Study On Landslide Susceptibility Assessment In Hanbin District Of Ankang City

Landslide susceptibility assessment has important theoretical and practical significance for the prevention and prediction of regional landslide disasters.This study is based on the geological hazard risk investigation in Hanbin District of Ankang City.Taking ArcGIS software as the spatial data analysis platform,the landslide susceptibility assessment in Hanbin District is studied by means of statistical analysis and mathematical modeling,and the following main conclusions and achievements are obtained.(1)The main types of landslides in Hanbin area are small and shallow accumulation layer landslides,which are mainly distributed in river valleys and low hilly areas with elevations below 1000m.Autumn is the period of high incidence of landslides.(2)Through the analysis of the influencing factors and correlation of landslides in Hanbin District,such as slope elevation,slope aspect,distance from faults,distance from rivers,normalized difference vegetation index,the rock and soil type,average annual rainfall,peak acceleration of ground motion,distance from roads and so on,are the main factors affecting the susceptibility of landslides.Taking the slope unit as the assessment unit,the information model is used to analyze the contribution of each assessmnet factor to the formation of landslide.It is concluded that landslides are most likely to occur when the elevation is less than 350m,the slope is 0-15°,the slope direction is south,the normalized difference vegetation index is 0.35-0.40,the rock and soil type is expansive soil,the annual rainfall is 950-1000mm,the peak acceleration of ground motion is 0.10,and the distance from river,faults and roads is 0-300m,0-600m and 0-600m respectively.(3)Six assessment models such as support vector machine,bagging,random forest,information-support vector machine,information-bagging and information-random forest are used to evaluate the susceptibility of landslides in Hanbin area.The results of landslide susceptibility zoning of each assessment model are reasonable under the test of landslide point density and landslide frequency ratio.The assessment accuracy of the six models is compared and analyzed by success rate and prediction rate curve and Kappa coefficient,It is concluded that the the area under the success rate curve(AUC Value)is random forest model(0.939)>bagging model(0.934)>information-bagging model(0.908),information-random forest model(0.902)>information-support vector model(0.768)>support vector machine model(0.747).The AUC values under the prediction rate curve were as follows:random forest model(0.742)>bagging model(0.734)>information-random forest model(0.732)>information-bagging model(0.728)>information-support vector machine model(0.702)>support vector machine model(0.697).The value of Kappa coefficient is random forest model(0.556)>bagging model(0.549)>information-random forest model(0.541),information-bagging model(0.519)>information-support vector model(0.466)>support vector machine model(0.444).Therefore,it is considered that among the six assessment models constructed in this study,the random forest model has the highest assessment accuracy.