Landslide Susceptibility Assessment in Zigui County Based on Heterogeneous Ensemble Model

Machine learning models have been extensively applied to assess landslide susceptibility, but the effectiveness of susceptibility assessment varies from model to model. To overcome the limitation of insufficient evaluation precision and generalization ability of a single model. Taking Zigui County as the study area, and nine assessment factors including elevation, NDVI, and lithology et al, are collected. RF, SVM, and ANN models were used as the base model to establish a heterogeneous ensemble learning model. By comparing the accuracy, precision, recall, AUC values of four models, the model with the highest accuracy will be determined to draw the susceptibility map of Zigui County. The results suggest that the heterogeneous ensemble model has the best evaluation accuracy and the AUC values are improved by 0.03, 0.032 and 0.226 respectively compared with the base model. The landslide susceptibility map created from this model can provide help to Zigui County for planning and management of geohazards.