Claim vs evidence
'95% chance the AMOC collapses': what the number means
No study we read says there is a 95 percent chance that the Atlantic overturning circulation (AMOC) collapses before 2100. The widely repeated '95%' is the width of an uncertainty interval around an estimated collapse year in one 2023 paper, which was corrected in 2025 (current text: central estimate 2065, interval 2037-2109). The IPCC has medium confidence that an abrupt collapse will not happen before 2100, while weakening is very likely. Direct measurements at 26 N show strong variability and a decline since about 2010; they are not a forecast.
What is known, forecast and unknown
- Observed: RAPID 26 N array, 2004-2024, shows a decline since about 2010-11 and strong variability (rapid.ac.uk); the 2005-2023 mean transport is 16.9 Sv (as stated in Portmann et al. 2026).
- Observed: North Atlantic sea-surface temperature records since 1870 (used indirectly in the 2023 paper).
- Model-based, not observations: IPCC AR6 (weakening very likely; no abrupt collapse before 2100, medium confidence) [a4]; Baker et al. 2025 (34 models: weakening 20-81 percent, mean 54 percent, no collapse; collapse unlikely this century) [a6]; Portmann et al. 2026 (51 +/- 8 percent weakening by 2091-2100, SSP2-4.5) [a7]; Ditlevsen and Ditlevsen 2023 (statistical extrapolation: tipping year 2065, 95 percent interval 2037-2109) [a2]. These studies differ in method and in how close they put a tipping point.
- Whether sea-surface temperature tracks the AMOC well enough to date a collapse.
- How close the real AMOC is to a tipping point (models disagree and may be biased toward stability).
- Whether the 2026 committed-collapse preprint holds up: not read.
What the AMOC is
The AMOC is a system of Atlantic currents that carries warm water north near the surface and cold water south at depth. It is not the same thing as the Gulf Stream, which is wind-driven [a1].
Where '95%' comes from
Ditlevsen and Ditlevsen (2023) used North Atlantic sea-surface temperatures since 1870 as an indirect signal of the AMOC and extrapolated a trend toward a tipping point. In the current text the central estimate is 2065 with a 95 percent confidence interval of 2037-2109 [a2]. A 95 percent interval is a range for the year, not a 95 percent probability of collapse. The paper names no emissions scenario and assumes a steady drift; it says it cannot rule out other mechanisms [a2].
The 2025 correction
In August 2025 the authors corrected a code error that changed the tipping-time estimate by 8 years; figures, a table and estimates in the text were updated [a3]. The earlier coverage quoted about 2057, consistent with that shift, but the notice itself does not list the old and new values.
What other work says
The IPCC's sixth assessment has medium confidence that there will not be an abrupt collapse before 2100, and says weakening is very likely [a4]. A 2024 model study simulated a tipping event and argued the present AMOC is on a tipping course; its methods are debated [a5]. In 34 climate models forced by an abrupt quadrupling of CO2, the AMOC weakened by 20 to 81 percent (mean 54 percent) 90 years after the forcing but did not collapse; the authors conclude a twenty-first-century collapse (weakening below 6 Sv) is unlikely [a6].
A 2026 study (Portmann et al.) uses observations to constrain climate models and estimates a weakening of 51 +/- 8 percent (their 90 percent interval) by 2091-2100 relative to 1850-1900 under the SSP2-4.5 scenario, against 32 +/- 37 percent unconstrained. That is a weakening, not a collapse, and the authors caution that the interval rests on strong assumptions [a7]. A 2026 preprint on committed collapse is not peer reviewed and was not read [a8].
What is observed
The RAPID array at 26 N has measured the overturning since 2004 and shows variability from days to a decade, with a sustained decline after 2010-11 [a9]. The proxy used for the 2023 estimate (sea-surface temperature) is questioned by other scientists [a10].
Claims and ratings
Each statement is rated against the named source on the date it was read. The rating describes the statement, not the person who made it. “Forecast as issued (not rated)” means the source is shown, with no verdict.
- a1Verdict: confirmed
The AMOC is not the Gulf Stream
- Said by
- Susanne Ditlevsen in a Q&A (via search)
- What the source says
- coverage conflated the two
- Source
- Q&A with the authors (Down To Earth)
Source date: 2023-08 · Retrieved: 2026-10-30 - How it was read
- search summary only; page opened on 2026-10-30 (HTTP 200, title confirmed)
- a2Verdict: unsupported as stated (an interval for a year, not a probability)
95 percent chance the AMOC collapses before 2100
- Said by
- a video explainer; repeated in media
- What the source says
- Abstract: 'We estimate a collapse of the AMOC to occur around mid-century under the current scenario of future emissions'; the transition is 'most likely to occur around 2037-2109 (95% confidence interval)'. Maximum-likelihood method (Strang splitting): tipping year 2065, 95% interval 2037-2109. A second, moment-based method gives 2057. Table 1 (sea-surface-temperature proxy minus 1x, 2x or 3x the global mean): 2123 (2075-2288), 2065 (2037-2109), 2068 (2037-2101). The paper assumes its model is approximately correct, says 'we cannot rule out that other mechanisms are at play', does not exclude a partial collapse, and uses a sea-surface-temperature proxy. The 95 percent is the confidence interval of the estimated tipping year, not a 95 percent probability of collapse.
- Source
- Warning of a forthcoming collapse of the AMOC, Nature Communications 14, 4254
Source date: 2023 (text version after 2025 correction) · Retrieved: 2026-10-30 - How it was read
- full text opened (PMC10368695) and the corrected-publication PDF read by the PM (user-supplied); figures and supplement not read
- a3Verdict: confirmed (notice read; 'earlier press values' from the user's note)
The 2023 collapse estimate was corrected; 2057 and 2025-2095 are the earlier figures
- Said by
- authors; earlier press coverage
- What the source says
- Author Correction (Nat Commun 16:7794, 2025-08-21): a coding error in the last step of the Strang-splitting estimator moved the maximum-likelihood estimate by 8 years; Figs. 5-7, Table 1 and the estimates in the text were updated. 2057 now belongs to the moment-based method; 2057 and 2025-2095 are the values in the 2023 press coverage (press release not opened). The notice lists no old or new values.
- Source
- Author Correction: Warning of a forthcoming collapse of the AMOC
Source date: 2025-08-21 · Retrieved: 2026-10-30 - How it was read
- correction notice opened in full (PMC12370995); the PM saw the 8-year statement on the Nature correction page; the uncorrected PDF and the 2023 press release were not opened
- a4Verdict: confirmed (secondary quote)
The IPCC rates abrupt collapse before 2100 as not expected (medium confidence); weakening very likely
- Said by
- IPCC AR6, as quoted by Drijfhout et al. 2025
- What the source says
- 'medium confidence that there will not be an abrupt collapse before 2100'; abrupt = a few decades or less
- Source
- IPCC AR6 WG1 as quoted in Drijfhout et al. 2025 (Environmental Research Letters, DOI 10.1088/1748-9326/adfa3b)
Source date: 2025 · Retrieved: 2026-10-30 - How it was read
- AR6 quote seen in a search summary of this paper; the paper page opened on 2026-10-30 (HTTP 200, title confirmed) but its text was not read; AR6 chapter not opened
- a5Verdict: partly supported (model-based, contested)
Physics-based early warning: the AMOC is on a tipping course
- Said by
- van Westen, Kliphuis, Dijkstra
- What the source says
- tipping simulated in one model (CESM); indicator from freshwater transport at the Atlantic southern boundary; reliability of methods debated
- Source
- Physics-based early warning signal shows that AMOC is on tipping course, Science Advances 2024
Source date: 2024-02 · Retrieved: 2026-10-30 - How it was read
- search summary only; the arXiv PDF link answers (HTTP 200, application/pdf, 2026-10-30) but was not read; arXiv version differs from the Science Advances version
- a6Verdict: confirmed (full text read by the PM)
A complete AMOC collapse this century is unlikely
- Said by
- Baker et al., Nature 638, 987-994, published 26 Feb 2025 (corrected publication 2025)
- What the source says
- In 34 CMIP6 models forced by an abrupt quadrupling of CO2 held for 150 years, the AMOC weakened by 20 to 81 percent (mean 54 percent) 90 years after the forcing and 'does not collapse in the model experiments considered here'. In the seven models that also ran a 0.3 Sv freshwater experiment, the weakening was 50 to 80 percent in both experiments (means 56 and 61 percent). Southern Ocean wind-driven upwelling sustains the weakened AMOC; a Pacific overturning circulation emerges in almost all models but is too weak to balance it. Conclusion: 'a twenty-first century AMOC collapse (defined here as weakening to below 6 Sv) is unlikely'. The '90 years' are model years after the forcing in an idealised experiment, not calendar years.
- Source
- Continued Atlantic overturning circulation even under climate extremes (Baker, Bell, Jackson, Vallis, Watson, Wood)
Source date: 2025-02-26 · Retrieved: 2026-10-30 - How it was read
- full text read by the PM from the user-supplied PDF (21 pages); DS did not open the article
- a7Verdict: confirmed (PDF read by the PM)
Observational constraints project about 50 percent AMOC weakening by the end of the century
- Said by
- Portmann et al., Science Advances 12, eadx4298, 15 April 2026
- What the source says
- Abstract: climate models suggest 32 +/- 37 percent weakening by 2100 (90 percent probability, SSP2-4.5, CMIP6); the best observational constraint (ridge-regularized regression on 19 observed variables) gives 51 +/- 8 percent, about 60 percent stronger than the multimodel mean. Table 1 (SSP2-4.5, 32 models): constrained 2091-2100 AMOC 8.1 +/- 1.4 Sv against a constrained 1850-1900 reference of 16.4 +/- 1.7 Sv; weakening 50.7 +/- 8.3 percent (unconstrained 32.0 +/- 36.7). Constrained, other scenarios: SSP1-2.6 19.5 +/- 25.4, SSP3-7.0 48.4 +/- 5.0, SSP5-8.5 57.7 +/- 5.4 percent. 'Weakening' is the 2091-2100 mean relative to 1850-1900. The authors say the 90 percent interval 'does not rigorously represent a probability interval' and 'should be treated with caution' (models treated as independent and equally plausible, observation error neglected). RAPID 2005-2023 mean: 16.9 Sv. An estimate of weakening, not a probability or a date of collapse.
- Source
- Observational constraints project a ~50% AMOC weakening by the end of this century
Source date: 2026-04-15 · Retrieved: 2026-10-30 - How it was read
- 16-page PDF read by the PM (user-supplied); DS did not open the article
- a8Verdict: unable to verify (needs the preprint text)
We may already be committed to collapse
- Said by
- Holden et al., preprint (reported in the news)
- What the source says
- Reported: large ensembles suggest a significant probability of committed collapse; not peer reviewed. Not read: no preprint text or link was available.
- Source
- news summaries (New Scientist via search) (link: Not verified)
Source date: 2026 · Retrieved: 2026-10-30 - How it was read
- news summaries only; the preprint text is needed to read the probability, the model and the scenario
- a9Verdict: confirmed (observation)
Observations show a sustained decline
- Said by
- RAPID programme
- What the source says
- 'sustained decline since' the 2010-11 downturn; variability from days to a decade; observing period 2004-2024
- Source
- RAPID-MOC
Source date: page current 2026-10-30 · Retrieved: 2026-10-30 - How it was read
- home page opened; strength in Sverdrups not stated there
- a10Verdict: partly supported (contested)
Sea-surface temperature is a poor proxy for AMOC strength
- Said by
- scientists quoted in Mother Jones (via search)
- What the source says
- 'deeply skeptical that SST is actually a proxy of AMOC'; early bucket measurements questioned; the paper itself calls SST an indirect fingerprint
- Source
- Mother Jones report on the study
Source date: 2023-07 · Retrieved: 2026-10-30 - How it was read
- search summary only
How to check
- Source
- BODC Published Data Library: 'Atlantic meridional overturning circulation observed by the RAPID-MOCHA-WBTS array at 26N from 2004 to 2024'
- Url
- https://www.bodc.ac.uk/data/published_data_library/
- Edmed
- https://www.bodc.ac.uk/resources/inventories/edmed/report/6178/
- Format
- NetCDF
- Coverage
- 2004-04-01 to 2024-11-30 (EDMED entry; another release listed 2024-03-27 as the end: check the version)
- Update
- versions released every one to two years, not daily; EDMED marks 'ongoing: no'
- Licence
- licence field blank: confirm with BODC before redistributing
- Honest card
- Yes, as an observation-only series with its end date and version shown; no forecast or collapse claim; needs BODC permission for redistribution
- Verified
- EDMED page and rapid.ac.uk opened; the data file itself was not downloaded
Data
No numeric series embedded: RAPID transport values were not obtained from the data files (2005-2023 mean 16.9 Sv as stated in Portmann et al.).
Caveats
- Verified by opening the source: the current text and corrected-publication PDF of Ditlevsen and Ditlevsen 2023, the Author Correction notice, the full text of Baker et al. 2025 and the PDF of Portmann et al. 2026 (the last two read by the PM from user-supplied files), rapid.ac.uk and the BODC EDMED page. Not opened: the AR6 chapter, the van Westen paper and the Holden preprint (summaries only).
- The collapse years are statistical estimates under stated assumptions, not predictions of a date; the Baker '90 years' are model years in an idealised experiment, not calendar years.
- This card describes statements, not the people who made them.
References
- Ditlevsen, P., and Ditlevsen, S. (2023). Warning of a forthcoming collapse of the Atlantic meridional overturning circulation. Nature Communications 14, 4254. DOI 10.1038/s41467-023-39810-w (full text read)
- Author Correction (2025). Nature Communications 16, 7794. DOI 10.1038/s41467-025-63201-y (read)
- IPCC AR6 WG1 (2021), as quoted by Drijfhout et al. (2025), Environ. Res. Lett., DOI 10.1088/1748-9326/adfa3b (summary only)
- RAPID-MOCHA 26N array: https://rapid.ac.uk/ ; BODC EDMED report 6178 (opened)
- Baker, J. A., Bell, M. J., Jackson, L. C., Vallis, G. K., Watson, A. J., and Wood, R. A. (2025). Continued Atlantic overturning circulation even under climate extremes. Nature 638, 987-994 (full text read by the PM)
- Portmann, R., et al. (2026). Observational constraints project a ~50% AMOC weakening by the end of this century. Science Advances 12, eadx4298, https://www.science.org/doi/10.1126/sciadv.adx4298 (PDF read by the PM)
- van Westen, R. M., Kliphuis, M., and Dijkstra, H. A. (2024). Science Advances (summary only)