geosimulation :: innovative geospatial simulation and analysis but innovative people

Home | Book | Research | Publications | Bio | Press | Geosimulation Labs
Dr. Paul M. Torrens, Department of Geography, University of Maryland, torrens at geosimulation dot com

Background material on the following topics is available on this site:

An introduction to geosimulation
Geosimulation is a catch-all phrase that can be used to represent a new wave of spatial simulation modeling that has come to the fore in very recent years. Besides traditional urban modeling and simulation, the intellectual roots of geosimulation derive from recent developments in computer science and geographic information science. The geosimulation approach draws together a diversity of theories and techniques, offering a unique perspective that traditional simulation has commonly lacked: a view of urban phenomena as a result of the collective dynamics of interacting objects, often represented at the scale of individual households, people, and units of real estate and at time-scales approaching "real time".

Geosimulation has similarities with traditional modeling approaches such as microsimulation, but there are several factors that distinguish it from its predecessors.

The first aspect regards the depiction of spatial units. While traditional urban models have focused on aggregate partitions of urban space, essentially modifiable spatial units, geosimulation-style models are often run with discrete and spatially non-modifiable objects at a "microscopic" scale, e.g., houses, lots, householders, and landowners.

The second feature relates to the portrayal of spatial relationships. Traditionally, geographic simulations are constrained by limiting assumptions in the methodologies used to build them. Spatial interaction modeling, for example, describes one form--and one form only--of spatial interaction: flows of matter and information between aggregate spatial units. Microsimulation often deals with individual units, but they are modeled in isolation; the interactions between units are not generally considered. The individual-oriented models characterized by the geosimulation approach contrast by concentrating on the interactive behavior of elementary geographic objects in a limitless variety of ways, whether this interaction takes the form of flows, or other spatial relationships such as action-at-a-distance, diffusion, aggregation, etc. In addition, relationships that might be observed at higher scales (such as at intra-urban levels) can be modeled as collections of these elementary units, assembled from the bottom-up.

The third characteristic is concerned with the treatment of time. Traditionally, urban models have included simple proxies for time: cross-sectional data for one snapshot in time, or longitudinal data that offers several snapshots, but with little information about the intervening periods. In contrast, geosimulation-style models offer the opportunity for the construction of dynamic simulations, often at time-scales approaching "real time". This has important implications for the range of hypothese that can be explored in simulation.

The fourth characteristic has to do with the goals of simulation. Geosimulation-style modeling marks a departure from the traditional goals of simulation as a predictive exercise. Newer approaches, at least those developed thus far, tend to be designed as scenario-exploring simulations:"tools to think with".

Research work in geosimulation mostly focuses on techniques to improve spatial simulation technology: the derivation of new algorithms for spatial processes, new methodologies for conceptualizing spatial entities and the relationships between them, the application of simulation models to real-world problems, and new software for experimenting with geographical systems. A lot of the ideas in geosimulation are not unique to the discipline. Many ideas are abstracted (read: "poached") from other fields, largely computer science, physics, chemistry, mathematics, economics, ecology, and biology. Of these disciplines, computer science is particularly relevant, especially artificial intelligence, artificial life, and object-oriented programming. Complexity studies and associated notions of emergence, self-organization, and adaptive systems have also been very influential. In addition, studies in individual-oriented modeling commonly make use of developments within the geographical sciences, including geocomputation, geographic information science, and spatial analysis. In many cases, however, much of the methodologies borrowed from outside of geography have been developed in non-spatial contexts and much of the innovation in geosimulation stems from adapting these technologies for explictly geographical applications.

Geosimulation and urban geography
Urban geography is an interesting field to be interested in from a simulation standpoint. First, cities are bewilderingly complex systems to try to comprehend, let alone recreate in a computer. Nevertheless, we have a relatively good idea--at least theoretically--of how they work. Second, urban geography (well, urban planning really) is one of the few areas where there is a rationale for building simulation models (aside from a desire to get paid for playing SimCity). In the United States, for example, there is a legal justification, at the Federal level, under the Clean Air Act Amendments (CAAA) of 1990 and the Transport Efficiency Act for the 21st Century (TEA-21) of 1997, that strongly suggests that cities build simulation models of both land-use and transport to forecast their compliance with air quality standards.
For the most part, "traditional" large-scale urban simulation can be understood to suffer from a series of limitations. (If you want to read more about these models, I have a working paper on the subject--a really long beast of a one--online: How land-use and transportation models work.) The most immediate limitations of "standard" urban simulation models are that they are:

  • Centralized in nature: often, it is assumed that all activity in the city revolves around the downtown
  • Relatively static: time moves in "snapshots", sometimes of several years
  • On an unsteady theoretical footing: they commonly contain very limiting assumptions
  • Highly aggregate: model developers often break a city into a few hundred units in a model
  • (Unnecessarily) complicated: the inner workings of models are not easily conveyed to users
  • Difficult to interpret: their results are not always easy to digest

In short, there is much room for improvement.

Geosimulation focuses its attention on a "new wave" of simulations for urban systems, mostly designed as cellular automata (CA) or multi-agent systems (MAS). These models offer a number of important innovations over their ancestors:

  • They are disaggregated: interactions can be simulated between individuals and actual building units
  • They are inherently dynamic: they operate in "real time" (in the James Gleick sense of the phrase!)
  • The algorithms that they use can be derived directly from theoretical ideas of how cities work
  • They are still complicated, but they make more intuitive sense: they don't draw yawns at parties
  • They are often displayed as visual environments, and as such are a lot easier to interpret

They are also a lot more fun to work with!

As super-cool as these new models are, however, there are some things that they don't do well. CA and MAS are, for the most part, bottom-up models: lots of activity goes on between small-scale entities at a very micro-scale and this "emerges" up, often in quite ordered patterns, at the macro-level. A lot of things work like this in the city, but some things don't. Importantly, CA and MAS are not all that good at handling the properties of urban systems that operate from the top-down, such as the designation of land-uses, planning controls, or the introduction of new public infrastructure (like highways). The "old-style" models that I just attempted to discredit actually do an adequate job of handling top-down systems.

Geosimulation looks to improve the old by sprucing it up with the new. I take the old-style models and, essentially, "fill-in" where they finish off, taking them down to the level of the individual and the building (well, really, taking individuals and buildings up to the level of the zone). It's a bit more complicated than that, but that's it in a nutshell.



Projects >>

Dynamic physics for built infrastructure

moving agents through space and time

Moving agents through space and time

modeling riots

Modeling riots

Validating agent-based models

Machine-learning behavioral geography

Accelerating agent-based models

megacity models

Megacity futures

Immersive modeling

Space-time GIS and analysis

A toolkit for measuring sprawl

space-time GIS

Modeling time, space, and behavior

simulating crowd behavior

Simulating crowd behavior