Google DeepMind's new weather model delivered highly accurate forecasts during the 2025 Atlantic hurricane season, surpassing traditional systems. The US Global Forecasting System, by contrast, performed poorly. Preliminary data highlights the AI's edge in tracking and intensity predictions.
The 2025 Atlantic hurricane season, which involved 13 named storms, concluded with notable advancements in forecasting technology. Google DeepMind’s Weather Lab began releasing cyclone track forecasts in June 2025, and the model quickly proved superior. According to preliminary analysis by Brian McNoldy, a senior researcher at the University of Miami, Google's DeepMind model (GDMI) achieved the lowest mean position errors across forecast horizons from 0 to 120 hours.
In comparisons, the US National Weather Service's Global Forecast System (GFS, denoted as AVNI) ranked as the worst performer. At 120 hours, Google's model reported an error of 165 nautical miles, while the GFS reached 360 nautical miles—more than double. The AI system even outperformed the National Hurricane Center's official forecasts (OFCL) and consensus models like TVCN and HCCA. It also excelled in intensity forecasting, capturing fluctuations in hurricane strength.
This marks the first full season for Google's AI, which relies on neural networks rather than traditional physics-based computations. 'The beauty of DeepMind and other similar data-driven, AI-based weather models is how much more quickly they produce a forecast compared to their traditional physics-based counterparts that require some of the most expensive and advanced supercomputers in the world,' said Michael Lowry, a hurricane specialist and author of the Eye on the Tropics newsletter. 'Beyond that, these ‘smart’ models with their neural network architectures have the ability to learn from their mistakes and correct on-the-fly.'
The GFS's poor showing remains unexplained. Lowry noted, 'It’s not immediately clear why the GFS performed so poorly this hurricane season.' Speculation points to potential data collection lapses from government cuts, though this would likely affect other models too. The model's upgrades since 2019 appear to have failed, widening the gap with competitors. Official evaluations from the National Hurricane Center are pending, but early results suggest AI will transform hurricane forecasting.