Dealing with "time" dim differently

[BSchilperoort]

Currently when you regrid data, you only need to specify the target dataset:

ds_regridded = ds_source.regrid.linear(ds_target)

And the dims common to ds_source and ds_target are regridded, except if a dimension is named "time". That one is dropped out. This is of course a bit arbitrary.

Other ways to deal with this would be:

1. Only regrid dimensions specified by user with an arg/kwarg (a bit cumbersome).
2. Force users to modify the ds_target object so it only includes the dimensions that they actually want to regrid (like 1. but without the input argument)
3. Regrid all common dimensions, except for ones specified by users through a kwarg.
4. Like 3., but automatically drop any dimensions with a non-int/float coordinate values (while warning the user).

To me option 4 sounds like a good compromise between the current behavior and being less arbitrary. It would also make it easier to deprecate the automagical ignoring of dimensions as users are already being warned.

---

This issue arose from a comment by @dcherian in #46 (comment)_

[dcherian]

Doesn't xarray-regrid expect 'latitude' and 'longitude' here? You could use cf-xarray https://cf-xarray.readthedocs.io/en/latest/quickstart.html#indexing though it'd be pretty easy to just implement lat/long matching using this attribute table: https://github.com/xarray-contrib/cf-xarray/blob/7072f70292efcc73030653209b44a367510766e0/cf_xarray/criteria.py#L43

[BSchilperoort]

Doesn't xarray-regrid expect 'latitude' and 'longitude' here?

Latitude and longitude are not required. It works on any rectilinear grid.

The latitude coordinate values are used in the conservative routine to correct the weights, if they're present. For the other methods this is not needed.

[slevang]

Yeah I think 3/4 should be flexible enough to handle all cases without imposing an extra burden on typical usage patterns. xarray already uses both exclude and exclude_dims in a number of places, either of those would work.