Ultrafine Particles and PM 2 . 5 at Three Urban Air Monitoring Stations in Northern Taiwan from 2011 to 2013

In this study, long term measurements of PM2.5 and ultrafine particles (UFPs) for daily average mass concentration at Zhongshan (ZS), Sinjhuang (SJ), and Jhudong (JD) urban air monitoring stations were conducted from 2011 spring to 2013 autumn. The results showed that daily average UFPs mass concentrations in spring (average at 3 stations: 1.58 ± 0.74 μg m) and summer (average at 3 stations: 1.59 ± 0.53 μg m) were higher than those in autumn (average at 3 stations: 1.02 ± 0.28 μg m) and winter (average at 3 stations: 1.04 ± 0.48 μg m) due to the impacts by heavy traffic emission and new particle formation event. The effective density (ρeff) and dynamic shape factor (χ) for ultrafine particles (UFPs) were found to be 0.68 ± 0.16 g cm and 2.06 ± 0.19, respectively, suggesting that the particle morphology was irregular shape. Based on the calculated ρeff and χ, the average number and surface area concentration ratio of UFPs to those of PM2.5 at these monitoring stations was determined to be 89.0 ± 5.5% and 42.1 ± 12.8%, respectively, suggesting that UFPs contribute significantly to the health-relevant PM2.5 aerosol fraction in these stations.


INTRODUCTION
Ultrafine particles (UFPs), defined as particles with the diameter d p less than 0.1 µm (or called PM 0.1 ), are drawing public concerns because they are cleared less quickly and completely from the lung than larger PMs under chronic or repeated exposure and may translocate across the lung epithelium into the circulatory system, leading to potential adverse effect on the cardiovascular system and other organs (HEI, 2013).The adverse effect could be attributed to their number and surface area concentrations, and concentration of bounded toxic air pollutants on them (Chen et al., 2010a;Leitte et al., 2011;Hata et al., 2013).Among all PM fractions, nanoparticles were found to cause the major health risk because of low clearance and rate and high deposition efficiency in the respiratory system (Gorbunov et al., 2013).In order to understand the major sources and evolution of UFPs to facilitate the prediction of air quality and develop mitigation strategies, it is very important to conduct long term measurements of UFPs.
In the past, numerous studies used the Fast Mobility Particle Sizer (FMPS) (Westerdahl et al., 2009;Sabaliauskas et al., 2012;Betha et al., 2014;Kim et al., 2015) or the Scanning Mobility Particle Sizer (SMPS) coupled with the condensation particle counter (CPC) (Chen et al., 2010a;Breitner et al., 2011;Jayaratne et al., 2011;Wang et al., 2011;Xu et al., 2011;Gao et al., 2012;Young et al., 2012;Cheung et al., 2013;Young et al., 2013;Cheng et al., 2014) to investigate the number concentrations of UFPs.Most of the researches have shown that both local rush hour traffic emission and new particle formation (NPF) play key roles in the diurnal variation of UFPs concentrations, in which two concentration peaks in the morning and evening could be attributed to traffic emission, while one peak in the afternoon is due to NPF (Young et al., 2012;Cheung et al., 2013;Young et al., 2013;Betha et al., 2014;Nikolova et al., 2014).
In addition to pollutant sources, meteorological factors including wind speed, wind direction, precipitation, relative humidity, and temperature all have effects on seasonal and long-term variation of UFPs concentrations (Morawska et al., 2008;Wang et al., 2011;Young et al., 2012;Cheng et al., 2014).However, the seasonal variation and long-term measurements of UFPs number concentrations in Taiwan were limited only the hourly average and diurnal patterns of UFPs number concentration in central and northern Taiwan were reported (Young et al., 2012;Cheng et al., 2014).
There are many previous long-term PM 2.5 studies.For example, Yu et al. (2013) conducted a long term PM 2.5 study for a full year from January 1 to December 31, 2010 and indentified seven source types by PMF analysis, which included secondary sulphur (26.5%), vehicle exhaust (17.1%), fossil fuel combustion (16%), road dust (12.7%), soil dust (10.4%), biomass burning (11.2%), and metal processing (6.0%).Contributions of different sources varied by different seasons with the combustion source types, including biomass burning and fossil fuel combustion, being significantly higher in the fall and winter compared to those in the spring and summer.Huang et al. (2014) found that the increased mass concentrations on high PM days were mainly caused by air pollutant transport from the outside-Hong Kong regions, in a full year study from March 2011 to February 2012 at HKUST AQRS (Air Quality Research Supersite).The major PM 2.5 components (crustal materials, organic matter, soot, ammonium sulfate, ammonium nitrate, and non-crustal trace elements) accounted for 90% of the measured mass with sulfate being the most abundant (32.0%), followed by organic matter (23.5%) and ammonium (11.8%).PMF analysis revealed that secondary sulfate formation process (annual average of 31%), biomass burning (23%), and secondary nitrate formation process (13%) were the three dominant contributing sources to the observed PM 2.5 throughout the sampling year.
Regional dust transport has been shown to influence local PM 2.5 concentrations.Remoundaki et al. (2013) found that the concentrations of crustal elements abrupt increased during Saharan dust transport events in December 2010 in Athens, Greece.The sum of all measured chemical components accounted for about 75% of PM 2.5 concentrations in which SIA (secondary inorganic aerosols) and carbonaceous material (OM + EC) contributed almost equally for about 30% in the PM 2.5 mass, and while the dust contribution was significant, which could be up to 23% of PM 2.5 mass during dust transport events, it was only about 5% in absence of such events.In Asia, Asian dust storm (ADS) not only increase the coarse particle concentrations, but also bring the fine and ultrafine particles to Taiwan (Liang et al., 2013).Liang et al. (2013) applied the PCA model to identify the potential source categories by measuring ambient 10-500 nm particle number concentrations, size distributions and composition data in Taipei, Taiwan and found that three factors estimated during an ADS were vehicular emissions (52%), dust storm (24%), and primarily gasoline vehicles (12%), while during non-dust periods, the three factors were vehicular emissions, secondary sulfate and nitrate (40%), combustion processes and traffic-related emissions (29%), and road dust (25%).
Long-term measurements for daily average fine particles (PM 2.5 ) and UFPs mass and number concentration in northern Taiwan are still lacking.Therefore in this study, PM 2.5 and UFPs at urban air monitoring stations in northern Taiwan from May 2011 to May 2013 were investigated.In order to characterize the effective density (ρ eff ), dynamic shape factor (χ), and number concentration of UFPs, a MOUDI and a SMPS were further deployed at ZS station in May 2014 to measure the mass and number distributions simultaneously for atmospheric aerosols with 5.6 ≤ d p ≤ 209.1 nm.The mass concentrations of UPFs were then converted to number concentrations to investigate the ratio of UFPs/PM 2.5 in different metrics by using the calculated ρ eff and χ at ZS station for 3 stations.The sources and longterm trends of UFPs number concentrations were finally discussed.

Sampling Sites and Experimental Methods
The mass concentration measurements for PM 2.5 and UFPs were conducted at the Zhongshan (ZS) air monitoring station in Taipei City, the Sinjhuang (SJ) air monitoring station in New Taipei City, the Judong (JD) air monitoring station in Hsinchu County, which are shown in Figs.1(a), 1(b), and 1(c) respectively.In addition to traffic emissions as major air pollution sources, dust storm with air pollutants from mainland China transported by the north-eastern monsoon and biomass burning from Southeastern Asia also impact on the air quality at these stations.The samples at the ZS station were collected at the rooftop of a 4-floor building located on the campus of Xin-Xing Junior High School (25°03'41"N, 121°31'36"E) in Taipei (population: 2.70 million as of Dec. 2014), which is 30 m S of Minquan E. Rd, 100 m W of Xinsheng Highway, 1.5 km S of No. 1 National Freeway, 2 km SW of Taipei International Airport, and 2 km N of Taipei Main Station.
The SJ station is located at the rooftop of a four-floor building in Fu-Jen Catholic University (25°02'16"N, 121° 25'59" E) (Fig. 1(b)) in New Taipei city (population: 3.97 million as of Dec. 20154).The traffic emission sources mainly come from Zhongzheng Rd., Zhongshan Rd., and No. 1 National Freeway, which is 600 m N, 300 m S, and 1.5 km S from SJ station, respectively.The SJ station is also influenced by industrial emission from the New Taipei Industrial Park (3.5 km SW), Taishan Industrial Park (3.5 km S), Gong'er Industrial Park (6 km SE).The JD station is located at the rooftop of a four-floor building in Tatung primary school in Zudong (24°44'32"N, 121°05'31"E) (Fig. 1(c)), which is a relatively small town (population: about 96000 as of Dec. 2014) surrounded by mountains.The main sources of particulate matters could come from the No. 3 National Freeway and the local No. 68 expressway which are 6000 m NW and 200 m SW, respectively, away from the JD station.
PM 2.5 and UFPs mass concentration measurements were conducted from May 3 2011 to Oct. 31 2013.Each sample was collected for 24 hours and the number of samples was 55, 55 and 62 at ZS, SJ and ZD stations, respectively.The total number of samples taken range from 8-24 per season per station.The sampling strategy of this study is to use a MOUDI to determine UFP mass concentrations and mass distributions, a Dichotomous sampler to determine PM 2.5 concentrations, and chemical analysis is to determine chemical compositions of UFP and PM 2.5 .To characterize the effective density and dynamic shape factor of UFPs and submicron particles for the conversion of number concentrations into mass and surface area concentrations, a SMPS was collocated with the MOUDI for measuring UFPs number distributions at ZS station with the time resolution of 3 minute from Apr. 22 to 25 and May 14 to 18 in 2014.Details of the strategy and methods were documented in our earlier paper (Chen et al., 2010b) and are only briefly discussed here.
In the MOUDI, the 3.2 µm cutsize stage was replaced with 2.5 µm cutsize stage and the 10-th stage was not used.Thus, the cutsizes of the MOUDI stages were 18, 10, 5.6, 2.5, 1.8, 1.0, 0.56, 0.32, 0.18, and 0.1 µm, respectively, and the after filter measured UFP concentrations.In the MOUDI, silicone grease (KF-96-SP, Topco Technologies Corp., Taiwan) coated aluminum foils were used as the impaction substrates in 0-9 stages to reduce solid particle bounce.Teflon filter was used as the impaction substrate in the 10 th stage (Teflon R2PL047, Pall Corp., New York, USA) and after filter (Teflon R2PJ047, Pall Corp., New York, USA) for chemical analysis for soluble ions and elements.The samples collected were weighed to determine mass distributions.The mass distribution function based on particle size d ai was calculated by using the multi-modal size distribution model (Seinfeld and Pandis, 1998) as: where j is the number of mode under lognormal distribution, d g,i is the geometric mean diameter (GMD), σ g,i is the geometric standard deviation (GSD), and M i is the mass concentration.
The Dichot (Dichotomous PM 10 sampler, Model SA-241, Andersen Inc., Georgia, USA) was used to sample PM 2.5 and coarse particles (PM 2.5-10 ).Teflon filters (Teflo R2PL037, Pall Corp., New York, USA) were used in fine and coarse particle channels for analyzing soluble ions and elements.In this paper, only PM 2.5 data will be reported.
Before chemical analysis, all Teflon samples of MOUDI, Dichot and filter blanks were conditioned at least 24-h in a temperature and humidity controlled room (22 ± 1°C, 40 ± 5% RH) and then weighed by using a microbalance (Model CP2P-F, Sartorius, Germany).After that, each Teflon filter was cut in two equal halves using a Teflon coated scissor.One half was analyzed by an ICP-MS (Model 7500 series, Agilent Technologies, Inc., USA) for elements, including both crustal materials (CM) (Na, Mg, Al, K, Ca, Fe and Si) and trace elements (TE) (S, Zn, Ni, Cu, Mn, Sr, Ag, Ba, Pb, V, Cr and Ti).The other half was analyzed by an ion chromatograph (IC, Model DX-120, Dionex Corp, Sunnyvale, CA) for ionic species, which are F -, Cl -, NO 3 -, SO 4 -2 , NH 4 + , Na + , K + , Mg +2 and Ca +2 .CM can be calculated from elements by the following equations (Chow et al., 1994a;Marcazzan et al., 2001;Hueglin et al., 2005): To characterize the effective density and dynamic shape factor of UFPs and submicron particles, a SMPS (Model 3936, TSI Inc., MN, USA) equipped with Nano-DMA (TSI Model 3085) and Ultrafine Water-based Condensation Particle Counter (UWCPC, TSI Model 3786) was collocated with the MOUDI to monitor the number distributions of ambient particles from 5.6 to 209.1 nm.

Particle Effective Density (ρ eff ) and Dynamic Shape Factor (χ)
Most of previous studies focused on measuring the ρ eff of submicron particles with a selected diameter (McMurry et al., 2002;Park et al., 2003;Geller et al., 2006;Spencer et al., 2007).Only few studies investigated the size-resolved ρ eff of UFPs by using the SMPS-electrical low pressure impactor (ELPI) (Kannosto et al., 2008) or SMPS-MOUDI (Hu et al., 2012).The present study used the SMPS-MOUDI to conduct the field measurement in which ρ eff of UFPs in a certain size range was calculated based on the equation described in Hu et al. (2012) where C(d mi ) is the slip correction factor as the function of d mi ; C(d ai ) is the slip correction factor as the function of d ai ; ρ 0 is the unit particle density (1 g cm -3 ); λ is the mean free path air molecules.ρ eff,i of UFPs can be calculated by Eq. ( 4) based on the iteration method.Eq. ( 5) is only valid for spherical particles.For nonspherical particles, the dynamic shape factor (χ ) and the volume equivalent diameter (d ve,i ) should be used for converting the aerodynamic diameter to the mobility diameter (DeCarlo et al., 2004).The iteration method was used for solving χ by the following equations: where n'(d pi ) is the number concentration calculated based on an assume value of χ i ; C(d ve,i ) is the slip correction factor as the function of d ve,i .In the iteration, χ i was obtained when n'(d pi ) was equal to the measured number concentration in the size range i.The surface area concentration s'(d pi ) in the size range d pi was computed directly from the number concentration as s'(d pi ) = n'(d pi ) (/4 d pi 2 ).UFP surface area concentration was calculated as the sum of surface area concentrations in all size ranges below 100 nm.

Meteorological Conditions
Average trace gas concentrations and meteorological parameters at the three air monitoring stations during the present study are shown in Table S1 in Supplementary Materials.The differences in temperature between seasons are insignificant compared to mid-or high-latitude regions because Taiwan has a subtropical climate.The largest difference in temperature between summer and winter was only 14.7°C occurring at SJ station.The rainfall at three stations occurred frequently during the sampling period.In general, the concentrations of criteria pollutants except O 3 were lower in JD than the other two stations.

Quality Assurance and Quality Control (QA/QC)
Fig. 2 shows the comparison of mass concentrations measured by the MOUDI with those measured by the Dichot at the ZS, SJ, and JD stations.Good agreement was obtained between the MOUDI and the Dichot mass concentrations with the slope of 1.03 (R 2 = 0.94), 0.94 (R 2 = 0.94), and 1.0 (R 2 = 0.98) at ZS, SJ, and JD stations, respectively.This indicates that the quality of PM 2.5 measurement is good and the accuracy of UFP concentrations measured by the MOUDI is guaranteed.Fig. 3 shows the anion-cation balance of water soluble ions in terms of equivalent concentrations.

Seasonal Variation of PM 2.5 and UFPs Mass Concentration
The seasonal variations of the sampling data at each air monitoring station are shown in from December to February, respectively.At ZS station, the UFPs average mass concentrations in spring (2.2 ± 0.7 µg m -3 ) and summer (1.9 ± 0.7 µg m -3 ) are higher than those in autumn (1.2 ± 0.3 µg m -3 ) and winter (1.2 ± 0.5 µg m -3 ).
It could be attributed to the impact by heavy traffic emission from Xinsheng Highway and NPF due to photochemical nucleation, which depends on the intense solar radiation during warm period (Young et al., 2012).Similar results were also found at SJ and JD stations, where high average UFPs mass concentrations of 1.6 ± 0.5 and 1.2 ± 0.5 µg m -3 were measured in summer, respectively, followed by those in spring, autumn, and winter, as shown in Table 1.It is noted that UFPs mass concentrations measured in JD were lower than those at ZS and SJ stations because JD station is surrounded by mountains with less UFPs sources from traffic emission and anthropogenic activities.In addition to the primary particles emission, another possible source of UFPs at the JD station is the NPF produced by the oxidation of biogenic sources from the forest (O'Dowd et al., 2002) located in Jianshi Township (~15 km SE from JD station).For PM 2.5 , the average mass concentrations in spring, autumn, and winter at three stations were found to be higher than those in summer mainly due to the meteorological effect and partly due to the impact of PMs transported from China on PM 2.5 mass concentration in Taiwan (Lee et al., 2011).Similar result was also found by Shen et al. (2010), in which the seasonal highest PM 1 mass concentrations was found in winter due to emissions from coal and biomass burning, followed by spring, autumn, and summer.
In addition to local sources and long-range transported PMs, precipitation, wind speed, and wind direction also affect UFPs and PM 2.5 .For example, on Feb. 15 and Sep.13, 2012, the average mass concentration of PM 2.5 at ZS station decreased from 23.0 to 7.2 µg m -3 and 24.5 to 8.8 µg m -3 , due to heavy precipitation of 5.8 and 29.8 mm in rain fall, respectively.Similar results were also found in the urban area of Pakistan where the heavy monsoon rainfall during July and August results in lower PM 2.5 concentration in summer than that in winter (Rasheed et al., 2015).A decrease in UFPs from 1.8 to 0.4 µg m -3 was also observed on Feb. 15, 2012 due to washout effect (Morawska et al., 2004;2008).However, an opposite variation trend was found for PM 2.5 and UFPs during rain events of 16.8 and 38.4 mm in rain fall on May 14 and 15, 2011, respectively, in which PM 2.5 mass concentration decreased from 31.9 to 10.3 µg m -3 , while UFPs slightly increased from 0.9 to 1.1 µg m -3 .It is probably because higher saturation ratio of semivolatile species combined with pre-existing particles existed in reduced temperature and heavy precipitation to form new particles, resulting in increase in UFPs concentrations (Charron and Harrison, 2003).Another possible explanation is that the heavy rain event washed the pre-existing PM out and thus reduced the coagulation sink and favored the presence of UFPs.On Aug. 14 and 15 of 2012, the wind speed (3 m s -1 ) was larger than the average wind speed (1.8 m s -1 ), resulting in minimum PM 2.5 and UFPs mass concentrations during the sampling period due to high coagulation rate, good air mixing, particle deposition, scavenging (Morawska et al., 2008).
Table 2 shows 3-year average concentration ratios of water soluble ions (Ions), organic matters (OM), elementary carbon (EC), TE and CM concentrations to PM 2.5 and UFPs.The ion concentration accounts for 15.2 ± 1.6, 19.2 ± 1.5, and 21.0 ± 2.1% of UFPs, while the TE concentration is 3.2 ± 0.9, 3.2 ± 1.3, 4.5 ± 1.0% of UFP concentration at the ZS, SJ, and JD stations, respectively.The OM in UFPs was not measured in this study and might contribute significantly in UFPs.
The TE concentration is 4.3 ± 3.6, 3.0 ± 1.4, and 3.0 ± 1.8% of PM 2.5 concentration at the ZS, SJ, and JD stations, respectively.The ion concentration contributes 37.8 ± 4.1, 40.0 ± 3.0, and 46.4 ± 1.6% to PM 2.5 , which is similar to that measured at the roadside in Chen et al. (2010b) (Table 2).Ion concentration accounted for more fraction in PM 2.5 than that in UFPs because UFPs are mostly freshly generated primary or secondary PMs by vehicles, while PM 2.5 are aged ambient particles, which contain more condensed water to absorb inorganic salts (Chen et al., 2010a).It is also noted that PM 2.5 sampled at the JD station contains more soluble ions than those in other stations because of less traffic contribution to PM 2.5 at this station, suggesting that particles are aged through photochemical reaction and thus more secondary sulfates, nitrates, and other inorganic salts are formed (Lin et al., 2007;Chen et al., 2010a).
CM including Al, K, Fe, Ca, Mg, Ti, and Si contributed 9.0 ± 7.4, 8.5 ± 7.7, and 9.5 ± 8.3% to UFPs at ZS, SJ, and JD, respectively, and 8.3 ± 8.0, 8.4 ± 6.2, and 10.6 ± 8.8% to PM 2.5 at ZS, SJ, and JD, respectively.This suggests that no substantial difference in CM fraction in either UFPs or PM 2.5 and CM is also an important composition in UFPs.

PM 2.5 and UFPs Seasonal Variation of Mass Distribution
Figs. 5(a)-5(c) and Fig. S1 of Supplementary Materials show seasonal average mass distributions at ZS, SJ, and JD stations from 2011 to 2013.In the 3-year average mass distributions all stations show diurnal bimodal distributions except those at ZS station in spring 2011 (Fig. 5(b)) and SJ station in summer 2012 (Fig. 5(c)).In Fig. 5(a), the 3-year average accumulation and coarse mode mass median aerodynamic diameters (MMAD) were 0.40 and 5.64 µm at ZS station, 0.44 and 5.47 µm at SJ station, and 0.42 and 5.38 µm at JD station, respectively.No significant difference was found all stations.In spring 2011 at ZS station (Fig. 5(b)), the MMAD was measured to be 0.89 µm, which was larger than the average MMAD due to the impact of sandstorm event from Inner Mongolia Autonomous Region which also transported PM 2.5 from China.In summer 2012 at SJ station (Fig. 5(c)), the mass distribution shows a single coarse mode with a relatively low mass concentration compared to other seasons.Large MMAD of 6.34 µm could be attributed to the influence of sea salt transported by Typhoon Tembin.Na + and Cl -concentrations in PM 2.5 were both 0.35 µg m -3 in August 2012, which were higher than the average concentration of 0.19 µg m -3 in 2012, provided further evidence of sea salt transported by Typhoon Tembin.At ZS and SJ stations, the peak concentration of the accumulation mode is slightly higher than that of the coarse mode, while the coarse mode concentration is higher than that of the accumulation mode at JD station.Again, it is because ZS and SJ stations are more influenced by the UFPs emitted from traffic and industries than JD station.

Number Size Distributions of Particles with d p ≤ 209 nm
Fig. 6 shows calculated ρ eff,i and χ i based on the experimental data measured at ZS station.The average ρ eff,i at ZS station was 0.68 ± 0.16 g cm -3 for UFPs (Fig. 6(a)) and 1.06 ± 0.32 g cm -3 for PM 0.1-0.18(Fig. 6(b)), while χ i was 2.06 ± 0.19 and 1.45 ± 0.16 for UFPs (Fig. 6(a)) and PM 0.1-0.18(Fig. 6(b)), respectively.Calculated χ i at ZS were larger than 1.0, suggesting that the shape of particles was  irregular (DeCarlo et al., 2004).This conclusion could be proven by the TEM images of UFPs sampled by using a personal nanoparticle sampler (PENS) (Tsai et al., 2012) at ZS station, as shown in Fig. 7. Results of EDX (Energydispersive X-ray spectroscopy) analysis shown in Table S2 of Supplementary Materials indicated that C was the most abundant composition of all elements (74.16-90.56%by weight) in nanoparticles, followed by Cu (4.19-9.03%),O (1.15-3.67%),Al (0-12.89%),and Si (0.5-2.04%), suggesting that OM was the major constituent in UFPs.The chemical compositions of CE at ZS station were different from those measured by using EDX in India, where the Mg/Al, Si/Al, K/Al, Ca/Al, Mn/Al, and Fe/Al were 0.44 ± 0.22, 1.96 ± 0.90, 0.65 ± 0.22, 1.52 ± 0.40, 0.84 and 1.54 ± 1.67, respectively (Agnihotri et al., 2015).Fig. 8 shows daily average particle number concentrations at ZS (Fig. 8(a)), SJ (Fig. 8(b)), and JD stations (Fig. 8(c)).ρ eff,i and χ i obtained at ZS station were assumed applicable for calculating number concentrations at the other two stations because the difference in 3-year average concentration ratios of chemical species to PM 2.5 and UFPs (Table 2) and seasonal average mass distribution among three stations (Fig. 5) were not significant.In three year period, UFPs daily average number concentration was 1.7 ± 0.8 × 10 4 , 1.4 ± 0.6 × 10 4 , and 1.2 ± 0.6 × 10 4 # cm -3 , respectively, while PM 2.5 daily average number concentration was 1.8 ± 0.8 × 10 4 , 1.6 ± 0.6 × 10 4 , and 1.3 ± 0.6×10 4 # cm -3 at ZS, SJ, and JD stations, respectively.The UFP concentration determined at ZS station was very close to 1.7 ± 1.4 × 10 4 # cm -3 measured at National Taiwan University in 2012 by Cheung et al. (2013).The variation of UFP number concentrations with different seasons and stations is apparent in Fig. 8.It shows UFP concentration is more abundant in ZS than SJ station and the least abundant in JD station, reflecting the influence of different traffic emissions at these stations.The UFP concentration is elevated in summer and spring than that in winter and autumn in all stations.In comparison, the trend of PM 2.5 mass concentration is different with summer being the lowest and winter being the highest.This suggests that secondary aerosol formation in warm seasons is the major cause for the increase in UFP number concentrations.Fig. 9 shows the average ratio of UFPs/PM 2.5 for mass, surface area, and number concentration at ZS (Fig. 9  PM 2.5 in 3-year average of all stations, the UFPs/PM 2.5 ratios for surface area and number concentration were as high as 42.1 ± 12.8% and 89.0 ± 5.5%, respectively.The dependence of the ratios on different stations and seasons is also similar to that in Fig. 8 with ZS having the highest UFPs/PM 2.5 ratio.The 3-year average UFPs/PM 2.5 number concentration ratio was shown to be comparable to 81 ± 8% and 89.0 ± 5.5% during cold and warm periods, respectively, measured at urban sampling stations in central Taiwan (Young et al., 2012).This result reveals that UFPs contribute significantly to particle number but also to surface area in PM 2.5 in the urban air.Long term measurements for physical and chemical properties of UFPs in the atmosphere should be further conducted for better health risk assessment in the future.

CONCLUSION
This study investigated the long term daily average mass concentrations UFPs and PM 2.5 from 2011 spring to 2013 autumn at Zhongshan (ZS), Sinjhuang (SJ), and Jhudong (JD) air monitoring stations.Particle number distributions at ZS station in 2014 were further measured by using a Scanning Mobility Particle Sizer (Model 3936, TSI Incorporated, St. Paul, MN, USA, SMPS) to examine the effective density (ρ eff ) and dynamic shape factor (χ) of UFPs.The results showed that heavy traffic emission and new particle formation event resulted in higher UFPs mass concentrations in spring and summer than those in autumn and winter.Based on the calculated average ρ eff (0.68 ± 0.16 g cm -3 ) and χ (2.06 ± 0.19), the number and surface area concentration ratios of UFPs/PM 2.5 at urban air monitoring stations were found to average 89.0 ± 5.5% and 42.1 ± 12.8%, respectively, and varied by seasons and stations.This suggests that UFPs contribute significantly to the healthrelevant PM 2.5 fraction and warrant serious attention in the future.

Fig. 5 .
Fig. 5. Seasonal average mass distribution at ZS, SJ, and JD stations from 2011 spring to 2013 autumn.
(McMurry et al., 2002)ensity in the size range from d m1 to d m2 , i is 9, 10, or after filter representing the 9 th stage, 10 th stage, and after filter of the MOUDI, respectively; d m is the electrical mobility diameter; n(d pi ) is the number concentration of size d pi measured by the SMPS; MC i is the mass concentration measured by the MOUDI in the size range from d a1 to d a2 in which d a represents the aerodynamic diameter; VC i is the volume concentration calculated from the number distribution measured by the SMPS in the size range from d m1 to d m2 .The range of d a1 -d a2 was changed to d m1 -d m2 by using the following equations(McMurry et al., 2002):

Table 1 .
Seasonal average mass concentrations of PM 2.5 and UFPs at ZS, SJ, and JD stations.

Table 2 .
3-year average concentration ratios of chemical species to PM 2.5 and UFPs at ZS, SJ, and JD stations.