A Comparative Study On Evaluation Methods Of Landslide Susceptibility Based On Different Evaluation Units

Landslide disaster seriously affects social development and stability,plays a leading role in all geological disasters,and the harm caused by it cannot be ignored.At present,the situation of landslide prevention and control in China is still grim,and the measures of disaster prevention and mitigation still focus on prevention.Therefore,it is the first choice to evaluate the susceptibility of landslide prevention.In this study,Luonan County of Shaanxi Province is taken as the research area.Based on the detailed investigation data of geological disasters,landslides are selected as the research object.On the basis of comprehensive analysis of the developmental characteristics and rules of landslides in the research area,ArcGIS,MATLAB,SPSS and other software are used as the research tools.Carried out a comparative study on the landslide Susceptibility evaluation method in the study area under the condition of different evaluation units(30 m grid unit,60 m grid unit and the slope unit)based on the classical machine learning model(NBC,LDA,SVM and KNN)and integrated learning(Bagging,AdaBoost and RF)model.For the three selected evaluation units,the applicability and evaluation accuracy of the model were analyzed,which is of guiding significance for the government to formulate the planning of landslide disaster prevention and control and the zoning of landslide susceptibility.The main achievements are as follows:(1)Based on the analysis of the development characteristics and distribution of landslides in the study area,two typical accumulation layer landslides were selected,and their basic characteristics and influencing factors were analyzed emphatically.The results show that:the landslide development type in the study area is mainly accumulation landslide;the scale is mainly medium and small;it is affected by rainfall and human engineering activities obviously;and it shows significant rules in region and time.(2)Through the principal component analysis,correlation analysis and the multicollinearity analysis of preliminary selection of evaluation factors,on the basis of the analysis results to cut out the small weight,or the strong correlation factors and keep the rest of the evaluation factors for its using statistical analysis of the relationship between the landslide and each evaluation factor,the analysis indicates that retain the residual evaluation factor control of the landslide,landslide liability in the study area was established evaluation index system of factors.(3)The evaluation models of the above seven kinds of landslide susceptibility based on three different evaluation units were established respectively,and the partition maps of the landslide susceptibility degree of each method were generated.The rationality of the evaluation results was tested by mathematical statistical method.The results show that the three evaluation units show high prediction accuracy in the above seven models,and all meet the relevant standards of rationality test,which indicates that the evaluation units and evaluation models selected in this study have played a good effect and the partition is reasonable.(4)The ROC curve and Kappa coefficient were used to compare and test the prediction accuracy and consistency testing degree of the three different evaluation units in the 7 models selected above.The comprehensive comparison shows that the optimal combination of evaluation unit and evaluation model in this study is the RF model based on 30m grid element.This combination has the most significant effect on the partition of landslide susceptibility,and the prediction accuracy and consistency test degree both reach a very high level(AUC=0.916,k=0.923),which can be used as the combination of evaluation unit and evaluation model with the highest prediction accuracy and the most reasonable partition in this study.