HurriLab’s post-storm analysis reports that Google DeepMind’s AI model suite came closest to Hurricane Isaias’s landfall location and forecast its timing within a few hours more than three days ahead. But track and intensity are different forecast tasks: National Hurricane Center director Michael Brennan said the official NHC forecast had 20% to 40% more skill than DeepMind’s for Isaias’s intensity.
The National Hurricane Center reported that Isaias made landfall near Destin, Florida, at about 8:30 p.m. CDT on October 9, 2026. The storm had maximum sustained winds of 105 mph and an estimated minimum central pressure of 971 millibars.
How the track forecasts compared
HurriLab evaluated forecast ensembles, groups of model runs that represent possible storm scenarios. Its fair energy score weighs track error and how closely ensemble members agree, while adjusting for differences in ensemble size.
In HurriLab’s preliminary Isaias analysis, the traditional European IFS ENS and U.S. GEFS ensembles had track errors more than twice their respective five-year averages. The analysis also found that European AI and GFS AI systems outperformed their traditional counterparts on track for this storm.
| Forecast task | Reported result for Isaias | Systems involved |
| Track | DeepMind was closest to the landfall location and within a few hours of the actual landfall time in a forecast made more than three days ahead. | Google DeepMind’s AI model suite; traditional IFS ENS and GEFS ensembles had track errors more than twice their respective five-year averages. |
| Intensity | Brennan said the official NHC forecast had 20% to 40% more skill than DeepMind’s. HurriLab reported that the NHC forecast had the lowest overall intensity errors. | Google DeepMind and the official NHC forecast; HurriLab also reported strong results from HWRF and HMON beyond 36 hours. |
These are separate measures, not a single contest with one overall winner. A track forecast estimates where a storm will move; an intensity forecast estimates how strong it will be. The reported results say DeepMind did especially well on Isaias’s track, while the official NHC forecast performed better on intensity.
Why the official forecast used more than one model
The NHC uses guidance from multiple AI and traditional models. Human forecasters interpret those predictions and decide what to communicate. Brennan has described the agency’s approach as combining the most useful model guidance because each storm presents different forecasting challenges.
That process matters when reading an AI forecast result: the model comparison concerns guidance, while the official forecast is produced through a broader forecasting process. Michael Lowry, a meteorologist whose analysis discussed the Isaias comparison, has also described AI models as difficult to interpret. A close forecast can be useful without explaining the physical reasons behind a storm’s behavior.
What Isaias says about AI forecasts
One storm cannot establish how a system will perform across hurricanes. NPR reported that Brennan said Google DeepMind did not perform well tracking Hurricane Simon in the eastern Pacific. The contrast with Isaias is a reminder that results can vary by storm and by forecast task; it does not provide a general success rate.
For a separate look at Google DeepMind’s WeatherNext Cyclones work, see NeoTeo’s earlier coverage.
