Multi-instance timeseries capability for standalone CUPiD

[mollymwieringa]

Description of changes:

• Please add an explanation of what your changes do and why you'd like us to include them.

These changes take first steps toward issue #396 by enabling the run_timeseries.py tool to find model history files produced by a multi-instance-enabled case of CESM (e.g. cice_000X.h instead of cice.h). It involves changes to run_timeseries.py and adding a multi-instance example with its own config file to standalone CUPiD.

These changes will help enable efficient postprocessing of many history files for multi-instances cases into more storage-efficient timeseries files, reducing the amount of overhead required by multi-instance simulations. This benefit is particularly relevant to data assimilation experiments, which use multi-instance simulations to run large ensembles that output history files at high temporal frequencies.

Outstanding work on this project during draft PR stage:

• testing as a part of the CESM workflow (requires updates to my CESM version to include CUPiD in the workflow, needs changes to cesm_postprocessing.sh and maybe generate_cupid_config_for_cesm_case.py)
• addition of a diagnostic notebook that can plot multi-instance timeseries as an ensemble with an ensemble mean

All PRs Checklist:

• Have you followed the guidelines in our Contributor's Guide?
• Have you checked to ensure there aren't other open Pull Requests for the same update/change?
• Have you made sure that the pre-commit checks passed (#8 in Adding Notebooks Guide)?
• Have you successfully tested your changes locally when running standalone CUPiD?
• Have you tested your changes as part of the CESM workflow?
• Once you are ready to have your PR reviewed, have you changed it from a Draft PR to an Open PR?

[mnlevy1981]

I'll wait for you to mark this as "ready for review" before going through it in detail, but one thing that jumped out is a way to avoid needing to add ninst to every configuration file.

I didn't really look through the new config.yml file, but I noticed it is pointing to output in your scratch space -- I'll make sure you have access to /glade/campaign/cesm/development/cross-wg/diagnostic_framework/CESM_output_for_testing/ and then you can copy the files you need over (right now it looks like we're testing on 21 years of ensemble output, maybe we could reduce that to just a few years to save disk space?)

[mnlevy1981]

If you use ninst = timeseries_params.get("ninst", 1) then examples without ninst in the time series block will default a single instance.

[mollymwieringa]

Okay, good idea! I've added this into my branch-- it'll come through once everything else is ready for review.

[mnlevy1981]

👍

[mollymwieringa]

@mnlevy1981 Just a note that the test case I set up definitely shouldn't be processing 21 years-- I think the 3 simulations are each only a month long. I'll double-check, but let me know if that's still too much storage!