(EXPERIMENTAL) NOAA FourCastNet Global Forecast System (FourCastNetGFS) (EXPERIMENTAL)

The FourCastNet Global Forecast System (FourCastNetGFS) is an experimental system set up by the National Centers for Environmental Prediction (NCEP) to produce medium range global forecasts. The model runs on a 0.25 degree latitude-longitude grid (about 28 km) and 13 pressure levels. The model produces forecasts 4 times a day at 00Z, 06Z, 12Z and 18Z cycles. Major atmospheric and surface fields including temperature, wind components, geopotential height, relative humidity and 2 meter temperature and 10 meter winds are available. The products are 6 hourly forecasts up to 10 days. The data format is GRIB2.

The FourCastNetGFS system is an experimental weather forecast model built upon the pre-trained Nvidia’s FourCastNet Machine Learning Weather Prediction (MLWP) model version 2. The FourCastNet (Bonev et al, 2023) was developed by Nvidia using Adaptive Fourier Neural Operators. It uses a Fourier transform-based token-mixing scheme with the vision transformer architecture. This model is pre-trained with ECMWF’s ERA5 reanalysis data. The FourCastNetGFS takes one model state as initial condition from NCEP 0.25 degree GDAS analysis data and runs FourCastNet with weights from the pretrained FourCastNet by Nvidia. Unit conversion to the GDAS data is conducted to match the input data required by FourCastNet and to generate forecast products consistent to GFS.

The input data generated from the GDAS data as FourCastNet input is provided under the forecast data directory. Example of file names is:

input_2024022000.npy

There are 40 files under each directory covering a 10 day forecast. An example of file name is listed below

fcngfs.t00z.pgrb2.0p25.f006

Please note that this NOAA GFS machine learning Model was produced using a code package released by ECMWF’s ai_models_fourcastnetv2 plugin. For information on ai_models_fourcastnetv2 plugin, please visit their github page listed in the documentation and license sections of this page.