Incorporating compound temporal precipitation dynamics to enhance landslide susceptibility modeling

This study presents an advanced landslide susceptibility assessment for China by integrating compound temporal precipitation (the combined effects of long‑term antecedent rainfall and short‑term triggering precipitation) into machine‑learning models. Using national landslide inventory data from 2014 and high‑resolution environmental covariates, the authors develop and test several LightGBM‑based models to quantify how precipitation dynamics influence landslide occurrence across different regions and seasons. The research demonstrates that incorporating compound temporal precipitation substantially improves predictive accuracy, particularly in spatial generalisation, and reveals strong spatial and seasonal heterogeneity in landslide sensitivity. 

The study highlights the need for early warning systems and risk‑informed planning to incorporate both long‑term wetness and short‑term rainfall extremes, as these jointly shape landslide triggering mechanisms. It recommends that future disaster risk reduction strategies integrate compound precipitation indicators into operational forecasting frameworks, prioritise regions with high hydrological sensitivity, and expand monitoring during peak rainfall seasons. The study also suggests improving landslide inventories, enhancing temporal coverage of precipitation data, and tailoring regional thresholds to local geomorphological conditions to strengthen preparedness and climate‑change adaptation efforts.