FourCastNet — Global Weather Model with Adaptive Fourier Neural OperatorsNew

FourCastNet (Pathak et al., 2022) is a global, data-driven, high-resolution weather model built on Adaptive Fourier Neural Operators (AFNO). At 0.25-degree resolution it produces a week-long global forecast in under two seconds, matching the ECMWF IFS on large-scale variables and outperforming it on fine-scale fields such as precipitation.

Its speed lets practitioners generate rapid, inexpensive ensembles with thousands of members, improving probabilistic forecasting. It is an early, influential demonstration that neural operators can rival operational NWP at global scale.