The MINI cases

MINI is a hand-checkable ice sheet: an 11 × 11 grid of 600 km pixels, small enough that the whole state fits on a screen and a wrong answer is visibly wrong. It exists to catch mistakes that a full ice sheet would hide in its own noise.

cd manual-tests/MINI
python scalars_MINI.py --model MINI1 --exp exp0   # one case
python run_MINI.py                                # all four, with a summary

Both read their input from the committed test-data/ tree, so they run in a fresh checkout with nothing else set up. Output goes to manual-tests/MINI/output/.

The cases

Two grids and two experiments:

Grid

Size

Origin

Notes

MINI1

11 × 11

−3,040,000 m

the primary test grid

MINI0

12 × 12

−3,340,000 m

offset by half a cell, for CDO remapping tests
exp0

ice loss that includes partially floating cells, so the grounding line falls inside a cell and the two grids resolve it differently.

expg

grounded ice only. With no floating ice and no grounding-line migration the three sea-level methods are describing the same physics, so they must agree exactly – which makes this the sharpest check in the suite. Any disagreement is a bug in one of them.

The pair of grids is what makes remapping testable: the same ice on a grid offset by half a cell must give the same total ice volume change, while VAF may legitimately differ by a few per cent where the grounding line is unresolved. tests/test_mini_smoke.py asserts both.

How MINI differs from the real processing

scalars_MINI.py is a separate, deliberately simpler script, not a mode of ismip7-scalars:

  • the first timestep of the experiment is the reference; there is no separate historical run;

  • the standard densities from slc.sl_constants are used directly, with no params.nc;

  • area weighting is dx² directly;

  • it writes the volume diagnostics (slc_Vtot, slc_Vgr, slc_Vfl) that the full processing does not.

It shares the ismip7_scalars.slc package with the real processing, which is the part worth cross-checking; the rest is scaffolding around it.

Regenerating the input files

manual-tests/MINI/setup/ holds the scripts that built the committed inputs – derive_exp0.py, check_masks.py, and the CDO/NCO shell scripts for remapping and grounding. They need CDO and NCO on the path, which is why they are not part of the test suite. The files they produce are committed under test-data/, so nothing routine needs them.