Google DeepMind's WeatherLab hurricane forecasting system outperforms all other forecasting models in predicting the path of Hurricane Erin.
The hurricane forecasting system 'WeatherLab,' which Google DeepMind launched in June 2025 in collaboration with the National Hurricane Center, predicted the path of Hurricane Erin, which struck the US mainland in August 2025, with extremely high accuracy, surpassing all other forecasting models.
Google's AI model just nailed the forecast for the strongest Atlantic storm this year - Ars Technica
WeatherLab is a hurricane forecasting system launched in June 2025 by Google DeepMind in collaboration with Google Research and the National Hurricane Center. The system features the latest experimental AI-based tropical cyclone model based on the latest probabilistic neural network models, and a demo has been released showing it can make predictions with a fairly high degree of accuracy.
Google launches 'Weather Lab' that uses AI to predict and warn of typhoons, hurricanes, and cyclones - GIGAZINE
Andrew Brady, an AI engineer who developed the weather forecasting system STORM-Net , reported that WeatherLab's accuracy in predicting the path of Hurricane Erin was 'astonishing.' Brady described the results as 'astonishing.'
I chose Google Deepmind (GDMI) against a slightly different group of models that I thought were more representative (except no EMXI because it's not in the public decks). For track, GDMI was best through 72 h, beat TVCN at all times, but trailed HAFS after 72 h. Not bad at all.
— James Franklin ( @franklinjamesl.bsky.social ) August 26, 2025, 9:58 AM
[image or embed]
James Franklin, former head of the National Hurricane Center's Hurricane Specialist Unit, published a graph comparing the track prediction results (error) of leading models with those of other models, showing that the GDMI (Google DeepMind Weather Lab model) (shown in red) had the lowest error up to 72 hours after the forecast began.
I chose Google Deepmind (GDMI) against a slightly different group of models that I thought were more representative (except no EMXI because it's not in the public decks). For track, GDMI was best through 72 h, beat TVCN at all times, but trailed HAFS after 72 h. Not bad at all.
— James Franklin ( @franklinjamesl.bsky.social ) August 26, 2025, 9:58 AM
[image or embed]
We can see that GDMI had the least error not only in predicting the path but also in predicting the intensity up to 72 hours.
For intensity, GDMI again beat everything else through 72 h, beat the consensus at all time periods, but trailed one of the HAFS after 72 h. Again, pretty impressive.
— James Franklin ( @franklinjamesl.bsky.social ) August 26, 2025, 10:01 AM
[image or embed]
The news site Ars Technica writes that WeatherLab's model has already demonstrated skill on par with the best physics-based models, and that with further improvements, it could become the 'gold standard' for certain types of weather forecasting.
Related Posts:
in Note, Posted by logc_nt