Processing a whole ensemble¶

ismip7-scalars-ensemble walks a submission tree, finds every experiment it can process, pairs each projection with its historical run, and runs ismip7-scalars once per experiment. Each run is a separate process, so one bad submission cannot take the batch down.

Point --modelpath at the ice sheet directory of the submission tree, the one holding a folder per group. On NIRD that is the second GrIS in the path:

ismip7-scalars-ensemble --region GrIS \
  --modelpath /nird/datalake/NS5011K/ISMIP/ISMIP7/GrIS/ISMIP7_output/ISMIP7_submissions/GrIS \
  --datapath /nird/datalake/NS5011K/ISMIP/ISMIP7/Output-Processing/Data/GrIS \
  --params-path /nird/datalake/NS5011K/ISMIP/ISMIP7/Output-Processing/Models/GrIS \
  --dry-run

--dry-run prints the command planned for each experiment and runs none of them. Start there.

What gets processed¶

An experiment directory four levels below the model path, for example NORCE/CISM16x-MAR312-p50/CORE/C001, whose lithk file names the same region, group, model and configid as the directory it sits in.

Anything else is logged and skipped: ad-hoc directories like old_CORE, filenames that break the ten-field convention, a file that says it belongs to a different configid than its directory. A mislabelled file is never processed as though it were something else.

How projections are paired with a historical run¶

The CORE experiment table bundled with the package gives each configid its scenario and forcing ESM. A projection is paired with the historical run driven by the same ESM: CESM projections with C001, MRI projections with C002. Historical and ctrl experiments are their own reference.

For a configid the table does not know, such as a future ESM or PPE experiment, the driver pairs an even configid with the odd one below it. If that gives nothing, the experiment is skipped with a reason rather than measured against itself, which would produce a plausible-looking series that is wrong. Pair those by hand with ismip7-scalars --hist-configid.

Options¶

Option	Meaning
`--region`	AIS or GrIS; required
`--modelpath`	ice sheet directory of the submission tree; required
`--dry-run`	print the planned commands, run nothing
`--exp-group`	only this experiment group
`--groups`, `--models`, `--configids`	comma-separated filters
`--datapath`, `--params-path`, `--outpath`, `--histout`, `--basins`	passed through to each run
`--core-csv`	use a different experiment table
`--python`	interpreter to run each experiment with
`--log-dir`	where the logs go; default Output/logs

Logs¶

One log per experiment, plus a summary naming every experiment and what happened to it:

ISMIP7 ensemble run -- GrIS -- 20260301T142530
modelpath: /nird/.../ISMIP7_submissions/GrIS
units: 24   ok: 19   skipped: 4   failed: 1

  NORCE/CISM16x-MAR312-p50/CORE/C001                       OK   [GrIS_NORCE_..._C001.log]
  NORCE/CISM16x-MAR312-p50/CORE/C003                       OK   [GrIS_NORCE_..._C003.log]
  VUW/PISM1/CORE/C005                                      SKIP: ... -- missing topg for exp in ...
  ...

The driver exits 0 unless something went wrong with the driver itself. An experiment that could not be processed is a line in the summary, not a failure of the batch, so read the summary.