Adding an lbc_lnk changes results in NEMO4.0.4 at ORCA025

[timgraham]

Context

While investigating a restartability issue in our coupled model at NEMO4.0.4 I discovered some unexpected behaviour, where adding an extra lbc_lnk on ub and vn at the end of dynnxt changed the results of the model. As the difference originates along the north fold, I suspect that this is only a problem on the ORCA025 and ORCA12 grids.

Analysis

In SBC ssu_m=ub(:,:,1) and in SI3 u_oce=ssu_m.

The change in the results occurs in ice_dyn_rhg_evp where the variable u_oce is used. u_oceV is calculated as:
u_oceV(ji,jj) = 0.25_wp * ( u_oce(ji,jj) + u_oce(ji-1,jj) + u_oce(ji,jj+1) + u_oce(ji-1,jj+1) ) * vmask(ji,jj,1) in a do loop from j=2,jpjm1. At at least one grid point along the north fold (in processors on the right hand side of the grid) I found that the value of u_oce(ji,jj+1) changes if an lbc_lnk is added on ub.

Recommendation

If it is confirmed that this is a general problem and not caused by a Met Office branch then an lbc_lnk call needs to be included.

Longer term, I wonder if it would be possible to test both types of north fold setup in routine testing.

[clem]

Hi Tim,
I tried to reproduce the issue with orca2 but it failed. Could you tell why your configuration is specific wrt orca2?
nn_fsbc, ln_nnogather...?
Or do you think it is a (mpi) model decomposition issue?

[timgraham]

Hi Clem,
I was thinking that ORCA025 used a different north fold to ORCA2 but looking at the NEMO book again I've realised they are the same (it's just ORCA1 that uses the F point fold). In that case I'm not sure why it would be different in my setup. Just to clarify - the model is now restartable without this change but adding the lbc_lnk changes the results (with or without a restart).
Here are some settings that may be relevant:

jpni=24,
jpnj=25,
ln_nnogather=.true.

nn_fsbc=1

I haven't checked if this is just a problem with the coupled model yet. I'll see if I can test it in our ocean only setup.

[timgraham]

Hi Clem,
We have now confirmed that this is an issue in our ocean only model at ORCA025 as well.
Tim