unstructured grid Error

[HannahGaenslen]

Dear ESMValTool Team,

When I was testing the monitoring recipe (recipe_monitor.yml) for a CNRM-CM6-1 dataset I encountered the following error:

ERROR   esmvalcore.preprocessor:338 Failed to run preprocessor function 'mask_landsea' on the data
[<iris 'Cube' of sea_surface_temperature / (degC) (time: 180; -- : 294; -- : 362)>]
loaded from original input file(s)
['/data/scratch/globc/dcom/CMIP6/CMIP/CNRM-CERFACS/CNRM-CM6-1/historical/r1i1p1f2/Omon/tos/gn/latest/tos_Omon_CNRM-CM6-1_historical_r1i1p1f2_gn_185001-201412.nc']
with function argument(s)
mask_out = 'land'
…
ValueError: Use of shapefiles with irregular grids not yet implemented, land-sea mask not applied.

How can I fix it?

I thought of adding another preprocessor for irregular grids to the recipe. I found this _regrid_esmpy.py preprocessor, but I don’t know how to use it correctly since I didn’t found any documentation on it.

Thank you for your help.
Cheers,

Hannah

recipe_monitor.txt
main_log_debug.txt

[schlunma]

Hi @HannahGaenslen!

To mask data on irregular grids, you need the land/ocean fraction (sftof) variable for that dataset. In your case, ESMValTool does not find that (see l. 104 in the debug log):

2023-05-04 14:44:07,939 UTC [375195] WARNING esmvalcore._recipe:445 Missing data for fx variable 'sftof' of dataset CNRM-CM6-1

You could first try to download the data automatically with ESMValTool using the command line option --offline=False (for v2.7.0, the one that you are using). However, in v2.7.0, ESMValTool is not very smart in finding this data. It uses the same facets you use for the "main" variable (in your case, exp: historical, ensemble: r1i1p1f2, grid: gn). Unfortunately, for many CMIP6 models, not every experiment contains the corresponding sftof files (e.g., some provide it only in the piControl experiment).

For this reason, we completely changed how these "supplementary variables" are handled in v2.8.0 (see here for an overview). Chances are high your recipe will just work without changes in v2.8.0 (you might have to use the automatic download using --search_esgf=when_missing in v2.8.0). Alternatively, you could specify different facets for the supplementary data, see here for more details.

[HannahGaenslen]

Hi 👋,

Thanks for your answer! I managed to fix the warning, by defining the grid for the sftlf variable (the model I use, doesn't produce sftof). Instead of using these "supplementary variables" as in v2.8.0 (I'm currently using v2.7.0) I specified the variable in the preprocessing step. Unfortunately, the error remains the same even if it's possible to load in the land fraction (sftlf).

@mpmoine and I tried the approach of adding a regridding preprocessor, but this also didn't fix the error. We think the reason is that the default order of the preprocessors calls mask_landsea before regrid. Is there any way to change the order?
This is how the preprocessor looks right now:

nino34: &index
    regrid:
      target_grid:
        start_longitude: -170.
        end_longitude: -120
        step_longitude: 1
        start_latitude: -5.
        end_latitude: 5.
        step_latitude: 1
      scheme: unstructured_nearest
    mask_landsea:
      mask_out: land
      fx_variables:
         sftlf:
          grid: gr
    area_statistics:
      operator: mean
      fx_variables:
        areacello:

Cheers,
Hannah

main_log_debug.txt
recipe_monitor.txt

[schlunma]

You can define a custom preprocessor order by using custom_order=true in the recipe, but I am afraid that regridding does not help here: When using supplementary variables, ESMValTool tries to add these before any other preprocessing steps like regridding in add_fx_variables. However, since the sftlf variable uses the model's atmospheric grid (which is different to the oceanic grid your main variable uses), it cannot be added:

2023-05-09 16:00:35,749 UTC [300308] DEBUG   esmvalcore.preprocessor._ancillary_vars:63 Dimensions of mlotst and sftlf cubes do not match. Discarding use of fx_variable: cannot broadcast shape (128, 256) to shape (180, 294, 362)

The only solution to this is using the land/sea fraction defined on the oceanic grid, usually called sftof.

Note that this comment assumes v2.7.0 of ESMValTool, but the main conclusion also applies to v2.8.0.

[HannahGaenslen]

Okay, thank you.
Since the model I specifically wanted to use here doesn't produces the variable, we are now working on generating sftof so it can be used in ESMValTool.