Arctic sea ice in TCo319-Dart simulation

[qiangclimate]

(1) there are checkboard patterns in sea ice concentration and thickness in some seasons (seemingly in melting season of 1995; visible in both concentration and thickness; checking another arbitrary year, 2000, we got the same findings.)

(2) sea ice is close to be free in August, 1995 already

(3) sea ice seems to be biased thin

(4) No checkboard patterns seen in Southern Ocean so far

(5) No lead features in sea ice Thickness fields in both Arctic and SO. (although 5km resolution, we expect to see sth in sea ice thickness)

Below are daily sea ice concentration for 4 days (representing different seasons) in 1995:

Below are daily sea ice thickness for 2 days in 1995: Day 91 Day 350

Southern Ocean daily sea ice thickness for 2 days in 1995 (no checkbord patterns):

[qiangclimate]

(6) Looking at sea ice thickness in the Southern Ocean, I feel something uncommon to me, but cannot tell what it is exactly. Perhaps something like being very discontinuous, either due to not converged sea ice solver, or imprint of jumping forcing (on sea ice). Just uncommon.

[koldunovn]

Can you show your namelist.ice?

[qiangclimate]

this is the Aleph simulation.

[qiangclimate]

(7) Looking at sea ice concentration on day91 in 1995, a winter day, we see there is no sea ice in Barents Sea and part of Kara Sea, implying that very possibly too much Atlantic Water heat enters the Arctic. In a real world, winter sea ice edge at this location is largely determined by ocean heat.

[qiangclimate]

(8) Looking at Arctic sea ice thickness on day350 in 1995, I guess that, if sea ice drift direction is not made wrong by not-converged sea ice solver, then we could think about winds in the model. Sea ice is not pushed/accumulateed to Greenland/CAA, and there is a strong stream of sea ice flowing into Beaufort Gyre. Anyway, sea ice solver remains as a possible reason.

[JanStreffing]

As mentioned yesterday, we see the imprint of atmospheric spectral ringing, which is a common feature for models with a spectral core like ECHAM6, or OpenIFS. Similar features can be seen e.g. in: https://doi.org/10.1029/2017MS001233

In the past the ringing artifacts were large scale. As our atmosphere is higher resolution, the artifacts become smaller. TCO grids are more prone to ringing than Tq grids.

[qiangclimate]

(a) Did EC-Earth see the same feauture in their sea ice?

(b) "If" TCo12xx super high resolution case does not show such checkboard mode, do we get some hints?

(c) "if" such checkboard mode really only happens when sea ice melts, do we get some hint? (We checked the sea ice just taking arbitrary days, so it is not a definite conclusion about the seasonality of the checkboard yet)

[mzapponi]

this is the namelist.ice:

&ice_dyn whichevp = 2 pstar = 30000.0 ellipse = 2.0 c_pressure = 20.0 delta_min = 1e-11 evp_rheol_steps = 120 alpha_evp = 250 beta_evp = 250 c_aevp = 0.15 cd_oce_ice = 0.0055 ice_gamma_fct = 0.5 ice_diff = 0.0 theta_io = 0.0 ice_ave_steps = 1 /

&ice_therm sice = 4.0 h0 = 0.5 emiss_ice = 0.97 emiss_wat = 0.97 albsn = 0.77 albsnm = 0.65 albi = 0.65 albim = 0.55 albw = 0.066 con = 2.1656 consn = 0.31 /

[tsemmler05]

(1) there are checkboard patterns in sea ice concentration and thickness in some seasons (seemingly in melting season of 1995; visible in both concentration and thickness; checking another arbitrary year, 2000, we got the same findings.)

(2) sea ice is close to be free in August, 1995 already

(3) sea ice seems to be biased thin

I have seen the checkboard patterns in atmospheric parameters before over the Northern Hemisphere - assumed that this may be caused by spectral resolution. But why only over the Northern Hemisphere? Because of the more pronounced orography (more land areas over the Northern Hemisphere compared to the Southern Hemisphere)?

Yes, the sea ice over the Arctic is reasonable in winter but too thin and of very limited extent in summer very early. Therefore, we need to use slightly higher albedo values. This will increase Arctic summer sea ice while leaving Southern Ocean sea ice largely unchanged (as seen from the tuning simulations). Three years of test data are under /scratch/awiiccp5/ctl1950alb

[tsemmler05]

(a) Did EC-Earth see the same feauture in their sea ice?

At the EC-Earth meeting the issue was discussed in the paleo context. TCO95 (and also TCO159? Not sure about this anymore) gave these features, TL159 did not give these features. EC-Earth paleo group discussed to move back to TL159 resolution; not sure if they did it at the end.

EC-Earth people looked at atmospheric parameters, not sure anymore if also at sea ice. But, I guess, this is related to each other.

[JanStreffing]

As I mentioned, TCo grids have this issue more than Tq grids. TL grids have it even less. So far I decided against TL grids, for three reasons:

1. Using TL grids we spend the majority of resources on high spectral resolution, giving us 1/4 as many surface points to exchange across the air-sea interface.
2. ECMWF has last used TL grid operationally in 2016. All new developments and projects like DE, NextGEMS, EERIE, and WW are using TCo grids.
3. The spurious mass gain is larger with TL grids, making them less suited for long simulations:

https://www.researchgate.net/profile/Christian-Kuehnlein/publication/297695132_A_new_grid_for_the_IFS/links/56e03f5108aee77a15fe8fca/A-new-grid-for-the-IFS.pdf

[trackow]

Tido, can you check which of the forcing fields does have this spectral ringing? e.g. heat fluxes, stresses, or Is it all of them?

[tsemmler05]

It seems like turbulent heat fluxes show the pattern more in the Northern Hemisphere than in the Southern Hemisphere. For the north-south and east-west stresses one sees a checkboard pattern all over the globe. For the radiation fluxes it does not become too obvious, probably because of the strong gradients in these fields.

I have 30 year averages computed here: /scratch/awiiccp5/ctl1950d/outdata/oifs/atm_remapped30ymean*_2000_2029.nc. I just viewed them in ncview. Both sshf and slhf show the issue over the Arctic.

[qiangclimate]

A general question is, could we solve this issue for CMIP7? Is it an openIFS issue, or a coupling issue?

[JanStreffing]

I would say it's a TCo grid issue. TCo grids have a high gridpoint resolution for a given spherical harmonic resolution. This prompted the EC-Earth community to go back to TL grids. We could follow this path and use for example TL511 instead of TCO319. The cost should be comparable. We would go from 31km to 39km. I can make a test to see what the SYPD impact would be. I'll get back to this once I have made such a test.

My main concern, is that ECMWF has moved away from TL grids ten years ago (cy40r1).

[helgegoessling]

Has there been further discussion/progress on this?

Some thoughts:

I agree with Jan that moving away from TCo back to TL is not a good idea. It would come at the big risk of unexpected side effects and bad surprises, given that the last decade of developments has been focussed on TCo configurations.

Has someone checked whether the patterns are also visible on the original TCo grid? I mean, without even transforming them to a regular grid, e.g., via CDO -r option? (I guess they should not be viewable in ncview without that transformation.) Wondering whether there might still be just some sort of aliasing goin on.

The fact that no issue is visible in the radiation fluxes seems plausible given that the radiation lives in grid-point space and is affected by the dynamics only very indirectly. The winds obviously live in the spectral space, but even for the turbulent fluxes I could imagine that, although they live in grid-point space, they are more directly affected by the winds (think of boundary-layer turbulence), and so may inherit some artifacts.

[JanStreffing]

I have not worked on this part in the meantime.

Has someone checked whether the patterns are also visible on the original TCo grid? I mean, without even transforming them to a regular grid, e.g., via CDO -r option? (I guess they should not be viewable in ncview without that transformation.) Wondering whether there might still be just some sort of aliasing goin on.

There is no regularization here. We are looking at output on the native FESOM2 DART mesh. The interpolation happens directly from the reduced Gaussian TCO319 to DART. The remapping is done in OASIS through SCRIP with these options:

GAUSWGT D SCALAR LATITUDE 1 9 2

So the FESOM2 surface fluxes are computed from the gauss weighted differences of the 9 nearest atmospheric datapoints, with a variance of 2, which means the Gaussian distribution is very wide, makeing the method close to DISTWGT from my understanding. I thought that this meant we are already smoothing out this effect, but perhaps I'm mistaken here.

I found in the past, that I was able to minimize interpolation error for TCO95-CORE2, but reducing the variance to 0.1, and increasing the number of neighbours to 25. However this setup was not stable for higher resolutions. But perhaps we can try something in between. e.g. 9 Neighbours and variance of 0.2 or 0.5. We might just get an even clearer remapping of the pattern onto the FESOM grid though.

The fact that no issue is visible in the radiation fluxes seems plausible Have we established this? I think we should still plot this. This can be done on a short run with oasis expout. We will be able to plot the fluxes before and after remapping, as oasis sees/generates them.

I will block myself one day (Thursday) this week to work on this.

• [ ] Plot the fluxes from oasis
• [ ] Try different remapping settings
• [ ] Compare to TCO1279-DART sea ice