranged on a weekly cycle, they can be sensitive to the balance between weekend
and weekday flights. With these aspects looked into and actual traffic flows taken into
account, very specific topics such as optimizing traffic flows to facilitate access to a
tourist attraction or adjusting the schedule for inter-city commuters can be studied.
It should be noted that only passenger transportation was considered in this study,
with freight transportation traffic flows omitted. Both industry and daily life relies
heavily on freight transportation (with online shopping increasingly taking a signif-
icant share of retailing activity), however it needs to be discussed separately from
passenger transportation as it differs quite markedly in terms of infrastructure, vehi-
cles, networks, logistics and operation. However, acquiring data may be problematic:
records of freight routes (as mentioned in the Introduction section) are mostly kept
by government agencies or commercial companies as confidential, as these data are
potentially commercially valuable. Improved and increased efforts may be required
for obtaining adequate, quality data collection in this regard.
Finally, in our future work we intend to take into account the differences in scale
and significance amongst different cities. The increased transport connections to
more populated and active regions will provide greater opportunities in the future.
This is clearly reflected in the result of the node accessibility calculations; introducing
a weighting mechanism either using reference data or a recursive approach such as
Eigenvector centrality (Bonacich and Lloyd 2001) will help improve the model’s
performance.
482 Y. F a n e t a l .
References
Alperovich G (1984) The size distribution of cities: on the empirical validity of the rank-size rule.
J Urban Econ 16(2):0–239
Bonacich P, Lloyd P (2001) Eigenvector-like measures of centrality for asymmetric relations. Soc
Netw 23(3):191–201
Borgatti SP (2005) Centrality and network flow. Soc Netw 27(1):55–71
Eluru N, Chakour V, El-Geneidy AM (2012) Travel mode choice and transit route choice behav-
ior in Montreal: insights from McGill University members commute patterns. Publ Transp
4(2):129–149
Ettema D, Ashiru O, Polak JW (2004) Modelling timing and duration of activities and trips in
response to pricing policies. Percept Mot Skills 96(1):1013–1014
Freeman LC (1979) Centrality in social networks’ conceptual clarification. Soc Netw 1(3):215–239
Gabaix X (1999) Zipf’s law and the growth of cities. Am Econ Rev 89(2):129–132
Jin F, Wang C, Li X (2010) China’s regional transport dominance: density, proximity, and accessi-
bility. J Geogr Sci. https://link.springer.com/article/10.1007/s11442-010-0295-6
Katz L (1953) A new status index derived from sociometric analysis. Psychometrika 18(1):39–43
Mackenzie KD (1966) Structural centrality in communications networks. Psychometrika
31(1):17–25
Ma J, Fan ZP, Huang LH (1999) A subjective and objective integrated approach to determine
attribute weights. Eur J Oper Res 112(2):397–404
Marchiori M, Latora V (2000) Harmony in the small-world. Phys A 285(3):539–546
Rich DC (1978) Population potential, potential transportation cost and industrial location. Area
10(3):222–226
Sabidussi G (1966) The centrality index of a graph. Psychometrika 31(4):581–603
Smith KV, Phillips CT, Lewis RJ (1970) Network evaluation of complex transportation systems.
Transp Res 6(1):103–111
Straatemeier T (2008) How to plan for regional accessibility? Transp Policy 15(2):0–137
Zhang L, Lu Y (2007) Regional accessibility of land traffic network in the Yangtze river delta. J
Geog Sci 17(3):351–364
Zhao K, Zhao H (2011) Integration of coal min resources and its influence on the economic devel-
opment of Shanxi Province. Coal Econ Res 11:17–22
Chapter 26
Taxi Behavior Simulation
and Improvement with Agent-Based
Modeling
Saurav Ranjit, Apichon Witayangkurn, Masahiko Nagai
and Ryosuke Shibasaki
Abstract Taxi services, despite deemed as a convenient form of commuting, are
challenged by many issues. The issues can be categorized based on the perspectives
of drivers and passengers. Regarding the issue from the driver’s perspective, taxi
drivers are working longer hours. However, the revenue generated does not justify
their increased working hours, (i.e. working longer hours with less revenue), which
further implies that drivers are not getting enough passengers. By contrast, from the
perspective of passengers, the prime issue with taxi services is that passengers are
rejected or denied service. In this research, we aim to establish a taxi behavior simu-
lation model for an existing conventional taxi operation and introduce optimization
for this type of taxi operation. The evaluation between the two models shows, that
by introducing optimization to the usual taxi behavior such as in providing greater
flexibility in selecting a passenger, an improved service can be achieved for both taxi
drivers and passengers.
Keywords Agent-based simulation and modeling ·Taxi behavior modeling ·
Optimization ·Taxi service level improvement
S. Ranjit (B)·A. Witayangkurn ·R. Shibasaki
Center for Spatial Information Science, The University of Tokyo, 5-1-5,
Kashiwanoha, Kashiwa-shi, Chiba 277-8568, Japan
e-mail: ranjits@iis.u-tokyo.ac.jp
A. Witayangkurn
e-mail: apichon@iis.u-tokyo.ac.jp
R. Shibasaki
e-mail: shiba@csis.u-tokyo.ac.jp
M. Nagai
Graduate School of Sciences and Technology for Innovation, Yamaguchi University,
2-16-1 Tokiwadai, Ube, Yamaguchi 755-8611, Japan
e-mail: nagaim@yamaguchi-u.ac.jp
© Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3- 030-19424- 6_26
483
484 S. Ranjit et al.
1 Introduction
With the growing advancement in the field of Global Positioning System (GPS)
technology, the utilization of GPS in the field of science has increased significantly
in the past recent years. One of the prominent fields which have benefited from GPS
technology is spatial information science. Spatial information is data that provides
information concerning geographic location and its corresponding features. The field
of transportation is one of the sectors that has benefitted most with GPS embedded
technology and its utilization of spatial data. Whether it is applied in the navigation
of, or the tracking of a vehicle within a road network, GPS technology has been
forefront of providing real-time spatial information which further helps in both real-
time and strategic decision making. However, such technology is not just limited
to navigation or tracking. In recent years, many large cities such as New-York and
Beijing have started embedding GPS device in vehicles, such as taxis to collect traffic
information (Yuan et al. 2013). Such vehicles are primarily known as a floating car
or a probe vehicle.
Taxi services are ubiquitous all over the world and are generally deemed to be
fast and convenient forms of commuting, particularly in big cities. Bangkok, the
capital of Thailand, is no exception with more than approximately 100,000 taxis
in operation daily. Despite Bangkok having a large fleet of taxis, there still exists
problematical service delivery issues with taxi services. Peungnumsai et al. (2017)
identified issues relating to taxi services through a detailed survey of taxi drivers in the
city of Bangkok. The issues can be categorized based on the particular perspectives
of both drivers and passengers. The key issue from the drivers’ perspective, is that
they are working longer hours. However, the income or revenue generated does not
justify their longer working hours, with falling rates of income per hours worked,
which suggests that the drivers are not serving sufficient volumes of taxi passengers.
By contrast, passengers’ perspective, the key issue with taxi services is that
passengers are rejected or denied access to taxi services that they are entitled to
access (Zhang and Wang 2016). The previous study on Bangkok’s taxi services by
(Peungnumsai et al. 2018) suggests that there is a lack of authentic data available to
determine the number of passengers rejected by taxi drivers. However, there exist
many complaints regarding rejection of passenger attempts to access services offered
by taxi drivers, which is evident on social media, forums and news channels. It also
suggests that most of the rejections occurred due to mismatch of passenger desti-
nation choice as well as rejecting the requested destination where there was a risk
that the return trip would not be able to secure a passenger fare. For both drivers and
passengers, passenger ‘findability’ seems to be the root cause of service dissatisfac-
tion. One of the ways to minimize the issue of difficulty with passenger findability
is to optimize the operation of the taxi service. This could be resolved through the
application of a micro-level simulation of taxi behavior, which is based on utilizing
quantitative data evidence, detailing existing taxi operations, so that an optimization
of taxi operations can be determined. However, there is a lack of a properly estab-
lished model based on quantitative data that would provide an understanding of the
26 Taxi Behavior Simulation and Improvement … 485
Fig. 1 The need for taxi behavior simulation based on quantitative data analysis. Image Source
Thairath (2017)
existing and usual taxi operations. When dealing with ways to improve the service
level of taxi operations, a challenge may occur without a proper taxi behavior model,
as a micro-level simulation with quantitative data evidence plays an important role
in understanding the behavior of the actual service. Figure 1shows a graphical illus-
tration of the existing issue as it applies to current taxi operations. The two primary
goals of a taxi business are to provide good taxi services to passengers and maintain
commercially viable business operations. However, the evidence from the data is that
there are issues related to both providing a good standard of service to passengers as
well as ensuring profitability of operations. In addition, when passengers are rejected
by a taxi service, it results in high dissatisfaction levels with taxi services.
In this research, Agent-based simulation and modeling were utilized to model
taxi behavior in Bangkok and surrounding provinces. Agent-based modeling, which
works on state-rule-input architecture (Torrens 2010), where each taxi behaves as an
agent, interacts with the environment to capture dynamic behavior through recon-
structing complex patterns as defined through behavior rules (Baster et al. 2013).
Agents, as described by spatial and temporal parameters, interact with an environ-
ment which in turn provides a set of behavior rules that directs the outcome or a
goal of the simulation. The aim of the agent-based simulation model is to establish
a taxi behavior simulation model for the existing usual taxi operation and introduce
optimization of the usual taxi operation, to provide improved taxi operation.
The optimized model will have the ability for the driver to choose a passenger
depending upon the passenger’s origin and destination as well as available demand in
the region. The optimization model is used to provide a recommendation to the taxi
drivers to determine which passenger would be a better choice, via a mobile phone
application. The hypothesis behind the model is that when the driver can choose the
passenger, then the level of passenger rejections would be drastically minimized and
486 S. Ranjit et al.
in choosing a passenger, a driver would be provided with a degree of freedom in
their choice of taxi service tasks, thereby potentially improving their income earning
potential.
A past study has suggested that taxis tend to work in various service zones.
Peungnumsai et al. (2018) noted that there were three established service zones
where taxi tend to work, based on taxi interconnectivity within the city’s geograph-
ical locations. Hence, by providing the driver with the ability to choose their pas-
senger, drivers could easily provide a taxi trip to a passenger within their localized
service zones thereby improving service efficiency. The proposed optimization tech-
nique, therefore, focuses on allowing drivers to decide which passenger they should
provide a service to. Hence, this research has the objective of improving taxi opera-
tions through quantitative data analysis obtained from GPS sourced probe data from
Bangkok’s taxi fleet.
2 Literature Review
A wide range of study has been made possible in the field of spatial information
science with the constant development and advancement for collecting trajectory
data in space and time (Sadahiro et al. 2013). Technologies such as loop detectors
are available for retrieval of various traffic data but are limited to specific sections
of the road and are not readily available throughout the regions. However, the use
of a probe car utilizes the running vehicle to gather various traffic information and
has been seen as a primary component of Intelligent Transportation System (ITS)
technology for vehicle behavior modeling (Bischoff et al. 2015; Cheng and Nguyen
2011; Miwa et al. 2004).
Taxis with GPS sensors, in cities like New York and Beijing, collect spatial and
temporal data to be processed for extracting traffic information (Yuan et al. 2013).
Mobility intelligence from taxis is now considered as an essential factor that pro-
vides assistance in maximizing profit and reliability for every possible trip scenario
amongst taxi drivers (Moreira-Matias et al. 2013). As such, micro-level simulation
models are needed to understand the stochastic dynamics of taxi behavior. Informa-
tion from a simulation can be further analyzed for optimization by adjusting param-
eters like demand and supply as well as the taxi dispatching process (Maciejewski
et al. 2016).
In recent years, for many areas of application, such as flow evacuation, traffic, and
customer flow management (Bonabeau 2002), Agent-based modeling and simulation
(ABMS) are now being implemented. Agent-based modeling and simulation describe
the dynamic action of an entity that is, agents governed by behavior rule and properties
(Abar et al. 2017; Cheng and Nguyen 2011; Grau and Romeu 2015), to emulate
natural behaviors.
26 Taxi Behavior Simulation and Improvement … 487
3 Taxi Behavior Modeling State
Taxi behavior modeling consists of multiple states. A simple state diagram for taxi
behavior is presented in Fig. 2. When the taxi collects a passenger, the taxi state could
be described as the ‘Taxi Pick Up State’. When the taxi has a passenger, then the taxi
must go to the passenger destination place to drop off the passenger. The ‘Taxi Drop
Off State’ is defined as the state when the taxi has dropped off their passenger. Next,
the ‘Free Movement State’ defines when taxis are free to move in various directions
to search for passengers after the ‘Taxi Drop Off State’. In the Free Movement State,
a taxi could move to 9 cardinal directions including staying at the same location.
During this state, taxis are searching for passengers. As the taxi finally picks up their
passenger, then the taxi state is changed back to the ‘Taxi Pick Up’ state. The state
cycle continues until the taxi operation stops.
It is inferred that taxi behavior is a dynamic discrete time-dependent event within
the spatial and temporal domain, with customer pick-up, customer drop-off, cruising
and parking described in the ‘taxi state’ diagram. Simulation can help understand
such behavior which simplifies the real-world system for application in mathemat-
ical models. Agent-based simulation and modeling help capture dynamic behavior
through reconstruction of intricate patterns with a set of behavior rules. In this regard,
in agent-based modeling, rather than modeling, the system is based on a single equa-
tion, with the complex individual agent behaving more naturally with a collection
of autonomous taxi agents with rules governing them (Bonabeau 2002). Such a
model can highlight the effect of a change in taxi services and its impact on driver
income profitability through optimization which is introduced to the existing usual
taxi operation simulation model. Understanding the causality, such as the impact on
the behavior of the taxi service where there is a variation of a number of agents in the
region of high or low demand areas, helps manage taxi fleets better with regard to
the operation cost with direct impact on the income generated. The result is that the
utilization of spatial and temporal information from the probe vehicle can be an asset
for governing different aspects of urban management as such vehicles are opera-
tional throughout many cities. A taxi behavior simulation model implicitly describes
Fig. 2 Taxi behavior modeling state
488 S. Ranjit et al.
the behaviors of taxi operation at the city level where there is an existing usual taxi
operation. The improved taxi operation is achieved by optimization introduced to the
existing usual taxi operation.
The main research tasks are summarized as follows:
•Establishment of a taxi behavior simulation model that describes the existing usual
taxi operation, through quantitative data analysis obtained from taxi GPS probe
data.
•Optimize the taxi behavior model for existing usual taxi operations, to improve
incomes of drivers and the service levels for passengers.
4 Data Preprocessing
4.1 Probe Data
Vehicle probe data are data from vehicles collected through a communication network
(Nagashima et al. 2014) which are being widely used for various Intelligent Trans-
portation Systems (Liu et al. 2008). In this research, the vehicle data was obtained
from taxis that are operating in and around Bangkok, provided by Toyota Tsusho
Nexty Electronics (Thailand) Co., Ltd. in Bangkok, Thailand. Probe data from 10,000
taxis with a sampling time of 3 or 5 s, was collected between 1 June 2015 and 31 July
2015, for this analysis. The probe data collected belongs to the set of trajectories,
generated by taxis moving in geographical space, such that trajectory Ti={p1,p
2,
p3,…,p
j}, where pj=(xj,y
j,t
j), such that xj=longitude, yj=latitude, and tj
=timestamp. Data obtained from the GPS probe vehicle cannot always have guar-
anteed accuracy. The reason for this limitation could relate to the error in the GPS
measurement device itself or due to external environmental factors such as signal
obstruction from a building, multi-path error or even ionospheric error. An error, in
general, could be a systematic error or a random error (Zheng et al. 2014). Before
proceeding to any analysis of the probe data, the data had to be ‘cleaned’ of erroneous
and outlier data. Different map matching algorithm were compared together using
Open Street Map (OSM) as base road network. The probabilistic approach of a map
matching algorithm was chosen for its high accuracy in matching (Ranjit et al. 2017)
and for mapping GPS data to a road network along with removing outlier GPS data.
4.2 Road Network and Grid Network
As for the road network, the Open Street Map data of Thailand was taken and
topological error cleaning was subsequently performed (Ranjit et al. 2017). The
road network is represented by R such that R ={r1,r
2,r
3,r
4,…,r
n}, where r1,r
2,r
3,
r4,…,r
nis each road segment. A small grid size of 500 m ×500 m formed the basis
26 Taxi Behavior Simulation and Improvement … 489
of the grid network, which preserved the spatial characteristics of the grid (Nam et al.
2016) such that, the grid network was represented by G ={g1,g
2,g
3,g
4,…,g
m},
where g1,g
2,g
3,g
4,…,g
mdenotes each grid or cell.
Based on the probe GPS data with the road network and grid network, multiple
simulation variables were derived (Ranjit et al. 2018). These variables were later
used during the simulation process. The derived variables were as follows.
Stay Point Cluster: In order to conduct the taxi behavior simulation, the initial
location and starting time of taxi agents needs to be initially defined. The initial
location and starting time for each taxi agent was extracted based on the stay point
cluster and kernel density of the cluster timestamp within the given spatial location.
Stay point denotes the location where the vehicle (taxi) had stayed or stopped at
some location for a given interval of time, such as a parking place or a gas station,
or while looking for a passenger, as shown in Eq. 1. Each of the stay point cluster
locations extracted depicts the start location for each taxi during the simulation. As
for the stay point cluster, a grid-based DBSCAN algorithm (Ester et al. 1996;Gan
and Tao 2015; Wong and Huang 2016) provided a means to extract the cluster
Stay Point =(Panchor ,Psuccessor )<Dthreshold &
(Panchor,Psuccessor )>Tthreshold (1)
where, Panchor are the anchor points, Psuccessor are the successor points with Dthreshold
as threshold distance and Tthreshold as threshold time.
The Taxi Origin and Destination (OD): Origin and destination of the passenger
trips from the taxi probe data needs to be extracted and evaluated to model the
destination of the passenger during the simulation. Taxi origin and destination or
simply OD refers to the location where the taxi picked up and dropped off the
passenger or customer (Gonzales et al. 2014). Based on OD, transition probability
helps determine the passenger destination trip in the simulation such as given in
Eq. 2.
∀g∈Gt,P(gO→D)=TripO→D
TripO
(2)
where, PgO→Dis the origin-destination probability for all of the grid g that belongs
to G at time interval t, such that TripO→Dis the total number of passenger trips
between the origin grid O and the destination grid D, and TripOrepresents all the
passenger trips that originated at grid O at time interval t.
Taxi Demand: In the taxi modeling state of ‘free movement’ state, taxis are con-
tinually looking for passengers by moving to 9 cardinal directions including staying
at the same location. The demand probability of success for a given location and
time interval, for the usual taxi operation simulation model, defines whether the taxi
would get the passenger or not during passenger searching process. Equation 3shows
the demand probability of success.
490 S. Ranjit et al.
∀g∈Gt,P(dm)g=Og
Vg
(3)
where, P(dm)gis the probability of success for all of the grid g ∈G at time interval t,
such that Ogand Vgare the total number of demands generated and the total number
of recorded vacant taxis at grid g ∈G and time interval t, respectively.
Network Travel Time: Modeling of taxi behavior is subject to the movement of
a taxi from a passenger pick up location to a passenger drop off location as well as
including its free movement while searching for passengers. Both movements are
related to how a taxi travels between the locations along with considering travel time
for the road network. Equations 4and 5show the estimated average road network
segment speed and the average grid network speed used for estimating taxi travel
time during the simulation.
∀r∈Rt,¯sr=Sp∈r
Np∈r
(4)
∀g∈Gt,¯sg=Sp∈G
Np∈G
(5)
where, ¯srand ¯sgare the average speed on the road network segment r ∈R, and grid
network g ∈G, respectively at a time interval of t, such that Sp∈rand Sp∈gare
the sum of the speed of all the points p with Np∈rand Np∈gas the total number of
points that belongs to its respective network. In this regard, Eq. 6shows the road
network travel time during the simulation.
Tr:ˆp∈t=rdistance
¯sr
rdistance
¯sg
(6)
where, Tr:ˆp∈tis the road network travel time with rdistance as the road network segment
distance, such that for any point ˆp that appears on road network segment r ∈R and
grid network g ∈G at time interval t during simulation, it would require Tr:ˆp∈tunit
time to cross or complete the road network segment.
Free Movement Taxi: The taxi state diagram shows that when the taxi drops
the passenger to their destination, the taxi changes its state to the ‘free movement
state’, that is a taxi with no passenger. In this state, the taxi has a total of 9 pos-
sible cardinal directions to choose from to search for the passengers. The cardinal
directions are north (337.5°–22.5°), northeast (22.5°–67.5°), east (67.5°–112.5°),
southeast (112.5°–157.5°), south (157.5°–202.5°), southwest (202.5°–247.5°), west
(247.5°–292.5°), northwest (292.5°–337.5°) or it remains in the same place. Based
on this assumption, Eq. 7shows the directional probability of a vacant taxi searching
for passengers in the usual taxi operation simulation model.
∀g∈Gt,P(d)g=ng
Ng
(7)
26 Taxi Behavior Simulation and Improvement … 491
where, P(d)gis the direction probability for vacant taxi movement, moving to direc-
tion d, for all of the grid g ∈G at time interval t, such that ngand Ngare the number
of vacant taxi points moving to direction d, and the total number of vacant taxi points
in grid g ∈G, and the time interval of t, respectively.
5 Taxi Behavior Modeling (Usual Taxi Operation)
The conceptual design of Agent-Based Modeling (ABM), is based on State-Rule-
Input architecture (Torrens 2010), as Fig. 3shows. The conceptual design of agent-
based modeling consists of a State ‘S’, Rule ‘R’ and the Input ‘I’. The taxi spatial
and temporal variable is the Input (I) of agent-based modeling. The stay point cluster
extracted from the probe GPS data provides the agent spatial and temporal param-
eters. The agent interacts with an environment which provides the Rule (R) to the
agent. Multiple variables extracted from the probe GPS data including free movement
taxi, taxi trip origin-destination, taxi passenger demand, routing and interpolation for
the network travel time, defines the modeling environment.
The rule then guides the agent to reach the goal which is defined by agent State (S).
In general, two states define the agents. The first state is the taxi without a passenger
and the second state is the taxi with a passenger. Of course, there is the state when the
taxi is looking for a passenger, but it is also the state without a passenger. The goal of
an agent, which is the next state, depends upon the current state and the rule provided
by the environment. Both the input agent and environment are updated based on the
Fig. 3 Conceptual design of agent-based modeling
492 S. Ranjit et al.
goal reached. The updating of the environment provides the indirect interaction of
an agent amongst each other.
The proposed agent-based simulation and modeling recreate the real taxi behavior
that is, the existing usual taxi operation. The improvement of the taxi service for both
drivers, regarding monetary profit, as well as for passengers, regarding the service
level of the taxi, is then defined through optimizing parameters derived from the
usual taxi operation model.
6 Taxi Behavior Modeling (Optimized)
Optimization is the process that finds the best possible way to use the available
resources while not violating any given constraints (Lindfield and Penny 2017).
Optimization is vital to solving the various mathematical problems in many disci-
plines (Rothlauf 2011) that would help find solutions which are optimal with regard
to this goal. Hence, taxi operation optimization is also a key aspect of how taxi drivers
could improve or increase their income with improving passengers’ service level.
The usual taxi operation simulation model helps provide a better understanding of an
existing taxi operation. While the usual taxi operation simulation model shows the
existing state of the taxi operation behavior, the optimized model shows improved
taxi behavior. The optimization of the taxi operation could help reduce or minimize
the issues faced by both passengers as well as the driver. Hence, the advantage of the
optimization of the taxi operation is not just beneficial for the taxi driver but also for
the passenger as well.
Past literature and research have proposed a different method for the optimization
of taxi operations. In recent years, there is an increment in the utilization of the GPS
trace for predicting a mobility pattern (Castro et al. 2012) that included the mobility
of the taxi. Analysis of the spatial-temporal behavior of taxi services could help urban
planners and managers (Deng and Ji 2011) achieve better decision making processes
in the transportation management sector. A time location social model where the
three-dimensional properties of a city’s dynamics are considered to predict the dis-
tribution of passengers was proposed by Yang et al. (2016). The model recommended
top N locations to the driver based on historical traffic data where drivers would have
a high chance of finding passengers. A two-layer model as described in Tang et al.
(2016) provided a recommendation to taxi drivers searching for passengers. The first
layer of the model provided a recommendation to the taxi driver as to which zone to
go to for a passenger pick up. The model used was the DBSCAN algorithm to cluster
pick up and drop off records and described the attractiveness of the pick-up location.
The second layer of the model then provided a routing decision for the taxi driver.
As for the routing behavior, the model implemented a Path Size Logit (PSL) model
considering both travel time and distance. On the other hand, demand prediction for
the next time frame has been proposed using the deep-learning algorithm as quoted
in Yao et al. (2018).
26 Taxi Behavior Simulation and Improvement … 493
Different scenarios are available for the case of optimization of taxi operation
services. The first scenario is how the driver picks up the passenger. In this scenario,
the driver can choose which passenger to pick and which not to pick. The reason
behind having a driver choose the passenger is that the refusal rate would decrease
drastically. As mentioned earlier, one of the reasons for passenger rejection is due to
a mismatch of driver destination preferences compared to a passenger’s destination.
It has also been noted that taxi drivers tend to work in their own service zone. Given
such an existing scenario with taxi drivers having the ability to select their passengers,
taxi drivers could easily serve a passenger within their localized service zone as well
as passengers moving through to different service zones. The second scenario is the
routing strategy for the taxi driver, where a taxi chooses a route based on whether it
has a passenger or not. Choosing the optimum route also plays an important role in
how a taxi driver could efficiently run the service whilst minimizing operational costs.
However, in this study, the scope here is limited to passenger choice optimization
only.
6.1 Passenger Search Policy
Optimization or improvement of driver behavior is introduced based on passenger
choice. Given a taxi agent in a grid cell ‘g’ at a given time ‘t’, the taxi agent starts
to search for passengers in and around the grid which is also known as the search
space. An initial search space consists of a 3 ×3 grid cell each of 500 m ×500 m.
If there are no passengers within the initial search space, then the search space is
increased to 5 ×5 grid cell and so on. During simulation, search time is increased
for each increment of the search space. The passenger is removed from the available
passenger demand so that another taxi cannot select that passenger as they get picked
up. When multiple passengers are available within the search space, the passenger
is chosen based on a cost function. Passenger Selection is made based on a simple
cost function as shown in Eq. 8.
Cost Function =Pick up distance
Trip distance (8)
where, Pick up distance is the distance from the taxi agent to the passenger origin
and the Trip distance is the passenger trip distance from the passenger origin to the
passenger destination. Based on the cost function, the lower cost function value of the
passenger would be the better recommendation for the taxi drivers. The optimized
model is similar to the on-demand taxi service but with a recommendation regarding
which passenger to pick up.
Figure 4shows an example of the passenger search policy for the optimized
simulation model. At first, a taxi present at a grid location is considered as the
previous trip drop off location. Hence, the state of the taxi, as defined by taxi modeling
state, is the ‘free movement’ taxi. However, a different passenger search policy
494 S. Ranjit et al.
Fig. 4 Passenger search policy
is implemented, to search for the passenger in the optimized simulation model as
compared to the usual taxi operation simulation model. In the optimized or the
improved model, the taxi starts to search for passengers at different search levels.
In the given example, the taxi searches for a passenger at the 1st level of the 3 ×
3 grid. The search space is increased to the 2nd level of the 5 ×5gridasthereis
no passenger available at the 1st level of the 3 ×3 grid. For this search level of the
5×5 grid, there are two available passengers Pa and Pb. However, passenger Pa
is closer to the taxi agent when compared to passenger Pb. Also, passenger Pa trip
distance is longer when compared to passenger Pb. With this available information
and evaluation of the cost function, passenger Pa would have a lower cost value
than passenger Pb and hence the choice of the passenger for the taxi agent would
be Pa. In this example, it is also worth noting that, to pick up passenger Pa is easier
as Pa is within the same road network. However, passenger Pb is in the other road
26 Taxi Behavior Simulation and Improvement … 495
network making it difficult to access as compared to passenger Pa. Although both
Pa and Pb are in the same grid level, passenger Pa would be the preferred choice of
pick up. Current optimization policy implements a simple passenger choice method
for taxi drivers to select passengers. However, an additional variable such as a taxi
driver service zone could also be added within the simulation to make the passenger
selection more robust.
As for both the usual taxi operation simulation and optimized simulation, a dif-
ferent number of taxi agents were taken: 3000 taxi agents; 5000 taxi agents; and
10,000 taxi agents. The results from the simulation were then taken for the case
of evaluating the existing usual taxi operation simulation model and the optimized
model. The use of three different number of taxi agents i.e. 3000 taxi agents, 5000
taxi agents and 10,000 taxi agents, was to establish the sensitivity of the agent-based
model itself. Regardless of the number of agents selected, the model should adapt to
generate a similar simulated result. A high-performance computing platform utiliz-
ing the Apache Hadoop/Hive based distributed system (Witayangkurn et al. 2013)
was used for simulation computation.
6.2 Optimization Indicator
In order to quantitatively understand the optimized taxi operation, indicators explain-
ing the optimized service needs to be appropriately defined. The indicator would fun-
damentally distinguish between the existing usual taxi operation simulation model
and the optimized improved model. Three different indicators chosen for passenger
choice optimization are as follows.
Indicators for optimization:
•Reduced passenger waiting time
•Reduced travel distance without passenger
•Increasing the number of passenger trips of the taxi drivers.
Derived indicator for optimization:
•Improved taxi driver daily income.
7 Model Evaluation
Evaluation of the model is an essential aspect of determining its effectiveness. Taxi
behavior simulation for existing usual taxi operation simulation model and an opti-
mized or improved model was done for three different number of taxi agents (3000
taxi agents, 5000 taxi agents and 10,000 taxi agents). Depending on the requirement,
the number of taxi agents could be increased or decreased. The existing usual taxi
operation simulation model is correlated as driving without assistance and an opti-
mized service is correlated as driving with assistance for the purposes of evaluation.
496 S. Ranjit et al.
7.1 Passenger Waiting Time Evaluation
The evaluation between the usual taxi operation simulation model and the optimized
simulation model was done for passenger waiting time. Figures 5,6and 7show the
evaluation result for the three different taxi agents size simulation. Figure 5shows
the evaluation result conducted for the 3000 taxis agent simulation. For the usual
taxi operation simulation model, many taxis had to wait for more than 60 min to get
to their next passenger. For some taxis, the waiting time to get the next passenger
was more than 2 h and more. By contrast, for the optimized simulation model, the
waiting time to get their next passenger was significantly reduced. Similarly, Figs. 6
and 7show the waiting time to get the next passenger for the simulation of 5000 taxi
agents and 10,000 taxi agents respectively. For this case also, the usual taxi operation
simulation had many taxis with passenger waiting times of more than 60 min, while
for the optimized simulation, the passenger waiting times were reduced significantly.
When mentioning the waiting time, if the waiting time to get their next passenger
was higher it would automatically reduce the number of passengers the taxi driver
could have served and hence would result in reduced income. The taxi operation
was deemed as improved based on the indicator evaluating the waiting time in the
optimized simulation.
Fig. 5 Passenger waiting time (WT) comparison with 3000 agents
Fig. 6 Passenger waiting time (WT) comparison with 5000 agents
26 Taxi Behavior Simulation and Improvement … 497
Fig. 7 Passenger waiting time (WT) comparison with 10,000 agents
7.2 Distance Travel Without Passenger Evaluation
The evaluation for the daily distance traveled by taxis without any passengers,
between the usual taxi operation simulation model and optimized simulation model,
conducted shows how distance traveled without a passenger is affected. Figures 8,9
and 10 show the evaluation result for the three different taxi agents size simulation.
Figure 8shows the evaluation result conducted for the 3000 taxis agent simulation.
For the usual taxi operation simulation model, the number of taxis driving more than
60 km daily, without passengers, was significantly large. A considerable number of
taxis were even driving more than 100 km without any passengers. By comparison,
for the optimized simulation model, the number of taxis driving more than 60 km
daily, without passengers, was significantly reduced.
Similarly, Figs. 9and 10 show the daily distance traveled by taxis without pas-
sengers for the simulation of 5000 taxi agents and 10,000 taxi agents respectively.
Similarly, in this case, the usual taxi operation simulation showed a large number of
taxis running with more than 60 km daily without a passenger, while for the opti-
mized simulation, this number was reduced significantly. This can also be inferred
that as the daily distance traveled by taxis, without passengers, tends to reduce, then
taxi operations tend to improve in terms of their operational costs.
Fig. 8 Daily distance travel (DT) without passenger comparison with 3000 agents
498 S. Ranjit et al.
Fig. 9 Daily distance travel (DT) without passenger comparison with 5000 agents
Fig. 10 Daily distance travel (DT) without passenger comparison with 10,000 agents
7.3 Drivers’ Income Evaluation
The evaluation for the daily income of the taxi drivers, between the usual taxi oper-
ation simulation model and the optimized simulation model, conducted shows how
many factors in the improvement level could be achieved in terms of income genera-
tion. Figures 11,12 and 13 show the evaluation results for three different taxi agents’
size simulation. Figure 11 shows the evaluation result conducted for the 3000 taxis
agent simulation. For the usual taxi operation simulation model, the number of taxis
working daily, with income less than 2000 Baht (Baht =Monetary unit of Thailand),
was significantly large. Many taxis were even working with a daily income of less
than 1000 Baht, which is a very low income considering the fuel and maintenance
costs that the drivers need to invest. By comparison, for the optimized simulation
model, the number of taxis working daily, with income less than 2000 Baht was
significantly reduced.
Similarly, Figs. 12 and 13 shows the daily income of taxi drivers for the simulation
of 5000 taxi agents and 10,000 taxi agents respectively. For this case also, the usual
taxi operation simulation showed a large number of taxis working daily with income
of less than 2000 Baht, while for the optimized simulation, there was a significant
increase in the number of taxis working daily with income more than 2000 Baht. In
this regard, the optimized model simulation showed improvement in taxi operation
as many taxis were getting more income, which was the sole purpose of the model.
26 Taxi Behavior Simulation and Improvement … 499
Fig. 11 Drivers’ daily income (IN) comparison with 3000 agents
Fig. 12 Drivers’ daily income (IN) comparison with 5000 agents
Fig. 13 Drivers’ daily income (IN) comparison with 10,000 agents
7.4 Passenger Service Level Evaluation
The evaluation for the daily passenger trips, between the usual taxi operation simu-
lation model and optimized simulation model, conducted shows how the number of
trips varied for the two models. The daily number of trips for the taxi drivers can be
directly correlated with the income generated as well as with the passenger service
level. Figures 14,15 and 16 show the evaluation results for the three different sample
sizes of taxi agents’ simulation. Figure 14 shows the evaluation result conducted for
the 3000 taxis agent simulation. As can be seen, for the usual taxi operation simu-
lation model, the number of daily passenger trips was significantly less. Many taxis
500 S. Ranjit et al.
Fig. 14 Taxi passenger trips (PT) per day comparison with 3000 agents
Fig. 15 Taxi passenger trips (PT) per day comparison with 5000 agents
Fig. 16 Taxi passenger trips (PT) per day comparison with 10,000 agents
were even getting less than 10 passenger trips, which directly corresponds to their
income as mentioned previously. By contrast, for the optimized simulation model,
the number of daily passenger trips increased considerably as many taxis were getting
more than 20 passenger trips.
Similarly, Figs. 15 and 16 show the daily passenger trips for the simulation of
5000 taxi agents and 10,000 taxi agents respectively. For this case as well, the usual
taxi operation simulation showed a large number of taxis had few daily passen-
ger trips, while for the optimized simulation, the number of daily passenger trips
increased considerably. Regarding the daily passenger trips, the optimized model
showed improvement in the taxi operation significantly as many taxis were getting
26 Taxi Behavior Simulation and Improvement … 501
a significant number of passenger trips. This also suggests that the taxi service level
had improved as the increment in the number of trips directly relates to the reduction
of the passenger rejection rate.
Optimization of taxi behavior is an essential aspect for improving taxi drivers’
profit. Simple optimization that focuses on reducing waiting times and increasing
trips, could significantly improve profitability. An improvement method, such as
selecting passengers through a smartphone application, could benefit both taxi driver
income as well as improve passenger service levels.
8 Conclusion
Innovative new service delivery approaches for taxi services in the form of various
ridesharing apps such as Uber, Grab and Lyft taxis are becoming popular in many
countries. Introduction to such ridesharing apps has already resulted in a positive
impact on passenger service levels as mentioned in a report by Wallsten (2015). The
report stated that the number of complaints registered against taxis had started to
follow a decreasing trend as ridesharing app came into service. Hence, improvement
of taxi services through a mobile app recommendation system could further benefit
both taxi drivers as well as passengers. However, to make an improvement to any
system, understanding the existing usual working behavior is essential.
Modeling of taxi service for usual taxi operations shows existing taxi behaviors
at the city level. A data-driven agent-based simulation model provides a way to
simulate taxi behaviors in a large-scale urban area with the taxi probe vehicle data.
Such simulation and modeling of taxi services can provide multiple information
sources on existing working behavior. Information such as taxi drivers daily income,
the daily number of trips taxi drivers get, daily travel distances and the time that the
taxi driver has without passengers, can give an indication of the current service level
and shows the areas where improvement can be made to achieve a better taxi service.
On the other hand, optimization of taxi behavior, based on indicators derived
from the usual taxi operation simulation, provides an essential aspect for improv-
ing the taxi driver’s income and profit. Even with simple optimization that focuses
on reducing waiting times and increasing passenger trips, this could help improve
profitability. Introducing simple improvements such as selecting passenger through
a smartphone application could benefit both taxi drivers’ income as well as improve
the quality of passenger service level. An optimization model provides a recom-
mendation to taxi drivers to pick up suitable passengers. The result is that the taxi
driver can now improve their daily income by getting more passenger trips as well
as reducing their waiting time to get their next passenger and reducing the travel
distance without taking a passenger. This situation has an impact on both drivers as
well as passengers as the monetary gain of drivers is improved and the passenger
rejection rate is reduced. All this suggests that with a simple change in the passenger
search strategy, improvement of overall taxi operation could be achieved. Lastly, an
optimization model could be further improved by introducing route choice behavior
502 S. Ranjit et al.
along with passenger selection choice with proper fleet management that could be
introduced in the model by comparing the relationship between supply and demand,
for efficient driving. In addition, having fore knowledge of taxi drivers’ service zone
could further help optimizing the system to provide a better recommendation system
for taxi drivers.
Acknowledgements This research was facilitated and funded by Shibasaki Laboratory (http://
shiba.iis.u-tokyo.ac.jp/) of The University of Tokyo. This research was partially supported by Toyota
Tsusho Nexty Electronics (Thailand) Co., Ltd., by providing Taxi Probe Data for research purpose.
References
Abar S, Theodoropoulos GK, Lemarinier P, O’Hare GMP (2017) Agent based modelling and
simulation tools: a review of the state-of-art software. Comput Sci Rev 24:13–33. https://doi.
org/10.1016/j.cosrev.2017.03.001
Baster B, Duda J, Maciol A, Rebiasz B (2013) Rule-based approach to human-like decision simulat-
ing in agent-based modeling and simulation. In: 17th International conference on system theory,
control and computing, ICSTCC 2013; joint conference of SINTES 2013, SACCS 2013, SIMSIS
2013—Proceedings, pp 739–743. https://doi.org/10.1109/icstcc.2013.6689049
Bischoff J, Maciejewski M, Sohr A (2015) Analysis of Berlin’s taxi services by exploring GPS
traces. In: International conference on models and technologies for intelligent transportation
systems, MT-ITS 2015, (June), pp 209–215. https://doi.org/10.1109/mtits.2015.7223258
Bonabeau E (2002) Agent-based modeling: methods and techniques for simulating human systems.
Proc Natl Acad Sci 99(suppl. 3):7280–7287. https://doi.org/10.1073/pnas.082080899
Castro PS, Zhang D, Li S (2012) Urban Traffic modelling and prediction using large scale taxi
GPS traces. In: Kay J, Lukowicz P, Tokuda H, Olivier P, Krüger A (eds) Pervasive computing.
Springer, Berlin, Heidelberg, pp 57–72. https://doi.org/10.1007/978-3-642-31205-2_4
Cheng SF, Nguyen TD (2011) TaxiSim: a multiagent simulation platform for evaluating taxi fleet
operations. In: 2011 IEEE/WIC/ACM international conferences on web intelligence and intelli-
gent agent technology, vol 2, pp 14–21. https://doi.org/10.1109/wi-iat.2011.138
Deng Z, Ji M (2011) Spatiotemporal structure of taxi services in Shanghai: using exploratory
spatial data analysis. In: 19th International conference on geoinformatics, pp 1–5. https://doi.org/
10.1109/geoinformatics.2011.5981129
Ester M, Kriegel H, Sander J, Xu X (1996) A density-based algorithm for discovering clusters a
density-based algorithm for discovering clusters in large spatial databases with noise. In Proceed-
ings of 2nd international conference on knowledge discovery and data mining (KDD-96)
Gan J, Tao Y (2015) DBSCAN revisited: mis-claim, un-fixability, and approximation. In: Proceed-
ings of the 2015 ACM SIGMOD international conference on management of data, pp 519–530.
https://doi.org/10.1145/2723372.2737792
Gonzales E, Yang C, Morgul F, Ozbay K (2014) Modeling taxi demand with GPS data from taxis
and transit. Retrieved from http://transweb.sjsu.edu/PDFs/research/1141-modeling-taxi-demand-
gps-transit-data.pdf
Grau JMS, Romeu MAE (2015) Agent based modelling for simulating taxi services. Proc Comput
Sci 52(1):902–907. https://doi.org/10.1016/j.procs.2015.05.162
Lindfield G, Penny J (2017) An introduction to optimization. Introduction to nature-inspired opti-
mization. https://doi.org/10.1016/b978-0-12-803636-5.00001-3
Liu K, Yamamoto T, Morikawa T (2008) Study on the cost-effectiveness of a probe vehicle system
at lower polling frequencies. Int J ITS Res 6(1):29–36. https://trid.trb.org/view/888064,http://
www.its-jp.org/journal/papers/50.pdf
26 Taxi Behavior Simulation and Improvement … 503
Maciejewski M, Salanova JM, Bischoff J, Estrada M (2016) Large-scale microscopic simulation of
taxi services. Berlin and Barcelona case studies. J Ambient Intell Hum Comput 7(3):385–393.
https://doi.org/10.1007/s12652-016-0366-3
Miwa T, Sakai T, Morikawa T (2004) Route identification and travel time prediction using probe-car
data. Int J ITS Res 2(1):21–28
Moreira-Matias L, Gama J, Ferreira M, Mendes-Moreira J, Damas L (2013) On predicting the
taxi-passenger demand: a real-time approach. In: Correia L, Reis LP, Cascalho J (eds) Progress
in artificial intelligence, vol 8154 LNAI. Springer, Berlin, Heidelberg, pp 54–65. https://doi.org/
10.1007/978-3-642-40669-0_6
Nagashima Y, Hattori O, Kobayashi M (2014) Improvement of traffic signal control using probe
data. SEI Technical Rev (78): 44–47. https://global-sei.com/technology/tr/bn78/,https://global-
sei.com/technology/tr/bn78/pdf/78-09.pdf
Nam D, Hyun K, Kim H, Ahn K, Jayakrishnan R (2016) Analysis of grid cell–based taxi ridership
with large-scale GPS data. Transp Res Rec: J Transp Res Board 2544(March 2017):131–140.
https://doi.org/10.3141/2544-15
Peungnumsai A, Witayangkurn A, Nagai M, Arai A, Ranjit S, Ghimire BR (2017) Bangkok taxi
service behavior analysis using taxi probe data and questionnaire survey. In: Proceedings of the
4th multidisciplinary international social networks conference, pp 1–8. https://doi.org/10.1145/
3092090.3092117
Peungnumsai A, Witayangkurn A, Nagai M, Miyazaki H (2018) A taxi zoning analysis using large-
scale probe data: a case study for metropolitan Bangkok. Rev Socionetwork Strat. 12(1):21–45.
https://link.springer.com/article/10.1007/s12626-018-0019-4
Ranjit S, Nagai M, Witayangkurn A, Shibasaki R (2017) Sensitivity analysis of map matching
techniques of high sampling rate GPS data point of probe taxi on dense open street map road
network of Bangkok in a large-scale data computing platform. In: 15th International conference
on computers in urban planning and urban management
Ranjit S, Witayangkurn A, Nagai M, Shibasaki R (2018) Agent-based modeling of taxi behav-
ior simulation with probe vehicle data. ISPRS Int J Geo-Inf 7(5):177. https://doi.org/10.3390/
ijgi7050177
Rothlauf F (2011) Optimization problems. In: Design of modern heuristics, pp 7–45. https://doi.
org/10.1007/978-3-540-72962-4
Sadahiro Y, Lay R, Kobayashi T (2013) Trajectories of moving objects on a network: detection of
similarities, visualization of relations, and classification of trajectories. Trans GIS 17(1):18–40.
https://doi.org/10.1111/j.1467-9671.2012.01330.x
Tang J, Jiang H, Li Z, Li M, Liu F, Wang Y (2016) A two-layer model for taxi customer searching
behaviors using gps trajectory data. IEEE Trans Intell Transp Syst 17(11):3318–3324. https://
doi.org/10.1109/TITS.2016.2544140
Thairath (2017) Bangkok public transport complaints: taxis top list—refusing customers still champ!
Retrieved from https://news.thaivisa.com/article/15623/bangkok-public-transport-complaints-
taxis-top-list-refusing-customers-still-champ
Torrens PM (2010) Agent-based models and the spatial sciences. Geogr Compass 4(5):428–448.
https://doi.org/10.1111/j.1749-8198.2009.00311.x
Wallsten S (2015) Has uber forced taxi drivers to step up their game? The Atlantic. Retrieved from
https://www.theatlantic.com/business/archive/2015/07/uber-taxi-drivers-complaints-chicago-
newyork/397931/
Witayangkurn A, Horanont T, Shibasaki R (2013) The design of large scale data management for
spatial analysis on mobile phone dataset. Asian J Geoinf 13(3):17–24
Wong DWS, Huang Q (2016) Sensitivity of DBSCAN in identifying activity zones using online
footprints. Proc Spatial Accuracy 2016:151–156
Yang Q, Gao Z, Kong X, Rahim A, Wang J, Xia F (2016) Taxi operation optimization based on big
traffic data. In: Proceedings—2015 IEEE 12th international conference on ubiquitous intelligence
and computing, 2015 IEEE 12th international conference on advanced and trusted computing,
504 S. Ranjit et al.
2015 IEEE 15th international conference on scalable computing and communications, vol 20, pp
127–134. https://doi.org/10.1109/uic-atc-scalcom-cbdcom-iop.2015.42
Yao H, Wu F, Ke J, Tang X, Jia Y, Lu S et al (2018) Deep multi-view spatial-temporal network for
taxi demand prediction. CoRR. https://arxiv.org/abs/1802.08714v2
Yuan NJ, Zheng Y, Zhang L, Xie X (2013) T-finder: a recommender system for finding passengers
and vacant taxis. IEEE Trans KnowlData Eng 25(10):2390–2403. https://doi.org/10.1109/TKDE.
2012.153
Zhang S, Wang Z (2016) Inferring passenger denial behavior of taxi drivers from large-scale taxi
traces. PLoS ONE 11(11):1–21. https://doi.org/10.1371/journal.pone.0165597
Zheng Z, Rasouli S, Timmermans H (2014) Evaluating the accuracy of GPS-based taxi trajectory
records. Proc Environ Sci 22:186–198. https://doi.org/10.1016/j.proenv.2014.11.019
Index
A
Accessibility, 9,13,21,30,31,70,77,81,86,
95,130,133,181,182,205,206,209,
210,212,214,217,218,230,361,391,
392,401,407,410,411,415,421,446,
463–468,470–472,474,476–479,481
Accumulated minimal cost, 477
Activity-based transportation model, 207
Adaption, 421
Age, 3,4,54,55,58,60,63,64,66,76,96,
144,186,196,197,200,210,250,269,
284,287,296,297,358,359,365,395,
396,448
Aged-care facility, 390,391,394,396,399,
401
Agent-Based Modeling (ABM), 491
Agent-Based Modeling and Simulation
(ABMS), 486
Air quality, 2,205–207,213,215,220
Application Programming Interfaces (APIs),
73,96
Applied statistical method, 390
Architecture, 10,86,144,227,231,250–254,
256,258–263,267,268,270,279,485,
491
Area-based learning, 10,283–285,287,
289–291,294,295,297–301
Augmented Reality (AR), 10,110,283–286,
291,292,294–300
Australia, 2,4,5,11,13,21,140,141,150,
159,160,271,290,306,347,348,350,
361,373,374,383–385
B
Bangladesh, 406
Baseline condition, 410,418,420
Behavior rules, 485–487
Benefit-cost analysis, 215
Betweenness centrality, 8,12,85–92,94,95,
405–422,465
Bicycle infrastructure, 67
Bicycle policy, 376,377,385
Bicycle ride safety, 374
Big data, 4,6,7,18,24,30,54,57,101,109,
119,120,133
C
Carbon emissions, 6,9,25,27,28,126,
179–183,185,186,188–190,192,193,
195–197,199–201,348,376
Case study, 9,12,55,67,141,142–144,146,
149,150,152,153,156–159,211,212,
220,322,348,374,378,405,413,443,
444,448
Cellular Automata (CA), 166–168,206,208,
209,211,213,214,447
Centrality, 12,90,406,408,410,412,413,
420,421,465,466,468,472,477,478,
481
China, 8,9,13,78,79,81,99–101,103–107,
109,111,117–120,122,124,165,
179–181,183,191,200,201,408,465,
466,472,474,476–478
Cities, 1–9,14,17–25,27–32,36,47,53–55,
59,62,67,69,75,77,81,85–87,95,97,
©Springer Nature Switzerland AG 2019
S. Geertman et al. (eds.), Computational Urban Planning and Management
for Smart Cities, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3-030-19424-6
505
99,101,104,105,108–110,112,120,
124,126,139–141,150,152,154,164,
187,199–201,224–227,259,268,272,
273,275,277,279,306,307,322,347,
348,350,351,358,374,376,385,391,
395,408,414,420–422,443–445,463,
464,466–468,473,474,476–478,481,
484,486,487
Climate change, 2,3,5,19,30,35,36,39,43,
47–50,123,145,160,164,186
Climate resilience, 7,48,49
Climate-sensitive planning, 148,159
Climate variability, 36,42,45,48,50
Cloud computing, 3,4
Clustering, 20,21,70,188,189,192,249,251,
255,260,262,413,427,428
Collaboration, 7,10,69,71,78,80,100,101,
106,110,112,144,151,159,160,230,
267–269,271,277,278,280
Collaborative learning, 10,268,269,271,272,
278–280
Collaborative planning, 119,140,279
Collaborative teaching, 10,270,279
Commercial activity, 87
Communication, 4,7,18,22,30,69,70,
72–74,77,78,80–82,105,149,154,
157,158,232,267,268,270,271,273,
277,278,280,488
Communicative planning, 81
Commuting, 11,28,31,59,134,181,182,184,
185,187,188,191–200,348–354,
357–361,483,484
Complexity, 13,101,159,176,422,464
Computational planning, 1,6,8,13
Consistency analysis, 166
Convolutional Neural Network (CNN), 252,
253
Cost function, 493,494
D
2D, 287,290,291,295
3D, 10,267–280,284–287,289,290,296
Dashboard, 5,31
Data analytics, 18
Database, 21,60,71,91,186,214,285,289,
295,307,309,310
Data cleaning, 55,58
Data processing, 211–213,395,396
Data protection, 7,81
Data storage, 211
Decision making, 6,7,47,70,220,447,484,
492
Deep learning, 10,249–251,268,278,279
Demand, 9,24,30,31,45,49,59,62,67,76,
120,121,126,132–134,180,206,207,
209,211,213,214,216,218,220,223,
240,246,305,306,308,314,315,318,
391,408,413,426,444,447,485–487,
489,491–493,502
Density, 8,37,76,85,86,88,90,94,95,142,
145,149,151,152,154,166,181,182,
184,186,187,189,190,197,199,237,
270,348,358,374,385,392,393,396,
398–401,434,450,489
Design, 3,9–11,18–22,28,47,49,71,79,87,
97,102,106,117,118,120,122–130,
134,139–141,148–150,154,157–160,
229–231,244,250,251,255,256,
258–260,262,263,267–271,278,279,
285,286,305,306,314,373–378,381,
385,464,491
Destination, 11,12,54,57,62,64–66,76,132,
305–309,311,313–315,361,363,364,
405–422,448,455,468–471,473,474,
484,485,487,489–491,493
Digital Elevation Model (DEM), 142,212
Disaster management, 406,410
Discrete choice modelling, 354,360
Disruption, 12,405–407,411,412,418–420
Distance, 12,20,22,31,56,57,85,106,166,
167,181,182,185,188–193,198–200,
209,260,296,307–309,311–313,317,
318,343,348,349,353,354,358,362,
391–393,396–399,401,402,410,428,
430,431,443,446,450,455,459,465,
466,469,474,489,490,492–495,497,
498,501
Distribution analysis, 129,412,420
Docking station, 54,56–58
Dubai, 9,10,223,225,227–229,231–234,
236–242,244–246
E
Earthquake, 406,415,439,447
Ecological, 37,47,106,117,121,124,126,
127,134,142,174
Economic, 2,4,5,9,11,19,21,27,37,40,47,
103,107,124,130,143,146,148,158,
175,176,179,183,198,201,206,208,
214,224,226,227,232,347,348,352,
359,361,390,463,464,466,476–478,
480
Ecosystem, 30,49,76,77,106
Electricity, 3,26,27,142,180,188
Elman Recurrent Neutral Network (ERNN),
11,321,322,332–344
506 Index
Emergency
management, 410,420,444,446,447
Emission, 9,22,25,27,28,43,133,179,180,
182–188,192–201,213,214,220
Energy, 3,4,6,19,23,26–28,30,36,43,44,
124,125,149,164,179–182,184,229,
231,244,478
Environment, 3,4,6,8,10,28,39,42,43,45,
47,48,54,63,67,74,107,108,117,
118,121–131,142,145,157,164,180,
184,214,215,218,220,223–225,229,
244,267–273,275–280,284,322,
349–351,354,359–361,374,376,385,
460,485,491,492
Equilibrium, 208,209,211,215,456
Evacuation planning, 12,444,446–448,459
Evaluation, 10,12,13,75,117,124–127,129,
130,132,133,140,145,156,159,214,
215,238,242–246,267–269,280,283,
298–301,322,338,341,376,381,382,
389–392,394–396,399–403,421,430,
432,435,465,483,494–499
Experiment, 12,105,124,253,263,267,268,
271–273,275,278,279,425,427,432,
435–440
Expert, 25,79,144,159,229,245,271,322,
451
F
Failure simulation, 406
Feedback, 10,99,103,110,131,133,141,
151,153,209,267,273,279
Flow analysis, 87,88,90–96
Forecast Analysis Zone (FAZ), 206
Forecasting, 11,47,207,209,214–216,307,
321,322,327,329,332–335,338–344
Framework, 8,9,47,119–122,124,127,128,
131,134,135,139–141,156,157,159,
165,166,168,175,205,208,214,215,
245,246,250,451,467,481
Fuzzy C-means clustering, 12,425,427,440
G
Gaming, 270,279
Gender, 54,55,58,60,63,66,67,76,296,
359,365
Geodesign, 8,9,139–146,148,149,152,153,
156–160
Geographic Information (GI), 77,390
Geolocation, 86,88,96
Geospatial, 35,86,188,402,403
Giant Connected Component (GCC), 406,411,
414,415
Google, 189,252,256,270,290,297,309,
379,436,465
Governance, 1,3,6–8,19,47,78,100–112
Graphic User Interface (GUI), 296
Graph network, 89
Greenfield, 7,18–20,32,148
Greenhouse, 26,36,43,205–207,213,348
Grids, 107
Gross Domestic Product (GDP), 19,164,175,
474,476–478,480
H
Health, 28,31,54,109,110,121,132,133,
143,144,164,213,242,243,270,351,
354,362,363,389–391,410,413,415
Hospital, 12,25,143–145,149,198,391,406,
407,409–416,418,420,421,446
Household, 5,9,23,26,27,108,179–186,
188–193,195–201,205–207,209–218,
220,306,316,349–351,359,365,368,
448
Housing, 22,49,70,100,106,112,122,145,
149,151,152,180,182,198,200,201,
246,349,354,360,396,407,408,414,
421,448
Human activity, 36,74,75,87,92
Human-oriented, 121
Hungary, 55
I
Income, 13,31,76,180,181,186,191,
195–197,199–201,207,210,224,350,
351,359,367,391,484,486–488,492,
495,496,498–501
India, 7,35–37,47,49
Indicator, 5,54,128,176,188–190,205,218,
220,233,257,272,336–338,340–343,
391,392,401,406,465,477,479,495,
496,501
Individual activity patterns, 7,69,70,72,74,
80
Industry
mix, 21
Information and Communication Technology
(ICT), 6,7,17,18,20–22,27,29,31,
32,80,100,102,103,106,108,110,
112,268,284
Information infrastructure, 17,20
Informed urbanization, 1,6
Infrastructure, 3,4,7,10,12,18–20,22,27,
29,30,48,49,86,106,110,141,142,
144,145,149,152–154,156–158,175,
223–227,231,242,243,246,306,317,
Index 507
349,350,354,373,376,378,382,406,
420,445,448,450,463,465,478,481
Innovation, 1,3,5,86,99,104,105,109–112,
224,230,279,322
Integrated land use transport planning, 8,139
Integrated platform model, 205,207
Interactive, 6,71,92,102,144,214,268,279
Interconnectivity, 486
Internet, 2–5,18,19,70,71,81,109,110,256,
259,439,463,465,466,474
Intersection design, 374–376,378,381,385
Iran, 13,350,443–445,448,456,458
Isochrone, 408,410,413
Italy, 406
J
Japan, 12,284,289,389,390,395,396,402,
426,439
Job centers, 408,414,421
K
Knowledge, 1,2,6,8,14,20,67,71,79,95,
99,100,102,112,141,148,157,158,
169,180,183,263,285,306,308,361,
376,450,502
Korea, 6,7,18–23,26,27,29,31,32,181,
225,307
L
Landscape, 7,9,69,72,76,77,80,128,129,
144,149,154,163–165,167–176,259,
287,290,291
Land spatial distribution, 217
Land-use
planning, 119,131–133,183,201,206
Latitude, 90–92,96,311,488
Legacy plans, 229,230,232
Livability, 122
Localized service zones, 486
Longitude, 90–92,96,311,488
M
Magnitude of disruption, 407,419
Mapping, 5,86,141,164,253,290,297,299,
390,488
Market failure, 67
Mass transit, 139
Matrix, 163,170,212,241–244,253,255–257,
305,318,323,325,328,329,335,341,
343,344,451,452,468,472
Median Share Ratio (MSR), 12,389,391–394,
398–403
Mega event, 9,10,223–227,229,230,246
Metadata, 91,95,96
Metrics, 5,6,9,140,158,163,165–168,170,
171,175,408–410
Micro-ecology, 8,122–126
Micro-level analysis and simulation, 481,
484–486
Micro-scale planning, 8,118,119
Mitigation, 375
Mixed Reality (MR), 10,283–286,291–293,
296–300,363
Mobile
social media, 75,86
source emissions, 9,205,207,213,219,
220
Mobility, 3,6,42,53,55,133,180,190–192,
209–211,225,229,232,246,305,306,
347–350,360–362,486,492
Modal shift, 11,347–352,354,356,358–361
Modeling/modelling, 1,7,11,12,47,69,70,
72,74,80,88,95,164,176,206,207,
214,233,236,238,244,269,270,272,
273,275–277,279,329,333,341,349,
354,356,360,373,382,385,391,447,
464,485–487,489–493,501
Monitoring, 19,23,25,30,95,101,107,109,
141,164,207,469
Multi-agent model, 206,208
Multi-attribute ranking method, 408
Multi-modal, 242,352
Multinomial Logit Model (MNL), 208,209,
213
Multi-path error, 488
N
Natural disaster, 450
Navigation, 12,275,307,425,443,459,
464–466,484
Nepal, 12,405,406,413,415,418
Network
analysis, 165,169,175,176,446,465,472
Nightlife, 62
O
Occupation, 85,154
Open data, 4–7,23,24,30,35,36,47,58,145,
156
Open source, 5,156
Open Street Map (OSM), 91,92,323,324,
448,488
Optimization model, 485,501
Origin-Destination (OD)
matrix, 306,307,314,316,318
Orthogonal grid pattern, 55
Outputs, 9,139,141,143,156,158,159,214,
215,220,244,255,259,354,455
508 Index
P
Participation, 7,19,29,30,69,70,72,73,
77–82,100,102,108–110,271,275,
350,391
Participatory planning, 80
Passengers
search policy, 493,494
service level, 499,501
waiting time evaluation, 496,497
Passive defense, 446
Path Size Logit(PSL) model, 492
Pattern analysis, 342
Physical space, 7,27,29,127
Planning framework, 464
Planning practice, 79
Planning process, 19,35,79,82,229,446
Planning support, 13,82,117–119,121,122,
132–134,272
Planning Support Science (PSScience), 13
Point of Interest (PoI), 86,90,92,95,473
Policymaking, 180,201,214
Population, 2,3,7,8,12,13,18–20,36,37,
44,47–50,54–56,69,70,81,82,90,96,
108,124,139,140,144,145,148,150,
152,160,164,175,182–184,189,190,
197,207,209,217,218,220,224,
233–236,246,307,316,317,322,348,
358,389–391,393–403,443–446,448,
450,452,456,473
Portal, 35,37,107,145
Post-event planning, 223,225
Practitioner, 55,67,157–159
Pricing model, 54
Privacy, 31,69,70,81
Probability, 210,211,257–259,356,357,359,
360,392,393,396,398–401,412,
418–420,489–491
Process, 9,11–13,22,30,54,55,58,78,87,
88,92,100–102,104,107,110,111,
125,139–141,143,145,149,152,
156–159,163–166,171–174,176,201,
209,211,213,215,220,229,244–246,
253,257,260,262,263,269–271,273,
279,295,296,321,329,334,335,338,
342,344,394,407,415,421,433,444,
446,447,455,456,486,489,492
Profitability, 485,487,501
Public activity, 8,85
Public open data, 390,402
Public space, 48,85,86,130,133,231
Public transport, 11,24,54,57,67,134,152,
154,181,193,199,226,305–309,318,
348,349,351,352,358,359,361,363,
376,460,465
Q
Qualitative, 31,175,241,245,351
Quality of life, 3,6,21,22,24,27,29,30,32,
80,109,122,272,405
Quantitative, 117,118,127–129,172,175,
211,447,484–486,488
R
Random disruption scenario, 411,412,414,
415,421
Ranking, 4,408,451
Real-time, 5,6,19,22,24,30,56,70,74,88,
91,95,158,272,273,277–280,377,
427,466,474,484
Real-world representation, 411,415
Recovery plan, 405,418
Regional trafficflow, 474,477
Regional travel, 12,426–428,430,432,
434–436,439,440
Regression analysis, 9,63,179,197,200,335
Reliability, 306,318,342,351,363,446,486
Relief activity, 12
Rescue activity, 429
Resilience, 406,420–422
Responder, 426,428,430,431,434,436,439,
440
Road hierarchy, 213,218,408,410
Road network, 12,13,133,199,214,220,324,
325,343,391,407,408,413,444,446,
448,455,456,460,484,488–490,494,
495
Route optimization, 445,447
Route preference, 474
Routing, 12,13,362–369,408,443,444,448,
455,456,459,491–493
S
Sample size, 58,257,262,316,317,342,411,
415,416
Scenario
planning, 6,215
Search space, 493,494
Sensitivity analysis, 207,311,318
Sensor, 18,22–24,75,87,101,377,379
Service level, 485,488,492
Shared-bike travel, 7
Shelter, 12,13,133,410,444,446–448,450,
452,454–456,459
Shortest path, 86–90,92,443–445,448,455,
459,465
Sightseeing, 10,283–285,290,291,297–300
Simulation, 9,12,13,87,117–119,124,127,
131–133,206,211,217,220,350,394,
405,407,411,415,416,425,427,
Index 509
432–434,439,440,447,455,460,
485–490,492–501
Single service, 12,405,413
Smart
mobility, 1,4,11,75
Smart card, 11,133,305–309,316–318
Smart cities, 1,3–8,13,14,17–20,22–24,
27–32,70,78,80,100,102,122
Smart devices, 53
Smart environment, 22
Smart governance, 8,99–112
Smart people, 6,17,28,30,100,229
Smartphone, 70,86,286,297,369,436,439,
501
Social activity, 62,70
Social media, 6–8,10,69–82,85–88,95–97,
101,110,158,272,283–285,287,290,
295,297,299,300,484
Software, 37,72,87,110,139,144,157–159,
166,169,212,233,236,272,275,278,
279,285,286,291,294,296,300,318,
377,378,381,382,385
Space-Time Autoregressive Integrated Moving
Average (STARIMA), 11,321,322,
328–332,334,340–344
Space-time prediction model, 342,344
Spatial allocation, 214
Spatial correlation analysis, 94,188,199,200
Spatial dynamics, 9,163,164,176,208
Spatial information science, 283,484,486
Spatial patterns, 124,125,132,198,199,206,
218,220
Spatial planning, 49,118
Spatio-temporal, 11,74,88,322,324,335,
338,344
Spatio-temporal information, 10,283–286,300
Speed limit, 376,408–410,413
Stakeholder, 6,81,82,101,102,104,110,154,
209,229,238,241,245,246,271,277,
279,420
Structural analysis, 421
Subscription, 60
Supply, 45,66,126,149,209,211,213,229,
246,413,427,486,502
Sustainable development, 9,13,123,206,227,
244,246,348
Sustainable transport, 54,349,359
System
bike-sharing system (BSS), 54–56,58,60,
62,64,67
geographic information system (GIS), 10,
24,30,31,37,43,75,80,119,128,
130,133,141,144,158,159,189,215,
272,283–287,289–291,295,296,300,
390,391,394,446,447,459
global positioning system (GPS), 67,133,
283,291,294,296,300,312,318,466,
474,484,486,488,489,491,492
intelligent transportation system (ITS), 486
management, 22–25,27,28,30,275,377,
382
micro-scale planning support system
(mPSS), 119
planning support system (PSS), 8,118–120,
123,134,140,149,158,160
recommendation system, 501,502
urban system, 87,88,176
visual analytics system (VAS), 95
T
Taxi behavior modeling and simulation, 13,
483–485,487–492,495,501
Taxi operation, 483,485,487–501
Taxi service, 13,483–487,492,493,501
Team communication, 267
Technology, 1–3,5,7–9,13,14,18,28,31,32,
49,54,81,86,99–103,105,106,
109–112,118,119,124,133,139,141,
143,144,156,227,232,260,278,376,
484,486
Temporal, 11,35–37,86,123,125,127,165,
166,170,172,220,263,305–307,322,
325,328,329,333–336,342,344,403,
485–487,491,492
Thailand, 484,488,498
Time, 2,3,5,7,9,11,13,19,24,27,28,31,
35,36,40,41,43,53–59,61–64,66,72,
73,78,80,85,86,88,89,91,93,95,
107,118,122,123,130–134,141,148,
154,157,158,160,163,165,169–172,
175,176,187–189,191–193,200,205,
209,213,218,220,242,251,263,272,
278,279,291,292,296,300,306–310,
312,313,321–329,334,335,340,
342–344,352,353,361–363,366,378,
381,382,402,403,409–411,413,415,
420,426–428,432–436,439,440,443,
444,447,455,456,459,463–466,
468–470,474,477,486–493,495,496,
501
Tool, 3,5,6,72,73,77–79,87,97,118,119,
128,140,141,144,156,160,189,192,
193,212,223,225,267–273,277,279,
280,342,376,377,385,390,413,420,
422,444,446,455,459,463,465
Topography, 460
510 Index
Traffic
analysis zones (TAZ), 206
emissions, 213,214,218,220
flow, 11,24,25,214,216,321–325,327,
329,334,335,340,341,344,373–375,
377,408,421,444,446,466,476,481
forecasting, 321
Transformational infrastructure, 139,174
Transparency, 4,101
Transport
planning, 8,139,153,305,352
Transportation modelling, 1,236,447
Transportation network, 12,183,205,207,
212,216,217,223,405–408,410,414,
420,421,444,465,478
Transportation planning, 76,134,206,214,
225,314
Travel behavior, 9,53,54,57,58,133,181,
182,189
Travel demand model, 9,205
Travel mode, 150,181,184,185,190,193,
194,350,354,358
Travel patterns, 7,53,54,307,313,318,325,
347,360
Travel route extraction, 467,468
Travel time, 9,184,189,191,205,212,214,
242,323,354,378,382,409,410,413,
428,430,434,468–471,474,490–492
Trip chain model, 11,305–309,311–314,
316–318
Trip duration, 55,58,63–66
Trust, 99,104,105,108,109,111,112
U
Ubiquitous city, 17,19,22,26,30
Uncertainty, 2,451,471
United Kingdom (UK), 3,349–351
United States of America (USA), 408
Urban context, 29,99,103,105,109,110,112,
421
Urban design, 8,18,20,21,27,29,86,117,
118,121–124,127–130,134,141,145,
268,277
Urban development, 22,32,36,37,42,45,
47–50,79,104,106,117,118,121,
129,140,141,146,148,149,154,158,
159,164,213,224–227,229,238,244
Urban governance, 101,118
Urban growth, 9,36,140,163–165,168,172,
175,176,209
Urbanism, 4,87,97,180
Urbanization, 99,100,103,105,106,112,120,
164–166,174–176,180,201
Urban land use, 7,69,74,75,80,163,172,176
Urban morphology, 164
Urban planning, 2,7,10,13,25,27,36,48,49,
69–73,77,79–82,87,97,99,101,106,
111,117–119,121–124,131,157,164,
176,263,267,268,279,361
Urban space, 7,22,28,32,70,85–88,96,118,
121,122,127,128,131–134
Urban sprawl, 9,163–165,167,172,175,176,
348
Urban zoning, 209
Usability, 269,275,277,278
Usefulness, 70,112,285,298–301
V
Validation, 166,207,211,213,307,316,317,
338,342,382,474
Variables, 36,39,50,74,76,105,175,176,
195,196,316,328,333,335,344,354,
356–359,451,464,489,491
Vehicle behavior modeling, 486
Vehicle Hours Traveled (VHT), 205,218–220
Virtual, 10,78,267–280,283–285,290,292,
294,296,300,427
Virtual Learning Environment (VLE), 10,267,
268
Virtual Reality (VR), 10,269,276,283–288,
290–292,296–300
Virtual social, 71,74
Virtual world, 267,271,273,275,277,278
Visualization, 86,92,131,252,253,262,270,
413,465,466
W
Weather conditions, 59,62
Web application, 12,286,425,427,436,440
Weighting route properties, 471
Workshop, 159
Z
Zoning, 132,142,214,246
Index 511
