As illustrated in Fig. 2, there were four haze episodes during the field measurement in Beijing, which were 7 to 11 October, 17 to 20 October, 22 to 26 October, and 29 October to 1 November.

The highest concentration of PM2.5 occurred at 18:00 on 25 October (the third episode), with a value of 469 µg m-3. Although the haze was severe, the peak value was much lower than that (900 µg m-3) during the haze in January 2013 (Zhang et al., 2014). Before the peak, increasing PM2.5 concentrations were observed. The four episodes had similar accumulation and dispersion patterns, in which the increasing slopes for the PM2.5 values in four episodes resembled each other, and the pollutants were dispersed within a few hours. The rates of increase of PM2.5, which were represented by the slope of PM2.5 concentrations, were calculated. To reduce the influence of the diurnal variation of PM2.5, the first and last peak values in each haze episode were chosen in the calculation. Thus, the slope r was rPM2.5,µgm-3h-1=[PM2.5]max-[PM2.5]minT(hour). The results were shown in Table 2.

In the four haze episodes, the rates of increase of the concentration of PM2.5 were 4.03, 4.44, 4.16 and 2.00 µg m-3 h-1. The first three hazes had similar rates of increase, and in the last haze episode, the concentration of PM2.5 increased at half the rate of the other three hazes. Gaseous pollutants (SO2, NO2, CO, and O3)also showed obvious increases during haze episodes, except for O3 in the fourth episode (Fig. 3). Because a heat supply was unavailable in October in Beijing, the emission of SO2, which was mainly from coal combustion, was much less than the emission of NO2, which was mainly from vehicle exhaust. The average mass concentration of NO2 was 73.1 µg m-3, which was much higher than that of SO2 (9.8 µg m-3). This result may not represent all the cities in the NCP, since large emission of SO2 could be found in heavy industry cities like Tangshan and Handan (Hu et al., 2014). O3 was a product of the photochemical reaction between nitrogen oxides (NOx) and volatile organic compounds (VOCs). High concentrations of NO2 resulted in a high concentration of O3. However, O3 remained at a low level during the fourth episode, especially on 30 and 31 October, whereas PM2.5 and other gaseous pollutants reached their peaks at this time, which indicated that the photochemical reaction during the fourth episode was not strong and that heterogeneous reactions played a major role in the haze formation in the fourth episode. All of the gaseous pollutants showed daily variation and reached a maximum on 18 October. These were also related to the variance in the height of the PBL, which will be discussed in Sect. 3.2.4.

Like the other severe hazes (Tao et al., 2014a, b) in last 2 years, the haze in October 2014 also influenced China to a large extent (Fig. 4). According to the PM2.5 spatial distribution over China, the haze originated from the North China Plain (NCP) and developed in the southwest and northeast directions. A similar situation occurred in September 2011, when the haze originated from Beijing and developed in the same directions (Liu et al., 2013b). Moreover, the haze influenced a large area of middle China and northeastern China; for example, the concentration of PM2.5 in Harbin, which is a large city in northeastern China, peaked at 664 µg m-3 during the third haze episode.

The temporal variation of the aerosol scattering coefficient at ambient environment bsp(RH), the absorption coefficient by NO2 bag, the aerosol absorption coefficient bap, the atmospheric extinction coefficient at ambient environment bext(RH) and the SSA were shown in Fig. 5. Data for bext(RH) were complete for the whole observation period, whereas the other three parameters were not available until 15 October. With clear diurnal variation, bsp, bap and bext increased day by day during the haze episodes and decreased sharply at the end of these haze episodes. Because bsp(RH) was calculated using bap and bext and bap was relatively small compared to bext, the bsp(RH) had nearly the same temporal variation as bext. The maximum bext was 5611 Mm-1 at 22:00 on 25 October, and the average bext reached 1069 Mm-1. Both data were much higher than those of other studies (Garland et al., 2008; Jung et al., 2009). The SSA in Fig. 5 was calculated with the aerosol extinction at ambient environment coefficient, bsp(RH). Aerosol representing fresh emission always has a low SSA, whereas aged aerosol has a higher SSA (Garland et al., 2008). As a result, the SSA showed an increasing trend during the haze episodes. Four sudden decreases in the SSA were observed during the dawn of 16, 22, and 26 October and 1 November. The first two decreases occurred at the beginning of haze episodes, which represented large quantities of freshly emitted aerosol. Since local emissions in Beijing showed a steadier pattern, the sudden change in SSA indicated pollutant transport from the vicinity of Beijing (discussed in Sect. 3.2.2). The other two decreases occurred at the end of haze episodes when strong winds blew over Beijing, and the pollutants were largely removed (discussed in Sect. 3.2.3). Aged aerosol was cleared and new aerosols increased in a short time (Guo et al., 2014). Thus, SSA decreased sharply.

Water-soluble ions are a primary component of aerosols and play a major role in the hygroscopic growth of aerosols. The temporal variations of major components (organic matter, SO42-, NO3-, NH4+, Cl- and black carbon (BC)) in PM1 were shown in Fig. 6. Organic matter contributed most to the PM1, followed by NO3-, SO42-, NH4+, BC and Cl-. The mass concentration of SO42- was always higher than that of NO3-, especially in the winter time, when the heat supply was prevalent in northern China (Zhao et al., 2013). Although Beijing was replacing coal with natural gas for a heat supply, the influence from the vicinity was still enormous. However, in this study, the mass concentration of NO3- was always higher than that of SO42-. The concentration of NO3-, which was primarily transferred from NOx, represented the contribution of motor vehicle sources. The heat supply was not available during October in the NCP, and the motor vehicles remained at a similar level throughout the year, which implied that motor vehicles played a more important part in the hazes during autumn. During the haze periods, the percentage of NO3- and NH4+ increased, whereas the percentage of organic matter continued to decrease. When strong wind blew pollutants away, the percentage of organic matter increased sharply. On the contrary, the percentage of NO3- decreased in a short time and did not return for a certain time. SO42- varied similarly to NO3- but in a much milder pattern. After the strong wind, even the concentration of SO42- decreased, and the return time was much shorter than that of NO3-. This indicated that even though the concentration of SO42- was lower than that of NO3-, the oxidation rate from SO2 to SO42- was faster. It has been reported that the existence of high levels of NOx might accelerate the reaction from SO2 to SO42- (He et al., 2014).

Even though all of the components in PM1 increased during the haze events, the accumulation pattern might be different for each component. Comparatively, the increasing pattern of SNA (sulfate, nitrate, and ammonium) was likely from local emissions. Diurnal variation of the concentration of SNA also existed, but it was not as significant as that of organic matter, Cl- and BC. There was also no sudden increase in the concentrations of SNA. They were more likely to accumulate stably with high RH and a stagnant atmosphere. When the RH was high and the atmosphere was stable, gases such as SO2 and NOx transform to SNA at a fast rate. A clear increase in the SNA percentage could be seen in the pie charts in Fig. 6. The values for SO42-, NO3- and NH4+ in PM1 increased from 8.48, 20.55 and 9.46 % in the non-haze period to 12.7, 27.5 and 12.14 % in the haze episode, respectively. Continuous increasing of SNA indicated that the formation of new SNA during the haze episodes contributed most of the formation of the haze. During the hazes in January 2013, high conversions from the gas phase of SO2 and NOx to the particle phase of SO42- and NO3- were found, and heterogeneous formations of SO42- and NO3- were considered to be important, especially during low visibility episodes (Quan et al., 2014).

SOR (sulfur oxidation ratios) and NOR (nitrogen oxidation ratios) were important factors, showing that gaseous species would be oxidized to secondary aerosols in the atmosphere (Sun et al., 2006). They were widely used in the analysis of the secondary transformation of aerosols. PM2.5 and SOR, NOR were highly related. As we found in the Fig. S1 in the Supplement, PM2.5 was well fitted with SOR and NOR. The correlation coefficient was 0.62 between PM2.5 and SOR and 0.79 between PM2.5 and NOR. It meant SOR and NOR could be higher with higher concentration of PM2.5. The temporal variations of SOR and NOR were shown in Fig. 7. SOR was mostly higher than 0.2, and NOR was mostly higher than 0.1, indicating intense secondary formation of SO42- and NO3- (Fu et al., 2008). SOR and NOR increased during the haze episodes with accumulated pollutants. Furthermore, SOR increased more quickly than NOR. To compare the rate of increase of SOR and NOR, the slope of the SOR and NOR in the observed haze were calculated. To reduce the influence of the diurnal variation of SOR and NOR, the first and last peak values in the figure were chosen in the calculation. Thus, the slope r was rSOR,h-1=SORmax-SORminT(hour). rNOR,h-1=NORmax-NORminT(hour). The results were shown in Table 3.

In the three observed haze episodes, r(SOR) was 3.4, 1.6, and 4.2 times of r(NOR), which indicated faster production of SO42-, even though the concentration of SO42- was lower than that of NO3-. Meanwhile, after the strong wind, which decreased SOR and NOR sharply, low SOR still existed, whereas NOR was nearly 0. These findings explained the shorter return time of SO42- after the haze episodes.

Biomass burning became prevalent in the NCP during the autumn harvest. The combustion was primarily conducted in the open field. Pollutants, such as BC and CO, were emitted on a large scale and influenced the air quality not only in the emission region but also in the downstream city. Hence, biomass burning in the surrounding provinces was an important cause of the hazes in the autumn in Beijing. Fire points in China, based on data from MODIS Terra and Aqua satellites, on 6 October were shown in Fig. 8a. On 6 October, 267 fire points were found in China, among which 29, 163 and 30 fire points were found in Hebei, Henan and Shandong provinces, respectively. In total, 1957 fire points, which were caused by biomass burning in the whole of October, were found in China, among which 54, 26 and 57 fire points were found in Hebei, Henan and Shandong provinces, respectively. Although the fire points were more strictly controlled in 2014 compared to 2013, the influence of biomass burning still could not be neglected.

The backward trajectories of Beijing during the first haze episode reflected how the biomass burning influenced Beijing (Fig. 8b). A total of nine backward trajectories of 48 h were drawn with the HYSPLIT model online version. The backward trajectories started from 17:00 (LST) on 11 October and restarted a new trajectory every 12 h. The air mass during the first haze episode mainly came from the south and southeast, originating from Hebei, Henan, Shandong and even Anhui Province. The pollutants from biomass burning in these provinces were transported to Beijing. Once the meteorological conditions were stagnant, haze formed and was aggravated in this region. A similar situation occurred in the haze in 2007 (Li et al., 2010). Therefore, biomass burning was a tough challenge for air pollution control in the autumn.

Based on the temporal variation of each component in Fig. 6, the organic matter, Cl- and BC had similar variation patterns. In addition to clear diurnal variations, which were caused by the diurnal development of the PBL, a sudden increase before each haze period was found for the concentration of organic matter, Cl- and BC. On 18 October, the height of the PBL (468.7 m) was 21.1 % higher than that on 17 October (386.9 m), but the concentrations of organic matter, Cl- and BC were 6, 6 and 4 times the values of the day before. Organic matter, Cl- and BC were emitted from biomass burning; abnormal high values and a sudden increase in organic matter, Cl- and BC indicated spatial transport of pollutants from straw burning. Consequently, regional pollutant transport was important for haze formation in October 2014, and straw burning was a significant pollution source. However, the concentrations of organic matter, BC and Cl- in haze period were found lower than those in nonhaze period. It was because the increase of organic matter, BC and Cl- was obvious in the beginning part of haze, after which the increase of new SNA were predominant.

Pressure systems can influence the wind and precipitation of a region. Surface weather maps of East Asia at 02:00 (LST) on 7–10 October during the first haze episode were shown in Fig. 9. The NCP was dominated by a weak high-pressure system on 7 October, which lasted for the following 2 days. The weak high-pressure system resulted in low surface wind and relatively stagnant weather, which was unfavorable for the dispersion of air pollutants. The high-pressure system slowly moved towards the northeast. Meanwhile, the Mongolia anticyclone (a low-pressure system) moved towards and encountered the high-pressure system on 9 October, which brought wind and caused a small decrease in PM2.5 on the dawn of 9 October. However, the weak high-pressure system dominated the NCP on 10 October, and the weather became stagnant again until 11 October, when another strong Mongolia anticyclone moved to the NCP, and the first haze episode ended.

The wind fields at 02:00 on 10, 20, 24, and 31 October in the NCP region were shown in Fig. 10, representing typical days in the four haze episodes. Generally, the wind speeds were slow during the haze episodes. In addition, the wind slowed sharply around Beijing city in all four figures. For example, on 31 October, the wind over the NCP came from the east, from the Bohai Sea. The wind separated into two directions when it encountered Beijing, one blowing to the north and the other blowing to the south. The wind over Beijing maintained a low speed. A similar phenomenon was observed in January 2013 when a severe haze occurred in Beijing (Tao et al., 2014b). Moreover, the wind on 10 October was smooth, blowing from the southeast of Beijing and then turning to the northeast. The wind around Beijing was clearly slowed and became strong after blowing over Beijing. The city acted as a large obstacle for the wind, slowing the wind speed, disturbing the wind direction and affecting other properties of the wind (Miao et al., 2009). The wind on 20 October was more complex. Winds from the southwest and the northwest blew toward and converged at Beijing. The winds from the two directions weakened each other, creating stagnant conditions in Beijing. The pollutants in the vicinity were brought to Beijing, accumulating and reacting further (Zhao et al., 2013). On 24 October, the wind was similar to that on 10 October, with the exception of the disordered directions around Beijing city.

Changes in wind pattern influenced the near-surface aerosol concentration (Pal et al, 2014). The wind rose diagram overlaid with the PM2.5 concentrations from 5 October to 2 November were shown in Fig. 11. The winds blew mainly from three directions: northwest, northeast, and southwest. In each direction, the higher the wind speeds were, the lower the PM2.5 concentrations were. The average wind speed was 1.1 m s-1, which was slightly higher than that (0.9 m s-1) in September 2011, when another haze episode occurred (Liu et al., 2013b). Furthermore, the PM2.5 concentrations were relatively high within the wind speed limit: 1 m s-1 for wind from the northwest, 1.5 m s-1 for wind from the northeast and 3 m s-1 for wind from the southwest. For example, when the wind was from the northeast and the wind speed was lower than 1.5 m s-1, nearly all of the PM2.5 concentrations exceeded 75 µg m-3 (the national secondary standard of PM2.5) and vice versa. Nevertheless, when the wind blew northwest, with wind directions concentrating between 270 to 315∘, the wind speed limit decreased to 1 m s-1. Thus, with similar wind speeds, the concentration of PM2.5 was lower. Zhang et al. (2015) made a research on relationships of evolution of PM2.5 concentrations and meteorological conditions. In Beijing city, winds from the northeast and southwest would result in higher concentration of PM2.5, and winds from the northeast would lead to the lowest recorded concentration of PM2.5 in autumn and winter. It could also be clearly found in the haze during January 2013 in Beijing (Yang et al., 2015). However, winds in January 2013 distributed more equally in every direction, while winds in October 2014 were concentrated in a specific direction. Differences between every wind direction were less in January 2013. High concentration (over 75 µg m-3) of PM2.5 could also be found at wind speeds over 2 m s-1 when wind blowing northwest.

The development of the PBL, which is mainly influenced by air temperature and dynamics, affects the vertical dispersion of pollutants (Pal et al., 2015). When the PBL height is low, the pollutants stay at the surface layer and maintain a higher level. This can easily lead to haze formation on the ground. As depicted in Fig. 12, the heights of the PBL between the two haze episodes were relatively high, and distinct descents of the PBL height were found during the four haze episodes. The lowest PBL at noon was 384 m on 17 October. It was only 14 % of the PBL height before the haze (2718 m on 15 October). Because it was difficult to break through the PBL, the local pollutants were compressed and accumulated near the surface, leading to a high concentration of PM2.5. This was not similar to cases reported in other haze episodes (Yang et al., 2015). The temperature during each haze episode in October 2014 maintained a constant level or even increased (Fig. 2). Normally, a higher temperature, which will enhance the PBL development, is conducive to higher PBL. However, the formation of hazes in October 2014 was influenced by many factors. The dynamics of the atmosphere is another important factor from the local perspective. The deficiency of horizontal movement of the atmosphere affected the vertical development of the PBL. The highest PBL occurred on 16 October, when a cold and fast wind blew over Beijing. The cold and fast air mass from Mongolia was an effective removal mechanism of the haze not only in the horizontal direction but also in the vertical direction. Once the cold air from the northwest blew over the NCP, it not only promoted the horizontal dispersion of pollutants but also accelerated the vertical diffusion of pollutants because the height of the PBL was rapidly increased.

Four haze episodes in October 2014 were characterized by higher RH. The RH was over 90 % during the severest haze period, and it sometimes reached 100 % in October 2014. The high RH aggravated the haze.

The relationships between RH and SOR and between RH and NOR were illustrated in Fig. 13. SOR and NOR were highly correlated with RH. The correlation coefficients of SOR and NOR with RH were 0.79 and 0.55, respectively, which were much higher than those in a previous study (Han et al., 2014), indicating the particularly high importance of RH during the haze episodes in this study. SOR reached a minimum when RH was approximately 40 %. When RH was > 40 %, SOR increased with increasing RH, whereas SOR decreased with increasing RH when RH was < 40 %. The conversion from SO2 to SO42- requires water vapor as a reactant. Hence, when RH was > 40 %, the high RH became a major factor accelerating the production of SO42-. The conversion is an endothermic reaction, which indicates that high temperature can promote the conversion. RH and temperature have an inverse correlation. When the RH was lower than 40 %, water was not adequate to accelerate the conversion. As a result, the temperature became the most important factor in the reaction. Therefore, when the RH was lower than 40 %, SOR decreased with decreasing temperature but not with increasing RH. The formation of NO3- was more complex than that of SO42-, so the correlation of NOR and RH was much weaker than that of SOR and RH. However, a clear positive correlation was still observed. No minimum value was found for NOR, and high RH can promote NOR.

The measured fRH values at ambient RH during the observed period were depicted in Fig. 14. The curve was similar to those of other studies, indicating an increasing tendency of fRH with increasing RH. Usually, the exponential relationship between fRH and RH can be fitted by an empirical function: fRH=1+a(RH/100)b. In this study, the curve-fitting parameters a and b were 3.79 and 6.10, respectively. fRH values at an RH of 80 %, which had an average value of 1.97 in this study, were extracted for comparison with other studies. When the RH was 80 %, the aerosol particulate scattering coefficient was nearly 2 times of that in dry conditions, which was relatively large compared with other studies (Liu et al., 2013a). Consequently, the RH contributed much more to aerosol extinctions in October 2014, leading to more severe hazes.

Visibility dependences on the mass concentration of PM2.5 at different RH intervals were shown in Fig. 15. Overall, the visibility decreased rapidly with increasing PM2.5 concentration. In addition, with increasing RH, the visibility decreased faster. For example, when the RH was less than 30 %, and the mass concentration of PM2.5 was lower than 85 µg m-3, all of the visibilities were over 10 km. However, when the RH was greater than 90 %, no visibility was more than 10 km in this observation. To quantify the relationship between visibility, the PM2.5 concentration and RH, Eq. (9) was used: bextRH=bspRH+bap+bsg+bag=QSP×PM2.5×(1+a×(RH100)b)+Qap×PM2.5+34, where Qsp and Qap were the mass scattering and absorbing efficiency, which were the ratios of bsp and bap to PM2.5, respectively. The average values of Qsp and Qap in this study were 3.10 and 0.42 m2 g-1. Curves of the dependence of visibility on the PM2.5 concentration at different RH intervals were shown in Fig. 15. The calculated curves showed the same trend as the dotted figure: the higher the RH was, the faster the visibility decreased as the concentration of PM2.5 increased. When the concentration of PM2.5 reached 75 µg m-3 (the national secondary standard of PM2.5), the visibility surpassed 10 km only when the RH was lower than 60 %. To control haze, keeping PM2.5 under the national second standard alone was sufficient.

In previous studies, a temperature decrease was often reported (Liu et al., 2013b). It not only deprived the dynamics of the PBL development, but also resulted in less heat turbulence, which was unfavorable for pollutant dispersion. Moreover, negative aerosol radiation forcing (ARF) was always recorded, indicating the feedback mechanism between radiation and aerosol loading (Quan et al., 2014). However, in this study, even though the solar radiation was reduced during the haze episodes, the temperature was steady or even increasing over a longer temporal range (several days). Water vapor, a greenhouse gas, had a vital effect on the atmospheric thermal balance. The short-wave radiation from the sun was not absorbed by water vapor, but the long-wave radiation from the Earth can be largely absorbed by it. As a result, even less solar radiation reached the Earth's surface, and the radiation from the Earth supplied increasing heat to the atmosphere with increasing RH. Increasing temperature accelerated the chemical reaction rate of aerosols and aggravated the haze. This situation could also be found in the haze in January 2013, but little research has focused on the temperature variation. Figure 16 depicted how RH influenced the haze formation. Water vapor in the atmosphere played a vital role in the formation of haze, which could not only accelerate the chemical transformation of secondary pollutants but also lead to hygroscopic growth of aerosols. Furthermore, as an important greenhouse gas, it absorbed surface radiation, altering the thermal balance of the atmosphere, which finally affected haze formation. RH in autumn was much higher than winter (Dong et al., 2013). It highly increased the rate of secondary reaction and hygroscopic growth. Thus, when only considering RH, with the same level of emission haze in autumn would be more severe. However, in reality a larger quantity of emissions and increasing RH in last decade in winter (Cheng et al., 2015) resulted in heavier haze in the winter season.