The future of hurricane forecasting is AI : NPR
As the Atlantic hurricane season draws to a close, the most destructive storm of the year was Hurricane Melissa. The Category 5 hurricane hit Jamaica in late October. It was the most violent storm ever to hit Jamaica, killing dozens of people and decimating many neighborhoods. The forecast was uncertain a few days before arrival on land. But one model in particular succeeded perfectly.
Ricardo Makyn/AFP via Getty Images
hide caption
toggle caption
Ricardo Makyn/AFP via Getty Images
MIAMI — The Atlantic hurricane season, which officially comes to an end Sunday, has lived up to predictions that it will be an active year.
There were 13 named storms and three Category 5 hurricanes. But for the first time in a decade, a hurricane did not make landfall in the United States.
The season’s most destructive hurricane, Melissa, was one of the strongest storms on record in the Atlantic. It hit Jamaica with winds of 300 km/h, devastating communities and killing dozens of people.
However, a week before the hurricane made landfall, forecast models disagreed on its direction. One model that got it right – accurately predicting Melissa’s track and Category 5 intensity – was a new hurricane model based on Google’s DeepMind AI.
James Franklin, former branch chief of the National Hurricane Center, analyzed the performance of forecast models this year and says Google’s DeepMind outperformed them all. “The model worked very, very well, which was very impressive,” he says. “This is the best guidance we’ve seen this year.”
Artificial intelligence has been used in weather forecasting models for some time. Google’s DeepMind, however, marks a significant step forward, suggesting that AI may soon overtake the physics-based models that meteorologists have long relied on.
Models like the Global Forecast System – GFS – developed by NOAA, are based on equations that calculate how wind, humidity and heat move through the atmosphere. Models use these equations to predict what might happen in the atmosphere, including the path and intensity of hurricanes.
Track errors for the 2025 Atlantic hurricane season from the most common forecast models and official NHC forecasts (OFCL – black line). Google DeepMind (GDMI – red line) had the lowest overall forecast error.
James Franklin
hide caption
toggle caption
James Franklin
AI models like Google’s DeepMind, on the other hand, know nothing about physics but instead focus on history. “They were developed to go back and look at historical records and discover patterns and relationships about what happened in the past in very subtle ways that a person wouldn’t be able to extract on their own,” says Franklin.
To develop their hurricane model, Google engineers worked closely with scientists at the National Hurricane Center and the Cooperative Institute for Atmospheric Research at Colorado State University. Kate Musgrave, research scientist at ACEI, analyzes the performance of AI-based models, including the one developed by Google.
She says that in the past, AI models worked well with one part of hurricane forecasting: following the path of a storm, “because that is governed by large-scale influences in the atmosphere. However, the intensity, the strength of a storm, is not captured well in AI models.” But Google’s model did a very good job of predicting intensity, she said, because it added historical data detailing the evolution of past hurricanes.
Musgrave thinks AI modeling could be the future not only for hurricanes, but also for predicting other weather phenomena, from tornadoes to cold snaps.
As for hurricanes, as AI models develop, she believes meteorologists will be able to predict their path and intensity earlier than ever before, which is a vital improvement. She says: “As coasts become more and more populated, we need more time to get people out of the way. Thus, longer-term forecasts become increasingly important.
People walk on a flooded street after Hurricane Melissa in Petit-Goave, Haiti, October 30, 2025. The storm, a devastating Category 5 with sustained winds of 185 mph, killed dozens of people in the Caribbean.
Clarens Siffroy/AFP via Getty Images
hide caption
toggle caption
Clarens Siffroy/AFP via Getty Images
The National Hurricane Center has adopted the new Google DeepMind model, referencing it in many of its forecast discussions, particularly when tracking Hurricane Melissa.
Wallace Hogsett, manager of science operations at the National Hurricane Center, says, “I think it’s clear at this point that AI will be a component of the hurricane forecasting process in the future. » Additional AI models are being developed by NOAA and the European Center for Medium-Range Weather Forecasts. He says: “I expect this pace of innovation to continue. »
But for forecasters accustomed to looking at data on wind, pressure, humidity and sea surface temperature and how they are interpreted by physics-based models, Franklin, a former NHC forecaster, says relying on AI can be destabilizing.
“AI models are like a black box for a forecaster,” he says. “A lot of data comes in. You get a forecast that comes out. But you don’t really know how it happened.”
Although AI models will become increasingly important, Franklin and Musgrave don’t expect them to replace long-standing physics-based models or the judgment of experienced forecasters.
