Content uploaded by Shaoying Li
Author content
All content in this area was uploaded by Shaoying Li on Jan 06, 2021
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tgis20
International Journal of Geographical Information
Science
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tgis20
Understanding the modifiable areal unit problem
in dockless bike sharing usage and exploring the
interactive effects of built environment factors
Feng Gao , Shaoying Li , Zhangzhi Tan , Zhifeng Wu , Xiaoming Zhang ,
Guanping Huang & Ziwei Huang
To cite this article: Feng Gao , Shaoying Li , Zhangzhi Tan , Zhifeng Wu , Xiaoming Zhang ,
Guanping Huang & Ziwei Huang (2021): Understanding the modifiable areal unit problem in
dockless bike sharing usage and exploring the interactive effects of built environment factors,
International Journal of Geographical Information Science
To link to this article: https://doi.org/10.1080/13658816.2020.1863410
Published online: 05 Jan 2021.
Submit your article to this journal
View related articles
View Crossmark data
RESEARCH ARTICLE
Understanding the modiable areal unit problem in dockless
bike sharing usage and exploring the interactive eects of
built environment factors
Feng Gao
a
, Shaoying Li
a,b
, Zhangzhi Tan
c,d
, Zhifeng Wu
a,b
, Xiaoming Zhang
e,f
,
Guanping Huang
a
and Ziwei Huang
a
a
School of Geography and Remote Sensing, Guangzhou University, Guangzhou, China;
b
Southern Marine
Science and Engineering Guangdong Laboratory, Guangzhou, China;
c
School of Intelligent Systems
Engineering, Sun Yat-sen University, Guangzhou, China;
d
Shenzhen Urban Transport Planning Center Co.,
Ltd, Shenzhen, China;
e
Guangzhou Urban Planning and Design Survey Research Institute, Guangzhou,
China;
f
Guangdong Enterprise Key Laboratory for Urban Sensing, Monitoring and Early Warning,
Guangzhou, China
ABSTRACT
Understanding the inuence mechanisms of dockless bike-sharing
usage is essential for land use planning and bike scheduling strat-
egy implementation. Although various studies have been carried
out to explore the impact of built environment (BE) factors on bike-
sharing usage, few studies have examined the modiable areal unit
problem (MAUP). Moreover, previous studies mainly focused on the
separate eect of each factor but neglected the interactions
between these factors. Taking Shenzhen, China as the case, this
study lls these two gaps by employing the geographical detector
method to examine the MAUP in dockless bike-sharing usage as
well as the interactive eects of BE factors. The results revealed that
the inuences of most BE variables are sensitive to the spatial areal
units, which have informed urban planners what built-environment
factors should be paid more attention to at certain spatial scales.
Additionally, through the comparisons between single eect and
interactive eect, this study revealed some interesting ndings that
can provide scientic basis for temporal rebalance strategy for the
innovative and high-density metropolis in China.
ARTICLE HISTORY
Received 27 December 2019
Accepted 8 December 2020
KEYWORDS
Geographical detector;
interactions; dockless bike
sharing; built environment;
social media data; Modifiable
areal unit problem
1. Introduction
In the past decades, public bike-sharing has grown rapidly and has spread across cities
worldwide due to its benets in providing convenient connections to the transit stations,
alleviating trac congestion, bringing healthy benets, etc (Shaheen et al. 2010, Rixey
2012, Zhang et al. 2014, Fishman 2015, Faghih-Imani and Eluru 2015, 2016, Wang et al.
2016a, El-Assi et al. 2017, Ma et al. 2018, Chen et al. 2019). With the recent boom of the
sharing economy, the dockless bike-sharing system, which allows users to rent a bicycle
through a smart-phone application, has dramatically expanded around the world (Shen
et al. 2018, Xu et al. 2019). Understanding the impact factors of dockless bike-sharing
CONTACT Shaoying Li lsy@gzhu.edu.cn
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE
https://doi.org/10.1080/13658816.2020.1863410
© 2021 Informa UK Limited, trading as Taylor & Francis Group
usages has two important meanings. On the one hand, bike-sharing travel is usually
encouraged by urban planners dues to the fact that bike-sharing systems have numerous
benets, such as promoting public transit use, reducing trac congestion, and increasing
physical activity and health (Shaheen et al. 2010, Fishman 2015, Faghih-Imani and Eluru
2016, El-Assi et al. 2017, Chen et al. 2019). Examining the inuencing mechanisms of bike-
sharing usages is essential because it can provide implication for urban planners who aim
to promote bike usages, in terms of land use development, road design, and so on
(Faghih-Imani and Eluru 2015, Wang and Zhou 2016, Wang et al. 2016a, Shen et al.
2018). On the other hand, unlike public bike-sharing which requires users to rent and
return bicycles at xed docking stations, dockless bike-sharing provides stationless rental
services, making the bicycles more convenient and exible to use (Si et al. 2019). However,
without accurate bicycle usage prediction and eective scheduling, the exibility oered
by dockless bike-sharing would lead to the mismatch between bicycle supply and
demand, which may bring about some urban issues (Pan et al. 2019, Si et al. 2019). For
example, when the supply of bicycles exceeds the demand, the problems such as over-
whelming public space would arise. On the contrary, if the supply is less than demand, it
will result in service insuciency. Hence, understanding the inuence mechanisms of
dockless bike usage can provide a scientic basis for bicycle prediction and scheduling,
which is essential to improve the management and services of dockless bike-sharing.
Examining the relationship between dockless bike-sharing usage and built environ-
ment (BE) factors is of great signicance in many aspects including urban planning and
cycling facilities design, bike scheduling strategy, bike-sharing service promoting and
bike-sharing usage prediction and simulation (Shen et al. 2018). Though there is a large
body of previous studies trying to understand the eects of BE factors on bike-sharing
usage ((Buck and Buehler 2012, Kim et al. 2012, Fishman 2015, Faghih-Imani and Eluru
2016, Wang et al. 2016a, 2016b, El-Assi et al. 2017, Shen et al. 2018, Wang and Lindsey
2019), few studies have taken the modiable areal unit problem (MAUP), a well-known
problem in geography research (Openshaw 1984), into consideration in the process of
data aggregating and modeling. However, studies have illustrated that MAUP is an
essential and fundamental issue in travel behavior analysis (Zhang and Kukadia 2005,
Mitra and Buliung 2012, Hong et al. 2013; Clark and Scott 2014, Yang et al. 2019, Zhou and
Yeh 2020). Furthermore, none of the previous studies examined the interactive eects of
factors, instead, the individual eects of factors on bike usage are quantied via regres-
sion coecients (Noland et al. 2016, Faghih-Imani and Eluru 2016, Zhang et al. 2017, Shen
et al. 2018), which is insucient to understand the dockless bike-sharing usage in such
a complicated built environment.
The aim of this study is to ll the above two research gaps with the geographical
detector model which is a spatial statistical method (Wang et al. 2010) by taking
Shenzhen, China as a case study to address the following research questions:
(1)How does MAUP aect the bike-sharing usage mechanism modeling results? How
do the BE factors perform at dierent spatial scale units?
(2)How are the interactive eects of BE factors on dockless bike-sharing usage com-
pared with the separate eect with the ideal spatial unit?
To answer these questions, rst, the factor detector of geographical detector model
was used to analyze the spatial associations between dockless bike-sharing usage and
BE factors with dierent spatial areal units, and examine the eect of MAUP. The results
2F. GAO ET AL.
can help land-use planners understand what BE factors should be paid more attention
to at dierent spatial scales, and help determine the suitable spatial scale to better
understand the interactive eects of BE factors for dockless bike-sharing usage.
Second, the interactive detector from the geographical detector model was used to
explore the interactions between these BE factors with the ideal spatial areal units. The
interactive eects were further compared with the separated eects, which is mean-
ingful to bike rebalance strategy. The rest of this paper is organized as follows: section 2
will review the related works; section 3 will present the study data and introduce the
methodology and datasets; section 4 will present the results and address some mean-
ingful ndings; and section 5 will summarize the study and provide suggestions for
further research.
2. Related works
The factors inuencing the usage of bike-sharing are complicated. An increasing
number of studies have been undertaken to explore this issue from dierent aspects,
including the social-demographic (Ogilvie and Goodman 2012, Zhao et al. 2015,
Wang et al. 2016a), weather and calendar events (Gebhart and Noland 2014,
Corcoran et al. 2014, Meng et al. 2016), and built environment (BE) (Cervero and
Kockelman 1997, Buck and Buehler 2012, Kim et al. 2012, Faghih-Imani and Eluru
2015, 2016, Wang et al. 2016a, 2016b, El-Assi et al. 2017, Shen et al. 2018, Wang and
Lindsey 2019). Among these, the BE impact on travel behavior has become the most
heavily researched subject in urban planning and travel behavior research (Buck and
Buehler 2012, Kim et al. 2012, Faghih-Imani and Eluru 2015, Wang and Zhou 2016,
Wang et al. 2016a, 2016b, El-Assi et al. 2017, Shen et al. 2018, Xu et al. 2019), and
many researchers have attempted to provide explanations about why BE factors
might be expected to impact travel behaviors (Cervero and Kockelman 1997, Wang
and Zhou 2016, Shen et al. 2018). These studies indicated that BE characteristics are
strongly associated with bike-sharing usage, and the inuence mechanisms are
complex, which need more attention and research eorts.
The measurement of the usage of bike-sharing and the models applied vary in
existing studies due to dierent research purposes. First, in terms of measurement of
bike usage, bike usage data used for analysis were measured at dierent geographic
scales which were usually dened by dierent ways in dierent studies without MAUP
addressed. Studies on the inuencing factors of bike-sharing usage fall into two main
categories: bike-sharing with xed stations and free-oating dockless bike-sharing. The
notable dierence between them when modelling their relationships with potential
factors is the denition of the spatial statistical unit of variables. For public bike-sharing
studies, the dependent variables were counted at docking stations and the independent
variables were linked to the stations’ service area dened by Thiessen polygon (Noland
et al. 2016) or buers (Rixey 2012, Wang and Lindsey 2019). For dockless bike-sharing
studies, data were usually aggregated to shnet cell, and the spatial areal units were of
dierent sizes among dierent studies (Shen et al. 2018, Mooney et al. 2018, Xu et al.
2019, Zhu et al. 2020). However, the MAUP received little attention in existing bike-
sharing-related research.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 3
Previous studies have shown that dierent areal units may generate inconsistent
results in the relationship between BE and travel behavior (Zhang and Kukadia 2005,
Mitra and Buliung 2012, Hong et al. 2013, Clark and Scott 2014, Yang et al. 2019, Zhou
and Yeh 2020). For example, Mitra and Buliung (2012) found that both the spatial scale
and zoning method aect the relationship between BE and active school transporta-
tion by comparing the results of buers of four distances and two types of census
boundaries. Clark and Scott (2014) proved that the relationship between active travel
and the BE is aected by the MAUP by comparing the models results of 14 geogra-
phical scales. Yang et al. (2019) found that the relationship between trip-chaining
behavior and the BE is dierent with several dierent spatial units. For bike-related
study, with massive and high-precision bike-sharing usage GPS record data, the eects
of the MAUP are nonnegligible in the results of data aggregating and modeling. It is
essential to examine the eects of the MAUP in BE-bike-sharing usage relationship
which can inform urban planners what factors are more important to promote bike-
sharing usages at certain spatial scales. Moreover, the MAUP should be considered and
carefully addressed to determine the ideal spatial units for explore the inuencing
mechanisms of bike-sharing usage.
Second, in terms of methods in modeling the relationship between bike usage
and BE factors, previous studies focused on the individual eect of each factor on
bike-sharing usage based on the regression coecients but neglected the interactive
eect. In these existing studies, non-spatial statistical methods (Buck and Buehler
2012, Kim et al. 2012, Faghih-Imani and Eluru 2015, Wang et al. 2016b, El-Assi et al.
2017), and spatial regression models (Noland et al. 2016, Faghih-Imani and Eluru
2016, Zhang et al. 2017, Shen et al. 2018) were used. The results have enriched the
understanding of how BE factors aect bike-sharing usage. However, these studies
mainly discussed the impact of the single BE factor on bike-sharing usage, the
interactive eect between BE factors have not examined. As the eects of BE on
dockless bike-sharing usage are complicated and should not be explained by each
factor separately and simply, examining the interactive eects between BE factors on
bike usage might help understand that mechanism closer to the actual situation and
mining the meaningful spatiotemporal characteristics hidden behind.
To better understand the relationship between dockless bike-sharing usage and
BE factors, this study tried to address the MAUP issue and examine the interactive
eects by employing the Geographical detector. Geographical detector is a spatial
statistical method that can eectively explore both the individual inuence and
interactive eect of geographical factors based on spatial variances analysis (Wang
et al. 2010). This method has been widely used in geographical variation studies,
such as the health risk assessment (Liao et al. 2010, Wang and Hu 2012, Ding et al.
2019, He et al. 2019), the risk assessment of the Wenchuan earthquake in China (Hu
et al. 2011), the inuencing mechanism of planting patterns on uoroquinolone
residues (Li et al. 2013), as well as the driving forces and their interactions of built-
up land expansion (Ju et al. 2016) and the relationship between dissection density
and environmental factors (Luo et al. 2016). However, it has rarely been applied in
travel behaviors or transportation studies for the exploration of the interactive
eects of BE factors on travel behaviors.
4F. GAO ET AL.
3. Data and methodology
3.1. Study area
Shenzhen (Figure 1) is located in southern China with the area of 1997.47 square kilo-
meters. As one of the largest cities in China, this city is a link and bridge connecting
Hong Kong and the Chinese mainland. As a highly urbanized and modern metropolis,
Shenzhen’s public transportation system has gradually developed and matured with 8
metro lines and 854 bus lines operating in 2018. Besides, Shenzhen is one of the earliest
cities in China to put dockless bike-sharing into operation.
Dockless bike-sharing emerged in Shenzhen in early 2016, and subsequently this
venture experienced explosive growth. In September 2017, the average daily use of bike-
sharing in Shenzhen reached 5.173 million. The rapid expansion of dockless bike-sharing
has changed the daily traveling mode of residents, played a remarkable role in solving the
‘last kilometer’ travel problem, and has reduced trac congestion. Simultaneously, dock-
less bike-sharing has also created problems such as congestion caused by random
parking, spatial and temporal mismatches between supply and demand, and an excessive
presence of bicycles because of the vicious competition between operators. Thus,
Shenzhen is a good case study area for investigating the complex inuencing mechan-
isms of both the origin and destination involved with dockless bike-sharing usage from
a spatial perspective. It will provide valuable insight into the planning and strategic
management of sharing bikes.
3.2. Geographical detector model
The geographical detector is a kind of spatial statistic model proposed by Wang et al., which
has been widely used to quantify the inuencing eects of potential driving factors on
geographical phenomena based on spatial variance analysis (Wang et al. 2010, 2017, Liao
Figure 1. Study area.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 5
et al. 2010, Wang et al. 2016b). The geographical detector model is applied in this study based
on the assumption that the spatial distribution of the origin or destination of bike-sharing
resembles their potential driving factors. The geographical detector model consists of four
detectors (sub models) including factor detector, interactive detector, risk detector, and
ecological detector. Factor detector mainly addresses the question of ‘What are the determi-
nants of the geographical phenomena?’. Interactive detector addresses the question of ‘Do
the determinants operate individually or interconnectedly?’. Risk detector addresses the
question of ‘Are the geographical phenomena of two sub regions signicantly dierent?’.
Ecological detector addresses the question of ‘What is the dierence of the impacts between
two explanatory variables?’. As the main purpose of the study is to understand the determi-
nants of the dockless bike-sharing usage and the interactive eects as well as the MAUP in it,
factor detector and interactive detector were employed to examine which factor has a more
important impact on the use of bike-sharing and how dierent the pairs of factors interact
with each other.
3.2.1. The factor detector
The function of factor detector is to calculate the PD (power determinant) value to
quantitatively assess the impact of potential driving factors on the spatial pattern of the
origin or destination of dockless bike-sharing. In this study, PD value is dened as the
dierence between one and the ratio of accumulated dispersion variance of the origin
or destination of bike-sharing over each sub region to that of over the entire study
region:
PD ¼1PL
h¼1Nhσ2
h
Nσ2¼1SSW
SST (1)
ssw ¼X
L
h¼1
Nhσ2
h(2)
SST ¼Nσ2(3)
where N refers that a study area consists of N units, which is stratied into h = 1, 2, . . .,
L stratum; and stratum h consists of Nh units; σ2 and σ2
h denote the global variance of the
dependent variable of the study area and the variance of the dependent variable in the
sub-areas; SSW and SST denote within sum of squares and total sum of squares, respec-
tively. The value of PD lies between zero and one. A higher PD value means the driving
factor has a stronger contribution to the spatial pattern of the origin or destination of
bike-sharing. In this study, PD values indicate the consistency of the spatial patterns
between the origin or destination of bike-sharing and its potential driving factors.
3.2.2. The interaction detector
The interaction detector determines whether two individual factors enhance or weaken
each other by comparing their combined contribution, as well as their independent
contributions(Wang et al. 2010). The model classies the interactive relationship between
two factors into seven types as follows:
6F. GAO ET AL.
Nonlinear enhance :PD A \B
�>PD Að Þ þ PD Bð Þð Þ
Independent :PD A \B
�¼PD Að Þ þ PD Bð Þð Þ
Bi enhance :Max PD Að Þ;PD Bð Þð Þ<PD A \B
�<ðPD Að Þ þ PD Bð Þ
Uni enhance=weaken :Min PD Að Þ;PD Bð Þð Þ<PD A \B
�<Max PD Að Þ;PD Bð Þð Þ
Nonlinear weaken :PD A \B
�<Min PD Að Þ;PD Bð Þð Þ
(4)
3.2.3. The MAUP test
For dockless bike-sharing without xed stations, a grid system is suitable for statistics of
bike-sharing usage. However, as the scale of grid changes, the results of bike-sharing
usage mechanism modeling dier greatly. This is regarded as the scale eect, one of the
modiable areal unit problems (MAUP) that generally exists in geographical studies
(Jelinski and Wu. 1996, Zhou et al. 2018, Zhou and Yeh 2020). Another MAUP beyond
that is the zoning eect, in which various conclusions might occur when rearranging the
zones of the given set of areal units using dierent methods (Jelinski and Wu. 1996, Ju
et al. 2016). To understand how do BE factors aect the dockless bike-sharing usage with
dierent spatial scale units and zoning methods, both scale eect and zoning eect are
tested to examine the MAUP before the geographical detector model is applied in this
work.
First, the scale eect is tested for two main purposes: 1) to examine how do the BE
factors perform at dierent spatial scale units, which would inform urban planners what
BE factors should be paid more attention to at dierent spatial scales; 2) to determine the
suitable spatial scale to better understand the inuencing mechanisms of dockless bike-
sharing usage for bike rebalance strategy. The range of PD values and the stability of their
ranks can reect the scale eect on the results of the geographical detector model. In
consideration of the extent of the study area and the spatial resolution of the multi-source
data, ten grid sizes (from 100 m to 1000 m, with an interval of 100 m) are selected to test
the scale eect on the PD values and their ranks. Second, the zoning eect of the
geographical detector is tested to help choose the zoning method for bike-sharing
usage mechanism modeling. In order to test the zoning eect, these three commonly
used zoning methods are selected: the natural breaks (NB) method (Brewer and Pickle
2002), the equal interval (EI) method, and the quantile (QU) method (Cao et al. 2013).
3.3. Datasets and variables
We obtained the real-time GPS data from dockless bike-sharing scheme operators includ-
ing Mobike, Ofo, Bluegogo, Ubike and Xiaoming Bike. All of them were the major bike-
sharing operators in Shenzhen in 2018. The GPS dataset used in this study ranges from the
8
th
of October, 2018 to the 14
th
of October, 2018 and it consists of ve weekdays and
a two-day weekend. The bikes’ unique ID, time stamps and the GPS location of both the
origin and destination (OD) of every trip were continually recorded then categorized
by hour. The raw collection contains over 6 million riding trips, with some redundant
records however. To sort out the real usage records, some necessary pre-processing steps
are needed. First, redundant coordinate records of stationary bikes and errors from GPS
drifting were removed. Next, some unrealistically long or short distance or duration trips
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 7
were also excluded, which might be from bicycle reallocation and maintenance per-
formed by the operators.
Large amount of literatures explore the association between travel behavior and built
environment (Ewing and Cervero 2001, 2010). Based on previous studies on bike-sharing
(Buck and Buehler 2012, Kim et al. 2012, Faghih-Imani and Eluru 2015, Wang and Zhou
2016, Wang et al. 2016a, El-Assi et al. 2017, Shen et al. 2018, Li et al. 2020a), four categories
of potential inuencing factors were selected to represent the built environment from
dierent aspects including accessibility, facilities and land use as well as population
distribution. These four types of factors contribute to the bike-sharing usage, while
simultaneously interacting with each other (Figure 2).
According to the Athens Charter, living, work, recreation, and transportation are four
important functions of a city. In this study, accessibility factors were selected to represent
the transportation function of city with three potential factors including roads density,
distance to metro stations and distance to bus stops. Moreover, facilities factors were used
Figure 2. The selected BE variables of bike-sharing usage.
8F. GAO ET AL.
to reect the city functions of living, work and recreation, which are also the urban
residents’ daily commuting purposes. Facilities factors including WORK, LIVING and REC
were extracted from POIs datasets, and the denition of them was shown as Table 1
below. However, land use and activities of a POI can be complicated and mixed. Here,
from the perspective of travel behavior, POIs were classied as Table 1 shown for the
reason of visit quantity and purpose. For example, restaurants and retailing stores were
dened as the category of REC rather than WORK since that rstly these facilities are
typical work places but in terms of quantity, consumers are the groups with much more
visits, rather than employees. Besides, land use mixture was also measured through POIs
data based on POIs types in Table 1. The Shannon entropy index was applied to evaluate
the degree of land use diversity (Shannon 1948) as
H¼ Xipilognpi(5)
where H represents the value of entropy; pirepresents the percentage of the ith type of
POIs; n is the number of types.
Given the literature review and data availability, eight variables were selected as the
potential BE-related driving factors of bike-sharing usage. Since the cycling path is still
under construction in Shenzhen until November 2019, the variable of cycling facilities was
not incorporated in this study.
In 2018, for this study, ve types of data were obtained: (1) Road data were originally
collected from OpenStreetMap (https://www.openstreetmap.org/). After careful examina-
tion, the bike-rideable roads were extracted. (2) Metro stations and bus stops location
data were obtained using the geocoding interface of Baidu Map API (http://lbsyun.baidu.
com/). These two distance variables were calculated via Network Analyst of ArcGIS 10.2
(Esri, Redlands, California, US). (3) The three facilities variables were extracted from a POIs
dataset in 2018 from Baidu Map (http://lbsyun.baidu.com/). Also, the land use variable
was also calculated based on POIs dataset. (4) The dynamic population distribution data
were collected from the Tencent LBS service platform (https://heat.qq.com/document.
php) which is the largest social media service in China, with a spatial resolution of 25 m
and a temporal resolution of one hour, ranging from the 8th of October, 2018 to the 14th
of October, 2018. With its advantages of substantial mass of users records and high spatial
and temporal resolution, it has been involved in various spatial studies to reect the
dynamic population (Chen et al. 2017, 2018, Niu et al. 2017, Yao et al. 2017, Song et al.
2019, Li et al. 2020b, 2020c). The population distribution data in the same period as the
dependent variable (dockless bike-sharing usages) was used to represent the POP
variable.
Table 1. The definition of POIs category and type.
Category Type Examples
WORK Company and office Companies, offices, industrial zone, business center, science park
Government organizations Department offices, post offices, police offices
LIVING Residential Residential quarters, Unit dormitories
Hotels Hotels, lodgings
REC Restaurant Restaurants, Cafes
Retail Shopping malls, supermarkets, book stores, department stores
Recreation Parks, KTVs, movie theaters
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 9
4. Results and discussion
4.1. How does MAUP aect the bike-sharing usage mechanism modeling results?
To examine the eect of the MAUP on the relationship between BE factors and the
dockless bike-sharing usage, both the most suitable and meaningful scale and the
information hidden behind the changes of model results with dierent spatial scales
and zoning method should be carefully discussed. In the study, the number of classes of
each factor was set at ve for both scale eect and zoning eect. Both scale eect and
zoning eect were tested with ten candidate grid scales ranging from 100 m to 1000 m
(Figure 3). Figure 4(a,b) show the PD values of each factor and the rank of them. It can be
obviously seen from Figure 4(a) that the PD values of all factors tend to increase with
increasing grid size, which is consistent with the result of Ju et al. (2016)’s study on driving
force of built-up land expansion with use of geographical detector. However, the relative
importance of factors should be discussed by the ranks of them, which is nonnegligible
(Ju et al. 2016).
For scale eect, rst, some interesting ndings are generated from the ranking results
of factors with dierent grid scales (Figure 4(b)), which could provide valuable insights
into urban planners who aim to promote bicycle usage. These ndings can be summar-
ized as follows: 1) The ranks of dierent factors show dierent relationships at dierent
grid scales, which indicates that dierent grid scales generate inconsistent results in the
inuence of these BE factors on dockless bike-sharing usage. Among these factors, BUS
Figure 3. Distribution of bike-sharing and its aggregation with ten candidate grid scales.
10 F. GAO ET AL.
and ROAD are stable factors with least changes in the ranks of PD values. This indicates
that the inuence of these two factors on dockless bike-sharing usage are less sensitive to
grid size. However, the ranks of other factors’ PD values are more sensitive to the grid
scales, including METRO, WORK, LIVING, REC, MIX and POP. It is suggested that the
planners should pay attention to the spatial scale in the planning of these scale-
sensitive factors. 2) For the variables regarding to facilities, such as LIVING, WORK and
REC, they are less inuential to bike-sharing usages as the scale becomes smaller,
especially when the grid is smaller than 600 m. This implies that at a small grid scale
less than 600 m, increasing the density of these facilities does not necessarily lead to
signicant increase in dockless bike-sharing usage. 3) With regard to MIX variable, the
Figure 4. Scale effect on the results of geographical detector (PD values and the ranks of factors).
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 11
relative importance decreases with the increase of grid scale, when the scale is less than
600 m. It hints that the land use mixture planning at a smaller spatial unit is more
meaningful to promote dockless bike-sharing use. 4) For POP and METRO factors, the
relative importance decreases with increasing grid size, when the scale is more than
600 m. This suggests urban planners to pay much attention on the planning of these two
factors at grid size less than 600 m.
Second, the scale eect was also tested to help choose the appropriate spatial scale for
the inuencing mechanisms analysis in the next sections. Taking both PD values and their
ranks into consideration and comparison, 600 m was determined as the spatial scale to
analyze the individual and interactive factors of dockless bike-sharing usage which is
helpful for bike rebalance strategy for three reasons. Firstly, the growth rate of PD values is
relatively high with grid scale smaller than 600 m, while it begins to slow down with grid
scale larger than 600 m (Figure 4(a)). Secondly, the ranks of PD values experience great
changes with grid scale smaller than 600 m, and then tend to be relatively stable when
scale is smaller than 600 m (Figure 4(b)). Thirdly, as a type of human mobility behaviors,
bike-sharing usage and its mechanism study should be based on a possibly ne scale level
to better characterize the built environment accurately, while oversized grid scales might
hide some realistic spatial heterogeneity of dockless bike-sharing usage.
For zoning eect, the PD values dier as classication method changes. Table 2 shows
dierent PD values with various kinds of zoning methods, and the results of zoning eect
that the natural break method is the optimal zoning method with highest PD values,
followed by equal break and quantile method. The geographical detector model is based
on the spatial variance analysis, and the natural break is the method designed to dene
the optimal arrangement of values into dierent intervals by minimizing each interval’s
average deviation within class and maximizing it between classes. Arbitrary zoning
methods might mislead the actual relationship between geographical phenomena and
its inuencing factors (Hu et al. 2011). Previous studies noted that various methods can be
used to classify numerical variables into type variables in the data processing of the
geographical detector, and the criteria to select the optimal zoning method are the PD
values of the results (Wang et al. 2010). Hence, the natural break method was selected as
the zoning methods in the following analysis.
Table 2. The zoning effect of the geographical detector.
Category Variable Range Cutting values Method PD value
Accessibility Metro [10.19, 43.6] 5.1, 11.3, 19.1, 29.5, 43.6 NB 0.2933
8.9, 17.6, 26.3, 34.9, 43.6 EI 0.2350
1.9, 4.8, 9.2, 16.5, 43.6 QU 0.2445
Facilities Work [0, 768] 33, 101, 198, 383, 768 NB 0.2342
153.6, 307.2, 460.8, 614.4, 768 EI 0.2101
0, 8, 42, 98, 768 QU 0.2270
Land use Mix [0, 1] 0.17, 0.50, 0.68, 0.84, 1 NB 0.1178
0.2, 0.4, 0.6, 0.8, 1 EI 0.1013
0, 0.58, 0.73, 0.84, 1 QU 0.1003
Population Pop [0, 239.3] 14.2, 38.4, 70, 116.1, 239.3 NB 0.3143
47.9, 95.7, 143.6, 191.4, 239.3 EI 0.2151
1.0, 9.9, 27.5, 55.8, 239.3 QU 0.2259
12 F. GAO ET AL.
4.2. How are the interactive eects of BE factors on dockless bike-sharing usage
compared with the single eect?
The factor detector was conducted to examine the relative importance (single eect) of the
BE factors to dockless bike-sharing usage. Figure 5 shows and compares the PD values of
origin and destination of dierent periods on weekdays and the weekend, from which we
get some ndings as following. First, regarding facility variables, there are signicant
dierences in their PD values for dierent periods. On weekdays, the PD value of LIVING
on origin is higher than that on destination at morning peak and noon period, while it is
higher on destination at evening peak and night period. Dierently, on weekend, the PD
Figure 5. The PD values of factors of different period on weekday and weekend.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 13
value of LIVING is higher on origin for all the periods except night time. This indicates that
on workdays, the commuting travel time of residents is relatively regular. On weekend,
residents tend to be more available and exible to start to use bike-sharing near living
facilities before night time. These results match human daily activities and travel behaviors,
which are consistent with our common knowledge. Second, in terms of the three accessi-
bility variables (METRO, BUS, ROAD), METRO shows the strongest impact on the usages of
dockless bikes including both origin and destination, as expected. It hints that transfer with
metro is the main transfer mode of dockless bike-sharing usage in Shenzhen, rather than
transfer with bus. This corresponds well with previous case studies in New York (Faghih-
Imani and Eluru 2015), Seoul (Kim et al. 2012) and Saint Paul, Minnesota (Wang et al. 2016a),
which conrms a commonly existing phenomenon in many metropolises that the accessi-
bility of metro stations is positively correlated with bike-sharing usage. However, like the
previous studies, it is dicult to provide further information about the inuence mechan-
isms of dockless bike-sharing usage just through the individual impact of METRO factor. This
suggests further research on the interactive eects of METRO and other facility variables.
The interaction detector was further applied in the study with a superior advantage of
quantifying the interactive inuence of factors on dockless bike-sharing usage. In this
study, 28 pairs of interactions were calculated between eight factors, and the top three
interactions of four periods on weekdays (Table 3) and weekend (Table 4) were presented.
The strongest three interactions for each period are mainly the METRO interacting with
facility factors which vary in terms of day of week, time of day and origin or destination.
However, it should be noted that we will not compare the dierences of the interaction
Table 3. The interactive detector results of factors on weekdays.
Days Time
O/
D Ranks Factors
Interactive
PD
Enhancement with
METRO compared
with the single
effect
Weekdays 8–9am O 1 REC ∩ METRO 0.424 70.93%
2 LIVING ∩ METRO 0.421 36.38%
3 POP ∩ LIVING 0.400 /
D 1 WORK ∩ METRO 0.344 84.48%
2 REC ∩ METRO 0.333 82.51%
3 POP ∩ METRO 0.322 35.13%
12–13pm O 1 REC ∩ METRO 0.469 70.02%
2 WORK ∩ METRO 0.468 77.85%
3 LIVING ∩ METRO 0.450 43.17%
D 1 REC ∩ METRO 0.431 74.30%
2 LIVING ∩ METRO 0.427 41.21%
3 WORK ∩ METRO 0.420 88.74%
18–19pm O 1 WORK ∩ METRO 0.427 72.13%
2 POP ∩ METRO 0.402 28.33%
3 REC ∩ METRO 0.400 78.72%
D 1 LIVING ∩ METRO 0.396 41.86%
2 REC ∩ METRO 0.386 82.10%
3 POP ∩ METRO 0.377 29.56%
21–22pm O 1 REC ∩ METRO 0.485 62.50%
2 LIVING ∩ METRO 0.449 39.42%
3 POP ∩ REC 0.447 /
D 1 LIVING ∩ METRO 0.441 33.44%
2 POP ∩ LIVING 0.427 /
3 REC ∩ METRO 0.420 70.67%
14 F. GAO ET AL.
PD values of the same interactive factors in dierent periods, as the fact of the usage as
well as the spatial characteristics of dockless bike-sharing usually vary with time. For
example, it can be found that the interactive PD value of ‘METRO∩WORK’ is higher for bike
destination (D) at morning peak time on weekend (0.372, in Table 4) than that on week-
days (0.344, in Table 3). This implies that 37.2% of the spatial distribution of bike destina-
tion at morning peak time on weekend is consistent with that of WORK factor interacting
with METRO, whilst 34.4% of that at morning peak time on weekend is consistent with
that of WORK factor interacting with METRO. However, we must realize that the bike
usage is much lower on weekend than weekdays, and the spatial distribution is also quite
dierent between dierent periods. Hence, we focus on what factors interact better in the
same period, from which we get some interesting ndings as following.
First, regarding morning-peak of weekdays, the largest interaction for bike origin is REC
interacting with METRO, followed by LIVING interacting with METRO. Although REC has
relative lower PD value than LIVING and METRO from the single factor detector results, the
interactive eect of METRO and REC factor get 70.93% enhancement compared with the
single eect. It was noted that the restaurant POI accounts for over 77% of the REC data we
used. This may imply that in addition to residential areas, the areas with high dense of
restaurants around metro stations are usually in high demand at morning peak on week-
days, which should not be neglected when relocating bikes. Additionally, though WORK
factor does not show very important inuence on bike destination from the single factor
detector results, the strongest interaction for bike destination is WORK interacting with
METRO, followed by REC interacting with METRO. This indicates that available bikes from
Table 4. The interactive detector results of factors on weekend.
Days Time O/D Ranks Factors
Interactive
PD
Enhancement
with METRO
compared with
the single effect
Weekend 8–9am O 1 REC ∩ METRO 0.392 68.19%
2 LIVING ∩ METRO 0.387 34.80%
3 POP ∩ LIVING 0.355 /
D 1 REC ∩ METRO 0.394 75.29%
2 LIVING ∩ METRO 0.379 46.47%
3 WORK ∩ METRO 0.372 103.48%
12–13pm O 1 REC ∩ METRO 0.505 68.38%
2 LIVING ∩ METRO 0.497 38.81%
3 POP ∩ LIVING 0.460 /
D 1 LIVING ∩ METRO 0.448 38.70%
2 REC ∩ METRO 0.446 70.53%
3 POP ∩ LIVING 0.405 /
18–19pm O 1 REC ∩ METRO 0.486 67.29%
2 LIVING ∩ METRO 0.458 42.82%
3 WORK ∩ METRO 0.444 92.36%
D 1 LIVING ∩ METRO 0.419 37.88%
2 REC ∩ METRO 0.415 71.52%
3 POP ∩ LIVING 0.391 /
21–22pm O 1 REC ∩ METRO 0.469 60.72%
2 LIVING ∩ METRO 0.442 37.30%
3 POP ∩ LIVING 0.424 /
D 1 LIVING ∩ METRO 0.446 38.44%
2 REC ∩ METRO 0.422 44.54%
3 POP ∩ LIVING 0.418 /
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 15
other places should be relocated to those metro stations with high dense of companies and
restaurants around before morning peak.
Second, in terms of the noon period of weekdays, over 40% of the spatial distribution of
bike usage (origin/destination) is consistent with that of three facility factors intersecting
with METRO. The interactive eect of REC and METRO as well as that of WORK and METRO
get greatly improved, although the individual eects of REC and WORK are much lower than
LIVING. This interactive result indicates that the areas with high dense of these three facilities
around metro stations should not be ignored for bike relocating at noon of weekdays.
Third, the top three interactions for bike origin (destination) at evening peak time of
weekdays is similar to those for bike destination (origin) at morning peak time. The bike
rebalance strategy for evening peak should be the opposite of that for morning peak. It
should be noted that we could not only focus on the restaurant POI of REC factors for
evening peak. Other retail, recreational POI also should be paid attention to when
relocating bikes at evening peak time.
Fourth, regarding the night time for weekdays, though the individual eect of REC is
lower than LIVING, it shows great interactive enhancement of REC with METRO both for
bike origin and destination. Additionally, ‘POP ∩ REC’ and ‘POP ∩ LIVING’ are the strong
interactions for bike origin and destination respectively. This suggests the operators to
pay more attention on high density areas of recreational facilities around metro stations,
high density of population and recreational facilities, high density of residential facilities
and population, as well as the metro station exits with high dense of recreational facilities
around, as these areas are usually in high demand for bike trips at night time.
Last, for weekend, ‘REC ∩ METRO’ and ‘LIVING ∩ METRO’ are two strongest interactions
for all the four periods. This indicates that available bikes should be relocated at the high-
density area of living or recreational facilities around metro for weekend. It can be found
that the interactive eect of WORK and METRO for bike destination at morning peak of
weekend enhances by over 100% compared with the single eect of WORK, although this
interactive PD value ranks third in this period. Similarly, WORK ∩ METRO is the third
strongest interaction for bike origin at evening peak of weekend, but the interactive PD
value is 92.36% higher than the individual eect of WORK. These ndings imply overtime
working phenomenon on weekend in Shenzhen. Moreover, the results also imply that the
people who work overtime on weekend still rely on metro and connecting dockless
bicycle when there is a metro station near the company. Hence, the need of dockless bike-
sharing usage for commuting should not be ignored at the peak time on weekend.
5. Conclusions and future work
This study employed the geographical detector model to examine the inuencing mechan-
isms of BE factors on the usage of dockless bike-sharing, so as to provide insights into land
use planning and dockless bike rebalance strategy. The major contributions and ndings
from this study can be summarized as follows: (1) To test the MAUP in BE-related bike-
sharing studies, the scale eect and zoning eect were conducted to explore the eect of
MAUP in the inuences of built environment factors on dockless bike-sharing usage. (2) This
study proves that the geographical detector is an eective method for examining the
interactive eects of built-environment factors on dockless bike-sharing travel. This method
can be employed to other BE-travel behavior relationship studies with careful consideration.
16 F. GAO ET AL.
Some interesting ndings were generated from the present study, which can provide
decision-support for urban planning and bike rebalance. First, the results of MAUP eect
revealed that the inuence of most BE variables, such as METRO, WORK, LIVING, REC, MIX
and POP, are sensitive to the spatial areal units. This suggests urban planners who aim to
promote dockless bike-sharing usages to pay more attentions on the spatial scale in the
planning of these built-environment factors. The inconsistent results in the relative
importance of these factors with dierent grid scales could inform urban planners what
built-environment factors should be paid more attention to at certain spatial scale.
Second, the comparisons between individual eect and interactive eect conrm the
importance of interacting eect for bike rebalance strategy, as the individual eect is not
sucient. The results revealed some interesting ndings, which have not been explored in
previous studies and have provided valuable insights into bike rebalance for such an
innovative and high-density metropolis in China.
The study proves that the MAUP does aect the dockless bike-sharing usage mechan-
ism modeling results. The key to determining the most suitable areal unit for data
aggregation and modeling with geographical detector model is to understand the
changes of trend and characteristics of the modeling results (level of PD value and its
relative stability) with dierent areal units. This study contributes to the existing literature
of travel behavior and human mobility analysis by applying the geographical detector
model in mechanism analysis with careful consideration of MAUP. A suitable and mean-
ingful spatial areal unit for analyzing the eects of built-environment factors on dockless
bike-sharing usage in Shenzhen might be 600 m scale grid. Though it would vary in
dierent case study around the world, the procedure to select the suitable and mean-
ingful spatial areal unit scale in this study could provide scientic basis for related analysis.
Despite the merits of this study, we have to acknowledge some limitations which
remain to be addressed in future research. First, the BE factors that may aect dockless
bike usage for physical purposes have not been considered in this work. In the future, this
work should be extended by obtaining additional variables such as sports facilities, and
greenways, to reveal their relationships with dockless bike usage. Second, the inuence
mechanisms of dockless bike-sharing have been analyzed from a spatial perspective, and
bike scheduling suggestions have been put forward in this study. Future studies should
be further performed to build a direct forecasting model to provide quantitative details
for the dynamic spatial scheduling of dockless bike-sharing.
Data and codes availability statement
The model datasets and geographical detector model software that support the ndings of the
present study are available in gshare at https://doi.org/10.6084/m9.gshare.11438502.
Disclosure statement
No potential conict of interest was reported by the author(s).
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 17
Funding
This study was supported by Guangzhou Science and technology project [Grant No. 201904010198],
the National Natural Science Foundation of China [Grant No. 41871290, 41401432], Key Special
Project for Introduced Talents Team of Southern Marine Science and Engineering Guangdong
Laboratory (Guangzhou) [GML2019ZD0301].
Notes on contributors
Feng Gao is a master student at Guangzhou University and works in the area of geospatial big data
analysis and urban planning.
Shaoying Li is an associate professor at Guangzhou University and works in the area of geospatial
big data analysis and urban study, transportation and land use.
Zhangzhi Tan receives the Ph.D. degree in Sun Yat-Sen University and works in the area of
transportation study.Zhangzhi Tan receives the Ph.D. degree in Sun Yat-Sen University and works
in the area of transportation study.
Zhifeng Wu is a professor at Guangzhou University and works in the research area of urban human
settlements.
Xiaoming Zhang is the director of Transportation Planning and Design Department of Guangzhou
Urban Planning Survey and Design Institute, and works in the area of transportation development
strategic planning, comprehensive transportation planning and trac information model develop-
ment and application.
Guanping Huang is a master student at Guangzhou University and works in the area of big data
analysis and urban study.
Ziwei Huang is a master student at Guangzhou University and works in the area of big data analysis
and urban economic study.
ORCID
Feng Gao http://orcid.org/0000-0003-0398-4255
Shaoying Li http://orcid.org/0000-0002-4703-5660
References
Brewer, C.A. and Pickle, L., 2002. Evaluation of methods for classifying epidemiological data on
choropleth maps in series. Annals of the Association of American Geographers, 92 (4), 662–681.
doi:10.2307/1515294
Buck, D. and Buehler, R., 2012. Bike lanes and other determinants of capital bikeshare trips.
Washington, USA: Transportation Research Board Meeting.
Cao, F., et al., 2013. Optimal discretization for geographical detectors-based risk assessment.
Mapping Sciences & Remote Sensing, 50 (1), 78–92. doi:10.1080/15481603.2013.778562
Cervero, R. and Kockelman, K., 1997. Travel demand and the 3Ds: density, diversity and design.
Transportation Research Part D, 2 (3), 199–219. doi:10.1016/S1361-9209(97)00009-6
Chen, Y., et al., 2017. Delineating urban functional areas with building-level social media data:
a dynamic time warping (DTW) distance based k -medoids method. Landscape and Urban
Planning, 160 (Complete), 48–60. doi:10.1016/j.landurbplan.2016.12.001
Chen, Y., et al., 2018. Emerging social media data on measuring urban park use and their relationship
with surrounding areas—A case study of Shenzhen. Urban Forestry and Urban Greening, 31, 130–
141. doi:10.1016/j.ufug.2018.02.005
18 F. GAO ET AL.
Chen, Y.J., et al., 2019. Optimal pricing and availability strategy of a bike-sharing rm with
time-sensitive customers. Journal of Cleaner Production, 228 (10), 208–221. doi:10.1016/j.
jclepro.2019.04.151
Clark, A. and Scott, D., 2014. Understanding the impact of the modiable areal unit problem on the
relationship between active travel and the built environment. Urban Studies, 51 (2), 284–299.
doi:10.1177/0042098013489742
Corcoran, J., et al., 2014. Spatio-temporal patterns of a public bicycle sharing program: the eect of
weather and calendar events. Journal of Transport Geography, 41, 292–305. doi:10.1016/j.
jtrangeo.2014.09.003
Ding, Y.T., et al., 2019. Using geographical detector technique to explore the impact of socio-
economic factors on PM 2.5 concentrations in China. Journal of Cleaner Production, 221,
1480–1490. doi:10.1016/j.jclepro.2018.11.159
El-Assi, W., Salah Mahmoud, M., and Nurul Habib, K., 2017. Eects of built environment and weather
on bike sharing demand: a station level analysis of commercial bike sharing in Toronto.
Transportation, 44 (3), 589–613. doi:10.1007/s11116-015-9669-z
Ewing, R. and Cervero, R., 2001. Travel and built environment: a synthesis. Transportation Research
Record, 1780, 87–114. doi:10.3141/1780-10
Ewing, R. and Cervero, R., 2010. Travel and the built environment. Journal of the American Planning
Association, 76 (3), 265–294. doi:10.1080/01944361003766766
Faghih-Imani, A. and Eluru, N., 2015. Analyzing bicycle-sharing system user destination choice
preferences: Chicago’s Divvy system. Journal of Transport Geography, 44, 53–64. doi:10.1016/j.
jtrangeo.2015.03.005
Faghih-Imani, A. and Eluru, N., 2016. Incorporating the impact of spatio-temporal interactions on
bicycle sharing system demand: a case study of New York citibike system. Journal of Transport
Geography, 54, 218–227. doi:10.1016/j.jtrangeo.2016.06.008
Fishman, E., 2015. Bikeshare: a review of recent literature. Transport Reviews, 36 (1), 92–113.
doi:10.1080/01441647.2015.1033036
Gebhart, K. and Noland, R.B., 2014. The impact of weather conditions on bikeshare tirps in
Washington, DC. Transportation, 41 (6), 1205–1225. doi:10.1007/s11116-014-9540-7
He, J.H., et al., 2019. Exploring the regional dierences of ecosystem health and its driving factors in
China. The Science of the Total Environment, 673, 553–564. doi:10.1016/j.scitotenv.2019.03.465
Hong, J., Shen, Q., and Zhang, L., 2013. How do built-environment factors aect travel behavior? A
spatial analysis at dierent geographic scales. Transportation, 41 (3), 419–440. doi:doi:10.1007/
s11116-013-9462-9
Hu, Y., et al., 2011. Geographical detector-based risk assessment of the under-ve mortality in the
2008 Wenchuan Earthquake, China. PloS One, 6 (6), e21427. doi:10.1371/journal.pone.0021427
Jelinski, D.E. and Wu., J., 1996. The modiable areal unit problem and implications for landscape
ecology. Landscape Ecology, 11 (3), 129–140. doi:10.1007/BF02447512
Ju, H., et al., 2016. Driving forces and their interactions of built-up land expansion based on the
geographical detector–a case study of Beijing, China. International Journal of Geographical
Information Science, 30 (11), 2188–2207. doi:10.1080/13658816.2016.1165228
Kim, D.J., et al., 2012. Factors inuencing travel behaviors in bikesharing. Washington, USA:
Transportation Research Board Meeting.
Li, S.Y., et al., 2020b. The varying patterns of rail transit ridership and their relationships with ne-
scale built environment factors: big data analytics from Guangzhou. Cities, 99, 102580.
doi:10.1016/j.cities.2019.102580
Li, S.Y., et al., 2020c. Spatially varying impacts of built environment factors on rail transit ridership at
station level: a case study in Guangzhou. China. Journal of Transport Geography, 82. doi:10.1016/j.
jtrangeo.2019.102631
Li, W.W., et al., 2020a. Understanding intra-urban human mobility through an exploratory spatio-
temporal analysis of bike-sharing trajectories. International Journal of Geographical Information
Science. doi:10.1080/13658816.2020.1712401
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 19
Li, X., et al., 2013. Inuence of planting patterns on uoroquinolone residues in the soil of an
intensive vegetable cultivation area in northern China. Science of the Total Environment, 458-460C
(3), 63–69. doi:10.1016/j.scitotenv.2013.04.002
Liao, Y., et al., 2010. Spatial analysis of neural tube defects in a rural coal mining area. International
Journal of Environmental Health Research, 6 (20), 439–450. doi:10.1080/09603123.2011.568801
Luo, W., et al., 2016. Spatial association between dissection density and environmental factors over
the entire conterminous United States. Geophysical Research Letters, 43 (2), 692–700. doi:10.1002/
2015gl066941
Ma, X.W., et al., 2018. Understanding bikeshare model as a feeder to metro by isolating
metro-bikeshare transfers from smart card data. Transport Policy, 71, 57–69. doi:10.1016/j.
tranpol.2018.07.008
Meng, M., et al., 2016. Eect of weather conditions and weather forecast on cycling travel behavior
in Singapore. International Journal of Sustainable Transportation, 10 (9), 773–780. doi:10.1080/
15568318.2016.1149646
Mitra, R. and Buliung, R.N., 2012. Built environment correlates of active school transportation:
neighborhood and the modiable areal unit problem. Journal of Transport Geography, 20 (1),
51–61. doi:10.1016/j.jtrangeo.2011.07.009
Mooney, S., et al., 2018. Freedom from the station: spatial equity in access to dockless bike share.
Journal of Transport Geography, 74, 91–96. doi:10.1016/j.jtrangeo.2018.11.009
Niu, N., et al., 2017. Integrating multi-source big data to infer building functions. International Journal
of Geographical Information Science, 31 (9), 1871–1890. doi:10.1080/13658816.2017.1325489
Noland, R.B., et al., 2016. Bikeshare trip generation in New York City. Transportation Research Part A:
Policy and Practice, 94, 164–181. doi:10.1016/j.tra.2016.08.030
Ogilvie, F. and Goodman, A., 2012. Inequalities in usages of a public bicycle sharing scheme:
socio-demographic predictors of uptake and usages of the London (UK) cycle hire scheme.
Preventive Medicine, 55 (1), 40–45. doi:10.1016/j.ypmed.2012.05.002
Openshaw, S., 1984. The modiable areal unit problem. Norwick: Geo Abstracts University of East
Anglia.
Pan, L., et al., 2019. Deep reinforcement learning framework for rebalancing dockless bike sharing
systems. In: 33rd AAAI Conference on Articial Intelligence/31st Innovative Applications of
Articial Intelligence Conference/9th AAAI Symposium on Educational Advances in Articial
Intelligence, Assoc Advancement Articial Intelligence, Honolulu, Hawaii, USA, 1393–1400.
Rixey, R., 2012. Station-level forecasting of bikesharing ridership: station network eects in three
U. S. systems. Transportation research record. Journal of the Transportation Research Board.
doi:10.3141/2387-06
Shaheen, S., Guzman, S., and Zhang, H., 2010. Bikesharing in Europe, the Americas, and Asia: past,
present, and future. Transportation research record. Journal of the Transportation Research Board,
2143 (1), 159–167. doi:10.3141/2143-20
Shannon, CE., 1948. A mathematical theory of communication. The Bell System Technical Journal.
doi:10.1002/j.1538-7305.1948.tb00917.x.
Shen, Y., Zhang, X., and Zhao, J., 2018. Understanding the usages of dockless bike sharing in
Singapore. International Journal of Sustainable Transportation, 12 (9), 686–700. doi:10.1080/
15568318.2018.1429696
Si, H.Y., et al., 2019. Mapping the bike sharing research published from 2010 to 2018: a scient metric
review. Journal of Cleaner Production, 213, 415–427. doi: 10.1016/j.jclepro.2018.12.157.
Song, Y., et al., 2019. Dynamic assessment of PM2.5 exposure and health risk using remote sensing
and geo-spatial big data. Environmental Pollution, 253 (Complete), 288–296. doi:10.1016/j.
envpol.2019.06.057
Wang, D.G. and Zhou, M., 2016. The built environment and travel behavior in urban China:
a literature review. Transportation Research Part D: Transport and Environment, 52, 574–585.
doi:10.1016/j.trd.2016.10.031
Wang, J.F., et al., 2010. Geographical detectors-based health risk assessment and its application in
the neural tube defects study of the Heshun region, China. International Journal of Geographical
Information Science, 24 (1), 107–127. doi:10.1080/13658810802443457
20 F. GAO ET AL.
Wang, J.F. and Hu, Y., 2012. Environmental health risk detection with GeoDetector. Environmental
Modelling and Software, 33, 114–115. doi:10.1016/j.envsoft.2012.01.015
Wang, J.-F., Zhang, T.-L., and Fu, B.-J., 2016b. A measure of spatial stratied heterogeneity. Ecological
Indicators, 67, 250–256. doi:10.1016/j.ecolind.2016.02.052
Wang, J.Y. and Lindsey, G., 2019. Do new bike share stations increase member use: a
quasi-experimental study. Transportation Research Part A: Policy and Practice, 121, 1–11.
doi:10.1016/j.tra.2019.01.004
Wang, X., et al., 2016a. Modelling bike share station activity: eects of nearby businesses and jobs on
trips to and from stations. Journal of Urban Planning and Development, 142 (1), 04015001.
doi:10.1061/(ASCE)UP.1943-5444.0000273
Wang, Y., et al., 2017. Identifying the determinants of housing prices in China using spatial regres-
sion and the geographical detector technique. Applied Geography, 79, 26–36. doi:10.1016/j.
apgeog.2016.12.003
Xu, Y., et al., 2019. Unravel the landscape and pulses and cycling activities from a dockless
bike-sharing system. Computers, Environment and Urban Systems, 75, 184–203. doi:10.1016/j.
compenvurbsys.2019.02.002
Yang, L., Hu, L., and Wang, Z., 2019. The built environment and trip chaining behavior revisited: the
joint eects of the modiable areal unit problem and tour purpose. Urban Studies, 56 (4),
795–817. doi:10.1177/0042098017749188
Yao, Y., et al., 2017. Mapping ne-scale population distributions at the building level by integrating
multisource geospatial big data. International Journal of Geographical Information Science, 31 (6),
1220–1244. doi:10.1080/13658816.2017.1290252
Zhang, H., Shaheen, S.A., and Chen, X., 2014. Bicycle evolution in China: from the 1900s to the
present. International Journal of Sustainable Transportation, 8 (5), 317–335. doi:10.1080/
15568318.2012.699999
Zhang, M. and Kukadia, N., 2005. Metrics of urban form and the modiable areal unit problem.
Journal of Transport Geography, 1902, 71–79. doi:10.3141/1902-09
Zhang, Y., et al., 2017. Exploring the impact of built environment factors on the use of public bikes at
bike stations: case study in Zhongshan, China. Journal of Transport Geography, 58, 59–70.
doi:10.1016/j.jtrangeo.2016.11.014
Zhao, J., Wang, J., and Deng, W., 2015. Exploring bike sharing travel time and trip chain by gender
and day of week. Transportation Research Part C: Emerging Technologies, 58, 251–264. doi:10.1016/
j.trc.2015.01.030
Zhou, X. and Yeh, A.G.O., 2020. Understanding the modiable areal unit problem and identifying
appropriate spatial unit in jobs–housing balance and employment self-containment using big
data. Transportation. doi:10.1007/s11116-020-10094-z
Zhou, X., Yeh, A.G.O., and Yue, Y., 2018. Spatial variation of self-containment and jobs-housing
balance in Shenzhen using cellphone big data. Journal of Transport Geography, 68, 102–108.
doi:10.1016/j.jtrangeo.2017.12.006
Zhu, R., et al., 2020. Understanding spatiotemporal heterogeneity of bike-sharing and
scooter-sharing mobility. Computers, Environment and Urban System, 81, 101483. doi:10.1016/j.
compenvurbsys.2020.101483
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 21
