乡镇滑坡灾害精细化易发性评价——以万州区熊家镇为例  

Refined susceptibility assessment of landslides in township——Taking Xiongjia Township, Wanzhou District as an example

作　　者：黄倩雯 刘庆丽 杜娟 吴名华[1] 黄肖萍 钱未 HUANG Qianwen;LIU Qingli;DU Juan;WU Minghua;HUANG Xiaoping;QIAN Wei(Badong National Observation and Research Station of Geohazards,China University of Geosciences(Wuhan),Wuhan 430o74,China;Geological Enoironment Monitoring Station of Wanzhou District,Chongqing 404100,China;School of Environmental Studies,China University of Geosciences(Wuhan),Wuhan 430078,China;Centre for Severe Weather and Climate and Hydro-Geological Hazards,China University of Geosciences(Wuhan),Wuhan 430078,China;Jiangci Tianjiu Geological and Mineral Construction Group Co.,Ltd.,Yingtan 335000,China)

机构地区：[1]中国地质大学(武汉)湖北巴东地质灾害国家野外科学观测研究站,湖北武汉430074 [2]重庆市万州区地质环境监测站,重庆404100 [3]中国地质大学(武汉)环境学院,湖北武汉430078 [4]中国地质大学(武汉)极端天气气候与水文地质灾害研究中心,湖北武汉430078 [5]江西省天久地矿建设集团有限公司,江西鹰潭335000

出　　处：《安全与环境工程》2024年第3期188-197,216,共11页Safety and Environmental Engineering

基　　金：国家自然科学基金面上项目(42172318)。

摘　　要：地质灾害精细化易发性评价是乡镇尺度地质灾害风险防控的基础,探究适应于乡镇级评价精度需求和精细化调查数据的易发性评价方法成为研究的重点。以万州区熊家镇为研究区,在精细化地质灾害调查的基础上,将滑坡灾害细分为厚层堆积层滑坡和浅层堆积层滑坡,分别对其进行斜坡单元划分,并基于灾害的成因机制构建差异化的易发性评价指标体系;针对乡镇尺度评价区灾害样本数量少、难以采用统计和机器学习类评价方法的问题,引入多分类逻辑回归模型进行小样本区的易发性评价;综合滑坡单元、非滑坡单元和存在一定变形、但尚未形成整体滑移的不稳定斜坡单元,突破传统的0、1二元变量,构建包含“中间态”变量的易发性评价灾害样本集,从而实现对灾害样本的有效扩充和对不同程度易发性的量化评价。结果表明:浅层堆积层滑坡高和极高易发区主要位于研究区北部顺向坡以及中部、南部的人类活动集中区域,且70.59%的调查变形点位于高和极高易发区;厚层堆积层滑坡极高和高易发区主要分布于西部坡度较缓的河流两岸,评价结果精度AUC值为0.823。本研究对乡镇及其同等尺度地质灾害风险调查与评价具有参考价值。The study of landslide susceptibility at the township is of great significance for subsequent local hazard risks prevention and control.Exploring the susceptibility assessment method suitable for the accuracy requirements of township-level assessment and refined survey data has become the focus of research.Xiongjia Township in Wanzhou District is taken as the study area.On the basis of refined geological hazard investigation, the landslide hazards are subdivided into deep-seated colluvial landslides and shallow colluvial landslides, and the slope units are divided respectively.Based on the landslide mechanisms, different susceptibility assessment indexes are proposed.Aiming at the challenge of insufficient landslide samples in the township scale area and the difficulty of using statistical and machine learning assessment methods, a multi classification logistic regression model is introduced to assess the susceptibility of study area with insufficient samples.Combining landslide units, non-landslide units and unstable slope units with certain deformation but no overall slip, the traditional binary variables of 0 and 1 are broken through, and the sample set including “intermediate state” variables is constructed, so as to realize the effective expansion of landslide samples and quantitative assessment of different degrees of susceptibility.The results show that the extremely high and high susceptibility areas of shallow colluvial landslides are mainly located in the north of the study area and the central and southern areas with intense human activities, and 70.59% of the defor-mation slopes are located in high-extremely high susceptibility areas.The extremely high and high susceptibility areas of deep-seated colluvial landslide are mainly distributed on both sides of the river with gentle slope in the west, and the accuracy AUC of the evaluation results is 0.823.This study has reference value for the investigation and assessment of landslide risk in township scale and the equivalent scales.

关 键 词：厚层堆积层滑坡 浅层堆积层滑坡 易发性评价 多分类逻辑回归模型 

分 类 号：X43[环境科学与工程—灾害防治] P642.22[天文地球—工程地质学]