Characteristics and Meteorological Factors of Severe Haze Pollution in China

A severe haze pollution incident caused by unfavorable weather conditions and a northern air mass occurred in eastern, northern, northwestern, and southwestern China from January 15 to January 22, 2018. To comparatively analyze variations in PM2.5 pollution, hourly monitoring data and 24 hmeteorological data were collected. Air quality observations revealed large spatiotemporal variation in PM2.5 concentrations in Handan, Zhengzhou, Xi’an, Yuncheng, Chengdu, Xiangyang, and Jinan. ,e daily mean PM2.5 concentrations ranged from 111.35 to 227.23 μg·m3, with concentration being highest in Zhengzhou. Hourlymean PM2.5 concentration presented multiple U-shaped curves, with higher values at night and lower values during the day. ,e ratios of PM2.5 to PM10 were large in target cities and the results ofmultiscale geographic weighted regressionmodel (MGWR) and Pearson correlation coefficients showed that PM2.5 had a significant positive or negative correlation with PM10, CO, NO2, and SO2. ,e concentration of PM2.5 was closely related to the combustion of fossil fuels and other organic compounds, indicating the large contribution of secondary aerosols to PM2.5 concentrations. ,e analysis of meteorological conditions showed that low temperature, low wind speed, and high relative humidity could aggravate the accumulation of regional pollutants in winter. Northwestern trajectory clusters were predominant contributions except in Jinan, and the highest PM2.5 concentrations in target cities were associated with short trajectory clusters in winter. ,e potential sources calculated by Weight Potential Source Contribution Function (WPSCF) and Weight ConcentrationWeighted Trajectory (WCWT)models were similar and the highest values of theWPSCF (>0.5) and theWCWT (>100 μg·m3) were mainly distributed in densely populated, industrial, arid, and semiarid regions.


Introduction
Severe haze pollution events occur frequently in winter in China and are dominated by PM 2.5 (particulate matter with aerodynamic diameters no larger than 2.5 µm). Particulate matter pollution adversely affects human health and atmospheric visibility [1][2][3]. In addition, the direct and indirect radiation lead to climate change and disturb the structure and function of ecosystems [4,5]. In densely populated and industrially developed areas, daily PM 2.5 concentrations in winter frequently exceed the Class II category of the National Ambient Air Quality Standards (NAAQS, 75 μg m − ).
by Jiang et al. [7] with a positive matrix method showed that the average annual concentrations of PM 2.5 and PM 10 were the highest in winter and the lowest in summer. Niu et al. [8] studied the temporal and spatial variation and chemical composition of PM 2.5 in the Guanzhong Plain from March 2012 to March 2013. e average daily PM 2.5 concentration was 134.7 μg·m − , exceeding the Class II category of the NAAQS.
Severe atmospheric pollution is closely related not only to emission sources but also to adverse meteorological conditions, terrain, pollutant transport pathways, and chemical reactions in atmosphere [9][10][11]. Human activities and meteorological conditions are the primary factors leading to variation in pollutant concentration [12,13]. For instance, if the aridity index and average annual temperature increased by 1%, PM 2.5 concentrations would increase by 66.9 and 35.7, respectively [14]. But PM 2.5 -heavy pollution is often accompanied by high relative humidity [15,16]. Moreover, stable weather conditions, low temperature, and wind speed can also aggravate the accumulation of regional pollutants in winter. PM 2.5 and PM 10 become very high in the postmonsoon season in Kolkata; PM concentrations are observed to be the lowest during the monsoon seasons; meanwhile, the NO 2 and CO concentrations demonstrate similar seasonal fluctuations [17].
Long-distance transportation among regions also plays an important role in PM 2.5 pollution. In recent years, the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model and the TrajStat model [18] have become critical tools for studying long-distance transport and potential pollutant sources. Many scholars have traced the transmission path of CO, O 3 , SO 2 , PM 2.5 , PM 10 , and other gaseous pollutants using HYSPLIT and TrajStat models [19][20][21]. Filonchyk and Yan [22] quantitatively investigated the causes of severe haze during spring and winter seasons in northwest China based on the backward trajectory and the HYSPLIT model. e results showed that the movement of air masses in the north, northwest, and west of China is the main cause of haze during spring and winter in northwest China. However, these studies only focused on the transport direction and passing area in the near surface and did not carry out statistical analyses of PM 2.5 concentrations in backward trajectories and clustering trajectories.
Previous studies have focused on the spatial and temporal distribution of PM 2.5 in a single city as well as influence factors and potential sources, which often neglects the impact of pollutant transport among cities. To better understand the impact of pollutant transport in different cities, real-time pollutant data and meteorological data of major Chinese cities were collected in this research during the haze period in January 2018. We investigated the correlations between air pollution and meteorological conditions and their spatial variation. We computed Weight Potential Source Contribution Function (WPSCF) and Weight Concentration-Weighted Trajectory (WCWT) models to quantify potential source distributions in different cities. is research aimed to provide a reference for the local government to manage sudden air pollution incidents, propose a method for studying long-distance transport, and identify potential sources of air pollutants in PM 2.5 haze pollution.

Air Quality Observations and Data Quality.
Observational pollution data from January 15 to January 22, 2018, were used in this study. Real-time data were provided by the China National Environmental Monitoring Center (http://www.cnemc.cn) after being validated, with hourly concentrations of six major pollutants: PM 2.5 , PM 10 , sulfur dioxide (SO 2 ), nitrogen dioxide (NO 2 ), carbon monoxide (CO), and ozone (O 3 ). We calculated daily mean concentrations of PM 2.5 in seven cities (Handan, Zhengzhou, Xi'an, Yuncheng, Chengdu, Xiangyang, and Jinan) (Figure 1(b)). We used the threshold of 75 μg·m − as the highest PM 2.5 concentration for acceptable air quality according to the Class II category of the NAAQS (GB3095-2012). Mean PM 2.5 concentration in 30 provincial capitals and seven cities was calculated with data from 302 monitoring sites during the pollution period (Figure 1(c)). As shown in Table 1, each city had set up several air quality monitoring sites, most of which were located in urban areas and some in suburban and rural areas as background sites. Daily mean PM 2.5 concentrations were calculated when valid data were available for more than 20 h during the day and values were greater than zero [23]. We adopted the same method as the government reports daily concentrations of air pollutants to the public, averaging the concentrations at all sites in each city to represent the daily mean concentration of the city.

Meteorological Data.
Hourly meteorological data from January 15 to January 22, 2018, were obtained from the National Meteorological Information Center of the China Meteorological Administration (http://data.cma.cn/) and used to analyze the relationship between meteorological conditions and air pollution. Hourly meteorological data included temperature, relative humidity, 2 min wind speed, 2 min wind direction, and sea-level pressure. Mean values of each meteorological parameter were calculated using data for all monitoring sites in a region.

Multiscale Geographically Weighted Regression.
MGWR is a significant improvement on GWR because it allows one to study relationships at varying spatial scales and covariate-specific bandwidths to be optimized [24]. MGWR can be defined as where (u i ,v i ) represents the geographical coordinates of i city, β bwi is the regression coefficient, and ε i represents a random error term. We used the MGWR2.2 software to undertake all calibrations (https://sgsup.asu.edu/sparc/ mgwr). e spatial kernel function type is Bisquare, the bandwidth search type is Golden, and the parameter initialization type is GWR estimation [25].

Inverse Distance Weighted
Interpolation. e inverse distance weighted interpolation method is based on the principle of similarity. e spatial variation of PM 2.5 mass concentration in China during the pollution period was drawn in ArcGIS 10.3. We used the coordinate information of 1,436 air quality monitoring stations in 338 cities as the "input point feature" and used daily PM 2.5 mass concentration data at each site as the Z value field. We set the maximum number of adjacent features and the minimum number of adjacent features to 15 and 10, respectively.

Backward Trajectory Statistics and Calculation.
Trajectory cluster calculation was carried out with TrajStat [26] http://www.meteothinker.com/downloads/index.html). TrajStat provides two clustering options, Euclidean distance or angle distance. In this study, we used angle distance as we intended to use the backward trajectories to determine how the air mass reached the center of the monitoring point. e angular distance between two backward trajectories is defined as 12 is the mean angle between two backward trajectories. e variables X 0 and Y 0 define the position of the study site.

WPSCF Analysis.
e Potential Source Contribution Function (PSCF) algorithm identifies source regions based on airflow trajectories analysis and has been widely used to identify potential source areas for high-concentration pollutants at receptor sites [27]. e area covered by the

Advances in Meteorology 3
backward trajectory is divided into equal i × j grid cells. e PSCF value for the ij th cell is defined as where n ij is the total number of trajectory endpoints that fall in the ij th grid cells and m ij is the total number of trajectory endpoints for which the monitored pollutant concentration exceeds a threshold value in the cells (Kong et al., 2013; [28]). In this study, the grid cell size was 0.5°× 0.5 latitude-longitude and we defined 75 μg·m − as the threshold value of PM 2.5 mass concentration. To account for uncertainty, PSCF values were multiplied by an arbitrary weight function W ij [29,30]. e weighting function reduced PSCF values when the total number of the endpoints in a cell was fewer than three times the average number of endpoints for all cells. We calculated WPSCF values to identify the possible source areas of PM 2.5 in a region.

WCWT Analysis.
e PSCF method calculates the proportion of pollution trajectories in a grid, reflecting the potential influence of the grid on the receptor site. Whether pollutant concentrations at the monitoring site are only slightly higher or much higher than the criterion, grid cells have the same PSCF value and it can be difficult to distinguish moderate pollution sources from major sources. In the Concentration-Weighted Trajectory (CWT) method [31], each grid cell is assigned a weighted concentration by averaging the sample concentrations that have associated trajectories crossing that grid cell as follows: where C ij is the average weighted concentration in the ij th cell, l is the index of the trajectory, M is the total number of trajectories, C l is the concentration observed on the arrival of trajectory l, and τ ijl is the time spent in the ij th cell by trajectory l. e influence coefficient W ij is also used in the WCWT method (WCWT ij � C ij × W ij ).  Northern cities receive relatively weak effects from monsoons, slower wind speeds, and small amount of precipitation, resulting in drier air and more sandstorms, which more often occur in winter, so it is easier for pollutants to enter the air. In contrast, lower temperatures are not beneficial to pollutants diffusion, retain pollutants near the surface, and lead to high concentrations [32]. A similar phenomenon was observed in the previous study [33]. Table 2 shows the average concentrations of the air quality index (AQI) and variations in six major pollutants. e areas with the most severe air pollution (AQI ≥ 115) were distributed in central, northwestern, and northeastern China. ese patterns may be caused by excessive pollutant emissions from coal-fired heating, biomass burning, and industrial combustion in winter [34,35]. e AQI of Shijiazhuang, Jinan, Taiyuan, Zhengzhou, Xi'an, and Chengdu reached 170.5, 200.5, 116.2, 260.0, 228.8, and 117.1, respectively. And the air quality levels reached a moderate or severe pollution grade, which have significant impacts on human health. Figure 2 shows the spatial distribution of daily mass concentrations of PM 2.5 during the pollution period. Elevated concentrations were mainly located in central, northwestern, and eastern China and concentrated in southern Hebei, western Shandong, southern Shanxi, northern Henan, central Shaanxi, and central Sichuan because of the high emissions from fossil fuel combustion and adverse weather conditions. Overall, PM 2.5 pollution in central and eastern China is more severe than in northern China due to distinct emissions sources, weather features, and source apportionment [36]. PM 2.5 pollution in Handan (Hebei Province), Zhengzhou (Henan Province), Xi'an (Shaanxi Province), Yuncheng (Shanxi Province), Chengdu (Sichuan Province), Xiangyang (Hubei Province), and Jinan (Shandong Province) was highest among major polluted cities, with daily mean PM 2.5 concentrations of 212.6 ± 95.5 μg·m − , 227.2 ± 109.9 μg·m − , 186.7 ± 59.0 μg·m − , 207.8 ± 83.1 μg·m − , 111.4 ± 45.4 μg·m − , 219.3 ± 87.8 μg·m − , and 163.5 ± 64.7 μg·m − , respectively.

Daily Mean PM 2.5 Concentrations.
Daily mean PM 2.5 concentrations in the seven major polluted cities in Figure 3(b) exceeded the Class II category of the NAAQS. Hourly data were used to examine daily variability in PM 2.5 and identify potential emission sources [37]. Trends in hourly mean PM 2.5 concentrations in polluted cities were similar and exhibited multiple U-shaped curves, with higher values in the early morning (0:00-05:00) and at night (19: 00-23:00) and lower values during the middle of the day (12: 00-15:00). Such daily patterns can be explained by enhanced emissions from heating, unfavorable meteorological conditions, and variations in topography.
A comparative study of seven major cities found that the PM 2.5 hourly concentrations showed a steady trend from day to night in Xi'an and Chengdu, which can be explained by the static stability of the atmosphere [38]. During the pollution episode, the mean wind speed was 1.65 and 1.62 m·s −1 in Xi'an and Chengdu, respectively, illustrating that the weak wind was conducive to the diffusion of pollutants. e temperature was 2.2°C and 9.3°C in Xi'an and Chengdu, respectively, and relative humidity was 61% and 87%, respectively. us, low temperature, low wind speed, and high relative humidity may have led to the accumulation of PM 2.5 in winter.
High concentrations of PM 2.5 appeared at midday (12: 00-13:00) in Handan and Yuncheng (Figure 3(a)), which can be explained by high emissions from coal heating, cooking, and transportation [4,37,39]. e lowest PM 2.5 concentrations were observed in the afternoon, when the boundary layer becomes larger and the wind speed increased. After 17 : 00, PM 2.5 concentrations started to increase in Handan, Zhengzhou, Xi'an, Yuncheng, and Xiangyang because of decreasing wind speed and increasing vehicle emissions. PM 2.5 pollution emitted from diesel truck traffic, which is allowed only during nighttime, additionally increased PM 2.5 burden because the emission factors of heavy-duty vehicles are six times higher than those from light-duty vehicles [40].

Correlations between PM 2.5 and Other Gaseous Pollutants.
e Pearson correlation coefficient (r) was used to investigate the relationship between PM 2.5 and PM 10 , CO, NO 2 , O 3 , and SO 2 using hourly data (Table 3). e analysis results showed a strong positive correlation (r > 0.9) between PM 2.5 and PM 10 in the seven polluted cities, indicating that a significant fraction of the PM 2.5 was secondary PM, such as ammonium sulfate, secondary organic aerosol, or fugitive dust, which typically have broader regional distributions than anthropogenic primary pollutants [39]. We found large values of the ratio of PM 2.5 to PM 10 in Handan (0.67), Zhengzhou (0.79), Xi'an (0.70), Yuncheng (0.65), Chengdu (0.68), Xiangyang (0.87), and Jinan (0.64), which indicates the large contribution of secondary aerosols to PM 2.5 concentration in these regions. e ratio of PM 2.5 to PM 10 also shows that PM 2.5 is the main component of PM 10 . Similar results were reported in the previous study [41]. A positive correlation between PM 2.5 and CO was observed (r > 0.6), which revealed that the CO emission process is accompanied by the emission of fine particles. e correlation coefficients between PM 2.5 and NO 2 and SO 2 in Handan were very high (PM 2.5 and NO 2 : r � 0.88; PM 2.5 and SO 2 : r � 0.64). is may be caused by the large amount of emissions from power plants, urban dust, and the combustion of fossil fuels. In Chengdu, the correlation coefficients of PM 2.5 with NO 2 and SO 2 were 0.55 and 0.64, respectively, mainly due to adverse weather conditions restricting diffusion and chemical conversion of traffic pollutants. However, correlations between PM 2.5 and NO 2 , O 3 , and SO 2 were lower in other cities (Table 3).
To further discuss the relationship between PM 2.5 and other pollutants, we investigated the impact of other pollutants on PM 2.5 by using the statistical advantage of the MGWR model that each regression coefficient was based on Advances in Meteorology   Tables 4 and 5. In terms of the number of effective parameters, from the analysis of global regression results, the goodness of fit (R 2 ) was 0.908 (p < 0.05), and the residual sum of squares (RSS) was 51.56 (Table 4). According to the local regression results, the local R 2 of all selected cities exceeds 0.80 (p < 0.05). ese regression results showed that the MGWR model uses fewer parameters to get the regression results closer to the true value, which could be used to evaluate the relationship between PM 2.5 and other pollutants. It can be clearly found from

Meteorological Condition Analysis.
e multiscale interaction of meteorological conditions affects air quality in a complex way [13]. Previous studies have shown that meteorological factors play an important role in the daily variation of pollutant concentrations [13].
e Pearson product-moment correlation coefficient between the hourly mean concentrations of major pollutants and local meteorological parameters (wind speed temperature, relative humidity, and sea-level pressure) in the seven cities is shown in Figure 4. In general, the correlation coefficients showed little difference among cities, indicative of regional pollution characteristics [20]. PM 2.5 , CO, PM 10 3 and temperature, which are affected by atmospheric turbulence and influence regional pollutant concentrations [3]. Generally, pollutant concentrations decrease with increasing wind speed. Deng et al. [42] and Wang et al. [43] found that low primary pollutant concentrations result in high O 3 concentrations and cause a positive correlation between wind speed and O 3 concentration. We found a positive correlation between O 3 concentration and wind speed in Handan, Xi'an, and Jinan (Figure 4(b)). We also found a negative correlation between pollutant concentrations and wind speed in Zhengzhou and a positive correlation between PM 2.5 , CO, PM 10 , NO 2 , and O 3 with wind speed in Xi'an, possibly because strong winds can stir up dust. O 3 concentration was weakly positively correlated with relative humidity in the seven cities (Figure 4(c)). Primary pollutant concentrations were negatively correlated with relative humidity in Handan, Yuncheng, Xiangyang, and Jinan and positively correlated in other cities (Figure 4(c)), which indicates that lower relative humidity is unfavorable for scrubbing gaseous pollutants. Pollutant concentrations were negatively correlated with sea-level pressure in all cities except Xiangyang, where a positive correlation was observed, possibly for two reasons. First, the atmosphere would be stable under the low air pressure, leading to the inversion layer taking place easily and the air convection slowing down, thus resulting in higher concentrations of atmospheric pollutants. e increase in air pressure may have led to the enhancement of air advection as well as an increase in wind speed, which plays a positive role in the diffusion of air pollutants. PM 2.5 concentrations in the study cities were influenced not only by local emissions but also by the surrounding pollution region. To further study the influence of wind speed and direction on pollutant diffusion, the wind rose and the distribution of hourly PM 2.5 concentrations, wind speed, and wind direction in the target cities were calculated (Figures 5 and 6). During the pollution episode, the lowest wind speeds were found in Handan. And the wind speeds associated with north and northwesterly winds were low in Handan, Zhengzhou, and Yuncheng, while weak winds were associated with the south and southeasterly directions in Xi'an, Chengdu, and Xiangyang. In addition, east winds were weak in Jinan ( Figure 5). As shown in Figure 6, high PM 2.5 concentrations were associated with weak winds, less than 2-3 m·s −1 . Previous studies have shown that low wind speeds can stimulate the accumulation of gaseous pollutants [44], indicative of the influence of regional transport from the surrounding polluted regions.

Statistical Analysis of Trajectory Clustering.
Cluster analysis is a widely applied multivariate statistical analysis technique. According to the similarity principle of trajectory space, large numbers of backward trajectories are divided into different transport groups or clusters [45]. e calculated backward trajectories were divided into five main trajectory clusters from the total spatial variance using the HYSPLIT and TrajSat models. Main transport pathways were divided into nine categories according to the results of the trajectory clusters (Figure 7): long northwest (LNW: Handan C2, Zhengzhou C3, C4, and C5, Xi'an C2 and C3, Yuncheng C2, C3, and C4, Xiangyang C3 and C5, and Jinan C3 and C4), short northwest (SNW: Xi'an C4 and C5 and Jinan C1), long north (LN: Handan C3), short north (SN: Handan C4 and C5, Zhengzhou C2, Chengdu C4, and Xiangyang C4), short southwest (SSW: Handan C1, Zhengzhou C1, Xi'an C1, and Jinan C2 and C5), long southwest (LSW: Yuncheng C1 and C5 and Xiangyang C2), long west (LW: Chengdu C1 and C3), short east (SE: Chengdu C2), and short south (SS: Chengdu C5 and Xiangyang C1). e northwestern trajectory clusters (LNW and SNW) passing through some natural sources of aerosol emissions including northern Xinjiang, southern Inner Mongolia, northwestern Gansu, and central Shannxi were predominant and accounted for dominant trajectories 72.9%, 55.1%, 80.2%, 90.6%, 38.5%, and 41.2% of clusters, respectively, in Handan, Zhengzhou, Xi'an, Yuncheng, Xiangyang, and Jinan. LW pathways made a large contribution of 53.1% in Chengdu. Particle matters accumulated more in short trajectories. As Table 6 shows, the highest mean PM 2.5 concentrations were associated with SSW (Handan C1, Xi'an C1, and Jinan C2 and C5), SN (Zhengzhou C2 and Xiangyang C4), LSW (Yuncheng C1 and C5), and SS (Chengdu C5) pathways. Our results are consistent with those from a study from Perrone et al. [46], which found that a longer airflow trajectory had a faster speed not conducive to particle deposition according to the principle of dynamics. Figures 8 and 9 show the 72 h backward trajectories and potential sources of PM 2.5 with WPSCF and WCWT. e warm-colored areas of the map represent the main potential sources of pollution that had a significant effect on PM 2.5 concentration.

Source Analysis.
e cold-colored regions represent minor potential sources of pollution. Figures 8 and  9 show areas with WPSCF > 0.5 and WCWT > 100 μg·m − (i.e., where main potential sources were concentrated). PM 2.5 pollution in these regions was mainly caused by artificial emissions. In areas with WPSCF < 0.5 and WCWT < 100 μg·m − (i.e., secondary potential sources), PM 2.5 pollution was related to the presence of arid and semiarid areas and desert areas, such as the Badain Jaran Desert, the Ulan Buh Desert, the Kubuqi Desert, and the Tengger Desert in Inner Mongolia, where dust storms occur frequently [47,48]. e identity and distribution of the main potential sources of PM 2.5 in each polluted city differed. Our results suggest that regional transport from the northwest and south of Handan plays a dominant role in the formation of pollution events. Pollution sources were mainly distributed in southern Hebei, south of Shanxi, and north of Shaanxi, Advances in Meteorology 9 which are the main coal-burning areas in winter.  distributed around Xi'an and were caused by motor vehicle emissions, thermal power plants, and coal burning for heating in winter. Ningxia and Gansu may also be important potential sources of PM 2.5 pollution in Xi'an since these areas are the main sources of dust in winter. Central Gansu, Ningxia, and Shaanxi were the main potential sources of pollution in Yuncheng. e high potential source regions for Chengdu were located in surrounding areas due to unfavorable meteorological conditions in winter. High WPSCF and WCWT values were mainly concentrated in the south of Henan where population density and biomass burning are high. e high potential source regions for Jinan were distributed in southwestern Shandong, northeastern Henan, east of Shandong, and south of Hebei [50].   Figure 7: Spatial distribution of cluster means in seven Chinese cities, with the large red dots representing cities.

Conclusions
Air pollution is a serious environmental and societal problem in China. is study mainly analyzed the characteristics of a severe air pollution incident that occurred from January 15 to January 22, 2018, in seven major polluted cities in China. We also analyzed the effects of meteorological conditions and identified potential pollution sources. is study provides an important scientific basis for the design of pollution control strategies in this region. e most serious air pollution was observed in Handan, Zhengzhou, Xi'an, Yuncheng, Chengdu, Xiangyang, and Handan, Xi'an, Zhengzhou, and Yuncheng and PM 2.5 , PM 10 , CO, and SO 2 concentrations were positively correlated with temperature, while the opposite occurred for NO 2 and O 3 concentrations in Handan, Xi'an, Zhengzhou, Yuncheng, and Jinan. Correlations with relative humidity showed significant regional differences. PM 2.5 , PM 10 , CO, and SO 2 were negatively correlated with relative humidity in Handan, Yuncheng, Chengdu, and Jinan, while NO 2 and O 3 were negatively correlated with wind speed. We identified calm meteorological conditions as one of the main factors causing the haze event. e analysis of transport contributions indicated that the northwestern trajectory yielded the greatest PM 2.5 contributions, ranging between 41.2% and 90.6% in target cities, whereas the highest mean PM 2.5 concentrations were generally associated with SSW, SN, LSW, and SS pathways. e potential pollution sources calculated by WPSCF and WCWT models were very similar and the highest values of WPSCF (>0.5) and WCWT (>100 μg·m − ) were distributed in densely populated and industrial areas.

Data Availability
Data used in this paper can be obtained from Chao He (he_chao@whu.edu.cn) upon request.

Conflicts of Interest
e authors declare that they have no conflicts of interest.