A data driven efficient framework for the probabilistic slope stability analysis of Pakhi landslide, Garhwal Himalaya

The stability of slope is usually characterized by many parameters which are mostly uncertain in nature. Deterministic approach is usually followed to calculate the factor of safety of a slope, but it does not depict the true state of the slope. Hence, a probabilistic approach is a better alternative, which can quantify the probability of failure of a slope under uncertain input parameters. In the present study, slope stability assessment of Pakhi landside is carried out using finite element modelling (FEM) to ascertain the stability conditions of the multilayer slope under both deterministic and probabilistic framework. The multilayer configurations of the profiles are established from the electrical resistivity tomography (ERT). The shear strength reduction (SSR) method is employed to determine the critical strength reduction factor of the slope considering four random variables, namely, cohesion ( c ), angle of internal friction ( ϕ ), Poisson’s ratio ( ν ) and elastic modulus ( E ) of each individual layers. The deterministic factor of safety values along two considered profiles namely, section X–X ′ and Y–Y ′ are calculated as 1.41 and 1.25, respectively. A data driven machine learning algorithm is used to build a computationally efficient surrogate model to perform Monte Carlo Simulations (MCS). MCS are performed for two different values of coefficient of variation, i.e., 5% and 15% for all the four random variables of all the layers. The proposed method has no idealization regarding the layering configuration and the failure surface. Probabilistic analysis has been made exhaustive and computationally efficient. The probabilistic analysis indicates good adherence with the recent landslide incident in the field. Further, the analysis indicates that the proposed methodology is favourable and useful tool for the system reliability analysis of landslide slopes. Research highlights Electrical resistivity tomography (ERT) interpretations could delineate sub-surface geometry of lithological layers of the landslide slope. Numerical simulation is performed using the finite element modelling. Sampling of random variables is done using the Latin Hypercube sampling technique. Probabilistic analysis is performed using the MARS-based surrogate model for Monte-Carlo simulation (MCS) with two values of CoV , viz., 5% and 10%. Efficiency of machine learning algorithm to incorporate with the stand alone FE code is demonstrated to build a surrogate model for an efficient probabilistic slope stability analysis.