A tale of heat and climate in NorESM3-dev beta simulations

[adagj]

The current version of NorESM3dev is rather cold, especially over land and at high latitudes. That has not always been the case, e.g. in beta01 the surface climate was much warmer. We want to pin down the model development causing the cooling. To do so we will

• Make a component overview table with beta changes, parameter updates and links to experiments (@TomasTorsvik , @oyvindseland , @matsbn , @kjetilaas )

---

Table on beta changes, parameter updates, links to experiments

(Contact @TomasTorsvik if you need access to edit)

---

• Focus on these key variables and how they have changed with beta versions:

1. PRECIP
2. Meridional slope in temperature in winter
3. Precip over South Pacific and SO
4. Arctic sea ice thickness
5. Antarctic sea ice
6. Land warming
7. OHT
8. SWCF and LWCF

• Make a combined beta11 + beta01 (CAM) simulation to isolate effects of CAM developments @adagj

Please contribute with results in this discussion
@DirkOlivie @kjetilaas @matsbn @TomasTorsvik @oyvindseland @MichaelSchulzMETNO

[kjetilaas]

A quick comparison between our current version and b01. Here is the latitudinal temperature profile (only ne16/LM):

Beta_01 (#195) temperature over land:

And Beta_11 (#338) temp over land:

A couple of quick observations:

• b01 is indeed warmer, but has the same temperature patterns compared to CRU-1900
• It seems too warm in much of the Arctic compared to CRU-1900 and CMIP6mean.
• It has the same SH temperature profile, with way too cold Antarctic.

[adagj]

The impact of ocean albedo schemes. Since Beta01, we started using Taylor albedo over ocean. I tested a simulation (#339) with

ocean_albedo_scheme = 0

in user_nl_cpl
The figures below shows the beta01 albedo (scheme 0) minus Taylor (scheme 1, and what we ar ecurrently using). We can see that going back to the "old" albedo scheme warm the ocean and especially the Arctic, but not a lot

ADF vs obs
ADF vs #332
noresm diag vs #332

[TomasTorsvik]

We are very much lower on zmconv_c0_ocn and a bit lower on clubb_c8 than what we were for beta01. Looking at the PPE heatmap, it looks like these parameters mostly cancel each other out, i.e. an increase in zmconv_c0_ocn can be compensated by an increase in clubb_c8, to maintain a stable RESTOM. Would it be an option to try increasing both of these closer to the beta01 values?

[kjetilaas]

I have been looking closer at the history of our simulations vs lat. temp profile:

This figure is a bit messy, but if you look closer there are a few interesting things to take from this:

• It seems the main change happened between b01 and b02, e.g. between n1850.ne16pg3_tn14.noresm3_0_beta01.20250701 #176 and n1850.ne16_tn14.noresm3_0_beta02-run2_yr001.20250829 #226. I haven't found out what exactly changed here that could have caused this. There are no clear smoking guns on the land side that I could find.
• There is one b01 simulation with a much warmer Arctic (orange "-.", n1850.ne16pg3_tn14.noresm3_0_beta01.20250704 #184). This has quite different parameters, including dust_emis_fact = 4.60D0 (compared to 6.1D0 in the other simulations).

I think we should investigate further what happened between b01 and b02. But maybe we could also test with lower dust_emis_fact? Would that be reasonable, @DirkOlivie?

@adagj @oyvindseland @TomasTorsvik

[kjetilaas]

PS: the SH does not seem to have changed much, so whatever solution we find for the Arctic based on these simulations will probably not solve the Antarctic.

[adagj]

Thanks @kjetilaas . I'm a bit confused... sorry. But do you plot n1850.ne16pg3_tn14.noresm3_0_beta10.Run7.OcnFlxScheme2.20260207 twice? And Run7, is that hidden? Can you exclude the rest of the Ocean flux scheme simulations and only keep n1850.ne16pg3_tn14.noresm3_0_beta10.Run7.OcnFlxSch0.Gust.20260207 ? Maybe it is easier to see the effect. It looks smaller than what the ADF shows, but it is probably not - just hard to separate the lines :-)

[kjetilaas]

Here is a version with fewer cases where I have also tried to align the years with your plot @adag. It looks to me like I have smaller difference in my plot than yours (I don't know why), but I think the big picture is the same. Your Gust simulation warms the model compared to default Run7, but the difference is small compared to the difference between b01 and b02.

[oyvindseland]

Will test the dust parameter. The syntax of the dust tuning is a bit confusing by the way. The scaling of the emissions = 1 / dust_emis_fact so a lower dust_emis_fact = higher emissions.

[TomasTorsvik]

What happened between b01 and b02:

• BLOM version v1.10.0 -> v1.11.4 : ChangeLog
• CAM version noresm3_0_011_cam6_4_085 -> noresm3_0_013_cam6_4_085 : ChangeLog
• CLM version ctsm5.3.045_noresm_v9 -> ctsm5.3.045_noresm_v11 : ChangeLog
• FATES version sci.1.85.1_api.40.0.0_noresm_v3 -> sci.1.85.1_api.40.0.0_noresm_v4 : ChangeLog
• CICE version noresm_cice6_6_0_20250522_v1 -> noresm_cice6_6_0_20250522_v2 : ChangeLog
• MOSART version mosart1.1.08_noresm_v0 -> mosart1.1.12_noresm_v0 : ChangeLog
  
• CMEPS version cmeps1.0.39_noresm_v4 -> cmeps1.0.39_noresm_v6 : ChangeLog

From ChangeLog file:

==============================================================

Tag name: noresm3_0_beta02
Date: 20-08-2025
One-line Summary: New default settings for noresm3_0_beta02 release

Details:
 - Update ccs_config to ccs_config_noresm0.0.51
 - Update BLOM to tag v1.11.4
   - Increase default vertical layers for hybrid coordinates from 56 to 72
   - Adaption of reference potential densities enabled by default
   - Improve CPPM transport method for tracers
   - Change default mixed layer stratification method to Bodner et al. (2023)
   - HAMOCC: Adjust N-cyle parameters and set new default parameters
 - Update CAM to tag noresm3_0_013_cam6_4_085
   - Introduce new history flags to contain all CAM output in NorESM3
 - Update CICE to tag noresm_cice6_6_0_20250522_v2
 - Update CMEPS to tag cmeps1.0.39_noresm_v6
   - Enable COARE algorithm by default for fully coupled NorESM compsets
 - Update MOSART to tag mosart1.1.12_noresm_v0
   - Updates NorESM to use ESCOMP mosart1.1.12
 - Update git-fleximod to v1.0.2

==============================================================

Tag name: noresm3_0_beta01a
Date: 13-08-2025
One-line Summary: Bug fixes for emission-driven runs

Details:
 - Update CMEPS to cmeps1.0.39_noresm_v5
   - Fix field check for fco2 from land
 - Update CLM to ctsm5.3.045_noresm_v11
   - Fix grazing and fire_closs updates
   - New parameter file
   - Update FATES to sci.1.85.1_api.40.0.0_noresm_v4
 - Added NE1850fates-nocomp compset for prognostic co2
 - New test for NE1850fates

[oyvindseland]

As far as I can see the CAM version update is actually CAM version noresm3_0_011_cam6_4_085 -> noresm3_0_014_cam6_4_085
not noresm3_0_013_cam6_4_085

Nevertheless the CAM update should actually not change the answer (bit-identical)

[TomasTorsvik]

@oyvindseland - thanks for checking! The .gitmodules file at noresm3_0_beta02 points to the CAM tag noresm3_0_013_cam6_4_085, so that's what you get with git-fleximod. But in any case, it should not make a difference if the CAM updates are not answer changing.

[oyvindseland]

Sorry you were right. It is just that I tested 014 in the same structure. --> Correct is noresm3_0_013_cam6_4_085 as you wrote

[oyvindseland]

I have tested updating only the CAM tag in beta01 to beta02 and the the results are bit-identical.

[adagj]

Thanks @TomasTorsvik ! I think it is very clear from the log: Enable COARE algorithm by default for fully coupled NorESM compsets

[adagj]

This figure is showing the difference from running with CESM default (in beta10; #340 ) and COARE (in beta10; #332 ) - switching to COARE cools the Arctic more than 1 degC

[adagj]

Link to the ADF https://ns2345k.web.sigma2.no/datalake/diagnostics/ADF/adagj/n1850.ne16pg3_tn14.noresm3_0_beta10.Run7.OcnFlxSch0.Gust.20260207_456_485_vs_n1850.ne16pg3_tn14.noresm3_0_beta10.Run7.2026-01-26_456_485/website/index.html

[adagj]

I have been looking into #194 which is the simulation @matsbn compared with at the latest development meeting, and it says in the med.log (med_aofluxes_init) ocn_surface_flux_scheme is 0 , so it was indeed using CESM default.

[TomasTorsvik]

It seems we have a convincing argument for going back to using ocn_surface_flux_scheme = 0.

[matsbn]

Interesting that ocn_surface_flux_scheme = 0 warms the simulation in the Arctic (maybe also in sea-ice covered regions of Southern Ocean). Over open ocean, however, SSTs are generally lower and comparing the the final 20 years (466-485), global mean surface temperature seems to be 0.18K colder.

"Old" diagnostics here: https://ns2345k.web.sigma2.no/datalake/diagnostics/noresm/matsbn/n1850.ne16pg3_tn14.noresm3_0_beta10.Run7.OcnFlxSch0.Gust.20260207/

[matsbn]

This is adding to the comment #344 (comment) with a completed 30 year simulation, exploring the impact of reverting back to the BLOM configuration as of summer 2025 in otherwise a recent ne16 configuration. Important to note is the BLOM configuration tested keeps the changes that we believe avoided the unrealistic Southern Ocean variability present in beta01 simulations. I will come back to that in later comments.

Years 11-30 has generally much colder surface temperature than #332

Even colder when comparing years 11-30 to the ne16 beta01 spinup (#191):

I guess the conclusions in #344 (comment) still holds.

[lisesg]

SH storm-track activity for the baseline run with ocean_albedo_scheme=0 (#339) compared to the run with ocn_surface_flux_scheme = 0 & add_gusts = .true. (#340):

SH storm-track activity for the baseline run (#339) compared to the run with ocn_surface_flux_scheme = 0 & add_gusts = .false. (#341):

SH storm-track activity for the baseline run (#339) compared to the run with ocn_surface_flux_scheme = 2 (#342):

Summary of area-mean RMSE (values shown in left titles subplots above), red cells indicate that the RMSE values are larger compared to the baseline and blue that they are smaller:

[lisesg]

NH storm-track activity for the baseline run with ocean_albedo_scheme=0 (#339) compared to the run with ocn_surface_flux_scheme = 0 & add_gusts = .true. (#340):

NH storm-track activity for the baseline run (#339) compared to the run with ocn_surface_flux_scheme = 0 & add_gusts = .false. (#341):

NH storm-track activity for the baseline run (#339) compared to the run with ocn_surface_flux_scheme = 2 (#342):

Summary of area-mean RMSE (values shown in left titles subplots above), red cells indicate that the RMSE values are larger compared to the baseline and blue that they are smaller:

[matsbn]

Trying to understand the impact of BLOM changes going from beta01 and later beta releases, I did some additional 30 year long simulations:

• n1850.ne16pg3_tn14.noresm3_0_beta01.blom56layer.20260211 #351: Test the impact of reverting back to the beta01 configuration but keeping the BLOM changes that we believe avoided the unrealistic Southern Ocean variability.
• n1850.ne16pg3_tn14.noresm3_0_beta01.blom72layer.20260211 #352: Test the impact of reverting back to the beta01 configuration but using a beta11 version BLOM with settings of recent simulations.

Comparing surface temperature for years 11-30 of #351 with beta01 (#191), shows much colder Southern Ocean and a slightly colder global mean temperature. Some warming in NH is seen. The impact of the BLOM changes to avoid the unrealistic Southern Ocean (SO) variability gives a large and interannually stable winter sea-ice extent in SO, explaining parts of the cooling seen going from beta01 to later versions, but far from all.

Comparing surface temperature for years 11-30 of #352 with beta01 (#191), shows some of the same as above, but the warming of NH and cooling of SH is enhanced.

When comparing years 11-30 of #352 with years 521-550 of the recent beta10 simulation #332, one gets an impression of the impact of using identical recent BLOM setup, but beta01 and beta10 for other components. Here the SST difference over open ocean is modest, while beta01 is much warmer over land and in the Arctic.

It should be mentioned that short simulations #344, #351 and #352 start from CLM, BLOM, iHAMOCC and CICE initial conditions. This is consistent when comparing to beta01 (#191), while #332 has gone through a long spinup period.

[adagj]

I have transferred 30 years of data for #347 and 10 years of #348 to /nird/datalake/NS9560K/noresm3/cases/ if someone has the time to dig into CAM beta01. I also prolonged the #340 , to see the impact on sea ice. Diagnostics here: ADF and noresm