Fix zonal_mean missing-coverage NaNs and dtype promotion

test/grid/integrate/test_zonal.py

@@ -2,6 +2,7 @@
 import dask.array as da
 import numpy as np
 import pytest
+import warnings
 
 import numpy.testing as nt
 
@@ -84,6 +85,44 @@ def test_lat_inputs(self, gridpath, datasetpath):
         assert len(uxds['psi'].zonal_mean(lat=1)) == 1
         assert len(uxds['psi'].zonal_mean(lat=(-90, 90, 1))) == 181
 
+    def test_zonal_mean_missing_latitudes_nan(self, gridpath, datasetpath):
+        """Zonal mean should return NaN (not zeros) when no faces intersect a latitude."""
+        grid_path = gridpath("ugrid", "outCSne30", "outCSne30.ug")
+        data_path = datasetpath("ugrid", "outCSne30", "outCSne30_vortex.nc")
+        uxds = ux.open_dataset(grid_path, data_path)
+
+        # Restrict to a narrow band so most requested latitudes have no coverage
+        narrow = uxds["psi"].subset.bounding_box(lon_bounds=(-20, 20), lat_bounds=(0, 10))
+
+        with warnings.catch_warnings():
+            warnings.filterwarnings("error", category=RuntimeWarning)
+            res = narrow.zonal_mean(lat=(-90, 90, 10))
+
+        below_band = res.sel(latitudes=res.latitudes < 0)
+        assert np.all(np.isnan(below_band))
+        assert np.isfinite(res.sel(latitudes=0).item())
+
+        with warnings.catch_warnings():
+            warnings.filterwarnings("error", category=RuntimeWarning)
+            res_cons = narrow.zonal_mean(lat=(-90, 90, 10), conservative=True)
+
+        below_band_cons = res_cons.sel(latitudes=res_cons.latitudes < 0)
+        assert np.all(np.isnan(below_band_cons))
+        assert np.isfinite(res_cons.sel(latitudes=5).item())
+
+    def test_zonal_mean_int_data_promotes_dtype(self):
+        """Integer inputs should be promoted so NaNs can be stored."""
+        grid = ux.Grid.from_healpix(zoom=0)
+        faces = np.where(grid.face_lat > 0)[0]  # only northern hemisphere
+        uxda = ux.UxDataArray(
+            np.ones(grid.n_face, dtype=np.int32), dims=["n_face"], uxgrid=grid
+        ).isel(n_face=faces)
+
+        res = uxda.zonal_mean(lat=(-90, 90, 30))
+
+        assert np.issubdtype(res.dtype, np.floating)
+        assert np.isnan(res.sel(latitudes=-90)).item()
+
 def test_mismatched_dims():
     uxgrid = ux.Grid.from_healpix(zoom=0)
     uxda = ux.UxDataArray(np.ones((10, uxgrid.n_face, 5)), dims=['a', 'n_face', 'b'], uxgrid=uxgrid)

uxarray/core/zonal.py

@@ -26,12 +26,19 @@ def _compute_non_conservative_zonal_mean(uxda, latitudes, use_robust_weights=Fal
     shape = list(uxda.shape)
     shape[face_axis] = len(latitudes)
 
+    if np.issubdtype(uxda.dtype, np.integer) or np.issubdtype(uxda.dtype, np.bool_):
+        # Promote integers/bools so we can represent NaNs
+        result_dtype = np.float64
+    else:
+        # Preserve existing float/complex dtype
+        result_dtype = uxda.dtype
+
     if isinstance(uxda.data, da.Array):
-        # Create a Dask array for storing results
-        result = da.zeros(shape, dtype=uxda.dtype)
+        # Pre-fill with NaNs so empty slices stay missing without extra work
+        result = da.full(shape, np.nan, dtype=result_dtype)
     else:
         # Create a NumPy array for storing results
-        result = np.zeros(shape, dtype=uxda.dtype)
+        result = np.full(shape, np.nan, dtype=result_dtype)
 
     faces_edge_nodes_xyz = _get_cartesian_face_edge_nodes_array(
         uxgrid.face_node_connectivity.values,
@@ -46,6 +53,14 @@ def _compute_non_conservative_zonal_mean(uxda, latitudes, use_robust_weights=Fal
 
     for i, lat in enumerate(latitudes):
         face_indices = uxda.uxgrid.get_faces_at_constant_latitude(lat)
+
+        idx = [slice(None)] * result.ndim
+        idx[face_axis] = i
+
+        if face_indices.size == 0:
+            # No intersecting faces for this latitude
+            continue
+
         z = np.sin(np.deg2rad(lat))
 
         fe = faces_edge_nodes_xyz[face_indices]
@@ -59,14 +74,16 @@ def _compute_non_conservative_zonal_mean(uxda, latitudes, use_robust_weights=Fal
 
         total = w.sum()
 
+        if total == 0.0 or not np.isfinite(total):
+            # If weights collapse to zero, keep the pre-filled NaNs
+            continue
+
         data_slice = uxda.isel(n_face=face_indices, ignore_grid=True).data
         w_shape = [1] * data_slice.ndim
         w_shape[face_axis] = w.size
         w_reshaped = w.reshape(w_shape)
         weighted = (data_slice * w_reshaped).sum(axis=face_axis) / total
 
-        idx = [slice(None)] * result.ndim
-        idx[face_axis] = i
         result[tuple(idx)] = weighted
 
     return result