[MITgcm-support] linear EOS and Tref

[Martin Losch]

Hi,
I have a puzzle to solve, maybe someone can help:
I use the linear EOS with a constant Tref = 1 and rhoConst=rhoNil=1035. 
I have an inital temperature of 0.1+0.01*random_noise[0 1]. With a heat 
flux of 200W/m^2 at the surface and non-hydrostatic dynamics I get 
convection after a while, everything is fine, scaling laws seem to be 
good etc.
Now I repeat the experiment with a different initial temperature of 
3+0.01*random_noise[0 1]. I thought that this should not change the 
solution (provide that the "random_noise" is the same), because all 
that matters dynamically is the 3D gradient of (T-Tref) which is 
actually independent of Tref. In my case I have T+deltaT-Tref, where 
deltaT=2.9=constant, so effectively I have a different Tref. But 
contrary to my expectations the solution does change and, what is 
particularily bad, the convection regime does not follow general 
scaling laws (e.g. average of w^2 should scale with sqrt(Q/f),etc) 
anymore and the temperature skewness no longer has a smooth vertical 
profile etc. All in all the solution looks weird.
(BTW,  theta never reaches freezing (-1.8degC).)

Any clue what could be going on?

Martin