[MITgcm-support] Re : Water at the bottom of NA below freezingpoint

[Jeff Scott]

Hi Martin and others,

Continuing this discussion...

I lay out my case for an advection-related problem as follows:

It is not just me that seems to a problem with anomalously cold temps, but 
Gus, Dror, and Christopher have chimed in, seeming to offer agreement of 
some sort. That being said, it is not entirely clear that we are all 
dealing with the same problem - I'm a little puzzled by Gus' comment that 
he is "losing half the heat content". I'm not sure exactly how that was 
determined, but sounds like a massive loss, whereas I'm focusing on 
gridpoints that are colder than they should be (although this does spread 
in some cases).

I don't think there is much in common between our setups, other than being 
coarse res; none of them are running my specific coupled configuration. In 
my setup, some of the problem gridpoints are fully covered with seaice 
(THSICE) all or at least part of the year. If there were a problem with 
THSICE and/or the coupling, why would the problem be so localized?

But I think the biggest argument I can offer is that when I use SOM, the 
problem largely goes away: minimum temps are generally much closer to the 
-2 C, roughly the coldest possible icefreeze temp. As I mentioned, I found 
SOM to add some diffusion to the system (ala what Christopher noted), but 
it also added some "wiggle" to the velocity field, particularly noticeable 
on a MOC contour plot.

Here is the minimum temps in two configurations, by vertical level, 
separated into NH and SH:

4x4x22
Depth   Min SH  Min NH
-----   ------  ------
5	 -1.8	 -1.7
18	 -2.3	 -1.7
35	 -2.8	 -1.7
60	 -3.2	 -1.7
95	 -3.2	 -1.7
142	 -2.4	 -1.3
205	 -1.9	 -0.4
285	 -1.6	 0.0
385	 -2.2	 0.0
508	 -1.4	 0.0
655	 -1.4	 0.0
830	 -1.5	 -2.9
1038	 -1.5	 -0.7
1282	 -1.7	 0.0
1570	 -1.6	 0.0
1905	 -1.7	 0.0
2292	 -1.7	 0.0
2738	 -2.0	 -0.0
3245	 -1.7	 -0.8
3820	 -1.8	 -1.0
4468	 -1.8	 -1.2
5192	 -1.7	 -0.9


2.5x2x22
Depth   Min SH  Min NH
-----   ------  ------
5	 -1.8	 -1.8
18	 -1.8	 -1.8
35	 -1.7	 -1.8
60	 -1.7	 -1.7
95	 -1.4	 -1.6
142	 -0.6	 -1.5
205	 0.0	 -3.9
285	 0.0	 -4.0
385	 0.0	 -1.6
508	 0.0	 -1.8
655	 0.0	 -3.3
830	 0.0	 -1.9
1038	 0.0	 -1.8
1282	 0.0	 -1.7
1570	 0.0	 -1.8
1905	 0.0	 -1.8
2292	 0.0	 -1.9
2738	 0.0	 -1.9
3245	 0.0	 -1.9
3820	 0.0	 -1.8
4468	 0.0	 0.0
5192	 0.0	 0.0


(min of 0.0 is really 0.0 or above)


In the 4x4x22 case, the show-stopper is the problem in the top 100 meters 
- these "bad" temps occur only at one point, just east of the stub of land 
jutting N of Antartica into the Drake Passage. It would seem this is a bit 
of a stagnation point.  But note there is also a "bad" point at 830m in 
the NH. I did look at the SH bad point at every time step; at no time is 
the surface below the freezing point, at least as dumped in the end of the 
k-loop in thermodynamics.F (the the coldest point seems always to be at 65 
or 95m depth). FWIW, the anomalous cold spot increases during summer and 
fall, decreases during winter and spring, at least for the year I 
examined.

In the 2.5x2x22 case, the SH is fine everywhere, but there are isolated 
cold points at 285 and 655m in the NA. But unlike the 4x4x22 case, these 
don't seem to spread and influence the whole deep ocean, and the model is 
not filling up with sub-zero temps. (ergo I tossed the 4x4x22 setup until 
I could solve this problem).

I also have a 4x4x15 case w/o KPP, and it exhibits the problem at one 
point in the NA, although the SH is again fine.

In common here are strong flows near rough, jaggedy topography. Perhaps 
there is some issue in how I'm making the topography? (I do fill in 
"holes", but otherwise I not trying to smooth it out). Perhaps I should 
simply try to smooth out topography where the problems occur.

My experience with the partial cells was simply that: I spent a week 
playing with them, and this problem became much, much worse (and I think 
the MOC also became more jaggedy too), so it was a pretty obvious 
conclusion that this was exacerbating my problems. I offer no reasons 
why...

Other issues/ideas: perhaps decreasing the time step would help... 
although this is a rather costly solution. Moreover, playing with the 
timestep can be difficult, as you then have to worry about the ratio of 
deltaTmom and deltaTtracer, among other things that might be tempted to 
blow up (in addition to technical issues between atmos, ocean, and 
coupling timesteps). But perhaps my issues with partial cells help point 
the finger toward the time step? (My time steps are 14400/480 for 2.5x2, 
28800/900 for 4x4)

FWIW, I did turn off temp and salt advection, and the cold anomaly 
diffuses away into the background on the order of months.


Jeff





On Wed, 12 Mar 2008, Martin.Losch at awi.de wrote:

> Hi all,
>
> I just cannot believe, that the advection (scheme) alone can produce, 
> what you are observing. Advection schemes can be poor and can produce 
> overshoots/undershoots where there are steep tracer gradients. In the 
> abyssal ocean you generally do not have these steep gradients. Also I 
> have never observed the ocean filling with cold water in my uncoupled 
> runs (with sea ice!!), and if I have observe temps below freezing in the 
> past, I could *always* trace this problem back to numerical stability 
> issues, eg. the timestep was too large, etc.
>
> I run passive tracers with very steep gradients (0 in one box and 1 in 
> the next initially), and I get overshoots on the order of 1e-6 for the 
> flux limited schemes at most. Certainly not enough to explain your cold 
> temperatures.
>
> Again, my suspicion is, that very cold water is produced probably at the 
> surface (heat flux), which is then immediately convected to lower 
> depths, at the surface being replaced by warmer water. The new warm 
> surface waters will again be cooled (e.g. atmospheric temperatures may 
> be too cold, problems with humidity?) and convect, etc. That way you can 
> easily fill the bottom of the ocean with cold water. Inaccurate 
> advection schemes are much less efficient for doing that (Jean-Michel, 
> say something). In coupled ice-ocean simulations I had situations where 
> I made mistakes with units etc of prescribed atmospheric fields 
> (temperature and salinity), that lead to dramatic ice growth, but never 
> to supercool temperatures in the deep ocean. Diagnose you heat fluxes 
> (eg., diagnostic oceFlx, but there are others that describe the full 
> temperature tendency) and see if it is balanced over longer periods 
> (year to years), I bet it isn't.
>
> I don't see why turning off partial cells should help in this situation, 
> either. Generally, solutions near topography are smoother with partial 
> cells, the only disadvantage I see is that sometimes the cfl-criterion 
> for w is more restrictive with partial cells, but I have not observed 
> that in coarse resolution runs, yet.
>
> In order to rule out the advection schemes as culprits I suggest that 
> you try turning off the advection of tracers altogether. That will give 
> you a completely different circulation in the ocean as only diffusion 
> will distribute tracers, but for a test it should be fine. To do that 
> set tempadvection=.false., saltadvection=.false., in data, PARM01 If you 
> still get the low temperatures, then you know that it's not the 
> advection schemes. If not, I'll shut up (probably).
>
> Gus, what I forgot to mention in my previous email: instead of 
> "useOldFreezing" I recommend using "allowFreezing". With this flag you 
> only reset the surface temperature to -1.9, this acts as a minimal 
> sea-ice model and no supercool temperatures can enter the ocean. Given, 
> I am right about the advection schemes, then there shouldn't be any 
> water below freezing in the ocean.
>
> Scott, I don't think 22layers in 4x4 or 2x2 degress is a problem. I do 
> exactly that (actuatlly 23layers with 10m near surface), and the old 
> ecco solutions are 2x2 with 23 layers (10m near the surface).
>