A precipitation downscaling framework for regional warning of debris flows in mountainous areas

A timely warning system for debris-flow mitigation in mountainous areas is vital to decrease casualties. However, the lack of rainfall monitoring stations and coarse resolution of satellite-based observations pose challenges for developing such a debris-flow warning model in data-scarce areas. To offer an effective method for the generation of precipitation with fine resolution, a machine learning (ML)-based approach is proposed to establish the relationship between precipitation and regional environmental factors (REVs), including normalized difference vegetation index (NDVI), digital elevation model (DEM), geolocations (longitude and latitude) and land surface temperature (LST). This approach enables the downscaling of 3B42 TRMM precipitation data, providing fine temporal and spatial resolution precipitation data. We use PERSIANN-Cloud Classification System-Climate Data Record (PERSIANN-CCS-CDR) data to calibrate the downscaled results using geographical differential analysis (GDA) before applying the calibrated results in a case study in the Gyirong Zangbo Basin. After that, we calculate the rainfall thresholds of effective antecedent rainfall (Pe)—intraday rainfall (Po) based on the calibrated precipitation and integrate these thresholds into a susceptibility map to develop a debris-flow warning model. The results show that (1) this ML-based approach can effectively achieve the downscaling of TRMM data; (2) calibrated TRMM data outperforms the original TRMM and downscaled TRMM data, reducing deviations by 55% and 57%; (3) the integrated model, incorporating rainfall thresholds, outperforms a single susceptibility map in providing debris-flow warnings. The developed warning model can offer dynamic warnings for debris flows that may have been missed by the original warning system at a regional scale.