Landslide Susceptibility Mapping Based On Sampling Technology And Bayesian Spatial Logistic Regression

China is a country with mountainous and hilly,and because of its large population,there are a large number of people living in the mountains.Due to the influence of rainfall,tectonic movement,climate and other factors,landslide has become one of the most frequent and most harmful geological disasters in mountainous areas,which has caused great threats and damages to people's lives,property and living environment,and restricted the sustainable development of mountainous economy.According to the survey,the number of landslide disasters in China accounted for 68.27% of the total number of all geological disasters in 2019.Therefore,it is of great practical significance to make reliable and effective landslide prediction.Landslide susceptibility assessment is a vital basic work in disaster prevention and mitigation,which can predict the probability of occurrence of landslide in a certain region according to the topographic conditions,ecological environment,geological environment and other conditioning factors.The results of landslide susceptibility assessment can show potential areas prone to landslides,which can provide scientific guidance information for managers.While the current methods of landslide susceptible assessment have a good performance,it's still far from perfect and a lot of work to do to improve the performance of the methods.Firstly,the processing of imbalance landslide data is the basis for the calculation of landslide susceptibility assessment model,and few current studies had considered this problem in improving the performance of landslide susceptible assessment model.In addition,as the spatial data,landslide data computed in model was only considered by its attribution information,whose spatial structure information was not full used,thus limiting the improvement of the performance of the model.In view of the above problems,this paper takes Sichuan Province as the research area,adopts undersampling,over-sampling and under-over sampling methods to balance landslide data,and adopts Bayesian Spatial Logistic Regression model,considering the spatial structure information of landslide data,so as to improve the performance of landslide susceptibility assessment.The main research work and results are as follows:(1)After screening by one-variate analysis,eight condition factors,including slope,aspect,Normalized Difference Vegetation Index(NDVI),rainfall,distance from faults,distance from river,landcover and lithology,were selected as indicators of susceptibility.(2)The attribute data of each indicators extracted by fishnet is considered as the original dataset,and three balanced datasets were obtained by using under sampling,over sampling and under-over sampling techniques respectively.(3)According to the established assessment indicators,the original dataset and three balanced datasets are used to calculate the ordinary Logistic Regression model and Bayesian Spatial Logistic Regression model respectively.In Bayesian Spatial Logistic Regression model,the accuracy of under-over sampling techniques is the highest among the three resampling methods,and its AUC value is 0.859,which is 0.108 higher than that of ordinary logistic regression model.In this study area,Bayesian Spatial Logistic Regression model has better analysis ability for spatial data,which provides a reference for the application of spatial statistical model in the evaluation of landslide susceptibility.(4)Through a comparative,the result obtained from the combination of under-over sampling techniques and Bayesian Spatial Logistic Regression model was selected to mapping the landslide susceptibility in Sichuan Province.The study area was divided into four zones classified as very low(<0.25),low(0.25?0.5),moderate(0.5?0.75)and high(>0.75),and the distribution characteristics of landslide susceptibility was analyzed in Sichuan Province.