Current Issue

The ozone data analyzed in this study have been measured at air quality monitoring stations nationwide by UV photometric method since 1989. The data used in the study were collected from monitoring stations in 16 cities and province for recent 10 years: Seoul (16 stations), Incheon (6 stations), Busan (10 stations), Daegu (4 stations), Daejeon (3 stations), Gwangju (4 stations), Gyeonggi (28 stations), Ulsan (7 stations), Gangwon (3 stations), Gyeongbuk (10 stations), Gyeongnam (7 stations), Jeonbuk (4 stations), Jeonnam (6 stations), Chungbuk (3 stations), Chungnam (3 stations), Jeju (3 stations). The locations of each atmospheric monitoring station in 16 the cities and provinces are shown in Fig. 1. More detailed information including longitude and latitude for each monitoring station are shown in Table S1. The hourly ozone concentrations at each monitoring station collected and managed by the National Air Quality Data Center (NAQDC), National Institute of Environmental Research (NIER), were used to calculate daily 1 hr maximum concentrations by city from 2002 to 2011. Every 1 hr ozone concentration at each station was averaged to determine 1 hr ozone concentrations and daily 1 hr maximum concentrations for each city. The time scale used in the ozone trend analysis varied from 5 minutes to a day, but daily 1 hr maximum concentration was most widely used in the previous studies (Thompson et al., 2001). Also, according to the human health and air pollution related studies, acute exposal to ozone cause several health effects such as arterial vasoconstriction, heart rate decreasing, pulmonary function, mortality (Brook et al., 2002; Gold et al., 2000; Folinsbee et al., 1988; Touloumi et al., 1997). Therefore, daily 1 hr maximum concentration was used in this study for further analyses except for general statistical analysis. Meteorological data at the Regional Meteorological Observatory for each city and province were obtained from the Korea Meteorological Administration at the same resolution for the same time period. Meteorological factors of interest were daily maximum ambient temperatures (T_max), mid-day average relative humidity (RH), morning and afternoon average wind speeds (WS_m and WS_a), morning surface temperature differences (T_ds), deviation in temperature from a 10-year monthly mean at 850 mb (T_d850). These factors were the most important meteorological variables selected by statistical analysis between ozone concentration and various meteorological sub-parameters. LRT factors of interest were 24 hr HYSPLIT transport directions and distances (T_Dir and T_Dis) as shown in Table 1. Details on the related methods were described in Camalier et al. (2007). Some of these variables were directly observed, while others were calculated based on observed data or other parameters. In addition, transport directions and distances were determined based on the hybrid singleparticle lagrangian integrated trajectory (HYSPLIT) model simulations (Draxlerand Hess, 1997). The HYSPLIT model was run to calculate 24 hr backward trajectories from each surface station. The trajectories were started at noon LST at a height of 300 m (i.e., within the mixed layer). The statistical summary of the meteorological and LRT parameters are given in Table 2.

Fig. 1. 
The locations of each atmospheric monitoring station and a representative monitoring station in 16 cities and provinces.

To investigate the characteristics of ozone trends during a whole year including each season (spring, summer, fall and winter), the seasonal 5^th%, 33^th%, 50^th%, 67^th%, and 95^th% ozone concentration were considered. The ozone rate of change (ppb per year) over 2002-2011 was calculated separately for each ozone percentile through the simple linear regression. The trends were calculated with a reference year of 2002.

In previous studies, various statistical methods have been applied to analyze the long term variation of tropospheric ozone concentration such as time series decomposition, KZ-filter, GAM, DLM, and GLM (Shin et al., 2012; Development Core Team, 2009; Lee et al., 2008; U.S. EPA, 2002; R Schmidt, 2001; Cressie, 1993; Hastie et al., 1990). Because the long term trends of ozone is influenced by meteorological conditions, it is important to deduce the long term trends excluding changing meteorological factors. Many previous studies attempted to infer the actual fluctuations of ozone levels by substituting meteorological factors (Camalier et al., 2007; Zheng et al., 2007; Thompson et al., 2001; Davis et al., 1998; Bloomfield et al., 1996). GLM is a large class of statistical model for linear combination of expected variables including dependent variables and common errors. GLM explains the relationships between variables selected from urban ozone and broad-ranged meteorological variables (Camalier et al., 2007). Because meteorological variables that affect ozone concentration as well as wind directions and distances calculated from the HYSPLIT model are used in GLM modeling, GLM can possibly diminish the impact of meteorological factors or LRT if these values are adjusted. In this study, these characteristics of GLM were used to investigate factors influencing the ozone trends.

GLM models used to evaluate meteorological adjusted ozone trends are generally expressed as follows:

Where, “i” represents the highest daily ozone concentration in an 8-hour period, “j” each meteorological factor, and “k” the year of the measurement. “β₀” represents the average value of total ozone, “f” the smoothing function related to “j”, “W_d” days in the week, and “Y_k” the annual average concentration of ozone during the year “k” that takes into account meteorological conditions. “g” is the connecting function that represents the correlations between all variables on the right-hand side of the equation.

The meteorological factor is known to play an important role in long-term fluctuation of pollution levels (Kong et al., 2006; Wise and Comrie, 2005; Dueñas et al., 2002; Abdul-Wahab et al., 1996; Cox and Chu, 1993). In particular, concentrations of ozone formed through photochemical reactions fluctuate significantly depending on meteorological factors such as temperature, wind direction, and stability. If only ozone level is taken into account, periodic fluctuations due to meteorological factors, LRT, and precursor emissions will comprehensively impact the long term trends. Since GLM can be used to selectively remove meteorological factors and movement in air parcels in the long term trends, and only consider the impact of LRT or precursor emissions, GLM was used to analyze the long term trends in ozone concentration.

As shown in Fig. 1, there are many sampling sites in each city and province. To characterize the long term trends of ozone by area, it is difficult to deal with all data in individual monitoring station. Thus, representative monitoring station was selected and their data was used for further trend analysis of each city and province. Among atmospheric monitoring stations in each cities and province, stations that have sufficient data with data continuity for the period from 2002 to 2011 were selected, and ozone concentration data collected from the stations in each city and province were analyzed using KZ-filter, correlation matrix, cluster analysis, and Kriging method for selecting representative monitoring station.

Firstly, KZ-filter was used to calculate annual changes in each monitoring station after eliminating seasonal fluctuations and random variables. Annual changes, which were extracted by KZ-filter, in each monitoring station were compared with those of other monitoring stations within same city and province to check similarities with other stations by correlation analysis. And the result was displayed in a correlation matrix. Additionally, to examine the spatial distribution of ozone annual changes, Kriging method was applied. Also, cluster analysis is used for identifying the homogenous groups of monitoring stations based on their similarity of annual changes. Along with the results from cluster analysis, correlation matrix, and Kriging method, representative monitoring stations, where correlation coefficient was high in the largest cluster group with spatially homogeneity, in each city and province were selected.

Among the 16 cities and province, the process of selecting representative monitoring stations in Seoul is detailed in Shin et al. (2015). Summary of the methodology used for selecting the station in Seoul is as follows. A total of 16 monitoring stations that had accumulated data over 10 years were selected out of the 25 monitoring stations in Seoul. In order to select representative monitoring stations, KZ-filter was applied in calculating the annual changes in each monitoring station from the time-series data composed of complicated cycless after excluding seasonal changes and random variables from short-term fluctuations. After calculation of the actual long term fluctuation patterns using the KZ-filter method, the correlation among each monitoring station was analyzed and compared to each other. The correlation between monitoring stations was somewhat high, ranging from 0.52-0.87. The highest correlation was observed at Sinjeong monitoring station (0.87) followed by Sadang (0.86) and Beon (0.82). Together with the correlation matrix, cluster analysis can be used to select representative monitoring stations by grouping several monitoring stations in each region. Based on the cluster analysis, 16 monitoring stations in Seoul were grouped into 4 clusters. Kriging method linearly combines actual values for interpolation, and when predicting values, the method reflects not only the deviation from the actual values but also the correlation coefficient with each neighboring value. Therefore, the method was used to select representative monitoring stations based on the interpretation of the spatial distribution of ozone concentrations. The results are similar to those of correlation matrix calculated by applying KZ-filter as well as the cluster analysis. Correlations among Sinjeong, Bangi, and Beon showed the highest spatial distribution when the Kriging method was applied.

The various methods applied to select representative monitoring stations in Seoul show similar results. The result of correlation matrix shows high correlations of Sinjeong, Sadang, and Beon. Based on the cluster analysis, 16 monitoring stations in Seoul were grouped into 4 clusters. Sinjeong, Sadang, and Beon formed a cluster. Similar to the above-mentioned two methods, the Kriging method again shows high correlation of Sinjeong, Bangi, and Beon.

The study found that the station at Sinjeong represented the long term trends of ozone better than 15 other monitoring stations. For other cities and province, the same procedure was applied and some of the results are shown in Fig. S1 and Fig. S2. Table 3 shows representative monitoring stations selected along with the correlation with other stations within each city and province after applying the above-mentioned methods to the monitoring stations in the 16 cities and province. Fig. 1 also shows locations of representative atmospheric monitoring stations in 16 cities and province.

Annual average ozone concentrations calculated by hourly ozone concentrations including night time data in the 16 cities and provinces are given in Table 4. As shown in Table 4, the annual average concentration of ozone in 16 cities ranged between 0.014 ppm and 0.043 ppm for the period with concentration increasing from 0.022 ppm in 2002 to 0.025 ppm in 2011. The 10 yearaverage ozone concentration was the lowest at 0.018 ppm and the highest at 0.036 ppm in Seoul and Jeju, respectively. Jeju is the background area surrounded by sea with lowest population density and vehicle registration, whereas Seoul is the biggest city with the largest vehicle registrations in Korea (KOSIS, 2014). Thus, it is assumed that the ozone concentrations in Seoul are the lowest because of NO_x titration. The spatial distributions of ozone percentile concentration for 10 year are shown in Fig. 2. For 5^th%, 50^th%, 95^th% ozone concentration, Jeju shows the highest concentration among all other area. For 5^th% and 50^th% ozone, south coastal areas show relatively high ozone concentration. However, for 95^th% ozone, inland areas show relatively high ozone concentration as well as south coastal area. Seoul metropolitan areas show relatively lower ozone concentration than other area. NO_x emissions for Seoul and Jeju for 2001 are shown in Table 5. NO_x emissions in Seoul are 4.3-10.0 times higher than those in Jeju. As mentioned above, it is assumed that the NO_x titration cause the ozone concentration difference between two sites.

Fig. 2. 
The spatial distribution of 5^th%, 50^th%, and 95^th% ozone concentration for 2002-2011.

The national air quality standards for ozone are 0.06 ppm for 8 hours (99 percentile) and 0.10 ppm for 1 hour (999 per mill) in Korea. Considering the annual average ozone concentrations (0.007-0.026 ppm) in the UK in 2012 and the daily maximum ozone concentration (0.024 ppm) in Spain for 10 years (2001-2010), ozone concentrations are relatively high in Korea (Santurtún et al., 2015; DEFRA). However, ozone concentrations in Korea are slightly lower than those in China. In previous studies, surface ozone concentration varied from 0.016 to 0.053 ppm (monthly median) and from 0.018 to 0.047 ppm in eastern China (1991-2006) and southern China (1994-2007), respectively (Wang et al., 2009; Xu et al., 2008). Long term variations of ozone for each city and province of the whole year and each season by each percentile are shown from Fig. 3 to Fig. 7. And, ozone rate of changes are listed up in Table 6. As shown in Fig. 3 and Table 6, ozone rate of changes for the whole year are -0.04 (IC) - 1.55 (SE) ppb yr^-1 for 95^th% ozone, -0.14 (JN) - 1.12 (SE) ppb yr^-1 for 67^th% ozone, -0.16 (JN) - 0.95 (GJ) ppb yr^-1 for 50^th% ozone, -0.07 (JN) - 0.83 (JJ) ppb yr^-1 for 33^th% ozone, -0.19 (GW) - 0.64 (JJ) ppb yr^-1 for 5^th% ozone. Considering the whole year ozone concentration changes, Seoul has the highest increase ratio for higher ozone concentration (95^th%, 67^th%). On the other hand, IC is the lowest increasing area for the highest ozone concentration (95^th%) and JN is the lowest increasing area for mid-range ozone concentration (67^th%, 50^th%, and 33^th%). JJ is the highest increasing area for lower ozone concentration (33^th% and 5^th%).

Fig. 3. 
Long term trends of percentile ozone of whole year in each city and province (dark pink: 5^th%; blue: 33^th%, green: 50^th%, purple: 67^th%, sky blue: 95^th%).

The changing ratio of ozone is from -0.13 ppb yr^-1 (IC) to 1.87 ppb yr^-1 (JJ) in spring, from -0.24 ppb yr^-1 (JN) to 1.79 ppb yr^-1 (SE) in summer, from -0.02 ppb yr^-1 (CB) to 1.58 ppb yr^-1 (JB) in fall, and from -0.82 ppb yr^-1 (DJ) to 1.01 ppb yr^-1 (JJ) in winter for 95^th% ozone concentrations (Figs. 4-7, Table 6). For 67^th% ozone concentrations, the changing ratio of ozone is from 0.02 ppb yr^-1 (IC) to 1.77 ppb yr^-1 (SE) in spring, from -0.18 ppb yr^-1 (JN) to 1.38 ppb yr^-1 (DJ) in summer, from -0.30 ppb yr^-1 (JN) to 0.89 ppb yr^-1 (US) in fall, and from -0.35 ppb yr^-1 (JN) to 0.93 ppb yr^-1 (JJ). For 50^th% ozone concentrations, the changing ratio of ozone is from 0.03 ppb yr^-1 (IC) to 1.81 ppb yr^-1 (SE) in spring, from -0.18 ppb yr^-1 (JN) to 1.19 ppb yr^-1 (DJ) in summer, from -0.27 ppb yr^-1 (JN) to 0.84 ppb yr^-1 (GJ) in fall, and from -0.32 ppb yr^-1 (JN) to 0.88 ppb yr^-1 (JJ). For 33^th% ozone concentrations, the changing ratio of ozone is from 0.28 ppb yr^-1 (JN) to 1.59 ppb yr^-1 (SE) in spring, from -0.11 ppb yr^-1 (JN) to 0.97 ppb yr^-1 (DJ) in summer, from -0.28 ppb yr^-1 (CB) to 0.80 ppb yr^-1 (JJ) in fall, and from -0.36 ppb yr^-1 (CB) to 1.01 ppb yr^-1 (JJ). For 5^th% ozone concentrations, the changing ratio of ozone is from -0.03 ppb yr^-1 (DJ) to 1.29 ppb yr^-1 (JJ) in spring, from -0.16 ppb yr^-1 (DG) to 0.58 ppb yr^-1 (JJ) in summer, from -0.27 ppb yr^-1 (CN) to 0.81 ppb yr^-1 (JJ) in fall, and from -0.22 ppb yr^-1 (CN) to 0.70 ppb yr^-1 (JJ). According to Xu et al. (2008), long term trend of surface ozone in eastern China for 25 years (1991-2006, -0.79--0.33 ppb yr^-1) has been decreasing at a moderate rate since the early 1990s. However, the ozone concentration increased by an average rate of 0.55 ppb yr^-1 in the southern China (1994-2007) (Wang et al., 2009). Even though long term trend of ozone are varied in China due to different status of economic development and related precursor emission within China, it was recognized that the precursor emission cause the changing of ozone long term trend. According to Cooper et al. (2012), spring season ozone trends above western North America for 1995-2011 (3-8 km above sea level) was -0.03-1.29 for 95^th %, 0.15-0.71 for 67^th%, 0.14-0.68 for 50^th%, 0.12-0.66 for 33^th%, -0.11-0.65 for 5^th%. Considering this result, the ozone increasing ratio in the Korean Peninsula is higher than those in higher ozone areas in the US even though there are some differences by area. Also, considering the area with the highest increasing ratio (1.29-1.87 ppb yr^-1) for each percentile ozone concentration in spring, increasing ratio of ozone in the Korean Peninsula are higher than those reported for downwind of Asia (0.29-0.97 ppb yr^-1) reported by Cooper et al. (2010).

Fig. 4. 
Long term trends of percentile ozone of spring in each city and province (dark pink: 5^th%; blue: 33^th%, green: 50^th%, purple: 67^th%, sky blue: 95^th%).

Fig. 5. 
Long term trends of percentile ozone of summer in each city and province (dark pink: 5^th%; blue: 33^th%, green: 50^th%, purple: 67^th%, sky blue: 95^th%).

Fig. 6. 
Long term trends of percentile ozone of fall in each city and province (dark pink: 5^th%; blue: 33^th%, green: 50^th%, purple: 67^th%, sky blue: 95^th%).

Fig. 7. 
Long term trends of percentile ozone of winter in each city and province (dark pink: 5^th%; blue: 33^th%, green: 50^th%, purple: 67^th%, sky blue: 95^th%).

In most areas (SE, DG, IC, DJ, GG, GW, CB, CN, and JB), ozone concentration increasing is the highest in summer, and other areas (GJ, JN, GB, and JJ) show the highest ozone increasing in spring for 95^th% ozone. For 67^th% ozone, the highest ozone increasing trends are shown in summer with ratio (0.23-1.38) in IC, DJ, US, GG, GW, CB, CN, and JB. In all other areas except for BS, US and GN, the highest ozone increasing are shown in spring at 0.36-1.77. For 50^th%, 33^th%, 5^th% ozone, the highest increasing ratio for each city and province in spring are absolutely higher (50^th% 0.26-1.81; 33^th% 0.28-1.59; 5^th% 0.01-1.29) than those of other seasons (50^th% 0.32-1.19; 33^th% 0.97; 5^th% -0.02-0.43).

The factors affecting the trends of ozone concentration in each region were investigated using the results of the GLM. All factors (T_max, RH, WS_m, WS_a, T_ds, T_d850, T_Dir, and T_Dis) used in the GLM have impact on the ozone concentration. However, the main factors affecting the trends of ozone concentration are different in each region due to different characteristics of ozone formation or transport related with considering factors. This study chose top 2 factors, which is highly correlated with ozone concentration changes, determined by F-statistics in each region. Fig. 8 shows the main 2 factors affecting on the trends of ozone concentration in the 16 cities and province. The highest correlating factor with ozone in each city and province is shown in Fig. 8(a). Pink and yellow represent the region where the highest correlation factor is T_max and LRT related factors such as T_Dir and T_Dis, respectively. The second highest correlation factor with ozone in each city and province is shown in Fig. 8(b). Pink, yellow, green, and blue represent the region where the second highest correlating factor is T_max, transport related factors (T_Dir or T_Dis), RH, and wind speed (WS_m or WS_a), respectively.

Fig. 8. 
The main influence factors affecting on the trends of ozone concentration; (a) the highest correlating factor, (b) the second highest correlating factor.

In the case of the Seoul metropolitan area (SMA), which consists of Seoul, Incheon, and part of Gyeonggi province surrounding Seoul and Incheon, maximum temperature was the biggest factor affecting ozone concentration, and LRT related factors such as transport direction and transport distance was the biggest factor in other cities and province. The second biggest factor was relative humidity in Seoul and Incheon, LRT related factors in Gyeonggi, and relative humidity, maximum temperature, and wind speed in all other cities and province.

This research calculated only the annual changes in ozone concentrations excluding meteorological fluctuations and long range transport fluctuations by stepwise applying the meteorological factors and LRT related factors by GLM. The meteorologically adjusted and LRT related factors adjusted long term trends of ozone in cities and province are shown in Fig. 9. Fig. 9 shows the annual changes in ozone concentrations (RAW) at the atmospheric monitoring stations in the 16 cities and province along with meteorologically and long range transport adjusted (MET&LRT adjusted) and only meteorologically adjusted (MET adjusted) datasets to investigate the effects of precursor emissions and LRT. The differences between ozone concentration (RAW) and adjusted concentrations were -7.6% and -6.2% as the largest difference in 2003 among 10 years for MET&LRT adjusted and MET adjusted concentration for Seoul, respectively. In the other cities and province, the differences between RAW and adjusted concentration (MET&LRT adjusted and MET adjusted) were relatively high in 2003. Year 2003 shows, excluding the effect of meteorological factors in almost regions, that ozone concentration increased in 11 city and province among all 16 areas, which leads to the conclusion that meteorological factors contributed to the decrease of ozone concentration. As such, anomaly detection of temperature and humidity, meteorological factors that have a high impact on ozone concentration, is shown in Fig. 10. Each anomaly is calculated as the difference between the average for the given year and the previous 9-year average. Negative value and positive value are lower and higher than average, respectively. Ambient temperature and relative humidity in year 2003 is lower and higher than those for previous 9-year, respectively. On the other hand, ambient temperature and relative humidity in year 2004 is higher than those for previous 9-year. Compared to year 2004, year 2003 showed considerably lower temperature (negative) and higher relative humidity (positive) nationwide. According to Camalier et al. (2007), 1% increase in relative humidity leads to a decrease in ozone by 0.5-1.5 ppb. When relative humidity is high, the excited singlet (O(¹D)) oxygen atom, which is formed by ozone photolysis, produce more OH radical (Seinfeld and Pandis, 2006). The reactions of OH radical with NO₂ are higher than those with VOCs in VOCs-limited region such as Seoul (Shin et al., 2013). The OH-NO₂ reaction removes OH radicals from VOC oxidation cycle. Thus, the further ozone formation of ozone is retarded. Therefore, it can be inferred that the ozone formation was deterred by photochemical reaction due to low temperatures and high humidity, which has contributed to the decrease in ozone concentration.

Fig. 9. 
The meteorologically adjusted and meteorologically & long range transport adjusted long term trends of ozone in each city and province.

Fig. 10. 
Temperature and relative humidity anomalies for ozone seasons of 2003 and 2004.