Comprehensive observations were carried out at the Chengdu comprehensive observation station of atmospheric combined pollution (30.63∘ N, 104.08∘ E). The observation equipment was placed on the top of a building, approximately 25 m from the ground, and there was no obvious pollution source within approximately 200 m. The site is located in south section 1 of Yihuan Road, Wuhou District, Chengdu (Fig. 1), and traffic emission sources may be the main pollution emission source around the observation station. This is a typical residential, traffic and commercial mixed area that represents the characteristics of the urban atmospheric environment. Chengdu is also a megacity in the Sichuan Basin of southwest China, as well as an important part of the Chengdu–Chongqing region, which is among the regions with serious air pollution in China. As shown in Fig. 1, the Sichuan Basin also has high aerosol optical depth (AOD).

During the research period, online experimental monitoring instruments were used to obtain the observation data with an hourly resolution (1 h). The equipment list is shown in Table 1. Data quality control and assurance are an important part of the atmospheric comprehensive observation experiment, and this result is described in detail in the Supplement (Fig. S1–S4 in the Supplement).

To clarify the conversion of gaseous pollutants to secondary aerosols, the nitrogen oxidation ratio (NOR) and sulfur oxidation ratio (SOR) were used to reflect the conversions of NO2 and SO2 to NO3- and SO42-, respectively (Sun et al., 2014; Y. R. Yang et al., 2015). These ratios can be calculated using Eqs. () and (): 1NOR=nNO3-/(nNO3-+nNO2), 2SOR=nSO42-/(nSO42-+nSO2), where n is the molar concentration.

The ISORROPIA-II thermodynamic model was used to analyze the variation in interaction characteristics among aerosol chemical compositions (Fountoukis and Nenes, 2007; Guo et al., 2017; Ding et al., 2019). Temperature (T), relative humidity (RH) and the concentrations of Na, SO42-, NH3, NO3-, Cl, Ca, K and Mg were input into the ISORROPIA-II thermodynamic equilibrium model. In this study, we used the “forward problems” mode to run the model, assuming that the aerosols were in a “metastable” state (salts do not precipitate under supersaturated conditions). At the time of data input, NH3 data were the sum of NH3 and NH4+. Previous studies had shown that the model has better performance when the RH is greater than 30 %, and some studies also believe that the model performance is greater than 40 %, so this study maintains the RH at higher than 40 % when data are input (Ding et al., 2019; Guo et al., 2016). The simulated data and observed data were compared and analyzed, and the observation data of NH3 were consistent with the input data of the model. The linear regression fitting slope of NH3 was 0.96 (R2=0.98), which showed that the run result of the model had good reliability and performance (Ding et al., 2019). Simultaneously, the aerosol water content (AWC) was calculated, and the sensitivity of the interaction between aerosol chemical compositions (NSA) and the pH of aerosols was analyzed (Ding et al., 2019; Fountoukis et al., 2009). The pH was calculated using Eq. (): 3pH=-log⁡10Haq+=̃-log⁡101000Hair+AWC, where Haq+ (molL-1) is the concentration of hydronium ions in liquid water of atmospheric particulate matter, which can be calculated by the Hair+ and AWC (µgm-3) outputs from the ISORROPIA-II thermodynamic equilibrium model (Ding et al., 2019; Guo et al., 2017).

We used the conditional probability function (CPF) to analyze the characteristics of pollutants under the influence of wind direction (WD) and wind speed (WS). The analysis results using CPF were obtained using the R programming language, named openair. This function can be defined as CPF=mθ,j/nθ,j, where mθ,j is the number of samples in the WD interval θ and WS interval j with mixing ratios greater than some “high” pollution concentration, and nθ,j is the total number of samples in the same WD–WS interval (Uria-Tellaetxe and Carslaw, 2014). Usually, a higher given high pollution concentration (percentile) is chosen, such as the 90th percentile, which will mask the lower percentile pollution concentration source contributions. In this work, to obtain a more complete contribution of pollution sources, a range of percentile values, 0–25, 25–50, 50–75 and 75–100, were selected for the CPF calculation.

The potential source contribution function (PSCF) is based on an analysis of pollution sources given the air mass backward trajectory and can be used to judge the long-distance regional transport of pollutants (Ji et al., 2019). In this study, MeteoInfoMap and TrajStat (Wang et al., 2009) were used, and the model simulation data input model was provided by the National Oceanic and Atmospheric Administration (National Oceanic and Atmospheric Administration, 2019, last access: 17 June 2020); these data were calculated to the 24 h backward trajectories at the observation site at a height of 500 m every 1 h from 1 January 2015 to 31 December 2017 (UTC+8). The calculated domain for PSCF was a range of 20–50∘ N, 75–115∘ E, and a grid cell with a resolution of 0.5∘×0.5∘ was divided. The PSCF could be defined using Eq. (): 4PSCFij=MijNijWij, 5Wij=1.0(Nij≥3Nave)0.7(3Nave>Nij≥1.5Nave)0.4(1.5Nave>Nij≥Nave)0.2(Nave>Nij), where PSCFij is the value for the ijth grid cell and Mij is the total number of endpoints in the ijth grid cell, with pollution concentrations at the observation site (30.63∘ N, 104.08∘ E) that are greater than a given threshold value (the 75th percentile was selected for gaseous pollutants). Nij is the number of backward trajectory endpoints in the ijth grid cell (0.5∘×0.5∘) during the simulation period. Therefore, the PSCF reflects the two-dimensional planar position distribution characteristics of potential sources, not the three-dimensional characteristics that reflect the transmission of pollution. To reduce the uncertainty in Nij, the empirical weight function Wij was introduced in Eq. (), where Nave is the average of Nij during the simulation period (Ji et al., 2019; Zhang et al., 2017; Wang et al., 2009).

The annual average mass concentration of NSA and its proportion in PM2.5 are shown in Table 2. The annual averages of PM2.5 were 67.78, 71.88 and 59.68 µgm-3, corresponding to 2015, 2016 and 2017, respectively. However, the pollution of PM2.5 in Chengdu was much higher than the annual secondary guideline value (35 µgm-3, Ambient air quality standards/GB3095-2012) and the World Health Organization annual guideline value (10 µgm-3). The same PM2.5 pollution problem was also a serious problem in Beijing and Nanjing (Ji et al., 2019; Zheng et al., 2019). The annual average mass concentration of NSA also changed significantly, and the difference was large. The Mann–Whitney U test showed that the variation in NO3- was nonsignificant (p>0.05), and SO42- and NH4+ had obvious significance from 2015 to 2017 (p<0.05), indicating that NO3- had not decreased significantly, and there was an increase in 2017 compared to 2015. SO42- continued to decline, and NH4+ was also lower in 2017 than in 2016. Notably, SO42- and NH4+ decreased significantly in 2017 compared with 2015, but the variation in NO3- was nonsignificant. Meanwhile, the annual averages of NO3-/SO42- were 0.95, 1.02 and 1.45 for 2015, 2016 and 2017, respectively, indicating that the contribution of NOx emissions sources to PM2.5 was increased compared with that of SO2 emissions sources (Li et al., 2017; Wang et al., 2015). As shown in Table S1 in the Supplement, from 2013 to 2017, the emissions of NO2 in Chengdu were obviously higher than those of SO2, but PM2.5, NO2 and SO2 all decreased due to the implementation of the Air Pollution Prevention and Control Action Plan launched by the Chinese government and a more detailed pollution control plan launched by Sichuan Province. From 2015 to 2017, the measures taken by Sichuan Province in the coordinated reduction of multiple pollutants have been continuously strengthened, and the scope of management and control has been continuously expanded, for example, in the improvement of desulfurization, denitrification and dust removal technologies in key industries, from accelerated improvement in 2015 to deeper improvement in 2017. The process of eliminating small coal-fired boilers began in 2015 and was completed in 2017, when the ultra-low-emission coal-fired power plant transformation was promoted. In terms of vehicle emission control, we accelerated the elimination of yellow label vehicles (general term for gasoline vehicles with emission levels lower than the national I emission standard and diesel vehicles with emission levels lower than the national III emission standard when new vehicles are finalized) and “old vehicles” (the emission level does not meet the national stage IV emission standard) in 2015 and basically completed the elimination of “yellow label” vehicles in 2017. The quality supervision of oil products has also been improved, and non-road mobile machinery pollution control requirements were proposed in the 2017 plan (The People's Government of Sichuan Province, 2015, 2016, 2017, last access: 17 June 2020). Compared with 2015, NOx and SO2 decreased by 5.98 % and 32.35 %, respectively, in 2017, which shows that the treatment of NOx and SO2 emissions has achieved remarkable results, of which the SO2 emission reduction effect is the best, followed by that of NOx. The effect of this emission reduction is due to air pollution prevention measures, especially measures of “electricity instead of coal” and “natural gas instead of coal” (refers to increased use of electricity and natural gas in the residential sector to reduce coal combustion).

The chemical compositions of PM2.5 from 2015 to 2017 vary with concentration, as shown in Fig. 2. With the accumulation of PM2.5 in the atmosphere, the concentration of NSA also increased significantly, but the proportion of NSA in PM2.5 decreased (Fig. 2a and b). The variation trend of OC, EC and metal elements with increasing PM2.5 concentration is similar to that of NSA (Fig. 2c and d), and this variation trend of OC and EC is consistent with the results of long-term observation research carried out in Beijing (Ji et al., 2019). With the accumulation of PM2.5 concentration, NSA, OC, EC and metal element concentrations have an increasing trend, but their ratio with PM2.5 gradually decreases, indicating that other compositions have a higher contribution. This result, on the one hand, may be since some chemical compositions such as ions and dust have not been included in the statistics; on the other hand, the unknown component may also have a high contribution characteristic to PM2.5 (Zhang et al., 2015; Huang et al., 2014). For the contribution characteristics of unknown components of PM2.5, studies in some regions of China show that the contribution of higher PM2.5 concentration is higher than that of lower PM2.5 concentration (Huang et al., 2014; Li et al., 2017; Geng et al., 2019). When PM2.5 was less than 50 µgm-3 and greater than 250 µgm-3, the mass concentrations of NSA were 11.57 and 90.06 µgm-3, respectively, and the proportions were 37.78 % and 31.45 %, respectively. Comparing Fig. 2b and d, it was found that NSA was always the main contributor in the entire process of PM2.5 accumulation, which was significantly higher than the proportions of OC and EC (Ji et al., 2019; K. Liu et al., 2019). In the accumulation process of PM2.5 concentrations greater than 50 µgm-3, NO3- accounts for a high proportion in NSA and is stable at approximately 14 %, and the proportion of SO42- and NH4+ continues to decrease (M. Liu et al., 2019; Wang et al., 2016). When the PM2.5 concentration was less than 50 µgm-3, the concentration of SO42- was higher than that of NO3-, and the concentration of NH4+ was lower than the NH4+ concentration of PM2.5 at 50 to 100 µgm-3, possibly due to SO42- concentration being higher than the NO3- concentration, forming more chemically stable (NH4)2SO4 (Guo et al., 2017). In addition, when the PM2.5 was less than 50 µgm-3, low RH and strong solar radiation were also important ways to generate sulfate (Yao et al., 2018).

The monthly variation characteristics of NSA from 2015 to 2017 are shown in Fig. 3. At the beginning and end of each year, the pollutant concentration is relatively high and relatively low in the middle of each year. The meteorological conditions also have obvious monthly variation characteristics (Fig. S5a, b in the Supplement); from April to August, they have higher WS and lower RH, which is not only conducive to the dilution and diffusion of pollutants but also reduces the chemical conversions of pollutants by aqueous phase and influences the formation of secondary inorganic aerosols (Wang et al., 2016; Ji et al., 2019). Overall, the concentrations are higher in January and December and lower in July and August. The highest monthly average NO3- reached 19.98 µgm-3 in January 2017, and the highest monthly average SO42- and NH4+ were 22.08 and 12.66 µgm-3 in January 2015, respectively. The lowest concentrations of NSA appeared in August 2017, which were 1.96, 3.07 and 1.62 µgm-3. The gaseous precursors of NSA also have obvious monthly variations, and the NOx and SO2 trends were similar to those of NO3- and SO42- (Figs. 3 and 4). NH3 emissions were significantly different, with increases in warmer months (April–July) and colder months (September–December). On the one hand, NH3 volatilization was promoted by relatively high Ts (Fig. S5c); on the other hand, the use of agricultural fertilizers and livestock farming were also important sources of NH3 in China. Second, from urban regions, fossil fuel combustion and motor vehicle emissions also contribute significantly (X. Liu et al., 2013; Pan et al., 2016). Notably, NH3 increased significantly from April to December 2017 compared with 2015 and 2016, especially during low-T months (Fig. 4c). The results of an analysis of the monthly concentration variation in pollutants indicate that the implementation of pollution reduction and control measures should be strengthened at the beginning of each year (January to March) and the end of the year (October to December).

The seasonal variation in NSA is shown in Fig. S6 in the Supplement, and the concentration in winter was much higher than that in summer. NO3- only declined in spring and summer from 2015 to 2017, with an increase in autumn and winter (Fig. S6a). Seasonal variations in NH4+ were similar to those of NO3-, with higher concentrations in winter and the lowest in summer (Fig. S6c). This may be because higher Ts and WSs can not only promote the decomposition of NH4NO3 in summer but also promote the dilution and diffusion of pollutant concentrations (Guo et al., 2017; An et al., 2019). There is a significant downward trend in SO42-, which continues to decrease in spring, summer and winter from 2015 to 2017 (Fig. S6b). In autumn, the concentration was the highest in 2016, and it was significantly lower in 2017 than in 2015 and 2016. The variation amplitude of NSA and gaseous pollutants in cold months was significantly higher than that in warm months (Figs. 3, 4 and S6). This higher variation amplitude may be due to the differences in pollutant accumulation and scavenging processes. This finding also indicates that the instability of local pollutant emissions and regional transport during cold months was affected by meteorological conditions (Li et al., 2017; Ji et al., 2018). The large variation amplitude of pollutants in different months, similar to the changes in the Beijing–Tianjin–Hebei region of northern China and Chengdu, is due to the accumulation and removal of pollution by meteorological conditions and pollutant emissions (Ji et al., 2019; Qin et al., 2019; H. Zhang et al., 2019).

From 2015 to 2017, the concentration of NSA was higher in the daytime than in the evening (Fig. 5a), and similar results were found in different seasons (Fig. 5b), which may be due to the combination of pollutant emissions and meteorological conditions. As shown in Fig. S7 in the Supplement, from 09:00 to 11:00 UTC+8, the concentrations of SO2, NOx, NH3 and CO increased significantly, indicating that the primary emission of pollutants was relatively strong. At this time, although RH is in a declining stage, it still has a relatively high atmospheric humidity (approximately 65 %), and O3 and NO2/NO also occasionally show an increasing trend, indicating that the atmospheric oxidizability has also increased (Figs. S7 and S8 in the Supplement). This situation also provides favorable conditions for the formation of secondary aerosols and promotes the accumulation of NSA (Cheng et al., 2016; Wang et al., 2016; Sun et al., 2014). In addition, before 10:00, relatively low WS will enable easy pollutant concentration accumulation. In contrast, the higher WS in the afternoon may be the main factor for the decrease in pollutant concentration (Figs. 5 and S8). Photochemical reactions may also be one of the factors in the formation of NSA, and the concentration of O3 peaks at approximately 15:00, which may be affected by the free radicals generated by photochemistry. At approximately 19:00, the ratio of NO2/NO reached its highest value, and the concentration of NO2 also increased significantly (Song et al., 2018; Zhu et al., 2019). At night, with the increase in RH (Fig. S8), dissolved ozone, free radicals, hydrogen peroxide and NO2 can catalyze SO2 to form secondary aerosols through an aqueous phase reaction (Zhang et al., 2015; An et al., 2019). The seasonal diurnal variation in NSA is shown in Fig. 5b. The concentration of NSA in winter was obviously higher than that in summer, and the diurnal variation range was larger. The concentrations in spring and autumn were closer, but the diurnal variation in spring was larger than that in autumn. The larger diurnal variation range not only indicates serious pollution but also indicates the importance of other factors affecting air quality, such as meteorological conditions and secondary aerosol conversion conditions (Ji et al., 2019; Y. Yang et al., 2015). The peak value of the NSA seasonal diurnal variation also varies in different seasons. The peak value appears at approximately 13:00 in winter, approximately 10:00 in spring and summer, and approximately 12:00 in autumn, possibly due to the influence of meteorological conditions. In previous studies in Beijing–Tianjin–Hebei and the Pearl River Delta, the concentration of pollutants was affected by meteorological factors, and it was usually lower in the daytime than at night. In the Yangtze River Delta, the peak usually occurs in the morning, but in our study, the concentration was higher in the daytime than at night (Peng et al., 2011; Wang et al., 2018; J. Guo et al., 2017). In addition to the diurnal variations in WS and atmospheric humidity, some studies have shown that due to the unique topographical structure of the Sichuan Basin, the atmospheric circulation between the Qinghai–Tibetan Plateau, Yunnan-Guizhou Plateau and Sichuan Basin and the meteorological conditions of the Chengdu region are affected, such as the characteristics of air mass transport and typical “night rain” (more precipitation at night than in the day) under the influence of atmospheric circulation (Y. Zhang et al., 2019; L. Zhang et al., 2019).

The weekly variation in NSA is shown in Fig. S9 in the Supplement. During the overall observational period, workdays (Monday to Friday) showed higher variations than weekends (Saturday and Sunday), with the highest variation being on Tuesday and the lowest being on Sunday. Despite the difference in mean values between Tuesday and Sunday, nonparametric tests show that the difference in mean values was nonsignificant (Mann–Whitney U test, P>0.05). As shown in Fig. S9, the average trends of NO3- and NH4+ were consistent from Monday to Sunday. The correlation coefficient was 0.94 (P<0.01) from 2015 to 2017, which indicates that they have a common source and that vehicle emissions also have an important contribution to NH4+ (Pan et al., 2016). The average NO3-, SO42- and NH4+ concentrations from 2015 to 2017 were 9.21, 8.64 and 5.64 µgm-3 on workdays and 8.56, 8.33 and 5.29 µgm-3 on weekends, respectively. The average values of NOx, SO2 and NH3 were 42.43, 4.35 and 20.39 ppb on weekdays and 39.60, 4.34 and 19.67 ppb on weekends, respectively. Similarly, the mean difference between NSA and gaseous precursors (NOx, SO2 and NH3) was not significant by the Mann–Whitney U test on weekdays and weekends. Population standard deviation comparisons of NO3-, SO42- and NH4+ showed that workdays had higher standard deviations than did weekends, with 7.96, 6.04 and 4.35 on weekdays and 6.76, 5.69 and 3.88 on weekends, respectively. It could also be seen from the box plot of NSA weekly variation that the concentration range on working days was slightly larger than that on weekends (Fig. S10 in the Supplement). Analysis of the diurnal variation in NSA gaseous precursors on weekdays and weekends shows that the variation trend is relatively consistent (Fig. S11 in the Supplement), and the concentration of NOx on weekdays will be slightly higher at the peak of 09:00 to 10:00 than on weekends, which may be affected by the morning rush hour of vehicles. In this study, NSA and gaseous precursors are also slightly higher on weekdays than on weekends, which indicates that in Chengdu's air pollution prevention and control actions, the management of relevant industries and departments should be strengthened on weekdays.