An Observation-Based Model for Secondary Inorganic Aerosols

An observation-based model for secondary inorganic aerosol (OBM-SIA) is developed to determine the sensitivity of formation of sulfate (SO4) and nitrate (NO3) to changes in their precursors. The model incorporates CB05 chemical mechanism with inclusion of two recently discovered OH enhancement pathways and essential aqueous phase chemical reactions, thermodynamic equilibriums for gas-aerosol phase apportionment and size distribution of SO4 and NO3. A sequence of present time-frame observations of precursors and particle compositions are used to drive the simulation and to determine responses to perturbed emission rates of precursors. OBM-SIA obviates the need for uncertain emission inventories and boundary layer dynamic conditions, and makes use of data that are increasingly available due to recent advances in online instruments for various gaseous and aerosol components, thereby offering a cost-effective tool for the analysis of SIA-precursor relationships in the atmosphere. OBM-SIA is applied to hourly gaseous and particulate composition data during a wintertime pollution episode in Hong Kong. The major pathway responsible for the production of NO3 is the reaction of OH and NO2 in the gas phase, while the gas phase oxidation of SO2 by OH and aqueous phase oxidation of S(IV) by O3 contribute most significantly on SO4 production. NO3 production is more sensitive to the reduction of NOx and anthropogenic VOCs. Reduction of SO4 formation is however most sensitive to SO2 reduction and it becomes more effective as more SO2 is reduced. The work illustrates the utility of OBM-SIA in suggesting evidence-based control strategies for effective reduction of targeted SIAs.


INTRODUCTION
Ambient particle formation and accumulation in the atmosphere involve a series of physical and chemical processes in the lower troposphere.The particles can be a result of direct emissions from anthropogenic/natural sources (primary particles).They can also be formed from reaction between certain pollutants and components of the natural atmosphere, e.g., sulfur dioxide (SO 2 ), nitrogen oxides (NO x ) with various atmospheric oxidants (e.g., OH and H 2 O 2 ).Particle production and accumulation is critically dependent upon physical parameters that characterize the boundary layer, such as temperature, relative humidity (RH), solar radiation, wind profile, and mixing height.In addition, regional transport of particles and their precursors can also affect particle levels in a particular locale.

Sulfate (SO 4
2-) and nitrate (NO 3 -) are known secondary particle components, thereafter collectively called secondary inorganic aerosol (SIA).SIA typically constitutes a major part of fine particulate matter (PM) mass in areas with high emissions of SO 2 and NO x .For reduction of SIA, formulation of effective control strategies requires a quantitative understanding on the relationships between SIA and their precursors.The relationships are typically location-dependent and time-dependent as a result of oxidants being the key reactants in the formation of SIA.Specifically, it is critical to understand the reduction potential of SIA corresponding to certain levels of reduction in a precursor, or the combination of precursors under the typical atmospheric conditions of a study region.
Emission-based air quality models are traditionally used in quantifying the particle-precursor relationship within a given area and, by extension, in developing strategies for particle abatement within the area (Cardelino and Chameides, 1995;Che et al., 2011).These models use emission inventories and numerical representations of transport and chemistry processes in the boundary layer to predict precursor and ambient particle concentrations within a given air shed.
However, because of significant uncertainties associated with emission inventories as well as numerical representations of boundary layer dynamics and transport (Zheng et al., 2009(Zheng et al., , 2010)), the application of emission-based model to the particle abatement problem is known to be associated with significant level of uncertainties.
Recognizing the significant scientific challenges inherent in emission-based models to simulate particulate-precursor concentrations, we have developed an alternative approach that is based on the ambient observations of SIAs and their precursors.The model is mechanistically complete and timedependent and it resolves chemical processes, thermodynamic equilibrium and size distribution of secondary inorganic species (i.e., SO 4 2-, NO 3 -and NH 4 + ).We refer to this model as an Observation-Based Model for Secondary Inorganic Aerosols (OBM-SIA) to differentiate with the observationbased model for ozone (O 3 ) developed by Cardelino and Chameides (1995).Because OBM-SIA uses observed precursors and particle compositions as inputs, it has the advantages of circumventing the reliance on emission inventories and not requiring the simulation of boundary layer dynamics.A third advantage is that OBM-SIA does not require major computer resources to operate and makes use of data that is increasingly available from the air quality monitoring sites.
In this study, we first describe the structure of OBM-SIA, then evaluate its performance and ability to characterize secondary formation and accumulation processes by comparison with field measurements of atmospheric radicals in the Pearl River Delta (PRD).Next, we present results of OBM-SIA analysis of a wintertime pollution episode in Hong Kong as an example of model's application.The results help us to gain insight into important physical and chemical processes for SO 4 2-and NO 3 -formation in Hong Kong and to propose effective PM 2.5 control strategy for Hong Kong.

STRUCTURE AND DEVELOPMENT PRINCIPLE OF OBM-SIA
OBM-SIA shares the general framework with the Observation-Based Model for O 3 (Cardelino and Chameides, 1995), with significant expansions made to describe the much more complicated atmospheric processes of particles.OBM-SIA consists of three modules, a mass conservation module, a particle formation and phase conversion module, and a source estimation and perturbation module.(1) The mass conservation module specifies the temporal variation of atmospheric chemical species, based on mass conservation for each species.(2) The particle formation and phase conversion module estimates production potential of total SO 4 2-and NO 3 -(the sum of gas-and particulate-phase H 2 SO 4 and HNO 3 ) and their partitions among gases, fine aerosols (PM 2.5 ) and coarse aerosols (PM >2.5 ).This module also allows the investigation of contributions of different processes to the formation of SO 4 2-and NO 3 -in PM 2.5 .
(3) The source estimation and perturbation module estimates transport of secondary particles and performs sensitivity studies in evaluating effectiveness of different strategies on reducing SO 4 2-and NO 3 -in PM 2.5 .Fig. 1 depicts the relationship between the three modules and the data flow in OBM-SIA.We next describe the chemical mechanism and deposition treatment incorporated in the model, followed by the details of the three modules.

Chemical Mechanism
Chemical Mechanism incorporated in OBM-SIA is the CB05 mechanism plus two recently discovered OH enhancement pathways (Li et al., 2008;Hofzumahaus et al., 2009).CB05 is an upgraded version of CBIV chemical -and the contributions by different pathways.In the sensitivity run, relative incremental reactivities (RIR) of SO 4 2-and NO 3 -are calculated to assess the impact of reduction in certain precursors on the abatement of ambient secondary inorganic aerosol levels.*Meteorological Parameters.mechanism, with updated rate constants and additional lumped and explicit model species (USEPA, 2006).The CB05 mechanism involves 57 species in the gas phase, 8 in the particle phase and 12 in the aqueous phase (Table 1).Initial concentration of the species is given in Table C2 of the supplementary materials by running OBM-SIA over 24 h during normal conditions in HK.
OH radicals are primarily formed through the photolysis of O 3 and HONO, and govern the oxidation of various atmospheric trace gases.A recent study showed that the known OH sources are not enough to sustain the observed reaction rate of OH (Hofzumahaus et al., 2009).Taking into recent advances in understanding of OH production, we include in our mechanism additional OH production pathways: the amplified trace gas removal mechanism (R1-R2) (Hofzumahaus et al., 2009).
In R1 and R2, U is an unknown species acting as a reactive equivalent of 0.85 ppb of NO.
Laboratory experiments and field studies have demonstrated the importance of the heterogeneous kinetics of gas-phase species in interacting with aerosol surface (Ravishankara, 1997;Wagner et al., 2008;Pradhan et al., 2010).In OBM-SIA, heterogeneous reaction of N 2 O 5 hydrolyzed onto the surface of aerosols (R3), calculated using the method of Lin and Cheng (2007), is included in the gas phase reaction mechanism.This reaction is important in terms of producing HNO 3 at night (Hu and Abbatt, 1997;Seinfeld and Pandis, 2006, p. 180).Heterogeneous production of SO 4 2-is not considered due to the acidic nature of the ambient aerosols, the absence of sea-salt, dust and HOCl/HOBr particles in the modeling regions, and large uncertainties associated with this pathway.
The reaction of H 2 O 2 interacting with dust aerosol The non-traditional chemical formulas are adopted to be the same as those used in CB05 mechanism.They are either abbreviations for single VOCs, the sum of a few VOCs, or the sum of certain functional groups.The corresponding relationships with real VOCs are given in Table C1.b Base denotes base run.Sens denotes sensitivity run.c PNA: Peroxynitric acid (HNO 4 ).
particles is also included in OBM-SIA.This reaction might be an important sink of H 2 O 2 that 'has been, until recently, poorly constrained' (Pradhan et al., 2010).However, to the best of our knowledge, probability of this reaction on the urban aerosols has not been reported.As dust particles generally account for 1/10 of urban PM 10 in the PRD, we therefore approximate uptake coefficients (γ H 2 O 2 ) to be 1/10 of the values for Gobi aerosols at 70% RH (6.03 × 10 -4 ).
Photochemical reactions are essential for the production of O 3 and secondary particulate in the atmosphere.In the OBM for O 3 , the photolytic rate constants are calculated as a function of zenith angles determined by geographical location and time of year.Apart from solar zenith angle, the photolytic rate could be also influenced by the pollution level, e.g., the amount and type of aerosol particles and absorbing gases present in the air, the air molecules and the surface albedo.For example, Koepke et al. (2010) estimated NO 2 photolysis frequencies within different types of urban streets and showed that the photolysis frequencies of NO 2 reduced to less than 20% for narrow skyscraper streets, to about 40% for typical urban streets and to about 80% for garden streets, in comparison with the values for open horizon.
In OBM-SIA, atmospheric NO 2 photolysis frequency, J(NO 2 ), is firstly calculated using the equation proposed by Trebs et al. (2009), where α represents site-dependent UV-A surface albedo (set as 0.05 in this study); G is the measured solar radiation rate (W/m 2 ); constants B 1 and B 2 are 1.47 × 10 -5 W -1 m 2 /s and -4.84 × 10 -9 W -2 m 4 /s, respectively.A 5-order polynomial regression equation, based on zenith angles as a function of clear sky J(NO 2 ), is then obtained to estimate the substituted hourly zenith angles.Rate constants for other photolytic reactions are then calculated from the derived substituted hourly zenith angles.

Deposition
Deposition is the ultimate pathway to remove trace gases and particles from the atmosphere (Seinfeld and Pandis, 2006, p. 900-926).In current OBM-SIA version, only simplified dry deposition process is considered for several important species listed in Table 2. Similar simplications were adopted in Vayenas et al. (2005) and Huang et al. (2011).Dry deposition of gaseous H 2 SO 4 is not included considering its extremely low partial pressure in the atmosphere.Wet deposition process is neglected, because no precipitation was observed during the sampling period.

Mass Conservation Module
Pre-treatment of the Observation Data OBM-SIA uses the hourly concentration of the observed atmospheric species as well as the meteorological condition data to calculate the formation potential of secondary SO 4 2-and NO 3 -.A cubic spline is employed to interpolate these hourly data (and, by extension, any missing data points that may be present in a given dataset) into concentrations with a temporal resolution in a scale of characteristic time of photochemical reactions, e.g., a few thousandths of a second.Instead of using the actual concentrations, observed VOC species are transformed into their corresponding functional groups in the CB05 mechanism using the recipe of Yarwood et al. (2005).
Calculation of the Unspecified Species (Step I in Fig. 1) The species in the chemical mechanism are separated into specified and unspecified species.Specified species refer to those whose concentrations are directly measured and used as inputs into OBM-SIA, while unspecified species refer to those whose concentrations are calculated by the model, e.g., radical species such as OH and HO 2 .Table 1 lists the specified and unspecified species in this study.
Calculation of gas-phase unspecified species is done by integrating a coupled set of differential equations with the following form: where C A is the concentration of species A; P A and L A are chemical production and destruction rates of A, respectively; the third term on the right side represents dry deposition rate of A, where v A is the dry deposition velocity and H(t) is the mixing height at time t; the fourth term on the right represents the loss by dilution as the mixing height expands, where C A FT is the free tropospheric concentration of species A (assumed to be zero for all unspecified species); and the far-right term represents the loss of A due to gas-aqueous phase transfer.Solution of the time-dependent differential equations is accomplished in the model using a standard Gear routine.Here, we want to clarify that in this paper, "formation" represents the net effect of two competing processes, i.e., "production" for positive contributions and "removal" for negative losses.

Aqueous Phase Reactions
Significant formation of SO 4 2-is known to derive from in-cloud (fog) processing followed by cloud (fog) evaporation in different environments (Seinfeld and Pandis, 2006, p. 308-334).In OBM-SIA, simplified aqueous phase chemical mechanism is included to account for the aqueous phase production of SO 4 2-, as shown in Table 3. Cloud coverage, expressed in eighths of sky covered by cloud, is obtained from observations at the Hong Kong International Airport.Cloud processing is assumed to occur in the upper half of the mixing layer, and aqueous phase reaction occurs in clouds with water-to-air volume ratio of 5.0 × 10 -7 (Warneck, 2003).Aqueous phase pH is an important parameter that governs the solubility of gases and activity of aqueous phase reactions (especially for production of SO 4 2-).A previous local study showed pH of the cloud water in Hong Kong varied between 3.6 and 4.6 (Tanner and Tam, 2006).Cloud water pH is assumed to be 4 in this study and the sensitivity of modeling results in response to varying cloud pH is examined in Appendix B of the supplementary materials.
In the presence of cloud (fog) droplets, transfer of chemical species between gas and aqueous phases occurs.The transfer of a species can be described by the following equations (Schwartz, 1986;Warneck, 1999;Huang et al., 2011). (3) where k t (s -1 ) is the transfer coefficient; c g (mol/dm 3 ) and c a (mol/L) are gas-and aqueous-phase concentrations of a species, respectively; L is the total liquid water content (vol/vol); R g (L/atm/mol) is the gas constant and H is the Henry's law constant.In OBM-SIA, a series of coupled ordinary differential equations is used to treat the processes explicitly, with the values of diffusion coefficients, mass accommodation coefficients, Henry's law constants and several other parameters from Warneck (1999).Meantime, concentrations of Fe (III) and Mn (II) in the cloud droplets are assumed to be typical values of 0.3 µM and 0.03 µM, respectively.
In current OBM-SIA version, only simplified aqueous phase reactions known important to formation of SO 4 2- (Seinfeld and Pandis, 2006) is included as shown in Table 3.Further efforts would be laid on improve comprehensive aqueous chemical mechanisms in OBM-SIA.5).As such, formation rate of H 2 SO 4 , calculated here, should be considered as the potential formation rate of H 2 SO 4 , rather than an actual value.

Particle Formation and Phase Conversion Module
Formation rate of HNO 3 (g) is calculated using Eq. ( 6): Definition of each term in Eq. ( 6) is similar to its corresponding term in Eq. ( 5).The new term, C HNO 3 V HNO 3 /H(t), represents the dry deposition rate of HNO 3 (g).This term does not occur in Eq. ( 5) because of the extremely low mixing ratio of H 2 SO 4 (g) in the ambient conditions.Gaseous HNO 3 has high propensity to stick to most surfaces due to zero surface resistance.The deposition velocity of HNO 3 (g) is therefore prominent.A number of studies have concluded that dry deposition of HNO 3 (g) contributes significantly to removal of HNO 3 (g), sometimes even greater than the wet deposition (Meyers et al., 1991;Janson and Grannat, 1999).Dry deposition velocity of HNO 3 (g) have been reported in the range of 1-13 cm/s under various conditions.Typical values of 4 cm/s on land and 1 cm/s over ocean (Seinfeld and Pandis, 2006, p. 904) are adopted in OBM-SIA.

Phase Partition of Total SO 4 2-and NO 3 -(
Step III in Fig. 1) The calculated total SO 4 2-and NO 3 -are partitioned into gas phase and particle phase, the latter consisting of particles of fine and coarse sizes.H 2 SO 4 , once formed, is totally partitioned into the fine particles.For HNO 3 , the partition into coarse particles is first determined based on the measured size distribution of NO 3 -(based on 8 h MOUDI observations).A thermodynamic model is then applied to partition the remaining HNO 3 into gas and fine particles.Considering the fast running speed and easy accessibility, ISORROPIA (Nenes et al., 1998) is incorporated into OBM-SIA for simulation of phase conversion.One implicit assumption of ISORROPIA is that thermodynamic equilibrium is reached between gas and particle phase compositions.This assumption is validated by examining the gas and particle pair of data collected at Panyu, a residential town in the center of the PRD during the same sampling period (see Appendix C in the supplementary materials for details).

Source Estimation and Perturbation Module
Source Term Estimation (Step IV in Fig. 1) In this module, time-dependent source functions of specified species, e.g., Σ SO 4 2-(t) and Σ NO 3 -(t), are determined.As a 0-d model, OBM-SIA resolves source functions as the combined result of local emissions and transport into and out of the sampling site.Particle SO 4 2-and NO 3 -in the PRD are predominately secondary in nature, their primary emissions can therefore be neglected.In view of this, "transport function" is used to replace "source function" hereafter, and transport function essentially represents regional chemistry processes (chemistry production or loss outside the sampling point).Contributions from transport function can be approximated by Eq. ( 7) and Eq. ( 8).
where (∂C SO 4 2-/∂t] obs and [∂C NO 3 -/∂t)] obs represent the observed concentrations of SO 4 2-and NO 3 -, respectively, as a function of time, after smoothing with a cubic spline; and the remaining term represents local production of SO 4 2-and NO 3 -in fine particles calculated from step III.
Source Perturbation (Steps V-X in Fig. 1) By integrating the calculated formation rates and transport function over the period of interest, another set of calculation is carried out to rebuild the ambient SO 4 2-and NO 3 -in scenarios that one type of precursor is controlled, as expressed in Eq. ( 9) and Eq. ( 10).FT is the free tropospheric concentration of species A (assumed here to be zero).Similar definitions are adopted in the equation for NO 3 -(i.e., Eq. ( 10)).

 
Sensitivity studies are conducted by examining the incremental reactivity (IR) and relative IR (RIR) of each control run and comparing ambient concentrations of SO 4 2-and NO 3 -in fine particles in scenarios before and after control strategies are carried out.The IR x , determined by Eq. ( 11), gives an absolute estimate of the changes of SO 4 2-and NO 3 -molecules in fine particles in response to the reduction of one molecule of precursor X.The RIR X , calculated using Eq. ( 12), gives an estimate of the effectiveness of reducing the concentration of precursor X over that of formation of SO 4 2-and NO 3 -in fine particles.
where F(X) represents the formation of SO 4 2-or NO 3 -in fine particles corresponding to precursor X at concentration S(X); F(X -X) represents the formation of SO 4 2-or NO 3 corresponding to precursor X at concentration S(X) -S(X).S(X) is the ambient concentration of X at the time of interest.Control runs with the highest RIR value and the largest difference in SO 4 2-and NO 3 -levels are deemed the most effective way in reducing SO 4 2-and NO 3 -.

MODEL EVALUATION
In an observation-based model such as OBM-SIA, almost all significant species in gaseous and particle phases are directly measured and used as inputs to drive the model, traditional model evaluation by comparing model prediction with observations is therefore not applicable.Radical species are typically not measured due to their low concentrations and the associated significant measurement challenge.They are consequently "unspecified species" in the OBM-SIA.Comparison of the simulated radical species with measurements would therefore be considered a meaningful way in evaluating the performance of OBM-SIA.It was fortunate that OH and HO 2 radicals were measured at a site in the PRD region during an intensive field campaign in the summer of 2006 (Hofzumahaus et al., 2009).We took advantage of this rare data opportunity to evaluate the OBM-SIA model performance.
In this OBM-SIA evaluation exercise, the input data are extracted from a series of peer-reviewed manuscripts (Chan et al., 2006;Garland et al., 2008;Hua et al., 2008;Hofzumahaus et al., 2009) or typical values in the PRD, e.g., mixing height of 1000 m, are used.VOC concentrations (except benzene and isoprene) during the sampling campaign are not reported.We therefore use VOC profiles measured at industrial suburban locations in the PRD during the same season (Chan et al., 2006).Temporal variations of anthropogenic and biogenic VOCs are taken to be the same as the temporal variations of benzene and isoprene, respectively.In this calculation, substituted zenith angle is calculated based on the photolysis rate of NO 2 (Hofzumahaus et al., 2009).We apply one day as the model spin-up time.
Fig. 2 compares the observed and the model-calculated diurnal patterns of OH and HO 2 radicals.Two modelpredicted profiles are included; one is generated using CB05 mechanism without the two OH enhancement pathways and the second using CB05 plus the two OH enhancement pathways (i.e., the default OBM-SIA chemical mechanism).The OH radicals predicted by the default OBM-SIA chemical mechanism capture the observed diurnal pattern and agree with the observed OH profile within the measurement uncertainty.In comparison, the modeled OH profile without the inclusion of the two OH enhancement pathways is mostly outside the lower bound of the measurement uncertainty and the predicted maximum OH concentration is 50% lower than that predict by the default OBM-SIA mechanism.The results demonstrate the capability of OBM-SIA in depicting ambient OH radical levels and their variations.The discrepancy between the prediction and the observation may result from (1) substituted rather than the actual VOCs profile being used; (2) the current mechanism not comprehensive enough to portray OH characteristics; and (3) large OH measurement uncertainty.For the simulations of HO 2 radicals, model predictions obtained using CB05 mechanism and the default OBM-SIA chemical mechanism are comparable and both fall within the measurement uncertainty range.

APPLICATION OF OBM-SIA
Though OBM-SIA is applied on a ten-day long field measurement campaign in winter in HK, in this section, we only present detailed OBM-SIA analysis of SIA formation on two days, a normal day and an episode day.The purpose is to illustrate the applicability and usefulness of OBM-SIA under two different atmospheric conditions encountered at our study location.More extensive OBM-SIA analysis covering a wider range of atmospheric conditions will be examined in a future paper.

Observation Data and Other Input Parameters
A measurement campaign was conducted in Tung Chung, a mixed residential/commercial district in the western part of Hong Kong and 2 km southeast of the Hong Kong International Airport, during 17-26 December 2009.Hourly data of criteria gaseous air pollutants (O 3 , CO, SO 2 , and NO x ) and PM 2.5 mass concentrations are made available through the air quality monitoring program by the Hong Kong Environmental Protection Department (HKEPD) (HKEPD, 2009a).Hourly VOC data were obtained with an on-line Gas Chromatography (HKEPD, 2009b).PM 2.5 ionic chemical composition (SO 4 2-, NO 3 -, NH 4 + , K + , and Na + ) at an hourly resolution was measured using a semi-continuous Particle-Into-Liquid System (PILS) (Weber et al., 2001;Xue et al., 2011).Size distributions of the SIA components averaged over 8-hour were obtained with a 10-stage Micro-Orifice Uniform-Deposit Impactor (MOUDI) followed by ion chromatography (IC) analysis on water extracts of the impactor substrates (Huang et al., 2006).
A high PM episode occurred on 25 December, with the PM 2.5 mass reaching 147.7 µg/m 3 during 16:00-17:00 (Local time, the same as below).In comparison, the average PM 2.5 mass was 46.4 µg/m 3 during the same hours on the other sampling days.Around 75% of the elevated PM 2.  A summary of the input data and their measurement methods is given in Table 4. Notably, for certain species not measured at Tung Chung, we used the observations made in locations of approximate site characteristics in Hong Kong or other locations in the PRD.Among them, HNO 3 (g), HONO(g), and NH 3 (g) are measurements made in Panyu, a suburban district 25 km away from Guangzhou urban center.Carbonyl data are measurements in Tsuen Wan (an urban residential/commercial district), Hong Kong in the same month (Louie et al., 2013).Sensitivities of modeling results on these approximations are examined in Appendix B of the supplementary materials.

Formation of NO 3 -and SO 4 2-on a Normal Day and the PM Episode Day
The various pathways contributing to NO 3 -and SO 4 2formation are tabulated in Table 5. Fig. 4 shows in bar graphs the contributions of individual pathways to the total formation rate of NO 3 -during daytime (7:00 to 19:00) on 21 and 25 December.The upper part of each bar graph presents the production rates of NO 3 -by different pathways; the middle part presents the removal rates of NO 3 -by different pathways; the lower part shows the net contributions from local formation and transport function.Fig. 5 shows the same information for SO 4 2-.Discussion on nighttime results is presented in supplementary materials (Appendix D of the supplementary materials).

Simulation of Daytime NO 3 -
The total production rate of NO 3 -during daytime of 25 December reaches 2.73µg/m 3 /h, significantly higher than that during daytime of 21 December (0.67 µg/m 3 /h) (Figs. 4(a) and 4(b)).The major pathway responsible for the production of NO 3 -is the reaction of OH and NO 2 in the gas phase (P N 2).This reaction contributes 85% and 68% of the total production rate of NO 3 -during daytime of 21 and 25 December, respectively.The absolute production rate is 1.86 µg/m 3 /h on 25 December, over 3 times the value on 21 December.This is largely due to the elevated levels of reactants, NO 2 (43.5 ppb on 21 December vs. 68.6 ppb on 25 December) and OH (2.48 × 10 5 molecule/cm 3 on 21 December vs. 4.00 × 10 5 molecule/cm 3 on 25 December).
The enhanced NO 2 level may be associated with increased emissions from vehicles during the Christmas holiday going to and from the nearby airport while the enhancement of OH is attributed to the faster cycling of HO 2 and OH through chain reactions (R1-R2, R4-R10) in the presence of higher levels of VOCs (as organic peroxyl radical, RO 2 ), CO, NO x and U (Lee et al., 1998(Lee et al., , 2000)).CREs NO HNO CRO    The second important pathway during daytime, as indicated in Figs.4(a) and 4(b), is the reaction of N 2 O 5 with H 2 O (P N 1) in the gas phase, on aerosol surface and/or in the cloud water.Compared with P N 2, the contribution of this process at 0.09 µg/m 3 /h is much less important during normal day, but increases rapidly to 0.71µg/m 3 /h during the episode day.Although NO 3 as one of the precursors of N 2 O 5 is readily photolyzed by daylight (R12-13), recent studies highlighted the importance of P N 1 to daytime atmospheric chemistry.Geyer et al. (2003) observed daytime NO 3 concentration of up to 5 pptv by long-path differential optical absorption spectroscopy (DOAS) three hours before sunset in a polluted urban area near Houston, Texas.They suggested that NO 3 chemistry could play an important role in producing NO 3 -in polluted megacities.Brown et al. (2005) observed daytime N 2 O 5 mixing ratio of up to 20 pptv from the NOAA P-3 aircraft during the New England Air Quality study in the summer of 2004.They speculated that hydrolysis of N 2 O 5 could increase the daytime conversion  rate of NO x to HNO 3 by a maximum of 13% with respect to OH + NO 2 .It is further argued that in winter, when the thermal equilibrium between NO 3 and N 2 O 5 shifts strongly in favor of the latter, effects of P N 1 during the daytime could even dominate over NO 2 oxidation by OH.With the significant accumulation of NO 2 during the pollution episode day, daytime N 2 O 5 mixing ratio reaches 17.5 pptv (by R14-R15).As a result, a notable increase of production rate of P N 1 is expected.
In comparison, pathways of P N 3, 5, 6, 7, 8, 9 are relatively inefficient, with the combined contribution less than 7% of the total production rate.
The total removal rate of NO 3 -is 0.09 and 0.54 µg/m 3 /h during daytime of 21 and 25 December (Figs.4(a) and 4(b)), respectively.Majority of the removal is attributed to dry deposition of HNO 3 (g) (R N 1) and NO 3 -(R N 2).The much higher removal rate of R N 1 highlights the importance of dry deposition of HNO 3 (g), which is further linked to the removal of aerosol NO 3 -due to the thermodynamic equilibrium between HNO 3 (g) and NO 3 -.Reactions of HNO 3 with OH radicals (R N 3) and photolysis of HNO 3 (R N 4) are insignificant removal pathways, contributing less than 2% of the total removal rate.
The lower parts of plots in Fig. 4 present the net formation rate of ambient NO 3 -contributed by local chemistry and transport function.The local formation rate during daytime of 25 December is 2.19 µg/m 3 /h, around four times that of 21 December (0.58 µg/m 3 /h).The combined production rate exceeds the total loss rate on both the normal and the episodic days, indicating local accumulation of NO 3 -in the atmosphere.The excess of NO 3 -production is more prominent on the episodic days, suggesting the importance of local chemistry in contributing to the pollution episode.Negative transport functions on both days indicate that Tung Chung is a source region of NO 3 -.

Simulation of Daytime SO 4 2-
The contributions of various pathways to the total formation rate of SO 4 2-during daytime are shown in Fig. 5(a) for 21 December and in Fig. 5(b) for 25 December.Similar to NO 3 -, the total production rate of SO 4 2-during daytime of 25 December (0.24 µg/m 3 /h) is significantly larger than that on 21 December (0.06 µg/m 3 /h), indicating the elevated local production during the pollution episode.The only gas-phase production pathway, i.e., oxidation of SO 2 by OH (Ps1), contributes 32% and 28% of the total production rate during daytime of 21 and 25 December, respectively.The remaining SO 4 2-is formed through aqueous phase reactions.
Among the aqueous-phase production pathways, the oxidation of S (IV) by O 3 (Ps2) and O 2 with catalysis of iron (Ps4) make the most significant contributions.In comparison, aqueous-phase oxidation by H 2 O 2 (Ps3) contributes only 7% and 5% of the total SO 4 2-production on 21 and 25 December, respectively.Such small contributions are initially unexpected as a number of studies have shown S (IV) oxidation by H 2 O 2 in clouds to be a significant SO 4 2-formation pathway (Heikes et al., 1996;Stein andLamb, 2002, 2003).But a consideration of the high NO x conditions encountered in our study could explain the low contribution of the H 2 O 2 oxidation pathway.Under high NO x conditions, the reaction of NO 2 and OH (R11) dominates the removal of HO x , suppressing the HO 2 radical combination reaction (R16) to form H 2 O 2 (Kleinman, 1994;Stein andLamb, 2002, 2003).
It is noted that the aqueous oxidation of S (IV) by NO 2 (Ps6) accounts for around 10% of the total SO 4 2-production during the pollution episode day.Contribution of Ps6 is typically considered negligible in the past studies because of limited water solubility of NO 2 and considerably lower NO 2 seen in other locations (Venkataraman et al., 2001).In addition, this pathway is strongly dependent on aqueousphase pH, with increasing rates at higher pH.Pandis and Seinfeld (1989) indicate that this pathway might be important when the aqueous phase (cloud and/or fog) is neutralized.Our results suggest the possible significance of Ps6 during pollution episodes characteristic of high levels of NO 2 and SO 2 .We expect that this reaction is potentially important in other mega cities of China where strong primary emissions of NO 2 and SO 2 are common (Tan et al., 2009;Cheng et al., 2011).
The only removal process of SO 4 2-considered in OBM-SIA is dry deposition.The total removal rate of SO 4 2-is smaller than that of NO 3 -on both normal and episodic days, indicating that SO 4 2-could travel longer distance than NO 3 -.It is also discovered that the total production rate of SO 4 2-is at least one order of magnitude lower than the transport function during the pollution episode, which indicates most of the observed SO 4 2-is from non-local sources.

Contributions of Local Chemistry and Transport Function during Pollution Episode Day
Fig. 6 shows the diurnal patterns of NO 3 -and SO 4 2-on 25 December and concentration contributions by local chemistry and transport function.Contributions from local chemistry are calculated by integrating formation of NO 3 or SO 4 2-over the given time.Contributions from the transport function, on the other hand, are calculated as the differences between the observations and contributions from production due to local chemistry.We note that OBM-SIA as a 0-d model cannot describe the chemistry outside the sampling site.Actually, the separation of local chemistry and transport can be regarded as the separation of local and regional chemistry as there is no primary emission source for SO 4 2-and NO 3 -.The 24-hour on 25 December could be divided into four periods of distinct NO 3 -and SO 4 2characteristics.Period I, from 0:00 to 8:00, is characterized by low and relatively stable ambient concentration of NO 3 -and SO 4 2- . Local chemistry contributes -0.39 µg/m 3 of NO 3 -and -0.01 µg/m 3 of SO 4 2-.Period II, between 8:00 and 14:00, is signified of increasing NO 3 -and SO 4 2- , with NO 3 -peaking around noon time while SO 4 2-keeping increasing.In period II, ambient NO 3 -increases by 14.59 µg/m 3 , which is entirely contributed by the local production as a result of increasing solar actinic flux and gas precursors.Negative contribution by the transport function (-3.85 µg/m 3 ) suggests that NO 3 -is transported from Tung Chung to its adjacent regions.The increase magnitude in SO 4 2-is almost the same as that in NO 3 -, but the majority of the increase is by transport function.Period III is from 14:00 to 17:00, during which both NO 3 and SO 4 2-increase rapidly and reach their highest values in the day, with 39% of NO 3 -increase contributed by local production.In comparison with Period II, formation of NO 3 slows down as a result of decreasing solar actinic flux.This is however offset by higher NO 2 concentrations, leading to a local NO 3 -production rate of 3.06 µg/m 3 /h, almost the same as that in Period II.As for SO 4 2-, only 5% of the concentration increase is attributable to local chemistry.Period IV is from 17:00 to 24:00.During this period, NO 3 and SO 4 2-are rapidly transported out of Tung Chung.If Periods II and III are considered as the period responsible for the pollution episode, local production contribute 78% of NO 3 -increase and 5% of SO 4 2-increase, while transport function accounts for the remaining.This conclusion is broadly consistent with previous understanding that regional transport plays a crucial role for SO 4 2-levels in Hong Kong especially during winter when the prevailing winds are from the north/northeast, while NO 3 -is more local in origin (Louie et al., 2005;Ho et al., 2006;Yuan et al., 2006).

Response of NO 3 -and SO 4 2-to the Reductions in Precursors
Sensitivity study is carried out to determine the effectiveness of reducing certain precursors in abating ambient SO 4 2-and NO 3 -levels.In this respect, RIR values and the maximum levels of SO 4 2-and NO 3 -are used as indicators, after control strategies are conducted.Recall that the RIR X function represents the changes of SO 4 2-and NO 3 -in response to one percent reduction in precursor X concentration.Thus, the larger the RIR X value for a given precursor X is, the more effective the reduction of this precursor X in leading to reduction of SO 4 2-and NO 3 formations.
Due to the nonlinear nature of the SO 4 2-and NO 3 chemistry, RIR functions are dependent on the magnitude of the precursors change.A series of calculation is therefore needed to investigate RIR variations at different precursor reduction levels, as shown in Fig. 7.For reduction of NO x , it is assumed that NO and NO 2 share the same change rate when NO x control is conducted.
The sensitivity study indicates that NO 3 -formation during daytime of 25 December is more sensitive to the reduction of NO x and anthropogenic VOC (AVOC) (Fig. 7(a)).At the ambient AVOC level, the RIR AVOC is approximately 0.66.This means that each percent control of AVOC is modeled to result in 0.66% reduction in NO 3 -formation.At the ambient NO x level, the RIR NOx value is calculated to be 0.53, suggesting that reduction of NO x is less effective than reduction of AVOC.The calculation indicates that the significance of NO x and AVOC control would decrease as more precursors have been reduced.The efficiencies of NO x and AVOC control would drop to half when 60% of the precursors have been reduced.In comparison, the values of RIR SO2 and RIR BVOC are much smaller, indicating control of SO 2 and reduction in biogenic VOC (BVOC) would have little effect on NO 3 -reduction. 2-, and (c) SIA production in response to one percent reduction in precursor X, i.e., RIR X .RIR X is calculated for reduction in NO x , SO 2 , AVOC, and BVOC.radical that would have been lost through the reaction of R11 in response to decreasing NO x .The extra OH radicals react with SO 2 to form H 2 SO 4 in the gas phase (Ps1) or react with CO and VOCs to increase the abundance of HO 2 and subsequently favoring the formation of H 2 O 2 (R20) (Stein and Lamb, 2002).Both processes increase the production of SO 4 2-.Since production of H 2 O 2 is proportional to square of the HO 2 concentration (R20), the enhancements of SO 4 2-formation by Ps3 is more significant compared with that by Ps1.When NO x is reduced by over 60%, Ps3 becomes the dominated pathway for the formation of SO 4 2-.This confirms the importance of this pathway in a relatively low-NO x environment.Meantime, contribution from Ps2 (S (IV) + O 3 ) increases slightly as NO x concentration is lowered.This is a result of O 3 formation in a VOC-limited regime.Decreasing the availability of NO x would favor O 3 formation.
The RIR tendency for the sum of SO 4 2-and NO 3 -are similar to those of NO 3 -, as shown in Fig. 7(c).We therefore can conclude that total mass of SIA is more sensitive to the control of NO x and AVOC.This is reasonable, as the majority of NO 3 -is contributed by local production, while SO 4 2-is mostly contributed by transport function.Fig. 9 presents the predicted maximum hourly concentrations of SO 4 2-, NO 3 -and SIA under different control scenarios.Similarly, it can be seen that the most efficient strategy for controlling ambient SO 4 2-is to reduce SO 2 and for NO 3 -is to reduce AVOC.

Sensitivity of Uncertain Parameters
A series of sensitivity tests are conducted to investigate the sensitivity of the OBM-SIA modeling results to a number of parameters that are used in this case study.Detailed description of the results is presented in Appendix B in the supplementary materials.Briefly, the OBM-SIA predictions of SO 4 2-are very sensitive to cloud droplet pH, moderately sensitive to changes in formaldehyde and HONO(g) concentrations, and insensitive to other examined parameters.The predictions of NO 3 -are moderately sensitive to the dry deposition velocity of HNO 3 (g) and the levels of formaldehyde and HONO(g).The sensitivity study results highlight the importance of accurate quantification of atmospheric aqueous pH and HNO 3 (g) deposition rate.In situ measurement of HONO and formaldehyde concentrations are also recommended in future field experiments.In order to improve model's performance, onsite observations of these species are needed in future studies.

Major Assumptions of OBM-SIA and Limitations
The readers are advised that OBM-SIA necessarily adopts  a series of simplifying assumptions, including (1) transport is assumed negligible when resolving concentrations of the unspecified species by Eq. ( 2); (2) transport function is assumed unvaried when the control strategies is conducted; and (3) thermodynamic equilibrium is assumed constantly achieved in the sensitivity studies; (4) since OBM-SIA is a 0-D model, it uses surface observations to the entire PBL.As a result, OBM-SIA is more applicable to well mixed atmospheric conditions, e.g., haze (Meng et al., 2008) and fog events.In addition, as OBM-SIA does not directly predict pollutant concentrations, traditional model evaluation based on comparison between model prediction and in-situ observation is not applicable.Hence, OBM-SIA should be considered a diagnostic model that provides a useful and independent evaluation on the results obtained by emissionbased models (EBM).OBM-SIA cannot replace the EBM in prognostic modeling to calculate the pollutants level or to estimate precursor reduction needed to bring an area into attainment (Cardelino and Chameides, 1995).

SUMMARY
An Observation-Based Model for Secondary Inorganic Aerosols (OBM-SIA) has been developed to evaluate the formation of SIA and response of SIA to precursor reductions.The model resolves chemistry processes, thermodynamic equilibriums and size distribution (coarse mode and fine mode) of fine particle SO 4 2-and NO 3 -.OBM-SIA uses a sequence of present time-frame observations of precursors and particle compositions and meteorological data to drive the simulation, ensuring that the calculations are carried out for the proper combinations of NO x and VOCs conditions.This modeling approach has the unique advantage of obviating the need for uncertain emission inventories and boundary layer dynamic conditions.
Utility of the model is demonstrated by carrying out an OBM-SIA analysis of observation data of gaseous precursors, SIA components, and relevant meteorological parameters collected in a residential/commercial neighborhood site in Hong Kong during a wintertime period (17-26 December, 2009).The model analysis indicates that the major pathway responsible for the production of NO 3 -is the reaction of OH and NO 2 in the gas phase, while the oxidation of SO 2 by OH in the gas phase and S (IV) by O 3 in the aqueous phase make the most significant contributions to SO 4 2-production.OBM-SIA analysis also allows a comparison of contributions of various SIA formation pathways during a normal day and a high PM episodic day encountered during the observation period.The results suggest that local chemistry contributed more significantly to ambient NO 3 -and SO 4 2-levels on the episode day than normal days.Local production contributes to 78.4% of NO 3 -elevation and 5.0% of SO 4 2-elevation in the pollution episode hours.The remaining is contributed by regional transport.In another words, regional transport is responsible for SO 4 2-in Hong Kong while NO 3 -is more local in origin during the winter period.A series of sensitivity studies is carried out using OBM-SIA to compare the effectiveness of different precursor control options in reducing ambient SO 4 2-and NO 3 -levels for atmospheric conditions encountered during the PM episode.The simulation results show that formation of NO 3 -is more sensitive to the reduction of NO x and AVOC concentrations.For each percent reduction in NO x and AVOC, 0.53% and 0.66% reduction in mass of NO 3 -formation would be expected, respectively.Reduction of SO 4 2-formation, on the other hand, is most sensitive to reduction of SO 2 and it would become more effective as more SO 2 has been reduced.Nonlinear relationship between NO x control and SO 4 2reduction is predicted, mostly due to the complex chemistry of NO x and atmospheric oxidants, e.g., HO x radicals.

ACKNOWLEDGMENTS
This work was supported by the Research Grant Council of Hong Kong (615406), Joint Funding of the National Science Foundation of China -Guangdong Province on Key Projects (U1033001), and the Fok Ying Tung Graduate School (NRC06/07.SC01).We thank Dr. Peter K.K. Louie of Hong Kong Environmental Protection Department for providing the VOC speciation data at Tung Chung and Prof. Dui Wu of Institute of Tropical and Marine Meteorology, China Meteorological Administration for providing the gas and particle data at Panyu.documented in the field of studying NOx removal using scrubbing technologies in power plants (Shen and Rochelle, 1998;Dahlan et al., 2006;Barrelo et al., 2008;Hu et al., 2010).
Laboratory experiments have established that the reaction is faster under higher pH conditions (Lee and Schwartz, 1983;Clifton et al., 1988;Littlejohn et al., 1993) and in the presence of O 2 (Littlejohn et al., 1993;Shen and Rochelle, 1998).
In a study to investigate stoichiometry of this reaction, Lee and Schwartz (1983) bubbled NO 2 through a HSO 3 -solution and measured the change of [H + ].Their results suggest that the overall reaction is mainly: Reaction R1 likely proceeds through a free radical chain mechanism (Nash, 1979;Littlejohn, et al., 1993;Shen andRochelle, 1998, Tursic et al., 2001).The reaction steps were suggested to be (Shen and Rochelle, 1998): ) in a product study provides strong evidence for the radical chain mechanism (Littlejohn et al., 1993).Sulfite radicals subsequently under a series of reactions to yield SO 4 2- (Seinfeld and Pandis, 2006, p332).The overall mechanism for the reaction of NO 2 with S(IV) is compiled in Table A1.1.27E+07*a H+ +6.4E-05/ a H+ +55 3 a :There are other reactions and species in this mechanism, but they are expected to be less important than those listed here.b :(1) Takeuchi et al, 1977; (2) Seinfeld and Pandis, 2006, p332, and references therein;(3) Connick et al., 1993   with the aqueous phase pH, as oxidation of sulfite (R1) is much faster than the oxidation of bisulfite (R2) and more of S (IV) exists in the form of sulfite at higher pH.k 1 is calculated to be 1.7×10 5 M -1 S -1 at pH = 4, 3.0×10 5 M -1 S -1 at pH = 4.6, and 2.4×10 6 M -1 S -1 at pH=6.
We subsequently calculated k 1 at different NO 2 (up to 100 ppb) and SO 2 levels (up to 45 ppb) at two pH conditions, pH 4 (Fig. A2) and pH 5 (Fig. A3).Figs.A2 and A3 suggest that mathematically, k 1 could be described as an exponential function of the abundance of NO 2 and SO 2 .For the purpose of deriving an empirical relationship between k 1 and SO 2 and NO 2 concentrations, we use the following equation to link k 1 with NO 2 and SO 2 : We arbitrarily set y to be 0.5, then determine x by fitting k 1 estimated by Eq. 3 with the boxmodel predicted k 1 .It is found at pH=4 and pH=5, Eq.3 can be expressed as Eq. 4 and Eq. 5, respectively: (Eq.5) Figs. A4 compares the model predicted k 1 and the approximations by Eq. 4 and Eq. 5, clearly showing the two equations provide excellent approximations for the box-model calculated k 1 values.We subsequently use Eq. ( 4) and Eq. ( 5) to obtain k 1 values in OBM-SIA.
Table B1 Sensitivity of predicted local contributions of SO 4 2-and NO 3 -in fine particulates during daytime of 25 December on the uncertain parameters in OBM-SIA a a : The concentration unit is μg m -3 and the percentage numbers in parentheses denote percentage changes from the results predicted in base scenario.b : Two average HONO diurnal profiles are used: one is 30% percentile curve and the other one is 80% percentile curve.c : The high and low limits of the cloud pH are suggested by Tanner and Tam (2006).
It is noticed that the OBM-SIA predictions of SO 4 2-are very sensitive to cloud droplet pH, with the percentage change from the base case larger than >50%; moderately sensitive (10%> variation >50%) to the changes of formaldehyde and HONO (g) concentrations, and insensitive to other parameters (Table B1).The modeled NO 3 -is moderately sensitive to the dry deposition velocity of HNO 3 (g) and the levels of formaldehyde and HONO (g).
Photolysis of HONO and formaldehyde are known to be important to ambient OH levels (Su et al., 2008;Cheng et al., 2010).Because their concentrations were not measured at our sampling site, their observations at other locations in Hong Kong or PRD were adopted.The sensitivity study results suggest that on-site concentrations HONO and formaldehyde will be very valuable in future field experiments.
As discussed in the main text, the measured cloud pH in Hong Kong ranges from 3.6 to 4.6 (Tanner and Tam, 2006), and a pH of 4 is assumed in the OBM-SIA calculations.As the cloud pH has a potential significant impact on the SIA concentrations, we next examine the sensitivity of SO 4 2-production on the pH values in more details.The range of pH from 3.6 to 4.6 in Hong Kong is adopted from Tanner and Tam (2006).Contributions by the local process are calculated by OBM-SIA, while contributions by the transport function are determined as the differences between the observations and the contributions by local processes.Positive contributions by transport function refer to transport into the sampling site, while negative contributions refer to transport out of the sampling site.
The strong dependence of aqueous phase production of SO 4 2-on pH is expected as all the SO 4 2-formation pathway in the aqueous, except for the H 2 O 2 oxidation pathways, are known to be pH-dependent (Seinfeld and Pandis, 2006, p308-334).The pH dependence is derived from the fact that dissociated constituents of gaseous SO 2 (HSO (Eq.9) As shown in Fig. B2, H HSO3- * and H SO32- * increase by 1-2 orders of magnitude as the cloud pH increases from 3.6 to 4.6.This illustrates that more alkaline solutions favor the dissolution of SO 2 , which in turn enhances the production of SO 4 2-in aqueous phase.
Another reason for the pH-dependence is that the aqueous phase oxidants (e.g.dissolved O 3 , NO 2 ) react much more rapidly with SO 3 2-and HSO 3 -than with SO 2 H 2 O.For example, The oxidation rate of SO 3 2-and HSO 3 -by dissolved O 3 is 1.5×10 9 and 3.5×10 5 M -1 S -1 , respectively, significantly higher than that of SO 2 H 2 O (2.4×10 4 M -1 S -1 ).The implication of this finding of strong pH dependence is that a buffering effect of the formation of SO 4 2-could be predicted in atmospheres of high NO x and sufficient neutralizing capacity.When the acidic gases (e.g., SO 2 ) are reduced, oxidation rate from S (IV) to SO 4 2could be enhanced due to rising cloud water pH.This effect may partly explain the discrepancy between the still increasing SO 4 2-concentration despite significant drop in SO 2 ambient concentration achieved through stringent control efforts on SO 2 emissions in the PRD (Louie et al., 2005a(Louie et al., , 2005b;;So et al., 2007).

Fig. 1 .
Fig. 1.Schematic diagram of OBM-SIA.In the base run, observed concentrations of gas and particle species and meteorological parameters are used to estimate formation of SO 4 2-, NO 3-and the contributions by different pathways.In the sensitivity run, relative incremental reactivities (RIR) of SO 4 2-and NO 3 -are calculated to assess the impact of reduction in certain precursors on the abatement of ambient secondary inorganic aerosol levels.*Meteorological Parameters.
Fig. 2. Comparison of predicted OH and HO 2 radicals by the OBM-SIA with observations in the Pearl River Delta reported by Hofzumahaus et al. (2009).

Fig. 3 .
Fig. 3. Time series of various observation data in Tung Chung on 21 December (normal day) and 25 December (high PM episode day).(a) PM 2.5 and its chemical compositions, (b) VOCs, and (c) Criteria gaseous air pollutants including NO 2 , NO, SO 2 , CO, and O 3 .In panel (b), observed VOC species have been transformed into the functional groups used in the CB05 mechanism, according to the recipe ofYarwood et al. (2005).Full names of the VOC species are given in TableC1of the Supplementary Materials.

Table 4 .
Data sources and determination of parameters to drive OBM-SIA.Species Data Sources Temperature, Relative Humidity Solar Radiation Auto-weather station NO, NO 2 , O 3 , CO, SO 2 Gas analyzers Mass concentration of PM 2.5 , PM 10 Tapered Element Oscillating Microbalance (TEOM) PAR, ETH, OLE, TOL, XYL, ISOP, TERP Ong), NH 3 (g) Approximated by measurements at Panyu, a suburban district of Guangzhou FORM, ALD 2 Approximated by measurements in the same month at Tsuen Wan, an urban site in Hong Kong ETOH, CH 4 , ETHA, H 2 , O 2 Adopted with typical values (in ppm) in the atmosphere: ETOH = 0, CH 4 = 1.90, ETHA = 1.50 × 10 -3 , H 2 = 0.55, O 2 = 2.00 × 10 5 H 2 O(g) Calculated from observed Temperature and Relative Humidity Water-to-air volume ratio of cloud 5.0×10 -7 (Warneck, 2003) Mixing Height 400-1000 m, typical values in winter in Hong Kong based on LIDAR measurements at Yuen Long, an urban location 19 km northeast to Tung Chung U 8.5 × 10 -4 (ppm) (Hofzumahaus et al., 2009) Cloud coverage On-site observation at the Hong Kong International Airport Size distribution of SO 4 2-and NO 3 -On-site measurements by MOUDI, 8-hr resolution Table 5.Production and loss pathways of HNO 3 + NO 3 -and H 2 SO 4 + SO 4 2-considered in the OBM-SIA.Seq.Pathways to produce or remove NO 3

Fig. 4 .
Fig. 4. Contributions of individual pathways to the total formation rate of NO 3 -in the fine particles.Production and removal pathways are tabulated in Table 5. D denotes daytime, from 7:00 to 19:00 local time.

Fig. 5 .
Fig. 5. Contributions of individual pathways to the total formation rate of SO 4 2-in the fine particles.Production and removal pathways are tabulated in Table 5. D denotes daytime, from 7:00 to 19:00 local time.

Fig. 6 .
Fig. 6.Diurnal variations of (a) NO 3 -and (b) SO 4 2-concentrations in the fine PM and modeled relative contributions by local chemistry and transport function on 25 December 2009.Numbers in parenthesis show the formation rates (μg/m 3 /h).Contributions from the local chemistry are calculated by integrating formation of NO 3 -or SO 4 2-over the given time.Contributions from the transport function are calculated as the differences between the observed values and contributions from the local chemistry.
Fig. 7.The changes of (a) NO 3 -, (b) SO 4 2-, and (c) SIA production in response to one percent reduction in precursor X, i.e., RIR X .RIR X is calculated for reduction in NO x , SO 2 , AVOC, and BVOC.

Fig. 9 .
Fig. 9. Model predicted maximum hourly concentration of (a) NO 3 -, (b) SO 4 2-, and (c) SIA in the fine particles in response to different extents of reduction in precursors.The base scenario uses observations on the PM episode day (i.e., 25 December 2009).

Fig.
Fig. A1 shows example calculated k 1 as a function of reaction time under four different pH conditions, all assuming SO 2 = 30 ppb and NO 2 = 100 ppb.The calculated k 1 quickly stabilizes after ~60s of simulated reaction time, indicating the adjustment is completed with 60 s.The pH-dependence of k 1 is clearly demonstrated in Fig. A1.Apparently, k 1 increases Fig. A2 The calculated effective rate constants of NO 2 +S (IV) in contours as a function of NO 2 and SO 2 concentrations at pH =4 for the aqueous phase.Red open circles represent the effective rate values corresponding to NO 2 and SO 2 daily average conditions at Tung Chung on individual days during December 2009.The red filled triangles represent the effective rate values corresponding to SO 2 and NO 2 concentrations observed during the episode hours on 25 December 2009.

Fig
Fig. B1 shows the contribution of SO 4 2-by local processes and transport functions as a function of cloud pH during Periods II and III on 25 December 2009.It is revealed that the contribution by local process increases rapidly along with cloud pH.In Period III, the contribution of local process exceeds that of transport when cloud pH is higher than 4.5.
Fig.B1 SO 42-in fine particles contributed by the local process and transport function as a function of cloud pH at (a) 8:00-14:00 and (b) 14:00-17:00 on 25 December 2009.The range of pH from 3.6 to 4.6 in Hong Kong is adopted fromTanner and Tam (2006).Contributions by the local process are calculated by OBM-SIA, while contributions by the transport function are determined as the differences between the observations and the contributions by local processes.Positive contributions by transport function refer to transport into the sampling site, while negative contributions refer to transport out of the sampling site.

Fig
Fig. B2 Variation of
Fig. E RH and cloud coverage (CV) on December 21 and December 25, 2009 at Tung Chung

Table 1 .
Specified (S) and unspecified (U) chemical species in the OBM-SIA.

Table 2 .
Dry deposition velocities of important species.
SO 4 by dilution as the mixing height expands.Due to a lack of direct measurements, it is simply assumed that cloud formation rate equals to the evaporation rate and thus, formation rate of H 2 SO 4 can be estimated by Eq. (

Table A1 . Reaction mechanism for the S (IV) -NO 2 aqueous phase reactions a Reactions Reaction constant at 298K(M -1 S -1 ) E/R(K) Ref. b
SO2 is the Henry's law constant.P SO2 is the partial pressure of SO 2 , and K 1 and K 2 denote the first and second acid dissociation equilibrium constants (shown in Table3) for dissolved SO 2 .The effective Henry's law coefficient for SO 2 , H HSO3-   

Table C1 :
Full names of the chemical species in Table 1 of the main text

Table C2
Initial concentration of chemical species to drive current OBM-SIA a: Aqueous phase species concentration would quickly achieved by gas-aqueous phase equilibriums.S-18