Constructing regional catchment-specific triggering threshold for runoff-generated debris flows: a case study in the Bailong River

Constructing catchment-specific rainfall thresholds is crucial for high-accuracy debris flow forecasting. However, rapid methods for determining thresholds across multiple catchments remain limited due to the difficulty of relating catchment characteristics to peak discharges. This study explores such relationships using hydrological simulations and machine learning techniques, aiming to enable the rapid estimation of catchment-specific rainfall thresholds. We conducted a series of hydrological simulations with varied rainfall amounts to calculate peak discharges for 267 catchments in the Bailong River Basin, China. The rainfall amounts and corresponding peak discharges were fitted using three types of mathematical functions: linear, polynomial, and power-law. Machine learning models were then applied to predict the fitted coefficients for each catchment based on thirteen catchment features related to runoff generation and concentration. The Extra Trees Regressor (ETR) was used for catchments larger than 10 km², while CatBoost Regressor (CBR) was employed for smaller ones. This modelling framework allows the derivation of functional relationships between rainfall and discharge for each catchment, which were subsequently used to estimate rainfall thresholds for debris flow initiation. The proposed method shows strong potential for improving forecasting accuracy and regional applicability, and it can be further enhanced with more hydrological observations and long-term monitoring data.