Of Cells and Cities: a Comparative Econometric and Cellular Automata Approach to Urban Growth Modelling

摘要

This paper presents a comparative assessment of two distinct urban growth modeling approaches. The first urban model uses a traditional Cellular Automata methodology, based on Markov transition chains to prospect probabilities of future urban change. Drawing forth from non-linear cell dynamics, a multi-criteria evaluation of known variables prospects the weights of variables related to urban planning (road networks, slope and proximity to urban areas). The latter model, frames a novel approach to urban growth modeling using a linear Logit model (LLM) which can account for region specific variables and path dependency of urban growth. Hence, the drivers and constraints for both models are used similarly and the same study area is assessed. The comparative approach of both these model introduces the region of the Algarve, the farthest south region of Continental Portugal, which has been since the sixties victim to excessive urban sprawl, bringing specific vulnerability to the coastal region. Within the European context, the CORINE Land Cover project is used for assessment in the multi-temporal spatial landcover layers for 1990, 2000 and 2006. Both models are projected in the segment of Faro-Olh～ao for 2006 and a comparative assessment to ground truth is held. The calculation of Cohen's Kappa for both projections in 2006 allows for an assessment of both models. This instrumental approach illuminates the differences between the traditional model and the new type of urban growth model which is used. Both models behave quite differently: While the Markov Cellular Automata model brings an over classification of urban growth, the LLM responds in the underestimation of urban sprawl for the same period. Both excelled with a Kappa calculation of over 89%, and showed to have fairly good estimations for the study area. One may conclude that the Markov CA Model permits a riper understanding of urban growth, but fails to analyze urban sprawl. The LLM model shares interesting results within the possibility of identifying urban sprawl patterns, and is therefore an interesting solution for some locations. Another advantage of the LLM is directly linked to the possibility of establishing probability for urban growth. Thus, while the traditional methodology shared better results, LLM can be also an interesting estimate for urban patterns from an econometric perspective. Hence further research is needed in exploring the utility of spatial econometric approaches to urban growth.