Assessment Of Landslide Susceptibility Based On Ensemble Learning Algorithm In Xi'an City

The role and significance of landslide susceptibility assessment in human engineering activities is obvious.Taking the landslide disaster in xi?an city as the research object,on the basis of data collection,field investigation and experimental analysis,ArcGIS,MATLAB and SPSS software were used to evaluate the susceptibility of landslide disaster in xi?an city,and the following results and conclusions were obtained:(1)This paper analyzed the geological environment conditions in the study area,development characteristics,distribution rules,formation conditions and influencing factors of landslide hazards,and put forward that the main control factors of landslide hazards in the study area are landform,lithology,geological structure and groundwater action,and the inducing factors include rainfall and human engineering activities.(2)The evaluation indexes of landslide hazard susceptibility in xi?an city were determined,including elevation,slope,stratigraphic age,distance to the fault,rainfall,stream power index,topographic wetness index,normalized differential vegetation index,land use type and distance to the road.This paper analyzed the spatial distribution law of evaluation factors and the occurrence of landslide hazards,and established the evaluation factor index system of landslide hazards in xi?an city.(3)Three ensemble learning algorithms of random forest,bootstrap aggregating and boosting tree were used to predict the hazard vulnerability of landslide in xi?an city.The results show that prediction accuracy of random forest,bootstrap aggregating and boosting tree are 90.60%,89.52% and 87.60%,respectively.According to the method of natural discontinuities,the landslide susceptibility index in the study area is divided into five grades: very high,high,medium,low and very low.According to the statistics of zoning results,the areas of very high—high prone areas predicted by random forest,bootstrap aggregating and boosting tree models account for 26.45%,26.60% and 34.34% of the total area of the study area,and the number of landslides account for 71.71%,69.86% and 70.33% of the total landslide,respectively.The results of three algorithms are satisfactory,among which the random forest algorithm is better than the other two.(4)The receiver operator characteristic(ROC)curve and the Kappa coefficient was used to compare and test the performance of three models,the results show that the area under curve(AUC)values of random forest,bootstrap aggregating and boosting tree models are 0.955,0.954 and 0.866,respectively,the Kappa coefficient are 0.816,0.790 and 0.752,three evaluation models have higher prediction performance,including random forest model performance is more superior than other two kinds of model.