Hybrid Bayesian network-based landslide risk assessment method for modeling risk for industrial facilities subjected to landslides

Abstract

Industrialization exposes more petrochemical facilities in the slope-industrial interfaces (SIIs) to the impact range of landslides, demanding practical assessment approaches for addressing associated damage and its potential consequences. However, accurately predicting the expansion and upgrade of landslide hazards in industrial plants is challenging, involving a series of cascading-event trigger-response analyses. Accordingly, this paper develops a hybrid Bayesian network-based landslide risk assessment (HBN-LRA) model to evaluate the landslide risk on storage tanks in SIIs. This model decomposes the landslide risk into three submodules: slope stability, failure of targets, and proactive remedial measures and updates. It transmits the landslide risk through conditional dependence between subsystems. The results of applying the HBN-LRA model toward quantifying the landslide risk in a typical SII area indicate that the distance from the slope in the system risk factors directly determines storage tank damage. Landslide risk is sensitive to geological and geomorphic conditions, such as soil depth, cohesion, and unit weight; their relative importance all exceed 0.15. Tests on the case slope demonstrate that reducing the drainage paving distance from 10 to 8 m can improve slope stability by 50%. This result highlights the potential of the proactive remedial module in evaluating and designing slope drainage systems.