CORE2 mesh review

[JanStreffing]

While working on another issue I recently made a number of plots for the Mediterranean Sea and was surprised by what I saw. I only looked at the Mediterranean in detail. Maybe such issues exist elsewhere?

It seems that in the entire basin the coastline is pushed back by one element. Perhaps to keep the Straight of Gibraltar open? The result are some very strange coastlines e.g. in Italy. Fairly major islands like Crete go missing completely. I don't think we would be able to approximate East-West-Meditaranean gradients correctly with such wide connections.

There are other choices about CORE2 like missing marginal seas:

More missing islands and missing straighs:

Considering that CORE2 will be workhorse for long AWI-ESM simulations of warm climates for the foreseeable future, I would like to start a review of this mesh. I think the timing is good, as @PengyangSong is currently setting the mother mesh for CORE2+ Ice cavities. Is the CORE2 mesh still best the way it is, or should we work on a CORE2-NG mesh that best suits our needs?

First round of questions from my side would be:

• What magnitude of errors do we make by these simplifications?
• What would be the cost (in terms of timestep) if we remove these simplifications?
• How close are these to our research areas? e.g. Focus on Romania for observations. If we ever want to compare to model results missing the Black Sea would be questionable.

Adding @patrickscholz, @dsidoren, @helgegoessling, @PengyangSong , @christian-stepanek, @tsemmler05, @chrisdane, @ogurses, @niceniulu, @trackow

[christian-stepanek]

Some notes to your questions:

1. What magnitude of errors do we make by these simplifications. Difficult to say, may depend on the case. I think that these simplifications of the coast line were just not noticed. The mesh generator is kind of predisposed towards creative interpretation of the bathymetry data set that one feeds into it. When generating a Pliocene mesh I noted that details of the land sea mask are very difficult to implement. The algorithm was notorious in removing small land masses, and I had to work a lot to keep important features (like a closed Isthmus of Panama) correctly represented in the mesh. Furthermore, and that was quite puzzling to me, there were far-field effects. When I modified topography around the Isthmus to keep the land bridge in the mesh intact, I suddenly noticed that the mesh generator reacted to such a localized orography change my modifying some coastline features far away (I remember some changes happening in the Indian Ocean, although the modification of orography just happened in Middle America).
2. I guess the costs should be negligible as we would only cause minor changes in the size of the total ocean - or did I misunderstand the question?
3. Independently of what are our research areas, it would be great to be always as close to reality as possible. Research areas may change.

[JanStreffing]

2. I guess the costs should be negligible as we would only cause minor changes in the size of the total ocean - or did I misunderstand the question?

i believe inclusion of marginal seas was sometimes omitted, because the straights leading in and out are prone to model blowups and can potentially limit the maximum timestep.

[ogurses]

If you are not interested in regional climate, I do not see any reason to go finer resolution in the Mediterranean. There are several spots where the intermediate/deep waters are formed and several straits (Otranto/Messina/Gibraltar, etc.) to be resolved for correct circulation. Rhodes Gyre and Gulf of Lions are the regions of convection . If you trace the Mediterranean overflow into North Atlantic back, you end up in Eastern Mediterranean (i.e., Levantine basin).The salinity of the overflow is actually decided there.

If you want tackle a 50m. resolution through Turkish Straits, then you can add Black Sea into your system (You need to spin up with a few seconds timestep in the beginning). Please consider to have a proper initial condition for the Marmara Sea beforehand otherwise you need a log time to have correct two-layer stratification there.

[JanStreffing]

If you are not interested in regional climate, I do not see any reason to go finer resolution in the Mediterranean. There are several spots where the intermediate/deep waters are formed and several straits (Otranto/Messina/Gibraltar, etc.) to be resolved for correct circulation. Rhodes Gyre and Gulf of Lions are the regions of convection . If you trace the Mediterranean overflow into North Atlantic back, you end up in Eastern Mediterranean (i.e., Levantine basin).The salinity of the overflow is actually decided there.

I see. I wasn't thinking really thinking about increasing resolution. I was surprised how the with the existing resolution we have entire 40-50 km edge length elements full of ocean in regions where a look at a map tells me there is land there. It looks to me like some parameter for where there should be ocean an where not was off for the Mediterranean.

If you want tackle a 50m. resolution through Turkish Straits, then you can add Black Sea into your system (You need to spin up with a few seconds timestep in the beginning). Please consider to have a proper initial condition for the Marmara Sea beforehand otherwise you need a log time to have correct two-layer stratification there.

I understand adding the Turkish Straights would be a potential bottleneck. We are representing the passage between Denmark and Sweden already, which is ~3.7km wide. We do so with an element that has about 25km edge length:

Do you think that that it would be a larger error to have ~25km wide Dardanelles and Bosporus, than to have no Black Sea in the model?

I'll have a look later at our flow through rates around the Indonesian Archipelago.

[ogurses]

I see. I wasn't thinking really thinking about increasing resolution. I was surprised how the with the existing resolution we have entire 40-50 km edge length elements full of ocean in regions where a look at a map tells me there is land there. It looks to me like some parameter for where there should be ocean an where not was off for the Mediterranean.

I think people were concerned at that time more about what comes out of the Mediterranean and how it integrates with the North Atlantic at mid-depths rather than the Med. itself (Coasts, east-west salinity gradient). I don't know, is it still the case?

[ogurses]

Do you think that that it would be a larger error to have ~25km wide Dardanelles and Bosporus, than to have no Black Sea in the model?

It won't work. 1) Minimum width of Bosphorus is 700 m (Causes hydraulic jump there). 2) There is significant ssh gradient between the Back Sea and the Mediterranean which should be kept for the water exchange between the basins (Black Sea ca. 18psu, Med > 38 psu). You can however include the effect by practically defining the Dardanelles as a river source into the North Aegean.

[JanStreffing]

Do you think that that it would be a larger error to have ~25km wide Dardanelles and Bosporus, than to have no Black Sea in the model?

It won't work. 1) Minimum width of Bosphorus is 700 m (Causes hydraulic jump there). 2) There is significant ssh gradient between the Back Sea and the Mediterranean which should be kept for the water exchange between the basins (Black Sea ca. 18psu, Med > 38 psu). You can however include the effect by practically defining the Dardanelles as a river source into the North Aegean.

Is it that different from the Baltic Sea? Here we have outflow width of the Øresund at 3700 over which we have a salinity gradient from ~10 to 30 PSU. Still we opted to include the Baltic with in CORE2 with Øresund width of ~25km.

I had a look at our CMPI6 MR simulation with glob mesh that includes the straights and the Black Sea. I plot the last year of the historic simulation. Three observations.

1. We approximate the Turkish straights as ~20km wide.
2. The salinity gradient between Black Sea and Aegean Sea is ok at ~17 PSU
3. The absolute salinity values are through the roof in both the Black and Mediterranean Sea.

I assume the latter could be Straights of Gibraltar issues or P-E errors. Or both.

Compare to AWI-CM3 with CORE2, again last year of the historic simulation.

We have the same high salinity bias in the Mediterranean with the different atmospheric model, suggesting Straights of Gibraltar. To me the glob mesh geometry looks like something we could approximate on CORE2 without having much wider channels than we in glob, and without resorting to 50m edge length triangles.