BUG: pd.concat(axis=0) can override crs information (#2067)

CHANGELOG.md

@@ -28,7 +28,8 @@ Deprecations and compatibility notes:
 Bug fixes:
 - Accessing `gdf.geometry` where the active geometry column is missing, and a column named `"geometry"` is present 
   will now raise an `AttributeError`, rather than returning `gdf["geometry"]` (#2575) 
-
+- Combining GeoSeries/GeoDataFrames with ``pandas.concat`` will no longer silently
+  override CRS information if not all inputs have the same CRS (#2056).
 
 Version 0.11.1 (July 24, 2022)
 ------------------------------

geopandas/array.py

@@ -1317,7 +1317,7 @@ class GeometryArray(ExtensionArray):
         ExtensionArray
         """
         data = np.concatenate([ga.data for ga in to_concat])
-        return GeometryArray(data, crs=to_concat[0].crs)
+        return GeometryArray(data, crs=_get_common_crs(to_concat))
 
     def _reduce(self, name, skipna=True, **kwargs):
         # including the base class version here (that raises by default)
@@ -1388,3 +1388,26 @@ class GeometryArray(ExtensionArray):
             else:
                 return False
         return (self == item).any()
+
+
+def _get_common_crs(arr_seq):
+
+    crs_set = {arr.crs for arr in arr_seq}
+    crs_not_none = [crs for crs in crs_set if crs is not None]
+    names = [crs.name for crs in crs_not_none]
+
+    if len(crs_not_none) == 0:
+        return None
+    if len(crs_not_none) == 1:
+        if len(crs_set) != 1:
+            warnings.warn(
+                "CRS not set for some of the concatenation inputs. "
+                f"Setting output's CRS as {names[0]} "
+                "(the single non-null crs provided)."
+            )
+        return crs_not_none[0]
+
+    raise ValueError(
+        f"Cannot determine common CRS for concatenation inputs, got {names}. "
+        "Use `to_crs()` to transform geometries to the same CRS before merging."
+    )

geopandas/tests/test_merge.py

@@ -54,11 +54,6 @@ class TestMerging:
         self._check_metadata(res)
         exp = GeoDataFrame(pd.concat([pd.DataFrame(self.gdf), pd.DataFrame(self.gdf)]))
         assert_geodataframe_equal(exp, res)
-        # check metadata comes from first gdf
-        res4 = pd.concat([self.gdf.set_crs("epsg:4326"), self.gdf], axis=0)
-        # Note: this behaviour potentially does not make sense. If geom cols are
-        # concatenated but have different CRS, then the CRS will be overridden.
-        self._check_metadata(res4, crs="epsg:4326")
 
         # series
         res = pd.concat([self.gdf.geometry, self.gdf.geometry])
@@ -66,6 +61,48 @@ class TestMerging:
         assert isinstance(res, GeoSeries)
         assert isinstance(res.geometry, GeoSeries)
 
+    def test_concat_axis0_crs(self):
+
+        # CRS not set for both GeoDataFrame
+        res = pd.concat([self.gdf, self.gdf])
+        self._check_metadata(res)
+
+        # CRS set for both GeoDataFrame, same CRS
+        res1 = pd.concat([self.gdf.set_crs("epsg:4326"), self.gdf.set_crs("epsg:4326")])
+        self._check_metadata(res1, crs="epsg:4326")
+
+        # CRS not set for one GeoDataFrame, but set for the other GeoDataFrame
+        with pytest.warns(
+            UserWarning, match=r"CRS not set for some of the concatenation inputs.*"
+        ):
+            res2 = pd.concat([self.gdf, self.gdf.set_crs("epsg:4326")])
+            self._check_metadata(res2, crs="epsg:4326")
+
+        # CRS set for both GeoDataFrame, different CRS
+        with pytest.raises(
+            ValueError, match=r"Cannot determine common CRS for concatenation inputs.*"
+        ):
+            pd.concat([self.gdf.set_crs("epsg:4326"), self.gdf.set_crs("epsg:4327")])
+
+        # CRS not set for one GeoDataFrame, but set for the other GeoDataFrames,
+        # same CRS
+        with pytest.warns(
+            UserWarning, match=r"CRS not set for some of the concatenation inputs.*"
+        ):
+            res3 = pd.concat(
+                [self.gdf, self.gdf.set_crs("epsg:4326"), self.gdf.set_crs("epsg:4326")]
+            )
+            self._check_metadata(res3, crs="epsg:4326")
+
+        # CRS not set for one GeoDataFrame, but set for the other GeoDataFrames,
+        # different CRS
+        with pytest.raises(
+            ValueError, match=r"Cannot determine common CRS for concatenation inputs.*"
+        ):
+            pd.concat(
+                [self.gdf, self.gdf.set_crs("epsg:4326"), self.gdf.set_crs("epsg:4327")]
+            )
+
     def test_concat_axis1(self):
 
         res = pd.concat([self.gdf, self.df], axis=1)