Semicontinuous measurements of water-soluble organic carbon (WSOC) and organic carbon (OC) in PM2.5 were made at Gosan, Korea, in March-April 2005. On average, the WSOC/OC mass ratio for all air masses observed at Gosan was 0.30 ± 0.12. WSOC correlated well with CO (r2 = 0.54) in Chinese outflow, suggesting that a major part of the observed WSOC and/or their precursors was of combustion origin. The relationship between the increase of WSOC and O3 suggests that the observed WSOC was mostly secondary product. To interpret the measured organic compounds, thermal analyses of organic standards were made in the laboratory. Thermograms of a single standard of water-soluble organic species showed that carbon that evolved at high temperatures (600°–870°C) was generally associated with water-soluble compounds having high molecular weights (MWs) on the order of hundreds, while carbon that evolved at low temperatures (<300°C) generally had MWs of less than ∼180 g mol−1. Positive matrix factorization (PMF) analysis revealed three organic compound groups (low, medium, and highly refractory compounds) based on the OC thermograms. On average, highly and low refractory compound groups accounted for 79% and 21% of the WSOC mass, respectively, at Gosan. Highly refractory compound groups significantly contributed to WSOC regardless of air mass origin. The results of the laboratory experiments imply that a large fraction of these highly refractory compound groups was likely associated with high MW compounds. For water-insoluble organic carbon(=OC–WSOC), medium and low refractory compound groups accounted for 60% and 40%, respectively, consistent with the results of the laboratory experiments.
 Water-soluble organic carbon (WSOC) can significantly alter the hygroscopicity of aerosols and can be important in determining the ability of particles to act as cloud condensation nuclei (CCN) [Novakov and Penner, 1993; Saxena et al., 1995; Facchini et al., 1999]. A major source of WSOC is considered to be secondary organic aerosols (SOA), which are formed by oxidation of volatile organic compounds (VOCs) followed by condensation on existing particles and/or nucleation. Moreover, recent studies suggest that in-cloud oxidation processes may provide a pathway for production of oxalic acid [Warneck, 2003; Crahan et al., 2004], which is generally the most abundant dicarboxylic acid in the atmosphere [Kawamura and Sakaguchi, 1999]. Recent field studies have shown that bulk WSOC and estimated SOA have similar chemical characteristics in an urban area [Miyazaki et al., 2006; Kondo et al., 2007]. Over regions remote from primary sources, WSOC also includes aged primary components of OC as well as SOA [Fuzzi et al., 2006] and thus accounts for a significant portion of the OC mass.
 However, recent field studies have also indicated that SOA are underestimated in current models, both at the surface [de Gouw et al., 2005; Volkamer et al., 2006] and in the free troposphere [Heald et al., 2005]. Knowledge of the relative abundance of WSOC and its chemical characteristics provides insight into processes leading to SOA and CCN, which have important impacts on both regional air quality and global climate.
Decesari et al.  proposed that WSOC can be separated into three main classes according to the following acid/base characteristics: neutral compounds, monocarboxylic/dicarboxylic acids, and polyacids. In particular, polyacidic compounds are often referred to as humic-like substances (HULIS) [e.g., Havers et al., 1998; Kiss et al., 2002; Mayol-Bracero et al., 2002]. HULIS species in atmospheric aerosols, have molecular weights (MWs) estimated to be on the order of several hundreds. HULIS exhibit some structural properties (e.g., spectra) similar to those of humic/fulvic acids that naturally occur in terrestrial and aquatic environments [Duarte et al., 2005]. However, they have a number of dissimilarities in their physical and chemical properties (e.g., surface activity, droplet activation, and MWs) [Gelencsér et al., 2002; Graber and Rudich, 2006]. In ambient aerosols, the estimated fraction of HULIS in WSOC in the fine mode is highly variable. For instance, it ranges from 15–36% in biomass burning aerosols in the Amazon basin [Mayol-Bracero et al., 2002], to a high of 55–60% at European background sites [Krivácsy et al., 2001], although the definitions of HULIS are different depending on the analytical methods.
 Several recent field studies have made total and speciated OC measurements over the western Pacific region. WSOC measured in airborne filter samples during the Asian Pacific Regional Aerosol Characterization Experiment (ACE)–Asia showed that the WSOC fraction contributed from 10 to 50% of the total carbon (TC) mass [Mader et al., 2004]. Kawamura et al.  reported that total dicarboxylic acid (C2-C5) mass accounted for on average 1.8% of the total organic carbon (OC) sampled by aircraft measurements over the east Asia/Pacific region in spring. In east Asia, however, the organic compounds identified in previous studies accounted for <20% of OC [e.g., Simoneit et al., 2004; Yang et al., 2005]. Field data on WSOC in ambient air are still lacking, especially for remote areas in east Asia, and the chemical characteristics of the major fraction of WSOC over this region are not well understood. Knowledge of the high MW fractions of WSOC compounds in the atmosphere, in particular, would improve our understanding of the sources and formation processes of SOA. The high MW organic fractions may also affect the activation of particles to form cloud droplets [Charlson et al., 2001]. However, their role in cloud droplet growth is still not well quantified, mainly because the chemical and physical properties of real organic compounds in atmospheric aerosols are poorly characterized [Dinar et al., 2006].
 Semicontinuous measurements of WSOC and OC aerosols in PM2.5 were made at Gosan on Jeju Island, Korea, in March and April 2005 during the East Asian Regional Experiment (EAREX) in the framework of the United Nations Environment Programme (UNEP)/Asian Brown Clouds (ABC) project. In this paper, we present the results of the WSOC measurements and examine the WSOC composition from the characteristics of its thermal evolution. The following sections present results based on the (1) thermal analysis of organic standards obtained in the laboratory, (2) temporal variations in WSOC, (3) correlation of WSOC with chemical tracers, and (4) characterization of the observed WSOC in relation to the thermal properties of OC.
 Mass concentrations of OC and EC were measured using a semicontinuous EC/OC analyzer (Sunset Laboratory, Inc., Tigard, OR, USA) [Birch and Cary, 1996; Kondo et al., 2006; Miyazaki et al., 2006]. This instrument collected ambient aerosol particles on a quartz-fiber filter for 45 min at a flow rate of 8 L min−1 and then analyzed them using the thermal-optical-transmittance (TOT) method for 15 min. The carbonaceous compounds on the filter sample were thermally converted to carbon dioxide (CO2), which was subsequently measured by nondispersive infrared absorption (NDIR). Pure helium (He) and 10% oxygen (O2) in He (He/O2) gases were used as the carrier gases for detecting OC and EC, respectively.
 The relative amounts of total evolved carbon allocated as OC and EC are sensitive to the temperature protocol used for the analysis [Schauer et al., 2003]. In the present study, we used a temperature protocol based on that proposed by the National Institute for Occupational Safety and Health (NIOSH) [Birch and Cary, 1996]. Figure 1 shows a typical thermogram obtained at Gosan during the study period. The sample on the filter was heated in four temperature steps; the temperature profile was 300°C for 75 s, 450°C for 60 s, 600°C for 60 s, and 870°C for 90 s. The OC evolved during these steps was operationally defined as volatile OC. In this study, the OC concentrations detected in temperature steps of 300°C, 450°C, 600°C, and 870°C are defined as OC1, OC2, OC3, and OC4, respectively.
 In the He atmosphere stage of analysis, a fraction of OC was charred or pyrolyzed. Such pyrolyzed organic carbon (PC) can absorb light, reducing the transmittance monitored at 670 nm through the filter (Figure 1). This step was followed by exposure of the samples to the He/O2 atmosphere in the 870°C temperature step. The PC and EC evolved at this step, increasing the transmittance. To split the amount of total carbon into OC and EC, the carbon that evolved to bring the transmittance back to its prepyrolysis level (dash-dotted line in Figure 1) was assumed to be equal to the PC. Here, PC is defined as OC that evolves in the He/O2 mode. Previous studies have shown that WSOC is prone to charring and that the PC/WSOC ratio tends to increase with WSOC loadings [Yu et al., 2002; Andreae and Gelencsér, 2006]. Consequently, the PC/WSOC ratio can provide a measure of the amount of charring relative to WSOC, which affects the speciation of OC and EC. In this study, the average measured PC/WSOC ratio at Gosan was as low as 0.07 ± 0.04.
2.2. Laboratory Experiments
 The thermal characteristics of OC, such as the position of thermogram peaks, are mainly controlled by the chemical composition of the organic aerosols and heating temperature [e.g., Gelencsér et al., 2000; Yu et al., 2004]. Thermal analytical methods to study the thermal characteristics of particle samples have been widely used [e.g., Ellis and Novakov, 1982; Grosjean et al., 1994]. Moreover, the OC fraction evolved at each temperature step has been used recently in source apportionment studies [e.g., Kim et al., 2004]. To investigate the thermal characteristics of OC in terms of the MW of organic species under the current operating conditions, thermograms were obtained in the laboratory using different single standards of organic compounds relevant to atmospheric aerosol (more than 20 organic species). Standards of water-soluble organic species were dissolved in purified water and were then applied by pipette to prebaked blank quartz filters. For water-insoluble organic species, a single standard was directly applied to the filter without using any solvent. In this laboratory experiment, we defined the evolution peak in the He/O2 mode as OC5. OC5 represents the fraction of OC that is resistant to volatilization in the He mode (including carbon that undergoes charring) and requires the same conditions as EC evolution. This section summarizes key results of these experiments, with details provided in Table 1.
Table 1. Summary of the Laboratory Experiments Using Standards of the Organic Speciesa
Most of these species are water-soluble organic compounds; six species at the bottom were examined as representative of water-insoluble organic compounds. Concentrations of each species were converted to ambient air concentrations under the sampling conditions in the field.
For the definition of OC5, see text.
Suwannee River fulvic acid and humic acid are produced by the International Humic Substance Society and are desalted (H+-saturated) by cation exchange.
Figure 2 shows typical thermograms of single organic species obtained in the laboratory using the same operating temperature protocol as that used in the field study. The water-soluble organic species shown are oxalic acid (MW = 90 g mol−1), levoglucosan (MW = 162 g mol−1), azelaic acid (MW = 188 g mol−1), and Suwannee River fulvic acid (SRFA). Depending on the analytical method, the MW of SRFA has been determined as ranging from hundreds to 1,000 Da [e.g., Hur and Schlautman, 2003]. For example, Dinar et al. , using a UV-VIS spectrometer, estimated the average MW of SRFA to be 780 Da. In our study, oxalic acid and SRFA were used to represent low and high MW water-soluble organic species, respectively. The thermal evolution of oxalic acid was mostly completed in the OC1 stage (<300°C), where OC1 accounted for ∼95% of the total OC detected. For the other standards of water-soluble organic species examined, almost all the species with MWs <∼180 g mol−1 evolved mostly at the OC1 stage (Table 1). For levoglucosan, 62% of the mass evolved at the OC1 stage, while the remaining mass evolved at the OC2 (18%), OC3 (8%), and OC4 (9%) stages. In addition, a small fraction (3%) of PC formed. These thermal profiles of levoglucosan are similar to those obtained by Subramanian et al. . On the other hand, a significant amount of azelaic acid evolved at the OC3 (32%) and OC4 (65%) stages. For SRFA, about 71% of the carbon showed peaks at the highest temperature stages (600°–870°C) both in the He (35%) and He/O2 (36%) modes. Because SRFA does not contain EC and the laser signal decreased in the He mode, the carbon in the He/O2 mode represents the part of SRFA that underwent charring to form PC during the analysis [Yu et al., 2004]. Our experiments also showed that the organic fractions that underwent charring (i.e., OC5 and some fractions of OC2-4) generally increased with the increasing MW of water-soluble organic species (Table 1). For example, thermograms of dicarboxylic acids revealed that the fractions of carbon that evolved at the highest temperature ranges of OC4 and OC5 increased as carbon numbers increased from oxalic acid (C2) to azelaic acid (C9). In summary, the present results indicate that, for the species examined in this study, the complete evolution of a single water-soluble organic compound generally requires higher temperature conditions with increasing MW.
Yu et al.  obtained similar experimental results for the relationships between the thermograms and MWs of several water-soluble organic compounds. Their thermograms of organic standards of ethanolamine, oxalate, levoglucosan, and humic acid with temperature profiles were similar to those in the present study. Andreae and Gelencsér  also reported on the thermal evolution of high MW compounds at high temperatures in the presence of O2, although their analytical method differed from ours. They presented a thermogram of a humic acid sample (typically dark brown to black substances) as a function of heating temperature in an oxidative atmosphere, demonstrating that complete evolution required temperatures as high as ∼700°C with a thermal peak centered near 600°C.
 For water-insoluble organic species, thermograms of nonacosan, docosanol, and hexadecanoic acid are shown in Figure 2b. These organic compounds have been found to dominate identified water-insoluble organic species (generally MW >200 g mol−1) over the western Pacific [Simoneit et al., 2004]. These species evolved mostly in the OC2 stage, with small fractions evolving in the OC3 stage. For species with MW <160 g mol−1, water-insoluble organic carbon evolved mostly in the OC1 stage (Table 1), similarly to the water-soluble organic species examined here. Generally, these water-insoluble organic standards produced little charring compared to the water-soluble organic species (Table 1). These results are consistent with the thermogram features for water-insoluble organic compounds obtained by Yu et al. .
 Thermograms were also obtained by changing the concentrations of organic standards for several species. The resulting shapes of the thermograms were not significantly affected by the amount of organic species within a range of 0.5–11 μg C m−3 for at least the standards examined in this study. However, considering the limited number of species examined, there are still uncertainties in the effect of the concentrations of organic species in ambient aerosols on the thermogram peaks.
2.2.2. Multicomponent Thermograms
 When interpreting thermograms obtained in field measurements, it is important to prove that the thermograms of various organic compounds can be regarded as a linear combination of the individual species under the same thermal conditions. Thus we also obtained thermograms of a mixed solution of water-soluble organic compounds for comparison with the direct sum of the individual thermograms obtained in section 2.2.1. Figures 3a and 3b show typical examples of the thermograms. About 85% of the mixed solution of three dicarboxylic acids (oxalic acid, fumaric acid, and adipic acid) evolved in the OC1 stage (Figure 3a). Both the temperature range of the evolution and the peak area were clearly similar to those of the sum of the individual peaks. Although the position of the peak of the mixed solution at the OC1 stage slightly shifted at higher temperatures by ∼20°C compared to that of the sum, the peaks were still within the same volatility region (OC1) in the thermograms. The thermograms of the mixed solution of oxalic acid, azelaic acid, and SRFA were also generally similar to that of the sum of their individual thermograms (Figure 3b). The thermograms of the mixed solution and the direct sums of individual thermograms for more than 20 combinations (for species listed in Table 1) generally showed little difference under our operating conditions.
 Previous studies have suggested that mixing with inorganic species can shift the thermal evolution of both organic compounds and EC. For example, ammonium sulfate ((NH4)2SO4), typically the most abundant inorganic species in the fine mode found in Asian pollution outflow [e.g., Topping et al., 2004], can shift the thermal evolution of OC [Andreae and Gelencsér, 2006]. Moreover, there is some evidence that mixing with other inorganic species can also catalyze oxidation and shift the thermal evolution of EC to lower temperatures in response to the presence of metals such as sodium (Na+) in an oxidizing atmosphere [Grosjean et al., 1994]. To investigate these possible effects, standard solutions of organic species were mixed with those of inorganic species to obtain their thermograms. Figure 3c shows thermograms of oxalic acid with (NH4)2SO4 compared to that of oxalic acid. The thermogram of ammonium oxalate ((COONH4)2) is also shown for comparison. Although the evolution temperatures of characteristic peaks differed slightly (shifting by ∼15°C), the peaks of the three components were still within the same temperature range (OC1) of the thermograms. Figure 3d shows thermograms of SRFA and SRFA with sodium chloride (NaCl). The thermograms of SRFA with the addition of NaCl exhibited a downward temperature shift by about 20°C for OC5 in the He/O2 mode. Experiments with two concentrations of NaCl (∼50% and 100% of SRFA in mass) yielded identical temperature shifts. The effect of Na+ as a catalyst lowering the oxidation temperature of EC (PC in this case) in an oxidizing atmosphere is consistent with the results of a previous study [Grosjean et al., 1994]. However, in our experiment, NaCl did not significantly change the peaks of evolved SRFA in the He mode. These experimental results indicate that the mixture of these inorganic species with organic species examined here did not significantly bias the definition of OC1-4 under the operating conditions in this study. We can therefore assume that the thermograms of measured OC in ambient aerosols represent a linear combination of each of the OC component thermograms in the present study.
 It should be noted that organic compounds do not always simply evolve at a given single temperature because they may undergo thermally driven reactions before they volatilize (e.g., by charring) as shown in the experiments. Nevertheless, our results indicate that the complete evolution of water-soluble organic compounds generally requires higher temperature conditions with increasing species MWs, under the operating conditions of our study. The thermal characteristics of organic species described in this section were subsequently used for the chemical interpretation of factors obtained by positive matrix factorization (PMF) analysis, as described in section 6.
3.1. Site Description and Synoptic-Scale Meteorology
 Measurements of aerosol and gas species were performed at the Gosan ground superstation (33.17°N, 126.10°E) on Jeju Island, located approximately 100 km south of the Korean peninsula. The measurements were made during the ABC-EAREX 2005 campaign from 17 March to 4 April 2005. Jeju has no large local industrial sources and is thus an ideal location for monitoring polluted air masses from east Asia [Carmichael et al., 1997]. The sampling site was located ∼60 m above sea level on the western edge of Jeju Island. Trailers containing each instrument were situated about 10 m inland from the sea cliffs.
 Previous studies of aerosol characterization have reported that the air masses arriving at Gosan in spring are frequently influenced by anthropogenic sources in China, Korea, and Japan, as well as by Asian dust storms [e.g., Huebert et al., 2003]. During our observation period, the Siberian High dominated over the continent (centered near ∼50°N, 100°E), and transient midlatitude cyclones migrated from southwest to northeast of the Gosan site every 4 to 5 d [e.g., Sawa et al., 2007]. This led to a dominant flow of northerlies to northwesterlies at the surface at Gosan, suggesting that the observed air masses were subject to regional influences from the Asian continent during the study period. Indeed, previous studies have typically observed the largest enhancements of anthropogenic aerosols within the boundary layer (BL) over the western Pacific in association with postfrontal BL outflow from China in spring [e.g., Jordan et al., 2003; Miyazaki et al., 2005].
 Lidar observations showed that the height of the BL top over the study area varied at 1–2 km during the observation period (N. Sugimoto et al., unpublished data, 2005). We therefore defined the mean BL height over the sampling region as 1.5 km (pressure altitude of ∼850 hPa) for use in air mass classification based on back trajectories, as described in section 5.1.
 Measurements of WSOC were made using a Particle-into-Liquid Sampler (PILS) [Weber et al., 2001; Orsini et al., 2003] followed by online quantification of WSOC every 6 min by a TOC analyzer (Model 810; Sievers, Boulder, CO, USA) [Sullivan et al., 2004; Miyazaki et al., 2006]. Ambient aerosol was sampled at a flow rate of 16.7 L min−1 by the PILS, which used a steam saturator to grow the aerosol to sizes collectable by inertial impaction. The carbonaceous compounds in the liquid sample were then quantified online using the TOC analyzer at a flow rate of 0.3 mL min−1. In the TOC analyzer, organic compounds dissolved in deionized water are oxidized to form carbon dioxide (CO2) using an ultraviolet (UV) lamp and ammonium persulfate as a chemical oxidizing agent. The CO2 formed is then measured using a membrane-based conductivity detection technique. The concentrations of dissolved TOC in the present study ranged between 25 and 320 μg C per 1 L of water. For this study, we technically define WSOC as the particles sampled by the PILS and detected by the TOC analyzer after penetrating a liquid filter (a polypropylene syringe filter 17 mm in diameter) with a pore size of 0.22 μm. A 20 cm long parallel plate denuder composed of absorbent surfaces impregnated with activated charcoal [Eatough et al., 1993] was placed upstream of the PILS to remove gaseous carbonaceous interferences.
 Zero levels (blanks) of WSOC were measured automatically for 30 min every 4 h by passing the sample air through a Teflon PTFE membrane filter with a 47 mm diameter upstream of the PILS. The zero levels ranged between 17 and 24 μg C per 1 L of water during the study period. The uncertainty in the WSOC measurements was estimated to be between 5 and 9%. Comparison of PILS-WSOC measurements with WSOC manually extracted from 12-h-integrated quartz filters for a 6-d period in Tokyo showed agreement within 12% [Miyazaki et al., 2006].
 The sample line from the inlet to the PILS was a 4-m-long stainless steel and copper tube with a 1-mm inner diameter. A PM2.5 (2.5 μm diameter cutoff size) cyclone (URG-2000-30EH; URG Corp., Chapel Hill, NC, USA) was used. In a previous study, size-segregated, filter-based measurements of WSOC at the same site in spring showed 75% of the WSOC mass on average in the fine (<1.5 μm) mode [Topping et al., 2004]. That study indicated that the WSOC mass were dominantly in the PM2.5 range at the sampling site.
3.2.2. Inorganic Aerosols
 The inorganic bulk composition of aerosols was also obtained using another PILS and ion chromatography analysis [Orsini et al., 2003; Takegawa et al., 2005] with another inlet system including a PM2.5 cyclone. The mass concentrations of sulfate (SO42−) and other inorganic compounds were obtained every 15 min. The SO42− measurements discussed here had a detection limit and uncertainty of 0.002 μg m−3 and 12%, respectively.
 We estimated the concentrations of non-sea-salt (nss) SO42− assuming that all of the observed Na+ at the sampling site originated from NaCl in sea salt. On average, the fraction of nss-SO42− was ∼97% of the total SO42− in the PM2.5 mode for the entire period, indicating that ss-SO42− was a minor contributor to the total SO42− in the PM2.5 mode. We therefore use the term SO42− instead of nss-SO42− in this paper.
3.2.3. OC, EC, and Black Carbon (BC)
 Measurements of OC and EC were made using the instruments described in section 2.1. Measured OC and EC data were available for the period 24 March to 4 April. The inlet for air sampling was equipped with a PM2.5 cyclone (Model URG-2000-30EHB; URG Inc.) and denuder identical to that used for measuring WSOC. Background levels of OC were periodically checked (four data in total) by a particle filter placed upstream of the denuder. The average background level of OC was 0.94 ± 0.60 μg C m−3, at which the adsorption of any remaining organic vapors is supposed to be the most critical factor causing the artifact of OC. The detection limits of these measurements were 1 μg C m−3 and 0.2 μg C m−3 for OC and EC, respectively. The overall accuracies of the OC and EC measurements were estimated to be 16 and 22%, respectively, from uncertainties associated with the sensitivity calibration, aerosol sampling, and temperature protocol [Kondo et al., 2006; Miyazaki et al., 2006]. We defined water-insoluble organic carbon (WIOC) measured at Gosan as WIOC = OC − WSOC and estimated the WIOC uncertainty to be 26%, by combining the errors of the OC and WSOC measurements [Miyazaki et al., 2006].
 The black carbon (BC) concentrations were measured by light absorption at a wavelength of 565 nm using a particle soot absorption photometer (PSAP, Radiance Research, Seattle, WA, USA) equipped with a PM2.5 inlet cyclone. The PSAP was operated with a 1-min time interval at a flow rate of 1.0 L min−1. The inlet was heated to 400°C to effectively volatilize nonrefractory aerosol components (e.g., sulfate and nitrate) [Kondo et al., 2006]. The specific mass absorption coefficient of BC (σBC) was determined to be 8.9 m2 g−1 by comparison of the light absorption measured by the PSAP with a heated inlet and the mass concentration of EC measured in Tokyo [Kondo et al., 2006]. The detection limit of BC measurements was 0.05 μg m−3 for 1 min. Mass concentrations of BC measured by the PSAP with the heated inlet were compared with those of EC measured by the Sunset Laboratory semicontinuous EC/OC analyzer for a 12-d period within the observation period. The mass concentrations of EC agreed with those of BC to within 2% (slope = 0.98, r2 = 0.97). This agreement suggests that the catalytic effect of inorganic species shown in 2.2.2 can be negligible in our determination of EC concentration. Here, we use BC concentrations instead of EC concentrations because the BC data were obtained for a longer period. This paper reports 1-h averaged OC and BC concentrations. All of the aerosol instruments used in this study measure bulk chemical compositions and cannot measure the mixing state of individual particles.
3.2.4. CO and O3
 Carbon monoxide (CO) concentrations were measured using a NDIR instrument (APMA-360 model, Horiba, Kyoto, Japan) at a flow rate of 1 L min−1, with an integration time of 1 h [Sawa et al., 2007; Tanimoto et al., 2007a]. The background (zero) signals were routinely measured every hour by passing the sample air through a CO removal catalyst (Sofnocat; Molecular Products Ltd., Essex, U.K.) to produce CO-free air. Calibrations were performed every other day by supplying a CO standard (1 part per million by volume [ppmv] CO in air). The overall uncertainty of the hourly CO measurements was estimated to be ∼10% at a CO mixing ratio of 150 ppbv.
 Ozone (O3) was measured using a commercial UV absorption O3 monitor (Model 1100, Dylec, Japan) with an integration time of 1 h [Tanimoto et al., 2007b]. The precision of the O3 instrument was estimated to be ∼1 ppbv. All of these gas species data were merged onto the time interval of OC sampling (∼45 min).
4. Temporal Variations of WSOC
Figure 4 shows time series of WSOC mass concentrations and other measured parameters. The time series provides insights into the range of these parameters during the study periods and the amount of data available for each measurement. The WSOC concentrations ranged from levels below the detection limits to a maximum value of about 8.3 μg C m−3 (Figure 4a). The average WSOC concentration observed throughout the study period was 1.2 ± 0.9 μg C m−3.
 The temporal trend of WSOC concentrations tracked those of SO42− (Figure 4a), BC, and CO concentrations (Figure 4b), which ranged between 0.5–20 μg m−3, 0.2–4.0 μg C m−3, and 200–800 ppbv, respectively. The trend of O3 concentrations, which ranged between 40–90 ppbv, was also similar to those of these species (Figure 4c). High concentrations of WSOC (>2 μg C m−3) were observed on (1) 17–18 March, (2) 22–23 March, (3) 28 March, and (4) 30 March to 1 April. During periods 1–3, a surface low was developing just south of the Korean peninsula with a cold front stretching southwestward over the East China Sea [Sawa et al., 2007]. The sampling site was located behind a surface cold front associated with the low-pressure system, causing the transport of pollutants inland from northeast China. During these periods, precipitation and a decrease in surface pressure (Figure 4d) were observed in association with the passage of cold fronts; such fronts typically passed every 4–5 d at the study site. Increased concentrations of the above species were observed just after the passage of a cold front; the WSOC and CO concentrations were greater than 2 μg C m−3 and 400 ppbv, respectively. From 31 March to 2 April, a high-pressure system was stationary over eastern China, as reflected by the relatively stable surface pressure (Figure 4d). This stationary system brought air from the northwest to Jeju Island and accumulated air masses over the sampling site without precipitation.
 The variability in the WSOC levels could be interpreted as resulting not only from transport from an upwind region, but also from local production of WSOC. Miyazaki et al.  showed diurnal variation in WSOC, with a maximum during the day at an urban site in Tokyo, which variation was interpreted as the dominance of local production of WSOC. To examine the relative degree of local photochemistry, median diurnal profiles of WSOC and other species are shown in Figure 5. The diurnal profiles of WSOC, SO42−, and O3 have no distinct patterns. This suggests that the effects of local photochemistry on the temporal variations in WSOC, which is usually most intense in the afternoon, were relatively weak compared to variations due to transport of air masses. Therefore, as illustrated in Figure 4, the temporal variations in the WSOC mass concentrations observed at Gosan were mainly driven by regional meteorology rather than by local photochemistry. The diurnal variations in BC also showed no clear patterns, suggesting insignificant effects of local emissions at the sampling site.
5. Possible Origins and Processes of WSOC
5.1. Air Mass Classification
 To characterize the observed WSOC in the different air masses encountered during the study period, 3-d kinematic back trajectories were calculated. The trajectories were calculated every 2 h for air masses starting from the sampling site at pressure altitudes of 950 hPa using European Centre for Medium-Range Weather Forecasts (ECMWF) data and a calculation program developed by the National Institute of Polar Research, Japan. In the following analysis, we focus on the period 25 March to 3 April, because a complete data set including OC was obtained for this period. The trajectories were classified into three broad categories according to the regions of air mass origin; maritime (MR), free tropospheric (FT), and Chinese (CH) air masses. Figure 6 shows typical trajectories for the three cases. MR air masses were defined as those transported over the East China Sea and/or Yellow Sea without passing over the continent (Figure 6a). The MR air masses had spent at least 2 d over the sea prior to sampling. FT air masses were defined as those transported from the free troposphere (at a pressure altitude greater than 850 hPa) over the continent by northwesterly winds; these air masses then descended to the BL over the Yellow Sea (Figure 6b). The CH air masses originated over the northeastern part of China and were advected to the sampling site within the BL (Figure 6c). The average transport time of the CH air masses was 1.5 ± 0.8 d from the coastal region (120°E) to the sampling site. The air mass classification results identified 62% of all the trajectories as CH air masses, 21% as FT and 17% as MR.
 Average values of the measured species according to the back trajectory categories are summarized in Table 2. Th CH air masses had the largest average concentration of WSOC (1.8 μg C m−3), whereas the MR and FT air masses had relatively low concentrations (<1.0 μg C m−3). The CH air masses also had the largest mean concentrations of BC, CO, O3, and SO42−, reflecting the strong influence of anthropogenic emissions. Notably, the average SO42− concentration in CH air masses was as high as 5.9 μg m−3. This value is close to the average PM2.5 SO42− concentration of 5.4 μg m−3 previously measured at Gosan in spring [Lee et al., 2001]. Streets et al.  estimated that Chinese emissions of anthropogenic species dominate those of other Asian regions, contributing 41% of BC, 42% of CO, 42% of NOx, and 59% of SO2 to the total emissions in Asia. The present results show the consistency between the measured values of the chemical species and the classification of air masses by trajectory.
Table 2. Mean Values and Standard Deviations of Each Species for the Three Categories Defined by Back Trajectoriesa
Air Mass Category
WSOC, μg C m−3
WIOC, μg C m−3
OC, μg C m−3
BC, μg m−3
SO42−, μg m−3
ND indicates the number of data points used.
Chinese (CH), ND = 121
1.8 ± 0.7
3.2 ± 1.3
5.4 ± 1.7
0.34 ± 0.13
2.1 ± 0.8
338 ± 68
67.9 ± 9.6
5.9 ± 3.1
Free tropospheric (FT), ND = 42
0.8 ± 0.4
2.5 ± 1.2
3.3 ± 1.2
0.26 ± 0.11
0.7 ± 0.2
229 ± 24
57.1 ± 5.1
1.6 ± 0.4
Marine (MR), ND = 33
0.3 ± 0.2
3.0 ± 1.4
3.3 ± 1.4
0.11 ± 0.08
0.4 ± 0.2
204 ± 24
45.3 ± 6.7
1.6 ± 1.2
1.2 ± 0.9
3.1 ± 1.4
4.6 ± 1.8
0.30 ± 0.14
1.5 ± 0.9
303 ± 81
62.3 ± 12.4
4.8 ± 3.3
5.2. WSOC/OC Ratio
 The water-soluble fraction of total OC provides valuable clues to the composition and chemical processes of organic aerosols. Figure 7 shows temporal variations of WSOC and OC concentrations together with the WSOC/OC ratios. The average OC concentration was 4.6 ± 1.8 μg C m−3, ranging from 2 μg C m−3 to a maximum value of ∼10 μg C m−3. Throughout the study period, the measurements captured short-term temporal variations of OC and WSOC on timescales of several hours to a day. The temporal trends of the OC concentrations were similar to those of the WSOC concentrations (r2 = 0.77). The observed WSOC/OC ratios ranged from 0.02 to 0.66, with average WSOC/OC ratios of 0.30 ± 0.14. The ratios tended to increase when the concentrations of WSOC increased, with ratios larger than 0.50 mostly observed for the CH air masses.
Figure 8 shows the frequency distributions of the WSOC/OC ratios for each air mass (as classified by the back trajectory). Overall, the WSOC/OC ratios had broad distributions (Figure 8a), suggesting that different sources and/or chemical processes contributed to the OC levels observed at Gosan. The WSOC/OC ratios showed clear differences in each air mass (Figure 8b), with the average ratio higher in CH air masses (0.34 ± 0.13) than in FT (0.26 ± 0.11) and MR (0.11 ± 0.08) air masses (Table 2). The average WSOC/OC ratio in the CH air masses obtained in this work is similar to that found for an urban area of Tokyo in summer (∼0.35) [Miyazaki et al., 2006] and at an urban site in China (∼0.30) [Yang et al., 2005]. The ratio is also within the range of values (0.10–0.50) reported by Mader et al.  for airborne filter samples obtained over downwind regions of China in spring. More comprehensive summaries of WSOC/OC ratios from previous studies available in the literature are given by Mader et al.  and Jaffrezo et al. .
 The WSOC/OC (WIOC/OC) ratio in the FT air masses was relatively smaller (larger) than that in the CH air masses. The mass concentrations of WSOC and SO42− were 60–70% lower in the FT air masses than in the CH air masses (Table 2). These species are susceptible to removal (e.g., by washout), particularly during transport from the boundary layer to the free troposphere [Park et al., 2005; Heald et al., 2006]. In contrast, the concentration of WIOC, which is less susceptible to washout, was lower in the FT air masses by only 20%, which likely resulted in the low WSOC/OC (high WIOC/OC) ratio in the FT air masses.
5.3. Correlations Between WSOC and Chemical Tracers
 This section discusses several chemical tracers used to investigate possible sources of the observed WSOC. Figure 9 shows the relationship between WSOC and CO concentrations, with CO used as an inert tracer of incomplete combustion. Overall, WSOC correlated well with CO concentrations (r2 = 0.72). The correlation coefficient was high in the CH air (r2 = 0.54), while no significant correlation was seen in the MR air (r2 = 0.06). The high positive correlation between WSOC and CO in the well-defined CH air masses suggests a link between combustion sources and the origin of WSOC. The observed slope (dWSOC/dCO) in the CH air masses was 7.3 × 10−3μg C m−3 (ppbv CO)−1. This value is comparable to the slope between OC and CO (11.9 × 10−3μg C m−3 (ppbv CO)−1) within the urban plume of a major city in eastern China, as shown by aircraft measurements [Maria et al., 2003], assuming a WSOC/OC ratio of ∼0.50 [Mader et al., 2004].
 To investigate changes in WSOC mass in the CH air masses, the relative increase in WSOC concentrations compared to that of CO was calculated on the basis of the relationship between WSOC and CO. ΔWSOC and ΔCO denote quantities in which background values of WSOC and CO have been subtracted from observed WSOC and CO values, respectively, and ΔWSOC/ΔCO is differentiated from the slope between the two species (dWSOC/dCO). ΔWSOC/ΔCO is better suited than absolute WSOC concentrations for detecting changes in the WSOC mass of CH air masses because, compared with WSOC, ΔWSOC/ΔCO changes much less through mixing with background air during transport. Background values of 0.2 μg C m−3 and 180 ppbv for WSOC and CO, respectively, were in approximately the lowest fifth percentile of WSOC and CO values for all the air masses. In Figure 10, ΔWSOC/ΔCO and WSOC/OC are plotted as a function of the O3 concentrations for the CH air masses, where O3 is used as an indicator for photochemical processing. In Figure 10 (left), ΔWSOC/ΔCO shows positive correlations with O3 concentrations (r2 = 0.45). The WSOC/OC ratio also tended to increase with increasing concentrations of O3 (Figure 10, right). These positive correlations suggest that a major fraction of the observed WSOC included secondary products formed via photochemical processes, which are also linked to O3 production.
 Previous studies have reported that significant fractions of particles from biomass burning (BB) contain water-soluble organic mass [e.g., Novakov and Corrigan, 1996; Mayol-Bracero et al., 2002]. It is difficult to quantitatively determine the contribution of BB to the observed WSOC from our measurement data alone. However, the weak correlation between WSOC and potassium (K+) (r2 = 0.27; not shown) as a tracer of BB emissions in the CH air masses suggests that BB sources contributed to the observed WSOC to some degree during the study period. This agrees with the results of Han et al. , who inferred that ∼20% of fine aerosol (particle diameters of 0.56–2.5 μm) sampled at Gosan in spring was attributable to BB sources based on various chemical tracers.
6. Thermal Characterization of WSOC
 As described in section 2.2, the thermal characteristics of OC (i.e., the positions of thermogram peaks) were closely related to the MW of WSOC components. Comparison of the measured WSOC with the thermal characteristics of OC, therefore, provides insights into the chemical compositions of the measured WSOC from the viewpoint of their thermal stability.
 It should be noted that carbonate carbon (e.g., calcium carbonate), typically found in dust plumes, was found to peak sharply in the 870°C temperature step in the He mode [Birch and Cary, 1996], although it was a small contributor (3% on average) to submicron OC over the western Pacific during ACE-Asia [Lim et al., 2003]. To reduce the uncertainty of the OC thermogram associated with calcium carbonate, data with Ca2+ concentrations >1 μg m−3 (13% of the total data) were excluded from the current analysis. One may argue that most of the Ca2+ in an aged polluted plume is in the form of Ca(NO3)2 or CaSO4. In these excluded data, the sum of SO42− and NO3− in equivalence was nearly balanced with NH4+ with a slope of 1.09 (r2 = 0.99). The Ca2+ amount was substantially larger than the amount of excess anion(=SO42− − NH4+ + NO3−) available for reacting with Ca2+, with a slope of 3.3 in molar equivalence. This indicates the presence of carbonate in these air masses and that at most 30% of Ca2+ was in the form of Ca(NO3)2 and/or CaSO4.
6.1. Relationship Between WSOC and OC at Each Temperature Step
Figure 11 shows relationships between WSOC and OC1-4 during the study periods. The mean values of OC1-4 are summarized in Table 3. The mean mass concentration of OC4 was the largest (2.0 ± 0.7 μg C m−3) of all the OC components regardless of air mass origin. The average ratio of OC4/OC was as high as 0.41 ± 0.06, while those of OC1/OC, OC2/OC, and OC3/OC were 0.22 ± 0.05, 0.19 ± 0.03, and 0.15 ± 0.02, respectively. The mass concentrations of OC1-4 showed highly positive correlations with those of WSOC (r2 = 0.50–0.74). In particular, the correlation coefficient between OC4 and WSOC (0.74) was much higher than that between OC4 and WIOC (r2 = 0.37; not shown). These results suggest that a large fraction of the observed WSOC was likely associated with thermally refractory compounds.
Table 3. Mean Values and Standard Deviations of OC1-4, PC, and PC/WSOC for the Three Regimes Defined by Back Trajectories
OC1, μg C m−3
OC2, μg C m−3
OC3, μg C m−3
OC4, μg C m−3
PC, μg C m−3
PC/WSOC, μg C μg C−1
1.2 ± 0.2
1.0 ± 0.4
0.8 ± 0.4
2.4 ± 0.7
0.07 ± 0.04
0.04 ± 0.02
Free tropospheric (FT)
0.6 ± 0.1
0.8 ± 0.2
0.6 ± 0.2
1.8 ± 0.5
0.08 ± 0.06
0.13 ± 0.08
0.7 ± 0.2
0.6 ± 0.2
0.4 ± 0.2
1.1 ± 0.3
0.03 ± 0.02
0.09 ± 0.07
1.0 ± 0.3
0.9 ± 0.3
0.7 ± 0.4
2.0 ± 0.7
0.06 ± 0.04
0.07 ± 0.04
 One of the possible uncertainties in defining the peak area of OC4 is mainly introduced by evolution of original EC in the 870°C step of the He mode, caused by oxygen supplied by mineral oxides in the particle mixture [Chow et al., 2001; Subramanian et al., 2006]. This effect could lead to overestimation of OC4 (underestimation of EC). However, as described in section 3.2.3, EC was highly correlated with BC (r2 = 0.97; σBC = 8.9 m2 g−1) with the EC/BC slope being 0.98, suggesting insignificant loss of EC during the thermal analysis. Bond and Bergstrom  suggested a σBC value of 7.5 ± 1.2 m2 g−1 for uncoated BC particles, which is 16% lower than that used in this study. With a σBC value of 7.5 m2 g−1, the average EC/BC ratio decreases to 0.83. Even if we assumed that 17% (from the EC/BC slope) of the original EC evolved in the 870°C step of the He mode, this would contribute a maximum of ∼20% of OC4 based on the average EC/OC4 ratio of 0.74 ± 0.21; OC4 would still be the largest contributor to OC.
6.2. PMF Analysis
 To characterize the observed WSOC and WIOC by a linear combination of each OC component thermogram, we used a positive matrix factorization (PMF) method [Paatero and Tapper, 1994]. PMF is a least-squares-based factor analysis model that numerous studies have used to identify the sources of airborne particles [e.g., Polissar et al., 2001; Hopke, 2003]. Details of PMF can be found elsewhere [Paatero, 1997]. The method does not require a priori information about the concentration profiles (i.e., thermal properties in this study). In the PMF analysis for contribution estimation, the mass concentrations of six species (OC1, OC2, OC3, OC4, WSOC, and WIOC) were input as variables into the PMF model with 196 samples. Note that PMF modeling does not regard these variables as random quantities [Paatero, 1997]. Therefore variables in PMF modeling do not necessarily have to be “independent” values since the terms dependence or independence are used to characterize random variables (P. Paatero, personal communication, 2007).
6.3. Thermal Component Groups
 The PMF analysis found that three factors provided the best solution and reproduced more than 93% of the measured values for each species in the six-variable system. Figure 12 presents profiles of the resolved factors and the contributions of OC1, OC2, OC3, and OC4 for all of the data analyzed. Factor 1 (F1) was strongly influenced by the mass of WIOC (1.2 μg C m−3), followed by OC1 and WSOC. Among OC1-4, OC1 was the largest contributor to F1, suggesting a link of this factor with low refractory or more volatile organic compounds. Factor 2 (F2) was also dominated by WIOC (1.5 μg C m−3), with OC2 and OC3 additionally providing large contributions. On the other hand, there was no contribution from WSOC to F2. Factor 3 (F3) was mainly attributed to the masses of OC4 (0.9 μg C m−3) and WSOC (1.0 μg C m−3), as anticipated from the strong correlation between the two parameters shown in Figure 11 and discussed in section 6.1. On the basis of the contributions of OC1-4 to each resolved factor, F1, F2, and F3 were interpreted as low, medium, and highly refractory compound groups, respectively.
Figure 13 shows time series of the mass contributions of the individual identified factors to the WSOC and WIOC mass concentrations. Table 4 summarizes the average values of each factor's contribution according to the classification of air masses. On average, the low refractory compound group (F1) accounted for 21% of WSOC by mass (Figure 13a). As shown in Figure 2 and discussed in section 2.2, carbon that evolved at lower temperatures (<300°C) was generally associated with low MW compounds under the current operating conditions. Therefore this result implies that the average mass fraction of low MW compounds to the total WSOC was likely small. This finding agrees with previous studies at Gosan. For example, Kawamura et al.  showed that low-MW dicarboxylic acids (C2-C6) accounted for 6% of the total carbon (TC). The mass fraction of water-soluble saccharides with MWs equal to or less than 180 g mol−1 (e.g., glucose and glycerol) was as low as that of dicarboxylic acids at Gosan [Simoneit et al., 2004].
Table 4. Mean Ratios and Standard Deviations of Each Factor to the WSOC and WIOC Mass Concentrations for the Three Categories
Air Mass Category
0.25 ± 0.11
0.75 ± 0.11
0.52 ± 0.17
0.48 ± 0.17
Free tropospheric (FT)
0.09 ± 0.07
0.91 ± 0.07
0.17 ± 0.13
0.83 ± 0.13
0.63 ± 0.23
0.37 ± 0.23
0.29 ± 0.11
0.71 ± 0.11
 The most pronounced result was that the highly refractory compound group (F3) contributed significantly (∼79%) to the WSOC mass concentrations. The fractions of F3 to WSOC were as large as 75% and 91% in CH and FT air masses, respectively (Table 4). From thermograms of WSOC, previous studies inferred that 40–50% of the WSOC was refractory at urban sites in China [Yu et al., 2004; Yang et al., 2005]. The present result indicates that highly refractory compounds were significant contributors to WSOC not only in the polluted CH air masses but also in relatively clean FT air masses. Interestingly, the PC/WSOC ratios of the FT air masses (0.13 ± 0.08) were also larger than those of the CH and MR air masses (Table 3). Considered together with the predominant contribution of F3, WSOC in the FT air masses likely contained a larger mass fraction of organic compounds that were refractory and/or prone to char, compared to the CH and MR air masses. However, those compounds were neither speciated into functional groups nor characterized at the molecular level in this study. Even in the MR air masses, F3 accounted for 37% of WSOC, while the fraction of F1 to WSOC was 63%. The relatively large fraction of F1 in MR air masses can be interpreted as a large fraction of low-MW WSOC in marine aerosols. This is in agreement with results by Cavalli et al. , who reported that low MW monocarboxylic/dicarboxylic acids (MDA) accounted for ∼50% of the total WSOC identified, while polycarboxylic acids (PA) accounted for 22% in submicron marine aerosols sampled over the eastern Atlantic.
 Our results suggest that regardless of air mass origin, significant amounts of the observed WSOC were associated with highly refractory compounds. The highly refractory compounds that evolve at the OC4 stage can be interpreted as (1) carbon that may be originally refractory and associated with high MW compounds and/or (2) carbon that may undergo charring during the thermal analysis followed by evolution in the He mode prior to the He/O2 mode [Yu et al., 2002]. The former interpretation is supported by the experimental results discussed in section 2.2. With regard to the latter interpretation, organic compounds that undergo charring generally include saccharides (e.g., levoglucosan), cellulose [Statheropoulos and Kyriakou, 2000], and other high MW water-soluble organic species [Huebert and Charlson, 2000], some of which were also discussed in section 2.2. These water-soluble organic compounds have high MWs, generally with values in the hundreds. Therefore these results coupled with the results of our laboratory experiments imply that a large fraction of the highly refractory compound groups in the WSOC mass is likely associated with high MW organic compounds. This is supported by the study of Decesari et al. , who showed that polyacids accounted for 33–40% of the sum of identified WSOC in aerosols at Gosan. Their techniques included filter samples analyzed with ion-exchange chromatography and proton nuclear magnetic resonance (1H NMR); the WSOC identified accounted for 88% of the total WSOC in their study. The present results provide valuable clues regarding the sources and formation processes of SOA mass over the Asian region. They may also have important implications for understanding the possible effects of WSOC on their activation to form cloud droplets [Charlson et al., 2001; Dinar et al., 2006]. We note that the data may not be representative for conditions over the region in spring because measurements occurred over an approximately 2-week period and may have been influenced by synoptic-scale events.
 For WIOC, F2 accounted for 61% of WIOC, whereas F1 accounted for 39% on average (Figure 13b). The dominance of F2 and F1 in WIOC indicates that a significant fraction of the observed WIOC was linked to OC2 and OC1, consistent with the results of the laboratory experiment shown in Table 1 and discussed in section 2.2. Previous studies have reported that most WIOC identified from urban sources is composed of the products of incomplete combustion, such as aliphatic hydrocarbons, long-chain ketones, alkanols, and polycyclic aromatic hydrocarbons (PAHs) over the western Pacific [e.g., Simoneit et al., 2004]. In fact, WIOC in CH air masses was positively correlated with CO (r2 = 0.30; not shown). These results suggest that incomplete combustion products were possible sources of the observed WIOC in CH air masses. On the other hand, WIOC showed poor correlations with CO (r2 < 0.01; not shown) in FT and MR air masses. It should be noted that the average WIOC mass concentration in MR air masses (3.0 ± 1.4 μg C m−3) was similar to that in CH air masses (3.2 ± 1.3 μg C m−3), despite clear differences in the levels of anthropogenic tracers (e.g., BC or CO) in the two air masses (Table 2). One possible explanation is that the WIOC in MR air masses may be attributed to organic matter transferred from the ocean surface into the atmosphere [Oppo et al., 1999]. Some evidence points to the occurrence of WIOC in marine aerosols (mainly lipidic species), for which marine biogenic origin has been postulated [e.g., Gogou et al., 1998; Mochida et al., 2002]. Simoneit et al.  reported that 1–45% of the total identified organic compound mass (<16% of OC) sampled over the mid-to-western Pacific was marine-derived lipids, due mainly to fatty acids (<C20). Supporting this hypothesis, WIOC showed a positive correlation with Na+ (r2 = 0.36; not shown) in the MR air masses.
 Semicontinuous measurements of aerosol WSOC and OC in the PM2.5 mode were conducted at Gosan on Jeju Island, Korea, during ABC-EAREX2005. In addition to the field measurements, thermal analyses of organic standards were made in the laboratory for the interpretation of the measured organic compounds. Thermograms of a single standard of water-soluble organic compounds showed that carbon evolved at high temperatures (600°–870°C) was generally associated with high molecular weight (MW) compounds with MWs on the order of hundreds. In contrast, carbon that evolved at low temperatures (below 300°C) was generally associated with organic compounds having MWs lower than ∼180 g mol−1. These results indicate that under the thermal conditions in this study, complete evolution of a single water-soluble organic compound generally required higher temperature conditions with increasing MW. The experiments also showed similar thermogram peaks of multicomponents to those of the sum of individual peaks, suggesting that the observed thermograms can be regarded as a linear combination of the individual species under identical operating conditions. On the other hand, a single standard of water-insoluble organic compounds evolved mostly at temperatures between 300° and 600°C.
 On average, the observed WSOC/OC ratio for all air masses was 0.30 ± 0.12 μg C/μg C at Gosan. The ratio was 0.34 ± 0.13 μg C/μg C in air masses originating from northeastern China, as identified by back trajectories. The WSOC concentrations correlated well with CO (r2 = 0.54) in the Chinese outflow, suggesting that the observed WSOC and/or their precursors were of combustion origin. In the Chinese outflow, the relationship between the relative increase of WSOC and O3 suggests that the observed WSOC was mostly a secondary product.
 A positive matrix factorization (PMF) analysis was applied to the observed mass concentrations of WSOC, WIOC, and thermal components of OC. Three organic compound groups were resolved by PMF and interpreted to be low, medium, and highly refractory compound groups on the basis of the thermal characteristics of OC. On average, highly refractory compound groups were the largest contributor to the WSOC, accounting for 79% of WSOC by mass. Low refractory compound groups accounted for 21% in the air masses observed during the study period. Highly refractory compound groups significantly contributed to WSOC, not only in the Chinese outflow but also in marine-influenced air masses. The combined results of the field measurements and laboratory experiments imply that a large fraction of the highly refractory compound groups was likely associated with high MW compounds at Gosan during the observation period. Medium and low refractory compound groups accounted for 60% and 40% of the WIOC, respectively, consistent with the thermograms of the organic species examined in the laboratory.
 The authors thank M. Kuwata, Y. J. Kim, C.-S. Hong, and other Korean colleagues for their support of the field experiments. R. J. Weber is greatly acknowledged for helpful comments on this paper. We thank J. C. Nam for providing the meteorological data for Gosan. This work was funded by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT).