A prototype of catchment-scale digital twin systems (cDTS) for debris-flow early warning

Debris flows as fast-moving and water-saturated sediment masses are particularly hazardous in alpine areas due to their high destructive power and poor predictability. We still do not fully know under what conditions debris flows occur and how to predict them. The most common method for predicting debris flow in warning systems and hazard assessment uses precipitation intensity and duration thresholds. However, these do not provide accurate and quantitative predictions of debris flow occurrence and are subject to high uncertainty due to limited data. Thanks to recent developments in novel monitoring technologies that have led to an unprecedented data explosion, it is now time to address these knowledge gaps innovatively and interdisciplinaryly. To this end, we develop a scalable and transferable catchment Digital Twin System (cDTS) that combines the latest knowledge from geomorphology, remote sensing, and hydrology to derive probabilistic rainfall intensity-duration (ID) thresholds from limited observations. The cDTS is a physics-informed genetic machine learning framework based on partially known physics, sparse and noisy data, and nonlinear dynamical networks. We test this framework on a small catchment in the Italian Dolomites to determine probabilistic thresholds for the occurrence of debris flows. The new rainfall thresholds are a negative exponential function controlled by infiltration capacity instead of a power law relationship. The cDTS is a lighthouse case for applications of the digital twin in geoscience, helping improve early warning system performance by providing timely, evidence-based information to the public and policymakers.