Investigation of Biomass Burning Chemical Components over Northern Southeast Asia during 7-SEAS / BASELInE 2014 Campaign

This study investigates the chemical components of biomass burning (BB) aerosols obtained from Doi Ang Khang (DAK; near BB source) and Chiang Mai University (CMU; an urban location) over northern Southeast Asia in dry season (March to mid-April) 2014. PM2.5 (particulate matter with an aerodynamic diameter less than or equal to 2.5 μm) samples were collected over a 24-h sampling period as a part of the Seven South East Asian Studies (7-SEAS)/BASELInE (BB Aerosols & Stratocumulus Environment: Lifecycles & Interactions Experiment) campaign. The collected aerosols were analyzed for mass concentrations of ions, metals and levoglucosan. The influence of air mass movements on aerosol species was also analyzed. The average PM2.5 mass concentrations at DAK (80.8–83.3 μg m) and CMU (90.7–93.1 μg m) were not significantly different (p > 0.05) and well correlated (r = 0.8), and likely originated from similar source origins. The number of fire hotspots was particularly high during 20–21 March (greater than 200) and, consequently, peaks of PM2.5 were recorded at both sites. The most abundant elements at both sampling sites were K (49–50% of total elements), Al (26–31%), Mg (16%) and Zn (4–7%), whereas SO4 (30–38% of total ions), NO3 (13–20%), Na (16–20%) and NH4 (14–15%) were the most abundant ions. Concentrations of levoglucosan and K (BB tracers) were well correlated (r = 0.5 for CMU and 0.7 for DAK) confirming that the PM2.5 detected in these areas were mainly influenced by BB activity. Principal component analysis (PCA) revealed that BB, road traffic, agricultural activity and soil re-suspension were plausible sources of PM2.5 over the study locations. Apart from local sources, the influence of long-range transport was also investigated by way of three-day backward trajectory analysis.


INTRODUCTION
Atmospheric aerosols play crucial roles in the earthatmosphere system and atmospheric chemistry (Andreae and Crutzen, 1997).Aerosols originate from both natural (e.g., oceans, volcanic eruptions, wind-blown dust, and forest fires) and anthropogenic (e.g., industrial activities, traffic emissions, fossil-fuel and open burning) sources.Biomass burning (BB) is a significant source of trace gases and particulate matters (PM) in the troposphere (Chang et al., 2015;Tian et al., 2015).Over Southeast Asia, this activity usually peaks in the dry season (Huang et al., 2013;Hyer and chew, 2010;Permadi and Kim Oanh, 2013;Mahmud, 2013), which includes both forest fires and open burning of crop residues for preparation of planting areas (Chantara et al., 2012;Kim Oanh et al., 2011;Kim Oanh and Leelasakultum, 2011).BB emissions severely affect human health (IPCC, 2007;Chuesaard et al., 2014;Sevimoglu and Rogge, 2015), as well as regional/global climate (Arola et al., 2007;Pratt et al., 2010;Alonso-Blanco et al., 2014).Chemical profiling of BB aerosols is essential to distinguishing various source origins, and can be used to evaluate effects on health and climate (Lee et al., 2016).
Several extensive studies have been conducted to characterize BB aerosols world-wide (e.g., Popovicheva et al., 2015;Lee et al., 2016 and references therein), including studies over the Indochina region: Myanmar, Thailand, Laos, Cambodia, and Vietnam (e.g., Lee et al., 2011;Chuang et al., 2013;Lee et al., 2016).Chuang et al. (2013) characterized the aerosol chemical properties from near-source BB in northern Indochina during the Seven South East Asian Studies (7-SEAS)/Dongsha Experiment, and reported a significant contribution by the smoldering state of softwood burning.Aerosol chemical profiling of near-source BB smoke in Sonla, Vietnam has been also characterized, and the dominance of carbonaceous contents is reported from 7-SEAS Campaigns in 2012 and 2013 (Lee et al., 2016).The signature of BB aerosols at a receptor site in East Asia (i.e., Mt.Lulin), due to long-range transport from the Indochina region, has been explained and verified by longterm observations (Chuang et al., 2014) and also by modelsimulated results (Chuang et al., 2016b).Pani et al. (2016a) recently investigated the radiative impact of upper-layer BB aerosols from the Indochina region over northern South China Sea.Moreover, Pani et al. (2016b) reported a detailed estimation of direct aerosol radiative effect for near source BB aerosols over northern Indochina during the spring of 2013.An accurate understanding of aerosol chemical characterization is crucial for accurately estimating aerosol and climatic impacts.
The chemical composition (i.e., carbonaceous fractions, water insoluble/soluble ions and heavy metals) and size distributios of BB aerosols mainly depend on the burning materials, type, and topography of the burning location and local meteorological conditions.All types of BB have common tracers, such as the potassium ion (K + ) (Cachier et al., 1991), levoglucosan, mannosan and galactosan (Jung et al., 2014).Moreover, soil heating will occur during open burning.When the soil temperature is sufficiently elevated, a greater number of particles and chemicals, such as NO x , N 2 O, Al 2 O 3 , SiO, CaO, FeO, Fe 2 O 3 and TiO 2 will be released from the soil into the air (Williams et al., 1992;Jimenez et al., 2006).
Upper northern Thailand has been facing severe air pollution annually during dry season (February-March) due to intensive BB activities over local and neighboring regions.Emission inventory studies over Chiang Mai (CM) City conducted by the Pollution Control Department in 2002 revealed that 97% of total PM originated from various local sources (i.e., 89%, 5.4% , and 2.3% from forest fires, solid-waste burning and agriculture-residue burning, respectively) and only 2.5% from longer-range sources (Kim Oanh and Leelasakultum, 2011).A report from the Land Development Department (http://www.ldd.go.th/),Thailand in 2010 showed that about 80 % of the land-use pattern in Chiang Mai Province was forest area, and that paddy and crop fields contributed only 5.5% and 4.5%, respectively.Sillapapiromsuk et al. (2013) reported that BB emissions (About 705 tons of PM 10 in 2011) over the region were higher from forest burning (79%), followed by crop-field (15%) and rice-field (5%) burning.Kim Oanh and Leelasakultum (2011) also reported that 80% of northern Thailand hotspots in the 2007 dry season were detected in forest areas and 20% in the agricultural areas (mixed swidden cultivation, paddy field, and corn plantation).
7-SEAS/BASELInE (BB Aerosols & Stratocumulus Environment: Lifecycles & Interactions Experiment) was conducted in spring 2013-2015 over northern Southeast Asia to explore impacts of BB aerosols on atmospheric processes (Lin et al., 2013).This study has proven the most extensive campaign ever conducted for characterizing the physical and chemical properties of BB aerosols in northern Southeast Asia.This study aimed at finding signature chemical profiles from the near-source BB aerosol, which references the levoglucosan, elemental and ion compositions of BB aerosols in PM 2.5 collected over two different locations (i.e., at near-source regions of BB activities and an urban location) in northern Southeast Asia.Statistical approaches were applied to compare the aerosol chemical profiles.Moreover, identification of different source regions contributing to the BB aerosol samples was done by three-day back-trajectory analysis.

Sampling Sites
As shown in Fig. 1, two sampling sites in Chiang Mai Province, northern Thailand were selected to represent near-source and urban areas.The first sampling site was located at a meteorological station in Doi Ang Khang (DAK), Fang District (19°55ꞌ57.61ꞌꞌN,99°2ꞌ42.61ꞌꞌE,1,534m above mean sea level; MSL).The site was near the Myanmar border, about 125 km north of Chiang Mai City, and situated on top of a mountain surrounded by forest and agricultural fields.In dry season, open burning is intensive in the area nearby.Therefore, the site was selected to represent a near-source sampling site for BB.The second site was on the rooftop of the four-story Research Institute for Health Science (RIHS) building at Chiang Mai University (CMU;18°47ꞌ43.63ꞌꞌN,98°57ꞌ28.17ꞌꞌE,354 m MSL).This site is in Chiang Mai City, representing an urban site.

PM 2.5 Sampling and Analysis
PM 2.5 samples were collected in the dry season from March to mid-April 2014 during the 7-SEAS (cf: Lin et al., 2013) 2014 spring campaign.The samples were collected using mini-volume air samplers (MiniVol, Airmetrics, USA) at a flow rate of 5 L min -1 .Two air samplers were set up at each site.Sampler No.1 was used for conducting PM 2.5 sampling for ion and elemental analysis.A pure quartz fiber filter (Whatman's, UK, Ø 47 mm) was used and equally divided into two halves for further analysis.Sampler No. 2 was used for collecting PM 2.5 samples for levoglucosan analysis.Teflon fiber filters (Whatman's, UK, 2 µm, Ø 46.2 mm) were thus used.Samples were collected for 24 hours starting at 9 am daily.The filters were stored in desiccators filled with silica gels before and after the sampling for at least 24 hours prior to being weighed using a microbalance (MX5, Toledo, Switzerland), with a sensitivity drift of 7 × 10 -6 .Each filter was weighed three times in a controlled room (temperatures: 27 ± 2°C and relative humidity: 41 ± 5%).Subsequently, filters were kept in a freezer until extraction.
PM 2.5 mass concentrations at CMU were compared with values obtained from an automatic active sampler (Tapered Element Oscillation Microbalance; TEOM) at the air quality

Sample Extraction and Analysis
One half of the quartz filter was used for analysis of ion composition and the other half for assessing elemental composition.

Extraction and Analysis of PM 2.5 -Bound Ions
Half of a filter sample was extracted with 45 mL of deionized water at 35°C for 30 min using an ultrasonicator (Elma, Germany; Sillapapiromsuk et al., 2013).The extracted solution was filtered with a cellulose acetate membrane (pore size 0.45 µm, diameter 13 mm) and analyzed for cations (Na + , NH 4 + , K + , Ca 2+ and Mg 2+ ) and anions (Cl -, NO 3 -and SO 4 2-) by an Ion Chromatograph (882 Compact IC plus, Metrohm, Switzerland).The ion concentration extracted from blank filters (Number of samples: n = 3) was used for a background subtraction.

Extraction and Analysis of PM 2.5 -Bound Elements
The second half of the quartz filter was extracted with 4 mL of aqua-regia (3:1, v/v, HCl to HNO 3 ) in a doublelayer Teflon digestion bomb at a temperature of 140°C for 4 hours (Wang et al., 2003).After extraction, the solution was cooled down to room temperature and then adjusted with 2% HNO 3 to 25 mL in a volumetric flask.Extracted solutions were filtered with a nylon membrane (0.45 µm pore size, 13 mm diameter) and stored at 4°C prior to analysis.Nine elements (Al, Cd, Cu, K, Mg, Mn, Ni, Pb and Zn) were analyzed using an inductive coupled plasmaoptical emission spectrometer (ICP-OES; Optima 3000, Perkin Elmer, Germany).The sample solutions were measured in triplicates, while quality control of the ICP performance was done by use of a mixed standard solution (1.25 mg L -1 ) for peak area comparison at every 15-20 sample injections.Blank samples were also extracted and analyzed using the same method as for the samples.A median value was used for background subtraction from each sample.

Extraction and Analysis of Levoglucosan
The method of levoglucosan extraction from the particulate samples was developed from Dhammapala et al. (2007).The sampling filters were spiked with 20 µL of 1000 mg L -1 methyl β-D-xylopyranoside (internal standard for levoglucosan) sonicated in 15 mL ethyl acetate for 30 minutes at a controlled temperature (~10°C).The extracted solution was filtered through a glass filter membrane (pore size 0.45 µm) and evaporated until nearly dried under a stream of air in a water bath at 35°C.The residue was added with 1 mL of ethyl acetate.Only 100 µL of the solution was drawn and dried under a stream of nitrogen.After that, 100 µL of derivatizing reagent (N-methyl-Nbis(trimethylsilyl)-trifluoroacetamide and 2% N-(Trimethylsilyl)imidazole (v/v)) were added and incubated at 70°C for 1 hour.The derivatized sample was then analyzed by gas chromatography with a flame ionization detector (GC-FID).

Data Quality Control Quality Control for Ion Analysis
Precision of the ion analysis was tested in terms of repeatability and reproducibility five times by injections of mixed ions standard solutions.Relative standard deviation (%RSD) was used to characterize instrument precision.Repeatability and reproducibility for cation analysis were 0.6-2.3% and 0.7-2.0%,respectively, while those of anions were 0.2-4.3% and 0.7-4.6%,respectively.The instrument detection limits (IDL) values of the IC were 0.012-0.024mg L -1 for cations and 0.002-0.022mg L -1 for anions.Moreover, accuracy of the ion analysis was performed by spiking of the mixed-ion standards (0.5 mg L -1 ) onto a half of the quartz filter before extraction (Sillapapiromsuk et al., 2013).High-percentage recoveries of all ions (89-107%) obtained, which confirmed the relatively high accuracy of the analysis.

Quality Control for Elemental Analysis
Accuracy of the elemental analysis was performed using Standard Reference Material (SRM) urban dust 1648a (National Institute of Standard & Technology; NIST).Five sets of about 100 mg SRM were extracted using the same method as the samples.High recoveries (83-104%) of Cd, Cu, Mg, Mn, Ni, Pb and Zn were obtained.For other elements (Al, Cd, Cu, K, Mg, Mn, Ni, Pb, and Zn), the mixed standard solution (0.5 µg mL -1 ) was spiked onto the filter (n = 5).The recoveries were also high (83-111%).Precision of the method was obtained from five injections of 1.25 µg mL -1 mixed-standard solution into the ICP-OES.Relatively good precision in terms of %RSD was obtained for both repeatability (0.6-7.8%) and reproducibility (3.1-11.9%).Moreover, the values of IDL for elemental analysis by ICP-OES were identified with five injections of a low concentration (0.02 mg L -1 ) of the mixed standard used for construction of corresponding calibration curves.The standard deviation (SD) of the five injections of each element was multiplied by three.The IDL values of the ICP-OES for all analyses ranged from 0.001 to 0.076 mg L -1 .

Quality Control for Levoglucosan Analysis
Efficiency of the levoglucosan analysis method was tested by spiking method to obtain recoveries of levoglucosan at three levels of standard concentration (2, 10 and 100 mg L -1 ).The average percent recoveries obtained from those three concentrations ranged between 75-97%.The precision of levoglucosan analysis using GC-FID was relatively high, with %RSD values of 5.8 (repeatability) and 2.4 (reproducibility).The IDL value of the GC-FID for levoglucosan analysis was 0.3 mg L -1 .

Meteorological Data
Meteorological parameters, including temperature (T), pressure (P), relative humidity (RH), and wind speed (WS) at CMU, were obtained from an automatic weather station operated by the Thai Meteorological Department located in Chiang Mai (about 3 km from the CMU).Meteorological parameters for DAK were obtained from an automatic weather station (Vaisala Weather Transmitter: WXT520, Finland).

Back Trajectories
Three-day back trajectories arriving at the CMU and DAK sampling sites were performed using the National Oceanic and Atmospheric Administration's Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model version 4 (Draxler and Hess, 1998) for qualitatively assessing the transport of aerosols from their source regions to the receptor location.Global Data Analysis System (GDAS) meteorological data (0.5° × 0.5°) were used for the calculations.Trajectories were calculated daily from 1 to 31 March 2014 starting at 06:00 UTC and 18:00 UTC from altitudes of 500, 1000, 1500 and 2500 AGL, clustered to characterize the primary trajectory direction and orientation arriving at the receptors.

Data Analysis
Data obtained from the experiments were compiled and analyzed by the SPSS statistical program (Version 14.0 for Windows).A T-test was employed for comparison of pollutants from near-source and urban sites.Moreover, their correlation coefficients (r) were also calculated.Principal component analysis (PCA) was used to identify possible sources of pollutants detected.A log-transform (log (x+1)) approach was employed for data transformation to obtain a normal distribution before PCA analysis.Varimax rotation principal component patterns for levoglucosan, metal and ion concentrations of both CMU and DAK sites were solved.Eigenvalues higher than 1 were used for identification of pollutant sources.The maximum percentages of total variance were used as the components.Loading determining the most representative for each component, and generally at a value higher than 0.5, was selected (Ayers and Yeung, 1995).Moreover, correlation between PM 2.5 concentrations and the number of hotspots provided by the Fire Information for Resource Management System (FIRMS) was investigated.The number of hotspots was derived from satellite images collected with the Moderate Resolution Imaging Spectroradiometer (MODIS) on board NASA's Aqua and Terra satellites (NASA/University of Maryland, 2002).Each hotspot/active fire location represented the center of an approximately 1 km pixel flagged as containing one or more actively burning hotspots/fires.Chiang Mai (CM) province was specified and a period of time was set.After that, the number of hotspots was counted using QGIS.
Daily variation in PM 2.5 mass concentrations over the study sites obtained from both the gravimetric and TEMO methods are shown in Fig. 2. PM 2.5 concentrations started increasing at the beginning of March, decreasing in early April.Noticeably, more than 85% of the 24-h PM 2.5 concentrations from both sampling sites exceeded the Thailand National Ambient Air Quality Standard (50 µg m -3 ).PM 2.5 concentration during the low-burning period (February and April 2014) obtained from the PCD station in CM city was also plotted against those measured values.The results reveal that average PM 2.5 concentrations observed in the low-burning period (35.2-41.1 µg m -3 ) were about 2-3 times lower than those in the study period.This is mainly due to intensive BB activity during this period, resulting in high production of airborne PM.In addition, the geography (basin surrounded by high mountain ranges) and meteorological conditions favored the formation of a temperature inversion in the lower atmosphere during the dry season, which create stable atmospheric conditions that constrain vertical air flow and pollutant trapping near the surface (Wallace et al., 2010).This condition enhanced the confinement of BB aerosols near the surface, resulting in high values of mean surface mass concentration (Pani and Verma, 2014).
At the same period with our study, the aerosol vertical distribution was determined by using NASA MicroPulse Lidar Network (http://mplnet.gsfc.nasa.gov/;Welton et al., 2001;Campbell et al., 2002), that found at DAK, with 532 nm extinction coefficient profiles solved.They revealed a decreasing mixed layer height and stable synoptic conditions (i.e., high pressure and weak wind speed).Combined with intense regional forest fires, these conditions led to stagnant air masses that enhanced local haze episodes (Wang et al., 2015).In our study, the highest PM 2.5 mass concentration was observed on 20 March 2014 (189 µg m -3 ).Based on MicroPulse Lidar observation, the peak of the mean 532 nm aerosol extinction near the surface was ~0.5 km -1 , and a top aerosol height reaching approximately 0.5 km AGL confirming thermal capping at this level.Comparing to the day with lower PM 2.5 concentration (10 March 2014; 85 µg m -3 ), the peak of mean aerosol extinction was ~0.3 km -1 , while the aerosol height extended from the surface to approximately 1.5 km AGL, indicating greater vertical mixing.
PM 2.5 concentrations obtained from the two sampling sites, using different techniques, were analyzed for correlation of each pair.Results of the mini-volume air sampler and TEOM were strongly correlated at DAK (r = 0.964-0.975)and CMU (r = 0.877-0.913).Surprisingly, the concentrations measured at the near-source site (DAK) were also highly correlated (r = 0.850 and 0.817) with those measured at the urban site (CMU), indicating possible influence of the same source and similar transport pathways of BB plumes.
The meteorological parameters over CM City during January-May 2014 were plotted to illustrate their seasonal trend and variation during the study period dominated by BB activities (smoke-episode and marked as red colored square block (specific-period) in Fig. 3).Average values of T, P, RH and WS at CM city and DAK were 27.7 and 21.9°C, 1013 and 1008 hPa, 44 and 39% and 1.7 and 2.6  m s -1 , respectively during the specific period.Winds shifted from south to north, with an average wind speed of 1.7 m s -1 .Precipitation during the sampling period was 13.0 mm at CM City and 23.7 mm at DAK, respectively.Pollutant transport, including PM 2.5 , from CM City to DAK would take approximately twenty hours.This means that the CM region could plausibly be one of pollutant sources contributing to air quality at DAK.
Based on the meteorological patterns (Fig. 3), anomalous conditions occurred during the smoke episode, except for RH.Open BB in agricultural area is nominally performed when the weather is dry.Apart from that, forest fires often occur during dry season, leading to large open burning areas.Such conditions are thus strongly correlated with high PM 2.5 concentrations during the smoke period.Therefore, the period was divided into three sub-periods (Period I: 9-15 March, Period II: 16-22 March, and Period III: 25 March-2 April).
During Period I, average PM 2.5 mass concentrations varied between 81.5 ± 19.0 to 85.9 ± 15.3 µg m -3 at CMU and 77.1 ± 13.2 to 86.6 ± 15.4 µg m -3 at DAK.During Period II, average mass concentrations varied as 118.9 ± 33.8 to 131.7 ± 52.1 µg m -3 at CMU and 125.7 ± 42.5 to 129.2 ± 39.4 µg m -3 at DAK.The highest PM 2.5 mass concentrations at DAK and CMU were observed during 20 and 21 March, respectively.After this, concentrations decreased as local meteorological conditions evolved.In particular, there was a sudden change of RH values at DAK from 28% on 20 March to 77% on 23 March that corresponded with a decrease in temperature.The indicated behavior of hygroscopic particles exposed to increasing RH is to adsorb moisture, becoming saturated droplets as size increases, and thereafter as humidity increases still further to grow larger and more dilute (Orr et al., 1958;WMO/GAW, 2003).
At CMU, the highest PM 2.5 was found on 21 March (177-223 µg m -3 ), dropping dramatically on 22 March due most likely to an increase of wind speed.The result is consistent with that of Cheng et al. (2009) (Taichung City, Taiwan), which reported that a similarly-high PM 2.5 concentration was likely due to agricultural waste burning under calm or no winds (average WS = 0.5 m s -1 ).Moreover, it rained on 24 March (5.9 mm at CMU and 16.2 mm at DAK).Therefore, PM 2.5 concentration reduced to 58-61 µg m -3 at CMU and 63-71 µg m -3 at DAK on 25 March through washout effects.After the rain, open burning started again and PM 2.5 gradually increased.During Period III, average PM 2.5 mass concentrations were 92.5 ± 16.5 to 93.1 ± 16.2 µg m -3 at CMU and 83.7 ± 15.6 to 87.2 ± 19.6 µg m -3 at DAK, which were similar to those in Period I.
An important determining factor observed during each period was the number of satellite-derived fire hotspots, used to represent total area and/or amount of open burning.Correlations between daily hotspot numbers in CM province and PM 2.5 concentrations at CMU and DAK were 0.486 and 0.621, respectively.Numbers of hotspots detected in CM province was 398 (Period I), 684 (Period II) and 573 (Period III), each proportional to PM 2.5 concentrations in those intervals.The number of hotspots was particularly high (> 200) during 20 and 21 March when the PM 2.5 peak was recorded.

Concentrations of PM 2.5 -Bound Ions
The daily variation and average concentration of ion species in the PM 2.5 samples are shown in Fig. 4 and Table 1.According to the pattern of PM 2.5 concentration at both sites (Fig. 3), the study duration could be justifiably divided into three periods; Period I, lower concentrations of PM 2.5 (86-87 µg m -3 ) and total ions (17-18 µg m -3 ); Period II, higher concentrations of PM 2.5 (129-132 µg m -3 ) and its ion content (21-25 µg m -3 ); Period III, lower concentrations  of PM 2.5 (84-95 µg m -3 ) and ion content (22 µg m -3 ).The maximum concentrations of total ions were found on 20 March at DAK (38 µg m -3 ) and on 21 March (32 µg m -3 ) at CMU.The average concentrations of most ions at the CMU site were slightly higher than those at DAK.Since the CMU site is situated in a basin adjacent to the Suthape mountain foothill (Fig. 1), pollutants can accumulate at the foothill since winds are blocked by the mountains.This result in relatively high pollutant concentrations in the city compared with the DAK site, which is situated on top of the mountain where air can flow more freely.
Variation in PM 2.5 and its bounded total ion concentrations over DAK and CMU are depicted in Supplemental Fig. S1.It was found that time variation of total ion and PM 2.5 concentration for both sites were similar.Correlations between concentrations of PM 2.5 and total ions at CMU and DAK were 0.528 and 0.624, respectively.The major anion constituents at both DAK and CMU sites in descending order were SO 4 2-> NO 3 -> Cl -, while cations were Na + > NH 4 + > K + .The pattern was similar with Chantara et al. (2009), in which the major soluble species during BB in CM city were SO 4 2-> NO 3 -> NH 4 + > K + > Cl -.Sulfate was the greatest contributor to total ions in both the CMU (7.81 ± 2.60 µg m -3 ) and DAK (6.20 ± 2.51 µg m -3 ) samples.Referring to the emission inventory covering the area of CM Municipality in 2010, reported by Sopajaree et al. (2011), the major source of urban SO 2 is fossil fuel combustion, primarily from traffic sources (59%).Moreover, point sources were also responsible for a significant proportion (39%) as large amounts of diesel generators were used for boilers and industrial processes.Fossil fuels contain sulfur that can be transformed to particulate SO 4 2- (Hu et al., 2008).Chantara et al. (2012) revealed that the dominant ion species over a 5-year period (2005)(2006)(2007)(2008)(2009) found in CM was SO 4 2-(> 39% of total ion concentrations).Its highest concentration was found in 2007, which was the year that a severe haze episode occurred in Northern Thailand.The results presented here are consistent with that study.Moreover, SO 4 2-was found to be related with Zn at the CMU site (r = 0.826), confirming traffic influence in the local urban area (Lin et al., 2005).However, Sillapapiromsuk et al. (2012) revealed that SO 4 2-was the major ion found in PM 10 samples released from BB. Emission factors of ions emitted from leaf litter burning listed in descending order were SO 4 2-> K + > NO 3 -> Cl -> NH 4 + .Pengchai et al. (2009) analyzed PM 10 and its chemical composition collected from the ambient air of CM City using a source-apportionment model.It was mentioned that SO 4 2-could be formed from a reaction between vehicle exhaust gases and vegetative burning during BB periods.Hence, sources of SO 4 2-found at the CMU site are fossil fuel combustion (traffic) and BB.
Noticeably, SO 4 2-concentrations measured at the two sampling sites during 22 March-9 April were significantly higher than the early March period (1-21 March), unlike PM 2.5 .Correlations of SO 4 2-and PM 2.5 concentrations during 1-21 March were relatively strong at both CMU (r = 0.623) and DAK (r = 0.745), revealing particulate SO 4 2through gas phase reaction.On the other hand, there was no correlation found during 22 March-9 April, caused by reductions of PM 2.5 and corresponding increases of SO 4 2-.This is likely due to increasing of %RH caused by rain.

Moderate correlation between SO 4
2-and %RH was found at both CMU (r = 0.568) and DAK (r = 0.646).This result is consistent with Cheng et al. (2014), in which concentration of SO 4 2-increased along with the increase of RH.This implies that RH played an important role in the formation of particulates.
Peaks of SO 4 2-were mostly observed in foggy/cloudy days, when the oxidation of SO 2 could be significantly promoted due to aqueous processing.Moreover, the aqueous-phase processing due to high humidity could accelerate the transformation of SO 2 to acids via the oxidation pathways, such as H 2 O 2 and O 3 oxidation (Wang et al., 2016).Another pathway of particulate SO 4 2-is the oxidation of SO 2 through gas phase reaction (Wang et al., 2016).Moreover, 24-h air mass back trajectories at 0 m AGL were performed to assess sources of SO 4 2-at DAK.It was found that during 1-19 March (before increasing of SO 4 2-) air masses originated from the western continental areas relative to the DAK site.At the end of March (33% of sample) and in early April (19%), air masses originated from the south through CM City.
Nitrate was also measured as a major ion during this study.Potential sources of NO 3 -are traffic emissions (Minguillon et al., 2014), BB (Ruy et al., 2007;Chantara et al., 2012) and agricultural activities (Ahlgren et al., 2008;Gurjar et al., 2015).In this study, high correlations between NO 3 -and PM 2.5 concentrations were found at both CMU (r = 0.719) and DAK (r = 0.911).NO 3 -and BB tracers (K + and levoglucosan) were also well correlated at both sites (Table 2), revealing that NO 3 -probably originated from BB activities.However, the average NO 3 -concentration of DAK samples (4.13 ± 2.86 µg m -3 ) was higher than that of CMU (2.65 ± 1.13 µg m -3 ), which may be due to DAK being in closer proximity to intensive BB.Moreover, the site is surrounded by agricultural activities, where nitrogen fertilizer has generally been applied.
Long-range transport could also increase the loading of NO 3 -at DAK (downwind site), especially when air masses pass over open BB areas.The rsesults agree well with Chuang et al. (2016a), who reported that aerosol concentrations can be enhanced by production of secondary aerosol components en route from source areas to downwind sites.On the other hand, lower NO 3 -loading could be caused by NO 3 -thermal volatilization under higher temperature conditions.Higher temperature might also reduce the formation of NO 3 -in accumulation mode (Li et al., 2014).In fact, the fractions of nitrate in fine particles generally show a negative correlation with temperature (Zhao and Gao, 2008).Each of these factors likely contributed to the results presented here.The average NO 3 -concentration at DAK (higher altitude and lower air temperature) was higher than that of CMU, located in the city with higher air temperature.As both SO 4 2and NO 3 are two major water soluble ions in aerosols, high emissions of SO 2 and NO x , their transformations to SO 4 2-and NO 3 -, and their long/medium-range transport would have a significant impact on the air quality (Wang et al., 2016).
NH 4 + ranked as the most abundant species among cations found in PM 2.5 .Average NH 4 + concentrations at CMU (3.03 ± 0.83 µg m -3 ) and DAK (2.92 ± 1.00 µg m -3 ) were nearly the same.NH 4 + concentration was highly correlated with SO 4 2-at both sites, but with NO 3 -only at DAK (Table 2).At CMU, NH 4 + levels were lower than NO 3 -until after the rain event when the order reversed (Fig. 4(b)), which reconciles why NH 4 + and NO 3 -showed no correlation at CMU.The decreasing of NO 3 -concentration was probably due to RH increasing during the rain event.A negative correlation (r = -0.401) of NO 3 -and RH was in fact found, while RH was positively correlated with NH 4 + (r = 0.366).After the rain, NH 3 can be released from soil through agricultural activities, consequently leading to relatively high amounts of NH 4 + .However at CMU, moderate correlation between NH 4 + and NO 3 -before and after the rain were found (r = 0.689 and 0.542, respectively).
NH 3 acts as the major alkaline species for the neutralization of SO 4 2-and NO 3 -in PM 2.5 .NH 3 is emitted by a large number of sources, such as volatilization from animal waste, agricultural processes including fertilizers, BB (including forest fires and agricultural waste burning) and fossil fuel combustion (Krupa, 2003;Chantara et al., 2012;Sillapapiromsuk et al., 2013).Chantara et al. (2012) reported that NH 3 represents a major gas found in CM, mainly emitted from agricultural activities.In this study, NH 3 was probably affected by both agricultural activities and intensive open burning during the dry season, especially at DAK. High correlations between NH 4 + , NO 3 -and SO 4 2were found, indicating that NH 4 + likely originated from the neutralization of NH 3 by HNO 3 and H 2 SO 4 and then formed NH 4 NO 3 and (NH 4 ) 2 SO 4 (Kang et al., 2010;Kong et al., 2014).The correlation of NH 4 + and SO 4 2-was higher than NH 4 + and NO 3 -(Table 2) meaning that (NH 4 ) 2 SO 4 was a major compound present at both sites.The result agrees well with Chantara et al. (2012), which reported high correlation (r > 0.777) between NH 4 + and SO 4 2-.Average K + concentrations were 2.06 ± 0.60 µg m -3 at CMU and 1.82 ± 0.68 µg m -3 at DAK.K + is well known as a tracer of BB (Andreae, 1983;Chow, 1995;Rye et al., 2007).Concentrations of PM 2.5 and K + were highly correlated at both CMU (r = 0.707) and DAK (r = 0.786), indicating the high potential of representing the same source (BB).Maximum concentrations of PM 2.5 at CMU and DAK (223 and 189 µg m -3 ) and their major cations, K + (3.53 and 3.46 µg m -3 ) and NO 3 -(6.05and 14.67 µg m -3 ), were found on the same day.It has been reported that concentrations of K + and NO 3 -increased during BB episodes (Cheng et al., 2014;Prenni et al., 2014).In this study, high correlations between K + and NO 3 -were found at both sites (r = 0.764 at CMU and r = 0.804 at DAK).Moreover, significant correlations between K + and K element were also found at CMU (r = 0.833) and DAK (r = 0.502).The ratio values of K + /K were 0.62 (CMU) and 0.68 (DAK).Apart from that, correlations between K + and levoglucosan were also relatively high (0.539 at CMU and 0.686 at DAK).Levoglucosan is considered a unique molecular tracer of BB aerosols.This result indicates that PM 2.5 found at both sampling sites was predominantly the result of BB.
Another important ion is Na + , which consists mostly of crustal origins (Tao et al., 2013;Jaafar et al., 2014;Wang et al., 2016) and was present in relatively high concentrations  during the BB period.Open burning increases soil temperature and, when it is sufficiently elevated, the chemicals can be released from the soil into the air (Williams et al., 1992;Jimenez et al., 2006).Sillapapiromsuk et al. (2013) reported that Na + was emitted from BB. Concentrations of Na + were relatively high at both sampling sites, and thus they may have been released from heated soil and burned biomass.Noticeably, average Na + concentration of DAK samples (4.07 ± 1.05 µg m -3 ) was significantly higher than that of CMU (3.23 ± 1.59 µg m -3 ), presumably because DAK is located near the BB source.The average Na + concentration was about 3-4 times higher than Cl -at both sites.This result agrees with the study conducted by Jaafar et al. (2014), which reported that concentrations of Na + were 4-15 times that of Cl -for fine dust (PM 2.5 ), though much less was found as part of the coarse fraction.Unlike Na + , Cl -was a major contributor to the coarse fraction.In terms of geography, the study sites were located far from the sea, which explains the lack of correlation between Na + and Cl -concentrations.Therefore, sources of Cl -at both sites are plausibly from BB (e.g., Chow, 1995).Sillapapiromsuk et al. (2013) revealed that burning of agriculture residues (maize residue and rice straw) emitted high Cl -concentrations, and was related to the use of fertilizers and herbicides in the field.Paraquart dichloride (C 12 H 14 N 2 Cl 2 ) is widely used as a herbicide in maize plantation in Thailand, in addition to glyphosate, 2,4-D, ametryn and atrazine (Panuwet et al., 2012).Tao et al. (2013) also reported high concentrations of Cl -in PM 2.5 emitted from BB (wheat straw and rape straw) at Chengdu, China.However, the emission of Cl -from agriculture residues burning was higher than that of leaf litter caused by forest burning (Sillapapiromsuk et al., 2013).

Concentrations of PM 2.5 -Bound Elements
Nine elements (Al, Cd, Cu, K, Mg, Mn, Ni, Pb and Zn) were extracted from PM 2.5 samples collected at the CMU and DAK sites.Concentrations of PM 2.5 and its elemental composition are shown in Table 1.Mean concentrations of these total elements were 6.62 and 5.47 µg m -3 for CMU and DAK, respectively.The element with the highest mean concentration at both sites was K (3.30 µg m -3 at CMU and 2.68 µg m -3 at DAK), while Cd and Pb were not detected (lower than the IDL values).High concentrations of K originate from BB (Chow, 1995;Reche et al., 2012;Zhao et al., 2013).Correlations between concentrations of PM 2.5 and associated chemical components were analyzed (Table 2).The K element and PM 2.5 concentrations were well correlated at both CMU (r = 0.766) and DAK (r = 0.685) as well as another BB tracers (K + and levoglucosan).
Both sites showed the same relative percentage ratio of total extracted elements in descending order (CMU and DAK, respectively), which were K (49.8% and 49.1%) > Al (26.3% and 30.6%) > Mg (15.9% and 15.6%) >> Zn (6.9% and 3.7%).Most of the elements, except Zn, were observed in similar concentrations between sites.Zn concentration at CMU was significantly higher (p < 0.05) than that at DAK.This is because Zn likely originates locally from brake or tire wear on automobiles (e.g., Lin et al., 2005;Jaafar et al., 2014).Therefore, it was likely higher at the urban site (CMU).Only low concentrations of Cu, Mn and Ni were found in the samples, accounting for < 0.5% of the total elements.Cu is predominantly caused by brake dust and fossil fuel combustion, while Ni is a naturally-occurring element in the Earth's crust (Cempel and Nikel, 2006;Zhang et al., 2013;Nava et al., 2015).Al, Mg and Mn originate from soil dust (Yongjie et al., 2009;Kang et al., 2013;Enamorado-Báez et al., 2015;Nava et al., 2015) and were found to be well correlated at both sites.

Concentrations of Levoglucosan
To determine the impact of BB emissions on ambient aerosols, it is essential to correctly identify BB tracers.Common BB tracers are levoglucosan and K + (Cachier et al., 1991;Jung et al., 2014).Average concentration of levoglucosan at DAK (1.38 µg m -3 ) was slightly higher than at CMU (1.13 µg m ), with no significant difference (p > 0.05).Noticeably, the daily trends of levoglucosan, PM 2.5 , K + and K were similar (see Fig. S2 in the Supplement).Correlations between levoglucosan and PM 2.5 were significant at both sites (Table 2), indicating the influence of BB sources.However, the results showed that their correlation at DAK site was higher than CMU, again confirming that DAK is the near-BB source site.Moderate correlations between levoglucosan and K + at CMU (r = 0.539) and DAK (r = 0.686) were also found.This means that high PM 2.5 concentrations during the specific period were influenced by BB.

Principal Components Analysis (PCA)
Levoglucosan and the dominant species of elemental and ion composition of PM 2.5 were used as dependent variables in a principal components analysis (PCA) study to determine their possible sources.Varimax-rotation factor loadings with absolute values greater than 0.50 are presented and marked bold (Table 3).Patterns of factor loadings for PM 2.5 chemical composition in PCA over CMU and DAK sampling sites are listed in the same Table .PCA analysis at CMU can be classified into three components.The first part (32% of variance) showed high loading of SO 4 2-, NH 4 + , Ca 2+ , Al and Zn.SO 4 2-is a source associated with the formation of secondary sulfate aerosols from combustion sources (Tao et al., 2013;Huang et al., 2014), while NH 4 + originates from agricultural activity (Zhang et al., 2007).Zn probably originated from vehicles and fossil fuel combustion (Nava et al., 2015;Longhin et al., 2016), while Al and Ca 2+ mostly characterize crustal origins (Al-Khashman, 2009;Tao et al., 2013;Wang et al., 2016).The second component reflected high loadings of NO 3 -, K + and levoglucosan, representing the contribution of BB (Ruy et al., 2007;Zhang et al., 2007;Jung et al., 2014;Huang et al., 2014).Levels of K + and NO 3 -have been shown to increase during agricultural waste burning periods (Ezcurra et al., 2001).The third component contained high loadings of Na + and Mg, which are mainly soil characteristics (Nava et al., 2015;Wang et al., 2016).Moreover, a high negative loading of Cl -is shown, meaning + and Ca 2+ , which likely came from combustion sources, agricultural activities and soil re-suspension, respectively.A high loading of Cl -with negative value was also found in this component.Al and Mg come from crustal source, which were the major contributors to the third component.The last component contained only Zn, which is identified as road dust released from tire wear and diesel fuels (Adachi and Tainosho, 2004;Blok, 2005;Longhin et al., 2016).The results in this study indicate that sources of PM 2.5 at CMU and DAK are primarily BB, road traffic, agricultural activities and re-suspension of soils.

Back Trajectories
Three-day backward trajectories in March 2014 were calculated using the HYSPLIT model to evaluate possible source regions of air masses arriving at CMU and DAK.Trajectories starting at a specific altitude and time were clustered (see Fig. S3 in the Supplement).All back trajectories for both CMU and DAK sites were classified into three patterns: Pattern I reflect air masses originating from the Arabian Sea or India passing over the Andaman Sea and Myanmar to the sites; Pattern II represent air masses originating from the Andaman Sea near the southern part of Myanmar or Gulf of Moattama in southwest Thailand, passing through the southern region of Myanmar; Pattern III characterizes air masses originating from the east and southeast directions passing through some parts of Thailand to the sites.
The primary direction of those air masses transported to the CMU site at an altitude of 500 m and a 06:00 UTC starting time moved across the sea from the northwest (48%) and southwest (44%) directions.This pattern was found consistent at higher altitudes as well.The distance from source region to the site increased with altitude level.At 1000 m altitude, air masses were from the southern part of Myanmar (44%), northwest (33%) and Andaman Sea bending through southern Myanmar (22%).At 1500 m altitude, air masses were from the Andaman Sea (63%) and east coast of India (37%).Only at 2500 m did trajectories originate from the southeast, from Andaman Sea (48%), east of Thailand (33%) and the Arabian Sea (19%).
At 500 m and 18:00 UTC, the main air mass trajectories originated in the Andaman Sea, bending toward the southern part of Myanmar before reaching CMU (93%).The same pattern was found at 1000 m, with 81% contribution of trajectories from the west.At 1500 m, transport was from the west (India and Andaman Sea; 67%).At 2500 m, it was from the southwest direction (Pattern II), varying from south of Myanmar (63%) to the Arabian Sea (37%).At DAK, 85% of air masses arriving at 500 m and 06:00 UTC were from the southwest passing through Myanmar (Pattern II), though this scenario occurred less at higher altitudes (44-78%).This result was consistent with those trajectories starting at 18:00 UTC.Then, the primary orientation was from the southwest, with a noticeably higher contribution (41-93%).Major air mass movement to CMU and DAK typically passed through southern Myanmar and then over some part of Thailand (typically Mae Hong Son Province), which corresponded with a relatively high distribution of fire hotspots.The high pollutants at both sampling sites again reinforce the fact that the two sites are located near active BB sources (see Fig. S4 in Supplement).

CONCLUSIONS
The present study reports the chemical composition of biomass burning (BB) aerosols collected from two ground sites subject to near-source burning (Doi Ang Khang, Thailand; DAK) and urban downwind transport (Chiang Mai University, Thailand; CMU) over northern Southeast Asia during the dry season 2014 under 7-SEAS/BASELInE measurement campaign.The mass concentrations of chemical species (levoglucosan, ion and elemental compositions) and their distinctive concentration gradients in particulate matter of less than or equal to 2.5 µm diameter (PM 2.5 ) at both sites are investigated, statistically analyzed, and compared.The influence of regional air mass movements on observed aerosol species at both sites is also analyzed.The enrichment of PM 2.5 mass concentration and tracer concentration gradients confirm the influence of BB activity.
High concentrations of PM 2.5 were found at both CMU and DAK, and both sites shared relatively consistent patterns in evolution of observed species during the study period.Average PM 2.5 mass concentrations at DAK and CMU were strongly correlated, influenced directly by a relatively common set of regional BB emission characteristics.Ions including SO 4 2-, NO 3 -, Na + and NH 4 + and elements (i.e., K, Al and Mg) were the major species found at both locations.SO 4 2-, NO 3 -and NH 4 + accounted for higher than 64% of total ions.The high correlation between SO 4 2-, NO 3 -, and NH 4 + confirms the in-situ secondary formation of these species.High correlations between PM 2.5 and their chemical composition were found, including K + , NO 3 -, K, and levoglucosan.Aerosols from local BB activities (open burning of agricultural waste and forest fire) were predominant over the study area, followed by the influence of anthropogenic activities and urban vehicle emissions.Principal components analysis confirmed that BB, road traffic and soil dust were the main sources of PM 2.5 in this area.Air mass trajectory analysis showed that the contribution of major pollutants to observations at the CMU and sites was influenced by transport through the southern regions of Myanmar and subsequent advection to the southwest over Mae Hong Son Province, Thailand.
During haze episodes in Chiang Mai province, pollutants are spread everywhere covering the whole area in the dry season.The serious situation is regarded to health risk.This is especially true for PM 2.5 , as they can penetrate deep into the lungs and cause serious health problems.Future studies on the presence of toxic compounds such as polycyclic aromatic hydrocarbons (PAHs) in PM 2.5 during BB episodes in ambient air would provide useful information for regional human health risk analysis.

Fig. 1 .
Fig. 1.Location of the sampling site and land use pattern in northern Southeast Asia.

Fig. 2 .
Fig. 2. Daily variation in PM 2.5 mass concentrations over the study sites obtained from both the gravimetric and tapered element oscillation microbalance method.The bars denote the daily precipitation amount.

Fig. 3 .
Fig. 3. Daily variation in meteorological parameters and the PM 2.5 mass concentration over Chiang Mai city (CM city) and Doi Ang Khang (DAK).The red colored square block is specific-period (smoke-episode).

Fig. 4 .
Fig. 4. Daily variation of relative humidity and ion species in the PM 2.5 samples collected from Chiang Mai University (CMU) and Doi Ang Khang (DAK) sites.

Table 1 .
Composition of PM 2.5 samples collected from Chiang Mai University (CMU) and Doi Ang Khang (DAK) sites.

Table 2 .
Correlations between concentrations of PM

Table 3 .
Summary of factor loadings in principal component analysis over Chiang Mai University (CMU) and Doi Ang Khang (DAK).