Introduction

High loadings of particulate matter (PM) can influence air quality, human health, and the regional and global climate (Cao 2012; Wang et al. 2014a; Baumer et al. 2008; Lohmann and Feichter 2005; Pöschl 2006; Squizzato et al. 2017). China has suffered from severe air pollution in recent decades due to the rapid industrialization and urbanization (Bi et al. 2015; Liu et al. 2016; He et al. 2016). Xi’an, located in the northwest of China, is not only the economic and cultural center of the northwest region but also one of the world’s four great ancient capitals of civilization. With a rapid growth in population and transportation, Xi’an is also facing dire PM pollution problems (Cao et al. 2012b; Chen et al. 2016; Wang et al. 2016a).

Information on the chemical composition of particles is vital to tracing the sources of PM, and a direct relationship exists between particle size and chemical composition (Chen et al. 2014; Healy et al. 2013). Numerous studies have been conducted to characterize the physicochemical properties, sources, and atmospheric processing of aerosol particles in Xi’an (Chen et al. 2016; Wang et al. 2013; Wang et al. 2017; Wang et al. 2014b). Filter-based analysis has been the main method of studying PM in Xi’an. For example, Wang et al. (2015) analyzed samples of PM2.5 filters from different sites in Xi’an and discovered that the loaded PM2.5 had eight sources, including coal combustion, motor vehicles, and biomass burning. Zhang et al. (2015) shows that secondary aerosol formation is crucial to haze formation and that relative humidity and stagnant meteorological conditions promote aqueous transformation and increase the loadings of secondary aerosol during haze periods. Traditional offline measurement techniques generally collect samples on filters, which is time-consuming, and potentially decay or degeneration of the chemical species in filters may occur during their transport and storage (Liu et al. 2016). Therefore, a real-time or online measurement tool is urgently launched to investigate the properties of atmospheric aerosols.

Single-particle aerosol mass spectrometry (SPAMS) can provide real-time information on the chemical composition and mixing state of single particle (Cahill et al. 2012; Sultana et al. 2017; Sun et al. 2013; Laskin et al. 2012; Zhang et al. 2014; Yan et al. 2018). Several studies employing SPAMS have been performed in Chinese cities such as Shanghai (Huang et al. 2013; Zhang et al. 2014; Mu et al. 2013; Yang et al. 2012), Guangzhou (Bi et al. 2015, 2011; Zhang et al. 2015), and Beijing (Li et al. 2014; Liu et al. 2016). These studies are beneficial for deep understanding the chemical composition, sources, and mechanism of atmospheric processing of aerosol particles. Studies on single particles have shown the presence of numerous types of particle with complex mixing states in urban areas (Zhang et al. 2013; Hu et al. 2018). The physicochemical properties (e.g., chemical composition, size, and mixing state) of these particle types are unique and influenced by various factors including meteorological conditions and emission sources (Zelenyuk et al. 2008a; Moffet and Prather 2009). Therefore, the behaviors of these particle types in the atmosphere were different; this probably contributes to their distinct effects on the environment and climate.

SPAMS has been employed to study single particles produced by biomass burning, organic particles, and the mixing states of secondary species (Xu et al. 2017; Li et al. 2014; Zhang et al. 2013, 2015). However, the variability of single particles and the mixing states of secondary species with different particle types during heavy pollution events in Xi’an have been rarely reported. In this study, SPAMS was used during October 2015 to characterize the physical and chemical properties of single atmospheric particles in Xi’an. We analyzed the mass spectral and size distributions of single-particle species during pollution periods. This thorough investigation of the different mass spectrum signatures, possible sources of different particle types, and the mixing states of secondary species will improve the understanding of haze formation and evolution in Xi’an.

Methodology

Study site

The largest city in northwest China, Xi’an is located on the Guanzhong Plain at the southern edge of the Loess Plateau. The resident population of Xi’an is more than eight million. Because of rapid economic development and urbanization over past decades, Xi’an has been suffering from serious PM pollution (Cao et al. 2012c; Shen et al. 2011). An intensive single-particle aerosol measurement campaign was implemented during 9–26 October 2015 on the roof (approximately 10 m above ground level) of the Institute of Earth Environment, Chinese Academy of Sciences (34.23° N, 108.88° E). This site is located in a residential area in the southwest of downtown Xi’an (Fig. S1). To the east is Fenghui South Road (approximately 55 m), on which the traffic is moderate, whereas to the south is Keji Yi Road (approximately 20 m), which has less traffic.

SPAMS and data analysis

Particles were measured using a single-particle aerosol mass spectrometer (SPAMS 0515 Model, Hexin Analytical Instrument Co., Ltd., Guangzhou, China). Ambient air was drawn into the spectrometer at a flow rate of 75 mL min−1 using a conductive silicone tube with an inner diameter of 6 mm and length of approximately 1 m. The particles were dried using a silica gel tube before entering the spectrometer to prevent condensation of water vapor in the sampling line. A PM2.5 cyclone (URG Corp., USA) was installed to exclude coarse particles. An additional pump was employed to shorten the residence time of air in the sampling tube. The principle of SPAMS was described in detail previously (Li et al. 2011). Briefly, particles 0.2–2.0 μm in size are effectively drawn into the vacuum system through a ~ 0.1-mm critical orifice resulting from a pressure drop from ~ 760 to 2.2 Torr. The particles were then focused and accelerated to specific velocities on the basis of their aerodynamic diameters, which were determined by two orthogonally orientated continuous neodymium/yttrium aluminum garnet (Nd:YAG) diode laser (λ = 532 nm) separated by a fixed 6-cm distance. Subsequently, the particles were desorbed/ionized at the source region by a pulsed 266-nm Nd:YAG laser. Positive and negative ions were identified simultaneously by the time-of-flight mass spectrometer. In this study, the energy of the desorption/ionization laser was 0.58 ± 0.023 mJ per pulse, which can produce an energy density of 1.06 × 108 W cm−2. Before sampling was performed, a series of standard polystyrene latex spheres (TSI Inc., Knoxville, TN, USA) of known diameters (0.23, 0.32, 0.51, 0.73, 0.96, 1.4, and 2.0 μm) was used to calibrate the measurement of particle aerodynamic diameter. Positive and negative mass spectra were calibrated using lead nitrate and sodium iodide, respectively.

SPAMS converts all acquired mass spectra into a list of peaks at given mass-to-charge ratios (m/z) using several related calculation formulas. The obtained spectral peaks of each single particle subsequently analyzed using YAADA 2.1 (http://www.yaada.org), which is a MATLAB-based software toolkit (version 2012a), for further processing of particle sizes and chemical components. A detailed description of an adaptive resonance theory-based neural network algorithm (ART-2a) is given by Song et al. (1999). Briefly, each particle is first compared to all the peaks of area and m/z. If the particles’ peaks of area and m/z meet the vigilance criterion, these particles are of a similar type. The software repeatedly searches for a particle’s area and m/z to determine if particles have similar mass spectra. The analysis will be complete once the network has performed a given number of iterations. In this study, a vigilance factor of 0.8, learning rate of 0.05, and 20 iterations were set during data processing. A total of 1,120,797 mass spectra out of 4,906,525 sized particles with size ranging from 0.2 to 2.0 μm were obtained by SPAMS during the whole sampling period. Using ART-2a, 930 clusters out of 2534 clusters were obtained by ART-2a and manually merged into six major classes for further analysis. Each particle cluster was named by the major ion pattern in the positive and negative mass spectra.

Complementary data

Air quality index (AQI), PM2.5 mass concentration, temperature, and relative humidity (RH) data were derived from the Online Detection and Analysis Platform of Chinese Air Quality (www.aqistudy.cn). Wind speed (WS) and wind direction (WD) data were collected from Weather Underground (www.wunderground.com).

Results and discussion

Overview

The hourly variation of the PM2.5 mass concentration, AQI, total particle counts, and meteorological conditions during the entire campaign period are displayed in Fig. 1. According to the Technical Regulation on the Ambient Air Quality Index (on trial) (HJ633-2012; Table S1), two typical sampling periods were identified: clean days (CDs), corresponding to 9–13 October (AQI < 100), and polluted days (PDs), corresponding to 14–22 October (AQI > 100).

Fig. 1
figure 1

Temporal profiles (in 1-h resolution) of PM2.5 and visibility and meteorological parameters, including temperature, relative humidity (RH), wind direction (WD), and wind speed (WS), during 9–26 October 2015

Full size image

For PDs, high values of AQIs and PM2.5 were observed frequently, and the highest AQI (231) and PM2.5 mass concentration (181 ± 38 μg m−3) occurred on 20 October at 13:00. The particle count was higher on PDs than CDs. As illustrated in Fig. S2, the total particle count measured using SPAMS was strongly positively correlated with the PM2.5 loading on CDs (r = 0.81), but no correlation was discovered on PDs (r = 0.12). This may be attributed to the limit of the SPAMS detection rate. Given a certain SPAMS detection velocity, because the particle number is much greater on PDs than CDs, numerous particles may go undetected.

As shown in Fig. 1, the diurnal cycles of temperature and RH show a regular anti-correlation throughout the sampling period. On CDs, the RH was relatively low, with an average of 57%. The winds were mainly from the northwest, and the average WS was 2.1 m s−1, with a maximum value of 5.0 m s−1. The meteorological conditions were conducive to the dispersion of pollutants, leading to a low PM2.5 mass concentration (35.3 μg m−3) on CDs. On PDs, the winds were mainly blowing from the east and had low WS (1.6 m s−1), which favored pollutant accumulation. Moreover, the RH (72%) was elevated, which promoted the formation of secondary aerosols through aqueous reactions (Sullivan et al. 2009; Chen et al. 2016). Therefore, the PM2.5 loadings accumulated and increased, with an average concentration of 106 μg m−3. The sudden decrease in PM2.5 concentration on 23 October was caused by precipitation, which results in a scavenging effect on particles. Overall, meteorological conditions were discovered to play a crucial affect in pollution formation.

Classification and mass spectrum characteristics

The average positive and negative mass spectra of the six clusters of particles are collected and shown in Fig. 2 during the sampling period, and the number fraction of each group is summarized in Table 1. Groups were identified on the basis of their chemical characteristics and with reference to the particle types identified in published articles.

Fig. 2
figure 2

Average mass spectra of K-secondary, BB-fresh, BB-aged, dust, metal-containing, OC, ECK, EC–NaK, and ECOC particles during the sampling periods

Full size image
Table 1 Number fraction of single-particle classes during the campaign
Full size table

1.K-secondary particles

This particle type has a positive mass spectrum with strong K+ peaks (m/z = 39, 41) and a negative mass spectrum with strong nitrate (m/z = − 46, − 62) and sulfate (m/z = − 80, − 97) peaks. This particle type has been commonly found in research (e.g., Chen et al. 2016, 2017). Bi et al. (2011) indicated that the positive spectra of K-secondary particles usually contain strong potassium signals, and the negative spectra contain strong sulfate and nitrate signals. Chen et al. (2017) also discovered that the spectra of this group have remarkable signals of potassium, sulfate, and nitrate but weak signals for organic fragment ions in positive spectra.

2.BB particles

The average mass spectrum of BB particles contains strong K+ and Na+ signals and moderate signals of Cn+ (such as m/z = ± 24, ± 36, ± 48, ± 60), CnHm+ (such as 26[C2H2]+, 29[C2H3]+, 39[C3H3]+, 43[C3H7]+, 55[C4H7]+, 57[C4H9]+), 97[HSO4]−, 80[SO3]2−, 62[NO3]−, 46[NO2]−, 42[CNO]−, 35, 37[Cl]−, and 26[CN]−, which are consistent with previously reported identification criteria for BB particles (Chen et al. 2016, 2017; Healy et al. 2010). Although potassium is a marker of BB particles, the instrument employed was extremely sensitive to K+ (Gross et al. 2000). Thus, ion fragments of levoglucosan (e.g., 45[CHO]2−, 59[C2H3O]2−, and 71[C3H3O]−), CNO−, and CN− should also be observed simultaneously to identify BB particles (Silva et al. 1999; Bi et al. 2011). Two types of BB particle were identified: aged BB particles (BB-aged) and freshly emitted BB particles (BB-fresh). The spectrum of BB-aged particles had considerably stronger signals for nitrate and 18[NH4]+ than that of BB-fresh particles; however, the signals for levoglucosan, CNO−, CN−, and Cl− were weaker on PDs than CDs. The larger signal for 18[NH4]+ and smaller signal for levoglucosan in BB-aged particles indicate that such particles may be transformed from fresh BB particles when these fresh particles absorb volatile matter and soluble gases during atmospheric aging processes (Pratt et al. 2011; Zauscher et al. 2013).

3.Metal-containing particles

Particles containing strong signals for metals (e.g., 51[V]+, 52[Cr]2+, 54,56[Fe]2+, 55[Mn]2+, 63,65[Cu]2+, 64,66,68[Zn]2+, and 206,207,208[Pb]2+) and their oxides are defined as metal-containing particles. As shown by the number fractions listed in Table 1, Fe-, Pb-, and V-containing particles contributed higher proportions (13.9–56.3%) to the total number of metal-containing particles than other metal types (0.7–2.2%). Moreover, minor amounts of Zn, Mn, Cu, and Fe were present in the form of hybrid molecules with Pb on CDs (0.6–7.3%) and PDs (0.5–3.7%). The negative mass spectrum of the metal-containing particles has strong signals of mainly nitrate and sulfate. However, some Fe-containing particles have silicate and phosphate signals in the negative mass spectrum, and the spectrum of Pb-containing particles has a strong signal for nitrate but a weak signal for sulfate. Pb-containing particles have also been observed in other urban areas of China (e.g., Beijing, Shanghai, and Guangzhou) and are considered to originate from fossil fuel combustion, industrial processes, and waste incineration (Zhang et al. 2009; Ma et al. 2016). Furthermore, atmospheric Pb may serve as a tracer of local emission sources. Pb–Zn–Cl and Pb–Zn–K–Na particles could be associated with waste incineration (Moffet et al. 2008b) and steel manufacturing (Reinard et al. 2007). V-containing particles are a signature of heavy fuel oil combustion (Zauscher et al. 2011; Healy et al. 2009) and may be emitted from heavy trucks at night. The internally mixed ECOC and Pb (ECOC-Pb) particles may be influenced by anthropogenic activities (Bi et al. 2015; Ma et al. 2016).

4.Dust

Particles with obvious signals of 40[Ca]2+, 56[CaO]+, 79[PO3]−, and 76[SiO3]− are classified as dust particles. It should be noted that CaO+ and Fe2+ both have m/z = 56; they are distinguished by whether the signal corresponds to Fe isotope (m/z = 54) and FeOH+ (m/z = 73) or calcium (m/z = 40) and CaOH+ (m/z = 57) (Silva et al. 2000). The typical m/z signal peaks of other crustal element oxides (e.g., 16[O]−, 17[OH]−, 24[Mg]2+, 27[Al]3+, and 56[Fe]2+), organic fragments (e.g., 43[C2H3O]+), and Cn+ are observed in the average mass spectra of dust particles, and the ion fragments of Cl−, CN−, CNO−, Cn−, nitrate, and sulfate are present in the negative mass spectrum. Fugitive dust is abundant in Xi’an because of the city’s geographical location and dry weather (Cao et al. 2013). Fe-containing particles may originate from dust or industrial emissions. If the spectrum of Fe-containing particles contains signals of silicate and phosphate, these particles can be classified as dust; otherwise, they are defined as metal-containing particles (Dall’Osto et al. 2016).

5.Elemental carbon–related particles

Particles that have positive or negative mass spectra with clear m/z for Cn± are defined as elemental carbon (EC)–related particles. Although EC particles are chemically inert in the atmosphere, they easily act as a core attracting organic or primary inorganic particles during aging processes. Thus, EC ion fragments can be divided into several types, mainly ECK, EC–NaK ECOC, and minor fresh EC particles (Zhang et al. 2013). Notably, unlike that of pure EC particles, the spectrum of fresh EC has weak ion peaks of organic carbon (OC) and sulfate.

6.Organic carbon particles

The signals of OC particles in their positive mass spectrum mainly include those corresponding to various hydrocarbon ion fragments of CnHm+ (e.g., m/z = 27[C2H3]+, 29[C2H3]+, 39[C3H3]+, 43[C3H7]+/[C2H3O]+, 55[C4H7]+, and 57[C4H9]+) and aromatic species (e.g., m/z = 51[C4H3]+, 63[C5H3]+, 77[C6H5]+, and 91[C7H7]+). The OC particle mass spectrum obtained in this study is similar to that presented in previous studies. For example, Dall’Osto and Harrison (2012) discovered signatures of cations (m/z = 27, 29, 43, 51, 57, and 63) in the positive spectrum of organic fragments. Wang et al. (2015) identified 27[C2H3]+ and 43[C2H3O]+ as the major OC peaks, and the median peaks of m/z at 51, 63, and 77 were attributed to aromatic compounds.

Characteristics of particles during pollution periods

The percentages of each type of particle are summarized in Table 1 and list the different chemical compositions on CDs and PDs. On CDs, BB particles were the largest contributor (60.7%) to the total particles, followed by EC-related particles (20.0%), with all other types of particle together accounting for 18.7%. By contrast, EC-related particles were the largest contributor (42.3%) on PDs, followed by BB particles (32.4%) and K-secondary particles (12.4%), with other types of particles together contributing 12.9%. These results indicate that accumulation of BB and EC-related particles is the main reason for pollution in Xi’an. The diurnal variation of the six particle types is plotted in Fig. 3. The highest number fraction of K-secondary particles was observed at 15:00, which indicates that photochemical oxidation plays a critical role in the evolution of these particles. The number fraction of BB-aged particles peaked at 18:00, which may have been influenced by the cooking activity in the suburbs surrounding Xi’an, where biofuels (e.g., wood and crop residues) are essential household energy sources (Zhu et al. 2012). The increased number of BB particles observed during 16:00–23:00 may be attributed to the accumulation of particles under the low boundary layer at night. The ion fragments of levoglucosan were fewer on PDs than CDs, as revealed by the mass spectrum of BB particles; this may indicate that the parts of BB particles that were levoglucosan, 26[CN]−, and 42[CNO]− possibly have decayed or even vanished due to undergoing a series of atmospheric aging processes (Pratt et al. 2011; Li et al. 2014). Alternatively, the detection limitation of SPAMS may have led to fewer ion fragments in this case. For example, Zelenyuk et al. (2008b) demonstrated that thicker secondary coatings on particles can reduce the efficiency of the SPAMS laser for ablating/ionizing core material; Reinard et al. (2007) suggested that secondary species such as sulfate and nitrate are preferentially ionized, masking the presence of other species. BB-fresh particles can be converted into BB-aged particles, and after further aging, parts of these BB-aged particles are merged into K-secondary particles. This partly accounts for the number fraction of K-secondary particles being higher on PDs than CDs and the number fraction of BB particles being lower on PDs.

Fig. 3
figure 3

Diurnal concentration patterns for K-secondary, EC–NaK, ECOC, ECK, BB-aged, metal-containing, dust, OC, Fe-containing, Pb-containing, and NaK particles. The y-axis for the BB-aged particle plot is to the right; that for all other particle types is to the left

Full size image

EC-related particles mainly include ECK (3.8% on CDs versus 21.2% on PDs), ECOC (10.1% versus 15.5%), and EC–NaK (6.0% versus 5.6%). As illustrated in Fig. 3, the number concentrations of ECK and ECOC particles both decreased after 10:00, which is consistent with the superior dispersion conditions after sunrise. By contrast, the diurnal pattern of EC–NaK shows a slightly increasing trend in the afternoon, which indicates different sources of EC–NaK compared with those of ECOC and ECK particles. EC-related particles have multiple sources including biomass burning and fossil fuels (e.g., motor vehicle emissions and coal combustion) (Healy et al. 2011; Wang et al. 2016c). Li et al. (2014) suggested that EC–NaK is likely to originate from local traffic emissions, whereas ECOC may come from both local traffic emissions and biomass burning.

The diurnal variations of the metal-containing and dust particles were similar (Fig. 3). The amount of dust particles began to rise after 08:00 and peaked at 19:00. This diurnal variation was roughly consistent with the pattern of WS. Moreover, the largest value at 19:00 may have been related to resuspension dust caused by rush hour traffic in the evening. Fe-containing particles were present mainly in the metal-containing and dust particles. The diurnal trends for the metal-containing and dust particles were consistent with that of Fe-containing particles, indicating that Fe-containing particles may be a major contributor to metal-containing and dust particles.

SPAMS can determine size for particles of 0.2–2.0 μm, but particles larger than 1.0 μm generally have low transfer efficiency (Li et al. 2014). Figure S3 shows the number size distribution of the sized and hit particles during the campaign. Monomodal log-normal distributions were discovered for sized and hit particles, with both having peak values at ~ 470 nm. The different contributions of each particle type to the size-resolved distribution were identified using the particles’ chemical composition and physical properties. As shown in Fig. 4, the particle types is variable with diameters. The distributions of the carbonaceous types (e.g., BB particles, OC, ECOC, and EC) were concentrated mainly in the condensation mode (< 500 nm), whereas droplet mode (> 500 nm) particles were dominated by inorganic types (e.g., metal-containing particles, dust, and NaK). The size distribution of the K-secondary particles on CDs exhibited a slight peak at 350 nm, but on PDs, the peak value was increased to 510 nm. This may be explained by enhanced accumulation during periods of high PM2.5 loading leading to the growth of particles. The size distribution of BB particles on PDs shows a peak value at 520 nm, which is slightly larger than that for CDs (500 nm).

Fig. 4
figure 4

Size-resolved number concentration of each particle types on CDs and PDs

Full size image

The different types of EC-related particles show distinct size distributions. The distributions of EC–NaK, ECOC, and ECK particles are concentrated mainly in the ranges of 500–600, 400–600, and 500–700 nm, respectively, and these particles have larger number fractions on PDs than on CDs. The mass spectral intensities of secondary species (e.g., 18[NH4]+, 43[C2H3O]+, 62[NO3]−, and 97[HSO4]−) in these types of particles collected on CDs were weak but strong when the particles were collected on PDs, particularly the nitrate and sulfate signals (Fig. 2). ECK and ECOC particles contributed different number fractions to EC-related particles on CDs and PDs. The number fraction of ECK particles sized 500–700 nm was clearly larger for PDs than CDs. Previous studies indicated that a large number of ECK particles in a small size range provide sufficient EC cores for OC formation through the condensation of semivolatile species (Khalizov et al. 2009). Thus, an increase in the number fraction of ECOC was observed on PDs.

The coarse particles collected included dust and metal-containing particles (e.g., Pb and Fe), which had a wider size range—larger than 550–880 nm. This size range is in agreement with particles reported to be present in a suspension of soil dust (Guo et al. 2007). NaK particles represent a minor fraction in the super-micrometer particles, which may reflect crustal element sources (Qin et al. 2012). Fe-containing particles mainly originate from the local iron and steel industry, dust (Dall’Osto et al. 2014; Hu et al. 2018), or fly ash (Zhang et al. 2014).

Mixing state of the particles

The mixing state of primary and secondary species among the major particle types is illustrated in Fig. 5. The criteria for peak searching in the SPAMS dataset were adopted from those in previous studies (e.g., Silva et al. 1999; Moffet et al. 2008a), and a summary is presented in Table S2. The primary species contained chloride, levoglucosan, cyanide, and silicate, whereas the secondary species included C2H3O+, NH4+, nitrate, and sulfate. The peaks of m/z equal to 35, 37[Cl]− and 113, 115[K2Cl]+ were selected to reflect the mixing state of chloride in the atmosphere. K2Cl+ can be used as a tracer of fresh BB particles because aged BB particles contain more K2SO4 and KNO3. In this study, chloride was mixed mainly with BB-fresh particles (35.5% on CDs versus 54.8% on PDs), BB-aged particles (11.2% versus 5.2%), EC–NaK (37.5% versus 22.8%), ECOC (17% versus 9.6%), Pb (64.3% versus 24.6%), dust (24.6% versus 11.6%), and metal-containing particles (29.1% versus 8.5%). Notably, the lower chloride content of some types of particle (e.g., BB-aged, EC–NaK, ECOC, dust, and metal-containing particles) on PDs than on CDs confirmed that chloride can be removed from the particle phase by the uptake of HNO3 or H2SO4 to form HCl gas during aging processes (Silva et al. 1999; Zauscher et al. 2013).

Fig. 5
figure 5

Mixing state of primary and secondary markers associated with the different particle groups for CDs and PDs, indicated by the coloration. LEV levoglucosan

Full size image

The m/z of 45[CHO2]−, 59[C2H3O2]−, and 71[C3H3O2]− usually represents the ion fragments of levoglucosan. The number fractions of BB-aged particles mixed with levoglucosan (32.4% on CDs versus 16.9% on PDs) during different periods were smaller than those of BB-fresh particles (61.3% and 69.9%). This may be explained by levoglucosan decaying through atmospheric oxidization processes (Li et al. 2014; Pratt et al. 2011) and the limitation of ionization efficiency of SPAMS under high particulate loadings, which are common for PDs (Reinard et al. 2007; Zelenyuk et al. 2008b). Noticeably, CN was abundant in the BB particles, especially in the BB-fresh particles for both CDs (84.5%) and PDs (99.2%). Meanwhile, the proportion of CN mixed with inorganic matter (e.g., K-containing, NaK, metal-containing particles, and dust) was larger for CDs than PDs; however, the opposite was discovered for the EC-containing particles (e.g., ECK, EC–NaK, and ECOC). The high percentage of carbon-containing particles mixed with CN may have been caused by the presence of both carbon and nitrogen in the particles (Silva et al. 1999).

The m/z of 60[SiO2]− and 76[SiO3]− is representative of silicon. The number fractions of dust-related and Fe-containing particles mixed with silicon were 30.9% and 26.6% on CDs and 22.5% and 19.1% on PDs, respectively, indicating that silicon mainly originated from fly ash and mineral dust. The higher proportion of silicon in the metal-containing (21.7% on CDs versus 16.5% on PDs) and EC–NaK (9% versus 7.5%) particles may have been caused by an abundance of silicon particles emitted from coal combustion (Bein et al. 2006). Compared with CDs, the number fraction of silicon in those particle types was lower on PDs, indicating that stagnant weather conditions lead to a decrease in the contribution of dust sources.

The existence of secondary species in different types of particles can provide useful information regarding the chemical processes the particles undergo in the atmosphere. The secondary inorganic species are represented by the markers with m/z at 18, − 62, and − 97, and the secondary organic species are reflected by m/z at 43, which can be formed by gas-to-particle conversion or oxidation reactions (Kroll et al. 2012; Ng et al. 2011; Pratt et al. 2011). In this study, ammonium was mainly mixed with 14.3% of OC, 12.0% of ECK, 6.8% of ECOC on CDs and 22.4% of OC, 12.0% of ECK, and 17.7% of ECOC on PDs. Furthermore, more than 40% of the number fractions of BB-aged, OC, and ECOC particles were mixed with C2H3O+. Nitrate was found to be abundant in all particles, especially in the inorganic species (> 80%) (e.g., K-secondary, NaK, metal-containing, and dust particles) during the entire campaign, whereas sulfate tends to be mixed with carbonaceous materials (e.g., OC and ECOC). Compared with CDs, the mixing percent of sulfate and carbonaceous particles was more increased on PDs (from 85% of OC and 84% of OCEC on CDs to 92% and 90% on PDs, respectively), and the proportion of nitrate in inorganic components, such as K-secondary particles (93.3% on CDs versus 93.9% on PDs), NaK (79.8% versus 94.9%), metal-containing particles (87.5% versus 93.4%), and dust (69.8% versus 84.8%) was enhanced. Under polluted environments, SO2 oxidation leads to high sulfate production rates and promotes the formation of organic matter on aqueous particles (Wang et al. 2016b). Studies have indicated that NH4+ tends to neutralize SO42− first and then NO3− (Hara et al. 1999; Li and Shao 2009). For high PM2.5 loading in Xi’an, the NH4+ available may not be sufficient to neutralize the SO42− and NO3− present (Zhang et al. 2011). Under these circumstances, NO3− combines with other cations such as Ca2+, Na+, and K+.

Conclusions

This study employed SPAMS to investigate the size distributions, compositions, and mixing states of single particles collected in October 2015 in Xi’an, China. CDs and PDs were compared on the basis of the AQI and meteorological conditions. The PM2.5 mass concentration was discovered to be 35.3 and 106 μg m−3 on CDs and PDs, respectively. A total of 1,120,797 particles were ionized using SPAMS. The ART-2a neural network algorithm was used to classify particle types, and ~ 96% of the total particles could be clustered into six groups: K-secondary, BB, metal-containing, dust, EC-related particles, and OC particles. BB and EC-related particles were the dominant types on both CDs (60.7%, 20.0%) and PDs (32.4%, 42.3%).

The aging of particles is enhanced under stagnant weather conditions and changes the chemical composition of aerosols, leading to variation in their contributions to the total particles. Weak or absent primary tracers of levoglucosan, Cl, and CN and strong signals of secondary sulfate and nitrate species revealed the increased content of K-secondary and EC-related particles. In general, biomass burning, traffic exhaust, and coal combustion are the possible sources of K-secondary, EC-related, and OC particles. The particles with differing compositions had size dependence. However, most particles were larger on PDs than on CDs due to the particle condensation during aging processes. Clearly, the physical and chemical properties of particles are changed under different pollution conditions. The results of the present study indicate that accumulation of biomass-burning and EC-related particles is the main reason for the pollution in Xi’an. Our study provides clearer understanding of haze formation in the area, especially for those wishing to control pollution effectively in the urban areas of Xi’an.