Modeling the Atmospheric CO₂ Concentration in the Beijing Region and Assessing the Impacts of Fossil Fuel Emissions

Abstract

Reducing anthropogenic fossil fuel CO₂ (FFCO₂) emissions in urban areas is key to mitigating climate change. To better understand the spatial characteristics and temporal variations in urban CO₂ levels in the Beijing (BJ) region, we conducted a long-term CO₂ simulation study by using the Weather Research and Forecasting WRF-Chem model and CO₂ observation data. To assess the model performance, three representative sites with high-precision CO₂ observation data were chosen in this study: the rural regional background Shangdianzi (SDZ) site, the suburban Xianghe (XH) site, and the urban BJ site. The simulation results generally captured the observed variations at these three sites, but the model performed much better at the SDZ and XH sites, with mean biases of −0.7 ppm and −2.3 ppm, respectively, and RMSE of 12.3 ppm and 21.4 ppm, respectively. The diurnal variations in the model results agreed well with those in the observed CO₂ concentrations at the SDZ and XH sites during all seasons. In the meanwhile, the diurnal variations in the modeled FFCO₂ were similar to those in the CO₂ observation with a positive bias at the BJ site, which may have been caused by higher emissions especially in winter. Moreover, both the modeled FFCO₂ and biospheric CO₂ (BIOCO₂) have positive correlations with the observed CO₂ concentration, whereas the planetary boundary layer height (PBLH) and observed CO₂ concentration exhibited negative correlations at all sites. In addition, the contributions of FFCO₂ and BIOCO₂ to CO₂ varies depending on the seasons and the location of sites.

1. Introduction

With the increase in the concentration of greenhouse gases in the atmosphere, global warming has become a major issue [1]. Carbon dioxide (CO₂) accounts for the greatest proportion of greenhouse gases, almost all regions worldwide are exhibiting significant climate warming, and the increasing atmospheric CO₂ concentrations is currently the main contributor to global warming [2]. Since the Industrial Revolution, the average atmospheric CO₂ concentration at the Earth’s surface has increased from lower than 300 ppm to approximately 410 ppm [3]. Fossil fuel CO₂ (FFCO₂) emissions have become the main type of anthropogenic CO₂ emissions into the atmosphere, and their contribution is continuously increasing [4]. Although many efforts have been made to estimate FFCO₂ emissions, high uncertainties still exist in national and global assessments [5,6,7].

FFCO₂ emissions can be determined using a bottom-up method. In this method, the fossil fuel usage in each source sector can be combined with the estimated carbon content in each fuel type to obtain estimates of FFCO₂ emissions [8]. The spatiotemporal FFCO₂ emission datasets obtained by using the bottom-up method can clearly reveal the footprint of human activities, and the highest emissions occur near urban centers and related power plants [9,10]. An alternative FFCO₂ research method is the top-down method, in which global or regional transport models are used [11,12,13]. Network observation experiments have shown that there are factors causing biases in atmospheric transport models [14,15]. Previous studies have indicated that regional high-resolution models can better capture measured CO₂ signals and perform better at capturing diurnal CO₂ variations [16]. Recent research on urban-scale FFCO₂ emissions has benefited from the application of in situ observation strategies focusing on urban- and medium-scale transportation models [17,18,19,20]. In a high-resolution simulation experiment with multiple nested domains centered on Maryland, it was found that atmospheric transport errors in urban areas are not limited to a specific time of day [21].

The FFCO₂ emissions in China accounted for 28% of the global emissions in 2016. Notably, cities play an important role in global carbon emissions, especially in urban central areas, accounting for 67–76% of the global CO₂ emissions [22]. Recent research has revealed that there are differences in FFCO₂ emissions between different types of cities. The FFCO₂ concentration in the coastal city of Xiamen is lower than that in the inland cities of Xi’an and Beijing [23,24]. The Beijing, Tianjin, and Hebei (JJJ) region is the largest urbanized area in northern China and is facing considerable pressure to reduce emissions. Although there are carbon emission inventories for the JJJ region, comprehensive assessments of CO₂ emissions in districts and counties are still limited, which is crucial for the implementation of mitigation strategies [25]. As the capital of China, Beijing has more than 21.5 million permanent residents and more than 6.4 million motor vehicles at the end of 2019 [26,27], where both energy consumption and CO₂ emission levels significantly increased [28,29,30]. Although there has been a significant increase in international attention to global climate change in recent years, its urgency is often underestimated. The challenges in China are difficult, especially in regard to the use of fossil fuels [31]. To overcome these challenges, CO₂-related research has developed.

High-resolution simulations conducted in the BJ region revealed that the FFCO₂ emissions in Beijing mainly originate from industrial and residential sources, while weather patterns play an important role in transport [32]. However, few studies have focused on the long-term simulation performance for FFCO₂ sources during different seasons. In this paper, we combined high-precision CO₂ observation data from different in-situ sites and high-resolution long-term simulation results for the BJ region.

The purpose of this study was to improve the understanding of the temporal variation in the CO₂ concentration in the Beijing region at different sites and to determine the corresponding seasonal changes. Furthermore, we explored the factors influencing the CO₂ concentration, and future urban CO₂ simulation research directions were provided.

2. Materials and Methods

2.1. WRF-Chem Model

In this study, a modified version of the Weather Research and Forecasting (WRF)-Chem model (V3.7.1) was used to simulate the CO₂ concentration in the atmosphere above the JJJ region of China. The total CO₂ concentration comprises 3 components, namely, background CO₂ (BGCO₂), FFCO_2, and biospheric CO₂ (BIOCO₂), including vegetation photosynthesis and respiration, microbes, and animal and human respiration. FFCO₂ emissions were simulated by using the emission inventory of the National Development and Reform Commission (NDRC) and the Multiresolution Emission Inventory for China (MEIC). The anthropogenic emission inventory of the NDRC exhibits a high spatial resolution of up to 0.015° × 0.015°, and the total emission amount in the JJJ region is provided by the province. Moreover, emission sources are classified as industrial, cement, power, heating, residential life, and other types [33]. In the MEIC, emission sources are divided into industrial, residential, power, and transport types, and this inventory exhibits a horizontal resolution of up to 0.25° × 0.25° [34].

BIOCO₂ emissions were simulated by the Vegetation Global Atmosphere Soil (VEGAS) model. The VEGAS model is a typical dynamic global vegetation model (DGVM) that can be used to simulate the growth of different plant types and emissions from animals and human beings. In this model, vegetation absorbs CO₂ in the atmosphere through photosynthesis and transfers carbon to the soil through processes such as withering and tree fall. The soil carbon pool is continuously transported into the atmosphere via a land carbon cycle encompassing processes such as combustion and degradation [35]. In this study, BIOCO₂ was simulated by the VEGAS model, with a temporal resolution of one hour and a spatial resolution of 1° × 1°. BGCO₂ were simulated by the GEOS-Chem model, with a horizontal resolution of 4° × 5° and 47 vertical levels [36,37]. The time intervals of both the VEGAS and GEOS-Chem models were set to one hour for chemical and transport processes.

We analyzed the atmospheric CO₂ concentration from March 2019 to March 2020 in the BJ region. The WRF-Chem model was configured with a spacing of 9 km × 9 km and was centered at 116.5° E and 39.3° N (Figure 1b). The six-hour initial and boundary meteorological conditions were obtained from the National Oceanographic and Atmospheric Administration (NOAA) National Centers for Environmental Prediction (NECP) final (FNL) 0.25° × 0.25° reanalysis data. Table 1 provides a summary of the model configuration settings.

For analyzing the difference between model and observation at the same location, the four grid points closest to the observation position were selected for linear interpolation to one point.

2.2. CO₂ Observations

The three triangles in Figure 1c denote the locations of the high-precision sites in the Beijing (BJ) region. The BJ site is located in a high-urbanization area surrounded by dense residential buildings. The measurement data came from the observation of the 325 m meteorological tower (49 m above sea level) at an altitude of 80 m. Within 1 km around the site, the vegetation height ranges from 15–20 m, the vegetation coverage rate ranges from 10%–18%, and the height of buildings ranges from 70–200 m [38,39]. The Xianghe (XH) site is located in the southeastern suburbs approximately 50 km away from Beijing. The land use types are mainly croplands and irrigated farmlands. Within 1 km of the XH site, the buildings around the site are mainly residential houses with a height of under 20 m [40]. The measurements data came from the 60 m height of the XH site. The Shangdianzi (SDZ) site is located at the center of the JJJ region, 150 km away from Beijing. It exhibits a temperate semi-humid monsoon climate. The surrounding land use types are forestland, orchards, and farmland. There is only one village 0.8 km south of the site, and there is no obvious industrial emission source within 30 km [41]. The data of the 16 m-high SDZ site are used in research.

The Picarro cavity ring-down spectroscopy (CRDS) G2301 analyzers (Picarro, Santa Clara, CA, USA) were produced in the United States and installed at three sites to measure CO₂ mole fraction. The original data were calibrated by a gas system with sample air, target gas, and calibration gas [40]. All data were obtained from China Meteorological Administration (CMA) and had been strictly quality controlled.

2.3. Statistical Parameters

To compare the simulated and observed CO₂ concentrations, the mean bias (MB) and root mean square error (RMSE) were calculated as follows:

$$MB={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N}\left(M_i-O_i\right)$$

(1)

$$RMSE={\left({\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N}{\left(M_i-O_i\right)}^2\right)}^{{\displaystyle \frac{1}{2}}}$$

(2)

where $M_i$ and $O_i$ denote the simulated and observed values, respectively, and N is the number of data points. To illustrate the difference between the model and observation results, the simulation results were interpolated to in situ locations.

$$C_{FFE}= {\displaystyle \frac{\frac{1}{N}{\sum }_{i=1}^{N}FFC{O}_2}{\frac{1}{N}{\sum }_{i=1}^{N}FFC{O}_2 + \frac{1}{N}{\sum }_{i=1}^{N}BIOC{O}_2}}$$

(3)

$$C_{BIO}= {\displaystyle \frac{\frac{1}{N}{\sum }_{i=1}^{N}BIOC{O}_2}{\frac{1}{N}{\sum }_{i=1}^{N}FFC{O}_2 + \frac{1}{N}{\sum }_{i=1}^{N}BIOC{O}_2}}$$

(4)

We used the means of the modeled FFCO₂ and BIOCO₂ levels during a certain period to determine their corresponding contributions to the variation in the modeled CO₂ concentration. In the above equations, $C_{FFE}$ and $C_{BIO}$ are the contributions of FFCO₂ and BIOCO₂, respectively.

Table 1. Configuration settings of the WRF-Chem model.

Option	Configuration
Simulation period	January 2019 to January 2021
Simulation region	Beijing–Tianjin–Hebei region, China (Figure 1)
Domain center	116.5° E and 39.3° N
Horizontal resolution	9 km × 9 km
Vertical levels	39 vertical levels (from the surface to 50 hPa)
Microphysics scheme	WRF single-moment 5-class scheme [42]
Boundary layer scheme	Yonsei University scheme [43]
Surface layer scheme	MM5 similarity scheme [44]
Land surface scheme	Unified Noah land surface model [45]
Longwave radiation scheme	Rapid radiative transfer model (RRTM) longwave scheme [46]
Shortwave radiation scheme	Goddard shortwave scheme [47]
Meteorological field	NCEP 0.25° × 0.25° reanalysis data

Table 1. Configuration settings of the WRF-Chem model.

Option	Configuration
Simulation period	January 2019 to January 2021
Simulation region	Beijing–Tianjin–Hebei region, China (Figure 1)
Domain center	116.5° E and 39.3° N
Horizontal resolution	9 km × 9 km
Vertical levels	39 vertical levels (from the surface to 50 hPa)
Microphysics scheme	WRF single-moment 5-class scheme [42]
Boundary layer scheme	Yonsei University scheme [43]
Surface layer scheme	MM5 similarity scheme [44]
Land surface scheme	Unified Noah land surface model [45]
Longwave radiation scheme	Rapid radiative transfer model (RRTM) longwave scheme [46]
Shortwave radiation scheme	Goddard shortwave scheme [47]
Meteorological field	NCEP 0.25° × 0.25° reanalysis data

3. Results and Discussion

3.1. Modeled CO₂ Concentration

The variation trend of the simulated CO₂ concentrations was similar to that of the observations, and the time intervals of high and low CO₂ concentrations matched well. However, during some periods, the simulated CO₂ concentrations could hardly match the observed values. For example, during the period from March to April 2019, the observed CO₂ concentrations at the BJ site (Figure 2a) were low, while the simulation concentrations were relatively high. Similar situations frequently occurred in winter. At the BJ site, the simulation results were more consistent during the second half of spring and the first half of summer, while there were mostly small deviations during other periods (Figure 2b). Despite the differences between the BJ and XH sites, the simulated values were lower than the observations at the XH site during the corresponding period from March to April 2019. Notably, during the second half of spring and from May to June in summer, the simulated CO₂ concentrations were mostly lower than the observed values. In July, the simulated and observed values corresponded well. However, deviations were obtained in autumn and winter, although the overall simulation effect was better than that at the BJ site (Figure 2c,d). At the SDZ site, the simulated results were very consistent with the observed values from March to April in spring. In contrast, from May to August, the observed values were greater than the simulated values. Positive and negative deviations alternately emerged in autumn (Figure 2e,f).

The MB and RMSE were used to study the differences between the observed and modeled CO₂ concentrations. Over the 2019–2020 period, the MB values at the BJ, XH, and SDZ sites were 25.2 ppm, −2.3 ppm, and −0.7 ppm, respectively. The RMSE values at the BJ, XH, and SDZ sites were 62.4 ppm, 21.4 ppm, and 12.3 ppm, respectively. Closer to the urban center, the MB and RMSE were greater, but these values could not directly indicate which site exhibited a better simulation performance. However, according to the results, the BJ site demonstrated the highest deviation, and the simulation results and observations greatly differed.

Notably, the maximum seasonal differences at the BJ, XH, and SDZ sites were 61.3 ppm, −5.8 ppm, and −4.4 ppm, respectively (Figure 2g–i). There were differences between the SDZ site and the other two sites. The maximum deviation at the SDZ site occurred in summer, whereas that at the other two sites occurred in winter. The deviations during the different seasons could be explained by weather processes, carbon budgets in the biosphere, and simulations of anthropogenic emissions. A comparative analysis of the standard deviation (SD) values revealed that the BJ site exhibited significantly greater difference (SD = 29.1 ppm) compared to the XH (SD = 7.9 ppm) and SDZ (SD = 6.1 ppm) sites. Based on the site location, it could be considered that the uncertainties in vegetation carbon budgets in summer and anthropogenic emissions in winter led to seasonal deviations. Moreover, the SDZ site is far from anthropogenic emission sources, where the main factor causing the highest deviation in summer is the vegetation carbon budget.

3.2. Diurnal Variation in the CO₂ Concentration

Figure 3 shows a comparison of the diurnal variations in the CO₂ concentrations in spring, summer, autumn, and winter in Beijing (BJ time or BJT), as well as the annual average, between the simulations and observations at the BJ, XH, and SDZ sites. In spring, the modeled CO₂ concentration at the BJ site was higher from 0 to 10 and 18 to 24 BJT, while the variations and amplitudes were very consistent from 10 to 18 BJT (Figure 3a). The maximum deviation occurred from 0 to 10 BJT, at approximately 10–20 ppm. The patterns of the observations and simulations in summer were similar to those in spring (Figure 3b). From 0 to 18 and 20 to 24 BJT, the simulated CO₂ concentration was higher than the observed concentration. The maximum deviation before sunrise reached approximately 40 ppm, and only the change trend and value were consistent between 18 and 20 BJT. In autumn (Figure 3c), the trend of the modeled and observed concentrations between 0 and 15 BJT was similar to that in summer, with higher simulated values than the observed values. The maximum deviation occurred slightly later than that in summer. The trends and values of the modeled CO₂ concentration were very consistent with those observed between 15 and 24 BJT. In winter (Figure 3d), the diurnal changes in the simulated CO₂ concentration greatly differed from those in the observed concentration. The simulated CO₂ concentration was much greater than the observed concentration, but the trend was similar to that of the observations in autumn. The diurnal variations in the simulated CO₂ concentrations in spring and summer were consistent with those in the observations at the XH site (Figure 3f,g). In autumn (Figure 3h), the simulated change range was slightly greater than that of the observations. From 0 to 10 BJT, the simulated values were greater than the observed values, with a maximum deviation of approximately 10 ppm. From 10 to 24 BJT, the simulated values were lower than the observed values, with a deviation of less than 10 ppm. In winter (Figure 3i), the simulation values were lower than the observations from 8 to 24 BJT, and the trends of the simulations did not match those of the observations. Starting at approximately 8:00, the simulated CO₂ concentration began to decrease, while the observed concentration decreased later. At the SDZ site, the simulation performance was satisfactory in spring (Figure 3k), autumn (Figure 3m), and winter (Figure 3n), and the model could suitably capture the diurnal change trend of CO₂. The change range of the simulated CO₂ concentrations in summer was significantly small (Figure 3l), while the change range of the observed CO₂ concentrations was large. The decrease in the CO₂ concentration from 6 to 12 BJT reflects the high photosynthesis of vegetation, but the simulation did not capture this decrease.

According to the overall annual average diurnal variations in CO₂ at the three sites, the model performance at the XH site was the best, followed by those at the SDZ and BJ sites. The diurnal variations in the simulations at these two sites were relatively consistent with those in the observations, but the valley value was overestimated at the XH site, and the peak value was underestimated at the SDZ site (Figure 3j,o). The simulation performance at the BJ site fluctuated the most.

3.3. Characteristics of the Modeled FFCO₂ and BIOCO₂

The relationships between the observed CO₂ concentrations and the simulated FFCO₂ emissions during the different seasons at the three sites are shown in Figure 4a–c, while the distributions of the observed CO₂ concentrations and simulated FFCO₂ are shown on the top and right coordinate axes, respectively. The CO₂ concentrations at all three sites were positively correlated with the simulated FFCO₂. The highest correlation coefficient (r) at the BJ site occurred in spring, with r = 0.66 and p < 0.01, indicating a high correlation between the modeled FFCO₂ emissions and the observed CO₂ concentrations. Moreover, the lowest correlation coefficient (r) occurred in summer, with r = 0.29 and p < 0.01, indicating that the CO₂ concentration was less correlated with the FFCO₂ emissions. In spring, the distributions of the observed CO₂ concentrations and simulated FFCO₂ values at the BJ site were relatively concentrated. The CO₂ concentration was mainly below 450 ppm, while the FFCO₂ value was mostly lower than 50 ppm. The distributions of both types of data were scattered in winter. At the XH site, the correlation was the lowest in summer, with r = 0.45 and p < 0.01. Although the XH site is not located in the urban center area, the CO₂ concentration was still correlated with the FFCO₂ value in winter, with r = 0.81 and p < 0.01. At this site, the distributions of the observed CO₂ concentrations and modeled FFCO₂ values were similar to those at the BJ site, but the modeled FFCO₂ value was less than 30 ppm. At the SDZ site, the correlation with FFCO₂ emissions was the lowest in summer, with r = 0.52 and p < 0.01, whereas the correlation was high in winter, with r = 0.88 and p < 0.01. Compared to that at the other two sites, the distribution of the observed CO₂ concentrations was significantly narrower, with values less than 430 ppm, while the distributions of the modeled FFCO₂ values during the different seasons were more similar than those at the other two sites. Although sites XH and SDZ occur far from urban areas, the variation in the CO₂ concentration was partly related to FFCO₂, which supports the important role of regional transport in determining FFCO₂ levels in the BJ region.

To investigate the contribution of FFCO₂ to the total CO₂ concentration in the simulations, the fractions of the modeled FFCO₂ at the three sites during the different seasons are shown in

Table 2. The percentages (%) of the modeled FFCO₂ values to the total simulated CO₂ concentrations at the Beijing (BJ), Xianghe (XH), and Shangdianzi (SDZ) sites during the different seasons.

	Spring	Summer	Autumn	Winter
BJ	12.0	18.1	26.4	43.5
XH	15.1	23.6	29.5	31.7
SDZ	16.4	26.3	31.6	25.7

. The lowest fractions among the various seasons at the three sites occurred in spring. The proportion at the BJ site was 12%, which is lower than that at the XH (15.1%) and SDZ (16.4%) sites. The second lowest fractions of FFCO₂ at the three sites occurred in summer, at 18.1%, 23.6%, and 26.3%, respectively. The fraction of the modeled FFCO₂ at the SDZ site in summer was not low. In autumn, the differences between the fractions of FFCO₂ at the three sites were small (26.4%, 29.5%, and 31.6%). The fractions at all three sites were relatively high. In addition, the SDZ site had the highest percentage during the various seasons. At the BJ site, the proportion of the simulated FFCO₂ in winter was considerable (43.5%), indicating a notable increase between autumn and winter. At the XH site, the fraction of the modeled FFCO₂ in winter reached 31.7%, which only increased by 2.2% (31.7%−29.5%). Interestingly, the fraction of the modeled FFCO₂ at the SDZ site in winter was 25.7%, a decrease of 5.9% from that in autumn (31.6%−25.7%).

According to the time series of the diurnal variations in the modeled FFCO₂ (Figure 5), the FFCO₂ values at the BJ, XH, and SDZ sites varied between 40 and 80 ppm, 10 and 30 ppm, and 5 and 10 ppm, respectively. Obviously, similar trends were observed between the modeled FFCO₂ values and observed CO₂ concentrations, especially at the SDZ site. The slight difference was that the FFCO₂ values at the BJ and XH sites varied greatly, with the highest value occurring at approximately 06:00 and the lowest value occurring from 15:00 to 16:00. The change rate at the SDZ site was relatively slow, and the value began to decrease at 03:00 and decreased to the lowest level before 15:00.

The time series of the modeled BIOCO₂ values at the three sites are shown in Figure 6a–c. Clearly, high negative BIOCO₂ values were observed at each site in summer. However, there were still positive BIOCO₂ values at the XH site in summer. According to the seasonal mean of the BIOCO₂ values at the three sites (Figure 6d–f), except for summer at the SDZ site, the BIOCO₂ values were positive. The mean BIOCO₂ value at SDZ in summer was −0.2 ppm, which is a negligible value. The seasonal mean BIOCO₂ value at each site was the highest in autumn. The values at the BJ, XH, and SDZ sites were 8.2 ppm, 8.9 ppm, and 6.2 ppm, respectively. Based on box plot analysis, the change in the BIOCO₂ value was relatively small in spring and summer. In contrast, the BIOCO₂ value changed the most in autumn. A comparative analysis of the standard deviation (SD) values revealed that there was no significant difference at each site. BJ and SDZ sites showed less difference during the Spring (BJ: 1.2 ppm; SDZ: 1.1 ppm), while XH site was in summer (1.4 ppm).

The relationships between the modeled BIOCO₂ values and observed CO₂ concentrations at the three sites are shown in Figure 6g–i. Interestingly, there was a positive correlation between the modeled BIOCO₂ values and observed CO₂ concentrations at the three sites. In particular, the r value at the XH site was the highest, at 0.64. The r values at the BJ and SDZ sites were 0.62 and 0.54, respectively (p < 0.01). From the perspective of the diurnal variations (Figure 7), the diurnal variations in the simulated BIOCO₂ values at the three sites agreed well with those in the observed CO₂ concentrations. The change trend at each site indicated a typical single-peak feature (a peak and a valley). The consistency at the SDZ site was the greatest. The reason is that the diurnal variation in FFCO₂ emissions was not obvious, which slightly impacted the diurnal variation in CO₂. Therefore, the variation in the CO₂ concentration at the SDZ site was dominated by that in the BIOCO₂ value.

We analyzed the contributions of FFCO₂ ($C_{FFE}$) and BIOCO₂ ($C_{BIO}$) to the change in the CO₂ concentration at the monthly scale, as shown in Figure 8. $C_{BIO}$ was negative because of the negative BIOCO₂ value provided by the notable carbon sink in the biosphere. For example, in June and July, $C_{FFE}$ contributed more than 100%. The months with the largest $C_{FFE}$ values were June, June, and July at the BJ (101.4%), XH (101.6%), and SDZ (117.1%) sites, respectively, and the lowest $C_{BIO}$ values occurred at the same time (BJ: −1.4%; XH: −1.6%; SDZ: −17.1%). Except for June and July, the $C_{BIO}$ value in the other months was positive. In contrast, the largest $C_{BIO}$ values occurred in October at all three sites (BJ: 15.7%; XH: 40.1%; SDZ: 50.2%). This indicated a positive contribution of BIOCO₂ emissions at the three sites. This contribution mainly originated from vegetation, microbes, animals, and human respiration.

3.4. Characteristics of the Modeled PBLH

To study the relationship between the planetary boundary layer height (PBLH) and the CO₂ concentration, the time series of the weekly CO₂ concentration and modeled PBLH at the three sites are shown in Figure 9a–c. The variations in the modeled PBLH at the three sites were similar, which shows that the simulated PBLH is relatively high in spring and summer, at approximately 1000 m, corresponding to a low CO₂ concentration. The modeled PBLH decreased in autumn and winter, reaching approximately 500 m most of the time, corresponding to a low CO₂ concentration. There was an obvious cross point of the trends of the modeled PBLH and CO₂ concentration in September, which indicates that they exhibit opposite variations during the different seasons. Additionally, the diurnal variations in both the modeled PBLH and observed CO₂ concentration (in BJT) at the three sites are shown in Figure 9d–f. The diurnal variations in the PBLH at the three sites were similar. The lowest value occurred at 06 BJT before sunrise, which was lower than 400 m. After sunrise, the PBLH began to slowly increase and reached its highest value from 14 to 15 BJT in the afternoon. The PBLH at all three sites reached up to 1400 m, while the BJ site exhibited the greatest height. Within the corresponding period, the CO₂ concentration increased (decreased) to its highest (lowest) value when the modeled PBLH reached its lowest (highest) value. There were obvious opposite trends in the diurnal variations between the modeled PBLH and the observed CO₂ concentration.

To evaluate the correlation between the modeled PBLH and CO₂ concentration at the three sites, scatter plots are shown in Figure 9g–i. The CO₂ concentrations at all three sites showed a negative correlation with the simulated PBLHs. At the XH site, r = −0.46, p < 0.01, indicating a negative correlation between the simulated PBLH and the observed CO₂ concentration. The r value at the BJ site was the same as that at the XH site, which is r = −0.46, p < 0.01. There was still a negative correlation between the modeled PBLH and the observed CO₂ concentration at the BJ site. In contrast, the r value at the SDZ site slightly differed from that at the other sites, with r = −0.44 and p < 0.01. The above results show that the modeled PBLF and the observed CO₂ concentration were negatively correlated at all sites.

4. Conclusions

High-resolution simulations of the CO₂ concentration from March 2019 to March 2020 were analyzed in this study to investigate its characteristics by using the WRF-Chem transport model. The model results were compared with high-precision measurement data from three observation sites in the BJ region.

Although differences were observed, most simulations reasonably corresponded to the observations. Throughout the 2019–2020 period, the simulated CO₂ concentrations at the XH and SDZ sites were more accurate than those at the BJ site. The deviation between the simulations and observations at site BJ was the greatest, and the MB was positive regardless of the season. Based on a comparison of the diurnal variations among the different seasons, the simulation results at the XH site remained highly consistent throughout the year. Anthropogenic factors are the primary causes for the significantly higher simulated CO₂ concentrations and amplitudes compared to observations at the BJ site, particularly in winter, resulting in poor model-observation consistency.

It is meaningful that through this study, we have evaluated the reliability of the model in the study of carbon emissions in urban areas, especially seasonal variations, which provides some references for the long-term CO₂ simulation.

The novelty of this study and the performance of the model were as follows:

• We quantified the FFCO₂ effects on atmospheric CO₂ concentrations from urban to regional background sites. There was a positive correlation between the modeled FFCO₂ and the observed CO₂ concentration at each site, particularly during spring and winter. The BJ and XH sites exhibited the greatest contributions of FFCO₂ to the total modeled CO₂ concentration.
• We separated the biosphere contribution to CO₂ variations. There was a negative correlation of modeled BIOCO₂ and observed CO₂ concentration in summer.
• We studied the impacts of meteorological factors on CO₂ variations. The negative correlation between modeled PBLH and observed CO₂ had been confirmed.

There are still several aspects that can be improved in our future research:

• Resolution: At present, the resolution of our simulation research is not fine enough for local scale. If we need to study some details in urban areas, such as street level, we should combine higher-resolution local inventory and model grids;
• The impact of high value point sources: through the elimination of the sources, we could quantitatively analyze its contribution to different sites;
• The impact of COVID: The mismatch between the model and observation are partly due to variations in carbon emissions during COVID. By adding factors affecting carbon emissions during COVID, the accuracy of simulation is expected to be improved.