DeepMind’s cyclone AI reportedly adds a day to hurricane forecasts

Google DeepMind’s WeatherNext Cyclones model reportedly delivered about a day more predictive accuracy for tropical cyclones in a 2023–2025 evaluation, while forecasting track, intensity and wind structure. That could give forecasters more time to prepare, though it is a model result, not a guarantee of extra warning in every storm.

Google DeepMind Watch analysis

What happened

Gizmodo reports that researchers evaluated the model with the US National Hurricane Center and other institutions using historical data from 2023 to 2025. The reported average improvement was roughly one extra day’s worth of predictive accuracy. DeepMind research scientist Tom Andersson told Gizmodo that the model was also tested during the 2025 Atlantic hurricane season.

WeatherNext Cyclones was designed to tackle a familiar forecasting difficulty: predicting where a cyclone will go while also estimating how strong it will become. Gizmodo says the model can generate up to 1,000 scenarios, an approach intended to capture rare but physically plausible “gray swan” storms. The outlet also reports that the model is open-source and does not require a supercomputer to run.

Why it matters

A better forecast of a storm’s path is only part of the job. Intensity and rapid strengthening matter too, because communities need useful warning before conditions deteriorate. A reported improvement in both areas could give forecasters another tool for weighing possible outcomes and communicating risk.

The concrete takeaway is promising but bounded: the reported extra day is an average from the evaluation, not a promise that every community will receive 24 additional hours’ notice. Forecasts remain uncertain, and a model’s value depends on how it performs in real operations and how forecasters use its outputs.

Our read

This is the kind of AI result that matters beyond a leaderboard: better cyclone forecasts can give people and emergency services more time to act. The exciting part is the reported gain; the part worth keeping an eye on is whether it holds across different basins, storms and forecasting teams. One more day is a substantial claim, so it deserves real-world scrutiny as well as applause.

What to watch

• Whether the reported performance is borne out across more storms and regions.
• How forecasters use the model’s multiple scenarios in operational warnings.
• Whether the reported open-source access leads to wider use by forecasting teams with fewer computing resources.

Discussion spark: Should forecasting agencies make an AI model with a reported average one-day accuracy gain part of routine operations, or require more evidence across regions before relying on it?

not affiliated with, endorsed by, or operated by Google or Google DeepMind