NOAA Deploys AI-Driven Global Weather Models in Major Forecasting Shift

The National Oceanic and Atmospheric Administration (NOAA) has launched a new generation of operational, artificial intelligence–driven global weather models, marking a significant shift in how forecasts are produced, delivered, and scaled. The agency says the new systems dramatically improve speed and efficiency while maintaining—or in some cases improving—forecast accuracy.

“NOAA’s strategic application of AI is a significant leap forward in American weather model innovation,” said NOAA Administrator Neil Jacobs. “These models represent a new paradigm, delivering improved large-scale and tropical track forecasts faster and at far lower computational cost.”

The newly deployed suite includes three interconnected models. The Artificial Intelligence Global Forecast System (AIGFS) uses AI to generate forecasts comparable to NOAA’s traditional Global Forecast System while consuming up to 99.7% less computing power. A full 16-day forecast can now be completed in roughly 40 minutes, allowing forecasters to access critical guidance much sooner. Early results show improved performance on large-scale weather patterns and reduced tropical cyclone track errors, though NOAA notes that storm intensity forecasting remains an area for improvement.

The Artificial Intelligence Global Ensemble Forecast System (AIGEFS) adds probabilistic forecasting through a 31-member AI ensemble. The model produces a range of possible outcomes similar to NOAA’s existing ensemble system but extends forecast skill by 18 to 24 hours while using just 9% of the computing resources.

The most innovative element is the Hybrid Global Ensemble Forecast System (HGEFS), which combines the AI-based ensemble with NOAA’s traditional physics-based ensemble into a 62-member “grand ensemble.” By blending AI and physical modeling approaches, NOAA says the hybrid system consistently outperforms either method alone and provides a more robust representation of forecast uncertainty. To NOAA’s knowledge, it is the first operational hybrid AI–physics ensemble system in the world.

The models are the product of Project EAGLE, a collaborative effort spanning NOAA’s National Weather Service, research laboratories, and external partners in academia and industry. The systems were initially built on Google DeepMind’s GraphCast model and then further trained using NOAA’s own data, significantly improving performance when paired with NOAA’s forecasting infrastructure.

While traditional physics-based models will remain essential, NOAA’s move signals a broader transition toward AI-augmented forecasting. For forecasters and the public alike, the shift promises faster guidance, clearer uncertainty ranges, and more timely warnings for high-impact weather events—delivered at a fraction of the computational cost of legacy systems.