Past, Present, and Future Variability of Atlantic Meridional Overturning Circulation in CMIP6 Ensembles

Coquereau, Arthur; Sévellec, Florian; Huck, Thierry; Hirschi, Joël J.-M.; Jamet, Quentin

doi:10.5194/egusphere-2025-17

Preprints

https://doi.org/10.5194/egusphere-2025-17

Preprints

27 Jan 2025

| 27 Jan 2025

Past, Present, and Future Variability of Atlantic Meridional Overturning Circulation in CMIP6 Ensembles

Arthur Coquereau, Florian Sévellec, Thierry Huck, Joël J.-M. Hirschi, and Quentin Jamet

Abstract. The Atlantic Meridional Overturning Circulation (AMOC) is a key component of the climate system, exhibiting strong variability across daily to millennial timescales and significantly influencing global climate. Sensitive to external conditions such as freshwater input, greenhouse gas concentrations, and aerosol forcing, important variations of the AMOC can be triggered by anthropogenic emissions. This study presents a comprehensive analysis of sources of AMOC variance in state-of-the-art climate ensemble models. By decomposing the effects of scenario, model, ensemble, and time variability, along with their interactions, through an Analysis of Variance (ANOVA), we identify three distinct regimes of AMOC variability from 1850 to 2100. The first regime, spanning most of the historical period, is characterized by a relatively stable AMOC dominated by internal variability. The second regime, initiated by AMOC decline at the end of the 20^th century and lasting until mid-21^st century, is governed by a transient increase of time variability. Notably, the direct effect of forcing differences remains muted all along this regime, despite the start of emission-scenarios in 2015. The third regime, beginning around 2050, is marked by the emergence and rapid dominance of inter-scenario variability. Throughout the simulations, model variability remains the primary source of uncertainty, influenced by aerosol forcing response, AMOC decline magnitude, and the physical variability. A key finding of this work is the evidence that internal variability decreases simultaneously with AMOC intensity and seems proportional to emission-scenario intensity.

Received: 02 Jan 2025 – Discussion started: 27 Jan 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 9804 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (9804 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

04 Mar 2026

Past, present, and future variability of Atlantic meridional overturning circulation in CMIP6 ensembles

Arthur Coquereau, Florian Sévellec, Thierry Huck, Joël J.-M. Hirschi, and Quentin Jamet

Earth Syst. Dynam., 17, 209–233, https://doi.org/10.5194/esd-17-209-2026,https://doi.org/10.5194/esd-17-209-2026, 2026

Short summary

Arthur Coquereau, Florian Sévellec, Thierry Huck, Joël J.-M. Hirschi, and Quentin Jamet

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-17', Anonymous Referee #1, 30 Jan 2025

This study investigates the variability of AMOC across scenarios, models, ensembles and time in CMIP6 models. The authors applied a novel method, and this part needs other reviewers’ expertise. It is a solid study, but it is difficult to see the key messages, and the implications and discussions of the results should be improved. At least the key points highlighted in the end should be better discussion in the middle sections. Some details are needed – for example, please elaborate on the initial conditions of the model ensemble (L51-53, Table 1), and the reason for incorporating the time dimension is unclear (L71-72).

Citation: https://doi.org/10.5194/egusphere-2025-17-RC1
- AC1: 'Reply on RC1', Arthur Coquereau, 13 Feb 2025
  
  We thank the reviewer for the feedback. We will address each point below and provide a revised version of the manuscript at the end of the interactive discussion.
  1. Key Messages
  We understand the concern of the reviewer about the visibility of our key messages. We will revise the manuscript to better highlight our main findings. Specifically, we will introduce the key points from the "Summary and Discussion" directly in the “Results” section.
  
  2. Initial Conditions
  We agree that further explanation of the initial conditions of the ensemble models is relevant. We will indicate in the revised manuscript that initial conditions are derived following a predefined strategy for the CMIP6 framework, starting with different years of the multi-secular preindustrial control run (known as piControl), which is run under fixed external forcing conditions from the year 1850 (Eyring et al., 2016). This strategy allows us to sample the phase-space and estimate the different possible trajectory evolutions.
  
  3. Time Dimension
  We agree with the reviewer that the introduction lacks an important motivation for incorporating the time dimension: the simple fact that adding the time dimension allows us to study how the temporal variability of the system changes over time. It allows us to study interannual-to-decadal variability through successive 30-year climate periods and estimate the evolution of this interannual-to-decadal variability (including trends). As an example, when the AMOC declines in CMIP6, this variation is a factor of variability for AMOC. However, when time variability is not taken into account, this contribution is not visible. Historically ANOVA has been applied without incorporating the time dimension.This can probably be explained by the fact that ANOVA is often used to characterize uncertainty, and that a trend common to all scenarios, models, and members is not a factor of uncertainty, whereas it is a factor of variability. Hence incorporating the time dimension allows us to generalize ANOVA methodology for variability study. This was mentioned in the “Method” section (l. 111-112), but we will explicitly express this simple idea of taking into account temporal variability and dynamical adjustment in the introduction of the revised manuscript.
  
  References:
  Eyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., and Taylor, K. E.: Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization, Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016, 2016.
  
  Citation: https://doi.org/10.5194/egusphere-2025-17-AC1
RC2:
'Comment on egusphere-2025-17', Anonymous Referee #2, 21 Feb 2025
I miss an interpretation of what I see in Fig. 1. The caption explains what each curve represents, but how to interpret the differences is unclear to me and not (well) discussed in the text.

In the same Fig. different y-axes are used, which is misleading. I would suggest to you use the same y-scale everywhere (the one from panel f), or at least the panel f scale in b,d,f and the panel e scale in a,c,e.

I think it is misleading to talk about the “large ensembles” and the “small ensembles” in many places, suggesting that differences in AMOC variability are due to ensemble size, while it is most probably to different models being analysed and more and less models available in the small and large ensemble. I suggest to use “The models with a large ensemble” vs “the models with a smaller ensemble”, or just ensemble A and ensemble B. This especially holds to the amount of decline. The much larger decline in the small ensemble is simply due to including models with larger AMOC sensitivity while the large ensemble by chance consists of 3 models that do not weaken much. This fact has nothing to do with ensemble size. On the other hand, when discussing the interactions I accept that ensemble size can be driving factor for differences in certain periods.

Please say what R T, S mean in the caption of Fig. 3. Figures should be stand-alone understandable without having to go bac to the text to see what they actually display.

Showing the interactions is of interest but I completely miss a physical picture of how I should interpret these interactions and what physical process they represent. What does it mean? Please explain in the text.

I would rephrase the conclusion between anthropogenic forcing intensity (which could be positive as well in terms of AMOC: aerosols!) and ensemble variance decrease or decrease of natural variability by saying there is a strong link between forced AMOC weakening and decrease in ….(l 318)

Discussion on line 319-320 and further down in 3.2.2. I see what you say in your figures, but don’t have a clear picture of what SRT interaction actually means in terms of physical processes, so this part is not meaningful to me (see also point 5).

My physical interpretation of 2.3 is that up to 2050 scenarios with stronger greenhouse forcing have smaller weakening in aerosol emissions and vice versa, making the impact on AMOC almost scenario independent, until the aerosol effect is gone, and we see the effect of greenhouse gas forcing alone.

Discussion around line 360. I disagree with the interpretation, see point 3.

Summary your point 2. I think what you see in phase 2 is also forced by SSP scenarios, but as said before the net forcing on AMOC is reasonably equal for the scenarios. SSP126 forces stronger AMOC weakening than SSP245 and SSP585 in the first half of this century because the aerosols are faster removed from the atmosphere in this scenario.
Citation: https://doi.org/10.5194/egusphere-2025-17-RC2
- AC2: 'Reply on RC2', Arthur Coquereau, 21 Mar 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-17/egusphere-2025-17-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-17-AC2

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-17', Anonymous Referee #1, 30 Jan 2025

This study investigates the variability of AMOC across scenarios, models, ensembles and time in CMIP6 models. The authors applied a novel method, and this part needs other reviewers’ expertise. It is a solid study, but it is difficult to see the key messages, and the implications and discussions of the results should be improved. At least the key points highlighted in the end should be better discussion in the middle sections. Some details are needed – for example, please elaborate on the initial conditions of the model ensemble (L51-53, Table 1), and the reason for incorporating the time dimension is unclear (L71-72).

Citation: https://doi.org/10.5194/egusphere-2025-17-RC1
- AC1: 'Reply on RC1', Arthur Coquereau, 13 Feb 2025
  
  We thank the reviewer for the feedback. We will address each point below and provide a revised version of the manuscript at the end of the interactive discussion.
  1. Key Messages
  We understand the concern of the reviewer about the visibility of our key messages. We will revise the manuscript to better highlight our main findings. Specifically, we will introduce the key points from the "Summary and Discussion" directly in the “Results” section.
  
  2. Initial Conditions
  We agree that further explanation of the initial conditions of the ensemble models is relevant. We will indicate in the revised manuscript that initial conditions are derived following a predefined strategy for the CMIP6 framework, starting with different years of the multi-secular preindustrial control run (known as piControl), which is run under fixed external forcing conditions from the year 1850 (Eyring et al., 2016). This strategy allows us to sample the phase-space and estimate the different possible trajectory evolutions.
  
  3. Time Dimension
  We agree with the reviewer that the introduction lacks an important motivation for incorporating the time dimension: the simple fact that adding the time dimension allows us to study how the temporal variability of the system changes over time. It allows us to study interannual-to-decadal variability through successive 30-year climate periods and estimate the evolution of this interannual-to-decadal variability (including trends). As an example, when the AMOC declines in CMIP6, this variation is a factor of variability for AMOC. However, when time variability is not taken into account, this contribution is not visible. Historically ANOVA has been applied without incorporating the time dimension.This can probably be explained by the fact that ANOVA is often used to characterize uncertainty, and that a trend common to all scenarios, models, and members is not a factor of uncertainty, whereas it is a factor of variability. Hence incorporating the time dimension allows us to generalize ANOVA methodology for variability study. This was mentioned in the “Method” section (l. 111-112), but we will explicitly express this simple idea of taking into account temporal variability and dynamical adjustment in the introduction of the revised manuscript.
  
  References:
  Eyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., and Taylor, K. E.: Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization, Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016, 2016.
  
  Citation: https://doi.org/10.5194/egusphere-2025-17-AC1
RC2:
'Comment on egusphere-2025-17', Anonymous Referee #2, 21 Feb 2025
I miss an interpretation of what I see in Fig. 1. The caption explains what each curve represents, but how to interpret the differences is unclear to me and not (well) discussed in the text.

In the same Fig. different y-axes are used, which is misleading. I would suggest to you use the same y-scale everywhere (the one from panel f), or at least the panel f scale in b,d,f and the panel e scale in a,c,e.

I think it is misleading to talk about the “large ensembles” and the “small ensembles” in many places, suggesting that differences in AMOC variability are due to ensemble size, while it is most probably to different models being analysed and more and less models available in the small and large ensemble. I suggest to use “The models with a large ensemble” vs “the models with a smaller ensemble”, or just ensemble A and ensemble B. This especially holds to the amount of decline. The much larger decline in the small ensemble is simply due to including models with larger AMOC sensitivity while the large ensemble by chance consists of 3 models that do not weaken much. This fact has nothing to do with ensemble size. On the other hand, when discussing the interactions I accept that ensemble size can be driving factor for differences in certain periods.

Please say what R T, S mean in the caption of Fig. 3. Figures should be stand-alone understandable without having to go bac to the text to see what they actually display.

Showing the interactions is of interest but I completely miss a physical picture of how I should interpret these interactions and what physical process they represent. What does it mean? Please explain in the text.

I would rephrase the conclusion between anthropogenic forcing intensity (which could be positive as well in terms of AMOC: aerosols!) and ensemble variance decrease or decrease of natural variability by saying there is a strong link between forced AMOC weakening and decrease in ….(l 318)

Discussion on line 319-320 and further down in 3.2.2. I see what you say in your figures, but don’t have a clear picture of what SRT interaction actually means in terms of physical processes, so this part is not meaningful to me (see also point 5).

My physical interpretation of 2.3 is that up to 2050 scenarios with stronger greenhouse forcing have smaller weakening in aerosol emissions and vice versa, making the impact on AMOC almost scenario independent, until the aerosol effect is gone, and we see the effect of greenhouse gas forcing alone.

Discussion around line 360. I disagree with the interpretation, see point 3.

Summary your point 2. I think what you see in phase 2 is also forced by SSP scenarios, but as said before the net forcing on AMOC is reasonably equal for the scenarios. SSP126 forces stronger AMOC weakening than SSP245 and SSP585 in the first half of this century because the aerosols are faster removed from the atmosphere in this scenario.
Citation: https://doi.org/10.5194/egusphere-2025-17-RC2
- AC2: 'Reply on RC2', Arthur Coquereau, 21 Mar 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-17/egusphere-2025-17-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-17-AC2

Peer review completion

AR – Author's response | RR – Referee report | ED – Editor decision | EF – Editorial file upload

ED: Reconsider after major revisions (17 Apr 2025) by Gerrit Lohmann

AR by Arthur Coquereau on behalf of the Authors (05 Jun 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (05 Aug 2025) by Gabriele Messori

RR by Anonymous Referee #3 (21 Aug 2025)

Suggestions for revision or reasons for rejection

Review of “Past, Present, and Future Variability of Atlantic Meridional Overturning Circulation in CMIP6 Ensembles”, by Cocquereau et al., second version.

The authors have done a great job in addressing most points raised by reviewers and editor and the paper has accordingly improved.
I still disagree, however, with their response on my points 8 and 10. I recommend acceptance with minor revision, urging them to give a more balanced discussion here.

Main point (3.2.2):
In my view, after 1980-2000 and up to ca. 2050 aerosol removal and increasing greenhouse forcing apear similarly important in driving AMOC decline, with ssp126, ssp245 and ssp585 showing rather similar trends in the early half of the 21st century that only diverge after 2050.
The authors counteract this by noting that the globally averaged radiative forcing of the two is very dissimilar and my argument is not consistent with the evolution of the radiative forcing imbalance diverging quickly after 2015 in the ssp scenarios. I do not dispute this fact, but the authors seem to overlook that the two have a very different spatial pattern in terms of radiative forcing, with aerosols inducing a much more heterogeneous pattern with a strong maximum over the Atlantic NH sector, thus affecting deep mixing in the SPG much more effectively than global warming for the same globally averaged radiative imbalance caused by these two forcings.
They can find ample evidence for this in the two Menary and the Robson papers cited in this ms and maybe could give a look to the Deser et al. paper as well:
https://journals.ametsoc.org/view/journals/clim/33/18/jcliD200123.xml

Minor point
I didn’t understand the last sentence of the abstract. It reads as if internal variability is proportional with emission scenario intensity, but it also decreases proportional to AMOC intensity. The latter, however, is more inversely proportional to emission scenario intensity (I think the authors mean radiative forcing), thus I think IV also is inversely proportional to this (decreasing simultaneously with AMOC strength) and the decrease in IV is proportional to the emission scenario intensity.

Hide

RR by Anonymous Referee #4 (22 Sep 2025)

Suggestions for revision or reasons for rejection

Review of Past, Present, and Future Variability of Atlantic Meridional
Overturning Circulation in CMIP6 Ensembles
Arthur Coquereau, Florian S.vellec, Thierry Huck, Joel J.-M. Hirschi, and Quentin Jamet

This manuscript proposes an analysis of variance of the AMOC in the CMIP6 ensemble of historical and future simulations. The analysis is based on the ANOVA decomposition, previously used with 3 dimensions and here analysed in 4 dimensions. The ANOVA method has already been applied to explain the AMOC spread in terms of intermodal, intermember and inter-scenarios. The main novelty of the paper is to introduce the time dimension as an additional one, so as to assess the evolution of time variability along future scenarios and to compare uncertainty due internal variability as compared to the other sources of spread. Furthermore, one of the strength of ANOVA is to quantify the interactions among the various dimensions
The paper proposes an interesting view on an already well described result which is the spread of AMOC in climate simulations. The approach allows to formalize some assessment and sounds thus promising. Nevertheless, I propose that this paper is strongly revised before being published. My arguments are not so much on the science itself, but more on the way it is presented, argued and related to previous work.
- the manuscript generally does not relate clearly to what was already shown/known in the literature. Some major papers assessing the spread and uncertainty in AMOC simulation are not cited (Buckley et al 2016 https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2015RG000493, Baker et al 2023 https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GL103381, Jackson et al 2023 https://royalsocietypublishing.org/doi/epdf/10.1098/rsta.2022.0187, …, citing here only the main review ones). The fact that the effect of the scenarios on AMOC emerges only some decade after the end of the historical period was for example already discussed in previous studies.
- Seminal papers such as Hawkins et al 2009 are also not discussed neither in the introduction not in the discussion. It should be introduced right at the beginning as the most classical method used to separate various contributions to spread, and then in the conclusion to compare results and advances.
- Adding time variability as an additional dimension is interesting but adds a real complexity as to what is called variability. I strongly advise the authors to check the vocabulary more precisely and possibly use the term “spread” when relating to the models or members dimension in particular.
- Description of the figures and the Conclusions are sometimes very wavy and unclear.
o See for example comments on Fig 2 below. I have had difficulties to follow?
o Other example: L295 I am not convinced with the argument that the different behaviour at the end of the period is due do different sensitivity to external forcings? . Why would the large ensembles exhibit similar sensitivity and not the small ones? Averaging or not internal variability can also lead to very different behavior (see for example Bonnet et al 2021 https://www.nature.com/articles/s41467-021-26370-0 discussing this point). This also somewhat contradicts something that is said later in the text (l448-450) that there is not intrinsic differences between models that have performed large and small ensembles. One way to back up your assertion would be to test more carefully the effect of the ensemble size, by checking the results of large ensemble taking only a subset of members. This is partly done in B5 but I suggest a bootstrapping on more selections of members for more robust results. Finally, l296: I don’t understand the argument of the reference period, please clarify.
o Scenarios and time variability: some elements of the external forcing are known to influence time variability: volcanoes, modulations of the atmospheric aerosols loadings etc. A discussion of how these components are treated in the scenarios and implications for the results is needed.
- The English level is in general quite poor and sometimes makes the understanding pretty hard. I recommend a thorough reading by an English-native. I corrected a few obvious grammar mistakes below, but the improvement required is more general.

Specific comments
L6 «the numerous terms » : which ?
L8-10: what is the difference between what you call “internal variability” and “time variability” (same question in other instances in the text)
L10 forcing differences: do you mean scenarios?
L12: model variability:the models don’t vary. Do you mean “spread”?
L40: the link between AMO and subpolar gyre is not really expalained
L40: change-> changes
L68: I have had a hard time understanding the concept of dimensions here during my first reading. I don’t think it was introduced before. Lease clarify.
L70: some comments on the result of Zhang et al 223, not just their methods, is needed.
l. 75-77 I don’t understand
l93-95: why is this a problem? Model ensembles averaging 1 member / model exist and are found to be robust for certain features. In this case, one does not care which member model is averaged with which. The numbering contains nothing physical
l109-110: I still don’t understand the dimensions at this stage
l139: vary -> varies
l144: why main? The direct effect?
L147: Nr is not introduced I think
L160: (7) is just another formulation of (6) right? I don’t understand the phrasing ‘similarly…’
L161: “described” -> “named after”
L170: the discussion of Hawkins et al and Lehner et al should be moved to the introduction
L187: want -> wants
L195 and around: I don’t understand the articulation between eq (10) and (11). I am expecting that there is one mathematical computation for the variance, so that either (10) or (11) is correct. Here, you seem to say that this decomposition is partly empirical. Please clarify. If (10) overestimates the total variance, how can it be written as an equality?
L208-210: related to that, do you mean here that (11) is not physical? Why? Furthermore, why does (11) prevent from comparison with traditional variance computation? What are the traditional variance computation? Do you mean just R?
L222 more than what?
L223 “phase shift”: what are you referring to here? A phase shift between time series? Which ones? Where do you see that?
L236 is B2 computed also for the synthetic model
L247 is “mixture” the correct term? It sounds a bit like a recipe no?
L258 I am not sure the “first regime” was introduced at this stage
L269: “slow”: do you mean “weak”?
L269: associated with increased aerosol concentration: this is an hypothesis proposed by some studies that you do not demonstrate here. Please temperate the affirmation
L276: studies by Weijet et al, Jackson et al etc should be cited here
L284-285: the first increase that I see occurs in 1930-1050. Is this what you call the 2nd half of the century? And this is not a typical aerosol period. PLease clarify
L290: why do you attribute this later increase in small ensembles to the same causes? How do you know?
L299” model variability”. This term is very confusing. “spread” rather?
L319: “analysis of the model-associated interactions”: where is this done?
L326: “significance of the results”: where is this at all done?
L326 “analysis of the model-associated variability”: you have just said that this is not done here for these small ensembles?
L349 “In this context, internal variability dominates: Over what? Why? I don’t understand
L355 Please specify which panel you are looking at here. Why not commenting the large ensembles as well?
Fig 5 is largely redundant with Fig 3. Isn”t there a way to merge the two?
L371 “This period” -> “the first century of the historical period”
L379 please check the grammar of this sentence
L387: “slight differences”: quantify and be more precise. What type of perturbations are these? In general, discussion of how forcings that can induce variability (volcanoes, aerosols etc) are treated in the scenarios is welcome
L388 “pseudo ensemble members”: I don’t understand (or I am not convinced) why the effect is different from large ensembles?
L418: this sentence is very confusing in my view. Please clarifiy what is declining and increases: AMOC strength, variance ; the effect of time variability on the total variance, …
L430: “be thus directly” -> thus be direcly
L436 and below: “mode associatedl variability” is a misleading phrasing. “variance associated with model differences”, of “inter-model spread”?
L440 “This follows” -> “It follows”
L448-450: see general comment above + are you sure it is due to inherent characteristics of small ensembles? Larger modelling groups generally are the ones having enough resources to run large ensembles… Or run large ensembles on their model version that is more robust, more carefully adjusted etc. R. Knutti et al have probably discussed this idea of model democracy
L450: please recall what “reference” you are refering to here
L451 please check the grammar (and the position of the “,”)
L452: I don”t understand this point and I don’t understand how B3 shows this
L465 what are the first two ways”?
L468-469: a verb is missing in this sentence
L471-472: here as well
L497: I think you mean scenario spread rather than variability. See general comment above.
L 499: please be more temperate: “we argue that tis AMOC decline is therefore a lagged response to…
L 501: do you think there is also a change in the frequency of internal variability?
L525-527: isn’t that point obvious, given that eache simulations are just one realization of the time variability ?

Hide

ED: Reconsider after major revisions (25 Sep 2025) by Gabriele Messori

AR by Arthur Coquereau on behalf of the Authors (05 Dec 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (08 Dec 2025) by Gabriele Messori

RR by Anonymous Referee #3 (05 Jan 2026)

RR by Anonymous Referee #4 (04 Feb 2026)

Suggestions for revision or reasons for rejection

2nd review of Past, Present, and Future Variability of Atlantic Meridional Overturning Circulation in CMIP6 Ensembles Arthur Coquereau, Florian Sévellec, Thierry Huck, Joel J.-M. Hirschi, and Quentin Jamet

Before all, I would like to apologize the fact that I did this review with much delay. Thank you to the authors for considering my comments and addressing most of them. The manuscript has clearly improved. I still have 1 major comment, and a general comment about the level of English. Yet, given my delay, I leave to the editor the decision to let me read again or not the manuscript. Sorry again.
The line numbering refers to the version with the modifications highlighted.

- I am still not convinced by the way the authors treat intermediate period (end of historical up to 2050 roughly):
1. I am not convinced by the modifications of the text to my comment 36. The text does read as if the authors were attributing the variance peak in the 1940s to the aerosols (around line 334), while (i) I don’t think aerosol forcing had really began then (I am ready to be proven wrong), and (ii) the authors acknowledge themselves later in the text (460-462) that “1900-1950 window is … a few decades.. before the increase of the aerosol forcing. “
2. Although I reckon that the manuscript is already relatively long, I think that ideally, the authors should be much more precise in their analysis of the various components of the historical and scenario forcing, the way each forcing acts, regionally and temporally etc. It is also ok not to do that, since many points are already addressed in the manuscript, but in this case, the conclusions should clear been taken down a bit. (l. 570 of the corrected manuscript)
3. I remark that Reviewer#1 also had a major point of the use of the aerosol forcing to justify or not divergence before 2050. While the authors argue that “analyses of the local radiative imbalance over the North Atlantic (10°-60°N, 90°W-0°E) also show an early and rapid separation among scenarios”, I think that their figure proves that this is not true for scenarios ssp 245 and ssp585. This is an illustration of what I consider as a bit of looseness and shortcuts in the attribution of detected changes to scenarios or not.
In general, these remarks relate to a lack of physical analysis to back up some of the hypothesis and conclusions. This may be ok, given the statistical advances proposed in the paper but I think it should be better acknowledged

- My second main comment still concerns the level of English language. I am not native myself, but I had the impression that in several instances, the subtility of the reasoning and argumentation was not well reflected in the text because of too poor language level. I have suggested some changes below, the editorial team may do the rest later

Specific comments
l. 7 components -> contributions ?
l. 36: acknowledge the fact that nevertheless, these paleo-events are not yet fully understood?
l. 44: the regional effect of aerosols should be underlined in my view
l. 58: it is not an improvement f models, but rather an improvement of the protocol or of the CMIP project itself
l. 78: At this stage I still do not understand which dimension you are referring to (This is defined later In section 2)
l. 82: specify ensembles of what?
l. 85: has changed -> changes?
l. 89-90: I would rather say: In the case of the historical period, the time dimension further allows to detect….
l. 103: remove therefore
l. 141: I don’t think both terms “factor” and “effect” are needed here, choose one
l. 165: “at the origin: do you mean “at the beginning”?
l. 177: again, are both terms “effect” and “contribution” needed?
l. 187: “in the paper” -> “hereafter”
l. 202: “this method”: please specify. Do you mean “their method”?
l.208: “especially” -> “at least”
l. 215 “the difficulty of capturing” -> “their difficulty to capture”
l. 232: “would be accounted” -> “are accounted”
l. 234: “would exceed” -> “exceeds”
l. 250-251: “in the present study” goes after “analyze”
l. 262 “appears” -> “can appear”
l. 280: “During” -> “Over”
l. 285 and following: Following one of my previous comment, the authors have changed “mixed” into “combined” in 1 instance. Yet, the verb “mix” is used in several other instances in this paragraph and I still find it misued and confusing. I would consistently change It into “combine” all along.
l. 299: “While in the 21st century, the transfer occurs from…”
l. 300: “I would add “that can be detected” after “variability”
l. 311: the 2 references are linked (authors in common” -> I would remove “often”
l. 321: “used” -> “use”
l. 333 “reached” -> reaches”
Around line 334: As said above, there is no reason to attribute changes to aerosols around 1940, and the signature in small ensembles appears earlier than in large ensembles. Same remark l. 404: why aerosols?
l. 335: “due” -> “linked”?
l. 337: “observed” is misleading, as nothing is observed here. “detected”?
l. 349: I don’t nderstand why specifically the smaller ensemble and the link with the previous sentence
l. 353-355: this sentence is too long, cut it in 2
l. 369This is for example a place where I believe that specific differences among the scenrtios should be discussed
l. 374: “variability” -> “variance”?
l. 375 “In that subsection” until the end of parag: I would remove
Section 3.2.1 and elsewhere in the text: I find the term “ensemble “ a bit misleading to characterize the spread associated to R. Why not “realization”, more intuitive and consistent with R?
l. 408: add “the” before “number”
l. 411: into “the details”
l. 442-443: “Indeed…”: this point has already been made several times
l. 444: I don’t think the “However” is justified”
l. 445: What are these small perturbations, how small are they exactly, due to which forcing etc. Here, we have a very wavy argument
l. 446: “amplification by the chaotic nature of the system”: similarly, this is not shown and not quantified. It is very hypothetical and approximative in my view
Also I am wondering here whether the authors have taken into account the fact that the size of the ensembles differ from historical to scenario periods. Sorry if I missed that.
l. 460-462: This is inconsistent with the argument of aerosol influence on the small ensembles variance in 1940 that was given further up
l. 468: remove “therefore”
l. 483: remove “therefore”
l. 484 “In small ensembles” should come at the beginning of the sentence
l. 493: in my view, the first “variability” should be “variance” and the second one “spread”
l. 495-496: do you mean here that the historical forcings were not anthropogenic? But what about your argument on the aerosols then? Please clarify
l. 502: “same” as what”? The 2-step increase is not clear to me , please clarify
l. 503-505: please add temporal ranges to (i) and (ii)
l. 505: already said several times, consider remove to condense a bit the text
l. 512: any physical clues on the origin of the different AMOC sensitivities?
l. 535 “terms”
l. 538 “aims at representing” -> “represents”
l. 550: isn’t it worth insisting on the fact that physically, the T effect IS DUE TO anthropogenic forcing?
l. 559-560: Similarly, this does not mean that the effect of external forcing is absent, but rather that all scenarios are equivalent. I have the impression this could be misinterpreted. I suggest to clarify.
l. 570: as said above in my general comments, I am not sure this is fully true.
Also, physical proceses of AMOC decline and typical timescale of AMOC response should be considered
l. 582: “facilitates”

Hide

ED: Publish subject to minor revisions (review by editor) (07 Feb 2026) by Gabriele Messori

AR by Arthur Coquereau on behalf of the Authors (19 Feb 2026) Author's response Author's tracked changes Manuscript

ED: Publish subject to technical corrections (23 Feb 2026) by Gabriele Messori

AR by Arthur Coquereau on behalf of the Authors (25 Feb 2026) Manuscript

Journal article(s) based on this preprint

04 Mar 2026

Past, present, and future variability of Atlantic meridional overturning circulation in CMIP6 ensembles

Arthur Coquereau, Florian Sévellec, Thierry Huck, Joël J.-M. Hirschi, and Quentin Jamet

Earth Syst. Dynam., 17, 209–233, https://doi.org/10.5194/esd-17-209-2026,https://doi.org/10.5194/esd-17-209-2026, 2026

Short summary

Arthur Coquereau, Florian Sévellec, Thierry Huck, Joël J.-M. Hirschi, and Quentin Jamet

Viewed

Total article views: 1,344 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
1,055	257	32	1,344	42	62

HTML: 1,055
PDF: 257
XML: 32
Total: 1,344
BibTeX: 42
EndNote: 62

Views and downloads (calculated since 27 Jan 2025)

Month	HTML	PDF	XML	Total
Jan 2025	54	19	1	74
Feb 2025	117	29	4	150
Mar 2025	57	12	2	71
Apr 2025	38	11	2	51
May 2025	38	8	3	49
Jun 2025	28	10	5	43
Jul 2025	48	11	2	61
Aug 2025	87	2	0	89
Sep 2025	375	5	1	381
Oct 2025	35	11	1	47
Nov 2025	45	24	4	73
Dec 2025	25	37	1	63
Jan 2026	64	42	4	110
Feb 2026	38	32	2	72
Mar 2026	6	4	0	10

Cumulative views and downloads (calculated since 27 Jan 2025)

Month	HTML	PDF	XML	Total
Jan 2025	54	19	1	74
Feb 2025	117	29	4	150
Mar 2025	57	12	2	71
Apr 2025	38	11	2	51
May 2025	38	8	3	49
Jun 2025	28	10	5	43
Jul 2025	48	11	2	61
Aug 2025	87	2	0	89
Sep 2025	375	5	1	381
Oct 2025	35	11	1	47
Nov 2025	45	24	4	73
Dec 2025	25	37	1	63
Jan 2026	64	42	4	110
Feb 2026	38	32	2	72
Mar 2026	6	4	0	10

Viewed (geographical distribution)

Total article views: 1,343 (including HTML, PDF, and XML) Thereof 1,343 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 04 Mar 2026

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (9804 KB)
Metadata XML

Short summary

Using statistical methods and a set of ensemble climate models, we decompose the sources of Atlantic Meridional Overturning Circulation (AMOC) variance. Three distinct phases of physical variability are identified: from 1850 to 1990, internal variability dominates; from 1990 to 2050, dynamical adjustment related to AMOC decline takes over; after 2050, differences between forcing scenarios become dominant. Beyond these physical factors, model variability remains the major source of uncertainty.


Total:	0
HTML:	0
PDF:	0
XML:	0