Warning! The scale of land-use CA is changing! 1. What kind of panacea are cellular automata?
Geographic modeling is a cool issue today, and Cellular Automata (CA) is a cool issue in geographic modeling. Several reasons of this popularity can be suggested: the idea of CA is clear and can be implemented with minimal programming expertise; much high-resolution spatial data has recently become available; public interest in environmental and land-use forecasting is on the upturn. The last few years have seen a burst of publications in the field, with many applications demonstrating apparent correspondences between the model and recent real world data. On the basis of such fits, models are applied for scenario investigation and prediction.
But why, in reality, does this fit always occur? Does it mean that one or more CA models -such as, for example, the popular constrained CA of White and Engelen (White & Engelen, 1997;White & Engelen, 2000) -are universal tools, sufficient for adequate representation of every city's or region's land-use dynamics?
A possible avenue of explanation for the correspondence between model and reality are immediately obvious: (1) the dynamics of land-use patterns -or at least of the characteristics that we use to describe them -are not, apparently, very complicated; (2) the CA models possess sufficient degrees of freedom with respect to the model's spatial and temporal resolution, neighborhood size and form, parameters of the transition rules and the form of their analytical presentation; and (3) whether done analytically or by trial and error, one can always tune the model to fit a case study. This is reassuring yet worrying. An analogy with multiple non-linear regression comes to mind and raises reservations regarding the extrapolation we seek when modeling system dynamics. 'Nothing new under the sun', might be the response of those who recall Douglas Lee's ''Requiem for Large Scale Modeling'' (Lee, 1973).
This in turn begs the question, will the contemporary boom come to a similar unseemly end and share the fate of comprehensive modeling? I do not think so.
2. Even simple CA are complex and should be treated with care
The unfortunate experience of comprehensive modeling should encourage us to simplify our models in every respect possible. Yet the prevailing feeling regarding land-use