The field campaign was conducted in the environment monitoring station of Nanjing Olympic center (32∘0′33.00′′ N, 118∘44′9.53′′ E) from 13 to 29 April 2015. Details of the sampling site are shown in Fig. S1 in the Supplement. The site was surrounded by residential buildings, close to a few urban arterial roads (∼ 85 m northwest of Huangshan Road, ∼ 200 m northeast to Mengdu Street and ∼ 425 m southwest of Xinglong Street). There are also a restaurant (∼ 50 m), a student cafeteria (∼ 300 m), and the Nanjing Cigarette Factory (∼ 480 m southeast) around the site.

The sampling inlet was installed outside the fifth floor of the building (∼ 12 m above the ground), with a PM2.5 cyclone (URG Corp., Chapel Hill, NC, USA) to remove coarse particles. Ambient particles were dried (RH < 10 %) via a diffusion dryer filled with silica gel before entering into the SP-AMS. The sampling line (∼ 2 m long) was assembled using stainless steel tubing and proper fittings. Air flow was controlled at around ∼ 5 L min-1, with a flow rate into the SP-AMS at ∼ 80 cm3 min-1.

The SP-AMS can measure non-refractory (NR) PM1 components including ammonium, nitrate, sulfate, chloride, and organics similar to other types of AMS via a thermal Tungsten heater. Moreover, it can also measure rBC and inorganic/organic species that coated on the rBC cores, as it is equipped with an intracavity Nd:YAG laser vaporizer (1064 nm) (Onasch et al., 2012). During this campaign, the instrument was switched between “laser on” and “laser off” settings, and between V-mode (better for mass quantification) and W-mode (better chemical resolution, ∼ 5000 in this study), with one cycle including six menu settings (M1: laser on V-mode; M2: laser off V-mode; M3: laser on W-mode; M4: laser off W-mode; M7: laser on PToF-mode; M8: laser off PToF-mode). Each menu was set to 2.5 min, thus a full running cycle lasted for 15 min. The PToF-mode was under V-mode, but was tuned in particular for measuring particle sizes. The Tungsten heater was always turned on and kept at ∼ 600 ∘C.

The SP-AMS, in conjunction with a scanning mobility particle sizer (SMPS) (TSI inc., Shoreview, MN, USA) was calibrated for mass quantification (e.g., ionization efficiency) using size-selected (250 and 300 nm) monodisperse ammonium nitrate particles following the procedures detailed in Jimenez et al. (2003). Pure ammonium sulfate was used to determine the relative ionization efficiency (RIE) of sulfate to nitrate (Setyan et al., 2012). Quantification of rBC was calibrated using Regal Black (REGAL 400R pigment black, Cabot Corp.) particles according to the procedures reported in Onasch et al. (2012). Note that the aqueous solution of Regal Black was sonicated during calibration to maintain a relative stable aerosol flow. RIEs of ammonium, nitrate, sulfate, chloride, organics and rBC were determined to be 3.15, 1.05, 1.20, 1.3, 1.4, and 0.33, respectively. On the other hand, particle sizing was calibrated using standard polystyrene latex (PSL) spheres (Duke Scientific Corp., Palo Alto, CA, USA) across 100–700 nm range. Flow rate was also calibrated prior to the measurement.

Concentrations of gaseous species, e.g., carbon monoxide (CO) (Model T300, Teledyne API, USA), ozone (O3) (Model EC9810, Ecotech Pty Ltd, Australia), nitrogen dioxide (NO2) and sulfur dioxide (SO2) (Model LGH-01, Anhui Landun, China), and meteorological data including air temperature (T), relative humidity (RH), visibility (km), wind speed (WS) and wind direction (WD) were acquired at the same site. PM2.5 and PM10 mass concentrations were also recorded (BAM-1020, Met One Instruments, Inc., USA), in parallel with the SP-AMS measurement.

The SP-AMS data were post-processed by using the Igor-based standard ToF-AMS Analysis Toolkit SQUIRREL v1.56D and PIKA v1.15D, available at http://cires1.colorado.edu/jimenez-group/ToFAMSResources/ToFSoftware/index.html. Note all mass concentrations reported here were calculated from the HR fitted results on V-mode data. A collection efficiency is typically used to account for the particles that are not collected and measured by the instrument; this is due to the particles lost during passage through inlet, time-of-flight chamber and bouncing from the vaporizer. For the SP-AMS, the CE of the laser vaporizer is mainly governed by particle divergence, while for the Tungsten vaporizer, the CE is governed mainly by the bouncing effects (Matthew et al., 2008). A CE value of 0.5 is valid and used commonly for the AMS measurements for most environments (Canagaratna et al., 2007). Nevertheless, Middlebrook et al. (2012) further investigated this issue, and found that high aerosol acidity, high mass fractions of ammonium nitrate, and high sampling line RH can increase the CE, and composition-dependent CE parameterization. For our data set, we found that the composition-dependent CE rather than a constant CE = 0.5 has negligible effects on the quantification of aerosol species, as the particles were neutralized (Fig. 3a), the mass fraction of ammonium nitrate were < 40 % in almost all cases, and also the sampling line RH was below 10 %. And in fact, the PM1 mass concentrations quantified by using the composition-dependent CE correlate a bit worse with the PM2.5 concentrations than ones using CE = 0.5. Thus, a constant CE of 0.5, in keeping with many other AMS studies, was employed for this data set.

Unless specified, the concentrations of ammonium, sulfate, nitrate, chloride, and organics are from M2 setting (Tungsten vaporizer only), while the rBC data is from M1 setting (dual vaporizers: Tungsten + laser) in this paper. The meteorological data (RH, T, WS, WD, and visibility), concentrations of gas-phase species (CO, NO2, SO2, and O3) and PM2.5 were averaged into hourly data for comparisons with the SP-AMS data. The data reported are at local time, e.g., Beijing (BJ) Time.

Positive matrix factorization (PMF) (Paatero and Tapper, 1994) was applied on the high-resolution mass spectra (HRMS) of organic aerosol (OA) obtained under laser off W-mode (M4 setting) to elucidate the OA sources/processes. We used the PMF evaluation tool version 2.08A (downloaded from: http://cires1.colorado.edu/jimenez-group/wiki/index.php/PMF-AMS_Analysis_Guide) (Ulbrich et al., 2009) to investigate the PMF results by varying the number of factors (from 2 to 8 factors) and rotations (“fpeak”, from -1 to 1 with an increment of 0.1). Only ions with m/z less than or equal to 180 were included in the analyses. Following the instruction detailed by Zhang et al. (2011), the 4-factor solution (at fpeak=-0.1) was chosen as the optimal solution, as the 3-factor solution cannot separate the hydrocarbon-like OA (HOA) and cooking OA (COA) (Fig. S2). For the 5-factor solution (Fig. S3), Factor 2 and Factor 4 are clearly a split of the SVOOA from the 4-factor solution (r2=0.89 and slope of 1.05, Fig. S3c); Factor 2 of 5-factor solution also shows much weaker correlations with nitrate than SVOOA of 4-factor solution does (r=0.07 vs. 0.49). A summary of the key diagnostic plots are provided in Fig. S4. Detailed discussion of the PMF results is presented in Sect. 3.5. Note we found no significant differences between the PMF source apportionment results from the HRMS of OA (without rBC) obtained with dual-vaporizers setting (M3 setting) and current results (M4 setting, Tungsten vaporizer only), as the OA HRMS acquired under these two circumstances were very similar (details in Sect. 3.4).

Note that the elemental ratios shown throughout the paper were all calculated based on the method proposed by Aiken et al. (2008) (referred to as A-A method). Recently, Canagaratna et al. (2015) improved this methodology by using specific ion fragments as markers to calculate the O / C and H / C ratios (referred to as I-A method). The I-A method increased the O / C ratio, H / C ratio, and the OM / OC ratio from the values calculated from the A-A method, on average, by 28, 10, and 8 %, respectively (Fig. S5). In this work, we used the results from the A-A method for consistency and comparisons with previous AMS measurements.

The temporal variations of meteorological parameters, concentrations of the gas pollutants, concentrations, and mass fractions of different PM1 components, and the PM2.5 mass loadings (from Met one BAM-1020) over the sampling period are illustrated in Fig. 1. During this study, the mean temperature was 18.5 ∘C, RH on average was 64 %, and wind predominantly blew from the southeast and southwest (Fig. S6). The SP-AMS PM1 concentrations ranged from 5.1 to 97.9 µg m-3, with an average of 28.2 µg m-3. Note this average PM1 concentration is significantly lower than those observed during summer (38.5 µg m-3), autumn (46.4 µg m-3) and winter (89.3 µg m-3) (Y. J. Zhang et al., 2015, 2016), showing that the air during springtime in Nanjing is cleaner than in other seasons. The variations of PM1 concentrations also match very well with PM2.5 concentrations (Pearson's r2=0.72), and on average PM1 accounts for ∼ 54 % of the PM2.5 mass. This ratio appears to be a bit low, likely due to the uncertainty of CE of the SP-AMS.

The average PM1 composition is shown in Fig. 2a. The most abundant component is found to be organics (45.0 %), following by sulfate (19.3 %), nitrate (13.6 %), ammonium (11.1 %), rBC (9.7 %) and chloride (1.3 %). Figure 2b further shows changes of the PM1 chemical compositions in different concentration bins. It can be seen that although most PM1 mass loadings are within 10–40 µg m-3, high loading periods tend to have higher mass contributions from organics and rBC, and less contributions from secondary inorganic species, indicating that high PM events were influenced significantly by local fresh emissions.

The molar ratio of inorganic anions (sulfate, nitrate, and chloride) to cations (ammonium) is 1.05 (Fig. 3a) (Zhang et al., 2007b). Considering that a small fraction of sulfate, nitrate, and chloride are possibly associated with metal cations, such as Na+, K+, and Ca2+, etc. it can be concluded that the NR-PM1 was neutral throughout the study. On the other hand, the molar ratio of inorganic anions to ammonium is on average 1.17 (Fig. 3b) when dual vaporizers are on. This may be partially due to variations of ionization/collection efficiencies of the measured species as the addition of a laser beam may change the distribution of vaporized species inside the ion chamber, and also because of the detection of sulfate, nitrate, and chloride bonded with metal cations under the dual vaporizers. These species do not evaporate on the Tungsten vaporizer under the laser-off mode. Indeed, more metal signals were observed with the dual vaporizers, as shown in Fig. S7.

Figure 2c shows the average diurnal changes of organics, sulfate, nitrate, chloride, and rBC. Sulfate concentrations are slightly higher during daytime than during nighttime, indicating a significant contribution from photochemical reactions. Sulfate also shows the least variations among all species, reflecting its regional behavior. Except for sulfate, all other species present a dual-peak pattern, with one peak in the early morning and another one in the early evening. The peaks of rBC and organics are likely due to local traffic/cooking activities (see details in Sect. 3.5), while the behavior of nitrate is likely driven by the thermodynamic gas-particle partitioning: NH3(g) + HNO3(g) ↔ NH4NO3(p) as it shows good anti-correlations with the diurnal changes of temperatures (r=-0.72 for nitrate vs. T). The good correlations between the diurnal cycles of nitrate and RH, in particular during nighttime, suggest a nighttime formation pathway of nitrate, e.g., N2O5+ H2O = 2HNO3 and HNO3+ NH3= NH4NO3. Furthermore, we calculated the diurnal variations of the equilibrium constant of NH4NO3 (Kp,AN) (Seinfeld and Pandis, 2006; Young et al., 2016) in Fig. 2c. The Kp,AN displays a similar trend as nitrate (r=0.68), providing strong evidence that nitrate variations were governed mainly by the thermodynamic equilibrium. Chloride shows similar behavior as nitrate, indicating it is driven by the equilibrium NH3(g) + HCl(g) ↔ NH4Cl(p) as well (r=-0.76 for chloride vs. T). Therefore, when temperature rises, more NH4NO3 and NH4Cl can dissociate into gaseous NH3, HNO3 and HCl, mass loadings of particle-phase nitrate and chloride decrease correspondingly, and vice versa.

In order to further elucidate the formation processes of sulfate, we calculated the oxidation ratios of sulfur (fS) (Fig. 4a), defined as fS=nSO42-/ (nSO42-+nSO2) (Xu et al., 2014), indicating the conversion of SO2. Here nSO42- and nSO2 are the molar quantities of particle-phase sulfate, and gas-phase SO2, respectively. Diurnal variations of fS and RH are presented in Fig. 4b, and Fig. 4c shows variations of sulfate and nitrate concentrations with RH. The diurnal profile of fS shows a negative correlation with that of RH (r=-0.52), and mass concentrations of sulfate even drop under high RH conditions, indicating an insignificant role of aqueous-phase processing for sulfate formation during this campaign. On the other hand, the fS reaches a maximum around 3 pm. Note the afternoon rise of fS and sulfate may be affected by the down mixing of sulfate formed earlier, however, since concentrations of all other aerosol species that mix with sulfate decrease significantly, we postulate that the increase of fS likely suggest the photochemical production of sulfate in the afternoon.

The campaign-averaged mass-based size distributions, fractional contributions, and diurnal size distributions (image plots) of the major PM1 species are depicted in Fig. 5 (temporal variations of the mass-based size distributions of these PM1 species over the whole measurement period are provided in Fig. S8). Note the size distribution of rBC in these plots were scaled from the size distribution of m/z 24 (C2+), as other major rBC ion clusters may be heavily influenced by other ions, such as C+ signal but from organics at m/z 12 (C+), HCl+ signal at m/z 36 (C3+), SO+ signal at m/z 48 (C4+), C2H4O2+ signal at m/z 60 (C5+). It also should be noted that, although the AMS is able to capture the bulk of atmospheric accumulation mode particles (Canagaratna et al., 2007), the right side of size distributions may be affected by the incomplete transmission of larger particles limited by the SP-AMS inlet (in particular, the supermicron ones).

As can be expected, all inorganic species (sulfate, nitrate, chloride, and ammonium) display a unimodal distribution with an accumulation mode peaking ∼ 550 nm (vacuum aerodynamic diameter, Dva; DeCarlo et al., 2004), since they were mainly formed from secondary reactions. The organics has a much broader size distribution across from ultrafine (< 100 nm) to supermicron meter range, with a small sub-peak centering ∼ 120 nm in addition to the major peak at ∼ 440 nm, indicating influences from both primary and secondary emissions. On the contrary, size distribution of rBC behaves very differently from other components, which peaks at 90–200 nm range, reflecting clearly that it mainly originates from primary emissions. Overall, the small particles predominantly consist of organics and rBC, which together account for more than 90 % of the ultrafine particle mass. Mass contributions from inorganic species increase significantly with the increase of particle size, and they dominate masses of particles larger than 400 nm (Fig. 5b).

In line with the diurnal mass loadings of the PM1 species shown in Fig. 2c, the diurnal size distribution of sulfate is generally stable, with masses concentrating in the 400–700 nm range throughout the day (Fig. 5c); while the size distributions of nitrate, chloride, and organics present clear enhancements in the 300–700 nm range during early morning and early evening due to increased mass concentrations of these species during these two periods. The size distribution of rBC is also enhanced during the morning and evening hours, but it extends to a much smaller size range (< 100 nm).

In order to figure out the major species that are responsible for the visibility degradation, here we employed the IMPROVE method to reconstruct the light extinction coefficients (bext). bext values are derived from the measured visibility: bext=3.91/Vs (Kong et al., 2015), where Vs stands for the visibility (in meter). The following IMPROVE formula (Yang et al., 2007) was used: bext=3f(RH){[(NH4)2SO4]+[NH4NO3]+[NH4Cl]}+4[OM]+10[BC]+1[soil]+10, where f(RH) is a RH-dependent empirical coefficient which considers the effects of water uptake by inorganic salts on the light extinction; the f(RH) values used here were taken from Malm and Day (2001), which were regressed from the Great Smoky data set. [(NH4)2SO4], [NH4NO3], [NH4Cl], [OM], and [BC] represent the mass concentrations of ammonium sulfate, ammonium nitrate, ammonium chloride, organics and black carbon directly from the SP-AMS measurements (in µg m-3) ([(NH4)2SO4] =1.375× [SO42-], [NH4NO3] =1.29× [NO3-] and [NH4Cl] =1.51× [Cl-]). Since the SP-AMS cannot accurately measure soil components (e.g., various metals/metal oxides/metal salts), the term [soil] was set to zero during calculations.

By using this method, the reconstructed visibilities match reasonably well with the measured values (r2=0.50) as shown in Fig. 6a. Figure 6b shows the time series of the measured and reconstructed extinction coefficients throughout the whole sampling period. It should be noted that, on average, the measured PM1 species are only able to explain ∼ 44 % of the light extinction. This is likely due to that (1) as shown earlier, the SP-AMS measured PM1 only occupies ∼ 54 % of the PM2.5 mass; (2) we did not include contributions from soil components, coarse particles and also some gas-phase species (such as NO2); (3) although the influences of water are partly included through f(RH) for inorganic salts, the water uptake by organic species are not considered explicitly, which can be significant especially for the SOA under high RH conditions (Duplissy et al., 2011; Denjean et al., 2015). Indeed, as shown in Fig. 6a, reconstructed visibilities appear to deviate more significantly from the measured visibilities under high RH than ones under low RH conditions, suggesting the importance of particle-bounded water on visibility degradation. The pie chart in Fig. 6b presents the average relative contributions of different components to the light extinction of PM1. The largest contributor is organics which accounts for 37.7 %, followed by ammonium sulfate (25.1 %), rBC (20.7 %), ammonium nitrate (15.1 %) and a minor contributor of ammonium chloride (1.4 %).

The unique laser vaporizer of SP-AMS allows it to detect rBC and species coated on the rBC core including both non-refractory and refractory organics; thus comparison between the OA mass spectra obtained with dual vaporizers and Tungsten vaporizer settings can infer some information regarding the chemical features of refractory organics, which were unable to be determined by any other types of AMS. As shown in Fig. 7a and b, the OA obtained with the dual-vaporizers setting have slightly higher oxygen-to-carbon (O / C) ratio (0.28 vs. 0.27), nitrogen-to-carbon (N / C) ratio (0.033 vs. 0.032) and lower hydrogen-to-carbon (H / C) ratio (1.50 vs. 1.52) than the corresponding elemental ratios of OA obtained with the Tungsten vaporizer only. This result indicates that refractory organics are likely more oxygenated than the non-refractory organics, and for this data set it is mainly due to a higher fractional contribution from C2H3O+ (see the inset of Fig. 7a). This is different from the results on laboratory-generated nascent soot, where larger fCO2+ (i.e., the fraction of total organic signal contributed by CO2+) was observed with the dual-vaporizers setting, indicating the variability of the chemical compositions of refractory organics.

It should be noted that, accurate determination of refractory organics is very difficult because: (1) a large portion of refractory organics cannot be detected by the SP-AMS if they did not coat on rBC cores; (2) to accurately measure the species only coated on rBC cores, the Tungsten vaporizer has to be physically removed, otherwise the vaporizer temperature is still around 150 ∘C even its power is turned off, and the non-refractory organics that do not coat on rBC cores can still be measured and complicates the analyses; (3) the CE and IE values for different species may vary under different vaporizer settings, so that direct subtraction of organics measured under Tungsten-only setting from the organics measured under the dual-vaporizer setting may not represent the real refractory organics; (4) some ions measured under dual-vaporizer setting are likely induced by the laser itself rather than the 70 ev electron impact. For example, a series of fullerene-like carbon clusters can be generated by the laser itself, even though they do not really exist in the atmosphere (J. Wang et al., 2016; Onasch et al., 2015). This laser-induced ion formation scheme may work for other organics, thus making it even more difficult for identifying the refractory organics. Further studies are essential to investigate this issue.

Overall, the O / C ratio (0.27) of OA in Nanjing during springtime is a bit lower than those observed at other urban locations in China – for instances, 0.30 in Shenzhen (He et al., 2011), 0.31 in Shanghai (Huang et al., 2012), 0.33 in Lanzhou (Xu et al., 2014) and 0.34 in Beijing (Zhang et al., 2014), and much lower than those at rural sites – for instances, 0.47 in Kaiping (Huang et al., 2011) and 0.59 in Changdao (Hu et al., 2013). As O / C ratio is a good indicator of the aging degree of OA, the relatively low O / C level indicates a significant contribution from fresh emissions in Nanjing aerosols during springtime. Accordingly, the non-refractory OA (pie chart in Fig. 7b) is dominated by hydrocarbon CxHy+ ions (51.2 %) rather than the oxygen-containing ion fragments (37.4 % of CxHyO1+ and CxHyO2+).

The scatter plot of f44 (mass fraction of m/z 44 to the total OA) vs. f43 (mass fraction of m/z 43 to the total OA) (a.k.a., triangle plot) (Ng et al., 2010) was often used to investigate the oxidation degrees of OA. As presented in Fig. 8, most OA reside in the bottom end of the triangular region, again pointing out the less-oxygenated behavior of the OA. Since the HRMS can separate different ions at the nominal m/z, we also examined the fCO2+ vs. fC2H3O+ space and illustrated it in Fig. S9 - many OA locate outside the triangular region, yet still close to the bottom. Moreover, m/z 60 (mainly C2H4O2+) is a significant fragment ion of levoglucosan, which is well known as the biomass burning aerosol tracer (Alfarra et al., 2007). However, as f60 (mass fraction of m/z 60 to the total OA) is very low in OA (average ±1σ=0.4±0.06 %), indicating no biomass burning influences on the OA properties during springtime in Nanjing.

In order to further elucidate the sources and evolution processes of OA, we performed PMF analyses and identified four OA components, including two primary OA (POA) factors – a traffic-related hydrocarbon-like OA (HOA) and a cooking-related OA (COA), and two secondary OA factors – a semi-volatile oxygenated OA (SV-OOA) and a low volatility OOA (LV-OOA). Details about their characteristics are discussed below.