Black Carbon , and Particle Size-Resolved Number Concentrations and the Ångström Exponent Value of Aerosols during the Firework Display at the Lantern Festival in Southern Taiwan

The Yanshui Beehive Firework Festival is a traditional folk activity in Southern Taiwan held during the Lantern Festival, and it is the third largest folk celebration in the world. During this festival, more than 200 firecracker towers with hundreds of thousands of firecrackers are ignited, posing a risk to public health because of an abrupt increase in particulate matter concentrations within a short period. In this study, real-time variations of PM2.5 (particles with an aerodynamic diameter less than or equal to 2.5 μm), black carbon (BC), and particle number concentrations were monitored before and during the firework display to understand the effect of the firework display on the short-term air quality. The hourly average concentrations of PM2.5, BC, total particle number, and ultrafine particle number during the firework display (episode period) were 146.9 μg m, 2639 ng m, 3.37 × 10 # cm, and 1.18 × 10 # cm, respectively. These values were 6.9, 2.3, 5.9, and 3.7 times greater than those during the same period on reference days (nonepisode period), respectively. The measured ultraviolet BC (UVBC) and BC concentrations indicated that aerosols were bound with ultraviolet-absorbing organic compounds, which were abundant, during the episode period. BC aerosols during the episode originated from vehicular traffic and firecracker burning, and the absorption Ångström exponent value was 1.4. The particle number size distribution during the episode period showed a major accumulation mode and a minor Aitken mode of 180 and 63 nm, respectively. This particle number size distribution pattern was considerably different from that in the nonepisode period. During the episode period, particle coagulation played a crucial role in removing particles in the nucleation and Aitken modes in the ambient air at high particle number concentrations.


INTRODUCTION
Epidemiological studies have consistently reported that short-term exposure to high concentrations of particulate matter (PM) can increase cardiovascular morbidity and mortality (Franchini and Mannucci, 2007;Pope et al., 2011;Rice et al., 2013).Dominici et al. (2006) demonstrated that every 10 µg m -3 increase in PM 2.5 (particles with an aerodynamic diameter less than or equal to 2.5 µm) could increase the risk of congestive heart failure by 1.3%.Pope et al. (2006) also showed that every 10 µg m -3 rise in PM 2.5 could increase the risk of acute coronary syndrome by 4.5%.Therefore, short-term air pollution during unusual episodes, such as dust storms, firework displays, and biomass burning, is a major public health hazard.
Pyrotechnic displays at celebration events often considerably degrade the short-term air quality (Vecchi et al., 2008;Camilleri and Vella, 2010;Nasir and Brahmaiah, 2015).The burning of fireworks generates a large amount of gaseous pollutants and suspended particles, and the impact of fireworks on visibility and human health is particularly evident at pyrotechnic events (Huang et al., 2012;Nasir and Brahmaiah, 2015).On the basis of a firework episode in Milan, Italy, Vecchi et al. (2008) noted that pyrotechnic displays could increase the concentrations of particles in the size range 0.5 and 1.0 µm by up to 6.7 times in 1 h.Zhang et al. (2010) found that the particle concentration during the peak hour of the firework display during a Chinese New Year firework event in Shanghai, China, was approximately three times greater than that on the preceding day.
The Yanshui Beehive Firework Festival is a traditional folk activity in Southern Taiwan held during the Lantern Festival, and it is one of the most popular religious events in Taiwan.Furthermore, it is the third largest folk celebration in the world.During the firework display, more than 200 firecracker towers with hundreds of thousands of firecrackers are ignited simultaneously.The ignition produces a cacophony resembling that of hundreds of thousands of bees streaming out of their hives.The event attracts a large number of visitors every year.Nevertheless, it poses a risk to public health because of an abrupt increase in PM concentrations.Because the firework display is performed close to the ground, its effect on air quality is more severe than aerial pyrotechnic displays.Chang et al. (2011) showed that the PM 10 (particles with an aerodynamic diameter less than or equal to 10 µm) and PM 2.5 concentrations during the beehive firework display can reach approximately 430 and 250 µg m -3 , respectively, which are 10 times the normal values.Furthermore, Lin et al. (2014) demonstrated that the PM 10 and PM 2.5 concentrations at a downwind location relative to the location of a beehive firework display were 570 and 430 µg m -3 , respectively, which were 4.6 and 12.5 times greater than the daily mean values specified by Taiwan national standards (125 µg m -3 for PM 10 and 35 µg m -3 for PM 2.5 ), respectively.Additionally, the aerosols produced by the beehive firework displays are high in Cl -, K + , and Mg 2+ ions, which are related to the materials used in the fireworks (Chang et al., 2011;Lin et al., 2014).
According to epidemiological studies, exposure to PM 2.5 is associated with adverse health effects, including respiratory and cardiovascular diseases (Pope et al., 2004;Dominici et al., 2006).Toxicological studies have also linked such exposure to oxidative DNA damage (Risom et al. 2005).In particular, toxicological studies have suggested that ultrafine particles (UFPs) may be more harmful to health than large particles because small particles have a much larger adsorbed or condensed surface area than large particles with the same mass and can be deposited in alveoli, where they interact with epithelial cells (Brown et al., 2002;Delfino et al., 2005;Oberdörster et al., 2005).Black carbon (BC) is also a major constituent of atmospheric aerosols, and it has an appreciable impact on global climate change, apart from adversely affecting human health (Power et al., 2001;Rich et al., 2005;Jacobson, 2010;Bond et al., 2013).Therefore, the risk to human health from exposure to high PM 2.5 , BC, and particle number concentrations within a short period could be severe.Except for PM 2.5 , information on the effect of the firework display on ambient BC and particle number concentrations is limited.In this study, real-time variations of PM 2.5 , BC, and particle number concentrations were monitored before and during a Yanshui beehive firework display to understand the effect of the display on the shortterm air quality.The characteristics of the particle number size distribution before and during the display were also determined.Moreover, the absorption Ångström exponent α was used as an index to determine the types of BC emission sources.The results can be used to address public health concerns and as a reference for managing similar events in the future.

Monitoring Site, Monitoring Equipment, and Data Collection
Yanshui, located in Southern Taiwan, is a district in Tainan City.It is famous for its firework display at the Lantern Festival, and it has a population of 26,000 and an area of 52 km 2 .It is estimated that more than 100,000 tourists, which is four times the local population, visit Yanshui on the day of the Lantern Festival (Chang et al., 2011).
Fig. 1 shows the monitoring locations.The location of the firework display was the downtown area of Yanshui.In this study, the monitoring instruments were set up on the rooftop of a building of Yanshui Junior High School (23°19′17.5″N,120°15′50.3″E).This monitoring site was located to the southwest of the main firework display site, and the distance between the two sites was approximately 200 m.A Grimm Series 1.108 Aerosol Spectrometer (Grimm Technologies, Inc., Douglasville, GA, USA) was used to determine the ambient PM 2.5 concentration.A TSI Model 3938 Scanning Mobility Particle Sizer (SMPS) spectrometer (TSI, Inc., Shoreview, MN, USA) was used to measure particle number size distributions for particles with diameters in the range of 13-750 nm.The measured particle number size distributions were fitted to distribution functions by using DistFit software (Chimera Technologies, Inc., Forest Lake, MN, USA) for evaluating geometric diameters and geometric standard deviations.The total particle number concentrations for the particle size range of 13-750 nm were calculated from the raw data of the SMPS.Additionally, two AE31 Rack Mount Aethalometer instruments (Magee Scientific Corp., Berkeley, CA, USA) were used to measure BC mass concentrations in parallel at two different aerosol sampling flow rates.The AE31 Aethalometer can be used to evaluate BC mass concentrations at seven different wavelengths: 370, 470, 520, 590, 660, 880, and 950 nm.Furthermore, a correction algorithm developed by Cheng and Yang (2015) was applied for minimizing measurement artifacts in the determination of BC mass concentrations.Inlets of the monitoring instruments were located at heights of approximately 15 m from the ground level and 1.5 m from the rooftop floor.The instruments were operated continuously from February 11-15, 2014, and the logging interval for all measurements was set at 5 min.Local meteorological data were recorded using a Vantage Pro 2 TM weather station (Davis Instruments Corp., Hayward, CA, USA).The wind speed and wind direction were measured at a height of 3 m above the rooftop floor.
The main firework display was performed between 20:00 and 23:00 (3 h) on February 14, 2014.Therefore, the data monitored during this period was treated as corresponding to an air pollution episode (episode period), and the data monitored at the same period of time on February 11 and 12 were treated as representing the reference air quality status for the area (nonepisode period).Differences in amounts of pollutants (or meteorological conditions) between different monitoring periods were tested using an independent sample t test.The significance level was 0.05 for all statistical tests.The strength of correlations between different pollutants

BC Correction
Aethalometer measurements of BC may be subject to optical interference (Liousse et al., 1993;Petzold et al., 1997;Reid et al., 1998;Bond et al., 1999;Weingartner et al., 2003), and it is possible to correct the artifacts to a certain extent by using numerical methods (Weingartner et al., 2003;Arnott et al., 2005;Virkkula et al., 2007;Collaud Coen et al., 2010).Recently, a new correction model was developed for determining artifacts in BC measurements (Cheng and Yang, 2015); the model is based on two aerosol loading rates.In the current study, this correction model was adopted.Accordingly, the two Aethalometer instruments were used to measure BC concentrations at sampling flow rates of 6 and 2 L min -1 .During the monitoring period, the filter tapes in the two instruments were shifted automatically to expose a pristine spot on them exactly every 8 h.This was done for ensuring that the two sampling spots had a fixed start and end time for the sampling, for facilitating internal comparison.
Before correction, the measured BC at the sampling flow rate of 6 L min -1 was significantly smaller than that at the sampling flow rate of 2 L min -1 by approximately 9% for the 370 nm wavelength (p < 0.001).However, the measured BC at the two sampling flow rates did not differ significantly for the 880 nm wavelength (p = 0.421) at this monitoring site.These measurement results demonstrated that the loading effect on the BC measurements at the shorter wavelength was stronger than that at the longer wavelength.After the BC measurements were corrected using the model of Cheng and Yang (2015), the corrected measurements for the two sampling flow rates could be exactly equal (p = 0.494 for the 370-nm wavelength; p = 0.492 for the 880-nm wavelength), indicating that the loading effect on BC measurement results could be minimized.Jeong et al. (2004) noted that BC concentrations measured at a wavelength of 370 nm could be considerably higher than those measured at a wavelength of 880 nm during a fire episode because of a marked increase in ultravioletabsorbing organic compounds.This finding points to the abundance of organic compounds in the particles released into the atmosphere during a forest fire.In the current study, the BC mass concentrations determined at the wavelengths of 370 and 880 nm were treated as ultraviolet black carbon (UVBC) and BC, respectively, to distinguish particles bound to ultraviolet-absorbing organic compounds (Wu et al., 2007;Sandradewi et al., 2008;Wang et al., 2011;Cheng et al., 2014).Furthermore, BC mass concentrations evaluated at the wavelengths of 470 and 950 nm were used to determine the absorption Ångström exponent, which can be used as an index to identify the types of BC emission sources (Kirchstetter et al., 2004;Day et al., 2006;Sandradewi et al., 2008).
The absorption Ångström exponent α can be computed from the aerosol light absorption at two different wavelengths in the ultraviolet to infrared spectral region.In this study the aerosol light absorption coefficients at 470 and 950 nm wavelengths was used: where b abs,λ=470 and b abs,λ=950 are the absorption coefficients of aerosol at 470 and 950 nm wavelengths, respectively.
The absorption coefficient of aerosol can be determined from the change rate of ATN: where A is the area of the sample spot, Q is the sampling flow rate, C is a light enhancement parameter, it is associated with multiple scattering of the light beam at the filter fibers in the unloaded filter, and it is strongly dependent on the filter material (Weingartner et al., 2003).In this study, C was set as 2.14 for the quartz filter (Weingartner et al., 2003).The original ATN value was measured from Aethalometer, and it was also modified with the correction model (Cheng and Yang, 2015).

Meteorological Conditions during the Beehive Firework Display and during the Same Period on Reference Days
Fig. 2 presents the time variation of meteorological conditions at the monitoring site during the whole monitoring period and on February 14, 2014.The average wind speed, principal wind direction, temperature, and relative humidity were 1.8 m s -1 , NNW-NNE, 13.8°C, and 71.3%, respectively, during the whole monitoring period.The daily average wind speed, principal wind direction, temperature, and relative humidity were 2.1 m s -1 , NNW-NNE, 13.3°C, and 73.5%, respectively, on February 14, 2014.Except for the wind speed, the meteorological conditions did not vary appreciably during the firework display (episode period).The wind speed during this period was between 0.9 and 1.9 m s -1 .Table 1 summarizes the meteorological data during the episode period and at the same period of time on the reference days (nonepisode period).The meteorological conditions did  not differ appreciably between these two periods, indicating that the effect of meteorological conditions on air quality could be ignored in this study.Moreover, Cheng et al. (2014) noted that in Southern Taiwan, the average wind speed, prevalent wind direction, temperature, and relative humidity in the winter season (December-February) were 1.8 m s -1 , NNW-NNE, 21.9°C and 71.9%, respectively.Except for the temperature, the meteorological conditions recorded in this study were similar to those observed by Cheng et al. (2014).Thus, the meteorological conditions during the monitoring period were typical of the winter season in Southern Taiwan.

Effect of the Beehive Firework Display on Ambient PM 2.5 Concentration
Fig. 3 shows the time variations of 5-min and 1-h PM 2.5 concentrations at the monitoring site during the whole monitoring period and on February 14, 2014.The average hourly PM 2.5 concentration during the whole monitoring period was 21.5 µg m -3 , and the interquartile range was 13.1-23.6µg m -3 .Measurement results show that PM 2.5 concentrations at this monitoring site varied irregularly, influencing by the local meteorological conditions, such as wind speed and the stability of the atmosphere boundary layer.An abrupt increase in the PM 2.5 concentration during the episode period can be clearly observed.After the episode period, the PM 2.5 concentrations decreased to the normal level in a few hours.Furthermore, a small peak in the 5min PM 2.5 concentration can be observed between 16:00 and 18:00, corresponding to the preview period preceding the firework display on February 14, 2014.Additionally, another small peak in the 5-min PM 2.5 concentration is observed in the morning between 8:00 and 10:00 on February 13, 2014.This peak likely corresponded to an open field burning of agricultural wastes at the upwind site of the monitoring location.The temporal variation of the 5-min PM 2.5 concentration during the episode period was high, and it was associated with the change in the wind direction.Table 2 presents the 5-min maximum and hourly average PM 2.5 concentrations for the episode and nonepisode periods.The hourly average PM 2.5 concentrations during the episode and nonepisode periods were 146.9 and 21.2 µg m -3 , respectively.The PM 2.5 concentration during the nonepisode period was similar to that commonly observed in suburban areas.According to the research results of Dominici et al. (2006) and Pope et al. (2006), the risks of congestive heart failure and acute coronary syndrome increased by 16.3% and 56.6%, respectively, during the episode period, comparing to those during the nonepisode period.In the current study, the 5-min maximum PM 2.5 concentration during the episode period increased to 593.9 µg m -3 , which was 22 times the value during the nonepisode period.Chang et al. (2011) observed that the PM 2.5 concentration during a Yanshui beehive firework display reached 250 µg m -3 , and Lin et al. (2014) found PM 2.5 concentrations at upwind and downwind locations to be 165 and 430 µg m -3 , respectively, during a Yanshui beehive firework display.The PM 2.5 concentrations recorded in the current study during a Yanshui beehive firework display did not match with those measured by Chang et al. (2011) and Lin et al. (2014).The mismatch could result from the difference in the monitoring locations, data resolutions, and meteorological conditions.However, all these measurement results consistently revealed that a large amount of PM can be produced during a beehive firework display, degrading the air quality in the surrounding areas.

BC Concentration and Absorption Ångström Exponent Value during the Beehive Firework Display
Fig. 4 shows the time variations of the 5-min and 1-h UVBC and BC concentrations at the monitoring site during the whole monitoring period and on February 14, 2014.The average hourly UVBC and BC concentrations during the whole monitoring period were 1283 and 1078 ng m -3 , respectively, and the interquartile ranges were 863-1511 and 768-1345 ng m -3 for UVBC and BC concentrations, respectively.Measurement results show that time variations of hourly UVBC and BC concentrations at this monitoring site are also irregularly, but they are not complete similar to that of the hourly PM 2.5 concentration (R Pearson = 0.692 for UVBC vs. PM 2.5 ; R Pearson = 0.616 for BC vs. PM 2.5 ).Despite the episode period, the remarkable peaks in the 5-min UVBC and BC concentrations also can be observed between 8:00 and 10:00 on February 13, 2014 and between 16:00 and 18:00 on February 14, 2014, respectively.Moreover, the UVBC concentrations are considerably greater than the BC concentrations during the episode period, indicating that ambient particles in this period were bound to ultraviolet-absorbing organic compounds, which were abundant.Table 3 presents the 5-min maximum and hourly average UVBC and BC concentrations during the episode and nonepisode periods.The hourly average UVBC concentrations during the episode and nonepisode periods are 4224 and 1280 ng m -3 , respectively, and the hourly average BC concentrations during the episode and nonepisode periods are 2639 and 1149 ng m -3 , respectively.During the nonepisode period, the UVBC concentration was close to the BC concentration, indicating that the aerosol was bound to organic compounds that showed less ultraviolet absorption.Furthermore, the 5-min maximum UVBC concentration during the episode period increased to 15558 ng m -3 , which was 11 times the value in the nonepisode period.The 5-min maximum BC concentration Table 2. PM 2.5 concentrations during the episode and nonepisode periods.
during the episode period increased to 7783 ng m -3 , which was six times the value in the nonepisode period.The UVBC concentration was considerably higher than the BC concentration by approximately two times during the episode period.Fig. 5 shows the time variation of hourly BC/PM 2.5 ratios during the whole monitoring period.The average hourly BC/PM 2.5 ratio during the whole monitoring period was 6.1%, and the interquartile range was 4.0-7.9%.The high ratio values of the BC/PM 2.5 usually can be observed during the afternoon hours, and low ratio values of the BC/PM 2.5 occur during the predawn hours at this monitoring site.The probably reason is that reduced BC aerosols production and their removal by deposition decreases the BC aerosols load during the predawn hours.The BC concentration increased less than the PM 2.5 concentration during the episode period, comparing to those in the nonepisode period.The BC concentration was only 1.8% of the PM 2.5 concentration during the episode period, and the BC/PM 2.5 ratio was 5.5% in the nonepisode period.Past results have indicated that a relatively large amount of other compounds exist in the aerosols produced by a firework display (Wang et al., 2007;Moreno et al., 2010;Chang et al., 2011;Lin et al., 2014).For example, Chang et al. (2011) and Lin et al. (2014) found that the aerosols produced were remarkably high in Cl -, K + , and Mg 2+ ions compared with the background aerosols.Wang et al. (2007)  and Moreno et al. (2010) demonstrated that concentrations of elements such as Sr, Mg, K, Ba, and Cu were far greater in airborne dust during firework episodes.The presence of these highly toxic metals was related to the materials used in the fireworks.
The absorption Ångström exponent has been used as an index for determining the types of BC emission sources (Kirchstetter et al., 2004;Day et al., 2006;Sandradewi et al., 2008).Kirchstetter et al. (2004) noted that the absorption Ångström exponent value ranged between 0.8 and 1.1 for diesel soot, and Day et al. (2006) found that the absorption Ångström exponent values for fresh wood smoke aerosol were between 0.9 and 2.2 and strongly dependent on the type of wood and burning conditions.Sandradewi et al. (2008) observed that the absorption Ångström exponent values for vehicular traffic and wood burning were 1.1 and in the range of 1.8-1.9,respectively; a high absorption Ångström exponent value was observed for wood combustion aerosol because the aerosol was bound to compounds showing strong absorption in the ultraviolet region.Fig. 6 shows the time variation of hourly absorption Ångström exponent values during the whole monitoring period.The average  hourly value of the absorption Ångström exponent during the whole monitoring period was 1.17, and the interquartile range was 1.08-1.24.The highest hourly value of the absorption Ångström exponent was recorded in the morning between 8:00 and 10:00 on February 13, 2014 and it was high as 2.18.This peak value was corresponded to a positive Delta-C (UVBC-BC), suggesting BC aerosols was bound to organic compounds showing strong absorption in the ultraviolet region.At this situation, it was likely an open field burning of agricultural wastes occurred during this period at the upwind site of the monitoring location.Additionally, the positive Delta-C values also could be found from the afternoon to evening hours through the whole monitoring period.These positive Delta-C values likely corresponded to an increase of household activities and residential wood burning emissions after midday.In this study, the hourly absorption Ångström exponent values during the episode and nonepisode periods were 1.42 and 1.13, respectively.The measurement results showed that the sources of BC aerosols during the beehive firework display were vehicular traffic and firework burning.Traffic congestion always occurred during this period, with a large number of visitors entering the downtown area.Compared with the episode period, a higher hourly absorption Ångström exponent value (1.63) was observed during the preview period preceding the firework display.In this period, the main source of BC aerosols was firework burning because the effect of vehicular traffic on the BC concentration was relatively small because most visitors arrived in the downtown area after 18:00, after leaving their workplaces.Measurement results also indicated that vehicular traffic was the major source of BC aerosols at the monitoring site during the nonepisode period.However, the BC concentration (1149 ng m -3 ) during the nonepisode period was lower than that in urban environments.Cheng et al. (2014) noted that the BC concentration was 3330 ng m -3 in the Kaohsiung urban area in Southern Taiwan and that the BC/PM 2.5 ratio was approximately 11%.

Particle Number Concentration and Size Distribution during the Beehive Firework Display
Fig. 7 shows the time variations of 5-min and 1-h total particle and UFP number concentrations at the monitoring site during the whole monitoring period and on February 14, 2014.The total particle and UFP number concentrations in the size ranges of 13-750 and 13-100 nm, respectively, were calculated from the raw data of the SMPS.The average hourly total particle and UFP number concentrations during the whole monitoring period were 6.76 × 10 3 and 3.64 × 10 3 # cm -3 , respectively, and the interquartile ranges were 4.55 × 10 3 -6.91 × 10 3 and 2.36 × 10 3 -3.98 × 10 3 # cm -3 for total particle and UFP number concentrations, respectively.Time variations similar to that of the UVBC concentrations were observed for the total particle and UFP number concentrations (R Pearson = 0.882 for total particle vs. UVBC; R Pearson = 0.871 for UFP vs. UVBC).The total particle number concentration increased more than the UFP number concentration during the episode period, indicating that most particles existed in the atmosphere during this period had a size greater than 100 nm.Table 4 presents the 5-min maximum and hourly average of the total particle and UFP number concentrations during the episode and nonepisode periods.The hourly average total particle number concentrations during the episode and nonepisode periods were 3.37 × 10 4 and 5.71 × 10 3 # cm -3 , respectively, and the hourly average UFP number concentrations during the episode and nonepisode periods were 1.18 × 10 4 and 3.15 × 10 3 # cm -3 , respectively.Previous studies have found the mean UFP number concentrations in urban areas to range between 1.0 × 10 4 and 2.0 × 10 4 # cm -3 (Noble et al., 2003;Hussein et al., 2004;Matson, 2005;Cheng et al., 2013).The UFP number concentrations near a major highway have been observed to be greater than 10 5 # cm -3 , which is considerably higher than those in urban areas by 5-10 times (Hagler et al., 2009;Buonanno et al., 2009;Cheng et al., 2010).Moreover, over 80% of airborne particles in urban air are in the UFP size range (Shi et al., 2001).The UFP number concentration during the nonepisode period at the monitoring site was only 3.15 × 10 3 # cm -3 in the current study, considerably lower than that previously observed at urban sites.Fig. 8 shows the time variation of hourly ratios of the UFP number concentration to the total particle number concentration during the whole monitoring period.The average hourly ratio of the UFP number concentration to the total particle number concentration during the whole monitoring period was 54.8%, and the interquartile range was 49.4-60.8%.The high ratio values of the UFP-to-Total particle usually can be observed during the afternoon hours, corresponding to photochemical nucleation processes under higher temperature conditions during noon hours.The ratios of the UFP number concentration to the total particle number concentration were 34.9% and 55.6%, respectively, during the episode and nonepisode periods, indicating that the particle number concentrations during the episode period Table 4. Particle number concentrations during the episode and nonepisode periods.

Monitoring period Nonepisode period a
Episode period b Time Total particle number, # cm -3 5 min max.8.07 × 10 3 1.15 × 10 5 14.3 Hourly average 5.71 × 10 3 3.37 × 10 4 5.9 UFP number, # cm -3 5 min max.were dominated by particles with a size greater than 100 nm.Furthermore, the 5-min maximum total particle number concentration during the episode period reached 1.15 × 10 5 # cm -3 , which was 14 times the value in the nonepisode period.This number concentration for the episode period was comparable with that at a busy traffic site and could severely affect human health.
Table 5 shows the particle number concentrations for the episode and nonepisode periods for seven selected particle size ranges: 18-32, 32-56, 56-100, 100-180, 180-320, 320-560 nm, and other sizes (comprising 13-18 and 560-750 nm).The particle number concentrations were dominated by particles in the size range of 100-180 nm (30.3%), 180-320 nm (25.2%), and 56-100 nm (19.8%) during the episode period.By contrast, the particle number concentrations were dominated by particles in the size range of 56-100 nm (28.1%), 100-180 nm (20.2%), and 32-56 nm (19.1%) during the nonepisode period.Fig. 9 shows the time variation of particle number size distributions at the monitoring site during the whole monitoring period and on February 14, 2014.According to measurement results, the particle number size distribution patterns for different periods were remarkably different.During the firework display and preview period, the dominated modes of the particle number size distributions were 180 and 118 nm, respectively, on February 14, 2014.Additionally, a dominated mode could be observed at 23 nm during afternoon hours on February 12, 2014.At this situation, the UFP-to-Total particle ratio was the highest in the whole monitoring period (Fig. 8).This dominated mode of the particle number size distribution was probably caused by photochemical nucleation processes.Fig. 10 shows the average particle number size distributions measured in the episode and nonepisode periods.The measured particle number size distribution was determined using the DistFit software.Table 6 lists the geometric diameters and geometric standard deviations of the particle number size distributions.The fitting results showed that the number size distribution pattern at the monitoring site could be accurately simulated by two lognormal models.The particle number size distribution patterns for the episode and nonepisode periods were considerably different.The particle number size distribution during the episode period exhibited two modes, a major accumulation mode and a minor Aitken mode of 180 and 63 nm, respectively.During the nonepisode period, the particle number size showed a two-mode distribution, with the main mode at 78 nm and a small mode at 342 nm.This pattern was similar to that of the average particle number size distribution during the whole monitoring period.The average hourly size distribution of the particle number during the whole monitoring period also exhibited a two-mode pattern, a major Aitken mode and a minor accumulation mode of 76 and 301 nm, respectively.Measurement results suggested that particle coagulation played a crucial role in removing particles in the nucleation and Aitken modes in the ambient air, which showed a high  Little information is available on the correlation between BC mass and size-resolved particle number concentrations.According to measurement results, hourly BC mass concentrations were positively correlated with hourly particle number concentrations in the size range of 56-100 nm (R Pearson = 0.810), followed by 32-56 nm (R Pearson = 0.760) and 100-180 nm (R Pearson = 0.733) during the whole monitoring period.Except the episode period, measurement results indicate that most of the BC in aerosols at this monitoring site could range from 32-180 nm in size.However, the correlation between hourly UVBC mass and particle number concentrations was stronger than that between hourly BC mass and particle number concentrations.The hourly UVBC mass concentrations were significantly positively correlated with hourly particle number concentrations in the size range of 56-100 nm (R Pearson = 0.895), followed by 100-180 nm (R Pearson = 0.854) and 32-56 nm (R Pearson = 0.828) during the whole monitoring period.Fig. 11 shows the time variations of hourly particle number concentrations in the size range of 32-180 nm and UVBC mass concentrations at the monitoring site during the whole monitoring period.Despite the poor relationship between particle number and UVBC mass concentrations in the episode period, the measurement results demonstrate that time variation of the hourly particle number concentrations in the size range of 32-180 nm was remarkably similar to that of the UVBC concentrations (R Pearson = 0.910) during the whole monitoring period.

CONCLUSIONS
The hourly average concentrations of PM 2.5 , BC, total particle number, and UFP number during the episode period were 146.9 µg m -3 , 2639 ng m -3 , 3.37 × 10 4 # cm -3 , and 1.18 × 10 4 # cm -3 , respectively.These values were 6.9, 2.3, 5.9, and 3.7 times greater than those during the nonepisode period, respectively.Moreover, the 5-min maximum concentrations of PM 2.5 , BC, total particle number, and UFP number during the episode period were 21.9, 5.8, 14.3, and 6.0 times greater than those during the nonepisode period.
The measurement results showed that a large amount of PM was produced during the beehive firework display, degrading the air quality in the surrounding areas.The PM concentrations at the ground level near the display site were far higher than the values measured at the monitoring site, which was located on the rooftop of a 15-m-high building.Therefore, the risk posed by the high short-term PM concentration to public health should be addressed.
The measurement results showed that the ratio BC/PM 2.5 was only 1.8% during the episode period, considerably lower than the ratio during the nonepisode period.This low ratio indicates that a relatively large amount of other compounds existed in the aerosols generated by firework burning.The measured UVBC and BC values showed that these aerosols were bound to ultraviolet-absorbing organic compounds, which were abundant.The sources of BC aerosols during the beehive firework display were vehicular traffic and firework burning, and the absorption Ångström exponent value was 1.4.Furthermore, the particle number concentration during the episode period was governed by particles in the size range of 100-180 nm (30.3%), 180-320 nm (25.2%), and 56-100 nm (19.8%), and the particle number size distribution exhibited two modes: a major accumulation mode and a minor Aitken mode of 180 and 63 nm, respectively.This particle number size distribution pattern differed considerably from that for the nonepisode period.During the episode period, particle coagulation played a crucial role in removing particles in the nucleation and Aitken modes in the ambient air, which showed a high particle number concentration.

Fig. 2 .
Fig. 2. Time variations of meteorological conditions at the monitoring site (a) during the whole monitoring period and (b) on February 14, 2014.

Fig. 3 .
Fig. 3. Time variations of 5-min and 1-h PM 2.5 concentrations at the monitoring site (a) during the whole monitoring period and (b) on February 14, 2014.

Fig. 4 .
Fig. 4. Time variations of 5-min and 1-h UVBC and BC concentrations at the monitoring site (a) during the whole monitoring period and (b) on February 14, 2014.

Fig. 5 .
Fig. 5. Time variation of hourly BC-to-PM 2.5 ratios at the monitoring site during the whole monitoring period.

Fig. 6 .
Fig. 6.Time variations of hourly absorption Ångström exponent value, UVBC and BC concentrations at the monitoring site during the whole monitoring period.

Fig. 7 .
Fig. 7. Time variations of 5-min and 1-h total particle (TP) and UFP number concentrations at the monitoring site (a) during the whole monitoring period and (b) on February 14, 2014.

Fig. 8 .
Fig. 8. Time variation of hourly UFP-to-Total particle ratios at the monitoring site during the whole monitoring period.

Fig. 9 .
Fig. 9. Time variation of particle number size distributions at the monitoring site (a) during the whole monitoring period and (b) on February 14, 2014.

Fig. 10 .
Fig. 10.Average particle number size distributions during the episode and nonepisode periods.

Fig. 11 .
Fig. 11.Time variations of hourly particle number concentrations in the size range of 32-180 nm and UVBC concentrations at the monitoring site during the whole monitoring period.

Table 1 .
Meteorological data for the episode and nonepisode periods.