A meta model for time-variant reliability analysis of unsaturated slope under rainfall

Purpose With the rise in rainfall-induced landslides due to climate change, landslide-prone areas increasingly encounter natural catastrophes during rainy seasons. Conventional geotechnical analyses and empirical models are unable to effectively capture the unsaturated behaviour of natural slopes under the inherent uncertainties of soil properties. This study aims to develop a probabilistic framework to assess the safety of slopes under rainfall events while accounting for both geotechnical variability and dynamic effects of rainfall. Design/methodology/approach The proposed framework treats geotechnical parameters of each soil layer and the rainfall duration and intensity as random variables. A time-dependent factor of safety (FoS) is computed using a finite element model in GeoStudio, capturing the transient behaviour of slope stability. This data is then used to train a hybrid surrogate model that combines regression and artificial neural network, allowing for efficient prediction of FoS over time. The surrogate model is coupled with the Monte Carlo simulation to estimate the probability of slope failure as a function of time. A comprehensive reliability sensitivity analysis is also conducted on key soil parameters to assess their impact on slope stability. Findings The results are presented as exceedance probability curves that illustrate the evolution of FoS and failure probability over time. Additionally, the time-dependent sensitivity analysis reveals the significant influence of permeability on the dynamic safety of slopes, demonstrating the importance of incorporating transient and uncertain conditions into slope stability assessment. Originality/value This study introduces a novel time-dependent probabilistic framework that integrates machine learning techniques with finite element modelling to evaluate slope safety. The use of a hybrid surrogate model, tailored to handle the dynamic and nonlinear nature of the problem, offers a more accurate and computationally efficient alternative to conventional methods, particularly for slopes in unsaturated soil conditions under variable rainfall scenarios.