Temporal Characteristics of Ozone (O3) in the Representative City of the Yangtze River Delta: Explanatory Factors and Sensitivity Analysis

Ozone (O3) has attracted considerable attention due to its harmful effects on the ecosystem and human health. The Yangtze River Delta (YRD), China in particular has experienced severe O3 pollution in recent years. Here, we conducted a long-term observation of O3 in YRD to reveal its characteristics. The O3 concentration in autumn was the highest at 72.76 ppb due to photochemical contribution and local convection patterns, with its lowest value of 2.40 ppb in winter. O3 exhibited strong diurnal variations, showing the highest values in the early afternoon (15:00–16:00) and the minimum in 07:00–08:00, specifically, peroxyacetyl nitrate (PAN) showed similar variations to O3 but PAN peak usually occurred 1 h earlier than that of O3 due to PAN photolysis. A generalized additive model indicated that the key factors to O3 formation were NO2, PAN, and temperature. It was found that a certain temperature rise promoted O3 formation, whereas temperatures above 27 °C inhibited O3 formation. An observation-based model showed O3 formation was VOCs-limited in spring and winter, was NOx-limited in summer, and even controlled by both VOCs and NOx in autumn. Thus, prevention and control strategies for O3 in the YRD are strongly recommended to be variable for each season based on various formation mechanisms.


Introduction
O 3 is a typical secondary pollutant with a complex formation mechanism involving a series of chemical reactions among volatile organic compounds (VOCs), oxides of nitrogen (NO x , NO + NO 2 ) and carbon monoxide (CO) [1]. O 3 , as an important indicator of photochemical pollution, plays a central role in the oxidation of chemical and climate-relevant trace gases in the troposphere [2]. O 3 pollution has become a serious air quality problem affecting human health, vegetation, biodiversity, and climate worldwide as O 3 concentrations have increased significantly since the second half of the 20th century [3,4]. According to a government report in China in 2020, O 3 is the only air pollutant that maintained a rising trend during the last 5 years and O 3 pollution is another urgent environmental problem in China, except for haze [5].
It has been noted that O 3 levels increased by 30% to 70% in the temperate and polar regions of the Northern Hemisphere from 1896-1975 [3]. Despite policies to reduce precursor emissions, O 3 concentrations have remained high; therefore, in-depth studies of the factors influencing O 3 formation are critical to controlling ozone pollution. In addition to precursor substances, meteorological factors also have an influential effect on ambient O 3 concentration [6,7]. Peroxyacetyl nitrate (CH 3 C(O)O 2 NO 2 , PAN) in the atmosphere serves also as a reliable and scientific indicator of photochemical pollution [8]. PAN acts as Int. J. Environ. Res. Public Health 2023, 20, 168 2 of 14 a temporary reservoir for NO x and radicals, which can be transported to distant areas to redistribute NO x as well as influence O 3 production on a regional or even global scale [9]. Recently, some related studies focused on severe photochemical smog events in China with a relatively short period of measurement [10,11], but most of them focused on the events occurring in Beijing and the Pearl River Delta (PRD), China [10,12,13].
The Yangtze River Delta (YRD) is the region with the highest degree of urbanization and industrialization in China, and the coal-based energy system not only supports urbanization and industrialization but also contributes to serious regional air pollution problems [14]. Since 2017, O 3 has become the most significant air pollutant in the YRD [5]. As the integrated development of the YRD has become a national strategy [15], the new situation of air pollution prevention and control makes it necessary to conduct an in-depth study on the O 3 characteristics in the YRD to promote the sustainable development of the YRD.
So far, researchers have studied the spatial and temporal variations of O 3 , the mechanism of its formation, and the influencing factors [16][17][18]. Previous research has shown that O 3 in the upper troposphere has increased annually across Europe from 1995 to 2013 [19]. In the troposphere with little UV radiation, it has been widely established that NO 2 photolysis at wavelengths ≤424 nm becomes the main source of atomic oxygen and contributes to O 3 formation. The main feature of O 3 formation is the nonlinear dependence of O 3 production on its precursors, i.e., NO x and VOCs [2]. Several studies have proven that there is a complex photochemical interaction between O 3 and PM 2.5 and that PAN photochemistry has both negative and positive effects on O 3 production [2,9]. In addition, some studies have found that the correlation between O 3 and meteorological factors varies by season and region [16].
Generally, the available literature provides an essential foundation for ozone research. However, many studies focused on a single air pollutant, and few considered the synergistic and coordinated effects of multiple pollutants in a comprehensive manner [15]. Thus, in this study, we performed a one-year continuous observation of O 3 , PAN, other pollutants, and meteorological parameters to provide further insights into the formation mechanism of ambient O 3 in the YRD, China in 2021 in a typical city located in the YRD, China, Shaoxing city, which is the core city of the Great Hangzhou Bay Area, near Shanghai and Hangzhou [20]. Here, a generalized additive model (GAM) was used to synthetically quantify the complex nonlinear relationships between O 3 and multiple parameters, specifically including PAN for the first time. Compared with machine learning techniques, the GAM can uniquely quantify trends in O 3 concentrations, which is better for understanding and controlling pollution [16]. Additionally, an observation-based model (OBM) was used to investigate the sensitivity of O 3 production in different seasons, evaluating the effects of precursor reduction on O 3 production. Therefore, this study can provide a comprehensive understanding of O 3 formation and scientific evidence for the prevention and control of O 3 pollution in the YRD, China.

Observation Site
A field observing campaign was continuously conducted from January-December 2021 at an Atmospheric Observation Supersite (120.62 • E, 30.08 • N) in Shaoxing shown in Figure 1, located on the rooftop of an approximately 15 m-high building. The observation site is surrounded by residential areas and administrative offices, with well-developed traffic and no obvious industrial pollution sources, which can be considered as a typical urban site in the YRD, China. March, April, and May are considered as spring season, June, July, and August as summer, September, October, and November as autumn, and December, January, and February as winter in this paper according to the climate in YRD, China.

Measurement Apparatus and Methods
Concentrations of atmospheric O3, NOx, SO2, and CO were measured by instruments (USA i-series 49i, 42i, 43i, and 48i, Thermo Fisher Scientific, Waltham, MA, America), while PM2.5 is sampled on a tapered element oscillating microbalance (TEOM1405, Thermo Fisher Scientific, Waltham, MA, America). We calibrated each of these instruments periodically on a monthly basis. Meteorological data (temperature (T) and relative humidity (RH)) were obtained from an on-site meteorological station.
PAN concentration was determined through a PAN analyzer (PAN, Met Con Inc., SN, German) containing gas chromatography with an electron capture detector (GC-ECD), a sampling and calibration unit, and a computer control unit. It is the reaction of NO and acetone under UV light to produce PAN standard gas. During the observation period, calibration was performed once a week. The PAN was detected every 5 min with a detection limit of 50 ppt. The overall uncertainty of the measurement was estimated to be ± 3%.
Ambient VOCs were measured online by a cryogen-free automated gas chromatography (GC) system equipped with a flame ionization detector (FID) and mass spectrometer detector (MSD) with a temporal resolution of 1 h (Lu et al., 2022). A total of 94 VOC components were identified and measured during the course of this study. Detailed descriptions of the principles, performance, quality assurance, and quality control (QA/QC) processes of the online GC-MS/FID system are available in a previously published paper [20].

Generalized Additive Model
GAM, an extension of the additive model, is a flexible and free regression model that can make more reasonable nonlinear fittings than traditional statistical models [21]. It is widely used to reveal the complex nonlinear relationships between air pollutants and contributing factors in some air pollution studies [22,23]. In this study, GAM was applied to

Measurement Apparatus and Methods
Concentrations of atmospheric O 3 , NO x , SO 2 , and CO were measured by instruments (USA i-series 49i, 42i, 43i, and 48i, Thermo Fisher Scientific, Waltham, MA, USA), while PM 2.5 is sampled on a tapered element oscillating microbalance (TEOM1405, Thermo Fisher Scientific, Waltham, MA, USA). We calibrated each of these instruments periodically on a monthly basis. Meteorological data (temperature (T) and relative humidity (RH)) were obtained from an on-site meteorological station.
PAN concentration was determined through a PAN analyzer (PAN, Met Con Inc., SN, German) containing gas chromatography with an electron capture detector (GC-ECD), a sampling and calibration unit, and a computer control unit. It is the reaction of NO and acetone under UV light to produce PAN standard gas. During the observation period, calibration was performed once a week. The PAN was detected every 5 min with a detection limit of 50 ppt. The overall uncertainty of the measurement was estimated to be ± 3%.
Ambient VOCs were measured online by a cryogen-free automated gas chromatography (GC) system equipped with a flame ionization detector (FID) and mass spectrometer detector (MSD) with a temporal resolution of 1 h (Lu et al., 2022). A total of 94 VOC components were identified and measured during the course of this study. Detailed descriptions of the principles, performance, quality assurance, and quality control (QA/QC) processes of the online GC-MS/FID system are available in a previously published paper [20].

Generalized Additive Model
GAM, an extension of the additive model, is a flexible and free regression model that can make more reasonable nonlinear fittings than traditional statistical models [21]. It is widely used to reveal the complex nonlinear relationships between air pollutants and contributing factors in some air pollution studies [22,23]. In this study, GAM was applied to analyze the relationship between O 3 and some factors including PAN, VOCs, PM 2.5 , NO, NO 2 , CO, T, and RH, respectively. Its basic form is as follows [24]: where µ is the response variable; g(µ) is the "link" function; α is the intercept; x 1 , x 2 , and x n are the impact factors; f 1 (x 1 ), f 1 (x 1 ), and f n (x n ) are the smooth functions of the impact factors; and β is the residual.

Observation-Based Model
An observation-based model (OBM) was used to simulate the net O 3 production rate and the sensitivity mechanism of O 3 production in this study [25]. The model is informed by observations of 94 VOCs, trace gases (O 3 , NO x , and CO), and the meteorological parameter as boundary conditions for simulating atmospheric photochemical processes. The relative incremental response (RIR) was calculated using Equation (2) to evaluate the relative contribution of the precursors to O 3 formation [18]: where, X represents a specific precursor of O 3 , including VOCs, NO x , and CO, respectively. P O 3 is the O 3 formation potential from 07:00 a.m. to 19:00 p.m., which is the net amount of O 3 production rate during the evaluation period and can be obtained from the OBM; ∆X represents the change in X concentration; S(X) means the observed concentrations of species X, which represents the combined impacts of regional traffic and on-site emissions;

∆S(X)
S(X) means the relative change of S(X) [26].

Seasonal Variation
The temporal variations and statistical description of each observed element during the observation period are displayed in Figure 2 and Table 1, respectively. The measured daily mean concentrations of O 3 ranged from 2.40 to 72.76 ppb, with an annual average of 30.27 ppb, which was higher than those reported in other cities such as Xiamen (28.11 ppb) [27], Shenzhen (27.3 ppb) [28], and Melbourne (20 ppb) [29]. The annual levels of PAN, VOCs, NO, NO 2 , and SO 2 were 0.81 ppb, 26.18 ppb, 9.18 ppb, 13.82 ppb, and 2.55 ppb, respectively. The levels of PM 2.5 and CO were 26.66 µg·m −3 and 0.62 mg·m −3 , respectively.
The mean concentration of O 3 in autumn (36.16 ppb) was significantly higher than in all other seasons, with the maximum daily concentration also occurring in autumn (72.76 ppb), unlike in Chengdu and Beijing, but the same as the previous result from Shanghai in the YRD [17,27,30]. It reflects the local synoptic flow pattern, which is the product of the interaction of the East Asian monsoon, tropical cyclones, and the land-sea breezes over the YRD [31]. The average O 3 concentration was the lowest in winter, which was due to weaker photochemical reactions at low ultraviolet radiation. Figure 2 shows there is a significant correlation between O 3 and PAN (p < 0.05), but the lowest mean PAN concentration occurred in summer (0.59 ppb). It was explained by the location of the observatory in the YRD, which was influenced by the East Asian summer monsoon that brought clean, humid air masses and diluted PAN, none of which were conducive to photochemical production (Li and Fan, 2022). It is noteworthy that the PAN had the highest average level in the spring (0.94 ppb), which was inconsistent with some previous reports [32,33]. We attribute this to low photodegradation efficiency and accumulation of long-term non-methane volatile organic compounds (NMVOCs) in the free troposphere during winter [34,35]. Photochemistry became active in early spring and accumulated NMVOCs promoted PAN accumulation, resulting in the highest PAN levels in spring, matching the mean level of VOCs in this study, which was highest in winter (39.03 ppb).  The averaged values for PM 2.5 , NO 2 , and CO were significantly higher in winter than in other seasons, at 40.87 ppb, 21.46 ppb, and 0.71 ppb, respectively. It could be caused by weak convection in the winter, leading to higher concentrations of accumulated pollutants [36]. NO and SO 2 were concordant with O 3 , with average concentrations highest in the autumn. Meanwhile, the ratio value of NO/NO 2 was greater than 1.0 (3.05), indicating that less O 3 consumption occurred in the NO 2 photolysis cycle in autumn [37].

Diurnal Variation
The average diurnal variation patterns for O 3 , PAN, and some other pollutants as well as meteorological parameters during the 4 seasons of 2021 are shown in Figure 3. O 3 as a secondary pollutant showed the highest value during the early afternoon (15:00-16:00) and the lowest value at 07:00-08:00. Temporal variations in solar radiation and temperature were considered major drivers of such diurnal variations in O 3 levels [2]. PAN has a similar diurnal pattern to O 3 , reaching a maximum between 11:00 and 14:00 in all seasons, then decreasing during low solar radiation, and a minimum in the early morning (06:00-08:00, indicating the dominance of local photochemistry during the observation period [11]. Specifically, the PAN peak usually occurred 1 h earlier than that of O 3 , presumably resulting from the increasing decomposition rate of PAN with increasing temperature [27]. The variation between maximum and minimum values of PAN in summer was the highest (0.94 ppb) while was the smallest difference in winter (0.70 ppb), which was a net growth pattern that also indicates that the lifetime of PAN increases with decreasing temperature. Contrastingly, NO x , CO, and VOCs levels showed a diurnal pattern opposite to O 3 ( Figure 3). The diurnal variation of NO 2 exhibited a bimodal distribution, with a peak in the morning, followed by a decrease in NO 2 concentration due to photolysis, and subsequent accumulation of NO 2 at night due to primary emissions. The peak NO x and CO levels in the morning were strongly related to vehicle emissions during the morning rush hour. The trend of VOCs concentration was the same as the daily variation of NO 2 , with a higher concentration in the morning, followed by a gradual decrease, but then a higher concentration at night, which was associated with the lower photochemical losses at night and the accumulation of primary emissions of pollutants. Anyway, a close correlation between precursor emissions and human activities (e.g., transportation) can be seen in the observed areas.

The Influencing Factors of O 3 Using the GAM
Eight parameters were selected as explanatory variables (PAN, VOCs, PM 2.5 , NO, NO 2 , CO, T, RH) and O 3 concentration as the response variable. The multi-factorial correlation analysis was performed using the GAM and the results are shown in Figure 4 and Table 2.   (Figure 4e), which was consistent with previous results from Beijing [16], but the degree of freedom (df) of NO 2 in this study was 1, indicating that a large proportion of O 3 was directly produced by NO 2 photolysis [38]. The effect of PAN on O 3 was also not negligible, showing a nonlinear positive correlation between the two with a narrow confidence interval (CI) (Figure 4a). In general, PAN inhibits O 3 formation by competing with O 3 precursors and terminating free radical chain reactions [11]. However, the positive correlation results implied that PAN may also promote O 3 production by providing more RO 2 radicals and increasing the oxidation capacity of the atmosphere in the presence of sufficient NO x [33]. Therefore, controlling vehicle emissions can reduce NO x levels and effectively mitigate the O 3 -promoting effect of PAN.
The Edf of T and RH were both greater than 1 (Figure 4g,h), demonstrating a nonlinear relationship with the response variable. When T < 27 • C, the O 3 markedly increased with rising T, implying that a certain range of heating can promote the photochemical reaction of O 3 . In contrast to the direct linear relationship of many studies [16,39], the increase in temperature above 27 • C inhibited O 3 formation. This inhibition of O 3 formation at high temperatures was not a coincidence, as a similar situation was observed by the University of California [40]. This phenomenon was driven by atmospheric chemistry and ecosystem-climate interactions due to the strong function of an e-folding decrease of PAN at high temperatures, as well as in areas with strong sources of isoprene and NO x , where chemistry is more VOCs-limited could experience a decrease in O 3 at high levels of temperature [40,41]. In addition, high temperatures may enhance surface heat flux and convective mixing, thereby increasing the atmospheric boundary layer height and diluting the O 3 concentration [42]. When RH < 55%, the effect of RH on O 3 concentrations did not change significantly, and when RH > 55%, O 3 levels decreased remarkably with the increase of RH due to the interception effect of RH on precursors and the fact that O 3 was dissolved in atmospheric water droplets and self-degraded at high relative humidity [43].
As levels of VOCs, PM 2.5 , and CO increased, O 3 concentrations initially increased and then gradually decreased (Figure 4b,c,f). Higher PM 2.5 levels contributed to increased O 3 levels through the scattering effect of PM 2.5 in a certain range, but excessively high PM 2.5 levels reduced terrestrial ultraviolet, leading to the inhibition of photochemical reactions and hence lower O 3 levels [21,44]. CO and VOCs had little effect on O 3 and it seems reasonable to assume free radical reactions with NO x dominated in the region. NO displayed a complex relationship with O 3 , but was generally negatively correlated due to their susceptibility to reaction [2]. HO 2 in a high-NO atmosphere promotes the oxidation of NO to NO 2 , and NO consumes peroxyacetyl radicals to generate NO 2 , promoting O 3 formation [9,39]. In summary, the multifactorial GAM is more interpretable and simulates more realistic O 3 trends in the atmosphere. It demonstrates that in the YRD NO 2 and PAN have the greatest influence on O 3 levels, followed by T and RH.

Sensitivity of O 3 Formation
In this study, the RIR values calculated by OBM for the precursors in all seasons are shown in Figure 5. The RIR values for VOCs were significantly higher than those for NOx in spring and winter (Figure 5a,d), with a negative RIR value for NO x in the winter (−0.36), indicating O 3 production in the observing area was mainly controlled by VOCs. Notably, the RIR of BVOC (isoprene) in winter was only 0.01, which can be attributed to the fact that plant branches became bare in winter and BVOC emissions were greatly reduced, which, together with low temperatures and weak radiation in winter, caused the effect of isoprene on O 3 formation to be lower [45]. Moreover, formaldehyde (FORM) and xylene (XYL) showed the top two RIRs for O 3 in spring and winter, revealing their dominance in the O 3 generation. Therefore, reducing VOCs in these two seasons is more effective for controlling O 3 pollution. Additionally, previous studies have concluded that anthropogenic primary sources (e.g., vehicle emissions and industrial activities) contributed most to FORM in the spring and winter, that biological sources contributed more in the summer and autumn, and that the major sources of XYL were traffic and industry [46,47]. O 3 production in summer was more sensitive to NO x , with a RIR of 0.34. Toluene (TOL) and XYL of AVOCs appeared to have negative values and decreases in their concentrations will instead lead to an increase in O 3 concentrations. In autumn, O 3 formation displayed a high sensitivity to simultaneously VOCs and NOx (RIR (VOCs) : 0.24, RIR (NOx) : 0.29), while the effect of CO on O 3 formation was negligible (RIR (CO) : 0.01). Further analysis showed reducing TOL has an adverse impact on O 3 formation and FORM needs to be prevented and controlled. In a nutshell, O 3 formation was in the VOCs-limited in spring and winter, controlled by NO x in summer, and even controlled by both VOCs and NO x in autumn, and FORM emissions have to be emphasized throughout the year.
Empirical Kinetics Modeling Approach curves were simulated and plotted to investigate the impacts of precursors reduction on O 3 formation ( Figure 6). In other words, the relationship of P(O 3 ) with relative changes of S(VOCs) and S(NO x ) can be expressed by isopleth diagrams for P(O 3 ). The mean P(O 3 ) levels varied considerably over the four seasons, with estimates of 262 ppb, 165 ppb, 205 ppb, and 77 ppb, respectively. In spring, a 10% reduction in S(VOCs) resulted in a decrease of 13 ppb in P(O 3 ), and a 10% reduction in NO x only resulted in a reduction of 2 ( Figure 6a). In winter, O 3 levels gradually decreased with an increasing reduction ratio when only VOCs was reduced; however, O 3 levels progressively increased when only NO x was reduced, especially when the reduction ratio reached to 40% (Figure 6d). It indicated that the regime was in the VOCs-limited in spring and winter as the results of the RIRs. During summer, O 3 formation was more sensitive to NO x , with an increase in the percentage of NO x reduction leading to a notable reduction in O 3 levels, while VOCs reduction required a large percentage of reduction. For autumn, the S(VOCs) and S(NO x ) data point was close to the ridge line, indicating the point was in a transition regime where significant NO x reductions can be achieved in the short term but easily transition to the NO x -limited regime. Thus, stringent control of VOCs pollution ought to be implemented in parallel with collaborative regional prevention and control of NO x to facilitate long-term control of O 3 . However, many studies have not studied the seasonal sensitivity differences in depth and finally only obtained that the study area was in the VOCs-limited control or NO x -limited [48,49]. Based on the above conclusions, it is necessary for YRD to dynamically adjust its prevention and control strategy in accordance with the characteristics of the O 3 formation mechanism.
centrations will instead lead to an increase in O3 concentrations. In autumn, O3 formation displayed a high sensitivity to simultaneously VOCs and NOx (RIR(VOCs): 0.24, RIR(NOx): 0.29), while the effect of CO on O3 formation was negligible (RIR(CO): 0.01). Further analysis showed reducing TOL has an adverse impact on O3 formation and FORM needs to be prevented and controlled. In a nutshell, O3 formation was in the VOCs-limited in spring and winter, controlled by NOx in summer, and even controlled by both VOCs and NOx in autumn, and FORM emissions have to be emphasized throughout the year. Figure 5. The observation-based models (OBM) calculated relative incremental reactivity (RIR) for O3 precursors (green) and specific species (red) in (a) spring, (b) summer, (c) autumn, and (d) winter during the daytime (07:00-19:00). BVOCs and AVOCs stand for biological VOCs and anthropogenic VOCs, respectively. ETH, PAR, ALD2, FORM, TOL, OLE, and XYL stand for ethylene, alkanes, aldehydes other than formaldehyde, formaldehyde, toluene, alkenes other than ethylene, and xylene, respectively. If the RIR value is positive, the reduction of precursors contributes to O3 reduction, while a negative value means that precursor reduction may lead to an increase in O3 concentration. Figure 5. The observation-based models (OBM) calculated relative incremental reactivity (RIR) for O 3 precursors (green) and specific species (red) in (a) spring, (b) summer, (c) autumn, and (d) winter during the daytime (07:00-19:00). BVOCs and AVOCs stand for biological VOCs and anthropogenic VOCs, respectively. ETH, PAR, ALD2, FORM, TOL, OLE, and XYL stand for ethylene, alkanes, aldehydes other than formaldehyde, formaldehyde, toluene, alkenes other than ethylene, and xylene, respectively. If the RIR value is positive, the reduction of precursors contributes to O 3 reduction, while a negative value means that precursor reduction may lead to an increase in O 3 concentration.

Conclusions
Long-term O 3 observations in the YRD in 2021 displayed strong seasonal variations with a maximum in autumn (72.76 ppb) due to the metrological interaction and the lowest O 3 level in the low-radiation winter (19.03 ppb). O 3 levels displayed an obvious cyclic pattern of diurnal variation, with O 3 showing the highest values in the early afternoon (15:00-16:00) due to vivid photochemical reactions and the lowest values in the 07:00-08:00. PAN presented a similar diurnal pattern to O 3 ; however, the rate of decomposition of PAN increased with increasing temperature, resulting in the peak of PAN usually occurring 1 h earlier than the peak of O 3 precursors (NO x , CO, and VOCs), which, in contrast to O 3 , showed a diurnal pattern, with the lowest levels in the afternoon and the maximum in the night or the morning peak.
Furthermore, GAM revealed key factors affecting O 3 levels were NO 2 , PAN, and T. A large fraction of O 3 was produced directly by NO 2 photolysis, and PAN contributes to O 3 production by providing more RO 2 radicals and increasing the oxidation capacity of the atmosphere in the presence of sufficient NO x . Thus, reducing vehicle NO x emissions can effectively mitigate the O 3 -promoting effect of PAN. It was found a certain temperature rise promoted the photochemical reaction of O 3 , whereas rising temperatures above 27 • C inhibited O 3 formation. It is strongly recommended to target control in different seasons according to various O 3 formation mechanisms. Based on the RIRs, FORM needs to be emphasized all year round.
This study extends the understanding of O 3 pollution in the YRD region, integrates the coordinated effects of multiple parameters on O 3 production, and quantifies the contribution of PAN to O 3 formation for the first time, and proposes seasonal control of various precursors which are significant guidelines for photochemical pollution control in the YRD region, China.