Seosan, the target area of the GMAP-2020 campaign, is a small city with a population of 174 780 in 2017; it is accessed via three expressways to the east and four national highways cross the city. It is located in midwestern South Korea, and is affected by >300 emissions point sources including LPSs. Coal-fired power plants including Taean, Dangjin, the Hyundai Dangjin steelworks, and the Daesan petrochemistry industrial complex (LPS1–LPS4, respectively, in Fig. 1) have the highest emissions rates in South Korea. The Hyundai Dangjin steelworks (LPS3) and Taean and Dangjin power plants (LPS1 and LPS2) emit 10.5, 11, and 8.8 Gg of NOx per year, respectively. Although Seosan accounts for only 1.8 % of the population of Seoul, its NOx emissions (10.2 Gg yr-1) account for 13.2 % of its total NOx emissions. The transportation sector of Seosan is a far greater NOx source than the industrial sector of Seoul (ratio of 99:1); however, within Seosan, the industrial sector is on par with the transport sector (52:48; https://www.air.go.kr/en-main.do, last access: 22 August 2022).

During the past decade, the annual mean NO2 level in Seosan has been 17 ppb, which is approximately half of that in Seoul (31.2 ppb). NO2 exhibits strong seasonal variation, reaching a minimum in summer and maximum in winter, due to meteorological factors and greater energy use during winter (Kim and Kim, 2020). Therefore, the timing of the GMAP-2020 campaign was well suited to tracking pollution.

Pandora measures the UV and visible wavelengths (280–525 nm) of direct sunlight with a spectral resolution of 0.6 nm, to determine the vertical column density of NO2, O3, and HCHO (Hermanet al., 2009). For measurements in Dobson units (DU; 1 DU = 26.9 Pmol cm-2), column NO2 has a very high signal-to-noise ratio (700:1) and very high precision (0.01 DU) for clear skies (Herman et al., 2009). The vertical column density of NO2 can be determined using DOAS software (Van Roozendael and Fayt, 2001). Pandora direct-sun measurements are advantageous in that the air mass factor is simplified, and therefore is dependent only on the geography for a known solar zenith angle.

From retrieved Pandora measurements, tropospheric and total (= tropospheric + stratospheric) vertical column densities are both available. However, it should be noted that appreciable uncertainties cannot be neglected in the tropospheric NO2 profiles obtained from Pandora instruments, particularly for the high aerosol-loading areas such as East Asia. In this background, we used total vertical column densities in the present study, and also confirmed that they have a high correlation with the tropospheric column densities observed in our study period, with little change in stratospheric column density in space and time at the local scale.

Four Pandora instruments were installed at sites to the south of LPSs (Fig. 1) during the GMAP-2020 campaign, i.e., at Seosan Daehoji, Seosan Dongmun, Seosan city council, and Seosan super site (PA1–PA4 in Fig. 1). The presence of clouds reduces vertical column density precision by decreasing the number of photons arriving at Pandora instruments within a fixed integration time. Therefore, the retrieved Pandora measurements were cloud-screened using an observed cloud cover of 0.6. Cloud cover was provided by the Korea Meteorological Administration (KMA), and the precision improvement afforded by cloud screening was verified by comparing each Pandora-derived vertical column density with the median vertical column density, with and without cloud screening within the inter-comparison period.

At PA4, the operating period was extended to cover almost the entire year (12 November 2020–30 October 2021) including the GMAP-2020 campaign period, and the Pandora spectra were processed into vertical column density data for trace gases using the standard NO2 algorithm in BlickP software provided by PGN (Cede, 2019). The resultant PC-NO2 data were obtained from the PGN website (https://pandonia-global-network.org, last access: 22 August 2022) for the 1-year period from 12 November 2020 to 30 October 2021, and were also used as PC-NO2 statistics for PA4, in this study.

Hourly average data for SI-NO2 over a period of 1 year were obtained from Ministry of Environment AQM network stations in Seosan: Pandori, Leewon, Taean, Dongmoon, Seongyeon, and Daesan (AQM1–AQM6, respectively, in Fig. 1). The Seosan super site (PA4/AQM1) provided hourly data for NO and NOy via an NO-DIF-NOy analyzer (42i-Y; Thermo Scientific, Waltham, MA, USA), and for PM2.5 chemical species using an ambient ion monitor (AIM; URG 9000D, URG Corp., Chapel Hill, NC, USA). Weekly zero and span checks were conducted for NOy calibration to ensure that differences between checks remained <3 %. Water-soluble ions in aerosol and gaseous species were measured hourly using an AIM, and ion mass balance was used to ensure data quality under the quality control procedures of the AQM network installation and operation guidelines (NIER, 2021).

Aircraft measurements were conducted during the GMAP-2020 campaign period, and nine flights were conducted on 8 d (26, 27, and 28 November and 1, 6, 8, 9, and 12 December 2020). The horizontal and vertical distributions of NO2 and O3 over Seosan were measured during GMAP-2020 using an NO2 monitor (T500U; Teledyne, Thousand Oaks, CA, USA) and an O3 analyzer (TEI49C; Thermo Scientific) onboard the Cessna Grand Caravan 208 B. These instruments had response times of <40 and <20 s, and detection limits of 40 ppt and 1 ppb, respectively. The flight paths included a raster mode over all of Seosan at a height of 500–700 m and a profiling mode from 500 m to 1.5 km over PA1 and PA4 (Fig. 2).

Ground-based hourly observation data for meteorological variables were obtained from Seosan Automated Synoptic Observing System (ASOS) stations maintained by the KMA, and wind and temperature profile data were obtained twice daily (00:00 and 12:00 UTC) via a rawinsonde instrument at the Osan World Meteorological Organization upper air measurement station (47122) near Seosan. Due to time constraints of the sonde measurements, information on PBLH variation was obtained from Unified Model (UM) simulation results provided on the KMA website (https://afso.kma.go.kr, last access: 22 August 2022).

During the GMAP-2020 campaign, a 3D sonic anemometer (CPEC200; Campbell Scientific, Logan, UT, USA) was also installed on the rooftop at PA4 for turbulent flux measurements at the city–atmosphere interface (Hong et al., 2019). All wind components and sonic temperatures were measured at a 10 Hz sampling rate, and ground-level sensitive heat flux was measured directly using a 30 min averaging period. Quality controls such as double rotation, spike removal, and outlier filtering were also applied.

We examined the synergy of PC and SI data obtained during the GMAP-2020 campaign, and combined these measurement data to evaluate air quality in Seosan, South Korea. We attempted to interpret the meteorological and photochemistry data measured during GMAP-2020, and to demonstrate that caution is required when attempting to study the vertical structure of air pollutants using either surface observations or satellite data only, particularly in industrial areas.

First, we examined the combined use of year-long PC-NO2 and SI-NO2 measurements, and investigated the factors modulating their correlation. Numerous studies have examined the correlations between chemical species, including aerosols (Thomspon et al., 2019; Wang et al., 2019; Kim et al., 2012, 2018; Jo and Kim, 2013; Sanchez et al., 1990). Wang et al. (2019) reported that aerosols moderately correlate with NO2 due to the frequent occurrence of lifted layers probably related to the transport of pollutants. Jo and Kim (2013) differentiated haze types using the trajectory speed and direction and different synoptic conditions. In this background, we hypothesized that their differences in PC-NO2 and SI-NO2 were due to meteorological conditions, and performed k-means and agglomerative hierarchical cluster analyses of various meteorological variables. Clustering is the grouping of objects that are alike and are different from the objects belonging to other clusters. As a first step, k-means clustering was applied to find smaller clusters until each object was classified in one cluster. Subsequently, agglomerative hierarchical steps are applied to make up for the shortcomings of k-means clustering, in which once merging (or splitting) is done, it can never be undone. More details are found in Venkat Reddy et al. (2017). We used XLSTAT software (Addinsoft, Paris, France) for the cluster analysis with eight meteorological variables representing local and synoptic circulations in the cluster analysis: surface wind speed (Wsfc), 925 hPa temperature (T925), sea level pressure (Psfc), pressure tendency (dPsfc/dt), 850 hPa wind speed (W850) and its north–south and east–west components (NS850 and EW850), and 500 hPa geopotential height (GPH500). We subtracted 30 d moving averages from all data to account for typical seasonal variation. Monthly averages were used for PC-NO2 analysis due to the limited availability of hourly data.

Correlations between PC-NO2 and SI-NO2 were analyzed in each meteorological group and the impact of photochemistry was interpreted based on case-specific features. We also investigated correlations in association with near-surface micrometeorological variables such as PBLH in each meteorological group.

The yearly PC-NO2 statistics at four Pandora sites (PA1–PA4) are summarized in Table 1. The total averaged PC–NO2 over all sites was 0.45 DU during GMAP-2020, which is well above the typical values (0.1–0.2 DU) for Anmyeondo (the location is shown in Fig. 1), a representative background site (Herman et al., 2018). Although site PA3 is located in a rural area, it nevertheless exhibited the highest PC-NO2 amounts, suggesting that plumes were frequently transported from nearby point sources and/or urban areas.

Scatter diagrams of hourly PC-NO2 and SI-NO2 measurements from Pandora sites PA1–PA3 (GMAP-2020) and PA4 (yearly measurement; 12 November 2020–30 October 2021) are shown in Fig. 3a. These hourly data had a fair 1:1 linear relationship (R=0.41), implying the overall uniformity of NO2 profiles, whereas the linear relationship with PC-NO2 weakened as SI-NO2 levels increased (Fig. 3a). It appears that the SI-NO2 has a distinct diurnal change despite the same PC-NO2, and higher variable surface NO2 levels may result from the relatively weaker linear relationship between PC-NO2 and SI-NO2. To explore these anti-correlation cases further, we selected the lower and upper bounds of the tendencies; these are plotted in Fig. 3b, which shows that PC-NO2 was positively correlated with SI-NO2 on 24 February 2021 (R=0.88), while a negative correlation occurred on 21 April 2021 (R=-0.88), indicating a wide range of case-specific correlations. The negative correlation on 21 April (Fig. 3b) implied that the nonhomogeneous NO2 distributions vertically were partially due to the photochemical process. For example, the decrease in PC-NO2 despite an increase in SI-NO2 might have occurred because NO2 is removed by photochemical loss; it can occur more severely in the upper atmosphere with high OH concentrations. Another possible reason is the occurrence of lifted layers related to pollutant transport, yielding sharp changes in vertical concentration from the surface to the upper layer. The case-specific discussion follows.

Our k-means cluster analysis distinguished three groups with the lowest within-group variance and largest among-group variance. Among the total of 141 cases, 47, 66, and 28 were classified into groups 1–3, respectively. Thus, group 2 had the largest proportion of cases (47 %) and group 3 had the smallest (20 %). The combination of meteorological components in group 1 indicated the end of a high-pressure system (Psfc > 0, dPsfc/dt<0), with southerly winds (NS850 > 0) bringing warmer air (T925 > 0) to the region, leading to stable atmospheric stratification and weak surface winds (Fig. 4). This group 1 meteorological mode appeared to result in very weak NO2 ventilation, which produced the highest PC-NO2 and SI-NO2 values. Group 3 showed the opposite trend, with strong northerly winds bringing colder air into the region, leading to an unstable atmosphere and stronger surface winds, and ultimately decreasing PC-NO2 and SI-NO2 to their lowest levels.

SI-NO2 was approximately twice as high in group 1 than group 3, whereas PC-NO2 showed no significant difference (Fig. 4a). We hypothesized that PBLH might also differ significantly under these micrometeorological conditions; therefore, we further explored daily maximum PBLH simulated by the Global Forecast System (GFS) and Lagrangian backward trajectories obtained from Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) and GFS system for the 141 cases. The mean simulated PBLH in Seosan, our study area, was 942.1±405.3 m for 2020, which was similar to the annual mean daily maximum PBLH (1013.6 m) in Osan (Lee et al., 2013). However, the simulated PBLH differed significantly among the three groups (767.0±304.8, 923.2±335.3, and 1280.6±501.2 m for groups 1–3, respectively). The PBLH for group 3 was 1.7-fold higher than that for group 1 (Fig. 4k). We also detected significant differences among the three groups in synoptic components of the lower troposphere including W850, and in local meteorological parameters such as the sea breeze index (SBI) suggested by Biggs and Graves (1962), which is defined as, SBI=U2CpΔT, where U is Wsfc (Fig. 4c), Cp is specific heat, and ΔT is the temperature difference between T925 and the sea surface temperature. Thus, the SBI represents the ratio between inertial (ρU2/2) and buoyance forces (ρgcpΔT), where ρ is air density and g is gravity, and its value provides an indication of the likelihood of local circulation events such as sea breezes; at higher SBIs (i.e., SBI > 3), sea breezes cannot overcome the prevailing wind, whereas lower SBIs (i.e., 0 < SBI < 3) can indicate strong sea breezes.

In the example shown in Fig. 4l, the SBI for groups 1–3 was 0.1±4.5, 0.1±9.2, and -0.2±12.5, respectively. Most SBIs in group 1 ranged from 0 to 3, indicating that group 1 corresponded to the dominant local circulation (LD), whereas the SBIs in group 3 had the lowest frequencies comparing 1 and 3, which corresponded to a dominant synoptic-scale circulation (SD). Group 2 can be considered a mixture of local and synoptic-scale circulation (MD). These results indicate that Seosan may experience frequent LD conditions (with sun on 1/3 of the days of the year), with infrequent SD conditions (1/5 of all days).

Data collected via aircraft during GMAP-2020 are summarized in Table 2. A total of nine aircraft measurements were conducted during the campaign period (12 November 2020–20 January 2021). Four of nine flights were conducted under LD conditions, and the remaining flights (except that on 27 November 2020) were conducted under MD conditions. No aircraft measurements were consistent with SD conditions during the GMAP-2020 campaign.

We examined spiral segments from each flight over Seosan during 11:00–17:00 KST to exclude marginal effects of diurnal variation in NO2 (Fig. 2). The overall results indicated that the vertical O3 profiles were relatively constant in the PBL, whereas NO2 profiles appeared to be highly dependent on meteorological conditions. We compared data collected during flights conducted under LD (one flight) and MD conditions (two flights) during the GMAP-2020 campaign, to examine differences in the vertical structures of the PA and SI observations.

Aircraft measurements of vertical NO2 and O3 profiles for flights FL-5 (6 December) and FL-6 (8 December) under LD conditions are shown in Fig. 7, along with 24 h backward trajectories starting at different altitudes (100, 500, and 1000 m). All observed NO2 profiles shown in Fig. 7 appeared to have generally exponential curves, with anomalous features at higher altitudes. For example, when vertical turbulent mixing prevailed within the PBL (O3 profile, Fig. 7b), the data were fitted with an exponential vertical curve, and the anomalous NO2 layer aloft was found to have a height of 1.5 km, which was higher than the estimated PBLH of 1.2 km. HYSPLIT 24 h backward trajectories starting at 12:00 KST showed that all air mass from the surface to the lower free atmosphere was transported over the Yellow Sea via the Shandong Peninsula (Fig. 7c). This finding suggests that the anomalous NO2 layer aloft was not produced locally (i.e., from local LPS emissions), but instead traveled via long-range regional-scale transport. This transport of NO2 across the region was also discussed and might be particularly high during the winter when the NOx lifetime is relatively longer (Stohl et al., 2002; Wenig et al., 2003; Lee et al., 2013). According to Anmyeondo lidar measurements for 6 December (http://kalion.kr, last access: 22 August 2022), the anomalous NO2 layer aloft corresponded well to an aerosol layer that appeared at ∼1.0 km at approximately 12:00 KST, persisting until 2200 KST. However, based on a cross-comparison of our data, high surface levels of SI-NO2 (>∼4 ppb; Fig. 7a) were influenced more by local LPS than by that in the atmosphere aloft due to long-range transport (Fig. 7a).

Aircraft measurements for flight FL-7 (9 December) under LD conditions are shown in Fig. 7b. The NO2 vertical profile exhibited an exponential curve, with an anomalous peak at ∼600 m immediately above the top of the simulated PBL. HYSPLIT backward trajectory data starting at 12:00 KST showed that the non-surface air had a different origin from the surface air (Fig. 6d), indicating that the anomalous NO2 plume likely traveled from coal-fired power plants in a nearby industrial city (Taean) northwest of Seosan. This finding indicates a distinct vertical structure of higher NO2 at the surface due to strong local emissions, whereas lower NO2 levels were observed at higher altitudes, with anomalously high NO2 levels in some layers aloft due to medium-range transport from nearby areas. Thus, despite the limited number of aircraft measurements, the elevated anomalous NO2 structure that was observed intermittently led to a negative correlation between PA-NO2 and SI-NO2. The discrepancies imply that vertical profile distribution study should proceed cautiously when only surface measurements are obtained under LD meteorological conditions.

Aircraft measurements were conducted under MD conditions on flights FL-1 (26 November), FL-3 (28 November), and FL-8 (12 December) (Fig. 8). We applied several regression models (linear, exponential, and polynomial) to three vertical structures, and obtained two distinct NO2 vertical profile patterns from the surface to the PBLH: decreasing linearly for FL-1 and FL-8 (Fig. 8), and constant with altitude for FL-3 (Fig. 8b). None of the three cases showed anomalous layers above the PBLH, similar to the exponentially declining profiles obtained under LD conditions (Fig. 7). These vertical structures observed under MD conditions may have been induced by strong vertical mixing within the PBL, supplemented by prominent surface photochemical losses at the same time. The vertical O3 profile during FL-1 showed a decoupled structure, with different patterns within and above the PBL (Fig. 8d); however, the other 2 d showed uniform distributions, with no particular anomalous features between the upper PBL and surface atmosphere (Fig. 8b, c, e, and f). The observed daily maximum sensible heat fluxes measured at Seosan (Fig. S3) were much higher for FL-3 (175.9 W m-2) than FL-1 and FL-8 (118.9 and 102.0 W m-2), suggesting that vertical turbulent mixing was much more prominent during FL-3. These chemical and physical characteristics are all related to MD conditions. Thus, the higher coefficient of determination (R2=0.64) obtained under MD conditions (Fig. 5b) has an important bearing on the absence of irregular or anomalous layers aloft, with little variation regardless of the shape of the curve (Figs. 7 and 8).

Figure 9 shows examples of PC-NO2 and SI-NO2 diurnal variation under LD (FL-5 and FL-7) and MD (FL-1 and FL-8) conditions, and Fig. 10 shows latitudinal mean distributions for FL-5 and FL-7, based on the aircraft measurement data shown in Figs. 7 and 8. PC-NO2 was decoupled from SI-NO2 on 2 d, FL-5 and FL-7, which were both classified as having LD conditions (Fig. 9a and b), whereas good vertical mixing and uniform NO2 distribution were observed on the remaining 2 d, FL-1 and FL-8, which showed MD conditions (Fig. 9c and d). According to our analysis of the aircraft measurements (Fig. 7), the poor correlations between PC-NO2 and SI-NO2 captured by FL-5 and FL-7 were mainly due to an NO2 polluted layer transported aloft, as described in Sect. 4.3.