Google's GenCast represents a significant advancement in weather forecasting, outperforming ECMWF's ENS model with superior accuracy in predictions extending up to 25 days ahead.
Trained on four decades of historical data, GenCast has outperformed ENS in 97.2% of tests and reached 99.8% accuracy for forecasts 36 hours ahead or more.
As an ensemble model, GenCast provides 50+ predictions simultaneously, dramatically reducing forecast generation time to just 8 minutes, compared to hours for conventional models.
The model is being released as open source, allowing broad access and enabling improvements from broader development communities, potentially revolutionizing the field of meteorology.
Collection
[
|
...
]