Google DeepMind Introduces GenCast AI Model: A Revolution in Weather Forecasting

Google DeepMind has developed GenCast, an AI system that predicts weather faster and more accurately than traditional methods.

GenCast outperformed the leading weather prediction system, ENS, in 97.2% of cases, particularly excelling at forecasting tropical cyclone trajectories 12 hours earlier than current methods.

The model was trained on weather data from 1979 to 2018, using machine learning to identify complex weather patterns. This approach enables GenCast to generate a 15-day forecast in just 8 minutes on a Google Cloud TPU v5, compared to the hours of computation required by traditional models running on supercomputers.

However, the model was tested against the 2019 version of ENS, which has since been improved, and its 12-hour intervals may not suffice for tasks requiring more frequent updates, such as wind energy assessments.

Google DeepMind has made GenCast’s code open source, allowing experts to test and enhance the technology, paving the way for new advancements in meteorology.