Global Biogeochemical Cycles

N isotopic composition of dissolved organic nitrogen and nitrate at the Bermuda Atlantic Time-series Study site

Authors


Abstract

[1] To better constrain the dynamics of the dissolved organic nitrogen (DON) pool and the role of N2 fixation in the nitrogen cycle at the Bermuda Atlantic Time-series Study (BATS) site, we measured the 15N/14N ratio of total nongaseous nitrogen (TN) in the upper 250 m and of nitrate in the upper 1000 m of monthly water column profiles from June 2000 through May 2001. The annually averaged TN δ15N in the upper 100 m is 3.9‰, which is greater than thermocline nitrate (2–3‰ at 250 m) and similar to literature values for shallow sinking nitrogen at BATS (3.7‰). We discern no seasonal variation in TN δ15N, which suggests that most of the DON pool is recalcitrant on this timescale. The TN data require a δ15N for the sinking flux that is similar to previous measurements, suggesting that N2 fixation is a minor component of new nitrogen at BATS. Small but measurable differences in the concentration and 15N/14N of total organic nitrogen (TON) between the surface and subsurface (∼250 m) suggest that subsurface remineralization of ∼0.25 μM of the surface TON acts to lower the 15N/14N of nitrate in the thermocline at BATS.

1. Introduction

[2] Although decades of study have characterized the carbon and nitrogen budgets of the oligotrophic Sargasso Sea [Jenkins, 1988; Michaels et al., 1994a; Bates et al., 1996; Lipschultz et al., 2002], fundamental aspects remain poorly understood. In particular, studies at the Bermuda Atlantic Time-series Study (BATS) site have revealed a drawdown of dissolved inorganic carbon (DIC) from surface waters during the summer months for which one cannot account in terms of either the nutrients required to support the biological removal of the DIC or the export production that should result from the DIC decrease [Michaels et al., 1994a; Gruber et al., 1998]. In the absence of significant nitrate (NO3) input during this period, N2 fixation has been proposed as a means of fueling phytoplankton growth to remove the DIC. This hypothesis is supported by regional geochemical observations, including the high concentration of NO3 relative to phosphate (PO43−) in the North Atlantic thermocline [Michaels et al., 1996; Gruber and Sarmiento, 1997] and the low-15N/14N of NO3 in these same waters [Altabet, 1988; Karl et al., 2002] (see also Liu et al. [1996] for similar evidence from the Pacific).

[3] Despite these regional observations, local geochemical measurements have provided arguments against N2 fixation as the missing N source for the BATS summertime DIC drawdown. At BATS, Altabet [1988] observed a similar 15N/14N for the N sinking into shallow sediment traps (PNsink) and the NO3 being supplied to the surface from the thermocline (δ15N of 3.7 and 3.5‰, respectively; δ15N, in per mil versus atmospheric N2 = {[(15N/14N)sample/(15N/14N)atm] − 1} × 1000). Since newly fixed N has a δ15N in the range of ∼−3–0‰ [Hoering and Ford, 1960; Minigawa and Wada, 1986; Carpenter et al., 1997], Altabet [1988] concluded that there was little room in the isotope budget for N2 fixation to be quantitatively important as a source of N for new production at BATS.

[4] One caveat to this analysis is that dissolved organic nitrogen (DON) was not included as a potential flux term in the isotope budget. Since the work of Altabet [1988], it has been shown that N2 fixers can shunt a major fraction of newly fixed N into the DON pool, significantly increasing the ambient DON concentration ([DON]) [Karl et al., 1992; Capone et al., 1994; Glibert and Bronk, 1994]. Thus DON could potentially represent a significant flux of low-δ15N N out of the surface, similar to the seasonal C concentration dynamics of dissolved organic carbon (DOC) at BATS [Carlson et al., 1994]. If this occurs, it could allow the hypothesis that N2 fixation is the missing N source for the DIC drawdown at BATS to be consistent with the isotope data of Altabet [1988]. However, Hansell and Carlson [2001] have shown that, unlike DOC concentration, [DON] at BATS is remarkably stable throughout the year and not clearly different between the mixed layer and the underlying thermocline. Thus, if DON is important in the 15N budget for the upper 100 m, in the absence of a concentration gradient, there would need to be a large DON δ15N difference between the surface and subsurface waters.

[5] The low 15N/14N of NO3 in the thermocline of the North Atlantic and elsewhere in the subtropics has been interpreted as a sign that N2 fixation is occurring at rates sufficient to cause the accumulation of a sizable NO3 excess in the low-latitude thermocline (relative to expectations of NO3 concentration based on PO43− concentration and Redfield ratios [Liu et al., 1996; Brandes et al., 1998; Karl et al., 2002]). However, Altabet's [1988] measurements indicate that not only is N2 fixation not a dominant source of new N at BATS, but the PNsink δ15N at BATS is too high to generate the observed 15N-depletion of the thermocline NO3. That is, while the PNsink δ15N is essentially equal to the NO3 δ15N (3.7‰ and 3.5‰, respectively [Altabet, 1988]), the low NO3 δ15N at BATS ultimately requires a low-15N N source to lower its δ15N from that of deep ocean NO3 (∼5‰). One might suspect that the low δ15N of thermocline NO3 at BATS results from preferential remineralization of low-δ15N N as the sinking flux transits through the thermocline. However, the PNsink δ15N from sediment traps at different depths does not appear to support this hypothesis, since the PNsink δ15N does not typically increase with depth [Saino and Hattori, 1987; Altabet, 1988], and often appears to decrease with increasing depth [Altabet et al., 1991]. Thus we lack the direct mechanism by which the BATS thermocline NO3 develops its low δ15N. This is disconcerting for our efforts to make use of the NO3 δ15N data as a complement to the NO3 concentration to PO43− concentration ratio data.

[6] The goal of this manuscript is to add total N (TN) (specifically, DON, the dominant surface TN term) and a new suite of NO3 isotope measurements to the N isotope data that already exist for BATS, in order to increase the isotopic constraints on the N cycle there. Three questions are of central importance in this regard: (1) Does the DON δ15N at BATS and its variability indicate dynamic behavior in the DON pool not evident in the DON concentration data alone, in particular, in response to N2 fixation or the DIC drawdown? (2) How does the addition of the DON and new NO3 data change the N isotope budget constructed by Altabet [1988]? Specifically, does the budget now require significant inputs from a low-δ15N N source such as N2 fixation? (3) Do TN measurements indicate a means for transporting low-δ15N N from the surface to the thermocline that can lower the δ15N of NO3 in the thermocline at BATS?

2. Sampling Site and Sampling Protocol

[7] Samples were collected monthly at the BATS site, 31°50′N, 64°10′W, between March 2000 and May 2001 on board the R/V Weatherbird II. Samples were typically collected at depths of 0, 40 or 60, 100, 140 or 160, 200, 250, 300, 400, 500, 600, 800, and 1000 m; the same hydrocasts were sampled as part of the BATS core program. Seawater samples were collected in 60-mL HDPE bottles that had been soaked overnight in acid and rinsed with deionized water. Each bottle was rinsed with sample water three times before being filled. Samples were frozen at −20°C until analysis. NO3 concentration ([NO3]) and NO3 δ15N measurements were made on samples collected between 0 and 1000 m. The concentration and δ15N of TN was measured on samples from 0 to 250 m, and also 300-m samples for August 2000 and October 2000, when 250-m samples were not collected.

3. Methods

3.1. Nitrate

[8] The sum of the concentration of NO3 and nitrite (NO2) was measured by reduction to nitric oxide (NO) in a heated solution of acidic V(III), followed by chemiluminescent detection of the NO [Braman and Hendrix, 1989]. With the exception of rare wintertime events [Lipschultz et al., 1996, Lipschultz, 2001], NO2 is typically scarce in the water column at BATS and is thus subsequently disregarded. The NO3 δ15N was analyzed by quantitative bacterial reduction of NO3 to nitrous oxide (N2O) using a strain of denitrifier that lacks N2O reductase activity, followed by automated extraction, purification, and analysis of the N2O product by an isotope ratio mass spectrometer (the “denitrifier method” [Sigman et al., 2001; Casciotti et al., 2002]). Individual analyses are referenced to injections of N2O from a pure N2O gas cylinder and then standardized using an internationally accepted NO3 isotopic reference material (IAEA-N3, with a δ15N of 4.7‰ [Bohlke and Coplen, 1995]). Replicate analyses are generally consistent with a standard deviation for analysis of ±0.2‰. One difficulty with analysis of samples with 1–2 μM NO3 was producing standards uncontaminated with NO3; at this low concentration, a 0.1 μM NO3 background in the solvent water can yield a substantial isotopic difference from the NO3 standard.

3.2. Total Dissolved N

[9] We describe here a new method for the natural abundance-level isotopic analysis of total dissolved N (TDN, the sum of NO3, NO2, ammonium, and DON) in seawater. The method is a straightforward coupling of the commonly used “persulfate oxidation” method for TDN conversion to NO3 [Solorzano and Sharp, 1980; Bronk et al., 2000] with the denitrifier method for the isotopic analysis of NO3 described above.

3.2.1. Oxidation of TDN to NO3

[10] To oxidize TDN to NO3, 12 mL of sample is added to a boro-silicate glass test tube with Teflon lined cap. To this sample is added 2 mL of a persulfate oxidizing reagent (POR), which is made up daily with 6 g certified ACS-grade NaOH dissolved in 100 mL of deionized water (DIW), followed by the addition of 6 g certified ACS-grade K2S2O8 (potassium persulfate) (POR recipe adapted from Solorzano and Sharp [1980]). The potassium persulfate is recrystallized three times following a procedure described by Grasshoff et al. [1999]. Immediately after adding the POR to the sample, screw caps are closed tightly, and samples are autoclaved for 55 min on a slow vent setting. Finally, in each batch of samples, three test tubes with 12 mL of the POR are also autoclaved with the samples to determine the concentration and δ15N of the N contamination associated with the POR (subsequently referred to as the “reagent blank”).

[11] After the samples cool, the sum of the NO3 and NO2 concentration ([NO3] + [NO2]) of the samples and of the persulfate reagent is measured by chemiluminescence (further referred to as [NO3], since [NO2] was typically undetectable). For the samples, [TDN] is calculated by mass balance by subtracting the [NO3] of the autoclaved POR from the [NO3] of the post-oxidized sample, taking dilution into account. [DON] is calculated by subtracting the [NO3] of the seawater sample before the persulfate treatment from the [TDN].

[12] We found it necessary to be vigilant about contamination at virtually every step and in the use of every reagent, and to monitor the reagent/procedural blanks for each “batch” of samples [Hopkinson et al., 1993]. Glassware was washed with copious DIW, then soaked in soap and 10% HCl baths, and washed again with DIW after all steps. All nonvolumetric glassware was then combusted at 450°C for 4 hours. Volumetric glassware and black phenolic screw caps with Teflon liners were rinsed in DIW, then in a 10% HCl bath overnight, rinsed with DIW again, and dried at ∼50°C. Our water deionization system includes an UV lamp for oxidizing organic matter. Persulfate oxidation of our DIW indicated that its [TDN] was typically ∼0.5 μM.

[13] In the surface waters of the Sargasso Sea and of many other subtropical and tropical regions, dissolved inorganic nitrogen (DIN, the sum of NO3, NO2, and ammonium) is sufficiently scarce to be considered absent for our purposes. The [NO3] + [NO2] for surface water samples (≤100 m) collected for this study was consistently below the detection limit of our methods (<0.1 μM), except in August 2000, March 2001, and April 2001, when the concentrations of some samples were between 0.1 and 1.0 μM. In shallow thermocline waters (≤300 m), where the [NO3] is typically less than the [DON], dual analyses of TDN and NO3 allow the concentration and isotopic composition of DON to be calculated by mass balance.

[14] Since none of the BATS samples were filtered, the concentration and isotopic analyses of “TDN” also include suspended particulate organic N (PNsusp) and are thus truly TN measurements. Because PNsusp concentrations ([PNsusp]) represent <10% of the total N concentration ([TN]) at BATS [Michaels et al., 1994b; Michaels and Knap, 1996], we often refer to our analyses as being of TDN (or of DON in the surface). Nevertheless, it will be suggested in section 5.4 that variations in [PNsusp] and PNsusp δ15N with depth are apparent in our data, and we are careful to use the term “TN” (rather than “TDN”) when PNsusp is relevant. In the upper 100 m, where [DIN] is extremely low, TN is essentially equivalent to total organic nitrogen (“TON,” the sum of DON + PNsusp). In the subsurface, the concentration and δ15N of TON is estimated by differencing NO3 from TN, as would be done to estimate DON from TDN and NO3 measurements in samples devoid of particles.

3.2.2. TDN δ15N Analysis

[15] After the oxidation of TDN to NO3, the samples and aliquots of autoclaved POR were converted to N2O using the denitrifier method [Sigman et al., 2001; Casciotti et al., 2002]. Because the POR that is added to the DON sample creates an extremely alkaline solution, the pH of the sample must be lowered into a range that is suitable for bacterial conversion by adding 6 N HCl (made from certified ACS-grade HCl) until the sample pH is between 3 and 6. The volume of sample aliquots for bacterial reduction to N2O was adjusted to yield either 10 or 20 nmols N. The bacterial medium was prepared with three times the amount of potassium phosphate buffer described by Sigman et al. [2001] to provide extra buffering of the medium pH, given the high normalities of NaOH and HCl involved in the oxidation protocol.

[16] As with the concentration measurements, mass balance calculations are required to (1) correct for the effect of the POR blank on the sample TDN δ15N and (2) extract the DON δ15N from the measurements of TDN δ15N and NO3 δ15N in a given sample. These calculations require concentration and δ15N analysis of (1) the NO3 in the unoxidized sample, when it is measurable, (2) the POR alone, and (3) the oxidized sample.

3.2.3. DON Oxidation Efficiency

[17] Table 1 lists the degree of completeness to which common DON standards were oxidized to NO3 in both DIW and seawater (“SW”) and/or 0.5 M NaCl solution using persulfate oxidation and compares our results with other published values. Standards include: glycine, urea, 6-amino caproic acid (ACA), EDTA, and 4-aminoantipyrine (Sigma part A1-4328), most of which were selected for comparison because they have been used previously for testing the efficacy of [DON] methods [Bronk et al., 2000, and references therein]. In all cases, recoveries of DON standard as NO3 were equal to or greater than other published values, with the exception of 4-aminoantipyrine, a cyclic compound with three nitrogen atoms, which has been noted previously as a difficult compound to oxidize (Table 1) [Walsh, 1989; Hopkinson et al., 1993; Bronk et al., 2000]. Differences in reported oxidation efficiencies of 4-aminoantipyrine may result from the use of different compounds marketed under the common name “antipyrine” (e.g., compare work of Suzuki et al. [1985] with that of Bronk et al. [2000]).

Table 1. DON Standard Oxidation Yields
CompoundStandard in DIWaStandard in DIW (Others')bStandard in 0.5 M NaCl SolutionStandard in SW (Others')b
  • a

    Standard in DIW is as calculated from a 0.0, 2.5, 5.0, 7.5, and 10.0 μM concentration series of each standard, in DIW and SW or 0.5M NaCl solution where applicable.

  • b

    Standard in DIW is from Bronk et al. [2000, and references therein].

  • c

    Reported values reflect the 95% confidence level.

  • d

    N/R denotes: not reported.

Glycine102 ± 14%c92.9–98.5%97 ± 5%N/Rd
Urea100 ± 7.5%98.7–105.6%109 ± 25%87.5–100%
ACA102 ± 6%N/RN/RN/R
EDTA105 ± 0.5%N/RN/RN/R
Antipyrine42 ± 2.3%45–68.1%N/RN/R

3.2.4. DON δ15N Precision and Accuracy

[18] The ability of the persulfate and denitrifier methods to yield precise and accurate DON δ15N data was tested by measuring the δ15N of the model compounds glycine and ACA at a range of concentrations (0–10 μM) in deionized water and/or 0.5M NaCl solution. The isotopic composition of these model compounds was determined by direct combustion and isotopic analysis of the undiluted compound salts in the laboratory of M. A. Altabet, yielding δ15N values of 9.6‰ and 4.6‰ for the glycine and ACA standards, respectively. In the tests reported here, the measured δ15N follows the expected trend with compound concentration when the reagent blank is taken into account (Figure 1). The agreement between the measured and calculated TDN δ15N indicates that (1) the oxidation of these model compounds is adequately complete and/or nonfractionating (preventing significant preferential loss of either the light or heavy isotope during the oxidation), (2) the blank in this method is due almost entirely to N in the POR, and (3) the reagent blank is constant and characterizable within a set of oxidations. It must be stressed that the POR [N] and δ15N may vary from one “batch” of reagent to another (Figure 1), so that the [N] and δ15N of the reagent blank must be measured with each batch of oxidations.

Figure 1.

Accuracy of DON δ15N method tested using laboratory standards and reagent blank characterization. (a) The δ15N versus N concentration ([N]) added for ACA concentration series (0, 2.5, 5, 7.5, and 10 μM) in DIW, with ACA δ15N = 4.6‰ (dashed line). Curved line represents the calculated TDN δ15N using the known ACA and measured [N] and δ15N of the POR (0.3 μM when diluted by sample, and 3.6‰, respectively). That the measured TDN δ15N (pluses) fall along the expected TDN δ15N trend indicates that (1) the oxidation of these model compounds is complete and/or nonfractionating, (2) the “blank” in this method is largely N in the POR, and (3) the concentration and isotopic value of the blank is constant and accurately characterized. (b) δ15N versus [N] added for filtered seawater amended with a concentration series (0, 2.5, 5, 7.5, 10 μM) of glycine, δ15N = 9.6‰ (dashed line). Curved line represents TDN δ15N calculated as above (blank [N] when diluted by sample = 0.4 μM, blank δ15N = 1.1‰). The measured TDN δ15N (asterisks) fall along the calculated TDN δ15N line. The background N in this experiment includes N in the POR as well as natural DON in the filtered seawater matrix. (c) The δ15N versus quantity of POR added (as a factor of the standard POR addition) for analyses of 5 μM ACA in DIW. The solid and dashed lines represent POR δ15N (−6.7‰) and ACA δ15N (4.6‰), respectively. Crosses are the measured TDN δ15N, and open circles represent the corrected δ15N after correction for the reagent blank. As the quantity of the reagent is increased, the δ15N approaches the δ15N of the reagent. It should be noted that the [N] of the POR in this experiment, 9.1 μM in the undiluted reagent, is a value typically four times greater than any POR used to generate the BATS data shown in Figures 2 and 3. The large isotopic difference between the sample and POR also contributed to the large blank correction in this experiment.

[19] Since the tests shown in Figure 1 were conducted, significant improvements have been made in blank reduction, blank characterization, and the mechanics of the oxidation step, and these were in place for all of the BATS measurements reported below. Previously, the reagent blank [N] was occasionally very high (as in Figure 1c, for which the POR [N] was 9.1 μM), necessitating a large blank correction. Subsequently, the POR [N] has been reduced during the recrystallization process by drying the potassium persulfate crystals in the presence of indicating Drierite in a vacuum dessicator. Since this step was taken, no data have been generated using a POR with a [N] > 4 μM. Currently, the POR [N] ranges from 1.2 to 2.5 μM (effective concentration of 0.2–0.4 μM when diluted by the sample), and the POR δ15N ranges between −14 and 10‰. The oxidation step has been improved by using ∼15 mL test tubes (Fisher part 14-933-1A) which reliably maintain a seal during autoclaving. During the tests run in Figure 1, test tubes of different sizes and brands would occasionally leak or vent during the autoclaving step, causing “fliers” in [DON] and/or DON δ15N.

[20] The reproducibility of the DON δ15N method can be assessed by repeatedly analyzing a given seawater sample. Shown in Table 2 are the results of replicate [DON] and DON δ15N analyses of the four BATS samples that yield the smallest and largest standard deviations for all of the [DON] and DON δ15N measurements reported here. Additionally, the median [DON] and DON δ15N standard deviations for all analyses made on samples collected at one depth are reported in Table 2. Frequently, standard deviations include not only replicate analyses on a single sample but also analyses of duplicate samples collected at the same depth.

Table 2. Replicate BATS DON Analyses
 Depth, m[DON], μMStandard Deviation [DON], μMDON δ15N, ‰Standard Deviation DON δ15N, ‰n
  • a

    Smallest standard deviation for [DON] for entire data set was measured on February 2001 100-m samples.

  • b

    Largest standard deviation for [DON] for entire data set was measured on July 2000 60-m samples.

  • c

    Smallest standard deviation for DON δ15N for entire data set was measured on July 2000 0-m samples.

  • d

    Largest standard deviation for DON δ15N for entire data set was measured on January 2001 100-m samples.

  • e

    Median standard deviations for entire data set are listed for both [DON] and DON δ15N analyses.

Feb. 20011003.80, 3.790.01a3.35, 3.570.162
July 2000604.95, 4.370.41b3.59, 3.630.032
July 200004.28, 4.320.024.30, 4.320.01c2
Jan. 20011004.13, 4.08, 4.200.064.68, 4.08, 3.910.41d3
Mediane  0.2 0.2 

[21] Our [DON] measurements for BATS surface (upper 100 m) samples are within the same range as those previously made at the same site by UV oxidation; our typical range is from 3.8 to 5.0 μM, with total range of 3.4–5.9 μM, whereas Hansell and Carlson [2001] report a concentration range of 4–5.5 μM TON. Although there is no external check for the [DON] or DON δ15N of the samples reported here, a low median standard deviation for replicate [DON] and DON δ15N analyses of a single sample (0.2 μM and 0.2‰, respectively, Table 2) suggests that oxidation is complete and/or nonfractionating and thus yields an accurate assessment of the DON δ15N of these seawater samples.

4. Results

4.1. Nitrate

[22] At depths of ≥800 m, NO3 δ15N is close to the nominal mean ocean value of ∼5‰ [Sigman et al., 2000] and decreases upward from 800 m to ∼2.5‰ at 250 m, while the [NO3] decreases from ∼22 μM at 800 m to ∼2.5 μM at 250 m (Figures 2 and 3). At 250 m, NO3 δ15N appears to change gradually during the sampling period. NO3 δ15N is typically <2.5‰ from May through August of 2000 and >2.5‰ from November 2000 through February 2001 (Figures 234). This variation appears to be associated with a slight increase in [NO3] at the same depth (Figures 234). [NO3] decreases to <0.1 μM typically by or somewhat below 100 m. In many of the shallow sections of the profiles, there is also evidence for an increase in NO3 δ15N from ∼200 m to ∼100 m, presumably due to isotope fractionation during NO3 assimilation by phytoplankton. Generally, there is more variability in the NO3 δ15N near the top of the water column, which is to be expected from the large relative variations in [NO3] at this depth and from the various degrees to which the isotope effect of NO3 assimilation will be expressed when NO3 supply and NO3 consumption are occurring in a complex sequence.

Figure 2.

At BATS, (a) [NO3] (pluses) and [TON] (solid triangles) and (b) NO3 δ15N (pluses) and TON δ15N (solid triangles). NO3 data are from March 2000 through May 2001. TON data are from June 2000 through May 2001. Altabet's [1988] PNsusp δ15N (−0.2‰) (solid circle) and PNsink δ15N (3.7‰) (solid circle) are shown for reference. Replicate analyses were performed on individual samples. In addition, at roughly half the sampling depths, replicate samples were collected. Plotted values are averages of all analyses at a given depth.

Figure 3.

(a) [TON], (b) TON δ15N, (c) [NO3], and (d) NO3 δ15N at BATS from June 2000 through May 2001.

Figure 4.

(top) The δ15N and (bottom) [N] versus month (June 2000 through May 2001). Top panel shows 0 m TON δ15N (solid triangles), 40 m TON δ15N (solid diamonds), 100 m TON δ15N (solid squares), and 250 m NO3 δ15N (solid circles). Bottom panel shows 0 m [TON] (open triangles), 40 m [TON] (open diamonds), 100 m [TON] (open squares), and 250 m [NO3] (open circles).

4.2. TN Concentration and δ15N

[23] As also observed by Hansell and Carlson [2001], the range of measured [TON] in the upper 100 m reported here is quite limited, with a mean concentration of 4.2 μM and standard deviation for the entire data set in the upper 100 m of 0.5 μM, with no distinct seasonal or depth related trends (Table 3, Figures 234). The TON δ15N in the upper 100 m ranges from 3.5 to 4.5‰, and, as with [TON], no seasonal trends in TON δ15N are recognized (Figures 234). The only sample deviating from these values is the 0-m January 2001 sample, with a [TON] of ∼6 μM and TON δ15N of ∼2‰. This sample yielded similar values when it was analyzed four times, twice each from two sample bottles collected from the same hydrocast bottle, indicating that the values are real, although perhaps the result of hydrocast bottle contamination.

Table 3. Average Concentration and δ15N for NO3, TN, and TON at BATS
N PoolDepth, mAverage [N], μMStandard Deviation [N],a μMAverage δ15N, ‰Standard Deviation δ15N, ‰
  • a

    Standard deviations apply to entire suite of measurements made on samples collected between June 2000 and May 2001.

  • b

    TON is calculated by differencing NO3 from TN.

TN0–1004.180.513.920.48
TN2506.320.523.570.24
NO32502.390.572.650.32
TONb2503.930.384.130.39

[24] The annually averaged [TON] decreases by 0.25 μM from 100 to 250 m, with an average [TON] concentration of 3.93 μM at 250 m (0.38 μM, 1 SD) (Table 3). The Kruskal-Wallis test for nonparametric data [Desu and Raghavarao, 2003] indicates that the difference in [TON] between the upper 100 m and 250 m is significant beyond the 99% level. Between 100 and 250 m, the TON δ15N increases slightly but significantly from a mean of 3.9‰ between 0 and 100 m to a mean of 4.1‰ at 250 m (0.39‰, 1 SD). The Kruskal-Wallis test for nonparametric data indicates that the 0–100 m and 250 m TON δ15N data are significantly different from each other above the 93% confidence level.

[25] While there are no other bulk marine DON/TON δ15N values in the literature that can be directly compared with those reported here, we can compare our results with those of related analyses. Benner et al. [1997] report a δ15N of 6.6‰ for ultrafiltered DON from surface waters at BATS, which is higher than our average surface water bulk DON δ15N of 3.9‰. The ultrafiltered material generally represents 30% of total DON in open ocean waters, implying that the remaining smaller size fractions of DON have a δ15N of roughly 2.7‰. Feuerstein et al. [1997] report bulk DON δ15N of −1.2 to +5.8‰ in Great Lakes water, interestingly different from our marine data. In a meeting abstract, Abell et al. [1999] report bulk DON δ15N of 1–2‰ for surface water samples from the Pacific Ocean, lower than the results reported here.

5. Discussion

5.1. Stability of DON δ15N at BATS

[26] The lack of seasonal change in [TON] in the upper 100 m at BATS, observed previously by Hansell and Carlson [2001], implies that the bulk surface DON pool is not dynamic on the timescale of seasons. However, this measure, by itself, leaves open the possibility that there are large but balanced inputs to and losses from the surface DON pool. The stability of the DON δ15N in the upper 100 m at BATS observed in this study argues against this possibility, in that large inputs and outputs would need to be isotopically balanced. For instance, if 1 μM N were simultaneously added to and removed from a 4 μM DON pool, and the δ15N of the DON removed was 3‰ lower than the DON added, then the δ15N of the DON pool would increase by 0.75‰. Although we cannot rule out the possibility of similar isotope compositions for DON that is produced and consumed, no processes or feedbacks have yet been described that would maintain this isotopic similarity for DON inputs and outputs. Thus we infer from the isotopic stability that the rates of input and output are small.

[27] The refractory nature of DON revealed by its isotopic stability is perhaps expected given the previous [DON] measurements [Hansell and Carlson, 2001], as well as recent bulk dissolved organic matter (DOM) Δ14C age measurements [Loh et al., 2004]. If DON is present and stable at such relatively high concentrations at BATS, it is hard to imagine that much of it could be labile, unless the system was limited by something else (e.g., P). Of course, while the stability of both its concentration and δ15N makes a strong case for the generally recalcitrant nature of the bulk DON at BATS, it does not rule out the possibility of large inputs of a class of DON that is immediately and completely consumed. Ammonium is a good example of such dynamic behavior: It is an important N source for phytoplankton at BATS, but its concentration is so low that it is not important in our TN analyses [Lipschultz, 2001].

[28] The static behavior of DON at BATS represents a constraint on the potential N sources for the biological removal of DIC during the summer at BATS. A 2.5 μM N decrease is required to fuel the observed DIC removal over the upper 150 m from April through December [Michaels et al., 1994a], and calculations for the DIC removal restricted to the upper 25 m require as much as 6 μM N to remove the commensurate DIC, taking into account additional DIC sources such as advection [Gruber et al., 1998]. Since there is no appreciable DIN in the surface waters or identified fluxes of DIN that could be immediately consumed during this time period [Lipschultz, 2001], we consider other sources of N. While it is clear that some phytoplankton can use certain kinds of DON as a primary N source [Palenik and Morel, 1990; Capone et al., 1994; Mullholland et al., 2002; Berman and Bronk, 2003], no changes in [DON] of this magnitude are observed at BATS. The lack of change in [DON] through the summer season would thus require that N must be added to the DON reservoir to offset this loss if DON were fueling the DIC drawdown. The isotope data provide the additional constraint that the N added to and removed from the DON pool would need to have very similar δ15N. Revisiting the calculation above, if 2.5 μM DON is consumed from the bulk 4 μM DON pool and the DON consumed is 3‰ higher than the 2.5 μM DON that must be simultaneously added back to the DON pool, then the bulk DON δ15N would decrease from 4‰ to 2‰, which is clearly not observed.

[29] N2 fixation is considered a plausible explanation for the summertime DIC drawdown at BATS [Bates et al., 1996; Gruber et al., 1998]. Again, it would seem unlikely for a hypothetical input of N to the DON pool via N2 fixation, with its presumably low-15N N, to have the resulting change in δ15N of the bulk DON pool be closely balanced by DON consumption. While fractionation during degradation/consumption of DON is certainly possible [Macko et al., 1987; Waser et al., 1998], it would be highly fortuitous if it led to loss of N with the same δ15N as the input of DON from N2 fixers. Thus, if N2 fixation is responsible for the summertime drawdown of DIC, it seems unlikely that it is supplying the needed N through the bulk DON pool. Given earlier evidence for DON production by N2 fixers [Karl et al., 1992; Capone et al., 1994; Glibert and Bronk, 1994], we are left with three possible conclusions: (1) The fraction of DON produced by N2 fixers at BATS is completely consumed as rapidly as it is produced, so that it never becomes a significant fraction of the resident TON pool measurable by monthly sampling, although it would appear in the N exported from the surface (see below); (2) the link between N2 fixation and DON production observed elsewhere [Karl et al., 1992; Capone et al., 1994; Glibert and Bronk, 1994] does not apply at BATS; or (3) N2 fixation rates are not sufficient to fuel the DIC drawdown at BATS, as suggested by in situ measurements [Orcutt et al., 2001] (Table 4).

Table 4. Estimated N Fluxes to Surface Watersa
N SourceFlux, mol N m−2 yr−1N2 Fix/Sum New NbReference
  • a

    Table is after Lipschultz et al. [2002].

  • b

    Estimates of N2 fixation are divided by the sum of the estimate of the flux of NO3 from below from Jenkins and Doney [2003] and the estimate of N2 fixation being considered.

  • c

    Amount of N required to close the DIC imbalance takes the DIC drawdown of 40 μmol/kg that occurs over the upper 25 from 9 May to 16 October [Gruber et al., 1998]. This N flux is calculated by converting the units to mol C m−2, and dividing by the Redfield C/N ratio of 6.6. This value assumes DIC is only removed during the 160-day period considered, and thus does not multiply by (365/160), so the flux is not “per year,” but “per period.”

  • d

    Value denotes amount of N required to account for the Ono et al. [2001] estimate of shallow remineralization (between 100 and 250 m) of 1.53 mol C m−2 period−1 (240 days) at BATS, calculated by dividing the C flux term by the Redfield C/N ratio of 6.6. This value only refers to the time period considered (April 16 through December 12), and is thus not an annual rate.

  • e

    Value denotes amount of N required to fuel a DIC drawdown of 2.16 mol C m−2 from April to December [Michaels et al., 1994a]. The N flux is calculated by dividing the DIC value by 6.6 and represents a flux “per period” (275 days).

Wet and dry deposition0.01 Prospero et al. [1996]
NO3 from below0.84 Jenkins and Doney [2003]
N2 fixation0.0152%Orcutt et al. [2001]
N2 fixation0.0455%Hansell et al. [2004]
N2 fixation0.0728%Gruber and Sarmiento [1997]
N flux required for DIC drawdown0.16c16%Gruber et al. [1998]
N flux required for DIC drawdown0.25d22%Ono et al. [2001]
N flux required for DIC drawdown0.33e28%Michaels et al. [1994a]

5.2. Absolute Values of [DON] and DON δ15N

[30] As is observed elsewhere in the oligotrophic open ocean [Abell et al., 2000], DON constitutes the largest pool of fixed N at BATS. Because of the high concentration of this N pool in this otherwise N-scarce region, determining the δ15N of the bulk DON pool has been a priority for researchers [Druffel and Williams, 1992; Capone, 2001]. The data in Figure 2 show that at BATS, TON δ15N is higher than the δ15N of all other N pools and fluxes at BATS. Altabet's [1988] measurements of the average PNsusp δ15N (−0.2‰) and of the average PNsink δ15N (3.7‰) are lower than the average TON δ15N in the upper 100 m and at 250 m (3.9‰ and 4.1‰, respectively, Table 3). Moreover, the TON δ15N is higher than the δ15N of NO3 in the shallow thermocline (∼2–3‰ at 250 m).

[31] More work will be required before we understand the processes that set the absolute values of [DON] and DON δ15N, both in the surface and subsurface. One useful albeit overly simplistic hypothesis is that there is little fractionation in the production or consumption of DON, so that newly produced DON is essentially isotopically identical to the new N supplied to the surface ocean. Even in this simplest case, the interpretation of the bulk surface DON δ15N is complicated by the observation that at BATS, there is only a small vertical gradient in [DON] between the upper 100 m and 250 m, which implies that a large fraction of the surface DON is simply DON that is mixed up from below. If most of the surface DON in the open ocean comes from the upward mixing of subsurface DON, then the surface DON δ15N will have a long response time to surface processes and/or changes in the N cycle.

[32] Since very little work has been done on DON δ15N [Feuerstein et al., 1997; Benner et al., 1997; Abell et al., 1999] or on the isotope dynamics of DON transformations [Macko et al., 1987; Hoch et al., 1996; Waser et al., 1998], we cannot at this point rule out the possibility of multiple isotope fractionations associated with DON production and consumption; indeed, we expect them. Given this situation, it remains unclear what the absolute δ15N of surface or subsurface DON signifies when considered in isolation, as has been pointed out previously for other forms of organic N [Altabet, 1988; Karl et al., 2002]. Consequently, we focus on what can be learned by including DON in the N isotope budget at BATS, which does not require such an understanding of DON isotope systematics.

5.3. N Isotope Budget at BATS

[33] Building on the conceptual model of Altabet [1988], we develop an isotope budget for the upper 100 m at BATS that includes only N fluxes from the underlying thermocline. This represents a test of the premise that N2 fixation is not a prominent fraction of the new N supplied to the euphotic zone at BATS (Figure 5, Table 4). The 0–100 m depth interval is examined because it is the typical euphotic zone depth throughout the year at BATS, and is thus the depth range over which the DON pool should be the most dynamic. A typical maximum winter mixed layer depth is 250 m [Michaels et al., 1994b; Michaels and Knap, 1996; Hansell and Carlson, 2001], and is thus a reasonable choice to represent the TN flux up from below. The vertical (one-dimensional) N mass and isotope budget includes the following three fluxes: TN mass flux up from 250 m, TN mass flux down from the upper 100 m to 250 m, and the PNsink flux, represented by the terms ND, NS, and P, respectively, in equation (1), where Q is the flux term representing the annually averaged vertical exchange of water,

equation image

Note that ND and NS represent TN concentrations and thus include NO3, DON, and PNsusp. A similar equation can be applied to the 15N budget at BATS, approximated here using δ15N rather than 15N/14N,

equation image

where δ15ND, δ15NS, and δ15NP represent the δ15N of TN at 250 m, the δ15N of TN in the upper 100 m, and the δ15N of PNsink, respectively. Substituting equation (1) into equation (2) yields the following expression for PNsink δ15N:

equation image
Figure 5.

N fluxes and transformations at BATS. The water column is divided into the upper 100 m and 100–250 m. Of the 6.3 μM TN pool with δ15N = 3.6‰ at 250 m, 3.9 μM is a recalcitrant DON pool with δ15N = 4.1‰ and cycles essentially unchanged between upper 100 m and 250 m. The remaining 2.4 μM of the 250-m TN pool is NO315N = 2.8‰), which mixes up to the surface waters, where it is completely consumed, and then partitioned into two fluxes. The larger of the two N fluxes out of the upper 100 m is PNsink, which has a calculated δ15N of 3.0‰, based on the values reported here. Altabet's [1988] measured δ15N for PNsink (3.7‰) is also shown for reference. The other, much smaller N flux out of the upper 100 m is from a labile TON pool (0.25 μM, 1.3‰) that is respired back to NO3 at 250 m after being mixed out of the upper 100 m; this flux is most likely composed of PNsusp (see text).

[34] From June 2000 through May 2001, the average [TN] between 0 and 100 m is 4.2 μM with a standard deviation of 0.5 μM, and the average TN δ15N is 3.9‰ with a standard deviation of 0.5‰ (Table 3). The average [TN] from June 2000 through May 2001 at 250 m is 6.3 μM with a standard deviation of 0.5 μM, and the average δ15N of TN is 3.6‰, with a standard deviation of 0.2‰ (Table 3). The standard deviations for these annual averages include not only errors associated with the measurements, but also any systematic (e.g., seasonal) changes that may occur. From equation (3), we calculate an expected PNsink δ15N of 3.0‰. This is somewhat lower than, although perhaps not significantly different from, the annually averaged δ15N of 3.7‰ measured for sediment trap materials collected at 100 m at BATS in 1986 [Altabet, 1988] (Figure 2). From this calculation, we see that our isotope budget requires no additional input of low-δ15N new N to the surface water at BATS beyond the terms included in this budget. If N2 fixation were an important source of new N at BATS, our calculated PNsink δ15N should be greater than the measured value to reflect a missing N source with a low δ15N not included in our fixed N balance. Thus our work supports the conclusions of Altabet's [1988] budget that the NO3 flux up from below is the dominant source of new N at BATS and that N2 fixation is too small an input to be apparent in the isotope budget. That our budget yields an estimated PNsink δ15N that is less than the measured value of 3.7‰ results from the δ15N of the NO3 that we measure at 250 m (a component of the 250 m TN term), which is 1‰ lower than that reported by Altabet [1988].

[35] The result of our budget is consistent with in situ measurements of N2 fixation at BATS [Orcutt et al., 2001], which, when compared with flux estimates of NO3 to the mixed layer, are on the order of 2% of new N on an annual basis (Table 4). The higher geochemical estimates of N2 fixation [i.e., Gruber and Sarmiento, 1997] represent ∼8% of the new N flux to BATS (Table 4), which might also be difficult to resolve within the surface ocean N isotope budget described here. For example, assuming a δ15N of −1‰ for newly fixed N [Minigawa and Wada, 1986; Carpenter et al., 1997], an 8% contribution from N2 fixation to the annual N budget would lower the PNsink δ15N by ≤0.3‰ relative to the PNsink δ15N that one estimates on the basis of N supply solely from NO3 from below. Since the fluxes of NO3 from below and N2 fixation should be temporally separated at BATS [Michaels et al., 1994a; Bates et al., 1996; Lipschultz et al., 2002], the seasonal invariance of [DON] and DON δ15N may be more sensitive indicators of the limited importance of N2 fixation than the annually integrated N isotope budget described in this section. However, this only applies if N2 fixation measurably augments the DON pool (see section 5.1).

[36] Returning to the summertime DIC drawdown at BATS, we consider what corresponding fluxes of N, in the form of N2 fixation, are required to account for the DIC removal. Gruber et al. [1998] estimate that 40 μmol C kg−1 are removed between April and October in the upper 25 m at BATS. On the basis of the Redfield ratio N equivalent, this magnitude of C removal requires a N2 fixation flux of 0.16 mol N m−2 between April and October. This is roughly 16% of the new N on an annual basis at BATS, is an order of magnitude greater than in situ N2 fixation flux measurements made at BATS [Orcutt et al., 2001], and is more than twice the geochemical N2 fixation estimate of Gruber and Sarmiento [1997] (Table 4). Nevertheless, the amount of N2 fixation required to resolve the Gruber et al. [1998] DIC imbalance may be a minimum estimate, for two reasons. First, N2 fixers may have a lower C:N than Redfield ratio values [McCarthy and Carpenter, 1979; Carpenter et al., 2004], thus requiring a larger flux of N than that calculated in Table 4 to remove a given amount of DIC. Second, Gruber et al. [1998] consider DIC removal over only the upper 25 m, even though the mixed layer is often deeper than 25 m throughout that time period, and DIC may be removed below this depth during portions of the time period considered [Michaels et al., 1996]. Ono et al. [2001] estimate a C flux of 1.53 mol C m−2 period−1 (240 days) between 100 and 250 m from April to December at BATS, which, when applying Redfield C/N ratios, corresponds to a N flux of 0.25 mol N m−2 period−1 (Table 4). This estimate, which is greater than that of Gruber et al. [1998], may also represent a lower bound, as it does not detect any exported N that sinks below 250 m.

[37] An upper estimate of the amount N2 fixation contributes to new production at BATS is provided by the DIC drawdown estimate of Michaels et al. [1994a] of 2.16 mol C m−2 period−1, which is calculated over the upper 150 m from April through December. This amount of DIC removal requires a N2 fixation flux of ∼0.33 mol N m−2 period−1, equal to almost 40% of the NO3 flux up from below (Table 4), which was clearly not detected in the isotope budget.

[38] In summary, given that the various DIC drawdown estimates require a N input equivalent to ∼20% of the annual new N supply at BATS, a N2 fixation flux large enough to reconcile the summertime DIC drawdown appears to be inconsistent with the N isotope budget results presented here, unless extreme deviations from Redfield C:N ratios for export production are invoked [Orcutt et al., 2001; Anderson and Pondaven, 2003]. This sense of deviation is not apparent in sinking particles collected at BATS [Conte et al., 2001] or in the C:N ratios of N2 fixers [McCarthy and Carpenter, 1979; Carpenter et al., 2004]. The differences in the mean concentrations and seasonal cycles of DOC and DON imply a non-Redfieldian stoichiometry for summertime DOM at BATS [Hansell and Carlson, 2001], and previous work has shown that DOC is effectively exported from the euphotic zone during winter mixing [Carlson et al., 1994]. However, this C loss term is not adequately large to explain the C drawdown at BATS [Michaels et al., 1994a].

5.4. Organic N and NO3 δ15N at BATS

[39] The low NO3 δ15N in the thermocline of the Sargasso Sea and other subtropical regions should be relevant for understanding the oligotrophic N cycle. As mentioned previously, the low δ15N of NO3 in the thermocline, both at BATS and elsewhere, has been interpreted as evidence of N2 fixation [Liu et al., 1996; Brandes et al., 1998; Karl et al., 2002]. Working to increase the low NO3 δ15N in the thermocline are both diapycnal and isopycnal mixing, which introduce NO3 into the thermocline with a δ15N close to ∼5‰, typical of NO3 in the deep ocean [Sigman et al., 2000]. Thus we expect a significant source of low-δ15N N to the BATS thermocline, such that a combination of this N source and the input of deep ocean NO3 yield a NO3 δ15N of 2–3‰ at ∼250 m depth. However, the δ15N of shallow PNsink measured at BATS is 3.7‰ [Altabet, 1988], which is somewhat greater than (not less than) the δ15N of NO3 in the shallow subsurface at BATS. Moreover, data from sediment traps at multiple depths show no evidence for preferential remineralization of 14N as the sinking flux passes through the water column [Saino and Hattori, 1987; Altabet, 1988; Altabet et al., 1991]. Thus, sinking N and its remineralization do not appear to represent the vehicle by which the low-δ15N N is accumulated in the shallow thermocline.

[40] Given the large isotope effect of nitrification (∼15–35‰ [Mariotti et al., 1981; Casciotti et al., 2003]), it has been postulated that nitrification can significantly lower the δ15N of NO3 in subsurface ocean waters [Sutka et al., 2004]. However, nitrification can only have this effect if the ammonium (NH4+) released into the thermocline does not eventually end up in the thermocline NO3 pool. NH4+ concentrations are low throughout the year at BATS (∼30 nM), indicating that direct NH4+ export by circulation is not significant [Lipschultz, 2001]. Assimilation of NH4+ is perhaps the most feasible means for nitrification to affect the NO3 δ15N, by providing a pathway by which 15N-rich NH4+ could be diverted away from nitrification. However, the organic N resulting from this assimilation of NH4+ must be exported from the shallow thermocline; otherwise, this putative high-15N organic N would eventually end up in the thermocline NO3 pool. As mentioned above, there is no evidence for the creation or export of high δ15N particles from the thermocline. While we cannot rule out the loss of 15N-rich DON from the thermocline to the deep ocean, this seems unlikely. Moreover, NH4+ assimilation by microbes may have an isotope effect comparable to that of nitrification, as is the case for assimilation by eukaryotic phytoplankton [Waser et al., 1998], in which case, NH4+ assimilation would not even represent a possible conduit for the escape of high-δ15N N from conversion to NO3. Thus it seems safe to conclude that nitrification is not an important driver of the low NO3 δ15N observed at BATS.

[41] Here we consider the implications of our TN measurements for the missing source of the low-δ15N NO3 in the thermocline at BATS, focusing on our observation of differences between the TON (calculated by differencing the TN and NO3 measurements) in the upper 100 m (“surface”) and at 250 m. The difference in [TON] between the surface and at 250 m describes the amount of TON generated in the upper 100 m (subsequently referred to as “semi-labile”) that is eventually mixed downward and respired back to NO3 at 250 m,

equation image

where the average [TON] in the upper 100 m, the average [TON] at 250 m, and the [NO3] at 250 m resulting from remineralization of the downward-mixed surface TON are represented by the terms [TONS], [TOND], and [NO3 remin], respectively. The same expression can be written for 15N, again approximating 15N/14N as δ15N,

equation image

where δ15N-TONS is the δ15N of TON in upper 100 m, δ15N-TOND is the δ15N of TON at 250 m, and δ15N-NO3 remin is the δ15N of the component of surface TON that is remineralized to NO3 at ∼250 m. Substitution of equation (4) into equation (5) yields an expression for the δ15N of the remineralized semi-labile surface TON,

equation image

[42] [TONS] and [TOND] are 4.18 μM and 3.93 μM, respectively, which indicates a net addition of 0.25 μM TON in the upper 100 m (Table 3, Figure 5). The δ15N of TON in the upper 100 m and at 250 m are 3.9‰ and 4.1‰, respectively (Table 3, Figure 5). This indicates that the 0.25 μM TON added in the upper 100 m has an isotopic composition of ∼1.3‰ (Figure 5). Upon the annual vertical mixing of this upper 100 m TON pool down into the subsurface (∼250 m), the TON added in the upper 100 m is respired back to NO3 (equation (4)), adding 0.25 μM NO3 with a δ15N of 1.3‰ to the thermocline. Thus the annual net production of low-δ15N TON in the upper 100 m and subsequent oxidation of this semi-labile TON in the subsurface appears to provide a vehicle for the transport of low-δ15N N into the subsurface at BATS.

[43] The difference in [TON] and TON δ15N between the upper 100 m and at 250 m, while subtle, is consistent with the previously observed depth variations in PNsusp. Between 0 and 100 m, [PNsusp] is ∼0.3–0.4 μM, and is ∼0.1–0.2 μM at 250 m [Michaels et al., 1994b; Michaels and Knap, 1996;Michaels et al., 1996]. Additionally, our calculated δ15N of the remineralized NO3, ∼1.3‰, is generally consistent with the δ15N of the N that appears to be lost from PNsusp as it is mixed down from the upper 100 m and respired in the subsurface (∼−1‰, calculated from data of Altabet [1988, 1989]). Thus the low-δ15N TON being remineralized to NO3 in the subsurface is likely the PNsusp that is generated in the upper 100 m.

[44] Given the subtle month-to-month variations of the data in hand, we tentatively note that the surface-to-subsurface gradients in both [TON] and TON δ15N are weakest in the late winter (January–April 2001; Figure 3). This is roughly consistent with the interpretation that winter mixing supplies new semi-labile TON (i.e., PNsusp) to the subsurface, with subsurface remineralization causing a subtle decrease in [TON] and increase in TON δ15N at 250 m during the rest of the year. Temporal variations in subsurface NO3 δ15N are roughly consistent with this interpretation, as the highest NO3 δ15N appears to occur from January to April, as would be expected if vertical mixing homogenizes the upper water column. The June-to-August minimum in NO3 δ15N observed here is slightly early relative to the period of peak N2 fixation, late summer and early fall [Orcutt et al., 2001], which raises some doubt about the role of N2 fixation-driven N export in generating the low δ15N of thermocline NO3. However, we are discouraged from close interpretation of the temporal changes in NO3 δ15N because there is a strong correlation in depth between NO3 δ15N and [NO3] (Figures 2 and 3), and we observe a shoaling of high-[NO3], high-δ15N water at 400–600 m depth during the winter and spring of 2001 (Figure 3), which may erode the low NO3 δ15N from below.

[45] The degree to which the mixing-driven flux of semi-labile TON from the surface to the subsurface can influence the δ15N of the NO3 in the thermocline depends on the rate at which the remineralized NO3 accumulates in the subsurface relative to the timescale on which NO3 communicates with the higher latitude surface waters and with the deep sea. By combining the N flux balance (equation (1)) with literature data for the sinking flux [Michaels et al., 1994a] and with our data on the vertical gradient in [TN], we estimate a vertical mixing rate between 250 m and the upper 100 m (Q in equation (1)). This value is combined with the vertical gradient in [TON] to estimate a rate of oxidation of PNsusp to NO3 in the shallow thermocline of ∼0.1 μM yr−1 (calculation not shown). This rate is low if the low-15N NO3 of the thermocline is generated entirely since the water last left the surface (over a period of ∼1–8 years [Jenkins, 1980]). However, the water in the Sargasso Sea thermocline may be recycled multiple times through the North Atlantic thermocline circulation [Jenkins and Doney, 2003], so the low-δ15N NO3 may have a much longer time to diverge from the high δ15N of deeper water NO3. Thus the net production of low-15N TON in the upper 100 m and subsequent oxidation of this semi-labile TON in the subsurface may accumulate at a rate capable of explaining the low-δ15N of the NO3 in the shallow thermocline at BATS. This question could be addressed more directly with a water column transect that sampled thermocline water as it ages away from its site of subduction, analogous to the use of N* by Gruber and Sarmiento [1997].

[46] The recognition of the semi-labile PNsusp flux as the form in which low δ15N-N is transported from the surface into the thermocline does not address the ultimate question of what the low-δ15N N source is to the surface. It has been posited that N2 fixation is this source. As described above, the flux of N2 fixation required to fit observations of elevated N* in the Sargasso Sea thermocline is sufficiently small that it is not clearly in conflict with the N isotope budget for the euphotic zone at BATS that we present here, which does not require any input from N2 fixation. While the vertical N mass and isotope budget we have constructed for the shallow thermocline at BATS is in balance for the temporal and spatial scales considered above, a N isotope imbalance remains when considering the fluxes of N between the thermocline and the deep ocean. At 1000 m, the δ15N of the NO3 being incorporated into the thermocline from the deep ocean (∼5‰, Figures 2 and 3) is significantly higher than the δ15N of the sinking N (∼3‰ [Altabet et al., 1991]), such that an input of low-δ15N N to the upper ocean is required.

[47] While N2 fixation occurring elsewhere in the basin is the most straightforward explanation for this perceived imbalance, there are other sources of low-δ15N N to the subtropical Atlantic surface waters that may contribute to the low δ15N of thermocline NO3. Specifically, while precipitation fluxes of fixed N are small, they are of the same order of magnitude as the in situ N2 fixation rates measured at BATS (Table 4). More importantly for the isotopic budget, the δ15N of NO3 in Bermuda rain varies seasonally (range of −14 to 2‰) but has a low annual mean, flux-weighted δ15N (∼−5‰ [Hastings et al., 2003]). The δ15N of TN in the Bermuda rain appears to have similar characteristics, with a low annual average, flux-weighted δ15N (∼0‰ (A. N. Knapp and M. G. Hastings, manuscript in preparation, 2005)). Thus the δ15N of these precipitation fluxes clearly mimics the δ15N signal of N2 fixation and may contribute significantly to the low δ15N of NO3 observed in the thermocline at BATS.

[48] Above, we indicate that the concentration and isotopic imbalances in the upward and downward annual fluxes of TON in the upper 250 m at BATS could be explained by the PNsusp term. According to this view, which borrows heavily from Altabet [1988], NO3 supplied to the surface produces two downward fluxes of N. The first is a small PNsusp flux transported by annual mixing; the PNsusp at the surface has a low δ15N because of the uptake by phytoplankton of the low-δ15N NH4+ that is excreted by plankton and cycles within the euphotic zone [Checkley and Miller, 1986]. The second, much larger flux is as PNsink, which has a higher δ15N [Checkley and Miller, 1986]. Both of these components are eventually reoxidized to NO3 in the thermocline (Figure 5). If this view of the PNsusp component of the TON flux is correct, it suggests that essentially all of the DON in the surface originates in the subsurface, since the surface DON would be indistinguishable in concentration and δ15N from the DON at 250 m (Figure 5). These results suggest that at BATS, any labile DON that is produced is completely and rapidly consumed and incorporated into PN. However, the lack of any variation in [DON] or DON δ15N in this oligotrophic region begs the question, under what conditions will the resident surface ocean DON pool show dynamic behavior?

[49] Here we have only considered a one-dimensional N mass and isotope balance for the upper 100 m at BATS. This balance does not address a number of potentially important processes that may affect our measurements, such as lateral advection, isopycnal transport, and/or communication between the deeper ocean and the upper 250 m. As described by Hansell et al. [2004], much of the N* signal observed at BATS may be generated to the south and east of BATS by the remineralization of N2 fixer biomass and transported along isopycnals to the Sargasso Sea. This process would also be expected to lower the δ15N of thermocline NO3 at BATS. Additionally, Jenkins and Doney [2003] describe a “nutrient spiral” in the Sargasso Sea in which low-δ15N N would be entrained in the upper thermocline waters and cycled on timescales longer than those examined here. Consequently, it is likely impossible to understand the origin of high N* and low NO3 δ15N in the thermocline at BATS solely by studies at this one location. The same concerns apply to DON in the surface; given its apparently low degree of reactivity at BATS, its properties at any given site may well be influenced by transport within the gyre circulation. Transect studies should help to clarify the interaction of transport with N biogeochemistry in the subtropical and tropical Atlantic.

6. Conclusions

[50] We report a robust method for TDN δ15N analysis that can be used for DON δ15N measurement in NO3-free and low-NO3 waters. This method yields [DON] and DON δ15N measurements that converge upon the concentration and δ15N of lab standards, as well as previously published [DON] measurements made at BATS by UV oxidation. Blanks are small and well characterized, allowing for accurate measurement of DON δ15N down to ∼1–2 μM DON. With its relatively broad applicability, this TDN δ15N method can be used to help constrain nitrogen budgets, both in freshwater and marine systems. Future work to improve this method might focus on quantitative removal of NO3 prior to the DON oxidation step, so that DON can be measured without the differencing of TDN and NO3; any such removal step must be tested for contamination and for alteration of the sample DON.

[51] At BATS, the new method was used to measure TON δ15N in the upper 250 m over a year, and revealed a relatively static surface TON pool, with shallow TON δ15N of ∼4‰ throughout the year. The TN concentration and isotope data require a δ15N for the sinking flux that is similar to previous measurements of sediment trap materials [Altabet, 1988], suggesting that N2 fixation is too small a fraction of the annual new N supply to be apparent in the N isotope budget for the upper 100 m at BATS. While BATS is often taken as a typical example of an oligotrophic subtropical environment, its relatively high latitude leads to conditions that promote significant exchange between the euphotic zone and the shallow thermocline, such that the upward supply of NO3 from below is much higher at BATS than it is farther south. As a result, even the relatively high N2 fixation rate suggested by recent geochemical studies [Gruber and Sarmiento, 1997] would represent less than 10% of the new N supply to the BATS euphotic zone over the course of a year. Thus it is perhaps expected that N2 fixation is not a dominant term in the N isotope budget at BATS, and it is not clear that the isotope budget is in direct conflict with geochemical estimates of N2 fixation. Looking forward, we are motivated to study lower latitude, more stratified regions, where a given rate of N2 fixation should represent a larger fraction of the annual new N supply [Lipschultz et al., 2002].

[52] While our results are not necessarily in conflict with N*-based estimates of N2 fixation, they do appear to pose a problem for the hypothesis that N2 fixation fuels the summertime DIC drawdown at BATS. As discussed above, the rate of N2 fixation required to account for the apparent DIC drawdown is quite large and probably should have been apparent in our N isotope budget. Moreover, the lack of variation in both [TON] and TON δ15N throughout the year indicates that bulk DON does not accumulate at BATS due to N2 fixation, nor is it the major source of N for phytoplankton during the period of DIC drawdown. If isotope approaches for estimating N2 fixation are to be pursued further at BATS, it is recommended that the δ15N of PNsink be revisited; this was last measured by Altabet [1988] using sinking particle collection techniques that have subsequently been improved. Moreover, vertical migration at BATS may play an important role in the N isotope budget and requires attention [Villareal et al., 1999; Steinberg et al., 2002]. Finally, N fluxes other than N2 fixation and NO3 from below, such as precipitation [Hastings et al., 2003], should also be considered in the context of N mass and isotope budgets at BATS as well as in other oligotrophic regions.

[53] We report evidence for the transfer of low-δ15N N from the euphotic zone to the thermocline, via the downward mixing of a small, low-δ15N fraction of TON that is remineralized in the shallow thermocline (∼250 m). The depth gradients in TON concentration and δ15N are consistent with those of PNsusp [Altabet, 1988, 1989; Michaels et al., 1994b; Michaels and Knap, 1996; Michaels et al., 1996], suggesting that PNsusp is the low-15N N being mixed down and then remineralized. If so, this suggests that DON is invariant throughout the year (in concentration and δ15N) over the upper 250 m at BATS, and that this pool is recalcitrant on the timescale of a few years.

Acknowledgments

[54] We thank Deborah Bronk and Bess Ward for methodological advice, and Michael Bender, Matt Reuer, and Jorge Sarmiento for discussions. The BATS staff kindly collected all samples. This work was funded by a DOD ASEE/NDSEG graduate fellowship to A. N. K., by U.S. NSF Biocomplexity grants OCE-9981479 (to D. M. S., through the MANTRA project) and DEB-0083566 (to Simon Levin), and by British Petroleum and Ford Motor Company through the Carbon Mitigation Initiative at Princeton University. This is BBSR contribution 1653. Figure 3 was generated by Peter DiFiore with the application “Ocean Data View,” provided by R. Schlitzer of AWI, Bremerhaven, Germany.

Ancillary