Downloading GFS Forecasts
Important
As of February 2026, NOAA’s legacy NOMADS OpenDAP endpoint used by earlier versions of EarthSHAB has been deprecated. EarthSHAB now downloads GFS data directly using the updated NOAA distribution endpoints via GRIB access and converts them locally into NetCDF format.
Forecast data is retrieved at 0.25° spatial resolution (GFS 0p25 product) and saved locally as NetCDF files (see Unidata netcdf API). This allows:
significantly faster batch simulations
reproducibility
offline trajectory execution
GFS Forecast Structure
GFS forecasts are produced:
4 times per day: 00Z, 06Z, 12Z, 18Z
Temporal resolution: * 0–120 hours: 1-hour or 3-hour increments (depending on product) * 120–384 hours: 3-hour increments
Availability constraints:
Forecasts typically become available ~3–4 hours after cycle time
NOAA retains approximately 9–10 days of past forecasts
Each forecast extends up to 384 hours (16 days) into the future
Output NetCDF file contains:
Dimensions: * time * lev (pressure levels) * lat * lon
Variables:
hgtprs (geopotential height)
tmpprs (temperature)
ugrdprs (u wind)
vgrdprs (v wind)
Saving GFS Forecasts
saveNETCDF.py downloads a selected GFS forecast and converts it into a NetCDF file formatted specifically for EarthSHAB. This is backwards compatible with the previously downloaded GFS forecasts from NOMADS.
The forecast download is controlled via the config file. Key parameters include:
forecast_start_time: the start time of the gfs forecast downloaded from the server.Must be one of: 00, 06, 12, 18 UTC
Must be within NOAA retention window (~9 days)
Must not be too recent (accounts for ~3–4 hour upload lag)
netcdf_gfs["lat_range"]: How many 0.25 degree indicies to download cenetered aroudn thestart_coordnetcdf_gfs["lon_range"]: How many 0.25 degree indicies to download cenetered aroudn thestart_coordnetcdf_gfs["download_days"]: Determines how far into the forecast horizon to retrieve data. Maximum: ~16 daysnetcdf_gfs["start_coord"]: Initial latitude / longitude of the simulation. Also used as the center of the spatial subset of forecast data.