Volume 117, Issue C11
Free Access

Oxygen trends over five decades in the North Atlantic

I. Stendardo

Corresponding Author

I. Stendardo

Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland

Now at Institut für Umweltphysik, Abteilung Ozeanographie, Universität Bremen, Bremen, Germany

Corresponding author: I. Stendardo, Institut für Umweltphysik, Abteilung Ozeanographie, Universität Bremen, DE-28359 Bremen, Germany. ([email protected])Search for more papers by this author
N. Gruber

N. Gruber

Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland

Search for more papers by this author
First published: 03 November 2012
Citations: 65


[1] We investigate long-term trends in dissolved oxygen in the North Atlantic from 1960 to 2009 on the basis of a newly assembled high-quality dataset consisting of oxygen data from three different sources: CARINA, GLODAP and the World Ocean Database. Oxygen trends are determined along isopycnal surfaces for eight regions and five water masses using a general least-squares linear regression method that accounts for temporal auto-correlation. Our results show a significant decrease of oxygen in the Upper (UW), Mode (MW) and Intermediate (IW) waters in almost all regions over the last 5 decades. Over the same period, oxygen increased in the Lower Intermediate Water (LIW) and Labrador Sea Water (LSW) throughout the North Atlantic. The observed oxygen decreases in the MW and IW of the northern and eastern regions are largely driven by changes in circulation and/or ventilation, while changes in solubility are the main driver for the oxygen decrease in the UW and the increases in the LIW and LSW. From 1960 until 2009 the UW, MW, and IW horizons have lost a total of −57 ± 34 Tmol, while the LIW and LSW horizons have gained 46 ± 47 Tmol, integrating to a roughly constant oxygen inventory in the North Atlantic. Comparing our oxygen trends with those of the oceanic heat content, we find an O2 to heat change ratio of −3.6 ± 2.8 nmol J−1 for the UW, MW and IW, and a ratio of −2.8 ± 3.4 nmol J−1 for the LIW and LSW. These ratios are substantially larger than those expected from solubility alone.

Key Points

  • O2 changes significantly over the last five decades in the North Atlantic
  • The upper ocean has lost O2, while the lower waters have gained O2
  • O2/heat change ratio is larger than that expected from solubility

1. Introduction

[2] Dissolved oxygen tends to respond very sensitively to climate variability and change because any perturbation in sea-surface temperature not only changes the solubility of dissolved oxygen, but also alters upper ocean stratification in a way that tends to amplify the solubility effect [Najjar and Keeling, 2000; Gruber et al., 2001; Keeling and Garcia, 2002; Plattner et al., 2002; Bopp et al., 2002; Keeling et al., 2010]. This amplification is a result of ocean stratification affecting the supply of oxygen from the near-surface ocean to the ocean's interior, i.e., altering the transport from the place where it is produced by photosynthesis or reset by air-sea exchange to where it is consumed by heterotrophic respiration [Sarmiento and Gruber, 2006]. This high sensitivity to climate forcing makes oxygen one of the best candidates for detecting and better understanding the link between global warming and the resulting biogeochemical and physical changes in the ocean [e.g., Joos et al., 2003; Deutsch et al., 2005a, 2005b; Körtzinger et al., 2006; Brennan et al., 2008; Keeling et al., 2010]. The amplification also implies that the ocean likely will lose a substantial amount of oxygen in the coming decades and centuries in response to global warming, a process termed “ocean deoxygenation” [Keeling et al., 2010]. Global models suggest a loss of between 1 to 7% for this century for a business as usual scenario [Matear and Hirst, 2003; Frölicher et al., 2009] with an oxygen to heat change ratio of about −5 nmol O2 J−1 [Keeling et al., 2010].

[3] Several studies reported long-term decreases in the oxygen content in the interior ocean [e.g.,Bindoff and McDougall, 2000; Emerson et al., 2001; Johnson and Gruber, 2007; Whitney et al., 2007; Stramma et al., 2008, 2010; Helm et al., 2011] (see summary in Keeling et al. [2010]) consistent with the expectation of a warming ocean losing oxygen. However, the vast majority of studies reported trends over 20 years or less, and only a few looked at trends over 40 years and more [Watanabe et al., 2003; Whitney et al., 2007; Stramma et al., 2008]. Great caution needs to be used when interpreting trends over 20 years or less in the context of global warming [Gruber, 2009]. Interannual-to-decadal variations are substantial, and will influence the determination of long-term trends, particularly when trend estimates (as in most of these studies) are based on relatively limited data. This is especially the case for analyses based on a few repeated transects as they usually have substantial temporal gaps between them, making them prone to projecting interannual variations erroneously onto long-term trends [Gilbert et al., 2010]. Thus, the determination of whether the oceanic content of oxygen has decreased in response to the observed global ocean warming requires not only an extension of the analysis backward in time to cover at least 30 to 40 years, it also requires a good temporal resolution in order to be able to distinguish between natural fluctuations and long-term changes. This can only be achieved by combining data from a large number of different cruises and by analyzing them on a region-by-region basis [see, e.g.,Stramma et al., 2008; Gilbert et al., 2010].

[4] Another crucial requirement for the reliable determination of long-term trends is the availability of internally consistent data. The combination of data from a large number of sources makes this requirement particularly challenging, especially for oxygen, as there exists no certified reference material or absolute standards [Emerson et al., 1999]. This internal consistency is usually achieved by applying a secondary quality control procedure to the data [Johnson et al., 2001; Key et al., 2004, 2010], i.e., data from different cruises are compared in the deep ocean at cross-over locations resulting in the determination of adjustments that are then applied on a cruise-by-cruise basis [see, e.g.,Tanhua et al., 2010]. While this secondary quality control does not guarantee the high level of accuracy needed for the computation of the saturation state, for example, it makes the data suitable for long-term trend studies [Stendardo et al., 2009]. Some projects such as CARINA (CARbon In the Atlantic) [Key et al., 2010] and GLODAP (GLobal Ocean Data Analysis Project) [Key et al., 2004] have already assembled a large collection of hydrographic data that were subjected to first and secondary quality control procedures, but their temporal coverage is limited.

[5] In this paper we augment the oxygen data from the CARINA and GLODAP data sets with selected cruises from the World Ocean Database 2005 (WOD05) [Boyer et al., 2006] (National Oceanographic Data Center NODC) in order to generate an internally consistent data set for the entire North Atlantic covering the period from 1960 until 2009, i.e., covering nearly 50 years. This permits us to reliably determine long-term trends and to investigate the causes for these trends including the relationship with the observed warming of the ocean [Levitus et al., 2005, 2009, 2012].

[6] We focus on the North Atlantic for several reasons. First, prior analyses of the oxygen changes in the North Atlantic have revealed substantial changes [Johnson and Gruber, 2007; Stramma et al., 2008]. However, these analyses were spatially restricted or covered only the last 20 years. Second, the North Atlantic is subject to substantial interannual to decadal changes in connection with the North Atlantic Oscillation (NAO) [Visbeck et al., 2003] with potentially important implications for oceanic oxygen [Frölicher et al., 2009]. Third, and finally, the North Atlantic has the best data coverage of all ocean basins, making it possible to reliably determine trends over an entire basin.

[7] The longest trend analyses of oxygen in the Atlantic were conducted by Stramma et al. [2008], who focused on the oxygen minimum zones (OMZ) of the tropical North and South Atlantic. They found that the OMZ in these regions intensified and expanded over the last 50 years, similar to the signals they identified in the other oceanic basins. They suggested that this could be caused by anthropogenic climate change, especially since they found corresponding decreases also in the Pacific and Indian Oceans. However, no formal attribution was undertaken. Long-term oxygen changes are not restricted to the tropical Atlantic. Further north in the domain of the subpolar gyreJohnson and Gruber [2007] reported a substantial decrease in the oxygen content of the Mode and Intermediate waters. They attributed these changes largely to changes in the NAO, which evolved from largely negative states in the 1960s to positive ones in the late 1980s/early 1990s, but since then has decreased again to near neutral and negative states [Hurrel et al., 2003]. No connection of this loss of oxygen to anthropogenic climate change could be made, since the analyzed data record covered only 15 years. At the same time, model simulations predict substantial decreases in the oceanic oxygen in the North Atlantic in response to anthropogenic climate change until the end of the century, only partially offset by increases in some regions of the tropical Atlantic [Frölicher et al., 2009]. But Frölicher et al. [2009]demonstrated also that the North Atlantic is one of the regions where the natural variability of oxygen is largest so that detection and attribution studies will be more difficult there. Here, we do not attempt to conduct such a formal detection and attribution, but rather provide the long-term observational basis required for such future studies.

[8] A better understanding of long-term oxygen variations and changes in the North Atlantic may help also to better decipher changes associated with the Meridional Overturning Circulation (MOC) [Brennan et al., 2008] and its linkage to subpolar gyre circulation [Häkkinen, 2001], Labrador Sea Water (LSW) formation and spreading [e.g., van Aken et al., 2011; Yashayaev et al., 2007; Kieke et al., 2009] and its important role in the uptake of anthropogenic carbon [Gruber, 1998; Sabine et al., 2004; Steinfeldt et al., 2009].

2. Methods

2.1. Data Sets

[9] Our analysis is based on three data sets: CARINA [Key et al., 2010], GLODAP [Key et al., 2004], and WOD05. CARINA is a collection of 98 hydrographic cruises for the Atlantic Ocean (188 in total for the entire ocean) spanning 29 years (from 1977 to 2006). 81 of these cruises contain oxygen observations, all of which underwent a careful secondary quality control, resulting in the application of adjustments to 23 of them. Most of the adjustments, computed as multiplicative factors, were in between ±1% and ±2%, while only five cruises needed an adjustment larger than 2% [Stendardo et al., 2009]. GLODAP contains 48 cruises in the Atlantic Ocean with a time span of 15 years starting from 1978. All GLODAP cruises contain high quality oxygen data that were also subjected to a secondary quality control [Key et al., 2010]. CARINA and GLODAP are internally consistent, since GLODAP was used as a reference during the secondary quality control procedures of CARINA [Stendardo et al., 2009]. WOD05 contains 5645 cruises in the Atlantic with at least one oxygen observation, spanning the period from 1960 to 2004. While the spatio-temporal coverage of this data set is very good, these data underwent only a relatively rudimentary primary quality control (seeBoyer et al. [2006] for details), and no adjustments for cruise offsets were applied. This precludes the use of the WOD05 for trend determination without proper data selection/adjustments.

[10] Rather than attempting to adjust the entire WOD05 oxygen data set, we used CARINA data to identify and select those cruises from WOD05 that are consistent with the CARINA oxygen data. The selection procedure consisted of a similar method used to determine the adjustments of the CARINA data set during the secondary quality control [Tanhua et al., 2010]. In detail, we computed the oxygen offsets between CARINA and WOD05 measurements on a cruise-by-cruise basis through a cross-over analysis using the MATLAB routines provided by CARINA (http://cdiac.ornl.gov/oceans/CARINA/Carina_inv.html). The results from the cross-over analysis were used to calculate the adjustment values for each cruise of the WOD05 using an inversion based on a Weighted Damped Least Square method (WDLSQ) (all the details of the methods are described inTanhua et al. [2010]). The weight applied in the cross-over analysis included a term that is inversely proportional to the temporal gap between two cruises in order to account for potentially real changes [Stendardo et al., 2009]. Only those cruises that required an adjustment of less than 1%, i.e., whose adjustment factor was between 0.99 and 1.01, were selected. All other cruises were discarded. This resulted in the identification of 184 cruises with high-quality oxygen data spanning the period 1960 to 2000. Duplicate cruises, i.e., those included in more than one of the three sources, were sub-selected with the highest priority given to CARINA and then GLODAP. In order to further extend the length of the record, we added one recent high-quality cruise conducted in 2009, i.e., Merian cruise MSM12/3 that sampled the subpolar North Atlantic [Rhein et al., 2009].

[11] The combined data set contains 331 cruises, 22,849 stations, and a total of 465,998 oxygen observations (251 cruises, 13,224 stations and 239,703 oxygen observation for the North Atlantic from 30°N to 65°N) spanning the period from 1960 until 2009 and with a good spatial coverage (Figure 1a). CARINA and GLODAP dominate the period from 1980s to 2005, while WOD05 contributes most of the data for the period from 1980 until 1995 and also permitted us to extend the record back to 1960 (Figure 1b).

Details are in the caption following the image
(a) Map of the North Atlantic showing the stations contained in the different data sets (green for WOD05, blue for GLODAP, violet for CARINA, and red for the cruise MSM12/3). Also shown are the eight regions: LS (Labrador Sea), IS (Irminger Sea), IB (Iceland Basin), RT + WEBn (Rockall Trough and Western European Basin North), NFL (Newfoundland Basin), NAB (North American Basin), MAR (Mid Atlantic Region), and WEBs (Western European Basin South). (b) Number of observations per year in the different data sets within the North Atlantic region shown in Figure 1a.

2.2. Trend Analysis

[12] We defined 8 regions of investigations based on basin topography and circulation pattern, listed in Figure 1a. The definition of such large regions permits us to obtain a good temporal coverage in each region, but comes at the expense of a potential aliasing effect emanating from a possible uneven spatial distribution of the data within the region over time. We attempted to eliminate this potential aliasing by adjusting each oxygen observation in our data set to the center of the region. This spatial adjustment was determined from the gridded WOCE Global Hydrographic Climatology [Gouretski and Koltermann, 2004] by computing the climatological difference between the oxygen concentration at the grid-point and grid depth closest of each observation, and the mean oxygen concentration of the entire region at that depth. These adjustments were then subtracted from each oxygen measurement. These adjustments ranged between −18.1μmol kg−1 and 18.7 μmol kg−1 (5th and 95th percentiles), while 50% of the adjustments were smaller than 0.1 μmol kg−1.

[13] The trend analysis was performed on isopycnal layers on the basis of potential density (σ1) referenced to 1000 dbar pressure and with a layer thickness of 0.1 kg m−3 until the 32.4 kg m−3 layer, and a thickness of 0.05 kg m−3 until 32.5 kg m−3. Data shallower than 100 m depth and occurring at densities lighter than the surface density in winter time were excluded from the analysis in order to avoid seasonal signals. The winter time density at each location was taken from the World Ocean Atlas 2001 [Conkright et al., 2002]. We binned the data to annual increments on each layer, in order to avoid potential aliasing effects from the fact that certain years contain many more observations than others. The resulting data distribution by layer and region (Figure S1 in the auxiliary material) reveals overall a good temporal coverage for each analyzed σ1layer, but also some pronounced gaps. These gaps impact the determination of interannual variations, but our tests based on subsampling the data indicate that they are not causing major problems regarding the determination of the long-term trends.

[14] We divided the σ1 layers into five density horizons: Upper Water (UW) for 30.2 ≤ σ1 < 31.3 kg m−3, Mode Water (MW) for 31.3 ≤ σ1 < 31.8 kg m−3, Intermediate Water (IW) for 31.8 ≤ σ1 < 32.3 kg m−3, Lower Intermediate Water (LIW) for 32.3 ≤ σ1 < 32.4 kg m−3 and finally Labrador Sea Water (LSW) for 32.4 ≤ σ1 < 32.45 kg m−3. No trends were considered below these densities, as the secondary quality control used for the CARINA and GLODAP syntheses assumed temporally invariant oxygen concentration below 3000 m [Tanhua et al., 2010; Stendardo et al., 2009]. This permitted them to use these deep data to determine the offset adjustments, but also resulted in the removal of any deep trends, should such have existed.

[15] Within these density horizons we can identify the main Atlantic ocean water masses. The upper water masses characterized by thick layers of nearly uniform properties (temperature, salinity and density) [McCartney and Talley, 1982; Brambilla and Talley, 2008] are the Mode Waters consisting of Subpolar Mode Water (SPMW) in the subpolar North Atlantic and Subtropical Mode Water (STMW) in the subtropical gyre of the North Atlantic. The SPMW, which is found throughout the supolar gyre, is characterized by densities between σθ = 26.9 kg m−3 (σ1 = 31.3 kg m−3) and σθ = 27.2 kg m−3 (σ1 = 31.7 kg m−3) and temperatures between 14°C and about 11°C [Talley et al., 2011]. Thus, our MW horizon corresponds largely to SPMW. Underneath the MW lies the Intermediate Water (IW) often characterized by extrema in one or more hydrographic parameters [van Aken, 2000]. These water masses include the Sub-Arctic Intermediate Water (SAIW), the Mediterranean Sea Outflow Water (MOW) and the Antarctic Intermediate Water (AAIW). Intermediate waters like the MOW and the SAIW have a mean potential density of aboutσ1 = 32.15 kg m−3, with MOW associated with a strong salinity maximum and seen as a deep marker of subtropical influence [Johnson and Gruber, 2007]. In contrast, the SAIW originates in the western boundary current of the subpolar gyre and can be considered as a deep marker of subpolar influence [Johnson and Gruber, 2007]. The AAIW instead originates in the Southern Ocean and it moves northward and is brought into the northeastern North Atlantic by the North Atlantic Current (NAC). These three water masses occupy our IW horizon. The LSW density horizon corresponds exactly to the domain of the Labrador Sea Water, which is formed through deep convection in the Labrador Sea [Yashayaev et al., 2007].

[16] The oxygen trends were computed for each region and density horizon using a general least-squares linear regression method, where we included the effect of autocorrelation explicitly in the estimation procedure [Pinheiro and Bates, 2000]. Namely, we fitted the following model to the O2 data for each layer k making up the density horizon:
where t is the year, t0 is 1985, ak is the intercept of a particular layer, and where we assume the residual εto have a multi-variate normal distribution with a mean of zero and a correlation matrix, whose off-diagonal elements include the temporal autocorrelations. We estimated those by assuming an exponential model of the semivariance,γ, i.e.,
where s is the distance (in time) between the observations in each layer, r is the range (i.e., the temporal decorrelation length) and g is the “nugget” effect, which incorporates the fact that data at lag zero are not perfectly correlated with each other due to “noise” and measurement uncertainties. For each density horizon, this resulted in the estimation of 3 + n parameters, i.e., bgra1, …, an, where nis the number of layers in each horizon. A likelihood ratio test revealed that the inclusion of the autocorrelation was needed in about 20% of all cases. In order to be internally consistent, we fit the same model with autocorrelation to all data. The significance of the trend was determined by a Student-t test with a threshold of 95%.

[17] Implicit assumptions in this approach are that the trend over the entire water horizon is the same and that all layers have the same temporal decorrelation length. We checked on these assumptions by repeating the above analysis by fitting the data layer by layer (n = 1). In that case, we set the nugget to zero.

[18] When fitting the data over an entire horizon, we also assume that the data on the different layers are independent, i.e., we don't include a vertical autocorrelation. This is justified because we are analyzing trends on density layers, which eliminates most of the effects of processes that can create such vertical autocorrelations, such as eddies and fronts.

2.3. Temperature, O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0009, AOU, and O*2

[19] In order to determine the potential processes driving the changes in oxygen, we also analyzed trends in the saturation concentration of O2, i.e., O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0007, trends in the apparent oxygen utilization (AOU), and trends in the quasi-conservative tracer O*2 [Gruber et al., 2001]. The analysis of trends in O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0007permits us to assess the role of heat exchange at the air sea-surface, as this affects the saturation concentration and ultimately the uptake/loss of oxygen at the sea-surface (the effect of changes in freshwater fluxes is very small and can be neglected). We compute O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0007 from the concurrently measured temperatures and salinities using the formulation of Weiss [1970]. Use of the newer formulation by Garcia and Gordon [1992] would have given nearly identical results for the trends.

[20] Trends of −AOU urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0003 indicate the contribution of processes other than warming/heating on the trend in oxygen, i.e., primarily the rate of oxygen consumption from the remineralization of organic matter, i.e., the biological driver, and the rate of transport and mixing of the water mass [Sarmiento and Gruber, 2006]. Thus we separate the trends into the following two contributors:

[21] An important caveat in this separation is the assumption that surface ocean oxygen fully equilibrates with the atmosphere in response to heating/cooling of the surface. This does not always occur, particularly not in the high-latitudes during winter time [Ito et al., 2004]. As a result, one would tend to overestimate the heat flux driven component trend and underestimate the trend component stemming from biology and/or transport/mixing.

[22] A corollary separation can be achieved by analyzing the trend in the quasi-conservative tracer, O*2 = O2urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0010, where urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0011 is the oxygen to phosphorus ratio of biological uptake/release [Keeling and Garcia, 2002]. O*2 is a tracer that reflects the O2gained or lost by a water parcel through air-sea gas exchange, irrespective of whether this exchange occurs as a result of biological consumption and production of O2 in the surface ocean or whether it is the result of heating and cooling the surface [Gruber et al., 2001; Keeling and Garcia, 2002]. This second separation thus consists of the following:
where the “no-gasex” component includes all processes at the surface and interior that are not associated with the exchange of O2across the air-sea interface. This is primarily biology and transport and mixing. In contrast, O*2, i.e., the “gasex” component corresponds directly to the amount of O2 that the ocean is losing or gaining from the atmosphere, i.e., it is that part of the marine O2 that leaves an imprint on atmospheric oxygen [Gruber et al., 2001; Keeling and Garcia, 2002]. The biological production and consumption at the surface add or remove oxygen from the water that can create an air-sea disequilibrium of oxygen and thus lead to air-sea gas exchange [Gruber et al., 2001] that consequently affect the O*2. A reduction of the transport of this biologically produced oxygen from the surface into the ocean's interior would tend to cause a larger loss of this biological oxygen to the atmosphere, and thus causing a reduction in O*2. An increase of this transport would tend to increase O*2. However, cooling and warming at the surface have also the potential to alter O*2strongly since they can cause an air-sea disequilibrium of oxygen as well.

[23] We used the phosphate observations from our data sets to compute O*2. While a secondary quality control was undertaken for phosphate in the GLODAP and CARINA data sets, no additional checks were performed on the phosphate data from WOD05. In the CARINA dataset, the overall accuracy of the phosphate is about 2.6% [Tanhua et al., 2009] compared to 0.8% for O2 [Stendardo et al., 2009], making the determination of O*2 rather uncertain. The analysis is further hampered by the lack of phosphate data for many cruises that have O2 data. This is especially the case for many of the earlier cruises, therefore restricting the length of the O*2 trends substantially. In addition, the above separation is sensitive to the assumption of a constant stoichiometric ratio between oxygen and phosphorus. Although this stoichiometric ratio varies relatively little during remineralization [Anderson and Sarmiento, 1994; Gruber et al., 2001], it nevertheless adds uncertainty to the trends in O*2. As a result of these limitations, we will be using O*2 trends in a qualitative manner only.

2.4. Inventory Changes

[24] We computed the changes in the oxygen (and heat) inventories by multiplying the rates of change with the volume of the respective water mass on each analyzed σ1 layer. In the standard case, the volumes were determined from the WOCE Global Hydrographic Climatology [Gouretski and Koltermann, 2004], thereby neglecting temporal changes in the volumes of the water mass or layer under consideration. To compute inventories by depth intervals, we first transformed the rates of change by σ1 layer to rates of change by depth, and then integrated them vertically. The transformation from σ1 to depth was also based on the WOCE Global Hydrographic Climatology.

[25] We recognize that the assumption of temporally constant volumes is potentially problematic, especially since a number of studies have shown substantial changes in the distribution of water masses in the North Atlantic, particularly in that of the Labrador Sea Water [Kieke et al., 2009; Yashayaev et al., 2007]. We made an attempt to include this effect by computing the inventories separately for each decade and then assessing the trends. To this end, we first constructed decadal mean vertical density profiles for each region by adjusting all density data for the spatial mean pattern in the same way as done for O2. We then computed the thickness of each isopycnal horizon and determined its rate of change by linear regression. The impact of this volumetric change on the column inventory was then computed by multiplying the total change in isopycnal height with the mean O2 anomaly of that horizon. This anomaly was determined by subtracting the average O2concentration for the entire region from the long-term mean O2 profile for that region. We use the vertical anomaly instead of the actual O2 concentration since only differences in the O2 concentration between the horizons matter. This is because the sum of the losses and gains over all isopycnal horizons owing to the exchange of the mean O2 concentration is zero. The final result, i.e., the change in inventory stemming from changes in isopycnal volumes, was obtained by multiplying the changes in the column inventory with the area of each region as a function of depth. Given the large uncertainties in the reconstruction of the changes in the density field, our confidence in these inventory change estimates is low, so that we report the values computed under the assumption of constant volumes and then investigate the impact of the changes in isopycnal volumes on these estimates.

3. Long-Term Oxygen Trends

3.1. Overview

[26] We find substantial and statistical significant long-term changes in the oxygen concentration for many regions and many of the density horizons over the last 49 years starting from 1960 (Figures 2a, 2d, 2g, 2j, and 2m). In this plot and in the related Figure 4, the trends were converted to concentration changes over the 49 years of the records, with the statistical significance expressed in the figures with different symbols. Because of the sparsity of data and the elimination of data above the winter time outcropping density, some regions do not have estimates for all density horizons (Figure S1 in the auxiliary material).

Details are in the caption following the image
O2 concentration changes over the last 49 years for the five horizons and their drivers. (a, d, g, j, and m) The changes in oxygen and the changes induced (b, e, h, k, and n) by solubility (represented by the urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0006) and (c, f, i, l, and o) by changes in circulation/ventilation and remineralization (represented by the −ΔAOU), respectively. Shown are the results for the different water masses: UW (Figures 2a–2c), MW (Figures 2d–2f), IW Figures 2 g–2i), LIW (Figures 2j–2 l), and LSW (Figures 2 m–2o). Statistically significant concentration changes are represented with filled circles, while the not significant ones are plotted with open circles.

[27] Our results reveal that the UW (which can be identified only in the four southern regions of the North Atlantic), has lost oxygen in all four regions with a maximum of −13.5 ± 7.7 μmol kg−1 in the MAR region (Figure 2a). Also the MW has lost oxygen especially in the northern and eastern regions (RT + WEBn and WEBs), with a maximum of −18.6 ± 9.4 μmol kg−1 in the RT + WEBn region (Figure 2b). At the same time, the MW in the southwestern regions (NAB, NFL and MAR) has gained oxygen with a total change of up to 18.7 ± 10.2 μmol kg−1 in the NAB region. Also the IW layer shows a distinct east–west difference in trends. In the eastern part of the North Atlantic (IB, RT + WEBn, MAR and WEBs) oxygen has decreased on average over the past 50 years with a maximum decrease observed in the MAR region of −13.4 ± 4.2 μmol kg−1. On the other hand, O2 has increased in the northwestern and western part of the basin (LS, NAB), although not statistically significant. The LIW and LSW horizons show generally small increases in the oxygen concentration, although not statistically significant with the exception of the MAR and IB region in the LSW water horizon, where O2 increased by more than 9.5 ± 7.8 μmol kg−1.

[28] Overall, we find a complex pattern of trends with the upper layers of the ocean, corresponding to the domains of the UW, MW and IW horizons, having lost oxygen, and with the deeper layers, representing the LIW and LSW horizons, revealing no substantial changes or having gained a small amount of oxygen. The weighted average oxygen decrease over the UW, MW, and IW water mass horizons is −4.8 ± 2.2 μmol kg−1 for the last 49 years. The deeper layers (LIW and LSW) gained a statistically not significant 3.6 ± 3.7 μmol kg−1 over this period.

[29] Our trend analysis by water mass horizon assumed a relatively homogeneous trend over the layers making up each horizon. Figure 3 confirms that this is largely the case, although this figure also reveals how variable the trends are with density.

Details are in the caption following the image
Isopycnal oxygen trends from 1960 through 2009 for the 8 analyzed regions. The colors of the lines correspond to the colors of the regions. The shading around each lines represents the trend ±2 standard deviation (2σ). The gray boxes represent the σ1-based density horizons defined in this study: UW (Upper Water), MW (Mode Water), IW (Intermediate Water), LIW (Lower Intermediate Water), and LSW (Labrador Sea Water).

3.2. Potential Causes

[30] The separation of the trends into a solubility (O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0007) and into a biology and circulation/ventilation (AOU) driven part, (see equation (3)) reveal that both sets of processes contribute to them, with the AOU contribution overall dominating (Figure 2).

[31] Similar to the oxygen trends, the −AOU trends show negative values in the domain of the UW, IW and MW, and positive values in the MW in the southern regions and no significant in the LIW and LSW (Figures 2c, 2f, 2i, 2l, and 2o). One main difference exists on the shallower isopycnal layers in the southern regions, where the decreases in oxygen are larger than those in −AOU. In the NFL, for example, −AOU contributes only (a statistically insignificant) −6.9 ± 9.2 μmol kg−1 to the overall oxygen decrease of −15.0 ± 7.3 μmol kg−1. The most important difference between the trends in O2 and −AOU exists on the southeastern LSW horizons, where the statistically significant increase in O2 is not associated with a trend in −AOU.

[32] The trends in oxygen solubility dominate the oxygen trends only in a few areas (Figures 2b, 2e, 2h, 2k, and 2n), i.e., in the upper horizons in the southern regions (NFL and MAR) where the decrease in oxygen is largely caused by a decrease in oxygen solubility, and in the deeper isopycnal horizons where increasing oxygen solubility explains most of the observed increase in oxygen. The strongest changes in oxygen solubility are found in the MW and IW horizons in the NFL and MAR regions with a maximum value in the NFL of −16.9 ± 7.0 μmol kg−1. This strong negative change is, however, compensated by an even stronger positive change from −AOU, so that the overall change is slightly positive.

[33] We next discuss the oxygen changes by horizon, aiming to disentangle the contributions from physical and biological processes in more detail. To this end, we include also the analysis of the changes in O*2 (Figure 4), which permits us to determine the contribution of air-sea exchange to the O2 changes. The latter process responds quite sensitively to changes in upper ocean convection, so that we can use changes in O*2 as a qualitative tracers of this process. In addition, O*2 changes indicate also to what degree changes in oxygen have been transmitted into the atmosphere.

Details are in the caption following the image
Same as Figure 2, but for the gas-exchange component of oxygen, i.e., O*2. Regions with insufficient data are shaded in grey.

3.2.1. Upper Water Horizon

[34] The substantial long-term decrease in oxygen of about −4.3 ± 3.1μmol kg−1 observed in the entire Upper Water horizon and especially in the central regions of the southern North Atlantic (MAR) is largely due to a reduction in solubility stemming from the concurrent warming (Figures 2a and 2b). This reduction is enhanced by the negative trend in −AOU (Figure 2c), which is by itself not statistically significant, but nevertheless increases the trend by several ten percent. This reduction in −AOU (increase in AOU) is likely primarily caused by the warming induced decrease in the ventilation of this particular isopycnal horizon, although a biological contribution cannot be excluded. The O*2 changes (Figure 4a) are even more negative in most regions of this isopycnal horizon, indicating that most of the oxygen reduction was transmitted by outgassing into the atmosphere, increasing oxygen there.

3.2.2. Mode Water Horizon

[35] In the Mode Water horizon, the clear separation between a large decrease in oxygen in the eastern and northern regions of the North Atlantic, and a strong increase in the southwestern regions, is almost entirely driven by the changes in −AOU (Figure 2f). Solubility damps the circulation and biology-driven changes in the south-central regions (NFL and MAR) (Figure 2e).

[36] The long-term decreases in the eastern and northern regions are rather steady in time (Figure 5), with a hint of a possible acceleration since the early 1990s. This permits us to put these changes in the context of the study of Johnson and Gruber [2007] who investigated the oxygen decrease in the SPMW observed at 20°W between 1988 and 2003. These authors linked the oxygen loss in this water mass to variations in the NAO with the observations in 1993 reflecting conditions after a period of relatively high NAO, and those in 2003 reflecting conditions after a period of much lower NAO. This shift in NAO caused the SPMW to be colder, fresher, denser, and hence containing more oxygen in 1993, and warmer, saltier, lighter and hence containing less oxygen in 2003. They also identified a substantial contribution of −AOU to the changes, which they suggested to be primarily the result of a northwest movement of the subpolar gyre and a corresponding shift in the position of the NAC and the associated front between the higher and lower oxygen flavors of SPMW. While our oxygen data concur with the oxygen decrease between the early 1990s and early 2000s (Figure 5), they indicate almost no contribution from solubility changes. Instead, our data attribute the oxygen reduction to a trend in −AOU, irrespective of whether the change corresponds to the entire period or just to the period from 1993 to 2003. Thus, while the longer-term perspective confirms the conclusions ofJohnson and Gruber [2007] with regard to the role of −AOU, our new data suggest no substantive contribution from solubility.

Details are in the caption following the image
Timeseries of trends in oxygen and its components for selected regions of the mode-water density horizon. (a–c) Trends for O2, (d–f) those for O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0007, and (g–i) those for O*2. Actually plotted are the anomalies relative to the long-term mean on each of the isopycnal layers that contribute to the mode-water density horizon. Figures 5a, 5d, and 5g depict the results for the RT + WEBn region. Figures 5b, 5e, and 5h depict the results for the WEBs region. Figures 5c, 5f, and 5i depict the results for the NAB region. The symbols are the anomaly from the median value of the linear trend for each isopycnal layer that forms the MW horizon, while the thick line is the linear trend. In the lower left corner is the trend with the uncertainty inμmol kg−1 decade−1.

[37] Johnson and Gruber [2007] argued that given the corresponding changes in temperature, salinity, potential vorticity, and the depth structure of the observed oxygen changes, the reduction in −AOU likely does not come from an increase in the rate of biological oxygen demand, but is more likely a result of circulation and ventilation changes. Our results support this conclusion, although we lack the concomitant observations to separate the role of biology clearly from the role of circulation. Nevertheless, the general decrease in O*2 in the northeastern North Atlantic and especially in the RT + WEBn region (Figure 4b) would support an important role of ventilation, referred here to all processes supplying atmospheric O2to the ocean interior including air-sea gas exchange and the transport of O2across the base of the surface mixed layer. This is because in the absence of cooling or warming, one could explain all changes by invoking a long-term trend toward enhanced stratification, which tends to reduce O*2, and also tends to increase the residence time of waters on this isopycnal horizon with regard to the surface, explaining the concomitant reduction in −AOU and oxygen.

[38] In the southern part of the southeastern North Atlantic (WEBs), we observe a slight decrease (although not significant) of O*2 and decrease in the oxygen. O urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0007did not change, while −AOU decreased. In this case, the oxygen changes are likely driven by changes in remineralization and/or circulation, with not much contribution from the gas-exchange.

[39] The southwestern and south-central regions of the North Atlantic (MAR, NFL and NAB) exhibit a strong positive trend in −AOU (Figure 2f), suggesting a strong increase in ventilation/circulation and/or a decrease in biological consumption of oxygen. Warming in the MAR and NFL regions reduce the −AOU-driven trend substantially (Figure 2e), causing the overall trend to be statistically not significant (Figure 2d). No such dampening effect exists in the NAB region, leading to a substantial and significant oxygen increase there. The absence of significant changes in O*2 suggests that little O2 was exchanged with the atmosphere, so that the additional O2in the interior must have other sources. A strong reduction of biological oxygen demand can be likely excluded, since in such a case, one would expect a strong decrease in oxygen demand throughout the water column with the strongest reduction occurring in the upper ocean. This is not the case since −AOU shows little trend in the UW horizons in the southwestern and south-central regions (Figure 2c). This leaves stronger ocean circulation/ventilation as the primary candidate for explaining the increase in oxygen in the mode waters of the southwestern North Atlantic. We will discuss this further below.

3.2.3. Intermediate Water Horizon

[40] The east–west difference in the oxygen changes in the Intermediate Water horizon (Figure 2g) is entirely driven by the trends in −AOU (Figure 2i) while the solubility-driven trends modify the magnitude in the central Atlantic (Figure 2h). The trends are overall similar to those observed in the overlying Mode Waters, although generally weaker and mostly statistically insignificant (Figures 2d–2f). What differs, however, are the trends in O*2 (Figure 4), where the IW waters show nearly everywhere an increase in O*2, albeit not a significant one. Thus gas-exchange appears to have a small positive impact on oxygen on this horizon. Using the same argument as above regarding the unlikely role of biology in causing the −AOU and hence the oxygen changes, this leaves ocean circulation as the most likely candidate for explaining the differential trends in oxygen in the intermediate waters.

[41] Johnson and Gruber [2007] reported also oxygen decreases for the intermediate waters in the eastern North Atlantic between1993 and 2003. Our data confirm this finding and suggest that this decrease is actually just a part of a much longer trend, spanning the entire period from the early 1960s to the present (Figure 6). Johnson and Gruber [2007] interpreted the changes in this density range also in the context of the shifts in the NAO with cold, fresh (and high oxygen) intermediate waters of northwestern origin being replaced by warmer, saltier, and less oxygenated intermediate waters of more southern origin (MOW and AAIW). Our slight negative trend in oxygen solubility in the WEBs region support this interpretation, although the overall oxygen trend in this region suggest a much stronger forcing from −AOU. This suggests that the stronger northward penetration of both MOW and AAIW, which dominate the intermediate water domain, might have been a process that has been ongoing since the 1960s.

Details are in the caption following the image
Same as Figure 5, but for selected regions for the intermediate water horizon. Shown are the results for (a, d, and g) the IB region, (b, e, and h) the WEBs region, and (c, f, and i) the MAR region.

3.2.4. Lower Intermediate and Labrador Sea Water Horizons

[42] In contrast to the overlying water masses, the waters of the LIW and LSW horizons exhibit an increase in oxygen nearly throughout the entire North Atlantic (Figures 2j and 2m) although only in two regions in the LSW the increase is statistically significant. Most of the increases are driven by an increase in oxygen solubility caused by the long-term cooling of these waters (Figures 2k and 2n). The contribution of −AOU is generally not statistically significant, but nevertheless enhances the solubility-driven signal in most regions (Figures 2l and 2o).

[43] The long-term increase in oxygen in the LIW and LSW horizons is mostly determined by the increase in oxygen saturation (decrease in temperature) between the 1960s and the 1990s (Figure 7), while the trends have flattened thereafter. This is in agreement with the results of Dickson et al. [2002], who described that from 1966 to 1992 there was an overall cooling of the entire water column of the Labrador Sea. This is a consequence of an intensification of winter-time convection in the Labrador Sea over this period, which leads to the formation of cold, fresh, well oxygenated waters. Hydrographic time series reveal two periods of intensified LSW production, from 1972 to 1976, and from the late1980s to 1997 [McCartney and Talley, 1982; Wallace and Lazier, 1988; Sy et al., 1997]. This evolution brought deepening convection and ultimately formed LSW that was fresher, colder, deeper and denser than at any other time in history of deep measurements there [Dickson et al., 2002]. Since the late 1990s, winter time convection has been relatively steady and even began to decrease recently [Lazier et al., 2002], disrupted only by an event of particularly deep convection in the winter 2007–2008 [Vage et al., 2009; Yashayaev and Loder, 2009]. This led to a slight warming of the LSW in the Labrador Sea as seen by the small downward trend in the oxygen saturation in this region (Figure 7) [Lazier et al., 2002].

Details are in the caption following the image
Same as Figure 5, but for selected regions for the Labrador Sea Water horizon. Shown are the results for (a, d, and g) the LS region, (b, e, and h) the IB region, and (c, f, and i) the WEBs region.

[44] These signals produced locally in the LS region are then transported into the other regions of the North Atlantic along the circulation path of the LSW leading to a relatively homogeneous distribution of the trends throughout the North Atlantic. Nevertheless, the finite spreading time from the Labrador Sea to the other regions leads to deviations, particularly in how the recent cessation of the trend in the Labrador Sea is shared in the other regions. Spreading times from the Labrador Sea are about 6.5 years to the western Mid Atlantic Ridge, 7.5 years to the North Atlantic's eastern side [Koltermann et al., 1999; Kieke et al., 2009], about 4–5.5 years to the Western European Basin [Sy et al., 1997], about 5 years to the Icelandic Basin [Yashayaev et al., 2007], and about 0.5–2 years to the Irminger Sea [Sarafanov, 2009]. Considering this time lapse the signal of increasing oxygen with decreasing temperature until the 1990s and vice versa after the late 1990s can actually be seen in the Irminger Sea but gets attenuated and lost in regions further to the south (Figures 7b and 7c). The relatively quick transfer of the Labrador Sea water properties into the northeastern North Atlantic is likely the reason why Johnson et al. [2005] saw only relatively small changes in oxygen along 20°W between 1988 and 2003.

3.3. Climate Change Versus NAO

[45] Are the oxygen changes we have identified over the past five decades in the North Atlantic already a manifestation of anthropogenic climate change or are they a result of long-term trends in natural modes of climate variability, such as the North Atlantic Oscillation, which itself can be affected by climate change [Shindell et al., 1999; Visbeck et al., 2003]? Without a formal detection and attribution process, we cannot give a definite answer, but our detailed analysis of the drivers underlying the trends provide us a basis for a preliminary assessment.

[46] The NAO represents the main mode of interannual atmospheric variability in the North Atlantic. It is characterized by positive or negative phases that are associated with corresponding changes in wind stress, wind stress curl, and heat and freshwater fluxes [Visbeck et al., 2003]. The NAO has a substantial influence on the rate and intensity of deep convection in the Labrador Sea, and the rate of formation of mode waters [Dickson et al., 2002]. More precisely, during periods of a positive NAO state, the oceanic heat loss in the winter is intensified in the subpolar North Atlantic, while this loss is reduced in the subtropical North Atlantic. This create a negative temperature anomaly in the subpolar gyre and a positive temperature anomaly over most of the western subtropical gyre [Visbeck et al., 2003]. Also the mean path of the North Atlantic current changes in association with the NAO, with positive phases characterized by an intensification and southeastward expansion of the subpolar gyre. During negative phases, generally the opposite changes occur.

[47] During the last 50 years the NAO index evolved from strong negative values in the late 1960 to the early 1970s to near neutral values during the late 1970s until late 1980s. Then in the late 1980s, the NAO index entered a prolonged period where it remained in its strongly positive phase, which lasted until 1995. Thereafter, the NAO index started to decrease to lower positive and sometimes negative values. This decreasing phase is relatively recent, so that the trend of the NAO index over the past 50 years is still positive.

[48] On the basis of what is known about the impact of the NAO on the North Atlantic's temperature, ventilation and circulation, many of the oxygen changes we identified in our study can be related to the overall trend in NAO toward a more positive state.

[49] The primarily ocean warming-driven decreases in oxygen in the Upper Waters in the south-central regions of the North Atlantic (NFL and MAR), can be linked to the trend toward more positive phases of the NAO, which caused a warming of the northern subtropical gyre. The reduced O2 concentration in the mode water horizon in the eastern North Atlantic can also be linked to the NAO, as more positive phases of this variability mode tend to reduce the formation of mode waters, explaining the strong −AOU and O2trends in this region. The same NAO-driven process can explain the circulation-driven reduction in the intermediate water horizon in the eastern North Atlantic. And finally, it is well established that the strong cooling and hence the increase in oxygen solubility and oxygen in the lower intermediate and Labrador Sea water are a consequence of the long-term trend in the NAO.

[50] But there are also some aspects where it is unclear whether the NAO is the main driver. First, we can only provide a tentative link to the NAO when attempting to explain the strong −AOU driven increase in the oxygen content of the mode waters of the southwestern regions, which we interpreted to be primarily caused by enhanced circulation. One possible explanation is that this increase is connected to the onset of stronger convective activity in the Labrador Sea, which enhanced the formation of well oxygenated water in the density range of MW, which then pushed southward across the Great Banks into the southwestern region. At the same time, periods of high NAO are usually associated with low rates of subtropical mode water formation [Joyce et al., 2000], which produces waters of lower density than our MW horizon, but may still influence our trends. Second, it is unclear why the intermediate waters of the western part of the North Atlantic show such a differing trend from those in the east, and third, it is difficult to reconcile the relative contribution of changes in circulation and changes in ventilation to explain the changes in the mode and intermediate waters.

[51] The latter question is related to the interpretations provided by Johnson and Gruber [2007], who emphasized the role of circulation changes in explaining the oxygen changes in the subpolar mode waters between 1993 and 2003. They argued that much of the change they identified could be explained by a northwestward movement of the North Atlantic current associated with the transition from a high to a low NAO state, permitting low oxygen waters to penetrate further north in the eastern side of the Atlantic. This would imply that the oxygen would have been even lower than today in the 1960s, when the NAO was lower compared to today. This is clearly not the case, as the mode waters show a substantial loss of oxygen from 1960 until today, and the same is true for the intermediate waters. This means that either the circulation changes are less important than ventilation or that NAO unrelated factors are the dominant driver for the oxygen changes in the mode waters. One possible explanation is that the northwestward/southeastward shifts of the North Atlantic current occurs in a manner unrelated to the NAO, but rather in response to changes in the wind stress curl over the North Atlantic, as recently suggested by Häkkinen et al. [2011]. They showed that these shifts occurred roughly every 15 years, with the most recent period of a northwestward shift and associated enhanced northward penetration having taken place between the early 1990s and the early 2000s, i.e., during the period analyzed by Johnson and Gruber [2007]. However, no long-term trend exists in these shifts so that they cannot explain the 50 years trends, although they likely contribute to the fluctuations around the linear trends over this period. In particular, they may contribute to the acceleration of the decrease in oxygen in the mode waters of the eastern regions seen after the 1990s (Figure 5).

[52] We cannot assess whether any of these NAO unrelated oxygen changes are a manifestation of anthropogenic climate change, but we note that model simulations up to the end of this century suggest a relatively homogeneous deoxygenation of the North Atlantic in response to climate change. This means that some of the oxygen decreases in the upper three horizons could be related to the climate change induced warming of the North Atlantic, but it is rather unlikely that the oxygen increases in the western part of the basin in the mode and intermediate water horizons are a result of such processes. Furthermore, some of the trends in the NAO may itself be a manifestation of climate change, although this is not well established [Hurrel et al., 2003; Visbeck et al., 2003].

4. Changes in Oxygen Inventory and Its Relationship to Heat Uptake

[53] Summing over all regions and over the upper three density horizons (UW, MW, and IW), the upper North Atlantic has lost −56.8 ± 34.0 Tmol of oxygen between 1960 and 2009 (Table 1). This loss is offset by a not statistically significant gain of 45.9 ± 46.6 Tmol in the two deeper horizons (LIW and LSW), resulting in an insignificant total loss across all considered density horizons of −10.9 ± 57.7 Tmol. Although the upper horizon shows a stronger decrease in oxygen (−4.8 ± 2.2 μmol kg−1) compared to the increase in the lower horizon (3.6 ± 3.7 μmol kg−1), the substantially larger volume of the LIW and LSW horizons (124 × 1014 m3 versus 99 × 1014 m3), leads to a near cancelation of the oxygen inventory over all horizons. The spatial distribution of the oxygen inventory changes follows largely that of the oxygen concentration (compare Figure 8 with Figure 2), but with distinct shifts in emphasis on a regional scale induced by the volume weighting.

Table 1. O2 Inventory Changes, Heat Content Changes, Volumes and O2 Concentration Changes Over the Five Water Mass Horizons Discussed in This Study for Each Regiona
O2 Inv. Changes (Tmol)
UW 2.5 ± 1.2 −2.7 ± 4.9 3.7 ± 2.3 −0.3 ± 0.4 9.3 ± 5.6
MW −1.2 ± 2.1 10.4 ± 6.1 1.6 ± 3.4 6.5 ± 5.0 5.4 ± 7.8 13.5 ± 5.8 −11.6 ± 13.2
IW 1.3 ± 5.7 −2.0 ± 5.0 −15.5 ± 26.2 −8.9 ± 9.6 −1.2 ± 5.1 6.9 ± 7.9 11.9 ± 4.2 4.5 ± 3.5 36.0 ± 30.9
LIW 0.9 ± 16.6 0.4 ± 7.5 2.9 ± 17.6 9.6 ± 8.5 0.7 ± 13.0 0.0 ± 20.0 2.7 ± 11.1 5.3 ± 5.3 22.5 ± 37.9
LSW −0.7 ± 17.6 0.6 ± 6.5 2.8 ± 6.3 2.5 ± 4.7 1.1 ± 5.8 5.2 ± 15.1 5.4 ± 4.8 6.5 ± 6.2 23.4 ± 27.1
Heat Content (1022J)
UW 0.09 ± 0.05 −0.07 ± 0.18 0.14 ± 0.06 0.03 ± 0.01 0.19 ± 0.20
MW 0.01 ± 0.03 0.00 ± 0.10 0.28 ± 0.11 0.02 ± 0.23 0.56 ± 0.17 −0.09 ± 0.12 0.78 ± 0.35
IW −0.00 ± 0.07 −0.02 ± 0.09 −0.06 ± 0.36 −0.16 ± 0.26 0.30 ± 0.28 0.02 ± 0.39 0.53 ± 0.15 −0.01 ± 0.08 0.62 ± 0.69
LIW −0.01 ± 0.50 −0.06 ± 0.10 −0.15 ± 0.53 0.25 ± 0.22 0.11 ± 0.29 −0.13 ± 0.43 −0.09 ± 0.19 0.19 ± 0.17 −0.77 ± 0.96
LSW −0.10 ± 0.14 −0.06 ± 0.07 0.12 ± 0.07 −0.10 ± 0.40 −0.02 ± 0.11 −0.11 ± 0.17 0.16 ± 0.13 −0.21 ± 0.24 0.88 ± 0.55
Volume (1014m3)
UW 1.6 5.3 2.8 0.6 10.4
MW 1.1 6.1 2.4 3.8 7.3 8.9 29.6
IW 2.4 3.6 9.3 11.4 4.9 8.1 9.4 9.9 59.1
LIW 11.8 7.0 9.6 8.7 7.4 11.9 11.7 6.7 74.7
LSW 6.0 4.5 5.8 4.9 5.4 9.0 7.8 6.4 49.7
O2 Conc. Changes (μmol kg−1)
UW 15.0 ± 7.3 −6.0 ± 8.9 13.5 ± 7.7 −3.2 ± 5.3 4.3 ± 3.1
MW −10.4 ± 17.3 18.6 ± 9.4 5.4 ± 13.5 18.7 ± 10.2 10.2 ± 13.8 14.7 ± 5.7 −4.1 ± 5.0
IW 5.9 ± 29.2 1.1 ± 13.7 −14.4 ± 27.2 −6.9 ± 9.9 0.3 ± 10.0 7.3 ± 9.4 13.4 ± 4.2 5.5 ± 3.4 5.3 ± 3.1
LIW 0.8 ± 14.2 1.2 ± 14.1 2.7 ± 18.1 11.0 ± 8.7 −1.4 ± 10.8 −0.6 ± 17.5 1.0 ± 10.0 12.1 ± 10.2 2.9 ± 5.1
LSW −1.3 ± 29.0 1.8 ± 15.2 9.5 ± 7.8 4.9 ± 9.1 5.0 ± 12.8 5.5 ± 14.4 urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0008 9.8 ± 16.5 4.7 ± 5.3
  • a In bold are the results that are statistically significant (p < 0.05).
Details are in the caption following the image
Same as Figure 2, but for vertically and temporally integrated changes in oxygen and in its components, i.e., urn:x-wiley:01480227:media:jgrc12605:jgrc12605-math-0006 and −ΔAOU.

[54] We computed also the changes in oxygen and oxygen inventory as a function of depth by projecting the isopycnally-determined oxygen trends onto a depth coordinate. From 100 m down to 700 m, corresponding largely to the depth range occupied by the UW, MW and IW horizons, the mean oxygen loss between 1960 and 2009 amounts to −4.9 ± 1.8μmol kg−1 giving a total loss of −28.8 ± 10.7 Tmol (Table 2). In contrast, from 700 m to 2750 m oxygen increased, on average by 1.2 ± 1.3 μmol kg−1, yielding a total insignificant oxygen gain of 22.8 ± 24.6 Tmol (Table 2). The relative errors of these depth-based inventories are larger due to the uncertainties associated with the projection from density to depth.

Table 2. O2 Inventory Changes, Heat Content Changes, Volumes and O2 Concentration Changes Over the First 700 m Depth and From 700 m to 2750 m for Each Regionsa
O2 Inv. Changes (Tmol)
100–700 m 4.6 ± 5.3 −1.7 ± 3.1 7.7 ± 3.6 13.5 ± 3.7 3.5 ± 1.6 4.4 ± 3.8 2.0 ± 4.6 13.4 ± 3.4 28.8 ± 10.7
700–2750 m 0.4 ± 10.3 0.9 ± 4.7 7.2 ± 6.6 7.3 ± 5.8 3.7 ± 7.7 4.8 ± 16.1 9.3 ± 6.9 7.7 ± 6.0 22.8 ± 24.6
Heat Content (1022J)
100–700 m −0.03 ± 0.10 0.01 ± 0.06 −0.09 ± 0.10 −0.03 ± 0.06 0.29 ± 0.09 −0.12 ± 0.16 0.67 ± 0.09 −0.03 ± 0.04 0.67 ± 0.27
700–2750 m −0.17 ± 0.19 0.16 ± 0.06 0.31 ± 0.21 0.38 ± 0.26 0.39 ± 0.25 −0.36 ± 0.38 0.29 ± 0.14 0.69 ± 0.23 1.38 ± 0.65
Volume (1014m3)
100–700 m 5.5 3.9 6.4 8.1 4.7 9.1 9.9 9.2 57.0
700–2750 m 17.9 13.6 18.1 24.3 17.8 33.3 31.6 31.6 188.2
O2 Conc. Changes (μmol kg−1)
100–700 m 8.0 ± 9.3 −4.2 ± 7.9 11.6 ± 5.5 16.1 ± 4.5 7.3 ± 3.4 4.7 ± 4.0 1.9 ± 4.5 14.1 ± 3.6 4.9 ± 1.8
700–2750 m 0.2 ± 5.6 0.6 ± 3.4 3.9 ± 3.6 2.9 ± 2.3 2.0 ± 4.2 1.4 ± 4.7 2.9 ± 2.1 2.4 ± 1.9 1.2 ± 1.3
  • a In bold are the results that are statistically significant (p < 0.05).

[55] There is an important caveat in these inventory changes as we have neglected so far the potential contribution of the changes in volume of the different density horizons. Our attempt to consider this contribution suggests an inventory loss in the upper ocean of a somewhat larger amplitude (by ∼4 Tmol in the UW, MW, and IW horizons), and a somewhat smaller gain in the lower parts of the considered ocean (by ∼6 Tmol in the LIW and LSW horizons). These changes represent a roughly 10% modification of our O2 inventory change estimates and are within the uncertainty ranges inferred from the temporal trends in O2. These volume change induced modification of the O2 inventory changes are a consequence of our analyses revealing an overall small increase in the thickness of the upper layers by about 15 m on average over the 49 year period (range from −29 to +53 m between the regions), and a decrease of the thickness of the deeper layers by the same magnitude (also the same range, as the thicknesses of the layers were computed over a temporally constant overall depth range, i.e., between 100 m and the mean depth of the σ1 32.45 kg m−3for each region). However, these estimates are rather uncertain due to our limited capability to reconstruct the three-dimensional time-varying density field in the North Atlantic over the last 50 years. As such reconstructions become available, it will be relatively straightforward to use our oxygen trends to determine improved estimates of the inventory changes.

[56] Interestingly, despite the uncertainty introduced by the potentially changing isopycnal volumes, we find that the total amount of oxygen in the North Atlantic has remained essentially unchanged over the last 50 years. But this hides the fact that over this period, oxygen has been redistributed substantially between the upper and lower layers of the North Atlantic.

4.1. O2 to Heat Change Ratio

[57] Using our temperature data to determine the heat content changes over the eight regions, we obtain for the upper three water horizons (UW, MW, and IW) an overall O2 to heat change ratio of −3.6 ± 2.8 nmol J−1 (Table 3). As the isopycnal volumes cancel in the computation of this ratio, this estimate is not sensitive to the changes in these volumes. The O2/heat change ratio for the LIW and LSW together is smaller in magnitude, i.e., −2.8 ± 3.4 nmol J−1 and not significant at the 95% confidence level. The O2/heat ratios computed by depth yield a similar depth trend, with an even stronger difference between the upper and lower parts of the ocean, i.e., −4.3 ± 2.4 nmol J−1 in the upper 700 m, and −1.6 ± 1.9 nmol J−1 between 700 and 2750 m (Table 3). These consistently negative ratios confirm results from model studies [Plattner et al., 2002; Matear and Hirst, 2003; Oschlies et al., 2008; Frölicher et al., 2009] and data-based considerations [Najjar and Keeling, 2000; Gruber et al., 2001; Keeling and Garcia, 2002] that changes in the oceanic O2 content are strongly tied to changes in the oceanic heat content. Based on the relationship of O*2 with temperature and seasonal cycle of oxygen and heat fluxes, Keeling and Garcia [2002] computed a global mean ratio of approximately −5 nmol J−1 for thermocline waters with a range of −2 to −10 nmol J−1, depending on region and mean temperature. Models run under strong climate change scenarios gave ratios between −5 to −6 nmol J−1 [Keeling et al., 2010].

Table 3. O2/Heat Ratios (nmol J−1) for the Three Water Masses and the Ratios Derived From the Effect of the Solubility Alone by Using the O2 Saturation in Third Column
O2/Heat Ratio O2/Heat Ratio Sol. Alone
100–700 m −4.3 ± 2.4 −1.6
700–2750 m −1.6 ± 1.9 −1.9
UW-MW-MW −3.6 ± 2.8 −1.5
LIW-LSW −2.8 ± 3.4 −2.0

[58] The differences in the O2 to heat ratio between the upper and deeper parts of the ocean confirm the trends reported by Keeling and Garcia [2002], who also reported a substantially more negative ratio for thermocline waters (−5 nmol J−1) in comparison to the deeper waters associated with the formation of North Atlantic Deep Water (−3 nmol J−1). Interestingly, this trend of less negative ratios at lower temperatures and more negative ratios at warmer temperatures is not expected on the basis of the relationship of O*2 with temperature [Gruber et al., 2001; Keeling and Garcia, 2002], as this would suggest the opposite trend. This may suggest that temporal changes in the O2 to heat ratio are not simply a projection of the spatial relationship between O2 and heat uptake/loss.

[59] Comparing our computed ratios with those expected from solubility alone (about −1.5 nmol J−1 for the upper density horizons and about −2.0 nmol J−1 for the deeper horizons) reveal an enhancement factor of about 2.4 in the upper horizons and about 1.4 for the deeper horizons. Thus our upper ocean findings support the results of Keeling et al. [2010] who reported an enhancement factor of around four for thermocline waters. In contrast, Frölicher et al. [2009] found in their model study only an enhancement by a factor of two, more inline with our lower enhancement factor in the deeper density horizons.

[60] All in all, our oxygen to heat change ratios support the long-standing hypothesis that ocean warming will lead to an over-proportional loss of oxygen. Our results are unique because they demonstrate this behavior for the first time on the basis of long-term trends, which is especially important when these ratios are used for estimating past or future changes in the oceanic oxygen inventory [Keeling and Garcia, 2002; Plattner et al., 2002; Matear and Hirst, 2003; Frölicher et al., 2009; Keeling et al., 2010].

5. Summary and Conclusions

[61] By building a high-quality oxygen data set for the North Atlantic, we estimate secular trends in dissolved oxygen over much longer periods and over much broader regions than hitherto possible. A general least-square linear trend analysis finds that the upper layers of the North Atlantic occupied by the UW, MW and IW lost on average about −4.8 ± 2.2μmol kg−1 of oxygen between 1960 and 2009. This loss is not uniformly distributed but mostly concentrated in the eastern and northern North Atlantic. In contrast, the southwestern regions of the North Atlantic have actually gained oxygen. The contribution of the driving mechanisms also differs from region to region. The decrease of oxygen in the UW is driven by a warming induced reduction in solubility. The oxygen decreases observed in the northern and eastern part of the MW and IW are probably driven mainly by changes in circulation and ventilation, while the decrease observed in the central part of the North Atlantic in the IW is a combination of both circulation/remineralization and solubility changes, with the solubility dominating. The increase of oxygen observed in the southwestern region is driven mainly by circulation changes. The trend analysis also shows an increase in the oxygen concentration in the domain of the LIW and LSW with an average (not significant) oxygen gain of 3.6 ± 3.7 μmol kg−1. The increase is rather uniformly distributed across all regions and is mainly driven by changes in solubility.

[62] When integrating the oxygen changes across all horizons and regions, an essentially unchanged oxygen inventory emerges. This hides, however, the fact that this lack of change is actually the result of a near cancelation of a loss of −56.8 ± 34.0 Tmol in the upper layers and a gain of 45.9 ± 46.6 Tmol in the deeper layers. Given the fact that the oxygen gain of the deeper layers is a result of a long-term trend of the NAO towards a more positive state, and that this trend has reversed sign since the mid-1990s, it is intriguing to speculate that the deeper layers of the North Atlantic likely could begin to lose oxygen in the coming decades. Some first indication of this reversal in the trend can be seen in the Labrador Sea, and likely will be spreading into the other regions in the coming years. The recent reversal of the NAO trend may also begin to reverse the trends in the upper density horizons in the coming decades. However, if global warming continues unabated, the resulting strong heating and the associated increase in stratification will likely overwhelm this effect, resulting in a possible future where the entire North Atlantic will be losing oxygen. Given our diagnosed scaling of the oxygen to heat change ratio of −3.6 ± 2.8 nmol O2 J−1 for the upper density horizons (UW, MW, and IW) and −2.8 ± 3.4 nmol O2 J−1 for the lower density horizons (LIW and LSW), any projected warming of the Atlantic Ocean will make this loss substantial. Thus, our analysis strongly supports the notion that if anthropogenic climate change continues to evolve unabated, the ocean is bound to deoxygenate with poorly understood consequences for marine life. This is a source of concern, especially when considering that ocean deoxygenation is not occurring in isolation, but together with ocean acidification and ocean warming [Gruber, 2011].

[63] Therefore, it is crucial to continue with the oxygen sampling in order to document the evolution of oxygen in the North Atlantic and to further improve the spatial and temporal coverage of the observations. For the time being, the only approach yielding oxygen data with sufficient accuracy is ship-based, but oxygen sensors on Argo floats are continuously improving their performance, making them very valuable candidates for future monitoring of long-term oxygen trends [Johnson et al., 2009; Gruber et al., 2010].


[64] This work was supported by funds from ETH Zurich. We thank the numerous scientists and analysts responsible for the collection, and the analysis of the large amount of data that form the GLODAP, CARINA and WOD05 datasets. We are indebted to Mark Payne for helpful advice regarding the statistical analyses and for providing us with the necessary tools. We also thank Monika Rhein for sharing the data from the cruise MSM12/3. We are grateful to Toste Tanhua and the CARINA synthesis teams for their contributions and for sharing the many routines developed during this effort. Three reviewers provided insightful and careful comments on this paper. Their help is much appreciated.