Distribution Law And Risk Assessment Of Landslide Triggered By Typhoon And Rainstorm

Landslide disaster is one of the most frequent and harmful disasters in the world,according to statistics,90%of landslides are induced by rainfall.Typhoons enter southeast China every year.Rainstorms are often accompanied by Typhoons,which trigger a large number of landslide and debris flow geological disasters.In 2019,the heavy rainfall brought by the strong typhoon "Lichema" triggered many geological disasters in Ningguo City,Anhui Province,most of which were landslides.Therefore,the statistical analysis of rainfall landslides in the study area,the establishment of landslide database,and the study of landslide distribution law and regional landslide susceptibility are conducive to the prevention and control of local landslide disasters and the protection of people's lives and property.Taking Ningguo City as the research area,this paper compares the satellite images before and after the typhoon,visual interpretation the landslides during the typhoon and rainstorm,and establishes the landslide database of Ningguo City.On this basis,the influence of various influencing factors on landslide is analyzed,the spatial distribution law of landslide is obtained,and the landslide susceptibility zoning in the study area is carried out by logical regression model and analytic hierarchy process.Based on Bayesian thought and machine learning model,the probability of landslide under different rainfall is calculated,and the regional landslide risk zoning map is drawn.The main contents are as follows:(1)Establishment of landslide database based on visual interpretation of satellite images.According to the 3M resolution planet satellite images before and after typhoon rainstorm,414 landslides were obtained in the study area by artificial visual interpretation method.The landslide area is classified,combined with the spatial distribution of landslide,the spatial distribution characteristics of "many points,small area and wide distribution" of landslide are obtained.On this basis,the landslide disaster database in the study area is constructed to analyze the relationship between the spatial distribution of landslide and influencing factors.(2)Preliminary treatment of landslide susceptibility.In this paper,the grid unit with wide applicability and high precision is selected as the evaluation unit,and the grid size is 30m×30m.According to the previous literature research results and the comprehensive situation of the study area,the relationship between landslide and various influencing factors is studied:Elevation,slope,slope direction,distance from fault,stratum lithology,distance from water system,distance from road,Normalized Difference Vegetation Index(NDVI)and rainfall.Reclassify each impact factor and make collinearity analysis,the results show that the variance expansion factor(VIF)is less than 10,which meets the requirements of subsequent logistic regression models.(3)Logistic Regression(LR)model and(Analytic Hierarchy Process(AHP)combined with ArcGIS software are used to evaluate and verify the landslide susceptibility in the study area.The weight value of each factor is obtained by logistic regression model and analytic hierarchy process,the susceptibility distribution map of the study area is obtained by overlaying layer operation with grid calculator in ArcGIS.The consistency test and rationality test were used to verify the AHP model,and the rationality test and Receiver Operating Characteristic curve(ROC)curve were used to verify the LR model,both have high accuracy.According to the susceptibility zoning map,the study is divided into:extremely high risk area,high risk area,medium risk area,low risk area and very low risk area.The results are in good agreement with the actual situation.(4)Based on Bayesian method and machine learning model,the distribution map of real landslide probability in the study area is obtained,and the landslide probability under different rainfall is discussed.Samples are selected based on Bayesian method,and points are taken evenly in the study area.50 points are taken every 1km2,with a total of 150100 points,the samples falling into the landslide area are landslide samples,and the other samples are non landslide sample points,based on the machine learning model,the real probability of landslide occurrence in the study area is obtained and passed the ROC curve test,AUC(area under curve)is 0.886,so the model is very reliable.Based on this method,the real probability of landslide occurrence in the study area under different rainfall conditions is calculated,and the landslide risk zoning map is drawn.Fig.48 Table 14 Ref.88...