Google DeepMind and Google Research have introduced WeatherNext 3, an advanced global weather AI model that generates hourly forecasts at high spatial resolutions by learning directly from real-time satellite observations. This update provides forecasts roughly five times sharper than its predecessor, WeatherNext 2, enabling more localized predictions for critical weather events.
- WeatherNext 3 produces forecasts at 5-kilometer resolution for surface variables like temperature and moisture, 10 kilometers for other surface variables, and 25 kilometers for atmospheric variables like wind speed.
- The model ingests live geostationary satellite mosaics alongside historical analysis to bypass the six-hour data lag associated with traditional numerical weather prediction models.
- It achieves significant improvements in precipitation forecasting accuracy, showing a Continuous Ranked Probability Score (CRPS) improvement of up to 60% against IMERG baselines.
- The system includes specific predictions for renewable energy, such as 100-meter wind speeds and cloud cover, to assist grid operators and developers.
- WeatherNext 3 is being integrated into Google Search, the Gemini app, Google Maps, and Google Earth Engine, offering up to 50% more accurate precipitation forecasts for longer-term planning.
The model aims to make reliable, high-fidelity weather intelligence accessible globally, particularly benefiting regions historically underserved by high-resolution forecasting due to computational costs.