Towards hydrometeorological thresholds of reservoir-induced landslide from subsurface strain observations

Abstract

Synergistic multi-factor early warning of large-scale landslides is a crucial component of geohazard prevention and mitigation efforts in reservoir areas. Landslide forecasting and early warning based on surface displacements have been widely investigated. However, the lack of direct subsurface real-time observations limits our ability to predict critical hydrometeorological conditions that trigger landslide acceleration. In this paper, we leverage subsurface strain data measured by high-resolution fiber optic sensing nerves that were installed in a giant reservoir landslide in the Three Gorges Reservoir (TGR) region, China, spanning a whole hydrologic year since February 2021. The spatiotemporal strain profile has preliminarily identified the slip zones and potential drivers, indicating that high-intensity short-duration rainstorms controlled the landslide kinematics from an observation perspective. Considering the time lag effect, we reexamined and quantified potential controls of accelerated movements using a data-driven approach, which reveals immediate response of landslide deformation to extreme rainfall with a zero-day shift. To identify critical hydrometeorological rules in accelerated movements, accounting for the dual effect of rainfall and reservoir water level variations, we thus construct a landslide prediction model that relies upon the boosting decision tree (BDT) algorithm using a dataset comprising daily rainfall, rainfall intensity, reservoir water level, water level fluctuations, and slip zone strain time series. The results indicate that landslide acceleration is most likely to occur under the conditions of mid-low water levels (i.e., < 169.700 m) and large-amount and high-intensity rainfalls (i.e., daily rainfall > 57.9 mm and rainfall intensity > 24.4 mm/h). Moreover, this prediction model allows us to update hydrometeorological thresholds by incorporating the latest monitoring dataset. Standing on the shoulder of this landslide case, our study informs a practical and reliable pathway for georisk early warning based on subsurface observations, particularly in the context of enhanced extreme weather events.