Google DeepMind AI Beat Physics Models on Hurricane Isaias
TL;DR
- Hurricane Isaias made landfall near Destin, Florida as a Category 2 storm in early October 2026 on a track Google DeepMind's AI model had called while physics models pointed toward Louisiana.
- The National Hurricane Center incorporated Google DeepMind's model into official forecasts in 2025, and it topped Isaias's path forecast but trailed the NHC blend on storm intensity.
- Meteorologist Michael Lowry called it 'the best performing computer model that we've ever seen,' while University of Miami researcher Brian McNoldy said forecasters cannot explain why it works.
Hurricane Isaias came ashore near Destin, Florida as a Category 2 storm in early October after Google DeepMind's AI weather model forecast landfall on the western Florida Panhandle while conventional physics-based models initially pointed toward southeastern Louisiana, NPR reports.
"That model has been, by far and away, the best performing computer model that we've ever seen," said Michael Lowry, a Miami-based meteorologist who has advised FEMA and the National Hurricane Center. The Hurricane Center began incorporating Google DeepMind's model into its official forecasts in 2025, and by the end of that Atlantic season it already ranked among the top predictors of both path and strength.
The limits showed up on intensity. DeepMind's model topped the field on Isaias's track, but it trailed the NHC's blended forecast in projecting how strong the storm would be at landfall.
What unsettles the forecasters using it is that they can't explain the gains. "It's something we're not used to – to have a model, the new kid on the block, very quickly becoming the best model," said Brian McNoldy, a hurricane researcher at the University of Miami, who added that "we don't know why they might get a forecast right. We also don't know why they might get a forecast wrong." Lowry noted the AI models "don't really know anything about the atmosphere other than the current picture," learning patterns from historical storm data rather than solving physics equations in real time.
NHC Director Michael Brennan said the agency's answer is a blend. "Every storm has different forecast challenges," he told NPR, saying forecasters "take that consensus approach and use that blend of the best models" and layer in human judgment to translate the output for the public. Lowry said the pace of AI gains is what he cannot ignore: advances "that would have taken us 10 years, a decade or more, to make" now arrive in months or weeks. The storm lands inside a run of coverage we've been tracking closely, with Google the subject of 194 AI Weekly stories in the last 90 days, though few of them sit inside a federal life-safety mission the way this one does.
Originally reported by npr.org
Read the original article →Original headline: Google DeepMind's AI Weather Model Pinned Hurricane Isaias on Florida Panhandle Days Before Physics Models