[Mitgcm-support] Re: more Obcs ruminations.

[mitgcm-support at dev.mitgcm.org]

Hi Jake,

Thanks for the tips: I've done some more investigation of the problem
with the help of information you provided.

cg2dTargetResidual doesn't affect the problem. Thus I've decide to go with
1e-8 as you do.

You're right useEnergyConservingCoriolis isn't helpful here. I've turned
it off again.

The flux limiting advection schemes work nicely, but I did an adjoint
sensitivity calculation with this advection scheme and it blows up. It
gives 10^69 when I was expecting 1!

I tried a every different combination of diffusion coeffs without any
success. I think I've got rid of these oscillations and they pop up
somewhere else.

Smoothing the topography is not helpful as you suspected.

My latest fix which seems to be working so far is to enable KPP. I had it
disabled because it wasn't working well with the adjoint. I noticed that
the oscillations are associated with cooler surface layers than subsurface
layers in the winter/spring. (I never see the oscillations in the
summer/autumn). KPP appears to eliminate this and the associated
oscillations. Jake could you remind me of the modifications you made to
KPP to get it work with adjoint?

It seems like all the solutions I have cause problems with the adjoint at
the moment!

Thanks

Dan