The measurements of particle hygroscopicity and effective density were conducted from 21 December 2014 to 13 January 2015 at the Department of Environmental Science and Engineering on the main campus of Fudan University (31.30∘ N, 121.5∘ E). It can be considered as a representative urban site for Shanghai. There are many dwelling quarters and commercial blocks in surrounding area. About 400 m away from the measurement site, there is the Middle Ring Line, one of the busiest elevated roads in the city.

At a supersite about 100 m away from the Environmental Building, PM1.0 was monitored using a Thermo Scientific^™ 5030 SHARP monitor. Trace gas pollutants were monitored using Thermo Scientific^™ i-series gas analyzers (43i for SO2, 49i for O3, 42i for NO/NO2/NOx), and meteorological data were monitored using an automatic meteorological station (model CAWS600, Huayun Inc., China) (Yin et al., 2015). The data of PM2.5, PM10, and CO were released by the Shanghai Environmental Monitoring Center. The height of the planetary boundary layer (PBL) was computed online using the NCEP Global Data Assimilation System (GDAS) model (http://ready.arl.noaa.gov/READYamet.php).

Particle size distribution, hygroscopic growth factor (GF), and effective density were measured using a custom-built HTDMA-APM system (Fig. 1). The custom-built HTDMAs mainly consist of two long DMAs (3081L, TSI Inc.), a humidifier (PD-50T-12MSS, Perma Pure Inc.), and a condensation particle counter (CPC, model 3771, TSI Inc.). A detailed description of the HTDMA is available in Ye et al. (2009). In this observation, particle number size distribution in the range of 14–600 nm and hygroscopic growth at 83 % RH for particles with dry diameters of 40, 100, 220, 300, 350, and 400 nm were determined by HTDMA in turn. The determination of effective density by DMA-APM was described previously (Yin et al., 2015; Pagels et al., 2009). Briefly, a combined system consisting of a compact APM (model 3601, Kanomax Inc.) and a CPC (model 3775, TSI Inc.) was connected to the sample tubing through a three-way electrical switch behind the upstream DMA (DMA1). The APM comprises two coaxial cylindrical electrodes rotating at the same angular velocity. Charged aerosol particles of a certain diameter sized by DMA1 are axially fed into the annular gap between the electrodes and experienced an outward centrifugal force from the particle rotating and an inward electrostatic force from the high-voltage field between the electrodes. Particles pass through the APM and are sent to the CPC when the two forces are balanced. The mass of particles that pass through the APM is determined by the rotation rate and the applied voltage. Effective densities for dry diameters of 40, 100, 220, and 300 nm were determined by the method of DMA-APM in this study. The HTDMA-APM was operated alternatively in HTDMA mode and then DMA-APM mode, for every 40 min.

Before the field observation, the HTDMA-APM was calibrated using 40–450 nm NIST-traceable polystyrene latex sphere particles and ammonium sulfate. The measured HTDMA data were inversed with the TDMAinv algorithm to obtain the actual GF distribution. This is because the raw data are only a skewed and smoothed integral transform of the actual GF probability density function (GF-PDF) (Gysel et al., 2009). The hygroscopicity parameter κ was derived from the GF data after inversion with the TDMAinv algorithm according to the κ-Köhler theory (Petters and Kreidenweis, 2007).

A SPAMS (Hexin Analytical Instrument Co., Ltd., China) installed in the same room with the HTDMA-APM system was used to obtain the chemical and size information of individual particles in the range of 0.2–2 µm. Detailed information on SPAMS is available in Li et al. (2011). Briefly, ambient particles are drawn into a vacuum chamber through an aerodynamic focusing lens and accelerated to a size-dependent terminal velocity. Sized particles are desorbed and ionized by the pulsed desorption/ionization laser (Q-switched Nd: YAG, λ=266 nm) at the ion source region. Both positive and negative mass spectra for a single particle are recorded by a bipolar time-of-flight spectrometer. The single particle information was imported into YAADA (version 2.11, www.yaada.org). Based on the similarities of the mass-to-charge ratio and peak intensity, particles were classified using the ART-2a method.

Cascade impactor aerosol samples for offline analysis were collected at the roof platform of the Environmental Building using a 10-stage MOUDI sampler (micro-orifice uniform deposit impactor, model 110-NR, MSP Corp., USA). A detailed description of the sampling, pretreatment, chemical analysis, and quality control of this system is available in Tao et al. (2016). Briefly, cascade impactor samples were collected every 24 h using the PALL7204 quartz filter as the collection substrate. Each filter was weighted with a BP211D electronic balance at 25 ± 1 ∘C and 40 ± 2 % RH. The water extract of each sample was analyzed using an ion chromatograph (Metrohm 883 basic IC plus, Switzerland) equipped with a third-party column heater (CT-100, Agela Corp., China). Seven anions (F-, Cl-, NO2-, Br-, NO3-, SO42-, and PO43-) were resolved using a Metrosep A Supp 5-250/4.0 column at 35 ∘C with an eluent of 3.2 mmol L-1 Na2CO3 + 1.0 mmol L-1 NaHCO3. Six cations (Li+, Na+, NH4+, K+, Ca2+, and Mg2+) were separated by a Metrosep C4-250/4.0 column at 30 ∘C with an eluent of 1.7 mmol L-1 HNO3 + 0.7 mmol L-1 2,6-pyridine dicarboxylic acid.

Figure 2 shows the temporal variations of PM mass loading during the winter observation (21 December 2014 to 13 January 2015). The official data of PM2.5 and PM10 were blank on some clean days. Meteorologically, our measurement was deployed in a typical winter period. The average concentrations of PM1.0, PM2.5, and PM10 were 57 ± 37, 87 ± 67, and 129 ± 78 µg m-3, respectively. About 62 % of hourly averaged PM2.5 concentrations exceeded 75 µg m-3 of the Chinese Grade II guideline (GB 3095-2012), indicating heavy particle pollution in Shanghai during wintertime. The PM episodes exhibited a clear periodic cycle of ∼ 5 days. A similar feature was previously observed in Beijing (Guo et al., 2014). At the beginning of each cycle, the PM1.0 level was below 35 µg m-3. Generally, the difference between the concentrations of PM1.0 and PM2.5 during clean days was less significant than that in haze periods. Occasionally the measured PM2.5 concentrations were larger than those of PM10, possibly due to system error. However, the particle mass concentration began to increase in the next few days, with PM1.0 and PM2.5 peaking at over 100 and 200 µg m-3, respectively. During the end of each PM episode, the change in weather conditions played a key role in the decrease of particle concentration. As shown in Fig. S1 in the Supplement, the prevailing winds on haze days were from the northwest. The prevailing winds during two clean periods (25–27 December and 12–14 January) were northeasterly, bringing clean air mass from the East China Sea. Two cold fronts from the north swept Shanghai on 31 December and 6 January, bringing gale and lower temperatures which favored the dispersion of atmospheric pollutants.

Figure 3 illustrates the daily concentrations of sulfate, nitrate, and ammonium (SNA) as a function of PM1.0 mass loading. In general, the sum of concentrations of SNA increased linearly as the PM1.0 mass loading increased. It is noticeable that the SNA / PM1.0 ratio slightly fluctuated around 0.28, regardless of the pollution level. Because soil dust and sea salt made a negligible contribution to the fine particle mass concentration in this study, the almost constant ratio of SNA / PM1.0 indicates that SNA and carbonaceous aerosols (including soot and organic matter) synchronously increased during the haze events. As the PM1.0 concentration increased, the concentration of nitrate increased more rapidly than sulfate so that it became the most abundant ionic species at PM1.0 > 40 µg m-3. This finding indicates that NOx contributed more to haze formation in Shanghai compared to SO2. Generally, the visibility decreased with the increase in PM concentration, indicating photochemical activity began to weaken as the development of haze events. The large increase in nitrate concentration may be attributable to heterogeneous reaction on the preexisting particles. Nitrate formation is highly dependent on the surface area of preexisting particles and is favored under NH3-rich conditions (Chu et al., 2016). In contrast, Han et al. (2016) reported that the mass ratio of nitrate to sulfate decreased with the increase of PM2.5 level and that the sources of sulfate contributed more to the haze formation in Beijing than mobile sources. This finding suggests that the haze formation mechanism in Shanghai is likely different from that in Beijing. VOCs and NOx are exclusively from local emissions whereas regional transport is a big source of SO2 under stagnant atmosphere due to different atmospheric lifetimes among SO2, NOx, and VOCs (Guo et al., 2014). Considering the relatively smaller contribution of sulfate, our results reveal that the accumulation and secondary transformation of local emissions likely played a dominant role in this haze formation.

Figure 4a displays a box chart of the mean hygroscopicity of each hygroscopic GF distribution for different sizes. Considering all of the GF distributions collectively, the hygroscopicity parameter κ increased with an increase of the dry diameter, with a mean κ of 0.161 at 40 nm and 0.338 at 300 nm. Assuming a two-component system of a model salt (ammonium sulfate, κm=0.53) and an insoluble species (κ=0), the volume fraction of hygroscopic species (ε) can be obtained based on the Zdanovskii–Stokes–Robinson mixing rule. The average ε was 0.3 for 40 nm particles, suggesting that the primary particles or initial growth of freshly generated particles were dominated by non-hygroscopic species. In contrast, the 300 nm particles were extremely aged, with more-hygroscopic species.

Generally, HTDMAs measure dry particles smaller than 300 nm due to technical limitations, and it is common that particle hygroscopicity increases with the increase of particle size (Liu et al., 2014; Swietlicki et al., 2008). The increase of particle hygroscopicity with particle size was attributed to the addition of more-hygroscopic SNA (Swietlicki et al., 2008; Ye et al., 2010). The very few measurements for dry particles larger than 300 nm showed different size dependencies. Gasparini et al. (2006) reported that particle hygroscopicity first increased and then decreased with the increase of particle size, peaking at the diameter of 300 nm. In contrast, Z. J. Wu et al. (2016) reported that particle hygroscopicity increased with particle diameter in the range of 35–350 nm. In this study, the determination size range was extended to 400 nm and the mean κs of 300, 350, and 400 nm particles were nearly equal. We attribute the different size dependencies of hygroscopicity among various measurement sites to the total emissions of SO2 and NOx, gas precursors of hygroscopic sulfate and nitrate. It is noticeable that the 5th percentile hygroscopicity decreased for dry diameter larger than 300 nm, likely due to the presence of the smallest dust particles (Gasparini et al., 2006). The variability of hygroscopicity parameter κ was much greater for 40 nm particles. The particle population with κ<0.1 was attributed to fresh traffic particles (Ye et al., 2013). The considerable percentile of κ<0.1 indicates that the 40 nm particle population was sometimes dominated by near-hydrophobic particles.

Figure 4b displays a box chart of median effective density for different particle sizes. The median effective density varied in the narrow range of ρeff=1.35–1.41 g cm-3 for 40–300 nm particle population. The size dependency of particle effective density varied in the literature. Hu et al. (2012) and Yin et al. (2015) reported that effective density of the particles increased as particle size increased while a opposite trend was observed by Geller et al. (2006) and Spencer et al. (2007). The different trends were attributable to the variable fraction of lower-density mode particles (ρeff<1.0 g cm-3). The densities of the secondarily produced (NH4)2SO4, NH4HSO4, and NH4NO3 are ∼ 1.75 g cm-3. The effective density of organic aerosols varies mostly in the range of 1.2–1.6 g cm-3, depending on their source origins (Malloy et al., 2009; Turpin and Lim, 2001; Dinar et al., 2006). The lower-density particles with ρeff<1.0 g cm-3 were attributable to fresh or partially aged traffic-related particles because the number fraction of the lower-density group in urban area was found to be consistent with the concentration of NO (indicator of traffic) (Levy et al., 2013; Rissler et al., 2014). Although the dominant accumulation-mode particles have an effective density greater than Aitken-mode ones, the presence of a lower effective density group associated with traffic emissions might decrease the mean effective density to a value lower than that of Aitken-mode particles (Levy et al., 2014). Yin et al. (2015) reported that effective density distributions were dominated by a single peak in the previous observation. In contrast, a lower-density peak below 1.0 g cm-3 was often present in this observation, decreasing the mean effective density of externally mixed aerosols.

As shown in Fig. 2, the PM episode from 7 to 12 January was a representative case of severe haze formation and elimination processes. It can be divided into clean (7 January), transition (8 January), haze (9–11 January), and post-haze (12 January) periods. During the transition from the clean period to haze period (7 to 8 January), both PM1.0 and PM2.5 concentrations increased slightly, with an average PM1.0 / PM2.5 ratio of 0.65. A sharp increase in PM2.5 (of 125 µg m-3) was observed from 06:00 to 12:00 LT on the morning of 9 January. During the haze period, the concentration of PM2.5 exceeded 115 µg m-3 (medially polluted level, HJ633-2012) for 63 h. On 11 January, the hourly PM2.5 concentration exceeded 250 µg m-3, corresponding to the severely polluted level.

Figure 5 displays the temporal profile of particle size distribution, along with the measured PM1.0 concentration during the representative PM episode. The calculated PM concentrations (PMcal) were obtained based on the particle size distribution and average effective density of 1.39 g m-3 in the range of 14–600 nm measured in this study. It is noticeable that the temporal trends in mass concentrations of PMcal and PM1.0 are highly consistent. In contrast to the fact that particle size distribution was dominated by nanoparticles during the clean period, the burst of Aitken-mode particles and subsequent continuous growth to approximately 200 nm in diameter was observed three times during the haze period, indicating that the presence of numerous larger particles is likely responsible for the severe particle pollution (Guo et al., 2014). The importance of larger particles in haze formation is also illustrated by the contour plot of the particle volume size distribution. The difference of particle number concentration between transition and haze periods was less significant, whereas the volume concentration increased considerably during the haze period. This feature clearly demonstrates that the haze formation was closely correlated with particle growth and elevated number of larger particles.

Interestingly, the particle mass concentration was sensitive to variations of wind speed and PBL. During the transition and haze periods, the wind speed decreased considerably with insignificant change in prevailing wind (Fig. S1). This finding indicates that outside transportation became less and less significant. It is noteworthy that the temporal evolution of the particle mass concentration was inversely correlated with the PBL height. The decreasing PBL provided a stagnant atmosphere that favored the accumulation of local emissions. This finding reveals that the severe haze pollution was likely triggered by the adverse meteorological conditions. The impact of decreasing PBL height on haze formation can also be evidenced by the variations of trace gaseous species (Fig. S2). During the PM episode, the concentrations of NO2, SO2, and CO displayed variation trends similar to that of the particle concentration. The fluctuations of trace gas concentrations were caused by primary emission and secondary processes. Noticeably, the concentration of NO increased dramatically in rush hours during the haze period whereas it fluctuated slightly during the clean period, indicating that local emissions were easily accumulated under stagnant atmosphere. In addition, the maximum concentration of O3 remained considerably higher during daytime, whereas it decreased significantly at night. The most plausible explanation is that O3 was consumed rapidly by the accumulated trace gases, such as NOx, and VOCs.

Figure 6 shows the averaged hygroscopicity and effective density for different pollution periods of the PM episode. Regardless of the pollution period, the nearly hydrophobic particles were externally mixed with some hygroscopic particles. During the clean period, the more-hygroscopic particles dominated the 40 nm particle population, indicating that the near-hydrophobic primary particles were rapidly dispersed due to atmospheric dilution. The number fraction of the near-hydrophobic group for different sizes increased as the PM episode developed, indicative of the increasing accumulation of local emissions. Notably, the increase of the near-hydrophobic particles with the evolution of the PM episode become less significant as particle size increased, indicating that primary emission exerted a more significant impact on smaller particles than on larger ones. The median diameter of nascent traffic particles from various gasoline sources ranged between 55 and 73 nm with an average of 65 nm (Momenimovahed and Olfert, 2015). Therefore, the number fraction of the near-hydrophobic particles larger than 200 nm is not sensitive to the accumulation of traffic emissions.

Interestingly, the variations of particle effective density for different sizes are in good agreement with the hygroscopicity. The dominant peak of effective density distribution appeared at ρeff=∼1.5 g cm-3 for 40 nm particles in the clean period, indicating that they are highly aged with hygroscopic inorganic salts (Yin et al., 2015). As the episode developed, the mean density shifted to lower values, indicating the increasing contribution of lower-density carbonaceous materials. The averaged density distribution was broadened as the episode developed, suggesting that it could be deconvolved into two groups and that the number fraction of the low-density group increased. This finding revealed that the lower-density particles are less hygroscopic whereas the larger-density group corresponds to the more-hygroscopic one. In addition, the variations of hygroscopicity and effective density coincided with the evolution of PBL height, indicating that the increasing accumulation of local emissions due to adverse atmospheric conditions is likely responsible for the enhancement of those near-hydrophobic and lower-density particles.

Figure 7 displays the temporal profiles for contributions of EC (including bare EC and OC-coated EC), OC, sulfate, and nitrate determined by SPAMS. Obviously, the relative contribution of nitrate increased as the episode developed. In contrast, the relative contribution of sulfate displayed an opposite trend. This feature is comparable with the aforementioned results of SNA, thus further highlighting the important role of nitrate in haze formation in Shanghai. The number fraction of EC particles generally increased during the haze period, peaking at midnight on 9 and 10 January. It should be pointed out that the measured number fraction possibly underestimated the contribution of EC particles because the dominant size range of fresh traffic particles is below the detection limit of SPAMS (0.2–2.0 µm). This finding provides good support for the increase of near-hydrophobic and lower-density particles as the episode developed. Niu et al. (2016) reported that the number ratio of secondary particles to soot in haze samples was higher than that collected in the clean days in Beijing. Our finding is comparable to their results. In contrast, the number fraction of pure OC decreased during the pollution event. The possible explanation is that the condensation of organic matter was favored on the large amount of preexisting EC particles or that photo-oxidation of VOCs was minimized due to lower solar radiation.

As shown in Fig. 5, three “banana-shaped” evolutions of the particle size distribution were identified in the representative PM episode. The banana-type contour plot of particle size distributions is a typical characteristics of NPF events and traditionally regarded as one of the most important criteria for identifying NPF (Xiao et al., 2015; Dal Maso et al., 2005; Levy et al., 2013; Zhang et al., 2012). Atmospheric NPF is often defined by the burst of nucleation-mode particles and subsequent growth of the nuclei to larger particles (Zhang et al., 2012; Kulmala et al., 2012). Gas-phase sulfuric acid produced via oxidation of SO2 by OH radical plays a dominant role in the NPF events. NPF is typically completely suppressed when preexisting particles is abundant, because gas-phase sulfuric acid is rapidly lost to the surfaces of preexisting aerosols (Zhang et al., 2012). In addition to sulfuric acid, low-volatility organic species and interaction between sulfate and organics are important for NPF (Zhang et al., 2004; Zhao et al., 2009). However, the possibility of NPF can be ignored in this study due to the absence of the burst of nucleation-mode particles and the high concentration of PM1.0. The burst of Aitken-mode particles in the current study may be attributable to rapid accumulation of traffic emissions during rush hours under stagnant atmospheric conditions. The banana-shaped particle growth in the time evolution of particle size distribution from the Aitken-mode size range to accumulation-mode size range was primarily due to coagulation and condensation processes. This feature provided an excellent opportunity to reveal the chemical mechanism of particle growth during the PM episode.

The first banana-shaped evolution of the particle size distribution occurred from approximately 05:00 to 15:00 LT on 9 January, with an increase of the particle number concentration (Ntotal) from 1.7×104 to 3.4×104 cm-3 followed by a decrease until 17:00 LT (Period 1). The second banana-shaped evolution occurred from approximately 18:00 LT on 9 January to approximately 12:00 LT on 10 January (Period 2). The Ntotal increased from 2.1×104 to 4.2×104 cm-3 within 3 h, followed by gradual decrease of Ntotal in contrast to a continuous increase of the particle mass concentration. During the growth process, the mode diameter of the particle population increased from below 40 nm to approximately 200 nm. The third banana-shaped evolution began in the evening rush hours on 10 January, with the continuous increase of PM mass concentration for 12 h (Period 3). The latter two banana-shaped evolutions lasted long enough to trace the changes in hygroscopicity and effective density due to particle growth.

Figure 8 illustrates the evolution of particle hygroscopicity and effective density during periods 2 and 3. During the initial stage, the measured GF and effective density distributions were both bimodal, with a dominant peak at GF = ∼ 1.0 and ρeff=∼1.0 g cm-3, respectively. In a previous study, we found that the number fraction of near-hydrophobic particles varied with the traffic exhaust (Ye et al., 2013). Moreover, laboratory studies showed that the effective density of 50 nm vehicle particles was approximately 1.0 g cm-3 (Olfert et al., 2007; Park et al., 2003; Momenimovahed and Olfert, 2015). These findings indicate that the initial burst of Aitken-mode particles is attributable to the presence of enhanced traffic-related emissions. In contrast, the number fraction and GF of the more-hygroscopic group increased with the growing particle size, indicating the addition of hygroscopic inorganic species. The variation of the effective density of the particles was similar to that of the hygroscopicity, indicating the increase of high-density materials. In general, inorganic sulfate and nitrate are more hygroscopic and denser than soot particles or organic aerosols (Yin et al., 2015). These findings suggest that secondary sulfate and nitrate increased with the growing particle size, indicating the importance of the conversion of SO2 and NOx in particle growth. This conclusion is supported by the largest SNA concentration in PM1.0 during the PM episode (31.3 µg m-3 on 10 January and 23.8 µg m-3 on 11 January). Considering that the concentration of nitrate was much higher than that of sulfate during the haze event, the increase of hygroscopicity was dominated by the addition of nitrate.