Corresponding author: R. C. Hamme, School of Earth and Ocean Sciences, University of Victoria, P.O. Box 1700 STN CSC, Victoria, British Columbia, V8W 2Y2, Canada. (firstname.lastname@example.org)
 Some estimates of the budget of bioavailable nitrogen (fixed-N) suggest that the oceanic nitrogen cycle is grossly out of balance. We use observations of dissolved N2/Ar ratios along abyssal flow paths in the ocean to investigate fixed-N loss by benthic denitrification, one of the largest uncertainties. Dissolved N2/Ar in the deep-sea increases from the North Atlantic to the North Pacific. At depths >3500 m, this can be explained by the mixing of low N2/Ar source waters from the North Atlantic with high N2/Ar source waters from the Southern Ocean. Benthic denitrification in sediments bathed by these abyssal waters is below the detection limit. However, measureable increases in N2/Ar at depths of 2000–3000 m between the subtropical and subarctic North Pacific, regions that share the same source water, must be caused by benthic denitrification. The Cascadia Basin, with high denitrification rates and connection to the open North Pacific, is a likely source.
 Nutrient availability controls surface productivity in much of the ocean. The ocean inventory of nitrogen-containing nutrients, such as nitrate and ammonia, is largely controlled by inputs from N2 fixation and losses from denitrification that returns fixed-N to N2 [Gruber, 2004; Codispoti, 2007]. The global rate of denitrification has been estimated at 230 to 450 Tg N yr−1 or more, while the rate of nitrogen fixation and other inputs has been estimated at about 250 Tg N yr−1 [Gruber, 2004; Codispoti, 2007; DeVries et al., 2012a]. The larger denitrification estimates [Codispoti, 2007; Brandes and Devol, 2002] would result in an oceanic nitrogen cycle that is grossly out of balance, with net loss rates potentially consuming all oceanic fixed-N in ~2500 years. An imbalance of this size seems inconsistent with a relatively constant global nitrate inventory over the last 10,000 years, based on the stability of the nitrogen isotope content of sedimentary organic matter [Altabet, 2007] and on atmospheric pCO2 recorded in ice cores [Gruber and Sarmiento, 1997]. Estimates of the rates of both fixed-N sources and sinks have increased over the past 20 years, but estimates of benthic/sedimentary denitrification have increased fastest [Liu and Kaplan, 1984; Middelburg et al., 1996; Codispoti, 2007] and are the most poorly constrained [DeVries et al., 2012a]. To determine if the nitrogen cycle is in balance and the impact of any imbalance on global productivity, it is essential to bring all available tracers of these processes, particularly benthic denitrification, to bear on quantifying the fixed-N budget.
 A relatively recent tracer that can be used to constrain the marine fixed-N budget is the dissolved N2/Ar ratio. Several reaction pathways remove fixed-N, but all result in the conversion of fixed-N to N2 if they go to completion [Codispoti, 2007]. (In this paper, we refer to all pathways that produce N2 as denitrification.) Observations of N2/Ar have already been used in oxygen minimum zones in the Arabian Sea and the eastern tropical Pacific to estimate water-column denitrification at depths above ~1000 m [Devol et al., 2006; DeVries et al., 2012a]. Below these depths, the open ocean is sufficiently oxygenated to prevent fixed-N loss in the water column. However, oxygen concentrations may decrease enough within the sediment to allow significant benthic denitrification.
 In this paper, we explore the use of dissolved N2/Ar measurements in the water column, outside the major oxygen minimum zones, to constrain benthic denitrification in the deep ocean. New deep water is created in just a few locations globally, with N2/Ar signatures characteristic of the physical processes during formation. If benthic denitrification rates are significant, the N2/Ar ratio will increase away from these source regions along flow paths through the abyssal ocean, as N2 diffuses from the sediments into the overlying water column. To date, benthic denitrification estimates have been mainly derived from combining water-column estimates with benthic to water-column denitrification ratios based on isotopic constraints [e.g., Brandes and Devol, 2002; Altabet, 2007] and from sedimentary diagenesis models [e.g., Middelburg et al., 1996; Bianchi et al., 2012]. Observations of N2/Ar offer a means at an independent estimate of benthic denitrification.
 New mass spectrometric techniques have resulted in the ability to make the extremely high-precision measurements of dissolved N2/Ar necessary to observe the small changes outside major denitrifying zones. We present data from 30 cruises collected over the period 2000–2010. Water samples were introduced from Niskin bottles, through CO2-flushed tubing, into evacuated 160 mL flasks until half full [Emerson et al., 1999]. These flasks were pre-poisoned with dried HgCl2 and sealed with Louwers-Hapert O-ring stopcocks. CO2 was held in the flask neck with a vinyl cap to further preserve the sample from atmospheric contamination. Water in the flask was equilibrated with the headspace for at least 8 h in a constant-temperature rotating bath and then removed. Headspace gases were purified through liquid N2 and concentrated into valve-sealed tubes with liquid He. Samples were analyzed for O2/N2/Ar ratios on an isotope ratio mass spectrometer (Finnigan MAT 251 or Delta XL at UW, MAT 253 at UVic) against standards of similar composition. Minor corrections were made for O2 differences between samples and standard [Emerson et al., 1999].
 On samples from cruises after 2004, a 36Ar spike was added and the Ar concentration was determined by isotope dilution [Hamme and Severinghaus, 2007]. In the other samples (Hawaii and the two sites in the North Atlantic), the Ar concentration was determined from O2/Ar ratios and O2 concentrations from Winkler titrations. We find a mean 0.4% ± 0.5% offset between these methods.
 We express our measurements as a normalized ratio, the N2/Ar supersaturation, calculated as the anomaly from the ratio expected at equilibrium:
where (N2/Ar)equil is the N2/Ar ratio at equilibrium with the atmosphere for the potential temperature and salinity of the water mass [Hamme and Emerson, 2004] and ∆N2/Ar is presented in percent (×100). We also present individual gas supersaturations in the same way, e.g., ∆N2 = (N2)meas / (N2)equil − 1, and presented in percent (×100).
 Samples were collected and analyzed in duplicate. Air leaks into flasks increase the ∆N2/Ar ratio but can be excluded based on duplicates. We only present data where both duplicates were available (i.e., neither sample lost) and where duplicates agreed to within a standard deviation of 0.17% for ∆N2/Ar (2.5 times the final pooled standard deviation). This reduced the size of the data set by 18%. Precision in the reduced ∆N2/Ar data set is characterized by a 0.068% pooled standard deviation of duplicates. Absolute Ar measurements are inherently less precise than ∆N2/Ar, so for these, we applied a restriction that duplicates agree to within a standard deviation of 0.35%. The pooled standard deviation of ∆Ar duplicates by the spike method was 0.14%.
 Observations collected globally over the last decade show that ∆N2/Ar in the deep subarctic North Pacific (>1500 m) is 0.36% higher than in the deep North Atlantic (Figure 1). These data further show a measurable 0.17% increase between tropical/subtropical and subarctic sites in the North Pacific in the 2000–3000 m depth range. We particularly highlight data near Bermuda in the Atlantic, because this station integrates multiple North Atlantic water masses and was a particularly precise data set, and near Hawaii and Station P (Papa) in the Pacific, because these areas represent older ocean water and data at these locations can be averaged from multiple cruises (12 and 5 cruises, respectively). The differences between the Atlantic and Pacific are even more striking when comparing sites at the same neutral densities rather than depths (Figure 1c).
4 Processes Controlling Dissolved N2/Ar
 Both biological and physical processes affect N2 concentrations in the ocean. Our use of the normalized ∆N2/Ar ratio removes some of the physical effects. Dissolved gas concentrations are highly correlated to temperature due to the temperature dependence of gas solubility. Normalization by the concentration of the gas expected at equilibrium removes this temperature effect. During water-mass formation, changing atmospheric pressure, rapid cooling of surface waters, and injection of air bubbles affect gas saturations [Hamme and Severinghaus, 2007]. Both atmospheric pressure and cooling affect N2 and Ar supersaturations similarly, causing a strong correlation between ∆N2 and ∆Ar in our data, with a slope near 1 (Figure 2a). The generally low atmospheric pressure at water-mass formation regions affects the saturation of all gases equally. The rapid cooling that characterizes water-mass formation creates undersaturation with respect to equilibrium with the atmosphere because gases are more soluble at colder temperatures [Hamme and Severinghaus, 2007]. However, the temperature dependency of N2 and Ar solubilities are similar, so cooling also affects these gases almost equally. Because of these similarities, the ∆N2/Ar ratio is relatively insensitive to both atmospheric pressure and cooling (Figure 2b).
 The only physical process that dramatically alters ∆N2/Ar is injection of excess air. Breaking waves push bubbles beneath the water where higher pressure forces gases into solution. Melting at depth in the ocean of glacial ice shelves containing air bubbles creates a similar effect [Schlosser et al., 1990; Hohmann et al., 2002]. Essentially, air injection adds gases with an N2/Ar ratio of 84 to a dissolved pool with a ratio near 37. Of all the processes that affect ∆N2/Ar ratios observed in the deep sea, net denitrification and bubble injection are the most likely to cause large changes (Figure 2b).
 We observed the highest ∆N2/Ar ratios in the densest waters of the Southern Ocean and South Atlantic and the lowest deep water ∆N2/Ar in the North Atlantic (Figure 1c). Melting of glacial ice shelves around Antarctica injects air into Southern Ocean waters [Schlosser et al., 1990; Hohmann et al., 2002] and could enhance the Southern Ocean ∆N2/Ar ratio. A rough estimate of the contribution of glacial meltwater required to create the observed signal demonstrates this. Using a mean air content of Antarctic glacial ice of 0.11 cm3 g−1 [Hohmann et al., 2002], we estimate that a 0.17% addition of pure glacial meltwater would raise the dissolved ∆N2/Ar ratio by 0.6%, the observed difference between the deep North Atlantic and the densest water we measured in the Southern Ocean. Estimates of glacial meltwater contributions to Antarctic Bottom Water (AABW) are 0.08%–0.14% based on the δ18O of H2O but with significant uncertainty [Weiss et al., 1979; Jacobs et al., 1985]. The potential effect is the right order of magnitude to explain the high ∆N2/Ar ratios in the deep Southern Ocean.
5 Source Water Contributions
 In our observations, the deep Pacific Ocean has a ∆N2/Ar ratio midway between the deep North Atlantic and the densest waters of the Southern Ocean (Figure 1). To what extent can this be explained by mixing of source waters? The deep Pacific is filled from the south by Circumpolar Deep Water, a combination of AABW, North Atlantic Deep Water (NADW), and intermediate water masses at shallower depths [Broecker et al., 1985]. A potential temperature-salinity diagram shows that our data captures much of the typical mixing trend among water masses in the Southern Ocean (Antarctic deep trend in Figure 3a), though we did not sample pure AABW.
 The Circumpolar Deep Water mixing trend has a much larger change in temperature than in salinity, so we isolate the effects of mixing by plotting potential temperature vs. ∆N2/Ar. An approximately linear trend connects low ∆N2/Ar / high potential temperature NADW with high ∆N2/Ar / low potential temperature near-AABW (Figure 3b) and suggests that pure AABW has a ∆N2/Ar of 1.9%. Because of similar temperature dependencies, ∆N2/Ar would mix linearly within 0.01% for these waters. The deeper samples from all our profiles (>3500 m) lie on this linear trend. We conclude that simple mixing of different source waters, which are differentially affected by air injection during water-mass formation, can explain the deepest ∆N2/Ar data.
 A simple calculation demonstrates that low rates of benthic denitrification in abyssal sediments should be detectable by this method over the long time scales of ocean circulation. Suppose benthic denitrification removed fixed-N at a rate of 18–28 Tg N yr−1 in sediments deeper than 3000 m; 18 Tg N yr−1 is the estimate from Middelburg et al. , 19 Tg N yr−1 from Bianchi et al. , and 28 Tg N yr−1 from DeVries et al. [2012b]. We would then expect to observe ∆N2/Ar increasing at a rate of 0.25%–0.39% per 1000 years of abyssal circulation.
where 6 × 10−4 mol N2 kg−1, the solubility of N2 at a potential temperature of 1.6°C and salinity of 34.7, converts the N2 concentration increase to a N2/Ar increase, and 4.27 × 1020 kg is the mass of the ocean below 3000 m [Becker et al., 2009]. An increase of this magnitude would create a measureable deviation from our source water mixing line (Figure 3b), so abyssal rates are likely lower than this between the North Atlantic near Bermuda and the North Pacific.
6 Denitrification Contributions
 Our measurements also demonstrate a 0.17% difference in ∆N2/Ar between the tropical/subtropical and the subarctic North Pacific, centered at neutral densities of 27.9 and depths of 2000–3000 m (Figure 1). With 12 repeat cruises from the subtropics near Hawaii and 5 from the subarctic at Station P, the standard deviation of the mean for each depth bin at these sites is 0.03–0.05 %, allowing us to detect this small ∆N2/Ar difference between sites. The circulation of the North Pacific at these depths is complex with recirculations and return flow of deep Pacific water to the south. However, appreciable deep water does not form in the North Pacific, so approximately the same source water must feed all the deep North Pacific sites. Given this, the observed difference in ∆N2/Ar between the subtropical and subarctic North Pacific must be caused by biological input from denitrification. Moreover, the signal we observe in ∆N2/Ar is supported by observations of lower nitrate relative to phosphate concentrations at these depths in the subarctic North Pacific suggesting removal of fixed-N [Deutsch et al., 2001], though this is near the detection limit for the nutrient-based method.
 The reason for a denitrification signal at 2000–3000 m is not immediately obvious. The depth of the ∆N2/Ar difference is much deeper than either the oxygen minimum in the subarctic North Pacific at ~1000 m or the anoxic waters in the eastern subtropical North Pacific where water column denitrification occurs. On the other hand, if benthic denitrification throughout the North Pacific caused this denitrification signal, we might expect it near the bottom of the water column, where the largest proportion of seafloor is [Becker et al., 2009].
 Instead, the location of the subtropical/subarctic ∆N2/Ar difference suggests a denitrification source at mid-depth, with either the Cascadia Basin or the Bering Sea as potential candidates. The Cascadia Basin is located offshore of the state of Washington and contains the largest swath of seafloor at 2000–3000 m in the open North Pacific. The Aleutian Islands separate the Bering Sea from the open Pacific, but deep water exchanges through Kamchatka Strait with outflows from the Bering Sea into the western subarctic North Pacific as deep as 3000 m [Reed et al., 1993]. Significant rates of benthic denitrification have been observed at depths of 2000–3000 m in both locations [Engström et al., 2009; Lehmann et al., 2005]. Additionally, the Bering Sea is depleted in fixed-N throughout its full water column [Gruber and Sarmiento, 1997; Deutsch et al., 2001]. The contribution of these regions to the open ocean can be traced through dissolved silica concentrations because both the Cascadia Basin and Bering Sea have high silica in their deep waters. Hydrographic data shows a plume of high silica at 2000–2500 m stretching west from the Cascadia Basin into the North Pacific, but there is no indication of a similar plume from the Bering Sea [Talley and Joyce, 1992; Johnson et al., 2006]. This pattern suggests that water from the Cascadia Basin reaches our study site at Station P, whereas Bering Sea water does not. This makes Cascadia Basin the more likely source of the deep denitrification signal in the subarctic North Pacific, but more N2/Ar data is needed to confirm this.
 The global ∆N2/Ar data set presented here, unprecedented in precision and global coverage, shows clear differences among regions. The patterns in ∆N2/Ar in the deepest waters (>3500 m) from the North Atlantic through the Southern Ocean and into the North Pacific can be explained by the mixing of abyssal source waters with different initial signatures. However, differences at depths of 2000–3000 m between the subtropical and subarctic North Pacific must be caused by biological input of N2 during denitrification because these waters have the same origin. It should be possible to further deconvolve the impact of source waters vs. biological denitrification using measurements of other noble gases that are sensitive to air injection [Hamme and Severinghaus, 2007]. Our interpretation of this data set introduces the utility of this new tracer, dissolved ∆N2/Ar, to constraining denitrification outside the confined areas of the world's oxygen minimum zones. To obtain rates, these measurements will need to be paired with other inert gas tracers and accurate estimates of water mass movement, likely using circulation models.
 We thank C. Stump, K. Giesbrecht, and D. Wilbur for their expertise in making these measurements. We also thank J. Klymak, N. Gruber, and an anonymous reviewer for comments that improved our interpretation. This research was supported by NSF OCE-1029299 to S.R.E. and NSERC DG 328290–2006 to R.C.H.