Summary

Using OMIP-type simulations with two OGCMs, ICON-O and FESOM, the authors investigate the influence of the vertical mixing parameterization on the simulation of the tropical Atlantic, with both the KPP and TKE schemes tested with several parameter settings. While the models do show sensitivity to the choice of mixing scheme and parameter setting, the choice of model has a larger impact. Furthermore, all simulations show relatively similar bias patterns, including a weaker than observed equatorial SST gradient, which suggests that the vertical mixing scheme may not be the major cause of these biases.

The manuscript is well written and presents a thorough analysis of the performance of the two models. One concern I have is that one year of simulation (the year 2015) may not be sufficient to reliably assess model performance. I presume the high-resolution simulations are too expensive to conduct long-term integrations, but in that case one may wonder if simulations with lower resolution but longer integration time would have been more suitable. Itemized comments follow below.

Major Comments

1) It is not clear whether one year is long enough to reliably assess model performance. Uncertainty in both the observations and the single-realization simulations could be comparable to the biases you are trying to examine. If long-term OMIP simulations at lower resolution are available for ICON-O and FESOM you could check how representative a single year is by comparing the bias of individual years with that of the long-term mean. Alternatively, you could also look at OMIP simulations from other modeling centers to get a rough idea.

2) In the introduction, you refer to previous studies suggesting that the biases in CGCMs and their corresponding OMIP simulations are similar. But the biases you show in the ATL3 region are actually rather atypical, with temperatures lower than observed and the minimum occurring one month early. In many CMIP6 models, on the other hand, the SSTs in the ATL3 are too warm and the minimum is reached too late.

The discrepancy could be due to the high model resolution, as stated by the authors, but this cannot be assessed without a corresponding CGCM simulation. Do the authors have such simulations available?

3) Related to comment 2: How large are the biases in ICON-O and FESOM compared to those seen in typical CMIP6 models? A quick look at the AWI-CM-1-1-MR piControl simulation in the CMIP6 archive, which uses FESOM, suggests that SSTs are about 2K too cold in the western equatorial Atlantic and 2K too warm in the east. Again, it would be instructive to compare the biases in the two OGCMs with their corresponding CGCM simulations.

4) Prigent and Farneti (2024) have recently examined the performance of OMIP simulations in the tropical Atlantic. It should be instructive to compare with their results. Are the biases you see in your high resolution simulation similar to the OMIP simulations?

5) SST biases in the equatorial Atlantic have a strong seasonality, with the most severe bias occurring in JJA. While the seasonality of this bias can be partly inferred from Figs. 5 and 6, it would be helpful to see a longitude-time section of the equatorial SST bias as well.

6) In their conclusions the authors state that vertical mixing can only explain a limited amount of the biases seen in CGCMs, and that biases in atmosphere-ocean interaction likely play an important role. An alternative hypothesis, however, is that AGCM biases are a major source (e.g., Richter and Xie 2008, Wahl et al. 2011; Richter et al. 2012; Voldoire et al. 2018). The results shown here may be consistent with this hypothesis. The authors should discuss this.

7) What do the authors conclude about the prospect of reducing tropical Atlantic biases?

Minor Comments

1) l. 31: The study by Song et al. (2015) could also be cited here.

2) l. 99: Please define EVP.

3) Figures 7 and 8: The depth of the thermocline should be indicated in all panels.

4) Figure 11: The values of the vertical axes should be -0.05 instead of -0.5.

5) Section 6: How can the bulk formula affect SST and the EUC, but not MLD?

References

Prigent, A. and R. Farneti, 2024: An assessment of equatorial Atlantic interannual variability in Ocean Model Intercomparison Project (OMIP) simulations. Ocean Sci., 20, 1067–1086, https://doi.org/10.5194/os-20-1067-2024.

Richter, I., S.-P. Xie, A. T. Wittenberg, and Y. Masumoto, 2012: Tropical Atlantic biases and their relation to surface wind stress and terrestrial precipitation. Climate Dyn., 38, 985–1001, doi:10.1007/s00382-011-1038-9.

Song, Z., S.-K. Lee, C. Wang, B. Kirtman, and F. Qiao, 2015: Contributions of the atmosphere–land and ocean–sea ice model components to the tropical Atlantic SST bias in CESM1. Ocean Modell., 96, 280–296, doi:10.1016/j.ocemod.2015.09.008.

Voldoire, A., E. Exarchou, E. Sanchez-Gomez, T. Demissie, A.-L. Deppenmeier, C. Frauen, K. Goubanova, W. Hazeleger, N. Keenlyside, S. Koseki, C. Prodhomme, J. Shonk, T. Toniazzo, A.-K. Traoré, 2019, Role of wind stress in driving SST biases in the Tropical Atlantic, Clim. Dyn., DOI:10.1007/s00382-019-04717-0, 53(5), 3481-3504.

Wahl, S., M. Latif, W. Park, and N. Keenlyside, 2011: On the tropical Atlantic SST warm bias in the Kiel climate model. Climate Dyn., 36, 891–906, doi:10.1007/s00382-009-0690-9.

Summary:

The authors compare different mixing schemes in two global models to systematically address the effects of using different prescriptions of mixing on mean and seasonal variability of temperature and state variables in the tropical Atlantic. In short, while variability between models is more significant than variability between mixing schemes, the authors find that mixing schemes can affect the representation of smaller-scale phenomena; for instance only the TKE scheme reproduces diurnally-varying deep cycle turbulence. This provides more generalized insight compared to previous studies who focused on single locations with only one model. I think this work would be of interest to modelers and be a good fit in GMD.

Major comments:

1. The model resolution in the Atlantic is 10-13 km (and 50km outside the tropical Atlantic for FESOM). I agree that the resolutions in the region of interest are comparable and don’t suspect that this is a problem, but I think some discussion should be added on this. That is, whether there is sufficient resolution to resolve the spatial variability in this region – While mesoscale features like tropical instability waves are probably resolved, smaller-scale filaments, etc. might not be and it is unclear how this would influence mixing.
2. The current set of metrics used focuses primarily on comparing long-term mean values. While there is some comparison of seasonal variability (for SST), I think it might be important to state how well the models recreate the seasonal cycle, particularly because of strong variability associated with the equatorial current system. For example, section 4.3 discusses how well models represent the mean equatorial current, but does not include information on how well the seasonal variability is represented (even though we know from observations that it is significant). This might also provide insight into the physical mechanisms. (Or, based on the minimal influence on the seasonal cycle of SST, maybe it is similar for all runs. In any case, I think this should be discussed.)
3. I think there should be some numerical quantitative results included in the text. While quantitative results can be deduced from the figures, I think that is not easy for some readers to do. I pointed out a few places that I think quantitative results would improve the text in my line-by-line comments.

Line-by-line minor comments and suggestions:

L1 – I would reword to not state “e.g.” in the abstract. At minimum there should be a comma.

L27-33 – This section maybe can be shortened or removed… The main focus of the paper is mixing, not fixing the atmospheric or ocean parameters, right?

L48-56 – Move 1 paragraph earlier? – because this is a main focus of the manuscript

L62 – delete “for example”

L65 – “specific region” instead of “specific bias”?

L68-69 – Strange syntax. Maybe say something like “Previous studies typically only use a single model and thus it is unclear whether those results are universally applicable”?

L79 – I prefer “in the present study” to “this study”. It is less ambiguous.

L83 – remove “e.g.”

L99 – spell out what EVP is

L100 – what are the boundaries for the “equatorial Atlantic” with 13 km resolution rather than 50km? That would strengthen this argument.

L101 – move this to after you discuss the ICON-O resolution.

L115-116 – “we agreed on” – strange syntax, please reword

L122-125 – This should be elaborated on. I think the reason is because with coarser resolution, smaller-scale instances of shear instability are averaged out and thus mixing occurs at apparently higher Ri.

L129 – Eliminate “we want to”. You are making that comparison now, it is not a future research focus.

L131-144 – This is a very important section and provides motivation for your work. In fact, as I was reading through, I was wondering if there would be any difference if all parameters were the same in both models. Perhaps it would be a good idea to move this earlier to emphasize that point.

L206+ - I’m having a hard time seeing some of the patterns in the figures that you state in the text when scrolling back and forth between the text and figures. I think labeling individual figure panels and referencing them in the text might help make this section more readable.

Another question – for some of the runs, it is clear in Figure 2 that the bias in the equatorial region is significantly different from areas to the north and south. That may be outside the domain of study, but might be something to discuss since it is so obvious in the figures.

L213 – Add justification for averaging between 4N and 4S. I think this is just because of the cold tongue location.

L237-238 – “considerably stronger”. I agree with the result itself, but in this (and other places), I think it would be insightful to include quantitative results. For example, saying 50% stronger (or however much it is) would give some better insight into the differences and I think be a more useful result.

L247 – ITCZ

L245-269 (and earlier..) – Clarification question: Models typically have a warm subsurface bias in the Atlantic cold tongues. Figures 4-6 show that other than in the far eastern Atlantic, SSTs are much colder in all model runs than the observations. Would be helpful to add a sentence or two to clearly state that typical model biases are strongly depth dependent. Maybe this should go earlier in the text (line 64??)

L261 – Why? I think here and in other places, as a reader I would appreciate some speculation on what is causing some of the differences between the runs.

L276 – Remove “actually” – it sounds like you are expressing surprise, but I think we would expect similar results for both models

This is another place where I think numerical, quantitative results would be helpful (yes, they can be deduced from the figures, but I think it is better if they are in the text for readers to easily see).

L283 – Mention why you specifically discuss 23W – because that’s where the PIRATA mooring is

L289-300 – Important information, but maybe this should be in the introduction?

L314-315 – “between 0.1 and 0.2 seems best”- Agreed, but different c_k produce more realistic results between ICON-O and FESOM. Might be helpful to state the specific values that work best in the text for each model.

L350 – “the high stratification band is weaker” – Is this true? From Fig 10 it looks to me that it is thinner, but not necessarily weaker.

L362 – “several reasons” – If it is just those two then say that.

L370-372 – True, but the atmospheric variations have a larger impact on fluxes than the diurnal jet/ocean current because wind varies more. Maybe reword this as to not overstate the impact.

Fig 11 – I like the comparison between obs and the models. But I think the differences between model runs (e.g. discussed starting at L380) are very difficult to see other than in the top center plot. Perhaps a change in the color scale would make it easier to see these changes.

L381 – At different points in the text you use “parametrization”, “parameterization” and “parameterization”. All might be acceptable, but be consistent.

L390 – might be worth mentioning that DCT is diurnally-varying here at the start

L398 – Not sure I completely agree. There are periods where it looks like from Fig 12 that FESOM does not show DCT.

L400-402 – I think a more needs to be added to reconcile these points. The FESOM runs seem to best represent DCT, but the ICON runs (including KPP, which poorly represents DCT) resolve the downward propagation. Aren’t these related, so shouldn’t the same runs represent both well? In other words, doesn’t this imply that the model is reproducing DCT in the FESOM cases for the wrong reason/physics? You state later (and Fig 14 very clearly shows) that the actual K values are closer to the observations in the FESOM cases, so maybe that has something to do with it, but still I think more should be said here.

L407 – Again, it’s more complicated than that. I don’t think there’s a single run that accurately represents all elements of DCT. So I would avoid using “satisfying” or similar terms unless you can quantify what that means.

L425 – move this to line 422? Might be better to first say what is different as a result of different bulk formulae, since I think that is the more important point

On another note, I think it would be helpful to explain a bit more on what terms in the bulk formula might cause these differences, referring again to appendix A.

Another question- I understand what you’re doing is a sensitivity experiment, but isn’t it typically unreasonable to replace bulk formula with those from another model that was tuned to different formulae? Maybe it is worth reiterating that this is a sensitivity test and I_TKE_02_FBF is not meant to get “realistic” results. (Or, I may be wrong here. In that case please clarify.)

L442 – break up into multiple sentences

L477 – this sentence should be quantitative, especially considering the result is different from some past research

L489 – “biases are not sensitive”

General comments for the discussion and conclusion-

1) Overall I think the authors have done a nice job contextualizing their work with previous studies.

2)  A key conclusion is that the differences between models >> the differences between vertical mixing schemes within models. It would be helpful to have some sort of quantitative measure of how much these differences are. This might make it easier to also compare to any previous studies, i.e., Did the previous studies really show a greater difference between mixing schemes? Or did they show quantitatively the same effect and just perceived it as greater because they only looked at 1 model/process/variable? I think that’s an important distinction to make.

3) I’d also like to see a bit of speculation on what physics are different between the models that cause the large differences. To be fair, this is touched on in a few places earlier, but I think a cohesive discussion would be useful.