How Much Does Transportation Matter?

Several methods have been developed for forecasting land use change, with varying degrees of sensitivity to the influence of transportation networks. The simplest types of models for forecasting land use change are Markovian models (1–3) such as Markov chain models, which tend to treat land use change as a stochastic process. Assuming that rates of change between land use types are more or less constant from one period to the next, Markovian models project land use transitions forward to any given future date, eventually reaching an equilibrium distribution of land uses. These models tend to have a limited ability to incorporate transportation networks and other spatial features, except as states (e.g., land use types) in the model. More often, they are applied to analyses of land cover change. Cellular and agent-based models have recently gained greater acceptance as tools for simulating land use change in urban areas. Advances in computational power and data storage have facilitated the development of models that disaggregate urban space to a greater degree and can operate with individuals or land parcels as the units of analysis, rather than zones. These include microsimulation models of urban development (4), as well as models based on a cellular automata framework (5, 6). Cellular automata models emphasize neighbor effects and dynamic interactions between agents (with land use cells as agents), while microsimulation models treat individual households and firms as agents and attempt to simulate their behavior in terms of location and travel choices. Microsimulation models of land use are often coupled with transportation models and are integrated into larger urban simulation models (4, 7). Despite these methodological advances, regression models continue to be a popular method for modeling and simulating land use change. Indeed, many simulation models with a land use component use regression methods, either in the form of discrete models of land use change (8, 9) or within hedonic or bid-rent frameworks for land prices (4, 10). Regression models allow the identification of exogenous variables, which are thought to influence patterns of development. The variables can represent physical and social influences on development (11, 12), neighborhood effects (12, 13), or the effects of transportation and accessibility (14, 15). It is these latter effects that are of the greatest interest in the current context. While regression techniques have been used previously to identify the correlates of highway network growth in terms of land use and population characteristics (16), here we will focus our attention on relationships running in the opposite direction, that is, changes in land use types as a function of transportation and other attributes.