Comparison of Ozone and PM2.5 Concentrations over Urban, Suburban, and Background Sites in China

Surface ozone (O3) and fine particulate matter (PM2.5) are dominant air pollutants in China. Concentrations of these pollutants can show significant differences between urban and nonurban areas. However, such contrast has never been explored on the country level. This study investigates the spatiotemporal characteristics of urban-to-suburban and urban-to-background difference for O3 (Δ[O3]) and PM2.5 (Δ[PM2.5]) concentrations in China using monitoring data from 1171 urban, 110 suburban, and 15 background sites built by the China National Environmental Monitoring Center (CNEMC). On the annual mean basis, the urban-to-suburban Δ[O3] is −3.7 ppbv in Beijing-Tianjin-Hebei, 1.0 ppbv in the Yangtze River Delta, −3.5 ppbv in the Pearl River Delta, and −3.8 ppbv in the Sichuan Basin. On the contrary, the urban-to-suburban Δ[PM2.5] is 15.8, −0.3, 3.5 and 2.4 µg m−3 in those areas, respectively. The urban-to-suburban contrast is more significant in winter for both Δ[O3] and Δ[PM2.5]. In eastern China, urban-to-background differences are also moderate during summer, with −5.1 to 6.8 ppbv for Δ[O3] and −0.1 to 22.5 µg m−3 for Δ[PM2.5]. However, such contrasts are much larger in winter, with −22.2 to 5.5 ppbv for Δ[O3] and 3.1 to 82.3 µg m−3 for Δ[PM2.5]. Since the urban region accounts for only 2% of the whole country’s area, the urban-dominant air quality data from the CNEMC network may overestimate winter [PM2.5] but underestimate winter [O3] over the vast domain of China. The study suggests that the CNEMC monitoring data should be used with caution for evaluating chemical models and assessing ecosystem health, which require more data outside urban areas.


Introduction
Surface ozone (O 3 ) and fine particulate matter (PM 2.5 ) are two major pollutants in China (Qu et al., 2018;Shu et al., 2019). O 3 is formed by photochemical reactions between nitrogen oxides (NO x ) and volatile organic compounds (VOCs) (Sillman, 1999). Short-term exposure to high O 3 levels can increase the risk of respiratory and cardiovascular mortality, and long-term exposure even at low levels can affect human health (Turner et al., 2016;Mills et al., 2018). In addition, high O 3 concentrations ([O 3 ]) dampen leaf photosynthesis through stomatal uptake, inhibiting plant growth (Gregg et al., 2003) and decreasing ecosystem productivity (Yue et al., 2017). The exposure in China is greater than other, developed countries such as the U.S., Europe, and Japan . PM 2.5 is another major pollutant in China, especially in urban areas, due to high local emissions and regionally transported aerosols (Zhang and Cao, 2015). Haze episodes with high PM 2.5 concentrations ([PM 2.5 ]) can cause adverse health problems (Chow, 2006;Pope III and Dockery, 2013) and reduce net primary productivity of plants by limiting radiation and precipitation (Yue et al., 2017).
Observations have shown large differences in air pollution between urban (cities and megacities) and nonurban (suburban, rural, background, and remote) areas. In urban areas, greater volumes of traffic and residential activities increase anthropogenic emissions, such as carbon dioxide (CO 2 ) and NO x (Gregg et al., 2003;Pataki et al., 2006). In addition, the greater density of roads and buildings in urban areas changes the surface albedo and heat capacity, causing stronger heat-island effects than in nonurban areas (George et al., 2007). These differences can have substantial impacts on the contrast of O 3 and PM 2.5 between urban and nonurban regions. Studies have shown that nonurban [O 3 ] are usually higher than those in urban areas (Dueñas et al., 2004;Banan et al., 2013;Han et al., 2013;Yang et al., 2014;Tong et al., 2017). However, exceptions are also found in that the summer average [O 3 ] in urban Beijing is 33.4 ± 0.4 ppbv higher than in clean regions (Ge et al., 2012). Model results show that nonurban O 3 is sensitive to NO x , while urban O 3 is sensitive to both NO x and VOCs (Sillman et al., 1993;Xing et al., 2011;Jin and Holloway, 2015;Wang et al., 2017), leading to large uncertainties in the urban-to-nonurban difference of O 3 (Δ[O 3 ]). The urban-to-nonurban difference of PM 2.5 (Δ[PM 2.5 ]) is less complicated. With more primary and secondary pollutants produced in cities, the urban PM 2.5 level is usually higher than that in nonurban areas (Putaud et al., 2004;Barmpadimos et al., 2011;Bravo et al., 2016;Xu et al., 2016;Zheng et al., 2018).
In previous studies, the difference in air pollution between urban and nonurban areas has tended to be explored for a city (Han et al., 2013;Wang et al., 2015;Tong et al., 2017;Huang et al., 2018;Zheng et al., 2018;Zhao et al., 2019), several cities (Xue et al., 2014), or a certain region, such as the North China Plain (Xu et al., 2016), Yangtze River Delta , or Pearl River Delta (Zheng et al., 2010). However, the urban-to-suburban difference has not been compared among different regions. Since the year 2013, more and more suburban sites have been built to monitor regional pollution levels in contrast to urban sites. In this study, we investigate the differences of O 3 and PM 2.5 between urban and suburban areas in China using observations from a ground-based monitoring network during 2015-18. We pay particular attention to the spatial distribution and temporal characteristics of such differences. In addition, we use pollution data from 15 background sites built by the China National Environmental Monitoring Center (CNEMC) to compare O 3 and PM 2.5 concentrations over urban, suburban, and background sites in China. Details regarding the monitoring network are explained in the next section. Section 3 compares the pollution levels between urban and suburban areas, and attempts to interpret the causes. Section 4 compares the pollutant concentrations of urban and suburban sites with those of background sites. And lastly, section 5 discusses and concludes the study's main findings.

Site-level data
We use data from the CNEMC of China, including concentrations of O 3 , PM 2.5 , NO 2 and SO 2 from 1614 observation sites (http://www.cnemc.cn/sssj/). The time span of these sites ranges from 1 January 2015 to 31 December 2018. For data quality control, we choose 1281 sites with less than 20% missing data for O 3 and PM 2.5 during the monitoring period, thus ensuring these sites have relatively complete records during 2015-18. In addition, we use the daily maximum 8-h-average O 3 concentrations ([MDA8]) and PM 2.5 concentrations of background sites in 2017.
According to the requirements by the Ministry of Ecology and Environment (MEE, http://www.mee.gov.cn/), the national ambient air quality monitoring network includes three types of sites, including evaluation, comparison and background sites. Evaluation sites are obliged to be built in urban areas and distributed equally to cover the whole city. Comparison sites are built more than 20 km away from main pollution sources and urban centers, and background sites are built even farther (> 50 km) away. As is shown in Fig. 1, there are 1171 urban sites (evaluation sites), 110 suburban sites (comparison sites), and 15 background sites. The sites are densely located in the central and eastern parts of China, while those in the west and northeast are sparsely distributed. The locations of urban and suburban sites seem to overlap because they are usually only around 20 km away from each other. Most of the background sites are located in natural scenic areas, almost completely free of anthropogenic emissions.

Gridded data
CO 2 and NO x emissions are good indicators of anthropogenic activities (Gregg et al., 2003;Pataki et al., 2006). We use emissions data of CO 2 and NO x from the Multiresolution Emissions Inventory for China (MEIC, http://meicmodel.org) in 2016, which has a resolution of 0.25° × 0.25°. MEIC is a bottom-up emissions inventory that provides anthropogenic emissions of over 700 sources in China using a technology-based method . We derive sitelevel emissions from MEIC through their locations, and compare their differences between urban and suburban sites.

Four selected regions
We select four megacity clusters in China-namely, Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), Pearl River Delta (PRD), and Sichuan Basin (SCB)-as the major domains for analysis. These clusters have also been highlighted by the Chinese government as needing to reduce their air pollution . BTH includes 9 cities with 65 urban sites and 5 suburban sites; YRD includes 26 cities with 110 urban sites and 9 suburban sites; PRD includes 9 cities with 50 urban sites and 2 suburban sites; and SCB includes 21 cities with 75 urban sites and 14 suburban sites (Fig. S1 in the Electronic Supplementary Material, ESM). We use all of the urban and suburban sites to compare and quantify the pollutant concentrations within a region.

Comparison of urban and suburban emissions
We compare the probability density function (PDF) of CO 2 and NO x emissions between the urban and suburban CNEMC sites (Fig. 2). For CO 2 , 76% of suburban sites show emissions lower than 10000 t km −2 yr −1 , and this percentage is higher than that of urban sites (65%). In contrast, 19% of urban sites have emissions higher than 20000 t km −2 yr −1 , which is only 5% for the suburban sites.
For NO x , 80% of suburban sites show low NO x emissions of less than 20 t km −2 yr −1 , and this percentage is also higher  than that of urban sites (65%). The PDF shows that anthropogenic emissions are generally higher in urban than suburban areas, suggesting different pollution levels between urban and suburban regions.

Urban-to-suburban differences of air pollution
We focus on air pollution in the summer (June-July-August, JJA) and winter (December-January-February, DJF) during 2015-18. Figures 3a-b show the urban and suburban [MDA8] in the four regions. On average, the [MDA8] is higher in summer, with the highest level in BTH and the lowest in PRD (Fig. 3a). In contrast to summer, both the urban and suburban [MDA8] shows a peak in PRD but low values in BTH in winter. The low summertime MDA8 in South China is associated with large quantities of precipitation that wash out precursors in this season (Wang et al., 2017), while the high summertime MDA8 in North China is related to the high temperatures and solar radiation . [PM 2.5 ] in summer and winter, respectively. In summer, [PM 2.5 ] is highest in BTH and lowest in PRD, consistent with the distribution of [O 3 ] in the same season. In winter, the lowest urban and suburban [PM 2.5 ] are found in PRD, but the highest values are found in BTH for urban and YRD for suburban areas. Such a winter distribution of [PM 2.5 ] generally resembles its summer pattern, except that both the average level and variability are much larger in the cold seasons. The lowest urban and suburban [PM 2.5 ] in PRD are related to fewer coal-based industries and favorable meteorological conditions for atmospheric dispersion and dilution (Zhang and Cao, 2015). In comparison, the highest [PM 2.5 ] in BTH is associated with the stagnant weather (Chen et al., 2008), high local emissions (Zhang and Cao, 2015), and frequent regional transportation (Huang et al., 2014).

Figures 3c-d show the urban and suburban
To quantify the urban-to-suburban differences of air pollution, we subtract the average concentration of all sub-  urban sites from that of urban sites in the same region and detect the significance of the difference using the Student's t-test (significance level: < 0.05) ( Fig. 4 and Table S1). The Δ[MDA8] is negative for almost all regions, indicating that the suburban [MDA8] is higher than that in urban areas, except for YRD in summer (Δ[MDA8] = 2.7 ppbv). For YRD, high emissions of biogenic and anthropogenic VOCs (Liu et al., 2018) and the substantial NO x reductions (He et al., 2017;Song et al., 2017) convert a VOC-limited regime to a mixed sensitive environment (Jin and Holloway, 2015), leading to a positive (though nonsignificant) urban-to-suburban Δ[MDA8] via the higher urban NO 2 level (Fig. 4c). In contrast to MDA8, the Δ[PM 2.5 ] is generally positive in the four regions (Fig. 4b), suggesting that concentrations of urban PM 2.5 are usually higher than in suburban areas. Negative but nonsignificant Δ[PM 2.5 ] values of −0.04 (summer) and −0.6 μg m −3 (winter) are found in YRD.
We calculate the urban-to-suburban differences in the → NO 2 , SO 2 and NO 2 to O 3 ratio, and the PM 2.5 to PM 10 ratio (Figs. 4c-f), to determine the possible reasons for the differences of MDA8 and PM 2.5 . It should be noted that the [MDA8] between urban and suburban areas is significantly different only in BTH (−7.0 ppbv) and SCB (−6.3 ppbv) during winter. In these two regions, the urban NO 2 concentrations are significantly higher than the suburban ones by 6.0-10.0 ppbv (Fig. 4c). As O 3 can be titrated by NO via the reaction NO + O 3 NO 2 + O 2 (Sillman, 1999;Murphy et al., 2007), the higher level of NO 2 in urban areas indicates strong conversions from NO to NO 2 (Tong et al., 2017), leading to higher O 3 loss and lower [MDA8] in urban areas (Fig. 4a). This is also evidenced by the highest NO 2 to O 3 ratios over urban sites in BTH and SCB (Fig. 4e)  4c-d). Furthermore, the PM 2.5 to PM 10 ratio over urban sites is larger than in the suburbs (Fig. 4f), suggesting that secondary formation of fine particles contributes more than primary emissions in the urban areas of BTH.

Temporal variations of urban-to-suburban differences
We quantify the diurnal, weekly, seasonal, and interannual variations of the urban-to-suburban differences of O 3 and PM 2.5 in the four regions (Fig. 5). The hourly [O 3 ] is used to study the diurnal variation (Fig. 5a). During daytime, the absolute Δ[O 3 ] peaks at 0800 LST (local standard time) in most sub-regions, especially for BTH (−8.7 ppbv) and SCB (−5.5 ppbv), likely due to high NO x emissions from traffic in the rush hour (Dominguez-Lopez et al., 2014). Traffic emissions are also an important driver for Δ[PM 2.5 ], the peak of which (21.0 μg m −3 ) is found at 0800 LST in BTH (Fig. 5e, Table S2). In addition, high values of Δ[PM 2.5 ] may be caused by relatively low boundary-layer heights (Zhang and Cao, 2015) and weaker turbulence (Miao et al., 2016) in urban areas.
We use the daily [O 3 ] to study the weekly variations of urban-to-suburban differences (Fig. 5b). The ozone weekend effect (OWE) indicates that the daily mean [O 3 ] (not [MDA8]) is lower on weekdays than weekends owing to lower anthropogenic NO x emissions at weekends (Tong et al., 2017). However, our results do not find the OWE in all sub-regions, except for the urban areas in YRD and PRD, where the differences between weekday and weekend [O 3 ] are nonsignificant (Table S3). For suburban areas, a positive Δ[O 3 ] between weekdays and weekends is found, with a maximum difference of 6.8 ppbv in YRD. No significant differences of Δ[PM 2.5 ] are found between weekdays and weekends (Fig. 5f).
Both the Δ[MDA8] (Fig. 5c) (Table S2). On an annual mean basis, the suburban [MDA8] is higher than the urban value by 3.7 ppbv in BTH, 3.5 ppbv in PRD, and 3.8 ppbv in SCB. In comparison, the [PM 2.5 ] in suburban areas is lower than the urban value by 15.8 μg m −3 in BTH, 3.5 μg m −3 in PRD, and  Table S2.

μg m −3 in SCB.
The variations of urban-to-nonurban differences of air pollution are related to the ambient pollution levels. Figure 6 illustrates the variations of Δ[MDA8] at different ranges of urban [MDA8] on a daily basis from 2015 to 2018. In BTH and SCB, the median Δ[MDA8] shifts from a negative to positive value with an elevated urban [MDA8], suggesting that the increase of [MDA8] at urban sites is faster than at suburban sites. In summer, the urban [MDA8] can be either high (e.g., sunny days) or low (e.g., rainy days) on different days. As a result, the positive and negative Δ[MDA8] values may offset each other, leading to a limited average Δ[MDA8] (Fig. 4a). In winter, the urban [MDA8] is usually low, leading to a strong and negative Δ[MDA8] in these sub-regions (Fig. 5c). In comparison, the Δ[PM 2.5 ] changes from near zero to more positive values with the increase of urban [PM 2.5 ] in all four regions (Fig. 7). As the [PM 2.5 ] rises, there is an overall increasing trend and variability of Δ [PM 2.5 ]. This suggests that the [PM 2.5 ] in urban areas grows faster compared to in suburban areas during pollution episodes, and the Δ[PM 2.5 ] is linearly dependent on the urban [PM 2.5 ]. As a result, the Δ[PM 2.5 ] shows large positive values during winter season, when the urban [PM 2.5 ] is usually high (Fig. 4b).

Comparison of air pollutants over urban, suburban and background sites
In total, there are 10 background sites in the central-east- ern China region (18°-43°N, 100°-125°E), the number of which is much smaller than that of urban and suburban sites. Here, we compare the annual mean [MDA8] and [PM 2.5 ] at these background sites to the nearby urban and suburban sites within a 2° × 2° grid cell (Table 1 and Table S4). On average, the background [MDA8] (37.8-55.0 ppbv) is higher by 12% than in urban areas, and by 5% than in suburban areas. We find better correlations of [MDA8] between the background and suburban sites (R = 0.8) than those between the background and urban sites (R = 0.4) (Fig. 8). In contrast, the [PM 2.5 ] over background sites (6.7-33 μg m −3 ) is lower by 45% than in urban areas, and by 30% than in suburban areas (Fig. 9), with a higher correlation coefficient between background and suburban sites (R = 0.8). As for the regression fits, suburban values are closer to the background concentrations for O 3 , consistent with the findings in previous studies (Tong et al., 2017;Huang et al., 2018). We further calculate the Δ[MDA8] and Δ[PM 2.5 ] for summer (JJA) and winter (November-December, ND) between urban and background sites in 2017. Results show that the absolute urban-to-background (urban minus background) differences of [MDA8] and [PM 2.5 ] are much larger in winter than summer (Fig. 10), consistent with the seasonal variations of urban-to-suburban differences (Fig. 4). In summer, a moderate contrast of air pollution (−5.1 to 6.8 ppbv for Δ[MDA8] and −0.1 to 22.5 μg m −3 for Δ[PM 2.5 ]) is found between urban and background sites (Table S5). However, such a contrast is much larger and more significant in winter (−22.2 to 5.5 ppbv for Δ[MDA8] and 3.1 to 82.3 μg m −3 for Δ[PM 2.5 ]). Exceptions of positive Δ[MDA8] are found at sites 4, 5, 6 and 9 in JJA (Fig. 10a), suggesting that the sign of urban-to-background Δ[O 3 ] is not uniform on the country level during summer.

Discussion and conclusions
We compare our results with previous studies and find both agreements and differences. Studies have shown that diurnal variations of urban [O 3 ] are opposite to those of NO x emissions, which peak at 0800 LST during the rush hour (Zheng et al., 2010;Al-Rashidi et al., 2018). Such variations may result in negative peaks of Δ[O 3 ] within a diurnal circle, consistent with our findings (Fig. 5a). We do not find an OWE in BTH (Table S3), though some weekend effects in Beijing have been reported by . The positive Δ[MDA8] during spring and summer in YRD is quite different from those over the three other regions (Fig. 5c), likely because of regionally high emissions of both biogenic and anthropogenic VOCs (Liu et al., 2018) and the substantial NO x reductions (He et al., 2017;Song et al., 2017) transform a VOC-limited regime to a mixed sensitive environment over YRD (Jin and Holloway, 2015) and result in the increase in urban O 3 . In terms of interannual variation, Huang et al.(2018) found that the urban [O 3 ] in Shenzhen, a city in PRD, increased faster than its nonurban counterpart during 2012-17, leading to a decline in the absolute Δ[MDA8]. A similar trend is also present in our analyses for the whole PRD domain during 2015-18 (Fig. 5d).
For PM 2.5 , Zheng et al. (2018)  at 1100 LST and 2300 LST respectively. The first occurs three hours later than our results for 2015-18. At the seasonal scale, the absolute Δ[PM 2.5 ] was found to be larger in winter than in summer, because of urban emissions from heat-ing in the cold season, which is consistent with our results (Fig. 5g).    Table 1. The black dots denote that the difference is statistically significant (P < 0.05). 2013). Such a trend is more significant in urban regions than in nonurban areas (Lin et al., 2018), explaining the downward trends of Δ[PM 2.5 ] in BTH, YRD, PRD, and SCB, especially after the year 2016 (Fig. 5h).
It is important to acknowledge that there are some uncertainties in our analyses. One major source of uncertainty originates from the classification of urban and suburban sites based on the official documents of the MEE. It has been more than a decade since these suburban sites were set up in 2007. During this period, urbanization has been increasing rapidly, which may have turned some suburban sites, which were originally built far away from urban centers and pollutant sources, into urban sites (Yan et al., 2010;Yang et al., 2013). This may result in reduced differences of air pollutants like O 3 and PM 2.5 between urban and suburban areas from year to year (Figs. 5d and h). Furthermore, the number of suburban sites built by the CNEMC is far fewer than urban ones, leading to biases in interpolations and comparisons. After trying to use other information, such as administrative divisions and satellite-based land-cover types/built-up percentages, we found that the classification based on the MEE definition is the most effective way to distinguish urban and suburban sites. It is still valid for our study period because urban emissions with this classification are larger than the suburban emissions for 2016 (Fig. 2). In China, the paucity of background observation sites is a limitation for research into the effects of O 3 on ecosystems, as background information is needed for comprehensive validations of modeled O 3 (Yue et al., 2017). We find that the urban-to-background differences of O 3 are not significantly different by a large majority during summer, suggesting that data of urban sites from the CNEMC network can be directly used to study ecological effects of O 3 that are also mostly concentrated in summer.
In this study, we analyze the differences of O 3 and PM 2.5 between urban and suburban areas in four megacity clusters (BTH, YRD, PRD and SCB) at different time scales (diurnal, weekly, seasonal and interannual). We find that the differences vary in time and space, but the pattern whereby the suburban [MDA8] is higher and the urban [PM 2.5 ] is higher, dominates. However, obvious seasonal variations are observed. Both the urban-to-suburban and urban-to-background pollution shows a more significant contrast in winter (Figs. 4 and 10). According to national statistics (http://www. stats.gov.cn/), the total urban area was 2 × 10 5 km 2 in China in 2018, which was only 2% of the national total area. As a result, air quality monitoring sites built mainly in urban areas can reasonably capture country-level pollution in summer but may overestimate the national average [PM 2.5 ], and underestimate the [MDA8] in winter.