Comparison of Total Column and Surface Mixing Ratio of Carbon Monoxide Derived from the TROPOMI/Sentinel-5 Precursor with In-Situ Measurements from Extensive Ground-Based Network over South Korea

: Atmospheric carbon monoxide (CO) signiﬁcantly impacts climate change and human health, and has become the focus of increased air quality and climate research. Since 2018, the Troposphere Monitoring Instrument (TROPOMI) has provided total column amounts of CO ( C TROPOMI ) with a high spatial resolution to monitor atmospheric CO. This study compared and assessed the accuracy of C TROPOMI measurements using surface in-situ measurements ( S KME ) obtained from an extensive ground-based network over South Korea, where CO level is persistently affected by both local emissions and trans-boundary transport. Our analysis reveals that the TROPOMI effectively detected major emission sources of CO over South Korea and efﬁciently complemented the spatial coverage of the ground-based network. In general, the correlations between C TROPOMI and S KME were lower than those for NO 2 reported in a previous study, and this discrepancy was partly attributed to the lower spatiotemporal variability. Moreover, vertical CO proﬁles were sampled from the ECMWF CAMS reanalysis data (EAC4) to convert C TROPOMI to surface mixing ratios ( S TROPOMI ). S TROPOMI showed a signiﬁcant underestimation compared with S KME by approximately 40%, with a moderate correlation of approximately 0.51. The low biases of S TROPOMI were more signiﬁcant during the winter season, which was mainly attributed to the underestimation of the EAC4 CO at the surface. This study can contribute to the assessment of satellite and model data for monitoring surface air quality and greenhouse gas emissions.


Introduction
Major sources of atmospheric carbon monoxide (CO) include the incomplete combustion of fossil fuels, biomass burning, and the oxidation of methane and non-methane hydrocarbons, predominately activated by the hydroxyl radical (OH). CO is removed by photochemical oxidation, which consumes OH during the process [1,2], thus affecting the atmospheric cleansing capacity [2] and lifetime of methane (CH 4 ) [3,4]. In addition, this reaction produces greenhouse gases such as carbon dioxide (CO 2 ) and tropospheric ozone (O 3 ); therefore, CO is regulated by worldwide air quality standards and is designated a significant greenhouse gas with a radiative forcing of 0.23 W m −2 [5]. The lifetime of CO varies from weeks to months [6], which is long enough to persist through horizontal and vertical transport but too short to be well mixed globally. Owing to the moderate lifetime of CO, it is frequently utilized as a tracer for the propagation of pollution [7,8]. For these reasons, the Monitoring Atmospheric Composition and Climate (MACC) project of the Global Monitoring for Environment and Security (GMES) program prioritized CO as an important chemical species for air quality and climate studies [9]. Nadir-viewing passive sensors provide global distributions of CO retrievals from either near-infrared or thermal-infrared (TIR) radiances. Since the first measurement of CO during four flights of the space shuttle between 1981 and 1999 [10], the measurement of pollution in the troposphere (MOPPIT) has provided decades of global CO retrievals since 2000 from the 1-0 CO absorption band at 4.7 µm [11]. These TIR measurements are sensitive to CO in the middle troposphere and depend on the spectral resolution and thermal contrast in the lower troposphere. The Atmospheric Infrared Sounder [12] onboard the Aqua launched in 2002, the Tropospheric Emission Spectrometer (TES) [13] onboard the Aura launched in 2004, and the Infrared Atmospheric Sounding Interferometer [14] onboard the Meteorological Operational (METOP) also utilize this TIR absorption band of CO.
For clear atmospheric conditions, the shortwave-infrared (SWIR) earth-radiances near the first overtone 2-0 absorption band of CO (between 2.30-2.39 µm) is negligibly affected by scattering in the atmosphere but is dominated by atmospheric absorption and surface reflectance. Therefore, SWIR measurements are sensitive to the total column amount of CO along the light path, making CO retrievals using these wavelengths suitable for detecting emission sources of CO. In addition to the more recent progress of the MOPITT using its SWIR measurements [15], the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography [16] on the Envisat satellite has provided continuous time series of global CO SWIR measurements since 2002. Worden et al. (2010) combined the TIR and SWIR measurements of MOPITT to retrieve global CO trends and assessed its theoretical information content, which showed increased retrieval sensitivity near the surface compared with those using a single band [17]. In October 2017, the TROPOspheric Monitoring Instrument (TROPOMI) onboard the Sentinel-5 Precursor (S5P) of the European Space Agency (ESA) was launched and continues to measure CO using SWIR radiances with higher spatial resolution and better radiometric performance [18]. Moreover, the TROPOMI allows for the detection of weak regional sources, such as individual wildfires, from its daily overpasses.
The retrieval sensitivity of CO near the surface is critical for the operational use of satellite data for air quality and climate applications, as its emissions and major chemical interactions occur within the boundary layer. In addition to the nadir-viewing sensors, instruments on solar occultation satellites such as the Atmospheric Chemistry Experiment-Fourier Transform Spectrometer (ACE-FTS) on board SCISAT [19][20][21], or of a limb viewing geometry including the Michelson Interferometer for Passive Atmospheric Sounding (MI-PAS)/ENVISAT [22] and Microwave Limb Sounder (MLS) /Aura [23] provide informative CO profile retrievals. However, CO retrievals from these sensors are limited by their lower horizontal resolution and coverage compared with those from the nadir viewing instruments; therefore, they are not suitable for accurately identifying regional emissions. Retrievals using both SWIR and TIR radiances show promising results with high sensitivity near surfaces [17]; however, to the best of our knowledge, these retrieval data are not currently available as an operational product. To overcome the limitations of satellite products, previous studies have combined model simulations and column retrievals from satellites to derive surface concentrations of aerosols [24] and to trace gases [25]. Zhang et al. (2020) reported that the annual mean values of the MOPITT CO over Asia decreased significantly at a rate of 0.58 ± 0.15% per year from 2003 to 2017 and associated this decrease with reduced biomass burning over southeast Asia during the spring season [26]. Similar results were reported by Buchholz et al. (2021), who demonstrated a decreasing global CO trend of approximately 0.5% per year between 2002 and 2018 based on MOPITT data. They also attributed the significant decline in CO over Northeast China from 2002 to 2018 to improvements in combustion efficiency [27]. Zheng et al. (2018) suggested that decreased CO emissions in China from four primary sectors (iron and steel industries, residential sources, gasoline-powered vehicles, and construction materials industries) could be responsible for 76% of the inversion-based trend of east Asian CO emissions [28]. Kang et al. (2019) estimated that the anthropogenic contribution of CO decreased to approximately 94% from 2001 to 2011 over east China [29]. Figure 1 shows the mean total column amounts of CO over east and southeast Asia for 2019 from the TROPOMI, which were binned to a 0.05 • × 0.05 • horizontal grid. To calculate the average values, the CO data with a quality flag greater than or equal to 0.5, were sampled. As shown in this figure, significant amounts of CO prevailed over east China throughout 2019, which also affected downwind regions, including the Korean peninsula [30]. Jeong and Hong (2021) derived surface-level NO 2 by combining the TROPOMI and reanalysis data to assess long-term exposure for epidemiological studies [25]. They compared the estimated NO 2 with an extensive ground-based network over South Korea managed by the Korean Ministry of Environment (KME). To the best of our knowledge, only a few studies have compared satellite-retrieved and ground in-situ CO measurements, despite their significance for assessments of satellite retrievals [31,32]. This study is a follow-up study of [25], which aimed to compare and assess CO products of the TROPOMI (C TROPOMI ) for complementing surface measurements using an extensive ground-based network over South Korea, and thereby to contribute to the improvement of our understanding of the air quality impacts of CO and provide a guideline for climate studies.

TROPOMI Total Column Density of CO
The TROPOMI is the unique payload of the S5P satellite mission and has measured reflected solar light by the Earth using two spectrometer modules since 2017: one covering the ultraviolet-visible (270-495 nm) and near-infrared (675-775 nm) spectra and the other covering the SWIR between 2305 and 2385 nm. The SWIR spectrometer was developed by Surrey Satellite Technology Limited, United Kingdom, and has a spectral resolution of approximately 0.25 nm with a sampling resolution of approximately 0.1 nm. The TROPOMI also measures the Sun directly through the irradiance port and internal diffuser for calibration [18,33].
The SWIR measurements of the TROPOMI feed the Shortwave Infrared CO Retrieval (SICOR) algorithm to retrieve total CO column amounts and effective cloud parameters (i.e., cloud optical thickness and cloud center height) [34,35]. The SICOR algorithm is based on the SCIAMACHY heritage [36] and is improved for cloudy and aerosol-loaded atmospheres. The inversion utilizes a profile-scaling method based on monthly averaged vertical profiles of CO from the global chemistry transport model, version 5 (TM5) [37]. Moreover, it generates vertically integrated columns of CO with an averaging kernel for each retrieval [35], which are tested extensively using SCIAMACHY measurements and cover the TROPOMI spectral range with a similar spectral resolution [38]. The SICOR algorithm consists of two steps. In the first step, the SICOR algorithm retrieves the total amount of CH 4 from the TROPOMI radiances between 2315 and 2324 nm to filter optically thick clouds and aerosols assuming a non-scattering atmosphere. A full-physics algorithm retrieves C TROPOMI in the second step from radiances between 2324 and 2338 nm. The CH 4 retrievals from the first step were used to derive the effective cloud parameters at this stage. One of the merits of the SICOR algorithm is that it provides reliable retrievals for cloudy conditions because the sensitivity of the measurement to the CO above the cloud is utilized to retrieve C TROPOMI assuming a certain vertical profile shape from the TM5 [35]. In addition, the high reflectance of the cloud enhances retrieval sensitivity.
Borsdorff et al. (2018) compared C TROPOMI with the European Center for Medium-Range Weather Forecasts (ECMWF)/Integrated Forecasting System (IFS) products of the Copernicus Atmosphere Monitoring Service (CAMS), which assimilates IASI and MOPITT observations of CO [35,39]. Both CO observations show a marginal mean difference of 3.2 ± 5.5% with a Pearson correlation coefficient (r) of 0.97. Martínez-Alonso et al. (2020) compared C TROPOMI to the MOPITT and airborne (ATom, Atmospheric Tomography mission) datasets, which showed excellent agreement with a mean bias of less than 3.73% [40]. C TROPOMI also showed good agreement with ground-based Total Carbon Column Observing Network (TCCON) measurements, with a mean bias of about 6.2 ppb [41]. In general, the accuracy and precision of the CO data product meets the level 2 user requirements: within an accuracy of <15% and a precision with ≤10%.

Surface Network of CO Measurements
The KME has monitored particulate matter, NO 2 , CO, O 3 , and SO 2 since the 2000s from extensive surface air quality monitoring stations in South Korea. In 2019, 569 stations measured the surface mixing ratios of CO. These stations are predominately situated at ambient locations in urban and rural areas far from major roadways and typically deployed on the roofs of public buildings with fewer than five stories. To monitor roadside air quality, several stations (41 in 2019) are situated near major roads with a height of approximately 2.5 m above the ground level. The KME measures CO mixing ratios based on a nondispersive method using CO analyzers (model 3008, Dasibi Environmental Corp.; US Environmental Protection Agency reference method RFCA-0488-067) with a lower detection limit of 0.1 ppm and response time of 120 s. Linearity of the detector is better than 1%, and span drift is about ±1% for 24 h and ±2% for one week. Instruments are inspected monthly. The standard inspection procedure consisted of a two-step process: first, abnormal samples were screened based on the conditions of the instrument (i.e., calibration, inspection, or malfunction). Next, data exceeding the normal range or rate of change were screened [42,43]. Five minutes of temporal resolution of the raw data was averaged hourly after the quality assurance procedures and then reported to the public [42].

ECMWF Atmospheric Composition Reanalysis 4
The 4th generation global CAMS reanalysis data of the ECMWF (EAC4) assimilates the total column CO, tropospheric column NO 2 , aerosol optical depth, and total column/profiles of O 3 from satellite retrievals to furnish the three-dimensional fields of these species [44]. The EAC4 covers the period from 2003 with a three-hour temporal resolution and a horizontal resolution of approximately 80 km (0.75 • × 0.75 • ) at 60 vertical model grids. The EAC4 assimilates the MOPITT TIR total column CO (TCCO, Version 6) retrievals that are sensitive to those in the mid and upper troposphere [45]. We sampled the vertical profile shape of the CO from EAC4 to convert C TROPOMI to a surface-mixing ratio (S TROPOMI ) for comparison with the surface measurements (S KME ). Table 1 summarizes the measurement parameters for CO used in this study, obtained from different sources.

Comparison of Spatial Distributions of CO from TROPOMI and Ground Network
The mean values of C TROPOMI for 2019 over South Korea are shown in Figure 2 and were binned to a comparable resolution of the TROPOMI (0.05 • × 0.05 • horizontal grid). Panel (a) of Figure 2 represents South Korea, and panels (b) to (d) focus on the most significant emission areas of CO in the domain of panel (a). In general, high C TROPOMI values were observed in eastern South Korea, where the low C TROPOMI vales in Figure 2a were predominately observed over mountainous areas. Figure 2b depicts the values over the Seoul metropolitan area, where more than half of the Korean population (~26 million) is distributed. As shown in this figure, the TROPOMI clearly indicate high values of the C TROPOMI over Seoul, Incheon, and active ironworks in Dangjin. One of the largest industrial complexes in Gwangyang and the ironworks in Pohang resulted in a significant CO burden, as shown in Figure 2c,d, respectively. Large amounts of C TROPOMI over the western sea of the Korean peninsula are likely associated with trans-boundary transport from East China [30] (also see Figure 1). The annual mean value of the CO surface mixing ratio measured by the KME network in South Korea is shown in Figure 3. The domains of the panels in Figure 3 are the same as those shown in Figure 2. As shown in this figure, the KME network was densely distributed over highly populated areas, particularly in cities situated in the Seoul metropolitan area (Figure 3b). Such strategic distribution of the ground-based network is efficient for monitoring NO 2 , which is predominantly emitted from transportation in South Korea [25]. However, these network spatial distributions are not optimal for monitoring CO, as this compound is predominately emitted from industrial activities. As indicated in Figures 2 and 3, a vast number of stations over the Seoul metropolitan area demonstrate the spatial distribution of C TROPOMI (indicated by the comparison of Figures 2b and 3b), whereas the sparse distribution of surface measurements detected limited areas of the emission sources (as shown by comparing the lower panels of Figure 2 with those of Figure 3). Satellite retrievals, such as TROPOMI, can efficiently complement such limitations of ground-based networks.
The circles and squares in Figure 2 indicate ambient and roadside monitoring stations, respectively; Jeong and Hong (2021) reported significantly higher values of NO 2 from the roadside stations than the nearby ambient monitoring sites [25]. By comparing the values of the circles and squares in Figure 2b, we determined that unlike NO 2 , the CO mixing ratios measured at roadside stations did not show significant differences from those at ambient monitoring stations. This is likely due to the relatively longer lifetime of CO; the emitted burden of CO remains in the atmosphere for a sufficient period to be well-mixed within a boundary layer over the Seoul metropolitan areas. Therefore, C TROPOMI is likely to experience less horizontal heterogeneity within its footprint but is more closely related to boundary layer height.  Figure 4a compares the annual mean values of C TROPOMI and S KME over the KME stations. C TROPOMI values within ±0.025 • from each ground station were averaged for spatial collocation. Note that the C TROPOMI and S KME are not linearly comparable due to the spatiotemporal variabilities of vertical profile. However, as a major fraction of CO supposed to be distributed within the mixing layer, we expect such comparison may provide a primitive but basic assessment of the satellite retrievals before converting the C TROPOMI to surface mixing ratio for direct comparison. The green circles and red rectangles in Figure 4a represent the ambient urban/rural monitoring stations and roadside stations, respectively. The RMSE denotes the root-mean-square error, and the MBE represents the mean bias error. For a similar comparison for NO 2 , the annual mean values of the TROPOMI and surface measurements show a high correlation (r = 0.84), particularly over the ambient monitoring sites (r = 0.88) [25]. However, the correlation between C TROPOMI and S KME was lower (r = 0.37), partly attributed to the lower variability of CO compared to that of NO 2 . As discussed, the comparison for the roadside monitoring stations did not show a notable difference from that of the ambient sites (revealed by comparing the green circles and red squares in Figure 4a). Spatiotemporally coincident samples (C TROPOMI within ±0.025 • of the KME stations and S KME within ±30 min of the TROPOMI overpass time) of C TROPOMI and S KME in 2019 are compared in Figure 4b, and show a slightly lower correlation (r = 0.33) than that in panel (a). The correlation coefficients between C TROPOMI and S KME at each KME station in 2019 are shown in Figure 5. In general, a higher correlation appeared near the emission sources of CO owing to the higher retrieval sensitivity of the TROPOMI and variability of CO. For NO 2 , the correlations between the TROPOMI retrievals and KME measurements over the roadside stations (squares) were significantly lower than those over the ambient stations (circles) because of their higher spatiotemporal variability near the source areas [25]. Such differences were not observed for CO, as shown in this figure, which was attributed to its relatively longer lifetime, as shown in Figure 3. The TROPOMI retrievals utilize the profile-scaling method based on monthly averaged vertical profiles of CO from the TM5 [35,37], thus the ratio of S KME to C TROPOMI is assumed to be higher near strong emission sources: the CO mixing ratio at the surface of these areas is likely higher than that at ambient (well-mixed) stations. The ratios over the KME stations are shown in Figure 6 and were significantly high near emission sources in South Korea. The ratios along the coastal line were highly variable at each station, and could likely be attributed to complex boundary layer processes occurring over these areas (see Figure 6c,d). A similar complexity was observed for NO 2 [25], which again emphasizes the importance of intensive field campaigns combined with model simulations over these areas (e.g., ozone water-land environmental transition study [46]).

Estimation of CO Surface Mixing Ratio from TROPOMI and CAMS Reanalysis Data
To derive surface air quality from satellite data, Jeong and Hong (2021) utilized the ratio of surface mixing ratios to total column amounts from the EAC4, which are multiplied by C TROPOMI as follows [25]: where S TROPOMI is the estimated surface CO mixing ratio from C TROPOMI , and S EAC4 and C EAC4 are the surface mixing ratio and total column amount of CO from EAC4, respectively. As the CAMS model system (for EAC4) and TM5 (for C TROPOMI ) utilizes the same chemical mechanism, which is a modified and extended version of the CB05 [47,48], we expected the biases that arose from the different averaging kernels to be minimized. Furthermore, some of the systematic biases (e.g., emission inventory) could be canceled out because the ratio of S EAC4 to C EAC4 was relatively more accurate than their absolute values.  Figure 7 shows the annual statistics of the CO vertical profiles from the EAC4 of longitudes from 125 • to 131 • and latitudes from 33 • to 39 • . The black line with circles depicts the mean values of the CO mixing ratio at each layer, the dark gray area indicates the standard deviation (±σ), and the light gray area shows the data range (minimum and maximum values) at each level. The mean values with a ±σ of S KME for entire stations are indicated by the red circle with an error bar (466 ± 218 ppb), which was significantly higher than that of the EAC4 (193 ± 95 ppb). This difference is partially attributable to the spatial coverage of the KME network; most of the ground stations are located near urban areas or large emission sources, whereas the EAC4 values in this figure were calculated from data over the entire target region of South Korea. The mean (±σ) values of C TROPOMI over this target domain and over the KME stations were 24.8 (±3.6) × 10 17 molec. cm −2 and 25.3 (±4.5) × 10 17 molec. cm −2 , respectively. The horizontal heterogeneity of CO within a TROPOMI pixel is relatively small due to its moderate lifetime; therefore, the spatial coverage of the KME stations does not fully explain the difference between S KME and S EAC4 . Turquety et al. (2008) [49] compared the Laboratoire de Météorologie Dynamique, zoom; version 4 (LMDz) and Interactive Chemistry and Aerosols; version 2 (INCA) model simulations [50,51] to the Measurement of Ozone and Water Vapor on Airbus In-Service Aircraft (MOZAIC) aircraft-based in-situ profiles [52] over Asia, and reported relatively lower biases of CO from the model in the troposphere, suggesting the underestimation of CO emissions. The uncertainties of CO emissions over South Korea in the EAC4 may have propagated errors in the CO vertical profiles, particularly near the surface, which could have affected the difference between S KME and S EAC4 . The surface mixing ratios of CO from EAC4 and derived from TROPOMI were compared with the KME measurements in Figure 8. A comparison between Figures 4 and 8 reveals that the correlations of the CO surface mixing ratios between the different sources showed a higher correlation (r = 0.48-0.51) than that between S KME and C TROPOMI . Moreover, S TROPOMI shows a slightly higher correlation with S KME than that between S EAC4 and S KME with a statistical significance (z-score of about 4.04). The slope of the regression line between the S TROPOMI and S KME also shows slightly better consistency than that between the S EAC4 and S KME (t-value of about 6.9). Accordingly, the RMSE and MBE values between the S TROPOMI and S KME were lower than those between the S EAC4 and S KME , which quantifies the benefit of using TROPOMI to derive the surface CO mixing ratio. However, such agreement between the S TROPOMI and S KME was lower than that for NO 2 using the identical technique over the same spatiotemporal domain [25], and the low bias of S TROPOMI (MBE = −187.6 ppb) compared to S KME was still significant with respect to the average value of S KME (466 ± 218 ppb) which is discussed at following figures. Monthly mean values of C TROPOMI (red circles) and C EAC4 (green squares) are shown in Figure 9a. The dark colors indicate the mean values over the KME stations, and the lighter colors depict those over the entire domain of South Korea (longitudes from 125 • to 131 • and latitudes from 33 • to 39 • ). Over the KME stations, the mean C EAC4 values were lower than those of C TROPOMI throughout the year by approximately 10% (2.6 × 10 17 molec. cm −2 ). This could be partly attributed to the lower spatial resolution of the EAC4 (i.e., approximately 80 km) compared to that of the TROPOMI (i.e., approximately 7 km), as the greater collocated pixels of the EAC4 for each site may contain a greater fraction of background areas around the KME stations. The average values of C EAC4 and C TROPOMI over broader and identical spatial domains experienced these sampling issues to a lesser degree, as demonstrated by the light colors in Figure 9a. The mean values of C EAC4 over the entire target domain were 7% lower than those of C TROPOMI . Regarding similar comparisons, Borsdorff et al. (2018) reported biases of approximately ±15% depending on the region (see Figure 2 of [35]), and we expected that these biases were within the uncertainty ranges of C TROPOMI and C EAC4 . In addition, the monthly variations over the KME stations showed an excellent correlation (r = 0.98), as shown in Figure 9a.
As shown in Figures 7 and 8, significant underestimations of S EAC4 (by approximately 46%) and S TROPOMI (by approximately 40%) compared to S KME were also observed in the monthly mean values of these parameters throughout the year (Figure 9b). The black diamonds in Figure 9b depict the monthly mean values of S KME for 24-h measurements, and the blue squares indicate those for the TROPOMI overpass time. In general, the S KME values were high in winter and low in summer, despite the C TROPOMI peak observed in March (revealed by comparison of Figure 9a,b). The high C TROPOMI values in March were likely associated with active biomass burning over southeast Asia, whereas the S KME peak in January was attributed to the stable boundary layer during this period. The mean S KME values from 24-h samples and from the TROPOMI overpass time showed slight differences of approximately 6-10% in spring and winter but comparable values in summer, which was attributed to diurnal boundary layer development. Note that the NO 2 from the KME at the TROPOMI overpass time was consistently lower by approximately 23% than the 24-h mean values because of a combination of its chemical processes and boundary layer development during the daytime [25]. The monthly mean S EAC4 values showed generally similar tendencies (r = 0.89), but with significantly low biases throughout the year, particularly in winter. Such relatively low biases of S EAC4 resulted in similar degrees of underestimation of S TROPOMI , as presented in Figure 9b.

Summary and Discussions
This study aimed to assess C TROPOMI using an extensive ground-based network over South Korea to derive the surface mixing ratio of CO over the globe, key information in understanding its role in the regional air quality climate. Our analysis reveals that the CO concentration over South Korea is persistently affected by both local emissions and trans-boundary transport, emphasizing the importance of satellite-based remote sensing over the region. The TROPOMI accurately detected major sources of CO over South Korea (e.g., Seoul, Dangjin, Pohang, and Gwangyang), complementing the spatial coverage of ground-based networks. In general, the correlations between C TROPOMI and S KME (r = 0.33 for all coincident samples, r = 0.37 for annual mean values at each site) were lower than those for NO 2 reported in a previous study [25], and this observation was partly attributed to the lower spatiotemporal variability. Moreover, higher correlations were observed near the emission sources. We utilized vertical profiles from EAC4 to convert the total column amounts of CO from TROPOMI to the surface mixing ratio. This converted S TROPOMI was directly compared to S KME , which showed a significant underestimation of approximately 40%, with a moderate correlation of approximately 0.51. The relatively low biases of S TROPOMI were more significant in winter and were associated with the underestimation of S EAC4 . Turquety et al. (2008) also reported a significant underestimation of CO (by approximately 49% below 850 hPa) from the LMDz-INCA model compared to the MOZAIC aircraft measurements over highly polluted areas in Bangkok, Thailand. They suggested that part of this underestimation could be attributed to the relatively low horizontal resolution of the model (i.e., 3.75 • in longitude 2.5 • in latitude) [49], which may not accurately resolve highly polluted areas. Moreover, Khan et al. (2017) also reported a significantly low bias of the MOPITT CO compared to the KME measurement in Seoul, South Korea, by approximately 35%, with a low correlation of 0.28 [53]. To the best of our knowledge, the factors affecting the low biases of the surface CO mixing ratio from satellites and models over this region remain uncertain. Intensive field campaigns combining various chemistry models of high spatial resolution (comparable to that of the TROPOMI) and in-situ profile measurements (e.g., from aircraft or unmanned aerial systems) may help to better understand these discrepancies. Moreover, multi-band retrievals of CO using both SWIR and TIR [17] may also help to detect the surface burden of CO more efficiently over a broader region. One of the important merits of this study is that this method is applicable to other regions (e.g., other Asian or developing countries, where in-situ measurements are sparse) as the EAC4 and TROPOMI provides relatively uniform quality over globe. However, comparison studies between the satellite retrievals and surface measurements are essential for broader regions to understand uncertainties in the assumed CO profiles and emissions.
Author Contributions: Conceptualization, methodology, formal analysis, investigation, resources, data curation, writing, and editing by U.J. Software, validation, resources, writing and editing, visualization, and funding acquisition by H.H. All authors have read and agreed to the published version of the manuscript. Institutional Review Board Statement: Not applicable.