The Fate of Sedimentary Reactive Iron at the Land‐Ocean Interface: A Case Study From the Amazon Shelf

Reactive iron (Fe) oxides in marine sediments may represent a source of bioavailable Fe to the ocean via reductive dissolution and sedimentary Fe release or can promote organic carbon preservation and long‐term burial. Furthermore, enrichments of reactive Fe (sum of Fe oxides, carbonates and sulfides normalized to total Fe) in ancient sediments are utilized as a paleo‐proxy for anoxic conditions. Considering the general importance of reactive Fe oxides in marine biogeochemistry, it is important to quantify their terrestrial sources and fate at the land‐ocean interface. We applied sequential Fe extractions to sediments from the Amazon shelf to investigate the transformation of river‐derived Fe oxides during early diagenesis. We found that ∼22% of the Amazon River‐derived Fe oxides are converted to Fe‐containing clay minerals in Amazon shelf sediments. The incorporation of reactive Fe into authigenic clay minerals (commonly referred to as reverse weathering) is substantiated by the relationship between Fe oxide loss and potassium (K) uptake from sedimentary pore waters, which is in agreement with the previously reported Fe/K stoichiometry of authigenic clay minerals. Mass balance calculations suggest that widely applied sequential extractions do not separate Fe‐rich authigenic clay minerals from reactive Fe oxides and carbonates. We conclude that the balance between terrestrial supply of reactive Fe and reverse weathering in continental margin sediments has to be taken into account in the interpretation of sedimentary Fe speciation data.

Even though clay minerals are generally not considered part of the highly reactive Fe pool, it is currently unknown if authigenic clay minerals are dissolved by the standardized sequential extraction scheme, which is widely applied to determine Fe HR in marine sediments and to evaluate early diagenetic Fe cycling in modern marine sediments (e.g., Baldermann et al., 2015;Henkel et al., 2016;Lenstra et al., 2019;Scholz, Schmidt, et al., 2019) and paleo-redox conditions in the geological record (e.g., Poulton & Canfield, 2011;Poulton et al., 2010;Raiswell et al., 2018). In this study, we evaluate the impact of authigenic clay formation (i.e., reverse weathering) on sedimentary Fe speciation across the Amazon shelf, which is a well-known type locality for reverse weathering Aller, 1995, 2004;Spiegel et al., 2021). We also test the hypothesis that reverse weathering may reduce the proportion of reactive Fe in continental margin sediments relative to river suspended particles.

Study Area
The Amazon River is characterized by high discharge rates of water and suspended sediment (annual discharge of 5,444 km³ of water (Dai et al., 2009), containing 1,200 ⋅ 10 12 g of suspended sediment (Milliman and Sivitsky, 1992)). The majority of the particulate load discharged by the Amazon River originates from the Andes (Meade et al., 1985), whereas a minor fraction is derived from the tropical lowlands. Therefore, Amazon suspended sediments contain considerable amounts of primary silicate minerals (e.g., mica and chlorite) and soil-derived clay minerals (e.g., kaolinite and montmorillonite) (Gibbs, 1967). Previous studies reported particulate total Fe to aluminum ratios (Fe T /Al) between 0.42 and 0.48 at different locations along the Amazon River and VOSTEEN ET AL.

10.1029/2022GC010543
3 of 18 on the Amazon shelf (Martin & Meybeck, 1979;Poulton & Raiswell, 2000;Sholkovitz & Price, 1980), which is similar to the Fe T /Al of 0.44 of the average upper continental crust (Taylor & McLennan, 2009). Poulton and Raiswell (2002) reported an average reactive Fe to total Fe ratio of 0.47 for Amazon River particulates. Despite the low prevalence of highly weathered sediment sources for the Amazon River particulate load, this value is high in comparison to most other rivers in their data base, which indicates that Fe speciation is dominated by intensely weathered material from the Amazon lowlands.
The Amazon continental shelf is exposed to the North Brazil Current (NBC) with flow speeds of 40-80 cm s −1 across the shelf and strong tidal currents reaching up to 200 cm s −1 current speed (Candela et al., 1992;Nittrouer et al., 1986Nittrouer et al., , 1995. The suspended sediment plume discharged by the Amazon River is advected toward the northwest by a combination of these forces and wind-induced surface waves (Kuehl et al., 1986;Nittrouer et al., 1986). High sedimentation and sediment mass accumulation rates prevail on the Amazon shelf (up to 10 cm yr −1 and 6.9 g cm −2 yr −1 , respectively) (Kuehl et al., 1986). The grain size of sediments deposited on the inner shelf (<70 m water depth) is primarily clayey silt or silty clay, while the outer shelf (70-100 m water depth) is characterized by a relict sand layer excavated by current erosion (Gibbs, 1973;Kuehl et al., 1986;Nittrouer et al., 1983). The highly dynamic current system on the Amazon shelf results in frequently occurring resuspension events, which continuously rework up to 150 cm of the upper sediment package (Kuehl et al., 1986).

Sample Collection and Pore Water Analyses
Sediment samples were collected during research cruise M147 RV Meteor (Koschinsky et al., 2018). This cruise to the discharge area of the Amazon River took place from late April to the end of May 2018 during the period of high riverine discharge. A multiple corer (MUC) was deployed at several stations across the Amazon shelf and slope. This study focuses on two transects comprising eight stations in total. One transect represents a section from the Amazon River mouth to the open ocean (offshore transect, stations 024, 023, 085, 117, 037, and 041; Figure 1) and another transect reflects a section parallel to the shore following the sediment plume toward the northwest (along shore transect, stations 023, 089, and 093; Figure 1). Suspended particulate matter in near-bottom water was collected near the along shore transect ( Figure 1; Table 1). The particulate matter samples were obtained using a trace metal-clean CTD (station 095) or a single O.T.E. bottle on a wire (stations 104, 107, 112, 114). Additionally, solid phase sediment samples from two locations on the Amazon deep-sea fan were included in this study (GeoB 4417-5 and GeoB 4409-2 in Figure 1) to extend the offshore transect toward the open ocean. These samples were obtained with a gravity corer during RV Meteor cruise M38-2, which were subsampled with syringes at 5 cm depth intervals (Bleil et al., 1998).
The MUCs were sampled in a cool lab at 12°C at the shallower stations and 4°C at the deeper station 041 on board of the research vessel immediately after retrieval. The bottom water overlying the sediment within the core liners was sampled with a pre-cleaned (HCl) syringe and filtered through a pre-cleaned 0.2 μm polycarbonate (PC) syringe filter (stations 023 to 041) or a 0.2 μm cellulose acetate (CA) syringe filter (stations 085 to 117). Sediment cores were sub-sampled into pre-weighed plastic cups in intervals of 1 cm in the uppermost 6 cm and in intervals of 2 cm in the depth interval between 6 cm and the bottom of the cores. At six of the selected stations (024, 041, 085, 089, 093, and 117) a second sediment core from the same deployment was sampled for pore waters within a glove bag filled with nitrogen gas. The depth intervals were adjusted to visually distinguishable sediment layers (mainly based on sediment color), such that 12 samples were adequately spaced across the core length. The sediment was transferred into pre-cleaned 50 mL centrifuge tubes and centrifuged at 4°C or 12°C at 3,500 rpm for 40 min. Afterward, the centrifuge tubes were transferred to another nitrogen filled glove bag and pore water samples were filtered through pre-cleaned 0.2 μm PC syringe filters (stations 023 to 041) or 0.2 μm

of 18
CA syringe filters (stations 085 to 117). The pore water samples were acidified to pH < 2 with HCl (sub-boiling distilled). All solid phase sediment and water samples were stored at 4°C for further analysis in the home laboratory after the cruise.
Pore water nitrate concentrations were measured on board within 24 hr after sampling with a SEAL QuAAtro continuous flow auto analyzer. The measurements were calibrated with an eight-point-calibration-curve, which was checked against reference materials for nutrient concentrations in seawater (CRM Lot. CG and CRM Lot. BW; see Table S1 in Supporting Information S1) before and after each run.
Depending on the concentration range, dissolved Fe was determined by inductively coupled plasma optical emission spectrometry (ICP-OES, Varian 720-ES) or mass spectrometry (ICP-MS, Agilent 7500). For correction of instrumental mass bias and signal drift, samples and standards were spiked with indium. Samples and standards were diluted with 3% HNO 3 and calibration standards were mixed with additional NaCl to be comparable to the sample matrix.
Upon recovery, the O.T.E. bottles for sampling of suspended particulate matter were carried into a trace metal clean sampling container. The water containing the suspended particulate material was directly filtered through acid washed 25 mm 0.2 μm polyethylene sulfone filters (Supor, Pall Gelman) by pressurizing the O.T.E. bottles (1.5 bar N 2 ). The filters were subsequently stored in the dark at −20°C.

Solid Phase Analyses
Sediment samples were weighed, freeze-dried and weighed again to determine water content and porosity. To determine total concentrations of Fe, aluminum (Al) and potassium (K), freeze-dried and ground sediment samples were completely digested following standard procedures (e.g., Scholz et al., 2011). In brief, 100 mg of sediment were digested in an acid mix consisting of 2 mL 40% HF, 2 mL 65% HNO 3 and 3 mL 60% HClO 4 at 185°C for 8 hr. Subsequently the acids were evaporated to dryness at 190°C and 1 mL of HNO 3 was added and Note. For MUC and gravity core stations, sampling depth equals water depth. Salinity, temperature and bottom water oxygen as well as total organic carbon (TOC ± SD) data are presented, if available. A sequential extraction method was applied to determine operationally defined Fe fractions in Amazon shelf sediments. In the original studies on the reactive Fe content of river suspended particles and continental margin sediments, reactive Fe was defined based on a single-step sodium dithionite extraction (Poulton & Raiswell, 2002. To ensure comparability of our data to previously reported reactive Fe concentrations for river suspended particles (including those from the Amazon) and continental margin sediments with oxic bottom water, we also performed this single-step sodium dithionite extraction on Amazon shelf sediments (Canfield, 1989). In brief, 70 mg of freeze-dried sediment were treated with 10 mL of sodium dithionite solution (50 g L −1 ) for 2 hr at room temperature in an overhead shaker (Raiswell et al., 1994). The solution was subsequently separated from the residue by centrifugation at 4,000 rpm for 4 min. By applying this method Fe oxide minerals are selectively dissolved and extracted but Fe carbonates or magnetite (Poulton & Canfield, 2005), which may form during early sediment diagenesis (Aller et al., 1986;Karlin et al., 1987;Vuillemin et al., 2019;Walker, 1984) are not quantitatively extracted. The results from the single-step sodium dithionite extraction will be abbreviated Fe D in the following (accordingly, previously published ratios are referred to as Fe D /Fe T ).
We further applied the multi-step Fe extraction scheme (Poulton & Canfield, 2005), which is more commonly applied in recent studies to evaluate early diagenetic Fe cycling and paleo-redox conditions in the geological record (e.g., Alcott et al., 2020;Henkel et al., 2016;Raiswell et al., 2018;Scholz, Schmidt, et al., 2019). This extraction scheme consists of three steps, which are summarized in Table 2. Upon completion of the extraction, the extraction solutions were separated from the residue by centrifugation at 4,000 rpm for 4 min. Recent studies (Hepburn et al., 2020;Slotznick et al., 2020) demonstrated that the mineral phases extracted by the individual steps differ considerably from those defined in the original study by Poulton and Canfield (2005). Therefore, we follow previous studies and denominate individual fractions according to the extraction chemicals rather than the originally intended target phases (Table 3) (Henkel et al., 2016). In the original study, the first extraction step was intended to dissolve crystalline Fe carbonates from ancient sedimentary rocks. For modern marine sediments, a more gentle 24 hr acetate extraction at room temperature was recommended instead (Poulton & Canfield, 2005). Recent studies have demonstrated that the 48 hr acetate step at 50°C can extract much but not all Fe carbonate and, in addition, leads to dissolution of some of the Fe oxides and possibly some Fe contained in clay minerals (e.g., nontronite) (Hepburn et al., 2020;Slotznick et al., 2020). Given that Fe carbonates are abundant on the Amazon shelf (Aller et al., 1986) and since we intended to obtain a maximum estimate of the reactive Fe abundance for comparison with paleoenvironmental studies, we decided to apply the 48 hr acetate extraction at 50°C. It needs to be kept in mind that some Fe oxide and Fe containing clay minerals may have been extracted during this step.
The element concentrations of the solutions obtained from the total digestion, the single-step sodium dithionite Fe extraction and the sequential Fe extraction methods were measured by ICP-OES (Varian ICP 720-ES). The certified reference standards MESS-3 and PACS-3 as well as our in-house standard OMZ-2 were used to determine the reproducibility and accuracy (total element concentrations) of these measurements (Table 4). The long-term laboratory averages for OMZ-2 were calibrated against the standards of Alcott et al. (2020).
Acid volatile sulfide (mainly Fe monosulfide, FeS) and pyrite were extracted from the sediment using the chromium reduction method (Canfield et al., 1986). In brief, 0.5 g of sediment were mixed with 8 mL of 6 M HCl to dissolve FeS. The gaseous H 2 S released was precipitated in a bubble trap filled with 10 mL of 5% zinc acetate solution. Afterward, the residue was mixed with 15 mL of chromium (II) chloride solution and heated to 175°C for 1 hr to dissolve pyrite.   (Poulton & Canfield, 2005) 6 of 18 the sediment can be calculated stoichiometrically from the amount of released S 2− , which was determined via photometric measurement of the amount of ZnS formed within the bubble trap. Following most studies on Fe speciation in modern and ancient marine sediment, individual Fe fractions (Fe ac , Fe dith , Fe oxal, and Fe py ) are summarized as Fe HR and normalized concentrations that are based on the multi-step extraction are referred to as Fe HR /Fe T .
A combined version of the Fe extraction method described above and the ones published by Huerta-Diaz and Morse (1990) and Zegeye et al. (2012) Poulton and Raiswell (2002). b Poulton and Canfield (2005). c Zegeye et al. (2012). d Huerta-Diaz and Morse (1990).  in the suspended particulate matter. Carbonate-associated Fe and poorly crystalline Fe oxides (ferrihydrite) were extracted with 2 mL 0.5 M HCl for 1 hr at room temperature (Fe HCl ). Crystalline Fe oxides (goethite and hematite) and magnetite were extracted as described above for Fe dith and Fe oxal except that only 2 mL of extractant were used. For the extraction of Fe silicates, the remaining residue was treated with 2 mL 10 M HF for 1 hr and after a first separation with another 2 mL 10 M HF for 16 hr. Afterward, 1 g of solid boric acid was added to the mixture to dissolve solid fluorides and the extraction was continued for 8 hr. After separation, the residue was washed with 2 mL boiling MilliQ. The solutions of these three steps (2 mL HF, 2 mL HF + boric acid, MilliQ wash) were combined and represent the silicate fraction (Fe HF ). Thereafter the residue was treated with 2 mL concentrated HNO 3 for 2 hr and washed with 2 mL pure water to extract pyrite (Fe HNO3 ). The concentrations were measured by ICP-MS (Agilent Technologies 7500 Series). Using this method, Fe T and Al were calculated as the sum of the Fe and Al in individual fractions. The combined fractions Fe HCl , Fe dith , Fe oxal, and Fe HNO3 represent the Fe HR fraction within suspended particulate matter. An overview about the various sequential extraction schemes, calculation of Fe HR for different sample types and abbreviations is given in Table 3.
For the determination of total organic carbon, freeze-dried sediment was weighed into a silver cup and carbonate carbon was removed via acidification with 0.25 N hydrochloric acid. The measurement was then performed via flash combustion in an Elemental Analyzer (Euro EA).

Pore Water Data
Pore water Fe concentrations ranged from 0.06 to 990 μM ( Figure 2). The lowest Fe concentrations were generally measured close to the sediment surface. In most of the analyzed sediment cores (stations 024, 085, 117, and 093) Fe concentrations increased with sediment depth within the first 2 cm and reached concentration maxima between 255 μM (085) and up to 990 μM (093). At station 089 the increase of dissolved Fe concentrations started at a depth of 20 cm and reached 546 μM at the bottom of the core. The pore water Fe profile at station 041 significantly differed from the others and was characterized by a narrow dissolved Fe peak of only 5 μM at 1.5 cm depth and low Fe concentrations in the rest of the core. At station 093 the dissolved Fe concentration decreased again below ∼15 cm sediment depth. Iron concentrations within the uppermost pore water samples (0-1 cm) were similar to Fe concentrations in the overlying bottom water.
Pore water nitrate concentrations ranged from below 0.05-75.2 μM. At most of the stations, a peak of dissolved nitrate was observed above the Fe maximum. All sediment cores displayed additional maxima of dissolved nitrate in between or below peaks of high dissolved Fe concentrations.

Solid Phase Data
The total Fe content in the sediment solid phase (Fe T ) ranged from 0.9 wt% to 5.4 wt% ( Figure 3). The downcore variability of Fe T decreased from the stations close to the Amazon River mouth (024 and 023; 1.6, and 2.4 wt% Fe T range) to the distal ends of the offshore and alongshore transects (GeoB 4417-5 and station 093; 0.5 and 0.4 wt% Fe T range, respectively). The total Al content ranged from 1.6 wt% to 10.7 wt% with a variability similar to that of the total Fe content. Sedimentary K content ranged from 0.6 wt% to 2.4 wt% ( Table S3 in Supporting Information S1). Pore water Fe and nitrate concentrations for stations 024, 085, 117, and 041 located on the offshore transect and stations 089 and 093 located on the alongshore transect. The distance from the intersection station 023 is noted behind the respective station number (station 024 has a negative distance as it is located shoreward of station 023). Vertical arrows at the top x-axes depict bottom water concentrations. Note differing x-axis scales. All presented data can be found in Table S2 in Supporting Information S1.

of 18
The Fe T /Al of sediment samples from most of the sediment cores ranged from 0.47 to 0.56 (mean of 0.50 ± 0.02) (Figure 4). Samples from station 037 exceeded these values reaching Fe T /Al of up to 1.93. All analyzed samples exceeded the Fe/Al signature of average upper continental crust (0.44; Taylor & McLennan, 2009) but were generally consistent with the range of Fe T /Al previously reported for Amazon River suspended sediment (0.46; Poulton & Raiswell, 2002, 0.47;Martin & Meybeck, 1979). Apart from station 037, Fe T /Al values were approximately constant across the entire Amazon continental margin and no spatial trend of enrichment or depletion was identified (Figure 4). The sedimentary K/Al of most of the Amazon shelf sediment samples ranged from 0.22 to 0.30 (mean of 0.24 ± 0.02) (Figure 4). Only samples from station 037 exceeded this range with K/Al ranging from 0.38 to 0.47. All analyzed sediment samples exceeded the K/Al ratio of Amazon River suspended sediment (0.16; Martin and Meybeck (1979)).
The sediments of station 037 differed from the other stations in that they were characterized by exceptionally low Fe T concentrations but higher Fe T /Al and K/Al compared to all other stations and Amazon River suspended . Fe D and Fe T are shown as a solid blue and a solid back line, respectively. The distance from the intersection station 023 is noted behind the respective station number. All presented data can be found in Table S3 in Supporting Information S1. 9 of 18 sediment. This station was located within a belt of sandy sediments on the outer Amazon shelf, which was previously reported to represent a relict sand layer meaning that the material does not correspond to the fine-grained suspended material discharged by the modern Amazon River (Adams et al., 1986;Kuehl et al., 1986;Milliman and Barretto, 1975). Therefore, samples of station 037 will not be further considered in the discussion.
Reactive Fe recovered from sediment samples by the single-step sodium dithionite extraction (Fe D ) ranged from 0.69 to 2.06 wt% (Figure 3). The proportion of Fe D in the total Fe pool ranges from 0.28 to 0.41 (0.36 ± 0.03 on average), which is about 22% lower than the Fe D /Fe T of Amazon suspended sediment of 0.47 reported by Poulton and Raiswell (2002). Within the individual fractions of the multi-step sequential extraction (Figure 3), the sodium dithionite extractable Fe (Fe dith ) was the largest fraction reaching concentrations ranging from 0.37 to 1.79 wt% representing 49.7%-73.9% of the combined Fe HR fraction. The sodium acetate extractable Fe (Fe ac ) concentrations ranged from 0.09 to 0.74 wt% representing a share of 9.4%-29.1% of the Fe HR fraction. The ammonium oxalate and oxalic acid extractable Fe concentrations (Fe oxal ) ranged from 0.08 to 0.31 wt% representing a share of 9.4%-17.4% of the Fe HR -phases. Only the deepest samples of stations 093 and 117 showed measur-  (Martin & Meybeck, 1979;Poulton & Raiswell, 2002). Solid lines depict the Fe T /Al of upper continental crust (Taylor & McLennan, 2009) and the Fe D /Fe T of average continental margin sediments with oxic bottom waters (Poulton & Raiswell, 2002). The bathymetry along the offshore transect and location of sediment core stations is shown in (d). Note that core GeoB 4409-2 is projected onto the offshore transect; its actual location is further southwest (see Figure 1). All presented data can be found in Table S3 in Supporting Information S1. 10 of 18 able concentrations of Fe py . In these samples the concentrations ranged between 0.03 and 0.12 wt% representing a share of 1.9%-4.9% of the Fe HR -fraction. Variations in the total Fe concentrations (Fe T ) within the analyzed sediment cores can mainly be attributed to concentration changes within the Fe HR fractions, especially the Fe dith and the Fe ac fractions. The sum of Fe ac and Fe dith was similar to Fe D and also the downcore variability of the sum of Fe ac and Fe dith was similar to Fe D .
Combining all the reactive Fe concentrations obtained by the multi-step sequential extraction yields Fe HR from 0.85 to 2.66 wt%, which is about 23% higher than sedimentary Fe D . The Fe HR /Fe T of Amazon shelf sediments ranges from 0.37 to 0.55 with a mean of 0.47 ± 0.03). This mean value is equal to the Fe D /Fe T of Amazon River suspended sediment (Poulton & Raiswell, 2002). The Fe HR /Fe T of Amazon shelf, slope and deep-sea fan sediments were all higher than the Fe D /Fe T of average continental margin sediment underlying oxic bottom water of 0.28 ± 0.06 (Poulton & Raiswell, 2002). Across the entire Amazon continental shelf (excluding station 037), Fe HR /Fe T remained close to the average and no spatial trend was observed (Figure 4).
The suspended particulate matter sampled at the water column stations near the along shore transect was characterized by Fe T /Al of 0.72-0.89 (Table S4 in Supporting Information S1). These Fe T /Al ratios of outer shelf suspended particles were higher than previously reported Fe T /Al for Amazon estuary and inner shelf suspended matter (0.42 ± 0.04 and 0.46 ± 0.05; Sholkovitz & Price, 1980). The Fe HR /Fe T of the suspended particulate matter samples ranged from 0.70 to 0.76.

Early Diagenesis Across the Amazon Shelf
Consistent with previous studies (Aller et al., 1986, high pore water Fe concentrations were found in most of the pore water samples analyzed in this study (Figure 2). This observation along with low concentrations of solid phase Fe sulfide (Figure 3) implies that biogeochemical cycling within Amazon shelf sediments is dominated by Fe reduction whereas net sulfate reduction and burial of reduced sulfur are only of subordinate importance (Aller & Blair, 1996;Aller et al., 2004Aller et al., , 2010. Sedimentary organic carbon concentrations on the Amazon shelf (Table 1) are theoretically high enough to fuel intense organic carbon degradation by sulfate reduction (e.g., Arndt et al., 2013;Jorgensen, 1982) leading to Fe sulfide precipitation. The predominance of Fe-rich (i.e., ferruginous) conditions in pore water despite relatively high organic carbon concentrations can be explained by high solid phase Fe concentrations and, more importantly, frequently occurring remobilization of the uppermost sediment package. During such resuspension events the diagenetic sequence is reset by re-oxidation of hydrogen sulfide to sulfate coupled to Fe reduction thus contributing to the maintenance of extensive Fe-rich zones in the pore water (Aller et al., 1986(Aller et al., , 2010. The mixing and homogenization of the uppermost sediment package (up to 150 cm) Kuehl et al., 1986) by resuspension and deposition events are also reflected by the pore water Fe and nitrate profiles observed in this study. Nitrogenous conditions (presence of nitrate but no oxygen, Fe and hydrogen sulfide) (Canfield & Thamdrup, 2009) are generally expected to be present at the sediment surface above the ferruginous zone (Froelich et al., 1979). The occurrence of multiple nitrogenous zones above, within and/or below ferruginous zones (stations 024, 085, 089 and 093 in Figure 2) is likely a transient feature related to previous resuspension events.
No significant gradient of dissolved Fe concentrations between the bottom waters and uppermost pore water samples was observed at any of the study sites. Consequently, and as expected in an area with a fully oxygenated water column (Table 1), a diffusive flux of dissolved Fe across the sediment-water interface can be excluded. Pore water Fe profiles indicate that dissolved Fe transported upwards by diffusion is reprecipitated at the sediment surface. It is possible that resuspension of the upper sediment parcel mobilizes Fe-rich particles and/or pore water dissolved Fe into the oxic water column. Furthermore, macrofauna could potentially transfer dissolved Fe across the sediment-water boundary despite oxic conditions at the sediment surface (Severmann et al., 2010). Such a mechanism could be facilitated by complexation of dissolved Fe by organic ligands or through formation of colloids (Homoky et al., 2011(Homoky et al., , 2021. Consistent with all of these mechanisms of Fe remobilization, suspended particulate matter samples from the outer Amazon shelf are characterized by elevated Fe HR /Fe T in comparison to shelf sediments. However, only a small number of suspended particulate matter samples were analyzed within this study. Thus, the overall importance of these processes cannot be quantified.

Diagenetic Redistribution Within the Reactive Iron Pool
The Fe within Amazon shelf sediments originates from suspended particles and dissolved Fe discharged by the Amazon River. Given that Fe is more mobile than Al in ferruginous pore waters, any loss or gain of Fe relative to these sources can be evaluated based on Fe T /Al ratios. Previously reported Fe T /Al ratios for Amazon suspended particles varied considerably, presumably due to differing sampling times and locations. Poulton and Raiswell (2000) reported an Fe T /Al of 0.46 for Amazon River suspended sediment whereas Martin and Meybeck (1979) reported a slightly higher Fe T /Al of 0.48. Gaillardet et al. (1997) reported significantly higher Fe concentrations within suspended particles of the Amazon River than the two afore mentioned studies, which could imply higher Fe T /Al. However, no Al data were published in this study so that Fe T /Al cannot be calculated.
In order to compare Fe T /Al ratios of Amazon shelf sediments to the ratios of riverine suspended particles, the potential impact of dissolved Fe precipitation within the estuarine salinity gradient needs to be considered. The annual discharge of dissolved Fe (F Fe,diss ) from the Amazon River can be calculated as follows: where FH 2 O,river refers to the annual discharge of water from the Amazon River (5,444 km 3 yr −1 = 5.44 ⋅ 10 15 L yr −1 ; Dai et al., 2009) and m Fe is the molecular mass of Fe (55.9 g mol −1 ). The reported concentration of dissolved Fe within the Amazon River water ([Fe] diss,river ) varied in previous studies (e.g., Aucour et al. (2003): 1.8 μM; Bergquist and Boyle (2006): 2.5 μM). The resulting calculated annual discharge of dissolved Fe (F Fe,diss = 0.55 ⋅ 10 12 and 0.76 ⋅ 10 12 g yr −1 for 1.8 and 2.5 μM, respectively) can be compared to the annual discharge of solid phase Al and Fe.
The annual discharge of solid phase Al from Amazon River suspended particles (F Al,part ) is the product of the annual discharge of suspended sediment of the Amazon River (F part = 1,200 ⋅ 10 12 g yr −1 ; Milliman & Syvitski, 1992) and the concentration of the Al within the suspended particles ([Al] part = 121.1 mg/g or 115 mg/g; Raiswell, 2000 andMartin &Meybeck, 1979, respectively): This calculation yields a F Al,part of 145 10 12 g yr −1 (using [Al] part from Poulton & Raiswell, 2000) or 138 ⋅ 10 12 g yr −1 (using [Al] part from Martin & Meybeck, 1979). An annual discharge of solid phase Fe within riverine suspended particles (F Fe,part ) of 67 ⋅ 10 12 g yr −1 and 66 ⋅ 10 12 g yr −1 can be calculated by analogy to F Al,part (Equation 2), that is, by using [Fe] part instead of [Al] part ([Fe] part = 55.9 mg/g (Poulton & Raiswell, 2000) or 55 mg/g (Martin & Meybeck, 1979)). By summing up the total Fe discharged by the Amazon River via suspended particles and dissolved Fe, an expected Fe T /Al ratio of the Amazon shelf sediments can be calculated ((Fe T /Al) sed,exp ): Based on this calculation, the expected Fe T /Al of Amazon shelf sediments ranges from 0.47 to 0.49, which is within error of the observed Fe T /Al of 0.50 ± 0.02. The above calculations are associated with a large uncertainty. Particulate Fe concentrations within the lower reaches of the Amazon River might be higher (Gaillardet et al., 1997) than those published by Poulton and Raiswell (2000) and Martin and Meybeck (1979), which may imply a higher Fe T /Al in discharged particles. On the other hand, our calculation neglects any Al precipitation within the estuarine salinity gradient. Previous studies have shown that dissolved Fe and Al removal may be mediated by a different set of processes within different areas of the estuarine system (Mackin & Aller, 1984;Takayanagi & Gobeil, 2000). Furthermore, this decoupling continues on the shelf where sediments may represent a source or a sink for dissolved Al, depending on riverine discharge, bottom water turbulence and, thus, sediment resuspension dynamics (Mackin & Aller, 1984). Considering these uncertainties, we assume that the Fe T /Al of Amazon shelf sediments is largely consistent with the composition of the source material indicating little net gain or loss of Fe from the sediment relative to river-derived particles.
The Fe D /Fe T of Amazon shelf sediments observed in this study (0.36 ± 0.03) is lower than the average Fe D /Fe T ratio reported for Amazon River suspended particles (0.47; Poulton & Raiswell, 2002). In contrast, the Fe HR / Fe T of 0.47 ± 0.03 obtained by the sequential extraction is similar to the Fe D /Fe T of riverine suspended particles (Figure 4). In general, reactive Fe within river suspended particles comprises Fe oxide minerals, which 12 of 18 are contained in the Fe D fraction (Poulton & Raiswell, 2002). In contrast, within marine sediments Fe oxide minerals may be reductively dissolved and the dissolved Fe released may either be lost across the sediment-water interface or be re-precipitated as authigenic Fe carbonate, sulfide and silicate minerals (e.g., Aller et al., 1986;Canfield, 1989;Scholz, Severmann, McManus, Noffke, et al., 2014). These authigenic mineral phases are not extracted by a simple dithionite leach. However, they are at least partly recovered by the more elaborated Fe extraction scheme yielding Fe HR (Poulton & Canfield, 2005). The mismatch in Fe D /Fe T between Amazon shelf sediments and river suspended particles but close similarity of Fe HR /Fe T of shelf sediments and Fe D /Fe T of river suspended particles are therefore indicative of an internal redistribution of Fe within the Fe HR pool on the Amazon shelf.

Impact of Authigenic Mineral Formation on Sedimentary Iron Speciation
Pyrite concentrations in Amazon shelf sediments are generally too low to explain the mismatch between Fe D /Fe T and Fe HR /Fe T (Figure 3). However, the Fe dith extraction step dissolved roughly 20% less Fe than the single-step dithionite extraction (Fe D ) within the Amazon shelf sediments. This observation indicates that a fraction of Fe D was already dissolved by the Fe ac extraction step, which was originally intended to recover Fe contained in carbonates (Poulton & Canfield, 2005). Recent studies revealed that the Fe ac extraction can also partly dissolve Fe oxide minerals (e.g., hematite) and even Fe-containing clay minerals such as nontronite (Hepburn et al., 2020;Slotznick et al., 2020). The formation of nontronite can occur within Fe-reducing sediments and is potentially mediated by microbial extracellular polymeric substances (Harder, 1976;Ueshima & Tazaki, 2001). The redistribution of riverine Fe D to sedimentary Fe ac is thus not only due to the formation of authigenic Fe carbonates, but can also be related to the extraction of Fe from hematite and/or clay minerals such as nontronite. Geochemical evidence for the formation of such clay minerals within Amazon shelf sediments has previously been presented based on the same sample set as analyzed within this study. Spiegel et al. (2021) reported pore water profiles that are indicative of coincident removal of dissolved silica and potassium (K), which is consistent with the formation of authigenic silicate minerals via reverse weathering. Spiegel et al. (2021) also reported elevated sedimentary K/Al across the entire Amazon shelf (redrawn in Figure 4b) and quantified sedimentary K uptake by multiplying excess K relative to Amazon River suspended material by the annual discharge of particulate material. Their findings are consistent with early studies of reverse weathering and sedimentary K uptake in the study area (e.g., Michalopoulos & Aller, 1995Michalopoulos et al., 2000). Michalopoulos and Aller (1995) determined an average stoichiometry of Fe to K ((Fe/K) clay ) of 0.94 within clay minerals formed by reverse weathering on the Amazon shelf. We can utilize this value to evaluate the potential impact of authigenic silicate formation on sedimentary Fe speciation. To this end, sedimentary K uptake will be compared to the deficiency of sedimentary Fe D on the Amazon shelf relative to riverine suspended Fe D .
As a first step, the solid phase K data produced for our sediment samples were corrected for pore water K concentrations given that K dissolved in pore water was transferred to the solid phase during the freeze-drying procedure (see Section 3.2). The K concentration within the wet sediment (K T,wet ) was calculated from the K concentrations of the freeze-dried sediment (K T,dry ) and the determined water content (u): The mass of pore water dissolved K within the total wet sediment sample (K pw,wet ) was calculated from the pore water K concentration (K pw ) and the water content (u): The corrected sedimentary K concentrations for the dry sediment samples (K sed,dry ) was then calculated from the difference of these two values divided by the relative amount of solid phase within the wet sediment (1−u): Ksed,dry = KT,wet − Kpw,wet 1 − Corrected solid phase K concentrations were on average 1.8% lower than uncorrected concentrations. The correction for station 023 was based on averaged pore water concentrations from station 089, as no original data was available for this station and bottom waters at station 023 and 089 had a similar salinity (see Table 1). Sedimentary 13 of 18 K concentrations of samples from the Amazon deep-sea fan (GeoB 4417-5 and GeoB 4409) could not be corrected because no water content data were available.
Following the pore water K correction, the solid phase K data were corrected for the cation exchange capacity (CeC) of Amazon suspended sediments. It was previously demonstrated that Amazon suspended particles take up K upon contact with seawater (Sayles & Mangelsdorf, 1979). As it is not known whether this loosely adsorbed K is incorporated into minerals on the Amazon shelf, we subtracted it from the solid phase K concentrations using a mean CeC of K of 2.0 Δmeq/100 g (Sayles & Mangelsdorf, 1979). The CeC-corrected data were between 3% and 9% lower than the solid phase K data that were only corrected for pore water K (4% lower on average).
The sedimentary K uptake (K xs , in mmol g −1 ) was calculated applying the corrected K data (K corr ) and the K to Al ratio of riverine particulate matter ((K/Al) part ): The (K/Al) part was calculated from data given by Martin and Meybeck (1979) (Al part = 115 mg/g, K part = 18 mg/g, i.e., (K/Al) part = 0.157) and m K is 39.1 g/ mol. Data for K corr and Al sed were obtained within this study (Al was not corrected for pore water concentrations, as these are neglectable low).
The deficiency of sedimentary Fe D on the Amazon shelf relative to river suspended sediment (Fe D,def , in mmol g −1 ) can be calculated from the sedimentary Fe D and Fe T concentrations as well as the Fe D to Fe T ratio within riverine particulate matter ((Fe D /Fe T ) part ): The (Fe D /Fe T ) part was taken from Poulton and Raiswell (2002;= 0.47) and m Fe is 55.9 g/mol. Concentrations of Fe D and Fe T were obtained within this study (not corrected for pore water concentrations because pore water Fe concentrations are negligible).
A plot of the calculated Fe D,def versus K xs data is shown in Figure 5. The Fe D,def and K xs of sediments from the Amazon shelf, slope and deep-sea fan are generally consistent with a slope of −0.94. Thus, the ratio of Fe D loss to K uptake is close to the Fe/K stoichiometry of authigenic clay minerals formed via reverse weathering (0.94; Michalopoulos & Aller, 1995). This observation highlights the importance of reverse weathering for the partitioning of sedimentary reactive species on the Amazon shelf. Furthermore, the observation of a relatively constant ratio of Fe D,def to K xs across the Amazon shelf indicates that authigenic Fe-rich clay minerals are relative evenly distributed across the entire shelf, probably due to continuous sediment resuspension and transport by tidal currents.
The formation of authigenic Fe carbonate minerals such as siderite (FeCO 3 ) within the ferruginous Amazon shelf sediments was previously demonstrated based on pore water saturation state calculations and Scanning Electron Microscopy (SEM) (Aller et al., 1986). More recent studies in modern ferruginous sediments demonstrated that authigenic siderite forms as a result of organic carbon degradation and the accumulation of dissolved inorganic carbon in pore waters in the absence of H 2 S (Vuillemin et al., 2019). Thus, authigenic carbonate formation is likely to contribute to the redistribution of Fe D to Fe ac within Amazon shelf sediments. Precipitation of authigenic carbonate and silicate minerals cannot be clearly distinguished based on the extraction method utilized within this study. However, considering the close correlation between Fe D loss and K uptake (Figure 5), the formation  Michalopoulos & Aller, 1995). All presented data can be found in Tables S3 and S5 in Supporting Information S1.

10.1029/2022GC010543
14 of 18 of authigenic clay minerals appears to be dominant for the redistribution of riverine Fe D to Fe ac within Amazon shelf sediments.
Since Fe ac and Fe dith cannot account for the entire riverine Fe D pool, a fraction of riverine Fe D must also be transferred to the Fe oxal pool (originally meant to represent magnetite; Poulton & Raiswell, 2005) within the Amazon shelf sediments. The formation of authigenic magnetite in marine sediments has for example, been attributed to extracellular production by iron-reducing bacteria within ferruginous sediments, intracellular production by magnetotactic bacteria in the nitrogenous-ferruginous transition zone and potentially Fe 2+ -oxidizing bacteria within the nitrogenous zone (Roberts, 2015). Therefore, a redistribution of river-derived Fe to the sedimentary Fe oxal fraction seems plausible in the extended ferruginous zones of Amazon shelf sediments. Additionally, recent studies demonstrated that the Fe oxal extraction step can also dissolve Fe silicate minerals such as berthierine or chamosite and nontronite (Hepburn et al., 2020;Slotznick et al., 2020). Chamosite and glauconite (which is a precursor of berthierine) may be formed authigenically in continental shelf sediments (Baldermann et al., 2015(Baldermann et al., , 2022van Houten and Purucker, 1984). Thus, the formation of authigenic Fe bearing silicate minerals discussed above likely contributed to the transfer of Fe D to Fe oxal .

Summary and Implications
The Amazon River is the world's largest river system draining an intensely weathered tropical terrain. Consequently, Amazon River suspended sediments are characterized by relatively high reactive Fe concentrations and Fe D /Fe T . In the Amazon shelf sediments, Fe D /Fe T ratios are diminished compared to the riverine suspended source material. In contrast, the Fe HR /Fe T of sediments on the Amazon shelf and deep-sea fan are identical to the Fe D /Fe T of Amazon suspended particles. The ratio of Fe D loss to K gain in Amazon shelf sediments is similar to the Fe/K stoichiometry of authigenic Fe silicate minerals formed during reverse weathering on the Amazon shelf.
Overall, our data are consistent with a scenario in which river-derived Fe oxides are converted to authigenic Fe silicate and carbonate minerals during early diagenesis. The conversion of Fe oxides to Fe silicate and carbonate minerals can explain the mismatch in Fe D /Fe T between Amazon shelf sediments and Amazon River suspended particles. Further research on the formation of Fe-rich silicate minerals at the land-ocean interface is required to evaluate the global relevance of reverse weathering processes as a sink for land-derived Fe oxides. The sequential extraction scheme by Poulton and Canfield (2005) has to be applied with caution in this context since an important fraction of the silicate-bound Fe is likely contained in the Fe ac and Fe oxal fractions of the Fe HR pool.
The original definition of reactive Fe (Berner, 1970(Berner, , 1984Canfield, 1989;Poulton & Canfield, 2011) focused on the reactivity of Fe minerals with respect to hydrogen sulfide on early diagenetic timescales. The commonly applied sequential extraction method to determine the highly reactive Fe content of marine sediments explicitly includes sulfide minerals formed from reactive Fe oxides in the water column or during early diagenesis (Poulton & Canfield, 2005). Authigenic silicate minerals are also formed from reactive Fe oxides during early diagenesis. According to our findings and those of previous studies, authigenic silicate minerals are at least partly recovered by the multi-step sequential extraction scheme for the recovery of the Fe HR pool (Hepburn et al., 2020;Slotznick et al., 2020). Furthermore, a number of recent studies have demonstrated that Fe-containing clay minerals may be dissolved, converted or precipitated by dissimilatory Fe reduction, pyrite formation, weathering and reverse weathering during early diagenesis (e.g., Baldermann et al., 2015Baldermann et al., , 2022Eroglu et al., 2021;Laufer-Meiser et al., 2021;Scholz, Schmidt, et al., 2019;Scholz, Severmann, McManus, Noffke, et al., 2014;Vorhies & Gaines, 2009). We therefore argue that the conventional separation of classic reactive Fe minerals (Fe oxides, carbonates, and sulfides) from allegedly unreactive Fe-containing clay minerals needs to be reconsidered. Depending on the scientific perspective of the respective study, it might be meaningful to broaden the term "reactive Fe." For example, if one is interested in the total amount of biogeochemically reactive Fe at the time of deposition (e.g., in paleo-environmental studies), it might be useful to explicitly include authigenic silicate minerals. Furthermore, if reductive remobilization of Fe oxides or the role of Fe oxides in organic matter preservation is concerned, it may be necessary to consider the impact of reverse weathering on Fe oxide concentrations. The role of authigenic silicate minerals in organic carbon preservation should be investigated in future studies.
Our findings have further implications for the use of sedimentary Fe speciation as a paleo-redox proxy. The commonly applied sedimentary Fe HR /Fe T threshold for water column anoxia of 0.38 (Poulton & Canfield, 2011) was originally based on a single-step sodium dithionite leach (Raiswell & Canfield, 1998). Amazon shelf sediments are characterized by Fe D /Fe T below this threshold. In contrast, Fe HR /Fe T in these sediments clearly exceeds 15 of 18 the threshold for anoxia despite well-oxygenated conditions in the overlying water column. The offset between Fe D /Fe T and Fe HR /Fe T may be explained by the different minerals recovered with the Fe D (Fe oxides) and the Fe HR (Fe oxides, carbonates and silicates) pools. A similar observation was made by Wei et al. (2021) on sediments originating from the subtropical mountainous island of Hainan in the northern South China Sea. These authors reported an offset between Fe HR /Fe T and Fe D /Fe T similar to that observed on the Amazon shelf in our study. We therefore argue that applying the anoxia threshold, which was originally defined based on a single step sodium dithionite leach as part of Fe speciation data that were generated applying the multi-step sequential extraction scheme, may result in misleading interpretations.
The Fe HR /Fe T of Amazon shelf sediments is higher than the Fe D /Fe T of average continental margin sediments but essentially identical to the Fe D /Fe T of Amazon river suspended particles (Poulton & Raiswell, 2002). Such elevated reactive Fe to total Fe ratios of river suspended particles represent a terrestrial signal related to high continental runoff and intense chemical weathering (Canfield, 1997;Poulton & Raiswell, 2002). On the Amazon shelf, this terrestrial signal in the Fe HR /Fe T data remains unaltered across the shelf, slope and deep-sea fan. We therefore suggest that sedimentary Fe HR /Fe T data may not strictly reflect redox conditions in the ocean but may also provide information about the terrestrial supply of reactive Fe as a function of chemical weathering and continental runoff. Past periods of global warming were often accompanied by intensified chemical weathering on land (e.g., the Late Permian (e.g., Cao et al., 2019;Sun et al., 2018) and Late Cretaceous (e.g., Föllmi, 1995;Pogge von Strandmann et al., 2013)), which likely increased terrestrial Fe oxide supply and contributed to elevated sedimentary Fe HR /Fe T during those times (Scholz, 2018;Scholz, Beil, et al., 2019). Furthermore, an increased terrestrial supply of Fe oxides may have amplified organic carbon preservation and burial during greenhouse episodes (Kennedy & Wagner, 2011). Importantly, we do not intend to negate the utility of Fe HR /Fe T as a marine paleo-redox indicator. Instead, we argue that considering the potential control of riverine Fe oxide supply and reverse weathering on sedimentary Fe speciation may provide additional information on paleo-environmental conditions.