DeepMind Unveils Hourly Weather Forecasting Model for Renewable Energy

Google DeepMind has launched WeatherNext 3, an advanced global weather AI model that generates hourly forecasts at multiple resolutions. This new model is designed to help grid operators manage wind and solar assets by providing accurate predictions of power output.

The WeatherNext 3 model uses a functional generative network (FGN) mesh transformer architecture, which is substantially scaled up from its predecessor, WeatherNext 2. The model trains on satellite imagery and historical NWP analysis, rather than relying solely on NWP-processed datasets. This approach reduces data lag to three to four hours, down from seven.

The model's hourly cadence and satellite-first training enable it to provide detailed forecasts of precipitation, temperature, and wind speeds at the ground level. It also includes high-resolution cloud-cover and solar radiation data, giving solar farm operators granular irradiance estimates for scheduling generation against grid demand.