NOAA puts AI global weather models into operation, led by a retrained GraphCast

NOAA/AIGFS

NOAA launched three operational AI forecast systems: AIGFS (one best-guess forecast), AIGEFS (a 31-member AI ensemble) and HGEFS (a 62-member hybrid of AI and physics runs). AIGFS started from Google DeepMind's GraphCast, fine-tuned on NOAA's own analyses. It matches or beats the physics GFS on large-scale patterns and storm tracks at a tiny fraction of the computing cost, but is weaker on hurricane intensity.

• A 16-day AIGFS forecast uses 0.3% of the physics GFS's computing and takes about 40 minutes.
• AIGEFS early results add 18 to 24 hours of useful forecast skill, at about 9% of the physics ensemble's compute.
• HGEFS beat both the AI-only and physics-only ensembles on most of NOAA's verification scores.
• Known weakness: v1.0 forecasts hurricane intensity worse than GFS; NOAA said later versions will fix it.