A particle swarm intelligence based cellular model for urban morphology evolution modelling : A case study in Jiading District of Shanghai

As a complex non-linear and dynamic process,full understanding of the urban morphology and evolution mechanism requires modelling.Due to its abilities of simulating and predicting a complex system,cellular automata (CA) have been increasingly used to capture the nature of urban evolution since the pioneering work of Tobler.More recently,intelligence methods were widely adopted to optimize geographical CA models.The similarity between the nature of self-organization of particle swarm optimizers (PSO) and the "bottom-up" approach of cellular models makes it particularly suitable for optimizing transition rules.Based on automatically searching the minimum differences between the simulation results produced by a CA model based on conventional logistic regression method and the actual pattern of urban morphology,this research integrates the PSO method and a CA model to stochastically optimize combination of parameters of CA rules and construct the PSO based CA model for urban expansion and evolution modelling.Based on Matlab,Visual Studio.Net and GIS,the PSO algorithm and the PSO-CA model were successfully implemented.By using the 17 years (from 1989 to 2006) historical remotely sensed images,the PSO-CA model was calibrated to simulate the urban expansion of Jiading District,Shanghai Municipality.Besides,the urban pattern of Jiading District at 2010 was projected with the PSO-CA model.Evaluated with a confusion matrix,the simulation results of the PSO-CA model obtained accuracies of 87.42% for the non-urban category,76.51% for the urban category,82.35% for overall,and 64.31% for the Kappa coefficient,which outperforms the logistic regression based CA model,with accuracies of 84.36% for the non-urban category,71.58% for the urban category,78.42% for overall,and 56.32% for the Kappa coefficient.This research have demonstrated that conventional transition rules were substantially improved by the PSO technique,which also can optimize a wide range of traditional CA models for urban evolution modelling