How best to proceed with using CMOR when including observational uncertainty information via variable_id suffixes

[dhegedus99]

Hello,

I am part of the REF delivery team, coordinating the use of observational datasets within the REF (Rapid Evaluation Framework, Github, website). The REF is currently only accepting datasets which have been approved by and published on ESGF via obs4MIPs. Obs4MIPs requires the observational datasets to be prepared by CMOR. Through various community engagement events, we have come up with a few requests for obs4MIPs in terms of the data formatting and the use of uncertainties (GH#167). This lead to very useful and informative discussions, and we identified that some of our requests might require updates to CMOR - which might not be feasible on short timescales.

In an ideal scenario, we would be able to record the uncertainty fields as ancillary variables within the same file following CF conventions (Example 3.3. Ancillary instrument data), but we think this is not be possible with CMOR yet. So my first question (in terms of longer term development): Is that something CMOR could be updated to accommodate? As in, have the main geophysical variable and ancillary uncertainty fields in the same file, following the CF conventions for including ancillary variables?

The alternative approach we have taken currently is including uncertainty information in separate files, but the variable_id for these additional files would be able to have certain suffixes (this is so that the REF can distinguish between the different files). We have come up with a list of suffixes we would like to be able to include for the REF (GH#188). For example, for precipitation, the file with the observations would have variable_id="pr", then the file with the standard error on the observations would have variable_id="pr_stderr". We weren't sure how best it would be for CMOR to handle this? Does it require some updates to CMOR, and do you think these would be substantial enough to limit this list to a small number of suffixes? Is it simpler and better for us to submit these variable_ids as a new variable when updating the obs4mips cmor tables? What would the developers of CMOR advise?

Please note, the REF is on a tight schedule, as we need to have datasets available within the next few weeks to start testing the metrics within the REF, but at the latest we would require the datasets to be available via obs4MIPs by 30th May (time of the Beta launch).

Thank you!

[durack1]

@dhegedus99 this is an interesting one. CMOR is built around the "CMIP data model" which since CMOR1/CMIP3 has been a single variable per file. This format is what we are planning toward for the CMIP7 contributions, and it's also the format that plays most cleanly with the ESGF system.

It would be possible to create REF specific CMOR tables, which define variables that you require, another option would be to augment the current https://github.com/PCMDI/obs4MIPs-cmor-tables to include these new variables, which would enable CMOR to write these - following the single variable per file format.

Pinging a couple of extra folks so they're in the loop @gleckler1 @taylor13 @mauzey1 @znichollscr @matthew-mizielinski

[durack1]

I will migrate this across to a discussion, as it's more philosophical CMOR usage than a code issue that requires CMOR changes