[MITgcm-support] model stability when increasing resolution

[Alexandre Pohl]

Dear users,

I use the MITgcm in a coupled setup with aim_v23 and thsice. I based my previous runs on the 'cpl_aim+ocn' verification case and everything was working pretty well with the CS32 grid. I recently adopted the CS48 grid, the bin files of which are provided by Forget et al. (doi:10.5194/gmd-8-3071-2015).

Since then, the model is unstable. It frequently blows up with this kind of error message :
(PID.TID 0010.0001) *** ERROR *** SUFLUX_POST: TS= 1.2635634795961E+04 out of range  100 400
The model diverges and the temperature exceeds the reasonable range at one of the corners of the cubed-sphere.

Reducing the atmospheric time step increases the stability but it is really expensive in terms of computer resources. In order to solve this issue, I modified the 'cpl_aim+ocn' verification case as follows:

1. I use the implicit gravity wave (implicitIntGravWave=.TRUE.) instead of z*, following this thread <http://mitgcm.org/pipermail/mitgcm-support/2007-November/005096.html> and based on the 'hs94.cs-32x32x5' verification case; together with the following parameters based on this thread <http://mitgcm.org/pipermail/mitgcm-support/2006-September/004324.html>: useAbsVorticity=.TRUE., SadournyCoriolis=.TRUE., selectKEscheme=3.
Based on the previous references, I also added  'momViscosity=.FALSE.', replaced 'saltAdvScheme=3' with 'saltAdvScheme=2' and I commented 'useJamartWetPoints' and 'staggerTimeStep'. I attach my 'input_atm/data' file to the email.
2. I also adopted 'ALLOW_ADAMSBASHFORTH_3' from the same 'hs94.cs-32x32x5' verification case.
3. I use 'nShapT=3' instead of 'nShapT=4'. If I understand well, the smoothing effect is stronger with this new parameter value. Please note that no major corner effect appears in ETAN after 75 model years. 
4. in case it could be important, I also changed a few things regarding the oceanic component: I use 'viscAhGrid=0.1' instead of 'viscAh=3E5' and I adopted a more reasonable vertical discretization.

As a result, the model seems much more stable so that I can use atmospheric and oceanic timesteps of respectively 300s and 1200s, which allows me to simulate 1 model year every 19 minutes, i.e., ca. 75 model-years per day using 73 CPUs.

So, my questions are:
1. Are the changes that I made reasonable ? Are there some parameters to change when changing the horizontal resolution that would allow me to improve the stability / performance of the model ?
2. What do you think about the performance of the model ? Does it seem correct ?

Any help / thoughts would be greatly appreciated !
Thank you,
Alexandre




__

Alexandre POHL
Postdoctoral Researcher
CEREGE, TECHNOPOLE ENVIRONNEMENT ARBOIS-MEDITERRANEE
BP80, 13545 AIX en PROVENCE, CEDEX 04, FRANCE
https://alexandrepohl.org <https://alexandrepohl.org/>






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20170223/ca06dcfe/attachment.htm>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: data.txt
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20170223/ca06dcfe/attachment.txt>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20170223/ca06dcfe/attachment-0001.htm>