Chemical Characteristics and Source Apportionment of PM2.5 and Long-range Transport from Northeast Asia Continent to Niigata in Eastern Japan

Seasonal intensive sampling was undertaken for two weeks during each of four seasons from May 2015 to February 2017 at Niigata-Maki station in Niigata, eastern Japan. Daily mean concentrations of PM2.5 ranged from 4.2 μg m to 33.4 μg m during the observation period, which were lower than Japanese Environmental Quality Standard for PM2.5 (35 μg m). The higher concentrations of SO4, NH4 and OC were observed in spring and summer, which may result from photochemical activity and secondary OC production. The major chemical components of PM2.5 at Niigata-Maki site were SO4, NO3, NH4, OCM, EC and crustal elements. Compared with data at other urban sites, a lower concentration of EC and NO3 and higher OC/EC ratio were observed at Niigata-Maki site, which may result from no significant stationary source and low vehicular traffic in the rural site. PM2.5 source apportionment was characterized by positive matrix factorization (PMF) analysis, and the results inferred four major sources: sea salt (10.2%), biomass combustion (18.9%), soil dust (13.2%) and secondary aerosol (44.4%). The potential source contribution function (PSCF) analysis suggested that the major sources of secondary aerosol and sea salts were domestic in southwest Japan and the Sea of Japan, whereas the sources of biomass combustion and soil dust in specific seasons were long range transportation from the Northeast Asian continent (NEA). Comparing with previous studies in western Japan, this study showed a large domestic contribution of southwest Japan for secondary aerosol, while a larger contribution of the NEA was observed in the previous studies. Significant contribution of biomass combustion from northeast China in autumn, and local area in Niigata and southwest Japan in the other seasons was uniquely observed in this study.


INTRODUCTION
Fine particulate matter (PM 2.5 ) is considered as an important environmental pollutant and has adverse effects on human health, including effects on the heart, nervous, and vascular system (Zeller et al., 2006;Chan et al., 2006;Leiva G et al., 2013). It is also well documented in various investigations that particulate matter has influence on human health in Japan (Murakami et al., 2006;Yamazaki et al., 2007). Moreover, PM 2.5 is proved to be linked to degradation of visual range, with the features of light extinction (Yang et al., 2007). To evaluate these effects, components of PM 2.5 need to be investigated at different temporal and spatial 939 NEA to western Japan (Kaneyasu et al., 2014). High levels of polycyclic aromatic hydrocarbons (PAHs) in total suspended particles (TSP) are transported from the NEA to western Japan (Coulibaly et al., 2015). In addition to observation studies, sensitivity analysis by using a regional chemical transport model demonstrated that contribution from central north China accounts for 50-60% of PM 2.5 at a remote island in western Japan except in summer (Ikeda et al., 2014).
Niigata City is the capital and the most populous city of Niigata Prefecture located in eastern Japan. It lies on the coast of the Sea of Japan and faces the Northeast Asian continent, and the climate is strongly affected by the circulation of the East Asian Monsoon over East Asia, covering eastern China, the Korean Peninsula and Japan. In winter it's dominated by cold currents from the northwest, while in summer the wet and humid currents of tropical origin prevail. Some studies reported that the effects of long-range transport on the areas located downstream of large emission sources play a major role in ambient air quality (Shimadera et al., 2009). Since the NEA is upwind with respect to Niigata during the winter monsoon period, PM 2.5 in Niigata is probably influenced by long-range transport from the NEA. However, few observation studies have focused on fine particles across the coast of the Sea of Japan from the NEA, such as transboundary transport of anthropogenic sulfur in PM 2.5 (Inomata et al., 2016), long-range transport from the NEA of biomass combustion (Huo et al., 2016), and long term variations of PAHs in total suspended particles at a background site on Noto Peninsula (Tang et al., 2015). Therefore, it is important to identify the sources of PM 2.5 in a coastal city of the Sea of Japan, such as Niigata, and contribution to long-range transportation from the NEA in order to provide scientific findings for evaluation of atmospheric impact of PM 2.5 in eastern Japan. In this research, a field observation study including seasonal intensive measurement of PM 2.5 was conducted from May 2015 to February 2017 at the Niigata-Maki national acid deposition monitoring station in Niigata, eastern Japan. A comprehensive analysis of thirty three chemical components (including EC, OC, Cl - , NO 3  -, SO 4 2-, Na + , NH 4 + , K + , Mg 2+ , Ca 2+ and 23 metallic elements) in PM 2.5 was implemented and the seasonal variations of chemical characteristics in PM 2.5 were elucidated. Then, statistical analyses of correlation matrix, positive matrix factorization (PMF) for annual and potential source contribution function (PSCF) for annual and specific season were conducted in order to identify major sources of PM 2.5 in Niigata, and also identify the long-range transport from the NEA.

The Monitoring Station
Niigata-Maki national acid deposition monitoring station (37°48'32", 138°52'10", 52 m altitude) is located at the foot of Mt. Kakuda (482 m a.s.l.), 1 km from the seashore, and 25 km southwest of the center of Niigata City, the capital of Niigata Prefecture, Japan, as shown in Fig. 1. This station was constructed by the Ministry of the Environment, Japan to monitor acid deposition and air pollutant concentration and to investigate acid deposition in rural areas on the coast of the Sea of Japan. There are no industrial sources near Niigata-Maki site, but a small community (approximately 1300 population) is located 2 km northwest of the station, and thus it is classified as a rural station. Air masses reaching the station are dependent on seasonal wind patterns, which are affected by the monsoon circulation: in winter the northwest cold currents prevail, while in summer they are replaced by the hot and humid currents of the Pacific Ocean.
For the reference data, we used PM 2.5 data at the Sadoseki national acid deposition monitoring station (38°15'02", 138°24'01"), which is classified as a remote station, and the Kameda air quality monitoring station (37°52'22", 139°06'05"), which is classified as an urban station. The Fig. 1. Location of the study monitoring station: Niigata-Maki and the reference sites: Sado-seki, Kameda, and Chiba. Li et al., Aerosol and Air Quality Research, 18: 938-956, 2018 940 data of Kameda and Sado-seki sites are cited from the web site of the Ministry of the Environment, Japan (MOEJ) (MOEJ, 2016). Furthermore, we also used the PM 2.5 continuous monitoring data at the Chiba site located in Tokyo Metropolitan Region of eastern Japan and classified as an urban station. It was conducted as a full year continuous monitoring of PM 2.5 by a research project (Ichikawa et al., 2015). Niigata and Tokyo metropolitan areas are located in eastern Japan. Niigata lies on the Sea of Japan side, whereas Tokyo metropolitan area lies on the Pacific Ocean side. The central mountains of Honshu (Main Island of Japan) are located between them. Therefore, the meteorological conditions and emission sources of both areas would be different, and it is scientifically meaningful to compare the characteristics between them. Moreover, a comprehensive analysis of chemical components in PM 2.5 was also implemented at the Chiba site, which is comparable to our observational results. Thus, we referred the data at Chiba site on behalf of Tokyo metropolitan area.

Sampling and Chemical Analysis
Daily PM 2.5 samples were taken for two weeks during each of four seasons from May 2015 to February 2017, and each daily sampling time was conducted from noon to noon of the next day. The sampling periods in this study were the same as the national intensive survey of PM 2.5 conducted by MOEJ as shown in Table 1. JFY denotes the Japanese fiscal year, which is the period from April to March. We collected a total of 106 daily samples during two years. PM 2.5 samples were collected on the PTFE membrane filters (Zefluor, 47 mm in diameter and 2.0 µm in pore size, Pall Co., USA) using a sequential FRM sampler (Model 2025, Thermo Fisher Scientific, Inc., USA), designed with a flow rate of 16.7 L min -1 and sequential 24-hour sampling for 14 days. The PTFE filters were used for mass concentration, water-soluble ions and metallic elements measurement. PM 2.5 samples were also collected on the quartz filters (2500QAT-UP, 8 × 10 inches, Pall Co., USA) by using a high volume sampler (HV-1000F, Sibata Scientific Technology Ltd., Japan) and PM 2.5 impactor (HVI-2.5, Tokyo Dylec Co., Japan), designed with a flow rate of 576 L min -1 . By using this high-volume sampler, daily PM 2.5 sampling was conducted for 7 days (core time). Before and after the core time, 4 days and 3 days continuous sampling were also conducted respectively, so the total sampling period using the high-volume sampler is the same as the sequential FRM sampler (14 days). We used the average concentration for daily data in continuous sampling period. The quartz filters were used for carbonaceous species measurement. The two samplers were set at 5 meters above the ground and 1.5 m above the rooftop. At the Sado-seki site, PM 2.5 was monitored by continuous β-ray absorption monitor Kimoto Electoric Co.,Ltd.,Japan). At Kameda and Chiba sites, the same type of FRM sampler as the Niigata-Maki site was used (2025i Sequential Air Sampler Thermo-Scientific Inc., USA, Flow rate: 16.7 L min -1 and 24 m 3 day -1 ).
The mass concentrations of sample filters were determined gravimetrically by using the electronic microbalance (MSE2.7S-000, sensitivity = 0.1 µg, Sartorius AG, Germany) under controlled temperature (20-23°C) and relative humidity (RH at 35-45%). Teflon filters were stored under controlled conditions of temperature and relative humidity for at least 24 hours before and after sampling, and then passed through a static electricity remover before being weighed at least twice. The maximum acceptable difference between the two readings was 3 µg, and all processes were performed in a weighing chamber (PWS-PM2.5, Dylec Co., Japan).
For analyses of water-soluble ion, half of a filter was extracted with 20 mL of Milli-Q water (18.2 MΩ · cm) for 30 minutes by using a shaker (SR-2w, Taitec Co., Japan). The extracted solution was filtered through a pre-washed membrane filter (A045A025A, 25 mm in diameter and 0.45 µm pore size, Advantec Toyo Co., Japan), and stored in a refrigerator at 4°C until chemical analysis. Three anions (Cl -, NO 3 and SO 4 2-) and five cations (Na + , NH 4 + , K + , Mg 2+ and Ca 2+ ) were determined by ion chromatography (ICS-2100 and ICS-1100, Thermo Fisher Scientific, Inc., USA).
Organic carbon (OC) and elemental carbon (EC) were analyzed by using a Desert Research Institute (DRI, USA) Model 2001A thermal/optical carbon analyzer. Collected quartz filters were punched into 0.505 cm 2 pieces to determine OC and EC by the IMPROVE method with the thermal optical reflection protocol (Chow et al., 2001). Depending on combustion temperature and matrix gas, carbonaceous components were fractionated into Organic carbon (OC1-OC4) and Elemental carbon (EC1-EC3), and the optical pyrolysis correction of OC (PyC) by reflectance of helium-neon laser. OC was defined as OC1 + OC2 + OC3 + OC4 + PyC and EC was as EC1 + EC2 + EC3 -PyC.
The concentrations of 23 metallic elements (Al, Fe, Ni, Cr, Zn, Mn, Co, Cu, V, Ga, As, Se, Rb, Sr, Sc, Mo, Cd, Sb, Cs, Ba, Pb, Ag and Hg) were analyzed using an Thermo Fisher XSeries 2 Inductively Coupled Plasma Mass Spectrometry (ICP-MS). A quarter of a PTFE filter was put into a PTFE pressure digestion container, and then dissolved in 6 mL of 70% nitric acid (HNO 3 ) and 3 mL of 50% hydrofluoric acid (HF) and 1 mL of 30% hydrogen peroxide (H 2 O 2 ). All digestions were performed in a microwave oven (Ethos 900, Milestone General Inc., USA), the steps of the digestion procedure are shown as Table 2. After the microwave digestion procedure, the digestion containers  were heated until the solution volume is reduced to approximately 0.1 mL so that all of HF is evaporated. Then, the samples were diluted with 1 mol L -1 HNO 3 to make a 15 mL solution, was added into both sample and standard solutions. All reagents used for the digestion procedure were ultra-pure grade quality or better, and all regent solutions were prepared using MILLI-Q water. All samples and standards were stored in metal-free centrifuge tubes and kept in a refrigerator at 4°C before analysis.

Quality Assurance (QA) and Quality Control (QC)
The detection limits were calculated as equivalent air concentrations of 3 times of the standard deviations for 8 laboratory blank samples. The detection limits of OC and EC were 0.26 and 0.08 µg m -3 , respectively. Working standard samples and ion balance were tested for the water soluble ion analysis by ion chromatography. The original working standard solution of artificial rain water containing 3 anions of Cl -, NO 3 and SO 4 2and 5 cations of Na + , NH 4 + , K + , Mg 2+ and Ca 2+ was obtained by 18th interlaboratory comparison project on wet deposition in Acid Deposition Monitoring Network in East Asia (EANET) (Network Center for EANET, 2016). Working standard solutions were prepared from the original solution by diluting 100 times with MILLI-Q water. The accuracy of water soluble ion analysis was checked by agreement between the measured values and the prepared values within 15% differences.
We also use ion balance for PM 2.5 component to check validity of analytical procedure for water soluble ions. The principle of electroneutrality in an extract of PM 2.5 sample requires that the total anion equivalents equal to the total cation equivalents. According to this principle, ion balance in the extract was checked by the below equation.
where C and A represents total anion and total cation equivalent concentrations (µeq L -1 ), and R represents ion balance, respectively. The seasonal variations of water soluble ion balance were shown in Fig. 2, and all seasonal ion balance values were satisfied with the required criteria (Network Center for EANET, 2010).
To check validity of analytical protocol for metallic elements, the Certified Reference Material (CRM) No. 28 Urban Aerosols were used (National Institute for Environmental Studies, 2015). The recoveries of metallic elements were determined by 9 CRM samples, only the recoveries of Sb (75%) less than 80%, and the other metallic elements ranged from 83 to 109%. Therefore, Sb concentration was determined by the recovery ratio, as shown in Table 3. The analytical results were in a good agreement with the certified or reference values within analytical error. Table 4 shows that the annual mean mass concentrations of PM 2.5 were 12.3 µg m -3 in JFY 2015 and 9.3 µg m -3 in JFY 2016, respectively. Although the monitoring was not conducted for a full year, these values were lower than the annual average of the Japanese Environmental Quality Standard (JEQS) for PM 2.5 (15 µg m -3 ). Daily mean concentrations of PM 2.5 ranged from 4.2 µg m -3 to 33.4 µg m -3 from May 2015 to February 2017, which were lower than the daily average of the JEQS for PM 2.5 (35 µg m -3 ), as shown in Fig. 3. Daily PM 2.5 mass concentrations at Niigata-Maki, Kameda and Sado-seki showed similar temporal trends throughout the observation period. The higher mass concentration days were observed in spring and summer, and the lower mass concentration days were observed in autumn and winter (Fig. 3). The seasonal means are discussed in the following paragraph. Daily PM 2.5 mass concentrations at Niigata-Maki strongly correlated with those at Kameda (r = 0.82, p < 0.01) and Sado-seki (r = 0.80, p < 0.01), which implies these sites are affected by similar sources and local meteorology (Fig. 4).

Temporal Variations of PM 2.5 Mass Concentration
We compared the seasonal averages of PM 2.5 mass concentrations with various meteorological parameters of rainfall (mm), wind speed (m s -1 ), and temperature (°C). The meteorological data was obtained from the nearby monitoring station of the Japan Meteorological Agency (JMA) (JMA, 2017), which is located 6 km southeast of the Niigata-Maki station. As shown in Fig. 5, the seasonal variations of rainfall, wind speed and temperature might effect on the variation of PM 2.5 concentration. The higher seasonal PM 2.5 mass concentration is likely to be observed in the higher temperature, windless and less rainfall weather. These meteorological conditions corresponded to higher PM 2.5 concentrations in spring and summer. Previous studies also showed that higher temperatures often enhance the chemical reactions of atmospheric pollutants (Sanchez-Romero et al., 2014;Zhang et al., 2015), that strong winds are favorable for the diffusion of air pollutants (Csavina et al., 2014), and that PM 2.5 was negatively correlated to the quantity of accumulated rainfall because of wet removal processes (Ouyang et al., 2015).

Chemical Characteristics of PM 2.5 Water Soluble Ions
Since the sampling site is located close to the seashore, it is important to identify the contributions of water soluble ions in PM 2.5 from non-marine and marine sources. Nonsea-salt ion concentration (nss-M) was estimated by the following equation assuming that Na + in particulate matter is solely originated from sea salts: where M is the target ion, K is the mass ratio of the target ion to Na + in seawater (Mg 2+ = 0.1190, Ca 2+ = 0.0382, K + = 0.0370, SO 4 2-= 0.2515, Cl -= 1.79) (Kennish et al., 1994).  , nss-K + , nss-Ca 2+ , nss-Mg 2+ and nss-Clto respect ionic concentrations in PM 2.5 were 97%, 94%, 34%, 14%, 2%, respectively. This result suggested that the major sources of SO 4 2and K + in PM 2.5 were non-marine sources, and Cl -, Mg 2+ and Ca 2+ were mainly originated from sea salts. Table 4, the annual average concentrations of total water-soluble ions (TWSIs) in JFY 2015 and JFY 2016 were 6.5 µg m -3 and 3.8 µg m -3 , which accounted for 52.8% and 40.9% of PM 2.5 mass concentration in JFY 2015 and JFY 2016, respectively. Clear seasonal variations of TWSIs were observed at Niigata-Maki from May 2015 to 945 February 2017. The seasonal mean concentrations of TWSIs were in the order of summer > spring > winter > autumn (Table 4). The concentrations of sulfate (SO 4 2-) and ammonium (NH 4 + ) were highest among all water soluble ions in PM 2.5 . The annual means of SO 4 2and NH 4 + were 58.0% and 18.5% of the concentration of TWSIs in JFY 2015, 53.1% and 18.6% in JFY 2016, respectively. The higher concentrations of SO 4 2and NH 4 + were observed in spring and summer. NH 4 + and SO 4 2are known as tracer species of secondary aerosol, and the increased photochemical activity was one of the possible reasons for enhanced sulfate concentration during the period (Husain et al., 1990). Additionally, the annual mean molar ratio of NH 4 + to SO 4 2was 1.7 and 1.9 in JFY 2015 and 2016, respectively, the values were close to the molar ratio of NH 4 + to SO 4 2in (NH 4 ) 2 SO 4 (2.0) . This indicated that NH 4 + and SO 4 2were mainly from secondary aerosol (NH 4 ) 2 SO 4.

Carbonaceous Species Analysis
Carbonaceous species in the atmosphere exist in three forms: organic carbon (OC), elemental carbon (EC), and carbonate carbon (CC). Because the ion balance of extract of PM 2.5 were satisfied with the required criteria, the contribution of carbonate to PM 2.5 was estimated to less than 5%. Therefore, total carbon (TC) in PM 2.5 was considered to be the sum of measured OC and EC (µgC) without pretreatment of removal of carbonate.
As shown in Table 4, the annual mean concentrations of TC were 2.4 µgC m -3 and 2.1 µgC m -3 , which accounted for 19.5% and 22.6% of PM 2.5 concentrations in JFY 2015 and 2016, respectively. OC dominated over EC, and the annual mean OC/EC ratios were 2.6 and 3.2 in JFY 2015 and 2016, respectively. The ratio of OC/EC is often used as an indicator of the sources for the carbonaceous aerosol. We compared the annual mean OC/EC ratios between the Niigata-Maki site with those at some urban sites in Yokohama (Khan et al., 2010), Chiba (Ichikawa et al., 2015) and Kameda (MOEJ, 2016) in Japan. The annual mean OC/EC ratios at the Niigata-Maki (2.6 and 3.2 in JFY 2015 and 2016) were higher than those at the urban sites of Yokohama (1.9), Chiba (2.5) and Kameda (2.2). The Niigata-Maki site is classified as a rural site where there is no significant stationary source and low vehicular traffic. Because EC is considered as a tracer of diesel emission or fossil fuel combustion of stationary sources, the higher OC/EC ratio at the Niigata-Maki also represented the characteristics of a rural site.
The seasonal EC and OC concentrations were summarized in Table 4. EC concentration showed no strong seasonal variation. However, clear seasonal variation of OC was observed at the Niigata-Maki site. The seasonal mean concentrations of OC were in the order of spring > summer > autumn > winter during JFY 2015 and 2016. This result is similar to the results of some previous observational studies in Japan. For examples, the contribution of secondary formation, especially in summer, is more important for organic aerosols in Saitama, a suburban area of Tokyo (Bao et al., 2009). Moreover, the seasonal average contributions of secondary OC to total OC in PM <2.1 were 60%, 75%, and 36% in spring, summer, and winter in Gunma Prefecture, a suburban area of Tokyo, respectively (Kumagai et al., 2010). The high contributions of secondary OC are attributable to high photochemical oxidant concentrations (> 100 ppb). The observed high OC concentrations in spring and summer may result from higher temperatures and more intense solar radiation, which provide favorable conditions for photochemical activity and secondary OC production (Khan et al., 2010).

Metallic Elements Analysis
In this study, 23 metallic elements (Al, Fe, Ni, Cr, Zn, Mn, Co, Cu, V, Ga, As, Se, Rb, Sr, Sc, Mo, Cd, Sb, Cs, Ba, Pb, Ag and Hg) in PM 2.5 were also measured during sampling period. The seasonal and annual mean concentrations and the crustal enrichment factor (EF) for each elements is summarized in Table 5. The EF has commonly been used as a first step in attempting to evaluate the strength of the crustal versus anthropogenic sources, defined as:  Furuta et al. (2005) suggested that Ni and Cr were originated from chemical industry sources, and Sb was from the combustion of plastic products and brake pad wear of automobiles. The EFs of Ni, Cr, Se were also high (55, 35, 6698) at the Hachimantai mountain range in northern Japan. This result suggested that Cr was originated from industrial sources, and the Se was from coal combustion or exhaust of vehicles (Kikuchi et al., 2010). These previous studies can be used for references to identify anthropogenic emission sources of the high EFs elements. As shown in Table 5, the annual mean concentrations of a total 23 metallic elements were 702.3 ng m -3 (JFY 2015) and 279.2 ng m -3 (JFY 2016), which were 5.7% and 3.0% of the annual means of PM 2.5 at Niigata-Maki site, respectively. The concentrations of iron (Fe) and aluminum (Al) were highest among 23 metallic elements in PM 2.5 . The major metallic elements Al and Fe to total metallic elements were accounted for 24.2% and 51.6% at Niigata-Maki site in JFY 2015, and 52.4% and 29.2% in JFY 2016, respectively. No obvious seasonal variation of metallic elements was The concentrations of Ag and Hg were under detection limits during sampling periods, and thus they were not shown in Table 5.
observed, but we found a remarkably high concentration event of Fe, Ni and Cr in winter JFY 2015. These metallic elements were recommended as markers for the metal processing industry, especially for steel works (Chen et al., 2014). The waste incineration was also a source of Ni and Cr (Reimann et al., 1998).

PM 2.5 Mass Balance
Summation of determined major chemical components (SO 4 2-, NO 3 -, Cl -, NH 4 + , Na + and EC) and estimated fraction (Organic carbon matter (OCM) and soil dust) were compared with PM 2.5 mass concentration to evaluate whether the mass balance discrepancy was significant. The following equation was used to estimate OCM including O, H, N compounds: where OCM is organic carbon matter; k is adjustment factor to account for non-carbon organic matter (1.4) (Bell et al., 2007); OC m is measured organic carbon; OC b is organic carbon for blank filters. Soil dust was estimated from the measured concentrations of Al, Mg, K, Ca, Fe, and the estimated concentration of Si. The silicon concentration was estimated from the average mass ratio of Si to Al in the earth's crust as Si = 3.41Al (Mason, 1966), the concentration of soil dust was calculated with the following equation (Hueglin et al., 2005).
Soil dust = 1.89Al + 1.66Mg + 1.21K + 1.40Ca + 1.43Fe + 2.14Si (5) Fig. 6 shows the mass balance between the PM 2.5 mass and the sum of the aerosol chemical components. The annual means PM 2.5 mass was 6.2% and 17.0% larger than the sums of the aerosol chemical components in JFY 2015 and 2016, respectively. The discrepancy between the PM 2.5 mass and the sums of the aerosol chemical components varies among different seasons, and the PM 2.5 mass exceeded maximum 30.8% in spring JFY 2016. On the other hand, the PM 2.5 mass was maximum 29.5% lower than the sum of the aerosol chemical components in winter JFY2015. This can be explained by the remarkably higher concentration of Fe. The EF value of Fe was 7.56 in winter JFY 2015, which implies that a significant portion of Fe was originated from anthropogenic sources. Since all of Fe measured concentrations were calculated soil dust by the Eq. (5), the portion of soil dust was irregularly higher, and thus the sum of the aerosol chemical component exceeded the PM 2.5 mass in winter JFY 2015. Previous studies suggested that the mass balance discrepancy and its seasonal shift may be due to a combination of aerosol water content retained on the conditioned filters, and volatilization losses of PM 2.5 (Hering et al., 1999;Pang et al., 2002). Fig. 7 shows annual average concentrations and annual average contributions of major components in PM 2.5 at Niigata-Maki site. The same data at Kameda, Chiba and Japan national-average are also shown as references. As shown in Fig. 7(b), water-soluble ions, OC and EC, which are major components of PM 2.5 , accounted for 52.8%, 14.5% and 5.4% of the annual means PM 2.5 mass concentration at Niigata-Maki site in JFY 2015, and accounted for 40.8%, 17.3% and 5.3% in JFY 2016, respectively. The sum of NO 3 -, SO 4 2and NH 4 + accounted for 82.0% and 80.6% of the water-soluble ions in JFY 2015 and 2016, respectively. The "Others" fraction is defined as PM 2.5 mass concentration minus the sum of water-soluble ions, EC and OC concentrations. The others accounted for 27.3% and 36.6% of the annual means PM 2.5 mass concentration in JFY2015 and 2016, respectively. This fraction is assumed to contain trace elements, organic matters other than carbon (OM), crustal elements of soil dust and analytical uncertainties of the measurement. OM was estimated by Eq. (4), OM = OCM -OC. Soil dust was estimated by Eq. (5). The annual sum of OM and soil dust accounted for 69.4% of the "Others" fraction, as shown in Fig. 6. These results showed that SO 4 2-, NO 3 -, NH 4 + , OCM, EC and crustal elements were the major chemical components of PM 2.5 in Niigata.

Major Components of PM 2.5
We compared the mass concentration and the contributions of major components of PM 2.5 among Niigata-Maki, Kameda (MOEJ, 2016), Chiba (Ichikawa et al., 2015) and Japan national average (MOEJ, 2016). As shown in Fig. 7(a) mean mass concentration of PM 2.5 at Niigata-Maki site was lower than Kameda, Chiba and Japan national average, and Chiba annual mean exceeded the annual JEQS mass concentration (15 µg m -3 ). These results are probably due to respective site locations that Niigata-Maki is a rural site and Kameda is an urban site in Niigata, and that the Chiba site is located in Tokyo metropolitan area.
As shown in Fig. 7(b), the ratio of total sea salt related ions of Cl -, Na + , Mg 2+ and Ca 2+ to the PM 2.5 mass at Niigata-Maki (8.2% in JFY 2015, 7.8% in JFY 2016) was obviously higher than Kameda (4.1%), Chiba (3.1%) and Japan national average (2.3%), which is mainly due to the site location being close to the coast of the Sea of Japan. The ratios of EC to the PM 2.5 mass at Niigata-Maki (5.4% in JFY 2015, 5.3% in JFY 2016 were lower than at Kameda (10.3%), Chiba (8.0%) and Japan national average (7.0%), and the OC/EC ratios of Niigata-Maki (2.6 in JFY 2015, 3.2 in JFY 2016) were higher than Kameda (2.2), Chiba (2.5), which may be due to fewer traffic emission source near Niigata-Maki site. Since EC is an indicator of primary emission, diesel vehicle exhaust is probably the main source of EC in urban areas (Bao et al., 2016). The ratios of NO 3 to PM 2.5 mass at Niigata-Maki were obviously lower than Kameda and Chiba which also may be due to fewer traffic emission sources. The above differences showed that the contribution of vehicle source to PM 2.5 would be lower at the Niigata-Maki site.

Source Apportionment of PM 2.5 from PMF Analysis
EPA PMF 5.0 was used to identify the potential sources of PM 2.5 . The aim of PMF model is to minimize the function Q. Missing values were replaced by the median concentration of a given species, with an uncertainty of four times the median (Brown et al., 2015). Values below method detection limit (MDL) are retained and related uncertainties are set at 5/6 of detection limit values. For values greater than the MDL, the calculation is based on a user provided fraction of the concentration and MDL, and the error fraction was suggested as 10% by the previous study of Paatero (2007). Uncertainty is defined as: The number of factors to be chosen will depend on the user's understanding of the sources impacting samples, number of samples, sampling time resolution, and species characteristics. We run the PMF5.0 with moving the factors from two to seven, the decrease of Q/Qexpected is illustrated from 13.8 to 7.9, as well as a smaller decrease with moving from six to seven factors (8.4-7.9). When the changes in Q values become smaller than the increase of factors, it can suggest that there may be too many factors being fit (Brown et al., 2015;Liu et al., 2017). This result indicated that six factors may be the optimal solution.
The error estimation results of classical bootstrap (BS), displacement of factor elements (DISP), and bootstrap enhanced by displacement (BS-DISP), which is due to the small size of the data set and limited number of factors. The source apportionment results were generally stable at six factors, with all factors mapped in BS in 100% of runs except for the vehicle and industrial source factors (mapped on 85% and 81% of runs), mapping over 80% of the factors indicates that the BS uncertainties can be interpreted and the number of factors may be appropriate (PMF 5.0 User Guide, 2016). No swaps occurred with DISP also showed that the solution is stable, and all BS-DISP runs were successful. PM 2.5 source apportionment was characterized by the following six factors, as shown in Fig. 8, and the contributions of each source to the PM 2.5 mass was shown in Fig. 9. Each factor could be characterized as follows.
Factor 2 was characterized by high positive loading of OC (85.9%), EC (52.9%) and K + (81.5%), which are generally considered as indicators of biomass burning, and K + is a widely used tracer of biomass burning source (Almeida et al., 2015;Yao et al., 2016). Thus, the factor 2 was identified as biomass combustion, and its contribution to PM 2.5 was 18.9%.
Factor 3 was characterized by high positive loading of Al (65.1%), Ga (54.8%), Sr (57.9%) and Ba (74.8%), which were mainly originated from crustal soil. As previously mentioned, the elements of EF < 2 suggest mainly natural crustal source origin, and those of EF > 5 suggest strong enrichment by non-crustal sources (Gao et al., 2002). The mean EF values of these elements during overall period were very low (Al,0.79;Ba,1.81;and Sr,0.43). This suggests that these elements are mainly originated from natural sources and mineral matter. It was also reported that trace elements of Ga and Sr are originated from re-suspension of road dust (Amato et al., 2012). Thus, the factor 3 was identified as soil dust, and its contribution to PM 2.5 was 13.2%.  Factor 4 was characterized by high positive loading of major components of SO 4 2-(71.7%) and NH 4 + (71.1%), and their typical source is secondary aerosol such as (NH 4 ) 2 SO 4 . Factor 4 was also represented by the high positive loading of metallic elements of V (69.4%), As (46.2%), Se (56.2%) and Pb (35.6%). V is a marker element of oil combustion (Okuda et al., 2006), and As, Se and Pb are the typical anthropogenic sources of coal combustion (Reimann et al., 1998;Wahlin et al., 2006). Many PMF source apportionment studies have reported a relationship between (NH 4 ) 2 SO 4 and particulate matter components of combustion origin, such as heavy metals (Kim et al., 2004). This is likely associated with SO 2 emissions from oil or coal-burning power plants forming secondary sulfate and reacting with gaseous NH 3 from a variety of sources. Thus, the factor 4 was identified as secondary aerosol and possibly associated with coal or oil combustion processes, and its contribution to PM 2.5 was 44.4%.
Factor 5 was characterized by high positive loading of Zn (52.3%), Sb (53.1%), NO 3 -(50.8%), As (28.6%) and Pb (29.2%), and their typical source is vehicular traffic. Zn is a marker element of tire wear (Wahlin et al., 2006), and Sb is known as a tracer of brake wear dusts (Iijima et al., 2008, Iijima et al., 2009. Mn and As are also emitted from re-suspension of road traffic-generated particles (Fabretti et al., 2009). Pb is originated from tire dust or fuel combustion of vehicle (Smichowski et al., 2008). Vehicular traffic is also a significant source of gaseous precursors (NO x ) and formed secondary aerosol of nitrate (de Gouw et al., 2009). Thus, the factor 5 was identified as vehicular traffic, and its contribution to PM 2.5 was 4.0%.
Factor 6 was characterized by high positive loading of Sc (92.9), Co (81.9%), Fe (62.9%), Cr (87.3%) Ni (86.5%), Cu (36.7%) and Mo (50.1%), which were possibly attributed to industrial activities. The common anthropogenic sources of Co, Cr, Ni, Cu and Fe include steel works and smelters. They are considered as markers for metal processing industry, especially for steel works (Reimann et al, 1998;Chen et al., 2014). Co and Ni are possibly originated from crude oil processing in refineries (Speight, 2014). Waste incineration is also a source of Ni and Cr (Reimann et al, 1998). Thus, the factor 6 was identified as industrial activity, and its contribution to PM 2.5 was 0.5%.
The PMF analysis demonstrated that the major sources of PM 2.5 at the Niigata-Maki site were identified to be sea salt, biomass combustion, soil dust and secondary aerosol. Comparing with previous studies in western Japan, the major sources of PM 2.5 in an industrial area of Hyogo Prefecture, located in the southern-central region of Honshu, were identified to be secondary sulfates (28.9%), vehicular traffic (20.8%), steel mills (7.8%), and secondary chloride and nitrate (7.0%) (Nakatsubo et al., 2014). The major sources of PM 2.5 in Okinawa were identified to be coal combustion (32.6%), secondary species (28.5%), sea salt and nitrate (19.1%), oil combustion (12.8%), and soil dust (7.0%) (Shimada et al., 2015). The high contribution of secondary aerosol was common between western Japan and our study, whereas low contribution of vehicular traffic and high contribution of biomass combustion in our study showed unique characteristics in the coast of the Sea of Japan. Indeed, our identified sources included both local and long range origin. In the next section, the significant origin area of the major sources of PM 2.5 will be identified by the PSCF analysis.

Emission Regions of PM 2.5 Major Sources by PSCF Analysis
We performed the PSCF analysis to identify the preferred atmospheric transport pathways from sources to receptors by the Meteoinfo1.4.3: GIS-based software that uses various trajectory statistical analysis methods to identify potential sources from long-term air pollution measurement data . 72-h backward trajectories were obtained by using the Meteoinfo1.4.3 model from the Global Data Assimilation System (GDAS) meteorological data (ftp://gus.arlhq.noaa.gov./). We used PSCF to estimate the back trajectories at arrival heights of 500, 1000, and 1500 m, and three-day back trajectories were used to calculate the PSCF values for 1° × 1° grid cells computed for arrival heights of 500, 1000, and 1500 m for each of the 106 daily mean samples. The studying field is from 30 to 50°N, and 120 to 150°E, which includes more than 95% of area covered by all the paths. Regions with PSCF values ranging from 0.5 to 1 were designated as probable source regions for each factor during the study period. The potential source regions of the four major sources identified PMF analysis are shown in Fig. 10. The detailed discussions for respective sources are shown below.

Sea Salt
Sea salt was characterized as major components of Cl -, Na + , Mg 2+ and Ca 2+ by PMF analysis, Niigata-Maki station lies on the coast of the Sea of Japan, 1 km from the seashore, and thus, the high value PSCF plots suggested that Sea of Japan was the main source area of sea salt, as shown in Fig. 10(a).

Biomass Combustion
Biomass combustion source was characterized as major components of OC, EC and K + by PMF analysis. As shown in Fig. 10(b), the high value PSCF plots suggested that local area in Niigata, southwest Japan, northeast China and Korea were the potential source areas. The biomass combustion emissions in the NEA are generally from agricultural waste burning after barley and wheat harvest (between late spring and early summer) and rice harvest (in the autumn) (Zang, 2015). Biomass burning activity in the NEA during these period were also estimated using a fire spot database, and the fire spots data are provided by the Fire Information for Resource Management System (FIRMS) (https://firms.mo daps.eosdis.nasa.gov/firemap/), the satellite image in Fig. S1 showed dense hot spots appeared in northeast China in autumn during the sampling period. To better identify the potential sources of aerosols originating from these regions, we used seasonal fire spots to compare with the autumn biomass combustion PSCF plots (Fig. 11). The high dense regions of hotspot in autumn JFY 2015 and 2016 corresponded to high PSCF areas. These results showed that the major source of biomass combustion was from northeast China in autumn, and local area in Niigata and southwest Japan were the main source areas in the other seasons. Our previous study also showed high concentration episode of water insoluble organic carbon and EC in precipitation that is synchronized with high number of fire spots in the NEA, which implies considerable contribution of biomass combustion in spring and autumn (Huo et al., 2016).

Soil Dust
Crustal soil source was characterized as key elements of Al, Ga, Sr and Ba by PMF analysis. As shown in Fig. 10(c), the high value PSCF plots suggested that local area in Niigata, northeast China, Korea and the Sea of Japan were Fig. 10. Potential source regions by using PSCF analysis for annual mean: (a) sea salt, (b) biomass combustion, (c) soil dust, and (d) secondary aerosol. Li et al., Aerosol and Air Quality Research, 18: 938-956, 2018 952 the potential sources considering all seasons. When we focus on the seasonal soil dust PSCF plots in winter (Fig. 12), the major source area of soil dust was northeast China. For the other seasons, contributions of local area in Niigata and the Sea of Japan were more remarkable.

Secondary Aerosol
Secondary aerosol was characterized as major components SO 4 2and NH 4 + by PMF analysis. As shown in Fig. 10(d), the high value PSCF plots suggested that southwest Japan and Korea were the potential sources. The PMF source apportionment results showed a relationship between (NH 4 ) 2 SO 4 and combustion origin. This was likely associated with SO 2 emissions from oil or coal burning power plants. Fig. S2 show that the thermal power plants are concentrated in the southwest of Japan and Tokyo metropolitan area. Coal or oil burning factories and volcanoes were also the potential source of SO 4 2in Japan, Itahashi et al. (2017) revealed that the contribution of SO 4 2originated from volcanoes is 28.7% in summer. The concentrations of SO 4 2and NH 4 + were obviously high in summer as shown in Table 4, when the Pacific high covers the mainland in Japan and the air mass is usually transported along the ridge of the Pacific high. Moreover, the central mountains in Japan seem to hinder transportation of air pollutants from Tokyo metropolitan area. In addition, for the summertime contribution of sulfate aerosol, many previous observation and numerical model simulation studies showed the large contribution from the Asian continent, especially China. For example, Aikawa et al. (2010) revealed the strong gradient of sulfate from west to east by comparing the observational data in several stations. This is the suggestion of long-range transport of sulfate aerosol over the Asian continent. The numerical modeling revealed that central China was the main SO 4 2source region in China (Kajino et al., 2011;Itahashi et al., 2017). In Japan, the concentration of SO 4 2was highest in summer and followed in spring, which the long range transportation of SO 4 2from central China also had the great impact on sulfate concentration during the two seasons (Itahashi et al., 2017). Ikeda et al. (2015) revealed Fig. 11. Potential source regions of biomass combustion in autumn by using PSCF analysis at Niigata-Maki. Fig. 12. Potential source regions of soil dust in winter by using PSCF analysis at Niigata-Maki. Li et al., Aerosol and Air Quality Research, 18: 938-956, 2018 953 that the relative contribution of China to the annual mean PM 2.5 concentration was estimated to be 50-60% in western Japan. Comparing with the previous studies showed that southwest Japan was the important source area contributing to secondary aerosol, which included local and long-range transport sources.
Table 4 also shows that the concentrations of SO 4 2and NH 4 + were higher in spring, and the back trajectory analysis results showed that the origins of the air masses arriving at Niigata-Maki were mainly from two regions (central China and southwest Japan), as shown in Fig. S3. Comparing with the above discussion on sulfate aerosol, which central China and southwest Japan were the important source areas in Japan. Therefore, it is also probable that the two regions had a great impact on secondary aerosols in Niigata during spring.
The above results showed that the major sources of secondary aerosol and sea salts are domestic in southwest Japan and the Sea of Japan, whereas the sources of biomass combustion and soil dust in specific seasons are long range transportation from the NEA. Comparing with previous studies of PSCF analysis in urban area of western Japan (Nakatsubo et al., 2014) and in remote area of southwestern Japan (Shimada et al., 2015), this study showed a large domestic contribution of southwest Japan for secondary aerosol, while a larger contribution of the NEA was observed in previous studies. Moreover, significant contribution of biomass combustion originating from local or long range transportation was observed on the coast of the Sea of Japan. This study showed the regional characteristics of PM 2.5 sources in eastern Japan.

CONCLUSION
In this research, a field observation study including seasonal intensive measurement of PM 2.5 was conducted from May 2015 to February 2017 at Niigata-Maki station in Niigata, eastern Japan. The annual mean mass concentrations of PM 2.5 were 12.3 µg m -3 (JFY 2015) and 9.3 µg m -3 (JFY 2016). Daily mean concentrations of PM 2.5 ranged from 4.2 µg m -3 to 33.4 µg m -3 at Niigata-Maki station during observation period, which were lower than Japanese Environmental Quality Standard for PM 2.5 (35 µg m -3 for daily average). PM 2.5 concentration variation was highly correlated with meteorological conditions. We found that the higher seasonal PM 2.5 mass concentration usually occurred with higher temperature, windless and less rainfall weather, such as during spring and summer. The higher concentrations of SO 4 2-, NH 4 + and OC also observed in spring and summer may result from higher temperatures and more intense solar radiation, which provide favorable conditions for photochemical activity and secondary OC production. Annual average data showed that the major chemical components of PM 2.5 were SO 4 2-, NO 3 -, NH 4 + , OCM, EC and crustal elements. Comparing with the data at the urban sites of Kameda and Chiba, lower concentrations of EC and NO 3 and higher OC/EC ratios were observed at the Niigata-Maki site. EC is considered as a tracer of diesel emission or fossil fuel combustion of stationary sources, and the vehicular traffic is also a significant source of gaseous precursors (NO x ) and formed secondary aerosol of nitrate. Therefore, the observational results suggested that there was no significant stationary source and low vehicular traffic around the site. The higher OC/EC ratio at the Niigata-Maki site is also representative of a rural site. PM 2.5 source apportionment was characterized by positive matrix factorization (PMF) analysis, and the results inferred four major emission sources: sea salt, biomass combustion, soil dust and secondary aerosol. The relative contributions of the four major sources to PM 2.5 concentrations at Niigata-Maki were 10.2% for sea salt, 18.9% for biomass combustion, 13.2% for soil dust and 44.4% for secondary aerosol, respectively. We performed a potential source contribution function (PSCF) analysis to identify potential source regions of the four major sources. The high value PSCF plots suggested that the Sea of Japan was the main source area of sea salt. The major source region of biomass combustion was from northeast China in autumn, and local areas in Niigata and southwest Japan were the main source areas in other seasons. The major source area of soil dust was northeast China in winter, while the contributions of local areas in Niigata and the Sea of Japan were more remarkable in other seasons. The secondary aerosol of (NH 4 ) 2 SO 4 was mainly originated from southwest Japan. The PMF analysis and correlation results showed a relationship between (NH 4 ) 2 SO 4 with oil or coal combustion source. Thermal power plants are concentrated in the southwest of Japan and Tokyo metropolitan area. Coal or oil burning factories and volcanoes were also the potential source of SO 4 2in Japan. The concentrations of SO 4 2and NH 4 + were obviously high in summer, when the Pacific high covers the mainland in Japan and the air mass is usually transported along the ridge of the Pacific high. Moreover, the central mountains in Japan will hinder transportation from Tokyo metropolitan area. Therefore, southwest Japan was the major source region of secondary aerosol.
Comparing with previous source apportionment studies in western Japan, this study showed the regional characteristics of PM 2.5 sources at Niigata-Maki station in eastern Japan. A large domestic contribution from southwest Japan was observed for secondary aerosol, while a larger contribution of long range transportation from the NEA was observed in previous studies. Significant contribution of biomass combustion from local or long range transportation was uniquely observed. These results and scientific findings can be used for an evaluation of the atmospheric impact of PM 2.5 at Niigata-Maki station in eastern Japan.