Assign new variables by standard names

[juseg]

This pull request implements ds.cf.assign(**standard_variables). The new method should:

• Assign variables by standard name as **kwargs or dictionary.
• Decide on the variable short names without user input (see below).
• Either handle or reject duplicate standard names (see below).
• Workaround xarray incompatibilities, currently CF accessor adds grid mapping as coordinate #513.

Also needed:

• Add documentation, where should it go?
• Add tests, including name duplicates.

I would welcome opinions on the two points below.

1. Decide about the variable short name

I propose this algorithm:

• If variable has a short name, absent from the dataset, use it.

ds.cf.assign(air_temperature=DataArray(name='tas')) -> 'tas'

• Otherwise, use the standard name as a short variable name.

ds.cf.assign(air_temperature=DataArray()) -> 'air_temperature'

• If name is already used in dataset, warn user, and add trailing underscores.

ds.cf.assign(
   air_temperature=DataArray(name=tas)).cf.assign(
   total_precipitation=DataArray(name=tas)) -> tas, tas_

• If name is taken by another variable in call, also add trailing underscores.

ds.cf.assign(
   air_temperature=DataArray(name=tas),
   total_precipitation=DataArray(name=tas)) -> tas, tas_

2. When dataset already contains standard name

ds = xr.Dataset()
ds = ds.cf.assign(air_temperature=0)
ds = ds.cf.assign(air_temperature=1)

I find it more difficult to decide what the method should do here.

• Override existing variable (same as Dataset.assign).
• Override but raise a warning.
• Assign new variable with same standard name.
• Assign new variable but raise a warning.
• Customize via keyword e.g. existing: ignore, override, raise, warn

Note: if we allow multiple variables with the same standard name, the resulting Dataset is technically valid, and ds.cf shows several variables associated with one standard name, while ds.cf[standard_name] fails with a KeyError`.

[dcherian]

I think the cf-xarray version of this only really makes sense when the assigned name is a standard name on one of the present variables, so (2) in your listing.

For (1), we should just forward on to Xarray, as usual.

if we allow multiple variables with the same standard name, the resulting Dataset is technically valid, and ds.cf shows several variables associated with one standard name, while ds.cf[standard_name] fails with a KeyError`

Yes, this is intentional. ds.cf[standard_name] will raise an error unless there is only one result, since that is the only way to return a DataArray. to get all, use ds.cf[[standard_name]]. Then you will get a dataset with all dataarrays with that standard name.

[dcherian]

Something like

cf-xarray/cf_xarray/accessor.py

Line 2473 in 52160a5

apply_mapper(_get_all, self._obj, key, error=False, default=[key])

should be used here. So

mapped_name, = apply_mapper(_get_all, self._obj, standard_name, error=False, default=[standard_name])

will do the translation to actual name in the dataset, if possible. Use this to create a new dictionary that you can then pass to self._obj.assign

[juseg]

Thank you for the feedback, I will be looking at this next week.