基于多种数理模型及其融合的察雅县城滑坡易发性评价

Landslide Susceptibility Evaluation in Chaya County Based on Multiple Mathematical Models and Their Fusion

不同滑坡易发性评价模型的选择将得到差异化的空间预测结果,筛选出最优的模型是提高滑坡空间预测精度的有效方法。以西藏察雅县城滑坡为研究对象,选取了3类8个评价因子,在采用确定性系数(certainty factor,CF)、频率比(frequency ratio,FR)、信息量(information,I)和证据权(weight-of-evidence,WoE)4种单一模型评价研究区滑坡易发性的基础上,同时使用归一化融合和主成分分析(principal component analysis,PCA)融合方法对4种单一模型的评价结果进行融合,得到6组滑坡易发性区划图。研究结果表明,6种模型从低易发区到极高易发区,频率比值均呈现出显著的增长趋势,且高、极高易发性等级区内频率比值之和与总值之比均超过97%,均有效地评价了研究区滑坡的易发性。通过构建受试者工作特征(receiver operating characteristics,ROC)曲线分析评价模型的效果,显示主成分分析融合模型、证据权模型和信息量模型具有较高的曲线下面积(area under curve,AUC),综合高、极高易发区频率比值结果来看,使用PCA融合后的评价模型效果更佳。研究填补了藏东高原山区多模型对比研究的欠缺,为该地区的滑坡预测和预防提供了一定的科学依据。

The selection of different landslide susceptibility evaluation models will result in differentiated spatial prediction results,and selecting the optimal model is an effective method to improve the accuracy of landslide spatial prediction.Taking the Chaya County landslide in Tibet as the research object,8 evaluation factors from 3 categories were selected.Based on the single model evaluation using certainty factor(CF),frequency ratio(FR),information value(I),and weight-of-evidence(WoE),the evaluation results of the four single models were fused using normalization fusion and principal component analysis(PCA) fusion methods to obtain 6 landslide susceptibility zoning maps in the study area.The research results show that the frequency ratio values of the 6 models show a significant increasing trend from low susceptibility areas to extremely high susceptibility areas.The sum of frequency ratio values in extremely high and high susceptibility level zones exceeds 97% of the total frequency ratio value,effectively evaluating the susceptibility of landslides in the study area.By constructing receiver operating characteristics(ROC) curves to analyze the effectiveness of the evaluation models,it is shown that the PCA fusion model,evidence weight model,and information value model have high area under curve(AUC).Overall,the PCA fusion model performs better when considering the results of frequency ratio in high and extremely high susceptibility zones.The study fills a gap in the comparative study of multiple models in the Tibetan Plateau mountainous area and provides a scientific basis for landslide prediction and prevention in the region.