In situ measurements were conducted at Hebei University of Engineering (36.57∘ N, 114.50∘ E) in Handan from 3 December 2015 to 5 February 2016. Our sampling site is situated at the southeast edge of urban Handan, on the roof of a four-story building (∼ 12 m high), surrounded by the school and residential area, ∼ 300 m north of South Ring Road, and ∼ 400 m northeast of Handa Highway (S313). The ambient temperature varied from -12.7 to 14.4 ∘C, with an average of 1.8 ∘C. The prevailing wind came from the northeast and southwest, characterized by low wind speeds (Fig. 1b).

The mass concentrations of non-refractory submicron aerosol (NR-PM1) – including organics, sulfate, nitrate, ammonium, and chloride – were measured in situ using an Aerodyne ACSM. The detailed description of this instrument can be found in Ng et al. (2011a). In brief, ambient air was sampled through a PM2.5 cyclone to remove coarse particles with diameters exceeding 2.5 µm and then traversed a 2 m long, 1/2 in. (outer diameter) stainless-steel tube at a flow rate of 3 Lmin-1 using an external pump. A Nafion dryer was installed before the ACSM to dry aerosol samples and maintain the RH below 30 %. Subsequently, only a subset of the flow at ∼ 85 ccmin-1 was sampled through a 100 µm critical orifice, focusing aerosol particles between 40 nm and 1 µm into the vacuum chamber via an aerodynamic lens. In our study, the ACSM mass spectrometer was operated at a scanning speed of 200 msamu-1 from m/z 10 to 150. By automatically switching 14 cycles between filter mode and sample mode, the time resolution for the ACSM data in this study was approximately 15 min.

Because of the limit of the vaporizer temperature (∼ 600 ∘C), the ACSM could not measure refractory species such as BC. Thus a multi-angle absorption photometer (MAAP, Thermo Scientific model 5012) was deployed for real-time measurement of BC concentration. The MAAP was operated at an incident light wavelength of 670 nm, with a PM1 cyclone and a drying system incorporated in front of the sampling line (Petzold and Schönlinner, 2004; Petzold et al., 2005). Online PM2.5 mass concentration was measured simultaneously using a heated Tapered Element Oscillating Microbalance (TEOM series 1400a, Thermo Scientific). Other collocated instruments included a suite of commercial gas analyzers (Thermo Scientific) to monitor the variations of gaseous species (i.e., CO, O3, NO, NO2, NOx, and SO2). Meteorological parameters – i.e., temperature, RH, pressure, wind speed (WS), and wind direction (WD) – were obtained by a Lufft WS500-UMB Smart Weather Sensor. The data reported in this paper are in Beijing Time (BJT: UTC+8).

The mass concentrations of non-refractory aerosol species and the spectral matrices of OA were processed using ACSM standard data analysis software (v1.5.3.5) within Igor Pro version 6.37. The detailed procedures have been described in Ng et al. (2011a). The default relative ionization efficiency (RIE) values were used for organics (1.4), sulfate (1.19), nitrate (1.1), and chloride (1.3), whereas the RIE of ammonium (6.28) was directly determined via analyzing pure NH4NO3 particles. To account for the incomplete detection of aerosol species, a default collection efficiency (CE) value of 0.5 was applied to the entire data set as aerosol particles were dried before ACSM sampling and the ammonium nitrate fraction was always lower than 0.4 during the whole period. Although previous studies have shown that aerosol particles may be slightly acidic during wintertime in the NCP region, particle acidity was not high enough to affect CE values substantially (Sun et al., 2016a). As shown in Fig. S1 in the Supplement, the mass concentrations of PM1 (= NR-PM1 + BC) correlated tightly with total PM2.5 mass loadings measured by TEOM (slope = 0.88, r=0.93). Compared to the results reported previously in this area (Sun et al., 2013a, 2014, 2015; Zhang et al., 2014; Hu et al., 2016), the ratio of PM1 to TEOM-determined PM2.5 in this work appeared to be a bit higher. The difference may be due to (1) the contribution of semi-volatile species to PM2.5 varied greatly among different periods and different locations, because TEOM is heated to 50 ∘C during the measurement, which might have caused significant losses of semi-volatile species, for example, ammonium nitrate and semi-volatile organics; and (2) the contribution of particles in the size range of 1–2.5 µm to the total PM2.5 might also change among different pollution episodes and different sites.

To determine potential sources of OA, the ACSM mass spectra were processed using the ME-2 algorithm implemented with the toolkit SoFi (Source Finder) developed by Canonaco et al. (2013). The so-called a value approach allows the user to introduce a priori information in forms of known factor profiles or time series to obtain a rather unique solution and thus reduce the rotational ambiguity of the PMF2 algorithm. The spectra and error matrices of organics were prepared according to the protocol summarized by Ulbrich et al. (2009) and Zhang et al. (2011). Given the interferences of the internal standard of naphthalene at m/z 127–129 and the low signal-to-noise ratio of larger ions, we only considered ions up to m/z 120 in this study. A reference hydrocarbon-like OA (HOA) profile, which is an average of multiple ambient data sets taken from Ng et al. (2011b), was introduced to constrain the model performance with a values varying from 0 to 1. Following the guidelines presented by Canonaco et al. (2013) and Crippa et al. (2014), an optimal solution involving four factors with a value of 0.1 was accepted. Detailed analyses of the factor time series, mass spectra, diurnal patterns, and correlations with external tracers can be found in the Supplement (Figs. S2–S6). Note that, before using the ME-2 engine, we also attempted to perform PMF analysis with the PMF2 algorithm for one to eight factors. The solutions were thoroughly evaluated following the recommendations outlined in Zhang et al. (2011), and the results of three- and four-factor solutions at fpeak=0 are shown in Figs. S7–S8. The three-factor solution indicates the identification of a coal combustion OA (CCOA), a biomass burning OA (BBOA), and an oxygenated OA (OOA). But the CCOA factor seems to be mixed with the signals from hydrocarbon-like components related to traffic emissions, which is especially evident given the two noticeable peaks in the diurnal profile of the CCOA factor during morning and evening rush hours. In the four-factor solution, the additional factor could not be physically explained and showed indications of factor splitting. Solutions with five to eight factors show further splitting and mixing of factors. Our inability to separate an individual HOA factor using the PMF2 algorithm is probably due to the minor contribution of traffic emissions in Handan, consistent with the fact that the PMF2 algorithm tends to have difficulty in accurately retrieving minor factors (Ulbrich et al., 2009).

Frequent and persistent haze episodes were observed during the campaign, especially from 16 to 25 December 2015, when an extremely polluted and long-lasting haze event occurred. Based on TEOM measurements, only 4 days met the US National Ambient Air Quality Standards (NAAQS, 35 µgm-3 for the 24 h average of PM2.5) and 13 days met the Chinese NAAQS (CNAAQS) Grade II (75 µgm-3 for the 24 h average of PM2.5) for the whole study period of 65 days. In other words, the daily average PM2.5 concentrations exceeded the US NAAQS and the CNAAQS on 94 and 80 % of the days, respectively (Fig. 2). On 22 December, the daily PM2.5 concentration reached the highest value of 725.7 µgm-3, leading to the first “red” haze alarm (http://www.cma.gov.cn/kppd/kppdsytj/201310/t20131028_229921.html) ever in Hebei Province. The meteorological conditions were stagnant with calm winds throughout the study period (WS usually less than 1.5 ms-1), although relatively high WS (generally > 1.5 ms-1) with cleaner air from the northwest of Handan occasionally interrupted the haze evolution process (Fig. 2b). The RH varied from 11.7 to 94.8 %, generally with higher values for more polluted periods and lower values during cleaner periods. No precipitation occurred throughout the entire campaign.

Hourly PM1 concentrations fluctuated dramatically from 4.2 to 700.8 µgm-3 (Fig. 2g). The average PM1 concentration was 187.6 µgm-3, more than twice as high as that observed in the well-known severe haze event occurring in Beijing in January 2013 (Sun et al., 2014; Zhang et al., 2014). Organics constituted a major fraction of PM1, contributing 45 % on average during this study, followed by sulfate (15 %), nitrate (14 %), and ammonium (12 %). The large fraction of organics in PM1 was comparable to previous observations in other areas of the NCP during wintertime (Sun et al., 2013a; Zhang et al., 2013; Huang et al., 2014). The average chloride contribution (9 %) is relatively high compared to that previously observed in the NCP region. Submicron nonrefractory chloride in the aerosol phase can be directly emitted from different sources (e.g., biomass burning and coal combustion) (Lobert et al., 1999; McCulloch et al., 1999) and formed in the atmosphere through gas-to-particle conversion (e.g., NH4Cl partitioning) (Baek et al., 2006). Considering that chloride demonstrated pronouncedly enhanced peaks at night and that it showed good correlations with CCOA and BBOA (r=0.72 and 0.80, respectively), a large fraction of chloride during wintertime was thought to be from primary emissions at night. On average, BC accounted for 5 % of total PM1. Its distinct peaks at morning and evening rush hours suggested that BC was mainly associated with traffic emissions. In the daytime, PM1 was dominated by secondary species because of active photochemistry, whereas the contributions of primary species were significantly increased at night, probably caused by enhanced primary emissions from fuel combustion coupled with shallow boundary layer height (Fig. S9).

Ambient CO is an indicator of the intensities of anthropogenic emissions. The hourly CO concentration was as high as 10 ppm during the study period, higher than those observed in other areas of China (Andreae et al., 2008; Quan et al., 2014; Yang et al., 2015). Interestingly, the temporal pattern of organics tracked well with that of CO (r=0.84, Fig. 2), implying that combustion emissions were a significant source of organic aerosols in Handan, i.e., traffic, coal combustion, and biomass burning. In addition, during severe haze episodes with high NOx and CO concentrations, ozone remained at nearly zero for several days instead of showing a regular diurnal variation, indicating active ozone titration by NO and a strong influence of primary emissions on haze pollution in this study.

In this study, three POA factors (HOA, BBOA, and CCOA) and one SOA factor (OOA) were resolved by analyzing the ACSM OA mass spectra using the ME-2 algorithm. OOA was the largest contributor to OA mass with an average fraction of 39 % (Fig. 3). The traffic-related HOA only accounted for 7 % of total OA, which was in accordance with the fact that PMF analysis performed with the PMF2 algorithm had difficulty retrieving it (see Sect. 2.3 for more details). On average, primary sources dominated the OA mass (61 %) during this winter study, consistent with the results from previous winter studies in the NCP region (Sun et al., 2013a, 2016a; Zhang et al., 2014; Hu et al., 2016). The discussion below focuses on the characteristics, sources, and processes of each OA factor.

Identifying the responsible emission sources and formation pathways during haze events is essential to effectively implement emission controls, especially considering the increased frequency of haze events during winter. In this study, the whole period is divided into polluted and non-polluted days, as described in Sect. 3.3.2. The average PM1 concentration during polluted days (211 µgm-3) was more than 3 times higher than that during non-polluted days (49 µgm-3). However, the average aerosol composition did not show obvious changes between these two types of days, indicating the synchronous increase of all aerosol species (Fig. 8a). Indeed, during polluted days, the average mass concentrations of all aerosol species except for BC were approximately 4 times as high as those during non-polluted days (Fig. 8b). Sulfate, CCOA, and BBOA showed the highest polluted / non-polluted ratios, which were 5.3, 5.0, and 5.5, respectively (Fig. 8b). Given the higher average RH on polluted days (average ±1σ=56.5±18.8 %) than on non-polluted days (average ±1σ=40.9±18.7 %), aqueous-phase processing likely has increased the production of sulfate (Wang et al., 2012; B. Zheng et al., 2015; Elser et al., 2016). During polluted days, the average oxidation ratio of sulfur (molar ratio of sulfate to sum of sulfate and SO2) was 0.27, higher than that on non-polluted days (0.16). On the other hand, the strong increases of CCOA and BBOA were possibly caused by enhanced gas-to-particle partitioning associated with high PM mass loadings during polluted periods (Mader et al., 2002). Interestingly, compared to aerosol species, CO showed a lower polluted / non-polluted ratio of approximately 2. A possible reason is that CO has a longer atmospheric lifetime than aerosol particles do; thus it has a more elevated regional background concentration. Note that the polluted / non-polluted ratios for SO2 and NOx were also lower than for the aerosol species. This is potentially a result of enhanced aqueous-phase oxidation of SO2 and NOx as well as more efficient wet deposition, since the more polluted periods were generally more humid.

Figure 9 further displays the average hourly variations of the mass fractions of aerosol species as a function of PM1 concentration. The nitrate fraction went up a bit and then showed a decreasing trend with increasing PM1 mass loading, whereas the contribution of sulfate increased from 12 to 20 % as PM1 concentration developed from 100 to 600 µgm-3. Since it was unlikely that the emission sources of the main gaseous precursors of these two species (i.e., NOx and SO2) had changed significantly during our study, the observed changes in aerosol compositions suggest different formation mechanisms of nitrate and sulfate during wintertime. The substantially elevated production of sulfate during high-PM episodes was likely attributable to higher ambient RH, which facilitated sulfate production through aqueous-phase reactions of SO2 (Kim et al., 2017; Li et al., 2016b; Sun et al., 2013b). The oxidation ratio of sulfur increased from 0.1 to 0.4 when PM1 concentration rose from ∼ 10 to 600 µgm-3. The mass fractions of different OA factors varied widely as PM1 concentrations increased. The contribution of HOA to total PM1 was minor and remained relatively stable across all mass loadings. However, the mass fractions of CCOA and BBOA increased nearly linearly, with PM1 concentrations rising from ∼ 20 to 300 µgm-3, and plateaued at higher aerosol loadings. OOA, a surrogate of SOA, showed the opposite PM-loading dependency, and its contribution decreased slightly with increasing PM1 concentration. The study of Sun et al. (2013a) in Beijing also found a growing contribution of CCOA and a declining contribution of OOA with increasing PM1 concentrations during wintertime. These results reveal the important role of POA in the development of high PM pollution during wintertime. Indeed, the scatterplot of OA vs. PM1 concentrations (Fig. 9c) demonstrates that higher mass fractions of organics in PM1 were associated with elevated POA contributions to total OA, especially when PM1 concentrations were more than 200 µgm-3 (Fig. 9c). Overall, the results here suggest that secondary formation of sulfate (mainly from SO2 emitted by coal combustion), and primary emissions of organics from coal combustion and biomass burning are important factors driving the development of winter haze pollution in Handan.

From 14 to 28 December 2015, an extremely severe haze episode occurred and was characterized by a steady buildup of air pollutants, including fine particles and CO, over a period of ∼ 5 days (17–21 December 2015) followed by approximately 4 days of heavy air pollution, during which the average CO mixing ratio was 6.7 ppm and the average PM1 concentration was 500.1 µgm-3 (Fig. 10). This episode ended on 25 December, during which winds from the northwest brought in cleaner air, leading to dramatic reductions of air pollutants. This type of evolutionary process has been frequently observed in Beijing during autumn and winter, and it is called “sawtooth cycles” by Jia et al. (2008). In this study, the whole haze cycle was divided into five stages: (1) a clean period (Stage 1), (2) an almost linearly increasing period of PM1 concentration (Stage 2), (3) a remarkably high pollution period lasting for 4 days (Stage 3), (4) an abruptly cleaned up period (Stage 4), and (5) another clean period as the start of a new cycle (Stage 5). As shown in Fig. 10, each stage was initiated by a sudden change in the WD and air masses from different regions via the HYSPLIT back trajectories (Draxler and Rolph, 2013). This indicates that meteorological changes are important driving forces during the evolution of haze episodes.

Stage 1 was characterized by high winds from the northwest, which brought clean air masses from Western Siberia. Aerosols associated with this air mass origin were largely free of high anthropogenic emissions and appeared to be aged with a high contribution of secondary species. Consistently, the CO concentration during stage 1 was relatively low. During stage 2, the WD changed and the WS was lower than 1 ms-1. The air masses from the northern and southern areas of Handan were influenced by high anthropogenic emissions in northern Hebei and Henan Province, respectively. Thus, the PM1 concentration steadily increased during this stage, with an average of 164.6 µgm-3. Stage 3 was dominated by southerly and northerly winds and really stagnant conditions with low WS. On 23 December air masses from the southern and northern areas of Hebei circulated around Handan, leading to the accumulation of air pollutants including PM and CO. The average PM1 concentration during stage 3 was 500.1 µgm-3, with the hourly maximum reaching as high as 700.8 µgm-3, much higher than that observed during the severe haze episode in Beijing in January 2013 (∼ 300 µgm-3; Sun et al., 2014). Accompanied with a high CO concentration (average of ∼ 7 ppm) during stage 3, O3 concentration remained at a very low level of almost zero and with minimal diurnal variations, suggesting that gas-phase oxidation might not be a dominant mechanism for haze formation. Moreover, stage 3 was characterized with high RH, exceeding 70 % most of the time, which would promote the aqueous-phase formation of secondary species. Indeed, a high mass fraction of secondary species, especially a notable increase in sulfate contribution, was observed during stage 3. During stage 4, due to the return of cleaner air masses long-distance transported from the northwest, air pollutant concentrations in Handan decreased dramatically and PM1 concentration decreased from 443.7 to 34.1 µgm-3 within only 12 h.

To further evaluate the influence of air mass origins on aerosol characteristics, we performed a cluster analysis of HYSPLIT back trajectories for the whole study period to elucidate the relationship between aerosol concentration or composition and different clusters. As shown in Fig. S12, the whole NCP region was heavily polluted, with high PM1 concentrations for all four clusters. Overall, the aerosol compositions were similar among different clusters. However, we indeed observed an important role played by winds in altering aerosol characteristics according to the above case study. Referring to the haze cycle analysis, we attempted to apply another classification method based on WD and WS. Periods with WS exceeding 1.5 ms-1 from the northwest of Handan were denoted as “NW_HWS”, whereas the remaining periods were classified as “others” (Fig. 11). As expected, the PM1 concentration of others was more than 6 times higher than that of NW_HWS. Secondary aerosol species (i.e., sulfate, nitrate, ammonium, and OOA) contributed 66 % of total PM1 for NW_HWS. As air masses associated with others were more strongly influenced by anthropogenic sources, the main primary species (i.e., HOA, BBOA, CCOA, and BC), accounted for a higher fraction of 32 % for others. These results highlight the importance of high winds from the northwest of Handan in alleviating PM levels and changing aerosol composition during wintertime.

As mentioned previously, the sulfate contribution during stage 3 was visibly enhanced under high RH, revealing the effects of RH on aerosol processing. Many previous studies have observed the increased production of secondary inorganic aerosol species through aqueous-phase processing. In this study, we used the oxidation ratios of sulfur and nitrogen, defined as fS=nSO42-/(nSO42-+nSO2) and fN=nNO3-/(nNO3-+nNOx), respectively, to explore the influence of RH on aerosol formation (Fig. 12). Under relatively dry conditions (RH < 50 %), both fS and fN were almost constant. However, when RH > 50 %, fS started to increase linearly, similar to the results observed by G. J. Zheng et al. (2015) in Beijing. In comparison, fN showed a small increase at RH 60–70 % and then decreased a bit when RH > 90 %, suggesting different roles of aqueous-phase reactions in the formation of sulfate and nitrate. Recently, studies of aqueous-phase chemistry have paid increasing attention to organic components. Ge et al. (2012) observed the strong enhancement of SOA during a fog event in the Central Valley of California during winter. Based on high-resolution mass spectra from an AMS, Sun et al. (2016a) retrieved an aqueous-phase-processed SOA (aq-OOA) that tracked well with RH in Beijing during wintertime. However, the mass fraction of SOA in total OA in this study remained relatively stable and showed no dependency on RH (Fig. 12c). The RH-binned bulk composition of submicron aerosol also only exhibited an obvious increase of sulfate at high RH (Fig. S13). One explanation for this observation is that the variations of SOA contribution may be largely interfered with by high fractions of POA across different RH values. Another explanation is that a portion of OOA formed through aqueous-phase reactions may be incorporated into fog droplets, which are too large to be transmitted into the ACSM aerodynamic lens, as reported by Ge et al. (2012). This explanation is consistent with the results obtained by studying a fog event in London, in which no increase in OOA concentration was detected by AMS measurement, whereas the single-particle mass spectrometry observed aqueous-phase SOA production (Dall'Osto et al., 2009).