The air quality observational data are used to identify the regional characteristics of the O3 episode in August 2013. Fifteen cities are selected as the representative research objects to better reflect the status of O3 pollution over the YRD region. The locations of these cities are shown in Fig. 1b, which contains Shanghai, eight cities in Jiangsu province (Changzhou, Nanjing, Nantong, Suzhou, Taizhou, Wuxi, Yangzhou and Zhenjiang) and six cities in Zhejiang province (Hangzhou, Huzhou, Jiaxing, Ningbo, Shaoxing and Zhoushan). The in situ monitoring data for the hourly concentrations of O3, CO, NO2, SO2, PM2.5 and PM10 can be acquired from National Environmental Monitoring Center (NEMC). The quality assurance and quality control (QA/QC) procedures for monitoring strictly follow the national standards (State Environmental Protection Administration of China, 2006). The hourly pollutant concentration for a city is calculated as the average of the pollutant concentrations from several national monitoring sites in that city, which can better characterize the pollution level of the city. In order to identify invalid or lacking data, a checking procedure for these data is performed following the work of Chiqueto and Silva (2010). Finally, only less than 0.2 % of the primary data are ignored in the calculation. Moreover, the observed data of total volatile organic carbons (TVOC) during 4–10 August at an urban site in Shanghai (SAES, 31.17∘ N, 121.43∘ E) are also used. They are provided by Shanghai Academy of Environmental Sciences. The sampling height is about 15 m, and individual VOC species are continuously measured every 30 min by two online high performance gas chromatograph with flame ionization detector (GC-FID) systems (Chromato-sud airmoVOC C2-C6 no. 5250308 and airmoVOC C6-C12 no. 2260308, France). The details for measurement and QA/QC can refer to Wang et al. (2013).

The weather charts and the observed surface meteorological records are used to analyze the synoptic systems during the episode. The weather charts for East Asia are accessible from the Korea Meteorological Administration. The hourly meteorological data at the observation sites of SH (31.40∘ N, 121.46∘ E) located in Shanghai, HZ (30.23∘ N, 120.16∘ E) in Hangzhou and NJ (32.00∘ N, 118.80∘ E) in Nanjing can be obtained from the University of Wyoming, where 2 m air temperature, 2 m relative humidity, 10 m wind speed and 10 m wind direction are available.

Meteorological and air quality observation data are also used to validate the reliability of simulations in this study. Comparisons of the modeling results with the observation data are performed in Shanghai, Nanjing and Hangzhou. Shanghai is the most populous city in China as well as a global financial and transportation center. Located to the northwest of Shanghai, Nanjing is the capital of the province of Jiangsu and the second largest commercial center in East China. Hangzhou is the capital of the province of Zhejiang and located to the southwest of Shanghai. These cities are the provincial capitals and the typical metropolis in the YRD region. They are highly urbanized and industrialized, and all suffer from severe O3 pollution.

In this study, WRF/CMAQ, which consists of WRF model version 3.4.1 and CMAQ model version 4.7.1, is applied to simulate the high O3 episode over the YRD region in August 2013. Developed at the National Center for Atmospheric Research (NCAR), WRF is a new generation of mesoscale weather forecast model and assimilation system. Numerous applications have proven that it shows a good performance in all kinds of weather forecasts and has broad application prospects in China (Jiang et al., 2008, 2012; X. M. Wang et al., 2009; Liu et al., 2013; Xie et al., 2014, 2016a; Liao et al., 2014, 2015; Li et al., 2016; Zhu et al., 2015). WRF provides offline meteorological fields as the input for the chemical transport model CMAQ. CMAQ is a third-generation regional air quality model developed by the Environmental Protection Agency of USA (USEPA). A set of up-to-date compatible modules and control equations for the atmosphere is incorporated in the model, which can fully consider atmospheric complicated physical processes, chemical processes and the relative contribution of different species (Byun and Schere, 2006; Foley et al., 2010). CMAQ has been widely applied in China and proven to be a reliable tool in simulating air quality from city scale to mesoscale (Li et al., 2012; Wei et al., 2012; Liu et al., 2013; Zhu et al., 2015).

The simulation run is conducted from 08:00 (local standard time, LST) on 2 August to 08:00 (LST) on 16 August 2013, in which the first 48 h is taken as the spin-up time. Three one-way nested domains are used in WRF with a Lambert conformal map projection. The domain setting is shown in Fig. 1. The outermost domain (domain 1, d01) covers the most areas of East Asia and South Asia, with the horizontal grids of 88×75 and the grid spacing of 81 km. The nested domain d02 covers the southeastern part of China, with the horizontal grids of 85×70 and the grid spacing of 27 km. The finest domain (domain 3, d03) covers the core areas of the YRD region, with the grid system of 70×64 and the resolution of 9 km. For all domains, there are 23 vertical sigma layers from the surface to the top pressure of 100 hPa, with about 10 layers in the planetary boundary layer. The detailed configuration options for the dynamic parameterization in WRF are summarized in Table 1. Additionally, the SLAB scheme that does not consider urban canopy parameters is adopted to model the urban effect. In order to reflect the rapid urban expansion in the YRD region, the default United States Geological Survey (USGS) land-use archives are updated by adding the present urban land-use conditions from 500 m Moderate Resolution Imaging Spectroradiometer (MODIS) data, based on the work of Liao et al. (2014, 2015). The initial meteorological fields and boundary conditions are from 1∘ resolution global reanalysis data provided by National Center for Environmental Prediction (NCEP). The boundary conditions are forced every 6 h.

With respect to the air quality model, CMAQ uses the same vertical levels and the similar three nested domains as those adopted in the meteorological simulation, whereas the CMAQ domains are one grid smaller than the WRF domains. The Meteorology Chemistry Interface Processor (MCIP) is used to convert WRF outputs to the input meteorological files needed by CMAQ. The Carbon Bond 05 chemical mechanism (CB05) (Yarwood et al., 2005) is chosen for gas-phase chemistry (CHEM), and the fourth-generation CMAQ aerosol module (Byun and Schere, 2006) is adopted for aerosol chemistry. The initial and outmost boundary conditions are obtained from the Model for Ozone and Related Chemical Tracers version 4 (MOZART-4) (Emmons et al., 2010), while those for the two nested inner domains are extracted from the immediate concentration files of their parent domains. The anthropogenic emissions are mainly from the 2012 Multi-resolution Emission Inventory for China (MEIC) with 0.25∘×0.25∘ resolution, which is re-projected for the grids of China in both domains. For the grids outside of China, the inventory developed for the Intercontinental Chemical Transport Experiment-Phase B (INTEX-B) by Zhang et al. (2009) is used. The natural O3 precursor emissions are calculated by the natural emission model developed by Xie et al. (2007, 2009, 2014), including NO from soil, VOCs from vegetation and CH4 from rice paddies and terrestrial plants. The biomass burning emissions are acquired from the work of Xie et al. (2014, 2016a).

The CMAQ modeling system contains process analysis module (PROCAN), which consists of the IPR analysis and the integrated reaction rate (IRR) analysis (Byun and Schere, 2006). IPR has the capability of calculating the hourly contributions of individual physical processes and the net effect of chemical reaction compared to the overall concentrations and thereby can determine the quantitative contribution of each process in a specific grid cell. The atmospheric processes considered in IPR include the horizontal advection (HADV), the vertical advection (ZADV), the horizontal diffusion (HDIF), the vertical diffusion (VDIF), the emissions (EMIS), the dry deposition (DDEP), the cloud processes with the aqueous chemistry (CLDS), the aerosol processes (AERO) and CHEM. IPR has been widely applied to investigate the regional photochemical pollution and has proven to be an effective tool to show the relative importance of every process and provide a fundamental interpretation (Gonçalves et al., 2009; Li et al., 2012; Liu et al., 2013; Zhu et al., 2015). In this paper, the period from 4 to 15 August is selected for the IPR analysis. With the aid of IPR, we assess the roles of the individual physical and chemical processes involved in O3 formation over the YRD region and further present those in the typical cities (Shanghai, Nanjing and Hangzhou).

Comparisons of the modeling results in the finest domain (d03) with the hourly observation data are performed for meteorological factors and air pollutants in Shanghai, Hangzhou and Nanjing. The correlation coefficient (R), the normalized mean bias (NMB) and the root-mean-square error (RMSE) are used to evaluate the model performance. These statistic values are calculated as follows: R=∑i=1N(Si-S‾)(Oi-O‾)∑i=1N(Si-S‾)2∑i=1N(Oi-O‾),NMB=∑i=1N(Si-Oi)∑i=1NOi×100%,RMSE=1N∑i=1N(Si-Oi)212, where Si and Oi represent the simulated and the observed value, respectively. N means the total number of valid data. Generally, the model performance is acceptable if the values of NMB and RMSE are close to 0 and that of R is close to 1.

Figure 2 shows the temporal variation of the hourly O3 concentrations observed in 15 typical cities over the YRD region from 00:00 (universal time coordinated, UTC) 4 August to 23:00 (UTC) 15 August in 2013. Obviously, from 7 to 12 August, high O3 concentrations over 93.5 ppb (approximately equal to the hourly national air quality standard of 200 µg m-3) have been frequently recorded in 13 cities, which means O3 concentrations in most cities over the YRD region exceed the national air quality standard. So, this high O3 pollution episode is a typical regional O3 pollution episode that can affect the people and the ecosystem in a large area. In general, for each city, there is a remarkable continuous growth in O3 concentrations before the O3 episode, followed by the lasting heavy O3 pollution period. Though the O3 concentrations in Shaoxing and Nanjing meet the national O3 standard, their time series still show the similar tendency to those of the other cities in the same region. The excessive level of O3 occurring in Huzhou, Jiaxing, Nantong, Yangzhou and Shanghai lasts for more than 6 consecutive days, reflecting the regional continuous characteristics of this O3 pollution episode.

According to the temporal variation characteristics of O3 illustrated in Fig. 2, the abovementioned 15 typical YRD cities can be classified into three categories: (1) the cities in the Southeast Coastal Region (SCR), including Shanghai, Suzhou, Jiaxing, Ningbo, Shaoxing and Zhoushan; (2) the cities in the Central Inland Region (CIR), including Hangzhou, Huzhou, Wuxi, Changzhou and Nantong; and (3) the cities in the Northwest Inland Region (NIR), including Nanjing, Yangzhou, Zhenjiang and Taizhou. The classification is primarily on basis of the observational facts that the maximum O3 concentrations occur on 10–11, 12 and 13 August and begin to synchronously decrease on 12, 13 and 14 August in SCR, CIR and NIR, respectively. As shown in Fig. 2, in the SCR, Zhoushan firstly exceeds the national O3 standard on 4 August, followed by Jiaxing, Shanghai, Suzhou and Ningbo. The peak hourly O3 concentration of SCR occurs in Jiaxing on 10 August, with the value up to 162.4 ppb. In the CIR, Huzhou is the first city exceeding the national O3 standard, followed by the order of Nantong, Changzhou, Wuxi and Hangzhou. The high-level O3 pollution in Huzhou lasts from 5 to 13 August. In Nantong and Changzhou, the maximum hourly O3 concentration reaches 167.1 ppb on 10 August and 166.1 ppb on 12 August, respectively. As for the NIR, Yangzhou, Zhenjiang and Taizhou successively exceed the national O3 standard. It is also noteworthy that the date when O3 concentration exceeds the national air quality standard in coastal region is ahead of that in inland regions, as is the date of O3 decrease. The different start time of O3 decreasing in different regions might be related to the strong southeast wind in accordance with the movement of Typhoon Utor, which is discussed in Sect. 3.2 in detail.

Table 2 presents the maximum and the average concentrations of O3 and NO2 in 15 YRD cities during 7–12 August 2013. It illustrates that the mean concentrations of NO2 in different YRD cities range from 7.7 to 24.5 ppb during the O3 episode, indicating the heterogeneity of the spatial distribution of O3 precursor emissions. For O3, the highest hourly concentration (167.1 ppb) occurs in Nantong, followed by 166.1 ppb in Changzhou and 162.4 ppb in Jiaxing. These values are all nearly 2 times the national air quality standard. It seems that O3 concentrations are higher in the cities around Shanghai, where the concentrations of O3 precursors are more adequate as well. High concentrations of O3 and its precursors imply that there may be stronger photochemical reactions.

Figure 3 demonstrates the hourly variations of the observed NO2 concentrations in Shanghai, Nanjing and Hangzhou from 4 to 15 August 2013 and the time series of TVOC observed at SAES in Shanghai from 4 to 10 August 2013. Obviously, there are two peaks in the diurnal cycles of NO2 and VOC at all sites, which should be related to the rush hours in cities. The photolysis of NO2 dominates O3–VOC–NOx chemistry after 08:00 and thereby makes the concentrations of precursors (NO2 and VOC) begin to decrease. Thus, the related reactions form O3 and increase its concentration until about 14:00. These diurnal variations of O3 and its precursors follow the typical patterns in the polluted areas and reflect the close relationships between O3, VOC and NOx (Wang et al., 2013; Xie et al., 2016b). Moreover, the daily variations of NO2 and VOC show good agreement with those of O3. For VOC, the concentration in Shanghai largely increases since 6 August, which corresponds well with the over-standard O3 concentrations since then (Fig. 2). For NO2, the higher values occur from 6 to 11 August in all cities, but the concentrations start to decrease on 12, 13 and 14 August in Shanghai, Hangzhou and Nanjing, respectively. It seems that the changes of O3 precursors (NO2 and VOC) are also affected by the movement of Typhoon Utor.

Favorable weather conditions have large impacts on the formation of severe O3 pollution (Huang et al., 2005, 2006; T. J. Wang et al., 2006; Jiang et al., 2008; Cheng et al., 2014; Hung and Lo, 2015). High-level O3 episodes often take place in hot seasons, when the meteorological conditions with high temperature and strong solar radiation are beneficial to the photochemical reactions of O3 (Lam et al., 2005). Figure 4 shows the variations of the surface meteorological parameters that are related to this photochemical pollution episode during 4–15 August, including 2 m air temperature, 2 m relative humidity, 10 m wind speed and 10 m wind direction at the meteorological sites in Shanghai (SH) of SCR, Hangzhou (HZ) of CIR and Nanjing (NJ) of NIR.

As shown in Fig. 4a, the hot weather at SH, HZ and NJ exists for nearly a week from 7 to 12 August, with the hourly maximum temperature reaching the value over 40 ∘C. Meanwhile, the variations of 2 m relative humidity show the negative correlation with those of 2 m air temperature. The minimum 2 m relative humidity at SH and HZ occur on 9 and 10 August respectively, with the value below 75 %. These minimum values are also lower than the values before and after the O3 episode, suggesting that high-level O3 episodes usually occur under the weather conditions with high temperature and low humidity. The value of 2 m relative humidity at NJ is relatively higher than those at SH and HZ and remains more stable. These extremely hot and dry weather conditions at SH, HZ and NJ are successively relieved on 12, 13 and 15 August, which coincides well with the reduction of surface O3 concentrations in Shanghai, Hangzhou and Nanjing (Fig. 2). With respect to the observed surface wind (Fig. 4b), the 10 m wind speed at SH and HZ is comparatively lower during the period of the O3 episode, while it is suddenly intensified after 12 August. Meanwhile, the wind direction is fluctuating from 7 to 12 August, while it maintains southeasterly wind after 12 August as well. The growth of wind speed is more distinct at SH, with the maximum value of approximately 10 m s-1. The wind speed at NJ has an obviously diurnal variation from 4 to 8 August, and the minimum value occurs on 10 August.

Figure 5 displays the weather charts for the 500 hPa layer over East Asia at 00:00 (UTC) on 6, 8, 10 and 12 August 2013, which can illustrate the main synoptic patterns causing the O3 pollution. Obviously, during the period of the selected O3 episode, the whole YRD region is under the control of the strong western Pacific subtropical high, which is stronger and extends much farther west than normal. The anomaly of the subtropical high might be the direct and leading cause of the abnormally high temperature shown in Fig. 4a (Peng et al., 2014). The intensity of the subtropical high is usually characterized by the area index, defined as the total number of grid points that have geopotential heights of 588 decameters or greater in the region of 110–180∘ E and northward of 10∘ N. As shown in Fig. 5, the 5880-m area covers most of southeast China, and the high pressure center (5920-m area) is located in the southeastern coastal areas as well as the surrounding sea areas, which means the subtropical high is very intensive. This high pressure strengthens and remains over the YRD region for several days (from 6 to 12 August), implying that the air subsides to the ground. The downward air acts as a dome capping the atmosphere and helps to trap heat as well as air pollutants at the surface. Without the lift of air, there is little convection and therefore little cumulus clouds or rains. The end result is a continual accumulating of solar radiation and heat on the ground, which may greatly enhance the photochemical reactions between the abundant built-up air pollutants.

The other weather system worthy of note is Typhoon Utor (shown in Fig. 5c and d). Typhoon Utor is one of the strongest typhoons in the 2013 Pacific typhoon season. It is formed early on 8 August, develops into a tropical storm on 9 August, undergoes an explosive intensification within a half of day and achieves typhoon status early on 10 August. After landing in Luzon of the Philippines on late 11 August, it reemerges in the South China Sea on 12 August. Typhoon Utor hits the land of Guangdong Province in China on 14 August and thereby is finally weakened into a tropical storm. In the end, it is ultimately dissipated on 18 August. It was reported that ozone episodes during the hot season are usually associated with the passage of tropical cyclones close to the territory (Huang et al., 2005; T. J. Wang et al., 2006; Jiang et al., 2008; Cheng et al., 2014; Hung and Lo, 2015). When a site is at the front of moving typhoon system, it can be controlled by the downward airflow induced by the typhoons' peripheral circulation. So, the typhoon system can cause the local weather around the site with high temperature, low humidity, strong solar radiation and small wind for a short time, before it is close enough to bring winds and rains. All these changes of meteorological conditions can help to form the severe continuous O3 pollution (Jiang et al., 2008). In this O3 episode, the YRD region may be influenced by the peripheral circulation of Typhoon Utor as well. Especially on 10–11 August, the downward airflow in the troposphere is significantly strengthened (shown in Fig. 7), which may enhance the buildup of heat and air pollutants and thereby result in worse air quality, shown in Fig. 2.

Moreover, with the approaching of Typhoon Utor from 12 to 14 August, the near-surface breeze over the YRD region gradually becomes the prevailing southeasterly or southerly wind (Fig. 5d), with the highest wind speed up to 6–10 m s-1 in Shanghai (Fig. 4). The strengthened wind can bring the clean marine air from ocean to inland and thereby effectively mitigate the O3 pollution. Meantime, Typhoon Utor also gradually affects the position and strength of the western Pacific subtropical high. As the typhoon continuously approaches and finally lands on Guangdong, the high pressure system is forced to retreat easterly and move northwards. When the high pressure center completely moves to the oceans, the YRD region is totally under the control of the typhoon system. In the end, the hot weather is relieved and the O3 pollution is mitigated. The coastal cities in CSR are closer to the typhoon system, so they are firstly influenced during this period. Thus, the wind at SH in CSR firstly changes, followed by HZ in CIR and NJ in NIR. In the same way, 2 m air temperature and O3 concentrations also successively decrease from southeast (SH in CSR) to northwest (NJ in NIR) owing to the scavenging effect.

To evaluate the simulation performance, the hourly modeling results during the period of 4–15 August 2013 are compared with the observation records. Table 3 presents the performance statistics, including the values of R, the NMB and the RMSE, which are all calculated for 2 m air temperature (T2), 2 m relative humidity (RH2), 10 m wind speed (Wspd10), 10 m wind direction (Wdir10), surface O3 concentrations and surface NO2 concentrations in SH, NJ and HZ.

As indicated in Table 3, the simulated results of surface air temperature and relative humidity from WRF show good agreement with the observations. The highest correlation coefficient of T2 is found to be 0.91 at SH, followed by 0.84 at NJ and 0.80 at HZ (statistically significant at 95 % confident level). The corresponding correlation coefficients for RH2 are 0.85, 0.83 and 0.78, respectively. The values of RMSE for T2 at SH, NJ and HZ are 4.15, 2.91 and 3.09 ∘C and those for RH2 are 19.3, 9.41 and 13.96 %, respectively. Our simulation underestimates T2 and overestimates RH2 to some certain extent, with the values of NMB for T2 at SH, NJ and HZ being -5.68, -5.98 and -6.53 % and those for RH2 being 12.64, 4.52 and 16.36 %. These biases might be attributed to the uncertainty caused by the SLAB scheme, which can underestimate temperature in summer (Liao et al., 2014). However, according to the relevant studies (Li et al., 2012; Liao et al., 2015; Xie et al., 2016a), this level of over- or underestimation is still acceptable. The wind components are closely related to the transport processes. As shown in Table 3, our modeling results of wind speed and direction basically reflect the characteristics of wind fields. For Wspd10, R is 0.77 at SH, 0.74 at NJ and 0.75 at HZ. Though the values of NMB (1.53, 5.92 and 9.21 %) and RMSE (2.18, 2.41 and 2.39) display that the simulated wind speeds are a little overestimated, the biases are still reasonable and acceptable. For Wdir10, the simulated values also fit the observation records well, with the R values of 0.63 at SH, 0.57 at NJ and 0.58 at HZ. Comparing the mean values from SIM and OBS, we can find that WRF model generally simulates the prevailing wind direction during this period. In summary, the abovementioned performance statistics numbers illustrate that the WRF simulation can reflect the major characteristics of meteorological conditions of this O3 episode, and the meteorological outputs can be used in the pollutant concentration simulation.

Figure 6 shows the comparisons between the modeling results from CMAQ and the observed hourly concentrations of O3 in Shanghai, Nanjing and Hangzhou during 4–15 August 2013. Obviously, the observations and the simulated results present reasonable agreement at each site, with the correlation coefficients of 0.81 to 0.83, NMB of -6 to 12.12 % and RMSE of 33.95 to 38.79 ppb. Moreover, the simulation also reproduces the diurnal variation of O3, which shows that the concentration reaches its maximum at around noontime and gradually decreases to its minimum after midnight. With respect to the O3 precursor, comparisons of NO2 concentrations between simulation results and observations show that the correlation coefficient at each city is about 0.6 (given in Table 3), which further proves that the process of O3 formation is captured reasonable well over the YRD region and throughout the episode. However, CMAQ overestimates NO2 and underestimates O3 in Shanghai, while it underestimates NO2 and overestimates O3 in Nanjing and Hangzhou. These biases of O3 and NO2 should mainly be attributed to the uncertainties in emissions of O3 precursors (NOx and VOCs) (Li et al., 2012; Liao et al., 2015; Xie et al., 2016). Because of the VOC-sensitive O3 chemistry in the daytime and NOx titration at night in the YRD region (Xie et al., 2014), higher estimation of NOx emission in Shanghai may lead to higher NO2 and lower O3 predictions, while lower NOx estimations in Nanjing and Hangzhou may result in lower NO2 and higher O3 modeling results. The undervalued NO2 and overvalued O3 in Nanjing and Hangzhou can also be related to the overestimations in WS10 and the negative biases in T2. Moreover, the uncertainties in nonlinear chemical reactions coupled in CMAQ may also have important effects on model predictions. For example, the modeling results cannot catch the low O3 values observed at night in Nanjing and Hangzhou (Fig. 6), implying there may be some imperfections in the nocturnal chemistry of CMAQ. Nevertheless, the performance of CMAQ model is comparable to the other applications (Gonçalves et al., 2009; Li et al., 2012; Zhu et al., 2015). Compared to these previous related studies, the simulation in this study attains an acceptable and satisfactory result. Thus, the consistency of simulation and observation demonstrates that the modeling results are capable of capturing and reproducing the characteristics and changes of photochemical pollutants and can be used to provide valuable insights into the governing processes of this O3 episode.

Figure 7 presents the daytime vertical wind velocity as well as the vertical distribution of O3 concentrations from 116.5 to 122.9∘ E along the latitude of 31.40∘ N (where Shanghai is located) during 7–12 August 2013. The simulation results clearly illustrate that there are strong downward airflows over the YRD region during the period of the regional high-level O3 pollution, which can be attributed to the fact that these areas are under the control of the subtropical high and the sinking airflow is predominant (as discussed in Sect. 3.2).

From 7 to 9 August 2013, except for the aforementioned regional sinking airflows, there are still some local thermal circulations continually occurring at the lower atmospheric layers (< 2 km) along the vertical cross section of HZ–NJ. These circulations are related to urban heat islands. Usually high pressures are accompanied by more stagnant and fair dry weather, so the upward and the downward flows caused by urban-breeze circulations can easily appear in the urban areas. For the vertical distribution of O3, its high concentrations (> 50 ppb) generally appear from the surface to 1.5 km height above the cities. As discussed in Sect. 3.2, air pollutants tend to be trapped on the ground due to the regional sinking airflows. Moreover, the local circulations over the cities make the urban areas to be the convergence zones, and thereby more air pollutants can be accumulated in and around these cities. Under the weather conditions induced by the subtropical high, such as high air temperature, stronger solar radiation and less water vapor, the chemical reactions between the built-up air pollutants can be enhanced to form the high-level O3 pollution. Additionally, Fig. 7a–c also show that there are maximum O3 concentrations (> 90 ppb) occurring near the surface in and around SH. This phenomenon should be explained by the fact that the coastal city (SH) is firstly affected by Typhoon Utor.

From 10 to 12 August, with the approaching of Typhoon Utor, the vertical air movements over the YRD region are not restricted at the lower atmosphere anymore. As shown in Fig. 7d–f, there are stronger downward airflows from the surface to the top of troposphere. As discussed in Sect. 3.2, the YRD cities are at the front of the moving typhoon system, so the peripheral circulation of Typhoon Utor may enhance the sinking of atmosphere, which can lead to higher air temperature, lower humidity and stronger solar radiation. Affected by the enhanced downward air movement as well as the relevant changes of meteorological conditions, O3 concentrations over the YRD region maintain a high pollution level, with the O3 concentrations over 60 ppb below the height of 1.5 km (Fig. 7d–f). Furthermore, the high value center of O3 concentrations (> 90 ppb) moves westwards during 10–12 August, implying that the peripheral circulation of Typhoon Utor can drive the air from the coastal areas to the inland areas.

The vertical changes of wind velocity and O3 concentrations above Shanghai, Hangzhou and Nanjing are further illustrated in Fig. 8. Similarly, the atmospheric subsidence can also be found in the troposphere (usually occur at more than 1 km above the surface) during the period of high-level O3 pollution. With respect to Shanghai, affected by the extremely high temperature, more active photochemical reactions lead to higher O3 concentrations in the whole atmospheric boundary layer. The downward airflows induced by the subtropical high trap and enhance the accumulation of surface O3 as time passes. Thus, high O3 concentrations are formed below 2 km above the urban areas of Shanghai, and the high concentration centers occur near the surface below 500 m. It is interesting that O3 concentration on 8 August is comparatively lower, which can be seen in Fig. 2 as well. This phenomenon can be explained by the fact that the transient upward airflow occurs at above 300 m over Shanghai and inhibits the accumulation of the O3 pollution at the surface (shown in Fig. 8a). Additionally, Fig. 8a also presents the possible effects of Typhoon Utor on the formation of O3. On 10 August, when the typhoon system approaches the eastern coastal areas of China, the sinking air above Shanghai is apparently strengthened and thereby enhances the intensity of O3 pollution as well as the scope of the pollution. However, after 12 August, when Typhoon Utor changes the wind and even impacts the subtropical high, high temperature is alleviated and the built-up O3 is transported to other places. Thus, the pollution is mitigated. As to Hangzhou (Fig. 8b), from 7 to 9 August, owing to weaker photochemical reactions, lower O3 concentrations than those in Shanghai are found in the boundary layer. However, the O3 concentration can exceed the national standard from 10 to 12 August (Fig. 2), which should be influenced by the typhoon system. The influence process is similar to the above discussion for Shanghai; that is, the upper downward airflows (over 1 km above the surface) are enhanced significantly since 10 August. However, for Nanjing, the O3 concentration does not exceed the national O3 standard during 7–12 August (Figs. 2 and 8), which should be attributed to the fact that Nanjing is far away from the coastal areas and thereby hardly affected by the downward flow in the typhoon periphery. Though the O3 concentration in Nanjing increases on 12 August, it should mainly be caused by the local photochemical reactions because the vertical movement below 2 km above Nanjing is dominated by upward airflows.

It also should be mentioned that the near-surface vertical velocities around these cities are much lower than those at higher altitudes (Fig. 8). Especially in the planetary boundary layer (< 1 km), lots of zero-velocity lines appear near the ground. This phenomenon may be related to the upward airflow caused by urban heat islands. Thus, the maximum centers of O3 occur near the surface below 500 m, and the vertical diffusion process plays a more important role in the accumulation of surface O3. The essential role of the vertical diffusion process in the O3 episode is similar to that reported by Zhu et al. (2015).