What attributes to keep?

[rcaneill]

gsw keeps attributes from the function parameters. What attributes should be kept, and what should be removed? As a short list, I think that:

• long_name should be removed by default, and maybe replaced by something for the non cf-compliant quantities?
• units should be removed by default, and replaced by the proper value if it exists
• standard_name should be removed by default, and replaced by the proper value if it exists

• This means a little refactoring of the wrapping, but probably not too hard work.

I am working on PR #20 to implement all functions, I'll keep the way I do for the moment and I'll replace what is needed depending on what we decide here.

[DocOtak]

My thoughts so far:

• These functions always return "new" DataArrays, so we should just entirely control/replace the value of the attributes and not worry about retaining the ones from the input (by default the attrs of the first argument are carried forward). There could be things like valid_range, comments, other user specific things that just won't apply to the result of the gsw functions.
• If we will wrap all the functions, then I think units will always be known even if a suitable standard name doesn't exist. For explicitly unitless values (e.g. practical salinity) we should use the unit of "1"
• standard_name should not be present unless we put a value there.
• Not CF, but in OceanSITES (and IIRC argo), they will add a reference_scale attribute to temperature and practical salinity with the values of "ITS-90" and "PSS-78" since the units must be respectively degC and unitless.

[rcaneill]

About the reference_scale, I don't think that we necessary need to constraint to CF, if it is a common attribute we can add it.

[DocOtak]

In some play around tests, I found that the name of the returned xr.DataArray retains the "name" attribute (the class instance attribute, not the attrs dict). We should probably replace that with something relevant. Perhaps the GSW name (e.g. the return from SA_from_SP would be a xr.DataArray with the name "SA")

The following is the current behavior using our testing example dataset:

>>> da = gsw.sigma0(SA=ds.SA, CT=ds.CT)
>>> da
<xarray.DataArray 'SA' (id: 3)>
array([27.17191038, 26.12820162, 24.03930887])
Coordinates:
  * id       (id) int64 0 1 2
Attributes:
    standard_name:  sea_water_sigma_t
    units:          kg m-3
>>> da.name
'SA'

[rcaneill]

This is indeed a problem... I'll add it along with the ideas of PR #22 it this suits implementation suits you