GOSI10.beta.6

[ukmo-cguiavarch]

Issue to document the development and assessment of GOSI10.beta.6 release.

GOSI10.beta.6:

• based on GOSI10.beta.5.
• including NEMO update/bug fixes
• RK3

[ukmo-cguiavarch]

Impact of NEMO update/bug fixes

• @dcalve has updated the GOSI10 branch to pick up NEMO updates and bug fixes.
• @isabellaascione has run u-dx333 to test the impact of the NEMO update. u-dx333 is based on GOSI10.beta.5 with the NEMO updates

Results:

• ocean validation notes u-dx333 vs u-dw515
• sea ice validation notes u-dx333 vs u-dw515

The NEMO update mainly impacts the sea ice (as expected as the bug fixes are mainly sea ice related). It results in reduced sea ice in Summer. In particular, there is a reduction in sea ice thickness for the bug fix version is in the area of erroneous landfast ice north of Greenland. Overall there is no unexpected impact and @emmafiedler is currently working on further sea ice updates and tuning (including looking at the landfast ice).

Daley mentioned that the changes in sea ice may be related to the bug fixes included in this issue snow deposition, but @edblockley has since clarified that "those changes are only related to deposition in the coupled model and so should have a null impact on the forced model".

[ukmo-cguiavarch]

Impact of RK3

• @isabellaascione has run u-dx574 to test RK3. u-dx574 is identical to u-dx333 but with RK3.

Results:

• ocean validation notes u-dx574 vs u-dx333
• sea ice validation notes u-dx574 vs u-dx333

Summary:

• RK3 has a large impact in the North Atlantic with a cooling from the surface layer up to 400m in the SPG and a warming below 200m in the Nordic seas (North East of Iceland). It is associated with a warming from the subsurface down to 600m in the tropical Atlantic (temperature atlantic section )
• RK3 also impact the salinity with freshening in the SPG from the surface down to 200m and salinification in Arctic (salinity atlantic section )
• RK3 impact on sea ice seems limited, the main change in an increased in sea ice in the GIN seas but it seems to be a response to the cooling in the ocean rather than a direct change in sea ice (Seasonal cycle of sea ice extent in the GIN seas )
• All those changes are associated with a redcution in AMOC (-2Sv ) and also a change in global temperature.

Those changes are concerning as in previous test with GOSI10.beta.4, we did not see significant impacts (little change in T,S or AMOC see marine assess GOSI10.beta.4 RK3 vs. MLF ). It means that the impact of RK3 is not consistent with the different configurations.

Actions

• @isabellaascione to extent the validation to produce the time series for the full simulations
• @ukmo-cguiavarch to test RK3 with GOSI10.beta.5: u-dy169 currently running under the nemo shared account (https://cylchub/services/cylc-review/cycles/nemo/?suite=u-dy169%2Frun1)
• @dcalve mentioned that with RK3 the recommanded settings for namdyn_spg were nn_bt_alpha=0.09 and nn_bt_flt=3. As we did not change these parameters for the upgrade to RK3, @ukmo-cguiavarch will test the impact of these changes: u-dy170 currently running under the nemo shared account (https://cylchub/services/cylc-review/cycles/nemo/?suite=u-dy170%2Frun1)

[ukmo-cguiavarch]

Impact of RK3 at GOSI10.beta.5 (preliminary results - 20y only)

Results from GOSI10.beta.6 show that RK3 has a larger impact (esp. in the North Atl/Arctic) than at GOSI10.beta.4, we discussed testing RK3 with GOSI10.beta.5 to exclude the NEMO update included in GOSI10.beta.6.

Results:
RK3/MLF comparison for the y15-20 of the simulations GOSI10.beta.5 and GOSI10.beta.6 show that RK3 has a similar impact (the impact is even larger at GOSI10.beta.5).

• RK3/MLF at GOSI10.beta.6 : Assessment of u-dx574_nemo (1973/12/01 - 1978/11/30) vs. u-dx333_nemo (1973/12/01 - 1978/11/30) see Temperature Atlantic cross section
• RK3/MLF at GOSI10.beta.5 : Assessment of u-dy169_nemo (1973/12/01 - 1978/11/30) vs. u-dw515_nemo (1973/12/01 - 1978/11/30) see Temperature Atlantic cross section
• VALNA results (below) show that the AMOC is consistently stronger with MLF (WARNING: only 25y of data)

Summary:

These experiments seems to indicate that the sensitivity to RK3 is not related to the NEMO update applied at GOSI10.beta.6. GOSI10.beta.4 and GOSI10.beta.5 use the same NEMO code so it seems that the namelist changes introduced at beta.5 made the model more sensitive to RK3.

[ukmo-cguiavarch]

Impact of recommended dynspg settings

GOSI10.beta.6 + RK3 dynspg settings vs. GOSI10.beta.6 + RK3: : Assessment of u-dy170_nemo (1973/12/01 - 1978/11/30) vs. u-dx574_nemo (1973/12/01 - 1978/11/30)

The comparison of GOSI10.beta.6 (y15-20) with and without the recommended dynspg settings does not highlight a significant impact, in particular the AMOC has a similar strength in both experiments see VALNA #41 (comment) .

[isabellaascione]

Impact of NEMO update/bug fixes

* [@dcalve](https://github.com/dcalve) has updated the GOSI10 branch to pick up NEMO updates and bug fixes.

* [@isabellaascione](https://github.com/isabellaascione) has run [u-dx333](https://code.metoffice.gov.uk/trac/roses-u/browser/d/x/3/3/3) to test the impact of the NEMO update. [u-dx333](https://code.metoffice.gov.uk/trac/roses-u/browser/d/x/3/3/3) is based on GOSI10.beta.5 with the NEMO updates

Results:

* [ocean validation notes  u-dx333 vs u-dw515](https://gws-access.jasmin.ac.uk/public/jmmp/valid_ocean/u-dx333_vs_u-dw515/assess.html)

* [sea ice validation notes u-dx333 vs u-dw515](https://gws-access.jasmin.ac.uk/public/jmmp/valid_ice/u-dx333_si3_vs_u-dw515_si3/assess.html)

The NEMO update mainly impacts the sea ice (as expected as the bug fixes are mainly sea ice related). It results in reduced sea ice in Summer. In particular, there is a reduction in sea ice thickness for the bug fix version is in the area of erroneous landfast ice north of Greenland. Overall there is no unexpected impact and @emmafiedler is currently working on further sea ice updates and tuning (including looking at the landfast ice).

Daley mentioned that the changes in sea ice may be related to the bug fixes included in this issue snow deposition

Validation notes over the full period 1958-2021 :

• ocean validation notes u-dx574 vs u-dx333

• sea ice validation notes u-dx574 vs u-dx333

• ocean validation notes u-dx333 vs u-dw515

Important :
the run u-dx333 was executed using revision 348143
the run u-dx574 is based on u-dx333 at revision 348143 , with RK3 turned on.

a change to the NEMO namelist was made later in the repository of u-dx333 , see comment #41 (comment)

[dcalve]

Further RK3 vs MLF experiments using GOSI10.beta.5

The idea here is to roll back the changes made between GOSI10.beta.4 and GOSI10.beta.5, to see which of these were responsible for the larger RK3 vs MLF differences in GOSI10.beta.5.

• GOSI10.beta.5 + GOSI10.beta.4 SI3 namelist parameters
  • MLF (u-dy738)
  • RK3 (u-dy739)
• GOSI10.beta.5 + GOSI10.beta.4 SI3 namelist parameters + GOSI10.beta.4 NEMO namelist parameters
  • MLF (u-dy741)
  • RK3 (u-dy742)
• GOSI10.beta.4 + GOSI10.beta.5 code changes up to (and including) the RK3 bug fix for penetrative solar radiation absorption over sea ice (see the NEMO issue)
  • MLF (u-dy743)
  • RK3 (u-dy746)

[dcalve]

I've changed the NEMO namelist to use SST-based river runoff temperatures everywhere (ln_rnf_tem=.false.) rather than at points corresponding to a mask (ln_rnf_tem=.true.).

This is because the temperature variable (rnftemper) in the runoff file (/data/users/nemo/GOSIdev/GOSI10p3/eORCA025/eORCA025_runoff_GO6_icb_r42.nc) is 1 at all points with runoff, i.e. the runoff is input at $1^oC$ everywhere. It should instead have had values of -999 at these points to indicate that the SST be used as the runoff temperature.

This was introduced at GOSI10.beta.5 and would have resulted in cooler SSTs around many rivers compared to GOSI10.beta.4. This can be seen most clearly in the Amazon:

@edblockley, @emmafiedler - As this affects GOSI10.beta.5 onwards, it is unlikely to be responsible for the Ob estuary issues as those were also present in GOSI10.beta.4.

[dcalve]

Presentation of the RK3 vs MLF issue at the NEMO Systems Team meeting on 23/04/26

• Slides
• GOSI10.beta.4 validation note (now on JASMIN)
• GOSI10.beta.5 validation note (now on JASMIN)

[dcalve]

Update on the RK3 vs MLF issue presented to the NEMO Systems Team on 04/06/26

• Slides
• Summary
  • The SI3 parameter changes drive the differences between RK3 and MLF in GOSI10.beta.5 wrt GOSI10.beta.4

• These parameter changes (particularly the landfast sea ice) drive a salinity drift that coincides with a strengthening of the AMOC

• RK3 moderates this impact- hypotheses for why this is the case are:

  • We are seeing chaotic, numerical divergence between MLF and RK3
  • RK3 (more stable) is able to mitigate the “dynamic instability/sensitivity” introduced by the SI3 changes, leading to a weaker AMOC vs MLF

• See the minutes from the recent JMMP meeting for more info