Google DeepMind's AI weather model correctly called where Hurricane Isaias would come ashore this month, picking the western Florida Panhandle while traditional physics-based models initially pointed the storm toward southeastern Louisiana. Isaias made landfall near Destin, Florida, as a Category 2 hurricane in early October.
The forecast is the latest data point in a run that has made the Google DeepMind model a fixture in official U.S. hurricane forecasting. Miami meteorologist Michael Lowry called it the gold standard model of the last few years.
“That model has been, by far and away, the best performing computer model that we've ever seen.”— Michael Lowry, Meteorologist
The National Hurricane Center began folding the Google DeepMind model into its official forecasts last year. By the end of the 2025 Atlantic season, it was already among the best models at predicting both the path and the strength of storms, according to University of Miami hurricane researcher Brian McNoldy.
Key facts
- 01Hurricane Isaias made landfall near Destin, Florida, as a Category 2 storm in early October 2026.
- 02Traditional models first pointed Isaias toward southeastern Louisiana; the Google DeepMind model picked the western Florida Panhandle.
- 03The National Hurricane Center began using the Google DeepMind model in its official forecasts in 2025.
- 04NHC Director Michael Brennan said intensity forecast errors are about half what they were at the time of Hurricane Katrina.
- 05Google DeepMind and ECMWF run the two most accurate AI weather models so far.
McNoldy described the speed of that rise as something forecasters are not used to: a new entrant quickly becoming the best model available.
“This pretty much came out of the gate in first place, and that's a pretty impressive feat.”— Brian McNoldy, Hurricane researcher, University of Miami
The approach is different from the models forecasters have relied on for decades. Traditional models take in real-time atmospheric observations and solve the physics equations that govern the atmosphere. AI models like Google DeepMind's lean instead on patterns learned from thousands of past storms, starting from the current picture of the atmosphere rather than the governing equations.
That design has a trade-off forecasters are open about. Lowry said researchers do not know why the AI models get a forecast right, or why they get one wrong, and McNoldy said that makes forecasters a little uncomfortable because they like to understand the reasoning behind an answer.
The Isaias result was strongest on track. The Google DeepMind model did not match the hurricane center's blended forecast on intensity. Google DeepMind and the European Centre for Medium-Range Weather Forecasts, known as ECMWF, operate the two most accurate AI models so far.
The hurricane center does not hand its forecast to any single model. It updates every six hours during an active hurricane and blends several AI and traditional models, a consensus approach NHC Director Michael Brennan said reflects the different challenges each storm presents. Brennan said human forecasters provide continuity and translate the numbers into hazard guidance people can act on.
Forecast skill is improving quickly. Brennan said intensity forecast errors are now about half of what they were at the time of Hurricane Katrina. Lowry said AI models can be updated in months or even weeks, while traditional models are updated annually at best, and that AI is producing gains in a year that would previously have taken a decade or more.
“If anything, I think it is a boon for our business.”— Michael Lowry, Meteorologist
Lowry said he does not feel an existential crisis about AI's arrival in forecasting.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
