We present and analyze a data set of subsurface N2O from a range of oceanic regions. Observed N2O concentrations are highest in the eastern tropical Pacific (ETP), intermediate in the northern Pacific and Indian Oceans, and relatively low in the Southern and Atlantic Oceans. Tongues of high N2O, which propagate along sigma surfaces, provide evidence that N2O from the ETP is exported widely. Correlation slopes of ΔN2O (the level above atmospheric equilibrium) versus apparent oxygen utilization (AOU) are found to be an unreliable gauge of the biological N2O yield per mole O2 consumed because the slopes are strongly influenced by mixing gradients. Most features of the subsurface data set are consistent with an N2O source dominated by nitrification, including the widespread, robust ΔN2O-AOU correlation and the lack of a widespread anticorrelation between ΔN2O and N*. In addition, ΔN2O/NO3− ratios tend to increase with decreasing O2 in a manner consistent with laboratory studies of nitrifying bacteria. The sensitivity of the nitrifier N2O/NO3− yield to O2 can explain much of the variability in ΔN2O/AOU observed in the ocean. A parameterization is derived for the instantaneous production of N2O per mole O2 consumed as a nonlinear function of O2 and depth. The parameterization is based on laboratory and oceanic data and is designed for use in ocean biogeochemistry models. It is coupled to a global dissolved O2 climatology and ocean carbon model output to estimate a total oceanic N2O inventory of 610–840 Tg N and a global production rate of ∼5.8 ± 2 Tg N/y.
 Nitrous oxide (N2O) is a trace gas that exerts a strong influence on climate and atmospheric chemistry. It is the third most important natural long-lived greenhouse gas after CO2 and CH4, and also affects climate through its role in stratospheric ozone chemistry [Crutzen, 1974; Nevison and Holland, 1997; Prather et al., 2001]. N2O is produced mainly by microbial activity at Earth's surface and destroyed primarily by photochemistry in the stratosphere. Atmospheric N2O has increased slowly since the late nineteenth century, and at an accelerated rate of ∼0.25%/year in the latter part of the twentieth century [Weiss, 1981; Machida et al., 1995; Battle et al., 1996]. The cause of this increase most likely involves an enhanced microbial N2O source associated with human agriculture [Bouwman et al., 1995; Nevison and Holland, 1997; Kroeze et al., 1999]. The natural sources of N2O include oceans, sediments, and terrestrial soils. Global inventories of these sources are still fairly uncertain, but estimates suggest that the ocean contributes about one fourth to one third of the natural N2O source [Nevison et al., 1995; Prather et al., 2001].
 N2O is produced as an obligate intermediate of microbial denitrification and as a trace by-product of microbial nitrification. These processes are inhibited by O2 and light, such that oceanic N2O production takes place primarily in subsurface waters [Horrigan et al., 1981]. Denitrification, the reduction of NO3− to N2, occurs under highly suboxic to anoxic conditions and results in the loss of fixed nitrogen from the ocean to the atmosphere. Oceanic denitrification is thought to be restricted primarily to continental margin sediments and to a few known O2 deficient regions in the eastern tropical Pacific and the Arabian Sea. Nitrification, a two-step process involving the oxidation of NH4+ to NO3−, is more widespread. It occurs optimally under aerobic conditions in association with the recycling of fixed nitrogen by primary producers and decomposers. In suboxic regions and in sediments, nitrification and denitrification are often closely coupled [Ward, 1986; Ward et al., 1989; Seitzinger, 1988; Capone, 1991; Naqvi et al., 1998].
 Apparent oxygen utilization (AOU) reflects the amount of O2 consumed by remineralization of organic matter and nitrification in a water parcel since its last contact with the surface. The strong correlation between AOU and ΔN2O (the level above atmospheric equilibrium) commonly observed in ocean depth profiles provides circumstantial evidence that nitrification is the dominant mechanism of oceanic N2O production [Yoshinari, 1976; Elkins et al., 1978; Cohen and Gordon, 1979; Butler et al., 1989]. However, some studies have ascribed the correlation to widespread N2O production via denitrification, which may take place in anoxic microzones of sinking particulate matter [Yoshida et al., 1989]. Other studies have argued that denitrification may be a net sink for N2O, on the basis of the “bite” in the N2O profile observed in O2 deficient regions of the water column, where denitrifiers consume N2O for use as an electron acceptor [Elkins et al., 1978; Cohen and Gordon, 1978, 1979].
 Resolving the mechanisms, processes, and properties influencing oceanic N2O production should aid in predicting how the ocean might respond to future changes. An N2O source dominated by nitrification, for example, would be expected to increase in response to changes that enhance ocean productivity. Preliminary model and observational studies of iron fertilization, which has been proposed as a method to help sequester CO2 in the deep ocean, have found that the greenhouse benefit of reduced atmospheric CO2 may be balanced or even outweighed by increased oceanic N2O production [Fuhrman and Capone, 1991; Law and Ling, 2001; Jin et al., 2002].
 In this paper we present a database of subsurface N2O and O2 measurements from a variety of oceanic regions. The database is large enough in scope to offer for the first time a global picture of the subsurface N2O distribution. It includes cross sections of N2O and the ΔN2O/AOU ratio from transects in the Atlantic, Indian, and Pacific Oceans. We examine correlations between ΔN2O and AOU on isopycnal surfaces in these different regions and employ simple end-member mixing models to screen out mixing effects. We make a similar examination of correlations between ΔN2O and N*, a tracer of fixed nitrogen sources and sinks [Gruber and Sarmiento, 1997]. We also investigate the influence of factors such as O2, temperature, and pressure on the apparent yield of N2O per mole O2 consumed. A major goal of this study is to derive an improved, data-based model of oceanic N2O production that will allow future changes in the oceanic source to be predicted with more confidence.
 N2O samples were collected at a range of depths in Niskin or CTD bottles. Water was drawn from the bottles into 20-mL vials, which were sealed and analyzed by an automated headspace sampler technique [Butler et al., 1988; Butler and Elkins, 1991]. The precision of the N2O measurements is estimated at ±2.0%. Temperature, salinity, and O2 were measured by standard methods. O2 was not measured on the BLAST II cruise.
 ΔN2O was computed by subtracting the saturation concentration at the local potential temperature and salinity. The saturation value was calculated with the N2O solubility coefficient of Weiss and Price  and an assumed atmospheric partial pressure of 308 natm for all cruises. Between 1987 and 1994, atmospheric N2O increased from ∼305 to ∼312 natm, with an ∼1 natm gradient between the Northern and Southern Hemispheres [Hall et al., 2002]. For deeper waters that may have equilibrated with the atmosphere several hundred years ago, atmospheric N2O would have been as low as 265–275 natm [Machida et al., 1995; Battle et al., 1996; Flückiger et al., 1999; Sowers, 2001]. The secular increase in atmospheric N2O during the twentieth century introduces an uncertainty of up to 10–15% into the calculation of ΔN2O. We have not attempted to account for this uncertainty, since it is complex [e.g., Warner et al., 1996] and does not significantly affect our conclusions. AOU was calculated by subtracting observed O2 from saturation O2 using the solubility coefficient of Weiss .
2.2. Supplementary WOCE O2 and Nutrient Data
 The NOAA N2O and O2 data were supplemented with O2, NO3−, and PO4−3 from selected World Ocean Circulation Experiment (WOCE) cruises (WOCE Hydrographic Program Public Data, 2001, available at http://whpo.ucsd.edu/whp_data.htm). The WOCE data were interpolated to the same depths or sigma-theta levels as the NOAA data. The WOCE cruise tracks generally overlap to within 5° of the NOAA cruise tracks. Since O2 was not measured on BLAST II, we have relied entirely on WOCE data for O2 and AOU values in the Atlantic. Although the overlap between the BLAST II and WOCE cruise tracks is particularly good (Figure 1b), the data sets are not contemporaneous, which creates a source of uncertainty in our interpretation of O2-N2O relationships in the Atlantic.
 We rely upon WOCE nutrients for all cruises to calculate the tracer N*, which can be used to identify water masses that have experienced net sources or sinks of fixed nitrogen [Gruber and Sarmiento, 1997; Deutsch et al., 2001]. N* behaves as a conservative tracer in regions where only remineralization and nitrification occur. Negative values of N* have been interpreted as denitrification sinks, while positive values have been interpreted as biological N2 fixation sources. N* is calculated as NO3− − rN:PnitrPO4−3 + 2.90, with an estimated analytical error of ±0.2 μmol/kg, where rN:Pnitr = 16 ± 1 [Gruber and Sarmiento, 1997; Deutsch et al., 2001].
 Dissolved N2O concentrations are markedly higher in the northern and tropical sectors of the Pacific and Indian Oceans than in the southern sectors or in the entire Atlantic Ocean. Results are shown along the σθ = 26.9 surface, which falls within the upper 600-m range of the water column where most of the N2O that exchanges with the atmosphere is produced [Suntharalingam and Sarmiento, 2000], and which displays some of the strongest N2O-O2 correlations in our database (Figure 2a). N2O values are at their maxima (typically 50–70 nM) in and around the Arabian Sea and the eastern tropical Pacific (ETP) and decline to ∼40 nM in the western tropical and North Pacific. The σθ = 26.9 surface passes through the core of the O2 deficient zones of the ETNP and Arabian Sea, where N2O is depleted by denitrifier consumption, but intersects the lower of the two N2O maxima in the suboxic margins that sandwich the shallower ETSP O2 deficient zone. In the southern sectors of the Pacific and Indian Oceans, N2O concentrations decline to only 10–20 nM. In the Atlantic Ocean, N2O concentrations are in the range of 10–20 nM at both northern and southern midlatitudes and climb to 25–30 nM in the tropics.
 The ratios of ΔN2O/AOU on the σθ = 26.9 surface display patterns similar to those for the absolute values of N2O (Figure 2b). ΔN2O/AOU provides a rough estimate of N2O production per mole O2 consumed, but also reflects mixing effects and possible undersaturation or supersaturation of O2 and N2O at the time of last contact with the atmosphere. ΔN2O/AOU ranges from a high of 0.2–0.3 nmol/μmol in the ETP and Arabian Sea to a low of only 0.01–0.05 nmol/μmol in the South Atlantic and South Indian Oceans.
 Depth cross sections along latitudinal transects show that N2O concentrations reach their maxima in the upper 600 m of the water column at low latitudes, typically peaking at 200–400 m depth (Figure 3). A deeper N2O maximum occurs between 500 and 1000 m at midlatitudes in the North Pacific. The RITS89 eastern Pacific transect at ∼105°W passes through the ETNP denitrification zone at 14°N and skirts along the western side of the ETSP denitrification zone at ∼10°S (Figure 3c). The highest value in the NOAA data set (115 nM) was measured at 200 m at 14°N, just above the 300–700 m zone of N2O depletion in the ETNP O2-deficient layer. N2O concentrations taper off significantly south of 20°–30°S in all the ocean transects. Along the BLAST II Atlantic transect, N2O also declines northward of ∼20°N (Figure 3d).
 The ΔN2O/AOU cross sections (Figure 4) are generally similar to the corresponding N2O cross sections (Figure 3), except that the ΔN2O/AOU maxima are some 50–100 m shallower than the N2O maxima, and the ratios decline more noticeably with depth. One striking feature of the ΔN2O/AOU transect in the Eastern Pacific (Figure 4c), which is less apparent in the N2O transect, is the tongue of high ΔN2O/AOU (>0.15) that follows the σθ = 26.9 surface southward to ∼50°S.
 Depth profiles of N2O, O2, and N* at individual stations show that O2 and N2O are generally anticorrelated, as has been reported extensively in the literature (Figure 5). However, the correlation is not always strong. Many profiles from the central and western tropical Pacific display a sharp spike in the N2O profile around 200 m that is unmatched by a comparable dip in O2 (Figure 5a). These spikes in N2O occur at σθ levels comparable to that of the 115-nM extreme maximum in N2O at 14°N in the ETNP (Figure 3c), which suggests that some of the N2O produced in the ETNP may be transported westward. Similar spikes in the southeast Pacific (Figure 5b) occur consistently on the σθ = 26.9 surface, suggesting transport from the lower N2O maximum below the ETSP O2 deficient layer.
 The relationship between N* and N2O is more complex and variable than the N2O-O2 relationship. In regions where N* < 0, such as the northern and tropical Pacific, N* and N2O tend to be anticorrelated (Figure 5a). In contrast, in regions of strong N2 fixation such as near 21°N in the North Atlantic, where N* > 0, N2O and N* appear positively correlated (Figure 5c). The N2O-N* relationship is ambiguous in most other regions, such as the Southern Ocean, where near-zero values of N* indicate neither strong N2 fixation nor denitrification.
4. Data Analysis and Discussion
 The N2O and ΔN2O/AOU patterns presented in section 3 demonstrate that N2O production varies throughout the ocean. Production is particularly strong in several “hot spot” regions in the eastern tropical Pacific and Arabian Sea. Outside of these hot spots, there is a significant background production, which itself varies with O2 and depth.
4.1. Export of N2O From “Hot Spots”
 Many studies, including the present one, have reported N2O depletion in the heart of the denitrification zones in the ETP and the Arabian Sea and high N2O production in the surrounding suboxic regions [Elkins et al., 1978; Cohen and Gordon, 1978, 1979; Codispoti and Christensen, 1985; Ward et al., 1989; Law and Owens, 1990; Naqvi et al., 1998]. A new finding that emerges from the current data set is that N2O from the hot spots in the ETP is exported far afield, significantly increasing the N2O concentration at distant stations from what it would have been owing to background production alone. The advective signal emerging from both the ETNP and ETSP appears to be positive at all SAGA II and RITS89 stations. This suggests that in terms of export to the rest of the ocean, the peripheral suboxic regions of these hot spots dominate any signal of N2O depletion from within the anoxic cores.
 The RITS89 data from the southeast Pacific show particularly striking evidence that N2O from the ETSP O2 minimum zone, which is centered at 10°–15°S [Gruber and Sarmiento, 1997], is exported as far south as 50°S. This conclusion is based on the southward-extending tongue of high ΔN2O/AOU (Figure 4c) and the sharp bulges in the N2O profile at southern stations (Figure 5b). These occur consistently on the σ = 26.9 isopycnal surface, which is the sigma level of the lower N2O maximum in the ETSP. Comparison of observed ΔN2O to the biological ΔN2O production term (estimated using the function of [O2], AOU, and depth developed later in section 4.3.3 (equation (9))) suggests that biological production associated with local O2 depletion can only account for ∼40–60% of observed ΔN2O along and above the σ = 26.9 isopycnal surface from about 22° to 44°S. This suggests that export of N2O produced in and around the oxygen minimum in the ETSP accounts for the remainder.
 A similar exercise along the 5°N zonal transect in the western Pacific from SAGA II suggests that biological production associated with local O2 depletion accounts for ∼45–60% of observed ΔN2O on and above the σ = 26.5 isopycnal surface as far west as 157°E. This result is consistent with export from the ETNP O2 minimum zone (Figure 3c, 14°N, and Cohen and Gordon [1978, 1979]), which is centered off the coast of Central America around 15°N, 105°–110°W. Depth profiles from the ETNP show that the upper N2O maximum occurs between σθ = 25.9–26.2, with severe N2O depletion occurring below.
 The advective signal from hot spots in the northern Indian Ocean is less clear. The profiles from this region [Law and Owens, 1990; Lal et al., 2000] show extensive denitrifier N2O consumption in O2-depleted zones, bordered by moderate to high values (typically 30–60 nM) in the surrounding suboxic zones. The N2O concentrations from Law and Owens , which extend south from the Arabian Sea along the 67°E transect, are in the range of 10–15 nM near the equator in the top 500 m of the water column. These values are much lower than the 40- to 50-nM concentrations measured by Butler et al.  at comparable O2 levels in the eastern Indian Ocean and western tropical Pacific (Figure 2a). The low N2O values in the western Indian Ocean suggest that N2O-depleted water from the Arabian Sea may mix out and dilute N2O to the south. On the other hand, the eastern Indian Ocean transect suggests that high N2O from the Bay of Bengal may be transported toward the equator (Figures 2a and 3a). Further study is required to understand these contrasting patterns.
4.2. Correlation Slopes and Mixing Gradients
 Up to this point, we have presented ΔN2O/AOU ratios calculated from individual data pairs. More commonly in the literature, the ΔN2O/AOU ratio has been quantified based on the linear correlation slope of a ΔN2O versus AOU cross plot that encompasses a horizontal and vertical range of data. The slope in turn has been interpreted as the biological N2O yield per mole O2 consumed [Najjar, 1992; Suntharalingam and Sarmiento, 2000]. A compilation of the linear regression slopes from various studies reveals a somewhat confusing array of values, ranging from 0.076 to 0.31 nmol N2O/umol O2, in which patterns across different ocean regions are not easily identified [Suntharalingam and Sarmiento, 2000, Table 1]. The intercepts of these regression slopes are variable and often do not pass through the origin, suggesting nonlinear effects. Below we summarize our examination of the correlation slopes of ΔN2O versus AOU and ΔN2O versus N* from the NOAA cruises. We focus on specific σ levels to reduce uncertainties caused by diapycnal mixing of water masses of different origin.
 When data from a wide geographical area are combined in one cross plot, they can represent a range of different biological ΔN2O/AOU yields, even if they are restricted to the same σ surface. The correlation slope of such a cross plot may reflect a mixing gradient between source waters of different characteristic compositions rather than the true average biological ΔN2O/AOU yield. End-member mixing models can be useful in separating mixing gradients from biological relationships [Takahashi et al., 1985; Anderson and Sarmiento, 1994; Gruber and Sarmiento, 1997; Deutsch et al., 2001]. The observed N2O concentration can be partitioned into “preformed” N2O resulting from conservative mixing between two end-members along isopycnal surfaces and a residual biological source/sink term, Δ[N2O]′. The latter represents the nonconservative departure of N2O from the straight mixing line due to biological production.
The prime symbol distinguishes this production term from the ΔN2O term discussed in previous sections, which involves only the correction for atmospheric equilibrium,
A similar distinction can be made between −Δ[O2]′ and AOU (i.e., −Δ[O2]).
 In applying the mixing model to the NOAA data, we encountered several problems that limit our confidence in the model results. These problems are illustrated by data from the BLAST II South Atlantic transect from 6°N–34°S (Figure 6), a region where the circulation regime in the main thermocline can be approximated by simple binary mixing along isopycnal surfaces between northern and southern end-members [Takahashi et al., 1985]. First, the property plots of N2O, O2, and N* versus S along the σθ = 26.9 surface in the South Atlantic display relatively little bowing, such that the Δ[N2O]′ term is small compared to the [N2O]preformed term (Figures 6a–6c). Second, the ΔN2O/AOU ratios at individual stations show a large gradient (Figure 6f). The individual ratios range from 0.03 to 0.11 nmol/μmol across the transect, suggesting that any correlation slope derived from these data would be a mean of disparate values. The correlation slope of the ΔN2O vs. AOU cross plot is 0.12 nmol/μmol, which is larger than any of the individual values, while the mixing corrected slope of Δ[N2O]′ versus −Δ[O2]′ is 0.06 nmol/μmol (Figure 6d). Similarly, the apparent ΔN2O versus N* correlation slope of +7.3 nmol/μmol is reduced by the mixing model to ∼0 (Figure 6e). We conclude that the apparent slopes of these cross plots are primarily due to mixing gradients and do not provide a reliable estimate of the ratio of biological N2O production to O2 consumption or N2 fixation. The “corrected” slopes may provide a better estimate. The Δ[N2O]′ versus −Δ[O2]′ slope in particular appears to represent a reasonable mean of the individual ΔN2O/AOU ratios along the transect. However, the uncertainty in the corrected slopes is still high, because of the statistically small sample of data available and the error in the small residual Δ[N2O]′, Δ[O2]′, and ΔN* terms. We encountered similar problems in our analysis of NOAA data from the eastern South Pacific and the eastern Indian Ocean.
 To summarize the ΔN2O/AOU analysis, we find that the biological N2O yield per mole O2 consumed cannot be calculated with great confidence from cross-plot correlation slopes. The essential problem is that the N2O yield is spatially variable. As a result, strong mixing gradients exist in the data that can overwhelm more subtle N2O production terms. The ΔN2O/AOU ratio differs from traditional Redfield ratios that relate O2 consumption to remineralization of NO3− and PO4−3 [Takahashi et al., 1985; Anderson and Sarmiento, 1994]. The latter tend to be more universally constant and thus lend themselves well to estimation with a correlation slope and mixing model analysis. In contrast, ΔN2O versus AOU cross-plot slopes reflect a combination of mixing gradients and variable biological relationships, which are difficult to separate. This is especially true when data spanning a wide geographical range are combined into the same plot. Restricting the analysis to a narrow geographical range is not a satisfactory solution either, because in that case there are often insufficient data to obtain a good statistical correlation. We conclude that the ΔN2O/AOU ratio calculated from individual data pairs offers at present the best available means to estimate the biological yield of N2O per mole O2 consumed. At the same time, we note that in some extreme cases, such as the export of N2O from hot spots in the ETP discussed in section 4.1, mixing and transport can as much as double the ΔN2O/AOU ratio over its value due to local biological production. Surface undersaturation or supersaturation creates an additional uncertainty in the ΔN2O/AOU ratio of up to 10–20% in some regions, e.g., the North Atlantic [Suntharalingam, 1997].
 The mixing model analysis of ΔN2O versus N* suggests that the apparent positive correlation in cross plots from the South Atlantic (Figure 6e) is an artifact of mixing gradients. Similarly, the anticorrelations between ΔN2O and N* observed in the Indian and eastern Pacific data sets appear to result mainly from mixing gradients. The lack of a clear ΔN2O′/ΔN* anticorrelation tends to argue against a significant widespread denitrification mechanism for N2O production, as in anoxic microzones of sinking organic matter. A further argument against a widespread denitrification N2O source is that there is little evidence of a ΔN2O-N* anticorrelation, even before correcting for mixing, in data ranges that are far removed from the ETP and Arabian Sea.
4.3. Factors Affecting the Biological ΔN2O/AOU Yield
 In this section we examine some of the factors that could contribute to the variability of the ΔN2O/AOU biological yield observed in the ocean (Figures 2b and 4). These include sensitivity to O2, temperature, depth, and nitrogen content of organic matter. Our goal is to develop a parameterization of N2O production for use in ocean biogeochemistry models. A parameterization with a strong basis in empirical data is important for predicting how the ocean N2O source will respond to future changes. O2 consumption is a principal flux that is already calculated by most ocean biogeochemistry models (R. G. Najjar and J. Orr, Design of OCMIP-2 simulations of chlorofluorocarbons, the solubility pump and common biogeochemistry, 1998, available at http://www.ipsl.jussieu.fr/OCMIP/phase2/simulations/design.ps) (hereinafter referred to as Najjar and Orr, online manuscript, 1998). If the O2 consumption term is multiplied by an appropriate parameterization for the biological N2O yield per mole O2 consumed, the models can be extended in a straightforward manner to include an N2O production term [Suntharalingam and Sarmiento, 2000; Suntharalingam et al., 2000].
 Dissolved O2 is the single most important influence on oceanic N2O. The evidence for its influence is based in empirical observations as well as theoretical considerations and laboratory measurements. O2 levels are generally high in the young waters of the Atlantic and Southern Oceans and low in the older waters of the tropical and northern Pacific and the northern Indian Oceans (Figure 7). Dissolved N2O concentrations and ΔN2O/AOU ratios display the opposite distribution pattern (Figures 2a and 2b). The relationship between O2 and the ΔN2O/AOU yield of nitrification is discussed in more detail below.
 Nitrification normally proceeds via reaction (R1), with a net consumption of 2 moles of O2 per mole NH4+ nitrified to NO3−. (R2) is a minor alternative pathway that consumes only 1 mole of O2 per mole NH4+ and results in N2O production.
Laboratory studies of the nitrifier Nitrosomonas sp. show that the N2O yield of nitrification increases with decreasing O2 [Goreau et al., 1980; Joergensen et al., 1984], as might be predicted from thermodynamical considerations, based on the stoichiometry of the two reactions. Here it should be noted that the ΔN2O/AOU yield is the product of the nitrifier N2O yield (ΔN2O/ΔNO3−) and the N:O2 Redfield ratio (ΔNO3−/AOU). Of the two multiplied factors, only the nitrifier yield ΔN2O/ΔNO3− is sensitive to O2 availability. The Redfield ratio ΔNO3−/AOU is generally regarded as a constant [Anderson and Sarmiento, 1994].
 N2O, O2 and NO3− observations in the Atlantic Ocean, using all data between σ = 26.0 and 27.1 from the BLAST II cruise, are qualitatively consistent with the laboratory data discussed above. ΔN2O/AOU and ΔN2O/ΔNO3− tend to increase with decreasing O2 (Figure 8), while ΔNO3−/AOU displays no particular correlation with O2. Here we approximate that ΔNO3− ∼ NO3− at any given depth, assuming that NO3− is near 0 in the surface ocean. This assumption is not necessarily valid in HNLC regions like the Southern Ocean and the subtropical North Pacific. We plot ΔN2O/AOU against O2 rather than AOU to focus on the role of dissolved O2 in determining the nitrifier N2O yield. AOU does not provide a clear measure of the actual O2 content of the water since it depends to a large extent on [O2]sat, which is sensitive to temperature. The σ = 26.0–27.1 range was chosen on the basis of our analysis of cross-plot correlation slopes on isopycnal surfaces in section 4.2, which suggested that ΔN2O/AOU is similar through this range. Below σ = 27.1, ΔN2O/AOU begins to decline significantly. Above σ = 26.0, AOU is often close to or less than 0, such that the ΔN2O/AOU ratio becomes sensitive to the denominator. The Atlantic data were chosen because they are more likely to reflect pure nitrification than data from the Pacific or Indian Oceans.
 To derive an empirical expression for the biological N2O yield as a function of O2, one cannot simply fit a straight line through Figure 8a. A linear function would tend to underestimate the sensitivity of the yield to O2 because it would not account for the fact that the observed ΔN2O/AOU ratio reflects the integrated history of a water mass as its O2 content changes from surface saturation to the observed depleted value at depth. To gain insight into what the actual instantaneous form, dN2O/−dO2, of the yield as a function of O2 might be, we turn to the laboratory study of Goreau et al. . This is the only study we are aware of that has directly measured the O2 sensitivity of the N2O yield of marine nitrifiers in aqueous solution.
Goreau et al.  measured the N2O/NO2− end product ratio of primary nitrification at a range of O2 tensions (Figure 9a). Nitrification proceeds in a sequence of primary (R1′) and secondary (R1″) steps.
Since no evidence exists for additional N2O production during secondary nitrification [Goreau et al., 1980], the N2O/NO2− ratio (R2/R1′) is more or less equivalent to dN2O/dNO3−, i.e., the instantaneous form of the ΔN2O/ΔNO3− yield defined above.
where a1 = 0.20 ([mol N2O/mol N] [μmol O2/L]−1) and a2 = −0.0004 (mol N2O/mol N). This fit captures the sharp increases in the yield in the vicinity of very low O2 and reproduces the Goreau et al. data to well within their range of uncertainty (Figure 9b). The fit tends to underestimate the yield at high O2, although the Goreau et al. data do suggest that the average dN2O/dNO3− yield is ∼50% lower at a saturated O2 level of 220 μmol/L than at a modestly depleted level of 110 μmol/L.
As might be expected, the integrated ΔN2O/AOU yield (equation (6)) is considerably less sensitive to O2 than the instantaneous dN2O/−dO2 yield (Figure 9c). A comparison of the integrated yield to the BLAST II data reveals that the theoretical and empirical curves are similar in shape and value (Figure 9d). The overlap of the curves can be interpreted to suggest that the empirical yields in the BLAST II data set are consistent with a nitrification source. This statement must be qualified with a number of caveats, given the uncertainty in the instantaneous and integral forms of the fit and in the other variables involved. The Goreau et al. study was conducted in the laboratory at 25°C and atmospheric surface pressure with a pure culture of nitrifying bacteria. In contrast, the BLAST II data were collected at an average temperature of 10°C (range 5°–16°C) at subsurface ocean pressures, and may have been subject to nutrient limitations and/or competitive stresses from other bacteria. In addition, owing to the decreasing solubility of O2 with increasing temperature, the maximum O2 concentration encountered by laboratory nitrifiers was [O2]sat,25°C = 220 μmol/L, whereas nitrifiers in cold ocean waters can be exposed to O2 levels of up to ∼350 μmol/L. This last consideration raises the point that equation (6) is inherently temperature sensitive, since it is a function of [O2] and [O2]sat. For identical values of a1, a2 and [O2], equation (6) predicts ΔN2O/AOU yields that are ∼25% lower if [O2]sat is calculated at 10°C rather than at 25°C (Figure 9d).
 Empirical values for a1 and a2 can be backed out from the BLAST II data by plotting observed ΔN2O/AOU versus (ln[O2]sat − ln[O2])/([O2]sat − [O2]) (Figure 10). According to equation (6), the slope of the plot should be a1 RN:O2, while the intercept should be a2 RN:O2. This exercise assumes that the actual instantaneous N2O yield in the ocean obeys the same functional dependence on O2 as the laboratory fit described by equation (4). This assumption is not necessarily true, but is as reasonable as any other we can make given the available information. A linear fit through the data in Figure 10 yields values of a1 = 0.26 ([mol N2O/mol N] [μmol O2/L]−1) and a2 = −0.0006 mol N2O/mol N.
 When the above exercise is repeated for the other NOAA cruises, the BLAST II, RITS89, and SAGA II ΔN2O/AOU yields all show a significant correlation to O2, and the empirical values of a1 and a2 obtained from plots of ΔN2O/AOU versus (ln[O2]sat − ln[O2])/([O2]sat − [O2]) agree reasonably well (within ±25% of the mean at a given O2 level) (Table 2). In contrast to the other three NOAA cruise data sets, the SAGA III data do not fit well to equation (6). Instead, they display a relatively constant ΔN2O/AOU value of 0.15 ± 0.10 nmol/μmol, with no obvious variation with O2, temperature, depth, or N*. One reason for this may be that the SAGA III data are all from the tropical Pacific, where zonal currents are strong and transport and mixing of high N2O from the “hot spot” O2 minima in the ETP help determine the locally observed ΔN2O/AOU.
Table 2. Parameters a1 and a2 Obtained From Fit of NOAA Data to Equation (6)
 We turn to the RITS89 cruise in the Eastern Pacific to consider how denitrifier N2O production or consumption may affect empirical ΔN2O/AOU yields and their sensitivity to O2. The RITS89 ΔN2O/AOU ratios tend to increase with decreasing O2 (Figure 11a), but, unlike BLAST II, they drop abruptly at very low O2. This is consistent with N2O consumption by denitrifiers at O2 < ∼4 μmol/L (Najjar and Orr, online manuscript 1998). Supplementary data sets from the O2 minimum zones in the ETSP, ETNP, and Arabian Sea [Cohen and Gordon, 1978, 1979; Friederich et al., 1985; Lal et al., 2000] confirm that ∼4 μmol/L is approximately the O2 level at which N2O depletion is observed to begin. At slightly higher levels of O2, between the point where N2O is completely consumed and where denitrification shuts off because of O2 inhibition, the N2O yield of denitrification can be huge (≫50%) [Joergensen et al., 1984]. In this same suboxic realm, nitrification also can persist [Ward et al., 1989], with a ΔN2O/ΔNO3− yield of up to 9% mol N/mol N [Goreau et al., 1980]. Nitrification and denitrification are often tightly coupled in this realm, such that distinguishing N2O production by one mechanism versus the other may not be possible, except perhaps with isotopic techniques [Yoshinari et al., 1997; Naqvi et al., 1998; Popp et al., 2002].
 Consistent with the above discussion, the RITS89 data show a steep rise in ΔN2O/AOU at suboxic O2 levels, just before their abrupt drop. The maximum yield observed is 0.43 nmol/μmol at O2 = 5 μmol/L. This compares reasonably well (within ∼25%) to the yield of 0.35 nmol/μmol predicted by equation (6), which is derived solely from considerations of nitrification (Figure 11b). However, we cannot rule out the possibility that an additional process such as denitrification may contribute to N2O production at this data point. If this N2O is widely exported to the rest of the Pacific Ocean, as suggested in section 4.1, then denitrification may still contribute significantly to the total oceanic N2O inventory, even if production is restricted to a small number of O2-deficient regions. A comparison of the RITS89 and BLAST II data shows that the RITS89 ΔN2O/AOU yields are some 50% larger at high O2 (Figure 11b). This is further evidence of the transport of N2O from the ETSP discussed earlier, which leads to large ΔN2O/AOU ratios at relatively high O2 in the RITS89 data set.
 To summarize section 4.3.1, we find that the O2 sensitivity of the nitrifier dN2O/dNO3− yield plays an important role in determining the variability in ΔN2O/AOU ratios observed in the ocean. We have derived a parameterization for the instantaneous dN2O/−dO2 yield as an inverse function of O2. The parameterization is guided by both laboratory and empirical data and assumes that N2O is produced primarily by nitrification down to a threshold level of O2 of ∼4 μmol/L, below which it is consumed by denitrifying bacteria. Just above this threshold, we are not able to distinguish between N2O production by nitrification and denitrification. Although additional production by denitrification may well occur, our inverse function appears adequate to describe the observed high yields, within a ±∼25% range of uncertainty. The detailed parameterization is presented in section 4.3.3, after modifications for pressure sensitivity.
4.3.2. Pressure and Temperature
 Both N2O and ΔN2O/AOU tend to decrease with depth [Butler et al., 1989; Suntharalingam and Sarmiento, 2000]. This tendency is clear when one examines depth profiles extending several thousand meters, and is present but less apparent in the upper 1000–1500 m (Figures 3, 4, and 5). Effectively, this means that N2O and ΔN2O/AOU tend to decrease at high pressure, since pressure increases linearly with depth in the ocean. In the upper 1000 m, the influence of pressure may be dominated and obscured by other factors, such as varying O2 levels. In addition, in the top ∼75 m, light inhibition of nitrification suppresses the ΔN2O/AOU yield.
Butler et al.  hypothesized that the decline in N2O with depth may reflect pressure inhibition of the primary nitrification pathway that leads to N2O production. At increasing ambient pressure, NO2−(aq), the normal aqueous end product of primary nitrification (R1′), may be increasingly favored thermodynamically over the alternative gaseous end product, N2O(g). Since (R1′) is followed by secondary nitrification to NO3−(R1″), the dN2O/dNO3− nitrifier yield theoretically should decrease with increasing pressure.
 The theoretical basis for a direct temperature effect on the N2O yield is less evident. The rate of most microbial activity increases with temperature, hence one might expect an increase in N2O production at warmer temperatures. However, since microbial O2 consumption also increases with temperature, the net effect of temperature on the ΔN2O/AOU ratio is unclear. The only study we are aware of that directly examined the effect of temperature on the dN2O/dNO3− yield of nitrification was that of Yoshida and Alexander , who reported a strong increase in the yield with temperature in laboratory studies of Nitrosomonas europaea. However, the yields they reported of 4–24% may have been contaminated by the presence of denitrifying bacteria [Goreau et al., 1980] and are incompatible with the empirical ocean yields discussed in section 4.3.1, which are generally <1%. Most other studies of the sensitivity of N2O emissions to temperature have been conducted in soils or sediments, where temperature's principal effect is to alter microbial activity and hence O2 availability, thereby indirectly influencing the N2O yield [Roever et al., 1998; Rudaz et al., 1999; Castaldi, 2000; Usui et al., 2001]. An additional indirect effect of temperature results from the temperature sensitivity of the O2 solubility coefficient. As noted in section 4.3.1, [O2]sat increases with decreasing temperature such that nitrifying bacteria inhabiting colder waters tend to be exposed to higher O2 concentrations than bacteria in warmer waters.
 The above discussion suggests that the effect of temperature on N2O emissions is difficult to separate and may result largely from its influence on O2. Similarly, the effects of temperature and pressure may be difficult to distinguish, since temperature as well as pressure tends to decrease with depth. We therefore opt to model the N2O yield as a function of O2 and depth, with the assumption that any additional influence of temperature will be inherently linked to the O2 and depth inputs. Our approach is somewhat different from that of Elkins et al.  and Butler et al. , who found that the ΔN2O-AOU relationship in ocean data sets can be modeled more successfully as a linear correlation when temperature is added as a second independent variable. Butler et al.  noted in their approach that temperature may serve mainly as a proxy variable. Here we model ΔN2O/AOU directly as a nonlinear function of O2 and depth.
 The mathematical form that best describes the functional dependence of the nitrifier yield on pressure is not evident from first principles. In addition to thermodynamic considerations, microbial kinetics also play a role in ways that are difficult to predict a priori [Butler et al., 1989]. We adopt the approach of Suntharalingam and Sarmiento , who modeled the decrease in ΔN2O/AOU as an exponential function of depth. To tighten the focus on depth sensitivity, we restrict our analysis to data below 1000 m, where gradients in O2 are relatively small. All individual station profiles that included at least four measurements below 1000 m are fit to
where z = depth (m) and A (mol mol−1) and k (m) are empirically determined constants.
 Most of the 43 profiles tested display a good fit to equation (7), although about one fourth do not. Among the good fits, the pre-exponential factor A is generally similar to the value of ΔN2O/AOU at 1000 m, while the exponential constant k typically ranges between 2000 and 4000 m, with an average value of ∼3000 m. Among the “poor fits,” many profiles display strong ΔN2O/AOU maxima in the upper water column that drops off sharply and then remains relatively constant with depth. This behavior could be consistent with the observation that nitrification tends to be concentrated in a fairly narrow depth interval, usually located near the bottom of the euphotic zone [Ward, 1986]. However, if nitrification accounts for a constant fraction of the oxygen consumed in remineralization, the intensity of nitrification will not necessarily lead to variations in the ΔN2O/AOU yield. Some isotopic studies have noted that the minimum in δ15N-N2O found at the bottom of the euphotic zone suggests a disproportionately high rate of N2O production there, which is not evident based on O2 or even N2O concentration profiles [Dore et al., 1998; Popp et al., 2002].
 Although many questions remain and not all profiles fit well to equation (7), the exponential decrease with depth appears to provide a reasonable description of the observed decrease in ΔN2O/AOU, given the present information. We therefore modify equation (4) to account for depth/pressure sensitivity (equation (8) below). We do not attempt to account for the integrated pressure history of each water mass from the surface to the observed depth because such an adjustment yields only modest changes in the fit parameters (Table 2), which are smaller than the uncertainty in equation (7) itself.
4.3.3. Parameterization for the Instantaneous ΔN2O/−ΔO2 Yield
Equation (8) gives our recommended parameterization for dN2O/−dO2. This equation describes the instantaneous N2O production per mole O2 consumption as a function of O2 and depth.
where dN2O/−dO2 is in units of mol mol−1; O2,crit = 4 ± 3 μmol/L; RN:O2 = 16:170 mol/mol [Anderson and Sarmiento, 1994]; a1 = 0.26 ± 0.06 [mol N2O/mol N] [μmol O2/L]−1; a2 = −0.0004 ± 0.0001 mol N2O/mol N; Zscale = 3000 m ± 1000; Z = depth in meters; and Zeuph = depth of euphotic zone in meters.
 The best guess values of a1 and a2 were determined by experimentation with the range of coefficients suggested by Table 2. The coefficients were entered into the depth-scaled, integral form of the fit,
using observed O2, temperature, and depth. The resulting mean, maximum, and minimum model curves span most of the observed ΔN2O/AOU yields, excluding the high yields at high O2 in the SAGA II and RITS89 data sets, which are probably biased high by transport and mixing influences from the eastern tropical Pacific (Figure 12).
 The parameter O2,crit was derived based on supplementary data sets from the O2 minimum zones [Cohen and Gordon, 1978, 1979; Friederich et al., 1985; Lal et al., 2000] and is consistent with the value commonly used in ocean biogeochemistry models (Najjar and Orr, 1998). Below the O2,crit threshold, equation (8) assumes that N2O production ceases. It does not account for additional denitrifier consumption of N2O that has been produced by nitrification.
 Our parameterization of the biological ΔN2O/AOU yield differs from previously published parameterizations in that it is oxygen sensitive across the entire spectrum of O2 concentrations. In contrast, the parameterization of Suntharalingam and Sarmiento  is constant for all O2, while that of Suntharalingam et al.  is constant for most values of O2 and only becomes O2 sensitive at O2 < ∼50 μmol/L. Our mean value of O2crit = 4 μmol/L is also larger than the 1 μmol/L assumed by Suntharalingam et al. .
Equation (9) can be used to estimate cumulative biological N2O production, ΔN2O, over the history of a water mass, providing a means of testing the parameterization against observed depth profiles. Using observed O2, temperature, salinity, and depth as inputs, equation (9) predicts ΔN2O values that agree reasonably well with observations from most of data sets listed in Table 1 (Figure 13). Observed and modeled ΔN2O profiles from the North and South Atlantic and North Pacific tend to agree best when mean to minimum values of a1 and a2 are used (Figures 13a and 13d), while profiles from the tropical and South Pacific tend to agree best when maximum values are used. This pattern could suggest an additional N2O production mechanism in the tropical and South Pacific, such as denitrification, or some other influence that is not captured by the parameterization. Equation (9) does a poor job reproducing observed ΔN2O in profiles in the heart of O2 deficient zones in the eastern tropical North Pacific (Figure 13b) and the Arabian Sea, where extensive denitrifier consumption of N2O occurs.
Equation (9) can also be applied in global budget estimates. Using O2 and AOU from the World Ocean Atlas (WOA) 1998 [Levitus and Boyer, 1998] and the range of a1 and a2 given above, equation (9) predicts a global ocean inventory of ΔN2O = 160–360 Tg N. Adding the global inventory of [N2O]sat of 450–480 Tg N (estimated using temperature and salinity from WOA 1998, and assuming that the ocean equilibrated on average with an atmospheric N2O partial pressure somewhere between 280 and 300 natm) yields a total ocean N2O inventory of 610–840 Tg N. This inventory is somewhat lower than the earlier estimate of 860–1080 Tg N derived from an ocean biogeochemical model [Suntharalingam et al., 2000]. It may be biased low by the tendency of equation (9) to underestimate ΔN2O below ∼2500 m (Figures 13a and 13c).
 A similar global calculation using O2, AOU, and O2 consumption rates from the NCOM-OCMIP ocean carbon cycle model [Doney et al., 2003] as inputs to the instantaneous form (equation (8)) suggests an annual N2O production rate of 5.8 ± 2 Tg N/yr. Approximately 70% of the production is predicted to occur between 30°S and 30°N, with only 15% occurring south of 30°S. This total N2O production rate probably overestimates the actual oceanic source of N2O by ∼1–3 Tg N/year, since it does not account for denitrifier N2O consumption in O2 deficient zones (Figure 13b) [Codispoti and Christensen, 1985; Naqvi and Noronha, 1991]. Keeping this last uncertainty in mind, our estimated annual N2O production rate falls within the range of the sea-to-air N2O flux, which has been estimated at 1–7 Tg N/yr based on surface N2O data [Butler et al., 1989; Nevison et al., 1995; Prather et al., 2001]. Our distribution of N2O production at first glance does not agree well with surface sea-to-air flux calculations, which estimate that ∼1/3 of the global oceanic source is emitted from the Southern Ocean [Nevison et al., 1995; Suntharalingam and Sarmiento, 2000]. However, the two distribution estimates are not necessarily incompatible, depending on the fraction of N2O produced at low latitudes that is ventilated in the Southern Ocean. A prognostic 3-dimensional ocean model simulation of N2O that includes a denitrifier consumption term would help to better address this question.
4.3.4. N Enrichment of Organic Substrate
 There is some evidence in the BLAST II data set that ΔN2O/AOU increases with substrate N-enrichment. This observation, which has not been previously reported, is suggested here by two depth profiles at 12°N and 20.8°N (in which O2 and N* were measured 6 years earlier than N2O) from the North Atlantic subtropical gyre, a region known for high rates of N2 fixation [Gruber and Sarmiento, 1997, and references therein]. N2O and N* appear positively correlated in both profiles (Figure 5c), and the maximum ΔN2O/AOU yield found in the Atlantic occurs in the gyre. Although the dN2O/dNO3− nitrifier yield should be relatively low in the well-oxygenated North Atlantic gyre, the Redfield ratio of N-rich organic matter derived from N2 fixers may increase from its typical 16:170 value [Gruber and Sarmiento, 1997]. This would effectively increase the ΔN2O/AOU yield observed in the gyre, since ΔN2O/AOU is the product of dN2O/dNO3− and the N:O2 Redfield ratio, integrated over the lifetime of the water mass (equation (4)). If a firm correlation between N-rich substrate and enhanced N2O production can be established, this influence can be readily incorporated into the parameterization for dN2O/−dO2 in equation (8). Rather than being treated as a constant, RN:O2 can be allowed to increase from 16:170 to an appropriate higher value.
 If N2 fixation influences N2O production, there are a number of interesting implications for the oceanic source of N2O. For example, ecosystem shifts that favor N2 fixing bacteria would tend to increase nitrification and hence N2O production [Dore et al., 1998]. In addition, N2O production might increase in response to iron fertilization, since N2 fixers are particularly sensitive to iron limitation because of their high enzymatic Fe requirements [Falkowski, 1997]. In the one available field study of iron fertilization, Law and Ring  detected a small increase in N2O production in the pycnocline relative to unfertilized patches. However, their study site in the Southern Ocean was located well outside the warm tropical and subtropical waters that support N2 fixation. They hypothesized that the N2O increase was most likely due to a microbial response that altered the dN2O/dNO3− yield.
 The global subsurface database presented here shows that the concentration of dissolved N2O is high in the tropical and northern Pacific and relatively low, although still supersaturated, in the Atlantic and Southern Oceans. Subsurface N2O is undersaturated with respect to the atmosphere only in a few O2 minimum zones in the eastern tropical Pacific (ETP) and Arabian Sea, where it is consumed by denitrification. More generally, N2O concentrations peak in the upper 500 m of the water column in the ETP, where typical maximum concentrations are 60–70 nM. In comparison, the concentration of N2O is only 25–30 nM at equivalent depths in the tropical Atlantic, and 10–15 nM in the Southern Ocean. Evidence from depth profiles and transect cross sections suggests that N2O from the ETP is advected zonally to the western tropical Pacific as far west as 157°E and meridionally as far south as 50°S.
 The equilibrium-corrected term ΔN2O is generally positively correlated with apparent oxygen utilization (AOU) in depth profiles and along isopycnal surfaces. However, one must use caution in interpreting the linear correlation slope of a ΔN2O versus AOU cross plot as the biological yield of N2O per mole O2 consumed. ΔN2O-AOU correlation slopes are often strongly influenced by mixing gradients. In addition, the relationship between N2O production and O2 consumption is inherently nonlinear. This is especially true in O2 minimum zones and likely is also true in waters that are only moderately O2-depleted. As a result, efforts to simulate past and future changes in the oceanic N2O source using ocean biogeochemical models will depend strongly on the models' underlying O2 distribution.
 Most of the evidence presented here suggests that oceanic N2O is produced mainly by nitrification. This evidence includes the widespread ΔN2O-AOU correlation, the lack of a convincing biological ΔN2O-N* anticorrelation in the Pacific and Indian Oceans, and the hint of a positive ΔN2O-N* correlation in North Atlantic regions known for high N2 fixation. A nitrification source is also indicated by the fact that ΔN2O/ΔNO3− and ΔN2O/AOU ratios in the ocean increase with decreasing O2 in a manner consistent with laboratory studies of nitrifying bacteria. Much of the variability in ΔN2O/AOU throughout the ocean, for example, the low values in the Atlantic relative to the Pacific, may be explained based on the sensitivity of the nitrifier N2O yield to O2. However, denitrification and/or coupled nitrification/denitrification cannot be ruled out as a significant source because these mechanisms may be responsible for producing some of the N2O that is widely exported from the ETP.
 We would like to thank NOAA ESDIM for the funding that finalized the RITS-89 cruise data, J. David Nance for his work on this data set, Brad Hall, Brad Halter, Chris Harth, and Jürgen Lobert for their assistance in collecting these data, the captain and crews of the Akademik Korolev, FS Polarstern, and NOAA Ship Discoverer for their assistance at sea, and the Atmospheric Chemistry Project of the NOAA Climate and Global Change Program for their support of the work of J. H. B. and J. W. E. We appreciate the funding of the RITS cruise of 1989 by the Radiation of Important Trace Gas Species Program (RITS) and D. L. Albritton of NOAA's Aeronomy Laboratory. We are grateful to Dileep Kumar, Cliff Law, Gernot Friederich, Martin Vollmer, and Lou Gordon for sharing their subsurface N2O data and to Scott Doney and Keith Lindsay for providing ocean carbon cycle model results. C. D. N. would like to thank Ray Weiss and NSF grant OCE0096404 for their support of her work.