Bug Report: Inconsistent execution of Data Plausibility checks and numpy.ndarray error

[Ayoubnac1]

Summary

When running the Data Plausibility section (DATA001 → DATA006) of cmip6.py of the cc-plugin-wcrp compliance checks on different CMIP6 variables, the behavior is inconsistent between files :

• For some variables (e.g. tas, pr), the Data Plausibility checks (DATA001–DATA006) are not executed at all.
• For others (e.g. sfcWind), the checks do run, but one of them throws the following error:
  'numpy.ndarray' object has no attribute 'filled'`
• Despite this error appearing in all datasets, only sfcWind shows the full [DATA001] to [DATA006] block in the report, while tas and pr skip the entire group.

Observed Behavior

• Example excerpt from the sfcWind report:

{
  "name": "[DATA002] FillValue/MissingValue plausibility for 'sfcWind'",
  "weight": 2,
  "value": [0, 1],
  "msgs": [
    "Warning: _FillValue and missing_value are the same.",
    "Error during FillValue check: 'numpy.ndarray' object has no attribute 'filled'"
  ]
}

But in the corresponding tas and pr reports:

• No DATA001–DATA006 entries appear at all.
• The error message is not logged, even though the same exception occurs internally.

Steps to Reproduce

1. Run the plugin on the three different CMIP6 files (daily frequency):

compliance-checker -t wcrp_cmip6:1.0 pr_day_NorCPM1_historical_r1i1p1f1_gn_19500101-20141231.nc -f json -o output_pr.json
compliance-checker -t wcrp_cmip6:1.0 tas_day_NorCPM1_historical_r1i1p1f1_gn_19500101-20141231.nc  -f json -o output_tas.json
compliance-checker -t wcrp_cmip6:1.0 sfcWind_day_NorCPM1_historical_r1i1p1f1_gn_19500101-20141231.nc  -f json -o output_sfcWind.json

2. Compare the generated JSON reports :

• sfcWind → [DATA001]–[DATA006] appear with partial errors.
• tas and pr → missing the entire [DATAxxx] block.

3. Notice the identical traceback message printed during execution:
   'numpy.ndarray' object has no attribute 'filled'

Attachments

the 3 cmip6 test files to download via this link : https://filesender.renater.fr/?s=download&token=dafaaff9-de23-4ef5-953d-86428f044218

[Ayoubnac1]

@JavierDiezSierra

[chantreux]

Thank you for your detailed report and implementation of the Data Plausibility checks in the cc-plugin-wcrp compliance checker.
We’ve reviewed the structure and your implementation, and now is more clear how the checks are organized. Now that we understand the repository better, we’ll proceed to test the workflow directly using the test files you provided. We will propose a solution to the inconsistency when running the Data Plausibility section of the cc-plugin-wcrp (cmip6.py).

In fact we have identified some checks that are not implemented yet and that we will add when the current tests are correctly running:

• DATA001BIS(Inf): NaN/Inf check using the Inf parameter (currently, it only checks for NaN).
• DATA002BIS(MissingValue): Fill/Missing value check using the MissingValue parameter (currently, it defaults to FillValue).
• DATA005BIS(IQR): spatial statistical outlier check with IQR and threshold=1.5 (currently, it defaults to Zscore).

[chantreux]

@Ayoubnac1 Thank you again for sharing the CMIP6 test files. I’ve attempted to download and unzip the files, but unfortunately, the archive seems to be corrupted or incomplete, as I’m unable to extract its contents.
Could you kindly verify whether the upload was completed successfully? If possible, would you be able to re-upload the files or provide an alternative download link? If a CDS download of the same CMIP6 model is equivalent to the data you sent, we can directly download it.

[Ayoubnac1]

HI @chantreux sorry for the late reply, I was on vacation last week. I did the merge, but I'm a little confused because there are now two versions of each check and I don't know which one is the right one. For the release, I removed the versions from the title to integrate them into cmip6.py, so now there are duplicates. Thanks in advance.
https://github.com/ESGF/cc-plugin-wcrp/tree/csic-develop/checks/data_plausibility_checks

[chantreux]

Thank you for your reply. I’ve looked into the issue and realized what caused the duplication problem.
I apologize for not noticing earlier that you had renamed the checks (e.g., from check_chunk_size_v141.py to check_chunk_size.py) while working on the main branch. When I merged the CSIC branch with main to incorporate the latest updates, the old filenames remained in the CSIC branch, resulting in duplicate files.
I’ve corrected this and opened a new pull request with the fix.

[Ayoubnac1]

@JavierDiezSierra @chantreux Thank you for your quick response. I tested it with the changes, and the error is no longer occurring. I also wanted to let you know that, after further consideration, we believe that the data checks would be better implemented in a separate plugin like the cmip6 and cordex plugins, so that users can choose whether or not they want to enable them.
I will prepare a new release with the updated package structure following this new paradigm.

[JavierDiezSierra]

@Ayoubnac1 many thanks for your feedback. Yes, I think it’s a good idea to include the data checks separately so the user doesn’t have to deal with so many options. If they’re interested in computing data checks, then they can install the plugin.