PMP modes of variability diagnostic fails with multi-file model datasets

[lewisjared]

Summary

The PMP extratropical modes of variability diagnostics systematically fail when model data is split across multiple files (e.g. one file per year), raising NotImplementedError("Only one model file is supported at this time.").

This affects all extratropical modes of variability diagnostics (NPO, NAO, SAM, PNA, PDO, etc.) from the pmp provider.

Error

File "packages/climate-ref-pmp/src/climate_ref_pmp/diagnostics/variability_modes.py", line 252, in build_cmd
    raise NotImplementedError("Only one model file is supported at this time.")

NotImplementedError: Only one model file is supported at this time.

Reproduction

Edit the CMIP6 test case to use source_id=AWI-ESM-1-1-LR

Then run via:
uv run ref test-cases run --fetch --force-regen --clean --provider pmp --diagnostic extratropical-modes-of-variability-npo --test-case cmip6

Root Cause

In packages/climate-ref-pmp/src/climate_ref_pmp/diagnostics/variability_modes.py:248-252, the build_cmd method collects model file paths via input_datasets.path.to_list() and then raises NotImplementedError if there is more than one file.

Other PMP diagnostics handle this already with a glob. Does this driver support that?

[lee1043]

Hi @lewisjared, thank you for posting this. Unfortunately enabling multiple model files as input (input parameter to take list) requires some work from upstream (i.e., PMP's MoV driver) side. Update in PMP side will lead to a newer PMP version (with CDAT removed while enso_metrics is still with CDAT, thus they will need a different separate virtual envs). But I am not sure if this to fit the REF launch timeline, as it may not the best timing to update PMP version when we have the REF launch next week. Would you share your thoughts?

[lewisjared]

The enso driver script is already vendored into the repo (

climate-ref/packages/climate-ref-pmp/src/climate_ref_pmp/drivers/enso_driver.py

Line 1 in 6968755

"""

). Is that a workaround for this? I agree we shouldn't take a new PMP version at this moment

[lee1043]

Just a brainstorming a temporary workaround for https://github.com/Climate-REF/climate-ref/blob/main/packages/climate-ref-pmp/src/climate_ref_pmp/diagnostics/variability_modes.py#L250-#L257

import xarray as xr

# Combine all files into one dataset along the 'time' dimension
if len(model_files) > 1:
    all_file_name = os.path.join(
        definition.output_directory, 
        f"{input_datasets}_{source_id}_{experiment_id}_{member_id}.nc"))
    ds_all = xr.open_mfdataset(model_files, combine='nested', concat_dim='time')   
    ds_all.to_netcdf(all_file_name)
    print(f"Combined {len(model_files)} files into {all_file_name}")
    modpath = all_file_name
else:
    modpath = model_files[0]

@lewisjared how about that?

[lewisjared]

My response didn't send apparently.

I'm not sure if simply removing the check will actually work? https://github.com/Climate-REF/climate-ref/blob/2cb3515e0302a530ad6fa89d4b7e57b5150253a9/packages/climate-ref-pmp/src/climate_ref_pmp/diagnostics/variability_modes.py#L250

The line after it apparently handles multiple files:

if isinstance(model_files, list):
    modpath = " ".join([str(p) for p in model_files])
else:
    modpath = model_files

I would start by removing L250 and adding a test with AWI to see if that works.

[lee1043]

I have tested it with simply removing the check as suggested, but unfortunately (and not surprisingly) it didn't solve as there to be the error from the upstream. Let me think a bit more on this what I can do, particularly without updating the PMP version in the REF.

[lee1043]

I found an working workaround -- please review #602 when you get a chance.