Google DeepMind has announced an update to its AI-based weather model, releasing WeatherNext 3. According to DeepMind, the new version produces more detailed forecasts for temperature, humidity and wind speed, driven largely by additional data used to train the system.
Resolution and update frequency
WeatherNext 3 generates hourly forecasts and, DeepMind says, improves spatial resolution from a 25-kilometer grid to a 5-kilometer grid. The higher-resolution, hourly outputs are intended to better capture local variability, such as humidity changes that can occur within a single kilometer.
Satellite and other data sources
Rather than relying solely on numerical weather prediction models, WeatherNext 3 ingests real-time satellite data that DeepMind describes as providing a “rich, continuously updated view of the atmosphere.” The model also integrates observations from sparsely located meteorological stations, improving forecasts for variables that vary over short distances.
The system combines precipitation data from NASA with Google’s own satellite analysis, which DeepMind says enhances its ability to predict rain and snowfall.
Relevance for renewable energy
DeepMind highlights that the new model should be particularly valuable for companies involved in renewable energy. WeatherNext 3 can forecast wind speeds at 100 meters altitude, a parameter relevant to wind turbine performance. It also supplies high-resolution cloud cover and solar radiation data that can help solar farm operators estimate expected generation.
Access and integration
Google will integrate the model into Search, Gemini and Maps. Developers can also incorporate WeatherNext 3 into other services via an application programming interface (API).
Notes
The update was reported by Engadget; DeepMind attributes the model’s finer spatial and temporal resolution to the new datasets and the use of real-time satellite observations.
