Assessment of excess nitrate development in the subtropical North Atlantic

Transcription

1 Marine Chemistry 106 (2007) Assessment of excess nitrate development in the subtropical North Atlantic D.A. Hansell a,, D.B. Olson a, F. Dentener b, L.M. Zamora a a Rosenstiel School of Marine and Atmospheric Science, University of Miami, Miami, FL, USA b European Commission, Institute for Environment and Sustainability, Joint Research Centre, Ispra, Italy Received 29 March 2007; received in revised form 7 June 2007; accepted 12 June 2007 Available online 20 June 2007 Abstract Geochemical estimates of N 2 fixation in the North Atlantic often serve as a foundation for estimating global marine diazotrophy. Yet despite being well-studied, estimations of nitrogen fixation rates in this basin vary widely. Here we investigate the variability in published estimates of excess nitrogen accumulation rates in the main thermocline of the subtropical North Atlantic, testing the assumptions and choices made in the analyses. Employing one of these previously described methods, modified here with improved estimates of excess N spatial gradients and ventilation rates of the main thermocline, we determine a total excess N accumulation rate of 7.8± molnyr 1. Contributions to excess N development include atmospheric deposition of high N:P nutrients (adding excess N at a rate of 3.0± mol N yr 1 for 38% of the total), high N:P dissolved organic matter advected into and mineralized in the main thermocline (adding excess N at 2.2± mol N yr 1 for 28% of the total), and, calculated by mass balance of the excess N field, N 2 fixation (adding excess N at 2.6± mol N yr 1 for 33% of the total). Assuming an N:P of 40 and this rate of excess N accumulation due to the process, N 2 fixation in the North Atlantic subtropical gyre is estimated at molnyr 1. This relatively low rate of N 2 fixation suggests that i) the rate of N 2 fixation in the North Atlantic is greatly overestimated in some previous analyses, ii) the main thermocline is not the primary repository of N fixed by diazotrophs, and/or iii) the N:P ratio of exported diazotrophic organic matter is much lower than generally assumed. It is this last possibility, and our uncertainty in the N:P ratios of exported material supporting excess N development, that greatly lessens our confidence in geochemical measures of N 2 fixation Elsevier B.V. All rights reserved. Keywords: North Atlantic; Nitrogen cycle; Nitrogen fixation; Dissolved organic matter; Atmospheric deposition 1. Introduction In the past decade, marine N 2 fixation has undergone significant re-evaluation, resulting in an approximate doubling in estimates of the global rate (Gruber and Corresponding author. Tel.: ; fax addresses: dhansell@rsmas.miami.edu (D.A. Hansell), dolson@rsmas.miami.edu (D.B. Olson), frank.dentener@jrc.it (F. Dentener), lzamora@rsmas.miami.edu (L.M. Zamora). Sarmiento, 1997; Gruber, 2004; Capone et al., 2005; Deutsch et al., 2007). In making atmospheric nitrogen available to the ocean biota as a nutrient, nitrogen fixation alleviates nutrient depletion in oligotrophic systems (Karl et al., 1997; Capone et al., 2005). It also compensates for oceanic loss of fixed N by denitrification (Codispoti et al., 2001), and possibly impacts atmospheric CO 2 budgets (Michaels et al., 2001). Increasing nitrogen fixation rate estimates have changed global ocean nitrogen budgets (Codispoti et al., 2001; /$ - see front matter 2007 Elsevier B.V. All rights reserved. doi: /j.marchem

2 D.A. Hansell et al. / Marine Chemistry 106 (2007) Table 1 Estimated rates of N 2 fixation in the North Atlantic Ocean Fixation rate (mol N yr 1 ) Geochemical techniques a Michaels et al. (1996) Gruber and Sarmiento (1997) Hansell et al. (2004) 3± Bates and Hansell (2004) This study Biological assessments b Carpenter (1983) Lipschultz and Owens (1996) Capone et al. (2005) Numerical modeling Coles et al. (2004) a Employing spatial variability in thermocline excess N. b Bottle incubations of diazotrophs (mostly Trichodesmium) and biomass surveys. Gruber, 2004), thereby leading to new questions on this marine biogeochemical process (Karl et al., 2002; Mahaffey et al., 2005). Global estimates of marine N 2 fixation are largely underpinned by the insights developed for the process in the subtropical North Atlantic Ocean. Gruber and Sarmiento (1997), for example, extrapolated their North Atlantic rate to the global ocean, and this global rate was subsequently employed by Codispoti et al. (2001) to suggest imbalance in the global ocean nitrogen budget. Since much of the foundational research on marine N 2 fixation has focused on the North Atlantic (Dugdale et al., 1961, 1964; Goering et al., 1966; Carpenter, 1973; Carpenter and McCarthy, 1975; Capone and Carpenter, 1982; Michaels et al., 1996; Gruber and Sarmiento, 1997; Mills et al., 2004; Capone et al., 2005), and rates in the North Atlantic guide estimates for the global ocean (Gruber, 2004), it is in this basin that we should have the most confidence in the fixation rates reported. Contrary to this expectation, estimates for N 2 fixation in the North Atlantic vary widely (Table 1). Investigations have included assessments of distributions of geochemical variables in the main thermocline, direct biological uptake techniques, and numerical models, but there is presently little basis for ascertaining which rates are correct. The geochemical approach assumes that mineralization of high N:P sinking biogenic particles derived from diazotrophs will result in nitrate concentrations in the main thermocline in excess of those predicted from phosphate concentrations and Redfield N:P stoichiometry (e.g., Redfield et al., 1963; Takahashi et al., 1985; Anderson and Sarmiento, 1994). Fanning (1987, 1992) first invoked N 2 fixation (and other processes such as atmospherically deposited nutrients) to explain the high N:P ratios in the thermocline waters of the North Atlantic subtropical gyre. To further evaluate the anomalous N:P signals, the derived variable N (N =NO 3 16PO ; as modified by Deutsch et al., 2001) was introduced by Michaels et al. (1996) and Gruber and Sarmiento (1997) as an index for NO 3 concentrations in excess (or in deficit) of that expected from the remineralization of PO 4 at the molar stoichiometry of 16:1. The spatial distributions of N in the main thermocline of the world's oceans generally reflect the distribution of oceanic N 2 fixation and denitrification (Gruber, 2004). Relatively elevated N values suggest net additions of N relative to P, while low values indicate net removal of N through denitrification. Following Michaels et al. (1996) and Gruber and Sarmiento (1997), Hansell et al. (2004) employed a similar measure of the excess of N relative to P in their calculation of DINxs (DINxs=NO 3 16PO 3 4 ). The difference between DINxs and N is the constant offset of 2.72, so that gradients in concentrations of excess N are equivalent whether reported as N or DINxs. The term excess N will be commonly used here to encompass both indices; the terms N and DINxs will be employed when those specific measures are discussed. Estimates of excess N accumulation rates in the subtropical North Atlantic by Michaels et al. (1996), Gruber and Sarmiento (1997), and Hansell et al. (2004) vary widely, so the true rates of accumulation remain uncertain. Rates determined by these authors, respectively, were mol N yr 1, mol N yr 1, and 2.3± mol N yr 1, for a fold range. The temporal variability of excess N in the North Atlantic (Bates and Hansell, 2004) will not explain the wide range implied by these results. Instead, it is probable that errors and/or biases exist in part from analysis methods and assumptions. Mechanisms other than N 2 fixation have been thought to account for only a small fraction of the total signal when estimated excess N accumulation rates are high. Michaels et al. (1996), for example, deemed dissolved organic matter (DOM) export (or, conversely, DOM import to the main thermocline) to be a relatively small contributor to excess N development, while Gruber and Sarmiento (1997) made a relatively small correction for excess N deposited from the atmosphere. But as estimates for total excess N accumulation rates themselves become smaller, inputs other than N 2 fixation can take on greater relative importance. In this paper, the role of N 2 fixation in creating the excess nitrate signature of the subtropical North Atlantic is reassessed. To do this, various inputs of high N:P material are quantified, including nutrients introduced

3 564 D.A. Hansell et al. / Marine Chemistry 106 (2007) by atmospheric deposition and by import of DOM to the main thermocline, with subsequent mineralization. Second, the assumptions used in previous geochemical analyses, and the impacts of these on the rate estimates for input of high N:P materials to the main thermocline of the subtropical North Atlantic, are investigated. The choices made by Michaels et al. (1996) for spatial gradients of N in the main thermocline of the subtropical North Atlantic, and for ventilation rates of the main thermocline, are evaluated. Similarly, the choices made by Gruber and Sarmiento (1997) and Hansell et al. (2004) for the volume of water over which volumetric rates of excess N accumulation should be applied are considered. Basin-wide excess N accumulation rates are then recalculated using the Michaels et al. (1996) method, employing improved estimates for nutrient gradients and thermocline ventilation rates from recent data sets. Finally, N 2 fixation rates required for the mass balance of excess N accumulation in the thermocline are estimated. CLIVAR (Climate Variability and Predictability) data, collected in 2003, are used to assess the excess nitrogen in the North Atlantic subtropical gyre. The major ocean circulation zones that define the gyre, as well as the locations of the CLIVAR lines included in the analysis, are shown in Fig. 1. Fig. 2 shows DINxs along one of those lines (A20) to be located in the depth range of m, lying on the density surface σ θ Calculating rates of excess N input from rates of accumulation Determining the rate of input of high N:P material required to generate an excess N signal in the thermocline requires knowledge of the N:P ratio of the sinking/ subducted material and the accumulation rate of the excess N. Simply, a sinking particle that is mineralized in the main thermocline at an N:P ratio = X will be observed as an increase of X-16 units of excess nitrogen. That is, each 1 unit of P exported and mineralized carries with it both the Redfield portion of N (16 units) and the excess N (X-16 units). The N input rate, carried by high N:P material, is calculated as N input rate due to high N:P material Accumulation rate of excess N ¼ ððn :P exp N:P Redfield Þ=N :P exp Þ ð1þ where N:P exp is the N:P molar ratio of the exported material and N:P Redfield is the Redfield N:P ratio. Eq. (1) is effectively identical to the tracer continuity equation derived by Deutsch et al. (2001; their Eq. (8)). Gruber and Sarmiento (1997) and Hansell et al. (2004) assumed an N:P of 125 for the exported material. Mahaffey et al. (2005) suggested that assuming an N:P ratio near 40 may be more appropriate, a change that increases by 45% the excess N input rates estimated from accumulation rates (Eq. (1)). Sañudo-Wilhelmy Fig. 1. General circulation of the subtropical North Atlantic with locations of CLIVAR sections used in the current study (A16N in the east, A20, and A22 in the west; all occupied in northern summer and fall 2003). The schematic of circulation is taken from the dynamic height field (0 700 m) of Stommel et al. (1978). The circulation components include the wind driven gyre and a southern shadow zone ventilated from the south by cross equator flow. In the northwest sector of the gyre lies the Gulf Stream recirculation (R), which is separated from the subduction zone in the eastern gyre by a separation point at S.

4 D.A. Hansell et al. / Marine Chemistry 106 (2007) Fig. 2. Distribution of DINxs (μmol kg 1 ) against A) depth (m) and B) density (σ θ ) on CLIVAR Repeat Hydrography section A20 (nominally 52 W). Positive DINxs values indicate N:P ratios in excess of Redfield expectations. et al. (2004) found that phosphate-replete Trichodesmium had N:P ratios very close to Redfield expectations. It is apparent that in nature the N:P ratio of diazotrophs is not conservative, demonstrating plasticity in elemental ratios (LaRoche and Breitbarth, 2005; Krauk et al., 2006; White et al., 2006). Likewise, the N:P of nutrients introduced by deposition from the atmosphere varies widely (Baker et al., 2003). Variability in the N:P of both these processes introduces uncertainty in rates estimated using Eq. (1). 3. Processes contributing to excess N accumulation In this paper, three nutrient input terms, all characterized by high N:P (N 16), that may contribute to the pool of excess N found in the main thermocline of the North Atlantic are assessed: the atmospheric deposition of nutrients, the import of DOM to the main thermocline with ventilation, and N 2 fixation. Potentially high N:P sinking particles formed by other processes are not considered. Prochlorococcus and Synechococcus, for example, can have elevated N:P ratios in oligiotrophic systems (Bertilsson et al., 2003; Heldal et al., 2003), but their contributions to excess N development in the thermocline is unknown. The small size of these cells, however, probably minimizes their contribution to the sinking flux. Produced and accumulated in the surface layer, DOM is exported to depth with ventilation of the ocean, whether by convection or subduction. There it is mineralized to contribute to the development of the excess N. Both N 2 fixation and atmospheric deposition add excess N directly to the surface layer, and each should result in the production of biogenic particles. For these two inputs, the processes ultimately moving excess N to the thermocline are not known (particle export is likely), but DOM may play a role as intermediary in the process. Only very small portions of surface DON are likely to have as its source N 2 fixation, however (Knapp et al., 2007; Meador et al., 2007) Nitrogen deposited to the subtropical North Atlantic from the atmosphere The atmospheric transport of various substances from the continents to the oceans is important in marine biogeochemical processes. The major focus recently has been on dust as a source of Fe because of the element's role as a limiting micro-nutrient in several ocean regions (Prospero et al., 1996; Jickells et al., 2005; Mahowald et al., 2005). The deposition of aerosols could also be important as a source of N-nutrients such as nitrate and ammonium, especially in the North Atlantic where the deposition of anthropogenic N is relatively high (Duce et al., 1991; Prospero et al., 1996; Galloway et al., 2004). The biological fixation of N 2 on the continents has doubled from the pre-industrial rate of mol N yr 1 to almost mol N yr 1 due largely to energy, fertilizer and crop production, and animal husbandry (Galloway, 1998). Total deposition of

5 566 D.A. Hansell et al. / Marine Chemistry 106 (2007) reactive N to the global open ocean is estimated at mol N yr 1 (Galloway et al., 2004). Nutrients introduced to the ocean by atmospheric deposition should leave signatures in the water column that are in several ways indistinguishable from those modifications on the nutrient field imposed by nitrogen fixation (see Mahaffey et al., 2003). While high N:P values have normally been taken as signs of nitrogen fixation in the North Atlantic, nitrogen deposited to the ocean surface from the atmosphere can similarly exhibit little associated P. N:P's of N30 are common (Duarte et al., 2006), with N:P N1000 reported in dry deposition to the subtropical North Atlantic (Baker et al., 2003) and values N500 in the Gulf of Aqaba (Chen et al., 2007). Chen (2004), in contrast, reported the N:P of soluble nutrients in dry deposition to the tropical North Atlantic at less than the Redfield ratio. Atmospherically deposited N can also have near-zero δ 15 N values (Hastings et al., 2003). Low δ 15 N values in suspended organic matter and zooplankton have commonly been taken as indications of N 2 fixation (Montoya et al., 2002; McClelland et al., 2003; Mahaffey et al., 2003). Both marine N 2 fixation and atmospherically deposited nutrients can leave these biogeochemical signatures, and both must contribute to the drawdown of sea surface inorganic carbon when ambient nutrients are depleted, a process primarily assigned to marine N 2 fixation (Michaels et al., 1994; Lee et al., 2002). The role of the atmosphere in generating elevated N:P ratios in the water column has been suggested previously for the North Atlantic and elsewhere. Pahlow and Riebesell (2000) suggested that increasing N:P in deep North Atlantic nutrients was due to increased deposition of atmospheric nitrogen from anthropogenic sources. As in the North Atlantic, the Mediterranean Sea is characterized by high nitrate to phosphate ratios ( 28:1) in the deep water. N budget analyses there suggest that deposition from the atmosphere is the most important source of excess nitrogen (Herut et al., 2002), while nitrogen fixation is an insignificant process in the system (Krom et al., 2004). To be determined is the extent to which the atmospheric deposition of nutrients contributes to the high N:P ratios and low δ 15 N values in the North Atlantic, and the extent to which the North Atlantic main thermocline serves as a sink for anthropogenic N transported via the atmosphere. Estimates for deposition of N to the North Atlantic (0 to 70 N) range from mol N yr 1 (Duce et al., 1991; Prospero et al., 1996; Galloway et al., 2004), with input to the tropical and subtropical sector estimated at mol N yr 1 (Duce et al., 1991). This rate is consistent with modeled deposition of 3.6 ± mol N yr 1 in the 0 40 N zone (estimate derived from Dentener et al., 2006; 30% uncertainty based on error reported for model estimates of N deposition over land vs. measured deposition rates), with Fig. 3. Distribution of total N deposition into the North Atlantic. The deposition fields of reduced (NHx) and oxidized (NOy) N were taken from Dentener et al. (2006) as the average of a 23 member model ensemble of global deposition. These deposition fields were compared to a variety of mainly land-based measurements since few measurements of deposition are available over the open ocean. The ocean is also a source of NH 3, which enters in the models as fixed fluxes from the Bouwman et al. (1997) inventory. These emissions, which are merely recycling of oceanic nitrogen, were subtracted from the deposition fields.

6 D.A. Hansell et al. / Marine Chemistry 106 (2007) strongest inputs downwind of NW Africa and the industrialized NE United States (Fig. 3). The role of marine P in negating the contributions to excess N development from addition of nutrients from the atmosphere requires evaluation. Marine P (P present in the surface ocean at the time of deposition) will lower the effective N:P ratio of the deposited nutrients, thus reducing the rate of excess N addition. The concentrations of DIP are low in the surface waters of the western subtropical North Atlantic, with disparate mean inorganic P concentrations of 0.48 nmol L 1 and 40 nmol L 1 having been reported by Wu et al. (2000) and Sohm and Capone (2006), respectively. Cavender-Bares et al. (2001) reported inorganic Pb10 nmol L 1 in the western subtropical Atlantic. If the lowest concentrations reported are correct, marine P may have little affect on excess N. The role of marine P remains unaccounted for here Export of high N:P dissolved organic matter with thermocline ventilation Away from continental influences (i.e., fluvial inputs), dissolved organic matter (DOM) is at its highest concentrations in the subtropical gyres of the open ocean (Hansell, 2002). In the surface layer of the North Atlantic gyre, dissolved organic carbon (DOC) concentrations are generally b80 μmol kg 1 (Hansell and Carlson, 2001; Hansell, 2002), dissolved organic nitrogen (DON) is b6 μmol kg 1 (Hansell and Carlson, 2001; Hansell and Follows, in press), and dissolved organic phosphorus (DOP) is b0.1 μmol kg 1 (Wu et al., 2000), for a molar C:N:P of 800:60:1. Net DOM production largely occurs where new nutrients mix into the euphotic zone (Hansell and Carlson, 1998). Ekman pumping supplies nutrients to the euphotic zone of the subpolar gyre and over the equator (Williams Fig. 4. Concentrations of A) nitrate (μmol kg 1 ), B) phosphate (μmol kg 1 ), C) DINxs (μmol kg 1 ), and D) DON (μmol kg 1 ) plotted against water mass age (pcfc11 age in years) between 20 and 50 N on A16N on the density surface σ θ

7 568 D.A. Hansell et al. / Marine Chemistry 106 (2007) Fig. 5. Upper ocean concentrations of DOP (μmol kg 1 ) between 32 and 35 N along 55 W in February DOP was measured by UV oxidation of the water; phosphate was measured using standard colorimetric methods. Surface concentrations bound the mean DOP of 0.07 μmol kg 1 reported by Wu et al. (2000) for that region. and Follows, 1998), while horizontal Ekman transport delivers resultant DOM from the sites of formation to the sites of ocean ventilation (Abell et al., 2000; Hansell, 2002; Hansell et al., 2002; Williams et al., 2006). Once exported, mineralization results in the loss of DOM (Doval and Hansell, 2000; Hansell and Carlson, 2001) and the accumulation of mineralization products (dissolved inorganic carbon, nitrate, and phosphate). Given the elemental composition of DOM (high N:P), the inorganic nutrients released with mineralization should accumulate at ratios greater than the Redfield expectation, thus contributing to development of excess N. Though there are relatively few DON and DOP data available from the North Atlantic, we can establish bounds for the contribution of exported DOM to excess N development with the existing data. As a first assessment, nitrate, phosphate and DON concentrations determined on the CLIVAR A16N line in 2003 are employed. Concentrations of these measured variables, along with the derived variable DINxs, are plotted against water mass age in Fig. 4. The nutrient concentrations and DINxs increased with age, while DON decreased by 1 2 μmol kg 1 N. DOP was not measured on A16N so the contribution of DOM to DINxs development cannot be quantified directly. To put bounds on the contribution of DOM export, DOP measured elsewhere in the North Atlantic subtropical gyre is used (Fig. 5). The ΔDOP from surface (i.e., the source water for subduction) to 500 m in the western Sargasso Sea was 0.05 μmol kg 1, similar to the vertical concentration changes in DOP reported by Landolfi (2005) across the subtropical North Atlantic. Fig. 6 was created to aid in understanding the DINxs increase expected with mineralization of varying concentrations of DON and DOP. According to Fig. 6,a 1 2 μmol kg 1 ΔDON (Fig. 4) and a ΔDOP (bounding the 0.05 μmol kg 1 value in Fig. 6) will result in a DINxs increase of μmol kg 1 N. As the measured increase in DINxs on CLIVAR A16N (Fig. 4C) was 0.5 μm, this calculation leaves open the possibility that much of the DINxs accumulation in that ocean sector is due to mineralization of exported DOM. In Section 5.1, we will assume that to be the case. An important role for DOM in DINxs formation was similarly suggested by Landolfi (2005), who found that up to 80% of DINxs formation could be due to the mineralization of high N:P DOM. DINxs development due to DOM mineralization should be strongest near the sites of ventilation (in recently subducted water), and considerably weaker as the water mass ages (note the relatively conservative DON concentrations at ages N15 years in Fig. 4D). There is little difference in DON concentrations between the oldest waters on the σ θ surface (Fig. 4D) and on the same density at the BATS site (Hansell and Carlson, 2001), indicating little additional DOM oxidation with further aging. As such, to increase DINxs concentrations in the main thermocline beyond the values seen on A16N, other inputs of DINxs are required elsewhere in the gyre (i.e., atmospheric deposition and N 2 fixation). 4. Assessment of previous estimates of excess N accumulation in the subtropical North Atlantic As mentioned, the several efforts to quantify the accumulation rate of excess N in the subtropical North Fig. 6. Plot of the estimated change in DON concentrations (ΔDON) required as a function of change in DOP concentrations (ΔDOP) to generate DINxs increases of 0.5, 1, 1.5, and 2 μmol kg 1 along an isopycnal surface. The box indicates the bounds of ΔDOP and ΔDON discussed in the text.

8 D.A. Hansell et al. / Marine Chemistry 106 (2007) Atlantic have widely divergent results. Prior to newly estimating this rate, it is instructive to first evaluate the assumptions employed, and identify possible errors, in the previously published analyses Michaels et al. (1996) Michaels et al. (1996) was the first important effort to estimate the rate of accumulation of excess N using geochemical data sets. Their approach was to evaluate N gradients on isopycnal surfaces in the main thermocline of the subtropical North Atlantic. They then employed tritium helium calibrated models of ventilation on those surfaces (Jenkins, 1980; Sarmiento, 1983) to estimate rates of N formation. The calculation performed for each density layer of interest was: Excess N accum: rate ðmol N yr 1 Þ ¼ DN ð mol m 3 Þventilation rateðm 3 yr 1 Þ ð2þ Both the N concentration gradients and the ventilation rates of the system used in Michaels et al. (1996) will be checked here. The N gradient employed by Michaels et al. (1996) was established between the regions where each density surface outcrops during winter (initial N values at the time of ventilation; with most data taken from GEOSECS (Geochemical Ocean Sections Study) and TTO (Transient Tracers in the Ocean) in the 1970's and early 1980's, respectively) and the BATS (Bermuda Atlantic Time-series Study) site near Bermuda taken as the final N value (nominally 32 N, 64 W; data collected during the early 1990's). The resulting N concentration differences (ΔN ) were reported to range from b1 μmol kg 1 at low (near σ θ 26) and high (σ θ N27.2) densities, to N3 μmol kg 1 at densities near σ θ 26.5 (Fig. 7). A first test of the N gradients employed by Michaels et al. (1996) can be made by comparison to the data employed by Gruber and Sarmiento (1997). The data Gruber and Sarmiento (1997) used (including data from GEOSECS and TTO) were well distributed across the subtropical North Atlantic, including both regions of ventilation and regions with the oldest thermocline water (greatest time since ventilation), and so should represent the full concentration ranges in N. In the density range σ θ = , their N concentration gradient was b1.5 μmol N kg 1, while at it was 2 μmol N kg 1 (Fig. 7). In approximately these density bands, Michaels et al. (1996) employed mean N gradients of 1.2 and 3.1 μmol N kg 1, respectively. Fig. 7. Delta excess N (μmol kg 1 ), the range of excess N concentrations on specific density surfaces between the sites of water mass ventilation and the oldest waters, as employed by Michaels et al. (1996) ( ), as estimated from Fig. 16 in Gruber and Sarmiento (1997) (ovals), and as indicated by the CLIVAR data shown in Fig. 8 (stippled boxes). In the lower density band, the N concentration gradient was similar in the two analyses, while at the higher density band the gradient employed by Michaels et al. was higher by one third. The Michaels et al. (1996) analysis gave a mean increase in N of 3.1 μmol N kg 1 for the σ θ surfaces (and 2.4 μmol N kg 1 for ), both higher than indicated in the Gruber and Sarmiento (1997) analysis (Fig. 7). A second test for N gradients employs recently collected CLIVAR data from the North Atlantic (unlike the data employed by Michaels et al. (1996) and Gruber and Sarmiento (1997), which were collected over more than a decade, the CLIVAR data were collected contemporaneously in 2003). Three density ranges reported here cover most of the main thermocline: σ θ ranges of , , and (Fig. 8). On the lightest density layer, the full excess N range was 2 μmol N kg 1, consistent with the ranges reported by Michaels et al. (1996) and Gruber and Sarmiento (1997) (both near 1.5 μmol N kg 1 ; Fig. 7). On the intermediate density layer, the range was μmol kg 1, while on the deepest density layer considered the N range was mostly μmol N kg 1. These ranges in the intermediate and higher density waters are lower than those employed by Michaels et al. (Fig. 7), in some cases up to a factor of two. Gruber (2004) also noted that the Michaels et al. (1996) N gradients are larger than more recent data support. By providing better spatial coverage and minimizing temporal aliasing, the CLI- VAR data (which indicate smaller ΔN gradients)

9 570 D.A. Hansell et al. / Marine Chemistry 106 (2007) Fig. 8. Covariations of DINxs concentrations and pcfc-11 ages on density surfaces A) σ θ , B) σ θ , and C) σ θ in the North Atlantic. Data are from CLIVAR lines A16N (triangle), A20 (square), and A22 (circle); from depths greater than 100 m. are likely to provide a more reliable indication of the true gradients than the two small regions employed by Michaels et al. (1996). The second component of Eq. (2) to be evaluated is the ventilation rates assigned for the density layers of interest. Michaels et al. (1996) used values from Jenkins

10 D.A. Hansell et al. / Marine Chemistry 106 (2007) (1980) and Sarmiento (1983), with those data presented in Fig. 9A. There is general agreement between the two data sets at σ θ N 26.5, but because rates are dissimilar at lighter densities, there is a wide discrepancy for cumulative ventilation of the σ θ band (Fig. 9B). The cumulative ventilation rates of that layer are 88 Sv and 34 Sv using the data of Jenkins (1980) and Sarmiento (1983), respectively, and this spread is the reason for the wide range in the Michaels et al. (1996) estimates of excess N development. The ventilation rates employed by Michaels et al. (1996) based on tritium data are higher by 2 7 fold compared to several other estimates of ventilation in the North Atlantic (Table 2), which range from 12 Sv to 16 Sv for similar density ranges. The volume flux estimates in Jenkins (1980) and Sarmiento (1983) are based on tritium-deduced residence times for thermocline waters. These estimates exceed the total Sverdrup circulation at 24 N in the North Atlantic (30 Sv; Schmitz et al., 1992), which is at the upper end of the expected Table 2 Estimates of the ventilation rates of the main thermocline in the subtropical North Atlantic Source Density range considered Rate (Sv) Jenkins (1980) Sarmiento (1983) Woods and Barkmann (1986) Speer and Tziperman (1992) Qiu and Huang (1995) Marshall et al. (1999) Haine et al. (2003) Valdivieso Da Costa et al. (2005) subduction rate. The convective addition of tritium onto surfaces within the Gulf Stream circulation may result in an overestimate of ventilation rates with the tritium data set. More investigation is required to explain the high estimates of ventilation rates in those early analyses. A final point about the Michaels et al. (1996) analysis is that they calculated the rate of excess N accumulation, but reported this rate as equivalent to N 2 fixation without consideration for the N:P ratio of the sinking material. Assuming an N:P ratio of 40 and employing Eq. (1) increases their estimate of N 2 fixation to mol N yr Gruber and Sarmiento (1997) Gruber and Sarmiento (1997) followed Michaels et al. (1996) with a broader, basin-wide survey of the distribution of N in the North Atlantic (largely using TTO and GEOSECS data and data from individual cruises). To estimate the basin-wide rates of excess nitrate accumulation, they plotted N distributions along isopycnal surfaces, adding water mass ages to estimate volumetric rates of excess nitrate accumulation on those surfaces. With volumetric rates in hand, they calculated the rate of excess nitrate accumulation in the entire North Atlantic by applying their rates of excess N production to the volumes of the density layers of interest. The calculation applied to each density layer was essentially: Excess N accum: rate ð mol N yr 1 Þ ¼½DN ð mol m 3 Þ=DAge ðyrsþš Volume ðm 3 Þ ð3þ Fig. 9. A. Ventilation rates for the main thermocline (σ θ ) in the North Atlantic from Jenkins (1980) and Sarmiento (1983). B. Cumulative ventilation rates (beginning at σ θ 26) using data from 4 sources (Jenkins, 1980 ( ); Sarmiento, 1983 ( ); Qiu and Huang, 1995 ( ); Haine et al., 2003 ( )). The two components of this calculation are evaluated here. The first component considered is the volumetric rate of accumulation of excess N (bracketed term in Eq. (3)). In general, the volumetric rates determined by Gruber

11 572 D.A. Hansell et al. / Marine Chemistry 106 (2007) Fig. 10. Cumulative volumes of water in the density range σ θ of the North Atlantic, starting at 10 N and ending at 45 N. Values are placed at midpoints of the 5 latitude bins for which volumes were estimated. and Sarmiento (1997) are similar to those determined independently by Hansell et al. (2004), who used data from the WOCE (World Ocean Circulation Experiment) program. Hansell et al. (2004) reported excess N production of 0.43 μmol kg 1 yr 1 near σ θ 26.0, very similar to the Gruber and Sarmiento (1997) rate of 0.41 μmol kg 1 yr 1 on σ θ Adjustment of the Gruber and Sarmiento (1997) estimate for the Deutsch et al. (2001) modification of the N equation increases their rate to 0.47 μmol kg 1 yr 1 (this adjustment makes gradients in N equivalent to gradients in DINxs). On σ θ 26.5, Hansell et al. (2004) found that excess N increased at a rate of 0.09 μmol kg 1 yr 1. In comparison, the rate estimated by Gruber and Sarmiento (1997) for that surface (using interpolation of their estimates on bounding density surfaces and adjusted for Deutsch et al., 2001) was 0.15 μmol kg 1 yr 1, or 65% higher than Hansell et al. (2004).Onσ θ 27.0, the rate of change in excess N estimated by Hansell et al. (2004) was 0.04 μmol kg 1 yr 1, or 53% of the (adjusted) rate estimated by Gruber and Sarmiento (1997). Reasons for the observed rate inequalities (different water mass age tracers used in each study and natural variability in the system) are discussed in Hansell et al. (2004). The second component of Eq. (3) investigated is the volume of water over which the Gruber and Sarmiento (1997) volumetric rates of excess N accumulation were applied in order to estimate the basin-wide rate of accumulation. Gruber and Sarmiento (1997) applied their volumetric rates to an area that exceeds the subtropical gyre in latitude and includes a portion of the Gulf of Mexico (10 50 N and W). As seen in Eq. (3), the rate of excess N accumulation scales directly with volume, so the choice of volume is critical. The volume on specific isopycnal surfaces varies with increasing latitudinal range. The increase in water volume (in the density band σ θ 26 27) for each 5 Fig. 11. Stocks of DINxs (mmol N m 2 ) in the range of σ θ Integration was conducted over the depths in which DINxs values were N0 μmol kg 1.

12 D.A. Hansell et al. / Marine Chemistry 106 (2007) latitude bin (from 10 N to 45 N) in the North Atlantic is mostly linear (Fig. 10). Employing the volume underlying N in Eq. (3), for example, doubles the estimate for the rate of excess N accumulation relative to the outcome if N is chosen (a zonal range that better reflects elevated N 2 fixation rates; Carpenter, 1983; Capone et al., 2005). As such, the choice of latitudinal range is a primary determinant for the magnitude of the rate estimate. To determine if the Gruber and Sarmiento (1997) choices for surface area (and water volume) are correct, we first consider the distributions of excess N (focusing on waters with DINxs concentrations N0 μmol N kg 1 ) in the North Atlantic. Fig. 2 shows the southern limit of excess N on A20 to be near 10 mn and the northern limit to be the south wall of the Gulf Stream at about 37 N. This distribution suggests that the northern boundary employed by Gruber and Sarmiento (1997) (50 N) is inappropriate. The distribution of the water column inventory of excess N (Fig. 11) confirms the distribution to be largely limited to 10 to b40 N. The simple presence of excess N, however, cannot be taken as evidence for the area over which excess N inputs occur. Taking the data from Fig. 8C asan example, excess N accumulates during the earliest ages of the density layer considered, but there is little further accumulation with increased aging of the water (in this case at ages N20 years). These data indicate that the youngest waters accumulate excess N, but as the water circulates and ages into the gyre (Fig. 1), it moves away from zones of high excess N input and is conservatively transported into the gyre. Hansell et al. (2004) addressed the issue of volume choice by estimating the areas and volumes where excess N accumulation was evident in the data (where increases in excess N with increasing age were seen). They argued that the areas where excess N development takes place were regional and not basin-wide, so the volumes under smaller (regional) areas were employed in their calculations of the basin-wide rate of excess N accumulation. The volumes they applied were 5 7 times smaller than the volumes employed by Gruber and Sarmiento (1997) over the same density intervals. This difference in volume choice played the primary role in the 8-fold difference in basin-wide rates of accumulation by the two analyses ( mol yr 1 by Gruber and Sarmiento (1997) versus 2.3± mol N yr 1 by Hansell et al., 2004). Unfortunately, the WOCE and CLIVAR coverage of the North Atlantic remains inadequate to provide high confidence in the actual spatial areas (and volumes) over which the volumetric rates should be applied. While the findings of Hansell et al. (2004) could not support the application of their volumetric accumulation rates to a wider swath of the basin, the data density is inadequate for defining the areas of accumulation with confidence. Large spatial gaps exist, for example, between WOCE hydrographic sections A16 and A20 (Fig. 2), where the accumulation of excess N is likely to be strong. The sparse sections also fail to resolve the gyre structure (Fig. 1). 5. Re-analysis of total excess N accumulation and N 2 fixation Eq. (3) cannot be used with confidence to calculate the total excess N accumulation rate since the choice of volumes to be employed is presently subjective. A more objective approach for estimating excess accumulation rates, one where water volumes need not be identified, is that of Eq. (2) as used by Michaels et al. (1996). The concentration ranges in excess N and the ventilation rates can be objectively determined and employed in Eq. (2). As demonstrated above, both of those variables were overestimated in the Michaels et al. (1996) analysis, but more accurate data can be used in a re-analysis. Two of the papers with lower estimates of ventilation (Qiu and Huang, 1995; Haine et al., 2003) provide rates on specific density surfaces, which are used here (along with estimates of the gradients in excess nitrate taken from Fig. 8) to calculate rates of accumulation of excess nitrate on those surfaces. This calculation repeats that of Michaels et al. (1996), but uses more reliable estimates of both ventilation rates and excess nitrate gradients. Using Qiu and Huang (1995) data results in rates of accumulation of excess nitrate of mol N yr 1, while the result using the Haine et al. (2003) model is mol N yr 1 (Table 3). The full range from these analyses ( mol N yr 1,ora central value of mol N yr 1 with half the range assigned as the error) are 3 10 fold lower than the estimate by Michaels et al. (1996), 2 3 fold lower than Gruber and Sarmiento (1997), and 2 4 times larger than Hansell et al. (2004) Marine N 2 fixation assessed by mass balance of the excess N field The rate of excess N added to the main thermocline by N 2 fixation will be calculated here using assumptions of mass balance of the summed excess N inputs. Assumptions are i) that all excess N has the same fate, that being accumulation in the main thermocline, regardless of input mechanism or pathway, and ii) that

13 574 D.A. Hansell et al. / Marine Chemistry 106 (2007) Table 3 Estimates for the rate of accumulation of excess nitrate in the main thermocline of the subtropical North Atlantic Density Ventilation rate a (Sv) Gradient in excess nitrate b (μmol kg 1 ) Rate of accumulation (mol N yr 1 ) Qiu and Huang (1995) Summed accumulation rate Haine et al. (2003) Summed accumulation rate a Ventilation rates from Qiu and Huang (1995) estimated from their Fig. 13; rates from Haine et al. (2003) estimated from their Fig. 6 (Experiment 1; ventilation tracer S vt ). b Excess nitrate concentration gradients taken from Fig. 8; data from lowest density ranges (σ θ b26.35) are not shown in Fig. 8 but were taken from the same CLIVAR data sets. no in situ processes (e.g., unequal mineralization length scales for N and P) play a role in generating excess N. In Section 5, the rate of excess N accumulation (taken to be equal to the sum of inputs by DOM export, atmospheric deposition and N 2 fixation) was estimated for the main thermocline. We now estimate the rates of inputs of excess N by atmospheric deposition and by DOM export. Imbalance between the sum of these two inputs and the total excess N accumulation rate leads us to an estimate of the input of excess N likely due to N 2 fixation. In Section 3.1, the rate of atmospheric deposition of bio-available N was calculated at 3.6± mol N yr 1. If the N:P of these nutrients is 1000 (Baker et al., 2003), this process (Eq. (1)) adds excess N to the main thermocline at a rate of 3.5± mol N yr 1, thus accounting for 44% of the total accumulation rate (total excess N accumulation =7.8± mol N yr 1 ; Section 5). An N:P of 100, which will be assumed here in order to complete the mass balance calculation, results in an excess N accumulation rate of 3.0 ± mol N yr 1 (38% of total accumulation). As for DOM export, the process should contribute to DINxs development near the sites of ventilation, and less so as the subducted water ages. The measured increase in DINxs on A16N (Fig. 4C) was 0.5 μm, a value that could be due fully to mineralization of exported DOM (see calculations in Section 3.2). Assuming that DOM was the primary source of DINxs on that line, we can estimate the upper bound of the contribution of DOM to excess N development in the main thermocline of the North Atlantic. A ventilation rate of 14 Sv (Table 3) and a DINxs development of 0.5 μm(fig. 4)is equal to 2.2± mol N yr 1 (assumed uncertainty of 50%). If DOM is the primary source of the signal, DOM oxidation contributes 28% of the 7.8 ± mol N yr 1 total excess N accumulation in the thermocline. Here the rate of accumulation of excess N in the main thermocline of the subtropical North Atlantic was calculated at 7.8± mol N yr 1. If the 3.0± mol N yr 1 deposited from the atmosphere to the tropical and subtropical North Atlantic accumulates as excess N in the thermocline, 38% of the total accumulated excess N could be due to atmospheric inputs. Export of DOM at 2.2± mol N yr 1 could account for 28%. The 33% balance (2.6± mol N yr 1 ; error propagated from errors of other inputs and total N accumulation) may ultimately be due to N 2 fixation. Given an excess N accumulation rate due to N 2 fixation of 2.6± mol N yr 1, we use a range of N:P ratios and Eq. (1) to estimate the N 2 fixation rates required to create this excess N signal. N:P ratios of 125 and 40 result in N 2 fixation rates of and mol N yr 1, respectively. The highest rate estimated here equals 26% of the recent direct estimate of N 2 fixation by Trichodesmium colonies ( mol N yr 1 ) in the North Atlantic by Capone et al. (2005), and is a much smaller fraction of possible basin wide N 2 fixation suggested by Davis and McGillicuddy (2006), who report that the biomass of Trichodesmium spp. commonly collected with nets greatly underestimates true biomass, resulting in gross underestimates of rates. 6. Uncertainties The estimate for N 2 fixation resulting from the mass balance calculation is 26% of a recent direct (biological) estimate of N 2 fixation in the North Atlantic by Capone et al. (2005). IfDavis and McGillicuddy (2006) are correct that rate estimates based on net collections of Trichodesmium are greatly underestimated, then the difference between geochemical evidence for N 2 fixation and direct biological estimates are grossly out of balance. Assuming that temporal variability is not the cause of the large difference in estimates, and assuming that at

14 D.A. Hansell et al. / Marine Chemistry 106 (2007) least one of these estimates for fixation is approximately correct, then either the estimate generated here based on geochemical mass balance is too low or the Capone et al. (2005) estimate is too large. To generate too low a rate here, the terms to calculate the total excess N accumulation rate (Eq. (2)) have to be too small. Most recent estimates of the ventilation rate are in agreement (Table 2), thus increasing our confidence in those values. The gradients in excess N concentrations are field measured values, which are generally supported by the Gruber and Sarmiento (1997) results using other data, so there is little room for a large error there. Confidence in the terms of Eq. (2) suggests that the total excess N accumulation rate estimated here is representative of the true value. This value, in turn, sets an upper limit to the sum of the inputs required to generate the excess N signal. Excess N input from N 2 fixation cannot, according to this logic, be larger than the total excess N input (using Eq. (3)). If DOM export and atmospherically deposited nutrients contribute to excess N development, then N 2 fixation has to be commensurately smaller than the total excess N accumulation rate. This mass balance analysis precludes N 2 fixation as large as that reported by Capone et al. (2005), unless a large part of the fixed N does not accumulate as excess N in the thermocline or some of the other assumptions made here are wrong. There is no evidence in the DINxs field that exported N fixed by diazotrophs is mineralized deeper in the water column, but perhaps the material is exported to the seafloor and retained there. Another possible fate of diazotroph N is to remain in the upper layer of the ocean, to be exported to the north with the thermohaline portion of the total gyre circulation (the net transport through the gyre). If the thermohaline circulation is taken to be 16 Sv, and if all of the N 2 fixation measured by Capone et al. (2005) is lost from the gyre by that mass transport, the upper layer total N concentrations would have to increase in the throughflow water by 3 μmol N kg 1 (i.e., the water entering the gyre from the south would have a total N concentration 3 μmol N kg 1 less than the water leaving the gyre to the north). This large change in total N has not been observed, but some fraction of that value might be reasonable. This scenario is countered, however, by Knapp et al. (2007) who reported that neither concentrations nor δ 15 N of DON varied in the North Atlantic relative to gradients in N 2 fixation (or atmospheric dust fluxes), which are interpreted as indicating that the diazotrophic N is largely exported vertically within the gyre circulation and not horizontally across the gyre boundaries. Perhaps the N:P ratio of organic matter exported as a result of diazotrophic production is much lower than the 40 assumed here. Gruber and Sarmiento (1997) demonstrated graphically that even small changes in the N:P ratio, when low (b40), can have exponential consequences on estimates of N 2 fixation rates. If the excess N accumulation due to N 2 fixation is indeed mol N yr 1, as calculated above, the N 2 fixation rate reported by Capone et al. (2005) requires an N:P of 19 (using Eq. (1)). If all of the excess N accumulation ( mol N yr 1 ) is due to N 2 fixation, such that DOM export and atmospheric nutrients contribute nothing, then an N:P of 31 would meet the Capone et al. (2005) estimate. While most N:P values resulting from N 2 fixation are higher than these, low ratios have been observed (Sañudo-Wilhelmy et al., 2004; Krauk et al., 2006). Given that small variations in the N:P force large changes in calculated nitrogen fixation rates, the integrity of the geochemical mass balance technique for assessing the process is challenged. It follows that the direct biological measures of N 2 fixation (e.g., Capone et al., 2005) cannot be confirmed by comparison to estimates based on geochemical fields of excess N. One important caveat in this analysis is that while denitrification is not thought to be significant in the thermocline waters, it has important impacts on the basin N budget through its actions on the shelves (Seitzinger and Giblin, 1996; Fennel et al., 2006). Introducing water with a negative excess N value to the thermocline will reduce the calculated excess N accumulation rate (hence the mixing models and corrections for southern component waters employed by Gruber and Sarmiento (1997) and Hansell et al. (2004)). The rates of denitrification in the North Atlantic shelves estimated by Seitzinger and Giblin (1996) and by Fennel et al. (2006) are mol N yr 1 and mol N yr 1 N, respectively. To influence our estimates of the accumulation rate of excess N in the gyre, shelf water with a strong denitrification signal (i.e., negative excess N values) would need to strongly ventilate the main thermocline. If this process is occurring, and if a strong denitrification signal is being introduced, then the data should demonstrate it. In Fig. 2 we see the negative excess N values at the far northern end of the Section A20, consistent with expectations for a denitrification signal coming off the shelf. But those waters are north of the Gulf Stream, and mixing into the gyre is required for impact on the main thermocline. If such mixing is occurring, then there should be evidence for this in Fig. 8. We would expect young waters with negative excess N values to be seen mixing into the gyre on the western sections (A22 and A20). Such water/mixing is

15 576 D.A. Hansell et al. / Marine Chemistry 106 (2007) not evident in Fig. 8. Young, low excess N water is located on the eastern most line only (A16N); we have assigned that signal to the export of high N:P DOM (Section 4.2) since that water is largely formed by the subduction of surface water. Further analysis is required to ascertain the impact of denitrification on estimates of excess N accumulation in the main thermocline. Finally, we do not know the mechanism of transfer of N and P from the diazotrophs (or from other biogenic particles formed from atmospheric nutrients) to the main thermocline. It is possible that nitrogen fixed by Trichodesmium sp. is transferred to other organisms (at unknown N:P) that are more readily exported. Here it was presumed that nitrogen fixed in the sea (as N 2 ) and that deposited from the atmosphere follow similar but unknown paths to the main thermocline. Until the paths are identified, we must leave open the possibility that still unidentified processes contribute to excess N development in the main thermocline, or that the main thermocline is not the destination for all of the excess N introduced to the surface layer. It may be that a fraction of the exported DOM has, as its ultimate source, N 2 fixation or atmospheric deposition, so that DOM export is a mechanism of export for those processes but not a uniquely separate process for excess N development. Neither Knapp et al. (2007) nor Meador et al. (2007), however, found evidence for N 2 fixation as an important source of dissolved organic N in the North Atlantic, as determined with stable isotope abundances of nitrogen. 7. Conclusions Estimates for the global annual rate of marine nitrogen fixation have doubled over the past decade, yet our best studied ocean basin (the North Atlantic) evidently has fixation rates that are considerably lower than most recent estimates based on geochemistry suggest, and upon which several global estimates are founded. Total excess N accumulation in the main thermocline of the North Atlantic is estimated at 7.8± mol N yr 1, with 38% of this due to atmospheric deposition of high N:P, low δ 15 N nutrients. High N:P DOM exported with subduction contributes an additional 28%, leaving 33% to N 2 fixation. Unless the N:P of exported material is very low, the rate of N 2 fixation determined here using geochemical tracers is considerably smaller than the rates recently determined using direct biological measures (Capone et al., 2005) or inferred by Trichodesmium biomass distributions (Davis and McGillicuddy, 2006), and the rate of accumulation of excess N in the main thermocline is considerably smaller than earlier estimates of that process (Michaels et al., 1996; Gruber and Sarmiento, 1997). Fig. 12. Primary locations of input by the three main processes hypothesized to contribute to the development of excess N in the main thermocline of the subtropical North Atlantic. The processes are N 2 fixation (NF), DOM export (DOM), and atmospheric deposition (AD). All of the processes will contribute to excess N development across much of the subtropical gyre, but only the primary sites of input for each is shown.