Surface in situ measurements of O3 and NO2 are collected from air quality monitoring networks of the National Institute of Environmental Research (NIER) (AirKorea, http://www.airkorea.or.kr, last access: 25 October 2022). This network measures hourly air pollutant (O3, NO2, CO, SO2) mixing ratios through chemiluminescence technology (Kley and Mcfarland, 1980). The KMA operates automatic synoptic observation system (ASOS) at 102 weather stations. The ASOS measurements are provided on five types of timescales (minutely, hourly, daily, monthly, yearly) via the KMA Weather Data Service (https://data.kma.go.kr/, last access: 25 October 2022). We used daily averages of air temperature, relative humidity, solar irradiance, total precipitation, wind speed, and wind direction.

Ozonesondes are balloon-borne instruments capable of measuring the vertical distribution of atmospheric ozone from the surface to balloon burst, usually near 35 km. The electrochemical concentration cell (ECC) sensor is the most widely employed. ECC ozonesondes have an uncertainty of 5 %–10 % and a precision of 3 %–5 % (Smit et al., 2007). In South Korea, ECC sondes have been regularly launched only at Pohang station every Wednesday in the afternoon (13:30–15:30 LT) since 1995. Ozonesonde measurements are reported in units of partial pressure (mPa) with vertical resolution of about 100 m by the Korea Meteorological Administration (KMA). Bak et al. (2019) demonstrated that Pohang ozonesonde measurements are a stable set of reference profiles for validating satellite products, with quality comparable to ECC ozonesonde measurements in Japan and Hong Kong. To improve the data quality, we screened out sounding measurements at balloon burst altitudes higher than 200 hPa and observations of either tropospheric ozone column values above 80 DU or stratospheric ozone column values below 100 DU.

Both OMI and MLS were launched on board NASA's EOS-Aura spacecraft in July 2004 and are still functioning in measuring the Earth's atmospheric composition. The Aura satellite crosses the Equator at ∼ 13:30 in the afternoon. OMI is a nadir-viewing imaging spectrometer capable of daily global mapping at a relatively high spatial resolution of 13 km × 24–48 km (across × along-track). MLS measures microwave thermal emission from the limb of Earth's atmosphere. Compared to OMI, MLS makes measurements at a good vertical resolution (∼3 km) in the upper atmosphere but at relatively coarse horizontal resolutions (∼165 km along the orbit track). Version 4.2 of the MLS standard ozone product is used in this study, only for the recommended vertical range from 261 to 0.025 hPa (Schwartz et al., 2015). We used OMI ozone profiles retrieved using the PROFOZ version 2 algorithm, which is in preparation for reprocessing OMI measurements to release a new version of the OMPROFOZ research product (Liu et al., 2010). This retrieval algorithm consists of wavelength and radiometric calibrations as well as forward modeling simulations, with an optimal estimation inversion in which a priori knowledge is optimally combined with measurement information to obtain a better estimate of the state (Rodgers, 2000). The measurement sensitivity inherently decreases toward the surface, with the increasing dependence of retrievals on the a priori information (Bak et al., 2013). OMI sensitivity is very low to surface ozone, with its maximum in the free troposphere (∼500 hPa) (Shen et al., 2019).

The Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) is NASA's latest reanalysis, spanning the satellite observing era from 1980 to the present (Gelaro et al., 2017). In addition to a standard meteorological analysis, a global O3 field is driven by atmospheric dynamics and constrained by satellite O3 measurements using the GEOS-5 atmospheric model and the data assimilation system. Since October 2004, MERRA-2 has assimilated total column ozone from OMI and stratospheric ozone profiles above 215 hPa from MLS. Note that OMI total column ozone is assimilated to account for the lower sensitivity of MLS measurements in the lower stratosphere, specifically in clouded scenes.

The CAMS reanalysis is the latest global reanalysis dataset of atmospheric composition produced by the Copernicus Atmosphere Monitoring Service (CAMS), covering the period from 2003 to the present (Inness et al., 2019). Compared to MERRA-2, multiple satellite measurements were assimilated for the CAMS reanalysis with ECMWF's Integrated Forecasting System. These included total ozone columns from SCIAMACHY, OMI, and GOME-2 as well as ozone profiles from MIPAS and MLS after 2005.

Both types of reanalysis data have similar temporal and spatial resolutions. The MERRA-2 system produces 3-hourly analyses at 72 sigma–pressure hybrid layers between the surface and 0.01 hPa, with a horizontal resolution of 0.625∘ × 0.5∘. The CAMS reanalysis data provide estimates every 3 h with a horizontal resolution of 0.75∘ × 0.75∘. The vertical resolution of the model consists of 60 hybrid sigma–pressure (model) levels from the surface to 0.1 hPa. In this study, we used CAMS global reanalysis (EAC4) monthly averaged fields at 25 pressure levels (1000 hPa to 1 hPa) as well as MERRA-2 monthly mean data at 42 pressure levels (1000 hPa to 1 hPa). Both datasets provide ozone profiles in the unit of mixing ratio.

Figure 1 shows both interannual and seasonal changes in daily ground-level concentrations of O3 averaged at six AirKorea sites located within Pohang for 16 years (2005–2020) in comparison with its primary precursor NO2. Pohang is a major industrial city on South Korea's east coast, with the largest population of northern Gyeongsang Province. In this analysis, hourly measurements in the afternoon (13:00–15:00 local time) are first averaged for a given calendar day and then smoothed by a 2-week moving average. The afternoon NO2 does not change much seasonally. However, the seasonal cycle of ozone is bimodal with peaks in early summer and fall. Ozone concentration rapidly increases from ∼30 ppb in January to primary peak values of ∼55 ppb on average during the period of late May to early June. The second peak of ozone occurs in fall, which is much lower than the major peak.

In wintertime, the annual minimum ozone concentrations have gradually increased by ∼10 ppb during last 15 years, whereas the annual maximum of summertime ozone has rapidly increased from ∼40 to 80 ppb in spite of the reduction of the NO2 amount by ∼15 ppb or more.

In order to avoid smoothing out important features of intra-summer variations in ozone and their association with synoptic weather patterns, daily ozone and meteorological variables are zoomed in for 2020 as a 1-week moving average (Fig. 2). The local maximum ozone concentrations are generally tied to the local warm, dry air and intense solar radiation before the rainy season starts.

The correlation between ozone concentrations and meteorological variables is quantitatively compared in Table 1 for summer and the post- and pre-summer periods, respectively. Solar insolation amounts are directly linked to ozone concentrations over all seasons (r= 0.51–0.91). The significant relationship between ozone and air temperature is also identified before and after summer seasons. However, in summer, ozone variations are rarely linked with temperature variations due to the intense precipitation suppressing ozone formation. Consequently, the local minimum ozone levels are tied to the local maximum relative humidity during the rainy season (r=-0.64). This indicates that both the depth and width of the summer trough could be highly variable, likely influenced by the strength and duration of the summer monsoon (Yang et al., 2014; Zhou et al., 2022). Note that the relative humidity is significantly influenced by air temperature, rather than the amount of water vapor in the pre- and post-summer periods. Therefore, in post-summer the correlation of ozone with relative humidity (r=0.59) is likely to arise from the correlation of ozone with air temperature (r=0.51). The rapid drop of ∼10 ppb in ozone from the end of July to early August is hardly explained by the meteorological factors mentioned above; the weather becomes warmer with other meteorological variables (precipitation and solar radiation) being relatively invariant. However, the prevailing wind is characterized by southwesterlies in early August, whereas the northwesterly winds were dominant in July and late August (see Fig. 3). This summer minimum could deepen with the inflow of a poor ozone air mass originating from the southern sea off the Korean Peninsula into inland areas.

To understand the seasonality of ozone profiles, ozonesonde measurements collected at Pohang station are climatologically averaged for each month and each pressure bin (∼0.5 km intervals). Ozonesonde soundings mainly measure ozone in the lower atmosphere below 10 hPa, while space-based limb soundings mainly measure ozone in the upper atmosphere above 215 hPa. However, both sounding measurements provide limited spatiotemporal information. OMI nadir measurements and reanalysis data provide daily global maps of ozone profiles, but the reliability of those data products should be assured before using them to interpret ozone variability and its linkage to the monsoon circulation. As shown in Fig. 4a, two kinds of seasonal patterns are identified with a bimodal structure of layer ozone partial pressures in the lower troposphere (LT), whereas there is a unimodal cycle in the upper troposphere and lower stratosphere (UTLS). The LT ozone concentrations peak in June and October with a global minimum in winter as well as a local summer minimum in late July and early August, which is consistent with surface measurements. The concentrations of UTLS ozone are relatively higher in March due to the stratospheric intrusion, while the minimum concentrations appear broadly over the summer and early fall due to the rise of the tropopause, which is a common feature of ozone in the extratropical UTLS (Gettelman et al., 2011; Rao et al., 2003). In order to quantify the similarity of seasonal variations, the correlation coefficient is calculated for temporal ozone changes between each layer and the top and bottom layer. As shown in Fig. 4b the seasonality of ozone at 50 hPa is significantly correlated down to ∼300 hPa, with a correlation coefficient larger than 0.8. In addition, ozone in the boundary layer is significantly correlated with the lower-tropospheric ozone up to 700 hPa (r>0.9) as well as the upper-tropospheric ozone up to ∼300 hPa (r= 0.7–0.8). This illustrates that 300 hPa could be regarded as a chemical barrier working as a boundary between the troposphere and stratosphere at Pohang.

In Fig. 5, monthly averaged ozonesonde profiles are presented for 2020 and compared as a reference to assess satellite measurements and reanalysis products. This contour map of ozonesondes clearly illustrates the intrusion depth of stratospheric air masses down to ∼300 hPa during spring months (Fig 5a). The mixing depth of ozone that forms near the ground level is also identified, which is bounded up to ∼400 hPa in the summer and ∼600 hPa in other seasons. The minimum ozone concentration is typically found just below the thermal tropopause. The August minimum of lower-tropospheric ozone vertically extends above ∼600 hPa. This air mass is much cleaner compared to the winter ozone concentration over the lower troposphere. The dominant factor suppressing the ozone formation is long-lasting summer precipitation from early July to mid-August in 2020 (Fig. 2). The southerly wind that blows on the observation site is relatively strong compared to June and July. Therefore, we could interpret the inland polluted air masses as likely to be diluted with inflows of maritime clean air masses as mentioned above. In the lower troposphere, a minor peak of ozone concentrations is also identified in spring, which is not visible in time series plots of surface measurements (Fig. 2). The springtime peak is mainly originated by fair weather accelerating the formation of ground-level ozone with the wintertime accumulation of ozone and its processors; it could be partly attributed to the dynamical processes transporting ozone-rich air from the UTLS and upwind areas. In Fig. 5b–f, OMI, MERRA-2, and CAMS ozone profiles are qualitatively evaluated with respect to the capability to reproduce the seasonality of ozone profiles at this location. The ozone minimum of the summer monsoon season is detected from all ozone products, but it is much broader than that in ozonesondes due to both the limited time resolution of ozonesonde measurements and the limited spatial resolution of OMI and reanalysis products. OMI also shows very good agreement with ozonesondes in terms of reproducing the boundary layer ozone extending up to the free troposphere and the low ozone concentration below the tropopause. In addition, the vertical gradient of ozone enhancement above the tropopause is consistently reproduced from OMI, ozonesondes, and MLS. The spring ozone peak near the surface is not detectable from OMI measurements due to the limited sensitivity to relatively shallow boundary layers compared to summer (Shen et al., 2019). In Fig. 5d, an OMI a priori profile is also presented to highlight the fact that the summer minimum is derived from the independent information of OMI measurements rather than a priori information. It also illustrates that the summer minimum is a regional feature of tropospheric ozone seasonality not represented in the climatological data in which long-term global measurements are composited as a function of month and latitude.

Both MERRA-2 and CAMS considerably overestimate ozone abundances in both the troposphere and stratosphere in spite of the fact that MLS measurements are commonly employed for assimilating stratospheric ozone profiles. In MERRA-2, the bimodal peaks (April and October) of the lower-tropospheric ozone are inconsistent with others (early summer, September). We also compare how each ozone product represents the tropopause against thermally defined tropopause heights using the World Meteorological Organization (WMO) definition (WMO, 1957). There is no universal method to define the ozonepause height, but threshold values of 100 to 150 ppb in ozone mixing ratios were used to discriminate stratospheric and tropospheric air masses (e.g., Hsu et al., 2005; Prather et al., 2011). In this paper, the 150 ppb value is selected due to similarities of thermal tropopauses with ozone surfaces of 150 hPa from ozonesonde measurements. As shown, the ozone surfaces at 150 ppb of reanalysis products are positioned in the free troposphere due to the overestimation errors. Both ozonesondes and Aura measurements show some consistency between their ozone and thermal tropopause pressures. In particular, OMI shows strong consistency with the fact that retrievals near the tropopause are largely constrained with the a priori state taken from the tropopause-based ozone profile climatology (Bak et al., 2013).

In this section, we focus on the ozone changes related to interannual meteorological variabilities, along with the evaluation of different ozone products. In Fig. 6, time series of the mean ozone mixing ratio in the lower troposphere (750–950 hPa) in August are compared. The summer monsoon typically ends in the late July and early August over the Korean Peninsula, and hence the ozone abundance in August is sensitive to the intensity and duration of the monsoon season. OMI and ozonesondes show a similar long-term change, except for much more fluctuation in time series of ozonesondes due to insufficient samplings (weekly observations) used in monthly averages. A noticeable correlation (r=-0.52) exists between wind speeds and ozone mixing ratios (ozonesonde). Low wind speed could enhance the accumulation of ozone precursors and the rate of ozone formation. Accordingly, both ozonesondes and OMI measurements detect higher ozone abundances in August from 2014 to 2017 when the wind speeds are relatively lower. As shown in Fig. 7a–c, where the monthly meteorological fields at 850 hPa in 2015 are presented from the MERRA-2 product, the western North Pacific Subtropical High (WNPSH) was broken in August, and hence the weather was likely to be calm and dry over the Korean Peninsula. Compared to the past few years, a lower amount of ozone is detected in 2020 from ozonesonde measurements. In August 2020, the lower-tropospheric southwesterly winds blow from the western North Pacific to the Korean Peninsula across the edge of the WNPSH, and the rain belt was activated over the Korean Peninsula (Fig. 7d–f). Therefore, the weather was windy and wet, suppressing ozone formation in August 2020.

MERRA-2 ozone shows no annual variation before 2020, unlike other ozone measurements and products. CAMS also shows higher ozone concentrations correlated with wind speeds, but these are less consistent with ozonesonde measurements compared to OMI. How the El Niño–Southern Oscillation (ENSO) cycle interacts with the East Asian monsoon has not been established. According to the Oceanic Niño Index, the 2015–2016 El Niño event, the warm phase of the ENSO, was one of the strongest events ever recorded, whereas the 2020–2021 La Niña event was also abnormally strong. There were a lot of unprecedented weather events in South Korea during these super El Niño and La Niña periods, such as unprecedented summer rainfall in 2020 and unprecedented summer heatwaves in 2015–2016 (Yoon et al., 2018). Therefore, we could relate the higher ozone amount in August 2015–2017 and the lower ozone amount in August 2020 to a climatic forcing on the strength and position of the WNPSH and hence the East Asian summer climate.