Current Issue

Carbon-fraction profiles of 43 organic materials, including n-alkanes, polycyclic aromatic hydrocarbons (PAHs), carboxylic acids, fatty acids, fatty alcohols, humic substances, sugars, sugar alcohols, and anhydrosugars (Table 1), were determined by thermal optical analysis. N-alkanes, fatty acids, fatty alcohols, PAHs, and carboxylic acids are known as major components of primary and/or secondary OA and measured widely (Seinfeld and Pandis, 2006; Gelencsér, 2004). Humic substances, sugars, sugar alcohols, and anhydro-sugars are high-molecule materials and originated from natural matter, which are not so measured widely but were prepared for the variety of materials. The availability and solubility of the reagent were also considered. Solutions of the organic materials were prepared as follows: n-alkanes, fatty acids, and fatty alcohols were dissolved in hexane (500 ppm), PAHs were dissolved in acetonitrile (100 ppm), and humic substances were dissolved in aqueous sodium hydroxide (313 ppm for humic acid and 800 ppm for fulvic acid), whereas water was used as the solvent for carboxylic acids (1000 ppm), sugars and sugar alcohols (2000 ppm), and anhydro-sugars (800 ppm for levoglucosan and 2000 ppm for galactosan). All reagents used, except humic substances, were special grade. Humic substances obtained from ando and brown forest soils in Japan (Watanabe et al., 1994) were supplied by the Japanese Humic Substances Society. The sample solutions (10 to 20 μL) were then applied to preheated quartz fiber filters (Pall, 2500QAT-UP, 8 mmφ) and analyzed by the IMPROVE analysis protocol by using a DRI model 2001A carbon analyzer (Chow et al., 1993). Seven fractions were determined: OC1 (120°C), OC2 (250°C), OC3 (450°C), and OC4 (550°C) in a He atmosphere, and EC1 (550°C), EC2 (700°C), and EC3 (800°C) in a 2% O₂ and 98% He atmosphere; however, the organic materials were not expected to include EC. The amount of pyrolyzed organic carbon (OCP) in each filter sample was also estimated by transmittance because of the sample solutions. Transmittance can sense the pyrolysis of the sample solution permeated into the whole cross section of the filter. Each sample was loaded into the analyzer as soon as the solvent had evaporated (evaporation time was no more than 30 min), and the organic solvents (hexane and acetonitrile) did not affect the analysis. The blank was as the same level as preheated filter (below 0.03 μgC) and the adsorption from ambient air was negligible. The analyzed carbon mass was 1.2 to 9.6 μgC except for sugars and sugar alcohols (9 to 20 μgC).

The carbon-fraction profile of each of the 43 organic materials was determined (Fig. 1). Note that the carbon-fraction profiles are analyzed by the IMPROVE protocol. The analyzed carbon amount in each organic material was generally consistent with the calculated amount. The general characteristics of the carbon-fraction profiles can be summarized as follows.

Fig. 1. 
Carbon-fraction profiles of the organic materials. Fulvic and humic acids A and B originated from ando and brown forest soils, respectively.

n-alkanes: OC1 was the dominant fraction in compounds with relatively low carbon numbers such as eicosane (C20), and OC2 was the dominant fraction in those with high carbon numbers such as triacontane (C30), hentriacontane (C31), and tetracontane (C40). These results are consistent with the known greater volatility of n-alkanes with fewer carbons. The proportions of OC3, OC4, and OCP in n-alkanes were extremely low. One of the index which indicate the characteristics of n-alkanes is the carbon preference index (CPI), and the value of CPI is large when biogenic aerosols derived from higher plants are rich due to predominance of odd carbon number n-alkanes, whereas the value of CPI is close to 1 when the contribution of aerosols from fossil fuel burning become larger, which imply that n-alkanes show neither odd nor even carbon number dominance in aerosols from fossil fuel burning (Gelencser, 2004). However, the carbon-fraction profiles of adjacent odd and even carbon number n-alkane pairs, such as pentacosane (C25) and hexacosane (C26), and triacontane (C30) and hentriacontane (C31), were similar.

PAHs: OC1 was the dominant fraction in benzo[a] anthracene (BaA), which has four benzene rings, whereas OC2 was the dominant fraction in the other PAHs, which have five or more benzene rings. Thus, this result is consistent with the known greater volatility of PAHs with fewer benzene rings.

Carboxylic acids: OC4 was the largest fraction in oxalic acid, which is generally the most abundant carboxylic acid in ambient aerosols (Kumagai, et al., 2010; Kawamura and Yasui, 2005; Kawamura and Ikushima, 1993). In the other carboxylic acids, OC3 was the dominant fraction. Unlike the n-alkane and PAH profiles, the carbon-fraction profiles of the carboxylic acids did not show any trend related to carbon number. OCP was also found in the dicarboxylic acids, although in small proportions.

Fatty acids: OC1, OC2, and OC3 were the dominant fractions in the fatty acids, but the proportions of each differed. Palmitic acid (C16 : 0) and stearic acid (C18 : 0) are the main fatty acids found in ambient aerosols (Seinfeld and Pandis, 2006; Gelencsér, 2004). In palmitic acid, the proportions of OC1, OC2, and OC3 were similar, whereas in stearic acid, OC3 was dominant, and OC1 was hardly present. In contrast, the dominant fraction in oleic acid (C18 : 1), an unsaturated fatty acid, was OC2.

Fatty alcohols: The dominant fraction in 1-octadecanol (C18) and 1-tetracosanol (C24) was OC2, whereas both OC3 and OC4 were major fractions in 1-triacontanol (C30). Thus, unlike the fatty acids, fatty alcohols with higher carbon numbers had higher proportions of OC3 and OC4.

Humic substances: The carbon-fraction profiles of fulvic acid were dominated by OC3, OC4, and OCP, and the proportion of OCP was notably higher than that in the other organic materials. The carbon-fraction profiles of humic acid were similar to those of fulvic acid, except that in humic acid the proportion of OCP was lower and the proportion of OC3 was higher. In addition, the humic acid profiles, unlike the fulvic acid profiles, included OC1.

Sugars and sugar alcohols: The carbon-fraction profiles of sucrose, galactose, fructose, mannose, mannitol, and arabitol were very similar, and the OC2 fraction was dominant. OCP was also present in small proportions.

Anhydro-sugars: The levoglucosan profile was dominated by OC2, OC3, and OC4, whereas in galactosan, OC1 was the dominant fraction.

To examine the influence of carbon number, the carbon-fraction profiles of organic materials with the same number of carbons (C18, C24, or C30) among n-alkanes, fatty acids, and fatty alcohols were compared. Among C18 materials, OC1 was dominant in octadecane (n-alkane), OC2 was dominant in oleic acid (fatty acid) and 1-octadecanol (fatty alcohol), but OC3 was dominant in stearic acid (fatty acid). Among C30 materials, OC2 was dominant in the n-alkane (triacontane); OC3 was the dominant fraction, and the proportion of OC1 was relatively high in the fatty acid (melissic acid); and in the fatty alcohol (1-triacontanol), OC3 and OC4 were the dominant fractions. Among C24 materials, similar to C18 materials, OC1 was dominant in the n-alkane (tetracosane), and OC2 was dominant in the fatty alcohol (1-tetracosanol). However, the carbon-fraction profile of the fatty acid (lignoceric acid) was similar to that of the C30 fatty acid (melissic acid), except with a larger proportion of OC3. Relationships among the profiles of n-alkanes, fatty acids, and fatty alcohols were not necessarily consistent, but the carbon-fraction profiles of fatty acids and fatty alcohols tended to include higher temperature fractions compared with those of the n-alkanes. In materials with fewer carbons (C18), the dominant fraction changed from OC1 in the n-alkane to OC2 and OC3 in the fatty acids and fatty alcohol, whereas in materials with more carbons (C30), the dominant fraction changed from OC2 in the n-alkane to OC3 and OC4 in the fatty acid and fatty alcohol.

In this study, a quartz fiber filter that the sample solutions were impregnated (followed by solvent evaporation) was used to determine the carbon-fraction profiles. The difference of the existing state of organic materials such as surface-to-volume ratio between impregnated solutions and airborne particles on a filter could affect the thermal behavior during the thermal optical analysis. A more realistic approach would be collecting particles on a filter produced by atomizing organic compound solutions. As for the sample analysis, the actual temperature at an analyzed filter sample is important to determine the carbon-fraction profiles. The difference between the setting temperature of IMPROVE protocol and the actual temperature at a filter sample is caused by the temperature control and the distance between a thermocouple and a filter sample. In the case of the DRI model 2001A carbon analyzer used in this study, the temperature control of OC1 (120°C) is not so fine compared to other fractions, though the difference was around 3 or 4°C in this study. With regards to the distance, the tip of the thermocouple is situated under the quartz boat that the filter sample is placed in the analyzer and thus the thermocouple is very close to the filter sample (around 2 or 3 mm). The variation of OC1 temperature affect to the reproductivity of OC1 fraction. Oher than that, the broad peak tailing of detector signal affect to the reproductivity of the OC fraction due to the procedure of dividing carbon-fraction by IMPROVE protocol. OC fractions are divided when the slope of the detector signal is closer to flat no matter whether the detector signal recover to the baseline or not, and thus the variation of dividing involve the following carbon fraction. Moreover, it is unavoidable that the relative standard deviation (RSD) of the minor carbon fraction (the fraction with low carbon mass) become larger (~150% in this study). However, the reproductivity of the minor carbon fraction less affect the carbon-fraction profile, because in this study the pattern similarity of the carbon-fraction profile is important (Section 3) and the major carbon fraction contributes to it much more than the minor carbon fraction. Several patterns of the carbon-fraction profile such as single fraction dominant (eicosane, tetracontane), double fraction dominant (hexacosane, oxalic acid, malonic acid, acetic acid, humic acid and sucrose), and triple fraction dominant (fulvic acid and levogulcosan) were examined, and the RSDs of the these major carbon fraction by 4 or 5 repetitions were relatively small such as OC1 of eicosane (1%), OC1 and OC2 of hexacosane (8% and 23%), OC3 and OC4 of oxalic acid (18% and 8%), OC3, OC4 and OCP of fulvic acid (7%, 12% and 23%). Among those fractions OC2 of hexacosane (23%) and OCP of fulvic acid (23%) were affected by the previous broad peak tailing. Note that the relatively worse RSD was found on levogulcosan (OC2 15%; OC3 23%; OC4, 25%).

For the CMB analysis, the carbon-fraction profiles of the 43 organic materials were first categorized into five types based on the results of a cluster analysis (Table 1). The cluster analysis was carried out by Ward’s method using normalized data and squared distances. Type I comprised OC1-dominant organic materials, namely, C18-26 n-alkanes, the PAH with four benzene rings (BaA), and galactosan. Type II comprised the OC2-dominant organic materials, C30-40 n-alkanes, PAHs with five or more benzene rings, oleic acid (C18 : 1 fatty acid), C18 and C24 fatty alcohols, the sugars and sugar alcohols, and levoglucosan. Thus, n-alkanes and PAHs were categorized into Type I or II depending on the number of carbons or number of benzene rings, respectively. Type III comprised OC3-dominant organic materials, namely, the carboxylic acids (except for oxalic acid), the fatty acids (except for C18 : 1), and humic acid. Type IV contained only oxalic acid (dominantly OC4), and Type V contained only fulvic acid (which included OCP as a major fraction). Although 1-triacontanol was also categorized individually, this was excluded from the CMB analysis, because the number of types must be five in order to solve the 5 by 5 matrix calculation in the CMB analysis and because 1-triacontanol is less abundant than either oxalic acid and fulvic acid in atmospheric aerosols.

Next, five carbon-fraction profile patterns were obtained by averaging the profiles of the organic materials in each type (Fig. 2); the standard deviations were used as the pattern uncertainties in the CMB analysis. Then, a CMB analysis using the five carbon-fraction profile patterns of the organic materials and particulate OC data from various kinds of emission sources was carried out. For this analysis, CMB 8.0 of the U.S. Environmental Protection Agency and an add-in software routine programmed in VBA for Microsoft Excel were used. Mathematical principle of the CMB model is simultaneous equations shown briefly as below.

Fig. 2. 
The five patterns of carbon-fraction profiles determined by averaging the profiles of the organic materials in each type.

C represents measured concentration of OC, and C_i represents measured concentration of the i-th OC fraction. S_j is the contribution of the j-th type of carbon-fraction profile. A_ij is the fractional abundance of the i-th OC fraction in the j-th type of carbon-fraction profile. In this section C_i was normalized to unity because only the fraction profile needs to be obtained. The calculations were carried out by the effective variance least-squares method with elimination of negative contribution.

For the particulate OC data on emission sources, measurements of particulate emissions in the Tokyo metropolitan area (Tokyo Metropolitan Government, 2012a) were used in the CMB analysis. The emission sources, including fossil fuel combustion, biomass burning, cooking, and soil and road dust, are summarized in Table 2. Particles in the primary emissions were collected from flues by an Andersen stack sampler with a PM_2.5 cut-off. In addition, condensable particles were secondarily produced from gaseous materials by the air dilution method (dilution ratio 20) with a 16.7-L residence chamber (residence time 10 s at ambient temperature), and were collected on a quartz fiber filter. Sample air from flues was first passed through a PM_2.5 cyclone; thus, fine primary particles took a role as the condensation cores. The exhaust gas temperature was above 100°C (107 to 313°C) for natural gas, heavy oil, and incinerators, whereas below 100°C (56 to 81°C) for biomass and furnaces of steel and glass scraps (Tokyo Metropolitan Government, 2012b). The temperature at the gas meter of condensable particle sampling was around the ambient air temperature (8 to 44°C, depending on the season and the environment of the sampling place). Automobile exhaust emissions were sampled by using a chassis dynamometer, which collects primary particles only without size separation, and no condensable particle data were available for kitchen emissions. Further, only primary soil and road dust particles were collected by an air sampler in a small chamber by the resuspension method. The carbon-fraction profiles of emission source particles were determined by the thermal optical method in the same way as the organic materials (Section 2). The emission source particles were measured once for each emission source, and thus the measured data was given as it is, and a very small value (10^-3) was uniformly given as the standard deviation due to avoiding impossibility of calculation with no deviation. Note that it is very rare for chemical data on size-separated primary particles and secondarily produced condensable particles from various kinds of emission sources to be measured by the same method, even if the measurement for each emission source is only once.

The results of the CMB analysis for particulate OC of the emission sources are shown in Fig. 3. The emission sources are roughly classified by fuel or burning material as following subsections.

Fig. 3. 
Percentages of the five types of organic materials in particulate OC of the emission sources listed in Table 2 for (a) primary and (b) condensable particles, estimated by the CMB analysis. An asterisk denotes particles not separated by size.

Ambient PM_2.5 samples were collected with a low-volume air sampler (Thermo, FRM-2025) on preheated quartz fiber filters (Pall, 2500QAT-UP, 47 mmφ) during 24-h sampling periods. The sampling was carried out at Kazo, a suburban site 50 km from central Tokyo, from October 2013 to August 2014, and around 20 samples were collected during each season. A honeycomb activated carbon denuder (Thermo, 3500) was used to prevent positive sampling artifacts.

OC and EC in the samples were analyzed by the thermal optical method with a DRI model 2001A carbon analyzer (Chow et al., 1993), and the IMPROVE analysis protocol was applied as described in section 2.1. The following seven fractions were estimated: OC1 (120°C), OC2 (250°C), OC3 (450°C), and OC4 (550°C) in a He atmosphere, EC1 (550°C), EC2 (700°C), and EC3 (800°C) in an atmosphere of 2% O₂ and 98% He. OCP was estimated from the reflectance of each filter sample.

The CMB analysis was carried out as described in section 3.1, using the averages and standard deviations calculated from the measurements obtained for each season.

The χ² and r² in the CMB analysis results were satisfactory (χ²≦0.1 and r²≧0.99) for all seasons; thus, the fit of the model to the measurement data was good. In all seasons, Type V accounted for about 50% or more of all OC, followed by Type II (30-40%). The organic material of Type V was fulvic acid alone in this study, but only one study has reported fulvic acid concentrations in aerosols in Japan (Yamanokoshi et al., 2014). According to Yamanokoshi et al. (2014), the average concentration of fulvic acid in aerosol particles with diameters smaller than 10 μm measured in Tokyo ranged from 0.6 μg/m³ (summer) to 1.3 μg/m³ (autumn) (from 0.3 to 0.6 μg/m³ as OC). For comparison, the average OC concentration in PM_2.5 in Tokyo during the four seasons ranged from 2 to 3 μg/m³. Although the measured particle size fraction and the measurement period and site were not the same for fulvic acid and OC, these results indicate that fulvic acid accounts for roughly 20% of OC in aerosols in Tokyo. Therefore, the estimated contributions of Type V at Kazo in this study are different from the fulvic acid contributions reported by Yamanokoshi et al. (2014) for aerosols measured in Tokyo. Yamanokoshi et al. (2014) have suggested that HULIS in Tokyo are produced by secondary photochemical reactions in summer and by biomass burning in autumn. In summer, concentrations of photochemical oxidants are higher at Kazo than in Tokyo because Kazo is located downwind of Tokyo, whereas in autumn, biomass burning is conducted much more frequently around Kazo than in Tokyo. Therefore, the concentration of fulvic acid in aerosols may be higher at Kazo than in Tokyo. Type V was a major component of both primary and condensable particles from biomass burning (Section 3). However, the carbon-fraction profile of Type V includes a high proportion of OCP (Fig. 2), and other organic materials with a high proportion of OCP may not have been considered as Type V in this study. Therefore, the carbon-fraction profiles of more organic materials need to be investigated.

Fig. 4. 
Percentages of the five types of organic materials in ambient particulate OC measured in Kazo, estimated by the CMB analysis.

Type II is composed of organic materials that originated from fossil fuel combustion (n-alkanes and PAHs), biomass burning (levoglucosan and sugars), and biogenic emissions (sugars and sugar alcohols). Fossil fuel combustion emissions likely contributed to the aerosols collected during all four seasons, whereas biomass burning is conducted in a wide area around Kazo mainly in autumn. Biogenic emissions generally increase during the season when vegetation is growing. Theodosi et al. (2018) have reported that mannitol concentrations, as well as glucose, fructose, and sucrose concentrations, increased throughout the growing season in aerosols sampled at a remote site in Crete. The seasonal differences in the Type II contribution may reflect increased biogenic emissions in summer (37%) and biomass burning emissions in autumn (34%). The percentage contributions of Type I, which were found only in autumn and winter, are reasonable, because the relatively more volatile materials of this type would form aerosol particles by condensation at low temperatures. Type III was found only in winter, and the highest Type IV contribution was also seen in winter. Type III consists of a mixture of various materials, mainly acids, and Type IV consists only of oxalic acid.

To verify the estimated organic matter composition, the comparison with the measured organic materials of the ambient samples is needed. The estimated organic matter composition is desired to be similar to seasonal trend of the measured organic materials that represent each type. However, considering that carboxylic acids are mostly produced photochemically, it would be difficult to explain the seasonal trends of Type III and IV. Also, the carbon-fraction profiles of more organic materials are needed to be analyzed in order to make better estimation of organic matter composition, especially because the only one compound was assigned to Type IV and V of the carbon-fraction profiles. In addition, as a limitation of this study, the carbon-fraction profiles are usable only for EC/OC measurements performed by IMPROVE protocol, since the database of emission source particles were compiled by IMPROVE protocol (Section 3), and ambient particles mostly analyzed by IMPROVE protocol in Japan. Different temperature protocols like a NIOSH (Peterson and Richards, 2002), EUSAAR2 (Cavalli et al., 2010) or newly used IMPROVE_A (Chow et al., 2007) have different temperature steps and so logically different carbon-fraction profiles. Nonetheless, this study would offer some clues of understanding the organic composition.