Thank you for your document “Simulated and Observed Transport Estimates across the Overturning in the Subpolar North Atlantic (OSNAP) Section”. It is a very useful contribution to our understanding and obviously contains a lot of analysis. My recommendations are mostly about exploring some relationships a bit more, putting results into context with some other studies and improving the clarity of the results. My suggestion is for major revisions – I do not think the revisions themselves are particularly serious, but am aware that they might take some time.

Major

1. There are a couple of places where the authors make the point that we don’t know that OW dominates OE for decadal and longer variability (L243, final sentence). This is true, but an obvious question is what do the models show? Do any/many of the models suggest that stronger variability can be seen at OW? Is there a relationship between decadal variability at OW and shorter timescale variability/mean state which might let us infer what the observations might show?
2. You split your models into high and low resolution, but it seems to me that there are 3 groups of about 6 models each – low resolution (>0.25 degrees), medium resolution (~0.25 degrees), and high resolution (<0.25 degrees). Does this split change anything about your conclusions of resolution?
3. Your timeseries plots are difficult to read because it’s difficult to tell some of the colors apart. Some possible options: use different line styles as well as colors; split the models into low, medium and high resolution, with legends for each separately.
4. The paper is quite long, however there is quite a lot of space taken up by description of plots, such as values for particular models etc. This might be important to highlight in some cases, but some parts of the text feel like a long description of the figure. Ideally this information could be read off the plot for someone interested, and more space could be used to discuss the more interesting findings such as: common model biases; differences with resolution; relationships between variables; possible causes and implications of specific differences (eg the authors do discuss the implications of having a better overflow representation in some models); how results fit with other findings. This is done sometimes, but not always. I’ve made some suggestions in the points below.

Minor

L52 Remove ‘general’ – I’m not sure what it means here.

L75 These reviews are not recent anymore!

L88 comma after weakening

L123 I think I’m correct in saying that fluxes may also be important over the Irminger sea as well as the Lab sea?

L125-131 The text implies that all models show a link between low frequency AMOC variability and the Labrador sea which is not true, certainly in coupled models - eg see https://doi.org/10.1007/s00382-023-07069-y

L200 I’m surprised that the authors do not also include a discussion of Yeager et al https://doi.org/10.1126/sciadv.abh3592 , which has a different mechanism for the interaction between the regions

L221 conclude that coupled models

L276 It’s not clear to me why MOM5 and 6 are listed as different models, but the NEMO versions are not.

L286 Define z* or reference something  to explain what it is

L378 I think the dz/dsigma should be inside the dx since the thickness is normally integrated over x as well.

L405 You quote densities as eg 24 kg/m3 rather than 1024 kg/m3 – you need to make this point somewhere.

L465 It looks to me that the black line is plotting the density of the maximum transport rather than a timeseries of the maximum transport itself.

L482 Do you mean annual mean of the maximum over depth of the monthly mean transports? It’s a bit long, but at the moment it isn’t clear what the maximum and average are taken over

L497 Is the weakening trend at OE also seen in most models? How does the overall trend compare to other studies eg https://doi.org/10.1038/s43017-022-00263-2 and references therein

L558 The reference to ‘high densities’ seems to suggest relatively high densities for what we are interested in, however it looks like you’re referring to densities where the transport is maximum here. Suggest rewording

L565 This suggest that in the observations the northwards transports have a narrower range of densities that the models. Is there anything which can be said about why that might be?

L632 What are the deeper/denser levels? Can you say anything about why this might be and the implications?

L650 This isn’t obvious to see from the plots – it looks like the scale is saturated for some plots. Could you do a quantitative comparison, eg comparing the mean of the north/south flows?

L654 maximum aggregated transports

L705 Is there also a relationship of mean strength or monthly variability with annual/decadal variability?

L709 simulated ensemble mean SDs

L710 and elsewhere. The SD should have units of Sv I believe

L732 clarity?

L738 This sounds like you’re saying that the densification is happening on the OSNAP section rather than that the upper/lighter waters are being densified to the north and then returning south. I think this needs a bit of clarification

L743 Why might the models be also freshening as well as cooling unlike the obs? Could this be related to too much mixing in overflows? Or something else?

L769 Note that UKMO25 captures the density though not T and S.

Fig 16 and others. It is has to see the deep values – it looks like the deep values are plotted first and then the shallower ones later, and in some cases it looks like they are plotting over the deep values and obscuring them. It might make the figures clearer to plot the shallower data first and deeper data last.

Section 8. There are a lot of points here which are not explored. There are other studies showing a stronger AMOC in more saline models eg https://doi.org/10.1175/JCLI-D-22-0464.1   https://doi.org/10.5194/os-17-59-2021 https://doi.org/10.1016/j.ocemod.2013.10.005 So discussion about how this seems a robust feature would be welcomed. Also some more discussion about density compensation would be useful. Are there any thoughts about why there is density compensation in some places but not others? What about density compensation in the observations? This paper suggests that there is density compensation for OW https://doi.org/10.1038/s41561-019-0517-1 How does that compare with what you see? Could biases in the models compared with the observations affect the density compensation? Finally you could do with some motivation about why you’re looking at Fig 18 as well as 17 and whether anything can be concluded from considering results together.

Fig 18 The caption is a little unclear. Is the y axis the total AMOC across both sections rather than OW and OE separately?

L886 total velocities -> net transports

L900 Are there differences in biases with resolution?

L929-935 I found this rather confusing – are you always using mean as time-mean and max as max over density? Is the point that you’re making here that the annual time mean of the maximum over density is not the same as the maximum over density of the annual time mean?  

L1109 ‘the same numerical schemes and parameterizations’ – but do they not have different ice models? This sentence implies that the overall models have similar set up other than resolution, but this isn’t true. Maybe just needs a comment here.

Appendix B

It really isn’t clear why the focus here is on two models only, particularly since the authors not that the differences are likely to be model dependent. What is the aim of this section? If it is to document differences then surely it would make sense to plot them all. Otherwise please explain the motivation/aim of the section.

This work provides a valuable benchmark for the modelling community for evaluating simulations at the OSNAP section. It draws on six different ocean GCMs and a wide range of model simulations. All models are consistent with the earlier findings of Lozier et al. (2019), showing that OSNAP East largely determines the strength and variability of the AMOC. This further supports the conclusion that convection in the Labrador Sea may not, as previously thought, be the primary driver of AMOC variability.

Although all models reinforce this central message, the model spread at OSNAP is, unsurprisingly, substantial. The analysis is comprehensive and highly valuable, and the paper is clearly structured and professionally written. The manuscript would be suitable for publication in GMD following the authors’ responses to the minor comments listed below.

line 55. worth mentioning here that they cancel each other at constant depth and density. 

line 116: Do you have any citations for this? I am unsure whether this statement is correct. One can certainly say many things about coupled versus forced AMOC simulations. Xu et al. (2019), if I remember correctly, reported different observations in their paper ‘On the variability of the Atlantic meridional overturning circulation transports in coupled CMIP5 simulations’ (https://doi.org/10.1007/s00382-018-4529-0). Gent (2018) also discussed this issue in ‘A commentary on the Atlantic meridional overturning circulation stability in climate models’ (https://doi.org/10.1016/j.ocemod.2017.12.006).

line 126: the findings by Ortega et al. 2012, 2021 were further confirmed using AWI-CM by Sidorenko et al. in AMOC variability and watermass transformations in the AWI climate model. Journal of Advances in Modeling Earth Systems, 13, e2021MS002582. https://doi.org/10.1029/2021MS002582

line 186: also in FESOM forced ocean by Sidorenko et al. (2020). AMOC, water mass transformations, and their responses to changing resolution in the Finite-volumE Sea ice-Ocean model. Journal of Advances in Modeling Earth Systems, 12, e2020MS002317. https://doi.org/10.1029/2020MS002317

line 188: I am a strong supporter of the work by Megann et al. (2021), which you cite above. Their interpretation is that Subpolar Mode Water formed in the northeastern Atlantic, which initially retains relatively high buoyancy, is advected into the Irminger Sea, where it loses buoyancy, and is subsequently transformed in the Labrador Sea—following further buoyancy loss—into Upper North Atlantic Deep Water before being exported southward. This view is consistent with McCartney and Talley (1982). Megann et al. (2021) also introduced a buoyancy-loss–ramped accumulation index that successfully reproduces the decadal variability of the AMOC, which could be a valuable approach to consider in the proposed follow-up study.

line 353: How would you expect the results to change if instantaneous sampling were used to compute the AMOC in density space?

line 590: If the transports within each bin are provided in Sv, I would suggest plotting the thin blue, red, and black lines in Figures 7–10 in a stepwise manner, since the transports are already aggregated within the respective bins (depth ranges).

line 708: Just a question: to what extent would a more realistic simulation of the Arctic Ocean improve the representation of the OSNAP transports?

line 785: Here you use the depth range above 700 m, whereas in the section above you considered depths deeper than 500 m for the T–S diagrams, in order to exclude the larger surface biases. How would Figures 17 and 18 change if deeper biases were considered instead?

line 834: More generally, in the context of model bias analysis, this suggests that the AMOC may transport water masses with similar (and less biased) densities that nonetheless occupy different regions of T–S space. This behaviour appears to be common across many regions of the global conveyor belt.

lines 874–876: Here, σ0 should also be explicitly mentioned and a reference to Fig. 6 added; otherwise, the reader may become confused.

line 886: The same correction as in the abstract applies here: both flows cancel each other at constant depth and density.

I wonder whether a figure similar to Fig. 2 in Lozier et al. (2019) would show comparable behavior across the models. Such a comparison could potentially serve as a useful benchmark. The same applies to their Fig. 3. Some brief discussion of this point would be nice to have.

This paper presents a coordinated comparison of modelled and observed overturning transports across OSNAP-West and -East sections, using a large number of forced OMIP-type simulations spanning a wide range of ocean/sea-ice models and resolutions. The paper is explicitly framed as descriptive and benchmarking, rather than mechanistic explorations, and is largely successful in achieving that goal. The topic is highly relevant for the ocean and climate modelling community, and it will provide valuable insights into how well ocean models capture the OSNAP overturning circulation in both depth and density space. The paper is overall well-written, and I only have some minor/technical comments.

• Near the end of the Introduction, the authors clearly state that the paper is “largely descriptive”, which is welcome; however, before that, there is heavy text with extensive discussions of, e.g., the control of Labrador vs. Irminger Seas, which might make it read like a hypothesis-driven AMOC mechanism paper. I am not sure if/how this could be improved, though - how about clarifying this somewhere earlier in the text? I am not sure if the authors share a similar view, and I will leave it to them to decide.
• The differences in results between sigma_0 and sigma_2 space (such as the relative contributions from OW) are important findings. For a broader audience, it would be beneficial if the authors, where appropriate, could expand and articulate a bit why using sigma_2 alters the results.
• L47, “...ocean - sea ice simulations…”
• L206, “compensation”
• L344-349, the authors claim that they “analyze the same forcing cycles for a given LR and HR set of simulations from the same group”; however, this does not seem to be the case in the following descriptions. For example, ANU10 (HR) is not in the same cycle as its LR counterparts of ANU1 and ANU25.
• L489, “FSU72”
• Figure 7 and Figure 8 (and other pairs of figures), here the figures are not separated as HR vs. LR (as in previous text and figures), but rather separated alphabetically - is there any rationale behind the change? It would be helpful to explain a bit, or at least mention this change in the text.
• Fig. 18, if I understand correctly, the MOC here is the total, i.e., OW+OE; is there a reason not to separate it, as in Fig. 17? Also, would it be useful to add the OSNAP to the scatter plots, as in Fig. 17?