High Resolution Integrated Modelling of the Spatial Dynamics of Urban and Regional Systems

WHITE, Roger (roger@riks.nl), Memorial University of Newfoundland, Department of Geography, St. John's, Newfoundland, Canada

Keywords: cellular automata, land use, integrated models, process models, spatial dynamics

An emerging branch of geocomputing involves the modelling of spatial processes. This work builds on the resources provided by remote sensing and GIS, and, to a lesser extent, spatial statistics, to show how spatial systems evolve in time. The fundamental logic of most spatial process models represents causal relations, and the models are, in a formal sense, predictive; they are also, to a lesser degree, empirically predictive. A variety of techniques are being used, the most important being traditional regionalized system dynamics approaches, multiagent systems (MA), and cellular automata (CA). The techniques are frequently combined to model processes operating at different spatial scales.

Cellular automata are dynamical systems defined on a raster space. Cell states typically represent land use and land cover, and the transition of a cell from one state to another depends on the states of cells within a neighbourhood of the cell. The cell space may either be homogeneous, in the case of theoretical applications designed to investigate basic properties of dynamic spatial systems, or inhomogeneous, in the case of most applications to actual geographical systems, where the inhomogeneities may represent such factors as suitability, accessibility, or legal restrictions on land use; thus, a cellular model may be thought of as running on top of a GIS. In a pure CA, the number of cells in a particular state is determined endogenously by the cellular dynamics; however, this is not realistic for many geographical applications, since the number of cells usually reflects the level of demand for the activity that is carried out on the cells. Most applied cellular models are constrained to generate particular numbers of cells in each state, with the target cell numbers determined exogenously, usually by another model.

Urban and regional models based on cellular automata give good representations of the spatial dynamics of land use and land cover, as judged by a variety of measures, from the Kappa index to various fractal dimensions and visual inspection. In a current application, a cellular model of The Netherlands as 500-m resolution is driven by a macro-scale dynamical spatial interaction model defined on 40 economic regions; this model is in turn driven by national planning projections and policy goals. Given the national totals, the macro scale model generates regional demands for population and economic activity. These demands are translated into demands for cell space, which the CA then attempts to allocate. In turn, information on conditions at the cellular level, such as the quantity and quality of land available to various activities and actual densities at the cellular scale, are returned to the regional model to modify parameter values there. Linking the two models operating at the two scales improves the performance of both; for example, regional population estimates are improved by about 65 percent when the macro scale model is linked to the cellular level. The purpose of this multiscale model, developed for the Netherlands Institute of Health and the Environment, is to permit the evaluation of national and regional policies in terms of their effects on the natural and human environment. This is only possible because the CA model is able to translate the national trends and policies into potential consequences at the micro scale where most effects will be experienced.

The results of high resolution modelling of spatial dynamics raise a number of methodological issues. One of the most pressing concerns evaluation of the results. In calibration and testing, a model-generated map must be compared to an actual land use/land cover map, but current pixel by pixel techniques are unable to capture patterns and often misrepresent the degree of similarity, while other measures, such as fractal dimensions, are too general. Approaches based on fuzzy logic and pattern recognition techniques show some promise. Another issue concerns predictability. To the extent that these models capture the evolving, innovative nature of real cities and regions, they cannot be strictly predictive, nor can they reliably be characterized by statistical measures because the statistical patterns may undergo punctuated evolution as the system's history unfolds. We know that the world is a complex mix of predictability, uncertainty, and novelty. A CA-based modelling approach captures this, but at the same time it implies a different kind of science, for which the methodology and standards have not yet been fully worked out.