Global oceanic oxygenation controlled by the Southern Ocean through the last deglaciation

Abstract

Ocean dissolved oxygen (DO) can provide insights on how the marine carbon cycle affects global climate change. However, the net global DO change and the controlling mechanisms remain uncertain through the last deglaciation. Here, we present a globally integrated DO reconstruction using thallium isotopes, corroborating lower global DO during the Last Glacial Maximum [19 to 23 thousand years before the present (ka B.P.)] relative to the Holocene. During the deglaciation, we reveal reoxygenation in the Heinrich Stadial 1 (~14.7 to 18 ka B.P.) and the Younger Dryas (11.7 to 12.9 ka B.P.), with deoxygenation during the Bølling-Allerød (12.9 to 14.7 ka B.P.). The deglacial DO changes were decoupled from North Atlantic Deep Water formation rates and imply that Southern Ocean ventilation controlled ocean oxygen. The coherence between global DO and atmospheric CO₂ on millennial timescales highlights the Southern Ocean’s role in deglacial atmospheric CO₂ rise.

Millennial variability of global ocean oxygen through the last deglaciation is likely controlled by Southern Ocean processes.

INTRODUCTION

Oceanic dissolved oxygen (DO) is supplied from air-sea exchange and photosynthesis in the photic zone, and it is consumed by organic carbon respiration throughout the water column. Because respired carbon accumulation and oxygen consumption are stoichiometrically linked during organic carbon degradation, DO reconstructions are particularly valuable for understanding marine respired carbon storage (1), which plays a major role in the glacial-interglacial variability of the partial pressure of atmospheric CO₂ (pCO₂) (1).

The most recent compilation of available DO proxy records (e.g., laminations, foraminiferal assemblages, redox-sensitive metals, bulk sedimentary nitrogen isotopes, and carbon isotopic composition records) captures this inverse relationship between marine carbon storage and DO (2–4). However, all presently available DO reconstructions are influenced by local variability in ventilation (e.g., changes in upwelling intensity) and/or export of organic matter to the sediment (5), making it challenging to determine the average global state of oceanic oxygenation. For instance, increased DO has been observed during the Bølling-Allerød [B-A; 12.9 to 14.7 thousand years before the present (ka B.P.); also known as the Antarctic Cold Reversal (ACR)] in the deep Atlantic (>2000 m) as a result of the recovery of Atlantic Meridional Overturning Circulation (AMOC) (2, 6), whereas the upper (<1500 m) Indo-Pacific experienced DO loss in the same interval (2). Spatially heterogeneous DO reconstructions thus complicate extrapolation from local to globally integrated oxygenation responses and how these relate to marine carbon storage and global climate change.

The seawater thallium isotopic composition (ε²⁰⁵Tl, normalized ²⁰⁵Tl/²⁰³Tl in parts per 10,000) is a promising qualitative indicator of the global DO content that can record transient oxygenation responses on multi-millennial (several thousand years) timescales (7, 8). The modern seawater thallium isotopic composition is homogenous at ε²⁰⁵Tl = −6.0 ± 0.3 (SD) due to its long residence time (~22 ka) compared to the deep ocean overturning time (7). Because of the relatively invariant isotopic composition (ε²⁰⁵Tl of −2) of oceanic Tl sources, seawater ε²⁰⁵Tl is mainly modulated by the oceanic Tl sinks on glacial-interglacial timescales, primarily through oxidative sorption of Tl onto manganese oxides that preferentially removes ²⁰⁵Tl relative to ²⁰³Tl from the ocean (9, 10) (Supplementary Materials). Manganese oxide burial is directly linked to oceanic oxygenation because manganese oxides require ambient oxygen to form (7). As ocean oxygen decreases and low-oxygen waters expand, reductive dissolution of manganese oxide in the water columns and/or sediments would increase the dissolved Mn(II) reservoir in the ocean and decrease ²⁰⁵Tl scavenged via Mn oxide burial, resulting in heavier seawater Tl isotopic compositions (higher ε²⁰⁵Tl values) (Supplementary Materials). Seawater ε²⁰⁵Tl variations are recorded in sediments deposited with reducing porewaters (11, 12), where complete removal of Tl from the ambient waters (likely by iron sulfides) leads to zero net isotopic fractionation (12). Oxygen minimum zone (OMZ) sediments (where reducing porewaters prevail) are therefore potential archives for past seawater Tl isotopic compositions and global oceanic oxygenation.

Here, we report the first globally integrated DO content reconstruction at high resolution (150 to 500 years) from the Last Glacial Maximum (LGM; 19 to 23 ka B.P.) to today using seawater thallium isotope reconstructions from the Arabian Sea OMZ (TN041-8PG and TN041-8JPC, 17°48.76′N, 57°30.34′E, 761-m water depth). The Arabian Sea OMZ intensity is closely linked with export productivity associated with monsoon-driven upwelling (13) and ventilation by intermediate water masses (e.g., Antarctic Intermediate Water in glacial times) (14). The cores TN041-8PG/JPC were deposited under persistent low-oxygen environments in the past ~30 ka, as evidenced by redox-sensitive trace metals and benthic foraminiferal surface porosity (fig. S1) (15). Fidelity of the authigenic sedimentary ε²⁰⁵Tl record with contemporaneous seawater values was confirmed through the decision tree constructed from a recent global core top calibration (Materials and Methods) (12). Authigenic sedimentary ε²⁰⁵Tl values also show decoupled changes from the local DO variations (fig. S1) (15), corroborating that the authigenic ε²⁰⁵Tl record is not controlled by local factors (e.g., ventilation or export productivity). Our results reveal lower global DO during the LGM compared to the Holocene (0 to 11.7 ka B.P.), as well as systematic oxygenation changes in response to millennial events across the last deglaciation. Most of the global DO rise from the LGM to the Holocene occurred in the Heinrich Stadial 1 (HS1) and the Younger Dryas (YD), but a brief period of deoxygenation characterized the B-A/ACR.

RESULTS AND DISCUSSIONS

Lower global ocean oxygen content in the LGM compared to the Holocene

In the Holocene, the authigenic ε²⁰⁵Tl values were relatively stable [ε²⁰⁵Tl = −6.1 ± 0.4 (SD), n = 31] and indistinguishable from the modern seawater value of ε²⁰⁵Tl = −6.0 ± 0.3 (SD), suggesting that global ocean DO has remained relatively constant through the Holocene (Fig. 1). At ~3.5 ka B.P., authigenic ε²⁰⁵Tl values briefly decreased to ε²⁰⁵Tl ~ −6.5, which could imply a short period of higher oceanic oxygenation. The exact cause of this higher DO interval remains to be investigated. In contrast, ε²⁰⁵Tl values indicate persistently lower global DO content in the last glacial period (18 to 32 ka B.P.), with an average ε²⁰⁵Tl value of −5.1 ± 0.3 (SD, n = 30). If we apply a steady-state mass balance model (Supplementary Materials), then the mean ε²⁰⁵Tl value during the last glacial period would correspond to an 18 ± 6% (SD based on analytical uncertainty of ε²⁰⁵Tl; Supplementary Materials) lower Mn oxide burial flux compared to the Holocene, possibly related to concurrent changes in dissolved Mn(II) of seawater. For example, lower glacial DO could have led to higher dissolved Mn(II) in the water columns, followed by reoxidation to Mn oxide under increased oceanic DO in the Holocene (1, 16). A smaller global oceanic DO reservoir during the LGM occurred despite the generally more oxygenated conditions in the upper ocean (<1500 m) at this time (Fig. 2) (2, 17). Therefore, the reduction of DO content in the deep ocean must have exceeded the DO rise in the upper ocean (2) during the LGM to drive a net decline of the global DO content, implying that the oceanic DO reservoir is primarily controlled by the deep ocean. This inference is consistent with the fact that ~63% of the ocean volume (and ~85% of the seafloor) resides in waters deeper than 1500 m (Supplementary Materials).

Fig. 1. Global oceanic oxygenation reconstructions compared with global climate and ocean circulation records.

(A) Atmospheric CO₂ concentrations from the Antarctic composite ice core records (54). ppm, parts per million. (B) authigenic sedimentary ε²⁰⁵Tl record from the core TN041-8PG/8JPC with the LOESS fit (Materials and Methods) shown in the solid red curve and the bootstrapped 2-SD envelope shown in the dashed red curve. Error bars are in 2 SD. When the measurement 2 SD is smaller than 0.3, the error bar was set to 0.3 (the long-term reproducibility of ε²⁰⁵Tl analyses, see Materials and Methods). ¹⁴C dates were shown in the yellow triangles. (C) Opal fluxes of TN057-13PC as an upwelling proxy in the Southern Ocean (28). (D) ²³¹Pa/²³⁰Th activity ratios from the Bermuda Rise as an indicator of AMOC strength from the core OCE326 GGC-5 (green) (43) and ODP Site 1063 (blue) (44). The error bar represents 2 SE. The production ratio of ²³¹Pa/²³⁰Th in the water columns is shown in the horizontal dashed line. (E) Radiocarbon age offsets between benthic foraminifera and the atmosphere (B-Atm) as a proxy for subsurface water mass ventilation from the Indian Ocean core SS172/4040 (42). The error bar is 2 SD. (F) The relative deviation (δ¹⁴R) between the deep water and the atmospheric Δ¹⁴C as an indicator for deep water ventilation from the Pacific Ocean at core MD97-2106 (47). Error bars are 1 SD. The HS1 and YD are shaded in light blue, whereas B-A/ACR is shaded in light red.

Fig. 2. Compilation of qualitative oxygenation changes between the LGM and the Holocene from localized redox proxies.

(A) The upper ocean sites (water depth ≤ 1500 m) are denoted by solid circles, whereas the deep ocean sites (water depth > 1500 m) are represented by hollow squares. Lower, higher, and ambiguous changes in bottom water oxygenation during LGM compared to the Holocene are shown in green, blue, and gray symbols, respectively. Whenever there is an overlap, blue overlies green, which overlies the gray symbols. The studied site (TN041-8PG/8JPC) is shown in the yellow star. Base map was generated using Ocean Data View (75). (B) The number of compiled sites that show lower and higher oxygenation in LGM relative to the Holocene binned by the water depth. The bottom green arrow shows the globally integrated oceanic DO content change indicated by reconstructed seawater thallium isotopic composition change.

Our record provides the first global picture of lower oceanic DO content during the LGM relative to the Holocene, and it is consistent with previously published records of, e.g., local oxygenation reconstructions in the deep ocean (2, 3, 18), ventilation records from radiocarbon (19, 20), and intermediate-complexity model simulations (21, 22). The global DO content can be modulated by oxygen solubility (controlled by oceanic temperature), export productivity from the photic zone that consumes oxygen in subsurface waters under respiration (i.e., the soft tissue pump), and ventilation (e.g., air-sea gas exchange of upwelled waters and subsurface water formation rates). Because the global DO content is controlled by the deep ocean (Fig. 2), we focus on the forcing mechanisms that may have dominated the deep ocean oxygen variability. The observed lower DO content during the LGM is opposite to the higher oxygen solubility in seawater at lower temperatures (23), ruling solubility out as a driver of oceanic DO changes from LGM to Holocene. The contribution of LGM export productivity to lowering deep ocean oxygen remains uncertain. Higher LGM export productivity has been observed in the Subantarctic Zone associated with enhanced dust input (fig. S4) (24), which could have drawn down ocean oxygen via respiration. However, lower export productivity in LGM has been found elsewhere (25–29). In addition, the difference between the planktic and benthic foraminifera stable carbon isotopic compositions (δ¹³C) in the LGM was recently shown to be similar to the modern values in the deep Pacific Ocean, suggesting that respiration of export productivity alone may not fully explain the observed reduced deep ocean oxygen (30). Deep ocean ventilation is primarily controlled by the strength of deep water formation in the North Atlantic [via North Atlantic Deep Water (NADW) formation] and Southern Ocean [via Antarctica Bottom Water (AABW) formation]. With glacial NADW potentially limited to the upper 2000 m, the LGM deep ocean was primarily occupied by AABW (31, 32), which forms along the Antarctica margin following poleward transport of the upwelled NADW in the Antarctic Zone. Less ventilated AABW is thus likely needed to drive a net reduction of deep ocean DO during the LGM. Glacial AABW was more isolated due to extensive sea ice (limiting air-sea gas exchange) and weakened upwelling (22, 28, 33, 34), and such isolation also allows continued oxygen consumption in the deep ocean. In addition, the remineralization depth could have been deeper due to slower organic carbon respiration rates under cooler LGM temperatures, which could have allowed additional export carbon sinking into the deep ocean (35–37). Despite the lack of observational evidence, this mechanism may further reduce the deep ocean oxygen content even without a substantial change of export flux or ocean circulation.

Millennial global ocean oxygen variability during the deglaciation

During the deglaciation, the seawater ε²⁰⁵Tl record reveals millennial variations of global oceanic oxygenation and Mn oxide burial in the ocean, which could have led to non–steady-state Tl isotope mass balance. The transient seawater ε²⁰⁵Tl changes can be simulated using a non–steady-state box model by perturbing Mn oxide fluxes (Supplementary Materials and fig. S5). Rapid deoxygenation in the global ocean in the B-A/ACR (Fig. 1) may have led to enhanced reductive dissolution of Mn oxides and Mn(II) supply from organic carbon respiration (38). An increased global deep ocean oxygen content relative to the LGM is evidenced by shifts toward more negative seawater ε²⁰⁵Tl values during the cold intervals in the Northern Hemisphere (HS1 and YD), which could have also triggered excessive Mn oxide burial fluxes due to an elevated Mn(II) reservoir in the low-oxygen LGM and B-A/ACR, respectively (Supplementary Materials). These distinct intervals of deep ocean DO changes are likely not related to temperature-driven solubility [as mean ocean temperature closely follows atmospheric pCO₂; (23) and fig. S1] or export productivity in the Antarctic Zone (Fig. 1) (28). As the subantarctic-sourced water primarily ventilates the upper ocean [<2000-m water depth; (39)], reduced export productivity (under declined dust input) and less complete nutrient consumption compared to the LGM (24, 40) also suggest that the Subantarctic Zone may not be a major contributor to the deep ocean oxygen content (fig. S4).

Deep ocean ventilation changes via NADW and/or AABW formation likely served as the primary driver of the observed deglacial DO variability. Strengthened NADW formation that supplied oxygen-rich surface water to the deep ocean resulted in a better ventilated deep Atlantic (2, 19, 41) and, possibly, Indian Ocean (42) during the B-A/ACR (Fig. 1E). A more oxygenated Atlantic, however, contrasts with our reconstructed global ocean oxygen content at the same time. We note that global DO variability from Tl isotopes through the deglaciation is anti-phased with NADW formation rates (Fig. 1) reconstructed from ²³¹Pa/²³⁰Th (43–45), arguing against an NADW ventilation control on the global deep ocean oxygenation. Despite recent studies debating whether NADW formation completely halted during the HS1 (46), a consensus remains that NADW formation was weaker during the deglacial stadial periods relative to interstadial periods. Because our data do not align with that framework, they instead point to AABW ventilation as the most probable means of modulating the global DO content during the deglaciation. This conclusion is consistent with a radiocarbon reconstruction in the deep southwest Pacific, which shows development of poorly ventilated water masses starting from the early B-A/ACR but increasingly ventilated conditions in the HS1 and YD (Fig. 1F) (47). Redox-sensitive trace metal (e.g., authigenic uranium and manganese) records from Atlantic sectors of the deep Southern Ocean also show similar oxygenation in YD and HS1 (1).

Deep Southern Ocean ventilation is likely related to migration and strength of Southern Hemisphere westerlies (48, 49). For example, southward displacement and intensification of westerlies during HS1 and YD would have increased upwelling and abyssal mixing in the Southern Ocean (50), leading to a less stratified water column and new AABW formation (48). A southward displacement of the westerlies would also have contributed to Antarctic Zone warming via southward eddy heat transport (49, 51). Subsequent reduction of Antarctica sea ice cover would have promoted air-sea gas exchange (52) and additional AABW ventilation (53). Conversely, during the B-A or ACR, expanded sea ice and northward migration of westerlies would have triggered the opposite oceanic DO response.

Coherence between reconstructed global ocean oxygen and atmospheric CO₂

The high-resolution ε²⁰⁵Tl reconstruction of the global DO content also shows notable coherence with atmospheric CO₂ concentrations (Supplementary Materials and Figs. 1 and 3; r = −0.74, P < 0.01), corroborating the ocean’s role in modulating pCO₂ variability throughout the last glacial period (at least to 32 ka B.P.). A smaller oceanic DO reservoir in the LGM suggested by the less negative ε²⁰⁵Tl values corresponds to lower pCO₂, whereas deglacial ε²⁰⁵Tl variability on millennial timescales displays nearly synchronous variability with the two-step pCO₂ rise (54). Transitions toward a higher global DO content correspond to pCO₂ increase during 15.5 to 17.5 ka B.P. (HS1) and 12.9 to 11.7 ka B.P. (YD), whereas the pCO₂ plateau during B-A/ACR was contemporaneous with the lowest global DO content in the deglaciation, suggesting a pause in CO₂ release from the ocean (55). The Southern Ocean control on the global DO reservoir would then imply that CO₂ outgassing from the Southern Ocean rather than the Atlantic Ocean likely dominated the oceanic contribution to the deglacial pCO₂ changes. As NADW formation recovered following the HS1 and the deep Atlantic became more ventilated in the B-A/ACR, a Southern Ocean control on deglacial ocean oxygenation thus also provides a plausible explanation for stagnant atmospheric pCO₂ at this time (19). However, NADW formation may still play an important role in regulating ocean carbon storage. For instance, NADW formation during LGM could have transported CO₂ into the deep ocean through entrainment into AABW (56). During the last deglaciation, abrupt changes of NADW formation rates may have triggered more rapid changes (e.g., sub-millennial) in atmospheric pCO₂ that cannot be resolved in our record (57, 58). We also caution that our global ocean oxygen reconstruction cannot constrain other controls of ocean carbon storage (e.g., carbonate pump and air-sea gas disequilibrium) that could have contributed to the deglacial pCO₂ change synergistically.

Fig. 3. Pearson correlation between atmospheric pCO₂ and authigenic Tl isotopic compositions.

The atmospheric pCO2 data are from (54). The Pearson correlation is statistically significant (P < 0.01). The LGM, HS1, B-A/ACR, YD, and the Holocene are denoted by green, purple, yellow, red, and blue circles, respectively.

Implications for future ocean oxygenation

Predicting future oceanic oxygenation under anthropogenic climate change remains challenging, but sedimentary archives may provide insights for future model development on how physical (e.g., ocean mixing and sea ice formation) and biogeochemical processes (e.g., productivity and respiration) will respond. Observations and model simulations suggest poleward migration and intensification of southern westerlies in recent decades and, possibly, in the future, which may lead to a more ventilated Southern Ocean (50, 59). However, observed freshening of AABW since the 1990s due to Antarctic meltwater could reduce deep ocean oxygenation (30, 60). Whether the deep Southern Ocean will become more oxygenated in response to the ongoing climate change is thus still under debate. Our reconstruction on the globally integrated ocean oxygen variability would then be consistent with recent model predictions implying a more ventilated Southern Ocean on centennial to millennial timescales under future warming (61), which may lead to higher global deep ocean oxygen content as ventilation outpaces ocean warming and stratification.

MATERIALS AND METHODS

Materials and age models

The cores TN041-8PG/8JPC (17°48.76′N, 57°30.34′E, 761-m water depth) were retrieved from the Oman margin through the US Joint Global Ocean Flux Study Arabian Sea Process Study. The intermediate waters that affect the Arabian Sea OMZ include the saline Red Sea Intermediate Water, Persian Gulf Water, and the southern-sourced Indian Central Water produced by mixing of aged Antarctic Intermediate Water (AAIW) and Indonesian Intermediate Water (62). Persian Gulf Water and Red Sea Intermediate Water are relatively oxygenated due to recent contact with the atmosphere before being exported into the Arabian Sea (63). The Indian Central Water, however, is nutrient-rich and low-DO due to oxygen consumption by organic carbon respiration in AAIW on its path to the Arabian Sea (62). During glacial times with low sea level, water exchange between the Red Sea and Arabian Sea could be greatly reduced, with a stronger AAIW ventilation in the Arabian Sea (14). Previous research has shown coeval responses of stronger AAIW ventilation in the Arabian Sea and the North Atlantic Heinrich events that weakened the thermohaline circulation (14). Bulk sediment samples were taken from the cores at a 4-cm resolution (~150 to 500 years, a total of 90 samples) and were freeze dried. The age model was taken from (15). Briefly, ages were estimated using the δ¹⁸O stratigraphy of Globigerinoides ruber first, and, then, seven accelerator mass spectrometry ¹⁴C dates were generated from an average of 300 to 350 individuals of G. ruber picked from the >250- or > 335-μm-size fractions, at the National Ocean Sciences Accelerator Mass Spectrometry Facility at Woods Hole Oceanographic Institution (WHOI). Planktic ¹⁴C dates were converted into calendar ages using the Marine20 curve (64) with an Arabian Sea regional marine radiocarbon reservoir age correction ΔR = 93 ± 61 years (65), using Bacon (rbacon package) that uses adaptive Markov chain Monte Carlo algorithm and Bayesian statistics (66).

Bulk elemental concentrations and enrichment factors

A fraction of freeze-dried sediments was ground into powder using an agate mortar and pestle for bulk elemental concentration and thallium isotope analyses. Sedimentary U and Al have been published by (15). Here, we summarize the bulk elemental concentration method. Samples were selected at a 16-cm resolution through the core. About 10 mg of bulk freeze-dried samples was digested using a mixture of 1 ml of concentrated nitric acid (HNO₃), 1 ml of hydrofluoric acid (HF), and 0.1 ml of hydrochloric acid (HCl) at 135°C. Aqua regia (3:1 mixture of HCl and HNO₃) was used to remove fluoride precipitates formed during the HF digestion. The dried sample was redissolved in 2% HNO₃ and diluted for analyses on the Thermo iCAP Q inductively coupled plasma mass spectrometer (ICP-MS) at WHOI. Two standard calibration curves were used, including rock [BHVO-2 (basalt) and AGV-2 (andesite), US Geological Survey] and high-purity ICP-MS elemental standards. Indium was used as an internal standard to monitor the instrument drift. MESS-3 (continental margin sediments, National Research Council of Canada) was used as the standard reference material to monitor accuracy, which yielded an external recovery of 96 to 113% for all certified elements. Long-term reproducibility of the elemental concentrations was monitored by the relative SD of repeated measurements of the MESS-3 digested with each batch of samples, which was <10%.

Authigenic Tl isotopic composition measurements

Thallium isotopic compositions [ε ²⁰⁵Tl = 10,000 × (²⁰⁵Tl/²⁰³Tl_sample − ²⁰⁵Tl/²⁰³Tl_{NIST SRM997})/(²⁰⁵Tl/²⁰³Tl_{NIST SRM997})] were determined for the authigenic component of the sediments. The authigenic component was extracted by immersion of approximately 100 mg of ground bulk sediments in 2 M nitric acid at 130°C for 12 to 15 hours, primarily targeting Fe sulfides (main host phase of Tl in the reducing environments) with minimal impacts from the detrital fraction (11, 12, 67). Authigenic fractions were then separated by pipetting out the supernatant following centrifugation. The supernatant was dried down and repeatedly digested with inverse aqua regia (1:3 mixture of HCl and HNO₃) and hydrogen peroxide (H₂O₂) at 130°C to remove organic matter. Last, samples were dissolved in 1 ml of 1 M HCl, and 30 μl of brominated water (Br₂-H₂O) was added to oxidize Tl(I) to Tl(III) for column chemistry (68). The standard reference material SCo-1 (Cody shale, US Geological Survey) was also processed in the same manner to monitor sample reproducibility. One mini-column with 100-μl anion exchange resin (AG 1-X8) was used to achieve efficient Tl separation following the well-established procedures (67–69). The resin was cleaned and preconditioned with 1.5 ml of 0.1 M HCl–5% SO₂, 1.5 ml of 0.1 M HCl, and 0.1 + 0.3 × 3 ml of 0.1 M HCl–1% Br₂-H₂O. Digested samples in 1 M HCl–3% Br₂-H₂O were loaded onto the columns, followed by the addition of 0.1 × 3 + 1.5 ml of solutions of 0.5 M HNO₃–3% Br₂-H₂O, 0.1 + 1.5 ml of 2 M HNO₃–3% Br₂-H₂O, and 0.1 ml of 0.1 M HCl–1% Br₂-H₂O to elute the sample matrix. Thallium was then eluted with 0.1 + 1.5 ml of 0.1 M HCl–5% SO₂ solution. The collected solution was then dried down and dissolved in 0.5 ml of 0.1 M HNO₃ + 0.1% H₂SO₄ solution. All Tl isotope analyses were performed on a Thermo Finnigan Neptune multi-collector ICP-MS at the WHOI Plasma Facility, using standard-sample bracketing and external normalization to Pb (68). Each sample were measured at least twice for ε²⁰⁵Tl and a long-term external reproducibility of ~ ±0.30 ε²⁰⁵Tl units (2 SD) (11, 70). SCo-1 standard measurements yielded a Tl isotopic composition of −2.83 ± 0.10 (SD, n = 2), which is identical to previously published values (7).

Authigenic sedimentary ε²⁰⁵Tl for the cores TN041-8PG/8JPC

Recent work has shown that quantitative Tl removal by sulfides can occur in the ambient sediment porewaters under manganous (no oxygen with Mn reduction) conditions (71). If quantitative Tl removal occurs at/near the sediment interface, then authigenic sedimentary ε²⁰⁵Tl faithfully preserves the overlying seawater ε²⁰⁵Tl value (12). A decision tree based on modern sediment core top data has recently been developed to test fidelity of sediment archives in preserving the seawater ε²⁰⁵Tl value (12), where Mn, U, and Ba enrichment factors (Mn_EF, U_EF, and Ba_EF) with respect to the upper continental crust (72) are used as input variables. The elemental enrichment factors of the core TN041-8PG/8JPC are calculated using the following equation

$${\text{Element}}_{\text{EF}}=\frac{{\left(\frac{\text{Element}}{\text{Al}}\right)}_{\text{sample}}}{{\left(\frac{\text{Element}}{\text{Al}}\right)}_{\text{UCC}}}$$

The calculated Mn_EF, U_EF, and Ba_EF (table S2) were passed through the decision tree and yielded 100% positive output, which implies that the sedimentary ε²⁰⁵Tl values measured in our cores record the overlying water column value. Despite the lower U_EF values during the LGM that indicates better oxygenated conditions [~10 to 15 μmol/kg higher; (15) and fig. S1], positive output from the decision tree implies that porewaters at/near the sediment-water interface were still sufficiently reducing for quantitative Tl removal. The persistent low-oxygen condition through the last LGM is also consistent with foraminiferal I/Ca reconstructions from the same core, which indicate that both LGM and Holocene local bottom water oxygen was <50 μmol/kg (15).

Postdepositional Mn remobilization is unlikely to affect authigenic ɛ²⁰⁵Tl in the studied core. If Mn oxide preservation had occurred at/near the sediment-water interface due to higher bottom water oxygen during the LGM, then preferential uptake of ²⁰⁵Tl upon sorption onto Mn oxides would have led to much heavier sedimentary ε²⁰⁵Tl [>−2, the marine Tl input value (7)] as has been observed in the Toarcian Peniche section (73). However, persistent U enrichment throughout the cores (fig. S1) suggests that porewaters were always reducing through the LGM and argues against preservation of Mn oxides at any time. Therefore, we conclude that the authigenic sedimentary ε²⁰⁵Tl should reflect the open ocean seawater throughout the record.

Statistical analyses of authigenic ε²⁰⁵Tl

The mean and SD values of authigenic sedimentary ε²⁰⁵Tl during LGM and the Holocene were obtained by bootstrapping the datasets 10,000 times in MATLAB. The LGM period in the core based on ε²⁰⁵Tl variations (18 to 32 ka B.P., before the HS1) was compared with the Holocene (0 to 11.7 ka B.P., after the YD) ε²⁰⁵Tl values. The two intervals show statistical mean values that are significantly different (fig. S2). The presented authigenic ε²⁰⁵Tl record has also been smoothed by a nonparametric regression [locally estimated scatterplot smoothing (LOESS) function]. The robust LOESS algorithm was used, which performs local regressions with weighted linear least squares and a second-degree polynomial model that assigns lower weights on the outliers. The span (i.e., fraction of the data points used for local regression) of LOESS fit was optimized using cross-validation (74). A total of 99 fractions (1 to 99% of the data used for LOESS regression) were tested to minimize the final residual error of the LOESS fit. For each fraction, the cross-validation process divided the original dataset into 10 splits, with each realization using 9 splits as the training data for obtaining the LOESS fit and the remaining 1 split as the test data to determine the residual. The optimized fraction was selected on the basis of the smallest cross-validated residual, which yielded a span of 14%. The best LOESS fit curve and the 2-SD confidence intervals were then obtained through 10,000 realizations generated by the bootstrapping approach.

To determine the correlation between atmospheric pCO₂ (54) and authigenic Tl isotopic composition over the past 32,000 years, the much higher resolution ice core pCO₂ composite record was first interpolated onto the ages of authigenic ε²⁰⁵Tl record for comparison. A Pearson correlation coefficient was then computed between the interpolated pCO₂ and the LOESS fit of authigenic ε²⁰⁵Tl, and a P value was used to determine the significance (Fig. 3). Cross-correlation of the two time series was also determined after adding 400-year time lags to one of the records (median temporal resolution in the authigenic Tl isotopic composition record) in MATLAB (fig. S3). These analyses reveal that the statistically most significant correlation between pCO₂ and ε²⁰⁵Tl is observed when no time lag between the two time series is applied. Such strong correlation also suggests that pCO₂ changes during the last deglaciation were closely linked with ocean ventilation that simultaneously also exerted control over the global ocean oxygen budget.

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S5

Legends for tables S1 and S2

Table S3

References

Other Supplementary Material for this manuscript includes the following:

Tables S1 and S2