Spatial variability of air pollutants in a megacity characterized by mobile measurements: Chemical homogeneity under haze conditions

. Characterization of the spatial distributions of air pollutants on an intracity scale is important for understanding localized sources, secondary formation, and human exposure. In this study, we conducted in situ mobile measurements for the chemical composition of fine particles, volatile organic compounds (VOCs), oxygenated VOCs (OVOCs), and common gas pollutants in winter in the megacity of Beijing. The spatial variations of these gaseous and particulate pollutants under different pollution conditions are investigated. During the less-polluted periods, a large spatial variability exists in the inorganic 15 composition of fine particles, suggesting a wide range of particle neutralization in Beijing. Significant spatial variations are also observed in the composition of organic aerosol (OA), which is mainly driven by local emissions of primary OA from vehicle and cooking exhaust. The spatial variations of VOCs and OVOCs vary by species. In general, hydrocarbon compounds show a large spatial variability driven by traffic emissions, while secondary OVOCs are more spatially homogeneous in concentration. Other gas pollutants show relatively low spatial variabilities, although hot spots of concentration frequently 20 appear which are plausibly caused by high-emitting plumes as well as fast on-road ozone titration. During the haze periods, the spatial variabilities of air pollutants are largely reduced because of the contribution of regional transport. Hydrocarbons and less-oxygenated OVOCs show good positive spatial-temporal correlations in concentration. More-oxygenated OVOCs show good positive correlations among themselves and moderate negative correlations with hydrocarbons, less-oxygenated OVOCs, and particulate components. The results highlight the potential role of chemical homogeneity on the SOA production 25 in the megacity under haze conditions. On the other hand, the spatial heterogeneity of air pollution calls a future need of using fine-resolution models to evaluate human exposure and pollution control strategies. strategies as clean air actions continue. The results also indicate different human exposures to particles and gaseous pollutants on the intracity scale, which requires fine-resolution models to evaluate. Moreover, our results suggest about 30-40% contributions of high-emitting plumes to the on-road concentrations of aromatic species during the clean periods compared with 9-13% during the haze periods. Under haze conditions, stagnant weather and predominantly contributions from regional transport largely reduce the spatial variabilities of particulate and gaseous pollutants in the city. More-oxygenated OVOCs show good positive correlation among themselves and moderate negative correlations with hydrocarbons, less-oxygenated OVOCs, and PM 2.5 components, suggesting a homogeneous photochemical production on the intracity scale. This chemical homogeneity may lead to local production of semivolatile oxidation products that may enhance the SOA production under winter haze conditions in the megacity.


Introduction
Air quality degradation has become a serious problem in developing countries (Huang et al., 2014;Khuzestani et al., 2017).
Cities in northern China suffer from severe haze events, especially in winter (Huang et al., 2014). High concentrations of 30 volatile organic compounds (VOCs) and nitrogen oxides (NOx = NO + NO2) contribute to high atmospheric oxidation capacity to produce ozone (O3) and secondary aerosol (Lu et al., 2019b;Wang et al., 2017). The mass concentrations of particles having aerodynamic diameter less than 2.5 µm (PM2.5) may reach up to several hundreds of μg m -3 and O3 concentrations may exceed 100 ppbv in megacities. Both of high exposures of PM2.5 and O3 are of great concern to human health Bell et al., 2006). 35 Tremendous efforts have been made to characterize air pollutants in populated urban areas in China Quan and Jia, 2020). In most studies, stationary measurements were conducted to obtain temporal variations of pollutants for the analysis of sources and formation mechanisms. Sulfate, nitrate, ammonium, and organic aerosol (OA) are the major chemical components of PM2.5 in cities in China . Their mass concentrations show decreases nationwide since 2013 as a result of the emission reduction of gaseous precursors such as sulfur dioxide (SO2), NOx, and VOCs 40 under clean air actions as well as the changes of meteorological conditions (Zhang et al., 2019). In Beijing, the contribution of secondary OA (SOA) has significantly increased since 2018, while the contribution of primary OA (POA) from burning of coal and biomass (biofuel) and cooking emissions have decreased (Lei et al., 2021). Characterization of organic precursors including VOCs and semivolatile and intermediate volatility organic compounds (S/IVOCs) is crucial for understanding the SOA production. Previous studies have shown that besides alkanes, aromatics and oxygenated VOCs (OVOCs) are abundant 45 in cities in China (Mozaffar and Zhang, 2020;Li et al., 2016;Guo et al., 2017). These VOCs may come from a variety of anthropogenic sources (e.g., vehicles, industries, solvent usage, coal and biomass burning, and household use of volatile chemical products etc.) and have great potential to form O3 and SOA.
On an intracity scale, local emissions, advection, and boundary layer evolution may affect the spatial distributions of air pollutants. Tower measurements have shown greater vertical gradients of particulate matter (PM) concentration and 50 composition during clean periods than haze periods in Beijing, which is attributed to the influence of both physical (e.g., regional transport, mixing layer height, and inversions etc.) and chemical processes (e.g., gas-particle partitioning and aqueous processing) (Zhou et al., 2018). Significant vertical gradients also present in the concentrations of VOCs in cities, which can be affected by transport of air mass, vertical diffusion, and species reactivity (Sun et al., 2018;Mo et al., 2020). Moreover, grid sampling and simultaneous measurements at multiple sites are often used to investigate the spatial variations of air 55 pollutants on sparse spatial scales. Previous studies show that the surface concentrations of VOCs and PM2.5 components vary because of different source contributions in metropolitan areas in China Song et al., 2021).
To better characterize the spatial distributions of air pollutants, mobile laboratory equipped with fast response in situ instruments have been widely used in Europe and United States (Kolb et al., 2004). For example, mobile measurements in https://doi.org/10.5194/acp-2021-651 Preprint. Discussion started: 24 August 2021 c Author(s) 2021. CC BY 4.0 License.
Pittsburgh and Zurich indicate different spatial variability of PM composition (Tan et al., 2014;Gu et al., 2018;Ye et al., 2018;60 Mohr et al., 2011;Elser et al., 2016). While primary carbonaceous sources perhaps lead to the large spatial heterogeneity of PM composition in Pittsburgh, a rather uniform distribution was observed in the metropolitan area of Zurich especially during thermal inversions over the Swiss plateau, highlighting the complexity of urban air pollution. However, the development and application of advanced mobile laboratory are lagged in China. Mobile measurements have been conducted to obtain the spatial distributions of SO2, NOx, carbon monoxide (CO), and black carbon (BC) as well as NO2 and SO2 column densities for the 65 evaluation of regional transport routes in North China Plain (NCP) and urban air quality (Wang et al., 2009;Wang et al., 2011;Zhu et al., 2016;Wu et al., 2018). Recent deployments of online mass spectrometers in mobile laboratory have successfully captured the spatial distributions of VOCs and PM2.5 composition in cities in China, which can be particularly useful for identifying emission sources (Wang et al., 2021c;Liang et al., 2020;Liao et al., 2021).
In this study, we conducted in situ sampling with gas analyzers and advanced mass spectrometers on a newly developed mobile 70 laboratory platform in winter in the megacity of Beijing. Spatial distributions of the mass concentrations of non-refractory PM2.5 (NR-PM2.5) components (i.e., OA, nitrate, sulfate, ammonium, and chloride), VOC and OVOC species, and common gas pollutants such as SO2, CO, NO, NO2, and O3 are obtained. The spatial variabilities of these gaseous and particulate pollutants under different pollution conditions are investigated to provide new insights to urban pollution and to improve our understanding of human exposure. 75

Methods
On-road measurements were conducted by the PKU mobile laboratory ( Figure S1 in the Supplement) on the 4th Ring Road of Beijing on 7-21 November 2018. The 4th Ring Road is a 65-km-long highway that passes through commercial, industrial, and residential areas. To show the spatial characteristics, the measurement area is categorized into 4 zones: Zone NE, Zone NW, Zone SE, and Zone SW ( Figure S2 in the Supplement). Zone SE and Zone SW are close to industrial facilities, while Zone NE 80 and Zone NW are mostly commercial and residential areas. The average traffic volume on the 4th Ring Road is about 4000-8000 vehicles per hour, and over 90% of the fleet consisted of light duty gasoline vehicles (LDGVs) that meet National Stage III to V emission standards.
The measurements started at around 9:00 a.m. and went one cycle after another till 4:00 p.m. Each cycle took approximately 70 min with a mean speed of 60 km h -1 . Self-contamination from the vehicle exhaust is negligible for a vehicle speed over 30 85 km h -1 (Liao et al., 2021). A suite of instruments was deployed in the mobile laboratory. Gas pollutants were detected by gas analyzers including NO2 (Teledyne, T500U), NO-NOx (Ecotech, EC9841A), SO2 (Ecotech, EC9850A), CO (Ecotech, EC9830A), and O3 (Ecotech, EC9810A). The chemical composition of NR-PM2.5 was measured by an Aerodyne time-of-flight aerosol chemical speciation monitor (TOF-ACSM) with PM2.5 lens and a capture vaporizer . The uncertainty for the mass concentrations of NR-PM2.5 and its components is about 30% (Canagaratna et al., 2007). The positive 90 https://doi.org/10.5194/acp-2021-651 Preprint. Discussion started: 24 August 2021 c Author(s) 2021. CC BY 4.0 License. matrix factorization (PMF) analysis was performed on the OA mass spectra by using the Igor PMF evaluation tool (PET, version 3.00B). Five OA factors were resolved including hydrocarbon-like OA (HOA), cooking-related OA (COA), and three oxygenated OA (OOAs). The common OA factors related to biomass burning (BBOA) or coal burning (CCOA) were not resolved in this data set and were perhaps mixed with OOAs (Liao et al., 2021). Their contributions to OA were however expected to be small because of the stringent emission control in NCP Duan et al., 2020). VOCs and 95 OVOCs were detected by an Ionicon proton transfer reaction-quadrupole ion guide time-of-flight mass spectrometer (PTR-QiTOF). The overall quantification uncertainties are <20% for calibrated species and 19% and 33% for uncalibrated species with known and unknown reaction rate constants (Kptr), respectively (Huang et al., 2019). Table S1 in the Supplement lists the Kptr values, the mean concentrations, and the tentative categorization of VOCs and OVOCs measured herein. Detailed information about the campaign, sampling system, instrument operation, calibration and data analysis has been described 100 elsewhere (Liao et al., 2021). Figure S3 in the Supplement shows the time series of mass concentrations of air pollutants measured at the PKU campus roof site during the campaign. This site is located between the 4th and the 5th north Ring Roads, representing a typical urban environment in Beijing (Cheng et al., 2021). In this study, we classified clean (< 35 µg m -3 ), haze (> 75 µg m -3 ), and non-haze (< 75 µg m -3 ) days based on the daily mean PM2.5 mass concentrations for the measurements. Moreover, we computed a batch 105 of 12-hour backward trajectories for the height of 3 m around the 4th Ring Road from the HYSPLIT model to investigate the influence of urban background and regional transport on our measurements during the haze day (Stein et al., 2015). We set the start time of the trajectory analysis at 9:00 a.m. and repeated the trajectory calculations every 1 hour until 4:00 p.m. ( Figure   S4 in the Supplement). In addition, a moving average window of 10 (i.e., 20 s) was applied to the entire data set for VOC and OVOC species. Concentration baselines for each 2-s point in the 20-s smoothed data were calculated as the 5th percentile 110 concentration within a rolling window of 60 (i.e., 120 s), for which the urban background conditions were identified (Apte et al., 2017). We then computed the peak fraction (PF) parameter as 1 minus the ratio of the mean baseline value to the observed mean concentrations. The time series of the concentrations of common VOC and OVOC species shows significant plumes above the baseline for the on-road measurements ( Figure S5 in the Supplement). PF values are used as a semi-quantitative indicator to show the contributions of the above-baseline concentrations to the observed concentrations of VOCs and OVOCs, 115 which may represent the intensity of transient emission from local sources (e.g., vehicle exhaust).

Spatial distribution of NR-PM2.5 and its components
A significant spatial variability of the mass concentrations of NR-PM2.5 and its chemical components is observed during the campaign. As described briefly by Liao et al. (2021), the mean ( one standard deviation) mass concentration of NR-PM2.5 for 120 non-haze days is 30.7  24.7 µg m -3 , in which OA is the most abundant component that contributes to about 52% of the NR-  Similarly, the measurements in Pittsburgh show a significant spatial heterogeneity of primary carbonaceous components such as HOA, COA, and BC and less spatial variabilities of OOAs . (Table 1), indicating a relatively greater spatial variability of 135 inorganic composition compared with organic composition. As shown in Fig. 1b, ammonium is the predominant inorganic component in Zones NW and SW, while nitrate contributes most in Zone SE. The CV values for ammonium are the highest in both mass concentration and mass fraction. The molar ratio of NH 4 + : (2× SO 4 2− + NO 3 − ) range from ~ 0 to 1.0, suggesting a wide range of particle neutralization on the intracity scale under clean-day conditions in winter Beijing, which is perhaps driven by the difference in chemical processes and local emissions of precursors. In NCP, the aerosol mass can be sensitive to 140 ammonia or nitric acid availability (Nenes et al., 2020). Traffic and agriculture (e.g., livestock) emissions are the main sources of ammonia in Beijing during the clean periods in winter (Sun et al., 2017;Pan et al., 2018). Traffic emissions depend on onroad vehicle type and volume and may vary greatly in the city. On the other hand, the concentrations of nitric acid depend on the concentrations of NOx and hydroxyl radical (OH) that are affected by local emissions and atmospheric oxidation and may also vary significantly in the city. Uniform solar radiation on the surface is not expected in the megacity because of the shades 145 of buildings as well as cloud covers. Therefore, the spatial heterogeneity of inorganic composition reflects spatially different thermodynamic regimes in the megacity environment, which may challenge the development of further pollution control strategies as clean air action continues. For comparison, the inorganic components have less spatial variability in Pittsburgh than in Beijing , which may be explained by the greater emissions of ammonia and NOx as well as higher OH concentrations in urban areas in China. inorganic (i.e., ammonium, sulfate, nitrate, chloride) and organic components (i.e., HOA, COA, and OOAs) are similar across the city (CV < 0.35), although a few hot spots of COA exist (Figure 1f). The similar chemical composition across the city is 155 consistent with the understanding of haze evolution in NCP during which regional transport becomes a main contributor to the PM pollution in Beijing. Over 60% of the PM1 and PM2.5 mass in Beijing can be contributed by regional transport during the winter haze episodes (Sun et al., 2014;Wu et al., 2021). The similar particle composition may suggest a chemical homogeneity on the megacity scale, and the north-south difference in mass concentration is perhaps driven by the differences in atmospheric dilution on the intracity scale (Sun et al., 2016;Chen et al., 2020). Moreover, the haze-day uniform spatial distribution of PM 160 composition is similar to the observations in the metropolitan area of Zurich when thermal inversions occur over the Swiss plateau and secondary pollution is built up (Mohr et al., 2011), highlighting stagnant metrological conditions as one of the key drivers of the intracity chemical homogeneity. In addition, the large spatial variabilities of the mass concentration and composition of NR-PM2.5 present in all measurement cycles during the clean day, and the compositions of NR-PM2.5 on the 4th Ring Road are similar in each cycle during the haze day, not limited to the noon cycle. 165 Figure 2 shows the spatial distributions of the mixing ratios of gaseous pollutants for the example noon cycles. The concentrations of SO2 are low (< 10 ppbv) throughout the campaign (Figure 2a). During the clean day, the SO2 concentrations are low in the north and relatively high in the west of the 4th Ring Road, which is different from the distributions of the NR-PM2.5 mass concentrations. The SO2 concentrations has a CV of 0.13. The low spatial variability of SO2 may be explained by 170 the predominant contribution of regional transport as a result of the relocation of steel industry and power plants out of Beijing and the ban of high emission vehicles from the road (Ge et al., 2018). Previous mobile measurements in Beijing captured specific transport events of SO2 from the south and found that the SO2 concentration decreased from Zone SE to Zone NW (Wang et al., 2011), which is consistent with our clean-day observations. During the haze day, the SO2 concentrations are lower than the concentrations observed during the clean day, which may be explained by the large conversion of gaseous SO2 175 to sulfate by photochemical and aqueous processes under winter-haze conditions (Wang et al., 2021b). The haze-day distribution of SO2 concentration is different from the clean-day distribution and has a greater CV value of 0.27. The 12-hour backward trajectories around the 4th Ring Road show that the air masses were mostly from the east in the morning, started shifting towards south at ~12:00 p.m., and then were mainly originated from the south around 4:00 p.m. ( Figure S4 in the Supplement). The spatial distribution of SO2 is therefore affected by the transport routes. Southwesterly, southeasterly, and 180 easterly flux pathways have been recognized as typical regional transport routes to Beijing (Chang et al., 2019). similarity of spatial distributions of these pollutants are consistent with the large contribution of regional transport to urban air pollution and the stagnant meteorology condition during the haze episode.

Spatial distribution of gaseous pollutants
The clean-day spatial distribution of NO2 concentrations is in general similar to that of NO, although more concentration hot spots exist (Figure 2d). The on-road NO2 are contributed by direct tailpipe NO2 emissions, secondary NO2 through on-road NOx titration, and background NO2. The tailpipe NO2 emissions are small for LDGVs (National Stage III and above) (Wu et 200 al., 2017). The urban background concentrations of NO2 measured at the PKU roof site on 18 November 2018 (clean day) are low at noon (5.2  0.6 ppbv). The high on-road NO2 concentrations (16.8-94 ppbv) are therefore mainly contributed by the fast on-road NOx titration, which is similar to the findings in North Carolina, U.S. . As shown in Fig. S7a in the Supplement, the clean-day on-road concentrations of O3 range from 3.2 to 15.4 ppbv, which are 3-4 times lower than the concentrations measured at the PKU roof site (40.0-44.5 ppbv). The spatial distribution of O3 concentration is opposite to 205 that of NO, which is consistent with the titration effect. The more-frequent appearance of hot spots in the clean-day NO2 distribution is likely because of the faster detection of NO2 by cavity attenuated phase shift technique than the chemiluminescence detection of NO. During the haze day, the spatial distribution of O3 concentration is in general opposite to that of NO but shows much lower concentrations (i.e., a few ppbv) than during the clean day, which can be explained by the stronger titration effect with higher NO concentrations. However, the concentrations of NO, NO2, and O3 are all high in Zone 210 NW. The backward trajectories suggest transport of air masses from the east and the south to Beijing. The high concentrations of air pollutants in Zone NW cannot be explained by regional transport and are perhaps related to local meteorological conditions affected by specific land surface conditions. The spatial distributions of VOCs vary by species. LDGVs emit aromatic hydrocarbons (e.g., benzene, toluene, ethylbenzene, and xylenes), some oxygenated species (e.g., acetaldehyde), and so on (Gentner et al., 2017). As expected, the on-road 215 concentrations of such primary VOC species are much greater than the concentrations measured at the urban and suburban sites in Beijing (Table S1 in the Supplement). hot spots for benzene and toluene are not the same, highlighting the difference in emissions from individual vehicles (Mo et 220 al., 2016). The toluene-to-benzene concentration ratio (T/B) is widely used to differentiate the VOC sources (Song et al., 2021; and references therein). According to literature, traffic emissions have T/B ratios from 1 to 3 depending on vehicle type and fuel composition. Industry emissions have a wide range of T/B ratios (1.4 to 5.8). Solvent emissions may have high T/B ratio of ~ 8.8. Biomass burning emissions typically have a low ratio of ~ 0.3. During the clean day, most of the concentration hot spots for benzene and toluene corresponds to T/B ratios of 1.1-2.6, which is consistent with vehicle emissions (Fig. S7b in the  225 Supplement). Some of the high T/B ratios (e.g., 2-3 in the south) are perhaps related to industry plumes. The T/B ratios for onroad background (not high-emitting plumes) are low (0.6-1.3), which may be explained by faster photochemical consumption of toluene than benzene. By contrast, the spatial variability for OVOCs (CV ~ 0.6) is much lower than that for hydrocarbons (Table 1), owing to their secondary nature because of the lack of localized sources in Beijing (Wang et al., 2021a). During the haze day, a remarkable enhancement of VOC concentrations is observed with mean enhancement ratios of 3.3-16.3 (e.g., 8.7 230 for benzene, 6.5 for toluene, 6.5 for C8-aromatics, 5.2 for methyl ethyl ketone, and 4.8 for propene) ( Figure S8 in the Supplement). The spatial distributions of benzene and toluene show consistent high concentrations in Zones NW, SW, and SE (Figure 2e-f) but low T/B ratios in those segments that indicate aged air masses because biomass burning sources are limited during the campaign (Fig. S7b) (de Gouw et al., 2005). The spatial distributions of benzene and toluene are different from the distributions of other air pollutants under haze conditions, suggesting that the influence of regional transport may vary by 235 species. Moreover, the haze-day spatial distributions of OVOC concentrations are quite homogeneous with relatively small CV values (i.e., 0.14 for Σ (Aldehyde & Ketones) and Σ (Acids & Anhydrides) in Table 1). The distribution of individual OVOCs is similar among species except for the OVOCs having 3-4 oxygen atoms in their formulae, which will be discussed further in Sect. 3.3. The calculated PF values suggest 30-40% contributions of high-emitting plumes to the on-road concentrations of aromatic species during the clean day compared with 9-13% during the haze day. For OVOCs (e.g., formic 240 acid), the contribution of high-emitting plumes to the on-road concentrations are comparable for the clean and haze days (i.e., smooth out some of the spatial variabilities caused by on-road vehicle emissions (Mellouki et al., 2015;Karl et al., 2018).
Under haze conditions, the spatial variability of hydrocarbons is slightly greater in the afternoon, which is probably because of the change of regional transport direction in the afternoon. The backward trajectories suggest that the air masses gradually 255 shifted from eastern to southern during the haze day ( Figure S4). For the two categories of OVOCs, the non-haze temporal variations are similar to that of hydrocarbons. Their CV values indicate much smaller spatial variabilities of OVOCs in the morning rush hour (8:00 a.m. to 10:00 a.m.) than that of hydrocarbons. This is consistent with previous source apportionment results that secondary production is the main contributor of most OVOCs in Beijing (Wang et al., 2021a). As the photochemical production proceeds during the day, the spatial variabilities of OVOCs increase and then decrease, similar to that of 260 hydrocarbons. On the other hand, the haze-day temporal and spatial variations of OVOCs are all similar to that of hydrocarbons (CV < 0.36), corresponding to the influence of regional transport. other during the clean day (Figure 5a). This is consistent with the predominant contribution of traffic emissions to the on-road concentrations. Secondary species like OVOCs and PM2.5 are more regional and do not correlate with these primary hydrocarbons ( Figure S9a in the Supplement). By contrast, the haze-day heatmaps show significantly different patterns ( Figure   5b and Figure S9b in the Supplement). Primary hydrocarbons not only correlate well with each other (Pearson's r > 0.7, α = 0.01) but also show good positive correlations with many less-oxygenated OVOCs (e.g., aldehydes and ketones with 1-2 270 oxygen atoms in their formulae (Table S1). These OVOCs may be contributed by both primary and secondary sources, and they can be formed and accumulated during the regional transport. The NR-PM2.5 components also correlate with these VOCs and OVOCs ( Figure S9b). Such good spatial-temporal correlations are consistent with the predominant contribution of regional transport to hydrocarbons, less-oxygenated OVOCs, and PM2.5.

Chemical heterogeneity under haze conditions
Interestingly, more oxygenated OVOCs (e.g., anhydrides and acids with 2-4 oxygen atoms in their formulae) that are mainly 275 contributed by secondary sources show good positive correlations with each other (r > 0.7, α = 0.01) and moderately negative correlations with hydrocarbons and less-oxygenated OVOCs as well as the NR-PM2.5 components. These species include CH2O2 (tentatively assigned as formic acid), C2H4O3, C3H4O3, C4H4O4, C4H4O3, C5H4O3, and C5H4O2. The spatial-temporal correlations for these more oxygenated OVOCs may be explained by a homogeneous photochemical production on the intracity scale. High oxidation capacity in winter in NCP has been reported (Lu et al., 2019a;Slater et al., 2020). Secondary OVOCs 280 are formed and accumulated along the transport route. Regional transport leads to relatively uniform spatial distributions of these species during the haze day (Figure 4). With a rather uniform distribution of hydroxyl radical which likely happens under winter haze conditions because of the thick cloud cover and increased particle scattering, the formation of more-oxygenated OVOCs from less-oxygenated OVOCs can be relatively homogeneous in the city. Although these more-oxygenated OVOCs https://doi.org/10.5194/acp-2021-651 Preprint. Discussion started: 24 August 2021 c Author(s) 2021. CC BY 4.0 License. may partition to the particle phase, the gas-to-particle partitioning are not expected to affect much the spatial variability of the 285 more-oxygenated OVOCs because similar chemical composition of PM2.5 present on the intracity scale under haze conditions. The secondary production and potential gas-to-particle partitioning explain negative correlations of the more-oxygenated OVOCs with their precursors (hydrocarbon and less-oxygenated OVOCs) and PM2.5. Moreover, the gas-particle repartitioning caused by temperature difference may affect the distribution of more-oxygenated OVOCs. This effect is however expected to be minor considering the OA loadings, vapor pressures of these more-oxygenated OVOCs, and possible temperature elevation 290 of 1-3 °C for the megacity Wang et al., 2020;Yang et al., 2013). Peroxyacetyl nitrate (PAN) is more sensitive to temperature and less stable than the other more-oxygenated OVOCs (Seinfeld and Pandis, 2016), which may explain lower correlation coefficient of its fragment (C2H4O3) in Fig. 5b.

Conclusions
In this study, we conducted on-road mobile measurements of air pollutants including NR-PM2.5 and its components, gaseous 295 pollutants, VOCs, and OVOCs under different pollution levels in winter in Beijing. During the clean periods, large spatial variabilities of the pollutants were observed, which can be largely attributed to the influence of localized sources. In particular, the inorganic composition of NR-PM2.5 vary greatly, indicating a wide range of particle neutralization on the intracity scale under less-polluted conditions in winter Beijing. For OA, the spatial variability is mainly driven by local HOA and COA emissions. The spatial variations of regular gas pollutants are relatively smaller than those of the PM2.5 components. The spatial 300 distributions of the concentrations of VOCs and OVOCs vary by species. Detected hydrocarbons (i.e. mostly aromatic species) are affected by on-road vehicle emissions, and thus show large spatial variabilities during the clean periods. The large spatial heterogeneity of the particle composition and VOCs may challenge the development of future pollution control strategies as clean air actions continue. The results also indicate different human exposures to particles and gaseous pollutants on the intracity scale, which requires fine-resolution models to evaluate. Moreover, our results suggest about 30-40% contributions 305 of high-emitting plumes to the on-road concentrations of aromatic species during the clean periods compared with 9-13% during the haze periods. Under haze conditions, stagnant weather and predominantly contributions from regional transport largely reduce the spatial variabilities of particulate and gaseous pollutants in the city. More-oxygenated OVOCs show good positive correlation among themselves and moderate negative correlations with hydrocarbons, less-oxygenated OVOCs, and PM2.5 components, suggesting a homogeneous photochemical production on the intracity scale. This chemical homogeneity 310 may lead to local production of semivolatile oxidation products that may enhance the SOA production under winter haze conditions in the megacity.
Data availability. Data presented in this manuscript are available upon request to the corresponding author.  0.61 0.14 n/a n/a Σ (Acids & Anhydrides) 0.63 0.14 n/a n/a Total 0.68 0.16 n/a n/a Others SO2 0.13 0.27 n/a n/a CO 0.26 0.25 n/a n/a NO 0.22 0.26 n/a n/a NO2 0.24 0.26 n/a n/a O3 0.22 0.24 n/a n/a https://doi.org/10.5194/acp-2021-651 Preprint. Discussion started: 24 August 2021 c Author(s) 2021. CC BY 4.0 License.