Satellite observation of regional haze pollution over the North China Plain

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2012jd017915, 2012 Satellite observation of regional haze pollution over the North China Plain Minghui Tao, 1,2 Liangfu Chen, 1 Lin Su, 1 and Jinhua Tao 1 Received 9 April 2012; accepted 29 April 2012; published 21 June [1] Haze clouds often form over the North China Plain (NCP) of eastern China, where large amounts of aerosol particles and their precursors are emitted. To obtain general insights into regional pollution, a large-scale, long-term study was conducted using A-Train satellite observations, ground measurements, and meteorological data. Contrary to previous analyses, most of the haze clouds appeared to form abruptly (within 2 3 h). Case studies show that natural sources contribute significantly to the formation of regional haze. Dust plumes can mix with local pollutants, causing smog clouds to form abruptly, while moist airflows can cause widespread haze-fog pollution. The combined observations revealed highly inhomogeneous haze clouds, in terms of both vertical and horizontal distribution, leading to clear discrepancies between site measurements near the surface and satellite observations at the top of the atmosphere. Surprisingly, prevailing dust plumes, which are closely connected with the haze clouds, were observed in winter. Airborne dust and water vapor transported from outside the region are the main drivers of regional haze over the NCP. Accumulation of local pollutants also leads to common occurrences of urban smog; however, the occurrence of most haze clouds shows no obvious correlation with local pollution. Local- and regional-scale haze pollution are common over the NCP, but they have differing formation mechanisms, and contrasting chemical and physical properties. The present findings improve our understanding of heavy pollution over eastern China and its links to climate. Citation: Tao, M., L. Chen, L. Su, and J. Tao (2012), Satellite observation of regional haze pollution over the North China Plain, J. Geophys. Res., 117,, doi: /2012jd Introduction [2] Haze is defined as a weather phenomenon where atmospheric visibility is less than 10 km, resulting from dense layers of aerosol particles such as acidic chemicals, organics, black carbon (BC), and fly ash as well as dust accumulated in the air, and with a lower relative humidity (RH) than fog. These small particles exhibit large spatial and temporal variation, have diverse optical properties, and represent a major uncertainty in climate research because of their significant impacts on the Earth s radiation balance and hydrologic cycle [Intergovernmental Panel on Climate Change, 2007; Kaufman et al., 2002; Ramanathan et al., 2001]. By scattering and absorbing solar radiation, haze layers can significantly cool the surface of the Earth while heating the atmosphere, which in turn affects convection and 1 State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing Applications, Chinese Academy of Sciences and Beijing Normal University, Beijing, China. 2 Graduate University of Chinese Academy of Sciences, Beijing, China. Corresponding author: L. Chen, State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing Applications, Chinese Academy of Sciences, Beijing , China. (minghuit@gmail.com) American Geophysical Union. All Rights Reserved /12/2012JD stability in the low troposphere [Menon et al., 2002; Ramanathan et al., 2007]. Heavy loading of aerosols can alter cloud properties and their lifetimes over polluted regions, and weaken monsoons [Lau and Kim, 2006; Li et al., 2011b]. Moreover, haze particles near the surface affect air quality and are a public health concern [Hoek et al., 2010]. [3] Since the late 1990s, a widespread haze layer, named the atmospheric brown cloud (ABC), has been observed over south Asia. This layer also exists in other regions, such as southern Africa, the Amazon Basin, and eastern China [Chameides et al., 1999; Ramanathan et al., 2007]. While aerosols in haze clouds have a strong influence on the radiation budget, surface evaporation, and monsoon rainfall over south Asia [Ramanathan et al., 2001, 2005], aerosol loading and optical properties vary greatly in space and time, as well as under different meteorological conditions [Gustafsson et al., 2009; Haywood et al., 2008; Li et al., 2011a]. Therefore, continuous and large-scale observations of all major sources of aerosols are essential for understanding their general climatic effects over any region, and for fully constraining numerical models of climate patterns [Kaufman et al., 2002; Li et al., 2007]. [4] Haze clouds often form over the North China Plain (NCP) of eastern China during the dry season (from October to March) and have been frequently detected by satellites (NASA Earth Observatory Natural Hazard) and ground 1of16

2 Figure 1. (left) Area location and (right) map of the NCP. The red square frame indicates location of the NCP and its surrounding terrain. observations [Lee et al., 2010; Ma et al., 2010]. As this is one of the world s most populated and fastest developing regions, excessive emissions of aerosol particles and their precursors lead to a high loading of pollutants, with marked radiative effects [Li et al., 2007; Lu et al., 2010; Wang et al., 2011]. Some intensive field studies, such as the East Asian Study of Tropospheric Aerosols and their Impact on Regional Climate (EAST-AIRC) [Li et al., 2011a] and Haze in China (HaChi, ) [Xu et al., 2011], have been conducted to understand the properties of these aerosols and their effects on the climate over eastern China. Numerous field studies have achieved substantial progress in understanding the chemical and physical properties of these aerosols [Guinot et al., 2007; Guo et al., 2010; Huang et al., 2010; Li et al., 2011] and their radiative effects [Li et al., 2007], as well as in developing numerical simulations of the effects of aerosols on climate [Chameides et al., 1999; Qian et al., 2009]. [5] Although many studies have examined the chemical and physical properties of haze particles [Huang et al., 2011; Li et al., 2011; Sun et al., 2006], these analyses have focused on particular case events and measurements from ground sites. However, aerosol loading and optical properties vary greatly between different regions due to source heterogeneity [Wang et al., 2011; Zhang et al., 2011]. Indeed, the spatially limited site measurements near the land surface provide information about local sources but may not reveal other sources contributing to the formation of widespread haze clouds. Furthermore, existing satellite studies over eastern China have concentrated on variations in aerosol loading, with less attention being given to haze events [He et al., 2011; Su et al., 2010]. Correspondingly, the general characteristics of the regional haze over the NCP, including essential information about its type, scale, and formation, are still unclear due to a lack of large-scale and continuous comprehensive observations. In particular, little is known about how the haze clouds change over space and time. [6] The recent A-Train satellites (Aqua, Aura, PARASOL, and CALIPSO) provide an unprecedented opportunity to observe haze clouds simultaneously from numerous perspectives [Hsu et al., 2006; Omar et al., 2009; Tanré et al., 2011; Torres et al., 2007]. Here, we present a general overview of variations in haze cloud over the NCP during the dry season using a two-year data set of satellite observations (October 2009 to March 2011) integrated with ground measurements and meteorological data. Different regional pollution events were identified and classified by considering meteorological conditions, synergetic satellite observations, and ground measurements. We analyze the optical properties and the mechanisms underlying the formation of typical regional haze events in detail. Unlike previous analyses, the sources, optical properties, and formation of the haze clouds are examined at a regional scale. Also discussed in detail are the characteristics of haze clouds over the NCP and the differences between urban haze and regional pollution. In doing so, we present the first largescale and long-term comprehensive insights into haze clouds over eastern China. 2. Data and Methods 2.1. General Description of the Study Area [7] This study was carried out during the dry season (from October to March), when pollutants often accumulate over the NCP due to weather conditions [Li et al., 2011]. The NCP is surrounded by mountains in the north and west, with vast deserts in northwestern China and Mongolia (Figure 1). Foggy days are frequent in October, with weak winds. Increased coal consumption during the heating season (November to March) releases abundant particle pollutants and their precursors [Cao et al., 2007]. During this season, northwesterly cold, dry air masses prevail at the top of the planetary boundary layer (PBL), with frequent temperature inversions and stagnant weather near the surface. In addition 2 of 16

3 to the meteorological conditions, mountains in the west and north of the plain, with altitudes of 1 2 km, influence the diffusion of pollution Satellite Data Sets [8] The Moderate Resolution Imaging Spectroradiometer (MODIS), on the morning satellite Terra and the afternoon satellite Aqua, measures reflected solar radiance and terrestrial emissions at 36 wavelength bands with resolutions between 250 m and 1 km. The two satellites pass eastern China twice daily with a swath of 2300 km. Heavy smog, dust storms, and cirrus clouds can be clearly discerned from MODIS true color images. These images represent direct and effective observations of daily variations in air pollution, including changes between morning and afternoon. [9] Both the dark-target and deep-blue aerosol data from the Collection 5.1 Aqua MODIS atmosphere Level 2 aerosol products were used in this study. The dark-target algorithm derives clear-sky aerosol optical depth (AOD) over land surfaces with dense vegetation, mostly within an error range of t [Levy et al., 2010]. The deep-blue algorithm can derive AOD over bright surfaces, such as deserts and areas without vegetation where the dark-target algorithm does not provide retrievals within an error of 0.3 t [Hsu et al., 2006]. The MODIS fire location products are used to identify agricultural biomass burning (ABB) smoke. [10] POLDER-3 was carried on the French microsatellite PARASOL, which is part of A-Train. POLDER can detect fine particles over land with a high surface reflectivity and spatial variability by using polarization information that is primarily obtained from atmospheric scattering processes [Tanré et al., 2011]. The POLDER fine mode AOD (FAOD) at 865 nm over East Asia is well correlated (R 0.92) with AERONET inversions [Su et al., 2010]. The high sensitivity of POLDER to fine aerosols (particle radius 0.30 mm), which are derived mainly from anthropogenic sources, is important in the detection of haze clouds. Although high FAOD values do not denote a low fraction of coarse particles, haze or smog with low FAOD values indicates that the dominant aerosols are coarse, such as dust particles. [11] The OMI sensor aboard NASA s EOS-Aura satellite measures Earth s reflectance in the visible and ultraviolet spectra ( nm) with a wide swath. The UV Aerosol Index (UAI) is a qualitative parameter that detects the existence of UV light-absorbing aerosols on the basis of the interaction between molecular scattering and aerosol absorption [Torres et al., 1998]. Because UAI is highly sensitive only to elevated UV-absorbing aerosols (at least 2 km above the surface), it can identify aerosols such as desert dust or biomass burning smoke with positive values, but returns small or positive values for clouds and weakly absorbing aerosols. The unique property of UAI is its ability to detect aerosol absorption over highly reflective surfaces, such as ice and snow, and even aerosols intermingled with or above clouds [Torres et al., 2007]. The OMI products show a high correlation (R > 0.66) with MODIS observations, especially the UAI [Ahn et al., 2008]. Here, we mainly use the level 2 G OMI UAI, with a 0.25 degree resolution. UAI values can detect properties of thick haze clouds with some sensitivity, especially in the upper part of the aerosol layer. [12] The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) aboard the CALIPSO satellite provides vertical profiles of elastic backscatter at 532 nm and 1064 nm, and of linear depolarization at 532 nm. Dust can be separated from other types of aerosols because its nonspherical particles have a high depolarization ratio [Liu et al., 2008; Omar et al., 2009]. The CALIPSO algorithms use optical properties of the aerosols integrated with surface type and layer altitude to determine the aerosol subtype. Six aerosol models were adopted here, including smoke (biomass burning), dust, polluted dust (dust + smoke), clean continental, polluted continental, and clean marine [Omar et al., 2009]. Comparisons between the CALIPSO aerosol subtype and the AERONET results show high agreement for coarse absorbing (dust) and mixed absorbing (polluted dust) aerosols (91% and 53%, respectively) but low agreement for fine absorbing and non-absorbing aerosols (37% and 22%, respectively) [Mielonen et al., 2009]. However, CALIOP s swath is very narrow (footprint diameter: 70 m) due to lidar detection. In this study, we analyzed the 5 km aerosol layer products through the haze cloud to validate other detection methods and to understand typical layer structures. Due to strong solar radiation during the daytime, only the nighttime data from V.3.01 were analyzed, and the daytime data were used for reference only Ground Measurements and Meteorological Data [13] The Chinese Environmental Protection Agency (EPA) monitors daily primary pollutants (PM 10,SO 2, and NO 2 ) in 86 major cities throughout China. The pollutants are measured in different standardized areas of each city, and at one background station that provides a representative value. Only daily average concentration data are available before All the sites are calibrated and validated within EPA standards. For example, PM 10 concentrations in most cities are measured with Tapered Element Oscillating Microbalance analyzers (TEOMs, model 1400a, Rupprecht & Patashnik, USA). Here, we choose measurements from four megacities (Beijing, Tianjin, Shijiazhuang, and Zhengzhou) in different parts of the NCP (Figure 1) to show typical variations in the local primary pollutants and to improve the satellite-based analysis. [14] Reanalysis data from the National Centers for Environmental Prediction (NCEP) and the Chinese Meteorological Administration (CMA) records were used to analyze weather conditions. The NOAA HYSPLIT trajectory model was employed to trace air masses to the NCP (R. R. Draxler and G. D. Rolph, HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) model access via NOAA ARL READY Web site, NOAA Air Resources Laboratory, Silver Spring, Maryland, 2011, HYSPLIT.php). Analysis of the vertical distributions of temperature and RH at meteorological sites such as Beijing and Zhengzhou can detect moist airflow, temperature inversion, and foggy weather over the NCP ( uwyo.edu/upperair/sounding.html). 3. Regional Observations of Haze Cloud Characteristics Over the NCP 3.1. Synergetic Estimation of Regional Haze Pollution [15] The CMA defines haze as having a visibility of <10 km and RH of 80%, and fog as having a visibility of <10 km and RH of >90%. However, RH usually varies 3of16

4 Figure 2. (top) AQUA MODIS true color images and (bottom) PARASOL FAOD at 865 nm of different regional pollution. (a) Agricultural biomass burning on 4 October (b) Blowing dust on 25 December (c) Clean fog on 1 December (d) Haze-fog on 4 November (e) Haze cloud on 21 January between <70% and >90% over daily timescales during radiation fog events. Ground measurements show that both aerosols and primary gases are present at high levels on hazy and foggy days over the NCP, and the polluted fog has similar radiative effects to haze [Li et al., 2011; Wang et al., 2009]. The EPA often reports such polluted foggy days as hazy-foggy weather. When satellites pass over the NCP at approximately 1:30 P.M. China Standard Time the haze-fog can change to haze during radiation fog events. Here, we consider both haze and haze-fog situations but exclude clean fog-dominant cases based on the CMA s criteria. Compared with urban smog, widespread haze clouds with various emission sources are difficult to measure and validate because surface sites are sparse. Although satellites cannot directly identify haze clouds from meteorological conditions, their synergetic observations provide an effective way to distinguish different events by their optical properties, mode of formation, and spatial and temporal variations. [16] Figure 2 shows samples of different events over the NCP that have been validated. Fog, blowing dust, ABB smoke, haze-fog, and haze show clear differences in MODIS true color images. The MODIS fire count indicated smoke from harvest burning at the beginning of October, when large amounts of fine particles were emitted, with high FAOD values (approximately 0.3) over the fires (Figure 2a). Blowing dust on 25 December 2009 showed an opposite pattern, with a low FAOD (<0.05) over the dust haze due to dominant coarse particles and strong winds. However, ABB was concentrated at the beginning of October and had a limited, although variable, spatial distribution, while intense dust events are mainly concentrated in the spring [Wu et al., 2009]. [17] Compared with notable dust events or ABB smoke, it is difficult for satellites to distinguish fog from haze over the NCP, where fog usually mixes with local pollutants. Figure 2c shows scarce radiation fog in the afternoon after strong winds, with a small amount of blowing dust. The contour of the translucent fog was clear, with a clean background. Larger FAOD values in haze-fog indicate higher levels of anthropogenic pollutants than those in fog (Figures 2c and 2d). Although high FAOD values demonstrate the existence of abundant fine particles, this does not mean that fine mode aerosols are dominant in the dense smog (Figure 2e). To further understand the pollution and to distinguish haze from haze-fog, additional information is required. [18] Due to the large-scale coverage of haze clouds and haze-fog smog, meteorological data from ground sites under these phenomena were used to validate and analyze these regional pollution events. Variations in RH on foggy and haze-fog days were similar, but differed from hazy days (Figures 3a 3c). Radiation fog usually occurs in the early morning near the surface and largely dissipates during the day, whereas the haze layer is thick and associated with relatively steady RH. The A-Train satellites passing over the NCP after midday can miss most of the radiation fog. However, advection fog or moist air masses can last for several days as a thick wet layer. UV Aerosol Index values were clearly different among fog, haze-fog, and haze clouds due to the heights of the layers and the optical properties of the pollutants (Figures 3d 3f). Based on these observations, sensitivity to elevated absorbing aerosols can be a good indication of whether fog is polluted. CALIPSO passes through the haze clouds can also supply information about layer structure using vertical detection. [19] The combined observations integrate both space and ground data, enabling the identification and evaluation of regional pollution. Despite their respective limitations, the different data sets provide comprehensive information for evaluating both spatial variations in, and the optical properties of, haze clouds. 4of16

5 Figure 3. (a c) Vertical variations of RH in Zhengzhou and Xuzhou and (d f) OMI UV Aerosol Index with wind fields at 850 hpa on 1 December 2006, 4 November 2010, and 21 January 2011, respectively Temporal and Spatial Variation of NCP Haze Clouds [20] To obtain an overview of regional haze, multiple data sets covering the dry seasons of 2009 and 2010 were obtained and analyzed. Haze clouds over the NCP exhibited large spatial and temporal variability during this period (Figure 4). To evaluate their spatial characteristics, haze clouds were divided into two groups according to size. Haze covering most of the plain was classified as type L (Large), whereas regional pollution that was concentrated over only part of the NCP was classified as type P (Partial). [21] Large differences in the duration of regional haze were recorded (Figure 5). Unlike local haze resulting from gradual accumulation, abrupt regional smog can cover the NCP within 2 3 h (Figures 5a and 5b) and clear within one day. In contrast, a durable haze-smog can persist for several days (Figures 5c 5e). Accordingly, regional haze with duration of less than one day was classified as type A Figure 4. Aqua MODIS true color images of different haze clouds. 5of16

6 Figure 5. (top) Terra and (bottom) Aqua MODIS true color images of haze clouds in the same day. (Abrupt), and haze events lasting for more than one day were classified as type D (Durative). [22] The classified regional haze events in 2009 and 2010 were further analyzed and compared (Figure 6). The impact of cloud over the NCP was limited during the dry season. Regional haze pollution over the NCP was most frequent and strongest in October, when haze clouds were present for more than 20 days, half of which were of the durative and large-scale types. Surprisingly, abrupt and partial haze was dominant during the heating season, which is unlike the typical pattern of urban pollution [Li et al., 2011]. Regional pollution events decreased in March. Haze clouds did not usually concentrate over urban/industrial regions but instead showed dramatic spatial variation. While detailed analyses were not performed, most regional haze was found to appear Figure 6. Frequency of regional haze events in different types. 6of16

7 Figure 7. (a) MODIS AOD combined dark-target and deep-blue retrievals and 24-h back trajectories in Beijing and Zhengzhou at different altitudes: 500 m (black), 1500 m (red), 2500 m (blue). (b) PARASOL FAOD at 865 nm. or become denser in the afternoon, based on comparisons of Terra and Aqua satellite images. [23] With the exception of March, there were more than 10 days of regional haze events over the NCP in each month of the dry season, and the abrupt and partial type accounted for a major fraction of these events. In certain regions, pollution was not as strong due to the mutable distribution of partial haze. Haze events in the dry seasons of 2009 and 2010 showed different trends. Given that there were no notable changes in local emissions, meteorological conditions and transport would have accounted for the different trends in different years Sample Analysis of NCP Regional Haze Pollution Abrupt Haze [24] Haze clouds developed abruptly over the NCP on 16 December 2008 and 6 November 2010 (Figures 5a and 5b). The Terra true color image shows no obvious regional pollution over the NCP at 11:10 A.M. (China Standard Time) on 16 December. When the Aqua satellite passed over the NCP at 12:50 P.M., dense smog had shadowed most areas of the plain. On the following morning, the smog had cleared. Similar pollution processes occurred on 6 November Such abrupt formation of regional pollution is unlike urban haze, which has been shown to have a clear accumulation process [Li et al., 2011; Ma et al., 2011]. [25] Although dense smog enveloped the ground with an AOD > 1.0, the PARASOL FAOD was moderate ( ) in most areas, except in urban/industrial regions (Figure 7). The fraction of fine particles in the smog was generally limited. High UAI values over the smog indicated the existence of large-scale UV absorbing aerosols, and that the top of the pollution layer was at a high altitude (Figure 8). Back trajectories and wind fields at 850 hpa (approximately 1.5 km) show northwestern air masses from the Gobi deserts. Thus, dust could be the main component of the UV-absorbing haze cloud. The higher FAOD than that in dust storms indicates a mixture of dust particles and local pollutants. Indeed, model calculations demonstrate that acid gases can fully coat dust particles within hours of moving into the polluted region [Ma et al., 2010]. [26] CALIPSO passes through the haze clouds at night reveal the vertical structure and optical properties of the 7of16

8 Figure 8. UV Aerosol Index with 850 hpa wind fields at 14:00 Beijing Time. (a c) Abrupt haze on 16 December 2008, 6 November 2010, and 22 December (d f) Durative haze during January Figure 9. CALIPSO Total Backscatter at 532 nm and Aerosol Subtype on 6 November 2010 and 16 December The black line in the map represents the orbit track. 8of16

9 Figure 10. Temporal variation of primary pollutants (SO 2,NO 2, and PM 10 ) in typical megacities over the NCP between October and December The pink pane indicates durative haze during 7 9 October thick layer (Figure 9). Prominent stratification of the haze layer was observed. There were three main segments in the thick haze layer on 6 November, with a pure dust layer covering the upper region, polluted dust in the middle region, and local industry pollutants concentrated near the land surface. Concentrations of PM 10 showed no obvious variation related to the inhomogeneous structure of the haze cloud (Figure 10), and daily averaging smoothed this abrupt process. Dust plumes seemed to mix well with the local pollutants on 16 December, and the fraction of anthropogenic pollutants increased in the southern part of the NCP in accordance with FAOD values. Ground measurements showed that the superposition of dust and anthropogenic aerosols resulted in much higher AOD values than those in pure dust aerosols [Wu et al., 2009]. The nighttime CALIPSO observations were made approximately 10 h after the daytime A-Train satellite observations, and the corresponding results show that such abrupt haze can last for more than half a day. [27] The spatial distribution of the haze clouds is also highly inhomogeneous. The height of the anthropogenic pollutant layer in urban areas was clearly higher than in rural regions (Figure 9). OMI UAI values were highest in urban/ industrial regions with abundant anthropogenic emissions (Figures 8a and 8b), and soot particles adhering to blowing dust can enhance the absorption of the haze layer [Guo et al., 2010]. The degree of mixing of dust and local pollutants increased from north to south, and can be affected by meteorological conditions, the scale of dust events, and transport pathways Durative Haze [28] Unlike abrupt haze, persistent regional smog settled over the NCP from 16 January 2010 for several days, and was associated with a general accumulation process. The haze pollution gradually spread from the eastern foot of the Taihang Mountains across the whole plain. Meteorological observations in Beijing and Zhengzhou showed RH levels were <72%, and continuous temperature inversions occurred on January. MODIS AOD values from 17 January indicate that heavy aerosol loading had extended to most parts of the NCP (Figure 7). FAOD values increased sharply in urban/industrial regions on the same afternoon, followed by the abrupt appearance of a brown haze over Bohai. Wind fields at 850 hpa showed that air masses from the northwest and southwest met over the north of the NCP during these two days, which may have prevented the diffusion of pollution. The local accumulation and pollutants transported from the southern NCP favored a concentration of haze pollution to the north. However, moderate FAOD values of around 0.15 and UAI values of >1.5 further indicate that coarse particles were dominant over rural regions (Figure 7). Although the FAOD values over Bohai were high, back trajectories suggest that dust from the Gobi deserts led to the abrupt brown cloud, with strong UV absorption. Thus, airborne dust from the northwestern deserts played a significant role in the haze during this period. 9of16

10 Figure 11. Temporal variation of primary pollutants (SO 2,NO 2, and PM 10 ) in typical megacities over the NCP between January and March The pink pane indicates durative haze during January [29] Dense smog completely obscured the land surface on 18 January. The smog was so dense that MODIS aerosol data were missing in several regions (Figure 7). Wind fields and meteorological data show that southerly moist air masses prevailed over the NCP on the morning of 18 January (Figure 8f). Continuous temperature inversions during the haze period then caused a strong accumulation of anthropogenic pollution near the surface (Figure 11). Concentrations of primary pollutants showed strong peaks, and the SO 2 levels in Beijing and Tianjin exceeded 200 mgm 3. The humid airflows caused rapid hygroscopic growth of soluble particles, and high FAOD values ( 0.3) indicate a high concentration of fine particles (Figure 7b). The CMA reported this widespread smog, which had a visibility of <1 km near the surface in many places, as haze-fog. Meteorological sites in Zhengzhou observed a wet layer (RH of >80%) from the surface to an altitude of 1.5 km, while the humid air masses arrived in Beijing above an altitude of 823 m. The main body of this dense smog may have been composed of a mixture of fog and pollutants; UAI values of >1.5 had a similar distribution as the dense smog, indicating intense fog pollution interactions. The occurrence of much higher UAI values over the urban/industrial regions of Beijing and Hebei also indicated a large accumulation of local pollutants, but the heavy pollution was mainly concentrated around the emission sources (Figure 8f). Back trajectories on 18 January showed that air masses from the Gobi deserts and Loess Plateau may have also contributed to this cloud of pollution. [30] Although CALIPSO did not pass the NCP during this period, elevated dust plumes were detected between the deserts and the NCP. Dust transport and the accumulation of local pollutants caused haze to form on January, and the transport of moist air masses on 18 January acted to strongly enhance and prolong the regional haze. Although a background of high-level emissions is essential for the formation of regional haze [Lu et al., 2010], an accumulation of local pollutants as urban haze is not necessary. Another large-scale, durable haze occurred between 7 and 9 October 2010, with generally stable concentrations of SO 2 and NO 2, and a strong peak of PM 10 at widely dispersed sites. This finding suggests that the transport of pollutants from outside the region was dominant (Figure 10), indicating in turn that the transport of dust plumes or moist air masses can play a significant role in the formation of haze clouds over the NCP. 4. Formation Mechanisms of Regional Haze Over the NCP 4.1. Atmospheric Conditions During the Haze Period [31] Meteorological conditions from October 2010 to March 2011 were analyzed to investigate the process of regional pollution formation. Figure 12 shows monthly NCEP wind fields at 850 hpa with an OMI UAI frequency of >1.0. Generally stagnant weather prevailed over the NCP in October and February, during which time durable 10 of 16

11 Figure 12. Monthly wind fields at 850 hpa and frequency of UAI > 1.0 from Oct 2010 to Mar and large-scale haze events often occurred. Examination of the daily wind data shows frequent southerly air masses or very slow wind speeds during these two months. Although the height of the mixing layer and increased emissions both contributed to the high concentrations of primary pollutants during this period, daily variations in surface pollution can be indicative of weather conditions at the bottom of the PBL (Figures 10 and 11). [32] Figure 13 presents the frequency of RH levels of >80% in Beijing and Zhengzhou within 2 km of the surface. Most of the haze-fog days occurred in October and February, when durative haze is frequent. Two different types of fog were found over the NCP during the observation period, classified according to their altitudes. The low type of fog was usually concentrated near the surface (at altitudes of <300 m), while the high type of fog could be up to 1 2 km thick, with an inhomogeneous structure. The low type of fog is likely to be radiation fog caused by the large range in daily temperature, with morning RH levels between 80% and 90% due largely to the mixture of local pollutants. Humid airflows with abundant water vapor can lead to several days of widespread thick fog accumulating over the NCP. The wet layer first appeared at high altitudes (>1 km) and then extended to the surface. Generally, RH levels of >90% appeared at high altitudes while RH levels near the surface were 80 90%. [33] Northwesterly winds passing over the Gobi deserts were dominant at 850 hpa over the NCP during the dry season except October and February, when abrupt haze was frequent. However, numerous peaks in the amount of primary pollutants suggested the common accumulation of local urban pollutants and stagnant weather conditions at the bottom of the PBL (Figures 10 and 11). Although meteorological sites recorded frequent temperature inversions and stagnant weather near the surface, cold and strong winds from the northwestern deserts were dominant in the upper part of the PBL. The weather was generally dry during these four months, with few days having RH > 60%. [34] Although Asian dust storms mainly occur in the spring, the UAI values indicate that dust sources in the Gobi deserts were highly active during the dry season (Figure 12). Sun et al. [2001] found that dust materials from the Gobi deserts can only be entrained to elevations of <3000 m, and are the main sources of dust deposited in eastern China. In contrast, dust materials from the Taklimakan Desert can be entrained to elevations of >5000 m, and are therefore transported over long distances. The prevailing northwesterly winds passing over the Gobi deserts may carry the active airborne dust to eastern China during the dry season Mixing of Dust and Anthropogenic Pollution [35] Dust particles play an important role in atmospheric photochemistry, acting as reactive surfaces for, or interacting with, anthropogenic pollutants [Dentener et al., 1996]. Asian dust can have considerable chemical reactivity and can mix with sulfur precursors as it passes over eastern China [Huang et al., 2010]. While springtime outflows of Asian dust are known to mix well with pollutants within the PBL [Itahashi et al., 2010], previous analysis has mainly focused on case studies of large-scale dust storms during the spring. [36] In the present study, airborne dust was active over the Gobi deserts from November to January and was followed by frequent high UAI values (Figure 12). CALIPSO observations show that pure airborne dust was much more frequent in spring than in winter, and the sedimentation of dust 11 of 16

12 Figure 13. (a) The frequency of different fog events in Beijing and Zhengzhou between October 2010 and March (b) Regional pollution followed by different fog events over the NCP. Arrowhead denotes general wind direction of wind fields at 850 hpa. is thought to be unrelated to diurnal cycles [Liu et al., 2008]. However, the OMI UAI indicates more active airborne dust over the Gobi deserts during winter. High UAI values were most frequent during the winter in both the Gobi deserts and in North China. The mixture of dust and pollutants could be one of the reasons for this discrepancy, because CALIPSO lidar is highly sensitive in identifying pure dust particles [Mielonen et al., 2009]. CALIPSO detected fewer dust events in the Gobi deserts, possibly because of its limited swath width (footprint diameter: 70 m) and the small scale of dust events during the winter season. Satellite observations and meteorological data show that prevailing northwesterly winds can transport dust plumes to the NCP (Figure 12). [37] Satellite observations demonstrate that air masses with dust plumes lead to abrupt regional smog via mixing with local pollutants. Dust transport in winter was more frequent but occurred on a much smaller scale than during spring dust storms. Compared with dust storms, the prevalent dust plumes resulted in greater mixing and perennial influences. Although the swath of CALIPSO is very narrow, superposition or mixing of dust plumes and anthropogenic pollutants was commonly evident in CALIPSO observations (Figure 9). Mineral aerosols appeared at high concentrations in both winter and spring, and constituted 35% of the annual PM 10 values [Zhang et al., 2011]. In urban Beijing, mineral aerosols accounted for 32 67% of total suspended particles (TSP) and 10 70% of PM 2.5 in the normal four seasons [Han et al., 2007]. Outside sources accounted for 62% of the total mineral TSP and 76% of the PM 2.5 in spring, and 69% and 45% in winter, respectively. Although rural areas have much lower concentrations of acid gases and anthropogenic aerosols than urban/industrial regions [He et al., 2012; Xu et al., 2011], the background concentration of the pollutants is still sufficient to fully coat dust particles transported to the NCP [Ma et al., 2010]. [38] The OMI UAI values cannot provide total coverage due to row anomalies from January 2009, and the NCP region is not covered for approximately one-third of the time since this date. Clouds and spatial variation of dust haze may also affect the statistics of the distribution of high UAI values in Figure 12. However, the general trends of the UAI values are in accordance with the synergetic observations. Dust plumes with high UAI values were transported to the NCP from the northwestern deserts every 2 3 days. Unlike the dust storms that are visible in satellite true color images, pure dust plumes at smaller scales had low AOD values over the clear areas. When dust passed over the urban/industrial belt in the NCP, dense smog consisting of polluted dust particles formed within several hours Haze-Fog Cloud Over the NCP [39] The CMA defines haze in terms of RH and visibility, but fog and haze are often difficult to distinguish in the NCP. Moist smog is relatively foggy in rural areas, yet hazy in 12 of 16

13 urban areas. During the dry season, few days of pure fog were observed over the NCP and haze-fog days appeared to prevail during wet weather. Hygroscopic growth of fine particles can enhance light extinction in the pollution layer [Jung and Kim, 2011], and frequent fog events were found in October and February, when most of the large-scale and durable haze occurred (Figure 13a). [40] Based on comparisons of haze clouds and variations in RH in Beijing and Zhengzhou, we found that many of the regional smog events were related to high RH. The southerly air masses with abundant water vapor can mix and interact with local pollutants. High concentrations of gaseous precursors and heterogeneous particle sources provide a basis for regional pollution [Li et al., 2011a; Lu et al., 2010]. The mountains to the north and west of the NCP also favor the accumulation of haze-fog clouds with continuous transport of water vapor. The first two images in Figure 13b show typical samples of regional pollution caused by moist air masses. Southeasterly airflows from the Yellow Sea, containing water vapor mixed with pollutants, on 19 October 2011 formed a thick haze-fog layer that accumulated over the Beijing and Tianjin areas. Large-scale southwesterly airflows prevailed over the NCP from 25 November 2011, leading to widespread haze-fog for nearly five days. [41] Radiation fog can also favor the formation of regional pollution, but this is usually limited compared with advection fog. A Terra MODIS true color image shows radiation fog in the central NCP on the morning of 30 October 2010 (Figure 13b). RH levels were >80% near the surface in Zhengzhou (altitude <200 m), but decreased to <50% in the afternoon. The moist and calm conditions of regional low fog can enhance the accumulation of local pollution. Polluted fog can transform into haze in the afternoon, depending on the meteorological conditions, the amount of fog, and local emission levels. Satellite observations showed that haze-fog was usually enhanced by prevailing dust plumes, resulting in much denser smog clouds. [42] Substantial haze-fog was observed near the surface when dense, durative haze clouds covered the NCP on 8 October 2010 (Figure 13b), although the dense polluted fog makes it difficult to identify the contribution of airborne dust. Although the top of the low fog/inversion layer was between 100 and 300 m, high UAI values indicate the existence of an elevated absorbing aerosol layer in the fog-haze cloud. CALIPSO detected dust plumes between the deserts and the NCP with northwest winds at high altitudes. Thus, dust transport can enhance the haze-fog pollution, resulting in thick and dense haze clouds. This view is supported by the generally stable concentrations of SO 2 and NO 2, and a strong PM 10 peak at widely dispersed sites (Figure 10). [43] It is worth noting that durable, dense smog still occurred when RH was <80% over the NCP. During the durable haze event of January 2010, humid airflows did not arrive in the NCP until the afternoon of 18 January. Ongoing dust transport and generally stagnant conditions near the land surface led to regional pollution during this period, and although weather conditions were mutable within the PBL, northwesterly winds were dominant at 700 hpa ( 3 km). Consequently, durable haze clouds resulted from various sources. Generally, moist airflows and regional radiation fog played a major role in the frequent smog clouds in October and February, aided by dust plumes. Compared with dust transport, smog clouds caused by moist air masses generally had a longer duration. 5. Discussion 5.1. Haze Over the NCP: Regional or Local? [44] Satellite observations demonstrate that the transport of dust plumes and large amounts of water vapor is the main driver of regional haze pollution over the NCP. A high background concentration of anthropogenic emissions provides an environment for the mixing and transformation of pollutants from various sources [He et al., 2012; Xu et al., 2011]. However, previous studies based on ground measurements have reported different results to those of the present study. Based on chemical analysis of the pollutants at one urban site, Li et al. [2011] suggested that the accumulation of anthropogenic pollutants was the main source of the regional brown haze. Photochemical smog and gray haze-fog are considered to be typical of regional air pollution over eastern China [Ma et al., 2011]. [45] Different scales of observation could explain discrepancies between satellite observations and ground measurements. For example, a typical photochemical smog was observed over the urban areas of Beijing on 5 November 2010 (Figure 14c), when brown haze from local vehicles and industries accumulated over the city. However, there was no obvious indication of regional pollution around Beijing in isochronous Aqua satellite images (Figure 14). The concentrations of SO 2,NO 2, and PM 10 were 21, 84, and 121 mg m 3, respectively, indicating that vehicle emissions were the main source of the brown haze. The urban photochemical smog was mainly concentrated near local sources and made a limited contribution to regional pollution. Compared with urban smog, the haze cloud resulting from heterogeneous sources is highly inhomogeneous in both its vertical and horizontal distributions. Consequently, ground samples from urban sites capture only some characteristics of the haze clouds, meaning that the degree to which local observations represent regional conditions must be considered carefully. [46] Primary pollutants (SO 2,NO 2, and PM 10 ) were found at high concentrations in major cities of the NCP, especially during the heating season (Figure 11). The accumulation of pollutants from urban/industrial regions causes common brown urban haze, as shown in Figure 14c. However, regional haze shows different trends to that of anthropogenic emissions (Figure 6), but is consistent with the regional transport of dust plumes and water vapor. Southerly winds, and the existence of mountains in the north and west, can favor the accumulation of industrial pollutants over the northern NCP, but few regional pollution events were dominated by anthropogenic sources. Most of the haze clouds were driven by a mixture of dust plumes or water vapor with local pollutants. Generally, local and regional haze pollution over the NCP have different formation mechanisms. [47] There are more anthropogenic pollutants in the haze clouds over urban/industrial regions than over rural areas [Ma et al., 2010; Xu et al., 2011]. Local urban smog can still occur near the surface when dust plumes or moist air masses exist over cities, and can act as part of the haze cloud. Because the distribution of haze clouds is variable, the influence of local urban smog may be stronger than that of regional haze in certain regions with high emission levels. 13 of 16

14 Figure 14. Urban haze over Beijing on 5 November 2010 observed from different scales. (a) 1 km Aqua MODIS true color image of the NCP. (b) 250 m Aqua MODIS true color image of Beijing area. (c) Photo of smog over Beijing urban areas taken at altitude of 550 m in western Fragrance Hill (the red arrowhead in Figure 14b). Both local and regional haze are prevalent over the NCP, but they differ in terms of scale, optical properties, and their effects on air quality and climate Discrepancies Among Studies of Haze Pollution [48] Haze clouds over the NCP show large spatial and temporal variation. These inhomogeneous structures and heterogeneous properties lead to discrepancies among studies. Unlike the ABC over south Asia, which prevails during the biomass burning season [Gustafsson et al., 2009], haze clouds over North China mainly result from the interaction of large-scale natural sources and local accumulated pollutants. Traditional views link the haze clouds with large anthropogenic emissions over eastern China, based directly on measurements near the land surface. In the present study, multisensor observations from space show that natural sources have strong effects on the formation of regional haze over the NCP. [49] The results of a coupled climate/air quality model suggest that regional haze results in a 5 30% reduction in the solar radiation that reaches some of China s most productive agricultural regions [Chameides et al., 1999]. However, observations from limited sites rarely fully constrain these models, and overestimations may occur if the sites are located near emission sources. Most of the haze events occur over part of the NCP, and the large spatial variation and abrupt formation processes may weaken the long-term climate effects of regional haze in some regions. The contribution of anthropogenic sources could also be overestimated due to the confusion urban smog with haze clouds. Urban haze is common in most of the major cities in eastern China due to high-level local emissions, but mostly this local accumulation of pollutants is inadequate to cause regional dense clouds. [50] The chemical and physical properties of aerosols show great differences among the regions of China [Li et al., 2011a; Wang et al., 2011]. Although satellites commonly observe dust plumes over the NCP in winter, few ground studies have reported dust transport events, due largely to the relatively small magnitude and large spatial variation of blowing dust. Fixed sites may miss many cases due to the temporally variable distribution of dust-haze clouds. Groundbased observation networks observe a high Angstrom exponent in northern China [Wang et al., 2011], while mineral aerosols account for a large fraction of PM 10 (35%) [Zhang et al., 2011]. The inhomogeneous structure of dust-haze smog, with fine anthropogenic pollutants near the surface and polluted dust at higher altitudes, can lead to discrepancies between ground measurements and satellite observations at the top of the atmosphere Uncertainties in Regional Pollution [51] The combined observations from both space and ground stations have provided the first overview of the haze pollution over eastern China. Although a chemical analysis was not performed, this study reveals the general state of haze pollution over the NCP, which is affected by both intense anthropogenic pollution and natural transport processes. The heterogeneous emissions, driven by various meteorological conditions and their interactions, lead to common forms of pollution, including local urban haze, regional haze-fog, and dust-haze clouds, which differ in their properties and in their effects on air quality and climate. Although the general variations and formation mechanisms of the haze clouds have been described in this study, uncertainties remain in terms of their chemical and physical properties, and their effects on climate, due to insufficient observations and complex interactions. [52] The applicability of satellite aerosol retrievals hampers advances in our understanding of haze clouds. We found that the MODIS and PARASOL AOD products cannot provide full retrievals of many haze clouds over the NCP, leading to an underestimation of the aerosol loading. The applicability of monthly mean satellite AOD data, which are usually used as a representation of spatial and temporal variations in aerosol loading, may be strongly affected by haze. It is necessary to closely examine whether satellite AOD trends represent variations in anthropogenic emissions, as used in some studies [Lu et al., 2010]. OMI row anomalies cause incomplete coverage, and UAI is unable to provide further quantitative information regarding dust, soot, and the mixture of these two materials. In 14 of 16