Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2026 Apr 17;12(16):eadx4298.
doi: 10.1126/sciadv.adx4298. Epub 2026 Apr 15.

Observational constraints project a ~50% AMOC weakening by the end of this century

Valentin Portmann  1   2 , Didier Swingedouw  1 , Omar Khattab  1 , Marie Chavent  2
Affiliations

Observational constraints project a ~50% AMOC weakening by the end of this century

Valentin Portmann et al. Sci Adv. .

Abstract

Climate models show considerable discrepancies in their future projections around the Atlantic, mainly due to uncertainties in the fate of the Atlantic Meridional Overturning Circulation (AMOC). Climate models suggest a reduction in AMOC strength of 32 ± 37% by 2100 (90% probability, Shared Socioeconomic Pathways 2-4.5 scenario, Coupled Model Intercomparison Project Phase 6). To refine this estimate and reduce its uncertainty, we use four different observational constraint methods. The best one, which provides the lowest leave-one-out error, integrates a large set of observable variables using ridge-regularized linear regression-a method unusual in climate science. It gives an estimate of the AMOC slowdown of 51 ± 8% (90% probability), i.e., a weakening ∼ 60% stronger than suggested by the multimodel mean. This refinement mainly results from correcting a bias in South Atlantic surface salinity, consistent with recent studies emphasizing its role in the proximity to an AMOC tipping point. This more substantial AMOC weakening has key implications for future adaptation strategies.

PubMed Disclaimer

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Fig. 1.
Fig. 1.. CMIP6 annual mean 26°N AMOC.
(A) The AMOC time series. (B) The AMOC time series relative to the 2005-to-2023 means. CMIP6 MMMs (curves) and model uncertainty at one SDs (shading) over different scenarios. The number of models used per scenario is given in parentheses. Only the historical (gray shadow), SSP1-2.6 (blue shadow), and SSP3-7.0 (red shadow) ensemble SDs are shown. These two future scenarios share the same 18 models used for this plot. Each climate model time series is obtained by averaging all its available members (see model name and number of members in tables S1 and S2). The green time series is the RAPID real-world observation between 2005 and 2023. This is an update of Intergovernmental Panel on Climate Change (IPCC) figure 4.6 (66).
Fig. 2.
Fig. 2.. CMIP6 uncertainty decomposition of the future AMOC.
The total uncertainty in the AMOC anomaly at 26°N during the 21st century is decomposed into three sources of uncertainty, based on Hawkins and Sutton (3), using 17 CMIP6 climate models. (A) The decomposition of the annual variance in sverdrup2 of the AMOC for each source. (B) The contribution to the total variance in percent for each source. Changes are relative to the 1970-to-2000 mean. The scenarios used are SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. Only one member per model is considered. This figure can be compared with that of Reintges et al. (15), who used CMIP5 instead of CMIP6.
Fig. 3.
Fig. 3.. Comparison of OC methods applied to the future AMOC.
Comparison of four OC methods with the MMM method applied to the 2091-to-2100 mean AMOC using either a single observable variable (left columns in solid) or multiple observable variables (right columns in dashed). The scenario considered is SSP2-4.5. linear regression stands for unregularized linear regression, while ridge regression stands for ridge-regularized linear regression. (A) Comparison of the estimates of the different methods and the 90% model uncertainty (±1.64 the SD of the method’s error, defined in Eq. 9). (B) Comparison of the mean leave-one-out error (defined in Eq. 10), of the different methods. The definition and difference between the model uncertainty and the leave-one-out error can be found in Materials and Methods. The best method in terms of leave-one-out error is shown in bold.
Fig. 4.
Fig. 4.. Impact of each observable variable of SSS and SST on the constrained estimate of the future AMOC.
(A) SSS. (B) SST. The constraint based on ridge-regularized linear regression corrects the unconstrained estimate of the future AMOC. This correction is decomposed for each of the observable variables and shown here in color. The correction induced by the past AMOC is not shown as it is very small. The exact values are given in the fourth column of table S3. The two most important groups of observable variables are salinity in the South Atlantic and temperature in the subtropical region.
See this image and copyright information in PMC

References

    1. Lemos M. C., Rood R. B., Climate projections and their impact on policy and practice. WIREs Clim. Change 1, 670–682 (2010).
    1. Heaphy L. J., The role of climate models in adaptation decision-making: The case of the UK climate projections 2009. Eur. J. Pilos. Sci. 5, 233–257 (2015).
    1. Hawkins E., Sutton R., The potential to narrow uncertainty in regional climate predictions. Bull. Am. Meteorol. Soc. 90, 1095–1108 (2009).
    1. Strobach E., Bel G., The contribution of internal and model variabilities to the uncertainty in CMIP5 decadal climate predictions. Climate Dynam. 49, 3221–3235 (2017).
    1. E. Ratter, R. D. Rivera-Arriaga, D. Susanto, Swingedouw, L. Tibig, “2019: Extremes, abrupt changes and managing risk,” in IPCC Special Report on the Ocean and Cryosphere in a Changing Climate, H.-O. Portner, D. C. Roberts, V. Masson-Delmotte, P. Zhai, M. Tignor, E. Poloczanska, K. Mintenbeck, A. Alegriia, M. Nicolai, A. Okem, J. Petzold, B. Rama, N. M. Weyer, Eds. (Cambridge Univ. Press, 2022), pp. 589–655.

LinkOut - more resources