An artificial intelligence weather model is changing what forecasters can see before a hurricane reaches land. WeatherNext, developed by Google’s DeepMind and Google Research, helped identify the threat from Hurricane Melissa earlier than existing models, giving communities more time to prepare.
The model’s performance, described in a paper published on Thursday in Nature, points to a practical shift in cyclone forecasting: not replacing forecasters, but giving them another tool when time matters most.
What WeatherNext Saw Before Landfall
In October 2025, a storm system formed over the Caribbean Sea. Forecast models did not agree on what would happen next. One possible outcome was a weaker storm moving toward Haiti. Another was a stronger system heading for Jamaica.
WeatherNext chose the more dangerous scenario. Five days before landfall, it predicted with 80 percent confidence that the storm system would strike Jamaica as a Category 5 hurricane.
That storm became Hurricane Melissa. It caused catastrophic flooding and landslides across Jamaica. The earlier AI forecast helped forecasters warn communities sooner, giving people and emergency planners more time to act before conditions deteriorated.
Researchers say the model can predict cyclones with unprecedented accuracy. On average, it gives forecasters a day more lead time than existing models. Put plainly, its forecasts three days out are as accurate as previous models’ forecasts two days out.
Why One Extra Day Matters
Hurricane forecasting is not only about drawing a line on a map. Emergency decisions must be made before the worst impacts arrive, and those decisions take time to organize.
Mike Brennan, director of the US National Hurricane Center, said, “Even a few hours can make a difference.” Evacuations, supply staging, and moving response resources all depend on forecasts that are accurate early enough to be useful.
He also described the value of extending reliable forecasts by a day: “Time is really golden when it comes to those types of decisions, so the ability to push forecast accuracy out as much as a day beyond what we’ve previously been able to do is really valuable.”
The researchers said that, historically, gaining a full day in forecast lead time would take a decade of work. That makes WeatherNext’s result notable, especially because hurricane decisions often carry high consequences if the forecast is wrong.
The Hard Part Is Intensity
Cyclones are difficult for AI because the most extreme events are rare. Machine learning systems usually need large amounts of training data, but rare weather disasters do not provide the same volume of examples as everyday weather patterns.
Ferran Alet, a research scientist at Google DeepMind and one of the paper’s lead authors, explained the approach this way: “We don’t have that much cyclone data, but we have a lot of weather data.” He added, “So what we did was train a model to be both good at weather as well as cyclones.”
Hurricanes also operate across very different scales. Kate Musgrave, tropical cyclone group lead at the Cooperative Institute for Research in the Atmosphere and an author on the paper, said storm track and storm intensity depend on different kinds of information.
Predicting track requires a broad view of the atmosphere, including details such as cold fronts and prevailing winds. Predicting intensity requires a much more local view, including atmospheric and ocean conditions near the storm.
Musgrave said earlier AI models had performed well on track, but struggled with intensity. That distinction matters because a storm can move from a lower category to a major hurricane quickly. Hurricane Melissa marked the first time the National Hurricane Centre was able to predict a Category 5 hurricane while the storm was still at Category 1.
A Powerful Model With Unanswered Questions
Before WeatherNext was used in live forecasts, researchers tested it on retrospective data. Musgrave said the results were strong enough that the team was skeptical the same performance would appear in real-time use. When forecasters began using it in operations, the model’s performance held up.
One surprising detail is that WeatherNext uses much lower-resolution atmospheric data than traditional models typically need to forecast storm intensity. Alet said the reaction from the community was shock, because the result suggests lower-resolution inputs contain more useful signal than previously believed.
The researchers do not yet fully understand what the model is detecting. Alet described it directly: “It’s a black box at the end of the day, but that gives physicists a signal that something is happening that was not previously understood.”
WeatherNext also does not produce a single forecast. It generates a range of possible storm scenarios, helping forecasters think through how small changes can lead to larger differences later. Last year, the AI model created 50 scenarios per storm; now, it generates 1,000.
Musgrave said that is beyond what existing numerical models can do with available computing power. For forecasters, the value is not just one answer, but a broader set of plausible outcomes to compare with other models.
Why Human Forecasters Still Matter
Brennan described DeepMind’s model as a valuable addition to the forecasting toolbox, but he cautioned against treating any one model as a guaranteed answer. A model that performed well last year, or for one storm, may not be the best guide for the next season or the next storm.
That is why the human role remains central. “A hurricane is not just a track or an intensity forecast,” Brennan said. “It requires experts to translate that into what the impacts are going to be—and it’s the impacts that kill people.”
Google DeepMind has also announced that it is open-sourcing the WeatherNext models used during hurricane season. Researchers will be able to use and improve them, and Alet hopes that broader access could reveal new insights into how cyclones work.
For now, WeatherNext’s biggest lesson is practical. Better AI forecasting can buy time. In hurricane response, that time can shape warnings, preparations, and decisions before a dangerous storm arrives.