[Mitgcm-support] Re: model comparison

[mitgcm-support at dev.mitgcm.org]

Detlef et al.,

The main difference is that (U,V) is staggered in time with
(S,T) in the old model and they are synchronous in the new
model. We did this to allow us to consolidate the communication
to one place (do_gterm_blocking.F). When we've made these
comparisons in the past we normally re-arrange the old code
to synchronous. I think the offset in salt is due to this.
Don't understand why rigid-lid fixes it!

There are quite a few other changes some of which are marked
with CPP flags in the new code (such as OLD_ADV_BCS) but we
haven't used them for sometime - the new model is different
in discretization of some terms like boundary conditions
and calculation of some grid geometry. There are undoubtedly
others which are not marked. I don't remember when we last
did a backwards comparison but it was way back probably
in the single digit checkpoints.

Oh, one other difference is the units have changed!

Alistair.

"Detlef B. Stammer" wrote:
> we are now down to a comparison of the two models number by number. And there
> starts the problem: The two models (old and new) behave substantially different
> at the very first time step with serious consequences for the later development.
> In the old code the tracers show a step (e.g. 0.1 in salt) at the first time
> step and then continue smoothly but offset from the new code. I can eliminate
> the step with rigid lid and my old rigid lid looks like the free surface in the
> new code. But we have to confirm this quantitatively.
> 
> Before we spend hours and days on solving this: Can you recall what you changed
> going from old to new in the initialization? Is there something missing in the
> old code that I can easily fix? In case you have a quick idea please let me know
> since we are some what puzzled.