A spatial error-based cellular automata approach to reproducing and projecting dynamic urban expansion

Urban systems are featured by spatial autocorrelation, which may produce clustering of model residuals when simulating urban expansion using cellular automata (CA). Accurate identification of spatial autocorrelation and reduction of residual clustering are essential to accurate CA modeling of urban expansion. We developed a new CA approach (CA(SEM)) using a spatial error model (SEM) that incorporates spatial autocorrelation. Using Zhengzhou City as a case study, we calibrated three types of CA models [e.g., logistic regression (Logit), spatial lag model (SLM) and SEM] from 2000 to 2010. Here, two important issues are the choice of the appropriate method (SLM vs. SEM) for urban expansion modeling and the applicability of CA(SEM) for projecting urban scenarios. We validated the CA(SEM) model from 2010 to 2017 and projected urban scenarios out to the year 2030 using this model. End-state assessment reveals that CA(SEM) yields a higher overall accuracy (91.4%) in the calibration, but lower overall accuracy (83.8%) in the validation. For change assessment, CA(SEM) yields a lower figure-of-merit (FOM; 31.8%) in the calibration but a higher FOM (35.2%) in the validation. We conclude that CA(SEM) can accurately simulate urban expansion at Zhengzhou considering the fit performance of urban land transition rules, and the accuracy assessment of urban patterns and expansion. Scenario prediction using CA(SEM) is therefore valuable for formulating useful urban planning regulations and in supporting sustainable urban development.