Encountered CoordinateMultiDimError at area_statistics when trying to plot timeseries plots

[stacyqxz]

Hi all,
I was trying to plot timeseries plots for siconc and sithick anomalies using CMIP6 datasets. For both variables, the same error of iris.exceptions.CoordinateMultiDimError occurred at the preprocessor area_statistics.

For sithick, I was using GFDL-ESM4. For siconc, datasets include BCC-CSM2-MR, CanESM5, FIO-ESM-2-0, GFDL-CM4, GFDL-ESM4, IPSL-CM6A-LR, MIROC6, MPI-ESM1-2-HR, MPI-ESM1-2-LR, MRI-ESM2-0. I attached the recipes and corresponding debug log files here.

Any idea how I can fix this error?
recipe_cmip6_siconc_anomaly.txt
siconc_main_log_debug.txt
recipe_cmip6_sithick_anomaly.txt
sithick_main_log_debug.txt

[zklaus]

Admittedly it is a bit hidden in the _log_debug.txt (perhaps it's more visible in the _log.txt?), but if you look a little bit above the exception, you will find the following:

2022-06-08 17:47:44,693 UTC [3276653] DEBUG   esmvalcore.preprocessor._area:260 Cell measure "cell_area" not found in cube sea_ice_area_fraction / (%)         (time: 318; -- : 92; -- : 168). Check fx_file availability.
2022-06-08 17:47:44,693 UTC [3276653] DEBUG   esmvalcore.preprocessor._area:263 Attempting to calculate grid cell area...
2022-06-08 17:47:44,693 UTC [3276653] ERROR   esmvalcore.preprocessor._area:279 fx_file needed to calculate grid cell area for irregular grids.
2022-06-08 17:47:44,695 UTC [3276653] ERROR   esmvalcore.preprocessor:327 Failed to run preprocessor function 'area_statistics' on the data
[<iris 'Cube' of sea_ice_area_fraction / (%) (time: 318; -- : 92; -- : 168)>]
loaded from original input file(s)
['/fs/site5/eccc/crd/crd_shared/ESGF_downloads/CMIP6/CMIP/IPSL/IPSL-CM6A-LR/historical/r1i1p1f1/SImon/siconc/gn/v20180803/siconc_SImon_IPSL-CM6A-LR_historical_r1i1p1f1_gn_185001-201412.nc',
 '/fs/site5/eccc/crd/crd_shared/ESGF_downloads/CMIP6/ScenarioMIP/IPSL/IPSL-CM6A-LR/ssp585/r1i1p1f1/SImon/siconc/gn/v20190119/siconc_SImon_IPSL-CM6A-LR_ssp585_r1i1p1f1_gn_201501-210012.nc',
 '/fs/site5/eccc/crd/ccdp/users/scrd901/migration/scrd901/data/ESGF_DOWNLOADS/CMIP6/ScenarioMIP/IPSL/IPSL-CM6A-LR/ssp585/r1i1p1f1/SImon/siconc/gn/v20190903/siconc_SImon_IPSL-CM6A-LR_ssp585_r1i1p1f1_gn_201501-210012.nc']
with function argument(s)
operator = 'mean'

This tells us that we need to tell ESMValTool about the cell areas for IPSL by specifying the areacello fx file. The details are explained here.

Hope that helps.

[stacyqxz]

I've tried adding areacello as a fx_variable under area_statistics, but it still gave me the same error.

[zklaus]

I am certain that the log file changes. Could you please provide the log and the recipe so we can figure this out? In particular, contrary to the CMIP6 specification, several modeling groups have decided not to provide the cell area files for every ensemble member and every experiment when it only depends on the model configuration; as a consequence it is often necessary to specify those facets as well to locate the correct cell measure.

[stacyqxz]

main_log.txt
Couldn't find the log file for that specific run. But this has the same error.

[zklaus]

What about the recipe? I suspect you need to modify the way you are presenting the areacello information.

[stacyqxz]

I think the same too. But the log file I just upload had no areacello specified in the recipe. If I added areacello in the preprocessor (recipe_cmip6_siconc_anomaly.txt), it'd give me a different error as in main_log_debug.txt.

[stacyqxz]

So given fx_variable didn't work and the tripolar ocean grids couldn't be collapsed over lat and lon, I found that the output of a preprocessor as follows is 2D + time.

  siconc_djf_change_mapping: &siconc_preproc
    regrid:
      target_grid: 1x1
      scheme: nearest
    extract_region:
      start_longitude: -145.5
      end_longitude: -44.5
      start_latitude: 35.5
      end_latitude: 89.5
    extract_season:
      season: DJF
    seasonal_statistics:
      operator: mean
    multi_model_statistics:
      span: full
      statistics: [mean]
      keep_input_datasets: true

In this way, I was able to do cube.collapsed(['lat', 'lon'], iris.analysis.MEAN) in the diagnostic script without regridding.

[zklaus]

I am happy if you can solve your problem, but what you write here does not check out. The first step in this preprocessor is regridding, so of course the problem then doesn't show up after that. Let's try to correct your cell area specification above to actually solve the problem.