Since the 1960s, dissolved oxygen concentrations in the ocean have declined by 2%, equating to a loss of approximately 6 Pmol of the ocean’s total oxygen content, estimated at 227 Pmol 1,7–10. In the shallow and intermediate depths (0-1200 m water depth) of the world’s oceans, this change can be largely (50–100% depending on depth) attributed to warming temperatures, which reduce the solubility of oxygen in the water column1. The expansion of so-called “oxygen minimum zones” threatens highly productive ecosystems on which food security and industry with value in the hundreds of billions of dollars (USD) rely on 9,11,12. However, the decline in oxygen in shallow and intermediate depths is a relatively small part (~ 27%) of the total loss1. The vast majority of oxygen loss (~ 73%) has taken place in the deep ocean1. This issue has received less attention than coastal and shallow waters, despite the pivotal role that this enormous reservoir of oxygen (165 Pmol) plays in the global carbon cycle 2,6,13. In the deep ocean, very little (~ 2%) of the oxygen loss can be attributed to temperature change and solubility1,7. Loss of this magnitude requires a change in deep ocean circulation and water mass production.
The formation of deep water masses exerts a strong influence on the geographic distribution of oxygen in the deep ocean (Fig. 1A). In polar regions, oxygenated water is mixed the downward due to cooling, brine rejection, and buoyancy forcing. The movement of this well-ventilated water through the deep is described by the Atlantic Meridional Overturning Circulation (AMOC) (Fig. 1A). Today, North Atlantic Deep Water (NADW), occupies 21% of the global ocean volume. It forms in the Labrador, Greenland, Icelandic, and Norwegian Seas and flows southward through the Atlantic Basin at a rate of ~ 15 Sverdrups (1 Sv = 106 m3 s-1) 14–17. Antarctic Bottom Water (AABW) is the deep oceans’ densest and most voluminous water mass, occupying 36% of the global ocean15. It is formed in the Weddell Sea and other coastal regions of Antarctica and flows northward away from continent at a rate of 12 Sv18,19. NADW and AABW having recently been in contact with the atmosphere, have relatively high oxygen content (Fig. 1A). Oxygen content declines with ventilation age as deep dwelling organisms use it to respire carbon (Fig. 1B)20.
Simulations have been used to suggest that the supply of oxygen to the deep ocean is threatened by declining deep water production in both the North Atlantic and the Antarctic regions3. Atmospheric warming is causing the addition of heat as well as fresh water from glaciers to the surface ocean, increasing its buoyancy5,21. Because deep water mass production is forced by densification, this polar warming is expected to feed-back negatively on ocean ventilation. Technical, logistical, and practical challenges with measurements combined with observations of a limited duration have made verifying the predictions made by simulations a challenge. Our best direct assessments of AMOC, have come from mooring arrays such as RAPID-MOCHA. The measurements showed a substantial decline in AMOC transport between 2004 and 20134,22–24. Since 2013, AMOC transport has stabilized and may be increasing25. No such moorings exist for Antarctic waters, but the salinity, density, and volume of AABW have declined over the past 50 years26–29.
Natural processes in polar regions have reduced the ventilation of deep water in Earth’s past, especially during glacial episodes of the Pleistocene epoch. It is proposed that the expansion of sea ice in deep water-forming regions reduces the capacity of the surface ocean to exchange oxygen and carbon dioxide with the atmosphere30–32. Also, paleo-reconstructions that leverage chemical properties of water masses (e.g., neodymium and carbon isotopes) show a substantially different glacial configuration of deep ocean ventilation, with a shallower glacial NCW and the vast expansion of nutrient-rich, oxygen-poor deep water sourced from the Southern Ocean, Southern Component Water (SCW, the glacial equivalent of AABW)33–35.
Reconstructions of deep water oxygen variability in Earth’s past are useful for understanding influences on ocean oxygen in the modern climate system. Methods of qualitatively reconstructing bottom water oxygen in the past include: 1) documenting the appearance and disappearance of various oxygen-tolerant and intolerant benthic species in the fossil record36,37 and 2) measuring the concentration of redox sensitive trace metals (Mn, Fe, U) bound to sediment particles by anthogenesis during burial38,39. There are two well-developed methods to quantitatively reconstruct deep ocean oxygen. The difference between epifaunal and infaunal δ13C ratios acts as a proxy for oxygen concentrations in bottom water because oxygen concentrations drive nutrient gradients in sediment pore waters with depth40,41. Another method leverages the inclusion of iodate into foraminiferal tests in oxidizing conditions42. Each of the four methods described above has its proxy-specific limitations. What they all share, however, is a laborious sample preparation and analytical process, which limits the resolution and breadth of proxy records in space and time.
We propose a new proxy method capable of qualitatively reconstructing bottom water oxygen fluctuations, that relies on the recurrence of distinct dark-colored, green, horizontal, mm scale banding in marine sediment (Extended Data Fig. 1). Color has long been recognized as an indicator of redox conditions43 with red being associated with oxidization, and green (or gray) associated with reducing conditions. The green color of these bands suggests an origin related to redox processes. But while these bands are more or less ubiquitous in sediments from across the world and geologic time, explanations for their occurrence vary substantially. The green bands have been proposed to be connected to glauconite formation44, volcanic eruptions45–48, terrestrial hydroclimate events49, sedimentation rate changes50,51, and early diagenesis52–57.
Our preferred hypothesis for their formation relies on early work on sediment pore water chemistry and diagenesis58, as well as a detailed magnetic survey of tropical Atlantic sediment cores detailed in Funk et al.56. Berner59 first described how the deposition of a thin organic rich layer during an anoxic event could lead to the accumulation of trace metals in or near that layer, especially iron hosted in pyrite. Later, Froelich58 described the typical cascading series of diagenetic reactions in pore waters, where organic material acts as a key catalyst and donor of electrons in the reduction of nitrate, manganese, iron, and sulfate. Wilson52,53 then described how the emplacement of an organic rich layer in sediment could disrupt the typical reaction series, producing steep redox gradients above it that are capable of immobilizing dissolved Fe2+ from porewaters below and concentrating it into a thin layer.
The best-known example of this phenomenon is the geochemical and color anomalies above organic-rich sapropel layers from the Mediterranean that have been "burned down" by oxic bottom waters and indicate their original thickness54,55. Indeed, there are extended intervals of Mediterranean sediment where the rhythmic reappearance of the sapropels is replaced with sapropel "ghosts" - color-banded sediment indicating the complete burndown of the organic material in the layer60.
While many bands appear green, red colored bands have also been noted61,62. The specific color of the bands is a secondary feature, thought to originate in the redox state of iron bound to the clay fraction of sediment50,63–65. Lyle66 demonstrated that the color of these marine sediments is rapidly reversible (10 minutes) from red to green in the presence of hydrogen peroxide (H2O2, an oxidant) or sodium hydrosulfite (Na2S2O4, a reductant).
The most thorough survey to date of green bands was conducted on a suite of 17 sediment cores from the Equatorial Atlantic that span the last 500 thousand years (ka)56. In this work, they used rock magnetics to diagnose the diagenetic dissolution of magnetite in dark colored features they called “relic diagenetic fronts” and “organic-rich layers”. They found that the occurrence of the bands was associated with precession maxima, glacial periods and most particularly glacial terminations. They attributed these layers to productivity pulses, concomitant water column deoxygenation, and the subsequent release of organic carbon from the sediment in oxidizing conditions.
While the work of Funk et al.56 promotes productivity pulses, we hypothesize that any event that causes a temporary deoxygenation of deep water can lead to the formation of these bands (Fig. 2). The low oxygen conditions cause increased efficiency of organic matter export to the seafloor sediments (Fig. 2A). A subsequent rejuvenation of oxygen in deep water is then required to respire the buried organic matter, leading to redox reactions that concentrate trace metals into the bands (Fig. 2B). We test this conceptual model with 1) a worldwide survey of the occurrence of diagenetic bands in shallow sediments and investigate the various environmental and geographic features that correspond to their presence and 2) a stratigraphic survey of green band formation in two sediment cores in disparate regions of the Atlantic Ocean basin, a region subject to highly variable deep water oxygen fluctuations across the ice ages cycles of the Pleistocene. The results indicate substantial variability in the oxygen content of the deep ocean across Pleistocene glacial periods and implicate bottom water oxygen as a driver of low frequency variability in the carbon cycle across the last million years.
A color transition from red to greenish gray is a common feature of marine sediment cores from around the world, a function of the depth to which dissolved oxygen in ocean bottom waters can penetrate down into sediments58,66,67. Sometimes, dark colored bands appear near these redox fronts56 (Extended Data Fig. 1). We surveyed sediment core images for these redox fronts marked by diagenetic banding using 2,121 marine sediment cores from the Atlantic, Indian and Pacific Oceans, encompassing available photos from sediment cores containing the sediment-water interface from every site and expedition from the International Ocean Discovery Program (IODP). These images were sourced from IODP’s Laboratory Information Management Systems database, hosted by Texas A&M University. Because this change in sediment color has been demonstrated to be rapidly reversible66, we describe these fronts as “active” in nature, allowing us to test the influence of bottom water oxygen and deep water masses on their geographic distribution. To qualify as a banded redox front, the sediment core needed to have banding located within 5 cm of an oxidative color transition, which we define as a shift from orange, red, or brown sediments (indicating oxygenated sedimentary porewaters), to green or gray sediments (indicating reducing conditions). The results of the survey are in Fig. 1 and Table 1.
Table 1
Active Redox Front Survey Results
|
Atlantic
|
Indian
|
Pacific
|
Arctic
|
Southern
|
Total
|
Sites
|
Banding Present
|
67
|
6
|
7
|
3
|
1
|
84
|
Total Surveyed
|
369
|
132
|
538
|
11
|
115
|
1165
|
Cores
|
Banded Front
|
108
|
10
|
9
|
2
|
4
|
133
|
Total Surveyed
|
673
|
234
|
981
|
20
|
213
|
2121
|
The survey shows that, while banding is a relatively rare phenomenon, the presence of banding is much more common in the Atlantic Ocean and nearly absent in the Pacific Ocean (Fig. 1A). The highest abundance of banding in the modern ocean is located in the North Atlantic Ocean. A meridional hydrographic section shows sections with banding are largely restricted to depths between 2000–4000 meters, corresponding to depths bathed in oxygen-rich deep water (Fig. 1B, 1D). The average modern bottom water oxygen concentration for a banded core is 245 ± 40 µmol/L (1σ) (Fig. 1C). Cores where banding is absent have a lower mean and a wider distribution (170 ± 70 µmol/L). This pattern suggests that high bottom water oxygen is a prerequisite for band formation, but not a guarantee of their presence.
Today, the upper oceans are entirely supersaturated with respect to calcite, while, mainly because of the increasing solubility of calcite with pressure, the deeper oceans are undersaturated. In the deep ocean, dissolved carbonate in the form of bi-carbonate ion (HCO3-) dominates the dissolved inorganic carbon (DIC) budget. The calcium carbonate (CaCO3) content of seafloor sediments typically reduces with increasing depth, especially below the calcite saturation depth68,69. However, this pattern also varies geographically with the age of bottom water as oxygen is consumed and respired carbon accumulates, pushing the solubility of calcite further towards dissolution70. Thus, distributions of DIC in the deep ocean largely mirrors bottom water oxygen, both evolving as a function of ocean ventilation age. The modern deep Atlantic Ocean is well-ventilated, carbonate ion-poor, and rich in oxygen, and the modern deep Pacific is poorly ventilated, carbonate ion-rich, and low in oxygen71.
Changes in the production and configuration of deep water masses in Earth’s past, impacting the distribution of oxygen and DIC in the deep ocean, have also left their imprint on calcite dissolution patterns. During past glacial periods, it’s widely suggested that AMOC experienced slowdowns, with NCW production weakening and shoaling34,72. At depths below 3000 meters, NCW was replaced by aged SCW which reduced deep-water oxygen levels throughout the global ocean and rendered the Atlantic Ocean more corrosive to CaCO373–75. The transition back to interglacial conditions was accompanied by a resurgence of well-ventilated NADW at great depths in the Atlantic. Redox fronts and banding have been noted to form around the deglacial transition in sediment cores. We test the influence of a deglacial paleo-environmental change on the formation of these bands with a compilation of sedimentary CaCO3 content data from core tops and from Last Glacial Maximum (LGM)-aged sediments76. Regional averages (according to the Longhurst biogeochemical provinces77) of Holocene and LGM CaCO3 concentrations are differenced to produce a regionally sensitive proxy for deep water ventilation change. On average, the distribution Holocene – LGM CaCO3 is centered near zero for core sites without banding (Fig. 1E), with a wide distribution (2.5 ± 14 wt %), reflecting a decrease in preservation in the Atlantic and better preservation in the Pacific. Sediment core sites with banding display a large increase in CaCO3 contents during the Holocene (19 ± 14 wt %) (Fig. 1E). We argue these data support our model for band formation because in a simple sense, sediments from a poorly ventilated, corrosive North Atlantic should have higher organic matter contents, either due to reduced dilution by CaCO3 or higher export efficiency from a deoxygenated water column (Fig. 2A). The post glacial rise in deep water ventilation (oxygen supply) would respire this carbon and create banding.
Bands forming at an active redox front (Supplementary Fig. 1) are universally orange, red, or brown. However, beneath the front, bands are often green, indicating burial to depths with reducing pore waters. The observation of green banding at depths beyond the reach of oxygen diffusion is much more common (Supplementary Fig. 1). These green bands can be found at sites that lack a band forming at an oxidative color transition (Fig. 1) and sites that lack an oxidative color transition at all. We demonstrate that fluctuations in bottom water oxygen are most likely responsible for the formation of these green bands by linking observations from our survey with stratigraphic records of the occurrence of these bands from two different sediment core sites in the path of modern NADW that extend back over the past 1.2 million years.
We present a record of the presence of green bands on a newly generated benthic foraminiferal oxygen isotope age model from sediment core Site U1474 (see methods, Extended Data Fig. 2), located in the Southwest Indian Ocean near Durban, South Africa (31°13.00′ S; 31°32.71′ E, 3045 m below sea level, Fig. 1). Proximity to the margin leads terrigenous sedimentation to dominate this region of the Natal Valley, with CaCO3 concentrations varying between 30–50%. The target for the core site was a contourite drift78, a sedimentary feature sculpted by bottom currents that at the present day flow northward below 2000 meters79. The physical and chemical signature of this water indicates its source in the North Atlantic Ocean, i.e NADW80. Paleoceanographic reconstructions based on carbon and neodymium isotopes from the nearby Cape Basin and Agulhas Plateau show that the reduced influence of NCW is a consistent feature of glacial periods across the Middle to Late Pleistocene, a feature shared with many cores across the Atlantic Basin81–84. We additionally document a record of green bands at North Atlantic Ocean Site U1313 (41°00.068' N, 32°57.438' W, 3426 m below sea level), a reoccupation of Deep Sea Drilling Program Site 60746,85. This site, with its well-developed age model85, paleoceanographic context, and location proximal to NCW formation allows us to distinguish between local and large-scale environmental phenomena driving the creation of green bands.
The green band records were generated with the assistance of the OpenCV library, implemented in a Python framework86. Image data is typically stored in a matrix with dimensions x-y-3, where x is the height, and y is the width. The third dimension stores the intensity of light in red, green, and blue. OpenCV has a function, inRange, that filters images for color, pixel by pixel, in the more intuitively designed Hue-Saturation-Value colorspace (HSV). The function accepts arguments Hmin, Smin, Vmin and Hmax, Smax, Vmax, which make a cube-shaped filter in 3D color space (Extended Data Fig. 3B). The function returns a binary matrix with the shape of the original image, where pixels that fit within the filter are coded true (Supplementary Fig. 3C-D). This matrix can then be summed along the depth axis of the core to create a record of the intensity of a color within the core. Our approach breaks up HSV color space into 10 x 10 x 10 blocks, which we called "channels". We used digitized versions of Munsell color chips to restrict the channels to a range of colors typically expressed by sediment, reducing the number of channels tested from 3600 to 627 (Extended Data Fig. 4). We used several core sections to test the sensitivity of each channel to the variable appearance of green bands in the sediment to find a combination of channels that maximize the intensity of the signal produced by green bands with depth (Extended Data Fig. 5). This set of filters (Extended Data Fig. 6A, 6B, Extended Data Table 1A and 1B) is then applied to each section image from the core site, and the resultant depth series from each channel is summed to generate a record which we call "green pixel percent."
The catalog of green bands with depth is created using the depth series of percent green pixels. Highs in the record, which should represent the occurrence of a green band, are then identified using a peak-finding algorithm. Sediment at those depths is then visually assessed and false positives are removed from the record.
Green pixel percent and depths flagged positively for the occurrence of a green band are plotted alongside an image of core section U1474F 5H3 in Fig. 3. Also shown are XRF core scan-derived ratios of Fe/Ti (Fig. 3D). This ratio acts as a proxy for diagenesis in sediments because iron is diagenetically mobile under reducing conditions, and titanium is not56,87. XRF core scanning at Site U1474 was conducted at a 2 mm resolution88, approximately 5–10 times the typical resolution for the long cores collected by the IODP89–92. This resolution is important, as most bands are much less than 2 cm in thickness.
Peaks in the green pixel percent consistently correspond to peaks in Fe/Ti throughout the core section, indicating the diagenetic origin of the bands (Fig. 3C). This relationship remains true throughout the length of the record at Site U1474, which is visualized in its entirety in Supplementary Document 1. We leverage the clear connection between color and the chemical fingerprint of diagenesis at Site U1474 to create records of shallow sediment diagenesis, proxied by green band occurrence at sites like Site U1313 that lack adequately resolved chemical data (Supplementary Document 2).
The results of the stratigraphic survey of green banding at Sites U1474 and U1313 are visualized in Figs. 4 and 5. We use two different binning techniques to highlight what we interpret to be the most salient patterns in the data. Figure 4(B-C) shows the results in terms of thickness of green bands (in cm) per thousand years of sediment accumulation for each site according to its age-depth model (see methods, Extended Data Fig. 2). This visualization highlights the frequency and severity of individual deoxygenation events and is useful for interpreting mechanisms leading to the development of bands on orbital-millennial timescales. Figure 5(B) shows the number deoxygenation events per MIS. With this visualization, we aim to determine drivers of long-term variability in the frequency of deoxygenation events across the Pleistocene.
We identified 239 bands at Site U1474, with a thickness totaling 288 cm, accumulating at an average rate of 2.3 mm/ka (Fig. 4B). At Site U1313, there are 142 bands with a thickness totaling 191 cm, a rate of 1.5 mm/ka (Fig. 4C). In both records, the bands appear with greater abundance in the first half of the record (1200 − 600 ka BP). This pattern is even more exaggerated at North Atlantic Site U1313. There appears to be substantial coherence between band formation at the disparately located core sites. This appears especially true during the earlier part of the record (1200–700 ka BP). During the latter half of the record, green banding at U1474 is more abundant than at U1313. To compare sites and test for synchronicity, we binned the banding data into 5 ka intervals at both locations. The binning interval accounts for uncertainty in the independent age-depth models. A total of ~ 55% of banded intervals with green bands at Site U1313 are accounted for in the record from U1474. The sum of the thickness from the intervals from U1313 that match banding at U1474 accounts for most of the total thickness of bands from the site (91%). This finding suggests a common, underlying process with regionally expansive presence influencing both regions and that the Southern Hemisphere site exhibits a greater sensitivity to this phenomenon.
The most salient observation in the recurrence of green band occurrence is that they are almost completely restricted to sediment from glacial periods. At Site U1474, 90% of the bands are placed in intervals when benthic foraminiferal 𝛿18O is higher than 4 per mil (Fig. 4B, 4C, Extended Data Fig. 7A-B). At U1313, that number is lower, at 76% (Extended Data Fig. 7C-D). The observation that these bands, called relic diagenetic fronts by Funk et al.56, are more abundant during glacial periods is consistent with their survey of Equatorial Atlantic sediment cores.
The literature suggests that these bands are a deoxygenation-reoxygenation phenomenon52–55. The most pronounced increase in deep water oxygen levels should occur at the deglaciations, providing an opportunity for a first order test of reoxygenation hypothesis. The link between deglaciations and green band formation is highly consistent at Site U1474. These bands are often the thickest and darkest in color, such as those at glacial Terminations 1, 2, 3, and 6. Lighter green bands were observed at Terminations 2 and 7, but no green layer was found at Termination 5 (Fig. 4B, Extended Data Fig. 8).
While several bands are likely attributed to post-glacial reoxygenation, many bands are positioned well before glacial terminations. Site U1474 has an average sedimentation rate of 4 cm/ka (Hall et al. 2017), so if the modern depth of the color transition at Site U1474, with a depth of 60 cm (Supplementary Fig. 1), reflects the extent of oxygen penetration into sediment pore waters under typical interglacial conditions, the post-glacial oxygen rise could only account for the oxidation of organic matter emplaced within ~ 15 ka of the glacial termination. Clearly, the post glacial deep water oxygen resurgence is responsible for the formation of many of the green bands, but the limited depths of oxygen penetration in pore water requires that many of the bands are formed by more frequent, short-lived deep ocean oxygen fluctuations during glacial periods themselves.
We test the influence of rapid bottom water oxygen variability on band formation using a quantitative, high-resolution proxy reconstruction of bottom water oxygen concentrations from sediment core site IODP U1385 (Fig. 4D, 37°34.285′N, 10°7.562′W, 2578 mbsl)73,90,93. The record is well dated, synchronized using a benthic oxygen isotope stratigraphy94. Green bands are notably absent from this core. We suggest this is a result of a relatively high sedimentation rate (10 cm/ka), which would quickly bury organic matter beyond the reach of oxygen in pore waters90. The oxygen proxy is based on the difference between carbon isotope ratios of epifaunal and infaunal benthic foraminifera73,93. Because many measurements fall out of the calibration range for this proxy (less than 255 umol/L), the record mostly displays deoxygenation events in the deep ocean. Despite the different geographical locations of the core sites, bottom water oxygen concentrations in the North Atlantic (via NCW ventilation) act as an excellent predictor of green band occurrence in the Indian Ocean, especially during the intervals between 1000 − 600 ka BP and 400–200 ka BP.
Orbital forcing may also be influencing patterns of band development within glacial periods. The glacial occurrence of green bands seems most aligned with axial precession, recurring with a roughly 23 ka period. For example, the dark green bands at 36.05 and 36.78 meters CCSF (Fig. 3) are spaced at 23 ka intervals, based on the age model resulting from benthic oxygen isotope stratigraphy. The phasing is consistent with minima in Northern Hemisphere summer insolation or maximum in Southern Hemisphere summer insolation (Extended Data Fig. 9A). Glacial bands from both Sites U1474 and U1313 appear to share this association with insolation forcing (Extended Data Fig. 9B). The bands are largely absent during interglacial periods, likely due to the continuous influence of well-ventilated NADW. However, several notable exceptions to the glacial interglacial pattern in banding (MIS 3, 7a, 13a, 17a, 17c) coincide with precession maxima. Again, the observation of a precession periodicity is consistent with the results of the magnetic surveys on relic diagenetic fronts from the Equatorial Atlantic Ocean56.
Funk et al.56 ascribed the formation of the bands to productivity pulses, but we prefer an explanation invoking variation in the source or ventilation of deep water masses. Key to this interpretation is the synchronicity of the development of the bands on a regional scale, represented in three different sediment cores from both hemispheres. Presently, Sites U1313, U1385, and U1474 are linked by the flow of well-ventilated NADW, and paleoceanographic surveys show consistent expansion of contraction of NCW/SCW water mass signals across the glacial cycles of the Pleistocene. Thus, as opposed to variability in productivity, which could be more influenced by local climate processes, we prefer to attribute the formation of green bands bottom water oxygen variability driven by either the contraction of a well-ventilated water mass or the expansion of a poorly ventilated water mass across the Atlantic basin.
The glacial interglacial cycle in the occurrence of green bands is imprinted on a lower frequency, long-eccentricity-paced (~ 400 ka) pattern of variability (Fig. 5B). Binning the number of green bands from both sites by MIS makes it easy to identify the two long-eccentricity cycles of the last million years, with high green band occurrence at approximately 700 ka BP and 300 ka BP ka BP (Fig. 5B). We argue that even though the bands form from low oxygen events, conversely, highs in green band abundance on longer timescales represent intervals where bottom water oxygen was high in the Atlantic Basin. This is because without an adequate supply of oxygen, organic matter rich layers would not be oxidized into green bands. Below we discuss other proxy evidence for, driving mechanisms of, and carbon cycle imprints of cyclicity in bottom water oxygen.
Shifts in bottom water oxygen are evident in the style of deposition of sapropels in the Mediterranean Sea95–97. Low oxygen conditions there permit the burial of these organic rich layers, and they occur in phase with precession variability in Northern Hemisphere insolation60,95–97. Most sapropels have lost some of their original thickness due to the “burn-down” of organic matter from their tops after oxygenated water returns to the depths of the Mediterranean Sea54,55. However, there are long intervals in the Mediterranean barren of sapropels. These periods, so called “Red Intervals” due to their color, are thought to be the result of persistently high bottom water oxygen60. The occurrence of these red intervals, binned by MIS’s, are plotted alongside green bands in Fig. 5C. The highs in the MIS green band frequency at 700 and 300 ka BP are matched by the abundance of red intervals in the Mediterranean. We argue that the environmental change that led to the formation of more green bands in the Atlantic Ocean and fewer sapropels in the Mediterranean Sea share a driving mechanism, higher bottom water oxygen.
Other proxy evidence supports our model for the evolution of bottom water oxygen across the Pleistocene. An 800 ka-long proxy record for oxygen variability exists in the Natal Valley in the form of authigenic uranium data from sediment core MD96-207798. Uranium is an effective proxy for the presence of dissolved oxygen because it is highly sensitive to redox conditions and is preserved, bound to sediment particles, only in low oxygen environments99. Authigenic uranium concentrations are high during glacial MIS 8, 10, and 12, suggesting particularly low oxygen during these intervals, aligning with the low green band abundance between 600 − 300 ka BP we observe in Site U1474 (Extended Data Fig. 10). The abundance of green bands Site U1474 also bears a striking resemblance to proxy records for carbonate dissolution, which are sensitive to the ventilation of deep water (Extended Data Fig. 11)100,101. While not a direct proxy for dissolved oxygen, carbonate dissolution is driven by the concentration of carbonate ion in seawater, which in the modern ocean is largely the inverse of bottom water oxygen71. Two records of one such proxy, foraminifera shell weights and sizes, from both the North Atlantic and the North Indian, record an extended interval of carbonate dissolution between 500 − 400 ka when green bands are largely absent from both core sites100,101 (Extended Data Fig. 11). The dissolution proxy indicates that during this time, the deep Atlantic and Indian Oceans were poorly ventilated, rich in respired carbon, and corrosive, much like the modern-day Pacific. The correspondence between low-frequency variability in green band formation and this dissolution proxy suggests, much like the core-top survey (Fig. 2), that green bands form in an environment more like the modern Atlantic.
The occurrence of green bands and red intervals corresponds to low-frequency variability observed in the benthic d13C stack (Fig. 5D). This connection yields insight into potential drivers forcing long-term variability in the carbon cycle. We approximated global carbon cycle fluctuations (i.e. changes in the size of the terrestrial biosphere and deep ocean carbon storage) with a stack of benthic d13C from 14 core sites across the global ocean (Extended Data Fig. 12). The mechanisms driving these global shifts have long been debated102–106. A prominent hypothesis related to the whole ocean carbon isotope swings is focused on productivity in the Southern Ocean. The carbon isotope ratio of the ocean should become heavier if total global surface ocean productivity and/or the "rain ratio" of isotopically heavy particulate inorganic matter (coccolithophore) to isotopically light particulate organic matter (diatoms) increases. It is postulated that this phenomenon is driven by eccentricity lows which promote coccolithophore growth via longer growing season duration102,104,106–108. Biogeochemical models and culture experiments109–111 suggest that lower seasonality leads to a longer growing season and promotes their evolutionary success.
However, it has also been proposed that high bottom water oxygen concentrations, acting to reduce organic matter preservation and change the burial fluxes of organic and inorganic carbon, could produce the same pattern in benthic d13C103. Simulations show that changing the burial fluxes of inorganic carbon and organic carbon with oxygen is the most plausible way to reproduce observed surface and deep ocean carbon isotope variability without unreasonable perturbations to the carbonate saturation horizon or pCO2103. To meet these conditions, rates of inorganic carbon and carbon burial must vary with a ratio of 1:1 based on the stoichiometry of these equations about the respiration of organic matter and dissolution of calcium carbonate:
CH2O + O2 ⇔ CO2 + H2O
CaCO3 + CO2 + H2O ⇔ Ca2+ + 2HCO3−
Simply put, increased bottom water oxygen makes the burial and removal of isotopically light carbon from the ocean reservoir less efficient, causing benthic 𝛿13C to decrease. The green band distribution we observe is phased with whole ocean 𝛿13C variability in a way that suggests bottom water oxygen likely played a key role in driving long-term carbon system dynamics. Notably, the Southern Ocean is a powerful lever on systems controlling bottom water oxygen via SCW production. Additionally, models also indicate that in the modern ocean, biological activity in the Southern Ocean is responsible for lowering ocean oxygen levels by as much as 20%112. Culture experiments suggest that low eccentricity ought to promote higher productivity109–111. This hypothesis is borne out in n-alkane abundances, a productivity proxy, from South Atlantic ODP Site 1090113. Glacial periods MIS 8, 10, 12, and 14 are characterized by high productivity in the subpolar South Atlantic Ocean (Fig. 5A). Increased coccolithophore production would raise d13C and drive down bottom water oxygen. The decline in bottom water oxygen would increase the burial of isotopically light organic carbon and further increase d13C.
In summary, we invoke three mechanisms acting on different timescales to explain the general pattern of green bands. First, to create an individual band, bottom water oxygen must decrease in order to increase the perseveration of organic matter within deep sea sediments and then resurge to respire that organic matter after burial. The timescale for the creation of a single band is on the order of thousands of years, limited by the sedimentation rate and the depth to which oxygen can penetrate in sediment pore waters. Our data provide limited constraints on the mechanism driving bottom water oxygen variability on these timescales. Potential sources include the variability in NCW, SCW or local changes in productivity. However, we argue that when the formation of bands is synchronized over a large region, deep water mass variability is the most reasonable driving mechanism. Second, bands are more abundant during glacial periods due to the reduction/shoaling of NCW, which reduces deep water oxygen content. The lower bottom water oxygen content during glacial periods must be closer to the threshold at which organic matter is preserved or respired, increasing both the number of bands and the sensitivity to bottom water oxygen fluctuations. Finally, we suggest that long eccentricity-paced variability in the abundance of bands could be driven by orbitally influenced productivity in the Southern Ocean that drive down bottom water oxygen.
The use of green bands as a proxy for the presence of bottom water oxygen is limited by the requirement that there be a certain threshold in dissolved oxygen reached after a low bottom water oxygen event in order to respire organic matter and create a green band. If such an event occurs, but the bottom water oxygen never rises above a certain threshold following the event, the event would not be recorded by the proxy because the organic matter in sediments is simply preserved, and no green band is created. The proxy is also limited in that it isn't a quantitative measure of oxygen concentrations. However, in comparison to the best proxies for dissolved oxygen concentrations, foraminiferal I/Ca ratios42 and difference in epifaunal-infaunal benthic carbon isotope ratios41,73, generating very high-resolution records of green band formation using computer vision and images of sediment cores is rapid and easy. This method is portable to new sites and potentially useful for understanding ancient deep water oxygen dynamics at high resolution in potentially much older sediment. We recommend using the data presented here as a training set to develop a computer vision model. This would facilitate the automatic detection of green bands and expedite the application of this proxy to the vast archive of sediment core images collected in the well-documented history of the International Ocean Discovery Program.
Our work grounds the connection of green band formation to bottom water oxygen with a global survey of sediment core tops and demonstrates how the occurrence of diagenetic banding in the sediments from the Atlantic Basin record deoxygenation-reoxygenation events caused by jump starts in the production of NADW. The proxy is limited by a threshold in dissolved oxygen concentrations that must be surpassed after an event in order to record it. This limitation, however, in and of itself reveals that a few Mid-Pleistocene glacial periods (MIS 8, 10, 12) were characterized by particularly low bottom water oxygen. This variability is roughly paced with the 400 ka band of orbital eccentricity, and our observations add evidence to the suggestion that deep ocean oxygen changes likely played a role in the carbon cycle variability across the Mid-Late Pleistocene.