| Landslide susceptibility evaluation can predict the occurrence degree of landslides in the region,and the results can provide important data support for landslide control.This paper takes Guiyang City as the research area,collects 422 landslide point data and 22 landslide influence factors,and studies the landslide susceptibility evaluation of Guiyang city from three aspects: the screening of landslide influence factors,the ensemble learning method,and the construction of landslide susceptibility evaluation system.In terms of the screening of landslide influencing factors,four groups of landslide influencing factors were selected by using Pearson correlation coefficient analysis,variance analysis and the incremental improvement method of Area Under Curve(AUC)of the receiver operating characteristic curve of the two screening methods.XGBoost and Light GBM models were used as evaluation models to evaluate the landslide susceptibility of the study area.In this paper,four indicators of AUC value,accuracy,sensitivity and specificity are used to evaluate the effect of the model.From the analysis of the experimental results,the four indicators after the incremental improvement of AUC can be improved by up to 2%,5%,7%,3 %,the AUC values of the evaluation model trained by the landslide impact factor obtained by the Pearson correlation coefficient screening method are 0.824 and 0.828,and the AUC values of the model trained by the screening results after the incremental improvement of AUC are 0.830 and 0.831;using variance The AUC values of the two models of the analysis and screening method were 0.829 and 0.834,and the AUC values of the evaluation model after the AUC incremental improvement were 0.846 and 0.849,respectively.The results show that the data screened by the AUC incremental improvement method can effectively improve the effect of landslide susceptibility evaluation.In terms of ensemble learning methods,four ensemble learning methods with different ensemble methods,including random forest,GBDT,Stacking and Voting,were used to evaluate the landslide susceptibility.These four ensemble learning models recombine basic learners into parallel,series,multi-level,and voting methods.For the new evaluation model,the AUC values evaluated by the four models are 0.856,0.868,0.874,and 0.881,respectively.Through comparative analysis,it can be seen that ensemble learning can effectively improve the effect of landslide susceptibility evaluation model,and the ensemble effect of multiple models is better than that of single model.For Guiyang City as the study area,the Vote ensemble method has the best results for landslide susceptibility evaluation.In terms of the construction of the landslide susceptibility evaluation system,a variety of programming languages are used to implement the landslide susceptibility evaluation and the WEB side result display platform.This paper uses Python and third-party libraries to package and combine the various parts of landslide susceptibility evaluation into a desktop landslide susceptibility evaluation system.An interactive software system that can visualize the entire process of landslide susceptibility evaluation.Finally,the landslide susceptibility evaluation results are displayed on the Internet using WebGIS technology,which allows users to view the landslide susceptibility evaluation results on the Internet,and can view the relevant information of the landslide points in the study area.Landslide susceptibility evaluation is an important part of landslide control,and the evaluation result display system on the network can be easily integrated into the comprehensive landslide disaster monitoring system to improve data support and technical solution construction for landslide disaster control in the era of big data.Figure [36] Table [7] Reference [101]... |
