Supplement of Formation and characteristics of secondary aerosols in an industrialized environment during cold seasons

1Jiangsu Key Laboratory of Atmospheric Environment Monitoring and Pollution Control, Collaborative Innovation Center of Atmospheric Environment and Equipment Technology, School of Environmental Sciences and Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China 2College of Chemistry and Environmental Engineering, Jiangsu University of Technology, Changzhou 213001, China 3Nanjing Tianbo Environmental Technology Co., Ltd, Nanjing 210047, China


Introduction
Aerosol particles can significantly affect the earth's climate (e.g., Pöschl, 2005;Carslaw et al., 2010), air quality (e.g., Chan and Yao, 2008;Shen et al., 2017;Heal et al., 2012) and human health (e.g., Shiraiwa et al., 2017;Pope and Dockery, 2006;Hu et al., 2017), etc.However, such effects are highly uncertain, in a large part due to varying physicochemical properties of the particles, including sizes, compositions, and sources, etc.The secondary species, which often dominate the fine particle mass (e.g., Zhang et al., 2007;Huang et al., 2014), are particularly less understood.Formation of secondary components, including both inorganic and organic species, is very complex and dependent heavily upon atmospheric environments and meteorological conditions.For example, recent studies demonstrate that nitrate formation can be governed by multiple mechanisms in different seasons and locations of China (Ge et al., 2017a;Yang et al., 2017).For sulfate, it is well known that the transformation of SO2 into sulfate can occur in both gas-phase and aqueous-phase via reactions with multiple oxidants (Seinfeld and Pandis, 2016).
Similarly, secondary organic aerosol (SOA) can be produced by both gas-phase and aqueous-phase (or multiphase/heterogeneous) reactions too (e.g., McNeill, 2015;Herrmann et al., 2015;Blando and Turpin, 2000;Lim et al., 2010).For examples, Zhang et al. (2018a) points out that both photochemical and aqueous-phase reactions can be important for the secondary organic carbon (SOC) in Shanghai during summertime, and aqueous-phase processing is more important to the SOC nighttime formation; Saffari et al. (2016) also shows the nighttime SOC formation in Los Angeles, while Ye et al. (2017b) shows that photochemical oxidation is important to the daytime SOC formation in Changzhou, China.
Due to a large variety of possible gaseous organic precursors, the formation pathways, yields and properties of SOA are complicated and much less clear compared with those of secondary inorganic aerosol (SIA).A large number of laboratory studies have been conducted to elucidate the characteristics of SOA produced by gas-phase oxidation (gasSOA) from certain precursors and the relevant mechanisms are continuously implemented into models (Ervens, 2015;Hallquist et al., 2009).Nevertheless, current models still cannot accurately reproduce the SOA mass or other properties, and the SOA produced in aqueous-phase (cloud/fog drops or aqueous aerosols) (aqSOA) is postulated to be a possible missing portion that can help reconcile the discrepancies (Ervens, 2015;Ervens et al., 2011;Heald et al., 2005).
As ambient conditions are more variable than the lab-controlled conditions, elucidation of secondary aerosol formation (including both SIA and SOA) and its influencing factors is more sophisticated.Traditional filter-based studies are limited in capturing the rapid evolution process, thus highly-time resolved measurements are necessary for a better interpretation of the secondary aerosol formation in ambient air (Wexler and Johnston, 2008).Recently, the Aerodyne Aerosol mass spectrometry (AMS) technique (Canagaratna et al., 2007) has emerged as a powerful tool, as it can quickly determine the concentrations, sizes and chemical compositions of fine aerosols (typically for submicron aerosols, PM1).In combination with factor analysis (e.g., Zhang et al., 2011), it is able to resolve a few types of SOA, which may reflect different SOA formation mechanisms and/or evolution processes .For example, the effects of aqueous-phase and photochemical processing on the SOA formation in Beijing were discussed through analyses of multiple datasets (Xu et al., 2017a;Sun et al., 2016;Sun et al., 2013a).Formation of the aqSOA from AMS was also observed in other locations (e.g., Ge et al., 2012;Gilardoni et al., 2016;Li et al., 2013).Overall, these AMS results have greatly advanced our understanding on SOA (e.g., Jimenez et al., 2009;Li et al., 2017;Spracklen et al., 2011;Philip et al., 2014).
As is well known, Eastern China is frequently experiencing severe haze pollutions, and Yangtze River Delta (YRD) region is one of the heavily polluted regions (e.g., Hu et al., 2014).Previously, the AMS has been applied for online characterization of the fine aerosols in Nanjing (Shen et al., 2014;Zhang et al., 2015a;Zhang et al., 2016;Zhang et al., 2017;Wang et al., 2016b) and Hangzhou (Li et al., 2018) of this region.The offline AMS technique is also developed to analyze filter samples collected in Changzhou (Ye et al., 2017c;Ye et al., 2017a), and Yangzhou (Ge et al., 2017b) in the YRD region.In this study, we conducted a field campaign by using the most advanced AMS version, soot-particle aerosol mass spectrometer (SP-AMS) (Onasch et al., 2012)  occurrence of fullerene soot, that was linked with industry emissions (Wang et al., 2016c).In this work, we focused on analyzing the characteristics of secondary aerosols in this special industrialized environment.

Sampling site and instrumentation
The sampling site and instrumentation was described in Wang et al. (2016c), thus only a brief summary is provided here.The site resided in the Campus of Nanjing University of Information Science and Technology (32°12′20.82″N,118°42′25.46″E),and the measurement period was February 20 to March 23, 2015.Particularly, the site was located west/southwest of an industrial zone (mainly petrochemical, chemical, iron and steelmaking, and power plants), close to the residential area and a few arterial roads (within a radius of a few kilometers) (Fig. 1).Therefore, the site was probably influenced by the mixed emissions from industry, traffic and cooking, etc.
Our main instrument was SP-AMS, which was carefully tuned and strictly calibrated for both mass quantification and sizes following the standard protocols (details in Wang et al. (2016c)).The SP-AMS was operated with a dual-vaporizer setup (with both laser and tungsten vaporizers), and the laser was switched alternately, thus we were able to determine the chemical compositions and size distributions of both non-refractory species (ammonium, sulfate, nitrate, chloride and organics) and refractory black carbon (rBC).Concentrations of PM2.5 and gaseous species (CO, NO2, SO2 and O3) were acquired from the nearest environmental monitoring site.The meteorological parameters, including air temperature (T), relative humidity (RH), wind speed (WS), wind direction (WD), solar radiation and visibility were provided by a meteorological station ~50m away from our site.

Data analyses
The SP-AMS data were processed by using the Igor Pro-based (Wavemetrics) standard ToF-AMS analysis toolkit SQUIRREL version 1.59D and PIKA version  1.19D (Sueper, 2015).All mass concentrations were calculated from the high resolution fitting of V-mode data (a mode sensitive to mass changes).The rBC concentrations were from V-mode with both laser and tungsten vaporizers on, while concentrations of other non-refractory species were from V-mode with the tungsten vaporizer only.The calibrated relative ionization efficiencies (RIEs) were used to account for the different instrument responses to different species.Also, the composition-dependent collection efficiencies (CE) (Middlebrook et al., 2012) were applied to consider the particle loss due to incomplete transmission, and particle bouncing from the tungsten heater.All these data were averaged into hourly data when comparing with the meteorological parameters or gaseous species.The data reported are at local time (Beijing time).
The high chemical resolution of SP-AMS allows us to separate different ions and derive the elemental ratios of OA including oxygen-to-carbon (O/C), hydrogen-to-carbon (H/C), nitrogen-to-carbon (N/C) ratios, and organic mass to organic carbon (OM/OC) ratios.The method proposed by Canagaratna et al. (2015)(referred to as I-A method) was used here unless otherwise stated.The I-A method is an update of the Aiken-ambient (A-A) method (Aiken et al., 2008), which improved the calculation of O/C and H/C ratios.Results from I-A method correlated very well with those from A-A method (Fig. S1 in the supplement), but increased the O/C, H/C and OM/OC on average by 27%, 9% and 8.5%, respectively.
Positive matrix factorization (PMF) (Paatero and Tapper, 1994) and the PMF Evaluation Toolkit version 2.08D (Ulbrich et al., 2009) were applied on the high resolution mass spectra of OA acquired under W-mode (a mode with high mass resolution, ~4000 in this work) with the dual-vaporizer setting.Six OA factors were identified, including three primary factors relevant with traffic (HOA), cooking were presented in details in Wang et al. (2016c).In this work, we focused on analyzing the features and behaviors of these OA factors, in particular the unique primary IOA and three SOA factors.

Overview of the PM1 characteristics
Figure 2 shows the time series of meteorological parameters (T, RH, solar radiation, WS, WD), concentrations of gaseous species (NO2, O3, SO2 and CO), fractional contributions of different PM1 components (sulfate, nitrate, chloride, ammonium, organics and rBC) to the total PM1, and different OA factors (HOA, COA, IOA, LSOA, SVOOA and LVOOA) to the total OA.The weather was overall humid during the sampling period with an average RH of ~70%.There were also a few precipitation events, for example, intermittently on February 28-March 1 and March 17-19, unfortunately the precipitation data were not recorded due to malfunction of the meteorological meter.The wind was not very strong (average 1.4 m/s) and mostly blew from east/northeast directions (Fig. S2).As a number of industrial plants located east/northeast of our site, the measured PM was expected to be influenced by industrial emissions.The campaign acrossed the later winter and early spring, thus the temperature varied significantly from 0 o C to 21 o C with a mean of 8.5 o C. The later days (March 13-23) were much warmer than the earlier days (12.5 o C vs. 6.0 o C).
The PM1 concentrations varied dynamically from 8.4 to 180.5 µg/m 3 , with an average of 46.3 µg/m 3 , and was dominated by inorganic components (68.4%) (Fig. 3a).Such mass contribution from inorganic species in PM1 was higher than most AMS measurements in both urban and rural sites of China (Li et al., 2017) and other countries (Jimenez et al., 2009), likely reflecting the SO2 and NOx emissions enhanced by industry at this specific location.rBC contributed 6.1% of PM1 mass.Organics, on average, was still the most abundant component but only took up 25.5% of PM1 mass.The average contribution from POA (=HOA+COA+IOA) (52%) was almost equal to that of SOA (=LSOA+SVOOA+LVOOA) (48%) (Fig. 3b).Taking together, the secondary components (SIA and SOA) occupied ~82% of the PM1 mass.This fraction is higher than those observed in PM2.5 in a few Chinese megacities during heavy haze periods (30-77%) (Huang et al., 2014).It points out that even in an environment that is expected to have significant primary emissions, secondary species still plays a major role to the aerosol pollution, indicating that the precursors (SO2, NO2 and SOA precursors) can be efficiently oxidized in such atmosphere.
The average size distributions of different species are shown in Fig. 3c.The major inorganic species all peaked in the accumulation mode (~650 nm Dva), representing their behaviors as secondary species.Relatively, the organics had a wider size distribution and peaked at a smaller size (~550 nm Dva), indicating it was a mixture of both primary and secondary species.Size distribution of rBC was significantly different from other components (peak Dva of 250-500 nm), as it was predominantly originated from primary sources.
We also compared the mass concentrations of PM1 determined by SP-AMS with the PM2.5 concentrations from the nearest monitoring station in Fig. 3d.Generally, they correlated well with each other (r 2 =0.70), and the PM1 on average occupied ~83% of PM2.5.This PM1/PM2.5 fraction is higher than those previously observed in urban Nanjing, for example, 54% during springtime (Wang et al., 2016b) and 63% during wintertime (Zhang et al., 2016), showing a more significant contribution from small particles in an environment affected by industry than the average urban case.

Sulfate and nitrate formations
As shown in Fig. 3a, sulfate and nitrate were abundant in PM1.By assuming they were associated with ammonium, (NH4)2SO4 and NH4NO3 together would dominate the PM1 mass (63.2%).In Fig. 4, we presented the variations of sulfate and nitrate concentrations as functions of RH and the odd oxygen (Ox=O3+NO2) over the entire sampling period.Typically, RH is an indicator of atmospheric moisture and is relevant with aqueous-phase/heterogeneous reactions (Tie et al., 2017), while Ox can be used to represent photochemical activities (Herndon et al., 2008).From Fig. 4a sulfate and nitrate concentrations increased substantially with the increase of RH, especially at RH>65%.Note the decreases at RH>90% were probably due to the scavenging effects of precipitation (we have no exact precipitation data so cannot accurately eliminate such influences).On the other hand, sulfate and nitrate both presented no clear increases with the increase of Ox (Fig. 4b).The image plots which describe the dependences of nitrate (Fig. 4c) and sulfate (Fig. 4d) on RH and Ox, also showed that high mass concentrations of sulfate and nitrate appeared mainly in the regimes with RH>65%, but distributed evenly across the changes of Ox.These results highlight a more significant role of moisture in enhancing the formation of both sulfate and nitrate than that of photochemical processing.Similar effects of RH on sulfate and nitrate were also observed in urban Jinan, China (Wang et al., 2012).
As is well known, ammonium nitrate is semi-volatile and water can enhance its thermodynamic gas-particle partitioning and dissolution in the particle phase.Our results clearly reveal that this "thermodynamically-driven" mechanism (Ge et al., 2017a) is critical in governing the nitrate variations in Nanjing.During nighttime， the heterogeneous hydrolysis of N2O5 can also contribute to nitrate production.It is worth to mention that the nitrate formation mechanisms may vary greatly, for example, during wintertime in Beijing, photochemical production of nitrate is more evident (Sun et al., 2013b).On the other hand, ammonium sulfate is non-volatile and thermodynamic partitioning affects its concentrations in a much lesser extent, however water as a reaction medium, may facilitate the oxidation of SO2 into sulfate (Seinfeld and Pandis, 2016).In a word, although high RH can promote formations of both nitrate and sulfate, the underlying detailed processes are different.
To further investigate the sulfate and nitrate formation mechanisms, we calculated the sulfur oxidation ratio (SOR = nSO4 2-/(nSO4 2-+nSO2)) and nitrogen oxidation ratio (NOR = nNO3 -/(nNO3 -+nNO2)).Here nSO4 2-, nNO3 -, nSO2 and nNO2 are the molar concentrations of particle-phase sulfate, nitrate, gaseous sulfur dioxide and nitrogen dioxide, respectively.Note since there are multiple gas-phase forms of nitrogen oxides (NO, N2O3, N2O4, N2O5, etc) rather than only NO2, the NOR actually should  5a and 5b.The NOR was elevated from 0.09-0.13(average: 0.12) at RH<65% to 0.19-0.22(average: 0.21) at RH>65%, yet there was no significant enhancement when RH increased from 65% to 95%.Different from NOR, SOR increased nearly linearly with RH from 0.07 at RH=35% to 0.50 at RH=95% by a large factor of 7.4.The mean SOR under RH>80% was 0.46, much higher than 0.23 observed in wintertime Beijing at RH=80-90% (Sun et al., 2013a), likely indicating a stronger aqueous-phase SO2 oxidation ability in Nanjing.The comparison between NOR and SOR shown here demonstrates that moisture in fact plays a more important role for sulfate formation than it does for nitrate.These results also suggest that by considering increases of particle-phase sulfate and nitrate alone, may not well reflect the RH effects, as nitrate concentrations actually increased more rapidly than that of sulfate with the increase of RH: From RH of 40 to 90%, nitrate increased from 5.1 to 16.6 µg/m 3 , while sulfate increased from 6.8 to 13.9 µg/m 3 .This result likely suggests that a significant portion of nitrate is due to gas-to-particle conversion rather than direct production in aqueous-phase.Previous studies show that the presence of NO2 actually can enhance aqueous-phase sulfate formation under humid/foggy/cloudy conditions in Nanjing (Xie et al., 2015) and Beijing (Cheng et al., 2016;Wang et al., 2016a), though the significance of this mechanism is heavily dependent upon the RH conditions (Liu et al., 2017;Guo et al., 2017;Zhang et al., 2018b).
Correspondingly, Figures 5c and 5d illustrate the variations of NOR and SOR as a function of Ox ceontrations.NOR showed no obvious dependence on Ox, suggesting that photochemical production of nitrate is not significant.While SOR displayed an overall decreasing trend with the increase of Ox concentrations.As high RH was associated with low solar radiation (Fig. 1) and low Ox concentrations (Figs.4a and     4b), this result, on the other hand, underscores the dominance of aqueous-phase processing over photochemical processing for sulfate production.In addition, both NOR and SOR did not vary significantly with changes of temperatures (Fig. S3b), again supporting that they were not significantly influenced by photochemical processes.Moreover, the aqueous-phase sulfate production is actually insignificant during springtime in urban Nanjing when the temperature is warmer and the air is drier, as revealed in our previous study (Wang et al., 2016b).Therefore, the findings point out that aerosol sulfate chemistry can be very different under different seasons in the same region.In addition, a measurement study conducted almost in the same period as this study in Beijing (Zhang et al., 2018c) reported the dominance of aqueous-phase production of sulfate, while nitrate was mainly produced by photochemical and heterogeneous reactions.These results indicate that the formation mechanisms of sulfate and nitrate can be different at different locations in the same season.

Industry-related OA
A specific OA factor -industry-related OA (IOA) was resolved by the PMF analysis in this study.This factor on average occupied ~1/6 of the total OA mass, equivalent to ~1.9 µg/m 3 .It was different from those from traffic and cooking, as their time series were significantly different (r 2 of 0.03 for HOA vs. IOA, and r 2 of 0.01 for COA vs. IOA).Also, across the sampling period, its temporal variations were relatively smaller than those of HOA and COA, indicating it was a persistent source in this region.
The ion-specified HRMS by six ion families is presented in Fig. 6.The IOA O/C ratio (0.44 from I-A method, and 0.33 from A-A method) was relatively high as a POA factor, but was still within a reasonable range.The most abundant ion category in IOA was the chemically-reduced hydrocarbon ions (46%, Fig. 6), but it also had a high contribution from oxygenated ions (especially CO2 + and C2H3O + ).Previously, an industry-related OA factor from the PMF analyses of both online measurement data (El Haddad et al., 2013)  It should be noted that, the IOA MS had also relatively higher fractions C2H4O2 + (m/z 60) and C3H5O2 + (m/z 73) than other factors did.These two ions are often used as biomass burning OA tracers as they can be produced by levoglucosan (Aiken et al., 2010).However, a recent study (Yan et al., 2018) showed that coal combustion could be a significant source of levoglucosan in China.The AMS mass spectrum of peat burning OA was also found to contain appreciable m/z 60 and m/z 73 signals (Lin et al., 2017).Considering that the winds during this campaign were mostly blew from the industry zone with coal (power plants and steel works) and oil burning (petroleum industry) (Fig. S2), and the open biomass burning was banned by the government in this region, the C2H4O2 + and C3H5O2 + signals were likely associated with industry emissions.Of course, more studies are required to further elaborate the characteristics of IOA, including molecular characterization of specific industry-related organic tracer compounds and measurement of the heavy metals which are typically associated with industry emissions.

A modified method to separate SOA from POA
Previously, Ng et al. (2010) proposed to use a triangle plot of f44 (ratio of m/z 44 to total signal of the OA factor) versus f43 (ratio of m/z 43 to total signal of the OA factor) as a diagnostic to describe different OA factors.Ambient OA in this f44:f43 space typically converges to the upper left corner with the ageing of OA.This is because f44 (mainly CO2 + ) is representative of highly oxygenated carboxylic acids while f43 (mainly C2H3O + ) is mainly produced from non-carboxylic oxygenated species.This plot is presented in Fig. 7b.For this dataset, separation of the factors (especially HOA, IOA, LSOA and SVOOA) was not very well by using this method.Chhabra et al. (2011)  as suggested by Heringa et al. (2012).As shown in Fig. 7c, the HOA and COA fell far from SVOOA and LVOOA, while IOA and LSOA were still very close to each other.
Since the method in fact still uses only a few ions, it is likely better to represent the average properties of OA by making use of the chemical information from more ions.
For this purpose, the six OA factors were projected into the fCxOy/fCxR (or fCxRO>1/fCxRO1) vs. fCx≥3R/fCx≤2R spaces (Figs.7d and 7e).Here fCxOy/fCxR is the ratio of all oxygen-containing ions (CxOy) to all oxygen-free ions (CxR), fCxRO>1/fCxRO1 is the ratio of ions with more than one oxygen atoms to those with only one oxygen atom, and fCx≥3R/fCx≤2R is the ratio of hydrocarbon ions at x≥3 to those at x≤2.As shown in Fig. 7a, with increase of oxidation degrees (O/C) of the OA factors, the contribution from Cx groups at x≤2 gradually increased, while contribution from hydrocarbon ions (CxR) with x≥3 decreased.The small ion fragments also contained more oxygenated ions, likely reflecting the abundance of small organic acids.This trend suggests that the ageing of OA proceeds with fragmentation of large molecules, and a fCx≥3R/fCx≤2R ratio can reflect such reaction pathway.On the other hand, the ageing of OA often involves the oxidation of hydrocarbon species, thus the fCxOy/fCxR ratio or fCxRO>1/fCxRO1 ratio may indicate the oxidation of OA.In both Fig. 7d and Fig. 7e, the SOA factors locate in the upper left region, and can be better separated from the POA factors than that by using the f44:f43 space.In particular, the different OA factors approached to the upper left corner in a consecutive order which was consistent with their O/C ratios, and the movement trajectory was also close to a straight line (y=1.89-0.39x,r 2 =0.85).This  (Ng et al., 2010), this method uses the bulk composition of OA.Our modified method proposed here seems to balance the level of chemical information, not only dependent upon a couple of specific ions, but also not rely on the general properties without incorporating detailed ion compositions.Of course, more tests by using more ambient datasets are strongly needed in the future to verify this graphical method.

Effects of aqueous-phase and photochemical processing
In this study, we identified three SOA factors (LSOA, SVOOA and LVOOA), which together occupied nearly half of OA mass.We plotted the mass concentrations and fractional contributions of the three SOA factors against RH and Ox, to investigate the effects of aqueous-phase and photochemical processing on their formations (Fig. 8).Generally, the concentrations of LVOOA presented an increasing trend with RH, and its mass fraction increased more obviously from only 4% at RH<40% to 32% at RH>90%, while those of LSOA and SVOOA showed no such trends.This result demonstrates the importance of aqueous-phase processing on LVOOA formation.On the contrary, both mass loadings and fractions of LSOA and SVOOA increased obviously with the increase of Ox, while those of LVOOA decreased significantly from 46% at Ox<50 µg/m 3 to 9% at Ox>140 µg/m 3 .Such relationships reveal that photochemical processing promotes the LSOA and SVOOA formations, but contributes negligibly or even hinders the formation of LVOOA.Note the LVOOA had the highest O/C ratio (0.74) among the three SOA factors.This is consistent with the facts that aqSOA typically has a high oxidation level as discovered by a number of lab studies (e.g., Herrmann et al., 2015), and the aqSOA products have low volatility in general and contribute significantly to LVOOA (Ervens et al., 2011) SVOOA were relatively fresher with O/C ratios of 0.44 and 0.66, respectively.Our results suggest that the relatively fresh SOA is closely associated with gasSOA (SOA produced from gas-phase photochemical reactions).These findings are in good agreement with those observed in urban Beijing (Xu et al., 2017a), likely representing the general characteristics of the AMS-resolved OOA factors.
For reference, we also checked the variations of POA factors (HOA, COA and IOA) against RH and Ox (Fig. S5).Results show that both concentrations and fractions of the POA factors overall had no clear responses to either RH or Ox increases, which are in accordance with their behaviors as POA -they were emitted directly into the air thus were not expected to be influenced by both aqueous-phase or photochemical processing.One exception was that COA concentrations seemed to increase with Ox, which was likely a coincidence as the lunchtime was exactly noon/early afternoon with strong photochemical activities.
To further investigate the influences of aqueous-phase and photochemical processing on the oxidation levels of SOA (O/CSOA), we calculated the O/CSOA based on the method proposed by Xu et al. (2017b) The combined effects of RH and Ox were further illustrated in Fig. 9. High SVOOA and LSOA concentrations appeared in the regimes with high Ox concentrations (Figs.9a and 9b) and high fractions of these two factors also tended to be accompanied more significantly with low RH conditions (Figs. 9d and 9e).Furthermore, it is known that different formation pathways have different impacts on the size distributions of fine aerosols.Secondary species from gas-phase photochemical reactions typically condense on smaller particles (condensation mode), while particles from aqueous-phase reactions in clouds/fogs or wet aerosols are larger (droplet mode) (Ervens et al., 2011;Kerminen and Wexler, 1995;Meng and Seinfeld, 1994).Enhancements of particle sizes were indeed observed when aqueous-phase reactions occurred (Gilardoni et al., 2016;Ge et al., 2012).In this regard, we investigated the average size distributions of OA, sulfate and nitrate at different RH levels (Fig. 10).Obviously, the peak sizes of OA, sulfate and nitrate all shifted towards larger sizes with the increase of RH.The OA size distribution peaked at 268 nm Dva (vacuum aerodynamic diameter) at RH<40%, and increased substantially to 694 nm Dva at RH of 88-92%.Because increase of RH did not elevate the three POA (Fig. S5), LSOA and SVOOA factors (Figs.8a and 8b), the growth of OA sizes can be attributed mostly to the formation of LVOOA, namely aqSOA.The peak Dva of sulfate increased from 418 nm (RH<40%) to 730 nm (RH=88-92%), and that of nitrate increased even more significantly -from 233 nm (RH<40%) to 694 nm (RH=88-92%).Small drops of the peak Dva occurring at RH>92% for all three species, were probably due to the scavenging effects of precipitation.Overall, the enhancements of peak sizes of OA, sulfate and nitrate by RH provide another evidence supporting the importance of aqueous-phase processing on secondary aerosol formation in this study.

Mass contributions at different pollution levels
We have demonstrated that moisture plays important roles to the formations of sulfate, nitrate and LVOOA, while photochemical processing is important for LSOA and SVOOA productions.Here we investigated contributions of these secondary components at different levels of PM1 and OA pollutions.As shown in Fig. 11a, mass fractions of nitrate increased from 16.5% during clean periods (PM1<10 µg/m 3 ) to ~26-27 % during polluted periods (PM1>60 µg/m 3 ) (there was a small drop from 27.1% at PM1 of 60-70 µg/m 3 to 25.9% at PM1>70 µg/m 3 ).Variations of sulfate contribution were relatively small from 25.2% at PM1<10 µg/m 3 to 21.1% at PM1>60 µg/m 3 .We also illustrated the mean RH and Ox concentrations at different PM1 levels in Fig. 11a.
With the aggravation of PM1 pollution, RH also significantly increased (from 57% to 83%), but Ox concentrations varied very little within a range of 82.3 -90.3 µg/m 3 .This result shows the importance of moisture to haze pollution as it promotes the formations of nitrate and sulfate.In fact, both NOR and SOR values increased with the increase of PM1 concentrations (Fig. S3a).As we showed earlier, nitrate concentrations increased more quickly with RH than that of sulfate, thus its fractions increased significantly, while the sulfate concentrations increased but its relative contributions decreased slightly.The faster increase of nitrate concentrations than sulfate during heavy pollution was also observed in Nanjing during 2014 (Wu et al., 2017), Nanjing (Zhang et al., 2015b) and Lin'an (Shen et al., 2015) during 2013 January in the YRD region.This is different from that in North China Plain, where sulfate often plays a more significant role in heavy haze formation.
For organics, the contributions continuously decreased from 32.3% at PM1<10 µg/m 3 to 23.6% at PM1>70 µg/m 3 (Fig. 11a).This result indicates that although high RH could enhance the production of a portion of SOA (namely the LVOOA), overall, the OA pollution was not governed by RH effects.Therefore, we calculated the mass fractions of different OA factors to total OA at different OA levels in Fig. 11b.Clearly, the LVOOA contributions changed from 25.2-29.0%at OA<10 µg/m 3 down to only 10.4% at OA>30 µg/m 3 , while the sum of LSOA and SVOOA increased substantially from 12.1% at OA<5 µg/m 3 to 45.1% at OA>30 µg/m 3 .This result shows that photochemical processing was more important than the aqueous-phase processing in exacerbating the OA pollution.Correspondingly, we found that the RH did not vary significantly as a function of OA concentrations (within 70-80%), while Ox concentrations increased obviously from 77.6 µg/m 3 at OA<5 µg/m 3 to 115.7 µg/m 3 at OA>30 µg/m 3 .This is somewhat opposite to the case of PM1, and partially explains the decrease of OA contributions to the heavy PM1 pollutions.In addition, the total POA mass fractions decreased from 62.7% at OA<5 µg/m 3 to 44.5% at OA>30 µg/m 3 .This is different from that found during springtime in urban Nanjing (Wang et al., 2016b), where the heavier OA pollution periods were accompanied by elevated POA contributions.
Moreover, we calculated fCO2 + /fC2H3O + and Y/X (Y=fCxOy/fCxR, X=fCx≥3R/fCx ≤2R) against OA concentrations in Fig. 11c.The fCO2 + /fC2H3O + modified the f44/f43 index (Ng et al., 2010) as it eliminates influences from other ions on m/z 44 and m/z 43.Both indexes decreased with OA concentrations, showing that the reduction of highly oxygenated LVOOA contributions and increase of moderately oxidized LSOA and SVOOA contributions compensated the decrease of fresh POA contributions.
Correspondingly, fC4H9 + and fCx≥3R/fCx≤2R increased with the increase of OA concentrations.As C4H9 + (m/z 57) is often used a POA tracer, and fCx≥3R/fCx≤2R also reflects the abundance of chemically reduced hydrocarbon ions (Section 3.3.2),this result again suggests the OA overall became less oxygenated when its pollution became heavier.All results underscore that the relatively fresh and photo-formed SOA (LSOA and SVOOA) plays dominant roles to the OA pollution in this industrialized environment.
In addition, we checked the variations of mass fractions of different components at different wind speeds (WS) and wind directions (WD) (Fig. S6).Generally, the changes were not dramatic, consistent with those of RH (67%-79%) and Ox (79-95 µg/m 3 ), which did not present clear increasing or decreasing trends against WS and WD as well.Such results indicate that the influences of secondary aerosol species seemed to be not affected significantly by the different air masses in this location.

Case studies
We investigated influences of secondary aerosol formation in two specific cases.
The first case was from 3:30pm March 1 to 3:30 pm March 2 (marked in Fig. 2).During this episode, the mean PM1 concentrations was 48.9µg/m 3 , and the summed mass contributions from (NH4)2SO4, NH4NO3 and LVOOA to PM1 was 74.4%, both higher than their corresponding campaign-average values (46.3µg/m 3 and 68.6%, Fig. 3a).This episode was characterized by significant productions of sulfate, nitrate and LVOOA (SNL).The total PM1 concentrations were very closely linked with the SNL formation (Fig. 12b) especially during later afternoon and nighttime.Correspondingly, the enhancement of SNL almost linearly responded to the increase of RH (r=0.96), but oppositely correlated with that of Ox (r=-0.86).High RH during nighttime were associated with high SNL concentrations, while high Ox concentrations during noon/early afternoon were accompanied by low SNL concentrations.
Another case was characterized by relatively significant photochemical formation of SOA, especially the LSOA.This episode lasted from the early evening of March 12 till early morning of March 16 (marked in Fig. 2).In fact, high LSOA concentrations mainly occurred in this period, which was on average 4.8 µg/m 3 , much higher than its average value of 0.38 µg/m 3 during other periods.This is actually one reason it was defined as LSOA since it was most likely a specific and localized event.We found it was mainly driven by photochemical processing (Section 3.3.3),but unable to identify the precursors whose emissions were particularly enhanced during this episode and led to the LSOA formation due to measurement limitations.During this period, the mean OA concentration was 18.3 µg/m 3 (Fig. 13a), much higher than its campaign-average value of 11.8 µg/m 3 ; mass fraction of LSOA and SVOOA (LSS) together was 43.6% (Fig. 13a), also higher than the campaign average of 26.4%.Similar as those in Fig. 8, we observed the enhancements of LSS mass concentrations and fractions at high Ox concentrations but reductions at high RH conditions (Fig. S7).

Conclusions
This work presents the field measurement results focusing on the formation and characteristics of secondary inorganic (mainly sulfate, nitrate) and organic species, by using the SP-AMS in suburban Nanjing during February -March, 2015.The site was surrounded by a large number of industrial plants.Nevertheless, under this industrialized environment, the PM1 was mainly comprised of secondary aerosol components (75.4%), including 63.2% from ammonium sulfate and nitrate, and 12.2% from SOA.This finding indicates that the gas-phase precursors such as SO2, NO2 and volatile organic compounds can be effectively transformed into particle-phase species.
Furthermore, moisture was found to play a major role in enhancing productions of sulfate, nitrate and the highly oxygenated portion of SOA (LVOOA, 45% of SOA), yet the detailed mechanisms were different.The moisture likely affected nitrate by enhancing its thermodynamic gas-particle partitioning and nocturnal heterogeneous production, while direct productions of sulfate and LVOOA in aqueous phase were more significant.In addition, the peak sizes of sulfate, nitrate and OA all shifted towards larger sizes with the increases of relative humidity, reflecting the effects of aqueous-phase processing too.On the other hand, the other two less oxygenated SOA factors (LSOA and SVOOA, together 55% of SOA) were mainly driven by photochemical processing.
Overall, the moisture-driven nitrate and sulfate productions were important to aggravate the PM1 pollution.The photo-chemically formed SOA was more important than the aqueous-phase SOA to OA pollution.The influences of these two formation pathways were demonstrated very clearly in two typical episodes.In addition, in this study we also provided two new findings.First, we separated a specific industry-related OA factor.Secondly, we proposed a modified graphical method to describe the evolution of OA.Both of them should be investigated and verified in the future with other AMS datasets.In summary, this work highlights that aqueous-phase chemistry is very important for sulfate and nitrate productions in an industrialized environment during cold seasons.We also show that the highly oxygenated SOA is very likely linked with aqueous-phase processing while the less oxygenated ones are associated with photochemical processing.

Data availability
The observational data in this study are available from the authors upon request (caxinra@163.com).
in suburban Nanjing during February -March 2015.The sampling site was near an industry zone, and indeed we reported earlier the Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.be smaller than those calculated here.The dependences of NOR and SOR on RH are shown in Figs.
and offline PM2.5 samples in Marseille (France)(Bozzetti et al., 2017) was reported.Similar to our IOA, the factor resolved by Bozzetti et al. (2017) had a high CO2 + peak (~10% of total) and a O/C ratio of 0.33.This Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.industry-relatedOA factor in Marseille was attributed to contributions from coke, steel and petrochemical facilities, which is also very similar to the industry zone adjacent to our site.
also shows that the movement paths of different SOA components in the f44:f43 space depend on the precursors.Generally speaking, f44 and f43 use only two fragments, they may sometimes not well represent the general Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.properties of OA.In addition, m/z 44 and m/z 43 may be significantly influenced by other ions, such as C3H7 + on m/z 43 and C2H4O + /C2H6N + on m/z 44.Heringa et al. (2012) proposed another method to discriminate different OA factors.Following this method, we first re-sorted the HRMS according to the number of carbon (C1-C7) and oxygenation state (Cx, CxO1, CxO>1) of each fragment ion.Each HRMS was grouped into 21 clusters and displayed in Fig. 7a.Then, we placed our six PMF-resolved OA factors into the two-dimensional space of C2/C2O1 vs. C3O1/C3O>1 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.result suggests that the (fCxOy/fCxR):(fCx≥3R/Cx≤2R) space is probably a more effective metric to separate different OA factors and demonstrate the evolution of OA.It should be mentioned that Heald et al. (2010) proposed to use a Van Krevelen diagram (H/C versus O/C) to describe the OA ageing, as shown in Fig. S4 for this dataset.Compared to the f44:f43 space . Our results clearly demonstrate the close link between LVOOA and aqSOA.The LSOA and Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.
. It represents the combined O/C ratio of the summed LSOA, SVOOA and LVOOA.The changes of O/CSOA and O/COA (the O/C ratios of total OA) versus RH and Ox were shown in Figs.8e and 8f.The O/COA showed no clear dependences on both RH and Ox, probably due to the mixing effects of POA.However, the O/CSOA tended to be larger at higher RH (especially at RH>60%), indicating the contribution of more oxygenated aqSOA; while the O/CSOA clearly decreased with increase of Ox concentrations (from 0.71 at Ox<50 µg/m 3 to 0.57 at Ox>140 µg/m 3 ), as more photochemical SOA with lower O/C ratios were generated.
Differently, both high concentrations and high fractions of LVOOA located in the Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.upper left corner characterized by high RH and low Ox conditions (Figs. 9c and 9f), again supporting the importance of aqueous-phase processing on LVOOA production.
More clearly, from noontime of March 15 to early morning of March 16, the LSS mass concentrations correlated positively very well with Ox (r=0.91) but meanwhile linearly decreased with RH (r=-0.90)(Figs.13b and 13c).Contrary to the first case (Fig. 12), high Ox concentrations during noon/afternoon associated with high LSS concentrations, while high RH during nighttime associated with low LSS Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-75Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 5 February 2018 c Author(s) 2018.CC BY 4.0 License.concentrations.These two cases are two typical examples demonstrating especially the aqueous-phase nighttime formation of SNL and the daytime photo-formation of LSS, respectively.

Figure 1 .Figure 2 .
Figure 1.(a) The sampling site and its surroundings.The solid blue point is the sampling site inside the campus of Nanjing University of Information Science and Technology (NUIST) (b).The orange stars mark the positions of the major plants adjacent to the site. 942

Figure 5 .
Figure 5. (a-b) Variations of nitrogen oxidation ratio (NOR)(a) and sulfur oxidation ratio (SOR) (b) as a function of RH (5% increment); and (c-d) NOR and SOR variations as a function of Ox concentrations (8 µg/m 3 increment) (the lines and solid triangles are the mean values, the lines in the boxes are the median values, the upper and lower boundaries of the boxes indicate the 75th and 25th percentiles, and the whiskers above and below the boxes indicate the 90th and 10th percentiles).

Figure 6 .
Figure 6.High-resolution mass spectrum of the industry-related OA (IOA) colored by six ion categories (the inset pie shows the relative mass contributions of the six ion categories).