EC-Earth3-Iron includes a characterization of the dust mineralogical composition at emission and explicitly traces the Fe and calcium-containing species. The relative amounts of eight different minerals, namely illite, kaolinite, montmorillonite, calcite, feldspars, quartz, gypsum, and hematite, are derived from the soil mineralogy atlas of Claquin et al. (1999), including the updates proposed in Nickovic et al. (2012). The atlas provides the soil mineralogical composition in arid and semi-arid regions of the world, distinguishing between two soil size classes (i.e., the clay size fraction up to 2 µm and the silt size fraction from 2 to 50 µm diameter). The mineral fractions emitted in the accumulation and coarse insoluble modes of TM5-MP are estimated from the soil mineralogy atlas based on the brittle fragmentation theory (BFT) from Kok (2011). BFT posits that the emitted particle size distribution is independent of wind and soil conditions and additionally allows for estimating the size-resolved mineral fractions (Pérez García-Pando et al., 2016; Perlwitz et al., 2015a, b). The resulting mineral mass fractions are then applied to the dust emission fluxes, as calculated online in the model, yielding the corresponding accumulation- and coarse-mode emission of each mineral. We note that although we derive the mineral dust fractions in each mode using BFT, we maintain the dependence of the ratio between the accumulation- and coarse-mode dust mass at emissions upon wind and soil conditions of the original dust emission scheme (Tegen et al., 2002).

In EC-Earth3-Iron, the different Fe-containing minerals are not prognostic variables (tracers). Instead, we trace the mineral dust Fe according to three dissolution classes, namely fast, intermediate, and slow Fe pools (Ito and Shi, 2016). No relationship of Fe dissolution with other elements is observed, however, for clays and feldspars, where the total Fe content of the minerals is very low (<0.54 %), and the Fe is in the form of impurities (Journet et al., 2008). For this, 0.1 % Fe content in total Fe-containing minerals is here assumed directly soluble as amorphous free iron impurities regardless of mineralogy (Ito and Shi, 2016). The emitted amounts of calcium (i.e., in calcite) and Fe (i.e., in illite, kaolinite, montmorillonite, feldspars, and hematite) are derived either from the average elemental compositions of minerals or based on experimental analyses (Journet et al., 2008; Nickovic et al., 2013). The respective average fractions applied to mineral dust sources of this work are listed in Table S1.

Fe is also emitted in the model from anthropogenic activities (including fossil and biomass fuels) and biomass burning (excluding biofuel combustion) following Ito et al. (2018). The Fe-containing fossil fuel and biofuel combustion emissions are estimated here by applying specific factors (i.e., per emission sector and per particle size) to the total particulate emissions (i.e., the sum of organic carbon, black carbon, and inorganic matter), as derived for this work based on estimates from Ito et al. (2018), for the Fe content in the sub-micrometer and super-micrometer combustion aerosols. The historical anthropogenic emissions are taken here from the Community Emissions Data System (Hoesly et al., 2018) and the historical fire emissions from the BB4CMIP6 dataset (van Marle et al., 2017). We note, however, that the estimate of Fe emission from metal smelting remains highly uncertain and that further work is needed (Rathod et al., 2020). As for the biomass burning, the iron fractions in the fine particles are related to the combustion stages of flaming (0.46 ± 0.51 %) and smoldering (0.06 ± 0.03 %) fires, while the averaged iron fraction is used for coarse particles (3.4 %) (Ito, 2011). The global mean ratio of 0.04 gFe gBC-1 for biomass burning in fine particles is consistent with that of 0.032 in the review paper by Hamilton et al. (2022). Fe-containing aerosol combustion emissions are considered to be insoluble (Ito, 2015), except for ship oil combustion, which is assumed to be mostly soluble, i.e., ∼ 79 % on average for the years 2000–2014. We note that the value of 79 % represents the high solubility of iron emissions in oil fly ash (Ito et al., 2021). Rathod et al. (2020) proposed a lower solubility in emissions (i.e., 47.5 % for iron sulfates), with an upper value, however, at ∼ 90 %. The year-to-year variation in anthropogenic combustion Fe-emission fractions follows Ito et al. (2018). On the contrary, for biomass burning Fe-emission fractions no such variation is provided. The average Fe fractions (per sector) for the years 2000 to 2014 applied to the total particulate carbonaceous emissions are also listed in Table S1.

EC-Earth3-Iron also includes OXL primary emissions from natural and anthropogenic wood-burning processes that mainly account for its rapid formation in the sub-grid plumes not represented in the model. Indeed, OXL is well correlated with elemental carbon and levoglucosan (Cao et al., 2017; Cong et al., 2015), which are observed at significant levels during biomass burning episodes in the Amazon (Kundu et al., 2010), suggesting that oxalic acid could be either directly emitted or formed rapidly via combustion processes. During biomass burning episodes, enhanced emissions of ionic species have been generally measured, indicating an average OXL mass concentration measured in plumes of ∼ 0.04 %–0.07 % w/w​​​​​​​ (Yamasoe et al., 2000). Furthermore, domestic wood combustion is a potential OXL source (Schmidl et al., 2008) since measurements indicate an OXL contribution to the total particulate concentrations of ∼ 0.09 %–0.28 % w/w. Gasoline engines may also contribute to total dicarboxylic acid mass emitted to the atmosphere (Kawamura and Kaplan, 1987), although their direct contribution to ambient OXL concentrations is generally found to be low (Huang and Yu, 2007) and is therefore neglected here. All in all, primary OXL sources are quite uncertain and, given the current estimates, may only have a limited impact on the calculation of its atmospheric concentrations (e.g., Myriokefalitakis et al., 2011).

The gas and particle equilibrium calculations of NH3/NH4+ and HNO3/NO3- have been substantially revised in EC-Earth3-Iron. In EC-Earth3-AerChem, EQSAM (Metzger et al., 2002) is used to determine the partitioning of NH3/NH4+ and HNO3/NO3-. In EC-Earth3-Iron, the ISORROPIA II thermodynamic equilibrium model (Fountoukis and Nenes, 2007) replaces EQSAM to determine the equilibrium between the inorganic gas and the aerosol phases. ISORROPIA-II calculates the gas–liquid–solid equilibrium partitioning of the K+-Ca2+-Mg2+-NH4+-Na+-SO42--NO3--Cl--H2O aerosol system and is used in the forward mode, assuming that all aerosols are in a metastable (liquid) state. The inclusion of sea salt and dust aerosols in the aerosol thermodynamic calculations has been shown to nevertheless substantially affect the ion balance and thus the partitioning of HNO3/NO3-and NH3/NH4+ species, especially in areas with abundant mineral dust and/or sea spray aerosols (Athanasopoulou et al., 2008, 2016; Karydis et al., 2016). In EC-Earth3-Iron nitrate aerosols are calculated for both the accumulation and coarse modes, in contrast to the bulk aerosol approximation used in the EC-Earth3-AerChem. For this, kinetic limitations by mass transfer and transport between the gas and the particulate phases in accumulation and coarse modes (Pringle et al., 2010) are considered, with ISORROPIA-II then re-distributing the respective masses between the gas and the aerosol phases. We note that Ca2+ from calcite is simulated prognostically in the model based on mineralogy maps (Sect. 2.2.1), in contrast to other crustal elements in soils that are calculated by assuming constant mass ratios to dust concentrations of 1.2 %, 1.5 %, and 0.9 % for Na+, K+, and Mg2+, respectively (Karydis et al., 2016; Sposito, 1989). For sea spray aerosols, mean mass fractions of 55.0 % Cl-, 30.6 % Na+, 7.7 % SO42-, 3.7 % Mg2+, 1.2 % Ca2+, and 1.1% K+ (Seinfeld and Pandis, 2006) are also applied.

The acidity levels of deliquescent aerosols are calculated in the model based on thermodynamic processes for accumulation and coarse particles. Aerosol acidity impacts the scavenging efficiency and the dry deposition of inorganic reactive nitrogen species due to changes in the partitioning of total nitrate and ammonium between the gas and aerosol phases and between the various aerosol sizes (Pye et al., 2020). Acidity levels also play a fundamental role in the aqueous-phase chemistry by controlling the dissociation reactions and thus the reactivity of the chemical mechanism. Indeed, aqueous-phase species, such as organic and inorganic acids, are oxidized with higher rates when they are dissociated. Nevertheless, in the case of the forward and reverse reactions, they typically occur fast and thus the concentrations of the reactants and the products are generally assumed to be in equilibrium in the global model due to its relatively long time step and large model grid. Note, however, that recent modeling studies showed that the metastable assumption produces pH values that are different from the stable assumption (e.g., regionally up to 2 pH units in the presence of crustal elements over dust sources, and roughly 0.5 pH units globally; Karydis et al., 2021). However, work to date, such as in Bougiatioti et al. (2016), Guo et al. (2019, 2015) and others identified in the review of Pye et al. (2020), has shown that the metastable solution tends to provide semi-volatile partitioning of pH-sensitive species (e.g., NH3/NH4 and HNO3/NO3) and aerosol liquid water content that is closer to observations – at least for when the relative humidity is above 40 %. For this reason, we assume that the most plausible estimates of acidity are to be obtained with the metastable assumption, and we base our simulations on that.

Under ambient atmospheric conditions, the water vapor uptake on aerosols depends on both the inorganic and organic components, along with the meteorological conditions (e.g., the temperature and the relative humidity conditions). ISORROPIA II does not, however, include water associated with organic aerosols, possibly leading to an underestimation of the aerosol hygroscopicity, especially within the boundary layer where the contribution of water-soluble organics to total aerosol mass can be substantial. For this, we account here for a contribution of aerosol water from organic particles in the acidity calculations,using a hygroscopicity parameter κorg=0.15 (Bougiatioti et al., 2016). In more detail, the particulate water due to the organics (Worg) that is added to the aerosol water associated with the inorganic aerosol as calculated from ISORROPIA-II (Winorg) is determined in the model as follows: 1Worg=ms⋅ρwρs⋅κorg(1RH-1), where ms is the soluble organic mass concentration (µg m-3) as simulated by the TM5-MP chemistry scheme, ρw is the water density (1 kg m-3), ρs is the organic aerosol density (1.4 kg m-3), and RH (0–1) is the relative humidity.

Cloud acidity is also an important factor for simulating the multiphase processes in the atmosphere. The in-cloud proton concentration is initially determined by the electro-neutrality of strong acids and bases (i.e., H2SO4, SO42-, methanesulfonate (MS-), HNO3, NO3-, and NH4+), and then the subsequent dissociations of CO2, SO2, and NH3 (Jeuken et al., 2001) are solved iteratively in the model. For the cloud acidity calculations, the liquid water content, and the respective cloud cover fraction (i.e., 0–1) are obtained from meteorology. Note, however, that the effect of mineral dust (especially calcium) on cloud proton concentrations is neglected here. This assumption may result in some overestimation of cloud acidity, although the overall impact should be small, particularly in dusty areas with a low presence of clouds. Another limitation is the omission of light gaseous organic acids (such as formic and acetic acids) in the cloud pH calculations, possibly leading to some underestimation in cloud acidity where their concentration is important.

The aqueous-phase chemistry scheme used in this work is based to a large extent on the Chemical Aqueous Phase Radical Mechanism (CAPRAM) (e.g., Deguillaume et al., 2004; Ervens et al., 2003; Herrmann et al., 2000, 2015). However, CAPRAM includes more than 70 aqueous-phase species, 34 equilibria for compounds that are present both in the gas and the aqueous phases, along with numerous photolytic and aqueous-phase reactions, also covering a large series of acid–base and metal–complex equilibria. Note that various updates may further extend the mechanism by including, among other processes, the oxidation of aromatic hydrocarbons (Hoffmann et al., 2018), the multiphase oxidation of DMS (Hoffmann et al., 2016), and the tropospheric multiphase halogen chemistry (Bräuer et al., 2013). For this, some reactions are considered here in a more simplified way based on various assumptions published in the literature. Indeed, the level of chemical complexity of such a detailed mechanism is beyond the computational resources available for three-dimensional global climate–chemistry simulations, and thus simplifications that preserve however the essential features of the aqueous mechanism are needed.

Aqueous-phase chemical transformations are considered at the interface and in the bulk, initiated mainly by free radicals and oxidants produced both via photochemical reactions and in dark conditions (Bianco et al., 2020). The sources of OH radicals in the aqueous phase, however, strongly differ from those in the gas phase, primarily because of the presence of ionic species and TMIs in the solution. OH radicals are the main oxidant in the aqueous phase, either produced directly in the aqueous medium or diffused from the gas phase (i.e., via a gas-to-liquid transfer). However, aqueous-phase oxidation can also be induced by non-radical species, such as ozone (O3) and hydrogen peroxide (H2O2). A characteristic example is the formation of SO42- in cloud droplets, via the oxidation of dissolved sulfur dioxide (SO2) by O3 and H2O2, with H2O2 nevertheless being the most effective oxidant (Seinfeld and Pandis, 2006), especially when the solution becomes acidic. Upon the absorption of SO2 in cloud droplets, the establishment of the equilibrium between the dissolved sulfur species in oxidation state four, i.e., SO2⚫H2O, HSO3- (pKa1=1.9), and SO32- (pKa2=7.2) (hereafter also referred to as S(IV)) is calculated in the model. Thus, depending on the availability of oxidants and the solution's acidity, the different S(IV) species can participate in the formation of S(VI) (i.e., dissolved sulfur in oxidation state six).

In EC-Earth3-Iron, the aqueous-phase sulfur scheme is applied both in cloud droplets and aerosol water, replacing the S(VI) production through the dissolved S(IV) oxidation in cloud droplets previously included in the EC-Earth-AerChem (van Noije et al., 2014, 2021). In more detail, besides the two classic reactions of bisulfite and sulfite with hydrogen peroxide and ozone included in EC-Earth3-AerChem, additional reactions of S(IV) oxidation via methyl hydroperoxide (CH3O2H), peroxyacetic acid, and with the hydroperoxyl radical (HO2)/superoxide radical anion (O2-) are considered. Nevertheless, in acidic solutions, the oxidation by peroxides, and especially H2O2, is significantly more important than other oxidants (Herrmann, 2003; Jacob, 1986). H2O2 is produced in the gas phase and can be rapidly dissolved in the liquid phase due to its high solubility. The dissolved H2O2 (as well as the organic peroxides, such as CH3OOH) can react rapidly with the HSO3-. However, the pH-independent reaction of HSO3- with CH3OOH (or other organic peroxides) is expected to be less important than H2O2 under typical cloud conditions due to the much lower solubility of CH3OOH. Note that the dissociation of H2O2 is neglected here since it is not expected to significantly influence the total H2O2 concentrations under typical tropospheric conditions (Herrmann, 2003; Jacob, 1986). In contrast, at a higher pH, the S(IV) oxidation by ozone tends to dominate the S(IV) oxidation (Seinfeld and Pandis, 2006). O3 oxidizes rapidly all three S(IV) forms in the aqueous phase, becoming significant at pH higher than 4 (Seinfeld and Pandis, 2006), even in the absence of light. S(IV) oxidation by O3 is also predicted to dominate S(VI) formation during winter in arctic regions due to the lack of photochemical production of OH and H2O2 at high latitudes, as well as the high anthropogenic SO2 emissions in the Northern Hemisphere (Alexander et al., 2009). Laboratory studies indicate that S(IV) compounds may be also oxidized in the aqueous phase via other pathways. For example, the aqueous S(VI) production can be enhanced by TMIs (Harris et al., 2013), such as the Mn(II) catalyzed oxidation of S(IV) by dissolved O2. In a global modeling study, Alexander et al. (2009) attributed 9 %–17 % of the total S(VI) production to the latter mechanism. However, such reactions would require several oxysulfur radicals as intermediates (e.g., Deguillaume et al., 2004; Herrmann et al., 2005), like a free radical chain mechanism initiated by reactions of HSO3-, SO32- with radicals and radical anions, or TMIs catalyzed via oxidation of several S(IV) compounds, which is not considered in our model. Thus, in the case of the sulfate radical anion (SO4-) production via the Fe(III) sulfate complex [Fe(SO4)]+ photolysis (Table S2), the sulfate radical anion is simply added to the S(VI) pool.

Gas-phase organics can be also oxidized in the interstitial cloud space, form water-soluble compounds like aldehydes, and rapidly partition into the droplets. In the presence of oxidants such as OH and NO3 radicals in the solution, the dissolved organics undergo chemical conversions and form low-volatility organics that remain, at least partly, in the particulate phase upon droplet evaporation (Blando and Turpin, 2000). The dissolved OH radicals react with organic compounds in the aqueous phase by hydrogen abstraction or electron transfer, forming alkyl radicals (R), which in the presence of dissolved oxygen further form peroxyl radicals (RO2). The OH oxidation of organic compounds in the aqueous phase can lead to either fragmentation or the formation of oxidized organic species, resulting overall in CO2. However, the recombination of organic radicals can also be a favorable pathway when the water evaporates, and thus the aqueous solution becomes more concentrated. Box model simulations have shown that the cloud processing of polar products from isoprene oxidation can be an important contributor to secondary organic aerosol (SOA) production (Lim et al., 2005). Indeed, laboratory measurements show that the aqueous-phase photooxidation of C2 and C3​​​​​​​ carbonyl compounds (Perri et al., 2009, 2010), such as glyoxal (Carlton et al., 2007, 2009), methylglyoxal (Altieri et al., 2008), glycolaldehyde, pyruvic acid (Carlton et al., 2006), and acetic acid (Tan et al., 2012) leads to the production of low-volatility DCAs, which are commonly found in atmospheric aerosols and clouds (Sorooshian et al., 2006).

In EC-Earth3-Iron, gas-phase species can be reversibly transferred to the aqueous phase and oxidized by radicals and radical anions. The partitioning of 15 organic species that exist in both phases are considered in the aqueous-phase mechanism, namely methyl-peroxy radical (CH3O2), methyl hydroperoxide (CH3O2H), formaldehyde (HCHO), methanol (CH3OH), formic acid (HCOOH), acetaldehyde (CH3CHO), glycolaldehyde (GLYAL; HOCH2CHO), glyoxal (GLY; CH(O)CH(O)), ethanol (CH3CH2OH), acetic acid (CH3COOH), methylglyoxal (MGLY; CH3C(O)CHO), hydroxyacetone (HYAC; CH3C(O)CH2OH), pyruvic acid (PRV; CH3C(O)COOH), GLX, and oxalic acid (H2C2O4). The aqueous-phase oxidation is taking place by the OH and NO3 radicals, as well as the CO3- radical anion. OH is either produced by photolytic reactions of dissolved compounds or via a direct transfer from the gas phase into the solution, as well as by Fenton reaction (Deguillaume et al., 2010). NO3 radicals are transferred from the gas phase, while the CO3- radical anion is produced mainly via the oxidation of hydrated CO2. In general, the aqueous-phase oxidation largely proceeds via OH radicals, followed by NO3 radicals under dark conditions, while the CO3- radical has an overall small impact on the oxidizing capacity of the solution.

Upon their transfer to the solution, aldehydes are considered to be in equilibrium with the corresponding diols. The hydrated aldehydes are oxidized via H-atom abstraction with radicals (OH, NO3) or radical anions (CO3-), followed by the elimination of HO2 in reaction with O2, leading overall to the formation of organic acids. Alcohols, such as CH3OH and C2H5OH, are also oxidized via an H-atom abstraction; the resulting α-hydroxy-alkyl radicals, however, are not explicitly resolved, but the direct formation of aldehydes (e.g., formaldehyde and acetaldehyde) is considered via the respective peroxyl radical reactions with molecular oxygen to yield HO2. Moreover, the glycolic acid (HOCH2COOH) production via glycolaldehyde oxidation is not also explicitly described in the aqueous-phase scheme, and only the direct production of GLX is considered (Lin et al., 2012; Myriokefalitakis et al., 2011). This assumption is expected to have a negligible impact on the overall chemical mechanism since the glycolic acid is rapidly oxidized into glyoxylic acid with its net in-cloud production being rather small (Liu et al., 2012).

After cloud evaporation, OXL and SO42- are considered to reside entirely in the particulate phase of the model. This approximation may nevertheless result in an overestimate of OXL (pKa1=1.23; (COO-)2, pKa2=4.19) concentrations, since low levels of gas-phase oxalic acid have been also observed in the atmosphere under favorable conditions (e.g., Baboukas et al., 2000; Martinelango et al., 2007). Note that other products, such as pyruvate, glyoxylate, and the oligomers from GLY and MGLY, are also considered to reside in the particulate phase upon cloud evaporation (Lim et al., 2005; Lin et al., 2012; Liu et al., 2012) and are thus added directly to the SOA pool of the model. However, in contrast to OXL and the low-volatility oligomers, the pyruvic and glyoxylic acids are allowed to be partially transferred back to the gas phase of the model when the cloud droplets evaporate.

For the present work, the aqueous reaction rate coefficients are taken (where available) from the available literature of the CAPRAM schemes and supplemented with reaction rates from laboratory and modeling studies (i.e., Carlton et al., 2007; Deguillaume et al., 2009; Lim et al., 2005; Sedlak and Hoigné, 1993). For the sulfur chemistry, the aqueous reaction rates are taken from Seinfeld and Pandis (2006). In the case of missing experimental data for temperature dependencies, the rate constants for T=298 K are only applied in chemistry calculations. O3, H2O2, NO3, HONO/NO2-, HNO3/NO3-, CH3O2H, Fe3+, [Fe(SO4)]+, and [Fe(OXL)2]- are photolyzed in the aqueous phase. Aqueous photolysis frequencies (where available) are taken from the gas-phase chemistry. For Fe species (e.g., Fe3+, [Fe(SO4)]+, [Fe(OXL)2]-), their maximum (i.e., noon at 51∘ N) photolysis frequencies as proposed by Ervens et al. (2003) are scaled based on the gas-phase H2O2 photolysis rates. A list of all aqueous and photochemical reactions included in the chemical scheme of this study is presented in Table S2, with the respective equilibrium reactions shown in Table S3.

A three-stage kinetic approach (Shi et al., 2011) is applied to describe the solubilization of the Fe-containing dust mineral pools (Ito and Shi, 2016), representing: (1) a rapid dissolution of ferrihydrite on the surface of minerals (i.e., fast pool), (2) an intermediate stage dissolution of nano-sized Fe oxides from the surface of minerals (i.e., intermediate pool), and (3) the Fe release from heterogeneous inclusion of nano-Fe grains in the internal mixture of various Fe-containing minerals, such as aluminosilicates, hematite, and goethite (i.e., slow pool). A separate Fe pool for combustion aerosols (Ito, 2015) is also considered in the model.

The dissolved Fe in the model is produced via dissolution processes in aerosol water and cloud droplets depending on the acidity levels of the solution (i.e., proton-promoted dissolution scheme), the OXL concentration (i.e., ligand-promoted dissolution scheme), and irradiation (photo-reductive dissolution scheme), following Ito (2015) and Ito and Shi (2016). The Fe release from different types of minerals thus depends on the solution acidity (pH) and the temperature (T), as well as on the degree of solution saturation. In more detail, the dissolution rates for each of the three dissolution processes considered can be empirically described (e.g., Ito, 2015; Ito and Shi, 2016; Lasaga et al., 1994) as follows: 2RFei=KipH,T⋅α(H+)mi⋅fi⋅gi where Ki (mol Fe gi-1 s-1) is the Fe release rate due to the dissolution process i, α(H+) is the H+ activity of the solution, and mi is the empirical reaction order for protons derived from experimental data. The functions fi and gi represent the suppression of the different dissolution rates due to the solution saturation state as follows: 3fi=1-(aFe3+⋅aH+-ni)/Keqi,4gi=0.17⋅ln⁡(aOXLaFe3+)+0.63, where αH+, αFe3+​​​​​​​, and αOXL stand for the solution's activities of protons, ferric cations, and OXL, respectively, as calculated each time step in the model, and Keqi (mol2 kg-2) is the equilibrium constant. The activation energy that accounts for the temperature dependence is derived as a function of acidity based on soil measurements (Bibi et al., 2014; Ito and Shi, 2016), i.e. 5EpH=-1.56×103⋅pH+1.08×104 Overall, the net Fe dissolution rate results from the sum of the three rates. All parameters used for the calculation of dissolution rates for this work are presented in Table S4.

The annual net chemistry production of OXL (Table 2a) in EC-Earth is 12.615 ± 0.064 Tg yr-1, which is lower than in ERA-Interim (18.116 ± 0.071 Tg yr-1). The difference is explained by a higher oxidizing capacity in ERA-Interim than in EC-Earth. ERA-Interim calculates higher OH concentrations in the tropical and subtropical troposphere (Fig. S1b). In contrast, zonal mean OH levels in EC-Earth are slightly higher in the extratropics, causing a more efficient oxidation of the OXL precursors such as GLY (Fig. S1d), GLYAL (Fig. S1f), MGLY (Fig. S1h), and CH3COOH (Fig. S1j) at higher latitudes, especially in the Southern Hemisphere (SH). Note that van Noije et al. (2014) also showed that the simulated oxidizing capacity in the previous version of EC-Earth (EC-Earth v2.4) was lower compared to a respective ERA-Interim configuration in large parts of the troposphere, due to the simulated lower temperatures (cold biases) and specific humidities. However, since sea surface temperatures (SSTs) and sea ice concentrations are prescribed in our EC-Earth atmosphere-only simulations, the long-term means of tropospheric temperatures and water vapor are not expected to differ significantly from ERA-Interim close to the surface levels, as also indicated by the low differences in the OH levels of the two simulations at low altitudes (Fig. S1b).

The production term of OH via the H2O pathway in EC-Earth is ∼ 5 % lower than in ERA-Interim due to a lower amount of water vapor being available to react with O(1D). In addition, a ∼ 6 % lower OH production through the H2O2 photolysis is simulated in EC-Earth. Note that H2O2 is an important driver of the aqueous-phase oxidizing capacity in the model, with about 80 % of the OH radicals in the liquid phase being produced by photolysis of the dissolved H2O2. In more detail, the lower atmospheric abundance of the gas-phase H2O2 in EC-Earth (∼ 11 %) leads to smaller H2O2 uptake in the aqueous phase (∼ 13 %) and thus to a slower oxidation of OXL precursors due to the respective lower dissolved OH radical production (∼ 19 %). Overall, the total OH production is ∼ 7 % lower in EC-Earth, which corresponds to a ∼ 18 % lower aqueous-phase OH production, resulting in a ∼ 30 % lower OXL net chemistry production compared to ERA-Interim.

The total OXL production is 15.5 Tg yr-1 in Lin et al. (2014) and 14.5 Tg yr-1 in Liu et al. (2012), both of these values are lower than our ERA-Interim estimates (20.8 Tg yr-1; Table 2a) but close to EC-Earth (15.0 Tg yr-1; Table 2a). The main reason for the lower chemistry production of other published estimates compared to our results is the contribution of the aqueous-phase glyoxal oxidation scheme proposed by Carlton et al. (2007) that is applied in our simulations. The oxidation of the glyoxal-derived high-molecular-weight products formed mainly in the cloud droplets is calculated to contribute significantly to the global OXL production in our model (Table 2a). This result is in line with Carlton et al. (2007), who indicated that the GLX pathway may not be the primary pathway for oxalic acid formation, but this is instead attributed to the rapid oxidation of GLY multifunctional products via the OH radicals (i.e., ​​​​​​​3.1 × 1010 L mol-1 s-1; Table S2). However, for ERA-Interim(sens), where no such reactions are considered, the total OXL chemical production is calculated on average 11.5 Tg yr-1 (Table 2); i.e., closer to the estimates of Lin et al. (2014) and Liu et al. (2012). On the other hand, our ERA-Interim net chemistry production calculations are close to the estimates of Myriokefalitakis et al. (2011) (i.e., 21.2 Tg yr-1) when no potential effects of the ionic strength (e.g., Herrmann, 2003) on OXL precursors are considered, although this is still lower since no Fe chemistry was considered in that latter study. Indeed, the enhanced aqueous-phase oxidation capacity due to the Fenton reaction increases both the production and the destruction terms of OXL in our model, leading to ∼ 7 % lower net OXL production and a lower (∼ 8 %) atmospheric abundance, respectively. Nonetheless, our calculations indicate that Fe chemistry impacts on OXL net production drastically, increasing the destruction of the dissolved oxalic acid by at least ∼ 50 %. The potential primary sources (0.373 ± 0.005 Tg yr-1) accounted for in the model (Table 2a) do not, however, significantly contribute to the simulated OXL atmospheric levels, and only a small fraction of OXL is calculated to be formed in aerosol water (∼ 6 %) for all simulations in this work.

Focusing further on the atmospheric sinks of OXL, roughly 13 % in ERA-Interim and 16 % in EC-Earth of the produced oxalic acid is oxidized into CO2 in the aqueous phase, mainly via the photolysis of the [Fe(OXL)2]- complex (∼ 55 %) and via OH radicals (∼ 45 %). The fraction of the total produced OXL that is destroyed in the aqueous-phase is higher than in Liu et al. (2012) by ∼ 7 %, where no Fe chemistry was considered, but lower compared to Lin et al. (2014) and Myriokefalitakis et al. (2011), where roughly 30 % of the produced OXL is oxidized into CO2 in the aqueous phase. Finally, a total average deposition rate of 18.5 Tg yr-1 is calculated in ERA-Interim, primarily due to wet scavenging (∼ 99 %), resulting in a global atmospheric lifetime of 5.7 d, which is close to Liu et al. (2012) and Lin et al. (2014) but higher compared to Myriokefalitakis et al. (2011) (∼ 3 d); this is probably because of the more intense OXL production at higher altitudes in our model.

The major pathways of global OXL production, both in ERA-Interim and EC-Earth, are the oxidation of glyoxal (∼ 74 %), followed by glycolaldehyde (∼ 11 %), methylglyoxal (∼ 8 %), and acetic acid (∼ 7 %). Glyoxylic acid is nevertheless an important intermediate species because it is directly converted to OXL in the aqueous phase upon oxidation. Other important findings concerning the chemical budgets are summarized below.

Glyoxal. About 70 Tg yr-1 GLY is produced in the gas-phase in ERA-Interim, similar to Lin et al. (2014), while in EC-Earth it is calculated 3 % lower. The global gas-phase production of the present work is higher than other global model estimates, e.g., about 56 Tg yr-1 (Myriokefalitakis et al., 2008), 40 Tg yr-1 (Fu et al., 2009, 2008), and 21 Tg yr-1 (Liu et al., 2012). This difference can be explained by the more comprehensive isoprene chemistry of the gas-phase scheme used here (Myriokefalitakis et al., 2020b). Indeed, isoprene secondary oxidation products (e.g., epoxides) are significant precursors of GLY in the atmosphere (Knote et al., 2014) and the contribution of isoprene epoxides (IEPOX) from the gas-phase isoprene oxidation is here considered as a pathway of GLY formation. Note that the oxidation of other biogenic hydrocarbons, like terpenes and other reactive organics, may also result in GLY formation, since their chemistry is lumped on the first-generation peroxy radicals of isoprene in the model (Myriokefalitakis et al., 2020b). Besides the biogenic hydrocarbon oxidation, the model considers GLY formation due to the oxidation of other organic species (e.g., Warneck, 2003), such as acetylene (4.8 Tg yr-1) and aromatics (18.8 Tg yr-1). In the gas phase, other hydrocarbons, like ethene, further contribute to the atmospheric production of GLY via their oxidation products, mainly glycolaldehyde (5.4 Tg yr-1). However, as in many modeling studies, additional primary and/or secondary glyoxal sources might be still missing in our model. Indeed, the elevated glyoxal concentrations over oceans that have been observed from space (e.g., Wittrock et al., 2006) would require at least 20 Tg yr-1 of extra marine sources to reconcile model simulations with satellite retrievals (Myriokefalitakis et al., 2008). Great uncertainties, however, still exist on these oceanic sources (Alvarado et al., 2020; Sinreich et al., 2010), and therefore the only glyoxal primary sources accounted for in the model are from biofuel combustion and biomass burning processes (e.g., Christian et al., 2003; Fu et al., 2008; Hays et al., 2002), overall resulting in about 7 Tg yr-1 on average in the model (Myriokefalitakis et al., 2020b). Glyoxal is rapidly destroyed in the atmosphere via photolysis (∼ 70 %), followed by its oxidation in the gas phase (∼ 15 %) and the aqueous phase (∼ 15 %). Roughly 5.4 Tg yr-1 of glyoxal is produced in the aqueous phase via the dissolved GLYAL oxidation in ERA-Interim, close to the Liu et al. (2012) calculations but somehow higher compared to EC-Earth. Overall, the net cloud uptake of glyoxal in ERA-Interim is 6.3 Tg yr-1, which is higher than the estimates from Liu et al. (2012) (1.6 Tg yr-1). As expected, this increase is due to the applied glyoxal oxidation scheme in the aqueous phase of our base simulations. Finally, 4.2 Tg yr-1 of glyoxal is removed from the atmosphere via wet scavenging (∼ 73 %) and dry deposition (∼ 27 %).

Figure 2a presents the annual mean (average 2000–2014) net chemistry production rates of OXL in EC-Earth and the respective absolute differences compared to ERA-interim (Fig. 2c). The maximum OXL production rates are calculated around the tropics and in the Southern Hemisphere, where both biogenic emissions (mainly isoprene) and the liquid cloud water are substantially enhanced (Fig. S2a). The Amazon region appears as the largest source of OXL, along with central Africa and Southeast Asia. At higher latitudes (>45∘ N) the lower cloud liquid water content and vegetation cover lead to a lower OXL production over Asia and North America. However, over highly populated regions in the Northern Hemisphere, such as in Europe, the US, and China, enhanced OXL production rates are calculated due to its anthropogenic precursors. Furthermore, a significant source of OXL is calculated downwind of land areas, such as the South Pacific and the tropical Atlantic Ocean due to the long-range transport of OXL precursors.

The illustrated differences in OXL production between ERA-interim and EC-Earth (Fig. 2c) are caused due to the adopted atmospheric dynamics (i.e., online calculated versus offline), as both simulations use identical prescribed anthropogenic and biogenic emissions (see van Noije et al., 2021). For instance, EC-Earth calculates higher cloud water concentrations at ∼ 800–600 hPa around the tropics (30∘ S–30∘ N) compared to ERA-Interim. In contrast, lower concentrations are derived aloft, with ERA-Interim presenting enhanced cloud water concentrations at ∼ 400 hPa (Fig. S2b). Moreover, due to the lower OH concentrations in the tropical and subtropical troposphere (Fig. S1b), EC-Earth gives lower OXL production rates, especially over intense biogenic emission areas. Overall, the difference in the oxidizing capacity of the atmosphere between the two configurations significantly impacts the aqueous-phase OXL production efficiency in the model.

Sulfate (SO42-) is the main inorganic aerosol species produced in the aqueous phase, and similar to OXL its production in the model mainly occurs in cloud droplets. In addition, these two species largely reside in the aerosol accumulation mode of the model (roughly 99 % for SO4-2 and 97 % for OXL). SO42- is a key species for determining atmospheric acidity, and therefore here we also present the sulfate budget in conjunction with that of OXL. Sulfate is produced both in cloud droplets and in aerosol water, with the production in aerosol water having a negligible contribution on a global scale. In contrast to OXL, for which no gas-phase production is considered, the gas-phase oxidation of SO2 via OH radicals contributes to the total SO42- concentrations with about 12.0 Tg S yr-1 (Table 2b). Our global estimate of the gaseous sulfuric acid (H2SO4) production is higher than in EC-Earth v2.4 (7.8 Tg S yr-1 averaged for the years 2000–2009; van Noije et al., 2014) but slightly lower than in the EC-Earth3-AerChem AMIP simulations (van Noije et al., 2021) used for the CMIP6 experiments (available in https://esg-dn1.nsc.liu.se/search/cmip6-liu/, last access: 11 June 2021) where 12.9 Tg S yr-1 of H2SO4 are produced (averaged for the years 2000–2014). These differences can be directly attributed to the OH radical production rates in the gas phase between the new and the previous chemistry versions of the atmospheric model, as have been discussed in Myriokefalitakis et al. (2020b). Despite the generally lower gas-phase OH radical levels (Fig. S1b), the slightly higher (∼ 8 %) global H2SO4 gas-phase production rate in EC-Earth than in ERA-Interim (Table 2b) can be attributed to the higher (∼ 6%) DMS emissions in EC-Earth (Fig. S4b) that contribute to the atmospheric SO2 levels over the ocean.

The aqueous-phase SO42- chemistry production from the oxidation of dissolved SO2 is 39.8 Tg S yr-1 in EC-Earth (Table 2b), which is higher than in EC-Earth v2.4 (29.3 Tg S yr-1; van Noije et al., 2014) and EC-Earth3-AerChem (32.5 Tg S yr-1). The higher SO42- chemical production is mainly due to the higher SO2 aqueous-phase oxidation rates by H2O2. In more detail, our calculations show that ∼ 84 % of the global SO42- production in EC-Earth is due to the dissolved SO2 oxidation via H2O2; 33.3 Tg S yr-1 is produced due to H2O2, which is higher compared to 23.9 Tg S yr-1 in van Noije et al. (2014). The dissolved SO2 oxidation via O3 (6.4 Tg S yr-1) is also higher than in EC-Earth v2.4 (5.4 Tg S yr-1). However, the contribution of CH3O2H to the SO42- aqueous-phase production is small (0.05 Tg S yr-1) in the model, with the HO2 contribution being practically negligible on the global scale (0.02 Tg S yr-1) for all simulations performed in this study. A total annual mean deposition rate of 52.5 Tg S yr-1 is simulated in EC-Earth, with wet scavenging dominating the total deposition rate (∼ 93 %). Note that 2.5 % of the sulfur in the SO2 emissions (1.6 Tg S yr-1) is assumed to be in the form of SO42- for all simulations, which accounts for its formation in the sub-grid plumes (Aan de Brugh et al., 2011; Huijnen et al., 2010). Overall, a global SO42- lifetime over deposition of 4.8 d is calculated in EC-Earth, which is lower than in ERA-Interim (6.6 d) but similar to the EC-Earth v2.4 estimate (4.9 d).

Figure 2b also shows the annual mean SO42- net chemistry production rates in EC-Earth. High SO42- production rates are calculated downwind of major anthropogenic SO2 emission hotspots, such as central Europe, the eastern US, India, Russia, and eastern Asia. Furthermore, relatively high production rates due to biomass burning and volcanic eruptions are calculated in South America, southern Africa, and Indonesia. Significant SO42- production is calculated over almost all oceanic regions due to the SO2 production via the gas-phase oxidation of marine DMS emissions (Fig. S4a). Compared to the ERA-Interim simulation, however, the SO42- production rates in EC-Earth are on average slightly higher over land in the tropics and extratropics. This increase can be attributed to combined effects that result in differences in chemical production and deposition rates (Table 2b). Some differences over oceans are nevertheless expected due to the differences in DMS concentrations, since DMS emissions are calculated online in the model based on sea surface temperature and wind velocity (Fig. S4b).

In EC-Earth, the total Fe (TFe) soil emissions result in 59.33 ± 1.22 Tg yr-1, while in ERA-Interim they are 48.96 ± 0.95 Tg yr-1. This difference results from the differences in wind speed between EC-Earth and ERA-Interim. EC-Earth produces higher dust emissions over large parts of the Middle East and Asia compared to ERA-Interim (Fig. S4f), which explains the differences in TFe emissions (TFeC emissions do not differ). However, most of the dissolved Fe from mineral dust in the model originates from atmospheric dissolution processes. In EC-Earth, FeD is primarily dissolved due to aerosol acidity at 0.31 Tg yr-1, followed by the ligand-promoted dissolution that additionally produces 0.17 Tg yr-1, while the photoinduced processes have a small impact on the global dissolved Fe release from dust, with 0.05 Tg yr-1 (Table 2d). Fe primarily resides (98.4 %) in the slow pool of Fe-containing dust aerosols in the model, in particular in the coarse mode, with about 1.0 % being emitted as nano-sized iron oxides (intermediate pool) and 0.5 % as ferrihydrite (fast pool). Thus, most of the dissolved Fe release originates from the heterogeneous inclusion of nano-Fe grains in the internal mixture of various Fe-containing minerals such as aluminosilicates, hematite, and goethite (∼ 66 %), followed by nano-sized iron oxides (∼ 24 %) and to a lesser extent by ferrihydrite (∼ 10 %). Note, however, that the Fe release from aluminosilicates, hematite, and goethite particles is a slower process compared to the other soil classes considered in the model, as dictated by the three-stage approach applied for this study (Table S4).

Fe emissions from combustion processes are estimated at 2.518 ± 0.105 Tg yr-1 in both simulations, with 0.012 Tg yr-1 being emitted as dissolved from the primary oil combustion processes. Roughly 0.274 ± 0.010 Tg yr-1 are released through Fe dissolution from combustion aerosols in EC-Earth, in good agreement with ERA-Interim (0.285 ± 0.011 Tg yr-1). The acid-promoted dissolution contributes ∼ 17 % and photo-reductive processes ∼ 16 % to the Fe release from combustion particles, thus most Fe release comes from the ligand-promoted dissolution. This result is in line with laboratory studies (e.g., Chen and Grassian, 2013), where the contribution of oxalate-promoted dissolution is several times larger than the proton-promoted pathway under highly acidic dark conditions. According to our calculations, the relative contribution of atmospheric processing to the combustion aerosol Fe solubilization (∼ 11 %) is significantly higher compared to that of crystalline dust minerals (∼ 1 %), in agreement with laboratory (e.g., Chen et al., 2012; Fu et al., 2012) and modeling (e.g., Ito, 2015; Ito and Shi, 2016) studies.

The annual mean dissolution rates of FeC and FeD in EC-Earth are presented in Fig. 3. For combustion aerosols, the maximum dissolution rates occur downwind of biomass burning sources and highly populated regions, such as South America and Central Africa, the Middle East, India, and China. High dissolution rates are more likely to coincide with high OXL concentrations (Ito, 2015). Indeed, the model calculates important dissolution rates near regions where the OXL production rates are enhanced (Fig. 2a), such as over the Amazon basin and central Africa, as well as downwind of these regions, as the combustion aerosols are transported to the open ocean, in agreement with observations (e.g., Sholkovitz et al., 2012). For the mineral dust aerosols, most of the FeD dissolution fluxes occur downwind of the major dust source regions (e.g., the Sahara and the Gobi Desert), where the atmospheric transport of anthropogenic pollutants, such as SOx and NOx, enhances atmospheric acidity; e.g., the Fe release from the dust minerals due to proton-promoted dissolution processes is enhanced over the Middle East. Significant dissolution rates are also simulated over the Atlantic Ocean at the outflow of the Sahara, as well as at the outflow of Asian desert regions to the Pacific Ocean. High rates due to the contribution of the organic ligand-promoted dissolution processes are calculated downwind of central Africa and the equatorial Atlantic Ocean, where the oxidation of biogenic hydrocarbons in the presence of cloudiness leads to enhanced OXL aqueous-phase formation rates. On the contrary, the efficiency of ligand-promoted dissolution is substantially suppressed near dust source regions due to the low OXL availability (Fig. 2a).

Figure 3 also presents the absolute differences between the ERA-Interim and EC-Earth annual mean Fe dissolution rates. ERA-Interim has significantly lower dissolution rates in the tropics (e.g., central Africa) and around the Equator, both for FeC (Fig. 3c) and FeD (Fig. 3d). This decrease is attributed both to the differences in atmospheric dynamics between the two model configurations and the suppression of the organic dissolution processes with lower OXL production. Indeed, Fig. 2c shows that in ERA-Interim, OXL production rates increase in the tropics, impacting Fe dissolution rates. In contrast, FeC dissolution rates increase in ERA-Interim over the Arabian Peninsula, India, and eastern Asia, due to fluctuations in OXL production and aerosol acidity. EC-Earth also shows lower FeD dissolution rates over the northern Pacific in the outflow of Asia. These differences are due to a higher aerosol acidity (i.e., up to ∼ 1 pH unit; Fig. S3d, f) in ERA-Interim due to changes in the buffering capacity of dust promoted by the higher calcite emissions in EC-Earth (Fig. S4f). This is especially the case for coarse dust aerosols where the majority of the Fe resides. EC-Earth also shows differences (positive or negative) with ERA-Interim in the Fe dissolution rates over oceanic regions (Fig. 3e, d), likely due to differences in SO42- production over oceans from marine DMS emissions (Fig. S4b) and the impact of sea salt emissions (Fig. S4d) upon the buffering capacity of the solution. All in all, the total DFe atmospheric source in EC-Earth, accounting for both primary emissions and atmospheric processing, is 0.806 ± 0.014 Tg yr-1 for the present day, well within the range of estimates presented in the model intercomparison study (0.7 ± 0.3 Tg yr-1) in Myriokefalitakis et al. (2018).

The annual mean AOD at 550 nm modeled in EC-Earth for 2000–2014 compares favorably with AERONETv3 direct-sun level 2.0 data (Fig. 4a). Overall, the model presents an nMB of -9 % and an nRMSE of 46 %, considering information from 738 AERONET sites. The regional analysis suggests a slightly better behavior in northern hemisphere regions (e.g., North America, Europe, East Asia) dominated by anthropogenic aerosols (normalized errors and biases, below 45 % and ± 10 %, respectively). The largest deviations from observed AOD occur in the Southern Hemisphere (e.g., South Africa, Australia, and Oceania) or remote regions. Over dust-dominated regions (e.g., North Africa, West Asia, and the Middle East) the model also behaves well (with normalized errors and biases below 45 % and ± 10 %, respectively). Selecting specifically dust-dominated sites for the comparison (Fig. 4b) and following the criteria explained in Sect. 2.5, EC-Earth slightly overestimates the retrieved AOD at 550 nm over North Africa (nMB = 21 %) and shows underestimations over sources in West Asia and the Middle East, as well as in transport regions such as Central America. In general, EC-Earth's ability to reproduce the annual mean AOD at 550 nm holds for dusty sites, with a normalized mean bias of 3 %, and a normalized root-mean-square error of 37 % over 38 sites. Overall, the average optical depth for dust at 550 nm (annual mean over the 2000–2014) yields a value of 0.032 ± 0.005, which falls well in the range of observationally based estimates based on in situ measurements, satellites, and global models (i.e., 0.030 ± 0.005; Ridley et al., 2016).

The comparison of model outputs with climatologies of dust surface concentration from the RSMAS and the AMMA campaign (Fig. 4c) yields slightly poorer results, with an nMB of 19 % and an nRMSE of 81 %, as an average of the 23 sites available. EC-Earth best reproduces dust surface concentrations over source regions, such as North Africa (nMB = 21.9 %, nRMSE = 37.7 %), shows underestimations in transport areas (e.g., Central America: nMB =-37 %; nRMSE = 39 %) and poorly represents the surface concentration in remote regions (e.g., the South Pacific and Southern Ocean, with nMB up to -98 % and nRMSE up to 113 %). The evaluation of the dust deposition field (Fig. 4d) shows both positive and negative biases over source and transport regions (see Table A1), with the deposited mass being generally underestimated, except for the Southern Ocean where the model tends to overestimate the observations. EC-Earth3-Iron may thus share the difficulties of many global models in representing the long-range transport of dust, particularly coarse particles downwind of dust sources (e.g., Adebiyi and Kok, 2020). As minerals in dust constitute the primary source of TFe to the atmosphere, the aforementioned discrepancies with respect to observations (e.g., higher concentrations over dust source areas and an underestimation of the dust deposition rates) are expected to also affect the representation of dust-related DFe in the model.

The averaged OXL surface concentrations in EC-Earth for the boreal winter (December, January, and February, i.e., DJF) and summer (June, July, and August, i.e., JJA) are presented in Fig. 5. OXL surface concentrations are distributed roughly between 60∘ S and 60∘ N, mainly in regions where intensive volatile organic compound (VOC) emissions from anthropogenic and biogenic sources coexist with cloud water. The highest OXL concentrations are calculated over tropical Africa, the Amazon Basin, eastern Asia, the eastern United States, and Europe, clearly showing the strong impact of OXL precursors (e.g., glyoxal) and the availability of cloud water. In the Northern Hemisphere, OXL concentrations are generally calculated higher in summer and lower in winter, indicating a strong impact of temperature and photochemistry on the production rate of oxalic acid in the aqueous phase. During DJF, the model calculates lower OXL concentrations over midlatitude and high-latitude regions, such as East Asia, central Europe, and the northern US. Over these highly populated regions, the aerosol water content is enhanced, following the increased SO4-2 production due to anthropogenic activities, and the aqueous-phase OXL production in deliquesce particles also contributes to OXL atmospheric concentrations. Furthermore, high OXL concentrations are calculated in the tropics for both seasons due to the photochemical activity and the intense sources of biogenic VOCs in these regions.

Figure 5 further presents the differences of OXL concentrations between EC-Earth and ERA-Interim. In general, OH levels in ERA-Interim are higher, which causes a more efficient oxidation of OXL precursors for both seasons. Moreover, ERA-Interim shows higher concentrations around the intertropical convergence zone (ITCZ) due to differences in meteorology between the two simulations, as discussed above. During boreal winter, some differences are observed over the subtropics of the Northern Hemisphere. Although OXL concentrations are very low over these latitudes, the relatively strong increase in liquid water that serves as a medium for OXL production, both for clouds (Fig. S2b) in higher altitudes (∼ 400 hPa) and at the surface for deliquescent particles (Fig. S2d, e) in the ERA-Interim simulation. In the vertical, OXL concentrations are distributed in the model from the surface to ∼ 400 hPa with a maximum at around 900 hPa. The zonal mean differences, however, indicate strong increases in the Southern Hemisphere (30∘ S–0∘​​​​​​​) during boreal winter (Fig. 5e). Compared to EC-Earth, ERA-Interim calculates higher OXL concentrations in the upper troposphere for both seasons (Fig. 5e, f), mostly due to more efficient transport of OXL precursors by deep convection into the tropical and extratropical upper troposphere. In the lower and middle troposphere, higher concentrations are calculated in ERA-Interim depending on the location and the season. The concentrations in EC-Earth are also lower than in ERA-Interim in the Northern Hemisphere (NH) extratropics during boreal summer due to a lower chemical production (Fig. 5f).

Figure 6 presents the comparison of the different model simulations performed for this work with OXL surface observations. OXL concentrations show a strong seasonal dependence, with maxima during the warm season due to the intense photochemical activity combined with the higher precursor abundance. Over the Mediterranean, and specifically the eastern part which is characterized by the long-range transport of air pollution and from surrounding urban centers (Kanakidou et al., 2011), the model underestimates the observed concentrations during winter at the Finokalia station in all simulations (Fig. 6a), either due to missing OXL primary and secondary sources or a too strong removal. During summer, ERA-Interim satisfactorily simulates the observed OXL levels, also representing the observed trend, which indicates that the model reproduces the mixing and aging of the air masses in the region under favorable meteorological conditions and intense solar radiation. EC-Earth calculates lower OXL concentrations than ERA-Interim due to the lower oxidizing capacity, thus underestimating the observed concentrations for all seasons. On the other hand, ERA-Interim(sens) tends to underestimate the observations for all seasons more than the other simulations, further indicating the important role of the secondary sources to OXL atmospheric concentrations in the region. At the Puy de Dôme site (Fig. 6c), which is located at 1450 m a.s.l., ERA-Interim underestimates the observed OXL concentrations, although it simulates them more realistically compared to EC-Earth, especially during summer (Fig. 2c). The seasonal variation in the area can be explained by the stronger upward transport of air masses during summer (Legrand et al., 2007), thus increasing the OXL production in the region. However, the model fails to represent the observed OXL levels, possibly due to missing sources. The importance of other production pathways not related to the aqueous-phase GLX oxidation is demonstrated in the comparison of the observed OXL levels with the ERA-Interim(sens) simulation. Again, ERA-Interim(sens) deviates more strongly from the measured values than other simulations. Nevertheless, this indicates that other species may further contribute to OXL production, such as the decay of longer diacids (e.g., azelaic and malonic acids) (Legrand et al., 2007) that are currently not included in the model. Another reason may be the impact of the enhanced cloud LWC in the region, implying a more intense cloud processing compared to other surface sites and thus a faster oxidation of oxalic acid into CO2 (Ervens et al., 2004) in the model. Finally, at the Azores (Fig. 6e), a site that is characterized by a marine environment, the model tends to underestimate the observed OXL concentrations most of the time, with ERA-Interim again presenting a better skill than other simulations. EC-Earth underestimates the observed concentrations more than ERA-Interim, especially during summertime, and ERA-Interim(sens) simulates the lowest OXL concentrations. The observed OXL levels in the region, however, can be explained either by the transport of pollutants from the continents or the photochemical production in the region. Thus, the illustrated differences against the observations between EC-Earth and ERA-Interim can be attributed to differences in the oxidizing capacity and in simulated transport of EC-Earth, such as the vertical mixing in the troposphere (e.g., van Noije et al., 2014) that has a further impact on OXL precursors like glyoxal. Furthermore, since the long-range transport is found to be relatively constant in summer and winter in the region (Legrand et al., 2007), other species of marine origin, such as the unsaturated fatty acids (e.g., linoleic and oleic acids) may also contribute as precursors to the OXL production, especially during summer, but are not included in the model. All in all, we acknowledge that other formation pathways of OXL, both primary or secondary, may exist in the atmosphere (e.g., Baboukas et al., 2000); for example, higher DCAs (such as malonic, succinic, glutaric, and adipic acids) may act as precursors for smaller dicarboxylic acids like OXL, both in the gas phase (e.g., Kawamura and Ikushima, 1993) and the aqueous phase (e.g., Ervens et al., 2004; Lim et al., 2005; Sorooshian et al., 2006) and could further contribute to atmospheric OXL concentrations.

In Fig. 6g, OXL observations reported in the literature are compared with monthly mean simulations. Due to the relatively low resolution of the global model (i.e., 3∘ × 2∘ in longitude by latitude), the spatial variability of urban emissions cannot be well resolved. Therefore, urban stations are omitted, and the comparison is limited to locations representative of background concentrations. All simulations tend to underestimate OXL observations, with lower biases in ERA-Interim (i.e., nMB =-46 %, nRSME = 110 %). As expected, for almost all sites the ERA-Interim simulation calculates the highest OXL concentrations and the ERA-Interim(sens) the lowest (i.e., nMB =-74 %, nRSME = 125 %). The latter indicates that additional production pathways need to be considered in modeling studies to capture the observed OXL concentrations. EC-Earth underestimates the observed concentrations more than ERA-Interim (i.e., nMB =-64 % and nRSME = 117 %) but less than ERA-Interim(sens), also highlighting the importance of the atmospheric oxidating capacity and the atmospheric dynamics in the OXL production. A summary of statistics for the evaluation of the simulated OXL concentrations for the different simulations is presented in Table A2. Overall, our analysis indicates that the model either misses OXL sources (primary and secondary) or overestimates OXL sinks, especially during winter. Thus, under relatively low temperatures and irradiation, the model may not be representative of the fast secondary OXL production in wood-burning plumes or the secondary production through species produced by the oxidation of emitted from vehicles and other anthropogenic activities, such as ethane and aromatic hydrocarbons.

The averaged SO42- surface concentrations as calculated in EC-Earth for the boreal winter and summer are presented in Fig. 7. During both seasons, high SO42-concentrations are simulated near or downwind major anthropogenic emission hotspots, where the vast majority of the surface SO2 emissions from anthropogenic origin occur (e.g., Tsai et al., 2010). Enhanced SO42- surface concentrations are also calculated downwind of biomass burning and volcanic eruptions, showing the overall impact of SO2 primary sources and the abundance of cloud water over these latitudes. Over the remote oceans, however, DMS oxidation may significantly contribute to the SO42- surface concentrations, as is also the case for the sulfur emissions over major shipping routes. In Fig. 7, the differences between EC-Earth and EC-Earth-AerChem in the averaged SO42- surface and zonal mean concentrations are also presented for both boreal winter (Fig. 7c, e) and summer (Fig. 7d, f). Considering, however, that the EC-Earth version developed for this work is based on the EC-Earth-AerChem model version, the illustrated differences are solely due to the applied chemistry schemes in the model. During boreal winter, the largest differences appear over eastern Asia, the Middle East, India, and central Europe. In EC-Earth-AerChem, the gas-phase SO2 oxidation by OH radicals is roughly 8 % larger than in EC-Earth (see Sect. 3.1.2), leading to an overall lower conversion of SO2 to sulfate in the aqueous phase. Moreover, in our simulations SO42- is also produced in deliquesced particles, which partly contribute to the SO42- atmospheric concentrations, especially over highly populated regions during the boreal winter when sulfur emissions due to anthropogenic activities are enhanced. On the contrary, during boreal summer some differences are illustrated (Fig. 7d) mostly over the Middle East where the EC-Earth-AerChem simulation results in lower SO42- concentrations, and over eastern Asia, where some higher concentrations in EC-Earth-AerChem occur. These negative and positive differences are attributed to differences in oxidizing capacity between the two models in both the gas and the aqueous phase. Finally, the zonal mean differences indicate higher concentrations in EC-Earth in the lower troposphere of the Northern Hemisphere during boreal winter (Fig. 7e) than for the SO42- surface concentrations. For boreal summer no important differences are presented here either (Fig. 7f).

Figure 6 further presents the model comparison with SO42- surface observations. Sulfate concentrations maximize under intense photochemical activity and high SO2 atmospheric levels; generally also suggesting a faster formation rate compared to oxalic acid (Ervens et al., 2004; Legrand et al., 2007). At the Finokalia station in the Mediterranean (Fig. 6b), the model overestimates the observed SO42-concentrations during boreal winter and summer in ERA-Interim, probably due to too high SO2 background concentrations because of a too strong long-range transport from surrounding regions. During winter, EC-Earth better reproduces the observations, probably due to the lower oxidizing capacity compared to ERA-Interim, but in late spring and early summer the active photochemistry in the region leads again to an overestimation of the observed concentrations. The EC-Earth-AerChem simulation leads to generally lower concentrations compared to our EC-Earth simulation, tending to somehow underestimate the observed concentrations except for autumn. At Puy de Dôme (Fig. 6d), ERA-Interim, which has a higher cloud liquid water content aloft and a more intense oxidizing capacity compared to EC-Earth, overestimates the observed SO42-concentrations in almost in all seasons. In contrast, EC-Earth better simulates the measured SO42- concentrations, although it seems to underestimate the observations during summer. At that site, the EC-Earth-AerChem calculations agree well with EC-Earth, although concentrations are again slightly lower. At the Azores site (Fig. 6f), ERA-Interim also simulates the observed concentrations well, following the observed annual cycle. In contrast, both EC-Earth and EC-Earth-AerChem underestimate the SO42- observations, especially during spring and summer. Note that the SO42- production in this marine site is attributed to the SO2 atmospheric levels both from air masses advected from industrialized regions and the local production due to the oxidation of marine DMS emissions. Thus, the differences between ERA-Interim and the other EC-Earth simulations presented in this study may indicate slower aging of the polluted air masses transported in the region. Finally, a comparison of the model's monthly mean predictions with a compilation of SO42- observations (n=3828) around the globe (Daskalakis et al., 2016) is presented in Fig. 6h. EC-Earth tends to overestimate the available SO42- observations (Table A2), presenting positive biases (i.e., nMB = 16 %, nRSME = 55 %) that are slightly lower than in ERA-Interim (i.e., nMB = 23 %, nRSME = 57 %). In contrast, EC-Earth-AerChem tends to slightly underestimate (Table A2) the observed concentrations (i.e., nMB =-5 %, nRSME = 57 %), showing a slightly lower correlation coefficient (R=0.70) than EC-Earth (i.e., R=0.76) and ERA-Interim (i.e., R=0.75).

Figures 8 and 9 present the averaged dissolved FeC (DFeC) and FeD (DFeD) surface concentrations, respectively, for DJF and JJA. EC-Earth calculates an annual global DFeC atmospheric burden of 0.002 Tg, while ERA-Interim calculates slightly higher concentrations (0.003 Tg) due to the more intense ligand-promoted dissolution rates (Table 2c). Elevated DFeC concentrations during boreal winter (Fig. 8a) are calculated over central Africa, eastern Asia, and India, where significant DFe concentrations (∼ 0.01–0.1 µg m-3) are associated with biomass burning and anthropogenic combustion emissions. During boreal summer (Fig. 8b), the maximum DFeC concentrations are calculated in the northern latitudes, in particular over the Mediterranean Basin, the Middle East, the western US, and China. The increase in the surface dissolved Fe concentrations over these regions, ranging from ∼ 0.01 to 0.1 µg m-3, clearly highlights the anthropogenic contribution due to the enhanced solubilization of Fe when the combustion aerosols are mixed with acidic and organic pollutants during atmospheric transport (Fig. 3b). Due to the intense biomass burning in the Southern Hemisphere (i.e., South America, central Africa, and Indonesia), the enhanced OXL production rates over such regions (Fig. 2a) lead to higher dissolved Fe concentrations. Figure 8a and b demonstrate that the geographic pattern of the DFeC concentrations may change overall from boreal winter to summer, following the biomass burning activity and the atmospheric processing of Fe-containing combustion aerosols of anthropogenic origin.

Mineral dust emissions mostly occur in the midlatitudes of the Northern Hemisphere (Fig. 9a, b), relatively close to where the vast majority of the population exists, and the anthropogenic emissions of acidic compounds dominate. For both seasons, high DFe concentrations from dust occur over the midlatitudes of the Northern Hemisphere, where the major dust sources are located. However, the equatorial maximum during boreal winter tends to shift to the north during boreal summer following the migration of the ITCZ (Fig. 9b). DFe from mineral dust aerosols maximize over the major dust source regions, with surface concentrations of roughly 0.1–1 µg m-3 (Fig. 9a, b) overall dominating the Fe burden. The outflow from these regions transports DFe over the global oceans, where secondary maxima of ∼ 0.01–0.1 µg m-3 are calculated, mainly over the Northern Hemisphere in the tropical Atlantic Ocean. The dissolved Fe associated with Saharan dust is nevertheless attributed to the long-range transport and the atmospheric processing that converts the insoluble Fe minerals to soluble forms.

The differences between the EC-Earth and ERA-Interim are illustrated in the averaged DFeC (Fig. 8c, d) and DFeD (Fig. 9c, d) surface concentrations. The differences in DFeC between the EC-Earth and ERA-Interim are well correlated with those of OXL concentrations (Fig. 5c, d), indicating the strong impact of ligand-promoted dissolution on the DFeC atmospheric load (Table 2c). Note, however, that the most important relative differences between the two simulations are calculated over regions with low dissolved FeC concentrations, and consequently the total burden does not change significantly (Table 2c). The differences in the DFe concentrations associated with mineral dust aerosols follow the general anomaly pattern of the two model configurations due to differences in transport that lead overall to higher (∼ 15 % globally) dust emissions in EC-Earth (Fig. S4f). We note, however, that the annual mean dust emission in EC-Earth3-Iron amounts to 1257 ± 26 Tg yr-1 (years 2000–2014 averaged), which falls in the lower range of the AEROCOM phase III models (Gliß et al., 2021) and is also at the low end of the range estimated by Kok et al. (2021) from inverse modeling (1200–2900 Tg yr-1) for dust aerosol with a geometric diameter ≤10 µm.

Figure 10 presents a comparison of the different model simulations with cruise observations of dissolved Fe concentrations. The spatial distributions of the DFe observations for the accumulation, coarse, and total suspended aerosols are shown in Fig. 10a–c, respectively. The median (mean ± standard deviation) DFe concentration in the accumulation mode amounts to 0.96 (3.68 ± 6.44) ng m-3 in the observations, while in EC-Earth and ERA-Interim it is 3.67 (5.79 ± 5.17) ng m-3 and 3.12 (5.25 ± 4.54) ng m-3, respectively. The respective observed concentration of DFe in the coarse mode is 0.85 (5.03 ± 10.90) ng m-3, while in EC-Earth and ERA-Interim the calculated values are around 1.74 (6.18 ± 7.29) ng m-3 and 1.04 (5.33 ± 7.30 ng m-3). Finally, the concentration of DFe in total suspended particles (tsp) is 1.93 (7.83 ± 19.03) ng m-3 in the observations, and it is 4.37 (12.51 ± 19.09) ng m-3 and 4.39 (11.63 ± 17.12) ng m-3 in EC-Earth and ERA-Interim, respectively. The correlation coefficients between the median values of the DFe cruise observations and the model results for EC-Earth and ERA-Interim are calculated as 0.49 (nMB = 38 %, nRMSE = 164 %) and 0.58 (nMB = 23 %, nRMSE = 145 %) for the accumulation aerosols, while for the coarse mode these values are 0.46 (nMB =-9 %, nRMSE = 193 %) and 0.59 (nMB =-25 %, nRMSE = 177 %) for EC-Earth and ERA-Interim simulations, respectively (Fig. S5). A summary of statistics for the evaluation of the simulated DFe concentrations for the EC-Earth and ERA-Interim simulations is also presented in Table A2.

The spatial distributions of the absolute differences between the simulated DFe concentrations in EC-Earth and the observations are also presented in Fig. 10d–e. The model shows a general overestimation of the observed DFe concentrations around the tropics (up to ∼ 10 ng m-3) but an underestimation at mid-to-high latitudes (up to ∼ 5 ng m-3). It is, however, unclear why the model is unable to capture the observed concentrations over such regions. Reasons could be the misrepresentation of coarse dust emissions, missing anthropogenic primary Fe emissions, weak secondary sources of the dissolved Fe (especially in the southern latitudes), or even systematic errors in the transport of coarse particles. Over the Pacific, EC-Earth better predicts the average DFe concentrations. For completeness, Fig. S5 also presents a comparison of the simulated and observed TFe concentrations, showing that the model better captures the TFe concentrations in the accumulation mode (Fig. S6g) than in coarse mode (Fig. S6h). Considering, however, that the TFe is mostly dominated by primary sources, the calculated differences to the observed concentrations (Fig. S6e–f) downwind continental sources should mainly depict errors in the emission parameterizations or a misrepresentation in the mineralogical composition of the larger Fe-containing soil particles.

The overestimation around the tropics and the northern latitudes is further illustrated by a comparison of the model predictions with observations as a function of latitude (binned at 2∘) (Fig. 10g–i). Although the dissolved Fe in the accumulation mode (Fig. 10g) is simulated well over the Southern Ocean, the model strongly overestimates the observed concentrations in the tropics, especially around the Equator, as well as in the northern extratropics. In contrast, for the coarse and the total suspended aerosols, an underestimation of the DFe aerosol observations in the southern latitudes (i.e., around 30–60∘ S) is clearly visible for both simulations (Fig. 10h and i), along with an overestimation around the tropics that is similar to that for the accumulation DFe aerosols. All in all, the differences between the two model configurations are relatively small (see Table A2).