Updated fix: remove_fill_values()

[agstephens]

Added import of cf_xarray and detection of bounds for each
coordinate. And then sets _FillValue to None to avoid writing
bounds with _FillValue to netcdf.

Pull Request Checklist:

• This PR addresses an already opened issue (for bug fixes / features)
  • This PR fixes issue #xyz
• Tests for the changes have been added (for bug fixes / features)
• Documentation has been added / updated (for bug fixes / features)
• HISTORY.rst has been updated (with summary of main changes)
• I have added my relevant user information to AUTHORS.md

• What kind of change does this PR introduce?:

• Does this PR introduce a breaking change?:

• Other information:

[agstephens]

Hi @ellesmith88, please check this PR that hopefully makes sure the coordinate bounds variables do not have _FillValue = NaN in output files (once fixed). Thanks

[ellesmith88]

@agstephens I recently added this fix into clisops https://github.com/roocs/clisops/pull/204/files which fixes the fill values for the coordinate variables bounds in the output files. I think we could probably remove the fill value fix in daops completely, apart from the fact that it is currently used in the decadal fixes, as I implemented it as a fix before we decided to put it all in clisops.

If you want to add the bounds fix here as well then I think it needs to be something like:

try:
     bounds = ds.cf.get_bounds(coord_id)
     bounds.encoding["_FillValue"] = None
except KeyError:
     continue

as a KeyError is raised, rather than returning None, if the coordinate doesn't have any bounds. I tested with the file /badc/cmip6/data/CMIP6/CMIP/IPSL/IPSL-CM6A-LR/historical/r1i1p1f1/Amon/rlds/gr/v20180803/rlds_Amon_IPSL-CM6A-LR_historical_r1i1p1f1_gr_185001-201412.nc

Also, cf-xarray would need to be added to the requirements.

I can make these changes if you want to keep this in daops.

[agstephens]

@agstephens I recently added this fix into clisops https://github.com/roocs/clisops/pull/204/files which fixes the fill values for the coordinate variables bounds in the output files. I think we could probably remove the fill value fix in daops completely, apart from the fact that it is currently used in the decadal fixes, as I implemented it as a fix before we decided to put it all in clisops.

If you want to add the bounds fix here as well then I think it needs to be something like:

try:
     bounds = ds.cf.get_bounds(coord_id)
     bounds.encoding["_FillValue"] = None
except KeyError:
     continue

as a KeyError is raised, rather than returning None, if the coordinate doesn't have any bounds. I tested with the file /badc/cmip6/data/CMIP6/CMIP/IPSL/IPSL-CM6A-LR/historical/r1i1p1f1/Amon/rlds/gr/v20180803/rlds_Amon_IPSL-CM6A-LR_historical_r1i1p1f1_gr_185001-201412.nc

Also, cf-xarray would need to be added to the requirements.

I can make these changes if you want to keep this in daops.

@ellesmith88 Your proposal sounds most sensible - that we just remove this as a dachar/daops fix because it is applied at the clisops level. Let's just fix it at source (in clisops) and keep everything else simpler.