Forecasting and mitigating landslide collapse by fusing physics-based and data-driven approaches

Deep-seated landslides represent one of the most devastating natural hazards on earth, typically creeping at inappreciable velocities over several years before collapsing at catastrophic speeds. They can have detrimental consequences to society, causing fatalities and prone to affect transportation infrastructures. Currently, limited data-driven tools allow a time-dependent assessment of these landslides, mainly linking their motion with groundwater variations or forecasting the collapse time using the empirical inverse velocity–time approach. In this study, we validate that monitoring the basal temperature of a creeping landslide, and fusing it with physics-based modeling, can offer predictive and control capabilities for the landslide's response. By following the theoretical suggestions of several works, we installed a thermometer on the sliding surface of a creeping landslide in Andorra. In this paper, we report the results of 3 months of recording and identify the basal stress–temperature space as the underlying unobserved phase space that determines the problem and allows us to forecast and control the landslide. Following our results, we validate our suggestion by applying the model to three other cases, where temperature measurements are unavailable: the Vajont (Italy) and Mud Creek (California) collapsed landslides and the active Shuping landslide (China). The study shows that physics-based models can be trained in the same phase space and offer forecasting and control capabilities. We anticipate our results to be the starting point for a new era in monitoring, controlling, and forecasting deep-seated landslides, aiming at alleviating their devastating impact on society.

Duke Scholars

Author Manolis Veveakis Civil and Environmental Engineering