基于X-多层感知器耦合模型的滑坡易发性评价——以贵州省松桃自治县为例

Landslide Susceptibility Evaluation Based on CoupledX-Multilayer Perceptron Model-a Case Study of Songtao Autonomous County of Guizhou Province,China

滑坡易发性评价是区域滑坡灾害风险评估的基础。当前主要滑坡易发性评价方法主要采用单一数据驱动模型,在实际应用中易出现漏报、误报问题。本文针对单一数据驱动模型的弊端,提出结合多层感知器(Multi-Layer Perceptron, MLP)构建耦合模型进行滑坡预测分析;选取贵州省松桃苗族自治县作为研究区,借助ArcGIS软件平台,将高程、坡度、坡向与起伏度等12个因子作为评价指标因子;采用归一化频率比(NFR)模型与信息量(I)模型对研究区进行易发性评价,再分别与MLP模型结合成为NFR-MLP、I-MLP耦合模型并开展滑坡区预测分析;将得到的易发性结果分为高、较高、中等、较低、低易发区五类;结合区划结果频率比、接受者操作特征曲线(ROC)线下面积AUC值以及新典型滑坡实例,检验模型的精确度与可靠性。结果表明:(1)精确度大小为:I-MLP耦合模型>I模型>NFR-MLP耦合模型>NFR模型。因MLP模型具备高度的容错性和鲁棒性,致使X-MLP耦合模型更加适应复杂多变的环境因素;(2)I-MLP耦合模型预测性能较为出众,相较于单一模型精度提升5.7%。本研究结果可为研究区地质灾害防治提供一定指导建议。

Landslide susceptibility evaluation is a prerequisite for regional geo-hazard risk mapping.Most past inves-tigation into landslide susceptibility tried to use a single data-driven model,which were prone to underreporting and misreporting in practical prewarning.In this study,a typical geohazard-prone area,Songtao Miao Autonomous County,Guizhou province,China was chosen to conduct a case study.A Coupled Multilayer Perceptron(MLP)model for landslide prediction was introduced to solve the drawback of a single data-driven model.It took twelve evaluation index factors including elevation,slope,aspect,and topographic relief into GIS software.A Normalized Frequency Ratio(NFR)model and an Information Value(I)model were separately used to draw a primitive delineation of susceptibility of the study area,and then they were combined with MLP model to create NFR-MLP and I-MLP coupled model to further analysis.The research area was divided into five zones:high,relatively high,medium,relatively low,and low suscep-tibility zones according to the evaluation.The accuracy and reliability of the two models were justified by combining frequency ratios of zoning results with AUC values under the receiver operating characteristic curve(ROC)line along with new typical landslide examples.We have the following findings:(1)All models could be ranked in order of accuracy:I-MLP coupled model>I model>NFR-MLP coupled model>NFR model.Because the MLP model had advantages in fault tolerance and robustness,making the X-MLP coupled model more suitable for evaluating complex and changeable geo-environment.(2)I-MLP coupled model had outstanding predictive performance,with 5.7%accuracy im-provement as compared with those from some single data-driven models.This susceptibility zoning results can pro-vide guidance for prevention and control of geological disasters in research areas.