Manufacturing
WeatherNext 3 is live across Google's consumer products as of September 3. This is the first time that Google DeepMind's AI weather forecasting model has directly powered the weather information users see in Google Search, Google Maps, and the Gemini app simultaneously.
The model has a measurable advantage over WeatherNext 2. Resolution improved from a 25-kilometer grid updated every six hours to 5-kilometer resolution updated every hour, a five times improvement in spatial sharpness combined with a six times improvement in update frequency, according to the company.
The leap is enabled by a fundamental change in what the model learns from: WeatherNext 3 ingests live geostationary satellite data on an hourly basis, bypassing the 6-hour data lag that traditional numerical weather prediction systems carry.
Most AI weather models, including WeatherNext 2, train on data produced by those traditional systems, inheriting their lag as a result. WeatherNext 3 trains directly on raw satellite observations, giving it a more current view of the atmosphere at the moment a forecast is generated.
"Weather is chaotic, and so small differences really start to perturb massively," said Ferran Alet, Staff Research Scientist Manager at Google DeepMind. "Machine learning targets the problem we are really solving, which is approximate noisy physics from incomplete information and finite compute, and so it learns patterns from a lot of data."
The model also forecasts 100-meter wind speeds, roughly at turbine height, for wind energy output planning, alongside high-resolution cloud cover and solar radiation estimates for solar farm production projections, according to the company.
