xarray¶

ds = f.to_xarray()                        # every field
ds = f.to_xarray(':(TMP|HGT):[0-9]+ mb:') # or the ones you pick

This needs xarray (install the xarray or all extra).

Layout¶

You get one variable per parameter, level kind and kind of forecast. Each is laid out along whichever of these vary within it:

time, the reference time;
member, the ensemble member;
step, the forecast step;
a level dimension named after the level kind (isobaricInhPa, heightAboveGround and so on), with layers described by both their bounds.

Coordinates and attributes follow CF conventions: 1-D latitude and longitude on lat-lon grids; 2-D ones, projection x/y and a CF crs grid mapping on projected grids; units, long names and times.

Fields outside the common layout¶

Fields on a different grid from the rest get their own spatial dimensions. Fields that no coordinate tells apart, such as the same field stored twice, go along a <name>_field dimension instead of overwriting each other.

Winds¶

to_xarray(earth_winds=True) turns grid-relative winds (HRRR, NAM and RAP's Lambert grids, rotated grids) to east and north before building the Dataset. See Regridding and winds.

NetCDF from the command line¶

fastgrib file.grib2 -match ':HGT:[0-9]+ mb:' -o heights.nc