Delivery of anthropogenic bioavailable iron from mineral dust and combustion aerosols to the ocean

Atmospheric deposition of anthropogenic soluble iron (Fe) to the ocean has been suggested to modulate primary ocean productivity and thus indirectly affect the climate. A key process contributing to anthropogenic sources of soluble Fe is associated with air pollution, which acidifies Fe-containing mineral aerosols during their transport and leads to Fe transformation from insoluble to soluble forms. However, there is large uncertainty in our estimate of this anthropogenic soluble Fe. In this study, for the first time, we interactively combined laboratory kinetic experiments with global aerosol modeling to more accurately quantify anthropogenic soluble Fe due to air pollution. Firstly, we determined Fe dissolution kinetics of African dust samples at acidic pH values with and without ionic species commonly found in aerosol water (i.e., sulfate and oxalate). Then, by using acidity as a master variable, we constructed a new empirical scheme for Fe release from mineral dust due to inorganic and organic anions in aerosol water. We implemented this new scheme and applied an updated mineralogical emission database in a global atmospheric chemistry transport model to estimate the atmospheric concentration and deposition flux of soluble Fe under preindustrial and modern conditions. Our improved model successfully captured the inverse relationship of Fe solubility and total Fe loading measured over the North Atlantic Ocean (i.e., 1–2 orders of magnitude lower Fe solubility in northernAfricanthan combustion-influenced aerosols). The model results show a positive relationship between Fe solubility and water-soluble organic carbon (WSOC) / Fe molar ratio, which is consistent with previous field measurements. We estimated that deposition of soluble Fe to the ocean increased from 0.05–0.07 Tg Fe yr in the preindustrial era to 0.11–0.12 Tg Fe yr in the present day, due to air pollution. Over the high-nitrate, low-chlorophyll (HNLC) regions of the ocean, the modeled Fe solubility remains low for mineral dust (< 1 %) in a base simulation but is substantially enhanced in a sensitivity simulation, which permits the Fe dissolution for mineral aerosols in the presence of excess oxalate under low acidity during daytime. Our model results suggest that human activities contribute to about half of the soluble Fe supply to a significant portion of the oceans in the Northern Hemisphere, while their contribution to oceans in high latitudes remains uncertain due to limited understanding of Fe source and its dissolution under pristine conditions.


Introduction
Changes in supply of nutrients such as bioavailable iron (Fe) from the atmosphere to the ocean have altered oceanic carbon uptake, but significant uncertainties remain on the magnitude of this effect (Ciais et al., 2013).Thus improved quantification of atmospheric delivery of bioavailable Fe is essential to quantify the long-term carbon sink (Jickells et al., 2005).The present study focuses on "potentially" bioavailable Fe, which includes colloidal materials and aqueous species (often operationally defined as soluble Fe).The response of the aerosol Fe solubility (i.e., soluble Fe / total Fe) to air pollution is a key uncertainty in our understanding of the biogeochemical cycle of Fe, marine ecosystem, and climate (Mahowald et al., 2009;Shi et al., 2012;Hajima et al., 2014).
Atmospheric processing of mineral dust has been hypothesized to be an important source of soluble Fe to the oceans because of acidic condition in aerosol water (Zhuang et al., Published by Copernicus Publications on behalf of the European Geosciences Union. 1992; Meskhidze et al., 2003).Previous chemical transport models have used mineral dissolution rates and stoichiometric numbers of Fe in minerals to estimate Fe release rates from mineral aerosols (Meskhidze et al., 2005;Solmon et al., 2009;Ito and Feng, 2010;Ito, 2012).Laboratory studies of mineral dust have demonstrated that the Fe release rates used in previous global models were much slower than the measurements during a typical aerosol lifetime of 2-7 days (Mackie et al., 2005;Shi et al., 2011).In laboratory works, the enhanced dissolution rates of elements from phyllosilicate minerals have been observed during the initial period when metals are incongruently dissolved in solution (e.g., Malmström and Banwart, 1997;Brandt et al., 2003).Much slower quasi-steady-state dissolution rates after 10-14 days are typically observed for aluminosilicate minerals in acid solutions (Amram and Ganor, 2005;Lowson et al., 2005;Golubev et al., 2006;Rozalén et al., 2008;Bibi et al., 2011).Recent atmospheric chemical transport models have adopted the initial period of enhanced Fe release rate for the proton-promoted dissolution (Ito and Xu, 2014;Myriokefalitakis et al., 2015).In previous studies which implemented the proton-promoted Fe dissolution with no organic ligand, the Fe dissolution was significantly suppressed due to the dust alkalinity, particularly in the Southern Hemisphere (Meskhidze et al., 2005;Ito and Feng, 2010;Johnson et al., 2010;Ito and Xu, 2014).Currently, ferric sulfate is treated as water-soluble Fe in oil combustion aerosols at emission (Ito, 2013(Ito, , 2015;;Myriokefalitakis et al., 2015;Wang et al., 2015).
Previous laboratory studies suggest that different acid types and photochemical reactions affect proton-promoted Fe dissolution rates of mineral dust, in addition to the types of Fe species associated with mineral source materials (Cwiertny et al., 2008;Fu et al., 2010;Rubasinghege et al., 2010).However, these studies were conducted in the absence of ammonium salts, such as sulfate, which are ubiquitous in aerosol water.Some of the anions in aerosol water are known to be effective inorganic ligands, which form the complexes with Fe in solution (Cornell and Schwertmann, 2003).In batch experiments, the mineral dissolution rate at high dust / liquid ratio can be influenced by a different ability of these anions to form soluble complexes with metals (Hamer et al., 2003).Thus the effect of decrease of the activity of Fe 3+ on the Fe dissolution rates via the formation of aqueous complexes needs to be assessed in laboratory experiments to constrain the degree of suppression used in models.
Recently, global atmospheric transport model studies have emphasized the role of oxalate for promoting Fe dissolution from Fe-containing aerosols (Luo and Gao, 2010;Johnson and Meskhidze, 2013;Ito, 2015;Myriokefalitakis et al., 2015) (see the Supplement).Moreover, oxalate-promoted dissolution of Fe is suppressed at low concentrations of oxalate near strong Fe sources (Ito, 2015), because excess oxalate is necessary to induce significant Fe dissolution (Chen and Grassian, 2013).These modeling studies highlighted the importance of oxalate-promoted Fe dissolution for mineral dust over the remote oceans.However, a constant oxalatepromoted dissolution rate with time for mineral dust has been prescribed in previous models regardless of different dissolution behaviors in different Fe types, due to a lack of experimental data for oxalate-promoted Fe dissolution kinetics.
Here, we conducted a series of laboratory experiments to examine how inorganic and organic ligands in solution (i.e., sulfate and oxalate) affect Fe dissolution rates in mineral dust.The experimental data were then used to derive a new Fe release scheme, which is implemented in a global chemical transport model to quantify the effect of atmospheric processing of mineral aerosols on Fe mobilization.This study incorporates the proton-and oxalate-promoted Fe dissolution schemes for the mineral aerosols in our model (Ito, 2015).We also examine quasi-photo-reductive dissolution scheme for mineral aerosols in a sensitivity simulation.Determination of Fe dissolution for different types of Fe requires three key parameters of Fe release rate, degree of suppression, and Fe content.We implement three-stage kinetic process for the Fe dissolution scheme to dust aerosols.We use the updated version of the mineralogical database for Fe content in soils (Journet et al., 2014).To assess model assumptions for Fe dissolution, the calculated Fe solubility is evaluated against field observations in relation to total Fe loading and water-soluble organic carbon (WSOC) over the North Atlantic Ocean (Wozniak et al., 2013(Wozniak et al., , 2015)).The model estimates "anthropogenic" soluble Fe supply from both dust and combustion sources to the oceans in association with changes in air quality based on the Intergovernmental Panel on Climate Change (IPCC) emission data set.We use the term "anthropogenic" soluble Fe here as primary soluble Fe from oil combustion aerosols and secondary soluble Fe from both dust and combustion aerosols, due to the changes in emissions from fossil fuel use and biofuel combustion between the preindustrial era and the present day (Ito et al., 2014).

Laboratory experiments
In this study, we used the same Tibesti dust sample as in Shi et al. (2011).We followed a similar methodology to that in Shi et al. (2011).Please see the Supplement for more details.In order to determine the Fe dissolution kinetics in dust aerosol water, which contains organic ligands (e.g., oxalate) and a high concentration of inorganic ions (e.g., sulfate), four sets of time-dependent dissolution experiments were performed, as summarized in Table 1.The Fe dissolution experiments include the following: -Experiment 1: a dust / liquid ratio of 1 g L −1 in 0.05 and 0.005 mol L −1 sulfuric acid solution only (i.e., no (NH 4 ) 2 SO 4 ) (pH = 1.3, ionic strength I = 0.15 M; pH = 2.1, I = 0.015 M).The pH values in highly acidic solutions were estimated from molality and the activity coefficient, which were calculated using the E-AIM III aqueous solution simulator (Wexler and Clegg, 2002).The pH values in highly acidic solutions and with high ionic strength (I > 3 M) were estimated from molality and activity coefficient, which were calculated using E-AIM III aqueous solution simulator (Wexler and Clegg, 2002).Thus the calculated pH values are different from molality for H + .
-Experiment 4: a dust / liquid ratio of 60 mg L −1 , 10 g L −1 and 50 g L −1 in 0.005 mol L −1 sulfuric acid solution (pH 2).The pH was continuously monitored during the experiments at four different dust / liquid ratios (i.e., 60 mg L −1 , 1 g L −1 , 10 g L −1 , and 50 g L −1 ), and once the pH change was more than 0.1 pH unit, acids were added to decrease the pH to 2. This aims to determine how different dust / liquid ratios affect the Fe dissolution kinetics in comparison with experiment (1) (1 g L −1 ).

Model description
This study uses the Integrated Massively Parallel Atmospheric Chemical Transport (IMPACT) model (Rotman et al., 2004;Liu et al., 2005;Feng and Penner, 2007;Ito et al., 2007Ito et al., , 2012Ito et al., , 2014Ito et al., , 2015;;Lin et al., 2014;Xu and Penner, 2012;Ito, 2015).The model is driven by assimilated meteorological fields from the Goddard Earth Observation System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO).Simulations were performed with a horizontal resolution of 2.0 • × 2.5 • and 59 vertical layers with a top boundary at 0.01 hPa using meteorological fields for the year 2010 (and 2011 for the comparison with the field measurements).
We run the model with emissions of primary aerosols and precursor gases of secondary aerosols such as sulfate, nitrate, ammonium, and oxalate for the preindustrial era and the present day to disentangle the naturally and anthropogenically perturbed components (Table 2), as described in Ito et al. (2014).The emission data sets for anthropogenic activities such as fossil fuel use and biofuel combustion are taken from the historical emissions for IPCC Fifth Assessment (AR5) report for the preindustrial era and the present day (Lamarque et al., 2010).The present-day estimates for mineral aerosols from arid and semiarid regions as well as combustion aerosols from biomass burning are used together with anthropogenic emission changes (Ito and Xu, 2014;Ito et al., 2015;Ito, 2015).The same natural emissions of dimethylsulfide (DMS), sulfur dioxide (SO 2 ), nitrogen oxides (NO x ), volatile organic compounds (VOCs), and ammonia (NH 3 ) are used for both periods in our simulations, as we use the same meteorological data set.Thus Fe dissolution due to natural acidity is not included in the "anthropogenic" fraction of soluble Fe, as in Ito and Xu (2014).
Previously, Ito and Xu (2014) used the mineralogical database compiled by Nickovic et al. (2012) and Fe content for hematite (69.9 %), illite (4.0 %), smectite (11 %), kaoli- Note: the parentheses represent the percentage of each source of Fe to total Fe.The initial Fe solubility (58 ± 22 %) is used to estimate primary soluble Fe emission for the oil combustion aerosols only (Ito, 2015).We also examined the initial Fe solubility of mineral dust (0.1 %) in an additional sensitivity simulation (Hand et al., 2004;Ito and Xu, 2014).Insoluble Fe can be transformed into secondary soluble Fe via atmospheric processing of Fe-containing mineral dust (see the text) and combustion aerosols (Ito, 2015).
Here, the updated global database of mineral composition and Fe content for hematite (69.9 %), goethite (62.8 %), illite (4.3 %), smectite (2.6 %), kaolinite (0.23 %), chlorite (12.3 %), vermiculite (6.71 %), and feldspars (0.34 %) in clay-and silt-sized soils (CASE 1 in Journet et al., 2014) was used to estimate the emissions of Fe and calcite in dust aerosols.The size distribution at emission follows the mass fractions of emitted soil particles in Kok (2011).The mass fluxes of mineral dust at emission are interpolated to represent four model size bins (radius: < 0.63, 0.63-1.25,1.25-2.5, and 2.5-10 µm) with the theoretical expression (Ito et al., 2012).The mineral fractions in clay-and silt-sized soils are also distributed in the four size bins following the brittle fragmentation theory after Scanza et al. (2015).All the Fecontaining minerals are found in the clay-sized soils, while only three minerals (i.e., goethite, chlorite, and feldspars) are in the silt-sized soils (Journet et al., 2014).Thus Fe content averaged in size bins 1-3 (3.6 %) is higher than the largest one (2.3 %), in contrast to constant Fe content (3.1 %) with size in previous version.As a result, global Fe emission from dust (69 Tg yr −1 ) is slightly smaller than that estimated in the previous version (79 Tg yr −1 ) (Ito, 2015).The sum of Fe emission in the smaller size from bin 1 to bin 3 in this study (25 Tg yr −1 ) is larger than that in the previous version (22 Tg yr −1 ).Consequently, smaller dust particles may transport more Fe to remote regions relative to larger particles.The mineral dust aging process with the formation of soluble materials in aerosol and cloud water (e.g., sulfate, nitrate, ammonium, and oxalate) is explicitly simulated in the model (Ito and Xu, 2014;Ito, 2015).The values of the pH of the aerosol water used in the calculations of the dissolution rates are estimated for all of the wet aerosols in each size bin, as in Ito and Xu (2014).The aqueous-phase chemical reactions for the formation of oxalate are the same as described by Lin et al. (2014), except for the treatment of the Fe chemistry in aerosol and cloud water as in Ito (2015).Thus Fe(III)-oxalate complex is the major form of Fe in modeled solution.The photolysis of Fe-oxalate complex can contribute to a significant oxalate sink in cloud water influenced by ship emissions (Sorooshian et al., 2013;Wang et al., 2014).On the other hand, a complexation of Fe(II) with stronger organic ligands from fossil fuel combustion may be important for the stability of Fe dissolved in rainwater (Kieber et al., 2005;Willey et al., 2015).Here, we focus on the acid mobilization of relatively insoluble Fe in Fe-containing minerals to soluble Fe.Because of the lack of knowledge regarding the specific ligands and formation rates of Fe-organic complexes, Fe chemistry is disabled in cloud and rainwater but implemented for Fe-containing wet aerosols in four size bins to obtain good agreement regarding oxalate with the observations over the ocean (see Fig. S3 in Ito, 2015).The deposition velocities of soluble Fe depend on the aerosol types and size bins, and they follow the aging of the parent aerosols in the atmosphere (Ito and Xu, 2014).Here, we developed a new Fe release scheme in mineral dust (see Sect. 4) to improve the previously used scheme (Ito and Xu, 2014).The new scheme is implemented in the IMPACT model for simulations in Sect. 5.

Development of a new Fe dissolution scheme based on new experimental results
In our model, Fe release from aerosols due to chemical processing is calculated based on an online simulation of aqueous-phase chemistry (Ito and Feng, 2010;Ito, 2012Ito, , 2015;;Ito and Xu, 2014).Ito and Xu (2014) developed an Fe dissolution scheme that considers the types of Fe species associated with mineral source materials, mainly based on the measurements by Shi et al. (2011).Following their studies, three Fe pools are characterized by ferrihydrite, nano-sized Fe oxides, and heterogeneous inclusion of nano-Fe grains in aluminosilicates (e.g., illite, smectite, and chlorite).Here we developed a new Fe dissolution scheme, which considers our laboratory experimental data sets regarding (1) the formation of Fe inorganic and organic complexes in solution and (2) the formation of surface complexes between oxalate and Fe oxides, following Ito (2015) for combustion aerosols.
Figures 1 and S1 in the Supplement demonstrate the effects of inorganic anions to form soluble complexes with Fe at different dust / liquid ratios on dissolution rates measured in acidic solution.Figure 1 shows that the pH = 2 (0.05 M H 2 SO 4 , 1 M (NH 4 ) 2 SO 4 , red triangles) and pH = 2.1 (0.005 M H 2 SO 4 , green diamonds) cases at 1 g dust L −1 have significantly different Fe dissolution rates.The one with high ionic strength (red triangles, Fig. 1) has a much higher dissolution rate than predicted by the small difference in the pH values.It is expected that one pH unit can lead to 3-4 times difference in dissolution rates as shown here and by Shi et al. (2011).The presence of complexing ions such as sulfate in this case has the potential to accelerate the dissolution rate by absorption or by complexation with Fe dissolved in solution (Cornell and Comparison of Fe solubility in solution (%) measured at two different dust / liquid ratios of 1 g L −1 (red triangles; pH = 2) and 10 g L −1 (blue squares; pH = 2) in 0.05 mol L −1 sulfuric acid solution with 1 mol L −1 (NH 4 ) 2 SO 4 (I = 3.15 mol kg −1 ), and dust / liquid ratios of 60 mg L −1 (black circles; pH = 2.1) and 1 g L −1 (green diamonds; pH = 2.1) in 0.005 mol L −1 sulfuric acid solution without (NH 4 ) 2 SO 4 .mann, 2003).We observed a good agreement of measurements between 1 g L −1 dust in 0.05 mol L −1 sulfuric acid solution with 1 mol L −1 (NH 4 ) 2 SO 4 (pH = 2.0, red triangles) and 60 mg L −1 dust in 0.005 mol L −1 sulfuric acid solution without (NH 4 ) 2 SO 4 (pH = 2.1, black circles).Thus the solution remains undersaturated with respect to Fe(III), because essentially all aqueous Fe(III) species (> 99 %) are complexed with sulfate (i.e., FeSO + 4 ) in 0.05 mol L −1 sulfuric acid solution with 1 mol L −1 (NH 4 ) 2 SO 4 (pH = 2.0) (Meskhidze et al., 2005;Ito, 2015).The higher dust / liquid ratio at 10 g L −1 (blue squares) exhibits lower Fe dissolution rate after the initial period of enhanced Fe release rate, possibly due to the re-adsorption of solution phase Fe onto the particulate phase, as the solution with (NH 4 ) 2 SO 4 is undersaturated with respect to Fe(III) (Spokes and Jickells, 1996;Bibi et al., 2011).
Our data indicate that the addition of complexing agents (i.e., sulfate in Fig. 1 and chlorite in Fig. S1) accelerated dissolution of Fe minerals by binding Fe released from the surface in solution (Zhang et al., 1985;Xu and Gao, 2008).Almost identical slopes were found between 1 g dust L −1 in 0.05 mol L −1 sulfuric acid solution with 1 M (NH 4 ) 2 SO 4 (pH = 2.0, red triangles) and 60 mg dust L −1 in 0.005 mol L −1 sulfuric acid solution without (NH 4 ) 2 SO 4 (pH = 2.0, black circles) during a typical aerosol lifetime (Fig. 1).Thus we chose dust / liquid ratios of 1 g dust L −1 in sulfuric acid solution with 1 M ammonium sulfate to represent a proton-promoted Fe dissolution scheme for mineral dust.
Figure 2 demonstrates the impact of oxalate on Fe dissolution rate (black circles).The addition of 0.03 M Na 2 C 2 O 4 accelerated the dissolution of Fe in 1 g dust L −1 solution, 0.05 M H 2 SO 4 or 0.005 M H 2 SO 4 , with 1 M (NH 4 ) 2 SO 4 .
Dissolved Fe concentration was 60 % higher at 72 h in the 0.05 M H 2 SO 4 and 1 M (NH 4 ) 2 SO 4 dust suspensions with oxalate.It was over 100 % higher at 72 h in the 0.005 M H 2 SO 4 and 1 M (NH 4 ) 2 SO 4 ) dust suspensions when added oxalate.The higher activity of protons can facilitate the oxalate-promoted dissolution process by protonating the hydroxyl (OH) groups at the surface of hydrous Fe oxides, thereby contributing to increasing the number of positively charged surface sites.This increase promotes ligand adsorption, and weakening of the Fe-O bond, which permits the reaction between oxalate and Fe sites by ligand exchange (Cornell and Schwertman, 2003;Ramos et al., 2014).Although the protonation of the ligands in solution increases at higher activity of protons, the adsorption mechanism of HC 2 O − 4 involves the loss of a proton during the ligand-exchange adsorption reaction or during the transfer process from bulk solution to the mineral surface (Yoon et al., 2004).Consequently, the amount of absorbed complex on the mineral surface is higher in solutions at pH < 7 when the overall charge at the reactive surface sites is positive, compared to that at lower activity of protons (Zhang et al., 1985;Xu and Gao, 2008;Lanzl et al., 2012;Ramos et al., 2014).The surface binding sites for adsorbed oxalate become saturated at high proton and high oxalate concentrations, and thus oxalatepromoted dissolution rates are almost independent of pH for mineral dust (Yoon et al., 2004;Cama and Ganor, 2006;Lanzl et al., 2012).
Experimental data in Fig. 3 demonstrate that the Fe release under higher dust / solution ratios is suppressed when the solution becomes supersaturated with respect to Fe(III), as observed by the decrease in the rate (black circles).At a low dust / liquid ratio of 60 mg L −1 at pH 2.1 (H + concentration of 0.01 mol L −1 ), Fe dissolution continued even after 800 h.The rate of Fe dissolution decreased substantially with increasing dust / liquid ratio.At a dust / liquid ratio of 50 g L −1 , Fe dissolution stopped at 180 h and only 1.2 % (10 µmol g −1 ) of the total Fe was dissolved, half of which released in the first hour.The calculated thermodynamic solubility of this Fe pool at pH 2 is −3.34 (mol L −1 on a log scale).This is comparable to the previously measured solubility of nanogoethite, which is −3.6 (mol L −1 ) at pH = 2 (see Fig. 7 in Shi et al., 2011).Thus this Fe pool likely includes ferrihydrite and reactive nano-Fe oxides aggregated on the mineral surface.This is consistent with higher Fe solubility of 1.2 % as compared to that of the highly reactive Fe (0.63 %) (Shi et al., 2011).Only 3.3 % of Fe was dissolved at a dust / liquid ratio of 10 g L −1 and pH 2, and Fe dissolution stopped at 180 h.The calculated solubility of the second Fe pool is −3.55 (mol L −1 ).The above experimental data set is used to determine the model parameters to predict the influence of solution saturation state on the Fe dissolution rates (see Table 3 and discussion below).
Based on above laboratory results (Figs. 2 and 3), Fe dissolution from mineral dust aerosols is treated explicitly as a kinetic process that depends on the pH, ambient tempera-   (Bibi et al., 2014).c m i is the reaction order with respect to aqueous phase protons, which was determined by linear regression from our experimental data in the pH range between 2 and 3 for proton-and oxalate-promoted dissolution schemes.d K eq is the equilibrium constant (mol 2 kg −2 ) (Bonneville et al., 2004;Jang et al., 2007).e n i is the stoichiometric ratio (Bonneville et al., 2004;Jang et al., 2007).The stoichiometric number of moles of Fe per mole of mineral is empirically determined for Fe dissolution rate as in the Eq. ( 1).f The formation of the amorphous Fe(OH) 3 (s) suppresses the oxalate-promoted dissolution from mineral aerosols in the base simulations, while no such effect was considered for quasi-light-induced reductive dissolution in the sensitivity simulation (i.e., f 3 = 1).
ture, the degree of solution saturation, and competition for oxalate between surface Fe and dissolved Fe in our model (Table 3).The net Fe dissolution rates ( RFe i in units of moles of dissolved Fe per Fe gram of Fe-containing mineral particle per second) for the proton-promoted (i = 1), oxalatepromoted (i = 2), and quasi-photo-reductive (i = 3) Fe dissolution schemes can be empirically described using the following equation, which is similar to the formulation applied for Fe-containing minerals (Zhang et al., 1985;Lasaga et al., 1994;Hamer et al., 2003;Meskhidze et al., 2005;Lanzl et al., 2012;Ito and Xu, 2014;Ito, 2015): where k i is the "far-from-equilibrium" (i.e., f i = 1 and g i = 1) Fe release rate (moles Fe g −1 s −1 ), a(H + ) is the H + activity, m i represents the empirical reaction order for protons, and f i and g i account for the suppression.The Fe release rate, k i , is estimated for the proton-and oxalate-promoted dissolution schemes by fitting the parameters to our measurements in sulfuric acid and ammonium sulfate (experiment 2 and 3) with and without oxalate (Fig. 2).Fe release from mineral dust under acidic conditions is characterized by initial rapid Fe release and subsequent slow Fe release (Desboeufs et al., 1999;Mackie et al., 2005;Cwiertny et al., 2008;Shi et al., 2011).Since the typical lifetime of mineral dust is about a week, the initial rapid Fe release rates are important for the atmospheric processing of mineral dust.A three-stage kinetic model is used to describe the Fe release behavior of mineral dust: the first stage is characterized by a rapid dissolution of hydrous ferric oxide (HFO) on the surface of minerals, the second stage is an intermediate stage of nano-sized Fe oxide dissolution from the surface of minerals, and the third stage is the Fe release from fine-grained materials, which are internally mixed with aluminosilicate particles, as the mineral surface is slowly dissolved.Here, we prescribe the content of HFO (0.65 %) and nano-sized Fe oxides (1.3 %) on the surface of minerals.The content of HFO is similar to that of the highly reactive Fe measured on the mineral surface and within the range of the first Fe pool from 0.5 to 2.9 % (Shi et al., 2011).The content of nano-sized Fe oxides is also within the range of the second Fe pool from 1.0 to 3.5 %.The proton-promoted dissolution rates for stage I and II are strongly dependent on pH.A comparable strong dependence on pH was also reported for nanosized hematite (Lanzl et al., 2012).The similarity in our Fe release rates to those of illite suggests that Fe is mainly released from the reactive surface on Fe-containing minerals by similar mechanisms to aluminosilicates (Fig. S2), which involve inward movement of dissolution from the grain edges (Brandt et al., 2003;Rozalén et al., 2008).Consequently, by applying Eq. (1) for mineral dust, which was previously applied for combustion aerosols (Ito, 2015), we can avoid the need to explicitly treat individual Fe-containing minerals to represent dissolution processes that occur on the timescale of dust aerosol lifetime.
The enhanced mineral dissolution in the presence of chelating ligands such as oxalate has been attributed to both the ligand adsorption at the surface of mineral and complex formation in solution (Drever and Stillings, 1997).The differences between with (i.e., measured overall rate) and without oxalate (i.e., proton-promoted dissolution rate) can be attributed to the surface complexation (i.e., oxalate-promoted dissolution rate).We observed rates of oxalate-promoted dissolution to be almost independent of pH for stage I and III (Table 3), but could not calculate the empirical reaction order for stage II from the measurements.Lanzl et al. (2012)  found that this value was nearly independent of pH (< 4.0) and particle size between 8 and 90 nm hematite (Zhang et al., 1985).Thus we apply the same value to stage II as in stage I.The calculations (red squares) reproduce the initial rapid Fe release and subsequent slow Fe release due to the proton-and oxalate-promoted Fe dissolution.The calculations also reproduce the enhancement in the Fe solubility due to the effects of oxalate under acidic conditions, compared with proton-promoted dissolution, as was observed for Arizona test dust in previous study (Chen and Grassian, 2013).The activation energy for k i is described by a function of pH and temperature for soils (Bibi et al., 2014).The Fe release is suppressed by the degree of saturation, f i , and competition for oxalate between surface Fe and dissolved Fe, g i (Ito, 2015).We apply the same equation for mineral dust, g i , as in Ito (2015).Thus excess oxalate is needed to form mononuclear bidentate ligand with surface Fe and promote Fe dissolution significantly (Chen and Grassian, 2013).The function f i (0 ≤ f i ≤ 1) is given by .The red squares are calculated using Eq. ( 1) from the equilibrium constant (mol 2 kg −2 ) used in this study at each hour.The black circles are our measured data.The fraction of total dissolved Fe present as Fe(III) is prescribed at pH = 2 (0.2) in this calculation to emulate the experimental conditions, while the photochemical redox cycling between Fe(III) and Fe(II) in solution is explicitly simulated in our global model (Lin et al., 2014).The large fraction of Fe(II) in solution under the dark conditions is likely associated with the Fe dissolution of Fe(II)-containing solids (Cwiertny et al., 2008).The initial fraction of Fe speciation is not critical in estimating the Fe redox speciation in aerosol water, because Fe(II) is quickly oxidized to Fe(III) in oxygenated water (Deguillaume et al., 2010).
in which a Fe is the concentration of Fe 3+ in aerosol water (mol L −1 ), n i is the stoichiometric ratio, and K eq i is the equilibrium constant measured.Results of previous laboratory experiments in batch experiments showed that oxalate had negligible effect on the Fe dissolution of hematite at higher pH values (≥ pH = 5) under dark conditions (Zhang et al., 1985;Xu and Gao, 2008;Lanzl et al., 2012).The decrease in proton concentration (e.g., during activation into cloud droplets or neutralization by carbonate) could lead to formation of the amorphous Fe(OH) 3 (s) that coats on the mineral surfaces (Shi et al., 2009(Shi et al., , 2015) ) and inhibits both the adsorption of oxalate and the detachment of surface Fe-oxalate (Zhang et al., 1985;Jang et al., 2007;Rozalén et al., 2014).This effect was not considered in previous modeling studies for mineral dust, leading to an overestimation of oxalate-promoted Fe dissolution in mineral dust at  (Luo and Gao, 2010;Johnson and Meskhidze, 2013;Myriokefalitakis et al., 2015).Paris et al. (2011) concluded that the light-induced reductive dissolution was not the principal process to explain the increase in Fe solubility under low dissolved Fe and low oxalate concentrations.However, the Fe release from the reactive surface on Fe-containing minerals was observed at higher pH values (≥ 5) under high dissolved Fe and high oxalate concentrations and irradiation conditions (Lanzl et al., 2012;Chen and Grassian, 2013).Two rate-limiting steps are possibly involved in apparent contradictions at higher pH: the adsorption of oxalate on the oxide surfaces at low dissolved Fe concentration (i.e., far from dissolution equilibrium) and the detachment of surface Fe-oxalate via photoinduced ligand-to-metal charge transfer as dissolution equilibrium is approached (Kraemer and Hering, 1997).To examine the uncertainty in Fe release at higher pH values, no pH effect on the suppression of quasi-photo-induced Fe dissolution for mineral aerosols was considered in the sensitivity simulation.In analogy to the combustion aerosols (Chen and Grassian, 2013;Ito, 2015), we apply the same rate constant to quasi-photo-induced dissolution as in oxalate-promoted dissolution and we set f 3 = 1 in Eq. ( 1).The quasi-photoinduced dissolution rate is calculated by scaling the photolysis rate of H 2 O 2 estimated in the model, following Ito (2015).The photo-degradation of oxalate due to photolysis of Feoxalate complexes is simulated for Fe-containing aerosols in aqueous chemistry (Lin et al., 2014;Ito, 2015).

Modeling results and discussion
The model-calculated concentrations of total and soluble Fe in aerosols have been extensively compared with field observations (Ito and Feng, 2010;Ito, 2012Ito, , 2013Ito, , 2015;;Ito and Xu, 2014).Here, model-calculated daily average surface concentrations of soluble Fe (red squares) were compared with the measurements (black circles) during the 2010 and 2011 US GEOTRACES cruises over the North Atlantic (Fig. 4) (Wozniak et al., 2013(Wozniak et al., , 2015)).The model-calculated variability at each latitude and longitude represents the daily variability during the sampling dates.The modeled soluble Fe concentration exhibits a latitudinal variability which is similar to that of the with low values over the remote ocean (< 0.5 ng m −3 ), intermediate values near the European continent, and high values near northern Africa (> 5 ng m −3 ) (Fig. 4a).The modeled soluble Fe concentration shows a longitudinal variability which is also similar to that of the measurements, with low values over the remote ocean (< 0.5 ng m −3 ), intermediate values near North America, and high values near northern Africa (> 2 ng m −3 ) (Fig. 4b).
Figure 5 displayed the daily averaged, model-calculated surface total aerosol Fe loading, Fe solubility, Fe / WSOC molar ratio, and dust / combustion ratio for soluble Fe (red  (Wozniak et al., 2013(Wozniak et al., , 2015)).The number of modeled data points ( 84) is larger than the measurements (37), because each daily average is calculated for each sampling date at each center of cruise location.
squares) over the 2010 and 2011 US GEOTRACES cruise tracks in comparison with the measurements (black circles) by Wozniak et al. (2013Wozniak et al. ( , 2015)).As previously reported in Ito (2013), the oil combustion from shipping mainly contributes to high Fe solubility at low Fe loading observed over the high-latitude North Atlantic Ocean (Fig. 5a).In this study, low Fe solubility near northern Africa was successfully simulated.The internal mixing of alkaline minerals with Fe-containing minerals in aqueous chemistry for mineral dust can lead to higher pH and thus suppress the Fe dissolution near the source regions, compared to the external mixing (Ito and Feng, 2010).The model predicts relatively higher pH values for dust aerosols except submicron particles near the source regions, because the dust alkalinity reservoir (i.e., calcite) is able to buffer the acidification (Fig. S3).While our model has incorporated the initial rapid Fe release rate in acid solutions with oxalate explicitly, the comparisons with observations support the suppression of Fe dissolution under low acidity and low oxalate concentration near the source region of dust aerosols.
The Fe / WSOC molar ratios in aerosols influenced by combustion aerosols are 2-3 orders of magnitude lower than those near northern Africa, which is also consistent with the observations (Wozniak et al., 2013(Wozniak et al., , 2015) ) (Fig. 5b).The averaged WSOC concentration in our model (330 ± 470 ng m −3 ) is consistent with the measurements (330 ± 290 ng m −3 ).The higher Fe solubility measured in water (pH = 5.5) for the excess WSOC with Fe-binding functionalities (e.g., -COOH, -NH 2 ) may suggest a potential role of the organic  Wozniak et al. (2013Wozniak et al. ( , 2015)).The number of modeled data points ( 84) is larger than the measurements (37), because each daily average is calculated for each sampling date at each center of cruise location.
compounds in aerosols for the delivery of Fe to the ocean in soluble form (Wozniak et al., 2013(Wozniak et al., , 2015)).Here, similar plots can be obtained even with a constant WSOC concentration at 330 ng m −3 (Fig. S4).The results indicate that the variability in Fe solubility is nearly independent of the variability in WSOC concentration, possibly because of the excess ligands to stabilize Fe in solution at low Fe loading.The use of our process-based model demonstrates that chemical reactions and mixing with combustion aerosols are the main mechanisms to cause the high Fe solubility at low Fe loading in the North Atlantic (Fig. 5c).As previously discussed in Ito (2013), this is consistent with the observations (e.g., Sedwick et al., 2007;Séguret et al., 2011).The suppression of Fe dissolution under low proton and low oxalate concentrations leads to the lower Fe solubility of mineral dust deposited to the ocean (0.64-0.71 %) as a global mean in the present day (Table 4), compared to those estimated for mineral dust by previous modeling studies (1.4-15 %; see Table 5 in Hajima et al., 2014).The Fe solubility for mineral dust varies spatially over the remote oceans, due to the proton-and oxalate-promoted and quasi-photoreductive Fe dissolution (Fig. 6).The Fe solubility ranges from 0.75 to 2 % over the North Atlantic and Pacific in the present day (Fig. 6a and c), which is relatively consistent with that used in conventional ocean biogeochemical models (1-2 %) (Jickells et al., 2005).The base simulations result in low Fe solubility (< 1 %) over the Southern Ocean in the present day and significant portions of the ocean in the preindustrial era, due to the suppression of Fe dissolution under low proton concentrations (Fig. 6a and b).In contrast, the sensitivity simulations for mineral aerosols (i.e., f 3 = 1) lead to higher Fe solubility (> 1 %) deposited to the remote oceans of high-nitrate, low-chlorophyll (HNLC) regions such as the subarctic North Pacific, the eastern equatorial Pacific, and the Southern Ocean when quasi-photo-reductive dissolution was considered at higher pH values (Fig. 6c and d).This is reflected in higher contribution of oxalate-promoted (i = 2 and 3) dissolution to total soluble Fe deposition in the sensitivity simulations for mineral aerosols, compared to that in the base simulations (i = 2) (Fig. S5).We note that higher contribution of proton-promoted dissolution (i = 1) near the source regions may include the effect of oxalate on Fe dissolution via the suppression of mineral dissolution as well as the soluble Fe content at emission.As a result, the protonpromoted dissolution scheme contributed the majority of soluble Fe deposition to the ocean, 90 % for the base case and 69 % for the sensitivity case.Since this Fe dissolution is not only due to the proton-promoted dissolution by definition, we also examined the effect of different assumption on the initial period of enhanced Fe release in an additional sensitivity simulation, in which we use 0.1 % for the initial Fe solubility of mineral dust (Hand et al., 2004;Ito and Xu, 2014).The model results show that the contribution of dissolution scheme to total soluble Fe deposition depends on the assumption on the initial Fe solubility at emission (Fig. S6).The model results suggest that the initial soluble Fe content from dust source regions such as South America (Patagonia), Australia, and southern Africa may be important for the supply of soluble Fe to the Southern Ocean.The proton-promoted dissolution scheme contributed 77 % for the additional sensitivity case, which is between our base and sensitivity simulations.
The annually averaged rate of deposition of soluble Fe from dust and combustion sources to the oceans is presented in Fig. 7a and d for the base and sensitivity simulations, respectively.The total Fe solubility (Fig. 7b and e) is higher than that calculated from dust only over significant portions of the open ocean.However, our modeled Fe solubility (0.1-0.5 and 0.2-0.7 % for the base and sensitivity simulations, respectively) is still low over the South Atlantic east downwind from the Patagonian dust source regions where previous modeled Fe solubility deposited to the ocean (1.4-2.0 % by Mahowald et al., 2009;0.5-0.6 % by Johnson et al., 2010) was significantly lower than that deduced from observations (7.5-20 % by Baker et al., 2013).Our modeled Fe solubility for dry deposition over the Atlantic (1.1 % ± 1.9 % and 1.2 % ± 2.0 %) is in good agreement with the measurement (2.1 % ± 2.2 %), while that for wet deposition (3.4 % ± 3.2 % and 3.6 % ± 3.3 %) is significantly lower than the measurement (10.4 % ± 4.6 %) (Baker et al., 2013)   monthly averaged Fe solubility (1-4 %) in wet deposition is an order of magnitude lower than that observed on the Kerguelen Islands in South Indian Ocean (82 % ± 18 % by Heimburger et al., 2013).This enhanced solubility may be due to unidentified reactive organic species in cloud and rainwater which contain Fe-binding functionalities (e.g., -COOH, -NH 2 ) such as humic-like substances from biomass burning and biologically derived materials from the ocean (Parazols et al., 2006;Deguillaume et al., 2014;Ito et al., 2014Ito et al., , 2015)).The role of humic-like substances in the complexation and dissolution of Fe oxides over a wide pH range has received considerable attention in recent literatures (Al-Abadleh, 2015).The multiple ligands with high affinity for Fe binding can wrest aqueous Fe from any Fe-oxalate complexes, allow the oxalate ligand to react with the surface Fe oxides, and assist the Fe-oxalate detachment from the surface Fe oxides at intermediate pH (e.g., pH = 5) (Cheah et al., 2003).Thus the consumption of oxalate due to photolysis of the Fe-oxalate complex may be limited due to complexation with stronger ligands in atmospheric water.On the other hand, functional groups on the humic molecule are less protonated at pH > 4 and increase the probability of coating of organic matters on the reactive mineral surfaces, and thus they inhibit the oxalate-promoted dissolution (Drever and Stillings, 1997).Clearly, more work is required to elucidate the underlying mechanisms that promote Fe dissolu- tion in cloud and rainwater over the Southern Ocean in future studies.
The contributions of anthropogenic soluble Fe deposition to the present day are examined in Fig. 7c and f for the base and sensitivity simulations, respectively.The soluble Fe deposition from both mineral dust and fossil fuel combustion sources due to changes in atmospheric pollution contributes more than half of the total soluble Fe deposition over significant portions of the open ocean in the Northern Hemisphere for the base simulations.The sensitivity simulations for mineral aerosols (i.e., f 3 = 1) lead to higher soluble Fe deposition in both the present day and the preindustrial era (Table 4) and result in lower anthropogenic soluble Fe deposition to the HNLC regions, due to almost no pH dependency of the quasiphoto-reductive dissolution.Of the total soluble Fe deposition from anthropogenic sources (excluding biomass burning) to the ocean (0.05-0.06 Tg Fe yr −1 ), 67-72 % is from dust sources and 28-33 % is from fossil fuel combustion (Table 4).In contrast, our model indicated higher contribution of biomass burning aerosols in the preindustrial era (42-55 % in total soluble Fe deposition to the ocean).However, significant uncertainties remain on the magnitude of this source strength (Luo et al., 2008;Ito, 2011Ito, , 2012Ito, , 2015;;Wang et al., 2015).It should be noted that anthropogenic soluble Fe input is also sensitive to the soluble Fe content at emission.The additional sensitivity simulation with the initial Fe solubility (0.1 %) for dust aerosols indicates smaller contribution of anthropogenic component near the source regions (Fig. S7).
Our estimate of total Fe deposition to the ocean (10 Tg Fe yr −1 ) is within the range of other models (Table 5).(Hajima et al., 2014).
Our estimates of soluble Fe deposition to the oceans (0.05 and 0.07 Tg Fe yr −1 from the base and sensitivity simulations, respectively) in the preindustrial era are in good agreement with that of Myriokefalitakis et al. (2015) (Table 5).The ratio of present-day soluble Fe deposition to the preindustrial era from the base simulation (47 %) is in good agreement with that estimated by Luo et al. (2008) (46 %), despite the use of various dissolution schemes, emission data sets, and atmospheric transport models.The sensitivity simulations, which include the quasi-photo-reductive dissolution for mineral aerosols (i.e., f 3 = 1), result in relatively small increases in the soluble Fe deposition in the global ocean (0.077 and 0.122 Tg Fe yr −1 as a global mean in the preindustrial era and present day, respectively).The global deposition is similar between our base and sensitivity simulations, mainly because of the suppression of proton-promoted (i.e., f 1 ) and oxalatepromoted (i.e., g 2 ) dissolution near the strong source regions of mineral dust.This is reflected in lower Fe solubility (1.1-1.2 % as a global mean), compared to those estimated by previous modeling studies (1.4-15 %; see Table 5 in Hajima et al., 2014).

Conclusions
We have developed a new scheme of Fe dissolution, which reproduced the proton-and oxalate-promoted dissolution behaviors of our experimental results for mineral dust.The batch dissolution experiments provided the Fe dissolution rates under far-from-equilibrium conditions and the parameters for degree of suppression as the saturation state approached equilibrium.Our model reproduced the slope of Fe solubility vs. total Fe loading measured over the North Atlantic Ocean.To investigate the uncertainty in the Fe dissolution associated with organic compounds, no pH effect on the suppression of quasi-photo-induced Fe dissolution for mineral aerosols was considered in the sensitivity simulations.The Fe release scheme allows us to reduce the number of mineral tracers for implementation in the Earth system models.At lower proton activity and lower oxalate concentration in aerosol water on dust particles near major dust sources, both protons and oxalate had no significant effect on the Fe dissolution in our model and thus resulted in the lower Fe solubility of mineral dust deposited to the ocean (0.64-0.71 %) as a global mean in the present day.Fe release under more acidic conditions in aerosol water due to air pollution resulted in significant increases in soluble Fe deposition over large portions of the open ocean in the Northern Hemisphere.
In our model, low Fe solubility is estimated for mineral dust (< 1 %) over the Southern Ocean downwind from the dust source regions.The differences between our base and sensitivity simulations for mineral aerosols are notable for low Fe loading over the remote HNLC regions such as the subarctic North Pacific, the eastern equatorial Pacific, and the Southern Ocean.However, the differences in Fe solubility between different simulations are generally smaller than the differences between different model estimates and measurements.A comprehensive comparison of model-predicted Fe and its related species with observations is therefore needed to elucidate the high Fe solubility observed under pristine conditions.
The Supplement related to this article is available online at doi:10.5194/acp-16-85-2016-supplement.

Figure 3 .
Figure 3.Comparison of Fe solubility in solution (%) predicted using Eq.(1) with the measured Fe dissolution rates at pH = 2, 0.005 M H 2 SO 4 , and dust / solution of (a) 60 mg L −1 , (b) 1 g L −1 , (c) 10 g L −1 , and (d) 50 g L −1 .The red squares are calculated using Eq.(1) from the equilibrium constant (mol 2 kg −2 ) used in this study at each hour.The black circles are our measured data.The fraction of total dissolved Fe present as Fe(III) is prescribed at pH = 2 (0.2) in this calculation to emulate the experimental conditions, while the photochemical redox cycling between Fe(III) and Fe(II) in solution is explicitly simulated in our global model(Lin et al., 2014).The large fraction of Fe(II) in solution under the dark conditions is likely associated with the Fe dissolution of Fe(II)-containing solids(Cwiertny et al., 2008).The initial fraction of Fe speciation is not critical in estimating the Fe redox speciation in aerosol water, because Fe(II) is quickly oxidized to Fe(III) in oxygenated water(Deguillaume et al., 2010).

Figure 4 .
Figure 4. Comparison of simulated (red squares) and observed (black circles) soluble Fe concentration (ng m −3 ) during the (a) 2010 and (b) 2011 US GEOTRACES cruises over the North Atlantic(Wozniak et al., 2013(Wozniak et al., , 2015)).The number of modeled data points (84) is larger than the measurements (37), because each daily average is calculated for each sampling date at each center of cruise location.

Fe
Figure 5. (a) Atmospheric loading of total aerosol Fe (ng m −3 ) versus Fe solubility for model estimates (red squares) and measurements (black circles) over the cruise tracks.(b) The Fe / WSOC molar ratio versus Fe solubility for model estimates (red squares) and measurements (black circles) over the cruise tracks.(c) The dust / combustion ratio for soluble Fe versus percent of soluble Fe in total Fe for model results over the cruise tracks.The measurements are obtained fromWozniak et al. (2013Wozniak et al. ( , 2015)).The number of modeled data points (84) is larger than the measurements (37), because each daily average is calculated for each sampling date at each center of cruise location.

Figure 6 .
Figure6.Ratio (%) of the soluble to total Fe deposition for mineral dust in the (a) present day from base simulations, (b) the preindustrial era from base simulations, (c) the present day from sensitivity simulations, and (d) the preindustrial era from sensitivity simulations.The formation of the amorphous Fe(OH) 3 (s) suppresses the oxalate-promoted dissolution from mineral aerosols in the base simulations, while no such effect was considered for quasi-light-induced reductive dissolution in the sensitivity simulation (i.e., f 3 = 1).

Figure 7 .
Figure 7. Deposition of soluble Fe (ng Fe m −2 s −1 ) from dust and combustion sources to the oceans in the present day (a) from base simulations and (d) from sensitivity simulations, Fe solubility in the present day (b) from base simulations and (e) from sensitivity simulations, and ratio of increase from the preindustrial era to present day to soluble Fe deposition in the present day (c) from base simulations and (f) from sensitivity simulations.The formation of the amorphous Fe(OH) 3 (s) suppresses the oxalate-promoted dissolution from mineral aerosols in the base simulations, while no such effect was considered for quasi-light-induced reductive dissolution in the sensitivity simulation (i.e., f 3 = 1).

Table 1 .
Summary of Fe dissolution experiments performed in this study.

Table 2 .
Global Fe emission (Tg Fe yr −1 ) estimated for different types of Fe-containing aerosols. Schwert-

Table 3 .
Ito (2015)used to calculate Fe dissolution rates for mineral dust, based on laboratory experiments.ki(pH, T ) is the pH-and temperature-dependent "far-from-equilibrium" Fe dissolution rate of Fe-containing mineral dust (moles Fe g −1 s −1 ) for each Fe dissolution scheme i.The parameters are fit to our measurements for African dust.The photo-induced dissolution rate of Fe compounds is scaled to the photolysis rate of H 2 O 2 calculated in the model, followingIto (2015).b E(pH) = −1.56× 10 3 × pH + 1.08 × 10 4 .The parameters are fit to the measurements for soils a

Table 4 .
Annual deposition rates of total and soluble Fe (Tg Fe yr −1 ) from different sources to the ocean in the base, sensitivity, and additional sensitivity simulations.
* Assuming an initial Fe solubility of 0.1 %.

Table 5 .
Comparison of total and soluble Fe deposition to the oceans (Tg Fe yr −1 ) from different studies for the preindustrial era and the present day.
*The values are taken from a compilation of literatures