Standardised and transparent model descriptions for agent-based models: Current status and prospects

Full text

Standardised and transparent model descriptions for agent-based
models: Current status and prospects
Birgit Müller
a
,
*
, Stefano Balbi
b
, Carsten M. Buchmann
c
, Luís de Sousa
d
, Gunnar Dressler
a
,
Jürgen Groeneveld
a
,
e
, Christian J. Klassert
f
, Quang Bao Le
g
, James D.A. Millington
h
,
Henning Nolzen
a
, Dawn C. Parker
i
, J. Gary Polhill
j
, Maja Schlüter
k
, Jule Schulze
a
,
Nina Schwarz
c
, Zhanli Sun
l
, Patrick Taillandier
m
, Hanna Weise
a
a
UFZ, Helmholtz Centre for Environmental Research eUFZ, Department of Ecological Modelling, Permoser Str. 15, 04138 Leipzig, Germany
b
Basque Centre for Climate Change (BC3), Alameda Urquijo 4, 4

, 48008 Bilbao, Spain
c
UFZ, Helmholtz Centre for Environmental Research eUFZ, Department Computational Landscape Ecology, Permoser Str. 15, 04138 Leipzig, Germany
d
Resource Centre for Environmental Technologies, Public Research Centre Henri Tudor, Technoport Schlassgoart, 66 rue de Luxembourg, P.O. Box 144, L-
4002 Esch-sur-Alzette, Luxembourg
e
School of Environment, University of Auckland, Private Bag, 92019 Auckland, New Zealand
f
UFZ, Helmholtz Centre for Environmental Research eUFZ, Department of Bioenergy, Torgauer Str. 116, 04347 Leipzig, Germany
g
Natural and Social Science Interface (NSSI), Institute for Environmental Decisions (IED), ETH Zurich, Universitaetstrasse 22, CH-8092 Zurich, Switzerland
h
Department of Geography, King’s College London, Strand, London WC2R 2LS, UK
i
School of Planning, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada
j
The James Hutton Institute, Craigiebuckler, Aberdeen AB15 8QH, UK
k
Stockholm Resilience Centre, Stockholm University, Kräftriket 2B, 10691 Stockholm, Sweden
l
Leibniz Institute of Agricultural Development in Central and Eastern Europe (IAMO), Theodor-Lieser-Str. 2, 06120 Halle (Saale), Germany
m
Department of Geography, UMR IDEES, CNRS, University of Rouen, 7 rue Thomas Becket, 76130 Mont Saint Aignan, France
article info
Article history:
Received 3 April 2013
Received in revised form
20 January 2014
Accepted 21 January 2014
Available online
Keywords:
Agent-based modelling
Domain specific languages
Graphical representations
Model communication
Model comparison
Model development
Model design
Model replication
Standardised protocols
abstract
Agent-based models are helpful to investigate complex dynamics in coupled humanenatural systems.
However, model assessment, model comparison and replication are hampered to a large extent by a lack
of transparency and comprehensibility in model descriptions. In this article we address the question of
whether an ideal standard for describing models exists. We first suggest a classification for structuring
types of model descriptions. Secondly, we differentiate purposes for which model descriptions are
important. Thirdly, we review the types of model descriptions and evaluate each on their utility for the
purposes. Our evaluation finds that the choice of the appropriate model description type is purpose-
dependent and that no single description type alone can fulfil all requirements simultaneously. How-
ever, we suggest a minimum standard of model description for good modelling practice, namely the
provision of source code and an accessible natural language description, and argue for the development
of a common standard.
Ó2014 Elsevier Ltd. All rights reserved.
1. Introduction
Agent-based models are argued to be helpful to investigate
complex dynamics in coupled humanenatural systems (Hare and
Deadman, 2004; Liu et al., 2007; Balbi and Giupponi, 2010;
Filatova et al., 2013). However, the production of research using
agent-based modelling has not been as efficient as it could be up to
now. Reasons include that model assessment, replication, and
comparison are hampered to a large extent by a lack of
*Corresponding author. Tel.: þ49 341 235 1708; fax: þ49 341 235 1473.
E-mail addresses: birgit.mueller@ufz.de (B. Müller), stefano.balbi@bc3research.
org (S. Balbi), carsten.buchmann@ufz.de (C.M. Buchmann), luis.a.de.sousa@gmail.
com (L. de Sousa), gunnar.dressler@ufz.de (G. Dressler), juergen.groeneveld@ufz.
de (J. Groeneveld), christian.klassert@ufz.de (C.J. Klassert), quang.le@env.ethz.ch
(Q.B. Le), james.millington@kcl.ac.uk (J.D.A. Millington), henning.nolzen@ufz.de
(H. Nolzen), dcparker@uwaterloo.ca (D.C. Parker), gary.polhill@hutton.ac.uk (J.
G. Polhill), maja.schlueter@stockholmresilience.su.se (M. Schlüter), jule.schulze@
ufz.de (J. Schulze), nina.schwarz@ufz.de (N. Schwarz), sun@iamo.de (Z. Sun),
patrick.taillandier@univ-rouen.fr (P. Taillandier), hanna.weise@ufz.de (H. Weise).
Contents lists available at ScienceDirect
Environmental Modelling & Software
journal homepage: www.elsevier.com/locate/envsoft
1364-8152/$ esee front matter Ó2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.envsoft.2014.01.029
Environmental Modelling & Software 55 (2014) 156e163

transparency in model descriptions. Further, code developed for
one project is rarely reused for other projects, even for closely
related research. To overcome these problems, standardised model
description protocols, ontologies and graphical representations
have been created. The various model description types have been
developed to achieve different purposes, including facilitation of
in-depth model comprehension, assessment, replication, design
and communication.
In this contribution we address the question of whether an ideal
standard for describing agent-based models exists. We first present
a classification of the prevalent types of model descriptions and
give an overview of their different purposes. We then review
available model description types, evaluating each on its utility for
the different purposes. Finally, we discuss advantages of combining
these different types, suggest a minimum standard of model
description for good modelling practice and discuss future chal-
lenges. Note that we set the focus on providing an adequate
description of the model itself and not on the description of model
results. Appropriate documentation of the model results is beyond
the scope of this paper (but see “Transparent and comprehensive
ecological modeling (TRACE) documentation”in Schmolke et al.
(2010), pp. 482 which suggests a standard for all parts of the
modelling process).
The idea for this article came about at a workshop at the 6th
International Congress on Environmental Modelling and Software
(iEMSs) 2012 in Leipzig, Germany, and the article reflects the per-
spectives of the participants, who are members of the integrated
social and environmental modelling communities.
2. Current state of the art: different types of model
descriptions in use
We classify the prevalent types of model descriptions in three
categories: natural language descriptions, formal language de-
scriptions and graphics (cf. Fig. 1 for an overview). In the following
paragraphs the different description types are briefly outlined:
Natural language descriptions present models in everyday
language with or without a prescriptive structure. The prescriptive
approach divides the model description into categories, each
explaining a particular part of the model. One example of such an
approach is the ODD protocol (cf. Grimm et al., 2010 and its
extension to include a description of human decisions in ABMs,
ODD þDinMüller et al., 2013). ODD describes the model in a hi-
erarchical way using three main categories: Overview,Design con-
cepts and Details that are themselves subdivided into several
subcategories such as (in the case of design concepts) sensing or
interaction. ODD is being widely used for the description of ABMs
(for examples see Balbi et al., 2013; Caillault et al., 2013; Marohn
et al., 2013; Smajgl and Bohensky, 2013). In contrast, a non-
prescriptive natural language description puts no constraints
regarding content and form of the model description on the
author (see exemplary model descriptions in Becu et al., 2003;
Deadman et al., 2004). Furthermore, non-prescriptive de-
scriptions can also be used to present the source code in a more
intuitive way. Examples are literate programming (cf. Knuth, 1984),
documentation generators such as Doxygen or Javadoc that
assemble source code comments into a structured document, or, in
principle, any form of source code documentation that uses natural
language.
Formal languages describe models in an abstract and self-
consistent way with formal syntax and semantics that avoid am-
biguity. Model descriptions written in formal languages may
therefore be used to describe important aspects of a model spe-
cifically. Formal languages that we consider here include ontol-
ogies, source code, pseudo code and mathematical descriptions.
An ontology can be defined as “an explicit specification of a
conceptualization”(Gruber,1993, p. 199) that describes entities and
their structural interrelationships, often using a hierarchical cate-
gorisation. They specifically allow logical inferences to be drawn.
Various formal languages are available for writing ontologies e
OWL (Web Ontology Language) being currently the most popularly
used (Horrocks et al., 2003; Grau et al., 2008). OWL has been argued
to improve the transparency of formal descriptions of model
structure in comparison with source code, since the latter is focused
on programmer and compilation convenience rather than using
logics to reflect common-sense perceptions (Polhill and Gotts,
2009). One example of ontologies applied to agent-based model-
ling is that of Christley et al. (2004). A second example is the MR
POTATOHEAD ontology developed by Parker et al. (2008), which
describes the components that appear in agent-based models of
land use/cover change. It identifies key model elements and their
alternative instantiations, based on a broad review of models. MR
POTATOHEAD has an OWL implementation which facilitates eval-
uating conceptual completeness.
Providing source code is another formal way to communicate
models. The following subcategories are listed according to their
readability, from cryptic to simple-to-read. Low-level program-
ming languages (e.g. assembly language) are characterized by their
strong linkage to the computer’s hardware and are often platform-
dependent. Though unlikely to be used for an entire ABM imple-
mentation, these can be useful for computationally intensive
functions where bespoke code improves on compiler optimisation.
Assembly language is necessary where higher-level programming
language libraries are not available for specialised hardware oper-
ations. For example, it is common in Linux distributions not to
provide C libraries for accessing floating point arithmetic utilities
stipulated by the IEEE 754 (1985) standard (IEEE, 1985). Polhill and
Izquierdo (2005, footnote 2) note that implementation of these
utilities using assembly language is necessary in a Cygwin
environment.
1
High-level programming languages in their basic form are
platform-independent (especially where governed by standards)
and improve the readability for the user by providing algorithmic
constructs such as loops or conditional statements. Popular ex-
amples of high-level programming languages are Java and Cþþ.In
addition, program-level tools extend the functionality of high-
level programming languages by “providing useful software li-
braries for building specific classes of models”(de Sousa and da
Silva, 2011, p. 170) and can further improve the readability of the
source code. Usually they are tailored to specificfields of modelling.
Fig. 1. Classification for structuring the prevalent types of model descriptions.
1
The utilities they implemented for this purpose are now available at https://
github.com/garypolhill/ieeefp.
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163 157

They can be further distinguished into tools that provide a domain-
specific language (e.g. NetLogo (Tisue and Wilensky, 2004)or
GAMA (Taillandier et al., 2012) for agent-based modelling) and
tools where the user has to write the application in a high-level
programming language (e.g. Repast (North et al., 2007) or Mason
(Luke et al., 2005)). In any case, provision of source code has been
argued to be a necessary condition of maintaining good scientific
practice in the publication of simulation models (e.g. Polhill and
Edmonds, 2007).
Pseudo code is a structured description of the model combining
natural language elements with formal language constructs (e.g.
loops). Gilberg and Forouzan (2004), p. xii, define pseudo code as
“[natural language]-like presentation of the steps needed to solve a
problem. It is written with a relaxed syntax .that hides the detail
[allowing the reader] to concentrate on the problem requirements.”
While such definitions can be given, to the knowledge of the au-
thors no common standard on formulating pseudo code exists.
Indeed, the provision of such a standard has been argued against on
the basis that it would then become another programming lan-
guage (Wikipedia, 2013), though there are stylistic conventions
(e.g. Smed and Hakonen, 2006), especially for operators and control
statements. The advantage of pseudo code is that it is independent
of the programming language and therefore the knowledge of a
specific programing language is not required to read and under-
stand the code (for examples cf. Roy, 2006; Perez and Dragicevic,
2010,p.227;Robinson et al., 2013, p. 134).
Mathematical descriptions provide an exact way to depict
model processes and states, usually with formulated equations
composed of strings of symbols. While being suited to describe
quantitative properties of the model, they are not able to commu-
nicate specific model concepts, such as underlying theories or
process scheduling. Mathematical descriptions can range from
general descriptions of model states (see equation (1) for the
calculation of the willingness to pay (WTP)) to specific equations
(see equation (2), cf. Filatova et al., 2009, section 3.12).
WTP ¼fðutility;incomeÞ(1)
WTP ¼utility$income
b
2
$utility
2
;with b¼constant factor (2)
Graphics use particular visualization techniques to illustrate
processes, structures, relationships, program flows, etc. They partic-
ularly support the understanding of qualitative properties of the
model, such as its structure. Graphics can be either formaldstrictly
adhering to pre-defined rules or protocols, such as UML (Unified
Modelling Language, cf. Object Management Group, 2011 and ex-
amples for class diagrams, activity diagrams and sequence diagrams
in Polhill et al., 2013), or non-formaldfollowing loosely-defined
principles or conventions, such as flow charts (cf. International
Organization for Standardization (ISO), 1985, examples in van Oel
et al., 2010; Zhang et al., 2011), Bayesian belief networks (Jensen,
2001; Aalders, 2008; Sun and Müller, 2013), decision trees
(Quinlan, 1986), cognitive maps (Eden, 1988; Kitchin, 1994)orcausal
loop diagrams (Maruyama, 1992). Non-formal graphics are often
used to get a first impression of the model concept (for examples see
Haase et al., 2010; Rebaudo and Dangles, 2013).
It should be noted that the different description types outlined
above are not necessarily mutually exclusive. For instance UML, as
part of the graphics category, is also an ontology. At the same time,
the MR POTATOHEAD ontology can be visualized using various
graphical approaches. However from our point of view a separate
category “graphics”underpins the importance of visualisation ap-
proaches for instance for the communication with stakeholders or
scientists from other disciplines.
3. Does an ideal standard for describing models exist?
3.1. Different purposes of model descriptions
Model descriptions can enable their users to meet various
different but related purposes as the models themselves (Kelly
et al., 2013). We propose the following as the most important
purposes: model communication, in-depth model comprehension,
model-assessment, -development, -replication, -comparison, the-
ory building and code generation (note some overlap occurs be-
tween purposes.). Here we briefly describe each of these purposes
to facilitate our review of model descriptions and the discussion of
an appropriate standard (the order chosen reflects the level of
generality, i.e. from general and universal to rather specific
purposes):
Communication of the model: Model descriptions serve as a
means of communication of the model to an audience that can
consist of scientists as well as stakeholders or people from outside
the research domain. These groups may need different information
(e.g. methodological details for specialists versus basic information
on the model’s entities and processes for stakeholders). They also
need to be approached in different ways, which poses a particular
communication challenge. One example is the interaction with
stakeholders which may benefit particularly from the use of easily
understandable visual tools. Apart from that: The communication
between model designer of the conceptual model and programmer
necessary for model implementation is a special but important case
of model communication.
In-depth model comprehension: This is the prime motivation
for model descriptions. The challenge is to allow a profound and
complete understanding of the model’s entities and processes.
Ideally, the reader can also relate real-world concepts to the model.
Model understanding is the precondition for most of the other
purposes. A detailed and thorough description of all model com-
ponents is thus essential. We emphasise here the degree of detailed
model understanding required by other scientists, experts and re-
viewers (in contrast to the purpose of communication which does
not necessarily imply overarching comprehension).
Model assessment: Assessment here means an evaluation of
the model with respect to its suitability to answer the research
question, the consistency between model design and sub-models,
as well as chosen spatial and temporal scales. This purpose is
particularly relevant for reviewers, but also researchers or policy
makers who want to use model results. To fulfil this purpose the
respective information (on research question(s), model purpose,
design and scales) must be given together with a clear statement of
the underlying model assumptions.
Model development: Following a prescriptive model descrip-
tion process can improve model development (‘model design’),
particularly when the model description is elaborated in parallel to
the model design. Ideally, describing the model helps the modeller
to adopt another (external) point of view of the model and can act
as a check list for completeness. To achieve this, a model descrip-
tion should follow a concise and strict structure which obliges the
author to describe all relevant aspects. A structured model
description can also facilitate and give guidance to the develop-
ment of models jointly with stakeholders (‘collaborative model
development’). In the case where the model description is expected
to assist model development up to the final implementation, a
rigorous protocol taking into account software and implementation
related issues is particularly helpful.
Model replication: Adequate model descriptions can enable
model replication. However, different levels of replication (see
Wilensky and Rand, 2007) may pose different requirements for the
model description. Exact quantitative replication of the results
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163158

(exactly the same numbers) requires much more detailed infor-
mation than statistical or qualitative replication. One extreme
example for such detailed information is the random seed used if
the model includes stochasticity. We use the term ‘qualitative’to
refer to replication that produces similar behaviour to the original
model and is robust to implementation details like random number
generators or the hardware used. Different metrics in model per-
formance assessment (see Bennett et al., 2013) can be used for
assessing behaviour similarity among the outputs of model repli-
cation. This usage is akin to the concept of ‘distributional equiva-
lence’in Wilensky and Rand (2007).
Model comparison: Model descriptions can facilitate model
comparisons with respect to concepts and techniques (for example
fitting or optimization algorithms). This can allow the reader to also
evaluate which model(s) are more or less appropriate for investi-
gating certain questions. To achieve this, a strict and complete set of
criteria (e.g. aim, scales, and processes) for comparison should be
part of the model descriptions, preferably in a standardised struc-
ture and format.
Theory building: By communicating the ideas behind a model, a
description can also aid in theory building. The challenge and
prerequisite here is to embed the respective model (concept) into
Table 1
Assessment on how well the purposes are met by the different description types (light grey elimited ability, middle grey emedium ability, dark grey ehigh ability, x not
applicable).
Natural
language
Formal
language
Graphics
Prescriptive
structure
Non-prescr.
structure
Ontologies
Source code
Pseudo code
Mathematical
descriptions
Formal
Non-formal
ODD
Verbal
descript.
OWL
MR POTA-
TOEHEAD
Low-level
progr. lan.
High-level
progr. lan.
Program-
level tools
UML
Exampl.a
Communication to
peers
Communication for
Communication for
education
stakeholders
In-depth
comprehension
Assessment
Development
design
Development
collaborative
Replication
Comparison
Theory
building
Code
generation xx
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163 159

the existing pool of theories and theoretical concepts. This can be
attained most easily when the description standard obliges the
author to summarize questions, applied theories, concepts, prin-
ciples and hypotheses. Such an evaluation of the described model
can reveal a lack of theoretical foundation for the model. Thereby,
model descriptions can support models in their role of challenging
existing theory (cf. Epstein, 2008). This, in turn, together with
evaluation of model results, can facilitate the creation and assess-
ment of new theoretical concepts and even new theory.
Code generation: Formal languages and graphic-based model
descriptions in particular can support model implementation
through (automated) source code generation. In case of model
reimplementation, code regeneration here implies a higher level of
formality, compared to most levels of model replication (see
above). Code (re)generation can be achieved if a description stan-
dard is formal, complete, and exact (not allowing any ambiguities),
while containing accurate information on entities and processes
plus their translation into code characteristics like classes and
methods. In this case, specialised software can then directly
generate the basic (code) structure from the model description
alone.
3.2. Matching purposes and types
In the following, we assess how well the purposes are met by
the different description types. For the assessment, we focus on the
potential of the description type rather than on how it is realised in
practice in our experience. The various description types fulfil the
purposes mentioned in the previous section to different degrees
(see Table 1 for an overview and Table A1 in the online appendix for
further details).
Communication of the model needs to be differentiated for
different target groups. (a) Communication to peers is achieved
with most of the description types, given the knowledge of the
respective description approach/programming language. (b)
Communication for education purposes is improved by e.g. natural
language descriptions, OWL, usage of program-level tools, pseudo
code and mathematical descriptions as well as non-formal
graphics. (c) Communication to stakeholders should not be too
technical, thus the suitable description types are limited to natural
language descriptions, OWL, and non-formal graphics. Non-formal
graphics are the only description type that can foster communi-
cation for all target groups, due to their large flexibility.
In-depth model comprehension that includes profound un-
derstanding of model entities and processes is fostered by natural
language descriptions, particularly with prescriptive structure
(such as the ODD protocol), but also by formal language de-
scriptions (i.e. OWL ontology, source code in a high-level pro-
gramming language and of program-level tools) as well as different
types of graphics. While being suited to describe quantitative
properties of the model, mathematical descriptions alone are not
able to communicate specific model concepts, such as underlying
theories or process scheduling.
Model assessment in general is facilitated by structured natural
language descriptions, ontologies and all types of graphics, i.e. UML,
and non-formal graphics such as cognitive maps, decision trees and
the like. Some specific types of model assessment esuch as
checking the consistency between model design and sub-models e
can more easily be carried out with ontologies or formal graphics.
Model development: (a) Model design by modellers and pro-
grammers is aided with prescriptive natural language descriptions
such as ODD, ontologies and usage of program-level tools, pseudo
code or mathematical descriptions; both formal and non-formal
graphics are also helpful. However, usage of non-formal graphics
and also program-level tools as a check list for the model design
process is limited, depending on the specific tool chosen. (b)
Collaborative development together with stakeholders on the
contrary is eased by non-prescriptive natural language descriptions
and formal and non-formal graphics.
(Quantitatively exact) model replication is difficult based on
mathematical descriptions without the provision of source code in
program-level toolsdequations alone do not guarantee replica-
tions (as discussed in Section 3.1). Usually detailed information of
the specific analytical or numerical procedures needs to be pro-
vided in form of script similar to high-level source code. However,
we want to highlight that although the provision of source code
technically facilitates model replication, it may circumvent the
consistency check between the conceptual model and its imple-
mentation (one purpose of model replication) by encouraging
“replicators”to simply copy the source code. In addition, for inde-
pendent model replication it is necessary to know the intention of
the modeller and therefore, we suggest that for independent model
replications, both source code and a natural language description
are provided. Qualitative model replication may be achieved also
with other model descriptions, such as MR POTATOHEAD, ODD,
UML and non-formal graphics.
Model comparison is made easier with prescriptive natural
language descriptions, ontologies, usage of program-level tools and
the provision of mathematical descriptions, if parameters are also
provided. Again, the specific focus of a model comparison will
indicate descriptions that are most suitable from this list (e.g.
comparison of the conceptual basis might be easier using pre-
scriptive natural language descriptions or ontologies rather than
mathematical descriptions).
Theory building is not well facilitated by model descriptions, as
most model description types do not ask for the theoretical back-
ground of the model, hypotheses to be tested, etc. One exception is
the prescriptive structure of the ODD þD protocol (Müller et al.,
2013), which asks for the theories underlying the model; while
theories are up to now not explicitly listed in the “basic principles”
section of the ODD protocol (Grimm et al., 2010). Another example
is presented in Schlüter et al. (2014). They develop a procedure to
document the theoretical background, the hypotheses and the as-
sumptions on which a model conceptualization is based in a
structured way. Furthermore, non-formal graphics are able to
convey information relevant for theory building and can thus
facilitate theory building without the constraints of formalised
graphics.
Code generation in the sense of automatically generating code
is enabled by formal graphics, such as UML (cf. Bersini, 2012)or
program-level tools facilitating generation of system models such
as SIMULINK or STELLA as well as OWL program-level tools. Apart
from that, pseudo code is often used to generate the structure of the
programme (e.g. Roy, 2006).
3.3. One size does not fit all
Our main conclusion from the analysis conducted above is that
the choice of a model description standard is purpose-dependent
and that no single model description type alone can fulfil all re-
quirements simultaneously. We have identified conflicting objec-
tives: a) to achieve a detailed model description that enables model
replication and b) to provide a concise and easy to communicate
model description. Furthermore, one should avoid making the
recommendations for model description more demanding than
necessary for its purpose. This is important for making the rec-
ommendations useful to a wide range of authors, which seems to
be a precondition for the establishment of a common standard of
model description.
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163160

Although the provision of information (especially source code)
is sometimes hindered by legal or other institutional reasons
(Polhill and Edmonds, 2007;Ince et al., 2012), we consider it
important to make the source code of a model available for three
reasons: firstly, it is the definitive implementation of the model, not
subject to ambiguities, omissions or inaccuracies associated with
verbal descriptions, secondly, because it provides the most direct
means of replicating model experiments, and thirdly, because it is
necessary to allow others to identify shortcomings in the imple-
mentation chosen by the author. The website www.openabm.org,
for example, provides an archive where model files can be uploaded
to share source code and/or model implementations.
Natural language description, especially when formalized in
standard protocols, such as ODD (Grimm et al., 2010 or its extension
focussing on human decision-making, ODD þD, Müller et al., 2013),
helps to make a connection between verbal descriptions of the real
world system underlying the model and the model itself. These
standard protocols can inform the scientific community whether
and how the model itself meets minimum scientific standards, and
what additional aspects or capabilities the model requires to meet
its specific purposes. In our view, the elements which should be
documented in the protocol in order to meet a minimum standard
of model description fall into two main categories: those that are
always needed to describe any system models (e.g. goal statement,
context/boundary setting conditions, unit and scale of analysis),
and those that are specific for an ABM, such as minimal charac-
teristics of the encoded agents (e.g. ‘heterogeneous’,‘autonomous’,
‘interactive’,‘reactive’and ‘adaptive’, cf. Benenson and Torrens,
2004), or system properties that ABMs are usually designed to
explain (e.g. ‘emergence’and ‘adaptation’, cf. Holland, 1995;
Bonabeau, 2002).
Standard protocols tell authors the information they need to
include in their model description, and they prime readers’ex-
pectations regarding what information they will find where. For
readers, this can facilitate the understanding of the assumptions
made in the implementation of a model. It further requires authors
to more fully open the “black box”of their model, potentially
revealing its weak areas and better contributing to scientific
progress. Therefore, we consider the use of a standard description
important.
In addition, graphical representations, such as UML diagrams,
can facilitate various purposes of model communication and un-
derstanding, and informal graphical representations are especially
beneficial for educational purposes and when working with
stakeholders. The close links between certain graphical represen-
tations and ontologies present a strong argument for the use of
ontologies in the model design phase (cf. Livet et al. (2010) for the
potential of ontologies for model building). However, the particular
mode of graphical representation to use is sensitive to the model in
question and to the intended audience. Hence, we do not recom-
mend it as a minimum standard for model description in journal
articles, but as an optional augmentation to the text and source
code.
4. Ways ahead
We recommend that researchers build on current examples of
good model descriptions, not only to describe their models trans-
parently, but also to strive for common standards in describing
ABMs, in order to contribute to comparability, model assessment
and replication, and theory development. However, many open
questions and challenges also remain, which need to be addressed
to improve model descriptions in the future.
Firstly, the standardisation of model descriptions is impeded by
the fact that it is extremely difficult to find a consistent terminology
across the many disciplines to which agent-based modelling is
applied (cf. Balbi and Giupponi, 2010). Researchers on multi- and
inter-disciplinary projects often report that differences in termi-
nology and vocabulary are an impediment to mutual understand-
ing (McConnell et al., 2011). A standardised description has the
potential to promote the use of a common terminology, through
suggesting and defining terminology such as agent and emergence
by a standardised description protocol, examples and guiding
questions therein.
Secondly, while there is a lot of common understanding about
the purposes of model descriptions, there are some aspects on
which we have found different perspectives. A major issue is
whether natural language description standards need to be
detailed enough to allow for replication, or whether such standards
should only facilitate understanding and communication, leaving
replication to the availability of the source code. On the one hand, if
we emphasize the view that natural language descriptions are
necessary to assess the consistency between the model and the real
world, then this might be an argument to make standardised de-
scriptions comprehensive enough to allow for replication. On the
other hand, the question arises of whether such comprehensive
descriptions might come at the cost of losing the readability of the
documentation when models are very large and/or complex. So-
lutions to this might come from hierarchical natural language
description such the ODD (starting from an overview and providing
details later) or distinguishing between different levels of replica-
tion (numerical, statistical, and qualitative) and from developing
large models over the course of several articles (Polhill et al., 2008;
Grimm et al., 2010). However, the debate has not yet come to a
conclusion.
In addition, although we have focused on model descriptions in
this article, there are similar challenges for the description of model
results. One attempt to address this issue is the use of narrative
approaches which, for a working model, can be useful to illustrate
characteristic (and specific) interactions between model agents and
explain how these interactions produce system-level dynamics
(Millington et al., 2012). Information about model outputs may be
relevant for theory building; for example, documentation about
hypotheses tested by the model and their results, or the results of
global sensitivity analysis.
Thirdly, there are institutional and cultural issues surrounding
the adoption and spread of standards. Should authorities promote
standards (e.g. by journals making them a publication requirement;
Polhill, 2010), or should they spread in an emergent process?
Another aspect is that the pressure for providing transparent model
descriptions might be greater if replicating a model to assess the
reliability of its results were a more common practice in the ABM
scientific arena. However, several institutional and cultural factors
impede such a development: journals do not insist on licenses that
enable software reuse, employers have an interest in protecting
intellectual property rights, there are no standard libraries for ABM,
and replication is not seen by everyone as innovative research.
Further, model replication is a resource-intensive undertaking, and
in an era of shrinking research budgets and university funding, it
may not be practical.
Finally, there have been several attempts to ease communica-
tion between modellers and facilitate reuse of models and model
components. Such reuse is seen as potentially decreasing start-up
costs and reducing barriers to entry to modelling, thus increasing
efficiency and speed of scientific progress in the field (Alessa et al.,
2006). Common platforms for ABMs and model-level tools have
been developed with these goals in mind. Contrary to program-
level tools, model-level tools allow “the usage of .simulation
models without requiring programming. These are pre-
programmed models, designed for specific application fields that
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163 161

can be parameterized by the user.”(de Sousa and da Silva, 2011,p.
170). Model-level tools can greatly facilitate communication since
each model can be described simply by its set of parameters and
inputs, using a single standard implementation. On the negative
side this sort of tools tends to be highly specialised, filling narrow
market niches; thus they are usually commercial tools whose in-
ternal implementation may not be open to independent scrutiny.
Although some platforms and program-level tools such as
Repast have become popular, there are still a plethora of different
platforms being used by the ABM developer community. The issue
of how to increase research efficiency in the field by helping to
facilitate code reuse remains an important one, but the large
number of platforms impedes developments in this direction to
some extent, and there seems to be no tendency towards agreeing
on one common platform that is used by everyone. Therefore, we
suggest it is worth working on a platform-independent standard
for model description, especially as such a standard should remain
relevant and useful even if common code bases are adopted in the
future.
The process of establishing such a platform-independent stan-
dard for model descriptions could be inspired by successful at-
tempts to establish standards in other domains: Model-Driven
Development (Selic, 2003) is an emerging approach proposing the
creation of domain specific lexica allowing for the simultaneous
development and documentation of models. It has proved suc-
cessful in domains parallel to ABM, notably with the SysML and
ModelicaML languages, thus pointing to a further avenue for
standardisation. To mention a second example, the Object Man-
agement Group (OMG) is a non-for-profit organisation of the
software industry that has developed several widely adopted
standards such as UML (Object Management Group, 2011), while
over the years evolving a rather intricate internal structure with a
multi-tier hierarchy and multiple ad-hoc boards at top level. A third
example is the Open Geospatial Consortium (OGC), which involves
the public sector, academia and industry to develop standards for
spatial data publication and sharing (Castronova et al., 2013). These
success stories highlight two points: First, some sort of formal and
well-defined organisation is needed to drive the process of speci-
fication and later on the diffusion of the standard. Second, well-
defined standards that address objective problems tend to be
swiftly adopted by software developers and the industry in general.
In the context of ABM, which is a domain still somewhat restricted
to academia, an organisation like the OGC seems more suitable.
This kind of organisation may be simpler for a small number of
volunteers to start working on a draft standard, drawing on plat-
forms such as www.openabm.org.
5. Conclusion
We have identified eight main purposes of model descriptions
and summarized our expert knowledge in an assessment of the
suitability of description types for different purposes. We conclude
that no single model description type alone can fulfil all purposes
simultaneously. Hence, we suggest a minimum standard consisting
of a structured natural language description plus the provision of
source code. Such description frame is particularly important for
academic purposes, favouring in-depth model comprehension and
model assessment. This echoes comments by other authors (e.g.
Polhill and Edmonds, 2007;Ince et al., 2012) that good modelling
practice entails both the provision of source code and an accessible
natural language description, ideally following a formalized stan-
dard such as ODD (Grimm et al., 2006, 2010). However, other
description types can strengthen model description substantially in
regard to specific purposes or target groups. For instance, graphics
are appropriate to facilitate the model communication, while
ontologies can foster model comparison and mathematics can
improve the possibilities of replication.
Every author should therefore tailor the usage and weight of one
or more description types according to the characteristics of the
model and the purpose of the publication, in the view of meeting
the above mentioned minimum standard. A joint effort of the ABM
community towards transparent and comprehensible model de-
scriptions through the use of standards would lead to a significant
advancement of the field by enhancing exchange of information
between peers and improving communication with model end-
users. Therewith, the potential of agent-based modelling to sup-
port problem-oriented analysis and governance of humanenatural
systems would strongly increase.
Acknowledgements
We thank all participants of the workshop “Human decisions in
agent-based models (ABM) for natural resource use - need for
protocols”at the 6th International Congress on Environmental
Modelling and Software (iEMSs) 2012 in Leipzig for their valuable
input to the discussions and Volker Grimm and three anonymous
reviewers for helpful comments on an earlier version of this
manuscript.
Financial support from the DFG (German Research Foundation)
in the framework of the Collaborative Research Centre 586 is
gratefully acknowledged by BM. JGP’s contribution was funded by
The Scottish Government Rural Affairs and the Environment Port-
folio Strategic Research Theme 1 (Ecosystem Services). JM’s
contribution was funded by a Leverhulme Early Career Fellowship.
MS acknowledges funding from the European Research Council
under the European Union’s Seventh Framework Programme (FP/
2007-2013)/ERC grant agreement no 283950 SES-LINK. ZS’s
contribution was supported by I-REDDþproject funded by the
European Community’s Seventh Framework Research Programme.
DP’s contribution was supported by the Canadian Social Science
and Humanities Research Council grant 410-2011-1340.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.envsoft.2014.01.029.
References
Aalders, I., 2008. Modeling land-use decision behavior with Bayesian belief net-
works. Ecol. Soc. 13.
Alessa, L.N.i., Laituri, M., Barton, M., 2006. An “all hands”call to the social science
community: establishing a community framework for complexity modeling
using agent based models and cyberinfrastructure. J. Artif. Soc. Soc. Simul. 9 (4),
14.
Balbi, S., Giupponi, C., 2010. Agent-based modelling of socio-ecosystems: a meth-
odology for the analysis of adaptation to climate change. Int. J. Agent Technol.
Syst. (IJATS) 2 (4), 17e38.
Balbi, S., Giupponi, C., Perez, P., Alberti, M., 2013. A spatial agent-based model for
assessing strategies of adaptation to climate and tourism demand changes in an
alpine tourism destination. Environ. Model. Softw. 45 (0), 29e51.
Becu, N., Perez, P., Walker, A., Barreteau, O., Le Page, C., 2003. Agent based simu-
lation of a small catchment water management in northern Thailand descrip-
tion of the CATCHSCAPE model. Ecol. Model. 170 (2e3), 319e331.
Benenson, I., Torrens, P.M., 2004. Geosimulation: Automata-based Modeling of
Urban Phenomena. Wiley, New York, USA.
Bennett, N.D., Croke, B.F.W., Guariso, G., Guillaume, J.H.A., Hamilton, S.H.,
Jakeman, A.J., Marsili-Libelli, S., Newham, L.T.H., Norton, J.P., Perrin, C.,
Pierce, S.A., Robson, B., Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013.
Characterising performance of environmental models. Environ. Model. Softw.
40 (0), 1e20.
Bersini, H., 2012. UML for ABM. J. Artif. Soc. Soc. Simul. 15 (1), 9.
Bonabeau, E., 2002. Agent-based modeling: methods and techniques for simulating
human systems. Proc. Natl. Acad. Sci. U. S. A. 99 (Suppl. 3), 7280e7287.
Caillault, S., Mialhe, F., Vannier, C., Delmotte, S., Kêdowidé, C., Amblard, F.,
Etienne, M., Bécu, N., Gautreau, P., Houet, T., 2013. Influence of incentive
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163162

networks on landscape changes: a simple agent-based simulation approach.
Environ. Model. Softw. 45 (0), 64e73.
Castronova, A.M., Goodall, J.L., Elag, M.M., 2013. Models as web services using the
Open Geospatial Consortium (OGC) Web Processing Service (WPS) standard.
Environ. Model. Softw. 41 (0), 72e83.
Christley, S., Xiang, X., Madey, G., 2004. An Ontology for Agent-based Modeling and
Simulation, Agent 2004 Conference on Social Dynamics: Interaction, Reflexivity
and Emergence. Citeseer.
de Sousa, L., da Silva, A., 2011. Review of spatial simulation tools for geographic
information systems, SIMUL 2011. In: The Third International Conference on
Advances in System Simulation, pp. 169e174.
Deadman, P., Robinson, D., Moran, E., Brondizio, E., 2004. Colonist household
decisionmaking and land-use change in the Amazon Rainforest: an agent-based
simulation. Environ. Plan. B Plann. Des. 31 (5), 693e709.
Eden, C., 1988. Cognitive mapping. Eur. J. Oper. Res. 36 (1), 1e13.
Epstein, J.M., 2008. Why model? J. Artif. Soc. Soc. Simul. 11 (4), 12.
Filatova, T., Parker, D., van der Veen, A., 2009. Agent-based urban land markets:
agent’s pricing behavior, land prices and urban land use change. JASSS J. Artif.
Soc. Soc. Simul. 12 (1).
Filatova, T., Verburg, P.H., Parker, D.C., Stannard, C.A., 2013. Spatial agent-based
models for socio-ecological systems: challenges and prospects. Environ.
Model. Softw. 45 (0), 1e7.
Gilberg, R.F., Forouzan, B.A., 2004. Data Structures: a Pseudocode Approach with C.
Thomson.
Grau, B.C., Horrocks, I., Motik, B., Parsia, B., Patel-Schneider, P., Sattler, U., 2008. OWL
2: the next step for OWL. Web Semant. Sci. Serv. Agents World Wide Web 6 (4),
309e322.
Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., Goss-Custard, J.,
Grand, T., Heinz, S., Huse, G., Huth, A., Jepsen, J.U., Jørgensen, C., Mooij, W.M.,
Müller, B., Pe’er, G., Piou, C., Railsback, S.F., Robbins, A.M., Robbins, M.M.,
Rossmanith, E., Rüger, N., Strand, E., Souissi, S., Stillman, R.A., Vabø, R., Visser, U.,
DeAngelis, D.L., 2006. A standard protocol for describing individual-based and
agent-based models. Ecol. Model. 198 (1e2), 115e126.
Grimm, V., Berger, U., DeAngelis, D.L., Polhill, J.G., Giske, J., Railsback, S.F., 2010. The
ODD protocol: a review and first update. Ecol. Model. 221 (23), 2760e2768.
Gruber, T.R., 1993. A translation approach to portable ontology specifications.
Knowl. Acquis. 5 (2), 199e220.
Haase, D., Lautenbach, S., Seppelt, R., 2010. Modeling and simulating residential
mobility in a shrinking city using an agent-based approach. Environ. Model.
Softw. 25 (10), 1225e1240.
Hare, M., Deadman, P., 2004. Further towards a taxonomy of agent-based simulation
models in environmental management. Math. Comput. Simul. 64 (1), 25e40.
Holland, J.H., 1995. Hidden Order: How Adaptation Builds Complexity. Addison-
Wesley, Reading, MA.
Horrocks, I., Patel-Schneider, P.F., Van Harmelen, F., 2003. From SHIQ and RDF to
OWL: the making of a web ontology language. Web Semant. Sci. Serv. Agents
World Wide Web 1 (1), 7e26.
IEEE, 1985. IEEE Standard for Binary Floating-Point Arithmetic. IEEE. Institute of
Electrical and Electronics Engineers, New York, NY, pp. 754e1985.
Ince, D.C., Hatton, L., Graham-Cumming, J., 2012. The case for open computer
programs. Nature 482 (7386), 485e488.
International Organization for Standardization (ISO), 1985. Information Processing
eDocumentation Symbols and Conventions for Data, Program and System
Flowcharts, Program Network Charts and System Resources Charts.
Jensen, F.V., 2001. Bayesian networks and decision graphs. Statistics for engineering
and information science. Springer 32.
Kelly, R.A., Jakeman, A.J., Barreteau, O., Borsuk, M.E., ElSawah, S., Hamilton, S.H.,
Henriksen, H.J., Kuikka, S., Maier, H.R., Rizzoli, A.E., van Delden, H., Voinov, A.A.,
2013. Selecting among five common modelling approaches for integrated
environmental assessment and management. Environ. Model. Softw. 47 (0),
159e18 1.
Kitchin, R.M., 1994. Cognitive maps ewhat are they and why study them. J. Environ.
Psychol. 14 (1), 1e19.
Knuth, D.E., 1984. Literate programming. Comput. J. 27 (2), 97e111.
Liu, J.G., Dietz, T., Carpenter, S.R., Folke, C., Alberti, M., Redman, C.L., Schneider, S.H.,
Ostrom, E., Pell, A.N., Lubchenco, J., Taylor, W.W., Ouyang, Z.Y., Deadman, P.,
Kratz, T., Provencher, W., 2007. Coupled human and natural systems. Ambio 36
(8), 639e649.
Livet, P., Müller, J.-P., Phan, D., Sanders, L., 2010. Ontology, a mediator for agent-
based modeling in social science. J. Artif. Soc. Soc. Simul. 13 (1), 3.
Luke, S., Cioffi-Revilla, C., Panait, L., Sullivan, K., Balan, G., 2005. MASON: a multi-
agent simulation environment. Simulation 81 (7), 517e527.
Marohn, C., Schreinemachers, P., Quang, D.V., Berger, T., Siripalangkanont, P.,
Nguyen, T.T., Cadisch, G., 2013. A software coupling approach to assess low-cost
soil conservation strategies for highland agriculture in Vietnam. Environ.
Model. Softw. 45 (0), 116e128.
Maruyama, M., 1992. A quickly understandable notation system for causal loops for
strategic decision makers. Technol. Forecast. Soc. Change 42 (4), 409e412.
McConnell, W.J., Millington, J.D.A., Reo, N.J., Alberti, M., Asbjornsen, H., Baker, L.A.,
Brozovic, N., Drinkwater, L.E., Drzyzga, S.A., Fragoso, J., Holland, D.S., Jantz, C.A.,
Kohler, T.A., Maschner, H.D.G., Monticino, M., Podestá, G., Pontius, R.G.,
Redman, C.L., Sailor, D., Urquhart, G., Liu, J., 2011. Research on Coupled Human
and Natural Systems (CHANS): approach, challenges, and strategies. Bull. Ecol.
Soc. Am. 92 (2), 218e228.
Millington, J.D.A., O’Sullivan, D., Perry, G.L.W., 2012. Model histories: narrative
explanation in generative simulation modelling. Geoforum 43 (6), 1025e1034.
Müller, B., Bohn, F., Dreßler, G., Groeneveld, J., Klassert, C., Martin, R., Schlüter, M.,
Schulze, J., Weise, H., Schwarz, N., 2013. Describing human decisions in agent-
based models eODD þD, an extension of the ODD protocol. Environ. Model.
Softw. 48 (0), 37e48.
North, M., Howe, T., Collier, N., Vos, J., 2007. A Declarative Model Assembly Infra-
structure for Verification and Validation, Advancing Social Simulation: the First
World Congress. Springer, pp. 129e
140 .
Object Management Group, 2011. UML Infrastructure Specification, v2.4.1.
Parker, D.C., Brown, D.G., Polhill, J.G., Deadman, P.J., Manson, S.M., 2008. Illustrating
a new ‘conceptual design pattern’for agent-based models and land use via five
case studies: the MR POTATOHEAD framework. In: Paredes, A.L., Iglesias, C.H.
(Eds.), Agent-based Modelling in Natural Resource Management. Universidad
de Valladolid, Valladolid, Spain, pp. 23e51.
Perez, L., Dragicevic, S., 2010. Modeling mountain pine beetle infestation with an
agent-based approach at two spatial scales. Environ. Model. Softw. 25 (2), 223e
236.
Polhill, J.G., Parker, D., Brown, D., Grimm, V., 2008. Using the ODD protocol for
describing three agent-based social simulation models of land-use change.
J. Artif. Soc. Soc. Simul. 11 (2), 3.
Polhill, J.G., 2010. ODD updated. J. Artif. Soc. Soc. Simul. 13 (4), 9.
Polhill, J.G., Edmonds, B., 2007. Open access for social simulation. JASSS J. Artif. Soc.
Soc. Simul. 10 (3).
Polhill, J.G., Gimona, A., Gotts, N.M., 2013. Nonlinearities in biodiversity incentive
schemes: a study using an integrated agent-based and metacommunity model.
Environ. Model. Softw. 45 (0), 74e91.
Polhill, J.G., Gotts, N.M., 2009. Ontologies for transparent integrated humanenatural
system modelling. Landsc. Ecol. 24 (9), 1255e1267.
Polhill, J.G., Izquierdo, L., 2005. Lessons learned from converting the artificial stock
market to interval arithmetic. J. Artif. Soc. Soc. Simul. 8 (2).
Quinlan, J.R., 1986. Induction of decision trees. Mach. Learn. 1 (1), 81e106.
Rebaudo, F., Dangles, O., 2013. An agent-based modeling framework for integrated
pest management dissemination programs. Environ. Model. Softw. 45 (0),141e
149 .
Robinson, D.T., Sun, S., Hutchins, M., Riolo, R.L., Brown, D.G., Parker, D.C., Filatova, T.,
Currie, W.S., Kiger, S., 2013. Effects of land markets and land management on
ecosystem function: a framework for modelling exurban land-change. Environ.
Model. Softw. 45 (0), 129e140.
Roy, G.G., 2006. Designing and explaining programs with a literate pseudocode.
J. Educ. Resour. Comput. 6 (1), 1.
Schlüter, M., Hinkel, J., Bots, P.W.G., Arlinghaus, R., 2014. Application of a SES
framework for model-based analysis of the dynamics of socialeecological
systems. Ecol. Soc. (in press).
Schmolke, A., Thorbek, P., DeAngelis, D.L., Grimm, V., 2010. Ecological modelling
supporting environmental decision making: a strategy for the future. Trends
Ecol. Evol. 25 (8), 479e486.
Selic, B., 2003. The pragmatics of model-driven development. IEEE Softw. 20 (5).
Smajgl, A., Bohensky,E., 2013. Behaviour and space in agent-based modelling: poverty
patterns in East Kalimantan, Indonesia. Environ. Model. Softw. 45 (0), 8e14.
Smed, J., Hakonen, H., 2006. Algorithms and Networking for Computer Games. John
Wiley & Sons.
Sun, Z., Müller, D., 2013. A framework for modeling payments for ecosystem ser-
vices with agent-based models, Bayesian belief networks and opinion dynamics
models. Environ. Model. Softw. 45 (0), 15e28.
Taillandier, P., Vo, D.-A., Amouroux, E., Drogoul, A., 2012. GAMA: a simulation
platform that integrates geographical information data, agent-based modeling
and multi-scale control. Princ. Pract. Multi-Agent Syst., 242e258.
Tisue, S., Wilensky, U., 2004. Netlogo: a Simple Environment for Modeling
Complexity, International Conference on Complex Systems, pp. 16e21.
van Oel, P.R., Krol, M.S., Hoekstra, A.Y., Taddei, R.R., 2010. Feedback mechanisms
between water availability and water use in a semi-arid river basin: a spatially
explicit multi-agent simulation approach. Environ. Model. Softw. 25 (4), 433e
443.
Wikipedia, 2013. Wikipedia: WikiProject Computer Science/Manual of Style eAl-
gorithms. http://en.wikipedia.org/wiki/MOS. PSEUDOCODE#Algorithms (last
accessed 12.03.13.).
Wilensky, U., Rand, W., 2007. Making models match: replicating an agent-based
model. JASSS J. Artif. Soc. Soc. Simul. 10 (4).
Zhang, B., Zhang, Y., Bi, J., 2011. An adaptive agent-based modeling approach for
analyzing the influence of transaction costs on emissions trading markets.
Environ. Model. Softw. 26 (4), 482e491.
B. Müller et al. / Environmental Modelling & Software 55 (2014) 156e163 163