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

[Martin Losch]

Hi Gus,

I think it's obvious now that Jeff's problem is very different from  
what you observe (and Dror?). From what you describe your surface  
heatflux removing too much heat from the ocean causing the surface  
temperatures to decrease -> convection -> deep ocean fills with cold  
water.

allowFreezing=.true. is a minimal ice model (hack) without any impact  
on the heat flux, but I will add heat in the surface layer  
(proportional to theta-tFreeze, if theta < tFreeze). So it's a good  
idea to go away from this in the future. I would turn it off for  
debugging as well.

surface relaxation is a heat source as well, but it's part of the  
overall balance and you can diagnose it (TRELAX is the name of the  
diagnostics, I believe).

Your parameters for GM and KPP look good. You may want to use  
GM_taper_scheme    = 'ldd97', because you are using KPP with GM, and  
that's what Large-etal (1997) did for their taper scheme.
Your vertical diffusivity could be a little smaller (1e-5), but we  
don't know this number very well.

For debugging, it may be a good idea to go without KPP and GM and put  
them back in, when you have sorted out hour heat flux? I strongly  
suggest diagnosing the various components, oceQnet, oceQsw, oceFreez,  
TRELAX.

Martin


On 13 Mar 2008, at 22:53, Gus Correa wrote:

> Hello Martin, Jeff, and Samar
>
> Thank you very much for your suggestions.
> I am not disputing what is causing the cooling, I just want to fix  
> the problem.
> Your help to do this is much appreciated.
> Also, it is not necessarily the same problem that Dror and Jeff  
> reported, I agree,
> despite some similarities.
>
> I'll be away this coming week, but I am very interested in your  
> suggestions.
> In case you reply to this message, I'll get back to you when I return.
>
> 1) I ran again the coupled ocean-mixed layer atmosphere for 50 years,
> changing the advection scheme to 3rd. order upwind, using a  
> staggered time step,
> turning off the shaved cells, and replacing useOldFreezing by  
> allowFreezing=.TRUE.
>
> I still get the cold temperatures.
>
> As Martin guessed, the cold water sinks from the surface.
> This happens around Antarctica
> (there is cold water at all depths there) and spreads across the  
> ocean bottom,
> regardless of  the new advection scheme and the removal of shaved  
> cells.
>
> So, I may need to go back and find what is wrong on the surface  
> (something that
> I've been doing for quite a while).
> One possibility is that my surface forcing shortwave dataset is bad
> (it is ERBE surface radiation budget, but I already used NCEP  
> Reanalysis, and ERA40,
> and the problems are about the same.)
> Another possibility is that the bulk formulas for longwave,
> sensible and latent heat are not very good (but they are more or  
> less standard).
> We use Trenberth wind speed on the bulk formulas, and I hope this  
> is an ok dataset.
> I also used other wind products but the problems weren't fixed.
> Another possibility is that there is a bug on my code.
> I already searched extensively for bugs, found and fixed several,  
> but of course there may be a bug.
>
> 2) I am trying to avoid any artificial heat source.
> I just learned from Martin that one of these is "useOldFreezing",
> which should be replaced by allowFreezing=.TRUE., which I will  
> adopt from now on.
>
> However, I presume another artificial heat source is surface  
> temperature relaxation to
> climatological data (i.e. a non-zero value for tauThetaClimRelax).
> I don't use surface temperature relaxation in the coupled runs.
>
> I just ran the (non-coupled) test case global_ocean.90x40x15 with  
> and without surface
> temperature relaxation, and the difference in the resulting  
> temperature field is very large on the
> first three or four levels.
>
> I wonder if any of you have done successful coupled runs without  
> surface temperature relaxation.
> (Our goal for the coupled runs is certainly to avoid surface  
> temperature relaxation.)
>
> 3) I would much appreciate if you can advise on what KPP and GM  
> parameters I should use.
> I am using what I could find on a few test cases and some  
> suggestions of others,
> but I am afraid I may be choosing bad values, and producing wrong  
> vertical and horizontal mixing.
> Aside from the atmosphere code itself, this could be yet another  
> cause of problems, right?
> I  don't know if the KPP and GM values below are reasonable for my
> 4deg X 4deg X 24 levels (10m spacing on the top 100m) grid.
>
> A) In the "data" namelist:
>
> viscAr=1.E-3
> viscAh=5.E5
> diffKhT=0.0
> diffKrT=3.E-5
> diffKhS=0.0
> diffKrS=3.E-5
> implicitDiffusion=.TRUE.,
> implicitViscosity=.TRUE.,
>
> B) In the data.kpp namelist (this namelist is actually empty)
> &KPP_PARM01
> &
>
> C) In the data.gmredi namelist:
>
>  GM_background_K    = 1.e+3,
>  GM_taper_scheme    = 'gkw91',
>  GM_maxSlope        = 1.e-2,
>  GM_Kmin_horiz      = 50.,
>  GM_Scrit           = 4.e-3,
>  GM_Sd              = 1.e-3,
>
>
> 4) Samar's suggestions/questions:
>
> A) I still didn't implement sea ice.  I am trying to solve the  
> current problem before I venture into more coding,
> and to keep it simple, as you suggested.
> B) My mixed layer atmosphere version uses exactly the same T grid  
> of MITgcm.
> The grid cell areas are the same.
> So I don't need to do spatial interpolation of atmospheric fluxes  
> that are passed to the ocean.
> I assume Matlab does a decent job interpolating standard datasets  
> to the T grid, and this is what I use off line.
>
> 5) Jeff's question:
>
> I determined the ocean heat content by integrating it spatially and  
> in time over a long 1000 year run.
> It decreases from ~ 3x10**25 J to ~ 1.5x10**25 J during this period.
> Then I saw the inordinate amount of cold water on the ocean bottom.
> Since you and Dror reported spurious cold bottom water also,
> perhaps caused by the advection scheme or bottom topography,
> I thought these two might be the cause of my cold water as well.
> However, other than this coincidence, our surface diagnostics were  
> showing imbalances in global mean Qnet,
> suggesting that the problem was on the surface.
>
>
> Thank you.
>
> Gus Correa