A paradox for air pollution controlling in China revealed by “APEC Blue” and “Parade Blue”

A series of strict emission control measures were implemented in Beijing and surrounding regions to ensure good air quality during the 2014 Asia-Pacific Economic Cooperation (APEC) summit and 2015 Grand Military Parade (Parade), which led to blue sky days during these two events commonly referred to as “APEC Blue” and “Parade Blue”. Here we calculated Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) and Ozone Monitoring Instrument (OMI) NO2 and HCHO results based on well known DOAS trace gas fitting algorithm and WRF-Chem model (with measured climatology parameter and newest emission inventor) simulated trace gases profiles. We found the NO2 columns abruptly decreased both Parade (43%) and APEC (21%) compared with the periods before these two events. The back-trajectory cluster analysis and the potential source contribution function (PSCF) proved regional transport from southern peripheral cities plays a key role in pollutants observed at Beijing. The diminishing transport contribution from southern air mass during Parade manifests the real effect of emission control measures on NO2 pollution. Based on the ratios of HCHO over NO2 we found there were not only limited the NO2 pollutant but also suppress the O3 contaminant during Parade, while O3 increased during the APEC.


Results
In order to evaluate the impact of emission control policy on air quality during Parade and APEC, three episodes are separately defined in this study: 1 st episodes is defined as the period of Parade (from August 20 th to September 3 rd 2015), in which the strict air quality policies were implemented at a regional scale; 2 nd and 3 rd episodes are defined as the "pre-Parade" from August 5 th to 19 th and the "post-Parade" from September 4 th to 21 st . Similarly, the three episodes of APEC are respectively defined as pre-APEC period (from October 24 th to November 2 nd ), APEC period (from November 3 rd to 12 th ) and post-APEC period (from November 13 th to 21 st ).
We use the retrieved NO 2 VCD from MAX-DOAS measurements to analyze the temporal variation in urban Beijing (40°N, 116°22ʹ 48″ E) for these two events (Fig. S1). In order to compare and validate with the OMI satellite data, the MAX-DOAS data are temporally averaged for the OMI satellite overpass time. Firstly, the NASA's OMI tropospheric NO 2 data products are employed, which are widely used in previous studies 5,19 . However, in view of the tropospheric air mass factors (AMF) of NASA's OMI tropospheric NO 2 products are calculated based on the monthly mean NO 2 profile shapes derived from the Global Modeling Initiative (GMI) chemistry transport model multiannual simulation 19,23 , which might not be fully representative to the situation in China, especially in Beijing. So we also utilize the USTC's NO 2 products which account for the local atmospheric conditions and use the WRF-Chem model with measured climatology parameter 24 and newest emission inventory 25 to simulate trace gas profiles. The time series and inter-comparison of three independent tropospheric NO 2 products are shown in Figs 1 and 2.
Generally, these three independent datasets showed a good agreement of during both Parade and APEC. A common trait, the NO 2 VCDs had been abruptly decreased, was found during the Parade and APEC periods. It should be noted that the USTC OMI NO 2 results present better correlation (r = 0.79 of Parade and r = 0.82 of APEC) than NASA OMI's (r = 0.71 of Parade and r = 0.80 of APEC) with ground-based MAX-DOAS measurement, due to the corresponding local trace gases profiles used in AMF calculation, and the OMI NO 2 SCD between USTC's and NASA's are quite close. Applying inappropriately trace gas profile to calculate AMF can cause up to 40% bias in previous study 26 .
However, from Table 1, we can find the MAX-DOAS measurements are systematically higher than the NASA and USTC OMI results. This systematic underestimation for OMI observations, which was also found in previous studies 18,19 , might be due to two main reasons. First, the grid cells' values of satellite observations might be not only contained strong emission sources areas for the research site, but also average over neighboring cleaner areas 18 . Second, the different profile of NO 2 and aerosols could cause a systematic underestimation of the real tropospheric NO 2 VCDs in the satellite retrievals 18,19 . Previous studies show that the OMI NO 2 columns were increased by 15-20% 27,28 , even up to 40% when a better estimated NO 2 profile was applied in the AMF calculation for the NO 2 column retrieval. Thus consistently in Figs 1 and 2, we found the difference of NO 2 column between USTC OMI results and MAXDOAS results are systematically 17% smaller than that of NASA's, due to adopting newest emission inventory 25 and measured local atmospheric conditions in AMF simulation.
As shown in Table 1, the averaged NO 2 VCDs measured by MAX-DOAS instruments were respectively 22.07, 12.50 and 25.00 (10 15 molec cm −2 ) during the periods around Parade. Compared with the pre-Parade, the averaged NO 2 VCDs during Parade exhibits a distinct reduction ratio about 43% ((pre-Parade -Parade)/pre-Parade). While the air quality was observed to rapidly plummet from the perspective of NO 2 with the end of Parade. The averaged NO 2 VCDs during post-Parade was twice than compared to Parade ((post-Parade -Parade)/Parade). In the same way, during the periods around APEC, the mean NO 2 VCDs measured by MAX-DOAS were respectively 39.01, 30.72 and 48.76 (10 15 molec cm −2 ). 21% ((pre-APEC -APEC)/pre-APEC) and 59% ((post-APEC -AEPC)/APEC) were separately represented the decrease of APEC (compared with pre-APEC) and the growth of post-APEC (compared with APEC). The tremendous reductions during both two events (Parade and APEC) demonstrate the Chinese stringent control policy had been worked effectively from the perspective of NO 2 for Beijing local.
However, the NO 2 results of MAX-DOAS could only reveal the variations around the site of Beijing urban areas during these two events. To more explicitly explore the distributions and variations in tropospheric NO 2 we propose to use the USTC's NO 2 products retrieved from the OMI to analyze in this study, which showes a quite good correlation with MAX-DOAS results. Figure 3 presents the NO 2 distributions in Beijing and its surrounding regions (including Hebei, Shanxi, Shandong Province and Tianjin Municipality) during three periods Scientific RepoRts | 6:34408 | DOI: 10.1038/srep34408 around Parade. Mean values were calculated in all of the three periods, i.e. pre-Parade, Parade and post-Parade, respectively. The spatial distributions of tropospheric NO 2 VCDs were substantially similar but still different in diverse time frames. High NO 2 VCD appeared on similar areas, including Beijing urban areas, Tianjin, southern Hebei, major areas of Shandong, and parts of Shanxi. And these areas which referred above are heavily industrialized and thus suffered by more intense anthropogenic emissions 5 . Compared with pre-Parade, we can easily find a sharp decrease over the urban Beijing, southern Hebei (including Tangshan city), Shandong and Shanxi during the Parade period. This phenomenon indicates that the strict emission control measures implemented in Beijing (nearly 2,000 industrial firms, including petrochemical and cement plants, suspended or cut production in Beijing local) and surrounding Provinces during Parade (based on Chinese media reports, http:// en.people.cn/n/2015/0907/c98649-8946581.html), which were even more stringent than APEC periods, did significantly decrease the NO 2 concentrations and improve the air quality of urban Beijing and its surrounding regions. Compared with Parade period, most of the areas, including Hebei, Shanxi, Shandong, especially for urban Beijing, had a rapid increase of NO 2 VCDs during the post-Parade without strict control measures (Fig. 3).
In contrast, the NO 2 VCDs had an apparent reduction only in Beijing during APEC period, and NO 2 VCDs were not having a obvious reduction over the other surrounding regions of Beijing compared with pre-and post-APEC 5 . The results also corroborated that strict emission control measures during APEC periods may be not strict enough or not work well in Beijing surrounding regions compared with Parade periods. While the OMI data are spatially averaged over the grid cells within 30km of the ground location around the Beijing urban area. The transparent blue square area represents the Parade period. And the grey shade square areas separately represent "Cycle 1" (Aug. 14 th to 21 st ), "Cycle 2" (Sep. 6 th to 9 th ) and "Cycle 3" (Sep. 12 rd to 18 th ) of MAX-DOAS results.
Scientific RepoRts | 6:34408 | DOI: 10.1038/srep34408 Discussion Role of regional source and meteorological impacts. Figure 1 show the time series of daily mean tropospheric NO 2 VCDs about Parade. The result showed an intriguing phenomenon that the NO 2 columns varied with time like a cycle. In the continual circulation, NO 2 columns could be increased abruptly in someday and sustain for one day or several days, and dropped sharply for a few days. We named this phenomenon as "fluctuation  In order to determine the formation of "fluctuation effect", we pick out three typical cycles named "Cycle 1" (From August 14 th to 21 st ), "Cycle 2"(From September 6 th to 9 th ) and "Cycle 3" (From September 12 rd to 18 th ) respectively ( Fig. 1). To find the trigger of peak and valley values, we analyzed the MAX-DOAS data during "Cycle 1" period by the cluster analysis of the 24-h air mass back trajectories (AMBTs) starting at 500 m at Beijing urban area (Fig. S2). Combined Fig. 1 with Fig. S2, we could find a close relationship between the directions of AMBTs and the "fluctuation effect". The directions of AMBTs changed from north to south with the increasing of NO 2 values (Fig. S2). Analogously, the NO 2 values would be decreased when the directions changed from south to north. For instance, on August 14 th and 15 th , the NO 2 columns (11.55 and 13.37 10 15 molec cm −2 , respectively) were in the bottom position of "Cycle 1", and the all AMBTs' direction were from the north or north-west by chance. The NO 2 columns (20.12 and 19.82 10 15 molec cm −2 , separately) on August 16 th and 17 th were increased compared with two days before, and the AMBTs' paths were also transited from north to south. Just like 70.8% AMBTs were from southwestward on August 16 th , and the 83.3% AMBTs' routes were from south or southwest on August 17 th . As we expected, 95.8% AMBTs were from south or southwest at the peak (46.22 10 15 molec cm −2 ) on August 18 th . As well, the process from the peak value down to the valley value experienced a period of transition (August 19 th ) which had a change from 4.2% west air mass increased to the 37.4% northwest air mass comparing with the "Peak" (August 18 th ). The corresponding NO 2 columns (30.33 10 15 molec cm −2 ) were also lower than peak value. The valley values (August 20 th and 21 st ) were appearing with the air mass directions all changed to the north. This result indicates the contamination of Beijing urban area was directly affected by the pollution transmission of Beijing south areas.
Similar results were also found in "Cycle 2" and "Cycle 3" (Figs S3 and S4). We could see the valley concentrations' air mass came from the north or northeast, and the peak values' air mass came from south or southwest. There's a special point that the 62.5% air mass were surrounded by the south outsikts and peripheral cities of Beijing on September 7 th . We can find the air mass were mainly from nearby southern Beijing during this day and their sojourn time of southern Beijing were more than the other days. It also meaned the Beijing area was not only influenced by the relatively distant southern cities, but also specially affected by southern neighboring areas of Beijing (just like the adjacency between Beijing and Hebei Province). Figure 2 plots the time series of daily averaged tropospheric NO 2 VCDs during the APEC periods. In contrast with Parade period, two peak values (Nov. 4 th and 7 th ) were emerged during APEC period which was implementing the strict emission control measures period. We analyze the corresponding MAX-DOAS data by the cluster analysis (Fig. S5). There're separately 79.2% and 83.3% AMBTs from south (Baoding, Langfang and Tianjin) on Nov. 4 th and 7 th . From the previous study 5 , we could find the regions of Tianjin and Hebei were still at a high level of NO 2 during APEC period, which could have an influence on transporting to Beijing.
As suggested, meteorological conditions and regional atmospheric transport should play a key role in affecting the column NO 2 levels. The NO 2 peak values of "fluctuation effect" were mainly influenced by the polluted southern air mass, and the reason which led to the NO 2 valley values is the clean effect of northern air mass. It also means the NO 2 pollution of Beijing was directly affected by the southern surrounding areas' atmospheric transport during the period without strict emission control measures.
Identified the potential sources by PSCF. To further demonstrate the regional impact, we have analyzed the averaged NO 2 VCDs measured by MAX-DOAS through the PSCF analysis in this study.The distributions of PSCF values in Beijing urban before, during and after the Parade are respectively shown in Fig. S6. For the pre-Parade and post-Parade period, due to the NO 2 at high levels during these two periods the high PSCF values' areas may represent the quintessentially potential emission sources for NO 2 in Beijing urban. Both of these two periods show that cells with high PSCF values appeared mainly in the Beijing south suburbs and the southwestern surrounding cities around Beijing. The southern cells PSCF values were apparently higher than northern cells, Scientific RepoRts | 6:34408 | DOI: 10.1038/srep34408 which indicated the potential source areas maybe contain Baoding and Langfang and other southwestward cities around Beijing. That also proved the air pollutant transport from the locations nearby Beijing rather than farther places. For the Parade period, even though the NO 2 VCDs were kept in a relatively low level during this period, we can also find there were some potential sources in Langfang and Tianjin according to Fig. S6. Synthesizes the above analysis, there is a common trait which indicates the NO 2 pollution of Beijing urban during three periods around Parade were mainly affected by southern outskirts of Beijing, southwestern and southeastern surrounding cities of Beijing (e.g. Liulihe for Beijing and Baoding, Langfang for Hebei Province, as well as Tianjin).
Moreover, we referred to many series of air pollutants data (http://pm25.in/) for determining the air conditions about southern cities of Beijing, including AQI, PM 2.5 , PM 10 , NO 2 , O 3 , SO 2 and so on. There are more than fifteen hundred sites on a nationwide scale in this website, and we selected 100 sites of data by sequence to present the pollutant concentrations during pre-Parade and post-Parade in this study (Tables S1 and S2). We found the air pollution (e.g. AQI, PM 2.5 , NO 2 and O 3 ) of Baoding and Langfang sites were at a quite high level compared to other sites. Especially for Baoding, the concentrations of AQI, PM 2.5 and PM 10  From the above, we can confirm that the NO 2 pollution in Beijing was mainly affected by the regional transport from the southern surrounding cities around Beijing (like Baoding, Langfang and Tianjin) which are the most potential NO 2 sources areas for Beijing. The ambient contamination "fluctuation effect" for NO 2 in Beijing was triggered by air mass directions. During the Parade, the controlling successfully decreased the impact of such regional source impact on NO 2 in Beijing.
We also analyzed the averaged NO 2 VCDs measured by MAX-DOAS through the PSCF analysis method during the periods around APEC (Fig. S7). It also revealed that high PSCF values with cells during pre-APEC were appeared on the southeastern suburbs of Beijing and southeastern surrounding cities of Beijing (just like Langfang and Tianjin). During the APEC period, the southeastern suburbs of Beijing and southeastern surrounding cities of Beijing (Langfang and Tianjin) were mainly source areas. Southern suburbs of Beijing, Baoding and Langfang were the primarily potential source areas during post-APEC. In account of the NO 2 columns were slightly different between pre-APEC period and post-APEC period, we also combined the MAX-DOAS data of pre-APEC and post-APEC period and took a PSCF analysis (Fig. S7). Which could also reveal the consequence similar to above. In general, we could confirm the NO 2 pollution in Beijing urban during the periods around APEC were mainly affected by the emission from southern and southeastern suburbs of Beijing, southern and southeastern cities of Beijing, especially from Tianjin, Langfang and Baoding.
Controlling impact on NO 2 . According to the foregoing analysis, the NO 2 columns would be seriously influenced by air mass in Beijing. Previous studies showed meteorological conditions were found to play a significant role in reducing pollution levels during the Olympic Games 29-32 . These results reflect the uncertainties in assessing the impact of emission controls on air pollutants over a certain area. For avoiding the influence of the meteorological factor, we had performed the cluster analysis of the 24-h air mass back trajectories for the periods around Parade which from August 5 th to September 21 st . And for better contrast, we divide these AMBTs into two types during three periods around Parade (Fig. 4, Table S3): 1. Southern air mass, it means that the AMBTs were almost from south, southeast and southwest (> 50%) for a whole day. 2. Northern air mass, which means more than 50% AMBTs were from north, northeast and northwest for a full day. From the Table S3 and Fig. 5, we compared the southern air mass with northern air mass, and could easily find the mean tropospheric NO 2 VCDs for southern air mass during both pre-Parade and post-Parade period (25.68 and 32.07 10 15 molec cm −2 ) were 2~3 times higher than northern air mass during these two periods (16.65 and 14.90 10 15 molec cm −2 ). These huge differences directly showed the importance of meteorological factors and southern regional atmospheric transport. Baoding, Langfang and other southern cities from Beijing were suffered more serious air pollution than Beijing (Table S1), and much more air pollution would be brought in Beijing local when the air mass passed through these areas. Compared with southern regions around Beijing, the northern areas were kept in a cleaner conditions. However, with a series of emission control policy implemented in Beijing and its southern surrounding cities during Parade period, the impact of air mass had been vanished. The mean tropospheric NO 2 VCDs for southern air mass during Parade (11.80 10 15 molec cm −2 ) was even a little lower than northern air mass (12.69 10 15 molec cm −2 ), and both of them stayed at a relatively low level. This phenomenon reflected the effectiveness of strict emission control measures. For southern air mass, compared with averaged NO 2 VCDs during pre-Parade (25.68 10 15 molec cm −2 ), the averaged NO 2 VCDs during Parade (11.80 10 15 molec cm −2 ) had been decreased about 54%. The averaged NO 2 VCDs during post-Parade (32.07 10 15 molec cm −2 ) was almost 3 times higher than during Parade period. These results directly proved the southern peripheral cities around Beijing plays a key role in pollutants transport for Beijing. This abruptly diminishing during Parade manifests the real effect of emission control measures and also shows the importance of emission controls which were implemented in cities to the south of Beijing. For northern air mass, the averaged NO 2 VCDs were respectively 16.65 and 14.90 10 15 molec cm −2 during pre-Parade and post-Parade period, both of which are a little more than during Parade (12.69 10 15 molec cm −2 ). And all of them stayed at a relatively low level for NO 2 pollutants, that means the regions surrounding north of Beijing were much cleaner than south. We haven't found a significant effect of transmission for northern air mass, and there were fewer emission controls implemented for the north and northwest of Beijing compared with southern regions. The NO 2 VCDs for the northern air mass had a little decrease during Parade period compared with pre-Parade period, which represents the probable effect of Beijing owns' emission Scientific RepoRts | 6:34408 | DOI: 10.1038/srep34408 measures due to eliminating the influence of the southern transmissions. However, according to the results of above, limiting the cities to the south of Beijing may be more important than Beijing local.

Change in O 3 versus NO 2 during and around the period of APEC and Parade.
In order to explore the chemical sensitivity of PO 3 during Parade periods and APEC periods in Beijing, we used the USTC OMI tropospheric NO 2 products and USTC recalibrated OMI total HCHO products to calculate the Ratio over Beijing and surrounding areas during these two events. And we also combined the Ratio and the corresponding variations of NO 2 and HCHO to analyze how O 3 varied as the function of Ratio.
From Fig. 6A, we can easily find the Ratio is between 1 and 2 (1.76) at Beijing urban site within 1° × 1° mean value (40°N, 116°22ʹ 48″ E) during pre-Parade, which indicates this area was at a mixed VOC-NO x -limited regime. With a series of strict emission control measures during Parade, the Ratio had changed from 1.65 to 3.71. It means the PO 3 conditions had also changed from mixed VOC-NO x -limited to a predominantly NO x -limited condition due to the sharp drop of NO 2 during Parade. It is observed that the tropospheric averaged O 3 VCDs had a decline (from 14.84 10 17 molec cm −2 during pre-Parade to 13.65 × 10 17 molec cm −2 during Parade) from the pre-Parade period transited to the Parade period, which was maybe caused by the rapid decrease of local NO 2 because the PO 3 was exactly stayed at NO x -limited chemistry. The southern air mass means that the AMBTs were almost from south, southeast and southwest (> 50%) for a whole day. Likewise, northern air mass means more than 50% AMBTs were from north, northeast and northwest. The black lines mean the AMBTs during the certain period for every hour (which have been signed in title). And the color lines present the 5 categories directions of AMBTs. Base map is from TrajStat 1.2.2 software (http://www.meteothinker.com).
Scientific RepoRts | 6:34408 | DOI: 10.1038/srep34408 After the strict emission control measures were lifted, the NO 2 returned to relatively high values as pre-Parade and the Ratio was also diminished (Ratio = 0.90, < 1), which indicated the PO 3 was turned into a VOC-limited condition during post-Parade. At this point, the reduction of mean HCHO VCDs, which was the normal seasonal decrease of HCHO, maybe was the primary reason caused by the O 3 kept decreasing correspondingly. Figure 7A shows the spatial variation of the Ratios over Beijing and surrounding areas during three periods around Parade. During pre-Parade, we found that most of Beijing urban areas were stayed at mixed VOC-NO x -limited (1< Ratio < 2) and part of eastern, southwestern and southern Beijing were stayed at both VOC-limited and mixed VOC-NO x -limited. And the other regions were presented to the NO x -limited. When the time was entering the Parade period, PO 3 was shifted to a predominantly NO x -limited regime (Ratio > 2) including Beijing urban and a major part of southern and southwestern areas. These results reveal that Chinese control policies had been worked effectively from the perspective of NO 2 , however, HCHO had not an apparent decrease over these areas. And the ratio had been turned into VOC-limited regime over Beijing and surrounding regions because NO 2 returned to relatively high values. The result which indicated PO 3 is VOC-limited in urban Beijing is also consistent with previous studies 33,34 .
Above of all, the emission control measures during Parade were not only work effectively for regional NO 2 pollution control patterns but also effective for O 3 controlling.
Similarly, Fig. 6B shows the Ratio and variations of O 3 , NO 2 and HCHO averaged VCDs at Beijing urban site within 1° × 1° mean value (40°N, 116°22ʹ 48″ E) during three periods around APEC. It is obviously that both pre-APEC and APEC were stayed at a mixed VOC-NOx-limited condition. From the pre-APEC to APEC, both NO 2 and HCHO had a certain reduction which should lead the O 3 diminishing. Conversely, the mean O 3 VCDs were increased slightly instead of decline. Interestingly, similar results were also occurred to the previous measurements which were probably caused by the decreasing NO-titration of ozone (i.e. NO+ O 3 → NO 2 + O 2 ) and regional transport of photochemically aged air 20,35 . And the similar results (the mean O 3 VCDs were increased slightly) were also presented to post-APEC, while the HCHO was decreased during VOC-limited condition. This result indicates that emission controls in this case maybe not strict enough or worked well to lessen the levels of ozone 20 .
We can find the spatial variation of the Ratios over Beijing and surrounding areas during three periods around APEC from Fig. 7B. Compared with pre-APEC, the Ratio was changed from around 0~1.5 (VOC-limited and mixed VOC-NO x -limited) to around 1~2.5 (mixed VOC-NO x -limited and NO x -limited) in Beijing areas during APEC due to the rapid NO 2 decrease in this period. However, in other regions especially for Beijing southern areas the Ratio was presented to a decreasing tendency, which indirectly reflects the NO 2 VCDs were even increased in the areas neighboring Beijing. It also suggests that the emission control measures were not effective to Beijing surrounding regions during APEC period. After the emission control measures lifted, the Ratio had a dramatically decrease overall due to the diminishing of HCHO and NO 2 growing broadly.
In summary, even though both Parade and APEC period had a series of strict emission control policies, the control effect of Parade was far more than APEC. The emission control measures of Parade were not only limited the NO 2 pollutant including Beijing and from outside adjacent areas but also suppress the O 3 contaminant. However, during the APEC the emission control measures had not been worked so effective for surrounding areas. These results and the additional information from Ratio may help government formulate some appropriate pollution control strategies. MAX-DOAS is a passive DOAS approach based on measurements of scattered sunlight at different elevation angles and zenith towards the horizon, and it could retrieve the NO 2 column densities by the DOAS algorithm 11 . The MAX-DOAS instrument at this site was collected by sequential measurements which were made at 8 different elevation angles (3°, 5°, 8°, 10°, 12°, 20°, 30° and 90°) of scattered sunlight. The Fraunhofer reference (FRS) was used to remove the solar Fraunhofer structure in the scattered sunlight, which was usually selected at the 90° elevation angle during noon of a clear day 11 . The QDOAS software 36 (http://uv-vis.aeronomie.be/software/QDOAS/) with NO 2 retrieval settings of MAX Plank institute for Chemistry (MPIC) (http://joseba.mpch-mainz.mpg.de/ mad_analysis.htm) was applied to analyze the spectra between 338 and370 nm. Before the spectral analysis, the effects of electronic offset and dark current are also removed by spectra measured at the same condition.
Finally, differential slant column density (DSCD) were obtained from the QDOAS software outputs, which means the slant column density (SCD) between the measured spectrum and the FRS. The tropospheric DSCDα can be expressed as (α : denotes the elevation angle): The SCD is also influenced by the length of light path and the observation geometry, thus it needs to be converted to the vertical column density (VCD) which is not affect by the observation geometry and light path and can be used to compare different measurements. The VCD is calculated with the air mass factor (AMF): According to the above equation, the SCDs at the angles of 90° and α can be described as: By substituting equalities (3) and (4) into equality (1), we can acquire 18 : Where the DAMF is the differential atmospheric air mass factor: In this study, thetroposphericAMFs were calculated using the Vector Linearized Discrete Ordinate Radiative Transfer (VLIDORT) model. To account for local atmospheric conditions, WRF-Chem model with measured climatology parameter 24   In this study, both NASA's and USTC's OMI tropospheric trace gases products are used. For NO 2 , the NASA's OMI tropospheric NO 2 products (http://disc.sci.gsfc.nasa.gov/Aura/), which use monthly mean NO 2 profile from Global Modeling Initiative (GMI) chemistry transport model 23 for VCD column conversion. USTC tropospheric NO 2 SCDs, which are retrieved by the DOAS 11 trace gas fitting algorithm using a nonlinear least-squares (NLLS) inversion technique from the OMI spectra. Consistent with MAX-DOAS measurements in study, to account for local atmospheric conditions, WRF-Chem model with measured climatology parameter and newest emission inventory 25 is also used to simulate trace gas profiles for VCDs products conversion from former calculated SCDs. For OMI HCHOVCDs, we recalculated them from the SCDs of NASA HCHO products 37 based on WRF-Chem model simulated profiles to account for local atmospheric conditions. The OMI tropospheric O 3 products 38 are download from (http://disc.sci.gsfc.nasa.gov/Aura/).

WRF-Chem model and Emission
Inventory. The numerical model adopted in this study is WRF-Chem version 3.7, which is an online-coupled chemical transport model considering multiple physical and chemical processes, including emissions and deposition of pollutants, advection and diffusion, gaseous and aqueous chemical transformation, aerosol chemistry and dynamics 39 . It is capable of simulating atmospheric chemistry on a regional scale and has been successfully applied in several of our previous studies 40,41 . In this work, the model domain covered East China and its surrounding area, centering at 35.0°N, 110.0°E with a 20 × 20 km grid resolution, as demonstrated in Fig. S8. There are 27 vertical layers from the ground level to the top pressure of 50 hPa. The 6 hourly Final operational global analysis (FNL) data with a 1° × 1° spatial resolution produced by the National Centers for Environmental Prediction (NCEP) was used as initial and boundary conditions of meteorological fields. In addition, NCEP's ADP global upper-air observations (NCAR archive ds351.0) were assimilated every 6 hours to enhance the meteorology reproduction. Key physical parameterization options for the modelling were the Noah land surface scheme to describe the land-atmosphere interactions 42 , the Lin microphysics scheme 43 with the Grell cumulus parameterization to reproduce the cloud and precipitation processes 24 , the YSU boundary layer scheme, and the RRTMG short-and long-wave radiation scheme 44 .
For the numerical representation of atmospheric chemistry, we used the CBMZ (Carbon-Bond Mechanism version Z) photochemical mechanism combined with MOSAIC (Model for Simulating Aerosol Interactions and Chemistry) aerosol model 45,46 . Both natural and anthropogenic emissions were included for the regional WRF-Chem modelling in the present work. Typical anthropogenic emissions were obtained from the Multi-resolution Emission Inventory for China (MEIC) database 25 , in which emissions sources were classified into five main sectors: power plants, residential combustion, industrial processes, on-road mobile sources, and agricultural activities. This database covered most of anthropogenic pollutants, such as SO 2 , NO x , CO, volatile organic compounds (VOCs), NH 3 , PM, BC, and OC. The biogenic VOC and NO emissions were calculated online by using the Model of Emissions of Gases and Aerosols from Nature (MEGAN) embedded in WRF-Chem 47 . More than 20 biogenic species, including isoprene, monoterpenes (e.g., α -pinene and β -pinene) and sesquiterpenes, were considered and then involved in the photochemistry calculation.
The simulation was conducted for 20 September to 20 November 2014 (Fig. S9), and 1 August to 26 September 2015 (Fig. S10), during which each run covered 24 hours. The chemical outputs from the preceding run were used as the initial conditions for the next run. First two weeks were regarded as the model spin-up period, so as to minimize the influences of initial conditions and allow the model to reach a state of statistical equilibrium under the applied forcing 48 . Cluster analysis and PSCF analysis. In this study, the 24h air-mass back trajectories arriving at Beijing urban site (40°N, 116°22ʹ 48″ E) 500 m above ground level (AGL) were calculated every hour for each day, which were computed by the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (http://ready. arl.noaa.gov/HYSPLIT.php) of National Oceanic and Atmospheric Administration (NOAA) 49,50 . The AMBTS could be used to identify the transport pathways of pollutants and potential source regions by the calculating Lagrangian path of air parcels in the chosen time scale 51 .
Cluster analysis is based on the AMBTs through statistical analysis, which could be proposed as a useful way to assess the potential sources of ambient 52 . In this study, cluster analysis was served for the periods both Parade and APEC through the software TrajStat 53 (http://www.meteothinker.com). The potential source contribution function (PSCF) analysis has been frequently used to identify the suspicious locations of emission sources that influence pollutant concentrations at the receptor site [54][55][56][57][58] , which is based on the estimates of the motion of AMBTs in time with contamination density measured at the receptor site. In this study, the AMBTs were distributed with the cells of 0.2° × 0.2°resolution grid. And the grid cells PSCF values were computed by counting the trajectory section endpoints terminating within each cell, including trajectories which are not only ending at the cell but also passing through the cell. The PSCF value could be described as: Where n ij represents the total number of trajectory section endpoints that fills into the ij th cell, and m ij is the number of section endpoints in the identical cell corresponding with trajectories associated with constituent values at certain receptor site surpassing a pre-specified criterion value 59 . In this study, the criteria values were the corresponding mean NO 2 VCDs of MAX-DOAS at the receptor site. Hence, cells with high PSCF values suggest these areas are likely to produce high pollutant values at the receptor sites, so they are sufficiently deemed to be probable source regions. To diminish the uncertainty of PSCF resulted from small n ij values, every PSCF value should be multiplied by an arbitrary weight function W ij 60 to better represent the uncertainty in the values for which n ij with small values. The weight function W ij is defined as: Where Avg presents the mean number of endpoints in each cell. The PSCF value would be reduced by the weight function when the sum of endpoints in a cell was less than around three times the mean value of the endpoints per cell 56 . In this study, the contributions of other atmospheric pollution source regions at Beijing urban site was identified by the PSCF analysis with the software TrajStat 53 .