It's called HRRR-Cast – an AI version of the current forecast model High-Resolution Rapid Refresh or HRRR (pronounced "her"), NOAA' s first regional experimental AI forecast system.
The HRRR has been the National Weather Service's short-term forecast model for ten years. It generates an 18-hour forecast every hour, and a new 48-hour forecast every six hours. The AI version is lightweight enough to run on a single laptop, unlike the traditional HRRR which requires a supercomputer.
Dr. Isidora Jankov is scientific computing branch chief in NOAA's Global Systems Laboratory. She spoke with Fox Weather recently about HRRR-Cast.
"We just actually did estimates the other day, and we plan to run 11 ensemble members that we can fit four ensemble members on one GPU; and it would take seven minutes to produce 48-hour forecasts."
That's considerably faster than the current traditional HRRR model.
Data-driven models like HRRR-Cast learn by analyzing vast amounts of similar historical weather data to identify patterns used to make predictions. Jankov sees the two working side by side.
"Traditional physical models are the key tool for us to understand physical processes, get to the scales that we haven't explored before, simulate these processes and eventually forecast them – which will then produce data sets that we can train AI models to produce solutions faster," she explains. "So, absolutely they need to work hand in hand."
Initial results show that this new AI model excels in producing realistic depictions of storm structure.
Related story:
NOAA Research develops an AI-powered sibling to its flagship weather model
Image compliments of NOAA Global Systems Laboratory
