Application and verification of a multivariate real-time early warning method for rainfall-induced landslides: implication for evolution of landslide-generated debris flows

Rainfall-induced landslides are a frequent and often catastrophic geological disaster, and the development of accurate early warning systems for such events is a primary challenge in the field of risk reduction. Understanding of the physical mechanisms of rainfall-induced landslides is key for early warning and prediction. In this study, a real-time multivariate early warning method based on hydro-mechanical analysis and a long-term sequence of real-time monitoring data was proposed and verified by applying the method to predict successive debris flow events that occurred in 2017 and 2018 in Yindongzi Gully, which is in Wenchuan earthquake region, China. Specifically, long-term sequence slope stability analysis of the in situ datasets for the landslide deposit as a benchmark was conducted, and a multivariate indicator early warning method that included the rainfall intensity-probability (I-P), saturation (S-i), and inclination (I-r) was then proposed. The measurements and analysis in the two early warning scenarios not only verified the reliability and practicality of the multivariate early warning method but also revealed the evolution processes and mechanism of the landslide-generated debris flow in response to rainfall. Thus, these findings provide a new strategy and guideline for accurately producing early warnings of rainfall-induced landslides.