Introducing WeatherNext 3, our most advanced and accurate global weather AI model

Google DeepMind and Google Research have introduced WeatherNext 3, an advanced global weather artificial intelligence model designed to deliver higher-resolution and more accurate forecasts. Unlike older models that rely on traditional physics simulations with a six-hour data lag, WeatherNext 3 trains directly on real-time geostationary satellite mosaics and sparse weather station data. This allows the system to generate new forecasts every hour, providing a global weather picture that is roughly five times sharper than the previous version.

The model produces surface variables like temperature and moisture at a 5-kilometer resolution, other surface variables at 10 kilometers, and atmospheric variables like wind speed at 25 kilometers. It also features breakthrough accuracy in precipitation forecasting, achieving significant improvements in medium-range predictions of rain and snow. Additionally, WeatherNext 3 includes specific variables engineered for renewable energy planning, such as 100-meter wind speeds for wind-energy output and high-resolution cloud cover and sun radiation for solar farms.

These advancements bring localized and reliable forecasting to billions of people, particularly benefiting regions in Latin America, Africa, and the Asia-Pacific that have historically lacked high-resolution coverage. The model is integrated globally into Google products and platforms, powering weather experiences across Google Search, the Gemini app, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine, while also being made available for developers and researchers via Google Cloud.