Bug in zdftke, zdfgls when used with zdfevd convection

[hadcv]

Enhanced diffusion coefficients when using the zdfevd module to boost both avt and avm (nn_evdm=1) cause excessive mixing under convection in the 'new' TKE scheme introduced in NEMO 3.2.

This version was written so that diffusion coefficients from the previous timestep are used to calculate the current TKE. However this inherits the arbitrary zdfevd-boosted coefficients, which greatly increases the production terms of the TKE budget, in places tripling the mixed layer depth.

This is mainly a problem for forced configurations, coupled configurations do not seem to be notably impacted. This is also a problem in zdfgls in theory, though this has not been tested.

Currently, this can be addressed in two ways (see attached figure):

1. Remove the contribution to avt/avm from zdfevd at the start of zdftke

2. Set nn_evdm=0 so that avm is not boosted

We (UKMO) currently follow method 1, but it is not clear whether the contributions from zdftmx and zdfddm should also be removed so that the TKE budget is energy conserving.

[hadcv]

Forced testing at ORCA1 resolution

[rbourdal]

Hi,

There is the same problem in the nemo version 3_4.
The problem is the same if zdftmx and/or zdfddm are actived but the impact is less critical.

Another solution is to use 4 another new 3D arrays to stock the turbulent coefficients needed (avm_zmix, avt_zmix, avmu_zmix, avmv_zmix, all declared in save). And you run tke or gls with these new coeff. At the end of TKE or GLS you make avm = avm_zmix, avt=avt_zmix,...
With this solution tmx and ddm have any impact on the turbulent closure solution.

note: enhanced vertical diffusion option should not be used with gls because stability functions are included in GLS working in instable cases.

[rblod]

We (paris) think tke has to be self-consistent in any case (tmx,ddm). Since it's quiet a huge bug I've fixed it for nemo 3.4 changeset:3406, following Romain suggestion. I haven't change GLS. A more extensive fix could be to remove avmu,avmv and recompute it when used (or have fortran substitute), in order to save memory. So I let the ticket open for now.

[rbourdal]

add same modifications than TKE to have any impact of evd,tmx,ddm on the turbulent closure solution

Committed revision 3798

See changeset:
https://forge.ipsl.jussieu.fr/nemo/changeset/3798/branches/2012/dev_MERGE_2012

[rbourdal]

and committed revision 3804 in:
​http://forge.ipsl.jussieu.fr/nemo/changeset/3804/branches/2012/dev_v3_4_STABLE_2012