The new forecasting model delivers more frequent updates, finer spatial resolution, and improved precipitation accuracy across Google’s consumer and developer platforms.
Google DeepMind and Google Research have introduced WeatherNext 3, an AI-powered global weather forecasting model designed to deliver more localized and frequently updated forecasts.
The system produces fresh forecasts every hour and reaches spatial resolution as fine as 5 kilometers, giving it significantly more detail than its predecessor. WeatherNext 3 is already being integrated into Google Search, Gemini, Google Maps, Google Maps Platform Weather API, and Google Earth Engine.
Forecasts refreshed every hour
WeatherNext 3 directly ingests geostationary satellite mosaics and ground-station observations, allowing it to generate updated forecasts throughout the day.
The hourly cadence is especially useful when weather conditions change quickly, such as during developing storms, fronts, or other extreme events. More frequent updates can provide a clearer view of how conditions are evolving rather than relying on forecasts produced at longer intervals.
More detail at the local level
The model can resolve surface temperature and moisture down to 5km or 0.05°, while surface winds are modeled at up to 10km or 0.1° resolution.
That finer spatial detail is intended to improve localized forecasting, particularly in areas where conventional models may struggle to represent smaller-scale conditions accurately. Compared with the previous generation, WeatherNext 3 operates at roughly five times finer resolution.
Improved precipitation forecasting
One of the biggest improvements is precipitation prediction.
WeatherNext 3 reduces error metrics by as much as 50% on key precipitation benchmarks compared with traditional numerical weather prediction baselines. For forecasts a day or more ahead, Google reports precipitation accuracy improvements of up to 50%, with the largest gains in areas where weather forecasting has historically been less reliable.
Those improvements could make longer-range weather information more useful for everyday planning, agriculture, transportation, emergency response, and other activities that depend heavily on rainfall forecasts.
Weather data for clean-energy operations
WeatherNext 3 also produces specialized parameters for the energy sector, including 100-meter turbine-height wind speeds, multi-layer cloud cover, and solar irradiance.
These data points can help grid operators and renewable-energy producers better anticipate wind and solar generation conditions, where small weather changes can affect power output and grid planning.
Already rolling out across Google services
WeatherNext 3 is already powering weather experiences across several Google products, including Search, Gemini, Google Maps, Google Maps Platform Weather API, and Google Earth Engine.
That broader integration means improvements in the underlying forecasting model can reach both everyday users and organizations building weather-dependent services on Google’s platforms.