Nitrite Accumulation and Nitrate Deficit
 In essence, the entire depth interval below 100 m and extending to our deepest sample at 1200 m potentially was a sink for nitrate, nitrite, and ammonium. Based solely on oxygen concentrations, heterotrophic denitrification may occur in the interval from 100–175 m down to 625–1025 m (Table 1). Anammox could occur roughly between 100 m and 1200 m at our study sites as oxygen concentrations in that interval are below the anammox threshold of O2 ~10 µM [Jensen et al., 2008] and seem to be limited mainly by the availability of ammonium.
 A plot of phosphate concentrations versus nitrate concentrations (Figure 4) shows that our samples have excess phosphate compared to the global average [Gruber and Sarmiento, 1997] and that most samples are nitrate deficient compared with the regional Redfield ratio adjusted for the preformed nitrate deficit of the Arabian Sea [Codispoti et al., 2001]. The nitrate deficit of virtually all samples illustrates that denitrification has depleted nitrate relative to phosphate in the entire water column down to our deepest samples at 1200 m.
Figure 4. Correlation of nitrate versus phosphate concentrations. All samples plot above the mean oceanic ratio of [NO3−] = 16 [PO43−] + 2.9 (dashed line; Gruber and Sarmiento, 1997), and most samples plot above the ratio of [NO3−] = 14.89 ([PO43−] − 0.28) * 0.86 typical for the Arabian Sea (solid line; Codispoti et al., 2001). Color coding of dots indicates the potential density σθ [kg m−3]. Strongest nitrate deficiencies are observed in PGW, ICW, and part of the RSW.
Download figure to PowerPoint
 There are local deviations in the extent of the ODZ and of the nitrate-deficient depth interval: oxygen supply by the PGW entrained into the ODZ water depth interval is very likely responsible for the elevated oxygen concentrations of >10 µM and the suppression of denitrification between 200 m and 350 m water depth at stations #944 and #947 (Figure 3a). O2 concentrations are between 30.5 and 0.7 µM in the PGW [Morrison et al., 1999], and a branch of this water mass has been observed to flow along the coast of Oman and to spread eastward into the central basin [Morrison, 1997; Prasad et al., 2001]. Furthermore, the NO3def is greater at the western and northwestern stations (#949, #950, and #957) than in the eastern part of the Arabian Sea (stations #953 and #955) (Table 1). The main reason for this difference is the larger nitrate deficit in the water masses below 300 m of the western stations, which is probably preformed [Mantoura et al., 1993]. Only a small amount of nitrite builds up at station #949, whereas it comprises more than 20% of the NO3def at station #957 (Table 1). Low nitrite concentrations have previously been reported from the western Arabian Sea, although oxygen concentrations are below the threshold for denitrification [Mantoura et al., 1993]. The virtual absence of nitrite in the western region (in which #949 is situated) may be due to either complete denitrification, which would then increase the NO3def or to nitrification/reoxidation to nitrate, or else to continuous consumption of nitrite by anammox.
Isotopic Evidence for Heterotrophic Denitrification and Parallel Nitrite Oxidation
 Denitrification is indicated by nitrite accumulation and by maxima of the two stable nitrate isotope ratios between 150 m and 400 m water depth. Midwater denitrification is a process acting on the 18O/16O and 15N/14N mixtures in source nitrate with similar fractionation factors 15ε and 18ε which are between 20 and 30‰ [Altabet et al., 1999; Barford et al., 1999; Granger et al., 2008; Sigman et al., 2009a; Voss et al., 2001]. The “Rayleigh” fractionation model is often used to calculate apparent fractionation factors from δ15NNO3 values [Altabet et al., 1999; Brandes et al., 1998]. This model assumes that the nitrate pool is advected into the ODZ and is not replenishment via isopycnal mixing. The fractionation factors obtained by the Rayleigh model are generally lower than by models taking diffusive replenishment of nutrients into account [Brandes et al., 1998]. Highest isotopic effects are obtained by the so-called “open system model” (Texts S1 and S2 and Figure S2 in the supporting information). When using the Rayleigh fractionation (described by Mariotti et al., 1988), the stable isotope value of the substrate increases according to
with δ15Ninitial being the stable isotope value of the initial product and f being the fraction of the original substrate remaining. The instantaneous product is calculated as
and the accumulated product is calculated as
 The same equations can be used for δ18ONO3 measurements, but they are not applied to δ18ONO3 + NO2 due to the sample storage artifact discussed above. Values for f are calculated according to
when using δ15NNO3 + NO2 and according to
when using δ18ONO3 and δ15NNO3.
 The 15ε and 18ε are obtained from the slope of ln(f) plotted against δ18Osubstrate or δ15Nsubstrate. The ε values obtained in our study are mostly higher than those found in incubation experiments [Barford et al., 1999; Granger et al., 2008] (Table 3). Most differences between δ15NNO3 and calculated δ15NNO2 (Δδ15N; Table 2) are even higher than our calculated 15ε (Table 3). Moreover, the 15ε/18ε of 1.2 is higher than the expected value of 1 (Table 3). Nitrate and nitrite isotopic compositions in the ODZ are thus not consistent with a sole denitrification effect and imply fractionating reactions other than or additional to denitrification as well as reactions that decouple nitrogen and oxygen isotopic compositions in nitrate and nitrite. The only process known to date which could explain both the low nitrite N-isotope values and a decoupling of N and O isotopic fractionation in nitrate is nitrite oxidation. This process shows a rare inverse isotope fractionation, with initial products being isotopically enriched relative to their source signature, depleting the source nitrite of heavy isotope species. A decoupling of oxygen versus nitrogen isotope signatures can occur due to different isotope effects for N and O during nitrite oxidation and due to the incorporation of oxygen from external sources [Buchwald and Casciotti, 2010; Casciotti and McIlvin, 2007; Sigman et al., 2005].
Table 3. Fractionation Factors ε by the Rayleigh Fractionation Model Derived From the Slope of a Plot of ln( f ) Against δ15NNO3 + NO2, δ18ONO3 or δ15NNO3, Respectively (See Text for Explanation of the Rayleigh Model)
|Nitrate + Nitrite|
 Plots of paired δ15NNO3 + NO2 and δ18ONO3 + NO2 values for depths ≥100 m, as well as δ15NNO3 and δ18ONO3, reveal a deviation from the expected 1:1 line at intermediate values (Figures 5a and 5b).
Figure 5. (a) δ15NNO3 + NO2 [‰] plotted against δ18ONO3 + NO2 [‰] and (b) δ15NNO3 [‰] plotted against δ18ONO3 [‰]. Color coding of dots indicates the potential density σθ [kg m−3]. The 1:1 line corrected for the deepwater offset between paired isotopes of 2‰ is indicated.
Download figure to PowerPoint
 A term to express this decoupling of nitrate N and O isotopes was defined by Sigman et al.  as
 For the ratio 15ε/18ε, the isotopic effects of denitrification with 15ε = 18ε are inserted so that 15ε/18ε is about 1, and any deviation of the Δ(15,18) from 0 should reflect processes during which δ15N and δ18O ratios are affected differently. We used the average deepwater nitrate stable isotope ratios from the individual stations, as we have no end member of source waters, and can assume that their stable isotope ratios are elevated above the oceanic average (see above). Most Δ(15,18)NO2 + NO3 and Δ(15,18)NO3 are negative in the interval from 100 m to 1000 m water depths, meaning that O isotope values are relatively more enriched than N isotope values. In the case of Δ(15,18)NO2 + NO3, this is partly an artifact of oxygen exchange of nitrite with water during sample storage, which can lead to elevation of δ18ONO2 toward and above the isotope value of ambient water due to equilibrium fractionation [Casciotti et al., 2007].
 To attain negative Δ(15,18)NO3, an external oxygen source is required. One important source of isotopically enriched oxygen is the exchange of oxygen between 18O-depleted nitrite and ambient water. This is an abiotic process depending on temperature, pH, and the residence time of nitrite, and it leads to a convergence of δ18ONO2 with the δ18O of water [Kool et al., 2007; Snider et al., 2010]. Recent experiments indicate that oxygen equilibration with water can affect 25% of nitrite [Buchwald and Casciotti, 2010; Casciotti et al., 2010]. This relative enrichment of δ18ONO2 in relation to δ15NNO2 can be transferred into the nitrate pool if nitrite is reoxidized.
 Differences in the negative kinetic effects on nitrogen and oxygen isotopes during nitrite oxidation, with N and O isotope effects of ~ −13‰ and ~ −7‰, respectively [Buchwald and Casciotti, 2010; Casciotti, 2009], would further deplete both nitrite isotopes and, additionally, increase δ18ONO2 relative to δ18NNO2.
 In our setting, however, values for Δ(15,18)NO3, where nitrite was removed by ascorbate treatment, are also negative and mostly close to Δ(15,18)NO2 + NO3. We consequently surmise that, barring some uncertainty regarding the abiotic equilibration of nitrite oxygen with ambient water, the two isotopes are truly decoupled (Figure 3b) owed to differences in biological processing. In the center of the nitrite maximum, the Δ(15,18) values are close to 0 suggesting that denitrification is the only significant process affecting Δ(15,18) and nitrification is virtually absent (Figure 3b and Table 2).
 The four stations with significant nitrite accumulation >1 µM in the SNM (Figure 3b) show minimum Δ(15,18)NO3 + NO2 and Δ(15,18)NO3 values at 150 m and 400 m water depth, i.e., above and below the zone with highest isotope values in the core of the denitrification maximum. We interpret this as a nitrification signal, which, as discussed above, differentially modifies nitrite oxygen and nitrogen isotopes, promoting a deviation of Δ(15,18) from zero. This observation agrees with the model of Anderson et al. , who assumed that nitrification occurs adjacent to the zone of denitrification, i.e., between 300 m and 400 m water depth and above 150 m, producing nitrate that diffuses back into the denitrification zone and further fuels denitrification. The tight spatial coupling of nitrification and denitrification may be due to the presence of microenvironments with locally higher oxygen concentrations or due to oxygen diffusion across the oxycline. The lower boundary of the denitrification zone coincides with a small but distinct oxygen peak at 400 m at all stations (Figure 3b), which we attribute to lateral advection of a relatively more oxygenated water mass such as PGW. Furthermore, little oxygen is needed for nitrification to occur: direct measurements of nitrite oxidation rates from the ETSP and the Benguela upwelling system reveal significant nitrite oxidation at O2 concentrations below 2.5 µM [Füssel et al., 2011; Lipschultz et al., 1990]. Nitrite oxidation is independent of prior ammonia oxidation if nitrite is provided by nitrate reduction.
 The production of low Δ(15,18) by nitrification under suboxic conditions helps to interpret the deviating nitrite concentration profiles at stations #949 and #950 (Figure 3b). Nitrite concentration in the SNM at #949 is very low, and the ODZ maximum values of δ15N are significantly lower than at the other locations. But the NO3def is up to 11 µM (Figure S1 in the supporting information), indicating that denitrification has been equally effective as at the other locations. The low nitrite concentrations and relatively depleted δ15NNO3 and δ18ONO3 between 150 m and 500 m water depth indicate that intense nitrification has indeed taken place as suggested by Lam et al.  and that the complete oxidation of depleted nitrite has lowered the δ15NNO3 and δ18ONO3. This unusual pattern may be due to better ventilation of the western part of the basin by PGW or ICW (see above). At station #950, where a mesoscale eddy has eroded the thermocline and mixed oxygen down to 220 m, the layer affected by the eddy is nitrite free and has relatively low isotope values but a pronounced Δ(15,18) minimum as a legacy of nitrification under initially suboxic conditions.
 Based on the concept that low Δ(15,18) thus marks zones of ongoing nitrification, Figure 6 is a schematic plot of variables and processes in the ODZ. Denitrification dominates in the zone between about 200 m and 300 m water depth, where nitrite accumulates and nitrate isotopic ratios attain maxima. This process may also be coupled to the anammox reaction, because ammonium produced by heterotrophic denitrification does not accumulate to any extent. Removal of nitrite by anammox explains the bimodal nitrite peak at locations #953–957, where we observe a relative concentration minimum in the core of nitrite accumulation and a slight isotopic enrichment of nitrate at about 250 m (Table 2). Nitrification is indicated by declining Δ(15,18)NO3 and low δ15NNO2 values in the upper ODZ just below the oxycline and below the core of the denitrification zone. Nitrification probably also adds to the bimodal pattern of nitrite accumulation that has maxima on either side of the denitrification layer. Because nitrite is an intermediate of both nitrification and denitrification, it may accumulate where microenvironments and diffusion allow close coupling of the two processes. Whereas the oxygen concentrations show a steep increase above 100 m, they drop to almost 0 µM below the small oxygen peak at 400 m. We believe that nitrification occurs only in a small layer around 400 m where oxygen remains available. Mixing and diffusion lead to the linear increases of Δ(15,18) above and below the two nitrification layers.
Figure 6. Schematic picture of the processes in the water column using station #953 as an example. The PNM is at the base of the mixed layer. The oxycline is the zone of sharply decreasing oxygen concentrations and of the upper δ15NNO3 + NO2 and δ18ONO3 + NO2 minimum. The zone dominated by denitrification is evident from nitrite accumulation, maxima of δ15NNO3 + NO2 and δ18ONO3 + NO2 as well as high Δ(15,18). Denitrification is coupled to anammox to remove the ammonia released during organic matter oxidation. Above and below this are zones dominated by nitrification indicated by minima of the Δ(15,18) which may co-occur with denitrification. The bimodal nitrite maxima may be a product of both nitrite accumulation during incomplete denitrification and nitrification. Removal of nitrite by anammox in the core of the denitrification zone could also explain the small nitrite minimum at 250 m.
Download figure to PowerPoint
Box-Model-Derived Mass Balance and Stable Isotopic Ratios in the ODZ
 In order to obtain a rough estimate of the amounts of reactive nitrogen taking part in nitrate reduction, denitrification, and nitrite oxidation, we produced a box model for the upper part of the ODZ (Figure 7 and Text S2 in the supporting information). It encompasses the secondary nitrite maximum and ranges from a water depth of 125 m to 350 m. The areal extent of the secondary nitrate maximum is 1.95 × 1012 m2 [Naqvi, 1987] so that the box volume becomes 4.38 × 1014 m3. The inflow and outflow of water are calculated by dividing the volume of water by the mean residence time of water within the ODZ. As the residence time of ODZ water, we used an estimate of 10 years [Olson et al., 1993]. Nitrate inputs are calculated by multiplying the total inflow with a nitrate concentration of 34 µM; the δ15NNO3 is assumed to be 7‰. The second source of nitrogen is organically bound nitrogen exported from the euphotic zone. There are a variety of equations allowing a calculation of the organic carbon (OC) flux at all water depths if the primary production (Pp) is known. Based on satellite-derived data on primary production and results derived from sediment trap experiments, such an equation was adapted to the condition found in the Arabian Sea (OC = 0.01 Pp2/water depth0.628) [Rixen et al., 2002]. The mean primary production in the Arabian Sea of 180 g C m2 yr−1 is used to calculate the organic carbon flux at the top (125 m) and the bottom (350 m) of the modeled box. The difference between these fluxes is the amount of organic matter remineralized within the box; this number is converted into nitrogen by using the Redfield ratio. Two different remineralization processes are considered: (i) nitrification which is a chain of reactions in the course of which NH4+ is converted via NO2− (ammonia oxidation) to NO3− (nitrite oxidation) and (ii) heterotrophic denitrification which can be described by two reactions which are the NO3− reduction (equation (12)) followed by the NO2− reduction to N2O and N2 (equation (13)):
Figure 7. Box model of N cycling in the Arabian Sea ODZ (see further explanation in the text and in Text S2 in the supporting information). Unitless numbers are fluxes in 1011 mol N yr−1. Nitrogen sources are advection of nitrate and sinking organically bound nitrogen. Nitrogen sinks are heterotrophic denitrification and lateral export of nitrate and nitrite (all in blue). Red colors indicate concentrations of nitrate and nitrite in the ODZ and their stable isotopic values and correspond with measurements at location #953.
Download figure to PowerPoint
 The formation of N2 and the outflow of reactive nitrogen are the two sinks balancing the nitrogen inputs. The N2O formation during denitrification is ignored because N2O, in spite of being an important green house gas, is quantitatively of low importance for the nitrogen budget in the ODZ.
 All together, 1000 model years were calculated to get the model into equilibrium whereby the conditions were changed in five steps after steady state was reached for the set conditions (Figure 8 and Text S2 in the supporting information). For the first 200 model years, there was no remineralization of organic nitrogen, and the mean NO3− concentration attained a value of 31 µM, NO2− concentrations were 0 µM, and δ15NNO3 was 7‰ as prescribed. In the second step after the model year 200, the entire organic nitrogen was decomposed and converted into NO3− via complete oxidation of organic nitrogen (complete nitrification). Accordingly, the NO3− concentration increased by approximately 4 µM, and the δ15NNO3 was reduced slightly, as the lighter 14N is preferentially decomposed. An isotopic fraction factor of 15ε = 2‰ [Möbius, 2013] was used as the isotopic effect of ammonification. In a third step, nitrification was reduced, and the remaining organic nitrogen was remineralized by denitrification. Consequently, the NO3− concentration decreased, and the NO2− concentration increased. Assuming a fractionation factor of 15ε = 25‰ for denitrification which is in the range of those determined experimentally and empirically [Altabet et al., 1999; Barford et al., 1999; Granger et al., 2008; Sigman et al., 2009a; Voss et al., 2001], the residual nitrate became heavier and reached values around 20‰ as measured in profile #953 (Table 2). Assuming furthermore that approximately 24% of the produced NO2− is not reduced to N2, the accumulating NO2− attains concentrations of around 5 µM which also agrees quite well with measurements of profile #953. However, the resulting δ15NNO2 is −6‰ which is much higher than the measured δ15NNO2 of ~ −16‰. Assuming, in a fourth step, that the entire available organic nitrogen was utilized by denitrification, the δ15NNO2 further increases. Considering, in the fifth step, that 60% of the produced NO2− is not reduced to N2 and that 40% of the accumulated NO2− is reoxidized to NO3− with an isotopic effect of −13‰ [Buchwald and Casciotti, 2010; Casciotti, 2009], the resulting δ15NNO2 values are similar to the ones measured in profile #953. Thus, a reoxidation of 24% of initial nitrite appears to be the process explaining the low δ15NNO2 values in the upper part of the ODZ.
Figure 8. Model run of 1000 years changing the conditions in 200 year steps: (I) Accumulation of nitrate concentrations by advective inflow. (II) Nitrification of settling organic matter so that the remineralized organic nitrogen is added to the nitrate pool. (III) Denitrification starts; part of the organic nitrogen is oxidized by nitrification, and part of it is respired by denitrification; 24% of the nitrite is not further denitrified to dinitrogen but accumulates in the water column. (IV) The entire organic matter decomposition is carried out by heterotrophic denitrification; nitrite is allowed to accumulate as in step III. (V) 40% of the nitrite produced during nitrate reduction is denitrified and released as dinitrogen gas, and 40% of the remaining nitrite is reoxidized to nitrate (i.e., 24% of initially produced nitrite). Concentrations of nitrate and nitrite (in μM) are shown in black; their respective stable isotopic ratios are shown in red. The assumed nitrification, denitrification, and nitrite oxidation rates are shown in Tg N yr−1.
Download figure to PowerPoint