AMOC transport variability and coherence: observations and models compared

Key points at a glance:

  • Thirty years of monitoring along the Norwegian continental slope and towards Svalbard shows ocean warming of 0.14 to 0.24°C per decade, with no matching change in the volume of water transported or its salinity.
  • The two main dense water overflows feeding the deep AMOC, the Denmark Strait and the Faroe Bank Channel, have stayed at ~3.2 and 2.2 Sverdrups (Sv) respectively since the mid-1990s, with no significant change.
  • The subpolar North Atlantic OSNAP array has measured an average overturning strength of ~16.5 Sv. Climate models produce between 9.4 Sv (coarse resolution) and 19.0 Sv (eddy-rich), depending on how finely they resolve ocean currents.
  • At 26.5°N, the RAPID array’s 20-year record shows finer-resolution models capture overturning strength and its seasonal cycle more accurately, but every model tested underestimates the heat and freshwater carried north.
  • The AMOC is far more consistent at South Atlantic latitudes than in the subpolar North Atlantic, and changes at high latitudes appear to reach lower latitudes 4 to 5 years later.

The Atlantic Meridional Overturning Circulation (AMOC) carries heat and salt north from the tropics and returns cold, dense water south at depth. Its strength varies from place to place and from year to year, and until recently, most observing arrays measured only one stretch of it in isolation.

EPOC Deliverable 1.3 brings together records from five monitoring arrays spanning the Arctic to the South Atlantic, some running for three decades, and sets them alongside three generations of climate model with the aim of establishing how consistently the AMOC behaves across the whole basin, and how well current models capture that behaviour.

The Arctic gateways: warming without a transport change

Warm, saline Atlantic Water enters the Arctic Ocean mainly through Fram Strait and the Barents Sea Opening, then cools and feeds back into the wider AMOC. EPOC’s earlier work on these gateways covered the 2004 to 2024 decline in heat transported into the Arctic (see our earlier work on the Arctic gateways for the full transport record).

Thirty years of measurements along the Norwegian continental slope and approaching Svalbard, from 63°N to 74°N, show consistent warming of 0.14 to 0.24°C per decade across four monitoring sections. The absence of any matching trend in transport volume or salinity is as significant as the warming itself.

Further north, a 27-year record of Atlantic Water temperatures in the West Spitsbergen Current, off Svalbard, shows a comparable warming rate of 0.20°C per decade, adding further evidence that the ocean is warming even where circulation strength is not.

Map and warming trend charts for four Norwegian Atlantic Current monitoring sections: Svinøy, Gimsøy, Barents Sea Opening and Bear Island
Left: Overview map showing the location of the four sections: Svinøy (SI), Gimsøy (GI), Barents Sea Opening (BSO) and Bear-Island (BI). Background shading shows 30-year average surface current speed. Centre: 30-year averages of absolute geostrophic velocities (shading), and temperatures (contours) at each section. The cyan polygon marks the location and extent of the Norwegian Atlantic Current core. Right: Time series of temperature (averaged over the respective core) and core transport for each section (12-months smoothed). Linear trends are drawn for temperature showing warming trends of 0.14 to 0.24°C per decade appear consistently across all four, shown alongside average current speed.

Dense water overflows into the North Atlantic

South of the Nordic Seas, a submarine ridge running from Greenland to Scotland holds back a reservoir of cold, dense water. This water spills over the ridge through two main channels: the Denmark Strait Overflow (DSO), between Greenland and Iceland, and the Faroe Bank Channel (FBC) Overflow, between the Faroe Islands and Scotland. Both feed the deep, southward-flowing branch of the AMOC.

Continuous observations extending back to the mid-1990s put the long-term mean transport at around 3.2 Sv for the Denmark Strait Overflow and 2.2 Sv for the Faroe Bank Channel Overflow. Neither shows a significant change over three decades of monitoring. Given ongoing debate about whether the AMOC is slowing, this stability in one of its deep branches is worth noting.

Time series chart of Denmark Strait and Faroe Bank Channel overflow transport, daily estimates and monthly means
Daily and monthly overflow transport through the Faroe Bank Channel (FBC; blue) and the Denmark Strait Overflow (DSO; purple), continuously monitored since the mid-1990s. Long-term average transport sits at -3.2 Sv for the DSO and -2.2 Sv for the FBC, shown as dashed lines.

Overturning in the subpolar North Atlantic

The Overturning in the Subpolar North Atlantic Programme (OSNAP) has monitored the full-depth circulation continuously since 2014, spanning from the Canadian shelf to Scotland. It provides a test of how well computer models represent the AMOC.

OSNAP measures an average overturning strength of 16.5 Sv. The coarsest resolution model tested (eORCA1) produces ~9.4 Sv; the medium-resolution, eddy-permitting model (eORCA025) produces ~13.1 Sv; the fine-resolution, eddy-rich model (eORCA12) produces ~19.0 Sv, closest to the observed value. Higher resolution generally helps, but not for every measure of transport.

One further finding stands out: the fine-resolution eddy-rich model (eORCA12) performs about as well as GLORYS12v1; a reanalysis product that blends model output with real observations. This suggests the extra data assimilation in GLORYS12v1 may add less value here than expected.

Monthly overturning strength at OSNAP, comparing four model resolutions against observations, 2014 to 2022
Monthly overturning strength at OSNAP, 2014 to 2022, comparing observations (black, with uncertainty envelope) against four model configurations: eORCA1, eORCA025, eORCA12 and the GLORYS12v1 reanalysis. Average overturning strength ranges from 9.4 Sv (eORCA1) to 19.0 Sv (eORCA12), against an observed average of 16.5 Sv.

Mid- and low-latitude arrays

At 26.5°N, the RAPID array has run since 2004, the longest continuous basin-wide AMOC record in existence. Finer-resolution models (eORCA12 and GLORYS12v1) capture both the mean overturning strength and its seasonal cycle more accurately than coarser configurations.

Some model simulations show a slight upward trend in overturning at this latitude, in contrast to the observed record. The reasons are still under investigation. Coarser models also underestimate the heat and freshwater carried north, pointing to biases that a weaker circulation alone does not explain.

Even the best-performing model has limits. GLORYS12v1 does well at 26.5°N but overstates Davis Strait transport by roughly a factor of two, a bias worth flagging for anyone using that product for Arctic-facing work.

At 26.5°N, the RAPID array has run since 2004, the longest continuous basin-wide AMOC record in existence. Finer-resolution models (eORCA12 and GLORYS12v1) capture both the mean overturning strength and its seasonal cycle more accurately than coarser configurations. Some model simulations show a slight upward trend in overturning at this latitude, in contrast to the observed record. The reasons are still under investigation. Coarser models also underestimate the heat and freshwater carried north, pointing to biases that a weaker circulation alone does not explain. Even the best-performing model has limits. GLORYS12v1 does well at 26.5°N but overstates Davis Strait transport by roughly a factor of two, a bias worth flagging for anyone using that product for Arctic-facing work.
Vertical overturning strength at RAPID (26.5°N) from 2004 to 2024. Observations (black) are compared against four model configurations: eORCA1, eORCA025, eORCA12 and the GLORYS12v1 reanalysis.

How connected is the AMOC across the ocean?

How consistently the AMOC behaves across different latitudes matters directly for observing system design. If changes at one latitude reliably predict changes at another, fewer arrays are needed to track the whole system.

The analysis finds coherence is much stronger in the South Atlantic than in the subpolar North Atlantic. In high-resolution coupled model simulations (MPI-ESM-ER), changes in the subpolar region appear to lead changes at lower latitudes by around four to five years at decadal timescales, a lag with practical consequences for how observing arrays should be interpreted together.

A further finding concerns the artificial smoothing built into some climate models. When this smoothing was reduced in HadGEM3-GC5 experiments, agreement between AMOC measures at different latitudes weakened substantially. The mechanism is still under investigation, but raises questions about how modelling choices shape our ability to simulate AMOC variability. Independent statistical estimates developed within EPOC broadly support the same pattern: coherence is strongest in the South Atlantic and weakest in the subpolar North Atlantic.

Lag in years between AMOC changes at different latitudes and the RAPID array, eddy-rich model simulation
Lag, in years, between AMOC changes at different latitudes and the RAPID array at 26.5°N, from an eddy-rich model simulation (MPI-ESM-ER). Positive values mean a latitude's AMOC changes lead the RAPID array; negative values mean it lags. Subpolar changes lead lower latitudes by around four to five years at decadal timescales.

Overall

Taken together, these findings advance understanding of how the AMOC varies across the full Atlantic basin, test how reliably current climate models reproduce what observations show, and point to where the most important gaps in knowledge and modelling capability remain. This gives EPOC, and the wider AMOC research community, a more reliable scientific basis for future assessments of the AMOC.

PDF icon

Want to learn more?

EPOC Deliverable 1.3 AMOC transport variability and coherence from observations and models. Report by Adam Blaker, May 2026.

The full report will be available for download when the EC review process is complete.