With the rapid economic growth and urbanization, an increasing frequency of haze episodes along with the air pollution has become of great concern in China during the last decade (Cao et al., 2016; Huang et al., 2014; Kulmala, 2015; Y. H. Wang et al., 2014, 2015). For example, during December 2016 a series of intense haze episodes took place in eastern China, characterized by surface PM2.5 concentrations exceeding 500 µg m-3 in several measurement sites in Beijing and its surrounding sites (http://www.mep.gov.cn/gkml/hbb/qt/201701/t20170102_393745.htm, last access: 11 December 2019). Severe air pollution has serious effects on human health. A recent study reported that the particulate matter has significantly decreased the life span of residents by as many as 5.5 years in northern China (Chen et al., 2013). On a global scale, the air pollution was estimated to cause over 3 million premature deaths every year (Lelieveld et al., 2015).

Increased emissions from fossil fuel combustion due to vehicle traffic, industrial activities and power generation, along with exceptionally strong secondary aerosol formation, were thought to be responsible for these haze episodes (Cheng et al., 2016; Huang et al., 2014; Pan et al., 2016; Petäjä et al., 2016; Zhang et al., 2015; Zhao et al., 2013). Meanwhile, the formation of intense haze episodes was considered to be affected by meteorological conditions (J. Wang et al., 2014; Quan et al., 2013; Zheng et al., 2016). For example, the mixing layer height is a key parameter that constrains the dilution of surface air pollution, and the development of the mixing layer is highly related to the amount of solar radiation absorbed by the air and reaching the surface (Ding et al., 2016; Stull, 1988; Sun et al., 2013; Tang et al., 2016). By using field measurements combined with model simulation, a positive feedback between aerosol pollution, relative humidity and boundary layer was found to be important in aerosol production, accumulation and severe haze formation in Beijing (Liu et al., 2018). Wang et al. (2018) found that PBL schemes in their atmospheric chemistry models are not sufficient to describe the explosive growth of PM2.5 concentrations in the Beijing–Tianjin–Hebei region due to the absence of an online calculation of aerosol–radiation feedback and/or a deficient description of extremely weak turbulent diffusion.

In this study, using unique measurements on the Beijing 325 m high meteorology tower, we show a clear relationship between mixing layer height and turbulent kinetic energy at the 140 m observation platform. We also present direct evidence of the feedback that relates the decreasing mixed layer height to increasing particulate matter concentrations, and this feedback is critical to the formation of intense haze episodes in Beijing.

A typical intense haze episode occurred during the heating season in urban Beijing during 17 to 22 November 2010. This episode was associated with synoptic stagnation in the North China Plain (Fig. S1 in the Supplement) and was characterized by low wind speeds and irregular wind direction (Fig. 1). Several meteorological variables had distinct temporal patterns during different stages of pollution, including reduced solar radiation and increased relative humidity during the most intense presence of haze (Fig. 1). The temporal patterns of PM2.5 concentrations were very similar at the two lower measurement heights (8 and 120 m, Fig. 1d), even though the concentration was clearly highest close to the surface. The PM2.5 concentration measured at 280 m behaved in a different way, especially during the most intense period of the haze when the mixed layer height was very low (Fig. 1e). The decoupling of the 280 m platform from the other two lower ones at low mixed layer heights is apparent in our 3-year measurement data set, especially when comparing O3 and NOx concentrations between the three measurement platforms (Figs. S2 and S5). During the haze period, the maximum PM2.5 concentrations at 8, 120 and 280 m were 505, 267 and 339 µg m-3, respectively. The higher maximum concentration at 280 m compared with 120 m can be ascribed to the transport of pollutants from the surrounding regions of Hebei and Tianjin provinces typical for polluted periods (Sun et al., 2013). The mixing layer height varied from 130 to 1640 m during the haze episode, ranging between about 200 and 500 m during the most intense period of the haze period on 18 November 2010 (Fig. 1e). The TKE was quite low during this intensive haze episode from 18 to 21 November, with an average value around 0.3 m2 s-2. However, the TKE increased significantly on the morning of 21 November as surface wind increased from 1.2 to around 6 m s-1, which was possible due to the movement of a cold front as shown in Fig. S1.

The vertical distributions of attenuated backscatter density obtained from ceilometer measurements indicate vertical mixing conditions accompanied by an inversion layer and high relative humidity in the surface as shown in Fig. 2. The strong inversion and high relative humidity occurred on the morning of 18 November 2010, with a lapse rate of 2 K / 100 m, relative humidity of 78 % and a northerly wind speed of around 2 m s-1 detected by the vertical sounding. The turbulent kinetic energy at 140 m was reduced to around 0.1–0.7 m2 s-2 due to decreased solar radiation, as presented in Fig. 1a. In this manner, the development of a mixing layer was significantly suppressed during the intense haze episode.

In order to demonstrate how the PM2.5 modifies the surface temperature, relative humidity and development of the mixing layer height, we performed two numerical simulation experiments, using the WRF model as a tool. We took the measurements during the intensive haze episode shown in Fig. 1 as an example. As shown in Fig. 3a, the variation of temperature and relative humidity showed pronounced daily variations, with higher and lower values, respectively, during daytime in both test and control experiments. However, the presence of aerosol in the test experiment clearly showed decreased surface temperature and increased relative humidity. The presence of aerosol reduces downward radiation reaching the surface, as a result of which the surface temperature and sensitive heat flux decrease and the development of the mixing layer height is suppressed (M. Li et al., 2017; Z. Li et al., 2017; Miao et al., 2016). Statistical results showed that the average relative humidity, surface temperature and mixing layer height were 8.2±3.4 ∘C, 40.5±11.6 % and 377.7±499 m, respectively, without the consideration of aerosol direct radiative forcing, whereas the consideration of aerosol directive radiative forcing changed these values to 7.1±3.1 ∘C, 40.6±11.7 % and 326.7±470.1 m, respectively. Our model results clearly demonstrate the pronounced role of aerosol particles in reducing the mixing layer height during this haze pollution episode.

In order to further illustrate how the mixing layer height modifies PM2.5 concentrations, we used 3 years of simultaneous winter-time air pollutant measurements in Beijing. We divided the observed PM2.5 concentrations into highly polluted and less polluted conditions using a threshold value of 75 µg m-3 for PM2.5 to distinguish between these conditions. This is consistent with the Chinese Environment Protection Bureau definition of a haze pollution event. With this threshold value, we found that 31 % and 69 % of the total measurement time corresponded to highly polluted and less polluted conditions, respectively. We plotted the PM2.5 data as a function of the mixing layer height at the three observation heights (8, 120 and 280 m) during both highly polluted and less polluted conditions and fitted an exponential curve to these data based on best fitting (Fig. 4). The PM2.5 concentration has a clear anti-correlation with the mixing layer height during the intense haze episodes. At all the measurement heights, the PM2.5 concentration increased as the mixing layer height decreased, and this pattern was very strong under polluted conditions (Fig. 4). We also tested the reciprocal fitting function for the data (Fig. S8). It overestimated the PM2.5 concentration when the mixing layer height was very low, as compared to the exponential fitting function (Fig. 4). This also indicates that a much higher PM2.5 concentration is needed in order to obtain a very low mixing layer height without the positive feedback. This can also be supported by the root-mean-square error (RMSE) of these two fitting methods. The RMSE of the exponential fitting is much smaller than the reciprocal fitting in any case (Table S1 in the Supplement).

It is worth noting that the increase was mainly from the PM1-2.5 fraction that increased from 42 % to 65 % as the mixing layer height decreased from more than 1400 m to lower than 300 m (Fig. S4). A major portion of particulate mass between 1 and 2.5 µm originates from secondary aerosol formation processes in urban air (Y. H. Wang et al., 2015; Zhang et al., 2015). As shown in Fig. S7, the concentration of NR-PM1 increased significantly from 12.1 to 56.4 µg m-3 when the variation of MLH decreased from more than 1400 m to less than 200 m. The reduction in solar radiation reaching the surface due to fine particle matter reduces the turbulent kinetic energy and the development of the mixing layer, as shown in Fig. 5. An exponential function between the turbulent kinetic energy at 140 m and mixing layer height was fitted. Based on this fit, the MLH roughly doubles from about 400 to 800 m when TKE increases from 0.1 to 1 m2 s-2. These are typical values of MLH during polluted conditions in Beijing.

The reduced sensible heat and TKE due to aerosol particles reduce the entrainment of relatively dry air into the mixing layer from above, which makes the air more humid within the mixing layer. This, together with the decreased surface temperature, increases the relative humidity (Z. Li et al., 2017). The increased relative humidity enhances the aerosol water uptake and promotes the formation of secondary organic and inorganic aerosol via aqueous-phase reactions (Liu et al., 2018; Wang et al., 2019), enhancing light scattering and causing further reduction in the intensity of radiation reaching the surface. All these factors suppress the development of mixing layer height and enhance the accumulation of air pollutants within the mixing layer. We ascribe part of the observed increase in PM2.5 and simultaneous decrease in the mixing layer height to the positive feedback associated with the particulate matter–mixing layer interaction (Petäjä et al., 2016; Ding et al., 2016), occurring at the same time as primary emissions and secondary formation are confined into a smaller volume of air. The feedback occurred at all three observation platforms and appeared to be most intensive at 8 m. In an urban environment, NOx originates mainly from local anthropogenic emissions, whereas the sources of particulate matter include both primary emissions and secondary formation (Ehn et al., 2014; Jimenez et al., 2009; Zhang et al., 2015; Zhao et al., 2013). As shown in Fig. S6, the median NOx concentration at 8 m was 250 % higher under highly polluted conditions compared with less polluted conditions as the mixing layer height decreased to 100–200 m, while the corresponding number for the PM2.5 concentration was 360 %.

The increase in the PM2.5 concentration from less polluted to highly polluted conditions is mainly due to concentrated particulate matter caused by a decreased mixing layer height, which is accompanied by primary particle emissions, secondary aerosol formation and feedback from particulate matter–mixing layer height interactions. Compared with the increased amounts of NOx, we can roughly estimate that a maximum of 110 % of the increased PM2.5 originates from secondary aerosol formation processes in this study. Of the remaining 250 % of the PM2.5 increase, potentially a large fraction originates from particulate matter–mixing layer height interactions, but we cannot quantify this fraction at the moment.

The development of a mixing layer height in an urban city is affected by the intensity of incoming solar radiation. Our measurement at the 325 m meteorology tower showed that the solar and ultraviolet radiation reaching the surface decreases considerably at increased pollution levels, which leads to a decreased TKE and, consequently, the suppression of mixing layer development. In turn, the shallowed mixing layer height further favours the enhancement of PM2.5 concentration and its precursor gases from both direct emissions and secondary formation. This feedback mechanism may be an important reason for the rapid increase in particulate matter from moderately polluted conditions to periods of intense pollution in an urban atmosphere as the strength increased when the PM2.5 concentration increased, although we cannot quantify the feedback amount exactly by observations currently. The particulate matter–mixing layer height feedback is probably a critical factor in the formation of intense haze periods from moderately polluted periods in Beijing and other polluted cities.