National Hurricane Center favors Google AI for a stronger Isaias forecast

The National Hurricane Center leaned heavily on Google DeepMind’s AI guidance for its early Hurricane Isaias forecasts, favoring a stronger storm than many traditional models predicted. CNN’s October 8 reporting describes a consequential choice: forecasters projected Category 2 strength from their first advisory, before the developing storm even had a name.

The early forecasts pulled in different directions

The disagreement covered both where Isaias would go and how strong it would become. WDSU meteorologist Derek Sibley’s comparison showed AI guidance favoring the eastern side of the forecast. Traditional models generally leaned west, with projected tracks over Mississippi and Louisiana. The Hurricane Center’s forecast leaned toward the AI guidance.

Those systems reach predictions differently. Physics-based models use equations describing the atmosphere to simulate future weather. AI models learn from past weather events; DeepMind’s also receives current weather data. During Isaias’s early development, many traditional models shifted between projections and generally predicted a weaker storm than DeepMind did.

The early intensity call departs from an older forecasting pattern. CNN described the center’s previous approach as raising predicted strength in steps as a storm intensified, which often left official forecasts behind its development. Brennan said the agency has made more aggressive intensity forecasts for several years, citing Helene and Milton as well as Melissa.

Melissa helped earn forecasters’ confidence

The decision drew on experience from the previous season. For Hurricane Melissa in 2025, forecasters predicted Category 5 intensity while it was still a Category 1 hurricane with 80-mph winds. They relied largely on DeepMind’s projections. CNN reported that the center anticipated Melissa’s rapid strengthening almost three days before its arrival in Jamaica.

There was also a training connection. The Hurricane Center worked with Google to train the model using its best-track analysis—the agency’s storm records. Director Michael Brennan credited that work with improving intensity predictions. Earlier AI models, he said, had learned from broad datasets that were not hurricane-specific and performed better on track than strength.

Sibley identified 2025 as the first hurricane season in which the center incorporated AI model guidance into its forecasts. His performance comparison also showed where DeepMind’s advantage stopped: it beat the center’s official track forecasts except at day five.

Sibley reported that DeepMind beat traditional American and European models on storm-track forecasts, especially 12 to 96 hours ahead. For intensity, DeepMind and the Hurricane Center were nearly equal, with Google doing better between 36 and 60 hours. Track performance was not an across-the-board advantage.

The AI models have shown more consistency

Michael Brennan, National Hurricane Center director, speaking to CNN about Isaias

Human forecasters still weigh the evidence

Brennan’s confidence extends beyond last season’s Atlantic storms. He told CNN that DeepMind had also accurately anticipated the intensity of many Pacific storms this season. CNN identifies WeatherNext 3 as the newest iteration of DeepMind’s weather model; the Isaias decision concerns its use in forecasting, not a new model launch.

The center still uses AI alongside physics-based models. Brennan also cited traditional model support and forecasters’ understanding of a storm’s environment as reasons for making stronger intensity predictions earlier. AI supplies another piece of evidence, not the entire judgment.

As of October 8, most computer models—not just DeepMind—pointed toward Isaias reaching the northeastern Gulf Coast and likely making landfall as a hurricane. The remaining test is the storm itself: that landfall expectation is still a prediction, rather than a completed outcome.