Probabilistic prediction of shallow landslides through a generalized linear mixed-effects model integrating geological and rainfall factors

Shallow landslides threaten infrastructure, ecosystems, and human life in mountainous and hilly regions. Accurate forecasting remains challenging due to complex environmental interactions and spatial heterogeneity. We develop a probabilistic multi-factor model based on a generalized linear mixed-effects framework (GLMM) to predict shallow landslide occurrence. The model incorporates lithology, soil type, normalized difference vegetation index (NDVI), slope, slope aspect, elevation, rainfall intensity and duration as fixed effects, while capturing spatial and temporal grouping via random effects. We apply the model to an extensive inventory in Guangdong Province, China, encompassing 2,167 shallow landslide events and 2,166 control cases from 2013 to 2022. Compared to rainfall intensity–duration (I–D) and event–duration (E–D) threshold models, the GLMM approach delivers significantly higher predictive performance, achieving over 80% accuracy while maintaining a strong balance between sensitivity and specificity, enabling more reliable alerts and fewer unnecessary evacuations. Rainfall intensity and duration exert the strongest positive predictors, while NDVI displays a marked negative effect. GLMM offers interpretable coefficients and accommodates grouped environmental data, making it suitable for regional-scale landslide risk assessment. Our results highlight the value of integrating multi-dimensional predictors and offer a reliable, interpretable tool for operational landslide early warning systems.