The present experiments were performed in the CESAM (French acronym for Experimental Multiphasic Atmospheric Simulation Chamber) atmospheric simulation chamber, which has previously been described in detail by Wang et al. (2011). In brief, CESAM is a stainless-steel chamber with a volume of 4.2 m3. Chamber illumination is accomplished using the borosilicate-filtered output of three high-pressure arc xenon lamps (4 kW, XPO 4000 W/HS, OSRAM), which provides a good reproduction of the solar energy distribution at the Earth's surface over the 290–700 nm wavelength region. The inner walls of the chamber are polished in order to provide good reflection inside the chamber and thus enhance the radiation homogeneity. During our experiments, the NO2 photolysis frequency JNO2 within the chamber was approximately 3 × 10-3 s-1, which corresponds to a solar zenith angle of ∼ 70∘ (Carter et al., 2005). The simulation chamber was maintained at room temperature (±1 ∘C) using a refrigerating liquid (70 % water / 30 % ethylene glycol), which was circulated in the double walls of the chamber. Temperature and relative humidity (RH) are monitored with a transmitter (HMP234, Vaisala) equipped with a thin-film capacitive humidity sensor (HUMICAP^®, Vaisala). During our experiments, the temperature accuracy is ±0.1 ∘C at 20 ∘C and the RH accuracy is ±1.9 % (up to 90 % RH).

Prior to each experiment, the chamber was evacuated to a secondary vacuum (typical pressure ∼ 4 × 10-4 mbar) and kept under vacuum overnight. The chamber was then filled to atmospheric pressure with a mixture of 200 mbar of oxygen (Air Liquide, ALPHAGAZ^™ class 1, purity 99.9 %) and 800 mbar of nitrogen produced from the evaporation of a pressurized liquid nitrogen tank (Messer, purity > 99.995 %, H2O < 5 ppm). The background conditions were typically: particles concentration < 0.1 µg m-3, ozone mixing ratio < 5 ppb, gas-phase organics < 5 ppb and RH < 1 %. To avoid contamination, a slight overpressure of about 5 mbar with respect to the atmospheric pressure was maintained during each experiment by adding nitrogen (Messer, purity > 99.995 %, H2O < 5 ppm).

All aging experiments were carried out under the same initial conditions with α-pinene and ozone reacting in the dark, with neither seeds nor OH scavenger. O3 was generated in an O2 flow using a commercial dielectric ozone SOA generator (MBT 802N, Messtechnik GmbH, Stahnsdorf, Germany) and introduced to the chamber through an injection port. Quantification of α-pinene was performed by evaporating precisely measured amounts of the terpene into a glass bulb held under vacuum. When the O3 concentration within the chamber reached ∼ 250 ppb, α-pinene (Aldrich, 98 %) was flushed from the bulb into the chamber in a flow of oxygen to a concentration of ∼ 200 ppb within the chamber. In all experiments, SOAs, formed directly after α-pinene injection, and its precursors were essentially consumed after 4 h of reaction.

In our companion paper, we observed changes in the oxidative degree and optical properties during the formation of SOAs, but these changes ceased after 9 h of reaction (Denjean et al., 2014b). In the present study, therefore, SOAs were allowed to evolve for 14 h prior to simulating atmospheric processing, which was in turn conducted over a period of 6 h. Three different aging regimes were applied after 14 h of α-pinene ozonolysis: (1) 700 ppb of ozone was introduced into the simulation chamber, (2) SOAs were exposed to light for 6 h in the absence of temperature control, during which time the temperature increased by ∼ 6 ∘C and (3) SOAs were exposed to light for 6 h under controlled temperature conditions (±1 ∘C). For comparison purposes, control experiments were performed: in these experiments, SOAs were allowed to remain in the chamber for 20 h under dark conditions. The specific experimental conditions associated with each aging regime are shown in Table 1. As discussed in our companion paper (Denjean et al., 2014b), the variation of SOA mass concentration between the experiments was attributed to different concentrations of α-pinene injected to the chamber. Despite this, the chemical, optical and hygroscopic properties of SOA were found to be very similar between the experiments. Therefore, the experiments were considered to be comparable for studying the effect of forcing on SOA properties.

During these experiments, the concentrations of α-pinene, ozone and VOCs were monitored using a Fourier transform infrared spectrometer (FTIR) from Bruker GmbH (Ettlingen, Germany) coupled to a multi-reflection cell with an optical path of 192 m. During the different forcings, the concentrations of gas-phase compounds were found to be below the detection limit. Ozone was also measured with a commercial instrument (Horiba APOA 370, Kyoto, Japan), with a detection limit of 0.2 ppb and a precision of 0.1 ppb.

In order to assist in the interpretation of experimental results, the α-pinene ozonolysis experiments were simulated using a box model that included the MCM (version 3.1) (Saunders et al., 2003) and the Generator for Explicit Chemistry and Kinetics of Organics in the Atmosphere (GECKO-A) (Aumont et al., 2005). The MCM v3.1 oxidation scheme for α-pinene contains 329 organic species and 973 reactions. The GECKO-A chemical scheme for α-pinene oxidation, which provides a more detailed description of the gaseous oxidation of organic species and takes into account minor reaction pathways not considered by the MCM, involves 5.7 × 104 organic species reacting according to 1.7 × 106 reactions (Valorso et al., 2011). Both models simulate gas–particle partitioning in terms of equilibria between the gas phase and an ideal liquid homogeneous condensed phase (Camredon et al., 2007).

Vapour pressures and boiling points for secondary organic species were estimated using the methods developed by Nannoolal et al. (2004, 2008), as they have been shown to provide the most reliable estimates for the purpose of SOA formation (Barley and McFiggans, 2010). Accretion, oxidation and photolysis in the condensed phase were not taken into account in this investigation in the models. In addition, recent studies (Renbaum-Wolff et al., 2013; Saukko et al., 2012; Denjean et al., 2014b) suggest that α-pinene-O3 SOA viscosity increases with aging. Due to possible kinetic limitations, heterogeneous chemistry and particle-phase reactions can be limited in highly viscous aerosol. The effect of viscosity on rate constants were not taken into account in the current MCM and Gecko-A models.

Time integration of the chemical schemes was solved using the two-step solver (Verwer et al., 1994, 1996). The gas–particle partitioning module was solved using the iterative method described in Pankow (2008). Simulations were initialized at a time point corresponding to α-pinene injection. This injection was implemented in the model as a constant flux over the injection time period that reproduced the observed concentration. Dilution within the chamber (arising from the periodical injection of N2 to compensate for instrumental sampling flows) was described as a measurement-constrained first-order process and applied in the simulations to both gas and aerosols. The temperature used in the simulations was that observed in the CESAM chamber during the experiments. Ozone-wall loss is significant in the CESAM chamber (Wang et al., 2011). The ozone loss rate employed in the simulations, therefore, was adjusted for each experiment in order to reproduce its measured decay.

Figure 1 shows the temporal evolution of SOA mass concentrations (normalized to the mass concentration at the beginning of each aging experiment and corrected for dilution within the chamber) during and after simulated atmospheric processing. As shown in this figure, the SOA mass concentrations in the control experiments decreased by ∼ 15 % over the 6 h duration of the experiment. Since these measurements were performed after the total consumption of O3, this mass decrease cannot be attributed to O3-induced fragmentation reactions. It is thus probable that the SOA mass decrease observed in the control experiments arose via losses of particle and gaseous compounds to the chamber walls. Similar behaviour was observed when SOA was exposed to ozone and to light under controlled temperature conditions, which suggests that these forcings did not lead to significant fragmentation or functionalization. On the contrary, exposure of SOA to light and increasing temperature led to a 40 % loss in total SOA mass concentration and, as shown in Fig. 2, to a change in the SOA number size distribution: as the temperature increased from 20 ∘C to 26 ∘C, the geometric mean diameter of the normalized SOA number size distribution decreased from 286 nm to 249 nm, which also indicates that significant evaporation of SOA particles occurred.

Since the CRI is related to the aerosol chemical composition, density, molecular weight and polarization (Liu and Daum, 2008), it was thus expected to be influenced by atmospheric processing. The imaginary part of the SOA CRI over the 370–950 nm wavelength range is shown in Fig. 3. Its value was almost zero at all the wavelengths studied, which indicates that SOA, even after simulated atmospheric processing, has a pure scattering effect in the visible to near-UV region. The formation of chromophores has been observed previously from the ozonolysis of biogenic terpenes (Bones et al., 2010; Laskin et al., 2010; Zhong et al., 2012). Some oligomers formed in particle-phase can contain long conjugated structures which increase the light absorption properties of SOA in the visible. Interestingly, while one might have expected the formation of chromophores during the aging processes, our results indicate that this phenomenon occurred only weakly for α-pinene–O3 SOA and/or that the specific absorption of any potential aging products was not significant. Unlike toluene (Nakayama et al., 2010) or limonene (Bones et al., 2010), α-pinene–O3 SOA products did not contain long conjugated structures which absorbed visible radiation, at least under our experimental conditions when conducted in the absence of NOx or seed particles.

The influence of atmospheric processing upon the real part of the SOA CRI is shown in Fig. 4. No discernible change in real CRI was observed when the SOA was exposed to O3 or to light at constant temperature. On the contrary, a significant increase in the real CRI was observed, 1.35 (±0.03) to 1.49 (±0.03), after photolytic forcing with increasing temperature. The hygroscopic properties of SOA were studied by measuring the GF at a constant relative humidity (90 ± 1 %). As shown in Fig. 5, the GF remained constant (∼ 1.04 ± 0.02) both during O3 exposure and during photochemical aging at constant temperature. These results are consistent with our observations of constant mass concentration and CRI of SOA, and confirm that no significant processing of the SOA occurred under these conditions. In contrast, a significant increase in the GF, from 1.04 (±0.02) to 1.14 (±0.02), was observed with increasing temperature.

The results presented in the previous two sections suggest that the exposure of the aging of SOA to light and increasing temperature resulted in changes in its physical, optical and hygroscopic properties, which are likely to be linked to changes in its chemical composition. Indeed, as shown in Fig. 6a, the increase in the O : C ratio of bulk SOA during its exposure to light and increasing temperature, 0.55 (±0.16) to 0.59 (±0.18), was much larger than that observed for SOA under control conditions. The increase in the O : C ratio was, however, within the measurement uncertainties (±30 %) estimated from Aiken et al. (2007). These uncertainties may be overestimated compared to the experimental variability and even experimental reproducibility observed in this study. In fact, we estimated the experimental uncertainties to be ±0.01 from the standard deviation of the experimental values before the forcing. A new parameterization of the O : C ratio derived from the AMS was recently presented by Canagaratna et al. (2014). Application of this new parameterization to our measurements resulted in a O : C ratio similar to that obtained with our modified fragmentation table in which the measured fragment H2O+ was assigned to organic compounds. This indicates that the two methodologies agree under very dry conditions (RH < 1 %). The average O : C ratio obtained with both parameterizations was 20 % larger than that obtained with the Aiken method and can be attributed to an underestimation in the Aiken method of the CO+ and H2O+ ions produced from many oxidized species.

The fragments f44, defined as the ratio of the m/z 44 (a major fragment of organic acids and hydroperoxides) signal to the total organic aerosol signal, and f43, defined as the ratio of m/z 43 (associated with less oxygenated groups, e.g. aldehydes and alcohols) signal to the total organic aerosol signal, have been widely used in laboratory and field studies as indicators of SOA functionality and degree of oxidation (Ng et al., 2010; Alfarra et al., 2013; Poulain et al., 2010; Pfaffenberger et al., 2013). As shown in Fig. 6b, in the experiment performed under illumination but in the absence of temperature control, the f44 / f43 ratio of SOA increased from 1.73 (±0.03) to 2.23 (±0.03) as the chamber temperature increased. This result implies a temperature-mediated increase in particle-phase oxidized species. On the contrary, only a small increase in the f44 / f43 ratio, from 1.93 (±0.03) to 2.03 (±0.03), was observed during the control experiment, which implies that the SOA composition in this experiment remained relatively constant.

During the aging of SOAs by photolysis with increasing temperature, we observed a decrease of the mass concentration (Fig. 1) which was associated with an increase of the f44 / f43 ratio of bulk particles (Fig. 6b). These evolutions may result from possible shifts of some semi-volatile organics to the gas phase as temperature increases. A second possible explanation would be the photochemical reactions that occur in the condensed phase combined with the evaporation of less oxidized compounds. In this case, photolysis could lead to a loss of semi-volatile and less oxidized compounds in the particle phase due to the fragmentation of condensed-phase species (Donahue et al., 2012; Henry and Donahue, 2012).

In order to investigate if the increase in temperature can explain the SOA mass decrease, the SOA formation and evolution were modelled with the detailed chemical schemes GECKO-A and MCM. In these models, the simulated SOA formation started at the initial step of α-pinene ozonolysis. The effects of temperature on phase partitioning were simulated with both models. The effects of photochemical reactions and heterogeneous reactions by O3 were ignored in these simulations. As shown in Fig. 7 for the initial SOA formation period, the time profiles of α-pinene and ozone concentrations were well reproduced by both models within the measurement uncertainties. Although the temporal profile of the SOA formation was also well reproduced by the models, the simulated mass concentrations of SOAs obtained using the GECKO-A and MCM models (302 and 337 µg m-3, respectively) were higher than those observed in the experiment (130 µg m-3). This overestimation of the simulated SOA mass concentration might be due either to an underestimation of the vapour pressure with the Nannoolal method used in this study or some oxidation processes occurring in the aerosol phase and leading to fragmentation not implemented in the model. In addition, an unavoidable consequence of simulation chamber measurements is the interaction of gases with the chamber wall surfaces. It is plausible that the VOC-wall losses affect the SOA mass concentration by being a sink for semi-volatile products preventing them from condensing on the aerosol phase.

Four experiments, each of which exhibited different temperature increases, were simulated using these models. SOA mass concentration loss was calculated as the difference between the mass concentration before and after each forcing. For the simulations, only the effects of the temperature on phase partitioning were simulated with both models. The effects of photochemical reaction on the gas and particle phase were omitted. A comparison between the observed and modelled loss of SOA mass concentration under these temperature conditions is presented in Fig. 8a. As shown in this figure, both the temporal profile and magnitude of SOA mass loss observed in these experiments were well reproduced by the GECKO-A and MCM models. This finding suggests that the observed heating-induced loss in SOA mass concentration can be explained by the temperature-dependent gas–particle phase partitioning of the semi-volatile components of SOA.

The measured and simulated O : C ratios for these experiments are shown in Fig. 8b. Before the forcing, both models predict an O : C ratio of 0.49, which is within the uncertainty of the experimental value (0.55 ± 0.16) obtained with the AMS. During the aging experiments, the O : C ratio simulated with the two models increased with increasing temperature; for example, experiment E200411, which showed the highest temperature increase (+6.5 ∘C; red in Fig. 8c), also showed the highest increase in simulated O : C ratio (from 0.49 to 0.51); on the contrary, experiment E060512 exhibited constant values for both temperature and simulated O : C ratio (yellow in Fig. 8b and c). These results suggest that the changes in SOA physical and chemical properties observed in these experiments resulted from the heating-induced evaporation of semi-volatile and less oxidized SOA species, which in turn modified the optical and hygroscopic properties of the condensed phase.