Skip to content

SMB water-budget diagnostics and conservation fixes for coupled runs - #95

Closed
whlipscomb wants to merge 4 commits into
lipscomb/separate_glide_glissadefrom
lipscomb/smb_conservation
Closed

whlipscomb wants to merge 4 commits into
lipscomb/separate_glide_glissadefrom
lipscomb/smb_conservation

Conversation

@whlipscomb

@whlipscomb whlipscomb commented Nov 9, 2026 •

Copy link
Copy Markdown
Contributor

Branch: lipscomb/smb_conservation
Base: lipscomb/separate_glide_glissade (PR #94, at 8b3bac8). Depends on PR #94 (stacked PR); the base will be changed to main after #94 is merged.

Matching CISM-wrapper PR: ESCOMP/CISM-wrapper #156 (stacked on CISM-wrapper PR #155).
Coupler change: ESCOMP/CMEPS PR #720.
4 commits, November 6-7, 2026

Summary

In CESM runs with an evolving Greenland ice sheet, the land model's meltwater exceeded the ice that CISM actually lost by about 40-60 Gt/yr, which shows up as a freshwater drift in the coupled system. This PR adds diagnostics that measure where SMB received from the coupler is not applied, and two changes that remove the imbalances found:

  1. SMB budget diagnostics (fb1379c, f02c5c8). New fields record, in each cell and globally, the SMB received from the coupler, the part that GLAD sets to zero before the dycore sees it, and the part the dycore does not apply. The budget closes exactly: received = applied + unapplied (GLAD) + unapplied (dycore).
  2. Fractional ice_covered (5fcc9d2). A new option, ice_covered_thck_scale in the [GLAD climate] section, makes GLAD send ice_covered = min(1, thck/ice_covered_thck_scale) instead of 1 wherever thck > 0. This removes almost all of the imbalance. The default, 0, keeps the binary field.
  3. Return unapplied positive SMB as solid runoff (028d39f). Positive SMB that the dycore does not apply (for example, removed by block_inception) is added to rofi, so that the water the land model removed as SMB is not lost.

The fractional ice_covered requires a coupler that accepts fractional values: ESCOMP/CMEPS PR #720 (ESCOMP/CMEPS#720). The matching CISM-wrapper changes are on ESCOMP/CISM-wrapper lipscomb/smb_conservation.

Where the imbalance came from

The diagnostics show that almost all of the imbalance is in cells holding only a trace of ice. GLAD reported any cell with thck > 0 as fully ice-covered, so the coupler gave it a full year of low-elevation glacier SMB (ablation of ~1 m/yr or more), and the land model melted that much glacier ice (glacier_region_melt_behavior = 'replaced_by_ice') and sent the water to runoff. CISM could melt only the few mm or cm present. The rest was lost in two ways:

  • in the dycore, as melt beyond the ice present (about a third), and
  • in GLAD, which applies its mask at each dycore time step within the coupling interval: once the trace ice has melted, GLAD sets the negative SMB of the now ice-free cell to zero for the rest of the year (about two thirds).

Ice re-formed in many of these cells (from bare-land accumulation and transport), so they switched between ice-covered and ice-free from year to year.

In a 1-year run (model year 53), 94% of the 62 Gt/yr imbalance was in cells with less than 10 cm of ice at the start of the year:

Start-of-year thickness Cells Imbalance (Gt/yr) Mean coupler SMB (m/yr ice)
0 < H < 1 cm 2,179 38.2 -1.20
1-10 cm 1,056 20.7 -1.42
10 cm - 1 m 256 3.1 -1.16
1-10 m 697 0.2 -0.10
>= 10 m 117,086 0.0 +0.27

With a fractional ice_covered, a cell with thin ice gets a blend of glacier and bare-land SMB, weighted by its ice cover. In a single-cell model, this reaches a steady state with no unapplied melt whenever the thickness scale exceeds the difference between bare-land and glacier SMB over one coupling interval; a thickness threshold, or the fraction of the year with ice present, instead gives cells that keep switching states.

Results

1850 I-compset runs (1850_DATM%CPLHIST_CLM60%BGC-CROP_SICE_SOCN_MOSART_CISM2%GRIS-EVOLVE, ne30pg3 land, gris4, DATM coupler history from a B run), 20 years (53-72) with the same forcing:

Control (binary) H0 = 2 m H0 = 4 m
CISM-side imbalance (Gt/yr) 36 to 53 0.01 to 0.16 0.00 to 0.10
Cells switching ice-covered/ice-free per year 280-340 22-57 14-32
CISM applied SMB - CLM QICE (Gt/yr) 31-48 1.5-7.5 1.5-7.2
Ice area with H > 10 m, year 72 (1000 km2) 1872.9 1872.8 1872.9

The remaining 1.5-8 Gt/yr between CLM QICE and CISM's applied SMB is not a conservation error: it is bare-land SMB (which QICE omits) and glacier SMB in coastal land cells partly outside CISM's ice mask, which CTSM handles conservatively. Away from the thin margin, the ice sheet is unaffected; the fixed runs keep trace and thin margin ice in steady states. We plan to make 4 m the CESM default once CMEPS PR #720 is in a CESM tag.

Changes

New output fields (glissade_vars.def; m/yr ice in output, each with a _tavg version):

  • acab_cpl: SMB received from the coupler, saved by GLAD before any masking
  • acab_unapplied_glad: SMB set to zero by GLAD (negative SMB in ice-free cells; any SMB in ice-free ocean cells)
  • acab_unapplied: SMB passed to the dycore but not applied (melt beyond the ice present; ocean cells; partly ice-covered cells; adjustments such as blo ck_inception)

New global scalars (kg/s, each with a _tavg version): total_smb_cpl_flux, total_smb_unapplied_glad_flux, total_smb_unapplied_flux. In coupled runs these are also written to the log, in kg/s or Gt/yr (as dm_dt_diag chooses).

New config option ice_covered_thck_scale in [GLAD climate] (m, default 0).

Solid runoff: rofi now includes the positive part of acab_unapplied. (Ice removed as ice caps, as thin ice or by forced retreat, removal_thck, and calved icebergs, calving_thck, were already included.)

Answer changes

  • Standalone runs: none (output and log files unchanged).
  • CESM, default settings (ice_covered_thck_scale = 0): the ice sheet is unchanged. rofi increases where positive SMB is not applied (about 2-5 Gt/yr for Greenland, mostly from block_inception). The new diagnostic fields are written only if requested (the CISM-wrapper branch adds them to the default history fields).
  • CESM with ice_covered_thck_scale > 0: answers change as described above. Requires CMEPS #720.

Testing

All on derecho.

  • Standalone suite (dome DIVA/SLAP/Glide on 1 task, dome DIVA/L1L2/local SIA/SSA on 4 tasks and BP on 8, DIVA and Glide restart tests, Halfar, ISMIP-HOM A and C, confined and circular shelves, slab, Glint, EISMINT-2), compared to the standalone baseline used for PR Separate the Glide and Glissade dycores; remove obsolete code #94: all output fields and log files identical.
  • CESM aux_glc subset with the CISM-wrapper lipscomb/smb_conservation branch (default settings, standard CMEPS), compared to the baseline from PR Separate the Glide and Glissade dycores; remove obsolete code #94 (SMS_D_Ly1.f09_g17_gris4.T1850Gg intel and gnu, ERS_Ly5.f09_g17_ais8.T1850Ga intel, ERI_Ly9.f09_g17_ais8gris4.T1850Gag gnu noevolve_ais): all builds, runs and restart/branch comparisons pass. The CISM history (h0i) files are identical to the baseline. The expected differences are rofi in the coupler history (glcImp_Fgrg_rofi, from the positive SMB now returned as solid runoff), the new annual h0a history files (no baseline counterpart), and the namelists (new ice_covered_thck_scale entry and history fields).
  • With fractional ice_covered (CISM-wrapper testmod cism/fractional_ice_covered, H0 = 4 m) and CMEPS #720, all PASS including exact-restart comparisons: ERS_Ly5.f09_g17_gris4.T1850Gg.derecho_intel, ERS_Ly4.f10_f10_ais8gris4_mg37.I1850Clm50SpGag.derecho_intel, SMS_D_Ly1.f09_g17_ais8.T1850Ga.derecho_gnu.
  • 1-, 5- and 20-year I-compset runs as above. A 1-year run with ice_covered_thck_scale = 0 and CMEPS #720 is bit-for-bit with the old CMEPS (coupler lnd2glc and glc export history, CISM state).

Notes for reviewers

  • The budget diagnostics for coupled runs should be useful beyond Greenland; e.g., for Antarctica the dycore's unapplied SMB will also include partly ice-covered calving-front cells.
  • A related pattern remains for basal melt: GLAD's rofl uses bmlt, not bmlt_applied. This probably matters little for Greenland and could be revisited for coupled Antarctic runs.

🤖 Generated with Claude Code

whlipscomb and others added 4 commits November 7, 2026 13:55
In CESM runs with an evolving ice sheet, not all of the SMB received
from the coupler is applied to the ice. This commit adds diagnostics
to measure the unapplied SMB and where it occurs.

New 2D fields (m/yr ice in I/O, m/s ice in the code), each with a
time-average version:
* acab_cpl: SMB received from the coupler, saved by GLAD before any
  masking.
* acab_unapplied_glad: SMB zeroed by GLAD before the dycore sees it,
  i.e. negative SMB in ice-free cells and any SMB in ice-free ocean
  cells. GLAD applies this masking at each ice dynamics time step
  within the coupling interval, using the current ice thickness.
* acab_unapplied: SMB passed to the dycore but not applied
  (acab - acab_applied), e.g. melt in excess of the ice present.

New global scalars (kg/s), each with a time-average version:
total_smb_cpl_flux, total_smb_unapplied_glad_flux and
total_smb_unapplied_flux. In each cell and globally,
  acab_cpl = acab_applied + acab_unapplied_glad + acab_unapplied
  total_smb_cpl_flux = total_smb_flux + total_smb_unapplied_glad_flux
                       + total_smb_unapplied_flux
For coupled runs, these totals are also written to the log file.

Standalone runs are unchanged, including their log files.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The SMB budget diagnostics did not close: in a 1-year CESM run with an
evolving Greenland ice sheet, the SMB received from the coupler exceeded
the sum of the applied and unapplied terms by 2.25 Gt/yr. The missing
SMB was positive acab removed by block_inception, which zeroes positive
acab outside the extended ice sheet mask after GLAD has passed acab to
the dycore, and before acab_unapplied was computed.

Now acab_unapplied is the acab received by the dycore, saved at the
start of glissade_prepare_climate_forcing before any adjustments
(overwrite_acab, block_inception, dthck_dt correction), minus
acab_applied. With this change the budget closes in each cell:
  acab_cpl = acab_applied + acab_unapplied_glad + acab_unapplied

Standalone answers are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GLAD sends the coupler an ice_covered field, used by the coupler to
choose between bare-land and glacier SMB for each CISM cell, and by the
land model to set its glacier areas. Currently ice_covered = 1 where
thck > 0, else 0. As a result, cells with a trace of ice (a few mm or
cm) are assigned a full year of glacier ablation, most of which cannot
be applied. In CESM runs with an evolving Greenland ice sheet, this
leaves about 60 Gt/yr of land-model meltwater with no matching ice
loss, and many such cells flicker between ice-covered and ice-free
states from year to year.

This commit adds a parameter ice_covered_thck_scale in the [GLAD
climate] section. If ice_covered_thck_scale > 0, then
  ice_covered = min(1, thck/ice_covered_thck_scale),
so that thin ice is treated as partly covering the cell. In a
single-cell model, this gives a steady state with no unapplied melt,
provided ice_covered_thck_scale exceeds the difference between the
bare-land and glacier SMB over one coupling interval.

The default, ice_covered_thck_scale = 0, keeps the current binary
ice_covered field, so answers are unchanged by default. Fractional
values require a coupler that supports them; the current CMEPS code
aborts unless ice_covered is 0 or 1.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… runoff

In coupled runs, the land model removes the SMB from its water budget
and sends it to GLAD. If the dycore does not apply a positive SMB, the
water is lost. In a CESM run with an evolving Greenland ice sheet, about
2.25 Gt/yr of positive SMB was lost this way, removed by block_inception
outside the extended ice sheet mask. Positive SMB could also be lost in
ice-free ocean cells and in partly ice-covered calving-front cells.

Now the positive part of acab_unapplied is added to rofi, the solid ice
runoff sent to the coupler, along with calving_thck and removal_thck.
Negative values of acab_unapplied (melt in excess of the ice present)
cannot be returned as runoff and are not included.

Also, correct a comment: icebergs are part of calving_thck, not
removal_thck. Both are included in rofi.

Ice sheet answers are unchanged; rofi changes where positive SMB is not
applied.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@whlipscomb

Copy link
Copy Markdown
Contributor Author

Superseded by #97 (merged).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants