HRRRCast: Advancing AI-Driven Weather Prediction at NOAA
Drs. Isidora Jankov & Daniel S. Abdi
GSL
3 December, 2pm
Abstract:
NOAA's Global Systems Laboratory (GSL) has developed HRRRCast, the first AI-powered, CONUS-wide, data-driven forecasting system at NOAA. Trained on three years of High-Resolution Rapid Refresh (HRRR) output, HRRRCast began as an emulator of the physics-based model and now provides competitive forecasts of reflectivity, precipitation, temperature, humidity, and wind out to 48 hours. The system has recently advanced to HRRRCast V2.0, featuring expanded vertical resolution, additional variables-including accumulated precipitation-and a nine-member ensemble executed in near-real time at EMC and visualized through DESI. Early results demonstrate strong skill and highlight the promise of AI-driven approaches for advancing operational weather prediction.