基于随机森林模型的藏东南地区滑坡易发性评价及主控因素分析  

Evaluation of landslide susceptibility in southeast Tibet based on a random forest model

作　　者：杜鹏 陈宁生[1,3] 伍康林 李志 张瀛玉龙 DU Peng;CHEN Ningsheng;WU Kanglin;LI Zhi;ZHANG Yingyulong(Institute of Mountain Hazards and Environment,Chinese Academy of Sciences,Ministry of Water Resources,Chengdu 610041,China;College of Engineering,Tibet University,Lhasa 850000,China;Academy of Plateau Science and Sustainability,Xining 810016,China;College of Engineering Sciences,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院、水利部成都山地灾害与环境研究所,成都610041 [2]西藏大学工学院,拉萨850000 [3]高原科学与可持续发展研究院,西宁810016 [4]中国科学院大学工程科学学院,北京100049

出　　处：《成都理工大学学报（自然科学版）》2024年第2期328-344,共17页Journal of Chengdu University of Technology: Science & Technology Edition

基　　金：国家自然科学基金—国际(地区)合作与交流项目(42361144880);国家重点研发计划(2023YFC3008301);中国电建核心技术攻关“揭榜挂帅”项目(DJ-HXGG-2022-02);中国科学院加德满都科教中心资助项目(131C11KYSB20200033);西藏自治区重大科技专项(XZ202101ZD0013G)。

摘　　要：为了获取藏东南地区的滑坡易发性分区、深入探究研究区滑坡发育的主控因素。本文采用随机森林模型对藏东南地区的滑坡易发性进行研究。首先通过现场调查、遥感解译和文献查询等手段,共发现研究区滑坡306处,其次使用频率比模型(FR)对研究区易发性做快速评判,在其低易发区进行非滑坡样本的选取,使得随机森林模型得到较好的原始数据集,并通过计算权重来制作研究区易发性图,分析滑坡发生的主控因素,最后绘制ROC曲线和计算线下曲线的面积(AUC)对模型的准确性进行验证。模型预测结果显示,研究区高易发区主要集中在易贡藏布、帕隆藏布交汇处和雅鲁藏布江大拐弯的地方。在对特征重要性因素排序中,前三分别是断层密度、高程和坡度,模型的AUC=0.940,因此有较好的准确性。根据结果分析,研究区因长期受到构造作用影响,地质构造复杂,东西两侧均是巨型断层,南侧为喜马拉雅前缘断裂带,断裂带发育,尤其在高易发区断层纵横交错,多重挤压造成岩石破碎,节理发育。因此,易贡藏布、帕隆藏布交汇处和雅鲁藏布江大拐弯区域为滑坡灾害高易发区具有合理性。该模型对藏东南地区滑坡的发生和灾害的治理提供了理论依据,对该地区防灾减灾提供了理论支撑。In this study,we investigated the main factors influencing the development of landslides in southeast Tibet by using the random forest model to determine the susceptibility of the area to them.We identified 306 landslides in the area through field investigations,the interpretation of remote sensing data,and the literature.Moreover,we used the frequency ratio model to quickly determine the susceptibility of the area to landslides,and chose non-landslide samples for regions with a low susceptibility to them to provide a suitable dataset to train the random forest model.The weights of the parameters were calculated to generate a map of the susceptibility of the area to landslides to analyze the main factors controlling their occurrence.The accuracy of the proposed model was verified by plotting its ROC curve and calculating the AUC of the offline curve.The results of predictions by the model showed that highly susceptible areas were mainly concentrated at the confluence of Yigong Zangbo and Palong Zangbo,and in the large bend of Yarlung Zangbo River.The density of faults,elevation,and slope were,respectively,the three most important features with regard to the susceptibility to landslides,and the AUC of 0.940 of the model showed that it was accurate.Our analysis showed that the study area had a complex geological structure owing to the long-term influence of tectonic action,with large faults on the east and west sides,and the Himalayan frontal fault zone on the south side.With the development of the fault zones,especially in the region that was highly susceptible to landslides,the faults intersected,and multiple extrusions led to the fragmentation of rocks and the development of joints.Therefore,the intersection of Yigong Zangbo and Palong Zangbo as well as the region containing the large bend of Yarlung Zangbo River were highly susceptible to landslides.The proposed model provided a theoretical basis for accurately predicting landslides and appropriately managing disasters in southeast Tibet.

关 键 词：藏东南地区 随机森林模型 滑坡易发性 断层密度 

分 类 号：P642.22[天文地球—工程地质学]