Google's DeepMind team this week announced an AI model for weather prediction called GenCast.
In a paper published in Nature, DeepMind researchers said they found that GenCast outperformed the European Center for Medium-Range Weather Forecasts' ENS, clearly the world's top operational forecasting system. Ta.
The DeepMind team then provided a better explanation of the technology in a blog post. Whereas previous weather models were “deterministic, providing a single best estimate of future weather,” GenCast is “composed of an ensemble of more than 50 predictions, each of which is likely to occur.” It represents weather trajectories and creates “complex probability distributions of future weather scenarios.”
As for the comparison with ENS, the team said they trained GenCast on weather data dating back to 2018, then compared predictions for 2019 and found that GenCast was more accurate 97.2 percent of the time.
Google says GenCast is part of its suite of AI-based weather models that it's starting to incorporate into Google Search and Maps. We also plan to release real-time and historical predictions from GenCast for anyone to use in their own research and models.