Landslide early warning systems from weather radar observations

Abstract

Landslides are one of the most widespread natural hazards, causing thousands of casualties every year and significant economic damage worldwide. In mountainous regions of Thailand, damages and fatalities due to landslides have increased over the last decades due to intensified human interventions.

This research aims to contribute to the development of a near real-time, joint early warning system for flash floods and landslides in Khao Yai National Park, Thailand. Spatial and temporal landslide occurrences in Khao Yai National Park were assessed. To extend the available landslide inventory, the detection of historical landslides was automated on the Google Earth Engine platform using remotely sensed relative changes in NDVI. Predisposing factors of landslide occurrence were determined using the frequency ratio method. A susceptibility map was then derived by linking the detected landslides to the predisposing factors: slope angle, aspect, land use, lithology, and distance to road. Validation with the landslide inventory shows excellent classification performance.

The temporal probability was assessed using a physically-based multi-hazard model. The hydrological model includes the driving hydrological processes for each grid cell. The streamflow model output was calibrated using observed nested river discharge events. The model performance was tested under spatially distributed precipitation forcing of varying quality. The results show good performance in predicting streamflow and landslide occurrence based on modeled antecedent conditions when using high-quality weather radar information as forcing.

This research demonstrates the effectiveness of integrating fully distributed physically-based modeling with high-resolution spatial rainfall observations from weather radar. The findings are expected to enhance the development of a multi-hazard early warning system and improve disaster risk management in northeastern Thailand.