DeepMind WeatherNext AI model adds about a day of hurricane forecast lead time, Nature study finds

Google DeepMind and Google Research's WeatherNext AI model can give cyclone forecasts roughly a day more lead time than existing models on average, according to a paper published Thursday in Nature and reported by Ars Technica. Researchers say predictions three days out are about as accurate as prior models two days out. Historically, they said, advancing forecasts by a day would take about a decade of conventional work. Ars Technica recounts Hurricane Melissa in October 2025, when WeatherNext five days before landfall predicted with 80 percent confidence that a Caribbean system would hit Jamaica as a Category 5 hurricane. Mike Brennan, director of the US National Hurricane Center, told Ars that even a few hours can matter for evacuations and staging supplies. Ferran Alet of Google DeepMind said the team trained the model to handle both general weather and rare cyclones because cyclone-specific training data are limited.