DeepMind's 'GenCast' AI Outperforms Traditional Weather Models

The new AI system offers faster, more precise mid-term weather forecasts but is seen as a complement rather than a replacement for physics-based methods.

DeepMind's 'GenCast' AI model can generate global 15-day weather forecasts in just eight minutes with improved accuracy over traditional methods.
The AI model demonstrated superior performance in predicting extreme weather events, tropical cyclone paths, and wind strength in over 97% of test cases.
Developed using 40 years of historical weather data, 'GenCast' was validated by independent reviewers and detailed in the journal Nature.
Experts at the German Weather Service (DWD) emphasize that AI models like 'GenCast' complement, rather than replace, physics-based forecasting systems due to their limitations in adhering to natural laws.
The DWD is testing its own AI-based tools to integrate with existing methods, aiming to enhance forecast precision across various weather variables.