Why did hurricane forecasters bet big on an AI model to predict Hurricane Isaias?

WeatherNext 3 gave forecasters a strong early signal on Isaias by learning patterns from historical weather and updating from live satellite observations.

For Hurricane Isaias—the Atlantic season’s first hurricane, during the 2026 super El Niño—National Hurricane Center forecasters leaned on Google DeepMind’s WeatherNext 3 while the European and GFS physics models were split, calling Category 2 strength from the first advisory. The AI does not solve atmospheric equations step by step. It learns relationships from decades of historical weather, then initializes every hour from live geostationary-satellite data at 5 km resolution; Google reports rainfall errors up to 50% lower than numerical baselines.

Trust also came from experience: the same model lineage anticipated Hurricane Melissa’s rapid intensification roughly three days before it reached Jamaica last season. WeatherNext now helps power weather information in Google Search, Gemini, and Maps, so sharper day-ahead and rainfall forecasts can reach ordinary users quickly. But one accurate storm does not eliminate uncertainty: human forecasters still compare models, observations, and local conditions, and official warnings and evacuation orders—not an app’s single forecast—should guide safety decisions.