A dynamic Bayesian network-based model for evaluating rainfall-induced landslides

Rainfall-induced landslides are among the most widespread hazards in the world. Forecasting slope stability when rainfall events occur is an important topic in the landslide disaster field. The link between rainfall events and slope instability is uncertain, because of the variabile rainfall infiltration mechanism within a slope mass. To reduce this uncertainty, a dynamic Bayesian network (DBN) is utilised to analyse the stability of a slope in time domains by representing the probabilistic relationships of variables [rainfall intensity, cumulative infiltration, pore water pressure (PWP), and factor of safety] and relating these variables to each other over adjacent time steps. This model improves the accuracy of objective evaluation by coupling parts of observation value against the uncertainty. Successively, a deterministic method (DM) with empirical or tested values in models and a logistics regression (LR) method totally based on the statistical results are employed to estimate slope stability in time domains. Through comparing the temporal slope instability evaluation of the three mentioned methods with the real scenes of an artificial slope, the results demonstrate that 1) the DBN can evaluate the stability of a slope closer to reality in time domains compared to the DM; 2) The DBN estimates the instability of the slope in accordance with the LR method and can include some physical mechanisms. The characteristics of the DBN make the method a potential way to evaluate slope stability.