Water column samples were collected using 10 L externally closing, Teflon-lined Niskin-1010X bottles deployed on an autonomous 1018 intelligent rosette system (TMR – trace metal rosette, specially adapted for trace metal work; General Oceanics Inc.). The polyurethane-powder-coated aluminium rosette frame was suspended on Kevlar rope which passed through a clean block with a plastic sheave (General Oceanics) and was lowered to a maximum depth of 1300 m. Bottles were tripped at preprogrammed depths using a pressure sensor as the TMR was being raised through the water column at approximately 0.5 m s-1.

All sample processing was carried out under an ISO class-5 trace metal clean laminar flow bench in a HEPA filtered-air clean container, with all materials used for sample handling thoroughly acid-washed. Samples were drawn through C-Flex tubing (Cole Parmer) and filtered in-line through 0.2 µm pore size acid-washed capsules (Pall Supor membrane Acropak 200 or Sartorius Sartobran 300 filters). The dissolved fraction is thus likely to contain colloids and small particles < 0.2 µm in diameter (Bowie and Lohan, 2009). All transfer tubes, filtering devices and sample containers were rinsed liberally with sample before final collection in 125 mL Nalgene LDPE bottles. Seawater samples were acidified within 24 h of collection using 2 mL of concentrated ultrapure hydrochloric acid (HCl, Seastar BASELINE grade) per litre of sample, resulting in an approximate final pH of 1.8, double bagged and stored for at least 24 h at ambient temperature until analysis.

Suspended particles for trace elemental analysis were collected using 11 large-volume in situ pumps (McLane Research Laboratories WTS6-1-142LV and Challenger Oceanics pumps), suspended simultaneously at prechosen depths following methods reported in Bowie et al. (2009). Up to 2000 L of seawater was filtered across a 142 mm diameter stack (134 mm diameter active area) consisting of a 53 µm nylon pre-filter screen (NYTEX) followed by a QMA quartz fibre filter (1 µm nominal pore size; Sartorius). The QMA filter was supported by a 350 µm polyester mesh, which was placed on top of the Teflon PFA grid of the pump housing. Prior to use, NYTEX screens were conditioned by soaking in 5 % H2SO4, rinsed 3× with Milli-Q-grade water, dried at ambient temperature under a laminar flow hood and stored in clean plastic Ziploc^® bags. QMA filters were conditioned for trace metal analysis (precombustion and acid cleaning) following Bowie et al. (2010). Upon recovery of the pumps, subsamples were taken from the QMA filters by using a circular plastic punch (14 mm diameter) and by cutting the nylon mesh using ceramic scissors. Filters were dried under a laminar flow bench and stored at -18 ∘C in acid-washed PCR trays until further analysis in the home laboratory. The 1–53 and > 53 µm size fractions were digested and analysed separately, and the particulate iron (pFe) reported here is the sum of both fractions. The ISPs were shown to be efficient in capturing large (> 53 µm) particles (Planchon et al., 2015).

Sinking particles for trace elemental analysis were collected using PPS3/3 free-floating sediment traps (Technicap, France), specially adapted for trace metals and deployed at 200 m. Traps were deployed for 5.3, 5.1, 1.9 and 1.5 days at stations E-1, E-3, A3-2 and E-5, respectively. The trap deployed at station R-2 was lost and not recovered. Traps drifted between 10 and 43 km over the course of the deployment. Full details of the trap deployments are given in Laurenceau-Cornec et al. (2015) and Planchon et al. (2015). Samples for trace elemental analysis were collected in three separate acid-washed cups (specifically for trace metals) containing a low trace metal brine solution (salinity ∼ 60), each opened for either 1, 3, 8 or 12 h (depending on the station). Upon recovery, cups were taken to a clean room and particles filtered off-line onto a 47 mm diameter, 2 µm porosity polycarbonate filter under gentle vacuum using a Teflon PFA unit (Savillex Corp., USA) equipped with a 350 µm pre-screen (to exclude zooplankton).

Dissolved Fe (dFe) was determined shipboard by flow injection analysis with chemiluminescence detection (FI-CL) using in-line preconcentration on an 8-hydroxyquinoline chelating resin (adapted from Obata et al., 1993, de Jong et al., 1998 and Sarthou et al., 2003). Dissolved Fe data were quality controlled against the SAFe (Sampling and Analysis of Fe) standard reference materials (Johnson et al., 2007). Full data including certification results and analytical figures of merit are reported in Quéroué et al. (2015).

Particulate Fe (pFe) was determined as follows. Sampled particles were acid extracted in 1 mL concentrated HNO3 (Seastar Baseline) for 12 h on a DigiPREP HP Teflon hotplate supplied with HEPA-filtered air (SCP Science) at 120 ∘C using 15 mL Teflon PFA Savillex vials. Digest solutions were diluted with 10 mL ultra high-purity water to 10 % HNO3 and spiked with 10 ppb indium as internal standard prior to analysis by sector field inductively coupled plasma mass spectrometry (Finnigan ELEMENT 2, Thermo Scientific), following Bowie et al. (2010). Blanks from replicate analysis of filters treated identically to the sample filters but without large volumes of seawater passed through them, were typically 2–3 % and < 1 % of the pFe sample concentrations for the ISP deployments and P-trap deployments, respectively. Recoveries from the analysis of the Community Bureau of Reference plankton certified reference material BCR-414 were excellent, with a 101 % recovery (n=3) for pFe. Full data are reported in van der Merwe et al. (2015).

For particulate organic carbon (POC) and particulate nitrogen (PN) analyses, QMA quartz filters from the ISPs were subsampled in a flow bench using a 14 mm diameter plastic punch and transferred to silver foil cups (Sercon brand p/n SC0037). Samples were also collected from the P-traps for POC and PN analyses (see Laurenceau-Cornec et al., 2015). Samples were treated with a 40 µL aliquot of 2 N HCl to remove carbonates (King et al., 1998), dried at 60 ∘C for 48 h and stored in a desiccator until analysis using a Thermo-Finnigan Flash EA1112 elemental analyzer (using sulfanilamide standards) at the Central Science Laboratory, University of Tasmania. The > 53 µm fraction was treated in the same way at the Vrije Universiteit Brussel after first transferring the material from one fourth of the screen, using pre-filtered seawater, onto 25 mm diameter, 1.0 µm pore size silver membrane filters (Sterlitech, Concord). Blank corrections for the pump samples were estimated from filters prepared identically but not deployed on the ISPs; for the trap samples this was done by re-filtering the pre-filtered seawater. All blank corrections were less than 2 % for all samples. The subsampling introduces uncertainties of 5–10 % from inhomogeneous filter coverage that exceeds the analytical uncertainty of ∼ 1 % in the POC analysis (Trull et al., 2015).

Trace metal clean seawater was collected from the mixed layer (20–40 m) using the TMR, was transferred into acid-washed polycarbonate bottles and 0.2 nmol L-1 (final concentration) of enriched 55Fe as FeCl3 was added (1.83 × 103 Ci mol-1 of specific activity, Perkin Elmer). Bottles were placed at in situ temperature in on-deck incubators continuously fed by surface seawater. Incubations were conducted for 24 h (sunrise to sunrise) at several light intensity levels (75, 45, 25, 16, 4 and 1 % of photosynthetically active radiation; PAR). For stations R-2, A3-1, E-1 and E-3, seawater was prefiltered on a 25 µm mesh size before 55Fe was added. After incubation, 300 mL of seawater was passed through 0.2 µm pore size nitrocellulose filters (47 mm diameter, Nuclepore). To determine intracellular Fe uptake rates, 55Fe not incorporated by cells was removed immediately after filtration using 6 mL of a Ti-citrate-EDTA washing solution for 2 min, then rinsed three times with 5 mL of 0.2 µm filtered seawater for 1 min (Hudson and Morel, 1989; Tang and Morel, 2006). The filters were placed into plastic vials and 10 mL of the scintillation cocktail “Filtercount” (Perkin Elmer) added. Vials were agitated for 24 h before the radioactivity on filters was counted with the Tricarb^® scintillation counter (precision < 10 %). Controls were obtained with 300 mL of microwave-sterilised seawater (750 W for 5 min) incubated and treated the same way. Subsamples for enumeration by flow cytometry were collected from each bottle just before the filtration step. Cells were fixed in glutaraldehyde (1 %) and kept frozen (-80 ∘C) until processing and analysis. Data were corrected by blank subtraction and Fe uptake rates normalised to the concentration of Fe in each incubation (in situ dFe and 55Fe added). Further details are given in Fourquez et al. (2015).

Since iron regeneration was not measured directly by experiment during KEOPS-2, we used the following approach to calculate iron regeneration fluxes. Bacterial Fe regeneration was estimated from bacterial turnover times determined from bacterial production and biomass (Christaki et al., 2014), assuming all loss of bacterial biomass through viral lysis and flagellate grazing resulted in the regeneration of Fe (Strzepek et al., 2005) and using a bacterial iron quota of 7.5 µmol Fe (mol C)-1 (Tortell et al. 1996). The mesozooplankton grazing contribution to Fe regeneration was assumed to be equal to the experimentally determined Fe regeneration during KEOPS-1 (Sarthou et al., 2008). The regeneration rates per mesozooplankton individual determined in Sarthou et al. (2008), were then multiplied by mesozooplankton abundance, calculated from the number of cells captured in a daily haul over 200 m during KEOPS-2 (Carlotti et al., 2015; values reported in Table 6 in Laurenceau-Cornec et al., 2015).

In the upper 100 m, we observed a salinity minimum (33.8) and temperature maximum (2.2 ∘C) characteristic of Antarctic Surface Water (AASW) overlying a layer of winter water (WW) at 180–200 m (Tmin of 1.6 ∘C) (Fig. 3a). Deeper in the water column, a Tmax of 2.5∘C at 500 m (associated with an oxygen minimum; not shown) was indicative of upper Circumpolar Deep Water (UCDW) overlying a salinity maximum of 34.8 at 1830 m in lower Circumpolar Deep Water (LCDW). Phytoplankton abundance was low (0.2 µg Chl a L-1; Lasbleiz et al., 2014) and dominated by diatoms, in waters with relatively high surface nitrate concentrations (> 25 µmol L-1; Blain et al., 2015), typical of Southern Ocean HNLC conditions (Lasbleiz et al., 2014).

Dissolved Fe concentrations were very low at the surface (< 0.1 nmol L-1) and increased with depth, averaging 0.3 nmol L-1 in LCDW and broadly tracking the nitrate profile. The pFe profile showed a similar structure to the dFe profile but with surface and deep water concentrations between 0.3 and 1.1 nmol L-1 (the deepest sample was 148 m above the seafloor). The exception was at 500 m, where, interestingly, we observed a dFe and pFe peak of 0.4 and 1.6 nmol L-1, respectively. Whilst this maximum may have arisen due to the enrichment of Fe in UCDW delivered from further south, we hypothesise that the Fe supply may have originated from subsurface sediments of the nearby Leclaire Rise (also known as Skiff Bank; Kieffer et al., 2002), a large seamount which rises to 250 m at 49∘50'S, 65∘00'E (approximately 140 km northwest of station R-2). Similar lithogenic inputs were also observed for other dissolved (Mn; F. Quéroué, personal communication, 2014, data not shown) and particulate (Mn, Al; van der Merwe et al., 2015) trace elements.

The dFe profile at the KEOPS-2 reference station R-2 is similar to the KEOPS-1 reference station C11 (with the exception of the R-2 enrichment in the 200–700 m depth strata; Fig. 4a), but it should be noted that the location of C11 was quite different – in HNLC waters to the southeast of the Kerguelen Plateau (51∘39'S, 78∘00'E) – and we had only 1 dFe data point in UCDW at C11. In contrast to the similarity of the dFe profiles, the pFe profile at C11 was generally lower than at R-2, with mean values through the water column of 0.2 ± 0.14 nmol L-1 (Andrew Bowie, unpublished data) compared to 0.53 ± 0.35 nmol L-1 for station R-2.

Stations A3-1 (Fig. 3b) and A3-2 (Fig. 3c) were in relatively shallow waters on the central plateau, and were impacted by plateau sediments and possibly fluvial and glacial runoff from the basaltic rocks of Heard Island ∼ 300 km upstream (van der Merwe et al., 2015; M. Grenier, personal communication, 2014). A pycnocline was observed at ∼ 190 m, above which the salinity (33.9) and nitrate (∼ 29 µmol L-1) were relatively constant. The mixed layer shoaled (from 165 to 123 m) and increased in temperature (from 1.7 to 2.2 ∘C) between the two visits to A3, consistent with springtime warming of surface waters. We believe that the water masses at A3-1 and A3-2 are comparable since surface waters move slowly in this region (Park et al., 2008, 2014a; Zhou et al., 2014); this was confirmed by REE data which indicated similar waters at both stations marked with fresh continental supplies and only modified by biological processes (M. Grenier, personal communication, 2014).

Surface chlorophyll images revealed that during the 28 days between the first and second visits to A3, a large diatom spring bloom developed mostly dominated by lightly silicified Chaetoceros spp. (surface Chl a increasing from 0.2 µg L-1 at A3-1 to 1.3 µg L-1 at A3-2; Lasbleiz et al., 2014), which likely resulted in the drawdown of dFe (mean mixed layer values decreasing from 0.3–0.4 nmol L-1 at A3-1 to 0.1–0.2 nmol L-1 at A3-2). The peak of biomass had passed by the time we sampled at A3-2, with the bloom starting to fade (Trull et al., 2015). Below the mixed layer, similar dFe profiles were observed during both visits to A3, with expected significant increases at depth towards the plateau floor (e.g. to 1.30 nmol L-1 at 480 m at A3-2; note that, due to operational constraints, there was no dFe data deeper than 340 m at A3-1). Such enrichments at depth were also observed in dissolved Mn and Co profiles (F. Quéroué, personal communication, 2014; data not shown) and dFe profiles from the occupations of station A3 during KEOPS-1 (Fig. 4b), indicative of plateau sedimentary supply.

The pFe profiles at A3 showed a similar structure to the dFe profile, with lower values at the surface (< 10 nmol L-1 at A3-1 and < 4 nmol L-1 at A3-2) and increasing with depth due to enrichment from bottom sediments (up to 33 and 14 nmol L-1 at 440 m at A3-1 and A3-2, respectively), and were on average 10 times greater than dissolved concentrations through the water column. The mixed layer pFe concentrations changed remarkably between the two visits, and the full water column integrated pool was ∼ 70 % lower at A3-2 than at A3-1. Interestingly, this change was also associated with a shift of particles from the 1–53 µm size range to the > 53 µm size range, with the larger class tripling in size (van der Merwe et al., 2015). The development of the large bloom between our two visits to A3, which consisted of a diatom community 50–210 µm in size (Trull et al., 2015), was likely responsible for converting the pFe within the surface mixed layer from the smaller size class to the larger size class. This may have been due to either (i) physical aggregation of the particles onto diatom aggregates and/or (ii) microbially driven conversion of small lithogenic Fe (1–53 µm) to bioavailable forms and incorporation into the large (> 53 µm) diatoms as biogenic Fe, with potentially some fraction of these larger particles exported to depths below the mixed layer, as previously discussed by Lam et al. (2006), Frew et al. (2006) and Planquette et al. (2011).

The spring (Oct-Nov) KEOPS-2 Fe profiles at station A3 showed a similar structure to those from the late summer (Jan-Feb) during KEOPS-1, with surface depletion, concentrations increasing with depth and enrichment just above the plateau seafloor (Fig. 4b). Through the water column, dFe was between 2 and 5 times greater during KEOPS-2 than KEOPS-1 and pFe was ∼ 10 times greater during KEOPS-2 (with the exception of the deepest samples). The lower values during KEOPS-1 were likely the result of biological uptake in surface waters and the export of Fe during the spring bloom prior to our arrival at the study site, combined with seasonal changes in the strength of the supply mechanisms to deeper waters at A3 (discussed in van der Merwe et al., 2015).

The E stations within the bathymetrically trapped complex recirculation system showed similar hydrographic and nutrient distributions below the mixed layer (Fig. 3d, e and f), which shoaled from 64 m at E-1 to 32 m at E-3 to 39 m at E-5, with some internal variability in water column structure at mid-depths. Surface waters warmed from 2.7 to 3.4 ∘C between the occupations of E-1 and E-5, although no significant nitrate drawdown was observed (Blain et al., 2015). Below AASW, a subsurface temperature minimum (Tmin, ∼ 1.7 ∘C) was observed between 180 m (E1) and 220 m (E5), characteristic of WW. The Tmin feature is associated with waters south of the PF, although the recirculation feature probably also received SAZ waters mixed in from the north (d'Ovidio et al., 2015). T, S and O2 characteristics indicated the presence of UCDW (∼ 600–700 m) and LCDW (deeper than ∼ 1300 m) deeper in the water column above the seafloor (Quéroué et al., 2015). Water parcel trajectories calculated from altimetry based geostrophic currents indicated that it took generally > 2 months for Fe-rich waters from the plateau to travel to the downstream plume site associated with the recirculation feature (E stations) (d'Ovidio et al., 2015). However shorter transport times are also possible due to episodic transport across the PF (Sanial et al., 2015).

Waters at the plume stations showed the largest spatial heterogeneity in surface biomass as revealed by the evolution of a mosaic of complex blooms seen in satellite images (see Supplement in Trull et al., 2015). We observed moderate surface Chl a levels ranging from 0.3–0.4 µg L-1 at E-1 and E-3 to 0.5–0.9 µg L-1 at E-5 (Lasbleiz et al., 2014), noting that as much as 50 % of the chlorophyll was below the mixed layer at the plume stations due to stratification of the upper water column in the warm, spring conditions. Unlike the plateau bloom dominated by large cells > 53 µm, the community in the plume E stations was more mixed (Laurenceau-Cornec et al., 2015), with cells present in both the 5–20 and 50–200 µm size classes (Trull et al., 2015). The E stations showed the highest C export fluxes of all regions as estimated from Th deficits, nitrate depletions and free-drifting sediment trap observations (Planchon et al., 2015; Trull et al., 2015; Laurenceau-Cornec et al., 2015).

Due to operational constraints, no dFe data were available at station E-1. The dFe vertical profiles at E-3 and E-5 were quite different, with a distinct surface enrichment to 0.4 nmol L-1 at E-3 above a minimum of 0.2 nmol L-1 at 100 m. This feature was absent at station E-5, where dFe was depleted to < 0.1 nmol L-1 at the surface, likely due to biological Fe uptake, which was highest at E-5 (1745 nmol m-2 d-1) compared to A3-2 (1120 nmol m-2 d-1) (Table 1) and E-4E (880 nmol m-2 d-1; data not shown), despite lower POC and primary production (see discussion below and Fourquez et al., 2015). Deeper in the water column (> 500 m) at E stations, dFe was broadly uniform (0.3–0.5 nmol L-1).

The pFe distributions at the three E stations were similar with a surface (35–40 m) enrichment (1.6–1.9 nmol L-1), a minimum at ∼ 100–200 m below the mixed layer (0.7–0.9 nmol L-1; broadly consistent with the Tmin layer), a maximum at 280–600 m (1.7–2.4 nmol L-1), and with evidence of enrichment near the seafloor at depths > 1800 m (up to 1.5–2.3 nmol L-1). By applying biogenic (using P) and lithogenic (using Al) normalisers to the data (see Sect. 3.2 below), surface pFe enrichment was roughly equally composed of biogenic and lithogenic Fe, whilst the 300–600 m maximum was predominantly composed of lithogenic Fe (> 100-fold greater than biogenic Fe at these depths). This lithogenic Fe was most likely from waters enriched by sediments and transported laterally eastward off the Kerguelen Plateau which sits at ∼ 530 m below the sea surface. There was no obvious change in pFe in surface or deep waters during the bloom evolution at the pseudo-Lagrangian E stations.

KEOPS-1 only occupied one station in the plume east of the Kerguelen Islands (A11 at 49∘09′ S 74∘00′ E). Dissolved Fe at A11 ranged from 0.09 nmol L-1 at the surface to 0.17 nmol L-1 at 1500 m (Blain et al., 2008b), and pFe ranged from 0.07 nmol L-1 at the surface to 0.81 nmol L-1 at 1500 m (Andrew Bowie, unpublished data); thus, it was much lower than our KEOPS-2 observations at the E site (noting that different sampling and digestion methods for pFe were used for the two cruises).

Iron and carbon pools were calculated by integrating the dissolved and particulate profiles down to the base of the surface mixed layer, defined as the depth where the potential density equalled the potential density at 10 m +0.02 kg m-3 (Park et al., 2014a). The mixed layer varied from 165 m at station A3-1 to 32 m at station E-3, consistent with the seasonal shoaling as surface waters warmed, but it remained deep (> 120 m) on the plateau throughout the study due to deep mixing as a result of several passing storms.

Integrated pools of both dissolved (∼ 5×) and particulate (∼ 10×) Fe were significantly greater on the plateau (station A3) compared to in the plume (station E), with stocks at the reference station R-2 lower still. Horizontal dFe supply from the plateau to the plume was either or both via (i) a geostrophic path looping along the northern side of the PF and then back into the recirculation feature (d'Ovidio et al., 2015) and (ii) direct Ekman flux transport of Fe-rich coastal water across the PF driven by westerly winds, as indicated by Ra tracers (Sanial et al., 2015). The latter process is supported by Lagrangian trajectories of water parcels derived from altimetry, which showed the PF was not a strong barrier to water mass movement, with transport of waters across the front taking place on timescales of days to weeks but being highly variable in space and time (d'Ovidio et al., 2015). The pFe pool showed the same variability as the dissolved pool at our three study sites and exceeded the dFe stocks at all sites by factors of approximately 19–26 (A3), 31 (E) and 6 (R-2), although it is estimated that only ∼ 2–3 % of the particulate pool can be converted into bioavailable forms by physically or biologically mediated dissolution (Schroth et al., 2009). If we assume that station A3-1 represented pre-bloom conditions and the integrated mixed layer pool of 54 µmol dFe m-2 was a good estimate of the winter stock, observations show that only 4 weeks later at station A3-2, almost 60 % of the winter stock had been drawn down to 21 µmol m-2. If annual variability is low, which may not always be the case (Grenier et al., 2015), by late summer > 90 % of the winter stock had been used with only 4.7 µmol dFe m-2 remaining in the surface mixed layer at A3 (KEOPS-1 data; Blain et al., 2007). We note that this drawdown is probably a conservative estimate since the winter dFe stock was probably an underestimation (as evidenced by lower dFe in the mixed layer compared to deep waters at A3-1; Quéroué et al., 2015). Biogenic iron (defined as the Fe associated with living phytoplankton and phytoplankton biodetritus) was calculated by assuming that all particulate phosphorus (P) was of biogenic origin and multiplying the mean particulate P concentration in the mixed layer at each station by a maximum intracellular Fe / P ratio of 1.9 mmol mol-1 for natural phytoplankton assemblages measured by Twining et al. (2004) for Fe-replete conditions. These calculations follow methods reported in Planquette et al. (2013) and assume that particulate P is not (in part) derived from local rock weathering. Van der Merwe et al. (2015) have tested this assumption using Fe / P ratios in the Kerguelen Islands basalts and the upper continental crust and note that the ∼ 1000-fold increase in pP within suspended particles can only be explained by pP produced in situ within the mixed layer from dissolved PO4-. Lithogenic Fe was calculated by assuming that all particulate aluminium (pAl) was of lithogenic origin and by multiplying the mean pAl concentration in the mixed layer by a lithogenic Fe / Al ratio of 0.36 mol mol-1, which is the mean value based on basaltic rocks from the Crozet region (0.51; Gunn et al., 1970) and crustal materials (0.2; Wedepohl, 1995). The selected Fe / Al ratio of 0.36 mol mol-1 is also very similar to that of 0.33 used extensively in earlier calculations (Taylor and McLennan, 1989) and reported for the deep Atlantic Ocean (Sherrell and Boyle, 1992). This approach is dependent not only on the chosen Fe / Al ratio but also assumes that processes other than biological assimilation, such as adsorption and scavenging onto organic particles, photoreduction, surface precipitation, and chemically and biologically driven dissolution, are not significant (Measures et al., 2008; Planquette et al., 2011, 2013; Ellwood et al., 2014). Since biogenic and lithogenic Fe were calculated independently, their sum may be less than the observed total particulate Fe concentration. This is likely due to plasticity in the chosen Fe / Al and Fe / P ratios and differential remineralisation rates for Fe, Al and P. Nevertheless, our estimates of biogenic and lithogenic Fe provide a perspective on the relative contributions to the total pFe pool.

Reference and plume waters contained roughly an equal fraction of biogenic and lithogenic Fe. The origin of this biogenic Fe pool will be a combination of biological uptake of dFe, physical adsorption onto suspended biological particles, and conversion from the lithogenic fraction (likely driven by microbes), with these processes operating on different timescales (Boyd et al., 2005; Frew et al., 2006; Planquette et al., 2011). By contrast, the plateau stations contain 19–69 times more lithogenic Fe than biogenic Fe, consistent with the supply from the nearby sediments of the plateau and Heard Island, as suspected by Zhang et al. (2008), van Beek et al. (2008) and discussed in Chever et al. (2010). The measurement of other geochemical “fingerprint” particulate tracers (such as Al, Mn) on the plateau confirmed the provenance of Fe supplied from the Kerguelen shelf sediments in the particulate phase (van der Merwe et al., 2015).

Vertical fluxes were calculated as follows. A vertical diffusivity (Kz) at the base of mixed layer of 10-5 m2 s-1 was used for the plume and reference site, and a Kz of 3 × 10-4 m2 s-1 was used for the plateau site, estimated from the Shih parameterisations (Shih et al., 2005) using the Thorpe scale method (Park et al., 2014b). These values are comparable to Kz values estimated for KEOPS 1 using the Osborn model (4 × 10-4 m2 s-1; Osborn, 1980) (Park et al., 2008a). Vertical diffusivity was multiplied by the vertical dFe gradient for each profile, which was determined using the linear part of the vertical profiles corresponding to the 150–200 m depth strata in Fig. 3a–f, consistent with calculations for KEOPS-1 (Blain et al., 2007). Vertical diffusivity of Fe was negligible at reference and plume sites but significant on the plateau due to both the higher vertical diffusivity and the steeper Fe gradient between 150 and 200 m.

Upwelling was defined as the vertical velocity (wek) multiplied by the dFe concentration at 200 m, which corresponds to the depth of the remnant winter water. The magnitude of vertical velocity in this region has recently been studied by Rosso et al. (2014), who used the MIT general circulation model to examine the sensitivity of the vertical velocity to the horizontal resolution. They found clear differences in wek due to the development of near-surface sub-mesoscale frontal structures that only their highest-resolution model was able to resolve. Rosso et al. (2014) reported vertical velocities for individual water parcels in excess of 100 m day-1 in the Kerguelen region, with wek stronger in the downstream plume. Both the horizontal and vertical circulations were much weaker over the plateau since it acts as a natural barrier to the strong ACC fronts coming from the west. Unfortunately, no seasonal cycle was included in the model forcing. Therefore the temporal root mean square of the vertical velocity reported in Fig. 12b of Rosso et al. (2014) was used for the plateau and plume sites (wek=0.5 and 1 m d-1, respectively), and a conservative value of wek=0.13 m d-1 for the open Southern Ocean (used by de Baar et al., 1995; originally reported in Gordon et al., 1977) was chosen for our reference station. Although wek was lower on the plateau compared to the plume, the higher dFe concentration at 200 m resulted in comparable estimates of upwelled Fe (Table 1).

Entrainment of Fe by episodic (intraseasonal) deepening of the mixed layer has rarely been taken into account in field studies (Frants et al., 2013) due to the absence of data characterising the short-term variability of the mixed layer depth, yet a recent compilation of observations (Nishioka et al., 2011; Tagliabue et al., 2014) and modelling studies (Mongin et al., 2008) suggests that entrainment could be a major vertical supply mechanism fuelling surface biomass (Carranza et al., 2015). We used more than 6000 vertical profiles of salinity and temperature collected in the KEOPS-2 regions of interest to estimate the seasonality of the mixed layer depth and its variability (Supplement Fig. 3). We derived the vertical supply of Fe by entrainment via hypotheses regarding the relation between the size of mixed layer depth excursions and their frequency (see the Supplement). Entrainment data based on transient deepening of the mixed layer was not available for station R-2; therefore we calculated this by multiplying the dFe concentration in winter water (which reflects the dFe concentration of the winter mixed layer) by the winter mixed layer depth (MLD) and assume that this entrainment event happens once per year. “Detrainment” at R-2 was accounted for by multiplying this new entrainment flux by the summer-to-winter MLD ratio.

For the vertical fluxes, in spring on KEOPS-2, entrainment was the dominant vertical Fe flux term on the plateau, delivering ∼ 70 % of the total vertical supply and tripling the total vertical flux in comparison to budgets that neglect this process. At the plume and reference sites, entrainment was comparable to the upwelling flux. Vertical diffusion accounted for 4–8 % of the total vertical supply on the plateau. In contrast, the contribution from dFe entrainment was much reduced in late summer on KEOPS-1 (42 and 8.8 nmol m-2 d-1 for plateau and plume, respectively) due to the deepening and weakening of the ferricline (Fig. 4b). The relative magnitude of the total vertical Fe supply terms at the three study sites was plateau > plume > reference (Table 1, row d).

For the lateral fluxes, the horizontal supply at reference station R-2 was assumed to be zero since HNLC waters upstream and downstream of this station contained similar dFe and pFe concentrations, and as phytoplankton growth and biomass was low at this site, there would be little biogenic Fe exported below the mixed layer. On the plateau, Fe supply at station A3 was taken from the steady-state box model of Chever et al. (2010), which used the horizontal dFe gradient and current velocities from Park et al. (2008a) to calculate the lateral flux of 180 nmol m-2 d-1 in the 0–150 m depth band above the plateau; note that this model used KEOPS-1 data. Lateral transport into the plume E stations was assumed to originate from Fe-fertilised plateau waters that were advected offshore (d'Ovidio et al., 2015). This value was estimated by assuming that horizontal stirring occurs in a Lagrangian framework and using altimetry-derived geostrophic velocities to determine transport across the plateau boundary. We also used a depth band of 0–150 m, considered as the winter mixed layer in the plume over the season. These estimates were combined with direct measurements of the dFe content of three different types of on-plateau stations to calculate the lateral flux over a 3-month supply period prior to the spring bloom, namely (i) two coastal stations near the Kerguelen Islands occupied on KEOPS-2 (stations TEW-1 and TEW-2), (ii) one coastal station close to Heard Island occupied during KEOPS-1 (station C1) and (iii) the central plateau station A3 considered here. This resulted in 5.4 × 107 mol Fe per day being injected into a plume size (defined at a threshold of > 0.3 µg Chl a L-1 and identified from satellite images) of 2.5 × 1011 m2 over 90 days in spring (full details of the calculations are contained in d'Ovidio et al., 2015). This equated to a lateral flux into the plume of 2400 nmol m-2 d-1 in the October–November period.

By combining our in situ Fe measurements with estimated ages of the water bodies in the plume, we calculate a first-order exponential scavenging removal constant between 0.041 and 0.058 d-1, which equated to a residence time of 17 to 24 days, consistent with estimates based on the Fe inventory and Fe export in free-floating traps (15–79 days; Laurenceau-Cornec et al., 2015). Since the total of the vertical and lateral fluxes in the plume were more than double those on the plateau, this may imply that the source waters supplying the plume from the northern Kerguelen Islands shelves (which had a uniquely narrow T-S class in surface waters; Grenier et al., 2015) were richer in Fe than the plateau further south at A3. This is supported by observations of dFe in the surface ocean at stations TEW-1 and TEW-2 (1.2–1.8 nmol L-1), which were close to Hillsborough Bay in waters only 86 m deep (Quéroué et al., 2015).

Considering only internal processes (diffusion, upwelling, entrainment, lateral transport) in supplying Fe to the surface mixed layer, the vertical terms dominated at the reference station; vertical terms were 6-fold greater than lateral terms on the plateau, whereas lateral advection was the dominant term in the plume (4–5-fold greater than the vertical terms). Since the particulate Fe stocks were abundant in surface waters (above the winter temperature minimum layer) and significantly higher than the dissolved pools (most notably on the plateau), it is likely that a fraction of the suspended lithogenic pFe from Heard Island or the Kerguelen Plateau sediments also contributed to the internal dFe supply and fuelled biological responses. This is discussed in more detail later.

Data on atmospheric Fe fluxes through dust deposition and the solubility of Fe in the dust for all three study sites were taken from the nearby land-based sampling site “Jacky” (49∘18′42.3′′ S, 70∘07′47.6′′ E; altitude 250 m) on the Kerguelen Islands, as reported in Heimburger et al. (2012, 2013a). Mean total Fe fluxes taken over the period 24 November 2008 to 07 September 2010 were 500 ± 390 nmol m-2 d-1 (Heimburger et al., 2013a), which was comparable to the Crozet region upstream (895 nmol m-2 d-1; Planquette et al., 2007) and the Southern Ocean sector south of Australia (288–488 nmol m-2 d-1; Bowie et al., 2009) but greater than the flux value estimated by Wagener et al. (2008) during KEOPS-1 (14–46 nmol m-2 d-1). The remoteness of the Kerguelen region means it receives low quantities of atmospheric material (Heimburger et al., 2012; Wagener et al., 2008), the majority of which is crustal in origin, such as desert dust from South America, South Africa or Australia (Prospero et al., 2002; Mahowald, 2007; Bhattachan et al., 2012), although local anthropogenic activities, rock outcrops and exposed soil may also impact dust fluxes.

Atmospheric fluxes were dominated by wet deposition (Heimburger et al., 2012). Heimburger et al. (2013b) calculated the mean “soluble” Fe deposition flux (defined as < 0.2 µm) using a median solubility of 82 ± 18 % in rainwater on the Kerguelen Islands. These high solubilities were attributed to the remoteness of the sampling location from dust sources resulting in strong cloud chemical processing during transport. However, the solubility of Fe dissolved in seawater at higher pH will be much lower (Schroth et al., 2009; Sedwick et al., 2007). Hence a conservative value of 10 % of Fe that is released into seawater was chosen (Baker et al., 2006; Mackie et al., 2006) for our budgets here, resulting in a soluble Fe atmospheric deposition flux to the Kerguelen region of 50 nmol m-2 d-1 (Table 1, row f). This value was lower than the internal vertical supply on the plateau (∼ 20-fold) and plume (∼ 10-fold), insignificant compared to the lateral supply to the plume but comparable to the lateral supply on the plateau. Although volcanic ash has not been considered here for atmospheric Fe supply, this term may have played an important role for primary productivity on the Kerguelen Plateau during the middle Miocene climate transition (Abrajevitch et al., 2014).

Downward Fe and C fluxes were measured directly in free-floating sediment P-traps at the plateau (A3-2) and plume (E-1, E-3, E-5) stations and estimated using the 234Th fluxes and Fe / Th ratios at the reference site (R-2) (Planchon et al., 2015). The sinking of pFe was by far the greatest loss term in our budgets, with 5746 nmol m-2 d-1 of total Fe exported from the mixed layer on the plateau, between 895 and 4579 nmol m-2 d-1 exported at the plume stations and 1302 nmol m-2 d-1 exported at the reference station. The flux of sinking pFe decreased from station E-1 to E-3 to E-5, concurrent with the seasonal progression of the bloom and indicating the mixed layer assemblages were efficiently recycling Fe under strong grazing pressure (Laurenceau-Cornec et al., 2015). The downward total pFe fluxes were greater than the sum of the vertical, lateral and atmospheric dFe supply on the plateau but were generally less in the plume.

Aluminium was used as a normaliser to estimate the fraction of lithogenic Fe in the exported material. The percentage lithogenic fraction of total pFe exported at the E stations remained much the same at each deployment (34–39  %), whereas the lithogenic fraction was a much larger component at A3-2 (51 %), reflecting the close proximity to sources of particulate material rich in Fe. The Fe / Al ratio of exported material was higher at E stations (1.0–1.1) and on the plateau A3-2 (0.87) compared to the Fe / Al ratio of lithogenically dominated particles (0.2; Wedepohl, 1995), confirming that a significant amount of exported Fe was biogenic in origin. Interestingly, the Fe / Al export ratios were similar to those associated with suspended particles at E stations (0.9–1.2) but lower than the Fe / Al of suspended particles at A3-2 (1.2). This suggests that the biota associated with the plateau bloom at A3 were more capable of efficiently recycling and retaining biogenic particulate Fe in the mixed layer (through rapid turnover to prevent aggregation and sinking) compared to lithogenic particulate Fe, which had a shorter residence time and was preferentially exported to depth. This may be due to greater ballasting of the lithogenic particles (Ellwood et al., 2014) and is consistent with other export studies which have shown that biologically processed particles have longer residence times than lithogenic particles in the mixed layer (Lamborg et al., 2008a). Since P may be lost from exported particles much faster than Fe due to bacterial remineralisation and zooplankton consumption (Schneider et al., 2003; Lamborg et al., 2008b), it was not appropriate to apply a biogenic normaliser to the P-trap data as this may underestimate the biogenic Fe component of particles captured in the traps.

Iron export fluxes were greater during the spring study of KEOPS-2 compared to the late summer study of KEOPS-1 (Table 2). This difference between the KEOPS studies was also observed in Fe uptake rates (Fourquez et al., 2015). Such observations may be simply related to the seasonal supply; in other words, greater Fe supply in spring resulted in greater Fe uptake and export. Determined pFe sinking fluxes were also greater than those observed during CROZEX study (Planquette et al., 2011), the SAZ-Sense expedition south of Tasmania (Bowie et al., 2009) and the FeCycle-I expedition east of New Zealand (Frew et al., 2006) and of similar magnitude to those reported by Bowie et al. (2001) during the Southern Ocean Iron Release Experiment (SOIREE ) (5.2 µmol m-2 d-1), but they were much lower than Ellwood et al. (2014) reported for FeCycle-II.

The export of particulate organic carbon (POC) into our P-traps followed the same trend as that of pFe at the E stations, decreasing from 7.0 ± 2.3 at E-1 to 2.0 ± 1.0 mmol m-2 d-1 at E-5 (Table 1). Despite the higher pFe vertical fluxes at A3-2, POC export was lower than the E stations. The C export fluxes at 200 m at A3-2 using our P-traps (Laurenceau-Cornec et al., 2015) were similar to results estimated from 234Th deficits by Planchon et al. (2015; 2.3 ± 0.7 and 3.8 ± 0.8 mmol C m-2 d-1, respectively). The comparison of these POC fluxes to results (for the A3 plateau site only) obtained during KEOPS-1 illustrates highly dynamic variations, reflecting the rapid decline in biomass during autumn (Blain et al., 2007). Specifically, P-trap measurements of POC fluxes at 200 m during KEOPS-1 decreased from 3.7 ± 0.3 to 1.3 ± 0.3 mmol C m-2 d-1 over two visits to A3 in February 2005 (Trull et al., 2008), whereas estimates based on 234Th at this time, reflecting the previous ∼ 30 days of export, suggested much higher values (25 ± 7 mmol C m-2 d-1; Savoye et al., 2008). These variations illustrate the difficulty of constraining budgets in temporally evolving systems, providing a cautionary note to our efforts. Additional discussion of temporal and spatial export flux variations during KEOPS-2 is provided in Laurenceau-Cornec et al. (2015) and Planchon et al. (2015).

Intracellular Fe uptake by phytoplankton and bacteria > 0.2 µm (Fourquez et al., 2015) was measured at stations A3-2 and E-5 when the bloom was rapidly growing (Cavagna et al., 2014). Iron uptake fluxes were similar on both the plateau (A3-2) and in the plume (E-5), ranging between 1120 and 1745 nmol m-2 d-1. If we assume that the Fe uptake rate of 28.1 pmol L-1 d-1 measured at E-5 (Fourquez et al., 2015) was conservative at E stations, 0.17 nmol L-1 of Fe could have been consumed in surface waters between the occupations of stations E-4E and E-5. This is consistent with the observed decrease in surface dFe concentrations from 0.19 to 0.06 nmol L-1 at E-4E and E-5, respectively (F. Quéroué, personal communication, 2013). The net and gross demand calculated at A3 during KEOPS-1 (204 and 408 nmol m-2 d-1, respectively; Sarthou et al., 2008) is approximately 3–5 times smaller than the intracellular Fe uptake at A3-2 during KEOPS-2 for a similar C biomass (mean value of 12.7 and 10.3 µmol L-1 POC in surface at KEOPS-1 and KEOPS-2, respectively; Cavagna et al., 2014), perhaps indicating luxury uptake as well as important differences in community composition and activity (primary production). These studies provide the opportunity to compare KEOPS-2 to KEOPS-1 data and generate a general picture of the seasonal progress from early spring to late summer, assuming that interannual and spatial variability is low, which may not be the case (Grenier et al., 2015).

The bacterial and mesozooplankton contributions to Fe regeneration were calculated separately (Table 3). Volumetric values varied between 0.06 and 0.59 pmol Fe L-1 d-1 and between 0.02 and 0.08 pmol Fe L-1 d-1 for bacterial and mesozooplankton Fe regeneration, respectively. The mesozooplankton rates were much lower than for KEOPS-1 because there were much fewer individuals (0.26–0.56 per L, compared to about 1–6 individuals per L for KEOPS-1; see Fig. 2 in Carlotti et al., 2008). Total Fe regeneration fluxes ranged from 10 (R-2) to 71 (A3-2) nmol m-2 d-1.

A similar Fe regeneration calculation was also performed based on the C budget by using the percentage of gross community production (GCP) that is remineralised for KEOPS-2 and results from Fe uptake experiments described above. This yielded higher Fe regeneration estimates in the range of 1–11 pmol L-1 d-1. Specifically, for station A3-2, 23 % of GCP was remineralised, and therefore the Fe regeneration flux in the mixed layer was 1119 nmol m-2 d-1. Similarly, for station E-5, 34 % of GCP was remineralised, resulting in an Fe regeneration flux of 504 nmol m-2 d-1. Since the Fe regeneration fluxes based on the C budget are much greater (∼ 16 times) than those calculated using the first approach, this suggests that the remineralisation efficiency for Fe regeneration appears to be less than that of C.

Iron regeneration fluxes can be compared with those from KEOPS-1 using the same first approach above. For station A3 on KEOPS-1, this resulted in a Fe regeneration flux of 1 pmol L-1 d-1 in surface waters. Malits et al. (2014) also calculated the release of bacterial bound Fe by viral lysis (0.42 pmol L-1 d-1), which was the dominant loss term during KEOPS-1 (Brussaard et al., 2008). This value compared to 1.5 pmol L-1 d-1 determined in zooplankton grazing experiments (Sarthou et al., 2008), suggesting that grazing and microbial Fe cycling were in a similar range, and the total Fe regeneration was between 2–3 pmol L-1 d-1 for KEOPS-1.

Importantly, Fe regeneration was much lower during the early compared to late bloom stage and was dominated by bacterial regeneration in spring (60–90 % of total Fe regeneration). Strzepek et al. (2005) estimated Fe regeneration rates (during FeCycle-II) for herbivores (16.5–18.4 pmol L-1 d-1), bacterivores (15–25.5 pmol L-1 d-1) and viruses (0.4–28 pmol L-1 d-1), which is equivalent to a total Fe regeneration rate of 1435–3236 nmol m-2 d-1 for a 45 m mixed layer. Bowie et al. (2009) estimated Fe regeneration to be 261–1206 nmol m-2 d-1 for the SAZ-Sense study. So our determined KEOPS-2 mixed-layer Fe regeneration rates (71 and 31 nmol m-2 d-1 at A3-2 and E-5, respectively) were on the lower end of the range reported in other sectors of the Southern Ocean, and clearly insufficient to meet demand (measured as Fe uptake) at all stations, indicating a reliance on new Fe supply. This is discussed in more detail below.