Strong Sensitivity of Aerosol Concentrations to Convective Wet Scavenging Parameterizations in a Global Model

This study examines the influences of assumptions in convective cloud parameterizations on global climate model simulations of aerosol concentrations and wet deposition. Two limiting cases regarding the effects of aerosol entrainment are considered. In the first case, aerosols entering convective clouds at their bases are the only aerosols 5 that are scavenged into cloud droplets, and are susceptible to removal by convective precipitation formation. In the second case, aerosols that are entrained into the cloud above the cloud base layer can activate, can collide with existing cloud droplets and ice crystals, and can subsequently be removed by precipitation formation. To facilitate this study, an explicit representation of the uptake of aerosol mass and number into 10 convective cloud droplets and ice crystals by the processes of activation, collisions, freezing and evaporation is introduced into the ECHAM5-HAM model. This development replaces the prescribed aerosol cloud-droplet-borne/ice-crystal-borne fractions of the standard model. The limiting case that allows aerosols entrained above cloud base to become cloud-droplet-borne and ice-crystal-borne reduces the annual and 15 global mean aerosol burdens by 30% relative to the other limiting case, and yields the closest agreement with observed global aerosol optical depth, and black carbon vertical profiles. Upper tropospheric black carbon concentrations differ by one order of magnitude between these limiting cases. Predicted convective cloud droplet number concentrations are doubled in the tropical middle troposphere when aerosols entrained 20 above cloud base are allowed to activate. These results show that aerosol predictions are strongly sensitive to the assumptions made regarding the uptake into cloud droplets of aerosols entrained above convective cloud bases. Additionally, wet deposition attributed to scavenging in convective clouds at temperatures where both the liquid and ice phase could co-exist is reduced by two to five-fold with the calculated 25 as opposed prescribed cloud-droplet-borne/ice-crystal-borne aerosol fractions of the standard model. The annual and global mean fractional contribution of convective wet deposition to total deposition is reduced from 15-30% to 5-20% for the prescribed and

global mean aerosol burdens by 30% relative to the other limiting case, and yields the closest agreement with observed global aerosol optical depth, and black carbon vertical profiles. Upper tropospheric black carbon concentrations differ by one order of magnitude between these limiting cases. Predicted convective cloud droplet number concentrations are doubled in the tropical middle troposphere when aerosols entrained 20 above cloud base are allowed to activate. These results show that aerosol predictions are strongly sensitive to the assumptions made regarding the uptake into cloud droplets of aerosols entrained above convective cloud bases. Additionally, wet deposition attributed to scavenging in convective clouds at temperatures where both the liquid and ice phase could co-exist is reduced by two to five-fold with the calculated 25 as opposed prescribed cloud-droplet-borne/ice-crystal-borne aerosol fractions of the standard model. The annual and global mean fractional contribution of convective wet deposition to total deposition is reduced from 15-30% to 5-20% for the prescribed and calculated cloud-droplet-borne aerosol fractions, respectively. We find that aerosol predictions are strongly sensitive to the assumptions regarding the prescribed versus calculated cloud-droplet-borne and ice-crystal-borne aerosol fractions in mixed liquid/ice phase convective clouds.

5
Aerosols play an important role in the climate system by influencing the Earth's radiation budget, directly by scattering and absorbing radiation, and indirectly by modifying cloud properties (Twomey, 1991;Charlson et al., 1992). Aerosols also have important impacts on global air quality (van Donkelaar et al., 2010), and human health (Dockery et al., 1993). As a result, the prediction of three-dimensional aerosol distributions is im-10 portant in both global climate, and air quality models. These distributions are strongly influenced by convective transport and wet scavenging in convective clouds. However, the representation of convective processes remains a major uncertainty for aerosol prediction in global models (Randall et al., 2003;Lohmann, 2008;Tost et al., 2010).
The parameterization of convective clouds in global models is a subject of ongo- 15 ing research efforts (Nober et al., 2003;Menon and Rotstayn, 2006;Lohmann, 2008). However, the aerosol-cloud interactions involving convective clouds are complex and difficult to capture in global models (Morales et al., 2011). Aerosols influence convective clouds since they act as cloud condensation and ice nuclei, and also by the semi-direct effect since they absorb radiation, which produces local heating that con- 20 tributes to cloud dissipation (Hansen et al., 1997;Ackerman et al., 2000). Conversely, convective clouds also influence three-dimensional aerosol distributions by processes such as aerosol wet scavenging and cloud processing (Engström et al., 2008). Further evidence of the uncertainty related to convective processes in global models is the wide disparity amongst these models in terms of the prediction of the contri- 25 bution of convective clouds to aerosol wet deposition. Textor et al. (2006) found that the predicted contribution of convective clouds to global and annual mean aerosol wet deposition ranged between 10 and 90%. Thus, there is no clear consensus on how greatly convective clouds contribute to aerosol removal from the atmosphere. The focus of this study is to examine how the assumptions made in convective cloud schemes can influence predicted aerosol concentrations and wet deposition.
Wet scavenging of aerosols in global models is commonly treated with prescribed 5 scavenging fractions. Global models often assume fixed values to represent the fraction of aerosols that are cloud-droplet-borne and ice-crystal-borne, and susceptible to removal by convective precipitation formation (Liu et al., 2001;Stier et al., 2005;Donner et al., 2011;Fang et al., 2011). The fraction of aerosol mass that is cloud-dropletborne is typically assumed to be near unity for accumulation and coarse mode aerosols 10 in warm phase convective clouds. However, for aerosol number in the nucleation and Aitken modes, and for both aerosol mass and number in mixed and ice phase clouds, these assumptions about the cloud-droplet-borne and ice-crystal-borne fractions vary considerably between models. Additionally, since clouds often occur at a scale smaller than the typical grid-box size of a global model, the precipitation fraction of the grid 15 box is also parameterized, often as a function of an assumed or parameterized updraft velocity (Liu et al., 2001;Stier et al., 2005). One goal of this study is to explore a more physical link between convective cloud microphysics and aerosol wet scavenging in a global model, and examine the influence of certain convective cloud assumptions on aerosol concentrations. 20 A key uncertainty related to convective cloud schemes in global models is the treatment of the influence of entrainment and detrainment processes on convective cloud droplet number concentration. The effects related to entrainment have been examined recently (Barahona and Nenes, 2007). Morales et al. (2011) developed an entraining droplet activation parameterization, and found that cloud droplet number concentration 25 in non-precipitating shallow cumulus clouds was over-predicted by 45% with an adiabatic parameterization that neglected entrainment effects. Considering both the liquid and ice phase, and deeper convective clouds, Fridland et al. (2004) found in a modeling study that allowing aerosols to entrain above cloud base, and act as cloud condensa-tion and ice nuclei, could enhance the number of ice crystals/cloud droplets in upper cloud regions by about an order of magnitude. Recent work has shown that observed ice crystal concentrations can be articificially high as a result of ice crystal shattering on aircraft measurement probes (Korolev et al., 2011). Neverthless, from a modeling perspective Fridland et al. (2004) did show that assumptions about the activation of 5 aerosols entrained above the cloud base can significantly influence the predicted number of cloud droplets/ice crystals in deep convective clouds. There remains uncertainty about how greatly deep convective clouds can entrain air above cloud base, and still maintain momentum. However, work by Romps and Kuang (2010) suggests that deep convective clouds can be highly diluted and still maintain momentum sufficient to reach 10 the tropopause. Despite these gaps in our knowledge, the representation of entrainment and detrainment rates is fundamental to convective parameterizations (Tiedtke, 1989).
The influence of entrainment on convective cloud droplet number concentration is often treated with empirical corrections to droplet activation schemes such as the Lin and 15 Leaitch (1997) scheme used by Lohmann (2008). One issue with these fixed empirical corrections is that the supersaturation experienced by entrained air parcels may not always evolve in the same manner for all convective clouds depending on the interplay of several factors including the following. 1) How completely does the entrained air parcel mix with the existing cloud? In the case of negligible mixing, there is low droplet surface 20 area in the entrained parcel, and supersaturations could be high enough to activate the entrained aerosols if the entrained parcel accelerates sufficiently. On the other hand if the entrained parcel completely mixes with the existing cloud, the high droplet surface area prohibits further increases in supersaturation such that the entrained aerosols can not activate. The supersaturation in the updraft may actually reduce if the entrained air 25 is drier, leading to droplet evaporation and a reduction of the already existing cloud droplet number. 2) How has rainout influenced the droplet population? In the case of rainout, the droplet surface area is lowered, and an entrained parcel could experience supersaturations required for activation if the parcel accelerates sufficiently. 3) How has detrainment influenced the cloud droplet number? In the case of a reduction in droplet surface area by detrainment, an entrained aerosol parcel could experience the required supersaturations for aerosol activation. 4) How much momentum dilution occurs as a result of entrainment, which limits the generation of supersaturation? 5) How much is the supersaturation in the updraft reduced as a result of the entrained 5 air being drier than the updraft? Thus, there are complex and potentially opposing effects related to the influence of entrainment and detrainment on the number of cloud droplets/ice crystals in convective clouds.
In this study, we compare a convective cloud droplet number concentration (CDNC) parameterization that allows adiabatic activation of aerosols at cloud base only, with 10 a revised CDNC parameterization that allows activation of aerosols entrained above cloud base. We use the ECHAM5-HAM model (Roeckner et al., 2003) since the model couples a detailed aerosol microphysics scheme (Stier et al., 2005) with a two-moment convective cloud microphysics scheme (Lohmann, 2008). To faciliate this study, we introduce an explicit representation of the cloud-droplet-borne/ice-crystal-borne aerosol 15 mass and number, which is based on the convective cloud microphysics of (Lohmann, 2008). We consider two limiting cases for aerosol wet scavenging in convective clouds. The first limiting case allows only those aerosols entering at the cloud base to become cloud-droplet-borne/ice-crystal-borne and susceptible to removal by convective precipitation formation. The second limiting case allows aerosols to entrain and activate 20 above the cloud base, and allows CDNC detrainment at all levels above the cloud base. We also compare to the standard ECHAM5-HAM model, which does not explicitly treat the cloud-droplet-borne/ice-crystal-borne aerosol fraction as a function of the cloud microphysics, but rather uses prescribed fractions. We examine the sensitivity of aerosol concentrations, burdens and wet deposition between these simulations. The following 25 section gives a model description. Results are summarized in Section 3. Section 3.1 examines the effects of these convective cloud schemes on aerosol concentrations. Section 3.2 considers the sensitivity of aerosol wet deposition to the convective cloud assumptions. Section 3.3 compares our simulated results with observations of global 6 aerosol optical depth, wet deposition, and aerosol vertical profiles.

Model description and development
The ECHAM5 model is a fifth generation atmospheric general circulation model (GCM) developed at the Max-Planck Institute for Meteorology (Roeckner et al., 2003), and evolved from the model of the European Centre for Medium Range Weather Forecast-5 ing (ECMWF). The model solves prognostic equations for vorticity, divergence, temperature and surface pressure using spheric harmonics with triangular truncation. Water vapor, cloud liquid water and ice are transported using a semi-Lagrangian scheme (Lin and Rood, 1996). Prognostic equations for cloud water and ice follow the two-moment cloud microphysics scheme of Lohmann et al. (2007). The model includes the cirrus 10 scheme of Lohmann and Kärcher (2002). Convective clouds, and convective transport are based on the mass-flux scheme of Tiedtke (1989) with modifications following Nordeng (1994). For this study, we have implemented the two-moment convective cloud microphysics scheme of Lohmann (2008). The solar radiation scheme has 6 spectral bands (Cagnazzo et al., 2007) and the infrared has 16 spectral bands (Mlawer et al., 15 1997;Morcrette et al., 1998).
Additionally, for this study, the GCM is coupled to the Hamburg Aerosol Model (HAM), which is described in detail in Stier et al. (2005). The aerosols are represented by seven log-normal modes, 4 soluble/internally mixed modes (nucleation (NS), Aitken (KS), accumulation (AS), and coarse (CS)) and 3 insoluble modes (Aitken (KI), accumulation 20 (AI),and coarse (CI)). The simulated aerosol species are sulfate, black carbon, particulate organic matter, sea salt and dust. The count median radius for each mode is calculated from the aerosol mass and number distributions in each mode. Aerosol mass and number are transferred between the modes by the processes of sulfuric acid condensation, and aerosol coagulation. All results presented in this study are from five 25 year free-running simulations, following a three months spin-up period, using climatological sea surface temperatures and sea ice extent. Aerosol emissions are taken from the AEROCOM database and are representative for the year 2000 (Dentener et al., 2006b). The aerosol emissions and the removal processes of sedimentation, and dry deposition are described in detail in Stier et al. (2005). The below-cloud and stratiform in-cloud scavenging schemes of Croft et al. (2009Croft et al. ( , 2010 are employed for this study.

Convective aerosol wet scavenging parameterizations 5
Here we describe the different convective wet scavenging, and cloud droplet number concentration parameterizations used in this study.

Original convective wet scavenging parameterization
The in-cloud aerosol scavenging parameterization for convective clouds in the standard ECHAM5-HAM model follows Stier et al. (2005). The scavenging of aerosols in 10 convective clouds is coupled to the mass flux scheme of convective tracer transport of Tiedtke (1989). In the convective updrafts, the fraction of aerosol mass and number that are cloud-droplet-borne and ice-crystal-borne are prescribed as a function of the aerosol mode alone for the purposes of the wet removal parameterization. These prescribed fractions, R i , are given in Table 1. The cloud-droplet-borne and ice-crystal- 15 borne fractions are assumed to be equal for each aerosol mode. The change in the i th tracer due to convective wet deposition at each model layer is where C liq i and C ice i are the concentrations of tracer associated with the liquid and ice phase, E liq and E ice are the fraction of liquid and ice water, respectively, that are 20 converted to precipitation.
For each model layer, a grid box mean deposition flux F dep i is found where F up is the grid box mean updraft mass flux. The grid box mean tracer tendency is where g is the acceleration due to gravity and ∆p is the model layer thickness in pressure units. The tracers deposition fluxes are integrated from the model top downward. 5 The mean updraft tracer flux for the i th tracer is recalculated as Finally, the fraction of evaporating precipitation is used to reduce the integrated tracer deposition flux as described in detail in Stier et al. (2005). This scavenging parameterization implicitly allows for aerosols that are entrained into the updraft above the 10 cloud base to become cloud-droplet-borne/ice-crystal-borne and removed by precipitation formation since the aerosol concentrations, C i , at each layer are adjusted by prescribed entrainment and detrainment rates.

New convective wet scavenging parameterization
The new, more physically detailed parameterization for convective aerosol wet scav- 15 enging is linked more closely to the convective cloud microphysics of Lohmann (2008). The convective cloud droplet number concentration (CDNC) following the Lohmann (2008) scheme allows for activation only of those aerosols that enter the cloud at the cloud base layer. The CDNC is transported upwards in the convective updraft. The microphysical conversion rates in the updraft include, autoconversion of cloud droplets 20 to form raindrops, heterogeneous contact and immersion freezing of cloud droplets, aggregation of ice crystals to form snow flakes, and accretion of raindrops with cloud droplets, and accretion of snow flakes with both cloud droplets and ice crystals. This scheme is described in detail in Lohmann (2008).
The activation scheme used throughout this study is the Ghan et al. (1993) scheme. For our study, this scheme is preferred over the Lin and Leaitch (1997) scheme since the Ghan et al. (1993) scheme does not implicitly account for the effects of entrainment on the number of activated droplets. The number of activated droplets is where N aer>25mn are the number of aerosols larger than 25 nm in radii, β is 0.0034 cm −4 s −1 , and w is the vertical velocity such that CAP E is the convective available potential energy, T KE is the turbulent kinetic energy and ω is the large-scale vertical velocity. 10 Our new parameterization of cloud-droplet-borne aerosols begins with a diagnosis of the mass and number of cloud-droplet-borne aerosols for each aerosol mode at the cloud base. The initial number of aerosols that are cloud-droplet-borne is equated to the convective CDNC at the cloud base. This number is apportioned between the aerosol modes, and separate cloud-droplet-borne mass fractions are calculated as de- 15 scribed in detail in Croft et al. (2010). The remaining interstitial aerosols can become cloud-droplet-borne or ice-crystal-borne by collision processes. The prescribed collision kernels of Hoose et al. (2008) are used for this study. The aerosol mass and number that are cloud-droplet-borne, ice-crystal-borne and in the interstitial phase for each mode are treated as separate variables in our model in order to calculate the 20 wet removal of the aerosol mass and number for each mode within the context of the convective tracer transport scheme. These auxiliary variables are not passed between time-steps in our model since the convective clouds collapse after each time-step. As the cloud droplets move up through the model vertical layers due to the updrafts, the cloud-droplet-borne aerosol mass and number are transported upwards, and are mod- 25 ified at each vertical level based on the microphysical conversion rates for freezing and precipitation formation. Aerosols are also released to the interstitial phase by evaporation due to the Bergeron-Findeisen process. Aerosols entrained above the cloud base can enter the cloud droplets and ice crystals by collisions, but not by activation, following the Lohmann (2008) convective cloud microphysics.
The equation that governs the cloud-droplet-borne aerosol mass as it moves up in 5 the modeled vertical layers for the j th aerosol mode and for the model level k is m j,k,CDCV = m j,k−1,CDCV +∆m j,k,coll −∆m j,k,f rz −∆m j,k,BF P −∆m j,k,auto −∆m j,k,acc (7) where m j,k−1,CDCV is the cloud-droplet-borne aerosol mass in the cloud droplets from the underlying model layer, and the mass change is ∆m j,k,coll due to collisions between cloud droplets and interstitial aerosols, ∆m j,k,f rz due to freezing, ∆m j,k,BF P due to 10 evaporation during the Bergeron Findeisen process, and ∆m j,k,auto and ∆m j,k,acc due to autoconversion and accretion, respectively. Similar processes are considered for the cloud-droplet-borne aerosol number. The processes that modify the ice-crystal-borne aerosol mass for the j th mode and for the model level k are where m j,k−1,ICCV is the ice-crystal-borne aerosol mass from the underlying model layer, and the mass change is ∆m j,k,colli due to collisions between ice crystals and interstitial aerosols, ∆m j,k,f rz due to freezing, and ∆m j,k,agg due to aggregation. There is a similar treatment for the aerosol number that is ice-crystal-borne. 20 The interstitial aerosol mass is where m j,k−1,inter is the interstitial aerosol mass from the underlying model layer. The interstitial aerosol number is treated similarly. Following this diagnosis of the cloud-droplet-borne and ice-crystal-borne aerosol, standard model, within the context of the convective tracer transport scheme. However, the cloud-droplet-borne and ice-crystal-borne fractions, R i , are explicitly diagnosed for each mode and each model level, and also separately for the liquid and ice phase. The cloud-droplet-borne aerosol mass fraction is R j,k,liq = m j,k,CDCV m j,k,CDCV + m j,k,ICCV + m j,k,inter (10) 5 and the ice-crystal-borne aerosol mass fraction is There is a similar treatment for the aerosol number tracers. The change of the j th tracer due to convective wet deposition at model level k is  (2008) did not explicitly account for the possibility of entrained aerosols to become activated above the cloud base, nor the explicit detrainment of the CDNC except at cloud top. For one concentrations and wet removal to different limiting assumptions that could be made by convective cloud schemes. Both limiting cases implement our new calculated convective cloud-droplet-borne and ice-crystal-borne aerosol fractions, which replace the prescribed fractions of Stier et al. (2005).
To implement this second limiting case, we use the model prediction of the number 5 of aerosols that are entrained into the updraft at each vertical level. We assume that 100% of these aerosols having radii greater than 25 nm can participate in the Ghan et al. (1993) activation scheme, and apply this activation scheme at each model level from the cloud base layer upwards. The number of newly formed cloud droplets is added to the cloud droplet number that is transported up from the underlying model 10 layer. The supersaturation required to activate eintrained aerosols could develop if the entrained air parcel accerlates and is exposed to low cloud droplet surface area either because there is negligible mixing with the existing updraft air, or rain-out or dilution has reduced the CDNC. These conditions are more likely representative of precipitating deep convective clouds. The prescribed entrainment rates in our model 15 for shallow, midlevel and penetrative convection are 1x10 −3 , 1x10 −4 , and 2x10 −4 m −1 , respectively. Further details about the calculation of entrainment and detrainment rates are in Tiedtke (1989) and Nordeng (1994). These additional terms due to activation of aerosols entrained above cloud base, and detrainment are used to adjust the convective CDNC and resultant cloud-droplet-borne 20 aerosol mass and number variables, (13) where ∆m j,k,act,ent is the cloud-droplet-borne mass change due to entrained aerosols becoming activated and ∆m j,k,det is the change due to detrainment of the cloud droplets. There is a similar treatment for aerosol number in the cloud droplets. 25 13 2.2 Model simulations Table 2 summarizes the model simulations that were conducted for this study. Simulation PF init is the control simulation with the standard ECHAM5-HAM model and with the convective microphysics of Lohmann (2008). There is entrainment and detrainment of aerosols along the convective updraft. The initial convective cloud droplet 5 number concentration (CDNC) is determined based on activation at cloud base, and this CDNC is transported upwards and modified only by freezing, evaporation, rain-out in the updraft, and cloud-top detrainment. For the aerosol wet removal, the fraction of aerosol mass and number that are cloud-droplet-borne and ice-crystal-borne are prescribed fractions (PF), which are given in Table 1 and follow Stier et al. (2005). Simulation CF init is identical to PF init except that the Lohmann (2008) convective cloud microphysics is used to determine the calculated fraction (CF) of aerosol number and mass that is either cloud-droplet-borne, or ice-crystal-borne as described in the previous section. Similar to simulation PF init, there is entrainment and detrainment of aerosols along the convective shaft, the initial convective CDNC is determined by ac- 15 tivation of aerosols at the cloud base, with modifications only by freezing, evaporation, rain-out and cloud-top detrainment. Aerosols do not activate above cloud base, but can enter cloud droplets and ice crystals by collisions. Figure 1 shows a schematic describing the model set-up for the calculated fractions (CF) simulations. The first limiting case simulation is CF pipe, which is similar to 20 CF init, except that aerosols are not allowed to entrain above cloud base and aerosols detrain only at cloud top, similar to the cloud droplets in simulation CF init. We calculate the cloud-droplet-borne and ice-crystal-borne aerosol fractions as described previously. The second limiting simulation is CF ed. This simulation is similar to simulation CF init except that the aerosols entrained above the cloud base are allowed to activate 25 and the cloud droplets are also allowed to detrain at all levels above the cloud base, similar to the aerosols in simulation CF init. As well, the aerosols are allowed to detrain and entrain above the cloud base at all model levels. The cloud-droplet-borne and ice-crystal-borne aerosol fractions are calculated as described in the previous section based on the convective cloud microphysics.

Effects of convective cloud schemes on aerosol concentrations
3.1.1 Effects of explicit calculation of aerosol into convective cloud condensate 5 We introduce an explicit calculation of the fraction of aerosol in the convective updrafts that is cloud-droplet-borne and ice-crystal-borne, based on the convective cloud microphysics of Lohmann (2008) (simulation CF init). This replaces the prescribed fractions (Table 1) used for the convective wet scavenging parameterization in the control simulation PF init. Figure 2 (panels b) and c)) shows the absolute and relative differences between simulation PF init and CF init considering the predicted annual, zonal mean vertical profiles of soluble/internally mixed accumulation mode aerosol mass. The soluble/internally mixed accumulation mode mass concentrations are increased by a factor of two in the lower tropical troposphere and by about one order of magnitude in the 15 upper tropical troposphere for simulation CF init relative to simulation PF init. This occurs since the aerosol wet removal for the simulation PF init is more vigorous than for simulation CF init. For simulation PF init, aerosols are entrained along the entire updraft shaft and 99% of accumulation mode aerosols in the updraft are assumed to be cloud-droplet-borne and ice-crystal-borne and susceptible to removal by precipitation 20 formation. On the other hand, for simulation CF init only those aerosols entering the updraft at the cloud base are allowed to become cloud-droplet-borne and ice-crystalborne as a result of acting as a cloud droplet nucleus, although aerosols entrained above the cloud base can enter the cloud droplets and ice crystals by collisions. These assumptions are consistent with the convective microphysics of Lohmann (2008). This 25 limits the aerosol wet removal since aerosols entrained above the cloud base do not activate to form additional cloud droplets. Wet removal budgets are examined in greater detail in Section 3.2. In Section 3.1.2 we also explore how greatly cloud droplet-aerosol collisions contribute to aerosol removal in our model by comparing simulation CF init with the limiting case simulation CF pipe, which does not allow aerosol entrainment 5 above the cloud base. Table 3 shows the global and annual mean aerosol burdens, lifetimes, and the aerosol optical depth (AOD) for the simulations PF init and CF init. Simulation PF init has lower aerosol burdens, lifetimes and AOD, about a factor of 0.6 relative to simulation CF init (and lowest amongst all simulations) as a consequence of more vigourous 10 wet scavenging.
3.1.2 Limiting cases: Aerosol pipe versus continuous aerosol entrainment/activation and detrainment The preceding subsection found a strong sensitivity of the predicted aerosol concentrations to the implementation of calculated cloud-droplet-borne fractions based on the 15 convective cloud microphysics of Lohmann (2008) relative to the use of the prescribed fractions of Stier et al. (2005). We now explore the sensitivity of our new wet removal parameterization to limiting assumptions that could be made for the parameterization of the convective CDNC. We focus on assumptions related to the possibility for entrained aerosols to activate above the cloud base. All of the CF simulations use the convective 20 CDNC to calculate the cloud-droplet-borne aerosol fraction in the updraft. So we first examine the sensitivity of the convective CDNC to the assumptions for our two limiting cases. Figure 3 shows the annual and zonal mean convective (CV) CDNC for the simulation CF ed, which revises the Lohmann (2008) convective CDNC parameterization to allow 25 activation of aerosols entrained above the cloud base, and also reduces the CDNC due to detrainment from the updraft shaft. The convective CDNC has a maximum at the Northern Hemisphere mid-latitudes. There is also a secondary maximum in the tropics. 16 The upper right panel of Fig. 3 shows that the CDNC is larger by up to a factor of two in the tropics for simulation CF ed compared to simulation CF pipe. Simulation CF pipe allows the activation of aerosols at cloud base only, similar to simulation CF init. This assumption strongly influences the tropical CDNC. As will be examined in Section 3.2, the tropical convective precipitation is also reduced for simulation CF ed relative to 5 simulation CF pipe. Our result is similar to the result of Fridland et al. (2004) who modeled increased cloud hydrometeor concentrations when aerosols above the cloud base were allowed to entrain and activate. The bottom panels of Fig. 3 show the contribution of activated, entrained aerosols to the CV CDNC, and the reduction due to CV CDNC detrainment, both multiplied by the vertical mass flux, for simulation CF ed. 10 There is a maximum CDNC detrainment for the Northern Hemisphere mid-latitudes where the CDNC has a maximum, and the entrainment effects on the CDNC have a maximum in the tropics where there is a maximum for the mass of soluble internally mixed mode aerosols as shown in Fig. 2. Figure 2 shows the vertical profiles of annual, zonal mean soluble/internally mixed 15 accumulation mode mass concentrations for the limiting case simulation CF ed, and also the absolute and relative differences between all four simulations. The accumulation mode mass concentrations have a maximum in the tropics, north of the equator due to the combination of biomass burning, and dust emissions (panel a)). The mass concentrations are increased near this maximum for all simulations relative to the sim-20 ulation PF init, which has the most vigorous aerosol wet removal by convective precipitation (wet deposition budgets are in the following subsection). The most pronounced absolute difference between the two limiting cases is in the tropics (panel h). Simulation CF ed has lower accumulation mode concentrations in the tropics by about 50% relative to CF pipe (panel i)) since the aerosol wet removal by convective precipitation 25 is more vigorous when aerosols entrained above cloud base are allowed to activate and become susceptible to removal by convective precipitation formation. This effect is greatest in the tropics where convective precipitation rates are the largest. Simulation CF pipe is most similar to simulation CF init since both of these simulations allow aerosol activation to form cloud droplets in only the cloud base layer. The upper tropospheric concentrations differ about one order of magnitude for simulations CF pipe and CF init relative to the standard model simulation PF init. Simulation CF ed is closest to the standard model simulation PF init, but has greater concentrations by about a factor of two in the upper troposphere. 5 Table 3 shows the global and annual mean aerosol burdens, lifetimes, and the aerosol optical depth (AOD) for these two limiting cases, in addition to those for the PF init and CF init simulations. Large differences exist between the two limiting cases. Burdens, lifetimes and aerosol optical depth are about 30% lower for the simulation CF ed as compared to CF pipe. Thus, the net influence of allowing aerosols entrained 10 above the cloud base to activate in the simulation CF ed yields more vigourous aerosol wet removal in the annual and global mean relative to the limiting case simulation CF pipe, which allows only aerosols entering the updraft at the cloud base layer to be susceptible to removal by precipitation formation. The aerosol lifetimes are similar between the CF init and CF pipe. This shows that collision processes above cloud base 15 make a limited contribution to the uptake of aerosols into the cloud droplets and ice crystals in our model. Allowing cloud droplet activation above cloud base is a strong controlling factor in our model, as is demonstrated by examining the aerosol burdens and lifetimes for simulation CF ed relative to both CF init and CF pipe. Simulation PF init is most similar to simulation CF ed since PF init implicitly assumes that 99% 20 of soluble accumulation and coarse mode aerosols entrained above the cloud base will become cloud-droplet-borne/ice-crystal-borne and susceptible to removal by precipitation formation. However, burdens, lifetimes and AOD are about 25% larger for simulation CF ed relative to simulation PF init. This can be attributed to less aerosol removal in the mixed liquid and ice phase clouds with temperatures between 238K and 25 273K as will be examined in further detail in the following section. 18 3.2 Sensitivity of wet deposition to convective assumptions Figure 4 shows the annual and global mean aerosol wet deposition attributed to convective precipitation relative to the total wet deposition (including both stratiform and convective precipitation), and also relative to the total aerosol deposition. Particulate organic matter is the aerosol species that has the most wet removal attributed to scav-5 enging in convective clouds relative to the total wet deposition (about 20 and 25% for simulations CF pipe and CF ed, respectively). Convective wet scavenging also makes the greatest contribution to total aerosol deposition for organic matter, followed by black carbon (about 25% and 20%, respectively for simulation CF ed). These species have the greatest emissions in regions of high convective precipitation, and do not have 10 fast dry deposition and sedimentation rates like dust or sea salt. Thus, the parameterization of convective wet scavenging is particularly relevant for the carbonaceous aerosols. For the carbonaceous aerosols and sulfate, the fractional contribution of convective wet scavenging to total removal is 20% greater for the simulation CF ed as compared to CF pipe and this contributes to the 30% shorter aerosol lifetimes as 15 shown in Table 3. All of our results for the fractional contribution of convective wet deposition to total wet deposition are in the lower end of the 10-90% range of Textor et al. (2006), and similar to the results of Fang et al. (2011) with the AM3 global model. The fractional contribution of convective wet deposition to total wet deposition varies with species and is greatest for simulation PF init (between 20 to 35%), and is least for 20 simulation CF pipe (between 10 and 20%). Thus, differences in the convective cloud assumptions related to the activation of aerosols entrained above cloud base, as for our simulations CF pipe and CF ed, can not entirely account for the 10-90% range of Textor et al. (2006). Other factors, such as the convective transport parameterization, can contribute to this discrepancy as was found by Tost et al. (2010). 25 Figures 5 and 6 show the annual mean geographic distribution of convective wet deposition for the five aerosol species for the simulation CF ed and also the absolute difference compared with the other limiting case simulation CF pipe. The convective wet deposition is increased by up to a factor of two in the tropics for all aerosol species except sea salt for the simulation CF ed as compared to CF pipe. This occurs since aerosols entrained above cloud base are allowed to become cloud-droplet-borne and susceptible to removal by precipitation formation for simulation CF ed, unlike for simulation CF pipe. Interestingly, this increase in wet deposition occurs despite the con-5 current decrease in the convective precipitation rate for several tropical locations such as in the tropical Indo-Asian and South American areas (bottom right panel of Fig. 6). Sea salt has decreased wet deposition for simulation CF ed relative to CF pipe, by up to a factor of two. For many geographic regions this is associated with a reduction in the precipitation rate. The bottom panels of Fig. 6 show the geographic distribution 10 of convective precipitation and the absolute difference between simulations CF ed and CF pipe. We note that the regions of convective wet deposition maxima are generally coincident with the precipitation maxima. The global and annual mean precipitation was 1.7 and 1.4 m yr −1 for the simulations CF pipe and CF ed, respectively. This precipitation change is associated with the changes to the convective CDNC parame- 15 terization between these two simulations.
Tables 4 -8 summarize the annual mean deposition budgets for the five aerosol species. The different convective cloud assumptions between the two limiting cases (CF pipe and CF ed) strongly influence the convective wet deposition budgets. The convective wet deposition attributed to warm and mixed phase clouds is greater by a 20 factor of about 3 for sulfate, carbonaceous aerosols and dust for the CF ed compared to the CF pipe simulation. As a result, the scavenging by ice phase clouds is reduced by more than one order of magnitude for simulation CF ed relative to CF pipe. While the warm phase scavenging is similar between CF ed and PF init, the mixed phase scavenging is lower for all CF simulations relative to simulation PF init (by about one half to 25 one third for simulation CF ed). Thus, the wet removal based on prescribed fractions as a function of aerosol mode alone, and applied across the entire cloud temperature range, is not consistent with the wet removal based on cloud-droplet-borne/ice-crystalborne fractions calculated from the convective cloud microphysics. Our results point to the importance of the convective cloud scheme assumptions in the prediction of the wet removal of aerosols attributed to convective precipitation. Figure 7 shows the annual mean geographic distribution of aerosol optical depth (AOD) from the MODIS/MISR/AERONET data set of van Donkelaar et al. (2010). The AOD 5 maxima are associated with dust emissions from Africa, and anthropogenic pollution sources from Asia and India. The simulation CF ed has the closest agreement with the observations. The convective scavenging for simulation CF pipe was less vigorous and this leads to an over-estimate of tropical AOD by up to a factor of three, particularly over the tropical oceans, and more remote to sources. Simulation PF init has the most 10 vigorous convective wet scavenging, but this underestimates the AOD, particularly associated with the African dust maximum by about 50%. Figure 8 shows a scatter plot of observed and modeled sulfate wet deposition. The observations are from the global dataset of Dentener et al. (2006a), but restricted to latitudes between 30 • S and 30 • N. All simulations are reasonable and have similar 15 scatter with about two-thirds of the modeled deposition being within a factor of two of the observations. The slope parameter shows that the model underestimates sulfate wet deposition in the tropical regions, which may indicate missing sources since sulfate production in convective clouds is not included in our model. The correlation coefficient is lowest for the PF init simulation (0.55) and only slightly improved for all 20 CF simulations (0.57-0.58).

Comparison with observations
Comparisons with aircraft observations of black carbon at tropical latitudes (Koch et al., 2009(Koch et al., , 2010 are shown in Fig. 9. The agreement between the observations and the model is best and similar for the simulation PF init and CF ed, except that PF init underestimates the concentrations in the upper troposphere by a factor of 2 relative 25 to the observations. The agreement with upper tropospheric observations is closer for CF ed. The black carbon concentrations in the upper troposphere differ by up to one order of magnitude between the two limiting cases (CF pipe and CF ed), and also 21 between the standard model simulation PF init and the calculated fractions simulation CF init. Thus convective clouds assumptions, particularly about how aerosols entrain and become cloud-droplet-borne above cloud base are shown to strongly influence predicted aerosol concentrations.

5
We examined the sensitivity of aerosol concentrations, burdens, lifetimes and wet deposition to limiting assumptions made by convective cloud schemes in a global climate model. To facilitate this study, we coupled the two-moment convective cloud microphysics of Lohmann (2008) to the aerosol wet scavenging parameterizations of the ECHAM5-HAM global climate model. Similar to many global models, the standard 10 ECHAM5-HAM model assumes prescribed fractions of the aerosol mass and number are cloud-droplet-borne and ice-crystal-borne, and thus susceptible to wet removal by convective precipitation formation. We introduced into the ECHAM5-HAM model an explicit representation of the uptake of aerosol mass and number into convective cloud droplets and ice crystals by the processes of activation, collisions, freezing and 15 evaporation to provide a closer coupling with the convective microphysics scheme of Lohmann (2008). This allowed an explicit calculation of the fraction of aerosol mass and number that was cloud-droplet-borne and ice-crystal-borne, and susceptible to convective wet deposition. The standard model had more vigorous convective wet removal, and the annual and global mean aerosol burdens were about 0.6 of those 20 with the revised convective wet deposition parameterization. Aerosol concentrations in the upper troposphere differed by up to one order of magnitude. The wet removal attributed to scavenging in convective clouds between temperatures of 238K and 273K where the liquid and ice phase could co-exist was reduced by a factor of two to five with the revised convective wet deposition parameterization, suggesting that implementa-microphysics. The parameterization of the influences of entrainment and detrainment on convective cloud droplet number number concentration (CDNC) in global models is associated with considerable uncertainty. As a result, two limiting cases were considered in this study. In the first case, the convective updraft was considered to behave like an isolated 5 pipe for aerosols and cloud droplets; aerosol activation was allowed only at the cloud base layer and detrainment of aerosols and cloud droplets/ice crystals occurred only at the cloud top. In the second limiting case, aerosols were allowed to entrain and detrain along the updraft, and the entrained aerosols could activate above the cloud base layer, and the CDNC could detrain the along the updraft. Allowing activation 10 of aerosols entrained above cloud base enhanced by a factor of two the annual and zonal mean convective CDNC in the tropics. Convective wet scavenging was more vigorous, which lowered by 30% the annual and global mean aerosol burdens and lifetimes. Aerosol concentrations differed by up to one order of magnitude in the upper troposphere between the two limiting cases. Closest agreement with observations of 15 the geographic distribution of the annual mean aerosol optical depth, particularly in the tropics, was found for the convective cloud scheme that allowed aerosols entrained above cloud base to activate and become cloud-droplet-borne/ice-crystal-borne. The sensitivity of aerosol concentrations to these convective cloud assumptions motivates the need for on-going observations coupled with modeling studies to better elucidate 20 the role of convective precipitation scavenging of aerosols in the global climate system.
The results of this study point to the importance of ongoing work to understand the kinematics of convective clouds, and convective entrainment and detrainment rates. Our results suggest that parameterizations of convective CDNC for global models should account for the possibility of aerosols entrained in the updrafts to become acti-of magnitude in the upper troposphere between our simulations) depending on the convective cloud assumptions related to entrained aerosols and their possibility to become cloud-droplet-borne/ice-crystal-borne. Scavenging parameterizations should be developed that account for the possibility of aerosols that are entrained above cloud base to become cloud-droplet-borne/ice-crystal-borne, and thus susceptible to wet re-5 moval, although concurrently detrainment processes must also be considered. In order to improve convective wet scavenging parameterizations in global models, ongoing field work and case studies are needed with an emphasis on examining how readily aerosols are entrained and become cloud-droplet-borne/ice-crystal-borne in the convective updrafts. These measurements must involve many varieties of convective cloud 10 conditions. Ultimately, improvements in convective cloud microphysics coupled with more physical wet scavenging parameterizations will improve the prediction of threedimensional aerosol distributions in our global models and help to resolve the ongoing uncertainty related to how convective clouds contribute to removing particulate matter from the atmosphere. 15 Twomey, S.: Aerosol, clouds, and radiation., Atmos. Environ., 25A, 2435-2442, 1991: Global estimates of exposure to fine particulate matter concentrations from satellite-based aerosol optical depth, Environ. Health Perspec.,118(6) Table 2. Simulation PF init has the same set-up as for simulation CF init, except that the cloud-dropletborne/ice-crystal-borne aerosol fractions are prescribed following Stier et al. (2005), not calculated based on the convective cloud microphysics of Lohmann (2008) as for the calculated fraction (CF) simulations.

Fig. 2.
The zonal and annual mean soluble/internally mixed accumulation (AS) mode mass concentration at STP for the simulation CF ed (top panel), and the absolute and percent differences between the four simulations (remaining panels). All simulations are described in Table  2. The colorscales change between the different panels. Fig. 3. The zonal and annual mean convective (CV) cloud droplet number concentration (CDNC) for the simulation CF ed, and CV CDNC difference between the two limiting cases (CF ed and CF pipe) (top panels). The contribution of activated entrained aerosols to zonal and annual mean CV CDNC, and the reduction in the CV CDNC due to detrainment (bottom panels) (both multiplied by the vertical mass flux) for the simulation CF ed. Simulations are described in Table 2 Fig. 4. The annual and global wet deposition attributed to convective (CV) precipitation relative to the total aerosol wet deposition attributed to both stratiform and convective precipitation (top panel), and the wet deposition attributed to convective (CV) precipitation relative to the total deposition (bottom panel) for each aerosol species. Simulations are described in Table 2.  5. The geographic distribution of annual mean convective wet deposition for the simulation CF ed and the absolute difference between CF ed and CF pipe for sulfate, black carbon and particulate organic matter. Simulations are described in Table 2. Fig. 6. The geographic distribution of annual mean convective wet deposition for the simulation CF ed and the absolute difference between CF ed and CF pipe for sea salt and dust, and the convective precipitation for simulation CF ed and the absolute difference between CF ed and CF pipe. Simulations are described in Table 2. Vertical profiles of black carbon (BC) concentration observations (OBS) from the aircraft data for tropical latitudes as described in Koch et al. (2009Koch et al. ( , 2010 and for the four simulation of this study as described in Table 2. Black dashed lines show observations from a different day of the aircraft campaign.   Stier et al. (2005), init: following initial (i.e. unmodified) assumptions of the standard ECHAM5-HAM as described above for PF init CF pipe Limiting case 1: CF: Calculated cloud-droplet-borne/ice-crystal-borne aerosol fractions as described for CF init, pipe: updraft is like a pipe for the aerosols, i.e. no entrainment of aerosols in the updraft above cloud base, detrainment only at cloud top, which is similar to the treatment of the convective CDNC following the cloud microphysics of Lohmann (2008) CF ed Limiting case 2: CF: Calculated cloud-droplet-borne/ice-crystal-borne aerosol fractions as described for CF init, ed: entrainment and detrainment explicitly influence the convective CDNC, the Lohmann (2008) convective CDNC parameterization is modified to allow activation of aerosols entrained above cloud base, and CDNC reduced by detrainment at all updraft levels Table 3. Global and annual mean aerosol burdens (Tg, except Tg S for sulfate) and lifetimes (days) in brackets after the burdens, and aerosol optical depth (AOD) for the four simulations presented in