Fix pandas' interpolate(fill_value=) error

[max-sixty]

Pandas no longer has a fill_value parameter for interpolate.

Weirdly I wasn't getting this locally, on pandas 2.1.0, only in CI on https://github.com/pydata/xarray/actions/runs/6054400455/job/16431747966?pr=8138.

Removing it passes locally, let's see whether this works in CI

Would close #8125

[max-sixty]

This indeed fixes the errors. I've been away for a while so don't want to self-merge quite yet, but will if no one comments otherwise in the next day or so...

[kmuehlbauer]

@max-sixty This is a regression bug in pandas, see pandas-dev/pandas#54920.

fill_value is still a kwarg for the scipy based interpolators. Only for the fill_methods this is an issue.

[max-sixty]

OK — though does it not being required for the tests to pass suggest we can merge it regardless?

[kmuehlbauer]

The tests pass because np.nan is the default fill_value in the scipy interpolators and maybe because fill_value is actually not needed for these specific tests. That would indicate, that the tests would need adaption to take fill_value correctly into account as needed for the scipy based interpolation methods.

Please have a look here at the scipy docs https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.LinearNDInterpolator.html#scipy.interpolate.LinearNDInterpolator

_Value used to fill in for requested points outside of the convex hull of the input points. If not provided, then the default is nan.

[max-sixty]

Right — agree the argument isn't doing anything in these tests. Is there a good reason to keep it vs. merging to fix the tests?

I also agree that we could have different tests which test for fill_value if that's important to test. (though would be an additional PR)

[kmuehlbauer]

@max-sixty Although upstream has milestoned this issue for the next release it might still take some time. Maybe it's easiest to move forward here and open an issue to create proper tests which take fill_value into account (optional PR)?