Kernel-Based Cellular Automata for Urban Simulation

Cellular automata (CA) can be used to simulate complex urban systems. Calibration of CA is essential for producing realistic urban patterns. A common calibration procedure is based on linear regression methods, such as multicriteria evaluation. This paper proposes a new method to acquire nonlinear transition rules of CA by using the techniques of kernel-based learning machines. The kernel-based approach transforms complex nonlinear problems to simple linear problems through the mapping on an implicit high-dimensional feature space for extracting transition rules. This method has been applied to the simulation of urban expansion in the fast growing city, Guangzhou. Comparisons indicate that more reliable simulation results can be generated by using this kernel-based method.