Review of Willeit et al: Surface buoyancy control of millennial-scale variations of the Atlantic meridional ocean circulation

Willeit and colleagues present a large ensemble of CLIMBER-X simulations with various combinations of continental ice sheet configurations and atmospheric CO2 concentrations. This unprecedented ensemble allows them to analyse the physical conditions that determine the forcing range in which CLIMBER-X produces DO-like, millennial-scale climate variability. They find that this "sweet spot" is controlled by the sign of the surface buoyancy flux north of 55N. Millennial-scale transitions between week and strong AMOC states occur when the buoyancy flux north of 55N is about to switch sign. A strong/present day like AMOC occurs when the buoyancy flux is negative and deep water formation takes place in the Labador and Nordic Seas. When the buoyancy flux switches sign, this modern-like deep water formation pattern becomes unsustainable. The conditions under which this sign switch occurs are controlled by the boundary conditions. LGM-like ice sheets tend to enhance buoyancy loss, while low CO2 concentrations tend to decrease it. The balance of the two effects seems to be well captured by CLIMBER-X as the strongest DO-like variability occurs at realistic MIS3-like boundary conditions.

Some of the conclusions are not exactly new, e.g. the cancellation of the effects of ice sheet size and CO2 concentration. However the range and combination of covered boundary conditions is unprecedented, and the results are very relevant for the DO- and wider CP community and thus definitely worthy of publication. It is also very much appreciated that the authors define a metric that could be used to compare the physical conditions that control the "sweet spot" across models. Before publication, I would ask the authors to provide more context in some parts and to address a few issues as outlined in my comments below.

**Major Comments**:

1. Introduction/Discussion: Please provide more context on what has already been suggested in terms of physical control on the sweet spot. At least Galbraith & de Lavergne (2019) and Klockmann et al (2018) provided some suggestions, e.g. the overall volume of Antarctic Bottom Water (AABW) present in the deep ocean, the density difference between AABW and North Atlantic Deep Water, presence of deep water formation in the Nordic Seas. Also spell out more directly how the additional CLIMBER-X simulations can help in pin-pointing the physical control across models. Because the physical control might also be model dependent.

2. I agree that the buoyancy flux analysis in this paper and the one in Klockmann et al (2018) cannot be compared directly one to one but at least a qualitative comparison should be possible and would actually strengthen the authors arguments even further. This could e.g. take place in the Discussion section. 

Overall, the mode transitions in the experiments with PI ice sheets in Klockman et al also occur when net buoyancy flux over their NAtl&LabSea region changes from buoyancy loss to buoyancy gain (Klockmann et al use density instead of buoyancy, so the sign is flipped). In their Nordic Seas region, the buoyancy flux is close to zero for the CO2 range where the transition takes place, so the Nordic Seas would not change the sign. This qualitative agreement makes the suggested metric M in the discussion of this manuscript even stronger. 

One interesting difference can be seen in the effect of ice sheets on the thermal component. In Klockmann et al, the stronger net buoyancy loss with glacial ice sheets is due to increased heat loss over the deep convection sites, while in the present study, it is due to the reduced freshwater input. I do not have an immediate hypothesis where this difference might arise from. Perhaps it is simply due to the different areas of integration. 

3. What is the role of sea ice in the buoyancy flux? Is the effect of freezing/brine release and melting included in the freshwater and heat budgets? Sea ice typically plays a big role in feedback loops regarding convection patterns. Even though it can be difficult to determine whether sea-ice is driving the change in the convection patterns or responding to it, it is still worth to be included more explicitly in the analysis.

4. It might be insightful to show the buoyancy flux also for the equilibrium simulations, e.g. in a similar style as Fig. 3 with buoyancy flux as the colour coding. That would help in linking the results from the transient and equilibrium simulations.

**Minor Comments**

l.3 "latitudinal reach" or "northward extent" instead of "latitude reach"?

l.38-40: see major comment 1

l.45: what is the climate-only setup? Are there other setups?

l.75: How sensitive is the model to the area where the noise is applied? Why is it applied only locally and not globally?

l.115: please also state the temperature changes over Greenland in the simulations and in the reconstructions. What does it mean if Greenland change is not capture well but the Iberian margin yes? 

l.123-124: "The heat transport [...]" What do you base this sentence on? Is it based on previous studies (if yes, please cite)? Or do you infer it from your results (if yes, please elaborate shortly)?

Fig.7: Please correct the caption. The interstadial sea-ice extent is drawn in dark teal and not grey

Fig.8: In the experiment description and in Fig.9 you mention a total of six noise amplitudes. Here you show only four. Why are 0.0625 and 0.125 not shown? Or did you not cover the full CO2 range for these amplitudes? If so please mention this in the experiment description.

Fig.9: Which CO2 concentration was used in the respective simulations displayed here? 

l.152: Please briefly state, how do you define stable here (and elsewhere in the manuscript). Also, how realistic are the deep convection patterns in CLIMBER-X given the very coarse resolution?

l.154: "two modes" instead of "two stable modes". The "stable" in the latter half of the sentence ("are stable under the same CO2") is sufficient.

l.161/Fig.10: what about the smaller oscillations that occur around 160ppm with interglacial ice sheets and around 240ppm with mid-glacial ice sheets? In these cases, the buoyancy flux does not change sign.

l.161/273: Is the Arctic Ocean included in the integral of the buoyancy flux?

Fig.11: What is the averaging period shown here? What does the grey circle around the North pole indicate?

l.169-188: This part is difficult to read with the many "increases" and "decreases". Try to spell out more specifically whether the listed factors induce a buoyancy loss or gain. It can become difficult to correctly interpret increase and decrease if a property (such as M) can have different signs with small or large absolute values. 

Fig.13 and related text: This figure is discussed very briefly, approximately with one and a half sentence. It might be worth to spend a few more words on this figure and to also make the connections between the left half and the right half clearer. Especially because the information in the right half has already been shown in Fig. 10 and 12. Also the relation between hosing and noisy freshwater forcing could be explained some more.

l.194-201: Compare diapycnal diffusivity results to previous work, e.g. diapycnal diffusivity seems to have played a key role in generating the DO oscillations under LGM conditions in Peltier&Vettoretti (2014).

Willeit et al. present and investigate DO-type millennial-scale oscillations from CLIMBER-X simulations. By analyzing North Atlantic surface ocean buoyancy fluxes, the authors provide further insight into the processes controlling convective stability and DO oscillations. The model indicates that transitions between different AMOC states occur when the buoyancy flux in the northern North Atlantic shifts from negative to positive, affecting convection patterns. Factors such ice sheet size, and CO2-induced cooling play crucial roles in stabilizing or destabilizing convection, shedding light on the mechanisms behind abrupt climate changes like DO events.  The investigation of AMOC stability properties presented here is very comprehensive. In addition to the role of ice sheet size and CO2, the effects of climatic noise and ocean diapycnal mixing were also studied. The manuscript is well written, the results are very interesting and I enjoyed reading it very much. In my opinion, the study should be published in CP after the following points have been addressed.

- Previous studies have focused on the role of orbital parameters in DO-oscillations (e.g. Zhang et al. 2021; Kuniyoshi et al. 2022). Willeit et al. used present-day orbital parameters. How does this influence the results? I suggest to add a short discussion.

- CLIMBER-X underestimates the amplitude of Greenland temperature variations. Please discuss possible causes of this shortcoming.

- The surface buoyancy flux analysis is very interesting. However, the authors do not explicitly consider the role of sea ice in controlling surface heat and freshwater fluxes. More discussion on sea ice effects would be necessary.

- The authors describe an important role of the Laurentide ice sheet in “blocking part of the Pacific-to-Atlantic atmospheric moisture transport” (line 183). However, there should be additional effects of the ice sheet on moisture transports, e.g. through weakening of the hydrologic cycle by cooling the atmosphere. Please add further discussion to this topic.

- The authors test the role of ocean diapycnal diffusivity and obtain interesting results. However, the discussion of the results comes up a little short here. Previous studies have explored effects of ocean mixing on AMOC stability. In particular, several studies showed that diapycnal mixing not only strengthens the AMOC but also enhances hysteresis width and the stability of the AMOC (e.g. Nof et al. 2007; Prange et al. 2003; Sijp and England 2006). I suggest to put the CLIMBER-X results into context considering previous work.

- Line 144: “...which cannot be done with GCMs resolving synoptic processes”. Yes, but in principle one could also add noise to the surface fluxes in GCMs. I suggest to rephrase to be more precise.

- Figure 7: Add more information to the figure caption (i.e. which boundary conditions were used in this specific experiment?).

- Equation (D2) in line 266 describes the surface buoyancy flux. I am wondering whether the model uses a real freshwater flux formulation or virtual salt flux. Please clarify.

References:

Sijp, W. P., & England, M. H. (2006). Sensitivity of the Atlantic thermohaline circulation and its stability to basin-scale variations in vertical mixing. Journal of Climate, 19(21), 5467-5478.

Prange, M., Lohmann, G., & Paul, A. (2003). Influence of vertical mixing on the thermohaline hysteresis: Analyses of an OGCM. Journal of Physical Oceanography, 33(8), 1707-1721.

Nof, D., Van Gorder, S., & de Boer, A. (2007). Does the Atlantic meridional overturning cell really have more than one stable steady state? Deep-Sea Research Part I, 54(11), 2005-2021.

Review of Willeit et al.

In this study, the authors explore the relation of surface buoyancy forcing and the initiation of the intrinsic oscillations (or threshold) of the AMOC in their earth system model. For this purpose, they conduct ensembles of simulations varying climatic forcing and atmospheric noise in the model. The model reproduces the modern and the LGM AMOC reasonably well. Also it reproduces the occurrence of the intrinsic oscillation of the AMOC under mid-glacial boundary conditions. The sets of experiments with different magnitudes of noise show the effect of noise in increasing the window of opportunity to cause the AMOC variability. Lastly, the authors explore the role of integrated buoyancy forcing (M) in predicting the initiation of the AMOC variability. Particularly, they show that the threshold type behavior of AMOC occurs when “M” approaches to zero.

I’d like to thank the authors for their effort in running so many exciting simulations. Especially, I find the experiments with different magnitude of noise very exciting since it is technically challenging to do so in AOGCMs! Furthermore, the authors are investigating an important question, “What controls the condition of the sweet spot?/Why DO cycles occur frequently under mid-glacial periods”, which is of highly interest to the readers of Climate of the Past. Therefore, I think these results should be published. However, while the presented figures are exciting, I feel that this study has lots of rooms for improvements in the writing part. For example, in the paper, “M” is introduced in a heuristic way, but the physical reasoning of why M can be a good index is not fully discussed and it is not also compared to existing literatures. Therefore I would recommend major revision. Below summarizes my criticism.

Best wishes,

Sam Sherriff-Tadano

General Comments

1. Why focus only over the North Atlantic?

The authors focuses on the role of buoyancy forcing over the North Atlantic building on their previous work (Ganopolski and Rahmstrof 2001). While I agree that the North Atlantic is a very important region, I’m aware that there are quite a few other studies claiming the importance of the buoyancy forcing or density over the Southern Ocean in controlling the glacial AMOC (Buizert and Schmitner 2015, Sun et al. 2020, Oka et al. 2021). Perhaps, for this particular model, the North Atlantic could be the most critical regions, but there are other modeling studies suggesting the importance of both North Atlantic and Southern Ocean. For example, Sun et al. (2020) showed the importance of density contrast between NADW and AABW, rather than the buoyancy flux itself, in controlling the glacial (LGM) AMOC. I think it would be reasonable to point out these previous studies and then explain why this study focuses only on the buoyancy flux over the North Atlantic.

2. Comparison with previous studies

I appreciate the authors effort in shorting the Introduction and Discussion, however I think the authors are missing important previous studies that tried to answer similar scientific question “What controls the condition of the sweet spot?/Why DO cycles occur frequently under mid-glacial periods“. For instance, previous studies have pointed out the potential importance of Antarctic temperatures (Buizert and Schmittner, 2015; Kawamura et al., 2017), Arctic sea ice (Loving and Vallis, 2005) or changes in surface winds by glacial ice sheets (Sherriff-Tadano et al., 2021a) in initiating the DO-like climate variability frequently during the mid-glacial period. None of the above mentioned studies have explored the role of integrated buoyancy flux over the North Atlantic, but I feel it is beneficial to describe these studies so that the readers can learn some of the history of this research topic.

3. Why is it better to integrate the buoyancy flux over the entire northern North Atlantic?

Here, I’m concerned about the role of sea ice as some of the other reviewers. Previous studies showed the importance of sea ice transport through the Denmark Strait and its melting over the NADW formation region in weakening the AMOC (Born et al. 2010, Vettoretti and Peltier 2018). However, when the buoyancy flux is integrated over the entire region, the spatial heterogeneity in the sea ice-related freshwater flux will be removed. Under this condition, it is not straightforward why M can be a good predictor for the initiation of sweet spot/threshold. Perhaps in this model, I speculate that following two points could be important; 1. Sea ice forms and melts at the same region, hence the sea ice-related freshwater flux is not so important in the first place, or 2. The regional contrast in salinity induced by sea ice formation and melting is removed by advection of salt by oceanic currents. Please discuss why it is better to integrate the buoyancy flux over the northern North Atlantic, rather than focusing over the NADW formation region.

4. Bit more discussion on the role of noise?

Fig. 8d and f reminded me of different characteristics of intrinsic oscillations obtained from CESM(Vettoretti et al. 2022)/MIROC(Kuniyoshi et al. 2022) and MPI (Klockmann et al. 2018). This could be very speculative, but if the authors agree, it might be interesting to point out the potential role of noise in causing different shapes of AMOC variability among models.

Specific Comments

L51: Please describe the climate sensitivity of the model here since it is one of the fundamental metric.

L96-97: Would be suitable to cite Eisenman et al. (2009) and Sherriff-Tadano et al. (2021b) since they discuss the role of changes in atmospheric freshwater flux by ice sheets in intensifying the AMOC.

L209-215: Fig. 10a and b show a threshold type behaviour of AMOC around 160ppm of CO₂ even when the value of M is negative. Is this related to the miss-choice of φ_M? If so, it would be worth discussing it here.

Reference

Born et al. (2010) https://doi.org/10.1007/s00382-009-0709-2

Buizert and Schmitner (2015) https://doi.org/10.1002/2015pa002795

Eisenman et al. (2009) https://doi.org/10.1029/2009pa001778

Ganopolski and Rahmstrof (2001) https://doi.org/10.1038/35051500

Kawamura et al. (2017) https://doi.org/10.1126/sciadv.1600446

Klockmann et al. (2018) https://doi.org/10.1175/JCLI-D-17-0859.1

Kuniyoshi et al. (2022) https://doi.org/10.1029/2021GL095695

Loving and Vallis (2005) https://doi.org/10.1029/2004pa001113

Oka et al. (2021) https://doi.org/10.1038/s43247-021-00226-3

Sherriff-Tadano et al. (2021a) https://doi.org/10.5194/cp-17-1919-2021

Sherriff-Tadano et al. (2021b) https://doi.org/10.5194/cp-17-95-2021

Sun et al. (2020) https://doi.org/10.1175/JCLI-D-19-0546.1

Vettoretti and Peltier (2018) https://doi.org/10.1175/JCLI-D-17-0559.1

Vettoretti et al. (2022) https://doi.org/10.1038/s41561-022-00920-7