The Google DeepMind team this week unveiled an AI model for weather prediction called GenCast.
In a paper published in Nature, DeepMind researchers say they found that GenCast outperformed the European Center for Medium-Range Weather Forecasts’ ENS (arguably the world’s best operational forecasting system).
And in a blog post, the DeepMind team provided a more accessible explanation of the technology. While previous weather models “were deterministic and provided a single best estimate of future weather,” GenCast “consists of an ensemble of more than 50 forecasts, each representing a possible weather trajectory, and “a complex array of future weather scenarios.” Produces a “probability distribution.”
Compared to ENS, the team said they trained GenCast on weather data going back to 2018 and then compared forecasts for 2019 and found that GenCast was 97.2% more accurate.
Google says GenCast is part of its suite of AI-based weather models and is starting to be integrated into Google Search and Maps. We also plan to make GenCast’s real-time and historical forecasts publicly available for anyone to use in their own research and models.
