South Korea's GEMS is the first satellite instrument in the GEO constellation and is monitoring AQ over Asia. GEMS was launched in February 2020 by Arianespace from the French Guiana Space Center. Like the TEMPO instrument, GEMS was built by Ball Aerospace & Technologies Corp. GEMS retrievals of O3, NO2, SO2, HCHO, glyoxal (CHOCHO), and aerosols are derived from ultraviolet–visible (UV–vis) measurements (Kim et al., 2020), and the cloud fraction necessary for data filtering is also available for each observation (Choi et al., 2020; Kim et al., 2024). Each day, the field of regard (FOR) shifts westward with the Sun; this provides measurements over India at the end of the day at the expense of losing coverage over Japan when the solar zenith angles become too large. Total column (TotC), stratospheric column (StrC), and (for some products) tropospheric column (TrC) values are retrieved in up to 10 hourly observations during daytime according to the season with a spatial resolution at Seoul of near 7 km × 7.7 km for gases and cloud, whereas this resolution is 3.5 km × 7.7 km for aerosol and surface reflectivity.

The main retrieval challenges for NO2 (e.g., Palmer et al., 2001; Bucsela et al., 2013; Lorente et al., 2017; Geddes et al., 2018; Zara et al., 2018; van Geffen et al., 2020) are (1) the conversion of the observed NO2 slant column density to an inferred vertical column density using an air mass factor (AMF) and (2) the separation of the StrC and TrC. These steps can both lead to significant uncertainties and biases as is well-documented in studies comparing the varying results from different NO2 retrieval algorithms using the same OMI (Ozone Monitoring Instrument; Levelt et al., 2018) satellite measurements (Zara et al., 2018). The AMF is primarily a geometric conversion based on the observation angles, but this must also consider other factors including cloud and aerosol information, terrain reflectivity, and vertical gas profile (Lorente et al., 2017). A CTM is used in this AMF calculation and also in the separation of the StrC and TrC (Lee et al., 2020; Geddes et al., 2018). The NO2 a priori profiles in this GEMS data version are provided by the GEOS-Chem model (Bey et al., 2001). Although beyond the scope of this paper, retrieval sensitivity studies and product validation for the new GEO composition measurements will be important to minimize any aliasing of diurnal variation in the retrieval input parameters onto the diurnal variation in the retrieved products themselves (e.g., Yang et al., 2023a; Kim et al., 2023; Szykman and Liu, 2023). This includes, for example, parameters that impact the AMF, a priori assumptions that change by location and time; angular dependences of surface and cloud reflectivities; vertical profiles of trace gases and aerosols; meteorology; and PBL evolution.

This work uses the publicly available Version 2 (V2.0) NO2 Level-2 data obtainable from the Korean National Institute of Environmental Research (NIER) (NIER, 2024). An example of the coverage is shown in Fig. 1. The data quality flags used are those recommended by the algorithm development team (Lee et al., 2022). These are as follows: FinalAlgorithmFlags ≤1; CloudFraction <0.3; and SolarZenithAngle and ViewingZenithAngle ≤70°. The nominal GEMS pre-launch accuracy for the NO2 TrC is 1×1015 (molec.cm-2) (Kim et al., 2020). The Level-2 data contain an error term for the NO2 spectral fit, but this does not account for the AMF calculation error and the conversion to vertical column. Analysis for TROPOMI (van Geffen et al., 2022) identified the latter as the main source of product error (around ±25 % over polluted regions). Comparison of the GEMS L2 NO2 with other measurements is further discussed in Sect. 4. Updates and enhancements to the GEMS retrieval algorithms are ongoing and will result in new data versions being released periodically.

TROPOMI is a push-broom imaging spectrometer on ESA's Sentinel-5 Precursor satellite in a Sun-synchronous orbit with a 13:30 local standard time Equator crossing (Veefkind et al., 2012). TROPOMI achieves close-to-global daily coverage at resolutions down to 3.5km×5.5km, depending on the species. NO2, HCHO, carbon monoxide (CO), SO2, O3, methane (CH4), aerosol, and cloud are retrieved from UV–vis and reflected shortwave infrared measurements. We use the TROPOMI operational Level-2 data from Collection 3 that are publicly available through the NASA Earthdata portal (Earthdata, 2024).

To capture time-resolved measurements of highly variable species such as NO2 in a coordinated manner, NASA initiated a large-scale global monitoring network of (quasi-)autonomous stations with the ground-based remote sensing spectrometer system called Pandora (Herman et al., 2009; Spinei et al., 2018). ESA joined this project in 2018 to form the Pandonia Global Network (PGN), which ensures systematic processing and dissemination of the data in support of AQ monitoring and satellite validation. Pandora instruments measure in the UV–vis range and retrieve column amounts of several air pollutants including NO2, HCHO, and O3. Data are publicly available (PGN, 2024).

The Multi-Scale Infrastructure for Chemistry and Aerosols (MUSICA) is a new community CTM for simulations of large-scale atmospheric phenomena in a global modeling framework, while still resolving chemistry at emission- and exposure-relevant scales (Pfister et al., 2020). In this work, we use MUSICA Version 0 (MUSICAv0), which is a configuration of the Community Atmospheric Model with Chemistry (CAM-chem) (Tilmes et al., 2019; Emmons et al., 2020) using a spectral element (SE) grid with regional refinement (RR) (CAM-chem-SE-RR) (e.g., Lauritzen et al., 2018; Schwantes et al., 2022). MUSICAv0 is run with a horizontal resolution of 0.0625° (∼7 km) over refined regions selected to cover the Korean and wider Asian domain (Jo et al., 2023) and allows near matching of the GEMS pixel resolution of 7km×7.7km over Seoul (see Fig. S1 of Jo et al., 2023). Chemical processes are all simulated in CAM-chem, which includes chemistry feedback on the meteorology (e.g., aerosol–cloud interactions). To reproduce the dynamics for the January and June 2023 months analyzed, the capability of CAM-chem to nudge the model meteorology to the GEOS-5 0.25° resolution reanalysis outside of the refined domain is used following Jo et al. (2023). Inside of the refined domain, the wind fields are calculated by the model. We stress that this work does not aim for exact model simulations of the GEMS data. Rather, the model is used to investigate the processes driving the observed NO2 diurnal variation.

Date-specific anthropogenic, biomass burning, and biogenic emissions are used in the simulations. The Copernicus Atmospheric Monitoring System version 5.1 (CAMS-GLOB-ANTv5.1) global emission inventory (Soulié et al., 2024) serves as the base anthropogenic emissions inventory along with the NIER/KU-CREATE inventory for East Asia and the Korean Peninsula that was produced for the Korea–United States Air Quality Study (KORUS-AQ) field campaign in May–June 2016 (Jang et al., 2019; Park et al., 2021; Crawford et al., 2021). Global biomass burning emissions are provided as 0.1° resolution daily averages by the Quick Fire Emissions Dataset (QFED) version 2.5_r1 (Koster et al., 2015) and Fire INventory from NCAR (FINN) v1.5 (Wiedinmyer et al., 2011). Biogenic emissions are simulated using the Model of Emissions of Gases and Aerosols from Nature (MEGAN) version 2.1 algorithm (Guenther et al., 2012), which is incorporated in the Community Land Model (CLM) (Lawrence et al., 2019) and calculated at each model time step using the model meteorology. Global inventories of anthropogenic emissions are usually provided as monthly means that are temporally interpolated to a particular day by the model. However, the diurnal variation in emissions also becomes important at the high spatial resolution of the MUSICAv0 refined grid. Diurnal emissions profiles have been derived for different sectors by country in the Emissions Database for Global Atmospheric Research (EDGAR) and can be applied to other inventories (Crippa et al., 2020; Jo et al., 2023).

For our simulations over Seoul (presented in Sect. 5.1), we consider the diurnal variation in emissions using profiles based on area–point and mobile sectors developed for KORUS-AQ and described in application to modeling over Seoul by Jo et al. (2023). This is shown in Fig. 2 and predicts a pronounced increase in emissions with daytime activity and rush-hour peaks in the mobile emissions. As is discussed in Sect. 5.1, the assumed diurnal profile of model emissions can make a significant difference to the calculated TrC. Here, we include Fig. 2 showing the shape of the diurnal emissions to help in the interpretation of the results of Sect. 5.1 (particularly Fig. 10).

Quantitative comparison of satellite retrievals and model simulations requires consideration of the measurement sensitivity to the target quantity and the retrieval a priori assumptions. These can be accounted for by applying the retrieval averaging kernels that formally relate the retrieved quantity to the true atmosphere (Rodgers, 2000). For retrieval–model comparisons of trace gas TrC from GEMS-like sensors, the averaging kernels allow a retrieval-consistent TrC to be derived from the model trace gas profile or, alternatively, a model-consistent TrC to be calculated from the retrieval by substituting the retrieval a priori trace gas profile used in the AMF calculation for the model profile (e.g., Boersma et al., 2016). However, there are known issues with the GEMS V2.0 processing of the NO2 averaging kernels and the values reported in the operational data files, such that these should not be used as per guidance from NIER. The intention of the GEMS project is to remedy this issue with the next GEMS Version 3 release. Any independent or alternative reprocessing of the GEMS data to calculate averaging kernels is beyond the scope of this work. As a result, we are unable to perform quantitative model comparisons of the GEMS V2.0 products at this time, and the comparisons with MUSICAv0 presented in Sect. 5 should be considered mainly qualitative for now.

An example of the 10 GEMS daytime retrievals of NO2 TrC over Northeast Asia (25–45° N, 115–130° E) for the relatively clear day of 15 June 2023 is shown in Fig. 3. High NO2 is seen over the Beijing region and the industrial areas of the North China Plain. Pollution is also high over Shanghai and the Yangtze Delta with another hotspot over Seoul in South Korea. The most striking first impression of this GEO perspective on atmospheric composition is how large the temporal variation in the pollution is with respect to magnitude as well as how much the spatial distribution shifts on an hour-by-hour basis. In certain locations, changing cloud fields do not permit for all 10 possible retrievals to pass the data filters. However, the hourly observations do provide at least some measurements, thereby demonstrating an advantage of the GEO perspective. During the winter months when the Sun is low in the sky, there are fewer GEMS hourly retrievals. An example is shown in Fig. 4 for 30 January 2023 NO2 TrC over Northeast Asia. Compared with Fig. 3 (which uses the same color scale), the NO2 burden is considerably higher because of reduced NOx photochemical loss and NO2 TrC buildup during the day. This results in less diurnal variation compared with the summer months and is discussed in Sect. 5. The east–west data stripes evident in Fig. 3 exist in this data version across the domain, similar to the spurious across-track variability issue for OMI. Zhang et al. (2023) comment that this is likely associated with the specific scan modes of GEMS as well as periodically occurring bad pixels.

We have developed quantitative measures of the magnitude of the GEMS NO2 TrC daily absolute and relative variation. The first of these – ADV (absolute daily variation) – represents the absolute change in NO2 TrC for a given day and pixel location as sampled by multiple hourly GEMS observations: 2ADVi,j=∑k=1n-1(TrCi,j,k-TrCi,j,k+1), where the pixel longitude and latitude are indexed by i and j, respectively, and k is the index of the GEMS hourly observation from 1 to n, with n being the number of useful cloud-free observations during the day. ADV is calculated for adjacent hourly observations, regardless of gaps due to missing data (gaps along a slope would have no effect on the resulting ADV value, although gaps where the slope changes sign may result in an underestimation). Figure 5 shows the monthly average of ADV for June 2023. Because this quantity depends on the number of useful cloud-free retrievals during the day, it is only calculated for days and locations for which there are at least five hourly observations; thus, it represents a lower bound on the total TrC change. No values are mapped east of ∼130° E, as there are too few daytime hourly observations at these locations. High variation usually coincides with locations of high TrC, such as industrial regions and cities, and illustrates the importance that time-resolved observations will have for characterizing changing NOx emissions and population pollution exposure. Chinese and Korean cities and Indian power facilities are particularly noticeable. Ship tracks between Hong Kong and Singapore as well as between Sri Lanka and the northern tip of Sumatra have been previously identified in GOME, SCIAMACHY, OMI, and TROPOMI NO2 data, among others (Beirle et al., 2004; Richter et al., 2004; Franke et al., 2009; Georgoulias et al., 2020). The ship tracks in Fig. 5 show diurnal variation, most likely because of horizontal dispersion with the increasing afternoon marine boundary layer. Averaging data temporally in this way at a given location reduces the contribution of transient emission or transport events that can produce significant day-to-day variation in the NO2 TrC. The averaged diurnal variation is then primarily dependent on emission and chemistry processes that occur on most days and, thus, indicates polluted urban regions in particular.

A second measure that we consider relates the magnitude of the GEMS NO2 TrC relative daily variation (RDV) over multiple hourly observations with respect to the value of the single observation that would be obtained from a LEO instrument such as TROPOMI. For a particular day and pixel location, the absolute deviation in the day's hourly observed NO2 TrC relative to the observation at t=13:45 LT (closest to the TROPOMI overpass time) is calculated and normalized by the t = 13:45 value: 3RDVi,j=1TrCi,j,t∑k=1n(TrCi,j,k-TrCi,j,t)2n, where t is the reference time (13:45 local standard time) and n is the number of observations at pixel location i,j excluding the observation acquired at the reference time. The daily RDV can then be averaged over the month, as shown for June 2023 in Fig. 6. The monthly averaged RDV is seen to be large, often > 50 % of the 13:45 LT value. The spatial distribution of RDV is similar to that of ADV shown in Fig. 5; however, in this case, it illustrates the diurnal uncertainties in emissions or exposure that might be expected from assuming estimates based on LEO observations. As for the absolute diurnal variation, because this quantity depends on the number of useful cloud-free retrievals during the day, it again represents a lower bound on the TrC variation. There are also no values at locations such as Japan and India where there is no 13:45 LT observation. The apparent high variation over the Pacific Ocean is a result of normalizing by the relatively low NO2 TrC values in this region.

The January 2023 monthly average of the GEMS NO2 TrC ADV and RDV over the Northeast Asia region are shown in Figs. A1 and A2, respectively. In general, reduced winter photochemistry results in a smaller diurnal variation in January compared with June. However, over polluted regions, the ADV in NO2 TrC has similar values due to the higher January TrC. A noticeable region of high absolute daily change is also seen in Cambodia which requires further investigation but may be explained by fires. The Fig. A2 RDV map coverage is limited because the 13:45 LT observation is only available between longitudes 113 and 132° E.

We have examined the time series of the GEMS NO2 TrC for various regions and seasons. Figure 7a shows the June 2023 time series spatially averaged over Northeast Asia (see Fig. 1 for the regional context). At this time of year, there are a maximum of 10 daylight hourly data points at the center of the GEMS FOR, with fewer points at the eastern and western edges. The number of hourly data points is further reduced by cloud filtering. On a day-to-day basis, the calculated average TrC appears noisy because of the changing amount and nonuniformity of cloud-free coverage, especially when polluted urban areas might be included in the spatial average on one day but not on the next. Because of this, the monthly time-averaged diurnal variation (Fig. 7b) is the quantity often considered. However, it is important to show the individual daily TrC after filtering for cloudy data to indicate the information that will usually be available for AQ applications. It should also be noted that cloudy missing points in the daily data tend to “flatten out” the apparent diurnal variation shown in the monthly time average; this suggests that this quantity be treated with care, as it may not capture the full dynamic range of the diurnal variation of individual days. Despite these considerations, a consistent diurnal cycle is seen on those days with multiple data points. This shows NO2 TrC decreasing through the morning hours, reaching a minimum in the early afternoon, and then increasing again late in the afternoon. Little difference is seen between weekdays and weekends at this spatial scale. As discussed in Sect. 5, we attribute this summer cycle primarily to photochemistry being the dominant driver of diurnal variation, as diurnal variation due to different local-scale emissions and meteorology is averaged out at this regional scale. This photochemical diurnal cycle is even more apparent when averaging over larger geographical regions, although different local times, solar zenith angles, and photolysis rates then complicate interpretation. A similar diurnal cycle is also seen in clean regions away from local sources or transported pollution plumes.

The NO2 diurnal variation is less consistent at the local scale where, in addition to photochemistry, changing emissions and, more importantly, meteorology determine day-to-day variability. Figure 7c shows the time series spatially averaged over Seoul, South Korea (37–38° N, 126–127.5° E; see Fig. 1 for the regional context). Prior to the launch of GEMS, stagnation events over Seoul had been seen to cause a buildup of pollution during the day, leading to an afternoon maximum in NO2 TrC. This was observed by the Geostationary Trace gas and Aerosol Sensor Optimization (GeoTASO) aircraft spectrometer (Leitch et al., 2014) during the KORUS-AQ field campaign (Judd et al., 2018; Crawford et al., 2021). GEMS observations sometimes show this same diurnal pattern; however, as discussed in Sect. 5, they more often show a photochemical 10:00–11:00 LT morning maximum in NO2 contributed by urban emissions, followed by a decrease, and then a small late-afternoon increase, as shown in the monthly time-averaged diurnal variation in Fig. 7d. The weekend values indicate a similar diurnal cycle to weekdays with smaller TrC. We note that other work has shown large differences in the NO2 diurnal variation seen by GEMS between weekdays and weekends for different Asian cities (J. Park et al., 2022).

Reduced diurnal variation is shown in the January 2023 GEMS time series spatially averaged over the Northeast Asia and Seoul regions, as shown in Fig. 8. As noted above, the apparent large day-to-day differences in TrC result primarily from the varying number of cloud-free data that enter the spatial average, especially for the large-scale Northeast Asia region. Over both regions, an increase in TrC is usually observed during the day because of reduced winter photochemistry, as discussed in Sect. 5. This is the case for both weekdays and weekends, although the weekday TrC values are now clearly greater, indicating the persistence of NO2 resulting from higher weekday NOx emissions.

We have used the MUSICAv0 model to simulate the NO2 TrC for the same Northeast Asia and Seoul regions discussed above for June and January of 2023, and the model setup is as described in Sect. 2.4. For each case, we performed two simulations as a sensitivity test of the assumed anthropogenic diurnal emissions profile: the first assumes constant anthropogenic emissions during the day (labeled “Base”), whereas the second uses the KORUS-AQ area–point and mobile sector diurnal emissions profiles described in Sect. 2.4 and shown in Fig. 2 (labeled “Diurnal”). The average daily NO2 TrC diurnal variation for weekdays is shown in Fig. 10. The time windows corresponding to the periods during which GEMS retrievals are shown in Figs. 7 and 8 are indicated by the unshaded hours. As is noted in Sect. 2.4, there is an issue with the GEMS V2.0 NO2 averaging kernels which should not be used following the guidance of NIER. As a result, the discussion here comparing MUSICAv0 simulations to the GEMS retrievals presented in Sect. 3.3 should be considered to be mainly qualitative. However, we believe that the model is still able to provide insights into the drivers of the NO2 TrC diurnal variation and that presentation of these results is worthwhile as it sets the stage for more quantitative analysis with the next GEMS data version.

We first consider the model results spatially averaged over the Northeast Asia study region and temporally averaged for weekdays in the months of June and January 2023 (Fig. 10a and c, respectively). The model patterns for the NO2 TrC diurnal variation are generally similar to the corresponding GEMS (Figs. 7b and 8b). However, as the model does not include cloud cover and reproduces a consistent daily pattern, the dynamic range of the diurnal variation is greater than the averaged GEMS data. The GEMS data are biased high relative to the model in both months with the difference being largest in January. This may be a GEMS V2.0 retrieval issue, as noted above, and/or an underestimation of NOx emissions in MUSICAv0. At this regional scale, the diurnal variation is similar for both the Base and Diurnal simulations, despite the very different anthropogenic diurnal emissions profiles, and suggests that photochemistry is the main driver. However, the magnitude of the diurnal variation during the GEMS retrieval window does depend on the emissions profile, ranging from 25 %–50 % in June to 6 %–17 % in January. We note that the GEMS NO2 TrC daily relative variation over Northeast Asia for June and January 2023 (Figs. 6 and A2, respectively) falls at the bottom of these ranges, suggesting that further work is needed to understand the greater model variation. The emissions profile also affects the local hour of the minimum NO2 TrC and is about 1 h earlier in the Diurnal case, which matches better with GEMS.

Over the Seoul region, the difference in NO2 diurnal variation between the Base and Diurnal simulations is large, especially in June (Fig. 10b). For the Base simulation in June, the shape of the diurnal variation shows a similar photochemical cycle to that described above for the Northeast Asia region. A difference is seen in that there is a NO2 morning peak around 10:00–11:00 LT, a decrease through midday, and then an afternoon minimum, like the diurnal variation seen in the corresponding GEMS data (Fig. 7d). This is discussed further in the next section. In contrast, the Diurnal simulation shows a very different NO2 buildup throughout the day and reflects the shape of the NOx diurnal emissions profile used in the model; however, the pronounced rush-hour peaks seen in Fig. 2 are not evident in the NO2 TrC average, even though they do clearly show up in the calculated surface concentration (not shown). Nevertheless, the fact that the average GEMS NO2 diurnal variation shown in Fig. 7d more closely resembles the constant NOx emissions of the Base simulation suggests that the hourly changes in the emissions profile may not actually be as large as indicated in Fig. 2. In January, both simulations show increasing TrC during the GEMS retrieval window, in agreement with the GEMS data in Fig. 8d. These simulations indicate that the modeled diurnal variation at the city scale will be very dependent on the diurnal profile of the assumed emissions, and the accurate characterization of the assumed emissions will be an important part of effectively using the hourly observations from GEO.

The underlying photochemical cycle is explained by considering the June 2023 Base simulation TrC NOx budget analysis shown in Fig. 11a for Seoul. After a buildup of NO2 during nighttime, photolysis begins with sunrise at 05:00 LT, and the morning decrease is mirrored by the increase in NO with the rise in the NO2 photolysis rate (jNO2) curve with increasing solar elevation. Although the NOx ratio is in a photochemical steady state on timescales of minutes, the diurnal change in solar irradiance drives a continual change in the ratio through the day. It should also be remembered that this calculation is for the TrC relevant to the GEMS retrieval and is, therefore, representative of a vertically weighted average, rather than surface values. After 10:00 LT, NO loss with the buildup of O3 pushes the NOx ratio toward NO2 at the same time that NOx is lost to the nitrogen reservoirs HNO3, and to a lesser extent, PAN. This balance results in a slight peak in NO2 around 11:00 LT; a continued decline due to NOx loss is then seen until the minimum at 16:00 LT. Decreasing photolysis results in a subsequent NO2 increase into nighttime with continuing NOx emissions. The underlying chemistry is similar for the corresponding June Diurnal simulation (not shown) only in this case, as the increasing NOx emissions during the day result in the replacement of most of the atmospheric NOx that is lost to the nitrogen reservoirs with a consequent gradual buildup of NO2.

In the January 2023 Base simulation shown in Fig. 11b, atmospheric NOx shows less variation under conditions of lower photolysis, lower O3, and less daytime conversion to HNO3. The NOx ratio depends mainly on the change in the jNO2 curve under conditions of limited photochemical activity, resulting in a 11:00 LT NO maximum and NO2 minimum (as shown in Fig. 10d). In the January Diurnal simulation (not shown), NO2 again builds up during the day following the increasing NOx emissions.

The MUSICAv0 monthly averaged daily NO2 TrC diurnal variation over Seoul (shown in Fig. 10b and d) results from averaging the diurnal variation during each day of June and January 2023. As noted before, the Base simulation usually shows an afternoon minimum, whereas the Diurnal simulation shows a buildup of pollution during the day and produces a late-afternoon NO2 peak. Both simulations indicate that day-to-day changes in the NO2 TrC magnitude depend primarily on meteorology. This is illustrated in Fig. 12, which shows the correlation of the average model NO2 TrC during the day with the model surface-layer wind (usually around ∼ 120 m) and also indicates the average wind direction over the city. This is shown separately in plots for June and January calculations for the 2 years of 2022 and 2023 together. For both months, there is a clear anticorrelation (R about -0.7) of NO2 TrC and wind speed. In June, the prevailing wind over Seoul is usually from the south. Shifts to winds from the west result in a lower wind speed and stagnant conditions over the city that permit a buildup of pollution, and this is reflected in the TrC value. Similar anticorrelation is seen in January, although the winds over Seoul during winter are mostly from the northwest and an occasional change to a weak anticyclonic pattern results in a low wind speed and NO2 TrC buildup.

A similar analysis for the modeled diurnal variation absolute daily change in NO2 TrC discussed in Sect. 3.2 does not show clear correlation (R of about -0.35) with wind speed in either month over Seoul. This suggests that the emission and chemistry processes discussed above are most important in determining the local diurnal variation and do not necessarily require stagnant meteorological conditions. We note that Yang et al. (2023b) did see correspondence between a low model wind speed and higher winter NO2 diurnal variation; therefore, further investigation using wind measurements would be useful. We have also examined the contribution of incoming transport to the NO2 TrC, although Seoul is unlikely to be a generally representative city in this respect, as it is large, fairly isolated, and has very high local NOx emissions. Upwind GEMS NO2 TrC values are usually about 3–4 times lower than the values retrieved over the city and have relatively flat diurnal variation, suggesting that incoming transport contributions to the Seoul NO2 TrC diurnal or day-to-day variations will be small. This might not be the case for longer-lived pollutants or when considering concentrations at a specific altitude, such as in the free troposphere (Jordan et al., 2020). In cleaner regions downwind of Seoul, the GEMS NO2 TrC measurements do show higher values because of plumes following a high-pollution day over the city. Understanding the role of meteorology is going to be important for our next step of relating the satellite retrievals of TrC to surface concentrations. Results from KORUS-AQ showed that meteorology and PBL dynamics play a large role in determining the extent to which the satellite and ground-based in situ views of pollutant diurnal variation can be reconciled (Crawford et al., 2021).