Accounting For Spatial Autocorrelation Into Regional Landslide Susceptibility Assessment

Landslide refers to the movement of a large number of rocks,cuttings or sediments along a slope under the influence of gravity.As one of the most common natural geological disasters,it often causes damage to infrastructure and natural ecology,resulting in casualties and economic losses.And landslide disasters in Chongqing occur repeatedly,where is one of the landslide-prone areas in China.Therefore,it is urgent to carry out research on regional landslide susceptibility assessment to provide some decision support for regional disaster prevention and reduction.In recent years,many quantitative methods have been used to explore the relationship between landslide occurrence and impact factors.Among them,logistic regression has been widely used and reported to be applicable for landslide research.However,previous landslide studies using logistic regression for landslide prediction rarely took spatial autocorrelation into account,which may lead to model misspecification errors,manifesting as model residuals with spatial autocorrelation.Therefore,this paper attempts to introduce eigenvector spatial filtering(ESF)into logistic regression to conduct regional landslide susceptibility assessment studies that take into account spatial autocorrelation for different scale landslide data sets.The aim is to reduce the negative impact of spatial autocorrelation on the quantitative assessment of regional landslide susceptibility based on regression models,so as to achieve accurate simulation and prediction of landslide disasters.The main research contents of the paper are as follows:(1)Research on regional landslide susceptibility assessment based on the ESF-LR method.In order to eliminate or reduce the interference of spatial autocorrelation on landslide analysis,the ESF-LR method for regional landslide susceptibility assessment was proposed.Without considering the amount of calculation and taking Wulong County as the research object,the landslide predisposing factors were determined and the multi-source acquisition of the factor data and the corresponding data processing were carried out.By introducing the ESF idea into logistic regression,the ESF-LR model for regional landslide susceptibility assessment was constructed through a series of operations such as non-landslide points generation,dimensional unification of factors,multicollinearity analysis,construction of spatial connectivity matrix,computation of eigenvalues and eigenvectors,stepwise selection of eigenvectors and regression modeling.Model evaluation and comparison were carried out,and the landslide susceptibility maps of Wulong County were prepared finally.The results indicate that the ESF-LR method can effectively eliminate spatial autocorrelation of residuals and significantly improve the performance of the model.(2)Research on regional landslide susceptibility assessment based on the improved fast ESF-LR method.In order to overcome the shortcomings of the ESF-LR method due to large computational demand and expand its applicability to large landslide data sets,this paper designed the improved fast ESF-LR method for landslide susceptibility analysis.The improved fast ESF-LR method is based on the proposed improved K-Means clustering algorithm which considers both sample density and distance between samples for sampling optimization,and performs fast approximation of the eigenvectors according to the low-rank approximation idea to accelerate the eigen-decomposition process.Taking Chongqing as the study area,landslide susceptibility modeling,model evaluation and landslide susceptibility mapping were carried out.While guaranteeing algorithm efficiency,the method can greatly reduce the negative effects of spatial autocorrelation,thereby improving classification accuracy and goodness of fit of the model.(3)Development of the Landslide Prediction Emergency Web System.Based on the foregoing landslide susceptibility assessment research,the system analysis and design were carried out from the aspects of demand analysis,overall design,structural design and functional design.And taking Wulong County as an application example,the interface of the Landslide Prediction Emergency Web System and functions such as map operation,pre-disaster landslide spatial prediction,post-disaster medical planning and data statistical analysis were developed for the integration and display of the results of regional landslide susceptibility analysis.