Interpreting Geostationary Environment Monitoring Spectrometer (GEMS) geostationary satellite observations of the diurnal variation in nitrogen dioxide (NO₂) over East Asia

Nitrogen oxide radicals $({\mathrm{NO}}_x\equiv \mathrm{NO}+{\mathrm{NO}}_{\mathrm{2}})$ emitted by fuel combustion are important precursors of ozone and particulate matter pollution, and NO₂ itself is harmful to public health. The Geostationary Environment Monitoring Spectrometer (GEMS), launched in space in 2020, now provides hourly daytime observations of NO₂ columns over East Asia. This diurnal variation offers unique information on the emission and chemistry of NO_x, but it needs to be carefully interpreted. Here we investigate the drivers of the diurnal variation in NO₂ observed by GEMS during winter and summer over Beijing and Seoul. We place the GEMS observations in the context of ground-based column observations (Pandora instruments) and GEOS-Chem chemical transport model simulations. We find good agreement between the diurnal variations in NO₂ columns in GEMS, Pandora, and GEOS-Chem, and we use GEOS-Chem to interpret these variations. NO_x emissions are 4 times higher in the daytime than at night, driving an accumulation of NO₂ over the course of the day, offset by losses from chemistry and transport (horizontal flux divergence). For the urban core, where the Pandora instruments are located, we find that NO₂ in winter increases throughout the day due to high daytime emissions and increasing ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio from entrainment of ozone, partly balanced by loss from transport and with a negligible role of chemistry. In summer, by contrast, chemical loss combined with transport drives a minimum in the NO₂ column at 13:00–14:00 local time (LT). Segregation of the GEMS data by wind speed further demonstrates the effect of transport, with NO₂ in winter accumulating throughout the day at low winds but flat at high winds. The effect of transport can be minimized in summer by spatially averaging observations over the broader metropolitan scale, under which conditions the diurnal variation in NO₂ reflects a dynamic balance between emission and chemical loss.

1 Introduction

The Geostationary Environment Monitoring Spectrometer (GEMS) satellite instrument was launched in February 2020 by the National Institute of Environmental Research (NIER) to observe air quality over East Asia. GEMS is the first geostationary instrument directed at air quality and provides hourly column measurements of several gases including nitrogen dioxide (NO₂) (J. Kim et al., 2020). NO₂ is part of the nitrogen oxides $({\mathrm{NO}}_x\equiv \mathrm{NO}+{\mathrm{NO}}_{\mathrm{2}})$ radical family, which is emitted by fuel combustion and whose chemistry plays a critical role in driving ozone (O₃) and fine particulate matter (PM_2.5) formation. NO₂ itself is of concern as an air pollutant. Loss of NO_x is by atmospheric oxidation by the hydroxyl radical (OH) and ozone, resulting in a lifetime of a few hours in summer and about a day in winter (Shah et al., 2020). The diurnal cycle of NO₂ measured from geostationary orbit offers unique information on the emission, chemistry, and transport of NO_x. Here we interpret the GEMS observations with the GEOS-Chem chemical transport model (CTM) to better understand the processes controlling this diurnal cycle.

Several studies have examined the diurnal variation in NO₂ in urban air using surface concentrations from air quality networks. The data typically exhibit bimodal maxima in the morning around 07:00–09:00 local time (LT) and in the evening around 19:00–21:00 LT, including over Beijing and Seoul (Cheng et al., 2018; H. C. Kim et al., 2020). This has been commonly attributed to high NO_x emission during morning and evening rush hours (Kendrick et al., 2015; Cheng et al., 2018), but urban NO_x emission inventories show little variation during daytime (Miao et al., 2020). Moutinho et al. (2020) found that the morning and evening NO₂ maxima could be driven by shallow mixing depths in contrast to the middle of the day and afternoon hours when surface heating maximizes the mixing depth. This diurnal maximum in mixing depth defines the planetary boundary layer (PBL) in daily contact with the surface. The PBL depth extends typically to 1–3 km altitude.

Ground-based measurements of NO₂ columns are available from the Pandora sun-staring spectrometer instrument network used for validating satellite observations (Herman et al., 2009; Kanaya et al., 2014; Judd et al., 2020; Verhoelst et al., 2021). Column measurements integrate concentrations from the surface to the top of the atmosphere and are therefore not directly sensitive to mixing depth. The Pandora network consists mainly of urban sites, where the NO₂ column and its variability are mainly within the PBL (Yang et al., 2023). The Pandora data from Seoul tend to show an increasing trend in the early morning followed by flat concentrations over the rest of the daytime, with less diurnal variation than NO₂ concentrations in surface air (Crawford et al., 2021). Nearby sites can show different diurnal variations, pointing to a major role of local transport in driving this variation (Chang et al., 2022; Kim et al., 2023).

Satellite observations of NO₂ from polar sun-synchronous low-earth orbit (LEO) have been made since 1995 starting with the GOME instrument (Martin et al., 2002) but observe by design at a single time of day. Several studies have combined observations from the SCIAMACHY or GOME-2 instruments observing in the morning at 09:00–10:00 LT and the OMI instrument observing in the afternoon at 13:00–14:00 LT to get some information on NO₂ diurnal variation. Boersma et al. (2008) found decreases from morning to afternoon over urban regions that they attributed to photochemical loss, as well as increases from morning to afternoon over tropical biomass burning regions that they attributed to a midday maximum in emissions. Boersma et al. (2009) found that the urban morning-to-afternoon decrease was largest in summer and absent in winter. Penn and Holloway (2020) found that NO₂ column ratios between morning and afternoon were lower than surface NO₂ concentration ratios, as would be expected from deeper vertical mixing in the afternoon. Ghude et al. (2020) found an important role for transport in driving morning-to-afternoon variations in NO₂ columns over urban India. Edwards et al. (2024) found that the diurnal variation in the NO₂ column from GEMS in June is driven by photochemistry at a regional scale and variability in emissions and meteorology at a local scale.

Here we analyze and compare the NO₂ diurnal cycles observed by GEMS over the Seoul and Beijing metropolitan areas in winter and summer. We compare them to the diurnal cycles observed by Pandora in the urban cores and to simulations with the GEOS-Chem CTM. We use GEOS-Chem to separate and quantify the roles of emission, chemistry, and transport in driving the NO₂ diurnal cycles observed from GEMS over different spatial scales. This work provides a basis for a more quantitative application of GEMS observations as top-down information on NO_x emissions and more generally for interpreting the diurnal cycle of NO₂ from geostationary orbit with application to the TEMPO instrument over North America launched in April 2023 (Zoogman et al., 2017) and the Sentinel-4 instrument over Europe to be launched in 2024 (Gulde et al., 2017).

2 Observations and model

2.1 GEMS data

GEMS is an ultraviolet–visible instrument measuring backscattered solar spectra at 300–500 nm (J. Kim et al., 2020). It was launched in February 2020 in geostationary orbit at a longitude of 128.25° E. We use hourly total NO₂ slant column density from the GEMS L2 NO₂ version 2.0 product at native 3.5 × 8 km² resolution for December–February (DJF) 2021/22 and June–August (JJA) 2022 (https://nesc.nier.go.kr/; NIER, 2023). The GEMS NO₂ algorithm uses differential optical absorption spectroscopy (DOAS) to fit backscattered solar spectra within the 432–450 nm range (J. Kim et al., 2020). This yields the slant column density along the light path (L2 data). We use all GEMS L2 NO₂ version 2.0 data that pass algorithm quality flag ≤ 112, final algorithm flag ≤ 1, solar zenith angle (SZA) < 70°, viewing zenith angle (VZA) < 70°, and cloud fraction < 0.3 (Lee et al., 2020).

The vertical column density of NO₂ is obtained by dividing the slant column density by an air mass factor (AMF) characterizing the photon path from the sun down through the atmosphere and back up to the instrument. The AMF depends on the viewing geometry and on the scattering properties of the atmosphere.

$$\begin{array}{}\text{(1)}& \text{AMF}={\text{AMF}}_{\mathrm{G}}{\int }_{\mathrm{0}}^{\mathrm{TOA}}w\left(z\right)S\left(z\right)\mathrm{d}z\end{array}$$

Here TOA is the top of the atmosphere, AMF_G is the geometric AMF defined by the solar zenith angle (SZA) and the satellite viewing angle (VZA) as AMF_G= sec(SZA) + sec(VZA), w(z) is the scattering weight that defines the instrument's sensitivity to NO₂ at altitude z, and S(z) is a normalized vertical profile of NO₂ number density called the shape factor (Palmer et al., 2001). Scattering weights are calculated with a radiative transfer model and increase with altitude (Martin et al., 2002; Yang et al., 2023). The shape factor is usually estimated with a CTM.

An alternative DOAS retrieval and AMF by Lange et al. (2024) improved the GEMS L2 NO₂ version 2.0 vertical column density product, which was biased due to using incorrect vertical profiles for AMF computation (Oak et al., 2024). Here, we use our own AMF. We use scattering weights at 448 nm compiled as a lookup table dependent on SZA, VZA, relative azimuth angle (RAA), surface albedo, cloud top pressure, and effective cloud fraction (Park and Kwon, 2020). We specify the shape factor with local NO₂ concentrations from the GEOS-Chem simulation described in Sect. 2.3 and extending to the mesosphere. In simulations of observations from the KORUS-AQ aircraft campaign over South Korea, Yang et al. (2023) showed that GEOS-Chem was successful in reproducing the NO₂ vertical profile observed below 5 km altitude and inferred from NO observations above. They found that the PBL extending to 2 km altitude accounted for over 95 % of the NO₂ tropospheric column and 80 %–91 % of the total NO₂ atmospheric column in the Seoul and Beijing metropolitan areas of interest here. The model correctly simulated the observed diurnal variation in the PBL NO₂ vertical profile over Seoul as driven by mixed layer growth. This resulted in a diurnal amplitude of 14 % for the AMF, peaking in the afternoon when mixing depth is maximum.

2.2 Pandora data

The Pandora instruments measure radiance at 280–525 nm (Herman et al., 2018) and fit total column NO₂ (including the stratosphere). There were two Pandora sites in Seoul and one in Beijing (40.0° N, 116.4° E) for the 2021–2022 period. The two Pandora sites in Seoul are at Seoul National University (Seoul – SNU; 37.5° N, 127.0° E) in the southern part of Seoul (Kim et al., 2021; Park et al., 2018) and at Yonsei University (Seoul – YSU; 37.6° N, 126.9° E) in the northern part of Seoul (Kim et al., 2017). The Beijing site is located on the north side of Beijing and a more detailed description is in Liu et al. (2024). We obtain the Pandora direct sun data from the Pandonia global network (http://pandonia-global-network.org; PGN, 2023). We exclude low-quality data (quality flag = 12) as recommended by PGN (2021).

2.3 GEOS-Chem model

We use GEOS-Chem CTM version 13.3.4 (https://doi.org/10.5281/zenodo.5764874, The International GEOS-Chem User Community, 2021) driven by assimilated meteorological data from the Goddard Earth Observing System – Forward Processing (GEOS-FP) with a horizontal resolution of 0.25° × 0.3125° (≈ 25 × 25 km²) over East Asia (24–52° N, 104–133° E) and 3-hourly boundary conditions from a global GEOS-Chem simulation with 4° × 5° resolution. GEOS-FP provides the finest spatial resolution available to drive GEOS-Chem. The model has 47 vertical levels including 14 vertical levels in the lower 2 km. Simulations were conducted for DJF 2021/22 and JJA 2022 with 6 months of initialization for each period.

Aside from emissions (see below), the simulation is the same as previously described by Yang et al. (2023) and features some modifications to the standard GEOS-Chem 13.3.4 to better reproduce KORUS-AQ aircraft observations over Korea in May–June 2016. These include aerosol nitrate photolysis, volatile chemical product (VCP) emissions and chemistry, and reduced HO₂ uptake by aerosol.

Simulations for 2022 require adjustment to NO_x emissions beyond the most recent emission inventories used in GEOS-Chem for China (MEIC for 2019; Zheng et al., 2021) and Korea (KORUSv5 for 2015; Woo et al., 2020). We apply for this purpose the surface NO₂ concentration trends for China from the Ministry of Ecology and Environment (MEE) network (https://quotsoft.net/air/; MEE, 2023) and for South Korea from the AirKorea network (https://www.airkorea.or.kr/web/last_amb_hour_data?pMENU_NO=123; KEC, 2022), focused mostly on urban sites. Mean 2022/2019 surface NO₂ concentration ratios in China are 0.91 in DJF and 0.83 in JJA, and mean 2022/2015 values in Korea are 0.70 in DJF and 0.51 JJA, which are applied to scale the anthropogenic NO_x emissions. We assume these scaling factors to be applicable to Beijing and Seoul.

Yang et al. (2023) found that the GEOS-Chem simulation during KORUS-AQ successfully reproduced important features of NO_x chemistry, notably the $\mathrm{NO}/{\mathrm{NO}}_{\mathrm{2}}$ ratio driven by photochemical cycling involving ozone and HO₂. Several other studies have evaluated the GEOS-Chem simulation of NO_x over East Asia. Park et al. (2021) found that GEOS-Chem successfully reproduced the NO_x vertical profiles observed during KORUS-AQ. Shah et al. (2020) found a good simulation of the seasonality of OMI NO₂ over China and its long-term trend. Liu et al. (2018) found that NO₂ diurnal variability at the MEE stations was well captured but the model was too low, as would be expected from the urban nature of the sites.

2.4 Diurnal variation in NO_x emissions

Figure 1 shows the diurnal cycle of NO_x emissions used by GEOS-Chem in Beijing and Seoul. MEIC for China provides monthly NO_x emissions separately for the transportation, residential, industrial, and power sectors, while KORUSv5 separates mobile, area, and point sources. Neither inventory specifies diurnal variations in emissions. In our work, we apply the diurnal pattern from Liu et al. (2019) for the power sector and Miao et al. (2020) for other sources in the MEIC inventory. For KORUSv5 we apply the diurnal pattern from Liu et al. (2019) for point sources, supported by results from Bae et al. (2021), and the industrial daily pattern from Miao et al. (2020) for area sources. We estimate the diurnal variation in mobile sources in KORUSv5 using hourly Seoul Transport Operation and Information Services (TOPIS, 2023) data on weekday total traffic and construction equipment activity.

Figure 1 Diurnal variation in NO_x emissions in Beijing and Seoul for DJF 2021/22 and JJA 2022. Local time is Chinese standard time (CST) for Beijing and Korean standard time (KST) for Seoul. Solar noon is at 12:08–12:27 CST in Beijing and 12:21–12:45 KST in Seoul. Values are for the white boxes in Fig. 2. Different colors represent different sectors, and the black line shows the total emission.

Figure 1 shows that emissions are dominated by industrial and transport sources in Beijing and by mobile (transport) sources in Seoul. Both sectors show a broad maximum between 07:00 and 18:00 LT that defines the overall diurnal cycle of emissions and is similar in winter and summer. There are no significant rush hour peaks in transport emissions, suggesting that the surface NO₂ maxima observed in early morning and evening are driven more by shallow mixing depths (Moutinho et al., 2020). Total NO_x emission in Beijing in winter is 30 % greater than in summer, driven by the industrial source and possibly due to workplace heating. There is less seasonal variation in Seoul where mobile sources are the largest emitters.

3 Intercomparison of total NO₂ columns

Figure 2 shows the total NO₂ columns over eastern China and South Korea observed by GEMS and simulated by GEOS-Chem during DJF 2021/22 and JJA 2022. The GEMS data are mapped on the 0.25° × 0.3125° GEOS-Chem grid. The yellow box delineates the Seoul Metropolitan Area (SMA). The zoomed-in black boxes are Beijing and Seoul, and the white boxes are the city centers where Pandora stations are located. The maximum NO₂ concentrations are in the city centers in summer but are shifted to the south in winter due to the prevailing winds and the long NO_x lifetime (Seo et al., 2021). We see from Fig. 2 that GEMS and GEOS-Chem have consistent spatial distributions and backgrounds, but GEOS-Chem over polluted regions is generally higher than GEMS except for Seoul in winter.

Figure 2 Total NO₂ columns over East Asia retrieved by GEMS and simulated by GEOS-Chem. The data are 3-month averages for December–July–February (DJF) 2021/22 and June–July–August (JJA) 2022 on the 0.25° × 0.3125° GEOS-Chem nested grid. The yellow rectangle delineates the Seoul Metropolitan Area (SMA; 36.6–38.1° N, 126.4–128.3° E). The zoomed-in plots show Beijing and Seoul, and the white boxes are the 0.25° × 0.3125° urban cores where the Pandora stations are located (black circles). Scales are different for DJF and JJA.

Figure 3 further intercompares GEOS-Chem and GEMS using the Pandora stations in Beijing and Seoul as evaluation metric. Previous GEMS evaluation with Pandora at the native pixel resolution of GEMS was presented by Kim et al. (2023). Here we conduct the evaluation on the coarser 0.25° × 0.3125° GEOS-Chem grid as most relevant for our work. GEMS and GEOS-Chem reproduce the diurnal and day-to-day variability observed by Pandora in DJF (R²= 0.87–0.90) and JJA (R²= 0.77–0.79). NO₂ column magnitudes also agree well with Pandora in winter, with linear regression slopes of 0.94 for GEMS and 0.90 for GEOS-Chem. Summer shows larger biases, reflecting differences between the SNU and YSU Pandora sites that cannot be resolved at the 0.25° × 0.3125° resolution of GEOS-Chem (there are few observations at the Beijing site in JJA). YSU is more polluted than SNU, which is in a mountainous area more remote from emissions. Overall, a comparison to Pandora supports the diurnal and day-to-day variability seen in the GEMS and GEOS-Chem data. The rest of our analysis focuses on the diurnal variability.

Figure 3 Intercomparison of GEOS-Chem, GEMS, and Pandora NO₂ columns for the Pandora sites in Beijing and Seoul. The figure shows scatterplots of daytime hourly data for DJF 2021/22 and JJA 2022. GEMS is mapped on the 0.25° × 0.3125° GEOS-Chem grid. Coefficients of determination (R²) and reduced major axis linear regressions are shown. The 1:1 line is dashed. The Beijing Pandora site has limited observations in JJA.

4 Diurnal variation in NO₂ columns on the urban scale

We start with an urban core analysis focusing on the white boxes shown in Fig. 2 for Beijing and Seoul, representing single 0.25° × 0.3125° GEOS-Chem grid cells where the Pandora stations are located. Scatterplot comparisons between GEOS-Chem, GEMS, and Pandora for these grid cells are shown in Fig. 3. Figure 4 shows the diurnal variation in the total NO₂ column observed from GEMS and Pandora and simulated by GEOS-Chem in Beijing. GEOS-Chem results are shown as averages for all days and for the subset of days when GEMS observations are available (generally limited by cloud cover). Wintertime NO₂ in all three datasets is flat from 10:00 to 11:00 LT and then increases from 11:00 to 14:00 LT. Summertime NO₂ decreases from 08:00 LT to a minimum at 13:00–14:00 LT and then increases to 16:00 LT, consistent between GEOS-Chem and GEMS. Pandora observations in the summertime are too limited to show.

Figure 4 Diurnal variation in total NO₂ column and driving processes in Beijing. The first row shows the average NO₂ columns observed by GEMS and Pandora and simulated by GEOS-Chem in DJF 2021/22 and JJA 2022 for the 0.25° × 0.3125° GEOS-Chem grid cell in the urban core where the Pandora station is located (white box in Fig. 2). GEMS observations are available for the hours indicated by symbols. GEOS-Chem results for the full diurnal cycle are shown as averages for all days and for the subset of days when GEMS data are available (generally limited by cloud cover). Pandora data are not shown for JJA due to a limited number of observations (Fig. 3). The second row shows the hourly tendencies in the GEOS-Chem NO_x budget (averaged for all days) for the planetary boundary layer (PBL) conservatively defined as extending up to 3 km altitude. The tendencies describe the contributions from individual processes to the NO_x budget as given by Eq. (3), with NO_x defined as ${\mathrm{NO}}_x\equiv \mathrm{NO}+{\mathrm{NO}}_{\mathrm{2}}+{\mathrm{NO}}_{\mathrm{3}}+\mathrm{2}{\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}+\mathrm{HONO}+{\mathrm{HNO}}_{\mathrm{4}}+{\mathrm{ClNO}}_{\mathrm{2}}$. The third row shows the PBL ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ column molar ratio in GEOS-Chem.

We used the GEOS-Chem budget tendency diagnostic to understand the drivers of the diurnal variation in NO₂ columns. This diagnostic tracks the mean mass-weighted changes in column concentrations after each model operation for any selected horizontal domain, vertical column, and time period. We focus on the PBL column conservatively defined as extending to 3 km altitude after verifying that altitudes higher than 3 km make negligible contributions to diurnal changes in the total model column. Within the PBL column we consider the budget of NO₂ as that of NO_x ($\equiv \mathrm{NO}+{\mathrm{NO}}_{\mathrm{2}}+{\mathrm{NO}}_{\mathrm{3}}+\mathrm{2}{\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}+\mathrm{HONO}+{\mathrm{HNO}}_{\mathrm{4}}+{\mathrm{ClNO}}_{\mathrm{2}}$) multiplied by the local ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ PBL column concentration ratio. This eliminates from the budget the fast interconversion reactions within the NO_x family and provides a more useful budget perspective. It allows us to consider NO_x emission as a source of NO₂ even though NO_x is emitted mainly as NO. The NO_x family is mainly contributed by NO and NO₂, and the main interconversion reactions defining the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio are

$$\begin{array}{}\text{(R1)}& {\displaystyle }\mathrm{NO}+{\mathrm{O}}_{\mathrm{3}}\to {\mathrm{NO}}_{\mathrm{2}}+{\mathrm{O}}_{\mathrm{2}},\text{(R2)}& {\displaystyle }\mathrm{NO}+{\mathrm{HO}}_{\mathrm{2}}\to {\mathrm{NO}}_{\mathrm{2}}+\mathrm{OH},\text{(R3)}& {\displaystyle }\mathrm{NO}+{\mathrm{RO}}_{\mathrm{2}}\to {\mathrm{NO}}_{\mathrm{2}}+\mathrm{RO},\text{(R4)}& {\displaystyle }\mathrm{NO}+\mathrm{XO}\to {\mathrm{NO}}_{\mathrm{2}}+\mathrm{X},\text{(R5)}& {\displaystyle }{\mathrm{NO}}_{\mathrm{2}}+hv\stackrel{+{\mathrm{O}}_{\mathrm{2}}}{⟶}\mathrm{NO}+{\mathrm{O}}_{\mathrm{3}},\end{array}$$

where RO₂ denotes organic peroxy radicals and X denotes halogen atoms. The net tendency for the PBL NO₂ column (Ω_NO2) can then be related to that of NO_x (${\mathrm{\Omega }}_{{\mathrm{NO}}_x}$) as

$$\begin{array}{}\text{(2)}& {\left[{\displaystyle \frac{\partial {\mathrm{\Omega }}_{{\mathrm{NO}}_{\mathrm{2}}}}{\partial t}}\right]}_{\mathrm{net}}=\mathit{\alpha }\left(t\right){\left[{\displaystyle \frac{\partial {\mathrm{\Omega }}_{{\mathrm{NO}}_x}}{\partial t}}\right]}_{\mathrm{net}},\end{array}$$

with

$$\begin{array}{}\text{(3)}& \begin{array}{rl}{\left[{\displaystyle \frac{\partial {\mathrm{\Omega }}_{{\mathrm{NO}}_x}}{\partial t}}\right]}_{\mathrm{net}}& ={\left[{\displaystyle \frac{\partial {\mathrm{\Omega }}_{{\mathrm{NO}}_x}}{\partial t}}\right]}_{\mathrm{emission}}+{\left[{\displaystyle \frac{\partial {\mathrm{\Omega }}_{{\mathrm{NO}}_x}}{\partial t}}\right]}_{\mathrm{chemistry}}\\ & +{\left[{\displaystyle \frac{\partial {\mathrm{\Omega }}_{{\mathrm{NO}}_x}}{\partial t}}\right]}_{\mathrm{transport}},\end{array}\end{array}$$

and where $\mathit{\alpha }\left(t\right)={\mathrm{\Omega }}_{{\mathrm{NO}}_{\mathrm{2}}}/{\mathrm{\Omega }}_{{\mathrm{NO}}_x}$ is the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ PBL column ratio. The terms on the right-hand side of Eq. (3) are updated by GEOS-Chem over its operator splitting time steps and are archived in the budget diagnostic as spatial and temporal averages. NO_x dry deposition is included in the emission operator, but its contribution is very small (Shah et al., 2020). α(t) is archived every hour for application in Eq. (2).

The second row of Fig. 4 shows the different NO_x budget terms from Eq. (3) over hourly time steps, with the net tendency as the left-hand-side term. The third row shows the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ PBL column molar ratios in GEOS-Chem. Each data point in the second and third rows (centered on the half hour) explains the change between the 2 successive hours shown in the first row. The GEOS-Chem diurnal variation in the NO₂ column in the first row reflects the net NO_x tendency combined with the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio. We see that the increase in the NO₂ column over the course of the day in winter reflects the dominant effect of daytime emissions, 4 times higher than at night and leading to NO_x accumulation. Chemical loss is slow in winter, and transport (flux divergence) is the main loss term. The flat trend of the NO₂ column from 10:00 to 11:00 LT corresponds to the diurnal minimum of the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio. This ratio increases over the rest of the day as the mixed layer deepens and the freshly emitted NO is exposed to higher ozone concentrations. The increase in the ratio contributes to the increase in the NO₂ column. The NO₂ column in GEOS-Chem thus peaks at 18:00 LT. During the night, the NO_x emission decreases, and the loss from transport leads to a decrease in the total NO₂ column. The ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio at night is only 0.55 mol mol⁻¹, despite no NO₂ photolysis, because of sustained NO emission and the slow rate of the NO+O₃ reaction (low ozone and low temperatures).

The opposite diurnal variation in NO₂ in summer reflects weaker daytime emission of NO_x and stronger chemical loss as shown by the GEOS-Chem budget analysis. Even though the emission term remains larger than the chemical loss term, there is also a negative transport term from ventilation. The chemical loss of NO_x peaks at 11:00–12:00 LT and then weakens, reflecting the noon maximum of OH concentrations (Logan et al., 1981) combined with the decreasing NO₂ concentration and explaining the slow recovery of the NO₂ column in the afternoon. The ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio is higher in summer than in winter and shows little variation during the daytime, reflecting the higher concentrations of O₃ and HO₂ radicals offsetting the effect of NO₂ photolysis. The daytime ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio averages 0.75 mol mol⁻¹ in summer, as compared to 0.50 mol mol⁻¹ in winter, contributing to the seasonality of NO₂ seen from space.

Figure 5 shows the same as Fig. 4 but for Seoul. The two Pandora stations show differences in NO₂ columns, particularly in summer, as previously shown in Fig. 3. They also show some differences in diurnal variation, particularly in winter, which we similarly attribute to local effects such as different emissions, wind speeds, and geography that cannot be resolved at 25 km resolution. The diurnal variations in GEMS and GEOS-Chem agree to within the ranges defined by data from the two Pandora stations. NO₂ columns in winter increase from 10:00 to 12:00 LT as in Beijing but then flatten in the afternoon, which we attribute in GEOS-Chem to stronger winds. NO₂ columns in summer show an increase from 08:00 to 10:00 LT, unlike in Beijing, because of larger emissions initially overwhelming the chemical loss term. There follows a decrease until 13:00–14:00 LT and a recovery in the later afternoon, similar to Beijing and driven by the same factors.

Figure 5 Same as Fig. 4 but for Seoul.

5 Separating the influences of emission, chemistry, and transport

We showed in Sect. 4 that the diurnal variation in the NO₂ column observed by GEMS on the urban scale reflects a balance between emission and transport in winter, as well as the added influence from chemical loss in summer. The transport term can be represented with a CTM in an inversion framework (Cooper et al., 2017), but simple quantification of the transport term on the urban scale can also be done from knowledge of the wind speed with a mass balance approach (Jacob et al., 2016).

Figure 6 illustrates the sensitivity of the NO₂ diurnal variation to wind speed in the wintertime GEMS observations over Seoul when chemical loss is a negligible term. A wind speed of 6 m s⁻¹ ventilates the 25 × 25 km² urban core on a timescale of 1 h. Here we segregate the data by GEOS-FP hourly wind speed at 850 hPa higher or lower than 6 m s⁻¹. The diurnal variations are very different at high and low wind speed and consistent between GEMS and GEOS-Chem. At high wind speed, the NO₂ column shows little diurnal variability because emission is balanced by transport. At low wind speed, NO₂ accumulates over the daytime hours because the transport term is weaker and does not keep up with emissions. Steady state between emissions and ventilation is finally reached at 16:00 LT, but the NO₂ column keeps increasing until 18:00 LT because of increasing ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ ratio. Edwards et al. (2024) found an anticorrelation between the wind speed and tropospheric NO₂ column concentrations consistent with our findings.

Figure 6 Same as Fig. 4 but for DJF 2021/22 in Seoul with data segregated by wind speed. Segregation threshold is 6 m s⁻¹ for the 850 hPa hourly wind speed in the NASA GEOS-FP meteorological data used as input to GEOS-Chem.

One can reduce the effect of transport by spatial averaging over a large domain, thereby increasing the ventilation timescale. Edwards et al. (2024) showed that regionally averaging the data over northeast Asia minimized the transport effect, though the interpretation of the result is more complicated due to averaging over diverse emission and chemical environments. Figure 7 shows the average diurnal pattern of the NO₂ column observed by GEMS and simulated by GEOS-Chem on the ≈ 150 km scale of the SMA (Fig. 2) in winter and summer. Again, the diurnal variations observed by GEMS and simulated by GEOS-Chem are consistent. NO₂ columns in winter increase over the course of the day in a more regular manner than on the 25 km urban scale (Fig. 5a) because the transport loss term is steadier and responds mainly to the change in the NO₂ column. The chemical loss term is not negligible, unlike on the urban scale, because its timescale of about 20 h is comparable to that of transport.

Figure 7 Same as for Fig. 4 but for the Seoul Metropolitan Area (SMA; 36.6–38.1° N, 126.4–128.3° E) corresponding to the yellow box in Fig. 2. Quantities are averages over all 0.25° × 0.3125° grid cells in the SMA.

We see from Fig. 7 that the transport term can be successfully marginalized on the scale of the SMA in summer because the chemical loss term is faster. The resulting SMA diurnal pattern of the total NO₂ column is consistent with that of Seoul (Fig. 5b) but with a flatter shape, and the early-afternoon minimum is now driven mainly by chemistry. The amplitude is greatly dampened because emissions are 5 times weaker when averaged over the SMA regional domain and because the chemical loss integrates over the residence time within the domain.

6 Conclusions

We used the GEOS-Chem model to interpret the diurnal variation in NO₂ columns observed from the GEMS geostationary instrument and Pandora ground-based spectrometers over Beijing and Seoul in December–January–February (DJF) 2021/22 and June–July–August (JJA) 2022. This was motivated by the need to understand the unique information offered by hourly geostationary satellite observations on the budget of NO_x through the contributions of emissions, chemistry, and transport to the diurnal cycle of NO₂.

The GEOS-Chem model used in this work had previously shown successful simulation of the NO₂ vertical profile and its diurnal variation over Seoul in the KORUS-AQ aircraft campaign, enabling reliable computation of the diurnal dependence of the air mass factor (AMF) that contributes to the diurnal variation in NO₂ observed from space. Here we used the diagnostic budget capability in GEOS-Chem to isolate the contributions of NO_x emissions, chemistry, transport, and the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ column ratio to the diurnal cycle of NO₂ columns. We also updated NO_x emissions to 2022 including their diurnal variations. The NO_x emissions for Beijing and Seoul are a factor of 4 higher in the daytime than at night, reflecting mobile and industrial sources, and show little variation during the daytime hours. We focused on the simulation of the total atmospheric NO₂ column rather than the tropospheric column, taking advantage of the stratospheric capability in GEOS-Chem, to avoid errors in the definition of the tropopause. Diurnal variation in the NO₂ atmospheric column in the two cities is mainly determined by the planetary boundary layer (PBL) up to 3 km altitude.

We investigated the diurnal variation in the NO₂ column at the 25 km urban scale over Beijing and Seoul. GEMS, Pandora, and GEOS-Chem show similar variability and diurnal variations. NO₂ columns in winter increase over the course of the daytime hours, reflecting accumulation from high daytime emissions offset by loss from horizontal transport (flux divergence), and are further enhanced by an increase in the ${\mathrm{NO}}_{\mathrm{2}}/{\mathrm{NO}}_x$ over the course of the day as ozone is entrained in the growing mixed layer. Chemical loss of NO_x in winter is too slow to play a significant role in the observed diurnal variation. In summer, by contrast, NO₂ columns decrease from 10:00 to 14:00 LT because of NO_x photochemical oxidation compounding the loss from transport.

We further examined the importance of transport for interpreting the diurnal variation in the GEMS urban NO₂ data by segregating the Seoul data by wind speed. In winter, the low-wind GEMS data (< 6 m s⁻¹) show a steady rise in NO₂ over the course of the day, while the high-wind data (≥ 6 m s⁻¹) show flat diurnal variation, consistent with the GEOS-Chem model. Transport plays an important role in the NO_x budget in both cases but cannot keep up with the high daytime emissions in the low-wind case.

We examined whether the role of transport in the diurnal variation in the urban NO₂ column could be reduced by spatial averaging of the data over the 150 km regional scale of the Seoul Metropolitan Area (SMA). The SMA data in winter show a steady increase over the daytime hours due to emissions, but the transport term remains the major sink of NO_x. The SMA data in summer show negligible loss from transport in daytime because the chemical loss term is much faster, but the diurnal amplitude is weak because of diluted emissions and long residence times for the air over the regional domain.

Our conclusions regarding the interpretation of the diurnal variation in NO₂ columns observed by GEMS can be extended to other instruments of the geostationary air quality constellation, such as TEMPO over North America, launched in April 2023, and Sentinel-4 over Europe, scheduled for launch in 2024. This work further lays the groundwork for the use of GEOS-Chem in inversions of the geostationary satellite data to infer NO_x emissions.