Characterization of PM Using Multiple Site Data in a Heavily Industrialized Region of Turkey

Source apportionment has most often been applied to a time series of data collected at a single site. However, in a complex airshed where there are multiple sources, it may be helpful to collect samples from multiple sites to ensure that some of them have low contributions from specific sources such that edges can be properly defined. In this study, samples were collected at multiple sites in the Aliaga region (38°40′–38°54′N and 26°50′–27°03′E) located in western Turkey on the coast of the Aegean Sea. This area contains a number of significant air pollution sources including five scrap iron-steel processing plants with electric arc furnaces (EAFs), several steel rolling mills, a petroleum refinery, a petrochemical complex, a natural gas-fired power plant, a fertilizer plant, ship breaking yards, coal storage and packaging, scrap storage and classification sites, large slag and scrap piles, heavy road traffic, very intense transportation activities including ferrous scrap trucks and busy ports used for product and raw material transportation. A total of 456 samples of PM10 at six sampling sites and 88 samples of PM2.5 at one site were collected for four seasons and the elemental composition was determined for 43 elements. The newest version of EPA PMF (V5.0) that has the capability of handling multiple site data was used for source apportionment. Eight factors were identified as iron-steel production from scrap (23.4%), resuspended and road dust (23.3%), crustal (20.5%), marine aerosol (14.4%), biomass and wood combustion (7.2%), salvage activities (4.7%), coal combustion (3.7%) and residual oil combustion (2.8%). The pattern of source contributions and conditional probability function analysis were consistent with the locations of the known sources. Thus, the multiple site data allowed for a comprehensive identification of the primary sources of PM in this region.


INTRODUCTION
Particulate matter (PM) may be emitted into the atmosphere from a variety of natural (i.e., soil erosion, sea spray, volcanic activities, natural forest fires) and anthropogenic sources (i.e., industrial activities, traffic emissions, residential heating, fossil fuel combustion including coal and biomass burning).In addition, they can be formed by the chemical transformation of organic compounds or inorganic gases in the atmosphere as secondary organic aerosol (SOA) (Kroll and Seinfeld, 2008) and secondary inorganic aerosol (SIA) (Belis et al., 2013).Atmospheric particulate matter with aerodynamic diameter smaller than 10 µm (PM 10 ) has been identified as one of the most significant of air pollutants in terms of its environmental and health impacts.Many studies indicated that PM 10 can affect the climate (Tainio et al., 2013), reduce the visibility (Polissar et al., 2001;Chang et al., 2009), affect other ecosystems (sea, soil, or vegetation) (Leung and Jiao, 2006;Odabasi et al., 2010;Hofman et al., 2013;Im et al., 2013), and high PM 10 concentrations may also play a role in the severity and incidence of respiratory diseases such as aggravated asthma, coughing and painful breathing, chronic bronchitis, and decreased lung functions (Ostro et al., 1999;Zheng, 2011;Chen et al., 2013).
Thus, the characterization of particulate matter (PM) is important for regulators and researchers because of its potential impact on human health and long-range transport crossing international border (Stefan et al., 2010;Kim et al., 2012).In order to have an efficient air quality management system and its regulatory approaches, it is necessary to have reliable air quality data and understand the spatial and temporal variations of PM and its compositions and sources.Several studies have been carried out in recent years to ascertain particulate matter's physical and chemical characteristics (Chen et al., 2003;Zheng et al., 2004;Braga et al., 2005;Jones and Harrison, 2006;Adamo et al., 2008;Huang et al., 2011;Han et al., 2014).
Source identification is an important step in air quality management.Therefore, receptor modeling has been widely applied to identify and apportion sources of particulate matter based on chemical species data collected at the receptor sites (Zabalza et al., 2006;Contini et al., 2012;Srimuruganandam and Shiva Nagendra, 2012;Tecer et al., 2012;Begum et al., 2013;Gianini et al., 2013).Positive Matrix Factorization (PMF) has become the most widely used receptor modeling approach (Paatero and Hopke, 2003).
Air pollution has become a severe problem in many parts of the world because of the intense industrial and urban activities (Chan and Yao, 2008;Yue et al., 2008;Stone et al., 2010;Lee et al., 2011;Vienneau and Briggs, 2013;Xue et al., 2013).Following rapid social and economic development over the past several decades, particulate matter and its elemental composition in urban and industrial regions has also become significant in Turkey.Several studies related to monitoring of the PM 10 have been performed at regional and national scales in Turkey (Karaca et al., 2005;Polat and Durduran, 2012;Kara et al., 2013), and the chemical composition and sources of atmospheric PM have also been studied in many parts of the country (Karakas et al., 2004;Yatin et al., 2000;Bayraktar et al., 2011;Durukan et al., 2013).The Anatolia located in the Eastern Mediterranean receives air masses from different directions (Ozturk et al., 2012).Major sources of atmospheric particles in the Mediterranean areas, including the area of the present study, are the long-range transport of mineral dust from deserts of North Africa (Herut et al., 2001), and industrial emissions from Eastern Europe (Sciare et al., 2003).Dogan et al. (2008) indicated that potential industrial source areas include the Balkan Countries, Ukraine, and regions located north of Ukraine.In addition, the local anthropogenic or industrial emissions, biogenic emissions, and sea salt are other significant PM sources for the region (Odabasi et al., 2002;Kocak et al., 2007;Turkum et al., 2008;Tecer et al., 2012).In some heavily industrialized regions of Turkey (i.e., Iskenderun, Gebze, and Aliaga), local sources are more dominant compared to other sources, and the PM concentrations can frequently exceed air quality standards (Karademir, 2006;Yatkin and Bayram, 2008;Odabasi et al., 2010;Pekey et al., 2010).
Over the past 30 years, the Aliaga region has undergone a rapid transition from an agricultural region to a heavily industrialized region, and has formed a complex industrial structure arising from iron-steel production, petroleum refining, petrochemical plants, and other industries (i.e., fertilizer plants, steel rolling mills).The establishment of industrial activities without proper planning followed by the subsequent population increase resulted in adverse impacts on the local ecosystem and these poses a potential threat to the human health.Therefore, it is necessary to identify the PM sources and apportion PM to those sources to provide efficient air quality management for this region.
The objectives of this study were (1) to determine the spatial and temporal variations of PM 10 and PM 2.5 concentrations, and (2) to identify the PM 10 sources using multiple sites data by EPA PMF (V5.0) in the Aliaga Region.PM samples (PM 10 from six sites and PM 2.5 from one site) were collected in the region during four sampling campaigns (summer, fall, winter and spring) performed between June 2009 and April 2010.PM samples were analyzed for trace elements using ICP-MS and the results were evaluated by PMF and CPF to identify their possible sources.

Study Area
The Aliaga region (38°40′-38°54′N and 26°50′-27°03′E) is located in the western part of Turkey on the coast of the Aegean Sea.It contains a number of significant air pollution sources including five scrap iron-steel processing plants with electric arc furnaces (EAFs), several steel rolling mills, a large petroleum refinery, a petrochemical complex, a natural gas-fired power plant, a fertilizer plant, ship breaking yards, coal storage and packing, scrap storage and classification sites, large slag and scrap piles, heavy road traffic, very intense transportation activities including ferrous scrap trucks and busy ports used for product and raw material transportation.Aliaga town with a population of ~60 000, several villages, agricultural areas, and some resort sites are also located within the region.The locations of the sampling sites, industrial activities and settlements in the study area are illustrated in Fig. 1.
PM 10 inlets that fulfill the US EPA specifications were used for PM 10 as well as PM 2.5 sampling (EPA, 2000).The particles enter the inlet at a flow rate 16.7 L/min by means of an air sampler pump supplying a constant volumetric flow rate.In this study, a Partisol™ 2025i-D Dichotomous air sampler (www.thermoscientific.com/AQI) which can simultaneously collect fine and coarse PM samples was used to collect PM 2.5 and PM 10 samples in Bozkoy site.Five PM 10 sampler systems [PM 10 inlet (Model PF 20630; Zambelli S.r.l.) with digital pump systems (ISODUST HF, PF 12000PE-01; Zambelli S.r.l.) (www.zszambelli.com)]were used at the remaining sampling sites.Teflon filters (PTFE) (47 mm, 2.0-µm pore size, Whatman) were used for sampling.The particulate masses were determined by weighing the filters before and after the exposure with a microbalance (Mettler-Toledo Model XP26, capable of weighing 1 µg).Quality assurance exercises were provided  Since the temperature and humidity in the weighing room may affect the moisture content of the filters, and thus its weight, filters were equilibrated in a conditioned room (50 ± 5% relative humidity and 20 ± 2°C temperature) for at least 24 hours before being weighed.The filters were digested in 8 mL HNO 3 (Merck, Suprapur® ) and 5 ml HCl (Merck, Suprapur® ) mixture with a microwave digestion system (MARS 5, CEM Corp.).Then the samples were diluted to 50 mL with deionized water (18.2MΩ/cm) and filtered through a 0.45 µm PTFE filter (Millipore) before analysis.Analysis of trace elements (Ag, Al, As, B, Ba, Bi, Ca, Cd, Ce, Co, Cr, Cu, Dy, Er, Fe, Ga, Gd, Hg, K, La, Li, Mg, Mn, Mo, Na, Nd, Ni, P, Pb, Pr, Rb, Sb, Se, Sm, Sn, Sr, Th, Tl, U, V, Y, Yb and Zn) were carried out using Inductively Coupled Plasma -Mass Spectrometry (ICP-MS) (Agilent 7700x, with HMI).Blanks were prepared simultaneously for a routine check for estimation of each metal in the reagents and blank filters.
For quality assurance, NIST standard reference material SRM 1648a Urban Particulate Matter was analyzed.Quality control/quality assurance procedures were applied during the sample preparation and analysis.The continuing check verification (CCV-1) standard solution (High Purity Standards, Charleston, SC) was used to check the validity of calibration curve during analysis.The limit of detection (MDL) of the method was defined as the mean blank mass plus three standard deviations (MDL = mean blank+3 SD).

Meteorological Conditions
The region is mainly classified within the local climate of the Mediterranean Sea that is characterized by hot, dry summers and cool, wet winters and springs.The annual rainfall reaches to 688 mm with the highest monthly rainfall of 131 mm falling during December (MGM, 2013).Monthly average temperatures in sampling periods were measured as 28.2, 21.0, 10.6 and 16.1°C for July, October, January and April, respectively.The prevailing winds are northwest and southeasterly in the region.The wind rose plots were generated using WRPLOT View (Lakes Environmental, Canada) for the four sampling periods.They are shown in Fig. S1 in the supplementary material.The annual prevailing wind directions were: WNW, 26.5%; NW, 13.3%; N, 13.0 and S, 8.8% and mean wind speed was 3.1 m/s during the sampling period.

Positive Matrix Factorization (PMF)
Positive Matrix Factorization (PMF) in the form of EPA PMF V5.0 was used in this study.This version is described in greater detail by Paatero et al. (2013).Uncertainties as well as the BDL concentrations and missing values were treated by the approaches described by Polissar et al. (1998).The uncertainties (σ ij ) for the ICP/MS data were estimated from the following: x DL   for samples below limit of detection where x ij is the determined concentration for species j in the i th sample, and DL j is the detection limit for species j.The coefficients 0.2 and 0.1 in Eqs. ( 1) and ( 2) were empirically determined (Zabalza et al., 2006).PMF has most often been applied to a time series of data collected at a single site.However, PMF generally requires a large number of samples to make a stable and reliable source identification.Also, in a complex airshed having multiple sources, it may be helpful to collect samples from multiple sites to ensure that some of them have low contributions from specific sources.Therefore, the newest version of EPA PMF (V5.0) that can handle multiple site data was used in the present study.

Conditional Probability Function
To identify directionally of local point sources, a conditional probability function (CPF) (Ashbaugh et al., 1985;Kim et al., 2004) was calculated using source contribution estimates resolved by PMF analysis and wind speed and direction data measured at the site.The same daily fractional contribution was assigned to each hour of a given day to match to the hourly wind data.Specifically, the CPF is defined as; where m Δθ is the number of occurrences from wind sector Δθ that are upper 25 percentile of the fractional contributions, n Δθ is the total number of observations from the same wind sector.In this study, the size of the wind direction sector was set to 11.25°.Calm winds (< 1 m/s) were excluded from this analysis.The sources are likely to be located in the directions that have high conditional probability values.

Kruskal-Wallis (K-W) Test
To assess the spatial and temporal variability of the PM and elemental concentrations, Kruskal-Wallis (K-W) tests were performed.The K-W test is a non-parametric test to assess whether the samples originate from the same distribution.It can determine if more than two samples are independent.The null hypothesis is that all the samples come from the same distribution based on non-significant differences in the population medians.If the significance level is smaller than 0.01, the null hypothesis is rejected and the alternate hypothesis is accepted.

PM 10 and PM 2.5 Concentrations
A total of 456 samples of PM 10 from six sampling sites and 88 samples of PM 2.5 from one site were collected over four seasons.Thus, spatial and seasonal variability of PM 10 and PM 2.5 were evaluated in the region.During the study, the annual average values of PM 10 concentrations were 39.9, 50.4, 53.6, 55.0, 54.1, and 49.8 µg/m 3 for Aliaga town, Helvaci, Bozkoy, Horozgedigi, Cakmakli, and the ship breaking yards, respectively.These annual PM 10 values exceeded the European Air Quality Directive, 2008/50/EC (EC, 2008) limit value of 40 μg/m 3 (a PM 10 limit value of 40 μg/m 3 was also set by the Air Quality Assessment and Management Regulation of Turkey) at all sites except Aliaga town.These results indicate that local industrial sources significantly affect the PM mass concentrations at the sampling sites.
The Horozgedigi, Cakmakli and Bozkoy sites were located in same area and were closest to the iron-steel plants with electric arc furnaces (EAFs) and steel rolling mills.In this area called the Nemrut area, many fugitive sources (i.e., large slag and scrap piles, unpaved roads, coal storage and packing, a very dense transportation activity of trucks) influence the entire region.The PM 10 levels during the sampling days were highest in this area, exceeding 100 µg/m 3 .Helvaci and Aliaga town sites were more distant (~10 km) from the Nemrut area.Helvaci had the high PM values because of urban activities (i.e., traffic, heating) and industrial activities since it was downwind to the industrial region.The lowest concentrations were measured in Aliaga town.The ship breaking site reflected the ship breaking activities.Lower values were measured in the winter due to the reduced activity and weather conditions.High concentrations were again measured in the spring season.PM 10 concentrations indicated similar seasonal patterns for all sampling sites.
Regarding the spatial and seasonal variability, there were significant differences in PM 10 concentrations among six sampling sites in the all seasons (p < 0.01) (Table 2).For the Bozkoy, Aliaga town, Cakmakli, and ship breaking sites, the K-W test indicated significant differences among sampling seasons (p < 0.001).Significant differences were not observed across seasons at Helvaci (p = 0.634) and Horozgedigi (p = 0.171).In summer, the PM 10 concentrations were slightly higher than those in other seasons, probably due to re-suspended PM from unpaved roads, storage piles, soil, and industrial facilities during this dry season.
Regarding the attainment of the daily limit value in EU Directive 2008/50/EC (EC, 2008), the 90.4 percentile value exceeded the limit value of 50 μg/m 3 at all sites, even in Aliaga town that had 17 exceedances (20% of samples).The exceedance ratio was approximately 50% for the rest of sampling sites during the sampling period.
The annual average PM 2.5 concentration was 28.3 µg/m 3 at Bozkoy site (Table 3).This value is above the air quality limit (25 µg/m 3 ) for PM 2.5 (EC, 2008).The seasonal PM 2.5 values varied between 12 to 48, 10 to 52, 11 to 77, and 12 to 55 µg/m 3 during summer, fall, winter, and spring, respectively.Seasonal PM 2.5 concentrations were not significantly different among the sampling seasons (p = 0.737).The highest PM 2.5 was in the summer season followed by winter and spring.The lowest average PM 2.5 concentration (27.1 µg/m 3 ) was measured in fall.
The PM 2.5 /PM 10 ratios were calculated from PM 2.5 and PM 10 concentrations measured at Bozkoy site.The ratio of PM 2.5 /PM 10 is used to assess whether PM 10 is dominated by anthropogenic or crustal sources.High PM 2.5 /PM 10 ratios (> 0.5) indicate that the anthropogenic sources contribute to PM 10 to a greater extent.In this study, the average PM 2.5 /PM 10 ratios showed a clear seasonal pattern ranging between 0.41 and 0.77.The ratio of PM 2.5 /PM 10 was higher in winter (0.77) compared to other seasons (Table 3).The higher ratios in winter were due to lower PM 10 values during the wet season when suspension and re-entrainment of soil were minimal.Winter rainfall was 119 mm compared with 48, 45, and 2 mm in spring, fall, and summer, respectively.Summer had the lowest PM 2.5 to PM 10 ratios since crustal sources (i.e., soil, wind entrainment, re-suspended dust) were the dominant sources in the region.The values in this area were slightly lower than the ratios measured in Alsasua, Spain (0.70 to 0.86) (Zabalza et al., 2006); Athens, Greece (0.45 to 0.78) (Pateraki et al., 2012); Agra, India (0.55 to 0.76) (Kulshrestha et al., 2009); and central Taiwan (0.56 to 0.72) (Fang et al., 1999).These results suggested that the coarse PM significantly contributed to total PM 10 mass in the study area.

Elemental Concentrations
Elements were determined in 229 PM 10 and in 48 PM 2.5 samples.Table 4 lists the average concentrations, standard deviations, and percent of missing and below detection limits observations of each element in PM 10 samples used for PMF analysis.The elemental concentrations in PM 10 and PM 2.5 were dominated by crustal and sea salt elements (Ca, Mg, Al, K, and Na).These elements were followed by Fe, Zn, Mn and Pb emitted from scrap handling and ironsteel production.Their concentrations were also high at all of the sampling sites, but the relative importance of these elements changed according to sampling site properties.
The elemental concentrations for PM 10 were usually higher in Bozkoy and at the ship breaking site compared to the other sites, especially for Fe, Zn, Pb, Mn, Cr, Cu, Ni, Sn, Sb, As, Cd, Ag, and Co that are related to scrap handling and iron-steel production.The Bozkoy site was mainly influenced from the activities located in the Nemrut area given its proximity and the prevailing wind direction.Cakmakli and Horozgedigi are also located within the Nemrut area.The ship breaking site reflected salvage facilities.The Cu concentrations were higher at this site since copper was emitted from recycling copper wires and electronic wastes (U.S. Geological Survey, 2013).K and Rb showed higher concentrations in Helvaci that may be attributed to the influence of biomass burning.The lowest concentrations were obtained in Aliaga town that is relatively far from the Nemrut area, but close to major roads, the refinery, and petrochemical complexes.Therefore, each site showed differences according to the influences of nearby sources.
Fig. S2 depicts the average trace element concentrations for all sampling sites.While the K-W test (p < 0.01) indicated that there were significant differences in the concentrations of Ag, As, Ba, Ca, Cd, Co, Cr, Cu, Er, Fe, Ga, K, Mg, Mn, Mo, P, Pb, Rb, Sb, Se, Sn, U, Y, and Zn, there were no significant differences for Al, B, Bi, Ce, Dy, Gd, Hg, La, Li, Na, Nd, Ni, Pr, Sm, Sr, Th, Tl, V, and Yb among the sampling sites.
The elemental concentrations (ng/m 3 ) were generally higher in spring, followed by summer and fall and they were the lowest in winter.Some elements indicated different patterns depending on the sampling sites.While the mean concentrations of K, Cr, V, Ni, Mo, Sb, Ag, Se, and Y in PM 10 were higher in summer, P, B, As, and Bi were higher in winter and fall.Table S1 lists the average seasonal elemental concentrations for PM 10 in the study area.
For PM 2.5 , the concentrations of Fe, Zn, Pb, Mn, Cu, V, Ni, Cd, Mo, Ba, Sb, and Se were higher in the summer and fall seasons, whereas As, B, Rb, Bi, Tl, U, Co, and Ga were higher in winter time.Some of the trace elements were found in concentrations below the detection limit in the PM 2.5 .Thus, seasonal comparisons are not possible for the fine fraction.The mean seasonal concentrations for PM 2.5 in Bozkoy are given in Table S2.More detailed discussion on the elemental concentrations is provided in the Supplementary Material.

Source Apportionment
The elemental composition of sources of particulate matter was resolved by PMF.To determine the optimal number of sources, 5−8 factors were examined.The model was run 20 times with a random seed to determine the stability of Q values; Q values were stable and all runs converged.The Q (robust) and Q (true) values were 2444.The Q values, the resulting source profiles, and the scaled residuals distributions were studied and the eight factor solution was identified based on the scaled residuals distributions and the interpretability of the resulting profiles (Belis et al., 2013;Wang et al., 2013b).All of the scaled residual distributions were approximately symmetric within the range of −3 and +3 that represents good agreement between the observed and predicted values (Friend et al., 2012;Li et al., 2013).In addition, the marker elements belonging to different PM sources resolved by PMF in previous studies were summarized in Table S4.
Eight factors were identified: iron-steel production from scrap (23.4%), re-suspended and road dust (23.3%), crustal (20.5%), marine aerosol (14.4%), biomass and wood combustion (7.2%), salvage activities (4.7%), coal combustion (3.7%) and residual oil combustion (2.8%).These factors represented only primary PM 10 sources since species such as sulfate and nitrate were not included.Also there were no measurements of carbonaceous species.The source profiles are depicted in Fig. 2, and Fig. 3 presents the time-series plots of the estimated daily contribution from each factor to the PM mass for different sampling sites.
The first factor was ascribed to iron-steel production from scrap.The factor was associated with high contribution of Fe, Mn, Pb, Zn, Sn, Ag, Cd, and Cr and with high concentrations of Fe, Mn, Zn, and Pb.These elements are marker elements for steel production from scrap (Thurston et al., 2011;Yatkin and Bayram, 2008;Mansha et al., 2012).These results were consistent with PM stack samples collected from electric arc furnaces in the study area.The elemental analysis for PM 10 and PM 2.5 in stack gas indicated that the PM contained substantial amounts of Fe, Zn, Mn, K, Mg, Ca, Al, Pb, Cd, Cr, Cu, and Sn.The average percentage contributions were determined as 20-30% for Zn, 20-27% for Fe, 2-4% for Pb, and 1-2% for Mn.
Steel production was a major source (23.4%) and dominated the sites near these plants.The highest contributions were in Bozkoy followed by Horozgedigi and Helvaci because of the prevailing northwesterly winds.The CPF plots (Fig. 4) showed the Nemrut area containing several steel production plants and steel rolling mills contributing to almost all of the sites.
The second source profile has high contributions of Ca, Ba, Co, Cu, Ni, Se, Mg, Mn, Se, Sr, and Zn (Escrig et al., 2009;Fabretti et al., 2009), and high concentrations of Ca, Fe, Mg, and Zn.It was defined as re-suspended and road dust from the Nemrut area where there are significant fugitive sources (i.e., paved and unpaved roads, slag piles, EAFs filter dust piles, and coal piles).Uncontrolled PM emissions from roads and piles by resuspension from loading and dumping activities are produced through wind entrainment.Relatively low contributions of some crustal elements (i.e., Al, K, and some rare earth elements) to this factor may be due to the fact that resuspended material from road dust and other fugitive sources is highly contaminated.Peaks were observed in summer and fall and the highest contributions were found at Bozkoy and Horozgedigi.These factor contributions were also seen in Cakmakli and Aliaga town.Cakmakli is located on the route of trucks carrying scrap from ports to plants and the major arterial roads (Canakkale-Izmir) pass through Aliaga town.The CPF plots in sampling sites mainly point to the Nemrut area (Fig. 5).
The third factor profile was dominated by high contributions of Al, Ca, Mg, Ba, Sr (Begum et al., 2005;Beuck et al., 2011) and lanthanides including La, Ce, Dy, Er, Nd, Pr, Sm, Th, Y, and Yb (Trapp et al., 2010;Gianini et al., 2012).The concentrations of Ca, Al, and Mg are high representing soil/crustal sources.It contributes 20.5% of the PM mass.Fig. 3 shows the seasonal variability.The peaks were observed during spring, when wind speeds were higher.Similar patterns are seen at all of the sites supporting an assignment of windblown soil.The relationships between soil species such as Al vs. Mg, Ca vs. Sr, Ba vs. Mg, Ca vs. Mg and lanthanides were strong (r 2 = 0.60-0.99)and statistically significant (p < 0.01).CPF plots (Fig. S3) indicated different directions for each sampling site since the areas influencing them were different.The CPF plots point to rural areas (agricultural) and Nemrut industrial area.
Factor 4 is characterized by Na, Mg, Al, Sr, and P (Aldabe et al., 2011;Kocak et al., 2011).It is associated with marine aerosol.The factor showed high contributions (14.4%) in this region since the sampling region is located on the Turkish shore of the Aegean Sea.In the seawater  samples collected from the seashore in the region, the highest concentrations were measured for Na, Mg, and K, followed by Al and P (unpublished data).High P concentrations in seawater could be due to wastewater discharges from the fertilizer plant and contaminated stream waters in the area.The time series plot indicates relatively constant contributions during summer and fall in Horozgedigi and Helvaci and during winter and spring in Cakmakli and the ship breaking.The Cakmakli and ship breaking sites were located on a peninsula and they were affected by different wind directions.The highest contribution was obtained in Cakmakli site.The contribution of Cr and Hg also appeared in this factor.Similar to P concentrations, Cr was measured at high levels in Aliaga Bay seawater (Unpublished data).Refinery wastewaters contain high amounts of Cr (Gerhardt and Maroney, 1994).Therefore, source of Cr in seawater might be the wastewater discharges from the refinery located in the area.On the other hand, Hg was frequently associated with chlorine-caustic soda production (SOLVAY, 2006) in petrochemical complexes.The petrochemical complex in the study area discontinued the mercury-cell process and utilized the membrane-cell process in July 2000, however sediment Hg concentrations remained high (unpublished data).Hg concentrations in seawater were not measured in the study area.However, they might be high due to sediment to seawater transfer of Hg.Marine aerosol did not represent clear directionality in the CPF plots owing to the sampling sites' regional directionality relative to the sea (Fig. S4).
The biomass and wood combustion factor (number 5) was clearly represented by the dominance of K and Rb in the profile (Santoso et al., 2008;Gianini et al., 2012;Wang et al., 2013a) as well as the high concentrations of Ca, Na and Mg.Crop residue burning on farmland especially wheat croplands is a common agricultural practice in this region.Crop residue burning is used to remove excess residue to facilitate planting and control pests and weeds prior to planting or reseeding (Dhammapala et al., 2006;Wulfhorst et al., 2006;Huang et al., 2012).Small-scale forest fires occur in the summer and contribute to this factor.In addition, wood is used as a fuel along with coal in residences in winter.Coal combustion represents another factor discussed below.Examining the time series plot, the summer peaks may be explained by crop residue burning, while small winter peaks indicate wood combustion for residential heating.This factor is high in Helvaci because it is surrounded by croplands (Fig. S5).
The sixth factor is characterized by Cu, Sb, Sn and Cd.
These elements, especially Cu and Sb have been widely used for copper wire (ATSDR, 1992;Hileman, 2002;Beavington et al., 2004).Copper recycling by removing the plastic insulation from copper wire and electronic wastes are performed on the scrap from ship breaking.This procedure is conducted by uncontrolled burning of the plastic coating in open areas.Although this is illegal, it is used to reduce costs.The time series indicates that the high contributions for this factor appear in ship breaking yards.Therefore, this factor represents the salvage activities.In Fig. 6, the salvage factor show clear directionality in the CPF plots at the ship breaking and Aliaga town sites located near the source area.It is not observed at the other sites because it is a relatively weak source.The elemental composition for factor 7 is dominated by As, B, Hg, Se, Tl, U, and Zn as seen in Fig. 2. Se, As, and Tl are often used as a marker elements for coal combustion (Ogulei et al., 2006;Zhou et al., 2009) and As has also been identified with Sb and Zn as markers for coal-fired power plant emissions (Moreno et al., 2007).The factor profile and high contributions in winter is attributed to coal combustion for residential heating.The time series shows the same pattern at all of the sampling sites except the ship breaking site.These impacted sites were located within the villages and Aliaga town.The coal combustion CPF plots did not indicate clear directionality because the sites are surrounded by residences (Fig. S6).
The last factor (factor 8) was associated with V and Ni that are signature elements for residual oil combustion emissions (Kim and Hopke, 2004;Mazzei et al., 2008;Kocak et al., 2009).This factor also contains Mo suggesting impacts from the refinery and petrochemical complexes (Bosco et al., 2005;Bozlaker et al., 2013).In addition, the contributions of La and Ce appeared in this factor as related to refinery activities (Alleman et al., 2010).The time series plot did not show clear seasonal differences, but the contributions were lower in the winter at all of the sampling sites.The CPF plots point to peninsula where the refinery and petrochemical complexes are located (Fig. 7).

CONCLUSIONS
In this study, spatial and seasonal variability of PM 10 and PM 2.5 were evaluated for four seasons in the Aliaga region by the samples collected at six sites for PM 10 and at one sites for PM 2.5 .The limit values for the protection of human health in European air quality Directive were exceeded for both PM 10 and PM 2.5 .The establishment of industrial activities without proper planning and subsequent population increase has caused the increased PM concentrations in this region.The significant PM sources such as iron-steel production and fugitive sources (i.e., paved and unpaved roads, slag piles, EAFs filter dust piles, and coal piles) have also contributed to the PM mass.This made necessary to take precaution for PM concentrations and to improve an efficient air quality management system.
The elemental concentrations were measured to be higher at sites located near industrial activities in Nemrut area and Fe, Zn, Pb, Mn, Cr, Cu, Ni, Sn, Sb, As, Cd, Ag, and Co that are related to scrap handling and iron-steel production has high concentrations compared with the rest of measured elements.
PM 10 sources using multiple sites data by EPA PMF (V5.0), having the capability of handling multiple site data, were determined for region.Eight factors were identified as iron-steel production from scrap (23.4%), re-suspended and road dust (23.3%), crustal (20.5%), marine aerosol (14.4%), biomass and wood combustion (7.2%), salvage activities (4.7%), coal combustion (3.7%) and residual oil combustion (2.8%).The pattern of source contributions and conditional probability function analysis were consistent with the locations of the known sources.Thus, the multiple site data allowed for a comprehensive identification of the primary sources of PM in this region.

ACKNOWLEDGMENTS
This study was funded by the "Assessment of current status of Aliaga industrial region for air pollution" project conducted by Dokuz Eylul University and supported by the Turkish Ministry of Environment and Urbanism and by the industries located in Aliaga region.We would also like

Fig. 1 .
Fig. 1.Locations of the sampling sites, industrial activities and settlements in the study area.

Fig. 3 .
Fig. 3. Time-series plots of the estimated daily contribution from each factor to the PM mass.

Fig. 4 .
Fig. 4. The CPF plot showing directions for iron-steel production from scrap.

Fig. 5 .
Fig. 5.The CPF plot showing directions for re-suspended and road dust.

Fig. S3 .
Fig. S3.The CPF plot showing directions for crustal and soil.

Fig. S5 .
Fig. S5.The CPF plot showing directions for biomass and wood combustion.

Table 1 .
Summary of the sampling information.

Table 2 .
Seasonal and annual PM 10 concentrations (µg/m 3 ) for the six sampling sites.

Table 3 .
Seasonal and annual PM 2.5 concentrations (µg/m 3 ) and PM 2.5 /PM 10 ratios in four seasons.

Table 4 .
PMF input data statistics for the elemental composition of PM 10 samples.
* S/N donates signal to noise ratio.

Table S3 .
The elemental concentrations (ng/m 3 ) in comparison with other industrial sites.

Table S4 .
The marker elements belong to different sources resolved by PMF.