Xingtai is located in the central-south part of the Beijing–Tianjin–Hebei region with the Taihang Mountains to the west (Fig. 1a). In this work, all measurements were made at the Xingtai National Meteorological Basic Station (XNMBS), a suburban site located approximately 17 km northwest of Xingtai City (37.18∘ N, 114.37∘ E; 180 m a.s.l.), from 30 April to 20 June 2016. The sampling site was influenced by mountain–plain winds during the study period. As shown in Fig. 1b, the wind direction showed clear day and night patterns with prevailing south-southeasterly winds in the day and west-northwesterly winds at night. The average wind speed was 4 m s-1 and the average temperature was 22.6 ∘C during the study period.

All instruments were placed in a container at the sampling site. The non-refractory PM1 (NR-PM1) chemical components including sulfate (SO4), nitrate (NO3), ammonium (NH4), chloride (Chl), and organics (Org) were measured in situ by an ACSM at a time resolution of 5 min. The ACSM was operated in the same way as in previous studies (Sun et al., 2016b; Zhang et al., 2016). A PM2.5 cyclone (Model: URG-2000-30ED) was supplied in front of the sampling inlet to remove coarse particles larger than 2.5 µm. The ambient air was drawn into the container through a 0.5 in. (outer diameter) stainless steel tube at a flow rate of 3 L min-1 using an external pump, of which ∼0.1 L min-1 was sub-sampled into the ACSM. The sampling height was approximately 2 m, and the particle residence time in the sampling tube was about 5 s. Aerosol particles were then dried by a silica gel diffusion dryer before sampling into the ACSM. Before the campaign, the ACSM was calibrated with pure ammonium nitrate particles following the standard protocols in Ng et al. (2011b). Because the ACSM does not measure refractory black carbon (BC), a seven-wavelength Aethalometer (model AE33, Magee Scientific Corp.; Drinovec et al., 2015) was used to measure BC.

The size-resolved particle number concentrations in the size range from 15 to 685 nm were measured in situ by a condensation particle counter (CPC, model 3775, TSI) equipped with a long differential mobility analyzer (DMA, model 3081A, TSI). The time resolution is 5 min. The total number concentrations (7–2000 nm) were measured by a mixing condensation particle counter (MCPC, model 1720, Brechtel). Other collocated measurements included the light extinction of dry PM2.5 at 630 nm measured by a cavity attenuated phase shift extinction monitor (CAPS PMext; Massoli et al., 2010); the mass concentration of PM2.5 measured by a six-channel particle counter (manufactured by Graywolf); and gaseous species of CO, NO, NOx, SO2, and O3 measured by gas analyzers (manufactured by ECOTECH). Meteorological parameters including ambient temperature (T), relative humidity (RH), wind speed (WS), wind direction (WD), precipitation, and solar radiation were also measured at the same site by the Xingtai Meteorological Administration.

The footprints of the selected episodes were determined using backward simulations from FLEXPART, a Lagrangian transport and dispersion model (Stohl et al., 2005). The model calculated the 36 h backward trajectories of 10 000 particles released every hour from the sampling site at a height of 180 m above sea level. The meteorological data driving the model were simulated by version 3.4 of the Weather Research and Forecasting (WRF) model with a 1 h time resolution and a 10 km spatial resolution. The WRF model was driven by initial and boundary conditions from National Centers for Environmental Prediction global reanalysis data.

The 72 h back trajectories at a height of 500 m were calculated every hour using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015) at the XNMBS. To investigate the chemical characteristics of aerosols from different source regions, a cluster analysis was then performed on the trajectories and three clusters were identified according to their similarities in spatial distributions.

In addition, non-parametric wind regression (NWR; Petit et al., 2017) was performed to evaluate the sources of local emissions and regional transport for PM1 aerosol species and OA factors. The NWR plots represent the probability that a specific compound or source is located in a certain wind direction.

The temporal variations of PM1 aerosol species and meteorological parameters (RH, T, WS, WD, and precipitation) are shown in Fig. 3. The average mass concentration of PM1 (equal to NR-PM1 + BC) was 30.5 µg m-3 and ranged from 0.2 to 140.1 µg m-3. Several pollution episodes usually lasting ∼2–3 days were observed during the study period, e.g., on 9–11, 17–23, 28–31 May, and 2–4 June. These pollution episodes were quickly cleaned, mainly by wet scavenging. The temporal variations varied differently among different chemical species. Organics showed dramatic variations, ranging from 0.01 to 101.5 µg m-3, and comprised the major fraction of PM1 for most of the time in this study. High concentration peaks of organics were frequently observed, likely due to the influences of local emissions. By contrast, sulfate concentrations obviously increased and remained relatively high during the pollution events, suggesting the important role of regional transport at the downwind site of Xingtai. The average PM2.5 mass concentration was 45.2 µg m-3. Although the average PM2.5 mass concentration was 15 % lower than that (53.3 µg m-3) measured at the urban sites in Xingtai, it exceeded the Chinese National Ambient Air Quality Standards by 29 %. These results suggest that the urban downwind sites also experience similar PM pollution events as the urban sites. For example, the daily PM2.5 exceeded the Chinese National Ambient Air Quality Standards 24 % of the time during this study. PM1 was highly correlated with PM2.5 (r2=0.95) and on average, comprised ∼64 % of PM2.5. The average mass concentration of PM1 is close to that measured in Xinzhou (35 µg m-3), a city in central China (Wang et al., 2016), but lower than that measured in 2013 in Xianghe (73 µg m-3), a rural site near Beijing (Sun et al., 2016a; Fig. S9). One possible explanation is the significant improvement in air quality during the last 4 years (Fig. S1).

On average, OA was the largest component of PM1, accounting for 38 % of the total PM1 mass, followed by sulfate (25 %), nitrate (14 %), ammonium (10 %), and BC (10 %; Fig. 1a). POA (equal to HOA + COA) and SOA (OOA) accounted for 22 % and 78 %, respectively, of the total OA mass. Together, ∼18 % of PM1 was comprised of primary-related materials (POA + BC) and 82 % was from secondary formation (NO3+SO4+NH4+Chl+SOA), indicating that aerosol particles from secondary aerosol formation processes dominated at the downwind site of Xingtai. Compared with aerosol composition in megacities in the NCP, e.g., Beijing (Hu et al., 2016), aerosol composition in this study showed substantially higher contributions of SOA (29 %) and BC (10 %), while the contributions of NO3 (14 %) and COA (4 %) were low due to the lesser amount of local traffic and cooking emissions. The nitrate contribution was similar to that observed at a suburban site in Xinzhou (Wang et al., 2016), but the SO4 contribution was relatively low (25 % vs. 32 %). One reason is that the higher RH (70 % vs. 52 %) in Xinzhou facilitated the formation of SO4. The PM2.5 concentration during this study period was more than twice as low as that in winter (151 µg m-3, Fig. S10), suggesting different sources between spring–summer and winter. For example, the winter season has significantly enhanced coal combustion emissions.

We also investigated the compositional differences between clean periods and polluted events (Fig. 3). Secondary inorganic aerosols (SIA) including SO4 (26 % vs. 21 %), NO3 (15 % vs. 9 %), and NH4 (10 % vs. 7 %) showed enhanced contributions to PM1 during polluted events, while a corresponding decrease was found for organics. For example, the contribution of SIA increased by 14 % from 37 % during clean periods to 51 % during polluted events, while organics and BC decreased by 9 % and 3 %, respectively. Although SOA dominated the OA composition during both polluted events and clean periods, SOA also contributed more to OA during polluted events (78 % vs. 72 %). Comparatively, POA (HOA and COA) was relatively more important than SOA during clean periods. These results suggest that PM at the downwind site of Xingtai was mainly affected by regional transport and secondary formation.

Aerosol species showed distinctly different diurnal patterns in this study (Fig. 4), indicating that the sources and formation processes of PM pollutants were different. The diurnal cycle of PM1 was characterized by peaks at ∼ 10:00 and ∼ 21:00 local time (LT). The first peak in the late morning was consistent with that of gas pollutants including CO, SO2, and NOx (Fig. S12). Through comparisons with the diurnal cycle of PM2.5 at the urban site in Xingtai (Fig. S13) and daily variations in WD (Fig. 1), the morning peak was mainly associated with the transport of pollutants from urban sites located to the southeast.

The high concentration of PM1 at night was clearly associated with enhanced primary emission concentrations, e.g., HOA and COA, although the shallower boundary layer may have also played a role. It is interesting to note that bext did not show a pronounced nighttime peak as did PM2.5, indicating that the extinction coefficients of primary aerosols were smaller than those of secondary aerosol species (Q. Wang et al., 2015). The diurnal pattern of PM1 only showed a pronounced peak at night during clean periods, consistent with those of HOA and COA. While the diurnal variations support the enhanced roles of primary emissions for PM during clean periods, they also indicate the lesser influences from urban emissions during daytime.

The diurnal pattern of organics overall resembles that of PM1 and was characterized by two pronounced peaks between 09:00–12:00 and 19:00–21:00 LT. However, the three OA factors showed different diurnal cycles. The average diurnal cycle of HOA showed a small morning peak and a pronounced nighttime peak. While the two peaks were comparable during polluted periods, only the pronounced nighttime peak was observed during clean periods. As shown in Fig. S11a, winds were dominantly from the south-southeast during polluted periods and mainly from the west-northwest during clean periods. These results suggest that the high morning HOA peak was mainly caused by the transport of pollutants from urban sites, while local traffic emissions became an important source of HOA at night. This is also consistent with what is seen in the corresponding bivariate polar plot (Fig. 5h), which shows high concentrations of HOA in regions to the south-southeast and to the north. The decrease in HOA during the day was mainly associated with the rising planetary boundary layer height.

The diurnal pattern of COA (Fig. 4h) was also similar to that observed in megacities (Sun et al., 2013; Crippa et al., 2013; Elser et al., 2016; Y. J. Zhang et al., 2015; Hu et al., 2016; Xu et al., 2016) and was characterized by two peaks around the mealtime hours, reflecting the influence of cooking emissions. However, the COA concentration in this study was much lower than that reported in megacities (e.g., peak concentration: 2.5 µg m-3 vs. 14 µg m-3 in Beijing; Hu et al., 2016), suggesting much lower cooking emissions at the suburban site. The corresponding bivariate polar plot further shows that the high concentration of COA mainly originated from restaurants and inhabitants' activities to the southeast of the site (Fig. 5i).

Overall, the diurnal cycle of OOA (Fig. 4i) was flat during both clean and polluted periods in this study, reflecting regional characteristics. This is also consistent with the wide distribution of OOA seen in the corresponding bivariate polar plot (Fig. 5j). Secondary inorganic species of nitrate and sulfate had different diurnal profiles. Nitrate had a pronounced diurnal cycle with much higher concentrations at night than during the day. As shown in Fig. 4d, the nitrate concentration decreased from 6.0 to 2.6 µg m-3 between 11:00 and 19:00 LT, which was mainly due to the evaporative loss of particulate ammonium nitrate particles due to high temperatures. Such diurnal cycles have been observed many times during the summer in megacities, e.g., Beijing (Sun et al., 2012), Nanjing (Ge et al., 2017), and Lanzhou (Xu et al., 2016), as well as those on other continents (e.g., Lanz et al., 2007). Nitrate also showed an increase in the early morning after sunrise (Fig. 4d). While the transport from urban sites played a role, this increase was mainly caused by daytime photochemical production when T was not high enough to substantially affect gas partitioning. During daytime clean periods, the diurnal cycle of nitrate was flat, with higher concentrations at night. Sulfate showed a much smoother diurnal evolution compared with nitrate (Fig. 4c), reflecting the regional characteristics of sulfate. Two small peaks were observed during the day. While the first peak between 09:00 and 11:00 LT was most likely from urban transport, consistent with regional sources from the southeast indicated by the corresponding bivariate polar plot (Fig. 5c), the second one was more likely from daytime photochemical production. Chloride accounted for a small fraction of the PM1 mass, yet it showed a strong diurnal cycle with a pronounced peak in the morning (Fig. 4e). This peak was noticeably similar to those of CO and SO2 (Fig. S11a and c), suggesting that the dominant source was from combustion emissions in the southeast.

Figure 6 shows the time series of total number concentration (15–685 nm, N15–685) and three different modes including the small Aitken mode (15–40 nm, N15–40), the large Aitken mode (40–100 nm, N40–100), and the accumulation mode (100–400 nm, N100–685), as well as particles in the range of 7–15 nm (N7–15) that were calculated from the difference between MCPC and SMPS measurements. The corrections for diffusion loss and multiple charges have been applied in SMPS data analysis. The average total number concentration was 11 200 ± 5800 cm-3, which is comparable to that measured in Shangdianzi (12 000 cm-3; Shen et al., 2011) and Yufa (10 200 cm-3; Peng et al., 2014) and slightly higher than that observed in Beijing (10 100 cm-3; Du et al., 2017). N15–685 showed a pronounced diurnal cycle with a clear increase during the day (Fig. 7e). A further analysis highlights that this increase was mainly driven by small and large Aitken mode particles, indicating the impacts of new particle formation and growth on the diurnal variations in particle number (Fig. S14). For example, the small Aitken-mode particles and ultrafine particles showed rapid daytime increases after sunrise by more than a factor of 5 during both polluted events and clean periods. Note that the number concentrations of N7–15 during clean periods was much higher than during polluted events (6600 cm-3 vs. 3300 cm-3, on average), which is comparable to the number concentration of small Aitken-mode particles. These results show that (1) new particle formation was much stronger on clean days than on polluted days, and (2) new particle formation also occurred on polluted days, with newly formed particles growing quickly due to the higher condensation sink (CS; 0.05 s-1). About 49 % of the total particle number was made up of large Aitken-mode particles. However, these particles accounted for a small fraction of the total volume concentration (5 %). We also note that the pronounced nighttime peak in N40–100 coincidently agrees with that of COA during clean periods, suggesting the influence of cooking emissions on large Aitken-mode particles. We calculated the particle number size distribution for two nights that experienced significant cooking emission events, i.e., 2 and 13 May. Narrow single-mode distributions peaking at ∼60 nm were seen (Fig. S15), supporting the influence of cooking emissions on large Aitken-mode particles. The diurnal cycle of N100–685 was relatively flat except for a small morning peak (Fig. 7d). This is also consistent with results from the corresponding bivariate polar plot (Fig. S16d). The number concentration of N100–685 during polluted events was more than a factor of 3–4 times that during clean periods, which contributes toward the major difference in particle number characteristics between polluted events and clean periods. One reason is the higher CS during polluted events, which facilitated the growth of particles.

Figure 8 shows that the average particle number concentration had a bimodal size distribution with the geometric mean diameter (GMD) peaking at 46 and 106 nm. Note that the peak diameter for the entire study was ∼62 nm, which is higher than that observed in urban Beijing (45 nm; Du et al., 2017). The higher CS (0.36 s-1 vs. 0.29 s-1) could be one of the reasons leading to more water vapor condensing on preexisting particles at the rural site. This is also consistent with the shift in the peak diameter from 57 nm during clean periods to 88 nm during polluted events. Although similar bimodal size distribution modes peaking at ∼46 nm (41 nm) and ∼109 nm (106 nm) were observed during clean periods and polluted events, respectively, the relative contributions of the two modes were largely different. While the particle number concentration was dominated by the small mode (56 %) during clean periods, the large mode was more important during polluted events, accounting for 73 % of the particle number concentration (Table 1). These results confirm the different roles of different mode particles between clean periods and polluted events.

New particle growth events (NPEs) were frequently observed during the study period. As shown in Fig. 9a, particle growth typically started at ∼ 08:00 LT and ended at midnight with an increase in GMD from ∼25 to ∼60 nm. This result is consistent with those previously reported for rural sites in the NCP (Wang et al., 2013) and urban sites, e.g., Beijing (Du et al., 2017). Note that the growth sizes of particles were overall larger than those observed in urban Beijing (from ∼22 to ∼55 nm), likely indicating a stronger aging process at the suburban site. The growth of particles tracked the diurnal cycle of CS, which showed a continuous increase from early morning to midnight. Although ACSM only has a good transmission for particles within the size range between 70 and 500 nm in diameter (Jayne et al., 2000), some particles from the small Aitken mode might not be detected by the ACSM. Simultaneous comparison between aerosol chemical composition and particle size distribution during the growth period make some sense. Aerosol composition seemed to significantly change during the growth period. As shown in Fig. 9a, OOA and sulfate were the only two species whose contributions increased, going from 26 % to 33 % and 27 % to 33 %, respectively, during the growth period (10:00–18:00 LT). Although the increases in sulfate and OOA were partly due to the decreases in nitrate and chloride because of the evaporative loss in the afternoon, the CO-normalized sulfate and OOA (SO4/CO and OOA/CO) also showed increases. Note that CO here was subtracted by a background value of 0.068 ppm that was calculated as the average of the lowest 5 % data in this study. These results show that both sulfate and OOA played important roles in daytime particle growth.

We also examined particle growth events on polluted days and clean days. As shown in Fig. 9b and c, the growth process of particles on polluted days started at ∼ 12:00 LT with the GMD increasing from ∼38 to ∼57 nm in 6 h. Particle growth started earlier (∼ 10:00 LT) on clean days with the GMD increasing from ∼27 to ∼41 nm. Particle growth on polluted days was faster than on clean days, which was likely due to the higher CS on polluted days. This is also consistent with more significant increases in SOA/CO and sulfate/CO on polluted days. The increase in CS between 09:00 and 12:00 LT on polluted days was associated with the corresponding increases in most aerosol species. However, the CO-normalized aerosol species did not show such an increase during this period suggesting that aerosol species and gaseous species were from the same air mass, i.e., transported in from urban sites. Such a pattern was not observed on clean days. Particle growth at night was more clearly seen on clean days than on polluted days, consistent with the increase in OOA/CO. However, such an increase was not observed for sulfate/CO on clean days, suggesting that OOA played a more important role than sulfate in particle growth at night.

We further calculated the particle growth rate (GR) of each growth event that lasted more than 3 h (Fig. 10a and Table S1). The particle growth rates were calculated using Eq. (1). GR=ΔDm/Δt, where Dm is the geometric mean diameter from the log-normal fitting of each size distribution and ΔDm is the increase in diameter during the growth period of Δt. The particle GR varied from 1.2 to 4.7 nm h-1, which generally falls within the range of values (1–20 nm h-1) based on observations from around the world (Kulmala et al., 2004; Yu et al., 2017). As indicated in Fig. 10b, GR was positively correlated with CS for most of the time, mainly for the fraction of OOA higher than 30 % (r2=0.61, Fig. S17a). As CS increased from ∼0.01 to 0.05 s-1, GR increased from 3 to 5 nm h-1. Higher CS is usually associated with lower particle GR due to faster consumption of condensable vapors. The positive relationship between GR and CS might indicate that the source of condensable vapors contributes to the increase in CS prior to the observation. Another possible explanation is that heterogeneous surface chemistry was more important than CS in growing aerosol particles in highly polluted environment (Kulmala et al., 2017). Figure 10b also showed some low GRs with high CS, which were characterized by low contributions of OOA. These results support the importance of the involvement of OOA in particle growth (Wu et al., 2017). Figure 10c further shows that GR was positively correlated with the concentration of sulfate during periods with low sulfate mass loadings (<3 µg m-3, r2=0.42, Fig. S17b), while periods with higher concentrations of sulfate had lower GRs. This is corroborated by Fig. S18, showing that GR decreased as the sulfate contribution increased. By contrast, GR was positively correlated with OOA/PM1 for most of the time. These results highlight that OOA played a more important role in particle growth than sulfate, although sulfate was also important during periods with low mass loadings. We further checked the dependence of GR on source region, but no clear relationship was found. For example, GR varied from 1.6 to 3.2 nm h-1 during periods with air masses from the north, 1.8–4.9 nm h-1 from the southwest, and 1.8–4.6 nm h-1 from the east-southeast (Fig. 11). This shows that GR had no clear dependence on air masses from different regions, although sulfate concentrations showed much difference.

Figure 12 shows the average composition and particle number distributions from different source regions. The air masses in cluster 3 (C3, 39 % of the time) were mainly from the southeast while the other two clusters were mainly from the northwest. The average PM1 concentration for C3 was 42 µg m-3, which is ∼30 % higher than that from the other two source regions. The high mass loadings for this cluster suggest that source regions to the southeast were responsible for the high PM pollution at the sampling site. This is also supported by the bivariate polar plots of aerosol species with high concentration regions in the southeast (Fig. 5). Among all clusters, the two dominant species in PM1 were SOA (26 %–28 %) and sulfate (26 %–29 %). Overall, the aerosol bulk composition for the three clusters from different source areas was similar. These results suggest that aerosol particles were relatively well mixed over the region around Xingtai. Particle number concentrations showed more differences among the different clusters. The two clusters from the northwest were both dominated by large Aitken-mode particles, on average accounting for 49 % and 51 % of the total particles, respectively. Although large Aitken-mode particles dominated the total particle number for C3, we also observed large increases in accumulation-mode particles (39 %) compared with the other two clusters (28 %–30 %). These results suggest that air masses from the northwest were relatively clean, which led to a more frequent occurrence of new particle formation and growth events, while those from the southeast with higher PM loadings tended to form more large particles due to the high CS.