Google has a new AI model and website for forecasting tropical storms

The Verge

a day ago

Google is using a new AI model to forecast tropical cyclones and working with the US National Hurricane Center (NHC) to test it out.
Google DeepMind and Google Research launched a new website today called Weather Lab to share AI weather models that Google is developing. It says its new, experimental AI-based model for forecasting cyclones — also called typhoons or hurricanes when they reach a certain strength — can generate 50 different scenarios for a storm's possible track, size, and intensity up to 15 days in advance. The NHC is working with Google to evaluate the effectiveness of the model.
The collaboration comes after the Trump administration and DOGE slashed the National Weather Service's staff and capacity for federal climate and weather research. Other companies and weather agencies are also exploring whether AI can improve forecasts, but technological advances so far don't eliminate the need for traditional weather models.
Google released a research paper today, which has yet to be peer-reviewed, on how its tropical cyclone model works. It claims that its model's predictions are at least as accurate as those of traditional physics-based models. We'll have to see what the National Hurricane Center's rating of it is as the Atlantic hurricane season churns through November.
For now, the aim is to strengthen NHC's forecasting in order to give people more accurate warnings and time to prepare for a storm. According to Google, its model's five-day predictions for cyclone tracks in the North Atlantic and East Pacific were 87 miles (140 km) closer, on average, to the storm's actual track than predictions from the European Center for Medium-Range Weather Forecasts (ECMWF) in 2023 and 2024.
Weather Lab's interactive website lets people see how AI models compare to the ECMWF's physics-based models. But Google is emphasizing that its website is just a research tool for now — not something the public should rely on for forecasts.
Google's cyclone model is trained on data from Europe's ERA5 archive, which includes hundreds of millions of observations collected by weather agencies around the world combined with predictions from a traditional weather model. The company also used ERA5 to train its previous AI weather prediction model GenCast. That model outperformed one of ECMWF's leading physics-based models 97.2 percent of the time, according to research published in the journal Nature in December 2024.
Animation showing the Google model's prediction for Cyclone Alfred when it was a Category 3 cyclone in the Coral Sea. The model's ensemble mean prediction (bold blue line) correctly anticipated Cyclone Alfred's rapid weakening to tropical storm status and eventual landfall near Brisbane, Australia, seven days later, with a high probability of landfall somewhere along the Queensland coast. Credit: Google
The company says it's also working with the Cooperative Institute for Research in the Atmosphere at Colorado State University and other researchers in the UK and Japan to improve its AI weather models.
The importance of real-world observations and older weather models in developing these new kinds of tools is one reason why AI is so far only poised to assist traditional weather forecasting instead of replacing it. Adjusting to a changing climate will also hinge on the ability to collect and analyze new data on increasingly extreme and erratic weather events.
How well the US can keep up with this kind of research, however, is a growing concern under the Trump administration. DOGE's rampage through the federal government has taken its toll at the federal agency that houses the NHC and the National Weather Service, the National Oceanic and Atmospheric Administration (NOAA). The National Weather Service reduced the number of weather balloon launches after staffing cuts, and NOAA is increasingly relying on weather balloon data from private companies. Project 2025 called for dismantling NOAA — which leads climate research on top of providing weather forecasts — and privatizing much of its services. Some advocates are raising alarm over the prospect of turning weather forecasts into a paid product instead of a free service available to anyone and everyone.
'For a long time, weather has been viewed as a public good, and I think, you know, most of us agree with that … Hopefully we can contribute to that, and that's why we're trying to kind of partner with the public sector,' Peter Battaglia, a research scientist at Google DeepMind, said in a press call when The Verge asked about concerns surrounding privatizing weather services.
Tellingly, Google's announcement today doesn't mention the climate crisis like the company has in previous launches for this kind of program. 'As climate change drives more extreme weather events, accurate and trustworthy forecasts are more essential than ever,' it said in a December 4 announcement for GenCast.