2.1 Sampling and Chemical Analysis

Particle phase cooking markers including sFAs and uFAs were measured from November 9 to December 9, 2018 at an urban site in Shanghai, China using a TAG module (Aerodyne Research Inc.) coupled with a gas chromatograph/mass spectrometer (GC/MS). A one-hour sample was collected at every odd hour (e.g., 1:00-2:00 a.m.), followed by on-line thermal desorption and in-situ derivatization before GC/MS separation and detection. No samples were collected during the even hours, as a cycle of sample collection and analysis took 2 h to complete. Details of chemical characterization can be found in our recent papers (He et al., 2020; Q. Wang et al., 2020).

2.2 Relative Rate Constant Analysis Method

Donahue et al. (2005) formulated the general expression to determine the relative rates for heterogeneous oxidation reactions of multi-component OA. The specific expression as applied to the ambient measurements of uFAs is derived, as given in Equation 1a and Equation 1b. The derivation details are provided in Text S1.

(1a)

(1b)

C_i and C_18:0 are the particle-phase concentration of species i and stearic acid (C18:0), respectively. Among the quantified sFA and uFA cooking markers, C18:0 is the least reactive, thus it was selected as the reference molecule for normalization. Using the concentration ratio eliminates the interference from atmospheric dilution and deposition. Fitting the ambient C_i/C_18:0 data versus t with an exponential function provides an estimate for k, the effective pseudo-first order decay rate (h⁻¹). is the second-order reaction rate constant of species i against an oxidant. is the average oxidant concentration in the aerosol.

3.1 General Characteristics of Ambient Cooking Aerosols in the Field Campaign

The relative abundance distribution of the series of sFAs and uFAs in the ambient data was largely similar to those measured for the real-kitechen cooking profiles (Text S2), strongly suggesting cooking emissions as their dominant source. Diurnal variations of concentrations and normalized concentrations of C18:2, C18:1, and t-C18:1 by C18:0 are shown in Figure 1. The normalized concentrations of all three uFAs showed the highest values at 19:00, coinciding with the dinner time; then started to decrease until 05:00 in the morning, reflecting the continuing loss of uFAs due to chemical degradation. We note that the influence of gas-particle partitioning on the target species is negligible due to their low volatilities (Text S3). Thus, this nighttime decay of particle-phase uFAs presents a unique opportunity of examining their degradation by heterogenous reaction with O₃, as the post-dinner nighttime (i.e., 19:00-05:00) was mostly free of compounding factors such as fresh emissions and degradation initiated by OH radical. In comparison, one can expect a rate constant analysis for the daytime data would be complicated, as the concentration variation is affected by both fresh cooking emissions throughout the day and chemical degradation from multiple oxidants (e.g., OH and O₃).

The sampling period was influenced by both local and long-range transport air masses according to the backward trajectory analysis (Text S4). Briefly, the first sampling sub-period (November 9–22) was mainly influenced by regional air mass from northern China, while the last sampling sub-period (November 23-December 3) was mainly influenced by local air mass. Day-by-day nightly average ozone concentrations were in the range of 2.3–20.9 ppb (Figure S6), with lower ozone observed under local air mass influence due to the enhanced NO concentrations from local vehicle emissions. Averaged over the whole campaign, the diurnal variation of ozone peaked around noon and early afternoon (Figure 1c), consistent with its photochemical formation origin. The bulk OA, monitored by an AMS (Q. Wang et al., 2020), showed fresher characteristics under local air mass in comparison with that under regional air mass influence. Specifically, the nightly average O/C ratios in the bulk OA were 0.35 ± 0.09 under local air mass while 0.44 ± 0.08 under long-range transport air mass. The contrast of the concentration ratios of uFAs at 19:00 (i.e., dinner time) among the two types of sampling days (Figure S10) also corroborated the AMS measurement data. For example, the highest concentration ratio of C18:1/C18:0 (2.5) occurred on November 30, in the period under local air mass influence, while the lowest value (0.4) occurred on November 11, in a period under regional air mass influence. The decay rates for the uFAs were calculated on an individual night basis, considering the night-to-night variations in ozone level and in air mass origins. We note that during the sampling campaign, consistent air mass origins within each night are observed, the relatively stable atmospheric conditions within the 10-h span make it possible to assume that the emitted cooking particles have the same composition during each night.

3.2 Effective Decay Rate Constant Analysis of uFAs Using the Relative Rate Constant Approach

In principle, the effective decay rate k as described in Equation 1b varies with time if one considers the changing aerosol matrix and the variation of ozone concentration over the examined time window of 19:00-05:00; the degree of variation depends on the variability of the underlying factors. After trying step-by-step fitting to examine the variation of k determined over different time intervals from 2 to 10 h (details in Text S5), we could categorize the measurement data in the individual nights into two scenario groups that can be described by a one-step degradation model and a two-step degradation model, respectively.

In the one-step model, one rate constant k is obtained by fitting data over the entire period of 19:00-05:00. In the two-step model, an initial rate constant k₁ is obtained for data between 0 and 4 hr (i.e., 19:00–23:00) and a k₂ for the subsequent 4–10 hr (i.e., 23:00–05:00). The reason of selecting t = 4 h as a separation point is empirical, based on that the rate constant before and after that point differed significantly and that the decay rate after 4 hr did not show further reduction (Text S5). On one hand, a large difference between k₁ and k₂ confirms the evolving reaction rate within the study time. On the other hand, if k₁ is close to k₂, it suggests a consistent rate constant during the examined time window. In this study, if k₁ equals or exceeds two times that of k₂, the two-step fitting will be used. Otherwise, the entire one night's data will be fitted by the one-step model, generating one k value, representing the average decay rate during the night. This approach significantly reduces the fitting residuals and better explains the ambient data (Figures S8 and S9). Overall, the ambient uFAs data are well fitted by either a one-step or two-step model, with the coefficient of determination R > 0.96. This result supports the assumption of the pseudo first-order decay rate of heterogeneous reaction.

Among a total of 22 nights' data, 17 nights are best described by the one-step model and 5 nights are best described by the two-step model for the degradation of C18:1. For C18:2, 6 nights are fitted by the two-step model. The individual night-by-night decay rate and the associated one standard deviation error are thus obtained and shown in Figure S10. Figure 2 shows examples of the fitting on two selective days. In the first example (November 14, 2018), a gradually decreasing trend was observed within the study time. Toward the end of the study time, all three uFAs were almost depleted. The backward trajectory on that day shows the influence from long-range transport air mass from northern China, indicating aged OA (Figure S4). In the second example (November 30, 2018), the concentration ratios of C18:1 and C18:2 decreased rapidly within the first 4 h, followed by a notably slower decrease in the last few hours. For example, 78% of C18:2 was depleted in the first 2 hr, while only 14% reacted in the second two hours and the decreasing trend afterward was not discernable. In this example, a single decay rate cannot fit the ambient data well, leaving a large residual (Figure S9) while the two-step fitting, with k₁ = 0.72 h⁻¹ and k₂ = 0.24 h⁻¹, explains the ambient data well. Note that k_1, the initial decay rate in the first 4 hr, is 3 times faster than the decay rate k₂ for the later hours. The backward trajectory on that day shows the influence from local circulating air mass, indicating fresh OA (Figure S4). The ozone concentrations under such a scenario are generally lower. The dissimilar behaviors in the two scenario groups reflect the different reaction kinetics on fresh and aged OA.

For t-C18:1, only 2 days' data are fitted by the two-step model while the one-step model describes better the other 20 nights. t-C18:1 being a trans-isomer, its steric arrangement is more similar to the sFAs than the cis-configured C18:1 and C18:2. This configuration difference is likely the underlying cause for the different kinetic behavior of t-C18:1, as the steric configuration could conceivably affect molecular interactions with the co-existing molecules on the particle.

3.3 Effective Decay Rate Constants Determined From Ambient Measurements

Figure 3a compiles the estimated effective rate constants for C18:2, C18:1, and t-C18:1, showing both the boxplot distributions and individual values. For the nights fitted with the two-step model, the gas phase ozone level is comparable during the two sub-time windows corresponding to k₁ and k₂. k₁ and k₂, however, differ significantly. On average, k₁ is 6 times higher than k₂ for C18:2, 4 times higher for C18:1, and 5 times higher for t-C18:1, respectively. Previous studies have revealed the importance of the viscosity of atmospheric OA on heterogenous reaction rates. The reaction can occur rapidly in low-viscosity particles throughout the bulk particle, while in viscous particles, heterogeneous chemistry may occur only very slowly and be confined to the particle surface (Reid et al., 2018). Previous chamber studies reported that OA upon aging would become more viscous and show increased mass transfer limitations (Renbaum-Wolff et al., 2013; Takhar et al., 2019). Plausibly, the large initial reaction rate could be due to the high surface concentration of uFAs and higher ozone solubility in fresh OA. The much slower reaction rate at later hours may be due to that enhanced SOA components in aerosol composition led to reduced diffusivity within the particle and decreased solubility for ozone (Huff Hartz et al., 2007). For the nights well-described with the one-step model, it is likely that the monitoring data did not capture the initial fast decay rate of the fresh OA under these days, and the aerosol sampled has been significantly aged.

Note that the k's incorporate the oxidant concentration (C_ox), as defined in Equation 1b, and k would linearly increase with C_ox, or gas-phase ozone, which is a closely associated quantity of C_ox. Indeed, a positive correlation of k and k₂ with measured gas phase ozone concentration was observed for the uFAs, with pearson correlation coefficient R_p of 0.72–0.78, as shown in Figure 3b. The ozone oxidant factor also explains the pattern of k > k₂ for the three uFAs (Figure 3a), as higher ozone was recorded under regional air mass influence (11–14 ppb) than under local air mass influence (3–8 ppb). Due to limited data set size (N ≤ 6) for generating k₁, we are unable to observe a discernable positive relationship between the decay rate constant for fresh OA (k₁) with ozone.

The data availability of multiple uFAs provides an opportunity to examine if the relative reactivities are consistent with their structural difference. Figure 3c presents a scatter plot of the effective rate constants determined for individual nights, C18:2 versus C18:1 in the left panel and t-C18:1 versus C18:1 in the right panel. Note that the underlying particle composition and mixing state varied from one night to another. Despite the variations, a linear relationship is evident, with R_p = 0.92 between the data of C18:2 versus C18:1 and R_p = 0.70 between the data of t-C18:1 versus C18:1. The linear relationships confirm that the derived from ambient measurements across many nights are mainly driven by the intrinsic reactivity of the individual uFAs. The slope indicates the relative reactivity, that is,

(2a)

(2b)

The reactivity order of C18:2 > C18:1> t-C18:1 is in accordance with their chemical structures. Both t-C18:1 and C18:1 have one C=C bond, but the trans-position of the C=C bond in t-C18:1 presents a larger steric hindrance to make the compound less accessible to ozone. This likely explains the lower reactivity of t-C18:1 than C18:1 toward ozone. C18:2 possesses two C=C bonds whereas C18:1 has only one, approximately doubling the rate constant for ozonolysis, based on the structure–activity relationship model (King et al., 1999). While the relative reactivity proportionality determined from our ambient data is lower than 2, it is in agreement with the relative reactive uptake coefficients of O₃ from previous chamber studies (1.4–1.6) (Hearn & Smith, 2004; Moise & Rudich, 2002; Thornberry & Abbatt, 2004). This result may indicate that, besides the chemical structure, other factors such as the diffusion limit could also be influential.

We also examined the correlation of the estimated effective rate constant with ambient relative humidity (RH) (Figure S11). No discernable dependence relationships could be observed. This is seemingly inconsistent with the active role of RH as recorded in laboratory studies. Vesna et al. (2009) reported that when RH increased from 0% to 80%, the loss rate of C18:1 increased while the yields of peroxide products showed a clear decrease. Weitkamp et al. (2008) on the other hand, found that a higher RH (60% vs. 0%) decreased the effective decay rate constant of both uFAs and cholesterol. In principle, water can influence the reaction in two ways. On one hand, higher RH facilitates more water uptake by polar constituents (e.g., SOA), hence leading to aerosol becoming less viscous and promoting heterogenous reactions via increased diffusivity. On the other hand, water can compete with the uFAs in their reactions with criegee intermediates, thus retarding the degradation reaction of uFAs. Previous studies have proposed that the carboxylic acids can also react with the criegee intermediates, thus be lost by this reaction as well as reaction with ozone (e.g., Ziemann, 2005). The nighttime RH in this work ranges from 60% to 100%, falling in the high RH range examined in previous chamber studies (Vesna et al., 2009; Weitkamp et al., 2008). According to Zhou et al. (2013), a phase change of SOA from semi-solid to more-liquid-like particles occurs at RH of 50%–70%. Under the humid condition in this work, aerosol morphology may have remained similar, rendering RH an insignificant influential factor for the uFAs loss rate.

To confirm that NO₃ or N₂O₅ was not the main oxidant leading to the observed decay of uFAs in the studied nighttime period, we examined the correlations of the effective rate constant k with NO₂ and O_x (O₃+NO₂) concentrations, which reflect the variation of ambient NO₃ (S. Wang et al., 2013). No positive correlation association was observed (Figure S12). Additionally, the relative uptake coefficient of NO₃ by C18:2 versus C18:1 was 1.77 (Zhao et al., 2011) and 2.06 (Gross et al., 2009) and of N₂O₅ was 2.69 (Gross et al., 2009). These values were notably higher than that obtained from the ambient data in this work (1.37 in Figure 3c). Thus, the cumulative evidence suggests that NO₃ and N₂O₅ were unlikely the major oxidants for uFAs oxidation in the nighttime in urban Shanghai.

3.4 Atmospheric Lifetime Estimation

With the knowledge of the pseudo first order reaction rates, we can estimate the atmospheric lifetime () of uFAs toward the heterogeneous reaction with O₃. They are listed in Table S1 and shown in Figure S13. Two lifetimes are estimated for the days that are described by the two-step model, for freshly exposed uFAs, and for partially protected uFAs. These days were mainly influenced by local air masses. At the nightly average ozone level of 3–8 ppb encountered on these nights, the average lifetime of freshly exposed uFAs was 3, 4, and 8 h for C18:2, C18:1, and t-C18:1, respectively. The low nightly ozone concentration level was likely a result of NO_x titration at our urban location. In comparison, the lifetime of the partially protected C18:2, C18:1, and t-C18:1 was longer by ca. one order of magnitude, with an average value of 22, 14, and 33 h, respectively, under similar gas phase ozone level (Table S1). On nights under influence of regional air masses, which were fitted with the one-step model, increased ozone exposure (11–14 ppb) was recorded, and the lifetimes of uFAs were accordingly shorter, at 4, 6, 8 h for C18:2, C18:1, and t-C18:1, respectively.

Many previous studies have used C18:1 as a model uFA in their laboratory investigations and subsequently estimated the lifetime of C18:1 using laboratory-derived kinetic data and certain O₃ exposure levels assumed for ambient conditions of 40–100 ppb (see Table 1). To compare with these lifetime estimates, we used the value for C18:1 derived in Section 3.2 and calculated the lifetimes under three O₃ exposure levels, 40, 50, and 100 ppb, to be 111, 89, and 44 min, respectively. Note that we have assumed a linear proportionality between particle-phase O₃ oxidant (C_ox) and gas-phase O₃ in the above calculation. The validity of this assumption may be subject to debate and awaits verification in future studies that cover a wider range of O₃ levels than was encountered in this work.

Our estimated lifetime of C18:1 is close to the lower limit from Weitkamp et al. (2008), estimated based on data using particles generated from residue grease and under an exposure of 40 ppb O₃ at 10% RH; also comparable to the study by King et al. (2009), examining the air–water interface matrix at an ozone level of 50 ppb, and the study by Ziemann (2005), examining pure oleic acid particles under exposure of 100 ppb ozone level at 10% RH. However, it is much longer than that reported for the simple aerosol matrix of liquid mixtures (e.g., a few minutes) in several studies (e.g., Knopf et al., 2005; Hearn & Smith, 2004). On the other hand, our estimated lifetime was significantly shorter than the results based on solid mixtures of pure and mixed monocarboxylic acid particles (44 min vs. 15 h under 100 ppb O₃) (Ziemann, 2005), as well as the estimate derived from residue meat grease particles under 50% RH (111 min vs. 24 h under 40 ppb O₃) (Weitkamp et al., 2008). The haphazard nature in the agreement between our ambient data-derived estimates and the previous laboratory data-derived estimates likely indicates the deficiency of the laboratory particle-ozone systems in simulating real-world conditions. This also highlights the necessity in producing more ambient measurements-derived kinetic data to test knowledge gained from the simpler synthetic systems.