Pre‐Industrial, Present and Future Atmospheric Soluble Iron Deposition and the Role of Aerosol Acidity and Oxalate Under CMIP6 Emissions

Atmospheric iron (Fe) deposition to the open ocean affects net primary productivity, nitrogen fixation, and carbon uptake. We investigate changes in soluble Fe (SFe) deposition from the pre‐industrial period to the late 21st century using the EC‐Earth3‐Iron Earth System model. EC‐Earth3‐Iron considers various sources of Fe, including dust, fossil fuel combustion, and biomass burning, and features comprehensive atmospheric chemistry, representing atmospheric oxalate, sulfate, and Fe cycles. We show that anthropogenic activity has changed the magnitude and spatial distribution of SFe deposition by increasing combustion Fe emissions and atmospheric acidity and oxalate levels. We report that SFe deposition has doubled since the early industrial era, using the Coupled Model Intercomparison Project Phase 6 emission inventory. We highlight acidity as the main solubilization pathway for dust‐Fe and oxalate‐promoted processing for the solubilization of combustion‐Fe. We project a global SFe deposition increase of 40% by the late 21st century relative to present day under Shared Socioeconomic Pathway (SSP) 3–7.0, which assumes weak climate change mitigation policies. Conversely, SSPs with stronger mitigation pathways (1–2.6 and 2–4.5) result in 35% and 10% global decreases, respectively. Despite these differences, SFe deposition increases over the equatorial Pacific and decreases in the Southern Ocean (SO) for all SSPs. We further observe that deposition over the equatorial Pacific and SO are highly sensitive to future changes in dust emissions from Australia and South America, as well as from North Africa. Future studies should focus on the potential impact of climate‐ and human‐induced changes in dust and wildfires combined.

2 of 21 CO 2 depends partly on ocean net primary productivity (NPP), that is, the rate of photosynthetic carbon fixation minus the fraction of fixed carbon used for respiration and maintenance by marine biota. Ocean NPP relies upon the availability of light and nutrients, for example, nitrogen, phosphorus, iron (Fe), and silica (Behrenfeld et al., 2007). A vast area of oceanic surface waters is depleted in Fe but not in other nutrients, for example, the Southern Ocean (SO), the eastern equatorial Pacific, and the subarctic Pacific (Boyd et al., 2005). In those regions, so-called high nutrient low chlorophyll (HNLC) regions, Fe is the limiting factor for phytoplankton productivity, and thus for NPP . Therefore the availability of Fe may significantly affect the ocean biological carbon export on a global scale.
The dominant Fe input to the open ocean is atmospheric aerosol deposition (Duce & Tindale, 1991). The major contributor to atmospheric Fe is mineral dust emitted from arid and semiarid regions, with an estimated present-day contribution of 95% of the total Fe aerosol burden. The remaining 5% is attributed to combustion sources, particularly anthropogenic combustion and biomass burning aerosols (Luo et al., 2008). However, not all Fe deposited to the ocean is directly bioavailable for marine biota. Much uncertainty exists about the physicochemical Fe forms that can be processed as nutrients (Baker & Croot, 2010;Jickells et al., 2005;Lis et al., 2015). It is widely assumed that soluble Fe (SFe) forms (e.g., aqueous, colloidal, or nanoparticulate) can be considered as bioavailable (Baker et al., 2006). Fe in dust (FeD) is considered to be largely insoluble at emission (≈0.1%) (Jickells & Spokes, 2001;K. S. Johnson, 2001;Mahowald et al., 2005;Schroth et al., 2009;Sholkovitz et al., 2012). On the other hand, the solubility of combustion-Fe (FeC), both from fossil fuels (FeF) and biomass burning (FeB) emissions, is known to be higher and is estimated to range from 8% to 81% depending on the fuel type or activity sector (Chuang et al., 2005;Rathod et al., 2020;Schroth et al., 2009).
Observational and experimental evidence point towards an increase in Fe solubility downwind of the sources (Rizzolo et al., 2017;Rodríguez et al., 2021;Zhuang et al., 1992). Acidic processing has been identified as a crucial Fe solubilization mechanism in the atmosphere, occurring at low pH in aerosol water (Desboeufs et al., 1999;Shi et al., 2015;Spokes et al., 1994;Zhuang et al., 1992). In addition, oxalate (hereafter OXL), can act as an organic ligand promoting Fe solubilization by effectively breaking the Fe-O bonds at the surface of the aerosol via the formation of ligand-containing surface structures (Li et al., 2018;Yoon et al., 2004). Photoreductive processes are also considered to be a non-negligible pathway to Fe dissolution, although their contribution might be lower than for acidic and OXL-promoted dissolution (Key et al., 2008). Aerosol acidity, atmospheric OXL, and therefore SFe estimates are governed by multiphase chemical processes. Such processes significantly impact the atmospheric cycles of inorganic species like sulfur (Hoyle et al., 2016;Steinfeld, 1998;Tsai et al., 2010) and hence sulfate (SO 2− 4 ), which is known to be the main control on aerosol liquid water content and aerosol acidity. Multiphase chemistry also acts as a complementary pathway for the formation of organic species related to Fe dissolution, such as OXL.
SO 2− 4 and OXL are, in fact, the most common species formed via aqueous-phase reactions of inorganic and organic origin, respectively (e.g., Carlton et al., 2007;Lim et al., 2010). SO 2− 4 is mainly produced via oxidation of dissolved sulfur dioxide (SO 2 ) (Steinfeld, 1998) and OXL is primarily formed through cloud processing of glyoxal and other water-soluble products of alkenes and aromatics of anthropogenic, biogenic, and marine origin (Carlton et al., 2007;Warneck, 2003).
Models along with observations can be used to better understand and constrain the atmospheric Fe supply to the oceans, but SFe concentrations and deposition measurements are scarce and heterogeneous (e.g., sediment traps, marine sediment cores, or direct deposition measurements from scientific cruises) (Schulz et al., 2012). Modeling is therefore key to analyzing geographical regions not covered by observations and making assessments on global scales to understand the different sources and processes affecting SFe deposition, as well as to assess their impacts on the ocean and the climate. However, estimating the atmospheric supply of SFe to the global ocean with models is challenging due to the variety and complexity of Fe forms in aerosols and the processes that alter their solubility. Over the last decade, models have seen advances in the representation of Fe emission sources and subsequent atmospheric processing. Early works neglected Fe sources such as combustion aerosols (Hand et al., 2004;M. S. Johnson & Meskhidze, 2013;Moxim et al., 2011), which have been later identified as relevant contributors to the atmospheric SFe (Guieu et al., 2005;Ito et al., 2021;Luo et al., 2008). Atmospheric dissolution has been represented with different levels of complexity in models. Simple approaches exist following first-order rate processing constants and considering a globally uniform 3.5% of Fe content in dust (e.g., Duce & Tindale, 1991;Hand et al., 2004;Luo et al., 2008). Mid-complexity representations allow for different types of acidic species to interact with dust, and consider mineral-specific dissolution rates (Ito & Xu, 2014;Meskhidze et al., 2005). Some models further account for OXL processing, even when the full complexity of the OXL BERGAS-MASSÓ ET AL.

10.1029/2022EF003353
3 of 21 formation in cloud water is not explicitly considered, but parameterized M. S. Johnson & Meskhidze, 2013;Scanza et al., 2018). More recently, complex schemes have been developed where both FeD and FeC are dissolved during atmospheric transport, multiphase chemistry is resolved explicitly including the OXL and sulfur cycles, and aerosol acidity is considered in both accumulation and coarse aerosol modes that account for aerosol microphysics (Myriokefalitakis et al., 2015(Myriokefalitakis et al., , 2022. While the present-day Fe cycle has been estimated in numerous studies, changes in the atmospheric Fe cycle that have occurred over the industrial period or that are expected to occur over the 21st century have been less explored, and only with simple or intermediate-complexity schemes. Modeling studies focusing on the present day estimate a global atmospheric dissolved Fe deposition flux into the ocean in the range 0.1-0.8 Tg-Fe yr −1 Ito & Shi, 2016;Ito et al., 2019;Luo et al., 2008;Myriokefalitakis et al., 2018;Scanza et al., 2018) (Table S1 in Supporting Information S1). Some studies estimate that pre-industrial Fe would have been at least 2 times lower, mainly due to lower emission of SO 2− 4 and nitrate precursors leading to a decline in proton-promoted solubilization (e.g., Myriokefalitakis et al., 2015); other studies, however, point towards higher values  due to a possible underestimation of wildfires in the pre-industrial era in commonly used emission datasets such as the Coupled Model Intercomparison Project Phase 6 (CMIP6) inventory . Despite the large uncertainties, it has been accepted that the rise in anthropogenic combustion emissions since pre-industrial times has increased the Fe atmospheric burden along with atmospheric acidity due to a drastic increase in SO 2 emissions (Hand et al., 2012;S. J. Smith et al., 2004). Future projections are even more uncertain as anthropogenic and fire emissions depend on hypothetical future human activities and the impact of land-use change and climate change on dust emission sources and fires is complex to estimate (Harris et al., 2016;Mahowald et al., 2009).
In this work, we aim to estimate the pre-industrial, present and future atmospheric delivery of SFe to the ocean by using the state-of-the-art Earth System Model (ESM) EC-Earth3-Iron (Myriokefalitakis et al., 2022). The atmospheric Fe cycle component in EC-Earth3-Iron contains numerous advances including a detailed atmospheric Fe solubilization mechanism that accounts for complex multiphase chemistry driving aerosol acidity and explicit representation of OXL. With our novel model capabilities, we estimate the SFe deposition into the ocean, and assess aerosol acidity, OXL and their effects upon Fe solubilization, while quantifying the contribution of natural and anthropogenic sources for pre-industrial and present-day conditions, and a range of future scenarios, following the Shared Socioeconomic Pathways (SSPs) of the CMIP6 . While our main focus is SFe and its drivers, our assessment of aerosol acidity and OXL in future scenarios has been understudied in previous literature, and their implications go well beyond the Fe cycle.
The manuscript is organized as follows: We first describe the model and the experimental setup (Section 2). We then present and discuss the Fe emissions in each simula and the corresponding simulated global aerosol acidity, OXL surface concentrations and the resulting SFe atmospheric deposition budgets and their distributions and source contribution for pre-industrial, present and future scenarios (Section 3). Finally, we summarize the relevant findings and discuss their implications, along with the plans for future research (Section 4).

Model Description
EC-Earth3 is an ESM comprised of modules, each representing a different Earth System component, that is, the atmosphere, ocean, sea ice, land surface, dynamic vegetation, atmospheric composition, and ocean biogeochemistry, which can be coupled in various model configurations according to different scientific needs (see Döscher et al., 2021, for details). In this study we apply the recently developed EC-Earth3-Iron model configuration (Myriokefalitakis et al., 2022), which is an extended version of the CMIP6 EC-Earth3-AerChem configuration (Van Noije et al., 2021). We perform all simulations in atmosphere-only mode, in which the atmospheric general circulation model, the Integrated Forecasting System (IFS) from the European Center for Medium-Range Weather Forecasts (ECMWF), is coupled to the atmospheric chemistry module, the Tracer Model version 5 release 3.0 (TM5-MP 3.0). With this setup, sea-surface temperatures (SSTs) and sea-ice concentrations (SICs) are prescribed as in the Atmospheric Model Intercomparison Project (AMIP) experiment (Döscher et al., 2021;Gates et al., 1999). For our analyses, that focus on the atmospheric cycle of Fe, the atmosphere-only mode is thought to constitute a valid and computationally efficient approach. 4 of 21 TM5-MP 3.0 represents interactive aerosols and tropospheric chemistry. In our setup, the gas-phase chemistry scheme is resolved by the MOGUNTIA chemical mechanism (Myriokefalitakis, Daskalakis, et al., 2020). SO 2− 4 , black carbon (BC), organic aerosols (OA), sea salt, and mineral dust microphysics are described by the modal aerosol scheme M7 (Vignati et al., 2004). M7 defines seven log-normal modes to represent the aerosols' size distribution and mixing state; four water-soluble modes (nucleation, Aitken, accumulation, and coarse) and three insoluble modes (Aitken, accumulation, and coarse). Natural emissions of mineral dust, sea salt, marine dimethyl sulfide (DMS), and nitrogen oxides from lighting are calculated online, while other natural emissions are prescribed (e.g., biogenic emissions of non-methane volatile organic compounds). Mineral dust emission is parameterized according to Tegen et al. (2002).
EC-Earth3-Iron includes a representation of the atmospheric Fe cycle, and explicitly calculates the dissolution of Fe in aerosol water and in cloud droplets, aqueous-phase OXL formation, and cloud and aerosol acidity, the latter in both accumulation and coarse modes. Its ability to represent tropospheric aerosols has been shown elsewhere (Myriokefalitakis et al., 2022), and remains similar to the standard EC-Earth3-AerChem version (Gliß et al., 2021), regardless of the substantial differences in the gas-phase and aqueous chemistry. For details and extended comparisons with observations of EC-Earth3-Iron we refer to Myriokefalitakis et al. (2022). Below we provide a summary of the main features related to the calculation of SFe.

Atmospheric Fe Cycle, Aerosol Acidity and Oxalate in EC-Earth3-Iron
EC-Earth3-Iron calculates the dissolution of Fe in aerosol water and in cloud droplets. Three different FeD pools are considered (Shi et al., 2011) according to their susceptibility to dissolve: (a) A fast dissolution pool that relates to ferrihydrite (i.e., hydrated ferric Fe oxide) on the surface of minerals; (b) an intermediate dissolution pool that considers nano-sized Fe oxides from the surface of dust minerals; and (c) a slow dissolution pool that takes into account the Fe release from heterogeneous inclusion of nano-Fe grains in the internal mixture of various dust minerals, such as aluminosilicates, hematite, and goethite. The model also takes into account an extra pool of Fe for the solubilization of aerosols produced by combustion (Ito, 2015;Myriokefalitakis et al., 2022).
This version of the model explicitly traces the three Fe pools and calcium (Ca) originated from mineral dust sources. The emitted dust-Fe (FeD) in the accumulation and coarse insoluble modes of each pool are based on the soil mineralogy of Claquin et al. (1999), including the updates proposed in Nickovic et al. (2012). The Fe content of each mineral is based on Nickovic et al. (2013). Brittle Fragmentation Theory (Kok, 2011) is used to have a better estimation of particle size distribution of each mineral at emission (Pérez García-Pando et al., 2016;Perlwitz et al., 2015aPerlwitz et al., , 2015b. Following Ito and Shi (2016), we assume an initial solubility (i.e., the fraction of soluble FeD, SFeD, over total FeD) of 0.1% for all Fe mineral soil emissions.
FeC emissions, including FeF and FeB emissions, are derived following Ito et al. (2018) and Hajima et al. (2019). FeC emissions are computed by applying specific emission factors to the total carbonaceous particulate emissions (i.e., the sum of organic carbon and BC emissions) for each aerosol mode considered and activity sector (i.e., energy, industrial, iron and steel industries, residential and commercial, shipping, waste treatment and biomass burning). FeC emissions are assumed here to be insoluble, except for ship oil combustion (≈80% solubility) (Ito, 2013). Prior studies have indicated some uncertainty regarding the solubility of Fe combustion aerosols at emission (Ito, 2012;Rathod et al., 2020); nonetheless, in our current investigation, we have adopted low solubility values at emission, as our model incorporates a rapid-dissolution pool for FeC that considers swift dissolution within the same emission gridcell. The FeF emission factors change year-to-year during the historical period , but are assumed to be constant in the future (set to the latest available value of the historical period). For FeB, the emission factors are kept constant following Ito et al. (2018).
The dissolution of Fe from mineral dust and Fe from combustion processes in each of the pools depends on the acidity levels of the solution (i.e., proton-promoted Fe dissolution), the OXL concentration (i.e., ligand-promoted Fe dissolution), and irradiation (photo-reductive Fe dissolution), following Ito (2015) and Ito and Shi (2016). Estimates of the degree of acidity of particles affecting proton-promoted Fe solubilization rely on the use of the thermodynamic model ISORROPIA II (Fountoukis & Nenes, 2007). ISORROPIA II is used to predict not only aerosol acidity (taking into account both basic and acidic species in both the fine and coarse modes) but also equilibrium gas-particle partitioning, liquid-phase activity coefficients, solid-liquid and liquid-liquid equilibria, dynamic mass transfer of semivolatile species and aerosol liquid water content. The modeling approach in ISORROPIA II does not consider single-ion activity coefficients that allow the calculation of pH as proxy 5 of 21 of acidity, but instead the pH values presented in this work are based on the free-H + molality. In-cloud proton concentration and hence cloud acidity are solved via aqueous-phase chemistry (Myriokefalitakis et al., 2022). OXL concentrations have a non-negligible influence on Fe solubilization, hence the formation of OXL is computed online in the model taking into consideration aqueous-phase chemistry (Myriokefalitakis et al., 2022). OXL is rapidly formed via biomass combustion processes in the atmosphere (Kundu et al., 2010). Even if the direct emission of OXL is very low compared to secondary formation, EC-Earth3-Iron includes OXL primary emissions obtained as a fraction of biomass burning and anthropogenic wood burning BC emissions, 0.763% (Yamasoe et al., 2000), and 0.863% (Schmidl et al., 2008), respectively.

Experimental Setup
In order to consistently quantify OXL concentrations, aerosol acidity and SFe deposition for pre-industrial, present and future climates, we perform five ensembles of atmosphere-only time-slice experiments. Each ensemble is composed of 30 different members where the atmospheric initial conditions have been created by applying infinitesimal random perturbations, with the aim of sampling the internal climate variability. The initial conditions for the atmospheric tracers, gas-phase and aerosols, use 2 years of spin-up to ensure realistic concentrations at the global scale, the third year of simulation is used to construct the ensembles presented in this work. Those were run with IFS and TM5-MP coupled, where potential feedbacks between the atmosphere and the ocean are neglected. Also, to maintain consistent conditions between the atmosphere and the ocean state, the aerosols and gas-phase species were not allowed to interact with the atmospheric state. The IFS horizontal resolution is T255 (i.e., a spacing of roughly 80 km), 91 layers are used in the vertical direction up to 0.01 hPa, and a time step of 45 min is applied. On the other hand, TM5-MP has an horizontal resolution of 3° in longitude by 2° in latitude and uses 34 layers to represent the vertical direction up to 0.1 hPa (≈60 km).
We simulated time-slices that represent pre-industrial (PI) and present-day (PD) conditions along with three different future scenarios based on Tier-1 CMIP6 Shared Socio-economic Pathways (SSPs) (O'Neill et al., 2016) (SSP1-2.6, SSP2-4.5, and SSP3-7.0) that respectively represent forcing levels of 2.6, 4.5 and 7.0 W/m 2 by the end of this century. The PD simulation considers climatological conditions based on CMIP6 historical for the 1985-2014 period and serves as a baseline for the assessment of past and future changes. To reproduce the PI climatological conditions, CMIP6 historical information for the 1850-1879 period is used. For the future scenarios, the climatological period considered is 2070-2099 (Table 1).
The future scenarios depict different pathways for socio-economic development and their potential impact on climate change. SSP1-2.6 (hereafter SSP126) represents a sustainable and optimistic pathway with expanded pollution controls, low emissions of Near-Term Climate Forcers (NTCF), and a reduced demand for energy-and resource-intensive agricultural commodities. SSP2-4.5 (hereafter SSP245) is a business-as-usual scenario with moderate population growth and continued growth of GHGs emissions, combined with some efforts to reduce pollutant emissions. SSP3-7.0 (hereafter SSP370) is a scenario with high NTCF emissions, particularly for sulfur dioxide (SO 2 ), and decreased global forest cover, along with increased population in low-and middle-income countries and ineffective policies to control air pollution and GHGs emissions.
Climatological monthly emission fields for the selected present-day and future 30-yr periods for both anthropogenic and natural species (not computed online) are generated specifically for each of the experiments from Earth System Grid Federation (ESGF) archives (https://esgf-node.llnl.gov/search/cmip6). For the pre-industrial period, fixed 1850 emissions are used. The historical anthropogenic emissions used for our PI and PD simulations are taken from the Community Emissions Data System (Hoesly et al., 2018) and the historical fire emissions from the BB4CMIP6 data set (van Marle et al., 2017). Future emission data for each scenario are detailed in Gidden et al. (2019).
The experiments use prescribed SSTs and SICs climatologies created from a selection of historical and scenario simulations performed with the coupled atmosphere-ocean version of EC-Earth3, all contributing to the CMIP6  (Koffi et al., 2016) and can be seen in Figure S1a.
6 of 21 exercise. For each of the selected periods (1850-1879 for the pre-industrial, 1985-2014 for the present, and 2070-2099 for the future scenarios), seven realizations available in the ESGF repository were used to account for potential differences due to the sampling of internal climate variability. The corresponding climatological ocean and sea ice boundary conditions for each experiment were produced by first averaging across the seven realizations and then, averaging in time each 30-year period, ending up with a climatological average for each calendar month.
Global estimates of dust emission largely vary across state-of-the-art ESMs (Gliß et al., 2021;Wu et al., 2020) ranging from 735 to 8,186 Tg yr −1 . Differences arise due to diverse representation of dust emission in models with different dependence on environmental and ambient factors (i.e., wind soil humidity, vegetation) or differences in the size distribution representation. Global and regional dust (and FeD) projections are even more uncertain (e.g., Ginoux et al., 2012;Kok et al., 2014;Mahowald, 2007) due to lack of confidence in future regional winds, precipitation, vegetation, and anthropogenic land-use change (Ginoux et al., 2012). In addition, Mahowald et al. (2002) and Yoshioka et al. (2007) show that the increase in dust surface concentrations observed between the 1960s and 1980s in Barbados is not captured by models and suggest that a change in dust source areas, to account for the creation of new deserts or human land use change, is required to match observations. In fact, most ESMs neglect the potential year-to-year evolution of relevant drivers of dust emission, for example, changes in dust source area extent due to changes in vegetation and/or land use (Kok et al., 2023). In particular, the EC-Earth3-Iron dust emission scheme makes use of yearly vegetation fields produced offline by the dynamic terrestrial vegetation module of EC-Earth, LPJ-Guess (B. Smith et al., 2001). LPJ-Guess fields are available for both the historical CMIP6 period and for the different future scenarios. However the dust emission scheme is mostly sensitive to surface roughness and cultivated fraction fields that consider exclusively intra-annual changes (Tegen et al., 2002). Therefore, dust projections with EC-Earth3-Iron depend primarily upon changes in simulated wind and soil humidity. To account for these uncertainties and eventual changes in source area extent in the future, we perform an additional set of sensitivity experiments where dust emissions from different regions are perturbed.
Some studies suggest an increase of dust loading in response to future warming (Kok et al., 2018), while others project a decrease in dust emissions in key semi-arid regions due to an increase in rainfall and vegetation (Pausata et al., 2020). Here, we explore the sensitivity of future SFe deposition to potential increases in dust emission. We define five simulations based on the SSP370 scenario for the period 2070-2099. In each simulation we double the dust emission flux from a different source region: North Africa (NAfr), Middle East (MEast), East Asia (EAsia), Southern Hemisphere (SH) and North America (NAm). The regions selected are based on Koffi et al. (2016) and are shown in Figure S1a. Selecting the SSP370 scenario allows us to assess the impact of the increased dust emissions in an environment rich in iron dissolution precursors. SSP370 is characterized by large emissions of reactive gases and aerosols, and the aerosols' acidity and OXL concentrations are expected to be the highest within our range of scenarios. Under these conditions, the impact of doubling FeD emissions on the soluble iron deposition will be maximized and less limited by the availability of dissolution precursors than in our other scenarios. As such, the responses to the perturbations in dust sources will lie on the upper end of the potential variability and their signal will be easier to identify and characterize.
Regarding data analysis, multiple analytical approaches have been employed in this study. Yearly budget calculations are performed for variables such as Fe emission, solubilization, and deposition or OXL surface concentrations. The spread of the ensemble of those budgets (e.g., the difference between members of a simulation) is shown through the standard deviation (σ) with respect to the ensemble mean. Differences in fields of extensive variables (e.g., SFe deposition) are shown as relative differences in %, taking the PD simulation as reference. The statistical t-test with a 95% confidence interval is carried out over those relative difference fields, and only significant relative changes are presented. On the other hand, aerosol pH values presented in this work are a diagnostic product and are not the same as the ones used by the model itself (see Text S1 in Supporting Information S1).
FeD emissions contribute the most to the total Fe emission burden in all simulations. Changes in FeD emissions between simulations follow the changes in dust emissions. Mean annual dust emission budgets range between 860 and 970 Tg dust/yr for our pre-industrial, present, and future estimates with some variability among runs of the same ensemble (≈±100 Tg Fe yr −1 ) ( Figure S2). These estimates fall in the lower limit of the simulated emission budget by other ESMs. This is partly due to current assumptions in the dust size distribution at emission and the lack of a super-coarse mode (Adebiyi & Kok, 2020;Gliß et al., 2021;Wu et al., 2020). In our simulations, wind strength plays a dominant role in controlling the variability of dust emissions. Specifically, EC-Earth3-Iron estimates stronger winds over the SH in the PD simulation than in any future scenario, and consequently higher FeD emissions ( Figure S3). This decrease is not noticeable at the global scale, where Northern Hemisphere (NH) sources dominate. Particularly, the North African dust emissions increase under low to moderate radiative forcing levels by the end of the century (SSPs 126 and 245), while they remain similar to present-day estimates under the highest forcing level (SSP370), consistent with the changes in surface winds ( Figure S3). Deposition records reveal a 55% ± 30% increase in dust load for PD compared to PI, according to Kok et al. (2023), with Asian dust sources contributing significantly to this rise. Despite this, our simulation and other climate models with prognostic dust cycles in the CMIP6 ensemble do not show this increase, indicating that dust emissions in models may not be sensitive enough to climate changes.

of 21
FeB emissions for the PI simulation do not contribute much to the total Fe emission (0.46 Tg Fe/yr). Our estimate is tightly linked to the prescribed fire emissions from the CMIP6 data set (van Marle et al., 2017), which are likely underestimated (e.g., Hamilton et al., 2018). Recent works have shown that, in contrast to what was previously assumed, increased population during the historical period lead to a reduction in burned area (Andela et al., 2017). This stems from models not accounting for the effect of human-induced land use changes, fragmentation of ecosystems, as well as the implementation of fire management practices in some regions (Andela et al., 2017;Knorr et al., 2014), and hence not being able to capture the variations in fire activity from the observationally constrained present-day information toward the pre-industrial period. Model estimates that account for the human-driven decline in fire emissions during the historical period, place PI FeB emissions from 3 to 5 times higher than our estimate, that is, between 1.5 and 2.7 Tg Fe/yr . Projected FeB emissions decrease in the three future scenarios considered with 0.33, 0.38, and 0.47 Tg Fe/yr for SSP126, SSP245, and SSP370, respectively. This decrease is observed for all regions and scenarios, except for East Asia for the SSP370 scenario, which is consistent with the trends defined in the CMIP6 emission inventory. Another weakness of the CMIP6 future fire emission estimates is that they do not incorporate the potential impact of vegetation and climate changes in fire regimes. Neglecting these factors may result in underestimated projections of FeB emissions. For instance, our estimates using CMIP6 data are approximately six times lower than those in Hamilton et al. (2020), who used the CMIP5-RCP4.5 intermediate scenario and fire emissions that accounted for the impacts of climate and vegetation changes (Ward et al., 2012). All in all, CMIP6 and other estimates are highly uncertain, as the human-vegetation-fire-climate feedbacks are still not well understood. Multiple recent studies have tried to better constrain future fire emissions, but with diverging results (Kasoar et al., 2022;.

OXL Concentrations and Aerosol Acidity
PD OXL primary emissions are estimated to be 0.36 Tg OXL/yr. The trend in OXL emissions for past and future projections with respect to the PD follows FeB emissions ( Figure S2). PD OXL net chemical production is 9.2 Tg OXL/yr, which is in the lower range of what has been reported in previous studies, that is, 9.0-27.3 Tg/yr (Lin et al., 2014;Liu et al., 2012;Myriokefalitakis et al., 2022). The PI simulation shows a drop in the annual global mean OXL net chemical production compared to PD estimates. However, an increase is seen for SSP245 and SSP370 where OXL net chemical production is especially boosted over areas where anthropogenic activities are expected to increase (e.g., East Asia, South America, and South Africa) (Figure 2). The primary mechanism of global OXL production is glyoxal oxidation (≈74%), followed by glycolaldehyde (≈11%), methylglyoxal (≈8%), and acetic acid (≈7%) (Myriokefalitakis et al., 2022). The primary sources of glyoxal is the secondary production through the oxidation of different volatile organic compounds of biogenic origin (mainly isoprene), followed by direct emissions from biofuel consumption and biomass burning. The observed differences in OXL net chemical production between the PI, PD, and future scenarios are consistent with changes in anthropogenic emissions between scenarios and regions.
Focusing in particular on SO 2− 4 , which plays a key role in determining atmospheric acidity, it primarily forms in the atmosphere from gas-phase precursors, specifically SO 2 , despite direct emission from certain sources. SO 2 emissions are 128 Tg SO 2 /yr in the PD simulation, which represents a 9-fold increase over the PI period. SO 2 emissions dominated by the energy and industrial sector are expected to decrease overall as the energy sector gets decarbonized, with a more abrupt general decrease for SSP126 (i.e., the optimistic scenario, see Figure S2) (Gidden et al., 2019). This is true with the exception of some regions under the more pessimistic scenarios (e.g., the Middle East and Central Asia, North Africa and the SH for SSP370) where an increase in the industrial demand is hypothesized ( Figure S2). SO 2− 4 net chemical production for PD is 149.9 Tg SO 2− 4 ∕ and follows the trends seen in SO 2 emissions; a 2.8-fold increase in PD estimates is seen compared with the PI while a decrease is projected for all three future scenarios, especially in the NH, where projected air quality mitigation strategies are expected to effectively reduce SO 2 emissions in most NH countries (Figure 2).
Aerosol pH and OXL concentrations, present notable differences among simulations (Figure 3). Mean aerosol pH values are the lowest (i.e., more acidic) during the PD period, both in the accumulation and coarse modes, 1.69 and 3.61, respectively. The PI simulation, with a more pristine atmosphere, is the one presenting higher pH values (i.e., less acidic), 2.16 and 4.34 as global means for accumulation and coarse modes. Projections follow the SO 2 emission and SO 2− 4 net chemical production trends discussed above, with the scenario representing a more sustainable pathway (SSP126) showing acidity values closer to PI estimates, 2.18 and 4.19, and the scenario with

of 21
ammonia in the fine mode, cations present in the coarse dust, such as Ca++, neutralize nitric acid in the coarse mode. Species found in sea salt and dust are then buffering the acidity of coarse aerosols, which are, for this reason, less acidic than fine ones ( Figure S4). In the coarse mode, slightly acidic pH values are simulated over ocean due to sea salt species, whereas over continental regions the pH is lower (i.e., close to 1), particularly where anthropogenic activity is intense. Regionally, our SSP370 scenario projects an increase in acidity over developing regions (e.g., Africa, South America and the Middle East and Asia), where population and energy and industry requirements are expected to grow in the near future.
Present-day estimates of OXL ( Figure 3l) show maximum surface concentrations over major biomass burning sources such as Central Africa, South America and Indonesia, as well as OXL net chemical production ( Figure 2b), but also downwind of those sources (up to 0.2 μg/m 3 on annual average) with a steeply decrease toward the poles. Overall, lower concentrations are found in the PI, with relative decreases between 10% and 30%, with except in certain equatorial regions of the Indian and Pacific Ocean, and Europe (Figure 3k), where up to 100% higher concentrations are found. In the future projections, OXL sharply increases in the Indian Ocean and equatorial Pacific. Also, OXL is expected to increase over and downwind of South Africa, southern South America and Australia. Under SSP370, the increases extend over other areas such as the North American continent, North Africa and the Asian and European continent. As OXL has mostly a secondary origin, changes in OXL surface concentrations between simulations are fundamentally driven by changes in OXL net chemical production (Figures 2a-2e), which ultimately depends on the abundance of organic precursors.

Atmospheric Fe Solubilization
Fe at emission is considered to be mostly insoluble; in EC-Earth3-Iron only 0.1% of the emitted FeD and 80% of the emitted Fe from shipping emissions are assumed to be soluble. In this section, FeF and FeB are discussed together as Fe from those sources are considered to have the same dissolution rates and treated in the EC-Earth3-Iron solubilization scheme as one pool (FeC) (Section 2. All in all, around 95% of atmospheric SFe results from atmospheric dissolution processes. FeD is primarily dissolved by acid dissolution, with a rate of 0.296 ± 0.007 Tg FeD/yr. Ligand-promoted dissolution additionally produces 0.138 ± 0.010 Tg FeD/yr and photo-induced processes have a small impact on the global dissolved Fe release from dust, with 0.039 ± 0.003 Tg FeD/yr (Figure 4). On the other hand, the main dissolution path for FeC is ligand-promoted dissolution with a rate of 0.189 ± 0.004 Tg Fe/yr. Acidic and photo-induced dissolution 12 of 21 each represent 17% of the FeC dissolution, with rates of 0.0478 ± 0.0011 Tg Fe/yr and 0.0476 ± 0.0011 Tg Fe/ yr, respectively (Figure 4). The primary mechanism for solubilizing FeC in our model is ligand-promoted dissolution, which is consistent with the solubilization scheme proposed by the work by Chen and Grassian (2013) on which our model is based. This pathway is further fostered by the combustion activities emitting both FeC and OXL precursors, which results in the maximum values of OXL production (Figure 2b) being spatially correlated with high solubilization values of FeC (Figures S5a-S5f). Nevertheless, our estimations of FeC solubilization represent an upper bound as recent experimental studies indicate a lower enhancement in Fe solubility than that reported by Chen and Grassian (2013) (Baldo et al., 2022).
Dissolution rates of FeC are 65% smaller in the PI scenario than in the PD. FeD dissolution is also smaller in the PI simulation, by 65% the acidic dissolution and by 20% the OXL-promoted and photo-induced dissolution. The drop in Fe dissolution is mainly driven by a reduction in direct FeC emissions (75%) (Figure 1) and a drop in OXL and SO 2− 4 net chemical production (≈13% and ≈64% respectively) (Figure 2). Different trends can be seen on our future scenarios of Fe dissolution compared to the PD. On the one hand, SSP126 shows a decrease in atmospheric Fe processing for both FeD and FeC, especially due to reduced acidic dissolution. Although FeD emissions in this scenario are 12.5% higher than in the PD, there is a relative reduction of nearly 60% in FeD acidic dissolution. The decrease is mainly driven by the drop in SO 2 emissions ( Figure S2), leading to a less acidic atmosphere (Figures 3c and 3h). On the other hand, SSP245 estimates a 12% increase in OXL-promoted and photoinduced Fe dissolution, and a relative decrease of 26% in Fe acidic dissolution. The drop in proton-promoted dissolution is driven by a decrease in aerosol acidity (Figure 3). The increase in OXL over some equatorial regions, where FeF emissions are projected to increase, leads to an enhanced ligand-promoted dissolution ( Figure S2d). In SSP370, the scenario with higher NTCF levels projected, Fe dissolution increases for both FeD and FeC as a result of all processing mechanisms. The increase in FeC solubilization is especially abrupt compared to the PD, with values ranging between 91% and 98% more for the different mechanisms. This results from a 78% increase in FeF primary emissions relative to PD (Figure 1), together with an OXL production and aerosol acidity increase around anthropogenic sources.
No substantial differences are evident when comparing the global solubilization budgets of the SSP370 scenarios with regionally perturbed dust emission with the base SSP370 simulation ( Figure S6). Despite the large increase in global FeD emissions in some experiments, that is, SSP370-NAfr (≈×2), SSP370-EAsia (≈×1.3), and SSP370-MEast (≈×1.2), minor changes are seen in Fe solubilization budgets. This is explained by the increase in calcium carbonate, which buffers acidity and therefore limits FeD solubilization (Myriokefalitakis et al., 2022).

Soluble Iron Deposition and Solubility
The Fe deposition in our PD simulation is 42 ± 5 Tg Fe/yr, with 12.1 ± 1.4 Tg Fe/yr deposited to the ocean. The SFe deposition is 0.721 ± 0.018 Tg SFe/yr, with 0.406 ± 0.011 Tg SFe/yr deposited to the ocean ( Figure 5). 70% of SFe deposited in the ocean comes from dust mineral sources, while the remaining 14% and 16% come from anthropogenic combustion and biomass burning sources, respectively. Since FeD emissions represent ≈95% of total Fe emissions (99.7% of directly SFe emissions), these results reflect the stronger atmospheric processing of FeC compared to FeD in present day conditions. As discussed in Section 3.3, this is due mainly to enhanced OXL-promoted dissolution in combustion aerosols. Our Fe and SFe deposition budgets in the ocean along with contribution of each source to the total deposition are within the range of previous studies ( Figure 5 and Table  S1 in Supporting Information S1) Ito & Shi, 2016;Ito et al., 2019;M. S. Johnson & Meskhidze, 2013;Luo & Gao, 2010;Luo et al., 2008;Myriokefalitakis, Gröger, et al., 2020;Myriokefalitakis et al., 2015Myriokefalitakis et al., , 2018Scanza et al., 2018).
Total Fe deposition budgets do not present significant differences among simulations (Figure 5a), ranging between 42 and 47 Tg Fe/yr, and 11 and 13 Tg Fe/yr over the oceans. This follows from Fe emission shown in Figure 1 being dominated by dust, whose global emission shows small variations between simulations. In contrast, SFe deposition budgets do show substantial variations among simulations with the lowest values reached in the PI, with 0.377 ± 0.015 Tg SFe/yr globally and 0.209 ± 0.009 Tg SFe/yr over ocean. Our PI SFe deposition compares well with prior studies (Figure 5b), except for Hamilton et al. (2020) in which SFe deposition over the ocean more than doubles our estimates, most likely due to the use of a different fire emission data set . A decline in SFe deposition is observed for the SSP126 and SSP245 scenarios compared to PD, 13 of 21 with 0.474 ± 0.013 Tg SFe/yr (0.265 ± 0.009 Tg SFe/yr) globally (over ocean) for SSP126, and 0.646 ± 0.019 Tg SFe/yr (0.360 ± 0.013 Tg SFe/yr) for SSP245. The projected reductions under these two scenarios are consistent with the drop in atmospheric Fe solubilization due to a decrease in SO 2 emissions and hence a decrease in aerosol acidity (Section 3.3). SSP370 scenario shows a clear increase in SFe deposition in comparison to all other scenarios, with 1.01 ± 0.03 Tg SFe/yr (0.56 ± 0.02 Tg SFe/yr) globally (over ocean). SFe deposition increases with increasing NTCF, and is almost doubled for SSP370 with respect to SSP126, while the SSP245 deposition falls in the middle. Although there are not many studies dealing with Fe deposition in the future, we can see that the estimates of our different scenarios also fall in the range of previous literature (Table S1 in Supporting Information S1).
PD estimates show maximum values of SFe deposition near the equatorial Atlantic downwind of dust mineral and biomass burning sources, and the north coast of the Indian Ocean where Fe comes from dust mineral sources and anthropogenic combustion (Figure 6b). In HNLC regions, such as the SO, the SFe deposition is lower than over the rest of the globe. The maximum solubility of Fe (i.e., the fraction of SFe over total Fe) at deposition (≈20%) occurs downwind of South African biomass burning sources, East Asian anthropogenic combustion sources, and remote equatorial regions of the Pacific, dominated by long-range transport of dust (Figure 6g). The higher solubility of Fe deposited over ocean compared to land is attributed to the longer lifetime of Fe-aerosols reaching the ocean, being thus more exposed to atmospheric processing. The PI simulation shows globally lower SFe deposition than the PD, except for some areas like South Africa (Figure 6a) where higher biomass burning emissions affect the solubility levels. Future scenarios show a decrease in SFe deposition in mid-and high-latitudes, but an increase in equatorial regions such as the equatorial Pacific, Atlantic, and Indian Ocean. Those increases are sharper and have a broader extension for SSP370, which is the future scenario with higher FeF emissions, aerosol acidity and OXL concentrations (Figure 6e). Solubility increases only in the SO and some regions of the Indian Ocean (e.g., the Bay of Bengal) for SSP126 and SSP245, while for SSP370 solubility increases in all regions but the North Atlantic. The increase in solubility for future scenarios could be driven by a change in Fe source contribution, likely related to a higher contribution of more labile Fe sources such as biomass burning and anthropogenic combustion. Those differences are reflected in the source contribution to SFe deposition for the different scenarios (Figure 7). In line with the emissions, the contribution of anthropogenic sources to SFe deposition is negligible, but the contribution from biomass burning sources is particularly high, especially in the SH. Overall, the NH SFe deposition is dominated by dust in all scenarios, although in some areas, such as the North Pacific or the North Indian Ocean, there is an increase in anthropogenic contribution for future scenarios, especially for SSP370. Moreover, for SSP370 the dust contribution is below 50% for the Indian and SO, which does not happen in any other scenario and basin.  (Table S1 in Supporting Information S1).  (Figures 8a-8d). However, those increases are more relevant in some regions especially downwind the perturbed sources. Perturbing North African dust sources causes a broader impact than changes in any other source, and leads to relative increases of up to 25% in SFe deposition with respect to the baseline SSP370 scenario in remote regions of the SH. In particular, our model shows that SFe deposition in HNLC regions, such as the SO and the equatorial Pacific, is sensitive not only to changes in Australian or South American dust sources, but also to changes in North African sources. In all examined perturbation scenarios, a reduction in solubility was observed in proximity to the disturbed sources. Such an outcome is expected when dust emissions rise due to two primary factors. First, Fe from dust sources is known to have relatively slower solubilization rates than other sources, implying that when its contribution to Fe deposition increases, Fe solubility at deposition decreases. Second, dust particles include calcium carbonate, which buffers acidity, thereby limiting the solubilization process.

Conclusions
Changes in climate and emissions can substantially modify atmospheric aerosol acidity, OXL production, and the strength and distribution of SFe deposition. Estimating these changes is crucial to assess future marine NPP and carbon and nitrogen cycles. Here, we have characterized the past, present and potential future SFe deposition with an ESM (EC-Earth3-Iron) that is equipped with a detailed representation of atmospheric Fe dissolution (Myriokefalitakis et al., 2022). In this way, the SFe deposition in EC-Earth3-Iron is expected to respond more realistically to changes in climate and emissions. Our experimental setup covers the PI period, the PD, and a range of possible future climates following different CMIP6 emission scenarios. These scenarios cover from substantially reduced anthropogenic emissions, associated with very ambitious mitigation strategies, to large increases in emissions related to a growing population, especially in low-and middle-income countries, and a resurgence of coal dependence. Our new model capabilities allow us to predict not only iron from different sources (FeD, FeF, and FeB) but also the precursors and processes controlling iron dissolution explicitly and interactively under changing climate conditions and emission levels.
Based on our calculations, the SFe deposition to the ocean has doubled since the early pre-industrial (PI) era, with values of 0.406 ± 0.011 Tg SFe/yr and 0.209 ± 0.009 Tg SFe/yr for the present day and PI period, respectively.  This trend is consistent with earlier studies Ito & Shi, 2016;Myriokefalitakis et al., 2015). The increase in SFe deposition since the PI can be attributed to a six-fold rise in fossil fuel emissions, which has led to an increase in FeF emissions and a rise in acidic solubilization in the atmosphere due to elevated SO 2 emissions. We project an increase in global SFe deposition of 40% by the late 21st century relative to PD under the low mitigation scenario SSP370, and a decrease of 35% and 11% under the optimistic SSP126 and business-as-usual SSP245 scenarios, respectively. In all simulations, total Fe emissions are dominated by dust sources with a contribution above 90%. Dust sources dominate as well SFe global deposition, but the contribution of FeC aerosols to SFe (≈30% during the PD) is enhanced relative to emissions as atmospheric processing is especially efficient for FeC aerosols. We find that in all simulations, ligand-promoted dissolution is the main FeC solubilization pathway and proton-promoted dissolution is the main one for FeD.
As ligand-promoted and proton-promoted dissolution are driven by aerosol acidity and OXL levels respectively, changes of those variables are in this work explored. Specifically, during the PI, low acidity levels (high pH) and OXL concentrations were observed, with a 15% reduction in OXL surface concentrations relative to the PD. While for the future, under the SSP370 scenario, OXL concentrations are projected to increase, with a 15% rise in OXL surface concentrations compared to PD and acidity levels are comparable to PD ones.
Dust sources dominate SFe deposition in the NH and globally. However, biomass burning and anthropogenic combustion emissions have a more crucial role in the SH. Anthropogenic combustion emissions do not contribute during the PI, but show significant contributions in the PD and future scenarios over the East Asian coast, Central American coast, and part of the South American region. SSP370 is the future scenario with the highest contribution of anthropogenic combustion sources to SFe deposition (especially in the Indian ocean). Biomass burning emissions dominate along the Southern African coast for all scenarios, especially in the PI. This shows that although dust sources are dominant, combustion sources are not at all negligible in regions such as the SO, known to be HNLC regions.
Our results suggest potentially large differences in the ocean response among future scenarios. However, there are patterns shared among all future scenarios that could have important implications for the ability of the ocean to capture carbon in the future. In all future scenarios, we obtain a decrease in SFe deposition over the Fe-limited SO, and an increase over the equatorial Pacific, also known as a HNLC region. The net effect of these changes in the global carbon capture is uncertain. Further analyses are planned where we will use our SFe deposition fields in a biogeochemistry model to understand the regional and global ocean NPP changes associated with future scenarios.
Past and future projected emissions are very uncertain and need further investigation. Recent studies suggest that CMIP6 probably underestimates PI fire emissions , and others also suggest large uncertainties in future fire emission estimates (Kasoar et al., 2022;. Uncertainties in fire emissions also affect the burden of precursors of OXL and therefore Fe dissolution. Potential changes in the spatial extent of dust sources due to changes in vegetation (Mahowald, 2007), land use (Ginoux et al., 2012), and biocrusts (Rodriguez-Caballero et al., 2022) are either poorly considered or not considered at all in ESMs. As seen in our perturbed dust experiments, HNLC regions such as the SO and equatorial Pacific could be very sensitive to those changes in dust emissions. Dust emissions associated with wildfires, where strong, turbulent fire-related winds most likely raise dust Wagner et al., 2018), are largely disregarded in current models. Additionally, by destroying vegetation, wildfires leave a bare source that often becomes a source of dust emission (Yu & Ginoux, 2022). Variations in these climate-sensitive, yet unaccounted, emissions could alter SFe deposition. Future observational and modeling studies should focus on better characterizing the evolution of fire and dust emissions and its interaction with other Earth System components to ultimately better represent the Fe cycle.

Data Availability Statement
The EC-Earth3-Iron code is available from the EC-Earth development portal (https://dev.ec-earth.org/) for members of the EC-Earth consortium. Model codes developed at the European Centre for Medium-Range Weather Forecasts (ECMWF), such as the IFS model code, are intellectual property of ECMWF and its member states. Permission to access the EC-Earth3-Iron source code can be requested from the EC-Earth community via the EC-Earth website (http://www.ec-earth.org/) and may be granted if a corresponding software license agreement is signed with ECMWF. The corresponding repository tag is 3.3.2.1-Fe. Due to license limitations of the model, only European users can be granted access. The model output data used in this study is available on the open repository Zenodo at https://doi.org/10.5281/ zenodo.7901172. 773051), the AXA Research Fund through the AXA Chair on Sand and Dust Storms at BSC, the Spanish Ministerio de Economía y Competitividad through the NUTRIENT project (CGL2017-88911-R), the European Union's Horizon 2020 research and innovation programme under grant agreement no 821205 (FORCeS), and ESA through the DOMOS project (ESA AO/1-10546/20/I-NB). We acknowledge the EMIT project, which is supported by the National Aeronautics and Space Administration Earth Venture Instrument program, under the Earth Science Division of the Science Mission Directorate. RLM received additional support from the NASA Modeling, Analysis and Prediction Program (NNG14HH42I). We also acknowledge the resources obtained on the Marenostrum4 supercomputer at BSC, granted through the PRACE project eFRAGMENT2 and RES project AECT-2020-3-0020, along with the technical support provided by BSC and the Computational Earth Sciences team of the BSC Earth Sciences Department. We also thank Pablo Ortega, Markus Donat and Ettiene Tourigny from the Climate Variability and Change team of the BSC Earth Sciences Department for their recommendations on the experimental setup. We further acknowledge the EC-Earth community and its AerChemMIP team. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies that support CMIP6 and ESGF.