Google DeepMind Releases Advanced AI Weather Model WeatherNext 3

Google DeepMind and Google Research have unveiled WeatherNext 3, a new artificial intelligence model for weather forecasting that they claim is the world's most advanced and accurate. The model directly learns from real-time observations to deliver more timely and localized predictions for weather events.

Leveraging raw satellite data, WeatherNext 3 generates hourly forecasts with high spatial resolution. It can predict key surface variables like temperature and humidity at a 5-kilometer resolution, with other surface variables at 10 kilometers and atmospheric variables such as wind speed at 25 kilometers. This level of detail is crucial for forecasting rapidly developing extreme weather.

The model is trained using sparse weather station observations combined with satellite data, enabling it to produce global forecasts on a 5 km grid while accounting for local terrain details. This capability is particularly significant for regions like Latin America, Africa, and Asia, where high-resolution forecasting has historically been cost-prohibitive. The developers expect this to provide billions of people and local businesses in these areas with localized, high-fidelity weather predictions.

WeatherNext 3 also includes specialized forecasts tailored for renewable energy production. It can predict wind speeds at a height of 100 meters, aiding in wind power generation estimates, and provide high-resolution data on cloud cover and solar radiation to help solar farms estimate ground-level light exposure.

These enhanced capabilities will now power weather information across various Google products, including Google Search, the Gemini app, Google Maps, Google Maps Platform Weather API, and Google Earth Engine.