基于多种机器学习模型的灵台县地质灾害易发性评估  

Geological hazard susceptibility assessment in Lingtai County based on multiple machine learning models

作　　者：安亚鹏 汪霞[1] 张芮 刘兴荣[3] 张国信 唐家凯[1,3] 周自强[3] AN Ya-peng;WANG Xia;ZHANG Rui;LIU Xing-rong;ZHANG Guo-xin;TANG Jia-kai;ZHOU Zi-qiang(College of Earth and Environmental Sciences,Lanzhou University,Lanzhou 730000,China;College of Water Conservancy and Hydropower Engineering,Gansu Agricultural University,Lanzhou 730070,China;Geological Hazards Prevention Institute,Gansu Academy of Science,Lanzhou 730000,China)

机构地区：[1]兰州大学资源环境学院,兰州730000 [2]甘肃农业大学水利水电工程学院,兰州730070 [3]甘肃省科学院地质自然灾害防治研究所,兰州730000

出　　处：《兰州大学学报(自然科学版)》2025年第1期84-90,98,共8页Journal of Lanzhou University(Natural Sciences)

基　　金：甘肃省科技重大专项计划项目(23ZDFA009);甘肃省省级重点人才项目;甘肃省科学院重点科技研发项目(2023ZD YF-03);甘肃省科学院科技产业化项目(2021CY09);甘肃省科技厅重大专项项目(2022-0103-SFC-0068)。

摘　　要：以灵台县为例,选取高程、坡度、坡向、地形起伏度、地层岩性、归一化植被指数、多年平均降雨量、河流缓冲区、土地利用和道路缓冲区10个因素作为地质灾害评价因子,利用逻辑回归(LR)、决策树(DT)和随机森林(RF)机器学习模型进行地质灾害易发性评价,用受试者工作特征曲线进行模型预测精度评价.结果表明,RF模型的地质灾害极高易发区包含57.32%的地质灾害点,高于LR和DT模型的54.88%和48.78%;RF模型的地质灾害点密度为0.47处/km^(2),高于LR和DT模型,表明RF模型在预测成功率上高于LR和DT模型.RF模型评价结果的曲线下面积为0.883,优于LR和DT模型,其中极高易发区和高易发区面积占比分别为4.9%和13.8%.We taking Lingtai County as an example,selected 10 elements,including elevation,slope,slope direction,topographic relief,stratigraphic lithology,normalized vegetation index,average annual rainfall,river buffer zone,land use and road buffer zone,as geological hazard assessment factors,and used three machine learning models,namely logistic regression(LR),decision tree(DT)and random forest(RF),to evaluate geological hazard susceptibility.The accuracy of model prediction was evaluated via the receiver operating characteristic curve,and the results showed that 57.32%of the geological damage points were in the highly prone area of RF model,which was higher than 54.88%and 48.78%of LR and DT models.The density of geological damage points of RF model was 0.47/km^(2),also higher than that of LR and DT models,indicating that the prediction success rate of the former being higher than that of the latters.The area under the curve of RF model was 0.883,which was better than LR and DT models.In the study area,the percentage of extremely prone area and high prone area was 4.9%and 13.8%,respectively.

关 键 词：易发性评价 逻辑回归模型 决策树模型 随机森林模型 受试者工作特征曲线 

分 类 号：P694[天文地球—地质学]