2.1 Site description

Beijing is a megacity with a population of over 21.7 million in 2023 (https://worldpopulationreview.com/world-cities/beijing-population, last access: 26 October 2023). The vast majority of the people in Beijing live within the 5th Ring Road (see Fig. 1). Beijing is located in the north of the North China Plain (NCP), with the Yan and Taihang mountains to the north and west, respectively. The NCP includes a highly industrialized and urbanized area where the weather conditions are often conducive to the accumulation of air pollution, resulting in the frequent occurrence of haze episodes (Sundström et al., 2012; Sun et al., 2014; Li et al., 2016). The effects of pollution in the NCP on the air quality in Beijing varies with weather conditions. Southern airflow transports pollution to the city, whereas northern airflow transports clean air from the north. Hence the air quality situation may vary strongly, depending on large-scale meteorological patterns (Hou et al., 2020). Furthermore, as the capital city of China, Beijing hosts the Chinese government and many national and international events. During such events, emissions from industry and traffic in and around Beijing are regulated to improve air quality (Wang et al., 2010; Wang et al., 2016; Fan et al., 2021; Chu et al., 2022).

Figure 1Digital elevation map of Beijing and surroundings. The solid black line is the provincial boundary and the solid gray line is the administrative boundary of Beijing on a district basis. The Beijing-RADI site is indicated by the black triangle, at about 12 km north of the administrative center of China (red pentagram). The enlarged panel shows Beijing with ring roads 2–6 (from the center to outside; there is no ring road 1) overlaid on the TROPOMI-derived NO₂ tropospheric VCD re-gridded to a resolution of 100 m × 100 m and averaged over the period from 1 August 2021 to 31 July 2022.

The AIRCAS laboratory building is located at the north side of Beijing between the 4th Ring Road and 5th Ring Road (see Fig. 1). The Beijing-RADI site is located at the roof of this building, at a height of 21 m above the ground. The site includes instrumentation for atmospheric observations such as a sun photometer, a lidar, and a suite of instruments for in situ measurements of aerosol properties, trace gas concentrations, and solar radiation. The Pandora instrument was added in 2021 to measure NO₂ VCDs and surface concentrations. Due to its location in a science park at the edge of the Olympic Park, the Beijing-RADI site is considered a combined urban and suburban background observatory with small local emissions and large influences from nearby highways, urban activities, and industrial pollution advected from the south of Beijing and the NCP.

2.2 Instrumentation and auxiliary data

2.3 Methodology

For the evaluation of the NO₂ observations, different methods are used, such as time series to show the variability on different timescales or the effects of external parameters such as wind speed, averaging to reduce short-time variability, and scatterplots for comparison with independent data sets. For the comparison between Pandora and TROPOMI NO₂ VCDs, the data need to be collocated. Furthermore, the difference in observation geometry needs to be taken into account. The Pandora bottom-up observations of direct and scattered (sky) solar radiation depend on the NO₂ vertical profile, the spatial distribution, and the SZA (Boersma et al., 2011; van Geffen et al., 2015). For example, for a situation when most of the total VCD is located below 2 km and the SZA is 45^∘, the total viewing area of Pandora with a FOV of ($\mathrm{2.6}\mathit{\pi }/\mathrm{180})\cdot (\mathrm{2}/\cos \left(\mathrm{SZA}\right)$) is about 128 m × 128 m. The TROPOMI top-down observations from space provide NO₂ VCDs with a spatial resolution of 3.5 km × 5.5 km. To match the observation volumes, the TROPOMI observations were resampled to a spatial resolution of 100 m × 100 m, using Google Earth Engine (GEE; https://developers.google.com/earth-engine/guides/scale, last access: 11 July 2023). For an accurate comparison between TROPOMI and Pandora, only high-quality data were retained for the collocated TROPOMI–Pandora pixels. For time collocation, Pandora data obtained within 10 min of (both before and after) the TROPOMI overpass time were averaged. The resulting collocated TROPOMI and Pandora data sets were quantitatively evaluated using scatterplots and the statistical metrics mean difference (MD), mean absolute difference (MAD), mean relative difference (MRD), standard deviation (σ), correlation coefficient, and fitted slope.

MD is calculated by averaging the difference between two observations, as shown in Eq. (1):

MAD, given by Eq. (2), is defined as the mean of the absolute differences between TROPOMI and Pandora:

MRD, Eq. (3), is the mean of the differences between TROPOMI and Pandora when normalized with Pandora's VCDs:

Positive and negative values of the MRD indicate the degree of overestimation or underestimation.

The Pearson correlation R (Pearson, 1895) is defined in Eq. (4).

The standard deviation σ is defined in Eq. (5):

The dilution factor, D_f, is defined in Eq. (6), where VCD_TROPOMI,0.1 (median) and VCD_TROPOMI,R (median) represent the median value of the TROPOMI VCD within a radius of 0.1 km and within a radius ρ around the location of Pandora, respectively (Chen et al., 2009; Griffin et al., 2019; Pinardi et al., 2020):

3.1 Pandora data overview

An overview of the Pandora data is presented in Fig. S1 in the Supplement, as time series for each month from August 2021 to July 2022, where the timescale is local time (i.e., China standard time, CST, locally called Beijing time, BJT, i.e., UTC + 8) throughout this paper. As an example, the data for November and December are enlarged in Fig. 2. Total NO₂ VCD, tropospheric NO₂ VCD, and surface NO₂ concentrations are plotted in different colors, and only high- and medium-quality data are included (DQ0, DQ1, DQ10, and DQ11). Time series including also low-quality data (DQ2 and DQ12) are presented in Fig. S2. Although the use of low-quality data is not recommended, we have added this figure in the Supplement to show the high data density throughout the year. Overall, the NO₂ concentrations are low in the summer and high in the winter, as is commonly observed in satellite data (van der A et al., 2006; Wang et al., 2019; Fan et al., 2021). For the tropospheric VCDs, including low-quality data adds 4681 data points to the total of 8620 data points during the 1-year study period (i.e., 54.3 % of all data are low quality). Among the total VCDs, 21 767 data points out of a total of 80 153 (27.2 %) are low quality. A comparison of Figs. S1 and S2 shows that low-quality data occur in particular during the winter months when the NO₂ concentrations are high and aerosol (PM_2.5) concentrations are also high, the boundary layer is shallow, the sky is often cloudy, and air quality is bad. These are conditions in which the data processing flags bad situations for some criteria, the reliability of the results decreases, and hence the DQ is increased to DQ2.

Figure 2Time series of NO₂ data measured with Pandora at the Beijing-RADI site for November and December 2022 (the horizontal axes give the day of the month): total VCD in green and tropospheric VCD in blue (both on left vertical axes in $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$) and surface concentrations in red (plotted on the right axis in µg cm⁻³) Only data with DQ0, DQ1, DQ01, and DQ11 are plotted. Note that the surface concentration scale has been chosen such that data are plotted at the bottom of the VCDs.

Download

3.1.2 Monthly mean NO₂ VCDs

As shown in Fig. 2, apart from diurnal and short-term variations, there is also a clear seasonal variation in the NO₂ VCDs. This is further illustrated in Fig. 3 where time series are plotted for the mean and median of monthly mean tropospheric VCDs (calculated from daily averages) derived from Pandora measurements during the period from August 2021 to July 2022. Only the data measured between 08:00 and 17:00 CST, corresponding to daylight hours during the winter, have been included, with DQ0, DQ1, DQ10, and DQ11. This selection of high-quality data reduces the number of days for which data are available but ensures reliability. A similar figure also including DQ2 and DQ12 data is included as Fig. S4, which shows that in the autumn and winter the NO₂ VCDs are high and that they are low in the spring and summer, in good agreement with the seasonal variation which is commonly observed in satellite data (van der A et al., 2006; Wang et al., 2019; Fan et al., 2021). The number of days for which data are available for each month is included in Table S1 and Fig. S2 (including DQ2 and DQ12). Comparison shows that mainly the monthly mean VCD for December 2021 is affected by the data selection: the number of days is reduced from 31 to 9, and apparently mainly data with high VCDs, i.e., during days with bad AQ (Sect. 3.1.1), were removed by the DQ2 criteria. As a result, the annual cycle of the mean tropospheric NO₂ VCDs is not as obvious in the high-quality data as in Fig. S4, which clearly shows the maximum occurring in the winter and the minimum in the summer. Such an annual variation pattern is in good agreement with near-surface observations of in situ NO₂ measurements in Beijing. The mean, median, minimum, and maximum values, as well as the standard deviation (σ), of the tropospheric NO₂ VCDs over Beijing for each month are summarized in Table 1. The difference between the median and mean values is significantly larger in the winter months than in the summer. The variability in the NO₂ VCDs over Beijing is much higher during the winter than in the summer. The minimum in February 2022, as compared to January and March, illustrates the effectiveness of the emission control policy in and around Beijing during the 2022 Winter Olympics (Chu et al., 2022). Time series of the monthly mean total VCDs are presented in Fig. S5 and show similar variation to that for the tropospheric VCDs in Fig. 3.

Figure 3Time series of the monthly mean and median tropospheric NO₂ VCDs over Beijing from August 2021 to July 2022. Only high-quality data (DQ0, DQ1, DQ10, and DQ11) are included, which limits the number of days for which data are available as indicated in Table 1. The blue-shaded area indicates the standard deviation for the monthly mean data. Monthly mean data were calculated from daily means.

Download

Table 1Statistics of the Pandora-derived monthly mean NO₂ tropospheric VCDs. Only data with DQ0, DQ1, DQ10, and DQ11 are included.

Download Print Version | Download XLSX

3.2 Impact of wind direction and wind speed on NO₂ VCDs in Beijing

The wind influences the concentrations of atmospheric constituents, and thus air quality, in several different ways. The wind transports aerosols and trace gases away from their sources and thus disperses them, leading to lower concentrations. This includes both vertical mixing due to wind-generated turbulence and advection (horizontal transport). This also implies that at very low wind speed the constituents accumulate, which leads to enhanced concentrations (Feng et al., 2014) and, in particular due to (photochemical) reactions, the formation of haze (An et al., 2019). The wind direction and its history indicate the transport pathways determined by large-scale weather phenomena (You et al., 2018; He et al., 2018; Hou et al., 2020; Li et al., 2021; W. Zhao et al., 2022). When the transport pathways cross source regions, the constituents are transported to the receptor point. Hence, depending on the wind direction, the wind transports polluted or clean air over long distances (Sundström et al., 2012).

To address the effect of advection on the NO₂ tropospheric VCDs at the Beijing RADI site, a polar map has been created of NO₂ tropospheric VCDs versus wind speed and wind direction (Fig. 4). To separate effects of wind direction and wind speed on NO₂ VCDs from the large diurnal variations (Fig. 2), data have been plotted at 13:00 CST only when the concentrations are close to their daily minimum. The data in Fig. 4 show that, overall, the tropospheric VCDs are smaller during northwesterly winds, in particular when wind speeds exceed about 4 m s⁻¹. This confirms that wind from these directions transport cleaner air to Beijing. As discussed in Sect. 2.1, the Yan and Taihang mountains are situated to the north and west of Beijing and the area between Beijing and the mountains is mainly agricultural with sparse population and little industry. Thus anthropogenic and industrial emissions over this area are small. Hence, during low wind speeds from this direction, relatively clean air is transported to Beijing. When wind speeds are higher, northwesterly winds from the Siberian plains bring clean air, thus greatly improving Beijing's air quality. As a result, the NO₂ concentrations in the northwesterly wind sector are generally low, as shown in Fig. 4. However, the data in Fig. 4 show some exceptions when NO₂ concentrations are high. Further analysis shows that these observations were all made during the winter and are likely due to NO_x emissions from natural gas companies located in the Changping District in the northwest of Beijing. Natural gas is provided for, e.g., heating in the winter, and NO_x is produced during the combustion process (Pan et al., 2023). Thus, in the winter, during northwesterly winds, NO₂ is transported to the Pandora site. This explains the observations of high NO₂ concentrations due to local emissions. (For more details of the high concentration number and time during northwest wind, please see Table S2.)

Figure 4Polar diagram of NO₂ tropospheric VCDs, wind direction, and wind speed during the TROPOMI overpass time. Wind speed is indicated by the radius of the circles, ranging from 0 to 12 m s⁻¹; wind direction corresponds to the angle in polar coordinates, clockwise from north. Note that the south wind direction indicates that the wind comes from the south.

Download

During southerly winds, the NO₂ VCDs are substantially higher. The high industrial activity and associated traffic and high degree of urbanization in the south of Beijing and the NCP lead to high emissions and, thus, during southerly winds, transport of large amounts of NO₂ to Beijing. In addition, during low wind speed, the pollutants are accumulated, resulting in the observed very high pollution levels at low wind speed, in particular during southerly winds.

3.3 Tropospheric contribution to total NO₂ VCDs

The total NO₂ VCDs are determined from the Pandora direct-sun measurements, and the tropospheric NO₂ VCDs are determined from the Pandora sky measurements. Hence the total and tropospheric NO₂ VCDs are independently determined and can be used to obtain information on atmospheric processes. The data in Fig. 2 show the strong diurnal variation in both the total and the tropospheric NO₂ columns, but the time series of the ratio of the tropospheric to the total NO₂ VCD (see Fig. 5, where the ratios for February are shown as an example, and Fig. S6 for all months) show that they do not vary in the same way. Because direct-sun measurements are made more frequently than sky measurements, the ratios have been calculated using the first tropospheric VCD sky measurement available after a direct-sun measurement and then averaged to obtain daily and monthly means (Fig. 6). The results show that the fraction of tropospheric NO₂ to total NO₂ VCDs varies between about 0.2 and 1.05 (tropospheric VCDs should be smaller than or equal to the total VCDs; a value larger than 1 indicates the uncertainties in the data). The large diurnal and day-to-day variations in these ratios indicate that different processes in the troposphere and above affect NO₂ concentrations. The data show that the tropospheric $/$ total ratio decreases to a minimum around the middle of the day; i.e., early in the day the tropospheric NO₂ VCDs decrease faster than the total before they increase in the afternoon. This is due to a combination of processes including sources and sinks of a (photo)chemical nature (Herman et al., 2009), transport influenced by meteorological phenomena such as variations in wind speed and wind direction (discussed above in Sect. 3.2), and variations in boundary layer height while also the temperature profile changes throughout the day, influencing reaction rates and chemical balance (see Kang et al., 2022, for a brief overview). Likely all of these are different between the troposphere and above and therefore influence the ratio of tropospheric $/$ total NO₂ VCD and its daily evolution.

Figure 5Time series of the ratios of tropospheric VCDs to total VCDs during February 2022. The horizontal axis gives the day of the month.

Download

Figure 6Box-and-whisker plot of the monthly mean ratios of NO₂ tropospheric VCDs to total VCDs from August 2021 to July 2022, with the number of valid data in each month plotted as a blue dot (right-hand axis). For each data point, the dot represents the mean value, the horizontal line represents the median value, the top and bottom edges of the boxes are the 25 % and 75 % quartiles, and the whiskers are the values corresponding to 10 % and 90 % of the data volume. The date format of the horizontal axis is year-month.

Download

The monthly mean ratios of the NO₂ tropospheric to total VCDs derived from the Pandora measurements at the Beijing-RADI site are plotted as time series in Fig. 6, together with the number of days for which data are available in each month. The results in Fig. 6 show that the monthly mean ratios vary between about 0.38 and about 0.75, and there is some inter-monthly variation. Considering only months with 10 or more days for which data are available, the monthly averages show that, on average, the tropospheric NO₂ contribution is 50 % to 60 % in the winter and 60 % to 70 % in the spring and autumn. The smaller ratio in the winter may be related to the frequent occurrence of haze days when tropospheric NO₂ is converted to fine particulate matter (Zheng et al., 2015; Xie et al., 2015; Wang et al., 2020), whereas the larger ratio in spring can be derived from reduced stratospheric concentrations due to enhanced solar shortwave radiation (Cheng et al., 2016; Müller, 2021). Similarly to ozone, stratospheric intrusion could be a possible reason for the springtime increase in tropospheric NO₂ concentrations (Lin et al., 2015) because the higher values in the stratosphere have been observed in many studies (Sioris et al., 2003; Hendrick et al., 2004; Preston et al., 1998). Also, the larger standard deviations in spring (especially in March when it was larger than 0.2) indicate a larger day-to-day variability than in other seasons, which may be due to more active photochemical reactions in response to enhanced radiation intensity.

3.4 Comparison of Pandora-derived surface NO₂ concentrations with ground-based in situ measurements

During the study period, the Thermo Scientific 42i-TL analyzer was operated side by side with the Pandora instrument during a period of 20 d in January 2022. The collocation of these two instruments provides an excellent opportunity for the evaluation of the ground-based NO₂ concentrations derived from Pandora sky measurements, using the in situ data from the Thermo Scientific 42i-TL analyzer as reference. Figure 7 shows the time series of these two data sets. The in situ observations were made continuously, 24 h per day, but the Pandora measurements were only made during daylight hours. In addition, there are many gaps in the Pandora time series because only the best-quality data (DQ = 0, 1, 10, 11) were retained. To fill part of these gaps, Pandora-derived total VCDs, for which many more high-quality data are available as shown in Fig. S2 and discussed in Sect. 3.1.1, have been included in Fig. 7.

Figure 7Time series of NO₂ surface concentrations measured using the Thermo Scientific 42i-TL analyzer (gray) and derived from the Pandora sky data (blue) and total VCDs (red) during 20 d: 10–30 January 2022. Dashed vertical yellow lines indicate 00:00 CST. Note that the NO₂ total VCDs are plotted on the secondary vertical axis and scaled to match the Thermo 42i-TL analyzer surface concentration data. The date format of the horizontal axis is year-month-day.

Download

The continuous time series of the ground-based in situ measurements shows that the NO₂ concentrations are usually high during the night, decrease after sunrise to a minimum in the afternoon, and increase thereafter. The observed diurnal variation in the NO₂ concentrations is due to complex photochemical reactions between atmospheric gases such as nitrogen oxides (NO_x), O₃, and volatile organic compounds (VOCs) (Law et al., 2002; Xue et al., 2016). On days with large variations in the NO₂ surface concentration, such as on 11, 14, 16, and 17 January 2022 (Fig. 7), the total VCDs measured with Pandora traced the surface concentrations well (note that in Fig. 7 the total NO₂ VCD data were scaled to match the Thermo 42i-TL analyzer surface concentrations). In addition, the data show that on 11 and 16 January, the total VCDs increased earlier than the surface concentrations, suggesting that the increase in the NO₂ VCDs on these two dates resulted from atmospheric transport. On 14 January, the surface concentrations changed before the total VCDs increased, suggesting the occurrence of an NO₂ source near the surface from where NO₂ was vertically mixed, resulting in the subsequent increase in the tropospheric VCDs.

The data in Fig. 7 further show that the Pandora-derived surface NO₂ concentrations are overall relatively low as compared with the in situ concentrations measured with the Thermo Scientific 42i-TL analyzer. This is further illustrated in the scatterplot of the Pandora-derived NO₂ surface concentrations versus the Thermo Scientific 42i-TL analyzer in situ data (Fig. 8). Figure 8 shows that there are two data regimes: for low in situ NO₂ concentrations, up to about 22 µg cm⁻³, the Pandora-derived surface concentrations trace the in situ concentrations but vary over a much smaller range (between about 5 and 10 µg cm⁻³). For larger NO₂ concentrations, the Pandora-derived surface concentrations are closer to the in situ data but with a substantial underestimation. Similar differences have been reported between NO₂ concentrations retrieved from MAX-DOAS measurements, which use a similar technique to that used for Pandora to derive surface concentrations, and NO₂ point measurements using instruments mounted on a mast at three heights above the surface (a.s.l.), which, like the Thermo Scientific 42i-TL analyzer, derive the NO₂ concentrations from chemiluminescence measurements (Kang et al., 2021). These authors report that MAX-DOAS substantially underestimates the in situ concentrations, by about 65 % at 60 and 160 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$ and by 33 % at 280 $\mathrm{m}\mathrm{a}.\mathrm{s}.\mathrm{l}.$, but do not offer an explanation. Below we discuss the Pandora and chemiluminescence data and influencing factors which may contribute to the observed differences.

Figure 8Scatterplot of NO₂ surface concentrations: Pandora data versus Thermo Scientific 42i-TL analyzer data.

Download

The Pandora surface concentration is derived from the ratio of the difference between the SCDs of NO₂ at the largest and smallest SZA and the same difference for air, i.e., (SCD_Pandora,max − SCD_Pandora,0)$/$(SCD_AIR,max − SCD_AIR,0) (Cede, 2021). In view of the exponential decrease in NO₂ concentrations with height as reported by Kang et al. (2021), the slant optical path at Pandora's maximum observation angle, which is slightly less than 90^∘, implies that the NO₂ concentration is integrated over a certain height with a negative NO₂ concentration gradient. As a result, the measured concentration is lower than that at the surface which is measured by the Thermo Scientific 42i-TL analyzer.

Furthermore, Pandora measures the average NO₂ concentration along the (slant) optical path, whereas the Thermo Scientific 42i-TL analyzer provides a local value at the location of the instrument (Kang et al., 2021) which may be more sensitive to local variations, such as short-term emissions by passing cars or turbulent perturbations of plumes.

Another factor influencing the comparison of Pandora and in situ measurements of NO₂ concentrations using a Thermo Scientific 42i-TL analyzer is that this instrument may overestimate atmospheric NO₂ concentrations due to interference from other atmospheric constituents which react with ozone to produce chemiluminescence, such as peroxyacetyl nitrate and nitric acid (Dunlea et al., 2007; Steinbacher et al., 2007) or alkenes (Alam et al., 2020). In particular, during the Beijing winter the nitrate concentrations are high (Luo et al., 2019; W. Wang et al., 2021) and may contribute to overestimation of the NO₂ concentrations measured by the thermo-chemiluminescence method.

3.5 Validation of TROPOMI NO₂ VCDs over Beijing using Pandora

The Pandora instrument and the PGN were designed for satellite validation, i.e., providing VCDs as independent reference for the validation and evaluation of satellite data. Here Pandora data are used for the validation of TROPOMI NO₂ total and tropospheric VCDs, resampled to a spatial resolution of 100 m × 100 m. Scatterplots of NO₂ total and tropospheric VCDs from TROPOMI versus Pandora VCDs and histograms of the TROPOMI–Pandora differences are presented in Fig. 9. For TROPOMI, OFFL data were used, and for Pandora, only high-quality data with DQ0, DQ1, DQ10, and DQ11 were used. The Pandora quality control limits the number of suitable data, especially in June, July, August, and September (Sect. 3.1). The scatterplots in Fig. 9a and c show the good correlation of both the TROPOMI-derived NO₂ total and tropospheric VCDs with the Pandora data. For the total VCDs, R=0.95, the slope of the least-squares (LSQ) fit is 0.96, and the MRD is −8.44 %, well below the expected deviation of 0.5 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$ + (0.2 to 0.5) ⋅ VCD_trop (van Geffen et al., 2022b). Figure 9a shows that for small values of the total VCDs, the data points are very close to the identity line, whereas for VCDs > 10 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$, the differences are larger and the data points are rather evenly scattered around the identity line. The histogram of the differences between the TROPOMI and Pandora NO₂ total VCDs in Fig. 9b shows a uniform distribution with the characteristics of a Gaussian distribution with the mean or median as the symmetry axis. On average, the mean TROPOMI total VCDs are slightly smaller than those from Pandora, with an MD of −1.16 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$, a MAD of 3.14 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$, and a standard deviation of 4.44 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$. This uncertainty is in part due to the large SZA, which leads to uncertainty in the air mass factor (AMF) (Herman et al., 2019; Griffin et al., 2019; Ialongo et al., 2020). In addition, a 10 % error in the AMF estimate due to cloud effects was reported in the ATBD (van Geffen et al., 2022b).

Figure 9Validation of TROPOMI total and tropospheric NO₂ VCDs, resampled to a spatial resolution of 100 m × 100 m, using Pandora observations as reference data: (a, c) scatterplots of total and tropospheric TROPOMI vs. Pandora data together with statistical metrics; (b, d) histograms of the differences between TROPOMI and Pandora NO₂ total $/$ tropospheric VCDs.

Download

The validation of the TROPOMI-derived NO₂ tropospheric VCDs is presented as scatterplots and histograms in Fig. 9c and d. As with the total VCDs, the TROPOMI tropospheric VCDs trace the Pandora data well and the scatterplot shows the good correlation between both data sets, with a correlation coefficient of 0.92, slightly smaller than for the total VCDs, and a somewhat smaller slope of 0.94, with an RD of 16.15 %. The histogram in Fig. 9d shows that the tropospheric VCDs are roughly symmetrically distributed around 0, and the number of large negative deviations (deviation < −10 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$) is slightly larger than the number of positive deviations (deviation > 10 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$).

The above analysis confirms the good performance of the TROPOMI NO₂ total and tropospheric VCDs and provides confidence for the use of the TROPOMI data. For this analysis, the TROPOMI data were resampled to a spatial resolution of 100 m × 100 m, similar to the Pandora observation area.

3.6 Spatial representativeness of Pandora vertical column density

Considering the good performance of the TROPOMI-retrieved NO₂ VCDs discussed in Sect. 3.5, these data have been used to evaluate the spatial representativeness of the Pandora observations at the Beijing-RADI site. Figure 10a shows the spatial distribution of the TROPOMI-derived NO₂ tropospheric VCD over a wide area centered at the Beijing-RADI site averaged over the first year of Pandora operations (Fig. 1). Large differences are observed with high tropospheric NO₂ VCDs over Beijing, Tianjin, and its highly industrialized surroundings and several other locations. Over the mountains to the north and west of the study area, the tropospheric NO₂ VCDs are much lower, by a factor of 10 or more. The Beijing-RADI Pandora site is located in the area of high NO₂ tropospheric VCD (as also shown in the enlarged panel of Fig. 1), while strong NO₂ gradients are observed toward the north and west. Obviously, because of the variation in the NO₂ VCD, the correlation between the TROPOMI and Pandora data is influenced by the choice of the area over which the TROPOMI data are averaged. In particular, this may happen when a separation between clean and polluted air occurs over Beijing, as reported, for instance, by Sun et al. (2017) (their Fig. 7). The effect of the size of the area over which TROPOMI data are averaged is illustrated in Fig. 10b, showing the annual mean (black) and median (red) tropospheric NO₂ VCDs averaged over areas centered at the Beijing RADI Pandora site with increasing radius. The data show that, for this specific case when the TROPOMI NO₂ VCDs are averaged over annual mean values, the area-averaged value remains constant within a radius of 10 km around the Pandora site and decreases when the radius is further increased. This is further illustrated by the data in Table 2, showing that for a radius larger than 10 km, the standard deviation (σ) increases as the radius increases. For less homogeneous NO₂ distributions, the area for which the situation at the site is representative will be much smaller than 10 km. This also implies that the location of the Pandora site influences the validation results. The data in Table 2 show that for a circle of 3 km, the σ is 0.223 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$ and for a circle of 5 km it is 0.282 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$. This means that when the Pandora site is at a different location in an overpass pixel, the 3.5 km × 5.5 km FOV produces an error in the NO₂ tropospheric VCD between 0.223 and 0.282 $\mathrm{Pmolec}.{\mathrm{cm}}^{-\mathrm{2}}$, i.e., between 1.7 % and 2 %.

Figure 10Illustration of the spatial variation in the NO₂ tropospheric VCD over a wide area around the Beijing-RADI (Pandora-171) site and the evaluation of the representativeness of the Pandora data for a larger area. (a) Spatial distribution of the TROPOMI-derived NO₂ tropospheric VCD re-gridded to a resolution of 0.1 km × 0.1 km over the study area indicated in Fig. 1, averaged over the period from 1 August 2021 to 31 July 2022. The three purple circles indicate a radius of 10, 30, and 50 km around the site. (b) Annual mean (black) and median (red) tropospheric NO₂ VCD averaged over circles centered at the Beijing RADI Pandora site with increasing radius, as a function of that radius (expressed in km), with variability (1 standard deviation) given as a light-blue area around the mean.

Table 2Statistical metrics for the tropospheric VCDs averaged over circles with increasing radius around the Pandora site.

Download Print Version | Download XLSX