September 2025

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;

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98