Seasonal characteristics , formation mechanisms and source origins of PM 2 . 5 in two megacities in Sichuan Basin , China

To investigate the characteristics of PM2.5 and its major chemical components, formation mechanisms, and geographical origins in the two megacities, Chengdu (CD) and Chongqing (CQ), in Sichuan Basin of southwest China, daily PM2.5 samples were collected simultaneously at one urban site in each city for four consecutive seasons from autumn 2014 to summer 2015. Annual mean concentrations of PM2.5 were 67.0± 43.4 and 70.9± 41.4 μg m−3 at CD and CQ, respectively. Secondary inorganic aerosol (SNA) and organic matter (OM) accounted for 41.1 and 26.1 % of PM2.5 mass at CD, and 37.4 and 29.6 % at CQ, respectively. Seasonal variations of PM2.5 and major chemical components were significant, usually with the highest mass concentration in winter and the lowest in summer. Daily PM2.5 concentration exceeded the national air quality standard on 30 % of the sampling days at both sites, and most of the pollution events were at the regional scale within the basin formed under stagnant meteorological conditions. The concentrations of carbonaceous components were higher at CQ than CD, likely partially caused by emissions from the large number of motorcycles and the spraying processes used during automobile production in CQ. Heterogeneous reactions probably played an important role in the formation of SO2− 4 , while both homogeneous and heterogeneous reactions contributed to the formation of NO−3 . Geographical origins of emissions sources contributing to high PM2.5 masses at both sites were identified to be mainly distributed within the basin based on potential source contribution function (PSCF) analysis.

analysis.Knowledge gained in this study provides scientific basis for making future emission control policies aiming to alleviate heavy PM 2.5 pollution in this unique basin.

Methodology
2.1 Sampling sites PM 2.5 samples were collected at two urban sites, one in Chengdu and another in Chongqing, the two largest cities in Sichuan Basin, southwest China.The two sampling sites are located 260 km apart (Fig. 1).The sampling site in Chengdu (CD) is located on the roof of a sixth-floor building in the Sichuan Academy of Environmental Science (104 • 4 E, 30 • 37 N) with no large surrounding industries but heavy traffic.The closest main road (Renmin South road of Chengdu) is about 20 m east of the sampling site.The sampling site in Chongqing (CQ) is located on the rooftop of Chongqing Monitoring Center (106 • 30 E, 29 • 37 N).The highway G50 is 250 m away from this sampling site.The two selected sampling sites are considered to represent typical urban environments in their respective cities (Tao et al., 2014;Chen et al., 2017).

Sample collection
Daily (23 h) integrated PM 2.5 samples were collected in four months, each in a different season: autumn (23 October to 18 November, 2014), winter (6 January to 2 February, 2015), spring (2 to 29 April, 2015), and summer (2 to 30 July, 2015).At both sites, PM 2.5 samples were collected in parallel on Teflon filters (Whatman, 47 mm) and quartz filters (Whatman, 47 mm).At the CD site, PM 2.5 sampling was carried out using a versatile air pollutant sampler (Wang et al., 2017).One channel was used to load the PM 2.5 sample on the Teflon filter for mass and trace elements analysis and the other one was equipped with a quartz filter for water-soluble inorganic ions and carbonaceous components analysis.The sampler was running at 15 L min −1 for each channel.At the CQ site, a low-volume aerosol sampler (BGI, frmOmni, USA) operating at a flow rate of 5 L min −1 was used to collect PM 2.5 samples on Teflon filter, and another sampler (Thermo Scientific Partisol 2000i, USA) with a flow rate of 16.7 L min −1 was used to collect PM 2.5 samples on quartz filter.A total of 112 samples and 8 field blanks, nearly equally distributed in the four seasons, were collected at each site during the campaign.In addition, three lab blank filters in each campaign were stored in clean Petri slides in the dark and analyzed in the same ways as the collected samples to evaluate the background contamination.
Before sampling, all the quartz filters were preheated at 450 • C for 4 h to remove the organic compounds.All sampled filters were stored in clean Petri slides in the dark and at −18 • C until analysis to prevent the evaporation of volatile compounds.Before and after sample collection, all the Teflon filters were weighed at least 3 times using an microbalance (Sartorius, ME 5-F, Germany) after their stabilization for 48 h under controlled conditions (temperature: 20-23 • C; relative humidity: 45-50 %).Differences among replicate weights were mostly less than 15 µg for each sample.

Chemical analysis
For the analysis of water-soluble inorganic ions, a quarter of each quartz filter was first extracted using ultrapure water in an ultrasonic bath for 30 min, and then filtered through a 0.45 µm pore syringe filter.Anions (SO 2− 4 , NO − 3 and Cl − ) and cations (Na + , NH + 4 , K + , Mg 2+ and Ca 2+ ) were determined using ion chromatograph (Dionex, Dionex 600, USA).Anions were separated using AS11-HC column with 30 mM KOH as an eluent at a flow rate of 1.0 mL min −1 .Cations were determined using CS12A column with 20 mM MSA (methanesulfonic acid) at a flow rate of 1.0 mL min −1 .Individual standard solutions of all investigated anions and cations (1000 mg L −1 , o2si, USA) were diluted to construct the calibration curves.The correlation coefficients of the lin-ear regression of the standard curves were all above 0.999.Field blanks were prepared and analyzed together with the samples and then subtracted from the samples.The concentrations of the water-soluble inorganic ions in the field blanks were in the range of 0.008-0.13µg m −3 .The relative standard deviation of each ion was better than 8 % for the reproducibility test.
Organic carbon (OC) and elemental carbon (EC) were measured by thermal-optical reflectance (TOR) method using a DRI OC / EC analyzer (Atmoslytic Inc., USA).The methodology for OC / EC analysis was based on the TOR method as described in Chow et al. (2007).For calibration and quality control, measurement with filter blank, standard sucrose solution and replicate analysis were performed.Blank corrections were performed by subtracting the blank values from the sampled ones.The concentration of EC in field blanks was zero, while OC was below 0.7 µg C cm −2 .The repeatability was better than 15 %.
The elements, including Al, Si, Ca, Fe, and Ti, were analyzed on a Teflon filter using X-ray fluorescence analyzer (Epsilon 5ED-XRF, PANalytical, Netherlands); the QA/QC procedures of the XRF analysis have been described in Cao et al. (2012).The gaseous species were continuously measured by a set of online gas analyzers, including an EC9850 SO 2 analyzer, 9841 NO / NO 2 / NO x analyzer, 9830 CO analyzer, and 9810 O 3 analyzer (Ecotech, Australia) at CD, and a Thermo 42i NO / NO 2 / NO x analyzer, 43i SO 2 analyzer, 48i CO analyzer, and 49i O 3 analyzer (Thermo Scientific, USA) at CQ.The mass concentrations of PM 2.5 were automatically measured by online particulate monitor instruments (BAM1020, Met One, USA, at CD and 5030 SHARP, Thermo Scientific, USA, at CQ). Hourly meteorological parameters, including ambient temperature (T ), relative humidity (RH), wind speed (WS) and direction, barometric pressure (P ), and solar radiation (SR), were obtained from an automatic weather station (Lufft WS501, Germany) at each site.Hourly precipitation data were recorded at the nearest weather station operated by China Meteorology Administration (http://www.weather.com.cn/).Planetary boundary layer height (PBLH) was obtained from the HYSPLIT model (http://ready.arl.noaa.gov/HYSPLIT.php).

Data analysis
The EC-tracer method has been widely used to estimate SOC (Turpin and Lim, 2001;Castro et al., 1999) 1.9, 2.8, 1.1 and 1.5 at CQ.The estimated SOC was only an approximation with uncertainties, e.g., from the influence of biomass burning (Ding et al., 2012).The coefficient of divergence (COD) has been used to evaluate the spatial similarity of chemical compositions at different sites (Wongphatarakul et al., 1998;Qu et al., 2015), which is defined as where x ij and x ik represent the average concentration for a chemical component i at site j and k, respectively, and p is the number of chemical components.Generally, a COD value lower than 0.2 indicates a relatively similarity of spatial distribution.
2.5 Geographical origins of PM 2.5 72 h air mass back trajectories were generated based on the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model using 0.5 • × 0.5 • meteorological data for the period of October 2014 to July 2015 when PM 2.5 measurements were made at both sites.Four trajectories at 04:00, 10:00, 16:00, and 22:00 UTC every day with the starting height of 300 m above ground level were calculated (Squizzato and Masiol, 2015).
PSCF is substantially a conditional probability that trajectories with pollutant concentrations larger than a given criterion passed through a grid cell (i, j ) (Ashbaugh et al., 1985;Polissar et al., 1999), which means a grid cell (i, j ) with high PSCF values is likely to be a potential source location of pollutants.PSCF is defined as follows, where n ij is the total number of endpoints falling in the grid cell (i, j ) and m ij denotes the number of endpoints that are associated with samples exceeding the threshold criterion in the same cell.To reduce the PSCF uncertainties associated with small n ij values, a weighting function was adopted as follows, where n ave is the average number of endpoints in each grid cell.
The trajectories coupled with daily pollutants concentrations were used for PSCF analysis, with the threshold criterion in PSCF analysis being set at the upper 50 % of PM 2.5 and other pollutants.The trajectory covered area was in the range of 20-45 • N and 90-120 • E and divided into 0.5 • × 0.5 • grid cells.
3 Results and discussion 3.1 PM 2.5 mass concentration and chemical composition

Overview
Table 1 presents seasonal and annual mean concentrations of PM 2.5 and its major chemical components at CD and CQ during the sampling periods.Daily PM 2.5 ranged from 11.6 to 224.7 µg m −3 , with annual average being 67.0 ± 43.4 µg m −3 at CD and 70.9 ± 41.4 µg m −3 at CQ, which were about 2 times the NAAQS annual limit.Secondary inorganic aerosol (SNA, the sum of SO 2− 4 , NO − 3 , and NH + 4 ) and carbonaceous species together represented more than 70 % of PM 2.5 mass at both sites (Fig. 2).The annual mean concentrations of SNA were 27.6 µg m −3 at CD and 26.5 µg m −3 at CQ, contributing 41.1 and 37.4 % to PM 2.5 mass, respectively.SO 2− 4 , NO − 3 , and NH + 4 accounted for 16.8, 13.6, and 10.8 %, respectively, of PM 2.5 mass at CD, and 17.2, 10.9, and 9.2 %, respectively, at CQ. Organic matter (OM), estimated from OC using a conversion factor of 1.6 to account for other elements presented in organic compounds (Turpin and Lim, 2001), was the most abundant species in PM 2.5 , accounting for 26.1 and 29.6 % of PM 2.5 mass at CD and CQ, respectively.In contrast, EC only comprised of around 6 % at both sites.The annual mean concentration of OC at CQ was 20 % higher than that at CD, while the annual mean concentration of EC at CQ was 25 % higher than that at CD.The annual mean concentration of fine soil (FS), calculated by summing the oxides of major crustal elements, i.e., Al 2 O 3 , SiO 2 , CaO, FeO, Fe 2 O 3 , and TiO 2 (Huang et al., 2014), was 6.7 µg m −3 (9.5 % of PM 2.5 mass) at CQ.It is noted that this was about 2 times that at CD (3.8 µg m −3 , 5.7 % of PM 2.5 mass).The minor components such as K + and Cl − constituted less than 5 % of PM 2.5 .The unaccounted-for portions of PM 2.5 reached 18.3 % at CD and 15.3 % at CQ, which were likely related to the uncertainties in the multiplication factors used for estimating OM and FS, other unidentified species, and measurement uncertainties.

Seasonal variations
Figure 3 shows the seasonal variations in mass concentrations of PM 2.5 and its major chemical components at CD and CQ.Seasonal variations of any pollutants were influenced by the seasonal variations in source emission intensities, atmospheric processes, and meteorological conditions.Unlike in northern China, there were no extensive coal combustion or wood burning for domestic heating in winter due to the warm temperature (around 10 • C on average) in the Sichuan Basin; hence, atmospheric processes and meteorological conditions played vital roles in the seasonal variations of PM 2.5 .On a seasonal basis, PM 2.5 mass was the highest in winter at both CD and CQ, which was 1.8-2.5 times those in the other sea- 1.9 ± 1.2 3.4 ± 1.9 0.6 ± 0.4 0.2 ± 0.2 1.5 ± 1.7 0.8 ± 0.4 1.6 ± 1.2 0.5 ± 0.5 0.04 ± 0.03 0.7 ± 0.9 K + 0.6 ± 0.4 1.2 ± 0.6 0.6 ± 0.5 0.5 ± 0.2 0.7 ± 0.5 0.5 ± 0.2 1.2 ± 0.7 0.5 ± 0.2 0.3 ± 0.1 0.6 ± 0.5 OC 10.4 ± 6.1 19.7 ± 8.4 6.3 ± 3.7 7.4 ± 1.5 10.9 ± 7.6 9.7 ± 4.7 24.2 ± 13.6 10.0 ± 5.1 8.5 ± 3.4 13.1 ± 10.0 EC 3.0 ± 2.1 6.3 ± 3.0 2.7 ± 2.3 2.5 ± 0.7 3.6 ± 2.7 3.8 ± 1.7 5.9 ± 3.2 4.7 ± 3.0 3.7 ± 1.5 4.5 ± 2.6 FS 3.2 ± 1.6 4.5 ± 2.0 4.8 ± 3.0 2.7 ± 1.5 3.8 ± 2.2 5.0 ± 2.8 6.3 ± 3.3 9.1 ± 7.6 6.5 ± 4.0 6.7 ± 5.0 sons.In contrast, its seasonal differences among the other three seasons were generally small, i.e., less than 40 %.Stagnant air conditions with frequent calm winds and low planetary boundary layer heights were the major causes of the highest PM 2.5 mass in winter (Table 1) (Liao et al., 2017;Chen and Xie, 2013;L. L. Li et al., 2017).All the major PM 2.5 components except FS followed the seasonal pattern of PM 2.5 mass with subtle differences.The highest FS concentrations were observed in spring at both sites.The relatively high wind speed and lower RH in spring were conducive for re-suspension of crustal dust and resulted in higher FS concentrations.In addition, frequent spring dust storms originated in the northwestern China was able to reach Sichuan Basin via long-range transport, causing the elevated FS concentrations (Chen et al., 2015;Tao et al., 2013).The highest contributions from FS to PM 2.5 mass were more than 10 %, appearing in spring at both sites.The majority of PM 2.5 components showed a summer minimum, which was caused by high planetary boundary layer height favoring pollutant dispersion and abundant precipitation favoring wet scavenging (Table 1).One exception was SO 2− 4 , which had a minimum in spring at CD and in autumn at CQ, likely due to the enhanced photochemical reactions associated with high temperature and strong solar radiation in summer.High O 3 concentrations in summer also supported this seasonal trend.It is also noted that the seasonal variations of NO − 3 were much larger than those of SO 2− 4 and NH + 4 .SO 2− 4 and SO 2 showed similar seasonal trends, with their concentrations 1.4-2.0times higher in winter than in the other seasons (Table 1).In contrast, the seasonal variations of NO − 2014).As shown in Fig. 2, the seasonal average contributions of SNA to PM 2.5 only varied within a small range from 39.5 to 43.2 % at CD, whereas in a relatively larger range from 31.0 % in summer to 37.1-41.5 % in the other seasons at CQ.The smaller contribution in summer at CQ was mainly due to the lower NO − 3 concentrations.At both CD and CQ, NO − 3 and NH + 4 showed the highest contributions in winter and the lowest ones in summer, whereas an opposite trend was found for SO 2− 4 .Both OC and EC exhibited the highest concentrations in winter at CD and CQ, around 1.9-3.1 times those in the other seasons.SOC was also the highest in winter at both sites, similarly to what observed for OC, which can be partly explained by the enhanced condensation process forming SOC under low temperature (Sahu et al., 2011;Cesari et al., 2016).In contrast, high temperature in summer favored gas-particle partitioning in the gas phase and thus limited the formation of SOC (Strader et al., 1999).The contributions of carbonaceous components generally followed the seasonal patterns of SNA, accounting for 26.7-38.8% of PM 2.5 mass.Among these carbonaceous species, OM showed the lowest fractions in PM 2.5 in spring (21.1 %) at CD and the highest value in winter (33.6 %) at CQ, while the percentages of OM in other seasons were similar at both sites, around 27 %.The seasonal variations of EC fractions were not obvious, with a slightly higher value in spring.

Similarities and differences between the two sites
Although none of the two sites alone can represent the whole region of the Sichuan Basin, the similarities in the characteristics of the major pollutants between the two sites should represent the regional-scale characteristics of urbanenvironment pollution, while the differences between the two sites should reflect the sub-regional characteristics of urban pollution.A comparison between the two sites in terms of seasonal-average concentrations of major chemical components is shown in Fig. 4 and discussed in detail below.Despite the 260 km distance between the two sampling sites, a moderate similarity was observed in autumn, winter, and spring on the basis of low COD values (0.15-0.18), indicating limited differences between the two urban environments in the Sichuan Basin and the similarities in major emission sources for both sites.The similar pollution patterns observed at both CD and CQ were likely to be related to the similar meteorological parameters and special topography of the basin, which is a closed lowland area surrounded by high mountains on all sides (Fig. 1).The mean elevation in the basin is about 200-700 m, while the surrounding mountains are around a range of 1000-3000 m elevation.The Tibetan Plateau lies close to the western Sichuan Basin, with an elevation above 4000 m.Such a plateau-basin topography forms a barrier for the dispersion of pollutants and causes air stagnation within the basin, thereby facilitating regional- scale pollution events in the basin; 72 h air mass back trajectory analysis (18:00 local time) showed that air masses reaching CD and CQ mainly originated from local areas in the basin (Fig. S1 in the Supplement), confirming the influence of the high mountainous surroundings of the basin.These results were consistent with those found in earlier studies in Chengdu and Chongqing (Tian et al., 2017;Liao et al., 2017), which suggested that air masses had short-range trajectories and primarily originated from inside the Sichuan Basin, highlighting the impacts of the special topography on PM 2.5 pollution.A similar case has also been found elsewhere, such as in the Po Valley, Italy (Ricciardelli et al., 2017).
It is worth noting that the COD values used to identify the similarities or differences of the two sites were calculated based on seasonal-average concentrations of all the components in PM 2.5 .However, if focusing on individual components, several chemical species in PM 2.5 differed by up to a factor of 2.5 in their season-average concentrations between CD and CQ, e.g., OC and EC in winter and spring, and Cl − and FS in all the four seasons.In summer, the differences for several major chemical components (FS, OC, SO 2− 4 , NO − 3 , and EC) between the two sites were larger than in the other seasons, causing a high COD value (0.33).These discrepancies were partly caused by the different atmospheric chemical processes, local sources and meteorological parameters between the two sites.Specifically, FS mostly deviated from the 1:1 straight line in all the seasons, with substantially higher concentrations at CQ than CD (Fig. 4).There was no significant difference in NH + 4 concentrations between CD and CQ, but there were considerable differences in SO 2− 4 and NO − 3 in spring and summer.SO 2 concentration was around 25 % higher at CQ than CD in spring and summer, which partially explains the site differences in SO 2− 4 .In contrast, NO 2 concentration was comparable at both sites in summer, but NO − 3 concentration was 58 % lower in CQ than CD.The site differences in NO − 3 concentration was caused by NH 4 NO 3 thermodynamic equilibrium controlled by ambient temperature and RH, instead of by its gaseous precursors.The equilibrium would be shifted toward the particulate phase when ambient RH was above the deliquescence relative humidity (DRH) of NH 4 NO 3 , and the dissociation constant decreased by about 1 order of magnitude when RH was above 75 % (Kuprov et al., 2014).DRH was calculated from temperature following Mozurkewich (1993).As shown in Table 1, the average temperature was comparable at CD and CQ during the summer period, hence leading to similar DRH values of NH 4 NO 3 , ranging from 59 to 64 % with an average value of 60.7 %.However, the ambient RH was substantially lower at CQ (61 %) than CD (72 %), causing lower NO − 3 concentration at CQ.As shown in Fig. S2, 53 % of the hourly data in summer have ambient RH lower than DRH at CQ, while this is the case for only 19 % of such data at CD, which explains the different NO − 3 concentrations between CD and CQ. Figure 4 shows higher concentrations of carbonaceous component (OC and EC) at CQ than CD in all the seasons except OC in autumn and EC in winter.OC and EC mainly originate from fossil fuel combustion and biomass burning.K + is usually regarded as a tracer of biomass burning (Tao et al., 2016).During the sampling campaign, no significant differences in K + levels were observed between CD and CQ (Table 1), suggesting that biomass burning was not be the major cause of the higher concentrations of carbonaceous component at CQ. Motorcycle traffic was likely a major source of volatile organic compounds (VOCs) in CQ since it is a famous mountain city where most people use motorcycles as daily transportation.The number of motorcycles was 2.0 million in Chongqing in 2014, which was much higher than that (0.7 million) in Chengdu (National Bureau of Statistics of China, 2015).Moreover, Chongqing has become China's largest automobile production base, which likely also emits VOCs from spraying processes.Higher concentrations of VOCs in CQ would cause higher concentrations of secondary organic carbon via photochemical reaction under high temperature or vapor condensation under low temperature.This hypothesis is supported by the large differences in OC concentrations in winter between the two sites.
Correlation analysis may also provide an insight into the similarities/differences between the two sites over an intensive sampling period.Good correlations between the two sites were found for daily SNA, OC, EC, and K + concentrations in autumn, winter, and spring (Table S1).However, for NO − 3 , a significant correlation was identified only in autumn, likely due to the strong impact of local vehicle emissions and the subsequent atmospheric processes forming NO − 3 .Similarly, a moderate correlation was observed just in winter for both Cl − and FS.In summer, weak or no correlations were identified between the two sites for almost all chemical components.

Pollution episodes and key chemical components
Pollution episodes during the campaign are highlighted with shaded areas in Fig. 5.These pollution periods (PP) were defined as daily PM 2.5 concentration being above the NAAQS guideline value of 75 µg m −3 .Similarly, the days with PM 2.5 concentration below 75 µg m −3 were characterized as clean periods (CP).A total of seven pollution episodes were identified during the campaign at each site.There were three longlasting pollution episodes occurred simultaneously at the two sites on 23-27 October 2014, 7(8)-26 January, and 26-28 (29) April 2015.A total of 34 and 31 pollution days were counted at the CD and CQ sites, respectively, accounting for 30.4 and 28.6 % of the entire sampling days (112 days).The number of pollution days at CD was 8, 21, 4, and 1 in autumn, winter, spring, and summer, accounting for 29.6, 75, 14.3, and 3.4 % of the total sampling days in each season, respectively, and at CQ they were 4, 19, 6, and 2, accounting for 14.8, 67.9, 21.4, and 6.9 %.Stagnant atmosphere and high RH were important factors causing PM 2.5 pollution events, as was found in this and earlier studies (G.J. Zheng et al., 2015;Chen et al., 2017;Liao et al., 2017).Compared with the clean periods, the pollution periods were usually characterized by low planetary boundary layer height and weak wind speed (Table S2), which suppressed pollutant dispersion vertically and horizontally.Temperature increased during the long-lasting pollution episodes, which promoted gasto-particle transformation, forming secondary aerosols.RH remained high (68-88 %) during pollution episodes (except in spring at CQ), although not much different from clean periods, which was also conducive for aqueous-phase reactions converting gaseous pollutants into aerosols (Chen et al., 2017;Tian et al., 2017).
Looking more closely at a regional-scale long-lasting pollution episode in winter, from 8 to 26 January 2015, the concentrations of PM 2.5 and major chemical components increased dramatically compared with clean periods (Fig. 6).PM 2.5 concentrations were more than 3 times higher at both sites, with the two dominant groups of components, SNA and OC, being 2.5-2.8 times higher at CD and 1.7-2.7 times higher at CQ.The enhancement of SNA and OC during pollution periods were similar at CD, but OC increased much more than SNA at CQ, indicating some different contributing factors to the high-PM 2.5 pollution at the two sites.Pollutants accumulation under stagnant meteorological conditions might be a main factor at CD based on the similar magnitudes of the enhancements of PM 2.5 and its dominant components, while additional processes should have increased OC more than other components at CQ.The percentage contributions of SNA to PM 2.5 were similar during clean and pollution periods: 38-41 % at CD and CQ (Fig. S3).However, the percentage contributions of OM to PM 2.5 decreased from 30.1 % on clean days to 27.5 % on pollution days at CD, and increased from 26.9 to 34.9 % at CQ. Concentrations of the individual SNA species (SO 2− 4 , NO − 3 , and NH + 4 ) increased by a factor of 2.5-3.3 on pollution days compared with clean days in all the cases (Fig. 6).But the percentage contributions differed among the species as NO − 3 increased and SO 2− 4 decreased on pollution days (Fig. S3).The percentage contributions of SNA and OM to PM 2.5 discussed above were different from those found in eastern coastal China and the North China Plain, where considerable increases were found for SNA and decreases for OM on pollution days than clean days (Tan et al., 2009;H. L. Wang et al., 2015;Quan et al., 2014;Zhang et al., 2015Zhang et al., , 2016;;Cheng et al., 2015).The pollution periods in eastern coastal China and the North China Plain were accompanied with sharp increases of RH, which would promote aqueous-phase formation of secondary inorganic aerosols and resulted in rapid elevation of SO 2− 4 and NO − 3 concentrations (G.J. Zheng et al., 2015;B. Zheng et al., 2015;X. J. Zhao et al., 2013;H. Li et al., 2017).In contrast, RH remained high during either clean or pollution periods in the present study, suggesting that high RH might not be the primary cause of the dramatic increase in PM 2.5 concentrations during the pollution period in the Sichuan Basin.Another point that needs to be mentioned is that, as shown in Fig. S1, local sources were the main contributors to the pollution episodes in the Sichuan Basin while sources outside local regions frequently contributed to pollution episodes in eastern coastal China and the North China Plain through long/medium-range transport (Gao et al., 2015;Hua et al., 2015;Q. Z. Wang et al., 2015).

Transformation mechanisms of secondary aerosols
In most cases, meteorological conditions, atmospheric chemical processes, and long-range transport are all responsible for PM 2.5 accumulation (G.J. Zheng et al., 2015).CO is directly emitted from combustion processes and is not very reactive.Its concentrations in the air are strongly controlled by meteorological parameters within a relatively short period, which makes it a good tracer that can be used for separating different dominant factors contributing to pollutants accumulation (G.J. Zheng et al., 2015;Zhang et al., 2015;Hu et al., 2013;Liggio et al., 2016).The impact of atmospheric physical processes on other pollutants can be revealed by scaling the concentrations of other pollutants to that of CO.For example, PM 2.5 was enhanced by a factor of 2.7 on pollution days at both sites, but the CO-scaled PM 2.5 (the ratio of PM 2.5 to CO concentration) only showed an enhancement of a factor of 1.6-1.8(Fig. 7), and the latter values were likely from the enhanced secondary aerosol formation.
As shown in Fig. 7, the CO-scaled SNA was 60-90 % higher on pollution days with individual species 40-120 % higher, even though their gaseous precursors (SO 2 and NO 2 , no data for NH 3 ) were only less than 30 % higher.This  3 / (n-NO − 3 +n-NO 2 )) were defined to evaluate the degree of secondary transformation (n refers to as the molar concentration) (Hu et al., 2014).NOR increased from 0.09 on clean days to 0.16 on pollution days at CD and from 0.07 to 0.14 at CQ. SOR increased only slightly, from 0.31 to 0.35 at CD and 0.28 to 0.35 at CQ.The CO-scaled SOC increased by a factor of 2.6 on pollution days at CQ, but no significant change was found at CD.The different patterns in SOC (or SOC / OC) than SNA (or SOR and NOR) suggested that secondary organic aerosol (SOA) production was of less important than SNA production at CD. SO 2− 4 is predominantly formed via homogeneous gasphase oxidation.In this pathway, SO 2 is firstly oxidized by OH radical to SO 3 , and then to H 2 SO 4 (Stockwell and Calvert, 1983;Blitz et al., 2003).Apart from homogeneous reaction, particulate SO 2− 4 can also be formed through heterogeneous reactions with dissolved O 3 or H 2 O 2 under the catalysis of transition metal and in-cloud process (Ianniello et al., 2011).HNO 3 is primarily produced from the reactions between NO 2 and OH radical during the daytime and later combines with NH 3 to produce particulate NO − 3 (Calvert and Stockwell, 1983).Particulate NO − 3 can also be formed through heterogeneous hydrolysis of N 2 O 5 on moist and acidic aerosols during nighttime (Ravishankara, 1997;Brown and Stutz, 2012).Similarly, SOA is mainly formed through photochemical oxidation of primary VOCs followed by condensation of SVOC onto particles as well as through aqueous-phase reactions (Ervens et al., 2011).While photochemical reactions are mostly influenced by temperature and oxidants amount, heterogeneous reactions always depend on ambient RH.To further explore the formation mechanisms of secondary aerosols, SOR, NOR and SOC / OC data were grouped with temperature (at 2 • C interval), RH (at 5 % interval) and daytime O 3 concentration (at 5 µg m −3 interval) bins (Fig. 8).An obvious increase in SOR with increasing RH was found at both sites, but this was not the case for temperature and O 3 concentration, suggesting heterogeneous processes played important roles in the formation of SO 2− 4 , as was suggested in many previous studies (Quan et al., 2015;B. Zheng et al., 2015;X. J. Zhao et al., 2013).Interestingly, SOR exhibited a decreasing trend with increasing O 3 concentration at O 3 concentrations lower than 15 µg m −3 and an opposite trend was found at O 3 concen- trations above 20 µg m −3 (Fig. 8).Additionally, high PM 2.5 concentrations were mostly associated with lower O 3 concentrations.This behavior might be explained by the complicated interactions between aerosol and O 3 .On the one hand, aerosols are generally considered as a constraining factor to O 3 production due to their absorption and scattering of UV radiation, which reduce solar radiation and consequently decrease photochemical activity.On the other hand, aerosols can provide an interface for the heterogeneous reaction, in accordance with O 3 consumption and secondary aerosol formation, which would result in decreased O 3 concentrations and increased secondary aerosols (G.J. Zheng et al., 2015).It was further found that the ambient RH remained high at low O 3 concentrations (Fig. S4), which was beneficial to SO 2− also revealed the possibility of the heterogeneous processes, although this cannot be verified directly due to the lack of high-resolution data.
The ratio of SOC / OC decreased with increasing temperature at CD but increased at CQ when temperature was lower than 10 • C.Although SOC / OC did not correlate well with RH, an opposite trend was also found between CD and CQ at high-RH conditions.Heterogeneous reactions seemed to be dominant in the formation of SOA at CD, whereas homogeneous reactions were prevalent at CQ. SOC / OC showed no apparent dependency on O 3 concentrations at either site, indicating more complex formation mechanisms of SOA than SO 2− 4 and NO − 3 .
3.2.3Geographical origins of high-PM 2.5 pollution PSCF analysis was applied to investigate the potential source regions contributing to high-PM 2.5 pollution.As can be seen from the PSCF maps in Fig. 10, all the pollutants including PM 2.5 and its chemical components as well as gaseous precursors had similar spatial patterns of potential source areas.Basically, all the major source areas for high pollutant concentrations were distributed within the basin.Long-range transport events as seen in the North China Plain and eastern coastal regions were not observed at CD and CQ (Zhao et al., 2015;Zhang et al., 2013).At CD, the major source areas in winter included the areas of the northeastern, southeastern, and southern Chengdu and in some areas of eastern Chongqing.A similar spatial distribution of PM 2.5 potential sources was also found by Liao et al. (2017) through PSCF analysis in winter 2013, in which high PM 2.5 concentrations were mostly associated with sources broadly located in the southeast of the basin, covering Neijiang, Zigong, Yibin, Luzhou, and east part of Chongqing.At CQ, the northeast area of Chongqing was identified as strong sources, where a number of industries were located, such as Changshou chemical industrial ozone.Overall, PM 2.5 pollution at CQ was characterized by significant local contribution from major sources located in or nearby Chongqing.In contrast, regional transport in Sichuan Basin from southeast, south, and southwest of Chengdu had a major impact on PM 2.5 pollution at CD.
Chemically resolved PM 2.5 data collected during four seasons at two urban sites in Sichuan Basin, southwest China, were analyzed in the present study.On about 30 % of the days, daily PM 2.5 exceeded the national air quality standard, with annual mean concentrations of 67.0 ± 43.4 and 70.9 ± 41.4 µg m −3 at CD and CQ, respectively.SO 2− 4 , NO − 3 , NH + 4 , OM, EC, and FS were the major chemical components of PM 2.5 , accounting for 16.8,13.6,10.8,26.1,5.4,and 5.7 % of PM 2.5 at CD,and 17.2,10.9,9.2,29.6,6.4,and 9.5 % at CQ, on an annual average, respectively.The concurrent occurrences of heavy pollution events at the two sites and similarities in pollutants characteristics between the two sites were mainly caused by the surrounding mountainous topography under typical stagnant meteorological conditions.Such a finding was also supported by back trajectory analysis, which showed that air masses reaching both sites originated within the basin and only traveled for short distances on heavy polluted days.Differences between the two sites with regards to several major chemical components provided evidence of sub-regional characteristics of emission sources and chemical transformation processes under different meteorological conditions.For example, an additional source factor from motorcycle traffic was identified for VOC emission in Chongqing, which led to higher OC concentrations, and lower relative humidity in Chongqing caused lower NO − 3 concentration in this city despite similar levels of its gaseous precursors in the two cities.The present study also identified different driving mechanisms for the PM 2.5 pollution episodes in the Sichuan Basin than in the other regions of China.For example, sharply increased relative humidity was thought to be one of the main factors causing high inorganic aerosol concentrations during the pollution periods in eastern coastal China and the North China Plain, while in the Sichuan Basin the special topography and meteorological conditions are driving forces for such events considering relative humidity was high throughout the year and did not differ much between pollution and clean periods.However, on an annual basis heterogeneous reactions might be more important in this basin than in the other regions of China due to the consistently high humidity conditions, as revealed in the case of SO 2− 4 formation in the present study.Future studies should use high-resolution data to verify the findings related to chemical transformation paths proposed here.More importantly, a detailed emission inventory of aerosol particles and related gaseous precursors in the basin should be developed promptly, which is needed for further investigating PM 2.5 formation mechanisms and for making future emission control policies.Source-receptor analysis using monitored chemical-resolved PM 2.5 data should be conducted to verify such emission inventories.

Figure 2 .
Figure 2. Seasonal and annual contributions of individual chemical components to PM 2.5 at CD (a) and CQ (b).

Figure 4 .
Figure 4. Seasonal mean concentrations of major components in autumn (a), winter (b), spring (c), and summer (d) at CD and CQ sites.

Figure 5 .Figure 6 .
Figure 5. Temporal variations of meteorological parameters, gaseous pollutants, and PM 2.5 during the campaign at CD (a) and CQ (b).Pollution episodes are highlighted by shaded areas.

Figure 7 .
Figure 7. CO-scaled concentrations of various pollutants and the values of SOR, NOR, and SOC / OC in winter at CD (a) and CQ (b).CP and PP are from the same period as Fig. 6.

Figure 8 .
Figure 8. Correlations of SOR, NOR, and SOC / OC against temperature (a), RH (b), and O 3 concentration (c) in winter at CD and CQ.

Figure 10 .
Figure 10.PSCF distribution of PM 2.5 , its chemical components, and gaseous precursors in winter at CD (a) and CQ (b).

Table 1 .
Meteorological parameters, annual and seasonal mean concentrations of PM 2.5 , gaseous pollutants, and major chemical components atCD and CQ during 2014-2015.na.means no data.