Mercury distribution and transport in the North Atlantic Ocean along the GEOTRACES-GA01 transect

Abstract. We report here the results of total mercury (HgT) determinations along the 2014 G eotraces Geovide cruise (GA01 transect) in the North Atlantic Ocean (NA) from Lisbon (Portugal) to the coast of Labrador (Canada). HgT concentrations in unfiltered samples (HgTUNF) were log-normally distributed and ranged between 0.16 and 1.54 pmol L−1, with a geometric mean of 0.51 pmol L−1 for the 535 samples analysed. The dissolved fraction (


Introduction
The ocean plays a central role in the global mercury (Hg) cycle. It receives Hg mainly from atmospheric deposition, whereas it disposes of it in deep marine sediments (e.g. Mason et al., 1994). In the meantime, the largest part of Hg is recycled in the atmosphere, while a smaller fraction pene-Published by Copernicus Publications on behalf of the European Geosciences Union.
trates the ocean interior via thermohaline circulation or the biological pump (see reviews by Mason and Sheu, 2002;Fitzgerald et al., 2007;Mason et al., 2012). Firstly, Hg reinjection to the atmosphere results from the formation of volatile elemental Hg via photoreduction and microbiological reduction of divalent Hg (e.g. Mason et al., 1995;Amyot et al., 1997). Secondly, Hg integration into the thermohaline circulation involves its solubilization in surface waters followed by the subduction of these water masses along isopycnals (e.g. Gill and Fitzgerald, 1988). Thirdly, the biological pump consists of Hg sorption onto biogenic particles produced in the euphotic zone. Then it conveys sinking materials at depth with possible Hg remobilization due to the particulate remineralization-dissolution process (e.g. Mason and Fitzgerald, 1993). The shape of observed vertical oceanic Hg profiles, characterized by increasing concentrations with depth, includes the marks of these different routes and is akin to nutrient-type profiles (Gill and Fitzgerald, 1988;Cossa et al., 2004;Lamborg et al., 2014;Bowman et al., 2015Bowman et al., , 2016. The Hg cycle is also known for being highly perturbed by human activities (e.g. Mason et al., 2012;Lamborg et al., 2014;Zhang et al., 2014;Amos et al., 2015). Modern Hg concentrations in the global atmosphere are more than 3 times the pre-industrial Hg concentrations, leading to increasing Hg concentrations in surface and intermediate oceanic layers, which remain to be precisely estimated. Despite these advances in knowledge of the Hg biogeochemical cycle, the key features of the Hg distribution among the principal oceanic water masses are still poorly documented. Recent enhancements in the precision of Hg analyses allow more reliable vertical Hg profiles in the water columns (e.g. Cossa et al., 2011;Lamborg et al., 2014;Heimbürger et al., 2015;Bowman et al., 2015Bowman et al., , 2016Munson et al., 2015;Cossa et al., 2017a, b). In addition, an original approach for the estimation of the anthropogenic fractions of Hg concentrations in oceanic waters has been proposed (Lamborg et al., 2014). Owing to these last methodological breakthroughs, significant advances in detailed Hg oceanic distributions are possible.
The North Atlantic Ocean (NA) plays an active role in the cycling of chemical species in the ocean because it is a region where deep water formation drives the Atlantic Meridional Overturning Circulation (AMOC) (Kuhlbrodt et al., 2007). Particularly in the subpolar NA, chemical properties, including Hg, are transported to the ocean interior; thus, NA offers a unique opportunity for studying the oceanic response to changes in atmospheric Hg deposition. The GEOTRACES-GA03 zonal and meridional transects, sampled in 2010 and 2011, covered the NA from east to west between 18 and 40 • N, from the coast of Africa to the coast of the USA. Here, we report the results of the GEOVIDE cruise along the GEOTRACES-GA01 transect, which targeted the NA from 40 to 60 • N, from Portugal to Newfoundland via the southern tip of Greenland (Fig. 1). This article provides (i) a highresolution description of the HgT distribution in the waters of the subpolar and subtropical gyres of the NA, (ii) characterization the HgT concentrations of the main water masses of the NA, (iii) an estimate of the temporal change of anthropogenic Hg in LSW and (iv) quantification of the HgT transport associated with the upper and lower limbs of the AMOC. These new data contribute to a refinement of the depiction of the Hg distribution in the NA waters and should allow further improvements in the oceanic Hg modelling.

Oceanographic context
A full description of the water masses along the GEOTRACES-GA01 transect can be found in García-Ibáñez et al. (2018). Briefly, the North Atlantic Current (NAC) conveys the warm salty surface waters from subtropical regions northwards to the subpolar regions, where they are cooled down by heat exchange with the atmosphere (Fig. 1). The intermediate and deep waters formed this way fill up the Global Ocean, initiating the southward-flowing limb of the AMOC (e.g. McCartney and Talley, 1984;Lherminier et al., 2010). In addition, the general circulation pattern is characterized by the subtropical and the subpolar gyres (Fig. 1).
In the subtropical gyre ( Fig. 1), several water masses can be identified. They are listed from top to bottom: (i) the mixed layer, (ii) the eastern North Atlantic Central Water (ENACW), (iii) the Mediterranean waters (MW), (iv) the Labrador Sea Water (LSW) and (v) the Lower North East Atlantic Deep Water (NEADW L ), which contains about 30 % Antarctic Bottom Water (AABW) (García-Ibáñez et al., 2015). The transformation of ENACW leads to the formation of different mode waters including the Subpolar Mode Waters (SPMWs) (McCartney and Talley, 1982;Tsuchiya et al., 1992;van Aken and Becker, 1996;Brambilla and Talley, 2008;Cianca et al., 2009). SPMWs are the near-surface water masses of the subpolar gyre of the NA characterized by thick layers of nearly uniform temperature, often denoted with temperature in subscript (SPMW 8 , for example). SPMWs are formed during winter convection at high latitudes due to atmospheric freshening of surface waters originating from the subtropical gyre . SPMWs participate in the upper limb of the AMOC and provide much of the water that is eventually transformed into the several components of North Atlantic Deep Water (NADW; Brambilla and Talley, 2008).
In the subpolar gyre, ocean-atmosphere interaction is particularly intense. The cooling down of subtropical waters produces dense waters, triggering the deepening of the mixed layer and further leading to deep convection. The main NA convection zones are located in the Labrador (LS), Irminger (IrS) and Nordic seas (NS) (Fig. 1)  components of the NADW, which constitutes the cold deep limb of the AMOC, flowing southward towards the Southern Ocean in the western Atlantic basin. LSW has been variably produced in the past 50 years, depending on the intensity of winter convection, linked to the intensity of the North Atlantic Oscillation (e.g. Rhein et al., 2002;Cianca et al., 2009;Yashayaev and Loder, 2016). Depths of winter convection in the LS vary from a few hundred metres (the early 2000s) to over 2000 m (early 1990s). The LSW is a thick layer in the LS but thins out as it travels south-westwardly. It spreads out into the entire NA, filling the subpolar gyre and entering the subtropical gyre. Within the subpolar gyre, LSW is marked by a salinity minimum above the ISOW. In both gyres, the well-ventilated LSW has a marked oxygen maximum.  (Table S1 in the Supplement) were occupied for hydrographic profiles (CTD, dissolved oxygen, nutrients), of which 29 included trace-metal sampling. Sampling and water treatment for HgT determination Cutter et al., 2017) were performed using ultra-trace techniques following the GEOTRACES recommendations. During the GEOVIDE cruise, an epoxy-coated aluminum rosette, equipped with 12 L GO-FLO (General Oceanics©) bottles initially cleaned following the GEOTRACES procedures (Cutter and Bruland, 2012), was deployed on a 6 mm Kevlar hydrowire. The rosette was also equipped with probes for pressure, conductivity, temperature, dissolved oxygen, fluorometry and transmission measurements (titanium SBE model 911-plus, Sea-Bird Electronics ® ). Specifically for Hg determination, all materials in contact with the seawater samples were made of Teflon or were Teflon coated, acid cleaned and rinsed with ultrapure water (Milli-Q, Millipore ® ) prior to utilization. Original vent fixture and sampling valves of the GO-FLO bottles were replaced with Teflon (PTFE) ones. GO-FLO bottles were subsampled under a laminar flow bench inside a clean tracemetal container. The efficiency of the high-efficiency particulate air filter (HEPA, 0.3 µm) in the container was checked with a Coulter counter during the cruise. All subsequent sample treatments (including filtration) and Hg analyses were also performed in clean class 100 containers. filtrations, acid-washed 0.45 µm polycarbonate membranes (Nuclepore) were preferred to cellulose acetate or polyethersulfone membranes proposed in the GEOTRACES protocols ( Fig. S1 in the Supplement). Subsamples were stored in Teflon bottles (FEP) until the on board HgT analyses, which occurred within 6 h after sampling.

Chemical analyses
In order to access all Hg species, the release of Hg from its ligands was achieved by a BrCl solution (50 µL of a 0.2 N solution is added to a 40 mL sample), and then the Hg was reduced with an acidic SnCl 2 solution (100 µL of a 1 M solution is added to a 40 mL sample). Potassium bromide (Sigma Aldrich, USA) and potassium bromate (Sigma Aldrich, USA) were heated for 4 h at 250 • C to remove Hg traces before making up BrCl solution with freshly doubledistilled HCl (Heimbürger et al., 2015). The generated Hg vapour was amalgamated into a gold trap and then released by heating into an atomic fluorescence spectrometer (AFS). We used two AFS systems in parallel (Tekran ® Model 2500, Brooks ® Model 3), both calibrated against the NIST 3133 (National Institute of Standards and Technology) certified reference material. This technique, initially described by Bloom and Crecelius (1983) and subsequently improved by Gill and Fitzgerald (1985), is now an authoritative procedure officialised by the US EPA as method 1631 (Environmental Protection Agency, 2002). The definitions of detection limit (DL), reproducibility and accuracy given here are adopted from Taylor (1987) and Hewitt (1989). Using a mirrored quartz cuvette (Hellma ® ) allowed for an "absolute DL", defined as 2 times the electronic noise magnitude, as low as 1.7 femtomoles. However, in practice for trace measurements, the DL is governed by the reproducibility of the blank values and calculated as 3.3 times the standard deviation of blank values. The blank values were determined on a purged Hg-free seawater sample spiked with reagents (i.e. BrCl and SnCl 2 ). The mean (± standard deviation) of blank values measured during the GEOVIDE cruise was 3.2 ± 1.0 femtomoles. Thus, for a 40 mL seawater aliquot, the DL expressed in HgT concentration was 0.07 pmol L −1 . The reproducibility (coefficient of variation of six replicate measures) varied according to the concentration level between 5 and 15 %. The accuracy of HgT measurements was tested using ORMS-5 certified reference material (CRM) from the National Research Council of Canada (http://www.nrc-cnrc.gc.ca/, last access: April 2018) and spiked to a purged Hg-free seawater sample. Measurements were always within the given confidence interval. To ensure good data quality and to continue previous efforts (Cossa and Courau, 1990;Lamborg et al., 2012), we organized the 2014 GEOTRACES intercalibration exercise for total HgT and methyl Hg as a part of the GEOVIDE cruise. The intercalibration sample was taken on 22 June 2014 in the LS at 49.093 • W, 55.842 • N, and 2365 m depth. The sample was sent out to 10 participating laboratories. This station was also planned as crossover station within the 2015 Arctic GEOTRACES effort (Canadian cruise) but has subsequently been changed to another location. Our results compare well with the consensus values, HgT = 0.63 ± 0.12 pmol L −1 , n = 8. We measured the 2014 GEOTRACES intercalibration sample twice for HgT and obtained 0.51 (22 June 2014, on board) and 0.58 pmol L −1 (30 October 2014, home lab).

Extended optimum multiparameter analysis
We used an extended optimum multiparameter (eOMP) analysis to characterize the water mass HgT UNF concentrations along the GEOTRACES-GA01 transect (García-Ibáñez et al., 2015, 2018. The eOMP analysis quantifies the proportions of the different source water types (SWTs) that contribute to a given water sample. The HgT UNF concentration of each SWT, [HgT UNF ] i , was estimated through an inversion of the SWT fractions given by the eOMP analysis. This approach was successfully applied to dissolved organic carbon water mass definitions in the NA (Fontela et al., 2016) where [HgT UNF ] j represents the measured HgT UNF concentration for each sample j , SWT j i the proportion of SWT i to sample j (obtained through the eOMP), [HgT UNF ] i the HgT UNF concentration for each SWT i (unknown), and ε j the residual. The 430 ε j s of the inversion presented a null mean and a standard deviation of 0.085 pmol L −1 (R = 0.84).

Mercury transport calculation
Velocity fields across the GEOTRACES-GA01 transect were calculated using an inverse model constrained by Doppler current profiler velocity measurements  Biogeosciences an overall mass balance of 1 ± 3 Sv to the north (Lherminier et al., 2007. The volume transport per SWT was computed by combining these velocity fields with the results of the eOMP (García-Ibáñez et al., 2018). Finally, the HgT UNF transport per water mass was calculated through Eq. (2): where T SWTi is the volume transport of SWT i, [HgT UNF ] i is the HgT UNF concentration for each SWT i (from Eq. 1), and ρ i is the density of the SWT i.
The inverse model configuration for the GEOVIDE cruise data is described in Zunino et al. (2017). The inverse model is based on the least squares formalism, which provides errors on the velocities and associated quantities such as the magnitude of the AMOC (estimated in density coordinate) and the  heat flux . The inverse model computes the absolute geostrophic transport orthogonal to the section. The Ekman transport is deduced from the wind fields averaged over the cruise period and added homogeneously to the upper 40 m . The transport estimates of the inverse model across the section have been validated by favourable comparisons with independent measurements (Gourcuff et al., 2011;Daniault et al., 2011;Mercier et al., 2015).

Results
Distributions of potential temperature, salinity, dissolved oxygen and silicic acid are given in García-Ibáñez et al. (2018), HgT UNF concentrations along the GEOTRACES-GA01 transect ranged from 0.16 to 1.54 pmol L −1 (n = 535), these data being log-normally distributed, positively skewed (Skewness = 1.1; Kurtosis = 2.1; Fig. S3) and having 97 % of the values lower than 1.00 pmol L −1 . The geometric mean and the median were 0.51 pmol L −1 , whereas the arithmetic mean and standard deviation were 0.54 and 0.19 pmol L −1 , respectively. These concentrations are within the range found along the GEOTRACES-GA03 transect (0.09-1.89 pmol L −1 , n = 605) that crossed the NA within the subtropical gyre from 18 to 40 • N (Bowman et al., 2015), but are lower than the range and the unusually high arithmetic mean determined in the South Atlantic along the GEOTRACES-GA10 transect (0.39-3.39 pmol L −1 , n = 375; Arne Bratkič, personal communication, 2017 and 1.45 ± 0.60 pmol L −1 ; Bratkič et al., 2016). The overall distribution of the HgT UNF concentrations along the GEOTRACES-GA01 transect is represented in Fig. 2. The main features of HgT UNF concentrations are an eastward increase from Greenland to Europe and a downward increase from subsurface to bottom waters. In addition, the lowest and highest (most variable) HgT UNF values were encountered in surface and subsurface waters, where Hg evasion to the atmosphere and high particulate matter concentrations may generate low and high HgT UNF concentrations, respectively. Out of the 141 filtered samples that were analysed, altogether, the filtered fraction of Hg (HgT F ) represents, on average, 78 % (range: 36-98 %) of the HgT UNF (Fig. 3). Excluding the upper 100 m, where the biogenic suspended particles are usually abundant, and the stations located on the shelf and slope, where particulate matter from continental sources are usually present, the HgT F fraction represents, on average, 84 % (range: 72-98 %) of the HgT UNF . In addition, in the LS, HgT F / HgT UNF mean ratios were rather low ranging 62-92 %, with a mean of 76 %. In fact, the primary production was high in spring 2014 in LS, and the winter convection, which reached 1400 m, conveyed surface particles at depth (Yashayaev et al., 2015;Lemaitre et al., 2017). If we exclude the LS from the HgT F mean computation, we obtain a mean percentage HgT F / HgT UNF ratio of 91 %, which is similar to values (∼ 90 %) obtained along the GEOTRACES-GA03 zonal and meridional transects (Bowman et al., 2015). In the following subsections, detailed descriptions of the HgT UNF profiles for the five following oceanographic environments are given: LS, IrS, Iceland basin (IcB), the eastern North Atlantic basin (ENAB) and Iberian Abyssal Plain (IAP). In the LS, the HgT UNF concentrations ranged from 0.25 to 0.67 pmol L −1 , with a mean (± standard deviation) of 0.44 ± 0.10 pmol L −1 (n = 113). Distribution, source and cycling of Hg in the LS have been described and discussed in detail in a companion paper (Cossa et al., 2017b). In summary, the highest HgT UNF concentrations were found in the waters of the Labrador Current (LC) receiving fresh water from the Canadian Arctic Archipelago and in the waters over the Labrador shelf and continental rise. In the LSW that formed during the 2014 winter convection, HgT UNF concentrations were low (0.38 ± 0.05 pmol L −1 , n = 23) and increased gradually with depth (up to > 0.5 pmol L −1 ) in the North East Atlantic Deep Water.

Iceland basin (stations 34-38)
HgT UNF concentrations in the IcB ranged from 0.18 to 0.65 pmol L −1 , with a mean of 0.46 ± 0.10 pmol L −1 (n = 51). In the top 100 m of the water column, HgT UNF concentrations were quite variable (0.25-0.62 pmol L −1 ), probably as a result of the counteracting importance of Hg evasion to the atmosphere, high particulate matter and/or complexing substance concentrations. West of the IcB (Station 38), contrasting HgT UNF levels were found on both sides at 500 m, characterized by a thermohaline gradient ( Fig. 2a  and b in García-Ibáñez et al., 2018). In the top waters, HgT UNF levels were depleted to 0.18 pmol L −1 , whereas below 500 m, they were much higher and converge to values close to what we found, at the same depths in the adjacent IrS (∼ 0.60 pmol L −1 , Station 40). In the bottom waters, constituted by more than 50 % of ISOW (García-Ibáñez et al., 2018), HgT UNF concentrations reached values > 0.50 pmol L −1 .

Eastern North Atlantic basin (stations 17-32)
The HgT UNF concentrations in the ENAB varied from 0.18 to 1.14 pmol L −1 , with a mean of 0.61 ± 0.18 pmol L −1 (n = 174). The ENAB, also named western European basin, is characterized by complex vertical stratification of the water column. The HgT UNF vertical profiles at all the stations of the ENAB were characterized by a complex but reproducible pattern depicting (i) two maxima peaks (the upper one at the subsurface, the lower within the intermediate waters), and below, (ii) a HgT UNF enhancement from 2500 m to the bottom (Fig. 2). The position and intensity of the peaks vary with longitude. The upper peak, which occurs within the top 200 m, is only 0.48 pmol L −1 at Station 29, but reaches 1.14 pmol L −1 at Station 19 (Fig. 2). The vertical position of maxima of the lower peak deepens eastwards, from 200 m down to 800 m, concurrently with an increase in its magnitude (Fig. 2). The position of the upper peaks suggests a relation with the position of the fluorescence maximum (data not shown), whereas the position of the lower peaks, which is close to the maximum of apparent oxygen utilization (AOU) that rose above 70 µmol L −1 (Fig. 2), suggests a dependence on the organic matter remineralization (see Discussion below). Between 1400 and 2500 m, in the layer corresponding to LSW, HgT UNF concentrations were quite uniform, with a mean concentration of 0.54 ± 0.04 pmol L −1 (n = 18). The HgT UNF concentration increased from 3000 m downwards to the sea bottom, consisting of NEADW L , where it reaches 0.95, 0.97, 1.03 and 1.13 pmol L −1 at stations 21, 19, 25 and 23, respectively.

Iberian Abyssal Plain (stations 1-15)
In the IAP, HgT UNF concentrations ranged from 0.19 to 1.54 pmol L −1 , with a mean of 0.69 ± 0.23 pmol L −1 (n = 94). The highest HgT UNF concentrations were measured in the upper 100 m near the shelf slope. At Station 2, the only station on the European shelf (bottom at 152 m), the HgT UNF concentrations increased from 10 m to the bottom, from 0.38 to 0.86 pmol L −1 , but did not differ from the open NA ocean levels. Offshore, at stations 1, 11, 13 and 15 (Fig. 2), the vertical distributions of HgT UNF presented a certain similarity with those of the eastern ENAB, but with an additional third deep peak. As in the eastern ENAB, the upper peak is associated with subsurface waters, and the second, centred around 800 m, is associated with the oxygen minimum of SPMW 8 . The third peak, centred around 1100-1200 m, is associated with the salinity maximum of the core of MW (Fig. S2). The presence of a HgT UNF peak in the MW was still visible westwards, at stations 17, 19 and 23, near 1100 m, as a shoulder of the main peak at 800 m (Fig. 2). Deeper within the water column, HgT UNF increased grad- the subpolar gyre (Fig. 2). A multipeak pattern was also observed in 1994 in the eastern Atlantic slope water column in the Celtic Sea (Cossa et al., 2004). The shape of the Hg profiles exhibited the same peaks in the same water masses as the ones observed in this study (i.e. SPMWs and MW). However, HgT UNF concentration levels measured 20 years ago were much higher, varying most often from 0.3 pmol L −1 in subsurface waters to more than 2.0 pmol L −1 at depth.

Biogeochemical and hydrographical controls on HgT distribution
The main paths of the Hg cycle in the open ocean can be briefly summarized as follows. Direct atmospheric deposition is the dominant source of Hg for the oceans, most of the deposited Hg is re-emitted in the atmosphere, and a minor Hg fraction is drawn down to the ocean interior with downwards convecting waters or associated with sinking particles. At depth, the dissolution of particulate matter, produced as a result of organic matter microbiological remineralization, remobilizes Hg from particles produced in the euphotic zone. The biological pumping/regeneration process results in a relationship between Hg concentrations and nutrient or dissolved oxygen concentration (or AOU), which are proxies of the organic matter remineralization (mainly the microbial respiration) that the sample had experienced since it was last in contact with the atmosphere. This biogeochemical behaviour, which is qualified of nutrient-like behaviour, is observed in the present study (Fig. 4). The correlation coefficient (R) between HgT F and the AOU, obtained from in situ measurements of dissolved oxygen and temperature, reached the highly statistically significant value of 0.87 (n = 141, p < 0.01). Similar behaviour was already observed in the water column near the shelf edge of the western European margin (Cossa et al., 2004) and elsewhere in the NA (Lamborg et al., 2014;Bowman et al., 2015). Thus, the present results confirm that biological uptake and regenerative processes appear to control a large part of the oceanic Hg distribution in the subpolar and subtropical gyres of the NA. Hydrological circulation may also impact the Hg distribution in the NA. We estimated the HgT UNF (and AOU) values of each SWT using eOMP (Table 1). The correlation coefficient between observed and predicted (eOMP-based) values calculated with Eq. (1) (Materials and Methods section) for HgT UNF is 0.71. The estimated HgT UNF concentrations vary significantly between SWTs from 0.32 ± 0.03 to 1.04 ± 0.02 pmol L −1 for the IrSPM to the NEADW L . However, a large part of the HgT UNF between SWTs is due to the regeneration process as suggested by the correlation coefficient (R = 0.82) of the linear relationship between HgT UNF and AOU (Fig. 5). Based on this model (HgT UNF = 0.0043 × AOU + 0.3547), we calculated mean corrected HgT UNF concentrations for each identified SWT for a zero AOU concentration. Corrected mean values range from 0.22 to 0.61 pmol L −1 in IrSPMW and NEADW, respectively (Table 1). This variation should result from the origin, the route and the age of each SWT. The corrected HgT UNF values of IrSPMW, PIM, SPMW, DSOW and LSW, which formed in the subpolar gyre and in the NS last winter, present very low and similar values, 0.22-0.31 (Table 1). The IrSPMW is the youngest SPMW that has formed in the IrS as a result of interaction between the air and the waters transported northwards by the NAC (e.g. McCartney and Talley, 1984); the low HgT UNF value found in the Ir-SPMW may result from a net Hg evasion in this region, consistently with the conclusion that western and central NA are a net source of Hg in the atmosphere (Mason et al., 2017). On the contrary, on the eastern NA side, where Hg deposition and evasion are rather similar (Mason et al., 2017), the ENACW shows a higher corrected HgT concentration (0.41 pmol L −1 , Table 1). The highest corrected HgT UNF mean concentration is calculated for NEADW L (0.61 pmol L −1 , Table 1), which is the dominant water mass in the bottom IAP. Its main core is below ∼ 3500 m depth and spreads down to the bottom (see Fig. 4 in García-Ibáñez et al., 2018). This water mass contains a significant component from the Southern Ocean (AABW), which is known to be Hg-rich (HgT AABW = 1.35 ± 0.39 pmol L −1 , Cossa et al., 2011). The same rationale can be drawn for the corrected HgT UNF concentration in MW (0.41 pmol L −1 , Table 1). Indeed, recent measurements in the waters of the western Mediterranean give HgT UNF values between 0.53 and 1.25 pmol L −1 within the layer that flows out of the Mediterranean Sea at the Strait of Gibraltar (Cossa and Coquery, 2005;Cossa et al., 2017a).  74-27.82, 1200-2000 m) between the Labrador Sea and the eastern North Atlantic basin. HgT Anth values were obtained according to the model by Lamborg et al. (2014). Young LSW corresponds to the LSW 2014 ) formed during winter 2013-2014. The insert shows the Hg concentration decrease in the troposphere over the North Atlantic during the last 20 years according to Soerensen et al. (2012). Table 1. Total Hg in unfiltered samples (HgT UNF ) vs. apparent oxygen utilization (AOU) concentrations of each source water type (SWT), calculated according to eOMP (Eq. 1) (see also García-Ibáñez et al., 2018). Corrected HgT UNF is a theoretical HgT concentration for a AOU concentration equal to zero, using the equation from

SWT
HgT UNF AOU Corrected HgT UNF (pmol L −1 ) (µmol L −1 ) (pmol L −1 ) In summary, the distribution pattern of HgT UNF along the GEOTRACES-GA01 transect, modelled by the mixing of SWTs (Fig. S4), stresses the importance of Hg scavenging by plankton and organic matter regeneration, but also shows that part of the Hg enrichment in certain SWTs, including MW and NEADW, is due to preformed Hg outside the NA. This type of result, characterizing the Hg concentrations in principal oceanic water masses, should contribute to refinements in model formulation and predictability.

Change in anthropogenic Hg in LSW
Evidence for a decrease in the Hg anthropization in the NA waters can be obtained from the comparison of the present results with those obtained 20 years ago with similar clean sampling and analytical techniques. In a companion paper  (Mason et al., 1998). Between 1993 and 2014 the decrease in HgT UNF concentrations would have been more than a factor of 2 (1.14 ± 0.36 pmol L −1 vs. 0.40 ± 0.07 pmol L −1 ). However, bearing in mind the uncertainty of the accuracy of early numbers, this magnitude of decrease cannot be taken for granted. To circumvent this difficulty, the approach proposed by Lamborg et al. (2014) can be used to estimate the anthropogenic Hg (Hg Anth ) concentrations in subsurface waters. Hg Anth is inferred from the difference between measured HgT UNF concentrations and the concentrations predicted based on a worldwide relationship between deep-ocean Hg concentrations and remineralised phosphate (Lamborg et al., 2014). There is a Redfield ratio of 141 between AOU and remineralized phosphate (Minster and Boulahdid, 1987), which is more representative for the NA than the global value of 170 proposed by Anderson and Sarmiento (1994). The LSW takes less than 20 years (Doney et al., 1997) to flow more than 3000 km eastward from the LS to the subtropical gyre of the NA. Along its path, LSW bears the record of Hg incorporation at the time of its formation; thus sampling along its flow path allows the observation of decadal variations in anthropogenic Hg inputs to the NA. In the NA, estimation of Hg Anth concentrations in the core of LSW, defined within potential densities of 27.74 and 27.82, account for 36 ± 0.07 % of the HgT UNF , and are one-third lower for younger waters (LS and IrS: 0.16 ± 0.11 pM) than for older waters (IcB and ENAB: 0.24 ± 0.06 pM) (Fig. 6). This 30 % decrease in Hg concentrations are consistent (i) with the observations of a temporal decrease of Hg in the marine boundary layer of the NA (Sprovieri et al., 2010;Weigelt et al., 2014) over the last two decades and with (ii) the estimated decline in Hg concentrations in subsurface waters of the NA estimated by models over the last few decades (e.g. Soerensen et al., 2012). This means that LSW that formed in the 1990s in the LS and is currently present in the ENAB received more Hg Anth from the atmosphere than the LSW 2014−2015 that formed during the 2014 winter. These results contrast with what can be deduced from the vertical profile of HgT UNF in the LS, where the Hg regeneration in the water column is sufficient to account for the Hg increase between the shallow LSW layer (LSW 2014(LSW −2015 ) and the deep LSW layer (LSW 1987(LSW −1994 ) (Cossa et al., 2017b). This discrepancy between these two deductions suggests that LSW, which is present in the eastern NA, is likely older (and more imprinted by legacy Hg Anth ) than the LSW currently present in the LS.

Latitudinal transport of Hg
The transport of HgT UNF per unit of water mass, calculated with Eq. (2) (Materials and Methods section), are given in Table 2. We also applied Eq. (2) separately to the upper and lower limbs of the AMOC and computed the transport of HgT UNF per water mass for the two limbs. The velocity fields across the Portugal-Greenland transect were calculated using an inverse model constrained by Doppler current profiler velocity measurements . The volume transport per SWT was computed by combining these velocity fields with the results of the eOMP (García-Ibáñez et al., 2018).
The mean (velocity-weighted) HgT UNF concentration of the water advected northwards within the upper limb of the AMOC is 0.55 pmol L −1 , whereas the one advected south-Biogeosciences, 15, 2309-2323, 2018 www.biogeosciences.net/15/2309/2018/ wards within the lower limb of the AMOC is 0.40 pmol L −1 . Across the Portugal-Greenland transect, there is northward HgT UNF transport within the upper limb of the AMOC (10.20 mmol s −1 ), and southward HgT UNF transport within the lower limb (7.12 mmol s −1 ), resulting in a net northward transport of 97.2 kmol yr −1 . Most of the HgT UNF southward transport is due to IrSPMW and ISOW displacements, whereas the HgT UNF northward transport is associated with ENACW and SPMW displacements ( Table 2). The Hg exchange across the LS section can be roughly estimated at 133 kmol yr −1 , using the mean southward water transport of the shelf edge LC as the Seal Island transect (Hamilton Bank near stations 77 and 78; 7.5 Sv, according to Han et al., 2008) and a mean HgT concentration of 0.56 pmol L −1 (Cossa et al., 2017b). Thus, from our snapshot study, the net Hg exchange across the GEOVIDE transect, which crosses the LS and the NA from Portugal and Greenland, would mean a loss in the Arctic of 36 kmol yr −1 . In comparison, Soerensen et al. (2016), based on a mass balance budget, estimated that "Arctic seawater is enriched in total Hg relative to inflowing waters from the North Atlantic and North Pacific Oceans at all depths, resulting in a 26 Mg a −1 (i.e. 130 kmol a −1 ) net loss from the Arctic via circulation".

Summary and conclusions
HgT UNF concentrations in the waters along the GEOTRACES-GA01 transect, which crossed the NA from 40 to 60 • N (Portugal to Canada), ranged from 0.16 to 1.54 pmol L −1 , but with 97 % of the values lower than 1.00 pmol L −1 and a geometric mean of 0.51 pmol L −1 (n = 535). The dissolved fraction (< 0.45 µm) of HgT, determined on 141 samples, averaged 78 % of the HgT UNF for the entire data set, 84 % for deep open seawaters and 91 % if the Labrador Sea data, where the primary production was high, are excluded. HgT UNF concentrations increased eastwards and downwards. The HgT UNF concentrations were similarly low in the subpolar gyre waters (∼ 0.45 pmol L −1 ), whereas they exceeded 0.60 pmol L −1 in the subtropical gyre waters, especially within NEADW L . The relationship between HgT F and AOU, which indicates a nutrient-like behaviour for Hg in the NA, attests to the influence of organic matter regeneration on HgT mobilization. The distribution pattern of HgT UNF along the transect, modelled by the mixing of SWTs, show Hg enrichment in MW and NEADW, and low Hg concentrations in younger water masses that formed last winter at high latitudes. Using the HgT UNF fraction unexplained by regeneration processes as a proxy for Hg Anth , we observed geographical trend in the Hg Anth in the LSW along its eastward journey in the NA. It was characterized by an eastward increase, which suggests that Hg incorporation in the downwelling waters of the LS has decreased over the last 20 years, parallel with the decrease in Hg concentrations in the NA troposphere. By combining the velocity fields with the results of the eOMP, a net northward Hg transfer of 97.2 kmol yr −1 across the Portugal-Greenland transect can be calculated as a result of the AMOC. Taking into account the southern Hg export with the LC, the net Hg exchange along the entire GEOVIDE transect would result in a loss in the Arctic of 36 kmol yr −1 .