Research On Zoning Of Landslide Susceptibility Based On XGBoost

The Xiangxi Autonomous Prefecture in Hunan Province is rich in precipitation,its landform types are diverse,dominated by middle and low mountains,and its geological conditions are complex,making it an area prone to geological disasters.According to the "Geological Disaster Prevention and Control Plan of Xiangxi Autonomous Prefecture(2011-2020)" and other statistics,Xiangxi Autonomous Prefecture includes a total of 708 hidden geological hazards,including landslides,debris flows,and collapses.Among them,there are 407 hidden landslides,accounting for geological hazards.57.49% of the total number of hidden danger points.This paper uses the actual landslide hazard point data in Xiangxi Autonomous Prefecture as the reference value,and combines the terrain and geomorphic data such as elevation and slope of the region,the geological structure data such as faults and stratigraphic age,and the geographical influencing factor data such as human activities such as roads and buildings.The distribution characteristics and development environment of landslide hidden danger points in this area are studied in detail and related analysis is carried out.The research findings are as follows:(1)The hidden dangers of landslides in Xiangxi Autonomous Prefecture mostly occur in the shale of the Xintan Formation with an elevation of 400?600m,a slope of 3?30°,a northwest direction,and a profile curvature of-0.6?1.4,Cambrian and Silurian.(Slate)and sandy slate.From the perspective of the age,lithology and geological structure of the landslide,the landslides in Xiangxi Autonomous Prefecture are mostly soil landslides,and the hazards of landslides are mainly small and medium-sized,accounting for 94.84%..(2)Based on the XGBoost and random forest algorithm,a landslide hidden danger point identification model was constructed.Among the 306 sample points in the test set,the XGBoost model has a model accuracy rate of 93.46%,a precision rate of 95.73%,a recall rate of 88.19%,and an F1 score of 91.80 %;In the 306 sample points of the test set,the random forest model is 92.48% accurate,91.27% accurate,90.91% recallable,and 90.91% F1 score.From the test set,the performance of the XGBoost model is better than the random forest.(3)By analyzing the importance ranking of the influencing factors of the XGBoost and Random Forest algorithm landslide hidden danger point identification models,it is shown that slope and building area are the two most important factors affecting the occurrence of landslides,of which the highest slope importance is 37.66%.This is because slopes with a certain slope give birth to landslides;the importance index of building area is 5.08%.The construction of houses usually involves excavation and filling.The size of the building area determines the size of the excavation and filling area to a certain extent.This will lead to the instability of the surrounding stratum or geological structure,resulting in a decrease in the stability of rock masses.(4)Using two machine learning methods,XGBoost and random forest,the landslide susceptibility zoning of different mapping units are compared.The comparative experimental results show that the landslide susceptibility zoning results based on the XGBoost model are most consistent with the actual situation.At the same time,the landslide susceptibility evaluation results obtained by the XGBoost model are analyzed.As the susceptibility level becomes higher,the proportion of the total number of units and the proportion of historical landslide hidden danger points gradually increase,and the landslide susceptibility is based on the slope unit.The division is more reasonable.(5)Using the Tree SHAP interpretation model,analyze the changes in the SHAP value of each feature.From all feature factors,select the first three with greater impact for targeted analysis.It can be seen that based on the XGBoost model,the slope,The area of cultivated land and the area of buildings and their contribution to the hidden danger points of landslides are obviously non-linear.The slope first increases and then decreases with the increase of the characteristic value.The area of cultivated land decreases monotonously with the increase of the characteristic value.The area of the building first increases and then gradually stabilizes with the increase of the characteristic value.