Abstract: Based on SOFM network modeling, an integrated regionalization system of cropland conversion pressures in China was established. Firstly, considering both supply and demand pressures, an index system of cropland conversion pressures was built, including 7 subsystems, and 10 factors at provincial level were selected as basic indicators. Second, through the factor analysis, 10 variables were compressed to 6 orthogonal factors. Thirdly, using these 6 factors as input variables, a SOFM neural network was set up. When the neural network was trained appropriately, it classified the data sets. As a result, the 31 provincial districts in China were classified into 9 groups, reflecting clear regional distinctions among the east, middle and west. On the basis of the classified results and such regionalization rules as all-around factors, relative consistency, regional conjugations and district integrality, the whole nation was divided into 4 regions of cropland conversion pressures and 25 sub-regions, and the result was shown as a regionalization map by GIS. At last, different management targets to sustainable cropland uses for different regions were put forward.

Key words: SOFM network, factor analysis, GIS, cropland, regionalization, China

摘要： 从资源禀赋和资源需求的角度，综合考虑粮食安全、建设占用、生态退耕和农业结构调整等对耕地资源的压力，建立了一套耕地资源压力评价指标体系，将归一化的指标数据经过因子分析处理后，基于SOFM人工神经网络模型进行无监督分类，以此为主要依据进行中国耕地压力综合分区。结果表明，中国耕地利用压力在空间上总体表现为东、中、西的区域差异，在SOFM网络分类的基础上，综合考虑综合性、相对一致性、区域共轭性、行政单元完整性等区划原则将全国分成4个耕地压力地带、25个耕地压力区，从而建立起中国耕地压力综合分区体系，并用GIS显示其空间分布。

关键词: SOFM人工神经网络, 因子分析, GIS, 耕地, 分区, 中国