不同采样策略下地震滑坡敏感性分析研究

Susceptibility analysis of earthquake-induced landslides with different sampling strategies

针对地震滑坡规律复杂,预测难度大,而传统地震滑坡敏感性分析采样策略研究不足的问题,分析不同采样策略对地震滑坡敏感性分析的影响。以多边形缓冲区作为采样区,并将滑坡的发生区域及前两者总和2个采样策略用于对比实验。选取4·20芦山地震中受灾严重的宝盛乡作为研究对象,根据采样策略对滑坡各个评价因子进行采样,并构建支持向量机模型定量计算敏感性指数,再利用自然断点法生成敏感性区划图。通过多种方法对结果进行分析。其中,以多边形缓冲区为采样策略得到的滑坡正确率94.44%,受试者工作特征曲线下面积99.1%,种子单元面积指数综合评价效果最佳,3项结果均占优势。结果证明该采样策略是有效可行的,可为后续防灾减灾工作提供依据。

Earthquake-induced landslides had complex regularity and were hard to forecast. Traditional susceptibility techniques of earthquake-induced landslides had simple sampling method. The differences from susceptibility analysis of earthquake-induce landslides were done by applying different sampling strategies. The polygon buffer zone was selected as sampling zone, and landslides zones and the sum zones of above two zones were used as comparison. Taking Baosheng in Lushan earthquake occurred on April 20, 2013 as the research area. The factors were extracted by sampling method. Then a two-class SVM was trained for calculating susceptibility index. Finally, landslide susceptibility map was done. Statistics, receiver operating characteristic curve and seed cell area indexes were used to analyze the results. The Results of polygon buffer zone were as follows： accuracy of landslide was 94.44%, the area of receiver operating characteristic curve was 99.1%, the seed cell area indexes were best. The result showed the sampling strategy of polygon buffer zone had obvious advantage. So the sampling strategy is feasible and better, it contribute to improve the accuracy of seismic landslide susceptibility analysis.