Introducing WeatherNext 3, our most advanced and accurate global weather AI model
Short summary
Google DeepMind released WeatherNext 3, a weather AI model that ingests live satellite data to generate hourly forecasts at 5-kilometer resolution—five times sharper than its predecessor. The model significantly improves precipitation prediction accuracy and brings high-resolution forecasting to underserved regions in Africa, Latin America, and Asia-Pacific.
- Generates hourly forecasts at 5km resolution for surface variables (temperature, moisture) instead of the previous 25km grid updated every 6 hours, with 60% better precipitation accuracy against satellite data
- Trains directly on real-time geostationary satellite observations rather than traditional physics-based numerical weather models, eliminating the 6-hour data lag that caused biases in fast-changing variables
- Includes specialized predictions for renewable energy (100-meter wind speeds for turbines, cloud cover and solar radiation) to help grid operators and clean energy developers match power generation with demand
- Now integrated into Google Search, Gemini, Google Maps, and Earth Engine, delivering up to 50% more accurate precipitation forecasts for 1+ day planning horizons
Full summary
Google DeepMind released WeatherNext 3, a weather forecasting AI that works at five times higher resolution than its predecessor. The model generates hourly forecasts at 5-kilometer resolution for surface variables like temperature and moisture, compared to WeatherNext 2's 25-kilometer grid updated every six hours. The jump in detail matters because weather changes dramatically over short distances—temperature and humidity can swing wildly across just a few kilometers, especially near coasts, valleys, and mountains. Traditional physics-based forecasting models can't capture these variations because they're trained on coarse atmospheric data and carry a six-hour lag from supercomputer simulations.
The breakthrough comes from training directly on live satellite data instead of those delayed physics models. WeatherNext 3 ingests real-time geostationary satellite imagery and sparse weather station observations, letting it produce a fresh forecast every hour grounded in the most recent observations available. This matters most for fast-developing weather—storms, rain systems, and wind shifts that emerge suddenly. The model shows dramatic improvements in precipitation forecasting, with up to 60% better accuracy on some metrics compared to existing satellite-based precipitation data. It also forecasts 100-meter wind speeds for wind turbines and cloud cover for solar farms, giving clean energy operators precise data for matching power generation to demand.
Google is rolling WeatherNext 3 into its existing products starting now—Search, Gemini, Maps, and Earth Engine all get the upgraded forecasts. Users planning a day or more ahead will see up to 50% more accurate precipitation predictions, with the biggest improvements in regions like Latin America, Africa, and Asia-Pacific that have historically lacked high-resolution forecasting because supercomputing costs made regional models prohibitively expensive. The raw forecast data is also available through BigQuery, Google Cloud Storage, and Earth Engine for researchers and developers to build on.