Seasonal dispersal of fjord meltwaters as an important source of iron to coastal Antarctic phytoplankton

Glacial meltwater from the western Antarctic Ice Sheet is hypothesized to be an important source of cryospheric iron, fertilizing the Southern Ocean, yet its trace metal composition and factors which control its dispersal remain poorly constrained. Here we characterize meltwater iron sources in a heavily glaciated western Antarctic Peninsula (WAP) fjord. Using dissolved and particulate ratios of manganese-to-iron in meltwaters, porewaters, and seawater, we show that glacial melt and subglacial plumes contribute to the seasonal cycle of bioavailable iron within a fjord still relatively unaffected by 15 climate change-induced glacial retreat. Organic ligands derived from the phytoplankton bloom and the glaciers bind dissolved iron and facilitate the solubilization of particulate iron downstream. Using a numerical model, we show that plumes generated by outflow from the subglacial hydrologic system, enriched in labile particulate trace metals derived from a chemically-modified crustal source, can supply the surface through vertical mixing, and that prolonged katabatic wind events enhance export of meltwater out of the fjord. Thus, we identify an important atmosphere-ice-ocean coupling 20 intimately tied to coastal iron biogeochemistry and primary productivity along the WAP.


Introduction
Warm temperatures are accelerating glacial retreat and increasing meltwater discharge, rapidly changing Earth's cryosphere (Mouginot et al., 2019;Rignot et al., 2013). Ranging from diffuse flows to waterfalls and streams, cryospheric meltwaters deliver dissolved and particulate material, altering coastal ocean biogeochemistry. Glacial meltwater enters the ocean 25 through surface runoff, direct melting of glacial ice (including icebergs), and discharge from liquid water reservoirs beneath glaciers, carrying iron (Fe) and other trace metals weathered from continental crust. In the surface ocean, the delivery of new Fe is critical for the growth of phytoplankton; and when enhanced, naturally or artificially, carbon is sequestered by the biological pump (Boyd et al., 2019). However, direct measurements of Fe in heavily glaciated fjords reveal that up to 90-99% of dissolved Fe (dFe) originating from glaciers is removed upon mixing with seawater due to estuarine-type removal 30 processes, including: precipitation of insoluble oxyhydroxides, adsorption to the surfaces of existing particles, and

Figure 1. Regional map of study region (red box, inset right) and model domain (dashed red box) with nearby Palmer Station and shelf station (Stn B). Bathymetric map of Andvord Bay with important stations labeled (GS = Gerlache Strait, AC = Aguirre
calculated as the difference between TDFe and dFe. Prior to analysis in the laboratory, these unfiltered acidified samples were vacuum filtered using acid-cleaned 0.4 µm polycarbonate (PC) filters in a Teflon filtration apparatus. Particulate samples were collected on 0.4 µm PC filters and stored at -20°C until complete digestion using an HNO3/HF mixture. The digestion method employed is described in Planquette and Sherrell (2013) and was applied to the US GEOTRACES GP16 total particulate trace metal sample set . 120 Acute attention to cleanliness was applied when sampling icebergs during small boat deployments in the fjord. Floating icebergs were sampled using a clean stainless-steel pickaxe and rust-free stainless-steel screwdriver and plastic mallet for chiseling pieces of ice. Samples were collected by slowly (engine idled) approaching the target piece of floating ice from downwind, limiting the chance of engine exhaust contamination. Each piece of ice was collected above freeboard (sea surface), to reduce the chance the ice was altered by seawater and rinsed with MilliQ prior to placing into acid-cleaned 2 125 gallon Ziploc polyethylene bags and storing at -4°C until sample processing. Prior to filtration, ice samples were removed from the freezer and left to melt at ambient shipboard temperatures. Once completely melted, a small incision was made on the Ziploc bags using a clean stainless-steel razor and contents poured into the Teflon filtration manifold or directly into sample bottles, thus collecting samples for dissolved, total dissolvable and particulate trace metal fractions.

Trace metal concentrations 130
Stored acidified filtered seawater samples were analyzed for Fe at Scripps Institution of Oceanography using flow injection with chemiluminescence methods described by Lohan et al. (Lohan et al., 2006). Dissolved Fe in the samples was oxidized to iron(III) for 1 h with 10 mM Q-H2O2, buffered in-line with ammonium acetate to pH ~3.5 and selectively pre-concentrated on a chelating column packed with a resin (Toyopearl® AF-Chelate-650M). Dissolved Fe was eluted from the column using 0.14 M HCl (Optima grade, Fisher Scientific) and the chemiluminescence was recorded by a photomultiplier tube (PMT,135 Hamamatsu Photonics). The manifold was modified based on Lohan et al. (2006). Standardization of Fe was carried out with a matrix-matched standard curve (0, 0.4, 0.8, 3.2, 10 nmol kg -1 added high purity Fe metal ICP spectrometry standard in 2% HNO3) using low-Fe open ocean seawater. Standards were treated identically to samples. Accuracy was assessed by repeated measurements of GEOTRACES coastal and Pacific Ocean reference seawater samples. Our measurements of GSC gave Fe = 1.391±0.115 (n = 19, over a three-month period, consensus 1.535±0.115). Our measurements of GSP gave Fe = 140 0.164±0.024 (n = 8, over a one-month period, consensus 0.155±0.045). Consensus values are from the most recent July 2019 compilation (geotraces.org). Precision, determined by replicated analyses of an in-house large-volume reference seawater sample within each analytical session, was typically ±5% or better. For the duration of these analyses, the average LOD (defined as 3x the standard deviation of the blank) was 0.036 (n = 10).
A subset of the seawater samples and all freshwater samples were run for Fe and Mn at Rutgers University using isotope 145 dilution-inductively coupled plasma mass spectrometry (ICP-MS) methods based on Lagerström et al. (2013) and similar to those described in Annett et al. (2017). Briefly, 10 mL aliquots of seawater samples were extracted using a commercially https://doi.org/10.5194/bg-2021-79 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. available automated SeaFAST pico system (Elemental Scientific, Inc.) after online buffering to pH approximately 6.5 using ammonium acetate buffer, achieving a 25-fold pre-concentration after column elution in 0.4 mL 1.6 M ultrapure nitric acid (Optima grade, Fisher Scientific) (Lagerström et al., 2013). Isotope dilution was used to standardize Fe, while Mn was 150 standardized using external matrix-matched standard treated identically to samples. The analysis of the concentrate was performed on an Element 2 sector-field ICP-MS (Thermo Fisher Scientific). Accuracy and precision (±3%, 1SD, for Fe and Mn) was assessed by repeated measurements of in-house large-volume reference seawater samples within each analytical session. Blanks averaged 51 pmol kg -1 for Fe (n = 59; LOD = 48 pmol kg -1 ) and 4 pmol kg -1 for Mn (n = 69; LOD = 4 pmol kg -1 ) for all analytical runs. A comparison of the seawater analysis methods employed here is shown in Fig. S1. In general,155 there is good agreement (average 11% and 6% difference late Spring and Fall, respectively) between the chemiluminescence and ICP-MS methods, comparable to the uncertainty of GEOTRACES consensus values from the intercalibration of 13 trace metal laboratories (for Fe, RSD 10%, https://www.geotraces.org/standards-and-reference-materials/). Total dissolvable trace metals and particle digests were analyzed using direct-injection ICP-MS methods using external standards and added In as a matrix and instrument drift corrector for the quantification of particulate Fe, Mn, aluminum (Al), and titanium (Ti) 160 concentrations (Annett et al., 2017).

Sediment cores and diffusive flux
Cores for this study were collected using a 12-barrel Megacore multi-coring device aboard the R/V Nathaniel B. Palmer cruise NBP16-01 in January 2016. See Taylor et al. (2020) for a complete account of coring efforts and Komada et al.
(2016) for a description of the pore water sampling procedures. Porewater dFe and dMn was determined colorimetrically 165 using the ferrozine and formaldoxime techniques, respectively (Armstrong et al., 1979;Burdige and Komada, 2020). For dFe, hydroxylamine-HCl (0.2% final concentration) was added to the samples before analysis, to reduce any dissolved Fe(III) in the samples to Fe(II). For dMn, a solution of hydroxylamine solution was added to an acidified (pH ~1-2) sample, and an EDTA solution was added to remove interference from a colored Fe complex. Porewater oxygen concentrations were measured using a polarographic microelectrode (Brendel and Luther 1995;Luther et al. 1998Luther et al. , 2008. A sequential extraction 170 technique (Goldberg et al., 2012;Poulton and Canfield, 2005) was used to determine sediment Fe speciation for the following fractions: Feox (highly reactive, poorly crystalline iron oxides), Femag (magnetite), Feprs (Fe in poorly reactive sheet silicates), FeT (total sediment Fe), Fepyr (Fe in pyrite), and finally FeU (unreactive pool under all treatments = FeT -(Feox + Femag + Feprs + Fepyr)). All extracts were analyzed for Fe by flame Atomic Absorption Spectrometry (for details see Burdige and Komada 2020). 175 In this study, we investigate the potential for efflux of dissolved trace metals as a source to the overlying water column.
Using equation 1, we can estimate the approximate sediment diffusive flux (Jsed) for dissolved porewater species. In this equation, is the porosity of the sediments, and was found to be on average 0.9 near the sediment surface. Porewater analyses of dissolved Fe and Mn in the Outer Basin (OB) cores reveal high variability in the top-of-core gradient ( #$ #% ) in 180 porewater Fe and Mn (Fig. S2). An average of two cores gives a gradient of ~21.9 µM cm -1 dissolved Fe and ~3.6 µM cm -1 dissolved Mn. Assuming a diffusion coefficient for Fe and Mn in free solution for seawater (DSW) at 0°C to be 3.15x10 -10 m 2 s -1 for Fe(II) and 3.02x10 -10 m 2 s -1 for Mn(II), we can then estimate the diffusion coefficient in the sediments (Dsed) by the following relationship (van Duren and Middelburg, 2001;Halbach et al., 2019): (2) 185

Iron-binding ligands
A subset of seawater samples was analyzed for dFe-binding ligands using single analytical window methods. The methods applied here are described extensively in Buck et al. 2016(Buck et al., 2018. Briefly, natural seawater samples were titrated with dFe (0-35 nM) in order to fully saturate the natural ligands. Following a 2 hour equilibration with the added Fe, a wellcharacterized ligand (salicylaldoxime, SA) was added to compete with natural dFe-binding ligands. The concentration of SA 190 used in this study to examine ligands was 25.0 µmol L -1 ( -"(/0) # = 115). After at least 15 minutes of equilibration, the Fe(SA)x electroactive complex was measured using adsorptive cathodic stripping voltammetry (ACSV) on a hanging mercury drop electrode (BioAnalytical Systems, Incorporated). Peak heights were measured using ECDSOFT and sensitivity was optimized in ProMCC (Omanović et al., 2015). A combination of traditional linearization techniques was used to determine the concentrations and strengths of natural ligands within the seawater sample using ProMCC (Omanović, 195 Garnier, and Pižeta 2015). The uncertainty on modeled complexation parameters was optimized using single or multiple ligand fitting. These methods were applied successfully to the GEOTRACES speciation data sets (Buck et al., 2015(Buck et al., , 2018. We calculate the capacity for the free ligand pool to bind Fe at equilibrium (Fitzsimmons et al., 2015), or aFeL', defined as: where eL is the difference between the total ligand concentration (Lt) and the dFe concentration, and K is the conditional 200 stability constant.

Numerical model simulations
Based on Hahn-Woernle et al. (2020), the ocean in the Andvord Bay region is modeled with the primitive-equation, finitedifference Regional Ocean Model System (ROMS, Haidvogel et al., 2008). The grid has a horizontal resolution of ~350 m and a terrain-following vertical coordinate system with 25 depth layers. Due to the changing terrain, the fixed number of 205 layers, and surface intensified resolution, the maximum thickness for deeper layers is 84.6 m and the minimum thickness for surface layers is 0.5 m (to better resolve e.g. the surface currents Bismark Strait, a passage to the continental shelf in the northwest, and along Gerlache Strait to the northeast (Fig. 1).
Boundary and initial conditions were derived from CTD and ADCP observations. The model is forced with tidal and meteorological data (from TPXO8 Egbert and Erofeeva 2002 [updated] and RACMO van Wessem et al., 2014, respectively) 210 and run from November 2015 for 5 months. After one month, the transient effects, based on dynamics and thermodynamics, were found to no longer be present, and the system was consistent. Only the final four sea-ice free months were analyzed (December through March). Processes like melting of icebergs and floating sea ice are not modeled directly, therefore such local freshwater sources are captured in a surface intensified meltwater input applied along the glacial boundaries. These new freshwater sources include also surface runoff and local melt of glacial ice, while precipitation and snowfall are 215 represented in the meteorological forcing. The applied volume transport of meltwater is a rough estimate based on few modeled results and observed data and results in an intensified meltwater input at the glacial fronts (for further details see Hahn-Woernle et al., 2020). To represent the seasonal cycle of temperature-induced melting the volume flux of inflowing meltwater follows a bell-shaped temporal distribution peaking at the end of January. We use this model to identify the potential supply pathways and estimate the hydrographic export of three Fe-rich sources in 220 Andvord Bay: surface glacial meltwater, subsurface subglacial plume, and deep water masses located within the inner basin.
For this purpose, we designed three model experiments with numerical "dyes" to track potential iron pathways: one, to track the current seasonal input of meltwater from glaciers in Andvord Bay (surface meltwater dye experiment) released along the glacial fronts in the inner fjord at 0-50 m depth (Fig. 1); and two additional experiments involving subsurface water masses in front of Bagshawe Glacier in Inner Basin A (IBA, 64° 53' 36'' S, 62° 34' 48'' W) at two different depths, one at ~100 m 225 and the other at ~300 m (subsurface and deep dye experiments, respectively). Due to the model geometry, the mean depths the subsurface and deep dyes were released were 107 (94-120 m) and 314 m (290-342 m), respectively. Covering two horizontal grid cells each (with different thickness), the subsurface and deep dyes had initial volumes of 5 x 10 6 and 11.3 x 10 6 m 3 , respectively. It follows from the experiment setup that the meltwater dye has a continuous source while the total amount of the other two dyes is a constant as long as they do not leave through the open boundaries of the model domain. 230

Results
We present seasonal results of Fe and Mn concentration and speciation, including a first assessment of Fe-binding ligands in a cold-based Antarctic fjord. Using porewater measurements on sediment cores collected in the fjord, we also present porewater Fe speciation and estimate the sedimentary efflux of dFe and dMn. Finally, the dispersal of Fe-rich sources is modeled to identify pathways for Fe supply and important dynamics contributing to their dispersal. 235

Seasonality and hydrography in Andvord Bay
In Andvord Bay (Fig. 1), seasonal changes in phytoplankton biomass were documented, as indicated by the proxy Chlorophyll-a, which shows a 10-fold concentration decrease across all taxonomic classes between the late spring and fall cruises (Pan et al., 2020). Associated with these changes in primary production, depletion of the surface macronutrients nitrate (N) and silicic acid (Si) were observed (Ekern, 2017). Increased Si concentrations within the inner fjord could be 240 driven by sedimentary processes, or weathering of the bedrock by contact with the 11 marine-terminating glaciers feeding into Andvord Bay Ng et al., 2020). Surface stocks of macronutrients were never exhausted (Fig. 2).
The phytoplankton community was dominated by small size classes, with very few large diatoms (Pan et al., 2020). The microplankton class was sparingly present in the Fall, however, benthic cameras captured a large sedimentation event of marine aggregates indicative of a large diatom bloom in late-January. The export of biogenic particles from the surface also 245 showed a distinct seasonality indicated by increased Chlorophyll-a pigment content in seafloor sediment cores (Ziegler et al., 2020), as well as higher respiration rates from chamber incubation experiments in the Fall compared to Spring (data not shown), although no indication of sulfate reduction was observed in sediment box and Kasten cores (2.3 m long), suggesting that oxygen, nitrate, and metal oxides were sufficient to oxidize organic matter within the upper sediments (C. Smith pers. comm.). 250

255
Derived glacial meltwater fractions (MWf, Fig. 2), based on salinity and oxygen isotopes of seawater, ranged from 0.75-2% in late Spring, and from 0.5-2.5% in the Fall (Pan et al., 2019). The fjord also exhibited a gradient in meltwater content, with highest MWf at the glacier terminus. Using a simple mass balance for the surface layer in Andvord Bay, we estimate an approximate meltwater input of 23600 m 3 d -1 in order to account for the observed changes in oxygen isotope ratios. This estimate is within the derived estimates of surface meltwater flux generated by warm atmospheric temperatures (1.4 x 10 4 to 260 1.2 x 10 5 m 3 d -1 ; Lundesgaard et al., 2020). MWf is strongly correlated with phytoplankton abundance within Andvord Bay; for a detailed discussion see Pan et al. (2019). We find that glacial meltwater impacts phytoplankton within the fjord, but the geographical influence of meltwater can extend across the shelf, hundreds of kilometers from the coastal inputs (Dierssen et al., 2002;Meredith et al., 2017).
Physical properties measured in the study region showed the dominant water masses in the fjord were Antarctic Surface 265 Water (cold fresh) and Bransfield Strait water (cold and salty) . However, during late Spring, greater influence of modified Upper Circumpolar Deep Water was observed outside of the fjord, indicated by its distinctly higher temperature at depth, but this water mass is prevented from entering the fjord due to a shallow sill near the fjord mouth in the Gerlache Strait (Fig. 2). Optical measurements recorded a change in the particle concentration and assemblage between the two cruises. Profiles of beam attenuation coefficient and particulate backscattering coefficient showed strong 270 seasonality (see Fig. 4  Strong buoyant plumes can drive circulation in fjords via the "meltwater pump", but without estimates of volume flux at the glacier grounding line, it is not possible to determine the effect of small amounts of basal and subglacial melt on circulation in Andvord Bay. While this process is described in-depth for Arctic glaciers, Andvord Bay differs in that ocean temperatures are approximately -1 °C at depth, too cold to ablate the glacier terminus, and neutral buoyancy is reached below the surface 280 layer (indicated by subsurface sediment plumes, Domack and Ishman 1993). However, two important consequences of these plumes are a flux of suspended particulate matter within subsurface "layers" as indicated by high beam attenuation coefficient and optical backscatter ( Fig. S3 in Pan et al., 2019), and general mid-water cooling found in the inner fjord ( Figure 8 in Lundesgaard et al., 2020). Downstream mixing mechanisms, such as flow over topographic features or wind induced upwelling, could displace plume water closer to the euphotic zone. 285

Water column trace metals
Dissolved Fe concentrations in the surface, defined as the upper ~20 m based on similar mixed layer depths (MLD) for both seasons , changed seasonally with an overall increase in dFe concentration in the Fall (Fig. 3). The average surface concentration during late Spring was 2.47±0.92 nM (n = 21), while in Fall it was 6.67±1.41 nM (n = 19).
Water column trace metals are presented in Table S1. These concentrations are within the ranges of dFe determined in prior 290 studies (1-31 nM) in the northern WAP region but indicate that large temporal variability exists in surface waters in this region (Hatta et al. 2013;Sanudo-Wilhelmy et al. 2002;Ardelan et al. 2010;Martin et al. 1990). The smaller range of surface concentrations during late Spring suggests that dFe was more tightly controlled by phytoplankton uptake, whereas in the Fall, patchiness among stations arises due to varying proximity to Fe sources and the effects of circulation and mixing.
Vertical profiles of dFe showed a steep increase to values greater than 10 nM at the deepest depths sampled during late 295 Spring, especially at stations located within the inner fjord and basins (Fig. 2, 4). In the subsurface (50-150 m), an enriched dFe source was present with average concentrations 3.68±1.52 nM in late Spring and 7.38±2.49 nM in the Fall. Deep water masses greater than 150 m deep had the highest average concentrations of dFe but a seasonal decrease in concentration was observed (8.79±4.75 nM in late Spring, 6.37±2.38 nM in Fall). The greatest concentrations of dFe were found in the inner fjord and basin stations, with the exception of one station located at the mouth of the fjord near Aguirre Channel (station AC 300 in Fig. 1). Water column concentrations were lower in the Gerlache Strait and fjord mouth. The general shapes of the profiles in late Spring are characteristic of a stratified water column, with dramatic ferriclines below the surface.  In the Fall, surface dMn was more than double that observed in the late Spring, but surface dFe showed a greater seasonal increase, such that the dissolved Mn:Fe ratio decreased overall and was more variable than in late Spring, when concentrations of dMn remained below 4.5nM, even at depth. Labile particulate Mn (LpMn = TDMn -dMn) showed strong 315 co-variation with LpFe and beam attenuation coefficient c(660). The comparatively high surface dissolved Mn:Fe ratios in late Spring were presumably due to intense biological drawdown of Fe during the vernal bloom, evidenced from low concentrations of dFe where phytoplankton biomass (as Chl-a) was highest (Fig. 5a). In the late Spring, dFe is anticorrelated with MWf ( Fig. 5c), whereas there was no significant trend between dFe, biomass and MWf variables in the Fall ( Fig. 5b,d). The correlation between dMn and dFe was stronger in the Fall, however, compared to the late Spring (Fig. 5e,f). 320 Labile particulate iron (LpFe = TDFe -dFe) concentrations were elevated in the inner basins in the late Spring and Fall, and strongly correlated with suspended particle concentrations, indicated by optical beam attenuation coefficient c(660) m -1 (Fig.   5n, 6). Average LpFe concentrations in the surface were comparable to surface waters in Ryder Bay (southern Antarctic Peninsula), where TDFe varied temporally from 57 to 237 nM (Annett et al., 2015). This comparison between LpFe and TDFe is valid since TDFe is much greater than dFe in these two coastal locations, hence it is a good approximation of LpFe. 325 The LpFe maxima were associated with high turbidity in the inner basins, reaching as high as 900 nM at 300m depth in the Fall (Fig. 6). Dissolved Fe and LpFe were highly correlated (r 2 = 0.48 late Spring n = 19; 0.77 Fall n = 28), implying active exchange between these pools (Fig. 5g,h). On average, dFe made up 3.1% (late Spring) and 4.6% (Fall) of the total (a) (b) dissolvable pool. The LpMn concentrations displayed similar seasonality to LpFe and similar association with total particles, but were more strongly correlated in the Fall (Fig. 5l). Dissolved Mn and LpMn were highly correlated (r 2 = 0.70 late Spring 330 n = 19; 0.79 Fall n = 28; Fig. 5i,j). On average, dMn composed 52% (late Spring) and 57% (Fall) of the total dissolvable pool.

Glacial ice and plume trace metals
Glacial ice and plume samples were analyzed for Fe, Mn, Al, and Ti concentrations, which are presented in Table 1. Three glacial ice samples were analyzed for dFe (71.52±121.31 nM) and dMn (49.43±82.64 nM). Visual inspection of Glacial Ice 3 and 4 showed these pieces contained low particle loads, while Glacial Ice 1 and 2 had a comparatively high content of dark colored coarse-grained particles. Hence, these and the "clean" glacial ice samples are indicative of the variability of trace 350 metal concentrations in icebergs found in Andvord Bay. Labile particulate trace metal concentrations were two orders of magnitude higher than the dissolved fraction based on two ice samples (40.71±85.58 µM LpFe, 3.64±5.06 µM LpMn). We did not determine labile particulate trace metals for Glacial Ice 3 and 4, thus these average labile particulate concentrations are skewed toward a high value. Total particulate trace metals showed similar concentration variability to the dissolved fraction (94.87±181.08 µM TpFe, 2.66±5.06 µM TpMn). For Glacial Ice 3 and 4, the concentration of dMn was greater than 355 TpMn. The ratios of labile and total particulate Mn:Fe were 0.061±0.002 mol:mol and 0.028±0.004 mol:mol, respectively.
Dissolved Al and Ti were not analyzed for these ice samples, but total dissolvable and total particulate samples were analyzed for Glacial Ice 1 and 2, and 1-4, respectively. We defined the refractory particulate trace metal concentration as the difference between the total particulate and total dissolvable fractions (RpTM = TpTM -TDTM). Total dissolvable Al and Ti average concentrations were skewed due to the heavy particle load present within Glacial Ice 1 and 2 (603.2±715.68 µM 360 TDAl, 20.75±27.06 µM TDTi). Total particulate Al and Ti had similar variability to the total dissolvable fraction and included all four glacial ice samples with averages of 428.11±790.46 µM TpAl and 13.41±25.68 µM TpTi, therefore the average total particulate concentrations were lower than the average determined for total dissolvable Al and Ti in Glacial Ice 1 and 2. We found the labile particulate concentration to be a valid comparison to total dissolvable since dFe concentration was on average 1.8±1.5% of TpFe concentration. Thus, the particulate fraction dominated trace metal speciation of total Fe, 365 Mn, Al, and Ti in glacial ice.  nM. The total dissolvable Al:Ti ratio was 64±6 mol mol -1 and the total particulate Al:Ti ratio was 39±1 mol mol -1 . The Al:Ti ratio is elevated above the crustal ratio (35 mol mol -1 ) in the total dissolvable fraction, suggesting a larger adsorbed fraction 380 for Al than for Ti. <&.1+-$2-03 ?@A@:BC ?@A@:BC *+,.2"%'"D&+"8&.

Glacial sediments
Solid phase Fe speciation of one sediment core from the outer basin station (OB, 64° 46' 46'' S, 62° 43' 57'' W, ~500 m, collected in January 2016), showed an enrichment of authigenic Fe oxides at the surface. Chemical treatments of the sediments with HCl dissolves poorly crystalline Fe oxy(hydr)oxides (ferrihydrite and lepidocrocite), which are found to be 385 10% of the total particulate Fe of the surface sediments in this location, compared to an average of 2% below 1.5 cm (Fig.   S4). In the surficial sediments, a larger portion of the Fe is associated with poorly labile sheet silicates (e.g. structural Fe(III) in clays, 36%), and a comparable fraction is refractory and is not liberated by any of the solution treatments (31%). Other fractions of particulate Fe are associated with more crystalline and thus less labile Fe oxides (goethite, hematite) and the minerals magnetite and pyrite. Porewater analyses were performed on two OB cores using colorimetric methods, revealing 390 high concentrations of dFe and dMn. Below the well-oxygenated layer (upper ~0.5 cm), but within the upper 10 cm, dFe reaches its peak concentration of 80 µM, while maximum dMn is 6 µM. Down-core from the peak, concentrations tend to decrease for both trace metals, but there is considerable variability between 15 and 25 cm, including several deeper local maxima. The average porewater concentration of dFe in the top 2.5 cm is 26 µM (Fig. S2). There is considerable difference in the porewater concentrations of the two OB cores indicating bioturbation of the sediments resulting in large variability on 395 small scales. Points excluded from the oxygen profiles were below the detection limit, while several samples were lost from the porewater profiles, represented as gaps in the vertical traces of dFe and dMn.

Fe-binding organic ligands
To gain insight into the speciation of dFe with the fjord, we analyzed seawater samples for Fe-binding ligands and to identify comparative strengths of organic Fe complexes (See Methods). Analysis of the ligands within Andvord Bay shows a down-400 fjord gradient in both quantity and quality (all ligand data presented in Table 2). In the late Spring, strong ligands  Table 2) is observed towards the GS, with increasing eL. 405 Within the fjord, weak ligands were detected at Inner Basin B (IBB), closest to Moser Glacier. In the Fall, total ligand concentrations (Lt) were elevated everywhere within the fjord, but the surface ligands were somewhat weaker compared to the late Spring. The greatest concentrations of ligands were found closest to the glaciers (range 11.18 -15.42 nM) and in the GS (12.00±2.94 nM). For both seasons, weak ligands were detected in the subsurface, but a greater concentration in the Fall suggested that these ligands have a local source within the fjord. Compared to other stations in the Fall, we found the plume 410 to contain a small excess of weak ligands (IBA, 110 m). Interestingly, the highest concentration of strong ligands (17.44±1.12 nM) among all sites was in deep water of Station IBA, at 280 m. This is the deepest depth sampled for Fe-binding ligands and the IBA bottom depth was 382 m. We found a down-fjord gradient in ligand strength at the surface, decreasing with distance from the inner basins ( -"2,-" $ 45+# = 11.95 at IBA, 11.03 at GS).

Table 2. Ligand concentrations and equilibrium constants detected in seawater samples. Fe' is the free (unbound) iron concentration. Lt is the total ligand concentration. logK is the conditional stability constant. eL is the excess ligand concentration (eL = Lt -[dFe]). logaFeL' is the complexation capacity. RFeL' is the ratio of Fe' of reoccupied stations, expressed as a percentage.
We determined the free (uncomplexed) Fe concentration (Fe' in Table 2) within samples analyzed for Fe-binding ligands. In 420 the surface, a greater concentration of Fe' was found in the Fall (8.74±6.43 pM, n = 7) compared to the late Spring

Dye experiments 425
To study the transport pathways for dFe, we use numerical passive dyes in the Hahn-Woernle et al. (2020) regional model of Andvord Bay (see Fig. 1 in Hahn-Woernle et al., 2020) to track three potential sources of dFe: surface glacial meltwater (0-50 m) from Bagshawe and Moser Glacier termini, neutrally-buoyant subsurface plume (100 m), and deep water located in IBA (300 m; as in Section 2.5). Due to numerous inputs and complex biogeochemical processes which result in observed dFe distributions in time and space, we simplify the problem by assuming no removal over the duration of simulated dye 430 experiments. We use this approach to illustrate the multiple transport pathways for dFe supply to the fjord and surrounding ocean from December through March (St-Laurent et al., 2017). The results are presented first for the surface meltwater experiment, followed by two fixed-volume experiments, referred to as subsurface and deep dye experiments.
Most of the surface glacial meltwater dye remains in the upper 100 m throughout the model run, and due to its proximity to the surface, it is quickly dispersed over a large region by relatively rapid surface currents. It takes about 10-15 days for the 435 surface meltwater to exit the fjord mouth, where most ends up in the central and northern Gerlache Strait after 120 days (Fig.   S5a).
The subsurface dye (100 m) is spread more rapidly than the deep dye (300 m). After 8 days, the subsurface dye reaches the fjord mouth, which is 4 days before the deep dye, implying it has a shorter residence time within the fjord compared to the deep dye. We loosely define residence time as the model timestamp at which a fixed fraction of dye remains within the fjord 440 domain. After 22 days, 25% of the subsurface dye has left the fjord, while it takes the deep dye almost twice as long (43 days). At the end of the 120 days long model run, less than 18% of the subsurface dye and over 30% of the deep dye remain in the fjord domain (Fig. S6a). Looking at the whole model domain in Fig. 1, which includes Andvord Bay and Gerlache Strait, only 59% of the subsurface dye and 75% of the deep dye are still present after 120 days. The missing 41% (25%) has mainly left the model domain through the Gerlache Strait to the north, where these waters mix with Bransfield Strait water 445 and subsequently with the southern Antarctic Circumpolar Front waters.
We analyzed the vertical distribution of the subsurface and deep dyes along the fjord mouth and horizontally over different depth layers. Within the first day, the subsurface dye spreads over the depth range of 20 to 125 m and the deep dye over 125 to 500 m (>1% of dye per depth layer). The subsurface dye leaves the fjord mainly within the upper 200 m. After 8 days, as the subsurface dye reaches the fjord mouth (Fig. S5b), the maximum concentration is still found close to its release depth at 450 100-125m. Over the next few days, surface layer concentrations (<20m) increase, but the highest concentration is soon found below 125m (after 2 weeks) (Fig. S6a).
The deep dye remains mainly below 200 m as it passes the fjord mouth (maximum water depth at the fjord mouth is 360 m).
After 12 days, as the deep dye reaches the fjord mouth, the maximum concentration is found below 300 m depth. In contrast to the subsurface dye, the deep dye remains longer in the proximity of the fjord mouth and on several occasions, re-enters the 455 fjord leading to a longer residence time within the fjord (Fig. S5c). The majority of the deep dye leaves the fjord at depths below 100 m and along the southwestern coastline. Both dyes, subsurface and deep, have low concentrations in the upper 100 m of the northeastern flank of the fjord mouth. This is due to the inflow of external water from the GS along the northeastern coastline. Throughout the run, the deep dye is confined to the inner basins of the fjord. In all cases, the dyes remain at higher concentrations and for longer periods in the subsurface fjord waters than in the surface layer, which shows 460 faster transport out of the fjord.

Iron sources in a heavily glaciated fjord
Due to the proximity to glaciers and influence of ice within Andvord, we hypothesized meltwaters to be an important source of Fe. We focus on quantifying dissolved, total dissolvable and particulate Fe and Mn, as well as total dissolvable and 465 particulate Al and Ti. Ratios of these elements are treated as proxies for contributions of various endmembers. Candidate endmembers include reducing sediments, weathered crustal material, and biogenic particles (Taylor and McLennan 1995;Twining et al., 2004). Where possible, we estimate fluxes of dFe. We begin by examining the relationship between glacial meltwater and dFe.

Role of surface glacial meltwater 470
Glacial meltwater at the surface has the potential to be a significant source of Fe to phytoplankton. There exists a weakly negative correlation between derived MWf and dFe at the start of the melt season (late Spring: r 2 = 0.29, n = 30; early-Fall: r 2 = 0.05, n = 13; Fig. 5c,d). One possible explanation is that increased meltwater at the surface leads to greater stratification and limits upwelling of Fe-rich deep water, with the effect augmented by removal processes, such as biological drawdown and scavenging of dFe onto sinking particles. Indeed, higher rates of primary production are associated with greater fractions 475 of meltwater in Andvord Bay (Pan et al., 2020). Since glacial meltwater is restricted to the surface, it constitutes a significant input of Fe to the surface throughout the growth season. While we observe high concentrations of dissolved and particulate trace metals within glacial ice, we note that the icebergs within Andvord were predominantly "clean" ice, with little sediment embedded in the ice, indicated by relatively low dFe and TpFe (for instance, Glacial Ice 3 and 4 in Table 1). Based on Fe:Al ratios in particles and average values for continental crust (Taylor and McLennan 1995), we estimate 87±22% (n = 4) of the 480 particulate Fe contained within Andvord icebergs is terrigenous in origin. This is consistent with mechanical weathering of continental crust followed by inclusion of the particles into the ice (freeze-in, Raiswell et al., 2018). Low Fe:Ti and Al:Ti ratios also reflect a continental crust source, but it is worth noting that Glacial Ice 2 had significantly more Mn and Al, relative to continental Fe and Ti. Further, Mn and Al solid speciation suggests there are high concentrations of Mn-and Aloxides, which may be formed when crustal material is altered (Raiswell et al., 2018). It is also possible that fjord sediments 485 were the source of particulate matter within Glacial Ice 2, which would correspondingly have higher Mn content (and higher Mn:Fe) than what is found in basal ice interacting with the subglacial environment (Hawkings et al., 2020). Continental crust material delivered to the ocean would contain a relatively low Mn content compared to Fe (Fe is 4% w/w in crustal material, while Mn is 0.08% w/w, Rudnick and Gao 2013).
Visual inspection suggests that the majority of the ice within Andvord has relatively low concentrations of particles, whereas 490 basal ice, with dark layers of sediment (Glacial Ice 1 in Table 1), will likely skew the average towards high values (Hopwood et al., 2019). A compilation of TDFe in icebergs in Antarctica estimated an average concentration of 24 µM (Hopwood et al., 2019). Our two measurements of LpFe in glacial ice are different (average for this study is 61±70 µM LpFe, n = 2) but are within the range of concentrations determined in the previous study. Thus, we use our average concentration ( Table 1)  including other freshwater sources that are not precipitation) and assuming the input of meltwater is distributed evenly over the fjord surface layer, we calculate fluxes on the order of 15.1 to 704 nmol m -2 d -1 for dFe and 10.4 to 487 nmol m -2 d -1 for dMn. Based on modeling work in this paper, it will become evident that meltwater released to Andvord does not stay within 500 the fjord. Additionally, significant metal loss might result from scavenging processes, transferring Fe to depth on sinking particle surfaces, rendering it inaccessible for phytoplankton uptake. Still, the availability of excess macronutrients within Andvord Bay (Fig. 2) means that substantial increases in the supply of trace metals from glacial meltwater would stimulate growth in the euphotic zone, if light were not limiting.

The nature of Fe in subglacial plumes 505
The inner basins consistently show higher beam attenuation and particle backscattering coefficients than mid-fjord and shelf stations (see Figure 3 in Supplementary Information in Pan et al. 2019). These signals are attributed to ultra-fine suspended sediments (<0.7-0.8 µm). The high particle backscattering coefficient in the surface at all stations in late Spring is due to the high concentrations of biogenic particles associated with the vernal bloom. Inner basins also show local maxima in beam attenuation coefficients at 70-150 m, as well as approaching the benthic boundary layer (Fig. 6). Sediments that originate 510 near the glacier terminus are carried upward in buoyant turbulent plumes, and spread laterally. This is consistent with the presence of glacial meltwater plumes, or "cold tongues", which originate at the glacier grounding line (described in Domack and Williams 2011), entrain deep water masses, and suspend sediments (Straneo and Cenedese, 2015). Since ocean temperatures remained below 0°C in Andvord (see Fig. 2), there is little to suggest basal melting of the ice, as is observed further south along the WAP. It appears reasonable on the basis of the evidence given above, that the subsurface plume 515 signature is subglacial in origin.
Total digestion and subsequent analyses of marine particles collected within the plume reveal high concentrations of weathered crustal sediments (82-86% of TpFe, 61-64% of TpMn), and also ingrowth of authigenic particles most likely consisting of precipitated Fe-and Mn-oxide phases (16-18% TpFe, 36-39% TpMn). These results suggest that the origin of plume particles is a chemically-altered crustal source. Labile particulate Fe is 82-100% of TpFe (Table 1). The Fe:Al and 520 Fe:Ti in plume particles (0.24±0.01 mol mol -1 and 9.25±0.24 mol mol -1 , respectively) were elevated above the average crustal ratios (0.2 mol mol -1 Fe:Al, 7 mol mol -1 Fe:Ti), which implies these samples are enriched in Fe relative to both crustal Al and Ti. In agreement with these results, particulate Al:Ti (39±1 mol mol -1 ) was elevated above crustal ratios (35 mol mol -1 ), indicating a large oxide fraction is associated with this particulate matter. This substantiates our claim that most of the Fe found in the plume is weakly adsorbed to particles and recently precipitated, since dilute HCl leaches liberate the most labile 525 forms of Fe, most likely as oxy(hydr)oxides (e.g. ferrihydrite) in addition to some Fe from clays. This could include oxides directly precipitated from the anoxic subglacial source, as well as a potential fraction of oxides derived from fjord sediments and porewaters entrained at the grounding line. This perhaps indicates a major difference between glaciers containing large volumes of subglacial meltwater that accumulate the products of more extensive reductive chemical weathering, and smaller glaciers situated on steep topography and which feed into fjords, such as those along the WAP. The long residence time and enhanced chemical weathering beneath large glaciers in west Antarctica (PIG, Thwaites Glacier) could result in large accumulations of dissolved trace metals in 540 subglacial outflow. However, subglacial discharge occurs at some distance from the open continental shelf waters because of the broad floating horizontal ice shelves, which make up about 45% of the Antarctic coastline and can extend 10s -100s km from the shelf (Schodlok et al., 2016). Our results suggest that assuming such high export efficiency to the coastal ocean (i.e., using endmember concentrations from glacial runoff and groundwaters as in Death et al., 2014) potentially overestimates dFe supply from anoxic subglacial environments because dFe rapidly precipitates after mixing with seawater. 545

Role of sediments
Analyses of Andvord Bay sediments reveal they are compositionally distinct from temperate fjords consisting of poorly sorted fine silt and clay, many dropstones, suspension deposits and ice-rafted debris (Eidam et al., 2019). Sediment accumulation rates are spatially variable, but a weak along-fjord gradient is present. These deposits suggest sluggish circulation, allowing for the deposition of sediments close to their source, likely through flocculation processes (Cowan and 550 Powell, 1990).

Profiles of beam attenuation coefficient show highest concentration of particles in the inner basins compared to other station
locations (see Figure 4 in Pan et al., 2019). There is little evidence for mechanical resuspension through gravity flows (i.e., turbidites) along the steep basin walls, yet such processes could be responsible for the near-bottom elevation in water column particles (Eidam et al., 2019). The presence of elevated particles in the inner basins is accompanied by the greatest 555 concentrations of dissolved and labile particulate Fe and Mn (Fig. 6), demonstrating the potential of resuspended fjord sediments as a source of dissolved trace metals.
Based on the core top porewater profiles, we estimate the sedimentary efflux to be 43.7 µmol m -2 d -1 for dFe and 7.2 µmol m -2 d -1 for dMn, due to diffusion alone (Fig. S2). This magnitude of flux was also observed in the shelf sediments in the vicinity of South Georgia Island in the SO (Schlosser et al., 2018). Abundant epibenthic fauna were observed within 560 Andvord Bay, which mix the sediments through bioturbation while consuming labile organic matter. The result is deviation from results based on diffusion alone. Taylor et al. (2020) used 234 Th as a proxy to investigate the effect of bioturbation on short timescales and found Andvord Bay sediments possess a high mixing coefficient down to 5 cm (Db = 36 cm 2 yr -1 ) consistent with greater deposition and subsequent utilization of organic carbon in the sediments. We believe this accurately reflects the conditions in this fjord: bioturbation by dense aggregations of epibenthic fauna within the basins. 565 These results are not surprising when compared to a global compilation of in situ measurements of sedimentary efflux of dFe, which is on average ~12 µmol m -2 d -1 for water masses located on continental margins and with O2 concentrations greater than 63 µmol L -1 (Dale et al., 2015). The bottom water oxygen concentration in Andvord Bay always exceeded 230 µmol L -1 . The bottom water O2 concentration for OB at the time sediments were cored, was 270 µmol L -1 . In the Ross Sea, Marsay et al. (2014) estimated spatially variable efflux spanning 0.028-8.2 µmol m -2 d -1 based on water column dFe profiles 570 (Marsay et al., 2014). Abundant epibenthic fauna found within Andvord (Ziegler et al. 2017(Ziegler et al. , 2020 would introduce oxygen to the upper few centimeters of the sediments through bioturbation and reduce the efflux of reduced metals (Severmann et al., 2010). Taylor et al. (2020)  Deep stations indicate a high mixing coefficient for sediments between 7 and 22 cm depth on timescales of 100 years 575 (Taylor, DeMaster, and Burdige 2020). The effect of this process is mixing of oxide-and organic carbon-rich surficial sediments further down in the core on short-to long-timescales. These flux estimates, together with solid phase speciation results, highlight the importance of rapid oxidation and precipitation occurring at the seawater interface, which effectively retain Fe as oxy-hydroxides within the sediments (Burdige and Komada, 2020;Laufer-Meiser et al., 2021). The Fe oxides are enriched within the penetration depth of oxygen (~0.5 cm, Fig. S2 inset) and once bioturbated downward, could be a 580 source of dFe following microbial cycling. Multiple local maxima of porewater dFe were observed deeper in the cores.
While dissimilatory iron reduction (henceforth, DIR) would be a source for Fe, oxidation of Fe with bottom water O2 and Mn(IV) are important sinks and exert a control on the dFe concentration of deep water masses. The deep inner basin water column samples had high dFe concentrations concomitant with high LpFe concentrations (Fig. 2, 6), suggesting some loss of porewater dFe to the water column and rapid formation of authigenic Fe mineral particles. Therefore, the fluxes calculated 585 from porewater profiles upper limit estimates because they do not account for oxidative losses at the sediment-water interface (e.g., Burdige and Komada 2020).
Due to weak midwater circulation, low tidal energy, and stratification of the surface, a disconnect between deep water masses enriched in dFe and the surface of Andvord Bay persists during prolonged quiescent periods. For these reasons, we believe most sedimentary-sourced Fe is restricted to deep water masses and therefore plays a minor role in dFe 590 concentrations within the upper water column. There is potential, however, for re-suspension and entrainment of surface sediments where subglacial meltwater discharges at the grounding line. Due to the low inferred volume of discharge this is likely a small contribution to the total particulate mass within the plume.
The Mn:Fe ratio is a useful signature of the source of dissolved and particulate trace metals in Antarctica and has been applied to the PAL LTER data set (Annett et al., 2017). Applying this same framework to our study, we find that water 595 column dissolved trace metals are heavily influenced by surface glacial ice melt and subglacial meltwater, and to a lesser extent, sediment sources within the fjord, irrespective of season, depth, and meteoric water input (Fig. 7). Due to the shorter residence time of dFe relative to dMn (i.e., inorganic oxidation of Mn is 10 7 times slower than Fe, Sherrell et al., 2018), we would expect the porewater dissolved Mn:Fe ratio to tend towards higher values once exposed to the seawater oxidative front. We therefore cannot rule out porewaters as a source of dMn to the water column. A similar process occurs within the 600 plume, where the elevated dissolved Mn:Fe (0.65 mol mol -1 ) relative to labile particulate Mn:Fe (0.024 mol mol -1 ) shows the effect of rapid conversion of Fe to authigenic mineral particles. Although we do not have comparable measurements for sedimentary labile particulate Mn, based on labile particulate Mn:Fe, we find that the water column labile particulate Mn:Fe ratio is precisely the same ratio as particles found within the subglacial plume, again irrespective of when and where the sample was taken (Fig. 8), suggesting plume particles remain suspended throughout the fjord water column. 605

Organic speciation of dissolved Fe 615
It has been hypothesized that excess ligands (eL = [Lt] -[dFe]) increase the solubility of particulate Fe phases (Gledhill and Buck, 2012;Tagliabue et al., 2019;Thuróczy et al., 2011;Wagener et al., 2012). The persistence of exchangeable pools of dFe would therefore be controlled primarily by particle assemblage and organic ligand complements, where pFe dominates total Fe speciation. We observe a modest increase between late Spring and Fall in the relative contribution of dFe to total Fe (4% to 5% of TDFe, respectively), implying dFe is controlled by scaling closely to LpFe (Fig. 5g,h) since both pools have 620 large interseason differences. This corresponds to an increase in eL between seasons (average 2.1±1.3 nM late Spring n = 9, 6.0±3.2 nM Fall n = 12). The ligands are likely produced during microbial high-affinity uptake or remineralization processes following the termination of a bloom (Gledhill and Buck, 2012;Hogle et al., 2016). The only subsurface sample to contain strong Fe-binding ligands is the deep inner basin adjacent to Bagshawe Glacier (IBA), possibly indicating these ligands have a sedimentary source. It appears, based on these results, ligands in Andvord Bay have the capacity to complex additional Fe 625 input, as well as prevent significant loss due to scavenging (Thuróczy et al., 2012). The nature of these ligands, taken together with the low concentration of dFe and abundance of LpFe within the plume, leads us to speculate that Fe minerals are the target for ligand-mediated mineral dissolution and perhaps microbial uptake, previously found to occur in deep-sea hydrothermal vent plumes (Li et al., 2014).
While we observe a seasonal increase in the excess ligand concentration, there is no significant change in the ratio of Lt:dFe 630 (late Spring 1.8±0.5, Fall 2.0±0.7). In the Amundsen sector, Thuróczy et al. (2012) found waters heavily influenced by the Pine Island Glacier to have Lt:dFe ratios <2.5 throughout the water column, with relatively weaker ligands compared with those found in the highly productive surface waters of the polynya. We too identify weaker Fe-binding ligands associated with the glaciers, and only at MB and Sill 3 did we observe elevated Lt:dFe (3.13, and 2.99 respectively, in the Fall). In the coastal zone of a remote island in the Bransfield Strait, an excess of strong Fe-binding ligands was observed, hypothesized to 635 indicate Fe-limiting conditions (Buck et al., 2010). Temperature and salinity profiles show a strong signature of Bransfield Strait water within Andvord . The presence of excess strong Fe-binding ligands at IBA and S3 during the bloom onset also correspond to elevated NO3 -:dFe (data not shown) above the threshold for potential Fe-limitation of coastal diatoms in the California Current transition zone (~10-12 µmol nmol -1 King and Barbeau 2011). The presence of strong Fe-binding ligands suggests an active microbial strategy in this coastal region to sequester additional Fe from 640 particulate phases during the bloom initiation.
The intense seasonality in primary production and the presence of an undersaturated ligand pool could further increase the bioavailability of particles for downstream communities, where particles within the water column are rare. We calculated the capacity for the free Fe-binding ligands to bind Fe (aFeL' = 1 + (eL • K)). Calculations of aFeL' are included for each sample in Table 2 as well as the Fe' inter-seasonal percent change for reoccupied stations (RFe'). We find the aFeL' increased between 645 late Spring and Fall at IBA, and Sill 4, while a decrease was found at IBB, Sill 3, and Gerlache Strait stations. While all reoccupied stations show an increase in the Fe' concentration (RFe'), the percent change is greatest where aFeL' decreased in the Fall. Thus, the seasonal increase in Fe' reflects the increase in dFe concentrations as well as lower complexation coefficient of weaker Fe-ligand complexes, which contribute most to dFe speciation in the Fall and are associated with surface waters adjacent to glaciers. 650 These first results of organic speciation of dFe in an Antarctic fjord highlight the importance of seasonal ligand sources in establishing the solubility of new Fe entering the coastal ocean. Seasonality in the ligand pool is not currently represented within SO biogeochemical models (Death et al., 2014;Oliver et al., 2019;St-Laurent et al., 2019;Raiswell et al., 2018;Person et al., 2019). Ligand-mediated complexation has the potential to greatly expand the spatial extent in which solubilization of particulate Fe occurs and could be critical for sustaining productivity over a larger geographical region 655 (Ardiningsih et al., 2020). Thus, the size, sinking rate, and composition of particles is critical to their lateral transport and reactivity over time with excess ligands. Our understanding of how cryospheric Fe is transformed after entering the coastal ocean is an important step towards understanding its impact on marine productivity and global biogeochemical cycles with associated feedbacks on climate. For the marine Fe cycle, these geochemical transformations control the bioavailability of Fe, while vertical advection and mixing supply this critical micronutrient to the surface ocean and the euphotic zone.

Surface meltwater sources and export
The surface glacial meltwater flux was estimated in a previous section, assuming the meltwater produced by warm air temperatures and solar irradiance is distributed evenly over the entire fjord area. We compared the observed and modeled contributions of surface glacial meltwater and subsurface sources to the surface dFe inventory at two key stations, Sill 3 (S3) and Gerlache Strait (GS). We assume that the concentration of dFe is a composite signature of three water masses (surface 685 meltwater dye, subsurface dye, deep dye), with varying relative contributions to the surface inventory. For example, in the late Spring, at S3, we observed a MWf of 0.0155 and the surface concentration of dFe was 2.49 nM (Fig. 3). Assuming a meltwater end member concentration of 71.52 nM (Table 1), we find that glacial meltwater contributes 1.09 nM (44% of surface stock) to the surface inventory. In the Fall, the MWf at S3 increased to 0.0226 (Figure 3d), which corresponds to a meltwater contribution of 1.59 nM dFe (35% of surface stock). In the GS, the same analysis reveals that in the late Spring, 690 the surface had a MWf of 0.0193, which contributed 1.38 nM dFe (105% of surface stock) and in Fall had a MWf of 0.0169, or 1.21 nM dFe (24% of surface stock). It could be the case that the meltwater signature observed in the late Spring in the GS did not originate from Andvord Bay, and thus, might have a different dFe content, but the dearth of measurements of dFe in Antarctic glacial ice prevents us from testing this. These results, apart from those for GS in the late Spring, suggest that one or several other sources contribute to the surface inventory of dFe, or, alternatively, that the glacial end member 695 concentration is too low. Given that the MWf varies from 1-2.5% within Andvord Bay during the time of sampling, it is expected that the input of glacial meltwater throughout the melt season would supply some dFe to the surface.
When we examine the time series derived from the model, we find the model consistently underestimates the contribution of meltwater to the surface (Fig. S7). The MWf does not exceed 0.0013 at either S3 or GS stations, and its seasonal maximum of 0.0046 is found at IBB in early February. Since processes like melting of drifting icebergs and sea ice cannot be captured 700 in the model, the applied meltwater flux is based on a simplified representation of all new freshwater sources except for precipitation in Andvord Bay. These sources include, for example, surface runoff and local melt of glacial ice exposed to the atmosphere. The flux which best recreates observed salinity and temperature profiles in Andvord Bay was achieved by a meltwater input of 0.15 GT over 4 months (Hahn-Woernle et al., 2020).
The overall low modeled meltwater fraction is likely a consequence of multiple factors of which we discuss three. First, the 705 meltwater was tracked only for the field season. The generally low salinity in the upper layer at the beginning of the season and the presence of meltwater dye at the end of the summer season (fjord average of 0.0003 MWf in upper 20m) suggested that meltwater can reside for multiple years in the fjord and cannot be fully captured by our meltwater dye. Second, local melt of glacial ice, e.g. floating icebergs, caused by a summertime surface heat flux, can have a strong impact on the MWf in the surface layer and is likely to be underestimated and not well-represented with the parameterization of the modeled 710 meltwater input. Third, only meltwater from the inner Andvord Bay is tracked and other sources are neglected. Based on other modeled meltwater dyes that track sources just outside Andvord Bay, the impact of the external sources is minor (maximum of 0.0003 MWf in early February) compared to the local sources, but they still contribute to the seasonal increase in MWf.
We extracted vertical profiles of MWf from the model at both stations and found that glacial meltwater originating from 715 Bagshawe and Moser glaciers reaches maximum concentration during the summer bloom (late-January 2016) at Sill 3, relatively constrained to the upper 25m (Fig. S8b). In early February, when the bloom was terminated, glacial meltwater concentrations in the fjord decreased due to a weakening meltwater input and lateral dispersal. The weakening input is supposed to reflect the seasonal cycle of ice melting. Ocean circulation dispersed the meltwater into the Gerlache Strait, as shown by a progressive increase in meltwater in the upper water column throughout the melt season (Fig. S8a). If the volume 720 flux of meltwater input is indeed correlated to the seasonal air temperature cycle, as it is parameterized in the model, the results in Fig. 3 would reaffirm that meltwater is an important control on the accumulation of phytoplankton biomass within Andvord Bay (Pan et al., 2020).
The effect of the wind in driving vertical fluxes will vary with wind direction and location within the fjord. The vertical velocity is analyzed for the observation site at Sill 3 and in front of Bagshawe Glacier (IBA). The latter site is an example 725 location for which katabatic winds are expected to lead to intensified upwelling and is also the location of the subsurface and deep dye experiments. Figure 9  Model results for Sill 3 are supported by late Spring observations of elevated dFe and low meltwater fraction at this station ( Fig. 3). We argue that these punctuated periods of upwelling could be a substantial source of dFe to surface waters in Andvord Bay. Further, this supply, together with the flux of glacial meltwater, provides dFe to fuel phytoplankton 735 community growth.
The efficiency with which wind events export the fjord surface water is explored in the glacial meltwater dye experiment. To account for the changing amount of meltwater in the fjord, export across the fjord mouth in Fig. 9c is given as the percentage of the total amount of dye present within the fjord to resolve the effect of katabatic winds on dispersal dynamics of Fe-rich sources. The meltwater dye experiences up to a 28-fold increased export into the Gerlache Strait during periods of strong 740 along-fjord wind, primarily through the surface. To analyze the correlation between along-fjord wind velocity and the relative meltwater export, we first apply a 24-hr Gaussian filter to the relative export of glacial meltwater (Fig. 9), to exclude tidal signals. Applying the same filter to the wind time series, we find the wind and export data are positively correlated (r = 0.628). The correlation between export and along-fjord winds supports the results by Lundesgaard et al. (2019) who found that katabatic winds control the export of fjord water. This has important implications for the dispersal of Fe-rich waters 745 downstream, which eventually mix with Fe-poor waters located on the continental shelf (Annett et al., 2017).

Subsurface and deep sources and export 755
Periods of vertical mixing are shown to occur during katabatic wind events . This could be an important mechanism for supplying additional dFe from the subglacial plume to the surface within the fjord. Prior to the wind event on December 11, the subsurface dye increases gradually in the upper 20m (Fig. S6b). With the onset of the wind event, the vertical transport of the subsurface dye into the upper 20 m intensifies and reaches a maximum of 32.7 x 10 3 m 3 d -  (Table 1) and 8.68 nM dFe for deep (~300 m) IBA waters in the late Spring, these periods of vertical mixing correspond to dFe fluxes of up to 2.81 nmol dFe m -2 d -1 and 0.36 nmol m -2 d -1 (3.17 nmol m -2 d -1 combined) based on the subsurface dye and deep dye, respectively. Following the katabatic wind event, which lasted approximately 11 days, 765 model results show that 36% of the subsurface dye has shoaled above 75 m, with 10% of dye found within the surface layer (<20 m, Fig. S6b). Of the deep water dye, less than 1% is found within the surface layer. The behavior of the deep water masses contrasts with that of the subsurface water, which corroborates the geochemical data suggesting an insignificant contribution of deep water masses to the surface hydrography and thus, to surface dFe inventory. The vertical fluxes estimated in this section are interpreted as a lower-bound for the contribution of the subsurface plume, since the modeled 770 subglacial plume is a fixed volume, when in reality, subglacial meltwater might be supplied continually throughout the melt season. Compared to the flux of surface glacial meltwater input, and the flux due to subsurface and deep water mixing, the upwelling flux generated by wind events is the largest by an order of magnitude.
The quicker export of the subsurface dye, and therefore the low surface concentration, is mainly due to its proximity to the ocean surface (Fig. S5b). The upper ocean is more subject to changes in the upper ocean dynamics and wind stress. In 775 contrast, the deep dye is exported more slowly and is more continuously released into the Gerlache Strait (Fig. S5c). These modeling results provide evidence for the flushing of fjord water to the Gerlache Strait which coincides with periods of intensified winds. Thus, katabatic winds are important both for replenishing the surface Fe concentrations from the subglacial plume as well as exporting Fe-rich surface waters. It is reasonable to assume that in the absence of a strengthened buoyancy-driven overturning circulation, sources from fjord sediments are limited in supplying the surface with dFe in 780 Andvord Bay.

Conclusion: Andvord Bay as a source of Fe
We have shown that in the absence of buoyancy-driven upwelling, the interaction of the ice sheet, atmosphere, and surface ocean, is important for resupplying the surface waters with Fe throughout the summer season, leading to enhanced productivity and sedimentation of carbon. Katabatic wind events result in pulsed export of the surface layer, while upwelling 785 and vertical mixing entrains subglacial plume water in the inner fjord. Observed surface concentrations of dFe in Fall lend support to the modeled dynamics (see Fig. 3). We summarize the findings of this study in a conceptual diagram showing important seasonal sources of Fe during the growth and melt season (Fig. 10). We highlight important processes in the diagram using circled number notation. We found ocean temperatures are cold 1 and do not melt the fronts of glaciers, but warm summer atmospheric temperatures contribute to the surface melting of glacial ice 2. Variability in dissolved and 790 particulate concentrations in glacial ice produces large uncertainties in the calculated flux. The speciation of Fe within glacial ice is mostly accounted for by refractory Fe-bearing particles 3. Only a fraction may be stabilized by excess organic ligands. Another source is fjord sediments 4, though there is considerable uncertainty shown in the magnitude of this flux because evidence indicates that a significant fraction of porewater Fe rapidly precipitates at the oxidative front forming a rich surface layer of Fe oxyhydroxides 5. Intense bioturbation of fjord sediments mixes the surface sediments downwards 795 fueling dissimilatory reduction processes. The dFe that escapes this sink enriches deep waters within the fjord. Small amounts of subglacial meltwater discharge enter the ocean and form turbid subsurface plumes 6. Within the plume, speciation is dominated by high concentrations of labile authigenic Fe-bearing particles that can be solubilized by Fe-binding organic ligands 7. Seaward-blowing katabatic winds 8 occur episodically and cause upwelling and vertical mixing supplying additional Fe to the surface phytoplankton community. These intense energetic periods facilitate the dispersion 800 and export of surface Fe and meltwater away from the fjord where it is advected downstream in the Gerlache Strait 9.
Given that the west Antarctic Peninsula hosts the greatest number of glaciomarine fjords on the continent, and multiple katabatic wind events occur throughout the year, single wind events can play a crucial role for the export of Fe. The modeled export of meltwater integrated over the week after the wind event on December 11 is 38x10 7 m 3 , which is about 43% of the meltwater input during the same time. For comparison, during the following week, with relatively calm wind conditions, 805 only 20% of the meltwater input is exported. We estimate the Fe export to be 272 mol dFe week -1 for this event. However, the warming climate may lessen the likelihood for pulsed export of meltwater-derived Fe by intensifying coastal currents due to declines in sea ice (Moffat et al., 2008), and reduced surface cooling, decreasing the velocity and frequency of katabatic winds over the west Antarctic Ice Sheet (Bintanja et al., 2014). The large variability in inferred dFe content of glacial meltwaters along the WAP (Annett et al., 2017) means that supply likely depends on fjord-specific processes and future changes in ice volume. Advected sources of dFe remain the largest contribution (~50%) to the inventory on the productive continental shelves (De Jong et al., 2015). Therefore, we believe that 820 a latitudinal assessment of WAP fjords could begin to address variable responses to ocean and atmospheric forcing in these https://doi.org/10.5194/bg-2021-79 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. productive ecosystems. Indeed, less than 160 km to the south of Andvord Bay, observations of warm modified UCDW intrusions and an invigorated "meltwater pump" present an alternative mechanism for sustaining local primary production (Cape et al., 2019).
The scope of our results should be highlighted. If we assume Andvord Bay is representative of a typical cold-based fjord, 825 and similarly, Barilari Bay is representative of a warm-based fjord (6% MWf at surface, Cape et al., 2019) then we can estimate the glacial meltwater export resulting from a single wind event for the entire western coast of the WAP (see Appendix A). A total of 3.6 x 10 10 m 3 (36 km 3 ) of surface glacial meltwater is exported seaward, which corresponds to 2.0 x 10 6 mol dFe. A modelling study estimated a total meltwater discharge for all of Antarctica to be 32.5 -97.5 km 3 yr -1 (Pattyn 2010). Thus, katabatic winds are highly efficient at delivering surface meltwater produced near the coast to the continental 830 shelves. However, this is small compared to the total basal melt production rate due to warm ocean temperatures for the largest ice shelves. Using highly accurate remote sensing topographic measurements Adusumilli et al. (2020) found that the major ice sheets have a steady-state meltwater production value of 1100±60 km 3 yr -1 . In a different modeling study, it was estimated 300 -800 km 3 yr -1 enters the SO accounting for observed trends in SO sea surface temperature, sea ice expansion, and sea surface height (Rye et al., 2020). The WAP feeds most directly into the Antarctic Circumpolar Current (ACC), 835 which advects modified coastal waters downstream to the productive Scotia Sea region, potentially magnifying the ecological impact of WAP fjord meltwater production. As the next wave of ocean biogeochemical models incorporate processes at the ice-ocean interface, better predictions of Fe supply to the ACC will be made.
In Andvord Bay, primary production will be sensitive to future changes in subglacial discharge as Antarctic glaciers continue to melt in response to oceanic and atmospheric warming . A greater flux of sediment is expected to be 840 released into the fjord, reducing light quality for primary producers, as part of a natural tidewater glacier cycle (Brinkerhoff et al., 2017). A key question outside the scope of this research is how the quantity and quality of Fe-binding ligands will change in the future. To a first approximation, decreased ligand concentrations associated with the phytoplankton bloom are expected to reduce efficacy of solubilization of particulate Fe and natural fertilization downstream resulting from this leaky fjord. This climatic trend is not yet realized within Andvord Bay (Eidam et al., 2019), but is expected to decrease dFe export 845 through increased scavenging and sedimentation, further resembling high-Arctic and temperate fjords (Hopwood et al., 2016).

Appendices Appendix A: Estimating total meltwater export from WAP fjords
In order to estimate the meltwater export resulting from a single katabatic wind event along the WAP, we first 850 identify two fjord types: 1) fjords where waters are below the freezing temperature (cold-based); and 2) fjords where intrusions of modified UCDW reach the glacier terminus and cause melting. This distinction leads to different MWf production rates. We use data collected from Andvord Bay as a basis for the amount of export occurring in cold-based fjords.
In this instance, a maximum MWf or 0.025 was observed, which corresponded to an export of 38x10 7 m 3 glacial meltwater and is based on the glacial meltwater dye export across the mouth of Andvord Bay integrated over the duration of a week-855 long katabatic wind event.
Meltwater runoff from glaciers due to warm atmospheric temperatures is parameterized as a function of number of days above a temperature threshold (Smith et al., 1998). The area of the glacier in contact with the atmosphere predicts how much meltwater is generated. We use this simple relationship with surface area and relate it to the MWf we observe, allowing us to estimate the fractional contribution from each glacier in Andvord Bay. As an example, Bagshawe Glacier has 860 an area of 250 km 2 , which is 48% of the total glacier area for this fjord, and so would be responsible for producing 48% of the surface glacial meltwater (~18.4 x 10 7 m 3 ). By dividing the total surface glacial meltwater export for a single katabatic wind event by the total area of glaciers in Andvord Bay, we calculate the export rate of meltwater in Andvord Bay glaciers to be 7.4 x 10 5 m 3 km -2 assuming glaciers have an equal rate of meltwater production per unit area. We use this rate as representative for cold-based type glaciers. 865 Since warm atmospheric temperatures in contact with the glacier surface cause production of meltwater, which enters the ocean as surface runoff, this seems a reasonable assumption. Additionally, intrusions of modified UCDW can reach the glacier terminus, causing slightly higher fractions of meltwater at the surface (~0.06 in Barilari Bay). Our general model results showed exchange with water outside of the fjord occurred during katabatic wind events, including inflow of water masses at depth located from outside of the fjord. Thus, these events are likely to enhance delivery of modified UCDW 870 to the glacier terminus (Jackson et al., 2014). We scale the meltwater export to the meltwater fraction since both Barilari and Andvord Bays had similar mixed layer depths. Also, ~40% export of meltwater during katabatic wind events in our model is reasonable compared to estimates for Arctic fjords (10-50%, Jackson et al., 2014). Based on the area of glaciers in Barilari, we calculate an export rate of meltwater for representative warm-based glaciers to be 10.2 x 10 5 m 3 km -2 . We extrapolate these rough estimates for all glaciers on the western coast of the WAP identified by Cook et al. (2016). All glaciers to the 875 south of Andvord Bay are considered warm-based, while those to the north are cold-based (Fig. A1). The area of each of the glaciers used here is published in Cook et al. (2016). Summing the entire volume export of surface glacial meltwater, we find that if all surface waters along the western coast of the WAP experienced a single katabatic wind event, reminiscent of the one recorded in Andvord Bay, a total of 3.6 x 10 10 m 3 (36 km 3 ) of surface glacial meltwater is exported towards the continental shelf (5 km 3 from cold-based glaciers; 31 885 km 3 from warm-based glaciers). This latitudinal difference is consistent with greater meltwater fractions found on the continental shelf in the southern lines of the PAL LTER grid (Annett et al. 2017). Based on a recent compilation of TDFe content in icebergs from Antarctica (Hopwood et al. 2019), and including two measurements from our study, we use a median concentration of 544 nM (n = 57). We then assume a conservative solubility of 10% of TDFe as the dissolved phase, which yields a dFe content of glacial meltwater to be 54.4 nM. This is close to our average dFe measured for three glacial ice 890 pieces in this study (~71 nM). We estimate a single wind event lasting one week on the western coast of the WAP corresponds to an export of 2.0 x 10 6 mol dFe. We realize this analysis does not take in to account the impact of shallow sills in fjords that might be important for restricting UCDW from entering the fjord mouth and interacting with glaciers. Invigorated upwelling due to buoyant plumes originating at the glacier face is expected to have a positive feedback on the melting of the glacier terminus by increasing the 895 delivery of modified UCDW to glaciers and enhancing melt (Cape et al., 2019). This may be driven by warm ocean temperatures, directly melting the face of the glaciers, or atmospheric warming could increase drainage of surface melt to the base of the glacier, resulting in subglacial discharge and buoyant plumes driving circulation. Directionality of the katabatic winds is an important parameter for wind forcing in fjords surrounded by steep topographic features (Lundesgaard et al., 2018). We have explored the possibility when one katabatic wind event per year occurs in the along-fjord direction 900 (seaward) for the entire western coast of the WAP. These mechanisms are fjord specific and deserve further attention due to the complex interactions between the ice, ocean, and atmosphere. We also concede that areal extent of glaciers may not be the most representative measure for meltwater production, when in fact glacier flow velocities might better correlate with meltwater production rates, and thus, meltwater export rates. However, the interplay between surface melt and the subglacial hydrological system, and thus flow rates could mean this is a sufficient, albeit rough assumption. Finally, large uncertainties 905 exist for the average glacial ice content of dFe and the degree to which TDFe may be solubilized and made bioavailable.
This analysis does not take into account the large quantities of solid ice (i.e., icebergs) exported via this mechanism.

Data availability
All CTD data from this study is available at U.