基于信息量-逻辑回归模型的怒江州滑坡灾害易发性评价

Evaluation of landslide hazard susceptibility in Nujiang Prefecture based on information-logistic regression model

为了解决用单一模型进行滑坡灾害易发性评价效果不佳的问题,本文构建了耦合模型对滑坡灾害易发性进行研究。选取典型高山峡谷区怒江傈僳族自治州(简称怒江州)为研究区;选用离断层距离、离河流距离、离道路距离、工程地质岩组、海拔、坡度、坡向、土地利用类型8个滑坡影响因子构建评价指标体系;运用二元逻辑回归法、信息量法、确定性系数法分别进行滑坡灾害易发性研究,构建了信息量-逻辑回归模型,并分别与上述三种方法各自所得结果作对比;结合均方根误差、最小值和平均值统计及受试者工作特征曲线进行检验与比较。结果表明:极高和高易发区主要分布在怒江、澜沧江、独龙江沿岸及道路两侧,符合研究区实际情况,表明评价指标和评价模型的选取具有合理性;所建信息量-逻辑回归模型(0.883)比信息量(0.874)、确定性系数(0.848)、二元逻辑回归(0.869)模型更具优势。

In order to solve the problem of poor evaluation of landslide disaster susceptibility by a single model,this paper constructs a coupled model to study landslide hazard susceptibility.Nujiang Lisu Autonomous Prefecture(Nujiang Prefecture for short),a typical alpine and canyon area,was selected as the research area,and eight landslide influence factors,including distance from the fault,distance from the river,distance from the road,engineering geological rock group,elevation,slope direction and land use type,were selected to construct an evaluation index system.The landslide hazard susceptibility was studied by binary logistic regression,information value and certainty factor method respectively.The information value-logistic regression model was constructed and compared with the results obtained by the single information value method,certainty factor method and binary logistic regression method respectively.The root mean square error,minimum value,mean value statistics and receiver operating characteristic curve analysis were combined to test and compare.The results showed that the extremely high and highly prone areas were mainly distributed along the Nujiang River,Lancang River,and Dulong River and on both sides of the road,which was consistent with the actual situation of the study area,indicating that the selection of evaluation indexes and evaluation models was reasonable.The model established in this paper(0.883)has more advantages than the single information value(0.874),certainty factor(0.848),and binary logistic regression(0.869)models.