An Aerodyne high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS) has been employed in a field sampling campaign at a suburban coastal site in Hong Kong in springtime 2011 to provide insights into the size-resolved chemical composition of nonrefractory submicron aerosol species. This is the first time that such detailed real-time measurements have been made in Hong Kong. The total nonrefractory PM1 was dominated by sulfate (51.0%) and organics (28.2%) with considerable acidity (average in situ pH = 0.95) and a characteristic bimodal particle size distribution with peaks at 200 and 570 nm of vacuum aerodynamic diameter (Dva). Source apportionment of organic aerosol yielded three characteristic aerosol fractions (hydrocarbon-like organic aerosol, semivolatile organic aerosol and low-volatile organic aerosol) with distinct temporal patterns and distributions in different particle size regions. The influence of air mass origin on species concentrations, particle size distributions and elemental ratios was investigated using backtrajectory analysis. Larger particle diameters, greater fractions of oxygenated organic aerosol and higher organic-to-carbon ratios were observed during coastal and continental air mass influence. Three major pollution events with elevated nonrefractory PM1 concentrations were observed in the sampling period, which were related to distinct meteorological and circulatory conditions. Accumulation and redistribution of local and regional pollutants were notable in a period of strong land-sea breeze over the Pearl River Delta region, with considerable photochemical activity and particle aging. Increased fractions of oxygenated organic aerosol were apparent in foggy conditions, illustrating the importance of aqueous phase oxidation processes in a cooler and more humid time period.
 Filter sampling and off-line analysis of ambient particulate matter are associated with various disadvantages including a low time resolution of 24 to 48 h per sample and potential influence of various artifacts, such as particle loss, reentrainment, and semivolatile vapor adsorption and desorption [Dillner et al., 2009; Stein et al., 1994; Turpin et al., 1994]. Real-time instruments, with temporal resolutions in the order of a few minutes or less per measurement, can reduce artifacts from changing environmental conditions and enable the capture of shorter-time scale species trends, such as diurnal variations and finer dynamics of pollution events. Over the past decade, aerosol mass spectrometry has gained acceptance as a fast, sensitive, versatile aerosol analysis method providing size-resolved compositional analysis of major organic and inorganic submicron particulate matter components in real time [Canagaratna et al., 2007; Drewnick et al., 2005]. Different versions of the Aerodyne aerosol mass spectrometer (AMS) have been employed in various locations in North America, Europe, and Asia in recent years [Aiken et al., 2010; Hagino et al., 2007; Hildebrandt et al., 2011; Lanz et al., 2007; Sun et al., 2010, 2011; Zhang et al., 2005]. This is the first time that an Aerodyne high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS) was employed for studying ambient aerosol in Hong Kong, and herewith presented are the findings of a month-long intensive campaign conducted at the recently established on-campus air quality supersite at the Hong Kong University of Science and Technology (HKUST) in April and May 2011.
 The Hong Kong Special Administrative Region (HKSAR), located in the southeast of the Pearl River Delta (PRD), is one of China's major manufacturing centers and one of the world's most densely urbanized areas with a population of more than 50 million people. Hong Kong is bordered by Shenzhen to the north but faces the South China Sea to the south, east, and west and encompasses a territory of about 1100 km2. The rapid urbanization and industrial development in the PRD have led to aggravating air pollution problems throughout the region. In Hong Kong, the number of low-visibility (<8 km) days has continuously increased since the 1980s and has seen a dramatic rise since the early 2000s [Chan and Yao, 2008]. Particulate matter has been recognized as a major factor in Hong Kong's urban air quality, affecting visibility [Zhuang et al., 1999] and human health [Thach et al., 2011]. Studies on particulate matter in Hong Kong have been traditionally based on filter sampling and off-line analysis including elemental and organic carbon (ECOC) analysis, gas chromatography-mass spectrometry, ion chromatography, gravimetric analysis, and absorptive colorimetry [Ho et al., 2003; Li and Yu, 2010; Yao et al., 2007]. Studies on the temporal and spatial variations of PM2.5 (particulate matter smaller than 2.5 micrometer in diameter, also referred to as fine suspended particles, FSP) and PM10 (particulate matter smaller than 10 micrometer in diameter, also referred to as respirable suspended particles, RSP) components [Ho et al., 2003; Ho et al., 2006; Louie et al., 2005; Yu et al., 2004] found that carbonaceous aerosol in the form of both organic carbon (OC) and elemental carbon (EC) contributed over 50% of FSP at urban and roadside sites. Diesel and gasoline exhaust, meat cooking, and biomass burning were identified as major sources of fine carbonaceous aerosol with fine mode OC mostly dominated by diesel and gasoline exhaust-related components [Hu et al., 2010; Li et al., 2003]. Generally, OC was observed to be higher in winter due to increased fuel consumption and biomass burning [Zheng et al., 2006]. Among the inorganic constituents, sulfate has been largely characterized as a regional pollutant originating at least in part from sources outside of Hong Kong, while fine particle nitrate is mainly associated with secondary formation from vehicle NOx emissions. Studies on the size distribution of ambient aerosol species in Hong Kong are mainly based on size-resolved sampling by cascade impactors or serial collocated samplers with different inlet cutoff size diameters. Sulfate and ammonium were observed as the dominant species in fine particles (Dp < 1.8 µm), while nitrate was mainly present in the coarse mode [Zhuang et al., 1999]. Solvent extractable organics were shown to be enriched in PM2.5 and to exhibit a seasonal dependency with higher concentrations and a shift toward larger particle size in the dry winter season [Zheng et al., 2008].
2.1 Sampling Site Description
 The sampling site is located on the campus of HKUST (22°20′N, 114°16′E) in the suburban area of Clear Water Bay at the east coast of Hong Kong. As a part of the HKUST supersite project, the AMS is stationed in a modular rooftop house sitting on a seawater pumping facility facing Port Shelter. There is an on-campus road leading to the pump house area which is however only sporadically used. The next major roads in the vicinity—Clear Water Bay Road and Hiram's Highway—are located just outside the university campus at about 750 m lateral and 75 m vertical distance uphill from the sampling site, blocked by hillside shrubs and trees as well as university buildings and student halls. A yacht club is located ≈2.5 km north of the campus in Hebe Haven with yachts, small boats, as well as fishing vessels regularly passing along the shoreline bordering one side of the university campus at a lateral distance of about 500 m. Satellite images depicting the location of the campus and its surroundings are appended in the supporting information (Figures S1 and S2). The HR-ToF-AMS was stationed in the modular sampling house together with various other instruments (for details, see http://www.envr.ust.hk/supersite/). Ambient air is sampled at 16.67 L/min through a PM2.5 cyclone at the roof that removes coarse particles into a stainless steel sampling port supplying the AMS as well as several other instruments, including a hygroscopic tandem differential mobility analyzer, a cloud condensation nuclei counter, and a fast mobility particle sizer with ambient air. The portion of sampled air designated for the AMS passes through a 1 m long diffusion drier (BMI, San Francisco, California) filled with silica gel to remove bulk gas and particle phase water and is subsequently sampled by the AMS at a flow rate of ≈80 cm3/min. Ambient sampling was conducted from 26 April 2011 throughout the month of May until the late morning of 1 June 2011. Other data presented in this work were obtained from collocated instruments which included a thermo-optical ECOC analyzer (Sunset Laboratory Inc., Tigard, Oregon), a gas analyzer system (Teledyne Instruments, City of Industry, California), and a monitor for aerosols and gasses in ambient air (MARGA) (Metrohm Applikon, Schiedam, Netherlands). Meteorological data were accessed at the Atmospheric and Environmental Real-Time Database of the Institute for the Environment of HKUST (http://envf.ust.hk/dataview/gts/current/).
2.2 HR-ToF-AMS Operation and Data Analysis
 The HR-ToF-AMS was alternated between combined V mode and PToF data acquisition, and sole W mode data acquisition with 300 s sampling time in each mode and 7 s pause between modes to allow adjustment of the ToF-MS voltages. Routine flow rate calibrations with a Gilian gilibrator (Sensidyne, Clearwater, FL) and PToF size calibrations with NanosphereTM polystyrene latex particles (Duke Scientific, Palo Alto, California) in the range of 80 to 800 nm were conducted before and after the sampling campaign. Ionization efficiency calibrations with pure ammonium nitrate particles were carried out on a weekly basis, while particle-free filtered ambient air using an in-line high-efficiency particulate air filter was sampled daily for 30–60 min to obtain background particle size distributions and mass spectra. The collected nonrefractory PM1 mass concentrations and size distribution data were treated in accordance with the general principles laid out in previous studies [DeCarlo et al., 2006; Jimenez et al., 2003], employing the standard WaveMetrics Igor Pro-based data analysis software SQUIRREL (Version 1.51H) and PIKA (Version 1.10H) available from the TOF-AMS Resource Web page (http://cires.colorado.edu/jimenez-group/ToFAMSResources/ToFSoftware/index.html) and using the default relative ionization efficiency (RIE) values of 1.2 for sulfate, 1.1 for nitrate, 1.3 for chloride, and 1.4 for organics [Canagaratna et al., 2007]. The RIE for ammonium was chosen as 4.0 based on the mean of RIE values obtained from the regular ionization efficiency (IE) calibrations during the sampling period. To account for drifts in the performance of the microchannel plate detector, a correction using the ratio of ionization efficiency (IE) over air beam was applied using the logged air beam data and ionization efficiency parameters from the weekly performed calibrations. To account for particle loss due to lens transmission, particle bounce, and beam broadening, a collection efficiency (CE) factor is commonly applied to AMS-derived mass concentrations. The CE for this work (CE = 0.5) is based on findings from previous studies [Middlebrook et al., 2012] and will be discussed subsequently in a later section. For a more detailed characterization of the organic fraction of nonrefractory aerosol, positive matrix factorization (PMF) was employed to resolve organic mass spectra into characteristic source profiles and their respective time series. PMF analysis was conducted on both V mode-derived unit mass resolution (UMR) and high-resolution W mode spectra [Ulbrich et al., 2009], with the input concentration and error matrices generated from SQUIRREL and PIKA, respectively.
 The AMS measurements spanned the period from 26 April to 1 June 2011, covering the whole month of May 2011 which was relatively dry and sunny with a mean relative humidity of 81% and an average temperature of 26.0°C. Cumulative rainfall at 186.7 mm was only about half of the long-term average for the month (http://www.hko.gov.hk/wxinfo/pastwx/mws.htm). In mid-May, the passing of a low pressure trough brought rain and thunderstorms with significant wet removal of aerosol, while the end of May was dominated by a continuous cool northerly airstream, leading to a significant accumulation of air pollutants and a major episodic pollution event in Hong Kong.
3 Results and Discussion
3.1 Overall Composition
 A synopsis of the most important measured and statistically resolved species as well as meteorological parameters is presented in Figure 1. During the campaign, total nonrefractory PM1 (NR-PM1) as measured by the AMS ranged from 0.8 to 72.4 µg/m3 with an average of 14.5 ± 9.7 µg/m3, which is within the upper range of measured concentrations in various other urban environments in the U.S. and Europe [Zhang et al., 2007b] but is significantly lower than what has been observed with AMS in other Chinese cities, e.g., Beijing, Shenzhen, and Kaiping [He et al., 2011; Huang et al., 2010, 2011b]. On average, sulfate and organics combined amounted to 79.2% of nonrefractory fine particle mass (Figure 2a) with average concentrations of 7.9 ± 4.8 and 4.4 ± 3.6 µg/m3 respectively. Nitrate on average contributed 4.1% of NR-PM1, ammonium 16.4%, and chloride 0.3%. This differs significantly from what has been reported at a rural background site in the PRD [Huang et al., 2011b], approximately 130 km to the west of the HKUST supersite, where both organic and sulfate concentrations were notably higher (both 11.2 µg/m3). Also, the fractional species distribution observed at the rural PRD site was markedly different, with much greater fractions of nitrate (10.7%) and organics (33.8%) but lower sulfate (33.8%), as opposed to the HKUST sampling site where the opposite (51.0% of sulfate and 28.2% of organics; Figure 2b) is found in this study. The difference in observed nitrate fraction may be due to the fact that sampling at HKUST and Kaiping took place in different seasons, with the latter measurements conducted in fall where temperatures are lower and more nitrate is expected to be found in the particle phase. The dominance of secondary sulfate as observed at the suburban HKUST site has been more commonly observed in rural and remote sampling sites [Zhang et al., 2007b] in other parts of the world but not as much in areas with significant influence of urban pollution.
3.2 Size Distribution
 The monthly averaged species size distribution (Figure 2c) exhibits an accumulation size mode diameter of 569 nm (vacuum aerodynamic diameter (Dva)), which roughly corresponds to a geometric diameter of 375 nm (see supporting information). The size mode is common to all measured nonrefractory (NR) species, suggesting internally mixed particles. As frequently observed in other studies, the organic fraction exhibits a clear shoulder into the smaller size regimen extending below 100 nm which is likely due to fresher organic aerosol of primary origin. This shoulder is also visible for all other NR species with generally bimodal distributions that can be deconvolved using lognormal peak fitting into a larger accumulation mode diameter centered at ≈570 nm (Dva) and a smaller-sized accumulation mode peak centered at ≈200 nm (Dva) (Figure 2d). Only 2% of sulfate was found in the smaller size mode. In contrast, 11% and 12% of organics and nitrate were present in condensation mode particles, respectively. This indicates that fresher particles were enriched in organics and nitrate. With the influence of a northwesterly continental air mass in late May, the smaller size mode became more distinct with the smaller and larger particle peaks in the organic fraction of NR-PM1 having almost equal magnitudes and with the mode peaks shifting to slightly smaller sizes centered at ≈150 and ≈550 nm. This indicates that particles were less aged and demonstrates the importance of fresh particle formation. When Hong Kong was under the influence of oceanic air mass, the accumulation size mode peak was less intense and shifted to smaller sizes, indicating a stronger influence of local pollution and a reduced impact of more aged and possibly long-range transported particulate matter. A more detailed analysis will be presented subsequently in the cluster analysis section (section 3.8).
3.3 Comparison of AMS, MARGA, and ECOC Analyzer Measurements
 The Environmental Protection Department of the HKSAR Government operated an Applikon MARGA instrument at the HKUST supersite, providing hourly concentrations of various inorganic anions and cations as well as related gas phase components. MARGA uses a wet rotating denuder to absorb gases from the sampled air and a steam jet aerosol collector to retrieve particulate matter. Two ion chromatography systems are used for species separation and yield analytical data for both cations and anions [Trebs et al., 2004]. The size cut for the AMS is roughly 1 µm due to the limitations of the AMS lens system, while MARGA has a size cut of 2.5 µm. Generally, MARGA and AMS observed similar variations of species concentrations over time (Figure 3), with overall R2 values of 0.80 for NH4+, 0.79 for NO3−, and 0.88 for SO42−, where hourly averaged AMS data were used to match the lower time resolution of MARGA. In terms of nominal mass concentrations, there is significant disagreement which varies with species. On average, the AMS measured only about 60% of SO42−, 50% of NH4+, and only 33% of NO3− as compared to MARGA. The 24 h averaged SO42− concentrations measured by MARGA agree well with filter-based measurements with a slope of 0.99 and R2 of 0.70 [Huang et al., 2013]. Hence, MARGA denuder efficiency is not an issue. It is, however, likely that the difference in size cut contributed to the difference between AMS and MARGA measurements. Furthermore, previous filter-based size-segregated measurements conducted at HKUST using micro-orifice uniform deposit cascade impactors have shown that nitrate at HKUST was mainly distributed in the coarse mode, with a mode diameter centered at 3.95 µm [Zhuang et al., 1999]. With a size cut of 2.5 µm, it is likely that MARGA detects larger fractions of coarse mode nitrate than the AMS. The tails of the AMS size distributions for SO42−, NO3−, and organics (Figure 2e) which were scaled relative to the peak of each species clearly show an enhanced trailing edge for nitrate. It is likely that this signifies the presence of coarse mode nitrate, but it cannot be completely ruled out that the tail may also be an artifact from delayed nitrate vaporization. Furthermore, as the lens transmission of the AMS decreases rapidly beyond a particle size of ≈1.5 µm (Dva), no reliable quantitative estimations can be made in this size range. Another factor is that a CE of 0.5 is adopted in the calculation of the AMS species mass concentrations. Sulfate is the overall dominating species throughout the sampling period, and nitrate mass contributions are small, at most 20%. It is, thus, likely that the actual CE is in the range of 0.3–0.4, as has been observed for dry ammonium sulfate-dominated particles which display a strong particle bounce effect [Matthew et al., 2008]. A CE of 0.3 (Figure S3a, supporting information) would yield a better agreement between MARGA and AMS in terms of measured particulate SO42−. For NO3−, a considerable difference would remain, with MARGA concentrations higher than those measured by the AMS due to the previously mentioned coarse mode nitrate. A minor discrepancy between AMS- and MARGA-derived NH4+ would also remain as some ammonium is expected to associate with nitrate. The sum of AMS- versus MARGA-measured inorganic species (i.e., ratios of the sum of NH4+, NO3−, and SO42−) also shows better agreement for the CE = 0.3 case (Figure S3b, supporting information), with an average ratio close to 1. Scattered data points with high ratios up to 2 mainly correspond to periods of low mass concentrations associated with rainfall events (wet scavenging) or high-wind conditions (dispersion).
 A Sunset thermo-optical carbon analyzer acquired hourly OC and EC concentrations (PM2.5 size cut) using an analysis protocol similar to the National Institute for Occupational Safety and Health protocol. Details on the instrument and the protocols are available elsewhere [Bae et al., 2004; Wu et al., 2011]. Using the average monthly ratio of organic matter to organic carbon (OM/OC) of 1.66 from the high-resolution data elemental analysis (see below), organic carbon concentrations were converted to organic matter concentrations and compared to the overall hourly average NR-PM1 organic concentrations. The agreement in terms of temporal trend (Figure 3) and measured value was good, with a slope of 0.88 and R2 of 0.87.
 Due to the different size cuts especially with respect to the influence from coarse mode nitrate and additional measurement uncertainties with the ECOC analyzer, e.g., substantial differences in EC and OC concentrations depending on the analysis protocol used [Wu et al., 2011], the AMS CE was purposely not based on the comparison with these two collocated instruments. Instead, the most commonly observed CE of 0.5 from previous AMS studies was chosen as recently summarized [Middlebrook et al., 2012]. Considering the interinstrument comparison just presented, with observed slopes consistently <1, the AMS-based mass concentrations reported in this study should be regarded as lower-bound estimates for NR-PM1 concentrations at the HKUST supersite.
3.4 Degree of Neutralization and Aerosol Acidity
 The ratio of measured ammonium to expected ammonium from the charge balance with sulfate, nitrate, and chloride is an indicator of the degree of neutralization of the sampled ambient aerosol [Pathak et al., 2004; Zhang et al., 2007a]. On average, ambient aerosol particles at the HKUST supersite appeared to be acidic with a slope of 0.77 and R2 of 0.99 (Figure 4a). As the color scale illustrates, the aerosol sampled at the HKUST supersite gradually became more ammonium deficient over the course of the intensive campaign, from a ratio close to 1 (neutralized sulfate) in late April to 0.75 at the beginning of June which implies equal amounts of sulfate and bisulfate in the aerosol [Zhang et al., 2007a]. This decrease in the ratio of measured to expected ammonium is related to decreasing fractional ammonium content over the month as shown in Figure 4b. As the ammonium ratio is merely an indicator of aerosol acidity, complementary in situ pH calculations (pHis) were performed using the following relation [Li et al., 2013]:
where ntot is the total number of moles of condensed phase species, fH+ is the molar fraction of protons, γH+ is the activity coefficient on mole fraction basis, and Vaq is the aqueous phase volume. The mentioned input parameters were generated by the Extended Aerosol Inorganic Model II (E-AIM II) [Clegg et al., 1998] with solid formation inhibited and using the inorganic mass concentrations as measured by the AMS. The calculated in situ pH did not change significantly over the sampling period, with a monthly average pH of 0.95 (Figure 4c). Lower pH was observed when Hong Kong was under the influence of cleaner oceanic air mass (13–14 May). A brief episode of pH values up to 4.5 was observed in late May, coinciding with a transition of air mass origin from coastal to north-north easterly continental. The lack of pH change as opposed to the decreasing trend in the ratio of measured to predicted ammonium indicates that this ratio was not a suitable indicator for aerosol acidity in this study. As apparent from the fractional aerosol composition (Figure 2b), significant fractions of nitrate in acidic ambient aerosol were found, which have also been observed in previous studies in Hong Kong and various other Chinese cities [Huang et al., 2011a; Pathak et al., 2004; Pathak et al., 2009, 2003].
3.5 Diurnal Patterns
 Diurnal patterns for all NR species in May are depicted in Figure 5a (25th and 75th percentile boxes, 5th and 95th percentile whiskers, median as line in box, and mean as solid colored line). The mean trend will be used for the following discussion. Sulfate showed only a very subtle increasing trend over the daytime hours, suggesting that despite long periods of sunny and relatively dry weather during the sampling period, photochemical production of SO42− was not significant. A comparison of hourly averaged sulfate and hourly averaged relative humidity (RH) does not indicate a significant correlation, suggesting that local aqueous phase oxidation of SO2 may not have been of major importance either. The observed SO42− is likely related to regional-scale production processes and subject to midrange and long-range transport (Figure 5b).
 Nitrate concentrations were subtly lower in the daytime, indicating that the combined effect of evaporation and more favorable dispersion due to the higher daytime planetary boundary layer was slightly stronger than the effect of the secondary photochemical production of nitrate. The subtle nighttime concentration increase was likely due to gas-to-particle partitioning as temperature decreased (Figure 5b). Organics showed a more complex diurnal pattern, with two peaks respectively in the morning and late afternoon accompanied by a more gradual concentration increase throughout the daylight hours. This is a combined result of the diurnal patterns of the individual organic aerosol groups as characterized by PMF which will be discussed in a corresponding later section. The diurnal pattern of ammonium follows closely that of sulfate, confirming its association largely with sulfate, since nitrate concentrations were generally low and nonrefractory chloride was negligible.
3.6 Elemental Analysis of Organic Aerosol
 The elemental composition of organic aerosol is determined from the high-resolution mass spectra, where unique ion species can be identified and from which elemental mass ratios as well as the OM/OC ratio are derived. The default calibration values (H/C = 0.910, O/C = 0.750, N/C = 0.960, and S/C = 1.000) based on earlier studies [Aiken et al., 2007, 2008] were used for the elemental analysis. The time series of observed elemental ratios are presented in Figure 6a, with mean values of 1.36 ± 0.11 for H/C, 0.40 ± 0.11 for O/C, and 0.012 ± 0.004 for N/C. The observed OM/OC ratio ranged from 1.26 to 2.08 with a mean value of 1.66 ± 0.14, which is higher than what has been typically assumed (OM/OC ≈ 1.4) for urban aerosol [Turpin and Lim, 2001] but agrees with other AMS-based reported values of ≈1.5 to ≈1.8 [Aiken et al., 2008; He et al., 2011; Huang et al., 2011b]. The observed OM/OC mean value from this work is significantly lower than the ratio of 2.1 ± 0.3 from a previous study [Chen and Yu, 2007] which employed a combination of thermal gravimetric and chemical analyses (ECOC analysis and ion chromatography) of filter samples. The diurnal variations of the elemental ratios are shown in Figure 6b. The H/C ratio exhibits two obvious peaks, respectively, in the morning at around 8 A.M. and in the late afternoon at 8 P.M. This reflects a rush hour pattern that is consistent with the pattern of intensified roadside traffic in Hong Kong, one of the most important sources of primary organic aerosol (POA) in the city. The O/C ratio shows an afternoon peak at around 3 P.M., consistent with high photochemical activity and production of secondary organic aerosol during daylight hours. As the OM/OC and O/C ratios are highly correlated (R2 = 0.99), the diurnal pattern of the OM/OC ratio closely follows that of the O/C ratio.
 The van Krevelen plot of H/C against O/C is used to examine possible aging mechanisms of organic aerosol. As apparent from Figure 6c, two separate trends were obtained. For the bulk of data from late April until 23 May, the observed slope is −1.01, close to observations from previous studies in Riverside (California), Central Amazon, and Mexico City [Heald et al., 2010]. As pointed out by Ng et al. , a slope of −1 suggests addition of a carboxylic group without fragmentation or simultaneous alcohol and carbonyl addition on different carbons as the dominant aging mechanism. A slope of −0.5 would imply carboxylic group addition combined with fragmentation. For the majority of the May data, the former process appears to be dominant. However, during the episodic event in the later part of May, the slope became notably more shallow (−0.85), indicating a change in aging mechanism with a shift toward greater influence of carboxylic acid addition together with fragmentation. It is also possible that different mixings of hydrocarbon-like organic aerosol (HOA) and oxygenated organic aerosol (OOA) may contribute to the observed shift. However, among all the time periods with high OOA contribution (HOA < 15%) to the overall organic aerosol, data in the late May episodic event follow a distinctively different trend as compared to the rest in the van Krevelen plot, and thus, a change in aging is regarded as the more likely explanation for the observed trend. A more detailed analysis on the degree of oxygenation of organic aerosol is presented in a separate study [Li et al., 2013].
3.7 Source Apportionment of Organic Aerosol
 Positive matrix factorization (PMF) is a commonly employed tool to deconvolve ensemble organic spectra from AMS data sets into characteristic organic aerosol (OA) components and their respective time series [Zhang et al., 2011]. PMF was performed on both unit mass resolution and high-resolution mass spectra with the number of factors ranging from 1 to 7. Based on the diagnostics results (see supporting information), an optimum final solution of three factors was chosen with FPeak = 0 and Seed = 0 for both the UMR and HR data sets. The qualitative analysis described in this section is based on the W mode-resolved factors due to the higher mass spectral resolution, while the quantitative analysis is based on V mode-resolved factors due to better signal intensity.
 The three identified factors (Figure 7a) include one hydrocarbon-like organic aerosol (HOA) factor and two oxygenated organic aerosol (OOA) factors, of which one is highly oxidized and resembles what has been termed low-volatility (LV-) OOA [Aiken et al., 2008] and the other is semivolatile (SV-) OOA containing signature ions of both primary organic aerosol (POA) and ions more typical of secondary organic aerosol (SOA). Despite bearing resemblance to HOA with a considerable fraction of saturated CxHy-type ions, it displays a significant abundance of more oxidized CxHyOz-type ions. The HOA factor mass spectrum is characterized by a dominance of saturated CxHy-type ions, most notably 43, 55, and 57, but also contains heavier mass fragments with m/z > 100. The elemental ratios of the three factors illustrate the resemblance between HOA and SV-OOA, which have an identical OM/OC ratio (1.356), a marginally different O/C ratio (0.153 for HOA and 0.159 for SV-OOA), but a more pronounced difference in H/C ratio (1.708 for HOA and 1.591 for SV-OOA). The LV-OOA factor is vastly dominated by CxHyOz ions and exhibits high OM/OC and O/C ratios of 2.195 and 0.804, respectively, as well as a low H/C ratio of 1.209. HOA and SV-OOA contribute 22.6% and 23.5% to total OA, while the remaining 53.9% is accounted for by LV-OOA.
 The diurnal patterns of the three organic aerosol components are depicted in Figure 7b. HOA exhibits two typical peaks, the first starting early morning at ≈6 A.M. and peaking at ≈9 A.M. and the second much broader one in the afternoon rising at ≈3 P.M. and peaking at ≈8 P.M. This reflects the distinct traffic pattern in Hong Kong, with the morning rush hour period between 8 A.M. and 10 A.M. and the evening rush hour period between 5 P.M. and 8 P.M. and corresponds to the diurnal trend observed in the H/C elemental ratio. The peak toward the end of the evening rush hour period reflects the fact that the university campus is located at some distance from the downtown area, with city outbound traffic passing by Clear Water Bay Road and Hiram's Highway at later stages of that rush hour period. The diurnal variations of HOA and NOx (measured by a Teledyne 200E gas analyzer; Figure 7d) agree well, substantiating motor vehicles as the primary emission source. This also explains the considerable fraction of NO3− in the smaller size mode mentioned earlier. LV-OOA has a clear noon-to-afternoon peak extending well into the night and coincides with the observed diurnal pattern of O3 (measured by a Teledyne 400E gas analyzer; Figure 7d), suggesting a relation to photochemical production processes, which prevail during daytime. The delayed drop in LV-OOA levels in the late hours after sunset is probably due to gas-to-particle repartitioning at lower nighttime temperatures. The diurnal pattern of SV-OOA does not correlate well with that of either of the inorganic or gas phase species but exhibits a two-peaked variation over daytime similar to HOA with peak positions shifted by 3–4 h relative to HOA at 1 P.M. and 11 P.M. In the late evening and nighttime, the diurnal patterns of SV-OOA and AMS-measured NO3− show similarity, supporting the semivolatile behavior of SV-OOA with gas-to-particle partitioning in the late hours after sunset. The relatively low degree of oxygenation and presence of ion fragments which are more typical of POA indicate that SV-OOA was still quite fresh, likely originating from other urban or industrial areas in Hong Kong, but had started to undergo oxidation during the transport to the campus site. Even though biomass burning is not very common in Hong Kong, it is a widespread agricultural practice in other parts of the PRD [Yuan et al., 2010]. Organic aerosol from biomass burning (BBOA) has also been shown to yield mass spectra that contain signatures from both POA and SOA [Aiken et al., 2009]. However, the resolved SV-OOA spectrum does not exhibit significant signals at the typical biomass burning marker m/z ratios, such as 60 or 73 [Alfarra et al., 2007], and hence, contributions from BBOA to SV-OOA do not seem very likely. Similarly, the possibility of the SV-OOA factor representing cooking-related organic aerosol has been evaluated. A small canteen located about 50 m away and uphill of the sampling site is the only source in the immediate vicinity but only opens in the evening hours. While this might explain the late evening diurnal pattern of the SV-OOA factor, it cannot explain the noontime peak. The persistence of notable amounts of SV-OOA and LV-OOA throughout a period of clean easterly air mass (from 5 September to 5 October) points out that considerable portions of the oxygenated organic aerosol are of local or near-regional origin. Despite its location in a suburban area, the HKUST supersite displays compositional features of downwind urban aerosol [Zhang et al., 2007b] with HOA accounting for a quarter of total organic aerosol on average (Figure 7b).
 As mentioned, PMF analysis was carried out on both V mode and W mode data which qualitatively agreed very well (Figure 7a) with similar spectral results. Plots of W mode-resolved organic aerosol contribution against V mode-resolved fractional organic aerosol contribution (Figure S4, supporting information) reveal good temporal agreement between V mode- and W mode-resolved organic aerosol factors with R2 > 0.85 for the oxygenated fractions (SV-OOA and LV-OOA) and a slightly lower R2 = 0.79 for HOA. The factor contribution varies more for HOA and SV-OOA with a wider spread of data points than for LV-OOA. This may be due to the fact that concentrations for HOA and SV-OOA were generally lower, leading to higher concentration uncertainty in the high-resolution W mode where signal fluctuations are higher and the signal-to-noise ratio is lower. Based on the UMR V mode and PToF data, an approximation of the size distributions of the three PMF-resolved organic aerosol factors can be made by examining the size distributions of the organic ion fragments at m/z 44 (CO2+ mostly associated with oxidized LV-OOA), m/z 43 (C2H3O+ from SV-OOA and C3H7+ from HOA), and m/z 57 (C4H9+ mostly associated with HOA) (Figure 7e). For all ion fragments, the bimodal distribution with peaks at ≈200 and ≈570 nm as previously described was observed. The organic fraction of m/z 44 is almost exclusively confined to the larger accumulation mode, whereas the organic fraction of m/z 57 has considerable contributions in both smaller and larger accumulation size mode, with a slightly larger peak in the smaller mode. The organic fraction of m/z 43 is found in considerable amounts in both modes, which may be due to the fact that both HOA and SV-OOA contribute ion fragments at that particular m/z ratio. m/z 43 was more heavily distributed in larger particles than m/z 57, which may be indicative of a greater contribution of the oxidized SV-OOA-related ion (C2H3O+) to the m/z 43 distribution.
3.8 Backtrajectory Analysis
 To investigate the influence of air mass origin on ambient particulate matter concentrations, backtrajectory calculations were carried out using the HYSPLIT4 (Hybrid Single Particle Lagrangian Integrated Trajectory, Version 4) model, available from the NOAA Air Resource Laboratory website (http://www.arl.noaa.gov/ready/hysplit4.html). The 72 h backtrajectories were calculated every 3 h (at 02:00, 05:00, 08:00, 11:00, 14:00, 17:00, 20:00, and 23:00 local time, UTC + 8) arriving at an altitude of 500 m in Hong Kong (22°15′N, 114°10′E), employing the Global Data Assimilation System (GDAS) 1° global meteorological data for the period of the intensive campaign. Clustering according to the backtrajectories' spatial distribution was performed using the inbuilt function of the HYSPLIT4 software with an optimum solution of six clusters (Figure 8a). Clusters 1 and 5 represent coastal north easterly air masses, while clusters 2, 3, and 4 comprise backtrajectories originated from oceanic areas to the south and east of Hong Kong. Cluster 6 contains the remaining backtrajectories with north and northwesterly continental air masses.
 Total AMS NR-PM1 mass concentrations were generally lower when Hong Kong was under the influence of air masses from the ocean, with measured mass concentrations only about half of those during periods of continental and coastal air mass influence (Figure 9a). It is apparent from the geographical depiction of the clusters that backtrajectories in cluster 2 were shorter, implying slower air masses than those in clusters 3 and 4. The observed mass concentrations are higher in oceanic cluster 2 than in clusters 3 and 4, likely due to increased accumulation of PM1 as these slow air parcels pass over Hong Kong and urban regions in adjacent mainland China. The observed changes in total mass concentration are largely related to organics and sulfate, as these species dominate NR-PM1, and thus display variations in relation to air mass clusters similar to total NR-PM1 (Figure 8b). The larger ranges of the total as well as individual species concentrations, as indicated by the 10% and 90% whiskers, in clusters 1, 2, and 6 are largely due to the previously discussed episodic pollution events. The variation of sulfate and organics with air mass origin is similarly reflected in the particle size distributions (Figure 8b). Peak areas—a measure of the total concentration—are larger in the land clusters (1, 5, and 6) than in the sea clusters (2, 3, and 4). Furthermore, observed main mode diameters are shifted toward a smaller size (≈400–450 nm) in the oceanic clusters 3 and 4, implying that particles were less aged and under greater local and regional influence. However, observed particle sizes in cluster 2 were larger and on a similar scale (Dva ≈ 500 nm) as in the continental and coastal clusters (1, 5, and 6). This is again likely due to the mentioned shorter backtrajectories and the associated longer residence time over Hong Kong and the PRD, permitting more time for the particles to grow and age before they reached the receptor site at HKUST.
 As discussed earlier, AMS-acquired mass size distributions at the supersite were generally bimodal, with the larger, major accumulation size mode in the range of 400–600 nm and the smaller, minor mode in the range of 100–200 nm of vacuum aerodynamic diameter. For organics, the smaller mode plays a more important role and remains relatively stable regardless of air mass origin (Figure 8c). This persistence confirms that the smaller mode is due to fresh particles of a permanent local source, such as local traffic emissions. The larger accumulation size mode of organics is of considerably smaller magnitude in the sea clusters, where the smaller and larger size mode contributed roughly equally to the overall size distribution. In contrast, the larger mode peak dominates overwhelmingly in the land cluster, indicating that a substantial fraction of organic components in times of coastal and continental air mass influence was associated with larger and more aged particles. These variations in the larger particle size mode illustrate the significance of long-range transport of aged particles during times of coastal and continental air mass for Hong Kong.
 The concentration trends of SV-OOA and LV-OOA roughly follow that of the total AMS NR-PM1 mass concentration, with oceanic clusters generally exhibiting lower concentrations. While the overall trend is similar for HOA, its magnitude of concentration variation is less pronounced than for the oxidized organic aerosol factors (Figure 9b). Examining the change of fractional contribution of resolved OA factors to total OA as a function of air mass origin (Figure 9c) reveals that there is a higher level of fresh HOA present during times of oceanic air mass influence, with average HOA fractions of 25%–30%. Conversely, more aged and highly oxidized LV-OOA is overly dominant during continental and coastal air mass periods at fractions >60% but decreases substantially when air originates from oceanic areas. The contribution of SV-OOA does not vary greatly with air mass origin and stays stable at ≈20% of total OA, indicating a rather constant formation. The observed trends confirm that aged LV-OOA associates strongly with coastal and continental air mass origins as its relative importance decreases by one third during times of oceanic air flow, favoring HOA in turn, whose fractional contribution is otherwise masked by the dominance of SV-OOA and LV-OOA. As mentioned, the nominal change in HOA concentration with changing cluster origin is not strongly pronounced, and hence, the higher observed fraction is due to receding LV-OOA concentration rather than increasing nominal mass concentration of HOA. The cluster analysis of PMF-resolved OA illustrates that most of the variability in OA concentration is due to changes in oxidized LV-OOA associated with air mass that has passed over land before reaching Hong Kong, while HOA and SV-OOA do not show as much variability in relation to air mass origin and are likely associated with local and regional sources. The relation between oxidized organic aerosol and air mass origin is similarly visible from the elemental ratio trends (Figure S6, supporting information). The average OM/OC ratio, an indicator of the degree of oxygenation with higher ratios implying more strongly oxidized organic aerosol, is generally above 1.7 in coastal and continental clusters and below 1.6 in oceanic clusters. This is the combined effect of a higher O/C and a lower H/C ratio found in land clusters (1, 5, and 6), indicating greater abundance of highly oxidized species and lower content of fresher hydrocarbon-like organic components. The opposite applies to ocean clusters where lower O/C and higher H/C ratios were observed. In these clusters, the abundance of more saturated and oxygen-poor organic compounds, as reflected in the higher H/C ratio, suggests that the organic aerosol components are more related to fresher emissions.
3.9 Episodic Events
 Elevated NR-PM1 concentrations or characteristic composition changes were observed on three separate occasions, covering the periods from 27 to 29 April, 14 to 17 May, and 26 to 30 May (Figure 1). Among these, the late May episode was the most severe pollution event recorded during the whole sampling campaign with NR-PM1 mass peaking at 72.4 µg/m3. The late May episode (Figure 10) was characterized by persistent northerly and northwesterly air masses passing over the PRD before coming into Hong Kong. Accumulation of local and regional pollutants was enhanced by slowing surface winds and the establishment of a well-defined land-sea breeze with a gradual daily reversal of wind direction, as evident from the changes in wind direction observable not only at HKUST, but also at Tung Chung (southwestern urban area) and Yuen Long (northern urban area) (Figure S5, supporting information). At HKUST, westerly winds were observed during nighttime, shifting northerly in the early hours. Daytime winds were first from the east turning southerly toward the afternoon and back westerly in the later evening to complete the cycle. Under such conditions, large-scale land-sea breeze effects over the whole PRD region can lead to an effective redistribution and accumulation of air pollutants [Lo et al., 2006]. An indication of the local and regional character of the accumulated aerosol is that among the organic aerosol, the slightly oxidized SV-OOA from fresher urban emissions increased most markedly during the episode. The marked increases in the concentrations of SV-OOA and LV-OOA suggest that oxidation of POA in the PRD region under the given circulatory conditions was efficient and likely related to photochemical processes as sunny and dry weather conditions prevailed over the course of the episode. The trends in elemental ratios (Figure 6a) support this observation: the OM/OC ratio continuously increased over the course of the episode, reaching its peak at OM/OC = 2.1, which was the highest value recorded during the whole campaign. This increase was caused by a rising O/C ratio, indicating a considerable rise in the degree of oxygenation of organic aerosol in this period.
 The late April episode was characterized by a subtle increase in total NR-PM1 concentration, largely due to increases in particulate organics and nitrate. This, however, led to a significant change in the average aerosol composition, with higher nitrate content and roughly equal fractions of organics and sulfate. Accounting for the increase in total organics were mainly the oxygenated organic aerosol factions, LV-OOA and SV-OOA. This trend is also reflected in the OM/OC and O/C ratios, which increased from 1.4 to 1.6 and from 0.3 to 0.4, respectively. Simultaneously, the H/C ratio decreased from 1.5 to 1.3. Altogether, this indicates a greater presence of aged and oxygenated organic aerosol. The said period was mainly foggy with RH > 90% and may be indicative of aqueous phase oxidation processes taking place.
 The mid-May event exhibited a significant increase in sulfate mass concentration and a high fractional contribution of aged LV-OOA, which constituted more than 60% of total organic aerosol during the period. Air masses originated from the northeast coast of China during that time, with 72 h backtrajectories extending as far as Shanghai. As no other meteorological anomalies were observed, it is likely that appreciable amounts of this aerosol were from farther northeasterly source areas in China and Taiwan and thus highly aged. This agrees with Ho et al.  who attributed observed secondary organic aerosol at a rural background sampling location in Hong Kong to long-range transport from the northeast. In contrast to the other two episode periods, the increase in OM/OC ratio from 1.55 to 1.7 in mid-May was caused by a decrease in H/C ratio alone (from 1.5 to 1.2) rather than a change in O/C ratio. This confirms the receding importance of fresh local emissions in favor of largely LV-OOA and that the observed degrees of oxygenation of SV-OOA and LV-OOA before and after the mid-May episode were not significantly different.
 A more detailed discussion of these pollution episodes is presented in a separate study [Li et al., 2013].
 Continuous real-time measurements with an Aerodyne HR-ToF-AMS at a suburban coastal location in eastern Hong Kong were carried out in springtime 2011. NR-PM1 was dominated by sulfate (51.0%) and organics (28.2%) with little contribution from nitrate (4.1%) and an average monthly total NR-PM1 concentration of 14.5 µg/m3. Instrumental intercomparison with a collocated Applikon MARGA and Sunset ECOC thermo-optical analyzer yielded good agreement in terms of temporal trends but significant discrepancies with respect to nominal concentrations. Uncertainties in AMS particle collection efficiency and the presence of coarse mode nitrate likely contributed to the observed differences. The calculated in situ pH based on E-AIM II predictions remained largely constant over the sampling period at an average pH of 0.95, implying considerable acidity. The measured species size distributions were bimodal with a major accumulation mode peak at ≈570 nm and a minor accumulation mode peak at ≈200 nm of vacuum aerodynamic diameter (Dva), with considerable fractions of organics and nitrate in the smaller mode. Source apportionment of organic mass spectra by positive matrix factorization (PMF) indicates the presence of three distinct organic aerosol fractions. Hydrocarbon-like organic aerosol (HOA) was attributed to local traffic emission with a contribution of 25% of total organic aerosol and a strong representation in smaller particles. Semivolatile oxygenated organic aerosol (SV-OOA) was characterized as a mildly oxidized organic aerosol of primary origin, likely emitted from a farther distant local source area. Lower-volatility oxygenated organic aerosol (LV-OOA) was considerably aged. It displayed a high degree of oxygenation and appeared to be related to photochemical activity. Its constituents were more confined to the larger size mode. Three major episodic pollution events were recorded during the sampling period attributed to different circulatory, meteorological, and air mass origin conditions. Backtrajectory analysis and air mass clustering showed a clear gradient of NR-PM1 concentration which was related to air mass origin: measured PM during oceanic air mass influence < PM during coastal air mass influence < PM during continental air mass influence. Shifts to smaller particle size modes during times of oceanic air mass influence illustrate greater importance of less aged and fresher local and regional aerosol during such periods. HOA and SV-OOA concentrations tended to show less variability in relation to air mass origin confirming their local nature, while higher LV-OOA concentrations were associated with coastal and continental air mass influence and related long-range transport of aerosol.
 This work was supported by the University Grants Committee (Special Equipment Grant, SEG-HKUST07 and General Research Fund GRF 600413) and the Environment and Conservation Fund (ECF) of Hong Kong (Project ECWW09EG04).