GraphCast: Learning skillful medium-range global weather forecasting

Global medium-range weather forecasting is critical to decision-making across many social and economic domains. Traditional numerical weather prediction uses increased compute resources to improve forecast accuracy but does not directly use historical weather data to improve the underlying model. In this talk we will be presenting GraphCast, a machine learning–based method trained directly from reanalysis data. It predicts hundreds of weather variables for the next 10 days at 0.25° resolution globally in under 1 minute. We compare GraphCast to the most accurate operational deterministic system (HRES) and show how its forecasts produce state of the art metrics, and support better severe event prediction, including tropical cyclone tracking, atmospheric rivers, and extreme temperatures. We also show how the approach can be extended to probabilistic forecasting to materialize similar improvements against ENS, a top operational ensemble forecast. These models are key advances in accurate and efficient weather forecasting and help realize the promise of machine learning for modeling complex dynamical systems.