Effective Density of Submicron Aerosol Particles in a Typical Valley City , Western China

Evolution of effective density of submicron aerosol particles and effects of firework displays and weather conditions on particle effective density in a typical valley city in northwestern China were evaluated by fitting the APS data to the SMPS data in the overlapping size region of the two instruments during September 2012 to August 2013. The monthly mean effective densities for submicron aerosol particles were found to vary from 1.3 g cm to 1.6 g cm depending on the month of year. In addition, the ranges of average submicron particle effective density were different on workdays, Saturdays, Sundays and holidays in different seasons with minimums in the morning and afternoon, especially observed in morning rush hours of weekdays, and a maximum during midday. The precipitation and strong wind resulted in particle effective densities increased by 0.1 g cm and 0.08 g cm, respectively, while effective densities of submicron aerosol particles decreased by 0.21 g cm due to the effect of dust. During the periods mainly affected by firework displays, the effective densities of submicron aerosol particles were 1.19 g cm, which was much lower than those for fireworkunaffected days, and the submicron particle effective densities decreased by 0.29 g cm due to firework displays.


INTRODUCTION
Regional or global climates were significantly affected by direct and indirect radiation effects of aerosol particles (Seinfeld and Pandis, 1998), and they also have serious health impact especially ultrafine particles (diameter smaller than 100 nm).Numerous epidemiological studies have found an association between particle number concentrations and cardiovascular and respiratory morbidity and mortality (Dockery et al., 1993;Laden et al., 2000).Ultrafine particles can easily enter in the bronchi and alveoli of the human respiratory system and deposit in the regions of the respiratory apparatus (Dockery et al., 1994;International Commission on Radiological Protection, 1994;Oberdörster et al., 2005).The main mechanisms of deposition of inhaled particles in the respiratory apparatus have diffusion, inertial impaction, gravitational settling (sedimentation) and interception.The diffusion, an important deposition mechanism for ultrafine particles, is mainly related to temperature and particle size, while the importance of other mechanisms for larger particles are related to density, size and retention time of the particles in the respiratory apparatus (International Commission on Radiological Protection, 1994;Oberdörster et al., 2005).Therefore, particle density may play a prominent role in the understanding of health effects of PM (particulate matter) pollutants.However, ambient particle density is directly affected by many factors such as sources, physical and chemical characteristics and the generation processes of aerosol particles.In addition, size-revolved particle density can be used to convert number to mass size distributions when particle shape is assumed as mobility and aerodynamic diameters are equivalent diameters (McMurry et al., 2002) and it also provide an association between mobility and aerodynamics sizes.The size of airborne particles varies over a range roughly from a few nanometers to hundred micrometers.The size-revolved particle density can be determined by combination of TDMPS-APS and MOUDI instruments (Hu et al., 2012) or DMA-APM approach (Geller et al., 2006).For example, Hu et al. (2012) showed that sizeresolved particle effective density increased with diameter from about 1.5 g cm -3 at the size of 0.1 µm to above 2.0 g cm -3 in the coarse mode in urban Beijing.In this study, the only average effective density of submicron airborne particles in urban Lanzhou was estimated due to limitation of instruments used.Thus, the effective densities determined from our study were limited to submicron airborne particles.
Particle effective density cannot be directly measured, but it can be deduced by applications of numerous dual and triple-instrument setups.To sum up, two methods were developed to calculate effective density of aerosol particles.The first method determined effective density based on the ratio of the aerosol mass to volume concentrations measured by different instruments, such as DMA (Differential mobility analyzers)-APM (Aerosol particle mass analyzer) (Charvet et al., 2014;McMurry et al., 2002;Park et al., 2003;Yin et al., 2015), DMA-Couette CPMA (Olfert et al., 2007), SMPS-APS + TEOM (Khlystov et al., 2004), SMPS + TEOM (Morawska, et al., 1999), DMPS + TEOM (Pitz et al., 2003(Pitz et al., , 2008a)), TDMPS-APS + MOUDI (Hu et al., 2012) and DMA-ATOFMS (Spencer et al., 2007).However, the instrument of TEOM (tapered element oscillating microbalance) measured the aerosol PM 2.5 mass concentration, so the comparison of the PM 2.5 particle volume concentration from the merged SMPS-APS or DMPS distribution with the TEOM PM 2.5 mass concentration has only provided us with an estimate of a bulk "effective" particle density (Khlystov et al., 2004).Moreover, the TDMPS-APS + MOUDI approach has also limits, not least the low size resolution of MOUDI cascade impactor and volatile material was lost from the MOUDI stages.During the summer more than 70% of nitrate was lost from the MOUDI stages (Cabada et al., 2004).The method of DMA-APM was introduced in detail by Charvet et al. (2014).This approach is time-consuming, requiring 1 h to provide the particle effective density for a given particle size.In addition, the APM cannot readily be used alone as it is based on a principle requiring the use of both a DMA and a particle detector, at least (Bau et al., 2014).The second method deduced effective density by combination of mobility and aerodynamic diameter measurements from various instruments in series or in parallel, such as DMA-ELPI (Electrical low-pressure impactor) (Maricq and Xu, 2004;Bau et al., 2014), SMPS + ELPI (Ristimaki et al., 2002;Virtanen et al., 2004), DMA-OPC-APS (Hand and Kreidenweis, 2002) and DMA + APS (Karg, 2000).The particle effective densities in various environments in the world were summarized in Table 1, and the sampling locations and used instruments were also given.The density of newly particles emitted from motor vehicles was generally lower than 1.0 g cm -3 , which was significantly lower than that of ambient particles (see Table 1).However, the particles from natural sources such as sea salt and mineral dust generally have densities far above 1.0 g cm -3 due to more compact structure compared with the fresh particles from diesel engines (Pitz et al., 2008a).Particle effective density has large spatio-temporal difference due to varying sources, physical and chemical particle composition among the cities (Pitz et al., 2003(Pitz et al., , 2008a)).However, few studies have been carried out to describe the particle effective density in China, with most of them focused on the developed cities such as Beijing (Gao et al., 2007;Yue et al., 2009;Hu et al., 2012) and Shanghai (Gao et al., 2007;Yin et al., 2015).The data given in the previous studies and our study were summarized in Table 2 to help the readers compare the different environments and fractions considered.For example, Gao et al. (2007) indicated that the densities of PM 2.5 (particles with aerodynamic diameter smaller than 2.5 µm) were 1.5 g cm -3 and 1.7 g cm -3 in Beijing and Shanghai, respectively.Yue et al. (2009) found particle apparent densities of PM 1 and PM 10 in Beijing to be about 1.4 3 g cm -3 and 1.68 g cm -3 , respectively.Based on the measurements of aerosol number, mass and chemical size distributions, Hu et al. (2012) estimated size-resolved ambient particle density in the winter season in Beijing, and found effective densities of ambient particles were 1.62 ± 0.38 g cm -3 for PM 1.8 and 1.67 ± 0.37 g cm -3 for PM 10 .Until now, there have been no relevant reports on the particle density in the cities in Northwestern China.In this study submicron particle effective density in urban Lanzhou was estimated using a scanning mobility particle sizer (SMPS model 3936, TSI, USA) and an aerodynamic particle sizer (APS model 3321, TSI, USA) in parallel, where particles were classified by mobility size with the SMPS and the aerodynamic sizes were determined by the APS.The objective of this study is to investigate the seasonal and diurnal variations of effective density of submicron aerosol particles in urban Lanzhou, and understand the effects of firework displays and weather conditions on particle effective density, as well as provide a basic knowledge for the submicron particle density in the cities in northwestern China.

METHODS
Fig. 1 shows the location of Lanzhou and the sampling site.The detailed descriptions can refer to Zhao et al. (2014).A study by Imhof et al. (2005) indicated that at 30 m above ground the background concentration was attained.So, our instrument captured the particle density in urban background of Lanzhou.Particle size distributions (10 nm-10 µm) were measured using scanning mobility particle sizer (SMPS model 3936, TSI, USA) and aerodynamic particle sizer (APS model 3321, TSI, USA) during 1 September 2012 to 31 August 2013.The SMPS and APS measure the size distribution of aerosols in the size range 10-1000 nm and 0.5-10 µm, respectively.The SMPS system (Model 3936, TSI, USA) consists of an electrostatic classifier (TSI Model 3080) and a particle detector (butanol-based Condensation Particle Counter, TSI Model 3775).The cutoff diameter and counting mode were 735 nm and single-particle count mode, respectively.Additionally, the instrument is fitted with an impactor (Stk 50 = 0.23, nozzle diameter = 0.508 cm) to reduce the influence of large, multiple charged particles on the mobility spectrum.The sheath and aerosol flows through the DMA were set to be 3 L min -1 and 0.3 L min -1 with a scan time of 295 s, respectively.Furthermore, the SMPS size distributions were corrected for diffusion losses according to Birmili et al. (1997).Lall and Friedlander (2006) found that SMPS cannot be directly used and must be corrected to obtain the surface area and volume distributions for aggregates.However, we calculated particle effective density based on number distributions, so aggregate correction did not use in this study.Additionally, although the size range of the APS is density-dependent (Wang and John, 1987), if one is sampling particles with density below 0.8 g cm -3 or above 2.0 g cm -3 or if one is primarily interested in particles below 1 µm, Stokes correction should be turned off (Aerosol Instrument Manager ® Software for APS Sensors, 2002).Thus, we turned off Stokes correction for APS by Aerosol Instrument Manager ® software during the experiment to improve the particle sizing accuracy of APS instrument.The working principles of SMPS and APS were also described detailedly by Zhao et al. (2014).The standardized quality control (QC) and quality assurance (QA) procedures for the measurement of particle size distributions using both TDMPS and APS instruments were firstly described by Pitz et al. (2008b).Referring to their proposed QC and QA procedures and the instructions of the manufacturer concerning maintenance and calibration, SMPS's mobility and the APS's time-of-flight responses were calibrated using monodisperse aerosols such as polystyrene latex (PSL) particles prior to their deployment in the field.The instruments normally maintained and calibrated by the manufacturer on a yearly basis.West Roads are main road in urban Lanzhou with traffic volumes higher than 2500 cars hr -1 , and the Second Thermal Power Plant is also a main central heating company in winter in Chengguan District of Lanzhou City, which is around 2.5 kilometers from our sampling site).
Furthermore, the operational parameters of the two instruments such as flow rates, temperatures, pressures, humidity and liquid levels were checked periodically and, if necessary, adjusted on a daily basis.The zero-points of the CPC and APS were checked monthly using a purge filter at the respective inlet.During the study period, the impactor of SMPS, and the inner and outer nozzle of the APS were also cleaned every day and every two weeks, respectively.Furthermore, the sample and sheath flowrates of SMPS and APS were examined periodically with two bubble flow meters (TH-ZM8 and TH-ZM10, Tianhong, China) to insure the good performance of the instruments.The measuring ranges of both butter flow meters were 0.1-2 L min -1 and 1-30 L min -1 , respectively.
The raw data of all devices were visually checked for plausibility and outliers resulting from malfunctioning were removed from the data set (Pitz et al., 2008b).After checking for plausibility and outliers, daily or hourly average was considered valid if 66% of the data were available due to weak representation of less data.For example, the particle number concentrations (PNC) during the daytime were generally higher than those during the nighttime due to effects of human activities, so daily mean PNC for missing more daytime data will be significantly lower than normal daily mean particle number concentrations.Additionally, particle effective density ranged from 0.6 g cm -3 to 3 g cm -3 for urban aerosol (Pitz et al., 2003).In order to obtain high quality PSD data and more accurately calculate particle effective density by combination of SMPS and APS, the daily or hourly mean data was considered valid if effective density ranged from 0.6 g cm -3 to 3 g cm -3 during the study period.Tenminute meteorological data such as air temperature, relative humidity and wind speed and direction were also obtained with an automatic meteorological station co-located with the sampling site.Beijing time (BT) (= UTC + 8) is used in this paper.
In this study, the hourly mean particle effective density in the overlapping size region of SMPS and APS was calculated by fitting the APS data to the SMPS data, which was similar to that presented by Karg (2000).This density is apt to characterize the sedimentation of accumulation mode particles.In the overlapping concentration interval N 1 , ..., N 0 the average mobility diameter D ̅ p and aerodynamic diameter D ̅ a are: Particle density and shape have important effects on diameter measurements (DeCarlo et al., 2004).A correction to the drag equation must be introduced to account for the reduction in drag that occurs when the relative velocity of the gas at the particle surface is nonzero (Hinds, 1999).This correction is implemented via the Cunningham Slip Correction Factor C c .Furthermore, the slip correction coefficient should be taken into account when effective density of submicron particles was calculated.Based on D ̅ p and D ̅ a , the average effective density can be calculated by the following formula (DeCarlo et al., 2004): where, ρ 0 and χ were standard density (1 g cm -3 ) and dynamic shape factor, respectively.In this study, the spherical assumption (χ = 1) was used due to the lack of observation of particle shape factor in urban Lanzhou.When the particles are mildly nonspherical, C c (D ̅ a ) was also thought to be roughly equal to C c (D ̅ p ) Therefore, the ratio of the slip correction factors

RESULTS AND DISCUSSION
The study period (September 2012-August 2013) was divided into four seasons, i.e., spring (March-May, 2013), summer (June-August, 2013), fall (September-November, 2012) and winter (December 2012-February 2013).In order to better study effective density of submicron aerosol particles in urban Lanzhou, the monthly mean particle number size distributions (PNSDs, dN/dlogD p ) representing particle number concentrations of each channel of the overall size range during September 2012 to August 2013 were first showed in Fig. 2. The error bars were not displayed to more clearly see monthly mean PNSD, and D p in that figure referred to mobility-equivalent diameter.The particle number concentrations in four modes (nucleation (10-50 nm), Aitken (50-100 nm), accumulation (100-1000 nm) and coarse modes (1-10 µm), dN) in each season were also counted in Table 3, which integrated PNSD in the corresponding size ranges.The mean number concentrations of particles in Aitken, accumulation and coarse modes were (11380.1 ± 6065.9) cm -3 , (8114.6 ± 5014.6) cm -3 and (12.1 ± 5.6) cm -3 in winter, which was significantly higher especially in January than those in the other seasons due to stable weather conditions.However, the mean particle number concentrations in nucleation mode in spring and summer were much higher than those in fall and winter with maximum number concentrations of 75447.7 cm -3 due to new particle formation events (Dall'Osto et al., 2011).Furthermore, the mean number of nucleation mode particles accounted for 57.4% and 54.6% of total number concentrations (10 nm-10 µm) in spring and summer, respectively, while that only accounted for 39.3% of total number concentration in summer.As it can be seen from Fig. 2, the mode diameters of PNSDs largely depended on the month of year.In general, the particles were coarser in heating season (November 2012-March 2013) than those in non-heating season.The mode diameters of PNSDs were in the size range 20-30 nm in September 2012 and April to August 2013, while for the period November 2012 to March 2013, the mode diameters were in 60-70 nm.
The finer particles in spring and summer were mainly related to more frequent new particle formation (NPF) events (Holmes, 2007;Dall'Osto et al., 2011).The NPF events can be understood by diurnal variations of particle number size distributions.An event of NPF was identified if there was a burst of particle number in the nucleation mode (10-50 nm) followed by subsequent particle growth (Dal Maso et al., 2005), which were reported in our sampling site of urban Lanzhou using the same data as this paper (Zhao et al., 2015b).NPF events mainly occurred in spring and summer due to favorable meteorological conditions and low condensation sink.Atmospheric aerosol formation consists of a complicated set of processes that include the production of nanometer-size clusters from gaseous vapours, the growth of these clusters to detectable sizes, and their simultaneous removal by coagulation with the pre-existing aerosol particle population.Currently, several mechanisms for new particle formation which proposed the importance of each mechanism are still the center of much debate.Proposed nucleation mechanisms include condensation of a binary mixture of sulphuric acid and water; ternary nucleation of sulphuric acid, water and a third molecule, most likely ammonia; ion-induced nucleation; secondary organic aerosol formation involving condensation of low-or non-volatile organic compounds and homogeneous nucleation of iodine oxides (Holmes, 2007).In addition, atmospheric particle formation events are significantly affected by environmental factors, such as temperature, humidity and the surface area of pre-existing particles (Zhao et al., 2015b).For example, the NPF events were easily occurred in spring and summer due to strong solar intensity, low RH (relative humidity) and high wind speed (Dall'Osto et al., 2011).

Evolution of Particle Effective Density
Fig. 3 gave annual variation of particle effective density during study period, and the span of the effective densities in four seasons were also provided in Fig. 4. The discrepancies of particle density between months can be attributed to changes in the proportions of particles from various sources present in the atmosphere (Morawska et al., 1999).For example, domestic heating was main pollution sources in winter in northern China.Freshly generated airborne particles from coal or biomass combustion processes for domestic heating consist mostly of carbonaceous materials and are found predominantly in the accumulation mode (100-1000 nm).The particle size distributions vary depending on combustion appliance, fuel type, combustion phase and  sampling method (Vu et al., 2015).An estimation of the average effective densities for ambient air submicron particles was found to vary from 1.3 g cm -3 to 1.6 g cm -3 depending on the month of year with the minimum and maximum densities of 1.31 g cm -3 and 1.62 g cm -3 in November 2012 and August 2013, respectively.The study by Pitz et al. (2008a) suggested maximum and minimum particle densities of PM 2.5 were in spring and winter, respectively.However, in our study average effective density of submicron aerosol particles in spring was (1.36 ± 0.22) g cm -3 , which was the lowest among the four seasons maybe due to our aim of investigating submicron particle density.Lanzhou city, located at the downstream of desert regions such as Taklimakan and Gobi Deserts, was affected easily by dust events.As one of the most air pollutants in spring in urban Lanzhou, dust particles with large density mainly concentrated on the aerodynamic diameter larger than 1 µm (Zhao et al., 2015c), which was beyond the submicron range of our study.In addition, the study of Pitz et al. (2008a) focused on urban air in Augsburg, Germany, while our measurements were in urban Lanzhou, China.Differences of energy structure and economic development level can also resulted in discrepancy of particle effective density between the two cities.Additionally, compared with other seasons, the variation of submicron particle density was weaker in winter in urban Lanzhou (see also Fig. 4).The low variation in submicron particle density, made it easy to estimate the mass concentration from the SMPS number concentration with an assumed density value.
In order to better understand the dependence of submicron particle density on the time of day, the diurnal variations of effective densities of submicron aerosol particles for workdays, Saturday, Sunday and holidays in different seasons during study period were presented in Fig. 5.The ranges of mean submicron particle effective density were different on workdays, Saturday, Sunday and holidays in fall, spring and summer with minimums in the morning and afternoon, which were related to traffic soot particles with low density due to their agglomerate structure (Yin et al., 2015), especially observed in morning rush hours of workdays, and a maximums during midday due to increased secondary particle production and the presence of more aged particles with the built-up of the convectively mixed boundary layer (Pitz et al., 2008a).However, the differences of average submicron particle effective density on workdays, Saturday, Sunday and holidays in winter were not obvious due to the stable weather conditions.The hourly mean effective densities of submicron aerosol particles on holidays were almost higher than those on workdays, Saturdays and Sundays in fall, winter and spring.The study about sizeresolved effective density indicated that particle density increased with diameter from about 1.5 g cm -3 at the size of 0.1 µm to above 2.0 g cm -3 in the coarse mode (Hu et al., 2012).Aerosol particles from motor vehicles mainly located at nucleation and Aitken modes (10-100 nm).Therefore, decrease of traffic volumes in urban area on holidays due to people driving out of the urban area for enjoying with family and friends can result in increases of particle effective density.However, particle densities on holidays were the lowest in summer especially before 08:00, which may be related to the fact that the holiday was the first day of strict single and double traffic restriction measures implemented by the local government (11 June-10 July 2013) and more vehicles can be on road due to fluke mind of people.
Particle effective density is largely affected by many factors such as sources, physical and chemical characteristics and the generation processes of aerosol particles and weather conditions.Furthermore, application of different approach to determine density maybe also causes some differences of calculated density.Particle densities in various environments in the world and in Beijing, Shanghai and Lanzhou, China were summarized in Tables 1 and 2, respectively.In order to understand the difference or specific characteristics of the effective density in Lanzhou compared to other cities or regions, the effective densities in four seasons in urban background of Lanzhou city from our study were also given in Table 2.As it can be seen from the two tables, particle densities have large discrepancy among the cities or regions due to varying pollution sources, aerosol characteristics and weather conditions.
The results from our study indicated that particle densities were between (1.36 ± 0.22) g cm -3 and (1.59 ± 0.18) g cm -3 in urban air of Lanzhou city, which were comparable to some cities such as Atlanta (McMurry et al., 2002), Texas (Hand and Kreidenweis, 2002), Pittsburgh (Khlystov et al., 2004) and Shanghai (Yin et al., 2015).However, particle effective densities in urban Lanzhou were significantly higher than those of diesel exhaust particles (Park et al., 2003;Maricq and Xu, 2004;Olfert et al., 2007) mainly due to their agglomerate structure.In addition, there are two major roads with traffic volume more than 2000 cars per hour near the sampling site, but our sampling site is on the roof of a 32-m high research building, high enough to reduce the effect of direct exhaust emissions from motor vehicles due to hygroscopic or coagulation growth of newly emitted particles during transportation.A study by Imhof et al. (2005) also indicated that at 30 m above ground the background concentration was attained.The seasonal variation of particle density was less analyzed.Pitz et al. (2008a) studied seasonal variation of PM 2.5 apparent particle density (ρ 2.5 ) in urban air of Augsburg, Germany using TDMPS-APS + TEOM approach.They found that mean ρ 2.5 over the two year period was 1.65 g cm -3 with maximum of 2.76 g cm -3 , which were much higher than the results of our study, which may be mainly related to the fact that dust particles were not considered in our study mainly focused on submicron ranges.Additionally, effective densities of submicron particles in summer and winter were much higher than those in spring and fall in urban Lanzhou, which were mainly due to more frequent new particle formation (NPF) events in summer (Zhao et al., 2015b) and more submicron particles from coal or biomass burning for domestic heating in winter in urban Lanzhou.However, Pitz et al. (2008a) indicated PM 2.5 apparent particle density was even higher in spring due to more frequent dust events in that season.The above analyses indicated particle density was significantly affected by sources.

Effects of Weather Conditions and Firework Displays on Particle Density
The mean diurnal variations of submicron particle effective densities for precipitation-, strong windy-, and dusty-affected days were compared with the other days excluded the days in the four seasons to investigate the effect of weather conditions on particle effective density (Fig. 6).The day was identified as precipitation-affected day if duration of precipitation was more than 5 hours, and the day was considered to be strong windy-affected day if wind speed was larger than the sum of corresponding monthly mean wind speed and half of standard deviation.The day was classified as dusty-affected day if PM 10 mass concentration significantly increased with strong winds and visibility was lower than 10 km.As illustrated in Fig. 6, although some particle densities for the affected days were between the sum and difference of corresponding hourly mean value and standard deviation for the other days, this study focused on comparison of mean particle densities between the affected days and the other days.Furthermore, mean particle densities for the other days had minimums in the morning and afternoon and maximum during midday in four seasons, which was in good agreement with the previous studies (Pitz et al., 2003(Pitz et al., , 2008a)).However, particle effective densities for precipitation-, strong windy-, and dusty-affected days did not obey the rule due to effect of weather conditions.The significance of difference between diurnal variations of mean densities for the affected days and the other days was also tested by Student's t-test with significance level of 0.05 (p = 0.05).
The particle densities of precipitation-, strong windyaffected days were higher than mean values for corresponding hours of the other days, indicating the scavenging effects of precipitation and strong wind on particle volume concentrations were more significant than on mass concentrations if particle effective density was defined as the ratio of mass to volume concentration.However, for dusty-affected days the submicron particle densities were lower than mean particle effective densities.The above analyses indicated that impact of dust events was in contrast to precipitation and strong winds.The precipitation and strong wind resulted in daily mean particle effective densities increased by 0.1 g cm -3 and 0.08 g cm -3 , respectively.The capture of aerosol particles by falling raindrop takes place with Brownian and turbulent shear diffusion, inertial impaction, diffusiophoresis, thermophoresis and electrical charge effects.Detailed studies of these processes (Greenfield, 1957;Slinn, 1983) have revealed that there exists a minimum in the collection efficiency of aerosol particles between sizes ranging from 0.1 to 2 µm diameter, which is often called the "Greenfield gap".Aerosol particles in accumulation mode were prone to remain distributed in the atmosphere even after heavy rainfall and snowfall (Chate, 2005;Zhao et al., 2015a).Therefore, precipitation (rain and snow) maybe resulted in increases of effective density of accumulation mode particles.The increased particle density during strongwind days was also possibly due to the prevailing wind from Northeast which brought polluted air masses from Baiyin city.However, effective densities of submicron aerosol particles decreased by 0.21 g cm -3 due to the effect of dust.Chun et al. (2001) also found that number concentrations of aerosol particles in accumulation mode decreased during dust periods, which may be caused by the fact that the increased coarse mode particles during dust events reduced the number concentration of accumulation mode through some processes such as inter-modal coagulation (Jung et al., 2002;Kim et al., 2007), leading to low particle density during dust events.
Until now, there is limited amount of literature describing the effective density of submicron aerosol particles during Chinese Spring Festival in cities in China, especially for northwestern China.So, in order to explore the impact of firework displays on particle effective density in a typical valley city, northwestern China, evolution of average particle densities affected (9 th -10 th and 24 th -25 th February 2013) and unaffected by firework setting off were given in Fig. 7. To reduce effects of traffic count differences between weekdays and weekends and snow, mean particle densities  for 26-27 January, 2-3, 16-17 and 23-24 February and 27-28 January and 3-4 February were compared with 9-10 (the Spring Festival) and 24-25 February 2013 (the Lantern Festival), respectively.As shown in Fig. 7, the particle effective densities were comparable before 18:00, while after 18:00 the particle densities for Spring and Lantern Festivals were much lower and maximally reduced by a factor of 0.4, especially during 18:00 9 th to 02:00 10 th and 20:00-21:00 on 24 th February.The significance of difference between diurnal variations of mean densities for the fireworkaffected days and the days unaffected by firework setting off was also tested by Student's t-test with significance level of 0.05 (p = 0.05).For the two periods mainly affected by firework displays, the mean effective densities of submicron aerosol particles were 1.19 g cm -3 , which was much lower than the mean particle densities for firework-unaffected days, indicating firework setting off led to decrease of submicron particle effective densities.The effective densities decreased by 0.29 g cm -3 due to the effect of firework displays.
Few studies showed that the most affected particles by fireworks were in the accumulation mode range (Wehner et al., 2000;Zhang et al., 2010;Zhao et al., 2014).Hu et al. (2012) and Yin et al. (2015) found that effective density of submicron particles increased with the diameter in urban areas of Beijing and Shanghai with minimum of 1.36 g cm -3 in Shanghai.Therefore, low particle density during firework displays was less affected by particle diameter, which may be mainly related to particle compositions.Additionally, Wang et al. (2007) analyzed chemical compositions of aerosol particles during the Lantern Festival in Beijing and showed that chemical compositions, such as Ba, K, Sr, SO 4 2-and NO 3 -, during bonfire night were five times higher than those during other nights.A similar diurnal pattern was observed between particle density and the contribution of NO 3 -and SO 4 2-to PM 1.0 , suggesting that density change in response to particle compositions.Thus, low particle density during firework displays was mainly related to particle compositions.

CONCLUSIONS
High time resolution aerosol particle number concentrations and size distributions (10 nm-10 µm) were measured using scanning mobility particle sizer (SMPS model 3936, TSI, USA) and aerodynamic particle sizer (APS model 3321, TSI, USA) during 1 September 2012 to 31 August 2013 at urban Lanzhou, northwestern China.The hourly mean particle effective density in the overlapping size region of SMPS and APS was calculated by fitting the APS data to the SMPS data to investigate variations of effective density of submicron aerosol particles and effects of firework displays and weather conditions on particle effective density.Some important conclusions were obtained.
The monthly mean effective densities for ambient submicron particles were found to vary from 1.3 g cm -3 to 1.6 g cm -3 depending on the month of year with the lowest and highest densities 1.31 g cm -3 and 1.62 g cm -3 in November 2012 and August 2013, respectively.Furthermore, the ranges of mean submicron particle effective density were different on workdays, Saturday, Sunday and holidays in different seasons with minimums in the morning and afternoon, especially observed in morning rush hours of weekdays, and a maximums during midday.Additionally, the mean effective density of submicron aerosol particles in spring was (1.36 ± 0.22) g cm -3 , which was the lowest among the four seasons.Compared with other seasons, the variation of submicron particle density was weaker in winter in urban Lanzhou.
The precipitation and strong wind resulted in increases of particle effective densities, and increased by 0.1 g cm -3 and 0.08 g cm -3 , respectively, while effective densities of submicron aerosol particles decreased by 0.21 g cm -3 due to the effect of dust.During the periods mainly affected by firework displays, the effective densities of submicron aerosol particles were 1.19 g cm -3 , which was much lower than the mean particle densities for firework-unaffected days, indicating firework setting off led to decrease of submicron particle effective densities, and decreased by 0.29 g cm -3 .

Fig. 1 .
Fig. 1.A map of Lanzhou and roads and stationary emission sources around the sampling site (the Tianshui and Donggang West Roads are main road in urban Lanzhou with traffic volumes higher than 2500 cars hr -1 , and the Second Thermal Power Plant is also a main central heating company in winter in Chengguan District of Lanzhou City, which is around 2.5 kilometers from our sampling site).

Fig. 2 .
Fig. 2. Monthly mean particle number size distributions (PNSD) during September 2012 to August 2013.D p referred to mobility-equivalent diameter.

Fig. 3 .
Fig. 3. Annual variation of particle effective density during September 2012 to August 2013.

Fig. 5 .
Fig. 5.Diurnal variations of submicron particle densities for workdays, Saturday, Sunday and holidays in different seasons during study period.

Fig. 6 .
Fig. 6.Diurnal variations of mean submicron particle density for precipitation-, strong windy-, and dusty-affected days, and for other days of fall, winter, spring and summer.

Table 1 .
Summary of particle effective densities in various environments in the world."-" and "+" stand for the instruments were used in serial and in parallel, respectively. *

Table 2 .
Summary of particle effective densities in various environments in Beijing, Shanghai and Lanzhou, China a .

Table 3 .
Statistics of particle number concentrations in different size ranges in four seasons.