Google DeepMind Unveils WeatherNext 3 AI-Powered Global Weather Forecasting Model

Google DeepMind has unveiled its latest AI-powered global weather forecasting model, WeatherNext 3. The new model uses live geostationary satellite data to generate hourly forecasts with key temperature and humidity predictions available at a resolution of around 5-kilometre. This is a significant improvement over the previous model, WeatherNext 2, which had a 25-kilometre grid and six-hour updates.

The new model can deliver up to 50% more accurate precipitation forecasts when planning a day or more ahead, according to Google. Its evaluations showed improvements of up to 60% against certain precipitation benchmarks. The model also trains on weather station observations to better reflect conditions on the ground.

WeatherNext 3 is being integrated into various Google services, including Search, the Gemini app, Google Maps, and Google Cloud. This integration will enable users to access more accurate weather forecasts in real-time. Google says that this model can also support applications such as agriculture and renewable energy by forecasting variables like wind speeds, cloud cover, and solar radiation.