National Hurricane Center leaned on Google DeepMind AI model to forecast Hurricane Isaias

The National Hurricane Center’s track forecast for Hurricane Isaias favored the eastward solutions of AI-based models, including the Google DeepMind ensemble mean, over physics-based models farther west, according to its Oct. 7 forecast discussion, issued about two and a half days before the storm struck the Florida Panhandle as a Category 2 hurricane on Oct. 9. In a post-storm review, weather.com meteorologist Jonathan Belles found the DeepMind model did well on track and intensity several days out, while noting that NOAA’s HMON and HWRF hurricane models also did well and that DeepMind missed a band of very strong winds on the storm’s western side. Preliminary verification by weather-tech firm HurriLab, reported by hurricane specialist Michael Lowry, found the physics-based European and American GFS ensembles too far west and a day or more late on landfall.