Abstract: Simulating land use change scenarios with cellular automata (CA) can help to the policy makers in understanding the mechanisms of land change, and support the spatial decision-making for the sustainable use of land resources. Genetic algorithm (GA), an intelligent approach originally conceived from the biological process of evolution, has the capability of minimizing the difference between simulated and observed land use patterns with optimum chromosomes (i.e., feasible CA parameters) obtained through a set of selection, crossover, and mutation operations. In this paper, GA-based CA model was developed, and applied to simulate the land use change in Jiaxing City of Zhejiang Province in 1992-2008. This model was calibrated with 6% (66 samples·km^-2) and 3% (33 samples·km^-2) samplings, and the simulation results were evaluated based on confusion matrix, Kappa coefficient, and landscape metrics analysis. Over 80% of the land use features generated by the GA-based CA model matched the observed classification of land features geographically, and much higher simulation accuracy could be obtained with a larger sample. The simulation accuracy and the landscape metrics for 2001 were better than those for 2008, suggesting a tendency that the model’s accuracy decreased over the simulating process.

Key words: land use simulation, cellular automata, multi-samples, genetic algorithm, landscape assessment