Research On Modeling Of Landslide Susceptibility Evaluation Based On Hierarchical Bayesian Method

China is a mountainous country,and China has a large population,many people have to live in the mountains.Landslide is one of the most dangerous natural disasters in mountainous areas,which causes great threats and damage to the life and property of mountain people,and restricts the sustainable development of China's economy.The landslide susceptibility mapping classifies the landslide susceptibility of the region according to the characteristics of each geographical environment element,provides an important reference for regional planning decision-making,and is an important content of landslide risk management.As a means of active defense landslide,landslide susceptibility mapping is of great significance to reduce the adverse impact of landslide disaster on the production and development of human society.In the transformation and development of the natural environment of mountainous areas,human beings need accurate and rich information to reduce the potential hazards caused by landslides.Landslide susceptibility maps as an important regional planning decision data plays an important role in landslide hazards.However,although there are many current landslide susceptibility mapping methods,most of them are not reliable enough,and they can be improved.At present,the evaluation of landslide susceptibility mapping mainly faces the following problems:First,the establishment of the evaluation index system of landslide susceptibility mapping is usually influenced by the level of professional knowledge of researchers,which may increase the uncertainty of the susceptibility evaluation results.Secondly,the geographic data of landslide susceptibility mapping not only contains attribute information but also spatial structure feature information.Most of the current research only uses the attribute information of geographic data,and does not consider its spatial structure feature information.Thirdly,in the actual landslide scenario,the contribution of the control factors to the landslide is not the same in different spatial locations.However,most of the current research holds that the contribution of the control factors to the landslide is equal in different locations.Fourth,the practical applications not only need the regional scale landslide susceptibility maps to plan the decisions,but also need the local information to provide the reference for the actual engineering construction.However,the current landslide susceptibility maps do not have stable local information,and also lack the method of obtaining stable local information.In view of above problems,this paper improves the reliability and richness of the local scale information contained in the landslide susceptibility map through model integration,so as to meet the needs of practical applications.Firstly,the GeoDetector is used to eliminate the redundant control factors and establish a scientific evaluation index system.Secondly,the spatial logistic regression model is established to fully exploit and utilize the spatial structure information and attribute information of geographic data,so as to improve the reliability of the landslide susceptibility mapping results to meet the high-accuracy requirements of landslide susceptibility maps.Thirdly,under the framework hierarchical Bayesian,a reliable evaluation of local landslide susceptibility is established by integrating GeoDetector and spatial variable coefficient model,so as to fit the contribution degree of different spatial location evaluation factors to landslide,so as to enrich the connotation of landslide susceptibility map to meet the needs of practical engineering application for local information of landslide susceptibility maps.Taking the data of the Duwen Highway basin as an example,the above method was verified.Our findings are as follows:(1)The GeoDetector screened 15 potential evaluation factors,of which 7 evaluation factors have a great influence on the spatial distribution of landslide occurrence,so they are included in the evaluation index system and used as input for subsequent models.The GeoDetector can quantitatively screen the evaluation indicators and can be used as a quantitative calculation tool for the evaluation index system.(2)The spatial logistic regression model simultaneously utilized the attribute information and spatial structure information of geographic data.The AUC of the evaluation is 0.93,which is 0.14 higher than the AUC of the traditional logistic regression model.In this study area,compared with the traditional logistic regression model,the spatial logistic regression model has higher reliability for spatial data,and provides a reference method for improving the reliability of landslide susceptibility evaluation map.(3)The AUC of the local scale landslide susceptibility evaluation model integrating the GeoDetector and the spatial variable coefficient model is 0.93,which indicates that the local scale landslide susceptibility mapping has higher reliability.Local-scale landslide susceptibility mapping enriches the information of landslide susceptibility map on local scale,and provides a new solution for landslide susceptibility mapping.(4)This study aims to improve the accuracy of the landslide susceptibility map and enrich the local information of the landslide susceptibility map.Through the model integration,two high reliability methods are obtained,which provides a reference scheme for landslide susceptibility mapping.