Research On Landslide Susceptibility Evaluation Based On Machine Learnin

Landslide disasters are characterized by large scale,wide distribution and strong destructiveness,causing incalculable losses to the national economy and people’s property security.According to the National Disaster Reduction Network and the Ministry of Natural Resources,the total number of geological disasters occurring between 2018 and 2021 is about 36,780,and landslides account for 84%,or about30,895,causing direct economic losses of 28,725 million.Therefore,it is important to construct a landslide sensitivity evaluation model and map it for landslide high incidence areas to avoid landslide disaster risks,promote regional economic development and ensure people’s property safety.Guizhou province has a typical karst landscape and abundant rainfall all year round,which possesses internal and external factors for landslide occurrence.between2019 and 2020,the number of landslides occurring accounts for 76.8% of the total number of geological disasters in the province,which is a landslide-prone area.This paper takes Dafang County,Guizhou Province as the research object,collects landslide data of Dafang County from 2015 to 2020,mines landslide sensitivity characteristics suitable for Dafang County,constructs landslide sensitivity evaluation model of Dafang County using various machine learning algorithms,and combines Arc Gis software for mapping expression.The main research contents of this paper are as follows.(1)Taking Dafang County as the study area,we analyzed and investigated the topography and geomorphology,meteorology and hydrology,stratigraphic lithology and geological structure of Dafang County through literature review and data collection,whose environment is characterized by complex topography and geomorphology,abundant rainfall,diverse stratigraphic lithology and intricate water system.Combining the commonality of global landslide causes and the characteristics of Dafang County,the 30m×30m raster is used as the minimum research unit to collect raster,text,vector and other data from different sources,and to organize and initially screen 1018 landslide samples and basic environmental data applicable to Dafang County.(2)Feature engineering of landslide dataset based on the basic environmental data of Dafang County.Layer overlay and calculation were performed with the base data using Arc Gis software to construct features affecting the landslide sensitivity of Dafang County and combined with Arc Gis for cartographic expression.The features were further selected by variance inflation factor(VIF)and information gain(IG),and the IG index of their rock depth,density,and NDVI were less than 0.01.After feature selection,the size of the original landslide dataset in Dafang County was determined as 1018 cases of samples and 12 features.(3)In order to solve the problems brought by the imbalance problem of data set on the model construction process,Borderline-SMOTE algorithm was used to sample the data set in the ratio of 4:3,while pre-processing such as data cleaning and transformation was combined with the characteristics of different algorithms.Six algorithms of logistic regression,support vector machine,random forest,XGBoost,Cat Boost and neural network were used to construct the landslide sensitivity model of Dafang County.The experimental comparison shows that the Cat Boost model is significantly higher than the other models,with an AUC index of 0.899,which is 1%-9% higher than the other models,and the standard deviation is 0.009.(4)To further improve the accuracy and robustness of the landslide sensitivity evaluation model in Dafang County,the best three models Cat Boost,NN and XGBoost were selected based on the AUC ranking results of individual models,and two integration ideas,i.e.,weighted average and Stacking,were used to fuse the three models and compare the evaluation results,in which the Stacking integration based The fusion model based on the Stacking integration idea improves 1.1% in AUC and1.3% in F1-score compared to Cat Boost.Finally,Arc Gis was used to map the landslide sensitivity evaluation of Dafang County.