Can We See Hurricanes Coming Sooner?

When a hurricane is heading towards a populated area, knowing where it will go is only part of the challenge. Forecasters also need to understand how powerful it could become, how far its damaging winds might extend and, crucially, how early they can be confident enough in those predictions for people to act.

New research into Google’s WeatherNext Cyclones suggests AI could help move some of that certainty earlier. Research published in Nature found that the model achieved an average lead-time advantage of at least a day when predicting the track, intensity and wind structure of tropical cyclones. In broad terms, its three-day forecasts could reach roughly the accuracy that leading operational models previously achieved at around two days.

That doesn’t mean every hurricane warning can suddenly be issued a day earlier. Forecasting involves bringing together different models, observations and expert judgement, and there will always be uncertainty around how a storm develops. But when preparations can involve evacuating communities, moving emergency resources and getting supplies into place, having useful information earlier creates something particularly valuable: more time to decide what to do.

When A Day Makes A Difference

Hurricane Melissa provided a striking example during the 2025 Atlantic hurricane season. Several days before the storm reached Jamaica, there was still uncertainty around how powerful it would become, but WeatherNext was already showing the possibility of extreme intensification. Five days before landfall, it predicted a Category 5 landfall in Jamaica with 80% confidence, rising to almost 100% three days before the storm arrived.

Melissa ultimately did make landfall in Jamaica as a Category 5 hurricane, but the interesting part isn’t simply that the prediction turned out to be right. It was that the possibility appeared strongly enough, early enough, to be useful while preparations could still be made.

The US National Hurricane Center used the AI forecasts alongside physics-based weather models, satellite observations, aircraft data and the judgement of experienced forecasters. Rather than replacing any of those existing sources, the model added another view of how the storm might develop. This helped forecasters build a picture of what Jamaica could be facing while authorities still had time to mobilise resources and coordinate evacuations.

Forecasting Before Certainty

Weather forecasting has always involved dealing with uncertainty. A hurricane doesn’t follow a predetermined route, and relatively small changes in atmospheric conditions can affect where it travels, how quickly it strengthens and which communities eventually find themselves in its path. Forecasters therefore look at a range of possible outcomes, with confidence usually increasing as the storm gets closer and more information becomes available.

That’s where the potential of AI becomes interesting. If better modelling can bring some of that confidence forward, the benefit isn’t simply a more accurate forecast on a screen. It’s a longer window between recognising that something serious could happen and reaching the point where it actually does.

For the people making decisions on the ground, that window matters;

• Evacuating a community takes time
• as does positioning emergency services
• preparing hospitals
• securing infrastructure
• and making sure people understand what’s coming.

An improvement that sounds fairly technical in a research paper can translate into additional hours in which those decisions are easier and safer to make.

There’s still no model that can tell us exactly what a hurricane will do days in advance, and AI doesn’t remove the uncertainty that comes with forecasting something as complex as the weather. What it may be able to do is help forecasters see the possibilities more clearly while there is still time to respond to them.