AMOC

The Atlantic Meridional Overturning Circulation (AMOC) is the ocean current system that transports warm tropical water northward and returns cold deep water southward. It carries ~1.25 petawatts of heat — roughly a quarter of the combined atmosphere–ocean poleward heat transport — and is the primary reason Western Europe is several degrees warmer than corresponding latitudes at the same longitude in the Pacific.

If it collapsed, the consequences would be severe for Europe, the North Atlantic, the ITCZ, and the global ocean–atmosphere circulation. Several recent papers have raised the question of whether a collapse is imminent, and the prediction windows are remarkably wide — anywhere from “as soon as 2025” (van Westen & Dijkstra 2024, narrowest estimate) to “very unlikely this century” (Baker et al. 2025).

This subsection walks through the data behind the controversy and asks a focused question: what would it take to narrow the prediction?

The headline prediction

Ditlevsen & Ditlevsen 2023 (Nature Communications) used the SST fingerprint from Caesar et al. 2018 — subpolar North Atlantic sea-surface temperature minus twice the global mean — and applied early-warning-signal statistics to detect a forthcoming tipping point. Their result:

Best estimate tipping year: 2057 (95% CI: 2037–2109)

The 72-year confidence interval is what makes this prediction controversial. A range that wide is essentially “could happen any time between now and the end of the century, give or take.”

Why is the window so wide?

The short answer: the dataset-quality story. The early-warning-signal method requires:

1. A long enough record (typically 50+ years) to establish baseline statistics
2. A high enough temporal resolution to detect changing variance and autocorrelation
3. A signal that is actually AMOC, not contaminated by other processes

The two main data sources have these properties in opposite ways:

The problem is the SST fingerprint is an indirect proxy. Sea-surface temperature in the subpolar gyre is influenced by wind variability, aerosol forcing, North Atlantic Oscillation phase, and other non-AMOC processes. This noise broadens the confidence interval.

The direct measurements are accurate but too short — RAPID’s 20-year record is below the 55-year optimal window for Ditlevsen & Ditlevsen’s estimator.

What the data actually shows

I downloaded the RAPID AMOC time series (2004–2024, v2024.1a) and the OSNAP 2014–2022 time series (Fu et al. 2025), and computed OLS trends on both:

This is exactly the decoupled behavior Fox et al. 2026 reported: the subtropical (RAPID) weakening is real and significant, but the subpolar doesn’t show a corresponding signal. The Ditlevsen prediction is consistent with the subtropical trend but doesn’t explain the subpolar null result.

Interactive: compute your own EWS

Below is the RAPID 12-month running mean AMOC time series, with sliding-window variance computed at multiple window sizes. Ditlevsen 2023 predicts variance should increase over time as the system approaches tipping. What does the RAPID record actually show?

What would actually narrow the prediction window

From the research dossier (28 papers reviewed):

1. Annually-resolved SST/SSS reconstructions from the subpolar gyre using corals and sclerosponges spanning 1700–present. This would replace HadISST (which begins 1870 and has known data quality issues in the early record) with a higher-fidelity, independently-dated record in the key AMOC fingerprint region.
2. Decadally-resolved Pa/Th records from the 20th century from high-sedimentation-rate cores. This would provide an independent check on whether the AMOC has actually been weakening (as opposed to the SST fingerprint showing a trend for non-AMOC reasons).
3. Longer RAPID-type direct measurements. Even 10–20 more years of RAPID data would significantly improve the ability to separate trend from decadal variability.
4. Machine-learning-based AMOC reconstructions from Argo profiles (demonstrated in a 2025 Egusphere preprint), which could extend the direct AMOC estimate backward using Argo’s spatial coverage in combination with historical hydrography.

In short: the wide prediction window is not a sign of model disagreement. It’s a sign of data scarcity.

The deeper scientific dispute

The honest assessment is that the scientific community is genuinely divided, and the 20-year RAPID record is too short to resolve the debate. Three positions are currently active:

The IPCC AR6 assessment reflects this uncertainty: medium confidence that AMOC will not abruptly collapse before 2100, but a downgrade from SROCC’s “very unlikely” — and low confidence in reconstructed 20th-century AMOC weakening trends because of weak agreement between proxies.

For the full citation table and 28-paper dossier, see the Paper page.

— Herman