Variability in Transport Pathways and Source Areas of PM 10 in Beijing during 2009 – 2012

The transport pathways and source areas of PM10 in Beijing were examined on the basis of a model-assisted analysis. Computed back trajectories were used to trace the air history. The aim of this work was to study the main source areas of PM10, the variability of transport pathways, and potential source areas in Beijing. The results reveal that the major potential source areas of PM10 in Beijing were Hebei, Shandong, Tianjin, northwest of Inner Mongolia, and Outer Mongolia. The main source areas of PM10 have changed from the northwest to the south and southeast of Beijing during 2009–2012. During the study period, the regional contributions of PM10 from Shandong, Tianjin and Henan increased, whereas those from Inner Mongolia and Mongolia decreased compared with 2003–2009. The northwest airflow is a key factor in extreme pollution episodes. Sand storm partly contributed to the PM10 concentration in fast northwesterly transport paths. PM10 concentrations in winter and spring were higher than autumn and summer. In spring and summer, Beijing was strongly affected by long-range transport. Long-range transport had a weaker effect on PM10 concentrations during autumn and winter. The clustering of back-trajectories and PSCF results indicate the need to reduce PM10 transport from areas surrounding Beijing, particularly from the south of Beijing.


INTRODUCTION
Particulate matter (PM) pollution is presently recognized as affecting climate change, visibility and human health, many scientists have investigated the method to prevent PM pollution (Tao et al., 2009;Sang et al., 2010;Matus et al., 2012;Sun et al., 2015;Liang et al., 2016).The regional transport of PM is of increasing concern in China (Sun et al, 2012;Lai, 2015).PM 10 regional transport in Chinese cities have been studied extensively in the last two decades (Wang et al., 2004;Cheng et al., 2011;Zhu et al., 2011;Li et al., 2012;Yan et al., 2015).The regional transport and source area of PM 10 vary from one region to another, depending on their geographic position and emission sources.Beijing is a megacity with a population of 21.516 million and is subject to severe PM 10 pollution (Li et al., 2013;Mo et al., 2014;Koo et al., 2015;Xie et al., 2015).Power plants, industry, domestic sources, transportation, agriculture, and biomass open burning were identified as the local PM 10 sources (Wang et al., 2008a).Many scientists have investigated that the PM 10 transported from Shanxi province, Tianjin municipality and Inner Mongolia could not be ignored.The total emission source contribution ratio from these regions was 26.65% (Huang et al., 2010).Wang et al. (2004) and Sun et al. (2008) indicated that PM 10 transported from northeastern of Beijing causes extreme pollution episodes.A study on the airspace over Kazakhstan revealed that the arid region in the southwest of Inner Mongolia was the main potential source of PM 10 in Beijing (Wang et al., 2004).A previous study reported that Horqin Sandy Land in northeastern China was a potential source of PM 10 (Zhang et al., 2003).However, high-pressure situations were predominant and the prevailing wind directions were north and northwest in spring.The transport in summer was slower and typically from the southeast, therefore, considerable seasonal variation of PM 10 regional transport has been observed in Beijing.
Rapid economic development and increasing number of vehicles play a crucial role in the increase of PM 10 pollution (Han et al., 2005;Tang et al., 2005;Huang et al., 2010).Hu et al. (2015) indicated that further investigation of PM 10 source is warranted because source contributions have been changing in Beijing.Severe haze events from particulate matter have replaced dust storms in Beijing (Wang et al., 2004;Huang et al., 2010).Dust storm is generally defined as a storm that carries a great deal of dust and sand lofted by strong wind.Haze is defined as the weather phenomenon which leads to atmospheric visibility less than 10 km due to the moisture, dust, smoke, and vapor in the atmosphere, while fog leads to atmospheric visibility less than 1 km.Most of the haze results from excess PM emitted by anthropogenic sources and particles produced via gas-to-particle conversion (Watson, 2002).A recent study on aggravated PM 2.5 pollution indicated that regional emission sources are located to the south and west part of Beijing (Wang et al., 2015).Besides, it is crucial to clarify regional emission resources of PM 10 , as the PM 2.5 and PM 10 share a similar potential source area and their ratio ranges from 0.5 to 0.9 (Wang et al., 2012).
The level of PM 10 depends on not only their emission sources but also meteorological parameters, such as wind direction, specific weather conditions (Zhao et al., 2011).Low-pressure wind flows are typically conducive to pollutant transport.Stable atmospheric conditions and relatively low scavenging by clouds and precipitation are additional factors influencing the effectiveness of the transport of aerosols (Quinn et al., 2007).Studies have shown that haze events are related to synoptic weather patterns.Li et al. (2015) indicated that wind velocity is negatively related to PM 10 concentrations in Beijing.Extreme pollution events were also frequent in Beijing.For instance, on Jan 12, 2013, the hourly average PM 10 concentration in urban areas was more than 700 µg m -3 (Liu et al., 2014).Zhang et al. (2003) proved that north and northeast airflows originate from the inland of China, and that the northern Pacific wind brings cold, dry, and clean air because of the specific terrain of Beijing (e.g., bordered to the southeast by the North China Plain and surrounded by mountains).However, effects of the long-range transport of PM 10 on extreme pollution events are poorly understood though there are some studies on the possible causes of high PM 10 episodes in Beijing.
Aiming to determine the contribution of regional PM 10 transport to extreme pollution episodes, and to identify the main sources and paths of air for such events, this study mainly investigated the variability in transport pathways and source areas of PM 10 in Beijing during 2009-2012.This study focused on the following factors: (1) where are the main source areas of PM 10 in Beijing?(2) Have the transport pathways and potential sources changed in Beijing during the last decade?(3) What are the key factors for extreme pollution episodes?

Data and Methods
Data were collected from January 2009 to December 2012 at Beijing Environment Protection Bureau web site.19 air quality monitoring stations were installed to monitor the concentrations of PM 10 .The stations are described in Fig. 1.More than 90% data were available for daily PM 10 Air pollution index (API) for the study period.The API values were converted to PM 10 concentration by:  Thus, the API is 500 when ρ (PM 10 ) ≥ 600 µg m -3 .The maximum value in an artificially established upper limit of PM 10 is 600 µg m -3 .Gridded meteorological data (GDAS) were obtained from the National Oceanic and Atmospheric Administration (NOAA) website (http://ready.arl.noaa.gov/archives.php).Among several meteorological data sources, including the North American (Eta) Data Assimilation System, GDAS one-degree archive and GDAS half-degree archive, the GDAS one-degree archive were selected.Meteorological parameters including air pressure, wind velocity, temperature, relative humidity and rainfall were obtained from the China National Meteorological Information Center (http://data.cma.gov.cn/article/getLeft/id/249/key Index/1.html).

Trajectory Cluster Analysis and PSCF Analysis
In this study, TrajStat software was used for cluster and PSCF analysis (Wang et al., 2009).This software incorporates trajectory statistics and the geographical information system, and allows users to customize the trajectory map and PSCF map, which has been widely used to explore the effect of air mass transport on particles (Wang et al., 2015;Xiao et al., 2015;Zhang et al., 2015b).
PM 10 concentrations were averaged from 19 air quality monitoring stations.A prior selection procedure was adopted to select episodes which were common to all or most monitoring stations simultaneously (Karaca et al., 2009(Karaca et al., , 2010)), aiming to eliminate data affected by local meteorology and emissions.The outliers with abnormally high PM 10 concentrations were excluded to minimize the local meteorology and emission effects.The latitude and longitude of the starting point was set as 39.54°N, 116.23°E, and the altitude of the receptor site was set as 200 m.
Receptor heights are commonly 50, 100, 200, 500, or 1000 m above ground level (AGL) because the boundary level of an urban atmosphere is considered to be 1000 m AGL at midday and considerably lower at night.In general, 50 m AGL appropriately presents the airflow pathways of local and regional pollutants (Song et al., 2008).PM 10 concentrations are typically high within the planetary boundary layer (PBL).Wehner et al. (2008) determined the PBL and selected 200 m AGL as the receptor height using a light detection and ranging system.Zhu et al. (2011) used 100, 200, 300, 500, and 1000 m AGL as the receptor heights and found no significant differences.Considering that air pollutant concentrations in Beijing were recorded at ground level, the receptor height of 200 m AGL is determined.Five-day (120 h) back trajectories arriving at 200 m AGL per 6 h (00, 06, 12, and 18 UTC) are calculated using TrajStat software.The total number of trajectories in the 4 years is 1275 with a missing rate of 4.04%.9 clusters provide the most accurate representation of air mass classifications according to the Euclidean distance between the trajectories (Fig. 2).Each cluster represents a specific transport direction and pattern.The seasonal proportion of each cluster was calculated (Fig. 3).Trajectories with PM 10 > 150 µg m -3 were classified as polluted trajectories.K-means clustering was used to classify the nine clusters into three categories (i.e., low, intermediate, and high) using IBM SPSS Statistics Version 19 (SPSS, Chicago, IL, USA).The trajectory cluster analysis indicated the direction of the potential source of PM 10 .
Potential contribution area of PM 10 was analyzed by Back trajectory.As previously reported, the use of trajectory has some limitations (Masiol et al., 2015).Considering the qualitative nature of cluster analysis results, PSCF analysis has been recognized as extremely useful in investigating potential source areas, which are indicated by high PSCF  values (Biegalski and Hopke, 2004).The PSCF value represents the probability of the air parcel residence time (Lin et al., 2002).On the basis of the results of trajectory and cluster analysis, the PSCF method was used to identify the potential sources of PM 10 .The location of the identified source ranged from 25°E-80°E to 35°N-150°N, and this area was divided into 25300 small cells (0.5° × 0.5°).The weight function W ij was set as follows: 80 1.00 20 80 0.70 10 20 0.42 10 0.05

Extreme Pollution Episodes Analysis
Extreme pollution episodes in Beijing have been reported to last as long as 5 days (Sun et al., 2014).During extreme pollution episodes, haze rapidly peaks and then sinks, of the reason which has not been analyzed (Zheng et al., 2015).Therefore, the key factors leading to extreme pollution episodes require further investigation.Here, an event with a daily PM 10 concentration more than 420 µg m -3 is assumed as the extreme pollution episodes.Clean day is defined for comparison (daily average PM 10 concentration < 50 µg m -3 ).A NOAA hybrid single-particle Lagrangian integrated trajectory HYSPLIT model (http://www.arl.noaa.gov/ready/hysplit4.html)was used to calculate air mass back trajectories.The 72-h back trajectory of each high and low PM 10 episode in Beijing was assessed at 6-h intervals.

Identified Transport Pathways and Sources Areas
Cluster analysis reveals that trajectories from various directions have distinct effects on the PM 10 concentrations.
The clusters were classified into 3 groups according to the average PM 10 concentrations: (1) low, clusters 2, 3, and 5; (2) intermediate, clusters 1 and 6-9; and (3) high, cluster 4 (Table 1).Extreme pollution episodes were commonly observed during winter and lower PM 10 concentrations in summer (Fig. 3).Clusters 1 and 2, which originated from Outer Mongolia and Xinjiang and then traveled southeast over Inner Mongolia and Hebei, were prevalent in winter and autumn, followed by spring and summer.Cluster 3 was predominant in autumn and winter; the PM 10 concentration of cluster 3 in summer was higher than those of clusters 1 and 2. Cluster 4, which originated from northern Inner Mongolia and then passed over Hebei, was predominant in spring and summer.Cluster 5 was a long northwest cluster and uniformly distributed among all four seasons.Cluster 7 originated near the Yellow sea at a lower altitude and primarily occurred in autumn.In addition, clusters 6 and 8 showed similar seasonal variation, and cluster 9 was predominant in spring.The highest PM 10 concentration was observed in cluster 2, followed by clusters 5 and 3 (Table 1).
The major PM 10 transport pathway was the south airflow.Cluster 3 represents the contaminated airflow from the south (Fig. 4(c)), which was the most frequent cluster (18.00%) and the polluted trajectory accounted for 27.38% (Table 1).Relatively weaker winds, higher temperature, higher relative humidity, and stable synoptic pattern prevailed in cluster 3 (Fig. 5), which seems adverse to pollution diffusion.In addition to unfavorable synoptic patterns, the overall high concentration of PM 10 may associate with large emissions from the south region, which includes two industrial areas: Hebei, a major steel producer in north China covering 18.88 km 2 with a population of 71.85 million, and the Loess Plateau.Notably, the average annual emissions of major pollutants in Hebei during 2009-2012 were 8.38 × 10 6 tons for ammonia nitrogen, 131.08 × 10 6 tons for sulfur dioxide, 108.14 × 10 6 tons for dust and smoke, and 122.33 × 10 6 tons for nitrogen oxide (Supporting information, Fig. S1).The second PM 10 transport pathway was the short and long northwest airflow.Among the high PM 10 concentration clusters, cluster 2 (Fig. 4(b)) and 5 (Fig. 4(e)) represent the short and long northwest airflow, respectively.Cluster 2 and 5 exhibit low temperature, moderate relative humidity and high air-pressure weather pattern for influence of Siberian high-pressure system.By comparison, cluster 5 exhibited high wind velocity, whereas wind velocity of cluster 2 was low.Notably, this synoptic pattern of clusters 2 and 5 can promote pollutant diffusion, whereas the PM 10 concentration was the highest.One possible explanation is the bare land and desert to the northwest of Beijing, which contribute to the maximum weighted average of PM 10 concentrations.In addition, the terrain of northwest China is mainly a desert zone, including the Badain Jaran Desert, Tengger Desert and Kumtag Desert (Fig. 1).According to the PSCF analysis results, the south and northwest region of Beijing are the main PM 10 sources.For the intermediate PM 10 concentration cluster, airflows from both cluster 6 (Fig. 4(f)) and 9 (Fig. 4(i)) originate from Russia and Mongolia, and differed in the length.Specifically, those in cluster 9 were longer than cluster 6, indicating that the wind velocity of cluster 9 was higher.Contributing sources of PM 10 in both clusters 6 and 9 were located in the northwest plateau region, Inner Mongolia, and arid and semiarid regions of Mongolia.The high PM 10 concentration can attribute to nearby areas that are bare of vegetation.Hence, a higher wind velocity can lead to higher PM 10 concentrations.Moreover, the temperature and relative humidity of cluster 6 were higher than those of cluster 9, whereas the air pressure was lower.Cluster 6 was predominant in winter, and cluster 9 mainly occurred in spring, resulting in differences in synoptic patterns between these clusters.In addition, cluster 9 was associated with the Mongolian high-pressure system, which is strongest before and after January and have strong effects on the entire Eurasian continent.The Mongolian high-pressure system, moving east in spring and disappearing in midsummer, has the widest coverage of all high-pressure systems in the northern hemisphere.Cluster 8 (Fig. 4(h)), which originated from Lake Baikal and passed over grassland and woodland in eastern Inner Mongolia with a high wind velocity and low relative humidity, had the lowest PM 10 concentration among the intermediate group.
The third transport pathway was the southeast airflows, which originated from the Jiaodong Peninsula, northwest plateau region, Inner Mongolia, arid and semiarid regions of Mongolia.Cluster 7 (Fig. 4(g)) represents the southeast airflow that originates from the Jiaodong Peninsula and passes through Langfang and Jinan.The average PM 10 concentration of cluster 7 was considerably high up to 130.33 ± 54.01 µg m -3 .According to the first global city air pollution survey report by the World Health Organization, PM 10 concentrations in Langfang (1016), Jinan (1039), and Beijing (1035) were significantly high among the 1082 surveyed cities.In addition to the primary particulate emissions, secondary particles also have an essential role in particle sources.Large emissions of secondary aerosol precursors (sulfur dioxide and nitrogen oxides) from ships were observed in the Yellow Sea.Shandong province was reported to have higher concentrations of secondary particles up to 53% than those in Beijing (Wu et al., 2009).On average, southeast airflow 7 was accompanied by the following weather patterns with low wind velocity and air pressure, high temperature and RH, which promote secondary particle production (Zhang et al., 2015a).
The northern airflow showed the lowest PM 10 concentration, which originated from Inner Mongolia and passing over Chengde, Chifeng, northern Inner Mongolia and Hulunbeier.Cluster 4 (Fig. 4(d)), which originated from the northeast of Beijing, had high wind velocity, relative humidity, medium temperature and air pressure.Two reasons explain the low PM 10 concentration in cluster 4, the first being favorable weather conditions.Beijing is a city with high terrain in the north and west and low terrain in the south and east.High air pressure and wind velocity were conducive to pollutant diffusion in Beijing.The second reason is the low anthropogenic emissions (Supporting information, Fig. S1).Zhang et al. (2015a) reported that weak winds with low air pressure and high relative humidity promote high PM 10 concentrations.The weather in Beijing is typically regulated by meteorological conditions such as wind directions and wind velocity (Wang et al., 2014).In conclusion, high concentrations of PM 10 appeared in the south, northwest and southeast region.Northern trajectories were associated with relatively low concentrations of PM 10 .

Airflow Trajectory Characteristics in Extreme Pollution Episodes
During 2009-2012, four days were heavily polluted: Dec 15, 2009; Mar 20, 2010; May 1, 2011 and Apr 29, 2012, with PM 10 concentrations of 600, 548, 562, and 540 µg m -3 , respectively.Three clean days were observed: Nov 23, 2011; Dec 17, 2012; and Dec 18, 2012, when PM 10 concentrations were less than 50 µg m -3 .It was found that most extreme PM 10 pollution events occurred in spring and winter and were affected by northerly and northwesterly winds (Fig. 6).The trajectories on clean days were mainly from Kazakhstan and Mongolia.Some local trajectories were also observed.Compared with the length of trajectories on polluted days, the length on clean days was short, indicating a lower wind velocity (supporting information.Table S1).This result is consistent with the cluster analysis that northwest trajectories typically occurred in spring and winter and were accompanied by strong winds.In conclusion, the results indicated that the northwest airflow has a major effect on extreme PM 10 pollution events in Beijing.Although haze was considered as the main reasons for the episodes of poor air quality in Beijing (Sun et al., 2004;Wang et al., 2008b;Cao et al., 2013), sand storm had partly contributed to the extreme PM 10 pollution events in fast northwesterly transport paths.

Seasonal Variations of PM 10 Concentration and Source Area
PM 10 concentrations during 2009-2012 exhibited considerable seasonal variation.Pollution in winter and spring were higher than autumn and summer.The average concentrations of PM 10 in spring, summer, autumn and winter were 140, 112, 128, 133 µg m -3 , respectively (Fig. 7).In spring, the trajectories were grouped into 4 clusters, of which cluster 1 was regarded as the polluted one and accounted for 32.9% (Fig. 8(a)).The average PM 10 concentration of cluster 1 was (164.9 ± 56.7) µg m -3 and   much higher than the mean.Fig. 9 shows the potential source areas of PM 10 in four seasons.Spring and summer appear to be the most severely seasons that affected by the longrange PM 10 transport.In spring, southern Hebei and Tianjin were the major potential source areas (Fig. 9(a)).Also some studies suggested that in spring the air masses originated from the Inner Mongol and the Gobi desert might carry dust, and the air masses originated from Shandong province mixed with more pollutants emitted from combustion on their pathways including Shanxi and Hebei Province, where large amounts of coal were used for heating (Zhang et al., 2007;Wang al., 2010).
In addition to Hebei and Shandong, Tianjin was a major potential source area in summer (Fig. 9 (Fig. 8(b)) with the highest PM 10 concentration (134.93 ± 49.55) µg m -3 had the similar pathway with cluster 1 in spring, which originated from Yantai, Shandong and through Hebei to Beijing.Wu et al. (2009) indicated that Hebei, Tianjin and Shandong were the main source areas for secondary ions, and regional secondary formation led to aerosol pollution in the summer of Beijing.Han et al. (2015) reported that regional secondary formation led to aerosol pollution in summer.
Long-range transport has a weaker effect on PM 10 concentrations during autumn and winter.In autumn, the pathway represented by cluster 1 (31.6%) was the most polluted ((151.44 ± 65.47) µg m -3 ) (Fig. 8(c)).Cluster 1 originated from Hebei province and its short length suggested low wind velocity.Low wind velocity and relatively low boundary layer height in autumn and winter may exacerbate the air pollution, contributing to the frequent severe haze events (Miao et al., 2015).
In winter, the air masses associated with cluster 2 (28.1%) led to the highest PM 10 concentration ((154.43 ± 75.71) µg m -3 ) (Fig. 8(d)).This cluster is a long-range trajectory which originates from Baikal Territory, go through Inner Mongolia, and reaches Beijing.The PSCF analysis also showed that Hebei and Inner Mongolia are the major potential source area of PM 10 (Fig. 9(d)).

Variability in Source Areas of PM 10
Trajectory analysis was carried out to identify the sources of PM 10 of Beijing during 2003-2009 and four transport pathways were identified: the northwest pathway, the south pathway, the V-shape southwest pathway and the southwest pathway.PM 10 concentration of southwest pathway was highest, which traveled over Loess Plateau and the west and south of Hebei province (Zhu et al., 2011) (Fig. 10(b)).The lowest PM 10 concentration was found for the southeast pathway and northern pathway.These results indicated that PM 10 mainly originated from the Loess Plateau, Shanxi Province, the west and south of Hebei province and Mongolia.
In addition, the result of the trajectory and PSCF analysis for PM 10 at Beijing during 2009-2012 is shown in Fig. 10(a).Three transport pathways were identified: the northwest pathway, the south pathway and the southeast pathway.The lowest PM 10 concentration was found along the northern pathway, which may result from low emissions.The probable areas of regional PM 10 sources are located in the south, southeast, and northwest of Beijing.Cells related to high PSCF values were likely to produce higher concentrations at the receptor site and assumed to be possible source areas.The major potential source areas with PSCF values of 0.7-1.0 were Hebei, Shandong and Tianjin.Moreover, central Inner Mongolia and eastern Mongolia were also potential source areas of PM 10 , where PSCF values were 0.5-0.7.
The major source area of PM 10 in Beijing during 2009-2012 is Hebei province, which is located in the south of Beijing.A previous study emphasized a major effect on PM 10 for trans-boundary transform from the south of Beijing (Han et al., 2015).Airflow from the south contributed a major proportion to high PM 10 concentrations.Meanwhile, a high relative humidity and temperature can promote the transformation of secondary particles, which are major factors of PM 10 pollution in Beijing.In addition, a sharp  (Zhu et al., 2011).increase of mobile vehicle population and fuel combustion with the development of urbanization was found during 2009-2012 (Supporting information, Fig. S1).
The second potential source area of PM 10 is Shandong Province.The mass fraction of PM 10 decreased from the dust source to the downwind coastal site, whereas those of secondary and carbonaceous components increased.Similar results were reported by studies conducted in Beijing (Sun et al., 2004;Han et al., 2015;Liu et al., 2015).In addition, other minor source areas were identified that the main cause of strong pollution events in Beijing was attributed to dust transportation from the northwest of Inner Mongolia and Outer Mongolia.The transmission dust in Beijing was mainly from north, west, and northwest paths (Zhang et al., 2009).
The southwest pathway from the Loess Plateau exhibits low PSCF value, indicating low contribution to PM 10 in Beijing.Remarkably, the regional contributions of PM 10 during 2009-2012 from Shandong, Tianjin and Henan increased, whereas those from Inner Mongolia and Mongolia clearly decreased compared with those during 2003-2009.The association between pollution sources and protective measures was assumed to be the main factor for the change in air quality (Zhang et al., 2011).Many protective measures were implemented in Beijing and the surrounding regions to ease emissions of atmospheric pollutants leading up to the 2008 Olympic Games, which played a major role in blocking the dust contribution from the northwest.Dust storm changes in governance areas were reported to be markedly affected by vegetation cover and relative humidity in the past 30 years, with a contribution of 32.70% and 44.50%, respectively (Qin et al., 2012).Furthermore, the decrease in the annual average wind velocity and windy days may be a major factor for the variability in long-range sources of PM 10 in Beijing (Supporting information, Table S2).However, a growing trend was observed for emissions of major pollutants (ammonia nitrogen, sulfur dioxide, dust and smoke, nitrogen oxides) in Beijing and the surrounding regions (Supporting information, Fig. S1).

CONCLUSION
Variability in transport pathways and source areas of PM 10 in Beijing were investigated by cluster trajectory and PSCF methods.Three transport pathways were identified: the northwest pathway, the south pathway, and the southeast pathway.The lowest PM 10 concentration was found along the northern pathway, which may result from low emissions.The results of PSCF analysis revealed that the major potential sources were Hebei, Shandong, Tianjin, and northwest of Inner Mongolia and Outer Mongolia.Compared with 2003-2009, the regional contributions of PM 10 from Shandong, Tianjin and Henan increased, whereas those from Inner Mongolia and Mongolia decreased.In addition, the northwest airflow had a major effect on extreme PM 10 pollution events in Beijing.In spring and summer, Beijing was strongly affected by long-range transport, whereas was weakly affected during autumn and winter.However, additional studies on chemical characteristics of PM 10 in Beijing are warranted to elucidate the variability in source areas of PM 10 .

Fig. 1 .
Fig. 1.Locations of air quality monitoring stations and the elevation map of Beijing, which ranges from 0 to 2249 m.The six colors indicate the following provinces and municipalities: Beijing, Tianjin Municipality, Inner Mongolia, Shandong Province, Shanxi Province, and Hebei Province.

Fig. 2 .
Fig. 2. Cluster analysis results during 2009-2012.The black lines show the nine clusters, which represent different airflow directions and patterns.The pentagram, circular, and triangle represent the locations of the Kumtag Desert, Badain jaran Desert, and Tengger Desert, respectively.

Fig. 5 .
Fig. 5. Meteorological values, including (a) wind velocity, (b) temperature, (c) air pressure, and (d) relative humidity of the nine clusters in Beijing.The central rectangle spans the first quartile to the third quartile, and the segment inside the rectangle shows the median, whereas the small squares represent the average.The dashes above and below the box represent the outlier cutoff points.The triangles above and below or overlapping the dashes show the maximum and minimum values, respectively.
The mean concentrations of PM 10 in four seasons from 2009 to 2012 were lower than those during 2003 to 2009.For example, the average mass concentrations of PM 10 were 164, 124, 151 and 152 µg m -3 in spring, summer, autumn and winter during 2003-2009, respectively, indicating the seasonal change of PM 10 of spring > winter > autumn > summer.

Fig. 6 .
Fig. 6.Seventy-two-hour back trajectories arriving at 200 m AGL every 1 h for the clean days and extreme pollution days using the NOAA on-line HYSPLIT4 trajectory model.(A) Three clean days (PM 10 < 50 µg m -3 ).(B) Four extreme pollution days (PM 10 > 420 µg m -3 ).Each color line shows 1-h trajectory.Horizontal lines at the bottom show the height of the trajectories.

Fig. 9 .
Fig. 9. Potential source areas of PM 10 in Beijing during the four seasons.

Table 1 .
Mean PM 10 concentration and the frequency of clusters based on all trajectories and polluted trajectories.NW L " represents the long northwest trajectory; "NW S " represents the short northwest trajectory.