September 2025 | Oceanography
Oceanography
THE OFFICIAL MAGAZINE OF THE OCEANOGRAPHY SOCIETY
VOL. 38, NO. 3, SEPTEMBER 2025
OCEANOGRAPHY IN THE AGE
OF INTELLIGENT ROBOTS
EVIDENCE FOR FRESHWATER-
DRIVEN AMOC CHANGES?
PERSPECTIVES ON MARINE
CARBON DIOXIDE REMOVAL
NUTRIENT FOOTPRINT OF
THE KUROSHIO CURRENT
IN THIS ISSUE
Oceanography | Vol. 38, No. 3
POWERING SCIENCE-BASED DECISIONS
FOR A BETTER OCEAN.
September 2025 | Oceanography
© MBARI 2023
contents
VOL. 38, NO. 3, SEPTEMBER 2025
58
5 QUARTERDECK. Enriching Readers’ Experience Through Oceanography Flipbooks
By E.S. Kappel
7 COMMENTARY. Democratize the Data: A New Way to Analyze and Design Ocean Models
By T.W.N. Haine
12 FEATURE ARTICLE. Is There Robust Evidence for Freshwater-Driven AMOC Changes?
A Synthesis of Data, Models, and Mechanisms
By S.K.V. Hines, N.P. Foukal, K.M. Costa, D.W. Oppo, O. Marchal, L.D. Keigwin, and A. Condron
24 FEATURE ARTICLE. Perspectives on Marine Carbon Dioxide Removal from the Global
Ocean Acidification Observing Network
By H.S. Findlay, R.A. Feely, K. Grabb, E.B. Jewett, E.F. Keister, G. Kitch, Y. Artioli, P. Bhadury, J. Blackford,
O. Crabeck, A. Ghosh, Y. Li, K.B. Lowder, S. Mehta, B. Van Dam, H. Beghoura, N. Karo, A.Z. Horodysky,
S. Hennige, S.M. Salaah, F. Ragazzola, and L. Wright-Fairbanks
40 FEATURE ARTICLE. Nutrient Footprint from the Origin of the Kuroshio Current to the
East China Sea Continental Shelf
By T.-H. Huang and C.-T.A. Chen
51 FEATURE ARTICLE. Mentors: The Hidden Beneficiaries of Mentoring
By M. Behl, S. Clem, C. Mouw, S. Legg, E. Hackett, K. Burkholder, K.B. Karnauskas, S.T. Gille,
L.A. Freeman, K. Venayagamoorthy, and J.L. Miller
60 ROGER REVELLE COMMEMORATIVE LECTURE. Oceanography in the Age of Intelligent
Robots and a Changing Climate
By C. Scholin
74 MEETING REPORT. Tools in Harmony: Integrating Observations and Models for Improved
Understanding of a Changing Ocean
By E.H. Ombres, H. Benway, K. Bisson, A.A. Larkin, E.A. Perotti, E. Wright-Fairbanks, J. Crosswell,
S. Dutta, C. Garcia, A. Gnanadesikan, K. Grabb, A. Fay, R. Jin, K. Kelly, H. Kwasniewski, A.K. Labossiere,
J. Lauderdale, J. Lee, Y. Lin, J.S. Long, A. Rufas, C. Schultz, N.D. Ward, and Y. Zhu
80 OCEAN EDUCATION. Drifter Challenge: A Low-Cost, Hands-On Platform for Teaching
Ocean Instrumentation and Sensing
By C. Xia, B. Champenois, F. Campuzano, and R. Mendes
88 THE OCEANOGRAPHY CLASSROOM. How To Get Factual Data and Articles: Surviving in
Today’s Online World
By S. Boxall
90 CAREER PROFILES. Brian Kennedy, Chief Scientist, Ocean Discovery League •
Paige E. Martin, User Training Team Lead, Australia’s Climate Simulator
93 BOOK REVIEW. The Ocean’s Menagerie: How Earth’s Strangest Creatures Reshape
the Rules of Life, by Drew Harvell
Reviewed by G.I. Matsumoto
Oceanography | Vol. 38, No. 3
CONTACT US
The Oceanography Society
1 Research Court, Suite 450-117
Rockville, MD 20850 USA
t: (1) 301-251-7708
info@tos.org
HAVE YOU MOVED?
Send changes of address to info@
tos.org or go to https://tosmc.
memberclicks.net, click on Login,
and update your profile.
ADVERTISING INFO
Please send advertising inquiries to
info@tos.org or go to https://tos.org/
oceanography/advertise.
CORRECTIONS
Please send corrections to
magazine@tos.org.
ON THE COVER
A mapping autonomous underwater
vehicle (AUV, left) is being recovered
after a seafloor survey in Arctic waters.
A portable remotely operated vehicle
(MiniROV) is on the deck of Canadian
Coast Guard Ship Sir Wilfrid Laurier.
Image © 2016 MBARI
SUBMIT A MANUSCRIPT TO
Oceanography
OBJECTIVE OF OCEANOGRAPHY
Oceanography is an open-access, peer-reviewed jour-
nal published quarterly by The Oceanography Society.
The journal presents significant research, noteworthy
achievements, exciting new technology, and articles that
address public policy and education. The overall goal of
Oceanography is cross-disciplinary communication in
the ocean sciences.
LEARN MORE
See the online Oceanography Author Guidelines for
a full listing of manuscript categories and descriptions,
publication fees, and details of the manuscript submis-
sion process.
https://tos.org/oceanography/
guidelines
WHAT GETS OUR ATTENTION
FEATURE ARTICLES (<7,000 words) provide an outlet for making signif-
icant advances in oceanography accessible to a broad readership. They
can include review papers that summarize the current state of knowledge
of a particular topic, synthesis papers that discuss new findings and how
they significantly revise our thinking about a topic, and more traditional
scientific research papers from across the full spectrum of ocean sciences.
BREAKING WAVES (<3,500 words) articles describe novel approaches
to multidisciplinary problems in oceanography. These provocative papers
present findings that have the potential to move the field of oceanography
forward or in new directions.
OCEAN EDUCATION (<3,500 words) articles should inspire teachers in
higher education to try new active, student-centered instruction (ranging
from short activities to curricula) and provide ideas/materials to do so.
DIY OCEANOGRAPHY (<3,500 words) articles share all of the relevant
information on a homemade sensor, instrument, or software tool(s) so that
others can build, or build upon, it. These articles also showcase how this
technology was used successfully in the field.
MEETING/WORKSHOP/CONFERENCE REPORTS (<3,500 words) describe
the goals, activities, and accomplishments of meetings/workshops/confer-
ences in all aspects of ocean science.
COMMENTARIES (<3,500 words) present analyses of issue of interest to
Oceanography readers, written by experts in the field. Unsolicited manu-
scripts are welcome.
RIP CURRENT – NEWS IN OCEANOGRAPHY (<1,500 words) articles
describe newsworthy items in the field of oceanography.
September 2025 | Oceanography
The Oceanography Society was founded in 1988 to advance
oceanographic research, technology, and education, and to dis-
seminate knowledge of oceanography and its application through
research and education. TOS promotes the broad understanding
of oceanography, facilitates consensus building across all the dis-
ciplines of the field, and informs the public about ocean research,
innovative technology, and educational opportunities throughout
the spectrum of oceanographic inquiry.
OFFICERS
PRESIDENT: Paula Bontempi
PRESIDENT-ELECT: Heidi Dierssen
PAST-PRESIDENT: Deborah Bronk
SECRETARY: Allison Miller
TREASURER: Susan Banahan
COUNCILORS
AT-LARGE: Leonard Pace
APPLIED TECHNOLOGY: Anna Michel
BIOLOGICAL OCEANOGRAPHY: Grace Saba
CHEMICAL OCEANOGRAPHY: Jun Nishioka
EARLY CAREER: Hilary Palevsky
EDUCATION: Leilani Arthurs
GEOLOGICAL OCEANOGRAPHY: Jon Lewis
OCEAN DATA SCIENCE: Jeremy Werdell
OCEAN SCIENCE AND POLICY: Eric Wade
STUDENT REPRESENTATIVE: Moronke Harris
JEDI COUNCILOR: Sheri White
EXECUTIVE DIRECTOR
Jennifer Ramarui
CORPORATE AND INSTITUTIONAL
MEMBERS
Baker Donelson
Bamfield Marine Sciences Centre
Ganey Science
Greenwater Marine Sciences Offshore
Integral Consulting Inc.
National Oceanography Centre
Sea-Bird Scientific
Sequoia
tos.org
EDITOR
Ellen S. Kappel, Geosciences Professional Services Inc.
ASSISTANT EDITOR
Vicky Cullen
DESIGN/PRODUCTION
Johanna Adams
ASSOCIATE EDITORS
Claudia Benitez-Nelson, University of South Carolina
Luca Centurioni, Scripps Institution of Oceanography
Grace Chang, Integral Consulting Inc.
Tim Conway, University of South Florida
Kjersti Daae, University of Bergen
Mirjam S. Glessmer, University of Bergen
Charles H. Greene, University of Washington
Alistair Hobday, CSIRO Environment
Helen R. Pillar, University of Texas at Austin
Carol Robinson, University of East Anglia
Amelia Shevenell, University of South Florida
Robert E. Todd, Woods Hole Oceanographic Institution
Peter Wadhams, University of Cambridge
Oceanography contains peer-reviewed articles that chronicle
all aspects of ocean science and its applications. The journal
presents significant research, noteworthy achievements, excit-
ing new technology, and articles that address public policy and
education and how they are affected by science and technol-
ogy. The overall goal of Oceanography is cross-disciplinary
communication in the ocean sciences.
Oceanography (Print ISSN 1042-8275; Online ISSN 2377-617X)
is published by The Oceanography Society, 1 Research Court,
Suite 450-117, Rockville, MD 20850 USA. Oceanography arti-
cles are licensed under a Creative Commons Attribution 4.0
International License, which permits use, sharing, adaptation,
distribution, and reproduction in any medium or format as long
as users cite the materials appropriately, provide a link to the
Creative Commons license, and indicate the changes that were
made to the original content. Third-party material used in arti-
cles are included in the Creative Commons license unless indi-
cated otherwise in a credit line to the material. If the material is
not included in the article’s Creative Commons license, users
will need to obtain permission directly from the license holder
to reproduce the material. Please contact Jennifer Ramarui at
info@tos.org for further information.
Oceanography
tos.org/oceanography
September 2025 | Oceanography
Oceanography | Vol. 38, No. 3
THE OCEANOGRAPHY SOCIETY’S
HONORS PROGRAM
One of the most meaningful TOS programs is celebrating our
colleagues’ accomplishments through the TOS Honors program.
Please take this opportunity to recognize a colleague, mentor,
team, or peer for their exceptional achievements and contribu-
tions to the ocean sciences.
Medals
WALLACE S. BROECKER MEDAL is awarded biennially to
an individual for innovative and impactful contributions to the
advancement or application of marine geoscience, chemical
oceanography, or paleoceanography.
The NILS GUNNAR JERLOV MEDAL is awarded biennially
to an individual for significantly advancing our knowledge of
how light interacts with the ocean.
The WALTER MUNK MEDAL is awarded biennially to
an individual for extraordinary accomplishments and novel
insights contributing to the advancement or application of
physical oceanography, ocean acoustics, or marine geophysics.
The MARY SEARS MEDAL is awarded biennially to an
individual for innovative and impactful contributions to the
advancement or application of biological oceanography, marine
biology, or marine ecology.
Fellows
Recognizing TOS members who have made outstanding and
sustained contributions to the field of oceanography through
scientific excellence, extraordinary service and leadership, and/
or strategic development of the field.
Awards
The TOS EARLY CAREER AWARD is presented bienni-
ally to up to three TOS Early Career members for significant
early-career research contributions and impact, and the poten-
tial for future achievements in the field of oceanography.
The TOS MENTORING AWARD is given biennially to an
individual for excellence and/or innovation in mentoring the
next generation of ocean scientists.
The TOS OCEAN OBSERVING TEAM AWARD is pre-
sented biennially to a team for innovation and excellence in sus-
tained ocean observing for scientific and practical applications.
tos.org/honors
NOMINATIONS ARE DUE OCTOBER 31, 2025
Honors Nomination Committee
In addition to receiving nominations directly from
community members, the TOS Honors Nomination
Committee provides an opportunity for members to
collaborate on suggesting and submitting nominations.
If you would like to apply to be part of this committee,
please apply by September 26, 2025. Learn more at:
tos.org/honors-nomination-committee
September 2025 | Oceanography
VIDEOS
• We recommend that videos not exceed a few minutes in length
• Videos must be in mp4 format and no larger than 100 MB
ANIMATED GIFS
• GIFS must be no larger than 5 MB
AUDIO
• Audio files must be in mp3 format and no larger than 100 MB
PHOTO GALLERIES
• Photos for a photo gallery must be in jpg or png format and no
larger than 5 MB per photo
• For best results, all photos should have the same dimensions
I look forward to sharing with readers even more media-
enhanced Oceanography articles in the future. I hope you will share
these articles broadly as well.
ENRICHING READERS’ EXPERIENCE THROUGH
OCEANOGRAPHY FLIPBOOKS
In my fall 2023 Quarterdeck column,
I introduced the digital flipbook ver-
sion of Oceanography to our readers.
These web-based publications make it
possible to page through entire issues
rather than click on individual arti-
cles from our table of contents pages.
Importantly, for the flipbook versions
of articles, authors can embed videos,
animations, audio files, and photo gal-
leries, enhancing the reader experi-
ence. I highlight these flipbooks again
because the slow adoption of these
multimedia assets by authors, which
can enhance articles’ appeal and reach
to a broader audience, has been sur-
prising. Wouldn’t it be more interest-
ing and informative to display clips of
simulations in the flipbook version of
your article rather than the same series
of static images that appear in the article PDF? An animation of
time series in addition to the stack of static plots? An animation
of a tsunami flooding forecast that shows a progression through
time much better than a series of screenshots? An interesting crit-
ter moving about or a video of data integration rather than the only
the frame grabs? Or even a video or a gallery of photos of fieldwork
taken by people on board the research vessel using their smart-
phones? We have already experimented with using a video for an
issue cover (see the flipbook version of the June 2025 issue) while
the static PDF cover was a merge of screenshots of a black smoker.
We at Oceanography do the work for you. When you submit
your article through our Scholastica portal you can upload addi-
tional media assets along with your manuscript and static figures.
If the video/audio files are too large to upload through Scholastica,
please provide a way for us to access them. Note that all additional
assets must be directly associated with a figure in the article—we
will embed the media within the figure when developing the flip-
book version of your article.
We have updated our Author Guidelines since I last wrote about
our flipbooks. The following are the most current parameters
regarding file type and size. Before submitting your media assets,
please double-check the Author Guidelines to confirm there have
been no further updates.
QUARTERDECK
Ellen S. Kappel, Editor
ARTICLE DOI. https://doi.org/10.5670/oceanog.2025.e311
If a picture is worth a thousand words, what is a video worth? Compare this video to the static screenshot
embedded in the pdf version of this column. Credit iStock.com/BlackBoxGuild
Oceanography | Vol. 38, No. 3
Image courtesy of ESA
OCEANOPTICSCONFERENCE.ORG
OCEAN OPTICS XXVII
SEPTEMBER 13–18, 2026 | GHENT, BELGIUM
It is our pleasure to invite you to attend Ocean Optics XXVII,
the twenty-seventh edition of the biennial international Ocean
Optics Conference. This event brings together a diverse ocean
optics community, including oceanographers, limnologists,
optical engineers, Earth observation scientists, resource manag-
ers, and policy professionals from across the globe, all united by
a shared passion for optics in aquatic environments.
Ocean Optics XXVII will be held at two exceptional venues in the
heart of Ghent. Wintercircus, a stunningly renovated innovation
hub centered around a dramatic circular arena, will provide the
space for the poster sessions, exhibits, and short courses. The
plenary sessions will be held just across the street in the iconic
Theaterzaal (theater) of the Arts Center VIERNULVIER, known for
its historic elegance and architectural grandeur.
REGISTRATION AND
HOUSING ARE OPEN
Register by January 14, 2026 to take
advantage of early bird rates.
Exclusive discounted hotel rates
are available on a first-come,
first-served basis.
oceansciencesmeeting.org
September 2025 | Oceanography
DEMOCRATIZE THE DATA
A NEW WAY TO ANALYZE AND DESIGN OCEAN MODELS
By Thomas W.N. Haine
INTRODUCTION
Simulation of ocean currents by numerical models has been revo-
lutionized by information technology advances in the last 50 years.
New discoveries have resulted from improved observing tech-
nologies, such as the global Argo network of autonomous profil-
ing floats (Riser et al., 2016; Argo, 2020) and satellite observations
of sea level (Lee et al., 2010; Vinogradova et al., 2025). Improved
ocean circulation models have also resulted in new discoveries
(Fox-Kemper et al., 2019; Haine et al., 2021), particularly those
based on better model grid resolution. The growth in ocean cir-
culation model fidelity brings challenges, however. One chal-
lenge concerns the difficulty of providing access to the very large
volumes of data ocean circulation models produce, and another
concerns the priorities for future cutting-edge ocean circulation
model simulations.
This commentary introduces and explains these topics and out-
lines some possible ways ahead. Developments in cloud storage
and cloud computing are providing open cyberinfrastructure plat-
forms that lower the barrier to data access. Open discussion on
future circulation model priorities is also beginning. These ser-
vices for, and engagement with, the oceanographic community
aim to make cutting-edge ocean current simulations as widely
accessible and as useful as possible.
GRID CELL AND DATA GROWTH
Global ocean general circulation models (OGCMs) show expo-
nential growth in grid cell resolution. This remarkable expansion
ultimately derives from Moore’s law, which states that the density
of microelectronic devices doubles every two years (Moore, 1975).
To illustrate, Figure 1 shows the number of horizontal grid cells
used to discretize the global ocean in five cutting-edge OGCMs
since 1980 (with black dots). The number of horizontal grid cells
doubles every 2.5 yr, keeping up with Moore’s law (some of the
increase in computer power is used to refine OGCM vertical res-
olution). Nowadays, cutting-edge OGCMs have horizontal resolu-
tions of around 1 km, with hundreds of millions of grid cells cov-
ering the surface of the global ocean.
Coupled Earth system models of the kind used to project
global climate change by the Intergovernmental Panel on Climate
Change (IPCC) also show exponential refinement of the horizon-
tal grid resolution in their ocean models (Figure 1, colored dots).
For these models, the doubling time is 3.7 yr, somewhat slower
than for OGCMs because other components of the Earth system
model compete for the computer speedup.
Observations of the global ocean have been revolutionized by
information technology advances too. Figure 1 shows, for example,
the number of annual deep stations with high-quality temperature
measurements (CTD stations deeper than 1,000 m). In the early
2000s, the rate of such observations increased by a factor of 10 as
the global Argo network came online. Today, about 100,000 deep
temperature stations are reported each year.
Consider next the relative rates of growth of OGCM resolution
and deep temperature measurements. Figure 1 shows that OGCMs
outstrip the observations, so there are now around 1,000 hori-
zontal grid cells for every deep temperature station. Put another
way, the average spacing between Argo CTD profiles is 300 km,
ABSTRACT. Ocean circulation models running on the latest supercomputers can cover the globe with resolutions of a few kilome-
ters. These virtual ocean datasets are increasingly realistic and provide insight into processes at scales that are inaccessible with conven-
tional observations. Because these datasets are far too massive for individual researchers to download and analyze, new cloud-based,
open-source, cyberinfrastructure resources are being developed. These tools provide a new analysis paradigm that is scalable, accessible,
and inclusive, and that democratizes access to ocean circulation model output. They also accelerate the pace of analysis of ocean models
and thereby increase the pace of discovery in oceanography. Another challenge concerns the priorities for next-generation ocean circu-
lation models. In particular, to improve circulation model simulations, how should increased supercomputer power be spent? Input on
this question from the oceanographic community is sought.
COMMENTARY
Oceanography | Vol. 38, No. 3
whereas the average spacing between cutting-edge OGCM grid
cells is 1 km. In this sense, cutting-edge OGCMs are becoming
unconstrained by data because the data are sparse compared to
the OGCM degrees of freedom (and notice that this is not true for
the ocean components of cutting-edge IPCC models). The unequal
growth of OGCM resolution and data density reflects the so-called
maturation of computational oceanography (Haine et al., 2021).
Cutting-edge OGCMs are thus becoming more and more valuable
as a resource in oceanography.
OGCM SOLUTIONS AND DATA ACCESS
LLC4320
For example, the 2016 black dot in Figure 1 is a model solution
called LLC4320 (the name refers to the latitude-longitude-cap
horizontal grid with 4320 × 4320 grid cells in each of 13 faces
that tile the global ocean; Rocha et al., 2016; Arbic et al., 2018).
The LLC4320 simulation provides hourly output for one year
in 2011–2012 using the Massachusetts Institute of Technology
OGCM code. A few similar solutions exist using other circula-
tion models and different configurations. Collectively, such solu-
tions are called “nature runs” or “digital twins” of the ocean cur-
rents (Boyes and Watson, 2022; Chen et al., 2023; NASEM, 2024;
Vance et al., 2024). They are useful for many purposes that include
understanding ocean dynamics, designing observing systems,
and machine learning.
Indeed, the oceanographic community is eagerly adopting
these cutting-edge OGCM solutions. To illustrate, the red dots
in Figure 2 show the number of papers each year that utilize the
LLC4320 solution. As in Figure 1, the y-axis of Figure 2 is loga-
rithmic, and straight lines indicate exponential growth. Thus,
Figure 2 shows that the number of LLC4320 papers per year has
grown roughly as an exponential with a doubling time of around
3 yr; dozens of papers now employ the LLC4320 simulation per year.
Despite this growing popularity, the data from LLC4320-type
cutting edge simulations are very challenging to use. The main
problem is the massive size of the datasets, which means that
access to these data is difficult and time-consuming. For LLC4320,
the total uncompressed data volume is four petabytes (one peta-
byte is 1015 bytes), and it takes many months to obtain accounts
on the NASA supercomputers where the LLC4320 simulation
was run. Moreover, the datasets are far too massive for individual
researchers to download and analyze personal copies.
POSEIDON PROJECT
Making the LLC4320 (and similar) simulation data easy to use is
therefore an important priority. Evidence from a neighboring field
in fluid mechanics shows the benefits of opening massive simula-
tion datasets to easy community access. Specifically, the blue dots
in Figure 2 show the number of papers each year that utilize the
Johns Hopkins Turbulence Database (JHTDB; Li et al., 2008). The
JHTDB is an open numerical turbulence laboratory that provides
free access to benchmark numerical solutions for various canonical
turbulence problems. Figure 2 shows that the number of JHTDB
papers per year has also grown exponentially, with a doubling time
of 3.0 yr. In total, more than 6 × 1014 individual model grid cells
have been queried using the JHTDB. A recent paper states that
FIGURE 1. Growth over time of the number of
horizontal grid cells in global ocean general cir-
culation models (OGCMs, see the black dots), the
number of horizontal grid cells in the global cou-
pled climate model from the Intergovernmental
Panel on Climate Change (IPCC, see the colored
dots), and the number per year of deep (greater
than 1,000 m depth) CTD stations. Note that the
y-axis is logarithmic and the straight red lines
indicate exponential growth (the doubling times,
τ2× are shown). The black dot in 2016 is for the
LLC4320 OGCM (see text and Figures 2 and 3).
The three-letter abbreviations in color refer to
the IPCC assessment reports. Modified from
Figure 2 in Haine et al. (2021)
2.5 yr
τ2× =
3.7 yr
τ2×
IPCC model
horizontal
grid cell #
Deep CTD
stations
per year
OGCM horizontal
grid cell #
SAR
FAR
AR4
AR5
AR6
TAR
109
108
107
106
105
104
103
103
104
105
Horizontal Grid Scale (m)
1980
1990
2000
2010
Number
2020
September 2025 | Oceanography
“since its publication, the JHTDB had become a gold standard and
an hypothesis testing tool in the turbulence community” (Shnapp
et al., 2023). This opening up of cutting-edge benchmark simula-
tions has been termed “democratizing the data.” In addition, such
databases significantly reduce carbon emissions by reusing extant
data rather than recomputing them (Yang et al., 2024).
Inspired by the JHTDB, an initiative called the Poseidon Project
has been democratizing the LLC4320 (and similar) OGCM data.
Figure 3 illustrates some key features of the Poseidon Project and
the modular workflows it supports. The left panel of Figure 3 is
a screenshot from the public Poseidon Viewer showing surface
relative vorticity in the LLC4320 North Atlantic Ocean. The first
Poseidon Project design goal is for users to access the data with
very low latency (time delay). The Poseidon Viewer achieves this
goal by visualizing the LLC4320 simulation data interactively,
including on mobile devices in a few seconds (try the Poseidon
Viewer interactive LLC4320 visualization tool).
The second Poseidon Project design goal is to provide a simple
software interface for accessing the data. The Poseidon Project (like
the JHTDB) is hosted on SciServer, which is a collaborative cloud
environment for analysis of extremely large datasets (Medvedev
et al., 2016). The SciServer supports Jupyter notebooks for data
analysis. The middle panel of Figure 3 shows a screenshot of a
SciServer Jupyter notebook using the OceanSpy Python software
to analyze LLC4320 data (Almansi et al., 2019). In this example,
a synthetic hydrographic section is being plotted. The OceanSpy
software is an interface to scalable, open-source tools from the
Pangeo community (which can be used directly in SciServer, for
example, by using xarray without the OceanSpy interface). The
right panel of Figure 3 shows trajectories of drifting particles in
the LLC4320 surface currents. The trajectories were computed in
a SciServer Jupyter notebook using the Seaduck Python software
(Jiang et al., 2023).
The third Poseidon Project design goal is to focus on final com-
putation and rendering of high-quality figures. SciServer achieves
these goals by performing data-proximate, lazy calculations (no
data downloads are necessary, although they are possible) and pro-
viding a robust, stable, fully functional programming environment
in the cloud. Thus, anyone with internet access can interact with
the LLC4320 data, make calculations, and produce publication-
ready figures. This is another sense in which the simulation data
are being “democratized” (made open to everyone).
INTERACTIVE
VISUALIZATION
SYNTHETIC OCEAN
OBSERVATION
LAGRANGIAN
TRAJECTORIES
FIGURE 3. The Poseidon Project makes high-resolution OGCM solutions publicly available, such as the global LLC4320 simulation. Users can interact with
the data using a mobile-friendly, interactive visualization tool and Python application programming interface software such as OceanSpy (Almansi et al.,
2019), which samples the OGCM data using synthetic oceanographic instruments, along with Seaduck (Jiang et al., 2023), which computes Lagrangian tra-
jectories. The data can also be accessed using Pangeo tools such as xarray. Run the Poseidon Viewer interactive LLC4320 visualization tool.
FIGURE 2. Growth over time of the number of papers per
year citing the LLC4320 global OGCM and the Johns Hopkins
Turbulence Database (JHTDB). Note that the y-axis is logarith-
mic (the τ2× doubling time for the annual JHTDB citations is
3.0 yr). The data are taken from the LLC4320 and JHTDB web-
sites as of March 2025.
Oceanography | Vol. 38, No. 3
10
FUTURE OGCM PRIORITIES
Returning to Figure 1, notice that the LLC4320 simulation is
already a decade old. Moore’s law has continued in the years since
NASA computed LLC4320, and the time is ripe to make a new
benchmark cutting-edge calculation. Extrapolating the OGCM
red line in Figure 1 suggests that such a new simulation could
have 3 × 109 horizontal grid cells, which corresponds to a horizon-
tal grid scale of 350 m. This resolution captures part of the unex-
plored regime of submesoscale dynamics in which rotational, iner-
tial, and buoyancy effects are all of similar importance (Taylor and
Thompson, 2023), and which is very hard to observe with current
oceanographic instruments.
Alternatively, the extra computational power could be spent
on other priorities. For example, the simulation could be run for
longer than one year at the same resolution as LLC4320. Or the
initial condition could be improved to avoid transient adjustments
during the simulation. The question is, what are the most import-
ant priorities and, in particular, how should the extra computa-
tional power be spent?
This question was asked during a town hall meeting at the
2024 Ocean Sciences Meeting. Participants in the town hall
responded to an online survey that asked them to rank 11 differ-
ent priorities for designing the next cutting-edge global bench-
mark OGCM simulation. Participants could also write in their
own priorities. Figure 4 shows the results of the survey, summa-
rizing the opinions of 44 respondents (the survey is still open—
take the survey).
The survey results show no consensus for future bench-
mark OGCM solutions because all the priorities were ranked as
important by some respondents and as unimportant by others.
Nevertheless, preferences are clear overall. The most highly
ranked priorities include longer run time and better horizon-
tal and vertical resolution. These priorities are relatively easy to
implement because they require little OGCM code development
and little pre-computation before the main OGCM code is run.
Better model spin-up/initial conditions and better air-sea forc-
ing are also highly ranked. These priorities are harder to imple-
ment because they involve improvements (which need to be
precisely defined) to input data from other large, complex mod-
eling systems. The four middle-ranked priorities are: better con-
straints to observations, better model parametrizations, better
model topography, and better mean circulation and stratifica-
tion. These are desirable scientific goals that are easy to state but
hard to achieve. One reason is that they involve detailed tuning of
OGCM parameters and input data, or improvements to OGCM
software. Another reason is that these priorities are interrelated
because, for example, improving the mean circulation probably
requires better parametrizations and topography, which will inev-
itably improve agreement with observations. Two priorities were
ranked as unimportant overall, namely an ensemble of LLC4320
runs (easy to implement) and better diversity in model code
(relatively easy to implement using existing OGCM systems).
Other priorities listed by a few respondents included adding
biogeochemistry, better documentation, and better comparison
with observations.
OUTLOOK
Given the ongoing advances in computational hardware, software,
and infrastructure, the time is ripe for a new cutting-edge OGCM
solution (or more than one) to be computed. Efforts like LLC4320
and the Poseidon Project require significant resources and there-
fore need broad support from academia, industry, funding agen-
cies, and non-professional oceanographers. To date, these efforts
have been supported by government agencies and private founda-
tions with standalone projects every few years. The need to sustain
open shared cyberinfrastructure like SciServer and digital twins
like LLC4320 is widely recognized (Barker et al., 2019; Grossman,
2023; Le Moigne et al., 2023; NASEM, 2024). The future sources of
support and the pathway for migrating from research project fund-
ing to community infrastructure funding are uncertain, however.
One notable example of a stable, long-term, cloud-based data
analysis environment for ocean sciences is the Mercator Ocean
International and Copernicus Marine Service resource, funded
by the European Commission. It provides real-time global ocean
hindcasts, analyses, and forecasts using ocean circulation models,
in situ and remote observations, and data assimilation (although
not presently at the LLC4320 horizontal resolution). Their focus is
on operational oceanography and the state of the ocean for diverse
stakeholders (von Schuckmann et al., 2024). Apart from aca-
demic users, people have applied the Copernicus Marine Service
PRIORITIES FOR FUTURE GLOBAL
BENCHMARK OGCM SIMULATIONS
FIGURE 4. Results from a 2024 Ocean Sciences Meeting survey on prior-
ities for the next benchmark global OGCM simulation. Forty-four respon-
dents ranked the priorities on the y-axis on a scale of 1 to 12 (1 is the top pri-
ority). The median value is shown with the dotted circle, the 25th and 75th
percentiles are shown with the thick bar, and the thin bars indicate maximum
and minimum values. “Other(s) (write in)” priorities included adding biogeo-
chemistry, better documentation and tutorials, and better evaluation with
observations. Take the survey.
September 2025 | Oceanography
11
to oil spill modeling, shipping route optimization, and maritime
tourism, to name a few. The value of such resources for catalyz-
ing research and expanding the community of users engaged with
ocean currents is tremendous.
As this commentary outlines, the track record of ocean model
advancements is remarkable, with no obvious end in sight. The
knowledge and tools for disseminating and analyzing massive ocean
current simulations currently exist. Decisions on future priorities
with broad community input and engagement are now required.
The prospects for future ocean model improvements and refine-
ment are very bright, and many are straightforward to implement.
REFERENCES
Almansi, M., R. Gelderloos, T. Haine, A. Saberi, and A. Siddiqui. 2019. OceanSpy:
A Python package to facilitate ocean model data analysis and visualization. Journal
of Open Source Software 4(39):1506, https://doi.org/10.21105/joss.01506.
Arbic, B.K., M.H. Alford, J.K. Ansong, M.C. Buijsman, R.B. Ciotti, J.T. Farrar,
R.W. Hallberg, C.E. Henze, C.H. Hill, C.A. Luecke, and others. 2018. A primer on
global internal tide and internal gravity wave continuum modeling in HYCOM
and MITgcm. Pp. 307–392 in New Frontiers in Operational Oceanography.
E. Chassignet, A. Pascual, J. Tintoré, and J. Verron, eds, GODAE OceanView,
https://doi.org/10.17125/gov2018.ch13.
Argo. 2020. Argo float data and metadata from Global Data Assembly Centre
(Argo GDAC), https://doi.org/10.17882/42182.
Barker, M., S.D. Olabarriaga, N. Wilkins-Diehr, S. Gesing, D.S. Katz, S. Shahand,
S. Henwood, T. Glatard, K. Jeffrey, B. Corrie, and others. 2019. The global
impact of science gateways, virtual research environments and virtual laborato-
ries. Future Generation Computer Systems 95:240–248, https://doi.org/10.1016/
j.future.2018.12.026.
Boyes, H., and T. Watson. 2022. Digital twins: An analysis framework and open issues.
Computers in Industry 143:103763, https://doi.org/10.1016/j.compind.2022.103763.
Chen, G., J. Yang, B. Huang, C. Ma, F. Tian, L. Ge, L. Xia, and J. Li. 2023. Toward digital
twin of the ocean: From digitalization to cloning. Intelligent Marine Technology and
Systems 1(3), https://doi.org/10.1007/s44295-023-00003-2.
Fox-Kemper, B., A. Adcroft, C.W. Böning, E.P. Chassignet, E. Curchitser,
G. Danabasoglu, C. Eden, M.H. England, R. Gerdes, R.J. Greatbatch, and others.
2019. Challenges and prospects in ocean circulation models. Frontiers in Marine
Science 6:65, https://doi.org/10.3389/fmars.2019.00065.
Grossman, R.L. 2023. Ten lessons for data sharing with a data commons. Scientific
Data 10:120, https://doi.org/10.1038/s41597-023-02029-x.
Haine, T.W.N., R. Gelderloos, MA. Jiminez-Urias, A.H. Siddiqui, G. Lemson,
D. Medvedev, A. Szalay, R.P. Abernathy, M. Almansi, and C.N. Hill. 2021. Is compu-
tational oceanography coming of age? Bulletin of the American Meteorological
Society 102(8):E1481–E1493, https://doi.org/10.1175/BAMS-D-20-0258.1.
Jiang, W., T.W.N. Haine, and M. Almansi. 2023. Seaduck: A Python package for
Eulerian and Lagrangian interpolation on ocean datasets. Journal of Open Source
Software 8(92):5967, https://doi.org/10.21105/joss.05967.
Le Moigne, J., M.M. Little, R.A. Morris, N.C. Oza, K.J. Ranson, H. Riris, L.J. Rogers, and
B.D. Smith. 2023. Earth System Digital Twin (ESDT) Architecture Framework. NASA
Technical Report, Earth Science Technology Office, NASA, 12 pp.
Lee, T., S. Hakkinen, K. Kelly, B. Qiu, H. Bonekamp, and E. Lindstrom. 2010. Satellite
observations of ocean circulation changes associated with climate variability.
Oceanography 23(4):70–81, https://doi.org/10.5670/oceanog.2010.06.
Li, Y., E. Perlman, M. Wan, Y. Yang, C. Meneveau, R. Burns, S. Chen, A. Szalay,
and G. Eyink. 2008. A public turbulence database cluster and applications to
study Lagrangian evolution of velocity increments in turbulence. Journal of
Turbulence 9:N31, https://doi.org/10.1080/14685240802376389.
Medvedev, D., G. Lemson, and M. Rippin. 2016. SciServer compute: Bringing analy-
sis close to the data. Pp. 1–4 in Proceedings of the 28th International Conference
on Scientific and Statistical Database Management - SSDBM ’16.ACM Press,
https://doi.org/10.1145/2949689.2949700.
Moore, G.E. 1975. Progress in digital integrated electronics. Pp. 11–13 in International
Electron Devices Meeting. IEEE, http://www.eng.auburn.edu/~agrawvd/COURSE/
E7770_Spr07/READ/Gordon_Moore_1975_Speech.pdf.
NASEM (National Academies of Sciences, Engineering, and Medicine). 2024.
Foundational Research Gaps and Future Directions for Digital Twins. The National
Academies Press, Washington, DC, 202 pp., https://doi.org/10.17226/26894.
Riser, S.C., H.J. Freeland, D. Roemmich, S. Wiffjels, A. Troisi, M. Belbéoch, D. Gilbert,
J. Xu, S. Pouliquen, A. Thresher, and others. 2016. Fifteen years of ocean obser-
vations with the global Argo array. Nature Climate Change 6(2):145–153,
https://doi.org/10.1038/nclimate2872.
Rocha, C.B., T.K. Chereskin, S.T. Gille, and D. Menemenlis. 2016. Mesoscale to
submesoscale wavenumber spectra in Drake Passage. Journal of Physical
Oceanography 46(2):601–620, https://doi.org/10.1175/JPO-D-15-0087.1.
Shnapp, R., S. Brizzolara, M.M. Neamtu-Halic, A. Gambino, and M. Holzner. 2023.
Universal alignment in turbulent pair dispersion. Nature Communications 14:4195,
https://doi.org/10.1038/s41467-023-39903-6.
Taylor, J.R., and A.F. Thompson. 2023. Submesoscale dynamics in the upper
ocean. Annual Review of Fluid Mechanics 55:103–127, https://doi.org/10.1146/
annurev-fluid-031422-095147.
Vance, T.C., T. Huang, and K.A. Butler. 2024. Big data in Earth science: Emerging prac-
tice and promise. Science 383(6688), https://doi.org/10.1126/science.adh9607.
Vinogradova, N.T., T.M. Pavelsky, J.T. Farrar, F. Hossain, and L.-L. Fu. 2025. A new
look at Earth’s water and energy with SWOT. Nature Water 3:27–37, https://doi.org/
10.1038/s44221-024-00372-w.
von Schuckmann, K., L. Moreira, M. Grégoire, M. Marcos, J. Staneva, P. Brasseur,
G. Garric, P. Lionello, J. Karstensen, and G. Neukermans, eds. 2024. 8th edition of
the Copernicus Ocean State Report (OSR8). Copernicus Publications, State Planet,
4-osr8, https://doi.org/10.5194/sp-4-osr8.
Yang, X., W. Zhang, M. Abkar, and W. Anderson. 2024. Computational fluid dynam-
ics: Its carbon footprint and role in carbon reduction. Journal of Renewable and
Sustainable Energy 16:055906, https://doi.org/10.1063/5.0217320.
ACKNOWLEDGMENTS
This work was supported by the National Science Foundation under grants 1835640
and 2103874, by the Institute for Data Intensive Engineering and Science at Johns
Hopkins University, and by the Alfred P. Sloan Foundation.
AUTHOR
Thomas W.N. Haine (thomas.haine@jhu.edu), Earth & Planetary Sciences,
Johns Hopkins University, Baltimore, MD, USA.
ARTICLE CITATION
Haine, T.W.N. 2025. Democratize the data: A new way to analyze and design ocean
models. Oceanography 38(3):7–11, https://doi.org/10.5670/oceanog.2025.e303.
COPYRIGHT & USAGE
This is an open access article made available under the terms of the Creative
Commons Attribution 4.0 International License (https://creativecommons.org/licenses/
by/4.0/), which permits use, sharing, adaptation, distribution, and reproduction in
any medium or format as long as users cite the materials appropriately, provide a
link to the Creative Commons license, and indicate the changes that were made
to the original content.
Oceanography | Vol. 38, No. 3
12
FEATURE ARTICLE
IS THERE ROBUST EVIDENCE FOR
FRESHWATER-DRIVEN AMOC CHANGES?
A SYNTHESIS OF DATA, MODELS, AND MECHANISMS
By Sophia K.V. Hines, Nicholas P. Foukal, Kassandra M. Costa, Delia W. Oppo,
Olivier Marchal, Lloyd D. Keigwin, and Alan Condron
INTRODUCTION
The Atlantic Meridional Overturning Circulation (AMOC) plays
a crucial role in regional and global climate. It transports mass
and heat to the Northern Hemisphere (e.g., Frajka-Williams et al.,
2019; Trenberth et al., 2019), is characterized by sinking at sev-
eral locations in the northern North Atlantic (e.g., Talley, 2013),
and thus provides a pathway for sequestering anthropogenic car-
bon for centuries to millennia (e.g., Gebbie and Huybers, 2012;
Brown et al., 2021). Here, we define the AMOC as the upper cell
of the meridional overturning circulation in the Atlantic Ocean. It
moves warm, saline waters northward where these waters lose heat
to the atmosphere, sink, and flow southward as colder and fresher
North Atlantic Deep Water (NADW). Due to positive feedbacks
involving the advection of salt by the northward-flowing branch,
the AMOC may be bistable, as suggested by simplified box models
of meridional overturning circulation (e.g., Stommel, 1961).
Paleoclimate data are consistent with the AMOC having more
than one equilibrium state, and they suggest that the AMOC has
abruptly changed in the past, sometimes in just a few decades. For
example, there is broad evidence from paleoclimate records that
AMOC existed for thousands of years in a reduced state during the
transition out of the last ice age (e.g., McManus et al., 2004; Lynch-
Stieglitz et al., 2014; Rafter et al., 2022), which may have driven
changes in atmospheric circulation, precipitation patterns, and
global surface temperature (e.g., Wang et al., 2001; Anderson et al.,
2009; Cheng et al., 2009; Clark et al., 2012). Some authors have
interpreted these intervals as times of AMOC collapse (McManus
et al., 2004), but paleo data cannot quantitatively reconstruct
the strength of the AMOC, so there is a reluctance within the
paleoceanographic community to use this term. Nevertheless, a
popular schematic in paleoclimate research represents the AMOC
in either an “on” state or an “off” state (Figure 1; Rahmstorf, 2002).
A vigorous, or “on,” state of the AMOC would correspond to the
meridional circulation in the modern Atlantic, which is on the
order of 15–20 Sv (1 Sv = 106 m3 s–1; Frajka-Williams et al., 2019).
A “collapsed,” or “off,” state of the AMOC could occur when surface
waters are not dense enough to sink deeply in the North Atlantic.
Importantly, the upper cell volume flux during a “collapse” can-
not be quantified by paleo data. In this paper, we do not define an
AMOC “collapse” as a complete cessation of circulation but rather
a large and persistent reduction in upper cell volume flux relative
to that of the “on” state.
Global climate models from the International Panel on Climate
Change (IPCC) Coupled Model Intercomparison Project 6
(CMIP6) predict that AMOC will “very likely” decline over the
twenty-first century due to anthropogenic forcing, but it is less
likely that the AMOC will collapse (though the term “collapse”
is not precisely defined in this context; Fox-Kemper et al., 2021).
Some reconstructions of North Atlantic sea surface temperature
and other oceanographic properties during the past ~100 years
were interpreted to mean that the AMOC has weakened during
this period (Thornalley et al., 2018; Caesar et al., 2021), but there is
still significant uncertainty, as other North Atlantic records show
conflicting signals (Kilbourne et al., 2022; Terhaar et al., 2025).
Time series of direct AMOC observations are not long enough
to confidently detect trends in the magnitude of the overturning
ABSTRACT. The Atlantic Meridional Overturning Circulation (AMOC) transports heat to high latitudes and carbon to the deep
ocean. Paleoceanographic observations have led to the widely held view that the strength of the AMOC was significantly reduced at two
intervals during the most recent glacial-to-interglacial transition, with global climate impacts. Climate models predict that the AMOC may
decline in the future due to anthropogenic forcing, but the time periods for modern observations are too short to detect recent trends with
high confidence. To understand the likelihood of future changes in the AMOC, it is important to understand the mechanisms that drove
past changes in AMOC strength. In this paper we review (1) the paleoceanographic proxy data that have led to the widespread view that
the AMOC sharply decreased for periods of several thousand years during the last deglaciation, (2) climate model simulations of the last
deglaciation, with particular attention to their use of fresh water to alter the AMOC, (3) the physical mechanisms that could have driven
past changes in the AMOC, and (4) how insights from past ocean change can inform our understanding of what may happen in the future.
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.
Oceanography | Vol. 38, No. 3
14
mechanisms of past AMOC changes as inferred from paleoceano-
graphic reconstructions and modeling studies, and (2) the implica-
tions of these changes for future AMOC variability.
PALEOCEANOGRAPHIC PROXIES
OF THE AMOC
Paleoceanographic data provide an avenue for extending the rel-
atively short instrumental record and for documenting the state
of the ocean during periods of past climate change. In particu-
lar, they provide a source of empirical information for assessing
the capacity for AMOC to undergo a drastic state change, such as
depicted schematically in Figure 1. We focus on the most recent
glacial-interglacial transition (also called the last “deglaciation”
or “Termination I”), which occurred following the Last Glacial
Maximum (LGM; ~22–18 ka; ka = thousands of years ago) and
ended at the start of the Holocene (10 ka), the current interglacial
period (see Lynch-Stieglitz, 2017, for a broader review of AMOC
proxy data during the last glacial period). During the deglacia-
tion, several abrupt cooling and warming events occurred in the
circum-North Atlantic that have been linked with, respectively,
AMOC decrease and increase through its role in transporting heat
to the high-latitude North Atlantic. After describing the deglacial
sequence of climatic events, we review the evidence that led to the
widely held view that deglacial climate oscillations were linked
to AMOC changes.
The first event, called Heinrich Stadial 1 (HS1; 18–14.7 ka), was a
North Atlantic cold interval notable for high iceberg discharge and
thought to be associated with reduced AMOC strength (Heinrich,
1988; Bond et al., 1992, 1993; Broecker et al., 1992; Broecker, 1994;
Hemming, 2004). Following HS1, the North Atlantic warmed
abruptly at the beginning of the Bølling-Allerød (BA, 14.7–12.6 ka),
thought to be associated with rejuvenation of the AMOC (T. Chen
et al., 2015). The BA was followed by another cold period, the
Younger Dryas (YD, 12.9–11.6 ka), which is also thought to be
associated with a weak AMOC (Broecker, 2003). Finally, the YD
concluded with another abrupt warming, at the beginning of the
Holocene, the relatively stable current warm period.
Paleoceanographic proxies used to make inferences about the
strength and/or structure of the AMOC (and/or the associated deep
counter-rotating cell) are often classified into two basic categories:
water mass proxies and kinematic proxies. Water mass proxies are
thought to record the distinct isotopic or chemical signature of dif-
ferent deep water masses, in particular, northern-sourced NADW
and southern-sourced AABW. Examples of water mass proxies are
the stable carbon isotope ratio (δ13C) of fossil benthic foramin-
ifera (W.B. Curry et al., 1988; Duplessy et al., 1988; W.B. Curry and
Oppo, 2005; Eide et al., 2017), the cadmium/calcium concentra-
tion ratio of fossil benthic foraminifera, from which the seawater
Cd concentration (CdW) is estimated (Boyle, 1988; Marchitto and
Broecker, 2006; Oppo et al., 2018), and the authigenic neodym-
ium isotopic composition (εNd) of sediments and deep-sea corals
(Frank, 2002; Goldstein and Hemming, 2003; Du et al., 2020).
Kinematic proxies are assumed to be more sensitive to flow rate
than water mass proxies. Examples include the radiocarbon age
of fossil benthic foraminifera and deep-sea corals (Keigwin, 2004;
Robinson et al., 2005), the protactinium-231 to thorium-230 activ-
ity ratio of bulk sediment, 231Pa/230Th (Yu et al., 1996; McManus
et al., 2004), and the mean size of sortable silt, SS
— (McCave et al.,
1995, 2017; McCave and Hall, 2006). Note that, albeit conceptually
useful, the distinction between water mass and kinematic proxies is
not without ambiguity: all water properties derived from measure-
ments in the sediment or deep-sea coral are affected by the flow
rate, which would make them “kinematic,” and kinematic proxies
reflect to some degree the composition of water masses.
All proxies are imperfect in the sense that proxy values may be
sensitive to multiple factors, other than the effects of water mass
composition and circulation rate, and each of them has limita-
tions that are necessary to consider when interpreting paleoceano-
graphic records. Some of the water mass tracers (δ13C of dissolved
inorganic carbon and CdW) are functions of biological activity.
The differences in composition between northern- and southern-
sourced deep water reflect regeneration of dissolved inorganic car-
bon and nutrients in the deep ocean as organic matter from the
surface is remineralized at depth. Thus, changes in biological activ-
ity can alter the spatial distribution of these tracers independently
of water mass or circulation rate change. The δ13C of dissolved
inorganic carbon is also affected by air-sea gas exchange (Lynch-
Stieglitz and Fairbanks, 1994; Lynch-Stieglitz et al., 1995).
Radiocarbon measurements on benthic foraminifera or deep-
sea coral samples are corrected for isotopic fractionation (includ-
ing biological fractionation), so biological activity should not
affect the distribution of these measurements. However, radio-
carbon is still a complicated tracer, because surface waters that
sink to depth in high-latitude regions are characterized by dif-
ferent initial radiocarbon values (Key et al., 2004). It takes about
a decade for the carbon isotopic ratios in the ocean mixed layer
to equilibrate with the atmospheric values (Broecker and Peng,
1974; Lynch-Stieglitz et al., 1995; Sarmiento and Gruber, 2006;
Jones et al., 2014). This equilibration time is longer than the resi-
dence time of surface waters in deep-water formation regions, par-
ticularly in the Southern Ocean (Bard, 1988). Processes such as
upwelling and the presence of sea ice, which reduces air-sea gas
fluxes (Prytherch et al., 2017), can lead to large differences between
the radiocarbon activity, or age, of the surface waters and that of
the atmosphere (“surface reservoir age”). Therefore, radiocarbon
records from benthic foraminifera and deep-sea corals reflect
both the water mass transit time from the surface (due to en route
radioactive decay) and the surface reservoir age. Some recent work
(Muglia and Schmittner, 2021) suggests that surface reservoir age
is the primary driver of deep radiocarbon distributions in the
Atlantic Ocean, thus making Atlantic radiocarbon values more a
water mass tracer than a kinematic tracer.
For neodymium isotopes, deep-water values are thought to be
dominated by conservative mixing, but sedimentary sources can
September 2025 | Oceanography
15
also alter isotopic compositions along deep-water flow paths, par-
ticularly in poorly ventilated basins, such as the deep Pacific and
Indian Oceans (Abbott et al., 2015; Du et al., 2018, 2020). Certain
types of sediment (particularly volcanic ash and ice-rafted debris)
can also be more reactive and prone to delivering non-conservative
additions of Nd to seawater (Wilson et al., 2013; Blaser et al., 2016;
Du et al., 2016).
The use of bulk sediment 231Pa/230Th as a circulation tracer relies
on the theoretical expectation that, while 231Pa and 230Th are pro-
duced at approximately uniform rates in the ocean (from the decay
of 235U and 234U, respectively), 231Pa is in general scavenged less
intensively by sinking particles than 230Th and hence is more sensi-
tive to circulation than 230Th (Henderson and Anderson, 2003). As
a result, the ratios of the two isotopes in sinking particles and sedi-
ment would be dependent on lateral transport of water (i.e., on the
AMOC), with stronger transport leading to lower 231Pa/230Th in
the underlying sediment. However, the 231Pa/230Th ratio of marine
particles in the water column has been found to vary with their
chemical compositions (e.g., Chase et al., 2002; Hayes et al., 2015)
and with sediment lateral redistribution (S.Y.-S. Chen et al., 2021),
complicating its use as an AMOC proxy.
One of the most widely cited reconstructions used as evidence
of AMOC change across the deglaciation is the 231Pa/230Th record
from the Bermuda Rise in the Northwest Atlantic (Figure 2e;
McManus et al., 2004). This record shows an abrupt increase in
231Pa/230Th to values close to the production ratio (which would
imply very little lateral flow out of the North Atlantic) during
HS1, and another smaller increase during the Younger Dryas. The
high 231Pa/230Th values during HS1 were attributed to a dramati-
cally weakened AMOC. Other 231Pa/230Th data from across the
North Atlantic broadly support this interpretation (Ng et al., 2018).
Compilations of benthic foraminifera δ13C from across the deep
Atlantic show low values during HS1 and an abrupt increase at the
start of the Bølling-Allerød (Figure 2g; Thiagarajan et al., 2014;
Lynch-Stieglitz et al., 2014; Lynch-Stieglitz, 2017), values that have
been interpreted as the resumption of a deep AMOC at the Bølling-
Allerød from a weaker state during HS1. Radiocarbon data from
the Northwest Atlantic also show an abrupt decrease in apparent
ventilation age at the start of the Bølling-Allerød from “older” val-
ues during HS1 and another pulse of old water at the YD (Figure 2f;
Robinson et al., 2005; Hines, 2017; Rafter et al., 2022). Compiled
εNd data are also consistent with a weakened AMOC during HS1
and the YD (Figure 2h; Pöppelmeier et al., 2019; Du et al., 2020),
although these data are less supportive of a fully collapsed AMOC.
The processes that might decouple variations in each proxy from
AMOC differ among proxies. Therefore, if these processes were the
dominant control on the deglacial variability in each record, we
would not expect them to correlate with one another. The finding
that many deglacial ocean circulation proxy records share com-
mon features at approximately the same times is apparent evidence
for changes in AMOC over the deglaciation. In other words, while
each proxy record could be explained by processes other than
circulation, the most parsimonious explanation for all the records
taken together would be that AMOC was abruptly reduced (or col-
lapsed) during HS1 and the YD.
This interpretation is also consistent with paleoclimate records
from terrestrial archives, including the oxygen isotopic composition
of Greenland ice cores (Figure 2a; North Greenland Ice Core Project
Members, 2004); the oxygen isotopic composition of Chinese spe-
leothems (Figure 2d; Wang et al. 2001; Cheng et al., 2009, 2016),
which records coeval shifts in atmospheric circulation patterns;
10
15
20
Age (ka)
HS 1
YD B/A
LGM
Holocene
-16
-14
-12
-10
εNd
-0.5
0.0
0.5
1.0
Benthic δ13C (‰)
1000
2000
3000
14C Ventilation Age (yr)
0.05
0.06
0.07
0.08
0.09
0.10
231Pa/230Th
-12
-10
-8
-6
Hulu cave δ18O (‰)
10
IRD (103 grains/g)
150
200
250
300
Atm. CO2 (ppm)
-50
-45
-40
-35
-30
NGRIP δ18O (‰)
FIGURE 2. Paleoclimate records across the deglaciation. (a) Northern
Hemisphere temperature from NGRIP δ18O of ice (North Greenland Ice Core
Project Members, 2004; Andersen et al., 2006; Rasmussen et al., 2014).
(b) Atmospheric CO2 from the West Antarctic Ice Sheet (Marcott et al., 2014).
(c) Ice-rafted debris concentration in the Northwest Atlantic at sites DY081-
GVY001 (solid) and EW9309-37JPC (dashed) (Zhou et al., 2021). (d) Hulu
cave δ18O (Cheng et al., 2016). (e) 231Pa/230Th from the Bermuda Rise (thin
lines: McManus et al., 2004; Lippold et al., 2009, 2019) and across the North
Atlantic (thick line: Ng et al., 2018). (f) Compiled deep Atlantic 14C venti-
lation age (Rafter et al., 2022). (g) Deep North Atlantic δ13C (as in Lynch-
Stieglitz et al., 2014; data from Hodell et al., 2008; Tjallingii et al., 2008;
Mulitza et al., 2008; Zarriess and Mackensen, 2011; Shackleton et al., 2000;
Skinner and Shackleton, 2004; Skinner et al., 2007). (h) εNd from the
Blake Bahama Outer Ridge (Pöppelmeier et al., 2019). YD = Younger Dryas.
B/A = Bølling-Allerød. HS 1 = Heinrich Stadial 1. LGM = Last Glacial Maximum.
IRD = Ice-rafted debris.
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.
September 2025 | Oceanography
17
FRESHWATER MECHANISMS FOR DRIVING
ABRUPT CHANGES IN THE AMOC
Given that geologic reconstructions suggest that HS1 and the YD
were not times of accelerated melting of Northern Hemisphere
ice sheets and elevated freshwater fluxes to the North Atlantic
(Figure 3a), alternative mechanisms for AMOC weakening at
these times must be sought. The mechanisms driving AMOC
reduction at HS1 and the YD need not have been the same, and
paleoceanographic data are consistent with different magnitudes
of AMOC change at each event, with HS1 thought to be the larger
and longer reduction of the two (Ng et al., 2018).
Heinrich events were associated with massive iceberg dis-
charges from the Laurentide Ice Sheet (Ruddiman, 1977; Heinrich,
1988; Broecker, 1994; Hemming, 2004), so it is possible that fresh
water from melting icebergs played an important role. Unlike the
deglacial meltwater that enters the ocean directly in liquid form,
icebergs can travel much farther from the coast before they dis-
integrate (Fendrock et al., 2022). The paths of large icebergs orig-
inating from terrestrial ice sheets can be tracked by ice-rafted
debris (IRD), which consists of coarse grains of continental origin
that are embedded in the icebergs and deposited on the seafloor
as the icebergs melt. In general, IRD is found most prominently
in marine sediment cores collected from between ~40°N and
50°N in the Atlantic Ocean (Ruddiman, 1977); however, smaller
amounts of IRD have been found much farther north, including
in the Nordic Seas and south of Iceland (e.g., Elliot et al., 2001;
Thornalley et al., 2010), and to the south on the Bermuda Rise
(Keigwin and Boyle, 1999). Therefore, the supply of fresh water
from melting icebergs could be a mechanism for reducing the
AMOC (Broecker, 1994). This hypothesis is supported by low
δ18O values measured in fossil planktonic foraminifera (indica-
tive of low salinity) from Heinrich layers within the main IRD
belt proposed by Ruddiman (Bond et al., 1992; Hemming 2004).
However, as yet, clear evidence of low salinity farther north has
not been found.
The Younger Dryas is another period of IRD deposition in the
North Atlantic (e.g., Zhou and McManus, 2024) and the Arctic
Ocean (e.g., Hillaire-Marcel et al., 2013; Lakeman et al., 2018),
although IRD fluxes at this time appear to have been smaller than
at HS1 (e.g., Zhou and McManus, 2024). While the YD is not asso-
ciated with a time of widespread ice sheet melting according to
sea level data and ice sheet models (Tarasov and Peltier, 2002;
Tarasov et al., 2012; Lambeck et al., 2014; Briggs et al., 2014), it
is notably associated with the abrupt draining of Lake Agassiz, a
proglacial lake formed by the melting Laurentide Ice Sheet that
sat at the boundary of Minnesota, North Dakota, Ontario, and
Manitoba (Broecker et al., 1989; Teller et al., 2002). It was initially
thought that Lake Agassiz drained east at the YD, directly into the
North Atlantic via the St. Lawrence River (Broecker et al., 1989;
Clark et al., 2001), but direct evidence for this has been elusive,
and more recent studies suggest that the lake instead drained north
into the Arctic via the Mackenzie River (Tarasov and Peltier, 2005;
Murton et al., 2010; Keigwin et al., 2018; Süfke et al., 2022). This
result is supported by model simulations, which show that fresh
water discharged into the ocean from the St. Lawrence River does
not immediately spread offshore but is instead transported away
from the subpolar North Atlantic in boundary currents, into the
subtropical gyre. Meltwater from the Mackenzie Valley into the
Arctic Ocean is more likely to reach deep-water formation regions
directly, regardless of whether the Canadian Arctic Archipelago is
ice-covered or open (Condron and Winsor, 2012).
The focus has often been on deep convection regions when it
comes to deglacial freshwater-driven perturbations of the AMOC,
whether the fresh water is delivered by icebergs or directly in liq-
uid form; however, recent physical oceanographic observations
and modeling indicate that capping water mass transformation
along the boundary currents or reducing the zonal density gradi-
ent across the mid-latitude North Atlantic (Buckley and Marshall,
2016; Yeager et al., 2021; Chafik et al., 2023; Frajka-Williams
et al., 2023) may be more important for disrupting the AMOC.
During the deglaciation, meltwater introduced to the western sub-
polar North Atlantic could have been entrained offshore along
the northern flank of the western boundary currents that consti-
tute the upper limb of the AMOC, including the Gulf Stream and
the North Atlantic Current (the eastward extension of the Gulf
Stream). This entrained meltwater could significantly alter the
density gradients across these powerful currents and hence reduce
their strength and associated heat transport (e.g., Yeager et al.,
2021; Madan et al. 2024).
0.0
0.2
0.4
0.6
0.8
Freshwater Flux (Sv)
Younger
Dryas
Heinrich
Stadial 1
GLAC-1D
TraCE-21k
10
15
20
25
AMOC (Sv)
-10
-12
-14
-16
-18
-20
Age (ka)
iLOVECLIM
TraCE-21k
B/A
FIGURE 3. Transient model simulations of AMOC across the deglaciation.
(a) Freshwater flux from ice sheet model simulation GLAC-1D with fresh-
water flux time series in the TraCE-21k model (Liu et al., 2009; Bouttes
et al., 2023). (b) AMOC strength calculated as the maximum streamfunction
between 20°N and 50°N below 500 m from TraCE-20k (Liu et al., 2009) and
iLOVECLIM (Bouttes et al., 2023).
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;