From 14 to 22 December 2016, complete haze pollution was observed in the Beijing area (see Table ). The generation, accumulation and elimination of PM2.5 were recorded. We can see that from 20 to 21 December, the hourly average PM2.5 concentration was almost maintained at approximately 400 µgm-3 over 48 h, which greatly exceeded the air pollution limits (i.e., 250 µgm-3) set by China's State Environmental Protection Administration. Figure  shows the concentration time series for PM2.5, wind speed and direction, temperature, relative humidity (RH), surface pressure and visibility for this period of heavy haze pollution.

Generally, visibility serves as a representative index of air quality and atmospheric diffusion capacity . Figure  shows that when the concentration of PM2.5 increased to high levels, visibility quickly deteriorated. Visibility on clean days was largely measured as greater than 10 km, and when the PM2.5 concentrations reached approximately 200–300 µgm-3, the visibility decreased to 2–5 km. Even when PM2.5 reached approximately 400 µgm-3, visibility dropped sharply to 1 km or to hundreds of meters. The surface pressure results suggest that air pressure levels decreased from approximately 1035 to 1023 hPa, and, in general, Beijing was controlled by a weak high-pressure system during the pollution episode. The RH taken from ground observations shows significant diurnal variations and an obvious anticorrelation between RH and temperature. From 20 to 21 December, the diurnal variation in temperature and relative humidity in heavy pollution was greatly suppressed, and a further analysis of MODIS images (see Fig. ) during this period shows that the pollution process was indeed accompanied by fog, while pollution formed in the south-central area of Hebei Province on 15 December 2016 and then spread across the whole Beijing–Tianjin–Hebei area on 18 December. Stratiform clouds appeared in areas surrounding Beijing on 21 December, but due to the high concentrations of pollutants (PM2.5 values approaching 400 µgm-3), mixed fog and haze appeared in Beijing. During the day, pollutants can scatter more solar radiation, while the ground receives less solar radiation, leading to the suppression of diurnal variations in temperature and relative humidity on the ground . An increase in RH occurs due to a decrease in temperature but is also the result of a surge in water vapor. For example, in the early morning, temperature differences observed between 17 and 20 December were minor, and the RH on 17 December was approximately 80 %, while the RH in the early morning of 20 December reached nearly 100 %, indicating an increase in water vapor levels in the Beijing area at this time. The surface wind speed during the pollution episode fell to almost less than 2 m s-1 and can be basically regarded as a stagnant weather system dominating ABL processes and resulting in poor air quality. From Fig. , we can see that there were cold fronts (strong northwesterly winds) on both 15 and 22 December, which advected pollutants away, resulting in good air quality. Between the fronts, PM2.5 levels slowly increased, as the air was stagnant (weak and variable winds in between) and the pollutants that were emitted locally slowly built up. The wind direction also seemed to be cyclical on each day in response to local mountain valley circulation around Beijing . According to other studies, stronger northerly winds occurring in the winter are the main mechanism through which pollution is removed, leading to good air quality .

The most basic definition of the ABL height is the height at which the influence of the Earth's surface on the lower troposphere disappears. This influence applies not only to conventional meteorological elements but also to turbulence quantities and even more for substances in the atmosphere such as aerosols, water vapor and nonreactive tracer gases . Levels of various pollutants and water vapor in the ABL are much higher than those found in the free atmosphere, and therefore there is often an obvious aerosol concentration gradient between the boundary layer and the free atmosphere. The extinction coefficient reflects the degree of aerosol particle scattering from lasers in the atmosphere . Thus, the ABL height can also be estimated from the extinction coefficient gradient. We used three popular methods – the gradient method (lidar_gra) , the standard deviation method (lidar_SD) and the wavelet method (lidar_wav) – to extract boundary layer heights from extinction coefficients (continuous Haar wavelet transformation was used in this paper, taking dilation parameter a=6). The ABL height determined by lidar is represented by Hc. In this study, the lidar_gra method applies the height of the atmosphere at which the gradient of the lidar extinction coefficient reaches its most negative value. The standard deviation of the extinction coefficient reflects the degree of lidar echo signal dispersion at different heights. The top of the planetary boundary layer constitutes the intersection between air in the boundary layer and the free atmosphere, which leads to a strong signal change at the top of the boundary layer. We define the height of the maximum standard deviation of signals as the ABL height. The lidar_wav method can also be used to detect abrupt changes in signals, so we use the Haar wavelet and take the height at which the wavelet coefficient is at its highest value as the height of the ABL. These methods are used to find the abrupt change in the extinction coefficient occurring at the top of boundary layer, though they present their own limitations.

Generally, the atmospheric boundary layer can be divided into a daytime convective mixing layer and a nighttime stable boundary layer. In the morning, the well-mixed convective boundary layer (CBL) is growing and often reaches its maximum height in the early afternoon. In the afternoon, the CBL gradually transforms into a neutral boundary layer. Figure  illustrates the evolution of ABL heights measured with lidar and radiosondes.

The determination of the ABL height by means of the lidar method is based on the vertical profiles of the extinction coefficient or the aerosol concentration. When concentrations of PM2.5 are high, the weakening effects of aerosol particles on lasers are stronger. ABL heights determined by lidar_gra and lidar_wav were almost the same, with a correlation coefficient of nearly 95 %. From 16 to 18 December and from 20 to 21 December, the ABL heights were approximately 500–750 m. Furthermore, the ABL height determined by the lidar_SD method was slightly higher than that derived from both methods. During a period of heavy pollution (20 to 21 December), the extinction coefficient quickly exceeded 3 km-1 at 250 m aboveground. Perhaps due to the accumulation of pollutants, Hc did not seem to decline on these days. When the atmosphere was relatively free of pollutants, such as on 15 or 22 December, the aerosol concentration was low and the extinction coefficient derived from the lidar system displayed no obvious signs of decline from the ground to the upper height. The ABL heights obtained by these methods based on the lidar system are clearly lower than those obtained by the other instruments. The ABL heights derived from extinction coefficient observed by lidar were unreliable when there was little pollution in the air. Therefore, the continuous observation of the ABL height can be achieved by means of other instruments or improved methods based on the lidar system.

Radiosonde instruments are the most widely used tools for conventional meteorological observation. The white points shown in Fig.  denote the ABL height determined from radiosonde data. The potential temperature (θ) determined by radiosonde technology is calculated from the following formula: θ=T+γdz, γd=0.00975 K m-1 ; T is the measured temperature. As noted in many previous studies, the most widely used approach for the determination of the ABL height and structure for the daytime and nighttime involves identifying local maxima in potential temperature vertical gradient profiles as measured by radiosonde devices , but this method is only appropriate to apply to the convective ABL . Since the pollution period we examine involves stagnant winter weather conditions and as our radiosonde data only apply to dawn and dusk periods (08:00 and 20:00 LT), a stable boundary layer often appears (see Fig. , for example) and the height of the stable boundary layer (SBL) is more difficult to determine . In this study, the level showing an obvious change in the potential temperature gradient and the profile of relative humidity were used to define the ABL height, as expressed by Hθ (see the blue dotted lines in Fig. ). Under stagnant and heavily polluted weather conditions, turbulence is more heavily suppressed than under normal weather conditions, and the top of the residual layer can also characterize the thickness of the stable boundary layer to some extent. We can also use the minimum value of the relative humidity (green curves shown in Fig. ) gradient to determine the height of the SBL. The atmospheric stratification of potential temperature and RH can affect the distribution of aerosol concentrations, which in turn affects the extinction coefficient. In Fig. , vertical profiles of the extinction coefficient observed by lidar during the same period are also given.

As shown in Fig. , the pollution episode was often accompanied by an inversion layer, as the vertical gradient of PT is positive, implying that the atmosphere was basically stable. The fact that the planetary boundary layer (PBL) was stable is unsurprising given the timing of radiosonde profiles at 08:00 and 20:00 local time (LT), which, respectively, occur approximately 30 min after sunrise and 3 h after sunset during the experimental period. Thus, a nocturnal inversion should barely be eroded by 08:00 LT (if at all, depending on the energy balance as insolation levels are low), and a nocturnal inversion should form by 20:00 LT. Due to this timing, these profiles are not representative of daytime conditions when pollutants are actively mixed. Midday profiles (noon local time or the early afternoon) would instead present instability and mixing. Air pollutants are generally blocked below the inversion layer and are not easily diffused to high levels. Figure a shows that Hθ at 20:00 LT on 16 December was approximately 690 m, where the potential temperature was approximately 280 K and the RH was approximately 20 %, and the extinction coefficient was also reduced to 0.7 km-1. Due to the cooling effects of surface longwave radiation, the ground inversion layer formed from the surface at a depth of approximately 100 m. At this time, the potential temperature gradient underwent an obvious change at 600 m. The inversion intensity levels below 600 m were weaker, and the height Hθ was approximately 690 m. The most negative value of the extinction coefficient gradient appeared at approximately 500 m at this time, and the extinction coefficient below 690 m was much higher than that observed above 690 m, indicating that aerosol particles were mainly concentrated below the inversion layer and that the Hθ calculated by radiosondes is basically consistent with Hc determined by lidar.

At 20:00 LT on 17 December, ground inversion started to form. Hθ was observed at approximately 680 m, and the potential temperature at this level was still at approximately 280 K, though the RH reached nearly 60 %. Below this height, the whole atmosphere layer had developed a high-humidity layer with an RH of nearly 80 % from the ground, and the corresponding extinction coefficient had also increased significantly. The extinction coefficient between 250 and 600 m was almost 3 km-1, revealing that the concentration of aerosols had increased significantly. Combined with the wind direction at this time (Fig. d), it is clearly observed that the transport of easterly winds moved considerable levels of water vapor from Bohai Bay (approximately 200 km east of Beijing), which promoted the hygroscopic growth of aerosol particles . At 08:00 LT on 21 December, the potential temperature distribution in the morning was different from that observed at 20:00 LT, and the surface temperature had begun to increase as solar radiation was received and the Hθ was approximately 410 m. The value below 300 m was nearly 95 %, while the maximum extinction coefficient, which exhibited bimodal features, reached nearly 4 km-1. The altitude at which the extinction coefficient reached peak levels in the lower layer was approximately Hθ. By means of analyzing and comparing Hc and Hθ values, it is apparent that when concentrations of PM2.5 were high, the accumulation of pollutants was mainly accompanied by the inversion layer in the atmosphere. The potential temperature gradient at the inversion layer is generally larger. Even though Hc reflects aerosol scattering information and Hθ denotes potential temperature characteristics, there is a strong correlation between them with a correlation coefficient of approximately 72 %. As shown in Fig. , Hθ was significantly higher than Hc determined by the three methods based on the lidar extinction coefficient.

The ground is the most important sink of atmospheric momentum, and the wind speed is zero at the Earth's surface. The ABL wind speed gradually changes from the Earth's surface to the geostrophic winds measured at high altitudes, and the wind information extracted from the WPR has been widely used to analyze wind characteristics. A comprehensive review of convective boundary layer heights is given by Seibert et al. (2000).

To analyze the influence of the boundary layer's wind structure on pollutants, we further discuss the representative wind speed and direction profiles of the three typical days for the unpolluted and polluted conditions. As is shown in Fig. , wind speeds below 1000 m did not exceed 6 m s-1 on 17 December. At 12:00 LT a typical “nose” profile distribution formed according to the wind profile, with a maximum value observed in the middle of the wind profile. The wind direction profiles show that the wind direction from the ground to approximately 750 m was northeasterly (0–90∘) and the wind direction observed above 750 m was northwesterly (270–360∘). Furthermore, the wind directions from 750 to 2000 m remained basically stable in the northwesterly direction, echoing geostrophic winds. The height of low-level maximum wind corresponded to the height where wind direction changed into geostrophic wind. In addition to 12:00 LT on 17 December, at four other times almost strictly northeastern winds formed below 1000 m, and winds began to transform into the northwesterly winds at different heights. Wind speeds increased to some extent on 21 December with typical “nose” type wind speed distributions observed at 00:00, 08:00 and 20:00 LT. From the ground to 2500 m steady southwesterly winds formed at 00:00 LT, and the maximum wind speed was approximately 900 m. At this time, the extinction coefficient was also very low above 750 m (shown in Fig. ), demonstrating that pollutants also formed below low-level maximum wind. At 12:00 LT the wind speed began to decrease slightly from the ground, and no obvious changes were observed beyond approximately 1100 m where wind speeds reached approximately 4 m s-1. Except at 00:00 LT on 21 December when southwesterly winds prevailed from the ground to high altitudes, the wind directions shifted to the northwest, though these wind speeds were less strong below 500 m, and wind speed maximum values formed at a height of approximately 1000 m.

Wind directions observed on 22 December were northwesterly from the low layer to the high layer, but the distribution of wind velocity profiles differed from that of 21 December. Wind speeds were based on no significant maximum value area, and the maximum wind speed of 500 m approached close to 12 m s-1. According to the extinction coefficient distribution (shown in Fig. ), the PM2.5 concentration was greatly reduced on this day. Hθ obtained by radiosondes were relatively similar at this time and far higher than Hc.

Based on high-resolution gradient observations (15-layer mean and 7-layer turbulence measurements), we can analyze the relationship between PM2.5 and low-level turbulence, average wind speed and temperature. As shown in Fig. , both turbulent kinetic energy (TKE) and friction velocity (u*) at 140 and 280 m were inversely correlated with ground level PM2.5 concentrations. TKE and u* can be calculated as follows : TKE=12u′2‾+v′2‾+w′2‾; u*=u′w′‾2+v′w′‾21/4. The maximum TKE levels observed on unpolluted days (15 to 16 December) reached approximately 7 m2 s-2 at 140 m while the TKE levels measured on hazy days (17 to 21 December) decreased sharply to very low values. After the start of the period of heavy haze pollution, TKE levels remained relatively low, and the change in TKE was not as significant when concentrations of PM2.5 increased from 200 to 400 µgm-3. On the other hand, the time series for u* is slightly different from that of TKE. It seems that the inverse correlation between u* and PM2.5 is more obvious than that of TKE for the heavy pollution period. In fact, even during the period of heavy haze, a slight fluctuation (diurnal variation) in PM2.5 concentrations can be observed, and the diurnal variation in u* follows the opposite pattern to that of the PM2.5 phase.

To understand the vertical structure characteristics of the ABL observed from the tower during the polluted and unpolluted periods, profiles for wind, potential temperature (θ), TKE and sensible heat flux (w′θv′‾) for the lower boundary layer are also given. At night (see Fig. ), the wind speed profile for the clear day basically follows a logarithmic distribution and the potential temperature changes little from the ground to approximately 300 m. For turbulent values, TKE gradually decreased from approximately 70 m, and sensible heat flux was basically negative. On polluted days, wind speeds from the lower layer to the upper layer were valued at less than 2 m s-1. At this time, the change in potential temperature was not very large from the ground to approximately 200 m, indicating that the atmosphere basically maintained neutral levels of stratification. A pronounced inversion layer cap is also observed to approximately 200 m from the surface. TKE levels were basically maintained at close to zero. Note that at this time the sensible heat flux measured at above 80 m remained at close to zero. Close to the ground, sensible heat flux was slightly positive, demonstrating that when pollution occurred and especially when the inversion layer existed, heat flux transport was suppressed. At night, surface longwave radiative cooling was restrained to a certain extent, and the weakening of turbulence activities aggravated pollution levels again.

In the daytime on 21 December 2016 (see Fig. ), the PM2.5 concentrations reached roughly 400 µgm-3, wind speeds were low, TKE remained at zero value and the potential temperature observed from the tower during the period of pollution basically denoted neutral stratification. The sensible heat flux was positive, but the value was basically measured as 0.02 K m s-1. At noon on 22 December, when the weather had improved, wind speeds were clearly higher. TKE still reached a maximum at 47 m. The influence of the urban canopy was stronger at heights of below 47 m. Unlike on the polluted day, levels of sensible heat flux were higher at this time and the lower layer reached a value of 0.1 K m s-1. The tower observation data clearly show that levels of sensible heat flux decreased significantly in the daytime during the haze episode because of the higher levels of solar radiation scattering by particles.

According to the concentration of PM2.5 observed, the haze pollution episode can be divided into different grades. Statistical mean values of surface visibility (Vis), wind speed (U), RH, ABL heights and turbulent fluctuations are also calculated. As is shown in Table , the statistical averages further confirm the conclusions of the previous analysis. Instead, when the concentration of PM2.5 is high, visibility and wind speed decrease while RH increases significantly. Our results show that due to the accumulation of aerosol particles, Hc is heightened slightly. The lidar results overestimate the ABL height at night . In this study the ABL height also increased due to the accumulation of pollutants during the period of heavy pollution. The turbulence levels observed also exhibit a decreasing trend occurring during the haze pollution episode, further demonstrating that turbulent activities are inhibited to a certain extent. We note that there appears to be only slight differences between the ABL height under slightly, moderately, heavily and seriously polluted conditions. Since the data examined in this study apply to only one period of heavy pollution, the minor differences observed may not be statistically different across categories except when compared to “good” air quality conditions.