Grouper's additional dimensions are non-optional

[aulemahal]

Generic Issue

• xclim version: master
• Python version: all
• Operating System: all

Description

Related to #1736 and discussion #1735. The Grouper add_dims was intended to allow the training of adjustments across dimensions. For example, train a method by mixing all realizations of a simulation. Of course, in such a case, the extra dimensions doesn't exist on the reference data. But the current code expects them to exist in many places.

What I Did

import xarray as xr
from xclim import sdba
from xclim.testing import open_dataset

sim = open_dataset('sdba/CanESM2_1950-2100.nc')
ref = open_dataset('sdba/ahccd_1950-2013.nc')
hist = sim.pr.sel(time=slice('1980', '2010'))
sim = sim.pr.sel(time=slice('2040', '2070'))
ref = ref.pr.isel(location=0).sel(time=slice('1980', '2010'))
ref = ref.drop_vars(['lon','lat'])

group = sdba.Grouper("time.dayofyear", add_dims=['location'])
loci = sdba.LOCI.train(ref, hist, thresh="1 mm/d", group=group)

What I Received

This specific example fails because:

• sdba.utils.map_cdf passes the group's dims without filtering them to apply_ufunc
• group.apply('mean', ref) tries to average the extra dim.

Surely, there are more spots where this happens. But fixing those two fixes the issue for LOCI.

Code of Conduct

• I agree to follow this project's Code of Conduct