Synoptic and meteorological drivers of regional ozone pollution events in China

Surface ozone (O3) pollution events are becoming more frequent and have recently emerged as a severe air pollution problem in China. However, the spatial–temporal distribution of surface O3, as well as its primary synoptic and meteorological drivers, remains poorly understood. The purpose of this study was to identify the key synoptic and meteorological drivers of O3 pollution in different regions of China. To achieve this goal, this study established meteorology overlaps of regional O3 pollution events in space and time and applied a comprehensive statistical model selection method for optimal synoptic and meteorological models, based on a newly released O3 dataset for 2015–2018. It was observed that extreme regional O3 pollution events (duration >7 d) occurred more frequently and exhibited a high co-occurrence frequency (>50%) with air stagnation (AS). Moreover, the beginning and end of 69% of the regional O3 pollution events coincided with regional daily maximum temperature changes. The intensity of AS is the dominant driver of O3 pollution event intensity across most of the six selected megacity regions. Although other meteorological drivers, such as the intensity of hot days (HD) and meridional wind of 10 m were also important, their impacts varied according to the region. Overall, increase in extreme AS and HD led to the worsening of regional O3 pollution events. These findings imply that mitigating regional O3 pollution should consider changing synoptic and meteorological conditions.


Introduction
Surface ozone (O 3 ) adversely affects human health, especially when its concentration is >100 μg m −3 (Word Health Organization 2018). In recent decades, China has become an epicenter for O 3 pollution (Zhang et al 2016). Since 2013, annual O 3 pollution days when O 3 concentrations above a certain threshold have been increasing in many Chinese cities (Wang et al 2017, Gong andLiao 2019). However, regional Chinese O 3 pollution events and their synoptic and meteorological drivers have remained poorly understood due to limited O 3 observational data availability, leading to a lack of reliable quantification (Zhao et al 2009). Understanding dominant O 3 event drivers is necessary to establish effective O 3 pollution event mitigation strategies.
Local meteorological variables, such as temperature (Bloomer et al 2009) and solar radiation flux (Otero et al 2016, can lead to severe O 3 pollution by enhancing the associated photochemical reaction rate. Temperature has been found to be one of the most likely variables to increase the summer maximum daily 8 h average Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. (Schnell and Prather 2017); however, few studies have quantified the contribution of meteorological extremes to Chinese regional O 3 pollution (Pu et al 2017, Gong and Liao 2019, Lin et al 2019. There is limited knowledge of the relationships between the changes in multiday O 3 pollution and meteorological variables over a region (Zhang et al 2017, Gong andLiao 2019). Considering changes in synoptic and meteorological conditions is important when trying to mitigate regional O 3 pollution events. Recent data should be used to explore the synoptic and meteorological drivers affecting Chinese regional O 3 pollution events.
This study investigated the response of regional O 3 events to different synoptic and meteorological variables across six Chinese regions (figure 1) by considering the spatial extent of these pollution events. The six regions included the Beijing-Tianjin-Hebei (BTH) super-city complex, the Yangtze River Delta (YRD), the Pearl River Delta (PRD), the Sichuan Basin (SCB), the Guanzhong Plain (GZP; also known as the Guanzhong Basin), and the West Side of the Strait (WSS).
The remainder of this paper is organized as follows. Data sources and methods have been described in section 2. The applicable synoptic and meteorological conditions of regional O 3 pollution events are summarized in section 3, along with the key drivers that affect event severity. Finally, study implications are discussed in section 4.

Study region
In China, severe O 3 pollution events have primarily occurred in economically developed regions, especially in large cities with significant populations . Therefore, this study focused on the six major Chinese urban agglomerations (table 1, figure 1). These included (1) the BTH, with 13 cities and 110 million people; (2) YRD, with 18 cities and 127 million people; (3) PRD, with 14 cities and 80 million people; (4) SCB, with 16 cities and 101 million people; (5) GZP, with 11 cities and 45 million people; and (6) WSS, with 20 cities and 93 million people.
For each region, its population are obtained from the National Bureau of Statistics (https://data.stats.gov. cn/index.htm), representing the population in 2019. Boundaries of each city in the six regions were obtained from the Ministry of Civil Affairs of the People's Republic of China (MCA; http://202.108.98.30/map).

Data
Observation data on MDA8 O 3 concentration since May 2014 were obtained from the China National Environmental Monitoring Center (CNEMC) network; data from March 2015 to February 2019 were used for this study. Moreover, data on the commonly used local meteorological variables that affected O 3 concentrations, including the daily maximum air temperature at 2 m (T max ), daily mean air temperature at 2 m (T mean ), daily sunshine duration (SD), daily surface air pressure (AP), daily precipitation (Pre), daily relative humidity (RH), and daily 10 m surface wind speed (WS) were acquired from the China National Meteorological Information Center (CNMIC, https://data.cma.cn/).

Defining regional ozone pollution events
The definition of a regional O 3 pollution event is summarized in table 2. For this study, 160 μg m −3 represents the O 3 pollution threshold, based on the Chinese O 3 Grade II air quality standard for an ambient air functional area (Ministry of Environmental Protection of the People's Republic of China 2012). Regional O 3 pollution event was defined as MDA8 O 3 concentrations>160 μg m −3 in over 25% of the cities within a region and MDA8 O 3 concentrations>160 μg m −3 in over 50% of the cities during a central extreme day (People's Government of Guangdong Province 2013). The central extreme day was defined as the day with the most O 3 -polluted cities during an O 3 pollution event (People's Government of Guangdong Province 2014). MDA8 O 3 event concentration was defined as the mean MDA8 O 3 concentrations for cities that had O 3 160 μg m −3 during the event. The term 'regional O 3 pollution event intensity' was used to quantify the severity of regional O 3 pollution events; it was the product of MDA8 O 3 concentration and the number of O 3 polluted cities during the event (table 2).

Statistical analysis
A subjective classification method (Huth et al 2008) was used to analyze SLP synoptics. Thus, synoptic pressure patterns that were applicable during regional events were identified using daily SLP isobaric charts. To simplify the classification and to reduce bias caused by subjective interpretations, only two common synoptic patterns (high-pressure and low-pressure systems) were used. The summer MDA8 and SLP were detrended by subtracting the 15 d moving averages (from pre-to post-7 days of the date), which helped establish the correlation between the detrended grid SLP and summer MDA8 for each region.
Meteorological variable changes were calculated for all the cities in each region. For each event, it was calculated from the day before to the first day and from the last day to the next day. The variables included T max , SD, SLP, Pre, daily specific humidity (SH), U10M, and V10M. Air stagnation (AS) was calculated using PBLH and theoretical boundary layer height (TBLH) , such that if PBLH was<TBLH and Pre<0.1 mm, AS would occur (AS=1); otherwise, AS would not occur (AS=0). TBLH was calculated from WS.
where, a, b, and c were parameters obtained from Wang et al (2018a), which varied seasonally (table S1 (available online at stacks.iop.org/ERC/3/055004/mmedia)). SH was calculated from the T mean , AP, and RH (Song et al 2012).
where, ES indicates the saturated vapor pressure (hPa) and is derived using equation (4): 273.16 17.269 mean mean ( ) Regional meteorology trends covering >50% of the cities in any one region were also investigated to explore the dominant meteorology drivers at regional O 3 pollution event onset and completion.
Multiple linear regression (MLR) was combined with the Bayesian information criterion (BIC) method to model the REI for each region, as a function of synoptic and meteorological variables. BIC considers both the goodness of fit and the number of variables in the model. BIC is a technique used to select the combinations of candidate synoptic and meteorological variables that have the best explanatory power (Anderson et al 1998), thus helping identify the predominant meteorological mechanisms associated with REI changes. The BIC model is formulated as follows: where o indicates the number of O 3 pollution events in each region, n denotes the number of selected variables in the MLR, y represents the matrix of the REI of each region, yˆstands for the estimated REI based on MLR involving a given subset of variables. The variables for the MLR were normalized by subtracting the mean and then dividing by the standard deviation.
All candidate variables used in model selection have been summarized in table 3, including AS, SD, SLP, Pre, SH, U10M, V10M, and hot days (HD, where T max 35°C) (Tan et al 2010, Meteorological Administration 2017. AS intensity is defined as the sum of the area of O 3 polluted cities with AS during a regional O 3 pollution event, whereas HD intensity is defined as the sum of the area of O 3 polluted cities with HD during a regional O 3 pollution event. In addition to HD and AS intensities, other meteorological variables were averaged, temporally (i.e., for every day of an ozone event) and spatially (i.e., over all the cities in a given region), during a given O 3 pollution event. Synoptic and meteorological variables of all regional event were normalized before calculating the MLR coefficients, which enabled a fair comparison of the contribution of dominant variables to regional REIs. We have tested all the meteorology combinations with more variables by BIC with MLR. The BIC values and R 2 of the models with two or one variables are similar to the models with more variables. Thus, to minimize the risk of overfitting while maximizing selected model robustness, only models with up to two variables should be considered.

Results
3.1. Identifying regional O 3 pollution events Among the 222 regional O 3 pollution events recorded for the six study regions during 2015-2018, the BTH experienced the maximum number of events (65), followed by the YRD (58). Approximately 95% of these events occurred during the warmer period (April-September) with sufficient sunlight and high temperature (figure 2). Although only four years of data were available, which were insufficient to conclude a significant trend, the number of O 3 pollution events tended to increase since 2015, with the most pronounced increase being observed daily 10 m meridional wind (m s −1 ) MERRA-2 SH daily specific humidity (g kg −1 ) CNMIC in the SCB and GZP. This finding was consistent with that of Gong and Liao (2019), who reported an increasing O 3 pollution trend in North China from 2014-2017. Regionally, O 3 pollution events in the six regions generally intensified between 2015 and 2018; longer regional events (>7 d) also became more frequent. In the BTH, 20% of the regional O 3 pollution events lasted between 8 and 18 d and covered 92%-100% of the cities in the region. The five worst BTH events occurred during June-July of 2017-2018 (Table S2). These five events affected all the regions during the central extreme day (table 1) In the YRD, 10% of the events lasted 8-13 d and covered 94%-100% of the cities. The five worst O 3 pollution events in the YRD occurred during May-September of 2015-2017. These events covered 83%-94% of the cities during the central extreme day, with MDA8 O 3 concentrations>189 μg m −3 .
In the GZP, three O 3 pollution events in 2017 lasted for more than a week. In the GZP and PRD, the worst O 3 pollution events occurred in 2017, and lasted 11-26 d; they covered 86%-91% of the cities, with MDA8 O 3 concentrations>187 μg m −3 (table S3).
These results confirmed that long-duration regional O 3 pollution events extended across almost all BTH, GZP, YRD, and PRD cities, and lasted for more than a week (MDA8 concentrations>180 μg m −3 ). However, none of the SCB and WSS events lasted longer than a week. Apart from the YRD, O 3 pollution events in the other five regions peaked in 2017 and 2018.
Fluctuations in the O 3 pollution event in the YRD were associated with extreme precipitation in the summer of 2016 (Herring et al 2018) and with typhoons in 2018 (Meteorological Administration 2018). In the other five regions, the strong western Pacific subtropical high (WPSH) and extreme regional heat waves associated with dry weather (Meteorological Administration 2017, 2018) may have contributed to 2017-2018 experiencing the worst regional O 3 pollution. Overall, irrespective of regional fluctuations, the worst regional events occurred in more recent years (i.e., 2017 and 2018), suggesting an overall acceleration in extreme O 3 pollution patterns over time (table S3). The six regional REIs differed in both magnitude and temporal distribution. As shown in figure 2, regional REIs in north China were higher than those in the south. Among the six regions, the WSS REI was only 7% of that in the BTH region across the study period.
Monthly BTH and GZP REIs peaked in June (figures 2(a), (e)), when increasing temperature, stronger solar radiation, and low humidity facilitated higher photochemistry activity levels (Hou et al 2014, Fang et al 2020. In contrast, monthly PRD REIs mostly peaked during August-October, and their worst annual O 3 pollution events occurred in September and October ( figure 2(c)). Moreover, monthly O 3 pollution event REIs in south China often peaked twice, that is in spring and fall, while experiencing a drop in summer. Although high temperatures in summer can intensify O 3 pollution, high precipitation and clear air brought by the East Asian summer monsoon can effectively inhibit extreme regional O 3 pollution in these regions (Li et al 2018a. Consequently, REIs in south China often decreased during the monsoon, while exhibiting peaks during premonsoon and/or post-monsoon.

Synoptic and meteorological conditions for O 3 pollution events
The changes in T max affected the onset and completion of most regional O 3 pollution events in the six regions (figure 3). Most O 3 pollution events occurred with increasing T max meaning that T max was higher during the first day of an event than the day before the event, whereas most pollution events ended with decreasing T max meaning that T max decreased from the last day to the next day of these events. Figure 3(a) demonstrates that 86% of events occurred in regions that were warm, with the highest frequency (96%) occurring in the GZP. Largescale increases in T max can lead to the release of more anthropogenic volatile organic compounds (VOCs) and can also change air circulation patterns (Ordonez et al 2005, Pu et al 2017. This results in an increase in photochemical processes, which causes regional O 3 pollution. Additionally, 69% of the events in the YRD and GZP started with regional warming and increasing SD trends ( figure 3(b)).
Owing to its direct and positive impact on the photochemical processes responsible for O 3 production, SD is also an important variable that influences the occurrence of regional events. Furthermore, 83% of the events in the study area ended with a regional cooling trend ( figure 3(c)). Using the SCB as an example, 87% of the events concluded with regional cooling and decreasing SD trends ( figure 3(d)). Overall, the start and finish of 69% of the O 3 pollution events in the study area coincided with regional increases and decreases, respectively, in T max .
Regional O 3 pollution events were strongly influenced by high-pressure systems from the ocean and by cyclones. As oceanic high-pressure systems affected 60% of the BTH, GZP, and YRD events, they were the dominant synoptic systems. Additionally, positive correlations between detrended SLP and summer MDA8 occurred in the Pacific (figure 4), which suggested that the pollution events were dependent on WPSHs.
Continuous WPSHs affected atmospheric circulation, water vapor levels, and O 3 transport (Ding et al 2013), leading to O 3 accumulation and regional O 3 pollution events , Dong et al 2020. Low-pressure systems in Northeast China also influenced O 3 pollution in the BTH region ( figure 4(a)) and affect 45% of the events. The downward airflows at the peripheries of Northeast cold vortexes favored regional O 3 collection and stagnation, giving rise to pollution events.
On the other hand, tropical cyclones were the dominant synoptic systems for pollution events in South China, with 60% of pollution events in PRD and WSS being association with typhoons (cyclones). In South China, negative correlations in the western Pacific showed that tropical cyclones were closely associated with O 3 pollution (figures 4(b)-(c) and (f)). This phenomenon occurred due to the combined effects of high Additionally, 65% of the SCB events occurred when there were low-pressure systems in the Qinghai-Tibet Plateau. Positive correlations in southwest China (figure 4(d)) indicated that Tibetan Plateau heat lows were able to influence regional O 3 pollution. The hot dry days caused by weak east heat lows also reportedly exacerbate regional O 3 pollution (Wang and Li 2015).
It was concluded that WPSHs and tropical cyclones with weak hot air circulation represented synoptic patterns that were typically present for most of the O 3 pollution events analyzed in the study area.
To further explore the synoptic and meteorological characteristics of regional O 3 pollution events, cooccurrence frequencies for O 3 pollution days for all the events with candidate variables were examined. Figure 5 shows that AS was a common meteorological condition, often co-occurring (>50%) in most of the six regions (92% of the cities). Its highest frequency (100%) was found in 50% of the WSS cities ( figure 5(a)). Given that AS inhibits horizontal and vertical mixing in the lower troposphere, it leads to the quick generation and collection of O 3 pollutants at the regional level Wang 2001, Garrido-Perez et al 2018). As for HD, lower cooccurrence frequencies (<50%) accounted for 98% of the six regions ( figure 5(b)), although 87% of longduration events had both AS and HD.
Concurrent with stagnant air and dry hot days, meteorology extremes were crucial synoptic features in the regional O 3 pollution events. The co-occurrence frequency of strong SD with low precipitation (i.e., Pre<0.1 mm) (General Administration of Quality Supervision 2012) and O 3 pollution days was>50% in most areas of the GZP ( figure 5(c)). Given the positive effect of dry air with strong SD on O 3 generation (Rizk 1992, this combination was associated with the occurrence of pollution events.
The co-occurrence frequency for O 3 pollution days and a south wind (SW) V10M was>70% in the BTH ( figure 5(d)). This dominance showed that SW was a key meteorological feature of these events, transporting significant amounts of O 3 from neighboring southern regions into the BTH (Ma et al 2011.
Overall, the results confirmed that AS was the dominant synoptic condition for most events, while SW pollution transport had a non-negligible impact on regional O 3 pollution in North China.

Key synoptic and meteorological drivers of O 3 pollution events
Key synoptic and meteorological variables were identified for the REI of the five regions (table 4). The six pollution events of the WSS region were not analyzed, as the sample size was extremely small for MLR (Harrell 2015). All regression coefficients were statistically significant for the five regions (p<0.01).
MLR models developed for all five regions explained 75%-95% of the variability in regional O 3 pollution event intensity (table 4). The optimal MLR model in the BTH showed that the contributions from AS and HD intensity were positive for REI. This result was similar to that of Ma et al (2019), who confirmed that heat waves with AS enhanced the contribution of temperature to regional extreme O 3 events in the north China Plain. Recent studies have also suggested that HD with AS can effectively increase O 3 generation and accumulation rates by controlling VOC emissions and peroxyacetyl nitrate chemistry (Jacob et al 1993, Garrido-Perez et al 2018. In 2017-2018, high AS and HD intensities were observed during the top seven worst regional O 3 pollution events. However, in the PRD, increasing AS intensity co-occurring with decreasing V10M aggravated regional O 3 pollution event intensity. Owing to the clean air mass carried by the south wind and O 3 transport caused by the north wind (caused by typhoons), positive V10M (i.e., the south wind) was established to reduce O 3 , whereas negative V10M (i.e., the north wind) could make O 3 conditions worse . This result supported the hypothesis that high AS intensity in the presence of weak north winds caused by tropical cyclones could lead to extreme regional O 3 pollution events.
In the YRD, SCB, and GZP regions, AS intensity was the only dominant regional O 3 pollution event intensity driver. For the YRD, AS strengthened by either WPSH or cyclones raised temperatures and triggered strong photochemical reactions, which lead to extreme regional O 3 pollution events (Gao et al 2020). For the SCB and the GZP, when large-scale AS phenomena controlled the mountain-basin areas on warm and sunny days, intense O 3 -related physical and chemical reactions quickly generated extreme regional O 3 pollution (Li et al 2018b, Ning et al 2020. For each region, the regression coefficient for AS intensity was higher than that for other variables, meaning that AS intensity was the dominant driver influencing temporal REI variability.
Overall, the results indicated that meteorological extremes made vital contributions to regional O 3 pollution event intensity variation across the study area. Large-scale, dry weak air convection influenced most of the regional O 3 pollution events. Local meteorological variables, such as wind direction and strong SD, also affected regional O 3 pollution events.

Conclusion and discussion
This study examined O 3 pollution events in the six economically developed regions of China occurred during 2015-2018 and identified their key drivers. O 3 pollution events were the most common during this period. Moreover, AS, under the influence of WPSH or tropical cyclones was associated with regional O 3 pollution events in the study area, with AS intensity being the dominant REI driver in most regions. Additionally, the rise and fall of T max induced the beginning and end of an O 3 event in most regions. In China, increasing meteorological extremes, such as the AS  and heat waves (WMO 2019, Wang et al 2019) created suitable meteorological conditions for generating regional O 3 pollution. The short time span of available O 3 monitoring data represented a limitation on the results presented in this study. A longer O 3 observation time series could provide more definitive results, allowing the analysis of synoptic and meteorological drivers, as well as their impacts on the interannual variability in surface O 3 in China, to be presented in a more robust manner.
Global warming and urban expansion could lead more frequent meteorology extremes. In that case, there will be more frequent O 3 pollution in the future . Thus, mitigating O 3 pollution in China and other countries will involve controlling O 3 precursor emissions. Comprehensive air quality management plans should be implemented by considering weather conditions and emissions to control intense and large-scale O 3 pollution. For example, mandating greater use of public transport during hot days that coincide with air stagnation would be beneficial.