[Mitgcm-support] AIM and NCEP files

[mitgcm-support at dev.mitgcm.org]

Hello,

Some news about AIM with NCEP input files:
(with old input files, 10 yr with full cell and 10 yr
with partial cell will be finished tomorrow) 

with partial cell, deltat=450s, 
Q : Advection 3rd O (version with geometric factor) and no AB
T : Advection 2nd O.
run 14 months and blows up (I analyzed this crash, 
strong W at the corner near Japan + feedback on T
=> blows up).
This was with the finite volume form of calc_phi_hyd,
(like the previous runs we did on LatLon grid).

I change to the energy conserving form, runs slightly longer 
(> 15 months) but finally blows up.

I start to run with the new 3rd O advection scheme for T:
blows up very rapidly (20 days).

Then I return to full cell: with deltaT = 600s, does not
last longer (20 days). This is surprising since:
a) with 2nd-O Adv for T, Full Cell and deltaT = 600s,
  the model runs fine, at least more than 3 yr (I stopped it after).
b) H&S is OK with the same topography, Partial Cell,
  3rd O advection for T, with deltaT=450s (10 yr) and now
  with deltaT=600s (> 2yr).

full cell & 3rd O : the model blows up near Australia, 
at an other corner, with beautiful oscillations (might be 
a Pb with AB), and strong W (even at the surface); then T 
starts to be wrong and it ends. No Pb with Q except very 
close to the end.

some idea:
* 3rd O might not be good with AB ? (not sure, but with Q
it seems to work fine without AB) 
* the BC of this 3rdO scheme cause Pb with AIM (but
then why is it OK with H&S ? may cause AIM adjustment
that induce Pb with AB.
* I don't see any reason to keep AIM G_terms in AB.

Next:
4thO Adv. for T (it is running with AIM & Full Cell, >160d done).
3rdO Adv. for T with no AB. otherwise, test for T the same 
                  3rd-O formulation that is now used for Q.

But I don't want to reduce again the time step because it 
takes a long time to get a 10yr run.

Jean-Michel