Samples were collected on RV Meteor cruise M107 in June 2014 during nine deployments at seven locations (two stations were occupied twice) along a cross-shelf transect at 18∘20′ N on the Mauritanian shelf in the ETNA (Fig. 1). The bottom depths of stations varied between 50 m on the shelf to 1136 m furthest offshore. Seawater sampling was carried out using a trace-metal-clean CTD (TM-CTD, Sea-Bird SBE25) rosette frame equipped with 24 trace-metal-clean samplers (12 L, Ocean Test Equipment (OTE)). The CTD frame was attached to plastic-coated nonconductive steel cable and deployed using a carousel auto-fire module (AFM, Sea-Bird) that closed the bottles at predefined depths. After recovery, the bottles were transferred to a clean-laboratory container and pressurized to 0.2 bar overpressure using filtered N2 gas. Samples were collected unfiltered for total dissolvable (TD) TM measurements and filtered using a 0.2 µm cartridge filter (Acropack 500, Pall) for dissolved (d) TMs and iodide. Trace-metal samples were collected in acid-clean 125 mL low-density polyethylene (LDPE) bottles (Nalgene), and iodide samples in opaque 60 mL high-density polyethylene (HDPE) bottles (Nalgene). Trace-metal samples were acidified to pH 1.9 using ultra-clean HCl (UpA, Romil) and stored (double bagged) for >6 months before preconcentration and analysis. Samples for iodide measurements were stored frozen at -20 ∘C until analysis.

Samples for the determination of radium isotopes (223Ra; t1/2=11.4 d; 224Ra; t1/2=3.7 d) were obtained using in situ filtration pumps (Challenger Oceanic) following the procedures described in Charette et al. (2015) and Henderson et al. (2013). Briefly, each in situ filtration pump was equipped with two particle filters (70 and 1 µm) and two Mn dioxide (MnO2)-impregnated cartridges (CUNO Micro Klean III acrylic) on which dissolved Ra adsorbs. In this work, 224Ra/223Ra ratios are shown, which were analyzed from the first cartridge. The pumped water volumes varied between 1000 and 1700 L and flow rates were 10–15 L min-1. For the determination of Ra in surface waters (∼5 m water depth), about 200–300 L of seawater was pumped into several 120 L plastic barrels followed by filtration over MnO2-coated acrylic fibers (Mn fibers).

Determination of Co, Mn, Fe, Cd, Pb, Ni, and Cu was carried out as described in Rapp et al. (2017). Briefly, samples were preconcentrated using an automated preconcentration device (SeaFAST, Elemental Scientific Inc.) equipped with a cation chelating resin (WAKO; Kagaya et al., 2009). Samples were UV digested prior to preconcentration to breakdown metal–organic complexes, which would cause an underestimation of the determined TM concentrations. Samples were in-line buffered to pH 6.4±0.2 using 1.5 M ammonium acetate buffer before loading onto the resin. The pH buffer was prepared using an ammonium hydroxide solution (22 %, OPTIMA grade, Fisher) and acetic acid (glacial, OPTIMA grade, Fisher) in deionized water (Milli-Q, Millipore), adjusted to pH 8.5. Retained TMs were eluted from the resin using 1 M distilled HNO3 and collected in 4 mL polypropylene scintillation vials (Wheaton). The acid was distilled from supra-pure HNO3 (SpA grade, Romil) using a sub-boiling perfluoroalkoxy polymer (PFA) distillation system (DST-1000, Savillex). Preconcentration was performed within a clean laboratory (ISO 5) and all sample and reagent handling was performed within the same laboratory in an ISO 3 laminar flow bench with a HEPA (high-efficiency particulate air) filter unit. Preconcentrated samples were analyzed by high-resolution inductively coupled plasma mass spectrometry (HR-ICP-MS, ELEMENT XR, Thermo Fisher Scientific) using isotope dilution for Fe, Cd, Pb, Cu, and Ni and standard additions for Co and Mn. SAFe (sampling and analysis of iron) reference seawaters S and D2 were analyzed with each analytical run and concentrations produced were in good agreement with consensus values (Table 1).

Leachable particulate (Lp) concentrations were calculated as the difference between total dissolvable and dissolved concentrations. The limit of quantification (LOQ) for the Lp concentrations was determined as the sum of the analytical standard deviations of TD and dissolved concentrations. Extended uncertainty calculations were performed using the Nordtest approach (Naykki et al., 2015) accounting for random as well as systematic errors (Rapp et al., 2017). The Lp fraction represents the particulate fraction which is readily dissolvable in the acidified samples during storage at pH 1.9 for 6 months and therefore does not contain any refractory particle components. This more labile fraction of particulate TMs mainly includes TMs in organic and/or biogenic particles, adsorbed to particle surfaces, and TM oxides and/or oxyhydroxides (Hurst et al., 2010).

Aluminum concentrations were determined in surface water samples for all stations along the transect and at two stations (stations 3 and 8) for the entire water column. Samples were analyzed for Al using the batch lumogallion method (Hydes and Liss, 1976). Acidified samples were buffered manually with a 2 M ammonium acetate buffer (Romil, UpA) to a pH between 5.1 and 5.2. The buffer was prepared using ammonium hydroxide (Romil, UpA) and acetic acid (Romil, UpA) in deionized water (Milli-Q, Millipore). Buffered samples were spiked with a 2 mg L-1 lumogallion (TCI) solution allowing the complexing agent to be in excess. The lumogallion solution was prepared in 2 M ammonium acetate buffer (Romil, UpA). After spiking, samples were heated up for 1.5 h at 80 ∘C in an oven (Heratherm, Thermo Scientific) and left to cool down overnight at room temperature to allow the formation of a fluorescent Al complex. Samples were measured using a fluorescence spectrophotometer (Cary Eclipse, Agilent). The samples were measured with excitation and emission wavelengths of 465 and 555 nm, respectively. The excitation and emission slits were set to 10 nm. The plastic cuvettes used for the measurements were precleaned in a 2 M HCl (trace-metal grade, Fisher) for at least 24 h. In between samples, the cuvette was thoroughly rinsed with deionized water followed by actual sample. The same cuvette was used during an analytical session. All samples were analyzed in duplicate and the concentrations calculated from the peak heights via standard addition. Samples and reagent natural fluorescence was monitored by analyzing their content in the absence of the complexing agent. The standards were prepared in low-trace-metal seawater from a 500 nmol L-1 stock standard solution prepared from a 1000 ppm Al standard solution (Merck Millipore). A typical calibration had the following standard concentrations: 0, 10, 20, 40, and 60 nmol L-1. GEOTRACES reference seawater (GS) was run with a mean average Al value of 27.76±0.17 nmol L-1 (n=4; consensus value 28.2±0.2 nmol L-1).

Frozen samples were defrosted overnight at room temperature prior to analysis for iodide by cathodic stripping square-wave voltammetry after Luther et al. (1988). The voltammetry unit consisted of a voltammeter stand (663 VA, Metrohm), an autosampler (863 Compact Autosampler, Metrohm), and an automatic burette (843 Pump Station, Metrohm) for automated spike addition. The system was controlled by Computrace software (797 VA; Metrohm).

Oxygen, salinity, nutrients, turbidity, and chlorophyll fluorescence were measured during 62 CTD deployments (including some repeated deployments at the same location) along the 18∘20′ N transect using a Sea-Bird SBE 9 CTD rosette system equipped with double sensor packages for O2, salinity, and temperature and 24 Niskin samplers (10 L; OTE). Turbidity and chlorophyll a were measured with a combined WETlabs turbidity and fluorescence sensor that was attached to the CTD. The output of both sensors was corrected using the calibration provided by the manufacturer. Throughout this article, turbidity data are presented in nephelometric turbidity units (NTU). The noise level of the sensor in our data set was found to be lower than 0.14 NTU. Oxygen sensor data were calibrated by Winkler titration (Hansen, 2007; Winkler, 1888; Sommer et al., 2015) on 348 discrete water samples that were collected from the OTE samplers. Oxygen sensor data were initially processed using calibration coefficients provided by the manufacturer. Subsequently, O2 sensor data were fitted to the O2 concentrations determined by the Winkler titration method using linear functions for temperature, O2, and pressure (i.e., depth). An uncertainty of 1.5 µmol kg-1 was determined for O2 concentrations. Onboard nutrient measurements of nitrite (NO2-), nitrate (NO3-), phosphate (PO43-), and silicic acid (Si(OH)4) of the discrete water samples were conducted using a QuAAtro autoanalyzer (Seal Analytical) according to the study by Grasshoff et al. (1983).

Apparent oxygen utilization (AOU) was calculated as the difference between saturation concentrations of O2 and measured O2 concentrations. The saturation concentration of O2 was calculated after the Weiss methods (Weiss, 1970) using the R package marelac (Soataert et al., 2016), taking into account salinity and temperature.

Aboard the ship, the Mn cartridges and Mn fibers were washed with Ra-free tap water to remove any residual sea salt and particles. Ra was removed from the tap water by slowly (<1 L min-1) passing it through a Mn-fiber-filled cartridge. Afterwards, both cartridges and fibers were partially dried with filtered compressed air to remove excess water. The samples were analyzed for 223Ra, 224Ra, and 228Th using a radium delayed coincidence counting (RaDeCC) system (Moore and Arnold, 1996). For the efficiency calibration of the RaDeCC, 227Ac and 232Th standard solutions were used, and the calibration followed the procedure described in Scholten et al. (2010) and Moore and Cai (2013). Counting errors were propagated following Garcia-Solsona et al. (2008). Excess 224Ra (224Raex), i.e., the 224Ra activity corrected for 228Th-supported 224Ra, was calculated by subtracting the 228Th activity from the 224Ra activity. The 228Th activity was measured 3 weeks after the first measurement of 224Ra, when the initial 224Ra had decayed. As we measured only the first Mn cartridge and the Mn cartridges do not adsorb radium quantitatively, we report here only 224Raex/223Ra ratios.

In order to advance understanding of the role of benthic Fe supply to the productive surface waters of the upwelling region, vertical diffusive fluxes (Eq. 1: left term, right-hand side) and wind-induced vertical advective fluxes (Eq. 1: right term, right-hand side) were estimated. On the continental margin below the surface mixed layer, solutes are transferred vertically toward the near-surface layers by turbulent mixing processes and by vertical advection forced by Ekman divergence (e.g., Kock et al., 2012; Milne et al., 2017; Rhein et al., 2010; Steinfeldt et al., 2015; Tanhua and Liu, 2015): 1Jz=Kz∂[TM]∂z+w⋅Δ[TM]. Here, Kz is the turbulent eddy diffusivity (m2 s-1), ∂[TM]/∂z the vertical gradient with depth (z) of the TM concentration [TM] (µmol m-4), Δ[TM] is a TM concentration difference (µmol m-3), and w represents vertical velocity (m s-1). Average advective and diffusive TM fluxes were calculated for a depth interval from the shallow O2-depleted waters to surface waters. The exact depth interval varied for each station (see Table S2) due to differences in the depths where TM samples were collected. The upper depth (8–29 m) was always in layers with enhanced chlorophyll a fluorescence, although for some stations the upper depth was below the surface mixed layer.

Diffusive Fe fluxes were determined by combining TM concentration measurements from the TM-CTD stations with nearby measured microstructure profiles. The microstructure measurements were performed with an MSS90-D profiler (S/N 32, Sea & Sun Technology). The loosely tethered profiler was optimized to sink at a rate of 0.55 m s-1 and equipped with three shear sensors, a fast-response temperature sensor, an acceleration sensor, two tilt sensors, and conductivity, temperature, depth sensors sampling with a lower response time. At TM-CTD stations with bottom depths less than 400 m, 18 to 65 microstructure profiles were available at each station. At deeper stations, 5 to 12 profiles were used. Standard processing procedures were used to determine the rate of kinetic energy dissipation (ε) of turbulence in the water column (see Schafstall et al., 2010, for detailed description). Subsequently, Kz values were determined from Kρ=ΓεN-2 (Osborn, 1980), where N is stratification and Γ is the mixing efficiency for which a value of 0.2 was used. The use of this value has recently been shown to yield good agreement between turbulent eddy diffusivities determined from microstructure measurements and from tracer release experiments performed in our study region (Köllner et al., 2016). The 95 % confidence intervals for station-averaged Kρ values were determined from Gaussian error propagation following Schafstall et al. (2010). Finally, diffusive fluxes were estimated by multiplying station-averaged Kρ with the vertical gradient of the respective TM solute, implicitly assuming Kz=Kρ.

The vertical advective flux by Ekman divergence requires determination of vertical velocity in the water column that varies with depth and distance from the coastline. Convincing agreement between vertical velocities derived from Ekman divergence following Gill (1982) determined from scatterometer winds and from helium isotope disequilibrium within the Mauritanian and Peruvian coastal upwelling regions was found by Steinfeldt et al. (2015) (see their Fig. 4). In their study, vertical velocities were parameterized as (Gill, 1982) 2w=τyρfLre-x/Lr, where τy represents the alongshore wind stress, ρ the density of sea water, x the distance from maximum Ekman divergence taken here as the position at 50 m bottom depth on the shelf, and Lr the first baroclinic Rossby radius. The parameterization results from considering the baroclinic response of winds parallel to a coastline in a two-layer ocean (Gill, 1982). The baroclinic Rossby radius Lr=f-1gρ2-ρ1ρH1H2H1+H2 (ρ1/2 and H1/2 are density and thickness of the surface and lower layer, respectively) was found to be 15 km from hydrographical data collected during the cruise. Similar values were determined by Steinfeld et al. (2015) in the same region. Using average alongshore wind stress from satellite data (0.057 Nm-2, determined from daily winds from Remote Sensing Systems ASCAT C-2015, version v02.1 (Ricciardulli and Wentz, 2016) at 18∘22.5′ N, 016∘7.5′ W using τy=ρairCdv2, where v represents alongshore wind, Cd is drag coefficient for which 1.15×10-3 was used (e.g., Fairall et al., 2003), and ρair is density of air) for June 2014, maximum vertical velocities of 3.7×10-5 m s-1 were determined for the shelf region (50 m water depth), which decayed offshore to 1.7×10-6 m s-1 at the position of the 1000 m isobath at 18∘ N. As these vertical velocities describe the magnitude of upwelling at the base of the mixed layer, additional corrections need to be considered for deeper depths. Here, we approximated the vertical decay of w as a linear function which diminishes at the ocean floor.

The calculation of the vertical advective flux supplying solutes from the shallow O2-depleted waters to surface waters requires knowledge of a concentration difference Δ[TM] associated with the upwelling flux. Ideally, the vertical length scale over which the concentration difference is determined can be diagnosed as the TM concentration variance divided by its mean vertical gradient (e.g., Hayes et al., 1991). However, in our study TM concentration time series data are not available. Previous studies have used a vertical length scale of 20 m to calculate the concentration differences between the target depth and the water below (e.g., Hayes et al., 1991; Steinfeldt et al., 2015; Tanhua and Liu, 2015). For our calculations, we chose to use a smaller length scale of 10 m following Hayes et al. (1991), which results in vertical advective TM flux presumably on the lower side of possible values.

All figures were produced in R (version 3.4.3). Data gridding in Figs. 2 and 3 was performed using the Tps function within the fields package in R (Nychka et al., 2016).

The cruise was conducted in June 2014 along a transect crossing a narrow shelf off the Mauritanian coast at 18∘20′ N. The vertical structure of the OMZ in this region is characterized by a deep OMZ at about 400 m depth and a shallow OMZ at about 100 m depth (Brandt et al., 2015). Coastal upwelling of nutrient-rich deep water occurs as a result of offshore transport of surface waters caused by a northeast trade wind component parallel to the coast. While north of 20∘ N upwelling persists throughout the year, upwelling south of 20∘ N, including the Mauritanian upwelling region, undergoes seasonal changes in upwelling strength (Barton et al., 1998), with strongest upwelling occurring between December and April. The seasonal variability is mainly driven by changes in wind forcing associated with the migration of the Intertropical Convergence Zone (Lathuilère et al., 2008). During the cruise period, cold upwelled waters with temperature less than 20 ∘C were still present on the shelf and upper continental slope (Thomsen et al., 2019, their Fig. 1) indicating active upwelling.

The eastern boundary circulation consists of the Mauritania Current (MC, Fig. 1) flowing poleward at the surface against the equatorward winds and of the poleward undercurrent (PUC) flowing in the same direction at depths between 50 and 300 m (Barton, 1989; Klenz et al., 2018; Mittelstaedt, 1983; Peña-Izquierdo et al., 2015). Both currents supply cold, O2- and nutrient-rich waters of predominantly South Atlantic origin (South Atlantic Central Water, SACW) to the coastal upwelling region (e.g., Mittelstaedt, 1991, 1983; Peña-Izquierdo et al., 2015). In response to the changing winds, the eastern boundary circulation likewise exhibits a pronounced seasonal variability (Klenz et al., 2018; Stramma et al., 2008a). The strongest poleward flow is observed during the relaxation period between May and July when alongshore, upwelling-favorable winds weaken but wind stress curl is at its maximum (Klenz et al., 2018). During the upwelling season in boreal winter, the circulation more closely resembles the classical eastern boundary circulation regime, with a weak PUC flowing beneath an equatorward coastal jet (Klenz et al., 2018; Kounta et al., 2018). At deeper levels (300–500 m depth), flow was found to be equatorward during both seasons. The shallow (<300 m depth) boundary circulations turn offshore at the southern flank of the Cape Verde frontal zone (CVFZ) (e.g., Tomczak, 1981; Zenk et al., 1991) at about 20∘ N, separating SACW from more saline and O2-rich central waters formed in the North Atlantic (NACW). The circulation in June 2014 was typical for a relaxation period characterized by strong poleward flow over the shelf and the upper continental slope between the surface and 250 m depth (Klenz et al., 2018; Thomsen et al., 2019). During the later parts of the cruise, the core of the MC moved offshore and reduced poleward flow was observed near the shelf break. Periods of elevated northward flow on the Mauritanian shelf inhibits the onshore near-bottom supply of low oxygen but nitrate-rich waters onto the shelf with consequences for benthic nitrogen cycling (Yücel et al., 2015).

Meridional sections of water mass properties and O2 concentrations from around 18∘ N showed that waters with an enhanced SACW proportion advected from the south as well as NACW coming from the north have higher O2 concentrations than the ambient waters (Klenz et al., 2018). The mixture of SACW and NACW waters found in the thermocline particularly during boreal winter, previously identified as a regional water mass and termed the Cape Verde SACW (SACWcv) by Peña-Izquierdo et al. (2015), is a signature of an older water mass with lower O2 concentrations than those of SACW or NACW due to a longer residence time and O2 consumption through remineralization. Elevated pelagic oxygen consumption levels at the Mauritanian continental margin were recently determined by Thomsen et al. (2019). During the transition period in May through to July, upper central waters (50–300 m depth) are dominated by SACW accounting for 80 %–90 % of the water masses in the boundary current region (Klenz et al., 2018).

The SACW transported poleward within the boundary circulation is supplied by the zonal North Equatorial Counter Current (NECC) and North Equatorial Undercurrent (NEUC), which flow eastward at about 5∘ N (Brandt et al., 2015) before diverging into a northward and a southward flowing branch in front of the African coast.

As a result of interactions between tidal currents, topography, and critically sloping upper continental slope topography (e.g., Eriksen, 1982), the Mauritanian upwelling region is known for elevated nonlinear internal wave activity resulting in enhanced mixing in the water column of the upper slope and shelf region (Schafstall et al., 2010). Vertical fluxes of nutrients driven by mixing processes are amongst the largest reported in literature, although lower than in the Celtic Sea (Tweddle et al., 2013) and the lower St. Lawrence estuary (Cyr et al., 2015).

The CTD and microstructure deployments were performed along the east–west transect in the period 8 to 27 June (2014) (Fig. 1). Oxygen concentrations reached a deep minimum of 40–50 µmol kg-1 at about 400 m and a shallow minimum of 30–50 µmol kg-1 at about 50–100 m (Fig. 2), which is in agreement with previous studies (Brandt et al., 2015; Thomsen et al., 2019). Mixed layer depths ranged from 10 to 22 m during the cruise. Salinity was highest at the surface (ca. 36.02) and generally decreased with depth to a minimum of 34.71 at around 1000 m. Nitrate (NO3-) concentrations in the surface mixed layer varied between 0.1 and 11.3 µmol L-1 and phosphate (PO43-) between 0.15 and 0.91 µmol L-1. NO3- and PO43- concentrations increased with depth to a maximum of 47.6 and 3.2 µmol L-1, respectively (Fig. 2).

Over a time period of 19 d, two trace-metal stations along the transect at water depths of 170 m (18.23∘ N, 16.52∘ W; first deployment: 12 June; second deployment: 21 June) and 189–238 m (18.22∘ N, 16.55∘ W; first deployment: 24 June; second deployment: 26 June) were reoccupied. Minimum O2 concentrations of 30 µmol kg-1 observed before 15 June increased to 50 µmol kg-1 after 19 or 24 June, depending on the location. This oxygenation event, captured in ocean glider measurements, is discussed in detail by Thomsen et al. (2019). Variability in oxygen concentrations observed further offshore was attributed to physical transport of SACW into the region (Thomsen et al., 2019). In contrast, closer to the coast, enhanced pelagic oxygen consumption rates were determined that significantly contribute to the variability in observed oxygen concentrations (Thomsen et al., 2019). Short-term variability in oxygen concentrations has also been observed further south in nearshore Senegalese waters where an anoxic event was likely attributed to the offshore advection of a decaying diatom bloom (Machu et al., 2019).

The sediments in the study area contain a large amount of carbonate, biogenic silica, and quartz (Hartman et al., 1976). The fraction of sand and mud varies largely depending on bottom depth, with sand comprising between 7 % and 70 % of the dry weight (Dale et al., 2014). The particulate organic carbon (POC) content varies between 0.55 wt % at shallow depth (66 and 90 m) and increases to 3.3 wt % at 1108 m depth (Schroller-Lomnitz et al., 2019). A more detailed description of the sediments underlying our study region and sediment parameters collected on the same cruise, including Fe(II) concentrations and Fe/Al ratios, are given in Schroller-Lomnitz et al. (2019).

Dissolved Fe and LpFe concentrations ranged between 0.97–18.5 and 1.6–351 nmol L-1, respectively (Fig. 3a, b). Surface waters (5–29 m) had lowest dFe (0.97–4.7 nmol L-1) and LpFe (1.6–35.9 nmol L-1) concentrations, whereas highest concentrations were present on the shelf close to the seafloor (up to 18.5 nmol L-1 dFe and 351 nmol L-1 LpFe). Enhanced concentrations of both Fe fractions at any given station were observed at depths with low O2 concentrations (30–60 µmol O2 kg-1). A similar distribution pattern was observed for dCo, with concentrations between 0.069 and 0.185 nmol L-1 (Fig. 3c). In contrast, LpCo concentrations varied from below the limit of quantification (LOQ) up to 0.179 nmol L-1 and were generally highest in surface waters and close to the coast (Fig. 3d). Compared to dFe, the concentration range of dCo was much narrower and enhanced concentrations were observed over a broader depth range and further offshore.

Surface dFe and dCo concentrations were low, presumably due to enhanced biological uptake. No clear increasing trend in dFe and dCo with depth was observed, indicating that processes other than, or in addition to, remineralization influenced their distributions. Elevated concentrations were found close to the sediments and within low-O2 waters. This suggested a benthic source of Fe and Co under O2-depleted conditions and offshore transport along O2-depleted water filaments, which is in agreement with previous studies (e.g., Baars and Croot, 2015; Hatta et al., 2015; Noble et al., 2012). Our sharper onshore–offshore gradient of dFe concentrations compared to dCo in O2-depleted waters shows that oxidation and removal mechanisms or scavenging rates were faster for Fe than Co (Noble et al., 2012). Previously reported dFe concentrations in coastal regions of the tropical North Atlantic were lower than we observed, between 0.5 and 6.3 nmol L-1 (Hatta et al., 2015; Milne et al., 2017; Wuttig et al., 2013). However, all these samples were collected at a greater distance from the coast. In the near-coastal Oregon and Washington shelf bottom water dFe concentrations were similar to our study under equivalent O2 concentrations (18.7–42.4 nmol L-1 dFe, 42–61 µmol kg-1 O2; Lohan and Bruland, 2008), whereas in the euxinic waters from the Peruvian shelf region, dFe concentrations were more than an order of magnitude higher, exceeding 200 to 300 nmol L-1 (Schlosser et al., 2018; Scholz et al., 2016). Similar dCo concentrations to our study were observed in the North Atlantic and South Atlantic, with highest concentrations of ∼0.16 nmol L-1 present within O2-depleted waters (Noble et al., 2012, 2017).

Dissolved Mn concentrations ranged between 0.46–13.8 nmol L-1 and LpMn from below the LOQ to 4.4 nmol L-1 (Fig. 3e, f). Highest dMn and LpMn concentrations were observed in surface waters, generally decreasing with depth. Additionally, concentrations were highest on the shelf and decreased offshore. The dMn concentrations were generally elevated within and below the deeper O2-depleted waters with 0.70–1.34 compared to 0.46–0.91 nmol L-1 just above. The increased dMn concentrations within the deeper O2-depleted waters (∼ 350–500 m depth) indicate a benthic source, similar to Fe and Co, which is in accordance with previous studies (Noble et al., 2012). However, in the shallow O2-depleted waters (∼ 50–200 m depth), this effect is not resolvable due to high surface concentrations, which were maintained by photo-reduction of Mn oxides to soluble Mn(II) that prevents loss of Mn from solution (Sunda and Huntsman, 1994). Reported dMn concentrations in the North Atlantic and South Atlantic were lower than in our study, with concentrations <3.5 nmol L-1 in surface waters and around 0.5–1 nmol L-1 dMn within the OMZ (Hatta et al., 2015; Noble et al., 2012; Wuttig et al., 2013). As for dFe, these lower reported values can also be explained by sampling stations positioned at further distance from the coast and removal of dMn via biological oxidation processes with distance from the source (Moffett and Ho, 1996).

Dissolved Cd and Ni concentrations were lowest in surface waters with 0.022–0.032 nmol Cd L-1 and 2.6–2.8 nmol Ni L-1 and showed an increasing trend with depth to maximum values of 0.60 and 5.8 nmol L-1, respectively (Fig. 3g, m). Leachable particulate Cd concentrations were from below the LOQ to 0.20 nmol L-1, and LpNi concentrations from below the LOQ to 1.7 nmol L-1. A large fraction of Ni (72 %–100 %) was present in the dissolved form. The majority of LpNi samples were below the LOQ (>70 % of the data) and LpNi is therefore not included in Fig. 3. LpCd concentrations were highest close to the coast and decreased offshore (Fig. 3h). In surface waters close to the coast the LpCd fraction was dominant with up to 84.3 % of the entire Cd pool (d + Lp). The fraction of LpCd in surface water beyond the shelf break (including stations 2, 1, and 9) contributed still up to 54.3 % of the Cd pool, whereas below 50 m only 0 %–12.8 % of TDCd was in the Lp phase beyond the shelf break. In contrast to Fe, Co, and Mn, no increases in Cd and Ni were observed near the seafloor and within the O2-depleted waters indicating that Cd and Ni concentrations are mainly controlled by remineralization of sinking organic matter, which is typical for these two nutrient-like TMs (Biller and Bruland, 2013). Similar distributions with concentrations between 0 and 1000 m water depth ranging from ∼ 2 to 5.5 and from ∼ 0 to 0.55 nmol L-1 for dNi and dCd, respectively, were observed during the GEOTRACES transect GA03_w in the tropical North Atlantic (Mawji et al., 2015; Schlitzer et al., 2018).

Dissolved Cu concentrations in surface waters ranged between 0.63 and 0.81 nmol L-1 (Fig. 3i). Concentrations increased with depth to around 1.37 nmol L-1 at 700 m depth close to the seafloor, whereas highest observed concentrations further offshore were 0.95 nmol L-1 at the greatest sampled depth of 850 m. These results indicate that in addition to remineralization processes of sinking biogenic particles, the distribution of Cu is influenced by inputs from the seafloor. This is in accordance with previous studies, suggesting that Cu is released from continental shelf sediments under oxic and moderately reducing conditions (Biller and Bruland, 2013; Heggie, 1982), whereas no increase in Cu concentrations near the seafloor was observed at low bottom water O2 concentrations (O2 < 10 µmol L-1; Johnson et al., 1988). A decrease in Cu concentrations in the bottom boundary layer was also reported with a seasonal decrease in O2 in summer from a minimum of 70 µmol L-1 O2 in May to 40 µmol L-1 O2 in August, suggesting a decrease in sedimentary release of Cu (Biller and Bruland, 2013). In strongly reducing sediments and the presence of H2S, Cu forms inorganic sulfides and precipitates, which may explain reduced sedimentary Cu release under low bottom water O2 concentrations (Biller and Bruland, 2013). Therefore, the sediment source of dCu might show a different dependency on bottom water O2 concentrations than dFe, dCo, and dMn explaining the distinct distribution of dCu. Concentrations of LpCu were from below the LOQ to 0.61 nmol L-1 with enhanced levels at station 4 close to the coast and at mid-depths of the three stations furthest offshore (9, 5, and 2) (Fig. 3j).

Observed dPb concentrations were lowest in the surface waters at 9–14 pmol L-1 and increased with depth to 29–86 pmol L-1 below 600 m depth (Fig. 3k). Lead is not considered a nutrient-like TM (e.g., Boyle et al., 2014), but our observations indicate a release of Pb from sinking particles following remineralization. The concentration range and depth distribution is similar to reported distributions further offshore at about 21∘ W (Noble et al., 2015). These authors suggested that increased concentrations of up to 70 pmol L-1 between 600 and 800 m depth were related to the influence of Mediterranean Outflow Waters (MOW). Additionally, increased Pb concentrations in proximity to sediments have been attributed to the benthic release of historic Pb through reversible scavenging from particles and the release of dPb associated with Fe/Mn oxyhydroxides during reductive dissolution of those oxides in anoxic sediments (Rusiecka et al., 2018). The major source of Pb to the ocean is atmospheric dust deposition from anthropogenic emissions (Bridgestock et al., 2016; Nriagu and Pacyna, 1988; Véron et al., 1994) with a recent indication of reduced anthropogenic Pb inputs to surface waters in the eastern tropical Atlantic under the north African dust plume (Bridgestock et al., 2016). Low surface-water concentrations on the Mauritanian shelf indicate low atmospheric inputs of Pb to this region. LpPb was from below the LOQ to 27 pmol L-1 and the distribution of LpPb was similar to that of LpFe, with subsurface maxima within O2-depleted waters (Fig. 3l), and this may indicate increased scavenging of dPb in these layers, which might be associated with Fe-containing particles.

In general, sediment-derived TM concentrations decrease with distance from the shelf and with time that passed since the water mass has been in contact with the sediments due to water mass mixing and removal processes such as precipitation and scavenging (Bruland and Lohan, 2006). Radium isotopes can be used as a tracer for benthic sources. The major source of Ra to the ocean is input from sediments through the efflux of pore water, sediment resuspension, and submarine groundwater discharge (Moore, 1987; Moore and Arnold, 1996; Rama and Moore, 1996). Due to the distinctive half-lives of the different Ra isotopes (e.g., 224Ra (t1/2=3.66 d) and 223Ra (t1/2=11.4 d)) and their conservative behavior in seawater, it is possible to quantify the time that has passed since a parcel of water was in contact with the sediments using the following equation by Moore (2000): 3A224A223obs=A224A223ie-λ224τe-λ223τ and solved for water mass age (τ) 4τ=ln⁡A224A223obs-ln⁡A224A223iλ223-λ224, where A224/A223 is the activity ratio of 223Ra and 224Ra, with the subscript obs for the observed seawater ratio and the subscript i for the initial groundwater endmember ratio, and λ223 and λ224 are the decay constants in units of per day (d-1) for 223Ra and 224Ra. The ratio 224Ra/223Ra is not affected by dilution assuming there is no mixing with waters having significantly different 224Ra/223Ra ratios.

Highest 224Raex/223Ra activity ratios were observed close to the seafloor (Fig. 3n). The average 224Raex/223Ra ratio in proximity to the sediment source (<2.60 m above seafloor) was 4.9±1.5 and was similar to the reported ratios for shelf waters off South Carolina (224Raex/223Ra = 4.4±1.5; Moore, 2000). The 224Raex/223Ra ratios decreased away from their benthic source due to decay (224Raex/223Ra = 0–0.5 in surface waters). Ratios close to the seafloor were relatively constant along the transect at bottom depths <600 m, whereas dFe, dCo, and dMn concentrations varied largely in the bottom samples. This suggests that factors, which are not influencing the Ra distribution, impacted the distributions of dFe, dCo, and dMn, with a likely influence of enhanced O2 concentrations reducing sediment release or increasing removal rates of these metals at water depths between 200 and 400 m. At around 800 m bottom depth, 224Raex/223Ra ratios were slightly elevated and coincided with increased dCo, dFe, dMn, and dCu concentrations despite O2 concentrations >70 µmol kg-1. This suggests that the enhanced TM concentrations at this location were influenced by a strong sediment source which may be related to the presence of a benthic nepheloid layer as indicated by an increase in turbidity in proximity to the seafloor. An elevated 224Raex/223Ra ratio of 3.5±0.6 was observed at about 16.65∘ N and 80 m water depth (bottom depth 782 m) and coincided with a local maximum of dFe, dMn, and dCo and reduced O2 concentrations. These observations indicate that the waters with the local maximum of dFe, dMn, and dCo have been in relatively recent contact (12–20 d assuming initial pore water 224Raex/223Ra ratios between 18 and 38; Moore, 2007) with sediments, likely originated from south of our transect as a result of a strong poleward flow (Klenz et al., 2018), and the dynamic current system in this region can cause local and short-term variability in the transport of sediment-derived TMs.

Principal component analysis (PCA) was performed (using the RDA function within the vegan package in R; Oksanen et al., 2018) to investigate different groups and correlations in the data set. Dissolved TMs (Fe, Mn, Co, Ni, Pb, Cu, and Cd), nutrients (silicic acid, nitrate, and phosphate), dissolved O2, apparent oxygen utilization (AOU), depth, and iodide concentrations (Supplement Fig. S1) were utilized in the PCA. Radium data were not included in the PCA as the number of available data points for 224Raex/223Ra was much lower than for the other parameters. Surface waters shallower than 50 m were excluded from the PCA to remove the influence of localized atmospheric deposition and photochemical processes, which in particular influence Mn and iodide distributions. The PCA generated three principal components (PC) with eigenvalues larger than 1, with PC1 explaining 53.6 % and PC2 25.5 % of the total variance in the data set (together 79.1 %). Inclusion of PC3 in the analysis explained only 6.8 % more of the variance.

The first PC group is formed by dCd, dCu, dNi, and dPb (Fig. 4), which are associated with depth, AOU, nitrate, and phosphate. This indicates that the distributions of Cd, Cu, Ni, and potentially Pb are controlled by organic matter remineralization processes. This is in agreement with strong Pearson correlations, R>0.9, for the relationships of dCd and dNi with depth, nitrate, and silicic acid (Supplement Table S1). Weaker correlations with major nutrients were observed for dPb (R>0.6) and dCu (R>0.4), potentially due to additional remineralization or removal mechanisms for these elements (e.g., prior atmospheric inputs and water mass transport, Pb; sediments, Cu and Pb; and scavenging). The second group of TMs is composed of dFe, dCo, and dMn that are associated with elevated iodide and turbidity, and low dissolved O2 (Fig. 4). Iodide (I-) is the reduced form of iodine (I), which is typically present as iodate (IO3-) in oxygenated subsurface water. Both I forms are present as soluble anions in seawater. Due to a relatively high redox potential (pE ∼ 10), iodine is one of the first redox-sensitive elements to undergo reduction under suboxic conditions, and is therefore a useful indicator for active reductive processes (Rue et al., 1997). Despite their role as micronutrients, Fe, Mn, and Co do not correlate with nutrients indicating that processes other than remineralization controlled their distributions.

The anticorrelation with O2 (also shown in Fig. S2) and correlation with iodide support the notion that Fe, Co, and Mn distributions were strongly influenced by water column O2 concentrations, presumably through (i) enhanced benthic metal fluxes from anoxic sediments and (ii) decreased oxidation rates in the overlying water column under O2-depleted conditions. This is also supported by elevated benthic Fe(II) fluxes observed at the seafloor within the shallow OMZ, with benthic fluxes of 15–27 µmol m-2 d-1 (Schroller-Lomnitz et al., 2019).

Variability in the redox-sensitive metals, Fe, Mn, and Co, was not fully explained by either O2 or iodide concentrations; Pearson correlations with O2 were -0.55, -0.61, and -0.58, respectively (Supplement Table S1). As shown before, other factors such as water mass mixing and age, the amount and type of particles present, and remineralization all likely impact their dissolved concentrations. Consequently, such a complex chain of factors and processes means that one variable alone is unlikely to explain the behavior of Fe, Mn, and Co.

The main sources of TMs in our study region are sedimentary release and atmospheric dust deposition (e.g., Rijkenberg et al., 2012). Also, release of TMs via organic matter remineralization may have an important influence on the distribution of TMs. In the following, we discuss the relative influence of remineralization, atmospheric dust deposition, and sedimentary release on the supply of Fe, Co, and Mn to surface waters.

Particles in the water column can comprise either a source or a sink of dissolved TMs. In the top 50 m of the water column a large part of the LpTMs may be part of living biological cells (e.g., phytoplankton) or organic detritus and can enter the dissolved TM pool by remineralization (Bruland and Lohan, 2006). Additionally, LpTMs may be part of lithogenic phases from Saharan dust and sediment particles, or authigenic phases. Authigenic phases are formed in situ by TM adsorption onto particle surfaces or by the formation of amorphous TM oxides and hydroxides (e.g., FeO(OH) in the mineral structure of goethite) (Sherrell and Boyle, 1992), processes referred to as scavenging. The extent of scavenging processes is largely influenced by the amount and type of particles present (Balistrieri et al., 1981; Honeyman et al., 1988).

Iron was mainly present in the size fraction >0.2 µm with TDFe concentrations being 0.44–44.5 times higher than dFe (<0.2 µm) (Fig. 6a). To investigate the influence of particle load on the distribution between dissolved and particulate phases, the fraction of Lp (Lp/TD) TMs and Lp concentrations are plotted against turbidity for Fe, Co, and Mn (Fig. 6b, c). A small fraction of LpFe of around 60 % was observed at lowest turbidity. As turbidity increases from 0.1 to 0.2 NTU, the LpFe fraction increased to >90 %. This suggests that the fraction of LpFe is tightly coupled to the particle load. Iron adsorption onto particles has been demonstrated to be reversible with a constant exchange between dissolved and particulate fractions (Abadie et al., 2017; Fitzsimmons et al., 2017; John and Adkins, 2012; Labatut et al., 2014). Furthermore, offshore transport of acid-labile Fe particles formed by scavenging (oxidation and adsorption) of dissolved Fe originating from a benthic source was observed in the North Pacific (Lam and Bishop, 2008) and may contribute to the bioavailable Fe pool. Therefore, an important fraction of Fe may be transported offshore, adsorbed onto particles, and can enter the dissolved pool by cycling between dissolved and particulate phases.

The LpCo fraction ranged between 0 % and 75 %, and the fraction and concentration of LpCo showed linear increases with turbidity, indicating an influence of particle load on Co size fractionation, similar to Fe. In contrast to Fe and Co, the fraction of LpMn varied between 3 % and 40 % and did not show a correlation with turbidity, whereas LpMn concentrations showed an increase with turbidity. This indicates that an increased presence of particles coincided with enhanced LpMn levels, but that the particle load did not substantially influence the distribution between dMn and LpMn phases and particles therefore did not contribute to the dMn fraction. This suggests that particles did not play a major role in transport of dMn, which agrees with a study on hydrothermal vent plumes, where the distribution of the dMn plume was decoupled from the distribution of the particulate Mn plume (Fitzsimmons et al., 2017).

The increase in LpFe concentrations with increasing turbidity was weaker in the surface waters compared to water depths below 50 m (Fig. 6c). This suggests a large additional LpFe source at depth with either a higher Fe content of particles or the presence of different sizes of particles causing different responses in turbidity measurements. The large additional LpFe source at depth is likely associated with benthic dFe inputs, with a subsequent transfer to the particulate phase by scavenging. Enhanced turbidity at depth may also indicate sediment resuspension, which would result in the release of TM-containing particles from sediments and enhanced release of dTMs from sediment pore water. The effect of sediment resuspension is discussed in more detail below (Sect. 3.6.2).

In contrast to Fe, the increase in LpCo and LpMn concentrations with turbidity was similar in surface waters and below and suggests less variability in the composition of the particulate Co and Mn phase throughout the water column with a potentially weaker influence of sediment release on the distribution of particulate Mn and Co. A weaker influence of sediment release might be influenced by a weaker release of Co and Mn from sediments in the dissolved form and slower oxidation rates compared to Fe, in particular for Co (Noble et al., 2012), resulting in a slower conversion into the particulate phase. Such an interpretation based on turbidity data alone, however, is very hypothetical and would require further investigation of particulate TM species composition in this area.

Large temporal changes in O2, turbidity, and redox-sensitive TMs were observed within a short timescale of a few days at two repeat stations, station 3A/3B and station 8A/8B (Figs. 7 and S5).

Stations 3 and 8 were sampled twice with a period of 9 d between both deployments for station 3 (Fig. 7a) and 2 d for station 8 (Fig. 7b). At station 3, O2 concentrations in the upper 50 m were very similar between both deployments, whereas below 50 m O2 increased from 30 µmol kg-1 during the first deployment to 50 µmol kg-1 9 d later. At the same time, turbidity below 50 m had decreased from 0.35 to below 0.2, and dFe concentrations from a maximum of 10 to 5 nmol L-1 9 d later. In addition, dMn and dCo concentrations decreased from 5 to 3 and from 0.14 to 0.12 nmol L-1, respectively. Particularly large changes were also observed for LpTM concentrations with a decrease from 147–322 to 31–51 nmol L-1 for LpFe, from 0.066–0.114 to 0.015–0.031 nmol L-1 for LpCo and from 1.24–2.64 to 0.16–0.54 nmol L-1 for LpMn. In contrast, no changes in temperature and salinity of the water parcel occurred below 50 m (Fig. 7a). Similar changes in O2 and turbidity were observed at station 8. During the first deployment a local minimum in O2 below 30 µmol kg-1 was present between 105 and 120 m water depths which coincided with a maximum in turbidity of 0.4 (Fig. 7b). In contrast O2 concentrations and turbidity during the second deployment were relatively constant (50–60 µmol kg-1 O2 and turbidity 0.2) below 50 m. At the depth of the local O2 minimum and turbidity maximum, concentrations of dFe, dMn, and dCo were elevated during the first deployment with concentrations of 9.4±2.1 nmol dFe L-1, 3.7±0.6 nmol dMn L-1, and 0.145±0.033 nmol dCo L-1 in comparison to 4.6±1.0 nmol dFe L-1, 2.6±0.5 nmol dMn L-1, and 0.122±0.028 nmol dCo L-1 at similar depth during the second deployment.