SOA formation from gas-phase emissions of five different heated cooking oils was investigated in a potential aerosol mass (PAM) chamber, which has been described in detail elsewhere (Kang et al., 2007, 2011; Lambe et al., 2011a, 2015). Briefly, a PAM chamber is a continuous-flow stainless steel cylindrical reactor using high and controlled levels of oxidants to oxidize precursor gases to produce SOA. The volume is approximately 19 L (length 60 cm, diameter 20 cm). High OH exposures were produced through the photolysis of ozone irradiated by a UV lamp (λ=254 nm) in the presence of water vapor. Ozone was produced by an ozone generator (1000BT-12, ENALY, Japan) via irradiation of pure O2. The OH concentration was controlled by the flow rate of ozone in the PAM chamber, which was approximately 40 ppm prior to dilution. The ozone concentration in the PAM reactor was adjusted to five different levels, ranging from 0.4 to 2.7 ppm. The total flow rate in the PAM chamber was set at 3 Lmin-1 by a mass flow controller, resulting in residence time of 380 s. The corresponding upper limit of OH exposure at these operating conditions was 1.7×1011 moleculescm-3s, which is equivalent to 1.3 days of atmospheric oxidation, assuming an ambient OH concentration of 1.5×106 moleculescm-3 (Mao et al., 2009). The upper limit of OH exposure was determined by measuring the decay of SO2 (model T100, TAPI Inc, USA), following previous procedures (Kang et al., 2007; Lambe et al., 2011a). Peng et al. (2016) found that non-OH chemistry, especially reactions with O3, may play a role in the oxidation flow reactors for consumption of VOCs. According to Klein et al. (2016a), emissions of VOCs from heating cooking oils were dominated by saturated and unsaturated aldehydes. In this study, the ratio of O3 exposure to OH exposure ranged from 1.5×105 to 1.9×105, relatively lower than tropospheric values (Schmidt et al., 2014). At this O3exp / OHexp, ozonolysis of saturated and unsaturated aldehydes was negligible since the ratios of their ozonolysis rate constants to OH rate constants were in the range of 10-9 to 10-7. (Grosjean et al., 1993; Atkinson and Arey, 2003). Thus reactions of VOCs with O3 played a negligible role in this study. Before and after each experiment, the PAM reactor was cleaned by exposure to a high concentration of OH until the mass concentration of background particles was less than 5 µgm-3.

The PAM chamber was designed with a large radius and a small surface-to-volume ratio to minimize wall effects. The transmission efficiency for particles at a mean mobility diameter (Dm) larger than 150 nm was greater than 80 % (Lambe et al., 2011a). The wall loss of particles was considered to be small, as the particles larger than 150 nm accounted for greater than 70 % of the aerosol mass (Fig. 1). Transmission efficiency of gases in the PAM chamber indicates that vapor wall losses in the PAM chamber are negligible (Lambe et al., 2011a).

A schematic of the experimental setup is shown in Fig. 2. The tested vegetable oils, purchased from a local supermarket, included canola (rapeseed), corn, sunflower, peanut and olive oils. For each experiment, 30 mL of vegetable oil was heated at approximately 220 ∘C for 20 min in a 500 mL Pyrex bottle on an electric heating plate. Note that visible smoke was observed during heating of olive oil, possibly because the temperature was above the smoke point of olive oil. This high temperature may result in increased emissions of large aldehydes from olive oil, but may not significantly change the relative composition of emissions from other oils with higher smoke points (Klein et al., 2016a). Prior to introduction to the PAM chamber, particles from the heated oil emissions were removed using a Teflon filter. An unheated 2 m Teflon tube was used as the transfer line. The residence time in the transfer line was less than 2 s, resulting in wall losses of VOCs less than 5 % according to Liu et al. (2015). After 10 min of heating, the UV lamp was turned on and the emissions were exposed to high OH levels for approximately 1 h. Once the UV lamp was turned off, the PAM reactor was flushed continuously using pure N2 and O2 until the aerosol mass was below 3 µgm-3. Then the experiment was repeated at another OH level. The RH and temperature of the PAM outflow were measured continuously (HMP 110, Vaisala Inc, Finland) and stabilized at 65–70 % and 19–20 ∘C, respectively. The adjustment of RH was achieved by passing the pure N2 and O2 through water bubblers. Blank experiments were conducted in the absence of cooking oils under similar conditions to quantify the amount of aerosols formed from matrix gas when exposed to different OH levels.

POA emitted from heated cooking oils was also characterized in this study. For each test, 30 mL of vegetable oil was heated to 240 ∘C for 2 min in a pan on an induction cooker. The emissions, after passing through a mixing chamber of 36 L, were first diluted by a Dekati diluter (DI-1000, Dekati Ltd., Finland) by a factor of approximately 8. Then 0.15 Lmin-1 of the total diluted flow was introduced to the PAM chamber, achieving a final dilution ratio of approximately 160. No ozone was introduced to the PAM chamber during measurement, and the UV lamp was off. Temperature and RH were similar to those of the SOA formation experiments.

A scanning mobility particle sizer (SMPS, TSI Incorporated, USA, classifier model 3082, CPC model 3775) was used to measure particle number concentrations and size distributions. Particle size ranged from 15 to 661 nm. An aerosol density of 1.4 gcm-3 was assumed to estimate the SOA mass from the particle volume concentration (Zhang et al., 2005). For the SOA formation experiments, the contribution from background organic aerosols was subtracted from the total organic aerosols. The maximum concentration of background organic aerosols was 8.4 µgm-3, almost negligible compared with the dozens to several hundreds of µgm-3 of SOA formed in this study. The organic aerosol composition was characterized by a high-resolution time-of-flight aerosol mass spectrometer (HR-TOF-AMS, abbreviated as AMS hereafter, Aerodyne Research Incorporated, USA) (DeCarlo et al., 2006). A silica gel diffusion dryer was connected to the sampling line to remove water. The residence time in the dryer was approximately 8 s, sufficient to reduce the RH to less than 30 %. The instrument was operated in the high-sensitivity V mode and high-resolution W mode alternating every 1 min. The toolkit Squirrel 1.57I and Pika 1.16I were used to analyze the AMS data. The molar ratios of hydrogen to carbon (H : C) and oxygen to carbon (O : C) were determined with the improved-ambient method (Canagaratna et al., 2015). The ionization efficiency of AMS was calibrated using 300 nm ammonium nitrate particles. The particle-free matrix air, obtained by passing the air through a HEPA filter, was measured for at least 20 min before each experiment to determine the signals from major gases. The collection efficiency (CE) was corrected by comparing AMS mass concentrations to concurrent SMPS mass concentrations, following the methods of Gordon et al. (2014) and Liu et al. (2015). The value of CE varied from 0.38 to 0.78 in this study. Note that particles were not dried prior to SMPS measurements, which might lead to an overestimate of SOA mass due to the uptake of water by organics. Lambe et al. (2011b) investigated the cloud condensation nuclei activity of PAM-generated SOA and found that the hygroscopicity parameter κorg was linearly correlated with O : C ratios. Based on their κorg–O : C relationship, we estimated an upper limit of κorg to be 0.089 in this study. The overestimate of SOA mass due to water uptake were thus determined to be less than 18 % (Petters and Kreidenweis, 2007; Pajunoja et al., 2015).

The SOA production rate (PR) was expressed as micrograms (µg) of SOA produced per minute (min), calculated using the following equation, similar to calculation of emissions rates of primary particles from cooking (Klein et al., 2016a): PR=SOA×DR×F, where [SOA] is the SOA concentration in µgm-3, DR is the dilution ratio and F is the flow rate in m3min-1 of the carrier gas. All gas-phase emissions from heated cooking oils were assumed to be transported into the PAM chamber.

Emission rates are commonly used to normalize PM emissions from cooking activities (Torkmahalleh et al., 2012; Gao et al., 2013; Klein et al., 2016a, b). Here, the adoption of SOA PR, similar to emission rates, facilitates the normalization of SOA production from cooking and direct comparison of the amount of primary emitted and secondary formed particles. Though SOA yields were not determined due to the lack of VOC concentrations, we believe that SOA PR is a useful metric for the estimation of SOA production from cooking and can be used for comparison among different studies. Note that PR is highly related to the experimental condition, especially OH exposure and temperature of the cooking oil.

In Fig. 3, we plot the time series of RH, ozone and organic aerosol concentrations during the aging of gas-phase emissions from heated peanut oil. As described above, the ozone concentration prior to dilution was stable at approximately 40 ppm. The pulse of RH was caused by disconnection of the introduction line when changing the Teflon filter. During the initial 10 min of heating, the mass concentration of organics was close to the detection limit of the instrument, indicating that POA emissions were thoroughly removed by the Teflon filter. During these periods of experiments where OH radicals were not present, we found that ozone chemistry had a negligible influence on SOA formation in this study. Immediately after oxidation was initiated by turning on the UV lamp, substantial SOA was formed, and its concentration stabilized after about 20 min. The SOA concentration subsequently reported is the average for the steady period.

Figure 4 shows SOA concentration as a function of OH exposure and photochemical age in days during the aging of gas-phase emissions from different heated cooking oils. The OH exposure ranged from 2.7×1010 to 1.7×1011 moleculescm-3s, corresponding to 0.2–1.3 days of photochemical age, assuming 24 h average ambient OH concentrations of 1.5×106 moleculescm-3 (Mao et al., 2009). For all experiments, the SOA concentration almost linearly increased from 41–107 to 320–565 µgm-3 as OH exposure increased. This linear increase has also been observed from vehicle exhaust at a similar range of OH exposures (Tkacik et al., 2014). Typically, VOCs are oxidized through functionalization reactions to produce less volatile organics that readily condense to form SOA. Upon further oxidation, fragmentation reactions and cleavage of carbon bonds can occur and form more volatile products that reduce SOA levels (Kroll et al., 2009). In this study, functionalization reactions dominated SOA formation as reflected by the increase in SOA concentrations shown in Fig. 4.

The slope of the fitted straight line to the SOA data was calculated to estimate the efficiency of different cooking oils in producing SOA (Table 1). The efficiency of SOA production, in ascending order, was peanut oil, olive oil, canola oil, corn oil and sunflower oil. The slope of sunflower oil was 3.82×10-15 µgmolecules-1s-1, more than 2 times that of peanut oil. The different slopes might be related to the emission rate and composition of VOCs from various cooking oils. Table 1 presents the type of fat content of the different cooking oils. It should be noted that the organic vapors studied here were not the specific fats present in the raw oils but the thermal breakdown products of fat lipids. Unsaturated fat accounts for 75–88 % of the total fat content. A multivariate linear regression was used to relate the SOA production efficiency to the fat content of cooking oils. The intercept was set to zero. The resulting equation was Y=2.62×10-17X1+4.71×10-17X2, where Y is the SOA production efficiency (µgmolecules-1s-1); X1 and X2 represent the content of monounsaturated fat (%) and omega-6 fatty acid (%) in cooking oil, respectively. The SOA production efficiency was strongly correlated (R2=0.97, p< 0.05) with the content of monounsaturated fat and omega-6 fatty acids. This indicated that the major SOA precursors from heated cooking oils were related to the content of monounsaturated fat and omega-6 fatty acids in cooking oils. Moreover, omega-6 fatty acids dominated the contribution to SOA production. Omega-6 fatty acids are a family of poly-unsaturated fatty acids that have in common a final carbon–carbon double bond in the n-6 position, counting from the methyl end (Simopoulos, 2002). The peroxyl radical reactions of omega-6 fatty acids might emit long-chain aldehydes (Gardner, 1989), which have been suggested as potential SOA precursors (Chacon-Madrid et al., 2010).

The average SOA PR from gas-phase emissions of the five cooking oils at an OH exposure of 1.7×1011 moleculescm-3s was calculated to be 1.35±0.30 µgmin-1. Torkmahalleh et al. (2012) found that primary PM2.5 emission rates for peanut, canola, corn and olive oils heated at 197 ∘C ranged from 3.7 to 54 mgmin-1. He et al. (2004) reported a PM2.5 emission rate for frying in vegetable oils of 2.68±2.18 mgmin-1. The SOA PR determined in this study was negligible compared with primary PM2.5 emission rates for heated cooking oils and frying in vegetable oils. However, our results may underestimate SOA production from cooking under real-world conditions. First, recent studies have demonstrated that the oxidation of IVOCs and SVOCs evaporated from POA could produce significant SOA (Donahue et al., 2006; Jimenez et al., 2009). In this study, POA from heated cooking oils was filtered. SVOCs and IVOCs might not evaporate from the filter given that they might be at saturation as the aerosol was cooled after the emissions. Second, emissions of SOA precursors will be enhanced when cooking food compared with heating cooking oils alone. For instance, long-chain aldehyde emissions from frying processes can be 10 times those of heated oil (Klein et al., 2016a). Large amounts of monoterpenes will be emitted when frying vegetables or cooking with herbs and spices (Klein et al., 2016a, b; Liu et al., 2017). These enhanced emitted precursors may significantly enhance SOA production. Finally, laboratory and tunnel studies indicate that SOA production from typical precursors and vehicle exhaust peaks at OH exposures higher than 5.0×1011 moleculescm-3s (Tkacik et al., 2014; Lambe et al., 2015). The relatively lower OH exposures in this study compared with typical conditions in the atmosphere may lead to the underestimation of cooking SOA.

Figure 5 shows high-resolution mass spectra of POA and SOA at an OH exposure of 2.7×1010 moleculescm-3s from heated canola oil. Other oils have similar mass spectra, as reflected in the good correlations shown in Table 2. The mass concentration of POA was approximately 35 µgm-3 for canola oil. The prominent peaks in POA from canola oil were m/z 41 and 55, followed by m/z 29 and 43. The m/z 41, 43 and 55 were dominated by C3H5+, C3H7+ and C4H7+ ion series, consistent with the previous observation by Allan et al. (2010). The m/z 29 was instead dominated by ion CHO+, which can be used as a tracer for organic compounds with alcohol and carbonyl functional groups, as a result of thermal decomposition of the oils (Lee et al., 2012). For the SOA mass spectra, the dominating peaks were m/z 28 and 29, followed by m/z 43 and 44. The m/z 28, 29, 43 and 44 were dominated by CO+, CHO+, C2H3O+ and CO2+, respectively. For all cooking oils, the mass fractions of m/z 28 and 44 in SOA were higher, while the mass fractions of m/z 55 and 57 in SOA were lower than those of the corresponding POA. The increase in mass fractions of the oxygen-containing ions in SOA mass spectra indicated the formation of oxidized organic aerosols.

The correlation coefficients (R2) between POA and SOA unit mass resolution (UMR) spectra of heated oil and COA resolved by positive matrix factorization (PMF) analysis (Lee et al., 2015) were calculated and summarized in Table 2 to evaluate their similarities. The POA mass spectra between different cooking oils exhibited strong correlations (R2> 0.97) and agreed well with the ambient COA factor obtained at roadside sites in the commercial and shopping area of Mongkok in Hong Kong (Lee et al., 2015). The SOA mass spectra between different cooking oils displayed good correlations (R2> 0.94), suggesting a high degree of similarity. The mass spectra of cooking SOA also greatly resemble POA and field-derived COA in ambient air, with R2 ranging from 0.74 to 0.88. Kaltsonoudis et al. (2016) also observed that the ambient COA factor in two major Greek cities in spring and summer strongly resembled the aged SOA from meat charbroiling in a smog chamber.

Fragments derived from the AMS data have been extensively used to explore the bulk compositions and properties of ambient organic aerosols (Zhang et al., 2005; Ng et al., 2010; Heald et al., 2010). Here, we use the approach of Ng et al. (2010) by plotting the fractions of the total organic signal at m/z 43 (f43) vs. m/z 44 (f44). The m/z 43 signal is abundant in C3H7+ and C2H3O+ ions, indicating fresh, less oxidized organic aerosols. The m/z 44 signal, usually dominated by CO2+ and formed from the thermal decarboxylation of organic acids, is an indicator of highly oxygenated organic aerosols (Ng et al., 2010).

In Fig. 6, we plot f43 vs. f44 of cooking SOA and SOA data from gasoline (Presto et al., 2014; Liu et al., 2015) and diesel (Presto et al., 2014) vehicle exhaust measured in a smog chamber, together with the triangle defined by Ng et al. (2010) based on the analysis of ambient AMS data. The ambient low-volatility oxygenated OA (LV-OOA) and semi-volatile OOA (SV-OOA) factors fall in the upper and lower regions of the triangle, respectively. Ng et al. (2010) proposed that aging would converge the f43 and f44 toward the triangle apex (f43=0.02, f44=0.30). In this study, the f43 and f44 ranged from 0.06 to 0.10 and from 0.05 to 0.07, respectively; they mainly lie in the lower portion of the SV-OOA region. As shown in Fig. 6, SOA from gasoline and diesel vehicle exhaust at a similar range of OH exposures had f44 values of 0.11–0.12. Compared with vehicle exhaust, SOA formed from gas-phase emissions of heated cooking oils was less oxidized. The potential SOA precursors from heated cooking oils might be long-chain aldehydes, which are less volatile than SOA precursors such as aromatics and long-chain alkanes from vehicle exhaust. Generally, the presence of additional methylene and aldehyde reduce compound vapor pressure by factors of 3 and 22, respectively (Pankow and Asher, 2008). For example, the vapor pressure of n-tridecanal is approximately 14 % of that of n-tridecane at 25 ∘C, as predicted by the group-contribution model (Pankow and Asher, 2008). A single polar moiety of first-generation products from long-chain aldehydes will have low enough volatility to condense, while more volatile aromatics and long-chain alkanes require more functionalization to form SOA (Donahue et al., 2012). Therefore, SOA formed from heated cooking oils was less oxidized. For each cooking oil, there was little change in f44 and a slight increase in f43 as OH exposure increased. The increased SOA mass may facilitate the partitioning of more volatile organics, leading to a slight increase in f43 and little change in f44. This is consistent with the observation of previous studies that the f44 of SOA from aromatics and monoterpenes varied little and that f43 increased slightly for SOA mass loadings higher than 100 µgm-3 (Ng et al., 2010; Kang et al., 2011).