Global Biogeochemical Cycles

Data-based estimates of suboxia, denitrification, and N2O production in the ocean and their sensitivities to dissolved O2


  • Daniele Bianchi,

    Corresponding author
    1. Department of Atmospheric and Ocean Sciences, Princeton University, Princeton, New Jersey, USA
    2. Department of Earth and Planetary Science, McGill University, Montreal, Quebec, Canada
      Corresponding Author: D. Bianchi, Department of Earth and Planetary Sciences, McGill University, 3450 University St., Montreal, QC H3A 2A7, Canada. (
    Search for more papers by this author
  • John P. Dunne,

    1. Geophysical Fluid Dynamics Laboratory, NOAA, Princeton, New Jersey, USA
    Search for more papers by this author
  • Jorge L. Sarmiento,

    1. Department of Atmospheric and Ocean Sciences, Princeton University, Princeton, New Jersey, USA
    Search for more papers by this author
  • Eric D. Galbraith

    1. Department of Earth and Planetary Science, McGill University, Montreal, Quebec, Canada
    Search for more papers by this author

Corresponding Author: D. Bianchi, Department of Earth and Planetary Sciences, McGill University, 3450 University St., Montreal, QC H3A 2A7, Canada. (


[1] Oxygen minimum zones (OMZs) are major sites of fixed nitrogen removal from the open ocean. However, commonly used gridded data sets such as the World Ocean Atlas (WOA) tend to overestimate the concentration of O2 compared to measurements in grids where O2 falls in the suboxic range (O2 < 2–10 mmol m−3), thereby underestimating the extent of O2 depletion in OMZs. We evaluate the distribution of the OMZs by (1) mapping high-quality oxygen measurements from the WOCE program, and (2) by applying an empirical correction to the gridded WOA based on in situ observations. The resulting suboxic volumes are a factor 3 larger than in the uncorrected gridded WOA. We combine the new oxygen data sets with estimates of global export and simple models of remineralization to estimate global denitrification and N2O production. We obtain a removal of fixed nitrogen of 70 ± 50 Tg year−1 in the open ocean and 198 ± 64 Tg year−1 in the sediments, and a global N2O production of 6.2 ± 3.2 Tg year−1. Our results (1) reconcile water column denitrification rates based on global oxygen distributions with previous estimates based on nitrogen isotopes, (2) revise existing estimates of sediment denitrification down by 1/3d through the use of spatially explicit fluxes, and (3) provide independent evidence supporting the idea of a historically balanced oceanic nitrogen cycle. These estimates are most sensitive to uncertainties in the global export production, the oxygen threshold for suboxic processes, and the efficiency of particle respiration under suboxic conditions. Ocean deoxygenation, an expected response to anthropogenic climate change, could increase denitrification by 14 Tg year−1 of nitrogen per 1 mmol m−3 of oxygen reduction if uniformly distributed, while leaving N2O production relatively unchanged.

1. Introduction

[2] Nitrogen is a major constituent of living organisms, and a limiting macronutrient over vast regions of the surface ocean. Molecular nitrogen (N2) is not readily usable by most marine organisms, and must be fixed by specialized prokaryotes before being incorporated into biomass. Nitrogen fixation by autotrophic diazotroph communities is the major oceanic source of biologically available nitrogen in the open ocean [Galloway et al., 2004]. Conversely, denitrification processes in the water column and sediments, including canonical denitrification and annamox, represent the main sink of nitrogen from fixed forms to N2 [Brandes et al., 2007]. The removal of fixed nitrogen in the open ocean is tightly linked to the oceanic oxygen distribution. The activity of denitrifying bacterial communities is strongly inhibited by oxygen concentrations at even a few percent of saturation levels [Codispoti et al., 2001]. As a consequence, fixed N losses are observed only in the most oxygen-depleted parts of the water column, in the suboxic core of the major oxygen minimum zones (OMZs).

[3] The oxygen distribution further affects the cycling of nitrogen by modulating the production of oceanic nitrous oxide (N2O). A byproduct of nitrification and denitrification pathways, N2O is a powerful greenhouse gas that affects the Earth's energy balance and climate. The ocean is responsible for approximately a quarter to a third of the preindustrial flux of N2O to the atmosphere [Galloway et al., 2004], with estimates between 4 and 10 Tg year−1 [Codispoti et al., 2001; Nevison et al., 2003; Galloway et al., 2004]. The production of N2O during nitrification occurs across a wide range of oxygen concentrations, and is enhanced at low oxygen levels [Nevison et al., 2003]. N2O is also a final product of nitrifier denitrification [Poth and Focht, 1985]. Additionally, denitrification processes consume N2O in oxygen-deficient waters [Nevison et al., 2003]. As a result of enhanced production under hypoxic conditions, generally assumed to occur at O2 < 60 − 120 mmol m−3 [Stramma et al., 2008], and consumption under suboxia (O2 < 2 − 10 mmol m−3 [Codispoti et al., 2005]), local minima of N2O appear at the core of intense OMZs, accompanied by local maxima in the surrounding oxycline regions [Codispoti and Christensen, 1985]. Overall, N2O sources appear to dominate over sinks in low-oxygen waters. As a consequence, open ocean OMZs account for 25–75% of the oceanic N2O production [Suntharalingam et al., 2000; Nevison et al., 2003].

[4] Despite recent developments in our understanding of the quality and magnitude of the pathways that make-up the oceanic nitrogen cycle (see Table 1), a number of open questions still remain [Brandes et al., 2007]. In particular, disagreement exists as to how close the marine nitrogen cycle is to a balance between N2 fixation and denitrification, and whether any imbalance can be detected [Codispoti and Christensen, 1985; Gruber and Sarmiento, 1997; Codispoti et al., 2001; Gruber, 2004; Codispoti, 2007]. The major processes that must be considered are the removal of fixed nitrogen due to denitrification and burial in sediments and inputs from N2 fixation, river run-off and atmospheric deposition. While river runoff and atmospheric deposition are relatively well constrained [Galloway et al., 2004; Codispoti, 2007], major sources of uncertainty center on the magnitude of sedimentary and pelagic denitrification and nitrogen fixation. Studies based on diagenetic models of benthic respiration calibrated against global observations suggest sedimentary denitrification as high as 285 to 300 Tg N year−1 [Middelburg et al., 1996; Codispoti et al., 2001, Codispoti, 2007]. Other studies propose a sedimentary sink of approximately 180–190 Tg N year−1 [Galloway et al., 2004; Gruber, 2004; Deutsch et al., 2004]. Most estimates of pelagic denitrification rely on the biogeochemical interpretation of excess dissolved phosphate (PO43 −) relative to nitrate (NO3) in the ocean. The range of estimates varies between 65 and 70 Tg N year−1 [Gruber, 2004; Deutsch et al., 2004] up to approximately 150 Tg N year−1 [Codispoti et al., 2001; Codispoti, 2007]. Major sources of disagreement involve the interpretation of water column N2 excess measurements, as well as the lack of consensus on the stoichiometry of denitrification reactions [Codispoti et al., 2001]. Global nitrogen fixation rates of between 85 and 141 Tg N year−1 have been proposed based on extrapolation of in situ fixation rates [Galloway et al., 2004], interpretation of observed excess of NO3 over PO43− and water mass age [Gruber and Sarmiento, 1997; Bange et al., 2005; Deutsch et al., 2001], and combinations of nutrient observations and ocean circulation models [Deutsch et al., 2007]. Unfortunately, the large uncertainty associated with the major pelagic sources and sinks of fixed nitrogen seems to hinder any clear conclusion on the balance of the marine nitrogen budget.

Table 1. Estimates of Oceanic Nitrogen Budgets (Tg N year−1)
 Galloway et al. [2004]Codispoti [2007]Gruber [2008]This Study
Pelagic and benthic fixation121135135 ± 50 
River input487680 ± 15 
Atmospheric deposition333050 ± 20 
Total sources202241265 ± 55 
Sediment denitrification206300180 ± 50198 ± 64
Pelagic denitrification11615065 ± 2070 ± 50
Burial162525 ± 10 
N2O production464 ± 26.2 ± 3.2
Total sinks342481275 ± 55 
Total budget−140−240−10 ± 80 

[5] In this study we combine global syntheses of organic matter export [Dunne et al., 2007] and new oxygen data sets with simple models for water column and benthic remineralization [Martin et al., 1987; Nevison et al., 2003; Middelburg et al., 1996] to provide an independent estimate of the pelagic and sedimentary denitrification and N2O production. We place particular emphasis on reducing the uncertainties related to (1) the low-oxygen range of the global oxygen distribution, and (2) the value and range of the parameters that are thought to regulate remineralization processes. Our approach relies on applying a set of algorithms of biogeochemical activity to geographically explicit export fluxes and oxygen fields, which is similar to what is commonly done in global models of ocean biogeochemistry. Therefore our results have implications for ongoing efforts to develop comprehensive models of the Earth system.

[6] The rest of the paper is organized as follows. Section 2 discusses the particle export and oxygen data sets used for the calculations. Section 3 describes the rationale used to parameterize remineralization processes in the water column and sediments. Section 4 presents the results of the calculations and the analysis of the sensitivity and errors associated with the data sets and parameters. Section 5 discusses the implication of our results and concludes the paper.

2. Data Sets Used in This Study

[7] We employ the global estimate of particulate organic carbon (POC) export from the analysis of Dunne et al. [2007]. The flux of POC that sinks out of the euphotic zone is obtained by combining estimates of primary production with an empirical equation for the ratio of POC export to primary production developed by Dunne et al. [2005]. Primary production is estimated from satellite chlorophyll, temperature and photosynthetic available radiation using the algorithms of Behrenfeld and Falkowski [1997], Carr [2001], and Marra et al. [2003].

[8] We use in situ oxygen measurements from the GLODAP data set [Key et al., 2004] as well as the gridded global oxygen data set of the World Ocean Atlas 2005 (WOA05) [Garcia et al., 2006]. The GLODAP data set consists of a compilation of high-quality measurements collected mostly during the 1990s by the World Ocean Circulation Experiment. The WOA05 is a gridded data set obtained by merging and interpolating a large number of measurements from a variety of observational programs from the last decades, including WOCE. WOA05 provides a climatological view of the ocean oxygen distribution, and is extremely valuable for estimating global and basin scale integrated properties and for model-data intercomparisons. However, growing evidence suggests that WOA05 underestimates the volume of suboxic waters in the major OMZs [Fuenzalida et al., 2009]. Suboxic processes are generally assumed to dominate at oxygen concentrations below 2–10 mmol m−3 [Morrison et al., 1999; Codispoti et al., 2001, 2005; Bange et al., 2005]. This is unfortunately near the level of detection, and oxygen biases of the order of just a few mmol m−3 have the potential to significantly affect our interpretation of the distribution of suboxic conditions.

[9] We assess the accuracy of the distribution of low-oxygen waters in the gridded WOA05 product by sampling the WOA05 at the location and month of GLODAP measurements. Figure 1a shows a scatterplot of all GLODAP measurements versus the sampled WOA05. A scatter of the data around the 1:1 line would be expected due to the interpolation procedure and the larger data set used in the WOA05. However, we find that the gridded data appear to consistently overestimate in situ measurements at low oxygen. This is evident in measurement from all major OMZs. The mean of the oxygen residuals (calculated as difference between gridded and in situ oxygen) is approximately 6.4 mmol m−3 in the lower hypoxic range (0–60 mmol m−3), and around 80% of the sampled gridded data set values overestimate in situ oxygen.

Figure 1.

Scatterplot of in situ GLODAP versus gridded oxygen data. The gridded data sets were sampled at the location and month of in situ measurements by using linear interpolation. Only values below 20 mmol m−3 are shown. The red line is the 1:1 line. The statistics shown are means and standard deviations of the oxygen biases (gridded minus in situ oxygen) in the hypoxic range (O2 < 60 mmol m−3), and f is the fraction of data points that fall above the 1:1 line (biases > 0). The data sets shown are (a) WOA05, (b) objective analysis of GLODAP measurements following the mapping scheme by Gandin [1963], (c) objective analysis of GLODAP measurements following the mapping scheme by Barnes [1964], and (d) WOA05 with linear correction based on in situ GLODAP measurements.

[10] The disagreement between WOA05 and GLODAP may arise from a combination of factors such as: positive biases in early measurements, interpolation artifacts, smoothing during objective mapping, and the effects of variability in ocean circulation. The measurements included in the WOA05 were obtained with different analytical methods. It is known that early measurements from suboxic waters can be biased high by up to 5 mmol m−3 [Codispoti and Christensen, 1985; Naqvi et al., 2010]. On the other hand, analytical methods adopted in recent years appear to have biases and precision better than approximately 2 mmol m−3 in the suboxic range [Morrison et al., 1999].

[11] In addition to the analytical biases, interpolation techniques can introduce artifacts that can be particularly relevant in low oxygen regions. An interpolated value between undetectable (zero) and detectable (nonzero) will have nonzero oxygen at all intermediate values, whereas contiguous anoxic regions in the ocean can have zero oxygen over significant volumes. Objective mapping procedures tend to further truncate extrema. Therefore, interpolated values may overestimate oxygen near the zero-oxygen limit. Biases can also be related to variability in oxygen records. Observational and model-based studies documented substantial variability in the oxygen content of the ocean on interannual to multidecadal timescales, resulting from a combination of circulation and biological changes [Garcia et al., 2005; Deutsch et al., 2011]. In addition, variability on shorter timescales, for example due to internal wave and eddy activity, can affect the oxygen distribution particularly in the upper ocean. Vertical displacements of isopycnal surfaces in regions of strong oxygen gradients can result in large changes in the oxygen concentrations at a given depth. This variability, when averaged to obtain climatological means at fixed depth levels, tends to produce positive biases when oxygen profiles show negative curvatures (e.g., around oxygen minima).

[12] To address the biases in the gridded WOA05 in the suboxic and hypoxic ranges, we produced three additional gridded oxygen data sets focusing on the consistency between mapped and in situ oxygen within the major OMZs. All three methods implicitly assume that temporal variability over the measurement period (recent decades) has been small, and that all data represent a steady mean state. We generated the first two data sets (referred to as objmap-1 and objmap-2) from the in situ GLODAP measurements by using a mapping procedure similar to the one described by Key et al. [2004]. We first interpolated the oxygen profiles to 60 neutral density levels between 20 and 28.5 kg m−3 (using the neutral density definition of Jackett and McDougall [1997]). We followed with a horizontal mapping on a 1° resolution grid, using the objective analysis technique developed by Gandin [1963], as described by Sarmiento et al. [1982] for the data set objmap-1, and the iterative interpolation scheme developed by Barnes [1964] for the data set objmap-2. Finally, we converted the mapped oxygen data sets from neutral density to depth space by using annual mean climatological densities from WOA05. We obtained the third data set by applying an empirical correction to the gridded WOA05 based on in situ GLODAP measurements.

[13] Similar to Key et al. [2004], we adopted two different correlation length scales for the zonal and meridional directions. We adopted constant zonal and meridional length scales of 10 and 20 degrees and 15 and 7 degrees for the objmap-1 and objmap-2 data sets respectively. These length scales were chosen among different combinations, and appeared to be best suited to retain the signal from low-oxygen regions while keeping interpolation artifacts acceptably small. The horizontal mapping scheme used to generate the second data set (objmap-2) is similar to the one adopted by Garcia et al. [2006] to produce the gridded WOA05. Garcia et al. [2006] adopted a three-iteration scheme with variable correlation length scales, and interpolated a larger number of oxygen measurements averaged by 1° boxes. Our approach uses a smaller sample of the WOA05 measurements, constant correlation length scales, but a higher number (10) of iterations.

[14] The main difference between the mapping procedure adopted in this study and the ones adopted by Key et al. [2004] and Garcia et al. [2006], lies in the choice of neutral density instead of depth as vertical coordinate for the horizontal maps. A substantial contribution to oxygen variability at a given location and depth on sub-seasonal to interannual timescale is due to vertical and horizontal displacements of density layers due to physical processes ranging from internal waves and eddies to climate-driven variations [Garcia et al., 2005; Deutsch et al., 2011]. By choosing neutral density as the vertical coordinate we minimize the effects of density structure changes that would introduce significant oxygen variability in depth space.

[15] For the third data set (corrected-WOA05), we apply an empirical correction to the gridded WOA05 oxygen based on in situ measurements from GLODAP. We calculate the corrected oxygen by regressing the GLODAP oxygen onto WOA05 values sampled at the location and month of in situ-measurements, using a linear fit over the full range of oxygen values (O2corrected = 1.009 O2WOA05 − 2.523 mmol m−3; negative values were set to zero). This approach allows us to maintain the geographical structure of the gridded WOA05 data set while reducing the biases in the low-oxygen range relative to recent high-quality oxygen measurements.

[16] Figures 1b1d show scatterplots of the oxygen values from the gridded data sets compared to in situ GLODAP measurements. Whereas the WOA05 overestimates in situ measurements in the low-oxygen range, the gridded data sets obtained by objective mapping and by correcting the WOA05 approximate in situ oxygen values to a better degree, and the new data sets scatter more uniformly around the 1:1 lines. Mean positive oxygen biases remain; however they decrease from approximately 6.4 mmol m−3 to 3.8–4.2 mmol m−3. Figures 2 and 3 show the volumetric distribution of low oxygen waters for the gridded data sets, and the thickness of the suboxic water column (O2 < 10 mmol m−3). WOA05 shows a substantially lower volume of suboxic waters compared to the new data sets. Relatively more oxygenated waters show similar volumes for all the data sets. The volumes of low oxygen waters from the three main OMZs in the Indian and Pacific Oceans are summarized in Table 2. The volume of waters with oxygen below 20 mmol m−3 is approximately 1.4 times larger in the new gridded data sets compared to WOA05; the discrepancy increases to approximately 2.5–4 times when water volumes with oxygen below 5 mmol m−3 are considered (Table 2).

Figure 2.

Volumetric distribution of oceanic oxygen from different gridded data sets discussed in section 2. Water volumes have been binned according to oxygen concentration in 5 mmol m−3 wide bins, starting from 0 mmol m−3. The data sets shown are WOA05, objective analysis of GLODAP measurements following the mapping scheme by Gandin [1963], objective analysis of GLODAP measurements following the mapping scheme by Barnes [1964], and WOA05 with linear correction based on in situ GLODAP measurements.

Figure 3.

Thickness (meters) of the water column with oxygen lower than 10 mmol m−3. The data sets shown are (a) monthly WOA05, (b) objective analysis of GLODAP measurements following the mapping scheme by Gandin [1963], (c) objective analysis of GLODAP measurements following the mapping scheme by Barnes [1964], and (d) monthly WOA05 with linear correction based on in situ GLODAP measurements.

Table 2. Volume (106 km3) of Suboxic Waters for the Major OMZs in the Indian and Pacific Oceans for Two Oxygen Limits (20 and 5 mmol m−3)a
  North PacificSouth PacificIndian OceanGlobal
  • a

    The data sets shown are gridded WOA05, GLODAP data mapped using the procedure described by Gandin [1963], GLODAP data mapped using the procedure described by Barnes [1964], and gridded WOA05 with an empirical linear correction based on GLODAP in situ measurements.

WOA05O2 < 20 mmol m−
Objective mapping 1
Objective mapping 2
Corrected WOA05
WOA05O2 < 5 mmol m−30.670.040.341.05
Objective mapping 1 2.370.450.863.68
Objective mapping 2 2.410.611.114.13
Corrected WOA05 1.640.130.682.45

[17] The two interpolated data sets generated in this study are in good agreement with each other, but show overall larger suboxic volumes than the corrected WOA05. On a regional basis, the difference is most evident in the South East Tropical Pacific OMZ, where the corrected WOA05 shows a substantially smaller volume of suboxic waters than the data sets obtained by objective analysis. In this region, OMZ volumes from the two objectively mapped data sets are in good agreement with the recent independent assessment by Fuenzalida et al. [2009], based on mapping OMZ upper and lower boundaries. The authors find that the volume of waters with oxygen lower than 20 mmol m−3 is between 2.28 and 2.78 · 106 km3, close to our estimate of 2.3 to 2.5 · 106 km3 from the objective mapping results, but somewhat higher than the estimate of 1.9 · 106 km3 from the corrected WOA05.

[18] In concluding, we note that by adopting an interpolation procedure that uses neutral density as a vertical framework we addressed only the variability in oxygen distribution that arises from variability in the density structure of the ocean. Variability related to biological export of organic matter and respiration plays an additional role in controlling seasonal to interannual changes in the position and intensity of OMZ [Deutsch et al., 2011]. With the inclusion of high quality measurements from ongoing observational efforts targeted at measuring oxygen near the detection limit within major OMZs, we are optimistic that the biases and uncertainties that characterize current data sets will be further reduced.

3. Models of Respiration, Denitrification and N2O Production

[19] With this revised view of the oxygen distribution in open ocean OMZs, we turn to investigate respiration, denitrification and N2O production on a global scale. We start from the average estimate of export production from Dunne et al. [2007], and propagate it to the ocean interior by imposing an oxygen-dependent profile of remineralization. We choose the simple and widely used approach of a power law dependence of POC flux with depth, according to the “Martin curve” [Martin et al., 1987]:

display math

Here ΦPOC(z) and Φ0POC are the POC fluxes at depths z and z0 (75 m, see following discussion) respectively, and b is the Martin attenuation coefficient. Ocean sinking particles are made up of a heterogeneous mix of mineral and organic components with different physical and chemical properties that spans a continuous spectrum of sizes. The degradation of particles in the water column involves complex and relatively poorly understood physical and biological processes, including dissolution and coalescence, and close interactions with the marine biosphere [Stemmann et al., 2004]. The Martin curve is based on an empirical fit to observations of POC flux in the open ocean and does not imply any a priori mechanism of particle sinking and degradation. Alternative approaches to the Martin curve have been proposed in recent years, based on mechanistic models of the association of POC with ballast minerals [Armstrong et al., 2001; Klaas and Archer, 2002; Lutz et al., 2002]. Dunne et al. [2007] explored a suite of remineralization schemes that include the effect of different ballast fractions, including CaCO3, opal and lithogenic material, and found that adding a ballast component improved the regional patterns of bottom fluxes in the deep ocean compared to observations. While more sophisticated models of particle flux than the Martin curve can be justified on theoretical and observational grounds, they require a larger number of parameters that are poorly constrained, and the use of additional observational data sets to estimate the mineral fractions. Here, we prefer to avoid the additional uncertainty associated with these parameterizations by adopting the simpler empirical Martin curve approach, supported by a large number of observational studies [Berelson, 2001], and we analyze the sensitivity to this reduced set of parameters.

[20] Following Dunne et al. [2007], we assume that the export starts from an average constant depth z0 = 75 m, a value close to the average 1% light level usually used as a proxy for the depth of the euphotic zone. Substantial uncertainty exists on the validity of a single remineralization profile for different oceanographic provinces, and on the numerical value of the attenuation coefficient itself [Lutz et al., 2002; Berelson, 2001]. A larger number of observations than currently available are needed to detect statistically significant regional variations in particle flux attenuation with depth [Primeau, 2006]. However, a growing number of studies show a significant reduction in the attenuation of particulate organic matter fluxes in severely oxygen-depleted regions compared to the rest of the ocean [Devol and Hartnett, 2001; van Mooy et al., 2002]. Early analyses indicated attenuation coefficients in suboxic waters as low as 0.32 [Martin et al., 1987]. Recent studies suggest values ranging from b = 0. 36 to b = 0. 40 [Hartnett and Devol, 2003; van Mooy et al., 2002].

[21] We adopt a constant attenuation coefficient b = 0. 8 for oxygenated waters (O2 > O2subox). This value is smaller than the attenuation coefficient first proposed by Martin et al. [1987] (b = 0. 86), but is consistent with the synthesis of U.S. JGOFS flux data by Berelson [2001] (b = 0. 82 ± 0. 16) and with the analysis by Primeau [2006] (b = 0. 70 ± 0. 08). For suboxic waters (O2 < O2subox) we reduce the attenuation coefficient to a value bsubox = 0. 36, in agreement with the results by Hartnett and Devol [2003] and van Mooy et al. [2002]. Due to the limited observational constraints, we assume a simple step-function transition between oxic and suboxic conditions.

[22] It is not clear at what oxygen concentration the transition to suboxic conditions occurs. Evidence based on the accumulation of nitrite (NO2) suggests that denitrification is strongest at oxygen concentrations below approximately 2.5 mmol m−3 [Morrison et al., 1999; Codispoti et al., 2001; Bulow et al., 2010], whereas the NO3 deficit relative to PO43− could support values a few mmol m−3 higher [Codispoti et al., 2001]. Measurements exist suggesting N2O removal by denitrification at oxygen concentrations up to 40 mmol m−3 [Farías et al., 2009]. Yet, most evidence indicates that denitrification is strongly inhibited by oxygen concentrations above 5 mmol m−3 [Codispoti et al., 2001; Bulow et al., 2010]. The occurrence of water column denitrification in the presence of detectable oxygen is puzzling on biochemical and thermodynamical grounds, but could be supported by the existence of anoxic microenvironments within sinking aggregates [Codispoti et al., 2001]. In this study, we adopt the value O2subox = 4. 5 mmol m−3, corresponding to approximately 0.1 ml l−1, as the threshold for water column denitrification, and we explore the sensitivity to a range of concentration limits. We assume that above the suboxic threshold, respiration uses oxygen as the main oxidant. Furthermore, we assume that below the suboxic limit nitrate becomes the main oxidant, and that nitrogen reduction, including of nitrous oxide, proceeds until molecular nitrogen is formed.

[23] With these assumptions, we calculate pelagic denitrification, i.e., the removal of fixed N by denitrification, from the divergence of the POC flux within the suboxic water column, assuming a constant stoichiometric ratio rN:Cdenitr. A substantial uncertainty exists in the overall stoichiometry of denitrification processes. Early studies suggested values for rN:Cdenitr generally around or below 1. For example, Richards [1965] provide two different denitrification reactions corresponding to rN:Cdenitr = 85:106 and rN:Cdenitr = 110:106 respectively. Higher values for rN:Cdenitr have been proposed recently. Van Mooy et al. [2002] suggest rN:Cdenitr = 132.4:106, under the assumption that microbial communities responsible for denitrification preferentially utilize the amino acid component of particles. Here we adopt the stoichiometric ratio rN:Cdenitr = 120:106 of Gruber and Sarmiento [1997].

[24] We assess sedimentary denitrification by applying the metamodel developed by Middelburg et al. [1996]. The metamodel is based on a mechanistic diagenetic model of benthic biogeochemical processes calibrated against global observations, and predicts benthic denitrification as a function of the POC flux at the sediment-water interface. According to the metamodel, sediment denitrification (Densed) can be calculated from:

display math

where ΦbottomPOC is the flux of particulate organic carbon (mmol C m−2d−1) to the sediments.

[25] We estimate the production of nitrous oxide (N2O) from nitrification, by using the parameterization of Nevison et al. [2003]:

display math

Following Nevison et al. [2003], inline image, inline image, rN:Cnitrif = 16:106. As noted in the introduction, N2O is an obligate intermediate of denitrification reactions, and is further utilized as the oxidant as denitrification proceeds to reduce nitrogen to the molecular form. It is well established that denitrification in suboxic waters represent a sink of N2O [Codispoti and Christensen, 1985; Suntharalingam et al., 2000]. The magnitude of the sink for N2O produced outside the suboxic region ultimately depends on the exact sensitivity of nitrification and denitrification to the oxygen concentration, and on the magnitude of N2O fluxes across strong N2O and O2 gradients. Due to these uncertainties, we adopt the approach of Nevison et al. [2003] and set the N2O production term to zero below the suboxic threshold O2subox (here assumed to be 4.5 mmol m−3).

[26] A summary of the numerical values for the model parameters and the associated uncertainties is presented in Table 4.

4. Results

4.1. Global and Regional Estimates

[27] Figures 4a4d show maps of export, vertically integrated pelagic denitrification, benthic denitrification and vertically integrated N2O production. Regional and global integrals are summarized in Table 3. The uncertainty was estimated by running a Monte Carlo type sensitivity analysis. Approximately 10,000 runs were performed by simultaneously varying the model parameters, assuming a gamma distribution for each parameter (to prevent non-physical negative values) with mean and standard deviations from the literature (Table 4). The globally integrated POC export from Dunne et al. [2007] is 9.4 ± 3.5 Pg C year−1. Approximately 7.2 ± 2.7 Pg C year−1, 77% of export, is respired in the water column. The remaining 2.2 ± 0.8 Pg C year−1, reach the seafloor and are mostly respired in the sediments; only a small part is eventually buried [Dunne et al., 2007]. Due to a combination of high productivity and export, and more limited water column remineralization, shallow depths dominate sedimentary fluxes. Approximately 70% of the POC sedimentation (1.5 Pg C year−1) takes place in shelf regions (defined operationally as bottom depths shallower than 150 m).

Figure 4.

(a) Export from the euphotic zone from Dunne et al. [2007], (b) vertically integrated pelagic denitrification, (c) vertically integrated N2O production, and (d) benthic denitrification. Contouring is on a log10 scale. Averages from the three oxygen data sets described in section 2 are shown.

Table 3. Regional and Global Estimates of POC Flux to the Sediments, Pelagic Denitrification, Benthic Denitrification (Shelf Values for z < 150 m in Parentheses) and N2O Productiona
 POC Flux to the Sediments (Pg C year−1)Pelagic Denitrification (Tg N year−1)Benthic Denitrification (Tg N year−1)Pelagic N2O Production (Tg N year−1)
  • a

    Results are averages obtained by using three new oxygen data sets discussed in section 2. For comparison, global pelagic denitrification, benthic denitrification, and N2O production obtained with the uncorrected WOA05 data sets are 24, 187 and 6.0 Tg N year−1 respectively, as discussed in section 4.1. Errors were estimated with a Monte Carlo, as described in section 4.1.

North Pacific0.51 (0.37)21.546.5 (30.7)2.05
South Pacific0.31 (0.22)17.625.9 (17.3)1.29
Indian Ocean0.22 (0.13)31.221.1 (11.0)1.70
North Atlantic0.54 (0.40)47.0 (29.2)0.48
South Atlantic0.27 (0.20)20.8 (13.1)0.49
Southern Ocean (south of 50°S)0.06 (0.01)5.9 (1.6)0.14
Arctic Ocean0.27 (0.21)30.6 (21.4)0.04
Global2.2 ± 0.8 (1.5)70 ± 50198 ± 64 (124)6.2 ± 3.2
Table 4. Sensitivity of the Globally Integrated Pelagic Respiration, Denitrification and N2O Production to the Remineralization Model Parametersa
 Mean Value and Error Estimate Adopted for This StudyReferencesbSensitivity
Benthic POC Flux (Pg yr−1)Pelagic Denitrification (Tg yr−1)Benthic Denitrification (Tg yr−1)N2O Production (Tg yr−1)
Φ0POC9.4 ± 3.5 Pg yr−110.23 inline image7.5 inline image14.8 inline image0.66 inline image
z075 ± 20 m20.09 inline image7.5 inline image10.3 inline image0.34 inline image
b0.8 ± 0.13, 4, 5−0.15 inline image−5.4 inline image−19.3 inline image−0.014 inline image
bsubox0.36 ± 0.16, 7−0.02 inline image14.5 inline image−0.53 inline image−0.042 inline image
O2subox4.5 ± 2.0 mmol m−38, 9, 10, 11−0.002 inline image14.4 inline image0.34 inline image−0.35 inline image
rN:Cdenitr(120:106) ± (10:106)6, 12, 135.8 inline image16.5 inline image
rN:Cnitrif(16:106) ± (1:106)140.39 inline image
a10.26 ± 0.06150.28 inline image
a2− 4 · 10−4 ± 10−4150.25 inline image
ΔO2 −0.002 inline image−13.8 inline image−0.33 inline image−0.01 inline image

[28] We obtain a global N2O production of 6.2 ± 3.2 Tg N year−1, in agreement with the estimate of 5.8 ± 2.0 Tg N year−1 by Nevison et al. [2003]. N2O sources correlate strongly with the patterns of export production and water column respiration, reflecting N2O formation as a byproduct of nitrification. N2O production is further modulated by the oxygen distribution, as evident in the regional estimates in Table 3.

[29] Our results show significant denitrification in the cores of the Pacific and Indian Ocean OMZs, with a global pelagic denitrification of 70 ± 50 Tg N year−1, in agreement with the global estimates of 65 and 70 Tg N year−1 by Gruber [2004] and Deutsch et al. [2004], but about half the 150 Tg N year−1 proposed by Codispoti et al. [2001]. The Indian Ocean shows a large fixed nitrogen sink, and is responsible for about 31 Tg N year−1 denitrification, in line with 21 to 34 Tg N year−1 reported by Bange et al. [2005] and 21 ± 7 Tg N year−1 by Howell et al. [1997], but slightly below the estimates of 39 ± 13 and 41 ± 18 Tg N year−1 by Bulow et al. [2010] and Devol et al. [2006]. Pelagic denitrification in the Pacific Ocean is 39.1 Tg N year−1, with the Northern and Southern Eastern Tropical Pacific (NETP and SETP) regions accounting for 21.5 and 17.6 Tg N year−1 respectively. The fixed nitrogen sink in the NETP and SETP is similar to the estimates of 22 and 26 Tg N year−1 by Deutsch et al. [2001]. We do not see significant pelagic denitrification in the Atlantic Ocean, where oxygen concentrations are above the suboxic limit almost everywhere in open waters.

[30] While the estimates of open ocean respiration and N2O production appear particularly robust across the different oxygen data sets used in this study, some discrepancies can be found in the estimates of pelagic denitrification. The corrected WOA05 produces the smallest estimate (approximately 51 Tg N year−1). The data sets from objective mapping of GLODAP measurement with the Key et al. [2004] and Barnes [1964] schemes (objmap-1 and objmap-2) produce around 72 and 87 Tg N year−1 respectively. Differences in global open ocean denitrification arise from a combination of different suboxic water volumes (Table 2), and location in the three data sets.

[31] Overall, our analysis stresses the critical importance of capturing the lowest range of the oxygen distribution for estimating pelagic denitrification. When we perform the same calculations without correcting the gridded WOA05 for the biases in the suboxic range, we obtain less than 24 Tg N year−1 of global open ocean denitrification. The denitrification is even smaller if the annual climatological WOA05 data set is used instead of monthly climatologies (denitrification < 5 Tg N year−1) due to the biases introduced by time averaging in the low-oxygen regions. In contrast, N2O production shows a minimal sensitivity to the oxygen data set used. When the uncorrected WOA05 data set is used, we obtain approximately 6.0 Tg N year−1 N2O production, very close to the results from the new data sets.

[32] We obtain a global sedimentary denitrification of 198 ± 64 Tg N year−1. This result is relatively insensitive to the oxygen data set used, and is slightly higher than the estimate obtained with the uncorrected WOA05 data set (187 Tg N year−1). The largest share of benthic denitrification occurs in shallow regions, where the POC sedimentary flux is highest. Figures 5a and 5b show the cumulative depth distribution of sedimentary denitrification broken down by basin. Our global estimate of benthic denitrification is smaller than the estimate by Middelburg et al. [1996], 230–285 Tg N year−1, with most of the difference originating from deep-ocean regions. Middelburg et al. [1996] obtain approximately 100 Tg N year−1 denitrification in shelf sediments, similar to our estimate of 124 Tg N year−1 (Table 3). The disagreement is due to a combination of factors. Middelburg et al. [1996] calculate sedimentary POC fluxes as a function of depth alone, whereas we employ a geographically varying export data set based on satellite-derived chlorophyll observations, combined with a model of water column remineralization. The globally integrated sedimentary POC flux from Middelburg et al. [1996] is approximately 3 Pg C year−1, compared to our estimate of approximately 2 Pg C year−1. The depth-dependent organic carbon flux parameterization adopted by Middelburg et al. [1996], results in the same sedimentation rate in geographically distant provinces such as continental slopes and mid-depth regions away from continental margins. This is unjustified in the light of satellite-based export estimates (Figure 4a) and sediment data syntheses [Dunne et al., 2007] showing large regional gradients that cannot be captured by variations in water column depth alone.

Figure 5.

Cumulative benthic denitrification integrated from the surface to a given depth, divided by basin. (a) Depths between 0 and 250 m and (b) entire water column.

[33] Recent studies pointed out that the model by Middelburg et al. [1996] might overestimate denitrification in coastal and shallow environments [Fennel et al., 2009]. Differences could arise from the lack of processes such as sediment transport, re-suspension and respiration in Middelburg et al.'s [1996] model. Additional differences may relate to the lack of explicit representation of riverine organic matter inputs to costal waters. Our denitrification estimate for the North Atlantic shelf (29 Tg N year−1 for depths shallower than 200 m) is larger than the estimate of 20 Tg N year−1 by Seitzinger and Giblin [1996], derived from the correlation between benthic denitrification and water-sediment oxygen fluxes, but is in line with the model assessment of 32.2 Tg N year−1 by Fennel et al. [2006], which is based on the same oxygen-denitrification relationship. Yet, on a global scale, we obtain less denitrification in the shelves than the 250 Tg N year−1 proposed by Seitzinger et al. [2006]. These persistent differences highlight the difficulty of narrowing the uncertainty to better than ±30%, given the available constraints.

4.2. Sensitivity and Error Estimates

[34] The relative simplicity of the remineralization scheme adopted in this study allows us to perform an extensive analysis of the sensitivity of remineralization and N2O production to the model parameters. We run our remineralization scheme by varying one parameter at a time across a range of values inferred from the literature, while fixing the other parameters to their control values. The sensitivities are calculated as partial derivatives (using a central finite difference approximation) of the globally integrated quantities with respect to the parameter considered. For example, to estimate the sensitivity of water column denitrification (Denitr) to the attenuation coefficient (b), we use:

display math

We average the results from the three gridded oxygen data sets to obtain the sensitivities summarized in Table 4. Figure 6 shows the relative sensitivities of the global estimates of denitrification and N2O production to changes in the model parameters equal to one standard deviation.

Figure 6.

Relative sensitivity of pelagic denitrification, benthic denitrification and N2O production to the model parameters. Sensitivities are calculated as fractional change in the global integrals for a change in each model parameter equal to one standard deviation (see Table 4). For example, the relative sensitivity of pelagic denitrification (Denitr) to changes in the global POC export (ΔΦ0POC) is calculated as inline image. Sensitivities equal to 0 indicate no change.

[35] The global export of POC from the surface is the main driver of oceanic remineralization. Changes in Φ0POC affect pelagic denitrification and N2O production in a linear way by the model construction. The depth of export (z0) is the second factor controlling the flux of POC entering the ocean interior. An increase in z0 shifts respiration deeper in the water column and enhances sedimentary fluxes and benthic denitrification. At the same time, water column denitrification and N2O production increase due to the larger respiration in the subsurface hypoxic and suboxic layers.

[36] In the interior, POC fluxes are controlled by the shape of the Martin curve. An increase in the attenuation coefficient b steepens the remineralization curve and moves respiration shallower. More POC is respired between the surface and the suboxic layers, and less reaches the denitrification region in the core of OMZs, as well as the sediments. The opposite is true for an increase in bsubox, which acts to enhance the degradation of POC within the suboxic layers, enhancing water column denitrification.

[37] The suboxic threshold O2subox has a strong influence on water column denitrification. The impact is less marked on N2O production, and negligible on sedimentary processes. As O2subox increases, suboxic waters expand both in area and thickness, enhancing denitrification. We observe an approximately linear dependence of total denitrification on O2subox, with a sensitivity of 14.4 Tg N year−1 (mmol m−3)−1.

[38] By design, N2O production shows a nonlinear dependence on O2subox, with a strong intensification for lower values of O2subox [Nevison et al., 2003]. This is related to the inverse dependence of N2O sources on oxygen concentration. N2O production increases by approximately 2.3 Tg N year−1 when O2subox is reduced from 10 to 2 mmol m−3, and by additional 1.7 Tg N year−1 when O2subox is reduced between 2 and 0.5 mmol m−3. This suggests that the oxygen concentration at which denitrification becomes a dominant sink for N2O is an important factor controlling the overall N2O source in the ocean. Additional uncertainty (∼1.7 Tg N year−1) is associated with the variability of the coefficient a1 in the parameterization by Nevison et al. [2003].

[39] The approximate contributions of each model parameter to the errors in the final estimates can be obtained by combining the sensitivities with the uncertainties in Table 4. For example, the uncertainty of water column denitrification associated with the uncertainty in the attenuation parameter b can be estimated with the following linear approximation:

display math

After the uncertainties associated with the global POC export, uncertainties in the parameters controlling suboxic processes, O2subox and bsubox, dominate the picture.

4.2.1. Sensitivity to the Oxygen State of the Ocean

[40] The hypothesis that anthropogenic climate change could lead to a decrease in the ocean oxygen due to ocean warming and increased stratification has been gaining substantial attention over the last decade [Sarmiento et al., 1998; Keeling et al., 2010]. Modeling studies predict average global ocean deoxygenation between 2 and 12 mmol m−3 over the next century [Keeling et al., 2010]. However, recent studies suggest different responses for hypoxic and suboxic waters. While hypoxic volumes are projected to expand, suboxic volumes might in fact decrease [Frölicher et al., 2009; Gnanadesikan et al., 2012]. In addition to model results, recent observational syntheses indicate an expansion of major OMZs in the Atlantic and Pacific Ocean during the past 50 years, above the fluctuations of decadal variability [Stramma et al., 2008]. Changes in the oxygen content of the ocean and in the extent of OMZs such as the ones recently suggested could have significant impacts on the oceanic nitrogen and carbon cycles and on marine ecosystems.

[41] We investigate the first order effect of changes in the oxygen state of the ocean on water column denitrification and N2O production by imposing a uniform change in the oxygen distribution (ΔO2). This approximates a uniform warming or cooling of the ocean, with no change in circulation or ecosystem behavior. For example, a uniform warming by 1°C could result in a solubility decrease of oxygen by ∼5 mmol m−3 [Deutsch et al., 2011]. The sensitivities to ΔO2 are shown in Figures 7a and 7b and summarized in Table 4. The sharp transition between fully oxic and suboxic conditions at ΔO2=O2subox results in the reduction of open ocean denitrification to near zero for oxygen increase above O2subox. Our model shows a nearly linear effect of oxygen changes on open ocean denitrification (Figure 7a), with sensitivities between − 12 and − 14Tg year−1 (mmol m−3)−1 for ΔO2 < 4 mmol m−3. The effects of a uniform oxygen decrease on N2O production are negligible (Figure 7b). This is due to two factors. First, our model assumes no net production of N2O within suboxic waters (O2 < O2subox). Second, the volume of hypoxic waters increase approximately linearly above the suboxic threshold. Therefore, a uniform decrease in oxygen acts to shift the location of N2O sources, without altering the relative contribution of waters with different oxygen content. For oxygen changes above 4 mmol m−3, N2O production starts decreasing at a rate between 0.3 and 0.2Tg N year−1 (mmol m−3)−1, reflecting the overall reduction of hypoxic waters.

Figure 7.

Sensitivity of (a) pelagic denitrification and (b) N2O production (Tg N year−1) to uniform changes (ΔO2, mmol m−3) in the ocean oxygen distribution. The solid line and the shading show the mean and standard deviation from a series of Monte Carlo simulations.

[42] It is important to note that changes in oxygen distribution are unlikely to be geographically uniform as assumed in this sensitivity analysis, and additional physical, biogeochemical and ecological processes (e.g., changes in circulation and productivity, feedbacks between denitrification, nitrogen fixation and export production) are likely to complicate the response of the nitrogen cycle to oxygen changes [Keeling et al., 2010; Frölicher et al., 2009; Gnanadesikan et al., 2012].

5. Discussion and Conclusions

[43] Substantial uncertainty still exists regarding the extent of open-ocean OMZs and their impact on nitrogen cycling. We showed that widely used gridded data sets of oxygen distribution have biases in the low oxygen range that lead to an underestimate of the extent of suboxic conditions in the ocean. We account for these biases by generating three data-based gridded maps that combine WOCE era oxygen measurements to produce an updated view of OMZs in line with recent independent estimates. The new gridded maps are used in conjunction with spatially explicit estimates of export production and a simple model of water column remineralization to estimate global denitrification and N2O production.

[44] We obtain 70 ± 50 Tg N year−1 of denitrification in the open ocean and 198 ± 64 Tg N year−1 in the sediments. These values are compatible with the isotopic constraints placed on the ratio of non-fractionating (sedimentary) to fractionating (water column) oceanic denitrification processes [Brandes and Devol, 2002], corrected for the dilution effect of Deutsch et al. [2004]. The ratio implied by our estimates, 2.8, is consistent with the value of approximately 2.7 proposed by Deutsch et al. [2004]. Our results are close to the lower range of existing estimates in the open ocean [Gruber, 2004; Galloway et al., 2004], and revise the benthic denitrification proposed by Middelburg et al. [1996] down by one third through the use of spatially varying sediment fluxes. Together with independent estimates of fixed nitrogen inputs to the oceans, our estimates provide independent evidence supporting the idea of a historically balanced oceanic nitrogen cycle [Gruber, 2004]. The N2O production estimate, 6.2 ± 3.2 Tg N year−1, is in agreement with previous studies, and appears robust with respect to the model parameterizations and oxygen data sets. A sensitivity analysis shows that ocean deoxygenation, could lead to an increase in denitrification of approximately 14 Tg N year−1 per mmol m−3 of oxygen decrease, while leaving N2O production relatively unchanged.

[45] The uncertainty in our estimate originates from uncertainties in the export production and oxygen data sets and in the parameterizations of remineralization processes. An accurate depiction of the suboxic range of oceanic oxygen is critical for capturing the magnitude and distribution of water column denitrification, as biases of a few mmol m−3 can significantly impact the final estimates. There are additional sources of error that we did not directly address in this study. These include both methodological uncertainties related to the validity of the parameterizations adopted in this study, and uncertainties related to the processes that influence export and remineralization in the ocean. Among these, uncertainty could arise from climate variability in export production and oxygen distribution, and biogeochemical processes such as cycling of dissolved organic matter, active transport by vertically migrating zooplankton, and feedbacks between denitrification, nitrogen fixation and particle export.

[46] The method used in this study relies on propagating the satellite-derived export production from the euphotic zone into the ocean interior using a simple scheme of particle remineralization. Our parameterizations are similar to the ones used in the current generation of ocean biogeochemistry models. The uncertainties identified in this study show that it is critical that models capture the extent, location, intensity and variability of the major suboxic regions, a challenging task that will require accurate observations and simulations of both circulation and biogeochemistry of OMZs. Particular attention should be placed on the development and testing of new parameterizations that provide a reliable representation of organic matter respiration under hypoxia and suboxia. To this end, we need to expand our quantitative understanding of the biogeochemical processes and thresholds that characterize the transition between oxic, hypoxic and suboxic regimes in the water column. We urge the development of global data sets of oceanic oxygen that focus on the faithful representation of the suboxic range. This will require assembling high quality measurements from the major OMZs and applying mapping procedures that reduce interpolation artifacts to a minimum.