We are currently testing NEMO 5.0 for various CMIP6 experiments using the global eORCA1 and eORCA025 configurations, with the goal of comparing results against NEMO 4.2.2.
For eORCA1, we were able to run simulations with very similar results (using the same parameterizations) and encountered no major issues.
However, for eORCA025, both in ocean-only (OMIP) and coupled (ESM) simulations, we observed recurring instabilities approximately every 20 years in the Arctic Ocean. These manifested as surface salinity values reaching 100 psu or even “infinite”. When traceable, the source of these unphysical values was often located in the Gulf of Ob.
This issue did not occur in our NEMO 4.2.2 runs. In NEMO 5.0, the problem appeared regardless of whether we used the previous MLF scheme or the new RK3 scheme.
Upon further investigation, we suspect the ice_thd_sal subroutine may be involved[1]. With some modifications to the ice salinity parameterization, we were able to bypass some of the instabilities temporarily, but they eventually reappeared.
The only workaround that consistently stabilized the simulation was to set the maximum sea ice thickness to 15 m (rn_himax = 15.0). While this is less than ideal, it allowed the eORCA025 simulation to run stably, and the global results remained broadly consistent with those from NEMO 4.2.2.
I’m posting here to ask:
Has this behavior been observed by the NEMO development team?
Have other modelling centers experienced similar instabilities with the eORCA025 configuration in NEMO 5.0?
Can I respect different results by using the next minor version (v5.0.1)?
Thanks in advance for any insights or suggestions!
Best regards,
Nicolas Lambert Interim Ocean Group Coordinator Canadian Centre for Climate Modelling and Analysis (CCCma)
The ice salinity seems to be a good suspect because there is an inflow of freshwater in the Gulf of Ob that might interact in an unphysical way with the sea ice. ↩︎
We have encountered cases of salinity > 100 PSU recently but traced it down to negative runoff issues and depth spreading of the river mouth where sss_m or the specified river salinity is not representative of the the bottom cell value. So, not sure it applies to your case…
In our OMIP runs, we indeed have a restoring factor for the sss, but not for our coupled simulations. I’ve been doing some experiments around the treatment of the runoff, but so far I have not succeeded in making the model more stable.
Not sure this helps but just in case… I cannot speak with respect to NEMO5.0, but with NEMO4.0.1 (which also uses SI3), we have had spurious ocean super cooling (instead of sea-ice formation) in the Ob river delta because the ocean has very low salinity there, at least in our setup. In our case, it was related to using single precision in NEMO, and I suspect the problem occurs there, which gets very close to zero divided by zero, if sss_m is close to zero. We do not have problems with double precision though, so it might not be the same problem as you.
Is the ocean salinity very low in that area in your config as well? Lower in eORCA025 than in eORCA1?
Hi, The first configuration used nn_icesal = 2 with rn_sinew = 0.25 . I tried different options related to nn_icesal, and it did not change the outcome.
Can you attach the namelist_cfg and namelist_ice_cfg that you are using? I have just encountered some issues (weird low salinity patches at the surface) with the option ln_shuman=.true. combined with flather lateral boundary conditions in a regional configuration, but it may be unrelated to your issue.
Most of the problems were coming from the Gulf of Ob where a combination of very low salinity (close to zero) and sometimes high sea ice thickness was causing the model to crash on a division by zero. These modifications here seem to have made the model stable without any discernible impacts on the results. So we will use that for our future simulations, unless someone here see issues with our fixes.
Thanks you.
Nicolas Lambert Interim Ocean Group Coordinator Canadian Centre for Climate Modelling and Analysis (CCCma)
@lambertn I am just wondering if the fix in salinity drainage is sufficient to fix your bug or not. Because in theory at_i is always <= 1 and t_i cannot be = rt0
With the fixes, the simulation showed no instabilities for at least 25 years. We did not extend the simulation further, but we plan to complete the full OMIP experiment (310 years) in the upcoming months.
