September 2025 | Oceanography
13
circulation (Frajka-Williams et al., 2019), although a recent study
reported a slight decline in the AMOC at 26°N between 2004 and
2022 (Volkov et al., 2024). Thus, it is unclear whether the AMOC
has already responded to anthropogenic forcing.
The mechanisms by which the northward-flowing surface waters
are transformed into dense water masses and exported southward
are complex. Classically, thermal convection has been thought of
as a means to form dense water masses in the Labrador, Irminger,
and Greenland Seas (Broecker and Denton, 1989; Manabe and
Stouffer, 1995), but more recent studies show that deep convection
does not result in net sinking (Spall, 2004; Pickart and Spall, 2007).
Instead, sinking likely occurs in the boundary currents of mar-
ginal seas (e.g., Nordic and Labrador Seas) where those currents
interact with each other and with steep topography (Bower et al.,
2011; Gary et al., 2011; Katsman et al., 2018; Johnson et al., 2019;
Desbruyères et al., 2020). Convection likely exerts a strong influ-
ence on the properties of the deep waters through mixing with the
boundary currents, but it may not be the primary mechanism for
forming the deep waters. A similar process occurs farther south
where NADW interacts with the lower, counter-rotating cell of
Antarctic Bottom Water (AABW) originating from the Southern
Ocean. The interplay between the relative strength of the NADW
and AABW cells likely sets the depth of the AMOC and thus
impacts AMOC dynamics (Marshall and Speer, 2012).
Paleoceanographic reconstructions, simulations from numeri-
cal models, and data inversions can provide insight into ocean cir-
culation changes during periods of past climate change and into
the mechanisms responsible, but all approaches have their own
limitations. Marine archives, such as corals and sediment cores,
have limited spatial and temporal coverage, and proxy reconstruc-
tions have analytical, chronological, and interpretive uncertain-
ties. Paleoceanographic data can be used to estimate the spatial
distribution of oceanic properties (such as temperature, isotopic
compositions, and nutrient concentrations), but reconstructions
of AMOC are primarily qualitative. In contrast, numerical mod-
els can provide quantitative volume flux estimates, but they suffer
from their own limitations due to, for example, uncertainties in
surface boundary conditions (atmospheric forcing), initial condi-
tions, and parameterization of sub-grid-scale phenomena. Notably,
due to computational limitations, numerical ocean models applied
in climate research are generally characterized by coarse horizon-
tal resolution (on the order of 1°), which means that the mesoscale
and submesoscale ocean eddy fields are not explicitly resolved,
and coastal phenomena known to contribute to shelf-ocean
exchange are poorly or not represented. Finally, inverse methods
have been applied to combine paleoceanographic data and mod-
els to extract quantitative information about past ocean circula-
tion (e.g., LeGrand and Wunsch, 1995; Gebbie and Huybers, 2006;
Marchal and Curry, 2008; Burke et al., 2011; Amrhein et al., 2015;
Zhao et al., 2018; Marchal and Zhao, 2021). These applications
showed that firm inferences about past circulation states from
existing paleoceanographic data are difficult given the combined
limitations of data and model.
In this paper, we review the paleoceanographic data that have led
to the prevailing view of a weak AMOC for millennia (or longer)
during the last glacial-interglacial transition and climate model
simulations of these events. We also discuss the mechanisms that
could have driven past AMOC changes, with particular attention
to freshwater forcing. Finally, we discuss the extent to which exist-
ing observational and model results are relevant to current and
future changes in the AMOC, with particular emphasis on the pos-
sible role of background climate state. This review is distinct from
other recent reviews on similar topics (e.g., Lynch-Stieglitz, 2017;
Liu, 2023) through a focus on (1) the lessons learned about the
AMOC
lower cell
upper cell
‘AMOC’
lower cell
Southern
Ocean
Atlantic
Ocean
Nordic
Sea
Southern
Ocean
Atlantic
Ocean
Nordic
Sea
‘ON’ circulation state
‘OFF’ circulation state
FIGURE 1. Schematic of two different states of the Atlantic Meridional Overturning Circulation (AMOC). (a) A vigorous or “on” state, with a relatively deep and
strong upper cell, similar to the circulation in the modern Atlantic. (b) A collapsed or “off” state, with a relatively shallow upper cell and a larger lower (Antarctic
Bottom Water) cell. A number of paleoceanographic observations have been interpreted as reflecting a collapsed state of the AMOC, as in (b), during the
last deglaciation. The unlabeled contours and colors schematically represent water masses originating from the North Atlantic (orange) and Southern Ocean
(green), with darker colors qualitatively representing a greater fraction of the water mass.