Relative contributions of secondary organic aerosol formation from toluene, xylenes, isoprene, and monoterpenes in Hong Kong and Guangzhou in the Pearl River Delta, China: an emission-based box modeling study


  • Siyuan Wang,

    1. Department of Chemistry, Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
    Search for more papers by this author
  • Dongwei Wu,

    1. Division of Environment, Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
    Search for more papers by this author
  • Xin-Ming Wang,

    1. State Key Laboratory of Organic Geochemistry, Guangzhou Institute of Geochemistry, Chinese Academy of Sciences, Guangzhou, China
    Search for more papers by this author
  • Jimmy Chi-Hung Fung,

    1. Division of Environment, Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
    2. Department of Mathematics, Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
    Search for more papers by this author
  • Jian Zhen Yu

    Corresponding author
    1. Department of Chemistry, Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
    2. Division of Environment, Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
    • Corresponding author: J. Z. Yu, Department of Chemistry & Division of Environment, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong, China. (

    Search for more papers by this author


[1] Secondary organic aerosols (SOA) formed from common anthropogenic and biogenic volatile organic compounds (VOCs) account for a significant portion of organic particulate matter in the ambient atmosphere. The Pearl River Delta (PRD) in southern China, located in the subtropics and as a region with intensive manufacturing industries, has significant emissions of both anthropogenic and biogenic VOCs. Two recent SOA tracer-based measurement studies, one in Hong Kong (located at the mouth of the PRD) and the other at a site 20 to 50 km downwind of urban Guangzhou districts in the middle of the PRD, show a rather considerable difference in the relative SOA contributions from one group of two biogenic VOCs (isoprene and monoterpenes) and one group of anthropogenic VOCs, namely, toluene + xylenes. In Hong Kong, more SOA was formed from isoprene and monoterpenes than from toluene and xylenes, although the relative contributions of the two groups of VOCs were reversed at the site downwind of Guangzhou. An emission-based 0-D box model has been developed to investigate this issue. The emission inputs of major inorganic pollutants and VOCs are generated using the programs Sparse Matrix Operator Kernel Emissions and Model of Emissions of Gases and Aerosols from Nature for this region. Toluene/xylene emissions in Guangzhou are more than twice that in Hong Kong whereas isoprene and monoterpenes emissions were similar at the two locations. The model incorporates a CB05 chemical mechanism and gas–particle partitioning of condensable VOC oxidation products to simulate SOA formation from major VOCs including isoprene, monoterpenes, toluene, and xylenes. The model-simulated VOCs fall within the range of ambient observations, demonstrating reasonable representation of emissions and oxidation of VOCs. The model simulates the sum of the SOA formation from isoprene, monoterpenes, and toluene + xylenes. In Hong Kong, monoterpenes are the major contributor (up to 70%), followed by isoprene (14%–48%) and toluene + xylenes (15%–43%). In Guangzhou, toluene + xylenes contribute more to SOA than isoprene and monoterpenes (up to 76% from toluene + xylenes vs. 13%–44% from isoprene and 10%–45% from monoterpenes). The reasonable agreement between the simulated SOA for the target VOCs and the tracer-based measurements suggests that the significantly larger toluene + xylene emissions in Guangzhou could explain the substantial difference in relative SOA contributions by the two groups of VOCs in the two cities. This work has also identified a lack of good measurements of monoterpenes and their SOA tracers to be an important data deficiency in assessing the relative contributions of biogenic and anthropogenic VOCs to SOA in this region.

1 Introduction

[2] Secondary organic aerosols (SOA) formed from the atmospheric reactions of anthropogenic and biogenic volatile organic compounds (VOCs) account for a large portion of organic particulate matter. Numerous chamber studies have demonstrated that toluene, xylenes, isoprene, and terpenes are important VOC precursors for SOA formation [Pandis et al., 1991; Kroll et al., 2005; Kleindienst et al., 2006; Kroll et al., 2006; Ng et al., 2007a; Ng et al., 2007b]. Field measurements of their SOA tracers spanning a range of environments also suggest that these species contribute a significant fraction of identified SOA components [e.g., Kleindienst et al., 2007; Hu et al., 2008; Stone et al., 2009]. Current air quality models largely underpredict organic aerosol [Morris et al., 2006; Volkamer et al., 2006; Goldstein and Galbally, 2007], strongly suggesting that there are missing potential pathways of SOA formation and/or missing SOA precursors in the SOA formation mechanisms, which have been demonstrated through smog chamber studies [e.g., Robinson et al., 2007; Volkamer et al., 2009]. In recent years, much work has been done in the modeling community to include additional SOA sources in models, as well as in explaining the lack of closure between measured and modeled SOA [e.g., Matsui et al., 2009; Hodzic et al., 2010; Dzepina et al., 2011; Li et al., 2011]

[3] A number of approaches are available to estimate SOA formation in the ambient environment. Among them, an SOA tracer-based approach uses the relationship between certain VOC precursors and their SOA tracers derived from laboratory experiments, thereby providing the possibility of tracking SOA formation from specific SOA precursors [Kleindienst et al., 2007]. Although this method has its uncertainties rooted in oversimplifying the complicated SOA formation processes with a single chamber-derived tracer “yield,” it has the unique advantage of revealing the direct linkage between SOA formation and major precursors. Tracers derived from the oxidation of common VOC precursors including toluene, xylenes, isoprene, and monoterpenes have been used to estimate SOA formation from these VOCs in a number of studies [Kleindienst et al., 2007; Hu et al., 2008; Fu et al., 2009; Ding et al., 2012].

[4] Alternatively, emission-based and source-based approaches have been widely used in model simulations in which SOA formation is estimated through simulating the oxidation kinetics of SOA precursors and partitioning of condensable oxidation products into the aerosol phase [e.g., Pandis et al., 1992; Odum et al., 1996; Strader et al., 1999; Matsui et al., 2009]. The treatment of partitioning has evolved over time from simply assuming a sufficiently low vapor pressure of 2 ppt for each condensable product species [Pandis et al., 1992], to absorption partitioning equilibrium in an organic aerosol phase [Pankow, 1994], and more recently, to the reformulation of the Pankow theory by replacing the partitioning coefficient with an effective saturation concentration (C*), which is a function of total organic aerosol concentration [Donahue et al., 2006]. In this work, we use the approach by Strader et al. [1999] in that the partitioning is approximated by a pseudo ideal solution. Ca,i, the aerosol-phase concentrations of species i (in µg m−3) is calculated using the equation below:

display math(1)

where Ctotal,i is the total concentration of condensable gas i, C0 is the organic loading, and MW is the molecular weight of different components. Ci*, the effective saturation concentration of pure organic compound i, determines the potential of an individual condensable product partitioning into aerosol phase. Ci* is either experimentally determined or derived from theoretical work, and its temperature dependency is described by the Clausius–Clapeyron equation. This approach has been integrated into various models along with a choice of certain gas-phase chemical mechanisms (e.g., CB4, CB05, SAPRC99, and SAPRC07) to simulate SOA formation [e.g., Strader et al., 1999; Carter, 2010; ENVIRON, 2010]. This modeling approach also provides the capability of tracking the contributions of individual VOC precursors to SOA formation.

[5] The Pearl River Delta (PRD), an economically more developed region in China, features large anthropogenic VOC (AVOC) and NOx emissions and severe ozone episodes [Zhang et al., 2008]. Due to the highly concentrated industrial activities, high levels of aromatics have been frequently observed in PRD, significantly higher than the ambient measurements in many other locations around the world. Large contributions from gasoline exhaust, coating/painting, and the oil refinery industry make toluene the most abundant ambient VOC in many PRD cities [Liu et al., 2008a; Liu et al., 2008b]. For instance, measurements using an in situ online gas chromatography (GC)-flame ionization detector in the PRD region showed that in 13 out of 19 days, the daily maximum level of toluene exceeded 20 ppbv in Guangzhou (GZ), a densely populated city. In the same period in a rural site in PRD, Xin Ken, the daily peak of toluene levels exceeded 20 ppbv in 9 out of 16 measurement days [Wang et al., 2008]. The maximum toluene levels measured using canister GC-mass spectrometry (MS) are reported to be 36.91 and 54.61 ppbv in GZ and Xin Ken, respectively [Liu et al., 2008c].

[6] Hong Kong (HK) is a highly urbanized city, taking up a land mass area of 1104 km2. HK, at the mouth of the PRD, is frequently affected by pollution transported from the PRD to the north but its local AVOC sources are significantly less intense. In comparison with locations in the PRD, the ambient levels of aromatics in non-roadside environments in HK are much lower. For example, a 1 year measurement project in HK indicates that the average level of toluene is 1.03 to 4.34 ppbv in typical urban and rural locations [Guo et al., 2007]; another 2 year measurement in HK (Table 1) showed similar results, and the maximum level of toluene was 14.56 ppbv [Lau et al., 2010]. The typical levels of AVOCs in major PRD cities are comparable to those observed in roadside environments of HK with heavy traffic influence. Table 1 lists more ambient measurements, consistently illustrating the spatial gradient of these major AVOCs between HK and the PRD.

Table 1. Ambient VOC Measurements in HK and in the PRD Region (Results are Given in Average ± Standard Deviation Unless Specified Otherwise)
  • a Summer and autumn measurements in various sites in HK from 2006 to 2008, n = 95 [Lau et al., 2010].
  • b Ambient measurements at Tsuen Wan air quality station, an urban background station in HK, n = 7 [HKEPD, 2005].
  • c Roadside measurements at Hong Kong Polytechnic University and Mong Kok in HK, n = 10 [HKEPD, 2005].
  • d Averaged from Shing Mun Tunnel and Tsuen Kwan Tunnel experiments in HK, n = 29 [HKEPD, 2005].
  • e Ambient measurements in GZ (urban) and Xin Ken (rural), approximately 90 samples each [Liu et al., 2008c].
  • f Ambient measurement in GZ (urban, 42 samples) and Dongguan (urban, 48 samples) [Barletta et al., 2008].
  • g Sampling near a significant industrial point source or an industrial complex (industrial) in PRD, n = 15.
  • h Sampling in an area (away from any known point sources) with mixed industrial and urban sources (industrial–urban) in PRD, n = 25.
  • i Sampling in an area (away from any known point sources) with mixed industrial and suburban/rural sources (industrial–suburban) in PRD, n =38 samples.
  • g–i

    See [Chan et al., 2006] for detailed descriptions.

  • j

    Within this table, monoterpenes represent the sum of α-pinene and β-pinene.

  • k

    Xylenes are the sum of m-, p-, and o-xylene.

  • l

    N.A. = not analyzed; UCI = University of California, Irvine, CA; PKU = Peking University, China [herein blind intercomparison of 55 major NMHCs had been conducted with UCI, see Liu et al., 2008a].

Ambient Measurements in HK (unit: ppbv) 
Tap Mun (Rural) a1.05 ± 0.710.91 ± 1.530.58 ± 0.45N.A.l1.25 ± 1.360.43 ± 0.91UCI l
Central West (Urban) a1.24 ± 0.741.41 ± 0.700.19 ± 0.17N.A.1.96 ± 1.760.60 ± 0.71UCI
Tung Chung (Urban) a1.06 ± 0.721.03 ± 0.750.34 ± 0.26N.A.1.40 ± 1.910.44 ± 0.88UCI
Yuen Long (Urban) a1.38 ± 0.742.14 ± 0.980.26 ± 0.11N.A.2.65 ± 2.910.78 ± 0.86UCI
Tsuen Wan (Urban) b1.20 ± 0.541.74 ± 0.310.59 ± 0.250.03 ± 0.024.30 ± 1.990.91 ± 0.28UCI
Polytechnic University (Roadside) c2.93 ± 0.858.56 ± 1.120.56 ± 0.130.10 ± 0.0512.10 ± 3.463.40 ± 0.47UCI
Mong Kok (Roadside) c3.33 ± 0.658.47 ± 0.830.27 ± 0.120.06 ± 0.026.93 ± 1.872.27 ± 0.80UCI
Tunnel Experiments (Hong Kong) d3.16 ± 0.8525.90 ± 2.840.53 ± 0.390.04 ± 0.0211.2 ± 3.463.63 ± 4.37UCI
Ambient Measurements in the PRD Region (unit: ppbv) 
Guangzhou (Urban) e5.58 ± 3.346.55 ± 4.820.22 ± 0.170.18 ± 0.187.01 ± 7.331.98 ± 1.92PKU l
Xin Ken (Rural) e3.07 ± 1.262.68 ± 2.190.17 ± 0.150.17 ± 0.228.46 ± 9.942.65 ± 3.97PKU
Guangzhou (Urban) f1.89 ± 0.103.97 ± 0.421.63 ± 0.180.16 ± 0.105.87 ± 0.742.11 ± 0.69UCI
Dongguan (Urban) f1.60 ± 0.063.07 ± 0.270.68 ± 0.160.14 ± 0.026.13 ± 0.812.00 ± 0.17UCI
Industrial Areas (PRD) g2.3 ± 1.35.1 ± 3.80.6 ± 0.3N.A.13.5 ± 11.86.1 ± 7.1UCI
Industrial–Urban (PRD) h1.8 ± 0.62.6 ± 1.30.4 ± 0.2N.A.11.5 ± 11.63.4 ± 2.2UCI
Suburban (PRD) i1.4 ± 0.62.0 ± 1.10.5 ± 0.7N.A.7.3 ± 5.42.2 ± 2.1UCI

[7] Located in the subtropics, HK and the PRD feature a large coverage of vegetation, implying significant biogenic VOC (BVOC) emissions, e.g., isoprene and terpenes [Zheng et al., 2009]. Unlike AVOCs, BVOCs in HK and the PRD show small differences in their ambient concentrations, as shown in Table 1. In general, ambient isoprene levels are similar between HK and GZ, whereas the limited number of ambient measurements indicate that the concentrations of monoterpenes (α-pinene + β-pinene) are slightly higher in GZ than those in HK.

[8] The high abundances of AVOCs and BVOCs provide considerable potential for SOA formation. Two recent studies report SOA tracer measurements in this region, one in HK [Hu et al., 2008] and another in Wan Qing Sha, a site in a suburban district but downwind of the densely populated urban districts of GZ [Ding et al., 2012]. HK and Wan Qing Sha are less than 80 km apart, but the relative SOA contributions of one group of BVOCs consisting of isoprene and monoterpenes and one AVOC group consisting of toluene and xylenes were found to differ significantly. For the ease of discussion, we use the term “SOAtx+i+m” hereafter to represent the sum of SOA from toluene + xylenes, isoprene, and monoterpenes; SOAtx for SOA derived from toluene + xylenes; SOAi for isoprene-derived SOA; and SOAm for monoterpenes-derived SOA. In Wan Qing Sha, SOAtx accounts for 80% to 86% of SOAtx+i+m, much higher than the portion accounted for by isoprene (8%–20%) and monoterpenes (1%–5%) [Ding et al., 2012]. The results in HK are very different: monoterpenes contribute 38% to 62% of SOAtx+i+m whereas toluene + xylenes and isoprene account for minor fractions (SOAtx, 17%–29%; SOAi, 21%–33%) [Hu et al., 2008]. Can such a discrepancy be explained by the different characteristics in emissions between the two locations? In this work, we have developed an emission-based 0-D box model that incorporates emissions generated for this region using Sparse Matrix Operator Kernel Emissions (SMOKE) and Model of Emissions of Gases and Aerosols from Nature (MEGAN) to simulate the gas-phase chemistry and SOA formation. The simulation results will be evaluated against the two sets of aforementioned tracer-based SOA measurements. The objectives are to understand the relative importance of toluene, xylenes, isoprene, and monoterpenes to SOA formation in HK and GZ in the PRD region and to identify data deficiencies in VOC emissions or SOA estimation.

[9] In this work, we focus on the SOA formation from toluene + xylenes, isoprene, and monoterpenes. This does not suggest that these precursors can explain the total SOA in this region. In fact Hu et al. [2010] showed that SOAtx+i+m estimates using the tracer measurements were consistently lower than the total SOA derived using either chemical mass balance or positive matrix factorization analysis. In recent years, semivolatile and intermediate volatility VOCs (S/IVOCs) from primary emissions have been found to form significant SOA mass upon fast oxidation [e.g., Robinson et al., 2007]. Although there are challenges in predicting atomic O/C ratio [Dzepina et al., 2011], this mechanism is expected to contribute to the currently underestimated SOA mass. Primary emissions of S/IVOCs are expected to be appreciable in this region, and SOA from this group of VOCs need to be investigated for a more comprehensive assessment of SOA precursors in future studies.

[10] The PRD region features high relative humidity, high ambient NOx, and VOC loadings. GZ is the largest city in the PRD and is influenced by both vehicular and industrial emissions. GZ represents a typical urban condition in the PRD, where the ambient concentrations of NOx and VOCs (especially aromatics) exceed many other places around the world. This emission-based box modeling work, together with ambient SOA measurements, provides a first-order assessment of the emission data of the important VOC precursors. This could in turn facilitate the evaluations of emission input data and chemical mechanism for more comprehensive simulation of SOA formation using three-dimensional (3-D) air quality models. Insights gained in this study help us to understand aerosols in the PRD, which is a significant aerosol source region in East Asia.

2 Model Description

2.1 Model Components

[11] The diurnal variations of temperature, relative humidity, and mixing layer height in the 0-D box model are set to approximate typical summer conditions in HK, with the temperature varying from 291 K at night to 295 K at noon, relative humidity from 50% to 70%. Temperature and relative humidity have opposite diurnal patterns. Mixing layer height is assumed to have a minimum at 500 m in the nighttime and begins to rise at 6:00 am local time and peaks at noon with the maximum set at 1200 m. The diurnal patterns of the above parameters are simulated to vary as sinusoid functions of local time. Photolysis is assumed to take place under clear-sky conditions, in which solar zenith angle is calculated using the method described by Finlayson-Pitts and Pitts [1999]. Initial concentrations for major species are based on observations in HK. Base case emissions of anthropogenic pollutants are generated by SMOKE and emission fluxes of biogenic species are generated by MEGAN, the details of which are discussed in the next section. Dry depositions are adopted from those described by Huang et al. [2011]. The model is prerun for 24 h for the establishment of intermediate species.

[12] A condensed mechanism CB05 is used to simulate reactions of the gas-phase species, in which photolysis rates are derived from either the Jet Propulsion Laboratory (JPL) database or International Union of Pure and Applied Chemistry (IUPAC). SOA formation mechanism is based on the method used in CAMx v5.30 [ENVIRON, 2010], in which five VOC precursors, i.e., toluene, xylenes, isoprene, terpenes, and sesquiterpenes, were considered. In this work, due to a lack of relevant data in this region, sesquiterpenes are not included in either gas-phase chemistry or SOA formation mechanism. The aqueous-phase process is not included.

[13] The entrainment and subsequent dilution due to the increased mixing layer height is ignored in the model. In principle, this effect is determined by the term + (Ci0-Ci*) (dH/dt)/H in the rate equation [Seinfeld and Pandis, 2006], where Ci0 is the “background” concentration of species i, and Ci and H are the concentration and model mixing layer heights at calculation step t, respectively. With the diurnal variation of the mixing layer height in this model, the maximum value of the term (dH/dt)/H is approximately 1 × 10−4 s−1. The corresponding timescale (>2.8 h) is therefore faster than (but comparable to) that of dry deposition for HNO3 (~14 h) [Huang et al., 2011]. The treatment used here means that the modeled SOA reflects an upper limit with respect to additional possible processes such as dilution due to the increase of mixing layers. The associated uncertainty arising from this simplification would be similar to those associated with uncertainties in dry deposition. This simplification would not affect the relative contributions of different SOA components.

[14] The ordinary differential equations (ODEs) are solved numerically using the double-precision version of the Livermore Solver for ODEs [Hindmarsh, 2010], which facilitates the sparse matrix manipulation and significantly enhances computational efficiency for stiff problems. Strict and dynamic tolerance control strategies are used to maintain sufficient accuracy. The model framework has been evaluated by simulating a subset of reactions extracted from MCMv3.2 ( that consist of thousands of species and reactions. The same sets of reactions are simulated using FACSIMILE, a widely used commercial solver for ODE systems [Warneck, 2003; Hofzumahaus et al., 2009]. The model results from the two solvers have been compared and the discrepancy is found to be fairly small, indicating the reliable performance of the Livermore Solver for ODEs and the model framework.

[15] For the investigation of the nonlinear nature of certain processes (e.g., ozone or SOA formation), isopleths for species of interests as a function of emission factors of NOx and nonmethane hydrocarbons (NMHCs) are constructed by running the model with different model settings of NOx and NMHC emission factors. Values for the isopleths are taken at 15:00 (local time) on the third day for further analysis. The time 15:00 is chosen to capture the approximate maximums of HOx radicals, and ozone. OH and HO2 radicals would typically reach their daily maximum at around 12:00 to 13:00, whereas ozone tends to peak in the late afternoons (e.g., 16:00–17:00).

2.2 Emissions

2.2.1 Emission Scenarios for Hong Kong

[16] Because this 0-D box model is emission-based, the emission inputs for major species are given special attentions and details are described below. Hourly resolved 3×3 km gridded anthropogenic emissions are generated using SMOKE, whereas biogenic emissions are generated using MEGAN. The latest available emission inventories in HK and the PRD are used [Liu et al., 2008a; HKEPD, 2009; Zheng et al., 2009; Leung et al., 2010; Zheng et al., 2010]. Yuen Long, an urban area in the northern part of HK, is selected to represent HK in the box model. Emissions are based on nine-grid–averaged emission outputs centered at Yuen Long, and covering an area of 81 km2 surrounding it. The raw emission outputs from SMOKE and MEGAN are used as Base Case I in this study.

[17] Previous studies indicate that the major PM components in HK could be significantly elevated by regional transport from the PRD region, especially on days when meteorological conditions favor land–sea breeze circulation [Fung et al., 2005; Louie et al., 2005; Hagler et al., 2006; Lo et al., 2006]. Unlike 3-D chemical transport models, box models are incapable of simulating transport processes. To compensate for this deficiency inherent to the box model, we have increased the raw emissions by certain factors to represent local emissions plus transport input on days of significant regional influence (called regional days thereafter, for ease of discussion, following the convention set by Hu et al. [2008]). The multiplying factors were set according to the typical ambient levels measured on the regional days. More specifically, the NOx emission is multiplied by a factor of 4.5, CO by 3, methane and other AVOCs by 5, formaldehyde by 3, and other oxygenated VOCs remain unchanged. Biogenic emissions for isoprene remain unchanged, whereas emissions for monoterpenes are multiplied by 3 to match the ambient measurements of monoterpenes. The modified emissions are used as Base Case II.

[18] In summary, Base Case I is for days that are dominated by emissions local to HK whereas Base Case II represent days with significant regional transport (i.e., the high pollution days). The average daytime emission fluxes of NOx and major VOC species are given in Table 2. The emission fluxes for both NOx and VOCs in HK calculated in this work are generally lower than but still comparable to those reported by Huang et al. [2011], who estimated emission fluxes using SMOKE but with a previous release of emission inventories and source profiles for this region (Table 2).

Table 2. Averaged Daytime Emission Fluxes for NOx and Major VOC Species Obtained with SMOKE or MEGAN (Unit: Molecule cm−2 s−1)
  1. a

    Daytime averaged emission fluxes for typical urban HK and GZ in this work, generated by SMOKE/MEGAN with the latest emission inventories [Zheng et al., 2009; Zheng et al., 2010] and source profiles [Liu et al., 2008a; Leung et al., 2010].

  2. b

    Daytime averaged emission fluxes of anthropogenic species were estimated using SMOKE but with previous release of emission inventories and source profiles, and isoprene emission flux is calculated using SMOKE but monoterpenes was roughly assumed to be approximately 60% of isoprene [Huang et al., 2011].

HK Base Case I a2.29 × 10117.64 × 1093.60 × 10102.40 × 10117.53 × 1090.52 × 10111.16 × 1010
HK Base Case II a1.03 × 10123.82 × 10101.80 × 10112.40 × 10112.26 × 10102.56 × 10115.82 × 1010
GZ Base Case a2.04 × 10127.71 × 10102.38 × 10112.62 × 10113.44 × 10105.05 × 10111.43 × 1011
HK emissions adopted by Huang et al b1.99 × 10121.20 × 10101.80 × 10113.35 × 10112.00 × 10111.80 × 10113.00 × 1010

2.2.2 Emissions for Guangzhou

[20] Emissions of NOx, VOCs, and other major pollutants for GZ are also generated using SMOKE and MEGAN, and evaluated against observational data sets. Emissions from two downtown districts in GZ (Li Wan and Tian He) are averaged to represent the emissions of urban GZ. Emissions in each district are based on nine-grid–averaged emission outputs, covering an area of 81 km2 (similar to the treatment for Yuen Long, HK). As a fast-growing economic region with intensive manufacturing industries, the PRD has large industrial emissions (e.g., coating and solvent use, furniture manufacturing) in addition to vehicular emissions, resulting in rather high levels of aromatics in its major cities as well as in the surrounding rural areas [Chan et al., 2006; Liu et al., 2008b]. In recent years, control measures have been implemented to reduce emissions from high-polluting diesel-powered vehicles and motorcycles in GZ, making gasoline and liquefied petroleum gas vehicles the major mobile sources in GZ (Guangzhou Environmental Protection Bureau, Compared with HK, GZ has larger emission fluxes for both NOx and VOCs, especially toluene + xylenes (Table 2). The emission flux in GZ is approximately 2 and 2.5 times larger for toluene and xylenes, respectively, than those in the Base Case II scenario for HK. We note that the emission rates for isoprene and monoterpenes in GZ are only slightly higher than in HK (1.1 times for isoprene and 1.5 times for monoterpenes).

2.2.3 Emission Data Evaluation

[21] Ratios of selected VOCs in the emission inputs are compared with those for the known major sources in the region for the purpose of evaluating the reasonability of the emission flux data. In the following paragraphs, we examine ethene/toluene and ethane/ethene ratios.

[23] Ethene emissions are dominated by vehicular exhaust whereas toluene could be from both gasoline vehicles and solvent uses. Tunnel experiments [HKEPD, 2005] and source profile studies [Liu et al., 2008a] indicate that vehicular emissions typically have higher ethene/toluene ratios (i.e., HK tunnel, 2.31; PRD tunnel, 5.49; gasoline vehicle, 3.50), but the ratio in painting and solvent use sources is nearly zero. Data from the Hong Kong Environmental Protection Department indicate that local emissions of VOCs in HK are dominated by road transports (mainly gasoline vehicles) [HKEPD, 2009]. In the PRD, painting and solvent use accounts for a large fraction of ambient VOCs [Liu et al., 2008b]. Hence, the ratio of ethene/toluene is expected to be lower in the PRD than that in HK and such characteristics are captured in the SMOKE-generated emission data. The ethene/toluene ratio in emissions in HK (0.69–0.70) is higher than that in GZ (0.47), consistent with the known difference in VOC source characteristics between the two cities. The ethene/toluene ratio in GZ is very close to that reported in the PRD emission inventory (0.49) [Zheng et al., 2009].

[24] The ratio of ethane/ethene in the emissions in this work (HK, 0.21; GZ, 0.32) is lower than the ambient measurements (HK, 0.34–1.15; GZ, 0.45–0.15), consistent with the expected faster degradation of ethene than ethane in the atmosphere.

[25] The key factors influencing the estimates of BVOC emissions are land type/vegetation type, solar radiation, and temperature. Considerable spatial and temporal variations and large uncertainties are expected in the BVOC emissions estimates. Table 2 summarizes the biogenic emission fluxes in HK and GZ used in this study. The emission ratio of isoprene/monoterpenes is approximately 11 in HK and 7.9 in GZ. Only limited ambient measurements of monoterpenes are available in this region, as shown in the previous section and in Table 1. The available measurements reveal that the ambient ratio of isoprene/monoterpenes is 4.8 to 19.7 in HK and 4.6 to 10.1 in the PRD region. Ambient measurements could provide some indications for data quality of the emission estimates. However, we note that the ambient measurements of monoterpenes have considerable uncertainties for several reasons. First, monoterpenes (e.g., α-pinene, β-pinene, limonene) usually have ambient concentrations at least one or two orders of magnitude lower than isoprene, and the calibration concentration range adopted for other VOCs might not be appropriate for monoterpenes. Second, ozone oxidation is a considerable removal pathway for monoterpenes, whereas this plays a minor role for other VOCs. A commonly accepted analytical method for the VOC species is canister-sampling followed by GC-MS analysis. However, not all canister-sampling setups are equipped with ozone scrubbers to remove ozone from the sampled air. Those setups without ozone scrubbers would underestimate the ambient monoterpenes. With the consideration of the analytical uncertainties for monoterpenes, the biogenic emission fluxes calculated using MEGAN are generally consistent with the ambient measurements.

[26] With the emissions from SMOKE and MEGAN, the model-predicted concentrations of methane, ethane, toluene, xylenes, isoprene, and monoterpenes in HK and GZ fall within the ranges of ambient measurements whereas the model slightly overpredicts CO and formaldehyde in HK (Figures 1 and 2). The good agreement in the order of magnitude demonstrates that the emissions and photochemistry used in this model are reasonable for representing the typical conditions in HK and GZ.

Figure 1.

Comparison between measured ambient concentrations of CO and selected VOCs and their simulated daytime concentration ranges in HK under Base Case II scenario. Simulation results are shown in the form of daytime range (8:00–18:00, local time). The measured data are shown in mean ± standard deviation [HKEPD, 2005].

Figure 2.

Comparison between measured ambient concentrations of CO and selected VOCs and their simulated daytime concentration ranges in GZ. Ambient measurements from [Chan et al., 2006; Barletta et al., 2008; Liu et al., 2008c].

3 Results and Discussion

3.1 Basic Photochemistry

[27] The model simulates daily maximum concentrations of OH and HO2 radicals at 2.1 × 106 molecule cm−3 and 3.5 × 107 molecule cm−3, respectively, in good accordance with field observations in the PRD [Zhang et al., 2008]. The modeled ozone peaks in the late afternoon and reaches 40 to 65 ppbv, which is also typical in urban HK. Figure S1 shows the noontime value of ozone, ozone + NO2 as a function of emission factors of NOx and NMHCs (after prerun for 24 h) in HK, providing information about ozone/NOx/VOC chemistry in various scenarios of precursor emissions. The ozone formation sensitivity regimes (NO titration, VOC limited, and NOx limited) are clearly shown in this figure. As can be seen from Figure S1, ozone formation in HK is generally VOC-limited, so is GZ (data not shown), both consistent with previous studies in the region [Zhang et al., 2008; Li, 2011]. The characterization of HK and GZ being in the VOC-limited photochemical region can also be demonstrated in the OH radical variation in response to perturbations in isoprene emissions. In HK, a reduction of 20% in the isoprene emission leads to a 33% reduction in OH levels, whereas in GZ, the corresponding reduction in OH level is 26%.

3.2 SOA Formation, Yields, and Effects of Temperature and Organic Loading

[28] Chamber-derived SOA yields under simulated atmospheric conditions are commonly used to estimate the SOA formation in models. Chamber studies could hardly bracket the complexity of SOA formation processes encountered in the ambient environment. Such a highly simplified approach is recognized to have multiple drawbacks, such as the following: (1) failure in properly representing the higher-generation oxidation; (2) inability in accounting for heterogeneous reactions/multiphase reactions; and (3) likely missing certain SOA formation pathways. Despite the drawbacks, it provides a way to evaluate SOA formation from known SOA precursors and the influence of major gas-phase chemistry and some other processes. Figure S2 shows the simulated SOA yields from toluene and xylenes as a function of temperature and organic loading (M0). Temperature affects the effective saturation concentration and hence alters the gas-phase/aerosol-phase fraction of condensable oxidation products. Preexisting organics would also influence the extent of partitioning into the aerosol phase. From this figure, it is seen that an increase of temperature leads to a decrease in SOA yields, and the elevated M0 level would increase the SOA yields. These general trends are consistent with our understanding of SOA formation (as shown in Figure S2) [Odum et al., 1997].

[29] In the model simulation for HK, M0 is assumed to be 5 µg m−3 in Base Case I, and 20 µg m−3 in Base Case II, in reference to ambient measurements [Hu et al., 2008]. In Base Case I, the SOAtx+i+m concentration is simulated to reach 1.1 µg m−3 at the end of a 48-h simulation period. In the Base Case II scenario, the SOAtx+i+m reaches 4.3 µg m−3 at the end of the simulation. If the preexisting organics were reduced to 10 µg m−3, the SOAtx+i+m formation would decrease to 3.3 µg m−3 at the end of the simulation. The higher SOAtx+i+m predicted in Base Case II is also partially due to the emissions of NOx and VOCs 3 to 5 times higher than those in Base Case I. More detailed analysis of the emission dependence of SOA formation is given in the following section.

3.3 Emission Dependence of SOA Formation

[30] Figure 3 shows the modeled SOA formation as a function of emission rates of NOx and total NMHCs. As shown in this figure, an increase in NMHC emissions leads to more SOA formation. The relationship of NOx emission rates with SOA formation, however, is more complicated. In a NOx-sensitive regime, a reduction in NOx emission would lead to a decrease in OH levels. This results in a slight decrease in SOA formation, as OH levels largely control the initial oxidation of VOCs. However, in a NOx-saturated regime (i.e., VOC limited regime), a reduction in NOx emission would cause an increase in OH levels; thereby, an increase in the SOA formation follows. A few scenarios of changes in SOA formation in response to NOx emissions for HK are examined and summarized in Table 3. Because HK is generally in a NOx-saturated regime, a reduction of 20% in NOx emissions would lead to a 16% increase in SOAtx+i+m mass, in which SOAtx increases by approximately 38%, SOAi by 7%, and SOAm by 1.4%. Similar results are found for GZ.

Figure 3.

SOA formation potential as a function of emission factors of NOx and NMHCs in HK. The emission rates (dimensionless) are relative to the Base Case II emission rates shown in Table 2. For example, the data point (1,1) denotes the Base Case II for HK, whereas (2,1) means the NMHCs emissions are doubled whereas NOx emissions remain unchanged. The color map shows the noon value of OH radical concentration. Solid grayish lines are the SOA mass concentration isopleths and the numbers on the line are SOA concentrations in µg/m3.

Table 3. Response of SOA and Ozone Formation to Reduction in AVOC Emissions (In Each Scenario, Only AVOC Emissions are Changed, whereas the Emissions for BVOCs Remain Unchanged)
 EmissionAromatics-SOA (%) aIsoprene-SOA (%) aMonoterpenes-SOA (%) aSOAtx+i+m (%) aO3 (%) b
  1. a

    Reduction in mass concentration.

  2. b

    Reduction in mixing ratio.

HKAVOC × 0.8−27.9−2.4−1.4−9.8−35.1
AVOC × 0.5−55.5−4.7−4.2−21.0−57.5
NOx × 0.8+38.2+7.7+1.4+16.3+63.0
 NOx × 0.5+43.8+8.5+1.5+18.6+71.1
GZAVOC × 0.8−27.8−1.3−0.4−14.4−29.5
AVOC × 0.5−59.3−4.4−10.4−33.8−56.0
NOx × 0.8+45.3+5.6+0.6+27.8+73.6
NOx × 0.5+41.5+4.9−0.5+25.5+192.5

[31] For the variation of NMHC emission factor in Figure 3, both AVOCs and BVOCs are assumed to vary at the same proportion. Although the emissions of AVOCs can be controlled, BVOCs cannot. As such, Figure 3 is not suited to probe the sensitivity of SOA formation to VOC emission control. We therefore run separate simulations, assuming the same base case BVOCs but different AVOCs. The results are discussed in the following paragraphs.

[32] VOC source apportionment studies in the PRD region indicate that the predominant VOC sources are vehicular emissions (31%–51%) and coating and solvent usage (16%–44%) [Liu et al., 2008b]. In HK, these two sources contribute substantial fractions of ambient VOCs as well (vehicular emissions, 31%–54%; solvent usage, 11%–18%) [Lau et al., 2010]. In comparison with vehicle emissions control, the evaporation of fuel or solvents might be easier to control (limit the VOC content in paints, promote water-based solvents, etc.). The effect of anthropogenic emissions of VOCs on SOA formation is evaluated by examining the reduction in SOA in response to a 20% to 50% reduction in AVOC (roughly corresponding to the combined contribution by evaporation-related sources). As seen in Table 3, in response to a 50% reduction in AVOCs in HK, SOAtx is predicted to be reduced by approximately 56%, biogenic SOA by approximately 9%, and SOAtx+i+m by 21%. In GZ, in response to a 50% AVOC reduction, SOAtx is reduced by 59%, biogenic SOA by approximately 15%, and SOAtx+i+m by 34%. This simple calculation demonstrates that SOA reduction is not linearly proportional to the reduction in AVOCs due to the nonlinear nature of the VOC oxidation and SOA formation chemistry. These results are generally in line with those reported by Carlton et al. [2010] in their study of the interaction of anthropogenic versus biogenic emissions over the continental United States. Their modeling results showed that with a 50% reduction in the total controllable emissions of AVOCs, the total reduction in anthropogenic SOA, biogenic SOA, and total SOA would be approximately 50%, approximately 5%, and approximately 10%, respectively.

3.4 SOA Contributions and Comparison With Tracer-Based Measurements

3.4.1 Hong Kong

[33] Figure 4 compares the simulated SOA contributions from individual VOC precursors (aromatics, isoprenes, and monoterpenes) with the measured SOA in HK and PRD [Hu et al., 2008; Ding et al., 2012]. In the SOA tracer measurements, 2,3-dihydroxy-4-oxopentanoic acid was the tracer to estimate toluene SOA. Chamber studies suggest that this toluene tracer may also be formed from the oxidation of xylenes [Kleindienst et al., 2007]. Hence, the SOA derived from this tracer is generally regarded to include the SOA formation from both toluene and xylenes. The tracer-based measurements in HK were made at four locations. The SOA concentrations and contributions were found to show little differences among the four sites [Hu et al., 2008]. This lack of spatial variation has been observed for other secondary PM components (e.g., sulfate) in HK [Louie et al., 2005]. We therefore only show the average values of the four sites in Figure 4. The measured SOAtx, SOAi, and SOAm observed on days dominated by local emissions in HK are reasonably simulated by the model under Base Case I conditions (i.e., lower emissions and lower organic loading at M0 = 5 µg m−3). Simulations under Base Case II (i.e., higher pollutant emissions and higher organic loading at M0 = 20 µg m−3) show good agreement with the measured SOA amounts observed on regional days in both SOAtx+i+m and the contributions of individual precursors.

Figure 4.

Comparison of simulated SOA contribution and the tracer-based measurements in HK and GZ.

[34] Two simulation results in HK are worth noting. (1) In the Base Case I scenario, isoprene is the most significant contributor among the three precursors. When the isoprene emission is cut by half (ISOP × 0.5 scenario in Figure 4), the simulation agrees better with the measurements and the contributions of isoprene and toluene + xylene becomes comparable. (2) In Base Case II, the model predicts a dramatic increase in monoterpenes’ contribution (1.78 vs. 0.19 µg m−3 in Base Case I), primarily as a result of monoterpene emissions increase being 3 times that in Base Case I. If monoterpene emissions are doubled while keeping other conditions the same as Base Case II (i.e., TERP × 2 scenario in Figure 5), the modeled SOAm is very close to the tracer-based measurement. In the TERP × 2 scenario, the modeled monoterpene concentration is 0.029 to 0.085 ppbv, still quite reasonable compared with the ambient concentration of monoterpenes.

Figure 5.

Simulated time series of removal rate for SOA precursors (a). Simulated production rate for CGs (b) and for individual lumped SOA components (c) under the Base Case II scenario. All the rates are accumulated in each channel. Effective saturation concentrations for CGs are derived from CAMx User's Guide [ENVIRON, 2010].

[35] The modeling results suggest that monoterpenes contribute more to SOA than toluene + xylenes and isoprene in HK, despite the lower abundances of monoterpenes. This could be explained by tracking the SOA formation processes in the current SOA formation model. Figure 5 shows the time series of simulated removal rate of each SOA precursor (top panel), the production rates of condensable gases (CGs) and SOA components (middle and bottom panel) in Base Case II scenario. The removal rate for isoprene is the fastest, followed by toluene and xylenes, a result of the fact that the OH reaction rate of isoprene is nearly 20 times faster than that of toluene and 10 times faster than other aromatics. The removal rate of monoterpenes is rather small due to its low ambient concentration. However, unlike isoprene and aromatics, monoterpenes have a fast ozone oxidation pathway. The ozone oxidation pathway accounts for nearly half the total removal of monoterpenes, especially in the late afternoon when OH begins to decrease while ozone is at a rather high level. For the other VOCs, the ozone oxidation pathway only makes minor contributions to SOA.

[36] The production rate for each CG is determined by both VOC removal rate and the formation yields. The middle panel of Figure 5 shows the production rates for individual CGs from their precursors. Although the removal rate of monoterpenes is significantly smaller than those of isoprene and aromatics, the production rate of monoterpenes-derived CGs is noticeably more significant. This can be attributed to the fact that both OH and ozone oxidations of monoterpenes have much higher yields for CGs (i.e., CG5 and CG6 in the reaction equations in Figure 5a). The two CGs from monoterpenes have much lower effective saturation concentrations (see the table within Figure 5), that is, less volatile. When partitioning of CGs into aerosol is considered, the significance of monoterpenes to SOA formation becomes apparent. As shown in Figure 5c, the production rates for monoterpenes-derived SOA are comparable with isoprene and aromatics in the morning and early afternoon and exceed those from isoprene and aromatics in the late afternoon. In the end, the overall SOA production rate from monoterpenes is comparable with isoprene and toluene + xylenes, although the monoterpene emissions in Base Case II are only approximately 10% of isoprene emission and approximately 7% of the combined emissions of toluene and xylenes.

[37] Monoterpenes also undergo reactions with NO3 radical. This pathway is assumed in CB05 to form CGs at the same yields as the OH and ozone oxidation pathways. This channel is not shown in Figure 5, because it is much smaller than the other channels in both daytime and nighttime (when NO3 levels are higher).

3.4.2 Guangzhou

[38] The simulated SOA formation for the GZ Base Case (M0 = 20.0 µg m−3) is compared with the tracer-based SOA measurements at Wan Qing Sha in Figure 4. Wan Qing Sha is approximately 50 km downwind of GZ, downtown to the south and approximately 25 km from the closest suburban district of GZ (Panyu). The timescale of transport from urban GZ to Wan Qing Sha is estimated to be a few hours or less, consistent with the timescale of fresh SOA formation. Hence, the model-predicted SOA formation could be compared with SOA observed at Wan Qing Sha. On days influenced by transport from GZ, the measured total SOA at Wan Qing Sha from toluene/xylenes, isoprene, and monoterpenes is 8.38 ± 3.19 µg m−3, with the individual contributions at 6.08 ± 2.57 µg m−3 for SOAtx, 2.21 ± 2.00 µg m−3 for SOAi, and 0.10 ± 0.07 µg m−3 for SOAm. The simulated SOAtx+i+m in the GZ Base Case scenario is 7.64 µg m−3, and the contributions of the above precursors are 3.88, 1.59, and 2.17 µg m−3, respectively. The GZ Base Case predicts SOAi in reasonable agreement with the measured amounts, but underpredicts SOAtx and overpredicts SOAm. If the toluene emissions increased by a factor of 1.5 and the other conditions remain the same as the GZ Base Case (i.e., TOL ×1.5 scenario in Figure 4), the simulated SOAtx+i+m becomes 9.30 µg m−3, and toluene + xylene-derived, isoprene-derived, and monoterpene-derived SOA are 5.52, 1.61, and 2.17 µg m−3, respectively. The simulated air concentrations of aromatics are 14.1 to 26.5 ppbv (toluene) and 3.2 to 5.4 ppbv (xylenes) in the TOL ×1.5 scenario, still within the range of observed ambient concentrations.

[39] The simulated SOAtx and SOAi formation generally agrees with the tracer-based measurements, but the model-predicted SOAm (2.17 µg m−3) is significantly higher than the measurements (~0.1 µg m−3). The measured SOAm at Wan Qing Sha was considerably lower than that measured in HK (0.25 µg m−3 on local days, 3.51 µg m−3 on regional days). We note that the ambient levels of monoterpenes in GZ measured in the months from September to November in past studies were more than twice that measured in an urban location in HK in the months from June to November (Table 1) [Barletta et al., 2008; Liu et al., 2008c]. The air concentrations of monoterpenes simulated by the model are 0.015 to 0.05 ppbv for HK and 0.03 to 0.14 ppbv for GZ, respectively, falling in the range of the observed ambient concentrations (Figure 2). Therefore, we believe that such a discrepancy between the modeled and tracer-based measured SOAm in GZ and the large difference in SOAm between HK and GZ is less likely due to emissions being grossly overpredicted in GZ. A more likely cause is the difference in analytical techniques for the monoterpene SOA tracers, especially the selections of tracers and calibration surrogates in the two SOA trace studies. Tracers of monoterpenes used in the Wan Qing Sha measurements include cis-pinonic acid, pinic acid, 3-methyl-1,2,3-butanetricaric acid, 3-hydroxyglutaric acid, and 3-hydroxy-4,4-dimethylglutaric acid, and they were analyzed using GC coupled with quadruple MS detector (Agilent). Due to a lack of calibration standards for these tracers, pinic acid was used as the surrogate for quantifying the monoterpene tracers [Ding et al., 2012]. In the study by Hu et al., tracers for monoterpenes include 3-methyl-1,2,3-butanetricaric acid, 3-hydroxyglutaric acid, 3-hydroxy-4,4-dimethylglutaric acid, 3-isopropylpentanedioic acid, 3-acetyl pentanedioic acid, and 3-acetyl hexanedioic acid; samples were analyzed using GC coupled with ion-trap MS detector (Varian) and two surrogates, citramalic acid and pimelic acid, were used for the quantification of these tracers. The discrepancy in the measured SOAm highlights the importance of conducting intercomparisons of analytical methods for the SOA tracers.

3.5 Sensitivity Tests of SOA Formation

[40] The previous discussion shows that SOA formation in this region is highly dependent on the emissions of their VOC precursors. VOC emissions have considerable temporal and spatial differences and their emission inventories have large uncertainties, especially for BVOCs (Table 2). In addition, NOx, total NMHC, and organic aerosol loading can significantly influence the amount of SOA formed.

[41] To assess the multiple factors that influence SOA formation, we have carried out a comprehensive series of sensitivity tests in which SOAtx+i+m is simulated as a function of six factors. The following factors include (1) emission factor of NOx, varying from 0.7× to 1.3× Base Case; (2) emission factor of NMHCs (other than isoprene and monoterpenes), from 0.7× to 1.3× Base Case, isoprene and monoterpenes from 0.7× to 2.0× Base Case; (3) emission factor of isoprene, from 0.5× to 2.6× Base Case; (4) emission factor of monoterpenes, from 0.5× to 2.6× Base Case; (5) emission factor of aromatics (toluene + xylenes), from 0.5× to 1.7× Base Case; and (6) level of organic loadings at 5.0, 10.0, and 20.0 µg/m3. The sensitivity tests of varying the above six factors were carried out under conditions of both HK Base Case I and Base Case II and for the GZ Base Case. The statistics compiled by the Hong Kong Environmental Protection Department show that the maximum differences in pollutant emissions from 1990 to 2006 were 58% in total NOx emissions and 40% in total VOC emissions [HKEPD, 2009]. Therefore, it is reasonable to consider that the range of variations in these factors likely cover the majority of atmospheric conditions encountered in the region.

[42] Ternary distribution plots of SOA contributions from toluene + xylenes, isoprene, and monoterpenes are constructed to show their relative importance in HK (Figure 6a) and in GZ (Figure 6b). Simulation results are enclosed in the shaded areas. In HK, among the target VOC precursors, toluene + xylene contribution was in the range of 15% to 43%, isoprene contribution was 14% to 48%, and monoterpene contribution was 21% to 70%. Most tracer-based measurements, on both local days and regional days, fell within the model-simulated ranges (Figure 6). In GZ, the emissions-based box model indicate that toluene + xylene are a more important SOA contributor, accounting for 33% to 76% of SOAtx+i+m, whereas isoprene and monoterpene contribute 13% to 44% and 10% to 45%, respectively, to SOAtx+i+m. We note that most of the tracer-based measurements in Wan Qing Sha fall outside the model-predicted area in the ternary distribution plot due to the extremely low SOAm derived from the tracer-based measurements in Wan Qing Sha. Despite the discrepancy between the modeled and the tracer-derived SOAm, it is clear that toluene and xylenes play an important role in SOA formation in the GZ area, which is different from the situation in HK.

Figure 6.

Ternary distributions of percentage contributions of aromatics, isoprene and monoterpenes to the total SOA simulated under different emission conditions in (a) HK and (b) GZ. Simulation settings cover the following ranges of different factors: NOx emission factor (0.7–1.3) × Base Case; NMHC emission factor (0.7–1.3) × Base Case; aromatics (toluene + xylenes) emission factor (0.5–1.7) × Base Case; isoprene emission factor (0.5–2.5) × Base Case; and monoterpenes emission factor (0.5–2.5) × Base Case. The tracer-based SOA measurements in HK are from Hu et al. [2008] and those in GZ are from Ding et al. [2012]. The measurement data are color-coded by total SOA concentration.

[43] In principle, such ternary distributions are able to cover the effects on SOA formation from (1) emissions of SOA precursors, and (2) gas-phase chemistry and SOA formation processes. For other locations that have different emission profiles of VOC precursors, the ternary distribution patterns are expected to be different from those shown in Figure 6.

3.6 Limitations and Implications of This Work

[44] In the present work, an emission-based 0-D box model is used to simulate the SOA formation from isoprene, monoterpene, and toluene/xylenes for emission scenarios typical in HK and GZ. Limitations and implications of this work mainly include:

  1. Emission inputs. Similar to the model used in this work, many 3-D chemical transport models are emission based. The quality of emission inputs is essential to the overall model performance and heavily relies on other databases, such as emission inventories, source profiles, land use type, etc. Large uncertainties may exist in these emission-related data, which may propagate into the final model results. For example, Leung et al. [2010] estimated the BVOC emissions in HK using the latest land cover and emission factor databases and reported the isoprene/monoterpenes emission ratio to be approximately 17 (annual). This ratio is more than 10 times higher than the result reported 1 year ago [Tsui et al., 2009]. In addition, Lo et al. [2006] demonstrated the substantial importance of land use information in accurately modeling the spatial pattern of emissions as well as the meteorological fields in the PRD region. The most updated land use information for the PRD region is the 2003 version. As a rapidly developing region, the land use types in the PRD region is expected to have been altered during recent years. Emission inputs should be further evaluated against relevant data sets, such as ambient measurements, remote sensing data, to gauge their accuracy. In this work, the emission fluxes have been carefully examined against available ambient observations. As discussed previously, the current VOC measurement techniques are not optimized for low-concentration and reactive species such as monoterpenes. Hence, emission inputs are expected to be a major source of uncertainties for SOAm in this modeling work.
  2. Gas-phase mechanism. Our current understanding about gas-phase chemistry still has large gaps. For example, the overestimate in the RO2 and HO2 levels during nighttime by models reflects such knowledge deficiencies [Geyer et al., 2003; Mihelcic et al., 2003]. Condensed mechanisms are used in this work and in 3-D models. It is known that the condensed mechanisms have their inherent limitations, e.g., the oversimplification of oxidation mechanisms beyond the first-generation reactions and the chemistry of oxygenated VOCs (such as formaldehyde and glyoxal). For example, the fast formation of carbonyls from the oxidation of aromatics is not included in the condensed mechanisms [Volkamer et al., 2001; Jenkin et al., 2003].
  3. SOA formation mechanism and other SOA precursors. SOA formation in many chemical transport models adopts semiempirical approaches. Although the thermodynamic partitioning of low-/semi-volatile organic compounds produced from the oxidation of main VOC precursors (e.g., aromatics, isoprene, monoterpenes, and sesquiterpenes) are incorporated into the model, many other processes that may potentially lead to SOA formation are not included, such as the aging of organic aerosols, multiphase chemistry. As discussed previously, in this study, we focus on evaluating the relative importance of toluene + xylenes (anthropogenic) versus isoprene and monoterpenes (BVOCs) in this region. Chamber studies have demonstrated that oxidation of the VOC precursors considered in this work lead to glyoxal and methylglyoxal [Yu et al., 1995; Yu et al., 1997; Yu et al., 1999]. The potential importance of SOA formation from these water-soluble dicarbonyls via aqueous particle processes has been recognized [Volkamer et al., 2007; Ervens and Volkamer, 2010; Ervens et al., 2011]. Unfortunately, in CB05, the chemistry of glyoxal and methylglyoxal is not treated explicitly. Hence, SOA from the VOCs considered here could be underestimated, as the aqueous-phase processes of these water-soluble compounds may form SOA. We also note that the oxidation of S/IVOCs from primary combustion emissions, a group of VOCs of anthropogenic origin, could lead to the formation of significant SOA as well [e.g., Robinson et al., 2007]. This source of SOA has not been considered in the current study. Future studies incorporating the primary emitted gas-phase S/IVOC in the model are needed to assess the overall SOA formation from major anthropogenic precursors and the relative contributions of anthropogenic and biogenic SOA for this region.
  4. Effects of regional transport. An inherent weakness of the one-box model is its incapability to simulate regional transport. In this work, we considered regional transport effects on SOA in HK in a very crude manner by scaling-up the raw emissions from MEGAN or SMOKE by certain factors. Transport input to GZ was not considered mainly because GZ was determined to be a major source area influencing regional concentrations of PM2.5 and its organic component [Hagler et al., 2006]. Three-dimensional modeling studies are needed to realistically simulate the regional formation and transport of VOC/NOx and SOA. Because of a lack of realistic representation of transport processes in the box model, we note that a comparison of SOA amounts between the model results and the tracer-based measurements should be viewed with caution and having in mind uncertainties associated with ignoring the transport factor.

4 Conclusions

[45] We have developed an emission-based 0-D box model that incorporates CB05 chemical mechanism and gas–particle partition of condensable VOC oxidation products to simulate SOA formation from isoprene, monoterpenes, and toluene/xylenes in HK and in GZ in the PRD region, China. The emission inputs for major inorganic air pollutants and common VOCs are generated using SMOKE (anthropogenic species) and MEGAN (biogenic species) for this region. In Base Case scenarios, toluene + xylene emission fluxes in GZ are approximately double those in HK whereas emission fluxes of isoprene and monoterpenes are similar. The model-simulated VOCs fall within the range of ambient observations, demonstrating a reasonable representation of the emissions and oxidation of VOCs. In HK, monoterpenes are modeled to have more significant SOA formation than toluene/xylenes do. On the other hand, in GZ, the model indicates that toluene + xylenes are a more important SOA contributor than isoprene and monoterpenes. This considerable difference in their relative contributions to SOA formation between HK and GZ was observed in two recent SOA tracer-based measurement studies and correctly captured by the model. As such, it can be inferred that the higher emission fluxes of toluene and xylene could largely explain the dominant contributions of the group of AVOCs to SOA over the group of BVOCs in GZ. The model predicts that the reduction in AVOCs could lead to a reduction in both SOA from AVOCs and SOA from BVOCs, as SOA formation is generally in the VOC-limited regime under the high NOx conditions encountered in this region.

[46] The model-predicted SOA formation from toluene + xylenes and isoprene agrees well with estimates derived from ambient SOA tracer measurements in HK and at Wan Qing Sha, a site downwind of GZ. However, the agreement was poor for the monoterpenes-derived SOA, with the measured SOAm substantially lower than the model prediction. Our analysis identifies that uncertainty in analytical techniques for the SOA tracers is suspected to be a major factor responsible for the discrepancy. This modeling study suggests that more and better measurements of monoterpenes and their SOA tracers are needed to better assess the contributions of BVOCs to SOA in this region.


[48] This work is supported by the Research Grants Council of Hong Kong (622409) and Environment and Conservation Fund/Woo Wheelock Green Fund (ECWW09EG04).