Google DeepMind and Google Research have launched WeatherNext 3, their most advanced global weather AI model to date. The model leverages real-time satellite data instead of traditional physics simulations, enabling hourly updates and significantly sharper forecasts.
According to independent live evaluations by Brightband, WeatherNext 3 is the most accurate global weather model available. It generates forecasts at multiple spatial resolutions, including 5-kilometer resolution for key surface variables like temperature and moisture, 10 kilometers for other surface variables, and 25 kilometers for atmospheric variables such as wind speed. This represents a fivefold improvement in sharpness over its predecessor, WeatherNext 2.
The model's ability to ingest live geostationary satellite mosaics allows it to track rapidly changing weather systems with unprecedented precision. It also trains directly on sparse weather station observations, improving accuracy for local variations near coastlines, valleys, and mountains. This is particularly beneficial for regions in Latin America, Africa, and Asia-Pacific that have historically lacked high-resolution forecasting.
WeatherNext 3 introduces specialized predictions for renewable energy, including 100-meter wind speeds and solar radiation levels, aiding grid operators and developers in planning clean energy output. Precipitation forecasting has also seen major gains, with CRPS improvements of up to 60% against IMERG, 30% for MRMS, and 10% against rain gauge measurements.
The model is being integrated across Google products such as Search, Maps, and Gemini, and is available to developers and businesses via Google Cloud, BigQuery, and Earth Engine.