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 uncertain3,4,5,6, 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.

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) below7,8 (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.

a,b, Meridional sections showing the thermal (a) and haline (b) structure of the Atlantic. The inset in a shows the transect (red box) used to derive these sections. Hydrographic data are from World Ocean Atlas 2023 (WOA23)7,8 and were plotted using Ocean Data View44. Open circles show the locations of sediment cores from the Northwest Atlantic used in this study. Modern deep-ocean water-mass geometry is well resolved in salinity space (b). AAIW, Antarctic Intermediate Water.

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 production10,11. 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 overestimate4,5,13. 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 estimates16,17. 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, δ18Osw) 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 δ18Osw 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 calibrations20,21, 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 δ18Osw (Methods), which is correlated with salinity in the modern ocean.

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

a–c, Trace metal versus estimated growth temperature for multiple species of benthic foraminifera that all show similar temperature sensitivities (a; Methods, Extended Data Fig. 2 and references therein), G. affinis20 (b) and H. elegans21 (c). The shading and dashed lines represent 95% confidence (CI) and 95% prediction (PI) intervals, respectively.

Temperature and δ18Osw reconstructions

Mid-to-late Holocene multi-proxy temperature and δ18Osw estimates from similar depths between 1 km and 4.5 km show good agreement, generally showing temperatures and δ18Osw 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 data7,8 and ostracod-based Mg/Ca temperature estimates22,23,24 (Fig. 3a) from the mid-to-late Holocene. If we calculate density using temperature and a single δ18Osw–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 δ18Osw–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 δ18Osw, we now apply them to the LGM.

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

a–f, Vertical temperature (a,d) and δ18Osw (b,e) profiles, and temperature versus salinity (T/S) cross-plots (c,f) for the mid-to-late Holocene (MH; a–c) and the LGM (d–f). In b and f, the use of grey versus black axis labels denotes weaker (grey) versus more robust (black) proxy reconstruction. The filled coloured symbols in a,** b**,** d** and e represent the mean value for each depth (individual and mean monospecific temperature data and are shown in Extended Data Fig. 4), and associated errors bars are ±2 s.e. (Methods). All δ18Osw data are reported relative to the Standard Mean Ocean Water (SMOW) scale. The dashed black lines are locally weighted scatterplot smoothing lines (smoothing span, 1) through all foraminiferal temperature data from this study. The grey line and ribbon in a and b, respectively, denote the modern temperature from WOA237 and the δ18Osw structure of the Northwest Atlantic (in the absence of modern in situ δ18Osw measurements, a range of δ18Osw was derived using salinity data from WOA23 (ref. 8) and modern salinity–δ18Osw relationships (NADW, North Atlantic (NATL) and LSW25)). The dotted best fit line in d shows the shift to warmer temperatures, most probably owing to the influence of a deeper glacial subtropical gyre at Blake Outer Ridge14. Ostracod temperature data are derived from published benthic ostracod shell Mg/Ca ratios22,23,24. The glacial δ18Osw estimate at approximately 4.5 km is derived using published ostracod Mg/Ca temperature data and nearby published benthic foraminiferal δ18O data45,46. Symbol colours in c and f correspond to core water depth and associated errors are ±1 s.e. (Methods). Isopycnals of σ2 were calculated using modern temperature and salinity measurements from the Global Ocean Data Analysis Project (GLODAP, v2.2022)47 and the Gibbs seawater Oceanographic Toolbox (TEOS-10 standard)48. North Atlantic (20–60° N, 0–80° W) GLODAP (v2.2022) temperature and salinity measurements are also plotted as smaller coloured circles and coloured according to water depth in c. To aid comparison, c and f are offset by 1.1 PSU to account for the LGM–Holocene whole-ocean salinity difference, derived from the change in global sea level. gNADW, glacial North Atlantic Deep Water; gNABW, glacial North Atlantic Bottom Water.

Source data

Apart from site ODP-172-1055 (approximately 1.8 km), which is substantially warmer (about 5 °C), glacial temperature and δ18Osw 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 δ18Osw estimates, as well as with the few other independent data from the Northwest Atlantic22,23,24. 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 NADW5,13, rather than being occupied by distinct northern- and southern-source waters. If we calculate glacial densities using the modern NADW δ18Osw–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 δ18Osw–salinity relationships. For example, our data from sites at 3–4 km hint at being colder and having lower δ18Osw, 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 δ18Osw 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 (ΔTMg/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 (ΔTMg/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.

a,b, Calculated difference in deep-ocean temperatures (ΔT; a) and δ18Osw-ivc (Δδ18Osw-ivc; b) derived from benthic foraminiferal Mg/Ca- and Δ47-based estimates (dark blue and green, respectively) and published sedimentary pore-water δ18O-based estimates (light blue6; Methods; potential temperature was converted to in situ temperature using the Gibbs seawater Oceanographic Toolbox (TEOS-10 standard48); δ18Osw is reported relative to the SMOW scale). For the Northwest Atlantic, we exclude data from the subtropically influenced site ODP-172-1055 and the abyssal site KNR-197-10-17GGC. Glacial Northeast Atlantic data are from sediment cores RAPiD-10-1P, RAPiD-17-5P and BOFS17K (Extended Data Table 1). Associated errors bars are ±2 s.e. (Methods). Modern temperature data were taken from WOA23 (ref. 7) (grey), and in the absence of equivalent δ18Osw, we assume a modern δ18Osw of 0.20 ± 0.2‰ and 0.25 ± 0.05‰ for the Northwest and Northeast Atlantic, respectively (for example, ref. 47). As in Fig. 3, glacial δ18Osw has been corrected by −1.0‰ to account for changes in global ice volume (Methods). We also calculated ΔT and Δδ18Osw-ivc for the LGM and mid-to-late Holocene using paired samples where available; the resulting estimates show close agreement with our broader climatological comparison (ΔTMg/Ca = −1.7 ± 0.7 °C, Δδ18Osw-ivc(Mg/Ca) = 0.5 ± 0.2‰, n = 5; ΔT(Δ47) = −2.2 ± 2.0 °C, Δδ18Osw-ivc(Δ47) = 0.4 ± 0.5‰, n = 3; Methods and Source Data).

Source data

In comparison, average-ice-volume-corrected glacial δ18Osw, hereafter δ18Osw-ivc (Methods), is higher than equivalent modern δ18Osw in the Northwest Atlantic (Δδ18Osw-ivc (Mg/Ca) = 0.38 ± 0.14‰; Δδ18Osw-ivc(Δ47) = 0.23 ± 0.24‰; Fig. 4b) and Northeast Atlantic (Δδ18Osw-ivc(Mg/Ca) = 0.51 ± 0.19‰). This suggests that previous glacial deep North Atlantic δ18Osw 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 Atlantic16,17. Therefore, although deriving reliable palaeosalinity estimates for the North Atlantic during the LGM is challenging—owing to the spatial and temporal variability in the δ18Osw–salinity relationship26—higher glacial δ18Osw-ivc is probably due to a combination of variable δ18Osw–salinity relationships and saltier NADW relative to the Holocene.

We also compared our δ18Osw-ivc reconstructions with the limited number of available isotope-enabled LGM simulations, which produce a relatively wide range of δ18Osw-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 δ18Osw-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 reconstructions5,14, 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 δ18Osw-ivc data with equivalent glacial data from sites along the modern pathway of NADW and its near-surface source waters reveals that the high δ18Osw-ivc signature of glacial NADW can be traced along that same route. Figure 5 shows how high δ18Osw-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 δ18Osw-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 δ18Osw-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.

a, Map showing the location of core sites used to reconstruct surface-ocean (red filled circles) and deep-ocean (blue filled circles) δ18Osw-ivc from the glacial North Atlantic (Methods, Extended Data Table 3 and references therein, and Source Data). The numbered star (11) denotes the approximate position of our Northwest Atlantic transect). The red and blue arrows denote surface- and deep-ocean currents, respectively. 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. GS, Gulf Stream. b, Simplified schematic showing the potential upstream pathway of high δ18Osw-ivc NADW (δ18Osw is reported relative to the SMOW scale). The numbers along the arrow correspond to the numbered core sites in a. For consistency and where necessary, both planktic and benthic Mg/Ca temperature data were recalibrated using new and/or updated calibrations and foraminiferal calcite (δ18Oc) was corrected using species-specific corrections14 (Methods and Source Data).

Source data

Subtropical hydroclimate forcing

As our reconstructed glacial δ18Osw-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 δ18Osw (ref. 30), with negative P − E causing higher δ18Osw. Therefore, we infer that the high glacial δ18Osw-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 Caribbean31,32, 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 winds2,14. Although there are complexities in relating δ18Osw to salinity26, these model results suggesting lower P − E over the subtropical gyre provide support for our inference that the high δ18Osw 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 Atlantic10,11.

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