Science of the Total Environment 601–602 (2017) 556–570

Contents lists available at ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

Source apportionment and health risk assessment among specific agegroups during haze and non-haze episodes in Kuala Lumpur, Malaysia

Nor Azura Sulong a, Mohd Talib Latif a,b,⁎, Md Firoz Khan c, Norhaniza Amil d, Matthew J. Ashfold e,Muhammad Ikram Abdul Wahab f, Kok Meng Chan f, Mazrura Sahani f

a School of Environmental and Natural Resource Sciences, Faculty of Science and Technology, Universiti Kebangsaan Malaysia, 43600 Bangi, Selangor, Malaysiab Institute for Environment and Development (Lestari), Universiti Kebangsaan Malaysia, 43600 Bangi, Selangor, Malaysiac Centre for Tropical Climate Change System, Institute of Climate Change, Universiti Kebangsaan Malaysia, 43600 Bangi, Selangor, Malaysiad Environmental Technology Division, School of Industrial Technology, Universiti Sains Malaysia, 11800 USM, Penang, Malaysiae School of Environmental and Geographical Sciences, University of NottinghamMalaysia Campus, 43500 Semenyih, Selangor, Malaysiaf Environmental Health and Industrial Safety Program, School of Diagnostic Science and Applied Health, Faculty of Health Sciences, Universiti Kebangsaan Malaysia, Jalan Raja Muda Abdul Aziz,50300 Kuala Lumpur, Malaysia

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• PM2.5 concentrations during haze pe-riods were 3–5 times higher than non-haze.

• Secondary inorganic aerosols dominat-ed the compositions of PM2.5 duringhaze.

• NAME model and GFAS indicate fireemissions from Sumatera, Indonesia.

• Biomass burning and traffic emissionare main sources of PM2.5.

• Carcinogenic and non-carcinogenichealth risk affects adult and infant themost.

⁎ Corresponding author at: School of Environmental andE-mail address: talib@ukm.my (M.T. Latif).

http://dx.doi.org/10.1016/j.scitotenv.2017.05.1530048-9697/© 2017 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 9 February 2017Received in revised form 13 April 2017Accepted 16 May 2017Available online xxxx

Editor: D. Barcelo

This study aims to determine PM2.5 concentrations and their composition during haze and non-haze episodes inKuala Lumpur. In order to investigate the origin of the measured air masses, the Numerical Atmospheric-dispersion Modelling Environment (NAME) and Global Fire Assimilation System (GFAS) were applied. Sourceapportionment of PM2.5 was determined using Positive Matrix Factorization (PMF). The carcinogenic and non-carcinogenic health risks were estimated using the United State Environmental Protection Agency (USEPA)method. PM2.5 samples were collected from the centre of the city using a high-volume air sampler (HVS). The re-sults showed that the mean PM2.5 concentrations collected during pre-haze, haze and post-haze periods were24.5 ± 12.0 μg m−3, 72.3 ± 38.0 μg m−3 and 14.3 ± 3.58 μg m−3, respectively. The highest concentration ofPM2.5 during haze episode was five times higher than World Health Organisation (WHO) guidelines. Inorganiccompositions of PM2.5, including trace elements and water soluble ions were determined using inductivelycoupled plasma-mass spectrometry (ICP-MS) and ion chromatography (IC), respectively. The major trace ele-ments identified were K, Al, Ca, Mg and Fe which accounted for approximately 93%, 91% and 92% of the overallmetals' portions recorded during pre-haze, haze and post-haze periods, respectively. For water-soluble ions, sec-ondary inorganic aerosols (SO4

2−, NO3− and NH4

+) contributed around 12%, 43% and 16% of the overall PM2.5mass

Keywords:Transboundary smoke hazeBiomass burningPM2.5 aerosolsPMFCarcinogenic and non-carcinogenic health risk

Natural Resource Sciences, Faculty of Science and Technology, Universiti KebangsaanMalaysia, 43600Bangi, Selangor,Malaysia.

557N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

during pre-haze, haze and post-haze periods, respectively. During haze periods, the predominant source identi-fiedusing PMFwas secondary inorganic aerosol (SIA) andbiomass burningwhere theNAME simulations indicatethe importance of fires in Sumatra, Indonesia. The main source during pre-haze and post-haze were mix SIA androad dust aswell asmineral dust, respectively. The highest non-carcinogenic health risk during haze episodewasestimated among the infant group (HI = 1.06) while the highest carcinogenic health risk was estimated amongthe adult group (2.27 × 10−5).

© 2017 Elsevier B.V. All rights reserved.

1. Introduction

Smoke-haze episodes in Southeast Asia (SEA) occur almost everyyear (Nichol, 1998; See et al., 2006). Haze is defined as low visibility con-dition (b10 km) where the human eye's ability to differentiate betweenobjects and the surroundings becomes restricted (Sun et al., 2006; Tan etal., 2009). Thick smoke-induced haze has enveloped SEA regions, mostnotablyMalaysia, Singapore, Indonesia and Thailand, as a result of recur-rent slash-and-burn agricultural activities (Betha et al., 2013). Biomassburning, particularly within peat soil areas in Sumatra and Kalimantan,Indonesia, contributes to high emissions of fine particles and atmospher-ic gases into the atmosphere (Abas, 2004; Heil and Goldammer, 2001;Muraleedharan et al., 2000; Zhao et al., 2013). Chemical interactions be-tween primary gases, organic substances and particles contribute to highlevels of secondary aerosols along the trajectories from the burning areas(Beringer et al., 2003; Heil and Goldammer, 2001; Koe et al., 2001;Rastogi et al., 2014). A previous study has shown that transboundaryhaze episodes in Malaysia are not caused by external sources alone butare in fact exacerbated by local emissions (Afroz et al., 2003) in additionto El-Niño's Southern Oscillation (ENSO) phenomenon which prolongsthe dry season (Aiken, 2004; Jones, 2006; Juneng and Tangang, 2005).Meanwhile for non-haze periods, vehicular emissions have been foundto account for over 30% of the total PM2.5 emissions in the urban areasof the Klang Valley (Rahman et al., 2011).

A detailed understanding of the sources of particulatematter (PM) isessential if PMpollution is to be effectively controlled andmanaged. Thereceptor models which are often used for source apportionment analy-sis are principal component analysis-multiple linear regression (PCA-MLR) (Thurston and Spengler, 1985), chemical mass balance (CMB)(Watson et al., 2001), positive matrix factorization (PMF) (Paatero,1997) and Unmix model (Lewis et al., 2003). In Malaysia, PM2.5 andsource apportionment studies during non-haze period have been car-ried out by Khan et al. (2016a, 2016b) using the PMF method wheremotor vehicle emissions coupled with biomass burning were identifiedas the predominant source in a suburban area. Another study undertak-en by Amil et al. (2016) used PMF to apportion sources in the urban-industrial environment of Petaling Jaya, Malaysia during haze andnon-haze periodswhere themain source during hazewas secondary in-organic aerosols (SIA) coupledwith biomass burning. In a source appor-tionment study carried out by Rahman et al. (2015), PMF was alsoapplied in order to apportion PM2.5 in an urban area of the Klang Valley,Malaysia and listed vehicle emissions as the primary source of hazeevents. Ee-Ling et al. (2015) determined the sources of PM2.5 usingPCA-MLR and revealed the predominant sources in Kuala Lumpur, Ma-laysia, asmotor vehicles and soil dust. In Singapore, Balasubramanian etal. (2003) performed PCA and listed soil dust, industrial factors, biomassburning and automobile emissions, sea salt and oil combustion as thefactors contributing to PM2.5 aerosols.

Haze occurrences have often been associatedwith adverse health ef-fects. A study conducted by Nasir et al. (2000) claimed that 285,227asthma attacks, 3889 cases of chronic bronchitis in adults, 118,804cases of bronchitis in children, 2003 respiratory hospital admissionsand 26,864 emergency room visits were reported during the 1997haze episode in Malaysia. Another study by Brauer and Hisham-Hashim (1998) revealed that therewas a significant increase in hospitaladmissions for asthmatic and other respiratory-related problemsduring

regional smoke-haze episodes in SEA. In addition, a study by Sahani etal. (2014) revealed that there were significant associations betweenhaze events and natural and respiratory mortality. The most importantissue to be addressed during haze periods is consequently public healthconcerns, especially those relating to population which have been clas-sified as “sensitive”, for example, children (below 12 years old), the el-derly (senior citizens) as well as immunocompromised persons (Afrozet al., 2003). Through health risk assessments, the nature and magni-tude of these health risks can be characterized. In Malaysia, Khan et al.(2016a) successfully conducted health risk assessment on PM2.5 aerosolcollected from a suburban area. The results of which demonstrated thatthe non-carcinogenic health risk was at a safe level while the lifetimecancer risk was slightly above the acceptable level.

Malaysia experienced deterioration in air quality between Septemberand October 2015. This was due to a transboundary haze episode whichoriginated fromSumatra andKalimantan, Indonesia andwas transportedby a south-westerly wind. 34 out of 52 stations throughout the countryrecorded an Air Pollutant Index (API) of over 200 (unhealthy) on 15 Sep-tember 2015, which was the worst haze day recorded (DOE, 2015). Thisstudy aims to determine the concentrations of PM2.5 in Kuala Lumpurand its inorganic compositions (trace elements and water-soluble ions)during pre-haze, haze and post-haze periods. Numerical Atmospheric-dispersion Modelling Environment (NAME) together with the GlobalFire Assimilation System (GFAS) was used to examine how PM2.5 emit-ted by fires in the region was transported towards Kuala Lumpur. ThePMF method was used to apportion possible sources of PM2.5 withthese three different periods. This study also estimated the carcinogenicand non-carcinogenic health risks among specific age groups during pre-haze, haze and post-haze episodes. The quantitative estimation of carci-nogenic and non-carcinogenic health risks will highlight the need to ef-fectively control and manage the sources of PM2.5.

2. Materials and methods

2.1. Study area

The sampling took place in the middle of Kuala Lumpur city centre(3° 8′ 20.4108″ N, 101° 41′ 12.678″ E, 56 m above sea level). The PM2.5

aerosol sampling was conducted on the rooftop (30 m above groundlevel) of a building within the compound of Universiti Kebangsaan Ma-laysia Kuala Lumpur's campuswhich is located onRajaMudaAbdul AzizRoad, Kuala Lumpur. Kuala Lumpur itself is Malaysia's capital city. It islocated within the Klang Valley and situated in the middle of the westcoast of theMalaysian Peninsula. In 2015, Kuala Lumpur had an estimat-ed population of 1.78 million people in an area of just 243 km2 with anaverage annual population growth rate of 2.4% (DOSM, 2016). The sam-pling site was 10, 30 and 40 m away from three congested roads inKuala Lumpur city centre, namely: Chow Kit Road, Raja Muda AbdulAziz Road and Pahang Road, respectively. The sampling site can be con-sidered as an urban background site with influence by traffic emissions.Supplementary 1 shows the location of the sampling site.

2.2. PM2.5 field campaign

Taking into account the history of haze episodes in Malaysia whichpredominantly occur during the southwest monsoon (i.e. June–

558 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

September every year), we deliberately chose to do sampling in the sec-ond half of the year. Sampling was undertaken for eight consecutivemonths, namely June–January 2016, with nine samples taken permonth. The period between June andAugust 2015 (n=27)was catego-rized as a pre-haze event since a haze event occurred in September 2015and October 2015 (n = 18). The period between November 2015 andJanuary 2016 was categorized as a post-haze event (n = 27). Hazedays were defined as the days with PM2.5 mass concentration of35 μg m−3 and above (Balasubramanian et al., 2003; See et al., 2006;Xu et al., 2014) and visibility b 10 km (Sun et al., 2006; Tan et al.,2009). A total of 72 field filters plus eight field blank filters were collect-ed during the sampling period. A high volume PM2.5 sampler (Tisch Co.,USA)was deployed for sampling alongwith the use of quartzmicrofiberfilter paper (20.3 cm×25.4 cm,Whatman™, UK). The sampler operatedat a flow rate of 1.13 m3 min−1. The samples were collected on week-days (Monday to Friday) andweekends (Saturday and Sunday) at dura-tion of 24 h per sample.

2.3. Gravimetric mass measurement

First of all, the quartz filter paper was pre-baked (wrapped in alu-minium foil) at 500 °C, for 4 h in a furnace (Carbolite Eurotherm 301Controller, UK) to remove carbonaceous contaminants. The quartz filterpaper was weighed before and after sampling using a five-digit elec-tronic microbalance (Mettler Toledo, USA) with a reading precision of0.01 mg. The filter paper was stored in a desiccator for 24 h both beforeand after sampling so as to reach humidity equilibrium (conditioning)of between 40% - 45%. The collected samples were rewrapped in alu-minium foil and stored in a freezer at under−18 °C in order to preventevaporation of volatile components before chemical analysis.

2.4. Microwave-assisted acid digestion and trace elements analysis

The microwave-assisted digestion system was used to extract vari-ous fractions of the trace elements (Kaufmann and Christen, 2002;Lopez-Avila et al., 1994; Pan et al., 2000). Details of the procedureused were as per described in Khan et al. (2016a). In short, a portion(1/16) of shredded filter sample was placed in a Teflon digestion vessel(DAP-60 K, Germany) made of tetrafluoromethoxylene (TFM) (SK-10,Milestone, Germany). Then, 12 ml of analytical grade HNO3 (65%,Merck KGaA, Germany) and 3 ml of analytical grade H2O2 (40%, MerckKGaA, Germany) were added to the vessel under a fume hood as the di-gestion reagents. Next, the Teflon vessel was capped and placed in ami-crowave digester (Start D, Milestone, Germany) to dissolve the samples.Six samples were placed in the microwave at a time. The microwavefunctioned in three steps: 1) 20 min of ramping to 180 °C; 2) 15 minof a steady state at 220 °C; and 3) 10 min of cooling down to 60 °C.After the procedure, all Teflon vessels were placed in cold water.The aim of this procedure was to reduce the residual pressure insidethe Teflon vessels. The digested samples were then filtered directlyinto 50 ml in a volumetric flask using a disposable syringe filter,0.20 μm pore (Pall Gelman Laboratory, MI, USA), coupled with a50 cm3 ml−1 Terumo syringe (Terumo®, Tokyo, Japan). Ultrapurewater (UPW) with a resistivity of 18.2ΩM/cm (Easypure® II, ThermoScientific, Canada)was added to the mark. Then, the samples were sub-jected to trace elements analysis using ICP-MS (Perkin Elmer ELAN9000, USA). In order to calibrate the instrument, multi-element calibra-tion standards 2 and 3 (CertiPUR® PerkinElmer, USA) were used withdifferent calibration curves each time. The analysis consisted of twomodes, namely: high concentration metals (Al, Ca, Fe, Mg, Mn and Zn)and low concentration metals (Ag, As, Ba, Cd, Co, Cr, Cu, K, Ni, Pb, Rb,Sr and V). After being subjected to ICP-MS, the vessels were then soakedwith acid bath (6% HNO3) for 24 h before being cleaned using ultrapurewater.

2.5. Water soluble ion analysis

A portion (1/8) of filter sample was cut into pieces. The shredded fil-ter paper was placed in a 20 ml centrifuge tube with a screw cap. UPWwas added up to the mark. The sample was then extracted for 40 minunder ultrasonic conditions using an ultrasonic bath (Elmasonic S70H,UK). After that, the solution was centrifuged at 2500 rpm speed for30 min (Kubota 5100, Japan). The extract was then filtered directlyinto a 50 ml volumetric flask using a disposable syringe filter of0.20 μm pore (Pall, USA) and a 50 cm3 ml−1 Terumo syringe. UPWwas added up to the mark. The extract was next divided into two vialtubes, each containing 15 ml, for anionic and cationic analysis.

In order to measure anions (Cl−, NO3− and SO4

2−), ion chromato-graph (Metrohm, 850 Professional IC, Switzerland) with Metrosep ASUPP 5 (4 × 150 mm) columns was used. For cations (Na+, NH4

+,Mg2+, Ca2+ and K+), ion chromatograph (Metrohm, 850 ProfessionalIC, Switzerland) with Metrosep C 4100 (4 × 100 mm) columns wasused. An eluent of 2.7 mM Na2CO3 and 0.3 mM NaHCO3 was preparedand used for the detection of anions with a pump flow rate of1.5 ml min−1. For the analysis of cations, 2.6 mM methane sulphonicacid solution was used as an eluent with 1 ml min−1 pump flow rate.Calibration curves, identified from five different points of 0.5, 1, 2, 5and 10 ppm certified standard solutions (CertiPUR® Merck, Germany),were used. Ions were identified based on their retention time.

2.6. Quality assurance and quality control

In this study, blanks (field, laboratory and reagent blanks)were usedin each analysis in order to correct the concentration of the chemicalcompositions of the samples. The field blank filters were handled inthe same way as an actual sample filter (except that the sampler wasturned off) and reweighed as a quality control (QC) to detect anyweightchanges due to filter handling. In addition, Urban Particulate MatterSRM 1648a produced by the National Institute of Standards and Tech-nology (NIST), USA was used as a reference material to determine therecovery percentage of the trace elements using the same method asused with the field samples. The method detection limit (MDL) for thetrace elements was calculated as three times the standard deviation often replicates of the reagent blank. A similar approach was used fortheMDL of the water-soluble ions, but with field blanks. In order to cal-culate the percentage recoveries of water-soluble ions, cation and anionstandards from (CertiPUR® Merck, Germany) were used. The percent-age of recovery for trace elements and water-soluble ions with theirMDLs are presented in Supplementary 2.

2.7. Backward trajectories and fire hotspots

To understand the origin of the measured air masses, the NumericalAtmospheric-dispersion Modelling Environment (NAME) (Jones et al.,2007), a Lagrangian particle dispersion model, was used. For eachday of the study period (June 2015–January 2016), use of NAMEwas adopted to calculate batches of 480,000 inert backwardtrajectories. The trajectories started at the latitude-longitude coordi-nates of the measurement site in Kuala Lumpur within an altituderange of 0–100 m and ran for five days. Every 15 min, the locations ofall trajectories within the lowest 100 m of the model atmospherewere recorded on a grid with a horizontal resolution of 0.5625° longi-tude by 0.375° latitude. From this information, the sensitivity of themeasured air mass to emissions occurring in the previous 5 days withina particular grid cell could be derived (units [g m−3] / [g m−2 s−1], i.e.s m−1). The trajectories were calculated using three-dimensional mete-orological fields produced by theUnited Kingdom'sMet Office's Numer-icalWeather Prediction tool, the UnifiedModel (UM). These fields had ahorizontal grid resolution of 0.35° longitude by 0.23° latitude and 59vertical levels below ~30 kmandwere available at 3 h intervals. Verticalvelocities were obtained from the UM and available at grid nodes. The

559N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

sub-grid scale process of turbulence was parameterised in NAME(Morrison and Webster, 2005).

In conjunctionwith the NAME calculationsmade, the Global Fire As-similation System (GFAS) (Kaiser et al., 2012) v1.2 PM2.5 emissions data(downloaded from: http://join.iek.fz-juelich.de/access) was used in thisstudy. This dataset relies on fire radiative power (FRP) observations ob-tained from the Terra and Aqua satellites byModerate Resolution Imag-ing Spectroradiometer (MODIS) instruments. Hertwig et al. (2015)provided a detailed discussion of the uncertainties associated with fireemission products such as GFAS in the context of modelling studies fo-cused on Southeast Asia. For example, they note that while GFAS doesaccount for peat soil fires, the detailed distribution of peat soil withinSumatra is not captured accurately. In addition, GFAS uses a persistencealgorithm in an effort to correct for obstruction of hot spots (e.g. bycloud) during satellite overpasses (Hyer et al., 2013).

By combining the emission sensitivities derived from NAME withthe GFAS emissions it is possible to calculate a modelled PM2.5 concen-tration, due only to emissions occurring within the five days timescaleof the backward trajectories, at the Kuala Lumpur measurement site(dimensionally, s m−1 × g m−2 s−1 = g m−3). For consistency withour backward trajectory calculations, the daily GFAS emission fieldswere averaged over thefive days duringwhich the trajectories travelled(e.g. trajectorieswhich started on 1 Junewere combinedwith emissionsaveraged over 28 May–1 June). Given the uncertainty associated withthe spatial and temporal variations in PM2.5 emissions fromfires, thene-glect of PM2.5 from sources apart from fires, and the lack of detailed lossprocesses during atmospheric transport,we concentrated on qualitativecomparisons between simulated and observed PM2.5.

2.8. Meteorological data, API and statistical analysis

All meteorological data, such as temperature, rainfall, relative hu-midity, visibility, wind speed and directionwere obtained from PetalingJaya Station which is located about 10 km from the sampling site andmaintained by the Department of Environment (DOE). Meanwhile,theAir Pollution Index (API) datawas collected fromBatuMuda Station,which is the nearest meteorological station to the sampling site and lo-cated in Kuala Lumpur city centre. API in Malaysia is calculated usingfive major air pollutants, namely, particles with a diameter size ofb10 μm (PM10), sulphur dioxide (SO2), nitrogen dioxide (NO2), surfaceozone (O3) and carbon monoxide (CO). The highest sub-index fromeach pollutant was selected for API for the period of time in question.Statistical analysis, including Pearson correlation and OneWay Analysisof Variance (ANOVA), was performed using IBM SPSS Statistics Version23.0. Pearson correlation was used to measure the strength and direc-tion of the linear association between two variables. In this study, thePearson correlation test was also used to identify the association be-tween PM2.5 concentrations and API as well as PM2.5 concentrationsand meteorological parameters, the relationships between PM2.5 con-centrationsmeasured by HVS and PMF, aswell as the link between con-centrations of K+ and nss-SO4

2− in PM2.5. The ANOVA test wasconducted to examine whether there were any statistically significantdifferences between the means of two or more independent groups. Inthis study, the ANOVA test was also conducted to compare the concen-trations of PM2.5 and water-soluble ions between three haze cycles.

2.9. Source apportionment analysis using PMF

Source apportionment of PM2.5 in the study area was performedusing the PositiveMatrix Factorization (PMF) 5.0model from theUnitedStates Environmental Protection Agency (US EPA). PMF has been usedas the tool to apportion possible sources contributing to the emissionsprofile of the sampling location in many other studies (Ballinger andLarson, 2014; Hasheminassab et al., 2014; Martins et al., 2015; Park etal., 2012; Srimuruganandam and Shiva Nagendra, 2012; Stanimirovaet al., 2011; Wang et al., 2009). To identify the set of factors which

represent the major sources of emissions, measured concentrations atthe receptor sites were used. PMF is a multivariate factor analysis thatdecomposes a matrix sample data into two matrices which are factorcontributions and factor profile (Paatero, 1997). The PMF model calcu-lation is presented by the following equation:

Xij ¼ ∑N

k¼1gik � fkj þ eij

where Xij is the concentration of chemical species jmeasured on samplei, N is the number of factors contributing to the samples, fkj is the con-centration of species j in factor profile k, gik is the relative contributionof factor k to sample i, and eij is error of the PMF model for the speciesj measured on sample i. The aim is to determine the values of gik andfkj that best reproduce Xij. The values of gik and fkj are adjusted until aminimum value of Q is found. Q is shown in the equation below:

Q ¼ ∑n

i¼1∑m

j¼1

Xij−∑Nk¼1gik � fkjσ ij

!2

where n is the number of samples,m is the number of species and σij isthe uncertainty of ambient concentration of species j in i. The uncertain-ty value of each sample variable was determined through the use of thefollowing equation:

σ ij ¼ 0:01 Xij þ X j� �

where σij is the estimation of the measurement error for the jth speciesin the ith sample, Xij is the observed elements concentration and Xj isthe mean value. Further details of the estimation of uncertainty can befound elsewhere (Khan et al., 2015). 5% of additional uncertainty wasadded to recover errors in gravimetric mass measurements, calibrationcurves preparation or from filter papers preparation.

The PMFmodel was run based on two data sets whichwere concen-tration data and uncertainty data. Before running the model, all datawas pre-treated and validated. For example, if the concentration wasbelow the MDL, it was replaced with half of the MDL and if the specieshad N50% of the dataset below the MDL, it was discarded. The finaldataset consisted of 27 elements, including PM2.5 mass. The signal-to-noise ratio (S/N ratio) for each species was calculated based on boththe concentration data and uncertainty data of the species. Generally-speaking, species were divided into three different groups relating totheir S/N ratio. As modified from Amil et al. (2016) and Khan et al.(2016a), the species were considered ‘strong’ if S/N N 2, ‘weak’ if theS/N is 0.2 ≤ S/N ≤ 2, and ‘bad’ if, S/N ≤ 0.2. From the PMF results, NH4

+

was set as a ‘weak’ species. Additionally, PM2.5 mass was set as ‘weak’to ensure it did not affect the overall PMF performance. We systemati-cally tested various numbers of possible factors to obtain the best andmost meaningful factor analysis outputs, considering factor solutionsfrom four to seven factors with a default seed value and FPEAK = 0.

Furthermore, we conducted PMF analysis for four different groupswhich were (1) pre-haze (2) haze (3) post-haze and (4) the overallsampling period dataset. For the overall sampling period, five factorswere identified as the most suitable source profiles of PM2.5 in KualaLumpur's city centre. These five factors were chosen based on the low-est Q (robust) of 1628.5with theQ (true)=Qexp value of 1.18, after 447computational steps and the convergence of the PMF results. Followingthe identification of five source profiles from the overall dataset, the re-maining three datasets were treated in the same manner. For the pre-haze dataset, PMF was resolved on 15 runs with a seed value of 5 withfive factors, based on the lowest Q (robust) of 268.5 with the Q(true) = Qexp value of 0.83 after 544 computational steps and the con-vergence of the PMF results. The haze dataset was resolved on 3 runswith a seed value of 2 with the lowest Q (robust) of 141.3 with the Q(true) = Qexp value of 0.58 after 309 computational steps and the con-vergence of the PMF results. Meanwhile, for the post-haze dataset, five

560 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

factors were identified as the best solution which was resolved on 20runs with a seed value of 5 with the lowest Q (robust) of 194.2 withthe Q (true) = Qexp value of 0.72 after 594 computational steps andthe convergence of the PMF results.

2.10. Health risk assessment

In this study, the health risk assessment was divided into two parts,namely carcinogenic and non-carcinogenic health risk. Supplementary3 shows the exposure factors in calculating the exposure dose basedon the method modified by Jamhari et al. (2014). A detail discussionof the health risk assessment was added in the Supplementary 4.

3. Results and discussion

3.1. Concentration of PM2.5 during pre-haze, haze and post-haze episodes

Overall PM2.5 mass, trace elements and water-soluble ionic concen-trations for pre-haze, haze andpost-haze episodes are presented in Sup-plementary 2. The daily variability of PM2.5 concentration is simplifiedin Fig. 1. The mean concentrations of PM2.5 during pre-haze, haze andpost-haze were 24.5 ± 12.0 μg m−3, 72.3 ± 38.0 μg m−3 and 14.3 ±3.58 μg m−3, respectively. The one-way analysis of variance (ANOVA)test was used in order to compare concentrations between the three pe-riods of sampling, namely; pre-haze, haze and post-haze episodes. Sta-tistically significant differences (p b 0.05)were foundbetween the threegroups. A Tukey post-hoc test revealed that PM2.5 concentrations duringthe haze period were significantly higher (p b 0.05) compared to pre-haze and post-haze. During the haze period, all samples exceededWHO guidelines and the standards set by the USEPA while 38% of thesamples exceeded the Malaysia Ambient Air Quality Standards 2015(IT-1). However, during pre-haze, only 42% of the samples exceeded24 h PM2.5 WHO guidelines while 12% of the samples exceeded theUSEPA. During post-haze, none of the days exceeded even the WHOguidelines. The average concentration of PM2.5 recorded in this studywas higher when compared to a study by Pinto et al. (1998) in 1997(62.1 μg m−3) and Amil et al. (2016) in Kuala Lumpur on similarhaze episodes in 2011 and 2012 (61.2 ± 24.3 μg m−3). In Singapore,Betha et al. (2014) reported the highest concentration of PM2.5

(329 μg m−3) during a haze episode in 2013.In addition to ground measurements, the simulated PM2.5 concen-

tration in Kuala Lumpur was also calculated by combining NAME emis-sion sensitivity and GFAS wildfire emissions for the average of theprevious 5 days as shown in Fig. 1. In close agreement with the mea-surements, the figure shows that the simulated onset of the haze epi-sode occurred in early September 2015, peaked on 15 September2015 and ended late October 2015. The simulated results also indicatethat measured exceedances of the WHO guidelines in the pre-haze

Fig. 1. Daily variability of PM2.5 concentration (ground) and PM2.5 simulated by combining NAM

period are linked to fire emissions, and that Kuala Lumpur was not af-fected by a biomass burning source of PM2.5 in the post-haze period.These results underscore the importance of biomass burning in drivingregional haze episodes, and provide further evidence, in support ofstudies such asHertwig et al. (2015), that NAMEcan accurately simulatethe main features of these events. When compared on a day-to-daybasis, the PM2.5 concentration from the ground measurement wasoften higher than the PM2.5 concentration estimated by NAME-GFAS.However, there were certain days when the PM2.5 concentration esti-mated by NAME-GFAS was greater in comparison to the groundmeasurement.

The concentration of PM2.5 was associated with the direction of theprevailing winds, as shown by the wind roses (Supplementary 5) andNAME emission sensitivity with GFAS wildfire emissions (Fig. 2). Fig. 2shows examples for three days of the study period. The NAME calcula-tions for 12 June (pre-haze) and 12 September (haze) show transportpaths which are typical for the time of year, with air masses passingover Sumatra on route to Kuala Lumpur. On 12 June, there was little re-gionalfire activity and associated PM2.5 emissions. In contrast, on 12 Sep-tember emissions fromwidespread fires in Sumatra were transported tothe measurement site in Kuala Lumpur. By 10 December 2015 (post-haze), the Northeast Monsoon had begun and fire activity in SEA wasmuch reduced. Overall, periods with an increased number of hotspotslead to the increased concentration of PM2.5 (Kusumaningtyas andAldrian, 2016). Also, the concentration of particulate matter dependson the El Niño/Southern Oscillation (ENSO) which directly affects con-vective activity, atmospheric stability and rainfall (Juneng et al., 2011).Haze particles, coupled with local particle emissions, increased the par-ticulate concentration in the atmosphere (Afroz et al., 2003). Duringthe non-haze period, traffic emissions were found to attribute to almost70% of the total particulate matter emissions in the city (Awang et al.,2000). However, the concentration of PM2.5 measured during the post-haze event was low because it occurred during the northeast monsoonwhen the intensity of rainfall is much higher compared to the southwestmonsoon. The wet deposition process diminished the particles from theatmosphere (Artaxo et al., 2002; Fan et al., 2004; Liss and Johnson, 2014;Varikoden et al., 2011).

3.2. Relationship between PM2.5 concentrations collected during pre-haze,haze and post-haze episodes with API and meteorological parameters

Fig. 3 shows the meteorological parameters measured during pre-haze, haze and post-haze episodes. In order to identify the relationshipbetween meteorological parameters and API towards PM2.5 mass, thePearson correlation test was conducted. The results showed that duringthe pre-haze episode, there was a positive correlation (r = 0.562;p b 0.05) between API and the PM2.5mass. Likewise, during the haze ep-isode, there was also a positive significant correlation (r = 0.466;

E emission sensitivity and GFAS wild-fire emissions for the average of the previous 5 days.

Fig. 2.NAME emission sensitivity (left column) andGFASwildfire emissions (right column) for selected dayswithin the pre-haze (12 Jun 2015, top), haze (12 Sep 2015,middle) and post-haze (10 Dec 2015, bottom) periods. The emission sensitivity values are derived from 5 day backward trajectories, and the wild-fire emissions are averaged over the same 5 days. Themeasurement site in Kuala Lumpur is marked with a red circle. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

561N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

p b 0.05) between API and the PM2.5 mass. Although we measured thePM2.5 concentration, PM2.5 can be treated as a fixed weight fraction ofPM10 since PM2.5 is a component of PM10. Amil et al. (2016) found thatthe average PM2.5/PM10 ratio for the Klang Valley was 0.72 ± 0.18 witha good correlation between them (r = 0.963; p b 0.0001) and betweenthe PM2.5 mass and API (r= 0.763; p b 0.001). Their study also demon-strated that PM2.5 was positively correlated with API during haze with aPM2.5/PM10 ratio of 0.74. Thus, it is suggested that PM2.5 is directly pro-portional to API. Meanwhile, for the post-haze episode, there was aweak correlation (r=0.176; p b 0.05) between API and the PM2.5 mass.

With respect to the relationship between visibility and the PM2.5

mass, there was a strong negative correlation (r = −0.692; p b 0.005)during the pre-haze period. In addition, during the haze episode, astrong negative correlation (r=−0.631; p=0.005)was also noted be-tween visibility and the PM2.5mass. Thisfinding indicates thatwhen thePM2.5 concentration increases, visibility decreases. Low visibility hasbeen found to relate to haze events (Ma et al., 2014). Fujii et al.(2015) stated that on unhealthy air quality days, the visibility measured

in Petaling Jaya, Selangor, Malaysia dropped to below 2.7 km. Further-more, Pui et al. (2014) reported that an increase of 500–800 μg m−3

of PM2.5 could reduce visibility to b100m. Suspended particles absorbedand scattered light leading to visibility impairment (Pui et al., 2014). Ac-cording to Kusumaningtyas et al. (2016), low visibility conditionsurrounded the area of peatland burning in Central Kalimantan, Indone-sia as a result of high aerosol loading. Strong evidence suggests thatPM2.5 is an important source of visibility impairment in both urbanand remote areas (USEPA, 2009). During the post-haze episode, therewas no significant correlation (r=−0.253; p N 0.05) between visibilityand the PM2.5 mass detected. A poor correlation between visibility andPM2.5 in tropical areas like Malaysia is always expected. This is becausevisibility is affected by the amount of moisture and water vapor in theatmosphere in addition to the quantity of particles and trace gases pres-ent. A high relative humidity usually affects the visibility variability par-ticularly during post-haze period.

A significant negative correlation betweenwind speed and the PM2.5

mass was observed during both the post-haze (r = −0.456; p b 0.05)

0

50

100

150

200

250

300

1

10

100

1000

10000

100000

AP

I in

dex

visi

bilit

y (m

)

Visibility (m) API

pre-haze haze post-haze

0

0.5

1

1.5

2

2.5

0

10

20

30

40

50

60

70

80

90

100

2015

-06-

1120

15-0

6-15

2015

-06-

1920

15-0

6-23

2015

-06-

2720

15-0

7-01

2015

-07-

0520

15-0

7-09

2015

-07-

1320

15-0

7-17

2015

-07-

2120

15-0

7-25

2015

-07-

2920

15-0

8-02

2015

-08-

0620

15-0

8-10

2015

-08-

1420

15-0

8-18

2015

-08-

2220

15-0

8-26

2015

-08-

3020

15-0

9-03

2015

-09-

0720

15-0

9-11

2015

-09-

1520

15-0

9-19

2015

-09-

2320

15-0

9-27

2015

-10-

0120

15-1

0-05

2015

-10-

0920

15-1

0-13

2015

-10-

1720

15-1

0-21

2015

-10-

2520

15-1

0-29

2015

-11-

0220

15-1

1-06

2015

-11-

1020

15-1

1-14

2015

-11-

1820

15-1

1-22

2015

-11-

2620

15-1

1-30

2015

-12-

0420

15-1

2-08

2015

-12-

1220

15-1

2-16

2015

-12-

2020

15-1

2-24

2015

-12-

2820

16-0

1-01

2016

-01-

0520

16-0

1-09

2016

-01-

1320

16-0

1-17

Win

d sp

eed

(m/s

)

Tem

pera

ture

(0 C

) / R

H (

%)

/ rai

nfal

l (m

m)

Temperature (0C) RH (%) Rainfall (mm) Wind speed (m/s)

pre-haze haze

post-haze

Fig. 3. API and meteorological factors during different haze cycle measured in the sampling campaign.

562 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

and haze episodes (r = −0.461; p = 0.05). PM2.5 is inversely propor-tional with wind speed, such that when the wind speed is high, thePM2.5 mass decreases. This is because the wind dilutes and dispersesPM2.5, which results in a reduction in PM2.5 concentrations. Windspeed and wind direction play a significant role in the severity of thesmoke haze area (Dotse et al., 2016). In a study by Charron andHarrison (2005), the PM2.5 concentration was found to be reduced byapproximately 30% when the wind was stronger. It was also reportedthat when the wind speed was greater (N9 ms−1), the PM2.5 massdropped from about 25 to 18 μg m−3. In our study, no significant corre-lation (r = 0.281; p N 0.05) was observed between PM2.5 and windspeed during the pre-haze period. In addition, no significant correlationwas found between temperature, rainfall and relative humidity and thePM2.5 collected during each haze cycle.

3.3. Concentrations of trace elements andwater soluble inorganic ions dur-ing pre-haze, haze and post-haze periods

The concentrations of trace elements and water soluble inorganicions during pre-haze, haze and post-haze episodes are presented inFig. 4. The results show that the major elements during pre-haze, hazeand post-haze episodes were K, Al, Ca, Mg and Fe. They accounted forabout 93%, 91% and 92% of the total metals portion collected duringpre-haze, haze and post-haze, respectively. Other trace elements ana-lyzed in this study were As, Ni, Co, Cr, Pb and Cd which are categorizedas carcinogenic agents (Class 1, 2A and 2B carcinogens) as well as othertrace elements, such as Cu,Mn, Ba, Zn, Rb, Sr and V. The total concentra-tion of elements analyzedwere 971±387ngm−3, 1499±575 ngm−3

and 778 ± 328 ng m−3 during pre-haze, haze and post-haze, respec-tively. A comparison of the single component concentrations of the ele-ments showed that most of them experienced a slight increase duringhaze compared with pre-haze and post-haze, except for K whichshowed a significant increase. Inversely, metals such as Zn, Cu, Ba andSr showed a lower concentration during haze compared to non-haze.

The concentration of K during haze event (604 ± 301 ng m−3, 238–1285 ng m−3) increased 3 times compared to pre-haze (208 ±118 ng m−3, 13.8–579 ng m−3) and post-haze event (195 ±73.7 ng m−3, 95.3–351 ng m−3). K is in fact an inorganic tracer and agood indicator of biomass burning (Pachon et al., 2013). The concentra-tion of K during pre-haze and post-haze episodesmay be contributed toby other sources, such as mineral dust (Amato et al., 2009) or road dust(Apeagyei et al., 2011) since this element has been associated withother sources and the sampling site is in close proximity to busy roads.

The one-way ANOVA test results suggested that there was a signifi-cant difference (p b 0.05) between the concentrations of water-solubleinorganic ions (SO4

2−, NO3− and NH4

+) measured during the pre-haze,haze and post-haze events. A higher concentration of SO4

2−, NO3− and

NH4+, which are classified as secondary inorganic aerosols (SIA), was

shown during the haze episode compared to the pre-haze and post-haze episodes. SIA contributed about 12%, 43% and 16% to the overallPM2.5 mass during pre-haze, haze and post-haze, respectively. The con-centration of SO4

2− and NO3− during the haze period increased by a fac-

tor of 10 compared to the concentration recorded during the pre-hazeand post-haze episodes. Meanwhile, NH4

+ recorded during the haze ep-isode increased 17.5–35 times compared to the pre-haze and post-hazeepisodes. The concentration of the biomass burning tracer K+, was also4.5–15 times higher compared to pre-haze and post-haze episodes. An-other minor water-soluble ions (Cl−, Na+, Ca2+ and Mg2+) comprised2%, 2% and 4% of PM2.5 mass during the period of pre-haze, haze andpost-haze, respectively.

3.4. Source apportionment using PMF during haze and non-haze (pre-hazeand post-haze)

PMF analysis was performed separately for the four data sets; pre-haze, haze, post-haze and the whole sampling period. Overall, fivesources of PM2.5 were identified using US EPA PMF Version 5.0. Thechemical profiles and the percentage contributions of each source for

Fig. 4. Concentration of elemental and ionic species based on pre-haze, haze and post-haze period.

563N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

Fig. 5. PMF resolved source factor profiles for PM2.5 mass of (a) overall data (b) pre-haze (c) haze and (d) post-haze.

564 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

565N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

each haze cycle are shown in Fig. 5. The PM2.5 concentration was mea-sured by HVS and the concentrations predicted via PMF (during thewhole sampling period) were regressed. The results showed that theyhad a significant correlation with each other (r2 = 0.73, p b 0.05), as isdemonstrated in Supplementary 6. During the calculation stage of thesource determination, a total of 72 samples with a data matrix of72 × 27 (data points × variables) was applied using the PMF model.The optimum number of factors or sources was chosen through a seriesof validation processes as suggested by Brown et al. (2015) and Paateroet al. (2014) and applied to a recently published article by Khan et al.(2016b). Emphasis was also given to generate physically meaningfulsource profiles even though the output result showed mixed sources.A study by Amil et al. (2016) in Kuala Lumpur also observed a similar re-sult of mixed sources for PM2.5 using the PMF 5.0 model.

3.4.1. Pre-haze (main source: mixed SIA and road dust)During the pre-haze period from June to September 2015, the highest

source contribution originated from mixed SIA and road dust, making acontribution of 32.4% to the PM2.5 mass. This source was characterizedby a significant amount of Ca2+ (74.9% of Ca2+ mass), Mg2+ (49.5% ofMg2+ mass), NO3

− (44.1% of NO3− mass) and SO4

2− (40.3% of SO42−

mass). A sampling campaign of PM10 undertaken in Megapolis, Greecesuggested that Ca originated from road re-suspension (Manousakas etal., 2015). The abundance of Ca in their samples represented urban roaddust as a source, as also suggested by Pey et al. (2013). Furthermore, sec-ondary particulates of SO4

2− andNO3−were produced during clear days as

the result of oxidation of the SO2 and NO2 emitted by vehicles. The pro-duction of SO2 and NO2 is possible since the sampling site is located inan urban area where air quality is influenced by a high quantity of localtraffic. The second highest source contribution was found to be trafficemissions and biomass burning (27.5%), represented by a high V (57.8%of V mass), Pb (52.6% of the Pb mass), Rb (47.2% of the Rb mass), K(47.0% of the K mass), Sr (41.8% of the Sr mass) and Cu (43.3% of the Cumass). Both Khan et al. (2016a) and Amil et al. (2016) suggested thatthe high level of V and Pb in Bangi and Petaling Jaya, Malaysia, are the re-sult of traffic emissions. Likewise, a high level of Cu indicated that trafficwas the source of fine particulate matter in Beijing (Yang et al., 2016).Pb and Cd were similarly identified in vehicle exhaust emissions in astudy conducted by Jin et al. (2015) while a high concentration of V indi-cated the contribution of fuel emissions (Yatkin and Bayram, 2008). Suchthat, the location of the sampling site near to main roads (Raja MudaAbdul Aziz, Chowkit, Pahang andTunRazak Roads) in Kuala Lumpur, con-tributed to the release of traffic-related sources. A large number of vehi-cles use these roads every day of the week, including weekends.According to Rahman et al. (2011), the main source of air pollutants inKuala Lumpur is vehicle emissions. Meanwhile K can be associated withbiomass burning (Pachon et al., 2013; Saarikoski et al., 2007). 25.8% ofthe total PM2.5 mass was contributed to by mineral dust, followed byaged sea salt (11.9%), and mixed road dust and industrial emissions(2.3%). Mineral dust has been identified as the source profile due to thehigh concentration of K (46.3% of K mass), Al (46.1% of Al mass) and Rb(27.0% of Rb mass). In a study by Amato et al. (2009), K, Al and Rb wereidentified as mineral/crustal dust. Aged sea salt was also characterizedas having a high concentration of concentration of Cl− (39.5% of Cl−

mass) and NO3− (29.5% of NO3

− mass). In a study by Khan et al. (2016b)andAmato et al. (2009), Cl− andNO3

−were identified as the predominantsources of aged sea salt. It can therefore be concluded that during the pre-haze period the sources predominantly originated from SIA and road dustemissions. SIA was also contributed to by traffic via the release of NOxprecursors from vehicles. This in turn eventually generated secondaryparticles which contributed to particulate matter (PM).

3.4.2. Haze (main source: SIA and biomass burning)Through PMF analysis, themost significant sources identified during

the haze period were secondary inorganic aerosol (SIA) and biomassburning (32.8% of PM2.5 mass). These were characterized by a high

loading of NO3− (61.3% of the NO3

− mass), Cl− (60.1% of the Cl− mass),SO4

2− (53.8% of the SO42− mass), NH4

+ (58.6% of the NH4+ mass) and

K+ (41.1% of the K+ mass). Khan et al. (2016b) reported that SO42−

and NH4+ originated from SIA. Similar results were obtained by Shen

et al. (2009) where, during an air pollution episode, K+, NH4+, SO4

2−

and NO3− were enriched in fine particles. Likewise, Xu et al. (2012)

also found a high concentration of NH4+, SO4

2− and NO3− in PM2.5. In

this study, the heavy smoke (haze episode) that occurred between Sep-tember and October 2015 released a significant amount of K+ into theatmosphere, which is the chemical tracer for biomass and peat burning(Currie et al., 1994). A study undertaken by Hays et al. (2005) also ob-served a significant amount of K+ during a biomass burning periodthus demonstrating its role as a chemical tracer of biomass burning.Other ions released during biomass burning were NH4

+, SO42−

and NO3−. Supplementary 7 shows the correlation plot between the

concentrations of K+ and nss-SO42− measured during the pre-haze,

haze and post-haze periods. During haze, K+ was strongly correlated(r2 = 0.94, p b 0.01) with nss-SO4

2− indicating that both ions sharedthe same source whichwas the biomass burning. Li et al. (2003) report-ed a significant correlation between K+ and SO4

2− and NH4+ which can

be released during biomass burning. However, this study found no cor-relation between NH4

+ and K+ during the haze period. For the pre-hazeand post-haze episodes, a weak relationship was shown between thesetwo ions. Overall, 25.4% of the total PM2.5 released during hazewas con-tributed to by industrial emissionswith a significant release of Ba (67.3%of the Ba mass), K (45.8% of the K mass), Sr (40.8% of the Sr mass), Cd(34.3% of the Cd mass) and V (29.8% of the V mass). Cd is the sourceof industrial emission (Murillo-Tovar et al., 2015) while Pandolfi et al.(2011) claimed that V is released from industrial emission. Other factorsidentifiedweremineral dust (22.3% of the total PM2.5 during haze), traf-fic emissions (17.8% of the total PM2.5 during haze) and sea salt (1.8% ofthe total PM2.5 during haze). Mineral dust was characterized by a highinfluence of Mg (39.8% of Mg mass), Ca (32% of Ca mass) and Ca2+

(52% of Ca2+mass). Khan et al. (2016b) suggested thatMgwas releasedfrommineral dust in their sampling location. Traffic emissions were in-dicated by a high concentration of V (39.4% of V mass) and Pb (39.0% ofPb mass) (Lestari and Mauliadi, 2009). Undoubtedly, the most impor-tant source of PM2.5 during the haze period was from biomass burning,which originated in Indonesia. This finding is further confirmed by thebackward trajectories obtained using the NAME model. The release ofprecursor gases, such as SO2, NO2 and NH3, from biomass burning gen-erated the formation of SIA in the atmosphere. In addition, a high levelof the biomass tracer K+demonstrated that the dominant source duringthe haze period was biomass burning.

3.4.3. Post-haze (main source: mineral dust)The results from PMF analysis revealed that the main source during

post-haze wasmineral dust (31.1% of the PM2.5mass). This was indicat-ed by the significant amount of Al (52.1% of the Al mass), Ca (35.4% ofthe Camass) andMg (30.8% of theMgmass). FromNovember 2015 on-wards, therewas a constructionwork takingplaces next to the samplingsite. Therefore, it is anticipated that construction activities had a high in-fluence on samples taken during that period of time. Studies haveshown that construction materials may contain an excess of the ele-ments of Al, Ca and Mg (Rodríguez et al., 2007; Song et al., 2006). Themineral dust source has been identified with each of the events at thissite. The resuspension of mineral or soil dust from the renovation andconstruction activitiesmight be a possible reason. Themixture ofminer-al dust in other source profile may be due to the domination of dust inthe tropical environment (Ferreira-Baptista and De Miguel, 2005). Thesecond highest source for the post-haze period was sea salt coupledwith industrial emissions (29.6% of the PM2.5mass), these are attributedto a high release of Cl− (62.8% of the Cl−mass), NO3

− (49.4% of the NO3−

mass), Ni (49.2% of the Ni mass), Cr (48.5% of the Cr mass) and Na+

(32.7% of the Na+mass). In a study undertaken in theWesternMediter-ranean by Bove et al. (2016), Na+ and Cl− were assigned to represent

566 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

the sea salt component in the study area. Na+ and Cl− were also classi-fied as a marine aerosol in a study conducted by an urban roadside inChennai City, India (Srimuruganandamand Shiva Nagendra, 2012).Ma-rine aerosols are produced by breaking waves bringing air bubbles tothe sea surface. For this study, the sea salt is thought to have originatedfrom the Malacca Straits which are located about 40 km from the sam-pling site. The Cl−/Na+ ratios during the pre-haze, haze and post-hazeperiods were 0.12, 0.23 and 0.09, respectively. These ratios suggestthat a high chlorine deficiency occurred during those three periodswhereby a high relative humidity was observed. Similar observationswere discussed by Amil et al. (2016) and Khan et al. (2010) whereby ahigh relative humidity was found to contribute to chlorine deficiency.Meanwhile, Sowlat et al. (2012) reported that the Cr element was re-leased by metal manufacturing processes and smelting plants neartheir sampling site. The contribution of the industrial emissions to thesampling site are most likely to have originated from the industrial cen-tre of Petaling Jaya, Selangor as this is located only 20 km from the sam-pling site. Many factories are based there including electronics,manufacturing, rawmaterial processing plants, high-technology indus-tries, alongwith customer service-based industries. Other sources iden-tified during the post-haze period were traffic emissions, SIA coupledwith traffic and biomass burning coupled with mineral dust. Eachsource contributed 16.4%, 15% and 7.9% to the PM2.5 mass, respectively.Traffic emission was indicated by Ba (60% of Ba mass) and Sr (42% of Srmass). SIA with traffic factor was dominated by Pb (51% of Pb mass),SO4

2− (43% of SO42− mass) and Cu (40.8% of Cu mass). Biomass burning

coupled with mineral dust is contributed by high influence of Ca (48.0%of Ca mass), K+ (42.0% of K+) and NO3

− (44.4% of NO3− mass).

3.4.4. Whole sampling period (main source: SIA and biomass burning)The first factor contributing to the site was sea salt, which accounted

for 18.6% of the total PM2.5 concentration collected during the completesampling campaign. This factor was assigned to the sea salt source dueto the presence of Ca2+ (55.9% of the total Ca2+ mass), Na+ (49.1% ofthe total Na+ mass), Cl− (41% of the total Cl− mass) and Mg2+ (40.0%of the total Mg2+). Pearson correlation analysis revealed that the corre-lation betweenNa+ and Cl− indicates the possible presence of sea spraycontribution to PM2.5 (r=0.69, p b 0.05). The second factor was identi-fied asmineral dust as it showed ahigh concentration of Ca (55.1% of theCa mass), Zn (49.0% of the Zn mass) and Fe (30.8% of the Fe mass) andaccounted for 4.7% of the overall PM2.5mass. The third and largest factorfor the site was mixed biomass burning and SIA. This factor accountedfor 38.5% of the total PM2.5 mass with a high contribution of K+ (79.3%of the K+ mass), NH4

+ (78.0% of the NH4+ mass), SO4

2− (73.2% of theSO4

2−) and NO3− (50.2% of the NO3

−). Factor 4 was classified as trafficemissions due to the high explained variation of Ba (69.0% of the totalBa mass), Sr (68.3% of the total Sr mass), Pb (61.0% of the total Pbmass), Cd (49.9% of the total Cd mass), V (48.7% of the total V mass),Cu (40.5% of the total Cumass) and Rb (38.6% of the total Rbmass). Traf-fic emissions were found to originate from both vehicles and road dust.They accounted for 22.4% of the total PM2.5 mass and were the secondlargest contributor to PM2.5 in the study area.With a 15.9% contribution,factor 5was recognised as industrial emissions due to a high abundanceof Cr, Mn and As (57.4%, 55.1% and 52.6% of their elemental mass,respectively).

3.5. Average daily doses (ADD) and non-carcinogenic health riskassessment

The average daily dose (ADD) for different age groups is shown inSupplementary 8. The ADD is used to calculate the non-carcinogenichealth risks via inhalation of Cr, Mn, Ni, Cd and As in PM2.5. The resultsof the hazard quotient (HQ) and hazard index (HI) are shown inTable 1. The health risk assessment for non-carcinogenic elements wasfurther discussed in three categories: (1) haze cycle (haze and non-haze) (2) specific age groups and (3) toxic elements. Based on the

haze cycle, the estimated HI during haze period was slightly highercompared to non-haze period. According to Betha et al. (2014), the par-ticles deposition in the respiratory system becomingworse during hazedays compared to non-haze days, hence extended period of exposure tofine particles can cause deleterious health effects to the targeted popu-lations. Usually, haze occurred for a few days per year but, as was thecase in this study, it lasted for two months. Based on the specific agegroups, the pattern of health risks posed by the population was asfollows: infant N toddler N adolescent N children N adult. The infantgroup showed ahigher health risk compared to the other age groups be-cause they were categorized as a sensitive population as their immunesystem was not as fully developed as the adults group. In addition, in-fants tend to become ill more easily than older people. According toYang et al. (2013), the likelihood of children experiencing non-carcino-genic health effects was also higher when compared to adults. In refer-ence to the toxic elements, the inhalation of Cr in PM2.5 showed that HQwas slightly above the acceptable limit during both the pre-haze (HQ=1.10× 100) and hazeperiods (HQ=1.02× 100). The trend forHQposedby toxic elements was as follows: Cr NMn N Cd N As N Ni. The source ap-portionment method identified that Cr, Mn, As and Ni originated fromindustrial emissions meanwhile Cd originated from traffic emissions.So we can manipulate and control both the industrial and traffic emis-sions as the alternative way to reduce the health risks posed by popula-tions as the result of inhalation to toxic elements in PM2.5. It is veryimportant to effectively control air quality in order to reduce healthrisk (Kim et al., 2015).

3.6. Lifetime average daily dose (LADD) and carcinogenic health risks

The lifetime average daily dose (LADD) was used to estimate excesslifetime cancer risks (ELCR) for the carcinogenic elements: Pb, Cd, Cr, Ni,As and Co via PM2.5 inhalation. The LADD values are shown in Supple-mentary 9. The excess lifetime cancer risks (ELCR) exhibited by the dif-ferent elements in PM2.5 are shown in Table 2. Health risk assessment ofthe carcinogenic elements can be divided into (1) haze cycles (haze andnon-haze) (2) specific age groups and (3) toxic elements. Based on thehaze cycle, the cumulative ELCR posed by carcinogenic elements duringhaze was higher compared to non-haze. The risk of developing cancerdepends on many factors, including an individual's genetics, the naturein which an individual is exposed to a carcinogen, alongwith the lengthand intensity of exposure (USEPA, 2016). Our research found that thecarcinogenic risk increases during haze compared to non-haze and themost affected age group was the adult group, for which the ELCR in-creased from 1.92× 10−5 to 2.27 × 10−5. Although haze periods alwaysoccur over a short period of time, the haze period in 2015 lasted for twomonths. So, the length and intensity of the haze was intensified andeventually increase the health risk estimation. The repeated occurrenceof the haze for almost every year poses a long term effects on publichealth. The highest ELCR estimation during haze was posed by theadult group (2.27 × 10−5) where 2–3 individuals in 100,000 are likelyto develop cancer in their lifetime due to exposure to haze smoke aerosol.The lowest ELCR estimation was posed by the infant group during thenon-haze period (1.20 × 10−6) indicating that 1–2 individuals in1,000,000 are likely to develop cancer in their lifetime as a result of expo-sure to urban PM2.5 aerosol. A study by Betha et al. (2014) also found thatthe ELCR estimated in haze aerosol was higher compared to urban aero-sol. In terms of specific age groups, the estimated health risk followed theorder of: adult N toddler N adolescent N children N infant. Adult(18 N 70 years) showed the higher health risk compared to the youngerage group because theywere, in comparison, exposed to the carcinogensfor a longer period of time. Such that, the longer the exposure time, thehigher the concentration of carcinogenic elements accumulated in thebody. Hu et al. (2012) reported that the carcinogenic risk posed by theadult group was greater when compared to the children's group as a re-sult of the inhalation of carcinogenic elements. In term of carcinogenic el-ements, the pattern of Cr N As N Co NNi N CdN Pbwas observedwhere the

Table 1Hazard quotient (HQ) and hazard index (HI) of elemental species in PM2.5.

Elements RfC (mg m−3) HQ

Infant0–b1 year

Toddler1–b6 years

Children6–b12 years

Adolescent12–b18 years

Adult18–b70 years

Pre-hazeCr 8.00 × 10−6 1.10 × 100 8.59 × 10−1 5.50 × 10−1 5.91 × 10−1 3.40 × 10−1

Mn 5.00 × 10−5 3.60 × 10−2 2.80 × 10−2 1.80 × 10−2 1.93 × 10−2 1.11 × 10−2

Ni 2.00 × 10−4 5.70 × 10−3 4.46 × 10−3 2.86 × 10−3 3.07 × 10−3 1.77 × 10−3

Cd 1.00 × 10−5 9.67 × 10−3 7.52 × 10−3 4.82 × 10−3 5.18 × 10−3 2.98 × 10−3

As 5.00 × 10−5 6.57 × 10−3 5.11 × 10−3 3.28 × 10−3 3.52 × 10−3 2.03 × 10−3

∑HI 1.16 0.90 0.58 0.62 0.36

HazeCr 8.00 × 10−6 1.02 × 100 7.93 × 10−1 5.08 × 10−1 5.46 × 10−1 3.14 × 10−1

Mn 5.00 × 10−5 2.94 × 10−2 2.29 × 10−2 1.47 × 10−2 1.58 × 10−2 9.07 × 10−3

Ni 2.00 × 10−4 3.51 × 10−3 2.73 × 10−3 1.75 × 10−3 1.88 × 10−3 1.08 × 10−3

Cd 1.00 × 10−5 2.79 × 10−3 2.17 × 10−3 1.39 × 10−3 1.49 × 10−3 8.60 × 10−4

As 5.00 × 10−5 2.81 × 10−3 2.19 × 10−3 1.40 × 10−3 1.51 × 10−3 8.68 × 10−4

∑HI 1.06 0.82 0.53 0.57 0.33

Post-hazeCr 8.00 × 10−6 5.87 × 10−1 4.57 × 10−1 2.93 × 10−1 3.15 × 10−1 1.81 × 10−1

Mn 5.00 × 10−5 4.53 × 10−2 3.52 × 10−2 2.26 × 10−2 2.42 × 10−2 1.40 × 10−2

Ni 2.00 × 10−4 3.67 × 10−3 2.86 × 10−3 1.83 × 10−3 1.97 × 10−3 1.13 × 10−3

Cd 1.00 × 10−5 8.38 × 10−3 6.52 × 10−3 4.18 × 10−3 4.49 × 10−3 2.58 × 10−3

As 5.00 × 10−5 5.29 × 10−3 4.11 × 10−3 2.63 × 10−3 2.83 × 10−3 1.63 × 10−3

∑HI 0.65 0.51 0.32 0.35 0.20

∑HI Haze 1.06 0.82 0.53 0.57 0.33∑HI Non-hazea 0.91 0.71 0.45 0.49 0.28

Note: RfC is the reference concentration; RfC for Cr andMnwere referred from Integrated Risk Information System (IRIS) (http://www.epa.gov/iris); RfC for Ni from Agency for Toxic Sub-stances and Disease Registry (ATSDR) while RfC for Cd and As from California Environmental Protection Agency.

a Indicates the average of non-carcinogenic health risk during pre-haze and post-haze episode.

567N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

ELCR of Cr was significant but acceptable (values were slightly above thethreshold value of 1 × 10−6). Besides that, Cr toxicity and carcinogenicity

Table 2Excess lifetime cancer risks (ELCR) of carcinogenic elements in PM2.5.

Elements IUR Excess lifetime cancer risk (μg m−3)−1

(μg m−3)−1 Infant0–b1 year

Toddler1–b6 years

Pre-hazePb 1.20 × 10−5 7.53 × 10−10 2.93 × 10−

Cd 1.80 × 10−3 2.49 × 10−9 9.67 × 10−

Cr 1.20 × 10−2 1.51 × 10−6 5.89 × 10−

Ni 2.40 × 10−4 3.93 × 10−9 1.53 × 10−

As 4.30 × 10−3 2.02 × 10−8 7.85 × 10−

Co 9.00 × 10−3 1.41 × 10−8 5.48 × 10−

∑ 1.56 × 10−6 6.05 × 10−

HazePb 1.20 × 10−5 1.50 × 10−10 5.83 × 10−

Cd 1.80 × 10−3 7.17 × 10−10 2.79 × 10−

Cr 1.20 × 10−2 1.40 × 10−6 5.44 × 10−

Ni 2.40 × 10−4 2.41 × 10−9 9.37 × 10−

As 4.30 × 10−3 8.65 × 10−9 3.36 × 10−

Co 9.00 × 10−3 5.38 × 10−9 2.09 × 10−

∑ 1.42 × 10−6 5.50 × 106

Post-hazePb 1.20 × 10−5 4.91 × 10−10 1.91 × 10−

Cd 1.80 × 10−3 2.15 × 10−9 8.38 × 10−

Cr 1.20 × 10−2 8.06 × 10−7 3.13 × 10−

Ni 2.40 × 10−4 2.52 × 10−9 9.80 × 10−

As 4.30 × 10−3 1.62 × 10−8 6.31 × 10−

Co 9.00 × 10−3 8.29 × 10−9 3.22 × 10−

∑ 8.35 × 10−7 3.25 × 10−

∑ ELCR Haze 1.42 × 10−6 5.50 × 10−

∑ ELCR Non-hazea 1.20 × 10−6 4.65 × 10−

Note: IUR is the Inhalation Unit Risk; IUR for Cd, Cr, Ni and As were obtained from IRIS (http://(http://www.epa.gov/reg3hwmd/risk/human/rbconcentration_table/Generic_Tables/docs/mas

a Indicates the average of carcinogenic health risk during pre-haze and post-haze episode.

was directly related to on Cr (VI). In this study, only the total of Cr as op-posed to Cr (VI) was determined and the carcinogenic risk of Cr was

Children6–b12 years

Adolescent12–b18 years

Adult18–b70 years

9 2.25 × 10−9 2.42 × 10−9 1.21 × 10−8

9 7.44 × 10−9 8.00 × 10−9 4.00 × 10−8

6 4.53 × 10−6 4.86 × 10−6 2.43 × 10−5

8 1.18 × 10−8 1.26 × 10−8 6.31 × 10−8

8 6.04 × 10−8 6.49 × 10−8 3.24 × 10−7

8 4.22 × 10−8 4.53 × 10−8 2.26 × 10−7

6 4.65 × 10−6 5.00 × 10−6 2.49 × 10−5

10 4.49 × 10−10 4.82 × 10−10 2.41 × 10−9

9 2.15 × 10−9 2.31 × 10−9 1.15 × 10−8

6 4.18 × 10−6 4.49 × 10−6 2.24 × 10−5

9 7.21 × 10−9 7.74 × 10−9 3.86 × 10−8

8 2.59 × 10−8 2.78 × 10−8 1.39 × 10−7

8 1.61 × 10−8 1.73 × 10−8 8.63 × 10−8

4.23 × 10−6 4.55 × 10−6 2.27 × 10−5

9 1.47 × 10−9 1.58 × 10−9 7.87 × 10−9

9 6.45 × 10−9 6.92 × 10−9 3.46 × 10−8

6 2.41 × 10−6 2.59 × 10−6 1.29 × 10−5

9 7.54 × 10−9 8.10 × 10−9 4.04 × 10−8

8 4.86 × 10−8 5.22 × 10−8 2.60 × 10−7

8 2.48 × 10−8 2.66 × 10−8 1.33 × 10−7

6 2.50 × 10−6 2.68 × 10−6 1.34 × 10−5

6 4.23 × 10−6 4.55 × 10−6 2.27 × 10−5

6 3.58 × 10−6 3.84 × 10−6 1.92 × 10−5

www.epa.gov/iris) and IUR for Pb and Co were cited from RSL Tables for US EPA Region 9.ter_sl_table_run_JUNE2015_rev.pdf).

568 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

deemed acceptable. Overall, the integrated risks of carcinogenic elementswere within acceptable level (1 × 10−6–1 × 10−4).

4. Conclusions

The study showed that the PM2.5 mass collected during pre-haze,haze and post-haze events in Kuala Lumpur were significantly different(p b 0.005). The average concentration of PM2.5 was recorded duringhaze (72.3 ± 38.0 μg m−3), followed by pre-haze (24.5 ±12.0 μg m−3) and post-haze (14.3 ± 3.58 μg m−3). The highest PM2.5

concentrations during haze were 5 times higher than WHO guidelines,3.9 times higher than the US EPA standards and 1.8 times higher thanthe Malaysian Ambient Air Quality Standards 2015 (IT-1). The NAMEmodel and GFAS emissions clearly showed that during haze, abundantfire activity was detected in Sumatra, Indonesia following which, theair masses with PM2.5 particles were transported to the sampling sitein Kuala Lumpur city centre. The concentration of PM2.5 recorded duringthe haze episode had a good correlation with the Malaysian Air Pollut-ants Index(API) (r = 0.466; p b 0.05) and significantly reduce the visi-bility (r = −0.631; p = 0.005).

Themajor elements in PM2.5 compositions during pre-haze, haze andpost-haze episodeswere K, Al, Ca,Mg and Fe. They accounted for approx-imately 93%, 91% and 92% of the total elements portion collected duringpre-haze, haze and post-haze, respectively. The concentration of trace el-ements during pre-haze, haze and post-haze comprised 4%, 2% and 5% ofthe PM2.5mass, respectively. The SIA component, namely SO4

2−, NO3− and

NH4+, were dominated by the composition of PM2.5. These three ions

were found to contribute to 43% of the overall PM2.5 mass during hazecompared to pre-haze and post-haze, where they only contributed 12%and 16%, respectively. The source apportionment method (PMF) usedon the overall dataset indicates that the dominant sources of PM2.5

were SIA and biomass burning (38.5%), and traffic emission (22.4%).Hence, during non-haze days, local emissions, particularly from traffic,greatly influence the PM2.5 source contribution to this sampling area.

The carcinogenic health risk assessment reveals that theintegrated risk of carcinogenic elements for all of the age groupsduring haze and non-haze exposure was within the acceptablelimit (1 × 10−6–1 × 10−4). However, the non-carcinogenic healthrisk assessment showed that the infant group faced more significanthealth risk than the other age groups during haze (HI = 1.06). Thefinding of this study revealed that the carcinogenic and non-carcinogenichealth risks posed by populations were slightly higher during hazecompared to non-haze. The most affected age group was the adultgroup for which the ELCR increased from 1.92 × 10−5 (non-haze) to2.27 × 10−5 (haze). The lowest ELCR estimation was posed by the infantgroup during non-haze (1.20 × 10−6) indicating that 1–2 individuals in1,000,000 are likely to develop cancer in their lifetime due to exposureof urban PM2.5 aerosols.

Through this work, source apportionment and the associated carci-nogenic and non-carcinogenic health risks relating to inorganic PM2.5

compositions in an urban area of Kuala Lumpur during haze and non-haze periods were successfully investigated. However, as this studywas limited to inorganic compositions and did not cover the organiccomponent of PM2.5 compositions, further and more comprehensivestudies are necessary in order to focus not only on the inorganic butalso organic constituents of PM2.5. A large number of samples are need-ed for concrete assessment on source apportionment analysis. The in-clusion of submicron-sized particles will also be important in futureresearch due to increasing concerns over their effects on both humanhealth and the environment.When it comes to health effects, particular-ly during a haze period, apart from undertaking health risk calculations,the most effective way to identify the relationship between haze andhealth effects would be to conduct case-crossover studies to evaluatemortality risks along with toxicity studies to better understand thetoxicological impact of PM2.5 released during haze periods.

Acknowledgements

This study was funded by Universiti Kebangsaan Malaysia throughUniversity Research Grants DIP-2016-015 and the Ministry of HigherEducation through Fundamental Research Grant (FRGS/1/2015/WAB03/UKM/01/1). The authors gratefully acknowledge the Depart-ment of Environment (DOE) for the provision of the meteorological da-ta. The authors also acknowledge the School of Environmental andNatural Resources UKM,Mr. Ak. Jalaludin Pg. Awang Besar for the ICPMSfacilities, the technical staff of the Department of Chemical and ProcessEngineering UKM and also Mr. Mohd Razif Maafol and Mr. SalehuddinArifen for their assistancewith water-soluble ionic analysis.We also ac-knowledge use of the NAME atmospheric dispersion model and associ-ated NWP meteorological data sets made available to us by the MetOffice. We acknowledge the significant storage resources and analysisfacilities made available to us on JASMIN by STFC CEDA along with thecorresponding support teams. The authors also express their gratitudeto K. Alexander for proofreading this manuscript.

References

Abas, R.M., Rahman, N.A., Omar, N.Y.M.J., Maah, M.J., Abu Samah, A., Oros, D.R., Otto, A.,Simoneit, B.R.T., 2004. Organic composition of aerosol particulate matter during ahaze episode in Kuala Lumpur, Malaysia. Atmos. Environ. 38, 4223–4241.

Afroz, R., Hassan, M.N.A., Ibrahim, N.A., 2003. Review of air pollution and health impactsin Malaysia. Environ. Res. 92, 71–77.

Aiken, S.R., 2004. Runaway fires, smoke-haze pollution, and unnatural disasters in Indo-nesia. Geogr. Rev. 94, 55–79.

Amato, F., Pandolfi, M., Escrig, A., Querol, X., Alastuey, A., Pey, J., Perez, N., Hopke, P., 2009.Quantifying road dust resuspension in urban environment by multilinear engine: acomparison with PMF2. Atmos. Environ. 43, 2770–2780.

Amil, N., Latif, M.T., Khan, M.F., Mohamad, M., 2016. Seasonal variability of PM2.5 compo-sition and sources in the Klang Valley urban-industrial environment. Atmos. Chem.Phys. 16, 5357–5381.

Apeagyei, E., Bank, M.S., Spengler, J.D., 2011. Distribution of heavy metals in roaddust along an urban-rural gradient in Massachusetts. Atmos. Environ. 45,2310–2323.

Artaxo, P., Martins, J.V., Yamasoe, M.A., Procópio, A.S., Pauliquevis, T.M., Andreae, M.O.,Guyon, P., Gatti, L.V., Leal, A.M.C., 2002. Physical and chemical properties of aerosolsin the wet and dry seasons in Rondônia, Amazonia. J. Geophys. Res.-Atmos. 107 (n/a-n/a).

Awang, M., Jaafar, A.B., Abdullah, A.M., Ismail, M., Hassan, M.N., Abdullah, R., Johan, S.,Noor, H., 2000. Air quality in Malaysia: impacts, management issues and future chal-lenges. Respirology 5, 183–196.

Balasubramanian, R., Qian, W.B., Decesari, S., Facchini, M., Fuzzi, S., 2003. Comprehensivecharacterization of PM2.5 aerosols in Singapore. J. Geophys. Res.-Atmos. 108 (n/a-n/a).

Ballinger, M.Y., Larson, T.V., 2014. Source apportionment of stack emissions from researchand development facilities using positive matrix factorization. Atmos. Environ. 98,59–65.

Beringer, J., Hutley, L., Tapper, N., Coutts, A., Kerley, A., O'grady, A., 2003. Fire impacts onsurface heat, moisture and carbon fluxes from a tropical savanna in northern Austra-lia. Int. J. Wildland Fire 12, 333–340.

Betha, R., Pradani, M., Lestari, P., Joshi, U.M., Reid, J.S., Balasubramanian, R., 2013. Chemicalspeciation of trace metals emitted from Indonesian peat fires for health risk assess-ment. Atmos. Res. 122, 571–578.

Betha, R., Behera, S.N., Balasubramanian, R., 2014. 2013 Southeast Asian smoke haze: frac-tionation of particulate-bound elements and associated health risk. Environ. Sci.Technol. 48, 4327–4335.

Bove, M.C., Brotto, P., Calzolai, G., Cassola, F., Cavalli, F., Fermo, P., Hjorth, J., Massabò, D.,Nava, S., Piazzalunga, A., Schembari, C., Prati, P., 2016. PM10 source apportionmentapplying PMF and chemical tracer analysis to ship-borne measurements in theWest-ern Mediterranean. Atmos. Environ. 125, 140–151.

Brauer, M., Hisham-Hashim, J., 1998. Fires in Indonesia: Crisis and reaction. Environ. Sci.Technol. 32, 404–407.

Brown, S.G., Eberly, S., Paatero, P., Norris, G.A., 2015. Methods for estimating uncertaintyin PMF solutions: examples with ambient air and water quality data and guidance onreporting PMF results. Sci. Total Environ. 518, 626–635.

Charron, A., Harrison, R.M., 2005. Fine (PM2.5) and coarse (PM2.5-10) particulate matteron a heavily trafficked London highway sources and processes. Environ. Sci. Technol.39, 7768–7778.

Currie, L.A., Klouda, G.A., Klinedinst, D.B., Sheffield, A.E., Jull, A.J.T., Donahue, D.J.,Connolly, M.V., 1994. Fossil- and bio-mass combustion: C-14 for sourceidentification, chemical tracer development, and model validation. Nucl. Inst.Methods Phys. Res. B 92, 404–409.

DOE, 2015. Press Statement of Ministry of Natural Resources & Environment, 15 Septem-ber 2015 Air Quality Status. Department of Environment.

DOSM, 2016. Federal Territory of Kuala Lumpur. Department of Statistics Malaysia.Dotse, S.-Q., Dagar, L., Petra, M.I., De Silva, L.C., 2016. Influence of Southeast Asian Haze

episodes on high PM 10 concentrations across Brunei Darussalam. Environ. Pollut.219, 337–352.

569N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

Ee-Ling, O., Mustaffa, N.I.H., Amil, N., Khan, M.F., Latif, M.T., 2015. Source contribution ofPM2. 5 at different locations on the Malaysian Peninsula. Bull. Environ. Contam.Toxicol. 94, 537–542.

Fan, S.M., Horowitz, L.W., Levy, H., Moxim, W.J., 2004. Impact of air pollution on wet de-position of mineral dust aerosols. Geophys. Res. Lett. 31 (n/a-n/a).

Ferreira-Baptista, L., De Miguel, E., 2005. Geochemistry and risk assessment of street dustin Luanda, Angola: a tropical urban environment. Atmos. Environ. 39, 4501–4512.

Fujii, Y., Tohno, S., Amil, N., Latif, M.T., Oda, M., Matsumoto, J., Mizohata, A., 2015. Annualvariations of carbonaceous PM2.5 in Malaysia: influence by Indonesian peatland fires.Atmos. Chem. Phys. 15, 13319–13329.

Hasheminassab, S., Daher, N., Ostro, B.D., Sioutas, C., 2014. Long-term source apportion-ment of ambient fine particulate matter (PM2.5) in the Los Angeles Basin: a focuson emissions reduction from vehicular sources. Environ. Pollut. 193, 54–64.

Hays, M.D., Fine, P.M., Geron, C.D., Kleeman, M.J., Gullett, B.K., 2005. Open burning of ag-ricultural biomass: physical and chemical properties of particle-phase emissions.Atmos. Environ. 39, 6747–6764.

Heil, A., Goldammer, J., 2001. Smoke-haze pollution: a review of the 1997 episode inSoutheast Asia. Reg. Environ. Chang. 2, 24–37.

Hertwig, D., Burgin, L., Gan, C., Hort, M., Jones, A., Shaw, F., Witham, C., Zhang, K., 2015.Development and demonstration of a Lagrangian dispersion modeling system forreal-time prediction of smoke haze pollution from biomass burning in SoutheastAsia. J. Geophys. Res.-Atmos. 120, 12605–12630.

Hu, X., Zhang, Y., Ding, Z., Wang, T., Lian, H., Sun, Y., Wu, J., 2012. Bioaccessibility andhealth risk of arsenic and heavy metals (Cd, Co, Cr, Cu, Ni, Pb, Zn and Mn) in TSPand PM2.5 in Nanjing, China. Atmos. Environ. 57, 146–152.

Hyer, E.J., Reid, J.S., Prins, E.M., Hoffman, J.P., Schmidt, C.C., Miettinen, J.I., Giglio, L., 2013.Patterns of fire activity over Indonesia andMalaysia from polar and geostationary sat-ellite observations. Atmos. Res. 122, 504–519.

Jamhari, A.A., Sahani, M., Latif, M.T., Chan, K.M., Tan, H.S., Khan, M.F., Mohd Tahir, N., 2014.Concentration and source identification of polycyclic aromatic hydrocarbons (PAHs)in PM10 of urban, industrial and semi-urban areas in Malaysia. Atmos. Environ. 86,16–27.

Jin, X., Xiao, C., Li, J., Huang, D., Yuan, G., Yao, Y., Wang, X., Hua, L., Zhang, G., Cao, L., Wang,P., Ni, B., 2015. Source apportionment of PM2.5 in Beijing using positivematrix factor-ization. J. Radioanal. Nucl. Chem. 307, 2147–2154.

Jones, D.S., 2006. ASEAN and transboundary haze pollution in Southeast Asia. Asia. EuropeJournal 4, 431–446.

Jones, A., T., D., Hort, M., Devenish, B., 2007. The U.K. Met Office’s Next-Generation Atmo-spheric Dispersion Model, NAME III. Air Pollution Modeling and Its Application XVII.Springer, US, New York, USA.

Juneng, L., Tangang, F.T., 2005. Evolution of ENSO-related rainfall anomalies in SoutheastAsia region and its relationship with atmosphere–ocean variations in Indo-Pacificsector. Clim. Dyn. 25, 337–350.

Juneng, L., Latif, M.T., Tangang, F., 2011. Factors influencing the variations of PM10aerosol dust in Klang Valley, Malaysia during the summer. Atmos. Environ. 45,4370–4378.

Kaiser, J.W., Heil, A., Andreae, M.O., Benedetti, A., Chubarova, N., Jones, L., Morcrette, J.J.,Razinger, M., Schultz, M.G., Suttie, M., van der Werf, G.R., 2012. Biomass burningemissions estimated with a global fire assimilation system based on observed fire ra-diative power. Biogeosciences 9, 527–554.

Kaufmann, B., Christen, P., 2002. Recent extraction techniques for natural products: mi-crowave-assisted extraction and pressurised solvent extraction. Phytochem. Anal.13, 105–113.

Khan, M.F., Shirasuna, Y., Hirano, K., Masunaga, S., 2010. Characterization of PM2.5,PM2.5–10 and PM10 in ambient air, Yokohama, Japan. Atmos. Res. 96, 159–172.

Khan, M.F., Latif, M.T., Lim, C.H., Amil, N., Jaafar, S.A., Dominick, D., Mohd Nadzir, M.S.,Sahani, M., Tahir, N.M., 2015. Seasonal effect and source apportionment of polycyclicaromatic hydrocarbons in PM2.5. Atmos. Environ. 106, 178–190.

Khan, M.F., Latif, M.T., Saw, W.H., Amil, N., Nadzir, M.S.M., Sahani, M., Tahir, N.M., Chung,J.X., 2016a. Fine particulate matter in the tropical environment: monsoonaleffects, source apportionment, and health risk assessment. Atmos. Chem. Phys. 16,597–617.

Khan, M.F., Sulong, N.A., Latif, M.T., Nadzir, M.S.M., Amil, N., Hussain, D.F.M., Lee, V.,Hosaini, P.N., Shaharom, S., Yusoff, N.A.Y.M., Hoque, H.M.S., Chung, J.X., Sahani, M.,Tahir, N.M., Juneng, L., Maulud, K.N.A., Abdullah, S.M.S., Fujii, Y., Tohno, S., Mizohata,A., 2016b. Comprehensive assessment of PM2.5 physicochemical properties duringthe Southeast Asia dry season (southwest monsoon). J. Geophys. Res.-Atmos.14589–14611.

Kim, K.H., Kabir, E., Kabir, S., 2015. A review on the human health impact of airborne par-ticulate matter. Environ. Int. 74, 136–143.

Koe, L.C., Arellano, A.F., McGregor, J.L., 2001. Investigating the haze transport from 1997biomass burning in Southeast Asia: its impact upon Singapore. Atmos. Environ. 35,2723–2734.

Kusumaningtyas, S.D.A., Aldrian, E., 2016. Impact of the June 2013 Riau province Sumatrasmoke haze event on regional air pollution. Environ. Res. Lett. 11, 075007.

Kusumaningtyas, S.D.A., Aldrian, E., Rahman, M.A., Sopaheluwakan, A., 2016. Aerosolproperties in Central Kalimantan due to peatland fire. Aerosol Air Qual. Res. 16,2757–2767.

Lestari, P., Mauliadi, Y.D., 2009. Source apportionment of particulate matter at urbanmixed site in Indonesia using PMF. Atmos. Environ. 43, 1760–1770.

Lewis, C.W., Norris, G.A., Conner, T.L., Henry, R.C., 2003. Source apportionment of phoenixPM2.5Aerosol with the Unmix receptor model. J. Air Waste Manage. Assoc. 53,325–338.

Li, J., Posfai, M., Hobbs, P.V., Buseck, P.R., 2003. Individual aerosol particles from biomassburning in southern Africa: 2, compositions and aging of inorganic particles.J. Geophys. Res. 108, 8484.

Liss, P., Johnson, M.T., 2014. Ocean-atmosphere interactions of gases and particles. first ed.Springer-Verlag, Berlin Heidelberg, Berlin, Jerman.

Lopez-Avila, V., Young, R., Beckert, W.F., 1994. Microwave-assisted extraction of organiccompounds from standard reference soils and sediments. Anal. Chem. 66,1097–1106.

Ma, N., Zhao, C., Chen, J., Xu, W., Yan, P., Zhou, X., 2014. A novel method for distinguishingfog and haze based on PM2.5, visibility, and relative humidity. Sci. China Earth Sci. 57,2156–2164.

Manousakas, M., Diapouli, E., Papaefthymiou, H., Migliori, A., Karydas, A.G., Padilla-Alvarez, R., Bogovac, M., Kaiser, R.B., Jaksic, M., Bogdanovic-Radovic, I.,Eleftheriadis, K., 2015. Source apportionment by PMF on elemental concentra-tions obtained by PIXE analysis of PM10 samples collected at the vicinity of lig-nite power plants and mines in Megalopolis, Greece. Nucl. Inst. Methods Phys.Res. B 349, 114–124.

Martins, V., Moreno, T., Minguillon, M.C., van Drooge, B.L., Reche, C., Amato, F., de Miguel,E., Capdevila, M., Centelles, S., Querol, X., 2015. Origin of inorganic and organic com-ponents of PM in subway stations of Barcelona, Spain. Environ. Pollut. 1-12.

Morrison, N.L., Webster, H.N., 2005. An assessment of turbulence profiles in rural andurban environments using local measurements and numerical weather prediction re-sults. Bound.-Layer Meteorol. 115, 223–239.

Muraleedharan, T.R., Radojevic, M., Waugh, A., Caruana, A., 2000. Chemical characterisa-tion of the haze in Brunei Darussalam during the 1998 episode. Atmos. Environ. 34,2725–2731.

Murillo-Tovar, M., Saldarriaga-Noreña, H., Hernández-Mena, L., Campos-Ramos, A.,Cárdenas-González, B., Ospina-Noreña, J., Cosío-Ramírez, R., Díaz-Torres, J., Smith,W., 2015. Potential sources of trace metals and ionic species in PM2.5 in Guadalajara,Mexico: a case study during dry season. Atmosphere 6, 1858–1870.

Nasir, M.H., Choo, W.Y., Rafia, A., Md., M.R., Theng, L.C., Noor, M.M.H., 2000. Estimation ofhealth damage cost for 1997-haze episode in Malaysia using the Ostro model.Procedings of the Malaysian Science and Technology Congress (MSTC).

Nichol, J., 1998. Smoke haze in Southeast Asia: a predictable recurrence. Atmos. Environ.32, 2715–2716.

Paatero, P., 1997. Least squares formulation of robust non-negative factor analysis.Chemom. Intell. Lab. Syst. 37, 23–35.

Paatero, P., Eberly, S., Brown, S., Norris, G., 2014. Methods for estimating uncertainty infactor analytic solutions. Atmos. Meas. Tech. 7, 781.

Pachon, J.E., Weber, R.J., Zhang, X., Mulholland, J.A., Russell, A.G., 2013. Revising theuse of potassium (K) in the source apportionment of PM2.5. Atmos. Pollut. Res. 4,14–21.

Pan, X., Liu, H., Jia, G., Shu, Y.Y., 2000. Microwave-assisted extraction of glycyrrhizic acidfrom licorice root. Biochem. Eng. J. 5, 173–177.

Pandolfi, M., Gonzalez-Castanedo, Y., Alastuey, A., Jesus, D., Mantilla, E., de la Campa, A.S.,Querol, X., Pey, J., Amato, F., Moreno, T., 2011. Source apportionment of PM10 andPM2.5 atmultiple sites in the strait of Gibraltar by PMF: impact of shipping emissions.Environ. Sci. Pollut. Res. 18, 260–269.

Park, D., Oh, M., Yoon, Y., Park, E., Lee, K., 2012. Source identification of PM10 pollution insubway passenger cabins using positive matrix factorization. Atmos. Environ. 49,180–185.

Pey, J., Alastuey, A., Querol, X., 2013. PM(1)(0) and PM(2).(5) sources at an insular loca-tion in the western Mediterranean by using source apportionment techniques. Sci.Total Environ. 456-457, 267–277.

Pinto, J.P., Lester, D.G., Hartlage, T.A., 1998. Report on U.S. EPA Air Monitoring of Hazefrom S.E. Asia Biomass Fires, North Carolina.

Pui, D.Y.H., Chen, S.-C., Zuo, Z., 2014. PM2.5 in China: measurements, sources, visibilityand health effects, and mitigation. Particuology 13, 1–26.

Rahman, S.A., Hamzah, M.S., Wood, A.K., Elias, M.S., Salim, N.A.A., Sanuri, E., 2011. Sourcesapportionment of fine and coarse aerosol in Klang Valley, Kuala Lumpur usingpositive matrix factorization. Atmos. Pollut. Res. 2, 197–206.

Rahman, S.A., Hamzah, M.S., Elias, M.S., Salim, N.A.A., Hashim, A., Shukor, S., Siong, W.B.,Wood, A.K., 2015. A long term study on characterization and source apportionmentof particulate pollution in Klang Valley, Kuala Lumpur. Aerosol Air Qual. Res. 15,2291–2304.

Rastogi, N., Singh, A., Singh, D., Sarin, M., 2014. Chemical characteristics of PM 2.5 at asource region of biomass burning emissions: evidence for secondary aerosol forma-tion. Environ. Pollut. 184, 563–569.

Rodríguez, S., Querol, X., Alastuey, A., de la Rosa, J., 2007. Atmospheric particulate matterand air quality in the Mediterranean: a review. Environ. Chem. Lett. 5, 1–7.

Saarikoski, S., Sillanpää, M., Sofiev, M., Timonen, H., Saarnio, K., Teinilä, K., Karppinen, A.,Kukkonen, J., Hillamo, R., 2007. Chemical composition of aerosols during a major bio-mass burning episode over northern Europe in spring 2006: experimental andmodelling assessments. Atmos. Environ. 41, 3577–3589.

Sahani, M., Zainon, N.A., Wan Mahiyuddin, W.R., Latif, M.T., Hod, R., Khan, M.F., Tahir,N.M., Chan, C.-C., 2014. A case-crossover analysis of forest fire haze events and mor-tality in Malaysia. Atmos. Environ. 96, 257–265.

See, S.W., Balasubramanian, R., Wang,W., 2006. A study of the physical, chemical, and op-tical properties of ambient aerosol particles in Southeast Asia during hazy andnonhazy days. J. Geophys. Res.-Atmos. 111, (n/a-n/a).

Shen, Z., Cao, J., Arimoto, R., Han, Z., Zhang, R., Han, Y., Liu, S., Okuda, T., Nakao, S., Tanaka,S., 2009. Ionic composition of TSP and PM2.5 during dust storms and air pollution ep-isodes at Xi'an, China. Atmos. Environ. 43, 2911–2918.

Song, Y., Zhang, Y., Xie, S., Zeng, L., Zheng, M., Salmon, L.G., Shao, M., Slanina, S., 2006.Source apportionment of PM2. 5 in Beijing by positive matrix factorization. Atmos.Environ. 40, 1526–1537.

Sowlat, M.H., Naddafi, K., Yunesian, M., Jackson, P.L., Shahsavani, A., 2012. Source appor-tionment of total suspended particulates in an arid area in southwestern Iran usingpositive matrix factorization. Bull. Environ. Contam. Toxicol. 88, 735–740.

570 N.A. Sulong et al. / Science of the Total Environment 601–602 (2017) 556–570

Srimuruganandam, B., Shiva Nagendra, S.M., 2012. Application of positive matrix factori-zation in characterization of PM(10) and PM(2.5) emission sources at urban roadside.Chemosphere 88, 120–130.

Stanimirova, I., Tauler, R., Walczak, B., 2011. A comparison of positive matrix factorizationand the weighted multivariate curve resolution method. Application to environmen-tal data. Environ. Sci. Technol. 45, 10102–10110.

Sun, Y., Zhuang, G., Tang, A., Wang, Y., An, Z., 2006. Chemical characteristics of PM2.5 andPM10 in haze-fog episodes in Beijing. Environ. Sci. Technol. 40, 3148–3155.

Tan, J.-H., Duan, J.-C., Chen, D.-H., Wang, X.-H., Guo, S.-J., Bi, X.-H., Sheng, G.-Y., He, K.-B.,Fu, J.-M., 2009. Chemical characteristics of haze during summer andwinter in Guang-zhou. Atmos. Res. 94, 238–245.

Thurston, G.D., Spengler, J.D., 1985. A quantitative assessment of source contributions toinhalable particulate matter pollution in Metropolitan Boston. Atmos. Environ. 19,9–25.

USEPA, 2009. Integrated Science Assessment for Particulate Matter (Final Report). UnitedState Environmental Protection Agency, Washington, D.C.

USEPA, 2016. Human Health Risk Assessment. United State Environmental ProtectionAgency, Washington D.C.

Varikoden, H., Preethi, B., Samah, A.A., Babu, C.A., 2011. Seasonal variation of rainfall char-acteristics in different intensity classes over Peninsular Malaysia. J. Hydrol. 404,99–108.

Wang, D., Tian, F., Yang, M., Liu, C., Li, Y.F., 2009. Application of positive matrix factoriza-tion to identify potential sources of PAHs in soil of Dalian, China. Environ. Pollut. 157,1559–1564.

Watson, J.G., Chow, J.C., Fujita, E.M., 2001. Review of volatile organic compound sourceapportionment by chemical mass balance. Atmos. Environ. 35, 1567–1584.

Xu, L., Chen, X., Chen, J., Zhang, F., He, C., Zhao, J., Yin, L., 2012. Seasonal variations andchemical compositions of PM2.5 aerosol in the urban area of Fuzhou, China. Atmos.Res. 104-105, 264–272.

Xu, J., Tai, X., Betha, R., He, J., Balasubramanian, R., 2014. Comparison of physical andchemical properties of ambient aerosols during the 2009 haze and non-haze periodsin Southeast Asia. Environ. Geochem. Health 37, 831–841.

Yang, L., Cheng, S., Wang, X., Nie,W., Xu, P., Gao, X., Yuan, C., Wang,W., 2013. Source iden-tification and health impact of PM2.5 in a heavily polluted urban atmosphere inChina. Atmos. Environ. 75, 265–269.

Yang, H., Chen, J., Wen, J., Tian, H., Liu, X., 2016. Composition and sources of PM2.5 aroundthe heating periods of 2013 and 2014 in Beijing: implications for efficient mitigationmeasures. Atmos. Environ. 124, 378–386.

Yatkin, S., Bayram, A., 2008. Source apportionment of PM(10) and PM(2.5) using positivematrix factorization and chemical mass balance in Izmir, Turkey. Sci. Total Environ.390, 109–123.

Zhao, W., Cheng, J., Li, D., Duan, Y., Wei, H., Ji, R., Wang, W., 2013. Urban ambient airquality investigation and health risk assessment during haze and non–haze periodsin Shanghai, China. Atmos. Pollut. Res. 4, 275–281.