Skip to content

Allow fractional ice_covered values in the lnd2glc downscaling - #720

Merged
billsacks merged 2 commits into
ESCOMP:mainfrom
whlipscomb:lipscomb/fractional_ice_covered
Nov 10, 2026
Merged

billsacks merged 2 commits into
ESCOMP:mainfrom
whlipscomb:lipscomb/fractional_ice_covered

Conversation

@whlipscomb

@whlipscomb whlipscomb commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Description of changes

Allow the ice_covered field received from glc (Sg_ice_covered) to take fractional values in [0,1] in the lnd -> glc downscaling (med_phases_prep_glc_map_lnd2glc).

Previously, each glc grid cell had to be exactly ice-covered (1) or ice-free (0); glc_get_elevation_classes_with_bareland aborts otherwise. Fields sent to glc were vertically interpolated between elevation classes for ice-covered cells and taken from the bare land class for ice-free cells. Now each field sent to glc, and each water tracer, is the weighted sum

ice_covered * (value interpolated between elevation classes) + (1 - ice_covered) * (bare land value)

For ice_covered = 0 or 1 this gives the same values as before. The vertical interpolation indices and weights are unchanged. The mediator no longer calls glc_get_elevation_classes; that routine is unchanged.

The CMEPS changes here, along with the companion changes in CISM and CISM-wrapper, resolve CMEPS issue #713.

Specific notes

Motivation: water conservation in CESM runs with an evolving Greenland ice sheet. With a binary ice_covered, a CISM cell holding only a trace of ice (a few mm or cm) is treated as fully ice-covered, so it receives a full year of low-elevation glacier SMB (ablation of ~1 m/yr or more). CTSM melts that much glacier ice (glacier_region_melt_behavior = 'replaced_by_ice') and sends the water to runoff, but CISM can melt only the trace of ice present. In 1850 I-compset runs (CLM6 BGC-CROP, DATM CPLHIST from a B run, evolving GrIS at 4 km), this leaves 36-62 Gt/yr of meltwater with no matching ice loss, about 3,400 cells are affected, and many of them switch between ice-covered and ice-free from year to year.

With a companion CISM change (ESCOMP/CISM branch lipscomb/smb_conservation, PR to follow), CISM can send ice_covered = min(1, thck/H0) (new option ice_covered_thck_scale in CISM's [GLAD climate] section; default 0 keeps the binary field). In 5-year runs with H0 = 2 m and 4 m:

  • the CISM-side imbalance falls to < 0.05 Gt/yr;
  • the gap between CTSM QICE over Greenland and the SMB applied by CISM falls from 34-75 Gt/yr to 1.5-8 Gt/yr (the remainder is bare-land SMB, which QICE omits, plus CTSM glacier outside CISM's ice mask, which CTSM handles conservatively; not a conservation error);
  • the number of cells switching between ice-covered and ice-free each year falls from ~300-450 to ~10-50.

Other parts of the coupling already handle fractional ice cover:

  • the land-side elevation-class fractions (glc_get_fractional_icecov) set bare land to 1 - ice_covered, so CTSM's glacier areas follow the fraction;
  • the SMB renormalization still makes the global glc integral equal the land integral. (Accumulation and ablation are summed by the sign of each cell's net SMB, so the two separate factors can move slightly further from 1, but the total is conserved.)

A range check stops the run with an error if ice_covered is outside [0,1].

Write-up with budgets and results here:

Contributors other than yourself, if any: Bill Sacks (lnd2glc remapping analysis); Gustavo Marques, Kate Thayer-Calder, Brian Dobbins and Gunter Leguy (coupled water budget diagnosis).

CMEPS Issues Fixed (include github issue #): none

Are changes expected to change answers? (specify if bfb, different at roundoff, more substantial)
Bit-for-bit when glc sends ice_covered = 0 or 1, which is the current behavior of CISM and the default of the new CISM option. Answers change only when glc sends fractional values.

Any User Interface Changes (namelist or namelist defaults changes)?
None in CMEPS. Fractional values are enabled by a CISM option (ice_covered_thck_scale, CISM-wrapper namelist), in separate CISM and CISM-wrapper PRs.

Testing performed

All on derecho.

  • Bit-for-bit, binary ice_covered, in the target configuration. 1-year I-compset run (1850_DATM%CPLHIST_CLM60%BGC-CROP_SICE_SOCN_MOSART_CISM2% GRIS-EVOLVE, a%ne30np4.pg3_l%ne30np4.pg3_oi%null_r%r05_g%gris4, cesm3_0_alpha11d) with this change and ice_covered_thck_scale = 0, compared with the same case built from cmeps1.1.73: the coupler lnd2glc history (37 fields), glc export history (6 fields) and CISM end-of-year state (41 fields) are identical.
  • CISM aux_glc tests, compared to baselines, all PASS and bit-for-bit (this change applied to the CMEPS version used by the baselines):
    • SMS_D_Ly1.f09_g17_gris4.T1850Gg.derecho_intel
    • SMS_D_Ly1.f09_g17_gris4.T1850Gg.derecho_gnu
    • ERS_Ly5.f09_g17_ais8.T1850Ga.derecho_intel
    • ERI_Ly9.f09_g17_ais8gris4.T1850Gag.derecho_gnu.cism-noevolve_ais
    • ERS_Ly4.f10_f10_ais8gris4_mg37.I1850Clm50SpGag.derecho_intel
    • SMS_D_Ld5.f10_f10_ais8gris4_mg37.I1850Clm50SpGag.derecho_gnu.cism-test_coupling
    • ERS_D_Ld9.f10_f10_mg37.I1850Clm50SpG.derecho_gnu.cism-override_glc_frac
  • Fractional ice_covered. 5-year runs of the I case above with ice_covered_thck_scale = 1, 2 and 4 m (results above), and 20-year runs (years 53-72) with 2 m and 4 m and a control. Over 20 years the CISM-side imbalance stays at 36-53 Gt/yr in the control and < 0.2 Gt/yr with either H0; cells switching between ice-covered and ice-free fall from ~300/yr to ~20/yr (4 m); the ice area thicker than 10 m and the interior thickness are unchanged. We plan to make 4 m the CESM default (in CISM-wrapper) once this CMEPS change is in a CESM tag. Tests with fractional ice_covered (CISM testmod cism/fractional_ice_covered, ice_covered_thck_scale = 4.0), all PASS including the exact-restart comparisons:
    • ERS_Ly5.f09_g17_gris4.T1850Gg.derecho_intel.cism-fractional_ice_covered
    • ERS_Ly4.f10_f10_ais8gris4_mg37.I1850Clm50SpGag.derecho_intel.cism-fractional_ice_covered
    • SMS_D_Ly1.f09_g17_ais8.T1850Ga.derecho_gnu.cism-fractional_ice_covered

Hashes: CMEPS f88417f (this branch); CISM ESCOMP/CISM lipscomb/smb_conservation (028d39f0); CISM-wrapper lipscomb/smb_conservation.

Previously, the lnd -> glc downscaling required the ice_covered field
received from glc to be exactly 0 or 1 in each glc grid cell
(glc_get_elevation_classes_with_bareland aborts otherwise). Fields sent
to glc were vertically interpolated between elevation classes for
ice-covered cells, and taken from the bare land class for ice-free cells.

Now ice_covered can be any value in [0,1]. Each field sent to glc, and
each water tracer, is the weighted sum
  ice_covered * (value interpolated between elevation classes)
  + (1 - ice_covered) * (bare land value).
For ice_covered = 0 or 1, this gives the same values as before.

The motivation is water conservation in runs with an evolving ice sheet.
With a binary ice_covered field, a glc cell with a trace of ice (a few mm
or cm) receives a full year of low-elevation glacier ablation, most of
which the ice sheet cannot apply, while the land model sends the full
meltwater to runoff. In CESM runs with an evolving Greenland ice sheet,
this leaves about 40-60 Gt/yr of meltwater with no matching ice loss.
CISM can now send a fractional ice_covered field (ice_covered_thck_scale
> 0 in the [GLAD climate] section), which removes this imbalance.

The land-side fractions already allow fractional ice cover
(glc_get_fractional_icecov), and the SMB renormalization conserves the
global total with fractional values, so no other changes are needed.

Answers are unchanged when glc sends ice_covered = 0 or 1.

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

@billsacks billsacks left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks a lot for your quick work on this, @whlipscomb !

I have given this quite a bit of thought, and am happy with this change. I just have one request to revert the change to the comment in glc_elevclass_mod - see my rationale in my line comment below.

Some things I want to do before final approval:

  • Run full derecho aux_glc testing with this change, ensuring bit-for-bit results (I'll do this) (my impression is that you only ran a subset of testing)
  • Do a careful look for uses of the ice-covered fraction elsewhere in CMEPS to see if there are any implicit assumptions of 0/1 values - unless you have already done so

I also have some bigger-picture questions (which won't hold up this PR):

  • Does the new CISM use of fractional ice coverage often prevent fully melting the ice in a CISM cell? If so, is this a problem?
  • The land ice group should consider whether you feel it's important for the glacier areas in DGLC to match the areas when using CISM - i.e., if the fractional logic you're introducing in CISM should also be put in DGLC (pinging @Katetc on this one)

Comment on lines +279 to +282
! Note: This subroutine assigns a single elevation class to each point, so it
! cannot handle a continuous fraction between 0 and 1. The CMEPS lnd -> glc
! downscaling (med_phases_prep_glc_mod) no longer calls this subroutine; it
! accepts fractional ice_covered values directly.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you back out this comment in glc_elevclass_mod? I'm in the midst of moving this module, so this change will conflict (#721 and ESCOMP/EPS_Share#88). And once we bring in these changes to CMEPS and CISM, I think the right approach is to just go ahead and delete this now-unused routine so that nobody tries to use it by accident.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@billsacks: Yes, I will back out this comment in a small follow-up commit. I agree that a full aux_glc test is a good idea.

I relied on Claude to check for other uses of the ice_covered fraction. Claude didn't find any, but I think it would be good for you to confirm this.

Once I've done that small commit, I'll come back to your big-picture questions.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@billsacks: I just made the requested change and pushed the commit to my fork.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My answer to your first question above: Yes, to some extent, but I don't think this is a problem.

I’ll give some numbers from Claude. In 20-year Greenland runs with H0 = 4 m, about 1,750 more margin cells hold thin ice (0 < H < 4 m) than in the control, totalling about 20 km³ of ice (negligible compared to the total GrIS volume). These cells are only partly ice-covered from CTSM's point of view: the glacier area CTSM sees is about 1.8% smaller than in the control, because the binary scheme counted every cell with a trace of ice as fully glaciated.

Here’s a numerical example: Consider a CISM marginal cell with H = 1 m, whose surface elevation corresponds to the lowest glacier EC. Let H0 = 4 m, giving ice_covered = 0.25. Suppose CTSM computes an SMB of -2 m/yr for the EC and 0 for bare land. In that case, CISM receives an area-weighted SMB of -0.5 m/yr, causing the cell to thin to 0.5 m over the coupling interval of 1 yr. Suppose advection supplies enough ice to increase H by 0.5 m, balancing the ablation. Then, assuming steady ice flow in CISM and a steady climate in CTSM, the cell will maintain H = 1 m and keep passing ice_covered = 0.25. If the advection were reduced, then H and ice_covered would decrease gradually toward a new steady state instead of going straight to zero.

In summary, I see the change as a physically sensible sub-grid parameterization of the ice margin. CTSM sees a more realistic ice area; water is conserved; and flipping between ice-covered and ice-free cells is greatly reduced.

For the second question: My first instinct is to say yes. DGLC currently computes ice_covered from the dataset thickness with a threshold of zero, so any cell with H > 0 gets ice_covered = 1. With the fractional logic, a margin cell with H = 1 m would instead pass ice_covered = 0.25, and CTSM would pass back a weighted average of its bare-land SMB and its SMB in the lowest glacier elevation class, giving a more positive SMB than now. This seems to me like the right behavior, and it would keep DGLC runs consistent with runs coupled to CISM.

For our current Greenland dataset (greenland_4km_epsg3413_c251001.nc) the effect is much smaller than with an evolving ice sheet. By Claude’s count, the dataset has only 33 cells with H < 1 cm (vs ~2,000 in evolving runs), though about 2,800 cells have H < 4 m. Fractional cover with H0 = 4 m would reduce the glacier area CTSM sees by about 1.8%. A rough estimate suggests a few Gt/yr less ablation, not 50 Gt/yr as in the coupled case. (We could test this.) In DGLC runs this is a consistency question rather than a conservation question, since DGLC handles the SMB itself.

@Katetc: What would you like to add?

@Katetc Katetc Nov 9, 2026 •

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, Gustavo looked at cases using DGLC and found no conservation issues in those. It also helps that the smb given to DGLC is never "applied". It is adjusted to reduce negative smb with the ice-sheet wide hole-filling and then sent straight to ROF. I'm dis-inclined to add the fractional scaling at this moment as it seems unnecessarily complex for what DGLC does currently. If you want to start an issue around that and tag/assign me, that is fine though. We can think about that step more later.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Katetc, I agree that fractional scaling is not a high priority at this moment. I was thinking from an unhurried long-term point of view.

Restore cesm/nuopc_cap_share/glc_elevclass_mod.F90 to its version on
main, to avoid a conflict with CMEPS PR ESCOMP#721. The previous commit only
updated a comment in this file. The code changes in
med_phases_prep_glc_mod.F90 are unchanged.

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

Katetc commented Nov 9, 2026

Copy link
Copy Markdown
Contributor

I'll do the code review for this PR this evening, but everything I've read about it seems like a nice, easy solution to the problem. Nice work! I have two questions. First, would you update the description of this pr to mention that it does address CMEPS issue #713 , that way we can close that issue once this is resolved.
Secondly, how does this change affect Gunter's spin-up workflow for CESM3? I know that we need to merge 5 PR's before we worry about that, but it seems like we would want to spin-up Greenland with this slightly different climate (from the ice sheet's perspective). I know that Gunter uses some custom scripts to convert Cplhist smb to CISM-standalone SMB, and I'm just curious if this would throw any wrenches in his workflow.

@Katetc Katetc left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the code changes are targeted and safe.

@Katetc

Katetc commented Nov 10, 2026

Copy link
Copy Markdown
Contributor

Also, after looking at the code and the history of the file a little, we will probably want to talk to our NorESM collaborators about this. The changes are backwards compatible so this PR shouldn't impact them. But they will probably be interested in why we are making this change.

@whlipscomb

Copy link
Copy Markdown
Contributor Author

@Katetc, you asked a good question about the spin-up workflow for CESM3. I'm not sure that a new CISM spin-up is needed. With the fix, CTSM will pass similar values of QICE for bare-land and each EC. These values will be weighted differently for partially ice-covered cells, leading to changes in ice-covered area at the GrIS margin, but I don't expect the margin to advance or retreat significantly. Also, the area-integrated ice volume and acab_applied_tavg should be about the same. As stated in my email to Dave Lawrence, I think the main changes will be in the ocean, where we prevent spurious freshening.

The other consideration is where we are in the CESM spin-up process. I suspect that a new GrIS spin-up would require a new CTSM spin-up also. This combination might take more time than we can spare right now.

I'll ask @gunterl to comment as well.

@gunterl

gunterl commented Nov 10, 2026 via email

Copy link
Copy Markdown

@billsacks billsacks left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm happy with this now. This does seem like a very good fix for this problem!

I have looked through the relevant CMEPS code and don't see anywhere else where there's an assumption that ice-covered is 0 or 1.

My test jobs are stuck in the queue. Hopefully they'll run overnight. Once testing is good, I'll merge this.

@billsacks

Copy link
Copy Markdown
Member

@whlipscomb and @Katetc - I also agree that, for DGLC, this would be about consistency, not conservation. I opened an issue to track this (ESCOMP/CDEPS#458).

@billsacks

Copy link
Copy Markdown
Member

I have run the full derecho aux_glc testing with comparisons against baselines, from cesm3_0_alpha11d. All tests passed and were bit-for-bit, except for the expected failure of NCK_Ly3.f09_g17_gris20.T1850Gg.derecho_gnu COMPARE_base_multiinst.

Also, to test the water tracer code, I ran SMS_D_Ld2.f19_g17.X.derecho_intel.drv-water_tracers--drv-glc_avg_frequently. This also passed and was bit-for-bit.

@billsacks
billsacks merged commit 934d67a into ESCOMP:main Nov 10, 2026
1 check passed
@whlipscomb

Copy link
Copy Markdown
Contributor Author

@billsacks: Thanks for the quick turnaround on the review.

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.

5 participants