From the beginning of Hurricane Isaias, meteorologists at the National Hurricane Center made a big bet. Rather than relying on the classic physics-based models that many Americans are familiar with, such as European and GFS, they shaped their predictions to closely match the AI hurricane model developed by Google DeepMind. So far, that gamble has paid off.
This storm forecast reflects the increased reliability of AI weather models overall and marks an incredibly rapid transition to routine use of such products by human forecasters.
The heavy reliance on DeepMind meant the NHC expected Isaias to be a Category 2 from its first advisory on Tuesday, which didn’t even have a name yet. At that point, many physically-based computer models that relied on mathematical formulas describing how the atmosphere would behave to simulate future weather conditions were oscillating between different predictions, but generally predicted weak storms with different trajectories than what DeepMind and other AI models were showing.
If this prediction is correct, it would be a big win for Google and help people in the storm’s path. They would have had more time to prepare for a more powerful blow. Hurricane Isaias was the first hurricane of the tropical Atlantic season, largely due to the influence of Super El Niño in the Pacific Ocean, and set the record for the latest first hurricane of the season in an ocean basin.
NHC forecasters’ confidence in Google DeepMind models in particular (the latest version of which is WeatherNext 3) stems primarily from experience. Last season, the Hurricane Center accurately predicted the rapid intensification of Hurricane Melissa nearly three days before its devastating arrival in Jamaica, based largely on predictions from DeepMind. This is also because the Hurricane Center partnered with Google to train the model and learn more about its capabilities.
As with what is happening in many fields today, the world of meteorology is rapidly advancing with AI, but these models still have advantages when used alongside traditional physical models. AI models are trained based on past weather events and, in the case of DeepMind, also fed with current weather data.
⛅ Get CNN Weather in your inbox
The forecast is just beginning. We bring you expert coverage and the stories behind the weather. So you always know more than just the numbers. Sign up for our newsletter
NHC Director Michael Brennan said DeepMind did very well last season and accurately predicted the intensity of many Pacific storms this season.
Regarding the reliance on DeepMind for Hurricane Isaias, Brennan said that “the AI models are more consistent” from one model run to the next.
“It’s probably a combination of what we’ve seen looking at these physics-based models and some of the inconsistencies there, and the performance of AI guidance to date and the fact that AI guidance has proven to be fairly reliable,” Brennan said.
Part of that evidence comes from a courageous call made by hurricane center forecasters during last year’s Hurricane Melissa.
Last season, Hurricane Center forecasters, relying primarily on DeepMind’s clear predictions, made the first clear prediction that Hurricane Melissa would reach Category 5 strength when it was a Category 1 hurricane with winds of 80 miles per hour. In fact, by some measures, DeepMind even outperformed human forecasters for the entire 2025 Atlantic hurricane season.
As more tropical cyclones rapidly intensify due to global climate change, the Hurricane Center is making more bullish initial predictions for certain storms, including Isaias. Before the AI entered the prediction phase, the NHC took a more stepped approach, gradually increasing the intensity limit as the storm’s strength increased. As a result, official forecasts were often delayed.
That habit seems to have ended. Rapidly intensifying storms are now more common, and modern models can better predict them.
“We’ve had more aggressive intensity predictions across the board in recent years, and we feel confident in trying to make those more aggressive intensity predictions early on, like we did with Melissa last year, but also with Helen and Milton,” Brennan said.
He cited the support of physical computer models for rapid intensification, as well as forecasters’ understanding of the environment in which storms are forming, as additional factors that enable early predictions of rapid intensification.
“It’s not just one thing, but it’s certainly true that AI models are working very well in some cases,” Brennan said. “So this is another tool in the toolbox, so to speak, or another piece of evidence that can help support more aggressive intensity predictions,” he said of the AI model.
Brennan said forecasters aren’t being flippant about relying on DeepMind because they know the training data used for it.
“I think the most important thing is to make sure that the models are trained based on the most accurate data that we have, and we’ve been working with Google to help them train based on our best track analysis. That’s one of the reasons why I think we’ve improved our performance in terms of intensity prediction compared to some of the earlier AI models that did well on track but not on strength,” he said. These models were trained on large datasets that were not specific to hurricanes, he said.
Meanwhile, Hurricane Isaias is heading toward the northeastern Gulf Coast and is likely to make landfall as a hurricane, according to the consensus of most computer models, not just DeepMind.
