exposures-init-geometry (#890)

* wip

* climada.entity.Exposures rewrite __init__

* wip

* climada.entity.Exposures: refactor __init__

* adaptations to changed Exposures structure: gdf.latitude and gdf.longitude have been eliminated

* adaptations to changed Exposures structure: gdf.latitude and gdf.longitude have been eliminated

* adaptations to changed Exposures structure: gdf.latitude and gdf.longitude have been eliminated

* centroids.Centroids.from_exposures() will always work as geometry column is granted

* adaptation to eliminated lat/lon columns

* hazard.Centroids: region_id and on_land are not necessarily present in the data frame

* Exposures() requires geometry or lat/lon

* cangelog

* points_to_raster function getting more complicated,
considering dataframes with geometry column instead fo lat/lon

* pydoc argument description fix

* fix typo (bug really)

* exposures: remove meta from attributes

* introduce exposures properties, region_id etc.
consequently use region_id property where applicable

* rename and use hazard_impf and hazard_centroids

* lint

* Exposures: rename pmeta property to _meta

* fix changelog

* exposures: gracefully handle empty exposures

* exposures: linting

* Update climada/entity/exposures/base.py

Co-authored-by: Samuel Juhel <[email protected]>

* petals compatibility test: try and switch the petals branch

* fix merging mistake

* deprecate set_geomety_points

* deprecate set_geomety_points

* remove properties and methods from the pydoc attributes section

* typo

* typo

* deprecate set_geomety_points

* Exposures.__init__: no need for a temporary geodataframe
as u_coord.set_df_geometry_points does not use a scheduler anymore

* Exposures.base: amend pydocs

* Exposures.base: use get_impf_column in hazard_impf

* pydoc Exposures.__init__

* pydoc Exposures.__init__

* changelog updates

* deprecate set_geomety_points

* jenkins: align with petals

* exposures tutorial: listing attributes and properties

* exposures: don't do anything in check, update Tutorial

* format

* import Impact from climada.engine.impact
in order to avoid circular imports

* avoid circular imports

* pre-commit run --all-files

* setup: DEPS_TEST update

* format all

* exposures tutorial: add information about optional columns

* clean up tutorials

---------

Co-authored-by: Samuel Juhel <[email protected]>

Author: emanuel-schmid

.github/workflows/ci.yml

@@ -74,6 +74,6 @@ jobs:
     needs: build-and-test
     with:
       core_branch: ${{ github.ref }}
-      petals_branch: develop
+      petals_branch: feature/exposures_crs
     permissions:
       checks: write

CHANGELOG.md

@@ -13,12 +13,26 @@ Code freeze date: YYYY-MM-DD
 ### Added
 
 - `climada.util.interpolation` module for inter- and extrapolation util functions used in local exceedance intensity and return period functions [#930](https://github.com/CLIMADA-project/climada_python/pull/930)
+- `climada.exposures.exposures.Exposures.geometry` property
+- `climada.exposures.exposures.Exposures.latitude` property
+- `climada.exposures.exposures.Exposures.longitude` property
+- `climada.exposures.exposures.Exposures.value` property
+- `climada.exposures.exposures.Exposures.region_id` property
+- `climada.exposures.exposures.Exposures.category_id` property
+- `climada.exposures.exposures.Exposures.cover` property
+- `climada.exposures.exposures.Exposures.hazard_impf` method
+- `climada.exposures.exposures.Exposures.hazard_centroids` method
 
 ### Changed
 
 - Improved scaling factors implemented in `climada.hazard.trop_cyclone.apply_climate_scenario_knu` to model the impact of climate changes to tropical cyclones [#734](https://github.com/CLIMADA-project/climada_python/pull/734)
 - In `climada.util.plot.geo_im_from_array`, NaNs are plotted in gray while cells with no centroid are not plotted [#929](https://github.com/CLIMADA-project/climada_python/pull/929)
 - Renamed `climada.util.plot.subplots_from_gdf` to `climada.util.plot.plot_from_gdf` [#929](https://github.com/CLIMADA-project/climada_python/pull/929)
+- Exposures complete overhaul. Notably
+  - the _geometry_ column of the inherent `GeoDataFrame` is set up at initialization
+  - latitude and longitude column are no longer present there (the according arrays can be retrieved as properties of the Exposures object: `exp.latitude` instead of `exp.gdf.latitude.values`).
+  - `Exposures.gdf` has been renamed to `Exposures.data` (it still works though, as it is a property now pointing to the latter)
+  - the `check` method does not add a default "IMPF_" column to the GeoDataFrame anymore
 
 ### Fixed
 
@@ -27,6 +41,10 @@ Code freeze date: YYYY-MM-DD
 
 ### Deprecated
 
+- `climada.entity.exposures.Exposures.meta` attribute
+- `climada.entity.exposures.Exposures.set_lat_lon` method
+- `climada.entity.exposures.Exposures.set_geometry_points` method
+
 ### Removed
 
 ## 5.0.0

climada/engine/forecast.py

@@ -187,7 +187,7 @@ def __init__(
         if exposure_name is None:
             try:
                 self.exposure_name = u_coord.country_to_iso(
-                    exposure.gdf["region_id"].unique()[0], "name"
+                    np.unique(exposure.region_id)[0], "name"
                 )
             except (KeyError, AttributeError):
                 self.exposure_name = "custom"

climada/engine/impact.py

@@ -259,10 +259,7 @@ def from_eih(cls, exposures, hazard, at_event, eai_exp, aai_agg, imp_mat=None):
             date=hazard.date,
             frequency=hazard.frequency,
             frequency_unit=hazard.frequency_unit,
-            coord_exp=np.stack(
-                [exposures.gdf["latitude"].values, exposures.gdf["longitude"].values],
-                axis=1,
-            ),
+            coord_exp=np.stack([exposures.latitude, exposures.longitude], axis=1),
             crs=exposures.crs,
             unit=exposures.value_unit,
             tot_value=exposures.centroids_total_value(hazard),
@@ -733,9 +730,6 @@ def plot_raster_eai_exposure(
         cartopy.mpl.geoaxes.GeoAxesSubplot
         """
         eai_exp = self._build_exp()
-        # we need to set geometry points because the `plot_raster` method accesses the
-        # exposures' `gdf.crs` property, which raises an error when geometry is not set
-        eai_exp.set_geometry_points()
         axis = eai_exp.plot_raster(
             res,
             raster_res,

climada/engine/impact_calc.py

@@ -119,6 +119,7 @@ def impact(
         apply_deductible_to_mat : apply deductible to impact matrix
         apply_cover_to_mat : apply cover to impact matrix
         """
+        # TODO: consider refactoring, making use of Exposures.hazard_impf
         # check for compatibility of exposures and hazard type
         if all(
             name not in self.exposures.gdf.columns

climada/engine/test/test_impact.py

@@ -112,8 +112,7 @@ def test_from_eih_pass(self):
         np.testing.assert_array_almost_equal(imp.eai_exp, fake_eai_exp)
         np.testing.assert_array_almost_equal(imp.at_event, fake_at_event)
         np.testing.assert_array_almost_equal(
-            imp.coord_exp,
-            np.stack([exp.gdf["latitude"].values, exp.gdf["longitude"].values], axis=1),
+            imp.coord_exp, np.stack([exp.latitude, exp.longitude], axis=1)
         )
 
     def test_pyproj_crs(self):
@@ -987,9 +986,9 @@ def test__build_exp(self):
 
         imp = dummy_impact()
         exp = imp._build_exp()
-        np.testing.assert_array_equal(imp.eai_exp, exp.gdf["value"])
-        np.testing.assert_array_equal(imp.coord_exp[:, 0], exp.gdf["latitude"])
-        np.testing.assert_array_equal(imp.coord_exp[:, 1], exp.gdf["longitude"])
+        np.testing.assert_array_equal(imp.eai_exp, exp.value)
+        np.testing.assert_array_equal(imp.coord_exp[:, 0], exp.latitude)
+        np.testing.assert_array_equal(imp.coord_exp[:, 1], exp.longitude)
         self.assertTrue(u_coord.equal_crs(exp.crs, imp.crs))
         self.assertEqual(exp.value_unit, imp.unit)
         self.assertEqual(exp.ref_year, 0)
@@ -1000,9 +999,9 @@ def test__exp_build_event(self):
         imp = dummy_impact()
         event_id = imp.event_id[1]
         exp = imp._build_exp_event(event_id=event_id)
-        np.testing.assert_array_equal(imp.imp_mat[1].todense().A1, exp.gdf["value"])
-        np.testing.assert_array_equal(imp.coord_exp[:, 0], exp.gdf["latitude"])
-        np.testing.assert_array_equal(imp.coord_exp[:, 1], exp.gdf["longitude"])
+        np.testing.assert_array_equal(imp.imp_mat[1].todense().A1, exp.value)
+        np.testing.assert_array_equal(imp.coord_exp[:, 0], exp.latitude)
+        np.testing.assert_array_equal(imp.coord_exp[:, 1], exp.longitude)
         self.assertTrue(u_coord.equal_crs(exp.crs, imp.crs))
         self.assertEqual(exp.value_unit, imp.unit)
         self.assertEqual(exp.ref_year, 0)

climada/engine/test/test_impact_calc.py

@@ -50,8 +50,8 @@ def check_impact(self, imp, haz, exp, aai_agg, eai_exp, at_event, imp_mat_array=
     """Test properties of imapcts"""
     self.assertEqual(len(haz.event_id), len(imp.at_event))
     self.assertIsInstance(imp, Impact)
-    np.testing.assert_allclose(imp.coord_exp[:, 0], exp.gdf["latitude"])
-    np.testing.assert_allclose(imp.coord_exp[:, 1], exp.gdf["longitude"])
+    np.testing.assert_allclose(imp.coord_exp[:, 0], exp.latitude)
+    np.testing.assert_allclose(imp.coord_exp[:, 1], exp.longitude)
     self.assertAlmostEqual(imp.aai_agg, aai_agg, 3)
     np.testing.assert_allclose(imp.eai_exp, eai_exp, rtol=1e-5)
     np.testing.assert_allclose(imp.at_event, at_event, rtol=1e-5)
@@ -490,7 +490,12 @@ def test_minimal_exp_gdf(self):
     def test_stitch_impact_matrix(self):
         """Check how sparse matrices from a generator are stitched together"""
         icalc = ImpactCalc(
-            Exposures({"blank": [1, 2, 3, 4]}), ImpactFuncSet(), Hazard()
+            Exposures(
+                {"blank": [1, 2, 3, 4]},
+                geometry=[],
+            ),
+            ImpactFuncSet(),
+            Hazard(),
         )
         icalc.hazard.event_id = np.array([1, 2, 3])
         icalc._orig_exp_idx = np.array([0, 1, 2, 3])
@@ -524,7 +529,14 @@ def test_apply_deductible_to_mat(self):
 
     def test_stitch_risk_metrics(self):
         """Test computing risk metrics from an impact matrix generator"""
-        icalc = ImpactCalc(Exposures({"blank": [1, 2, 3]}), ImpactFuncSet(), Hazard())
+        icalc = ImpactCalc(
+            Exposures(
+                {"blank": [1, 2, 3]},
+                geometry=[],
+            ),
+            ImpactFuncSet(),
+            Hazard(),
+        )
         icalc.hazard.event_id = np.array([1, 2])
         icalc.hazard.frequency = np.array([2, 0.5])
         icalc._orig_exp_idx = np.array([0, 1, 2])

climada/engine/unsequa/calc_impact.py

@@ -240,7 +240,9 @@ def uncertainty(
 
         if calc_eai_exp:
             exp = self.exp_input_var.evaluate()
-            coord_df = exp.gdf[["latitude", "longitude"]]
+            coord_df = pd.DataFrame(
+                dict(latitude=exp.latitude, longitude=exp.longitude)
+            )
         else:
             coord_df = pd.DataFrame([])