wrong diagnostic of the mixed layer depth

[cetlod]

The mixed layer depth diagnosed by the model ( hmlp variable ) is underestimated ( minimun value is about 1.7 m for eORCA1 simulations ). However, the turbocline depth ( hmld variable ) used for the computation is correct.

The problem comes from these lines  :

      DO jj = 1, jpj

         DO ji = 1, jpi

            iikn = nmln(ji,jj)  ;   imkt = mikt(ji,jj)

            hmlp (ji,jj) = ( fsdepw(ji,jj,iikn  ) - fsdepw(ji,jj,MAX( imkt,nla10 ) ) ) * ssmask(ji,jj)    ! Mixed layer depth

         END DO

      END DO

Why the MAX( imkt,nla10 ) ( = nla10 without ISF ) ? 

Should not be the MIN( imkt,nla10 ) ?

[mathiot]

You are right, this is a bug. It should be fsdepw(ji,jj,imkt) and not fsdepw(ji,jj,MAX( imkt,nla10 ).

I will modify it in 3.6_STABLE and trunk.

[mathiot]

Fix committed in the revision 5938.
In NEMO_3.6 STABLE, no change in the results because gdepw_1d(2) is larger than 10.0 in GYRE and ORCA config (=> no change in hmlp) and in AMM12 (in 3.6 stable), hmlp is not used in ldfslp (only used as diagnostics).
In NEMO_trunk, solution of AMM12 changed because hmlp is used in ldfslp. The commit for the trunk will arrive during the merge or after (merge revision frozen).

[nicolasmartin]

ORCA1 results has changed, probably as expected by Christian, see the tab ORCA1 for 3.6 on Trusting page (ocean.output on 'SSS_min' value & 'TRACER STAT' section, solver.stat from it. 1 & tracer.stat from it. 3)

• Ada HPC

O1L3P_trust_00000240_restart.nc         : 1629 of 1902 records differ,  1378 of 1902 records differ more than 0.001
O1L3P_trust_00000240_restart_ice.nc     : 257 of 261 records differ
O1L3P_trust_00000240_restart_trc.nc     : 2887 of 3705 records differ,  1913 of 3705 records differ more than 0.001

• Curie HPC

O1L3P_trust_00000240_restart.nc         : 1629 of 1902 records differ,  1249 of 1902 records differ more than 0.001
O1L3P_trust_00000240_restart_ice.nc     : 257 of 261 records differ
O1L3P_trust_00000240_restart_trc.nc     : 3627 of 3705 records differ,  307 of 3705 records differ more than 0.001

Is it regular that we have such differences between Ada and Curie on tracer parameters ? Up to 10-5 on abs max difference.

[mathiot]

Correction of the bug changes results for configuration with surface resolution higher than ~10m (AMM12 in the SETTE test and ORCA1 75L + ...) and if the slopes used by the model depend of the mixed layer depth (ie XXXldf_iso).