Atlantic Multidecadal Variability since 1850 is largely externally forced

Liang, Yongyao; Hawkins, Ed; McCarthy, Gerard; Thorne, Peter

doi:10.5194/egusphere-2026-699

Preprints

https://doi.org/10.5194/egusphere-2026-699

Preprints

27 Feb 2026

| 27 Feb 2026

Atlantic Multidecadal Variability since 1850 is largely externally forced

Yongyao Liang, Ed Hawkins, Gerard McCarthy, and Peter Thorne

Abstract. Whether observed Atlantic Multidecadal variability (AMV) is truly an intrinsic internal mode of climate variability or an externally forced response remains contentious, with conflicting literature that North Atlantic SST variability arises from internal dynamics or external forcing. The availability of several single model initial-condition large ensembles (SMILEs) and new insights into potential biases in sea surface temperature (SST) variations offer a fresh opportunity to reassess this question. We show that SMILE ensembles provide strong evidence that AMV-like variability is largely externally forced. New insights into potential SST biases also raise questions about apparent early 20th Century oscillatory behaviour. SMILE models with stronger multidecadal variability show weaker agreement with observed AMV phasing, even in the best performing individual ensemble members, suggesting that large internal model variability may obscure the forced signal. We conclude that future variations in North Atlantic SST will very likely be driven primarily by future anthropogenic activities.

Received: 11 Feb 2026 – Discussion started: 27 Feb 2026

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, 2042 KB)

Supplement (1701 KB)

Download & links

Yongyao Liang, Ed Hawkins, Gerard McCarthy, and Peter Thorne

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-699', Anonymous Referee #1, 27 Mar 2026
This manuscript addresses a highly relevant and still unresolved question: whether the observed Atlantic Multidecadal Variability (AMV) since 1850 is primarily an internally generated mode of variability or largely an externally forced response. The paper argues that AMV-like variability is largely externally forced and that future North Atlantic SST evolution will therefore be driven mainly by anthropogenic forcing. A major strength of the paper is the use of several SMILEs together with a large CMIP6 ensemble, as well as the inclusion of observational uncertainty through different SST datasets. However, I find that the main conclusion is currently stronger than the supporting analysis. Overall, I think the manuscript has clear potential, but substantial revisions are needed. I therefore recommend major revisions. Specific comments are provided below:
Specific comments:
The separation of forced and internal variability is central to the paper. The manuscript appropriately acknowledges that AMV estimates are sensitive to methodological choices. However, the actual sensitivity analysis is limited to linear versus quadratic detrending, so the claim that the main conclusions are robust is currently demonstrated only within these two methods. The authors should either test at least one alternative estimate of the forced signal from a different methodological family, for example a regression-based approach using global-mean SST/ near-surface air temperature (SAT), or provide a stronger justification for why such alternatives are not considered appropriate here. At present, citing one paper regarding the limitations of regression-based methods is useful, but not enough on its own to establish methodological robustness.

The manuscript relies heavily on correlation-based comparisons. However, I could not find a significance framework. Given that the AMV indices are smoothed with a 10-year running mean, the effective number of independent samples is substantially reduced. The authors should therefore explicitly state how significance is assessed and account for serial autocorrelation when interpreting correlation coefficients. An effective-sample-size correction such as the Bretherton lag-1 autocorrelation adjustment (Bretherton et al., 1999) would be appropriate; alternatively, the robustness of the results could also be checked with a bootstrap or Monte Carlo approach that preserves low-frequency dependence. As it is now it is difficult to judge whether the reported differences in correlation across models and highlighted ensemble members are statistically significant.

The choice of the 1961–1990 climatology is not justified. The authors should justify this choice and discuss whether the results are sensitive to an alternative standard reference period (e.g., 1981 -2010).

The manuscript should clarify how much model resolution can realistically matter for the chosen diagnostic. Since all SST fields are remapped to a common 5° × 5° grid and then reduced to a basin-mean NASST index, the potential advantages of higher resolution are not directly assessed.

A more complete evaluation of the AMV should include its spatial fingerprint, not only time-series correlations of indices. In the literature, the positive phase of AMV is associated with a characteristic Atlantic SST pattern, including warm anomalies across much of the North Atlantic and, in some studies, a North-Atlantic warm / South-Atlantic cool contrast, together with broader SAT responses over surrounding continents (e.g., Bellomo et al., 2018; Deser et al., 2010; Otterå et al., 2010; Ruprich-Robert et al., 2017). I suggest that the authors complement their index-based correlation. This could be done with pattern-correlations between regression maps of detrended Atlantic SST and/or SAT onto their AMV index. This quantification would allow a more physically grounded comparison and could substantially strengthen the paper.

Bellomo, K., Murphy, L. N., Cane, M. A., Clement, A. C., & Polvani, L. M. (2018). Historical forcings as main drivers of the Atlantic multidecadal variability in the CESM large ensemble. Climate Dynamics, 50(9–10). https://doi.org/10.1007/s00382-017-3834-3
Bretherton, C. S., Widmann, M., Dymnikov, V. P., Wallace, J. M., & Bladé, I. (1999). The effective number of spatial degrees of freedom of a time-varying field. Journal of Climate, 12(7). https://doi.org/10.1175/1520-0442(1999)012<1990:TENOSD>2.0.CO;2
Deser, C., Alexander, M. A., Xie, S.-P., & Phillips, A. S. (2010). Sea Surface Temperature Variability: Patterns and Mechanisms. Annual Review of Marine Science. https://doi.org/10.1146/annurev-marine-120408-151453
Otterå, O. H., Bentsen, M., Drange, H., & Suo, L. (2010). External forcing as a metronome for Atlantic multidecadal variability. Nature Geoscience, 3(10). https://doi.org/10.1038/ngeo955
Ruprich-Robert, Y., Msadek, R., Castruccio, F., Yeager, S., Delworth, T., & Danabasoglu, G. (2017). Assessing the climate impacts of the observed atlantic multidecadal variability using the GFDL CM2.1 and NCAR CESM1 global coupled models. Journal of Climate, 30(8). https://doi.org/10.1175/JCLI-D-16-0127.1
Citation: https://doi.org/10.5194/egusphere-2026-699-RC1
RC2:
'Comment on egusphere-2026-699', Anonymous Referee #2, 19 Apr 2026
Atlantic Multidecadal variability since 1850 is largely externally forced, Liang et al.

AMV is a dominant mode of climate variability originating in the North Atlantic and influencing the regional and global climate in several ways. While AMV has been initially defined and described as an internal oceanic mode of variability recent literature has put forward several lines of evidence suggesting an important role for external forcing.
This manuscript proposes to explore this long-debated and intensively studied topic . The analysis is based on the use of several single-model large ensemble simulations which ensemble mean and some selected members are compared to the observed AMV. The authors conclude that AMV is primarily forced while strong multidecadal variability may mask the signal.

The manuscript is generally well written and illustrated. Nevertheless, the novelty of the main result and the argumentation of the secondary results (impact of internal multidecadal variability, early 20^th century data issue, ..) are in my view not strong enough for the manuscript to be published in its current state.

Distinction among models variability is not clear and not sufficiently illustrated. I am missing some quantification of frequency band as well as amplitude., and a comparison to observatiions. ‘strong multidecadal variability’ is too vague to make this a proper conclusion in my view.

The abstract is misleading on the assessment of SST biases vs apparent oscillatory behavior in early 20^th I don’t see where this point is assessed in the paper.

SMILEs are not such novel tools, they have been standing for a little decadal now (Deser et al 2020). Concluding on a simple ensemble mean is a little bit outdated in my view. How does this approach relate to the current initiative to isolate external forcing from simulatios and observations (e.g. https://journals.ametsoc.org/view/journals/clim/39/8/JCLI-D-25-0326.1.xml)? Also the difference to multimodel ensemble mean( cf for example IPCC report Fig. 3.40) should be better underlined. Finally, the authors use members selection which is an interesting use of SMILES. I suggest that they push this approach further (e.g. https://www.nature.com/articles/s41467-021-26370-0)

Specific comments
l.17 the sentence [New insights… behaviour] seems misplaced to me, move to after setting all the SMILE-based conclusions. Anyway, this assertion is not addressed in the text in my view. I would remove this sentence from the conclusion.
l.64-66: internal variability is also intensively addressed with control simulations. I think this should be mentioned and discussed here
l.71: Such king of simulations have already been described l. 65-66
l.88-89: the diversity of AMV representation emerging from the various datasets is an important of aspect of the paper. Thus I think the datasets should be briefly described. Most importantly since DCENT is a fairly novel and less standard dataset
l.109: it would not harm to show that masking has a minor effect, especially since the authors partly orientate this manuscript in an analysis of the observations cavevats.
l.114-119: As recognized by the authors themselves, AMV index strongly depends on the detrending method. The authors cite a few previous studies showing that, they should at least acknowledge that there are many other ones. In fact, I think a detailed discussion on what the AMV is supposed to represent, what does it mean to remove the warming trend, when one aims at showing the impact of external forcing? Why only the warming? Why only its linear part? I guess this can be related to possible biases in climate sensitivity of the models. Please discuss.

This manuscript: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022GL097794 about the impact of linear detrending of the AMV should also be discussed in my view.
l.130 : ERSST5 is eliminated because it differs from the other dataset. I don’t think this is a valuable reason, specifically since it was originally included to explore data diversity. If the authors decide to eliminate ERSST5 on such wavy argument, I suggest not mentioning it at all.
l.145 “on the existence of 2 ensemble of models, with limited decadal variability and substantial variability: this distinction should be shown and better explored. What does “limited” and “substantial” mean here? compared to what? Fig. 2 tends to suggest that the difference lies in the frequency more than the amplitude of the variability. Please clarify.
l.145: “in their ensembles”: do you mean ensemble mean? Or each members?
l.148-150: significance of correlations and all statistical estimates is crudely missing
Fig. 2: units are missing in the top panels. This is very important to realize whether the models reproduce the observed AMV magnitude.
l.166: I don’t agree that 0.84 is “considerably lower” than 0.79.
l.191: “This suggests”: I am not sure I agree with the logical link. This “could suggest” but this could also be by chance?
The paragraphs ends on suggesting that because DCENT correlates better with a single member of a climate model, it represents the reality more accurately. Isn’t the argument twisted here?
l.241-244 the sentence is not clear: what role do the models with limited intrinsic multidecadal variability play in the conclusion introduced by “indicating”?
l.253: I don’t understand why models with stronger multidecadal variability might be expected to better capture the observed variability? Perhaps you need to quantify this “stronger” (in terms of amplitude, multidecadal variance / total variance for example) and compare the frequency band with the one detected in observations?
l.266-267: I agree that the AMOC modulates the AMV. Yet in the argument presented by the authors, the AMOC can be modulated by external forcings, then modulating the AMV. I agree with this but I don’t understand how this relates to the general topic of the paragraph “discrepancies between observations and projections”.
Citation: https://doi.org/10.5194/egusphere-2026-699-RC2
EC1: 'Comment on egusphere-2026-699', Gerrit Lohmann, 20 Apr 2026

The manuscript addresses an important and long-standing question, but in its current form, it does not meet the standards required for publication. I agree with the referees that the main conclusion (that AMV is largely externally forced) is stated more strongly than can be supported by the analysis presented.
A key concern is the limited and insufficiently justified methodological framework used to separate internal and forced variability. The study relies heavily on ensemble means and simple detrending choices, without adequately engaging with well-established attribution methodologies. Foundational approaches to this problem, rooted in stochastic climate theory and detection–attribution frameworks are not meaningfully considered. Nor does the manuscript convincingly leverage the full range of model-based strategies available (e.g., control simulations, single-forcing experiments, or formal detection frameworks, or use of long-term data). As a result, the attribution claim rests on a comparatively narrow line of evidence.
In addition, the manuscript is insufficiently situated within the existing literature. A substantial body of work (both earlier and more recent) has addressed AMV attribution using alternative conceptual and methodological approaches (both from data and model side).
For these reasons, major revisions are required. A substantially strengthened methodological framework, clearer positioning within the literature, and more rigorous support for the central claims are necessary before the manuscript can be reconsidered.

Citation: https://doi.org/10.5194/egusphere-2026-699-EC1

Yongyao Liang, Ed Hawkins, Gerard McCarthy, and Peter Thorne

Supplement

https://doi.org/10.5194/egusphere-2026-699-supplement

Yongyao Liang, Ed Hawkins, Gerard McCarthy, and Peter Thorne

Viewed

Total article views: 486 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
281	186	19	486	56	19	31

HTML: 281
PDF: 186
XML: 19
Total: 486
Supplement: 56
BibTeX: 19
EndNote: 31

Views and downloads (calculated since 27 Feb 2026)

Month	HTML	PDF	XML	Total
Feb 2026	66	22	3	91
Mar 2026	91	84	11	186
Apr 2026	124	80	5	209

Cumulative views and downloads (calculated since 27 Feb 2026)

Month	HTML	PDF	XML	Total
Feb 2026	66	22	3	91
Mar 2026	91	84	11	186
Apr 2026	124	80	5	209

Viewed (geographical distribution)

Total article views: 494 (including HTML, PDF, and XML) Thereof 494 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 01 May 2026

Short summary

Using novel climate simulations and historical temperature estimates, we separated natural climate fluctuations from externally driven influences. We find that the timing of long-term temperature swings since the mid-nineteenth century is mainly controlled by external factors, while natural variability mostly affects their strength. We also show that biases in early temperature estimates can complicate interpretation of Atlantic Multidecadal Variability.


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