Downscaled Precipitation Forecasts for Advancing Prediction of Post-Fire Debris Flows

Rainfall-induced, post-fire debris flow (PFDF) hazards are increasing in frequency globally, with the US West a prime location. Communities downstream of steep, burned terrain are especially vulnerable to PFDFs. Research efforts have been focused on predicting the occurrence of a PFDF based on rainfall information from in-situ or remote sensors. While such approaches have been applied with a certain degree of success, they do not offer enough lead time for proper preparedness and response measures. In this work, we investigate the potential of integrating precipitation information from a global forecasting system with machine learning models for PFDF prediction. The proposed framework provides significant lead time (24-72 hours) in PFDF prediction and has the potential to be applied globally. Forecasted precipitation is based on the NASA Goddard Earth Observing System - Forward Processing (GEOS-FP) product. The spatial resolution of GEOS-FP estimates is improved via a statistical downscaling procedure that uses IMERG precipitation as reference. The proposed PFDF prediction framework is evaluated against the Multi-Radar Multi-Sensor (MRMS)-based predictions for a number of PFDF cases in the US West. Results from the comparative analysis demonstrate the significant improvement of PFDF predictions for the downscaled GEOS-FP version (relative to original GEOS-FP estimates) and highlight the potential of the framework for developing global PFDF predictions.