Analysis of Long-Range Transport Effects on PM 2 . 5 during a Short Severe Haze in Beijing , China

Comprehensively using Inverse Distance Weighted (IDW) analysis, Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model, three-dimensional (3D) cluster analysis, Weight Potential Source Concentration Function (WPSCF) analysis and other statistic methods, we mainly studied about the spatio-temporal variation, long-range transport and potential source regions of PM2.5 in Beijing during a short severe haze from Dec05 to Dec11, 2015. The results showed that the concentration of PM2.5 decreased from south to north of Beijing. PM2.5 accumulation in the short-severe haze had high correlation with calm and steady meteorological condition (high relative humidity (RH), low wind speed (WS), low boundary-layer temperature (BLT) and surface air pressure (SAP)). In addition, air-flow in different heights (500 m, 1500 m and 3000 m) had different effects on the haze episode and the air flows at 500 m had the greatest contribution of the air pollution. The potential sources were mainly from the desert in northwest of Beijing and the built-up areas in Jing-Jin-Ji zone. Higher WPSCF values (> 0.7) were mainly distributed in Hebei, west Shandong province (around 0.5) and south Tianjin (around 0.5).


INTRODUCTION
With rapid development of urbanization and industrialization in recent years, Beijing, as one of the most rapidly growth city in China, was severely influenced by the heavy haze, meaning composed of the fine particulate matter (i.e., with aerodynamic diameters not larger than 2.5 µm, or PM 2.5 ) (Ping et al., 2017).Urban PM 2.5 originated mainly from both natural and anthropogenic emission sources, such as soil dust, sea/road dust, fossil fuel combustion, biomass burning and agricultural activities (Belis et al., 2013;Zhang et al., 2015a, b).Moreover, PM 2.5 had a serious impact on the respiratory system and cardiovascular system and led to asthma, lung cancer, pulmonary disease and some other serious diseases (Pope et al., 2002;Huang et al., 2014).
There were many studies about the spatio-temporal analysis of PM 2.5 in different places.Pearce and Naeher (2009) used kriging with external drift (KED) to provide high resolution maps of PM for a downtown region of Cusco, Peru.Beckerman et al. (2013) researched the spatio-temporal variability of PM 2.5 by Bayesian Maximum Entropy (BME) interpolation in the contiguous United States on the national scale.However, to our knowledge, few researchers used Inverse Distance Weighted (IDW) to study the short-time spatial distribution of PM 2.5 in Beijing.Li et al. (2016) compared shape function (SF) and IDW in a real-time research.Ramos et al. (2015) developed a hybrid interpolation technique combining the IDW with Kriging and with KED to apply it to daily PM 2.5 levels.Li et al. (2014) once changed the Euclidean distances in IDW to the shortest wind-field path distances to characterize the impact of complex urban wind-field on the distribution of the particulate matter concentration.
Normally, a severe haze pollution episode was generally associated with serious anthropogenic emissions under the special meteorological condition.The heavily polluted PM 2.5 episodes were frequently related to high relative humidity levels (Huang and Tai, 2008;Barman et al., 2008;Massling et al., 2009;Cheng et al., 2015).And the stable synoptic meteorological condition related to relatively low temperature and low wind speed in winter aggravated the accumulation of pollutants (Sun et al., 2006;Zhao et al., 2013).With further in this study, we had interpreted the reason of that this severe haze episode occurred in Beijing during Dec05 to Dec11 combined with correlation analysis between PM 2.5 mass concentration and meteorological condition, rather than the macroscopic description of the meteorological condition.PM 2.5 was responsible for the formation of regional haze because of its relatively long suspension time in the air and special optical properties (Sun et al., 2013).And the regional long-range transport of PM 2.5 played an important role during the severe pollution events in Beijing (Zheng et al., 2015) so it's worthy to determine the characteristics of the long-range transportation and potential regional sources of PM 2.5 .In the aspect of pollutant long-range transportation and potential sources, the HYSPLIT model had become a robust tool in recent years.Nowadays, the backward trajectory model was used in many regions to track the airflow carrying different particles, such as pollen, surface ozone, dust, and PM 10 (Shan et al., 2009;Li et al., 2012;Zemmer et al., 2012;Cao et al., 2014).When it comes to the long-range transport of PM 2.5 in Beijing, Wang et al. (2015) used the cluster analysis from back trajectories to find the regional sources of PM 2.5 in Beijing in a long time scale from 2005 to 2010.Zhang et al. (2013) argued the source apportionment of PM 2.5 in seasonal perspective.Shen et al. (2016) found that the main air mass during haze episodes in spring was from the south-easterly directions by cluster analysis.However, those researches were mainly concentrated on long-time research in the near surface layer by back trajectory cluster analysis and there were few studies about the difference of the backward trajectories arriving at the layer of 500 m, 1500 m and 3000 m (Makra et al., 2011(Makra et al., , 2013)).And few studies focused on the change principles of the air-flow trajectories under the weather system in a short-time serious air pollution event (Wang et al., 2016).What's more, we also used the three-dimensional (3D) cluster analysis to show the vertical distribution of PM 2.5 in this study.The Potential Source Contribution Function (PSCF) was also put into usage in order to define the contribution of potential source regions.The results not only can provide the references to Beijing government to fight against the short-time severe haze and build the air pollution regional control zone, but also can provide a new perspective of long-range transport and regional potential source of pollutants research.

Study Location and Monitoring Data
Beijing (115°25′-117°30′E, 39°28′-41°05′N) is one of the metropolises in the world.It covers about 16410.54km 2 , including 1500 km 2 of urban settlement.Beijing locates in the North China Plain which dominates and covers most of the central urban area and industries in the south and east.The land slopes downwards gradually from the northwest to the southeast.Yanshan Mountains in the northern, Jundu Mountain in the northwestern and Xishan Mountain in the southwestern contains Beijing as a dustpan shaped terrain (Fig. 1).Beijing belongs to warm temperate semihumid continental monsoon climate.The mean annual temperature and precipitation are 14.1°C and 600-800 mm, respectively (http://www.gov.cn/).
35 air pollution (PM 2.5 ) monitoring stations were selected (Fig. 1), including 23 urban environmental assessment sites, one urban background site to reflect the ambient air quality, six cross-region transmission sites close to Beijing municipal boundary in six directions to monitor the transmission of pollutants between regions, five traffic sites at the edges of busy roads, covering all districts and counties, suburbs, towns, traffic roads, residential areas and others.
Daily mean PM 2.5 mass concentrations of Beijing from Dec05 to Dec11, 2015 were collected from Beijing Municipal Environmental Monitoring Center (available at: http://zx.bjmemc.com.cn/).The hourly mean PM 2.5 mass concentrations were downloaded from the online detection and analysis platform for air quality in China (available at: http://www.aqistudy.cn/).Meteorological data including boundary-layer temperature (BLT), wind speed (WS) and relative humidity (RH) were available from Wunderground Website (available at: https://www.wunderground.com/).

Trajectory Data
Because a single height backward trajectory had a large uncertainty and was of limited significance (Stohl, 1998), a three heights representation of the air masses arriving at Beijing were made via the analysis of a large number of atmospheric trajectories in this study.72-hour backtrajectories arriving at 500 m, 1500 m and 3000 m Above Ground Level (AGL) with a 0.5° × 0.5° latitude-longitude grid were calculated every 6h a day (00, 06, 12, and 18 h UTC) from Dec05 to Dec11, 2015.Daily meteorological data were obtained from Global Data Assimilation System (GDAS) provided by NECP and it could be downloaded from HYSPLIT website (available at: http://ready.arl.noaa.gov/HYSPLIT.php).
A well-mixed convective boundary layer, middle atmosphere layer and high atmosphere layer for the region were investigated in this study, because back-trajectories at the three typical heights had totally different characteristics on the PM 2.5 concentrations of the target site (McGowan and Clark, 2008).The lowest layer (500 m: active for the boundary layer (BL)) was approximately the height of the mixing layer and had been found to be a generally largest influence on the PM 2.5 concentration (Su et al., 2015).The middle layer (1500 m) represented the BL top and the highest (3000 m) was characteristic for the free troposphere (FT) height (Makra et al., 2011(Makra et al., , 2013)).

The Inverse Distance Weighted (IDW) Method
IDW was one of the most useful spatial interpolation methods for estimating the values of the attribute at a site by using the same attribute sampled at neighbor points.IDW assumed that each measured site had a local influence that diminished with distance.Since the pollutant concentration data was unavailable in most study locations, spatial estimations were increasingly used to estimate the geocoded ambient pollutant concentrations in many researches.In this paper, the spatial variations of PM 2.5 mass concentrations in Beijing each day from Dec05 to Dec11, 2015 were drawn using ArcGIS 10.3 software.We used the location information of 35 sites as the "input point feature" and used the monitoring data of each site each day as the "Z value field".Based on the format (1), the ArcGIS could calculate the value near each site automatically.
P was the wight of distance change.P was larger, the influence of the distant site was smaller.P equaled 2 in this study based on the variance theory.The quantity d i0 was the distance between the prediction location and each monitoring site.

Back Trajectory Clustering
Cluster analysis was a multivariate statistical analysis tool which was used to divide the backward trajectories into different transport groups or clusters (Xin et al., 2016).Using the 3D back trajectory cluster analysis at three heights (500 m, 1500 m and 3000 m), we identified the dominating backward trajectory groups or clusters and found the contributions of target areas in the following research.
To determine the direction where the air masses reached the site, we used the angle distance clustering method (Alain and Bottenheim, 1995;Wang et al., 2009;Li et al., 2012), which was defined as flowing: (2) 4) where d 12 was the mean angle between the two backward trajectories, X 0 and Y 0 were defined as the position of the backward trajectory origin point.X 1 (Y 1 ) and X 2 (Y 2 ) referred to backward trajectories 1 and 2, respectively.

The Weight Potential Source Contribution Function (WPSCF) Analysis
The PSCF was the conditional probability that a parcel with a certain level of pollutant concentrations arrived at a receptor site after having passed through a specific upwind source area (Hwang and Hopke, 2007;Byčenkienė et al., 2014).The zone of concern was divided into i × j small equal grid cells.The grid cell in this paper was a 0.5° × 0.5° latitude-longitude grid.y ij was the total number of end points that fell in the ijth cell and x ij was the number of end points that were associated with samples that exceeded the threshold criterion in the same cell (Kong et al., 2013;Liu et al., 2013).In this study, we defined 75 µg m -3 as criterion value of PM 2.5 mass concentration according to the Class II category of the National Ambient Air Quality Standards of China (NAAQS) (GB3095-2012).Those cells with high PSCF values were likely to contribute to high concentration events at receptor sites (Gao et al., 2014).Therefore, it was reasonable to assume that they might be possible source areas.
The PSCF was then defined as: Since the PSCF was computed as a ratio of the counts of selected events (x ij ) to the counts of all events (y ij ), it was expected that x ij would be relatively smaller than y ij .Values related to sparse trajectory coverage of the more distant grid cells might result in a PSCF values with high uncertainty in the apparent high value (Karaca et al., 2009).In order to minimize the uncertainty, an empirical weight function W ij was multiplied by the PSCF value.That was, WPSCF = W ij × PSCF.In this study, the WPSCF values were calculated to evaluate the potential sources of PM 2.5 in Beijing, based on 72h daily backward trajectories arriving at the study area at 0:00 (UTC) during the study period from Dec05 to Dec11, 2015.

Spatio-Temporal Analysis of PM 2.5
During the study period (Dec05 to Dec11, 2015), a serious pollution happened in North China which covered most regions in Beijing-Tianjin-Hebei zone.Affected by the heavy pollution, Beijing issued the first red alert on haze and fog this year, which lasted from 7am on Dec08 to 12am on Dec10.It's the first time that Beijing has issued a red alert for haze and fog since 2014 when Beijing adopted the emergency response program for air pollution.
According to Fig. 2, PM 2.5 mass concentration increased dramatically from 14pm Dec05 to 23pm Dec09 and then decreased sharply.Most of the hourly mean PM 2.5 mass concentration during the seven days was far more than 75 µg m -3 .The highest hourly mean PM 2.5 value loading in Beijing was 289 µg m -3 at 22 pm Dec09 and the lowest were 14 µg m -3 at 14pm Dec10 and 16 µg m -3 at 14pm Dec05.In Dec05, Dec10 and Dec11, because of the little snow which could deposit the PM 2.5 particles and even the component materials in the air, the hourly mean PM 2.5 value was lower than 75 µg m -3 in most of the time.There were also pronounced diurnal variations.Taking the daily variation of PM 2.5 on Dec06 as an example, the lowest value was at 8am because the affluent dew made the PM 2.5 particles depositing in the morning.PM 2.5 was relatively high after 16pm because of the quickly increasing vehicle emissions and then the PM 2.5 mass concentration had a slightly decreasing after the traffic restriction.During the night time, the PM 2.5 mass concentration increased again influenced by the inversion layer and the stable atmosphere.
The spatio-temporal distribution of PM 2.5 was illustrated in Fig. 3.The PM 2.5 mass concentration was completely high all around Beijing from Dec07 to Dec09.The PM 2.5 air pollution was heavier in the south areas than that in the north, because the urban center and industrial areas were in the south and southeast of Beijing where the local emissions such as vehicle emissions and heating systems could contribute to the heavy air pollution.The minimum values were around the northeast monitoring site of Beijing which was a regional transmission site mainly to monitor the pollution level of air-masses from the northeast of Beijing.And the maximum values were usually around the south, southeast and southwest regional transmission monitoring sites which mainly monitored the pollution level of airmasses from Tianjin and Hebei province.
High PM 2.5 concentration was a regional phenomenon.The accumulation process of PM 2.5 particles occurred successively from the south to north.During Dec05 to Dec11, 2015, a high pressure system separated from Siberia -Mongolia high pressure and the high pressure center moved from Mongolia to Beijing.As a consequence, with the influence of high pressure, the wind carrying PM 2.5 traveled from Hebei province and then back to Beijing.Because the Hebei province and Tianjin covered highly coal industrialized and densely populated areas, the mixed considerable pollutants emitted from these regions could be transported to Beijing with south air-masses.However, the special terrain condition which surrounded by Xishan Mountains, Jundu Mountain and Yanshan Mountain in the southwest, northwest and north was the barrier for the low air flows from northeast China.

Meteorological Influencing Factors
Large quantities of anthropogenic emission sources and special weather conditions were the main causes of air pollution in the winter of Beijing (Zhao et al., 2009(Zhao et al., , 2013)).Especially, this severe haze in Beijing had a strong correlation with BLT, RH, WS.However, the SAP had no significant correlation with PM 2.5 concentration in this study.The value for SAP was around 1023 Pa to 1034 Pa  and it was relatively smooth.RH and PM 2.5 followed the same trend but temperature and PM 2.5 , WS and PM 2.5 showed a negative correlation.The correlation coefficient (CORRCOEF) between the RH and the PM 2.5 concentration was 0.819 (α = 0.01) and the CORRCOEF between the WS and PM 2.5 concentration was -0.502 (α = 0.01).In this study, the correlation between temperature and PM 2.5 mass concentration was weak (-0.164) compared with the others, but it was also significant at the 0.05 level (2-tailed).In Fig. 4, the RH was high during the haze event.Most of RH values were greater than 70% and even up to 100%.The high RH in haze and fog episodes would accelerate the formation of secondary aerosols and aerosol hygroscopic growth which could aggravate the pollution level of the atmosphere (Li and Han, 2016;Sun, 2006).Therefore, the daily variation of PM 2.5 in Beijing tracked the pattern of RH.Previous studies showed that the RH and WS were the two most important factors influencing the concentration of PM 2.5 pollution (Gao et al., 2015).During the haze episode, the WS was about 0-2 m s -1 , illustrating that the weak wind was conducive to the diffusion of pollutants.And the BLT was around -2 to 4°C.Thus, the static stability atmosphere, relatively low BLT, WS and high RH in winter, would lead to the accumulation of PM 2.5 .In order to find out the principal component of the effects on PM 2.5 concentration, a methodology -factor analysis and special transformation was put into usage (Jahn and Vahle, 1968;Jolliffe, 1993;Matyasovszky et al., 2011;Makra et al., 2012).Through the hourly data for the "target variable" -PM 2.5 mass concentration and "influencing variables" -RH, WS, BLT and SAP for the severe haze period, the results showed that: the principal component one (Z 1 ) had a significant negative correlation with SAP (-0.419) and RH(-0.812)but had a significant positive correlation with BLT (0.802) and WS (0.755), which could be considered as atmospheric stability.The larger the value of Z 1 was, the more unstable the atmosphere was, and the smaller the value of Z 1 was, the more stable the atmosphere was.Through the correlation coefficient between the concentration of Z 1 and PM 2.5 could be seen that the atmosphere was more stable, PM 2.5 concentration was higher.Meanwhile, the contribution rate of Z 1 reached 51.17%, indicating that the main reason for the formation of PM 2.5 pollution in Beijing was calm and steady meteorological condition.Indepth analysis of the main components of Z 1 , the haze occurred often with low BLT, small WS, high SAP and high RH.These phenomenon was regarded as temperature inversion which often occurred in the winter morning.The principle component two (Z 2 ) had a significant positive correlation with SAP (0.868) and had a negative correlation with RH (-0.468) and had no significant correlation with BLT (0.158) and WS(-0.189).At the same time, Z 2 and PM 2.5 concentration showed a high degree of negative correlation, which represented the combination of SAP and RH -the lower the air pressure and the greater the humidity, the more prone to haze weather.Therefore, the Z 2 might mean the weak low pressure.During the study period, a high pressure occurred in Beijing, which promoted the accumulation of PM 2.5 .The principle three (Z 3 ) had a strong positive correlation with WS (0.597) and had a strong negative correlation with BLT (-0.514) and had no significant correlation with RH (-0.053) and SAP (0.194).Z 3 and PM 2.5 concentration showed a high degree of negative correlation, so Z 3 represented the combination of BLT and WS.Z 3 might mean the influence of the southwest wind in temperate zone.

Air-Flow Trajectories
The results of backward trajectories arriving at three different heights (500 m, 1500 m and 3000 m) were shown in Figs. 5, 6, and 7, respectively.Normally, different source regions mean different kinds of components in PM 2.5 and the length of air-flow trajectories mean the capacity of the deposition of PM 2.5 particles.The relatively longer air-flow trajectory had a fast speed and it's unfavorable for the deposition of particles based on the dynamic principle (Perrone et al., 2013).
In winter, Beijing was often influenced by cold anticyclone.Those conditions usually led to high levels of pollutant concentrations due to weak mixing and unfavorable for the dispersion of pollutants (Zhao et al., 2013).The movement of the high-pressure center caused the change of the source regions and length of air-flows.When the center of the surface high-pressure system persisted in Beijing from Dec06 to Dec09, 2015, the severe haze episode occurred.On Dec05 (Fig. 5(a)), there were two major source regions: the south of Siberia and the northeast of China (the Hulun Buir Sandy Land and the borders between the northeast of Inner Mongolia and Heilongjiang Province).These air masses distributed widely, had a relative longer path and stayed relative shorter time in Beijing, so they were hard to carry PM 2.5 particles to Beijing although the air-flows crossed large semi-arid and arid regions in Mongolia and Inner Mongolia.On Dec06 (Fig. 5(b)), the transport direction started to change from the northwesterly to the southerly of Beijing for the surface high-pressure got to Beijing.Most of the air flows became shorter and the air flows around Beijing made a small cyclone.And then on Dec08 (Fig. 5(c)), influenced by the extended high-pressure, Beijing was located in the center of the high-pressure.Under the synoptic condition, winds at the surface layer were very weak with the wind speed lower than 2.0 m s -1 in Beijing and even equal to 0 m s -1 sometimes, and the relative humidity was usually 100%.The high-pressure, weak winds and high relative humidity were unfavorable for the dispersion of pollutants (Gao et al., 2015;Wang et al., 2015;Zhang et al., 2015a) which caused the accumulation of pollutants in Beijing.Furthermore, the previous source regions totally shifted from northwesterly to southerly.The air flows mainly originated from the heavy industries and populated regions in Jing-Jin-Ji zone, where there were many anthropogenic emissions such as fossil fuels, biomass burning, coal combustion in winter, resulting the concentration of PM 2.5 increasing greatly.On Dec10 (Fig. 5(d)) the high surface pressure persistently disintegrated and the path of air flows became longer again and the direction started to turn to the southwest of Beijing, thus the PM 2.5 mass concentration decreased in Beijing.
On the top of the boundary layer (1500 m), the influence of the cold anticyclone became weaker so the variation of air-flow trajectories from Dec05 to Dec11 (Fig. 6) wasn't as obvious as that in the near surface layer (500 m).On Dec05 and Dec06 (Figs. 6(a) and 6(b)), the pathway of airflow trajectories was almost the same with that in the layer of 500 m except that most of the air-flow trajectories at 1500 m were longer than the air-flow trajectories at 500 m.On Dec08 (Fig. 6(c)), some air flows passed through the desert regions in Inner Mongolia: Badain Jaran desert, Ulan Buh desert, Kubuqi desert, Tengger Desert and Maowusu Desert, where dust storms frequently occurred (Dillner et al., 2006;Sun et al., 2006;).Though, dust storm often happened in winter and spring in North China, the dust storm effect the air quality in Beijing indirectly.The air masses passed those sand sources and carried the fine particles in the dust to Beijing in winter.In other words, the fine particles in dust storm were carried to Beijing by long distance transportation (Wang et al., 2016).And the other air masses were mainly from Shaanxi, Shanxi and Hebei province which would carry more crustal elements from anthropogenic emissions to Beijing.On Dec10 (Fig. 6(d)), the air-flow trajectories continued turned to the west and northwest of Beijing.The longest-range air flow was originated from the middle of Kazakhstan.And the others covered a wide region including Hebei, Shanxi, Henan, Hubei, Shaanxi, Sichuan, Ningxia, Gansu, Xinjiang province and Inner Mongolia.Therefore, the air flows at 1500 m could also carry dust of desert regions and the pollutants of North China Plain to Beijing to aggravate the severe haze.
On the free troposphere (3000 m) (Fig. 7), the temperate westerlies were the dominated weather system.On Dec05 and Dec06 (Figs.

Potential Sources
The backward trajectories can identify the movements of air masses but the trajectories can't identify the major potential sources.In order to find the main sources of PM 2.5 in Beijing during this severe short air pollution, we applied 3D cluster analysis to analyze the main backward clusters results and tried to identify the potential sources.The clustering pathways and a summary of essential cluster characteristics in different heights were given in Fig. 8 and air-flows types were labeled according to their overall direction and relative length.Note that short trajectories were indicative of slow moving air masses while extremely long trajectories did so for fast flows, which has a difference in direct implications for air quality (Perrone et al., 2013).The clustering results were significantly different in the three heights.
In the near surface layer (500 m) (Fig. 8(a)), three trajectory clusters of air-flow were found: long northwest (NW L , cluster 1 (500 m)), short northwest (NW S , cluster 2 (500 m)) and west (W, cluster 3 (500m)), which accounted for 14.29%, 28.57% and 57.14%, respectively.Those clusters were the most important back-trajectories that had a big influence on the severe winter haze in Beijing.The cluster NW L rose from the northeast border between Russia and Kazakhstan in 900 hPa, and then reached the borders between Mongolia and Inner Mongolia in 700 hPa and finally started to fall in Hebei province in 1000 hPa, having a relative smaller contribution due to the extremely long path.However, the shorter cluster NW S could carry the anthropogenic emissions to aggravate the air pollution in Beijing when the air-flow fell in the south Hebei province (750 hPa) after blocking by high mountains (the Yinshan Mountain and Da Hinggan Mountains) and created an arc movement back to Beijing (1000 hPa) at last.Besides, the cluster W originated from the near surface layer on the Badain Jaran Desert (950 hPa), and then crossed the arid and semi-arid grasslands in Inner Mongolia, so it could carry natural soil dust from the desert to Beijing since the soil dust were one of the main sources of PM 2.5 in several regions (Hwang and Hopke, 2007;Alolayan et al., 2013;Zhang et al., 2013).Two clusters were grouped in the layer of 1500 m (Fig. 8(b)), of which cluster 1 (1500 m) was regarded as a long-range flow (NW L ) from the northwest of Beijing and accounted for 28.57%, while cluster 2 (1500 m) (71.43%) was a relative short-range flow (NW S ).The cluster NW S originated from the middle of Mongolia in the relatively low height might carry dust mixed with more pollutants emitted from combustion on their pathways including Shanxi and Hebei province where large amounts of coal were used for heating.As a result, although the air-flows in the layer of 1500 m were relative longer compared with the air-flows in the near surface layer, they were also the contributors of long-range transportation of PM 2.5 .
In the layer of 3000 m (Fig. 8(c)), there was only single trajectory cluster.The cluster originated from 550 hPa in Siberia and went down, gradually.However, this air flow had a longer-range transport at the high height and was relatively clean.

Potential Sources Contribution
In (Fig. 9), the WPSCF was applied to identify the contribution of the potential sources by combining with each source apportionment values.The areas with warmer colors showed the higher WPSCF values and could be regarded as the main potential sources, which had a significant influence on PM 2.5 in Beijing.By contrary, the areas with cold colors represented small WPSCF values and could be donated as unimportant potential sources, which had small impact on PM 2.5 .
From Fig. 9, the highest WPSCF values outside Beijing were almost more than 0.5 and most of them were more than 0.7 and parts of them even reached to 1, which indicated that those places around Beijing were the main potential source areas.The potential source areas determined from the WPSCF analysis coincided with the emissions of PM 2.5 in the larger North China Plain, because there were generally dense populations in the south area of Beijing and numerous coal chemical industries in western Beijing.Moreover, some specific areas were highlighted as having greater potential contributions for long-range PM 2.5 transport during the short severe haze event.The areas with a WPSCF values greater than 0.7 were in Hebei province because the development of industry in Hebei province releasing a lot of anthropogenic contaminants (for example: aerosol particles) and those pollutants could be carried by south prevailing winds to Beijing.Besides, some areas in Shandong province and Tianjin could not be ignored

CONCLUSIONS
This severe winter haze in Beijing has lasted for five days.The PM 2.5 air pollution was heavier in the south of Beijing than in the north of Beijing generally.Furthermore, the air pollution dispersed from the south of Beijing.
Through the factor analysis, the calm and steady meteorology condition was one of the main factors of the haze event under the surface cold anticyclone.The PM 2.5 concentrations had a positive correlation with RH (0.819), but had a negative correlation with the BLT (-0.502) and WS (-0.164).
Through the backward trajectories analysis, the air-flows in different height (500 m, 1500 m and 3000 m) had different effects on the severe haze in Beijing.The air flows in the boundary layer (500 m) had the largest influence on the PM 2.5 concentration.The air flows on the top of the boundary layer (1500 m) were also the contributors of PM 2.5 long-range transports, but the air flows in the FT height (3000 m) had little influence on the haze event.Comparing the cluster results at 500 m, 1500 m and 3000 m by applying 3Dcluster analysis, we found that the air flows in lower height moving over a source area can carry more pollutants than that in the higher height.However, all the categories at the three heights had the same pathway indicating that the important potential source regions were the desert in Siberia, Mongolia and Inner Mongolia, and the industrial and urban areas in Jing-Jin-Ji zone especially Hebei province.The higher WPSCF values (> 0.7) were mainly distributed in Hebei, west Shandong province (around 0.5) and south Tianjin (around 0.5) in China, which suggested that those areas had the main contribution to this severe haze event in Beijing.