Evaluation of Landslide Susceptibility by Optimization Integrated Machine Learning Algorithm Based on Gradient Boosting: Take Both Banks of Yarlung Zangbo River and Niyang River as Examples

The geological structures on both banks of the Yarlung Zangbo river and the Niyang river are active, and landslides occur frequently. The landslide susceptibility assessment can effectively reduce the damage to human life and property caused by disasters. This paper studies the performances of Weighted Random Forests, XGBoost and LightGBM algorithms based on Gini coefficient in landslide susceptibility. Select 188 landslide samples and 7 influencing factors, and use the 50–fold cross–validation method to train the model. During the training process, the feature selection algorithm is considered at the same time, and the Bayesian method is used to optimize the hyperparameters. Analysis of forecast results at the level. The results show that landslide is most likely to occur within the elevation of 32～1 544 m and 2 722～3 752 m, the gradient of 30°～40°, and the distance of 200 m from the fault zone, river and road. The extremely high and high landslide prone areas account for 12.14% and 12.41% respectively, and the low and extremely low landslide prone areas account for 26.47% and 29.55% respectively. More than half of the areas in Nyingchi prefecture are not prone to landslide disasters. Among all models, LightGBM model performs best, with AUC value of 0.843 2, accuracy of 0.853 1, and F1 score of 0.834 5. Damu township and Bangxin township in Motuo county, Danniang, Lilong, Zhaxi Raodeng township in Linzhi county, Long village in Lang county, and Jiangda township in Gongbujiangda county are positioned in extraordinarily high–risk areas, with a excessive likelihood of landslides. Corresponding prevention and control measures should be taken in these areas.