基于SVM-LR融合模型的滑坡灾害易发性评价--以山阳县为例

Assessment of Landslide Susceptibility Based on SVM-LR Model:A Case of Shanyang County

科学地选取评价因子与构建评价模型是滑坡易发性评价结果是否准确与可靠的关键。以山阳县为例,基于区内滑坡孕灾环境、诱发因素、滑坡发育特征以及因子间相关性分析,选取高程、坡度、坡向、曲率、断层、水系、道路、降雨量、归一化植被指数(normalized difference vegetation index,NDVI)作为评价因子。分别采用逻辑回归(logistic regression,LR)模型、支持向量机(support vector machine,SVM)模型以及由以上两种模型通过简单线性回归(LN)组合形成的SVM-LR融合模型对研究区滑坡进行预测研究。预测结果表明,在SVM-LR融合模型下,滑坡点落入研究区极高-高易发区占滑坡总数的79.07%,极高-高易发区滑坡相对点密度为2.287,较其他两种模型有明显的提高。最后采用接收者操作特征(receiver operating characteristic,ROC)曲线对不同模型的预测性能进行检验,结果表明:LR模型、SVM模型、SVM-LR融合模型的成功率曲线下面积(area under the success curve,AUC)分别为0.717、0.807、0.831,预测率曲线下面积(AUC)分别为0.747、0.774、0.808,由此看出,SVM-LR融合模型的预测性能更为优越。研究结果可为山阳县当地有关部门采取防灾减灾措施提供帮助。

Scientific selection of evaluation factors and construction of evaluation models is the key to the accuracy and reliability of landslide susceptibility evaluation results.Taking Shanyang county as an example,the landslide hazard environment twas analyzed,inducing factors,landslide development characteristics and correlation among factors in the area,and elevation,slope,aspect,curvature,fault,river,road,rainfall,NDVI were selected as evaluation factors.Logistic regression(LR)model,support vector machine(SVM)model and SVM-LR stacking model formed by the combination of the above two models through simple linear regression(LN)were used respectively to predict the landslide in the study area.Finally,ROC curve was used to test the prediction performance of different models.The prediction results show that under the SVM-LR stacking model,the number of landslide points falling into the very high and high region in the study area is 79.07%,and the relative point density of landslide is 2.287,significantly higher than that of the other two models.The test results show that the area under the success curve(AUC)of the LR model,SVM model and SVM-LR stacking model are 0.747,0.774,0.808,respectively,and the area under the success curve of the LR model,SVM model and SVM-LR stacking model are 0.747,0.774,0.808,respectively.The results can provide help for the relevant departments in Shanyang county to take measures for disaster prevention and mitigation.