Sensitivity of transatlantic dust transport to chemical aging and related atmospheric processes

We present a sensitivity study on transatlantic dust transport, a process which has many implications for the atmosphere, ocean and climate. We investigate the impact of key processes that control the dust outflow, i.e., the emission flux, convection schemes and the chemical aging of mineral dust by using the EMAC model following Abdelkader et al. (2015). To characterize the dust outflow over the Atlantic Ocean, we distinguish two geographic zones: (i) dust interactions within the ITCZ (DIZ) and (ii) the adjacent dust transport over the Atlantic Ocean (DTA). In the latter zone, the dust loading shows 5 a steep and linear gradient westward over the Atlantic Ocean, since particle sedimentation is the dominant removal process, whereas in the DIZ zone aerosol–cloud interactions and wet deposition / scavenging processes determines the extent of the dust outflow. Generally, the EMAC simulated dust compares well with CALIPSO observations, however, our reference model configuration tends to overestimate the dust extinction at lower elevation and underestimates it at higher elevation. The Aerosol Optical Depth (AOD) over the Caribbean responds to the dust emission flux, only when the emitted dust mass is significantly 10 increased over the source region in Africa by a factor of ten. These findings point to the dominate role of dust removal (especially wet deposition) in transatlantic dust transport. Experiments with different convection schemes indeed revealed that the transatlantic dust transport is more sensitive to the convection scheme than to the dust emission flux parameterization. To study the impact of dust chemical aging, we focus on a major dust-outflow in July 2009. We use the calcium cation as a proxy for the overall chemical reactive dust fraction and consider the uptake of major inorganic acids (i.e., H2SO4, HNO3, 15 HCl) and their anions, i.e., sulfate (SO2− 4 ), bi-sulfate (HSO − 4 ), nitrate (NO − 3 ) and chloride (Cl −)) on the surface of mineral particles. The subsequent neutralization reactions with the calcium cation forms various salt compounds that cause the uptake of water vapour from the atmosphere, i.e., by chemical aging of dust particles leading to an increase of 0.15 in AOD under subsaturated conditions (monthly mean, July 2009). As a result of the radiative feedback on surface winds, dust emissions regionally increased. On the other hand, the aged dust particles, compared to the "non-aged" case, are more efficiently removed 20 by both wet and dry deposition, due to the increased hygroscopicity and particle size (mainly due to water uptake). The enhanced removal of aged particles decreases the dust burden and lifetime, which indirectly reduces the dust AOD by 0.05

(monthly mean). Both processes can be significant (major dust-outflow, July 2009), but the net effect depends on the region and level of dust chemical aging.

Introduction
In the past several decades, transatlantic dust transport has gained tremendous attention because of many important impacts on Earth's climate, human health and ecosystems. North African dust transport over the Atlantic Ocean has emerged as a major 5 contributor to the soil nutrient input to many islands in the Caribbean, the Bahamas (Muhs et al., 2007), Bermuda (Muhs et al., 2012) and in the Amazon Basin (Bristow et al., 2010;Ben-Ami et al., 2012;Abouchami et al., 2013). Dust deposition influences the oceanic and terrestrial biogeochemistry by the transport of nutrients such as iron (Ussher et al., 2013;Baker et al., 2013Baker et al., , 2010Jickells et al., 2005) and phosphorus (Nenes et al., 2011) that dissolve into the ocean water. The emission, transport, and deposition processes of the North African dust are strongly influenced by meteorology causing strong seasonal, inter-annual 10 and decadal variability (Mahowald, 2007;Mahowald et al., 2010). Large fractions of the dust emissions are carried across the west coast of North Africa up to the Western Atlantic (Prospero et al., 2014) and significant correlations exist between the dust and climate variables, such as sea surface temperature, the North Atlantic Oscillation (NAO), and the Madden-Julian Oscillation (MJO) (Ginoux et al., 2004;Wong et al., 2008;Guo et al., 2013). In addition, the African dust in the Sahara airlayer region influences the rates of rainfall in the Inter-Topical Convergence Zone (ITCZ) (Huang et al., 2009(Huang et al., , 2010, and its 15 radiative impacts can shift and widen the ITCZ northward (Bangalath and Stenchikov, 2015).
African dust is transported in great quantities to the Caribbean basin throughout the year, although the strong seasonal cycle shows the maximum transport of dust in boreal summer and the minimum in winter (Prospero et al., 2014;Yu et al., 2015).
The seasonality is corroborated by satellite measurements of Aerosol Optical Depth (AOD), which show extensive plumes of high AOD in summer extending from the west coast of Africa to the Caribbean, the Gulf of Mexico, and to the southern United 20 States (Hsu et al., 2012;Yu et al., 2013;Chin et al., 2014;Kim et al., 2014;Groß et al., 2015). The satellite data also indicate that the dust transport to the Western Atlantic in winter and spring is comparable, but the dust is largely confined to the southern latitudes of Barbados with a plume axis crossing the coast of South America in the region of French Guiana and Surinam. In addition, satellite data indicate a decrease of 50% in AOD and a decrease of 0.1-0.2 in the dust-only optical depth during the transport (Kim et al., 2014). The ITCZ acts as an efficient removal mechanism (Prospero et al., 2014) and thus as a barrier to 25 the transport of dust to the southern Atlantic (Huang et al., 2009(Huang et al., , 2010Adams et al., 2012). To characterize the transatlantic dust transport, many studies have used satellite observations (Liu et al., 2008;Ben-Ami et al., 2009, 2010Adams et al., 2012;Ben-Ami et al., 2012;Ridley et al., 2013;Alizadeh-Choobari et al., 2014;Kim et al., 2014;Yu et al., 2015, among others).
However, the estimation of the satellite-based dust flux has large uncertainties, primarily because of ambiguity associated with the derived dust-only optical depth (Yu et al., 2009 and the dust mass extinction efficiency. Both parameters are used for calculating the dust mass loading (Kaufman, 2005).
One cause of uncertainty is the chemical aging of mineral dust. For instance, the condensation of inorganic acids, such as nitric acid (HNO 3 ), can alter the particle size due to changes in hygroscopicity of the dust particles (Metzger et al., 2006;Karydis et al., 2016). HNO 3 , which is an oxidation end product of combustion processes and lighting NO x , and therefore ubiquitous in the atmosphere, readily reacts with the calcium of the mineral dust surface. The neutralization product, calcium nitrate, additionally takes up ambient water vapour, which can change the particle (wet) radius. This process of water uptake can become significant, since it already starts at a relative humidity as low a 50% (the relative humidity of deliquescence (RHD) of Ca(NO 3 ) 2 is 48% at T=298 K). In strong contrast, dust coating by sulphuric acid (H 2 SO 4 ) does not lead to such hygroscopic 5 particles, since the RHD of CaSO 4 is close to 100% (at any T). Thus, especially the coating by nitrates can determine the hygroscopicity of mineral dust particles in case of a polluted atmosphere (Bauer et al., 2007;Sullivan et al., 2007;Li and Shao, 2009;Tobo et al., 2009Tobo et al., , 2010Li et al., 2013). The growth of the particles increases the scattering cross sections and therefore alters the AOD, it indirectly affects the cloud scavenging efficiency (Lance et al., 2013;Wu et al., 2013;Li et al., 2013), overall potentially increasing the wet and dry removal of the dust particles (Abdelkader et al., 2015). 10 Therefore, the dust cycle and the associated impacts are found to be challenging for global and regional modeling, because the complex dust processes have to be parameterized using a suite of simplifications (Astitha et al., 2010;Nowottnick et al., 2010;Huneeus et al., 2011;Ridley et al., 2013;Kim et al., 2014;Gläser et al., 2015). Although most sophisticated atmospheric models can reproduce the transatlantic dust transport plumes, the patterns differ in magnitude and seasonality. Generally, the models show better performance in summer than in winter for the transatlantic dust transport (Huneeus et al., 2011). It has been 15 observed that large uncertainties particularly exist between model simulations of the dust deposition (wet and dry) (Schulz et al., 2012). The atmospheric models that are applied in the AeroCom model intercomparison activity (http://aerocom.met.no/) show that the mean normalized bias of the AOD varies within a wide range from -0.44 to 0.27 (Huneeus et al., 2011), which is caused by large discrepancies in the dust-related processes (emission, horizontal and vertical distributions and the parameterization of chemical aging) that affect the dust transport from Northern Africa over the Atlantic ocean . This 20 indicates that in these models the dust removal is very efficient during the transatlantic transport (Kim et al., 2014) and that the development of the model requires comprehensive representation of the dust related processes. Though the incorporation of satellite products helps improving the model results, a deeper understanding of the key factors that determine the transport of the dust is also required. This study aims at examining the factors that can affect the transatlantic dust transport, i.e., the emission flux, convection schemes and the chemical aging of mineral dust, by using the EMAC model. 25

Model Description
We use the EMAC (ECHAM5/MESSy2 Earth System Model) following Abdelkader et al. (2015). The EMAC model describes the tropospheric and middle atmosphere processes and their interactions with land and oceans considering various submodels (Joeckel et al., 2010) -those used in this study are listed in Table 1. The mineral dust particles are emitted in two log-normal distribution modes (accumulation and coarse) with median diameters of 0.5 µm and 5.0 µm and a modal standard deviation 30 of 1.59 and 2.0 for the accumulation and coarse modes respectively (Abdelkader et al., 2015). The anthropogenic emissions are based on the EDGARv4.0 inventory (Pozzer et al., 2012) and includes the greenhouse gases, NO x , CO, non-methane volatile organic compounds (NMVOCs), NH 3 , SO 2 , black carbon (BC) and organic carbon (OC) from fossil fuel and biofuel use. The monthly large-scale biomass burning emissions of OC, BC and SO 2 , are based on GFED version 3 (Global Fire Emissions Database) (van der Werf et al., 2010). The emissions drive a comprehensive atmospheric chemistry mechanism (Sander et al., 2005), which calculates major inorganic acids (H 2 SO 4 , HNO 3 , HCl) online with meteorology. Organic acids are not considered in this model setup since their concentrations over Sahara during dust outflow are very low, though, many modeling studies reported the uptake of organic acids by dust particles (Metzger et al., 2006;Möhler et al., 2008;Liu et al., 5 2013; Li et al., 2013;Alexander et al., 2015;Wang et al., 2015).
The chemical aging of the dust depends on the condensation of inorganic acids and the associated uptake of water vapor. This increases the dust particle mass, particle size and the removal rates, which tends to decrease the lifetime of chemically aged dust. The condensation of acids in our model yields the anions sulfate (SO 2− 4 ), bi-sulfate (HSO − 4 ), nitrate (NO − 3 ), and chloride (Cl − ), whereas the condensation of ammonia (NH 3 ) yields a semi-volatile cation, ammonium (NH + 4 ), that reacts with the 10 inorganic anions in competition with the mineral cations Na + , Ca 2+ , K + , Mg 2+ (Metzger et al., 2006). However, in this study the cations are considered as reactivity proxy for natural aerosols, such as sea salt, biomass burning, or mineral dust, where we follow Abdelkader et al. (2015) and use a fixed percentage. These fractions have been derived from a comprehensive sensitivity study (which will be presented in a separate study) to achieve the best agreement of the cation and anion concentrations with various station observations for the period 2000-2012 (see Section 3). The anion-cation neutralization products (salt 15 compounds), simulated by the aerosol thermodynamic models, ISORROPIA-II (Fountoukis and Nenes, 2007) or EQSAM4clim (Metzger et al., 2016), can alter the hygroscopicity of the atmospheric dust particles, but the effect strongly depends on the atmospheric residence time, region and concentrations of acids. Generally, dust chemical aging changes the solubility, which controls the water uptake and in turn alters the aerosol size distribution (Metzger et al., 2006). The water uptake is a key parameter and important for aerosol-radiation feedback, aerosol in-cloud processing (nucleation scavenging), and below-cloud 20 (impaction) scavenging. The EMAC scavenging processes include detailed pH-dependent aqueous phase chemistry (Tost et al., 2006a) which is fully coupled with the aerosol and gas-phase chemistry, liquid cloud water and ice crystals. In addition to the aerosol hygroscopic growth and scavenging, the dust size distribution can change by coagulation, and smaller particles can grow into larger sizes for both the soluble and insoluble aerosol modes (Pringle et al., 2010), whereas aerosol hygroscopic growth is only allowed in the soluble modes (Abdelkader et al., 2015). Dry deposition and particle sedimentation can remove 25 all particles from the atmosphere depending on the particle size (Kerkweg et al., 2006a). Thus, the representation of the dust cycle in our EMAC setup couples the dust emissions, loading, and lifetime with the radiative forcing and model dynamics. As a result, changes in the dust loading feed back to the surface wind speed, soil moisture, cloud formation and precipitation, and in turn the dust emission flux. Overall, the level of air pollution controls the dust cycle because it determines the level of dust chemical aging by inorganic acids and water vapor. A Newtonian relaxation approach is used to nudge the model meteorology 30 in the free atmosphere (i.e., above the boundary layer) to achieve a realistic simulation of the surface wind speed and tracer transport (Abdelkader et al., 2015). Nudging significantly improves the surface dust mass concentration over the Caribbean compared with dust observations (Astitha et al., 2012). The model spectral resolution is T106 (≈ 110 km) and for the longterm simulations it is T42 (≈ 280 km). Both model resolutions use 31 vertical levels. Figure 1 summarizes the representation of the dust cycle and air-pollution-dust-chemical-aging-radiation feedbacks in our EMAC model setup.

Long-term evaluation
This study aims at examining the key factors that affect the transatlantic dust transport for a major dust-outflow event in July 2009 with a model resolution of T106, which is presented in Section 4. Before we focus on the sensitivity study, we present in this section the key findings of a comprehensive model evaluation, which was performed for the period 2000-2012 with a coarser resolution of T42. For the long-term evaluation, we use the following satellite and ground station AOD products: period. Both the dust burden and the precipitation rate peak during the summer season (JJA), where the dust plume is located relatively far north from the equator, in agreement with remote sensing observations (Prospero et al., 2014;Yu et al., 2015).
During the winter season (DJF), the dust burden and the precipitation rate show a minimum, whereas during the spring season 15 (MAM), the dust plume and the ITCZ are shifted southward. In winter and spring, the dust transport shifts southward to 0 • -10 • N and affects South America significantly, whereas during summer, the dust transport occurs predominantly at 10 • -20 • N, substantially affecting the Caribbean (Yu et al., 2015). During boreal winter the enhanced precipitation over the Northern part of South America results in higher and localized dust scavenging because the precipitation along the dust transport from the Western Africa into the Caribbean is at minimum. In contrast, during boreal summer, the dust spreads to a larger extent into 20 the ITCZ because of the stronger emissions (Prospero et al., 2014) while it is subject to enhanced dust scavenging. The strong southward gradient of the dust burden (≈ 100 mg m −2 deg −1 ) is collocated with precipitation in the western part of the Sahel and the ITCZ region. During the winter months, dust is primarily scavenged over Southeast America. As a result, the extent of the dust outflow is primarily controlled by precipitation in the ITCZ region. Figure S1 in the Supplement shows the dry and the wet removal of the dust particles. It shows that the dry removal dominates the northern part of the dust outflow region, whereas 25 the wet removal dominates the southern part.
To indicate the region where the dust interacts with the ITCZ, we introduce the dust-ITCZ (DIZ) zone which is shown in Figure 3 -the DIZ is marked by a blue line, and the AERONET station locations used to evaluate the simulated AOD are included. In the DIZ region, the transatlantic dust transport is controlled by dust-cloud interactions and the dust scavenging is most efficient. Accordingly, we refer to the region of the pre-dominant dry removal process (sedimentation), as the DTA zone. 30 Summarizing the long-term evaluation results, Figure 4 shows: (i) the transatlantic dust transport region with skill score (Taylor, 2001) at each station (see Appendix A for the evaluation metrics), (ii) the time series of the six selected stations that provide long-term data with three stations each in the Caribbean (left) and around West Africa (right), and (iii) the corresponding scatter plots of both sides of the Atlantic Ocean include the observations from all stations. Additionally, the correlation coefficients (R) ( Table 2) are lower than the SS1, because R is more phase sensitive than the 10 SS1 (i.e., more sensitive to time lags between simulated and observed AOD). The higher R value for West Africa (0.61) compared with Caribbean (0.41) mainly results from the overall higher contribution of dust AOD to the total AOD. Typically, the Caribbean is strongly influenced by the uncertainty associated with long-range transport and the dust chemical aging, with potential failures causing a time shift of dust peaks during the transport. These differences are, however, best revealed by station peaks, especially at Dakar which is at the edge of the DIZ zone. Over the Caribbean, the model generally underestimates the AOD during the dust outflow periods, e.g., seen at the AERONET station La Parguera. This underestimation could be related to the representation of dust emissions and related processes in the source region of the West Africa (Huneeus et al., 2011;20 Shao et al., 2011;Cuevas et al., 2015), by overestimated removal during transport (Schulz et al., 2012;Prospero et al., 2014), or due to low-biased dust transport from the boundary layer into the free atmosphere (Khan et al., 2015). In addition, the underestimation of AOD could be also due to the missing fraction of giant mode particles (larger than 10 µm), which may contribute to an underestimation of AOD near the dust source region. However, giant particles are not transported far over long distances and hence not really relevant for the long-range transport and our sensitivity study on the emission flux and removal 25 mechanisms.

Sensitivity studies
To study the key factors that may affect the transatlantic dust transport, we focus in this section on a major dust-outflow event that occurred in July 2009. We study the impact of various key factors with a relatively high model resolution (T106). The dust outflow is, in terms of AOD ,close to most observations, as indicated by monthly means highlighted by the red bar in Fig. 4. 30 However, near the source region at Dakar and Capo Verde, the AOD observations are underestimated for this month. During this period a major outflow event occurred, and therefore it seems suitable to test various model parameters, i.e., (a) the dust emission flux, (b) the convection parameterization, and (c) the level of dust chemical aging. Figure 5 shows the dust burden and the total mean precipitation for July 2009 from the reference EMAC simulation, which includes the dust cycle and chemical aging as shown in Fig. 1. The simulated dust surface concentration reaches on average up to 600 µ g m −3 at Dakar, indicating that the model captures in principle the strong outflow event. Generally, two strong precipitation areas are visible with one peak centered at 15 • W with a monthly average of 20 mm day −1 , i.e., one at the coast of West Africa and the other peak area is located in the Caribbean at 50 • W. with a monthly average of 25 mm day −1 . These 5 precipitation maxima influence the dust loading. During transatlantic dust transport, the ITCZ represents a strong barrier for the dust outflow and therefore controls the meridional extent of the dust plume (Yu et al., 2015). The ITCZ acts as a major sink that depends on the amount of precipitation (Prospero et al., 2014;Schlosser et al., 2014) and the removal might be enhanced depending on the dust chemical aging (Abdelkader et al., 2015). Clearly, the precipitation within the ITCZ coincides with the steep gradient of the dust burden in the meridional direction over the Western Africa. Along the zonal extent of the dust plume, 10 the collocation of the dust plume and precipitation corroborates that the meridional extent of the dust is primarily controlled by the location of the ITCZ. Fig. S1 in the Supplement summarizes the monthly average dust removal during July 2009. Table 1a and table 1b in the supplement additionally show some evaluation metrics for the AOD of the sensitivity study over the West African and Caribbean stations.
Typically, African dust outflow reaches the Caribbean ≈ 5 days later (Gläser et al., 2015) and the surface dust concentration 15 is significantly lower at the Caribbean side compared to Western Africa. Figure 6 shows the time series of the size-resolved surface dust concentrations. Two main dust outflows on the 2 nd and 12 th July are simulated at the Capo Verde station, indicated by dust concentrations higher than or close to 300 µg m −3 (equivalent particle cutoff diameter of 5 µm) and another weaker dust outflow is simulated on 24 th July, indicated by a lower concentration peak around 100 µg m −3 . The former two dust outflows are seen at Dakar with twice the concentration (up to 600 µg m −3 ) at slightly different time periods due to different 20 transport. Eventually, the dust outflow reaches the Caribbean with a significant lower concentration of around 60 µg m −3 at the earth surface.
Despite chemical aging, the model simulates a majority of the dust particles in the insoluble coarse (ci) mode, which indicates that the dust particle concentration is high and/or the inorganic acids concentration is relatively seen too low for complete chemical aging. This is especially valid for strong dust outflows, such as studied here. On the other hand, the fraction of the 25 aged dust, i.e., the ratio of the coarse mode soluble to insoluble particles (cs/ci), is somewhat higher in the Caribbean because of the continuous chemical aging during long-range transport. The aged dust fraction over West Africa is about 10% of the total dust mass and twice that at the Caribbean sites. The same is true for the dust in the accumulation modes (ai and as), but the mass concentrations are an order of magnitude lower compared to the coarse mode concentrations and therefore they are not discernable at the linear scale. At higher elevations, this fraction can be different because of different dust and precursor 30 gas concentrations.
To investigate the vertical distribution, the simulated dust extinction is compared with the dust subtype classification of the CALIPSO retrievals. an elevation up to 7 km. The dust burden is very low (as indicated by the area plot) south of 10 • N, which coincides with a very low AOD observed by CALIPSO. Both EMAC and CALIPSO show that the dust plume is limited to the area between 14 • to 22 • N and the top of the dust layer is lowered to 5 km over the middle of the Atlantic. This is primarily a result of 5 the prevailing deposition (gravitational settling + wet removal), which is further discussed in the following sections. Once the dust reaches the Caribbean, the plume spreads over a considerably larger area, which extends from 5 • to 28 • N as a result of changes in meteorological conditions. The dust plume eventually reaches the Caribbean with a top layer height of ≈ 5 km. In  Table 3. The total dust mass, emitted during July 2009 within the region between 20 • W to 10 • E and 15 • N to 20 30 • N, is 0.6133 kg m −2 for the reference case.
The first test case (B1E1) represents a redistribution of emission bins between the coarse and accumulation modes so that dust particles are shifted from the coarse to the accumulation mode while conserving the total dust mass. In this case, a larger amount of dust in the accumulation mode is transported over extended distances compared with the reference case "EMAC".
"EMAC" considers the same total dust mass with a larger fraction in the coarse mode. Additional sensitivity runs, B1E2 to 25 B1E7, change the total dust emission flux by increasing the emission flux according to different factors shown in Table 3. The horizontal dust emission flux is described by Eq. 1 (Marticorena and Bergametti, 1995;Astitha et al., 2012) With the tuning parameter c = 1 representing the reference case "EMAC" following (Darmenova et al., 2009;Astitha et al., 2012), g is the gravitational acceleration, ρ air the air density, u * the friction velocity, u * t the threshold friction velocity. For 30 case B1E8, the horizontal mass flux is increased by a factor of 2.6 (parameter c in Eq. 1). The cases highlighted in Table 3 are shown in Fig. 8, whereas the other cases are shown in the Supplement (Fig. S3).
Due to the different dry and wet deposition characteristics of the accumulation and coarse mode particles, significant differences are expected. Figure 8 shows that the AOD time series at the selected AERONET stations are rather insensitive to the emission flux modifications except for case B1E3 (and B1E4, which is shown in the Supplement). This is valid for both sides of the Atlantic, where the AOD at the Caribbean stations seems even less sensitive than the AOD for the West African sites.
Only for the cases where the coarse mass flux is significantly increased (factor of 5.3), the AOD shows a higher sensitivity.

5
The large increase in the coarse mode mass for case B1E3 results in a significant increase in AOD (exceeding 2.0) on both sides of the Atlantic Ocean. Case B1E8 (modification of the horizontal mass flux) shows better agreement with the AERONET observations at both sides of the Atlantic Ocean despite the very high AOD values obtained on 21 July at Saada station. The model captures the AOD during the two dust outflow events (2 July and 12 July) at Capo Verde as well as the first dust outflow at Saada on 4 July. For the Caribbean sites, case B1E8 shows the best agreement with AERONET for the three stations.

10
The sensitivity simulations show that the accumulation mode fraction of the dust contributes much less to the AOD on both sides of the Atlantic Ocean because even an increase by a factor of 5.3 in the dust emission flux is not sufficient to match the observations. Instead, such an increase (by a factor of 5.3) in the emitted dust mass flux results-regionally and globally-in an unreasonable dust budget shown by Astitha et al. (2012). On the other hand, this sensitivity study shows that the AOD is more sensitive to the dust mass in the coarse mode and that the AOD over the Caribbean is much less sensitive to the 15 total dust emission flux. Clearly, the model sensitivity is higher for the West African sites because these AOD results are more directly controlled by the Saharan dust outbreaks. To match the elevation at which this outflow occurs is equally important. The comparison with the CALIPSO observations ( Fig. 7) reveals that EMAC overestimates the dust extinction at lower elevations, whereas the values at higher elevations are underestimated. This finding points to the strong contribution of dust removal during transatlantic dust transport, and is largely controlled by the convection scheme.

Convection schemes
The scavenging of dust particles by precipitation is another key factor that controls the transatlantic dust transport (Kim et al., 2014). In order to study the impact of the convection and the associated precipitation during the dust outflow, different convection schemes (implemented in EMAC by Tost et al. (2006b)) are compared. The default scheme (TIEDTKE convection with NORDENG closure) provides realistic water vapor distributions on the global scale, which is crucial for radiative transfer 25 processes and atmospheric chemistry (Tost et al., 2006bRybka and Tost, 2013). However, the radiative effect of aerosols has not been considered in these studies. Table 3 includes the sensitivity tests by using several convective schemes available in the EMAC model. The principal cases are shown in Fig. 9, whereas the other cases are shown in the Supplement (Fig. S4). Figure 9 depicts the AOD time series for the stations shown in Fig. 3 and shows a larger sensitivity to the convection compared to the emission flux parameterizations (Sec. 4.1). In particular, the AOD is more sensitively influenced over the 30 West African than over the Caribbean sites, which is primarily a result of the decreasing dust burden due to the removal of the dust during transport (Fig. 6). Generally, the AOD is underestimated at all stations, except for Saada, in the reference simulation (EMAC). During the period 20-25 July 2009, this significantly improves in the sensitivity simulations (B1T3 and B1T5). However, the model also simulates a dust outflow event that is not observed by the AERONET stations. Overall, over the Caribbean, case B1T5 (ECMWF operational convection scheme) yields the best results for all dust outflow events. The main differences between the schemes appear in the tropical region, while the maximum difference is obtained during boreal summer. For these conditions (location + time), the EMAC reference setup is associated with relatively large discrepancy in the precipitation amount (Tost et al., 2006b). As a result, the scavenging of aerosols, including dust particles, is overestimated due to the high precipitation rates. Consequently, this over-removal of the dust results in an underestimation of the AOD over 5 the Caribbean. First, the dust burden shows a very steep gradient westward over the Atlantic Ocean. This is mainly a result of dust removal by deposition (sedimentation and scavenging mechanisms) during long-range transport. Over the Atlantic (within DTA), this gradient is linear in the logarithmic scale, whereas the gradient is nonlinear over the Western and Eastern Atlantic (especially within DIZ). The dust burden over West Africa (east of 10 • W) is about 1000 µg m −3 but declines to 50 µg m −3 over the 30 Caribbean. The different parameterization schemes show more than a factor of 2 difference between the dust burden over Western Africa, and about a factor of 3 over the Caribbean. This is primarily a result of different precipitation rates and the associated differences in dust removal. The two precipitation peaks (over Western Africa and the Caribbean), shown in Fig. 5, are also seen in Fig. 11. They are, however, weaker because the averaging is performed over a wider area (dust plume) that is not associated with precipitation. The higher precipitation rate over the western and eastern parts of the Atlantic results in 35 enhanced dust scavenging. Over the Atlantic, the precipitation is lower and therefore the removal by sedimentation is stronger during July 2009 (≈2 g m −2 compared to ≈0.2 g m −2 , respectively). The elevated precipitation over the Caribbean causes maximum wet deposition. As a result, the dust burden is an order of magnitude lower over the Caribbean compared to West Africa. In addition, there is a clear anti-correlation between the dust burden and the precipitation amount over both sides of the Atlantic. The comparison of precipitation with TRMM observations reveals that the EMAC model gives more realistic results 5 over West Africa compared with the Caribbean for all convection schemes.
Second, the ADP (Fig. 11) illustrates the effect of convection schemes on the transatlantic dust transport. Over West Africa, the dust is already aged with ADP values between 0.2 and 0.4, whereas over the Caribbean the ADP values are with 0.3 and 0.5 only slightly higher. The lower ADP values over West Africa can be attributed to the higher dust loadings, which requires a much larger amount of condensable material to becomes fully aged. Over the Caribbean, the dust loading is considerably Thus, the convection sensitivity analysis points to a too strong removal mechanism of the mineral dust particles along transatlantic transport, when the default convection scheme is used in EMAC. In addition, the level of dust chemical aging 20 seems to control the efficiency of dust scavenging. Higher levels of aged dust, and higher precipitation amounts, significantly decrease the dust burden and thus the AOD over the Caribbean. This further suggests that modeling the transatlantic dust transports requires improved convection parameterization (i.e., more realistic precipitation rates), and in parallel a realistic representation of dust chemical aging.

Dust chemical aging 25
To further investigate the impact of the dust chemical aging on the transatlantic dust transport, this process was excluded for an additional sensitivity study. The level of dust chemical aging depends on the availability of condensable acids (see Sec. 2). For the "No Aging" case, the condensation of acids on insoluble dust particles is excluded, which suppresses water uptake by dust particles. Figure 12 shows the AOD time series at the AERONET stations on both sides of the Atlantic for the two cases, i.e., Aging and No Aging. Generally, the Aging case systematically shows a higher AOD compared to the No Aging case, which 30 emphasizes the importance of this process and the associated water uptake in agreement with the results of Abdelkader et al.
(2015). However, the dust chemical aging has a stronger impact on the AOD over West Africa, especially at the Capo Verde and Dakar stations during the two dust outbreaks discussed above. The Aging case shows about 0.2 higher AOD compared with the No Aging case as a result of the larger particle size and the associated water uptake. This increases the scattering cross section and thus the AOD. Over the Caribbean, the dust chemical aging shows a smaller impact on the AOD; the Aging case shows only about 0.05 higher AOD because of the lower contribution of the dust to the overall AOD values (which includes the contribution of other aerosol species, sea salt, etc., for instance). During the high dust outbreaks, the concentration of the soluble compounds required to coat such a large amount of dust is not available according to the EMAC model. The aged dust particles are removed more efficiently during transport and relatively more uncoated dust particles reach the Caribbean. As a 5 result, the dust chemical aging has a limited effect on the AOD over the Caribbean AERONET stations.  The difference between the two simulations decreases during transport, which is supported by the differences in the dust-only AOD. In contrast, the difference in the total AOD shows lower values over the dust source region compared with the Aging case, which indicates a significant contribution of the dust chemical aging to the total AOD.
Interestingly, the negative feedback between the AOD and the radiation scheme results in higher dust emission over the region from 10 • E to 0 • and thus causes a higher dust burden. The average dust emission during July 2009 over the region from "No Aging" case has higher dust optical depth as a result of the lower dust removal as compared with the "Aging" case. The difference is at a minimum within a region between 18 • N to 22 • N. However, the total AOD shows that the "No Aging" case leads to a lower AOD, which is significant over Western Africa and less pronounced over the Caribbean sites. Note that the AOD, as compared with AERONET stations, shown in Fig. 12 does not resolve this large difference because the AERONET stations are all located in the DTA region where the differences are obviously lower. 30 The substantial higher AOD for the "Aging" case (0.3 on monthly mean) primarily results from the dust chemical aging because of the associated water uptake. Figure 14 shows the monthly averaged burden for lumped gas-phase acids (sum of HCl+HNO 3 +H 2 SO 4 ) and the difference between both simulations. The figure also shows the corresponding lumped inorganic aerosol mass (sum of SO 2− 4 + HSO − 4 + NO − 3 + NH + 4 + Cl − + Na + + Ca 2+ + K + + Mg 2+ ) and the aerosol associated water mass. For the Aging case, the burden of acids is very low over the dust source region because of the uptake by dust particles -an important effect which has been also recently studied with the EMAC model by Karydis et al. (2016) for the nitric acid uptake (also included here). Consequently, the aerosol burden is higher over the dust source region and over the outflow region, because of the additional neutralization of the calcium ions by anions and the associated absorption of water vapor by the resulting calcium salts. As a result, the aerosol-associated water increases by more than 255 mg m −2 for the Aged case. The effect of dust chemical aging is a result of gas-aerosol partitioning that clearly affects the AOD. It is best observed in the 5 differences (right column of Figure 14), which reveal that the impact of dust chemical aging can be significant, but mainly due to the associated uptake of aerosol water. We refer to this effect as the "direct effect of dust chemical aging." In addition, we refer to the higher removal of aged dust (by both sedimentation and scavenging), and the consequently shorter dust lifetime, as the "indirect effect of dust chemical aging" -both effects are introduced in this study.
To obtain improved statistics for the effect of dust chemical aging, the same analysis (Aging versus No Aging) was applied to the entire evaluation period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) at lower T42 model resolution. Figure 15 shows the long-term meridional dust burden mean and the model precipitation for TRMM observations over the DTA and DIR zones (as discussed above). The No Aging case consistently shows higher dust burdens in the DIR zone as a result of more efficient scavenging for the Aging case.
Even for this long-term average, the dust burden is three times higher for the "No Aging" case than the "Aging" case over the Caribbean sites. However, the impact of scavenging in the Aging case is stronger in the region between 10 • W and 20 • W, 15 which corresponds with the precipitation peak in the West Africa region.

Conclusions
Tansatlantic dust transport is a major large-scale atmospheric phenomenon. Although the EMAC model mostly reproduces the dust pattern during the transatlantic dust transport, the dust loadings and AOD can deviate in magnitude and seasonality from observations. To examine the controlling processes, the dust outflow region has been divided into two subregions: (1) 20 the dust-ITCZ (DIZ) zone and (2) the adjacent dust transport over the Atlantic Ocean (DTA) zone. In the former, the dust is removed primarily by scavenging, whereas in the latter region sedimentation is predominant. Considering the two subregions allows the distinction of factors that affect the transatlantic dust transport.
Several sensitivity studies were conducted using the EMAC model following Abdelkader et al. (2015) -with a comprehensive setup which includes a fully coupled online dust emission scheme and an explicit chemical aging of the atmospheric dust 25 particles. First, the simulated AOD is sensitive to the emission flux parameterization, and even more to the choice of the convection scheme. The dust emission flux affects the AOD over West Africa more strongly compared to the Caribbean sites. On the other hand, the dust burden shows a very steep gradient westward over the Atlantic Ocean. This is mainly a result of dust removal by deposition (sedimentation and scavenging) during long-range transport. Over the Atlantic (within DTA), this gradient is linear in the logarithmic scale, whereas the gradient is nonlinear over the Western and Eastern Atlantic (especially within 30 DIZ). The dust burden over West Africa (east of 10 • W) is about 1000 µg m −3 but declines to 50 µg m −3 over the Caribbean.
The different convection parameterization schemes show more than a factor of 2 difference in dust burden over West Africa, and about a factor of 3 over the Caribbean. This is primarily a result of different precipitation rates and the associated differ-ences in dust removal. Overall, the dust outflow into the Caribbean is best represented by the ECMWF convection scheme, as a result of more realistic representation of precipitation within the ITCZ (compared to other schemes available in EMAC and relative to TRMM observations). The more realistic precipitation subsequently improves the dust removal (compared to the reference EMAC simulations) and subsequently the AOD on both sides of the Atlantic Ocean significantly within the DIZ zone, a region which is largely controlled by wet removal processes. Considering the dust chemical aging amplifies this effect.

5
To study the impact of dust chemical aging, we use the calcium cation as a proxy for the overall chemical reactive dust fraction and consider the uptake of major inorganic acids (i.e., H 2 SO 4 , HNO 3 , HCl) and their anions, i.e., sulfate (SO 2− 4 ), bi-sulfate (HSO − 4 ), nitrate (NO − 3 ), and chloride (Cl − )) on the surface of mineral particles. The subsequent neutralization reactions with the calcium cation forms various salt compounds that causes the uptake of water vapour from the atmosphere, which leads to the chemical aging of dust particles. Dust chemical aging changes the particle sizes because of the additional 10 amount of condensed inorganic acids and the associated uptake of water vapor by the neutralization products (salts). Therefore, the aged dust particles are larger and scatter light more efficiently, whereas they are more rapidly removed by dry and wet removal processes. To analyze these effects, we performed Aging and No Aging simulations, for which we distinguish between the direct and indirect effect of dust chemical aging on AOD.
In our senitivity simulations, the dust chemical aging shows the largest impact on the AOD over West Africa and on the dust 15 burden in the ITCZ. The larger impact on the AOD results from the increase in the aerosol burden (more than 120 mg m −2 ) due to the uptake of acids and associated water by the originally insoluble dust particles. This directly increases the AOD by 0.15 (monthly average). As a result of the radiative feedback on the atmospheric dynamics and circulation, the dust emission regionally increases. On the other hand, the aged dust particles are more efficiently removed in our EMAC reference setup compared with the non-aged dust particles case. The enhanced removal of aged particles decreases the dust burden and lifetime, 20 indirectly affecting the AOD. Both processes are significant and the net effect depends on the region and the level of dust chemical aging, which is controlled by the strength of the dust outflow and the collocated air-pollution levels. In order to improve the dust cycle in climate models, we recommend an explicit treatment of dust chemical aging, at least by considering the calcium cation as a proxy for the overall chemical reactivity of the mineral dust particles.
σ -Standard deviation of the model (σ m ) and the observation (σ o ) for variable (X i ) with average of (X) with N the number of observations:

30
-R -Correlation coefficient between the model (m) and the observations (o): r geometric mean of the model (r m ) and the observations (r o ).

Caribbean
West Africa Figure 1: Time series dust aerosol modes and AOD for di↵erent convection schemes  (Table 3).

Caribbean
West Africa Figure 1: Time series dust aerosol modes and AOD for di↵erent convection schemes  surface dust concentration, and dust burden for different convection schemes (2nd-4th row) highlighted in Table 3. The model precipitation and cloud cover agrees for our EMAC set-up best with TRMM and MODIS observations with the TIEDTKE (B1T3) and ECMWF (B1T5) convection schemes.