Abstract: The authors present an analysis of how scale issues affect a cellular automata model of land use change developed for a research area in Longhua Town, Shenzhen City. The scale dependence of the model is explored by varying the resolution of the input data in 1990 used to calibrate the model and changing the length of model simulating time. To explore the impact of these scale relationships the model is run with input datasets constructed at the following spatial resolutions: 30, 60, 90, 120, 150, 180, 210 and 240 m for simulating land use in 1995 and 2000. Three kinds of indicator, i.e. point by point accuracy, Kappa and real change accuracy are used to assess the scale sensitivity of the model. The results show that 1) the more fine the cell sizes are, the higher the accuracy of the simulation results; 2) path dependence of the isolated cells is an important source of the spatial scale sensitivity of CA model; 3) the specific geographical process in different periods of time is an important source of the temporal sensitivity scale of CA model. The results have great significance for the scale selection of CA model.

Key words: cellular automata, LUCC, scale, sensitive analysis, neural network, Shenzhen

摘要： 以深圳市龙华镇为案例区, 构建了土地利用/覆被变化的元胞自动机模型, 从时间和空间两个方面定量研究了LUCC模型的尺度效应。通过改变模型输入数据的空间分辨率和模型模拟的时间长度, 探讨了尺度对土地利用变化模型的影响。分别采用龙华镇1990年30, 60, 90, 120, 150, 180, 210和240 m空间分辨率的土地利用数据作为元胞自动机模型的输入, 模拟研究区1995年和2000年的土地利用变化状况以诠释CA模型内在的尺度依赖特征, 并依据模型的点对点模拟精度、Kappa系数、实际变化元胞的模拟精度3个指标评价了该模型的尺度敏感性, 最后分析了CA模型尺度敏感性的来源。结果表明: 空间尺度上, 元胞的空间分辨 率越高, 模型的模拟精度越高, 时间尺度上, 模型模拟的实际时间跨度越长, 精度越高; 孤立元胞的路径依赖是CA模型空间尺度敏感性的重要来源; 特定地理过程在不同时间段的发展规律是CA模型时间尺度敏感性的重要来源。研究结果对元胞自动机的尺度选择具有重要意义。

关键词: 元胞自动机, LUCC, 尺度, 敏感性, 人工神经网络, 深圳