Climate-impacted reliability analysis of earth slopes with spatially variable soils using active learning surrogate modeling

Climate-induced factors such as rainfall, relative humidity, evapotranspiration, and temperature affect the stability of earth slopes, posing risks to infrastructure and public safety. Understanding and assessing these effects is essential for effective slope management and hazard mitigation. This study establishes a probabilistic framework for evaluating earth slope stability under climatic influences, including five key steps. The first two steps involve problem formulation and numerical model development, followed by a deterministic analysis in the third step. The fourth and fifth steps incorporate probabilistic analyses using random variables (RVs) and random fields (RFs), respectively. Deterministic analyses are conducted using the Finite Element Method (FEM) and the Limit Equilibrium Method (LEM), while probabilistic assessments are carried out by applying reliability techniques, including the First-Order Reliability Method (FORM) and the Adaptive Polynomial Chaos-Kriging (A-PCK) model. The framework is applied to an earth slope in Belo Horizonte, Brazil, considering 33 years of climate data. The analysis focuses on the most critical 150-day period, marked by intense climatic and groundwater fluctuations, to capture evolving safety conditions and highlight the impact of extreme weather events on slope stability. Based on the deterministic, RV, and RF approaches, a table of stability level criteria is introduced to guide slope monitoring and modeling. The slope is evaluated through a three-tiered process, progressing from deterministic analysis to probabilistic assessments using RV and RF methods, with the slope’s safety updated at each step. This structured, multi-level methodology supports adaptive monitoring and informed decision-making, providing a practical tool for managing slope stability under dynamic climate conditions.