default fillvalue not parsed by xarray, results in negative values in connectivity

[veenstrajelmer]

xarray does not use the default netcdf4 fillvalue in decode_cf, only the explicitly set _FillValue attribute: pydata/xarray#7723 and pydata/xarray#2374 and pydata/xarray#2742

This raises "ValueError: connectivity contains negative values":

import xugrid as xu
file_nc = r'p:\archivedprojects\1210301-dflow-fm\MeinardTiessen\model\NEVLA\a19\DFM_OUTPUT_WS13_new_04\WS13_new_04_0000_map.nc'
ds = xu.open_dataset(file_nc)

For now, a workaround is to add the default fillvalue to the attributes before feeding to xugrid:

import xugrid as xu
import xarray as xr

file_nc = r'p:\archivedprojects\1210301-dflow-fm\MeinardTiessen\model\NEVLA\a19\DFM_OUTPUT_WS13_new_04\WS13_new_04_0000_map.nc'

ds = xr.open_dataset(file_nc)

#add default fillvalue if missing from connectivity array attrs
varn_conn = 'NetElemNode'
if ds[varn_conn].min()<0 and '_FillValue' not in ds[varn_conn].attrs:
    from netCDF4 import default_fillvals
    dtype_str = ds[varn_conn].dtype.str[1:]
    ds[varn_conn] = ds[varn_conn].assign_attrs({'_FillValue':default_fillvals[dtype_str]})
    #decode_cf again, fillvalues are replaced with nan and _FillValue attr is moved to encoding
    ds = xr.decode_cf(ds)

uds = xu.core.wrap.UgridDataset(ds)

We will probably wait on xarray to fix their issue. If it takes too long, we might want to implement a workaround.

This is also an issue for non-connectivity variables: Deltares/dfm_tools#490

[veenstrajelmer]

This will not be added in xugrid. It is temporarily added in dfm_tools in Deltares/dfm_tools#496 and it will be removed again once xarray handles default fillvalues.