Remove columns that are all NaNs in make_criteria (#1713)

<!--Please ensure the PR fulfills the following requirements! -->
<!-- If this is your first PR, make sure to add your details to the
AUTHORS.rst! -->
### Pull Request Checklist:
- [x] This PR addresses an already opened issue (for bug fixes /
features)
    - This PR fixes #xyz
- [x] Tests for the changes have been added (for bug fixes / features)
- [x] (If applicable) Documentation has been added / updated (for bug
fixes / features)
- [x] CHANGES.rst has been updated (with summary of main changes)
- [x] Link to issue (:issue:`number`) and pull request (:pull:`number`)
has been added

### What kind of change does this PR introduce?

* Methods such as kmeans clustering will crash if there are NaNs in the
data. This is something that should happen if only a subset of
realizations are NaNs, but we should be safe to remove the columns that
are NaN across all realizations.

### Does this PR introduce a breaking change?

* No


### Other information:

CHANGES.rst

@@ -4,7 +4,7 @@ Changelog
 
 v0.49.0 (unreleased)
 --------------------
-Contributors to this version: Trevor James Smith (:user:`Zeitsperre`), Pascal Bourgault (:user:`aulemahal`), Juliette Lavoie (:user:`juliettelavoie`), David Huard (:user:`huard`).
+Contributors to this version: Trevor James Smith (:user:`Zeitsperre`), Pascal Bourgault (:user:`aulemahal`), Juliette Lavoie (:user:`juliettelavoie`), David Huard (:user:`huard`), Gabriel Rondeau-Genesse (:user:`RondeauG`).
 
 Announcements
 ^^^^^^^^^^^^^
@@ -27,6 +27,7 @@ Bug fixes
 * Fixed bug with loess smoothing for an array full of NaNs. (:pull:`1699`).
 * Fixed and adapted ``time_bnds`` to the newest xarray. (:pull:`1700`).
 * Fixed "agreement fraction" in ``robustness_fractions`` to distinguish between negative change and no change. Added "negative" and "changed negative" fractions (:issue:`1690`, :pull:`1711`).
+* ``make_criteria`` now skips columns with NaNs across all realizations. (:pull:`1713`).
 
 Internal changes
 ^^^^^^^^^^^^^^^^

tests/test_ensembles.py

@@ -566,6 +566,15 @@ def test_make_criteria(self, tas_series):
         uncrit = crit.unstack("criteria")
         assert set(uncrit.dims) == {"realization", "lat", "time"}
 
+        crit = ensembles.make_criteria(ds.where(ds.var_a > 0))
+        assert crit.dims == ("realization", "criteria")
+        assert crit.criteria.size == 12
+        np.testing.assert_array_equal(crit.isnull().sum(), 0)
+        np.testing.assert_array_equal(crit.min(), 1)
+        uncrit = crit.unstack("criteria").to_dataset("variables")
+        assert set(uncrit.dims) == {"realization", "lat", "time"}
+        assert uncrit.time.size == 3
+
 
 # ## Tests for Robustness ##
 @pytest.fixture

xclim/ensembles/_reduce.py

@@ -57,6 +57,8 @@ def make_criteria(ds: xarray.Dataset | xarray.DataArray):
 
     `ds2` will have all variables with the same dimensions, so if the original dataset had variables with different
     dimensions, the added dimensions are filled with NaNs.
+    Also, note that criteria that are all NaN (such as lat/lon coordinates with no data) are dropped from `crit` to avoid issues with
+    the clustering algorithms, so the original dataset might not be able to be fully reconstructed.
     The `to_dataset` part can be skipped if the original input was a DataArray.
     """
 
@@ -113,6 +115,9 @@ def _make_crit(da):
     else:
         # Easy peasy, skip all the convoluted stuff
         crit = _make_crit(ds)
+
+    # drop criteria that are all NaN
+    crit = crit.dropna(dim="criteria", how="all")
     return crit.rename("criteria")