UgridDataset is converted to Dataset with assign_attrs

[veenstrajelmer]

Interesting bug that when constructing a uds from a ugrid2d instead of a ugrid2d.to_dataset(). With the latter, .assign_attrs() converts the UgridDataset to a Dataset:

import xugrid as xu
from meshkernel import MakeGridParameters, MeshKernel

# user input
crs = 'EPSG:4326' # coordinate reference system, only EPSG 4326 (WGS84) is currently supported by the ERA5 and CMEMS download scripts.

# domain and resolution
lon_min, lon_max, lat_min, lat_max = 105.8, 106.85, 17.75, 18.5
dxy = 0.1

# create base grid
make_grid_parameters = MakeGridParameters(
    angle=0,
    origin_x=lon_min,
    origin_y=lat_min,
    upper_right_x=lon_max,
    upper_right_y=lat_max,
    block_size_x=dxy,
    block_size_y=dxy,
    )

mk = MeshKernel()#projection=projection)
mk.curvilinear_compute_rectangular_grid_on_extension(make_grid_parameters)
mk.curvilinear_convert_to_mesh2d() #convert to ugrid/mesh2d

mesh2d_grid = mk.mesh2d_get()

crs = 'EPSG:28992'
xu_ugrid2d = xu.Ugrid2d.from_meshkernel(mesh2d_grid, crs=crs)

uds1 = xu.UgridDataset(xu_ugrid2d.to_dataset()) # this results in a UgridDataset after assign_attrs
uds2 = xu.UgridDataset(grids=[xu_ugrid2d]) # this results in a Dataset after assign_attrs

print(type(uds1))
print(type(uds2))

uds1 = uds1.assign_attrs({
    'institution': 'Deltares',
    'references': 'https://www.deltares.nl',
    })
uds2 = uds2.assign_attrs({
    'institution': 'Deltares',
    'references': 'https://www.deltares.nl',
    })

print(type(uds1))
print(type(uds2))

Gives:

<class 'xugrid.core.wrap.UgridDataset'>
<class 'xugrid.core.wrap.UgridDataset'>
<class 'xugrid.core.wrap.UgridDataset'>
<class 'xarray.core.dataset.Dataset'>

[Huite]

The difference here is that uds1 = xu.UgridDataset(xu_ugrid2d.to_dataset()) results in a UgridDataset with assigned coordinates.

On the other hand, uds2 = xu.UgridDataset(grids=[xu_ugrid2d]) consists of the grid and an empty dataset. In this case, the wrapping code decides that the Dataset has nothing in common with the topology and returns an ordinary xarray dataset.

So once again, this is something that would be resolved by having explicit indices...

I can fix this in a roundabout way by adding coordinates to the dataset, but that feels rather clumsy as well. For your case, I guess I can suggest the workaround via mutation:

uds1.attrs = uds1.attrs.update({
    'institution': 'Deltares',
    'references': 'https://www.deltares.nl',
    })

Then, make sure you add actual data that refers to the UGRID dimensions.

(Sidenote: it's interesting that this problem hasn't appeared earlier. Basically every initialization with only grids is doomed unless the first you do is assign an actual DataArray -- I suppose this shows that it's a relatively rare way of creating a UgridDataset.)

[Huite]

Different workaround: if you assign node coordinates immediately, it should be preserved too.

uds2 = xu.UgridDataset(grids=[xu_ugrid2d])
uds2 = uds2.ugrid.assign_node_coords()
uds2 = uds2.assign_attrs(...)

Anyway, another argument for the explicit indexes as far as I'm concerned...