Seasonal variations of ultrafine and submicron aerosols in Taipei , Taiwan : implications for particle formation processes in a subtropical urban area

The aim of this study is to investigate the seasonal variations in the physicochemical properties of atmospheric ultra-fine particles (UFPs, d ≤ 100 nm) and submicron particles (PM1, d ≤ 1 μm) in an east Asian urban area, which are hypothesized to be affected by the interchange of summer and winter monsoons. An observation experiment was conducted at TARO (Taipei Aerosol and Radiation Observatory), an urban aerosol station in Taipei, Taiwan, from October 2012 to August 2013. The measurements included the mass concentration and chemical composition of UFPs and PM1, as well as the particle number concentration (PNC) and the particle number size distribution (PSD) with size range of 4–736 nm. The results indicated that the mass concentration of PM1 was elevated during cold seasons with a peak level of 18.5 μg m in spring, whereas the highest concentration of UFPs was measured in summertime with a mean of 1.64 μg m. Moreover, chemical analysis revealed that the UFPs and PM1 were characterized by distinct composition; UFPs were composed mostly of organics, whereas ammonium and sulfate were the major constituents of PM1. The seasonal median of total PNCs ranged from 13.9× 10 cm in autumn to 19.4× 10 cm in spring. Median concentrations for respective size distribution modes peaked in different seasons. The nucleationmode PNC (N4−25) peaked at 11.6× 10 3 cm in winter, whereas the Aitken-mode (N25−100) and accumulation-mode (N100−736) PNC exhibited summer maxima at 6.0× 10 3 and 3.1× 10 cm, respectively. The change in PSD during summertime was attributed to the enhancement in the photochemical production of condensable organic matter that, in turn, contributed to the growth of aerosol particles in the atmosphere. In addition, clear photochemical production of particles was observed, mostly in the summer season, which was characterized by average particle growth and formation rates of 4.0± 1.1 nm h and 1.4± 0.8 cm s, respectively. The prevalence of new particle formation (NPF) in summer was suggested as a result of seasonally enhanced photochemical oxidation of SO2 that contributed to the production of H2SO4, and a low level of PM10 (d ≤ 10 μm) that served as the condensation sink. Regarding the sources of aerosol particles, correlation analysis of the PNCs against NOx revealed that the local vehicular exhaust was the dominant contributor of the UFPs throughout the year. Conversely, the Asian pollution outbreaks had significant influence in the PNC of accumulation-mode particles during the seasons of winter monsoons. The results of this study implied the significance of secondary organic aerosols in the seasonal variations of UFPs and the influences of continental pollution outbreaks in the downwind areas of Asian outflows.


Introduction
Due to the significant impact of particulate matter on human health and climate change, it is vital to understand the formation process of atmospheric particles (Charlson et al., 1992;Donaldson et al., 1998).A number of mechanisms have been proposed by which atmospheric particles are formed, includ-H.C. Cheung et al.: Seasonal variations of ultra-fine and submicron aerosols in Taipei, Taiwan ing binary nucleation, ternary nucleation and ion-induced nucleation for charged particles, under different environmental conditions (Kulmala, 2003;Kulmala et al., 2004Kulmala et al., , 2012)).Numerous studies have been conducted in different locations to elucidate particle formation processes under various environmental settings in the free troposphere, boreal forest and coastal areas, where new particles formation processes are observed frequently (Kulmala et al., 2004, Holmes, 2007).Recently, investigations were also carried out on new particle formation within the urban boundary layer (e.g., Cheung et al., 2013, and references therein), where particle formation was suggested to be mainly influenced by the photooxidation of SO 2 .Furthermore, formation of particulate matter by heterogeneous reactions of gases on dust particles was reported recently (Hsu et al., 2014;Nie et al., 2014).Previous investigations have indicated that the air pollutants, in both gaseous and particulate form, associated with the continental outflows of air masses, could have affected a wide region in east Asia and caused severe regional air pollution (e.g., Lin et al., 2004;Wang et al., 2003;Buzorius et al., 2004).However, the formation processes of ultra-fine particles (UFPs, d ≤ 100 nm) and submicron particles (PM 1 , d ≤ 1 µm) under the influences of continental outflows are not yet well understood.
In urban environments, major contributing sources of aerosol particles include vehicular exhausts (e.g., Pey et al., 2008;Pérez et al., 2010), industrial emissions (Gao et al., 2009) and new particle formation by photochemical reactions (e.g., Pey et al., 2009).Approximately 55-69 % of the total particle number concentrations (PNCs) were attributed to secondary aerosols during midday in several European cities (Reche et al., 2011).In Taipei, Taiwan, a subtropical urban area, Cheung et al. (2013) observed that there was a tenfold increase in nucleation-mode particle number concentration (N 9−25 , with size 9 < d < 25 nm) during new particle formation events compared to that contributed by the vehicle emission.Besides the local sources, air quality of east Asian countries is also strongly affected by the transport of air pollutants from mainland China during periods of winter monsoons (Cheung et al., 2005;Lin et al., 2004;Matsumoto et al., 2003).Lin et al. (2004) reported that the mass concentration of particulate matter (PM 10 ) due to the long-range transport associated with winter monsoons was 85 µg m −3 , about 79 % higher than that due to local pollution (∼ 47.4 µg m −3 ) in Taipei.Chemical composition of fine and coarse particles was measured during a winter monsoon period at Rishiri Island, near the northern tip of Japan, to study the transport of continental aerosols (Matsumoto et al., 2003).The results showed that higher levels of particle mass concentration were associated with the outbreaks of continental polluted air masses.In addition, Cheung et al. (2005) found deterioration in visibility around southern China during wintertime as indicated by a twofold increase in aerosol light scattering coefficient under the influences of winter monsoons.Chen et al. (2013) conducted a measurement of the particle number concentration at the background station on the mountain of central Taiwan in summer 2009 and autumn 2010.The result showed that, on the contrary, particle number concentrations were dominated by local sources rather than long-range transport.To date, most of the relevant studies mentioned above were limited to measurements in terms of PM 10 or PM 2.5 for a particular period.The seasonal variations of particles in either ultra-fine or submicron range have not been well illustrated.
A 1-year aerosol characterization experiment was conducted in the urban area of Taipei, Taiwan.The aim of this study is to attain a better understanding of the seasonal variations of ultra-fine and submicron particles and the factors affecting particle formation, particularly under the influences of Asian monsoon circulations.In this study, we analyzed number concentration and size distribution of aerosol particles, together with the mass concentration and chemical composition of UFPs and PM 1 measured during four seasonal campaigns (i.e., 24 October-15 November 2012, 4-24 January, 17 March-11 April and1-14 August 2013).The results of this study will contribute to the management strategies of the severe air pollution over the east Asia region.

Observation site and instrumentation
The measurements were conducted at Taipei Aerosol and Radiation Observatory (TARO, 25.02 • N, 121.53 • E), located in the downtown area of Taipei, Taiwan, during October 2012 to August 2013.The measurements were carried out for 2-3 weeks in each season (see Table 1 for measurement details).The aerosol observatory is located on the top floor of Building B of the Department of Atmospheric Sciences, National Taiwan University (ASNTU), which is ∼ 20 m a.g.l (above ground level) (Cheung et al., 2013).
Particle number size distribution (PSD) in the range of 4-736 nm was measured by two scanning mobility particle sizer (SMPS) systems.One was equipped with a long differential mobility analyzer (long DMA, model: TSI 3081, TSI Inc.) and a condensation particle counter (CPC) (model: TSI 3022A, TSI Inc.) to measure the particles from 10 to 736 nm, which was named long SMPS.Another one was equipped with a nano-DMA (model: TSI 3085, TSI Inc.) and an ultrafine water-based CPC (UWCPC, model: TSI 3786, TSI Inc.) to measure the particles from 4 to 110 nm, which was named nano-SMPS.The polydisperse particles were classified into selected monodisperse particles according to their electrical equivalent mobility by the DMAs.The number concentration of the monodisperse particles was then counted by the CPCs.Ambient air was drawn into the SMPS systems from outside the building through a 0.635 cm (inner diameter) conductive tube, and a sampling duration of 5 min was adopted for each PSD measurement.The SMPS systems' flow rates (10 3 no.cm −3 ) (10 3 no.cm −3 ) (10 3 no.cm −3 ) (10 3 no.cm −3 ) (10 3 no.cm −3 ) Autumn 24 Oct-15 Nov 2012 13.9 (9.9-19.4)8.6 (5.8-11.8)3.9 (2.6-5.16.6 (9.2-26.7)6.9 (4.5-10.4)6.0 (2.5-11.3)13.7 (7.9-21.4)3.1 (0.5-5.1) 0.44 0.87 were checked weekly during the sampling period and the accuracy of the particle sizing of the DMAs was checked using polystyrene latex (PSL) spheres before the campaigns.Operation details are referred to Cheung et al. (2013).Size-segregated aerosol samples were collected by a pair of Micro-Orifice Uniform Deposition Impactors (MOUDI,model: 110,MSP Corp.).Taking the advantage that the cut diameter of the ninth MOUDI impaction stage was exactly 100 nm, the tenth impaction stage (with a cut diameter of 56 nm) of each MOUDI was removed to allow the postfilter function as a collector of UFPs (Marple et al., 1991).The sampling flow rate of MOUDI sampler was 30 L min −1 .In addition, a pair of PM 1 samplers, each consisting of a standard aerosol sampler (PQ-200, BGI Inc.) and a PM 1 sharp cut cyclone, were deployed to collect PM 1 samples with 16.7 L min −1 sampling flow rate.For both UFPs and PM 1 sampling arrangements, one of the paired samplers was equipped with Teflon filters, whereas another was equipped with quartz fiber filters.The Teflon filter samples were used for gravimetric measurements.The quartz filter samples were deployed for the analysis of soluble ions (Na + , NH + 4 , K + , Ca 2+ , Mg 2+ , Cl − , NO − 2 , NO − 3 , SO 2− 4 ) using ion chromatograph (IC), and carbonaceous components (i.e., organic carbon, OC, and elemental carbon, EC) in the aerosols using a DRI-2001A carbonaceous aerosol analyzer with IMPROVE-A protocol (Chow et al., 2007).Details of the in-lab analysis are as described previously (Salvador and Chou, 2014).Both the PM 1 and UFPs were collected with double-layered quartz filters (i.e., QBQ setup) and the artifacts due to adsorption of gaseous components were corrected as suggested by Subramanian et al. (2004).The sampling duration of each sample set (for both MOUDI and PQ-200 samplers) was from 14:00 to 12:00 LT (22 h), and a total of 69 and 75 sets of UFPs and PM 1 samples were collected during the entire investigation period (sample sets collected in autumn, winter, spring and summer were 20, 15, 25, and 9 sets for UFPs, and 21, 16, 25 and 13 sets for PM 1 , respectively).
Moreover, to assist the data interpretation, the hourly average mass concentration of PM 10 , the mixing ratio of trace gases (i.e., NO x , SO 2 and O 3 ) and the meteorology parameters (i.e., wind direction/speed and ultraviolet B (UVB, wavelength: 280-315 nm)) from the Guting air quality station of Taiwan Environmental Protection Agency, which is about 1 km from TARO, were analyzed in this study.The details of instrumentation setup for trace gas measurements are referred to Cheung et al. (2013).

Data processing and analysis
The PSD of 4-736 nm presented in this study was combined from two sets of SMPS data, where the nano-SMPS corresponded to the size range of 4-50 nm, and the long-SMPS corresponded to the size > 50 nm.The diffusion loss of the particles during the sample transport in the tubing was corrected according to the algorithm proposed by Holman (1972).Particle number concentrations for different size ranges were then calculated from the SMPS measurements.

Classification of new particle formation and calculation of the particle growth and formation rates
A new particle formation (NPF) event is defined as the increase of the number concentration of nucleation-mode particles, which are growing into Aitken-and/or accumulationmode size range (≥ 25 nm) and last for a few hours until they coagulate on the pre-existing aerosol and/or other surfaces in the atmosphere.The calculation of particle growth rate was represented by the rate of geometric median diameter changes during the period of nucleation-mode particles growing through 25 nm (Cheung et al., 2013).The formation rate (J ) of nucleation-mode particles for each NPF event was calculated for the particle size ranging from 4 to 25 nm according to the method of Dal Maso et al. (2005).The formation rate is defined as the sum of the apparent formation rate  (dN 4−25 /dt) and the coagulation loss rate during the NPF event.It should be noted that the reported apparent particle formation rate is expected to be smaller than the actual nucleation rate, since some fraction of formed nuclei are always scavenged by coagulation into larger pre-existing particles before they grow larger by condensation (Lehtinen et al., 2007).The work done by Kulmala et al. (2012) is referred to for an overview of the methodology on the measurement of the nucleation of atmospheric particles.

Back trajectory analysis
Backward trajectories were calculated using the HYSPLIT model (Hybrid Single Particle Lagrangian Integrated Trajectory, Version 4.9) of NOAA (National Oceanic and Atmospheric Administration) (Draxler, 1999) for TARO during the sampling period, in order to trace the origins of the air masses.Seventy-two-hour back trajectories were calculated twice per day at 00:00 and 12:00 LT with a height setting of 200 m a.g.l.It should be noted that the grid resolution of the meteorological data used for the calculation of back trajectories is 1 • × 1 • , which is not enough to trace the detailed air mass passage over the scale of the study region and, therefore, the trajectories only provide an indication of the region from which the air mass was originated.

Particle number concentration and size distribution in respective seasons
As mentioned above, the air quality of Taipei is significantly affected by both the local vehicular exhausts and long-range transport of air pollutants, of which the latter is dominated by meteorological factors.The information on the meteorological conditions, particularly the wind pattern, is important to elucidate our results and is thus presented here.The back trajectories of the air masses for TARO are illustrated in Fig. 1 (left panel).The results showed that northeasterly winds prevailed in autumn and winter seasons, passing through the Asian continent before reaching Taiwan, whereas southerly winds prevailed in summertime.The air masses observed in the spring period were found to mainly be associated with Asian continental outflows and occasionally associated with the southerly flows.This observation agreed with the surface wind direction measured in the Taipei area (see Fig. 1, right panel), where northeasterly winds dominated during the period from November 2012 to May 2013, and southerly winds prevailed from May to August 2013.The particle number concentrations in various size ranges during each season are summarized in Table 1.Relatively higher total PNCs (N 4−736 ) were observed in spring and winter with median values of 19.4 × 10 3 and 17.4 × 10 3 cm −3 , respectively, followed by that of summer (16.6 × 10 3 cm −3 ) and autumn (13.9 × 10 3 cm −3 ).This result is comparable to the previous measurements conducted in urban Taipei where the seasonal means of PNCs (10 < d < 560 nm) ranged from 11.0 × 10 3 to 17.0 × 10 3 cm −3 (Cheng et al., 2014).Figure 2 illustrates the number, surface and volume size distributions of the aerosol particles.The geometric mean diameter (GMD) of each PSD mode was retrieved from the number concentration data.The GMDs of the nucleation, Aitken and accumulation modes were found to be 10.4-12.8, 26.5-38.4and 91.8-159.0nm, respectively.Details of particle number concentration and GMDs of each fitted mode in four seasons are listed in Table S1 in the Supplement.
In addition, the fitted GMDs of surface distribution were found to be 77.4 and 293 nm for autumn, 22.1, 68.9 and 228 nm for winter, 77.4 and 253 nm for spring and 12.9 and 268 nm for summer, respectively (not shown in the figures).In winter and summer seasons, one of the fitted surface GMDs was located at nucleation mode, showing the significant contribution of nucleation-mode particles in these two seasons.Bimodal volume distribution was obtained for all seasons where the fitted volume GMDs were 96.3 and 372 nm for autumn, 71.8 and 275 nm for winter, 99.5 and 339 nm for spring and 99.5 and 237 nm for summer, respectively.The GMD of first volume mode was relatively stable in each season (i.e., 71.8-99.5 nm), but a smaller GMD (237 nm) for the second volume mode was observed in summer.The results implied that a higher fraction of particles could have evolved from smaller size range (i.e., nucleation and Aitken modes) into accumulation mode, which coincided with our observation that NPF events occurred mostly in summer (see Sect. 3.4).Furthermore, this seasonal variability agrees with our previous findings that the growth rate of newly formed particles was correlated with the photolysis of ozone, an indicator of photochemical activity (Cheung et al., 2013).The causes responsible for the observed seasonal variations in PNCs will be detailed in the following sections.This was different from that observed in urban Beijing where a relatively larger GMD was observed in accumulation mode due to the enhancement of condensation by higher photochemical activities in summer, but without significant seasonal variations in Aitken-mode distribution (Wu et al., 2008).
It was revealed that the nucleation-mode particles were predominant in the PNCs during autumn, winter and spring in the study area, whereas a distinct size distribution pattern was observed in summertime.In summer, the fraction of nucleation (N 4−25 /N 4−736 ) decreased to 0.44 (see Table 1) and the Aitken-mode PNCs increased to be comparable to that of the nucleation mode, whereas the N 4−25 /N 4−736 ratios for other seasons ranged from 0.56 to 0.77 (see Table 1).An observation from another aspect is that the PNC of nucleation mode (N 4−25 ) peaked in winter and reached the minimum in summer; whereas the PNCs of Aitken mode (N 25−100 ) and accumulation mode (N 100−736 ) reached their maxima in summertime.The changes in the size distribution in summer season were most likely due to the seasonally enhanced pho-tochemical production of condensable vapors that, in turn, contributed to the growth of aerosol particles in the atmosphere.S2 in the Supplement.

Mass concentration and chemical composition
The seasonal means of UFPs ranged from 0.73 to 1.64 µg m −3 , with an annual average of 1.01 µg m −3 .The measured mass concentration of UFPs of the present study was comparable to that in the urban area of Los Angeles, United States (0.80-1.58 µg m −3 , Hughes et al., 1998), and relatively higher than that in urban Helsinki, Finland (average 0.49 µg m −3 , Pakkanen et al., 2001).For the chemical composition, OC was found to be the major mass contributor, which accounted for 29.8 % (seasonal means ranging from 26.9 to 33.4 % for various seasons) of average mass concentration of UFPs.The EC was the second major component with an average mass contribution of 5.1 % (seasonal means 2.4-7.6 %), followed by sulfate (SO 2− 4 ) at 4.3 % (seasonal means 3.4-6.4%) and nitrite (NO − 2 ) at 2.9 % (seasonal means 0.9-7.3%).In addition, a large fraction of mass was contributed by the group of "others", which consisted of mineral (K + , Ca 2+ , PO 3− 4 and Mg 2+ ), sea salt (Na + and Cl − ) and unidentified species.The results showed that, on average, mineral and sea salt components attributed only 3.5 % (seasonal means 2.0-6.0 %) to UFPs mass concentration.Thus a substantial amount of UFPs remained unidentified, which likely included hydrogen and oxygen associated with OC.The conversion factors used to estimate the average molecular weight per carbon in particulate organic matter varied depending on the characteristic of aerosols.A lower factor value, 1.2, was usually suggested for saturated organic molecules, while a higher value, 1.6, was adopted for watersoluble compounds consisting of multifunctional oxygenated groups; and even higher factor values were suggested for aged aerosols which contained a higher portion of low and semi-volatile products of photochemical reactions (Turpin and Lim et al., 2001).The high unidentified mass fraction implied that the photochemical production of secondary organic aerosols was a significant process responsible for the elevated levels of UFPs observed in this study.
The results presented above indicated that UFPs exhibited a distinct seasonal variability and composition from PM 1 in the study area.The highest concentration of UFPs was observed in summer (1.64 µg m −3 ) and the lowest in winter (0.73 µg m −3 ).This result may be attributed to the stronger photochemical activities in summer, which could have enhanced the formation of secondary organic aerosols.Consequently, the mass concentration of OC increased from 0.20 µg m −3 in winter to 0.47 µg m −3 in summertime.It is noteworthy that the mass concentration of sulfate in UFPs also peaked in summer (64 ng m −3 ), suggesting enhancement in the photo-oxidation of SO 2 .Cheung et al. (2013) found that photo-oxidation of SO 2 was the major mechanism for the formation of new particles in Taipei, Taiwan, and the production of condensable vapors was also dominated by photo-oxidation.The covariations in sulfate and OC revealed in this study further suggested that secondary organic compounds were the major condensable matter contributing to the growth of newly formed particles.
While the organics predominated in the mass concentration of UFPs, which included nucleation-mode and Aitken-mode particles, the measurements of PM 1 in this study suggested that sulfate was the major constituent of accumulation-mode aerosols.In contrast to the seasonal variation of UFPs, the mass concentration of PM 1 reached a maximum at 18.5 µg m −3 in spring and exhibited a mini-mum at 11.6 µg m −3 in summer.The PM 1 differences between spring and summer were mostly due to declined ambient levels of sulfate, nitrate and ammonium ions.As a result, the mass contribution of the three inorganic ions in PM 1 reduced from 55.7 to 46.2 % and, conversely, the mass fraction of OC increased from 10.2 to 14.3 %.The seasonal characteristics of PM 1 concentration and composition were mostly attributed to the changes in the origin areas of background air mass, which shifted from the Asian continent to the western Pacific Ocean during summertime (see Fig. 1).Our previous studies reported that the fine particulate matter (PM 2.5 ) transported on the Asian outflows to northern Taiwan maximized in springtime and were enriched in sulfate, nitrate and ammonium (Chou et al., 2008(Chou et al., , 2010)).The seasonal variability of PM 1 found in this study was consistent with the previous observations for PM 2.5 and thereby suggested the significance of Asian outflow aerosols to the PM 1 budget in the downwind areas of the Asian continent.

Seasonal characteristics of photochemical production
In order to study the influences of photochemical production of particles, the measurements of PNC and PSD were analyzed per daytime (07:00-17:00 LT) and nighttime (17:00-07:00 LT), respectively (see Fig. 4).In urban environments, the possible sources influencing the PNC and PSD are complicated, which include not only the direct emission from primary sources but also the interaction between the newly formed particles, pre-existing particles and condensing vapors through condensation and coagulation processes.Nevertheless, these processes occurred throughout the day and will not dominate in the differences between daytime and nighttime PNCs as observed in this study.It was assumed that the photochemical reaction was the major attributing factor to the observed diurnal differences in PNC.Since the particles in nighttime were mainly emitted from the vehicular exhausts and the elevated PNCs in daytime were due to both the primary and secondary sources of the particles in the study area (Cheung et al., 2013), a larger difference between the PNCs observed in daytime and nighttime indicated stronger influences of photochemical production on the PNCs.The most striking seasonal feature shown in Fig. 4 is the large difference between daytime and nighttime PSD in summer as indicated by the low N 4−736 (nighttime)/N 4−736 (daytime) ratio; whereas higher ratios were observed in other seasons.
In addition, the diurnal variation of particle size distribution (see Fig. 5) provided further information about the variations in PSD.Two nucleation bursts were distinctly observed in morning and afternoon traffic peak hours in autumn, winter and spring, while a typical PSD pattern of nucleation events (an increase of nucleation-mode particle concentrations with subsequent growth in particle size) was dominant in summer.This result is as expected because the photochemical production of nucleation-mode particles is more intense during warm seasons (Cheung et al., 2011).Moreover, as discussed in the previous section, the photochemical reactions could produce condensable organics, which allows the newly formed nucleation-mode particles to grow into the Aitken mode.The relatively small differences between the daytime and nighttime N 4−736 in autumn and winter indicated that the photochemical contribution in PNCs had declined as compared to that in summertime.Nevertheless, the contribution of vehicle emission was also significant, especially during colder seasons and when photochemical reactions were less intense.This will be discussed in detail in Sect.3.5.

Factors affecting new particle formation (NPF)
As shown in a previous study, the NPF events were frequently observed in summer, which subsequently induced a notable increase in N 4−25 in Taipei (Cheung et al., 2013).The frequency of NPF events was found to be 10 out of 84 measurement days and the events were observed in autumn (1 out 23 days), spring (3 out of 26 days) and summer (6 out of 14 days) seasons.the NPF events, a nonlinear relationship between these two parameters was usually observed during the daytime (Cheung et al., 2013).The results showed that clear NPF events were observed often in summer and occasionally in spring, but rarely in autumn and winter in the study area.The average particle growth and formation rates were found to be 4.0 ± 1.1 nm h −1 and 1.4 ± 0.8 cm −3 s −1 , which are comparable to those measured in other urban studies in Asian cities such as Hong Kong (average 6.7 nm h −1 , Wang et al., 2014) and Beijing (average 5.2 nm h −1 , Wang et al., 2013), and also within the range observed in other nucleation studies across the globe (∼ 1-20 nm h −1 , Kulmala et al., 2004).The particle growth and formation rates of each case are listed in Table 2. Table 3 summarizes the averages of N 4−25 , PM 10 , the H 2 SO 4 proxy (as UVB • SO 2 / condensation sink) and wind speed for each season.Petäjä et al. (2009) calculated the H 2 SO 4 proxy with a pre-factor value, k, and used it to estimate the actual sulfuric acid concentration.The estimation of a site-specific k value requires an actual measurement of H 2 SO 4 , which is not available in this study area.The proxy value calculated in this study was therefore only used as an indicator of particle production strength contributed by H 2 SO 4 .The dominating factors associated with the frequent particle formation in summertime were the low PM 10 concentration (35.6 µg m −3 ) and high H 2 SO 4 proxy (493.1 ppb W m −2 s).The association of sulfuric acid pro- duction and the NPF events agreed with the elevated mass concentration of sulfate in UFPs during summertime (shown in Table S2), as well as the results of previous urban studies (Woo et al., 2001;Cheung et al., 2013).This strongly suggested that the new particle formation was mainly driven by the photochemical oxidation of SO 2 under low condensation sink conditions (Gao et al., 2009;Nie et al., 2014), where the SO 2 could be transported from the upwind area on the summer monsoons (see Fig. 1d).Contrarily, the absence of particle formation events in wintertime could be attributed to the declined photochemical production of H 2 SO 4 as well as the suppression of NPF by particles transported from the Asian continent (Lin et al., 2004).The results of this work evidenced that low PM 10 concentration and high sulfuric acid production favored the particle formation process in urban areas.Nevertheless, it should be noted that condensing vapors other than sulfuric acid, for example volatile organic compounds, could also contribute to the observed particle formation, which requires further investigation.
The scatter plot between UVB • SO 2 and condensation sink is depicted in Fig. 7. Relatively higher UVB • SO 2 values were obtained during NPF events.Notably, there was a group of data with high UVB • SO 2 / CS but low UVB • SO 2 where no NPF event was observed.This implied that there could be a threshold level of UVB • SO 2 for NPF in the study region.However, some exceptions existed in the data set and suggested that the parameters driving NPF have not been well accounted for and need to be studied further.It was also noticed that an Asian outflow event occurred on 7 April 2013 during which an atypical NPF was observed (labeled with a  black dot in Fig. 7).This could be relevant to the secondary particle formation on dust surface under the influence of the long-range transport of air mass.This will be discussed in further detail in Sect.3.6.

Influences of local emission on PNCs
Vehicle emission is known to be the major source of particulate matter in urban environment, particularly during nighttime.In order to investigate the relationship between the vehicular exhausts and PNCs, the scatter plots of NO x (as an indicator of vehicle emission) against N 4−25 , N 25−100 and N 100−736 during nighttime were examined for winter and summer periods (see Fig. 8).The values of the Pearson correlation coefficient (r) and the slope of linear regression between NO x and PNCs are summarized in Table 4.
The highest r values were found in both the plots of NO x against N 25−100 for winter (r = 0.88) and summer (r = 0.87).This result suggested a strong linear correlation between the vehicle emission and the N 25−100 which coincided with the results from previous studies (e.g., Morawska et al., 2008).During wintertime, stronger correlation was found between NO x and N 4−25 (r = 0.84) and N 25−100 (r = 0.88) compared to that between NO x and N 100−736 (r = 0.38).In contrast, high r values were obtained between NO x and all particle modes in summer (r = 0.70-0.87).The robust correlation of NO x and N 4−25 , as well as NO x and N 25−100 , suggested that local vehicle emission was the predominant source of UFPs throughout a year.These results coincided with previous studies on the size distribution of vehicle exhaust particles, which were found to be 20-130 and 20-60 nm, respectively, for diesel and petrol engine vehicles (Harris and Maricq, 2001;Ristovski et al., 2006).However, the PNCs of accumulation-mode particles (N 100−736 ) in winter were dominated by NO x -independent sources, which were most likely related to the pollution outbreaks from the Asian continent.Lin et al. (2004) indicated that the longrange transported air mass was characterized by a high level of PM 10 and a low mixing ratio of NO x due to its short atmospheric lifetime.Interestingly, a moderate correlation between the PNC of accumulation-mode particles (N 100−736 ) and NO x was also observed in summer.Given that the Asian outflow ceased in summertime, this correlation evidenced a substantial contribution of local sources, particularly vehicular emissions, to the PNC of accumulation-mode particles in Taipei, Taiwan.
The slope values can serve as a relative emission factor of particles per NO x , which indicates the degree of influence of vehicle emission on the PNCs (Cheung et al., 2013).The corresponding slope values for N 4−25 , N 25−100 and N 100−736 , were found to be 279, 163 and 18 cm −3 ppb −1 in winter, and 239, 330 and 155 cm −3 ppb −1 in summer.A larger sum of slope values (724 vs. 460 cm −3 ppb −1 ) was found in summertime compared to the winter period, evidencing a greater influence of vehicle emission on particle number concentration.The seasonal effects on the emission ratio of PNCs and NO x are rather difficult to address due to the complexity of different controlling factors, such as formation mechanisms and meteorological conditions.For example, Nam et al. ( 2010) reported a negative exponential correlation between the PM / NO x ratio in vehicle emission and ambient temperature, and suggested that the impact of ambient temperature on particulate matter was larger than that on NO x .Nevertheless, the observed differences in the PNCs / NO x ratios for winter and summer periods of this study necessitate further investigations on the formation mechanisms of aerosol particles in urban areas, in particular the nucleation and the Aitken modes.

Influence of long-range transport
During the seasons of winter monsoons, i.e., from autumn to spring, continental outflows have been frequently observed in Taipei, which is indicated by the stable northeasterly wind and increase of O 3 level (Lin et al., 2004).Previous studies of long-range transport of air pollutants on air quality of northern Taiwan showed that an elevated PM 10 was observed under the influence of continental outflows (Lin et al., 2004;Chou et al., 2004).Figure 9 S3.
As shown in Fig. 9, the diurnal variations of PSD during the LRT event exhibited two N 4−25 peaks associated with the morning and afternoon traffic rush hours; whereas the PNCs of the Aitken-mode particles remained at a low level.The results suggested that the influences of local vehicle emission on PNCs were still in place; whereas growth of particles due to secondary production of condensable vapors could have been suppressed, as NPF was rarely observed during the LRT events.It is noteworthy that a weak dust transport event was observed on 7 April 2013 where a nucleation event was observed in the PSD, evidencing that secondary formation of particles could have had occurred.However, the dominating diameter of particles was ∼ 40-50 nm at the initial stage of the event.The nucleation event lasted from ∼ 06:00 until 21:00 LT, when the northeasterly wind prevailed.The PM 10 and O 3 increased from minima of 44 µg m −3 (at 06:00 LT) and 25 ppb (at 05:00 LT) to the daily maxima of 92 µg m −3 (at 17:00 LT) and 61 ppb (at 16:00 LT).This result showed that the NPF process could have occurred in the upwind area where newly formed particles were transported to the study site, or heterogeneously formed particles were released from the dust surface during the long-range transport of air pollutants (Nie et al., 2014).
The average PSDs for LRT and non-LRT cases are shown in Fig. 10.The GMDs of the nucleation, Aitken, and accumu-  lation modes in PSD were found to be 10.4,37.2 and 158 nm for LRT and 11.4,30.4 and 114 nm for non-LRT cases, respectively (see Table S1 for detailed fit results of the PSD for LRT and non-LRT cases).The PNCs of different modes observed in non-LRT events were 8.6 × 10 3 cm −3 (nucleation mode), 9.3 × 10 3 cm −3 (Aitken mode) and 2.6 × 10 3 cm −3 (accumulation mode).The PNCs of LRT events were 9.2 × 10 3 cm −3 (nucleation mode), 4.0 × 10 3 cm −3 (Aitken mode) and 1.3 × 10 3 cm −3 (accumulation mode), respectively.The nucleation-mode PNC observed in non-LRT was comparable with that in LRT events, whereas significantly higher PNCs for the Aitken mode and accumulation mode were observed during non-LRT periods.This was attributed to the lower average wind speed (and hence poor dispersion) during non-LRT events (1.5 ± 0.8 m s −1 ) than that for LRT events (3.0 ± 0.8 m s −1 ).In contrast to the increase in PM 10 usually observed during LRT episodes (e.g., Lin et al., 2012), the relatively lower PNCs suggested that the number concentration of submicron particles, in particular UFPs, was dominated by local emissions.This agreed with the observation of seasonal mass concentration of UFPs that peaked in summertime when Taiwan was isolated from the influences of continental air mass.

Conclusions
The mass concentration and chemical composition of ultrafine particles (UFPs) and submicron particles (i.e., PM 1 ) as well as the particle number concentration (PNC) and size distributions (PSD) with size ranging from 4 to 736 nm were measured during four seasonal campaigns in the period from October 2012 to August 2013 at TARO, a subtropical urban aerosol station in Taipei, Taiwan.Distinct seasonal variability and chemical composition of UFPs and PM 1 were revealed.The UFPs were mostly composed of organic matter and reached maxima in summer, whereas the PM 1 composition was dominated by ammonium and sulfate, and exhibited a seasonal peak in spring.
It was found that the total PNC was significantly elevated during cold seasons, which was mostly caused by the high level of nucleation-mode particles (N 4−25 ).Conversely, both the Aitken-mode (N 25−100 ) and accumulation-mode (N 100−736 ) PNCs reached their respective maxima in summertime.Consistent correlation without significant seasonal variations was found between the UFPs (i.e., nucleation-and Aitken-mode particles) and NO x , suggesting that local vehicle emission was the major source of UFPs in the study area The declined correlation between NO x and N 100−736 in winter was likely due to the influences of air pollution associated with the Asian outflows.
The elevated level of UFPs in summer was attributed to the increase in the concentration of Aitken-mode particles (N 25−100 ).It was revealed from the measurements of PSD that a large number of nucleation-mode particles could have evolved into the Aitken mode during summertime, which was most likely relevant to the photochemical production of condensable vapors that, in turn, could have contributed to the growth of particles in the atmosphere.Moreover, the chemical measurements suggested that the constituents of the condensed materials in UFPs were mostly organic matter, implying the significance of secondary organic aerosols in the ambient UFPs.
A total of 10 new particle formation (NPF) events occurred out of 84 measurement days in this study, which were observed in autumn (1 out 23 days), spring (3 out of 26 days) and summer (6 out of 14 days) seasons.The prevalence of NPF in summer agreed with the highest H 2 SO 4 proxy and lowest PM 10 observed in this study, which provided favorable atmospheric conditions for new particle formation.The average particle growth and formation rates for the NPF events were 4.0 ± 1.1 nm h −1 and 1.4 ± 0.8 cm −3 s −1 , respectively, which were comparable to those measured in previous urban studies.
As exemplified above, the characteristics of various physicochemical properties of particles investigated in this study, and the occurrence of NPF, exhibited a strong seasonal variability, which was co-influenced by long-range transported particles during the seasons of winter monsoons and the strong photochemical activities in summer.The results of this study are critical for the authorities involved in urban development and health impact assessment, and the environmental policy makers who are tackling the severe atmospheric pollution in the east Asia region.
The Supplement related to this article is available online at doi:10.5194/acp-16-1317-2016-supplement.

Figure
Figure 3a and b illustrate the average chemical composition and mass concentration of UFPs and PM 1 , respectively, for each season.Details of the mass concentration and chemical composition of UFPs and PM 1 are listed in TableS2in the Supplement.The seasonal means of UFPs ranged from 0.73 to 1.64 µg m −3 , with an annual average of 1.01 µg m −3 .The measured mass concentration of UFPs of the present study was comparable to that in the urban area of Los Angeles, United States (0.80-1.58 µg m −3 ,Hughes et al., 1998), and relatively higher than that in urban Helsinki, Finland (average 0.49 µg m−3 , Pakkanen et al., 2001).For the chemical composition, OC was found to be the major mass contributor, which accounted for 29.8 % (seasonal means ranging from 26.9 to 33.4 % for various seasons) of average mass concentration of UFPs.The EC was the second major component with an average mass contribution of 5.1 % (seasonal means 2.4-7.6 %), followed by sulfate (SO 2− 4 ) at 4.3 % (seasonal means 3.4-6.4%) and nitrite (NO − 2 ) at 2.9 % (seasonal means 0.9-7.3%).In addition, a large fraction of mass was contributed by the group of "others", which consisted of mineral (K + , Ca 2+ , PO 3− 4 and Mg 2+ ), sea salt (Na + and Cl − ) and unidentified species.The results showed that, on average, mineral and sea salt components attributed only 3.5 % (seasonal means 2.0-6.0 %) to UFPs mass concentration.Thus a substantial amount of UFPs remained unidentified, which likely included hydrogen and oxygen associated with OC.The conversion factors used to estimate the average molecular weight per carbon in particulate organic matter varied depending on the characteristic of aerosols.A lower factor value, 1.2, was usually suggested for saturated organic molecules, while a higher value, 1.6, was adopted for watersoluble compounds consisting of multifunctional oxygenated groups; and even higher factor values were suggested for aged aerosols which contained a higher portion of low and semi-volatile products of photochemical reactions(Turpin and Lim et al., 2001).The high unidentified mass fraction implied that the photochemical production of secondary organic aerosols was a significant process responsible for the elevated levels of UFPs observed in this study.As shown in Fig.3b, average PM 1 was estimated to be 14.7 µg m −3 (seasonal means 11.6-18.5 µg m −3 ) in this study, which is similar to the results of a previous study in urban Taipei (average 14.0 µg m −3 , Li et al., 2010).The measured PM 1 level is relatively higher than that of the urban areas of Phoenix, United States (5.9 µg m −3 ,Lundgren  et al., 1996), and Helsinki, Finland (6.1 µg m −3 , Vallius et  al., 2000).For chemical composition, sulfate was the major mass contributor of PM 1 (average 39.0 %, seasonal means

Figure 2 .
Figure 2. Size distribution of particle number (upper panel), and surface and volume (lower panel) concentrations measured in (a) autumn, (b) winter, (c) spring and (d) summer (by curve fitting).

Figure 3 .
Figure 3. Seasonal average concentration and composition of (a) ultra-fine (UFPs) and (b) submicron (PM 1 ) particles observed at TARO in Taipei, Taiwan, from autumn 2012 to summer 2013.

Figure 5 .
Figure 5.Diurnal variation of particle number size distribution in each season.From the lower panel to the top panel: autumn, winter, spring and summer.

Figure 6 .
Figure 6.Scatter plots between hourly N 4−25 and NO x observed in (a) autumn, (b) winter, (c) spring and (d) summer at the TARO site during the period of 07:00-17:00 LT.

Figure 8 .
Figure 8. Scatter plots for hourly average PNCs vs. NO x measured during the time period of 20:00-04:00 (LT) in (a) winter and (b) summer, with classification of various particle size ranges.
depicts a longrange transport (LRT) pollution event observed at TARO during this study.The wind direction changed from westerly/northwesterly to northeasterly at 21:00, 24 March which continued until 06:00, 26 March.During this period, the O 3 mixing ratio remained at a moderate level (∼ 30-55 ppb) and PM 10 increased from 10.0 to 98.0 µg m −3 .It should be noted that the variations of measured pollutants were not solely influenced by long-range transport, but also partly by the variation of local pollution and boundary dynamics.In this section, we attempt to analyze the PSD/PNC under the influences of continental pollution outbreaks.The periods of the respective LRT events are listed in Table

Figure 10 .
Figure 10.Averaged PSDs for LRT and non-LRT episodes measured during the seasons of winter monsoons.Dashed lines illustrate the PSD of each individual mode.

Table 1 .
Median and quartile ranges (Q1-Q3) of the PNCs measured in each season.The size ranges of the PNCs are represented by the subscripted number.For example, N 4−25 represents the number concentrations of the particles from 4 to 25 nm.The fractions of N 4−25 and N 4−100 of total PNCs are presented in the last two columns.

Table 2 .
Time periods defined as the new particle formation events and the particle growth and formation rates.

Table 3 .
Average of N 4−25 , PM 10 , UVB, SO 2 , condensation sink (CS), H 2 SO 4 proxy and wind speed of different seasons.Standard deviation values are shown in parentheses.(Note: the data of the rainfall observations were not used in the calculations.) 1 ) Local vehicle emission also dominated the accumulation-mode PNC in summer, but not in wintertime.