The experiments were conducted in the atmosphere simulation chamber SAPHIR (Simulation of Atmospheric PHotochemistry In a large Reaction chamber). SAPHIR is a double-wall Teflon chamber with a volume of 270 m3. Details of the chamber have been previously described (Rohrer et al., 2005; Bohn et al., 2005). The chamber uses natural sunlight for illumination and is equipped with a louvre system, which can be used to simulate dark processes. For the experiments described here, various instrumentations were used to characterize gas phase and the particulate phase species.

Chamber parameters like temperature, relative humidity, flow rate, and photolysis frequencies were also recorded. The actinic flux and the corresponding photolysis frequencies were provided from measurements using a spectral radiometer (Bohn et al., 2005; Bohn and Zilken, 2005).

The number concentration and size distributions of aerosol were measured by a scanning mobility particle sizer (SMPS; DMA model 3081/CPC model 3785, TSI Shoreview, USA) and separate condensation particle counter (CPC; model 3786, TSI) to allow for detection of nucleation particles down to 3 nm.

The chemical composition of aerosol was measured by a high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS; Aerodyne Research Inc., USA). To characterize the degree of oxidation of aerosol, the oxygen to carbon ratio (O/C), and f44 (fractional contribution of m/z 44 to the total organics signal) were obtained from the mass spectra. The O/C (corrected for the minor influence of gaseous components) was derived by the elemental analysis of mass spectra obtained in the high mass resolution mode (W-mode) of the mass spectrometer as described by Aiken et al. (2007) and Aiken et al. (2008). The contributions of gas phase CO2 and water vapor to m/z 44 and to m/z 18, respectively, were characterized using a CO2 and H2O analyzer (Picarro, Santa Clara, USA). The values were subtracted to obtain the particle signals before the elemental analysis (Allan et al., 2004).

Droplet activation and droplet growth were measured using a size scanning CCN method as described previously (Buchholz, 2010; Zhao et al., 2010). This method, also known as Scanning Mobility CCN analysis (SCMA; Moore et al., 2010), has been successfully used in a number of previous studies (Asa-Awuku et al., 2008, 2009, 2010; Padro et al., 2007; Engelhart et al., 2008, 2011). The measurement was done by coupling a differential mobility analyzer (DMA; model 3081, TSI Shoreview, USA) with a cloud condensation nuclei counter (CCNC; Droplet Measurement Technique, USA) and condensation particle counter (CPC3786, TSI). Before entering the instruments, the particles were dried using a silica gel diffusion drier (gradually drying to ∼ 10 % relative humidity, RH) with a residence time of around 3 s. Particles then passed through the DMA and the outgoing air was split into two paths connecting to the CCNC and CPC, which measure the CCN and cloud nuclei (CN) concentrations, respectively. The flow rate of the CCNC is around 0.5 L min-1 with a sheath to aerosol flow ratio of 10. The residence time in the CCN column is around 24 s with the time in the final supersaturation slightly shorter (Lance et al., 2006). The DMA scanned over a size range between 10 and 450 nm while the supersaturation (SS) remained constant. And four to five different supersaturations in the range of 0.1–1.3 % were used depending on the particle sizes. From the measurement, CCN activation fraction over size and the dry activation diameter (or critical dry diameter, Dcrit) were obtained using a method as described in Buchholz (2010). Briefly, for each particle size, the CN and CCN concentrations measured were used to calculate the activation fraction (af=CCN/CN). Before af was calculated, the measured CN and CCN concentrations were corrected for multiple charged particles. To separate the single from the multiple charged particles, the fraction of multiple charged particles was calculated according to a Boltzmann charge distribution using the measured size distribution (Wiedensohler, 1988). Then, af was determined for each charge class separately and fitted with a Gaussian error function (Rose et al., 2008). The dry activation diameter at the set SS is the turning point of this function.

For each SS at least three full scans were performed and the resulting Dcrit were averaged. For the calibration of SS, Dcrit of ammonium sulfate at various SS was measured and compared to theoretical data in the literature (Rose et al., 2008). The set SS was corrected according to the theoretical data. From the CCN data, the hygroscopicity parameter κCCN was calculated according to the one parameter representation of the Köhler equation proposed by Petters and Kreidenweis (2007). The error bars of κCCN were estimated using the standard deviation of Dcrit from three duplicate scans. A higher hygroscopicity parameter κ indicates a more hygroscopic material, i.e., cloud droplet activation at lower SS for particles of a given size or at smaller size for a given SS.

The hygroscopic growth of the aerosol was measured using a home-built hygroscopic tandem differential mobility analyzer (HTDMA). The details of the HTDMA were described previously (Buchholz, 2010; Zhao et al., 2010). Particles were selected using the first DMA and then were exposed to a prescribed relative humidity to measure the growth factor. Hygroscopic growth was measured at different RH. The sizes of the humidified particles were determined by the second DMA, which was operated in a scanning mode in combination with a CPC (model 3022A, TSI). The size selected aerosol flow and the sheath air flow of the second DMA were humidified at room temperature (25–30 ∘C depending on the surroundings) to almost the same RH with the sheath air at slightly higher RH. The second DMA was kept in a thermo-insulated box, which was cooled to 20 ∘C. Both aerosol and sheath air flow were cooled down to the same temperature before entering the second DMA, and thus the RH increased to its final value. The residence time of particles at the final humidity is approximately 30 s before they entered the SMPS operated with sheath air of the same RH. The hygroscopic growth factor (GF) was calculated as the ratio of the size of the wet particle (Dwet) to the selected dry size (Ddry). The HTDMA was calibrated using ammonium sulfate aerosol by comparing with the theoretical growth curve (Rose et al., 2008). From hygroscopic growth factor at 90 % RH, the hygroscopicity parameter, κHTDMA, was calculated according to Petters and Kreidenweis (2007). The error bars of κHTDMA were estimated using the standard deviation of the growth factor at (90±1) % RH of at least three duplicate scans.

SOA samples were collected on PTFE (Polytetrafluoroethylene) filters at the end of different experiments to obtain detailed insight into the chemical composition of the aerosol particles. The details of sample collection and analysis are described in Emanuelsson et al. (2013) and Kristensen and Glasius (2011). Before the filters, the air passed through an annular denuder coated with XAD-4 resin to remove gaseous organic species. The filters were extracted and analyzed using a Dionex Ultimate 3000 HPLC system coupled through an electrospray (ESI) inlet to a q-TOF mass spectrometer (micro-TOFq, Bruker Daltonics GmbH, Bremen, Germany), which was operated in both positive and negative mode. Pinonic acid, cis-pinic acid, terpenylic acid, diaterpenylic acid acetate (DTAA) and 3-methyl butane tri-carboxylic acid (3-MBTCA) were quantified using authentic standards.

For SOA from part of the experiments (experiment nos. B3, AB4, AB6 as in Table 1), samples were also collected on quartz fiber filters and analyzed by ultra-high-resolution mass spectrometry (UHRMS). In this analysis, the aerosol samples were extracted as described elsewhere (Kourtchev et al., 2013). The extracts were analyzed using an ultra-high-resolution LTQ Orbitrap Velos mass spectrometer (Thermo Scientific, Bremen, Germany) equipped with a TriVersa Nanomate robotic nanoflow chip-based ESI source (Advion Biosciences, Ithaca NY, USA). The Orbitrap MS instrument calibration, settings, and mass spectral data interpretation are described in Kourtchev et al. (2014). The mass accuracy of the instrument was below 1.5 ppm and the instrument mass resolution was 100 000 at m/z 400. The negative ionization mass spectra were collected in three replicates and in this study, only ions that appeared in all three analytical replicates were kept for evaluation.

The VOCs were measured by a high-resolution proton transfer reaction-mass spectrometer (HR-PTR-MS; Ionicon, Innsbruck, Austria) (Jordan et al., 2009) and gas chromatography coupled to a mass spectrometer (GC–MS; PerkinElmer, Waltham, USA) (Apel et al., 2008; Kaminiski, 2014).

The OH concentration was measured directly using laser-induced fluorescence (LIF) (Fuchs et al., 2012). The OH radicals inside the chamber are mainly formed by the photolysis of HONO formed via a photolytic process on the chamber walls, and to a minor fraction by O3 photolysis (Rohrer et al., 2005). From the OH concentration, the OH dose was calculated and used as a common abscissa in order to better compare different experiments. The OH dose is the integral OH concentration over time that gives the accumulated OH concentrations to which gases and particles were exposed in the course of experiment; 1 h exposure to a typical atmospheric OH concentration of 2×106 molecules cm-3 is then equivalent to an OH dose of 7.2×109 molecules cm-3 s.

The experimental procedures have been described elsewhere in details (Emanuelsson et al., 2013; Flores et al., 2014) and only a short description is given here. The chamber was typically humidified to 60–70 % RH in the beginning of the experiment and relative humidity can vary in the range of 30–70 % due to the ambient temperature change and the dilution by the flow to compensate the sampling loss. In a typical experiment, VOC was added to the chamber and then the roof was opened to start the photooxidation. In some experiments, O3 was added. In all the experiments, particles formed by homogeneous nucleation and no seed was added. In the BSOA experiments, a monoterpene mixture of α-pinene and limonene with a molar ratio of 1:1 was used as the representative BSOA precursors and its photooxidation induced BSOA formation. Ozone was added to initialize BVOC oxidation and particle formation. In the ASOA experiments, toluene or xylene was used as the representative ASOA precursors. In the mixed SOA experiments (ABSOA), AVOC and BVOC were added, either simultaneously or sequentially to investigate the potential effect of adding order.

In total, three BSOA experiments (including one using α-pinene as precursor), seven ASOA experiments, and six mixing experiments (ABSOA) with biogenic and anthropogenic precursors were analyzed in detail and the summary for these experiments is given in Table 1. In two experiments (nos. AB1, AB2), AVOC was added 6.3 h before the BVOC. In experiments with sequential VOC addition, the second VOC was added 1–2 h after the SOA mass concentration generated from the first addition reached its maximum. Accordingly, the time lag was longer when AVOC was added first due to its lower reactivity. In two experiments (nos. AB3, AB4), BVOC was added 2.5 and 5 h before the AVOC, respectively. In the other two experiments (nos. AB5, AB6), BVOC and AVOC were added simultaneously into the chamber. In the ABSOA experiments, the mass fraction of ASOA in the total aerosol was estimated using a method based on the aerosol mass yield and VOC consumed as described by Emanuelsson et al. (2013), where ideal mixing of ASOA and BSOA components was assumed. Assuming the same density for ASOA and BSOA, the mass fraction of ASOA is equal to its volume fraction.

Figure 4 shows the hygroscopicity (κHTDMA) of BSOA, ASOA, and ABSOA. κHTDMA of BSOA was between 0.03 and 0.06, and increased slightly with OH dose. κHTDMA of ASOA was around 0.09–0.1, significantly higher than that of BSOA. Subsaturated hygroscopic growth of ASOA and BSOA was obviously different whereas their CCN activity was basically similar. The influence of the SOA types on the hygroscopic growth is different from their influence on CCN activity. The comparison between the water uptake in the subsaturated conditions from hygroscopic growth and that in the supersaturated conditions from CCN activity is discussed in the Sect. 3.3. κHTDMA of ASOA did not change much with the OH dose and κHTDMA of ASOA from different aromatic precursors (toluene, xylene, and benzene) were similar (nos. A1, 2, 4, 5 in Fig. 4).

The higher κHTDMA of ASOA can be related to the chemical composition represented by O/C. The O/C of BSOA was about 0.3–0.5, distinctively lower than that of ASOA (0.7–0.8). O/C has been found to correlate with κHTDMA for various SOA systems (Jimenez et al., 2009; Massoli et al., 2010; Duplissy et al., 2008, 2011; Lambe et al., 2011). The same arguments as for CCN activity apply here. ASOA reached much higher O/C at the same OH dose compared to BSOA (Fig. S4) (Emanuelsson et al., 2013) because first generation products of AVOC have a smaller carbon number and higher vapor pressure compared to BVOC. Thus, first generation products of AVOC need more oxidation steps before starting to condense significantly on particles as discussed in Sect. 3.1. In addition, constituents of aromatic ASOA generally have lower molecular weights than BSOA molecules here. We can expect that ASOA components have on average a lower molar volume and thus ASOA has higher κHTDMA assuming all other parameters are the same for ASOA and BSOA (cf. Appendix A), since κ is by definition the ratio of the molar volume of water to the average molar volume of the solute.

The κHTDMA observed for ABSOA formed with AVOC and BVOC added in various orders were in the range 0.03–0.06, close to or slightly higher than the κHTDMA of BSOA. The observed O/C range of ABSOA was slightly higher than the O/C range of BSOA, but partly overlapping. Since ASOA had much higher κHTDMA than BSOA, ASOA enhanced the κHTDMA of ABSOA and the extent of enhancement depended on its fraction.

The enhanced κHTDMA due to ASOA fraction was also reflected clearly in the ABSOA experiments when AVOC and BVOC were sequentially added. In the experiment when ASOA was formed first, the SOA showed higher κHTDMA, around 0.09 (Fig. 5a). When BVOC was added to the system, ASOA was converted to ABSOA with a significant BSOA fraction (e.g., 70 % within 2.5 h) and κHTDMA decreased from 0.09 to 0.04 with the formation of the BSOA components. Meanwhile the degree of oxidation of ABSOA decreased significantly as indicated by the decrease of f44 (from 0.23 to 0.1). For the experiments when AVOC was added to BSOA system, an effect was recognizable; however, κHTDMA only increased slightly (Fig. 5b). This was because the reaction of aromatics with OH and SOA formation was slow and the fraction of ASOA did not exceed 10 %. Accordingly, only a slight increase of f44 was observed (from 0.10 to 0.12) even with concurrent aging, consistent with the minor effects of the ASOA component on the chemical composition of ABSOA due to its low fraction.

Since ASOA has higher κHTDMA, mixing of ASOA with BSOA may directly enhance κHTDMA due to a simple linear mixing. In order to understand the role of ASOA components in enhancing κHTDMA of ABSOA, κHTDMA was also examined as a function of the ASOA fraction (as shown in Fig. 6). In the ABSOA experiment, two main factors affect the hygroscopicity: aging of the BSOA components and increasing fraction of ASOA components. Therefore, the OH dose is examined to account for the effect of aging. In Fig. 6, the dashed lines connect the κHTDMA of pure BSOA and pure ASOA of the same OH dose at a series of OH doses varying from fresh to aged SOA. Such a graph can help to detect whether the κHTDMA of ABSOA can be described by a simple linear mixing of the κHTDMA of BSOA and ASOA components with respect to their volume fraction or where non-linear response of κHTDMA is effective. For each OH dose, a dashed line connects pure BSOA and pure ASOA at the given OH dose (represented by the size of marker). This line defines the expected κHTDMA range of ABSOA with varying ASOA fraction at given OH dose. If the κHTDMA of ABSOA can be described by a linear combination of the κHTDMA of pure ASOA and BSOA components in respect of their volume fraction, the κHTDMA data point should be on the line corresponding to the given OH dose of that data point and should increase with ASOA fraction along the line due to the higher κHTDMA of ASOA. If a succession of points from one experiment cross dashed lines (i.e., points beyond the line corresponding to the OH dose of those points), it would indicate κHTDMA cannot be explained by a linear combination.

For ABSOA, several cases with non-linear effects were observed. For ABSOA in experiment no. AB1 and no. AB2 where AVOC was added first, κHTDMA was significantly lower than the values from the linear combination of pure ASOA and BSOA components (much below the lines corresponding to the OH doses of the data points). For ABSOA in experiment no. AB5 when AVOC and BVOC were added together, κHTDMA did not change significantly in spite of a significant increase of ASOA fraction. In the beginning, κHTDMA of ABSOA was higher than the value from a linear combination, whereas in the end, κHTDMA was lower than the value from a linear combination of pure systems. These cases indicate that the observed κHTDMA of ABSOA cannot be explained by a simple linear combination of pure ASOA and BSOA systems. There seems to be some additional effects such as oligomerization, which altered the chemical composition of ABSOA and thus affected κHTDMA. Moreover, for the ABSOA in experiment no. AB5, κHTDMA remained largely unchanged in spite of continuous oxidation and increase of ASOA fraction, both enhancing hygroscopicity. This further indicates that the possible oligomerization, which should decrease the Raoult term and thus hygroscopicity, compensates the effect of photochemical aging, which enhances hygroscopicity, consistent with the discussion in Sect. 3.1.

Morphological effects can also play a role. If the ASOA and BSOA components were not well mixed in the aerosol particles in the experiments with sequential VOC additions, there would be more BSOA components on SOA particle surface in experiments no. AB1 and no. AB2. This could affect the κHTDMA and contribute to the non-linear effect. But this cannot explain the non-linear effect in the experiment with VOCs added simultaneously. In addition, if ABSOA forms a glassy state, the lower diffusivity in the particle may hinder water uptake thus decreasing κHTDMA. Although similar growth kinetics of SOA to (NH4)2SO4 was observed in supersaturated conditions, in subsaturated conditions the water diffusivity in the particle may be limited thus limiting water uptake.

The ABSOA filter samples from experiment no. AB4 and no. AB6 were extracted and analyzed for oligomers. We observed the oligomer formation in these samples (Fig. S6). Oligomer in SOA has been found by a number of studies (Gao et al., 2004; Noziere et al., 2015; Tolocka et al., 2004; Kalberer et al., 2004; Kourtchev et al., 2014, 2015). Small multi-functional products from aromatics oxidation (Hamilton et al., 2005; Jenkin et al., 2003; Johnson et al., 2005) may promote oligomerization between ASOA and BSOA components. But we did not find indications that ABSOA contained more dimers compared to BSOA. This can be attributed to the low ASOA fraction ≤ 5 % in experiments no. AB4 and no. AB6 (estimated using the method as in Emanuelsson et al., 2013). The low ASOA fraction was caused by the low OH concentration and low chemical turnover of the aromatics in these experiments because high concentrations of VOC were used in order to generate enough particle mass for optical measurement (Flores et al., 2014). The low fraction of ASOA resulted in little oligomer formation by the interaction between the ASOA components and BSOA components. In the future experiments, conditions that can form comparable fractions of both ASOA and BSOA and thus are favorable to ASOA and BSOA interaction such as oligomerization are preferred. Therefore, relatively higher AVOC concentration and higher OH concentration (as in experiment no. AB5) are desirable.

The hygroscopicity parameter κ was obtained from CCN and HTDMA measurements in supersaturated and subsaturated conditions, respectively. For all SOA types studied here, there is a significant gap between κHTDMA and κCCN. κHTDMA was significantly lower than κCCN with κHTDMA/κCCN around 0.3–0.7 (Fig. 7). The ratio of κHTDMA/κCCN for BSOA and ABSOA was lower than that of ASOA, which is closer to 1. This means that there is a smaller gap between κHTDMA and κCCN for ASOA compared to BSOA and ABSOA.

The closure between κCCN and κHTDMA of SOA has been studied and discussed by a number of previous studies with varying results (Dusek et al., 2011; Alfarra et al., 2013; Good et al., 2010; Duplissy et al., 2008; Juranyi et al., 2009; Prenni et al., 2007; Massoli et al., 2010; Hansen et al., 2015; Wex et al., 2009; Whitehead et al., 2014). The discrepancy between κCCN and κHTDMA found here can be attributed to several possible reasons as discussed in the literature (Prenni et al., 2007; Massoli et al., 2010; Frosch et al., 2011; Good et al., 2010; Alfarra et al., 2013; Wex et al., 2009; Whitehead et al., 2014; Petters et al., 2009; Dusek et al., 2011). An important reason is the presence of slightly soluble compounds. These compounds only dissolve partly in the subsaturated condition while they can dissolve completely in the supersaturated conditions due to more water available. Therefore, κ is underestimated to varied extent in the subsaturated condition. ASOA components here seemed to have higher solubility compared to BSOA components, and thus the gap between κCCN and κHTDMA was smaller than that of BSOA.

Surface tension can also play a role in this discrepancy. κ was calculated using the surface tension of pure water (0.073 N m-1). If the surface tension of the droplets is lower than that of water, κ would be overestimated. While κHTDMA is not so sensitive to the change of surface tension, κCCN is more sensitive to surface tension at the point of activation according to the Köhler equation (Petters and Kreidenweis, 2007). The surface tension under the subsaturated conditions is assumed to be lower than that under the supersaturated conditions due to the more concentrated organics in the droplets under the subsaturated condition (Prisle et al., 2008). If the surface tension effect for BSOA would be larger than for ASOA, i.e., lower surface tension in supersaturated conditions, this would lead to a higher κCCN for BSOA. While the surface tension effect in subsaturated conditions is small, i.e., κHTDMA is relatively constant, higher κCCN of BSOA results in a larger discrepancy between κHTDMA and κCCN. However, surface active organics can be enriched at the surface to such a high extent that the Raoult term is significantly diminished (Prisle et al., 2008). This difference can compensate the overestimation by using the surface tension of water and the compensating effects make using surface tension of water be a reasonable choice.

Furthermore, the κ-Köhler model does not account explicitly for changes in non-ideality of a solution as a function of saturation ratio, i.e., water activity. Instead, κ might be not constant through the whole range of water activity (Petters and Kreidenweis, 2007). In addition, different aerosol behaviors such as evaporation and condensation of organics in HTDMA and CCN instrument and limited diffusivity of water in case of glassy particles can contribute to the discrepancy (Whitehead et al., 2014; Asa-Awuku et al., 2009; Irwin et al., 2010; Topping and McFiggans, 2012; Topping et al., 2013; Duplissy et al., 2009).