Space, Complexity, and Agent-Based Modeling

There is considerable interest in agent-based modeling as a tool to understand better the dynamics of complex systems. Particular attention has been focused by the land-change science community, but there has also been a good deal of effort in fields including epidemiology (Teweldemedhin et al, 2004), finance (LeBaron, 2000), computational sociology (Macy and Willer, 2002), ecology (Grimm et al, 2005), and computational economics (Tesfatsion, 2002). There are several notable prior syntheses and collections of agent-based modeling work. Gimblett (2002) and Janssen (2003) edited two early collections of agent-based model (ABM) research centered on complex human ^ environment sytems. Parker et al (2003) summarized applications of ABMs to land-use and landcover change, including how the field transitioned from abstract `toy' models to those more directly tied to real-world applications. Brown and Xie (2006) organized an issue of the International Journal of Geographic Information Science that included a number of agent-based applications with a particular emphasis on the spatial dynamics within the models. And, as evidence of the evolution in the application of agent-based approaches from abstract systems to real-world ones, Janssen and Ostrom (2006) collected a series of works of ABMs in the journal Ecology and Society that were directly supported by different types of empirical data. There are at least two fundamental reasons why ABMs are appealing tools for studying complex systems. First, ABMs explicitly incorporate agent interactions and the properties that emerge at higher levels of observation from these interactions. The importance of these agent interactions varies from system to system, but in many cases these relationships provide key insight into the behavior of complex systems. Second, ABMs enable modelers to represent agents with heterogeneous properties. Instead of every cell in an urban growth model being governed by the same land-change dynamics, the fitting process of an ABM can enable diverse spatial dynamics to be discovered through modeling. For example, landowners in south-central Indiana have varying responses to the same land-use decision-making context (Evans and Kelley, 2004). Recent innovations in agent-based modeling have produced yet more sophisticated representations of complex systems. A large body of work, particularly in land-change science, has focused on households as the primary agent of study. Now we see more diverse types of agents being portrayed, such as individuals (human and otherwise), villages, viruses, or voters. As we learn more about the kinds of modeling possible with agent-based approaches, we will likely incorporate a greater variety of agents in our models. And with this evolution will hopefully come a greater understanding not only of village-to-village interactions or landowner-to-landowner interactions, but also of interactions that go up and down across scales and representational levels (eg landowner to village, village to landowner). While there has been great progress in the agent-based modeling community (see, for example, the manuscripts here and those in the special issues and edited collections noted above), there remain a number of key challenges that point to reasonable next steps in the research enterprise. Researchers have made greater efforts to validate ABMs than was the case during the inception of ABM applications (Janssen and Ostrom, 2006). However, we should continue to consider the types of validation being Guest editorial Environment and Planning B: Planning and Design 2007, volume 34, pages 196 ^ 199