Why does reduced vertical resolution decrease DSW formation?

[adele-morrison]

From our discussion on Friday: We see much reduced DSW formation in the case with a 5m thick top level vs the 1m thick top level. However, we also change resolution over the whole top 500m in the thicker level case, so it's hard to pin down exactly why we're seeing the change. One hypothesis is that the increase in salinity from sea ice formation initially occurs in the top level (before convection takes place and mixes it downward). So in the case with the 5m top level, the surface salinity is not as salty as in the 1m case. This could then lead to changes in convection driven by the different stratification in the two cases. The SWMT metric will also bin the same sea ice buoyancy fluxes into different density classes in the two cases (with the 5m case having more surface area with lower surface density?).

We could potentially test this idea by running another simulation where we only change the top level thickness. Maybe one day when we have more compute available (unless Paul you have excess compute this quarter?)...

[fabiobdias]

I've just plotted the first 4 years of sea-ice thickness maps (monthly averages) from the DSW control (01deg_jra55v13_ryf9091_DSW) and the 5m thick top layer (01deg_jra55v13_ryf9091_5mdz) experiments against the original 01deg_jra55v13_ryf9091.

It seems the 5mdz experiment (lower right panel shows 5mdz minus control) has substantially more sea ice near the "coast" in the Weddell, which I guess indicates smaller coastal polynyas and subsequently less salt input? Also, it looks like the Weddell is affected all year-long while other regions (mainly Ross-West Antarctica sector) differ mostly during winter.

One hypothesis could be that a less salty surface layer in 5mdz facilitates sea-ice formation to the point coastal polynyas are less frequent in the Weddell Sea.

https://github.com/willaguiar/DSW-collaborative-project/assets/8631257/3b68ca50-6171-4ea7-aa4a-bac95d384347

[adele-morrison]

@willaguiar just a note that it would be good to post the new results with only the top layer changed in this issue.

Today Will and I discussed putting on a new run to chew up some of our excess compute this quarter. This would be identical to the control, but adding in 1 extra vertical layer by dividing the top cell (~1m thick) into two layers of ~0.5m thickness. i.e. test if the sensitivity to top layer thickness is linear.

I was also wondering how sensitive our vertical resolution results are to the choice of the KPP mixing scheme. KPP has been shown to be sensitive to vertical resolution. Would it be worth seeing if our vertical resolution results are robust if we use the ePBL scheme in panan-01?

[wghuneke]

I'm in favour of a test for the mixing scheme, as mentioned in my feedback from the COSIMA poster session. @adele157, are you suggesting to repeat the vertical resolution test in PanAnt (i.e. adding a second model to our analysis) or stay with OM2-01, but add sensitivity tests by changing the parameters in the mixing parameterisation?

[adele-morrison]

Yes I was thinking to repeat the top cell thickness tests in panan with KPP and ePBL (so 4 runs in panan). I don’t think we can do ePBL in ACCESS-OM2? But that’s a lot of new runs, so perhaps we should discuss in person whether it’s worth it. We have plenty of compute, it’s more the person time it would take to set them up and run them.

On Wed, Sep 13, 2023 at 8:00 AM, Wilma Huneke @.***> wrote:

I'm in favour of a test for the mixing scheme, as mentioned in my feedback https://github.com/willaguiar/DSW-collaborative-project/issues/23 from the COSIMA poster session. @adele157 https://github.com/adele157, are you suggesting to repeat the vertical resolution test in PanAnt (i.e. adding a second model to our analysis) or stay with OM2-01, but add sensitivity tests by changing the parameters in the mixing parameterisation?

— Reply to this email directly, view it on GitHub https://github.com/willaguiar/DSW-collaborative-project/issues/20#issuecomment-1716549319, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACA44U6Z3U2LRQZGK56HUODX2DLP7ANCNFSM6AAAAAAYBQNBEE . You are receiving this because you were mentioned.Message ID: @.***>

[willaguiar]

Summary of 5mtop simulation vertical grid: Previously, we were discussing why reducing the vertical resolution alters DSW formation. Hypothesis would be that [a] a thicker surface cell (5m thick on 5mdz instead of 1m on ctrl) requires stronger salt fluxes to form DSW, or [b] a less refined vertical grid can alter mixing along dense overflow path, making waters too light to become DSW.

We ran an additional experiment (5mtop) in which the only change is in the top cell ( which has 5m thick). As you can see in the fig, below 5m depth, all centres, and edges of cells are the same between 5mtop and the control (21mbath)

SWMT on the shelf in 5mtop simulation: Comparying the SWMT on the shelf between the control case (21mbath), reduced vertical resolution case (5mdz) and reduced top cell (5mtop), we can see that 5mtop shows the same reduction in transformation into waters denser that 27.8 than 5mdz

So it seems that the thickness of the grid in the top 5m of the water column is what is altering DSW formation [a]

Simulations comparison by sector, shelf only: I additionally separated the SWMT by sectors, and compared the last year of all simulations. As you can see, it seems that in the 5mtop case we also have the whole SWMT curve moving towards lighter densities in all sectors. Specifically in Weddell sector, we have no waters crossing the 27.8 sigma0 level by the end of the simulation.

Additional note is that the SWMT for CM2atm moves to way more dense values in the Weddell sea (more on that on another issue).

We are currently running a case with refined top cell ( 0.5m instead of 1m) to see if this cell thickness control is proportional/linear. I am also interested in making the KPP and ePBL tests, so we can see if this sensitivity changes depending on the mixing scheme.

[willaguiar]

We now have a new simulation run that goes in the opposite direction of 5mtop, e.g., the top cell is reduced from 1m dz, to 0.5m dz. here is what we find for the new simulation

Results for halfmtop simulation

---

The case with thinner 0.5m top cell seem to have more transformation towards waters denser than σ₀=27.8 kg m^-3 (top black line dashed), when compared to the control (21mbath) case. The whole SWMT curve also seem to shift to denser values halfmtop simulation also has an increase in cross-slope transport (below) when compared to the control, and by the fine year of the simulation, the DSW export sits between the control case and the CM2atm case (right plot is zoomed into the peaks). Most of this increase in density in halfmtop experiment happens in Weddell and East Antarctic sectors: And according to the age at the bottom, this also results in increased DSW export in the Weddell Sea. The anomalies in age tracer in halfmtop look almost opposite to 5mtop: .

[adele-morrison]

Nice that it's consistent! Now just to figure out why.

On Thu, 19 Oct 2023 at 10:33, Wilton Aguiar @.***> wrote:

We now have a new simulation run that goes in the opposite direction of 5mtop, e.g., the top cell is reduced from 1m dz, to 0.5m dz. here is what we find for the new simulation

The case with thinner 0.5m top cell seem to have more transformation towards waters denser than σ0=27.8 kg m-3 (top black line dashed), when compared to the control (21mbath) case. The whole SWMT curve also seem to shift to denser values [image: SWMTsimulations_comp_circumpolar_yr1909_sigma0_total] https://user-images.githubusercontent.com/70033934/276425280-981e7b69-1d6a-49bd-84d7-b76ef04e58bb.png halfmtop simulation also has an increase in cross-slope transport (below) when compared to the control, and by the fine year of the simulation, the DSW export sits between the control case and the CM2atm case (right plot is zoomed into the peaks). [image: CSMT_yr1909_multiexperiment_sigma0_1000m] https://user-images.githubusercontent.com/70033934/276430672-69d3fb45-bace-4244-8b0d-13172a63a65d.png Most of this increase in density in halfmtop experiment happens in Weddell and East Antarctic sectors: [image: SWMT_by_sctor] https://github.com/willaguiar/DSW-collaborative-project/blob/main/figs/SWMTsimulations_comp_by_sector_sigma0_total.png?raw=true And according to the age at the bottom, this also results in increased DSW export in the Weddell Sea. The anomalies in age tracer in halfmtop look almost opposite to 5mtop: [image: ageob] https://github.com/willaguiar/DSW-collaborative-project/blob/main/figs/age_ontheshelf.png?raw=true .

— Reply to this email directly, view it on GitHub https://github.com/willaguiar/DSW-collaborative-project/issues/20#issuecomment-1769579013, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACA44U46MJQMTQSKJVFPTCTYABRM3AVCNFSM6AAAAAAYBQNBEGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRZGU3TSMBRGM . You are receiving this because you were mentioned.Message ID: @.***>

[willaguiar]

By now we have already ~7 years of run with MOM6 5mtop and MOM control. Here is what we find for these simulations

---

SWMT MOM6 DSW formation (Shelf WMT) does seem to decrease, but at a quarter of the magnitude of what we find for MOM5. Although the MOM6 5mtop is not completed, it seems that the SWMT is rather stable already. For the line plot in [e] I used the transformation averaged between 90% and 65% of the maximum transformation volume in the last year of each 1mtop simulation. The blue lines in a-d shows where this is located

---

DSW outflow The changes in DSW outflow are rather similar in magnitude between MOM5 and MOM6,i.e., about 2 SV decrease until 1996. However, the DSW outflow in [e] could be a little overestimated for OM2, because the density layer I used for the average sits right in that weird dip (~27.8) that happens in $\sigma_0$ but not in $\sigma_2$. (We had a little discussion on this before, and this southward flow could be because in $\sigma_0$ CDW can inflow a denser layers). So maybe what we measured for OM2 could be a mix of change in DSW outflow and CDW inflow.

---

DSW outflow changes per region Below is the cumsum of the DSW outflow on the 1km isobath in the last year of each simulation, for the density layers explained in the top fig. On MOM5 (OM2), we see a complete shutdown of DSW outflow in Prydz bay and Weddell, and a decrease in outflow in Adelie and Ross. In MOM6, DSW outflow is shutdown only Adelie and Prydz, while Ross and Weddell have a decrease in magnitude. It think the change in DSW outflow is still pretty big on MOM6, even tho the SWMT changes are smaller. So the decrease in DSW with top cell thickening in MOM6 still important!? (both what I think and a question)

[willaguiar]

I think this suggests that although KPP mixing scheme amplify the DSW sensitivity to the top cell thickness, the changes cannot be fully attributed to KPP. Perhaps this is enough motivation to run Panan/MOM6 with KPP and different top cell thickness?

[adele-morrison]

This is nice! My take away from this (particularly this last plot of the export: ) is that we see roughly the same sensitivity to top cell resolution in both KPP and ePBL configs. So I don't think we need to do any more KPP tests.

Perhaps we could run both the ACCESS-OM2 and panan ePBL cases out to 15 years, given we have lots of hours left this quarter (could put them on oz91)? That way we could get the export time series a bit more stable?

And check how the North Atlantic dense water formation (just the overturning?) changes in ACCESS-OM2. Then write up I reckon!