[MITgcm-devel] GM + KPP and non-local term

[Dimitris Menemenlis]

On Sunday 17 October 2004 11:13 am, Jean-Michel Campin wrote:
> Can someone who knows something about KPP implementation 
> comment about this point ?

Jean-Michel, I do not claim to know anything about KPP implementation,
but since I am responsible to a large extent for the MITgcm KPP mess ...

> 2) I see how this KappaRT can generate problems, because 
> KPP does not know anything about what GM will predict for Kwz,
> Kwz can perfectly be very large, and in this case
> the non-local term (that is always treated explicitly) 
> might perfectly be very large also, leading to instabilities.

Non-local transport occurs only for convective conditions, so KPPdiffKzT is 
always very large (order 0.1 m^2/s).  What kind of problems do you expect?

> 1) I don't see why the non local term should have a contribution
> coming from GM-Redi (they are meant to represent 2 different 
> processes) and 

KPP is supposed to represent not just one process, but all processes that 
contribute to vertical mixing.  From my perspective, GM-Redi-KPP interactions 
are black magic.  Large et al. (and also MITgcm by default) taper GM-Redi 
coefficients when isopycnal slope is greater than 2%.  So GM-Redi coefficient 
is mostly zero or negligible when there is non-local transport.  There's lots 
of weird things about this formulation.  One is that you end up with zero 
horizontal/eddy mixing near the surface, where, intuitively one would expect 
eddy mixing to be strongest.  My guess is that we need to reformulate the 
whole darn thing at some point in time.  But until then, my inclination would 
be to leave as is.

D.

-- 
Dimitris Menemenlis <menemenlis at jpl.nasa.gov>
Jet Propulsion Lab, California Institute of Technology
MS 300-323, 4800 Oak Grove Dr, Pasadena CA 91109-8099
tel: 818-354-1656;  fax: 818-393-6720