Relatively warm deep-water formation persisted in the Last Glacial Maximum

Wharton, Jack H.; Kozikowska, Emilia; Keigwin, Lloyd D.; Marchitto, Thomas M.; Maslin, Mark A.; Ziegler, Martin; Thornalley, David J. R.

doi:10.1038/s41586-025-10012-2

Download PDF

Article
Open access
Published: 21 January 2026

Relatively warm deep-water formation persisted in the Last Glacial Maximum

Nature volume 650, pages 116–122 (2026)Cite this article

15k Accesses
1 Citations
51 Altmetric
Metrics details

Subjects

An Author Correction to this article was published on 27 January 2026

This article has been updated

Abstract

The Last Glacial Maximum (19–23 thousand years ago) was characterized by low greenhouse gas concentrations and continental ice sheets that covered large parts of North America and Europe¹. Glacial climate was therefore very different, with colder global mean temperatures and an increased Equator-to-pole temperature gradient, probably resulting in stronger westerlies². However, the state of the deep North Atlantic Ocean under these glacial climate forcings remains uncertain^3,4,5,6, particularly owing to the rarity of deep-ocean temperature and salinity constraints. Here we show that the temperature of the glacial deep (>1.5 km) Northwest Atlantic was approximately 0–2 °C (only 1.8 ± 0.5 °C (2 s.e.) colder than today), and, after accounting for the whole-ocean change, seawater δ¹⁸O was 0.3 ± 0.1‰ (2 s.e.) higher and can be traced back to the surface subtropics via the subpolar Northeast Atlantic and Nordic Seas. Together, our hydrographic data reveal the thermal and isotopic structure of the deep Northwest Atlantic and suggest sustained production of relatively warm and probably salty North Atlantic Deep Water during the Last Glacial Maximum. Furthermore, our results provide updated constraints for benchmarking Earth system models used to project future climate change.

Prevalent North Atlantic Deep Water during the Last Glacial Maximum and Heinrich Stadial 1

Article Open access 06 May 2025

Active Nordic Seas deep-water formation during the last glacial maximum

Article 24 October 2022

Rapid subsurface warming in the subpolar North Atlantic from freshening

Article Open access 27 March 2026

Main

At present, the deep North Atlantic Ocean is predominantly thermally stratified, with relatively warm (2–4 °C) and salty (34.9 practical salinity units (PSU)) North Atlantic Deep Water (NADW) occupying depths between approximately 1 km and 4 km, and colder (0–1 °C) and fresher (34.7 PSU) Antarctic Bottom Water (AABW) below^7,8 (Fig. 1). The hydrographic properties (temperature and salinity) of both NADW and AABW are determined by the processes that govern their formation. NADW is formed from the cooling and densification of relatively warm and salty surface Atlantic waters sourced from the western subtropics and carried northeastwards via the North Atlantic Current (NAC). By contrast, AABW is formed from fresher waters on the continental shelves around Antarctica via intense cooling and brine rejection from sea-ice growth⁹.

**Fig. 1: Modern Atlantic temperature and salinity.**

It remains uncertain whether and how the hydrographic structure of the deep North Atlantic changed during the Last Glacial Maximum (LGM). Modelling results suggest that the glacial climate and ice sheets would have driven stronger wind-driven gyre circulation, resulting in increased northwards salt transport and oceanic heat loss over the subpolar gyre, both of which would have favoured enhanced NADW production^10,11. Conversely, increased glacial sea ice and calving glaciers would have reduced oceanic heat loss, potentially supressing deep-water formation¹². Traditionally, palaeoceanographic nutrient proxies have been used to infer a glacial shoaling of the boundary between glacial NADW and AABW to around 2 km (ref. ³). However, recent modelling studies using stable carbon and oxygen-isotope ratios (δ¹⁸O (¹⁸O/¹⁶O) and δ¹³C (¹³C/¹²C)), nutrient and carbonate-ion concentration proxies (that is, Cd/Ca and B/Ca), and water-mass indicators (εNd (¹⁴³Nd/¹⁴⁴Nd); albeit with uncertainties regarding non-conservative behaviour and end-member non-stationarity) suggest this may be an overestimate^4,5,13. Recent results have also revealed a deeper northern subtropical gyre and associated subtropical mode waters (STMWs)¹⁴, and an abyssal deep-water mass of northern origin below approximately 5 km (ref. ¹⁵).

The associated hydrographic properties of the deep North Atlantic during the LGM are also uncertain—at present, palaeoceanographic reconstructions of temperature and salinity are limited to a small number of discrete sites. Among these, the most frequently cited are derived from sedimentary pore waters, which suggest that the glacial equivalents of NADW and AABW were both near freezing (−1.1 °C and −1.9 °C, respectively), with variations in salinity—rather than temperature—driving glacial stratification⁶. However, further investigations using inverse modelling have queried the assumptions underpinning these data (for example, the use of the prior condition that temporal changes in regional deep-ocean salinity scale with global benthic foraminiferal δ¹⁸O and mean sea-level history), highlighting the need to attribute larger uncertainties to these pore-water-based temperature and salinity estimates^16,17. Therefore, new palaeoceanographic proxy reconstructions are necessary to better constrain the hydrographic properties of the deep North Atlantic during the LGM.

Here we present geochemical reconstructions of seawater temperature and δ¹⁸O (henceforth, δ¹⁸O_sw) from 13 marine sediment cores collected at Cape Hatteras, Blake Outer Ridge, Bermuda Rise and Corner Rise in the Northwest Atlantic. These cores form a depth transect spanning water depths between approximately 1.5 km and 5 km, and are complemented by 3 additional cores retrieved from south of Iceland in the Northeast Atlantic (Fig. 1, Extended Data Fig. 1 and Extended Data Table 1). Not only do these data provide insights into the hydrography of the deep North Atlantic during the LGM but also they offer a means to constrain its vertical structure, given that both temperature and δ¹⁸O_sw behave as conservative tracers.

At present, our Northwest Atlantic sites are predominantly bathed by NADW, which is composed of Labrador Sea Water (LSW)—formed in the Irminger, Iceland and Labrador seas—and the downstream products of Iceland–Scotland Overflow Water and Denmark Strait Overflow Water, both formed in the Nordic Seas¹⁸. Our Northwest Atlantic core sites thus encompass the main export pathway for deep waters formed across the subpolar North Atlantic. In addition, sites from <2.5 km at Blake Outer Ridge allow us to reconstruct the properties of the deep subtropical gyre and associated STMWs, which extended to depths of 2 km to 2.5 km during the LGM¹⁴. Our Northeast Atlantic cores are situated along the main flow path of Iceland–Scotland Overflow Water as it transits through the Iceland Basin before combining with LSW and Denmark Strait Overflow Water to form NADW.

We reconstructed deep-ocean temperatures from the North Atlantic during the mid-to-late Holocene (2–6 thousand years ago (ka) before present (BP)) and the LGM by measuring trace-metal ratios (Mg/Ca and Mg/Li) in multiple species of benthic foraminifera (Methods), both of which are positively correlated with seawater temperature during calcification. In particular, we focused on aragonitic species, and for calcitic taxa, we focused on infaunal species, whose magnesium (Mg) partitioning during calcification is thought to be less affected by a low carbonate-ion saturation state (ΔCO₃²⁻; Methods), as they calcify under the influence of surrounding pore waters, which can be buffered¹⁹. For the aragonitic species, Hoeglundina elegans, and the deep infaunal species, Globobulimina affinis, we converted Mg/Li and Mg/Ca to temperature using published calibrations^20,21, respectively (Fig. 2). For Melonis spp. and Cassidulina neoteretis, we developed a single core-top calibration using data from previous calibration studies (Extended Data Fig. 2), which show that numerous low-Mg, shallow infaunal benthic foraminifera show similar temperature sensitivities (approximately 0.1 mmol mol⁻¹ °C⁻¹; Fig. 2 and Methods). The resultant temperature estimates from these different species are consistent and directly comparable, indicating no significant inter-species bias. We therefore averaged these multi-species data to derive mid-to-late Holocene and LGM mean-temperature estimates for each core, reducing the overall uncertainty (Methods). As ΔCO₃²⁻ may not be completely buffered in sedimentary pore waters, we also generated additional independent temperature estimates by measuring benthic foraminiferal clumped isotopes (Δ₄₇; Methods and Extended Data Fig. 3). Both trace-metal- and Δ₄₇-based temperature estimates were then combined with paired stable oxygen-isotope measurements¹⁴ to derive independent trace-metal- and Δ₄₇-based estimates of δ¹⁸O_sw (Methods), which is correlated with salinity in the modern ocean.

**Fig. 2: Benthic foraminiferal trace-metal temperature calibrations used in this study.**

Temperature and δ¹⁸O_sw reconstructions

Mid-to-late Holocene multi-proxy temperature and δ¹⁸O_sw estimates from similar depths between 1 km and 4.5 km show good agreement, generally showing temperatures and δ¹⁸O_sw ranging from 2 °C to 4 °C and from 0‰ to 0.25‰, respectively (Fig. 3a,b). These data also agree well with both modern observational data^7,8 and ostracod-based Mg/Ca temperature estimates^22,23,24 (Fig. 3a) from the mid-to-late Holocene. If we calculate density using temperature and a single δ¹⁸O_sw–salinity relationship, this incorrectly results in a density inversion with depth, with some sites offset from modern observations²⁵ (Fig. 3c and Methods). This is because different δ¹⁸O_sw–salinity relationships are applicable to the various subcomponents of NADW (for example, LSW versus overflow waters). To avoid this uncertain complexity, we do not make use of a density conversion to the mid-to-late Holocene data, nor is it possible to accurately do so for the glacial data either. Overall, these data imply that a modern-like circulation was prevalent during the mid-to-late Holocene, with relatively warm and salty NADW present down to at least 4 km in the Northwest Atlantic. Given that our multi-proxy data appear to be faithfully capturing in situ deep-ocean temperature and δ¹⁸O_sw, we now apply them to the LGM.

**Fig. 3: Hydrographic structure of the Northwest Atlantic during the mid-to-late Holocene and LGM.**

Apart from site ODP-172-1055 (approximately 1.8 km), which is substantially warmer (about 5 °C), glacial temperature and δ¹⁸O_sw reconstructions are relatively uniform, ranging from 0 °C to 2 °C and from 1.25‰ to 1.75‰, respectively (Fig. 3d,e). Similar to the mid-to-late Holocene, there is strong agreement—within the margin of error—between our Mg/Ca- and Δ₄₇-based temperature and δ¹⁸O_sw estimates, as well as with the few other independent data from the Northwest Atlantic^22,23,24. In particular, the relative warmth of site ODP-172-1055 (33° N) is consistent with the influence of a deeper subtropical gyre extending down to below approximately 2 km during the LGM¹⁴ (Fig. 3d,f). Our deeper data are also consistent with previous work that suggests that the glacial deep Northwest Atlantic was dominated by NADW^5,13, rather than being occupied by distinct northern- and southern-source waters. If we calculate glacial densities using the modern NADW δ¹⁸O_sw–salinity relationship, this again results in a density inversion (Fig. 3f), which implies that the glacial North Atlantic was filled with multiple modes of glacial NADW⁵, probably formed at different locations and characterized by different δ¹⁸O_sw–salinity relationships. For example, our data from sites at 3–4 km hint at being colder and having lower δ¹⁸O_sw, which may be consistent with a deep-water-formation region more affected by sea-ice formation and brine-rejection processes, such as proposed for the glacial Arctic Mediterranean⁵. Furthermore, our abyssal (>5 km) constraints indicate a temperature and δ¹⁸O_sw similar to our other deep North Atlantic data, which is consistent with the inference from previous carbon-isotope (δ¹³C and ¹⁴C) evidence of a northern-origin abyssal water mass in the Northwest Atlantic during the LGM¹⁵.

Warm and salty glacial North Atlantic

Notably, our reconstructed glacial temperatures from depths currently bathed by NADW (1.5–4 km, excluding ODP-172-1055) in the Northwest Atlantic, are on average less than 2 °C colder than modern temperatures at equivalent depths (ΔT_Mg/Ca = −1.63 ± 0.44 °C; ΔT_(Δ47) = −2.02 ± 0.95 °C; Fig. 4a). In addition, glacial temperature constraints from south of Iceland also suggest similarly modest cooling in the Northeast Atlantic (ΔT_Mg/Ca = −1.78 ± 0.56 °C; Fig. 4a). These data stand in contrast to glacial estimates derived from sedimentary pore waters⁶, which suggest near-freezing conditions at 2 sites (2.2 km and 4.6 km; Fig. 4a); thus, our results instead indicate that the deep North Atlantic remained relatively warm (approximately 0–2 °C) during the LGM.

Fig. 4: Summary of deep-ocean temperature and δ18Osw changes in the North Atlantic between the LGM and the modern ocean. — **Fig. 4: Summary of deep-ocean temperature and δ¹⁸O_sw changes in the North Atlantic between the LGM and the modern ocean.**

In comparison, average-ice-volume-corrected glacial δ¹⁸O_sw, hereafter δ¹⁸O_sw-ivc (Methods), is higher than equivalent modern δ¹⁸O_sw in the Northwest Atlantic (Δδ¹⁸O_{sw-ivc (Mg/Ca)} = 0.38 ± 0.14‰; Δδ¹⁸O_sw-ivc(Δ47) = 0.23 ± 0.24‰; Fig. 4b) and Northeast Atlantic (Δδ¹⁸O_{sw-ivc(Mg/Ca)} = 0.51 ± 0.19‰). This suggests that previous glacial deep North Atlantic δ¹⁸O_sw estimates derived from sedimentary pore waters were too low, probably owing to the greater methodological uncertainties and assumptions now recognized with this method, when applied to the North Atlantic^16,17. Therefore, although deriving reliable palaeosalinity estimates for the North Atlantic during the LGM is challenging—owing to the spatial and temporal variability in the δ¹⁸O_sw–salinity relationship²⁶—higher glacial δ¹⁸O_sw-ivc is probably due to a combination of variable δ¹⁸O_sw–salinity relationships and saltier NADW relative to the Holocene.

We also compared our δ¹⁸O_sw-ivc reconstructions with the limited number of available isotope-enabled LGM simulations, which produce a relatively wide range of δ¹⁸O_sw-ivc values for NADW (approximately −0.2‰ to 0.5‰; Methods and references therein). Of these, the iPOP2 (Parallel Ocean Program version 2) simulation²⁷ shows the best agreement with our proxy data, simulating high near-surface δ¹⁸O_sw-ivc values in the western subtropical North Atlantic (1‰), which feed through into NADW at depth. However, neither STMW nor NADW extend as deep in the simulations compared with proxy reconstructions^5,14, probably owing to limitations in the ability of models to simulate deep-water-formation processes, in part linked to their spatial resolution²⁸.

Sustained glacial deep-water production

A comparison of our glacial deep-ocean δ¹⁸O_sw-ivc data with equivalent glacial data from sites along the modern pathway of NADW and its near-surface source waters reveals that the high δ¹⁸O_sw-ivc signature of glacial NADW can be traced along that same route. Figure 5 shows how high δ¹⁸O_sw-ivc—recorded consistently by multiple planktic foraminiferal species that occupy and reflect the properties of the subsurface upper ocean that is the source for NADW—is traceable from the western subtropical Atlantic, northeastwards along the path of the Gulf Stream and NAC into the likely deep-water-formation regions of the glacial subpolar North Atlantic and Nordic Seas, then back south at depth into the deep Northwest Atlantic. We therefore infer that there was sustained deep-water formation in the subpolar North Atlantic (and southern intermediate-depth Nordic Seas) during the LGM, which is consistent with most glacial climate model simulations (for example, ref. ¹⁰). Furthermore, given that glacial Antarctic Intermediate Water (Extended Data Fig. 5a) and equatorial Atlantic surface waters²⁹, both of which feed the subtropical North Atlantic, were characterized by lower δ¹⁸O_sw-ivc (−0.4‰ to 0.3‰ and 0.4‰ to 0.5‰, respectively), we infer that processes occurring in the subtropics contributed to the particularly high δ¹⁸O_sw-ivc signature along the Gulf Stream–NAC–NADW pathway.

Fig. 5: High δ18Osw-ivc signature reconstructed at multiple sites in the North Atlantic reveals the upstream pathway of glacial Northwest Atlantic deep waters. — **Fig. 5: High δ¹⁸O_sw-ivc signature reconstructed at multiple sites in the North Atlantic reveals the upstream pathway of glacial Northwest Atlantic deep waters.**

Subtropical hydroclimate forcing

As our reconstructed glacial δ¹⁸O_sw-ivc from the Northwest Atlantic is approximately 0.5‰ higher than the Holocene, its source waters must have been subject to additional enrichment by hydrological fractionation processes such as negative precipitation minus evaporation (P − E). In the western subtropical Atlantic, the balance between precipitation and evaporation is probably the predominant control on δ¹⁸O_sw (ref. ³⁰), with negative P − E causing higher δ¹⁸O_sw. Therefore, we infer that the high glacial δ¹⁸O_sw-ivc reconstructed from the Gulf Stream region of the western subtropical gyre (Fig. 5) indicates that the regional P − E during the LGM was negative relative to the Holocene, owing to decreased precipitation and/or increased evapotranspiration. Such a glacial hydroclimatic regime is supported by terrestrial proxy data, indicating reduced precipitation over North America and the Caribbean^31,32, and climate models also consistently simulate negative P − E over the North Atlantic Subtropical Gyre during the LGM, primarily owing to increased evaporation driven by stronger, cold and dry glacial winds^2,14. Although there are complexities in relating δ¹⁸O_sw to salinity²⁶, these model results suggesting lower P − E over the subtropical gyre provide support for our inference that the high δ¹⁸O_sw surface- and deep-water values are recording the higher glacial salinity of the NAC and NADW, and that glacial deep-water production was sustained by the continued supply of salty (and warm) upper-ocean waters to the subpolar North Atlantic^10,11.

Surface–deep-ocean decoupling

In light of these relatively warm temperature constraints, a logical question that arises is why the deep North Atlantic did not cool further, given that much of the surface subpolar region was close to freezing during winter²⁹. Today, the deep Nordic Seas are near the freezing point (−1 °C to −1.5 °C) owing to deep open-ocean convection, and thus have little potential for further cooling. Actually, reconstructed glacial bottom-water temperatures from the intermediate and deep Nordic and Arctic seas instead suggest temperatures of 0–2 °C (up to 3 °C warmer than today³³; Extended Data Fig. 5c and references therein)—a consequence of expanded glacial sea ice limiting heat loss to the atmosphere and thereby reducing open-ocean convection. Previous work has shown that, despite the inferred reduction in Nordic Seas convection, the overflow of dense waters from the Nordic Seas into the subpolar North Atlantic persisted during the LGM⁵, probably driven by processes such as brine rejection or supercooling under glacial ice shelves³⁴. Although the overflow waters themselves were not colder than today, the slightly lower temperatures of their downstream products—as reconstructed in this study—probably reflect the entrainment of colder upper and intermediate waters within the glacial subpolar North Atlantic³⁵.

At present, most deep-water formation occurs south of the Greenland–Scotland Ridge, in the subpolar North Atlantic, through convection processes and entrainment¹⁸, and palaeoceanographic evidence indicates that this subpolar overturning was sustained and/or strengthened during the LGM, with an overall southwards shift in the locus of deep-water formation³⁶. This overturning would have been facilitated, in part, by the relatively high inferred salinity of the glacial NAC, which would have promoted deep convection before surface waters cooled to freezing point. In addition, simplified conceptual models have shown that as climate cools, it becomes increasingly difficult to form very cold NADW³⁷. This is because the buoyancy flux associated with surface cooling is reduced at lower temperatures, and the upper water column becomes more haline stratified, which inhibits deep convection. Consequently, deep-water formation shifts farther south to relatively warmer regions, producing deep waters with temperatures around 2–3 °C—broadly consistent with our glacial deep-ocean temperature reconstructions. A more extreme version of this process is the cooling of water at the northern edge of the subtropical gyre, forming deeper STMW³⁸. As STMW forms from warm subtropical waters, our data showing STMW temperatures of approximately 5 °C imply substantial heat loss—highlighting the importance of this low-latitude deep-water formation in glacial meridional heat transport.

This concept of continued glacial NADW formation aligns with a recent multi-proxy synthesis that concludes that glacial NADW was composed of multiple deep-water masses sourced from different regions throughout the North Atlantic and Arctic oceans and formed by a variety of mechanisms⁵. Here we show that the integrated downstream product of these different formation modes is hydrographically homogeneous, characterized by temperatures of 0–2 °C and a δ¹⁸O_sw-ivc of 0.25–0.75‰, the latter of which suggests that glacial NADW was probably also relatively salty compared with the modern.

LGM and future ocean implications

Our temperature and δ¹⁸O_sw transects reveal sustained production of relatively warm and probably salty NADW during the LGM. However, as δ¹⁸O_sw serves only as an approximate indicator of salinity, it remains unclear whether the glacial ocean was thermally—as it is today—or haline stratified, as inferred from pore-water measurements⁶. In light of recent estimates that the mean ocean temperature (MOT) during the LGM was approximately 2.3 ± 0.5 °C (1σ) colder than today³⁹, our data—showing that NADW was only 1.8 ± 0.5 °C colder—imply that other deep-water masses such as AABW or Pacific Deep Water must have cooled by more than NADW to account for the overall MOT decrease. Furthermore, given that the densest class of modern AABW found close to Antarctica (T_AABW-ANT = −0.5 °C (ref. ⁷)) is already near the freezing point of seawater (approximately −2 °C), its capacity for further cooling is limited; therefore, Pacific Deep Water (T_PDW = 1–1.5 °C) seems a more likely candidate for greater cooling. However, not all AABW, such as that found in the South Atlantic, is as cold (T_AABW-SATL = 0–1 °C), and thus some portions probably had greater cooling potential. Alternatively, and/or in addition, an increased relative volume of AABW globally, such as in the deep Pacific, could explain the glacial reduction in MOT⁴⁰, despite the relatively modest cooling of NADW. Moreover, it has been suggested that NADW supercooling was a trigger for glacial inception (for example, ref. ⁴¹); however, the presence of relatively warm NADW in the glacial North Atlantic implies that other processes in the Southern Ocean may have had a more important role in thermally isolating Antarctica.

This study demonstrates that, despite the colder climate state of the LGM, there was sustained production of NADW that was only 1.8 ± 0.5 °C colder than today. This persistence was probably driven by continued buoyancy loss, enabled by sufficient cooling of a steady supply of warm, salty water transported to mid-to-high latitudes via the NAC, in part maintained by the wind-driven gyre circulation¹⁰. Oceanographic observations of the twenty-first century⁴², and modelling studies of future climate scenarios⁴³, suggest that NADW production can occur at different locations than those that were typical during the twentieth century. The ability to accurately predict the future of NADW thus depends on climate models correctly simulating deep-water formation processes across a range of climatic and geographic settings—an area where our data can provide valuable test-bed constraints.

Methods

Age models and core sampling

The Northwest Atlantic Ocean samples are the same as those used by ref. ¹⁴; therefore, we adopt the same age models and same sampling strategies (Extended Data Table 1). For our Northeast Atlantic cores, we also use previously published age models^51,52. Stratigraphic information for each core, including multi-species benthic and planktic foraminiferal δ¹⁸O and age constraints, is also shown alongside our multi-proxy temperature estimates in Extended Data Fig. 6.

Multi-species and multi-proxy approach

As no single species of benthic foraminifera is present across all depths between 1.5 km and 5 km, we used a multi-species approach, measuring trace-metal ratios in multiple different taxa. We also took a multi-proxy approach, reconstructing deep-ocean temperatures using independent techniques: Δ₄₇ and trace-metal ratios, thus providing additional support for the individual proxy reconstructions.

Trace-metal analyses

Monospecific benthic foraminiferal samples consisting of between 5 and 15 individuals (where possible) were picked from the >250-μm sediment size fraction (where foraminiferal abundances were low, we also picked from the >212-μm fraction). Foraminifera were then gently crushed between glass slides, after which approximately one-third of the material was removed for isotopic analyses (these data were previously reported¹⁴). The remaining material was then loaded into 500-μl Bio-Rad polypropylene microcentrifuge tubes that had been pre-leached in hot 10% hydrochloric acid and used for trace-metal analysis.

All samples underwent oxidative and reductive cleaning following established methods⁵³, before each individual sample was analysed for a suite of trace and minor elemental ratios on a Thermo Finnigan Element2 magnetic sector inductively coupled plasma mass spectrometer at the University of Colorado, Boulder, as described in ref. ⁵⁴. Long-term ±1σ precision is 0.5% for Mg/Ca, 0.9% for Li/Ca and 4.2% for B/Ca. Mn/Ca, Al/Ca and Fe/Ca ratios were also measured to screen for potential contamination from detrital material and/or secondary phases. For low-Mg species (for example, Melonis pompilioides), samples with contaminant ratios >0.1 mmol mol⁻¹ were rejected. For high-Mg species, for example, G. affinis, this threshold was scaled accordingly⁵⁵. However, in cases where contamination indicators were only marginally above threshold and/or Mn/Ca values were elevated, but Mg/Ca was in good agreement with multiple coeval samples with low contaminant ratios, these data were retained (individual measurements are provided in the Source Data).

We also omitted all trace-metal data from the shallow infaunal benthic foraminifera Uvigerina peregrina, despite suggestions that it is less affected by ΔCO₃²⁻ (ref. ⁵⁶). This is because the global calibration poorly constrains the relationship between Mg/Ca and temperature (R² = 0.68)⁵⁷, particularly at the cold end of the calibration (<5 °C), which is most relevant to the deep ocean. The temperature sensitivity (approximately 0.07 mmol mol⁻¹ °C⁻¹) is also substantially lower than that of other shallow, low-Mg infaunal species and does not align well with our common calibration dataset (Extended Data Fig. 2f). As a result, U. peregrina Mg/Ca yielded implausibly warm glacial temperature estimates (>5 °C) at our core sites, which are not compatible with our clumped isotope temperature estimates from this species.

Trace-metal temperature calibrations

Deep-ocean temperatures were reconstructed using benthic foraminiferal Mg/Ca and species-specific calibrations taking the form: Mg/Ca = s × T + c, where T is the calcification temperature in °C, and s and c are the calibration slope (temperature sensitivity) and intercept, respectively (Fig. 2a,b). One-sigma uncertainties on individual monospecific temperature estimates were propagated, incorporating analytical error (assumed as a relative percentage error on Mg/Ca (see ‘Trace-metal analyses’)) and uncertainties in both the slope and intercept (Extended Data Fig. 4a,c,d).

Mid-to-late Holocene deep-ocean temperatures were also reconstructed using benthic foraminiferal Mg/Li, which has been shown to be less sensitive to carbonate-ion effects, especially in the aragonitic foraminifera, H. elegans⁵⁸. Mg/Li-based temperature estimates were derived first by dividing Mg/Ca by Li/Ca and using the H. elegans temperature–Mg/Li calibration²¹, which takes the form: Mg/Li = aT² + bT + c, where T is the calcification temperature in °C and a, b and c are the quadratic, linear and intercept coefficients, respectively (Fig. 2c). One-sigma uncertainties on individual H. elegans temperature estimates were propagated using Monte Carlo analysis (10,000 iterations), in which each Mg/Li value was perturbed according to analytical uncertainty, and each of the 3 calibration coefficients was randomly sampled from a normal distribution defined by its respective 1σ uncertainty. The quadratic equation was solved for each perturbed Mg/Li value and coefficient set, yielding 10,000 plausible temperature estimates per sample. The 1σ standard deviation of these simulated temperatures was taken as the uncertainty for that sample (Extended Data Fig. 4b).

Mean mid-to-late Holocene and LGM trace-metal-based temperature estimates for each core were derived by first averaging data from the same species within each core. The uncertainty on this mean was expressed as a total ±2 standard error (2 s.e.), combining the ±1 standard error of the mean (s.e.m.), calculated as the variability among replicate samples (n > 1), with the average 1σ propagated temperature measurement uncertainty (Extended Data Fig. 4e–g). Where n = 1, the 2 s.e. approximates to the 2σ propagated temperature measurement uncertainty (we note that this n = 1 uncertainty estimate does not account for variability among replicate measurements and thus probably underestimates the true uncertainty). When multiple different species were present in a core, we calculated a multi-species mean by averaging the species-specific means. The associated uncertainty was determined by propagating the individual species-specific uncertainties to the multi-species mean, treating temperature estimates derived from different foraminiferal species as independent estimates. We also combined data from cores at similar water depths (for example, 1057 and 15JPC; 1059 and 10JPC; 1060 and 6GGC) by averaging their mean values and propagating uncertainties using the same approach (Fig. 3a,d).

Common Mg/Ca temperature calibration

Previous calibration studies have shown that Melonis spp.⁵⁹, Bulimina spp.⁶⁰, Cibicides lobatulus⁶¹, Nonionella labradorica⁶² and C. neoteretis⁶³ show similar Mg/Ca temperature sensitivities (approximately 0.1 mmol mol⁻¹ °C⁻¹; Extended Data Fig. 2), suggesting a shared underlying mechanism governing Mg incorporation during calcification. Therefore, rather than selecting species-specific calibrations for different low-Mg, shallow infaunal species, we developed a common calibration for these five species using published core-top calibration data (Fig. 2a). To do this, we first corrected any non-reductively cleaned data⁶⁴, then converted each dataset to Mg/Ca and temperature anomalies (ΔMg/Ca and ΔT) to place them on a common scale. We then performed simple linear regression on the combined dataset, revealing a temperature sensitivity of 0.1 mmol mol⁻¹ °C⁻¹ (R² = 0.88). Finally, we calculated the average intercept for each species, assuming a common temperature sensitivity of 0.1 mmol mol⁻¹ °C⁻¹. Species-specific intercepts are shown in Extended Data Table 2.

Clumped isotope analyses (Δ₄₇)

To provide independent support for our trace-metal-based temperature reconstructions, we generated additional temperature estimates by measuring benthic foraminiferal clumped isotopes (Δ₄₇). These analyses were possible because of the high abundances of G. affinis, U. peregrina, C. neoteretis, Cibicidoides pachyderma and H. elegans in several of our cores. As this technique also yields foraminiferal δ¹⁸O (δ¹⁸O_c) data, we were also able to independently estimate paired δ¹⁸O_sw with these same samples. The benthic foraminifera were clean and showed no sign of post-depositional alteration from authigenic carbonate precipitation and/or diagenetic overgrowths or contamination from nannofossils and organic material (scanning electron microscope images of representative benthic foraminifera are provided as Supplementary Information); therefore, they were not cleaned or crushed before analysis.

A total of 575 stable and clumped isotope analyses were performed across 30 runs at Utrecht University. Samples weighing approximately 95 µg and 135 µg were prepared for analysis using a Thermo Scientific Kiel IV carbonate preparation device coupled to a Thermo Scientific MAT 253 mass spectrometer (conventional dual inlet method), and a Thermo Scientific 253 Plus mass spectrometer using the long-integration dual-inlet method, respectively. Each of the 30 analytical runs consisted of 46 samples, including 24 carbonate standards—ETH-1, ETH-2 and ETH-3—in a 1:1:5 ratio. These ETH standards differ in their δ¹³C, δ¹⁸O and Δ₄₇ composition and were used to calibrate the sample measurements, correct for δ¹³C and δ¹⁸O drift, and calculate empirical transfer functions. Δ₄₇ values were reported on the Inter-Carbon Dioxide Equilibrium Scale (I-CDES90)⁶⁵. Additional carbonate standards, MERCK and IAEA-C2, were measured alongside the samples to monitor the long-term reproducibility of the Kiel–253 Plus system, which showed a post-correction reproducibility of 0.033‰. To account for potential temporal drift, each run included both mid-to-late Holocene and LGM samples, alternating between the two where possible. This approach minimized drift-related bias and focused on reconstructing ΔT between the two time periods, thereby reducing reliance on the accuracy of any single Δ₄₇–temperature calibration.

In the Kiel IV, each sample was dissolved in phosphoric acid (104% H₃PO₄) at 70 °C, converting carbonate to carbon dioxide (CO₂). The evolved gas was cooled to −196 °C using two liquid-nitrogen traps to concentrate the CO₂ and remove excess water. It was then further purified at −40 °C using a PoraPakQ trap to eliminate possible organic contaminants. Negative pressure base lines were corrected for as described in refs. ^66,67. Pressure base lines were recorded at various m/z 44 intensities (0 V, 5 V, 10 V, 15 V, 20 V, 25 V) before each analytical run. The relationship between background signal and intensity was used to correct for nonlinearities in isotopologues measurements. Samples with extreme initial intensities (<11,000 V or >20,000 V) or high Δ₄₇ standard deviations were excluded. Standardized Δ₄₇ values were calculated using an empirical transfer function based on the offset between measured and accepted ETH values. To correct for subtle nonlinearity, an initial offset correction was applied using ETH-3 standards within a ±1,000 V range. For the MAT 253 Plus, most runs used ETH-3 data from two preceding and following runs to ensure at least ten standards in the target range.

δ¹⁸O_c values were normalized to Vienna PeeDee Belemnite using 15 ETH-3 standards preceding and following each sample. Species-specific offsets from isotopic equilibrium, were corrected for using published species-specific offsets¹⁴, and we also excluded any outliers based on corrected δ¹⁸O_c. We omitted data from samples with δ¹⁸O_c < 4.00‰ and δ¹⁸O_c > 5.00‰ and δ¹⁸O_c < 2.15‰ and δ¹⁸O_c > 3.15‰ for the LGM and mid-to-late Holocene, respectively. Raw Δ₄₇ values for each core and time period were averaged and converted to temperature using the calibration equation of ref. ⁶⁸: Δ₄₇ = (0.0397 ± 0.0011) × 10⁶/T² + 0.1518 ± 0.0128, where T is temperature in °C. Associated errors are reported as ±2 s.e., with standard error calculated as σ/√n, where σ is the standard deviation of Δ₄₇ values and n is the number of measurements for each core for each time period. For sites represented by two cores from similar water depths (for example, 1057/15JPC and 1059/10JPC), Δ₄₇ values from both cores were averaged before temperature conversion. This calibration of ref. ⁶⁸ was selected because (1) it yields mid-to-late Holocene temperature estimates that provide the best statistical match with modern in situ deep-ocean observations, whereas alternative calibrations yield poorer statistical agreement and/or implausible temperature estimates (Extended Data Fig. 3b,c), and (2) it is based exclusively on foraminifera and was produced in clumped isotope laboratories that use analytical procedures identical to those applied in this study

δ¹⁸O_sw

Mid-to-late Holocene and glacial δ¹⁸O_sw were calculated for each core using multi-proxy mean-temperature data (Fig. 2a,d), paired with benthic foraminiferal δ¹⁸O_c and the following linear equation: (δ_c – δ_sw + 0.27) = −0.224 ± 0.002 × T + 3.53 ± 0.02, where T is the calcification temperature in °C, δ_sw is δ¹⁸O_sw on the Standard Mean Ocean Water scale, and δ_c is δ¹⁸O_c on the PeeDee Belemnite scale⁴⁵. Multi-species mean Mg/Ca- and Mg/Li-derived temperatures were combined with previously published paired δ¹⁸O_c data¹⁴, whereas δ¹⁸O_sw derived from Δ₄₇ temperatures used δ¹⁸O_c measurements from clumped isotope analyses. δ¹⁸O_c from benthic foraminiferal species known to calcify in disequilibrium with seawater (for example, G. affinis) was corrected using empirically derived, species-specific offsets¹⁴. For each core, the ±2 s.e. uncertainty on each δ¹⁸O_sw estimate was calculated by propagating and combining uncertainties in quadrature, which included contributions from the ±2 s.e. uncertainty on the mean δ¹⁸O_c, the mean multi-proxy temperature estimate, and the slope and intercept of the δ¹⁸O_sw–temperature equation. When combining data from cores at similar water depths (for example, 1057 and 15JPC), we first averaged the species-specific δ¹⁸O_c estimates and then calculated the ±2 s.e. using the same error propagation approach as used for our temperature estimates.

δ¹⁸O_sw-ivc

To facilitate comparison with modern and mid-to-late Holocene δ¹⁸O_sw, we calculated ice-volume-corrected δ¹⁸O_sw (δ¹⁸O_sw-ivc) for the LGM by subtracting 1.0‰ to account for the global-ice-volume effect. This correction is based on a global mean change in δ¹⁸O_sw of 1.0 ± 0.1‰ (ref. ⁶⁹), derived from sedimentary pore-water measurements, including non-Atlantic sites, which are considered more appropriate for estimating global mean glacial δ¹⁸O_sw (ref. ¹⁶). Notably, this value also agrees well with other independent estimates (0.94 ± 0.18‰ (ref. ⁷⁰) and 1.05 ± 0.2‰ (ref. ⁷¹), based on simple numerical models of ice-sheet growth and benthic δ¹⁸O change at polar sites, respectively). To further evaluate this correction, we also estimated the global mean δ¹⁸O_sw change using the LR04 benthic foraminiferal stack⁷² and the most recent estimate of glacial MOT³⁹. Assuming a MOT change of 2.3 ± 0.5 °C and temperature sensitivity of 0.224‰ °C⁻¹ (ref. ⁴⁵), a glacial-to-modern benthic foraminiferal δ¹⁸O shift of 1.65‰, yields a global mean δ¹⁸O_sw shift of 1.13 ± 0.11‰ (1σ). This estimate is consistent, within uncertainty, with published values, supporting our use of the canonical 1.0‰ correction to derive δ¹⁸O_sw-ivc for the LGM.

Temperature–salinity plots

Multi-proxy temperature and δ¹⁸O_sw data were averaged to produce individual estimates for each core during the mid-to-late Holocene and LGM, with associated uncertainties expressed as ±1 s.e. after combining the error associated with each proxy (for example, Mg/Ca, Mg/Li, Δ₄₇). For the mid-to-late Holocene, where Δ₄₇ estimates were not available for all combined core pairs (for example, 1057/15JPC), multi-proxy data for both cores were averaged, and where multiple estimates from the same proxy were present (for example, Mg/Ca), their uncertainties were combined as described previously.

For the mid-to-late Holocene, δ¹⁸O_sw and its associated uncertainty were then converted to salinity using a single empirically derived δ¹⁸O_sw–salinity relationship²⁵, taking the form S = (δ¹⁸O_sw + c)/s, where S is salinity in PSU, and s and c are the calibration slope and intercept, respectively. For the LGM, δ¹⁸O_sw was first corrected for the global-ice-volume effect, then converted to salinity using the same δ¹⁸O_sw–salinity relationship, before a global +1.1 PSU offset was applied to account for the higher mean salinity of the glacial ocean, enabling direct comparison with modern salinity data. Whereas previous work has shown that different δ¹⁸O_sw–salinity relationships apply at different depths in the Northwest Atlantic¹⁴, probably reflecting distinct deep-water masses sourced from different regions across the subpolar North Atlantic, for simplicity, we apply the NADW-specific relationship (s = 0.51, c = 17.75), as it is the most appropriate for the majority of our core sites, that is, using this relationship yields the best agreement between our reconstructions and modern observations (Fig. 3c). Therefore, we are cautious not to over-interpret the derived salinity and density values, and instead focus primarily on the underlying δ¹⁸O_sw signal. Furthermore, owing to the large and regionally variable uncertainties in converting δ¹⁸O_sw to salinity⁷³, we do not propagate uncertainty using the errors associated with the slope and intercept of the δ¹⁸O_sw–salinity relationship; instead, we convert δ¹⁸O_sw uncertainties directly into salinity using a fixed slope and intercept.

Extended Data Figure 5 also includes published Atlantic glacial temperature and δ¹⁸O_sw estimates^{20,74,75,76,77,78,79,80}. However, given the uncertainty surrounding the glacial δ¹⁸O_sw–salinity relationships associated with the water masses bathing these sites, we do not convert these δ¹⁸O_sw data to salinity. The salinity axis and isopycnals are shown to enable visualization for the Northwest Atlantic transect data originally shown in Fig. 2c,f.

ΔT and Δδ¹⁸O_sw

Because we do not have a complete set of both mid-to-late Holocene and glacial temperature estimates for all our cores—in general, there are more mid-to-late Holocene data from shallower sites and glacial data from deeper sites—simply averaging all data from each time period would introduce an artificial cold bias. To avoid this, we calculated the mean LGM–modern deep-ocean temperature difference (ΔT) based on the individual ΔT values for each core site. This was done by first determining the ΔT for each core from between 1.5 km and 4 km depth (excluding subtropically influenced site ODP-172-1055), using the closest World Ocean Atlas 2023 (WOA23) data point to each site (it is noted that each WOA23 data point represents the annual average based on all valid data from 1955 to 2022)⁷. These per-core ΔT values were then averaged to derive regional ΔT estimates from Mg/Ca-derived temperature data for the Northwest Atlantic and Northeast Atlantic. The associated uncertainty (±2 s.e.) was determined by combining the uncertainty associated with each core’s ΔT in quadrature. For Δ₄₇, ΔT was calculated as a weighted average, with the number of individual measurements per core used as weights. We note that the mean ΔT ± 2 s.e. (0.94 °C) is almost identical to the mean ±2 s.e. associated with the mean of all Δ₄₇ glacial temperature estimates (0.96 °C, n = 284). We followed the same procedures to derive LGM–modern Δδ¹⁸O_sw; however, in the absence of modern in situ δ¹⁸O_sw measurements, δ¹⁸O_sw for each core was estimated using salinity data from WOA23 (ref. ⁸) and modern salinity–δ¹⁸O_sw relationships for NADW, North Atlantic (NATL) and LSW²⁵. These three estimates were then averaged to produce a modern δ¹⁸O_sw value for each core. To check for a potential proxy–climatology bias, we also calculated ΔT and Δδ¹⁸O_sw using paired LGM and mid-to-late Holocene data where available for both trace-metal-derived (n = 5) and Δ₄₇-derived (n = 3) reconstructions. To do this, we followed the same procedure outlined above; however, for mid-to-late Holocene sites with both Mg/Ca- and Mg/Li-based temperatures, we used their combined mean values. For 6GGC, where only Holocene data are available, we compared these with LGM estimates from the nearby site 1060, which is also located at approximately 3.5 km water depth.

Carbonate-ion effects

Previous work suggests that tthe bottom-water carbonate-ion (ΔCO₃²⁻) concentration affects the partitioning of Mg during calcification in epifaunal benthic foraminifera, with lower Mg/Ca where the carbonate-ion saturation state (ΔCO₃²⁻, defined as ΔCO₃²⁻ = CO₃²⁻_in-situ − CO₃²⁻_saturation) is low or undersaturated (<0 μmol kg⁻¹)^56,81. Although infaunal benthic foraminiferal Mg/Ca are generally considered to be less susceptible to undersaturation, pore-water ΔCO₃²⁻ is spatially and temporally variable, such that infaunal species may also be affected by ΔCO₃²⁻ (refs. ^20,82). To assess the potential impact of low ΔCO₃²⁻ on our infaunal benthic foraminiferal Mg/Ca ratios, we also reconstructed bottom-water ΔCO₃²⁻ using B/Ca ratios measured on Cibicidoides wuellerstorfi (where present) applying the calibration⁸¹: B/Ca = 1.14 ± 0.048 × ΔCO₃²⁻ + 177.1 ± 1.41, where ΔCO₃²⁻ is the carbonate-ion saturation state of seawater in μmol kg⁻¹. One-sigma uncertainties were propagated, incorporating analytical error (see ‘Trace-metal analyses’) and uncertainties in both the slope and intercept (Extended Data Fig. 7). Modern Northwest Atlantic ΔCO₃²⁻ was calculated using temperature, salinity, total alkalinity and total dissolved inorganic carbon data from nearby Global Ocean Data Analysis Project (GLODAP, v2022) stations⁴⁷ with PyCO2sys⁸³.

Overall, reconstructed ΔCO₃²⁻ is generally >0 μmol kg⁻¹, suggesting that the glacial deep Northwest Atlantic was oversaturated with respect to ΔCO₃² (Extended Data Fig. 7b). Although this implies that any ΔCO₃²⁻-effect-related suppression of Mg incorporation in our infaunal benthic foraminifera was probably minimal, ΔCO₃²⁻ at abyssal site KNR-197-10-17GGC is much lower (−37 ± 4 μmol kg⁻¹, n = 2), and it is also possible that pore waters may be more undersaturated than the overlying bottom waters²⁰. However, given that undersaturated conditions supress the Mg incorporation during calcification, any ΔCO₃²⁻ effect would bias Mg/Ca-derived temperatures towards colder values. Therefore, our reconstructed glacial deep-ocean temperatures may represent a conservative (cold) estimate and attempting to correct for any ΔCO₃²⁻ effect at this abyssal site would produce slightly warmer glacial temperatures.

Isotope-enabled models

To provide additional context for our proxy reconstructions, we compared our δ¹⁸O_sw-ivc estimates with available isotope-enabled simulations of the glacial Atlantic^{27,84,85,86,87,88}. These models generally simulate a slightly shallower glacial Atlantic Meridional Overturning Circulation, with NADW temperatures ranging from approximately 1 °C to −2 °C (refs. ^27,88) and δ¹⁸O_sw-ivc between −0.2‰ and 0.5‰ (refs. ^{27,84,85,86,87,88}). Although this is broadly consistent with the traditional view—based on palaeoceanographic nutrient proxies³—that the Atlantic Meridional Overturning Circulation shoaled during the LGM, and with pore-water-based estimates suggesting that glacial NADW was much colder⁶, these simulations are not consistent with more recent work⁵ and our temperature and δ¹⁸O_sw constraints from the North Atlantic. However, the iPOP2 model does reproduce comparably high δ¹⁸O_sw values, although these are restricted to the upper approximately 2 km of the North Atlantic²⁷.

Published data

Glacial deep-ocean temperature and δ¹⁸O_sw data and associated uncertainties were calculated following the same procedures used for our Northwest Atlantic data (Extended Data Table 3 and Source Data). Where appropriate, we used our common calibration to derive temperature, applying a 10% correction to any non-reductively cleaned trace-metal data (for example, Mg/Ca_reduc. = (Mg/Ca)/1.1) and appropriate δ¹⁸O_c offsets to benthic foraminiferal δ¹⁸O_c from species that calcify in disequilibrium with the surrounding seawater¹⁴. Consistent with our previous approach, U. peregrina-based temperature and δ¹⁸O_sw estimates were omitted from this compilation owing to concerns with the Mg/Ca temperature calibration of U. peregrina.

Glacial surface-ocean temperature and δ¹⁸O_sw estimates and associated uncertainties were also calculated from published Mg/Ca and δ¹⁸O_c data using species-specific Mg/Ca temperature calibrations (Extended Data Table 3), appropriate vital-effect corrections^89,90,91, and the updated δ¹⁸O–temperature relationship based on inorganic calcite precipitation experiments⁹¹ (Source Data). Although recent Mg/Ca temperature calibrations for planktic foraminifera include non-thermal influences such as whole-ocean salinity and pH, they do not account for spatial variations in local salinity⁹². For species with a known salinity effect on Mg/Ca (that is, Globigerinoides ruber and Globigerina bulloides), we iteratively solved for a salinity value that is self-consistent with the North Atlantic δ¹⁸O_sw–salinity relationship (s = 0.55, c = 18.98)²⁵. For both G. ruber and G. bulloides, we assume a glacial pH of 8.2 ± 0.2 (refs. ^92,93), recognizing that this conservative uncertainty is the primary contributor to the relativley large errors on the final temperature and δ¹⁸O_sw estimates (Source Data).

Data availability

The proxy data that support these findings are publicly available through the data repository Pangaea at https://doi.org/10.1594/PANGAEA.988210^94,95. GLODAP Bottle Data (version2.2022)⁴⁷ and World Ocean Atlas data^7,8 were downloaded from https://glodap.info/index.php/merged-and-adjusted-data-product-v2-2022/ and https://www.ncei.noaa.gov/products/world-ocean-atlas, respectively. Figure 1 was generated using Ocean Data View software⁴⁴. Source data are provided with this paper.

Code availability

Python code used to perform statistical analyses as part of this study are freely available on Zenodo at https://doi.org/10.5281/zenodo.17733604 (ref. ⁹⁶).

Change history

27 January 2026
A Correction to this paper has been published: https://doi.org/10.1038/s41586-026-10186-3

References

Clark, P. U. et al. The Last Glacial Maximum. Science 325, 710–714 (2009).
Article ADS CAS PubMed Google Scholar
Kageyama, M. et al. The PMIP4 Last Glacial Maximum experiments: preliminary results and comparison with the PMIP3 simulations. Clim. Past 17, 1065–1089 (2021).
Article Google Scholar
Curry, W. B. & Oppo, D. W. Glacial water mass geometry and the distribution of δ¹³C of ΣCO₂ in the western Atlantic Ocean. Paleoceanography 20, PA1017 (2005).
Oppo, D. W. et al. Data constraints on glacial atlantic water mass geometry and properties. Paleoceanogr. Paleoclimatol. 33, 1013–1034 (2018).
Article ADS Google Scholar
Blaser, P. et al. Prevalent North Atlantic Deep Water during the Last Glacial Maximum and Heinrich Stadial 1. Nat. Geosci. 18, 410–416 (2025).
Article ADS CAS PubMed PubMed Central Google Scholar
Adkins, J. F., McIntyre, K. & Schrag, D. P. The salinity, temperature, and δ¹⁸O of the glacial deep ocean. Science 298, 1769–1773 (2002).
Article ADS CAS PubMed Google Scholar
Locarnini, R. A. et al. World Ocean Atlas 2023. Volume 1: Temperature (NOAA, 2024).
Reagan, J. R. et al. World Ocean Atlas 2023. Volume 2: Salinity (NOAA, 2024).
Talley, L. D., Pickard, G. L., Emery, W. J. & Swift, J. H. in Descriptive Physical Oceanography 6th edn (eds Talley, L. D. et al.) 245–301 (Academic Press, 2011).
Sherriff-Tadano, S., Abe-Ouchi, A., Yoshimori, M., Oka, A. & Chan, W.-L. Influence of glacial ice sheets on the Atlantic meridional overturning circulation through surface wind change. Clim. Dyn. 50, 2881–2903 (2018).
Article Google Scholar
Klockmann, M., Mikolajewicz, U. & Marotzke, J. The effect of greenhouse gas concentrations and ice sheets on the glacial AMOC in a coupled climate model. Clim. Past 12, 1829–1846 (2016).
Article Google Scholar
Oka, A., Hasumi, H. & Abe-Ouchi, A. The thermal threshold of the Atlantic meridional overturning circulation and its control by wind stress forcing during glacial climate. Geophys. Res. Lett. 39, L09709 (2012).
Pöppelmeier, F., Gutjahr, M., Blaser, P., Keigwin, L. D. & Lippold, J. Origin of abyssal NW Atlantic water masses since the Last Glacial Maximum. Paleoceanogr. Paleoclimatology 33, 530–543 (2018).
Article ADS Google Scholar
Wharton, J. H. et al. Deeper and stronger North Atlantic Gyre during the Last Glacial Maximum. Nature https://doi.org/10.1038/s41586-024-07655-y (2024).
Keigwin, L. D. & Swift, S. A. Carbon isotope evidence for a northern source of deep water in the glacial western North Atlantic. Proc. Natl Acad. Sci. USA 114, 2831–2835 (2017).
Article ADS CAS PubMed PubMed Central Google Scholar
Wunsch, C. Last Glacial Maximum and deglacial abyssal seawater oxygen isotopic ratios. Clim. Past 12, 1281–1296 (2016).
Article Google Scholar
Miller, M. D., Simons, M., Adkins, J. F. & Minson, S. E. The information content of pore fluid δ¹⁸O and [Cl⁻]. J. Phys. Oceanogr. 45, 2070–2094 (2015).
Article ADS Google Scholar
Petit, T., Lozier, M. S., Josey, S. A. & Cunningham, S. A. Atlantic deep water formation occurs primarily in the Iceland Basin and Irminger Sea by local buoyancy forcing. Geophys. Res. Lett. 47, e2020GL091028 (2020).
Article ADS Google Scholar
Martin, W. R. & Sayles, F. L. CaCO₃ dissolution in sediments of the Ceara Rise, western equatorial Atlantic. Geochim. Cosmochim. Acta 60, 243–263 (1996).
Article ADS CAS Google Scholar
Weldeab, S., Arce, A. & Kasten, S. Mg/Ca-ΔCO₃²⁻_porewater–temperature calibration for Globobulimina spp.: a sensitive paleothermometer for deep-sea temperature reconstruction. Earth Planet. Sci. Lett. 438, 95–102 (2016).
Article ADS CAS Google Scholar
Marchitto, T. M., Bryan, S. P., Doss, W., McCulloch, M. T. & Montagna, P. A simple biomineralization model to explain Li, Mg, and Sr incorporation into aragonitic foraminifera and corals. Earth Planet. Sci. Lett. 481, 20–29 (2018).
Article ADS CAS Google Scholar
Cronin, T. M., Dwyer, G. S., Baker, P. A., Rodriguez-Lazaro, J. & DeMartino, D. M. Orbital and suborbital variability in North Atlantic bottom water temperature obtained from deep-sea ostracod Mg/Ca ratios. Palaeogeogr. Palaeoclimatol. Palaeoecol. 162, 45–57 (2000).
Article Google Scholar
Dwyer, G. S., Cronin, T. M., Baker, P. A. & Rodriguez-Lazaro, J. Changes in North Atlantic deep-sea temperature during climatic fluctuations of the last 25,000 years based on ostracode Mg/Ca ratios. Geochem. Geophys. Geosyst. 1, 1028 (2000).
Yasuhara, M. et al. North Atlantic Intermediate Water variability over the past 20,000 years. Geology 47, 659–663 (2019).
Article ADS CAS Google Scholar
LeGrande, A. N. & Schmidt, G. A. Global gridded data set of the oxygen isotopic composition in seawater. Geophys. Res. Lett. 33, L12604 (2006).
Rohling, E. J. Progress in paleosalinity: overview and presentation of a new approach. Paleoceanography 22, PA3215 (2007).
Zhang, J. et al. Asynchronous warming and δ¹⁸O evolution of deep Atlantic water masses during the last deglaciation. Proc. Natl Acad. Sci. USA 114, 11075–11080 (2017).
Article ADS CAS PubMed PubMed Central Google Scholar
Hirschi, J. J.-M. et al. The Atlantic Meridional Overturning Circulation in high-resolution models. J. Geophys. Res. Oceans 125, e2019JC015522 (2020).
Article ADS Google Scholar
Waelbroeck, C. et al. Constraints on surface seawater oxygen isotope change between the Last Glacial Maximum and the Late Holocene. Quat. Sci. Rev. 105, 102–111 (2014).
Article ADS Google Scholar
Benetti, M., Reverdin, G., Aloisi, G. & Sveinbjörnsdóttir, Á. Stable isotopes in surface waters of the Atlantic Ocean: indicators of ocean–atmosphere water fluxes and oceanic mixing processes. J. Geophys. Res. Oceans 122, 4723–4742 (2017).
Article ADS Google Scholar
Warken, S. F. et al. Caribbean hydroclimate and vegetation history across the last glacial period. Quat. Sci. Rev. 218, 75–90 (2019).
Article ADS Google Scholar
Bartlein, P. J. et al. Pollen-based continental climate reconstructions at 6 and 21 ka: a global synthesis. Clim. Dyn. 37, 775–802 (2011).
Article Google Scholar
Cronin, T. M. et al. Deep Arctic Ocean warming during the last glacial cycle. Nat. Geosci. 5, 631–634 (2012).
Article ADS CAS Google Scholar
Bauch, H. A. et al. A multiproxy reconstruction of the evolution of deep and surface waters in the subarctic Nordic Seas over the last 30,000 yr. Quat. Sci. Rev. 20, 659–678 (2001).
Article ADS Google Scholar
Thornalley, D. J. R., Elderfield, H. & McCave, I. N. Reconstructing North Atlantic deglacial surface hydrography and its link to the Atlantic overturning circulation. Glob. Planet. Change 79, 163–175 (2011).
Article ADS Google Scholar
Alley, R. B. & Clark, P. U. The deglaciation of the Northern Hemisphere: a global perspective. Annu. Rev. Earth Planet. Sci. 27, 149–182 (1999).
Article ADS CAS Google Scholar
Winton, M. The effect of cold climate upon North Atlantic Deep Water formation in a simple ocean–atmosphere model. J. Clim. 10, 37–51 (1997).
Article ADS Google Scholar
Hanawa, K. & D. Talley, L. in International Geophysics Vol. 77 (eds Siedler, G. et al.) 373–386 (Academic Press, 2001).
Seltzer, A. M., Davidson, P. W., Shackleton, S. A., Nicholson, D. P. & Khatiwala, S. Global ocean cooling of 2.3 °C during the Last Glacial Maximum. Geophys. Res. Lett. 51, e2024GL108866 (2024).
Article ADS Google Scholar
Bereiter, B., Shackleton, S., Baggenstos, D., Kawamura, K. & Severinghaus, J. Mean global ocean temperatures during the last glacial transition. Nature 553, 39–44 (2018).
Article ADS CAS PubMed Google Scholar
Barker, S. & Knorr, G. A systematic role for extreme ocean–atmosphere oscillations in the development of glacial conditions since the Mid Pleistocene transition. Paleoceanogr. Paleoclimatology 38, e2023PA004690 (2023).
Article Google Scholar
Våge, K., Papritz, L., Håvik, L., Spall, M. A. & Moore, G. W. K. Ocean convection linked to the recent ice edge retreat along east Greenland. Nat. Commun. 9, 1287 (2018).
Article ADS PubMed PubMed Central Google Scholar
Lique, C. & Thomas, M. D. Latitudinal shift of the Atlantic Meridional Overturning Circulation source regions under a warming climate. Nat. Clim. Change 8, 1013–1020 (2018).
Article ADS Google Scholar
Schlitzer, R. Ocean Data View: EPIC3 (ODV, 2022).
Marchitto, T. M. et al. Improved oxygen isotope temperature calibrations for cosmopolitan benthic foraminifera. Geochim. Cosmochim. Acta 130, 1–11 (2014).
Article ADS CAS Google Scholar
Henry, L. G. et al. North Atlantic ocean circulation and abrupt climate change during the last glaciation. Science 353, 470–474 (2016).
Article ADS CAS PubMed Google Scholar
Lauvset, S. K. et al. GLODAPv2.2022: the latest version of the global interior ocean biogeochemical data product. Earth Syst. Sci. Data 14, 5543–5572 (2022).
Article ADS Google Scholar
McDougall, T. J. & Barker, P. M. Getting started with TEOS-10 and the Gibbs seawater (GSW) oceanographic toolbox. Scoriapso WG 127, 1–28 (2011).
Google Scholar
Dalton, A. S. et al. An updated radiocarbon-based ice margin chronology for the last deglaciation of the North American Ice Sheet Complex. Quat. Sci. Rev. 234, 106223 (2020).
Article Google Scholar
Hughes, A. L. C., Gyllencreutz, R., Lohne, ØS., Mangerud, J. & Svendsen, J. I. The last Eurasian ice sheets—a chronological database and time-slice reconstruction, DATED-1. Boreas 45, 1–45 (2016).
Article Google Scholar
Thornalley, D. J. R., Elderfield, H. & McCave, I. N. Intermediate and deep water paleoceanography of the northern North Atlantic over the past 21,000 years. Paleoceanography 25, PA1211 (2010).
Barker, S., Kiefer, T. & Elderfield, H. Temporal changes in North Atlantic circulation constrained by planktonic foraminiferal shell weights. Paleoceanography 19, PA3008 (2004).
Boyle, E. & Rosenthal, Y. in The South Atlantic: Present and Past Circulation (eds Wefer, G. et al.) 423–443 (Springer, 1996).
Marchitto, T. M. Precise multielemental ratios in small foraminiferal samples determined by sector field ICP-MS. Geochem. Geophys. Geosyst. 7, Q05P13 (2006).
Barker, S., Greaves, M. & Elderfield, H. A study of cleaning procedures used for foraminiferal Mg/Ca paleothermometry. Geochem. Geophys. Geosyst. 4, 8407 (2003).
Elderfield, H., Yu, J., Anand, P., Kiefer, T. & Nyland, B. Calibrations for benthic foraminiferal Mg/Ca paleothermometry and the carbonate ion hypothesis. Earth Planet. Sci. Lett. 250, 633–649 (2006).
Article ADS CAS Google Scholar
Stirpe, C. R. et al. The Mg/Ca proxy for temperature: a Uvigerina core-top study in the Southwest Pacific. Geochim. Cosmochim. Acta 309, 299–312 (2021).
Article ADS CAS Google Scholar
Bryan, S. P. & Marchitto, T. M. Mg/Ca–temperature proxy in benthic foraminifera: new calibrations from the Florida Straits and a hypothesis regarding Mg/Li. Paleoceanography 23, PA2220 (2008).
Hasenfratz, A. P. et al. Mg/Ca-temperature calibration for the benthic foraminifera Melonis barleeanum and Melonis pompilioides. Geochim. Cosmochim. Acta 217, 365–383 (2017).
Article ADS CAS Google Scholar
Grunert, P. et al. Mg/Ca-temperature calibration for costate Bulimina species (B. costata, B. inflata, B. mexicana): a paleothermometer for hypoxic environments. Geochim. Cosmochim. Acta 220, 36–54 (2018).
Article ADS CAS Google Scholar
Quillmann, U., Marchitto, T. M., Jennings, A. E., Andrews, J. T. & Friestad, B. F. Cooling and freshening at 8.2 ka on the NW Iceland Shelf recorded in paired δ¹⁸O and Mg/Ca measurements of the benthic foraminifer Cibicides lobatulus. Quat. Res. 78, 528–539 (2012).
Article CAS Google Scholar
Skirbekk, K. et al. Benthic foraminiferal growth seasons implied from Mg/Ca–temperature correlations for three Arctic species. Geochem. Geophys. Geosyst. 17, 4684–4704 (2016).
Article ADS CAS Google Scholar
Kristjánsdóttir, G. B., Lea, D. W., Jennings, A. E., Pak, D. K. & Belanger, C. New spatial Mg/Ca-temperature calibrations for three Arctic, benthic foraminifera and reconstruction of north Iceland shelf temperature for the past 4000 years. Geochem. Geophys. Geosyst. 8, Q03P21 (2007).
Yu, J., Elderfield, H., Greaves, M. & Day, J. Preferential dissolution of benthic foraminiferal calcite during laboratory reductive cleaning. Geochem. Geophys. Geosyst. 8, Q06016 (2007).
Bernasconi, S. M. et al. InterCarb: a community effort to improve interlaboratory standardization of the carbonate clumped isotope thermometer using carbonate standards. Geochem. Geophys. Geosyst. 22, e2020GC009588 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Bernasconi, S. M. et al. Background effects on Faraday collectors in gas-source mass spectrometry and implications for clumped isotope measurements. Rapid Commun. Mass Spectrom. 27, 603–612 (2013).
Article ADS CAS PubMed Google Scholar
Meckler, A. N., Ziegler, M., Millán, M. I., Breitenbach, S. F. M. & Bernasconi, S. M. Long-term performance of the Kiel carbonate device with a new correction scheme for clumped isotope measurements. Rapid Commun. Mass Spectrom. 28, 1705–1715 (2014).
Article ADS CAS PubMed Google Scholar
Meinicke, N., Reimi, M. A., Ravelo, A. C. & Meckler, A. N. Coupled Mg/Ca and clumped isotope measurements indicate lack of substantial mixed layer cooling in the Western Pacific Warm Pool during the last ∼5 million years. Paleoceanogr. Paleoclimatol. 36, e2020PA004115 (2021).
Article ADS Google Scholar
Schrag, D. P. et al. The oxygen isotopic composition of seawater during the Last Glacial Maximum. Quat. Sci. Rev. 21, 331–342 (2002).
Article ADS Google Scholar
Clark, P. U. et al. Global mean sea level over the past 4.5 million years. Science 390, eadv8389 (2025).
Article CAS PubMed Google Scholar
Duplessy, J.-C., Labeyrie, L. & Waelbroeck, C. Constraints on the ocean oxygen isotopic enrichment between the Last Glacial Maximum and the Holocene: paleoceanographic implications. Quat. Sci. Rev. 21, 315–330 (2002).
Article ADS Google Scholar
Lisiecki, L. E. & Raymo, M. E. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ¹⁸O records. Paleoceanography 20, PA1003 (2005).
Rohling, E. J. & Bigg, G. R. Paleosalinity and δ¹⁸O: a critical assessment. J. Geophys. Res. Oceans 103, 1307–1318 (1998).
Article ADS CAS Google Scholar
Valley, S. G., Lynch-Stieglitz, J. & Marchitto, T. M. Intermediate water circulation changes in the Florida Straits from a 35 ka record of Mg/Li-derived temperature and Cd/Ca-derived seawater cadmium. Earth Planet. Sci. Lett. 523, 115692 (2019).
Article CAS Google Scholar
Umling, N. E. et al. Atlantic circulation and ice sheet influences on upper South Atlantic temperatures during the last deglaciation. Paleoceanogr. Paleoclimatol. 34, 990–1005 (2019).
Article ADS Google Scholar
Oppo, D. W. et al. Deglacial temperature and carbonate saturation state variability in the tropical Atlantic at Antarctic Intermediate Water depths. Paleoceanogr. Paleoclimatol. 38, e2023PA004674 (2023).
Article Google Scholar
Ezat, M. M., Rasmussen, T. L. & Groeneveld, J. Reconstruction of hydrographic changes in the southern Norwegian Sea during the past 135 kyr and the impact of different foraminiferal Mg/Ca cleaning protocols. Geochem. Geophys. Geosyst. 17, 3420–3436 (2016).
Article ADS Google Scholar
Marcott, S. A. et al. Ice-shelf collapse from subsurface warming as a trigger for Heinrich events. Proc. Natl Acad. Sci. USA 108, 13415–13419 (2011).
Article ADS CAS PubMed PubMed Central Google Scholar
Skinner, L. C., Shackleton, N. J. & Elderfield, H. Millennial-scale variability of deep-water temperature and δ¹⁸O_dw indicating deep-water source variations in the Northeast Atlantic, 0–34 cal. ka BP. Geochem. Geophys. Geosyst. 4, 1098 (2003).
Hasenfratz, A. P. et al. The residence time of Southern Ocean surface waters and the 100,000-year ice age cycle. Science 363, 1080–1084 (2019).
Article ADS CAS PubMed Google Scholar
Yu, J. & Elderfield, H. Benthic foraminiferal B/Ca ratios reflect deep water carbonate saturation state. Earth Planet. Sci. Lett. 258, 73–86 (2007).
Article ADS CAS Google Scholar
Doss, W., Marchitto, T. M., Eagle, R., Rashid, H. & Tripati, A. Deconvolving the saturation state and temperature controls on benthic foraminiferal Li/Ca, based on downcore paired B/Ca measurements and coretop compilation. Geochim. Cosmochim. Acta 236, 297–314 (2018).
Article ADS CAS Google Scholar
Humphreys, M. P. et al. PyCO2SYS: marine carbonate system calculations in Python. Zenodo https://doi.org/10.5281/zenodo.16420947 (2025).
Wadley, M. R., Bigg, G. R., Rohling, E. J. & Payne, A. J. On modelling present-day and last glacial maximum oceanic δ¹⁸O distributions. Glob. Planet. Change 32, 89–109 (2002).
Article ADS Google Scholar
Brennan, C. E., Weaver, A. J., Eby, M. & Meissner, K. J. Modelling oxygen isotopes in the University of Victoria Earth system climate model for pre-industrial and Last Glacial Maximum conditions. Atmos. Ocean 50, 447–465 (2012).
Article ADS CAS Google Scholar
Werner, M. et al. Glacial–interglacial changes in H₂¹⁸O, HDO and deuterium excess—results from the fully coupled ECHAM5/MPI-OM Earth system model. Geosci. Model Dev. 9, 647–670 (2016).
Article ADS CAS Google Scholar
Gu, S. et al. Assessing the potential capability of reconstructing glacial Atlantic water masses and AMOC using multiple proxies in CESM. Earth Planet. Sci. Lett. 541, 116294 (2020).
Article CAS Google Scholar
Caley, T., Roche, D. M., Waelbroeck, C. & Michel, E. Oxygen stable isotopes during the Last Glacial Maximum climate: perspectives from data–model (iLOVECLIM) comparison. Clim. Past 10, 1939–1955 (2014).
Article Google Scholar
Bemis, B. E., Spero, H. J., Bijma, J. & Lea, D. W. Reevaluation of the oxygen isotopic composition of planktonic foraminifera: experimental results and revised paleotemperature equations. Paleoceanography 13, 150–160 (1998).
Article ADS Google Scholar
Nyland, B. F., Jansen, E., Elderfield, H. & Andersson, C. Neogloboquadrina pachyderma (dex. and sin.) Mg/Ca and δ¹⁸O records from the Norwegian Sea. Geochem. Geophys. Geosyst. 7, Q10P17 (2006).
Kim, S.-T. & O’Neil, J. R. Equilibrium and nonequilibrium oxygen isotope effects in synthetic carbonates. Geochim. Cosmochim. Acta 61, 3461–3475 (1997).
Article ADS CAS Google Scholar
Gray, W. R. & Evans, D. Nonthermal influences on Mg/Ca in planktonic foraminifera: a review of culture studies and application to the Last Glacial Maximum. Paleoceanogr. Paleoclimatol. 34, 306–315 (2019).
Article ADS Google Scholar
Hönisch, B. & Hemming, N. G. Surface ocean pH response to variations in pCO₂ through two full glacial cycles. Earth Planet. Sci. Lett. 236, 305–314 (2005).
Article ADS Google Scholar
Felden, J. et al. PANGAEA – data publisher for earth & environmental science. Sci. Data 10, 347 (2023).
Wharton, J. et al. Mid-to-late Holocene and LGM multiproxy temperature, stable isotope, and seawater δ18O data from the North Atlantic [dataset bundled publication]. PANGAEA https://doi.org/10.1594/PANGAEA.988210 (2026).
Wharton, J. H. Relatively warm deep water formation persisted in the Last Glacial Maximum (python script). Zenodo https://doi.org/10.5281/zenodo.17733604 (2026).
The GEBCO_2014 Grid, version 20150318 (GEBCO Compilation Group, 2015).
Mawbey, E. M. et al. Mg/Ca–temperature calibration of polar benthic foraminifera species for reconstruction of bottom water temperatures on the Antarctic shelf. Geochim. Cosmochim. Acta 283, 54–66 (2020).
Article ADS CAS Google Scholar
Barrientos, N. et al. Arctic Ocean benthic foraminifera Mg/Ca ratios and global Mg/Ca-temperature calibrations: new constraints at low temperatures. Geochim. Cosmochim. Acta 236, 240–259 (2018).
Article ADS CAS Google Scholar
Sessford, E. G. et al. High-resolution benthic Mg/Ca temperature record of the intermediate water in the Denmark Strait across D–O Stadial–Interstadial cycles. Paleoceanogr. Paleoclimatol. 33, 1169–1185 (2018).
Article ADS CAS PubMed PubMed Central Google Scholar
Lear, C. H., Rosenthal, Y. & Slowey, N. Benthic foraminiferal Mg/Ca-paleothermometry: a revised core-top calibration. Geochim. Cosmochim. Acta 66, 3375–3387 (2002).
Article ADS CAS Google Scholar
Elderfield, H. et al. A record of bottom water temperature and seawater δ¹⁸O for the Southern Ocean over the past 440 kyr based on Mg/Ca of benthic foraminiferal Uvigerina spp. Quat. Sci. Rev. 29, 160–169 (2010).
Article ADS Google Scholar
Daëron, M. & Gray, W. R. Revisiting oxygen-18 and clumped isotopes in planktic and benthic foraminifera. Paleoceanogr. Paleoclimatol. 38, e2023PA004660 (2023).
Article ADS Google Scholar
Anderson, N. T. et al. A unified clumped isotope thermometer calibration (0.5–1,100 °C) using carbonate-based standardization. Geophys. Res. Lett. 48, e2020GL092069 (2021).
Article ADS CAS Google Scholar
Daëron, M. & Vermeesch, P. Omnivariant generalized least squares regression: theory, geochronological applications, and making the case for reconciled Δ₄₇ calibrations. Chem. Geol. 647, 121881 (2024).
Article Google Scholar
Ezat, M. M. et al. Deep ocean storage of heat and CO₂ in the Fram Strait, Arctic Ocean during the last glacial period. Paleoceanogr. Paleoclimatol. 36, e2021PA004216 (2021).
Article Google Scholar
El bani Altuna, N., Ezat, M. M., Greaves, M. & Rasmussen, T. L. Millennial-scale changes in bottom water temperature and water mass exchange through the Fram Strait 79°N, 63–13 ka. Paleoceanogr. Paleoclimatol. 36, e2020PA004061 (2021).
Article ADS Google Scholar
Thornalley, D. J. R. et al. A warm and poorly ventilated deep Arctic Mediterranean during the last glacial period. Science 349, 706–710 (2015).
Article CAS PubMed Google Scholar
Keigwin, L. D. Radiocarbon and stable isotope constraints on Last Glacial Maximum and Younger Dryas ventilation in the western North Atlantic. Paleoceanography 19, PA4012 (2004).
Pöppelmeier, F. et al. Influence of ocean circulation and benthic exchange on deep Northwest Atlantic Nd isotope records during the past 30,000 years. Geochem. Geophys. Geosyst. 20, 4457–4469 (2019).
Article ADS Google Scholar
Lippold, J. et al. Constraining the variability of the Atlantic Meridional Overturning Circulation during the Holocene. Geophys. Res. Lett. 46, 11338–11346 (2019).
Article ADS Google Scholar
Lippold, J. et al. Does sedimentary ²³¹Pa/²³⁰Th from the Bermuda Rise monitor past Atlantic Meridional Overturning Circulation? Geophys. Res. Lett. 36, L12601 (2009).
Carlson, A. E. et al. Subtropical Atlantic salinity variability and Atlantic meridional circulation during the last deglaciation. Geology 36, 991–994 (2008).
Article ADS CAS Google Scholar
Hall, I. R., Evans, H. K. & Thornalley, D. J. R. Deep water flow speed and surface ocean changes in the subtropical North Atlantic during the last deglaciation. Glob. Planet. Change 79, 255–263 (2011).
Article ADS Google Scholar
Peck, V. L. et al. High resolution evidence for linkages between NW European ice sheet instability and Atlantic Meridional Overturning Circulation. Earth Planet. Sci. Lett. 243, 476–488 (2006).
Article ADS CAS Google Scholar
Benway, H. M., McManus, J. F., Oppo, D. W. & Cullen, J. L. Hydrographic changes in the eastern subpolar North Atlantic during the last deglaciation. Quat. Sci. Rev. 29, 3336–3345 (2010).
Article ADS Google Scholar
Elderfield, H. & Ganssen, G. Past temperature and δ¹⁸O of surface ocean waters inferred from foraminiferal Mg/Ca ratios. Nature 405, 442–445 (2000).
Article ADS CAS PubMed Google Scholar
Hoff, U., Rasmussen, T. L., Stein, R., Ezat, M. M. & Fahl, K. Sea ice and millennial-scale climate variability in the Nordic Seas 90 kyr ago to present. Nat. Commun. 7, 12247 (2016).
Article ADS CAS PubMed PubMed Central Google Scholar

Download references

Acknowledgements

We thank the WHOI Seafloor Samples Laboratory, including E. Roosen, M. Starr, A. Patterson and G. Nissen, and H. Kuhlmann at the IODP Bremen Core Repository for help with core sampling; B. Atkinson, D. Fairman, I. Patmore and J. Shilland for general laboratory assistance; A. van Dijk and D. Efting for technical support in the clumped isotope laboratory; L. van Maldegem for assistance in the trace-metal laboratory; M. Stanley for help with scanning electron microscopy imaging; M. Daëron, W. Gray, D. Hodell, J. Holmes, J. Rae, L. Roberts, V. Taylor and the UCL OACD group for discussions; M. Wheeler for helping to design Fig. 5; and C. D’Alton and M. Irving for additional design help. J.H.W. was supported by the London NERC Doctoral Training Partnership grant (grant number NE/L002485/1). This project has received funding from NERC Project ReconAMOC (NE/S009736/1), the Leverhulme Trust, NSF grants OCE-1304291 and OCE-2233080, and EU Project 101059547-EPOC. This paper reflects only the authors’ views, and the European Union cannot be held responsible for any use that may be made of the information contained herein. For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any author accepted manuscript version arising.

Author information

Authors and Affiliations

Department of Geography, University College London, London, UK
Jack H. Wharton, Emilia Kozikowska, Mark A. Maslin & David J. R. Thornalley
Department of Earth Sciences, Utrecht University, Utrecht, Netherlands
Emilia Kozikowska & Martin Ziegler
Woods Hole Oceanographic Institution, Woods Hole, MA, USA
Lloyd D. Keigwin & David J. R. Thornalley
Department of Geological Sciences and INSTAAR, University of Colorado, Boulder, CO, USA
Thomas M. Marchitto

Authors

Jack H. Wharton
View author publications
Search author on:PubMed Google Scholar
Emilia Kozikowska
View author publications
Search author on:PubMed Google Scholar
Lloyd D. Keigwin
View author publications
Search author on:PubMed Google Scholar
Thomas M. Marchitto
View author publications
Search author on:PubMed Google Scholar
Mark A. Maslin
View author publications
Search author on:PubMed Google Scholar
Martin Ziegler
View author publications
Search author on:PubMed Google Scholar
David J. R. Thornalley
View author publications
Search author on:PubMed Google Scholar

Contributions

The project was conceived by J.H.W., M.A.M. and D.J.R.T. Marine sediment cores were collected by L.D.K. J.H.W. analysed and interpreted the trace-metal data, with contributions from T.M.M. and D.J.R.T. Clumped isotope data were generated by E.K. and analysed and interpreted by J.H.W., with contributions from E.K., M.Z. and D.J.R.T. Project supervision by D.J.R.T. J.H.W. wrote the first draft of the paper, with contributions from D.J.R.T. All authors contributed to discussion and the final version of the paper.

Corresponding author

Correspondence to Jack H. Wharton.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Map showing the location of cores used in this study.

Northwest Atlantic (a) and Northeast Atlantic (b). Ocean bathymetry is from the GEBCO_2014 global bathymetric grid (30 arc-second resolution)⁹⁷.

Extended Data Fig. 2 Species-specific benthic foraminiferal Mg/Ca-temperature calibration datasets used to develop our common calibration.

Published core-top Mg/Ca (mmol/mol) vs bottom water temperature (°C) are shown for six low-Mg, shallow infaunal benthic foraminifera: (a) Melonis spp.⁵⁹, (b) Bulimina spp.^60,98, (c) C. lobatulus⁶¹, (d) N. labradorica^62,99, (e) C. neoteretis^63,100, and (f) U. peregrina^{56,57,58,76,101,102}. Linear regressions were performed for each species using core-top data, often compiled from multiple studies. The resulting Mg/Ca-temperature relationships and associated R² values are shown. All species exhibit a temperature sensitivity of approximately 0.1 mmol/mol/°C, except U. peregrina, which shows a lower sensitivity (0.7 mmol/mol/°C). Inclusion of an outlier slightly increases the sensitivity ( ~ 0.01 mmol/mol/°C,) however, it remains lower than the other five species associated temperature sensitives. For C. neoteretis, we excluded calibration data from ref. ⁹⁹ due to the absence of a significant trend (R² = < 0.6). Furthermore, including these data results in a lower temperature sensitivity (approximately 0.06 mmol/mol/°C), which yields implausibly cold temperature estimates for relatively high Mg/Ca ratios. In comparison, using data only from refs. ^63,100, produces a higher temperature sensitivity more consistent with other shallow infaunal benthic foraminifera. Given the size of the Melonis spp. calibration dataset, we focused on data from the subtropical and subpolar North Atlantic (n = 82), most relevant to our study location.

Extended Data Fig. 3 Comparison of δ¹⁸O generated as part of this study with equivalent published data and Δ₄₇-based temperature estimates derived using different Δ₄₇-temperature calibrations.

a. Kernel density estimate plots showing the distribution of δ¹⁸O obtained from two independent analyses of benthic foraminifera from the same sediment samples for the mid-to-late Holocene and LGM (Δ₄₇, this study, blue; stable isotope analysis as per ref. ¹⁴, teal). p-values from Welch’s t-tests indicate no significant difference between the two datasets (in both cases p > 0.05), confirming that the analyses yield indistinguishable δ¹⁸O. Close agreement between the published data and published δ¹⁸O data from around the Northwest Atlantic suggests that the δ¹⁸O measured in this study were unaffected by contamination, e.g., from clays. b and c. Comparison of mid-to-late temperature estimates derived using different published Δ₄₇-temperature calibrations: M21 (equation (1))⁶⁸, DG23 (equation (5))¹⁰³, A21 (equation (1))¹⁰⁴, and DV24 (equation (28))¹⁰⁵. Dashed lines in b indicate the approximate modern in-situ temperature at each core site, based on GLODAP 2022 data⁴⁷. The M21 calibration was selected for its strong statistical agreement with modern deep ocean temperatures and methodological consistency, despite being derived, in part, from planktic foraminiferal Δ₄₇ data. This calibration was therefore applied to glacial samples as well.

Source data

Extended Data Fig. 4 Individual and mean monospecific temperature data from the North Atlantic.

Individual Northwest Atlantic Mg/Ca and Mg/Li data for the mid-to-late Holocene (a and b) and LGM (c), and associated ±2 s.e. (2 s.e. = 2 × 1σ (T); methods). Each coloured circle denotes a single monospecific measurement, based on a sample comprising material from 5-15 specimens. d. same as for c but Northeast Atlantic data. e-g, same as a-c but each coloured circle denotes the monospecific mean for each core, with associated ±2 s.e. (Methods).

Source data

Extended Data Fig. 5 Comparison of Northwest Atlantic transect, Northeast Atlantic, and published Atlantic and Arctic Seas temperature and δ¹⁸O_sw data.

a. Temperature-salinity cross plot as in Fig. 2f with additional published estimates of Atlantic glacial temperature and δ¹⁸O_sw, including associated ±1 s.e. (Methods). Data include cores located along the main pathway of modern NADW (open grey symbols; NS⁷⁷, IB, IB and RP (this study), GB⁷⁸), as well as cores in locations influenced by other, non-NADW water masses (open white symbols; NEADW/LDW⁷⁹, STMW/AAIW²⁰, AAIW (FS)⁷⁴, AAIW (DR)⁷⁶, AAIW (BM)⁷⁵, and LCDW⁸⁰). The former align closely with of our Northwest Atlantic data, consistent with our upstream tracing of NADW during the LGM. In comparison, the non-NADW cores are generally characterised by lower δ¹⁸O_sw. This suggests that (1) these sites were influenced by different glacial water masses, and (2) glacial NADW likely had higher δ¹⁸O_sw than these other glacial water masses. One exception is Gulf of Guinea core MD03-2707 (open white square)²⁰, which exhibits relatively high δ¹⁸O_sw. This is expected, given the site’s modern influence from a mix of AAIW and STMWs today, with the latter contributing to relatively high δ¹⁸O_sw. Notably, this core also lies on an approximate mixing line between AAIW and ODP-172-1055, which was bathed by STMWs during the LGM¹⁴. Iberian Margin core MD99-2234 (open white diamond)⁷⁹ and combined data from the Brazilian Margin (open white plus symbol)⁷⁵ also show relatively high δ¹⁸O_sw, likely reflecting the influence of Northeast Atlantic Deep Water in the Northeast Atlantic and southward flowing NADW and/or STMWs in the South Atlantic, respectively. b. Map showing the locations of cores corresponding to the published and Northwest Atlantic transect data presented in a (white shaded areas denote the approximate extent of the Laurentide Ice Sheet (LIS)⁴⁹ and Feno-Scandinavian Sheet (FIS) and British-Irish Sheet (BIS)⁵⁰ at 21.5 ka BP). c. Holocene and LGM deep ocean temperature reconstructions from the Nordic and Arctic Seas (black and grey edged symbols, respectively) derived from ostracod Mg/Ca³³ and benthic foraminiferal Mg/Ca^77,106,107 and benthic foraminiferal Δ₄₇ (ref. ¹⁰⁸; since data from the strict definition of the LGM is sparce, we also include ostracod data from last glacial period, i.e., Marine Isotope Stage 3 (MIS3)). Together, these records highlight that deep waters in the Nordic and Arctic Seas were warmer during the last glacial period than during the Holocene (Source Data). Associated error bars are ±2 s.e.

Source data

Extended Data Fig. 6 Stratigraphic overview of cores used in this study.

To highlight the relative downcore position of our new temperature constraints with respect to the mid-to-late Holocene and LGM, temperature proxy data are presented alongside downcore multi-species benthic or planktic foraminiferal δ¹⁸O data, as well as each cores’ corresponding age constraints. Most multi-species δ¹⁸O data were obtained from published sources, including cores 49JPC, 48JPC, 15JPC, and 10JPC¹⁴; 1055, 1057, 1059, and 66GGC¹⁰⁹;17GGC¹⁵; RAPiD-10-1P and RAPiD-17-5P⁵¹. However, benthic and planktic foraminiferal δ¹⁸O data from 6GGC, 2GGC, 1063, BOFS17K were generated as part of this study following the methods described in ref. ¹⁴ (Source Data). In the absence of downcore foraminiferal δ¹⁸O data from core 1060, new glacial δ¹⁸O (grey crosses) show that our Δ₄₇ analyses were performed on samples corresponding to the glacial benthic δ¹⁸O maxima. We also note that (1) all non-Cibicidioides spp. foraminiferal δ¹⁸O data have been corrected for vital effects, apart from U. peregrina data from cores 1055, 66GGC, 1057, and 1060, and (2) both benthic and planktic foraminiferal δ¹⁸O data were unavailable for cores 1055 and 1057; therefore, we use data from nearby cores KNR-140-2-51GGC and KNR-140-2-43GGC¹⁰⁹, which had been previously stratigraphically aligned¹⁴. Age constraints are taken from published age models (Extended Data Table 1) and comprise radiocarbon ages (filled black triangles) and stratigraphically aligned manual tie points (filled black circles). For each panel, grey shaded areas denote the approximate downcore position of the mid-to-late Holocene (left) and LGM (right). The thick black bars denote the depth range of samples used to generate Δ₄₇ based temperature estimates, as well as the corresponding temperature estimate (for measurements where cores were combined, e.g., 1057/15JPC, the temperature shown for both cores represents the combined estimate).

Source data

Extended Data Fig. 7 Reconstructed \({\boldsymbol{\Delta }}{{\bf{CO}}}_{3}^{2-}\) revealing the vertical \({\boldsymbol{\Delta }}{{\bf{CO}}}_{3}^{2-}\) structure of the deep Northwest Atlantic during the mid-to-late Holocene and LGM.

B/Ca derived \(\Delta {{\rm{CO}}}_{3}^{2-}\) for the mid-to-late Holocene (a) and LGM (b), and associated ±2 s.e. (2 s.e. = 2 × \((\Delta {{\rm{CO}}}_{3}^{2-})\); methods). Each open circle denotes a single C. wuellerstorfi measurement, based on a sample comprising material from 5-15 specimens (open yellow circles). Modern Northwest Atlantic \(\Delta {{\rm{CO}}}_{3}^{2-}\) (grey circles) derived from modern hydrographic measurements from the GLODAP (v2.2022) database⁴⁷. 0 μmol/kg denotes the threshold between oversaturated (>0 μmol/kg) and undersaturated (<0 μmol/kg) conditions⁵⁶.

Source data

Extended Data Table 1 Details of cores used in this study

Full size table

Extended Data Table 2 Species-specific common calibration intercepts, assuming a Mg/Ca-temperature sensitivity of 0.10 ± 0.003 mmol/mol/°C

Full size table

Extended Data Table 3 Details of cores and Mg/Ca-temperature calibrations used to derive estimates of glacial δ¹⁸O_sw-ivc shown in Fig. 5

Full size table

Supplementary information

Supplementary Information (download PDF )

Supplementary Information file containing scanning electron microscopy images of representative benthic foraminifera used for clumped isotope analysis; and an introductory section describing the relevant methods and main findings is followed by nine figures, each comprising four scanning electron microscopy images of monospecific specimens from a single core, accompanied by a caption describing the images.

Peer Review File (download PDF )

Source data

Source Data Fig. 3 (download XLSX )

Source Data Fig. 4 (download XLSX )

Source Data Fig. 5 and Source Data Extended Data Fig. 5 (download XLSX )

Extended Data Source Data Fig. 3 (download XLSX )

Extended Data Source Data Fig. 4 (download XLSX )

Extended Data Source Data Fig. 6 (download XLSX )

Extended Data Source Data Fig. 7 (download XLSX )

Rights and permissions

Open Access This article is 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 you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Wharton, J.H., Kozikowska, E., Keigwin, L.D. et al. Relatively warm deep-water formation persisted in the Last Glacial Maximum. Nature 650, 116–122 (2026). https://doi.org/10.1038/s41586-025-10012-2

Download citation

Received: 18 August 2025
Accepted: 04 December 2025
Published: 21 January 2026
Version of record: 21 January 2026
Issue date: 05 February 2026
DOI: https://doi.org/10.1038/s41586-025-10012-2

Subjects

Abstract

Similar content being viewed by others

Main

Temperature and δ18Osw reconstructions

Warm and salty glacial North Atlantic

Sustained glacial deep-water production

Subtropical hydroclimate forcing

Surface–deep-ocean decoupling

LGM and future ocean implications

Methods

Age models and core sampling

Multi-species and multi-proxy approach

Trace-metal analyses

Trace-metal temperature calibrations

Common Mg/Ca temperature calibration

Clumped isotope analyses (Δ47)

δ18Osw

δ18Osw-ivc

Temperature–salinity plots

ΔT and Δδ18Osw

Carbonate-ion effects

Isotope-enabled models

Published data

Data availability

Code availability

Change history

27 January 2026

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding author

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Extended data figures and tables

Supplementary information

Source data

Rights and permissions

About this article

Cite this article

Share this article

Search

Quick links

Temperature and δ¹⁸O_sw reconstructions

Clumped isotope analyses (Δ₄₇)

δ¹⁸O_sw

δ¹⁸O_sw-ivc

ΔT and Δδ¹⁸O_sw