基于确定系数法和信息量模型对新疆特克斯县滑坡敏感性评价的对比研究

Comparative Study on Landslide Sensitivity Assessment in Turks County of Xinjiang Based on Determined Coefficient Method and Information Quantity Model

在综合当地滑坡灾害发育特征和诱发因素基础上,选取黄土、坡度、坡向、剖面曲率、降雨等12个因子作为滑坡灾害易发性评价因子。利用确定系数法和信息量模型分别与逻辑回归结合构建了LLCF、LLSI两种模型作为滑坡敏感性评价方法。据模型计算结果,将滑坡灾害易发程度划分为极低易发区、低易发区、中易发区、高易发区和极高易发区。利用接收者操作特征曲线(ROC,Receiver Operating Characteristic Curve)及下围面积(AUC,,Area Under the ROC Curve)值分别对两种模型进行精度检验。结果表明:LLCF和LLSI模型在训练集上成功率曲线AUC值分别为0.9032和0.8411;在测试集上预测率曲线AUC值分别为0.8998、0.8715。整体上LLCF模型精度高于LLSI模型,为当地防灾减灾、国土资源等方面提供了有效的数据支持。

On the basis of synthetically considering the development characteristics and inducing factors of local landslide hazards,twelve factors,such as loess,slope gradient,slope direction,section curvature and rainfall,were selected to evaluate the susceptibility of landslide hazards.LLCF and LLSI models are constructed by combining deterministic coefficient method and information model with logistic regression respectively.According to the calculation results of the model,the vulnerability of landslide disasters can be divided into extremely low-vulnerable areas,low-vulnerable areas,medium-vulnerable areas,high-vulnerable areas and extremely high-vulnerable areas.The accuracy of the two models is tested by using ROC(Receiver Operating Characteristic Curve)and AUC(Area Under the ROC Curve)values.The results show that the AUC values of LLCF and LLSI models on the training set are 0.9032 and 0.8411 respectively,and the AUC values of predictive rate curve on the test set are 0.8998 and 0.8715 respectively.Overall,the accuracy of LLCF model is higher than that of LLSI model,which provides effective data support for local governments in disaster prevention and reduction,land and resources,etc.