Abstract: To indicate the comprehensive features of climatic change risk sources in Qinghai-Tibet Plateau, the Self-Organizing Feature Map, an artificial neural network model, was established. This research obtained the comprehensive analysis results of 91 sites of the Qinghai-Tibet Plateau area from 1971 to 2008, using the relative intensity of annual rainfall, surface dryness index, annual snow depth, annual snow days, and annual average wind speed as clustering variables and Self-Organizing Feature Map with number of categories of 4. The conclusion can be described as four types, namely snowing leading, elements equilibrium, rainstorm leading, and dryness and average wind speed leading. Further analysis indicated that the space distribution characteristics of these types corresponded with the high value areas of single indicator analyses indeed; however, comprehensive analysis contained rich and full-scale information to a higher degree and was also closer to the real conditions.

Key words: risk of climate change, clustering analysis, self-organizing feature map network, Qinghai-Tibet Plateau

摘要： 为了反映青藏高原气候变化风险源的综合特征及其地域差异, 构建自组织特征映射人工神经网络, 选用暴雨相对强度、干燥度指数、年均积雪深度、年积雪日数以及平均风速等作为聚类指标, 运用1971?2008年青藏高原地区91个站点的风险源数据训练网络。研究结果表明: 网络聚类结果存在一定的空间规律性, 同时结合对各个指标的分析, 聚类结果在综合各类指标的同时对某一类指标还有所偏重。通过统计分析和检验, 最后将4种类型分别描述为暴雨主导型、积雪主导型、干燥度和平均风速主导型及要素均衡型。进一步的分析表明, 这些类型的空间分布特征确实与单项指标的高值区吻合, 但综合分析包含了更加丰富和全面的信息, 更加接近真实的情况。

关键词: 气候变化风险, 聚类分析, 自组织特征映射网络, 青藏高原