Comparative Analysis of PM 2.5 and O 3 Source in Beijing Using a Chemical Transport Model

: For many years, Beijing has suffered from severe air pollution. At present, ﬁne particulate matter (PM 2.5 ) pollution in the winter and ozone (O 3 ) pollution in the summer constitute serious environmental problems. In this study, the combination of a comprehensive air quality model with particulate matter source apportionment technology (CAMx-PAST) and monitoring data was used for the high-spatial resolution source apportionment of secondary inorganic components (SNA: SO 42 − , NO 3 − , and NH 4+ ) in PM 2.5 ; their corresponding precursor gases (SO 2 , NO 2 , and NH 3 ); and O 3 in the winter and summer over Beijing. Emissions from residents, industry, trafﬁc, agriculture, and power accounted for 54%, 25%, 14%, 5%, and 2% of PM 2.5 in the winter, respectively. In the summer, the emissions from industry, trafﬁc, residents, agriculture, and power accounted for 42%, 24%, 20%, 10%, and 4% of PM 2.5 , respectively. The monthly transport ratio of PM 2.5 was 27% and 46% in the winter and summer, respectively. The regional transport of residential and industrial emissions accounted for the highest proportion of PM 2.5 . The regional transport of emissions had a signiﬁcant effect on the SO 42 − and NO 3 − concentrations, whereas SO 2 and NO 2 pollution were mainly affected by local emissions, and NH 4+ and NH 3 were mainly attributed to agricultural emissions. Industrial and trafﬁc sources were two major emission sectors that contributed to O 3 pollution in Beijing. The monthly transport ratios of O 3 were 31% and 65% in the winter and summer, respectively. The high-spatial resolution regional source apportionment results showed that emissions from Langfang, Baoding, and Tangshan had the greatest impact on Beijing’s air pollution. This work’s methods and results will provide scientiﬁc guidance to support the government in its decision-making processes to manage the PM 2.5 and O 3 pollution issues.


Introduction
Beijing, the capital of China, has undergone rapid socioeconomic change, which has placed tremendous pressure on the ecological environment and led to its deterioration. In recent decades, Beijing has suffered severe air pollution. At present, fine particulate matter (PM 2.5 ) pollution in the winter and ozone (O 3 ) pollution in the summer constitute severe environmental problems [1][2][3]. The unique terrain of Beijing, which is surrounded by the Yan Mountains to the northwest and the Taihang Mountains to the southwest, is conducive to trapping air pollution over the city [4]. To effectively control regional pollution and improve the air quality, central and local governments have implemented a series of air pollution control action plans since 1998. At that time, the annual ambient concentrations of SO 2 and NO 2 in Beijing were 120 µg m −3 and 74 µg m −3 , respectively. In 1998, policies were launched to reduce the emissions from coal-fired power plants, have been widely employed in China. Yu et al. [21] applied a PMF model to identify dust, coal combustion, secondary sulfate, and industrial emissions; vehicle emissions and secondary nitrates were identified as the two major categories of emissions in Beijing. Song et al. [22] used three receptor models (UNMIX, PMF, and CMB models), which were used to apportion the ambient volatile organic compounds (VOCs) in Beijing. All the models' results showed that emissions from gasoline-powered vehicles contributed the most to VOCs. In addition, the contributions of different emission categories and source regions to the PM 2.5 concentrations in the six central areas of Beijing were quantified by utilizing the particulate matter source apportionment technology (PAST) in the CAMx model [23]. Gao et al. [24] applied the WRF-Chem model combined with an online O 3 tagging method to study the O 3 contributions from different source regions over the Yangtze River Delta (YRD) region to the surface O 3 during a frequent wind-shifting period. The whole YRD was seriously affected by the four source regions-namely, YRD, AnHui, ShanDong and HeNan&HeBei-with relative mean contributions of 15.0%, 16.2%, 13.6%, and 9.0%, respectively. Wang et al. [25] determined the contributions of summertime O 3 in China using regional chemical transport models. Liu et al. [26] used a PMF method and revealed that soil dust, industry, secondary formation, and vehicle emissions were the main sources of PM 2.5 in the North China Plain. However, comparative analyses of PM 2. 5 and O 3 sources in Beijing are still lacking, and thus, the source apportionment of different PM 2.5 components and O 3 over Beijing, China's capital, is far from complete. Air pollutants are influenced by emissions, topography, and meteorology. An emissions inventory is fixed over a specific period, but meteorological conditions change both daily and hourly. Thus, it remains a challenge to address the complicated effects of pollution emissions, meteorological conditions, precursor gases, and secondary pollutant concentrations. To better control PM 2.5 and O 3 pollution, a comprehensive source apportionment study that reports different PM 2.5 components, O 3 , and their precursor gases is needed to provide a complete understanding. Additionally, it is crucial not only to know the contributions of emissions from different source categories but, also, to understand the detailed information of the regional transport contributions.
The objectives of this work are to study the high-spatial resolution source apportionment of secondary inorganic components (SNA: SO 4 2− , NO 3 − , and NH 4 + ) in PM 2.5 and their corresponding precursor gases (SO 2 , NO 2 , and NH 3 ), as well as O 3 , in the winter and summer over Beijing. Then, we introduced regional sensitivity coefficients to express the impacts of regional topography and meteorology on Beijing air quality. The CAMx model is applied over Northern China in combination with a PSAT probing tool to achieve this goal. The methods and simulation design are described in Section 2. The simulation results are compared with the monitoring data. High-spatial resolution source apportionment and regional transport results are presented in Section 3. This work's methods and results provide a scientific reference for supporting the government in their decision-making processes for the more targeted management of the PM 2.5 and O 3 pollution issues.

Model Description and Configuration
Meteorological conditions were simulated by employing the Weather Research and Forecasting (WRF) model (Version 3.4) [27]. The initial meteorological data and boundary conditions were derived from the National Centers for Environmental Prediction (NECP) Final Analysis reanalysis data. The New Thompson, Goddard short-wave, rapid radiative transport model long-wave, Yonsei University planetary boundary layer meteorology, Noah land surface model, and New Grell schemes were selected for the simulation. The CAMx model, version 6.0, was used to study the source apportionment in this study. CAMx is a kind of Eulerian photochemical dispersion model that allows comprehensive "one-atmosphere" assessments of air pollution to be conducted over spatial scales from neighborhoods to continents. The CAMx model is an offline 3D chemical transport model. In this work, the Carbon Bond 2005 (CB05) mechanism was used for the gas-phase chem-istry [28]. The aerosol chemistry scheme was chosen from the Regional Acid Deposition Model (RADM-AQ) [29]. The piecewise parabolic method (PPM) and Euler backward iterative (EBI) solvers were applied to the horizontal advection solution options and gas-phase chemistry solver options, respectively. Different PM 2.5 components, such as NO 3 − ; SO 4 2− ; secondary organic aerosols (SOAs); primary PM (primary organic aerosols, elemental carbon, crustal fine, crustal coarse, and other coarse particles); and O 3 , can be distributed in the PSAT module of CAMx. PSAT has been universally applied for source apportionment studies [30,31].
The coverage area of the WRF/CAM X model was a portion of Northern China. The horizontal resolution was 9 km, divided into 220 × 200 horizontal grid cells (as shown in Figure 1a). The modeling vertical resolution was divided into thirty logarithmic structure layers. These layers ranged from the surface to a fixed pressure of 100 mb. The base emission inventory input in the study used the 0.25 • Multiresolution Emission Inventory for China (MEIC) of 2016 (http://meicmodel.org, accessed on 15 March 2020). The MEIC included five categories of sources (i.e., industry, traffic, residential, power, and agriculture). neighborhoods to continents. The CAMx model is an offline 3D chemical transport model. In this work, the Carbon Bond 2005 (CB05) mechanism was used for the gas-phase chemistry [28]. The aerosol chemistry scheme was chosen from the Regional Acid Deposition Model (RADM-AQ) [29]. The piecewise parabolic method (PPM) and Euler backward iterative (EBI) solvers were applied to the horizontal advection solution options and gasphase chemistry solver options, respectively. Different PM2.5 components, such as NO3 − ; SO4 2− ; secondary organic aerosols (SOAs); primary PM (primary organic aerosols, elemental carbon, crustal fine, crustal coarse, and other coarse particles); and O3, can be distributed in the PSAT module of CAMx. PSAT has been universally applied for source apportionment studies [30,31]. The coverage area of the WRF/CAMX model was a portion of Northern China. The horizontal resolution was 9 km, divided into 220 × 200 horizontal grid cells (as shown in Figure 1a). The modeling vertical resolution was divided into thirty logarithmic structure layers. These layers ranged from the surface to a fixed pressure of 100 mb. The base emission inventory input in the study used the 0.25° Multiresolution Emission Inventory for China (MEIC) of 2016 (http://meicmodel.org, accessed on 15 March 2020). The MEIC included five categories of sources (i.e., industry, traffic, residential, power, and agriculture).

Simulation Design
PSAT is a species tagging method that marks source regions and emission categories based on the pollutants of interest designed by the user. Beijing was selected as the receptor region. The grid cell at the center of the city was regarded as the receptor. Five emission source categories and 13 source region emission groups were considered in the model. The five emission categories were labeled industry, traffic, residential, power, and agriculture. The entire domain was divided into 13 source regions: Beijing, Tianjin, Baoding, Tangshan, Langfang, Zhangjiakou, Chengde, Cangzhou, Shijiazhuang, Qinhuangdao, Hengshui, Xingtai, and Handan. The source regions were defined based on their geographical borders (shown in Figure 1b). The model simulation was applied for January and July of 2016 with the corresponding meteorological conditions. The month of January was chosen to represent the winter, and July was chosen to represent the summer. The emissions were based on the emission inventory of 2016. Regional sensitivity coefficients were examined based on the CAMx/PSAT simulation. Here, the same pollutant emissions were set in each simulation grid. The regional contributions to air pollution in Beijing were designed to represent the regional sensitivity coefficients. Table 1 summarizes the configurations of the model sensitivity simulation.

Simulation Design
PSAT is a species tagging method that marks source regions and emission categories based on the pollutants of interest designed by the user. Beijing was selected as the receptor region. The grid cell at the center of the city was regarded as the receptor. Five emission source categories and 13 source region emission groups were considered in the model. The five emission categories were labeled industry, traffic, residential, power, and agriculture. The entire domain was divided into 13 source regions: Beijing, Tianjin, Baoding, Tangshan, Langfang, Zhangjiakou, Chengde, Cangzhou, Shijiazhuang, Qinhuangdao, Hengshui, Xingtai, and Handan. The source regions were defined based on their geographical borders (shown in Figure 1b). The model simulation was applied for January and July of 2016 with the corresponding meteorological conditions. The month of January was chosen to represent the winter, and July was chosen to represent the summer. The emissions were based on the emission inventory of 2016. Regional sensitivity coefficients were examined based on the CAMx/PSAT simulation. Here, the same pollutant emissions were set in each simulation grid. The regional contributions to air pollution in Beijing were designed to represent the regional sensitivity coefficients. Table 1 summarizes the configurations of the model sensitivity simulation.

Model Performance
The results of the WRF model simulation were evaluated by data from the Meteorological Information Comprehensive Analysis and Process System (MICAPS), China's meteorological observation network. The simulation results of the air quality were evaluated by monitoring the air pollutant concentrations collected from the China National Environmental Monitoring Center (CNEMC). The CNEMC website (http://106.37.208.228:8082/, accessed on 15 August 2020) releases hourly air quality information for 367 monitored cities in China. The CNEMC website was also used for studying the spatial and temporal variations of air pollutant concentrations. The meteorological simulation results and air quality concentrations from Run 1 and Run 3 were evaluated. This research used the correlation coefficient (RC), normalized mean bias (NMB), and normalized mean gross error (NME) for the statistical analysis, according to the US Environmental Protection Agency (EPA) model evaluation protocol [32] and previous studies [33]. The surface PM 2.5 concentrations, O 3 concentrations, NO 2 concentrations, and SO 2 concentrations were determined by comparing the simulated results with the CNEMC monitoring data. The temperature at 2 m, wind speed at 10 m, and relative humidity (RH) at 2 m were compared with the MICAPS monitoring data. The evaluation statistics of the meteorological and air quality results in the winter and summer of 2016 are summarized in Table 2.  Figure 2a displays the daily simulated and monitored meteorological variables. The results of the meteorological variables-namely, the temperature at 2 m (T2), wind speed at 10 m (WS10), and relative humidity at 2 m (RH2), were verified. The WRF simulation results adequately captured the variations in T2 for each month, with RC values of 0.96 and 0.83 in the winter and summer, respectively. The NMB and NME also indicated an excellent model performance for T2. The NMB value of T2 was −0.22% in the winter and 0.52% in the summer. In the winter and summer, the NME values of T2 were 0.38% and 0.56%, respectively. The WRF simulation results fully reflected the change in wind speed, with average bias values of 16.92% and 46.44% in the winter and summer simulation periods, respectively. The RC of wind speed was 0.84 and 0.66, and the NME was 28.01% and 48.08% in the winter and summer, respectively. As shown in Table 2, the results showed small NMB values for the RH2 simulation of 6.58% in the winter and −11.98% in the summer. The RH2 simulation results had good RC results of 0.88 and 0.77 in the  Figure 2) effectively represented the regional pollution processes. All the simulation parameters in our study followed the guidelines of the U.S. EPA. The CAMx results could be used for the further simulation analysis of sensitivity source apportionment.
showed small NMB values for the RH2 simulation of 6.58% in the winter and −11.98% the summer. The RH2 simulation results had good RC results of 0.88 and 0.77 in the win and summer, respectively. Table 2 also summarizes the statistics for the PM2.5, SO2, N and O3 concentrations in the two seasons in Beijing. Figure 2b presents the daily simula and observed results of the PM2.5, SO2, NO2, and O3 concentrations. In the winter, the NM values for the results of the PM2.5, SO2, NO2, and O3 concentrations were 42.43%, 72.71 2.41%, and 28.93%, respectively. In the summer, the NMB values for the results of PM2.5, SO2, NO2, and O3 concentrations were 10.71%, 38.75%, −10.69%, and −24.74%, spectively. The NME values of the PM2.5, SO2, NO2, and O3 concentrations obtained fr the simulation were 63.96%, 80.51%, 32.96%, and 32.68% in the winter, respectively. T NME values of the PM2.5, SO2, NO2, and O3 concentrations obtained from the simulat were 38.83%, 48.81%, 17.05%, and 36.18% in the summer, respectively. These biases m be due to the following reasons. (1) A previous study showed that the uncertainty of 0. in the MEIC at some sites would influence the accuracy of the simulation results. (2) T uncertainties related to the meteorological simulation could affect the results. (3) Ad tionally, the inherent characteristics of the model can lead to a significant deviation in days of severe pollution. Overall, in this work, the WRF model and CAMx predicted variables reasonably well. The surface PM2.5, SO2, NO2, and O3 concentrations were w reproduced by the model. The RC values of the PM2.5, SO2, and NO2 concentrations tween the simulated and observed values were between 0.52 and 0.73. The model simu tion results (shown in Figure 2) effectively represented the regional pollution process All the simulation parameters in our study followed the guidelines of the U.S. EPA. T CAMx results could be used for the further simulation analysis of sensitivity source portionment.
(a) WRF model performance   Figure 3 presents the temporal variations of the hourly PM2.5, SO2, and NO2 con trations in Beijing during January and July of 2016. In addition, the hourly wind sp wind direction, planetary boundary layer height (PBLH), and humidity are also prese at the bottom and top of Figure 3. We found that the hourly variations of the SO2 and concentrations agreed with the PM2.5 variations, indicating that they might have o nated from the same emission pollution sources. The PM2.5, SO2, and NO2 concentra were negatively correlated with the wind speed. The higher the wind speed was  concentrations agreed with the PM 2.5 variations, indicating that they might have originated from the same emission pollution sources. The PM 2.5 , SO 2 , and NO 2 concentrations were negatively correlated with the wind speed. The higher the wind speed was, the smaller the air pollutant concentration. The wind speed contributed to the dispersion of air pollutants out of the region. The air pollutant concentrations were negatively correlated with the PBLH. The higher the PBLH was, the better the dispersion of air pollution. Additionally, the air pollutant concentrations were positively correlated with the RH at 2 m. The higher the humidity was, the more serious the pollution. A higher humidity favored the PM 2.5 precursors (i.e., SO 2 , NOx, NH 3 , and VOCs) and tended to form PM 2.5 through chemical reactions [34]. The higher the humidity was, the lower the PBLH and wind speed, and the more serious the pollution. The transformation and physical diffusion of air pollutants were affected mainly by the wind speed, wind direction, RH2, and PBLH. The higher the wind speed was, the better the diffusion of the air pollutants, and different wind directions led to divergences in terms of the transport of different pollution concentrations. The wind directions in Figure 3 show that the high concentration of air pollutants persisted under the condition of southerly winds, which was conducive to the transport of air pollutants from the surrounding areas to Beijing. The emission levels of Baoding and Shijiazhuang, which are located to the southwest of Beijing, were relatively high. The continuous transport of pollutants could lead to an increase in the PM 2.5 level. Previous studies also showed a pollution transport channel in Southwestern Beijing [35]. In addition to the ground observation data, the satellite observation data were also used to present the spatial distribution of regional pollution. Data products on air pollutants retrieved by satellite were from the Ozone Monitoring Instrument (OMI) aboard the National Aeronautics and Space Administration (NASA) satellite Aura (https://disc.sci.gsfc.nasa.gov/, accessed on 15 July 2021) [36]. The downloaded dataset was in HDF5 format and drawn into visual graphics by NCL software. Spatial distributions based on satellite observation results of NO 2 concentrations near the ground in January and July 2016 were shown Figure 4. Compared with the winter, the pollution situation slowed down in the summer. The spatial distributions of NO 2 pollution were also consistent in the whole process. Heavy air pollution originated from the southern part of the BTH region and spread toward the northern parts, accumulating in front of the Yanshan Mountains and Taihang Mountains in the western and northern domains, respectively. The cities of Shijiazhuang, Baoding, and Beijing and the southern parts of the BTH region experienced heavy air pollution. through chemical reactions [34]. The higher the humidity was, the lower the PBLH and wind speed, and the more serious the pollution. The transformation and physical diffusion of air pollutants were affected mainly by the wind speed, wind direction, RH2, and PBLH. The higher the wind speed was, the better the diffusion of the air pollutants, and different wind directions led to divergences in terms of the transport of different pollution concentrations. The wind directions in Figure 3 show that the high concentration of air pollutants persisted under the condition of southerly winds, which was conducive to the transport of air pollutants from the surrounding areas to Beijing. The emission levels of Baoding and Shijiazhuang, which are located to the southwest of Beijing, were relatively high. The continuous transport of pollutants could lead to an increase in the PM2.5 level.

Spatial-Temporal Variations of Air Pollution and Meteorological Factors
Previous studies also showed a pollution transport channel in Southwestern Beijing [35].
In addition to the ground observation data, the satellite observation data were also used to present the spatial distribution of regional pollution. Data products on air pollutants retrieved by satellite were from the Ozone Monitoring Instrument (OMI) aboard the National Aeronautics and Space Administration (NASA) satellite Aura (https://disc.sci.gsfc.nasa.gov/, accessed on 15 July 2021) [36]. The downloaded dataset was in HDF5 format and drawn into visual graphics by NCL software. Spatial distributions based on satellite observation results of NO2 concentrations near the ground in January and July 2016 were shown Figure 4. Compared with the winter, the pollution situation slowed down in the summer. The spatial distributions of NO2 pollution were also consistent in the whole process. Heavy air pollution originated from the southern part of the BTH region and spread toward the northern parts, accumulating in front of the Yanshan Mountains and Taihang Mountains in the western and northern domains, respectively. The cities of Shijiazhuang, Baoding, and Beijing and the southern parts of the BTH region experienced heavy air pollution.   The masses of SO4 2− , NO3 − , and NH4 + aerosols accounted for the largest proportion in PM2.5, especially under pollution conditions. In this work, the various tendencies of SNA in PM2.5 from the simulation results for January and July were calculated, as shown in Figure 5. The SNA simulation results were also compared with monitoring data from a previous study conducted by our group [37]. In January, the mean observed SNA concentration was 30 μg m −3 , and the mean simulated SNA concentration was 21 μg m −3 . The observation results showed that SO4 2− , NO3 − , and NH4 + accounted for 34%, 36%, and 30% of the total SNA, respectively, while the simulation results showed that SO4 2− , NO3 − , and NH4 + accounted for 36%, 38%, and 26%, respectively. In the summer, the mean observed SNA concentration was 35 μg m −3 , and the mean simulated SNA concentration was 26 μg m −3 . The observation results showed that SO4 2− , NO3 − , and NH4 + accounted for 32%, 36%, and 32% of the total SNA, respectively, while the simulation results showed that SO4 2− , NO3 − , and NH4 + accounted for 36%, 37%, and 27%, respectively. Nitrate accounted for the most significant proportion of SNA in both the winter and summer.

Source Apportionment of PM2.5
In this work, the analyses of the total PM2.5 concentration; SO4 2− ; NO3 −; NH4 + ; and their precursor gases (SO2, NO2, and NH3) were performed by the CAMx-PSAT model. The sources were separated into five categories: industry, traffic, residential, power, and agriculture. Figure 6 shows the results of the source apportionment for the total PM2.5 concentration. The results suggest that residents became a primary source of pollution in the winter in Beijing (as shown in Figure 6a). The sharp increase in emissions from residential  The masses of SO 4 2− , NO 3 − , and NH 4 + aerosols accounted for the largest proportion in PM 2.5 , especially under pollution conditions. In this work, the various tendencies of SNA in PM 2.5 from the simulation results for January and July were calculated, as shown in Figure 5. The SNA simulation results were also compared with monitoring data from a previous study conducted by our group [37]. In January, the mean observed SNA concentration was 30 µg m −3 , and the mean simulated SNA concentration was 21 µg m −3 . The observation results showed that SO 4 2− , NO 3 − , and NH 4 + accounted for 34%, 36%, and 30% of the total SNA, respectively, while the simulation results showed that SO 4 2− , NO 3 − , and NH 4 + accounted for 36%, 38%, and 26%, respectively. In the summer, the mean observed SNA concentration was 35 µg m −3 , and the mean simulated SNA concentration was 26 µg m −3 . The observation results showed that SO 4 2− , NO 3 − , and NH 4 + accounted for 32%, 36%, and 32% of the total SNA, respectively, while the simulation results showed that SO 4 2− , NO 3 − , and NH 4 + accounted for 36%, 37%, and 27%, respectively. Nitrate accounted for the most significant proportion of SNA in both the winter and summer.  The masses of SO4 2− , NO3 − , and NH4 + aerosols accounted for the largest proportion in PM2.5, especially under pollution conditions. In this work, the various tendencies of SNA in PM2.5 from the simulation results for January and July were calculated, as shown in Figure 5. The SNA simulation results were also compared with monitoring data from a previous study conducted by our group [37]. In January, the mean observed SNA concentration was 30 μg m −3 , and the mean simulated SNA concentration was 21 μg m −3 . The observation results showed that SO4 2− , NO3 − , and NH4 + accounted for 34%, 36%, and 30% of the total SNA, respectively, while the simulation results showed that SO4 2− , NO3 − , and NH4 + accounted for 36%, 38%, and 26%, respectively. In the summer, the mean observed SNA concentration was 35 μg m −3 , and the mean simulated SNA concentration was 26 μg m −3 . The observation results showed that SO4 2− , NO3 − , and NH4 + accounted for 32%, 36%, and 32% of the total SNA, respectively, while the simulation results showed that SO4 2− , NO3 − , and NH4 + accounted for 36%, 37%, and 27%, respectively. Nitrate accounted for the most significant proportion of SNA in both the winter and summer.

Source Apportionment of PM2.5
In this work, the analyses of the total PM2.5 concentration; SO4 2− ; NO3 −; NH4 + ; and their precursor gases (SO2, NO2, and NH3) were performed by the CAMx-PSAT model. The sources were separated into five categories: industry, traffic, residential, power, and agriculture. Figure 6 shows the results of the source apportionment for the total PM2.5 concentration. The results suggest that residents became a primary source of pollution in the winter in Beijing (as shown in Figure 6a). The sharp increase in emissions from residential

Source Apportionment of PM 2.5
In this work, the analyses of the total PM 2.5 concentration; SO 4 2− ; NO 3 −; NH 4 + ; and their precursor gases (SO 2 , NO 2 , and NH 3 ) were performed by the CAMx-PSAT model. The sources were separated into five categories: industry, traffic, residential, power, and agriculture. Figure 6 shows the results of the source apportionment for the total PM 2.5 concentration. The results suggest that residents became a primary source of pollution in the winter in Beijing (as shown in Figure 6a). The sharp increase in emissions from residential sources in the winter was attributed to the large amount of coal consumed for heating purposes in cold weather. Residential sources contributed 50 µg m −3 to the PM 2.5 concentration in January, comprising 38 µg m −3 from local emissions and 12 µg m −3 from regional transport. Industry became the second-largest source, contributing 24 µg m −3 to the PM 2.5 concentration, comprising 16 µg m −3 from local emissions and 8 µg m −3 from regional transport. Traffic emissions contributed 13 µg m −3 to the PM 2.5 concentration and were mostly from local vehicle contributions (11 µg m −3 from local emissions). Agriculture and power emissions contributed 4.4 µg m −3 and 2.1 µg m −3 , respectively. Agricultural sources mainly emitted NH 3 , which is an important precursor to PM 2.5 formation. The replacement of clean energy such as natural gas and the application of pipe end control technology in power plants, reduced the emissions of PM 2.5 , SO 2 , and NOx to very small levels [5]. Especially in Beijing, all coal-fired power plants have been shut down. Therefore, the local power emissions contributed only less than 1% to the PM 2.5 concentration. The emissions from residents, the industry, traffic, agriculture, and power accounted for 54%, 25%, 14%, 5%, and 2% of PM 2.5 , respectively. The monthly transport ratio of PM 2.5 was 27%. The regional transport of residential and industrial emissions contributed the most, accounting for 13% and 9% of the PM 2.5 concentration, respectively. In this work, the results of source apportionment were compared with those of previous studies. Li et al. [38] utilized the CAMx model to make predictions that the industry and residential sectors played a leading role in the PM 2.5 concentration in Beijing from 2006 to 2013 in the winter, with residential sources contributing the most (~50%), followed by the industry (~30%). According to a PMF model, Yu et al. [21] found that coal combustion and industrial emissions were the primary sources of PM 2.5 in the winter of 2016 in Beijing, accounting for 50% and 27%, respectively. Figure 6b presents the results of the source apportionment for the total PM 2.5 concentration in the summer. The results show that industrial emissions were a major pollution source in the summer in Beijing. Industrial sources contributed 35 µg m −3 to the PM 2.5 concentration, comprising 13 µg m −3 from local emissions and 22 µg m −3 from regional transport. Traffic was the second-largest source, contributing 20 µg m −3 to the PM 2.5 concentration, comprising 16 µg m −3 from local emissions and 4 µg m −3 from regional transport. Residential emissions contributed 17 µg m −3 to the PM 2.5 concentration, including 10 µg m −3 from local emissions. The emissions from the industry, traffic, residents, agriculture, and power accounted for 42%, 24%, 20%, 10%, and 4% of PM 2.5 , respectively. The monthly transport ratio of PM 2.5 was 46%. The regional transport of industrial emissions contributed the most in the summer. , and NOx to very small levels [5]. Especially in Beijing, all coal-fired power plants have been shut down. Therefore, the local power emissions contributed only less than 1% to the PM2.5 concentration. The emissions from residents, the industry, traffic, agriculture, and power accounted for 54%, 25%, 14%, 5%, and 2% of PM2.5, respectively. The monthly transport ratio of PM2.5 was 27%. The regional transport of residential and industrial emissions contributed the most, accounting for 13% and 9% of the PM2.5 concentration, respectively. In this work, the results of source apportionment were compared with those of previous studies. Li et al. [38] utilized the CAMx model to make predictions that the industry and residential sectors played a leading role in the PM2.5 concentration in Beijing from 2006 to 2013 in the winter, with residential sources contributing the most (~50%), followed by the industry (~30%). According to a PMF model, Yu et al. [21] found that coal combustion and industrial emissions were the primary sources of PM2.5 in the winter of 2016 in Beijing, accounting for 50% and 27%, respectively. Figure 6b presents the results of the source apportionment for the total PM2.5 concentration in the summer. The results show that industrial emissions were a major pollution source in the summer in Beijing. Industrial sources contributed 35 μg m −3 to the PM2.5 concentration, comprising 13 μg m −3 from local emissions and 22 μg m −3 from regional transport. Traffic was the second-largest source, contributing 20 μg m −3 to the PM2.5 concentration, comprising 16 μg m −3 from local emissions and 4 μg m −3 from regional transport. Residential emissions contributed 17 μg m −3 to the PM2.5 concentration, including 10 μg m −3 from local emissions. The emissions from the industry, traffic, residents, agriculture, and power accounted for 42%, 24%, 20%, 10%, and 4% of PM2.5, respectively. The monthly transport ratio of PM2.5 was 46%. The regional transport of industrial emissions contributed the most in the summer.
(a)  Figure 7a,b shows the source apportionment results for SO4 2− and its precursor gas SO2. SO4 2− and SO2 were mainly contributed by industrial and residential emissions. In the winter, the emissions from residents, industry, traffic, agriculture, and power contributed 47%, 39%, 10%, 0%, and 4% to the SO4 2− concentration, respectively. Residential sources contributed 32% of the SO4 2− concentration from local emissions and 15% from regional transport. The industry contributed 17% of the SO4 2− concentration from local emissions and 22% from regional transport. In the summer, the emissions from the industry, power, residents, traffic, and agriculture contributed 66%, 13%, 15%, 6%, and 0% to the SO4 2− concentration, respectively. Industrial sources contributed 9% of the SO4 2− concentration from local emissions and 57% from regional transport. In the winter, the emissions from residents, industry, traffic, agriculture, and power contributed 42%, 41%, 15%, 0%, and 2% to the SO2 concentration, respectively. Residential sources contributed 35% of the SO2 concentration from local emissions and 7% from regional transport. In the summer, the SO2 concentration was mainly from industrial emissions, comprising 19% from local emissions and 36% from regional transport. The regional transport contribution ratios of SO4 2− and SO2 were very different, which had a large impact on the SO4 2− concentration. The reason lies in that, different meteorological conditions not only affect the diffusion of pollutants but also affect the formation process. The regional transport ratios of SO4 2− were 42% and 74% of the total concentrations in the winter and summer, respectively. In contrast, the regional transport ratios were 18% (82% by local emissions) and 45% (55% by local emissions) of the total SO2 concentration in the winter and summer, respectively. Compared with the contribution of local emissions, the contribution of regional transport was smaller.  Figure 7a,b shows the source apportionment results for SO 4 2− and its precursor gas SO 2 . SO 4 2− and SO 2 were mainly contributed by industrial and residential emissions. In the winter, the emissions from residents, industry, traffic, agriculture, and power contributed 47%, 39%, 10%, 0%, and 4% to the SO 4 2− concentration, respectively. Residential sources contributed 32% of the SO 4 2− concentration from local emissions and 15% from regional transport. The industry contributed 17% of the SO 4 2− concentration from local emissions and 22% from regional transport. In the summer, the emissions from the industry, power, residents, traffic, and agriculture contributed 66%, 13%, 15%, 6%, and 0% to the SO 4 2− concentration, respectively. Industrial sources contributed 9% of the SO 4 2− concentration from local emissions and 57% from regional transport. In the winter, the emissions from residents, industry, traffic, agriculture, and power contributed 42%, 41%, 15%, 0%, and 2% to the SO 2 concentration, respectively. Residential sources contributed 35% of the SO 2 concentration from local emissions and 7% from regional transport. In the summer, the SO 2 concentration was mainly from industrial emissions, comprising 19% from local emissions and 36% from regional transport. The regional transport contribution ratios of SO 4 2− and SO 2 were very different, which had a large impact on the SO 4 2− concentration. The reason lies in that, different meteorological conditions not only affect the diffusion of pollutants but also affect the formation process. The regional transport ratios of SO 4 2− were 42% and 74% of the total concentrations in the winter and summer, respectively. In contrast, the regional transport ratios were 18% (82% by local emissions) and 45% (55% by local emissions) of the total SO 2 concentration in the winter and summer, respectively. Compared with the contribution of local emissions, the contribution of regional transport was smaller. Remote Sens. 2021, 13, x FOR PEER REVIEW 12 of 19 The source apportionment results for NO3 − and its precursor gas NO2 are shown in Figure 7c,d. NO3 − and NO2 were mainly contributed by industrial and traffic emissions. In the winter, the emissions from the residents, industry, traffic, agriculture, and power contributed 15%, 40%, 30%, 0%, and 15% to the NO3 − concentration, respectively. Residential sources contributed 9% of the NO3 − concentration from local emissions and 6% from regional transport. The industry contributed 15% from local emissions and 25% from regional transport. Traffic contributed 15% to the NO3 − concentration from local emissions and 15% from regional transport. The emissions from residents, industry, traffic, agriculture, and power contributed 14%, 37%, 42%, 0%, and 7% to the NO2 concentration, respectively. Residential sources contributed 12% to the NO2 concentration from local emissions and 2% from regional transport. The industry contributed 30% of the NO2 concentration from local emissions and 7% from regional transport. Traffic contributed 37% of the NO2 concentration from local emissions and 5% from regional transport. In the summer, the emissions from the industry, traffic, power, and residents contributed 49%, 34%, 15%, and 2% to the NO3 − concentration, respectively. Industrial sources contributed 9% of the NO3 − concentration from local emissions and 40% from regional transport. Traffic contributed 11% to the NO3 -concentration from local emissions and 23% from regional transport. The emissions from traffic, industry, power, and residents contributed 53%, 35%, 9%, and 3% to the NO2 concentration, respectively. Traffic contributed 42% to the NO2 concentration from local emission sources and 11% from regional transport. Similar to the results for SO4 2− , regional transport also had a significant effect on the NO3 − concentration, accounting for 59% and 75% of the total concentration in the winter and summer, respectively. As The source apportionment results for NO 3 − and its precursor gas NO 2 are shown in Figure 7c,d. NO 3 − and NO 2 were mainly contributed by industrial and traffic emissions. In the winter, the emissions from the residents, industry, traffic, agriculture, and power contributed 15%, 40%, 30%, 0%, and 15% to the NO 3 − concentration, respectively. Residential sources contributed 9% of the NO 3 − concentration from local emissions and 6% from regional transport. The industry contributed 15% from local emissions and 25% from regional transport. Traffic contributed 15% to the NO 3 − concentration from local emissions and 15% from regional transport. The emissions from residents, industry, traffic, agriculture, and power contributed 14%, 37%, 42%, 0%, and 7% to the NO 2 concentration, respectively. Residential sources contributed 12% to the NO 2 concentration from local emissions and 2% from regional transport. The industry contributed 30% of the NO 2 concentration from local emissions and 7% from regional transport. Traffic contributed 37% of the NO 2 concentration from local emissions and 5% from regional transport. In the summer, the emissions from the industry, traffic, power, and residents contributed 49%, 34%, 15%, and 2% to the NO 3 − concentration, respectively. Industrial sources contributed 9% of the NO 3 − concentration from local emissions and 40% from regional transport. Traffic contributed 11% to the NO 3 concentration from local emissions and 23% from regional transport. The emissions from traffic, industry, power, and residents contributed 53%, 35%, 9%, and 3% to the NO 2 concentration, respectively. Traffic contributed 42% to the NO 2 concentration from local emission sources and 11% from regional transport. Similar to the results for SO 4 2− , regional transport also had a significant effect on the NO 3 − concentration, accounting for 59% and 75% of the total concentration in the winter and summer, Remote Sens. 2021, 13, 3457 13 of 19 respectively. As SO 4 2− and NO 3 − are both secondary pollutants, their formation processes are relatively complex, and they are mainly rich in small particles, which are conducive to being transported over long distances. In contrast, regional transport accounts for 16% of the total NO 2 concentration in the winter and 32% in the summer. Compared with the contribution from regional transport, local emissions have more significant contributions to the NO 2 concentration. Among the local emissions, traffic emissions contributed most to the total NO 2 concentration. NH 4 + and NH 3 were mainly contributed by agricultural emissions, which accounted for 94% of the total NH 4 + concentration and 95% of the NH 3 concentration. A previous study also found that agricultural sources contribute to most ammonia emissions [39]. Residential and traffic emissions accounted for less than 6% of the total NH 4 + concentration and less than 4% of the NH 3 concentration. Compared with the contribution from local emissions, regional transport contributed less. Local emissions contributed approximately 80% of the NH 4 + concentration in Beijing in the two seasons. Figure 8 shows the source contributions and regional transport concentrations of the winter and summer monthly average O 3 concentrations. Industry and transportation sources were the two major emission sectors contributing to O 3 pollution in Beijing (as shown in Figure 8a). In the winter, the emissions from the industry, traffic, residents, power, and agriculture contributed 36%, 43%, 17%, 4%, and 0% to the O 3 concentration, respectively. In the summer, the emissions from the industry, traffic, residents, power, and agriculture contributed 49%, 38%, 7%, 6%, and 0% to the O 3 concentration, respectively. These results are generally consistent with the source apportionment results for NO 2 shown in Figure 7. Wang et al. [26] similarly discovered that industrial and traffic emissions were the main sources of O 3 in the summer in Beijing, accounting for 38% and 22%, respectively. The monthly transport ratio of O 3 was 31% and 65% in the winter and summer, respectively. Regional transport contributed the most in the summer. In the winter, industrial sources contributed 15% of the O 3 concentration from local emissions and 21% from regional transport. Traffic contributed 37% to the O 3 concentration from local emissions and 6% from regional transport. In the summer, industrial sources contributed 13% of the O 3 concentration from local emissions and 36% from regional transport. Traffic contributed 16% to the O 3 concentration from local emissions and 22% from regional transport. Liu et al. [40] found that, during an O 3 pollution episode in Beijing, the local emissions contributed 14% to the O 3 concentration. It has been proven that high concentrations of O 3 pollution are derived not only from local emissions but also from regional transport and even superregional transport. Figure 8b shows the regional transport source apportionment results for the O 3 concentration. Zhangjiakou was found to be the central transport region for Beijing O 3 pollution in the winter, and the transport contribution ratio to PM 2.5 was 27%. In the summer, Langfang, Tianjin, Tangshan, and Baoding were the major transport regions, with transport contribution ratios to O 3 of 10%, 10%, 8%, and 8%, respectively. As Derwent et al. [41] reported, O 3 can be transported thousands of kilometers in the atmosphere and can persist for long periods of time in the troposphere. Although Shijiazhuang, Hengshui, Handan, and Xingtai are far from Beijing, their emission contributions to O 3 pollution in Beijing were prominent, accounting for 5%, 5%, 4%, and 3% in the summer, respectively. These results show that regional collaboration is an important and necessary prerequisite for controlling O 3

Regional Sensitivity Coefficient Results
To better understand how the topography and meteorology affect Beijing's air pollution, the regional sensitivity coefficients were examined based on the CAMx/PSAT simulation. The sensitivity coefficients of the air pollutants for different regions in the winter and summer were obtained. The same pollutant emissions were set in each simulation grid, as described in Section 2.2. The results show that local emissions were the major sensitive area of pollution formation in the winter. In the winter, the regional sensitivity coefficients of PM2.5, SO4 2− , and NO3 − were 57%, 50%, and 60%, respectively. In the summer, the regional sensitivity coefficients of PM2.5, SO4 2− , and NO3 − were 43%, 26%, and 50%, respectively. These results illustrate that, in pollution control, more attention should be paid to local emission sources first, followed by the coordinated control of regional sources. Compared with SO4 2− and NO3 − , the local region of Beijing was more sensitive to their precursor gases. The regional sensitivity coefficients of SO2 and NO2 were 64% and 82% in the winter, respectively, and 56% and 58% in the summer. These results further confirmed that the contribution of regional transport had a more significant impact on the secondary inorganic aerosol concentrations than on the concentrations of their precursor gases. To clearly understand the region's sensitivity in generating air pollution, the sensitivity coefficients of the surrounding cities are shown in Figure 9. The results are divided into four levels: strong sensitivity, general sensitivity, weaker sensitivity, and insensitivity. In the winter, the sensitivity coefficients of Zhangjiakou, Chengde, and Langfang were more significant than those of the other cities. Zhangjiakou, Chengde, and Langfang, the regions characterized by strong sensitivity, are very close to Beijing (northwest, northeast, and south of Beijing, respectively). According to the meteorological records, northwesterly, northerly, and northeasterly wind directions were frequently observed in Beijing in January 2016 (as shown in Figure 10). Zhangjiakou and Chengde were located upwind of Beijing. The cities of Baoding, Tianjin, and Tangshan were characterized by a general sensitivity, while the cities of Cangzhou, Qinhuangdao, and Hengshui were characterized by a weaker sensitivity. Finally, due to their large distances from Beijing, the cities of Shijiazhuang, Handan, and Xingtai were characterized as insensitive to the air pollution in Beijing. In the summer of 2016, the southwesterly and northeasterly wind directions frequently occurred in Beijing. In this case, Langfang, Tianjin, Tangshan, and Baoding were the regions with strong sensitivity. As Zhangjiakou is downwind of Beijing in the summer, this city was insensitive.

Regional Sensitivity Coefficient Results
To better understand how the topography and meteorology affect Beijing's air pollution, the regional sensitivity coefficients were examined based on the CAMx/PSAT simulation. The sensitivity coefficients of the air pollutants for different regions in the winter and summer were obtained. The same pollutant emissions were set in each simulation grid, as described in Section 2.2. The results show that local emissions were the major sensitive area of pollution formation in the winter. In the winter, the regional sensitivity coefficients of PM 2.5 , SO 4 2− , and NO 3 − were 57%, 50%, and 60%, respectively. In the summer, the regional sensitivity coefficients of PM 2.5 , SO 4 2− , and NO 3 − were 43%, 26%, and 50%, respectively. These results illustrate that, in pollution control, more attention should be paid to local emission sources first, followed by the coordinated control of regional sources. Compared with SO 4 2− and NO 3 − , the local region of Beijing was more sensitive to their precursor gases. The regional sensitivity coefficients of SO 2 and NO 2 were 64% and 82% in the winter, respectively, and 56% and 58% in the summer. These results further confirmed that the contribution of regional transport had a more significant impact on the secondary inorganic aerosol concentrations than on the concentrations of their precursor gases. To clearly understand the region's sensitivity in generating air pollution, the sensitivity coefficients of the surrounding cities are shown in Figure 9. The results are divided into four levels: strong sensitivity, general sensitivity, weaker sensitivity, and insensitivity. In the winter, the sensitivity coefficients of Zhangjiakou, Chengde, and Langfang were more significant than those of the other cities. Zhangjiakou, Chengde, and Langfang, the regions characterized by strong sensitivity, are very close to Beijing (northwest, northeast, and south of Beijing, respectively). According to the meteorological records, northwesterly, northerly, and northeasterly wind directions were frequently observed in Beijing in January 2016 (as shown in Figure 10). Zhangjiakou and Chengde were located upwind of Beijing. The cities of Baoding, Tianjin, and Tangshan were characterized by a general sensitivity, while the cities of Cangzhou, Qinhuangdao, and Hengshui were characterized by a weaker sensitivity. Finally, due to their large distances from Beijing, the cities of Shijiazhuang, Handan, and Xingtai were characterized as insensitive to the air pollution in Beijing. In the summer of 2016, the southwesterly and northeasterly wind directions frequently occurred in Beijing. In this case, Langfang, Tianjin, Tangshan, and Baoding were the regions with strong sensitivity. As Zhangjiakou is downwind of Beijing in the summer, this city was insensitive.  Figure 11 shows the high-resolution regional source apportionment results. It was found that Langfang, Baoding, Tangshan, and Zhangjiakou were the major transport regions for Beijing air pollution in the winter, and the transport contribution ratios to PM2.5 were 8%, 5%, 4%, and 4%, respectively. The contributions to the PM2.5 concentration from Tianjin, Cangzhou, Chengde, Qinhuangdao, Shijiazhuang, and Hengshui were 1% to approximately 2%. In the summer, Langfang, Tangshan, Tianjin, and Baoding were the major transport regions for Beijing's PM2.5 pollution, and the transport contribution ratios to PM2.5 were 14%, 10%, 8%, and 5%, respectively. Since there were few northwest winds in the summer of 2016, Zhangjiakou's emission contribution was close to 0%. Comparing the regional sensitivity coefficients shown in Figure 9, the long-distance transport of pollution was increased. For example, the contributions from the cities of Shijiazhuang and Hengshui were more prominent, and even emissions from Xingtai impacted the air pollution in Beijing. The PM2.5, SO2, and NOx emission inventories released by the MEIC are given in Table 3 [42]. It is clearly shown that Tangshan and Baoding had high pollutant emissions. These cities have always been greatly affected and challenged by coal burning and the iron manufacturing industry due to their rapid economic development. By contrast, the pollutant emissions from Zhangjiakou and Chengde were relatively small. Therefore, to make considerable progress in improving the air quality of Beijing, regional  Figure 11 shows the high-resolution regional source apportionment results. It was found that Langfang, Baoding, Tangshan, and Zhangjiakou were the major transport regions for Beijing air pollution in the winter, and the transport contribution ratios to PM2.5 were 8%, 5%, 4%, and 4%, respectively. The contributions to the PM2.5 concentration from Tianjin, Cangzhou, Chengde, Qinhuangdao, Shijiazhuang, and Hengshui were 1% to approximately 2%. In the summer, Langfang, Tangshan, Tianjin, and Baoding were the major transport regions for Beijing's PM2.5 pollution, and the transport contribution ratios to PM2.5 were 14%, 10%, 8%, and 5%, respectively. Since there were few northwest winds in the summer of 2016, Zhangjiakou's emission contribution was close to 0%. Comparing the regional sensitivity coefficients shown in Figure 9, the long-distance transport of pollution was increased. For example, the contributions from the cities of Shijiazhuang and Hengshui were more prominent, and even emissions from Xingtai impacted the air pollution in Beijing. The PM2.5, SO2, and NOx emission inventories released by the MEIC are given in Table 3 [42]. It is clearly shown that Tangshan and Baoding had high pollutant emissions. These cities have always been greatly affected and challenged by coal burning and the iron manufacturing industry due to their rapid economic development. By contrast, the pollutant emissions from Zhangjiakou and Chengde were relatively small. Therefore, to make considerable progress in improving the air quality of Beijing, regional  Figure 11 shows the high-resolution regional source apportionment results. It was found that Langfang, Baoding, Tangshan, and Zhangjiakou were the major transport regions for Beijing air pollution in the winter, and the transport contribution ratios to PM 2.5 were 8%, 5%, 4%, and 4%, respectively. The contributions to the PM 2.5 concentration from Tianjin, Cangzhou, Chengde, Qinhuangdao, Shijiazhuang, and Hengshui were 1% to approximately 2%. In the summer, Langfang, Tangshan, Tianjin, and Baoding were the major transport regions for Beijing's PM 2.5 pollution, and the transport contribution ratios to PM 2.5 were 14%, 10%, 8%, and 5%, respectively. Since there were few northwest winds in the summer of 2016, Zhangjiakou's emission contribution was close to 0%. Comparing the regional sensitivity coefficients shown in Figure 9, the long-distance transport of pollution was increased. For example, the contributions from the cities of Shijiazhuang and Hengshui were more prominent, and even emissions from Xingtai impacted the air pollution in Beijing. The PM 2.5 , SO 2 , and NOx emission inventories released by the MEIC are given in Table 3 [42]. It is clearly shown that Tangshan and Baoding had high pollutant emissions. These cities have always been greatly affected and challenged by coal burning and the iron manufacturing industry due to their rapid economic development. By contrast, the pollutant emissions from Zhangjiakou and Chengde were relatively small. Therefore, to make considerable progress in improving the air quality of Beijing, regional Nevertheless, some uncertainties and limitations might bring the effect on the research results. Although CAMx-PSAT has been developed to quantify the source of air pollution and evaluate their contributions, which is affected by the uncertainties in terms of the meteorology simulations, parameterizations, and formulations used in the model. As for the MEIC, it is a bottom-up emission inventory model gradually developed for China. The limitations of research on the discharge coefficient, activity rates, and emission reductions measures would lead to uncertainties in the judgment of emission tendencies [43]. New chemical mechanism and more accurate emission inventory are needed to be the focus of future research.

Conclusions
The CAMx-PSAT model was employed in this work for the high-spatial resolution source apportionment of SNA in PM2.5, their corresponding precursor gases, and O3 in the winter and summer over Beijing. Both the contributions of different source categories of  Nevertheless, some uncertainties and limitations might bring the effect on the research results. Although CAMx-PSAT has been developed to quantify the source of air pollution and evaluate their contributions, which is affected by the uncertainties in terms of the meteorology simulations, parameterizations, and formulations used in the model. As for the MEIC, it is a bottom-up emission inventory model gradually developed for China. The limitations of research on the discharge coefficient, activity rates, and emission reductions measures would lead to uncertainties in the judgment of emission tendencies [43]. New chemical mechanism and more accurate emission inventory are needed to be the focus of future research.