Landslide Susceptibility Assessment Based On Multi-scale Geographically Weighted Regression

Landslides are one of the most major and destructive geo-hazards in the world,characterized by wide distribution,high frequency,great engineering damage,enormous loss and strong suddenness.Landslide susceptibility assessment is an important part of landslide geo-hazard risk management,which is essentially the category of geospatial data research.Therefore,from the perspective of spatial non-stationary modeling based on Multi-scale Geographically Weighted Regression(MGWR),this study conducts the landslide susceptibility assessment.The main research contents and conclusions are as follows:(1)Firstly,three spatial proximity expressions are constructed,which are based on grid-unit neighborhood,hexagonal neighborhood and slope-unit neighborhood,respectively.Using the segmentation metric function F value and the root mean square error to compare the neighborhood,choose the hexagonal neighborhood to express the spatial proximity.In addition,the hexagonal neighborhood is used to extract geographic information,thereby conducting spatial pattern analysis of historical landslides in the study area through exploratory spatial data analysis methods.It is obtained that it has a positive spatial clustering pattern and a trend of northeast-southwest distribution.(2)The landslide susceptibility assessment factors is selected,including elevation,slope,slope aspect,terrain relief,lithology,rainfall,distance to fault zones,distance to stream,distance to settlement,and vegetation coverage type.Factor analysis shows that there is a correlation within the assessment indexes.Moran’s index analysis was performed on the residual values obtained by the global regression,indicating that the global regression was invalid.(3)Based on spatial non-stationary scale differences and factor correlations,MGWR and PCAMGWR models were constructed,and landslide susceptibility mapping and assessment were performed based on the proposed models.The ROC curve,AUC value and other statistical indicators were used to validate the landslide susceptibility evaluation results and test the model performance,indicating that the PCAMGWR model had the best performance and the highest prediction accuracy of the landslide susceptibility results.In addition,the spatial non-stationary scale effect of landslide susceptibility and the spatial differentiation characteristics of assessment factors are analyzed,indicating that the factor correlation has an impact on the spatial non-stationary scale,the topographic relief has a local scale spatial effect,and other factors are regional Spatial effects of scale.(4)According to the results of landslide susceptibility zoning,the prevention and control zoning is carried out.From the national disaster prevention,comprehensive disaster prevention and key disaster prevention layer by layer,a hierarchical and step-by-step prevention and control strategy for landslides in Qingchuan County is constructed.According to the landslide prevention and control zones of different grades,specific prevention and control measures are proposed.The research results of this paper can provide a method for in-depth exploration of landslide susceptibility considering geospatial attributes in geological disaster risk management,and provide reference value for engineering site selection,route selection and construction in areas with frequent landslide geological disasters in my country.