Modelling Urban Growth Using Geographical Cellular Automata Based on Partial Least Squares Regression

Abstract:

Based on partial least squares regression, a novel geographical cellular automata model (PLS-CA) is proposed for simulating urban growth and expansion. In definition of CA transition rules, numerous highly correlated independent spatial variables are utilized for obtaining more actual simulation results. Conventional methods, such as multi-criteria evaluation (MCE) and principal component analysis (PCA), have difficult in remove the harmful effects of correlation. Using partial least squares regression (PLS) integrated with Geo-CA and GIS, a new CA model is created for optimizing the simulation of urban growth and expansion. The PLS-CA model has been successfully applied to simulate urban growth of Jiading district, Shanghai from 1989 to 2006. And the simulation results show that the accuracy of PLS-CA is higher than that of conventional CA models.