September 2025

Oceanography | Vol. 38, No. 3

16

and the atmospheric CO2 concentration recorded in Antarctic ice,

which in turn is sensitive to the interplay between the AMOC and

the lower AABW circulation cell (Figure 2b; Marcott et al., 2014).

A complication to this picture is the possibility that the atmosphere

can respond to a weakened AMOC by strengthening its meridi-

onal heat transport due to increased equator-to-pole temperature

gradients (Bjerknes, 1964). This feedback in the coupled ocean-​

atmosphere system is referred to as “Bjerknes Compensation” and

likely diminishes the signal in atmospheric-linked proxy records of

a weakened or collapsed AMOC. Despite this possibility, collective

paleoclimate data from both marine and continental archives are

consistent with AMOC weakening during both HS1 and the YD,

with a period of reinvigorated circulation during the BA. The HS1

and YD emerge, therefore, as key time intervals for investigating

AMOC changes and their driving mechanisms. Information from

these time intervals could in turn be used to inform our under-

standing of possible AMOC changes in future.

FRESHWATER FORCING IN TRANSIENT

MODEL SIMULATIONS OF AMOC DECLINE/

COLLAPSE ACROSS THE DEGLACIATION

To study deglacial climate variability, scientists have performed

and analyzed transient simulations with numerical climate models.

The most coordinated of such efforts is the Paleoclimate Modelling

Intercomparison Project (PMIP), where participating groups apply

climate models to conduct numerical experiments with prescribed

boundary conditions. The “Last Deglaciation” is one such experi-

ment, which simulates the period from 21 ka to 9 ka (Ivanovic et al.,

2016). Given its relatively long duration—about 12,000 years—

there are severe computational limitations to the spatial resolu-

tion of climate models that can be run to simulate the deglacial cli-

mate. The horizontal resolution of the ocean component of climate

models, such as those included in the most recent PMIP (PMIP4),

is too coarse (on the order of 1°) to explicitly simulate ocean eddies,

which play important roles in a wide variety of processes that are

thought to be crucial for AMOC—such as deep convection, lat-

eral restratification, and the dispersal and dilution of continental

freshwater. For example, recent observations around the convective

region of the Labrador Sea have confirmed that submesoscale pro-

cesses (smaller than 100 km) are critical to the restratification of

deep convective plumes (Clément et al., 2023), yet large-scale ocean

models with sufficient resolution can take years to run (Pennelly

and Myers, 2020). Eddies produced from the instability of buoyant

coastal currents formed by meltwater discharge may also be effec-

tive in transporting melt­water offshore (Marchal and Condron,

2025). To address the limitation due to coarse resolution, sub-grid-

scale processes (e.g.,  deep convection, dense overflows, coastal

eddies) are parameterized in the PMIP models, but this approach

can lead to inaccuracies in model sensitivity to freshwater fluxes,

with some models reported to be overly sensitive to fresh water

(Bouttes et al., 2023) and others not sensitive enough (Valdes, 2011;

He and Clark, 2022; Snoll et al., 2024).

Some model experiments (Liu et al., 2009; Menviel et al., 2011)

have explicitly used AMOC proxy records as tuning targets; in these

experiments, the temporal evolution of the freshwater flux into the

ocean is manipulated so as to qualitatively match the proxy records

(in both studies, the McManus et al. [2004] 231Pa/230Th record from

the Bermuda Rise and reconstructed Greenland temperature vari-

ations were used). The motivation for using freshwater forcing to

simulate the AMOC changes inferred from the proxy records is as

follows: over the deglaciation, continental ice sheets melted, lead-

ing to the release of vast amounts of fresh water into the ocean,

driving a sea level rise of ~130 m (Clark et al., 2009; Carlson and

Clark, 2012; Lambeck et al., 2014). The released fresh water could

have reduced the density of surface waters in deep-water forma-

tion regions of the North Atlantic, inhibiting deep convection

there and reducing the AMOC. Deglacial simulations by Liu et al.

(2009, “TraCE-21k”) and Menviel et al. (2011) reproduce this sce-

nario. Both simulations also match other paleoclimate reconstruc-

tions, in addition to those taken as evidence for AMOC changes

and used as tuning targets.

While deglacial simulations with prescribed freshwater forcing

can produce results that match paleoclimate records, the magni-

tude and timing of the freshwater fluxes assumed in these simu-

lations are not consistent with freshwater fluxes calculated from

the data-constrained deglacial reconstructions of continental ice

sheets (e.g., GLAC-1D: Tarasov and Peltier, 2002; Tarasov et al.,

2012; Briggs et al., 2014; and ICE-6G_C: Argus et al., 2014; Peltier

et  al., 2015; Ivanovic et  al., 2016). Both the simulations of Liu

et al. (2009; TraCE-21k) and Menviel et al. (2011) prescribe fresh-

water fluxes of approximately 0.2 Sv during HS1 that are nearly

twice as high as those predicted from GLAC-1D and ICE-6G_C

(Bouttes et al., 2023; Figure 3a). There are also significant off-

sets in the timing of the freshwater fluxes: Meltwater Pulse 1A, at

the beginning of the BA (Deschamps et al., 2012; Lambeck et al.,

2014), occurs earlier (by a few centuries) in the ice sheet recon-

structions than in the climate simulations, and the peak of melt-

water input in the ice sheet reconstructions occurs when fresh-

water flux is shut off in the TraCE-21k simulation. A similar result

holds true for Meltwater Pulse 1B, which roughly coincides with

the end of the YD.

In summary, while freshwater forcing has been used to drive

AMOC variability in climate models, the highest freshwater fluxes

assumed in the climate model simulations occur when freshwater

fluxes in the ice sheet models are believed to be relatively low. This

phenomenon is referred to as the “meltwater paradox” (e.g., Snoll

et al., 2024). Indeed, in other simulations forced with fresh­water

fluxes that are more consistent in magnitude and timing with

freshwater flux reconstructions, the AMOC does not collapse at all

or collapses at the start of the BA (Figure 3b; Bouttes et al., 2023;

Snoll et al., 2024). Thus, it appears that fresh water entering the

North Atlantic from the melting of the Laurentide Ice Sheet was

unlikely to be the driving mechanism for reducing the AMOC

during HS1 and YD.