Ozone pollution around a coastal region of South China Sea: interaction between marine and continental air

Abstract

Marine atmosphere is usually considered to be a clean environment, but this study indicates that the near-coast waters of the South China Sea (SCS) suffer from even worse air quality than coastal cities. The analyses were based on concurrent field measurements of target air pollutants and meteorological parameters conducted at a suburban site (Tung Chung, TC) and a nearby marine site (Wan Shan, WS) from August to November 2013. The observations showed that the levels of primary air pollutants were significantly lower at WS than those at TC, while the ozone (${\mathrm{O}}_{\mathrm{3}}$) value was greater at WS. Higher${\mathrm{O}}_{\mathrm{3}}$ levels at WS were attributed to the weaker NO titration and higher ${\mathrm{O}}_{\mathrm{3}}$ production rate because of stronger oxidative capacity of the atmosphere. However, ${\mathrm{O}}_{\mathrm{3}}$ episodes were concurrently observed at both sites under certain meteorological conditions, such as tropical cyclones, continental anticyclones and sea–land breezes (SLBs). Driven by these synoptic systems and mesoscale recirculations, the interaction between continental and marine air masses profoundly changed the atmospheric composition and subsequently influenced the formation and redistribution of${\mathrm{O}}_{\mathrm{3}}$ in the coastal areas. When continental air intruded into marine atmosphere, the ${\mathrm{O}}_{\mathrm{3}}$ pollution was magnified over the SCS, and the elevated ${\mathrm{O}}_{\mathrm{3}}$ ($>$ 100 $\mathrm{ppbv}$) could overspread the sea boundary layer $\sim$ 8 times the area of Hong Kong. In some cases, the exaggerated ${\mathrm{O}}_{\mathrm{3}}$ pollution over the SCS was recirculated to the coastal inshore by sea breeze, leading to aggravated ${\mathrm{O}}_{\mathrm{3}}$ pollution in coastal cities. The findings are applicable to similar mesoscale environments around the world where the maritime atmosphere is potentially influenced by severe continental air pollution.

Introduction

Ozone (${\mathrm{O}}_{\mathrm{3}}$) plays a central role in photochemical oxidation processes in the troposphere via direct reaction, photolysis and the subsequent reactions to produce the hydroxyl radical (Monks et al., 2015; Seinfeld and Pandis, 2016). As a strong oxidant, ${\mathrm{O}}_{\mathrm{3}}$ at surface level is recognized to be a threat to human health (WHO, 2003; Bell et al., 2007) and has a detrimental impact on vegetation (Fowler et al., 2009) and infrastructure (Kumar and Imam, 2013). Tropospheric ${\mathrm{O}}_{\mathrm{3}}$ is also the third most important greenhouse gas (IPCC, 2014) and is referred to a short-lived climate pollutant (Shindell et al., 2012).

To mitigate ${\mathrm{O}}_{\mathrm{3}}$ pollution in the troposphere, tremendous efforts from both scientific and regulatory communities have been made in the past three decades (NRC, 1991; NARSTO, 2000; Monks et al., 2015). The ${\mathrm{O}}_{\mathrm{3}}$ levels started to decrease at many locations, such as Jungfraujoch in Switzerland, Zugspitze in Germany, Mace Head in Ireland and parts of California and the eastern US (Lefohn et al., 2010; Cui et al., 2011; Parrish et al., 2012; Derwent et al., 2013; Lin et al., 2017). However, increasing studies showed that surface ${\mathrm{O}}_{\mathrm{3}}$ was elevated rapidly in East Asia in the last decade (Ding et al., 2008; Xu et al., 2008; Parrish et al., 2012; Xue et al., 2014; Zhang et al., 2014; Sun et al., 2016; Lin et al., 2017; Wang et al., 2017). For example, the observational data revealed that regional ${\mathrm{O}}_{\mathrm{3}}$ concentrations increased at a rate of 0.86 $\mathrm{ppbv}{\mathrm{yr}}^{-\mathrm{1}}$ in Pearl River Delta (PRD) from 2006 to 2011 (Li et al., 2014), at a rate of 0.56 $\mathrm{ppbv}{\mathrm{yr}}^{-\mathrm{1}}$ in Hong Kong from 2005 to 2014 (Wang et al., 2017), and even at a rate of 1.7–2.1 $\mathrm{ppbv}{\mathrm{yr}}^{-\mathrm{1}}$ (summertime only) at Mount Tai in central eastern China (Sun et al., 2016).

Hong Kong and the adjacent PRD are the most industrialized regions along the coast of South China Sea (SCS), and they suffer from serious ${\mathrm{O}}_{\mathrm{3}}$ pollution (Zheng et al., 2010; Guo et al., 2013; Ling et al., 2013). Numerous studies demonstrated that in addition to long-range transport (Chan, 2002; Guo et al., 2009; Wang et al., 2009) and local photochemical production (Ding et al., 2013b), tropical cyclones and mesoscale circulations are conducive to the occurrence of high-${\mathrm{O}}_{\mathrm{3}}$ events (Yin, 2004; Huang et al., 2005; Yang et al., 2012; Jiang et al., 2015; Wei et al., 2016). In a number of studies, the tropical cyclone has been considered as the most conducive driver to the occurrence of ${\mathrm{O}}_{\mathrm{3}}$ episodes in Hong Kong (Yin, 2004; Ling et al., 2013) because it generally causes peripheral subsidence, stagnation air and inversion layers, which favor the production and accumulation of ${\mathrm{O}}_{\mathrm{3}}$.

Mesoscale circulations (e.g., sea–land breezes (SLBs) and mountain–valley breezes) also play important roles in ${\mathrm{O}}_{\mathrm{3}}$ distribution and transport in the coastal cities like Hong Kong with complex topography and land cover (Liu and Chan, 2002; Ding et al., 2004; Lu et al., 2010; Guo et al., 2013). For instance, Guo et al. (2013) demonstrated that upslope winds brought pollutants including ${\mathrm{O}}_{\mathrm{3}}$ from low-lying areas to the peak of Mt. Tai Mo Shan (957 $\mathrm{m}$ a.s.l.) in Hong Kong. Ding et al. (2004) simulated a multi-day SLB-related ${\mathrm{O}}_{\mathrm{3}}$ episode and discussed the influence of SLB circulation on the transport of oxidant precursors, the residence time and re-entry of photochemical compounds. Lu et al. (2010) simulated the SLB in the 2003/2004 winter and revealed that the urbanization of Shenzhen might significantly enhance the sea breeze to the west of Hong Kong in the early afternoon, which worsened the local air pollution.

Both coastal human activities and marine atmospheric cyclic behavior can significantly affect the air pollution level in coastal urban environments (Adame et al., 2010; Velchev et al., 2011). Exploring SLBs provides an important way to understand the interaction between continental air and marine atmosphere, which has long been a focus of coastal air quality, global tropospheric chemistry and climate change research. Surprisingly, few studies investigated SLBs in Hong Kong although about 70 SLB days per year on average were observed in Hong Kong and the PRD region (Zhang and Zhang, 1997). Therefore, the association between mesoscale recirculation and air pollutants over the SCS and subtropical continental region is still not well established, which seriously limits our understanding of the interplay of continental and marine air masses in this region. Furthermore, previous ${\mathrm{O}}_{\mathrm{3}}$ studies carried out in this region neither paid enough attention to the variations of volatile organic compounds (VOCs, one important group of ${\mathrm{O}}_{\mathrm{3}}$ precursors) nor established any field measurements on an island, an ideal site for observation of marine air mass with less interference from local emissions, for understanding the ${\mathrm{O}}_{\mathrm{3}}$ pollution around the coastal region of the SCS (Parrish et al., 1998). So far, only a handful of studies intensely evaluated the chemical characteristics of air masses under various synoptic systems (Wang et al., 2005; Guo et al., 2009, 2013; Li et al., 2018).

This study aims to comprehensively characterize interaction between continental anthropogenic emissions and marine atmosphere over a coastal region of the SCS by concurrent measurements and in-depth analysis of air pollutants at a marine site over the SCS and a suburban site in Hong Kong. Firstly, the spatial and temporal variations of measurements are described to give an overall picture of the campaign as well as to directly evaluate how continental outflows polluted the marine atmosphere over the SCS. After that, the chemical and meteorological characteristics of air masses associated with high ${\mathrm{O}}_{\mathrm{3}}$ concentrations are explored. Finally, the interplay between the maritime and continental air masses and its influence on regional air quality are discussed.

Methodology

Sampling sites

Field measurements were carried out concurrently at a suburban site and a marine site over SCS (Fig. 1). The suburban Tung Chung (TC; 22.29${}^{\circ }$ N, 113.94${}^{\circ }$ E) site, part of the Hong Kong Environmental Protection Department (HKEPD) air quality monitoring network, is located in southwestern Hong Kong, about 3 $\mathrm{km}$ south of the Hong Kong International Airport at Chek Lap Kok with Hong Kong's urban center about 20 $\mathrm{km}$ to the southwest and Macau 38 $\mathrm{km}$ to the northeast. It is a newly developed residential town adjacent to the busy highway and railway lines. The sampling instruments were installed on the rooftop of a building with a height of 27.5 $\mathrm{m}$ a.s.l. A more detailed description of the TC site can be found in our previous publications (Cheng et al., 2010b; Jiang et al., 2010).

The marine site, Wan Shan island (WS; 21.93${}^{\circ }$ N, 113.73${}^{\circ }$ E), is located 40 $\mathrm{km}$ southeast of Zhuhai and is bounded to the north by the Pearl River Estuary, with a straight distance of about 44 $\mathrm{km}$ to TC. WS has an area of 8.1 ${\mathrm{km}}^{\mathrm{2}}$ and a population of about 3000 with sparse anthropogenic emissions at the island. The isolated island features a subtropical maritime climate. The measurement site was set up on the rooftop of the National Marine Environmental Monitoring Station with a height of about 65 $\mathrm{m}$ a.s.l.

High-${\mathrm{O}}_{\mathrm{3}}$ mixing ratios are frequently observed in Hong Kong in late summer and autumn (Ling et al., 2013) when tropical cyclones and the northeast monsoon prevail, respectively. During this period, WS is right in the downwind direction of TC, which facilitates the study of the interaction between the inland pollution and the marine environment.

Locations of the sampling sites (red stars) and the surrounding environment.

[Figure omitted. See PDF]

Measurement techniques

Measurements of trace gases and meteorological parameters

The sampling campaign was conducted from 10 August to 21 November across late summer and autumn in 2013. At WS, trace gases (i.e., NO${}_x$, ${\mathrm{O}}_{\mathrm{3}}$, ${\mathrm{SO}}_{\mathrm{2}}$ and $\mathrm{CO}$) were continuously monitored with a time resolution of 1 $\min$. $\mathrm{NO}$-${\mathrm{NO}}_{\mathrm{2}}$-NO${}_x$ was measured using a heated molybdenum ${\mathrm{NO}}_{\mathrm{2}}$-to-$\mathrm{NO}$ converter and a chemiluminescence analyzer (Thermo Environmental Instruments (TEI), model 42i) with a range of 0–200 $\mathrm{ppbv}$ and a lower detection limit of 0.40 $\mathrm{ppbv}$. It was noteworthy that the measured NO${}_x$ might include other oxidized reactive nitrogen that was converted by the molybdenum. Thus, the NO${}_x$ concentrations given below were considered the upper limits of their actual values (Dunlea et al., 2007; Ran et al., 2011). ${\mathrm{O}}_{\mathrm{3}}$ was monitored with a commercial UV photometric analyzer (TEI, model 49i) with a range of 0–0.050 to 200 $\mathrm{ppm}$ and a lower detection limit of 1.0 $\mathrm{ppbv}$. ${\mathrm{SO}}_{\mathrm{2}}$ was measured using a pulsed UV fluorescence approach (TEI, model 43S). $\mathrm{CO}$ was measured by a non-dispersive infrared analyzer based on gas filter correlation technology (API, model 300). Quality assurance and control procedures (e.g., instrumental maintenance and calibration) for these devices have been described elsewhere (Guo et al., 2009, 2013). Meteorological parameters, including temperature, relative humidity, solar radiation, wind speed and wind direction, were routinely monitored by a weather station (Vantage Pro 2 plus, Davis Instruments) with a time resolution of 5 $\min$. At TC, hourly data of the aforementioned trace gases and meteorological parameters were obtained from the HKEPD (http://epic.epd.gov.hk/ca/uid/airdata). Detailed information of the quality assurance and control protocols is available in the HKEPD report (HKEPD, 2015).

Sampling and analysis of VOCs

Concurrent VOC samples (i.e., non-methane hydrocarbons (NMHCs) and carbonyls) were collected on 21 selected days (including both non-${\mathrm{O}}_{\mathrm{3}}$ episodes and ${\mathrm{O}}_{\mathrm{3}}$ episodes) at both sites. These days were selected on the basis of weather prediction and meteorological data analysis for potentially high- and low-${\mathrm{O}}_{\mathrm{3}}$ days. An ${\mathrm{O}}_{\mathrm{3}}$ episode day is the day when the peak 1 $\mathrm{h}$ averaged ${\mathrm{O}}_{\mathrm{3}}$ mixing ratio exceeds 100 $\mathrm{ppbv}$ (Level II of China National Ambient Air Quality Standard). Please refer to our previous publication for details of this method (Guo et al., 2009).

The whole-air samples of NMHCs were collected using 2 $\mathrm{L}$ electro-polished stainless steel canisters. The canisters were cleaned, conditioned and evacuated before being used for sampling. A metal bellows pump was used to fill up the canisters with sample air over 1-hour integration (with a flow restrictor) to a pressure of 40 $\mathrm{psi}$. Seven 1-hour VOC samples (every 2 h during 07:00–19:00 inclusive) were collected simultaneously at each site. Intensive VOC sampling was also carried out at WS on 7 selected days (i.e., 3, 4, 9 and 22–25 October) with 11 1-hour samples (every 2 h during 01:00–22:00 inclusive). Totally, 311 valid VOC samples (144 at TC and 167 at WS) were collected in addition to about 5 $\mathrm{\%}$ field blanks and 5 $\mathrm{\%}$ parallel samples for quality assurance purpose. The speciation and abundance of 59 C${}_{\mathrm{2}}$–C${}_{\mathrm{11}}$ NMHCs in the canisters were determined by a preconcentrator (model 7100, Entech Instruments Inc., California, USA) coupled with an Agilent 5973N gas chromatography–mass selective detector–flame ionization detector (GC-MSD-FID, Agilent Technologies, USA). The detection limit of NMHCs was 3 $\mathrm{pptv}$ with a measurement precision of 2–5 $\mathrm{\%}$ and a measurement accuracy of 5 $\mathrm{\%}$. Detailed information of the analysis system and quality control and quality assurance for VOC samples can be found elsewhere (Simpson et al., 2010).

Carbonyl samples were collected using silica-gel-filled cartridges impregnated with acidified 2,4-dinitrophenylhydrazine (DNPH). Air samples were drawn through the cartridge at a flow rate of 0.4–0.6 $\mathrm{L}{\min }^{-\mathrm{1}}$ for 2 $\mathrm{h}$; the flow rate through the cartridges was monitored with a rotameter which was calibrated before and after each sampling. An ${\mathrm{O}}_{\mathrm{3}}$ scrubber was connected to the inlet of the DNPH–silica gel cartridge to prevent interference from ${\mathrm{O}}_{\mathrm{3}}$. In total, 227 carbonyl samples (124 at TC and 103 at WS) were collected with 5 and 6 samples per non-${\mathrm{O}}_{\mathrm{3}}$ and ${\mathrm{O}}_{\mathrm{3}}$ episode day (every 2 $\mathrm{h}$ during 07:00–18:00 inclusive), respectively. All cartridges were stored in a refrigerator at 4 ${}^{\circ }\mathrm{C}$ after sampling. The sampled carbonyl cartridges were eluted slowly with $<$ 5 $\mathrm{mL}$ of acetonitrile in the direction opposite to sampling flow into a 5 $\mathrm{mL}$ brown volumetric flask, followed by adding acetonitrile to a constant volume of 5 $\mathrm{mL}$. A 20 $\mathrm{\mu }\mathrm{L}$ aliquot was injected into the high performance liquid chromatography (HPLC) system through an autosampler. The operating conditions of the HPLC are shown in Table S1 in the Supplement. Typically, C${}_{\mathrm{1}}$–C${}_{\mathrm{9}}$ carbonyl compounds were measured efficiently with a detection limit of $\sim$ 0.2 $\mathrm{ppbv}$.

Observation-based model

A photochemical box model coupled with the Master Chemical Mechanism v3.2 (PBM-MCM) was applied to simulate the ${\mathrm{O}}_{\mathrm{3}}$ production at WS and TC for the VOC sampling days. The PBM-MCM model is a zero-dimensional photochemical box model combined with a near-explicit chemical mechanism consisting of 5900 species and 16 500 reactions, which fully describes the mechanisms of homogeneous reactions in the atmosphere (Jenkin et al., 1997, 2003; Saunders et al., 2003). The simulation was constrained by hourly data of meteorological parameters (i.e., temperature and relative humidity) and air pollutants ($\mathrm{NO}$, ${\mathrm{NO}}_{\mathrm{2}}$, $\mathrm{CO}$, ${\mathrm{SO}}_{\mathrm{2}}$ and 51 measured VOCs). Since the sampling interval was 2 $\mathrm{h}$ for each sample, cubic spline interpolation was used to derive VOC concentrations at each hour for modeling purpose. Please see our previous publication for details (Wang et al., 2017). $\mathrm{HONO}$ has been recognized as an important source of $\mathrm{OH}$, influencing ${\mathrm{O}}_{\mathrm{3}}$ formation significantly (Kleffmann, 2007). Since we did not measure $\mathrm{HONO}$ mixing ratios in this study, the average diurnal profiles of $\mathrm{HONO}$ observed at TC in autumn 2011 (Xu et al., 2015) and at a coastal background site (Hok Tsui, HT) in southeast Hong Kong in autumn 2012 (Zha, 2015) were applied to the photochemical simulations at TC and WS, respectively. Figure S1 in the Supplement shows the average diurnal cycles of $\mathrm{HONO}$ at TC and HT. The use of the aforementioned diurnal profiles might increase the uncertainty of model simulation. However, we believe that the newly introduced uncertainties could not be too high, because $\mathrm{HONO}$ observations at TC and HT were carried out 2 years and 1 year before the sampling campaign of this study, respectively. In addition, HT was comparable to WS in aspects of local emissions (nearly free of anthropogenic emissions), air mass category (mixed continental and marine air) and location (to the south of Hong Kong and on the SCS). It is noteworthy that the atmospheric physical processes (i.e., vertical and horizontal transport) were not considered in this model. In addition, the inherent uncertainty of NO${}_x$ measurement mentioned above might slightly affect the modeling results. The PBM-MCM model has been successfully applied in previous studies (Cheng et al., 2010a; Lam et al., 2013; Ling et al., 2014). Details of the model construction can be found in Saunders et al. (2003) and Lam et al. (2013).

WRF-CMAQ simulation and backward particle release model

In this study, the Weather Research and Forecasting (WRF v3.7.1) model (Skamarock et al., 2008) was used to simulate vertical and horizontal wind fields for various weather systems observed in this campaign and then provided meteorological parameters required by US EPA Community Multiscale Air Quality (CMAQ v4.7.1) model (www.epa.gov/cmaq). CMAQ is a three-dimensional Eulerian atmospheric chemistry and transport modeling system, which includes complex physical and chemical processes, such as physical transport and diffuse, gas and aqueous chemical transformation, and so on; it can treat multiple pollutants simultaneously from local to continental scales. A domain system composed of four nested grids (81, 27, 9, 3 $\mathrm{km}$) was adopted to better suit the simulation of mesoscale weather systems, as shown in Fig. S2. The domain with finest resolution (3 $\mathrm{km}$) covers the Pearl River Estuary region. Vertically, there were 31 sigma levels for all domains, with the model top fixed at 50 $\mathrm{hPa}$. The major selected physical schemes invoked in WRF and chemical mechanisms used in CMAQ are shown in Table S2. The input meteorological data were made using NCEP FNL (final) data with a horizontal resolution of 1${}^{\circ }$ $\times$ 1${}^{\circ }$ (https://rda.ucar.edu/). In addition, the geographical data were obtained from the Research Data Archive of National Center for Atmospheric Research (NCAR) (http://www2.mmm.ucar.edu/wrf/users/downloads.html). The emission inventories used in this study included the 2000-based Regional Emission Inventory in ASia (REAS) (Kurokawa et al., 2013) and the 2010-based Multi-resolution Emission Inventory for China (MEIC) (He, 2012), both of which were processed by the Sparse Matrix Operating Kernel Emission (SMOKE) model. The biogenic emissions were calculated by the Model of Emissions of Gases and Aerosols from Nature (MEGAN) (Guenther, 2006). The WRF modeling mainly focused on ${\mathrm{O}}_{\mathrm{3}}$ episodes with an additional 24 $\mathrm{h}$ preceding run as spin-up for each episode, and the integration was conducted separately. In addition, the spatiotemporal patterns of $\mathrm{CO}$ and ${\mathrm{O}}_{\mathrm{3}}$ were simulated by WRF-CMAQ during two ${\mathrm{O}}_{\mathrm{3}}$ episodes (see Sect. 3.4). The time series of the simulations and observations of $\mathrm{CO}$ and ${\mathrm{O}}_{\mathrm{3}}$ are shown in Fig. S3. Also, Table S3 gives the index of agreements (IOAs) between the simulated and observed meteorological parameters and air pollutants. Within the range of 0–1, higher IOAs represent better agreement between the simulated and observed values (Willmott, 1982). Here, IOA was between 0.51 and 0.84 for the simulation of meteorological parameters. Furthermore, it was not lower than 0.50 for primary air pollutants and reached 0.81 for ${\mathrm{O}}_{\mathrm{3}}$ simulation at both sites. The model performances were comparable to those reported in previous studies (Cabaraban et al., 2013; Wang et al., 2015). Therefore, we accepted the modeling results in view of the fact that the simulations were only used to qualitatively indicate the interactions between the continental and marine air in this study.

Backward particle release simulations were carried out using HYSPLIT model (Stein et al., 2015) for episode days at WS and TC sites during the entire sampling period (Draxler and Rolph, 2003). The backward particle release simulation, which considers the dispersion processes in the atmosphere, is capable of identifying the history of air masses (Guo et al., 2009; Ding et al., 2013a, b). In this work, we applied the model following a method developed by Ding et al. (2013b).

Results and discussion

Spatiotemporal variations

Table 1 summarizes the meteorological conditions and chemical species observed at WS and TC. Lower temperature (25.7 $\pm$ 0.1 ${}^{\circ }\mathrm{C}$) and higher relative humidity (82.8 $\pm$ 0.4 $\mathrm{\%}$) were recorded at the marine site (WS) compared to the suburban site (TC) ($p<\mathrm{0.01}$) (temperature: 26.7 $\pm$ 0.1 ${}^{\circ }\mathrm{C}$; relative humidity: 67.7 $\pm$ 0.5 $\mathrm{\%}$). At WS, the solar radiation (635.8 $\pm$ 46.9 $\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$) was much higher than that at TC (563.5 $\pm$ 46.1 $\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$, $p<\mathrm{0.01}$), while the average wind speed at TC (4.6 $\pm$ 0.1 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$) was significantly lower than that measured at WS (7.2 $\pm$ 0.2 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$). The lower wind speed at TC was related to the roughness of underlying surfaces. However, no statistical differences were found for the average wind direction (about 81${}^{\circ }$, northeast wind) at the two sites, indicating that the two sites were probably under the influence of similar air masses in most cases.

The $\mathrm{NO}$, ${\mathrm{NO}}_{\mathrm{2}}$, $\mathrm{CO}$, ${\mathrm{SO}}_{\mathrm{2}}$ and total VOCs (the sum of NMHCs and carbonyls) had lower average and maximum mixing ratios at WS than those at TC. The lower levels of primary air pollutants at WS were likely the results of fewer local emission sources, faster photochemical consumption (as discussed later) and/or more favorable dispersion conditions (e.g., higher wind speed). In contrast, ${\mathrm{O}}_{\mathrm{3}}$ was much higher at WS (Table 1), attributable to the enhancements by both meteorological and photochemical effects, as discussed in Sects. 3.2 and 3.3.

Descriptive statistics of meteorological parameters and trace gases at the two sites during the sampling period.

Parameter	WS			TC
	Mean $\pm$ 95 $\mathrm{\%}$ C.I.	Max.		Mean $\pm$ 95 $\mathrm{\%}$ C.I.	Max.
Temperature (${}^{\circ }\mathrm{C}$)	25.7 $\pm$ 0.1	32.8		26.7 $\pm$ 0.1	35.4
Relative humidity ($\mathrm{\%}$)	82.8 $\pm$ 0.4	98.9		67.7 $\pm$ 0.5	96.8
Solar radiation ($\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$)${}^{\mathrm{a}}$	635.8 $\pm$ 46.9	1026.8		563.5 $\pm$ 46.1	910.0
Wind speed ($\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$)	7.2 $\pm$ 0.2	23.8		4.6 $\pm$ 0.1	13.8
Wind direction (${}^{\circ }$)	81.3	–		80.9	–
${\mathrm{O}}_{\mathrm{3}}$ (ppbv)	51.3 $\pm$ 1.2	173.0		30.0 $\pm$ 1.0	159.9
$\mathrm{NO}$ (ppbv)	0.7 $\pm$ 0.1	21.0		14.0 $\pm$ 0.8	115.7
${\mathrm{NO}}_{\mathrm{2}}$ (ppbv)	4.3 $\pm$ 0.3	49.3		25.0 $\pm$ 0.6	104.2
$\mathrm{CO}$ (ppbv)	251.4 $\pm$ 6.5	727.7		560.5 $\pm$ 6.3	1047.9
${\mathrm{SO}}_{\mathrm{2}}$ (ppbv)	2.4 $\pm$ 0.1	12.2		5.9 $\pm$ 0.1	19.1
NMHCs (ppbv)	12.7 $\pm$ 1.1	32.9		17.7 $\pm$ 1.7	60.0
Carbonyls (ppbv)	7.9 $\pm$ 0.7	16.3		9.2 $\pm$ 0.7	26.5

${}^{\mathrm{a}}$ Average of the daily maximum solar radiation. C.I. denotes confidence interval.

Time series of local meteorological parameters and hourly mixing ratios of air pollutants at the two sites are illustrated in Fig. 2a–b. The temporal patterns of wind directions were generally similar at both sites, with the dominance of the southerly winds in August and northeastern winds between September and November. Occasionally, the northwesterly winds from the PRD region were observed.

This sampling campaign witnessed 17 ${\mathrm{O}}_{\mathrm{3}}$ episodes and 7 near-${\mathrm{O}}_{\mathrm{3}}$ episode days at TC, which refers to the days with maximum hourly mixing ratio of ${\mathrm{O}}_{\mathrm{3}}$ higher than 100 $\mathrm{ppbv}$ and within the range of 80–100 $\mathrm{ppbv}$, respectively (80 $\mathrm{ppbv}$ was Level I of China National Ambient Air Quality Standard for ${\mathrm{O}}_{\mathrm{3}}$). At WS, 21 ${\mathrm{O}}_{\mathrm{3}}$ episodes and 6 near-${\mathrm{O}}_{\mathrm{3}}$ episodes were recorded. Specifically, 13 ${\mathrm{O}}_{\mathrm{3}}$ episode days were simultaneously observed at the two sites, with the rest occurred exclusively at one site. On one hand, the primary air pollutants ($\mathrm{CO}$, ${\mathrm{SO}}_{\mathrm{2}}$ and NO${}_x$) generally increased during ${\mathrm{O}}_{\mathrm{3}}$ episodes, implying enhanced ${\mathrm{O}}_{\mathrm{3}}$ formation potentials from the precursors. On the other hand, ${\mathrm{O}}_{\mathrm{3}}$ episodes were always accompanied by the synoptic conditions, i.e., tropical cyclone (typhoon in the mature form) and continental anticyclone, and/or mesoscale circulations such as SLB, as detailed in Table S4. For example, the two multi-day ${\mathrm{O}}_{\mathrm{3}}$ episode events, i.e., 1–8 October and 19–27 October (highlighted in blue in Fig. 2), were strongly associated with continental high pressure. These episode days generally had high temperature, northerly winds and intensive solar radiation, with air flows largely from the inland or the coastal areas. Also, the mixing ratios of $\mathrm{CO}$, ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{SO}}_{\mathrm{2}}$ usually increased during these days, suggesting the accumulation of local air pollutants and/or the increasing contribution from regional transport. In contrast, ${\mathrm{O}}_{\mathrm{3}}$ episodes under the influence of tropical cyclones (highlighted in orange in Fig. 2) featured high temperature, strong solar radiation and typically calm or moderate northwesterly to northeasterly winds, except for typhoon Haiyan, which occurred on 9–12 November (discussed in Sect. 3.2.1). These conditions were all conducive to the formation and accumulation of ${\mathrm{O}}_{\mathrm{3}}$. Additionally, SLB was also an important factor regulating ${\mathrm{O}}_{\mathrm{3}}$ pollution in this region during ${\mathrm{O}}_{\mathrm{3}}$ episodes (Table S4). Detailed discussions can be found in Sect. 3.2.3.

Time series of trace gases and meteorological parameters observed for the sampling period of 10 August–30 September (a) and 1 October–21 November (b) at WS (green) and TC (red). The black line of 100 $\mathrm{ppbv}$ is the threshold for ${\mathrm{O}}_{\mathrm{3}}$ episode definition. The dates seriously affected by continental high pressure and tropical cyclones are shaded in blue and orange, respectively. Note that there are some data missing in these months due to extremely bad weather conditions and instrumental failure.

[Figure omitted. See PDF]

Meteorological influence on ${\mathrm{O}}_{\mathrm{3}}$ mixing ratios

Descriptive statistics of meteorological parameters during ${\mathrm{O}}_{\mathrm{3}}$ episode and non-episode days are summarized in Table 2. On episode days the wind speed and relative humidity were lower whereas solar radiation was stronger at both sites, suggesting that this type of weather condition was conducive to the formation and accumulation of tropospheric ${\mathrm{O}}_{\mathrm{3}}$. Furthermore, the wind direction during non-episode days was predominantly from the east (SCS), while during episode days the winds mainly came from the north and northeast, which might bring more pollutants from the urban areas of Hong Kong and inland PRD to the sampling sites. The characteristics of ${\mathrm{O}}_{\mathrm{3}}$ pollution under different weather conditions are discussed below.

Descriptive statistics (mean $\pm$ 95 $\mathrm{\%}$ C.I.) of meteorological parameters at the two sites during ${\mathrm{O}}_{\mathrm{3}}$ episodes and non-${\mathrm{O}}_{\mathrm{3}}$ episodes days.

Parameter	WS			TC
	${\mathrm{O}}_{\mathrm{3}}$ episode	Non-${\mathrm{O}}_{\mathrm{3}}$ episode		${\mathrm{O}}_{\mathrm{3}}$ episode	Non-${\mathrm{O}}_{\mathrm{3}}$ episode
Temperature (${}^{\circ }\mathrm{C}$)	25.3 $\pm$ 0.2	25.8 $\pm$ 0.1		26.3 $\pm$ 0.3	26.8 $\pm$ 0.2
Wind speed ($\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$)	5.3 $\pm$ 0.2	7.7 $\pm$ 0.2		3.7 $\pm$ 0.2	4.8 $\pm$ 0.1
Wind direction (${}^{\circ }$)	45.1	89.1		19.5	86.8
Relative humidity ($\mathrm{\%}$)	71.7 $\pm$ 1.2	85.7 $\pm$ 0.4		58.4 $\pm$ 1.4	69.6 $\pm$ 0.6
Solar radiation ($\mathrm{W}{\mathrm{m}}^{-\mathrm{2}}$)${}^{\mathrm{a}}$	723.2 $\pm$ 26.1	613.7 $\pm$ 57.6		699.0 $\pm$ 29.1	537.0 $\pm$ 53.1

${}^{\mathrm{a}}$ Average of the daily maximum solar radiation. C.I. denotes confidence interval.

Tropical cyclones

Tropical cyclone (low-pressure system) is one of the main meteorological conditions conducive to the occurrence of ${\mathrm{O}}_{\mathrm{3}}$ episodes in Hong Kong (Yin, 2004; Ling et al., 2013). In this study, 7 episode days and 3 near-episode days were closely associated with five tropical cyclones (i.e., Trami, Usagi, Wutip, Nari and Krosa) (Table S4 and Fig. S4). For example, Trami caused the worst ${\mathrm{O}}_{\mathrm{3}}$ episode on 21 August with the highest peak hourly ${\mathrm{O}}_{\mathrm{3}}$ mixing ratios of 160 and 173 $\mathrm{ppbv}$ at TC and WS, respectively. These episode or near-episode days usually appeared 1–2 days before the arrival of the tropical cyclones, because the large-scale peripheral subsidence of the tropical cyclones usually creates the meteorological conditions favorable to the formation and accumulation of ${\mathrm{O}}_{\mathrm{3}}$, such as inversion layer, high temperature, low humidity, intensive light and weak winds (Wang et al., 1998; Yin, 2004). The tropical cyclones also cause anticlockwise air flows at their outskirts, affecting the wind directions and subsequent the regional transport of air pollution. Figure 3 illustrates surface wind fields and air movement 2 days (i.e., 20–21 September) before the occurrence of Usagi as an example. It can be seen that when Usagi approached southeastern area of Hong Kong, it led to weak northeasterly and later northwesterly winds which potentially delivered ${\mathrm{O}}_{\mathrm{3}}$ and its precursors from highly polluted inland PRD region to the sampling sites (Yin, 2004; Wei et al., 2016; Wang et al., 2017). The wind speed was lower than 4 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$ at the sampling sites and in their surrounding area on 20 September (Fig. 3a), and it gradually increased the next day (21 September) with the approach of the tropical cyclone (Fig. 3b). It is noteworthy that the rarely occurring westerly and northwesterly winds caused tropical cyclones that resulted in unsynchronized occurrence of ${\mathrm{O}}_{\mathrm{3}}$ episodes between the two sites (Fig. 3c, d). Namely, high ${\mathrm{O}}_{\mathrm{3}}$ values were observed at TC only on 20 September, while ${\mathrm{O}}_{\mathrm{3}}$ started to increase at WS on the next day (21 September). This discrepancy might indicate the transport of ${\mathrm{O}}_{\mathrm{3}}$ and/or its precursors from terrestrial area to the offshore site driven by tropical cyclone.

Please note that not all tropical cyclones would cause high levels of ${\mathrm{O}}_{\mathrm{3}}$. For example, the tropical cyclone Haiyan observed on 9–12 November over the SCS did not cause high ${\mathrm{O}}_{\mathrm{3}}$ levels (Fig. 2b). Because the origin of Haiyan was at a lower latitude (southern Guam) and it moved on the waters southwest of PRD (Fig. S4), the anticlockwise air flow caused easterly and southeasterly winds in the north and northeast outer band of Haiyan. The winds originated from the SCS brought in clean marine air to the sampling sites, resulting in dilution and dispersion of local air pollutants.

Model-simulated 10 $\mathrm{m}$ wind vectors (arrows) and wind speed (shaded, unit: $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$), as well as the distribution of air mass concentrations (unit: $\mathrm{mass}{\mathrm{h}}^{-\mathrm{1}}{\mathrm{m}}^{-\mathrm{3}}$) at the 100 $\mathrm{m}$ height level simulated by HYSPLIT Lagrangian backward particle release model with WS and TC as the starting points 2 days (20 September 2013) before (a, c) and 1 day (21 September 2013) before (b, d) the arrival of Usagi.

[Figure omitted. See PDF]

Continental anticyclones

In addition to tropical cyclones, the continental anticyclone (high-pressure system) was frequently observed in the region, which often caused high ${\mathrm{O}}_{\mathrm{3}}$ concentrations. For example, two multi-day ${\mathrm{O}}_{\mathrm{3}}$ episodes (1–8 and 19–27 October) occurred at the sampling sites when there were intensive continental anticyclones and weak Western Pacific Subtropical High (WPSH) to the north of Hong Kong (see Fig. S5 as examples). The SLBs also occurred occasionally during the first several days of the two continental high-pressure systems when the synoptic winds were relatively weak (i.e., 2–5 and 19–21 October). Under the continental anticyclones, the northeasterly and easterly winds brought in densely polluted air from the inland (Fig. 2b) to the coastal areas of the SCS. For example, the $\mathrm{CO}$ mixing ratios were significantly elevated during these episode days, with an average of 409 and 683 $\mathrm{ppbv}$ at WS and TC, respectively, which were higher than other episode days. The continuous input of exotic air pollutants provided essential “fuel” to local photochemical production of ${\mathrm{O}}_{\mathrm{3}}$, leading to the severe multi-day ${\mathrm{O}}_{\mathrm{3}}$ episodes.

SLB circulation on 3–4 October 2013, showing surface wind pattern (arrows) and temperature (color) at 17:00 on 3 October (a) and at 04:00 on 4 October (b). Vertical cross section (taken over a longitude of 113.85${}^{\circ }$ E, mean of the longitudes of TC and WS) depicting the $v$–$w$ wind stream (arrow) and the index of $\mathit{\omega }\cdot \mathrm{100}$ (color) at 17:00 on 3 October (c) and at 04:00 on 4 October (d). For panels (c) and (d), the blue color (negative) and light red color (positive) present downward and upward winds, respectively. Panels (a) and (c) represent a sea breeze, and panels (b) and (d) show a land breeze. Note that $\mathit{\omega }$ is the vertical velocity in isobaric coordinates.

[Figure omitted. See PDF]

Sea–land breeze circulation

During the sampling period, SLB circulations in the study area were identified on 21 out of 104 sampling days. The occurrence frequency was comparable to that reported by Zhang and Zhang (1997), who discovered 70 SLB days in a year in the same region. In this study, 12 ${\mathrm{O}}_{\mathrm{3}}$ episode days were thought to be influenced by SLB (see Table S4), with 5 of them (27–28 September, 11–12 October and 1 November) under the dominance of tropical cyclones (i.e., Wutip, Nari and Krosa) and the other 7 days in association with the continental anticyclones. In addition to the effects of tropical cyclones and continental anticyclones discussed above, SLB also posed a non-negligible impact on ${\mathrm{O}}_{\mathrm{3}}$ pollution in these cases.

SLB circulation is driven by sea–land thermal difference and topographic conditions and usually happens when the synoptic winds are weak (Liu et al., 2002; Lo et al., 2006; Lu et al., 2009). On a typical SLB day, wind blows onshore during the day (sea breeze) and offshore in the evening (land breeze). However, the transition time of breezes in this study was found to vary in a wide range. The sea breeze switched to land breeze between 00:00 and 08:00 with a median of 03:00 for breeze shifting, and 11:00–18:00 with a median of 14:00 was the time when land breeze turned to sea breeze. Ding et al. (2004) also reported this phenomenon and pointed out that the start time of sea breezes in Hong Kong was generally delayed to noontime due to the synoptic northerly winds blowing from the continental areas to the SCS, particularly on ${\mathrm{O}}_{\mathrm{3}}$ episode days when northerly winds dominated in Hong Kong. For example, the sea breeze commenced at 15:00 on 3 October and transited to land breeze at 04:00 on 4 October (Fig. 4). Figure 4a and b depict the surface wind fields with a sea breeze and a land breeze, respectively. The vertical wind fields with the sea breeze and land breeze are presented in Fig. 4c and d, respectively. Surface and vertical SLB circulations were clearly seen in these panels of Fig. 4. The mesoscale circulations caused by SLB might promote the interactions between the continental (TC) and marine (WS) atmospheres. Specifically, the primary air pollutants observed at TC could be transported to WS by land breeze. Moreover, the air masses could return to TC after sufficient photochemical evolutions over the SCS, during which ${\mathrm{O}}_{\mathrm{3}}$ might also be elevated in the continental areas.

Descriptive statistics (mean $\pm$ 95 $\mathrm{\%}$ C.I.) of measured air pollutants and simulated $\mathrm{OH}$, ${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$ at the two sites during ${\mathrm{O}}_{\mathrm{3}}$ episodes and non-${\mathrm{O}}_{\mathrm{3}}$ episodes days.

Parameter	WS			TC
	${\mathrm{O}}_{\mathrm{3}}$ episode	Non-${\mathrm{O}}_{\mathrm{3}}$ episode		${\mathrm{O}}_{\mathrm{3}}$ episode	Non-${\mathrm{O}}_{\mathrm{3}}$ episode
${\mathrm{O}}_{\mathrm{3}}$ (ppbv)	74.3 $\pm$ 3.0	43.9 $\pm$ 1.0		44.1 $\pm$ 3.6	27.2 $\pm$ 0.8
O${}_x$ (ppbv)	81.6 $\pm$ 2.9	47.8 $\pm$ 1.0		83.3 $\pm$ 3.7	49.4 $\pm$ 1.0
$\mathrm{NO}$ (ppbv)	0.9 $\pm$ 0.2	0.5 $\pm$ 0.2		11.5 $\pm$ 1.4	14.5 $\pm$ 0.9
${\mathrm{NO}}_{\mathrm{2}}$ (ppbv)	8.5 $\pm$ 0.9	2.0 $\pm$ 0.5		39.2 $\pm$ 1.7	22.2 $\pm$ 0.6
$\mathrm{CO}$ (ppbv)	391.4 $\pm$ 9.1	209.4 $\pm$ 6.8		652.9 $\pm$ 16.0	541.9 $\pm$ 6.5
${\mathrm{SO}}_{\mathrm{2}}$ (ppbv)	4.3 $\pm$ 0.2	1.9 $\pm$ 0.1		8.1 $\pm$ 0.3	5.5 $\pm$ 0.1
NMHCs (ppbv)	17.7 $\pm$ 1.4	9.6 $\pm$ 1.2		20.2 $\pm$ 2.2	16.8 $\pm$ 2.1
Carbonyls (ppbv)	10.3 $\pm$ 0.8	5.4 $\pm$ 0.4		12.0 $\pm$ 1.3	8.1 $\pm$ 0.7
${\mathrm{NO}}_{\mathrm{2}}$ $/$ $\mathrm{NO}$ (ppbv $/$ ppbv)	12.7 $\pm$ 1.1	4.7 $\pm$ 0.5		3.4 $\pm$ 0.4	1.5 $\pm$ 0.2
Simulated $\mathrm{OH}$ ($\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}$)	5.5 $\pm$ 0.9	4.4 $\pm$ 0.6		1.5 $\pm$ 0.4	2.1 $\pm$ 0.3
Simulated ${\mathrm{HO}}_{\mathrm{2}}$ ($\times$ 10${}^{\mathrm{7}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}$)	29 $\pm$ 4.9	45 $\pm$ 4.2		2.0 $\pm$ 1.0	2.6 $\pm$ 0.7
Simulated ${\mathrm{RO}}_{\mathrm{2}}$ ($\times$ 10${}^{\mathrm{7}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}$)	19 $\pm$ 3.8	47 $\pm$ 6.2		1.2 $\pm$ 0.6	1.4 $\pm$ 0.4

${}^{\mathrm{a}}$ Average of the daily maximum solar radiation. C.I. denotes confidence interval. O${}_x={\mathrm{O}}_{\mathrm{3}}+{\mathrm{NO}}_{\mathrm{2}}$.

Chemical characteristics of air masses

Chemical composition

To inspect the chemical characteristics of air masses on ${\mathrm{O}}_{\mathrm{3}}$ episode days and non-${\mathrm{O}}_{\mathrm{3}}$ episode days, chemical species are statistically summarized at the two sites (Table 3). As expected, the levels of all pollutants (i.e., ${\mathrm{O}}_{\mathrm{3}}$, ${\mathrm{NO}}_{\mathrm{2}}$, $\mathrm{CO}$, ${\mathrm{SO}}_{\mathrm{2}}$, NMHCs and carbonyls) were significantly higher on ${\mathrm{O}}_{\mathrm{3}}$ episode days for both sites ($p<\mathrm{0.05}$), except for the comparable or even lower $\mathrm{NO}$ due to its titration to ${\mathrm{O}}_{\mathrm{3}}$ (see Sect. 3.3.2). Table S5 shows statistics of the top 10 NMHC and the top 3 carbonyl species observed during ${\mathrm{O}}_{\mathrm{3}}$ episodes and non-episodes at the two sites. The dominant species were quite similar regardless of episode or non-episode days at both sites. The higher concentrations of both primary and secondary pollutants on episode days than those on non-episode days were likely due to more intense photochemical reactions, more local pollutant accumulation as well as the regional transport of more highly polluted air masses. In contrast, the similar NMHCs composition at both sites during both episodes and non-episodes indicated somewhat interaction of air masses between the two sites regardless of ${\mathrm{O}}_{\mathrm{3}}$ levels.

It is worth mentioning that ${\mathrm{O}}_{\mathrm{3}}$ was much higher at WS than that at TC during both episodes and non-episodes ($p<\mathrm{0.01}$), with an average difference of 30.2 and 16.7 $\mathrm{ppbv}$, respectively (Table 3), though the levels of ${\mathrm{O}}_{\mathrm{3}}$ precursors (i.e., NO${}_x$ and VOCs) at WS were lower. Insight into VOC ratios found that ethene $/$ ethane (0.5 $\pm$ 0.04) and toluene $/$ benzene (2.2 $\pm$ 0.5) at WS were significantly ($p<\mathrm{0.05}$) lower than those at TC (0.7 $\pm$ 0.1 and 2.9 $\pm$ 0.4, respectively), likely indicating that the air masses at WS were more aged (Guo et al., 2007). Therefore, the higher ${\mathrm{O}}_{\mathrm{3}}$ at WS might be partially attributable to the aging of air masses (e.g., during the transport of continental air).

Influence of $\mathrm{NO}$ titration

Apart from the age of air masses, $\mathrm{NO}$ titration is another important factor influencing ${\mathrm{O}}_{\mathrm{3}}$ concentration. In areas with high $\mathrm{NO}$ levels, the $\mathrm{NO}$ titration (${\mathrm{O}}_{\mathrm{3}}$ $+$ $\mathrm{NO}$ $\to$ ${\mathrm{NO}}_{\mathrm{2}}$ $+$ ${\mathrm{O}}_{\mathrm{2}}$) is a main process consuming ${\mathrm{O}}_{\mathrm{3}}$. In this study, the average $\mathrm{NO}$ mixing ratio at TC was 14.0 $\pm$ 0.8 $\mathrm{ppbv}$, compared to 0.7 $\pm$ 0.1 $\mathrm{ppbv}$ at WS (Table 1). The much lower $\mathrm{NO}$ at WS implied weaker titration to ${\mathrm{O}}_{\mathrm{3}}$, which enabled the survival of more ${\mathrm{O}}_{\mathrm{3}}$ and caused substantial residual ${\mathrm{O}}_{\mathrm{3}}$ at WS, particularly at nighttime when there were no photochemical reactions (Figs. 2 and S6). Another direct evidence of $\mathrm{NO}$ titration effect was the trough of ${\mathrm{O}}_{\mathrm{3}}$ during the morning rush hours (06:00–07:00), together with an increase of ${\mathrm{NO}}_{\mathrm{2}}$ (Fig. S6). Furthermore, the total oxidants (O${}_x$ $=$ ${\mathrm{O}}_{\mathrm{3}}$ $+$ ${\mathrm{NO}}_{\mathrm{2}}$), which are usually adopted to take into account the NO titration influence, were comparable ($p>\mathrm{0.05}$) between TC and WS with mean values of 83.3 $\pm$ 3.7 and 81.6 $\pm$ 2.9 $\mathrm{ppbv}$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes and 49.4 $\pm$ 1.0 and 47.8 $\pm$ 1.0 $\mathrm{ppbv}$ during non-episodes, respectively (Table 3). This was reasonable in view of the interactions between the two sites. However, the remarkably higher ${\mathrm{O}}_{\mathrm{3}}$ and lower $\mathrm{NO}$ at WS indicated that $\mathrm{NO}$ titration was a determinant factor regulating the ${\mathrm{O}}_{\mathrm{3}}$ levels at both sites.

Moreover, $\mathrm{NO}$ titration is generally more significant on high-${\mathrm{O}}_{\mathrm{3}}$ days, resulting in higher ${\mathrm{NO}}_{\mathrm{2}}$ $/$ $\mathrm{NO}$ ratios due to the conversion of $\mathrm{NO}$ to ${\mathrm{NO}}_{\mathrm{2}}$ by ${\mathrm{O}}_{\mathrm{3}}$. Indeed, the mean ${\mathrm{NO}}_{\mathrm{2}}$ $/$ $\mathrm{NO}$ ratios increased from 4.7 $\pm$ 0.5 at WS and 1.5 $\pm$ 0.2 at TC during non-episodes to 12.7 $\pm$ 1.1 and 3.4 $\pm$ 0.4 during ${\mathrm{O}}_{\mathrm{3}}$ episodes, respectively, implying that more ${\mathrm{O}}_{\mathrm{3}}$ was titrated by NO during episodes. As a result, $\mathrm{NO}$ at TC was lower ($p<\mathrm{0.01}$) during ${\mathrm{O}}_{\mathrm{3}}$ episodes than during non-episodes (Table 3). It is noteworthy that $\mathrm{NO}$ at WS was on the same level between ${\mathrm{O}}_{\mathrm{3}}$ episode and non-${\mathrm{O}}_{\mathrm{3}}$ episode days ($p>\mathrm{0.05}$). This probably related to the weak titration at this marine site due to the trivial $\mathrm{NO}$ concentrations in both periods, as well as the counteracting effect of the increased transport of $\mathrm{NO}$ under northerly winds against the enhanced titration during ${\mathrm{O}}_{\mathrm{3}}$ episodes.

The aforementioned discussion demonstrated that $\mathrm{NO}$ titration played an important role in altering ${\mathrm{O}}_{\mathrm{3}}$ distribution, especially on ${\mathrm{O}}_{\mathrm{3}}$ episodes days. The lower $\mathrm{NO}$ (weaker $\mathrm{NO}$ titration) partially resulted in the higher ${\mathrm{O}}_{\mathrm{3}}$ concentrations observed at WS.

Atmospheric oxidative capacity and ${\mathrm{O}}_{\mathrm{3}}$ production rate

${\mathrm{O}}_{\mathrm{3}}$ formation is driven by the transformation and recycling of oxidative radicals, including $\mathrm{OH}$, ${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$, collectively referred to as RO${}_x$ hereafter. The production and loss rates of these radicals, and their equilibrium concentrations on the canister sampling days were simulated by the PBM-MCM model, as shown in Fig. S7. We noticed that WS featured significantly higher levels of these oxidative radicals on average ($p<\mathrm{0.05}$). The daytime (07:00–19:00 LT) average $\mathrm{OH}$ concentration at TC and WS was (1.5 $\pm$ 0.4) $\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}$ and (5.5 $\pm$ 0.9) $\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes, respectively. Consistently, ${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$ at WS were well above those at TC (Table 3). This pattern was also applicable between the two sites during non-episodes. Furthermore, while the difference in $\mathrm{OH}$ concentration became less on non-episode days, the gaps for peroxy radicals (${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$) between TC and WS widened, as listed in Table 3. From non-episodes to episodes, $\mathrm{OH}$ increased at WS alongside with the deceases of ${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$, likely indicating more conversion of ${\mathrm{HO}}_{\mathrm{2}}$ to $\mathrm{OH}$ by $\mathrm{NO}$, which is an important pathway leading to ${\mathrm{O}}_{\mathrm{3}}$ formation. Details about this were shown later.

To explain the inter-site differences of the concentrations of oxidative radicals and the variations between ${\mathrm{O}}_{\mathrm{3}}$ episodes and non-episodes, Fig. S7 also provides the breakdowns of the production and loss rates of $\mathrm{OH}$, ${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$ at TC and WS, separately. Overall, the reaction between ${\mathrm{HO}}_{\mathrm{2}}$ and $\mathrm{NO}$ dominated the production of $\mathrm{OH}$ at both sites, with the contribution of 69.4 $\pm$ 2.0 $\mathrm{\%}$ and 81.0 $\pm$ 1.5 $\mathrm{\%}$ at TC and WS, respectively. While the photolysis of $\mathrm{HONO}$ ranked second in the production of $\mathrm{OH}$ at TC (22.2 $\pm$ 2.1 $\mathrm{\%}$), the contribution of this pathway to $\mathrm{OH}$ production at WS (3.7 $\pm$ 0.6 $\mathrm{\%}$) was overstepped by ${\mathrm{O}}_{\mathrm{3}}$ photolysis (13.1 $\pm$ 1.6 $\mathrm{\%}$). This discrepancy was associated with the higher $\mathrm{HONO}$ and lower ${\mathrm{O}}_{\mathrm{3}}$ at TC (Fig. S1 and Table 3). As expected, the production rate of OH through ${\mathrm{HO}}_{\mathrm{2}}$ reacting with $\mathrm{NO}$ experienced the most significant increase from 1.4 $\pm$ 0.2 $\times$ 10${}^{\mathrm{7}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}{\mathrm{s}}^{-\mathrm{1}}$ during non-episodes to 3.6 $\pm$ 0.6 $\times$ 10${}^{\mathrm{7}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}{\mathrm{s}}^{-\mathrm{1}}$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes at WS, which explained more than 90 $\mathrm{\%}$ of the increase of the total $\mathrm{OH}$ production. In terms of the losses of $\mathrm{OH}$, reaction between $\mathrm{OH}$ and ${\mathrm{NO}}_{\mathrm{2}}$ was the largest sink of $\mathrm{OH}$ at TC. However, $\mathrm{OH}$-initiated oxidations of VOCs consumed most (52.7 $\pm$ 1.8 $\mathrm{\%}$) of $\mathrm{OH}$ at WS. This was reasonable in view of the much more abundant ${\mathrm{NO}}_{\mathrm{2}}$ at TC than at WS, in contrast to the smaller difference in NMHCs between the two sites (Table 3). Since $\mathrm{OH}$ can generally be recycled from the oxidation of VOCs, the lower $\mathrm{OH}$ at TC was likely caused by the lower ${\mathrm{O}}_{\mathrm{3}}$ photolysis and higher consumption of $\mathrm{OH}$ by ${\mathrm{NO}}_{\mathrm{2}}$, despite the more intensive $\mathrm{HONO}$ photolysis. The overall oxidation rate of VOCs by $\mathrm{OH}$ was employed to indicate the atmospheric oxidative capacity in previous studies (Elshorbany et al., 2009; Xue et al., 2016). In this study, we found that the oxidation rate of VOCs at TC (6.1 $\pm$ 2.1 $\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}{\mathrm{s}}^{-\mathrm{1}}$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes and 5.7 $\pm$ 0.9 $\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}{\mathrm{s}}^{-\mathrm{1}}$ during non-episodes) was remarkably ($p<\mathrm{0.05}$) lower than that at WS (${\mathrm{O}}_{\mathrm{3}}$ episode: 15 $\pm$ 2.5 $\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}{\mathrm{s}}^{-\mathrm{1}}$ and non-episode: 8.9 $\pm$ 1.3 $\times$ 10${}^{\mathrm{6}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}{\mathrm{s}}^{-\mathrm{1}}$). The results revealed that the atmospheric oxidative capacity at TC was weaker than at WS for both ${\mathrm{O}}_{\mathrm{3}}$ episodes and non-episodes, inconsistent with the findings of Elshorbany et al. (2009) and Xue et al. (2016), who concluded that the atmospheric oxidative capacity was higher in more polluted environments due to the fact that the atmospheric oxidative capacity is positively proportional to the VOCs and $\mathrm{OH}$ levels. Both Elshorbany et al. (2009) and Xue et al. (2016) reported very high mixing ratios of VOCs (e.g., toluene of 9.5 and 6.3 $\mathrm{ppbv}$, respectively) in the polluted cases, which explained the strong atmospheric oxidative capacity. However, in this study, it is more likely that the higher NO${}_x$ at TC consumed more $\mathrm{OH}$ and resulted in lower oxidative capacity than at WS, despite the slightly higher VOCs at TC (Table 3).

Daytime (07:00–19:00 LT) variations of the simulated production and loss rates of (a) RO${}_x$ and (b) ${\mathrm{O}}_{\mathrm{3}}$ at TC (left panel) and WS (right panel). ${\mathrm{O}}_{\mathrm{3}}$ episode days are highlighted in red background. The dates are not consecutive due to the discontinuous canister sampling of VOCs.

[Figure omitted. See PDF]

For ${\mathrm{HO}}_{\mathrm{2}}$, ${\mathrm{RO}}_{\mathrm{2}}$ reacting with $\mathrm{NO}$ accounted for 56.7 $\pm$ 1.1 $\mathrm{\%}$ and 60.7 $\pm$ 1.0 $\mathrm{\%}$ of ${\mathrm{HO}}_{\mathrm{2}}$ production at TC and WS, respectively. Oxidation of $\mathrm{CO}$ by $\mathrm{OH}$ was also an important pathway leading to ${\mathrm{HO}}_{\mathrm{2}}$ formation, second to ${\mathrm{RO}}_{\mathrm{2}}$ $+$ $\mathrm{NO}$ at both sites. At TC, ${\mathrm{HO}}_{\mathrm{2}}$ was almost exclusively depleted by $\mathrm{NO}$. However, 10.8 $\pm$ 1.8 $\mathrm{\%}$ and 6.5 $\pm$ 0.8 $\mathrm{\%}$ of the ${\mathrm{HO}}_{\mathrm{2}}$ losses were attributable to ${\mathrm{HO}}_{\mathrm{2}}$–${\mathrm{RO}}_{\mathrm{2}}$ and ${\mathrm{HO}}_{\mathrm{2}}$–${\mathrm{HO}}_{\mathrm{2}}$ reactions at WS, respectively, though ${\mathrm{HO}}_{\mathrm{2}}$ $+$ $\mathrm{NO}$ was responsible for the most fraction (82.7 $\pm$ 2.6 $\mathrm{\%}$) of ${\mathrm{HO}}_{\mathrm{2}}$ losses. We believe that the more significant self-consumption of peroxy radicals at WS was closely related to the low NO${}_x$ there, which hampered the transfer of oxygen atom from peroxy radicals to $\mathrm{NO}$ and further formation of ${\mathrm{O}}_{\mathrm{3}}$. This was confirmed by the enhanced losses of ${\mathrm{HO}}_{\mathrm{2}}$ through reactions with ${\mathrm{HO}}_{\mathrm{2}}$ itself and ${\mathrm{RO}}_{\mathrm{2}}$ from 3.0 $\pm$ 1.2 $\mathrm{\%}$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes to 24.9 $\pm$ 3.4 $\mathrm{\%}$ during non-episodes at WS, because NO${}_x$ was more scarce during non-episodes at this site (Table 3). Similarly, in contrast to the negligible influence of ${\mathrm{RO}}_{\mathrm{2}}$ reacting with ${\mathrm{HO}}_{\mathrm{2}}$ on ${\mathrm{RO}}_{\mathrm{2}}$ budget at TC, ${\mathrm{HO}}_{\mathrm{2}}$–${\mathrm{RO}}_{\mathrm{2}}$ reactions played important role in losses of ${\mathrm{RO}}_{\mathrm{2}}$ at WS, particularly on non-episode days (Fig. S7). When $\mathrm{OH}$, ${\mathrm{HO}}_{\mathrm{2}}$ and ${\mathrm{RO}}_{\mathrm{2}}$ were summed up, the production and loss rate of RO${}_x$ were obtained, as shown in Fig. 5a. Under such circumstance, the transformation and recycling pathways among these radicals can be neglected. For example, $\mathrm{OH}$-initiated oxidation of VOCs consumes $\mathrm{OH}$, which, however, generates ${\mathrm{RO}}_{\mathrm{2}}$. Therefore, these reactions were not considered as sources or sinks of RO${}_x$. On one hand, $\mathrm{HONO}$ photolysis was the largest source of RO${}_x$ at TC (53.7 $\pm$ 2.6 $\mathrm{\%}$), followed by the photolysis of HCHO (21.1 $\pm$ 1.6 $\mathrm{\%}$) and ${\mathrm{O}}_{\mathrm{3}}$ (18.7 $\pm$ 1.5 $\mathrm{\%}$). However, ${\mathrm{O}}_{\mathrm{3}}$ photolysis ranked the first among the sources of RO${}_x$ at WS with the contribution of 38.6 $\pm$ 2.3 $\mathrm{\%}$, higher than the contributions from $\mathrm{HCHO}$ photolysis (34.3 $\pm$ 1.4 $\mathrm{\%}$) and $\mathrm{HONO}$ photolysis (18 $\pm$ 2.5 $\mathrm{\%}$). On the other hand, while the reaction between $\mathrm{OH}$ and ${\mathrm{NO}}_{\mathrm{2}}$ served as the sole sink of RO${}_x$ at TC, it only explained 50 $\mathrm{\%}$ of RO${}_x$ sink at WS with the other half attributable to self-consumption of peroxy radicals.

Spatial distribution of $\mathrm{CO}$ at 08:00 (a), 14:00 (b), 19:00 (c) and 23:00 (d) on 21 August simulated by WRF-CMAQ, taken as an example of the “outflow” interaction pattern. Arrows in the figure represent the surface wind field.

[Figure omitted. See PDF]

Spatial distribution of ${\mathrm{O}}_{\mathrm{3}}$ at 08:00 (a), 14:00 (b), 19:00 (c) and 23:00 (d) on 21 August simulated by WRF-CMAQ, taken as an example of the “outflow” interaction pattern. Arrows in the figure represent the surface wind field.

[Figure omitted. See PDF]

Furthermore, the production and loss rates of ${\mathrm{O}}_{\mathrm{3}}$ were simulated (Fig. 5b). Despite the increased ${\mathrm{O}}_{\mathrm{3}}$ mixing ratio during episodes (Table 3), there was no significant change in net ${\mathrm{O}}_{\mathrm{3}}$ production between ${\mathrm{O}}_{\mathrm{3}}$ episodes (2.5 $\pm$ 1.0 $\mathrm{ppbv}{\mathrm{h}}^{-\mathrm{1}}$) and non-episodes (2.5 $\pm$ 0.5 $\mathrm{ppbv}{\mathrm{h}}^{-\mathrm{1}}$) at TC ($p>\mathrm{0.05}$), suggesting that regional transport might play critical roles in regulating ${\mathrm{O}}_{\mathrm{3}}$ levels at TC. In fact, previous studies (Huang et al., 2006; Jiang et al., 2008) have repeatedly confirmed that ${\mathrm{O}}_{\mathrm{3}}$ pollution at this site could be aggravated under northerly winds and/or downdraft on the periphery of typhoon. In contrast, the net ${\mathrm{O}}_{\mathrm{3}}$ production increased remarkably from non-episodes (2.8 $\pm$ 0.5 $\mathrm{ppbv}{\mathrm{h}}^{-\mathrm{1}}$) to ${\mathrm{O}}_{\mathrm{3}}$ episodes (6.6 $\pm$ 1.1 $\mathrm{ppbv}{\mathrm{h}}^{-\mathrm{1}}$) at WS. Obviously, ${\mathrm{O}}_{\mathrm{3}}$ production at WS was much higher than at TC during ${\mathrm{O}}_{\mathrm{3}}$ episodes, while they were comparable during non-episodes. This was likely due to the more abundant peroxy radicals (${\mathrm{RO}}_{\mathrm{2}}$ and ${\mathrm{HO}}_{\mathrm{2}}$) at WS than at TC, in addition to the increased NO${}_x$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes which enhanced the reactions between the peroxy radicals and $\mathrm{NO}$ (increasing ${\mathrm{O}}_{\mathrm{3}}$ formation). Insight into the ${\mathrm{O}}_{\mathrm{3}}$ production pathways found that the reaction rates of ${\mathrm{RO}}_{\mathrm{2}}$ $+$ $\mathrm{NO}$ and ${\mathrm{HO}}_{\mathrm{2}}$ $+$ $\mathrm{NO}$ were significantly enhanced from 1.6 $\pm$ 0.2 and 2.0 $\pm$ 0.4 $\mathrm{ppbv}{\mathrm{h}}^{-\mathrm{1}}$ during non-episodes to 3.2 $\pm$ 0.5 and 5.2 $\pm$ 0.9 $\mathrm{ppbv}{\mathrm{h}}^{-\mathrm{1}}$ during ${\mathrm{O}}_{\mathrm{3}}$ episodes, respectively. Our recent study (Wang et al., 2018) revealed that ${\mathrm{O}}_{\mathrm{3}}$ formation at WS was in a transition regime and much more sensitive to NO${}_x$ during non-episodes, when ${\mathrm{O}}_{\mathrm{3}}$ production through peroxy radicals reacting with $\mathrm{NO}$ was seriously limited by the low NO${}_x$. During ${\mathrm{O}}_{\mathrm{3}}$ episodes, with the increase ${\mathrm{O}}_{\mathrm{3}}$ precursors (particularly NO${}_x)$, these reactions were accelerated and the net ${\mathrm{O}}_{\mathrm{3}}$ production increased substantially. Detailed discussion on the ${\mathrm{O}}_{\mathrm{3}}$ photochemistry at WS can be found in our recent publication (Wang et al., 2018).

Spatial distribution of ${\mathrm{O}}_{\mathrm{3}}$ at 08:00 (a), 14:00 (b), 19:00 (c) and 23:00 (d) on 3 October, taken as an example of the sea–land breeze interaction pattern. Arrows in the figure represent the surface wind field.

[Figure omitted. See PDF]

Impact of air mass interaction on ${\mathrm{O}}_{\mathrm{3}}$ pollution in coastal areas

Driven by various weather systems (e.g., continental anticyclones, WPSH, tropical cyclones and SLBs), continental and marine air masses frequently interact with each other in the coastal areas. When continental air masses intrude into marine atmosphere, the chemical composition and atmospheric oxidative capacity over the marine atmosphere will be altered by the introduction of anthropogenic pollutants. Taking 21 August as an example, when the sampling sites (TC and WS) were under northwesterly to southwesterly winds caused by tropical cyclone (Fig. 2a), the maximum hourly ${\mathrm{O}}_{\mathrm{3}}$ reached 160 and 173 $\mathrm{ppbv}$ at TC and WS, respectively. Correspondingly, the primary air pollutants all stayed on high levels compared to those during non-episodes (Fig. 2a). Since WS was almost free of anthropogenic emissions, the great abundances of both primary and secondary air pollutants implied the influence of continental pollution on air quality at this site. Figures 6–7 depict the spatial distributions of $\mathrm{CO}$ and ${\mathrm{O}}_{\mathrm{3}}$ over the region of interest at selective times (08:00, 14:00, 19:00 and 23:00) on 21 August, respectively. $\mathrm{CO}$ is presented as an example of primary air pollutants emitted from anthropogenic sources. The spatiotemporal patterns of $\mathrm{CO}$ and ${\mathrm{O}}_{\mathrm{3}}$ were simulated by WRF-CMAQ. Noticeably, the model reproduced well the high level of $\mathrm{CO}$ in PRD region at 08:00, which was reasonable in view of the vehicular emissions in urban areas during morning rush hours. However, under the dominance of northwesterly winds in the morning, the center of high $\mathrm{CO}$ moved to the coastal areas. Even though the winds changed to southwesterly at noon, $\mathrm{CO}$ concentration over the SCS was still remarkably elevated according to the simulated results at 14:00. Further, the spatial distribution of $\mathrm{CO}$ at 19:00 and 23:00 confirmed the continuous movement of the polluted air masses away from the land under southwesterly winds. It should be noted that the increase of $\mathrm{CO}$ in PRD region at 19:00 and 23:00 was most likely caused by the vehicle emissions during evening rush hours. Overall, the dynamic distribution of $\mathrm{CO}$ in the study area clearly indicated the interaction between continental and marine atmospheres. As a result of the intrusion of continental air, a high level of ${\mathrm{O}}_{\mathrm{3}}$ was simulated over the SCS at 14:00 (Fig. 7b), which was comparable to the observed value (148 $\mathrm{ppbv}$) at WS. Moreover, ${\mathrm{O}}_{\mathrm{3}}$ was even higher over the SCS than that in continental area, mainly due to the more aged air masses, lower $\mathrm{NO}$ titration and higher oxidative capacity of the atmosphere (see Sect. 3.3). Consistent with $\mathrm{CO}$, the center of high ${\mathrm{O}}_{\mathrm{3}}$ moved away from the land. At 19:00, the ${\mathrm{O}}_{\mathrm{3}}$-laden air mass penetrated the SCS $\sim$ 300 $\mathrm{km}$, causing $\sim$ 8000 ${\mathrm{km}}^{\mathrm{2}}$ water area (8 times the area of Hong Kong) under high level of ${\mathrm{O}}_{\mathrm{3}}$ ($>$ 100 $\mathrm{ppbv}$). This case provided solid evidence of the transport of continental air masses to the SCS, which aggravated air pollution (particularly ${\mathrm{O}}_{\mathrm{3}}$ pollution) in this offshore area.

In contrast to outflow of continental air masses, the continental area near the coast could also be immersed by oceanic air masses. The arrival of oceanic air masses generally brings substantial marine-originated compounds (e.g., dimethyl sulfide) to the continent and significantly alleviates the anthropogenic air pollution there. In fact, this is one of the main reasons for low-${\mathrm{O}}_{\mathrm{3}}$ mixing ratio observed in the PRD region in summertime when southwestern winds prevail (T. Wang et al., 2009; Y. Wang et al., 2017). In this study, it was also found that winds over the ocean increased the concentration of dimethyl sulfide at TC (Fig. S8) and reduced the levels of almost all manmade air pollutants in many cases, mainly in summertime (Fig. 2a).

In contrast, sea breezes carrying elevated ${\mathrm{O}}_{\mathrm{3}}$ formed over the SCS might build up the terrestrial ${\mathrm{O}}_{\mathrm{3}}$ in the coastal area in some cases. Figure 8 shows the spatial distribution of ${\mathrm{O}}_{\mathrm{3}}$ over the study area on 3 October as an example of SLB regulating ${\mathrm{O}}_{\mathrm{3}}$ formation and distribution (see Fig. 4). Similar to the aforementioned scenario controlled by tropical cyclone, the simulated ${\mathrm{O}}_{\mathrm{3}}$ at 14:00 was generally higher over the SCS than in the terrestrial area, indicating the transport of polluted air masses from the land to the sea area. This was confirmed by the prevailing northeasterly winds in the morning (08:00 here). However, the ${\mathrm{O}}_{\mathrm{3}}$-laden air did not move far away from the land subsequently. Instead, it progressively approached the land, leading to increase of ${\mathrm{O}}_{\mathrm{3}}$ concentration in most parts of Hong Kong. This is because the wind direction in the coastal region changed from northeasterly to southeasterly at 17:00. Namely, the sea breeze appeared in late afternoon, which delivered the high ${\mathrm{O}}_{\mathrm{3}}$ formed over the SCS to the continental areas near the coast. In fact, the air quality monitoring stations deployed in southern Hong Kong by HKEPD also recorded the ${\mathrm{O}}_{\mathrm{3}}$ peak in the evening when ${\mathrm{O}}_{\mathrm{3}}$ could not be formed locally (Fig. S9), further confirming the recirculation of ${\mathrm{O}}_{\mathrm{3}}$-laden air from SCS to coastal areas of Hong Kong under sea breeze. However, the oceanic air did not penetrate further into inland PRD, which was likely stopped by the strong northeasterly winds dominated in the inland areas. Overall, it can be seen that SLB as a common interaction between marine and continental atmospheres played important role in regulating ${\mathrm{O}}_{\mathrm{3}}$ formation and distribution in coastal region of SCS, which is also applicable to other similar mesoscale environments throughout the world.

Conclusions

Coastal regions with dense population, economic prosperity and environmental pollution are common in the world. This study provided an overview of ${\mathrm{O}}_{\mathrm{3}}$ pollution in warm seasons around a coastal region of the SCS, focusing on the influences of interactions between marine and continental atmospheres on air quality in this subtropical region. The concurrent measurements of primary and secondary air pollutants at TC (a continental site) and WS (a marine site) from August to November 2013 indicated that ${\mathrm{O}}_{\mathrm{3}}$ was much higher at WS than that at TC, contrary to the more abundant primary air pollutants at TC. At the two sites, ${\mathrm{O}}_{\mathrm{3}}$ episodes and near-${\mathrm{O}}_{\mathrm{3}}$ episodes were frequently observed, which were closely associated with continental anticyclone, tropical cyclone and SLB. In addition to high temperature, strong solar radiation and weak wind, the aforementioned meteorological conditions all favored the transport of polluted air masses from continental areas to the SCS, during which the air pollutants were transformed with the aging of air masses. After arriving in the SCS, the land-originated air pollutants further involved in intensive photochemical reactions with the trait of low $\mathrm{NO}$ titration to ${\mathrm{O}}_{\mathrm{3}}$ and high ${\mathrm{O}}_{\mathrm{3}}$ production rate, leading to higher ${\mathrm{O}}_{\mathrm{3}}$ levels in marine atmosphere (WS) than in coastal cities (TC). In addition to the continental outflow that aggravated ${\mathrm{O}}_{\mathrm{3}}$ pollution over the SCS, SLB as a common interaction in coastal areas also often facilitated the recirculation of ${\mathrm{O}}_{\mathrm{3}}$ formed over the SCS to the continental areas, building up ${\mathrm{O}}_{\mathrm{3}}$ concentration in coastal cities under sea breeze. The findings can be extended to other similar regions to advance our understanding of ${\mathrm{O}}_{\mathrm{3}}$ pollution.

The data set is available to the community and can be accessed by request to Hai Guo ([email protected]) of the Hong Kong Polytechnic University.

The Supplement related to this article is available online at https://doi.org/10.5194/acp-18-4277-2018-supplement.

The authors declare that they have no conflict of interest.

Acknowledgements

This project was supported by the Natural Science Foundation of China (grant no. 41275122), the Research Grants Council (RGC) of the Hong Kong Government of Special Administrative Region (PolyU5154/13E, PolyU152052/14E, PolyU152052/16E and CRF/C5004-15E), the Guangdong special fund for science and technology development (2017B020216007) and partly by the Hong Kong PolyU internal grant (G-SB63, 1-BBW4 and 4-ZZFW). The authors thank HKEPD for provision of the air quality and meteorological data at TC site and are grateful to Po On Commercial Association Wan Ho Kan Primary School at Tung Chung and the National Marine Environmental Monitoring Station at Wanshan Island for their generous support of the field study. Contributions to field measurements by Kalam Cheung, Dawei Wang, Bo Liu, Nan Wang, Jiamin Ou, Huanghuang Yan and Xiaoxin Fu are also highly appreciated. The authors also gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for the provision of the HYSPLIT transport and dispersion model and/or READY website (http://www.ready.noaa.gov) and Tao Wang's group of the Hong Kong Polytechnic University for the provision of the average diurnal profiles of HONO at TC and HT. Edited by: Aijun Ding Reviewed by: two anonymous referees