Google DeepMind has published research demonstrating that its latest weather prediction model, GraphCast-2, now matches or exceeds the accuracy of traditional numerical weather prediction models at forecast horizons of up to 15 days. The model processes atmospheric data in under two minutes on a single TPU chip, compared to the hours of supercomputer time required by conventional physics-based models.
The breakthrough has significant implications for disaster preparedness, agriculture, and energy grid management, where accurate extended-range forecasts can drive substantial economic value. The US National Weather Service has begun pilot testing the model alongside its existing forecasting systems following a cooperative agreement signed in February.
