Surface ozone pollution in China: Trends, exposure risks, and drivers

Introduction Within the context of the yearly improvement of particulate matter (PM) pollution in Chinese cities, Surface ozone (O3) concentrations are increasing instead of decreasing and are becoming the second most important air pollutant after PM. Long-term exposure to high concentrations of O3 can have adverse effects on human health. In-depth investigation of the spatiotemporal patterns, exposure risks, and drivers of O3 is relevant for assessing the future health burden of O3 pollution and implementing air pollution control policies in China. Methods Based on high-resolution O3 concentration reanalysis data, we investigated the spatial and temporal patterns, population exposure risks, and dominant drivers of O3 pollution in China from 2013 to 2018 utilizing trend analysis methods, spatial clustering models, exposure-response functions, and multi-scale geographically weighted regression models (MGWR). Results The results show that the annual average O3 concentration in China increased significantly at a rate of 1.84 μg/m3/year from 2013 to 2018 (160 μg/m3) in China increased from 1.2% in 2013 to 28.9% in 2018, and over 20,000 people suffered premature death from respiratory diseases attributed to O3 exposure each year. Thus, the sustained increase in O3 concentrations in China is an important factor contributing to the increasing threat to human health. Furthermore, the results of spatial regression models indicate that population, the share of secondary industry in GDP, NOx emissions, temperature, average wind speed, and relative humidity are important determinants of O3 concentration variation and significant spatial differences are observed. Discussion The spatial differences of drivers result in the spatial heterogeneity of O3 concentration and exposure risks in China. Therefore, the O3 control policies adapted to various regions should be formulated in the future O3 regulation process in China.


. Introduction
Within the context of the yearly improvement of particulate matter (PM) pollution in Chinese cities, O 3 concentrations are increasing instead of decreasing and are becoming the second most important air pollutant after PM (1). According to the data published by the China General Environmental Monitoring Station, the daily maximum hourly average 90th percentile concentration of O 3 in 338 prefecture-level cities in China increased from 140.0 µg/m 3 in 2014 to 151.0 µg/m 3 in 2018, and the number of days exceeding the standard increased from 6.1% in 2014 to 8.4% in 2018, and the O 3 concentration in some regions has exceeded the secondary concentration limit (160 µg/m 3 ) for air quality in China (2). Longterm exposure to high O 3 concentrations not only affects urban air quality (3), damages human health (4), reduces food production (5), affects atmospheric radiation balance (6), and even influences global climate change (7). Due to its importance to the atmospheric . /fpubh. . environment and climate change, O 3 has received continuous attention from the scientific community and relevant regulatory administrations in the past decades.
To deeply understand the O 3 pollution in China, a large number of researchers have conducted extensive investigations on O 3 pollution levels, spatial and temporal patterns, trends, exposure risks, and drivers in China from different spatial and temporal scales over the past decade (8)(9)(10). For example, Gong et al. The numerous studies mentioned above are important references for a comprehensive assessment of the O 3 pollution in China, but these studies still have the following shortcomings. First, there is significant spatial heterogeneity in surface O 3 pollution, with a few individual cities or regions of O 3 pollution not being a substitute for the level of O 3 pollution in China. Second, there are potential spatial associations between exposure risk and health risk of populations to surface O 3 pollution, and unfortunately, previous studies have tended to ignore their interrelationships. Third, the effects of drivers on O 3 concentrations are spatially variable, and previous studies have tended to focus on the combined effects of drivers on O 3 , neglecting the spatial and temporal differences in the effects of drivers on O 3 concentrations. Therefore, the main objectives of this study include: (1) investigating the spatial and temporal patterns and trends of O 3 concentrations in China using trend analysis and spatial clustering based on a high spatial and temporal resolution O 3 concentration dataset; (2) examining the spatial and temporal associations of population exposure risk and health risk attributable to O 3 pollution using population exposure risk models and exposureresponse functions; and (3) revealing the drivers of differences in O 3 distribution in China from a spatial perspective based on a multi-scale geographically weighted regression (MGWR) model. This study has important practical implications for assessing the future health burden caused by O 3 pollution and its resulting health costs in China; meanwhile, it has important implications for how to equitably allocate healthcare resources and environmental management costs in the future planning and construction of healthy cities and smart cities in China.
. Materials and methods

. . Study area
This study focuses on China, including 31 provinces in the Chinese mainland, excluding Hongkong, Macau, Taiwan,

. . Data source
The daily maximum 8-h O 3 concentration (MDA8) reanalysis dataset of 10 × 10 km from January 1, 2013, to December 1, 2018, is from the tracking air pollution in China (http://tapdata. org/). The dataset is based on a machine learning algorithm and multi-data information fusion inversion. Its comprehensive construction combines ground monitoring data, satellite remote sensing data, high-resolution emission inventory data, air quality model simulation, and other multi-source data, which greatly improves the spatial and temporal accuracy of the data inversion results compared with the previous air quality reanalysis data (15). The daily O 3 concentrations in 360 prefecture-level cities in China during the study period were obtained from the China National Environmental Monitoring Center (http://www.cnemc.cn/sssj/). In order to reduce the error in the calculation of the health risk model, we calculated the 90th percentile concentration of the MDA8 O 3 concentration from the interannual scale based on the daily MDA8 O 3 concentration as the threshold.
The population size (Pop), the proportion of secondary industry to GDP (S_GDP), disposable income per capita (P_GDP), and soot emissions (Dust) for 360 prefecture-level cities in .

Research framework.
China during the study period were obtained from the China Statistical Yearbook (http://www.stats.gov.cn/tjsj/ndsj/#). The nitrogen oxide (NOx) and volatile organic compound (VOC) emissions were obtained from the China Multiscale Emissions Inventory Model (http://meicmodel.org/). The 1 × 1 km spatial resolution population data were obtained from the World pop dataset (https://www.worldpop.org/). The daily meteorological data were obtained from the China Meteorological Data Network (http://data.cma.cn/) during the study period. The meteorological data obtained in this study mainly include air temperature (Tem, • C), sea level pressure (Pa, Pa), relative humidity (Hum, %), 2-m mean wind speed (WS, m/s), 1-h precipitation (Pre, mm), and 10-min mean visibility (Vis, m).

. . Trend analysis
Trend analysis is usually used for the analysis of temporal dynamics of air pollutants to explore the interannual rate of pollutant changes (16). In this paper, the rate of change of O 3 concentrations in China from 2013 to 2018 was analyzed based on the trend analysis method, which is calculated as Equation (1): Where O 3 indicates the O 3 concentration of each cell; n indicates the time span, here the time span is 6; and i is the time unit.

. . Population exposure risk model
Previous studies have shown that significant heterogeneity in the spatial distribution of air quality concentrations and population density leads to major spatial differences in the exposure risks of populations to air quality (17). In addition, health risks due to exposure to pollutants are usually defined as a function of the multiplication of population density and pollutant concentration (18). Although the exposure risk intensity in the area can be quantified to some extent, it cannot distinguish the severity of the local area relative to the whole. To address this issue, we introduced a model for the relative exposure risk of the population attributable to O 3 exposure, as shown in Equation (2), which can evaluate the where R i indicates the risk of population exposure in grid cell i; P i indicates the number of exposed populations in grid cell i; C i indicates the O 3 concentration in grid cell i, and n indicates the total number of grid cells in the study area. To better reflect the spatial difference of relative population exposure risk, we categorized the population exposure risk as extremely low risk, low risk, lower risk, higher risk, high risk, and extremely high risk by using the reclassification method in ArcGIS10.8 software. The corresponding exposure risk values are R i = 0, 0 < R i ≤ 1, 1 < R i ≤ 2, 2 < R i ≤ 3, 3 < R i ≤ 5 and R i > 5, respectively. A higher value of R indicates a higher exposure risk.

. . Health risk model
In this study, a standard damage function was applied to estimate the population of premature deaths from respiratory diseases due to O 3 exposure. The specific equations are shown in Equations (3) and (4), and the relationships shown in the following equations have been extensively applied in previous studies (14,20,21).
where RR is the relative risk; (RR−1)/RR is the attributable fraction; x i is the O 3 concentration in a city i or grid i; x 0 is the threshold concentration; β is the exposure-response coefficient, which represents the additional health risk associated with an increase in unit O 3 concentration (22, 23); M is the number of premature deaths of respiratory diseases attributable to exposure to the O 3 environment; y 0 is the baseline mortality rate of respiratory diseases, and pop is the number of the exposed population. In this study, the mortality rate of respiratory diseases was obtained from the National Bureau of Statistics, where the crude mortality rate of respiratory diseases y 0 (

. . Multi-scale geographically weighted regression
Compared with the classical geographically weighted regression model (GWR), the MGWR model was a flexible regression model (28). Each regression coefficient was obtained based on local regression, and the bandwidth is specific. In addition, the GWR model uses weighted least squares in the fitting operation, while the MGWR model was equivalent to a generalized additive model (GAM), which could perform regression analysis on spatial variables with linear or non-linear relationships, and was also an effective tool for dealing with various complex non-linear relationships of spatial variables (29). Assuming that there are n observations, for observation i ∈ {1,2,3,. . . , n} at location (U i , V i ), the MGWR were calculated as follows (30): indicates the bandwidth used for calibration of the jth conditional relationship, ε i is the error term. In addition, the spatial kernel function type selected during the model operation is bisquare, the bandwidth search type is golden, and the model parameter initialization type takes GWR estimation as the initial estimation model.

. . Research framework
This study used the trend analysis method, spatial autocorrelation model, population exposure risk model, exposure-response function, and MGWR model to analyze the spatial-temporal pattern, exposure risk, health risk, and driving factors of O 3 concentration in China from 2013 to 2018. Firstly, we use the trend analysis method and spatial autocorrelation model to explore the changing trend and spatial-temporal distribution of O 3 concentration in China. Secondly, we selected the population exposure risk model and exposure-response function to investigate the population exposure risk and health risk attributed to O 3 pollution, and discussed their temporal and spatial correlation characteristics. Finally, we use the MGWR model to reveal the dominant factors of spatial distribution difference of O 3 concentration in China. Additionally, in this study we used O 3 concentration reanalysis data at 10 × 10 km resolution and population raster data at 1 × 1 km resolution to investigate the exposure risks and health risks attributed to O 3 pollution. To spatially match the 10 × 10 km O 3 concentration reanalysis data, we used the aggregation module of ArcGIS10.6 software to quantitatively change the spatial resolution of the 1 km×1 km population data. During the aggregation calculation, the output image element cell size was set to 10 × 10 km, i.e., 0.01 • ×0.01 • , and the nearest neighbor assignment method was selected for the aggregation technique. Figure 2 shows the research framework of this paper.
. Results  Figure 4A). The results of the hot spot analysis show that there is a significant hot spot (HH) region for O 3 concentration growth rate, which is mainly contiguous and focused in Shaanxi, Shanxi, central Inner Mongolia, Beijing-Tianjin-Hebei (BTH), southwest Liaoning, central Henan, eastern Hubei, Anhui, Jiangsu, and Shandong in China, which are the regions with the strongest O 3 growth rate in China. In addition, we found a significant cold spot area (LL) covering a large part of China (about 90% of the territory). These regions are mainly located in northeastern, southern, southwestern, eastern, and northwestern China, where the growth rate of O 3 concentration is relatively low and even decreasing regions are observed ( Figures 4B, C). The standard deviation ellipsometric analysis evaluated the overall variations in the spatial pattern of O 3 concentration growth rate from 2013 to 2018 in China ( Figure 4D). It can be found that the regions with significantly increased O 3 concentration growth rates are mainly concentrated in BTH, Shanxi, Shandong, Jiangsu, Jiangxi, Anhui, Hubei, Henan, and Shaanxi in China. This result also indicates that the above-mentioned regions are the primary contributors of O 3 during the whole study period in China. Meanwhile, the center of the median growth rate of O 3 concentration is located north of the standard deviation ellipse arithmetic center, indicating that the growth rate of surface O 3 concentration is greater in northern China than in southern China.

. . The population exposure risk and health risk
Overall, the total population exposed to O 3 > 160 µg/m 3 increased from 1.2% in 2013 to 28.9% in 2018, compared to a decrease in the population exposed to O 3 < 160 µg/m 3 from 7.2% in 2013 to 3.6% in 2018 ( Figure 5). Figures 6, 7 represents the spatial pattern of exposure risk levels attributed to O 3 pollution in 2013, 2015, and 2018. We found that most regions have remained at low (52.89-55.73%) or extremely low (19.48-20.48%) O 3 exposure risk levels over three time periods in China. From a temporal perspective, only 4.83% of the territory of the country was at high exposure to O 3 pollution in 2013, and this percentage increased to 6.45 and 7.19% in 2015 and 2018, respectively. Similarly, the area of the territory exposed to extremely high risk also exhibits a marked increasing trend, from 7.61% in 2013 to 9.62% in 2015 and further to 11.35% in 2018 (Figures 7A-C)   linear modeling regression process. Therefore, before conducting model regression analysis, to test whether there is multicollinearity between each explanatory variable, we use variance inflation factor (VIF) to test the multicollinearity problem between each explanatory variable, and previous studies have shown that when VIF ≥ 10, it indicates that there is a serious multicollinearity problem between the dependent variable and the independent variable. multicollinearity problem, which should be removed from the actual model operation. The collinearity test in this study was performed in SPSS 25.0 software and the results of the analysis showed that the range of VIF values for all explanatory variables was 1.000-9.765, which indicates that there was no cointegration between the dependent and independent variables. Table 1 indicates the diagnostic information of the MGWR model for the socioeconomic and meteorological factors. In terms of the number of valid parameters, the goodness-offit R 2 for the responses of socioeconomic and meteorological factors to O 3 concentrations are 0.861 and 0.799, respectively, and the residual sum of squares (RSS) is 136.297 and 136.51 µg/m 3 , respectively, with the absolute values of the deficit information criterion (AIC) and the log-likelihood value (Loglikelihood) < 5,000. These regression results indicate that MGWR uses fewer parameters to obtain regression results that are closer to the true values and can be fully used to assess the relationship between O 3 pollution and socioeconomic and meteorological factors.   Figure 9 indicates the spatial distribution of regression coefficients of socio-economic factors. The high values (>0.27) of regression coefficients for the total population are mainly located in North and East China, where the total population is significantly and positively correlated with its corresponding O 3 concentration. The influence of the share of secondary industry on surface O 3 in East and North China is significantly higher than that in other regions, and its regression coefficient exceeds 0.08. We also find that over 80% of the regional disposable income per capita is positively correlated with O 3 , with regression coefficients ranging from 0.07 to 0. 36

FIGURE
Probability distribution of the total population exposed to di erent O concentrations from to . The red and gray bars indicate the total population and the proportion of the population exposed to di erent O concentrations, respectively.  Figure 10 shows the spatial differences in the effects of various meteorological factors on O 3 concentration. It can be found that the temperature of cities in North, East, and Northeast China showed a significant (p < 0.05) positive correlation with O 3 concentration, with regression coefficients ranging from 0.23 to 0.49. The relative humidity was negatively correlated with O 3 concentration in all cities during the study period. Among them, cities in Heilongjiang, Jilin, Liaoning, Beijing, Tianjin, north-central Hebei, northwestern Shanxi, western Inner Mongolia, and northwestern Ningxia and northern Shaanxi showed a weak negative correlation between relative humidity and O 3 concentration with a non-significant (p > 0.05) regression coefficient of < −0.07. In contrast, cities in southern Zhejiang, southern Anhui, Jiangxi, central Hubei, Hunan, Chongqing, Guizhou, Yunnan, and cities in Fujian, Guangdong, and Guangxi regions showed a significant (p < 0.01) strong negative correlation between relative humidity (Hum) and its corresponding O 3 concentration with regression coefficients ranging from −0.18 to −0.15. Wind speed (WS) showed a significant (p < 0.05) negative correlation with O 3 concentrations in Heilongjiang, Jilin, Liaoning, Guangxi, southern Henan, Hubei, eastern Shandong, Jiangsu, Shanghai, Zhejiang, Sichuan and Chongqing regions, and northern Shanxi, with regression coefficients ranging from −0.02 to −0.06. It is particularly noteworthy that cities in BTH, southwestern Shanxi, northern Henan, central Shaanxi, Ningxia, southern Gansu, western Shandong, and Anhui have a significant positive correlation between their wind speed and O 3 concentration with regression coefficients >0.45. For air pressure, cities located in northern China showed a significant (p < 0.05) negative correlation between air pressure (Pa) and O 3 .

. . Spatial heterogeneity of O concentration drivers
There are strong spatial variations in the influence of different drivers on O 3 . Relative to lower population density regions, a larger population size implies more energy consumption and pollution emissions, meanwhile, it also further compresses the green area of cities, leading to a significant reduction in the ability of cities to mitigate air pollution, which better explains why the positive correlation between population size and O 3 concentration is significantly higher in densely populated northern and eastern China than in other regions (21,33). Previous studies have shown that industrial emissions are the predominant source of air pollution (34). Our study found that the share of secondary .
/fpubh. .   (35). In general, the higher the NOx emissions in cities, the lower the VOCs-NOx ratio. For example, the formation of O 3 in some cities located in Central and Northern China is often limited by VOCs (36,37). In these cities, the reduction of VOCs emissions decreases the formation of O3, but the reduction of NOx emissions increases the formation of O 3 . This chemical reaction tends to depend on the amount of VOCs and NOx emissions; the larger the emissions the more intense their reaction and the larger the O 3 emissions generated (38). In addition, industrial dust emissions indirectly affect solar radiation intensity by affecting atmospheric visibility, which further contributes to the O 3 photochemical reaction rate (39). Temperature is an important ambient condition for photochemical reactions, and higher temperatures can promote the rapid production of O 3 concentration, therefore, temperature and O 3 concentration are mostly positively correlated, especially in cities in Northern, Eastern, and Northeastern China where the solar temperature is higher in the warm season (40). The wind speed has a diffusion and transport effect on pollutants in the atmosphere.  warm-season burning winds, especially from June to August each year, when the burning winds blow from the mountains to the northern and western parts of the North China Plain, bringing dry the hot air further leads to a higher temperature in the region, which accelerates the photochemical reaction of O 3 production to some extent. Relative humidity has a negative correlation with O 3 concentration. Previous studies have shown that water vapor can not only absorb and release energy through changes in the aqueous phase but also undergo internal reactions, especially when controlling for other influencing factors, higher relative humidity leads to higher water vapor saturation, resulting in easy removal of O 3 and its precursors and lower O 3 concentrations (42). In addition, water vapor can reduce solar ultraviolet radiation through extinction mechanisms, thus affecting photochemical reactions and O 3 concentrations (43).

. . The O control policy implications
In summary, O 3 pollution in China is gradually increasing, and more and more of China's population is exposed to high O 3 concentration pollution. Scientific and effective reduction of O 3 concentration exposure levels in China is crucial to reduce population exposure risks (44). Under these circumstances, this study proposes policy recommendations on how to reduce O 3 concentrations in Chinese cities from the perspective of the drivers affecting the spatial distribution of O 3 and epidemiology. For O 3 pollution areas dominated by O 3 precursors (e.g., NOx, VOCs, and CO), the authorities can ensure that their emissions comply with government regulations by optimizing the industrial structure and reducing the emissions of O 3 precursors. Meanwhile, the governmental department should focus on the synergistic management of PM 2.5 and O 3 compound pollution. Research shows that NOx is not only an important precursor for O 3 generation but also an important precursor for PM 2.5 (45). Therefore, strengthening the NOx deep regulation and emission reduction is a key step to promote synergistic control. Furthermore, the O 3 abatement measures in the future should pay attention to different seasonal O 3 control measures and strengthen regional cooperation for O 3 pollution prevention.
For O 3 pollution regions dominated by meteorological factors, the department should forecast the variation of O 3 concentration due to the change of meteorological factors promptly, meanwhile develop a detailed O 3 pollution early warning program to reduce the risk of public exposure and explore a sustainable development path for O 3 pollution management in China. From .
/fpubh. . an epidemiological perspective, to protect public health and improve the status of O 3 pollution, it is crucial to establish studies of health effects attributed to O 3 exposure from a national perspective. In addition, it is important for relevant government departments to establish a mechanism to revise the National Ambient Air Quality Standards (NAAQS) for regulatory assessment and health risk prediction of future O 3 air quality standards in China (46).

. . Research limitations and future prospects
Surface O 3 distribution has strong spatial and temporal heterogeneity, and there are significant differences in O 3 concentrations with time scales. This study only focused on the interannual spatial variability characteristics of O 3 concentrations, neglecting the seasonal variability of O 3 concentration changes. Furthermore, due to the lack of basic research data and inadequate research methods, this study only focused on the number of premature respiratory deaths attributed to O 3 pollution in the assessment of health risks attributed to O 3 pollution, neglecting the all-cause premature death group. Additionally, using the same exposure risk coefficient (β) may lead to spatial errors in the estimated health risks due to significant spatial differences in O 3 exposure levels.  47), and their results found an average of 186,000 deaths from respiratory diseases due to O 3 pollution during the study period. This is slightly lower compared to our findings. A primary reason for this is that our study and Wang et al. (21) used different exposure response coefficients and critical thresholds. In addition, the interpolation of O 3 concentrations at large scales of pollution can also cause large errors in the assessment results. Therefore, in the future, we hope to conduct a detailed and comprehensive analysis of seasonal differences in O 3 pollution and all-cause health risks in China by utilizing more detailed surface O 3 monitoring data and meta-analysis methods. To provide a scientific basis for the improvement of O 3 pollution in China.

. Conclusions
In this study, we quantitatively investigated the spatial and temporal patterns, trends, population exposure risks, health risks, .
/fpubh. . and drivers of surface ozone in China from 2013 to 2018. We observed the annual average O 3 concentration of China increased significantly at a rate of change of 1.84 µg/m 3 /yr from 2013 to 2018 (p < 0.05, R 2 = 0.561). The significant increase was mainly distributed in East China, Central China, and North China. Meanwhile, the growth rate of O 3 concentration has a consistent and enhanced positive spatial autocorrelation (p < 0.05), and there are significant hot and cold spots areas. During the research period, there was an average of over 24,000 premature deaths from respiratory diseases attributed to O 3 exposure in China from 2013 to 2018, and the growth rate fluctuated at 1,178 per year (p < 0.05). Spatially, there was a consistency in the spatial distribution of exposure risk and health risk of populations exposed to O 3 . The results of the multi-scale geographically weighted regression model reveal spatial differences in the effect of various factors on O 3 concentration. The impact of the total population, disposable income, the share of secondary industry in GDP, and NOx emissions factors in eastern and northern regions are significantly greater than impacts in central and western regions. Meanwhile, we found that the effect of temperature on O 3 concentration in some cities in the north, east, and northeast is significantly higher than that in other regions, and relative humidity has a significant (p < 0.01) strong negative correlation with O 3 concentration in east, central, southwest and south China.

Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/supplementary material.