Overturning circulation, nutrient limitation, and warming in the Glacial North Pacific

Enhanced overturning circulation in the glacial North Pacific caused warming, nutrient limitation, and lower atmospheric CO₂.

Abstract

Although the Pacific Ocean is a major reservoir of heat and CO₂, and thus an important component of the global climate system, its circulation under different climatic conditions is poorly understood. Here, we present evidence that during the Last Glacial Maximum (LGM), the North Pacific was better ventilated at intermediate depths and had surface waters with lower nutrients, higher salinity, and warmer temperatures compared to today. Modeling shows that this pattern is well explained by enhanced Pacific meridional overturning circulation (PMOC), which brings warm, salty, and nutrient-poor subtropical waters to high latitudes. Enhanced PMOC at the LGM would have lowered atmospheric CO₂—in part through synergy with the Southern Ocean—and supported an equable regional climate, which may have aided human habitability in Beringia, and migration from Asia to North America.

INTRODUCTION

Ocean circulation exerts a fundamental control on heat transport, marine biological productivity, and ocean-atmosphere CO₂ exchange (1). Today, the North Atlantic and North Pacific have contrasting circulation regimes, with vigorous deep-water formation in the North Atlantic, but only limited local ventilation in the North Pacific (2–4). This drives distinct patterns of CO₂ exchange, nutrient supply, and heat transport in each basin (5) and represents a first-order asymmetry in global biogeochemistry and climate (Fig. 1). It has been proposed that the circulation of the North Pacific was markedly different in past cold climates, but no consensus on this change exists. Proxies for ocean ventilation suggest enhanced intermediate water formation under glacial conditions (6–11) and local deep-water formation during Heinrich Stadial 1 (12, 13). However, paired nutrient utilization and productivity proxies indicate reduced nutrient supply to the surface during cold climates (14, 15), interpreted as the result of more stratified conditions with reduced ventilation. As the Pacific contains around half of the water in the global ocean and 30 times more carbon than the atmosphere, the lack of consensus on its behavior in different climate states represents a major gap in our understanding of the global climate system. Understanding the glacial climate of this region is also of interest for understanding the conditions under which early humans first migrated from Asia to North America around the end of the last ice age (16). Here, we use a compilation of sediment core proxy data for ocean ventilation, biological productivity, temperature, and salinity (see the Supplementary Materials for details), paired with Earth system modeling experiments, to shed new light on the circulation of the glacial North Pacific and its regional and global impact.

Fig. 1. Modern hydrography and nutrient content of the northern Pacific and Atlantic Oceans.

(A) Sea surface salinity, with sea surface temperature contours for 5° and 10 °C. (B) Surface phosphate concentration, with a sea surface height contour to denote the subpolar gyre boundary. (C) Subsurface phosphate concentration along a zonal section indicated by the dashed line in (B), with pre-bomb radiocarbon age contours. Note the relative isolation of the subpolar gyre in the North Pacific, which allows pooling of cold, fresh water in the surface, and old, nutrient-rich waters to upwell from below. In contrast, the active overturning circulation of the North Atlantic flushes warm, salty, nutrient-poor water from the subtropics through the upper reaches of this well-ventilated basin. Salinity, temperature, and phosphate data are gridded annual averages from WOA09; radiocarbon data from Global Ocean Data Analysis Project version 2 [GLODAPv2 (101)]; and sea surface height from (159).

RESULTS AND DISCUSSION

Data compilation

To examine past changes in ventilation, we use proxies including δ¹³C in benthic foraminifera (6, 17), ¹⁴C offsets between benthic and planktic foraminifera (18, 19), and various redox tracers (7). These data indicate that waters below ~2000 m are poorly ventilated at the Last Glacial Maximum (LGM), with lower δ¹³C, ¹⁴C, and oxygen (Figs. 2A and 3). This is consistent with previous interpretations of the deep glacial Pacific as a likely store of glacial CO₂ (1, 20). In contrast, waters above ~2000 m show enhanced ventilation (Figs. 2A and 3). This is seen across all tracers, with 73% of the 72 sites above 2000 m indicating increased ventilation, 17% showing no change outside of 1σ uncertainty, and only 10% indicating reduced ventilation. This suggests a substantial increase in North Pacific Intermediate Water (NPIW) formation at the LGM (6). This result is further supported by profiles of δ¹⁸O in benthic foraminifera (6, 11): In contrast to the modern, where a smooth and modest increase in δ¹⁸O with depth suggests a dominantly diffusive regime, LGM profiles (6, 11) exhibit a notable transition at ~2000-m water depth (Fig. 3A). This indicates a sharp transition in temperature and/or δ¹⁸O_sw, which, as conservative tracers, provide evidence of a physical water mass boundary (21), with increased advection of NPIW overcoming the smoothing influence of diffusion. A local source of intermediate waters in the glacial North Pacific is also supported by meridional sections of benthic ¹⁴C age and δ¹³C (Fig. 3 and fig. S3), which show the spread of well-ventilated waters from the surface into the ocean’s interior.

Fig. 2. Reconstructions of changes in ventilation, export productivity, temperature, and salinity at the LGM relative to the Holocene.

Red indicates an increase at the LGM relative to the Holocene, blue a decrease, and white no clear change outside of 1σ uncertainty (see the Supplementary Materials for details of data compilation). (A) Ventilation proxies, including δ¹³C (circles), radiocarbon (diamonds), and a variety of redox tracers (squares), taken from sites above 20°N. (B) Export productivity proxies, including sediment core opal (circles) and biogenic barium (squares) contents. (C) Sea surface temperature proxies, including planktic foraminiferal Mg/Ca (circles) and alkenone saturation index UK’37 (diamonds). (D) Change in sea surface salinity, derived from paired δ¹⁸O and Mg/Ca data on planktic foraminifera, corrected for whole-ocean δ¹⁸O and salinity changes due to ice volume. At the LGM, North Pacific intermediate waters are better ventilated, and productivity in the subpolar North Pacific is lower. An increase in salinity is seen throughout the North Pacific, and although the subtropical gyre cools, the subpolar North Pacific shows an anomalous warming, despite peak glacial conditions. See fig. S1 for an alternative presentation of these data.

Fig. 3. Profiles and meridional sections of proxies for deep ocean circulation.

The δ¹⁸O profile (A) uses Cibicidoides spp. (circles) and Uvigerina spp. (squares), with Uvigerina data corrected for vital effects by –0.47‰ (160); the δ¹³C profile (B) uses Cibicidoides spp. only; profile data are from the NW Pacific (6). LGM δ¹³C data (in B and D) and δ¹⁸O are corrected for the –0.34 and 1‰ whole ocean changes in δ¹³C and δ¹⁸O_sw (17). Generalized additive model fits are shown for Holocene δ¹³C and δ¹⁸O (dashed lines) and LGM δ¹³C (solid line) (161). The thin dotted-dashed line through the LGM δ¹⁸O data in (A) is the Holocene δ¹⁸O fit with a +0.6 ‰ offset; while the Holocene δ¹⁸O profile increases smoothly with depth, the LGM data exhibit a marked transition at ~2000 m, indicative of a water mass boundary. ¹⁴C data in (C) are shown as ¹⁴C age relative to the contemporaneous atmosphere, with modern data from GLODAPv2 and LGM data from benthic foraminfera (see Materials and Methods for details). δ¹³C section data in (D) are from the western Pacific (17)—see inset map. Locations of sediment core data are shown in black dots and areas of poor data coverage in white, and further details are given in figs. S2 and S3.

To examine changes in sea surface temperatures (SSTs), we use Mg/Ca in planktic foraminifera and the alkenone saturation index U^K′₃₇. SSTs in the subtropical gyre are cooler at the LGM, as expected for glacial conditions, and there is a large midlatitude cooling due to the southward expansion of the subpolar gyre (Fig. 2C and fig. S4) (22). In contrast, LGM temperatures in the subpolar gyre are either similar to or warmer than today (80% of records show warmer temperatures or no significant change), with ~2°C warming observed in the western North Pacific (Fig. 2C and figs. S1C and S4) (23, 24). These warm SSTs require a circulation change to counter lower CO₂ and greater ice coverage (25) at the LGM and are consistent with enhanced heat transport by an invigorated Pacific Meridional Overturning Circulation (PMOC).

Sea surface salinity change at the LGM is estimated using δ¹⁸O in planktic foraminifera, corrected for temperature using Mg/Ca and for whole-ocean δ¹⁸O/salinity change. Increased salinity is indicated throughout the North Pacific subpolar gyre (Fig. 2D and fig. S4), which would have aided intermediate water formation.

To reconstruct changes in export productivity, we use sedimentary opal and biogenic barium contents (26). All data poleward of 40°N indicate a reduction in glacial export productivity (Fig. 2B and fig. S1B). However, tracers of nutrient utilization (e.g., δ¹⁵N) suggest that biological nutrient drawdown was more complete (15). For more complete nutrient drawdown to be achieved despite reduced export productivity, the pool of available major nutrients (phosphate, nitrate, and silicic acid) must have been smaller (1, 15). This is further supported by considering that micronutrient (e.g., iron) availability, which partially limits productivity in the North Pacific today, was enhanced because of higher dust fluxes (27). The decrease in export productivity in the LGM North Pacific thus requires a decrease in major nutrient supply.

Mechanisms for glacial nutrient limitation

Previous studies have suggested that nutrient limitation in the glacial North Pacific was driven by enhanced freshwater stratification in surface waters under a colder climate (14, 15). However, this suggestion is difficult to reconcile with our evidence for enhanced intermediate water formation (Figs. 2A and 3) and saltier, warmer surface waters (Fig. 2, C and D, and fig. S4). An alternative model is thus required for large-scale changes in nutrients and circulation in the northern subpolar basins.

To explore controls on circulation and nutrient supply in the North Pacific, we first consider modern phosphate concentrations in an adaptation of Warren’s box model [(2, 3); the Supplementary Materials], which was previously used to demonstrate that the low salinity that currently stratifies the subpolar North Pacific is a result of low evaporation rates relative to precipitation, as well as minimal exchange with the saltier subtropical gyre (2–4). Isolation of the subpolar gyre also contributes to high surface nutrients in the modern North Pacific, by limiting the input of low-nutrient water from the subtropics (fig. S5). Although salinity stratification limits the depth of convection, subsurface waters are still brought to the surface via wind and tidally driven mixing, wintertime convection, and Ekman suction (5). As these subsurface waters have been long isolated in the deep ocean, they are extremely rich in nutrients acquired through remineralization, with little input of low-nutrient subtropical water or newly ventilated intermediate water to dilute this potent subsurface nutrient reservoir (Fig. 1C). This contrasts with the North Atlantic, where a less zonal wind-stress pattern and an active overturning circulation supplies warm, salty, and nutrient-poor subtropical water to high latitudes (Fig. 1). Subduction of this water creates nutrient-poor intermediate and deep waters and effectively flushes nutrients from the upper reaches of the modern North Atlantic.

Thus, today we have two realizations of basin-scale nutrient dynamics and circulation at high northern latitudes: The Pacific is stratified and poorly ventilated, with cold, fresh, and nutrient-rich surface waters, supplied by upwelling of old, nutrient-rich waters from below, while the Atlantic is well ventilated, with a vigorous overturning circulation that flushes warm, salty, and nutrient-poor waters from the subtropics through its upper reaches. At the LGM, our data show that the North Pacific was better ventilated, had lower nutrients, and was saltier and warmer. This is consistent with a more Atlantic-like circulation regime, with enhanced overturning down to ~2000-m depth.

Modeling experiments

The ability of enhanced overturning to reduce nutrient supply to the surface of the glacial North Pacific is further supported by a series of experiments with an Earth system model (cGENIE) and tests with a simple box model (see the Supplementary Materials). We simulate Pacific overturning in cGENIE by reducing the prescribed transport of fresh water from the North Atlantic to the North Pacific. As the North Pacific gets saltier, its meridional overturning circulation increases (fig. S6), driving a decrease in surface and subsurface nutrients (Fig. 4 and figs. S5 and S7) due to increasing input of nutrient-poor subtropical waters. Although convective mixing increases and stratification is reduced (Fig. 4C), the dilution of the subsurface nutrient reservoir by freshly ventilated waters has a much larger effect (Fig. 4 and figs. S5 and S7; see Materials and Methods). This pattern is observed across different model base states and is also seen in the higher-resolution LOVECLIM and UVic Earth system models (fig. S8) (28). Recent work using the Paleoclimate Modelling Intercomparison Project (PMIP3) ensemble shows that Ekman suction in the glacial North Pacific was around 60% higher than today (29). Given this increase in upwelling, a substantial decrease in the nutrient content of the upwelled water is required to result in lower surface nutrient concentrations; our modeling results show this can be effectively achieved by enhanced overturning circulation.

Fig. 4. cGENIE Earth system model experiments illustrating the impact of changes in North Pacific overturning on nutrient concentrations.

Each symbol (A to C) represents a 5000-year-long model experiment under preindustrial (orange triangles) or glacial (blue diamonds) boundary conditions. Experiments are forced by decreasing the atmospheric freshwater flux into the North Pacific, which increases surface salinity and enhances PMOC (fig. S6). Output is plotted against the maximum meridional stream function in the North Pacific below 280 m. Phosphate concentrations are shown for (A) the surface layer (top 80 m) and (B) intermediate depths (928 to 1158 m) in the subpolar gyre (grid cells spanning 43°N to 66°N and 115°W to 225°W). Convection frequency (C), a count of convection events throughout the water column per model time step, is shown for the same region. Meridional sections of phosphate concentration (shading) at 165°W and Pacific meridional overturning stream function (contours) are shown in (D), for an experiment with a salinity forcing of −0.19 Sv and a PMOC of 8 Sv [indicated by a dashed line in (A) to (C)] that produces the best fit to our LGM data (Fig. 5 and fig. S10); anomalies of phosphate and overturning from the glacial base state are shown in (E); and the spread of North Pacific water at 1000-m depth, predominantly on the basin’s western boundary, is shown in (F).

Pacific overturning and biogeochemistry during other climate states

The presence or absence of a PMOC may help explain biogeochemical transitions in other time intervals. For example, the peak in nutrient supply during the Bølling-Allerød warm interval [14.7 to 12.9 thousand years ago (ka)] appears to be associated with an abrupt decrease in PMOC (12, 29), while the decrease in export productivity and increase in SST during intensification of Northern Hemisphere glaciation ~2.7 million years ago (14) could be explained by increased PMOC, analogous to the change from modern to LGM conditions. We note that, as a transient condition, the onset of deep-water formation can lead to a short-lived peak in surface nutrients and CO₂, as convection initially taps into high-nutrient subsurface waters before they have been replaced with well-ventilated low-nutrient waters (fig. S9); this remains a likely explanation for North Pacific pH, δ¹³C, and ¹⁴C excursions (12) during Heinrich Stadial 1 (HS1; ~17.5 to 16.0 ka). However, the typical pattern under steady-state conditions—both in paleo records and our modeling—appears to be that increased overturning drives a decrease in nutrient supply (Fig. 4).

The dynamics of a glacial PMOC

Increased ventilation of the North Pacific at the LGM would be facilitated by an increase in the salinity of subpolar surface waters (2, 3), as implied by the available data (Fig. 2D and fig. S4). Higher salinity in the glacial North Pacific may be driven by reduced net precipitation (P-E), due to a weakened hydrological cycle under globally cooler conditions (30), and a southward shift in storm track precipitation (31). Increased Ekman suction in the presence of the Laurentide ice sheet (29) would also bring up salt from subsurface waters. Salinity may be further elevated by increased exchange with the saltier subtropical gyre, due to an intensification of the westerly winds in response to the Laurentide ice sheet (22, 29), or as a Stommel feedback to an initial increase in overturning (32). Although an increase in SST would, in isolation, decrease the density of surface waters, this may be overcome by an increase in salinity, in part resulting from enhanced evaporation and divergent moisture transport (2); the combination of increased SST and increased North Pacific ventilation is indeed found in ocean circulation models (figs. S7 and S11). Last, it is possible that seasonal sea ice formation and brine rejection play a role (8, 25), in particular, in helping to set the local sites of downwelling, which are likely to be in the marginal seas (10). Enhanced sea ice production may be reconciled with warmer SSTs either by a difference in seasonality or if sea ice production occurs predominantly in coastal settings (8). Although basin geometry and a closed Bering Strait may, in part, limit salinity in the North Pacific (4), our proxy reconstructions show that local ventilation to intermediate depths is indeed possible, a result also found in the higher-resolution NCAR CCSM3 and GFDL CM2Mc models under some instances of colder climate conditions (figs. S11 and S12).

Comparing reconstructed δ¹³C profiles and other ventilation data with cGENIE output (Fig. 5 and fig. S10) suggests a PMOC of around 8 Sv at the LGM, which is also consistent with the changes seen in SST (fig. S7). Note, however, that the PMOC cell remains largely above 2000 m and within the Pacific basin in our cGENIE experiments (fig. S7), in contrast with Atlantic Meridional Overturning Circulation (AMOC) at similar overturning rates. δ¹³C values in the intermediate North Pacific also remain lower than those in the glacial North Atlantic (fig. S13). These observations may, in part, be explained by the much greater volume of the Pacific, which increases the residence time of northern-sourced waters within the basin, and allows remineralization and mixing with southern-sourced waters to exert greater influence. While glacial AMOC may have remained higher than PMOC, the LGM’s overturning circulation, with northern-fed cells in both basins to ~2000 m overlying southern-sourced waters, was substantially more symmetrical than today. It is possible that this more symmetrical glacial circulation is the most common configuration of the overturning circulation during the late Pleistocene, with today’s asymmetric mode limited to peak interglacial conditions.

Fig. 5. Simulated δ¹³C profiles in cGENIE compared to benthic δ¹³C data in the NW Pacific.

Data (6) are shown as symbols with a general additive model fit shown by the gray dashed line. cGENIE simulations are given by the colored lines. To aid comparison, all profiles are shown as δ¹³C anomalies from the mean δ¹³C for each profile. Left: Late Holocene data and the experiments run under preindustrial boundary conditions. Right: LGM data and experiments run under glacial boundary conditions. The legend lists the maximum North Pacific overturning for each experiment. The upper portion of the LGM δ¹³C profile shows the closest match to the simulation run with an Atlantic to Pacific freshwater forcing of −0.19 Sv, which has a PMOC of 8 Sv. As the lower portions of the profiles likely remain ventilated by Southern Ocean waters, and as we have not made any changes to the Southern Ocean in our simulations, some offset at depth is to be expected.

Impact of PMOC on CO₂

Enhanced PMOC would have increased the North Pacific’s influence on the glacial carbon cycle and climate (Fig. 6). Today, the North Pacific is a High Nutrient Low Chlorophyll zone, with an inefficient biological pump and high CO₂ outgassing. By reducing the carbon content of the high-latitude North Pacific, an invigorated PMOC would have helped stem this CO₂ leak, as indicated by boron isotope data, which show LGM surface ocean CO₂ close to equilibrium with the atmosphere (29). Furthermore, these changes in the surface ocean would have had a greater impact on the net ocean CO₂ storage, because enhanced intermediate water formation would lower the preformed nutrient content of the ocean’s interior. Combining our cGENIE results with a simple scaling between preformed phosphate and atmospheric CO₂ (33) shows that atmospheric CO₂ is relatively sensitive to modest increases in PMOC and suggests CO₂ drawdown on the order of ~16 parts per million (ppm) (see Materials and Methods and fig. S14). CO₂ drawdown may be further enhanced by PMOC’s impact on biogeochemistry in the Southern Ocean (34). Today, carbon-rich waters recirculated from the Pacific upwell in the Southern Ocean and contribute to high surface nutrients and CO₂ outgassing in this region. Replacing a portion of this upwelling water with well-ventilated glacial NPIW reduces the carbon and nutrient supply to the Southern Ocean surface (see Earth system model results in fig. S8), thus increasing the efficiency of the Southern Ocean biological pump and further lowering atmospheric CO₂. This process may help explain the similarity in nutrient utilization records from the North Pacific and Southern Oceans [the “polar twins” (35)]. If enhanced glacial PMOC results from increased sea ice and ice sheet size, then this would provide a feedback between Northern Hemisphere insolation and atmospheric CO₂, linking the orbital pacing of the ice ages to global glacial climate change.

Fig. 6. Schematic of ocean circulation and productivity in the modern and LGM North Pacific.

The modern North Pacific (left) lacks vigorous local ventilation (thin black arrows), because of low salinity in surface waters of the subpolar gyre (SPG). This is a result of high net precipitation (P-E) and minimal exchange with the saltier waters of the subtropical gyre (STG). The modern subpolar North Pacific is thus dominated by upwelling of nutrient- and CO₂-rich subsurface waters (gray vertical arrow), driving relatively high export productivity (green wavy arrows) and CO₂ outgassing. During the LGM (right), our data compilation suggests that ventilation at intermediate depths was enhanced, export productivity was reduced, and subpolar surface waters were saltier and warmer (red wavy arrows). This is consistent with an invigorated meridional overturning circulation, with enhanced formation of intermediate waters and advection of warm, salty, and nutrient-depleted subtropical waters to high latitudes, analogous to a shallower version of the overturning circulation seen in the modern North Atlantic.

Regional warming in Beringia due to enhanced PMOC

The relatively mild regional climate supported by heat transport from an enhanced PMOC may have implications for the first human migration to the Americas. Recent genetic studies suggest that the founding population of early Native North and South Americans were isolated from other, future Asian populations during the LGM, before dispersing into and through the American continents (16). This isolation is thought to have been located in northeastern Siberia or on the now submerged Bering Sea shelf, lasting between 7000 and 15,000 years (36) during what is known as the “Beringian standstill.” Some terrestrial records support the interpretation that this region was an ecological refugium, with spruce, birch, and alder pollen and fossil insects providing evidence of a relatively mild and maritime climate (37). However, given the extremely cold conditions further inland (38), it has remained unclear how habitable conditions could have been maintained. Our data and modeling suggest that an enhanced overturning circulation helped warm this region at the LGM, countering the influence of lower greenhouse gasses and higher albedo, to bring Pacific Beringian temperatures up to, or even warmer than, modern (Fig. 2C and figs. S1C and S4). Warming may have been particularly enhanced during LGM winters, as seen in CCSM3 (fig. S11B), which shows that for a mean annual temperature change of ~2°C in the glacial northwest Pacific (similar to our reconstructions), winter warming may have been around 6°C. An enhanced Pacific overturning circulation could have thus aided human habitation in Beringia during a period of refuge before migration to North America.

MATERIALS AND METHODS

Ventilation data

Sediment core proxy data for ventilation at the LGM includes δ¹³C data on benthic foraminifera, tracers of sedimentary redox, and radiocarbon age offsets between benthic and planktic foraminifera, taken from sites above 20°N in the North Pacific (with a version above 40°N shown in fig. S1A). Stable carbon isotope data (δ¹³C) (6, 9, 11, 39–53) on benthic foraminifera of the genus Cibicides are largely taken from the recent synthesis by Peterson et al. (17), supplemented with recent studies (9, 53). Where possible, we compare foraminiferal data for the LGM (19 to 23 ka) and late Holocene (0 to 6 ka). At sites where Holocene foraminiferal data are not available (24 of 62), we use seawater δ¹³C data taken from the PACIFic ocean Interior CArbon (PACIFICA) database (52), with values interpolated for these core sites using the three-dimensional estimation tool in Ocean Data View [ODV, (54)], except at Keigwin’s (6) depth transect in the NW Pacific and Max’s (9, 53) sites in the Bering Sea, where we use the accompanying water column measurements. Both approaches show an increase in δ¹³C at intermediate depths. To account for whole-ocean change in δ¹³C at the LGM (attributed to a smaller terrestrial biosphere), and thus better reveal patterns of change in circulation, we correct LGM δ¹³C values by +0.34‰ (17). We then calculate the LGM minus Holocene δ¹³C difference, with positive values indicating higher δ¹³C and increased ventilation at the LGM. We use a threshold value of ±0.1‰ for a significant increase/decrease in δ¹³C when comparing LGM and Holocene foraminiferal data, based on 1 SD of replicate samples of benthic foraminifera from the NW Pacific (6). When comparing LGM foram δ¹³C to modern water δ¹³C, we use a threshold value of 0.2‰, to account for calibration error [e.g., (42)] and the sparseness of water column δ¹³C data in this region.

Redox reconstructions (55–79) follow the approach and compilation of Jaccard and Galbraith (7), who estimated whether oxygen concentrations increased or decreased at the LGM (defined in their study to be 20 to 22 ka) compared to Holocene (5 to 10 ka). Redox proxies include trace metals, sediment laminations, assemblages of benthic foraminifera, and nitrogen isotopes from regions of modern denitrification. Following Jaccard and Galbraith (7), redox changes based on nitrogen isotopes are plotted at 400 m to reflect the depth of water column denitrification. We use the published thresholds for significant change in redox, quoted as approximately 1 SD.

Radiocarbon data (12, 20, 53, 70, 76, 80–100) are taken largely from syntheses by (13, 18–20). Glacial data shown in Fig. 2A consist of ¹⁴C age offsets between benthic and planktic foraminifera from the same sample and include data with calendar ages in the window 18 to 23 ka. Holocene benthic-planktic ¹⁴C data are rare in this region because of poor carbonate preservation, so we compare LGM benthic-planktic offsets to deep-surface offsets in water column ¹⁴C. These are calculated from nearby deep-water ¹⁴C data from GLODAPv2 (101) and a surface water age of 400 ¹⁴Cyr south of the subpolar gyre boundary (43°N) and 700 ¹⁴Cyr within the subpolar gyre. The threshold for a significant change at the LGM is the quadratic combination of ¹⁴Cyr 1-SE measurement uncertainties on LGM benthic and planktic ¹⁴C data and a 200-¹⁴Cyr uncertainty on modern deep-surface offsets, and averages 289 ¹⁴Cyr. The section view in Fig. 3C uses the datasets listed above with benthic data shown as age offsets from the contemporaneous atmosphere (102). Note that despite their varying influences and sources of uncertainty, each set of ventilation proxy data shows the same patterns of change at the LGM, with enhanced ventilation at intermediate depths and reduced ventilation in the abyss.

Productivity data

Biogenic opal and barium data (26, 59, 74, 76, 87, 103–118) were taken from the compilation of Kohfeld and Chase (26) and supplemented with recent data from (12, 15, 87, 119–122). LGM and Holocene mean values and SEs were calculated from data in the time windows 0 to 10 ka (Holocene) and 19 to 23 ka (LGM). A significant change is noted when the difference between the LGM and Holocene means exceeds the quadratic sum of their 1-SE values. Sedimentary opal (%) and barium (ppm) content are shown in Fig. 2B. Where available, published mass accumulation rate estimates of these biogenic components are shown in fig. S1B, largely based on sediment core age models (from δ¹⁸O and ¹⁴C) and dry bulk density (either measured or estimated); fluxes based on ²³⁰Th are indicated with bold symbol outlines (fig. S1B). Each of these treatments of the data indicates reduced export productivity in the LGM subpolar North Pacific.

Temperature and salinity data

Published records of planktic foraminiferal Mg/Ca and δ¹⁸O_calcite (24, 29, 82, 87, 123–127), U^K′₃₇ (112, 113, 128–135) covering the Holocene (0 to 10 ka) and LGM (19 to 21 ka) were collated for the North Pacific, spanning both the subpolar and subtropical gyres. All age models are as given in the original publication. All Mg/Ca were recalibrated (see below). U^K′₃₇ temperatures are as given in the original publication. For both Mg/Ca and U^K′₃₇, temperature change during the LGM is given as a difference to proxy-derived temperature in the Holocene.

We recalibrated the published Mg/Ca data using the “MgCaRB” approach, as detailed in (136), which accounts for changes in the nonthermal influences of salinity and pH downcore. For Globigerinoides ruber and Globigerina bulloides, we use the species-specific equations from (136). For Neogloboquadrina pachyderma, we use the species-specific sensitivities given in (137), which we incorporate into the MgCaRB protocol (note that N. pachyderma Mg/Ca appears insensitive to pH). While the direct temperature sensitivity of Mg/Ca in planktic foraminifera is ~6%/°C (136), because of the effect of temperature on pH through the disassociation constant of water (K_w), the “apparent” Mg/Ca temperature sensitivity is higher in pH-sensitive species (138). The MgCaRB approach accounts for the mean changes in surface salinity (due to sea level) and pH (due to atmospheric CO₂) downcore, iteratively solving Mg/Ca and pCO₂ to overcome the covariance induced by the thermal effect on K_w. Our approach does not account for local changes in salinity/pH; however, as local salinity and pH changes in both increase during the LGM, their effects on Mg/Ca work against each other, such that the bias on reconstructed temperature are likely to be small (a local 1–practical salinity unit (PSU) salinity and 0.05–pH unit increase would sum to zero temperature bias). Dissolution can also affect foraminiferal Mg/Ca, although this influence is thought to be relatively minor in lower-Mg planktic foraminifera such as N. pachyderma (139), the species analyzed most frequently in the high latitudes. LGM-Holocene changes in bottom water carbonate ion concentration are also relatively minor within the deep Pacific, decreasing by ~10 μmol kg⁻¹) during the LGM (140). Hence, although dissolution is unlikely to have a significant influence on changes in SST calculated from changes in Mg/Ca, the slight decrease in glacial bottom water carbonate ion concentration within the Pacific (140) would result in an apparent cooling in Mg/Ca SSTs. If we consider the sign of change in the secondary influences on Mg/Ca (local salinity, pH, and dissolution) from Holocene to LGM, then the warming signal that we see in Mg/Ca SSTs during the LGM is likely to be a conservative estimate; hence, the increase in ice volume–corrected seawater δ¹⁸O (salinity) that we see is also likely to be a conservative estimate. For U^K′₃₇, the change in temperature at each site was calculated using the calibrations given in the original publications (141, 142); the temperature range in this study is too low to be significantly influenced by the nonlinearity of U^K′₃₇ at >~24° (143).

The change in ice volume–corrected δ¹⁸O of seawater (δ¹⁸O_sw-ivc) was calculated from samples with paired measurements of Mg/Ca and δ¹⁸O_calcite. We use the change in temperature calculated from the Mg/Ca values (described above) and the temperature-δ¹⁸O fractionation factor of (144) to remove the temperature signal from δ¹⁸O_calcite. The effect of sea level was accounted for using the whole-ocean glacial δ¹⁸O_sw change of 1‰ (145) scaled to the sea level record of (146). Note that the ice volume correction on δ¹⁸O accounts for the fact that whole-ocean salinity was higher at the LGM; thus, an increase in δ¹⁸O_sw-ivc represents an increase in salinity beyond the whole-ocean salinity increase.

We calculate the change in temperature and δ¹⁸O_sw-ivc between the Holocene (0 to 10 ka) and LGM (19 to 21 ka) time slices for each record individually, with the change in temperature defined against the average for the Holocene. For five sites, we use the interval 0 to 11.5 ka for the Holocene and/or 17.9 to 21 ka for the LGM, and these sites are marked with a gray outline in Fig. 2C and fig. S1C; excluding these sites makes no difference to any of the conclusions drawn. We note that the use of data of early Holocene (which is relatively warm compared to the late Holocene and preindustrial; fig. S4) (147) within our Holocene time window means that the estimate of LGM warming is again likely to be conservative. An increase/decrease is indicated by a change of >±1°C for Mg/Ca, >±1.2°C for U^K′₃₇, and >±0.25‰ for δ¹⁸O_sw-ivc. These thresholds correspond to the uncertainties associated with the temperature proxies (136, 141, 142) and the equivalent in δ¹⁸O (144).

Time windows

Our data compilations take advantage of available syntheses of data from the North Pacific where available, supplemented with recently published data. Time slices selected for the LGM in these syntheses, as detailed above, are fairly consistent. Holocene time slices show more variety although, in each case, aim to most accurately represent postglacial Holocene conditions. For δ¹³C, this requires that the glacial-interglacial whole-ocean/atmosphere/biosphere change is complete, which was only achieved in the late Holocene. For some redox proxies, core-top values are liable to disturbance, so an early Holocene slice may be more appropriate. Late Holocene carbonate preservation in the North Pacific is poor, limiting Holocene proxy records based on planktic foraminifera. As a result, LGM benthic-planktic ¹⁴C data are compared to modern values from the water column. Last, as noted above, the use of early Holocene data within our LGM to Holocene temperature comparison is likely to underestimate the magnitude of LGM warming compared to preindustrial, given the warmth of the early Holocene (147), so the LGM warming signal seen in Fig. 2C and figs. S1C and S4 is likely to be conservative. We also provide a δ¹³C profile for a time slice ~3000 years older than the original LGM pick in fig. S2, which demonstrates that the enhanced ventilation signal is robust to age model uncertainties, and cannot be attributed to aliasing of an HS1 signal.

Box model

We use an adaptation of Warren (2) and Emile-Geay et al.’s (3) salinity budget for the North Pacific subpolar gyre to explore controls on salinity and nutrients in this region. The model considers a mass balance for water in a subpolar gyre box, representing latitudes 43°N to 63°N, longitudes 140°E to 120°W, and water depth 0 to 200 m, with inputs from the subtropical gyre (V_STG), upwelling (V_UP), net precipitation (P-E), and riverine runoff (R). Salinity of the subpolar gyre (S_SPG) is calculated from the ratio of inputs of salt to inputs of water

$$S_{\text{SPG}}=\frac{S_{\text{UP}}{V}_{\text{UP}}+S_{\text{STG}}{V}_{\text{STG}}}{V_{\text{UP}}+V_{\text{STG}}+(P-E)+R}$$

and depends on the salinity of subtropical input (S_STG) and upwelling water (S_UP). For the derivation of this equation, see (2) and (3). We follow the same approach to model phosphate concentration in the subpolar gyre (PO₄³⁻_SPG)

$${{\text{PO}}_4^{3-}}_{\text{SPG}}=\frac{{{\text{PO}}_4^{3-}}_{\text{UP}}{V}_{\text{UP}}+{{\text{PO}}_4^{3-}}_{\text{STG}}{V}_{\text{STG}}}{V_{\text{UP}}+V_{\text{STG}}+(P-E)+R}$$

using the volume transport terms and box definitions as described above and substituting salinity for phosphate concentrations in the subtropics (PO₄³⁻_STG) and in upwelled waters (PO₄³⁻_UP). Note that this formulation considers only phosphate supply by ocean advection and does not attempt to model removal by productivity. As a result, our modeled phosphate concentrations are equivalent to winter values before significant drawdown in the spring bloom.

We use Emile-Geay’s (3) estimates of advective transport from the subtropical gyre and upwelling, Wills and Schneider’s (148) estimates of P-E and R (taken from ERA-Interim reanalysis), and salinity and phosphate concentrations for subtropical and upwelled waters from World Ocean Atlas, computed using the box averaging tool in Ocean Data View (149). Following Warren (2), we consider only advection and do not attempt to account for the influence of diffusion or eddy flux, which have large uncertainties for salinity and are poorly constrained for phosphate. We checked this assumption with an estimate of phosphate eddy flux, based on the eddy transport terms determined by Emile-Geay (3) for salinity and using regional phosphate gradients. This made little difference (<5%) to our results, as the resulting values were small, and the supply of phosphate by vertical eddy flux is countered by the influence of nutrient-poor water from horizontal eddy flux. We use Warren’s approach (2) to obtain subtropical and upwelling salinity and phosphate values with minimal influence from eddy exchange with water from the subpolar gyre, by excluding values immediately adjacent to the subpolar gyre. Our subtropical surface values use a box spanning 24°N to 40°N, 135°E to 115°W, and depths 0 to 200 m, and our upwelling values are taken from 2000 m, with the same latitude and longitude bounds as the surface subpolar box. All model input values are given in table S2.

The model accurately reproduces the salinity of the modern North Pacific subpolar gyre (33.0). The model gives phosphate in the subpolar gyre of 2.0 μmol/kg, slightly higher than annual average values of 1.8 μmol/kg in the modern North Pacific, but matching February (prespring bloom) values of 2.0 μmol/kg. The influence of possible glacial conditions is shown in fig. S5 and table S1. Subpolar salinity increases when (i) net precipitation is reduced, (ii) upwelling is increased, and (iii) exchange with the subtropics is enhanced, particularly when subtropical salinity is elevated. Phosphate concentrations in the subpolar gyre are reduced by exchange with the subtropics and by decreasing the phosphate concentrations of upwelled waters and are increased by increasing the upwelling flux. However, the increase in the upwelling flux has less impact on surface phosphate than the reduction of the phosphate concentration of the upwelled water; indeed, the net effect of doubling vertical exchange while halving subsurface nutrients is similar to that of halving subsurface nutrients alone (fig. S5). Because the wind- and tidally driven transfer of water from the subsurface to the surface North Pacific is high, vertical nutrient gradients are relatively low, so further increase in vertical exchange (e.g., with reduced stratification) has less influence than decreasing the nutrient content of the upwelled water, as a result further substantiated by Earth system modeling (see below).

cGENIE Earth system model

To explore the impact of enhanced North Pacific overturning on nutrients and tracers in more detail, we ran a series of sensitivity tests with the “muffin” release of the cGENIE Earth system model (150). cGENIE consists of a three-dimensional frictional geostrophic ocean circulation model, two-dimensional sea ice and energy-moisture balance atmospheric models and incorporates representations of the marine geochemical cycling of carbon and other biologically mediated tracers (151). We use cGENIE with a 36 × 36 equal-area horizontal resolution and 16 vertical levels, with highest vertical resolution toward the ocean surface. Ocean physics and carbon cycle configuration are as described and evaluated in (150), with the addition of iron colimitation on marine productivity (152). Despite its coarse resolution, cGENIE successfully reproduces many of the large-scale features of marine biogeochemical cycles in the present-day (150, 151). Of particular relevance to this study, cGENIE predicts nutrient (PO₄³⁻) distributions in the North Pacific that are in agreement with present-day observations (151). The coarse resolution also allows cGENIE to be run efficiently and a broad parameter space to be explored: here, we show 28 × 5000–year runs for two sets of boundary conditions.

To test the potential influence of background climate state on our results, we ran experiments under preindustrial and glacial boundary conditions. To simulate a glacial climate state, we reduced the radiative forcing consistent with concentrations of the major greenhouse gases (CO₂, CH₄, and N₂O of 191 ppm, 350 parts per billion (ppb), and 216 ppb, respectively; 153–155). In addition, we increased the zonally averaged planetary albedo profile in the Northern Hemisphere and increased average ocean salinity by ~1 PSU, similar to that used in (12). Our experiments start from the end of 20,000-year equilibrium spin-ups carried out with prescribed atmospheric CO₂ of 278 ppm, δ¹³C of −6.50‰, and Δ¹⁴C of 0‰. The same prescribed atmospheric CO₂ concentration and isotopic compositions are used for both preindustrial and glacial boundary conditions, so that the size of the ocean-atmosphere carbon inventory is similar across our experiments. Radiative forcing by CO₂ is set independently of the actual concentration of atmospheric CO₂ calculated by the ocean-atmosphere biogeochemistry modules and is held constant at either preindustrial or glacial levels as described above.

To simulate an increase in North Pacific ventilation, we reduced the prescribed transfer of atmospheric fresh water from the Atlantic to the Pacific, which implicitly accounts for the net moisture transport between these basins (156, 157). We ran 28 total experiments under preindustrial and glacial boundary conditions, with a range of freshwater transport anomalies from 0.04 to −0.38 Sv. Experiments were run for 5000 years, which allows ocean circulation and nutrient concentrations to stabilize, with <2% change in mean global surface PO₄³⁻ over the last 500 years of the simulations.

The decrease in atmospheric freshwater transport from the Atlantic to the Pacific increases PMOC and reduces AMOC. The redistribution of surface salinity also changes deep-water formation in the Southern Ocean, in turn changing biological pump efficiency. Each of these processes may change atmospheric CO₂, and deconvolving their individual influences is beyond the scope of this study. To place broad constraints on changes in pCO₂ due to enhanced PMOC, we calculate changes in the ocean’s preformed phosphate inventory resulting from increased input of NPIW and use the theoretical scaling of Ito and Follows (33) to estimate the pCO₂ impact. We use a dye tracer to calculate the volumetric contribution of North Pacific waters to the ocean’s interior and use the end-member preformed phosphate composition of the grid box with the maximum contribution of NPIW. We assume that this NPIW replaces water in the Pacific’s interior that has a preformed phosphate of 1.4 µmol/kg (158) and calculate the resulting difference in preformed phosphate inventory by mass balance. The results of this mass balance calculation are shown in fig. S14, with contours showing the general relationship between NPIW volume, preformed phosphate, and pCO₂ change, and symbols showing the realizations found in our cGENIE simulations. With enhanced PMOC, the preformed phosphate content of NPIW decreases, as nutrients are flushed from the subpolar gyre, and this low-phosphate water occupies a greater ocean volume, thus increasing the net efficiency of the biological pump and decreasing atmospheric CO₂. To estimate the cGENIE scenario most representative of LGM conditions in the North Pacific, we compare modeled δ¹³C profiles in the NW Pacific to the data (Fig. 5 and fig. S10). An overturning rate of 8 Sv, found with an Atlantic to Pacific freshwater forcing anomaly of −0.19 Sv, best matches the shape of the LGM δ¹³C profile (Fig. 5) and also shows a good match to observed changes in δ¹³C, redox, and ¹⁴C across the basin (fig. S10). We thus estimate a change in pCO₂ of ~16 ppm as a result of enhanced PMOC at the LGM. However, we note that the influence of changes in ocean circulation and preformed nutrient inventory on pCO₂ may vary depending on circulation regime in the Southern Ocean (34), and that other processes, including sea ice and carbonate compensation, will also influence CO₂ but are not captured by this treatment. Further work is required to constrain the impact of PMOC on pCO₂ in more detail.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/50/eabd1654/DC1