Google DeepMind’s GenCast AI Sets New Benchmark in Weather Forecasting Accuracy

GenCast outperforms traditional models, offering faster and more precise 15-day weather predictions with significant implications for disaster preparedness and renewable energy.

GenCast, an AI weather model by Google DeepMind, surpasses the European Centre for Medium-Range Weather Forecasts’ ENS system in accuracy 97.2% of the time and 99.8% for forecasts beyond 36 hours.
The model uses a probabilistic ensemble approach, generating 50+ possible scenarios, which improves predictions for extreme weather events such as hurricanes, heatwaves, and high winds.
GenCast operates significantly faster than traditional physics-based models, producing 15-day forecasts in just 8 minutes on a single Google Cloud TPU, compared to hours on supercomputers.
Trained on four decades of historical weather data, GenCast provides high-resolution forecasts at 0.25° and is being made open-source to support global weather research and applications.
While not a replacement for traditional models, GenCast complements them, offering enhanced tools for disaster response, renewable energy planning, and improving public safety.

MIT Technology Review

Nature

Google