Random forest-based prediction of shallow slope stability considering spatiotemporal variations in unsaturated soil moisture.

With the increasing incidence of extreme rainfall driven by global climate change, geological hazards like landslides have become more prevalent. This study proposed an efficient framework that combined machine learning and physical models to enhance computational efficiency and reliability for regional slope stability predictions under extreme rainfall. The GEOtop model was employed to simulate volumetric water content (VWC) in unsaturated soil of an area in Singapore under maximum daily and maximum 5-day antecedent rainfall conditions. The result of analyses was then incorporated into Scoops3D for factor of safety (FOS) calculations. The random forest (RF) models were trained using VWC under maximum daily rainfall and applied to predict slope stability under maximum 5-day antecedent rainfall, with outcomes compared to those of Scoops3D. Statistical results and spatial distribution maps both showed that the proposed framework achieved comparable accuracy to Scoops3D at various depths while significantly improving efficiency. The findings also highlighted the critical role of surface soil moisture (at 0.05 m) in slope stability predictions. This framework demonstrates the potential of integrating machine learning and physical models for efficient slope stability prediction, as well as supports the integration of remote sensing or field-measured surface soil moisture data for dynamic predictions in unsaturated soil conditions.