Secondary organic aerosol (SOA), formed when oxidized products of volatile hydrocarbons condense, often comprises a substantial portion of the organic mass fraction of atmospheric aerosols. The prevalence of organic carbon aerosol on a global scale makes identifying significant sources of SOA an important task, as carbonaceous aerosol is known to strongly influence air quality and climate change. Model predictions of organic carbon aerosol concentrations have exhibited a low bias not present in coincident predictions of black (elemental) carbon, with this bias being attributed to under-prediction of SOA [Heald et al., 2005; Tsigaridis and Kanakidou, 2003].
 Isoprene (C5H8) is the second most abundant hydrocarbon emitted into the Earth's atmosphere after methane (∼500 Tg yr−1 [Guenther et al., 1995]). Although it has long been assumed that all its products remain in the gas phase, if isoprene were to yield even a small amount of aerosol, this would have a profound effect on global sources of organic aerosol. Biogenic volatile organic compounds other than isoprene, such as terpenes and sesquiterpenes, are presently believed to be the largest source of SOA mass on a global scale, with model estimates of the magnitudes of these sources ranging from 12–70 Tg yr−1 [Kanakidou et al., 2005]. Recent laboratory chamber studies of isoprene photooxidation show that SOA yields are 1–2% at high NOx levels [Kroll et al., 2005] and ∼3% at low NOx levels [Kroll et al., 2006]. Furthermore, organic aerosol collected in forested areas is strongly indicative of an isoprene precursor [Claeys et al., 2004a, 2004b; Ion et al., 2005; Kourtchev et al., 2005; Matsunaga et al., 2003]. The impact of such a potentially large source of carbonaceous aerosol necessitates careful investigation of the fate of isoprene oxidation products on a global scale.
 Claeys et al. [2004b] estimated SOA production from isoprene to be 2 Tg yr−1 by simply multiplying an estimate of global isoprene emissions by an observed yield of condensed polyols from isoprene; subsequent recognition of additional SOA production pathways increases this estimate [Claeys et al., 2004a]. Cloud processing of isoprene oxidation products alone has been calculated to contribute 1.6 Tg yr−1 of SOA [Lim et al., 2005]. Matsunaga et al.  estimated a source of SOA from isoprene in the range of 10–120 Tg yr−1; however, this study neglects the effects of temperature and background organic particulate matter concentrations on gas - particle partitioning, factors known to strongly influence SOA formation.
 Recent availability of data from laboratory chamber studies of isoprene oxidation [Kroll et al., 2005, 2006] allows us to now assess the global SOA forming potential of isoprene in a more fundamental manner. Several factors influence SOA formation, such as the ambient NOx concentration, RO2 concentration, temperature, and heterogeneous reactions [Limbeck et al., 2003; Czoschke et al., 2003; Edney et al., 2005; Kroll et al., 2006]. Until the mechanisms that govern these types of behavior are precisely known one must use empirical parameterizations based on actual laboratory data [Kanakidou et al., 2005].
 For inclusion of SOA in global models, the framework of the two-product model [Odum et al., 1996; Seinfeld and Pankow, 2003] provides a method for predicting the formation of SOA based upon empirical parameters determined from laboratory chamber studies even when the exact chemical nature of the aerosol products, or even the intermediate gas-phase oxidation products, are not known [Griffin et al., 1999b]. The model describes the oxidation of a parent hydrocarbon to produce two representative gas-phase products with stoichiometric coefficients α1 and α2. Subsequent partitioning of these products into the aerosol phase is governed by the availability of pre-existing organic aerosol and by their equilibrium partitioning coefficients, K1, K2, taking into account the temperature dependence of the partitioning coefficients using the Clausius-Clapyeron equation. At the moment condensation onto other (non-organic) aerosol species is not considered, though this would afford increased SOA formation from all species [Tsigaridis and Kanakidou, 2003].
 We simulate global SOA formation using the chemical transport model GEOS-Chem (version 7.2.4 with a horizontal resolution of 4° × 5° and 30 layers up to 0.01 hPa, GEOS-3 meteorological fields [Park et al., 2004]), previously implemented with a gas-particle partitioning model of SOA formation from terpenes [Chung and Seinfeld, 2002; Heald et al., 2005], updated here to include formation of SOA from oxidation of isoprene using parameters shown in Table 1. The α's and K's were derived from the final amount of SOA formed in chamber studies of isoprene oxidation by OH [Kroll et al., 2006] using the same method as Griffin et al. [1999a]. We assume reaction with OH is the only pathway for formation of SOA from isoprene. Though reaction with O3 or NO3 may also lead to SOA formation, the magnitudes of these sources are assumed to be minor, because an order of magnitude more isoprene reacts with OH than with O3 or NO3 on a global scale [Calvert et al., 2000]. We assign a molecular weight of 130 for the oxidation products from isoprene, which is that of tetrol, a compound prevalent in SOA that originates from isoprene [Claeys et al., 2004b].
|Product||αi||Ki, m3 μg−1b|
 An issue with empirical partitioning models is that the conditions of the chamber studies from which the yield parameters are derived may not be representative of atmospheric conditions. The main concern has been that NOx levels in these experiments tend to be larger than those in the troposphere. The experiments used to derive the yield parameters for isoprene given here were carried out under low NOx concentrations (<1 ppb) and at cooler temperatures more relevant to tropospheric conditions [Kroll et al., 2006]. Still, a single set of yield parameters may not fully represent SOA formation throughout the entire range of conditions present in the atmosphere—further laboratory and modeling studies are required to explicitly specify the dependence the SOA yield parameters on the chemical environment.
 Implementation of this model on global scales requires knowledge of thermophysical parameters that are not easily determined experimentally. The enthalpy of vaporization of SOA, ΔHv, is critical for extrapolating the equilibrium gas-particle partition coefficients to colder temperatures [Tsigaridis and Kanakidou, 2003]. The value of ΔHv depends upon the nature of the SOA and how it was formed [Offenberg et al., 2006], though there is not yet enough experimental data available to justify the use of more than a single value of ΔHv for all SOA. The base case value of ΔHv used here, 42 kJ mol−1 [Chung and Seinfeld, 2002], originally considered a lower estimate in comparison to values from similar studies which ranged as high as 156 kJ mol−1, is perhaps in fact quite reasonable, as recent experimental studies of the temperature dependence of SOA formed from α-pinene have placed ΔHv closer to the lower estimates [Offenberg et al., 2006; C. O. Stanier and S. N. Pandis, Measurements of the volatility of aerosols from alpha-pinene ozonolysis, submitted to Environmental Science and Technology, 2006]. The sensitivity of SOA predictions to the aqueous solubility of the oxidation product species, governed by an estimated average Henry's law constant of the oxidation products, H, has also been mentioned by Tsigaridis and Kanakidou , though the consequences of variations in H on global SOA predictions have not yet been explored. Loss of these products by wet removal depends strongly on H. Given that polyols resulting from isoprene oxidation are more soluble than many of the previously identified species in SOA, which were taken to have an average Henry's law constant of 105 M atm−1 (R. Sander, Compilation of Henry's law constants for inorganic and organic species of potential importance in environmental chemistry (version 3), http://www.mpch-mainz.mpg.de/∼sander/res/henry.html), we consider the effect of increasing the Henry's law constant of the oxidation products to 106 M atm−1, and, for comparison, decreasing it to 104 M atm−1.
 Model predictions of global yearly average SOA concentrations for the year March 2001–February 2002 are shown in Figure 1. We select this time period because it encompasses the ACE-Asia campaign, for which the observed amount of organic carbon aerosol in the free troposphere exceeds predictions by the base case model by a factor of 10–100 [Heald et al., 2005]. Figures 1a and 1c show the total concentrations of SOA generated by the existing (base case) biogenic VOCs (terpenes and OVOCs) at the surface and at 5.2 km, respectively. Figures 1b and 1d show total SOA concentrations when isoprene is included as an additional source of SOA. The difference between these two simulations is striking, most notably in the magnitude of the increases in the free troposphere, where typically more than 70% of the SOA is from isoprene. SOA concentrations increase by a factor of 1.5 to 3 in regions of relatively high SOA concentrations, and they increase by more than a factor of 10 in remote marine regions where SOA concentrations are small (<0.01 μg m−3), such as the Indian and South Central Pacific oceans.
 The yearly average total SOA burden (BT) and the net yearly SOA production (PT) are given in Table 2, where the total production is also broken down into contributions from isoprene (PI) and from the original set of VOCs (PO). The amount of SOA produced directly from isoprene is 6.2 Tg yr−1, almost as large as the original SOA source in the base calculation, 8.7 Tg yr−1. The presence of this much additional organic substrate enhances SOA formation from other sources by 17%. The total SOA burden more than doubles, and the lifetime of the SOA from isoprene (13.5 d) is twice that of the base case SOA (6.7 d). Results from a one month simulation with model resolution of 2° × 2.5° were equivalent.
|Sourcea||ΔHv, kJ/mol||H, M/atm||PO, Tg/yr||PI, Tg/yr||PT, Tg/yr||BT, Tg|
|O + I||42||105||10.2||6.2||16.4||0.39|
|O + I||42||106||10.1||6.1||16.2||0.38|
 Two factors give rise to the distinct distributions and lifetimes of the SOA formed from isoprene compared to the base case set of VOCs. Emissions of isoprene are generally much greater. As a result, isoprene is not completely oxidized near its sources, and substantial amounts of isoprene can be lofted to much greater altitudes. Also, gas-particle partitioning of the isoprene oxidation products is shifted less toward the particle phase than that of the products of the base case VOCs; hence, the lifetime of the isoprene oxidation products is also greater. The combined effect of these factors increases SOA precursor concentrations in the free troposphere where partitioning to the aerosol phase is enhanced owing to lower temperatures, leading to formation of SOA in regions where there was little in the base case. Although this increase alone is not enough to account for the discrepancy between predicted and observed tropospheric organic carbon aerosol in the region studied during the ACE-Asia campaign [Heald et al., 2005], it does significantly impact our global picture of organic carbon aerosol distributions.
 The total amount of isoprene predicted to be oxidized by OH is 209 Tg yr−1; the global isoprene SOA “yield” is 2.9%, which is essentially the same as those from the low-NOx chamber experiments (∼3%). We find that simply calculating the formation of SOA from isoprene from a direct calculation (wherein SOA is formed, irreversibly, as a constant percentage of the amount of isoprene that reacts) leads to lower SOA burdens than the two-product model, in contrast to previous studies comparing these methods [Lack et al., 2004; Tsigaridis and Kanakidou, 2003]. The reason for this discrepancy is, as noted earlier, a significant portion of the SOA from isoprene is formed from the semivolatile oxidation products that only condense substantially at lower temperatures, an effect that may not be as critical for modeling SOA from sources with greater yields.
 We examine SOA levels predicted by the base case model (without isoprene as a source of SOA) when using a reasonably larger value of ΔHv = 50 kJ mol−1 or when H = 104 M atm−1. Use of this value of ΔHv leads to a modest increase in the global SOA burden of 0.08 Tg, and average SOA concentrations in the troposphere increase by a factor of 2 to 3. Decreasing H increases the burden by almost 40%. While these are substantial consequences, the overall magnitude of these effects is still small compared to increases of SOA concentrations from isoprene, as shown in Figure 2. When isoprene is included as a source of SOA, increasing H to 106 M atm−1 has little overall effect, as the oxidation products are effectively completely soluble beyond H ≥ 105 M atm−1.
 Including isoprene as a source of SOA causes substantial increases in predicted SOA concentrations, particularly in the free troposphere and remote marine environments. A detailed comparison with measured organic carbon aerosol is now in order. This source of SOA may help explain observations of organic carbon aerosol, noted previously to be under-predicted by this (and others) model in these regions [Heald et al., 2005; Tsigaridis and Kanakidou, 2003], particularly considering recent revisions in estimates of isoprene emissions [Guenther et al., 2006]. This study highlights the need for further research into the chemical fate of the oxidation products of isoprene [Kroll et al., 2006] and the importance of developing SOA models that can explicitly represent condensation of oxidation products normally considered too volatile to contribute to organic aerosol formation [Donahue et al., 2006]. These results may have implications for climate change given the magnitude of the predicted top of the atmosphere radiative forcing of organic carbon in year 2100 climate [Liao et al., 2006].