The Impact of the Numbers of Monitoring Stations on the National and Regional Air Quality Assessment in China During 2013–18

We used data from the China National Environmental Monitoring Centre (CNEMC), including hourly observations of PM_2.5, O₃, and NO₂ from 1436 air quality monitoring stations. The occasional outliers of observations were filtered out by a fully automated outlier detection method (Wu et al., 2018). To ensure the time continuity of data, we selected stations with less than 20% missing data for PM_2.5 and O₃ observations each year. The network in 2013 is taken as the standard observation network (SON) representing the unchanging situation of sites, while the actual network in each year is regarded as the dynamic observation network (DON) representing the changing situations of sites. We evaluate the impact of the change in numbers of sites on PM_2.5 and O₃ concentration assessments by comparing the estimated pollution concentrations in the two different observation networks. Since some sites included in 2013 were absent in the following years, we removed these missing sites from the SON and DON to ensure that each site in the SON existed in the DON.

Air pollution in China presents regional characteristics (Chan and Yao, 2008; Tao et al., 2014; Rohde and Muller, 2015). Therefore, the research areas of this study include all of China and its six important city clusters, namely, the Sichuan and Chongqing areas (SC), Hubei and Hunan areas (HH), Fenwei Plain (FWP), Yangtze River Delta (YRD), Pearl River Delta (PRD) and Beijing-Tianjin-Hebei region (BTH).

The number of sites in the two observation networks after the abovementioned data preprocessing is shown in Table 1. From 2013 to 2015, the national number of sites experienced rapid growth increasing by 3.3 times from the original 307 sites to 1322 sites. The numbers of stations in six key city clusters also have notable increases. The numbers of stations in four regions (all but the BTH and PRD) increased significantly. From the spatial distribution of sites in 2013 and 2015 across the country (Fig. 1a) and 6 key regions (Fig. 2), it can be seen that the coverage of sites was relatively small and the distribution of sites was concentrated in 2013, while the coverage of sites was significantly larger and the sites were more evenly distributed in 2015.

Figure 1.  The distribution of (a) the sites in 2013 (blue dots) and the newly added sites in 2015 (red dots) in China national air quality monitoring network and the population in 2015 in China. (b−h) annual average fine particulate matter (PM_2.5) concentrations and (i−o) annual average concentrations of daily peak ozone (O₃_Max) in China and the six key city clusters from 2013 to 2018 with red lines indicating the SON, blue lines indicating the DON and the standard deviation of the annual averages (the left axis). The bar charts show the concentration differences between the DON and SON with blue indicating lower concentration for the DON, green indicating higher concentration for the DON (the right axis) and hatching indicating that the difference is significant at the 0.05 significance level (T-test).

Figure 2.  The distribution of the sites in 2013 (blue dots) and the newly added sites in 2015 (red dots) in (a) SC, (b) HH, (c) FWP, (d) YRD, (e) BTH and (f) PRD and the population in 2015.

Studies have shown that there is a positive correlation between population density and pollutant concentration (Wang et al., 2019b). Moreover, NOx and VOC emissions can reflect the urbanization, industrialization, and local pollutant emission levels of a region (Degraeuwe et al., 2017; Luo et al., 2017). The increase in the number of air quality monitoring sites in China is accompanied by changes in the monitoring area, population, and pollutant emissions within the monitoring region. Thus, analyses of the population and NOx, and VOC emissions surrounding the sites are also included in this study. We calculate the averages of total population and NOx and VOC emissions in a 3 km radius around each site using the 2015 population data in China provided by the Socioeconomic Data And Application Center (SEDAC, 0.008333° × 0.008333°, https://sedac.ciesin.columbia.edu/) and NOx and VOC emissions data from the Tsinghua MEIC emission inventory in 2014 (0.25° × 0.25°, http://meicmodel.org/, Li et al., 2017b; Zheng et al., 2018). However, due to the low resolution of MEIC, we cannot calculate the total emissions in a 3 km radius around each site. Therefore, this study divides the original 0.25° × 0.25° coarse grids into 0.005° × 0.005° fine grids with equal weight.

Since the number of sites changed from 2013 to 2015, we calculated the total population and NOx and VOC emissions in a 3 km radius around each site in the SON (in 2013) and DON (in 2015). The averages of total population and NOx and VOC emissions of all sites are shown in Table 2. For the whole country, the averages of total population and NOx and VOC emissions of all sites significantly decreased from 249 000 people, 1899 tons, and 2603 million mol in 2013 to 157 000 people, 1073 tons, and 1303 million mol in 2015, respectively. Except the BTH, the other five city clusters have different population and emissions reductions in a 3 km radius around sites from 2013 to 2015. The reductions were particularly obvious in the SC and HH.

Assuming that each station represents a 3 km radius region around it, as a whole, the total area and population covered by the observation network in China increased from 7977 km² and 65.35 million people in 2013 to 31 127 km² and 158 million people in 2018, respectively. The increases in the area and population covered by the observation network are conducive to obtain an accurate realization of air pollution and provide more important surface observation data for air quality assessments and air pollution research in China.

The number of sites in the DON increased significantly from 2013 to 2018. Most newly added stations were located in regions with relatively sparse populations and low emissions except the BTH. We discuss the impact of these newly added stations on assessments of PM_2.5 and O₃ concentrations in China in the following sections.

To assess the impact of the growth of the observation network on PM_2.5 concentration assessment, we first estimated the national annual average PM_2.5 concentrations in the SON with 307 sites and the DON with the temporally variable number of sites. Figure 1b shows the variation in the national annual average PM_2.5 concentration in the SON and DON from 2013 to 2018 and the differences (DON minus SON) between the two observation networks. The national annual average PM_2.5 concentration in China shows an obvious downward trend during 2013−18 in the SON and DON, which is consistent with previous studies (Zhai et al., 2019; Shao et al., 2021). The downward trends in the SON and DON are nearly the same, but the estimated PM_2.5 concentrations are different. The estimated national annual average PM_2.5 concentration in the DON is slightly lower than that in the SON by 0.6–2.2 µg m⁻³. The largest concentration difference between the two observation networks is in 2015. The reason for these differences may be that the newly added sites in the DON are located in regions with low population densities and emissions in comparison with sites in SON, since the PM_2.5 pollution is higher in urban centers covered by the SON (Wang et al., 2017b; Gao et al., 2020), consequently lowering the estimated national annual average PM_2.5 concentrations in the DON. This result shows that although more than 1000 sites were added to China national air quality observation network during 2013−18, the impact of these newly added sites on the national annual average PM_2.5 concentration assessment is relatively slight, which also indicates the strong regional characteristics of PM_2.5 pollution in China.

To further analyze the impact of observation network changes on PM_2.5 concentration assessments in China's different city clusters, we estimated the annual average PM_2.5 concentrations in the SON and DON in the six city clusters. Figures 1c-h show the annual average PM_2.5 concentrations in the SC, HH, FWP, YRD, BTH, and PRD in the SON and DON from 2013 to 2018 and the differences (DON minus SON) between the two networks. Similar to the country as a whole, in the six city clusters, the downward trends of the annual average PM_2.5 in SON and DON are nearly the same, but the estimated PM_2.5 concentration values are different. In the SC, HH, and FWP, the annual average PM_2.5 concentrations in the DON are lower than those in the SON by 1.4−6.0 µg m⁻³ (Figs. 1c-e). However, the annual average PM_2.5 concentrations in the DON are slightly higher than those in the SON by 1.1−3.7 µg m⁻³ in BTH (Fig. 1g, not significant at the 0.05 significance level) and 1.1−3.1 µg m⁻³ in YRD (Fig. 1f). In the PRD, the annual average PM_2.5 concentration differences between the two networks are so minor (<1 µg m⁻³) that they can be ignored (Fig. 1h). The possible reason for the different results among city clusters is that there are different changes in observation networks in these regions. In the SC, HH, and FWP, the newly added sites in the DON are mostly located in the regions with relatively low population densities and pollutant emissions (Table 2), causing lower PM_2.5 concentrations in the DON. However, the newly added sites in the DON in the BTH are mostly located in areas with relatively high population densities and pollutant emissions (Table 2), causing higher PM_2.5 concentrations in the DON. The number of newly added sites in the PRD is miniscule (Table 1) and does not have an obvious impact on PM_2.5 concentration assessment. In the YRD, the newly added sites in the DON are also located in the areas with relatively low population densities and pollutant emissions, but the changes in the population and pollution emissions are not as significant as those in the SC (Table 2). The PM_2.5 pollution in the northern YRD is more serious than that in the other parts of YRD (Ma et al., 2019). Many newly added sites in DON are in the northern YRD so that the estimated PM_2.5 concentrations in the DON are higher (Fig. 1f). From the above results, we can see that the impacts of observation network changes on PM_2.5 concentration assessment are not identical between city clusters and have certain regional characteristics.

O₃ is a critical pollutant for atmospheric compound pollution in China. It is quite important to explore the influence of observation network changes on O₃ concentration assessment. The O₃ concentration variation in the boundary layer is complex and affected by not only meteorological conditions and regional transport but also precursor concentrations, photochemical reactions, and VOC/NOx sensitivity (Finlayson-Pitts and Pitts, 1997). The impact of observation network changes on O₃ concentration assessments in China is likewise complicated.

To explore the impact of observation network changes on estimated O₃ concentration, in a manner similar to PM_2.5, we first estimated the national annual average daily peak O₃ (O₃_Max) concentrations in the SON and DON. Figure 1i shows the variation in the national annual average O₃_Max concentrations in the two networks from 2013 to 2018 and the differences (DON minus SON) between the two networks. The trends of national O₃_Max in the two observation networks are different and opposite during 2014−15. The trend of the estimated national O₃_Max during 2014−15 is downward in the DON, while it is slightly upward in the SON. Furthermore, the estimated national O₃_Max concentration values in the two networks are also notably different. The national O₃_Max concentrations in the DON are 1.4−5.1 µg m⁻³ lower than those in the SON and the standard deviation of the differences is 1.5 µg m⁻³. The largest concentration difference between the two networks is in 2015. Based on the above results, we can conclude that the newly added sites not only led to the opposite trends of estimated O₃ concentration in China in the two networks during 2014−15, but also made the estimated values of O₃ concentration in the DON lower than those in the SON. This provides a new perspective for understanding the declining trend of estimated O₃ concentration during 2014−15 in the DON (Meng et al., 2017). In studies of the trend of O₃ concentration in China, it is necessary to fully consider the impact of observation station changes on the assessment. In addition, the impact of changes in the location of observation stations on the O₃ concentration assessment should also be considered in the layout of observation stations.

The reason why the newly added sites in the DON have the above influences on the assessment of the national average O₃ concentration is relatively complex, and we will explain it in detail in section 3.2.3.

To explore whether the impacts of observation network changes on different city clusters are similar to those in the country as a whole, we estimated the O₃_Max concentrations of six regions in the SON and DON. Figures 1j-o show the annual average O₃_Max concentrations in the SC, HH, FWP, YRD, BTH, and PRD in the SON and DON from 2013 to 2018 and the differences (DON minus SON) between the two networks. The newly added stations in the DON affect evaluations of the long-term trend and O₃_Max concentration in all regions except the BTH and PRD. The trends of O₃_Max in the DON are upward in 2016 compared with 2015 in the SC (Fig. 1j), slightly upward in 2014 compared with 2013 in the FWP (Fig. 1l), and downward in 2015 compared with 2014 in the YRD (Fig. 1m). These trends are completely opposite to those in the same period in the SON. In the HH, the O₃_Max decrease in 2014 compared with 2013 and increase in 2015 compared with 2014 in the DON are more obvious than those in the same period in the SON (Fig. 1k). Moreover, the differences in the estimated annual average O₃_Max concentrations between the two observation networks are different in the six regions. In the SC and FWP, the O₃ concentrations in the DON are higher than those in the SON and the differences are up to 4.9 µg m⁻³ in 2017 and 8.1 µg m⁻³ in 2014 respectively (Figs. 1j and 1l). However, in the HH and YRD, the O₃_Max concentrations in the DON are lower than those in the SON and the differences are up to 8.2 µg m⁻³ in 2014 and 8.0 µg m⁻³ in 2015 respectively (Figs. 1k and 1m). These results show that, except in the BTH and PRD, the impacts of the newly added stations in the DON on O₃ concentration assessments in the SC, HH, FWP, and YRD are quite significant and distinct in different regions. When considering individual city clusters, we need to account for this impact including site changes.

The reason why the newly added sites in the DON have these different influences on O₃ concentration assessment in the six city clusters is relatively complex, and will also be explained in detail in section 3.2.3.

The above discussion reveals that the newly added sites in the DON affect the evaluations of long-term trends and annual average O₃_Max concentrations in the country as a whole and its six city clusters. Considering the complexity of O₃, we also estimated the differences in the monthly average concentration of NO₂ and O₃_Max between the two observation networks in the country as a whole and its six city clusters. Figure 3 shows the differences (DON minus SON) in the national monthly average concentrations of NO₂ and O₃_Max between the DON and SON from 2013 to 2018. Except in winter, the O₃ concentrations in the DON are lower than those in the SON (Fig. 3a). Furthermore, the NO₂ concentrations at all time periods after 2013 in the DON are significantly lower than those in the SON. Figure 4 shows the differences (DON minus SON) in the monthly average O₃_Max concentrations between the DON and SON in the six city clusters from 2013 to 2018. In the SC, HH, and FWP, the monthly average O₃_Max concentrations in the DON are higher in winter and lower in summer than those in the SON (Figs. 4a-c). In the YRD, the monthly average O₃_Max concentrations in the DON are almost equal to those in the SON in winter but lower than those in the SON in summer (Fig. 4d). In the BTH and PRD, the differences in the monthly average O₃_Max concentrations between the DON and SON are negligible (Figs. 4e and 4f). Figure 5 shows the differences (DON minus SON) in the monthly average NO₂ concentration between the DON and SON in the six city clusters from 2013 to 2018. Similar to the country as a whole, the NO₂ concentrations in the DON in the six city clusters are significantly lower than those in the SON (Figs. 5a-d) in all regions except the BTH and PRD. As a whole, the seasonal difference of the impact of the newly added sites in DON on the O₃ concentration assessment in regions is more significant than that in the whole country, especially in the SC.

Figure 3.  The differences in the national monthly average (a) O₃_Max and (b) nitrogen dioxide (NO₂) concentrations between the DON and SON from 2013 to 2018 in China. Red indicates higher concentration for the DON, blue indicates lower concentration for the DON, and hatching indicates that the difference is significant at the 0.05 significance level (T-test).

Figure 4.  The differences in the regional monthly average O₃_Max concentrations between the DON and SON from 2013 to 2018 in the (a) SC, (b) HH, (c) FWP, (d) YRD, (e) BTH and (f) PRD. Red indicates higher concentration for the DON, blue indicates lower concentration for the DON, and hatching indicates that the difference is significant at the 0.05 significance level (T-test).

Figure 5.  Same as Fig. 4 but for the NO₂ concentrations.

In the country as a whole and most city clusters, the estimated O₃ concentrations in the DON are higher in winter and lower in summer than those in the SON. We postulate a viable mechanism leading to this seasonal difference. In winter, when the particulate matter pollution is serious, the extinction effect of aerosols weakens the photochemical reaction of O₃ formation (Wang et al., 2019c, 2020). Given that in winter the titration of NO on O₃ greatly suppresses O₃ concentration (Zhang et al., 2014) and O₃ production is in VOC-limited regime (Kang et al., 2021), the lower NOx in DON than in SON (Fig. 3b) leads to higher O₃ concentration in DON than in SON. However, in summer, owing to the intense solar radiation, the photochemical reaction of O₃ formation is greatly strengthened (Geng et al., 2008; Zeng et al., 2018). Due to the fact that O₃ production is in NOx-limited or transition regimes in summer (Wang et al., 2021a) and the NOx in DON is lower than that in SON (Fig. 3b), the concentration of O₃ in DON is lower than that in SON. We also note that the VOC emissions in DON are lower than those in SON (Table 2). But the VOC emissions in MEIC inventory only include anthropogenic VOC emissions excluding the biogenic VOC emission. The estimated anthropogenic VOC emissions in China in 2017 using MEIC inventory were 28.5 Tg (Li et al., 2019b), while the estimated biogenic VOC emissions in China in 2018 using the Model of Emissions of Gases and Aerosols from Nature were 58.89 Tg (Li et al., 2020). The role of biogenic VOC emissions on tropospheric photochemistry cannot be neglected (Xie et al., 2007). We surmise that the VOC emissions in DON will greatly increase and the differences of VOC emissions between the SON and DON will shrink after adding the biogenic VOC emissions (Wang et al., 2021b). Therefore, the NOx emission differences might play a major role in the seasonal change of O₃ concentration difference between the two networks. In addition, the O₃ production is not only proportional to concentrations of VOC, but also to their reactivities with OH radicals. For example, the aromatics accounts for 45% of the total O₃ production, although the concentration of aromatics is only 25% of the total VOC concentrations in Shanghai, China (Geng et al., 2008). After building more VOC monitoring stations in the future, we can better explore the impact of VOC differences on the air quality assessment difference between the SON and DON.

The impact of the newly added sites in the DON on O₃ concentration assessment caused by the above mechanism is distinct in different regions. In the BTH and PRD, the changes in the observation network are so small that we can ignore this impact (Figs. 4e and 4f). In the YRD, the differences in population density and emissions between the SON and DON are not as significant as those in the SC (Table 2). Therefore, the above seasonal differences are not obvious in the YRD (Fig. 4d). In HH, although the changes in newly added sites are similar to those in the SC, this region is a major pollutant channel for the north-south transmissions in China (Lu et al., 2019b), where the pollution is greatly affected by external area transmissions. The influence of newly added sites in the DON on the O₃ concentration assessment in the HH is weakened, so the seasonal differences are not as significant as those in the SC (Fig. 4b). Such seasonal differences in these regions are ultimately reflected in the differences in the annual average O₃ concentration between the two observation networks (section 3.2.1 and section 3.2.2).

Based on the above results, we know that the impacts of the newly added sites on O₃ concentrations and trend assessments are significant and show seasonal and regional differences. When evaluating O₃ concentrations, we need to fully consider this seasonal and regional impact.

China national air quality monitoring network experienced a rapid development with the construction of a substantial number of new sites during 2013−18. This study analyzed the impact of the newly added sites on the long-term trends and concentrations assessments of PM_2.5 and O₃ by taking the network in 2013 as the SON and the actual network in each year as the DON. The results are as follows: (1) An increasing number of sites has had almost no influence on the long-term trends of estimated national and regional PM_2.5 concentrations in China but has slightly affected the estimated PM_2.5 concentrations. The newly added sites lead to the estimated national and most regional (in the SC, HH, and FWP) PM_2.5 concentrations being lower because the monitoring network expanded from the urban centers to the suburban areas with relatively low population densities and pollutant emissions, where PM_2.5 pollution is less. (2) The newly added sites significantly affect the assessments of O₃ the long-term trends and concentrations in China. The long-term trends of O₃_Max in the DON and SON are opposite in some years, which occurs in the country as a whole (2014−15), the SC (2015−16), FWP (2013−14), and YRD (2014−15). Moreover, the estimated annual average O₃_Max concentrations in the DON are lower in the country as a whole, the HH, and YRD, but higher in the SC and FWP than those in the SON. (3) The impact of the increasing number of sites on O₃ assessment exhibits seasonal differences. The estimated monthly national and most regional (in the SC, HH, and FWP) O₃_Max concentrations in the DON are higher in winter and lower in summer than those in the SON. The main reason for this seasonal difference is the difference of ozone sensitivity to precursors in winter and summer.

Of course, there are some uncertainties in this study. To ensure the time continuity of observation data, the setting of the missing data rate determines which observation stations to use in this study. We attempted to use different missing data rates to select stations and found that 20% is the most appropriate for ensuring that the station data are fully utilized, which is consistent with the previous missing measurement rate (Gao et al., 2020). For assessing the national concentrations by averaging the observations of all the sites, the uneven distribution of stations and short lifetime for PM_2.5 and O₃ will unavoidably introduce bias and different species will have different representative errors (Piersanti et al., 2015; Li et al., 2019c). Limited by the actual distribution of existing monitoring stations, we chose this method of averaging all sites as in the previous national assessment (Lu et al., 2020b). Moreover, in addition to the emission differences, the reason for differences of estimated PM_2.5 and O₃ concentrations between the SON and DON may also include pollution control policies (Zhang et al., 2020), meteorological factors (Tong et al., 2017; Ariya et al., 2018; Bilal et al., 2019) and regional transmission especially for regional air quality assessment (Huang et al., 2018). Additional research is needed to explore these possible reasons. Based on the conclusions of this study, we know that the numbers of monitoring stations affect the estimated PM_2.5 concentrations, O₃ long-term trends, and the estimated O₃ concentrations in China. When we assess the air quality, especially for O₃ assessments, we need to fully consider the impact of the changes in numbers of monitoring stations on the assessment. In addition, in the process of developing the observation network, the station layout needs to take the influence of the newly added sites on the assessment into account. We need to minimize such influence as much as possible so that the assessment results are more objective and accurate.

Acknowledgements. We acknowledge the use of surface air quality observation data from CNEMC. This research was supported by the National Natural Science Foundation (Grant Nos. 41875164 & 92044303) and the National Key Research and Development Plan (Grant No. YS2020YFA060022).