icvert variable converted to float when saving to NetCDF

[tomvansteijn]

The icvert variable in the nlmod model dataset is converted to float when saving to NetCDF. The variable has the attribute "_FillValue" set to -1. This attribute is lost when saving to NetCDF. The changes can cause problems when writing the DISV package.

Is this a NetCDF issue? Or is it caused by missing values in the icvert variable for certain coordinate combinations?

[OnnoEbbens]

It has to do with how NetCDF stores missing values in an integer array. For more information see this issue: #91

Did you encounter problems when writing the DISV?

[rubencalje]

So the saved NetCDF-file does contain integer values for the variable icvert, but when you read the NetCDF again using xarray, the integer values are converted to floats, and the values equal to _FillValue are set to NaN. To make sure this does not happen, read the NetCDF-file with the parameter mask_and_scale set to False. Like:

ds = xr.open_dataset(fname, mask_and_scale=False)

[tomvansteijn]

Thanks for the clarification @rubencalje. I meant to provide a working example but I haven't got to it yet. I trust this will solve the issue I had. Indeed the _FillValue was set to NaN, and NaN's were written to the DISV package files, with a Modflow error as a result. Good to include this solution (about the mask_and_scale flag) in the docs if you haven't already.

[OnnoEbbens]

Fixed by #238