MIAMI — The Atlantic hurricane season, which draws to an official close on Sunday, fulfilled forecasts it would 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 U.S.
The season's most destructive hurricane, Melissa, was one of the strongest Atlantic storms ever. It slammed Jamaica with 185 mph winds, devastating communities and killing dozens of people.
A week before the hurricane made landfall, however, forecast models disagreed on where it would go. One model that got it right — accurately predicting Melissa's path and its category 5 intensity — was a new one: Google's DeepMind AI-based hurricane model.
Loading...
James Franklin, a former branch chief at the National Hurricane Center, analyzed how the forecast models performed this year, and says Google's DeepMind outshone them all. "The model performed very, very well, which was very impressive," he says. "It was the best guidance we saw this year."
Artificial intelligence has been used in weather forecast models for some time. Google's DeepMind, though, marks a significant step forward, one that suggests AI may soon overtake the physics-based models meteorologists have long relied on.
Models like the Global Forecast System — the GFS — developed by NOAA, are based on equations that calculate how wind, moisture and heat move around in the atmosphere. The models utilize these equations to predict what might happen in the atmosphere, including the track and intensity of hurricanes.
AI models like Google's DeepMind, on the other hand, don't know anything about physics but focus instead on history. "They have been developed to go back and look at historical records and tease out patterns and relationships about what has happened in the past in very subtle ways that a person could not extract on their own," Franklin says.
To develop their hurricane model, engineers at Google worked closely with scientists at the National Hurricane Center and Colorado State University's Cooperative Institute for Research in the Atmosphere. Kate Musgrave, a research scientist at CIRA, analyzes the performance of AI-based models, including the one developed by Google.
She says in the past, AI models did well with one part of a hurricane forecast — tracking the path of a storm, "Because that is governed by large-scale influences in the atmosphere. Intensity, however, how strong a storm will be, is not captured well in the AI models." But the Google model did very well in forecasting intensity, she says, because it added historical data detailing how past hurricanes developed.
Musgrave believes AI modeling may be the future not just for hurricanes, but also in forecasting other weather phenomena, everything from tornadoes to cold snaps.
As for hurricanes, as AI models develop, she believes meteorologists will be able to forecast tracks and intensity earlier than ever before, a vital improvement. She says, "As the coastlines get more populated, we need more and more time to get people out of the way. So, forecasts further and further into the future become more important."
The National Hurricane Center has embraced the new Google DeepMind model, referencing it in many of its forecast discussions, particularly as it tracked Hurricane Melissa.
Wallace Hogsett, a science operations officer at the National Hurricane Center, says, "I think it's clear at this point that AI will be a component of the hurricane forecast process going forward." Additional AI models are under development by NOAA and the European Centre for Medium-Range Weather Forecasts. He says, "I expect that this pace of innovation will keep up."
But for forecasters used to seeing data on wind, pressure, humidity and sea surface temperature and how it's interpreted by the physics-based models, former NHC forecaster Franklin says relying on AI can be unsettling.
"AI models are something like a black box to a forecaster," he says. "A lot of data goes in. You get a forecast that comes out. But you don't really know how it came up with that."
Although AI models will become increasingly important, Franklin and Musgrave don't expect them to replace either long-standing physics-based models or the judgment of experienced forecasters.
Copyright 2025 NPR
