Chemical Characterisation of Organic Functional Group Compositions in PM 2 . 5 Collected at Nine Administrative Provinces in Northern Thailand during the Haze Episode in 2013

During the first two weeks of March 2013, agricultural waste burning, coupled with forest fires in northern regions of Thailand, steadily enhanced the levels of air pollution above the safety limit and produced an eyestinging, throat-burning, yellow-tinged haze that reduced visibility to less than 1,000 metres. A profusion of fine particles, measuring less than 10 micrometres in diameter and known as PM10, pervaded the atmosphere, reaching a peak on March 21st, 2013, at 428 μg m-3 at Mae Hong Son province. This value was over 3.6 times the acceptable safety ceiling of 120 μg m-3 and eventually prompted authorities to issue warnings against outdoor exercise. The monthly average concentrations of PM10 in March 2013 detected by the Pollution Control Department (PCD) of the Ministry of Natural Resources and Environment (MNRE) at Mae Hong Son (186±110 μg m-3), Chiang Rai (122±71


Introduction
During the first two weeks of March 2013, agricultural waste burning, coupled with forest fires in northern regions of Thailand, steadily enhanced the levels of air pollution above the safety limit and produced an eyestinging, throat-burning, yellow-tinged haze that reduced visibility to less than 1,000 metres. A profusion of fine particles, measuring less than 10 micrometres in diameter and known as PM 10 , pervaded the atmosphere, reaching a peak on March 21 st , 2013, at 428 µg m -3 at Mae Hong Son province. This value was over 3.6 times the acceptable safety ceiling of 120 µg m -3 and eventually prompted authorities to issue warnings against outdoor exercise. The monthly average concentrations of PM 10 in March 2013 detected by the Pollution Control Department (PCD) of the Ministry of Natural Resources and Environment (MNRE) at Mae Hong Son (186±110 µg m -3 ), Chiang Rai (122±71 whereas smaller particles are called "fine". The three modes correspond to the nuclei mode (particles below 0.1 µm), accumulation mode (0.1<D p <1 µm), and coarse mode (D p >1 µm) (Whitby and Sverdrup, 1980). Thus, fine particles include both the accumulation and nuclei modes. Because the definition of modes has been based only on the mass (or volume distribution) of particles, the boundaries between these modes are not precise. The location of modes is dependent upon whether the modes are based on the particle number or surface distribution. In contrast, aerosols in rural areas are primarily of natural origin but are moderately influenced by anthropogenic sources (Hobbs et al., 1985). Recent studies have shown an association between PM 2.5 and adverse health effects, such as lung cancer (Pope et al., 2002) and heart attacks (Brook et al., 2002). Large particles (D p >20 µm) deposit in the nasal passages, where they are removed by sneezing or swallowing. Finer particles, approximately 10 µm in diameter, can enter the respiratory system, and particles with diameters of less than 2.5 µm can penetrate even deeper into the lungs to the alveolar region, where they can remain for a long period of time (Harrison and Perry, 1986). In the urban atmosphere, polycyclic aromatic hydrocarbons (PAHs) are strongly associated with fine particles (Allen et al., 1996;Cecinato et al., 1999), although little data are available concerning the distribution of oxy PAHs within particulate matter (Allen et al., 1997). It is well known that PAHs are considered to pose potential health hazards because some PAHs are known carcinogens (Braga et al., 1999;Farmer et al., 2003). At ambient temperatures, two-ring naphthalene exists almost entirely in the gas phase, whereas five-ring PAHs and higher-ring PAHs are predominantly adsorbed onto particles. The intermediate three-and four-ring PAHs are distributed between the two phases (Zielinska et al., 2004).
Studies on the distribution of PAHs within the full size range of urban atmospheric particles found that 97% of the PAHs with five or more rings, which were all more or less completely particle-bound in the atmosphere, were associated with particles with aerodynamic diameters (d ac )<2.9 µm (Kauup et al., 2000). In the urban aerosols near the road measured in Guangzhou, China, approximately 57% of n-alkane and 62% of PAHs were found in the particle fraction with diameters<0.49 µm (Bi et al., 2004). The substance loading decreased steadily with increasing particle size. An average of only 1.2% of PAHs with five or more rings were found on large particles with d ac >8.6 µm. In Pasadena, California, Miguel and Frielander (1978) reported that 72.7% and 76.8% of B[a]P and 75.7% and 93% of Cor sampled between October and December 1976, respectively, were present in particles with aerodynamic diameters in the range of 0.05-0.26 µm. In Hamilton, Ontario, the total particulate PAHs ranged between 87% and 95% and were found on particle sizes of 1.1 µm<D<7.0 µm (Katz and Chan, 1980). The relatively high proportion of particulate PAHs (63% to 85%) measured in London were found in aerosols with aerodynamic diameters of less than 1.1 µm, and up to 95% of particulate PAHs in London were associated with particles under 3.3 µm in diameter (Baek et al., 1991).
Unlike in other developed countries, there are a limited number of studies concerning the chemical characterisation of gaseous species and aerosols in Thailand (Thumanu et al., 2009;Pongpiachan, 2013a;2013b;Pongpiachan et al., 2010Pongpiachan et al., , 2012a2012b;2012c, 2013a2013b). In this study, the authors postulate that the use of Fourier Transform Infrared Spectroscopy (FTIR), combined with numerous analogies of statistical analysis, assist in a better understanding of the distribution pattern of organic functional compositions of PM 2.5 , which can be subjected to variations in sources and meteorological conditions in northern Thailand. It is the objective of this study to demonstrate the general application of organic functional group analysis by FTIR as an innovative indicator to chemically characterise PM 2.5 at nine provinces in the northern region of Thailand "before" and "after" the haze episode in 2013. In addition, the incremental lifetime exposure to PM 2.5 and the application of FTIR spectral features as an alternative "Biomass Burning" proxy will be reviewed and discussed.

Monitoring sites and sampling period
Monitoring of PM 2.5 was conducted at nine locations and over several time periods. In all instances, the duration of each sample collection was 24 h. The sampling campaigns can be grouped according to the observation period during which they were conducted: campaign І was carried out before the "haze episode" in the winter of 2012 (i.e., from the 7 th to 22 nd of December 2012), whereas monitoring during campaign П was conducted in March 2013 (i.e., from the 4 th to 19 th of March 2013; see Table 1). Both monitoring campaigns were performed at nine observatory sites, namely, Chiang-Rai Province Observatory Site ( Figure  1). There were no obstructions in the vicinity of the sampling equipment, which was strategically positioned to be accessible to winds from all directions. Monitoring at group-1 monitoring stations (CROS, PYOS and NPOS), group-2 monitoring stations (LMOS, PHOS and UTOS) and group-3 monitoring stations (MHOS, CMOS and LPOS) was conducted synchronously every day from the 28 th of November to the 4 th of December 2012, from the 7 th to 13 th of December 2012 and from the 16 th to 22 nd of December 2012, respectively, as summarised in Table 1.

Sampling equipment
MiniVol™ portable air samplers (Airmetrics) were used to collect PM 2.5 over a 24 h period at nine sampling sites. The MiniVol's pump draws air at 5 litres minute -1 through a particle size separator (impactor) and then through a 47 mm filter. A 2.5 micron particle separation is achieved by impaction. Neither PM 10 nor TSP samples were collected for this study.

FTIR analysis
It is well known that Fourier Transform Infrared (FTIR) spectrophotometry provides at least three major advantages. These advantages include a multiplex advantage (i.e., the accumulation of the results is accomplished by measuring a spectrum of all wavenumbers with relatively high scanning speed); an aperture advantage (i.e., the sensitivity of FTIR results depends on the aperture area and the incident angle of light); and a wavenumber accuracy advantage (i.e., the laser emits extremely stable monochromatic light, generating a spectrum with high wavenumber accuracy). In this study, IR-Affinity-1 Shimadzu was chosen because it possesses the following traits: (i) high measurement sensitivity, (ii) measurement of samples with low transmittance, (iii) higher speed measurement, (iv) higher wavenumber accuracy and (v) highly accurate spectrum subtraction.
The process of FTIR transmission spectroscopy involves detecting the absorption of an infrared beam that is passed directly through the sample and through Quartz Fibre Filters (QFFs). Because the resulting absorption bands are unique to specific functional groups and because these bands are proportional to the amount of sample present, a functional group analysis can be performed using FTIR to provide quantitative and qualitative information. Subtraction of the blank (i.e., the spectrum of the empty QFFs) from the filtered sample was performed for each sample. Each spectrum was determined by averaging 16 scans at a resolution of 2 cm -1 . The spectra describe the absorbance of radiation as a function of wavenumber, which ranges from 4,000-400 cm -1 . For more details on the quantification of identified functional groups in PM 2.5 samples collected on QFFs, please refer to a previous publication by Krost and McClenney (1994).

Statistical analysis
SPSS version 13 was used to calculate the average values, standard deviations and principal component analysis with Varimax rotation of the data set.

Characteristics of FTIR spectra and their first derivatives
As illustrated in Figure 3 and Figure 4, PM 2.5-before the "Haze Episode" has more complicated FTIR spectra coupled with their first derivatives than those of PM 2.5-after the "Haze Episode" . The main IR absorption band of PM 2.5-before the "Haze Episode" in samples collected at PYOS, CMOS, NPOS and MHOS appears at approximately 405 cm -1 , 430 cm -1 , 463 cm -1 and 496 cm -1 , respectively. These FTIR spectra characteristics reflect more complicated chemical mixtures released from the unique emission sources of each province before the "Haze Episode". In normal conditions, both "anthropogenic activities" and "natural emissions" appear to play an important role in governing the organic functional group content of PM 2.5 . In contrast, the major IR absorption band of PM 2.5-after the "Haze Episode" in samples collected at MHOS, PYOS, PHOS and CROS occurs at approximately 405 cm -1 , 409 cm -1 , 409 cm -1 and 486 cm -1 , respectively. These results indicate that PM 2.5-after the "Haze Episode" is comparatively more homogeneous than PM 2.5-before the "Haze Episode" . This homogeneity was possibly caused by the overwhelming amount of biomass-burning aerosols present in the nine administrative northern provinces after the "Haze Episode".

Occupational exposure of outdoor workers to PM 2.5
To assess the health risks associated with the occupational exposure of outdoor workers to PM 2.5 , the incremental lifetime particulate matter exposure (ILPE) model was employed and defined as follows:
According to the methods for the derivation of inhalation dosimetry (US EPA, 1994), the inhalation rates of male and female outdoor workers were estimated to be 0.89 and 0.49 m 3 h -1 , respectively. The ILPE model was adapted from the probabilistic incremental lifetime cancer risk (ILCR) model, which was used to assess traffic policemen's exposure to PAHs during their work in China (Hu et al., 2007). The estimated ILPE levels in outdoor workers are summarised in Table 2. The predicted ILPE of PM 2.5 was consistently highest in both genders at MHOS after the "Haze Episode", with average values of 7,004±3,059 mg and 3,856±1,684 mg for PM 2.5

Figure 2. Spatial and Temporal Distribution of Wind Directions and Hot Spots in Thailand from 28 th February to 30 th March 2013 (ASEAN Regional Specialized Meteorological Centre, Singapore)
accumulated in male and female workers, respectively, over an exposure duration of 25 years. These increased concentrations of PM 2.5 after the "Haze Episode" may have several explanations, including not only vehicular and other industrial emissions but also agricultural waste burning in suburban regions. It is well known that uncontrolled biomass burning and forest fires have been found to be significant sources of ambient PM 2.5 in Mae Hong Son and other northern cities over recent decades, particularly during cold periods. An inversion is more likely to arise during the winter when the angle of the sun is very low, predominantly in mountainous provinces such as Mae Hong Son, and could theoretically be accountable for the rise of PM 2.5 content during the observation periods from the 13 th to 19 th of March 2013. The relatively low ILPE of PM 2.5 observed at CMOS (i.e., 2,348±1,200 mg and 1,292±661 mg for males and females, respectively) highlights the effects of anti-burning and smog prevention programs proactively enacted by local administration organisations since November 2011. It is also interesting to note that the ILPE of PM 2.5 at CROS after the "Haze Episode" (i.e., 3,064±1,031 mg and 1,687±568 mg for males and females, respectively) is approximately six times higher than that during normal conditions (i.e., 500±203 mg and 275±112 mg for males and females, respectively). This difference in the ILPE of PM 2.5 at CROS can be explained by the occurrence of comparatively more hot spots during the monitoring period after the "Haze Episode" (Figure 2).

Pearson correlation analysis
Pearson correlation coefficients of averaged FTIR spectra of PM 2.5 collected before and after the "Haze Episode" are displayed in Table 3. Only three combinations, namely CROS-LMOS (R=0.83), CROS-LPOS (R=0.88) and LMOS-LPOS (R=0.71), have R-values higher than 0.7 before major biomass burning events (Table 3) were observed after the forest fire events. As previously mentioned, these findings are consistent with the FTIR spectra results together with their corresponding first derivatives, emphasising the strong dominance of biomass burning over the chemical characteristics of organic functional groups in PM 2.5 (Figure 3 and Figure 4). It is worth mentioning that the R-values higher than 0.7 are combinations of administrative provinces that share the same border lines and/or vegetation types. In contrast, relatively low R-values (less than 0.3) were observed at administrative provinces that have strong dissimilarities in both geographical conditions and forest types, such as CMOS-MHOS (R=0.20), MHOS-PYOS (R=0.25), MHOS-PHOS (R=0.25) and MHOS-UTOS (R=0.22) (see Table 3).

Principal component analysis
PCA is employed as a multivariate analytical tool    :http://dx.doi.org/10.7314/APJCP.2013.14.6.3653 Chemical Characterisation of PM 2.5  to reduce a set of original variables (measured FTIR spectra in PM 2.5 samples) and to extract a small number of latent factors (principal components, PCs) to analyse relationships among the observed variables. Data submitted for this analysis were arranged in a matrix, where each column corresponds to one FTIR spectrum and each row represents the number of samples. Data matrixes were evaluated through PCA, allowing the summarised data to be further analysed and plotted. To enable further interpretation of potential sources of organic functional groups in PM 2.5 , a PCA model with three significant principal components (PCs) was calculated. Each PC of this model represented 43.06%, 13.67% and 9.85% of the variance, thus accounting for 66.58% of the total variation in the data.
These correlation coefficients of each PC are quite useful and provide valuable information that can be used to identify biomass-burning aerosols. However, these correlation coefficients, as well as source profiles, should be used with great caution because physiochemical processes can alter chemical composition distribution patterns during transport from the emission source to the receptor site. To minimise the above-mentioned uncertainties, the plots of three-dimensional PCs can be used as a tool to characterise aerosol types during the sampling period. The clearest features displayed in Figure 5 are described as follows: (i) 3D plots of B-CMOS, B-MHOS, B-PYOS, B-NPOS and B-PHOS, suggesting that the majority of PM 2.5 before the "Haze Episode" had similar organic functional compositions; (ii) 3D plots of groups of PM 2.5 after the "Haze Episode" are highly clustered, indicating that the majority of PM 2.5 shared comparable chemical compositions during the biomass-burning events; (iii) there are distinctly different sources of PM 2.5 before and after the "Haze Episode", indicating that FTIR spectra, coupled with PCA, can successfully distinguish "background aerosols" from "biomass-burning aerosols"; and (iv) the 3D plot of the B-LPOS strongly deviates from the group in both sampling campaigns, implying that other unique local emission sources might be significant in Lamphun province.
In conclusions, FTIR spectra of PM 2.5 collected before and after the "Haze Episode" at nine administrative provinces in northern Thailand were analysed by IR-Affinity-1 Shimadzu. In general, PM 2.5-before the "Haze Episode" samples have more complicated FTIR spectra than those of background aerosols, emphasising that "Biomass Burning" played an important role in governing aerosol chemical compositions in March 2013. The predicted ILPE of PM 2.5 indicated maximum values at MHOS in both genders after the "Haze Episode", with average values of 7,004±3,059 mg and 3,856±1,684 mg for PM 2.5 accumulated in male and female workers, respectively, over an exposure duration of 25 years. The relatively low ILPE of PM 2.5 observed at CMOS (i.e., 2,348±1,200 mg and 1,292±661 mg for males and females, respectively) highlights the effects of anti-burning and smog prevention programs proactively enacted by local administration organisations since November 2011. In conclusion, 3D plots of PCA, coupled with FTIR spectra, successfully discriminated "Biomass-Burning Aerosols" from "Background Aerosols", suggesting that FTIR analysis can be used as an alternative tool to characterise types of aerosols.