Contribution of Regional Transport to Surface Ozone at an Island Site of Eastern China

The surface ozone (O3) in non-urban areas can be strongly influenced by transported regional air pollutants from distant urban agglomerations. Comprehensively using the polar plot graphical technique, and backward trajectory and potential source contribution analyses, the variations of surface O3 and its potential source regions at an island site of Zhejiang, China, were analyzed based on data from June 2013 to October 2016. Relatively high hourly O3 levels (max.: 154.0 ppb) were often observed, with the total number of hourly O3 exceeding the national standard (75 ppb) being 583, indicating high health risks from O3 exposure. An obvious time-lag for the diurnal O3 peak was observed, with the daily maxima often occurring at 16:00. A decreased O3 level was observed in summer, which was probably due to the comprehensive influence of intense rainfall, high relative humidity and clean marine airmasses during this season. The transport of external air pollutants played a dominant role in affecting local O3, with relatively high O3 concentrations (> 50 ppb) being observed under high wind speeds (> 5 m s). Based on all the source identification analyses, the major regions contributing to the surface O3 lay to the north and northwest of the study area. Long-range transported air pollutants from the coastal provinces of the North China Plain may have significantly enhanced surface O3 at the study site, while contributions from local areas and areas to the south of the study site were rather small. A high incidence (64.3%) of O3 pollution was accounted for by the inflowing airmasses from outside of Zhejiang, which quantitatively confirmed the significant background transport of O3 to this province. Coordinated inter-regional control on pollutant emissions should be carried out to reduce O3 in Zhejiang Province.


INTRODUCTION
Surface ozone is a major trace gas which is of great importance for climate and air quality.It is mainly produced from its precursors such as carbon monoxide (CO), nonmethane volatile organic compounds (NMVOCs) and nitrogen oxides (NO x ) by photochemical reactions in the presence of sufficient solar radiation (Sillman, 1999).High levels of O 3 can adversely affect vegetation growth and human health due to its strong oxidizability and phytotoxicity (Monks et al., 2015).A large quantity of crop yield loss and premature human deaths have been reported to be O 3related (Fann et al., 2012;Feng et al., 2015).Besides, as the third major greenhouse gas, O 3 could play an important role in the process of global warming (Stocker et al., 2013).
With the increasing anthropogenic emissions of O 3 precursors, a steady rise in ground-level O 3 was widely observed or predicted (Nopmongcol et al., 2016), especially in the mid-latitude regions of northern hemisphere where background O 3 concentrations have increased continuously (0.5-2% per year) during past decades (Vingarzan, 2004).As the biggest developing country in the world, China has experienced tremendous changes in recent years.Continuous rapid urbanization and industrialization in China has made it a photochemically polluted region, with rising concentrations of O 3 and its precursors being a matter of growing concern for both public administrations and citizens (Ou et al., 2016).High levels of O 3 exceeding the National Ambient Air Quality Standard (NAAQS) of China have been frequently observed in developed urban areas (Li et al., 2016a;Tong et al., 2017a).
Given the damaging effects of surface O 3 on populations and ecosystems, and the increasing trends of its concentrations, it is critical to identify the possible sources of O 3 , which is fundamental to design effective strategies on emission controls.Surface O 3 has three major sources, i.e., in situ photochemical production, lateral boundary import, and downward transport from upper atmosphere (e.g., stratosphere) (Ma et al., 2002;Lin et al., 2015;Zhu et al., 2016).Due to the influences of meteorological conditions and the nonlinear relationships between O 3 and its precursors, it is usually complicated to accurately identify the sources of O 3 at a specific location.Over the past several decades, great efforts have been made on this subject by researchers worldwide.For example, using a regional threedimensional chemical transport model (3-D CTM), Berntsen et al. (1999) studied the transport of O 3 across the North Pacific and evaluated the influence of Asian emissions on the air quality of the United States; Ma et al. (2002) analyzed the contributions of various sources to the O 3 distributions in the troposphere over China during summertime based on a global 3-D CTM; Sudo and Akimoto (2007) quantified the contributions from different source regions to global distributions and budgets of tropospheric O 3 in the context of intercontinental transport using the tagged tracer simulation; Demuzere et al. (2009) and Pope et al. (2016) assessed the influences of the large-scale circulation on O 3 in the Netherlands and the United Kingdom, respectively, using the objective Lamb weather type approach; Shen et al. (2015) quantified the sensitivity of O 3 air quality in the eastern United States to the major patterns in synopticscale circulation by applying empirical orthogonal functions; Li et al. (2016b) estimated the cross-boundary transport of O 3 between Beijing and the North China Plain by applying comprehensively the bivariate polar plot graphical technique, the transport flux assessment and the potential source contribution function (PSCF) analysis.All these previous studies have provided important insights into the causes and control strategies of O 3 pollution.However, a majority of them were carried out in urban areas but few in suburban and rural areas, where the O 3 levels might be even higher due to the contributions of O 3 precursors from regional transport (Chiang et al., 2009;Kanaya et al., 2016).
Zhoushan is a small island city of Zhejiang Province in eastern China.The level of industrialization in the city is low with less emission of primary air pollutants into the atmosphere.Due to the influences of subtropical monsoon and strong sea-land breezes, local formation and accumulation of secondary air pollutants are also weak.Consequently, Zhoushan has been continuously one of the top ten cities with the best air quality in China in recent years (http://datacenter.mep.gov.cn/index).In contrast, the area near Zhoushan, i.e., the Yangtze River Delta (YRD) region, is more polluted with substantial air pollutants being emitted annually (Fu et al., 2013).YRD is one of the most developed and heavily-polluted city clusters of China, which consists of Shanghai Municipality and another 25 cities within Jiangsu, Zhejiang and Anhui Provinces (Fig. S1).The annual emissions of NO x and NMVOCs in the YRD region are estimated to be 2776 and 3822 kiloton, respectively, in 2010, and the VOC emissions increased by 110% from 2000 to 2010 (Fu et al., 2013).The majority of these air pollutants are accounted for by industrial and traffic emissions from large cities such as Shanghai and Suzhou (Fu et al., 2013).The significant differences in the pollution statuses between Zhoushan and its surrounding cities create a good condition for the study of regional influences on local air quality.
So far, no published studies about surface O 3 in the city of Zhoushan have been reported, although high levels of O 3 over the NAAQS were frequently observed during warm seasons (http://datacenter.mep.gov.cn/index).In this study, the O 3 pollution in Zhoushan over the period of June 2013-October 2016 was analyzed based on different techniques, including the bivariate polar plot graphical technique, the backward trajectory analysis, the potential source contribution function (PSCF) analysis, and case studies under shifting wind conditions.The main objectives are to determine the potential regions contributing to surface O 3 in Zhoushan, and to assess the influence of regional transport on O 3 pollution in this island city of eastern China.

Study Area
Zhoushan is one of the most important port cities for opening to the outside world of China.This city is famous for its developed marine fishery and port transportation.Zhoushan has the biggest fishing ground of China, and it is also a major component of Ningbo-Zhoushan port, which is the largest container port in the world with annual cargo throughput being higher than 900 million tons in 2016.As to the topography of Zhoushan, it is mountainous and about 70% of its land is covered by hills with altitudes of 200-300 m.The study site is located in the Administrative Service Center of Putuo District (ACPD) (29°57′N, 122°19′E), which is in the southeast of the city of Zhoushan (Fig. 1).This site is in a comprehensive area with a few commercial and residential buildings nearby.A mountain is 0.2 km away from the west of ACPD while the sea is 0.8 km away from the east of this site.Some small islands with relatively developed tourism are distributed along the southeast to the study site.
In terms of the climate conditions, the study area is subject to the sub-tropical monsoon with hot, humid summers and cool, damp winters.The annual mean air temperature (T), relative humidity (RH) and precipitation are 16.7°C, 79% and 1222.6 mm, respectively.The air temperature reaches its maximum (25.8-28.0°C)and minimum (5.2-5.9°C) in August and January, respectively.Intensive precipitation often occurs in June and August due to the influences of plum rain and typhoon weather.Annual mean wind speed is 3.5 m s -1 with northwest and southeast winds being dominant in winter and summer, respectively.

Data Collection
In-situ measurements of surface O 3 were carried out with a commercial trace gas instrument (Model 49i, Thermo Fisher Scientific Inc., USA).It functions on the absorption of ultraviolet light at 254 nm with the O 3 concentration being determined based on Lambert-Beer's law (Thermo Fisher Scientific Inc., 2017).Hourly average data monitored during the O 3 seasons (April-October) from June 2013 to October 2016 were used for analyses in this study.All times used are local time (China Standard Time, CST), which is 8 h ahead of the Coordinated Universal Time (UTC+8).In order to assess the influence of regional transport on surface O 3 in the study area, the hourly O 3 concentrations in the upwind cities of Jiangsu, Zhejiang and Fujian Provinces (Fig. S2) were collected from the data center of the National Ministry of Environmental Protection of China (MEPC) (http://datacenter.mep.gov.cn/).All the data released by MEPC were measured in-situ using commercial trace gas instruments (e.g., Model 49i, Thermo Fisher Scientific Inc., USA), which were maintained by third-parties to ensure the data quality.Meteorological data were also monitored on hourly basis using a portable automatic weather station (WS500-UMB, Lufft, Germany) established at ACPD.The meteorological variables applied in this study include air temperature (T, °C), relative humidity (RH, %), wind speed (WS, m s -1 ), wind direction (WD, °) and rainfall (mm).For all the monitoring data, only the days with a minimum capture rate on hourly data of 75% and the months with at least 20 days of valid data were applied to statistical analyses.

Transport Direction Analysis
The relationships between pollutant concentrations and wind conditions can provide useful information on the transport directions of air pollutants.For example, high concentrations of air pollutants associated with high wind speeds often indicate significant regional contribution from upwind areas (Grange et al., 2016).The transport directions of O 3 were analyzed in this study based on the bivariate polar plot graphical technique developed by Carslaw et al. (2006).Specifically, wind speed, wind direction and concentration data are partitioned into wind speed-direction bins with the mean concentrations being calculated for each bin.Then the wind components (u and v) are calculated with a generalized additive model (GAM) (Eq.( 1)) being used to fit the surface data and to extract the possible source features: where C i is the calculated pollutant concentration, s(u i , v i ) is the isotropic smooth function of i th value of covariate u and v, and e i is the residual.Based on Eq. ( 1), a smoothed surface is fitted to the wind speed-direction bins to create a continuous surface which can be plotted with polar coordinates.More details about this equation can be found in Yu et al. (2004), Carslaw et al. (2006) and Grange et al. (2016).All the analyses were made with Openair, an R package for air quality data analysis (http://www.openairproject.org/Downloads/OpenAirManual.aspx).

Back Trajectory and PSCF Analyses
The backward trajectories of airmasses arriving at the study site were calculated using the HYSPLIT-4 model from the National Oceanic and Atmospheric Administration Air Resources Laboratory (http://ready.arl.noaa.gov/HYSPLIT.php) and the meteorological data from the National Centers for Environmental Prediction Global Data Assimilation System (1 × 1°) (ftp://arlftp.arlhq.noaa.gov/pub/archives/gdas1).A tracking time of 72 h was used in this study with hourly trajectories at 50, 500 and 1000 m above ground level (AGL) being calculated from 00:00 to 23:00 (UTC) during the O 3 seasons (April-June for 2013-2016).A total of 57,240 hourly trajectories (19,080 for each height) were calculated, which were mainly used for the following analyses: (1) All the 57,240 hourly trajectories were applied to the PSCF analysis to determine the potential regions contributing to surface O 3 at ACPD; (2) 1249 hourly trajectories with the arriving time at ACPD of 14:00 CST (June 2013-October 2016), when the O 3 concentrations were usually high enough to pose threats to human health under suitable photochemical conditions (e.g., high air temperature), were applied to determine the influence of the inflowing airmasses to Zhejiang and (3) six hourly trajectories with the arriving time at ACPD of 14:00 CST were analyzed to further confirm the potential source regions of O 3 during the pollution episodes.As the trajectories with the starting height of 500 m AGL could represent the average flow field of atmospheric boundary layer (c.a.1000 m AGL in height), in which air pollutants are well mixed by horizontal and vertical advections, only the 500-m trajectories were used for the (2) and (3) analyses above.
PSCF analysis, which is a useful tool for determining the potential source of a given chemical constituent (Ashbaugh et al., 1985), was applied to identify the potential regions contributing to surface O 3 using the calculated trajectories of air parcels.The PSCF values can be regarded as the conditional probability that the concentrations of a given pollutant higher than the critical value are related to the passages of airmasses through a certain area during transport.It has been applied to identify the potential source regions of O 3 in different cities worldwide (e.g., Li et al., 2016b;Vellingiri et al., 2016).
In this study, the PSCF analysis was made using the Geographical Information System based software TrajStat (version 1.4.4R5),which is developed by Wang et al. (2009).The study domain for PSCF analysis was 2-75°N and 61-165°E with a resolution of 0.5 × 0.5°.The PSCF values could be calculated from the TrajStat software according to the equation as follow: where N ij is the total number of trajectory endpoints in the ij th grid cell and M ij is the total number of trajectory endpoints in the same cell with the pollutant levels at the sampling site being greater than a criterion value (Ashbaugh et al., 1985).The criterion value for O 3 was set at the 75 th percentile of the dataset for the study period from June 2013 and October 2016.To reduce the uncertainty in cells with small N ij , the PSCF values were multiplied by a weight function (W ij ), which was defined as follows (Wang et al., 2015):

Trajectory Analyses for the Air Flowing into Zhejiang
As Zhoushan is a border city in the northeast of Zhejiang Province (Fig. 1), the O 3 concentrations in the air parcels arriving at Zhoushan via East China Sea represent the background levels of O 3 flowing into Zhejiang Province.To assess the influences of background transport from outside of Zhejiang Province, afternoon (14:00 CST) 72-h back trajectories at ACPD over the study period were analyzed.According to Aleksic et al. (2016), only the trajectories, which spent less than three hours in Zhejiang Province before arriving at the study site, were used to eliminate the influence of pollutant emissions from within Zhejiang.
To explore the major pathways of airmasses flowing into Zhejiang, trajectories were clustered using TrajStat (version 1.4.4R5,http://www.meteothinker.com/downloads/index.html),which is a Geographical Information System (GIS)-based software developed by Wang et al. (2009).The clustering analysis was carried out based on Euclidean distances between trajectories, which were given by: where X 1 (Y 1 ) and X 2 (Y 2 ) refer to backward trajectories 1 and 2, respectively.Initially, each trajectory is defined to be a cluster.For the first iteration, two trajectories are paired.
For every combination of trajectory pairs, the cluster spatial variance, which is the sum of squared distance between the endpoints of cluster's component trajectories, is calculated.Then, the total spatial variance (TSV), the sum of all the cluster spatial variance, is calculated.The pair of clusters finally combined are the ones with the smallest change in TSV.The iterations continue until the last two clusters are combined.The percent change in TSV and number of clusters for each iteration are recorded.The final cluster number was determined by the point of inflexion on the curve of percent change in TSV vs. number of clusters (Fig. S3).More details about this methodology can be found in Draxler et al. (2014).

Overview of Surface O 3
Relatively high O 3 levels over NAAQS (Grade I: 75 ppb; Grade II: 93 ppb) were often observed in Zhoushan during the study period (Fig. 2).For each O 3 season (April-October) of different sampling years, the number of hourly O 3 greater than Grade I NAAQS (75 ppb) varied from 37 in 2013 to 214 in 2015, with a total of 583 being observed during the whole study period (Table 1).The maximum O 3 concentration during the whole study period was as high as 154.0 ppb (May 28, 2014), which was 61 ppb higher than the Grade II NAAQS (93 ppb).All these results indicated the potential health risks of O 3 pollution to local people at the study site.Similar high O 3 concentrations (≥ 150 ppb on hourly basis) were also observed in other cities (e.g., Hangzhou, Shanghai and Ningbo) of the same region (YRD) as this study (Xu et al., 2008; Li et al.,  2016a; Tong et al., 2017a).During warm seasons, the O 3 pollution could occur on the regional scale of eastern China (Wang et al., 2006).In view of the harmful effects of O 3 on human health, effective measures on the emission control of O 3 precursors around the whole region are urgently needed.
In general, the monthly variation of O 3 showed an opposite trend with that of air temperature (Figs. 3 and 4), with a significant negative correlation (Spearman's rank   correlation coefficient r = -0.26,p < 0.001) being observed between both variables for the whole study period.The O 3 levels were relatively high in spring (April and May) and autumn (September and October), and were obviously low in summer (June-August).This was contrary to the relatively high O 3 levels in summer, which have been widely observed worldwide (Dueñas et al., 2004;Mavroidis and Ilia, 2012).The possible reasons for the summer minimum of O 3 observed at ACPD are as follows: (1) In this study, higher precipitation was observed in summer (Fig. 4), which indicated cloudy and rainy weather that are unfavorable to the photochemical generation of O 3 .This might partly account for the low O 3 concentration observed during summertime.Similar effect of summer rainfall on surface O 3 has also been reported by Jo and Park (2005), Reddy et al. (2012) and Tong et al. (2017b).(2) The high relative humidity, which was observed simultaneously with the high air temperature in summer (Fig. 4), was favorable to the O 3 removal.As reported by Fiore et al. (2002), O 3 can be easily removed via reactions with water vapor under humid  weather conditions.With the increase of relative humidity relating to greater cloud abundance, the photochemical processes associated with O 3 generation will be slowed down while O 3 depletions will be enhanced (Londhe et al., 2008).This process of O 3 removal can be partly confirmed by the significant negative correlation between O 3 and RH (r = -0.49,p < 0.001) observed in this study.Therefore, the highest RH observed in summer at ACPD (Fig. 4) would probably contribute to the decrease in surface O 3 .(3) The study site was predominantly influenced by summer monsoon from June to August, when most (66.3%) of the airmasses arriving at Zhoushan came from the east and south marine area during the study period.These clean airmasses with less O 3 and its precursors might play an important role in reducing local O 3 concentration at the study site.The cleaning effect of inflowing marine airmasses, which resulted in the summer minimum of O 3 , has also been observed in other areas of East Asia (Zhang and Oanh, 2002).
As to the diurnal variation, hourly average O 3 concentration for the whole study period varied from 29.4 to 48.2 ppb (Fig. 5).The daily variation of O 3 concentration obviously lagged behind that of air temperature.The maximum value of O 3 occurred at 16:00, which is three hours later than the highest air temperature (13:00).Similar time lags for O 3 peaks were also observed in rural areas of other cities like Beijing and Hangzhou, China (Xu et al., 2011;Li et al., 2017).The possible reason for the O 3 time lag observed at ACPD might be the weak contribution from local sources but relatively strong regional contribution to surface O 3 in the study area.As reported by Fu et al. (2013), a considerable amount of VOCs (131.0 kiloton), which is close to the median (134.1 kiloton) among the 25 cities in the YRD of China, were emitted annually in Zhoushan.However, the NO x emission (13.3 kiloton) was the lowest of all the cities.The total emission amount of O 3 precursors (NO x + VOCs) was smaller in Zhoushan than most of other cities in the YRD, which indicated less potential for O 3 production from precursors emitted locally.The ratio of NO x to VOCs is an important factor in affecting O 3 formation (Sillman, 1999).Given the small emission amount of NO x , the in-situ O 3 production might also be low under an NO xlimited condition.Besides, the study site is located in a coastal area with relatively high wind speed of 3.9 m s -1 (diurnal average) during the study period.The in-situ photochemical production and accumulation of O 3 might be low as well, due to the cleaning effect of sea breeze.Therefore, the daily O 3 peak, which usually occurs at noon with the highest temperature, was inconspicuous due to the weak contribution from local emission sources.In comparison with the in-situ O 3 formation, the regional transport of air pollutants had played a more important role in the local O 3 increases (see the results from the PSCF and the Polarplot analyses in the latter part of the text).It may take longer time for the accumulation of surface O 3 to the maximum level after distant transport of air pollutants (Xu et al., 2011), which might account for the peak lag of O 3 observed in this study.

Associations of O 3 with Wind Direction/Speed
Based on the polarplot analyses, high levels (> 50 ppb) of O 3 were found to be associated with airmasses coming from north and northwest areas to the study site (Fig. 6(a)).Local production of O 3 was relatively weak with lower O 3 levels when wind speed was close to zero.In contrast, long-range transport of air pollutants had significantly contributed to surface O 3 at the study site, which can be inferred from the high O 3 concentrations (> 50 ppb) occurring under high wind speeds (> 5 m s -1 ) (Fig. 6(a)).These are similar to the results found in the coastal region of northeastern Iberian Peninsula, where high surface O 3 levels were not produced in situ but were largely contributed from horizontally advected flows (Gonçalves et al., 2009).The north region to ACPD is the marine area adjacent to the North China Plain (Fig. 1), where outflow of large amounts of O 3 precursors could take place during monsoon seasons with prevailing westerly winds (Zhang et al., 2002).As for the northwest region to the study site, it covers one of the most polluted provinces of China (i.e., Jiangsu) with large amounts of O 3 precursors being emitted annually (Li et al., 2016a;Wu et al., 2016).Long-range transported air pollutants from these two regions had probably contributed to the increase in O 3 levels in the study area.The relatively high O 3 concentrations (> 45 ppb) associated with strong north winds during nighttime (Fig. 6(c)) could also confirm this influence of regional transport.Given the large contribution of pollutant transport from other regions to the O 3 pollution in Zhejiang, the coordinated inter-regional control on pollutant emissions in the northern upwind areas will be quite necessary to help improving local and regional air quality.
In addition to the analyses concerning the whole study period, all the pollution episodes with O 3 concentrations continuously exceeding the 75 th percentile (51.8 ppb) of the dataset for 8 hours were also analyzed using polarplot technique.Higher O 3 concentrations were experienced with northwest winds when their speeds were higher than 6 m s -1 (Fig. 7  which suggested the critical role of photochemical activities on the O 3 pollution.However, the surface O 3 photochemically produced during the daytime could also accumulate at night, as indicated by the relatively high O 3 levels (50-70 ppb) during the nighttime of the pollution episodes (Fig. 7(c)).Similar nighttime accumulation of O 3 has also been reported in the coastal cities of Europe (e.g., Barcelona, Spain), where the O 3 transport from inland regions was influential (Valverde et al., 2016).
Besides northwest, another transport direction of O 3 during the pollution episodes was also detected to be the southeast when wind speeds were in the range of 2-6 m s -1 (Figs.7(a) and 7(b)).As the southeast region to the study area is in the East China Sea without large emission sources, the air pollutants transported from this direction were probably associated with the human activities (e.g., tourism and freight transportation) on the small islands surrounding ACPD (Fig. 1).It is worth mentioning that a civil airport and an international cruise terminal are located on the island to the southeast of ACPD with distances of 5.5 km and 8.8 km, respectively.The air pollutants emitted from the aircrafts and ships might contribute comprehensively to the O 3 increase in the study area when southeast winds prevailed over Zhoushan.More measurements on precursor emissions from these transportation vehicles are still needed in the future work to confirm this speculation.In contrast to the influence from short-distance transport of air pollutants, local contribution to O 3 pollution is relatively weak as indicated by the low O 3 concentrations at low wind speeds (Figs.7(a) and 7(b)).This is probably due to the few emission sources of O 3 precursors at the study site as well as the good dispersive conditions of atmosphere.

Potential Areas Contributing to Surface O 3
Based on the PSCF analysis, the major potential areas contributing to surface O 3 at ACPD (weighted PSCF > 0.4) were found to lie in the coastal region of the North China Plain and its adjacent marine areas (Fig. 8).All of them were located to the north and northwest of the study site, which is consistent with the results derived from the polarplot analyses discussed in previous section.The core areas with high probabilities (weighted PSCF > 0.5) were far away (> 400 km) from ACPD, which indicated the contribution of long-range transported pollutants to the O 3 at the study site.The regions to the west and southwest of ACPD (e.g., Zhejiang and Fujian Provinces) are unlikely to be the major potential source areas of O 3 , which can be indicated from the relatively low values (< 0.4) of weighted PSCF.Besides the coastal areas of the North China Plain, South Korea was identified as another major O 3 source region of ACPD (weighted PSCF > 0.5) (Fig. 8).It is independent from the source region near the North China Plain, and is also far away (> 600 km) from the study site.As a developed Asian country, South Korea is famous for its advanced manufacturing industries (e.g., steel production, automobile manufacturing and shipbuilding).Large amounts of precursors (e.g., NO 2 ; Fig. S2) are emitted into the atmosphere annually in South Korea (Han et al., 2009;Mijling et al., 2015), which might significantly contribute to the O 3 rise at ACPD via long-range transport.

Trajectory Analyses for the Inflowing Airmasses to Zhejiang Province
During the study period between June 2013 and October 2016, a total of 562 trajectories (44.9%) were identified as the inflowing ones.All these trajectories spent less than three hours in Zhejiang Province, which was determined visually from the endpoints of airmass trajectories.The O 3 data were available on 485 days of the inflowing days with 77 days of them being missing due to the power failure or instrument malfunction.
During the four O 3 seasons over 2013-2016, a total of 70 hourly O 3 concentrations (14:00 CST) greater than NAAQS Grade I (75 ppb) were observed.While under the inflow conditions at 14:00 CST, 45 observations of hourly O 3 exceeded 75 ppb, which accounted for 64.3% of the total exceedances over 75 ppb.This suggested that the O 3 pollution in the study area was predominantly contributed from external transport.As Zhoushan was a border city of Zhejiang Province, the O 3 concentrations in airmasses flowing into this city could represent the background levels in Zhejiang.The obvious O 3 inflow observed at ACPD indicated a high contribution of regional transport to surface O 3 in Zhejiang, especially for the cities close to Zhoushan.In order to meet the O 3 requirements of NAAQS in Zhejiang Province, the coordinated inter-regional prevention and control on pollutant emissions should be carried out.
For the purpose of identifying the major pathways of O 3 transported into Zhejiang Province via the study area, cluster analysis was made using the 485 inflowing trajectories.Four trajectory clusters were derived with two representing the airmasses from East China Sea (Clusters 1 and 4), and two representing the ones from the coastal region of the North China Plain (Clusters 2 and 3) (Fig. 9).The frequencies of airmass trajectories varied from 9.8% (Cluster 4) to 36.5% (Cluster 2), with more than half (53.7%) of them coming from north to the study area (Table 2).In terms of the average O 3 concentrations for different clusters, high levels (> 56.0 ppb) were observed  for the airmasses from the north region to the study site (Clusters 2 and 3) (Table 2 and Fig. 10(a)).The percentages of O 3 levels greater than the 75 th percentile (51.8 ppb) were also high (> 30%) for Clusters 2 and 3.Both the high frequencies and high O 3 levels for the northern clusters (2 and 3) indicated that the coastal area of the North China Plain was the major potential region contributing to the O 3 pollution in Zhejiang Province.This is consistent with the results derived from the polarplot and PSCF analyses in this study.In contrast, the O 3 concentrations in the airmasses coming from the southeast marine area (Cluster 4) were the lowest (avg.: 25.4 ppb), with no O 3 values over the 75 th percentile (51.8 ppb) being observed.The lack of emission sources of O 3 precursors in the southeast marine area could account for the relatively low O 3 levels for the airmasses from this direction.
As to the "polluted" northern clusters (2 and 3), the O 3 concentrations for Cluster 3 (avg.: 59.8 ppb) were significantly higher (p = 0.032) than those of Cluster 2 (avg.: 56.8 ppb) (Table 2).In comparison with the airmasses of Cluster 2, those of Cluster 3 had longer transport pathways through the whole Hebei Province (Fig. 9), which is one of the most polluted regions of China (Wang et al., 2013).This suggested that the airmasses associated with Cluster 3 might carry more O 3 precursors than those of Cluster 2, which could lead to the higher levels of O 3 via photochemical reactions during transport.Besides, O 3 concentration is usually negatively correlated with relative humidity due to the O 3 removal through reactions with water vapor (Fiore et al., 2002).In this study, a significant negative correlation between O 3 concentration and relative humidity was observed for the inflowing airmasses (r = -0.58,p < 0.001).Cluster 3 was characterized by the lowest relative humidity of the four clusters (Fig. 10(d)), which was more suitable for the accumulation of O 3 .Therefore, higher O 3 concentrations were observed for the airmasses represented by Cluster 3. In addition to the relationship between O 3 and relative humidity, it is also worth noting that the O 3 concentration had an opposite trend of variation to that of air temperature among the four clusters (Fig. 10), with the highest O 3 level but the lowest temperature being observed for Cluster 3.This was similar to the result observed in the monthly variations of O 3 and air temperature (Figs. 3 and 4).The city of Zhoushan has a typical subtropical climate with airmasses from northwest inland of China being prevalent in cold seasons.Therefore, low air temperature was more associated with the northwest airmasses (represented by Clusters 2 and 3), which could carry a considerable amount of precursor pollutants (e.g., VOCs) emitted from northern China (Wu et al., 2016).This might account for the opposite trends between O 3 and air temperature among clusters (Fig. 10).

Ozone Episode Analyses
In order to further identify the potential regions contributing to surface O 3 over the study area, typical episodes with obvious changes in wind directions were selected based on the following rules.Firstly, the days with hourly O 3 levels exceeding 75 th percentile (51.8 ppb) of the dataset for at least three consecutive hours (Period I) were selected.Secondly, the days with obvious and continuous changes in hourly wind directions (Period II) were selected.Thirdly, the intersection between Periods I and II was selected as the days of episodes.The O 3 concentrations in these episodes were analyzed along with wind speeds, air temperature, relative humidity, rainfall, and the 72-h back trajectories of airmasses with the arrival time at ACPD of 14:00 (CST), when the O 3 concentrations were relatively high.Besides, the O 3 levels in the neighboring provinces of Zhoushan (Fig. S2), where the back trajectories had passed through, were also analyzed to preliminarily explore the possible contributing mechanisms of external sources to local O 3 in the study area.

Case Study 1
Obvious variations in O 3 concentrations and meteorological variables including wind direction, wind speed, air temperature and relative humidity were observed from June 3 to June 8, 2014 (Fig. 11).The wind direction changed suddenly from northwest to southeast in the middle of this period.Before June 6, high O 3 concentrations were observed under relatively cold and dry weather conditions which were associated with strong northwest winds.This is especially obvious on June 3 and June 4, when the maximum hourly O 3 and wind speed reached as high as 90 ppb and 7 m s -1 , respectively.The O 3 concentration varied positively with wind speed before June 6 with the maximum values of both variables occurring almost simultaneously.This indicated that the regional transport of air pollutants had significantly contributed to the O 3 rise at the study site.Based on the back trajectory analyses, the airmasses to ACPD on June 3 and June 4 (14:00 CST) were found to pass through Shandong Peninsula, Jiangsu Province and Shanghai Municipality (Fig. S4), where large emissions of O 3 precursors (Li et al., 2016a;Wu et al., 2016) had probably contributed to the high levels of O 3 observed in the study area.As the airmasses were mainly in Jiangsu Province within 24 hours before arriving at the study site on June 3 and June 4 (Fig. S4), the O 3 levels in 13 cities of Jiangsu as well as Shanghai (Fig. S2) from June 2 to June 3 were analyzed on the regional basis (Fig. S5).The regional average O 3 levels varied from 23.5 to 55.1 ppb during the period of June 2-3, with the maximum hourly average O 3 concentration of 80.1 ppb being observed in the city of Xuzhou (XZ).This indicated that the direct pollutant transport from upwind cities in Jiangsu Province could contribute a considerable amount of O 3 to the study area.However, it is noteworthy that the O 3 levels at the study site were generally higher than those of the 14 upwind cities.This might be accounted for by the following two reasons.Firstly, most of the state controlled stations used for the O 3 analyses of the 14 cities are located in or near urban areas (http://datacenter.mep.gov.cn/),where the O 3 levels were usually lower than those in non-urban areas due to the titration of NO from traffic emissions (Londhe et al., 2008).Therefore, the average O 3 levels in the whole region of Jiangsu should be higher than the values calculated in this study, and closer to the levels at the study site.Secondly, offshore regions are important areas for O 3 formation due to the large supply of precursors from mainland cities (Zhang et al., 2002) and suitable photochemical conditions (e.g., high levels of chlorine radical) (Knipping and Dabdub, 2003).More O 3 might be produced in the lower marine atmosphere before the airmasses arriving at Zhoushan Islands, and finally resulted in the higher levels of O 3 at the study site.More studies based on long-term monitoring in the rural and marine areas along the transport pathways are still needed to confirm these assumptions in our future work.
Compared to the high levels of O 3 on June 3 and June 4, much lower values were observed on June 7 and June 8, when the relatively strong southeast winds (4.4 m s -1 on diurnal average) could significantly affect local air quality of Zhoushan (Fig. 11).As the airmasses arriving at Zhoushan on June 7 and June 8 mainly came from East China Sea (Fig. S4), the obvious decrease in O 3 levels during this episode was probably caused by the diluting effect of the clean and humid airmasses from the marine area.Based on the comparative analyses of O 3 levels during the period of June 3-8, the major potential region contributing to surface O 3 over Zhoushan was found to lie to the northwest of the study area, while the southeast marine area was probably a clean region with lower levels of O 3 in the atmosphere.This further confirmed the inferences on the potential source regions of O 3 in Zhoushan made in previous sections.

Case Study 2
During the period from May 9 to May 10, 2016, significant increases in O 3 levels and wind speeds occurred along with obvious changes in wind directions from east to northwest (Fig. 12).In this episode, the study site was under the influence of cyclone systems (Fig. S6), which are usually associated with cloudy and rainy weather conditions.The air temperature was relatively low, especially on May 10 when the cold and dry airmasses from northwest prevailed over the site (Fig. 12).It began to rain at late night of May 9 and lasted to the morning of May 10.These weather conditions indicated weak photochemical activities, which are not suitable for local O 3 formation.Given the relatively high levels of daily O 3 occurred under high wind speeds on both May 9 and May 10, surface O 3 at the study site was regarded to be largely contributed from external transport.
In the afternoon (14:00) of May 9, the airmass arriving at ACPD mainly passed through the coastal provinces of Southeast China (Fig. S7).The relatively low O 3 concentrations (avg.: 35.0 ppb) observed in the daytime of May 9 indicated the upwind provinces (e.g., Fujian and Zhejiang) in Southeast China contributed less O 3 to the study area.This can be confirmed by the lower O 3 levels in Fujian Province (6.9-33.2ppb) and southeast Zhejiang Province (7.7-30.2ppb) within 24 hours before the airmass arriving at the study site (Figs.S8(a) and 8(b)).Compared with the diurnal average O 3 concentration on May 9 (35.0 ppb), the value was obviously higher on May 10 (51.8 ppb), when the airmass came from northwest to the study area (i.e., Jiangsu Province and Shanghai Municipality) (Fig. S7).The average O 3 concentrations in this region varied from 13.8 to 57.2 ppb, with the peak values over 60 ppb in the southern part of Jiangsu Province (e.g., WX and CZ) (Fig. S8(c)).This indicated significant contribution of O 3 from Jiangsu to the study area.Based on the analyses of this episode from May 9 to May 10, 2016, the major potential areas contributing to the O 3 at ACPD were found to lie in the region northwest to Zhoushan (e.g., Jiangsu), Fig. 12. Plots of anomalies from the period averages for the hourly O 3 concentration, wind direction, wind speed, air temperature, relative humidity, and the hourly variation of rainfall at ACPD from May 9 to May 10, 2016.
while the southeast provinces (e.g., Fujian) were more likely to be a clean source region with low O 3 concentrations.

CONCLUSIONS
With surface concentrations frequently exceeding the NAAQS of China, the O 3 pollution is serious in the city of Zhoushan, posing high potential health risks to locals.Opposite trends of monthly variations were observed for the O 3 and air temperature, with obviously decreased O 3 occurring in summer, probably due to the influence of intense rainfall, high relative humidity and clean marine airmasses during this season.The transport of external air pollutants has played a dominant role in the rise of surface O 3 at this coastal site in eastern China, which is comprehensively indicated by the high O 3 concentrations (> 50 ppb) under high wind speeds (> 5 m s -1 ) from the north and northwest, the time lag for the occurrence of the maximum daily O 3 concentrations, and the high incidence (64.3%) of O 3 pollution under inflow conditions.
Although the coastal regions of China have been developed, the contributions of pollutants from these areas to the surface O 3 in Zhoushan clearly varied.Based on the source identification, the major regions contributing to the surface O 3 at ACPD were found to lie to the north and northwest of the study area.Long-range transported air pollutants from the coastal provinces of the North China Plain (e.g., Shandong and Jiangsu) as well as South Korea may have significantly increased the surface O 3 at the study site, while the coastal provinces to the southwest of the study area (e.g., Fujian) are relatively clean and produced less O 3 output to Zhoushan.In contrast to the contribution from regional transport, the local contribution to the surface O 3 was small, with much lower O 3 concentrations being observed under low wind speeds.
Significant background transport of O 3 to Zhejiang was confirmed, with nearly two thirds of the pollution episodes at ACPD being accounted for by the airmasses originating outside Zhejiang.This is especially true for the airmasses coming from north of the study area, with high levels (> 56.0 ppb) of O 3 being observed for the northern trajectory clusters.In order to meet the NAAQS on O 3 in the whole province of Zhejiang, the influx of pollutants from the coastal region of northern China must be reduced, which will require coordinated inter-regional control on pollutant emissions in the northern upwind areas.

Fig. 1 .
Fig. 1.Map of the sampling site of Administrative Service Center of Putuo District (ACPD) in Zhoushan, China.

Fig. 2 .
Fig. 2. Time series of hourly average O 3 concentrations at ACPD over the period of June 2013-October 2016.The horizontal solid and dash lines indicate the Grade I NAAQS (75 ppb), and the 75 th percentile of the O 3 observations (51.8 ppb), respectively.
TN days : the total number of days for the O 3 seasons (April-October) of each year; b N days : the number of days of the O 3 observations; c N hours : the number of hours of the O 3 observations; d N days > 75ppb: the number of days with the maximum hourly O 3 observations exceeding Grade I NAAQS (75 ppb); e N hours > 75ppb: the number of hourly O 3 observations exceeding Grade I NAAQS (75 ppb).

Fig. 3 .
Fig. 3. Variations in monthly average O 3 concentrations (solid circles) and daily average O 3 concentrations by month (hollow circles) at ACPD over the period of June 2013-October 2016.

Fig. 4 .
Fig. 4. Variations in monthly average air temperature, relative humidity and rainfall at ACPD over the period of June 2013-October 2016.

Fig. 5 .
Fig. 5.Diurnal variations of average O 3 concentration, air temperature and relative humidity at ACPD over the period of June 2013-October 2016.

Fig. 6 .
Fig. 6.Polar plots of average O 3 concentrations for the (a) whole day, (b) daytime, and (c) nighttime over the period of June 2013-October 2016 at ACPD.

Fig. 7 .
Fig. 7. Polar plots of average O 3 concentrations during the pollution episodes for the (a) whole day, (b) daytime, and (c) nighttime over the period of June 2013-October 2016 at ACPD.

Fig. 8 .
Fig. 8. PSCF maps of the O 3 concentrations greater than 75 th percentile value (51.8 ppb) from June 2013 to October 2016 at ACPD with trajectory heights of 50, 500 and 1000 m.The dashed circle lines indicate the distances (400, 800 and 1200 km) from ACPD.

Fig. 9 .
Fig. 9. Mean backward trajectory clusters and the trajectories associated with each cluster from June 2013 to October 2016 at ACPD.

Fig. 10 .
Fig. 10.Box plots of (a) O 3 concentration, (b) wind speed, (c) air temperature, and (d) relative humidity at ACPD for the four trajectory clusters.

Fig. 11 .
Fig. 11.Plots of anomalies from the period averages for the hourly O 3 concentration, wind direction, wind speed, air temperature, and relative humidity at ACPD from June 3 to June 8, 2014.

Table 1 .
Statistics of the O 3 observations during the study period.

Table 2 .
Statistics of the O 3 levels for the four trajectory clusters.Ratio represents the percentage of O 3 level over 75 th percentile which is 51.8 ppb. a