Ice runoff changes when coupling an evolving Greenland ice sheet

[billsacks]

I have been talking with @gustavo-marques @cecilehannay @wwieder @dlawrenncar, @gunterl and @Katetc over the past couple of weeks about changes in ice runoff around Greenland when changing from DGLC to CISM with an evolving Greenland ice sheet. I am starting this discussion to record some findings and other general notes about that. (@whlipscomb - also keeping you in the loop here.)

This is connected to run 382, run 384, and some subsequent runs.

@gustavo-marques 's finding was that, when switching from DGLC to CISM, the ice runoff from CTSM increased dramatically. We dug into trying to understand the cause of this increase and what could be done to mitigate it.

First, let's take a moment to understand the major sources of ice runoff over Greenland, both with DGLC and with CISM. With the caveat that I may be forgetting something, here are the major sources I can think of:

• With DGLC, the main source of ice runoff (positive or negative) is from the surface mass balance (SMB, also known as qice). There is an implicit assumption that the ice sheet is in equilibrium with DGLC, so all SMB goes to ice runoff. This ice runoff term goes through the GLC component. Negative SMB offsets positive SMB, both temporally (we work with annual totals) and spatially (positive values offset negative values across the ice sheet), so that, typically, there will only be a positive flux (unless the ice sheet as a whole is experiencing a negative SMB). (Note that negative SMB is also associated with a liquid runoff - i.e., melt - flux from CTSM.)
• With the ice sheet dynamics of CISM, the SMB flux is treated more realistically: Positive SMB creates more ice in CISM; negative SMB removes ice. Neither of these directly leads to ice runoff. Instead, there is a calving flux generated in CISM when ice reaches the ocean. (Negative SMB is also associated with a liquid runoff - i.e., melt - flux from CTSM.)
• With an evolving ice sheet in CISM, there is also a new term: A dynamic landunit conservation correction term in CTSM. This arises because, when portions of a land grid cell transition from vegetated / bare land to glacier (or vice versa) there are changes in the water and energy state that are very challenging for CTSM to reconcile locally. So, to conserve water and energy, CTSM does an accounting before and after these landunit transitions, then ensures system-wide conservation by generating liquid and ice runoff terms and a sensible heat flux, all of which can be either positive or negative. (On the history files, these are QFLX_LIQ_DYNBAL, QFLX_ICE_DYNBAL, and EFLX_DYNBAL.)

Through some investigation, we determined that the main source of the fluxes @gustavo-marques found was this last term - the dynamic landunit conservation correction, QFLX_ICE_DYNBAL. From spot-checking some points where this term is large, I determined that the main source of this term is from differences in the snow pack between CTSM's vegetated / bare land column and its glacier columns. Therefore, when the ice sheet advances or retreats, the grid cell average snow pack water content changes, and QFLX_ICE_DYNBAL compensates for this change to conserve water.

(More details: In many cases, the snow pack in a given CTSM column is either near its maximum value (10 m water equivalent) or near 0 (less than about 1 m water equivalent). Differences in snow pack depth between columns largely arise from the differences in elevations of these different columns; glacier columns can have either more or less snow pack than vegetated columns. In some spot-checked grid cells, I was able to reproduce the QFLX_ICE_DYNBAL flux pretty closely by just looking at the difference in average H2OSNO before and after the ice sheet area change at the year boundary, and then calculating the annual-average flux that would be needed to balance this change, as is done in CTSM.)

The consistently positive ice runoff fluxes observed by @gustavo-marques arise from the interaction between three behaviors in the model:

• (I haven't investigated this one carefully, so am going based on indirect evidence plus what I remember happened in CESM2.) Around the margins of the ice sheet, there is an oscillatory behavior where the ice sheet blinks in and out of existence on a 2-year cycle. I think this arises because: In years where there is no ice sheet in a given land grid cell, there is no capacity for the land model to generate negative surface mass balance. The ice sheet can then freely advance into that grid cell without being melted. Once it advances, however, in the next year the land model does generate a net-negative surface mass balance. This negative surface mass balance then melts the ice in that grid cell to a greater extent than the inflow of ice, resulting in zero ice in the following year.
• In these grid cells at the margin, we therefore get an annual oscillatory behavior of transitions between vegetated / bare columns and glacier columns. These transitions lead to changes in grid cell average snow pack (as described above), which in turn leads to compensatory QFLX_ICE_DYNBAL ice runoff fluxes.
• When averaged either temporally (across multiple years) for a given grid cell, or spatially (across all of Greenland) for a given year, I expect the net QFLX_ICE_DYNBAL flux to be near zero. However, we currently treat negative fluxes specially in CMEPS: In order to avoid sending negative fluxes to the ocean, we distribute these negative fluxes globally, reducing all positive values by some fraction in order to maintain the correct global sum while zeroing out negative values. This means that positive ice runoff lands regionally, whereas negative ice runoff is distributed globally. This asymmetry leads to the net-positive ice runoff fluxes @gustavo-marques observed.

We considered a few solutions or workarounds for this:

1. Completely zeroing these DYNBAL fluxes (run 384). However, this breaks conservation. While this loss of conservation is perhaps justifiable in early stages of the spinup (where we will be breaking conservation later anyway when we do offline spinups of the land and ocean components), we feel that this would be problematic in a transient scenario. And we don't want behavior to abruptly change when going from the spinup to the transient case, so we ruled this out as an option.
2. Setting CTSM's reset_dynbal_baselines flag to true. This flag resets what is considered the zero point for some pieces of the DYNBAL flux calculations. It turns out, though, that this doesn't help with the ice runoff problem discussed here, because the snow pack is deliberately excluded from these baseline calculations. Nevertheless, I still recommend doing this at one or more points along the spinup to reduce the sensible heat flux generated from these landunit transitions. Without setting this flag, we can get dynbal sensible heat fluxes up to 10s of W m-2 in select grid cells, persistent across the year; based on some quick looks, we can reduce these dynbal sensible heat fluxes by about a factor of 10. Setting this flag to true can be done at the start of a hybrid run; doing so breaks conservation at the beginning of that run, but then conservation is maintained from that point forward. There is more documentation of this flag here. It takes some thought to determine when you do / don't want to set this flag, but it feels safe to set it at this relatively early stage of spinup, when we're still going to have some conservation breakages later anyway when we spin up the surface components offline.
3. Turning off the zeroing / global spreading of negative ice runoff fluxes. This means that we will get some negative ice runoff fluxes in the ocean, which in some senses is undesirable, but the benefit is that the regional average ice runoff flux will be around zero, as it should be. We are exploring this in some recent simulations and I plan to add a namelist flag to CMEPS to facilitate this.

Longer-term, I have some ideas about how we could further reduce the DYNBAL fluxes from CTSM, and I will open a CTSM issue about that.

[micpetrini]

Hi @billsacks, @mvertens pointed me to this issue. Thanks for the detailed explanation and deep analysis. I am looking at the integrated ice runoff (variable QGLC_ICE_INPUT 'MOSART input ice runoff from glacier') over Greenland-ish* (*masking for lat-lon, as I don't have a Greenland mask on the 0.5 degree MOSART grid) for two identical NorESM runs, one with DGLC and one with evolving GrIS. We also see a very different behaviour, with values in the GrIS%EVOLVE case on the very high end than those found in the DGLC case, see screenshots below (the recurring zero values in the DGLC case is an issue we are looking at here, but not relevant for this discussion I think). Do you see an ice runoff increase of comparable magnitude in CESM, or is it way more?

In our GrIS%EVOLVE case we have physical calving, frontal ablation and other artificial quantities such as remove_ice_caps and force_retreat (not sure you are using that, we use it to mask out ice outside the present-day GrIS extent) contributing to ice_runoff. So even though I haven't yet tried to compute the actual budget (probably very complicated, but I'll make an attempt), qualitatively the behaviour I am seeing in our GrIS%EVOLVE makes sense, as we know that this budget is very likely dominated by force_retreat, which is dumping ice year after year around the margins (see third screenshot below) and don't have much of a interannual variability signal. If the signal is dominated by QFLX_ICE_DYNBAL fluxes, am I wrong assuming that we would see much more interannual variability in our curve?

DGLC

GrIS%EVOLVE

Rate of mass loss by forced retreat

[billsacks]

@micpetrini - thanks for sharing your findings. My understanding is that you're just looking at the ice runoff (calving) flux from GLC. Note that QFLX_ICE_DYNBAL comes from CTSM, and so wouldn't show up in that variable. You instead would need to look at the ice runoff flux from LND.

As for differences in the GLC flux between DGLC and CISM, I'll need to let @gunterl or @gustavo-marques comment on that. My understanding is that this flux wasn't hugely different between the two in CESM3 cases, but I only looked very briefly at a plot that @gustavo-marques showed, so I may be wrong.

[micpetrini]

thanks @billsacks for the reply. I cannot find in MOSART something like QLND_ICE_INPUT, could you please tell me which variables you are looking at in (I presume) CLM? I've just made the same timeseries analysis for frozen runoff as output from the ocean model (BLOM) for the two runs with DGLC and GrIS%EVOLVE; values don't match those in MOSART (they are two times higher, I need to check that I'm doing sums/weighting correctly), but in principle I see the same thing: values in the GrIS%EVOLVE case (RHS) on the very high end than those found in the DGLC case (LHS), see screenshots below.

I would like now to look at the same variables you are looking at + trying to estimate fluxes from CISM which would contribute to frozen runoff, to see if I find any match.

[billsacks]

@micpetrini - the analysis I mentioned from @gustavo-marques was looking at variables in MOM6. I'm not familiar enough with the MOSART history output to know what variables you could look at there. In CLM, you could look at QRUNOFF_ICE_TO_COUPLER.

I'm not surprised that you're seeing differences in ice runoff (calving) between a CISM case and a DGLC case. I would need to get my head back into this to be able to comment intelligently on this, so I may be making some mistakes in my thinking here, but my initial thought is: DGLC implicitly assumes a quasi-equilibrium state of the ice sheet, where annual ice runoff from DGLC is based on the net SMB over the ice sheet. Assuming net SMB is about the same in the CISM case as the DGLC case (which I hope is true!), I think your CISM plot is showing that you're getting calving fluxes that are a few times higher than the average net SMB. To me that indicates that the ice sheet is out of balance with the modeled climate. The mild downward trend in the CISM plot seems to support this - though it looks like it will take at least a few centuries for it to come into balance at this rate.

[fobryan3]

From Bill's original post:
"This means that we will get some negative ice runoff fluxes in the ocean, which in some senses is undesirable, ..."

It is not "in some sense undesirable", it is absolutely unacceptable. While I don't find the opaque explanation above helpful in understanding why the formulation of the CTSM/glacier model combo requires negative liquid or frozen runoff from another component model, the idea that the ocean should spontaneously create icebergs that then crawl up onto land preposterous and puts us at risk of having CESM3 exposed and labeled as an non-physical climate model by bad actors.

This is not just an opinion about physical plausibility. The consequences of this negative runoff error - liquid or frozen - are far more severe in MOM6 than they were in POP. Removing fresh water from MOM6 changes the concentration of EVERY ocean tracer - the volume of water changes but the mass of tracer does not. In POP we did not change the actual volume of water, we changed only the salinity when adding/subtracting fresh water. Asking the ocean to provide water to the land model will change the concentration of zooplankton or dissolved CO2 for example. Do we really want to see phytoplankton blooms around Greenland when you need some snow? Asking for a phase change in the water additionally changes the ocean heat content and temperature even further.

Any fixer required to offset inconsistencies or gaps in formulation within the land/glacier model combo should be offset within those models.

[billsacks]

@fobryan3 - I appreciate your sharing your perspectives here and reminding me / us of some of the issues with negative runoff.

I have been thinking of some possible alternative solutions on the land side - see ESCOMP/CTSM#4188. I will start a conversation with the original group about how we want to move forward given the time crunch for CESM3.