[an error occurred while processing this directive]

Integrating GIS, Models, and Predictions: The Background

Until comparatively recently, models were developed quite separately from their representation. With the development of full-fledged computer graphics in the 1980s—spatial representation becoming digital and visual for 2D through early desktop GIS and for 3D through CAD—there began various attempts at a strong coupling of desktop GIS with modeling, but for the most part, this was restricted to models designed in separate software but linked on the desktop.

In fact, models have been more heavily influenced by their use in participatory contexts, where visualization is of course important but where the predominant mode is to simply pick and choose from available software and engage in a loose coupling wherever such a coupling is required. Such is the modus operandi of planning support systems.

Insofar as models have been integrated with various representations and model types, the focus has been on a limited extension of one model type with its close neighbors rather than with major forms of representational or planning support systems, largely because the overhead of implementing a large-scale model in these systems was too great. It has been much easier to take elements from each of these related software packages and build these directly into models, a strategy demonstrated in the model in the main article describing long-term impacts of sea level rise and energy change in the Greater London region.

These sorts of models require good representation and predictive capabilities to input and output their data, and outcomes for rapid understanding and dissemination by scientists and stakeholders alike are usually regarded as large scale.

In an urban context, these are land-use transportation models (LUTM), sometimes referred to as land-use transport interaction (LUTI) models, that simulate the workings of the city system in terms of transportation flows between different land-use activities and the operation of housing and related markets in determining the location of activities at a cross-section in time.

These models have been widely developed since the 1960s, and as computers have gotten ever more powerful and spatial data ever richer, these models have grown in scale. There has been considerable integration of their various parts—for example, in transportation models, notions about integrated distribution and assignment have been widely advanced—while links to demographic and econometric forecasting at higher spatial scales in the form of demo-economic models have been explored. Links to environmental models are somewhat looser but in parallel, some of these model structures have been dissembled in the quest to simulate in ever more detail various subsectors, such as the retail system and the housing market.

These models are often termed operational, in that they are widely used in urban policy making—particularly in large cities—but are still quite distinct from the new generation of urban models that simulate finer-scale movement patterns and change, particularly local movement of individuals and specific changes in urban development. The former style of model is called an agent-based model, while the latter, which attempts to forecast the change in locational activity, is called a cellular automata. The key features of these models are that they are qualitative in their predictions, usually forecasting the spread or movement of development. They have little numerical forecasting of population transitions, travel demand, or housing market clearing as reflected in the prediction of supply or in the determination of prices. There has, however, been progress in stitching these kinds of models into desktop GIS through various plug-ins, such as agent-based modeling routines that interface with open source software.

There is little doubt, however, that one of the basic reasons it is difficult to couple different types of models to their representational software depends on the different professional expertise needed to effect such linkages. For example, linking traffic models to land-use models is hard enough because very different conceptions of these activities are required—land use depends on the urban economy, while transportation is reflected more in detailed design considerations and ideas about traffic flow, more the product of engineering. One of the reasons the development of agent-based modeling has become popular is because it tends toward no specific discipline, in that the conception of an agent and its interactions can be applied, at least at a casual level, to any kind of system. But this is also its Achilles' heel, as invariably, the detail in such models falls far short of that required for strong disciplinary development of theory or for professional policy-making purposes. Such models thus tend to be pedagogical rather than predictive.

Revolutionary Visualization

Now, however, the open nature of many new visualization technologies, particularly now on the Web in the form of online mapping, has spurred the development of all kinds of loose coupling that was hitherto largely unanticipated. There are many software products, some open source, that can now combine, for example, GIS and simulation, and this means that modelers have a cornucopia of possibilities when it comes to extending their models to embrace good representations and simulations.

—Michael Batty

See also "Integrated Models and Grand Challenges."

Contact Us | Privacy | Legal | Site Map