AI-Empowered Near-realtime Operational Prediction System of Landslides in Lao Cai province, Vietnam

Vietnam faces heightened vulnerability to severe climate change impacts, notably sea level rise, flooding, and landslides. In recent years, the northwest mountainous regions have experienced recurrent and widespread landslides during the rainy season (May to October), resulting in significant economic losses. This study focuses on the Lao Cai province, employing various data mining techniques—Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Random Forest (RF)—to spatially predict landslide hazards. Initially, a comprehensive landslide inventory map was constructed from diverse sources, pinpointing past landslide occurrences. Subsequently, multiple factors influencing landslides were considered, including slope angle, slope aspect, profile curvature, wetness index, lithology, Normalized Difference Vegetation Index (NDVI), soil type, soil moisture, road density, house density, and rainfall. Utilizing these factors, landslide susceptibility indexes were computed through the respective models. Validation, using landslide locations not utilized in the training phase, revealed that models employing Random Forest (RF) exhibited the highest prediction capability. The trained model was then applied to generate real-time forecasts of landslide susceptibility maps for up to 16 days, using bias-corrected Global Forecast System (GFS) precipitation data. This WebGIS operational prediction system enhances preparedness and awareness, facilitating improved mitigation strategies to mitigate the impact of landslides.