To investigate the role of gas-phase photolysis on SOA formation and yields, we compared experiments with photolysis on and off for four typical SOA precursors including α-pinene, toluene, and C12 and C16 n-alkanes (Table 2). In the “photolysis-off” experiment, photolysis of all organic species (except for formaldehyde and methyl hydroperoxide) is turned off, whereas all inorganic compounds (i.e., O3, NO2, NO3, H2O2, HONO, HNO3, HNO4) still undergo photolysis. Figure 1 shows the results for two different levels of NOx (0.01 and 10 ppb), OH (2 × 106 and 8 × 106 molecules cm-3) and background OA (1 and 10 µg m-3). Base case simulations (BASE) with and without photolysis of organics are shown in black. The results indicate that in a week of chemical aging with constant daylight or two weeks of equivalent atmospheric summertime exposure (representative of Boulder, 40∘ N, with the solar zenith angle of 45∘), the gas-phase photolysis leads to a 10 to 45 % decrease in SOA concentrations depending on the precursor. Most of the decrease occurs in the first 4 days of the simulation (8 equivalent atmospheric days). For a given species, the sensitivity to gas-phase photolysis is comparable under low- and high-NOx conditions, except for α-pinene SOA, which seems to be more sensitive under high-NOx conditions. Our results also suggest that, for most species, the resulting loss of SOA due to gas-phase photolysis is increasingly more important as the chemical processing time is increased during the first week atmospheric exposure (Fig. S1 in the Supplement).

The largest effect on yields (Table 3) is predicted for α-pinene SOA, with a reduction of 16 % under low NOx and 47 % under high NOx during the entire experiment. The reduction is relatively small during the initial 10 h, which is a typical duration of laboratory experiments, with a 2 % decrease at low NOx (10 % high NOx), and reaches ∼ 5 % (20 %) after 1 day, or 12 % (42 %) after 1 week of equivalent atmospheric ageing. For toluene oxidation, the effect of photolysis is more limited, and does not exceed 15 % for either low- or high-NOx conditions. We note, however, that the effect could be underestimated as the photolysis of nitroaromatic compounds, which are strong absorbers, is currently not represented in GECKO-A. For products of n-alkanes, the relative decrease in SOA yields is ∼ 10 % for C16 n-alkanes and ∼ 25–30 % for C12 n-alkane. The sensitivity to gas-phase photolysis is more important for C12 than C16 n-alkane, due to the fact that products of C16 n-alkane are partitioned to the particle phase in a much greater fraction after one generation of chemistry, where they are protected from gas-phase photolysis, whereas it takes several generations to produce substantial SOA from shorter-chain n-alkanes (Aumont et al., 2012).

The efficiency of the gas-phase photolysis will depend on the residence time of organic vapors in the gas phase, which can be significantly modified by the chemical environment. For instance, an increase in the OA mass available for the gas–particle partitioning is expected to enhance the partitioning of organics to the condensed phase, where they will be protected from gas-phase photolysis. Similarly, a faster gas-phase oxidation rate (higher OH) is expected to lead to more rapid generation of the low-volatility organic species which can condense into the particle phase, thus making them less vulnerable to the gas-phase photolysis. Sensitivity simulations with 4-fold higher OH concentrations (SENS_OH) and 10-fold lower background OA (SENS_OA) are performed to evaluate these effects (Fig. 1). As expected, SOA formation occurs more rapidly when a 4-fold increase in OH is considered. The decrease by an order of magnitude in the amount of the preexisting OA (and thus reduced gas–particle partitioning) also affects the amount of SOA formed. A large (∼ 75 %) decrease in SOA production is observed for toluene because a significant fraction of the predicted oxidation products have effective saturation mass concentrations (C*) in the 1–103 µg m-3 range (Fig. 4; see also Hodzic et al., 2014). The effect is more limited for SOA produced from other precursors (up to 30 %). For all precursor species, the sensitivity to photolytic reactions remains qualitatively similar (within 10 %) regardless of the OH and OA background values (Fig. S1 in the Supplement). As expected, a decrease in background OA concentrations leads in most cases to an enhancement of the SOA photolytic loss, whereas an increase in OH levels tends to result in a reduced SOA photolytic removal. We also note that the sensitivity to gas-phase photolysis is not significantly modified when diurnally variable photolysis rates are considered instead of fixed constant daylight conditions (see Fig. S3 in the Supplement).

Figure 2 shows the major functional groups in SOA molecules from various precursors. Fifteen families of functional groups are considered and they account for 54 to 65 % of the total SOA mass for α-pinene, 94 to 99 % for toluene, and for 70 to 90 % for C12 and C16 n-alkanes. Positional isomers are lumped into the same family of compounds. Ketone (K) and alcohol (O) moieties are present in the majority of the molecules, while hydroperoxides (H) are seen mainly at low NOx and nitrates (N) mainly at high NOx. Gas-phase photolysis leads to a decrease in most species, which seems to be particularly important for highly functionalized compounds containing multiple carbonyl and nitrate groups (e.g., HKKKK, HHKK, HHKKK, HKKK, NNKK, NNKO, where HHKK, for example, refers to a molecule that contains two hydroperoxide and two ketone groups). These species are formed by several generations of chemistry and are mainly found in the particle phase. Thus their decrease is more likely related to reductions in their precursor species due to photodegradation than to their direct loss by gas-phase photolysis. Some molecules containing alcohol groups (e.g., HHO, HKKO, HHKO) see an increase in their concentrations due to gas-phase photolysis (see also Fig. S2 in the Supplement). This increase can be explained by photolysis of hydroperoxides, which can lead to the formation of alkoxy radicals that can isomerize to form alcohols. Thus photolysis can contribute to SOA loss and to a lesser extent to its formation. Typically, photolysis of carbonyl compounds (ketone and aldehydes) tends to break the α-carbon bond on either side of the C = O group, leading to smaller more volatile fragments that are less likely to partition to the particle phase. On the other hand, photolysis of hydroperoxides and nitrates leads to elimination of -OH or -NO2, leading to alkoxy radicals, and potentially further functionalization of the carbon skeleton favoring formation of less volatile organic compounds that can partition more readily to the particle.

One of the highly uncertain factors that can influence the composition of SOA at low NOx is the choice of the rate for abstracting the H atoms from the carbon atom that is adjacent to the hydroperoxide (-OOH) group. As discussed in Sect. 2, in this paper we have used a lower activation factor of 3.5 for estimating the rate constant of that process, instead of the GECKO-A default value of 14 (Aumont et al., 2005). As shown in Fig. S4 in the Supplement this change does not affect the SOA production when the gas-phase photolysis of organics is turned off (∼ 8 µg m-3 for C12 n-alkane). However, the composition of SOA formed from n-alkanes is significantly modified, as is the effect of gas-phase photolysis on SOA yields (∼ 2-fold smaller when the value of 3.5 is considered). The main difference is found for HKKKK and HKKK molecules, which are much more abundant when the value of 14 is used. These molecules originate typically from the successive OH reaction, leading to hydroperoxide moieties under low-NOx conditions and their subsequent fast oxidation to a ketone moiety due to a the large activation factor used. Reducing this factor to 3.5 forces the OH to react away from the -OOH group. When the carbon backbone is sparsely functionalized, this increases the rate of production of multifunctional species, in particular multifunctional peroxides (e.g., HHO and HHK). However, when the carbon backbone is highly functionalized, this leads to more fragmentation, because in most cases the OH attack is now next to other functional moieties (e.g., multifunctional ketones).

Figure 3 shows the effect of gas-phase photolysis on oxygen to carbon (O / C) ratios of particles for the BASE run. For all cases, changes in O / C ratios (< 0.05) are minor. Slightly higher O / C ratios at low NOx were found in the presence of gas-phase photolysis as photolyzed fragments are typically smaller and more volatile carbon chains that need to undergo further oxidation to condense into particles. Our results are consistent with chamber studies by Wong et al. (2014) that observed small changes in O / C with an increase in more oxidized compounds (high O / C) in α-pinene SOA due to fast photodegradation of less oxidized particulate organics such as carbonyls. Changes in SOA composition due to gas-phase photolysis can also affect the volatility distribution of oxidized organic compounds. Figure 4 does not show a clear shift in volatility due to gas-phase photolysis but rather suggests that the SOA reduction is happening across a wide range of volatility bins.

Our explicit modeling results suggest that gas-phase photolysis leads in some cases to moderate changes in SOA yields (< 25 % for most precursors; < 45 % for high-NOx α-pinene), and small changes in volatility distribution and O / C ratios over 1 equivalent week of chemical aging in the mid-latitude atmosphere in summer or 2.5 weeks in winter. The implication in terms of SOA atmospheric lifetime is that gas-phase photolysis is a possible sink of intermediate organic vapors and thus SOA, although a smaller sink compared to dry deposition of these gaseous species (Hodzic et al., 2014; Knote et al., 2015). Indeed, the estimated summertime atmospheric lifetimes against photolysis of the SOA from the four precursors considered in our study range from about 10 days for α-pinene under high-NOx conditions (unlikely case) to more than a month for α-pinene under low-NOx conditions or for long-chain n-alkane species. These lifetimes are considerably longer than values reported by laboratory studies (e.g., Henry and Donahue, 2012). Current 3-D models typically represent the oxidation products as lumped surrogate species based on their volatility that can further age by OH oxidation but cannot photolyze due to the undefined chemical structure of those intermediate species. Our results suggest that omitting their gas-phase photolysis will likely result in reasonably small biases in SOA predictions over urban scales. However, errors could be significant at the global scale, in particular in the upper troposphere, where models have the tendency to accumulate SOA due to a less efficient wet removal.

In the GECKO-A simulations described above, once the organic molecules are partitioned to SOA they are protected from gas-phase photolysis. However, these molecules still contain numerous chromophores (Fig. 2) that absorb solar radiation and could undergo photolysis inside the particle. Optical absorption is also likely to be modified by the heterogeneous formation of high molecular weight compounds inside the particle (Graber and Rudich, 2006). As GECKO-A does not include condensed-phase photochemical reactions (or heterogeneous chemistry), the effect of particle-phase photolysis on SOA cannot be calculated directly. In this section, we consider a simple alternative empirical approach to examine the potential effect of particle-phase photolysis within GECKO-A.

We base our estimates of condensed-phase photolysis on the measured SOA mass absorption coefficient (MAC). Organic particles containing photolabile compounds have been shown to efficiently absorb light at actinic wavelengths. Recently measured MAC values range from 0.03 to 0.5 m2 g-1 for laboratory data (Lambe et al., 2013) or from 0.1 to 10 m2 g-1 for ambient urban measurements (e.g., Barnard et al., 2008) in the 300–400 nm wavelength interval. We use those measurements to estimate the condensed-phase photolysis of SOA. We represent the photolytic SOA loss as a first-order reaction, with effective reaction rate coefficient JSOA: JSOA=AF×[MAC]×[mc]×[QY], where AF is the actinic flux (photons m-2 s-1), MAC is the SOA mass absorption coefficient (m2 g-1), mc is the mass of one carbon atom (g) and QY is the quantum yield or the probability that absorbed photons will lead to bond cleavage and the loss of some mass from the particle. We assume that if each absorbed photon leads to the loss of one C atom, the quantum yield is equal to 1. We can scale Eq. (1) to known NO2 photolysis: JSOA=JNO2×AFJNO2×[MAC]×[mc]×[QY] The photolysis model TUV (v5.1; Madronich et al., 1993) was used to estimate the UV actinic flux (= 2 × 1020 photons m-2 s-1) and NO2 photolysis (= 9.7 × 10-3 s-1) over 280–400 nm at 1 km altitude and overhead sun, which combined with [mc] gives AFJNO2×mc=0.4 photons g m-2. Thus the resulting SOA photolysis rate can be written as JSOA= 0.4×JNO2×MAC×[QY]. To estimate the plausible range of JSOA values, we use the combinations of [MAC] × [QY] reported in the literature. Here we use MAC of 0.1 m2 g-1 as a lower limit for ambient aerosols. We note that for the chosen MAC value the light can penetrate the whole volume of the particle without being significantly attenuated by the absorbers (light attenuation was estimated to be less than 2–3 %; see discussion in Supplement Annex IV). QY has only been measured for a handful of species. Calvert and Pitts (1966) reported values of 0.01 (or 1 %) for photolysis of aldehydes in the aqueous phase. Lignell et al. (2013) reported values of 0.5 for cis-pinonic acid, which is one of the constituents of α-pinene SOA, whereas Wong et al. (2014) estimated an effective quantum yield of 1.2 ± 0.2 for the loss of organics in the case of α-pinene SOA photolysis. Given the range of values, here we use a conservative value of 0.01 (or 1 %) for QY. Thus our best estimate for JSOA is 0.04 % of JNO2. This value of JSOA is applied as a first-order removal rate for each photolabile species in the particle phase. The photo-fragments are no longer considered and the condensed-phase photolysis is considered here as an irreversible loss of one carbon atom. We note that this calculated value is 1–2 orders of magnitude lower that those reported by Henry and Donahue (2012), who estimated the photolytic loss of SOA as 2 % of JNO2 (total value of both particle and gas-phase photolysis and JNO2 of 3 × 10-3 s-1). In their experiments, Henry and Donahue (2012) argued that photolysis is more efficient (higher QY) in the gas phase than in the particle phase, where quenching and caging are more likely to occur and could cause rapid recombination of fragments. Therefore a lower QY may be expected in the particles, although it is unclear whether similar molecules are involved in photolysis in the two phases. We also note that photolysis of SOA is assumed to not occur at visible wavelengths (i.e., QY = 0 for λ > 400 nm).

Figure 5 shows that considering the above estimated condensed-phase photolytic loss of SOA (JSOA= 4 × 10-4 × JNO2= 3.2 × 10-6 s-1; lifetime of 7 days at equivalent JNO2 atmospheric exposure; see Table 1) in GECKO-A simulations leads to a 40–60 % decrease in SOA mass after 10 days of equivalent atmospheric aging for most species (Jmac run, Figs. 5 and S1 in the Supplement). A more limited decrease (15 %) is found for the high-NOx toluene SOA because the photolytic loss of nitroaromatic compounds, which are predicted to be the major SOA constituents (RVVO in Fig. 2), is not included.

For comparison, the effect of applying the gaseous photolysis rates for the corresponding species in the condensed phase is shown in Fig. 5 (Jmolecmax). Similar to the simulations above based on MAC, the photolysis here is again considered as an irreversible loss of organic carbon. The comparison between Jmac and Jmolecmax shows a fairly similar (within 20 %) reduction in SOA mass for most precursors. The Jmolecmax run considers that chromophores and quantum yields are similar in their gas- and particle-phase absorption, which is a crude assumption neglecting possible effects form oligomerization, caging and quenching. In the absence of particle-phase chemistry, Jmolecmax could be viewed as an upper limit to photolysis effects, but changes in absorption with particle aging complicate this simple interpretation as both enhancement and decrease in absorption have been reported (e.g., Shapiro et al., 2009; Bones et al., 2010; Updyke et al., 2012; Zhong and Jang, 2014). In-particle reactions are likely to modify the chemical composition of the particle and therefore change the quantity of photolabile species. For example, the works of Yee et al. (2012) and Schilling-Fahnestock et al. (2014) indicate extensive formation of peroxyhemiacetals in the SOA from dodecane oxidation at low NOx. However, the precursor molecules are third- and fourth-generation products, which contain additional, unfunctionalized ketone groups, which would still be susceptible to photolysis.

The overall SOA loss rate due to the combined effect of gas- and particle-phase photolysis (and ongoing OH chemistry) in GECKO-A runs was estimated for the Jmac simulations (see Table 4). Values range between 3.1 × 10-6 and 5.6 × 10-6 s-1, which translates to equivalent atmospheric SOA lifetimes of 4 to 8 days with regard to photolysis in the summer, except for high-NOx toluene SOA with a lifetime of 20 equivalent summer days, for which the effect of photolysis is likely underestimated in our simulations as discussed above. The estimated SOA lifetime with regard to photolysis is comparable to or even shorter than the typical ∼ 1-week aerosol atmospheric lifetime, which suggests that photolysis may be an important removal mechanism for atmospheric SOA. Atmospheric implications of our findings are further investigated in Sect. 3.3.

The above estimates for the α-pinene SOA photolytic loss rate cannot be directly compared with those of Henry and Donahue (2012) and Wong et al. (2014) due to several factors: for example, (i) the differences in the chemical composition of particles as the experiments typically use the SOA pre-generated by α-pinene ozonolysis for short (a few hours) exposure, whereas in our simulations the SOA is generated mainly by OH oxidation, and over a much longer time period (> 1 week); (ii) the possible evaporation of SOA in the laboratory experiments due to chamber heating under the UV lamps, which was not quantified in those experiments and which does not occur in our model simulations; and (iii) in-particle chemistry that could modify the composition and absorption properties of the SOA and which is not included in our model, and is generally not well understood.

For comparison with previous laboratory work, we ran the model under conditions similar to Henry and Donahue (2012), in which α-pinene was oxidized by ozone in the dark in the presence of hydrogen peroxide, and then the mixture was exposed to UV black lights as in Presto et al. (2005) and/or to OH oxidation. Figure 6 shows the sensitivity of the aerosol mass to assumed photolysis rates for these conditions. The experiment is performed for 5 h at the constant JNO2 of 3 × 10-3 s-1, which corresponds to 4 h of equivalent summertime atmospheric exposure (Table 1). Our results show that when exposed to only OH reactions (photolysis of organics was turned off), SOA concentrations increase by ∼ 35 % in 5 h of aging. On the other hand, SOAs exposed to only photolytic reactions (no OH) decrease by ∼ 7 % over the simulation period. The combined effect of both photolytic and OH reactions is therefore an overall increase in SOA concentrations by ∼ 30 %. As done previously, we also performed a sensitivity simulation in which all molecules can photolyze in both gas and particle phases at the gas-phase rate. Figure 6 shows that, with those reactions, the SOA concentrations are decreased by ∼ 50 % (Jmolecmax) in 5 h. These results suggest that most of the photolabile molecules are partitioned to the particle phase, where they are protected from gas-phase photolysis. The corresponding loss rate due to the combined effect of gas- and particle-phase photolysis and OH oxidation, is 3.4 × 10-5 s-1, which is within a factor of 2–3 of the values reported by Henry and Donahue (2012) and Wong et al. (2014).

We implement the estimated SOA gas-phase and in-particle photolytic loss rates within the GEOS-Chem global chemistry model (Bey et al., 2001) to investigate the regional and global effect of photolytic reactions on SOA concentrations. The GEOS-Chem model configuration used in this study is described in detail by Jo et al. (2013). In particular, SOA is modeled using volatility basis set approach with aging in which oxygenated semi-volatile organic compounds (SVOCs) formed by the gas-phase reaction of nine lumped hydrocarbon species (representing monoterpenes, sesquiterpenes, isoprene, and aromatic compounds) with OH, O3, and NO3 are partitioned between gas and particle phases using six volatility bins (with saturation vapor pressures ranging from 0.01 to 1000 µg m-3 at 300 K). Chemical aging of anthropogenic SVOCs with OH (with a rate constant of 4 × 10-11 cm3 molecules-1 s-1) is assumed to reduce the vapor pressure of the products by 1 order of magnitude. Model simulations are performed for year 2009.

Figure 7 shows the annual mean SOA concentrations predicted by the default GEOS-Chem run within the lower troposphere (below 5 km). The predicted continental background levels of SOA typically vary between 0.2 and 0.4 µg m-3, and the highest concentrations (> 1.5 µg m-3) are found over tropical forest regions of Africa and South America. Industrialized and urban areas in China, Europe and the USA feature SOA values significantly larger (0.5–1.5 µg m-3) than the background. These SOA values and spatial distribution are consistent with previous studies (e.g., Spracklen et al., 2011; Jo et al., 2013). Our results (Fig. 7b) suggest that gas-phase photolysis of intermediate semi-volatile organic compounds leads to a very small (< 4 %) reduction in SOA concentrations under ambient conditions. A much larger impact on SOA concentrations is predicted for in-particle photolysis. When the previously estimated photolytic loss of 0.04 % JNO2 is applied within the GEOS-Chem model, the annual mean SOA concentrations in the lower troposphere are decreased by ∼ 20–30 % over source regions, and up to 60 % over remote regions (Fig. 8a). The absolute decrease is ∼ 0.3 µg m-3 over land and ∼ 0.1 µg m-3 over oceans, with the highest absolute decrease of 0.6 µg m-3 coinciding with the maximum predicted SOA concentrations over Africa (Fig. 7c). As the quantum yield of the particle-phase photolysis and mass absorption coefficients are highly uncertain, here we also consider an order of magnitude higher photolytic loss rate of 0.4 % JNO2. As shown in Fig. 8c, this increase in JSOA results in a larger reduction of SOA concentrations in the lower troposphere reaching 50–70 % over land surfaces, and up to 70–90 % over water surfaces. In both cases, a strong spatial gradient is found between land and water surfaces, with larger relative reductions in SOA concentrations over oceans. This gradient is due to the continuous photolytic losses, the effect of which accumulates further away from source regions. Model results show that the relative decrease in SOA concentrations is ∼ 20–30 % stronger in the upper troposphere compared to the lower troposphere (Fig. 8b, d).

Our results suggest that photolysis of SOA, which is currently ignored in chemistry–climate and air quality models, could be an efficient removal process for organic particles. The diagnosed SOA tropospheric lifetime against photolytic removal (annual-average tropospheric mass burden divided by the annual tropospheric loss due to photolysis) ranges from 1 day for JSOA= 0.4 % JNO2 to 7 days for JSOA= 0.04 % JNO2, and is comparable to the lifetime associated with the SOA wet deposition, which ranges from 3.5 to 5.5 days in these model runs. This photolytic loss pathway is expected to play a particularly important role in regions where wet deposition is not very efficient such as the upper troposphere and lower stratosphere.