Atmospheric Pollution Research 2 (2011) xx-xx

Atmospheric Pollution Research

www.atmospolres.com

Characterization and source apportionment of particulate pollution in Colombo, Sri Lanka

M.C. Shirani Seneviratne 1, V.A. Waduge 1, L. Hadagiripathira 1, S. Sanjeewani 1, T. Attanayake 1, N. Jayaratne 2, Philip K. Hopke 3

1 Atomic Energy Authority, 60/460, Baseline Road, Orugodawatta, Wellampitiya, Sri Lanka

2 Central Environmental Authority, Battaramulla, Sri Lanka

3 Center for Air Resources Engineering and Science, Clarkson University, Potsdam, NY, USA

ABSTRACT

Because of increasing use of vehicles and other human activities in metropolitan Colombo, Sri Lanka, a study of long–term airborne particulate monitoring at two fixed sites was initiated. The specific objectives were to measure the elemental composition of the coarse and fine air particles, to identify the trends in pollution, to identify the main pollutant sources, and to quantify the source contributions. Samples of airborne particulate matter (PM) in the < 2.5 and 2.5–10 µm size ranges (PM2.5 and PM10–2.5) were collected using “Gent” stacked filter samplers at two urban sites in Colombo. The Air Quality Monitoring Station (AQM) of the Central Environmental Authority (CEA) operated during the period of May 2000 to December 2005. The second site at Atomic Energy Authority (AEA) operated from May 2003 to December 2008. Twenty–four hour samples were collected on weekdays. The fine filter samples were analyzed for 18 elements by ED–XRF. The annual averages for PM10, PM2.5, and black carbon (BC) at the AQM station during 2000 – 2005 ranged from 50 to 100, 16 to 32, and 8 to 15 µg/m3, respectively. The fine fraction data set including BC and major elements (Na, Mg, Al, Si, Cl, Fe, Zn, Ni, Cu, V, S, Br, Pb, Cr, K, Ca and Ti) was analyzed using EPA–PMF (Positive Matrix Factorization) to explore the possible sources of the PM at the two study sites. Four factors were found at both sites. The common sources are motor vehicles, road dust, biomass and sea salt. The Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPILT) back–trajectory model was used to explore possible long range transport of pollution. Smoke and soil dust transboundary events were identified based on fine Si and K in the data base in 2003 and 2004.

Keywords: Particulate matter Receptor modeling

Positive Matrix Factorization Sri Lanka

Article History:

Received: 05 July 2010 Revised: 28 September 2010

Accepted: 28 September 2010

Corresponding Author:

Philip K. Hopke Tel: +1-315-268-3861 Fax: +1-315-268-4410

E-mail: hopkepk@clarkson.edu

© Author(s) 2011. This work is distributed under the Creative Commons Attribution 3.0 License. 1. Introduction 1 2

Colombo (Latitude: 6° 49’ 35N, Longitude: 79° 51’ 45E) shown 3 in Figure 1 is the capital of Sri Lanka. It has an area of 37 km

2 and is 4

located on the west coast of the island. Its population is over 5 600 000 inhabitants. The city is always congested with a large 6 number of motor vehicles from both private and public transpor-7 tation systems. The harbor is also located in the city. The weather 8 is tropical with temperatures varying from 25 °C to 30 °C. 9

10 Fine particle air pollution (PM2.5, particles with diameters 11

≤ 2.5 m) has been related to adverse health outcomes (Dockery 12 et al., 1993), poor visibility in Asia (UNEP, 2002; Ramanathan et al., 13 2005), and long–range transboundary pollution transport (Cohen 14 et al., 2004). Because of the increased use of vehicles and other 15 human activities in Colombo and suburbs, a study of long term 16 airborne particulate monitoring at fixed sites was initiated in 2000 17 and 2003, respectively. 18

19 To protect public health, many countries have instituted air 20

quality standards that set maximum allowable concentrations. For 21 example, the United States Environmental Protection Agency 22 introduced a PM2.5 standard of 15 g/m

3 for annual average and 23

35 g/m3 for 24–hr maximum in 2006. In Australia, the PM2.5 goals 24

are 8 g/m3 for annual average and 25 g/m

3 for 24–hr maximum. 25

Sri Lanka has a fine PM2.5 permissible level of 25 g/m3 for annual 26

average and 50 g/m3 for a 24 hr maximum. 27

28

This paper presents the results of a multiple year particle 29 characterization study that includes elemental analysis by ED–XRF 30 and the application of data analysis methods for source 31 apportionment. In addition, air parcel back–trajectories were used 32 to identify possible source locations that contribute to the 33 observed concentrations of fine particle pollution at a sampling 34 site in Colombo. 35

36 2. Techniques and Methods 37 38

Sampling was conducted using a “Gent” stacked filter unit 39 particle sampler capable of collecting particulate matter in the 40 range of PM2.5–10 and PM2.5 size fraction (Hopke et al., 1997). 41 Samples were collected over 24–hour periods on the weekdays 42 using nuclepore filters with 8 m pore size for the course fraction 43 and 0.4 m pore size filter for the fine fraction. 44

45 Sampling site 1 is located at the Air Quality Monitoring Station 46

(AQM) in downtown Colombo near the main train station. The 47 samples were collected during the 24–hourly on week days with a 48 flow rate 15 to 18 L/min. Sampling site 2 is located at Atomic 49 Energy Authority (AEA) premises on the northeastern side of 50 Colombo. The sampler was placed on the flat area in the first floor 51 of the AEA building. The AEA building is located near the main road 52 leading to the Colombo metropolitan area. 53

54 Mass values were determined by weighing before and after 55

the sample collection with a 24–hour period of equilibration at 56

x Seneviratne et al. – Atmospheric Pollution Research 2 (2011) xx-xx

50% relative humidity on a microbalance. Mass was determined for 1 both the coarse and fine particle samples. 2

3

Figure 1. Maps of Sri Lanka (left) and Colombo (right) showing the locations 4 of the two sampling sites. 5 6

Energy Dispersive X–ray Fluorescence (EDXRF) spectrometry 7 has been an accepted method for the characterization of particle 8 pollution for many years (Markowicz et al., 1996). This method is 9 well suited for the analysis of filters containing only few hundred 10 micrograms of fine particulate air pollutant. The EDXRF analyses 11 were conducted at Clarkson University (Spectro XLAB–2000) for 12 the fine fraction samples. Single element MicroMatter standards 13 were used to develop the calibration parameters and samples of 14 NIST SRM 2783 were routinely analyzed with each batch of 15 samples to provide quality assurance of the elemental 16 concentrations. XRF analysis of the fine particle filters determined 17 the concentrations of the elements Na, Mg, Al, Si, Ca, Cl, K, S, Ti, V, 18 Cr, Mn, Fe, Ni, Cu, Zn, Br and Pb on fine filters. The coarse particle 19 filters were not subjected to chemical characterization. 20

21 A Stain Smoke Reflectometer (Diffusion Systems Ltd. Model 22

M43D) was used to measure the Black Carbon (BC) in the fine 23 filters assuming an average fine particle mass absorption 24 coefficient of 5.7 m

2/g. The elemental analysis methods are mature 25

and have been applied in a large number of prior studies. 26 27 Advanced data analysis methods such as Positive Matrix 28

Factorization (PMF) have also been routinely applied to such data 29 sets. US EPA PMF (US EPA, 2010) version 3.0 is a process that 30 quantitatively provides both fingerprints and daily time series plots 31 for major source contribution at a given site. These EPA–PMF 32 processes and their applications have been discussed in detail 33 elsewhere (Reff et al., 2007; Norris et al., 2008). 34

35 3. Results and Discussion 36 37 3.1. PM mass data 38 39

About 500 filters were collected during 2000–2008 from two 40 sites. Figure 2 shows the measured PM10 and PM2.5 mass concen-41 trations for this study period. The PM2.5 mass concentrations 42 peaked above 60 µg/m

3 in 2002 and 2003. Apart from these 43

increases, there are minor seasonal trends with little excess in the 44

average annual mass of fine particles and PM10 over eight years 45 period. 46

47 At the AQM station, the annual averages for PM2.5 showed an 48

increase to 2003 and a decline beginning in 2004. These values are 49 well above the US EPA air quality standards (15 g/m

3). The annual 50

averages for PM10 ranged from 65 g/m3 in 2000 to 53 g/m

3 in 51

2005, again above the value (50 g/m3) previously recommended 52

by the USEPA. 53 54 The variation of PM10, PM2.5 and BC collected at the AEA 55

premises are shown for the period of sampling 2003 to 2008 56 (Figure 2). The variation of BC is similar to that at the AQM site 57 during the period and annual average was around 10 g/m

3. At the 58

AEA station, the annual averages of PM2.5 and PM10 do not exceed 59 the USEPA recommended values, and showed no consistent 60 pattern in their variation. 61

62 3.2. Composition 63 64

The average PM2.5 fine particle composition results over the 65 study period from 2000 to 2008 obtained are shown in Tables 1 66 and 2. Mean values and their standard deviations as well as 67 median and maximum concentrations for each of the measured 68 chemical species are presented. These concentrations are similar 69 to those observed in other major cities across south and south-70 eastern Asia (Hopke et al., 2008). The variations in constituent 71 concentrations generally follow the variability in mass concentra-72 tions. 73

74 Table 1. Average elemental concentrations with the standard deviations of 75 PM2.5 at the AQM site for the period of 2000-2005 76

Constituent Mean ± SD Median Maximum

PM (μg/m3) 29.0 ± 15.0 28.7 110.0

BC (μg/m3) 12.2 ± 5.9 11.2 34.6

Na (ng/m3) 218.4 ± 169.5 163.1 1 018.2

Mg (ng/m3) 72.0 ± 46.1 58.7 263.1

Al (ng/m3) 113.3 ± 108.0 85.8 702.2

Si (ng/m3) 378.6 ± 252.1 319.8 1 683.5

S (ng/m3) 709.3 ± 511.4 585.8 3 583.7

Cl (ng/m3) 145.0 ± 164.9 106.3 1 718.7

K (ng/m3) 474.0 ± 360.8 361.9 2 254.2

Ca (ng/m3) 150.4 ± 165.8 113.1 1 741.5

Ti (ng/m3) 20.4 ± 16.0 16.6 129.1

V (ng/m3) 6.7 ± 9.9 3.5 68.0

Cr (ng/m3) 13.7 ± 11.6 10.4 90.1

Fe (ng/m3) 631.2 ± 486.9 523.5 3 247.3

Ni (ng/m3) 43.1 ± 35.0 35.0 253.9

Cu (ng/m3) 130.6 ± 106.6 106.4 706.4

Zn (ng/m3) 106.4 ± 91.8 84.5 656.6

Br (ng/m3) 16.0 ± 9.8 12.8 79.2

Rb (ng/m3) 4.7 ± 3.5 3.8 22.6

Pb (ng/m3) 37.6 ± 23.5 30.3 176.0

77 78

79 Figure 2. Annual mean concentrations of PM10, PM2.5, and BC at the two sampling sites. 80

Seneviratne et al. – Atmospheric Pollution Research 2 (2011) xx-xx x

1 In addition, pseudo–elements (Malm et al., 1994; Begum et 2

al., 2006) were calculated. These variables are combinations of the 3 measured composition values that help to estimate the likely 4 major source types. Soil estimates were obtained by summing the 5 masses of five elements Al, Si, Ti, Ca and Fe when converted to 6 their common oxides. Ammonium sulfate was estimated from the 7 sulfur concentration assuming a fully neutralized aerosol (Malm et 8 al., 1994; Begum et al., 2006). 9

10 Non–soil potassium is defined as: 11 12

(1) 13 Non–soil potassium is a very strong indicator of biomass 14

burning. The average reconstructed mass (RCM) obtained by 15 summing all the analytes (Malm et al., 1994; Begum et al., 2006). 16 This pseudo–element analysis estimates that the average fine 17 particle composition is 48% BC, 9% soil, 11% ammonium sulfate. 18 The average reconstructed mass (RCM), obtained by summing all 19 the analyses and comparing with the gravimetric mass, was 20 (70 ± 30) % showing good mass closure for the data (Begum et al., 21 2006). 22

23 Table 2. Average elemental concentrations with the standard deviations of 24 PM2.5 at the AEA site for the period of 2003-2008 25

Constituent Mean ± SD Median Maximum

PM (μg/m3) 22.9 ± 11.9 20.4 72.7

BC (μg/m3) 11.3 ± 3.6 11.7 23.4

Na (ng/m3) 109.6 ± 51.9 105.0 373.2

Mg (ng/m3) 32.2 ± 20.8 34.0 132.9

Al (ng/m3) 113.2 ± 107.1 71.4 543.1

Si (ng/m3) 210.9 ± 141.7 229.0 759.2

S (ng/m3) 484.7 ± 304.8 436.4 1 990.9

Cl (ng/m3) 107.5 ± 87.6 109.1 835.3

K (ng/m3) 292.2 ± 152.7 279.7 998.4

Ca (ng/m3) 113.2 ± 190.8 92.1 2 411.2

Ti (ng/m3) 12.3 ± 7.7 12.5 39.8

V (ng/m3) 5.3 ± 4.8 4.3 36.5

Cr (ng/m3) 5.5 ± 4.9 3.6 32.9

Fe (ng/m3) 256.3 ± 215.1 157.4 1 355.8

Ni (ng/m3) 17.0 ± 15.5 9.3 101.8

Cu (ng/m3) 53.3 ± 50.4 26.1 319.6

Zn (ng/m3) 51.3 ± 47.2 38.1 329.6

Br (ng/m3) 9.3 ± 6.1 9.8 34.5

Rb (ng/m3) 2.2 ± 1.5 2.3 11.3

Pb (ng/m3) 20.8 ± 12.4 22.5 72.4

26 Monthly box and whisker plots for the study period for 27

concentrations of soil and ammonium sulfate are given in Figures 3 28 and 4, respectively. The (+) sign denotes the monthly mean, the 29 horizontal bar represents the monthly median and the hatched 30 boxes contain 25

th to 75

th percentile range of the values for that 31

month. Outliers are shown as dots for the month. 32 33 It can be seen that the concentrations at the AQM site are 34

consistently higher than at the AEA site for both soil and sulfate. 35 Although sulfate is normally considered a regional pollutant, there 36 is significant sulfur in diesel fuels in Sri Lanka. The AQM site is 37 adjacent to the major railroad station in Colombo and the trains 38 are diesel powered. There is also substantial light and heavy duty 39 vehicle traffic in this area. There may also be an impact from 40 marine diesel emissions from ships in the harbor (Kim and Hopke, 41 2008). Thus, the higher source density at the AQM site led to 42 higher concentrations of both soil and sulfate. There were 43 particularly high AQM values observed in samples collected in 44 February. February is often a period when air quality is influenced 45 by long–range transport and this possibility will be discussed later 46

in this paper. Unfortunately only 10 samples were collected in this 47 month over the whole sampling period. 48

49

Figure 3. Monthly distributions of soil concentrations as displayed in box 50 and whisker plots. 51

52

Figure 4. Monthly distributions of Knon (smoke) concentrations as displayed 53 in box and whisker plots. 54

55 56 57

x Seneviratne et al. – Atmospheric Pollution Research 2 (2011) xx-xx

If the relative concentrations of soil and ammonium sulfate at 1 the AQM site are considered on a percentage PM2.5 mass basis, 2 they showed distinct seasonal trends. Soil was nearly twice as high 3 as a percentage of total mass in the drier months, while 4 ammonium sulfate showed the opposite trend being higher in the 5 wetter months. In Colombo, the monsoon period is from May to 6 September with heavy rainfall during this interval. Airborne soil is 7 largely resuspended road dust. During the dry season, large 8 quantities of dry dust can be resuspended from unpaved or 9 partially paved roads. Thus, it dominates the concentration relative 10 to sulfate during the dry season. The road emissions are reduced 11 during rainy periods. The rain washes soil from paved roads and 12 reduces emissions from unpaved roads. The sulfate emissions are 13 relatively constant through the year and represent a higher 14 percentage of the measured mass concentrations during the 15 monsoon period. 16

17 3.3. Source apportionment using EPA positive matrix factorization 18 (EPA – PMF) 19 20

EPA PMF Version 3 (US EPA, 2010) was used with PM2.5 21 elemental data from each site to determine the factors that 22 provided a mass apportionment among the different kinds of 23 sources. The fractional elemental contributions associated with 24 each profile, together with the source contributions to the total 25 fine mass were determined by PMF. A significant choice made by 26 the user is the number of factors. Solutions were chosen that 27 provided good fits to the data as determined by examination of the 28 distributions of the scaled residuals and are physically inter-29 pretable sources. 30

31 The fine fraction data sets from the two sites included BC and 32

major elements (Na, Mg, Al, Si, Cl, Fe, Zn, Ni, Cu, V, S, Br, Pb, Cr, K, 33 Ca and Ti) The AQM data set included 144 samples covering the 34 period of May 4, 2000 to December 29, 2005 while the AEA data 35 set had 183 samples covering the period of June 18, 2003 to 36 February 25, 2008. They were analyzed separately by PMF to 37 explore the possible sources of the PM. The apportionments of the 38 fine particle mass concentrations for the two sites are presented in 39 Tables 3 and 4. The analysis revealed only four factors for each site. 40 The broad source categories for both sites are motor vehicles, road 41 dust, biomass and sea salt. Plots of the source profiles and 42 corresponding mass contributions for the AQM and AEA sites are 43 presented in Supporting Material (SM) (Figures S1 to S4). 44

45 Table 3. Source apportionment for the fine particulate matter mass 46 measured at the AQM site 47

PMF Factor High loading Main Sources

Factor 1 48% BC, S, K, Fe, Pb Motor Vehicles

Factor 2 27% Al, Si,Ti, Ca, Pb, Fe, K Road Dust

Factor 3 21% BC, K Biomass Burning

Factor 4 4% Na, Cl, BC Sea Salt

48 Table 4. Source apportionment for the fine particulate matter mass 49 measured at the AEA site 50

PMF Factor High Concentrations Main Source

Factor 1 17% BC, S Motor Vehicles

Factor 2 9% Si,Ti, Ca, Pb, Fe, Ni, Cu, Zn, Pb, Br Road Dust

Factor 3 66% BC, K, S Biomass Burning

Factor 4 9% Na, Cl, BC, Ca Sea Salt

51 In both cases, fewer sources were resolved than has often 52

been observed in other locations (e.g., Chueinta et al., 2000; 53 Begum et al., 2004; Santoso et al., 2008). In these two locations, 54 the sites are surrounded by these major areas sources (roads, 55 housing) so that there is little variability in the mixture of sources 56 from day to day. Although there are differences in total emissions 57 from day to day, the relative proportions of many small sources do 58

not appear to vary sufficiently that they can be resolved in these 59 data sets. 60

61 The first factor for the AQM site contributes 48% of average 62

PM2.5 mass concentration and includes BC, S, and minor quantities 63 of metal species. The second factor includes soil elements with BC. 64 This factor with high Si, Na, Mg, Cl, Cr, Fe, Ni, Cu, Pb, Zn, and Br is 65 assigned as the road dust being produced primarily from local 66 traffic, and it contributes 27% of the mass. The third factor is 67 related to K, and BC concentrations in biomass burning which 68 contributes 21% of the average mass. Pb and Br are generally 69 associated with automotive exhaust and are significant in the fine 70 fraction (Seneviratne et al., 1999). Unleaded gasoline was intro-71 duced into Sri Lanka in 2003 (ADB–CAIA, 2006). Thus, the nature of 72 automobile emissions has been changing over the time interval of 73 these measurements. The final factor includes Na, Cl, along with BC 74 and contributes 4% to the mass. It also includes the highest 75 concentration of V along with Ni suggesting the impact of ship 76 emissions (Kim and Hopke, 2008). 77

78 Among the four factors found for the fine particulate matter in 79

AEA station, the profiles are similar to those at the AQM site, but 80 there are quite different source contributions. The highest 81 contributions were from biomass burning (66%). The facility is 82 surrounded by local housing that uses biomass for cooking. The 83 next highest contributions were from motor vehicle traffic (17%) 84 and road dust (9%). Although there are major roads in the vicinity, 85 there is much less traffic than at the central AQM site. Sea salt 86 contributed 9% of the average mass concentration. 87

88 3.4. Back–trajectory techniques 89 90

An important question is whether the observed concentra-91 tions arise entirely from local sources or if there are contributions 92 from distant sources through long–range transport. Recently, 93 Begum et al. (2010) have shown that examination of the highest 94 concentrations indicates that they are often associated with 95 transported particulate matter. The Hybrid Single Particle 96 Lagrangian Integrated Trajectory (HYSPLIT) back–trajectories 97 (Draxler and Rolph, 2010) were used to trace possible medium and 98 long–range transport of pollution to the AQM measurement site. 99

100 Soil dust and smoke transboundary events were identified 101

based on the pseudo–elements soil and Knon, an indicator of 102 smoke, in the data base from 2000 to 2006 (Figures 5 and 6). 103 Evidence of these events was identified. There are three major soil 104 events on November 9, 2000, January 14, 2004, and February 21, 105 2004. Air parcel back–trajectories beginning at noon on these 106 dates are presented in Figures S5–S7 (see the SM). According to 107 the NASA Natural Hazards, there was a dust storm that began on 108 the 15

th of February 2004 in the Arabian Peninsula (Qatar) (Figure 109

S6 in the SM shows a later storm observed in satellite images on 110 February 21, 2004). Samples were collected on February 21

st in 111

which soil could be estimated to be 10 μg/m3 even though the 112

source was 3 500 km from Colombo. 113 114 There were 7 large ”smoke” episodes with concentrations 115

above 450 ng/m3 observed in Figure 6 on July 12, 2000, February 4, 116

13, 14, 21, and 23, 2004 and January 1, 2006. The 2000 trajectories 117 (see the SM, Figure S9) extend westward into the area east of 118 Africa and may have collected Saharan and Arabian Peninsula dust 119 leading to high K in ratios relative to Fe that appear to be “smoke” 120 rather than “soil”. In February 2004, there was a more complex 121 situation. The trajectories on days when samples were collected 122 are shown in Figures S7, S10–S13 (see the SM). In general, these 123 trajectories pass over Northern India where smoke from 124 agricultural fires backs up against the Himalaya Mountains. They 125 then pass westward into areas where they might also accumulate 126 soil dust. The high K on January 1, 2006 was from fireworks and 127 thus, trajectories were not calculated. 128

129

Seneviratne et al. – Atmospheric Pollution Research 2 (2011) xx-xx x

Figure 5. Time series plot of the daily soil concentrations. 1

2

Figure 6. Time series plot of the daily smoke concentrations. 3

4 4. Conclusions 5 6

The chemical composition data of fine air particulate matter 7 (PM2.5) collected from Colombo (AQM) and Orugodawatta (AEA 8 site) were studied using EPA–PMF to explore the possible 9 emissions sources. Factors that have been resolved from two sites 10 show very similar chemical composition profiles. Additionally, few 11 different chemical compositions have been obtained due to local 12 difference. HYSPLIT back–trajectory technique is very handy tool to 13 get an identification of various transboundary events. The 14 influence of a dust storm in Qatar, Iran, on February 15, 2004, 15 could be clearly observed by this technique even though the 16 source region was 3 500 km from Colombo. Similarly, transport of 17 smoke from Northern India in December 2003 could be observed. 18

19 Acknowledgements 20 21

The work is financially and technically supported by RCA IAEA 22 Project no RAS/8/082 and RAS/7/013, Atomic Energy Authority and 23 Central Environmental Authority. The authors gratefully 24 acknowledge the NOAA Air Resources Laboratory (ARL) for the 25 READY website (http://www.arl.noaa.gov/ready.php) used to 26 calculate the trajectories used in this publication. 27

28 29 30 31

Supporting Material Available 32 33

Source profiles resolved from the data collected at the AQM 34 site (Figure S1), Source contributions resolved from the data 35 collected at the AQM site (Figure S2), Source profiles resolved from 36 the data collected at the AEA site (Figure S3), Source contributions 37 resolved from the data collected at the AEA site (Figure S4), Back 38 trajectory for November 9, 2000 showing transport from China and 39 Mongolia (Figure S5), Back trajectory for January 14, 2004 (Figure 40 S6), Back trajectory for February 21, 2004 (Figure S7), NASA 41 satellite image of Qatar area on February 21, 2004 (Figure S8), 42 Ensemble of back trajectories for July 12, 2000 (Figure S9), 43 Ensemble of back trajectories starting on February 4, 2004 (Figure 44 S10), Back trajectories starting at different heights on February 13, 45 2004 (Figure S11), Back trajectories starting at different heights on 46 February 14, 2004 (Figure S12), Back trajectory calculated for 47 February 23, 2004 (Figure S13). This information is available free of 48 charge via the Internet at http://www.atmospolres.com. 49 50 References 51 52 Angove, D.E., Cant, N.W., Bailey, G.M., Cohen, D.D., 1996. The application 53

of PIXE to the mapping of contaminants deposited on a monolithic 54 automotive catalytic converter. Nuclear Instruments and Methods in 55 Physics Research Section B: Beam Interactions with Materials and 56 Atoms 109, 563–568. 57

Asian Development Bank and the Clean Air Initiative for Asian Cities (CAI–58 Asia) Center (ADB–CAIA), 2006. Country Synthesis Report on Urban Air 59 Quality Management: Sri Lanka, www.cleanairnet.org/caiasia/1412/ 60 csr/srilanka.pdf. 61

Begum, B.A., Biswas, S.K., Markwitz, A., Hopke, P.K., 2010. Identification of 62 sources of fine and coarse particulate matter in Dhaka, Bangladesh. 63 Aerosol and Air Quality Research 10, 345–353. 64

Begum, B.A., Biswas, S.K., Hopke, P.K., Cohen, D.D., 2006. Multi–element 65 analysis and characterization of atmospheric particulate pollution in 66 Dhaka. Aerosol and Air Quality Research 6, 334–359. 67

Begum, B.A., Kim, E., Biswas, S.K., Hopke, P.K., 2004. Investigation of 68 sources of atmospheric aerosol at urban and semi–urban areas in 69 Bangladesh. Atmospheric Environment 38, 3025–3038. 70

Chueinta, W., Hopke, P.K., Paatero, P., 2000. Investigation of sources of 71 atmospheric aerosol at urban and suburban residential areas in 72 Thailand by positive matrix factorization. Atmospheric Environment 34, 73 3319–3329. 74

Cohen, D.D., Garton, D., Stelcer, E., Hawas, O., Wang, T., Poon, S., Kim, J., 75 Choi, B.C., Oh, S.N., Shin, H.J., Ko, M.Y., Uematsu, M., 2004a. 76 Multielemental analysis and characterization of fine aerosols at several 77 key ACE–Asia sites. Journal of Geophysical Research-Atmospheres 109, 78 art. no. D19S12. 79

Cohen, D.D., Stelcer, E., Hawas, O., Garton, D., 2004b. IBA methods for 80 characterisation of fine particulate atmospheric pollution: a local, 81 regional and global research problem. Nuclear Instruments and 82 Methods in Physics Research Section B: Beam Interactions with 83 Materials and Atoms 219-220, 145–152. 84

Dockery, D.W., Pope, C.A., Xu, X., Spengler, J.D., Ware, J.H., Fay, M.E., 85 Ferris, B.G., Speizer, F.E, 1993. An association between air–pollution 86 and mortality in six U.S. cities. New England Journal of Medicine 329, 87 1753–1759. 88

Draxler, R.R., Rolph, G.D., 2010. Hybrid Single–Particle Lagrangian 89 Integrated Trajectory Model (HYSPLIT), http:/www.arl.noaa.gov/ready/ 90 hysplit4.html. 91

Hopke, P.K., Cohen, D.D., Begum, B.A., Biswas, S.K., Ni, B., Pandit, G.G., 92 Santoso, M., Chung, Y.S., Davy, P., Markwitz, A., Waheed, S., Siddique, 93 N., Santos, F.L., Pabroa, P.C.B., Seneviratne, M.C.S., Wimolwattanapun, 94 W., Bunpropab, S., Vuong, T.B., Hien, P.D., Markowicz, A., 2008. Urban 95 air quality in the Asian region. Science of the Total Environment 404, 96 103–112. 97

x Seneviratne et al. – Atmospheric Pollution Research 2 (2011) xx-xx

Hopke, P.K., Xie, Y., Raunemaa, T., Biegalski, S., Landsberger, S., Maenhaut, 1 W., Artaxo, P., Cohen, D.D., 1997. Characterization of gent stacked 2 filter unit PM 10 sampler. Aerosol Science and Technology 27, 726–735. 3

Kim, E., Hopke, P.K., 2008. Source characterization of ambient fine particles 4 at multiple sites in the Seattle area. Atmospheric Environment 42, 5 6047– 6056. 6

Malm, W.C., Sisler, J.F., Huffman, D., Eldred, R.A., Cahill, T.A. 1994. Spatial 7 and seasonal trends in particle concentration and optical extinction in 8 the United–States. Journal of Geophysical Research-Atmospheres 99, 9 1347 – 1370. 10

Markowicz, A., Haselberger, N., Dargie, M., Tajani, A., Tchantchane, A., 11 Valkovic, V., Danesi, P.R., 1996. Application of X–ray fluorescence 12 spectrometry in assessment of environmental pollution. Journal of 13 Radioanalytical and Nuclear Chemistry 206, 269–277. 14

Norris, G., Vedantham, R., Wade, K., Brown, S., Prouty, J., Foley, C., Martin, 15 L., 2008. EPA Positive Matrix Factorization (PMF) 3.0 Fundamentals & 16 User Guide, U.S. Environmental Protection Agency, Office of Research 17 and Development, EPA 600/R-08/108, Washington. Available from: 18 http://www.epa.gov/heasd/products/pmf/pmf.html. 19

Ramanathan, V., Chung, C., Kim, D., Bettege, T., Buja, L., Kiehl, J.T., 20 Washington, W.M., Fu, Q., Sikka, D.R., Wild, M., 2005. Atmospheric 21 brown clouds: impact on South Asian climate and hydrologic cycle. 22 Proceedings of the National Academy of Science 102, 5326–5333. 23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

Reff, A., Eberly, S.I., Bhave, P.V., 2007. Receptor modeling of ambient 60 particulate matter data using positive matrix factorization: review of 61 existing methods. Journal of the Air and Waste Management 62 Association 57, 146–154. 63

Santoso, M., Hopke, P.K., Hidayat, A., Dwiana, D.L., 2008. Sources 64 identification of the atmospheric aerosol at urban and suburban sites 65 in Indonesia by Positive Matrix Factorization. Science of the Total 66 Environment 397, 229–237. 67

Seneviratne, M.C.S., Mahawatte, P., Fernando, R.K.S., Hewamanna, R., 68 Sumithrarachchi, C., 1999. A study of air particulate pollution in 69 Colombo using a nuclear related analytical technique. Biological Trace 70 Element Research 71–72, 189–194. 71

United Nations Environmental Programme (UNEP), 2002. The Asian Brown 72 Cloud: Climate and Other Environmental Impacts, 73 URL:http://www.rrcap.unep.org 74

U.S. Environmental Protection Agency (USEPA), 2010. EPA Positive Matrix 75 Factorization (PMF) 3.0 Model, http://www.epa.gov/heasd/products/ 76 pmf/pmf.html. 77

78