Air pollution is a significant issue affecting human health and the environment. Major air pollutants such as particulate matter (PM), nitrogen oxides (NO_x), ozone (O₃), and carbon monoxide (CO) become increasingly dangerous at higher concentrations. These pollutants cause respiratory and cardiovascular diseases and increase cancer risk over time (Kampa and Castanas, 2008). Air pollution is associated with reduced lung function, worsened asthma, and decreased life expectancy (Bernstein et al., 2004; Meo and Suraya, 2015), and also adversely affects ecosystems. Polluted air contaminates forests and water quality, acidifies soil, and disrupts ecosystem balance. Ozone can reduce plants’ photosynthetic capacity, potentially decreasing crop productivity (Taylor et al., 1994; Lovett et al., 2009; Greaver et al., 2012).

Air pollutants primarily originate from human activities. Fossil fuel combustion is a major source, emitting carbon dioxide (CO₂), CO, NO_x, sulfur oxides (SO_x), Volatile Organic Compounds (VOCs), and PM (van der A et al., 2008). Pollutants are emitted from transportation, heating, and industrial sectors, with varying characteristics by season. Spring sees increased yellow dust and fine dust, while summer experiences higher ozone levels (Jung et al., 2018). Fall accumulates pollutants due to atmospheric stability, and winter sees increased CO, NO_x, and PM due to heating. South Korean metropolitan areas face severe air pollution in winter due to fossil fuel use for heating (Nguyen et al., 2015).

To effectively monitor and manage air pollution that changes due to such complex and diverse factors, technology capable of continuously observing wide areas is necessary. In this context, satellite technology for detecting air pollutants is gaining attention. While traditional ground-based measurements provide high accuracy, they have spatial limitations and high costs (Rohde and Muller, 2015). In contrast, satellite-based air pollutant detection can simultaneously observe wide areas, helping understand global-scale air pollution phenomena (Martin, 2008). It also has the advantage of obtaining air quality information in areas with low population density or difficult access.

Satellite technology for detecting air pollutants has greatly improved over recent decades. In particular, sensors capable of measuring concentrations of major air pollutants such as aerosols, nitrogen dioxide, sulfur dioxide, and ozone have been developed (Duncan et al., 2014). Notably, the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor aboard National Aeronautics and Space Administration (NASA) Terra and Aqua satellites measures global Aerosol Optical Depth (AOD), playing an important role in understanding worldwide air pollution conditions (Levy et al., 2013). The Ozone Monitoring Instrument (OMI) sensor on the NASA Aura satellite significantly contributes to observing global ozone distribution (Levelt et al., 2018).

Additionally, the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis (ERA) Meteosat series provides AOD and volcanic ash information products at 10-minute temporal resolution through the Flexible Combined Imager (FCI) utilizing 16 spectral bands (Holmlund et al., 2021). South Korea Geo-KOMPSAT-2B (GK-2B) satellite monitors air pollution in East Asia at one-hour intervals, performing near-real-time monitoring (Choi et al., 2018).

Satellite observations offer great potential for air pollution monitoring through extensive spatial coverage and temporal continuity. However, limitations exist in accuracy due to technical constraints of satellite sensors, complex atmospheric scattering processes, and interference from surface characteristics. Therefore, reliability verification through comparison with ground observation data is required. Several studies have conducted comparative verification to assess satellite data reliability. Sherman et al. (2016) evaluated MODIS performance in North Carolina mountainous regions by comparing MODIS AOD with Aerosol Robotic Network (AERONET) AOD.

They found that MODIS tracks AOD variability well, but Terra satellite performance tends to degrade under low aerosol conditions. Dimitropoulou et al. (2020) evaluated the accuracy of tropospheric nitrogen dioxide (NO₂) columns measured by the Tropospheric Monitoring Instrument (TROPOMI) satellite by comparing them with ground-based Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) measurements in Brussels, resolving TROPOMI sensor NO₂ underestimation. Additionally, the accuracy of stratospheric ozone measurements was verified by comparing the Microwave Limb Sounder (MLS), which performs high vertical resolution stratospheric ozone measurements, with OMI satellite data. Results showed high accuracy with mean bias mostly within 5% and standard deviation within the 2–5% range (Liu et al., 2010).

Based on these previous studies, this research aims to evaluate the potential use of Geostationary Environment Monitoring Spectrometer (GEMS) environmental satellite data for effectively monitoring South Korea’s air pollution situation. Traditional ground monitoring networks have spatial limitations in realtime monitoring of wide areas. To complement this, this study analyzes South Korea’s air pollution situation using GEMS satellite data including NO₂, HCHO, ozone profile (O₃P), and AOD. In particular, we evaluate satellite data reliability and usability through correlation analysis with ground monitor measurements as reference values. Furthermore, this study analyzes the correlation between ground measurements and satellite data according to land cover ratios using mid-classification land cover maps. Through this, we assess the accuracy and applicability of satellite data across various land types including urban, rural, and forest areas. Through correlation analysis, we aim to examine the potential utility of GEMS satellite data for monitoring South Korean air pollution, which could contribute to the future development of air quality management systems following additional comprehensive validation studies.

South Korea is one of the countries with severe air pollution due to high industrialization and urbanization, with particularly high air pollutant concentrations in large industrial areas. South Korea’s representative air pollutants include PM₁₀, PM_2.5, NO₂, O₃, and CO, which can also generate secondary pollutants through complex chemical reactions in the atmosphere. To evaluate the reliability of GEMS satellite data that performs real-time detection of these substances, correlation analysis was conducted using GEMS satellite data, Airkorea ground observation data, meteorological variables from the Korea Meteorological Administration, and land cover data. Fig. 1 shows the distribution of 161 counties where correlation analysis was performed.

Figure 1. Distribution of 161 counties in Korea.

2.1. GEMS Data

GEMS is South Korea’s first geostationary environmental satellite that observes air pollutants in real-time over the East Asian region. This study uses the following GEMS Level 2 products: First, GEMS AOD is a dimensionless index indicating light attenuation by atmospheric aerosols, providing information about the amount and size distribution of aerosols in the atmosphere. Higher AOD values indicate higher atmospheric aerosol concentrations, signifying the presence of many particulate pollutants in the air. GEMS produces AOD at three wavelengths: 380 nm, 443 nm, and 550 nm. The 380 nm band, being in the ultraviolet region, is more sensitive to small aerosol particles, particularly useful for detecting Secondary Organic Aerosols (SOA) and sulfate aerosols (Bergstrom et al., 2007).

The 443 nm band, in the visible blue wavelength region, can detect various sizes of aerosol particles. It is effective in determining mid-atmospheric aerosol distribution and detecting marine aerosols and biomass-burning aerosols (Remer et al., 2005). The 550 nm band, in the visible green wavelength region, most closely corresponds to visibility perceived by human eyes. It particularly shows a high correlation with PM_2.5 and PM₁₀ concentrations and is most widely used in air quality research, as it receives relatively less influence from surface reflectance, making it advantageous for aerosol detection over land (Chu et al., 2003). Next, GEMS NO₂measures and provides tropospheric NO₂ concentrations.

NO₂ calculation is performed using the Differential Optical Absorption Spectroscopy (DOAS) method, calculating gas concentrations using light absorption data at multiple wavelengths, and estimating concentration through optical thickness according to the Beer-Lambert law. Main absorption cross-sections of NO₂ and O₃ are used to estimate accurate NO₂ concentrations. GEMS HCHO represents formaldehyde concentrations generated by the oxidation of VOCs in the atmosphere, serving as an important indicator of atmospheric chemical reactions. GEMS analyzes formaldehyde absorption characteristics in the 320–360 nm wavelength range using ultraviolet scattering data to calculate atmospheric HCHO vertical column density. This wavelength range minimizes interference from other gases like O₃ and NO₂ while accurately analyzing formaldehyde concentrations. GEMS O₃P represents the distribution of ozone concentrations according to atmospheric altitude. GEMS produces vertical ozone distribution information using backscattered ultraviolet measurements in the Hartley band (<~290 nm) and Huggins band (300–340 nm). In this study, GEMS data produced at 1-hour intervals during the daytime was collected and converted to monthly average data. The resulting monthly average images were then processed through Zonal calculations by 161 administrative districts of South Korea to produce monthly average data for each city/county.

2.2. Airkorea Data

This study utilized final confirmed air pollution data from Airkorea operated by the Korea Environment Corporation as ground measurement data. Airkorea measures and provides concentrations of various air pollutants in real-time at 1-hour intervals through general and special air pollution monitoring networks installed nationwide (https://www.airkorea.or.kr/web/). Airkorea air pollution data used for correlation analysis included CO, SO₂, NO₂, O₃, PM₁₀, and PM_2.5. PM is divided into PM₁₀ and PM_2.5 according to particle size, with smaller sizes having a greater impact on human health. PM_2.5 can penetrate deep into the lungs causing inflammation and affect the cardiovascular system through blood circulation. NO₂ is mainly emitted from traffic and industrial activities, causing respiratory diseases and acting as an ozone precursor. O₃ acts as a harmful pollutant in lower atmospheric layers, with concentrations tending to increase particularly in summer. CO is a colorless, odorless gas produced during incomplete combustion processes and can have fatal effects on human health at high concentrations.

These ground measurement data characteristics were averaged monthly and gridded through kriging. Fig. 2 shows the locations of 522 Airkorea monitoring stations used for kriging interpolation. While the monitoring stations are distributed across most cities and counties, their spatial distribution provides sufficient coverage for generating a continuous surface analysis. Subsequently, monthly average ground measurement data for 161 cities/counties were generated through zonal calculations.

Figure 2. The locations of 522 Airkorea monitoring stations used for kriging.

2.3. Land Cover Map

This study utilized large-scale land cover data provided by the Ministry of Environment (https://egis.me.go.kr/intro/land.do). The land cover data is geospatial data that systematically classifies nationwide land use states, providing important information reflecting various environmental characteristics and land use status. In particular, the large classification details major land cover categories, including urban, agricultural, forest, and water areas, making it useful for analyzing the distribution and characteristics of air pollutants. This study used this large-scale data to analyze the correlation between regional air pollution characteristics and land use patterns. Through this, we specifically evaluated the relationship between air pollutant concentrations and land cover types in certain regions and assessed the accuracy and applicability of satellite data according to various land types such as urban, rural, and forest areas.

The land cover map calculated the area ratios of urban areas, agricultural areas, forest areas, grasslands, wetlands, bare land, and water areas within each city/county. For example, quantifying land use characteristics of 161 cities/counties by calculating ratios like urban area 40%, agricultural area 30%, and forest area 30% for specific cities/counties.

2.4. LDAPS Data

LDAPS is a numerical forecasting system that performs eight times daily (00, 06, 12, 18UTC: 48-hour prediction, 03, 09, 15, 21UTC: 3-hour prediction) receiving boundary fields from the global forecasting model at 3-hour intervals. Its spatial resolution is 1.5 km, with 70 vertical layers up to approximately 40 km. The local forecasting model operates with its analysis-prediction circulation system using three-dimensional variational data assimilation techniques, and LDAPS output data is provided in three types: isobaric surface, model surface, and single surface data. This study used the K-Index variable from isobaric surface data and Boundary Layer Height (BLH), temperature, wind direction, and wind speed variables from single surface data.

K-Index is an indicator used to assess atmospheric instability, calculated considering vertical temperature distribution and humidity in the atmosphere. This index uses temperatures from three layers and dew point temperatures from two layers to diagnose instability factors in the middle and lower atmospheric layers. Eq. (1) shows the formula for calculating K-Index: where T₈₅₀ is the temperature at 850 hPa, T₅₀₀ is the temperature at 500 hPa, D₈₅₀ is dew point temperature at 850 hPa, T₇₀₀ is the temperature at 700 hPa, and D₇₀₀ is dew point temperature at 700 hPa.

Higher K-Index indicates higher convective activity potential, and during periods of high atmospheric instability, pollutants disperse more efficiently, reducing concentrations near the surface.

BLH refers to the height of the air layer directly influenced by the Earth surface and is an important indicator where most meteorological phenomena occur. BLH is closely related to pollutant dispersion, and as height increases, air quality tends to improve due to increased air volume available for mixing pollutants. In South Korea particularly, BLH rises in summer due to thermal heating and increased convection, facilitating pollutant dispersion, while in winter, BLH decreases, increasing pollutant concentrations near the surface (Li et al., 2021).

Temperature is a basic meteorological element representing the thermal state of the atmosphere, directly affecting chemical reactions related to secondary pollutant formation such as ozone. As temperature rises, photochemical reactions accelerate, increasing ozone production, and in urban areas particularly, high-temperature phenomena can further worsen air quality. In South Korea, strengthened photochemical reactions due to high summer temperatures lead to notably increased ozone concentrations (Sillman and Samson, 1995; Allabakash et al., 2022).

Wind direction and speed directly affect the movement and dispersion of air pollutants. Wind direction determines pollutant transport paths, with air quality in surrounding areas being significantly affected by wind direction from industrial complexes or large pollution sources. South Korea wind direction shows distinct seasonal differences, with westerly winds predominant in winter and more variable characteristics in summer. Increased wind speed promotes pollutant dispersion and dilution, positively affecting air quality improvement.

These meteorological elements act independently while also having interrelated effects, showing distinct seasonal variability. Particularly, the influence of these meteorological elements can vary by region depending on urbanization level and topographical characteristics, requiring a comprehensive approach considering such complex characteristics when analyzing air pollution correlations. Table 1 shows the data used in this study.

Table 1 . Summarizes the data used in the study.

Data	Variables	Spatial Resolution	Temporal Resolution	Resource
GEMS	AOD 380 nm	3.5 km × 8 km2	1 hour during daytime	National Institute of Environmental Research
	AOD 380 nm
	AOD 380 nm
	NO2	7 km × 8 km2
	HCHO
	O3P
Airkorea	CO	Points	1 hour	Korea Environment Corporation
	SO2
	NO2
	O3
	PM10
	PM2.5
LDAPS	K-Index,	1.5 km	3 hour	Korea Meteorological Administration
	BLH,
	air temperature,
	wind direction,
	wind speed
TCS	Traffic	Points	Daily	Korea Expressway Corporation
Land Cover Map	Large-scale land cover map	30 m	-	Ministry of Environment

2.5. Toll Collection System Data

Traffic volume data provided by the Korea Expressway Corporation includes daily and monthly traffic counts categorized by vehicle types (Class 1–5 and compact cars) at each Toll Collection System (TCS). In this study, the total traffic volume at each toll station was aggregated and converted to monthly average data. The processed monthly average traffic data were then used for kriging interpolation. The interpolated surfaces were then processed through zonal calculations across 161 administrative districts of South Korea to produce monthly average traffic data for each city/county. The traffic data includes comprehensive information such as collection date, hourly records, entrance/exit codes, Hipass system codes, highway operation codes, and business type codes, along with detailed vehicle class-specific traffic volumes.

2.6. Methodology

This study performed a systematic analysis to evaluate the reliability of GEMS satellite products. First, we collected GEMS satellite products (AOD, NO₂, HCHO, O₃P), Airkorea ground observation data (CO, SO₂, NO₂, O₃, PM₁₀, PM_2.5), Ministry of Environment land cover map, and LDAPS meteorological data (K-Index, BLH, temperature, wind direction, wind speed). In the data preprocessing stage, all data were converted to monthly averages and Airkorea point data were gridded through a kriging technique to unify temporal and spatial resolutions. Subsequently, zonal calculations were performed on all data for 161 administrative districts. HCHO underwent Min-Max normalization for consistency in analysis while other variables underwent Z-score normalization.

Finally, the accuracy of satellite data was verified by performing a correlation analysis between GEMS satellite products and ground observation data using representative values for each substance by city/county. Additionally, we calculated and analyzed ratios of land cover types by city/county to understand differences in satellite observation reliability according to land characteristics. We also confirmed the impact of atmospheric conditions on satellite observation accuracy through relationship analysis with meteorological conditions.

Fig. 3 is an overall flow chart of the research that shows the collection of GEMS satellite data, Airkorea ground observation data, LDAPS meteorological data, and land cover map, followed by data standardization and zonal calculations to perform correlation analysis between variables. Through this analysis, we quantitatively evaluated the reliability of GEMS satellite products and derived limitations and utilization methods of satellite observation according to meteorological conditions, providing basic data for effective utilization of GEMS satellite data.

Figure 3. Analysis processes used in this study.

2.7. Data Preprocessing

2.7.1. Z-score and Min-Max Normalization

In this study, standardized normalization methods were applied to analyze various air pollutant data from GEMS and Airkorea comprehensively. Z-score normalization was applied to all variables (AOD 380 nm, 443 nm, 550 nm, NO₂, O₃P, CO, O₃, NO₂, PM₁₀, PM_2.5) except HCHO, while min-max normalization was applied to HCHO considering its special distribution characteristics. Z-score standardization was performed through Eq. (2), where x is the original value, μ is the mean, and σ is the standard deviation. Min-max normalization was performed through Eq. (3). x is the original value, min (x) is the minimum value of x, and max (x) is the maximum value of x.

The main reasons for choosing these normalization methods are as follows: First, converting air pollutant data with different units and ranges to the same scale enables direct comparative analysis between satellite data and ground observation data, and prevents distortion due to scale differences between variables during correlation analysis. Second, Z-score normalization converts data to a normal distribution centered on 0 in standard deviation units, with ±1.96 values representing 95% confidence intervals, making outlier identification and analysis easier. Third, in spatial distribution visualization, expression with 0 as the median value can clearly show relative differences between regions, and statistically significant spatial patterns can be identified through the ±1.96 range, with color scale interpretation being intuitive. For GEMS HCHO, Min-Max normalization was applied as the data showed a very wide and asymmetric distribution. This enables more stable analysis by mitigating the influence of extreme values and ensures comparability with other variables through conversion to a 0–1 range. Additionally, it allows an intuitive understanding of statistical significance, enabling more effective interpretation and communication of research results.

2.7.2. Polygon-Based Data Standardization

In this study, for city/county unit analysis in South Korea, point-type data were rasterized using a kriging technique from geostatistics, and these raster images were analyzed by the administrative district unit using the spatial analysis technique of zonal statistics.

Kriging is an optimal spatial interpolation method that predicts values at unobserved points considering spatial autocorrelation of observed values. It can estimate predicted values and their uncertainty based on spatial structure identified through variogram analysis considering distance and directionality between observed values. It is particularly effective in estimating pollution levels in areas without monitoring networks as it can consider spatial continuity and regional characteristics of air pollutants. Ordinary kriging uses weighted linear combinations of surrounding observed values to predict values at unobserved locations, determining optimal weights that minimize error variance. The kriging process first calculates empirical variograms that quantify spatial correlation using distance and value differences between observation points. Subsequently, theoretical variograms are modeled by fitting theoretical models such as spherical, exponential, and Gaussian to empirical variograms, optimizing sill, range, and nugget parameters through Cressie (1985) weighted least squares method (Jeong et al., 2021).

Next, the zonal statistics technique was applied to aggregate raster analysis results by administrative district unit. This process converts values to representative values by administrative district by calculating average values of raster cells within city/county administrative boundaries, using the average of raster values within each administrative district boundary as the representative value for that administrative district.

In this study, the kriging technique was used to estimate air pollutant spatial distribution across South Korea using data from approximately 500 monitoring stations of Airkorea point-type data. For this, high-precision kriging results were obtained by determining variograms with optimal spatial autocorrelation using R autoKrige function. Subsequently, zonal statistics were consistently aggregated into 161 city/county unit spatial distributions for GEMS and LDAPS data in raster form and Airkorea data rasterized through kriging.

Based on standardized city/county unit data, correlation analysis was performed between GEMS satellite products and ground observation data, and satellite observation accuracy variability was evaluated according to land cover types and meteorological conditions. In particular, the impact of land cover characteristics such as urban areas, agricultural areas, and forest areas on satellite observation accuracy was analyzed, and relationships with meteorological conditions including boundary layer height, temperature, wind direction, and wind speed were identified.

Through correlation analysis between GEMS air pollutant data and Airkorea, LDAPS meteorological variables, TCS data, and land cover map (hereinafter Environmental Geographic Information System, EGIS) data, we evaluated the reliability of GEMS satellite data and reviewed its potential policy applications and atmospheric science applications. The analysis was conducted using monthly air pollution indicators by city/county from 2021 to 2023. Based on the constructed data, analysis was performed according to seasonal, regional, and land cover characteristics.

3.1. Seasonal Correlation Analysis for All Cities/Counties

First, seasonal correlation analysis was conducted for all 161 cities/counties. The analysis was performed using nine months of data, three months each from 2021 to 2023 (Spring: March-May, Summer: June-August, Fall: September-November, Winter: December-February). The Korean Peninsula shows distinct seasonality due to the East Asian monsoon, and its complex topographical characteristics combined with various weather patterns create significant spatiotemporal variability in the atmospheric environment. In particular, there are notable changes in atmospheric conditions by season, such as strong atmospheric circulation and heavy precipitation in summer, and the influence of cold continental high pressure in winter. These seasonal characteristics are formed through complex interactions of various meteorological elements such as temperature, humidity, wind fields, and precipitation, which significantly affect the spatiotemporal distribution of air quality concentrations. Therefore, analysis considering such seasonal variability is essential in validating satellite data reliability.

Spring in the Korean Peninsula is characterized by increased atmospheric instability as influence shifts from the winter Siberian high pressure to the North Pacific high pressure due to geographical characteristics. Additionally, it frequently experiences the effects of yellow dust originating from northern China and Mongolia regions. Dry surface conditions and strong surface heating cause atmospheric instability that promotes the long-range transport of dust particles, leading to rapid changes in atmospheric optical characteristics. The coastal regions are affected by sea and land breezes due to pressure system movement and mountainous terrain causing complex air flow patterns resulting in high complexity in air pollutant transport and dispersion processes.

Spring in the Korean Peninsula sees increased externallysourced fine dust like yellow dust due to westerly winds, pollen particle generation from plant blooming, and conditions favorable for secondary pollutant formation due to rising temperatures. Considering these complex atmospheric environmental characteristics of spring, performance evaluation of GEMS in detecting PM such as AOD, PM, and CO is required. Fig. 4 shows the correlation matrix between GEMS air pollutant data and EGIS data for the Korean Peninsula region in spring. First, GEMS AOD and LDAPS K-Index show a high negative correlation of about -0.6. K-Index is an atmospheric instability index where higher values indicate higher atmospheric instability.

Figure 4. Correlations between GEMS and EGIS for 161 cities during spring, focusing on GEMS AOD and Airkorea PM.

Due to this, when atmospheric instability is high, atmospheric circulation becomes more active, and when atmospheric circulation is active, the dispersion of airborne particles increases, causing AOD concentration to decrease. This mechanism is thought to explain the negative correlation between GEMS AOD and LDAPS K-Index. Next, GEMS AOD shows high positive correlations of about 0.57 and 0.6 with Airkorea PM_2.5 and CO, respectively. This relates to the AOD characteristic of indicating the degree of light scattering by atmospheric fine particles. PM is fine matter suspended in the atmosphere, and as these particles increase, light scattering increases, leading to higher AOD values. Additionally, CO is produced during incomplete combustion processes and is often emitted together with PM, showing similar behavior to PM, resulting in a high positive correlation with AOD. Particularly in spring, PM and CO from both long-range transport and local pollution sources tend to increase simultaneously, leading to more distinct correlations between these substances and AOD.

Fig. 5 shows the distribution of GEMS AOD 380 nm, Airkorea PM_2.5, and LDAPS K-Index. High concentrations of GEMS AOD and Airkorea PM_2.5 were observed in the Seoul Metropolitan Area and Gyeonggi Province regions. This is primarily attributed to large-scale pollutants such as yellow dust and fine dust that flow in due to westerly winds. Moreover, the air quality deterioration is exacerbated by the lack of washing effect due to low spring precipitation in the Korean Peninsula and atmospheric stagnation caused by continental high-pressure systems forming in the surrounding area (Lee et al., 2018). This atmospheric stagnation was also clearly evident in the LDAPS K-Index (Fig. 4c). The West Coast region of the Korean Peninsula shows lower atmospheric instability compared to the East Coast and Taebaek Mountains regions. Due to these stable atmospheric conditions, the phenomenon of atmospheric pollutants flowing in from external sources accumulating in these regions is intensified.

Figure 5. AOD and weather patterns in South Korea during spring: (a) GEMS 380 nm AOD, (b) Airkorea PM_2.5, and (c) LDAPS K-Index.

Summer in the Korean Peninsula is characterized by hot and humid weather conditions due to the North Pacific high-pressure system, which promotes photochemical reactions between NO_x and VOCs, leading to increased ozone concentrations in urban areas. Increased humidity during the monsoon season accelerates the formation of secondary aerosols such as sulfates and nitrates, and air pollutants tend to accumulate due to atmospheric stagnation after the monsoon season. Additionally, changes in heat wave and monsoon patterns due to climate change affect the spatiotemporal distribution of air pollution, while urban heat island effects and nighttime tropical nights increase pollutant concentrations.

Analysis results showed weak correlations between variables, which is related to the unique summer weather conditions of the Korean Peninsula (Fig. 6). The main factors are increased photochemical reactions and ozone formation of NO_x and VOCs occurring under strong solar radiation and high temperature conditions, increased precipitation during the monsoon season, and accumulation of pollutants near the ground due to atmospheric stagnation after the monsoon. Due to these complex factors, correlations between satellite and ground observation data are generally low in summer.

Figure 6. Correlations between GEMS and EGIS for 161 cities during summer.

To minimize these effects, monthly correlation analysis was performed for June, July, and August of each year (Fig. 7). Through relatively detailed temporal unit analysis, reducing the number of analysis variables enabled clearer identification of relationships between variables in complex atmospheric environments. Among summer air pollutants in Korea, ozone is evaluated as the substance requiring the most focused management. Ozone formation mechanisms and summer weather conditions are closely related. High temperatures and strong solar radiation in summer are directly related to ozone concentrations, particularly in June when the highest annual ozone concentrations are recorded. Analysis results showed GEMS O₃P observation data had a positive correlation of 0.59 with temperature, explaining that temperature rise promotes photochemical reactions of precursors like NO_x and VOCs, increasing ozone production. It also showed a high negative correlation of –0.71 with atmospheric stability, reflecting that O₃ concentrations remain high when the atmosphere is stagnant. Additionally, GEMS O₃P data showed a high correlation of 0.6 with Airkorea O₃ data, confirming the reliability of satellite observation data.

Figure 7. Correlations between GEMS and EGIS for 161 cities in June 2021, focusing on GEMS O₃P and LDAPS Air Temperature.

Fig. 8 shows the distribution of GEMS O₃P, Airkorea O₃, LDAPS temperature, and K-Index for June 2021. The GEMS O₃P and Airkorea O₃ observations in June 2021 showed generally high O₃ concentrations in the West Coast region, and the two observation values showed similar trends in most regions except for the Busan metropolitan area. This is primarily attributed to the promotion of photochemical reactions of precursor substances such as NO_x and VOCs due to rising temperatures and increased solar radiation in summer, particularly in June, and these characteristics were well reflected in the LDAPS temperature and K-Index distributions. Distribution analysis results showed high ozone concentrations in regions with high temperatures and low atmospheric instability, which is believed to be due to the continuous accumulation of ozone generated by atmospheric stagnation phenomena.

Figure 8. Ozeon and weather patterns in South Korea in June 2021: (a) GEMS O₃P, (b) Airkorea O₃, (c) LDAPS temperature, and (d) LDAPS K-Index.

Fall in the Korean Peninsula is influenced by relatively stable weather conditions and the inflow of cool, dry air. As temperature decreases and solar radiation reduces, the atmosphere stabilizes, limiting the vertical dispersion of pollutants, which can lead to increased concentrations. Particularly, as clear and calm weather becomes more frequent due to high-pressure influence, the atmospheric mixing height decreases, making it easier for air pollutants to accumulate near the surface. Yellow dust and foreign pollutants generated in China can also flow into the Korean Peninsula through westerly winds, potentially increasing concentrations of pollutants like fine dust.

Fall correlation analysis of air pollutants showed a negative correlation of -0.23 between GEMS NO₂ and LDAPS K-Index (Fig. 9). This is related to the fall characteristic atmospheric stagnation phenomenon. In fall, there is a strong tendency for atmospheric stabilization due to the influence of migrating highpressure systems, and as atmospheric stability increases, vertical mixing is suppressed, making it easier for NO₂ to accumulate near the surface. Particularly, temperature inversion phenomena due to low morning temperatures in fall further strengthen this atmospheric stagnation. Meanwhile, the high positive correlation between GEMS NO₂ and Airkorea NO₂ observations both demonstrates the reliability of satellite observations and shows characteristic distribution patterns of NO₂ in fall. In fall, there is a high positive correlation where NO₂ concentration in the atmosphere increases proportionally with increased vehicle traffic, showing that NO₂ emissions and formation mechanisms are closely related to traffic activity.

Figure 9. Correlations between GEMS and EGIS for 161 cities during fall, focusing on GEMS NO₂ and Airkorea NO₂.

Fig. 10 shows the distribution of GEMS NO₂, and Airkorea NO₂ concentrations along with traffic volume and LDAPS K-Index distribution in fall. NO₂ observed by both GEMS and Airkorea shows high concentrations in the Seoul Metropolitan Area and Busan Metropolitan Area, which display very similar spatial patterns to traffic volume distribution. Fall NO₂ concentrations are significantly influenced by anthropogenic emission sources such as traffic volume, which is confirmed by the high concentration distributions in the Seoul Metropolitan Area and Busan Metropolitan Area. Simultaneously, conditions of low atmospheric stability facilitate the accumulation of pollutants, contributing to increased NO₂ concentrations in urban areas. In particular, fall characteristic stable atmospheric conditions appear to further enhance NO₂ accumulation in urban areas.

Figure 10. NO₂ and weather patterns in South Korea during fall: (a) GEMS NO₂, (b) Airkorea NO₂, and (c) traffic volume.

3.2. Land Cover-Based Correlation Analysis

Next, we analyzed correlations between GEMS data and EGIS data according to land cover. Correlation analysis between GEMS data and EGIS data according to land cover is essential for understanding whether the spatial distribution and concentration of air pollutants appear differently according to land cover characteristics. This is because various land cover types, for example, urban areas, rural areas, and forest areas, can each differently promote emission patterns and movement and transformation of air pollutants.

3.2.1. Large Urban Areas

First, correlation analysis was conducted between air pollution data for regions with high urbanization ratios in the land cover map. Large urban areas have high frequency of air pollutant generation due to high population density and industrial activities, and high pollutant emissions due to increased traffic from large populations and concentrated industrial facilities. Additionally, climatic influences like urban heat island effects can further worsen air pollution levels. Particularly, air quality analysis in such urbanized areas plays an important role in understanding how land use characteristics affect air pollution and satellite observation accuracy. Through this, satellite observation data reliability can be evaluated according to land cover type, and effective air quality monitoring methods for urban areas can be proposed.

Fig. 11 shows the correlation matrix of January 2021 analysis results between GEMS satellite products and Airkorea ground observation data for the top 10 regions with high urbanization ratios in the land cover map. Analysis results showed that correlations between substances were very effectively reflected. First, LDAPS BLH variables show very low negative correlations with GEMS AOD variables, which occurs because when the atmospheric boundary layer is higher, pollutant dispersion is facilitated, causing a decrease in GEMS AOD concentration in the atmosphere. For this reason, a negative correlation between these two variables is evaluated. Next, high positive correlations were confirmed between GEMS AOD and PM substances and NO₂ in Korean major cities in winter. Such correlations are attributed to seasonal weather conditions and urban emission characteristics. In winter, the atmosphere maintains a stable state due to low temperatures limiting pollutant dispersion, and this creates an environment where pollutants easily accumulate near the surface as the atmospheric mixing height decreases. This causes increases in PM and NO₂ concentrations in the atmosphere, and GEMS AOD also shows correspondingly high concentrations. Additionally, in winter, atmospheric stability increases and mixing is limited, making it easier for pollutants to accumulate near the surface. Particularly in urban characteristics, high traffic volume, and heating demand are added, causing Formaldehyde and NO₂ concentrations from vehicle exhaust emissions to rise together. Under these conditions, GEMS HCHO and GEMS NO₂ coexist for long periods showing high positive correlation.

Figure 11. GEMS-EGIS correlations for 10 major cities, focusing on GEMS AOD and Airkorea PM_.

The 10 selected regions are Bucheon, Suwon, Seoul, Guri, Ansan, Incheon, Siheung, Seongnam, Anyang, and Busan, which were selected based on the highest proportion of urban used area in the land cover map. Of these, 9 regions are located in the capital area, with only Busan Metropolitan City outside the capital area (Fig. 12).

Figure 12. Distribution of urban dominant areas (>30%).

Fig. 13 shows the spatial distribution of winter GEMS AOD 380 nm, Airkorea PM₁₀, LDAPS BLH, and K-Index. GEMS AOD 380 nm and Airkorea PM₁₀ show a high positive correlation, with similar distribution patterns particularly observed in metropolitan areas. LDAPS BLH shows opposite distribution patterns as it has characteristics contrary to the generation and dispersion of air pollutants. In the case of the K-Index, while it is generally known to have a negative correlation with air pollution data, our correlation analysis showed a positive correlation (Fig. 10). However, this is difficult to interpret as a meaningful analysis result as it merely shows coincidentally similar patterns with AOD and air pollutant distribution, with the K-Index maximum value of –18.12 indicating very high atmospheric stability.

Figure 13. Air quality-weather patterns for 10 major cities: (a) GEMS AOD 380 nm map, (b) Airkorea PM₁₀ map, (c) land cover map, and (d) LDAPS BLH map.

3.2.2. Forest Dominant Areas

Forest dominant areas in South Korea’s Taebaek Mountains and South and East Sea regions play an important role in air quality and ecosystem health. Forests have the ability to absorb and reduce air pollutants, and analyzing air quality characteristics in areas with high forest ratios can scientifically verify forests’ air purification function. Additionally, by understanding how these regions’ characteristics affect air quality according to seasonal and weather conditions, we can understand forests’ environmental control function in response to climate change and contribute to air quality improvement in adjacent urban or industrial areas.

Analysis results show a high negative correlation between GEMS AOD and LDAPS K-Index and BLH (Fig. 14). This indicates that AOD concentration decreases as atmospheric instability is higher and the atmospheric layer available for pollutant dispersion is higher. Next, it shows a positive correlation between GEMS AOD and Airkorea pollutants. However, it shows relatively lower levels compared to large urban areas, which can be because forest areas have the function of absorbing or depositing atmospheric fine dust and pollutants, potentially reducing correlation with AOD as air pollutant concentrations become relatively lower in forest areas. Finally, GEMS HCHO and O₃ show negative correlations with Airkorea NO₂ and PM substances. This trend can be interpreted as HCHO and O₃ concentrations decreasing when NO₂ and PM concentrations increase in the atmosphere. NO₂ increase affects ozone formation, and PM can contribute to decreasing atmospheric formaldehyde concentration by increasing formaldehyde reactivity.

Figure 14. GEMS-EGIS Correlations for forest dominant areas.

70 cities/counties were selected as forest dominant areas, with the top 10 being Hwacheon, Yangyang, Samcheok, Jeongseon, Inje, Taebaek, Yeongwol, Hongcheon, Gunwi, and Pyeongchang, showing that most areas are distributed in Gangwon mountain regions (Fig. 15).

Figure 15. Distribution of forest dominant areas (>70%).

Fig. 16 shows the spatial distribution of spring GEMS AOD 380nm, forest-dominant area information from land cover data, Airkorea PM₁₀ and NO₂, and LDAPS K-Index and BLH in forest-dominant regions. In spring, the atmospheric boundary layer forms high in most regions, with particularly high atmospheric instability observed in the Yeongdong region and around the Sobaek Mountains including Jirisan. Both GEMS AOD and Airkorea PM₁₀ data show low values in most regions, suggesting relatively good air quality in these areas. Notably, these variables show a lower correlation compared to major metropolitan areas, which is analyzed to be due to the characteristics of forest regions. Forests play an important role in improving air quality as they have the ability to absorb and mitigate air pollutants such as fine dust, and are considered to contribute to reducing air pollutant concentrations by promoting atmospheric mixing and dispersion processes.

Figure 16. Air quality and weather patterns for forest-dominant areas: (a) GEMS 380 nm AOD, (b) Airkorea PM₁₀ , (c) LDAPS BLH, and (d) LDAPS K-Index.

3.2.3. Mudflat Areas

Mudflats are important ecosystem components with the ability to absorb and reduce pollutants, potentially contributing to regional air quality improvement. Particularly, mudflats located in Korea coastal areas are important natural resources responding to increasing air pollution problems from industrialization and urbanization, and identifying these mudflat functions enables more scientific and effective approaches to air quality management. Additionally, Korean mudflats receive high international evaluation for ecosystem services, as evidenced by their UNESCO World Natural Heritage listing.

Therefore, systematically analyzing correlations between air pollutants in mudflat ecosystems is an essential task for sustainable environmental management. The analysis was performed for Garorim Bay, which is a semi-closed bay located between Seosan City and Taean County in Chungcheongnam-do, with about 83% of its total area consisting of mudflats, making it the largest natural mudflat area on the west coast (Fig. 17).

Figure 17. Land cover map of the large mudflat areas in Seosan, South Korea.

Fig. 18 shows the correlation analysis matrix for spring between GEMS air pollutant data and EGIS ground observation data at the Garorim Bay location. As with previous analyses, GEMS AOD values show a negative correlation with LDAPS K-Index and BLH and also show a negative correlation with LDAPS Air Temperature. And it is evaluated to have a high positive correlation with Airkorea PM.

Figure 18. GEMS vs. EGIS correlation analysis matrix for mudflats areas. Focus on GEMS AOD and AK PM.

Fig. 19 shows the distribution of GEMS AOD 380 nm, LDAPS Air Temperature, and K-Index marking the location of Garorim Bay. The GEMS AOD value is evaluated to be approximately 0.35, while both air temperature and K-Index are evaluated to be low during this period.

Figure 19. Air quality and weather patterns for mudflats areas: (a) GEMS 380 nm AOD, (b) LDAPS K-Index, (c) Airkorea PM_2.5, and (d) Airkorea SO₂.

The Airkorea PM_2.5 and SO₂ values in this mudflat area are evaluated to be relatively higher compared to other regions (Fig. 20). This is analyzed to be due to the inflow of long-range transportable air pollutants from China, along with pollutants from industrial complexes in the Seoul Metropolitan Area and Chungcheong region being transported by westerly winds. Particularly, pollutants emitted from large-scale industrial facilities such as the Daesan Industrial Complex and Dangjin Thermal Power Plant located near Garorim Bay are also considered to be contributing factors.

Figure 20. Scatter plot of GEMS AOD vs. Airkorea PM_2.5 in industrial areas.

3.2.4. Industrial Area

The POSCO industrial complex in Pohang is the location of Korea’s largest steel mill, where various air pollutants generated from steel manufacturing processes significantly affect regional air quality (Fig. 21). Particularly, systematic correlation analysis is essential to understand the complex interactions and effects of SO_x, NO_x, and fine dust generated from steel processes. Through such analysis, causal relationships between pollutants can be identified, which can serve as scientific evidence for establishing effective air quality management policies and reduction measures.

Figure 21. Land cover map for the industrial complex area in Pohang, South Korea.

Fig. 22 shows the spring correlation analysis matrix between GEMS air pollutant data and EGIS ground observation data at the Pohang Industrial Complex location. Analysis results show that in industrial areas, there is a tendency for GEMS NO₂ and Airkorea CO to increase together. At this time, air quality observations at the Ulsan Industrial Complex showed CO concentrations of 0.35–0.45 ppm, only about 5% of the air quality standard (9 ppm), and GEMS NO₂ concentrations were also observed to be very low (Fig. 24). This suggests that air pollutant emission management is being effectively carried out in Korea industrial complexes.

Figure 22. Correlation matrix of GEMS vs. EGIS for industrial complex areas, focusing on GEMS NO₂ and Air Korea CO.

Figure 24. Scatter plot of GEMS NO₂ vs. Airkorea CO in industrial areas.

Fig. 23 shows the distribution of GEMS NO₂ and Airkorea CO marking the location of Pohang where the industrial complex is located. While the actual CO concentration is high at the location of the industrial complex, GEMS NO₂ is evaluated to be at low levels.

Figure 23. Concentration distribution of (a) GEMS NO₂ and (b) Airkorea CO in an industrial area.

Despite having relatively low spatial resolution, GEMS satellite observation data showed significant correlation and similar spatial distribution patterns when compared with kriged results from ground monitoring network (Airkorea) data. This suggests that while the GEMS satellite may not provide detailed concentration values for individual grids, it can effectively capture broad-scale distribution characteristics and spatiotemporal variability of air pollutants. Particularly considering the spatial constraints of ground monitoring networks, GEMS satellite data shows it can play a complementary role in understanding air quality conditions in monitoring blind spots and grasping broad-scale air pollution phenomena.

K-Index indicating atmospheric instability and atmospheric boundary layer height show close relationships with the spatial distribution and concentration changes of air pollutants observed by the GEMS satellite. Particularly, higher K-Index and boundary layer height facilitate vertical mixing of the atmosphere making pollutant dispersion easier, which is clearly confirmed in GEMS observation data. Pollutant concentrations observed by GEMS showed statistically significant correlations with these atmospheric stability indicators, suggesting that satellite observation data well reflects atmospheric dynamic characteristics.

Another notable point is that GEMS O₃P observations show consistent negative correlations with ground PM substances and NO₂ concentrations. This inverse correlation reflects the differential behavior of pollutants according to atmospheric vertical dispersion conditions, and consistent correlation patterns suggest that GEMS satellite data effectively represents the spatial distribution of air pollutants. These results suggest the possibility of building a more reliable air quality monitoring system through integrated utilization of satellite observation data and ground observation data in the future. Additionally, GEMS spatiotemporal continuity is judged to provide useful information for understanding air pollutant movement and dispersion patterns by complementing the spatial limitations of ground monitoring networks.

However, this study also has several limitations. These mainly stem from GEMS satellite data spatial resolution and seasonal characteristics. Due to the limited spatial resolution of GEMS, there were limitations in local spatial analysis based on detailed land cover maps or in identifying air pollution characteristics at the city/county level. Particularly, there were limitations in precisely analyzing high-concentration air pollution phenomena in industrial complexes located within individual cities/counties or air quality improvement effects of regions with natural purification functions like mudflats.

Additionally, South Korea’s unique weather conditions acted as important variables in seasonal analysis. Particularly, unusual weather like summer monsoons and typhoons, and complex atmospheric dynamic conditions in winter acted as factors making correlation analysis between GEMS satellite observation data and ground monitoring network data difficult. Such seasonal characteristics combined with various weather conditions, and the resulting complex behavior of air pollutants acted as factors limiting direct comparative analysis between satellite observation data interpretation and ground observation data.

Furthermore, while our seasonal analysis provided insights into temporal patterns, a more comprehensive temporal evaluation was needed. Although direct comparisons between GEMS and ground-based measurements are challenging due to differences in measurement units, we were able to analyze the temporal variation patterns of identical detection products using the complete three-year dataset (Fig. 25). Time series analysis revealed relatively strong positive correlations between GEMS NO₂ and ground-based NO₂ measurements (R≈0.7), as well as moderate correlations between GEMS O₃P and ground-based O₃ measurements (R≈0.5). These findings, despite the unit discrepancies between satellite and ground measurements, can serve as a foundation for future discussions and conclusions regarding GEMS accuracy assessment and validation methodology.

Figure 25. Comparison between GEMS products and Airkorea ground measurements (2021–2023).

In future research, several methodological improvements will be pursued to address the current limitations. To overcome the spatial resolution limitations of GEMS, we explore deep learning techniques such as super-resolution convolutional neural networks (SRCNNs) for spatial downscaling of satellite data, aiming to improve the resolution for more detailed regional analysis. For a better understanding of seasonal characteristics, we will integrate high-resolution numerical weather prediction (NWP) data with GEMS observations to better account for complex summer atmospheric conditions, particularly during monsoon periods. Finally, we intend to develop a more rigorous statistical framework by first standardizing the measurement units and establishing clear relationships between GEMS-derived products and ground-level pollutant measurements, followed by comprehensive uncertainty analyses. This standardization process will enable more effective quantitative validation through detailed statistical metrics across different temporal and spatial scales.

This study evaluated GEMS satellite air pollutant observation capability using GEMS satellite data, Airkorea ground-based observation data, LDAPS meteorological data, and the Ministry of Environment land cover map. Analysis results showed that despite spatial resolution constraints, the GEMS satellite effectively observes broad-scale distribution and spatiotemporal variability of air pollutants. Particularly GEMS AOD and O₃ showed high correlation with ground observation values and showed consistent correlations with meteorological variables like K-Index (atmospheric instability) and boundary layer height.

Seasonal analysis showed relatively high correlations in spring and fall. Especially in spring, a distinct positive correlation was observed between GEMS AOD and ground PM concentrations, and a negative correlation with K-Index showed that vertical dispersion characteristics of the atmosphere were well-reflected. However, correlation distinctiveness was difficult to find in some periods due to seasonal characteristics like summer monsoons and winter atmospheric stagnation.

In land cover type analysis, the high correlation between satellite data and ground-based observation data was confirmed in Urban areas. Particularly, a high positive correlation was observed between GEMS AOD and PM substances, NO₂, showing that urban area air pollution characteristics were well-reflected. Additionally, in forest-dominant regions, air pollutant concentrations were generally observed to be low, confirming forests’ air purification function. However, in regions with strong local characteristics like industrial complexes or mudflats, there were constraints in detailed analysis due to satellite spatial resolution limitations.

In analysis according to regional characteristics, GEMS data showed a significant correlation with ground observation values in both inland and coastal, mountain, and plain regions. Particularly in the west coast region, movement and accumulation phenomena of air pollutants due to spring westerly winds were effectively observed through satellite data.

This study shows that GEMS satellite data can be usefully utilized for monitoring South Korean air quality by complementing ground monitoring networks’ spatial limitations. Particularly, the possibility of GEMS satellite utilization was confirmed in monitoring broad-scale air pollution phenomena and tracking air pollutant movement paths. However, there are some constraints in microscopic spatial analysis or observation in specific seasons, suggesting additional research is needed to complement this. However, the results of this study can contribute significantly to the development of more intensive air quality management strategies, especially in areas with limited ground monitoring coverage, and policymakers can use the comprehensive spatial patterns revealed by GEMS to develop more effective regional air quality management strategies. It can help to establish quality improvement measures. Future effective air quality monitoring system construction is expected to be possible through the integrated utilization of satellite data and ground observation data.

This work was supported by a grant from the National Institute of Environmental Research (NIER), funded by the Ministry of Environment (ME) of the Republic of Korea (NIER-2024-01-02-040).

No potential conflict of interest relevant to this article was reported.