Rapidly assessing earthquake-induced landslide susceptibility on a global scale using random forest

Earthquake-induced landslides (EQILs) are an incredibly destructive geological disaster. Rapid landslide susceptibility assessments are indispensable and critical for risk analysis and emergency management. Previous studies mainly focus on the regional-scale assessment of EQIL susceptibility, while the global analyses of that are lacking. In this study, we constructed a global model for rapidly assessing earthquake-induced landslide susceptibility based on the random forest (RF) algorithm using globally available data. In total, 288,114 landslides from 16 high-quality EQIL inventories were utilized to develop the global landslide model. We split the data into 70% training dataset for model training and 30% testing data for model evaluation. We also used three blind test events to validate the model performance. The model showed excellent performance on the testing data (accuracy = 0.945, and AUC = 0.985). The RF model exhibited strong spatial generalizability and robustness, with an AUC exceeding 0.8 for each landslide inventory and showing good performance on the blind test events. The resulting landslide susceptibility maps also match relatively well with the actual landslide locations. Among the conditioning factors, modified Mercalli intensity (MMI), elevation and slope are the three most important conditioning factors. The susceptibility maps for each landslide event were produced. The developed RF model would be useful in studies of earthquake-induced land-slide susceptibility and emergency response after an earthquake. (c) 2021 Published by Elsevier B.V.

Automated summary

The global model constructed using the random forest algorithm can rapidly assess the susceptibility of earthquake-induced landslides. The model performed well on testing data and showed good performance on blind test events.