attributes added to mesh2d variable point to non-existent variables

[veenstrajelmer]

xu.open_dataset() adds several attrs to mesh2d that point to non-existing variables, like for instance 'boundary_node_connectivity': 'mesh2d_boundary_nodes'. This is an issue when using the file obtained from .ugrid.to_netcdf() in other processes.

MWE:

import xugrid as xu
import xarray as xr

file_nc = 'Grevelingen-FM_0000_map.nc'
file_nc = xu.data.adh_san_diego(xarray=True).encoding['source']

ds = xr.open_dataset(file_nc)
ds_mesh2d = ds.mesh2d.attrs

uds = xu.open_dataset(file_nc)
uds_mesh2d = uds.grid.to_dataset().mesh2d.attrs

uds_mesh2d_new = uds.grid.to_dataset().mesh2d.attrs
for key in ds_mesh2d.keys():
    uds_mesh2d_new.pop(key)
print(uds_mesh2d_new)

for varn in uds_mesh2d_new.values():
    try:
        uds[varn]
    except KeyError as e:
        print(f'KeyError: {e}')

Gives:

KeyError: 'mesh2d_nEdges'
KeyError: 'face'
KeyError: 'mesh2d_nMax_face_nodes'
KeyError: 'mesh2d_nBoundary_edges'
KeyError: 'mesh2d_boundary_nodes'
KeyError: 'mesh2d_face_faces'
KeyError: 'mesh2d'

[Huite]

I've come across this issue myself as well: after writing the netCDF, QGIS via MDAL will not accept these attributes.

I think you can make two cases for how these attributes should be interpreted;

• "This variable exists, under this name"
• "If it exists, or if you create this variable, it will be under this name"

I went with the second interpretation -- though probably mostly due to implementation convenience since you can always assume the variables are all available. It allows for a relatively easy configuring of a naming scheme.

Regardless, if the unused variables are removed during the to_dataset() call, that behavior is maintained and other software shouldn't trip up.