Updating Of Regional Landslide Susceptibility Based On Bayesian Method

China is a vast country with a diverse geological and geographical environment,with mountainous areas occupying a great part,and is one of the countries in the world where landslide disasters are more frequently occurring.In recent years,a commendable progress has been achieved with regard to the evaluation of regional landslide susceptibility based on statistical methods and machine learning models,as the study of single slope stability based on failure mechanism analysis.Nevertheless,the existing regional landslide susceptibility evaluation process is laborious and complicated,less consideration is given to the influence of the newly available landslide hazard data on the stability of surrounding slopes,and the landslide hazard warning level zoning map obtained is not of the characteristics of real-time.To address the problems of regional landslide susceptibility evaluation research,this paper takes Xunwu County,Jiangxi Province,a typical mountainous and hilly region in China,as the research area to carry out regional landslide susceptibility update study.The main study contents and related conclusions are as follows:(1)The data of 337 landslide hazards in the study area from 1980～2005 were collected,and the influence of geomorphological factors,basic geological factors,hydrological environmental factors and ground cover factors affecting landslide development were analyzed.Elevation,slope,slope direction,plane curvature,profile curvature,topographic relief,distance from water system,stratigraphic lithology,vegetation cover normalized index and building normalized index were selected as basic environmental factors,and the evaluation index system of landslide susceptibility in Xunwu County was constructed.(2)By normalizing the data in the study area,support vector machine,binary logistic regression and C5.0 decision tree model were used to evaluate the landslide susceptibility of Xunwu County,and the modeling effect and prediction accuracy of different models were compared and analyzed by ROC curves,statistical accuracy and grading results,and the results showed that the C5.0 decision tree model had the best prediction effect.(3)The calculation procedure of rainfall type slope stability based on grid cells was written,and the stability of regional slopes was analyzed from the physical mechanism level,then the cohesion,internal friction angle,soil thickness,initial groundwater level,soil saturation permeability coefficient and hydraulic diffusion coefficient were selected as uncertainty parameters,and the spatial correlation among the grid cells in the study area was characterized by random field theory.(4)A batch calculation program for slope stability based on the TRIGRS model was developed,and the Bayesian update of the regional landslide uncertainty parameters in Changning town was carried out using the latest landslide hazard data between 2006 and2020 to obtain a regional landslide probability distribution map.After incorporating the latest landslide location information,the posterior mean and standard deviation of cohesion and internal friction angle are significantly reduced compared with the a priori,and the closer to the landslide location,the higher the probability of landslide in the raster cell.(5)The regional landslide susceptibility distribution map incorporating the latest landslide hazard data was obtained by coupling the landslide hazard warning map obtained based on the landslide inventory database from 1980～2005 with the latest regional landslide probability distribution map,which provides important technical guidance for disaster prevention and mitigation management departments.