Atmospheric Chemistry and Physics the 2013 Severe Haze over Southern Hebei, China: Model Evaluation, Source Apportionment, and Policy Implications

Extremely severe and persistent haze occurred in January 2013 over eastern and northern China. The record-breaking high concentrations of fine particulate matter (PM 2.5) of more than 700 µg m −3 on hourly average and the persistence of the episodes have raised widespread, considerable public concerns. During that period, 7 of the top 10 polluted cities in China were within the Hebei Province. The three cities in southern Hebei (Shijiazhuang, Xingtai, and Handan) have been listed as the top three polluted cities according to the statistics for the first half of the year 2013. In this study, the Mesoscale Modeling System Generation 5 (MM5) and the Models-3/Community Multiscale Air Quality (CMAQ) modeling system are applied to simulate the 2013 severe winter regional hazes in East Asia and northern China at horizontal grid resolutions of 36 and 12 km, respectively , using the Multi-resolution Emission Inventory for China (MEIC). The source contributions of major source regions and sectors to PM 2.5 concentrations in the three most polluted cities in southern Hebei are quantified by aiming at the understanding of the sources of the severe haze pollution in this region, and the results are compared with De-cember 2007, the haziest month in the period 2001–2010. Model evaluation against meteorological and air quality observations indicates an overall acceptable performance and the model tends to underpredict PM 2.5 and coarse particulate matter (PM 10) concentrations during the extremely polluted episodes. The MEIC inventory is proven to be a good estimation in terms of total emissions of cities but uncertainties exist in the spatial allocations of emissions into fine grid resolutions within cities. The source apportionment shows that emissions from northern Hebei and the Beijing-Tianjin city cluster are two major regional contributors to the pollution in January 2013 in Shijiazhuang, compared with those from Shanxi and northern Hebei for December 2007. For Xingtai and Handan, the emissions from northern Hebei and Henan are important. The industrial and domestic sources are the most significant local contributors, and the domestic and agricultural emissions from Shandong and Henan are non-negligible regional sources, especially for Xingtai and Handan. Even in the top two haziest months (i.e., Jan-uary 2013 and December 2007), a large fraction of PM 2.5 in the three cities may originate from quite different regional sources. These results indicate the importance of establishing a regional joint framework of policymaking and action system to effectively mitigate air pollution …


Introduction
In February 2012, the Ministry of Environmental Protection (MEP) of China enacted the new National Ambient Air Quality Standards (CNAAQS), which are required to be enforced in many major cities and regions step by step before the nationwide enforcement becomes effective on 1 January 2016 (MEP, 2012).It sets the limits for the first time on fine particulate matter (PM 2.5 ), of 75 µg m −3 and 35 µg m −3 for daily and annual average concentrations, respectively.The limits for coarse particulate matter (PM 10 ) remain as 150 µg m −3 Published by Copernicus Publications on behalf of the European Geosciences Union.
for daily average but lower for annual average from a previous 100 µg m −3 to 70 µg m −3 (SEPA, 1996).However, it is reported that only ∼ 30 % of days on average in the 113 state key environmental protection cities can reach the new standards (MEP, 2013).
In January 2013, extremely severe, persistent haze occurred in eastern and central China, raising considerable public attention.This month has been reported as the haziest month in the past 60 years in Beijing (Lu et al., 2013).The daily concentrations of PM 2.5 at some sites in Beijing and Shijiazhuang (the capital city of Hebei Province) have been over 500 µg m −3 , which is 6.7 times the new CNAAQS.The corresponding air quality index (AQI) cannot even be calculated because the observed PM 2.5 concentrations exceeded the upper limit value defined in the calculation of AQI.Since then, the China National Environmental Monitoring Center (CNEMC) has published the 10 most polluted and clearest cities in China every month.The top 10 polluted cities in January 2013 included Xingtai, Shijiazhuang, Baoding, Handan, Langfang, Hengshui, Jinan, Tangshan, Beijing, and Zhengzhou.Seven out of the 10 cities are located in Hebei.The three cities in southern Hebei, i.e., Xingtai, Shijiazhuang, and Handan, are listed in the top four polluted cities by CNEMC in January 2013 and the top three in the most recent report of CNEMC on the assessment of urban air quality for January to June 2013 (http://www.cnemc.cn/publish/106/news/news_37027.html).Air pollution in Hebei has raised widespread public concerns in China and worldwide.
Hebei Province, located in northern China, has an area of 187 700 km 2 and population of 71.85 million.It extends west to the Taihang Mountains and south to the Yellow River, and encloses two megacities, Beijing and Tianjin.Most of the northwest area of Hebei is mountainous or hilly, and the central and south areas belong to the North China Plain (NCP).Hebei has a monsoon climate of medium latitudes, which has dry and windy springs, hot and rainy summers, and dry and cold winters (L.T. Wang et al., 2013).The prevailing winds are northwest in winter and southeast in summer.Hebei's gross domestic product (GDP) in 2011 was 2.45 trillion RMB, ranked 6th in China (NBS, 2012a).The major industrial manufacturers in Hebei include coal, power, steel and iron, ceramics, petroleum, and pharmaceutics.According to the Natural Bureau of Statistics (NBS), 24.0 % of steel, 14.5 % of coke and 6.9 % of cement in China were produced in Hebei in 2011 (NBS, 2012a, b).Most of these industrial manufacturers emit large amounts of air pollutants.The emissions of primary PM 2.5 , PM 10 , sulfur dioxide (SO 2 ), and nitrogen oxides (NO x ) from Hebei are estimated to be as high as 7.4 %, 7.5 %, 7.4 %, and 6.3 %, respectively, in the national total emissions in China, according to the widely used Intercontinental Chemical Transport Experiment (INTEX-B) emission inventory (Zhang et al., 2009).Very dense emission sources within Hebei is one of the causes for the severe air pollution occurring in Hebei.
The special location of the three southern Hebei cities, Shijiazhuang, Xingtai, and Handan, should be taken into account when analyzing the reasons for the extremely heavy pollution.They are located in the intersectional area of four provinces Hebei, Shandong, Henan and Shanxi.All of them are heavily populated, industrialized, and urbanized.The northwest prevailing wind in winter can bring pollutants from northern Hebei, Beijing, Tianjin, and Shanxi, and the southeast wind in summer may transport pollutants from Shandong and Henan.Shandong andHenan's GDP ranked 3rd and5th, respectively, in China in 2011 (NBS, 2012a).Shandong is the biggest agricultural province and Shanxi is a leading coal producer in China.The four provinces together produced 50.1 % of coke, 40.8 % of steel, and 22.6 % of cement in China in 2011 (NBS, 2012a, b).Note that China's national productions of coke, steel, and cement were about 60 %, 45.5 %, and 60 % of the world, respectively (NBS, 2012a).The four provinces thus produce about 30 % of coke, 18.6 % of steel, and 13.6 % of cement in the world.In the INTEX-B emission inventory, the total emissions from these four provinces account for 28.7 %, 27.9 %, 28.3 %, and 25.0 % in the national emissions of PM 2.5 , PM 10 , SO 2 , and NO x , respectively (Zhang et al., 2009).These industrial manufacturers create a heavy burden on air quality in the four provinces and surrounding areas.
The Community Multiscale Air Quality Model (CMAQ) system has been increasingly applied in the past few years to simulate air quality in China, in particular, under various emission reduction scenarios in support of the development of the emission control measures for the 2008 Beijing Olympics (e.g., Streets et al., 2007;Chen et al., 2007Chen et al., , 2008; L. T. Wang et al., 2008Wang et al., , 2010;;K. Wang et al., 2009;Fu et al., 2009;Zhou et al., 2010; S. X. Wang et al., 2011;Xing et al., 2011;Zhou et al., 2012;Gao and Zhang, 2012;Li et al., 2013).CMAQ has been evaluated and proven to be a creditable tool by several studies in Asia (e.g., M. Zhang et al., 2006;Uno et al., 2007;Fu et al., 2008; L. T. Wang et al., 2008, 2010, K. Wang et al., 2009;and Liu et al., 2010a, b).These studies, however, focused largely on either all of China or the three major polluted regions in China, NCP, the Yangtze River Delta (YRD), and the Pearl River Delta (PRD).In addition, among all 3-D regional air quality model applications in China, very few studies provide a comprehensive examination of the contributions of major source regions and source sectors to particulate matter (PM) concentrations in major receptor cities such as Beijing, Tianjin, and those most polluted cities in southern Hebei.Most of the previous studies involving Hebei aimed at the air quality improvement in Beijing and Tianjin, especially for the 2008 Beijing Olympics (Streets et al., 2007;Chen et al., 2007Chen et al., , 2008; L. T. Wang et al., 2008;Fu et al., 2009;Xing et al., 2011;Gao and Zhang, 2012).By performing CMAQ simulations with and without emissions from Hebei Province, these studies concluded that 20-40 % of air polluants in Beijing came from Hebei Province.There are only a few studies on air quality in Hebei.Most focused on the long-term statistical analyses of haze frequencies (Fan et al., 2005;Wei, 2010), and their statistical relationships with meteorological factors (Wei et al., 2010) and the concentrations of PM 10 (Zhang, 2009).Only one study used a 3-D air quality model to understand the formation mechanisms and sources of the severe haze pollution in Hebei (L.T. Wang et al., 2012).In L. T. Wang et al. (2012), the Mesoscale Modeling System Generation 5 (MM5) and CMAQ were applied to simulate air quality in southern Hebei in December 2007, the most polluted month during the period 2001-2010, and quantify the contributions of source regions to the PM 2.5 concentrations in Shijiazhuang and Xingtai.They found that haze pollution over southern Hebei was a regional problem, and the transboundary pollution played an important role during the haze period.
In this paper, the work of L. T. Wang et al. (2012) is extended in several aspects.First, haze frequencies in the seven representative cities in northern China are analyzed for the period January 2001 to February 2013, whereby January 2013 is found to be the most polluted month since 2001.Second, the regional haze events in January and February 2013 are simulated using the MM5-CMAQ system; the model predictions are evaluated against meteorological and real-time air quality observations in China.Third, the source contributions of major source regions to the PM 2.5 concentrations in the three most polluted cities, Shijiazhuang, Xingtai, and Handan during the two months are quantified and compared with the results for December 2007 of L. T. Wang et al. (2012).Finally, the contributions by each source sector in each source region are also analyzed to further identify the most influential contributors to the severe haze pollution in those three cities.The objectives of this study are to quantify the contributions of local and regional sources to the extremely severe haze pollution over the major cities in southern Hebei and provide scientific information for policymaking on the future air pollution control in this region at the province level and under a regional jointly controlled framework.
Compared to previous studies, this study has several distinct features.First, this work represents the most comprehensive 3-D air quality modeling study focusing on southern Hebei.Second, this work performs an extensive set of CMAQ simulations with the Brute Force method (BFM) (Dunker et al., 1996) during the most severe winter regional haze episodes to quantify source contributions from five major source sectors in six major source regions in northern China for three most polluted cities in southern Hebei, aiming at the understanding of major source contributors to the most severe air pollution episodes in this region.Third, the simulation of January and February 2013 is evaluated using the real-time observations of concentrations of PM 2.5 and PM 10 released by CNEMC since January 2013 at 496 individual monitoring stations in northern China.To our best knowledge, this work is the first study that directly uses such real-time hourly observational data from CNEMC for model evaluation in China.This data set allows the evaluation of the uncertainties in the spatial allocations of regional inventories, for the first time.Previous studies used the concentrations of PM 10 , SO 2 , and nitrogen dioxide (NO 2 ) derived based on average air pollution index (API) in major Chinese cities released by CNEMC.Such API-derived concentrations are not direct measurements but the average values of multiple representative sites within cities, and may contain some uncertainties.
The paper is organized as follows: Sect. 2 describes the modeling episode, configurations, and evaluation protocols.Section 3 presents the modeling results and their evaluation against the observations at the 496 monitoring sites in China.Section 4 examines the source contributions by six source regions and by five source sectors in each of the three receptor regions during January and February 2013 and compares them with the results of December 2007 in L. T. Wang et al. (2012).Section 5 summarizes major findings and limitations of this work.

Modeling domain and episode
As shown in Fig. 1, MM5 and CMAQ simulations are performed over two nested domains: East Asia with a grid resolution of 36 × 36 km (Domain 1) and an area in northeastern China encompassing Beijing, Tianjin, and the four provinces: Hebei, Henan, Shandong, and Shanxi at a 12 × 12 km grid resolution (Domain 2).The three key cities in southern Hebei, i.e., Shijiazhuang, Xingtai, and Handan are located at the center of Domain 2. January and February 2013 are chosen as the modeling period.The regional haze events during this period, in particular, January 2013, represent the most severe episodes with the highest PM 2.5 concentrations in China since 2001.A spin-up period of 5 days (27-31 December 2012) is used to minimize the influence of the initial conditions.

Model configurations and inputs
This work uses the same configurations as those of L. T. Wang et al. (2012) to ensure a consistency with the December 2007 MM5-CMAQ simulations.A brief description of model setup is provided below.More detailed descriptions can be found in L. T. Wang et al. (2012).MM5 model version 3.7 is used to generate meteorological fields for the CMAQ simulations.Overall, 23 sigma levels are used for the vertical grid structure from surface to tropopause (∼ 100 mb).The data sources for MM5 simulation include: the US Geological Survey database (ftp://ftp.ucar.edu/mesouser/MM5V3/TERRAIN_DATA/) for the terrain and land use data; the National Center for Environmental Prediction (NCEP) Final (FNL) Operational Global Analysis data sets for initial and boundary meteorological conditions; the NCEP Automated Data Processing (ADP) surface and upper air data for the four-dimensional data assimilation (FDDA).The analysisnudged parameters include wind, temperature, and the water vapor mixing ratio both in and above planetary boundary layer (PBL).
Two-way nesting is used for the MM5 simulation.The major physics options are the same as those of L. T. Wang et al. (2012), including the Kain-Fritsch cumulus scheme (Kain and Fritsch, 1993), the high-resolution Blackadar PBL scheme (Zhang and Anthes, 1982), the mixed phase (Reisner 1) explicit moisture scheme (Reisner et al., 1998), the cloud atmospheric radiation scheme for both longwave and shortwave radiation (Dudhia, 1993), and the force/restore (Blackadar) surface scheme (Blackadar, 1976;Deardorff, 1978).The Meteorology-Chemistry Interface Processor (MCIP) version 3.6 is applied to process the meteorological data for CMAQ.
The initial chemical conditions (ICON) for Domain 1 use the default clean profile available in CMAQ.The chemical boundary conditions (BCON) used for Domain 1 are the default profile available in CMAQ, and those for Domain 2 are generated from Domain 1.To calculate the photolysis rate for CMAQ using the photolysis rates processor (JPROC), the total ozone column data are obtained from the Ozone Measurement Instrument (OMI) (http://toms.gsfc.nasa.gov/ozone/ozone_v8.html)on the Aura satellite.
Emissions are extremely important to the accuracy of air quality modeling results.The Multi-resolution Emission Inventory for China (MEIC) (He, 2012) for the base year 2010 is used in this study.The MEIC inventory uses a dynamic, technology-based methodology for all the anthropogenic emissions of eight species, including SO 2 , NO x , carbon monoxide (CO), non-methane volatile organic compounds (NMVOCs), ammonia (NH 3 ), carbon dioxide (CO 2 ), PM 10 , and PM 2.5 in China for the years 1990 to 2010.MEIC was developed by the same group at Tsinghua University, China by continuously updating and improving the emission databases (Zhang et al., 2007(Zhang et al., , 2009;;Lei et al., 2011a).Compared with the previous inventories (e.g., Transport and Chemical Evolution over the Pacific (TRACE-P) inventory, or INTEX-B), the major improvements of MEIC include unit-based emission inventory for power plants (S.W. Wang et al., 2012) and cement plants (Lei et al., 2011b) and an online emission process database (http://www.meicmodel.org).It should be noted that for the simulation of the period January to February 2013 in this work, the total emissions may be underestimated, because of the economic increase during the period 2010-2013.The national GDP increased 9.3 % and 7.8 % in 2011and 2012, respectively (NBS, 2013).

Emission reduction scenarios and simulation design
In addition to the baseline simulation, CMAQ with the Brute Force method (BFM) (Dunker et al., 1996) is used to evaluate the source contributions to concentrations from specific source regions and sectors of interest.Namely, the contributions of a region or a sector are estimated by calculating the differences between the baseline simulation and the sensitivity simulation in which the anthropogenic emissions of that region or sector are set as zero (referred to as zero-out BFM).Strictly speaking, the BFM is a source sensitivity (SS) method (Burr and Zhang, 2011a, b), and it will not provide source apportionment (i.e., the sum of all source contributions equals the simulated baseline concentrations) if the relationship between the model input and output is non-linear, which is the case for 3-D PM modeling.Nevertheless, the zero-out BFM has been applied extensively in the past and is an effective way to approximate the contributions of various sources to support the policymaking in air pollution controls in major source regions in China (Streets et al., 2007;Chen et al., 2007;L. T. Wang et al., 2008;Fu et al., 2009;Zhou et al., 2010;Xing et al., 2011) and other regions in the world (e.g., Marmur et al., 2005;Burr and Zhang, 2011a;Zhang and Wu, 2013).
Although BFM with multiple scenarios is computationally expensive, following L. T. Wang et al. (2012), it is selected for source apportionment in this work because it is straightforward and can simulate the non-linear relationship among PM and PM precursors and indirect effects resulting from interactions between secondary PM species and non-direct gaseous precursors for large emission changes.It should be mentioned that another source sensitivity method using CMAQ is the embedded tool of Decoupled Direct Method in 3 Dimensions (DDM-3D) developed in recent years (Yang et al., 1997;Dunker et al., 2002;Cohan et al., 2005;Napelenok et al., 2006).It directly calculates the sensitivities of multiple sources and parameters at the same time of the baseline simulation, thus effectively reducing the computing time.However, DDM implemented in CMAQ only calculates the local first-order sensitivities for aerosols species, which cannot accurately predict the changes of chemical concentrations under large emission perturbations (i.e., larger than 40 %) (Zhang et al., 2005).A different approach for source apportionment is the tagged species method, i.e., some tracer species from specific source categories are added in the 3-D models, which undergo the same processes as the bulk species to track the contributions from those sources during the model simulation.The Tagged Species Source Apportionment algorithm (TSSA) (Tonnessen et al., 2007;Z. S. Wang et al., 2009) and the Particulate Precursor Tagging Methodology (PPTM) (US EPA, 2009) within CMAQ are two of those tools.They can attribute 100 % of concentration to emission sources but their main weakness lies in the omission of the indirect effects because of several assumptions for the secondary PM calculation, e.g., all secondary PM is linked to a specific precursor (sulfate to SO x , nitrate to NO x , and ammonium to NH 3 ), so that an increase in nitrate concentrations due to a reduction in SO 2 emissions cannot be reflected.In addition, CMAQ with TSSA or PPTM are not available for public use in the recent CMAQ versions including CMAQ v4.7.1.
Following L. T. Wang et al. (2012), six source regions in Domain 2 are selected in the regional contribution analysis (see Fig. 1).They include southern Hebei (SHB), northern Hebei (NHB), Shanxi province (SX), Shandong province (SD), Henan province (HN), and a two-city cluster region covering Beijing and Tianjin (BJTJ).Five major emission sectors are considered in sector contribution analysis including power plants (PO), industrial (IN), domestic (DO), transportation (TR), and agriculture (AG).Those five sectors cover about 93 % to 98 % of total emissions of PM 2.5 , PM 10 , SO 2 and NO x in the six regions of interest in winter.Only two sectors, biomass open burning and biogenic emissions, are not included, because although the open burning of crop residues is an anthropogenic and controllable emission in biomass open burning, it usually occurs in summer and autumn and is not considered as one of the major sources in this study.The biogenic emissions are not anthropogenic in nature and are considered to be uncontrollable in this study, although recent studies have quantified a controllable fraction of secondary organic aerosol resulted from biogenic volatile organic compounds in the southeastern US (Carlton et al., 2010) and East Asia (Glotfelty et al., 2013).For each of the six source regions, the emissions from each of the five major sectors are zeroed out individually (referred to as simulations PO, IN, DO, TR, and AG, respectively) and collectively (referred to as AL).The analysis of source sensitivities focuses on the three major receptor cities in southern Hebei, in other words, Shijiazhuang, Xingtai, and Handan, which are the most polluted cities in both this region and China during the most severe regional haze in January 2013.In addition, the source contributions of those three cities to each other are assessed, by zeroing out the emissions in each of the three cities, in order to understand their mutual influences.Therefore, a total of forty 2-month simulations, including 39 emission zero-out scenarios (= 6 source regions × (5 sectors + 1 AL) + 3) plus one baseline, are performed in this study.

Model evaluation data set and protocols
Table 1 summarizes the observational data used for model evaluation in this study.The meteorological data used in this study are obtained from the National Climate Data Center (NCDC) integrated surface database and the evaluations are performed for the five major parameters that affect air pollution: temperature at 2 m (T2), relative humidity at 2 m (RH2), wind speed at 10 m (WS10), wind direction at 10 m (WD10), and daily precipitation.For January to February, data every 1 or 3 h (most at 3 h) at a total of 371 sites within our domains are available.Engineering.b T2 -temperature at 2 m; RH2 -relative humidity at 2 m; WS10 -wind speed at 10 m; WD10 -and wind direction at 10 m.
The chemical concentrations evaluated in this study include PM 2.5 and PM 10 using observations from two data sets.The first one is the CNEMC's real-time database for pollutant concentrations.In L. T. Wang et al. (2012), the API data published daily by MEP were used to evaluate model performance following a number of past studies (Streets et al., 2007;Chen et al., 2007;Wang et al., 2008Wang et al., , 2010;;Fu et al., 2009;Liu et al., 2010a;Xing et al., 2011).This database is the only public accessible air quality data until 2011 when the real-time hourly concentrations of SO 2 , NO 2 , and PM 10 at all national sites in the key environmental protection cities started to be available on the web site of the CNEMC (http://58.68.130.147/air/).Since January 2013, the CNEMC began to release the real-time hourly concentrations of SO 2 , NO 2 , CO, ozone (O 3 ), PM 2.5 , and PM 10 at 496 national monitoring stations in 74 major cities in China.The 74 cities include all the capital cities of each province, municipalities, and all the cities within the Beijing-Tianjin-Hebei area (BTH), the YRD, and the PRD (http://113.108.142.147:20035/emcpublish/).Such real-time air quality data provide a much better database for model evaluation and are thus used in this work.However, the main difficulty in using this realtime database lies in the inaccessibility of the historic data of more than 24 h ago, as the data are no longer available on its web site after 24 h posting.For this reason, the data used in model evaluation are for the period 14 January-28 February, 2013, and those for 1-13 January 2013 are not available and thus not used in the model evaluation in this study.The second data source available to this study is the observations at a site located in the Hebei University of Engineering (refer to HEBEU) measured by the lead author's group since July 2012 (Wei et al., 2013).HEBEU is located in the southeast edge of the urban area of Handan, where the continuous PM 2.5 and PM 10 concentrations are measured.More detailed descriptions of the characteristics of the site and measurement method can be found in Wei et al. (2013).The observed hourly PM 2.5 and PM 10 concentrations at this site are used in the model evaluation for both January and February 2013.
The meteorological evaluation is performed in terms of domain-wide overall statistics.The statistical measures calculated include the mean bias (MB), the root mean square error (RMSE), the normalized mean bias (NMB), and the normalized mean error (NME), which are defined in Y. Zhang et al. (2006).The chemical evaluation is performed in terms of the overall statistics, spatial distribution, and temporal vari-ations at the representative cities, following the guidance by US Environmental Protection Agency (US EPA, 2007).In addition to NMB and NME, the mean fractional bias (MFB) and the mean fractional error (MFE) are analyzed and they are also defined in Y. Zhang et al. (2006).For both 36 km and 12 km domains, the observations of multiple sites are averaged, if they fall into the same grid, and are compared with the predictions extracted from that grid.For grid cells that only contain a single monitoring site, the observed value at this site is compared with the prediction in this grid cell.

Haze frequencies and visibilities in the representative cities
In L. T. Wang et al. (2012), haze frequencies in the seven representative cities in northern China, Shijiazhuang, Xingtai, Beijing, Tianjin, Taiyuan, Zhengzhou, and Jinan (see Fig. 1), were examined for the period 2001 to 2010, and December 2007 was found to be the haziest month during the 10 years.
The hazy days from 2011 to February 2013 are further identified in this study, according to the meteorological observations in those cities using the same criteria as those used in L. T. Wang et al. (2012).Figure 2  The number of hazy and foggy days that occurred in the seven cities in the 3 representative months, December 2007, and January and February 2013, are summarized along with the average monthly visibilities in Fig. 3.The foggy days are determined when it was recorded as "foggy weather" according to the WMO (Weather Meteorological Organization) meteorological code or when the relative humidity was above 90 % and the visibility was less than 10 km but there was no any other types of weather that influenced the visibility such as precipitation and sandstorm.Although January 2013 and December 2007 have the same total number of hazy days, several cities such as Shijiazhuang, Xingtai, and Zhengzhou had more frequent fog days in January 2013 than in December 2007 (8 vs. 1, 5 vs. 2, and 2 vs. 0, respectively).The total number of the foggy days in the seven cities is as high as 21 in January 2013, compared with 14 in December 2007 and only 3 in February 2013.Foggy weather not only indicates a very stable atmosphere that can facilitate accumulation of the pollutants, but also results in a growth of atmospheric aerosol particles due to water uptake.Persistently hazy/foggy weather will further reduce the incoming solar radiations that in turn result in a more stable atmosphere and an increased pollutant accumulation.February 2013 was much dryer than the other 2 months with only 1 foggy day occurring in Xingtai, Beijing, and Jinan.
As shown in Fig. 3a, January 2013 was the top polluted month in terms of the total frequencies of hazy/foggy weather among the 3 representative months in most of the seven cities, except for Xingtai and Taiyuan.For example, the number of hazy/foggy days in Tianjin is 17 in January 2013, which is much higher than 10 days in December 2007, and 9 days in February 2013.Zhengzhou, Shijiazhuang, Xingtai, and Tianjin were the most polluted cities in terms of their total numbers of hazy and foggy days, which were 24, 22, 17, and 17, respectively.
The monthly average visibilities of the seven key cities are calculated according to the meteorological observations at the 06:00 GMT (corresponding to 14:00 in Beijing local standard time) every day to avoid the influence of the radiation fog appeared in the morning, as discussed in L. T. Wang et al. (2012).As shown in Fig. 3b, the lowest visibility in terms of the monthly seven-city average was only 8.5 km, which occurred in January 2013.This number is 13.7 % lower than that in December 2007 (9.8 km) and 21.9 % lower than that in February 2013 (10.8 km).In terms of the haze/fog frequencies and the visibilities, air pollution in January 2013 is much more severe than those in the other 2 months, and those in February 2013 are the best among the 3 months analyzed.

Meteorological predictions
Table 2 shows the performance statistics for the meteorological predictions over Domain 1 at the 36 km resolution and Domain 2 at both 36 and 12 km resolutions.The nearsurface temperature is overall underpredicted in both domains.The MBs for the 36 km domain are −1.5 • C for both January and February.Predictions over Domain 2 at 12 km resolution in terms of MBs agree better with the observations than those at 36 km, with MBs of −1.1 • C and −0.9 • C for January and February, comparing to MBs of −1.2 • C and −1.0 • C, respectively, at 36 km.The large cold biases in T2 mainly occur in western China, especially over the Tibetan Plateau during both months, which were covered by thick snow layers.Such large cold biases reflect the limitations of MM5 in treatments of snow-related processes and snow effects (Zhang et al., 2011;Liu et al., 2010c).Relative humidity at 2 m agrees well with the observations with MBs ranging from −0.5 % to −8.0 % and NMBs ranging from −0.8 % to −12.6 % for both domains.It is slightly underpredicted, especially in Domain 2. Wind speed at 10 m is well reproduced in both domains, with MBs and NMBs ranging from 0.6 to 0.8 m s −1 , and 23.2 % to 33.8 %, respectively.For wind direction at 10 m, MBs for Domain 1 are −10.0• for January the corresponding spatial distributions of NMBs for PM 2.5 and PM 10 at 36 km horizontal grid resolution for January and February, 2013.Note the average for January is calculated from 14 January to 31 January due to the lack of data in 1-13 January.Square signs indicate the observations.and −12.9 • for February, and 7.1 • and −2.7 • for Domain 2 at 36 km resolution and 0.1 • and −8.1 • , respectively, at 12 km.It should be noted that the differences between the predictions and observations for wind direction may not reflect the real differences because it is a vector; e.g., the actual difference between 0 • and 360 • are zero instead of the calculated numeric difference of 360 • (Y.Zhang et al., 2006).Precipitation is underpredicted for both months, with NMBs of −31.6 % and −7.2 % for January and February for Domain 1, and −43.9 % and −34.7 % for Domain 2 at 36 km, and −44.2 % and −36.7 % at 12 km, respectively.The results at a fine grid resolution do not give a better performance.The likely reasons for moderate-to-large underpredictions in precipitation include the limitations in the model treatments, such as the cloud microphysics module used or inaccuracies in the model inputs, such as land use data over complex terrains, as discussed in several studies (e.g., Zhang et al. (2011) over China and Queen et al. (2008) and Wu et al. (2008) over other regions).Overall, our results are generally consistent with the 36 km and 12 km simulation results in the MM5 model evaluation study over eastern China by Zhang et al. (2011).

Domain-wide PM predictions at 36 km and 12 km grid resolutions
Figure 4a shows the overlay of the simulated and observed monthly concentrations of PM 2.5 and PM 10 of the 74 cities in Domain 1 in January and February 2013.The corresponding performance statistics are shown in Table 3, including the MFBs and MFEs, which are recommended as the key statistical performance measures by US EPA (2007).In this guidance (US EPA, 2007), a satisfactory performance is indicated by an MFB within ±60 %, and an MFE within 75 % proposed by Boylan (2005), and an MFB within ±50 % and an MFE within 75 % proposed by Morris et al. (2005).In this study, the MFBs within ±60 % and MFEs within 75 % are considered to be satisfactory performance.
In January, the simulated results indicate two heavily polluted areas.One is the large area in the central east of China, including southern BTH, the YRD, nearly all the areas of Shandong, Henan, Anhui, Hubei, and Hunan provinces.The other one is located in the Sichuan Basin.The predicted PM 2.5 concentrations in those areas are close to or higher than 150 µg m −3 , which are much higher than the CNAAQS daily average PM 2.5 standard of 75 µg m −3 .The simulated PM 10 concentrations in those areas show the same pattern as PM 2.5 with values close to or higher than 200 µg m −3 , which are much higher than the CNAAQS daily average PM 10 standard of 150 µg m −3 .In February, the heavily polluted area in the central east of China became smaller, only covering part of the BTH area, Shandong, Henan, and Hubei provinces.The heavily polluted area in the Sichuan Basin also becomes smaller in February, compared with January.Figure 4a shows that the predicted PM 2.5 and PM 10 agree well with the observations over the large, heavily polluted area in central east of China, and the PRD in both January and February.However, a notable underprediction occurs in northeastern and northwestern China, and some of the cities in the Sichuan Basin, consistent with previous studies (Liu et al., 2010a;Wang et al., 2010; L. T. Wang et al., 2012).This underprediction may be partially attributed to the spatial allocation of the emissions, and the lack of dust emissions.Several dust storm events were reported in northwest China in January 2013 (http://www.duststorm.com.cn/), which contributed to observed PM concentrations that are not captured by CMAQ due to a lack of an online dust emission module.Other possible reasons for underpredictions lie in the underpredictions in secondary aerosols such as sulfate, ammonium, and  secondary organic aerosols due to the limitations of model treatments.For example, heterogeneous chemistry of SO 2 on the surface of particles is not treated in CMAQ, which contributes significantly to the formation of sulfate when the surface of particles is large (e.g., Zhang and Carmichael, 1999;K. Wang et al., 2009).
As shown in Table 3, the model overall underpredicts PM 2.5 and PM 10 concentrations over Domain 1.The domainwide NMBs for PM 2.5 are −11.0 % and −9.7 % for January and February, respectively.The MFBs and MFEs are −19.5 % and 58.5 % for January, and −14.1 % and 62.1 % for February, respectively.All of them are within the criteria for a satisfactory performance mentioned above.PM 10 are more underpredicted, with NMBs of −32.0 % and −31.6 % for January and February, respectively.Figure 4b shows the spatial distributions of the NMBs for PM 2.5 and PM 10 concentrations for the 74 cities.Despite the significant underpredictions for PM 2.5 concentrations in northeastern and northwestern China in January, relatively few underpredictions occur in most of the cities in the BTH, the YRD, and the PRD regions.Possible reasons for these underpredictions include the biases in the meteorological predictions (e.g., the underprediction of temperature which leads to a lower photolysis rate in the formation of secondary inorganic and organic aerosols, and the overprediction of wind speed which blows off the pollutants), the uncertainties in the spatial allocation of the emissions, the lack of the dust emissions, and possible underestimations of emissions due to different base years of emission and simulation (2010 vs. 2013).The NMBs for the cities in the BTH area fall into the range of −41.9 % (Zhangjiakou) to 37.5 % (Zhengzhou) with an average of −14.0 %.The best agreement appears in Qinhuangdao (−3.8 %), Handan (5.3 %), and Cangzhou (6.5 %).In the YRD and PRD regions, the average NMBs are −2.4 % and −11.8 %, respectively.Obvious overpredictions occur in central China along the Yangtze River with the NMBs of 18.4 % (Changsha) to 224 % (Guiyang).The spatial distributions of NMBs of PM 2.5 and PM 10 in February are quite similar to those in January except for overpredictions in some cities in the PRD regions.The average NMBs for the cities in the BTH, the YRD, and PRD regions are −17.1 %, −7.4 %, and −5.4 %, respectively.The spatial distributions of NMBs for PM 10 concentrations are generally similar to those of PM 2.5 .The average NMBs for the cities in the BTH, the YRD, and the PRD regions in January are −37.9%, −18.5 %, and −25.1 %, respectively.The corresponding values for February are −38.2%, −19.7 %, and 18.0 %, respectively.Compared with PM 2.5 , the larger biases of PM 10 are mainly attributed to the lack of the dust emissions, which contribute to a large faction of coarse particles.
Figure 5a presents the predicted PM 2.5 and PM 10 concentrations from the 12 km simulation in Domain 2 overlaid with observations in January and February, 2013.The corresponding performance statistics are shown in Table 3.The simulations indicate an extremely severe pollution occurring in January over Beijing, Tianjin, southern Hebei, most of Henan, and the western areas of Shandong and Anhui provinces.The monthly average PM 2.5 concentrations over those areas are higher than 150 µg m −3 , which is twice the new CNAAQS standard for daily PM 2.5 concentrations.The concentrations in February are much lower, with the highest concentrations in southern Hebei, northern Henan, and some western areas of Shandong.The pollution pattern of PM 10 is similar to that of PM 2.5 ; with the concentrations in the heavily polluted area over 200 µg m −3 , which is also higher than the CNAAQS daily standard for PM 10 .
Overall, predictions agree well with the observations, especially in the heavily polluted areas.Nonetheless, it should be noted that the observations at some sites are significantly higher than 200 µg m −3 during that period, which can be reflected clearly in the spatial distributions of NMBs shown in Fig. 5b.The model significantly underpredicts PM 2.5 and PM 10 concentrations over the three cities in southern Hebei in both January and February, with NMBs of −36.4 % to 35.3 % in January and −35.9 % to −1.7 % in February.In the megacities such as Beijing and Tianjin, large overpredictions and large underpredictions occur at those sites that are close to each other, indicating that the model fails to capture the spatial distribution of observed concentrations in the two-city cluster region.In general, the concentrations at the urban sites are overpredicted and those at the rural sites are underpredicted.The observed concentrations are relatively uniform over those sites compared to predictions.The discrepancy between the spatial distributions of predictions and observations may be partially attributed to the uncertainties in the spatial allocation of emissions used.For example, some sources are spatially distributed into the grids using the population density as a surrogate; however, most of the industrial manufacturers in Beijing have been moved out of the urban areas in recent years, leading to inconsistencies between the actual and estimated spatial distributions of the emissions.The averaged NMBs for PM 2.5 of Beijing, Tianjin, Shijiazhuang, Xingtai, and Handan are −8.2 %, 40.6 %, 0.8 %, −18.4 %, and 8.8 % in January, and 9.4 %, 27.3 %, −19.5 %, −13.9 %, and −5.6 % in February, respectively.Notable overpredictions can be found in the area of the YRD, especially for PM 2.5 .The averaged NMBs in the cities in YRD for PM 2.5 are 20.6 % and 26.2 % for January and February, respectively.The domain-wide NMBs for PM 2.5 at 12 km resolution are 12.2 % for January and 15.7 % for February, and the MFBs and MFEs are 8.7 % and 51.1 % for January and 14.1 % and 56.8 % for February, respectively.
As shown in Table 3, the domain-wide NMBs for PM 10 over Domain 2 are −13.6 % and −10.4 % for January and February, respectively.The spatial distributions of the NMBs for PM 10 are similar to those of PM 2.5 in the BTH area.The averaged values for Beijing, Tianjin, Shijiazhuang, Xingtai, and Handan are −11.5 %, 0.5 %, −37.8 %, −47.0 %, and −22.9 % for January, and 7.3 %, 7.7 %, −39.5 %, −42.2 %, and −29.4 % for February, respectively.In the YRD region, different from PM 2.5 concentrations that are overpredicted at nearly all sites, PM 10 concentrations at some sites show slight to moderate underpredictions (NMBs of −0.1 % to −34.4 %).The averaged NMBs in this area are −15.3% for January and −0.8 % for February.
Table 3 also compares the model performance statistics for hourly PM 2.5 and PM 10 concentrations over Domain 2 from the simulations at 36 and 12 km grid resolutions.PM 2.5 concentrations are overall slightly underpredicted at the 36 km resolution compared with overpredictions at the 12 km resolution.The domain-average NMBs and MFBs are −2.5 % and −1.2 % at 36 km, and 14.0 % and 12.1 % at 12 km for the whole period 14 January to 28 February, respectively.NMEs and MFEs are 47.8 % and 50.0 % at 36 km, and 57.9 % and 54.7 % at 12 km.They are all within the benchmarks of MFBs < ±60 % and MFEs < 75 %.PM 10 concentrations are underpredicted at both grid resolutions, with NMBs and MFBs of −27.2 % and −26.9 % at 36 km, and −12.0 % and −13.2 % at 12 km, respectively.NMEs and MFEs are 48.4 % and 55.7 % at 36 km, and 53.3 % and 56.4 % at 12 km, respectively.While the use of a finer grid resolution changes PM 2.5 performance from a slight underprediction to a moderate overprediction, it notably reduces the underpredictions in PM 10 .

Site-specific PM predictions at the 12 km grid resolution
Table 4 summarizes the MFBs and MFEs for the hourly PM 2.5 concentrations in the representative cities in Domain 2. For the three cities, Shijiazhuang, Xingtai, and Handan, the MFBs and MFEs are within the recommended benchmarks at all sites.For Beijing, as discussed previously, overpredictions occur at the urban sites and underpredictions occur at the rural sites for both January and February.MFBs and MFEs at several sites are beyond the criteria.Nevertheless, the MFBs and MFEs for the average hourly PM 2.5 concentrations in Beijing are −14.2% and 51.8 % in January, and 12.8 % and 61.2 % in February, respectively.This may imply that the total amount of the emissions in Beijing is well estimated in terms of performance statistics using the MEIC emission inventory, despite the uncertainties in the spatial allocation of those emissions.The overall performance in Tianjin is much better than that in Beijing, as the MFBs and MFEs at all sites except for one urban site are within benchmarks.In general, the PM 2.5 concentrations in Tianjin are overpredicted.In Taiyuan, significant overpredictions occur at the two urban sites and underpredictions occur at one urban and one rural site.This may be partially attributed to the uncertainties in spatial allocation of emissions, as discussed previously.The model predictions agree well for the other two capital cities, Zhengzhou and Jinan, with all the statistics at all sites satisfying the criteria.

Temporal variations of PM in the three cities in southern Hebei
To further evaluate model performance for the three most polluted cities in southern Hebei of major concerns (i.e., Shijiazhuang, Xingtai, and Handan) and to understand the formation mechanisms of PM underlying the heavy pollution episodes, the hourly PM 2.5 and PM 10 concentrations from the 12 km simulation are compared with observations.The model's capability to capture the temporal variations of the PM 2.5 and PM 10 concentrations in the three cities are discussed below.

Shijiazhuang
Figure 6a-d shows the temporal variations of the observed and simulated hourly PM 2.5 concentrations at the four monitoring sites in Shijiazhuang.The observations indicate the occurrence of several very severe pollution episodes during the period 14 January to 28 February.The maximum hourly PM 2.5 and PM 10 concentrations reached as high as 756 and 1239 µg m −3 , respectively, both at an urban site (the Central Hospital).A persistent, extremely severe pollution event occurred during the period 15-30 January.During this time period, the concentrations rapidly built up in a very short time and resulted in five high-pollution episodes in 15 days, each persisting 2-5 days.However, the situation in February was somewhat different, with significantly lower concentrations during most days and less frequent pollution episodes (each lasting about 3-5 days).The observed average concentrations of PM 2.5 at the four sites were 258.2 µg m −3 in January and 210.4 µg m −3 in February, respectively, with both higher by factors of 2.44 and 1.81, respectively, than the daily standard for PM 2.5 .The model well reproduces the PM 2.5 concentrations at the two rural sites in Shijiazhuang.At the urban site of the Century Park, the model predictions generally agree with the observations, but a relatively better performance can be found in February than in January.At the urban site of the Central Hospital, PM 2.5 concentrations are significantly underpredicted, especially during the heavily polluted period in January.This underprediction may be attributed  to the uncertainties in the estimation of the local emissions (note that Shijiazhuang still has many chemical and pharmaceutical industrial manufacturers within or close to the urban areas, which is different from Beijing), the uncertainties and limitations of the model treatments in the formation and accumulation of secondary aerosols under high humidity conditions during that episode, as well as the uncertainties in the meteorological predictions (e.g., large cold bias in near-surface temperatures).In particular, in Y. X. Wang et al.'s (2013) study, sulfate and nitrate aerosols in PM 2.5 in Beijing were observed to be as high as 60-100 µg m −3 and 30-50 µg m −3 , respectively, on daily average during the period 11-15 January 2013, while in this work the corresponding sulfate and nitrate concentrations predicted in Beijing are 14-29 µg m −3 and 13-33 µg m −3 , respectively.The significant underpredictions indicate the deficiency of the model in treating the heterogeneous chemistry (e.g., missing the heterogeneous reactions of SO 2 to form SO 2− 4 ) that contribute significantly to the formation of secondary inorganic species on the surface of existing particles such as dust (K.Wang et al., 2009).
The PM 10 concentrations are overall significantly underpredicted, as shown in Fig. 7a-d, especially for the extremely polluted period, 15 to 30 January.This underprediction may be attributed to the lack of the dust emissions, the underestimation of emissions due to different base years of emission and simulation, and the uncertainties in the meteorological predictions (e.g., the overprediction of wind speed) during the extremely polluted period.MFBs and MFEs for the PM 10 concentrations during the entire period 14 January to 28 February are −41.6 % and 61.8 % at the Century Park, −66.6 % and 73.0 % at the Central Hospital, −27.6 % and 52.0 % at Fenglong Mountain, and −41.8 % and 57.3 % at the Northwest Headwater, respectively.

Xingtai
Figures 6e and f and 7e and f compare the temporal variations of the observed and simulated PM 2.5 and PM 10 concentrations at the two sites in Xingtai.The observed average concentrations of PM 2.5 at the two sites were 296.8 µg m −3 in January and 216.9 µg m −3 in February, respectively.The predictions of PM 2.5 concentrations in Xingtai overall agree well with the observations, at both urban and rural sites, especially in February.The MFBs and MFEs during the entire period for PM 2.5 are −20.1 % and 38.0 % at the urban site, and −5.3 % and 43.2 % at the rural site, respectively (see Table 4).The six observed pollution episodes in February are clearly reproduced.However, the PM 10 concentrations are significantly underpredicted at both sites for the same reasons mentioned previously, which is similar to the PM 10 concentrations in Shijiazhuang.The MFBs and MFEs during the entire period for PM 10 are −54.4% and 61.8 % at the urban site, and −42.3 % and 57.3 % at the rural site, respectively.

Handan
Figures 6g-i and 7g-i compare the temporal variations of the observed and predicted PM 2.5 and PM 10 concentrations at the three monitoring sites in Handan.The observed average concentrations of PM 2.5 at the three sites were 248.7 µg m −3 in January and 179.5 µg m −3 in February, respectively.Very few observational data are available at the rural site (East Wastewater Treatment Plant) for model evaluation.At the urban site of Handan Environmental Protection Bureau (EPB), the model well reproduces the PM 2.5 concentrations but the PM 10 concentrations are slightly underpredicted.The MFBs and MFEs at this site are 0.6 % and 37.0 % for PM 2.5 and −28.7 % and 40.2 % for PM 10 , respectively (see Table 4).
www.atmos-chem-phys.net/14/3151/2014/Atmos.Chem.Phys., 14, 3151-3173, 2014 At the urban site of HEBEU, the observations indicate three extremely severe pollution episodes occurring on 7-9, 10-12 and 13-14 January.During the period 10-12 January the observed maximum hourly concentrations of PM 2.5 and PM 10 reached as high as 825 and 1281 µg m −3 , respectively.The highest daily average PM 2.5 concentration of 663.7 µg m −3 occurred on 11 January, which was 8.8 times the new CNAAQS.The model fails to reproduce the extremely high concentrations (> 600 µg m −3 ) for all three episodes, although it does predict a high PM 2.5 concentration of 500 µg m −3 on 8 January.The MFBs and MFEs at this site are 15.5 % and 46.8 %, respectively.The observational data are only available at this site among the nine sites for the period 1 to 13 January, it is not clear if such large underpredictions also occur at the remaining eight sites.Possible reasons include the missing of episodic emissions from local sources, underestimating the regional transport from upwind regions (e.g., from NHB and SX when northwest winds dominate), and the neglect of heterogeneous chemistry on the surface of atmospheric particles that can increase the concentrations of secondary inorganic aerosols such as sulfate, nitrate, and ammonium (K.Wang et al., 2009).The temporal analyses showed clear differences in the observations for the two periods: 1-13 January when severe pollution occurred with high PM 2.5 and PM 10 concentrations and 14-31 January when the observed PM 2.5 and PM 10 concentrations are much lower.Therefore, in the following sections, the source contributions of PM 2.5 during the two time periods are examined and discussed separately.
4 Source contributions to PM 2.5 concentrations

Contributions by source regions
Following L. T. Wang et al. (2012), the emission sources within the SHB are considered as local sources for the three cities of Shijiazhuang, Xingtai and Handan, and the sources outside the SHB are considered as regional sources.In this section, the source contributions to Shijiazhuang, Xingtai, and Handan from the six regions of interest are analyzed for January and February 2013, and compared with the results of L. T. Wang et al. (2012) for December 2007, the most polluted month during the period 2001 to 2010.In addition, the inter-influence of the three cities to the PM 2.5 concentrations in each area are assessed.

Shijiazhuang
Table 5a shows the average contributions by source region to the PM 2.5 concentrations in Shijiazhuang, Xingtai, and Handan for December 2007, and January and February 2013.The local sources (SHB) contribute 75.8 %, 67.2 %, and 66.4 % on average to PM 2.5 in Shijiazhuang, during the period 1-13 and 14-31 January, and February, respectively, compared with 65.1 % for December 2007.It is the largest contributor among the six source regions included during all four periods.The second largest contributor is the NHB for all periods in 2013, with source contributions of 9.0 % for 1-13 January, 17.8 % for 14-31 January and 15.0 % for February.However, for December 2007, SX is the second largest contributor, contributing to 14.1 % of PM 2.5 in Shijiazhuang.For 1-13 and 14-31 January and February 2013, the contributions of SX significantly decrease to 7.2 %, 3.2 %, and 3.8 %, respectively.Relatively smaller contributions can be found for BJTJ, SD, and HN, but noticeable variations remain, for example, the contribution by BJTJ varies from 0.9 % in December 2007 to 3.4 % in 14-31 January 2013.The total regional contributions to the PM 2.5 concentrations in Shijiazhuang are 24.6 % in December 2007, 19.4 % in 1-13 January, 27.5 % in 14-31 January, and 27.8 % in February.The contributions from the total sources outside Hebei are 10.5 % in 1-13 January, 9.7 % for 14-31 January, and 12.8 % for February, compared with 17.4 % for December 2007.
For January 2013, the most polluted period, the higher contributions from the local sources are caused by a more stable atmosphere that favors the accumulation of the local pollutants than other regions.Figure 8 shows the average PBL height over Domain 2 for the four periods.During all the periods, PBL heights over the southern area of Hebei are the lowest within Domain 2. This indicates that even with the same emission densities as other cities, the southern Hebei cities may experience heavier air pollution than the others.This can be explained by the special terrain and local-scale meteorological phenomenon in this region, namely, the stable air blowing from the west crosses the Taihang Mountain, which sinks and meets the air blowing from the south or east directions over southern areas of Hebei.The convergence and confrontation of the two air flows result in a very stable atmosphere and accumulation of air pollutants (C.M. Wang et al., 2013).During the period 1-13 and 14-31 January there are large areas in southern Hebei, including Baoding, Shijiazhuang, and Xingtai, with the average PBL height of as low as or less than 160 m.The areas with low PBL height are smaller in December 2007 and February 2013.In February 2013, the average PBL heights are much larger than those in the other three periods.The increases in the contributions from NHB and BJTJ indicate more frequent north winds blowing the pollutants from the northern area to Shijiazhuang.A comparison of the estimations for December 2007 and January 2013 shows that even for the most polluted months, a large fraction of PM 2.5 in Shijiazhuang could originate from quite different regions (i.e., SX vs. NHB), which may mainly be due to the differences in meteorological conditions between the 2 months.The changes in the emissions between the year 2007 and 2013, which may result in some changes in atmospheric chemistry, could be another reason.It reveals the necessity of long-term modeling and analysis for the possibility of occurrence of severe haze under various meteorological conditions in future research, which should be taken into account in future policymaking on regional   1-13, 14-31, 2013 2007 1-13, 14-31, 2013 2007 1-13, 14-31 emission control strategies to effectively control the severe haze in Shijiazhuang.
Figure 9 presents the contribution ranges from each region to the PM 2.5 concentrations in Shijiazhuang.The contributions from the SHB, NHB, and SX have larger variations than those from other regions.For example, the highest local contributions can reach nearly 95 % and the lowest can be less than 20 %.As for the NHB, the contributions fall into the range of 0-60 %, and the median values are between 7 % and 30 % for 14-31 January.The contributions from BJTJ, SD, and HN are concentrated in smaller ranges, and might be considered as minor sources of the transboundary PM 2.5 in Shijiazhuang.
It should be noted that the several factors resulting in underpredictions in PM 2.5 concentrations may introduce uncertainties in the source apportionment results, for instance, the positive bias of wind prediction and lack of dust emissions will result in underestimations of the local source contribution, the lack of heterogeneous chemistry may lead to underestimation of some sources emitting more gas phase precursors, and uncertainties in spatial distributions of emissions could result in either negative or positive bias of local or regional contributions.

Xingtai
As shown in Table 5a, the source contribution in Xingtai is similar to that of Shijiazhuang, except for noticeable increases in the contributions from HN and SD.This is expected because Xingtai is closer to the two provinces than Shijiazhuang.The local sources contribute 77.1 % and 65.9 % for the period 1-13 and 14-31 January, respectively, which are higher than 64.0 % in December 2007 and 63.9 % in February 2013.The second largest contributor also changes from SX in December 2007 and 1-13 January to NHB for the period 14-31 January and February 2013, with the contributions from NHB of 13.0 % and 10.1 % for 14-31 January and February.Unlike Shijiazhuang where SX is the third largest contributor, HN is the third largest contributing region to PM 2.5 in Xingtai, with 5.7 % for January 14-31, and 7.6 % for February 2013.The average contributions from SX, SD, and BJTJ are respectively 3.6 %, 3.4 %, and 2.9 % for 1-13 January, 3.6 %, 3.4 %, and 2.9 % for 14-31 January, and 3.6 %, 5.8 %, and 3.2 % for February.The total regional contributions are 19.1 % for 1-13 January, 28.6 % for 14-31 January, and 30.3 % for February, which are slightly lower than 19.4 % for 1-13 January, and slightly higher than 27.5 % and 27.8 % for 14-31 January and February in Shijiazhuang.The total sources outside Hebei contribute 14.5 %, 15.7 %, 20.3 %, and 19.0 % for 1-13 and 14-31 January, February 2013, and December 2007, respectively.Figure 9 shows a significant increase in the contributions of HN, namely, the maximum value could reach around 30 % and the median values could fall into the range of 0.5 % to 9 % for the period 14-31 January 2013.Such contributions are sizeable and should not be ignored in the development of air pollution control strategies in Xingtai.Compared to Shijiazhuang, the contributions from NHB significantly decrease, but are not negligible (∼ 10 % on average) and should thus also be taken into account.Although the mean contributions from SX and HN are relatively small, their variations are notably larger than those for Shijiazhuang, indicating that those sources should not be neglected under some special conditions.The contributions from BJTJ are the smallest and mostly concentrated into a small range.It is considered to be a minor source to PM 2.5 in Xingtai.In summary, the source apportionment in Xingtai might be more complicated than that of Shijiazhuang because of its special location, smaller city area, and more complex emission environment outside the city.

Handan
Handan is located at the southern edge of Hebei Province, in the intersection area of Hebei, Shanxi, Henan, and Shandong.It is expected that the contributions from SD and HN are obviously larger than those for Shijiazhuang and Xingtai.As shown in Table 5, 73.1 % and 65.2 % of PM 2.5 originate from the local sources in 1-13 and 14-31 January, respectively, which are higher than 64.2 % in December 2007 and 63.0 % in February 2013.The contributions of NHB and HN to PM 2.5 concentrations are 10.3 % and 9.5 %, respectively, in 14-31 January, which are nearly two times their contributions in December 2007.This change may be attributed to the different meteorological conditions between the two months.In December 2007, more frequent eastern and western winds appeared over this area that increased the contributions of SX and SD but decreased those from NHB and HN (L.T. Wang et al., 2012).SD, SX, and BJTJ contribute to a total of 8.3 % of the PM 2.5 concentrations in 1-13 January, 9.4 % in 14-31 January, and 11.9 % in February, which are lower than 14.4 % in December 2007.During the two periods of 1-13 and 14-31 January, a total of 22.7 % and 29.2 % of PM 2.5 in Handan came from the regional sources, respectively, and 18.8 % and 18.9 % came from the sources outside Hebei Province, respectively.The regional contributions (i.e., total sources outside SHB) are 30.7 % and 24.8 % in February 2013 and December 2007, in particular, the contributions of the sources outside Hebei account for most of the regional contributions (i.e., 22.8 % and 19.8 %, respectively) indicating that the sources from northern Hebei play a relatively small role in high PM 2.5 concentrations in Handan.Therefore, regional joint prevention and control of air pollution are extremely necessary to improve air quality in Handan.
Figure 9 shows an increased importance of the contributions from HN and SD to air quality in Handan.Emissions from HN are of the same importance to those of NHB, in other words, the 75 % percentiles are above 15 % for 1-13 January, 12 % for 14-31 January, and 15 % for February.The maximum source contributions are as high as 64 %, 56 %, and 65 % for the three periods, respectively.Although the average contributions from SX and SD are not significant (2.9 % to 6.2 %), the large variations indicate their importance to air quality in Handan under certain atmospheric conditions (e.g., when slow or calm winds in the east-west direction dominated).BJTJ remains as a minor contributor to Handan, which is similar to the cases in Shijiazhuang and Xingtai.

Inter-influences of the three cities
Figure 10 shows the source contributions of three cities, Shijiazhuang, Xingtai, and Handan to the PM 2.5 concentrations in each city for the period 1-13 and 14-31 January, and February, to understand of their influences on each other.The source contributions of Xingtai and Handan to Shijiazhuang are 2.7 % and 1.5 %, respectively, on the whole period average.The maximum contributions are about 20 % for Xingtai, and 30 % for Handan.Overall, 63.8 % of PM 2.5 in Shijiazhuang comes from the emissions within this city.As to each period, the contributions of Shijiazhuang's own sources are 71.8 % for 1-13 January, 64.0 % for 14-31 January, and 59.9 % for February to Shijiazhuang.
The situations for Xingtai and Handan are quite different from that for Shijiazhuang.The source contributions from sources from Xingtai or Handan are obviously lower than those in Shijiazhuang.On average 46.0 % of PM 2.5 in Xingtai comes from the emissions within the city.The source contributions from Shijiazhuang and Handan are 9.0 % and 10.4 %, respectively.The maximum contribution could reach around 50 %, as shown in Fig. 10b.In Handan, the contributions of its own sources are 49.4 % on the whole period average, and Shijiazhuang and Xingtai contribute 4.7 % and 9.9 % on average, respectively.The highest contributions from Shijiazhuang and Xingtai could reach 40 % and 50 %, respectively.The deference between Shijiazhuang and the other two cities may be attributed to their locations (e.g., Shijiazhuang is in the north of the three cities and the northern wind was more frequent in this 2013 case), different emission densities, and different areas (e.g., Shijiazhuang is the largest city of the three and has the highest emission densities).It implies the necessity of joint pollution controls of the three cities.

Shijiazhuang
The average contributions by region and sector to the PM 2.5 concentrations in Shijiazhuang during the period 14 January to 28 February are presented in Table 5b.The industrial and domestic emissions are the top two major local sources, with the concentration contributions of 29.7 % and 24.6 %, respectively.Agricultural emissions, the major source of NH 3 , are the third largest contributor to PM 2.5 in Shijiazhuang.This is expected because ammonium (NH + 4 ) transformed from NH 3 is one of the most important compositions in atmospheric secondary inorganic aerosols.Transportation and power plants in local regions contribute 4.4 % and 0.6 % of the PM 2.5 concentrations, respectively.As for the other regions, industrial and domestic sources in the NHB, BJTJ, and SX are also major contributors, contributing a total of 11.3 %, 2.4 %, and 3.4 %, respectively, of the PM 2.5 concentrations in Shijiazhuang.Overall, 2.7 %, 0.2 %, and 0.5 % of PM 2.5 originate from the agricultural sources in NHB, BJTJ, and SX, respectively.Transportation and power plants in those regions are minor sources to PM 2.5 in Shijiazhuang, with nearly zero contributions.The agricultural sources from SD and HN become the second most important contributor (after the domestic sources), with the contributions of 0.9 % and 0.7 % to PM 2.5 in Shijiazhuang.SD and HN are both among the large agricultural provinces in China, and SD ranks the first among all provinces in terms of agricultural GDP.Industrial emissions in the two provinces only contribute 0.4 % and 0.3 % in the PM 2.5 concentration, respectively.Transportation and power plants are also minor sources.The total contributions from each sector, roughly estimated by adding the values from each region, are 0.4 % from power plants, 36.2 % from industrial, 38.0 % from domestic, 4.2 % from transportation, and 14.5 % from agriculture.The contribution ranges are plotted in Fig. 11a for 14-31 January and February separately, for each sector in each region.Despite the importance of the three sources, industrial, domestic, and agriculture, as discussed above, the contributions of industrial manufacturers slightly increase in SHB, BJTJ, and SD areas in February compared to January; however, the change in domestic sources in SHB and NHB are the opposite.The three major sectors in the local region should not be neglected in future controls of air pollution in Shijiazhuang, and also in NHB, BJTJ, and SX when regional joint controls are considered in policymaking.

Xingtai
As shown in Table 5b, the relative importance of the sectors in each region to the PM 2.5 concentrations in Xingtai is similar to Shijiazhuang.Industrial and domestic sources are the two most important sources in the local region, which contribute 27.3 % and 27.6 %, respectively, of the PM 2.5 con-centrations.Those two sectors from NHB, BJTJ, and SX are also small but not negligible, with the contributions of 2.2 %, 1.0 %, and 1.6 % for the industrial sector and 5.1 %, 1.5 %, and 1.9 % for the domestic sector, respectively.A difference between the contributions in Xingtai and Shijiazhuang lies in the notable increase in agricultural contributions from SD and HN, i.e., 1.3 % and 2.7 % of PM 2.5 are originated from agricultural sources in the two provinces, respectively.It should be noted that the contributions from agricultural sources of HN even exceed its domestic sources (2.4 %).It may imply that the NH + 4 in the secondary aerosols in Xingtai originates more from regional sources than in Shijiazhuang.The total contributions of each sector are 0.3 % from power plants, 34.2 % from industrial, 40.5 % from domestic, 1.8 % from transportation, and 16.8 % from agriculture.
As shown in Fig. 11b, the most significant difference between the source contributions in Shijiazhuang and Xingtai is the obvious increases in the contributions of the sources from the two provinces, SD and HN, especially for the three key sectors (industrial, domestic, and agricultural sectors).It indicates that despite the controls in the industrial and domestic emissions in those provinces (which are usually emphasized in previous control strategies), controls of the agricultural emissions in the neighborhood provinces should be further considered to more effectively mitigate PM pollution in Xingtai.

Handan
The two major local sources from SHB, industrial and domestic, contribute 28.8 % and 25.1 %, respectively, of the total PM 2.5 concentrations in Handan, followed by 8.8 % from agricultural sources.Given the special location of Handan, the contributions from SD and HN further increase compared to Shijiazhuang and Xingtai.The agricultural emissions of SD and HN contribute 1.4 % and 4.1 % in PM 2.5 , respectively, which are slightly higher than those from domestic sources (2.3 % and 3.7 %) and noticeably higher than those of industrial sources (0.7 % and 2.5 %).The total contributions of each sector are roughly estimated to be 0.3 % for power plants, 35.8 % for industrial, 38.1 % for domestic, 2.9 % for transportation, and 16.8 % for agriculture.
Compared with Shijiazhuang and Xingtai, a significant increase in the contributions of all the sectors in HN is found for Handan, as shown in Fig. 11c, especially for agriculture, domestic, and industrial.The industrial, domestic, and agricultural sources in HN contribute 2.5 %, 3.7 %, and 4.1 %, respectively, of the PM 2.5 concentrations in Handan.The SD's contributions are slightly higher in Handan than those in Xingtai.As discussed above for Xingtai, more effective PM controls in Handan should include not only the major sources in local regions, but also domestic, agricultural, and industrial emissions from SD and HN.

Conclusions
In this study, the haze frequencies are first analyzed for the seven representative cities in northern China, i.e., Shijiazhuang, Xingtai, Beijing, Tianjin, Taiyuan, Zhengzhou, and Jinan for the years 2001 to February 2013.January 2013 and December 2007 are identified as the top two most polluted months since 2001 in terms of the haze frequencies.However, more frequent foggy days and much lower visibilities indicate January 2013 to be the most polluted month.The MM5-CMAQ air quality modeling system is then applied to simulate the extremely severe haze events that occurred in eastern China in January and February 2013, using the MEIC inventory.The model performance is evaluated against the meteorological observations throughout off of China and against air quality observations at more than 400 national sites in the 74 major environmental protection cities.The model shows an overall acceptable performance and the MEIC inventory are proven to give a good estimation on total emissions of cities.The relative humidity at 2 m and wind speed at 10 m are well reproduced by the model, while temperature at 2 m and daily precipitation are underpredicted.PM 2.5 and PM 10 concentrations are underpredicted during extremely polluted episodes due to the biases in the meteorological predictions, the uncertainties in the spatial allocation of emissions, the lack of dust emissions and the weakness of the model in the treatment of the heterogeneous formation of secondary aerosols on the surface of atmospheric particles.
The contributions by source regions and by source sectors in each region, to the PM 2.5 concentrations in Shijiazhuang, Xingtai, and Handan (which are listed as the top three most polluted cities in China), are analyzed using CMAQ with the Brute Force method and the results are compared with those for December 2007.Even in the top two most polluted months (i.e., January 2013 and December 2007) a large fraction of PM 2.5 in those cities may originate from quite different regional sources.In January 2013, more PM 2.5 in the three cities came from the northern area, such as NHB and BJTJ.The contributions from SX significantly decrease and those from HN obviously increase in January 2013 compared with those in February 2007.In general, the regional sources contribute 27.5 % (Shijiazhuang) to 28.6 % (Handan) of the PM 2.5 concentrations in the three cities in January 2013.Within the three cities, only 4.2 % of PM 2.5 in Shijiazhuang comes from Xingtai and Handan, but 19.5 % of PM 2.5 in Xingtai and 14.6 % in Handan come from the other two cities.
Industrial and domestic emissions from the local sources contribute ∼ 28 % and 27 %, respectively, of the PM 2.5 concentrations in the three cities, followed by the local agricultural sources.The domestic and industrial sources in the NHB, BJTJ, and SX are the two most important contributors to PM concentrations in Shijiazhuang, Xingtai, and Handan.However, the top two largest contributors in SD and HN to PM 2.5 concentrations in Xingtai and Handan are domestic and agricultural emissions from SD and HN.The special locations, huge amounts of pollutants emissions, and highly complex emission environment, pose a significant challenge to mitigating air pollution in southern Hebei to meet the national standards for PM 2.5 and PM 10 concentrations.Air pollution control in the three major cities in southern Hebei indicates potential problems in the existing framework and mechanism of regional joint policymaking (e.g., lack of coordination among provinces for regional pollution control efforts), control strategies (e.g., ignoring agricultural NH 3 emissions control), and effective enforcement of control measures (e.g., lack of a mandatory implementation action system that is similar to the State Implementation Plan (SIP) in the US).
This study has several limitations.First, uncertainties (e.g., uncertainties in the spatial allocation) and limitations in emissions (e.g., the lack of an online dust emission module) may be propagated to the model predictions of PM 2.5 and PM 10 concentrations.Second, the observational data of PM 2.5 compositions can provide insights into the causes for underpredictions during the extremely polluted periods over southern Hebei.Such data, however, are unavailable for model evaluation in this work.Third, although BFM is ca-pable of quantifying the direct and indirect effects due to interactions among PM and PM precursors, it cannot attribute 100 % of atmospheric pollutants concentrations to source emissions as does a true source apportionment method.Better understanding can be expected in the future by intercomparing source contributions using both source sensitivity and source apportionment methods.Nevertheless, this study provides valuable scientific insights into the extremely severe haze that occurred over those cities and thus lays a foundation to the future policymaking, not only within the Hebei Province but also for the entire region of China, where severe pollution episodes frequently occur.

Fig. 1 .
Fig. 1.CMAQ modeling domains at a horizontal grid resolution of 36 km over East Asia (Domain 1 with 164 × 97 cells) and 12 km over an area in northeastern China (Domain 2 with 93 × 111 cells).The zoom-in area (Domain 2) shows the regions focused in the analysis of source contributions.It includes six regions of interest: southern Hebei (which includes three cities: Shijiazhuang, Xingtai, and Handan), northern Hebei (which includes the remaining areas in the Hebei Province), Shanxi Province (SX), Shandong Province (SD), Henan Province (HN), and a two-city cluster region covering Beijing and Tianjin (BJTJ).

Fig. 2 .
Fig. 2. The monthly average number of hazy days that occurred in the seven representative cities in Domain 2, including Shijiazhuang, Xingtai, Beijing, Tianjin, Taiyuan, Zhengzhou, and Jinan, during January 2001 to February 2013.The lines indicate the minimum and maximum number of hazy days in the seven cities.

Fig. 3 .
Fig. 3.The monthly total number of (a) hazy and foggy days and (b) the average visibilities in the seven representative cities in northern China in December 2007, and January and February 2013.The line in (b) indicates the minimum and maximum daily visibilities.

Fig. 4 .
Fig. 4. (a) Overlay of the simulated and observed monthly average concentrations of PM 2.5 and PM 10 of the reported 74 cities and (b) the corresponding spatial distributions of NMBs for PM 2.5 and PM 10 at 36 km horizontal grid resolution for January and February, 2013.Note the average for January is calculated from 14 January to 31 January due to the lack of data in 1-13 January.Square signs indicate the observations.

Fig. 5 .
Fig. 5. (a) Overlay of the simulated and observed monthly-average concentrations of PM 2.5 and PM 10 of the reported 235 sites and (b) the corresponding spatial distributions of NMBs for PM 2.5 and PM 10 at the 12 km horizontal grid resolution for January and February, 2013.Note the average for January is from 14 January to 31 January.Square signs indicate the observations.

Fig. 6 .
Fig. 6.The temporal variations of the observed and simulated PM 2.5 concentrations at the monitoring sites in Shijiazhuang, Xingtai, and Handan in January and February, 2013.(a-d) are the sites in Shijiazhuang, (e, f) are the sites in Xingtai, and (g-i) are the sites in Handan.

Fig. 7 .
Fig. 7.The temporal variations of the observed and simulated PM 10 concentrations at the monitoring sites in Shijiazhuang, Xingtai, and Handan in January and February, 2013.(a-d) are the sites in Shijiazhuang, (e, f) are the sites in Xingtai, and (g-i) are the sites in Handan.

Fig. 9 .
Fig. 9.The source contribution ranges by source regions to the PM 2.5 concentrations in (a) Shijiazhuang, (b) Xingtai, and (c) Handan in December 2007, and 1-13 and 14-31 January, and February, 2013.Dot signs indicate the maximum and minimum, and square signs indicate the average value.

Fig. 10 .
Fig. 10.The source contribution ranges by the three cities of Shijiazhuang, Xingtai, and Handan to (a) Shijiazhuang, (b) Xingtai, and (c) Handan in 1-13 and 14-31 January, and February, 2013.Dot signs indicate the maximum and minimum, and square signs indicate the average value.

Fig. 11 .
Fig. 11.Source contributions by region and sector to the PM 2.5 concentrations in (a) Shijiazhuang, (b) Xingtai, and (c) Handan from 14 January to February 2013.Dot signs indicate the maximum and minimum, and square signs indicate the average value.

Table 1 .
Observational data sets for model evaluation used in this study.
a NCDC -the National Climate Data Center integrated surface database.CNEMC -the real-time database from China National Environmental Monitoring Center.HEBEU -observations at the site of Hebei University of

Table 2 .
Performance statistics for meteorological predictions over Domain 1 at 36 km resolution and Domain 2 at 36 and 12 km resolutions.

Table 3 .
Performance statistics of PM 2.5 and PM 10 concentrations over Domain 1 at 36 km resolution and Domain 2 at 36 and 12 km resolutions.

Table 4 .
MFBs and MFEs for hourly PM 2.5 concentrations simulated at the 12 km grid resolution in the representative cities.
a These statistics are for 1-31 January.b These statistics are for 1 January-28 February.n -number of data pairs.c MFB > ±60 % and MFE > 75 % are considered to indicate a relatively poor performance in this study.

Table 5a .
The average source contributions (%) to the PM 2.5 concentrations in Shijiazhuang, Xingtai, and Handan by source regions:

Table 5b .
The average source contributions (%) to the PM 2.5 concentrations in Shijiazhuang, Xingtai, and Handan by source regions and sectors (for the period 14 January-28 February):