The National Oceanic and Atmospheric Administration has rolled out AI-driven forecasting models that it says deliver faster and more accurate weather predictions while slashing computing needs. The Artificial Intelligence Global Forecast System and its ensemble counterpart can generate 16-day outlooks in about 40 minutes using a fraction of the computing power of traditional physics-based models. NOAA officials said the systems, trained on decades of analysis data from legacy models, won’t replace numerical approaches but will run alongside them in a hybrid setup to extend forecast skill and speed delivery to forecasters. The agency estimates 91% to 99% lower real-time compute costs and potentially longer-lasting forecast value by up to a day, while cautioning that training the models is energy-intensive and that hurricane tracking and ensemble diversity still need refinement.
Related articles:
– Pangu-Weather: AI predicts global weather with high accuracy (Nature)
– ClimaX: A foundation model for weather and climate (Microsoft Research)
