A new global weather forecasting model is promising faster, more localized forecasts, with updates generated every hour and spatial resolution reaching as fine as 5 kilometers.

WeatherNext 3, developed by Google DeepMind and Google Research, uses live geostationary satellite imagery and ground station observations to generate forecasts without relying solely on traditional computationally intensive numerical weather prediction methods.

The model delivers surface temperature and moisture forecasts at resolutions of up to 5 kilometers, while surface wind forecasts can reach 10-kilometer resolution.

Developers said WeatherNext 3 can reduce errors on key precipitation benchmarks by as much as 50 percent compared with traditional numerical forecasting baselines.

It also produces specialized information for the renewable energy sector, including wind speeds at turbine height, cloud cover and solar irradiance, which can help operators anticipate electricity generation conditions.

The model is now being used to power weather information across Google Search, Gemini, Google Maps, Google Maps Platform’s Weather API and Google Earth Engine.

According to the developers, precipitation forecasts for users planning at least a day ahead can be up to 50 percent more accurate, with the largest improvements expected in areas where forecasting has historically been less reliable.