关键词: 空间直观模拟, 元胞自动机, 元胞邻域, 景观分析, 土地利用情景

Abstract: As one of the spatially-explicit simulation models, geographical cellular automata (Geo-CA) are able to simulate and project the scenarios of urban expansion and land use patterns. In the geographic simulation, neighbor configurations remarkably impact the mining of transition rules and the accuracy of simulated results. It is recognized that the impact of neighbors on results could be quantitatively analyzed in terms of key factors such as iteration processing and accuracy of simulation, landscape pattern, and model performances. With logistic regression based CA model, and Von Neumann and three types Moore neighbors respectively, the land use changes of Baoshan District, Shanghai from 1992 to 2008, were simulated. It is demonstrated that there are remarkable differences on attribute and spatial patterns between the simulation results produced by different cell neighbors. The number of newly added cells and the simulation accuracies are decreasing with the simulating process. However, the results generated by Moore 5×5 neighbors have higher accuracy than the results generated by Von Neumann, Moore 3×3, and Moore 7×7 neighbors. As for the landscape pattern, results produced by Moore 3/5 neighbors match with actual pattern obtained from classification maps which is better than those by Von and Moore7 neighbors. Besides, the requirement of computation increases and model performances decreases with the increase of number of neighbors. This study showed that it is appropriate to simulate the land use changes with the Moore3/5 neighbor configurations.

Key words: spatially-explicit modeling, cellular automata, cell neighbors, landscape analysis, scenarios of land use