Allow user override of Conventions attribute

[matthew-mizielinski]

At present the Conventions attribute can only be modified through changes to the Header entries within the MIP tables. There are discussions for CMIP7 of allowing CF-1.11, CF-1.12 and CF-1.13, but we would have to maintain multiple sets of MIP tables to support this (not something I am willing to do).

Would it be possible to allow the user to override the Conventions attribute at run time through an entry in the input JSON file?

It might also be worth adding a warning when using any of the quantization options if the Conventions version is below CF-1.12

[taylor13]

Would it be possible to allow the user to override the Conventions attribute at run time through an entry in the input JSON file?

Hi Matt,
Chris, Paul, and I (@mauzey1 @durack1) discussed this yesterday and I was tasked with raising an issue, which you've now done for me. Thanks!

I had suggested removing Conventions from the variable tables, but Paul pointed out that CMOR would then need to handle CMIP6 differently from CMIP7. (Of course other things in the CMOR tables have changed from 6 to 7, so perhaps that's not a difficult problem.)

If we did remove it from the variable tables and instead invariably set it In the "input" JSON file (example here), would you put it below line 58 (things that the user would not normally change) or above that?

(By the way, Matt, do you have a CMIP7 sample "input" file that you're willing to share? I'd like to see any changes.)

[taylor13]

There is related discussion in WCRP-CMIP/cmip7-cmor-tables#48 . Do we need to modify the CMIP7-cv.json file too?

[durack1]

Ideally, data for any project would be written to a specific convention (exactly what the "Conventions": "CF-1.7 CMIP-6.2" was meant to indicate - CF 1.7 with the CMIP 6.2 data spec. For this reason, it would make sense that the project_cv.json (currently the cmip7-cmor-tables/tables-cvs/cmor-cvs.json would contain this info, attempting to keep all data within a project consistent. If you move this to the user_input.json file, there is limited control, and pushes the requirement to validate this to the data publication process.

Of course, pragmatics may suggest you just want to open this wide up, but I would avoid loosening things too much if we can, and attempt to standardise at the project level.

Philosophically, the variable registry (or table files that define variables) should be oblivious to what data conventions are used to generate data using the variable definitions in the table files, but this needs to be thought about again for backward compatibility, in the case that folks want to continue to use the cmip6-cmor-tables, or mip-cmor-tables for anything, which to be honest is likely a bad idea, would be better just to rip the bandaid off and migrate all projects forward

[taylor13]

CV task team discussions wants to allow CF-1.11, CF-1.12, or CF-1.13 as options. Those who make use of lossy compression need features that are in CF-1.12, and those writing healpix grid data probably nee CF-1.13. But the CF checkers can only check files against 1.11 or earlier, so some people are uncomfortable specifying the later versions. On balance it was decided to allow any of the 3 options (which would appear in the CV as a list.

Since the choice among the 3 is up to the data provider, I think it best to allow "Conventions" to be specified in the "input" json file.