Enhanced sulfate formation during China’s severewinter haze episode in January 2013 missingfrom current modelsYuxuan Wang1,2,3, Qianqian Zhang1,4, Jingkun Jiang4,5, Wei Zhou4, Buying Wang4, Kebin He4,5,Fengkui Duan4, Qian Zhang4, Sajeev Philip6, and Yuanyu Xie1

1Ministry of Education Key Laboratory for Earth System Modeling, Center for Earth System Science, Institute for GlobalChange Studies, Tsinghua University, Beijing, China, 2Department of Marine Sciences, Texas A&M University at Galveston,Galveston, Texas, USA, 3Department of Atmospheric Sciences, Texas A&MUniversity, College Station, Texas, USA, 4School ofEnvironment, Tsinghua University, Beijing, China, 5State Environmental Protection Key Laboratory of Sources andControl of Air Pollution Complex, Beijing, China, 6Department of Physics and Atmospheric Science, Dalhousie University,Halifax, Nova Scotia, Canada

Abstract A regional haze with daily PM2.5 (fine particulate matters with diameters less than 2.5 μm)exceeding 500 μg/m3 lasted for several days in January 2013 over North China, offering an opportunityto evaluate models. Observations show that inorganic aerosols (sulfate, nitrate, and ammonium) are thelargest contributor to PM2.5 during the haze period, while sulfate shows the largest enhancement ratio of5.4 from the clean to haze period. The nested-grid GEOS-Chem model reproduces the distribution of PM2.5

and simulates up to 364 μg/m3 of daily maximum PM2.5. Yet on average, the model is a factor of 3 and 4lower in PM2.5 and fails to capture the large sulfate enhancement from the clean to haze period. A doublingof SO2 emissions over North China, along with daily meteorology corrections, would be required toreconcile model results with surface SO2 observations, but it is not sufficient to explain the modeldiscrepancy in sulfate. Heterogeneous uptake of SO2 on deliquesced aerosols is proposed as an additionalsource of sulfate under high-relative humidity conditions during the haze period. Parameterizing thisprocess in the model improves the simulated spatial distribution and results in a 70% increase of sulfateenhancement ratio and a 120% increase in sulfate fraction in PM2.5. Combined adjustments in emissions,meteorology, and sulfate chemistry lead to higher sulfate by a factor of 3 and 50% higher PM2.5,significantly reducing the model’s low bias during the haze.

1. Introduction

PM2.5 (fine particulate matters with diameters less than 2.5μm) is known to harm human health andpublic welfare by causing haze and visibility degradation. PM2.5 is a mixture of a variety of chemicalspecies (i.e., inorganic aerosols, organic aerosols, black carbon, dust, and sea salt), which all have differentclimate implications through radiation and cloud. The air quality guideline for PM2.5 suggested by theWorld Health Organization (WHO) is 25μg/m3 for 24 h mean [World Health Organization, 2005]. In recentyears, regional haze with PM2.5 levels exceeding tenfolds of the WHO standard has become the largest airquality concern in China. China established its first PM2.5 National Ambient Air Quality Standard in 2012(75μg/m3 for 24 h mean) and has begun to regularly monitor and report PM2.5 concentrations in majorcities since late 2012. Many recent publications present measurements of chemical compositions of PM2.5

and optical properties during the regional haze periods over China [Sun et al., 2006; Shen et al., 2009; Tanet al., 2009; Kang et al., 2013; Zhao et al., 2013; He et al., 2014]. The secondary inorganic aerosols (mainlysulfate, nitrate, and ammonium, SNA) and the organic matter (OM) are found to be the most importantspecies in PM2.5 during the haze [Wang et al., 2006; Zhao et al., 2013]. Recent observations from Beijingindicate that the fraction of secondary particles, including SNA and secondary organic aerosols (SOA),increases steeply during high PM2.5 episodes [Tan et al., 2009; Zhao et al., 2013]; and particularly, theproportion of SNA in PM2.5 increases as PM2.5 concentrations increase [He et al., 2014; Zheng et al., 2014].To better protect the health of millions of people, the challenge remains whether we understand theformation mechanism of high PM2.5 episodes well enough to guide the formation of effectivecontrol strategies.

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,425

PUBLICATIONSJournal of Geophysical Research: Atmospheres

RESEARCH ARTICLE10.1002/2013JD021426

Key Points:• Sulfate enhancement is the chemicaldriver for January 2013 winter haze

• Model simulates 364μg/m3 dailymaximum PM2.5 but low onsulfate fraction

• Heterogeneous oxidation identifiedas additional source of sulfateduring haze

Correspondence to:Y. Wang,yxw@tsinghua.edu.cn

Citation:Wang, Y., Q. Q. Zhang, J. Jiang, W. Zhou,B. Wang, K. He, F. Duan, Q. Zhang,S. Philip, and Y. Xie (2014), Enhancedsulfate formation during China’s severewinter haze episode in January 2013missing from current models,J. Geophys. Res. Atmos., 119, 10,425–10,440,doi:10.1002/2013JD021426.

Received 25 DEC 2013Accepted 13 AUG 2014Accepted article online 18 AUG 2014Published online 11 SEP 2014

Figure 1. (a–f ) Surface PM2.5 and (g–k) sulfate concentrations in January 2013. Observations of PM2.5 are shown in Figure 1a. Locations of Group A (black triangles)and Group B (red circles) cities are shown in Figure 1l. The black boxes in Figure 1b indicate North China (NC) and South China (SC). Please refer to Table 2 fordescriptions of different model simulations.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,426

January 2013 witnessed a regional haze with daily PM2.5 exceeding 500μg/m3 over North China, which lastedfor multiple days threatening the health of more than 150 million people [Jiang et al., 2014]. This month wasregarded to be the most polluted month since 2001 [Wang et al., 2014a]. While large SO2 emissions andmeteorological anomalies (such as the high-pressure system localized over this region) are suggested as causesof the severe haze [Yang et al., 2013], the formation mechanism of high PM2.5 during the haze is not wellunderstood. This January episode offers an interesting opportunity to test current understanding of sources andthe formation mechanism of PM2.5 in winter over this region as reflected in current state-of-the-art chemicaltransport models (CTMs). Previous model evaluations are typically based on observations taken in developedregions (e.g., US and Europe) where PM2.5 levels are substantially lower. Other than the occasions of dust stormand biomass burning plumes, current CTMs have rarely had the opportunity to be confronted with such highlevels of PM2.5 on a regional scale where anthropogenic emissions dominate. The first question to beinvestigated in this study is whether current CTMs can reproduce the timing, spatial scale, and concentrationlevels of PM2.5 in the haze episode of January 2013 in China. We show in what follows that the GEOS-Chem CTMhas some success in simulating PM2.5 in January 2013 but significantly underestimates the enhancement ofinorganic aerosols, particularly sulfate, during the severe haze period. Themodeling analysis, in combinationwithobservational constraints, is used to identify key processes contributing to the severe PM2.5 episode over China inJanuary 2013.

2. Observed and Simulated Characteristics of PM2.5 During the Winter Haze2.1. Observations

Daily PM2.5 mass concentrations in 74 Chinese cities in January 2013 were obtained from China Ministry ofEnvironment Protection (MEP) [http://www.pm25.in/]. Figure 1a shows the locations and monthly meansurface PM2.5 concentrations, and Figure 2a shows the latitudinal gradients of PM2.5 concentrations for the74 cities. Monthly mean PM2.5 exceeded 75 μg/m3 (China’s National Ambient Air Quality Standard for24 h mean PM2.5) at 64 cities. PM2.5 levels are higher over North China (NC; north of 34°N) than over SouthChina (SC) (Figure 1b shows the domains of NC and SC). The day-to-day variation of mean PM2.5 averagedover two city groups over NC (Groups A and B) is presented in Figure 3. The Group A cities (11 cities)

20 23 26 29 32 35 38 41 44 47

20 23 26 29 32 35 38 41 44 47

0

50

100

150

200

250

300

350

0.000.050.100.150.200.250.300.350.400.450.50

obsstdgamma_run_T2

(a)

latitude

(b)su

lfate

rat

io

Figure 2. (a) Surface PM2.5 concentrations and (b) sulfate fraction in PM2.5 as a function of latitude in China. PM2.5 obser-vations are shown in black, the standard model simulation in red, and the gamma_run T2 in blue. Please refer to Table 2 fordescriptions of different model simulations.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,427

are located over the edge of North China Plain or coastal regions, and the Group B cities (8 cities) are in theBeijing-Tianjin-Shijiazhuang city cluster (Figure 1l). The most severe pollution period is a 6 day episodeduring 10–15 January 2013, affecting both Groups A and B cities. During this episode, daily PM2.5 at fivecities exceeded 500 μg/m3 for multiple days and the maximum daily PM2.5 was 711 μg/m

3 in Xingtai.The mean PM2.5 for Groups A and B cities was 202 μg/m3 and 356 μg/m3 during this episode, respectively.We choose this severe haze episode as the focus of our study.

The MEP data does not provide chemical compositions of PM2.5. Chemical compositions of PM2.5 weremeasured at a site on campus of Tsinghua University in Beijing (TU site, lat 40°19′, lon 116°19′) during 9–17January 2013 [Cao et al., 2014]. The PM2.5 samples were taken with a high flow sampling instrument on a dailybasis, with chemical compositions analyzed in the laboratory. Daily mean mass concentrations of sulfate,nitrate, ammonium, organic carbon (OC), and elementary carbon (EC) observed at TU are presented in

0

50

100

150

200

250

300

1 4 7 10 13 16 19 22 25 28 31

1 4 7 10 13 16 19 22 25 28 31

0

100

200

300

400

500

600

(a)

T1

T2

T3

obs

std

emi

(b)

Figure 3. Time series of observed and simulated PM2.5 concentrations of the (a) Group A cities and (b) Group B cities overNorth China. Locations of the Group A and Group B (red circles) cities are shown in Figure 1l. Observations are shownin black, standard model simulation in red, the emission_run simulation in magenta, and the gamma_run simulations ingreen (T1), blue (T2), and cyan (T3). Please refer to Table 2 for descriptions of different model simulations.

9 10 11 12 13 14 15 16 170

40

80

120

160

200

240

280

320

date of January9 10 11 12 13 14 15 16 17

0

40

80

120

160

200

240

280

320

date of January

ECOCammoniumnitratesulfate

(a) observation (b) standard simulation

Figure 4. Daily mean mass concentrations of PM2.5 chemical constituents at the TU site during the time period of 9–17January 2013. (a) Observations and the (b) standard model simulation.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,428

Figure 4a. The sum of the five components is used to denote the PM2.5 mass at TU. The days before and afterthe haze period (10–15 January) are defined as the clean period (9, 16, and 17 January).

Table 1 presents the mean enhancement ratios (defined as the mean concentrations during the hazedivided by those during the clean period) of the PM2.5 chemical compositions at TU. All measured PM2.5

constituents had enhancements of more than a factor of 2 during the haze. The enhancement ratio is 4.2for SNA and 5.4 for sulfate, a much larger ratio than that of carbonaceous aerosols (OC and EC). Whenscaled to total PM2.5 mass (Figure 5a), sulfate had the largest increase from the clean period (15%) to hazeperiod (28%), whereas the fractions of OC and EC showed a decrease during the haze. This observationsuggests that secondary inorganic aerosols, especially sulfate, are the key chemical drivers for this particularwinter haze episode. This finding is in accordance with recent publications on chemical compositions of PM2.5

measured during other winter haze periods over China [Sun et al., 2006; Tan et al., 2009;Du et al., 2011;Han et al.,2013; Zhao et al., 2013; He et al., 2014].

2.2. Model PM2.5 Evaluation

We use the nested-grid GEOS-Chem CTM (version 9-01-01) with a spatial resolution of 0.5° × 0.667° [Chen et al.,2009] to simulate surface PM2.5 in January 2013, with chemical boundary conditions provided by theGEOS-Chem global simulation at a resolution of 4° × 5°. The nested-grid domain is 70°E–150°E, 11°S–55°N.GEOS-Chem includes a fully coupled treatment of tropospheric ozone-NOx-VOC-aerosol chemistry [Parket al., 2004]. Gas-aerosol phase partitioning of the sulfate-nitrate-ammonium-water system is calculatedusing the ISORROPIA II thermodynamic equilibrium model [Fountoukis and Nenes, 2007]. Formation ofSOA is implemented by Liao et al. [2007]. Mineral dust aerosols are simulated using the dust entrainmentand deposition mobilization scheme, as described by Fairlie et al. [2007]. Monthly anthropogenic emissions of

Table 1. Observed and Simulated Enhancement Ratio of PM2.5 Species at the TU Sitea

Sulfate Nitrate Ammonium SNA OC EC PM2.5

Observation 5.4 3.3 3.9 4.2 2.0 2.4 3.2Standard model simulation 2.1 2.1 2.0 2.1 1.8 2.3 1.9Emission_run 2.3 2.2 2.1 2.3 2.3 2.5 2.2Gamma_run T1 3.8 2.1 2.5 2.7 2.3 2.5 2.6Gamma_run T2 3.9 2.0 2.5 2.7 2.3 2.5 2.6Gamma_run T3 3.5 1.4 2.3 2.5 2.3 2.5 2.5

aThe enhancement ratio is defined as themean concentrations during the haze divided by those during the clean period.

clean haze clean haze clean haze

clean haze clean haze clean haze

0%

20%

40%

60%

80%

100%

ECOCNH4NO3SO4

(a) obs (b) std (c) emi

(d) T1 (e) T2 (f) T3

perc

enta

ge

0%

20%

40%

60%

80%

100%

0%

20%

40%

60%

80%

100%

0%

20%

40%

60%

80%

100%

0%

20%

40%

60%

80%

100%

0%

20%

40%

60%

80%

100%

perc

enta

ge

Figure 5. Mean mass fraction (%) of chemical components in PM2.5 during the clean and haze periods at the TU site: (a) observations; (b) the standard simulation;(c) the emission_run; (d) and the gamma_run T1, (e) T2, and (f ) T3. Refer to Table 2 for descriptions of different model simulations.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,429

NOx and SO2 over China are taken from the inventory of Zhang et al. [2009] for 2006, with adjustments of a57% increase in NOx and an 8% decrease in SO2 to represent the emissions in 2013 [Wang et al., 2013b]. NH3

emissions are from Streets et al. [2003] for 2000 with a reduction of 30% as recommended byHuang et al. [2012]for China and seasonality as implemented by Fisher et al. [2011]. Errors in the model representation of tooshallow nighttimemixing depths are corrected followingWalker et al. [2012]. The nested-grid GEOS-Chem CTMhas been evaluated and applied to the analysis of SNA and black carbon over China [Wang et al., 2013b, 2013a;Kharol et al., 2013]. Table 2 summarizes several different simulations conducted in this study.

Simulated surface PM2.5 concentrations are compared with observations both spatially and temporally(Figures 1–3). The model reproduces the observed distribution in surface PM2.5 with a spatial correlationcoefficient (R) of 0.63 and a mean bias of �39% (Figure 1). The maximum daily PM2.5 is 364μg/m

3 in themodel, a factor of two lower than the observed maximum although remaining remarkably high. The model isparticularly successful in predicting both the latitudinal gradient in PM2.5 across 30°N and the location ofmaximum concentrations over NC (Figures 1 and 2). The model has an R of 0.82 and a mean bias of �16%for cities over SC. For NC, the model has a better performance in simulating the day-to-day variation ofPM2.5 for Group A cities (R= 0.9, bias =�36%) than the Group B cities (R= 0.45, bias =�46%) (Figure 3). TheGroup B cities are located in themost polluted city cluster, and emissions from this region are underestimatedin the model. Similar to the GEOS-Chem model, the regional-scale Community Multi-scale Air Quality modelwas found unable to reproduce the high PM2.5 concentrations over Group B cities [Wang et al., 2014a].

Figures 4 and 6 compare observed and simulated chemical compositions of PM2.5 at TU on a day-to-day basis.The model is too low by several folds in simulating chemical compositions of PM2.5 during the haze period,yet the bias is comparatively much smaller during the clean period. For sulfate themodel is a factor of 4 too lowduring the haze. Since we only have observations from a single site in the urban area to compare with gridcell (~40×50 km2) average model predictions, the following analysis and discussion are emphasized onevaluating and subsequently improving the model’s ability to reproduce the key chemical features of the hazerather than the absolute concentrations.

First, the model fails to reproduce the large increase of sulfate fraction in PM2.5 during the haze (Figures 5aand 5b). The model shows this fraction to remain the same from the clean to haze period, suggesting amissing sulfate source during the haze. The model’s overestimate of the nitrate fraction in PM2.5 is caused bythe model’s underestimate of absolute concentrations of PM2.5. Second, the simulated enhancement ratiofor sulfate is 2.1, not only lower by a factor of 2.6 than the observed ratio (5.4) but not significantly differentfrom that of EC (2.3). EC is a primary pollutant with common anthropogenic sources as SO2 (e.g., residentialcoal and diesel vehicles), yet subject to little chemical processing in winter. The model’s lack of sulfateenhancement relative to EC suggests that the model’s discrepancy is more likely related to sulfate chemistryrather than physical accumulation (such as stagnation), as physical accumulation tends to exert similareffects on different aerosols. Third, the simulated fraction of sulfate in PM2.5 increases from north to south,a coincidence with the decrease of model bias in surface PM2.5 (Figure 2b). This implies an important roleplayed by sulfate regarding the model’s predictability of total PM2.5.

Table 2. A Summary of All the Simulations in This Study

Simulation Name Description

Standard simulation Model standard simulation with the 2013 inventoryDoubleSO2 run SO2 emissions from the standard simulation are doubled over NC and reduced by 30% over SC.

Changes are applied uniformly throughout January.Emission_run Total monthly emissions of SO2, OC, and EC (anthropogenic portions only) from the standard run

are increased by 100% over NC; SO2 emissions over SC are reduced by 30%; NOx and NH3emissions are the same as those in the standard run. The meteorology correction factors are

applied on a day-to-day basis to emissions of SO2, NOx, NH3, OC, and EC over NC.Gamma_run T1 Emissions are the same as in the emission_run; γ=10�4 when RH = 50%, γ=10�3

when RH = 50%; γ increases linearly with RH from 50% to 100% (equation (2)).Gamma_run T2 Emissions are the same as those in the emission_run; γ=10�3 when RH = 50%, γ=10�2

when RH = 50%; γ increases linearly with RH from 50% to 100% (equation (2)).Gamma_run T3 Emissions are the same as those in the emission_run; γ=10�2 when RH = 50%, γ=10�1

when RH = 50%; γ increases linearly with RH from 50% to 100% (equation (2)).

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,430

2.3. Evaluation of Model Meteorology

GEOS-Chem is driven by the assimilated meteorology from the Goddard Earth Observing System (GEOS-5) ofNASA Global Modeling Assimilation Office. Important meteorological factors relevant to haze formationinclude temperature, relative humidity (RH), wind speed, wind direction, and planetary boundary layer (PBL).Here we evaluate these meteorological factors in the model using meteorology observations from NationalMeteorological Information Center (NMIC) of China.

Table 3 summarizes the model’s performance in comparison with monthly mean meteorological variablesaveraged over a total of 676 NMIC sites in China. The comparison statistics includes the spatial correlation,normalized mean bias (NMB), and mean normalized bias (MNB) as introduced by Zhang et al. [2012]. Themodel shows a strong correlation with observed RH (R= 0.64) and temperature (R=0.96), with MNB andNMB within ±10% for both variables. The model shows a poorer correlation with observed wind speeds andtends to overestimate wind speeds under low wind conditions.

Figure 7 compared the observed and simulated meteorology at selected meteorological sites over NC.The model successfully reproduces the day-to-day variations of RH, temperature, and wind speeds withcorrelation coefficients larger than 0.8 (Figures 7a and 7b). The model shows a bias of within ±20% forboth RH and temperature, yet a relatively larger bias for wind speeds (NMB= 37.6%). The modelreproduces the frequency distribution of wind directions over NC with a correlation coefficient of 0.75. Asshown in both observations and the model, North and NNW winds are most frequent in January 2013over NC. As shown in Figure 7, the model discrepancy in the selected meteorological variables is

9 10 11 12 13 14 15 16 17

0

50

100

150

200

250

300

350

400

9 10 11 12 13 14 15 16 17

0

30

60

90

120

9 10 11 12 13 14 15 16 17

0

20

40

60

80

9 10 11 12 13 14 15 16 17

0

20

40

60

80

9 10 11 12 13 14 15 16 17

0

3

6

9

12

15

9 10 11 12 13 14 15 16 17

0.0

0.1

0.2

0.3

0.4

0.5

0.6

concentration,ug/m

3

obs

std

met

T1

T2

T3

(a) PM2.5

(b) sulfate

(c) nitrate (d) OC obs

std

met correction

DATE

(e) EC

CF,unitless

ECMWF

MODIS

GEOS-5

DATE

(f) cloud fraction over NC

Figure 6. (a) Daily mean PM2.5, (b) sulfate, (c) nitrate, (d) OC, and (e) EC concentrations at the TU site. Observations are inblack, the standard simulation in red, the emission_run in magenta, and the gamma_run simulations in green (T1), blue(T2), and cyan (T3). (f ) Daily cloud fraction over North China from the model (red), ECMWF reanalysis (blue), and MODISretrieval (green). Please refer to Table 2 for descriptions of different model simulations.

Table 3. Observed and Simulated Meteorological Variables in January 2013 at 676 Meteorology Monitoring Sites Fromthe National Meteorological Information Center (NMIC) of China

Observation Mean Model Mean Spatial Correlation Model MNB Model NMB

T (°C) �2.8 �3.0 0.96 6% �3.2%RH (%) 63.2 64.8 0.64 6.6% 2.5%Wind speed (m/s) 1.9 2.0 0.34 19.2% 4.6%

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,431

systematic across different days in January. Given that the model exhibits similar capacities in simulatingmeteorology for both the clean and haze periods, we argue that the model’s particular ineffectivenessin predicting PM2.5 and sulfate during the haze period cannot be attributed to its systematic bias inmeteorology. Additional evidence to support our argument is that the model successfully reproduces theobserved enhancement ratio of EC from clean to haze days (Table 1), which is mainly associated withchanging meteorological conditions.

The PBL height is a key parameter affecting pollutant concentrations near the surface. The GEOS-Chemmodel predicts that the average PBL height during the period 10–15 January 2013 is 275m over NC and290m over Beijing. Huang et al. [2014], based on the National Climate Data Center reanalysis, indicated anaverage PBL height of 234.2m over Beijing during the same period, comparable to the GEOS-Chem value.Wang et al. [2014], using aerosol lidar measurements at an urban site in Beijing (CAS/IAP site), reported thedaily maximum PBL height over Beijing of 400–800m during 10–13 January 2013. By comparison, meanafternoon PBL from GEOS-Chem is 573m over NC. Since we do not have sufficient observations of PBL toevaluate the model, we conducted a sensitivity analysis which indicates that a 50% decrease in the PBL depthduring the haze period (10–15 January) will result in a 64% increase of sulfate and a 61% increase of PM2.5

concentrations at the surface over NC. Although this shows a relatively large sensitivity, it is still far fromsufficient to explain a factor of 2.5 underestimation in simulated concentrations during the haze period. In theabsence of PBL observations, we will take an indirect approach in section 3.2 to compensate for potentialerrors in meteorology through the adjustment of precursor emissions of PM2.5.

3. Sulfate Formation During the Winter Haze

The analysis above presents that enhanced sulfate formation is one of the key chemical drivers contributingto the severe haze episode during 10–15 January 2013, and the model has difficulty reproducing this. Thissection will focus on reconciling sulfate simulations with the observational constraints presented abovethrough the analysis of precursor emissions and sulfate chemistry. Mechanistic understanding will be obtainedwith regard to the key factors responsible for sulfate enhancement during the haze period.

3.1. SO2 Simulation

Sulfate formation is determined by SO2 concentrations. NOx is another important precursor of inorganicaerosols. In this section we examine the model’s ability in simulating SO2 and NOx using daily concentrationsof SO2 and NO2 observed at 74 cities colocated with the PM2.5 sites. As shown in Figures 8a and 8b, the model

obsmodel

(a) RH,%

R = 0.93NMB =-12%MNB = -1%

(b) temperaure,oC

R = 0.92NMB = 15.6%MNB = 2%

(c) wind speed, m/s R=0.83NMB=37.6%MNB=31%

wind direction

obsmodel

(d) frequency of wind direction

Figure 7. Day-to-day variations of surface (a) RH, (b) temperature, and (c) wind speed and (d) frequency distribution ofwind directions over NC in January 2013. Observations of RH, temperature, and wind speeds are averages from 84 NMICmeteorology monitoring sites over NC. Frequency distribution of wind directions are compiled from 15 sites with closeproximity to the PM2.5 sites over NC. Observations are shown in black and model results in red. The model’s temporalcorrelations with observations and biases (MNB and NMB) for RH, temperature, and wind speed are also shown.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,432

underestimates mean SO2 over NC by 50% while overestimates it over SC by 30%, resulting in a spatialcorrelation of only 0.32 with observations. The overall model bias is�10% for SO2 concentrations over Chinaas a whole. When compared with monthly mean NO2 distributions, the model shows a better spatialcorrelation (0.5) and a smaller bias (�5%) (Figures 8c and 8d).

The day-to-day variations between observed and simulated SO2 and NO2 over NC are shown in Figure 9.In comparison with mean SO2 observations at Groups A and B cities, the model shows good temporalcorrelations of 0.6 and 0.85, and mean biases of �48% and �37%, respectively. The model has a relativelygood ability in simulating the day-to-day variability of NO2 over NC (Figures 9c and 9d) with a much smallerbias, suggesting a low bias in NOx emissions. The good temporal correlation with regard to SO2 and NO2

observations, particularly for Group B cities, provides additional evidence for the model’s ability in simulatingthe day-to-day changes in meteorology.

The underestimation of SO2 over NC is a legitimate reason for the model underestimation of sulfateconcentrations. To mitigate the mean bias in SO2, monthly total anthropogenic emissions of SO2 in themodel are increased by 100% over NC and decreased by 30% over SC. This is referred to as the “doubleSO2_run.”The resulting changes in simulated SO2 concentrations (Figures 9a and 9b) are sufficient to eliminatethe mean model bias of SO2 for January as a whole, although model discrepancy still exists at the daily leveland there is a 10% overestimation for Group B cities. A 100% increase of SO2 emissions results in a 60%increase of sulfate over NC. The impact of the increase in SO2 emissions on sulfate will be presented inmore detail in section 3.3.

3.2. Meteorology Correction

Anthropogenic emissions of SO2 are not expected to vary significantly on a day-to-day basis in January. Sincethe doubleSO2_run has successfully corrected for the mean bias of SO2, the remaining model discrepancy inSO2 at the daily level can be presumably attributed to model errors in meteorology (such as PBL and windspeeds) and transport which cannot be corrected directly. As indirect measures to correct for such modelerrors, day-to-day adjustment factors are applied to SO2 emissions from the doubleSO2_run in order to

10oN

15oN

20oN

25oN

30oN

35oN

40oN

45oN

50oN

10oN

15oN

20oN

25oN

30oN

35oN

40oN

45oN

50oN

80o E 120o E

0 8 16 24 32 40 48 56 64 72 80 [ppbv]

0 6 12 18 24 30 36 42 48 54 60 [ppbv]

(a) obs SO2 (b) std SO2

(c) obs NO2(d) std NO2

100o E

80o E 120o E100o E

Figure 8. Spatial distribution of monthly mean surface concentrations of (top) SO2 and (bottom) NO2 over China in January2013. Observations are shown in left, and the standard model simulations are shown in right.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,433

reduce themodel discrepancy in SO2 on a daily basis. These daily adjustment factors, referred to as meteorologycorrection factors (MCFs), are calculated for NC as a whole based on the relative difference between theobserved NC-mean SO2 concentrations (CSO2_obs) and simulated mean SO2 concentrations from thedoubleSO2_run (CSO2_doubleSO2_run) on individual days (t); that is, MCF(t) = CSO2_obs(t)/CSO2_doubleSO2_run(t).Figure 9e presents the times series of the derived MCFs over NC. They range from 0.6 to 1.5 and have a meanvalue of 1.0. The meteorology correction factors tend to be larger than 1.0 during days of high PM2.5 levels,yet they rarely exceed 1.4 on those days. This suggests that the model error in meteorology is a significantcontributing factor for the model underestimate of PM2.5 during the haze, but not a dominant one.

For validation, applying themeteorological correction factors to NOx emissions improves the NO2 simulation, asshown in Figures 9c and 9d. For example, after the meteorological correction factors are applied, thetemporal correlation of the model with daily NO2 observations at Group B cities increases from 0.46 to 0.81.This finding suggests that the meteorological correction factors derived from SO2 can be applied toanthropogenic emissions of other species in the model.

3.3. Emissions

In this section we construct an improved emission scenario incorporating the bias correction schemes appliedto SO2 emissions and meteorology derived above. First, the relative changes in SO2 emissions derived insection 3.1 (i.e., a +100% increase over NC) are applied to the total monthly emissions of not only SO2 but alsoOC and EC from the standard run. Emissions of NOx and NH3 are not adjusted because analysis of NO2

simulation indicates a small bias in NOx emissions, and NH3 emissions in winter are small. The reason whyOC and EC emissions in January are increased by the same percentage (e.g., +100% over NC) as SO2 is given bythe following considerations: (1) the model underestimates OC and EC at the TU site to a large extent similar tosulfate; (2) we do not have enough observations of OC and EC over NC to constrain their emissions separately(TU is only one site); and (3) OC and EC share some common sources as SO2 in winter (e.g., coal combustion).

1 4 7 10 13 16 19 22 25 28 31 1 4 7 10 13 16 19 22 25 28 31

1 4 7 10 13 16 19 22 25 28 31

1 4 7 10 13 16 19 22 25 28 31

1 4 7 10 13 16 19 22 25 28 31

0

20

40

60

80

0

40

80

120

160

10

20

30

40

50

60

0

20

40

60

80

100

0.4

0.6

0.8

1.0

1.2

1.4

1.6

SO

2con

cent

ratio

n,pp

bvN

O2c

once

ntra

tion,

ppbv

(a) group A SO2 (b) group B SO2

obsstd_rundouble_SO2 emissionemission_run

(c) group A NO2 (d) group B NO2

meteorology correct factor

DATE

fact

or

DATE

(e) met correct factor

0

Figure 9. Daily SO2 concentration at (a) Group A cities and (b) Group B cities; Daily NO2 concentration at (c) Group A citiesand (d) Group B cities; (e) the daily meteorological correction factors over NC derived from SO2 analysis. Observations are inblack, the standard simulation in red, the doubleSO2_run in dashed green, and the emission_run in magenta. Refer toTable 2 for descriptions of different model simulations.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,434

We acknowledge that such a simple adjustment of OC and EC emissions cannot bring the model to completelymatch with OC and EC observations at TU (Figures 6d and 6e), but it significantly mitigates the model bias atthat site. A thorough investigation of the model’s large discrepancy in OC and EC during the haze period isoutside the scope of this work.

Second, the meteorology correction factors are applied to introduce day-to-day changes in anthropogenicemissions of all species over NC, including SO2, NOx, NH3, OC, and EC. The day-to-day meteorologycorrection factors are applied on top to total monthly emissions of SO2, OC, and EC which have alreadybeen adjusted in the first step. After the two steps, adjusted emissions of PM2.5 precursors are used asmodel inputs and the resulting changes in simulation are referred to as the emission_run (Table 2).

Compared with the standard run, the monthly mean PM2.5 concentrations in the emission_run are higher byan average of 35% for NC and show a better spatial correlation with observations (Figure 1c, R= 0.65).Simulated sulfate concentrations are increased by 60% over NC (Figure 1h). The model bias of PM2.5

decreases from�36% (standard run) to�27% for Group A cities, and from�51% to�39% for Group B cities.It also shows that during the haze period simulated sulfate and PM2.5 concentrations at the TU site areincreased by 45% and 33%, respectively (Table 4).

Although the emission_run improves the simulation of PM2.5 concentrations and distributions, it stillcannot reproduce the important features of sulfate during the haze. The sulfate enhancement ratio in theemission_run increases to 2.3 from 2.1 in the standard run, still more than a factor of 2 lower comparedwith the observed value (5.4) but not different from that of EC (Table 1). The emission_run does not reproducethe fractional increase of sulfate in PM2.5 from the clean to haze period (Figure 5c).

3.4. Sulfate Formation

Reconciling the model results with the observed sulfate enhancement during the haze period can be achievedonly by increasing chemical formation of sulfate. In the standard simulation, sulfate formation includes in-cloudoxidation of SO2 by hydrogen peroxide (H2O2) and ozone (O3) and SO2 oxidation by OH in the gas phase.The reaction rate of aqueous phase SO2 oxidation in the cloud is typically faster than that of the gas phaseoxidation. The cloud fraction (CF) and related parameters such as cloud liquid water content (LWC) areimportant parameters that determine the in-cloud formation rate of sulfate. The day-to-day variation of CF overNC from the GEOS-Chem model is presented in Figure 6f. By comparison, CFs from the ECMWF (EuropeanCentre for Medium-range Weather Forecasts) reanalysis and the MODIS (Moderate Resolution ImagingSpectroradiometer) satellite retrievals are also shown. The mean CF during the period of 9–17 January in theGEOS-Chemmodel is similar to that from the ECMWF reanalysis, while MODIS indicates a much larger CF whichmay be due to the retrieval bias in winter [Zhao and Girolamo, 2006; Kotarba, 2009]. A notable difference isthat the model does not show an increase of CF on 14 January as indicated by ECMWF and MODIS. Sincethe in-cloud oxidation in GEOS-Chem is parameterized upon LWC, a sensitivity analysis with a 20% increase inthe GEOS-Chem LWC is conducted, which results in a 14% increase of simulated sulfate concentrations, still farfrom sufficient to explain the large bias of sulfate on 14 January.

As in many other CTMs, in-cloud SO2 oxidation by O2 catalyzed by transitional metal irons (TMI) [Alexanderet al., 2009] is not included in the GEOS-Chem simulation presented here. He et al. [2014] indicated that dustaccounted for an average of 10% in PM2.5 in Beijing during the period of 9–14 January 2013, whereas the

Table 4. Observed and SimulatedMean Concentrations (μg/m3) of Sulfate, Nitrate, and PM2.5 During the Clean and HazePeriods at the TU Site

Sulfate Nitrate PM2.5

Clean Haze Clean Haze Clean Haze

Observations 12.7 69.2 11.7 38.8 67.9 219.6Standard_run 5.3 11.3 16.5 33.6 36.6 73.5Emission_run 7 16.4 17.5 39.1 43.2 97.8Gamma_run T1 8.2 31 13.5 28.3 39.7 102.6Gamma_run T2 9.3 36.3 13 26 40.7 106.2Gamma_run T3 15.1 52.4 11.7 16.5 45.6 112.5

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,435

GEOS-Chem simulates only 10μg/m3 of dust concentrations in Beijing. However, the analysis above indicatesthat cloud fraction over NC during the severe haze episode is not significantly different from that duringthe clean conditions (Figure 6f). Therefore, additional in-cloud oxidation process is not expected to increasethe enhancement ratio of sulfate in the model.

The severe haze period is found to have higher RH (Figure 7a). Under high RH conditions that typically prevailduring the haze [Sun et al., 2013b], preexisting deliquesced aerosols provide a potential environment otherthan clouds for aqueous phase SO2 oxidation. A number of studies suggest that the conversion of SO2 tosulfate via heterogeneous reactions on the surface of aerosols can be very significant in the presence ofcatalyst (such as TMI) or alkaline gases like ammonia under appropriate RH [Chameides and Stelson, 1992;Berglen et al., 2004; Li et al., 2011; Wang et al., 2012b; Harris et al., 2013]. SO2 conversion rate on theseparticles is found to range from 1% h�1 [Kasibhatla et al., 1997; Turšič et al., 2003, 2004; Berglen et al., 2004]to 5% h�1 [Turšič et al., 2003]. However, this pathway is not well established mechanically and is not widelyadopted by current CTMs.

The heterogeneous reaction of SO2 on the surface of dust aerosols is relatively well studied in the literature[Usher et al., 2002, 2003; Song et al., 2007; Li et al., 2011;Wang et al., 2013b; Harris et al., 2013; He et al., 2014;Dentener et al., 1996; Zhang and Carmichael, 1999; Bauer and Koch, 2005; Fairlie et al., 2010; Wang et al.,2012a]. The complex heterogeneous reaction is usually parameterized in CTMs as a pseudo first-orderreaction of SO2, with the most important parameter being the uptake coefficient, γ [Jacob, 2000]. The reactionrate constant K (s�1) is calculated as follows [Song et al., 2007; Wang et al., 2012a]:

K ¼ R=Di þ 4= vi�γð Þ½ ��1� A (1)

where R is the aerosol radius (cm), Di is the gas phase molecule diffusion coefficient (cm2/s), vi is the meanmolecular speed (cm/s), and A is the aerosol surface area (cm2).

A wide range of the uptake coefficient γ has been given in the literature, from 10�4 to 0.1, and it is oftenassumed that γ increases as RH increases [Dentener et al., 1996; Zhang and Carmichael, 1999; Usher et al., 2002;Fairlie et al., 2010; Harris et al., 2013].

We adopt the parameterization in equation (1) to simulate the heterogeneous oxidation of SO2 ondeliquescent aerosols other than dust under haze conditions. Since the emission_run partially correctsthe mean bias of the model, the heterogeneous chemistry is added on top to the emission_run and isreferred to as the gamma_run. Considering the wide range of γ given in the literature, we conducted aseries of sensitivity simulations using three different ranges of γ: 10�4–10�3 (gamma_run T1), 10�3–10�2

(gamma_run T2), and 10�2–10�1 (gamma_run T3). We further assume a linear increase of γ with RH when RHis between 50% and 100%, as shown in equation (2):

γ ¼ γ_RH50% þ γ_RH100% � γ_RH50%ð Þ= 100%� 50%ð Þ � RH� 50%ð Þ (2)

where γ_RH50% and γ_RH100% represent the lower and upper range of γ, respectively, adopted in T1, T2, or T3. Thelower limit of 50% for RH is adopted from Sun et al. [2013b] which reported that the sulfur oxidation rateincreases almost linearly with RHwhen RH is above 50%.We assume the same γ for all the aerosol species.Whilesome studies assume a dependence of the uptake of SO2 by mineral dust on dust alkalinity [e.g., Liao et al.,2004], we do not implement this dependence in the heterogeneous uptake by nondust aerosols due to the lackof concrete evidence from laboratory studies.

Compared with the emission_run, the gamma_run simulations result in sulfate increases ranging from 18.5%(T1) to 137% (T3), and PM2.5 increases from 4% to 23%, respectively (Figure 1). The gamma_run T3 simulationreduces the model bias of PM2.5 to �8%. Spatially, the gamma_run simulations correlate better with PM2.5

observations (R=0.65–0.69 for T1 to T3, Figure 1) than the emission_run and standard simulation. This isbecause the parameterization introduces additional variability in aerosols associated with ambient RH. Sincetotal surface areas of aerosols and RH variability are smaller over SC than NC, the simulated sulfate concentrationsover SC increase (18%–100% from T1 to T3) less than those over NC (40%–150% from T1 to T3) due tosmaller surface areas of aerosols. As a result, the gamma_run successfully preserves the latitudinal gradientsin sulfate and PM2.5 (Figure 2b).

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,436

For Group A cities (Figure 3a), themonthlymean bias of PM2.5 decreases from�27% in the emission_run to�22%(gamma_run T1) and �11% (gamma_run T3). For Group B cities (Figure 3b), the bias is reduced from �39%(emission_run) to�36% (T1) and�26% (T3). The key feature of the gamma_run simulations is that they result inlarger increases of PM2.5 during the severe haze period compared to the monthly mean. This is because theparameterization introduces additional variability associated with ambient RH in addition to the hygroscopicgrowth of aerosols. The model still cannot capture the extremely high concentrations of PM2.5 observed on 11and 12 January at Group B cities. Wang et al. [2014b] suggest that the feedback of PM2.5 on PBL evolutioncan contribute about 30% of monthly mean PM2.5 in the Hebei-Beijing-Tianjin region where the Group Bcities are located. The absence of this mechanism in our model may be one reason why the model cannotcapture the highest PM2.5 concentrations. The model’s relative coarse resolution may be another reason.

Simulation results of PM2.5, sulfate, and nitrate at the TU site from the three gamma_run tests are summarized inTable 4, along with those from the standard run, the emission_run, and observations. T3 produces the largestsulfate, but nitrate concentrations from T3 are 50% lower than observations, suggesting that γ in T3 is probablytoo high, since excessive increase of sulfate removes free NH3 from making ammonium nitrate under theNH3-limited conditions inwinter. AlthoughT2 produces 31% less sulfate than T3 does, sulfate concentrations fromT2 during the haze period are a factors of 2 and 3 higher than those from the emission_run and the standardrun, respectively. T2 has a higher enhancement factor for sulfate than T3 (Table 1). Nitrate concentrations in T2are much higher than T3. There is a 17% difference in sulfate between T1 and T2 (Table 4).

Figures 5d–5f show that the gamma_run tests result in large increases of sulfate fractions in PM2.5. Sulfatefraction during the haze period increases from 16.3% in the emission_run to 30% in T1 and 34.3% in T2,similar to the observed fraction of 31.5%. The T3 test has a more than 50% mass fraction of sulfate in PM2.5,which is too high compared with observations.

Among the three gamma_run tests, we recommend T2 as the best range of γ from 10�3 (RH = 50%) to 10�2

(RH = 100%). Although T3 produces a factor of 3 higher sulfate than the emission_run, there is no increaseof PM2.5 concentration (�1%) and sulfate fraction in PM2.5 is overestimated due to an underestimate ofnitrate at TU. T1 results in 70% and 3% higher sulfate and PM2.5 concentrations than the emission_run,respectively. By comparison, sulfate concentrations in T2 are 106% higher than the emission_run and 193%higher than the standard run, leading to a significant reduction of the model bias. T2 also shows the largestenhancement ratio for sulfate (3.9) at the TU site.

3.5. Summary and Discussion

In summary, the gamma_run (T2) results in higher sulfate by a factor of 3 and 50% higher PM2.5 during the hazeperiods (Table 4), compared with the standard simulation. This brings the model to agree within 50% ofobservations at TU, a significant improvement from the low bias of factors of 4 and 5 in the standard simulation.Although adjustments to precursor emissions and meteorology are incorporated in the gamma_run setup,only through adding the new formation pathway of sulfate on deliquesced aerosols can the model reproducethe observed increase of sulfate fraction in PM2.5 during the haze period (Figure 5e) and the larger sulfateenhancement ratio than other species (Table 1), the two key chemical features of the haze which cannot besimulated by either the standard model or the emission_run.

Based on all the sensitivity simulations conducted above and comparisons with observations, we findemission is the major factor controlling the overall PM2.5 levels over Groups A and B cities in NC. Theemission_run produces 20% higher PM2.5 than the standard run, and the gamma_run contributes anadditional 10% increase. For sulfate, however, chemistry is the largest controlling factor. For NC as a whole,a 100% increase of SO2 emissions (i.e., emission_run) results in a 60% increase of sulfate, while the newheterogeneous pathway of sulfate formation produces an additional 70% increase. Analysis of sulfateconcentrations at the TU site during the haze periods reveals that chemistry contribution (150%) is evenlarger than emission (42%).

There is no simple way to separately quantify the impact of day-to-day changes in meteorology on PM2.5,

because the chemistry processes are influenced by meteorological conditions. To assess the role ofmeteorology, we conducted a full-month simulation for January 2012 using the same anthropogenicemissions of January 2013 as in the standard run. Simulated surface concentrations of PM2.5 are on average

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,437

27% higher in January 2013 than in January 2012, suggesting meteorology as an important contributingfactor responsible for the severe haze conditions in January 2013.

The gamma_run (T2) simulation, through combined corrections in emissions, meteorology, and sulfatechemistry, reduces the model discrepancy in sulfate and PM2.5 from a factor of 4 and 5 to within 50% ofobservations during the haze period. The remaining errors of themodel are probably due to other compoundingfactors. First, due to the NH3-limited conditions in winter, the increase of sulfate caused by the gamma_runwill result in a decrease in nitrate (Table 4). That explains why the gamma_run simulation results in less significantincrease of PM2.5 than it does for sulfate. Consideration of other types of positive ions in the model, such as Ca2+

and Mg2+, will allow for more PM2.5 in the gamma_run to further reduce the discrepancy between the modeland observations. Second, we cannot quantify or correct for all the potential errors in meteorology andmodel transport. Although the meteorological correction factors are implemented in the emission_run andgamma_run, these factors are derived from SO2 analysis, thus excluding certainmeteorological variables whichmay exert larger influence to aerosols than to SO2. The relative coarse resolution of the model may be anotherfactor. Third, the model’s discrepancy in simulating carbonaceous aerosols is not investigated in this work,which also contributes to a large bias in PM2.5.

4. Conclusion

The analysis of surface PM2.5 observations suggests that enhanced sulfate formation is one of the keychemical drivers contributing to the severe haze episode during 10–15 January 2013. The GEOS-Chemmodelhas difficulty in reproducing the large enhancement of sulfate concentrations and sulfate fraction in PM2.5

from the clean to haze days. Through analysis of SO2 and meteorology conditions, we derive correctionfactors for SO2 emissions and meteorology. A doubling of SO2 emissions over North China, along with dailymeteorology correction factors, is required in the model in order to achieve a low-bias simulation of surfaceSO2. However, the corrections on precursor emissions and meteorology cannot fully explain the model’sdiscrepancy in sulfate. Therefore, missing chemical formation of sulfate is further suggested as themost likelycandidate to reconcile the model results with observed sulfate enhancement during the haze period.

We propose that heterogeneous uptake of SO2 on deliquesced aerosols associated with high RH conditionsduring the haze period should be added in themodel as an important sulfate formation pathway. This pathwayis parameterized in the model using an uptake coefficient, γ, which is a function of RH, and depends on aerosolsurface areas. When RH increases, hygroscopic growth of atmospheric aerosols results in increasing watercontent and surface areas, promoting heterogeneous SO2 oxidation to occur on the surface of these aerosols.Since this reaction is first order in SO2, it leads to fast increase of sulfate aerosols which are highly hygroscopic,creating more surface area for such reactions under suitable RH conditions and causing rapid buildup ofsulfate during the haze period. A suitable range of γ is set between 10�3 and 10�2 on the basis of sensitivity testresults. Adding this chemical process to the model results in 70% and 120% increases of simulated sulfateenhancement ratio and sulfate mass fraction in PM2.5, respectively, therefore improving the simulatedconcentration levels and spatial distribution of PM2.5. The combined corrections to emissions, meteorology, andsulfate chemistry of the model (i.e., the gamma_run T2) result in a factor of 3 higher sulfate and a 50% increasein PM2.5 during the haze periods. This brings the model to be within 50% of observations during the haze, asignificant improvement compared with the factors of 4 and 5 low biases in the standard simulation.

We acknowledge that the model shows some deficiencies in simulating carbonaceous aerosols over China.This topic should be an important issue warranting further investigation.

ReferencesAlexander, B., R. J. Park, D. J. Jacob, and S. Gong (2009), Transition metal-catalyzed oxidation of atmospheric sulfur: Global implications for the

sulfur budget, J. Geophys. Res., 114, D02309, doi:10.1029/2008JD010486.Bauer, S. E., and D. Koch (2005), Impact of heterogeneous sulfate formation at mineral dust surfaces on aerosol loads and radiative forcing in

the Goddard Institute for Space Studies general circulation model, J. Geophys. Res., 110, D17202, doi:10.1029/2005JD005870.Berglen, T. F., T. K. Bern, I. S. A. Isaksen, and J. K. Sundet (2004), A global model of the coupled sulfur/oxidant chemistry in the troposphere:

The sulfur cycle, J. Geophys. Res., 109, D19310, doi:10.1029/2003JD003948.Cao, C., W. Jiang, B. Wang, J. Fang, J. Lang, G. Tian, J. Jiang, and T. F. Zhu (2014), Inhalable microorganisms in Beijing’s PM2.5 and PM10

pollutants during a severe smog event, Environ. Sci. Technol., 48, 1499–1507.

AcknowledgmentsThe PM2.5 observations were obtainedfrom the MEP’s daily air quality reports.The PM2.5 speciation data in Figures 4and 5 and model outputs are archivedat http://166.111.42.70. This researchwas supported by the National KeyBasic Research Program of China(2014CB441302), the CAS StrategicPriority Research Program (grantXDA05100403), the Beijing NaturalScience Foundation (8122025), andBeijing Nova Program(Z121109002512052). The authorsthank T.M. Fu, J.Q. Mao, and D.J. Jacobfor helpful discussions. We thank theanonymous reviewers for theirthoughtful and constructive commentsfor the manuscript.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,438

Chameides, W. L., and A. W. Stelson (1992), Aqueous-phase chemical processes in deliquescent sea-salt aerosols: A mechanism that couplesthe atmospheric cycles of S and sea salt, J. Geophys. Res., 96(D18), 20,565–20,580, doi:10.1029/92JD01923.

Chen, D., Y. Wang, M. B. McElroy, K. He, R. M. Yantosca, and P. Le Sager (2009), Regional CO pollution and export in China simulated by thehigh-resolution nested-grid GEOS-Chem model, Atmos. Chem. Phys., 9, 3825–3839, doi:10.5194/acp-9-3825-2009.

Dentener, F. J., G. R. Carmichael, Y. Zhang, J. Lelieveld, and P. J. Crutzen (1996), Role of mineral aerosol as a reactive surface in the globaltroposphere, J. Geophys. Res., 101(D17), 22,869–22,889, doi:10.1029/96JD01818.

Du, H., L. Kong, T. Cheng, J. Chen, J. Du, L. Li, X. Xia, C. Leng, and G. Huang (2011), Insights into summertime haze pollution events overShanghai based on online water-soluble ionic composition of aerosols, Atmos. Environ., 45, 5131–5137.

Fairlie, T. D., D. J. Jacob, and R. J. Park (2007), The impact of transpacific transport of mineral dust in the United States, Atmos. Environ., 41(6),1251–1266.

Fairlie, T. D., D. J. Jacob, J. E. Dibb, B. Alexander, M. A. Avery, A. van Donkelaar, and L. Zhang (2010), Impact of mineral dust on nitrate sulfateand ozone in transpacific Asian pollution plumes, Atmos. Chem. Phys., 10, 3999–2012, doi:10.5194/acp-10-3999-2010.

Fisher, J. A., et al. (2011), Sources, distribution, and acidity of sulfate-ammonium aerosol in the Arctic in winter-spring, Atmos. Environ., 45(39),7301–7318.

Fountoukis, C., and A. Nenes (2007), ISORROPIA II: A computationally efficient thermodynamic equilibrium model for K+-Ca

2+-Mg

2+-NH4

+-

Na+-SO4

2�-NO3

�-Cl

�-H2O aerosols, Atmos. Chem. Phys., 7, 4639–4659, doi:10.5194/acp-7-4639-2007.

Han, X., M. Zhang, J. Gao, S. Wang, and F. Chai (2013), Modeling analysis of the seasonal characteristics of haze formation in Beijing, Atmos.Chem. Phys. Discuss., 13, 30,575–30,610, doi:10.5194/acpd-13-30575-2013.

Harris, E., et al. (2013), Enhanced role of transition metal ion catalysis during in-cloud oxidation of SO2, Science, 340, 727–730, doi:10.1126/science.1230911.

He, H., Y. S. Wang, Q. Ma, J. Ma, B. Chu, D. Ji, G. Tang, C. Liu, H. Zhang, and J. Hao (2014), Mineral dust and NOx promote the conversion of SO2to sulfate in heavy pollution days, Sci. Rep., 4, 4172, doi:10.1038/srep04172.

Huang, K., G. Zhuang, Q. Wang, J. S. Fu, Y. Lin, T. Liu, L. Han, and C. Deng (2014), Extreme haze pollution in Beijing during January 2013:Chemical characteristics, formation mechanism and role of fog processing, Atmos. Chem. Phys. Discuss., 14, 7517–7556, doi:10.5194/acpd-14-7517-2014.

Huang, X., Y. Song, M. Li, J. Li, Q. Huo, X. Cai, T. Zhu, M. Hu, and H. Zhang (2012), A high-resolution ammonia emission inventory in China,Global Biogeochem. Cycles, 26, GB1030, doi:10.1029/2011GB004161.

Jacob, D. J. (2000), Heterogeneous chemistry and tropospheric ozone, Atmos. Environ., 34, 2131–2159.Jiang J. K., W. Zhou, Z. Cheng, S. Wang, K. He, J. Hao (2014), Particulate matter distributions in China during a winter period with frequent

pollution episodes (January 2013), Aerosol Air Qual. Res., doi:10.4209/aaqr.2014.04.0070.Kang, H., B. Zhu, J. Su, H. Wang, Q. Zhang, and F. Wang (2013), Analysis of a long-lasting haze episode in Nanjing, China, Atmos. Res., 120–121,

78–87.Kasibhatla, P., W. L. Chameides, and J. St. John (1997), A three-dimensional global model investigation of seasonal variations in the atmo-

spheric burden of anthropogenic sulfate aerosols, J. Geophys. Res., 102(D3), 3737–3759, doi:10.1029/96JD03084.Kharol, S. K., R. V. Martin, S. Philip, S. Vogel, D. K. Henze, D. Chen, Y. Wang, Q. Zhang, and C. L. Heald (2013), Persistent sensitivity of Asian

aerosol to emissions of nitrogen oxides, Geophys. Res. Lett., 40, 1021–1026, doi:10.1002/grl.50234.Kotarba, A. Z. (2009), A comparison of MODIS-derived cloud amount with visual surface observations, Atmos. Environ., 92, 522–530.Li, W., S. Zhou, X. Wang, Z. Xu, C. Yuan, Y. Yu, Q. Zhang, and W. Wang (2011), Integrated evaluation of aerosols from regional brown hazes

over northern China in winter: Concentrations, sources, transformation, and mixing states, J. Geophys. Res., 116, D09301, doi:10.1029/2010JD015099.

Liao, H., J. H. Seinfeld, P. J. Adams, and L. J. Mickley (2004), Global radiative forcing of coupled tropospheric ozone and aerosols in a unifiedgeneral circulation model, J. Geophys. Res., 109, D16207, doi:10.1029/2003JD004456.

Liao, H., D. K. Henze, J. H. Seinfeld, S. Wu, and L. J. Mickley (2007), Biogenic secondary organic aerosol over the United States: Comparison ofclimatological simulations with observations, J. Geophys. Res., 112, D06201, doi:10.1029/2006JD007813.

Park, R. J., D. J. Jacob, B. D. Field, R. M. Yantosca, and M. Chin (2004), Natural and transboundary pollution influences on sulfate-nitrate-ammonium aerosols in the United States: Implications for policy, J. Geophys. Res., 109, D15024, doi:10.1029/2003JD004473.

Shen, Z., J. Cao, R. Arimoto, Z. Han, R. Zhang, Y. Han, S. Liu, T. Okuda, S. Nakao, and S. Tanaka (2009), Ionic composition of TSP and PM2.5 duringdust storms and air pollution episodes at Xi’an, China, Atmos. Environ., 43(18), 3911–3918.

Song, C. H., C. M. Kim, Y. J. Lee, G. R. Carmichael, B. K. Lee, and D. S. Lee (2007), An evaluation of reaction probabilities of sulfate and nitrateprecursors onto East Asian dust particles, J. Geophys. Res., 112, D18206, doi:10.1029/2006JD008092.

Streets, D. G., et al. (2003), An inventory of gaseous and primary aerosol emissions in Asia in the year 2000, J. Geophys. Res., 108(D21), 8809,doi:10.1029/2002JD003093.

Sun, Y., Z. Wang, P. Fu, Q. Jiang, T. Yang, J. Li, and X. Ge (2013b), The impact of relative humidity on aerosol composition and evolutionprocesses during wintertime in Beijing, China, Atmos. Environ., 77, 927–934.

Sun, Y., G. Zhuang, A. Tang, Y. Wang, and Z. An (2006), Chemical characteristics of PM2.5 and PM10 in haze-fog episodes in Beijing, Environ. Sci.Technol., 40(10), 3148–3155, doi:10.1021/es051533g.

Tan, J.-H., J. Duan, D. Chen, X. Wang, S. Guo, X. Bi, G. Sheng, K. He, and J. Fu (2009), Chemical characteristics of haze during summer and winterin Guangzhou, Atmos. Res., 94(2), 238–245, doi:10.1016/j.atmosres.2009.05.016.

Turšič, J., A. Berner, M. Veber, M. Bizjak, B. Podkrajsek, and I. Grgic (2003), Sulfate formation on synthetic deposits under haze conditions,Atmos. Environ., 37, 3509–3516.

Turšič, J., A. Berner, B. Podkrajsek, and I. Grgic (2004), Influence of ammonia on sulfate formation under haze conditions, Atmos. Environ., 38,2789–2795.

Usher, C. R., H. Al-Hoshey, S. Carlos-Cuellar, and V. H. Grassia (2002), A laboratory study of the heterogeneous uptake and oxidation of sulfurdioxide on mineral dust particles, J. Geophys. Res., 107(D23), 4713, doi:10.1029/2002JD002051.

Usher, C. R., A. E. Michel, and V. H. Grassian (2003), Reactions on mineral dust, Chem. Rev., 103, 4883–4939.Walker, J. M., S. Philip, R. V. Martin, and J. H. Seinfeld (2012), Simulation of nitrate, sulfate, and ammonium aerosols over the United States,

Atmos. Chem. Phys., 12, 11,213–11,227, doi:10.5194/acp-12-11213-2012.Wang, K., Y. Zhang, A. Nenes, and C. Fountoukis (2012a), Implementation of dust emission and chemistry into the Community Multiscale

Air Quality modeling system and initial application to an Asian dust storm episode, Atmos. Chem. Phys., 12, 10,209–10,237,doi:10.5194/acp-12-10209-2012.

Wang, L. T., Z. Wei, J. Yang, Y. Zhang, F. F. Zhang, J. Su, C. C. Meng, and Q. Zhang (2014a), The 2013 severe haze over the southern Hebei, China:Model evaluation, source apportionment, and policy implications, Atmos. Chem. Phys., 14, 3151–3173, doi:10.5194/acp-14-3151-2014.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,439

Wang, X., W. Wang, L. Yang, X. Gao, W. Nie, Y. Yu, P. Xu, Y. Zhou, and Z. Wang (2012b), The secondary formation of inorganic aerosols in thedroplet mode through heterogeneous aqueous reactions under haze conditions, Atmos. Environ., 63, 68–76.

Wang, X., Y. Wang, J. Hao, J. Y. Kondo, M. Irwin, J. W. Munger, and Y. Zhao (2013a), Top-down estimate of China’s black carbon emissions usingsurface observations: Sensitivity to observation representativeness and transport model error, J. Geophys. Res. Atmos., 118, 5781–5795,doi:10.1002/jgrd.50397.

Wang, Y., G. Zhuang, Y. Sun, and Z. An (2006), The variation of characteristics and formation mechanisms of aerosols in dust, haze and cleardays in Beijing, Atmos. Environ., 40, 6579–6591.

Wang, Y., Q. Q. Zhang, K. He, Q. Zhang, and L. Chai (2013b), Sulfate-nitrate-ammonium aerosols over China: Response to 2000–2015 emissionchanges of sulfur dioxide, nitrogen oxides, and ammonia, Atmos. Chem. Phys., 13, 2365–2652, doi:10.5194/acp-13-2635-2013.

Wang, Z., et al. (2014b), Modeling study of regional severe hazes over mid-eastern China in January 2013 and its implications on pollutionprevention and control, Sci. China Earth Sci., 57, 3–13, doi:10.1007/s11430-013-4793-0.

World Health Organization (2005), WHO Air quality guidelines for particulate matter, ozone, nitrogen dioxide and sulfur dioxide.Yang, K., R. R. Dickerson, S. A. Carn, C. Ge, and J. Wang (2013), First observations of SO2 from the satellite Suomi NPP OMPS: Widespread air

pollution events over China, Geophys. Res. Lett., 40, 4957–4962, doi:10.1002/grl.50952.Zhang, L., D. J. Jacob, E. M. Knipping, N. Kumar, J. W. Munger, C. C. Carouge, A. van Donkelaar, Y. X. Wang, and D. Chen (2012), Nitrogen

deposition to the United States: Distribution, sources, and processes, Atmos. Chem. Phys., 12, 4539–4554, doi:10.5194/acp-12-4539-2012.Zhang, Q., et al. (2009), Asian emissions in 2006 for the NASA INTEX-B mission, Atmos. Chem. Phys., 9, 5131–5153, doi:10.5194/acp-9-5131-2009.Zhang, Y., and G. R. Carmichael (1999), The role of mineral aerosol in tropospheric chemistry in East Asia—A model study, J. Appl. Meteorol.,

38, 353–366.Zhao, G., and L. D. Girolamo (2006), Cloud fraction errors for trade wind cumuli from EOS-Terra instruments, Geophys. Res. Lett., 33, L20802,

doi:10.1029/2006GL027088.Zhao, X. J., P. S. Zhao, J. Xu, W. Meng, W. W. Pu, F. Dong, D. He, and Q. F. Shi (2013), Analysis of a winter regional haze event and its formation

mechanism in the North China Plain, Atmos. Chem. Phys., 13, 5685–5696, doi:10.5194/acp-13-5685-2013.Zheng, G. J., et al. (2014), Exploring the severe winter haze in Beijing, Atmos. Chem. Phys. Discuss., 14, 17,907–17,942, doi:10.5194/

acpd-14-17907-2014.

Journal of Geophysical Research: Atmospheres 10.1002/2013JD021426

WANG ET AL. ©2014. American Geophysical Union. All Rights Reserved. 10,440