摘要
广西平果市频发的地质灾害严重制约着市区的工程建设和生命财产安全。在充分收集和整理区域地质资料的基础上,通过遥感解译和现场调查,确定了平果市共发育251处斜坡类地质灾害,其中崩塌189处、滑坡62处。选择高程、坡度、坡向、曲率、工程地质岩组、距断层距离、土层厚度、距河流距离和降雨共9个因子作为评价因子,结合信息量和多层感知机分类器的优势,采用信息量和多层感知机分类器耦合模型对平果市斜坡类地质灾害进行易发性评价。斜坡类地质灾害易发性制图表明极高易发区占平果市面积的25.39%,主要分布于平果市的北部、中部和南部山区。通过ROC曲线对模型预测能力进行检验获得AUC=0.809,表明模型评价结果能够很好地预测研究区斜坡类地质灾害的发生。研究结果可为研究区的崩滑灾害风险评价和灾害防治提供科学依据。
Frequent geological hazards in Pingguo City,Guangxi Province,seriously restrict the engineering construction in this area.Through the remote sensing interpretation and field investigation,251 hillslope geological hazards,including 189 collapses and 62 landslides in Pingguo City are determined.All hazard points are randomly split into two subsets with 80%(200)and 20%(51)respectively for training and validation purposes.In addition,an equal number of non-hazard points were randomly selected from the hazard-free areas and then divided into a training dataset and a validation dataset.The hazard point is denoted as 1 and the non-hazard point is denoted as 0.Then,based on the geological environment conditions of Pingguo City,nine factors including elevation,slope angle,slope aspect,curvature,engineering geology groups,distance from faults,soil thickness,distance from rivers and rainfall are selected as hazard evaluation factors.Elevation,slope angle,slope aspect and curvature are topographic factors,which are produced from Advanced Land Observing Satellite(ALOS)digital surface model(DSM)with a resolution of 12.5 m.Distance from rivers is one of hydrological variables,also produced from the ALOS DSM.Engineering geology groups and distance from faults are geological factors,which are produced from a geological map at a scale of 1:200,000 provided by the China Geological Survey.Soil thickness is one of the factors related to land use,which is derived from domestic high resolutions remote sensing data such as Gaofen No.2 and Beijing No.2.Rainfall is an environmental factor,which is collected from the China Meteorological Administration.All relevant factors are converted into raster format with a 12.5-m resolution of ArcGIS software.Among these nine factors,engineering geology group is a categorical factor,whereas the other factors are continuous factors.In this study,continuous factors are reclassified into categorical factors based on natural breaks or equal interval methods.The tolerance(TOL)and variance inflation factor(VIF)are used to detect multicollinearity of all the factors.Generally,a VIF of less than 10 or a TOL of more than 0.1 indicates that all the factors are independent.Then,combined with the advantages of information value(IV)and multi-layer perceptron classifier(MLPC),the coupling model of IV and MLPC is used to evaluate the susceptibility of hillslope geo-hazard in Pingguo City,and the results are classified into five classes:very low,low,moderate,high,and very high.Finally,the receiver operating characteristic curve(ROC)and the area under the ROC curve are applied to carry out accuracy verification of the model.The main conclusions include:(i)Rainfall has the highest VIF value(1.163)and the lowest TOL value(0.86),both of which do not exceed the critical threshold values(TOL>0.1 or VIF<10).Thus,there are no severe collinearity problems among the 9 conditioning factors.(ii)The information values of the factor classes show that certain classes are conducive to the occurrence of hazards.The factor classes mostly prone to the occurrence of hazards are the elevation with 250-350 m,slope angle with 40°-50°,south slope,curvature less than-0.05,limestone and dolomite,0-500 m from the faults,soil thickness with 0-1m,the area 0-250 m from the rivers and rainfall with 1,373-1,428 mm.(iii)The zoning map of hillslope geo-hazard susceptibility in Pingguo City shows that the areas with very high,high,moderate,low and very low susceptibility respectively account for 25.39%,3.49%,4.34%,30.54%and 36.24%of the total area of the city.The areas with very high susceptibility are mainly distributed in the junction of steep mountains and plains,such as Liming town,Tonglao town and Bangwei town in the north of the study area,Haicheng town,Tonglao town and Pozao town in the middle and Xin'an town in the south.The hillslope geo-hazards in these five susceptible areas cover proportions of 64.54%,8.37%,1.59%,21.91%,and 3.59%,respectively,which indicates that the evaluation results of hazard susceptibility are highly consistent with the distribution of historical collapse and landslides,and the mapping results of hazard susceptibility are reliable and accurate.(iv)The frequency ratio of each susceptibility zone can be obtained by calculating the ratio of hazard point proportion and area proportion of each susceptibility class.The frequency ratio increases from the very low susceptibility area to the very high susceptibility area,which indicates that the coupling model can effectively determine the different susceptibility levels in the study area.(v)The performance of the model is comprehensively compared through the AUC and ROC curves with validation datasets.The AUC value for the IV-MLPC model is 0.809,showing that the model exhibits a satisfactory performance(AUC>0.8)in collapse and landslide spatial prediction.The research results can provide a scientific basis for the risk assessment and disaster prevention of collapse and landslide in the study area.
作者
王新伟
张漓黎
莫德科
叶宗达
江凡
WANG Xinwei;ZHANG Lili;MO Deke;YE Zongda;JIANG Fan(China Nonferrous Metals(Guilin)Geology and Mining Co.,Ltd.,Guilin,Guangxi 541000,China;Technical Innovation Center for Restoration Engineering of Mine Geological Environment in Karst Areas of South China,Ministry of Natural Resources,Nanning,Guangxi 530028,China;Guilin Survey Institute of Hydrogeological Engineering Co.,Ltd.,Guilin,Guangxi 541000,China;No.274 Geological Team of Guangxi,Beihai,Guangxi 536005,China)
出处
《中国岩溶》
CAS
CSCD
北大核心
2023年第2期370-381,共12页
Carsologica Sinica
基金
桂林市2018年市本级可持续发展重大专项项目“典型废弃采石场生态修复及综合利用关键技术集成与示范”(合同编号:20180101-2)
国家重点研发计划课题“漓江流域喀斯特自然景观修复与植被生态功能提升关键技术研发及试验示范(合同编号:2019YFC0507503)”专题“漓江流域典型喀斯特废弃采石场景观修复技术与示范”。
关键词
平果市
斜坡类地质灾害
信息量
多层感知机分类器
易发性分区
Pingguo City
hillslope geo-hazard
information value
multi-layer perceptron classifier
susceptibility zoning
