Samples were collected aboard the R/V Pourquoi Pas? during GEOVIDE. The cruise began on 15 May 2014 in Lisbon, Portugal, headed north towards Greenland, and then traversed south-west to St. John's, Canada, arriving on 30 June 2014 (Fig. 1). Seawater samples for δ30SiDSi analysis were collected using Niskin bottles attached to a standard rosette conductivity–temperature–depth (CTD) unit from 10 locations in the North Atlantic Ocean and Labrador Sea (Fig. 1 and Table 1). Samples were filtered through polycarbonate 0.45 µm filters (Millipore) and stored in acid-cleaned, low-density polyethylene bottles at room temperature. The DSi concentration of all samples collected (n=56) was determined via molybdate blue spectrophotometry (Strickland and Parsons, 1972). For the measurement of δ30SiDSi, DSi was extracted from the seawater by precipitating it as trimethylamine silicomolybdate, which was subsequently combusted to form SiO2 (De La Rocha et al., 1996). This SiO2 was dissolved at room temperature in polypropylene microcentrifuge tubes (1.5 mL) containing 23 M HF (Suprapur) to yield a final solution concentration of 0.23 M Si (e.g., 4 µmol of SiO2 dissolved in 17.4 µL of HF).

Every sample was further purified using ion exchange chromatography following Engström et al. (2006). Briefly, 17.4 µL of a 0.23 M Si solution was diluted in 7.7 mL of Ultra Hiqh Quality water (UHQ H2O; 18.2 MΩ-cm; Millipore Direct-Q) and loaded onto columns containing AG1-X8 resin (100–200 mesh size, BioRad) that had been preconditioned with 2 M NaOH. The sample matrix was eluted using 95 mM HCl + 23 mM HF followed by the elution of the purified Si using 0.14 M HNO3,+ 5.6 mM HF. All acids were Suprapur (Merck) and were diluted with UHQ H2O.

Silicon stable isotope composition (δ30Si) of the purified samples was measured using standard-sample bracketing combined with external normalization by doping the samples with magnesium (Mg) (e.g., Cardinal et al., 2003) on a Thermo Scientific Neptune multi-collector inductively coupled plasma mass spectrometer (MC–ICP-MS) at the Unité Géosciences Marines (Ifremer, Plouzané). Prior to the isotopic analysis, the purified samples were diluted with 0.16 M HNO3 (1 % HNO3) to 1 ppm Si, yielding a roughly 12 V signal on mass 28 at medium resolution (see Table 2 for additional information on operating conditions). All samples and standards (NBS28 and a 99.995 % pure silica sand (Alfa Aesar) used as a working standard) were passed through column chemistry and matrix-matched to give the same signal strength (within 10 %) and to contain the same amount of HF (generally 1 mM). Magnesium (1000 ppm, NIST SRM) was added to the samples and standards just prior to measurement at a final concentration of 0.1 ppm (Cardinal et al., 2003; Abraham et al., 2008). Si solutions were introduced into the plasma via an Apex desolvating system equipped with a PFA nebulizer (uptake rate = 100 µL min-1) without additional gas.

For each measurement, beam intensities at masses 25 and 26 (Mg) and 28, 29, and 30 (Si) were monitored in dynamic mode (i.e., switching between Si and Mg masses) for one block of 25 cycles of 8 s integrations. Five minutes of rinse with 2 % HNO3 followed each sample and each standard solution. Solutions were analyzed in medium-resolution mode (m/Δm>6000).

Using a standard-sample-standard bracketing technique, δ30Si values for the samples were expressed as follows: δ30Si(‰)=(30Si/28Si)/((30Si/28Si)standard)-1×1000. The Si isotope ratios in Eq. (1) above (30Si / 28Si and 29Si / 28Si) were corrected for mass bias by external normalization using Mg doping. For example, the corrected 30Si to 28Si ratio (30Si / 28Si)corr is (30Si/28Si)corr=(30Si/28Si)meas×(30SiAM/28SiAM)εMg, where (30Si / 28Si)meas is the measured ratio, and 30SiAM and 28SiAM are the atomic masses of 30Si and 28Si. εMg is calculated from the beam intensities on masses 25 and 26:

εMg=ln⁡(25MgA/26MgA)/(25Mg/26Mg)meas/25MgAM/26MgAM, where 25MgA / 26MgA is the expected ratio of the natural abundances of the isotopes, (25Mg / 26Mg)meas is the measured ratio, and 25MgAM and 26MgAM are the atomic masses of 25Mg and 26Mg respectively. Each measurement of a sample fell between two measurements of the standard, and each sample was measured three times. This total of three sample measurements and five standard measurements was repeated 2–3 times in each mass spectrometry session and used to calculate one replicate value of δ30Si and δ29Si. As discussed below, full chemistry replicates were routine for each sample (see Table S2 in the Supplement).

Interference-free measurement was ensured by checking that δ29Si and δ30Si for all samples was consistent with the mass-dependent fractionation line (δ29SiNBS28=0.52×δ30SiNBS28). The signal was optimized to reduce the 14N16O interference on m/z 30 to below 0.5 % of the 30Si peak. Measurements were performed on the low-mass side of the peak where interference is minimal. Blanks were maintained below 1 % of the main signal and were subtracted for each sample and standard. Long-term reproducibility and accuracy on δ30Si values of the analytical procedure were assessed using the standard deviation of 54 analyses of NBS28 and 29 analyses of a secondary reference standard (Silicon (IV) oxide, Alfa Aesar) generated over 6 years (±0.10 ‰, 2σ). Reproducibility of the full chemical and analytical procedure was estimated using at least one replicate of each sample (chemical preparation plus isotopic measurements) and average reproducibility on replicate δ30Si was ±0.10 ‰ (2σ). Measurements of Big Batch (n=3) produced an average value of -10.48±0.34 (2σ), well within the range of intercalibration values reported by Reynolds et al. (2007). Measurement of the US GEOTRACES intercalibration reference seawater standard from the Aloha Station (1000 m) gave a δ30SiDSi value of +1.16±0.16 ‰ (2σ, n=3), within the range of intercalibration values (1.24±0.20 ‰; Grasse et al., 2017). Measurement of the Canada/GEOVIDE GEOTRACES intercalibration samples, where duplicate samples at three depths and two stations were analyzed by two different laboratories (Ifremer Plouzané and ETH Zurich), conforming to the GEOTRACES intercalibration protocol for a cruise without a cross-over station, gave similar δ30SiDSi values (see Table 3). Note that ETH Zurich uses a different purification method (cation exchange resin; see de Souza et al., 2012b) and MC-ICPMS instrument (Nu Plasma 1700) than Ifremer Plouzané (described in Sect. 2.2). The methods for each laboratory that participated in the Canada/GEOVIDE GEOTRACES intercalibration study are also presented in the US GEOTRACES intercalibration study (Grasse et al., 2017).

In order to accurately examine the relationship between the distribution of δ30SiDSi and water masses, the results of the optimum multiparameter (OMP) analysis of García-Ibáñez et al. (2018) were used to identify the mixture of water masses present within each sample and their contribution to the DSi budget.

The upper layers of the GEOVIDE section were represented by the North Atlantic Central Waters (NACW), transported by the North Atlantic Current (NAC; Pollard et al., 1996), and Subpolar Mode Waters (SPMW), the end-product of the transformation of NACW through air-sea interaction (McCartney and Talley, 1982; Tsuchiya et al., 1992). To account for the change in the temperature of SPMW along the path of the NAC as the result of air-sea interaction, two SPMWs were differentiated: IcSPMW (Iceland-Subpolar Mode Water) and IrSPMW (Irminger-Subpolar Mode Water). The intermediate layers of the section were represented by LSW, Mediterranean Water (MW), Subarctic Intermediate Water (SAIW), and Polar Intermediate Water (PIW). LSW is the last stage of the transformation of SPMWs and forms in the Labrador and Irminger seas (e.g., Pickart et al., 2003; de Jong and de Steur, 2016; Fröb et al., 2016). MW enters the North Atlantic from the Mediterranean Sea through the Strait of Gibraltar (Ambar and Howe, 1979; Baringer and Price, 1997). SAIW originates in the Labrador Current by mixing of the NAC waters with LSW (Iselin, 1936; Arhan, 1990; Read, 2000). The deep layers of the section were represented by Denmark Strait Overflow Water (DSOW), Iceland–Scotland Overflow Water (ISOW), North East Atlantic Deep Water (NEADW) and Lower Deep Water (LDW). Overflow waters (DSOW and ISOW) form after the deep waters of the Nordic Seas flow over the Greenland–Iceland–Scotland sills and entrain Atlantic waters (van Aken and de Boer, 1995; Read, 2000; Dickson et al., 2002; Fogelqvist et al., 2003; Yashayaev and Dickson, 2008). NEADW is formed as a result of entrainment events that occur along the journey of ISOW through the Iceland Basin (van Aken, 2000). NEADW recirculates in the West European Basin and mixes with the surrounding waters, including the Antarctic Bottom Water (AABW) (van Aken and Becker, 1996), resulting in the formation of LDW.

Dissolved silicon (DSi) concentrations were measured throughout the water column (Fig. 2a, Table S1), but only the samples collected below 500 m are discussed since no δ30SiDSi samples were collected from within the upper water column. DSi concentrations below 500 m along the GEOVIDE section ranged from 7 to 47 µM (Fig. 2a). The stations located to the east of the Mid-Atlantic Ridge (MAR; STN 01, STN 13, STN 21, STN 26, and STN 32) show DSi increasing in concentration from <10 to 20–50 µM below about 2000 m (Fig. 2a). Stations located to the west of the MAR (STN 44, STN 60, STN 64, STN 69, and STN 70) show only slight increases in DSi concentration with depth, with most of the values falling between 9 and 12 µM (Fig. 2a). This difference relates to the distribution of water masses in the northern North Atlantic, with the predominance of the most egregiously Si-poor northern-sourced water masses (LSW, ISOW/NEADW, DSOW) predominating in the western Atlantic while abyssal layers in the eastern Atlantic have had more of a contribution from Si-rich southern-sourced waters (LDW) (see Sect. 4.1). A clear pattern in the DSi concentration throughout the water column is that the eastern profiles exhibit a more typically “nutrient-like” profile than in the western profiles (Table S1; Fig. 2a).

All GEOVIDE water column profiles have relatively high δ30SiDSi (+1.5 ‰ to +3 ‰) between 500–1000 m, and show a trend towards lower δ30SiDSi values with depth (although none significantly lower than +1 ‰) (Fig. 2b). Strikingly, the lowest values of δ30SiDSi occur at stations nearer to the Iberian margin (e.g., STN 01, STN 13: 0.95 ‰ and 0.98 ‰, respectively), while the highest values tend to occur at the upper depths of the profiles (500–1000 m) at stations north of 50∘ N and west of 20∘ W (stations 26, 32, 44, 60, 64, 69, and 77; Fig. 2b), mirroring the differences in DSi between these locations (Fig. 2a, Tables S1 and S2). Profiles of δ30SiDSi in the Labrador Sea (stations 64, 69, and 77) show high values (above +1.5 ‰) extending below 2000 m water depth. These high values reach the bottom within the central Labrador Sea (STN 69). Elevated δ30SiDSi values at depths between 1000 and 2000 m can be found at all stations in the Irminger Basin and extend eastwards into the western portion of the West European Basin (up to STN 26), but the mid-depths of the far eastern Atlantic are marked by lower δ30SiDSi values around +1 ‰. Near-surface water δ30SiDSi could not be evaluated due to the very low dissolved silicon (DSi) concentrations (<5 µM).

The deep Atlantic Ocean below 1000 m exhibits a wide variation in DSi concentrations, ranging from ∼10 µM in the mid-depths of the subpolar North Atlantic Ocean to ∼120 µM in the abyssal southern Atlantic. Since at least the work of Broecker and Takahashi (1980), it has been known that this variation is primarily brought about by the quasi-conservative mixing of DSi between Si-rich abyssal waters derived from the Southern Ocean and Si-poor waters of North Atlantic origin. The analysis of Sarmiento et al. (2007), which takes the effects of water mass mixing into account, has shown that the effect of opal dissolution on deep Atlantic DSi is resolvable, but plays a near-negligible role in controlling the deep DSi distribution.

The first systematic study of the Atlantic δ30SiDSi distribution (de Souza et al., 2012b) showed that the quasi-conservative behaviour of DSi is clearly reflected by the δ30SiDSi of the deep Atlantic Ocean. Surveying deep water over a wide range of latitudes within the Atlantic Ocean, they found that values of δ30SiDSi vary coherently from high values (>2.0 ‰) in the Si-poor waters that contribute to NADW to low values (1.2 ‰) in the Si-rich Southern Ocean deep waters (Fig. 3). The more recent work of Brzezinski and Jones (2015) found near-identical behaviour within Atlantic deep waters along a near-zonal transect across the subtropical North Atlantic (Fig. 3).

Our data agree with the systematics of these two studies, with each uncorrected dataset exhibiting similar linear regressions, except for the value of the y intercepts, and the slope of the current data being slightly exaggerated relative to the other datasets (see Fig. 3). Although the previously published data for the North Atlantic Ocean have nearly identical linear regressions, an offset of +0.11 ‰ relative to the de Souza et al. (2012b) data was observed and discussed by Brzezinski and Jones (2015), concluding that an analytical bias existed. Such offsets of order ±0.2 ‰ have been recognized to exist between seawater δ30Si data produced in different laboratories (Grasse et al., 2017); their origin remains unclear, although they may have to do with differences in sample processing and chemical purification. The offset of the new data to that of de Souza et al. (2012b), produced at ETH Zurich, is somewhat surprising given the good agreement in δ30SiDSi for six seawater samples analyzed both at Plouzané and Zurich, but a small offset to lower δ30SiDSi at Plouzané is consistent with the offset (0.1 ‰) in these two laboratories' mean δ30SiDSi values for the seawater reference Aloha-1000 (Grasse et al., 2017). Whilst not ideal for the determination of the absolute δ30SiDSi value for each basin, the existence of such interlaboratory offsets does not impair our ability to analyze the distribution of δ30SiDSi along the GEOVIDE transect, with the systematics of our data exhibiting similar behaviour to previously published studies (Fig. 3).

Factoring out the offset in absolute δ30SiDSi values and the greater variability in our data (see Fig. 3 for details), it is interesting to note that our deep North Atlantic samples exhibit essentially the same δ30SiDSi range (∼0.6 ‰; Fig. 3) as that observed over the entire latitudinal range of the Atlantic Ocean (de Souza et al., 2012b). Our dataset thus indicates that DSi in the North Atlantic Ocean is an important source of isotopic variability in the deep ocean. This is at least partially due to the transport of isotopically heavy DSi to the North Atlantic by northward-flowing Subantarctic Mode Water/Antarctic Intermediate Water (de Souza et a., 2012b, 2015) and its incorporation into NADW, e.g., during the formation of LSW. However, Brzezinski and Jones (2015) hypothesized that the Arctic Ocean may also play an important role in producing deep North Atlantic δ30SiDSi variability, via overflows across the Greenland–Iceland–Scotland ridge (i.e., DSOW and ISOW), which dominantly contribute to Lower NADW. It thus remains to be understood how the isotopic compositions of various precursors of NADW contribute to its isotopic signal. In the following, we discuss our δ30SiDSi dataset in the context of regional oceanography, in order to study the control of interacting interior water masses on the δ30SiDSi distribution of the high North Atlantic.

The GEOVIDE section intersects numerous water masses of various origins whose presence is reflected in the distributions of salinity, dissolved oxygen (O2), and potential vorticity (PV) along the section (Fig. 4a, b, c; dissolved O2 is presented as percent saturation, i.e., O2/O2sat×100, where O2sat is the saturation O2 concentration). Since δ30SiDSi was only measured at depths below 500 m, we focus on the intermediate and deep ocean water masses. In discussing the relationship between the δ30SiDSi distribution and water mass structure, we initially focus on the westernmost and easternmost sections of the GEOVIDE transect, where water masses are in their most unadulterated form along the transect, prior to discussing their extension into the mid-Atlantic.

Starting with the westernmost profiles, those of stations 77, 69, and 64 in the Labrador Sea, we see relatively well-oxygenated waters with low PV (Fig. 4c) extending to depths below 2000 m, reflecting the presence of LSW. This water mass, which contributes to NADW, is formed by deep convection in the Labrador and Irminger seas (e.g., de Jong and de Steur, 2016) and spreads across the North Atlantic at intermediate to middle depths (see Fig. 1). Two distinct types of LSW can be distinguished and are indeed visible in our profiles. There is an extremely low-PV (<4×10-12 m-1 s-1) and well-oxygenated (>90 % saturation) pycnostad extending from around 400 to 1200 m in the Labrador and Irminger basins, and a saltier, less well-oxygenated water mass observed from roughly 1500 to 2300 m. These two water masses have been called Upper and Lower LSW, respectively (e.g., Kieke et al., 2007), and reflect variability in the severity of heat loss and depth of convection in the Labrador Sea (Yashayaev et al., 2003, 2007) most likely associated with differences in atmospheric forcing during different phases of the North Atlantic Oscillation (NAO; Dickson et al., 1996; Lazier et al., 2002).

At the three stations in the Labrador Sea, Upper LSW has a δ30SiDSi of around +2 ‰ and a DSi concentration of <10 µM (Fig. 3). Lower LSW has slightly lower δ30SiDSi values (around +1.5 ‰) and slightly higher DSi concentrations (∼10–15 µM). These differences could be due to a slightly greater proportion of regenerated silica in these deeper layers, less frequently and intensively penetrated by deep convection, or to differences in the preformed properties of Upper and Lower LSW, a result of convection to greater depths during the formation of Lower LSW.

Below LSW, the central Labrador Sea (STN 69) exhibits an exemplary “stacking” of the water masses contributing to NADW. Specifically, an increase in salinity at 3000 m points to the presence of NEADW, a modified version of the eastern Atlantic overflow water mass ISOW (van Aken and de Boer, 1995) that has crossed into the western Atlantic at the Charlie-Gibbs Fracture Zone (van Aken, 2000). At this point, NEADW has a δ30SiDSi of +1.5 ‰ and a DSi concentration of about 15 µM, values essentially equal to that of Lower LSW since ISOW has entrained LSW during its journey from the Iceland–Scotland sill. At the very base of the water column, beginning at about 3500 m water depth, a decrease in salinity and increase in O2 saturation point to the presence of DSOW, which flows from the Arctic Ocean into the North Atlantic as a bottom-hugging overflow off the eastern coast of Greenland, and represents the densest water contributing to NADW (Dickson and Brown, 1994). Our δ30SiDSi sample is situated within the transition between NEADW and DSOW, with slightly lower DSi than in NEADW, but no distinguishable difference in terms of its δ30SiDSi value of around +1.5 ‰.

Moving eastwards along the GEOVIDE transect brings us next to STN 60, on the southeastern coast of Greenland. This relatively shallow station has sampled IrSPMW at its upper two depths, which have high δ30SiDSi values (2.8 ‰ at 1000 m and 1.8 ‰ at 1400 m) and relatively low DSi concentrations (around 10 µM), reflective of the mixing of nutrient-depleted surface waters into this water mass. Subpolar Mode Waters can be seen as precursors to LSW, as they form pycnostads of progressively greater density from east to west in the subarctic gyre, preconditioning the upper water column for deep-reaching Labrador Sea convection by producing a relatively unstratified water column (Brambilla and Talley, 2008). Their more elevated δ30SiDSi values than that of LSW imply that the entrainment of DSi during deep convection associated with LSW formation plays an important role in setting the final δ30SiDSi signature of this water mass. The deepest depth sampled at STN 60 (1800 m), on the other hand, is probably LSW, with its low PV, high O2, and δ30SiDSi around +1.4 ‰, similar to the value observed in the central Labrador Sea.

Moving all the way across the Atlantic to the easternmost portion of the GEOVIDE transect allows us to focus in on two more important interior water masses. One of the most striking features in the distributions of salinity and O2 is the tongue of salty, O2-poor water extending westward from the Iberian margin at about 1000 m water depth. This is predominantly MW that has entered the Atlantic through the Strait of Gibraltar (Iorga and Lozier, 1999). The one sample we have of predominantly MW is at 1000 m depth at STN 01, just off of the Iberian Peninsula, with a δ30SiDSi of +1.2 ‰ and a DSi concentration of about 10 µM. The δ30SiDSi value of +1.2 ‰ corresponds well to the value of +1.3 ‰ measured by Coffineau (2013) in samples from closer to the point of origin of this water mass.

The other water mass sampled for δ30SiDSi in the eastern Atlantic Ocean is an O2-poor, low-PV, Si-rich abyssal water mass that is present below about 3000 m. This is LDW (McCartney, 1992), which derives from northward-flowing AABW that has entered the eastern Atlantic Ocean via the Vema Fracture Zone (Mantyla and Reid, 1983; McCartney et al., 1991). The OMP results of García-Ibáñez et al. (2018) shown in Fig. 5b nicely illustrate that LDW (which they denote as NEADWL) is the dominant contributor to the DSi inventory of the deep eastern Atlantic. This is a direct result of LDW being rich in DSi (25–45 µM) when compared to the other water masses in the North Atlantic Ocean.

The influence of this Si-rich southern-sourced water mass is also clearly seen in the δ30SiDSi distribution (Fig. 2b): δ30SiDSi values below 3000 m in the far-eastern Atlantic (stations 01, 13 and 21) range from +0.95 ‰ to +1.3 ‰, significantly lower than values at similar depths in the western Atlantic, which is dominated by northern-sourced water masses. The low δ30SiDSi values we observe for LDW compare very well with the value of +1.2 ‰ observed in AABW in the South Atlantic (de Souza et al., 2012b), indicating that the Si richness of this water mass makes its δ30SiDSi value insensitive to mixing or opal dissolution as it flows northwards in the abyssal Atlantic Ocean. Interestingly, four of our deep samples from stations 1 and 13 have low δ30SiDSi, perhaps indicative of another unknown source of isotopically light DSi (see Fig. 3). However, this still needs to be confirmed.

The remaining three stations in the mid-Atlantic (22–38∘ W; stations 26, 32 and 44) are influenced by varying combinations of the water masses of northern and southern origin that were discussed above. The easternmost of these three stations, STN 26 at the edge of the Porcupine Abyssal Plain, provides an exemplary illustration of the interaction of these water masses. At a depth of 500 m, the δ30SiDSi value (+2.85 ‰) is our most elevated isotopic composition, which may be influenced by the SAIW, but this is difficult to argue since this sample site is the only partial sample from this water mass, At depths of 1400 and 2000 m, the water column at this station is dominated by Lower LSW, as reflected by the PV, O2 and salinity distributions (Fig. 4a, b, c). As in the Labrador Sea itself, Lower LSW bears an elevated δ30SiDSi value, here about +1.7 ‰, and a relatively low DSi concentration of ∼15 µM. This Si-poor water mass is underlain, at the very bottom of the profile (3500 m) by the Si-rich southern-sourced LDW (45 µM DSi) that bears a typically low δ30SiDSi value of +1.1 ‰.

The influence of dense LDW does not extend further west than the Porcupine Abyssal Plain, and thus at stations 44 and 32, Upper and Lower LSW give way to the denser ISOW (or its modified product, NEADW) with depth. As it flows over the Iceland–Scotland Ridge, ISOW mixes with more saline waters of the Atlantic thermocline to form NEADW that, being quite dense, comes to lie below LSW as it flows geostrophically along the western edge of the West European Basin (Fig. 1). The differences in DSi concentration and δ30SiDSi between these northern-sourced water masses are small: ISOW-influenced waters at depths of 2500–3000 m bear values of +1.4 ‰ to +1.5 ‰ for δ30SiDSi and 15–25 µM for DSi, whilst LSW is only slightly more DSi-poor and correspondingly higher in δ30SiDSi (10–12 µM and +1.4 ‰ to +1.9 ‰, respectively). The resemblance in DSi concentration and δ30SiDSi between these two water masses is also due to the entrainment of LSW into ISOW. Interestingly, the very base of the water column at STN 44 is occupied by the dense and Si-poor DSOW, where it has a dissolved silicon concentration of about 8 µM and a δ30SiDSi of +1.2 ‰. Although the DSi is typical for DSOW, the δ30SiDSi value of +1.2 ‰ is unexpectedly low for this water mass.

The most important isotope fractionation signal in marine DSi is produced by diatom DSi uptake in the surface ocean (De La Rocha et al., 1997; Varela et al., 2004; Sutton et al., 2013), due to the dominant importance of these phytoplankton for the marine Si cycle (Tréguer and De La Rocha, 2013; Hendry and Brzezinski, 2014). As a result, elevated values of δ30SiDSi can be produced only within the well-lit surface ocean, where photosynthesizing organisms can grow and silicify. This surface–ocean signal is communicated more broadly by the process of water mass subduction, i.e., the transport of surface water parcels into the ocean interior (Stommel, 1979). This is seen particularly clearly in our dataset, which spans a region in which exceptionally deep winter convection gives rise to mixed layers over 1 km deep in the Labrador Sea and Irminger Sea, injecting isotopically fractionated DSi into the ocean interior.

As can be seen from Figs. 4 and 5, there is a clear association of elevated δ30SiDSi values with the low-PV and high-O2 signal of LSW (e.g., STN 77, 502 m), the water mass that is produced by deep winter convection. Indeed, the eastward spread of these elevated values coincides remarkably well with the extension of LSW mapped by McCartney and Talley (1982) based on PV (as shown in Fig. 1). The influence of LSW on the North Atlantic δ30SiDSi distribution is also nicely illustrated by the depth profiles in Figs. 2b and 5, which show that, unlike the eastern Atlantic with low δ30SiDSi values at mid-depths, the central and western North Atlantic bears elevated δ30SiDSi values close to those observed within the Labrador Sea itself (stations 64, 69, and 77), which consists primarily of subducted surface water. Such high values at mid-depths are unique to the North Atlantic Ocean amongst the major open-ocean basins, and result from the local formation of deep waters from Si-depleted surface waters of the subpolar North Atlantic.

Thus, one proximal physical control on the North Atlantic δ30SiDSi distribution is the vertical transport of DSi from the surface ocean to mid-depths during LSW formation. Another physical control is shown by the close correlation between elevated δ30SiDSi and lower PV even within the eastern Atlantic Ocean, far from the region of deep convection (Figs. 4 and 5). This highlights the fact that the spreading of LSW as a result of the regional circulation transports its isotopic signal within the ocean interior, resulting in mid-depth δ30SiDSi values around +1.5 ‰ in regions where the physical signatures of LSW can be seen, documenting the importance of water mass structure on the marine δ30SiDSi distribution. Furthermore, this circulation pattern results in the direct interaction of this northern-sourced water mass with the southern-sourced LDW, producing strong local δ30SiDSi gradients whose systematics correspond nicely to the basin-scale systematics (60∘ S to 60∘ N) documented by de Souza et al. (2012b; see Sect. 4.1 and Fig. 3).

Elevated values of δ30SiDSi are also associated with the dense overflows from the Nordic Seas. Whilst our single sample of predominantly DSOW (STN 44, 2900 m) surprisingly bears a low δ30SiDSi value of +1.2 ‰, ISOW and its derivative NEADW bear similarly low DSi concentrations and similarly elevated δ30SiDSi values to LSW, reaching up to +1.5 ‰ in the abyssal Labrador Sea. They originate as dense bottom-hugging overflows of mid-depth Nordic Sea waters, influenced by the Arctic Ocean, that enter the North Atlantic across the submarine sills running between Greenland, Iceland and Scotland, and the elevated δ30SiDSi values of ISOW and NEADW reflect the isotopically heavy nature of the deep Arctic (Varela et al., 2016). Both Brzezinski and Jones (2015) and Varela et al. (2016) suggest that this feature results from the nature of the inflows to the Arctic Ocean, which receives isotopically fractionated DSi via the upper-ocean inflows from the Atlantic (and, to a lesser extent, the Pacific) due to the shallow sills that form its boundaries to these ocean basins. Observational and modelling studies indicate that these inflows are isotopically heavy primarily due to isotope fractionation during diatom DSi uptake in the Southern Ocean, although more proximal fractionation within the Atlantic and Pacific oceans most likely also plays some role (de Souza et al., 2012a, 2015).

Finally, interesting insights may be gained from a comparison of our Labrador Sea data with the only other published data from this region (de Souza et al., 2012b). Figure 6 compares data from the central Labrador Sea (STN 69) from the GEOVIDE study with literature data from within the Labrador Sea and slightly further south, in the vicinity of the Grand Banks (de Souza et al., 2012b), tracing LSW and NADW as they flow southwards. The three profiles agree within uncertainty at mid-depths and below, but diverge in the upper ocean at depths associated with Upper LSW. Since deep winter convection occurs up to depths of 1000–1500 m regularly within the Labrador Sea, this water mass is frequently ventilated locally, which may result not only in variable physical properties (as shown in Fig. 4) but also changes in its chemical characteristics, such as δ30SiDSi. However, care should be taken not to over-interpret such differences of ∼0.3 ‰, given the potential for δ30SiDSi offsets between laboratories of ±0.2 ‰, as discussed in Sect. 4.1 (Reynolds et al., 2007; Grasse et al., 2017).