Agent-Based Modeling

Agent-based modeling (ABM) is a technique used to build computer simulations. ABM allows for the creation of synthetic, but ultimately realistic, artificial geographic worlds in which events, phenomena, processes, and scenarios can be created and studied flexibly. ABM is an important tool in human geography employed in evaluating hypotheses and ideas that might not be easily experimented with, evaluating “what if” scenarios that cannot be tested otherwise, or relating to future conditions that cannot be sampled.

ABM is a part of a growing geographic methodology based on geocomputation and geosimulation. Both approaches mark a departure from traditional models focused on exchange of human geographic units between coarsely represented divisions of space. Newer models based on ABM are more likely to be built as simulations with massive amounts of intelligent geographic entities, each represented at the atomic scale, connected and interacting dynamically in space as complex adaptive systems.

Agent-based models belong to a family of models called automata. Automata have distinguished origins in pioneering work on digital computing during the 1930s and 1940s. Automata tools were first employed in geography as cellular models during the early 1970s, with the methodology evolving toward ABM during the 1990s. An automaton is a simple information processor just like the processors in digital watches and computers. Automata have some key properties that render them useful for model building. They have states that allow attributes to be encoded to them, changed, and stored. Automata have some representation of time that catches state conditions at discrete temporal points. They also contain transition rules that govern changes between states as time progresses. Rules are formulated as (computational or mathematical) functions that accept state information input from other automata, and this can be derived from neighboring automata within a specified local neighborhood of influence, as is characteristic with cellular automata.

Agent automata extend this basic framework, adding attributes borrowed from research on behavior and artificial intelligence. These attributes are very relevant for work in human geography. Agents are heterogeneous, contrasting with more traditional models that treat entities as “average individuals.” Agents are also proactive and may act to realize a goal or set of goals. They may have perception—the ability to sense other agents and environments—often based on an internal cognitive model. Importantly, agent interaction may take many forms: communication, active and intentional querying of other agents, human–environment effects, and so on. Agents are also adaptive and may change their rules of behavior based on experience within a simulation.

Agent tools are used in a variety of applications in human geography: pedestrian and crowd motion, vehicular traffic, residential mobility, gentrification, [Page 2]land use and land cover change, urban growth and sprawl, spatial epidemiology, civil violence, sociospatial segregation, and economic geography.

During recent years, research in this area has focused on applying agent-based models to new phenomena in human geography, and a growing integration between ABM and geographic information systems (GIS) and geographic information science (GIScience). In particular, authors have begun to develop geography-specific methodologies and toolkits based on ABM but with geography as a central building block.

...