Investigate use of Smagorinsky in eORCA025 GOSI10.beta.4 with RK3

[isabellaascione]

Rationale

Smagorinsky is tested in GOSI10.beta.4 with RK3 because the RK3 time-stepping scheme remains numerically stable when the Smagorinsky viscosity is activated, allowing meaningful evaluation of its impact. In contrast, previous tests carried out by Alex Megann at NOC show that the MLF scheme currently crashes with Smagorinsky enabled, preventing reliable testing. Using RK3 therefore provides a robust and controlled framework to assess Smagorinsky behaviour without introducing time-stepping–related instabilities.

The results of the investigation will inform on the use of Smagorinsky in future versions of GOSI10 configurations, most likely GOSI10.beta.5

Setting up of Smagorinsky

from Alex:

Settings from GOSI10p2.0 (based on 4.2.1):

&namdyn_ldf
ln_dynldf_blp=.true.,
ln_dynldf_hor=.false.,
ln_dynldf_iso=.false.,
ln_dynldf_lap=.false.,
ln_dynldf_lev=.true.,
ln_dynldf_off=.false.,
nn_ahm_ijk_t=32,
rn_csmc=3.0,
rn_maxfac=1.0,
rn_minfac=1.0,
/

Mixing diagnostics:

From Alex:
Settings from GOSI10p2.0 (based on 4.2.1):

&namtra_eiv
ln_ldfeiv=.true.,
ln_ldfeiv_dia=.true.,
nn_aei_ijk_t=21,
nn_ldfeiv_shape=2,
rn_le=100.0,
rn_ue=1.5,

`

    <file_group id="5d" output_freq="5d" output_level="10" split_freq="5d" enabled=".TRUE."> 

     <file id="file16" name_suffix="_grid_T" >
      <field field_ref="e3t" long_name="T-cell thickness" />
      <field field_ref="toce_con"     name="thetao_con" operation="average" freq_op="5d"  cell_methods="time: mean (thickness weighted)"   > @toce_con_e3t / @e3t </field>
      <field field_ref="soce_abs"     name="so_abs"       operation="average" freq_op="5d"  cell_methods="time: mean (thickness weighted)"   > @soce_abs_e3t / @e3t </field>
    </file>

    <file id="file18" name_suffix="_grid_V" description="ocean V grid variables" >
      <field field_ref="e3v" long_name="V-cell thickness"  />
      <field field_ref="voce"         name="vo"      operation="average" freq_op="5d" > @voce_e3v / @e3v </field>
      <field field_ref="voce_eiv"     name="voce_eiv" long_name="EIV ocean current along j-axis" />
    </file>

  </file_group>

`

Note: from NEMO5 ln_ldfeiv_dia no longer exists, and the mixing diagnostics are activated simply by adding them in the XIOS input file (file_def_nemo-oce.xml)

Experiments:

All experiments are using eORCA025 based on GOSI10.beta.4

Duration: 1958-2018 (60y)

• Control : u-dr287= RK3 30 min time step, NO Smagorinsky (this experiment is available without mixing diagnostics activated)

New Control: u-dw051 same as u-dr287, therefore science is the same, but with mixing diagnostics activated (experiment still running on 2026-01-15)

• u-dv431 = RK3 30 min time step, WITH Smagorinsky
• u-dv686 = RK3 20 min time step, WITH Smagorinsky
• u-dv512 = RK3 60 min time step, WITH Smagorinsky
• more experiments....

[isabellaascione]

Validation notes

Runs are from 1958-2018 .
Assessment period is 2008-2018

OCEAN

• RK3 30 min time step WITH Smagorinsky vs RK3 30 min time step NO Smagorinsky:

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv431_ocean_vs_u-dr287_ocean/assess.html

• RK3 20 min time step WITH Smagorinsky vs RK3 30 min time step WITH Smagorinsky:

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv686_ocean_vs_u-dv431_ocean/assess.html

• RK3 20 min time step WITH Smagorinsky vs RK3 30 min time step NO Smagorinsky:

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv686_ocean_vs_u-dr287_ocean/assess.html

• RK3 60 min time step WITH Smagorinsky vs RK3 30 min time step with Smagorinsky

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv512_ocean_vs_u-dv431_ocean/assess.html

• RK3 60 min time step WITH Smagorinsky vs RK3 30 min time step NO Smagorinsky:

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv512_ocean_vs_u-dr287_ocean/assess.html

ICE

• RK3 30 min time step WITH Smagorinsky vs RK3 30 min time step NO Smagorinsky:

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ice/u-dv431_si3_vs_u-dr287_si3/assess.html

• RK3 20 min time step WITH Smagorinsky vs RK3 30 min time step NO Smagorinsky:

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ice/u-dv686_si3_vs_u-dr287_si3/assess.html

• RK3 60 min time step WITH Smagorinsky vs RK3 30 min time step WITH Smagorinsky

https://gws-access.jasmin.ac.uk/public/jmmp/valid_ice/u-dv512_si3_vs_u-dv431_si3/assess.html

[isabellaascione]

Summary provided during JMMP monthly meeting 2026-01-15 :

JMMP_2026-01-15.pptx

[oceandie]

One quick comment about the RATIONALE: my suggestion to test RK3 with Smogorinsky was based on Daley's test with NEMO5 default physics (i.e., without GOSI10 specific settings), which showed that, when using Smagorinsky viscosity, simulations with RK60 and RK30 are much more similar and the subsurface heating in the Pacific that we saw was largely eliminated (see https://wwwspice/~daley.calvert/valid_ocean/marine_assess/u-du730_nemo_vs_u-du729_nemo/assess.html) - so the question was whether the same conclusion would apply also to our GOSI10beta4 configuration, and if yes, then this could allow us to use 60 min timestep in GOSI10@1/4 without degrading (too much) the solution.

From the results Isabella showed today, however, it seems that activating Smagorinsky viscosity with RK3 has already a large not negligible negative impact in comparison to the standard GOSI10beta4 with RK3 (see https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv431_ocean_vs_u-dr287_ocean/assess.html) ...

[isabellaascione]

Further plots with the results of the Smagorinsky RK3 60 min experiment
UPDATES_2026-01-16.pptx

[oceanaguiar]

One quick comment about the RATIONALE: my suggestion to test RK3 with Smogorinsky was based on Daley's test with NEMO5 default physics (i.e., without GOSI10 specific settings), which showed that, when using Smagorinsky viscosity, simulations with RK60 and RK30 are much more similar and the subsurface heating in the Pacific that we saw was largely eliminated (see https://wwwspice/~daley.calvert/valid_ocean/marine_assess/u-du730_nemo_vs_u-du729_nemo/assess.html) - so the question was whether the same conclusion would apply also to our GOSI10beta4 configuration, and if yes, then this could allow us to use 60 min timestep in GOSI10@1/4 without degrading (too much) the solution.

From the results Isabella showed today, however, it seems that activating Smagorinsky viscosity with RK3 has already a large not negligible negative impact in comparison to the standard GOSI10beta4 with RK3 (see https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dv431_ocean_vs_u-dr287_ocean/assess.html) ...

Those two experiments that Daley run (NEMO5 default physics) compare results for RK3_DT=30 and 60 min, both using Smagorinsky. Isabella run the same tests/assessment using (GOSI10.beta.4) plus compared the results with and without Smagorinsky: it's in the latter that the widespread T differences arise, the former looks (to me) similar to Daley's results. I don't think we have as yet reason to suspect there are differences between the NEMO5 and GOSI10. However, that doesn't change the fact that we need to understand/address the widespread differences when Smagoringsky is activated.

[AlexM62]

Here are the slides I showed this morning.
GOSI10_beta4 Smagorinsky numerical mixing.pptx