This paper demonstrates the feasibility of democratizing AI-driven global weather forecasting models among university research groups.
The paper highlights the use of NVIDIA's FourCastNetv2, an advanced neural network for weather prediction trained on a subset of the ECMWF ERA5 dataset.
The training of FourCastNetv2 involved 64 A100 GPUs and took 16 hours to complete, offering significant time and cost reductions compared to traditional NWP.
The paper provides insights on data management, training efficiency, and model validation, serving as a guide for other university research groups to develop AI weather forecasting programs.