Oceanography | Vol. 38, No. 3
18
While the timing of the highest meltwater delivery to the North
Atlantic across the deglaciation does not match the times when
AMOC appeared to be weaker (HS1 and the YD), the other mech-
anisms discussed above could have contributed to a weakening of
the AMOC. Thus far, climate models have not been able to accu-
rately simulate these processes due to the computational cost
required to resolve dynamical phenomena at small spatial scales for
long time periods. Although freshwater forcing is frequently used
as a convenient way to produce changes in the AMOC in models,
it is not the only mechanism that can drive variations in AMOC
strength. For example, other modeling studies using a coarse reso-
lution Earth system model suggest that abrupt AMOC oscillations
can arise from gradual changes in ice sheet height that modify the
wind field (Zhang et al., 2014) or atmospheric CO2 concentration
(Zhang et al., 2017).
HOW CAN PALEO OBSERVATIONS
INFORM MODERN UNDERSTANDING
AND FUTURE PREDICTIONS?
Future projections of AMOC strength from coupled climate mod-
els support a moderate decline but not a full collapse of AMOC
over the next 100 years (Fox-Kemper et al., 2021). However, these
estimates are only reliable if we understand the underlying physics
that drives an AMOC decline. As we discuss in the previous sec-
tion, there are still gaps in our understanding of what caused past
abrupt changes in the AMOC. The most recent deglaciation may be
a good past analog, because there is paleoceanographic evidence for
abrupt AMOC changes occurring on timescales of decades to cen-
turies, and a recent quantitative estimate of freshwater input from
iceberg melt during HS1 (Zhou and McManus, 2024) is comparable
to modern ice fluxes from the Greenland Ice Sheet (GIS; Bamber
et al., 2018). On the other hand, there were important differences
from our current climate state, including large areas of land and sea
ice cover. It has long been suggested that the AMOC is sensitive
to background climate state, and intermediate climate conditions,
with moderate CO2 concentrations, ice volumes, and temperatures,
are more conducive to millennial climate variability than peak gla-
cial or interglacial conditions (McManus et al., 1999; Sima et al.,
2004; Barker and Knorr, 2021). For example, abrupt climate oscilla-
tions known as Dansgaard-Oeschger (DO) Events were observed in
Greenland ice cores and North Atlantic sediment cores during the
middle of the last glacial period (~75 ka to 25 ka), and these have
been linked to variations in the AMOC (North Greenland Ice Core
Project Members, 2004; Andersen et al., 2006; Rasmussen et al.,
2014; Böhm et al., 2014; Henry et al., 2016). Several modeling stud-
ies have replicated this observation and found that the AMOC is
less stable under intermediate climate conditions (that is, neither
fully glacial nor fully interglacial; Ganopolski and Rahmstorf, 2001;
Sima et al., 2004; Galbraith and de Lavergne, 2019).
If there is evidence that the inherent stability of the AMOC is
dependent on background climate state, does that mean that the
mechanism(s) that drive AMOC change also vary with the mean
climate state? Unlike the deglaciation, no large continental ice sheets
cover North America or Eurasia today, and no ice-dammed lakes
are present to flood the subpolar North Atlantic. However, both the
GIS and Arctic sea ice are rapidly melting (The IMBIE Team, 2019;
Sumata et al., 2023; Greene et al., 2024), and the Beaufort Gyre has
been accumulating fresh water that could be released to the North
Atlantic more rapidly than melting ice sheets would do (Haine
et al., 2015). How these different freshwater sources (GIS, Arctic
sea ice, and Beaufort Gyre) could alter the AMOC under the mod-
ern climate conditions of the North Atlantic remains unknown.
Investigating AMOC variability during warm periods, such as
the current Holocene epoch, past interglacial periods, and even
farther into the geologic past, may provide more context for what
we might expect in the future. During the current Holocene epoch,
fresh water and ice were released from Hudson Bay at 8.2 ka
(Barber et al., 1999), causing global impacts (Alley et al., 1997).
Although it is difficult to detect a decade-to-century scale event in
the deep sea, there is some evidence for AMOC reduction at 8.2 ka
(Keigwin et al., 2005; Kleiven et al., 2008). These reconstructions
show different locations of freshwater delivery to the ocean during
the last deglaciation that may help us understand the relationship
between the location of freshwater input into the North Atlantic
and its impacts on the AMOC.
Today, the Greenland meltwater combines with outflow from
the Arctic Ocean through Davis Strait (B. Curry et al., 2014),
Hudson Strait (Straneo and Saucier, 2008), and Fram Strait
(Karpouzoglou et al., 2023) to carry large amounts (1–3 Sv) of
fresh, polar water masses into the coastal circulation system in the
subpolar North Atlantic (Foukal et al., 2020; Le Bras et al., 2021).
Much of this fresh water is retained on the continental shelves of
East Greenland and Labrador, but it can be transported into the
basin interior along West Greenland (Luo et al., 2016; Dukhovskoy
et al., 2019; Pacini and Pickart, 2023) and the Grand Banks (Jutras
et al., 2023; Fox et al., 2022; Furey et al., 2023; Duyck et al., 2025).
It is likely that the Grand Banks was the source of the large fresh-
ening event seen in the Iceland Basin in 2015 (Holliday et al., 2020)
and in the Irminger Sea in 2019 (Biló et al., 2022). However, nei-
ther how these events impacted the AMOC, nor how similar they
were to previous freshening events—the so-called great salinity
anomalies of the 1970s and the 1980s (Dickson et al., 1988; Belkin
et al., 1998)—is well understood.
Paleo freshwater discharge events may help elucidate the impact
of freshwater routing: current understanding suggests that HS1
originated in Hudson Strait, the YD originated in the Mackenzie
River, and the 8.2 ka event originated in Hudson Bay and prob-
ably reached as far as Cape Hatteras. Much of the recent work
on AMOC dynamics and stability (Boers, 2021; Ditlevsen and
Ditlevsen, 2023) has focused on model-based surface fingerprints
of AMOC variability (Rahmstorf et al., 2015; Caesar et al., 2018,
2021). But the suitability of this fingerprint for inferring AMOC
variability has been widely debated, and it is likely timescale
dependent (Little et al., 2020; Kilbourne et al., 2022; Li et al., 2022;