CDO

Hello James,

The NC_STRING data type is not handled correctly in CDO for the rotated_pole variable. We will fix the problem in the next CDO bugfix release 2.1.1.
In the first file (test-cordex.nc) the problem does not occur, because the first data variable (AWAP_qualitymask) in this file does not use the rotated_pole coordinate. This leads to the fact that also for the variable tasmax-bc the rotated_pole is not used.

Cheers,
Uwe

Hi James,

the variable tasmax-bc has the attribute grid_mapping = "rotated_pole" (this is ok) but the variable AWAP_qualitymask has the attribute grid_mapping = "Rotated_pole", this has to be "rotated_pole" too. This leads to the the warning message

Warning (cdfScanVarAttr): NetCDF: Variable not found - Rotated_pole

ncdump -h test-narclim.nc

...
    float AWAP_qualitymask(rlat, rlon) ;
        AWAP_qualitymask:_FillValue = 1.e+20f ;
        AWAP_qualitymask:standard_name = "quality_mask" ;
        AWAP_qualitymask:long_name = "AWAP quality mask" ;
        AWAP_qualitymask:units = "" ;
        AWAP_qualitymask:description = "Mask to define areas where AWAP might be unreliable" ;
        AWAP_qualitymask:grid_mapping = "Rotated_pole" ;
        AWAP_qualitymask:coordinates = "lon lat" ;
...
    string rotated_pole ;
        rotated_pole:grid_mapping_name = "rotated_latitude_longitude" ;
        rotated_pole:dx_m = 9783.618 ;
        rotated_pole:dy_m = 9783.618 ;
        rotated_pole:grid_north_pole_latitude = 60.31 ;
        rotated_pole:grid_north_pole_longitude = 147.63 ;
...

You can run the following Python script to make some corrections and write it to a new file test-narclim.nc. Then you can use it with CDO.

#!/usr/bin/env python
# coding: utf-8

import xarray as xr
from cdo import Cdo
cdo = Cdo()

infile = 'test-narclim.nc'

ds = xr.open_dataset(infile)

ds2 = ds.copy(deep=True)

rp_new = xr.DataArray(data=[0], coords=None, dims=None, attrs=ds2.rotated_pole.attrs)

ds2['rotated_pole'] = rp_new
ds2['AWAP_qualitymask'].attrs['grid_mapping'] = 'rotated_pole'

outfile = 'test-narclim_corrected.nc'

ds2.to_netcdf(outfile)

Thanks very much, Uwe and Karin! It looks like I blew off the grid_mapping variable prematurely 😅

Working with this data before cdo 2.1.1 comes out, I'm just doing temporal and ensemble statistics in cdo - and the final step where I'm doing field averages is in R, where I point it to the curvilinear grid variables rlat and rlon.

Is it likely that simply deleting the grid_mapping variable altogether (and potentially restoring it after my temporal and ensemble stats are calculated) is going to have any effect on my analysis?

(I can see that the grid_mapping var has important information on the projection, and it's required by CF-1.4 - see page 3 of https://is-enes-data.github.io/cordex_archive_specifications.pdf - but it's not clear to me that cdo is actually taking it into account when aggregating on time and across ensembles.) I've run my first analysis step and it doesn't seem like the grid is affected.

Karin Meier-Fleischer wrote in RE: "Not a valid data type or _FillValue type mismatch", ...:

Hi James,

the variable tasmax-bc has the attribute grid_mapping = "rotated_pole" (this is ok) but the variable AWAP_qualitymask has the attribute grid_mapping = "Rotated_pole", this has to be "rotated_pole" too. This leads to the the warning message...

Thanks Karin! I was able to fix this warning similarly by modifying the grid_mapping attribute with ncatted. It unfortunately didn't fix the error too, but I'm proposing a solution (for my use case) above in this post.

My mistake: I'm having success removing the grid_mapping attributes, not variable.

ncks -xv grid_mapping test-narclim.nc test-narclim-nogridmap.nc

cdo -selname,tasmax-bc test-narclim-nogridmap.nc test-narclim-result.nc

The grid_mapping attributes still seem to be in test-narclim-nogridmap.nc, so I'm not 100% sure why this seems to work.