Measurements were conducted in Dezhou, Shandong Province, China (37.4341∘ N, 116.3575∘ E), as part of the “Photochemical Smog in China” project, which aims to evaluate haze formation in China and its implications for air quality policies (Hallquist et al., 2016). Instrumental measurements were performed at the Meteorological Weather Bureau of Pingyuan from November 2017 to January 2018. This season is of specific interest because previous wintertime measurements in this area have indicated that NACs can contribute as much as 50 ng m-3 to the mass concentration of PM2.5 (particulate matter). The temperature and relative humidity during the measurement period ranged from -11.7 to 20.9 ∘C and from 129 %–99 %, with campaign averages of 2.2 ∘C and 50 %, respectively. The wind speed averaged 2.4 m s-1; a time series of the metrological conditions during the experimental campaign is shown in Fig. 1. The daytime mass concentration of particulate matter (PM1.0) measured with an aerosol mass spectrometer typically exceeded the European Air Quality allowable limit for PM2.5 (25 µg m-3), and there were more than 10 pollution episodes with high aerosol loadings (>100 µg m-3). Organic matter comprised 60 % of the PM1.0 on average, and it contributed as much as 80 % in several field measurements. Inorganic nitrate (NO3-) accounted for 20 % of the measured PM1.0 levels on average. The diurnal profiles of total particulate matter and organic matter were similar, with two distinct peaks at 08:00 and 19:00 (LT) (Fig. 1).

A Filter Inlet for Gases and AEROsols (FIGAERO) coupled to a time-of-flight mass spectrometer (ToF-CIMS) was utilized to characterize the NAC content of the gas and particle phases. This instrument is described in detail in previous publications (Lopez-Hilfiker et al., 2014; Le Breton et al., 2018, 2019). Teflon tubing and copper tubing were used as sample lines for the gas and particle phases, respectively. The ToF-CIMS was operated in negative ionization mode with iodide (I–) as the reagent ion. High-purity N2 air (99.9 % purity) was flown over a glass vial containing methyl iodide (CH3I) and into a TOFWERK type-P X-ray ion source (operated at 9.5 kV and 150 µA) to create ions to charge compounds (MH) entering the ion–molecule region (IMR). The resulting product ions were identified either as molecular adducts with iodide (MHI-) or deprotonated ions (M-) (Eq. 1): 1H2OI-+MH→nH2O+MHI-/M-.

The ToF-CIMS was optimized to have an average spectral mass resolution (m/Δm) of 3000. ToF spectra showing ions with mass-to-charge (m/z) ratios between 7–620 Da were acquired with a time resolution of 1 s and averaged over 1 min for data analysis The ToF spectra were mass calibrated using four frequently occurring ion peaks: iodide monomer (I-, 126.904 m/z), dimer (I2- 253.809 m/z), and trimer (I3-, 380.713 m/z) and the NO3- ion (61.988 m/z). The signal of the reagent ion (iodide, m/z=126.904) provided the information on the drift of the signal of the mass spectrometer (MS). The variabilities of the raw iodide signal during the field measurement were less than 10 % and 20 % for the gas- and particle-phase analysis, which indicated the minimal drift of the CIMS signal. During the post-processing of the data, all signals from MS were normalized to the signal of the reagent ion to account for the daily variations/drifts. Gas-phase blank analysis was performed during the post-campaign calibration of the instrument

Aerosol particles were collected for 30 min with a PM1.0 cyclone and deposited on a Zefluor^® PTFE membrane filter. Analytes were then desorbed by passing heated N2 gas over the filter, with a temperature cycle from room temperature to 200 ∘C over 20 min, followed by a 10 min soak at 200 ∘C to ensure desorption of the compounds from the filter (a typical desorption profile is shown in the Supplement). The NACs were quantified by doping the PTFE filter of the FIGAERO with known amounts of freshly prepared authentic standards. The standards were analysed using the same thermal desorption procedure as for the aerosol particles. NACs with no available standard were quantified by applying sensitivities for compounds with similar chemical structures.

A high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS) was used to measure the composition and size distribution of particles with diameters below 1.0 µm (PM1.0). These measurements provided mass concentrations of particle-bound non-refractory species such as organics, sulfates, nitrates, ammonium, and chloride with a 4 min average time resolution (DeCarlo et al., 2006). Volatile organic compounds (VOCs), including some precursors of NACs, were monitored using a combination of two online gas chromatographs (GCs) with a mass spectrometer (MS) and flame ionization (FID) detectors, resulting in the detection of over 100 VOCs. Photolysis rates of ozone (j(O1D)) and NO2 (JNO2) were measured using a commercial spectroradiometer, which was calibrated using a high-power halogen lamp after the field campaign.

A series of box-model simulations was conducted to clarify the mechanism of NC formation during the second period of the experimental campaign (see below), when secondary chemistry dominated the formation of NACs. Simulations were performed using AtChem (Sommariva et al., 2020; http://https://atchem.leeds.ac.uk/, last access: 27 January 2021), an online zero-dimensional box model, together with chemical reactions extracted from the Master Chemical Mechanism (MCMv3.3.1) via the website (http://mcm.leeds.ac.uk/MCM, last access: 27 January 2021; Jenkin et al., 2003; Saunders et al., 2003). AtChem was previously utilized to simulate the formation of formic acid and nitrophenol at a site dominated by oil and gas production (Yuan et al., 2016). In our simulations, NC was used as a representative NAC to clarify the potential contribution of compounds emitted during biomass burning (i.e. catechol) to secondary NAC formation. The strong dependence of NC production on the overall rate of secondary formation and its significant mixing ratio during the second period (74 ng m-3) made this compound a suitable representative NAC for this purpose. The MCM (v.3.3.1) assumes that NC is the sole product of catechol, which greatly reduces the complexity of the model's calculations. Measured concentrations of inorganic gases such as CO, O3, NOx, and selected volatile organic compounds were used to constrain the simulations (Table S3). Over 90 VOCs were measured during the field campaign in Dezhou. However, to minimize the computational cost of the modelling process, only the 10 VOCs with the highest mixing ratios were included in the model. Additionally, compounds with multiple isomers (e.g. xylene and trimethylbenzene) were treated as single species to further reduce computational cost. The VOCs with the highest contribution to OH reactivity in four Chinese cities (Tan et al., 2019) were also included in the model to properly simulate the major oxidation reactions in the atmosphere. The VOC concentration assumed in the simulations amounted to 75 % of the total VOC concentration measured in Dezhou. Overall, a total of 1195 species (intermediates, products, etc.) and 3705 reactions (oxidation, photolysis, etc.) were included in the mechanism. The contribution of traffic sources was also analysed; results for these sources are presented in the Supplement.

Figure 2 compares mass spectra collected during a strong BB episode (red lines) to those for a typical clean day (black lines) in Dezhou. The intensities of several molecular ion peaks increased significantly during the BB episode, particularly those at 138.019 and 154.014 m/z, which correspond to the deprotonated masses of NP (C6H4NO3-) and NC (C6H4NO4-), respectively. These ions exhibited 6- to 8-fold increases in signal intensity during the BB episode, clearly indicating substantial increases in the concentrations of the corresponding compounds. By analysing the difference between polluted and clear episodes, 16 ions related to nitro-aromatic compounds were identified (see Fig. 2), including nitrobenzoic acid, methoxy/methyl NP, and DNP. The exact positioning and assignment of the functional groups of the nitro-aromatic compounds could not be determined because ToF-CIMS cannot differentiate between isomers, i.e. compounds with the same molecular formulas. High-resolution fitting results for individual peaks are presented in the Supplement.

The campaign-average mixing ratios of NACs measured for the gas and particle phase were 1720 and 299 ng m-3, respectively. The measured fractions of the 16 NACs in the particle phase (Fp) ranged between 9 % and 28 %, with a mean of 16 %; these results are consistent with those obtained in an earlier study that applied the same measurement technique during springtime at a suburban site in Changping near Beijing, China (Le Breton et al., 2018). The overall contribution of particle-phase NACs to the total concentrations of organic matter (mean: 1.9 %; range: 0.0025 %–21 %) and total PM (mean: 1.1 %; range: 0.0013 %–11 %) varied substantially over the campaign, which may indicate that multiple NAC sources and formation pathways influence the concentration of NACs. Since the analytical technique used in this work cannot account for all NACs, the mass loadings reported here should be treated as lower limits. Nevertheless, the mean concentrations measured in Dezhou were higher than the mixing ratios of total NACs reported in other studies (Teich et al., 2017; Wang et al., 2018, 2019; Kahnt et al., 2013), possibly because this work examined a greater number of NACs (16 compounds). Furthermore, concentrations of NACs measured in winter are normally significantly higher than those measured in the summer because the boundary layer is usually more shallow, and BB (wood combustion) occurs more frequently in winter.

The diurnal profile of total gas-phase NACs exhibits two distinct peaks: a broad peak around midday, between 13:00–16:00, and another in the evening (around 20:00 local time), as shown in Fig. 3. The mean daytime concentration of gas-phase NACs in Dezhou (2200 ng m-3) was almost twice the night-time mixing ratio (1400 ng m-3), presumably because either the rate of NAC production was higher during the daytime or the rate of loss was lower. These results stand in contrast to those of an earlier study, in which only night-time peaks were observed due to the daytime photolysis of NACs (Yuan et al., 2016). Mean daytime and night-time concentrations of particle-phase NACs were similar (304 and 300 ng m-3, respectively) but with clear variability, as shown in the corresponding diurnal profiles: there were two diurnal particle-phase peaks (08:00 and 20:00 LT), but both were less pronounced than their gas-phase counterparts. These maxima coincided with the peaks in the diurnal profiles of organic and particulate matter, suggesting that levels of particle-phase NACs were linked to the general occurrence of ambient aerosols. This may be due to enhanced partitioning towards the particle phase caused by increases in the organic aerosol mass. The diurnal profile of particle-phase NACs was comparable with the observed profile of nitrocatechol detected from residential wood smoke (Gaston et al., 2016).

Figure 3 also shows the mass contribution of lumped NAC categories in the gas and particle phases. NACs were assigned to lumped categories based on their structural similarity to the most common NACs reported in previous studies. Both the gas and particle phases exhibited similar percentage contributions for each category. NP and its analogues accounted for almost half of the total NAC concentration in both phases, which was assumed to be due to the strong influence of primary emission from BB events. NC and methylnitrocatechol, both of which are commonly used as biomass tracers (Iinuma et al., 2010; Finewax et al., 2018), individually accounted for as much as 9 % of the total NAC concentration. Interestingly, the diurnal profiles of NP did not follow the general trend of the other measured NACs. This suggests that its formation pathway differs from that of other NACs such as NC; it may be that the contribution of secondary formation is greater than that of direct primary emission from BB for NP.

Previous studies indicated that NACs mainly originate from BB, traffic, or secondary formation in the gas or condensed phase, with minor contributions from coal combustion (Hanson et al., 1983; Wang et al., 2018; Yuan et al., 2016; Xie et al., 2019; Chow et al., 2016). While previous studies found that traffic has important effects on NAC levels (Cecinato et al., 2005; Tremp et al., 1993), its influence appeared to be limited in this case: there was a weak to negative association between typical automobile exhaust VOCs (benzene, toluene, and trimethylbenzene) (Zhang et al., 2018; Geng et al., 2008; Batterman et al., 2002) and the studied NACs. Furthermore, concentrations of the measured NACs did not peak during or shortly after periods of high traffic intensity, in contrast to results obtained at three sites in Europe during cold and warm seasons (Delhomme et al., 2010).

NACs may also form in aqueous phases, particularly when the atmosphere has a high liquid water content (Harrison et al., 2005b; Vidović et al., 2018, 2020). The contribution of aqueous-phase oxidation to NAC formation was found to be limited based on the negative relationship (r=-0.1 to -0.6) between the relative humidity (RH) and the mixing ratios of NACs in the gas and particle phases. An earlier study found that the relative contribution of aqueous-phase oxidation to NAC formation in Beijing increased as the ambient RH increased (Wang et al., 2019). The conditions in Dezhou during the measurement campaign were relatively dry (the campaign average RH was 50 %), with few days exceeding 70 % RH as shown in Fig. 1.

Given the minor contributions of traffic and aqueous-phase oxidation, most of the subsequent analysis focused on primary BB emissions and secondary formation in the gas and particle phases.

The measurement period included many intense BB episodes that increased the concentrations of some NACs, particularly in the particle phase. To verify the association between these elevated concentrations and the observed BB episodes, the atmospheric behaviours of levoglucosan and nitrous acid (HONO) and the ratio of NOx to NOy (NOx / NOy) were used as tracers of BB. Levoglucosan is a commonly used molecular tracer of BB that is superior to other markers such as K+ and black carbon (BC) because it is much less prone to interference from non-BB sources (Zhang et al., 2012; Simoneit et al., 1999). Concentrations of HONO increase during BB episodes because the rate of conversion of NO2 into HONO is elevated in BB plumes due to the presence of aerosols with high surface areas (Nie et al., 2015). The ratio of NOx and NOy reflects the freshness of the emissions from combustion; levels of NOx in plumes originating from local BB are high, typically resulting in a ratio close to 1. The atmospheric transformation of NOx leads to the formation of NOz components such as HNO3 and organonitrates, causing the ratio of NOx to NOy to deviate from unity. In Fig. 4, the 3-week measurement campaign is separated into four periods corresponding to four distinct NAC formation regimes. Regimes 1 and 3 are associated with strong BB episodes based on the profiles of the three previously mentioned tracers. Levels of NACs in the condensed phase mirrored those of levoglucosan, which exhibited two strong peaks at 08:00 and 20:00 LT. Moreover, during apparent BB events, the timing of the peaks in NAC concentrations agreed well with that of the peaks in the diurnal OM profile. This demonstrates the apparently strong contribution of NACs to submicron aerosols in the studied city in rural China

Figure 4 also shows the gas and particle concentration time series of three representative NACs (NP, NC, and DNP). NP and NC are frequently reported to be the dominant NACs in field- and laboratory-based BB studies (Xie et al., 2019; Wang et al., 2017, 2018). Figure 4 clearly shows that the three NACs behave in quite different ways under the regimes linked to strong BB episodes. In these regimes, gas-phase NP concentrations correlated strongly (r=0.8-0.9) with those in the particle phase. This may be because there was a common dominant source of NP in the gas and particle phases or because of fast partitioning of NP between these phases during BB events. There was also good agreement between the gas- and particle-phase time series for the methoxy/methyl (C7H7NO3) and ethoxy/ethyl (C8H9NO3) derivatives of NP during BB regimes. NP and its analogues thus appear to be good direct tracers of primary emissions from BB events in Dezhou.

Conversely, the correlations between the gas- and particle-phase concentrations of NC and DNP were very weak (r=0.2-0.3), indicating that BB events had different effects on the formation and partitioning of these NACs (see Supplement for a complete correlation analysis). During typical clean days, particularly from 20 to 30 December (regime 2), the average correlation coefficient between the gas and particle phases for all NACs fell to less than 0.5, possibly because the contribution of photochemical processes to NAC formation was high relative to that of primary BB sources.

Secondary NAC formation may also occur during periods of extensive BB, resulting in mixed contributions to the observed NAC concentration. Differences in the mixing ratios and mass concentrations of NACs between BB regimes and relatively clean regimes can shed light on the relative contributions of primary emissions from BB and secondary production during each regime type. This can be demonstrated by considering the average NAC concentrations under regimes 1 (a strong BB regime) and 2 (a non-BB regime); pronounced differences (>50 %) in NAC concentration between these regimes can be considered indicative of the influence of BB on NAC production in Dezhou. The average recorded signal intensities for levoglucosan in the gas and particle phases under regime 2 were 52 % and 72 % lower, respectively, than those for regime 1. Moreover, some compounds, particularly NP and its methoxy/methyl and ethoxy/ethyl derivatives, exhibited significantly reduced mixing ratios in both gas and particle phases under regime 2. This supports the position that these compounds may be useful direct tracers of BB in Dezhou. Some NACs exhibited lesser declines (<30 %) in the gas-phase mixing ratio, suggesting that BB may not be their dominant source; for example, they may be primarily formed via secondary production. This behaviour was observed for dinitrated aromatic compounds such as C6H4N2O5 and C7H6N2O5, which can be formed via nitration of mononitrates (C6H5NO3 and C7H7NO3) (Yuan et al., 2016; Vione et al., 2005). Such nitration processes are mainly driven by secondary photochemical or multiphase reactions, explaining the comparatively small difference in the mixing ratios of these compounds between regimes 1 and 2.

Under regime 2, concentrations of BB markers fell dramatically, indicating that the influence of primary biomass emissions was limited (as shown in Fig. 4). Here, the diurnal profiles of gas-phase NACs (as shown in Fig. 5) exhibit increases in concentration at 14:00–15:00 LT and a minor night-time peak at 20:00 LT. Similarly, in contrast to the events during regime 1, the peak in particle-phase NAC concentrations occurred also in the afternoon at 14:00–15:00 LT. These peaks in the daily mixing ratios NACs coincided with the daily peak ozone concentration. Secondary photochemical formation was therefore probably the dominant NAC formation process under regimes 2 and 4. This conclusion is supported by the fact that the coefficient of determination (r2) between ozone and nitrophenol (see Fig. 5) under regime 2 (r2=0.7) is substantially higher than that for the full data set including BB regimes 1 and 3 (r2=0.1). The most pronounced reductions in r2 were observed for compounds expected to originate mainly from primary sources (e.g. NP); for compounds expected to be formed mainly via secondary production (e.g. NC and DNP), the r2 with ozone remained relatively high throughout the campaign (see Fig. 5).

Secondary production of NACs can be linked to the presence of specific precursor compounds (Harrison et al., 2005a). Figure 6 shows the correlations between levels of NP, NC, and DNP and those of their proposed precursors – phenol (C6H6O), catechol/dihydroxybenzene (C6H6O2), and NP. Phenol and catechol are primarily formed by the pyrolysis of lignins and can be precursors for secondary formation of NACs, particularly during BB events (Yee et al., 2013; Finewax et al., 2018; Gaston et al., 2016). Levels of NACs correlated strongly (r2=0.7–0.8) with those of their primary precursors (i.e. phenol and catechol). This indicates that nitration of these precursor phenolic compounds in the presence of OH or NO3 radicals was an important route of NP and NC formation. The figure showing the correlation between precursors and final products also shows the observed ozone mixing ratio, which is a measure of secondary photochemical activity. This further underscores the significance of photochemical oxidation in the formation of NC and NP from catechol and phenol. A similar relationship was observed for the secondary formation of DNP via further oxidation of nitrophenol. DNP is formed by the reaction of nitrophenol with OH or NO3 radicals to form nitrophenoxy radicals (NO2C6H5O⚫), whose subsequent nitration yields DNP (Yuan et al., 2016).

As shown in Fig. 4, the ratio of NOx to NOy, which is an indicator of plume freshness, was lower under regime 2 than regime 1, suggesting that an older plume was sampled in the former case. This aged plume may have contained residual traces of regional photochemical smog containing phenol, catechol, and their derivatives that were formed as primary emissions during BB events outside the studied region.

The yield of NACs produced by secondary formation is known to depend on the NO2 concentration (Wang et al., 2018, 2019; Yuan et al., 2016). For instance, NP is formed by the nitration of phenoxy radicals (C6H5O⚫), which are themselves formed by the OH/NO3-mediated oxidation of phenol (Berndt and Böge, 2003). Mechanistically, the formation of NACs such as NP should be heavily dependent on the atmospheric concentration of NO2. However, NACs such as NP and nitrosalicylic acid were formed consistently in a mountainous region of China, even when the NO2 concentration was below 5 ppb (Wang et al., 2018). The campaign NO2 average for this work in Dezhou was 23 ppb, with daytime and night-time means of 17 and 26 ppb, respectively. These mixing ratios may have been high enough to sustain the nitration of aromatic VOCs. However, a negative correlation was observed between NO2 and NACs in the gas (ravg=-0.598) and particle phases (ravg=-0.116) under regime 2, when secondary formation was the dominant source of NACs. In aged air masses such as those sampled during regime 2, NOx will be transformed into nitrated compounds (and HNO3), which may explain this negative correlation.

To further investigate the secondary production of NAC during the experimental campaign, box-model simulations were performed to model NC formation and loss using the AtChem tool and atmospheric oxidation chemistry models from MCMv3.3.1. Figure 7 shows the reaction pathway for the formation of NC by catechol oxidation initiated by OH or NO3 radicals. Unlike in the case of NP, only one precursor – catechol – can generate the intermediates (i.e. CATEC1O, CATEC1O2, and CATEC1OOH) in NC formation. Sinks of NC are its further oxidation by NO3 or OH radicals, which lead to stable ring-opening products such as 2-oxoacetic acid. Photolysis and deposition/dilution of NC were also accounted for in the simulation because of their reported importance in the gas-phase atmospheric loss of formic acid and NP (Yuan et al., 2015, 2016). The photolysis frequency of NC used in the simulations was based on the reported value for NP (1.4 % of the photolysis frequency of NO2). A sensitivity analysis (see Supplement) of the box model against variation of the effective physical loss rates (due to dispersion and deposition) indicated that a high loss rate (1 h) provided the best estimate of the observed NC mixing ratios. Physical loss terms with equivalent lifetimes above 1 h (e.g. 3 h) overestimated the measured NC concentrations by at least 50 %. As a result, the tail of the modelled daytime peak extended well into the night when using low physical loss rates. The loss rate used in this work was higher than the rates used in previous box-model analyses of formic acid and NP (Yuan et al., 2015, 2016) but is reasonable given the low vapour pressure of NC (2.1×10-4 Pa) (Finewax et al., 2018), which favours partitioning into the condensed phase.

Figure 8 shows time series of the observed and modelled mixing ratios of nitrocatechol under regime 2. The simulated mixing ratio profile agrees reasonably well with the experimental data, as indicated by the mean ratio of the modelled concentration to the observed concentration (Model/Obsavg=1.25) and the coefficient of determination (r2=0.51) between the two data sets. These results clearly show the explicit dependence of the secondary formation of NACs such as nitrocatechol on the oxidation of thermal degradation and pyrolysis products of lignins (e.g. catechol) in aged plumes. The modelling procedure overestimated the observed concentration from 27 to 31 December, which was attributed to the elevated mass aerosol mass concentration and increased RH which would favour the partitioning of the gas-phase NACs to the particle phase and a potential loss by condensed-phase processes/deposition. Additionally, the presence of the simulated daytime peak confirms that the rate of daytime production of NC (source) exceeded its rate of photolysis (sink) during the second period of the field campaign. If the daytime loss rate of NC due to photolysis is disregarded, the mixing ratio will only increase by 10 %, clearly showing the weak contribution of photolysis to the overall loss of NC. The primary pathways of NC loss were thus oxidation by OH radicals and night-time oxidation by NO3 radicals.

Figure 8 also shows the diurnal profiles of the modelled and observed NC concentrations. The modelled profile features prominent peaks at 10:00 (shoulder), 16:00, and 20:00 local time, but the experimentally observed afternoon peak in NC levels occurred around 14:00 LT. This discrepancy can be explained by the change of wind direction from north-west to north-east observed after 14:00 LT. It should be noted that the parametrization of AtChem does not account for meteorological effects and that it only partially accounts for dispersion via the effective physical loss rate parameter. The three daily maxima were attributed to the contributions of different sources of the intermediate hydroxyphenoxy radicals (CATEC1O) throughout the day. As shown in Fig. 7, the nitration of CATEC1O radicals is the only source of NC, so the production of hydroxyphenoxy radicals will dictate the overall rate of NC formation under excess NOx conditions. Figure 9 shows the relative contributions of the three major CATEC1O formation pathways. Note that CATEC1O can also be produced through photolysis of hydroperoxylphenol (CATEC1OOH) and the reaction of hydroxyphenylperoxy (CATEC1O2) with NO3 and RO2 radicals, as shown in Fig. 7. However, these pathways account for less than 0.05 % of the total CATEC1O production and were therefore disregarded. The daytime shoulder peak of NC at 10:00 was due to OH radical oxidation of catechol and accordingly coincides with the diurnal peak in the OH concentration (∼ 4.5×106 molec cm-3). The major formation pathways of OH radicals in Dezhou during wintertime were HONO photolysis and the reaction of HO2 with NO, causing measured OH production to peak in the afternoon rather than at midday, as is more common. Ozone photolysis, which is typically a major source of OH radicals, made a negligible contribution under the studied wintertime conditions. The simulated diurnal profiles of major oxidants are presented in the Supplement. Additionally, the box model indicated that the high levels of CATEC1O at 13:00 were predominantly due to the reaction of NO with hydroxyphenylperoxy radicals (∼ 50 %). Finally, the elevated levels of NC at 20:00 were primarily attributed to the very efficient (∼ 90 % conversion) NO3 night-time chemistry after sunset (16:30).

The general sources of NACs have been explored in previous works (Wang et al., 2017, 2019). The 16 NACs found in Dezhou were further categorized based on their main formation routes: primary BB and secondary formation. As noted above, secondary processes were dominant under regime 2, whereas both routes contributed under regime 1. The NACs detected under regime 1 were further classified based on the correlation between their gas-phase and condensed-phase concentrations. A strong correlation between the gas and particle phase (rg/p) was taken to indicate either that primary BB was the dominant source of the NAC in question or that it underwent rapid partitioning between phases. Five NACs exhibited rg/p values above 0.75 and thus behaved like NP. The gas- and particle-phase concentrations fell by at least 55 % and 85 % when the major source of NACs shifted from primary BB under regime 1 to secondary formation under regime 2. However, these five compounds were also formed under regime 2, suggesting that secondary formation does contribute to their presence. To assess the impact of each formation pathway under regime 1, two approaches were used. In the first approach, it was assumed that the degree of secondary formation was similar under regimes 1 and 2, which is reasonable based on the average ozone levels under each regime (O3avg=20 ppb). Both regimes had similar total gas and particle-phase concentrations of DNP, a product only formed by secondary oxidation, further supporting the validity of this assumption. This first approach was referred to as the DNP method based on the similar DNP profiles observed under regimes 1 and 2. By subtracting the concentrations of the five BB compounds under regime 2 from those under regime 1, it was determined that primary BB combustion processes accounted for 70 % of the observed concentrations of these compounds. The second approach used to estimate the contribution of primary BB to the measured NAC concentrations involved using levoglucosan as a primary source tracer. This approach is analogous to the widely used EC (elemental carbon) tracer approach, in which EC is used to distinguish the primary organic carbon (POC) fraction from secondary organic carbon (SOC) in total organic carbon (OC) measurements (Day et al., 2015; Cabada et al., 2004). The high time resolution levoglucosan (lev.) measurements were performed using the same instrument and conditions as the NAC measurements, so they provided good data coverage, making lev. a suitable tracer for primary BB. The relative contributions of primary emission (BB) and secondary (sec) formation for each NAC were estimated using the following expression: 2[NAC]BB=[NAC]/[lev.]BB×[lev.]3NACsec=NACTot-NACBB.

Here, ([NAC] / [lev.])BB is the ratio of the concentration of the NAC to that of lev. during strong primary combustion emission, and NACBB and NACsec are the fractions of NACs generated through biomass burning and secondary production, respectively. NACTot and lev. are the measured concentrations of NACs and levoglucosan in ambient measurements, respectively. Using this approach, primary BB combustion processes were found to account for 60 % of the total production of BB-related compounds under regime 1, in good agreement with the estimate obtained using the DNP method.

The secondary compounds were categorized as such based on the weak correlations between their gas- and condensed-phase concentrations (rg/p<0.5) under regime 1 and their association with ozone (Δr2 (NAC–O3)), which is indicative of formation via secondary chemistry. For NACs mainly formed via secondary oxidation (e.g. nitrocatechol), the coefficient of determination (r2) between ozone and NACs under regime 2 was similar to that for the full data set. This suggests that the formation pathways of these compounds did not change over the measurement period, regardless of the occurrence of combustion episodes. Table 1 shows the classifications of 13 of the 16 NACs examined in this work, including their total concentrations under regimes 1 and 2. The classifications were further supported by the correlations between the concentrations of the 16 NACs (Fig. 10), which clearly divide the NACs into different groups based on their behaviour in the gas phase under regime 1. The data in Table 1 were used to estimate the relative contributions of primary and secondary processes under the two regimes (Fig. 10), revealing that the relative abundance of NACs associated with secondary formation processes increased to 40 % in both the gas and particle phases under regime 2.