Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 The fractionation of silicon (Si) stable isotopes by biological activity in the surface ocean makes the stable isotope composition of silicon (δ30Si) dissolved in seawater a sensitive tracer of the oceanic biogeochemical Si cycle. We present a high-precision dataset that characterizes the δ30Si distribution in the deep Atlantic Ocean from Denmark Strait to Drake Passage, documenting strong meridional and smaller, but resolvable, vertical δ30Si gradients. We show that these gradients are related to the two sources of deep and bottom waters in the Atlantic Ocean: waters of North Atlantic and Nordic origin carry a high δ30Si signature of ≥+1.7‰ into the deep Atlantic, while Antarctic Bottom Water transports Si with a low δ30Si value of around +1.2‰. The deep Atlantic δ30Si distribution is thus governed by the quasi-conservative mixing of Si from these two isotopically distinct sources. This disparity in Si isotope composition between the North Atlantic and Southern Ocean is in marked contrast to the homogeneity of the stable nitrogen isotope composition of deep ocean nitrate (δ15N-NO3). We infer that the meridional δ30Si gradient derives from the transport of the high δ30Si signature of Southern Ocean intermediate/mode waters into the North Atlantic by the upper return path of the meridional overturning circulation (MOC). The basin-scale deep Atlantic δ30Si gradient thus owes its existence to the interaction of the physical circulation with biological nutrient uptake at high southern latitudes, which fractionates Si isotopes between the abyssal and intermediate/mode waters formed in the Southern Ocean.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The oceanic biogeochemical cycle of silicon (Si) is tied to that of carbon by the important role played by diatoms – siliceous phytoplankton – in global new production [Smetacek, 1999; Ragueneau et al., 2000, and references therein]. Diatoms additionally tend to dominate the phytoplankton community in dynamic upwelling regions [Margalef, 1978; Jin et al., 2006] such as the Southern Ocean, which play an important role in the oceanic control of atmospheric pCO2 [Marinov et al., 2006; Gruber et al., 2009]. Silicon fluxes associated with diatom productivity and export are so large that they dominantly control the oceanic Si cycle [Tréguer et al., 1995]. Since diatoms preferentially incorporate the lighter isotopes of Si into their opaline frustules [De La Rocha et al., 1997], diatom uptake of Si in the sunlit surface ocean alters the stable isotope composition of dissolved Si in seawater (expressed as δ30Si; see section 2.2). As a result of this, seawater δ30Si values trace biogeochemical processes affecting Si. Thus, constraining the mechanisms that govern the oceanic δ30Si distribution allows direct inference of the pathways and processes by which Si is cycled within the ocean. This is particularly interesting in the context of an emerging paradigm of oceanic nutrient cycling [Sarmiento et al., 2004, 2007; Palter et al., 2010] that stresses the importance of lateral transport processes, modulated by biology in the Southern Ocean, in determining global oceanic nutrient distributions, and thus ultimately oceanic primary productivity [Sarmiento et al., 2004]. In addition, a better knowledge of the processes that control the modern δ30Si distribution will allow more robust interpretations of diatom opal δ30Si records from deep-sea sediment cores [e.g., De La Rocha et al., 1998; Pichevin et al., 2009], improving our understanding of how diatom productivity may have affected atmospheric pCO2 over glacial–interglacial cycles [Brzezinski et al., 2002; Matsumoto et al., 2002; Crosta et al., 2007]. Together, these two complementary issues provide strong motivation to study the modern oceanic δ30Si distribution, particularly in the Atlantic Ocean.
 In-depth studies of seawater δ30Si [Varela et al., 2004; Cardinal et al., 2005; Reynolds et al., 2006a; Beucher et al., 2008; Fripiat et al., 2011] have thus far focused on high-nutrient, low-chlorophyll regions that are the key areas for ocean–climate interaction and overlie opal-rich sediments. The Atlantic Ocean has been somewhat neglected, most likely due to its modest contribution to biological Si cycling in the modern ocean, reflected by a very low opal export flux for its area (5% of global export) [Sarmiento et al., 2007]. However, the fact that the primary formation regions of the deep and bottom waters that fill the global ocean are situated at its northern and southern end [Warren, 1981] makes the Atlantic highly relevant to the global oceanic Si cycle. Furthermore, as it contains the world's youngest and least Si-rich deepwaters (0–300 yr ventilation ages) [Matsumoto, 2007], a firm handle on the δ30Si distribution in the deep Atlantic is crucial to better resolve an interesting feature of the oceanic δ30Si distribution: it appears that the North Atlantic and North Pacific have distinct deepwater δ30Si values, which would contrast the homogeneity of the stable nitrogen isotopic composition of deep oceanic nitrate, δ15N-NO3 [e.g., Sigman et al., 2009]. However, the Atlantic end of this interbasin gradient is represented by only three depth profiles near Bermuda [De La Rocha et al., 2000]. This poor spatial resolution hinders a robust analysis of the mechanisms leading to the observed distribution, and modeling studies have delivered conflicting results in this regard [Wischmeyer et al., 2003; Reynolds, 2009].
 In order to characterize the δ30Si systematics of the Atlantic Ocean, we have analyzed a suite of Atlantic seawater samples from Denmark Strait to Drake Passage. A companion paper (G. F. de Souza et al., Strong physical control on silicon stable isotope distribution in the Atlantic thermocline, manuscript in preparation, 2012) discusses the δ30Si distribution in the main thermocline, while this paper focuses on the Atlantic δ30Si distribution below 2000 m water depth. We show that the deep Atlantic exhibits a strong meridional δ30Si gradient, which we relate to the interaction of the global meridional overturning circulation (MOC) with biological Si cycling processes at high southern latitudes, thereby providing evidence for the importance of a Southern Ocean source of nutrients to the low-latitude thermocline. Furthermore, we show that the distribution of seawater δ30Si values robustly confirms earlier interpretations of quasi-conservative behavior of Si at depth in the Atlantic Ocean [Broecker et al., 1991].
2.1. Sample Collection
 A total of 84 seawater samples are included in this study (Figure 1 and Table 1). These include 25 samples originally collected for analysis of neodymium and hafnium isotopic composition [Rickli et al., 2009] at 7 stations in the eastern Atlantic Ocean during expedition ANT XXIII/1 of R/V Polarstern (October–November 2005). Two detailed depth profiles (17 samples each) from the tropical Brazil and Angola Basins were collected during expedition KN192-5 of R/V Knorr (November–December 2007). A further 23 samples were collected at 6 stations along GEOTRACES section GA02 in the North Atlantic, during expeditions 64PE318 and 64PE319 of R/V Pelagia (April–May 2010). In addition, two samples from Drake Passage (R/V Polarstern expedition ANT XXIII/3) were procured. All samples were filtered onboard using 0.45 μM nitrocellulose (Polarstern), 0.45 μM polycarbonate (Knorr) or 0.2 μM cellulose acetate (Pelagia) filters. Samples from Polarstern expeditions were acidified onboard with 0.1% v/v distilled 9.5 M HCl [Rickli et al., 2009], while samples from expeditions KN192-5, 64PE318 and 64PE319 were acidified with 0.1% v/v distilled 6 M HCl in the laboratory at least 12 hr before preconcentration of Si.
Table 1. Location of Expeditions and Stations From Which Samples Were Collected for This Studya
 We typically process 64 nmoles (1.8 μg) of Si for each mass-spectrometric analysis of Si stable isotope composition. Prior to the chromatographic separation of Si from sea salt, Si is preconcentrated by coprecipitation with brucite using a method modified from Karl and Tien . This preconcentration decreases the associated salt cation and anion matrix by more than two orders of magnitude, permitting the use of small ion exchange columns. Our two-step precipitation procedure (similar to that used by Reynolds et al. [2006a]) achieves Si yields generally in excess of 99% (and always >97.5%): in a first step, ∼10% of seawater Mg is precipitated as Mg(OH)2 by adding ∼1% v/v of 1 M NaOH solution (semiconductor grade; Sigma-Aldrich) to the pH-neutralized seawater sample. This solution is shaken for 1 hr and allowed to react and settle for 24 hr before being centrifuged. The second precipitation step is carried out by further addition of 1 M NaOH to precipitate ∼10% of the remaining sample Mg, followed by shaking for 1 hr, settling for at least 24 hr and centrifuging. The supernatant is removed and the yield of the coprecipitation determined by photospectrometric analysis of supernatant Si concentration using the molybdate blue method [Strickland and Parsons, 1968]. The precipitate is then dissolved in a small volume of 6 M HCl and diluted with ultrapure water (>18.2 MΩcm) to 64 μM (1.8 ppm) Si, at which point it contains ≤53 mM Mg (i.e. ≤0.11 meq Mg/ml) as its main cationic matrix at a pH of 2–3. One milliliter of this solution is passed through a cation-exchange column with a retention capacity of 1.7 meq (1 mL AG50W-X8; BioRad Laboratories) to separate uncharged H4SiO4 from the cationic matrix. The uncharged and anionic matrix, which consists primarily of Cl−, SO42− and PO43−, is not removed by this method, such that the final solution contains ∼1.5 mM Cl− (mainly from HCl), ∼1 μM seawater SO42−, and a PO43− concentration that depends on the seawater P:Si ratio (which ranges from ∼0.02–0.2). Anion doping tests have shown that the presence of these anions does not result in analytical artifacts; in fact, the insensitivity of our analytical setup to the presence of SO42− has been previously documented by Georg et al. , a finding that contrasts with the matrix effects observed by van den Boorn et al.  using a different analytical system, perhaps due to the different sample introduction systems used. We verified the accuracy of our seawater δ30Si analyses using the standard addition method applied to isotopic analyses by Tipper et al.  (see section A in Text S1 in the auxiliary material).
 The purified Si solution is analyzed for Si stable isotope composition using a high-resolution multicollector inductively coupled plasma mass spectrometer (NuPlasma 1700; Nu Instruments, UK) in static mode by standard–sample bracketing, with one δ30Si analysis consisting of 5 bracketed measurements (36 × 5 s integrations each) of the sample. Detailed descriptions of the chromatographic separation and mass-spectrometric methods used are given by Georg et al. .
 Silicon stable isotope composition is reported as the permil deviation from the standard reference material NBS28, δ30Si, which is defined as:
The long-term external reproducibility of our δ30Si analyses is ±0.12‰ (2σSD), as estimated from the variance of ∼300 analyses of the secondary isotopic standard Diatomite [Brzezinski et al., 2006] over >3 yr. The average δ30Si value of Diatomite measured over this period is +1.22 ± 0.01‰ (2σSEM), consistent with the inter-laboratory comparison study of Reynolds et al. . Except in the case of near-surface samples with low amounts of Si, reported sample δ30Si values are the mean of at least 3, and up to 11, complete replicate analyses in at least 2 separate analytical sessions. External errors on the seawater data, reported as 2 standard errors of the mean (2σSEM), are usually ±0.07‰ or better (see Table S1 in the auxiliary material). Error bars shown in all figures are external 2σSEM. In the following, the notations Si, [Si] and δ30Si will refer to dissolved silicon (i.e. silicic acid), its concentration, and its stable isotope composition respectively.
 Depth profiles of δ30Si (see examples in Figure 2a) show the typical increase in δ30Si values toward the surface that is expected from the preferential uptake of lighter Si isotopes by diatoms in the surface ocean. Values of δ30Si range from approximately +1.2‰ in bottom waters to almost +3‰ in the surface mixed layer (Table S1 in the auxiliary material; all depth profiles are illustrated in Figures S1–S4 in Text S1 of the auxiliary material). The gradient toward higher δ30Si values is largest in the upper ocean, above the salinity minimum at ∼1000 m associated with Antarctic Intermediate Water (AAIW) in the South Atlantic. Below this, δ30Si gradients are much smaller, but δ30Si values generally continue to decrease with depth, with the exception of the northernmost North Atlantic (Irminger Basin; 64PE319 Stas 3 and 5), where δ30Si values increase slightly with depth near the very bottom of the water column (Figure 2b), associated with stronger gradients in temperature and [Si] (Table S1 in the auxiliary material). Furthermore, the entire water column exhibits a clear difference in δ30Si values between the North and South Atlantic (Figure 2). At temperate northern latitudes around 45°N, the western Atlantic (64PE319 Sta 11) exhibits lower [Si] and higher δ30Si values than the eastern Atlantic (ANTXXIII/1 Stas 3 and 6; Figure S2).
 A remarkably strong coherence in the δ30Si systematics of the entire sampled Atlantic is illustrated by the relationship between concentration and isotopic composition of Si. Graphs of δ30Si values versus the reciprocal of silicon concentration, 1/[Si] (Figure 3), reveal that samples from all regions exhibit the same relationship between these parameters, even when their δ30Si depth profiles exhibit clear differences (Figure 2). In Figure 3a, the existence of two line ararrays with different slopes is apparent. These arrays intersect at a δ30Si value of +1.55–1.6‰ and 1/[Si] of 0.09–0.06, corresponding to a potential density σθ of 26.75–27.00 (water depth 300–400 m), i.e. in the range of densities of Subantarctic Mode Water (SAMW) in the Atlantic [Larqué et al., 1997; Sallée et al., 2010].
 In the following discussion, we describe and discuss the deep Atlantic δ30Si distribution below 2000 m water depth (section 4.1), and draw upon the δ30Si signature of the intermediate and upper Atlantic to explain its ultimate origin in the context of the MOC (sections 4.2–4.4).
4.1. Quasi-Conservativity of Si
 It is a striking feature of the deep Atlantic δ30Si distribution that the North Atlantic water column exhibits consistently higher δ30Si values than the South Atlantic (Figure 2 and Table S1 in the auxiliary material). In a plot of δ30Si values against the reciprocal of Si concentration, 1/[Si], this meridional gradient is expressed as a linear array for water samples below ∼2000 m (Figure 3b). The existence of such a linear relationship is indicative of conservative mixing between two distinct reservoirs, or end-members (see, e.g., Albarède  for a derivation). The deep Atlantic δ30Si systematics thus strongly suggests that the distribution of Si in the deep Atlantic is controlled by binary mixing, with no significant sources or sinks. This would imply that Si is quasi-conservative in the deep Atlantic, which might intuitively seem unlikely for a biologically cycled nutrient. As we show below, however, it can be demonstrated that this is indeed the case in the deep Atlantic Ocean.
Broecker et al. , and more recently Sarmiento et al. , have shown that while the effect of opal dissolution on deep Atlantic [Si] can be resolved by careful analysis, this effect is of minor importance (<5% contribution) in controlling deep [Si]. Similarly, Anderson and Sarmiento  find that “the remineralization signal in the deep Atlantic is practically non-existent”. In the case of Si, this is partially due to the low Atlantic opal productivity [Sarmiento et al., 2007], but the leading-order control is the order-of-magnitude disparity in [Si] between the two water masses filling the deep Atlantic, i.e. North Atlantic Deep Water (NADW, [Si] < 20 μM) and Antarctic Bottom Water (AABW, [Si] ∼ 120 μM). This pronounced [Si] contrast means that admixture of Si-rich AABW to Si-poor NADW produces large [Si] changes, such that mixing of these two water masses exerts by far the strongest control on variations in Atlantic [Si] below a water depth of ∼2000 m. Silicon concentrations are thus closely correlated with conservative and quasi-conservative tracers of water mass mixing such as salinity [Holfort and Siedler, 2001] and PO*4 [Broecker et al., 1991] (see also section B in Text S1 in the auxiliary material) in the deep Atlantic, with the exception of a few local anomalies [e.g., van Bennekom and Berger, 1984].
 The influence of the contrasting northern and southern water masses on the deep Atlantic δ30Si systematics is apparent in Figure 3b. Low δ30Si values are associated with the high Si concentrations of AABW, while the highest δ30Si values are observed for the precursors of NADW, i.e. the dense Si-poor overflows from the Nordic seas (Denmark Strait Overflow Water and Iceland-Scotland Overflow Water, DSOW and ISOW) as well as Labrador Seawater (LSW) [Dickson and Brown, 1994]. These northern and southern deepwater sources thus represent the two isotopic end-members inferred above to be controlling the Atlantic δ30Si distribution. By calculating the deviation of measured δ30Si values from those expected from pure mixing of these end-members, we can quantify the degree of conservativity of Si. Following Gruber , we utilize the quasi-conservative tracer PO*4 [Broecker et al., 1991] to independently calculate the contribution of North Atlantic waters fNA to each sample (see section B in Text S1 in the auxiliary material). The δ30Si value expected from pure binary mixing, , is then:
where NA and SO refer to the North Atlantic and Southern Ocean components respectively, and and [Si]i are the isotopic composition and concentration of Si in the component i. The deviation of the measured value from this value is given by Δ , defined as:
Values of Δ for the deep Atlantic are, with some notable exceptions, generally smaller than the associated uncertainties (Figure 4b and section B in Text S1 in the auxiliary material). This close correspondence of measured and calculated δ30Si values demonstrates that Si is predominantly quasi-conservative in the deep Atlantic. The sample most strongly offset from the 1:1 line in Figure 4a is a sample from the Bay of Biscay, whose chemistry (PO*4, δ30Si, or both) is influenced by Mediterranean Overflow Water [Gruber, 1998]. There is a slight tendency of measured δ30Si values in the South Atlantic to be marginally lower than calculated values (Figure 4). These offsets are associated with samples from the deep Angola basin, where seawater neodymium isotope compositions appear to be affected by the Congo Fan [Rickli et al., 2009]. The observed slight δ30Si deviations from pure mixing might thus indicate that Si release from the Congo Fan [van Bennekom and Berger, 1984; Ragueneau et al., 2009] plays a minor role in modulating deep δ30Si values here.
 The strongest control on the deep Atlantic δ30Si distribution, however, is the physical circulation: Antarctic Bottom Water advects a low δ30Si value of around +1.2‰ into the Atlantic from the south, while high δ30Si values of +1.7‰ to +1.85‰ are introduced into the North Atlantic by the NADW-precursors. Mixing between these water masses results in the observed small δ30Si gradients with depth at individual stations, but also in the large-scale meridional δ30Si gradient. The propagation of the high δ30Si signal from the North Atlantic also traces the path of deepwater circulation: the midlatitude western Atlantic water column, which is strongly influenced by the deep western boundary current transporting NADW southwards, displays both lower [Si] and higher δ30Si values than the eastern Atlantic at the same latitude (Figure S2 and Table S1).
 By documenting the quasi-conservativity of Si and emphasizing the dominant influence of circulation on the δ30Si distribution, our data robustly confirm minimal influence of opal dissolution on the Si distribution in the Atlantic [Broecker et al., 1991; Sarmiento et al., 2007], and indicate that processes such as boundary exchange, which has been suggested to result in a significant flux of Si into the global ocean [Jeandel et al., 2009], are insignificant in determining the distribution of Si in the deep Atlantic.
4.2. Origin of δ30Si Signatures
 The distinct δ30Si signatures of North Atlantic and Southern Ocean waters are at odds with their homogeneity in δ15N-NO3 [e.g., Sigman et al., 2009]. Such a difference between these two broadly comparable nutrient isotope systems is surprising, and must reflect a dissimilarity in the oceanic cycles of Si and N. In the following, we first elucidate how the North Atlantic and Southern Ocean water masses obtain their distinct δ30Si signatures, and the implications that this has for the oceanic Si cycle. Comparison with the δ15N-NO3 system then allows us to suggest likely causes of the difference between oceanic δ30Si and δ15N-NO3 distributions.
Toggweiler et al.  introduced a cartoon representation of the ocean circulation that provides an illustrative schematic framework for the following arguments. They separate the MOC into two domains, or “loops”, as shown in Figure 5: a southern loop dominated by AABW, and a northern loop dominated by NADW, fed by northward flow through the thermocline. These two domains interact to form Circumpolar Deep Water (CDW) that upwells in the Southern Ocean, feeding both circulation loops. The prolonged exposure of the northern loop to biological activity at the surface leads to a strong depletion in nutrients. Below, we explain the origin of deepwater δ30Si values in the context of this schematic, beginning with the southern loop.
 Antarctic Bottom Water forms in the high latitudes of the Southern Ocean, primarily in the Weddell Sea [Orsi et al., 1999], through a series of interactions of surface waters with shelf waters densified by brine rejection [Foster and Carmack, 1976; Huhn et al., 2008]. Due to a combination of high upwelling rates and micronutrient- and light-limitation of plankton growth [e.g., Chisholm and Morel, 1991], these waters all contain high levels of macronutrients. The waters brought to the surface Antarctic by upwelling of CDW thus experience little Si depletion before they sink and are incorporated into AABW (Figure 5), such that their nutrient properties remain similar to those of CDW. Furthermore, the sinking waters entrain significant volumes of surrounding water as they traverse the slope, and experience strong mixing in the region of the Antarctic Circumpolar Current, such that they exit the Southern Ocean containing a considerable admixture of CDW [e.g., Mantyla and Reid, 1983]. Considering this formation process, it is unsurprising that the δ30Si value of AABW is indistinguishable from that of CDW at about +1.2‰ [Cardinal et al., 2005; this study], a value that in turn must reflect the balance of silicon inputs to and outputs from the Southern Ocean.
 The North Atlantic water masses that contribute to NADW (i.e. DSOW, ISOW and LSW), on the other hand, are formed by the buoyancy loss of surface waters from lower latitudes that are transported into the high-latitude North Atlantic by the Gulf Stream/North Atlantic Current [e.g., Hansen and Østerhus, 2000; McCartney and Mauritzen, 2001; Brambilla et al., 2008], as represented by the northern loop of Figure 5. Deep convection in the subpolar North Atlantic leads to the formation of Subpolar Mode Water, including its densest type, Labrador Seawater [McCartney and Talley, 1982; Brambilla et al., 2008]. Similarly, the Nordic overflows DSOW and ISOW are fed by light-to-dense conversions of waters of shallow North Atlantic origin in the Norwegian Sea [Mauritzen, 1996; Hansen and Østerhus, 2000; Eldevik et al., 2009]. In both of these cases, the surface waters that experience buoyancy loss have very low [Si] [Garcia et al., 2010], such that, although they may have elevated δ30Si values as a result of biological uptake of Si, they have limited potential to influence the δ30Si value of a water mass formed by deep convection. This is because the Si budget – and thus δ30Si signature – of the homogenized water mass formed by convection is strongly weighted (∼95%) toward the entrained subsurface waters. In the case of both LSW and the Nordic overflows, these subsurface waters possess North Atlantic characteristics [McCartney and Talley, 1982; McCartney and Mauritzen, 2001]. Thus, in order to trace the origin of the high δ30Si value of NADW, we must understand what controls the Si budget, and thus δ30Si value, of the North Atlantic subsurface.
 Since the magnitude of diapycnal mixing in the main thermocline is small [Ledwell et al., 1993; Toggweiler and Samuels, 1993; Toole et al., 1994; Schmittner et al., 2009], it is likely that the nutrients transported by the northward flow in the Atlantic thermocline are dominantly sourced from the waters that introduce nutrients into the thermocline in the south, i.e. SAMW and AAIW [Tsuchiya, 1989; Schmitz and McCartney, 1993; Sarmiento et al., 2004; Williams et al., 2006; Palter and Lozier, 2008; Palter et al., 2010]. The importance of a southern source of nutrients to the thermocline is also indicated by the δ30Si systematics: the intersect between the upper- and deeper-ocean δ30Si–1/[Si] relationships occurs at potential densities corresponding to these water masses. The intermediate cross-equatorial transport into the North Atlantic takes place in the σθ interval 26.8–27.2 [Schmitz, 1995, and references therein], which closely corresponds to the density range of SAMW and AAIW in the South Atlantic (σθ = 26.8–27.3 [Larqué et al., 1997; Sallée et al., 2010]). The corresponding samples in our South Atlantic dataset have δ30Si values ranging from +1.45‰ to +1.6‰, with an average of +1.5‰ when weighted by [Si] in the water mass. The NADW complex being transported southwards at 45°N has a concentration-weighted average only slightly higher than this (+1.6‰). In the context of the large-scale circulation represented by Figure 5, this similarity suggests that the 0.4‰ offset in δ30Si values between NADW and AABW is largely the result of the δ30Si difference between the water masses of Southern Ocean origin, i.e. between AABW and SAMW/AAIW.
 We test this hypothesis by performing a sensitivity test in a simple box model of the ocean, as shown in Figure 6a. This 8-box model is a modification of the 7-box model of Reynolds  (based on Toggweiler ), altered to more clearly separate the two high-latitude end-members that are our focus of interest here. As illustrated in Figure 6a, the circulation simulated by this model bears a close resemblance to the schematic view of Toggweiler et al.  (Figure 5). The model produces a [Si] gradient between the Si-rich deepwaters of the “Southern” Ocean (box ds) and the Si-poor North Atlantic deepwaters (box dn; Figure 6b). Although it overestimates [Si] in the surface Southern Ocean (boxes p and s), the model successfully captures the latitudinal gradients in surface [Si] and δ30Si values that are a key feature of the Southern Ocean Si distribution [Varela et al., 2004; Sarmiento et al., 2004; Fripiat et al., 2011]. The model also produces a large-scale δ30Si gradient between North Atlantic and Southern Ocean deepwaters (Figure 6b). Sensitivity analyses by Reynolds  have shown that this deepwater δ30Si gradient is a robust model feature whose presence is not dependent on the choice of parameter values (see also section D in Text S1 in the auxiliary material). In order to identify the ultimate source of this gradient, we perform a simple sensitivity test: by “switching off” isotope fractionation during utilization in the surface Southern Ocean, the model is prevented from creating an isotopic difference between intermediate and deep Southern Ocean waters. This approach has the advantage of leaving the elemental distribution of Si unaffected, which considerably simplifies the interpretation of model results. In this test configuration (Figure 6c), the δ30Si value of the North Atlantic is essentially identical to that of the deep Southern Ocean, despite isotope fractionation in the low- and high northern latitude surface ocean and a deep [Si]-gradient identical to that of the standard configuration. The presence of a deep δ30Si gradient in the model is thus dependent upon the introduction of a high-δ30Si signal into intermediate waters from the surface Southern Ocean. This model result thus supports the inference that the transport of Si with a high δ30Si value into the North Atlantic by the upper return path of the MOC must play an important role in producing the δ30Si difference between NADW and AABW. Silicon uptake and isotope fractionation by diatoms in the surface Southern Ocean thus effectively fractionate Si isotopes between the two MOC loops of Figure 5.
4.4. Si and N Cycles
 If the contrast in the δ30Si signatures of NADW and AABW is due in large part to a δ30Si difference between shallow and deep Southern Ocean waters, we should address (a) what in turn causes this difference, and (b) why this process does not result in a similar contrast in δ15N-NO3. A comparison of Southern Ocean δ30Si and δ15N–NO3 data is highly instructive in this regard (Figures 7 and 8) [Sigman et al., 2000; Cardinal et al., 2005; DiFiore et al., 2006].
Figure 7 shows δ30Si and δ15N–NO3 data from two Southern Ocean transects south of Australia [Sigman et al., 2000; Cardinal et al., 2005]. The δ30Si depth profiles in Figure 7b highlight the fact that elevated δ30Si values in the Subantarctic subsurface are associated with the salinity minimum of AAIW and shallower Subantarctic thermocline waters. Such a relationship is not observed to the same extent for δ15N–NO3 (Figure 7a), for which the isotopic signature of AAIW is not strongly different from that of the deep Southern Ocean. These observations clearly show that the process of AAIW formation influences its δ30Si and δ15N–NO3 signatures differently. Below, we explore the likely reasons for this contrasting behavior.
 Mass balance calculations by Fripiat et al.  indicate that the dominant contributor (97%) to the Si inventory of AAIW is Antarctic Surface Water (AASW) from the Polar Front (PF). Antarctic Surface Water is present from the PF southwards, and bears the properties of the Antarctic winter mixed layer [e.g., Rintoul and Bullister, 1999]. Within AASW, Si concentrations decrease strongly (∼30 μM) northward (i.e. toward the PF), while NO3 concentrations display only a modest decrease (∼3 μM [e.g., Garcia et al., 2010]). This markedly different behavior is likely due to a combination of the high silicification of Southern Ocean diatoms [e.g., Takeda, 1998] and the more efficient recycling of NO3 than Si in the surface and shallow subsurface [Dugdale et al., 1995; Brzezinski et al., 2003]. Importantly, the [Si] decrease in AASW toward the north is accompanied by increasing δ30Si values, while δ15N–NO3 values remain essentially constant [Sigman et al., 2000; Fripiat et al., 2011]. Thus at the Polar Front, AASW possesses an elevated δ30Si value of around +1.8‰ [Fripiat et al., 2011], but a δ15N–NO3 value that is virtually indistinguishable from that of the deep Southern Ocean (+5.6‰ [Sigman et al., 2000]). These isotope signatures are imparted to AAIW with virtually no alteration [Sigman et al., 2000; Fripiat et al., 2011]. Thus, the strong utilization of Si by diatoms south of the PF results in an elevated δ30Si signature in AASW that is incorporated into AAIW, while the modest drawdown of NO3 leads to a AAIW δ15N–NO3 signature that is very similar to that of the deep Southern Ocean.
 At levels shallower than AAIW, the δ15N–NO3 distribution in the Subantarctic subsurface may be further affected by complexities of the δ15N–NO3 system. Sigman et al.  found that δ15N-NO3 values in the Subantarctic thermocline were too low to be explained by mixing in the Subantarctic alone (Figure 8a). They suggested that this feature might be due to the influence of subtropical waters whose isotopic composition has been affected either by nitrogen fixation or by mixing with very N-depleted low-latitude surface waters. The same mixing calculation for the δ30Si data of Cardinal et al.  reveals that, unlike δ15N-NO3, δ30Si values in the Subantarctic thermocline are consistent with mixing between Antarctic and Subantarctic waters (Figure 8b), as also concluded by Fripiat et al. . Since mixing with low-nutrient subtropical surface waters should affect both δ30Si and δ15N-NO3 values in the Subantarctic thermocline to a similar degree, the discrepancy between the systematics of these two systems (Figure 8) suggests that the low δ15N-NO3 values in the Subantarctic thermocline result rather from the influence of low-latitude nitrogen fixation.
 The arguments above suggest a cause for the difference between the oceanic δ30Si and δ15N-NO3 distribution at the basin scale. First, the different degrees to which Si and NO3 are drawn down by biological production in the Antarctic Zone of the Southern Ocean, likely driven by the more efficient remineralization of NO3 in the shallow subsurface, results in the formation of AAIW that is markedly different in δ30Si, but not in δ15N–NO3, from the deep Southern Ocean. Additionally, δ15N–NO3 values in the Subantarctic thermocline appear to be depressed even at high latitudes due to the influence of regions of nitrogen fixation, a signal that is likely reinforced by “downstream” nitrogen fixation at low northern latitudes [Knapp et al., 2008] that further reduces the δ15N-NO3 of the Atlantic thermocline. No such complexity influences the Si isotope system, such that southern mode and intermediate waters not only (a) form with a higher δ30Si value than the deep Southern Ocean but also (b) can impart this high value to the entire Atlantic thermocline. The MOC-driven nutrient transport into the North Atlantic – the northern loop of Figure 5 – is thus significantly different in δ30Si, but not in δ15N-NO3, from the deep Southern Ocean. The δ30Si and δ15N-NO3 signatures of NADW thus ultimately reflect processes related to the interaction of biology and the physical circulation in the Southern Ocean (nutrient utilization, remineralization, water mass subduction) and the global low-latitude upper ocean (nitrogen fixation, thermocline ventilation, MOC closure).
5. Global δ30Si Distribution: Perspectives
 Although the focus of this paper is the biogeochemistry of Si in the Atlantic Ocean, it would be amiss not to briefly discuss this first comprehensive Atlantic δ30Si dataset in the global context. In their “first look” at seawater δ30Si values, De La Rocha et al.  argued for a diatom dissolution flux into deepwaters with, on average, a lower δ30Si value than seawater. Beucher et al. , in contrast, utilized a compilation of Southern Ocean and Pacific data to argue for a δ30Si value of the dissolution flux that is higher than that of deepwaters. This study has highlighted the important role, also recognized by De La Rocha et al. , that physical processes – i.e. deepwater formation and nutrient transport by the ocean circulation – play in setting deep ocean δ30Si values. We would thus argue that global deepwater δ30Si gradients cannot be interpreted quite as simply as either of these previous contributions has done. Rather, interpretation of the oceanic δ30Si distribution must consider the influence of both biological cycling and ocean circulation – that is, the remineralized and preformed contributions to the observed δ30Si distribution.
 Such analyses may be performed in the quantitative framework of global circulation models (GCMs), provided that these models conform to the observational constraint of a strong deepwater δ30Si gradient, which is not the case for the only published GCM study [Wischmeyer et al., 2003]. Reynolds  did reproduce this gradient with box models, but could not adequately explain the disagreement with GCM results. In fact, several plausible reasons exist for the inability of Wischmeyer et al.'s  model to produce a deepwater δ30Si gradient. Dutay et al.  have demonstrated that the physical model employed by Wischmeyer et al.  strongly underestimates the ventilation of intermediate waters, which our analysis would indicate are vital for the production of a large-scale deepwater δ30Si gradient. The poor intermediate water ventilation in Wischmeyer et al.'s  model may be due to its representation of lateral mixing [Maier-Reimer et al., 1993], which does not parameterize the advective effects of mesoscale eddies [Gent and McWilliams, 1990]. This eddy-induced transport plays an important role in the subduction of water masses in the Southern Ocean [Marshall, 1997], and its correct parameterization has been shown to improve GCM representation of the oceanic Si cycle [Gnanadesikan, 1999a; Gnanadesikan and Toggweiler, 1999; Dunne et al., 2007]. The coupled biological-physical model of Wischmeyer et al.  displays its strongest deviation from observational [Si] data in the surface Southern Ocean, suggesting that Si cycling in this region is inaccurately captured. Furthermore, their physical model employed an upwind tracer advection scheme [Maier-Reimer et al., 1993] that results in large numerical diffusion [e.g., Gerya, 2010]. This implicit strong mixing may inhibit large isotopic gradients at depth, as well as diluting any high-δ30Si signal in intermediate waters by strong vertical mixing with the deep. The combination of these dynamical and biogeochemical weaknesses may have prevented Wischmeyer et al.'s  model from forming southern intermediate and mode waters with the correct biogeochemical properties, such that it did not accurately represent the northward MOC-driven nutrient transport. Future modeling efforts should focus on reproducing the large-scale, first-order observational constraints that we have documented here.
6. Summary and Conclusions
 Our high-precision δ30Si dataset for the Atlantic Ocean demonstrates the strength of seawater δ30Si as a tracer of the biogeochemical cycling of Si in the sea. The stable isotope composition of dissolved silicon traces mixing between water masses of North Atlantic and Southern Ocean origin, which implies that Si has no significant sources or sinks at depth in the Atlantic Ocean. The contrasting δ30Si values of the North Atlantic and the deep Southern Ocean are, in turn, most likely the result of the fractionation signature introduced into the thermocline by the interaction of biological Si uptake with mode- and intermediate-water mass formation at high southern latitudes. Biological activity in the surface Southern Ocean thus imparts distinct δ30Si values to the two “loops” of Toggweiler et al.'s  MOC representation, with the large-scale interaction of these loops producing the basin-scale Atlantic δ30Si distribution.
 The Si isotope data presented here thus provide strong evidence for a regime of nutrient transport in the ocean that is closely coupled to lateral transports related to the MOC. In combination with our box-modeling approach, the δ30Si data robustly and independently corroborate previous inferences [Sarmiento et al., 2004] of the importance of the Southern Ocean's biogeochemical divide [Marinov et al., 2006] in determining global oceanic nutrient distributions, and thus the distribution and magnitude of oceanic primary productivity. Furthermore, the pathways of large-scale tracer transport in the ocean are directly relevant to the fundamental mechanisms driving the MOC [Gnanadesikan, 1999b]. The insights provided by seawater δ30Si data thus reach far beyond the oceanic Si cycle and touch upon fundamental questions in modern oceanography.
 The authors would like to thank the captain and crew of R/Vs Polarstern, Knorr and Pelagia as well as M. Rutgers van der Loeff as Chief Scientist on cruise ANT XXIII/1. P. Laan and J. de Jong are gratefully acknowledged for sampling on cruise 64PE319. Feedback from Nicolas Gruber helped to considerably improve an earlier version of this manuscript. Two anonymous reviewers are thanked for their constructive input. This work was supported by Swiss National Science Foundation grants 200021-116473 and 200020-130361.