Warming in one degree simulation is vertical grid dependent

[taimoorsohail]

Hi all,

I have been looking at maximum SSTs in the one degree ocean simulation, and I have noticed the maximum SST and latent and sensible heat fluxes scale with the thickness of the top grid cell. As you can see below, a dz_surface = 44 m yields a much higher maximum SST (~45C) compared with a dz_surface = 100 m. The seasonal signal is more pronounced (and more correct, when compared to EN4) when dz is greater.

I am wondering if this can help us pinpoint the issue in the ocean surface warming in the simulation?

cc @navidcy

[taimoorsohail]

For reference, these are the maximal SSTs in EN4, with the 90-91 period used in JRA55 in orange:

[taimoorsohail]

OK, after chatting to @navidcy , it seems the issue is that the surface heat fluxes are computed on the bulk temperature, i.e., the volume-averaged temperature in the top grid cell. This means that any changes to the grid cell thickness will impact the SST, which in turn impacts the flux formulations.

The solution would be to either compute fluxes on SkinTemperature, or to make the top grid cell very thin, ~1m. The thin surface grid cell would have the added benefit of allow shortwave penetration. In the current one_degree_simulation example, the top grid cell is ~50m, which means any exponential shortwave penetration scheme still dumps all the radiation into the top grid cell.

[taimoorsohail]

cc @simone-silvestri

[simone-silvestri]

Probably we should have the first grid cell around 5 meters, 1 meter seems quite a lot to handle in terms of vertical CFL for very shallow regions. It might slow down the computation significantly for a 1-degree model.

[simone-silvestri]

I am looking at some solutions of the 1/6th degree OMIP, and I think the issue here is not really the size of the vertical grid cells, but the shallow regions that are not masked when using a smaller grid size in the vertical. I'll explain, in my case, I use a vertical grid with 60 points. The spacing looks like this

With a surface dz of about 5 meters.
After 5 years of evolution, this is the SST

Where the maximum temperatures are found around an i-index of 1500. This is the meridional temperature profile at i = 1550

Where temperatures do not increase above 40 degrees. However, the maximum temperature is found in this very shallow region in the Filippines (about 150 meters depth) where SST is touching values up to 43 degrees.

The same happens near the estuary of the amazon river (which makes me believe that there might also be some problem with having a very fresh inlet.

I suspect that something like this is causing the difference between 44 meters and 100 meters spacing. We are exposing regions that have very little exchange with the rest of the ocean, are located near the equator, and heat up without really feeling the seasons. It would be interesting to check if also in the one degree these "hot spots" are located at river outlets.

However, this said, it seems that just monitoring maximum SST is not a good way to diagnose this specific SST problem.
Even if the SST might be higher, there are definitely very localized outliers.
Btw, @taimoorsohail this should solve the issue with SST increasing drastically on a moving grid

[glwagner]

How do those localized regions get so hot? What is the source of heat?

[simone-silvestri]

I think it has to do with river inflow that provides very fresh water at the surface that stagnates because there is no flow to remove the very fresh water and the temperature rises significantly.

[glwagner]

Ok, but it doesn't make sense to me that it would rise so significantly above the air temperature. I don't think land gets that hot (which also has very little vertical heat conduction). Are we sure the budget is balanced there?

[glwagner]

Also 150m is not that shallow! Boundary layer depths in the tropics may be just a few meters to ten meters. So wide swaths of the world ocean have mixing only down to a few meters. This really should be investigated, the story is not closed here.

[glwagner]

Latent heat fluxes should be totally bananas high with water temps at 42 deg, right? Seems like we should see massive evaporative cooling. Can you make the same plot with the latent heat fluxes and show that this hot spot also has elevated latent heat flux?

[simone-silvestri]

Ah, for sure, this was not meant to be a conclusive answer :) It was just to say that I think the vertical resolution is not necessarily the cause of the increase in sst, but just that the areas that are masked change if we change from 44m to 100m vertical resolution, leading to the appearance of these "shallow" zones, so the plot at the beginning of this issue might be a bit misleading, because the maximum of sst does not necessarily refer to the same geographical location.

[simone-silvestri]

This is the temperature flux at that location (the top boundary condition of temperature). It does not necessarily seem to be much different than the surroundings

[simone-silvestri]

I am a bit confused though, as I don't see the daily cycle which should be quite clear

I m going to take a deeper look at this multi-year JRA55 data

[simone-silvestri]

While we're at it, this is the salinity flux, dominated by precipitation and positive where ice is melting in the Arctic and negative where ice is forming in Antarctica (this is a snapshot in july)

[glwagner]

Also 40 deg SST is still way too warm, that is warmer than the hottest ocean temperature ever recorded?

[simone-silvestri]

For reference, these are the maximal SSTs in EN4, with the 90-91 period used in JRA55 in orange:

It looks like 42 degrees might be an allowable solution.

[glwagner]

Huh ok, that is good to know. I guess there is more variance and anomalous values in earlier periods where the data is not certain. What about average SST?

[simone-silvestri]

I guess I'll try reverting back to the repeat year forcing and seeing if something changes. And in the meantime, try to diagnose fluxes with the multi-year JRA55.

[glwagner]

If there is not a correlation between latent heat flux and SST we need to understand why. We can try fixing the friction velocity where the fluxes are computed (say u* = 0.1), so that does not vary independently. I guess for all we can tell the SST is correctly input into the turbulent flux solver. Something is awry though.