Extensive Spatiotemporal Analyses of Surface Ozone and Related Meteorological Variables in South Korea for the Period 1999–2010

Spatiotemporal characteristics of surface ozone (O 3) variations over South Korea are investigated with consideration of meteorological factors and timescales based on the Kolmogorov–Zurbenko filter (KZ filter), using measurement data at 124 air quality monitoring sites and 72 weather stations for the 12 yr period of 1999–2010. In general, O 3 levels at coastal cities are high due to dynamic effects of the sea breeze while those at the inland and Seoul Metropolitan Area (SMA) cities are low due to the NO x titration by local precursor emissions. We examine the meteorological influences on O 3 using a combined analysis of the KZ filter and linear regressions between O 3 and meteorological variables. We decomposed O 3 time series at each site into short-term, seasonal, and long-term components by the KZ filter and regressed on meteorological variables. Impact of temperature on the O 3 levels is significantly high in the highly populated SMA and inland region, but low in the coastal region. In particular, the probability of high O 3 occurrence doubles with 4 • C of temperature increase in the SMA during high O 3 months (May–October). This implies that those regions will experience frequent high O 3 events in a future warming climate. In terms of short-term variation, the distribution of high O 3 probability classified by wind direction shows the effect of both local precursor emissions and long-range transport from China. In terms of long-term variation, the O 3 concentrations have increased by +0.26 ppbv yr −1 (parts per billion by volume) on nationwide average, but their trends show large spatial variability. Singular value decomposition analyses further reveal that the long-term temporal evolution of O 3 is similar to that of nitrogen dioxide, although the spatial distribution of their trends is different. This study will be helpful as a reference for diagnostics and evaluation of regional-and local-scale O 3 and climate simulations, and as a guide to appropriate O 3 control policy in South Korea.

Abstract.Spatiotemporal characteristics of surface ozone (O 3 ) variations over South Korea are investigated with consideration of meteorological factors and timescales based on the Kolmogorov-Zurbenko filter (KZ filter), using measurement data at 124 air quality monitoring sites and 72 weather stations for the 12 yr period of 1999-2010.In general, O 3 levels at coastal cities are high due to dynamic effects of the sea breeze while those at the inland and Seoul Metropolitan Area (SMA) cities are low due to the NO x titration by local precursor emissions.We examine the meteorological influences on O 3 using a combined analysis of the KZ filter and linear regressions between O 3 and meteorological variables.We decomposed O 3 time series at each site into short-term, seasonal, and long-term components by the KZ filter and regressed on meteorological variables.Impact of temperature on the O 3 levels is significantly high in the highly populated SMA and inland region, but low in the coastal region.In particular, the probability of high O 3 occurrence doubles with 4 • C of temperature increase in the SMA during high O 3 months (May-October).This implies that those regions will experience frequent high O 3 events in a future warming climate.In terms of short-term variation, the distribution of high O 3 probability classified by wind direction shows the effect of both local precursor emissions and long-range transport from China.In terms of long-term variation, the O 3 concentrations have increased by +0.26 ppbv yr −1 (parts per billion by volume) on nationwide average, but their trends show large spatial variability.Singular value decomposition analyses further reveal that the long-term temporal evolution of O 3 is similar to that of nitrogen dioxide, although the spa-tial distribution of their trends is different.This study will be helpful as a reference for diagnostics and evaluation of regional-and local-scale O 3 and climate simulations, and as a guide to appropriate O 3 control policy in South Korea.

Introduction
Surface ozone (O 3 ) is a well-known secondary air pollutant, which affects air quality, human health, and vegetation.High O 3 concentrations have detrimental effects on respiration, lung function, and airway reactivity in human health (Bernard et al., 2001;Bell et al., 2007).In terms of mortality, Levy et al. (2005) have previously assessed that a 10 ppbv (parts per billion by volume) increase in 1 h maximum O 3 could increase daily mortality by 0.41 %.High O 3 concentrations could also reduce agricultural production.For example, Wang and Mauzerall (2004) reported that the East Asian countries of China, Japan, and South Korea lost 1-9 % of their yields of wheat, rice, and corn, and 23-27 % of their yields of soybeans due to O 3 in 1990.In addition, O 3 is one of the greenhouse gases for which radiative forcing is estimated as the third largest contribution among the various constituents in the troposphere (IPCC, 2007).Therefore, the spatially inhomogeneous distribution of O 3 due to its short chemical lifetime of a week to a month could induce strong regional-scale climate responses (Mickley et al., 2004).
In the Northern Hemisphere midlatitudes where population, industry, and transportation are concentrated, the background O 3 levels increased during the late-20th century due Published by Copernicus Publications on behalf of the European Geosciences Union.
to increases in anthropogenic precursors particularly nitrogen oxides (NO x ), but its trends show regional and temporal differences (Oltmans et al., 1998;Guicherit and Roemer, 2000;Vingarzan, 2004).Although the increasing trends of O 3 in Europe, the North Atlantic, North America, and Japan have flattened over the past decade (Oltmans et al., 2006(Oltmans et al., , 2013)), there have still been concerns about elevated O 3 concentrations in China owing to rapid economic growth and industrialization (Ding et al., 2008;Tang et al., 2009;Wang et al., 2009Wang et al., , 2012)).Such recent increases of O 3 in China can affect the regional background O 3 levels in East Asia by transboundary transport of O 3 and its precursors.For example, previous modeling studies have shown that the transport of O 3 from China by continental outflow is one of the major contributions of O 3 in Japan and South Korea (Tanimoto et al., 2005;Nagashima et al., 2010).Intercontinental transport of O 3 and its precursors originated in East Asia affects O 3 concentrations and air quality in remote areas even on a global scale (Akimoto, 2003).Experiments using a global climate-chemistry model with future emission scenarios by Lei et al. (2013) suggest that the increase in East Asian emissions will still be an important issue for the O 3 air quality in both East Asia and the United States.
Recently, several studies have focused on the relationship between O 3 levels and temperature, and suggested potential influences of the global warming and climate change on the high levels of O 3 (Jacob and Winner, 2009;Rasmussen et al., 2012, and references therein).Lin et al. (2001) calculated probability of daily maximum 8 h average O 3 exceeding 85 ppbv for a given range of daily maximum temperatures and reported that a 3 • C increase of the daily maximum temperature doubles risk of O 3 exceedances in the northeastern United States.In addition, Ordóñez et al. (2005) showed that high temperature extremes probably led to the high occurrence of severe O 3 episodes during the summer 2003 heat wave over Europe.These results imply the potentially large sensitivity of O 3 concentration and related air quality to the temperature increases (Jacob and Winner, 2009).In the model experiments by Lin et al. (2008), both averaged O 3 concentrations and frequencies of high O 3 episodes were predicted to increase in future over the United States and East Asia.Recent global climate-chemistry model experiments by Lei and Wang (2014) also implied the future O 3 increases in industrial regions due to more O 3 production by photochemical reactions and less O 3 removal by nocturnal odd nitrogen (NO y ) chemistry in warmer conditions.
In South Korea, one of the most densely populated countries in the world, both O 3 concentrations and high O 3 episodes have increased in recent decades despite efforts to regulate emissions of O 3 precursors (KMOE, 2012).Although the increase of O 3 levels in South Korea over the last 3 decades is mainly regarded as the results of rapid industrialization, economic expansion, and urbanization, there are other factors to be considered to explain the long-term increase in O 3 concentrations.For example, since the Ko-rean Peninsula is located on the eastern boundary of East Asia, downward transport of O 3 by the continental outflow considerably affects the high O 3 levels in South Korea (Oh et al., 2010).In addition, a recent warming trend related to global climate change could also be an important factor to increase O 3 concentrations in South Korea.Climate change is expected to increase both frequency and intensity of temperature extremes over the Korean Peninsula (Boo et al., 2006).Therefore, comprehensive understanding of the various factors affecting O 3 concentrations, such as local precursor emissions, transport of O 3 and its precursors from local and remote sources, and changes in meteorological fields related to climate change is required to guide environmental policies.
The present study aims to examine the spatiotemporal characteristics of the measured O 3 variations in South Korea with consideration of three timescales and various meteorological factors, using ground-measured data from 124 air quality monitoring sites and 72 weather stations for the 12 yr period of 1999-2010.We decomposed O 3 time series at each measurement site into different timescales of shortterm, seasonal, and long-term components by application of the Kolmogorov-Zurbenko filter (KZ filter) that has been used in previous studies (e.g., Gardner and Dorling, 2000;Ibarra-Berastegi et al., 2001;Thompson et al., 2001;Lu and Chang, 2005;Wise and Comrie, 2005;Tsakiri and Zurbenko, 2011;Shin et al., 2012).To investigate the meteorological impact on the O 3 levels, we applied the combined analysis of the KZ filter and linear regression model with the meteorological variables.In the short-term timescale, the possible effects of transport from the local and remote sources on the high O 3 episodes were explored by using the wind data.In the long-term timescale, the singular value decomposition (SVD) with nitrogen dioxide (NO 2 ) measurements was additionally applied to examine the effects of varying local emissions on the long-term O 3 trend.
The remainder of this paper is structured as follows.In the next section, we describe the observational data and analysis techniques used in this study.In Sect.3, we investigate the spatiotemporal characteristics of the decomposed O 3 time series and its relationship with meteorological variables over South Korea.Finally, the key findings are summarized in Sect. 4.

Data
The National Institute of Environmental Research (NIER) of South Korea provides hourly data of O 3 and NO 2 mixing ratios in the ppbv unit, which have been measured by ultraviolet absorption and chemiluminescence, respectively.We here select 124 urban air quality monitoring sites over South Korea, based on data availability for the period 1999-2010, and analyze hourly time series of O 3 and NO 2 from each site.It is noted that our current analyses exclude other data from roadside measurement sites where data can be directly affected by the vehicle exhaust emissions and suburban and background sites located around South Korea.Hourly meteorological data at 72 weather stations of the Korea Meteorological Administration (KMA) for the same period are also used to examine the effects of meteorological factors on the O 3 variations.The meteorological variables used in this study include common factors related to the O 3 variations such as temperature ( • C), surface insolation (W m −2 ), relative humidity (%), and wind speed (m s −1 ) (e.g., Ordóñez et al., 2005;Camalier et al., 2007;Jacob and Winner, 2009).Dewpoint temperature ( • C) and sea-level pressure (hPa) are additionally applied for multiple linear regression models as other previous studies have done (e.g., Thompson et al., 2001;Shin et al., 2012).Finally, wind direction (16 cardinal directions) is used to reveal its relationship with short-term changes in O 3 .Using the hourly data, we first calculated daily averages for O 3 (O 3 avg ), NO 2 (NO 2 avg ), temperature (T ), surface insolation (SI), dew-point temperature (TD), sea-level pressure (PS), wind speed (WS), wind direction (WD), and relative humidity (RH).We also obtained daily minimum O 3 (O 3 min ), daily maximum 8 h average O 3 (O 3 8 h ), and daily maximum temperature (T max ) from the hourly data set.
To investigate the relationship between O 3 and meteorological variables, it is desirable to use data observed at the same stations.However, not all air quality monitoring sites and weather stations are closely located in South Korea.Therefore, we assume that an air quality monitoring site can observe the same meteorological variables as those at a weather station if the distance between the two places is less than 10 km.Under this assumption, only O 3 data from 72 air quality monitoring sites and meteorological data from 25 weather stations are available to analyze the meteorological effects on the O 3 variation over South Korea.The insolation was measured only at 17 weather stations for the analysis period.Figure 1 shows geographical locations of the ground measurements used in the present study, together with colored topography based on the US Geological Survey (USGS) digital elevation model (DEM).

Decomposition of O 3 time series by KZ filter
The KZ filter is a decomposition method that can be used to separate time series into short-term, seasonal, and long-term components (Rao and Zurbenko, 1994).We applied the KZ filter to the O 3 time series by taking the moving average of window length m with iterating p times, which is denoted by KZ m,p .The KZ filter is basically a low-pass filter for removing high-frequency components from the original time series.Following Eskridge et al. (1997), the KZ filter removes the signal smaller than the period N, which is called as the ef-fective filter width.N is defined as follows: The KZ-filter method has the same level of accuracy as the wavelet transform method although it is a much easier way to decompose the original time series (Eskridge et al., 1997).In addition, time series with missing observations can be applicable to the KZ filter owing to the iterative moving average process.
For the clear separation of the components, we applied the KZ filter to the daily log-transformed O 3 as in Rao and Zurbenko (1994) and Eskridge et al. (1997), instead of the raw O 3 concentrations.While the short-term component separated by the KZ filter using raw O 3 data still shows clear seasonality, use of ln(O 3 ) makes the short-term component stationary and nearly independent of the seasonal influence by stabilizing variance (Rao and Zurbenko, 1994;Rao et al., 1997).Note that a temporal linear trend of log-transformed data is provided as percent per year (% yr −1 ) because the differential of the natural logarithm is equivalent to the percentage change.
The natural logarithm of the O 3 time series at each site denoted as [O 3 ](t) is thus decomposed by the KZ filter as follows: (2) [O 3 ST ] is a short-term component attributable to day-to-day variation of synoptic-scale weather and short-term fluctuation in precursor emissions.[O 3 SEASON ] represents a seasonal component related to the seasonal changes in solar radiation and vertical transport of O 3 from the stratosphere whose timescale is from several weeks to months.[O 3 LT ] denotes a long-term component explained by changes in precursor emission, transport, climate, policy, and economy over the entire period (Rao et al., 1997;Milanchus et al., 1998;Gardner and Dorling, 2000;Thompson et al., 2001;Wise and Comrie, 2005).Tsakiri and Zurbenko (2011) showed that [O 3 ST ] and [O 3 LT ] are independent of each other.Also, statistical characteristics of [O 3 ST ] are very close to those of white noise (Flaum et al., 1996) and therefore [O 3 ST ] is nearly detrended.
In this study, we used the KZ filter with the window length of 29 days and 3 iterations (KZ 29,3 ) following previous studies (e.g., Rao and Zurbenko, 1994) and decomposed daily ln(O 3 8 h ) time series at the 124 monitoring sites.KZ 29,3 removes the [O 3 ST ] with period shorter than about 50 days, following Eq.(1).We defined the filtered time series as a baseline ([O 3 BL ]) as in Eq. (3).
Equation ( 3) accounts for the multiplicative effects of shortterm fluctuations on the [O 3 BL ] due to the log transformation (Thompson et al., 2001).In other words, the exponential of  4) (Milanchus et al., 1998).
Since [O 3 BL ] is closely associated with meteorological fields, we built a multiple regression model with available meteorological variables as in Eq. ( 5), following previous studies (e.g., Rao and Zurbenko, 1994;Rao et al., 1995;Ibarra-Berastegi et al., 2001).Short-term variability of meteorological variables was also filtered out by KZ 29,3 .
In the multiple linear regression model, [O 3 BL ] is a response variable and the baselines of meteorological variables (MET BL (t) i ) are predictors.Also, a 0 , a i , and ε(t) denote the constant, regression coefficient of variable i, and residual of the multiple regression model, respectively.The residual term ε(t) contains not only the long-term variability of O 3 related to long-term changes in local precursor emissions but also seasonal variability of O 3 attributable to unconsidered meteorological factors in the multiple linear regression model.Thus, we applied the KZ filter with the window length of 365 days and 3 iterations (KZ 365,3 ) to ε(t) to extract the meteorologically adjusted [O 3 LT ] of which the period is larger than about 1.7 yr as follows: In Eq. ( 6), δ(t) denotes the seasonal variability of O 3 related to the meteorological variables unconsidered in the multiple linear regression model and/or noise.
Finally, [O 3 SEASON ] is obtained by sum of the combined meteorological variables regressed on [O 3 BL ] (a 0 + i a i MET BL (t) i ) and δ(t) as in Eq. ( 7).
Figure 2 shows a schematic representation of Eq. (2) using daily O 3 8h time series at the City Hall of Seoul for the period of 1999-2010.[O 3 SEASON ] in Fig. 2c clearly shows the typical seasonal cycle of O 3 in South Korea with high concentrations in spring, slight decrease in July and August, and increase in autumn (Ghim and Chang, 2000).The spring maximum of O 3 concentration in the Northern Hemisphere is generally attributed to episodic stratospheric intrusion (Levy et al., 1985;Logan, 1985), photochemical reactions of accumulated NO x and hydrocarbons during the winter (Dibb et al., 2003), accumulation of O 3 due to the longer photochemical lifetime (∼ 200 days) during the winter (Liu et al., 1987), and transport of O 3 and its precursors by the continental outflow (Carmichael et al., 1998;Jacob et al., 1999;Jaffe et al., 2003).However, frequent precipitation during the East Asian summer monsoon influences the decrease of O 3 concentrations in July and August (Ghim and Chang, 2000).

Spatial interpolation by AIDW method
Inverse-distance weighting (IDW) is a deterministic spatial interpolation technique for spatial mapping of variables distributed at irregular points.In this study, we adopted the enhanced version of the IDW, the adaptive inverse-distance weighting (AIDW) technique (Lu and Wong, 2008).While the traditional IDW uses a fixed distance-decay parameter without considering the distribution of data within it, the AIDW uses adjusted distance-decay parameters according to density of local sampling points.Therefore, the AIDW provides flexibility to accommodate variability in the distancedecay relationship over the domain and thus better spatial mapping of variables distributed at irregular observational points (Lu and Wong, 2008).
In the mapping of O 3 with spatial interpolation, there are ubiquitous problems such as spatial-scale violations, improper evaluations, inaccuracy, and inappropriate use of O 3 maps in certain analyses (Diem, 2003).The spatial mapping in the present study also has problems with the spatial resolution of the observations, which is not high enough to consider small-scale chemical processes and the geographical complexity of the Korean Peninsula (see Fig. 1).Most of the air quality monitoring sites are concentrated on the cities, and typical intercity distances are 30-100 km in South Korea while spatial representativeness of O 3 concentrations is possibly as small as around 3-4 (Tilmes and Zimmermann, 1998) or 5 km (Diem, 2003).In addition, mapping with a few monitoring sites combined with complex mountainous terrain can also distort the actual distribution of data, especially in the northeastern part of South Korea.Despite such limits, the spatial mapping in this study is still useful because we aim not to derive an exact value at a specific point where the observation does not exist, but to provide the better quantitative understanding of O 3 and related factors in South Korea, especially focused on the metropolitan and urban areas.

Spatial characteristics of O 3 and its trend in South Korea
Climatological daily average O 3 (O 3 avg ) and its temporal linear trends are represented in Fig. 3 and Table 1 using data from 124 monitoring sites distributed nationwide in 46 cities for the past 12 yr period.The spatial map of climatological daily average NO 2 (NO 2 avg ) is also shown in Fig. 3.
In Table 1, the cities are categorized into three geographical groups: 16 coastal cities, 14 inland cities, and 16 cities in the Seoul Metropolitan Area (SMA).We separated the SMA cities from the other two groups since the SMA is the largest source region of anthropogenic O 3 precursors in South Korea.The SMA occupies only 11.8 % (11 745 km 2 ) of the national area, but has 49 % (25.4 million) of the total population and 45 % (8.1 million) of total vehicles in South Korea.
It is estimated that approximately 27 % (0.29 Mt) of total NO x emissions and 34 % (0.30 Mt) of the volatile organic compound (VOC) emissions in South Korea were from the SMA in 2010 (KMOE, 2013).Therefore, the climatological NO 2 avg concentration in the SMA is much higher than that in other regions (Fig. 3b).
In general, O 3 concentrations are high at the coastal cities, low at the inland cities, and lowest at the SMA cities in South Korea.Along with Table 1, Fig. 3a shows that the 12 yr aver-age of O 3 avg is high at the southern coastal cities such as Jinhae (31.3 ppbv), Mokpo (30.3 ppbv), and Yeosu (28.1 ppbv), with the highest value at Jeju (32.6 ppbv), and low at the inland metropolitan cities such as Daegu (19.8 ppbv), Gwangju (20.5 ppbv), and Daejeon (20.7 ppbv), with lowest values at the SMA cities including Seoul (17.1 ppbv), Incheon (19.0 ppbv) and Anyang (16.8 ppbv).
Compared to the regional background concentration of 35-45 ppbv at five background measurement sites around South Korea (KMOE, 2012), the averaged O 3 concentrations in the SMA and inland metropolitan cities are much lower while those at the coastal cities are close to the regional background levels.In comparison with Fig. 3b, Fig. 3a shows that relatively lower O 3 avg regions are well consistent with relatively higher NO 2 avg regions.Substantial emissions of anthropogenic NO in the SMA and other inland metropolitan cities lead to NO x titration effects even in the absence of photochemical reactions during the night, and thus the averaged O 3 concentrations are depressed by 10-20 ppbv lower than the regional background concentration (Ghim and Chang, 2000).A recent modeling study by Jin et al. (2012) has suggested that the maximum O 3 concentrations in the SMA, especially in Seoul and Incheon, are VOC-limited.In the coastal region, however, low emissions of NO with dilution by the strong winds weaken the titration effect and result in the high O 3 concentrations.The dynamic effect of land-sea breeze is another possible factor of the high O 3 levels at the coastal cities. Oh et al. (2006) showed that a near-stagnant wind condition at the development of sea breeze temporarily contains O 3 precursors carried by the offshore land breeze    during the night, and following photochemical reactions at midday produces O 3 .The relationship between O 3 and wind speed and direction will be shown in Sects.3.2 and 3.5, respectively.
In terms of temporal trends, the surface O 3 concentrations in South Korea have generally increased for the past 12 yr as shown in Fig. 3c and Table 1.The averaged temporal linear trend of O 3 avg at 46 cities nationwide is +1.15 % yr −1 (+0.26 ppbv yr −1 ), which is comparable with observed increasing trends of approximately +0.5-2 % yr −1 in various regions in the Northern Hemisphere (Vingarzan, 2004).Compared with previous studies in East Asia, the overall increasing trend of O 3 in South Korea is smaller than recent increasing trends over China of +1.1 ppbv yr −1 in Beijing for the period 2001-2006 (Tang et al., 2009) and +0.58 ppbv yr −1 in Hong Kong for the period 1994-2007 (Wang et al., 2009) but slightly larger than increasing trend over Japanese populated areas of +0.18 ppbv yr −1 for 1996-2005 (Chatani and Sudo, 2011).
Several factors that could influence the overall increase of O 3 over East Asia were suggested by the following previous studies.Recently, Zhao et al. (2013) have estimated that the NO x emissions in China increased rapidly from 11.0 Mt in 1995 to 26.1 Mt in 2010, mainly due to the fast growth of energy consumption.The NO x and VOC emissions in South Korea also increased in the early 2000s.The estimated anthropogenic NO x and VOC emissions are 1.10 and 0.74 Mt in 1999 but 1.35 and 0.87 Mt in 2006, respectively (KMOE, 2013).Tanimoto et al. (2009) suggested that the O 3 increase results from such recently increased anthropogenic precursor emissions in East Asia.
However, model sensitivity simulations in Chatani and Sudo (2011) indicate that the changes in East Asian emissions can explain only 30 % of the O 3 trend.They have suggested the long-term variations in meteorological fields as a possible important factor although further studies are required.In particular, it is well known that insolation and temperature are important meteorological factors in O 3 variation.While insolation directly affects O 3 production through photochemical reactions, increased temperature affects net O 3 production rather indirectly by increasing biogenic hydrocarbon emissions, hydroxyl radical (OH) with more evaporation, and NO x and HO x radicals by thermal decomposition of the peroxyacetyl nitrate (PAN) reservoir (Sillman and Samson, 1995;Olszyna et al., 1997;Racherla and Adams, 2006;Dawson et al., 2007).Therefore, the O 3 increasing trend in Fig. 3c is possibly affected by changes in meteorological variables.
Figure 4 shows temporal linear trends of daily average temperature (T ) and insolation (SI).Despite the spatial discrepancy between trends of O 3 (Fig. 3c) and meteorological variables (Fig. 4), both temperature and insolation have generally increased in South Korea for the past 12 yr.The spatial mean of the temporal linear trend in temperature at 72 weather stations nationwide is approximately +0.09 • C yr −1 , which is much higher than +0.03 • C yr −1 for the Northern Hemispheric land surface air temperature for the period 1979-2005(IPCC, 2007)).This highly increasing trend of temperature in South Korea is probably due to the urban heat island effect with rapid urbanization.The averaged temporal linear trend of insolation at 22 weather stations nationwide is about +1.47 W m −2 yr −1 despite the decreasing phase of solar cycle during the 2000s.This is possibly caused by the reduction in particulate matter emissions due to enhanced environment regulation in South Korea during the recent decade (KMOE, 2012).
Although O 3 and related meteorological variables such as temperature and insolation have recently increased in South Korea, the spatial patterns of their temporal trends do not show clear similarity.In addition, the spatial distribution of O 3 trends is rather inhomogeneous even on a metropolitan scale.For instance, Table 1 shows a wide range of O 3 trends among the SMA cities from −1.25 % yr −1 at Gwacheon to +2.82 % yr −1 at Seoul.The spatial inhomogeneity in O 3 trends and the trend differences among O 3 , temperature, and insolation imply that the long-term O 3 trends in South Korea are not only affected by changes in meteorological conditions but also influenced by changes in local precursor emissions or transport of O 3 and its precursors.The local effects of precursor emissions on the long-term changes in O 3 are further examined in Sect.3.6.

Relationships between O 3 and meteorological variables
A multiple linear regression model is here adopted to explain relationships between O 3 8h and each of key meteorological variables such as T max , SI, TD, PS, WS, and RH.To exclude day-to-day short-term fluctuations or white noises from the original time series, KZ 29,3 was applied to each variable before the regression process and yielded baselines of each variable.As a result of the multiple linear regression, coefficients of determination (R 2 ) between baselines of O 3 8h and each meteorological variable, as well as adjusted R 2 for the multiple linear regression models, were calculated for 72 air quality monitoring sites distributed in 25 cities nationwide and summarized in Table 2. Combined meteorological effects on O 3 variations in each city are represented by adjusted R 2 , which adjusts for the number of predictors with consideration of the degrees of freedom.The nationwide average of adjusted R 2 is 0.51 and that of R 2 is 0.50 for SI, 0.29 for PS, 0.22 for T max , 0.14 for TD, 0.05 for RH, and 0.03 for WS, respectively.In South Korea, SI, T max , and TD generally show positive correlations with O 3 levels while PS is negatively correlated with O 3 variations.Since the short-term variability in each variable is excluded, the negative correlation between O 3 and PS is related to their seasonal cycle rather than continuously changing weather system of high and low.PS in South Korea, located on the continental east coast, is mostly affected by the cold continental high pressure air mass during the winter, when the O 3 concentrations are lowest.However, WS and RH show weak correlations with O 3 variations.The spatial distributions of R 2 for T max and SI, as well as the adjusted R 2 for the combined meteorological effects, are represented in Fig. 5.Both Fig. 5a and b show a common spatial pattern with high correlations at the inland and SMA cities and low correlations at the coastal cities.For instance, the average R 2 value with T max for the coastal cities is only 0.07, which is much smaller than 0.36 for the SMA cities and 0.30 for the inland cities.Also, the average R 2 values with SI are 0.60 for the SMA cities and 0.58 for the inland cities, but 0.35 for the coastal cities.Despite the similar pattern between Fig. 5a and b, the R 2 values of SI are much higher than those of T max because temperature affects net O 3 production rather indirectly compared to the direct influence of insolation on O 3 levels by photochemical production (Dawson et al., 2007, and references therein).The apparent R 2 differences among the three regions indicate that temporal variations of O 3 at the SMA and inland cities are much more sensitive to SI and T max than those at the coastal cities.The low dependence of O 3 on T max and SI at the coastal cities means that the photochemical reactions of precursors are less important for determining O 3 levels there compared to the SMA and inland cities.
The meteorological effects on O 3 at the inland, coastal, and SMA cities are also examined by daily minimum O 3 (O 3 min ).As represented in Fig. 6a and Table 3, the O 3 min is high near the coast, low at the inland cities, and lowest in the SMA.In the polluted urban area, the O 3 concentration reaches near-zero minima during the night since O 3 is reduced by NO x titration, a nocturnal NO y chemical process related to nitrate formation, and dry deposition in the absence of photochemical production.However, in the coastal region where the NO x concentrations are low (Fig. 3b), the Table 2. Coefficients of determination (R 2 ) between the baseline of daily maximum 8 h average O 3 (O 3 8h ) and baselines of six meteorological variables (T max , SI, TD, PS, WS, and RH) at 25 cities over South Korea for the period 1999-2010.Adjusted R 2 (Adj.R 2 ) between the baseline of O 3 8h and combined meteorological effects (a 0 + i a i MET BL (t) i ) are also represented.The cities are categorized into three groups: 10 coastal cities, 11 inland cities, and 4 cities in the SMA.Numbers in bold fonts indicate correlations significant at the 95 % level or higher.lower titration and nitrate formation at nighttime lead to the higher O 3 min levels.In addition, transport of O 3 from the regional background could also keep high levels of O 3 during the night (Ghim and Chang, 2000).Frequency distributions of O 3 concentrations in previous studies suggested that O 3 levels at the coastal cities such as Gangneung, Jeju, Mokpo, Seosan, and Yeosu are affected by the background O 3 transport, unlike Seoul where the effect of local precursor emissions is dominant (Ghim and Chang, 2000;Ghim, 2000).Therefore, combined effects of the low NO x levels and trans-port of the regional background O 3 influence the high O 3 min near the coast.Compared to the spatial distribution of R 2 between baselines of O 3 8h and T max or SI (Fig. 5a, b), O 3 min distribution in Fig. 6a shows high O 3 min at the coastal cities where the R 2 is low and low O 3 min at the inland cities where the R 2 is high.These opposite patterns suggest that the meteorological effects on the O 3 production are negatively correlated with O 3 min for the South Korean cities.The clear negative correlations are also shown in scatter plots (Fig. 6b, c).In both scatter plots, the three geographical groups of cities (blue for  the coastal cities, green for the inland cities and red for SMA) are well separated.Several industrial or metropolitan cities in the coastal region such as Changwon (CW), Busan (BS), and Ulsan (US) have a relatively low O 3 min compared to the other coastal cities. Larger NO x emissions in these southeastern coastal cities (Fig. 3b) induce lower O 3 min levels via NO x titration and a nocturnal NO y chemical process.Among the SMA cities, however, Ganghwa (GH) has much higher O 3 min compared to other SMA cities. Ganghwa is a rural county located on the northwestern coast of the SMA.Therefore, both small NO x emissions there and transport of regional back-ground O 3 from the Yellow Sea affect the characteristics of O 3 in Ganghwa.

Cities
The different meteorological effects on O 3 between the coastal and inland regions are further examined with wind speed.Daily average WS data over South Korea are averaged for 12 yr.The 12 yr averaged WS is summarized in Table 3 and presented in a spatial map in Fig. 7a, which shows high wind speed in the coastal region and low wind speed in the inland region.Figure 7b and c show that the averaged wind speeds at 25 cities are positively correlated with O 3 min and negatively correlated with the R 2 between O 3 and T max .In general, surface mixing and ventilation by high wind speeds reduce the near-surface precursors and thus decrease the photochemical production of O 3 .Therefore the relationship between high wind speed and high O 3 levels in the coastal region is attributable to the transport of background O 3 .However, the weaker wind speed induces a more effective photochemical reaction through the longer reaction time in stagnant conditions as well as more enhanced aerodynamic resistance to dry deposition (Jacob and Winner, 2009).Therefore, the effects of insolation and temperature on the O 3 productions become more important in the inland region where the wind speeds are lower.

Probability of O 3 exceedances related to temperature
Evaluating the probability of O 3 exceeding the air quality standard in a given range of temperature is useful to speculate about the potential sensitivity of O 3 concentrations to climate change (Lin et al., 2001;Jacob and Winner, 2009).
Here we calculate the probabilities that O 3 8h exceeds the Korean air quality standard of 60 ppbv (KMOE, 2012) as a function of the daily maximum temperature (T max ) for the coastal, inland, and SMA cities.Similar to the analyses in Lin et al. (2001) for the contiguous United States, Fig. 8 shows that the probabilities of O 3 exceedances increase with T max at the inland and SMA cities.For example, the probability of O 3 exceedances in the SMA is almost doubled by an increase of about 4 • C in T max and reach 27 % at 30 • C. In the coastal region, however, the probability of O 3 exceedance increases up to 12-13 % with a T max change from 10 to 20 • C and does not increase significantly for T max above 20 • C.This is consistent with the spatial feature of the meteorological effects on O 3 levels, which are high at the inland and SMA cities and low at the coastal cities as described in the previous section.Therefore, the probability of high O 3 occurrence will be more sensitive to the future climate change at the inland and SMA cities than at the coastal cities.In the previous modeling study by Boo et al. (2006), T max over the Korean Peninsula is expected to rise by about 4-5 • C by the end of 21st century, due to global warming.This suggests considerable future increases in exceedances of the O 3 air quality standard over South Korea, except over coastal regions, in the absence of emission abatement measures.in b at all cities (Table 4).Therefore, sum of [O 3 ST ] and [O 3 SEASON ] account for the most of O 3 variations.

Relative contributions of O 3 variations in different timescales
In Fig. 9a and b, the relative contributions of [O 3 ST ] and [O 3 SEASON ] show a strong negative relationship spatially.The relative contributions of [O 3 ST ] are generally larger at the coastal cities (53.1 %) than at the inland cities (45.9 %), whereas the relative contributions of [O 3 SEASON ] are smaller at the coastal cities (32.8 %) than at the inland cities (41.9 %).
Since [O 3 ST ] is related to synoptic-scale weather fluctuations (Rao et al., 1995(Rao et al., , 1997)), the large relative contributions of [O 3 ST ] at the coastal cities indicate the stronger effects of the eastward-moving synoptic weather systems there.Interestingly, the highest value of the [O 3 ST ] contribution appears at a northeastern coastal city, Gangneung.High and steep mountains on the west of Gangneung induce often warm, dry, and strong westerly winds, which is favorable to the clear sky and strong vertical mixing over the region.Since the westerly winds contain the precursors emitted from the SMA, the clear sky condition increases the O 3 levels during the daytime.In addition, the strong vertical mixing of high O 3 air from the free troposphere compensates the O 3 loss by titration during the nighttime.In the easterly winds, however, orographic lift often forms fog or clouds over the region and reduces the photochemical production of O 3 .Therefore, the combined effects of wind directions related to synoptic weather systems and topography increase the short-term variability of O 3 at Gangneung.
However, [O 3 SEASON ] is driven mainly by the annual cycle of meteorological factors such as insolation or temperature.Therefore, the large relative contributions of [O 3 SEASON ] at the inland cities are consistent with the higher impacts of temperature and insolation on O 3 there (Figs. 5, 9b).The highest and second-highest values of the [O 3 SEASON ] contribution appear at Andong and Wonju, located in the inland basin.Since the basin topography often traps pollutants and induces large annual ranges of temperature, seasonal variability of O 3 at the two cities is larger than that of other inland cities.
[O 3 LT ] explains less than 10 % of the total variances, but its relative contribution is considerable in the southwestern part of the Korean Peninsula as displayed in Fig. 9c.This is related to a relatively large long-term variability or trend in the region and is further discussed in Sect.3.6.We here further investigate the transport effect on the short-term variations of O 3 and the frequency of high O 3 episodes using exp[O 3 ST ] and WDs.As described in Sect.2.2, exp[O 3 ST ] is a ratio of the raw O 3 8h concentration to its baseline concentration in ppbv (exp [O 3 BL ]).Thus, the O 3 8h concentration is higher than the baseline O 3 8h concentration when exp[O 3 ST ] > 1.We classified every single value of exp[O 3 ST ] with the eight cardinal WDs during the months of frequent high O 3 events (May-October) at all available monitoring sites within each city.The probabilities of exp[O 3 ST ] > 1 for each WD were compared with the probabilities of exceeding the South Korean air quality standard of 60 ppbv for O 3 8h .Figure 10 shows exp[O 3 ST ] in the SMA cities (Seoul, Incheon, Suwon, and Ganghwa) with probabilities of exp[O 3 ST ] > 1 and O 3 8h > 60 ppbv for each WD.In Seoul, high O 3 episodes occur mostly during northwesterly winds although westerly and northeasterly winds predominate during the months of frequent high O 3 events (Fig. 10a, b).The high probability of high O 3 in northwesterly winds in Seoul is similar to those in other neighboring cities in the SMA, where the high probability also occurs during northwesterly winds in Incheon located in the west of Seoul (Fig. 10c, d), and during westerly winds in Suwon in the south of Seoul (Fig. 10e, f), and in Ganghwa in the northwest of Seoul (Fig. 10g, h).
The sea-mountain breeze can explain the prevalence of high O 3 episodes under westerly or northwesterly winds in the SMA.In the western coast of the SMA, there are many thermoelectric power plants (see triangles in Figs.11 and 12) and industrial complexes, which directly emit a large amount of O 3 precursors.Heavy inland and maritime transportation in those regions is also an important source of NO x and hydrocarbon emissions.Since the SMA is surrounded by the Yellow Sea in the west and a mountainous region in the east (see Fig. 1), the westerly sea breeze is well developed under O 3 -conducive meteorological conditions such as high temperature and strong insolation with low wind speed (Ghim and Chang, 2000;Ghim et al., 2001).In addition, locally emitted precursors and transported background O 3 from the west are trapped in the SMA due to the westerly sea breeze and the mountainous terrain in the east of the SMA.Therefore, the O 3 concentrations in the SMA increase in such O 3 -conducive meteorological conditions with near-westerly winds.
Another factor to increase the high O 3 probabilities in the near-westerly winds is long-range transport of O 3 and its precursors from China.For example, Ghim et al. (2001) reported some high O 3 cases in the SMA, which result from the transport of O 3 -rich air with strong westerly wind at dawn under overcast conditions.Oh et al. (2010) also showed that the elevated layer of high O 3 concentration over the SMA is associated with the long-range transport of O 3 from eastern China.As the mixing layer thickens over the SMA, the O 3 concentration can increase by up to 25 % via the vertical down-mixing process (Oh et al., 2010).Recently, Kim et al. (2012) showed that westerly winds also transport O 3 precursors such as NO 2 and carbon monoxide (CO) from China to South Korea.
Interestingly, the high O 3 probability in Ganghwa (Fig. 10g, h) shows a bimodal distribution with another peak in easterly wind.Considering that Ganghwa is a rural county on the northwestern coast of the SMA, the double peak of high O 3 probability in easterly and westerly winds shows the effects of both local-and long-range transport.-h, 12fh).The prevailing westerly wind of the synoptic-scale flow transports O 3 and its precursors from China to South Korea and thus increases the probability of high O 3 episodes as well as high O 3 concentrations.However, on a local scale, the high-probability regions of high O 3 correspond to those downwind of the thermoelectric power plants.For example, the high probabilities of high O 3 in the southeastern part of South Korea, downwind of power plants along the southeastern coast, also appear even during the easterly or southerly winds (Figs.11c-e, 12c-e).Therefore, the spatial features of the high O 3 probabilities for each wind direction could be associated with both the local effect of precursor emissions and long-range transport from the continent.

Long-term variation of O 3 and local precursor emissions
The  The increasing trends of O 3 are generally high in the SMA and southwestern region and low in the southeastern coastal region of the Korean Peninsula (Fig. 13a).This spatial inhomogeneity of the O 3 trends over South Korea is mainly caused by the long-term component trends (Fig. 13c) rather than the seasonal component trend (Fig. 13b).Therefore, the large spatial variability in local precursor emissions induced the spatial inhomogeneity of O 3 trends in South Korea.However, a relatively homogeneous distribution of the seasonal component trends implies that meteorological influences on the long-term changes in O 3 have little regional dependence nationwide.
Since the spatially inhomogeneous O 3 trends are related to the local precursor emissions, we also tried to investigate their relationship with NO 2 measurement data.To detect temporally synchronous and spatially coupled patterns between the long-term variations of O 3 and NO 2 , we applied the SVD to [O 3 LT ] and the long-term component of NO 2 ([NO 2 LT ]).[NO 2 LT ] was simply obtained by applying the KZ filter of KZ 365,3 to the log-transformed NO 2 time series.The SVD is usually applied to two combined space-time data fields, based on the computation of a temporal crosscovariance matrix between two data fields.The SVD identifies coupled spatial patterns and their temporal variations, with each pair of spatial patterns explaining a fraction of the squared covariance between the two space-time data sets.The squared covariance fraction (SCF) is largest in the first pair (mode) of the patterns, and each succeeding mode has a maximum SCF that is unexplained by the previous modes.
The first three leading SVD modes (singular vectors) of the coupled O 3 and NO 2 long-term components account for the SCF with 94.6 % of the total, of which the first, second, and third modes are 63.7, 23.6, and 7.3 %, respectively.Figure 14 displays the expansion coefficients (coupled spatial patterns) and their time series of the first mode along with a spatial map of the [NO 2 LT ] trends.The dominant first mode of the O 3 and NO 2 long-term components (Fig. 14a, b) is very similar to the spatial distributions of [O 3 LT ] trends (Fig. 13c) and [NO 2 LT ] trends (Fig. 14c), respectively.In  are needed.On top of that, especially in South Korea, biogenic precursor emissions are also potentially important for the analysis due to dense urban vegetation in and around metropolitan areas.Therefore, there remains the limitation of our current data analysis due to the lack of both VOC emission data and observations of atmospheric concentrations of VOCs in South Korea.

Conclusions
This study has investigated various spatiotemporal features and the interrelationship of surface O 3 and related meteorological variables over South Korea based on ground measurements for the period 1999-2010.A general overview of surface O 3 in terms of spatial distributions and its temporal trend is provided based on its decomposed components by the KZ filter.
In South Korea, the O 3 concentrations are low at the inland and SMA cities due to the NO x titration by anthropogenic emissions and high at the coastal cities possibly due to the dynamic effects of the sea breeze.The averaged O 3 levels in South Korea have increased for the period 1999-2010 with an averaged temporal linear trend of +0.26 ppbv yr −1 (+1.15 % yr −1 ).The recent increase of the O 3 levels in East Asia may result from the recent increase of anthropogenic precursor emissions and the long-term variations in meteorological effects.
We applied a linear regression model to investigate the relationships between O 3 and meteorological variables such as temperature, insolation, dew-point temperature, sea-level pressure, wind speed, and relative humidity.Spatial distribution of the R 2 values shows high meteorological influences in the SMA and inland regions and low meteorological in-fluences in the coastal region.The high meteorological influences in the SMA and inland regions are related to effective photochemical activity, which results from large local precursor emissions and stagnant conditions with low wind speeds.However, the low meteorological influences in the coastal region are related to large transport effects of the background O 3 and ventilation and dry deposition with high wind speeds.
In the SMA and inland region, the high O 3 probability (O 3 8h > 60 ppbv) increases with the daily maximum temperature rise.Specifically in the SMA, the most populated area in South Korea, the probability of the O 3 exceedances is almost doubled by a 4 • C increase in daily maximum temperature and reached 27 % at 30 • C. It is noted that the variations in O 3 exceedance probabilities according to the maximum temperature show an approximate logarithmic increase in the SMA and inland regions.It thus implies that these regions will experience more frequent high O 3 events in future climate conditions with the increasing global temperature.
The  The KZ filter is a useful diagnostic tool to reveal the spatiotemporal features of O 3 and its relationship with meteorological variables.General features revealed by the KZ filter analysis will provide a better understanding of spatial and temporal variations of surface O 3 as well as possible influences of local emissions, transport, and climate change on O 3 levels in South Korea.Our analyses can also be helpful as a reference for the evaluation of chemistry transport models and furthermore for establishing appropriate O 3 control policies.
The Supplement related to this article is available online at doi:10.5194/acp-14-6395-2014-supplement.
[O 3 LT ]   in Fig.2dshows that the O 3 concentrations at the monitoring site have increased in the past decade, irrespective of any change in meteorological conditions.It should be noted that [O 3 LT ] explains only 1.7 % of the total variance of [O 3 ] at this site as can be seen from its small ranges in Fig.2d, while relative contributions of [O 3 ST ] and [O 3 SEASON ] are 58.3 and 32.7 %, respectively.Therefore, the long-term changes in O 3 related to changes in local emission are only a small fraction of the O 3 variations.The relative contributions of each component are further examined in Sect.3.4.

Figure 1 .
Figure 1.Time series of daily maximum 8 h average ozone (O 3 8h ) at the City Hall of Seoul and its separated components such as (a) logtransformed O 3 8h time series ([O 3 ]) and its baseline ([O 3 BL ]), (b) short-term component ([O 3 ST ]), (c) seasonal component ([O 3 SEASON ]), and (d) long-term component ([O 3 LT ]) by applying the KZ filter.It is noted that the longer window length causes the larger truncation of the result (Wise and Comrie, 2005) since the KZ filter is an iterative moving average process.The baseline (red solid line) is superimposed in (a).

Figure 2 .
Figure 2. (a) Geographical location of South Korea, (b) 72 weather stations of the Korea Meteorological Administration (KMA) as blue circles, (c) 124 air quality monitoring sites of the National Institute of Environmental Research (NIER) as black dots, and (d) 72 air quality monitoring sites of NIER, which are located within 10 km from 25 weather stations of KMA over the South Korean domain.

Figure 3 .
Figure 3. Spatial distributions of 12 yr averaged concentrations of (a) daily average O 3 (O 3 avg ), (b) daily average nitrogen dioxide (NO 2 avg ), and (c) temporal linear trends of O 3 avg for the period 1999-2010 using data from 124 air quality monitoring sites (black dots) of NIER.

Figure 4 .
Figure 4. Spatial distributions of temporal linear trends of (a) daily average T and (b) daily average SI for the period 1999-2010 using data from 72 and 22 weather stations (black dots) of KMA, respectively.

Figure 5 .
Figure 5. Spatial distributions of R 2 between baselines of O 3 8h ([O 3 BL ]) and (a) daily maximum temperature (T max BL ) and (b) surface insolation (SI BL ).Black dots represent 72 air quality monitoring sites of NIER.(c) Spatial distribution of adjusted R 2 between [O 3 BL ] and combined meteorological effects (a 0 + i a i MET BL (t) i ).

Figure 6 .
Figure 6.(a) Spatial distribution of 12 yr averaged concentrations of O 3 min for the period 1999-2010 using data from 124 air quality monitoring sites (black dots) of NIER.(b) Scatter plot of R 2 between [O 3 BL ] and T max BL vs. O 3 min at 25 cities.(c) Scatter plot of R 2 between [O 3 BL ] and SI BL vs. O 3 min at 17 cities.City codes in red, green, and blue indicate the SMA, inland, and coastal cities, respectively.

Surface O 3 Figure 8 .
Figure 7. (a) Spatial distribution of 12 yr averaged daily average WS for the period 1999-2010 using data from 72 weather stations (black dots) of KMA.(b) Scatter plot of O 3 min vs.WS at 25 cities.(c) Scatter plot of R 2 between [O 3 BL ] and T max BL vs. WS at 25 cities.City codes in red, green, and blue indicate the SMA, inland, and coastal cities, respectively.

Figure 9 .
Figure 9. Spatial distributions of relative contributions of the (a) short-term component ([O 3 ST ]), (b) seasonal component ([O 3 SEASON ]), and (c) long-term component ([O 3 LT ]) to the total variance of original time series ([O 3 ]) using data from 72 air quality monitoring sites (black dots) of NIER.Note that the color scales are all different.

3. 5
Short-term variation of O 3 related to wind direction The short-term components of O 3 ([O 3 ST ]) account for a large fraction of total O 3 variation over South Korea.In Table 4, relative contributions of [O 3 ST ] range from 32.7 to 62.5 % and have a nationwide average of 49.8 %.Therefore, it is no wonder that high O 3 episodes are mostly determined by day-to-day fluctuation of [O 3 ST ].One considerable factor influencing the short-term variation of O 3 is wind.Shin et al. (2012) displayed [O 3 ST ] on the wind speed-direction domain and showed that the effects of episodic long-range transport and local precursor emission on the ambient O 3 concentrations could be qualitatively separated from [O 3 ST ].

Figure 10 .
Figure 10.Relationships between WD and exponentials of short-term components (exp[O 3 ST ]) during the months of frequent high O 3 events (May-October) at Seoul (a-b), Incheon (c-d), Suwon (e-f), and Ganghwa (g-h) in the SMA are represented in scatter plots of exp[O 3 ST ] vs. WD (a, c, e, and g) and probabilities of O 3 exceedances in each WD (b, d, f, and h).Red dots in scatter plots denote high O 3 episodes in which the daily maximum 8 h average O 3 (O 3 8h ) will exceed the air quality standard of South Korea (60 ppbv).Dashed lines in scatter plots denote the reference of exp[O 3 ST ] = 1.Probabilities of exp[O 3 ST ] > 1 and O 3 8h > 60 ppbv in each WD are represented as black thick lines and red thick lines, respectively.The black and red thin lines represent the 95 % of confidence intervals for each probability.We used O 3 data from 12 sites in Seoul, 6 sites in Incheon, 3 sites in Suwon, and 1 site in Ganghwa.
temporal linear trend of baseline ([O 3 BL ]) is almost the same as that of the original time series since the shortterm component ([O 3 ST ]) is nearly detrended.Therefore, the O 3 trend can be represented as a sum of the seasonal component ([O 3 SEASON ]) and long-term component ([O 3 LT ]) trends.The spatial trend distributions of [O 3 BL ] and its two separated components of seasonal and long-term components are shown in Fig.13.It is noted that the period used in Fig.13is shorter than the total period of original data because of truncation effect in the KZ filter process.The longterm component obtained by the KZ filter of KZ 365,3 loses 546 days at the beginning and end of the original time series.

Figure 12 .
Figure 12.Spatial distributions of probabilities that the daily maximum 8 h average O 3 (O 3 8h ) will exceed the air quality standard of South Korea (60 ppbv) for each WD: (a) N,(b) NE, (c) E, (d) SE, (e) S, (f) SW, (g) W, and (h) NW, respectively.Black dots denote 25 weather stations of KMA and triangles denote 26 major thermoelectric power plants in South Korea (blue triangle < 1000 MW, red triangles ≥ 1000 MW).
Fig. 14d, the strong coherence in the time series is observed between the first modes of the [O 3 LT ] and [NO 2 LT ] with a correlation coefficient of 0.98.The results of SVD analyses suggest that the long-term variations of O 3 and NO 2 in South Korea have similar temporal evolutions with different spatial patterns.The differences in spatial patterns of [O 3 LT ] and [NO 2 LT ] as shown in Fig. 14a and b require to be further investigated.Since the VOC emissions from industry, transportation, and the solvent usage in construction are large in South Korea (KMOE, 2013), further analyses of VOC measurements
O 3 time series observed at each monitoring site can be decomposed into the short-term, seasonal, and long-term components by the KZ filter.Relative contributions of each separated component show that the short-term and seasonal variations account for most of the O 3 variability.Relative contributions of the short-term component are large at the coastal cities due to influence of the background O 3 transport.In contrast, those of the seasonal component are large at the inland cities due to the high meteorological influences on the O 3 variations.The transport effects on the short-term component are shown in the probability distributions of both high shortterm component values and O 3 exceedances for each wind direction.During the months of frequent high O 3 events (May-October) in South Korea, the probabilities of both high

Figure 14 .
Figure 14.The first leading mode of SVD between the long-term components of (a) daily maximum 8 h average O 3 ([O 3 LT ]) and (b) daily average NO 2 [NO 2 LT ] for the period 2000-2009.(c) Spatial distribution of temporal linear trends of [NO 2 LT ].(d) Time series of the SVD expansion coefficient associated with [O 3 LT ] mode (blue line) and [NO 2 LT ] mode (red line).

Table 1 .
12 yr averaged concentrations and temporal linear trends of daily average O 3 (O 3 avg ) at 46 cities over South Korea for the period 1999-2010.The cities are categorized into three groups: 16 coastal cities, 14 inland cities, and 16 cities in the SMA.

Table 3 .
12 yr averaged of daily minimum O 3 (O 3 min ) concentrations and daily average WS at 46 cities over South Korea for the period 1999-2010.The cities are categorized into three groups: 16 coastal cities, 14 inland cities, and 16 cities in the SMA.
* Major metropolitan cities in South Korea.n/a: not available observations of wind speed.