基于不同因子分级的滑坡易发性区划对比--以万州区为例

Comparison of Landslide Susceptibility Mapping Based on Different Factor Classifications:Taking Wanzhou District as an Example

【目的】基于滑坡点属性与研究区全域点属性作为分类基础数据,对位于三峡库区的万州区滑坡易发性区划对比研究。【方法】选取高程、多年平均降雨量、地表切割深度、坡向、距道路距离、坡度、POI核密度、倾坡类型、岩性、归一化植被指数、曲率、地形起伏度、地形湿度指数等13个因子作为影响因子,分别对滑坡点属性与研究区全域点属性使用自然断点法进行因子分类,并通过两种分类结果构建模型预测研究区内滑坡易发程度的空间分布情况。【结果】以研究区全域点属性作为分类数据对应的曲线下面积(Area under curve,AUC)值为0.79,对应的模型测试集最高精度为0.755;以滑坡点属性作为分类数据对应的AUC值为0.85,对应的模型测试集最高精度为0.779;高程、地表切割深度和多年平均降雨量的重要性位于前3位,对滑坡的发生有较大影响。【结论】采用滑坡点属性确定因子分类标准建立的随机森林模型精确性和预测能力更优,可对万州区的滑坡灾害危险管理和影响因子分级研究提供参考。

[Purposes]Based on the attributes of the landslide point and the attributes of the entire study area as the basic data for classification,a comparative study of landslide sensitivity in Wanzhou district located in Three Gorges reservoir area is carried out,and a reference for the classification study of the influencing factors of landslide formation is provided.[Methods]Select 13 items including elevation,annual average rainfall,and others are used as landslide hazards as impact factors.The natural breaks classification used to classify the attributes of the landslide point and the whole area of the study area respectively,and the two classification results are used to construct a model to predict the spatial distribution of landslide susceptibility in the study area.[Findings]The area under curve(AUC)value corresponding to the classification data of the study area is 0.79,and the highest accuracy of the corresponding model test set is 0.755;the AUC value corresponding to the landslide point attribute as the classification data is 0.85,the highest accuracy of the corresponding model test set is 0.779;the importance of elevation,surface cutting depth,and annual average rainfall are in the top 3,which has a greater impact on the occurrence of landslides.[Conclusions]The random forest model established by the classification standard of landslide point attribute determination factors has good accuracy and predictive ability,and can provide reference for landslide disaster risk management and impact factor classification research in Wanzhou district.