Abstract
Agent-based modelling (ABM) presents a promising approach to investigating complex dynamic systems (e.g., urban systems) at the individual level over time, and thus has received increasing attention over the past few years in various research fields (Crooks et al., 2008). In many cases, ABM is involved in spatial analysis and modelling, as location is a key characteristic of an agent and is closely associated with many of its decisions and actions (e.g., trip destination choice). Therefore, there is a growing need for integrating ABM and Geographical Information Systems (GISs) to create spatially explicit Agent-based Models (or spatial Agent-based Models).
The book “Agent-based Modelling & Geographical Information Systems” is timely. It includes very detailed introductions to both ABM and GISs, and demonstrates how to develop, verify, calibrate and validate a spatially explicit Agent-based Model step by step, using free and popular software packages, such as NetLogo and QGIS (for ABM and GISs, respectively). This is particularly helpful for beginners (e.g., undergraduates) to easily and quickly learn about both ABM and GISs, as well as the ways to integrate them. Furthermore, I can also see several very interesting discussions on the recent challenges and opportunities in ABM and GISs. For example, big data (e.g., mobile-phone data and social media data) has become an important data source for understanding and modelling individual behaviours, and the discussion in this book on ABM and big data can offer useful advice and ideas. Therefore, the book should also be of interest to experienced researchers in ABM or spatial modelling.
The book starts with an overview of ABM and GISs (Chapter 1), which includes introductions to several relevant key terms, such as complexity, geographical systems and individuals, as well as the benefits of combining ABM and GISs.
Chapters 2–4 go into detail on Agent-based Models and Modelling. Specifically, Chapter 2 gives a more detailed introduction to ABM, discussing its advantages and limitations and presenting some typical Agent-based Models (e.g., SugarScape). Chapter 3 shows how to design and develop an Agent-based Model, with a focus on its essential elements, including the world (e.g., external systems), agents and their characteristics (e.g., age and income) and behaviours (e.g., decisions and actions), as well as the interactions between agents and the physical environment and between agents. These elements are well illustrated later within a classical segregation model. Special attention is paid here to the Overview, Design Concepts and Details (ODD) protocol, which is a widely accepted approach to documenting new ABM in detail. Chapter 4 starts with some basics about NetLogo and then demonstrates how to develop an Agent-based Model (specifically, simulating sheep eating grass) and further improve it by adding more variables and complex behaviours of agents.
Attention is then turned to GISs in Chapter 5, giving an overview of GISs, including their history, types and sources of data, temporal information, popular GIS software packages, and visualizations (or map types) for GISs. In particular, it lists several GIS tools, with a focus on QGIS, one of the most popular open-source products. Further, in Chapter 6, methods and tools for integrating ABM and GISs are described in detail. Apart from NetLogo, other well-known platforms for the integration of ABM and GISs, such as Swarm, MASON and Repast, are also introduced. Again, NetLogo is used here as an example for demonstration.
Chapters 7 and 8 present two specific groups of spatially explicit ABM: Human Behaviour and Networks, respectively. Simulating human behaviour remains a challenging endeavour, as agents are different from each other in terms of, for example, characteristics, emotions and experience, and they also interact with each other directly or indirectly. More importantly, all of these can change over time, which makes the simulation more difficult. Chapter 7 compares several classical behavioural frameworks (e.g., cognitive ones) and also introduces two general approaches to representing individual behaviours in ABM, namely mathematical approaches and conceptual cognitive models. Chapter 8 looks at both social and physical networks, as well as their basic properties (e.g., density) and types. Specific examples include transport networks (e.g., road and transit networks), online social networks (e.g., connections on Facebook) and offline social networks (e.g., friendships).
Chapters 9 and 10 are focused on the approaches to verifying, calibrating and validating GIS-based Agent-based Models. Specifically, Chapter 9 firstly introduces four groups of spatial statistics and algorithms, namely global statistics (e.g., R squared), visual comparisons (e.g., point map), description of point data (e.g., nearest neighbour) and local indicators of spatial association (e.g., GI*), which are essentially used to compare two spatial data sets and thus can be used for calibration and validation of spatial Agent-based Models. Chapter 10 then introduces several different ways to evaluate an Agent-based Model through verification, calibration and validation. However, it remains challenging to fully evaluate an Agent-based Model, because (1) in many cases, there is limited availability of disaggregate data to calibrate or validate an Agent-based Model at the individual level over time; and (2) parameterizing an Agent-based Model could be very time-consuming, especially for large-scale scenarios, which can take a long time for a single run (Zhuge and Shao, 2019).
Chapter 11 compares spatial ABM against several alternative modelling approaches from both theoretical and practical perspectives, including cellular automata, discrete event simulation, system dynamics and spatial interaction models. The theorical comparisons look at the number of levels, whether agents communicate with each other, and the complexity and number of agents. In the practical comparisons, an epidemic model is implemented using both ABM and the other alternative approaches above, so as to better understand their differences.
The book concludes with a discussion on challenges and future research directions in Chapter 12, which tends be of more interest to advanced researchers. Key challenges remain in the whole process of ABM, involving theory choice, data availability and model comparison, evaluation and dissemination. Some of the challenges are interconnected. For example, there are a number of Agent-based Models developed for residential location choices (Huang et al., 2014). However, little effort has been made to compare them in the same scenario. One of the reasons might be that many of them are not described in detail (for example, using the ODD protocol), and thus cannot be replicated. Furthermore, the model codes are generally not available online. In response to these challenges, we are expecting improvements in spatial Agent-based Models in future work: apart from ABM and big data discussed above, integration of different Agent-based Models, model uncertainty analysis and data assimilation will also likely receive more attention. These improvements are expected to make the models more realistic (in terms of individual behaviours), accurate (in terms of predictive ability) and applicable (in terms of case studies).
Overall, the book should be useful for both human and physical geographers who are seeking guidance and ideas on ABM and GIS. Readers can quickly learn about essential information, knowledge and skills in this domain. Although many of the examples in the book are more related to human geography, they should be useful for physical geographers as well, as human factors are closely associated with the physical environment. Therefore, the book would be particularly useful for those physical geographers attempting to explore the interactions between human and physical systems at the individual level. For example, an Agent-based Model of individual movement can be coupled with a hydrodynamic model using NetLogo to simulate the vulnerability of individuals to flooding within “what-if” scenarios considering different storm surge conditions and evacuation strategies (Dawson et al., 2011). Compared with other books on ABM or spatial modelling, this one tends to have much more detailed introductions to ways of developing a spatially explicit Agent-based Model from scratch, particularly using popular open-source software packages for ABM and GIS. Furthermore, readers can also benefit from the interesting and informative discussions on recent challenges and opportunities, as well as useful comparisons between different tools, theories and frameworks for spatial ABM.