A Path Toward Short-Term Probabilistic Flash Flood Prediction

Abstract

There are ongoing efforts to move beyond the current paradigm of using deterministic products driven by observational-only data to make binary warning decisions. Recent works have focused on severe thunderstorm hazards, such as hail, lightning, and tornadoes. This study discusses the one of the first steps towards having probabilistic information combined with convective-scale short-term precipitation forecasts available for the prediction and warning of flash flooding. Participants in the Hydrometeorology Testbed—MRMS Hydrology (HMT-Hydro) experiment evaluated several probabilistic-based hydrologic model output from the probabilistic Flooded Locations and Simulated Hydrographs (PRO-FLASH) system during experimental real-time warning operations. Evaluation of flash flood warning performance combined with product surveys highlighted how forecasters perceived biases within the probabilistic information and how the different probabilistic approaches influenced warnings that were verified versus those that were unverified. The incorporation of the Warn-on-Forecast System (WoFS) ensemble precipitation forecasts into the PRO-FLASH product generation provided an opportunity to evaluate the first coupling of sub-hourly convective-scale ensemble precipitation forecasts with probabilistic hydrologic modeling at the flash flood warning time scale through archived case simulations. The addition of WoFS precipitation forecasts resulted in an increase in warning lead time, including four events with ≥ 29 minutes of additional lead time but with increased probabilities of false alarms. Additional feedback from participants provided insights into the application of WoFS forecasts into warning decisions, including how flash flood expectations and confidence evolved for verified flash flood events and how forecast probabilistic products can positively influence the communications of the potential for flash flooding.