GFS: Missing levels for sigmaLayer and depthBelowLandLayer

cfgrib seems to miss some levels for the level types sigmaLayer and depthBelowLandLayer.

I’m working with this grib file: https://noaa-gfs-bdp-pds.s3.amazonaws.com/gfs.20210115/00/gfs.t00z.pgrb2.0p25.f003

wgrib2 -lev ~/Downloads/gfs.t00z.pgrb2.0p25.f003 shows four levels for each of these level types:

425:224793214:0-0.1 m below ground
427:225650288:0.1-0.4 m below ground
429:226508191:0.4-1 m below ground
431:227363072:1-2 m below ground

558:309354159:0.33-1 sigma layer
559:310143600:0.44-1 sigma layer
560:310922234:0.72-0.94 sigma layer
561:311736866:0.44-0.72 sigma layer

Reading the same file with cfgrib results in only two:

import cfgrib

ds = cfgrib.open_datasets("gfs.t00z.pgrb2.0p25.f003")

d = next(d for d in ds if 'depthBelowLandLayer' in d.variables)
d.depthBelowLandLayer
# <xarray.IndexVariable 'depthBelowLandLayer' (depthBelowLandLayer: 2)>
# array([0, 1])
# Attributes:
#     long_name:      soil depth
#     units:          m
#     positive:       down
#     standard_name:  depth

d = next(d for d in ds if 'sigmaLayer' in d.variables)
d.sigmaLayer
# <xarray.IndexVariable 'sigmaLayer' (sigmaLayer: 2)>
# array([0, 1])
# Attributes:
#     long_name:  original GRIB coordinate for key: level(sigmaLayer)
#     units:      1

d.r.shape
# (2, 721, 1440)

[alexamici]

@shahramn then this looks like an actual bug (or missing feature) in cfgrib as it only requests the level key and it accepts the default data type, that is int.

Thanks for the clarification.

[alexamici]

Levels will be read as floats in the next release.