Susceptibility assessment of earthquake-induced landslide by using back-propagation neural network in the Southwest mountainous area of China

Earthquake-induced landslides are destructive and hazardous geological disasters, and mountainous areas are prone to earthquakes. Therefore, it is essential to establish an earthquake-induced landslide susceptibility prediction model with good accuracy, robustness, and generalization capability for the southwestern mountainous areas of China. In this study, the 2008 Wenchuan Ms 8.0, 2010 Yushu Ms 7.1, 2013 Ya'an Ms 7.0, 2014 Ludian Ms 6.5, 2017 Jiuzhaigou Ms 7.0, and 81,513 landslide sites were analyzed. Thirteen influencing factors were selected: PGA, elevation, slope, aspect, curvature, fault type, distance to faults, distance to rivers, NDVI, TWI, lithology, land cover type, and distance to road. Weight and spatial distribution analyses were also performed. Based on a backpropagation (BP) neural network, a regional earthquake-induced landslide susceptibility prediction model was constructed for southwest China. Various methods have been used to evaluate the accuracy of the model. The results showed that the model had excellent prediction accuracy and applicability and that the AUC, accuracy, recall, F1-score, and precision of the model could be stabilized at approximately 0.9. The inclusion of more landslide events in the model can further improve its accuracy and robustness. The constructed model was applied to the fault intersection area, and it was found that most of the actual landslides fell in the predicted medium-high susceptibility area. The research results can enrich the theory of earthquake-induced landslide susceptibility assessment and provide valuable information regarding the location of landslide-prone areas for rapid post-earthquake emergency response in Southwest China.