不完备样本条件下基于支持向量回归模型的滑坡易发性评价

区域滑坡易发性评价对灾害中长期预测预报具有重要意义,在基于统计模型进行评价过程中,样本选取对评价结果有较大影响,构建较稳健的、受样本数量影响小的分析模型非常重要。本文以马来西亚热带雨林地区为例,选择坡度、坡向、地表曲率、地貌类型、岩性、构造、土地覆盖、道路和排水系统等9大要素作为评价因子,结合支持向量回归(SVR)模型计算研究区滑坡易发性指数,并探讨不完备样本条件下易发性评价方法,分析样本数量和评价精度之间的关系。结果显示,基于SVR模型进行该区滑坡易发性分析评价,其成功率验证法的描述精度约为95.9%;同时,样本数量的增减对分析精度影响较小;SVR方法是一种适于热带雨林地区高植被覆盖条件下的分析模型,可为今后同类地区的滑坡灾害管理工作提供支持。

null

Landslide hazard susceptibility relates to middle-and long-term predicting and forecasting,and it is very important to landslide managements.In the process of evaluation based on statistical model,the result is greatly influenced by landslide sample size,so the more conservative and less influencing model must be applied to the susceptibility evaluation in order to reduce the system error.The study area is located in Malaysia tropical rainforest,where nine factors were selected as topographic slope,aspect,surface curvature,geomorphology,lithology,structure,land cover,road and drainage and so on.The Landslide Hazard Susceptibility Index(LSI) was constructed based on support vector regression(SVR) theory,then the susceptibility evaluation methodology was discussed in incomplete sample conditions,and the relation between the sample size and the result accuracy was analysed too.The result show that the success-rate analysis accuracy based on SVR model was about 95.9%,an obviously high value;the fluctuation of sample size influenced the accuracy slightly;SVR was a better model suited to landslide hazard evaluation in high vegetation cover conditions,which could provide a technique support for landslide management in similar areas.