Corresponding author: J. Liu, College of Urban and Environmental Sciences, Peking University, Beijing 100871, China. (email@example.com)
 Secondary organic aerosols (SOA) constitute a significant fraction of ambient aerosols, but their global source is only beginning to be understood. Substantial evidence has shown that oxidation of water-soluble organic species in the liquid cloud leads to the formation of SOA. To evaluate this global source and explore its sensitivity to various assumptions concerning cloud properties, we simulate in-cloud SOA (IC-SOA) formation based on detailed multiphase chemistry incorporated into the newly developed Geophysical Fluid Dynamics Laboratory (GFDL) coupled chemistry-climate model AM3. We find global IC-SOA production is around 20–30 Tg·yr−1between 1999 and 2001. Depending on season and location, oxalic acid accounts for 40–90% of the total IC-SOA source (particularly between 800 hPa–400 hPa), and glyoxylic acid and oligomers (formed by glyoxal and methylglyoxal in evaporating clouds) each contribute an additional 10–20%. Besides glyoxal and methylglyoxal (extensively studied by previous research), glycolaldehyde and acetic acid are among the most important precursors leading to the formation of IC-SOA, particularly oxalic acid. Different implementations of cloud fraction or cloud lifetime in global climate models could potentially modify estimates of IC-SOA mass production by 20–30%. Dense IC-SOA production occurs in the tropical and midlatitude regions of the lower troposphere (surface to 500 hPa). In DJF, IC-SOA production is concentrated over the western Amazon and southern Africa. In JJA, substantial IC-SOA production occurs over southern China and boreal forest regions. This study confirms a significant in-cloud source of SOA, which will directly and indirectly influence global radiation balance and regional climate.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Organic aerosols (OA) constitute 20–90% of global fine particle mass, influencing solar radiation, global climate, human health and atmospheric visibility [Kanakidou et al., 2005; Zhang et al., 2007; Aiken et al., 2008; Kroll and Seinfeld, 2008; Hallquist et al., 2009; Liu et al., 2009; Lim et al., 2010]. According to the new IPCC “Representative Concentration Pathways” (RCP) emission projections, SO2 and BC sources could decrease by more than 80% and 50%, respectively, by the end of the 21st century, while the reduction of OA emissions is projected to be small [Intergovernmental Panel on Climate Change, 2008]. This indicates that OA could contribute a larger share of aerosols' composition and might play a more important role in future climate and air quality [Tsigaridis and Kanakidou, 2007]. Unlike primary OA (POA) which are released directly from fossil fuel, biofuel and biomass burning [Seinfeld and Pandis, 2006], secondary organic aerosols (SOA) are formed in the ambient air when volatile organic gases, such as isoprene, terpenes and aromatic species, are oxidized to form semi-volatile or low volatility organic species [Carlton et al., 2009]. Those oxygenated species will either rapidly condense onto pre-existing particles or nucleate to form new particles, leading to SOA mass accumulation [Chung and Seinfeld, 2002]. OA is an aggregate of hundreds to thousands of individual compounds [Turpin et al., 2000; Goldstein and Galbally, 2007], produced by gas-phase or even multiphase reactions. Detailed species based quantification of the evolution of all organic species in conjunction with multiphase derivation of SOA source is challenging. Consequently, global or regional models overwhelmingly rely on the partitioning theory and empirical bi-product, or volatility basis set parameterizations to model SOA [Odum et al., 1996; Farina et al., 2010; Lim et al., 2010]. However, large discrepancies are found between observed organic aerosol mass concentrations and model simulations, indicating a large missing source exists in our current understanding of organic aerosol formation [Tsigaridis and Kanakidou, 2003; Heald et al., 2005]. In addition, measured ambient OA usually exhibits much higher O/C ratios than that in directly emitted OA or formed in traditional smog chamber experiments [Aiken et al., 2008].
 All these studies suggest that in-cloud production of SOA is not a trivial source of OA, but the magnitude of associated uncertainties remains large. Unlike previous estimations which either rely on parameterized calculation, or focus on one specific IC-SOA species, this study explores the potential global IC-SOA sources and precursors based on new laboratory results, and investigates the effects of cloud properties on SOA production by implementing a detailed cloud chemistry scheme into the GFDL global coupled chemistry-climate model AM3. Specifically, we examine the extent to which assumptions of cloud lifetime, droplet size, cloud types, and cloud fraction could impact the amount of SOA mass produced. We describe the coupled gas and cloud chemistry insection 2, and quantify the global budget of IC-SOA insection 3. A variety of sensitivity simulations are conducted in section 4to explore the effects of cloud properties on IC-SOA production. The global distribution and seasonal variability of IC-SOA production are investigated insection 5. Finally, conclusions are drawn in section 6.
2.1. Model Description
 We simulate global IC-SOA production using the GFDL global coupled chemistry-climate atmosphere and land model AM3 [Donner et al., 2011], with new developments on cloud chemistry. AM3 has been developed from the previous generation GFDL AGCM AM2 [Anderson et al., 2004], with improvements in physics, dynamics, cloud and precipitation processes, and enhanced coupling to chemistry, aerosol, and land models. AM3 implements a finite-volume dynamical core [Lin, 2004] on a cubed sphere grid (projection of a cube onto the surface of a sphere, Putman and Lin ), replacing the latitude-longitude grid used in AM2. The model is configured with a horizontal resolution of approximately 200 × 200 km2and 48 hybrid vertical levels from the surface to 1 Pa (∼80 km). Besides a prognostic scheme of large-scale cloud condensate and volume fraction [Tiedtke, 1993; Rotstayn, 1997; Rotstayn et al., 2000], AM3 predicts both shallow cumulus clouds [Bretherton et al., 2004] and deep convective clouds [Donner, 1993]. Interactions between aerosols and large-scale liquid clouds are simulated using a prognostic scheme for droplet number [Ming et al., 2007], in which sulfate, organic matter, and sea salt aerosols are treated as CCN. Effect of aerosols on ice crystal number is not calculated. The radiative transfer algorithm and planetary boundary layer (PBL) parameterization (based on Lock et al. ) are essentially unchanged from AM2 [Anderson et al., 2004], except that hydrophilic BC and sulfate are assumed to be internally mixed for radiation calculation. AM3 includes detailed tropospheric and stratospheric chemistry with nearly 90 gas and aerosol species and more than 200 reactions. The gas-phase photochemistry mechanism is based onHorowitz et al.  and Austin and Wilson . A more detailed description of AM3 and an evaluation of simulated climate are given by Donner et al. .
2.2. Mechanism of SOA Formation From Cloud Process
 The standard version of AM3 simulates SOA from natural and anthropogenic sources. The natural source of SOA includes oxidation of terpenes emitted from plants with yield factors varying by latitude. The anthropogenic source follows Tie et al. , assuming a 10% yield of SOA from butane oxidation by OH [Donner et al., 2011]. In this study, we update the AM3 gas-phase chemistry from the MOZART-2 to MOZART-4 mechanism [Emmons et al., 2010], including a detailed photochemical mechanism for isoprene, mono-terpene and aromatic species (MOZART-4 chemistry includes a lumped species TOLUENE for both toluene and xylene [Emmons et al., 2010], and they will be differentiated in the future improvement). The associated reaction rates are updated according to JPL 2006 (http://jpldataeval.jpl.nasa.gov). Detailed description of the gas-phase SOA parameterization and evaluation of organic aerosols against a variety of surface and aircraft measurements are given in the follow-up paper. Global parameterized gas-phase SOA sources are estimated to be around 30 Tg·yr−1.
 The aqueous-phase oxidation of SO2 by H2O2 and O3 is parameterized following Tie et al. [2001, 2005]in the standard AM3. The use of similar methods to simulate the processes of IC-SOA production is difficult due to the complexity of organic aqueous-phase reactions. Therefore, we develop an aqueous-phase chemistry module for AM3 using a gas-liquid mass transport model and available aqueous phase reaction mechanisms. The IC-SOA chemistry uses an optimized cloud chemistry mechanism [Liang and Jacobson, 1999], merged with aqueous organic chemistry evaluated by previous studies [e.g., Lim et al., 2005; Ervens et al., 2008; Tan et al., 2009, 2010, 2011]. The merged gas- and aqueous-phase chemical mechanism is listed inTable 1 and Tables S1a–S1f in Text S1 in the auxiliary materials. Figure 1summarizes the primary paths for IC-SOA formation. The physical and chemical processes include the following.
 1. Gas-phase oxidation.Hydrocarbons (HCs), such as isoprene and terpenes released from natural sources as well as aromatic species (e.g., benzene, toluene, xylene) emitted from anthropogenic sources, are oxidized (e.g., by the OH radical) in the ambient air to form semi-volatile and low volatility organic species (some of which lead to the gas-phase formation of SOA) as well as water-soluble gases (WSG), such as glyoxal, methylglyoxal, glycolaldehyde, acetic acid, and hydroxyacetone.
 2. Gas-liquid transport. In the presence of clouds, WSG will diffuse onto the surface of cloud droplets and be absorbed into the cloud water.
 3. Liquid-phase oxidation. The dissolved WSG are oxidized by OH radical inside the cloud droplet, leading to the formation of lower volatility carboxylic acids, such as glycolic acid, glyoxylic acid, pyruvic acid and oxalic acid. These carboxylic acids will be eventually oxidized to CO2 if the lifetime of the cloud droplet is sufficiently long.
 4. Cloud evaporation and IC-SOA formation. When cloud droplets evaporate, the carboxylic acids produced within the cloud are assumed to be all remained in the aerosol phase as SOA. In addition, according to Loeffler et al.  and De Haan et al. , a portion of glyoxal and methylglyoxal will form low-volatility oligomers upon cloud evaporation. Therefore, we compute the mass production of six IC-SOA species (i.e., glycolic acid, glyoxylic acid, pyruvic acid, oxalic acid, and two classes of oligomers formed by glyoxal and methylglyoxal) and track these species as a new IC-SOA tracer, treated identically to gas-phase produced SOA in AM3.
 To solve the gas- and liquid-phase chemistry, either the two systems can be solved separately (e.g., calculate the gas-phase chemistry first and use its results as input for the aqueous-phase chemistry calculation) or the coupled system can be solved simultaneously. In this study, we follow the method ofJacob and convert aqueous-phase concentrations and reaction rates to the same units used for gas-phase chemistry (i.e., molecules per cubic centimeter of air):
where [Xi] and [Ci] are gas-phase and liquid-phase concentrations of speciesi in units of molecules/cmair3 and moles/Lwater, respectively; LWC is liquid water content (Lwater/Lair). Therefore, the chemical system is numerically solved without distinction between gas and aqueous-phase species.
 The mass transfer between gas and liquid cloud is governed by molecular diffusion [Schwartz, 1984], mass accommodation coefficient [Seinfeld and Pandis, 2006] and Henry's Law constant. The aqueous ionic equilibrium is assumed to hold everywhere inside the droplet since the characteristic time of the aqueous-phase dissociation reactions is short [Seinfeld and Pandis, 2006]. The absorption flux and volatilization flux of water-soluble species follows the studies of [Jacob, 1986; Liang and Jacobson, 1999]. The differences between the surface and bulk concentrations are ignored for most species, since the characteristic time for aqueous diffusion is shorter than their lifetimes against chemical decay or production [Seinfeld and Pandis, 2006]. However, due to the rapid liquid-phase reactions, surface concentrations of O3, OH and NO3 must be determined and corrected based on their bulk concentrations following Jacob .
 Cloud property information, such as cloud liquid water content, cloud droplet size and cloud volume fraction, are based on the large-scale cloud condensation simulated prognostically in AM3. The contribution of SOA from convective clouds (i.e., shallow cumulus cloud and deep convective cloud) is ignored in the base model configuration in this study, since SOA produced in convective clouds is assumed to be removed rapidly by convective precipitation. Convective scavenging of most SOA precursors is represented in the model. Cloud droplet size is assumed to be either 10μm everywhere (base) or, in sensitivity simulations, equal to 5 μm, 20 μm, or the droplet sizes simulated by the GCM prognostic cloud scheme. Given the coarse model resolution of AM3 (e.g., horizontal resolution is around 200 km × 200 km), cloud fraction (CF) is usually less than 1 in most locations. As a result, two calculations are performed in a partly cloudy grid box (i.e., 0 < CF < 1). The coupled gas and liquid-phase chemistry is used only for the cloudy fraction of the grid cell. For the non-cloudy part, only gas-phase chemistry is solved. The grid-box averaged concentration is estimated as a CF-weighted average of the cloudy and non-cloudy results. Cloud entrainment (i.e., air mass exchanges between cloudy and non-cloudy areas) is ignored in the default model configuration.
 The standard model time step of AM3 is 30 min. The default cloud lifetime (i.e., the time between cloud activation and evaporation) is set equal to the model time step. At t = 0, cloud droplets are assumed to be “instantaneously” formed at their nominal size and WSG are “quickly” scavenged by cloud droplets according to the effective Henry's law constants. Cloud pH value is determined dynamically by the dissociation of cloud condensation nuclei (CCN), the scavenging of acidic and alkaline gases (e.g., SO2, CO2, NH3, HNO3, HCOOH), the formation of liquid-phase sulfate and other carboxylic acids (e.g., CH3COOH), assuming that aqueous equilibrium and electroneutrality are continuously maintained. When a cloud droplet evaporates, the remaining aqueous-phase species will be transferred back to the gas-phase. Low volatility species such as ammonium sulfate and some carboxylic acids (e.g., glycolic acid, glyoxylic acid, pyruvic acid and oxalic acid) will be released back to the atmosphere as inorganic aerosols and SOA. In addition, 33% of glyoxal and 19% of methylglyoxal in the liquid phase are assumed to form oligomers upon cloud evaporation [De Haan et al., 2009]. The contribution of aqueous oxidation of methyl vinyl ketone (MVK) and methacrolein (MACR) is not incorporated in this study and could be an important source of IC-SOA based on recent laboratory studies [Chen et al., 2008; Zhang et al., 2010].
2.3. Model Configuration
 The emission inventories used for both gas and aerosol species are obtained from the database developed for IPCC AR5 studies [Lamarque et al., 2010]. Global total primary OM emissions in 2000 are approximately 20 Tg·yr−1 from anthropogenic sources, 37 Tg·yr−1 from biomass burning, and 15 Tg·yr−1 from oceanic emissions [O'Dowd et al., 2008]. Sea ice cover and sea surface temperature (SST) are prescribed using databases developed at the Hadley Center [Rayner et al., 2003]. We first conduct a four-year base simulation (using SST and emissions for 1998–2001, the first year of simulation is used for model spin-up) based on the default configuration for cloud chemistry in AM3. To test the dependence of in-cloud SOA production on cloud types and properties, we conduct additional sensitivity simulations listed inTable 2.
Table 2. Global Budget of In-Cloud Produced SOA
τ = 10 min
r = 5 μm
r = Predicted
r = 20 μm
Cldfr = 1
In-cloud SOA prod.
3. Global Budget
Table 2summarizes the global budget of IC-SOA, including total production, mass loading, dry deposition and wet removal. In addition, six IC-SOA species (i.e., glycolic acid, glyoxylic acid, pyruvic acid, oxalic acid and two classes of oligomers formed from glyoxal and methylglyoxal) are identified and their contributions to the total SOA mass are quantified. In the base simulation, global net chemical IC-SOA mass production (i.e., IC-SOA released upon cloud evaporation) is approximately 23 Tg·yr−1, of which more than 80% is removed by wet deposition annually. The corresponding atmospheric mass loading is 0.38Tg with an average atmospheric lifetime of 6 days. Among the six IC-SOA species, oxalic acid is the dominant component, accounting for nearly 60% of total IC-SOA mass production. The estimated annual oxalate formation is comparable to the work ofMyriokefalitakis et al. . Glyoxylic acid ranks second, accounting for nearly 15% of total IC-SOA source. In-cloud production of oligomers formed from glyoxal and methylglyoxal upon cloud evaporation are around 3.7 Tg·yr−1, 40% higher than the estimation of De Haan et al. . In-cloud production of glycolic acid and pyruvic acid are small.
 To better characterize the relative importance of chemical paths in the detailed cloud chemistry system, the rate of each liquid-phase reaction is archived. The production rate of oxalic acid by the oxidation of glyoxylic acid (i.e., Ra051–052 inTable 1; see Figure 1) is equal to 14.5 Tg·yr−1 (or 161 Gmol·yr−1). Most of the in-cloud produced oxalic acid will form SOA upon cloud evaporation. However, a small fraction of (∼7%) of the oxalic acid will be oxidized into CO2 in the liquid phase (Ra53–55). For glyoxylic acid, four reactions account for its production, with the dominant paths from the oxidation of acetic acid (51%, Ra047–048), glyoxal (37%, Ra043) and glycolic acid (11%, Ra049 and Ra050). The direct contribution from methylglyoxal is small (∼1%, Ra044). Among the 16.9 Tg·yr−1 (or 228 Gmol·yr−1) production of glyoxylic acid, nearly 71% is oxidized into oxalic acid, while the remaining is either partitioning to aerosol phase (i.e., IC-SOA, 21%) or gas phase (8%, eventually removed by dry and wet deposition). A recent laboratory study has confirmed the aqueous-phase formation of oxalate from acetic acid [Tan et al., 2011]. Glycolic acid is produced from glycolaldehyde (Ra041) at a rate of 3.0 Tg·yr−1 (or 50 Gmol yr−1), nearly half of which is further oxidized into glyoxylic acid (Ra049–050). Similarly, the primary source of pyruvic acid is from the oxidation of methylglyoxal (Ra044) and its dominant sink is formation of acetic acid (Ra045–046).
 The production of oligomers depends on the cloud abundance of glyoxal and methylglyoxal at the time the cloud evaporates. As shown in Figure 1, the main sources of cloud glyoxal are oxidation of glycolaldehyde (5.86 Tg·yr−1 or 101 Gmol·yr−1, Ra042) and gas-to-liquid exchange (1.57 Tg·yr−1 or 27 Gmol·yr−1). The main sinks of the in-cloud glyoxal are to form aqueous-phase glyoxylic acid (Ra043, 65%) and the self-associated oligomer SOA (35%). The large source of in-cloud produced glyoxal significantly enhances the mass production of glyoxal oligormers (a factor of 1.6 higher than that ofDe Haan et al. ). For methylglyoxal, its cloud source is predominantly gas-to-liquid transfer (4.1 Tg·yr−1 or 57 Gmol·yr−1), while the contribution from hydroxyacetone is trivial (<0.01 Tg·yr−1 or 0.2 Gmol·yr−1, Ra056). The cloud absorbed methylglyoxal is mostly oxidized by OH radical (72%) to produce pyruvic acid and glyoxylic acid (Ra044), and to form oligomers (28%) upon cloud evaporation.
Table 3summarizes the global budget of IC-SOA precursors. Oxidation of VOCs in the gas phase is the only source for each precursor (primary sources from anthropogenic or biomass burning emissions are not included in this study). As shown inTable 3 (also Figure 1), four WSGs (i.e., glyoxal, methylglyoxal, glycolaldehyde, acetic acid) directly contribute to the formation of IC-SOA. Their net cloud absorptions are: glyoxal (1.57 Tg·yr−1 or 27 Gmol·yr−1), methylglyoxal (4.1 Tg·yr−1 or 57 Gmol·yr−1), glycolaldehyde (9.06 Tg·yr−1 or 151 Gmol·yr−1), and acetic acid (6.96 Tg·yr−1 or 116 Gmol·yr−1). For glycolaldehyde, net global gas-phase production is about 22.2 Tg·yr−1 or 370 Gmol·yr−1, of which nearly 40% is absorbed and oxidized in cloud. The gas-phase production of acetic acid is around 78 Tg·yr−1, about 10% lower than the total acetic acid source in Paulot et al. , in which the primary source accounts for approximately 30%. For methylglyoxal, the gas-phase production is around 160 Tg·yr−1, comparable to Fu et al. . However, the gas-phase production of glyoxal is about 21 Tg·yr−1, lower than Fu et al.  by 16 Tg·yr−1, partially due to a lack of acetylene chemistry (which contributes ∼9 Tg·yr−1 of glyoxal in Fu et al. ). Approximately 6 Tg·yr−1of glyoxal is produced in the liquid phase in this study (Ra042). In addition, a number of studies indicate that oxidation of semi-volatile polycyclic aromatic hydrocarbons (PAHs), such as naphthalene and alkylnaphthalenes, could form additional glyoxal and methylglyoxal [e.g.,Wang et al., 2007; Chan et al., 2009], and their role in global production of IC-SOA should be clarified in the future work.
Table 3. Global Budget of IC-SOA Precursors (Gmol·yr−1)
Cloud net uptake
 The above analysis indicates that besides glyoxal and methylglyoxal (extensively studied by previous research), glycolaldehyde and acetic acid are among the most important precursors leading to the formation of SOA, particularly oxalic acid. If primary sources and additional secondary sources of these precursors (e.g., glyoxal and acetic acid) were considered, e.g., by including a more detailed treatment of oxidation of aromatic species [Nishino et al., 2010], IC-SOA production could be even larger. In addition, the relative contribution of carboxylic acids and oligomers to total IC-SOA mass could be substantially different than calculated in this study when their global budgets are improved with more complete chemical mechanisms and more accurate emission inventories.
4. Effects of Cloud Properties on SOA Production
 Liquid water in clouds is the media in which WSGs are oxidized, leading to the formation of inorganic and organic aerosols (could be internally mixed). Cloud properties determine the fraction of cloud volume, abundance of cloud water, size of cloud droplet, and duration of droplet cycle, which directly or indirectly influence cloud chemistry and production of IC-SOA. In this study, we conduct additional sensitivity tests to evaluate the extent to which changes in cloud properties could influence IC-SOA production.
4.1. Cloud Lifetime
 In the base simulation, cloud lifetime is assumed to be identical to AM3 model time step (30 min), longer than the typical cloud contact time for a cloud cycle (10 min) [Ervens et al., 2004]. To test the effects of cloud lifetime (i.e., the period between cloud activation and cloud evaporation) on SOA production, we conduct a simulation in which each cloud cycle is assumed to last only 10 min. As a result, three droplet cycles are involved in each model time step. A shorter cloud lifetime reduces the evolution period of water-soluble species in the liquid phase and increases frequency of cloud evaporation (which enhances the formation of oligomers from glyoxal and methylglyoxal). As shown inTable 2, a shorter cloud lifetime enhances SOA production by 20%. Specifically, oligomers are increased by 110% mainly due to the tripled evaporation frequency. As shown in Figure 1, the net absorptions of glyoxal and methylglyoxal are increased by 78% and 46%, respectively. For methylglyoxal, most of the increased absorption enhances the oligomer formation. For glyoxal, besides substantial increase in cloud absorption (78%), a higher production from glycolaldehyde (5%) and a lower loss to glyoxylic acid (−14%) account for the substantial increase in oligomer formation from glyoxal (86%). The net productions of glycolic acid, glyoxylic acid and pyruvic acid are increased by 74%, 69% and 38%, respectively (Table 2), as the reduction in loss rate exceeds that of production rate. Since the net production of oxalic acid is overwhelmingly determined by the loss rate of glyoxylic acid, a shorter cloud lifetime results in 26% decrease in oxalic acid.
 In sum, a shorter cloud lifetime substantially enhances the cloud absorption of glyoxal and methylglyoxal, which enhances the production of oligomers and first-generation carboxylic acids (i.e., oligomers, glycolic acid, glyoxylic acid and pyruvic acid), but slows down the formation of second-generation carboxylic acid (i.e., oxalic acid).
4.2. Cloud Droplet Size
 Given fixed cloud liquid water content (LWC), a smaller droplet size results in a larger total reaction surface and a shorter aqueous-phase mixing time, which enhances the gas-liquid transport and aqueous-phase reactions. In the base simulation, cloud droplet size is assumed to be 10μm everywhere. To test the sensitivity of SOA production to cloud droplet size, we conduct three sensitivity tests. Two of them assume the cloud radius is fixed at 5 μm or 20 μm everywhere and the third one uses online predicted droplet sizes. As shown in Table 2, halving cloud droplet radius slightly enhances the total SOA production (∼5%). Conversely, doubling droplet radius reduces SOA production by about 10%.
 In AM3, the droplet size of stratiform clouds and shallow cumulus clouds depends on aerosol activation as described by Ming et al. . The simulated cloud droplet radius ranges from 8 μm to 13 μm in most regions [see Donner et al., 2011, Figure 6]. As a result, replacing the fixed 10 μm cloud radius by model predicted droplet size has little impact on global SOA production (Table 2).
 These sensitivity tests indicate that global total in-cloud SOA production depends weakly on assumptions of cloud droplet sizes, and 10μm is a reasonable cloud droplet size for cloud chemistry.
4.3. Shallow Cumulus Cloud and Ice Cloud
 AM3 simulates large-scale stratiform cloud, shallow cumulus cloud and deep convective cloud [Donner et al., 2011]. For each cloud type, both water and ice amounts are resolved. In the base simulation, LWC and cloud fraction of stratiform clouds alone are employed in the cloud chemistry. To test the relative importance of other cloud types, we conduct three additional sensitivity runs. The first run combines the LWC and cloud fraction from both shallow cumulus cloud and stratiform cloud. The second run includes stratiform ice cloud in which ice is assumed to be a media identical to water. In the third run, both shallow cumulus cloud and stratiform ice cloud are included. As shown in Table 2, contribution of shallow cumulus clouds and stratiform ice cloud to SOA production is small. Incorporating both ice water and shallow cumulus clouds only enhance IC-SOA production by 10%. Therefore, the production of IC-SOA is dominated by large-scale liquid stratiform clouds.
4.4. Cloud Fraction
 The volume fraction of large-scale clouds is predicted prognostically by AM3, ranging from 0 to 1. In the base simulation, both gas-phase chemistry and coupled gas-aqueous phase chemistry are numerically solved for clear sky and cloudy sky in a model grid, and are merged by cloud fraction. We assume there is no entrainment between the cloudy and non-cloudy areas. Allowing exchange between cloudy and non-cloudy areas results in more oxidation of WSG in cloud and hence greater IC-SOA production. To test the effect of this assumption, we conduct another simulation in which 100% cloud entrainment (or CF = 1 for all cloudy boxes) is assumed. As shown inTable 2, sufficient cloud entrainment could potentially enhance global IC-SOA production by 26%, more than 65% of which is contributed from the increase in oxalic acid. Therefore, treatment of cloud fraction in global models is a sensitive parameter in estimation of IC-SOA or oxalate production.
5. Global Distribution and Seasonal Variability
 The seasonal variability and spatial distribution of IC-SOA production depend on many factors: location of cloud, cloud fraction, and availability of SOA precursors and oxidants. Therefore, regions with large emissions of VOCs (either anthropogenic or biogenic sources) in coincidence with clouds are ideal for IC-SOA formation.Figure 2shows the zonally averaged IC-SOA production in DJF, MAM, JJA and SON. In all seasons, IC-SOA production is distributed from the surface to ∼500 hPa with a maximum near 900 hPa, and horizontally between 70°S and 70°N. Comparing to the distribution of cloud liquid water content (Figure 3), IC-SOA production is potentially governed by the availability of cloud water. Specifically, the shape, vertical profile and maximum production layer overlays well the distribution of liquid cloud amount. Meridionally, four production centers are found near 30°S, 0°, 30°N and 60°N, shifting with season. In DJF, substantial IC-SOA is produced in the tropics and southern hemisphere between −30°S and 5°N, but little is produced at high latitudes. In JJA, besides the tropics and subtropics, a tremendous amount of IC-SOA is produced at boreal high latitudes between 45°N and 70°N.
Figures 4 and 5show the horizontal distribution of IC-SOA production and cloud liquid water content, respectively. As shown inFigure 4, distinct seasonal patterns are found for IC-SOA production. In tropics, such as central Africa and the Amazon, production of IC-SOA alternates with wet and dry seasons, and both spatial and seasonal patterns follow that of cloud (Figure 5). Over boreal forests, IC-SOA production varies significantly with the abundance of cloud water amount. In DJF, except Europe, low cloud liquid water in conjunction with little vegetation emission leads to the lack of both reaction media and SOA precursors, which substantially depresses the production of IC-SOA. Consequently, little IC-SOA is observed north of 45°N over both Asia and North America. In contrast in JJA, both vegetation emissions and liquid cloud amount increase substantially and tremendous amount of IC-SOA is produced over the same region.
 Over oceans, cloud water is typically abundant over North Pacific, North Atlantic and midlatitudes in the southern hemisphere. However, the IC-SOA production patterns are distinct between the northern and southern hemispheres. SOA precursors are transported along the midlatitude westerlies from eastern Asia and North America to north Pacific and north Atlantic, leading to significant IC-SOA production over oceans. Conversely, little IC-SOA is produced in the southern oceans due to a lack of SOA precursors.
Figure 6quantifies the seasonal IC-SOA production over ten defined continental regions as well as all oceans. Nearly 1/4 of IC-SOA is produced over oceans, indicating long-range transport of SOA precursors followed by oxidation in the liquid cloud is an important source of organic aerosols over oceanic regions. Among the ten regions, South America (SA, particularly the Amazon region) is the world's largest source of IC-SOA, contributing to 4.4Tg IC-SOA each year. Nearly 2/3 of the production occurs in the wet seasons. Similarly, central Africa (AF) and Southeast Asia (SE) are two important tropical sources, contributing to 2.2 and 2.0 Tg·yr−1IC-SOA, respectively. In the northern hemisphere except IN, IC-SOA production generally follows a similar seasonal variation pattern, highest in summer and lowest in winter. The summer to winter production ratio (SWR) is highest in the Former Soviet Union (FSU) region (∼39). While in North America (NA), Europe (EU), and East Asia (EA), the SWR ranges 4–7, indicating more VOCs are emitted in winter seasons than FSU. Over the Indian subcontinent, production of IC-SOA is related to the Indian monsoon. Production in the monsoon and post-monsoon seasons (June–November) is a factor of 4 higher than that in winter and pre-monsoon seasons (December–May). In the Middle East (ME), due to the lack of both cloud and vegetation emissions, IC-SOA production is very low, only 0.14 Tg·yr−1.
 Among the six IC-SOA species, oxalic acid is usually the dominant component (Figure 6) and glyoxylic acid ranks the second. However, in boreal winters, production of oxalic acid is greatly depressed over mid- and high-latitude regions (e.g., NA, EU, FSU and EA). This is because the production rate of oxalic acid (i.e., Ra052 inTable 1) depends strongly on temperature. For instance, the conversion rate from glyoxylic acid to oxalic acid at 20°C is a factor of 10 faster than at −20°C. As a result, the ratio of production between oxalic acid and glyoxylic acid in most places is around 4–6 in summer, but only 0.5–2 in winter. Over EU and FSU, the wintertime production of IC-SOA is dominant by oligomers formed from glyoxal since the yield of oligomers upon cloud evaporation is assumed to be fixed throughout the year. Therefore, more laboratory experiments are needed to explore the oligomers' yield under a variety of atmospheric conditions (e.g., cold and warm seasons).
Figure 7shows the distribution of atomic oxygen-to-carbon (O/C) ratios (characterizing the oxidation state of OA) of the in-cloud produced SOA. The O/C ratios of glycolic acid, glyoxylic acid, pyruvic acid and oxalic acid are 1.5, 1.5, 1.33 and 2, respectively. However, for oligomers, the O/C ratios are variable depending on precursors as well as the possible water loss during reactions [De Haan et al., 2009; Lim et al., 2010]. Here we assume oligomers formed from glyoxal and methylglyoxal are 0.83 and 0.5, respectively, based on the lower bounds estimated in De Haan et al.  and Lim et al. . The actual O/C ratios of these oligomers could be larger. As shown in Figure 7, the O/C ratios of IC-SOA usually range from 1.2 to 2.0 globally, much larger than these of primary OA or the organic aerosols formed in traditional smog chamber experiments [Aiken et al., 2008]. The high O/C ratios are mostly caused by the high share of oxalic acid. As shown in Figure 8, oxalic acid usually dominates the total IC-SOA production, particularly above 850 hPa (due to that the rapid decay of glyoxal and methylglyxoal at lower altitudes depresses in-cloud oligomers' formation at higher altitudes), consistent to the findings from aircraft measurements that oxalate is the largest contributor to above-cloud organic acids [Sorooshian et al., 2007]. These findings indicate that IC-SOA (mostly the oxalate), could significantly enhance the above-cloud O/C ratios globally.
 This study quantifies the global in-cloud production of secondary organic aerosols in the GFDL coupled chemistry-climate model AM3. A detailed cloud chemistry mechanism in conjunction with a gas-liquid mass transport model is employed and merged into the original gas-phase chemistry in AM3. In addition, climate model predicted cloud information, such as cloud water amount, cloud fraction and cloud droplet size, is passed online to the cloud chemistry scheme. Four carboxylic acids (i.e., glycolic acid, glyoxylic acid, pyruvic acid, oxalic acid) as well as two types of oligomers (formed from glyoxal and methylglyoxyal upon cloud evaporation) are simulated and their net chemical productions at the time cloud evaporates are recorded as IC-SOA production.
 Based on the IPCC-AR5 emission inventories and the observed SST and sea ice cover, we found that global in-cloud production of SOA is around 23 Tg·yr−1during the years around 2000, similar to the magnitude of gas-phase SOA production (20–30 Tg·yr−1) [Lack et al., 2004; Heald et al., 2008; Henze et al., 2008]. Significant IC-SOA production occurs from surface to about 500 hPa and peaks near 900 hPa, overlaying the layer of maximum cloud water content. Seasonal variation of IC-SOA production is determined by both VOCs emissions and availability of liquid cloud water. In DJF, more than 70% of IC-SOA is produced in the southern hemisphere, particularly over central and southern Africa and South America. Conversely in JJA, more than 70% of IC-SOA is produced in the northern hemisphere, mostly distributed over the boreal forest regions as well as southern China and northern India.
 Among the six IC-SOA species, oxalic acid dominates IC-SOA production in most locations and during most seasons, typically in the range 40–90%. Global total in-cloud production of oxalic acid is estimated to be 13.6 Tg·yr−1 (close to the estimate from Myriokefalitakis et al. ), followed by glyoxylic acid (3.5 Tg·yr−1), oligomers formed by glyoxal (2.6 Tg·yr−1, a factor of 1.6 higher than the estimate from De Haan et al. ), pyruvic acid (1.4 Tg·yr−1), glycolic acid (1.22 Tg·yr−1), and oligomers formed by methylglyoxal (1.17 Tg·yr−1). The most important water-soluble precursor accounting for IC-SOA formation is glycolaldehyde, of which nearly 151 Gmol·yr−1 (or 9 Tg·yr−1) is absorbed by cloud droplets and converted to glyoxal and glycolic acid and finally oxalic acid in the liquid phase. Acetic acid ranks second, with nearly 116 Gmol·yr−1 (or 7 Tg·yr−1) gas-phase acetic acid being absorbed by cloud droplets and converted to glyoxylic acid and oxalic acid in the liquid phase. Methylglyoxal and glyoxal are another two important IC-SOA precursors, with nearly 57 Gmol·yr−1 and 27 Gmol·yr−1captured by cloud, respectively. However, due to the low Henry's law constant, contribution of hydroxyacetone to IC-SOA formation is small. Since oxalic acid usually dominates IC-SOA production, particularly at higher altitude, the O/C ratios of IC-SOA are typically in the range of 1.2–1.9 in most locations, potentially explaining the discrepancy between the observed O/C ratios and these of primary OA or smog chamber measurements.
 Quantification of IC-SOA production depends on variety of assumptions embedded in cloud chemistry calculation. To understand the effects of cloud types and properties on IC-SOA production, several sensitivity simulations are conducted. Reducing cloud lifetime from 30 min to 10 min enhances global IC-SOA production by 20%. Except oxalic acid, productions of other five IC-SOA species are increased by 40–160%. Since cloud evaporation frequencies are tripled, oligomers formed from glyoxal and methylgloxal upon cloud evaporation are significantly increased (+110%). As a result, cloud absorption of glyoxal and methylglyoxal is increased by 80% and 45%, respectively. The cloud absorption of glycolaldehyde and acetic acid changes little (∼+5%). IC-SOA production is inversely related to cloud droplet size since a smaller droplet radius results in a larger cloud surface area which enhances gas-liquid transfer. Assuming a fixed cloud droplet size, a factor of 4 increase in cloud droplet radius from 5 um to 20 um only reduces the IC-SOA production by 15%, indicating IC-SOA production depends only weakly on cloud droplet size. In addition, IC-SOA production based on model predicted cloud radii is almost identical to that of using fixed 10μm radius. Moreover, adding more cloud water from shallow cloud or large-scale ice cloud enhances IC-SOA production, but the magnitude is relatively small (≤2.5 Tg·yr−1). Cloud entrainment determines the exchange of air between cloudy and non-cloudy area in each model grid, and thus the amount of gas-phase species exposed to cloud. Assuming cloud fraction to be 1 in all cloudy grids leads to 25% more IC-SOA production, indicating the treatment of cloud fraction is important for aqueous-phase chemistry in global chemical transport models or climate models.
 In sum, in-cloud production of SOA is potentially an important source of organic aerosols. The magnitude of mass concentrations is comparable to the predicted gas-phase SOA production. However, our results are subject to multiple sources of uncertainties, such as IC-SOA precursors, cloud chemistry mechanism, cloud amount/cover, and cloud properties (e.g., AM3 tends to underestimate cloud liquid water path by 5–15% over midlatitude North Pacific and even more over the tropical Pacific), as well as the assumptions for cloud activation and evaporation. Future laboratory research in the areas of cloud activation and evaporation chemistry at variety of atmospheric conditions, modeling work improving cloud prediction and gas and aerosol chemistry, as well as species based organic aerosol measurements, could help to reduce the uncertainties associated with the estimation of IC-SOA source.
 We thank Barbara Turpin for helpful suggestions on aqueous-phase mechanism of SOA and three anonymous reviewers for thoughtful comments. We also thank Renyi Zhang and Zhongming Chen for useful discussion on SOA cloud chemistry. We finally thank the Geophysical Fluid Dynamics Laboratory for computational resources. This paper is also partially supported by the National Natural Science Foundation of China (41130754).