Google releases WeatherNext 3 AI weather model with real-time satellite data

Google DeepMind and Google Research have released WeatherNext 3, a new AI weather forecasting model that utilizes real-time data from geostationary meteorological satellites. By bypassing the reliance on traditional numerical weather prediction models, which often suffer from a six-hour data delay, the new model can generate global forecasts every hour.

WeatherNext 3 offers a significant increase in resolution compared to its predecessor, WeatherNext 2. While the previous version operated on a 25km grid, the new model provides ground-level variables like temperature and humidity at a 5km resolution. This improvement allows for much clearer representation of temperature variations caused by local terrain.

The model also demonstrates enhanced accuracy in precipitation forecasting. By training on NASA satellite precipitation datasets and Google’s own radar data, the model has improved short-term precipitation prediction accuracy by up to 60% compared to NASA IMERG benchmarks. Additionally, the model includes new variables for renewable energy forecasting, such as wind speeds at 100 meters and solar radiation levels.

WeatherNext 3 is being integrated into Google Search, Gemini, Google Maps, and the Google Maps Platform API. Developers can access the data through BigQuery, Google Earth Engine, and Google Cloud Storage.