The monthly means and standard deviation of XCO2 and XCO at mid-latitude stations in the northern hemisphere are displayed in Fig. 4. These include FTIR stations in Beijing, Pasadena (34.14°N) (Wennberg et al., 2017), Xianghe (39.75°N) (Yang et al., 2020b), Karlsruhe (49.10°N) (Hase et al., 2017), Tsukuba (36.05°N) (Morino et al., 2018), and Paris (48.97°N) (Té et al., 2017). Table 1 summarizes the seasonal mean and standard deviation of XCO2 and XCO. There is no data recorded in February, which may cause a bias because anthropogenic XCO2 enhancements during the Spring Festival should be significantly lower than normal. The XCO2 values in Beijing ranged from 402 ppm to 423 ppm in 2019. The seasonal variation in XCO2 achieves its peak in winter (414.33 ± 2.65 ppm), followed by spring (413.58 ± 1.25 ppm) and autumn (412.09 ± 2.88 ppm), and is lowest in summer (407.87 ± 3.34 ppm). The features of the XCO2 seasonal variation in Beijing are similar to observations from other FTIR observation sites in the mid-latitude northern hemisphere. Intensity of photosynthesis, which is related to the latitude of the observation site, is the main reason for the seasonality in XCO2 variation. The peak and trough of monthly XCO2 means were found in December (415.72 ± 3.18 ppm) and August (404.87 ± 1.47 ppm), respectively, which differ slightly from other FTIR stations. The monthly XCO2 in Paris, Park Falls, Karlsruhe, and Xianghe stations reach a peak in March to April, whereas Pasadena and Beijing stations reach a peak in December. The monthly XCO2 of Beijing, Xianghe, Paris, Park Falls, Karlsruhe reach a trough in August, whereas Pasadena reaches a trough in September.
Figure 4. Monthly variations in (a) XCO2 and (b) XCO observed at Beijing, Xianghe, Karlsruhe, Pasadena, Tsukuba, and Paris during 2019. The geographical coordinates of each site are shown in the figure legend. The error bars are the monthly standard deviation of XCO2 and XCO.
Season XCO2 (ppm) XCO (ppb) Spring
(March to May)
413.58 ± 1.25 109.58 ± 5.49 Summer
(June to August)
407.87 ± 3.34 120.50 ± 4.28 Autumn
(September to November )
412.09 ± 2.88 112.37 ± 5.58 Winter
(December to February)
414.33 ± 2.65 115.78 ± 3.36
Table 1. Seasonal variability of average XCO2 and XCO in 2019 observed by EM27.
Values of XCO ranging from 85 to 192 ppb are observed in Beijing. The XCO in Beijing was higher than the values from other cities in the northern hemisphere and lower than that of the Xianghe station (Fig. 4b), implying the presence of high regional emissions in the Beijing–Xianghe region. As can be concluded from Table 1, XCO is highest in summer (120.50 ± 4.28 ppb), followed by winter (115.78 ± 3.36 ppb) and autumn (112.37 ± 5.58 ppb), and is lowest in spring (109.58 ± 5.49 ppb). The seasonal variation in Beijing is similar to the variation in Xianghe. The increase of CO during winter in Beijing could be from the increased heating time of vehicle catalysts at low temperatures (Han et al., 2009) and domestic heating. The seasonal variations in Karlsruhe, Pasadena, and Paris show XCO values that are highest in spring and lowest in summer and autumn. The XCO in summer for the Beijing-Xianghe region presents a seasonal variation opposite of the other three urban sites. Carbon monoxide (CO) reaches a minimum during summer at those sites due to strong ultraviolet radiation and high humidity, facilitating the formation of OH, which consumes more CO in the atmosphere (Té et al., 2016; Li et al., 2020). Carbon monoxide (CO) concentration was also found to be lowest in summer from 2006 to 2018 at the Shangdianzi (SDZ) in situ site, a rural region near Beijing (Li et al., 2020). This result is consistent with the Karlsruhe, Pasadena, and Paris sites. The high XCO values in summer observed in urban Beijing and Xianghe could be associated with strong anthropogenic emissions.
Carbon monoxide (CO) is co-emitted and co-transported to the observation site with CO2. The slopes of ΔXCO to ΔXCO2 reflect the overall combustion efficiency of the observed airmass. Figure 5 shows the daily regression slope and the Pearson correlation coefficients (R2) of ΔXCO and ΔXCO2 based on the diurnal variation method. Daily regression slopes are mostly around 10 ppb ppm–1. Daily correlation coefficients are generally larger than 0.5 on 78% of the observation days. Significant positive correlations between ΔXCO to ΔXCO2 in winter, spring, and autumn indicate that most air parcels originate from combustion sources. Approximately 61% of the daily correlation coefficients in summer (grey shaded region in Fig. 5) are small (< 0.1) and even negative, and the regression slope shows large uncertainties. This suggests that CO2 emissions are dominated by non-CO-related sources in summer. In situ observations near Beijing also captured the low correlation in summer (Wang et al., 2010). The main reason is that the CO2 signals were significantly mixed with enhanced biospheric CO2 uptake during the growing season, which could offset the anthropogenic emissions in urban areas. It is necessary for CO and CO2 to share a common combustion process when using CO as the anthropogenic tracer for CO2 to investigate regional combustion efficiency. One straightforward approach is to remove the observations in the growing season (Wang et al., 2010; Shan et al., 2019). Still, we found some strong correlation days in summer (Fig. 5b), so we excluded the non-CO-related observations (R2 ≤ 0.1). Before filtering the data, (ΔXCO:ΔXCO2)d observed for urban Beijing is 8.94 ± 0.13 ppb ppm−1 with an R2 of 0.80. Removal of the strong biologically-affected days contributed to the elevated (ΔXCO:ΔXCO2)d of 10.46 ± 0.11 ppb ppm−1 and an R2 of 0.86.
Figure 5. (a) The time series of the daily slope of ΔXCO and ΔXCO2 observed by EM27 over IAP. The error bar represents the confidence bounds for the slope estimates. (b) The time series of correlation coefficient (R2) of ΔXCO and ΔXCO2. The red triangles represent observations with R2 less than 0.1, and the blue circles represent R2 larger than 0.1. Summertime is indicated by grey shading.
We extended our analysis to comparisons of (ΔXCO:ΔXCO2)d based on other FTS station datasets of the mid-latitude northern hemisphere in and around 2019 (Table 2). The FTS stations are set up in either an urban or suburban environment with varied emissions sources: the Pasadena site is located on the southern coastal air basin of California with a population of nearly 17 million, the Tsukuba site is located in a highly urbanized city near Tokyo with a population of 2.28 million, the Paris site is in the most populous city in France (2.24 million), the Karlsruhe site is a smaller urban region surrounded by forest, which has a population of nearly 0.3 million. Stronger correlations between ΔXCO and ΔXCO2 exist with high regression slope values in the densely populated urban regions (Beijing, Pasadena, Tsukuba, Paris, and Karlsruhe). The values of (ΔXCO:ΔXCO2)d observed in Chinese cities (10.46 ppb ppm−1 in Beijing, 6.76 ppb ppm−1 in Xianghe) are significantly higher than stations in other nations. In addition, (ΔXCO:ΔXCO2)d in Beijing is the highest among all the cities we included, implying a high pollution per amount of CO2 emissions and relatively lower combustion efficiency over Beijing.
FTS station (longitude, latitude) Time period ΔXCO:ΔXCO2 (ppb ppm−1) R2 Beijing, CHN (39.98°N, 116.39°E) 2019.1−2019.12 10.46 ± 0.11 0.86 Xianghe, CHN (39.75°N, 116.96°E) 2018.7−2019.7 6.76 ± 0.70 0.52 Karlsruhe, DE (49.10°N, 8.44°E) 2018.1−2019.12 1.84 ± 0.21 0.52 Pasadena, US (34.14°N, 118.13°E) 2019.1−2019.12 4.06 ± 0.18 0.61 Tsukuba, JP (36.05°N, 140.12°E) 2018.1−2019.9 4.68 ± 0.22 0.58 Paris, FR (48.97°N, 2.37°E) 2019.1−2019.12 3.06 ± 0.06 0.76
Table 2. Comparison of ΔXCO:ΔXCO2 in different FTS stations close or within the urban area in the northern hemisphere.
According to the in situ observations, Wang et al. (2010) found a ΔCO:ΔCO2 value of 41.7 ppb ppm−1 at the Miyun background site in Beijing from 2007 to 2008. Han et al. (2009) derived a ΔCO:ΔCO2 value of 43.4 ppb ppm−1 during the 2006 winter at an urban site in Beijing. Using XCO retrievals from NASA/Terra Measurement of Pollution in the Troposphere (MOPITT) and XCO2 retrievals from the Japan Aerospace Exploration Agency Greenhouse gases Observing Satellite (GOSAT), Silva et al. (2013) estimated ΔXCO:ΔXCO2 in Beijing/Tianjin to be 43.5 ppb ppm−1 from 2009 to 2010. These values are significantly larger than our estimated value (10.46 ppb ppm−1). The reason could be that China has implemented pollution control policies since 2013. As a result, the combustion efficiency significantly increased as CO emissions decreased (Zheng et al., 2018b; Feng et al., 2019b; Li et al., 2020). Shan et al. (2019) estimated ΔCO:ΔCO2 to be 15.99, 8.02, and 5.67 ppb ppm−1 based on in situ, ground-based FTS, and satellite measurements (GOSAT, MOPITT), respectively, at Hefei from 2015 to 2016. Li et al. (2020) estimated ΔCO:ΔCO2 to be 25.5 ppb ppm−1 at the SDZ site near Beijing in 2018. The ΔXCO:ΔXCO2 value we calculated (10.46 ppb ppm−1) is close to the value estimated from ground-based FTS at Hefei in 2015−16 (8.02 ppb ppm−1) and is about 60% lower than the near-surface observed value at SDZ (25.5 ppb ppm−1). The value of ΔCO:ΔCO2 based on FTS (8.02 ppb ppm−1) at Hefei is about 50% lower than in situ observations (15.99 ppb ppm−1) (Shan et al., 2019). In situ observations only capture the near-surface signal of a small sampling area in the local planetary boundary layer. In contrast, the FTS observations detect the whole layer of the atmosphere, which may weaken the near-surface signal. The FTS observations have larger footprint compared to in situ observations and could be more representative to the regional flux (Wunch et al., 2016)
Transportation governed by weather conditions plays an important role in the day-to-day variations in CO and CO2 in Beijing (Feng et al., 2019a; Panagi et al., 2020). Air pollution is concentrated in Beijing’s southern and eastern parts (Feng et al., 2019a). We identified source regions for each observation based on X-STILT footprints. Pathways are characterized by sources in the northwest (NW) and north China plain (NCP) according to the year-round average 24-hour backward footprints (Figs. 6c and 6e), which share 62.93% and 26.72% of the observation days, respectively. Higher XCO2 and XCO occurred when air masses originated from the NCP region. Clean air masses originating from the NW are less affected by human activities, which may cause the observed decrease in XCO2 and XCO.
Figure 6. (a) Correlations of ΔXCO and ΔXCO2 in 2019. (b, d) Correlations of ΔXCO and ΔXCO2 in 2019 originating from the NW and NCP upwind sources. R2 is the correlation coefficient of ΔXCO and ΔXCO2. slope2019 is the regression slope of ΔXCO and ΔXCO2. (c, e) Maps of mean 24-hour backward footprint [ppm / (µmol m−2 s−1), lg(x)] with 0.25° resolution at IAP, Beijing, starting at 1200 LST in 2019, originating from the NW (c) and NCP (e). The closed blue line indicates the BTH area. Only footprint values larger 10−2 ppm / (µmol m−2 s−1) are displayed.
When an air mass passes over different source regions, the correlation between CO and CO2 shows different patterns. As shown in Figs. 6b and 6d, (ΔXCO:ΔXCO2)d originating from the clean region is 8.23 ± 0.1 ppb ppm−1 and from the polluted region is 11.46 ± 0.2 ppb ppm−1. Advection that brings air masses containing emissions from the NCP contributed to an elevated proportion of (ΔXCO:ΔXCO2)d, which exceeded the annual slope (10.46 ± 0.11 ppb ppm−1).
The regression slope of ΔXCO:ΔXCO2, based on the regional enhancement method (ΔXCO:ΔXCO2)r, is estimated to be 9.06 ± 1.89 ppb ppm–1, which is consistent with the value of (ΔXCO:ΔXCO2)d (10.46 ± 0.11 ppb ppm–1). The ratio (ΔXCO:ΔXCO2)d is calculated solely based on the diurnal variation observed by EM27. In contrast, (ΔXCO:ΔXCO2)r is estimated by subtracting the background value from the observations. Uncertainty in background value yields more uncertainty for (ΔXCO:ΔXCO2)r compared to (ΔXCO:ΔXCO2)d.
The estimated values for ΔXCO and ΔXCO2 from regional enhancement versus the background were used for comparison with the model simulation at hourly timescales (Figs. 7 and 8). The modeled ΔXCO2 (hereafter ΔXCO2,sim) was simulated from the MEIC and PKU anthropogenic emission inventories (ΔXCO2, MEIC and ΔXCO2, PKU). The observed enhancement ΔXCO2 (hereafter ΔXCO2,obs) was the difference between the observed urban XCO2 (XCO2,obs) and the XCO2 background (XCO2,back) from model. Summer data is excluded due to the unavailable background data for CO (detailed in the next paragraph). The same trend is shared by XCO2,back and XCO2,obs as shown in Fig. 7a.
Figure 7. (a) Measured XCO2 and CAMS background concentration. (Only 1100 to 1600 LST periods are displayed). (b−c) Hourly measured and modeled regional enhancement ΔXCO2 for each fossil-fuel emission inventory (b for MEIC, c for PKU).
Figures 7b and 7c show the correlation between ΔXCO2,obs and ΔXCO2,sim. The x-intercept of the linear fitting equation of (~1 ppm for MEIC and PKU) represents the ΔXCO2,obs value with no anthropogenic effect. The bias of ΔXCO2,obs and ΔXCO2,sim was mainly attributed to the error from emission inventories, background XCO2 values, and transport simulation. Both the observed and modeled ΔXCO2 are in good agreement, with correlation coefficients (R2) of 0.70 and 0.73 for MEIC and PKU, respectively. The slope of the fitting equation denotes the ratio of the observed ΔXCO2 change to the modeled ΔXCO2 change. Many previous studies attempted to use the slope value as a scale factor to evaluate and constrain regional CO2 emissions (Sargent et al., 2018; Shekhar et al., 2020; Yang et al., 2020a). The slope for MEIC (0.89 ppm ppm–1) is closer to the 1:1 line than PKU (0.61 ppm ppm–1). According to the regression slope value, MEIC underestimates approximately 11% of CO2 emissions surrounding Beijing and PKU underestimates approximately 49%.
Figure 8 denotes the correlation plots of observed ΔXCO (ΔXCOobs) and modeled ΔXCO (ΔXCOsim). The ΔXCOobs data is correlated to ΔXCOsim with R2 of 0.71 and 0.73 for MEIC and PKU, respectively. The minimum of the observed XCO (80.0 ppb) was taken as a constant XCO background value. The x-intercepts for MEIC (77.63 ppb) and PKU (80.79 ppb) show consistency and agree with the minimum of observed XCO. However, the constant background value could not capture the variation of the true XCO background, especially for summer with strong biological-influenced (detailed in section 3.1). Therefore, the ΔXCO data in Summer are excluded. The slopes for MEIC (1.3 ppb ppb–1) and PKU (1.35 ppb ppb–1) indicate an overestimation of approximately 30% and 35%, respectively, for CO emissions surrounding Beijing.
Many studies have compared the observed ΔCO:ΔCO2 with emission inventories (Turnbull et al., 2011; Tohjima et al., 2014; Shan et al., 2019). To calculate the simulated ΔCO:ΔCO2 with emission inventories on a regional scale, it is essential to know that the observational site is representative of the region. A few studies roughly specified a geometric bounding box outlining the region which influences the observed value (Wunch et al., 2009; Hu et al., 2019; Shan et al., 2019). However, ΔCO:ΔCO2 calculated based on the geometric bounding method is sensitive to the specified size of the enclosed area. A circle centered upon the EM27 observation site was specified as the influencing region. As the source area radius ranges from 50 to 500 km, ΔCO:ΔCO2 varies from 17.00 to 19.77 ppb ppm–1 for MEIC, 32.28 to 53.74 ppb ppm–1 for PKU, respectively. The results show great uncertainty among different inventories and influencing areas. This method is based on the assumption that each grid in the specific geometric region of the emission map contributes equally to the observed concentration.
The region of influence and the sensitivity of each influencing grid to the observations vary over time. Using surface hourly backward column footprints for each measurement is a common and robust way to quantify the sensitivity of the atmospheric concentration changes at the receptor to upwind source regions using units of concentration per unit flux (Turnbull et al., 2011; Tohjima et al., 2014). Each footprint is convolved with the corresponding hourly gridded emission inventories (PKU, MEIC). The modeled anthropogenic enhancement of CO and CO2 at the receptor site is the sum of contributions from the sensitive emission grid flux (detailed in section 2.3). The linear regression slopes of the modeled ΔXCO:ΔXCO2 based on PKU, MEIC, and observations are shown in Table 3. The outliers are excluded according to the three standard deviations criterion. Modeled ΔXCO shows a good relationship with ΔXCO2 with R2 of 0.97 for MEIC and 0.96 for PKU. The modeled data displays a slightly greater correlation than the observed ΔXCO and ΔXCO2 (R2 = 0.86 for the diurnal variation method, R2 = 0.83 for the regional enhancement method). The reason is that modeled values only take the anthropogenic influence of CO into account, ignoring the CO2-related but not the CO-related signal, such as the resident respiration of CO2. The observed ΔXCO and ΔXCO2 values based on regional background enhancement display the weakest correlation (R2 = 0.83) due to the uncertainty of the modeled background value. The simulated regression slope of ΔXCO and ΔXCO2 in 2019 is 14.91 ± 0.36 ppb ppm−1 for MEIC and 21.04 ± 0.70 ppb ppm−1 for PKU. The MEIC and PKU inventories are 42.54% and 101.15% higher than the observed value (10.46 ± 0.11 ppb ppm−1), respectively.
Dataset ΔXCO:ΔXCO2 (ppb ppm−1) R2 Emission inventories MEIC 2019 14.91 ± 0.36 0.97 PKU 2019 21.04 ± 0.70 0.96 Observations FTS (diurnal variation) 2019 10.46 ± 0.11 0.86 FTS (regional enhancement versus background) 2019 9.06 ± 1.89 0.83
Table 3. Comparison of the observed and simulated ΔXCO:ΔXCO2.
In recent studies, ΔXCO:ΔXCO2 based on EDGAR (PKU) emission inventories are about 256.59% (219.39%) larger than the values calculated from the FTS in Hefei 2015−16, and 207.86% (173.31%) larger than those in 2016−17 (Shan et al., 2019). Silva and Arellano (2017) found that ΔCO:ΔCO2 based on the EDGAR inventory was 50% higher than the value estimated by satellites in the megacities of China. Wang et al. (2010) found that the bottom-up estimate of ΔCO:ΔCO2 was 19.2% larger than the observations at Miyun, Beijing, during winter 2006. Despite the observation and comparative methods, the emission inventories in urban China might overestimate ΔCO:ΔCO2. The lack of consideration of CO2 emissions from the respiration of the residents in dense urban regions may lead to the overestimation of bottom-up-based ΔCO:ΔCO2 (Wang et al., 2010). Either the overestimation of CO anthropogenic flux or under-consideration of CO sinks are possible reasons for the elevated bottom-up estimates of ΔCO:ΔCO2 (Vardag et al., 2015; Shan et al., 2019).
The main reason why MEIC and PKU overestimated ΔXCO:ΔXCO2 surrounding Beijing is likely due to the overestimation of the regional CO emissions and underestimation of CO2 emission. The ΔXCO and ΔXCO2 discussed in section 3.4 are directly linked to the regional emission. The difference between the modeled and observed ΔXgas is directly proportional to the difference between the emission inventories and the actual emission. The deviation of the regression fitting equation with the 1:1 line shows the model-observed difference. The slope value of modeled ΔXCO to observed is less than one, suggestive of typical overestimations of CO emissions of 30% and 35% for MEIC and PKU, respectively. The underestimation of CO2 emissions magnifies the effects of overestimated CO emission, which contributes to the larger difference between the modeled ΔXCO:ΔXCO2 and the observed ratio. For MEIC, a relatively smaller underestimation of CO2 emissions makes the modeled ΔXCO:ΔXCO2 closer to observations.
|Season||XCO2 (ppm)||XCO (ppb)|
(March to May)
|413.58 ± 1.25||109.58 ± 5.49|
(June to August)
|407.87 ± 3.34||120.50 ± 4.28|
(September to November )
|412.09 ± 2.88||112.37 ± 5.58|
(December to February)
|414.33 ± 2.65||115.78 ± 3.36|