Labrador Sea ice runoff

[cecilehannay]

A series of coupled climate model runs (with beta05) showed unexpected differences in frozen ice runoff being routed from land to ocean via the MOSART river model. Some runs have near-zero ice runoff, while others have more substantial runoff.
This raised questions about the model’s configuration and the reliability of the results. These differences were also examined in whether the Labrador Sea froze early in some simulations.

There was a bug in beta05 (runs 122-129) that the glacier-to-runoff ice flux wasn't connected properly, which meant that all frozen runoff – positive or negative – that was supposed to be routed clm -> dglc -> mosart -> mom was being zeroed out in the dglc -> mosart step.

The treatment of snow capping is complicated. I believe that, in current configurations, snow capping over Greenland (positive SMB) is sent to DGLC, then routed as ice to MOSART; snow capping over Antarctica is sent as ice to MOSART (because we currently aren't using DGLC over Antarctica); and snow capping elsewhere is first melted and then sent as liquid to MOSART.

Once we implemented the DGLC model, this problem of what to do with snow capped ice actually became much more straightforward. Instead of a complicated reservoir system, we simply spread around positive mass flux to "fill the melted holes" in the ice sheet, a bit like a hole-filling algorithm for, say, water vapor in the atmospheric column. The end result of this is much less (mostly zero) negative water fluxes from the ice sheet and less positive fluxes during melt years. This was all implemented with the DGLC model last summer, so every simulation since, about September 2024, should have included this code. And I don't think it's been changed at all since then.

@wwieder @klindsay28 @olyson @gustavo-marques @dlawrenncar @slevis-lmwg @billsacks @Katetc

[dlawrenncar]

There was discussion this week about the Precip biases in the high latitudes, particularly Greenland / Canadian Arctic and whether or not this might be contributing to the Lab sea freeze. A positive P bias will generally lead to stronger runoff (ice or liquid). Here are land diagnostics plots for 20yrs averages from PI control runs for CESM2 vs a recent CESM3 run (127, high lat precip isn't changing a lot across recent runs, I believe). There does seem to be more precip and runoff over Greenland and more runoff over the Canadian Arctic, which I don't really understand since P is similar and LH went up

[adamrher]

@dlawrenncar we're currently testing artificially enhancing the sub-grid roughness over Greenland in an F-case, as was done in CESM2. This should lower the excessive precip rates over Greenland, as your plots show. I just finished making this modified topo file and so we might have some results in time for the Friday CAM7 meeting.

According to your plots, the Canadian Arctic biases are a somewhat different than Greenland. The precip rates are not that different from CESM2, but the runoff is much greater. The summer temperatures are cooler, and snow depths higher, suggesting that we may need to find a way to lock more water up in snow rather than change the precip rates.

[duvivier]

The precip bias would also lead to generally a fresher ocean surface regardless of the runoff aspect, and the fresher surface will freeze more easily.

[dlawrenncar]

@olyson Could you try to take a closer look to see if you can understand how in the Canadian Arctic that P is about the same between CESM2 and CESM3, but both runoff and LH/ET are higher in CESM3?

[olyson]

Sure.

[olyson]

This older version of the diagnostics package uses QOVER+QSURF+QRGWL for total runoff which evidently isn't consistent across model versions. The newer CESM_postprocessing version of the package had been updated to use QRUNOFF. I've switched to QRUNOFF for this comparison and the runoff is more comparable between model versions in an annual mean sense for Canadian Arctic. And annual mean P-E=R seems it would be more reasonable, just eyeballing it. But, CESM3 has much more runoff in June and much less in August than CESM2. I'll try making some plots of the runoff components.

[olyson]

Ok, I think I have the runoff components sorted. Below is a plot of runoff, runoff components, etc. Runoff = Surface Runoff + Sub-surface Runoff + QRGWL + Perched WT Drain (balances within E-9 mm/day). Note that in CESM2, the surface water runoff (QH2OSFC) was not included in the surface runoff variable (QOVER), while it is included in CESM3. So I've adjusted for that and in both cases Surface Runoff includes QH2OSFC.
So the differences in Runoff in June and August appear to be mainly caused by Surface Runoff which is partly caused by differences in QH2OSFC.
I'm not familiar with the surface water parameterization, so not sure why it is so different, maybe just snow melting behavior?

[swrneale]

Just looking back to CESM2 with the lab. sea freeze (the second one!) for context
Two things:
*All the patterns of LHFLX,SHFLX,LWCF and sfcs. are too far north consistent with Isla's jet stream analysis.
*The biased-high LHFLX and long-wave cloud forcing may also point to local cloudiness and excessive surface drag as issues.

[duvivier]

I find it's interesting that the LW Cloud flux and stress is not super different to the Southeast of Greenland in both simulations and is fairly different from the observations. I guess maybe in all CESM2 runs, even those that don't freeze over, there are some biases in these fields and it's more luck/chance that determines a freeze up, not these fields in particular? (I realize the freezing is happening more on the Southwest side of Greenland, but the sea ice does go towards the Southeast side and I still don't see a major change in the LW or stress fields).

[islasimpson]

Here are some maps of precip. Overall, I don't think the biases around Greenland are worse than what we had in CESM2. CESM2 definitely looked worse over the ocean around the southern tip of Greenland than the development runs do. In DJF, the precip biases (more precip) look worse over the land on the eastern edge of the Southern tip of Greenland but then CESM2 was worse over the ocean. In JJA I think we do have worse (positive) biases over North West Greenland. The nudging doesn't really improve much of the precip biases that we see, which I'm a little surprised about. I did go back and double check that the winds are being constrained, and they are. So, I don't think the nudging run is going to be that much help for this problem.

Annual mean:

DJF:

JJA:

[wwieder]

@adamrher has been able to reduce Greenland precip biases, but these changes also accelerate melt. Here I wanted to note that since CLM5 we made a number of changes to snow parameter values to melt snow faster and rescue Arctic vegetation from persistent snowpack that go beyond upplim_destruct_metamorph.

You can find these on lmwg_dev under the Dead Arctic Veg topic. Nearly all of these runs are I cases. I'll try summarizing these here, since reverting these changes would likely increase albedo & slow melt.

• upplim_destruct_metamorph higher values increase albedo and slow snow melt = 100 in CLM4 -> 175 in CLM5 & CLM6, see also lmwg_dev_#22

• xdrdt A5; higher values decrease albedo and accelerate snow melt = 1 in CLM5 -> 5 in CLM6 (see also lmwg_dev_#42.

---

• snicar_snobc_intmix blk, orig set to .false. in SNICAR PR -> CLM6 = .true.
• scvng_fct_mlt_sf B orig SNICAR PR = 1 ->CLM6 = 0.5
• snw_rds_refrz C orig SNICAR PR = 1000 -> CLM6 = 1500
• fresh_snw_rds_max D SNICAR PR = 204 -> CLM6 = 400
  Together blk_BCD can be seen in lmwg_dev_#37., more details & discussion from point runs are available in lmwg_dev_#36

---

• snw_rds_min E: had little effect, changed to 100 lmwg_dev_#41 and dropped from future tests

---

Finally we do have an old F2000 case from BEFORE we started mucking around with all these snow parameter in lmwg_dev_#13. Diagnostics show the following for Greenland temperature, runoff & precip.

[wwieder]

@adamrher you can add the following changes to user_nl_clm to revert snow albedo behaviors as listed above
paramfile = '/glade/derecho/scratch/wwieder/param_mods/ctsm60_params_revertSnow.c241119.nc'
snicar_snobc_intmix = .false.

specifically the following values are used on the modified parameter file:

• scvng_fct_mlt_sf=1.;
• snw_rds_refrz=1000.;
• fresh_snw_rds_max=203.;
• xdrdt=1

[olyson]

@wwieder , I was also researching snow albedo changes and I see that snow5d_thresh_for_onset = 0.2 in the parameter file. This used to be hardcoded and I believe the value at that time was 0.1

[wwieder]

Good catch, @olyson, I was considering that one of the plant / phenology changes that we included, but should have a relatively minor impact on snow albedo (especially in non-vegetated regions).

[olyson]

Got it, sounds reasonable.
Are we reverting the Sturm1997 parameterization in this experiment?

[adamrher]

I plotted the two F2000 runs you pointed me to. It looks like the Sturm param does facilitate more melt (but that run still looks better then where we are at now).

I can do two runs, one with Sturm and one w/o (both using the new param file)?