A dynamic Landslide hazard forecasting Framework for the Lower Mekong Region

In the Lower Mekong Region, landslides cause huge loss of life, infrastructure, and property every year. Despite the severity of landslide hazards, the region lacks a dynamic and near-real-time operational landslide hazard forecast system. In this study, a system was developed to predict landslide hazards for up to 2 days lead time. The model uses the Climate Hazards Center InfraRed Precipitation with Station data bias-corrected and downscaled version of National Centers for Environmental Prediction (NCEP) Global Ensemble Forecast System (CHIRPS-GEFS). NCEP-generated Climate Forecast System soil moisture data was incorporated as another dynamic variable. The model also included other predictors such as slope, relief, distance to roads, distance to rivers, and distance to faults. Landslide inventories spanning 2015-2020 were generated using a semi-automatic landslide detection - change detection method. A gradient-boosting model was trained for the period of 2015-2019 and validated using the 2020 inventory. The area under the curve (AUC) of receiver operating characteristic curve (ROC) was more than 98%, and the true positive rate was more than 75% for up to 2 days lead time. The model was also retrospectively simulated to illustrate the mean and seasonal variation of the landslide hazard probability. The long-term mean and mean seasonal hazard probability clearly showed the trend followed with the precipitation forecast. This tool joins the dynamic Landslide Hazard Assessment for Situational Awareness model for the Lower Mekong Region (LHASA-Mekong) model to provide an integrated view of the present condition and prediction of landslide hazards to support the decision-making on landslide risk management across the Lower Mekong