Iron deposition influences primary production and oceanic sequestration of atmospheric carbon dioxide (CO2). Iron has two oxidation states, Fe(II) and Fe(III), with Fe(II) being more soluble and available for oceanic phytoplankton uptake. The past proportions of soluble iron in aerosol dust remain unknown. Here we present iron speciation (Fe2+ and Fe3+) in the Antarctic Talos Dome ice core over millennial time scales. We demonstrate that iron speciation over the last 55 kyr is linked to increasing quantities of fine dust (FD) (0.7–5 µm) and intensified long-range dust transport. We propose that Fe(II) and Fe2+ production is principally enhanced in FD by photoreduction, although pH and organic complexation may also contribute to the speciation dynamics. During the Last Glacial Maximum, Fe2+ concentrations in dust increased by up to seven times more than interglacial levels, while Fe3+ only doubled. Cold and dusty climatic periods may increase the percentage of biologically available Fe(II) and Fe2+ deposited in the nutrient-limited Southern Ocean, allowing greater phytoplankton uptake and perhaps increased CO2 drawdown.
 Iron plays an important role in oceanic biogeochemistry [Jickells et al., 2005] due to its ability to limit the biomass of phytoplankton populations in large oceanic zones [Martin and Fitzwater, 1988] and consequent atmospheric carbon dioxide (CO2) drawdown [Joos et al., 1991]. The artificial addition of iron to equatorial Pacific Ocean waters (IronEx) [Coale, 1998] and the polar Southern Ocean (SOIREE) [de Baar et al., 2005] resulted in significant increases in phytoplankton productivity and abundance. Recent ocean fertilization experiments confirm that phytoplankton blooms caused by iron addition draw down considerable CO2 [Smetacek et al., 2012]. However, uncertainty remains regarding the net consequence of this biological pump on the atmospheric system [Fischer et al., 2010] and the extent to which iron controls this mechanism [Sunda, 2010] over glacial-interglacial cycles.
 Ice cores are an accurate archive of late Pleistocene climate, allowing detailed reconstructions of atmospheric composition and aerosol loading, which are important factors for understanding climate dynamics [Wolff et al., 2006]. Changes in the concentrations of many elements, including iron, have been studied in Antarctic sites, such as Dome C (75°06′S, 123°24′E; 3234 MASL) [Wolff et al., 2010] and Vostok (78°27′S, 106°50′E; 3488 MASL) [Gabrielli et al., 2005]. Iron speciation presents additional analytical challenges, and this is the first study to quantify iron species in any ice core. We determined iron speciation (Fe2+ and Fe3+) in Talos Dome (TD) (159°04′21″E, 72°47′14″S; 2315 MASL) ice core samples using flow injection analysis coupled with collision/reaction cell inductively coupled plasma mass spectrometry (FIA-CRC-ICP-MS) [Spolaor et al., 2012] to provide the first evaluation of iron speciation between glacial and interglacial climates with implications for carbon sequestration over millennial time scales.
 The biogeochemical properties of iron are strongly influenced by speciation [Trapp et al., 2010], with Fe(II) species being more soluble and bioavailable [Shaked et al., 2005] than Fe(III) [Baker and Croot, 2010]. Organic complexation may also strongly affect the solubility of iron [Rue and Bruland, 1995] and the oxidation rate of Fe(II). Marine algae more readily take up soluble iron, yet the fraction of soluble iron present in aerosol dust is still unknown [Aguilar-Islas et al., 2010; Baker and Jickells, 2006; Chever et al., 2010]. Observations demonstrate that dust origin influences the soluble iron fraction, with dust particles from arid desert soils presenting lower solubility compared to glacially weathered soils [Schroth et al., 2009]. Baker and Jickells  proposed that iron solubility is not fixed and changes predictably as an inverse function of dust concentration on both spatial and temporal scales, while Zhuang et al. [1992b] reported a change in iron speciation due to atmospheric travel time, with an increase in Fe(II) and solubility associated with remote sources. Long-range atmospheric transport strongly influences iron speciation, enriches Fe(II) by 50%–100% with respect to total source iron, and increases solubility by a factor of 5–17 [Zhuang et al., 1992b]. Variations in Antarctic iron concentrations are strongly linked to dust deposition and transport [Delmonte et al., 2010], where large-size particles (5–20 µm) are predominant during interglacials, while fine dust (FD) (0.7–5 µm) particles become predominant during glacials, indicating changes in the dust provenance and the travel time [Delmonte et al., 2010]. Although the net effect of glacial dust flux changes on oceanic primary productivity is still unclear [Wolff et al., 2006], syntheses conclude that this mechanism may account for 20 to 40 ppmv of the 80–100 ppmv CO2 changes observed over glacial-interglacial transitions [Fischer et al., 2010].
 Ice core samples were collected in 2005–2007 at TD. The site is an ice dome at the junction between the East Antarctic Plateau and the Transantarctic Mountains, located approximately 290 km from the Southern Ocean. The mean temperature is −41°C and snow accumulation is 80 mm we yr−1 [Frezzotti et al., 2004]. The ice sections were transported in a frozen state to the University of Venice and cut into 2 × 3 × 3 cm pieces using a modified commercial band saw and decontaminated using a rinsing technique [Ruth et al., 2008]. Iron speciation was determined by FIA-CRC-ICP-MS as described in detail by Spolaor et al. . This method enables the determination of leachable Fe2+ and Fe3+ in ice core samples using a Ni-NTA Superflow chelating system to separate the Fe species. At pH 2, the resin completely retains Fe3+ with no retention of Fe2+. After isolating and determining Fe3+, we oxidized the Fe2+ using H2O2 and determined the Fe2+ concentration as the difference between the two measurements. The samples were kept frozen at −20°C until immediately prior to analysis due to the rapid oxidation of Fe2+ to Fe3+ in liquid samples.
 Dissolvable Fe concentrations were determined in discrete samples of TD ice sampled for continuous flow analysis [Vallelonga et al.., 2013]. For each 1 m long ice sample, 15 mL was collected in Coulter Counter accuvettes at a rate of 0.5 mL/min. The sample preparation and analytical methods have been previously reported [Barbante et al., 1999]. Melted samples were acidified to pH 1 using sub-boiled distilled HNO3 (Romil, Cambridge, UK) and analyzed at least 24 h later by sector field ICP-MS (Finnigan™ ELEMENT2, Thermo Fisher Scientific Inc., Bremen, Germany) coupled to an APEX Q desolvating introduction unit (Elemental Scientific, Omaha, Nebraska, USA). The detection limit, defined as three times the standard deviation of the instrument blank, was 0.029 ng g−1, with precisions of 16% for Holocene samples and 10% for Last Glacial Maximum (LGM) samples. We use the methods published by Gaspari et al.  to report acid-leachable Fe concentrations, although it has been demonstrated that 1 month is required to completely dissolve Fe-bearing aerosol particles [Edwards et al., 2006]. The time period integrated by each sample varies from approximately 15 yr at 0.6 kyr before present (BP; 1950 CE) to 150 yr at 55 kyr BP. We compare our data to dust concentrations and particle size distributions in parallel sample aliquots [Delmonte et al., 2010].
 We employed statistical analyses to test the strength of the association between Fe species, dust concentrations, and size fractions (both fine and coarse) and sodium concentrations (a proxy for marine air mass inputs). A logarithmic transformation of all variables is necessary to account for the skewed distributions of the observed values. The association between iron species and the proxies was evaluated by linear regression. A residual analysis checked that the statistical assumptions for linear regression inference were satisfied. The strongest predictor of log(Fe2+) is the logarithm of FD (t = 3.970, p < 0.001), followed by the logarithm of sodium (t = 2.488, p = 0.018) and the logarithm of coarse dust (CD) (t = 2.114, p = 0.042). For log(Fe3+), the strongest predictor is the logarithm of FD (t = 3.181, p = 0.003) and to a lesser extent the log-sodium (t = 2.086, p = 0.044). CD is weakly associated with this iron species (t = 1.929, p = 0.063). The linkage between log(Fe2+) and sodium is likely due to the similar trend of both elements during the glacial. Despite the correlation, it is well demonstrated that dust is the main source of iron in Antarctica [Wolff et al., 2006], while the southern ocean is also influenced by other iron sources, including icebergs, subglacial meltwater and upwelling water masses [Raiswell and Canfield, 2012]. For Antarctic ice, sodium is widely used as a proxy for marine salt influxes. The ratio of total iron content to total FD load is shown in Figure 2, as is a smoothing spline reconstruction. We observe that there is little change in the ratio of iron to dust at TD, particularly over glacial-interglacial time scales, indicative of a constant dust-derived source of iron. Slightly enhanced and more variable iron/dust ratios can be observed during the most recent Holocene samples, which have been attributed to the deglacial activation of dust deflation zones local to TD [Delmonte et al., 2010].
3 Results and Discussion
 Our results demonstrate a difference in iron speciation from glacial to interglacial climates, with Fe3+ concentrations being greater than Fe2+ concentrations during the interglacial and the opposite occurring during the glacial (Figure 1). During the present interglacial (1–14 kyr before 1950 AD; hereinafter referred to as BP), Fe3+ concentrations were between 0.02 and 2.19 ng g−1 with an average concentration of 0.59 ng g−1. Fe2+ concentrations ranged from below detection limits of 0.01 to 1.89 ng g−1 with an average of 0.21 ng g−1 (Table 1). Holocene dust deposition at TD is characterized by a similar abundance of CD particles (5–20 µm) and FD particles (0.7–5 µm). CD has a mean concentration of 19.2 ng g−1, while FD has 21.1 ng g−1. CD is attributable to short-distance dust transport from local high-altitude ice-free areas of Victoria Land, while FD is attributable to distant emission sources, predominantly Southern South America [Delmonte et al., 2010]. In the last glacial period, evaluated here between 14 and 55 kyr BP, aerosol dust loading increased with particle size distributions shifting toward finer fractions (0.7–5 µm). During the LGM (14–32 kyr BP), the coarse particle percentage was usually less than 20% of the total dust mass [Delmonte et al., 2010], with the increased fluxes of fine aerosols attributed to long-range transport from deflation sources in Patagonia. The greater concentrations of glacial dust (up to 640 ng g−1 during the LGM) increased the fluxes of many elements to the Antarctic region [Wolff et al., 2010], including Fe2+ and Fe3+. Fe2+ concentrations increased by seven-fold during the LGM, from interglacial values of 0.21 to 1.57 ng g−1. We also observed an increase in Fe3+ concentration, with LGM values (1.23 ng g−1) approximately double those of the interglacial (0.59 ng g−1). The earlier glacial period of marine isotope stage (MIS) 3, evaluated between 32 and 55 kyr BP, was characterized by lower fine particle concentrations (76.5 ng g−1), although still higher than interglacial values (19.2 ng g−1), with greater inputs of CD particles. During MIS 3, Fe2+ concentrations (0.63 ng g−1) were three times greater than interglacial values, while Fe3+ concentrations were similar to interglacial values (0.79 ng g−1) (Figure 1).
Table 1. Summary of Average Fe and Dust Concentrations (ng g−1) in TD Icea
Dust (0.7–5 µm)
Dust (5–20 µm)
aThe Fe2+/Fe3+ mass ratio is also shown for the corresponding time periods. Enrichments beyond average Holocene values are shown in parentheses. LGM signifies Last Glacial Maximum and MIS 3 represents Marine Isotope Stage 3.
 The association of iron speciation ratios (Fe2+/Fe3+) with the presence of FD and CD fractions was tested. After the variables were log-transformed due to their skewed distribution, the results suggest that iron speciation is weakly associated with FD (t = 2.028, p = 0.051) and not associated with CD (t = 1.081, p = 0.2879). These results are interpreted cautiously given the small sample size. Low Fe2+/Fe3+ values were determined for the interglacial (average = 0.4, range = 0.02–2.1) and higher values for the LGM (average = 1.9, range = 0.10–8.6) and MIS 3 (average = 0.82, range = 0.21–2.47) (Figure 2). The LGM average was not greatly influenced by one sample (14 kyr BP) with an Fe2+/Fe3+ value of 8.6 because removing this sample changes the LGM average to 0.92, which remains substantially greater than the interglacial average.
 The enhancement of Fe(II) species in TD ice can be linked to the LGM period when dust fluxes were highest and the FD proportion was greatest. Antarctic glacial dust has been isotopically linked to deflation sources in southern South America [Delmonte et al., 2010]. We propose that the changing glacial-interglacial Fe speciation in Antarctica results from the production of Fe2+ by photoreduction of Fe3+ during atmospheric transport of FD.
 Our findings provide a link between FD particles and the enhancement of Fe(II) species in dust, both of which can linked to enhanced Fe solubility. Baker and Jickells  proposed that iron solubility of mineral dust is primarily controlled by the surface area to volume ratio of the particles. Trapp et al.  confirmed the proposal by demonstrating that iron solubility increases as its size fraction decreases (from 10 to below 2.5 µm) and attributed this to the enhancement of Fe(II). This is consistent with our finding that Fe2+ is associated with increased FD deposition in Antarctica. The 1–2 week transport time for glacial FD deposited in Antarctica allows time for the production of Fe2+ by photoreduction during atmospheric transport and/or cloud processing, as proposed by Zhuang et al. [1992a]. The greater increase of Fe2+ compared to Fe3+ during the glacial suggests that iron is present in a more soluble form (Figures 3a and 3b).
 Although there is extensive evidence that iron solubility is enhanced in the fine particle fraction of mineral dust aerosol, we consider also the potential for post-depositional alteration of the Fe2+/Fe3+ ratio in surface snow. Three different post-depositional processes are proposed to potentially alter Fe2+/Fe3+ ratios in Antarctica: photochemical processes in ice as well as the effects of organic compounds and the pH of the ice. Kim et al.  demonstrated the possibility for photochemical production of Fe(II) by solar irradiation in simulated glacier ice loaded with iron oxide particles. One can assume that photochemical Fe2+ production would be constant over glacial-interglacial climate cycles and can therefore consider that photochemically produced Fe2+ would lead to an elevated background concentration of Fe2+ in TD ice. As we observe that TD Fe2+ concentrations are sometimes <0.1 ng g−1, we consider photochemical production of Fe2+ to be insignificant compared to the glacial-interglacial variability observed in the record.
 Aside from Fe2+ production, we also consider the stability of Fe2+ in the presence of organic ligands and pH changes. Although pH [Millero, 1995] and organic complexation [Boye et al., 2001] are important factors controlling Fe solubility in seawater, much lower concentrations of organic carbon and hence ligands are observed in Antarctic ice [Antony et al., 2011; Federer et al., 2008]. We must also consider the possibility for Fe(II) to be reoxidized to Fe(III) after deposition in the ice matrix. Fe(II) stability is pH dependent [Bruland, 2001], where low pH tends to stabilize the Fe(II)/Fe(III) system. Mineral dust salts buffer Greenland ice core pH to values of 5.3 to 5.5 [Pasteris et al., 2011], although pH can decrease in the presence of volcanic acids. If Fe(II) is more stable at lower pH values, then our findings of enhanced Fe2+ during the dustier (and therefore of higher pH) glacial are likely to be conservative estimates. The presence of Fe(II) in atmospheric dust particles [Schroth et al., 2009; Zhuang et al., 1992b] further suggests that this species is stable following photoproduction. On the basis of these considerations, we suggest that photochemical production of Fe(II) in Antarctic ice is minimal with respect to dust Fe input, Fe(II) is stable in Antarctic ice on glacial-interglacial time scales, and the somewhat higher pH of Antarctic glacial ice, relative to Holocene ice, may lead to an underestimation of the glacial enhancement of Fe2+.
 We observe three non-LGM instances of Fe2+ enrichment in TD ice, one of which may be attributable to volcanism. The three greatest Fe2+/Fe3+ ratios reported here occurred at 4, 14, and 51 kyr BP, with respective Fe2+/Fe3+ ratios of 2.1, 8.6, and 2.4 (Figure 2). Each of these abrupt changes was due to increased Fe2+ concentrations rather than lowered Fe3+ concentrations. The sample with both the greatest Holocene Fe2+ concentration and the highest observed Fe2+/Fe3+ ratio dates to 14 kyr BP and is contemporaneous with a doubling of sulfate concentrations indicative of volcanism [Severi et al., 2011]. Multiple studies [Ayris and Delmelle, 2012; Duggen et al., 2009] have indicated that dust plumes from volcanic eruptions contain Fe in a highly soluble and thus more bioavailable form but did not determine Fe species. Hence, we consider it likely that this sample contains Fe2+-enhanced dust of volcanic origin. The speciation of this sample may also be controlled by the ice pH, which was lowered due to the presence of sulfate and potentially other volcanic acids. The influence of pH on Fe species stability and preferential molecular formation in ocean waters [Millero, 1995] and glacial ice [Spolaor et al., 2012] suggests the likelihood that volcanic acids promote the formation and retention of Fe(II) in dust. The two other Fe2+/Fe3+ peaks (4 and 51 kyr BP) are not associated with high concentrations of sulfate or any other analytes and hence may simply result from natural variability in the Fe2+/Fe3+ ratios found in TD dust (Figure 2).
 We identified greater concentrations of Fe2+ during the glacial period compared to Fe3+, with particularly enhanced Fe2+/Fe3+ ratios during the LGM. Increased Fe2+ concentrations during the LGM suggest increased deposition of more soluble iron to the Southern Ocean. Consequently, the presence of more bioavailable iron during the LGM could have strengthened the effectiveness of the biological pump, particularly between 14 and 32 kyr BP. The enhancement of Fe2+ in Antarctic ice during the glacial period is consistent with the photochemical formation of Fe2+ during long-range transport of FD (0.7–5 µm diameter). However, secondary post-depositional processes, the effect of aerosol pH, and the presence of organic compounds should be further investigated. We find evidence that Fe in dust from volcanic plumes may be highly soluble due to Fe2+ enrichment, consistent with recent oceanographic Fe solubility studies [Ayris and Delmelle, 2012; Duggen et al., 2009]. Our results support the hypothesis [Baker and Jickells, 2006] that during glacial periods the iron present in long-range transported dust is in a more soluble and bioavailable form than that present during interglacial periods, with the ability to influence primary productivity in the Southern Ocean.
 This work was supported by the University of Siena and IDPA-CNR (Institute for the Dynamics of Environmental Processes, National Research Council). P.V. acknowledges the support of an EU Marie Curie IIF Fellowship (MIF1-CT-2006-039529, TDICOSO). We thank all our colleagues from the University of Venice Environmental Analytical Chemistry Laboratory for instrumental assistance, technical information, and manuscript corrections. Roberto Zuliani is especially thanked for technical assistance and development of the software used for the manifold control. The Talos Dome Ice Core Project (TALDICE), a joint European program, is funded by national contributions from Italy, France, Germany, Switzerland, and the United Kingdom. Primary logistical support was provided by PNRA at TD. This is TALDICE publication n. 32. This work was also supported by funding to the Past4Future project from the European Commission's 7th Framework Programme (grant 243908) and is Past4Future contribution n. 42.