DeepMind, Google's AI research division, has developed a new weather forecasting system called GenCast. This AI system reportedly outperforms the European Centre for Medium-Range Weather Forecasts (ECMWF) ENS model in several areas, particularly in long-term predictions.
GenCast utilizes diffusion model technology, which trains the system by adding noise to existing weather models and then learning to reduce that noise for more accurate forecasts. It generates predictions in 12-hour intervals by merging current weather data with historical forecasts and supports ensemble forecasting to better evaluate uncertainty in predictions.
Despite its high accuracy, GenCast still faces challenges in predicting extreme weather events, such as hurricane intensity, and the DeepMind team is working to improve these aspects.
GenCast is available as an open-source project, facilitating academic collaboration and innovation in weather forecasting.
As climate change leads to more frequent natural disasters and extreme weather, the need for accurate and rapid weather predictions has never been more critical. GenCast aims to harness AI technology to provide higher-resolution forecasts while significantly reducing computational costs.