A comparison of three urban models of land use, transport and activity

Urban modeling is inherently an interdisciplinary endeavor, with a broad societal relevance. It is not surprising that it brings together researchers from a wide range of backgrounds, carrying with them the paradigms, toolsets and terminology of their disciplines. These backgrounds include economists and econometricians, geographers, planners and architects, civil engineers, computer scientists and many more. This paper makes a systematic comparison between three models that are all concerned with the organization of land use and activities in space and that all see proximity and accessibility as major determinants of that organization. Despite these strong conceptual similarities, the models are conceived from fundamentally different starting points and assumptions, and it is often thought that there are irreconcilable differences. The compared models are the MOLAND Cellular Automata model as introduced by White and Engelen [3]; the dynamic spatial interaction model UrbanSim [2]; and the MEPLAN regional economic land use and transport model as proposed by Echenique [1]. These models differ in their basic units: parcels of land in the cellular automata model and households and organizations in MEPLAN and UrbanSim. They also differ in their understanding of the predictability and regularity of the urban growth process: Central to the MOLAND model is the notion of complexity and self-organization, UrbanSim on the other hand assumes that rates of change follow continuous functions that may well be estimated using standard econometric techniques, and MEPLAN is based on a constant equilibrium in the demand and supply for land. This paper compares the strengths and weaknesses of these models; it also shows how the models make concessions to their basic structure in order to overcome weaknesses. It then emerges that in practice the models are more similar than their theoretical roots suggest and the remaining differences are to a lesser extent fundamental conceptual differences, but rather differences in emphasis, terminology and computational methods. The comparison of the models' strengths and weaknesses gives rise to the idea of a best-of-worlds hybrid solution. The paper concludes by setting out the basic components of such a solution and presenting a number of suggestions to make it work computationally.