OCEAN (DE)OXYGENATION A CROSS THE LAST DEGLACIATION INSIGHTS FOR THE FUTURE

. Anthropogenic warming is expected to drive oxygen out of the ocean as the water temperature rises and the rate of exchange between subsurface waters and the atmosphere slows due to enhanced


INTRODUCTION AND CONCEPTS
Oxygen is a sparingly soluble gas, and its scarcity in the ocean affects the welfare and behavior of marine animals in a large fraction of the ocean (Stramma et al., 2010). Although marine phytoplankton are responsible for about half of the planet's oxygen production, most of this photosynthetically produced dissolved oxygen quickly outgasses, and the concentration remains within a few percent of saturation over most of the ocean surface (Garcia et al., 2010).
Meanwhile, waters in the cold and dynamic environments of dense water formation tend to fall below oxygen saturation, given that it takes several weeks for gas exchange to compensate for the enhanced solubility stemming from cooling, combined with the often brief exposure of aged, undersaturated waters at the surface, particularly in the Southern Ocean (Gruber et al., 2001).
As a result, while most of the waters near the ocean surface have a "preformed" oxygen concentration (i.e., the concentration these waters had when last exposed at the ocean surface) close to that of oxygen saturation, it can deviate by more than 20% from this value in some waters, especially those that are ventilated at very high southern latitudes (Ito et al., 2004).
ABSTR ACT. Anthropogenic warming is expected to drive oxygen out of the ocean as the water temperature rises and the rate of exchange between subsurface waters and the atmosphere slows due to enhanced upper ocean density stratification. Observations from recent decades are tantalizingly consistent with this prediction, though these changes remain subtle in the face of natural variability. Earth system model projections unanimously predict a long-term decrease in the global ocean oxygen inventory, but show regional discrepancies, particularly in the most oxygen-depleted waters, owing to the complex interplay between oxygen supply pathways and oxygen consumption.
The geological record provides an orthogonal perspective, showing how the oceanic oxygen content varied in response to prior episodes of climate change. These past changes were much slower than the current, anthropogenic change, but can help to appraise sensitivities, and point toward potentially dominant mechanisms of change. Consistent with the model projections, marine sediments recorded an overall expansion of low-oxygen waters in the upper ocean as it warmed at the end of the last ice age. This expansion was not linearly related with temperature, though, but reached a deoxygenation extreme midway through the warming. Meanwhile, the deep ocean became better oxygenated, opposite the general expectation. These observations require that significant changes in apparent oxygen utilization occurred, suggesting that they will also be important in the future.

ACROSS THE LAST DEGLACIATION INSIGHTS FOR THE FUTURE
In the ocean's interior, respiration by the marine ecosystem consumes the oxygen transported downward from the surface, with the most rapid rates of net oxygen consumption occurring just below the euphotic layer. Oxygen consumption goes essentially to completion in oxygen minimum zones (OMZs) that occupy about 1% of the modern ocean volume (Bianchi et al., 2012), after which denitrification must step in to provide oxidizing potential, with a host of biogeochemical consequences (see below). The OMZs are typically located in regions without unusually high rates of O 2 utilization but where water is poorly connected to the deep wintertime convection regions (Figures 1 and 2) at which oxygenated surface waters are injected into the ocean's interior (Luyten et al., 1983;Karstensen et al., 2008

OCEAN DEOXYGENATION IN A WARMING WORLD
It is expected that as the ocean warms in the future, its oxygen content will decrease, particularly in the upper ocean, a phenomenon referred to as "ocean deoxygenation" (Keeling et al., 2010;Matear and Hirst, 2003). Recent observations seem to indicate relatively large changes in dissolved oxygen over the past decades in various ocean basins . But these changes remain subtle in the face of natural variability, so that great caution needs to be used when interpreting trends over 20 years or less in the context of global warming (Frölicher et al., 2009).
The future evolution of the ocean's oxygen content under anthropogenic climate change depends on three factors: (1) the degree of ocean warming, (2) changes in oxygen demand resulting from changes in growth and respiration rates, and (3) changes in ocean circulation and mixing. Projections with current-generation Earth system models as well as theoretical arguments suggest a consistent trend toward lower oxygen content of the global ocean (Sarmiento et al., 1998;Cocco et al., 2012;Bopp et al., 2013). For the high greenhouse gas emission scenarios (such as RCP 8.5 with a radiative forcing target of 8.5 W m -2 in year 2100), where surface ocean warming approaches 3°C, the total oceanic loss of oxygen by the year 2100 amounts to between 3% and 4%. But even for conservative emission scenarios (e.g., RCP 2.6), where surface warming remains below 1°C relative to present, the ocean is projected to lose about 2% of its current oxygen inventory by the end of this century (Bopp et al., 2013).
In nearly all models, this trend is predominantly driven by changes in ocean circulation and mixing, particularly the increase in upper ocean stratification, which allows the O 2 bio demand to accumulate in the ocean interior. This trend is to a substantial degree reinforced by upper ocean warming, which reduces O 2 sat . In fact, this reinforcement tends to lead to a rather consistent and uniform relationship between ocean heat uptake from the atmosphere and loss of oxygen from the ocean, with a ratio of about 4 to 6 nmol O 2 J -1 , as evidenced from both models (Plattner et al., 2002) and observations (Keeling et al., 2010;Stendardo and Gruber, 2012). In The most recent set of Earth system models project changes for the hypoxic class water volume (O 2 < 50 µmol kg -1 ) from -30% to +15%, and for the anoxic class (O 2 < 5 µmol kg -1 ) from -10% to more than +40%.
The reasons for these large regional differences in the future evolution of oceanic oxygen content are associated with different mechanisms that control the oxygen concentrations and different circulation/mixing time scales (Sarmiento and Gruber, 2006).
relatively coarse-resolution global Earth system models, resulting in different and largely inaccurate representations of modern oxygen minimum zones in these models (Cocco et al., 2012). These problems persist when the models are used to project future changes in these regions, limiting our ability to assess how OMZs will develop in the future. But perhaps the oxygen changes of the past ocean will help us to better constrain the future. Here, we make use of the three most common proxies: laminations, benthic foraminiferal species assemblages, and redox-sensitive trace metals (see Box 1).

Deglacial Changes in Oxygenation
The geological record provides a perspective on how oceanic oxygen content responded to prior episodes of climate change. These past changes tended to be much slower than the current, anthro-   (Adkins, 2013). The Holocene (i.e., the last 10,000 years), on the other hand, has been a period of relative climate stability (Marcott et al., 2013)   Latitude Latitude temperature dependence of the re mineralization rate of sinking organic matter (Matsumoto, 2007 (Boyle, 1988). Consistent with this observation, paleoceanographic proxy records from OMZs have been interpreted as showing general intensification of oxygen depletion across the deglaciation (Jaccard and Galbraith, 2012, and references therein). This occurred even though biological production may have decreased at low latitudes as a result of slackening wind-driven upwelling (Kohfeld et al., 2005).
In addition, the last deglaciation provides a unique perspective on transient changes between the two steady states.
Deglaciation did not follow a smooth, gradual progression;

BOX 1. TOOLBOX -BENTHIC OXYGENATION PROXIES
High sedimentation rate and extreme oxygen deficiency typically suppress sediment reworking by burrowing organisms. Absence of sediment mixing allows the seasonal cyclicity in marginal sediment supply to be preserved as thin, typically millimeter-scale laminations. The development of laminations depends on the nature of the sedimenting material and is only sensitive to oxygen at very low dissolved oxygen concentrations (i.e., close to anoxia). The absence of laminations could reflect a lack of regular variations in sedimenting material, but the presence of laminations is an unambiguous testimony of bottom water oxygenation levels < 5 µmol kg -1 (van Geen et al., 2003). Foraminifera dwelling at or just below the sedimentwater interface are particularly suitable for monitoring abrupt climate change and oxygen levels because many species are known to be opportunistic, rapidly responding to environmental change, including bottom water oxygenation and organic matter availability at the seafloor (Corliss, 1985).
Many trace elements are present in seawater either in soluble form or adsorbed onto particles. Under oxygen-depleted conditions, sedimentary redox-sensitive trace metal enrichments (such as V, Mn, Mo, Cd, Re and U) may occur through diffusion across the sediment interface and precipitation as mineral phases at reducing horizons within the sediment (Tribovillard et al., 2006). It has been generalized that redox-sensitive metals precipitate as authigenic mineral phases in sediments where oxygen penetrates to less than 1 cm (Morford et al., 2005), and dissolved oxygen levels of overlying bottom waters exceeding 50 µmol kg -1 are rarely observed (McManus et al., 2005).
All three proxies have their shortcomings and are sensitive to changes of oxygen over differing concentration ranges. In addition, their behavior will vary at each site, depending on local sediment dynamics. However, these individual weaknesses are largely independent of each other and, thus, including all three proxy types, we can infer a relatively robust qualitative representation of past changes in benthic oxygenation. A more quantitative assessment of the changes, however, is not yet possible. Because the oxygenation proxies are most sensitive at low oxygen concentrations (typically < 20% of saturation), the compiled oxygenation changes are biased towards the Indo-Pacific where oxygen is, on average, lower. In addition, most records are located in continental margin settings, where higher sediment accumulation rates provide the temporal resolution required to resolve millennialscale changes in bottom water oxygenation. Hemisphere. This interruption was one of many such AMOC oscillations that punctuated glacial ice core records with dramatic temperature swings between the two hemispheres, known as the bipolar seesaw (Stocker, 1998)  LGM; 22,000-20,000 years ago) and the Early Holocene (10,000-5,000 years ago), and (b) the Bølling/Allerød-Antarctic Cold Reversal (14,700-12,5000 years ago) and the Early Holocene (10,000-5,000 years ago). The multiproxy data compilation is plotted on top of the expected O 2 sat change. The symbols illustrate the proxies used to infer past changes in oxygenation, with squares corresponding to laminations, circles to benthic foraminifera species assemblages, and diamonds to redox-sensitive trace metals. Blue shadings indicate a relative decrease in oxygenation and orange shadings a relative increase in oxygenation. The multiproxy data compilation is plotted on top of the expected O 2 sat change estimates based on LGM-Holocene changes in sea surface temperature and salinity. LGM to Holocene B/A-ACR to Holocene Galbraith, 2013), which ventilate a large portion of the North Pacific thermocline. Second, it is plausible that the global ocean nutrient inventory decreased across the deglaciation, due to accelerating removal of nitrogen through denitrification , and diminishing input of iron as the atmosphere became less dusty (Winckler et al., 2008). Although it has been argued that a range of negative feedbacks would have prevented very large, durable changes in the nitrogen inventory (Gruber, 2004), it seems likely that nutrient inventories gradually shrank, to some degree, in response to deglacial warming, limiting export production. This suggestion is qualitatively consistent with proxy evidence for particularly high export production during the B/A-ACR in the sub-Arctic North Pacific (Kohfeld and Chase, 2011) and along the western coast of North America (Cartapanis et al., 2012).   (Matsumoto, 2007).
3. The deep ocean physical circulation changed over the deglacial warming in a way that is not captured by future Earth system model projections. This could be due to the poor representation and parameterization of Southern Ocean deepwater production and circulation in Earth system models (Russell et al., 2006). Alternatively, deep ocean circulation during the LGM may have differed fundamentally from modern circulation, either due to nonlinearities of ocean circulation at low temperature (Keeling and Visbeck, 2011)