Fix drop_sel for a MultiIndex

xarray/core/dataset.py

@@ -6065,6 +6065,8 @@ def drop_sel(
         Data variables:
             A        (x, y) int64 32B 0 2 3 5
         """
+        from xarray.core.dataarray import DataArray
+
         if errors not in ["raise", "ignore"]:
             raise ValueError('errors must be either "raise" or "ignore"')
 
@@ -6076,7 +6078,11 @@ def drop_sel(
             # is a large numpy array
             if utils.is_scalar(labels_for_dim):
                 labels_for_dim = [labels_for_dim]
-            labels_for_dim = np.asarray(labels_for_dim)
+            # Most conversion to arrays is better handled in the indexer, however
+            # DataArrays are a special case where the underlying libraries don't provide
+            # a good conversition.
+            if isinstance(labels_for_dim, DataArray):
+                labels_for_dim = np.asarray(labels_for_dim)
             try:
                 index = self.get_index(dim)
             except KeyError as err:

xarray/tests/test_dataset.py

@@ -2980,6 +2980,21 @@ def test_drop_multiindex_level(self) -> None:
             actual = data.drop_vars("level_1")
         assert_identical(expected, actual)
 
+    def test_drop_multiindex_labels(self) -> None:
+        data = create_test_multiindex()
+        mindex = pd.MultiIndex.from_tuples(
+            [
+                ("a", 2),
+                ("b", 1),
+                ("b", 2),
+            ],
+            names=("level_1", "level_2"),
+        )
+        expected = Dataset({}, Coordinates.from_pandas_multiindex(mindex, "x"))
+
+        actual = data.drop_sel(x=("a", 1))
+        assert_identical(expected, actual)
+
     def test_drop_index_labels(self) -> None:
         data = Dataset({"A": (["x", "y"], np.random.randn(2, 3)), "x": ["a", "b"]})