基于机器学习和InSAR技术的滑坡易发性动态评价  

Dynamic evaluation of landslide susceptibility based on machine learning and InSAR technology

作　　者：陈志波[1,2,3] 李鼎兴 陈澄 黄卫 唐雪峰 CHEN Zhibo;LI Dingxing;CHEN Cheng;HUANG Wei;TANG Xuefeng(Zijin School of Geology and Mining,Fuzhou University,Fuzhou 350116,China;Fujian-Taiwan Science and Technology Cooperation Base of Fujian Province on Intelligent Geo-environmental Engineering,Fuzhou 350116,China;Key Laboratory of Geohazard Prevention of Hilly Mountains,Ministry of Natural Resources of China,Fuzhou 350002,China;Key Laboratory of Geohazard,Fujian Province,Fuzhou 350002,China)

机构地区：[1]福州大学紫金地质与矿业学院,福建福州350116 [2]智能环境岩土工程福建省闽台科技合作基地,福建福州350116 [3]自然资源部丘陵山地地质灾害防治重点实验室,福建福州350002 [4]福建省地质灾害重点实验室,福建福州350002

出　　处：《自然灾害学报》2025年第2期56-65,共10页Journal of Natural Disasters

基　　金：国家自然科学基金项目(52278335);福建省自然资源厅科技创新项目(KY-070000-04-2022-021,KY-070000-04-2022-026);福州大学科研启动经费(XRC-23053)。

摘　　要：东南山地丘陵区因其特殊的气候条件与地质环境导致滑坡灾害具有群发性、突发性的特点。而现有的滑坡易发性评价方法多考虑静态的评价因子,无法反映东南山地丘陵区滑坡的动态特征。因此,需结合动态因子提高评价结果的时效性。文中以东南山地丘陵区的大田县为研究区,选取坡度、坡向、地表起伏度、地层岩性、归一化植被指数以及年均降雨量6个评价因子,通过信息量(information value,IV)模型与逻辑回归(logistic regression,LR)模型、随机森林(random forest,RF)模型进行滑坡易发性建模并进行合理性检验与精度检验。利用SBAS-InSAR技术反演研究区地表形变速率,结合滑坡易发性区划结果构建动态评价矩阵实现研究区的滑坡易发性动态评价。结果表明,随机森林模型的受试者工作特性(receiver operating characteristic,ROC)曲线下面积(area under curve,AUC)值最大(0.851),可用此方法进行东南山地丘陵区的滑坡易发性分区。在已知的滑坡中,13处滑坡稳定性较差,83处滑坡稳定性一般,150处滑坡较为稳定。通过绘制稳定性较差滑坡的时间序列形变曲线发现,东南山地丘陵区影响滑坡稳定性的主要因素为季节性降雨。The southeast mountainous and hilly areas are characterized by the occurrence of landslide disasters with a high frequency and suddenness due to their unique climatic conditions and geological environment.However,the current methods for evaluating landslide susceptibility mainly focus on static factors and do not adequately capture the dynamic characteristics of landslides in the southeastern mountainous region.Therefore,it is necessary to incorporate dynamic factors in order to enhance the timeliness of evaluation results.In this paper,Datian County in the mountainous and hilly region of southeast China were selected as the research area,and six evaluation factors including slope,slope aspect,surface fluctuation,stratigraphic lithology,normalized difference vegetation index and average annual rainfall were selected.Information value(IV)model,logistic regression(LR)model and random forest(RF)model were used to model landslide susceptibility,conduct rationality test and accuracy test.The surface deformation rate in the study area was derived using the SBAS-InSAR technique.By integrating the landslide susceptibility zoning results,a dynamic evaluation matrix was constructed to achieve dynamic landslide susceptibility assessment in the study area.The results show that the receiver operating characteristic(ROC)curve area under curve(AUC)value of the random forest(RF)model is the largest(0.851),and this method can be used to partition the landslide susceptibility in the southeastern mountainous region.The major findings revealed that among the known landslides,13 locations exhibited poor stability,83 locations had moderate stability,and 150 locations were relatively stable.Through the analysis of time-series deformation curves of landslides with poor stability,it was identified that seasonal rainfall was the primary factor affecting landslide stability in the southeast mountainous and hilly areas.

关 键 词：滑坡 东南山地丘陵区 INSAR 易发性评价 机器学习方法 

分 类 号：P642.22[天文地球—工程地质学] X43[天文地球—地质矿产勘探]