Abstract:

Simulation and quantitative analysis of urban land-use change dynamics are an effective way to understand the evolution of spatial structure in urban systems. Cellular automata (CA) has drawn continuous and increasing interest of researchers in the field of land use and land cover change simulation. Neighborhood rules are a core component of the urban CA model, with varied neighborhood effects among different land use combinations. Most urban CA models constructed with neighborhood rules consider only a small neighborhood scope under a specific spatial resolution. However, the extended enrichment factor indicates that there are still obvious neighborhood effects in large neighborhoods with a particularly long distance to the central cell. Based on a measured extended enrichment factor for a large neighborhood, we applied particle swarm optimization (PSO) to obtain the parameter settings of neighborhood rules, including various sub-neighborhoods at different distances within the large neighborhood. The extracted neighborhood rules were integrated into a widely used logistic regression urban CA model, Logistic-CA (LNCA), and a large neighborhood urban land use model, PSO-LNCA, was developed. Using Xiamen City as a study case, the PSO-LNCA model was implemented to simulate urban growth during the period between 1995 and 2010. The accuracy of simulated results by the model was evaluated with confusion matrix and Kappa coefficient. Accuracies for built-up land and non-built land and overall accuracy for 2010 are 83.9%, 91.7%, and 89.6%, respectively, and the Kappa coefficient for 2010 is 74.5%. The results show that the PSO-LNCA model achieved significantly higher simulation accuracy for built-up land and Kappa coefficient than the traditional urban CA model with a 3×3 kernel neighborhood (3.2% higher accuracy for built-up land and 4.5% higher for Kappa coefficient, respectively), and also generated relatively higher overall accuracy (1.8% higher). By integrating the extended neighborhood module, the simulation result generated by the PSO-LNCA model is closer to the actual space morphology and structure, compared with the traditional 3×3 kernel Logistic-CA model.

Key words: urban expansion, cellular automata, extended neighborhood effect, particle swarm optimization, Xiamen