Landslide Susceptibility Mapping under Extreme Events: Evidence from the October 2020 Event in Phuoc Son area, Danang city, Vietnam

Abstract

Landslides in mountainous regions of Vietnam predominantly occur during the rainy season, with high-intensity rainfall events frequently triggering widespread shallow landslides. On 28 October 2020, an extreme rainfall event, with daily precipitation reaching approximately 350–400 mm, caused numerous landslides in the mountainous Phuoc Son area, Da Nang City. Based on a landslide inventory compiled for this event and a comprehensive set of conditioning factors, this study investigates the influence of these factors on landslide occurrence. In addition, two machine learning algorithms, Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), were employed for landslide spatial prediction as well as landslide susceptibility mapping. Frequency analysis and Frequency Ratio (FR) results indicate that 12-hour accumulated rainfall played a dominant role in triggering this landslide event. Model evaluation shows that both models achieved good predictive performance; however, XGBoost outperformed RF, with the value of AUC up to 0.905. Based on the optimal XGBoost model, landslide susceptibility maps were generated under two scenarios: (1) using the spatial distribution of 12-hour accumulated rainfall and (2) applying a rainfall-triggering threshold of 320 mm/12h across the entire study area. The results indicate that, under Scenario 2, zones of high and very high landslide susceptibility expand markedly throughout the region, providing valuable information for disaster hazard prevention and land-use planning in mountainous areas.