A modeling framework for assessing the future changes in the occurrence of extreme rain-induced landslides

Abstract

Landslides triggered by extreme rainfall pose a serious threat to lives and livelihoods under climate change. However, the influence of climate change on the frequency and intensity of extreme rainfall and the resulting landslide hazard remains insufficiently investigated. Here we develop a novel framework to assess how future changes in extreme rainfall affect landslide occurrence, using a comprehensive dataset of landslide events in Guangdong province, China. We apply the XGBoost machine learning algorithm with a correction by feature interaction constraints to model landslide hazards based on various predisposing and triggering factors. We use a new extreme rainfall correction method to account for the global climate model-projected changes in rainfall patterns under different emission scenarios, based on which we perform a multi-model ensemble analysis to reduce the uncertainty of climate projections. We find that regions with high landslide probability are strongly associated with intense rainfall events, especially in steep and erodible river valleys. Our projections show that under a high-emission scenario, extreme rainfall events and landslide hazards will increase substantially in Guangdong, highlighting the urgency of adaptation of climate change and risk management. Our framework provides a valuable tool for understanding the interactions between climate change, extreme rainfall, and landslide occurrence, and for developing effective mitigation strategies.