degradation of North Atlantic solution in ORCA025 at 3.6

[davestorkey]

The solution in the North Atlantic seems to be significantly degraded at 3.6 compared to previous versions. From other tests I think this is due to the changes to the EEN momentum advection scheme in dynvor.F90. Attached pictures are 5-year means from a 10-year spinup forced with CORE2 atmospheric forcing. The SSH shows that at 3.6 the Gulfstream overshoots more at the separation point and then is too zonal at the "northward turning". This results in the return of the familiar cold bias in that region as seen in the SST plots.

[davestorkey]

5-year mean SSH from 10-year CORE2-forced ORCA025 run. Left: NEMO 3.4, right: NEMO 3.6 alpha

[davestorkey]

5-year mean SST from 10-year CORE2-forced ORCA025. Bottom left: SST anomaly for NEMO 3.5, bottom right: SST anomaly for NEMO 3.6, top right: difference 3.6 minus 3.5

[cbricaud]

I have the same issue in my ORCA025 run.
see NEMO_3.6_1990_1999_SSHmSSHmean.gif and NEMO_3.5_1990_1999_SSHmSSHmean.gif.
Clément

[davestorkey]

Gurvan views the change in the formulation of the EEN scheme as a numerical choice rather than a correction and recommended that we add the old version of the scheme as an option in NEMO 3.6 if we want to use it. I have now done this at revision 5029 of the trunk. The default option at 3.6 is still the new formulation.

The formulation of the vector invariant form of the momentum advection scheme in NEMO is the subject of ongoing investigation by Nicolas Ducousso, Mike Bell, Gurvan and others.

[jchanut]

Changes must be duplicated in the time splitting routine.

[davestorkey]

Thanks Jerome I forgot. I've put the changes to dynspg_ts.F90 in at 5032.