Stream: MOM6-dev

Topic: Diagnostic output

view this post on Zulip Michael Levy (Mar 18 2021 at 15:45):

Does MOM6 have equivalent diagnostics to any of the following from POP output? I'm working out the last kinks in available output from the MOM6 + MARBL compsets, and the following are diagnostics that POP computes that we may need from MOM as well

  1. FvPER_{tracer} and FvICE_{tracer}: I know these are tied to virtual fluxes, which I thought that MOM avoids, so we don't need them for any of the BGC tracers, right?
  2. RF_TEND_{tracer}: these are tied to the Robert Filter and can also be omitted, right?
  3. Horizontal diffusive flux diagnostics (HDIFE_{tracer}, HDIFN_{tracer}, and HDIFB_{tracer}): I assume these exist in MOM, but don't know where to look
  4. Last ones come from vertical mixing: flux across the interface from diabatic implicit mixing (DIA_IMPVF_{tracer}) and tendency from the KPP nonlocal mixing term (KPP_SRC_{tracer})... same as the horizontal mixing, I assume these diagnostics exist but don't know where

view this post on Zulip Keith Lindsay (Mar 18 2021 at 20:38):

It looks like per-tracer diagnostics are registered in subroutine register_tracer_diagnostics. MARBL is using diag_form=1, the default.

  1. These fields do not have an analogue in MOM, because MOM uses natural boundary conditions, so there are no virtual fluxes.
  2. These fields do not have an analogue in MOM, because MOM does not use the Robert filter. (It's kinda a pain to search for Robert in the MOM codebase, because of the many occurrences of "Robert Hallberg".
  3. I thought that MOM's neutral mixing scheme mixing between arbitrary layers in adjacent columns made such analogues not feasible. However, I see {tracer}_dfx and {tracer}_dfyas registered as diffusive zonal flux. So I'm confused. Maybe these aren't supported by the neutral mixing scheme, but are always registered. (I suspect {tracer}_dfy being labeled as diffusive zonal flux is a bug, and that it should be diffusive meridional flux.) I don't think there is an analogue for HDIFB_{tracer} with MOM's neutral mixing. The convergence of these fluxes are in the variables {tracer}_dfxy_conc_tendency and the related fields{tracer}_dfxy_cont_tendency.
  4. I don't see analogues for these in register_tracer_diagnostics or MOM_CVMix_KPP.F90, though the latter does have KPP_NLtransport_heat and KPP_NLtransport_salt, which appear to be analogues specific for temperature and salinity.

view this post on Zulip Michael Levy (Mar 19 2021 at 18:10):

Thanks @Keith Lindsay. For (3), are you suggesting I add {tracer}_dfx and {tracer}_dfy to replace HDIFE_{tracer} and HDIFN_{tracer}, respectively? Would it also be helpful to include {tracer}_dfxy_conc_tendency and {tracer}_dfxy_cont_tendency? It looks like POP includes the HDIF terms for tracers where we want to compute budgets: Fe, O2, DIC, DIC_ALT_CO2, DOC, and DOCr, and from your comment it isn't clear if the convergence terms are necessary.

I'll look into the KPP module this afternoon, thanks for pointing me in the right direction.

view this post on Zulip Andrew Shao (Mar 25 2021 at 16:09):

For the neutral diffusion scheme, the flux at the velocity point of a cell is a quantity that can be calculated. However, a problem in interpretation arises due to the nonlocality of scheme. Unlike other fluxes, the neutral diffusive flux is NOT symmetric between adjacent cells (e.g. the flux defined at the upstream point of j=1,k=1 is not equal to the flux of the downstream point at j+1,k=1.

view this post on Zulip Andrew Shao (Mar 25 2021 at 16:11):

To accurately capture this, we'd need to save 4 sets fluxes for each grid cell making it expensive to store

view this post on Zulip Andrew Shao (Mar 25 2021 at 16:12):

When we were debating this, the 3D convergence of the diffusive fluxes or the 2D column integral of diffusive fluxes seemed the most expedient way to enable most analysis applications

view this post on Zulip Andrew Shao (Mar 25 2021 at 16:14):

This is a long way of saying that dfx_2d and dfy_2d are well-defined and useable quantities, but that the 3D dfx and dfy aren't...we should probably disable the registration of those diagnostics if NEUTRAL_DIFFUSION is on

view this post on Zulip Michael Levy (Mar 29 2021 at 15:27):

Thanks for the additional info, @Andrew Shao. I talked to @Keith Lindsay a week and a half ago, and he pointed out that I was kind of blindly looking for 1-to-1 mappings from POP fields to MOM fields when really we want to make sure we include all the terms necessary to balance budgets... which may be done differently in MOM than in POP. So I think I'm going to wait until the first pass at the driver is on dev/ncar before coming back to these diagnostics

view this post on Zulip Andrew Shao (Mar 29 2021 at 21:29):

Gotcha, yeah I think the only diagnostic which we probably need to add is the dilution/concentration due to boundary freshwater fluxes

view this post on Zulip Keith Lindsay (Mar 24 2022 at 19:27):

FYI, I learned today from Alistair on the MOM6 slack channel that FMS, and hence MOM6, does not support taking time means of diagnostics across model restarts. It doesn't do diagnostic restarts. Some implications:
1) If you're generating monthly/annual means, your restart boundaries need to be on monthly/annual boundaries respectively.
2) We use annual means for some MARBL tracer budget terms, to reduce output volume. So we'll need to be restricted to running in yearly segments, or output these terms at a higher frequency and do a data reduction offline.
3) If you want to directly generate 5-day and monthly means simultaneously, you need to run in yearly segments. If you cannot run in yearly segments, e.g. hi-res runs, the alternative is to directly generate daily and monthly means, and then average the daily means to 5-day means. Alistair said this is what he does, with a sad emoji.

Last updated: May 16 2025 at 17:14 UTC