An improvement on CA model of logistic regression: A case study of Guangzhou
基于Logistic回归的CA模型改进方法——以广州市为例

Due to its simple structure and less input data, CA model of logistic regression is widely applied in urban simulation. However, data dependency has some impact on the accuracy. Therefore, an in-depth research should be conducted to modify the traditional model. This paper established an improved CA model of logistic regression in two major aspects. First, the urbanization factors were divided into forbidden constraint and general constraint. The input data were sampled only in general constraint, while the urbanization probability in forbidden constraint was set to be 0. Second, we reduced the data dependency of general constraint using principal component analysis in SPSS. In the case study of Guangzhou, the improved CA model was applied to simulate the urban growth from 2000 to 2008. Compared to the traditional CA model, the improved CA model made a 4% improvement both on model fitness and simulation accuracy, in which constraints division contributed a 3% improvement on overall simulation accuracy and a 6% improvement on non-urban simulation accuracy, while data dependency reduction gave a more reasonable explanation for urbanization mechanism. The study aimed to establish an improved CA model, which can mine a more reasonable urbanization mechanism, and provide more scientific support for urban planning and land management.