2.1 Doppler Lidar Observation

The StreamLine Doppler Lidar from Halo Photonics used in this paper was located at Physical Geography Experimental Station of The Chinese University of Hong Kong (22.415N°, 114.212E°; Figure 1). The deployed lidar is a Halo Photonics 1.5-μm pulsed Doppler instrument that had previously been used for exploring the PBL in the tropical and midlatitude environments (Pearson et al., 2010; Barlow et al., 2011; Harvey et al., 2013; Manninen et al., 2018). The lidar system was conceived to meet the need for remote sensing of motion and backscatter in the atmosphere, where naturally occurring aerosols and clouds act as a distributed target that backscatter the transmitted pulses. Derived variables include wind profiles, backscatter coefficient (β), turbulence parameters, and cloud base measurements. The lidar was set up to optimize the retrieval resolution in the boundary layer up to around 3-km altitude and operated in velocity azimuth display with its scanner configured to take wind profiles every 10 min. The spatial and temporal resolutions are 30 m and 1 s, respectively, when operating in vertical stare mode for the intervening periods.

Prior to its operation, a colocation comparison test has been conducted to validate our lidar. In the test, two other nominally identical Doppler lidars, which are, respectively, setting up to Guangzhou and Zhuhai in the PRD region, were operated simultaneously at a same location. All the three lidars are new instruments so they have been calibrated by the lidar manufacturer. The collected data by the three lidars were compared to show that the data were consistent within the 95% confidence interval indicating a satisfactory manufacturer calibration. The data collection period for this study was from March 2017 to January 2019, except for several intermittent months of unscheduled lidar maintenance for repairing by Halo Photonics, and routine maintenance checks for the lidar includes cleaning and leveling checks. In addition to the hourly automatic wiping of the lidar window, deep cleaning was carried out at a frequency of a week where the windows were given a deep clean with an optical cleaning solution. Routine leveling checks were carried out at a frequency of a week to correct for any settling of the system into the ground surface. The precision of the leveling was within ±0.1°.

The performance of the Doppler measurement mainly depends on the detector noise of instrument and the natural atmospheric variability within the resolution volume (Päschke et al., 2015), with an optimal performance under high signal-noise ratio conditions. As suggested by Manninen et al. (2018), this study used a signal-noise ratio of −20 dB as a threshold, below which data were rejected for obtaining reliable Doppler velocity samples at an approximate precision of <0.3 m/s. Vertical velocity variance was then calculated over a 1-min averaging period. According to previous studies (Barlow et al., 2011; Flamant et al., 1997), the gradient in backscatter was used to define the PBL top height, that is, the aerosol-based PBL height, at which the largest gradient between the free atmosphere and PBL occurs. The mixing layer height was defined as the height up to which a threshold of vertical velocity variance above 0.1 m²/s² was met (Barlow et al., 2011).

In addition, based on previous studies (ICAO, 2005; Manninen et al., 2018), wind shear was derived from the available horizontal wind during the velocity azimuth display scans of Doppler lidar as follows:

(1)

where Δz is the range gate (30m). u and v are the wind components.

2.2 Meteorological and Air Quality Observations

In addition, hourly PM₁₀ and PM_2.5 (particulate matter, defined as particles with an aerodynamic diameter less than or equal to 10 and 2.5 μm, respectively) data were downloaded from Environmental Protection Department of Hong Kong (http://www.aqhi.gov.hk/en.html). The observations at two stations [Sha Tin (ST) at 6-m height above sea level (asl) and Tai Po (TP) at 15-m height asl] near the lidar at Physical Geography Experimental Station were averaged to represent the surface air pollution around the lidar station. Data at Tap Mum (TP) at 15-m height asl (a background station) was used to aid the analysis of local and transport air pollutions (see Figure 1). There are many different levels of threshold that may be applied at different locations globally. For example, the Ministry of Ecology and Environment of China/Europe Environment Agency/Environmental Protection Department of Hong Kong categorized exceedance as PM₁₀ 24-hr average above 150/50/100 μg/m³. We chose the peak values in defining our pollution episodes as 100 μg/m³ is more reflective of the more intense pollution events that are of interest in this study, when our lidar observations were available.

According to previous studies (Borge et al., 2007; Guo et al., 2017; Osborne et al., 2019; Zheng et al., 2015), backward trajectories for identified particulate matter pollution episodes were computed using Numerical Atmospheric-dispersion Modelling Environment to identify the impacts of transboundary air pollution and local PM pollution in Hong Kong (Jones et al., 2007). Briefly, the Numerical Atmospheric-dispersion Modelling Environment model is a Lagrangian transport model that is driven by model forecast winds or reanalysis data with variable spatial resolutions (Walters et al., 2017). In this study, meteorological analysis data from the UK's operational global model provide the underlying meteorological driving data, that is, horizontal and vertical components of wind fields at a horizontal resolution of around 10–17 km (10 km post 11 July 2017, and 17 km prior to that) with 70 vertical levels.

Daily weather maps published by the Hong Kong Observation were used to identify the surface synoptic patterns which denote the weather conditions of the episodes. For the peak day of each of the nine identified episodes, the upper air sounding profiles at 08:00 and 20:00 local standard time at King's Park Meteorological Station of Hong Kong Observation were used to validate the lidar observations (Figure 1). This was done by comparing the wind speeds and wind directions from the soundings with those recorded by the lidar at the same time (see Figure S1 in the supporting information). The intercomparison results reveal a good agreement between the Doppler lidar data and the sounding wind profile (Figure S1). The agreement is comparable to the result of Päschke et al. (2015).

3.1 Identified Particulate Matter Pollution Episodes

During the study period (excluding our lidar maintenance period), nine elevated PM pollution (EP) episodes (EP1–EP9 in Figure 2) were identified, all of which reveal a daily peak of PM₁₀ concentration above 100 μg/m³. Our results show that the duration of PM₁₀ > 100 μg/m³ were shorter than 6 hr in all the episodes except in EP7 (~12 hr) and EP9 (~14 hr). The peaks of PM₁₀ concentrations at the background site (TM) during most episodes (except for EP2 and EP9) were also >100 μg/m³, indicating a minimal spatial variation of PM₁₀ during the episodes and thus implying that the episodes were probably related to transport of PM from outside to Hong Kong (Luo et al., 2018). For the weather conditions at the peak time in Table 1, moderate relative humidity (<80%) was observed in most episodes, except for EP5 and EP8, while near-surface wind speed (<3.0 m/s) were rather low in all cases. The near-surface winds in five episodes were less than 1.5 m/s and most included a northerly component. At the time of peak concentrations, surface PM₁₀ concentrations during the nine episodes were weakly correlated to wind speed and insignificantly correlated with relative humidity at the surface (Figures 3a and 3b).

In addition, for horizontal pollution features in the PM₁₀ peak days, a large-scale higher aerosol optical depth was also usually present around Hong Kong during all nine episodes, based on the combined observations of MODIS onboard Terra and Aqua (see Figure S3). The surface synoptic patterns (see Figure S2 and Table 1) in the PM₁₀ peak day were associated with the following classifications: (i) monsoon with a cold high-pressure system over mainland China and with wind convergence (related to EP1, EP5, and EP6, EP9); (ii) saddle-type pressure field (related to EP2), and during cold front passage (related to EP7); (iii) warm high pressure over western North Pacific extending over South China (related to EP3); and (iv) subsidence associated with peripheral circulation of tropical cyclones over northern part of the South China Sea or near Taiwan (related to EP4 and EP8). In general, the surface wind patterns show certain correlations with surface PM concentrations during the episodes, while surface air quality is also closely related to wind at higher altitudes based on the reanalysis data (Tong et al., 2018; Yim et al., 2019), so the vertical wind profiles in the PBL need further exploration.

3.2 Vertical Observations of Particulate Matter Pollution Episodes

Figures 4-6 show time series of lidar data during the nine episodes (over periods of 72 hr): 1-min averages of attenuated backscatter coefficient (log10 scale), 1-min vertical velocity variance, and wind speed every 10 min, indicating that various vertical structures of PBL during these episodes. Compared with time series of surface PM₁₀/PM_2.5, the backscatter coefficient in PBL usually increased with the increases of surface PM₁₀/PM_2.5 (Figures 2 and 4). Aerosol backscatter coefficients were higher at the PM₁₀ peak time than those at other times during these episodes, excluding the enhanced backscatter by clouds or trace rain based on the meteorological observations at King's Park. Moreover, the high values in near-surface level particulates extended up to approximately 1,000 m at peak time (Figure 4). The backscatter intensities at peak time for EP1, EP2, EP3, EP4, EP6 and EP9 were greater than those for EP5, EP7, EP8, and the vertical thickness of the aerosol backscatter layers varied among episodes, depending on emission sources or weather conditions, i.e. wind speed, turbulent mixing process, and cloud in aerosol-based PBL. For instance, at peak concentration time during the nine episodes, the PM₁₀ had a significantly negative correlation [correlation coefficient(r): r = −0.8652, p < 0.01] with aerosol-based PBL heights derived from vertical backscatter of lidar based on gradient method (Figure 3c). The maximum/minimum peak of PM₁₀ are associated with the lowest/highest PBL height of ~1.15/1.75 km for EP9/EP4, suggesting that the role of vertical mixing is important in controlling surface PM concentrations. Here we simply clarify the dependence of PM on aerosol-based PBL height in transport episodes. Note that the correlations also show seasonal variations in both surface PM and aerosol-based PBL height, while there are still some differences in both surface PM and aerosol-based PBL height for different cases in the same season.

It should be noted that the smaller the vertical velocity variance is, the more stable the atmosphere is. Stable atmospheric conditions are conducive to the accumulation of air pollutants in the PBL. From the time series of vertical velocity variance and wind speed, it can be clearly seen that different turbulent mixing intensities appeared during the nine episodes. At the time of peak concentrations, stronger turbulence with larger vertical velocity variances (>0.1 m²/s²) were evident in the whole PBLs in EP1, EP4, and EP7 (Figure 5), showing that higher mixing layer heights (Figure 3d) were probably related to strong transport under strong wind speed conditions (>7 m/s in the whole PBL; Figure 6). In EP2, EP5, EP6, and EP8, strong turbulence with larger vertical velocity variances (>0.1 m²/s²) only occurred at near surface (<500 m asl) with strong wind speeds at upper PBL before peak time (Figure 6). Very weak turbulence with smaller vertical velocity variances (<0.1 m²/s²) mainly appeared in the whole PBLs in EP3 and EP9 (Figure 3d), indicating that a stable condition with low mixing layer heights is conducive to local pollutant accumulations under very weak wind speed (<2 m/s in the whole PBL; Figure 6). The correlation between mixing layer heights and the surface PM₁₀ at peak time during the nine episodes also show a strong inversion correlation (r = −0.8316, p < 0.01). The maximum/minimum peak of PM₁₀ are associated with the lowest/highest mixing layer height of ~0.3/1.5 km for EP9/EP4. In general, before or at the peak of most episodes (EP1, EP2 EP4, EP5, EP6, EP7, and EP8), wind speed increased, and meanwhile, the wind direction changed to north (Figure S4) in the whole or upper PBLs, indicating that PM transboundary episodes in Hong Kong are usually accompanied with a relatively large wind speed from upstream high-emission regions of mainland China.

According to previous study (Osborne et al., 2019), we employed the Numerical Atmospheric-dispersion Modelling Environment backward trajectory analysis described in section 2 to explore the source of air mass that transported PM to Hong Kong. Figure 7 shows the horizontal and vertical components of 72-hr backward trajectory of our lidar in Physical Geography Experimental Station at The Chinese University of Hong Kong during the nine episodes. Based on the vertical aerosol backscatter, the starting trajectory heights were from 100 to 1,000 m with an interval of 100 m, and the initial times were set to the peak time of PM₁₀ of each episode. Generally, all the air masses of Hong Kong during these episodes originated from mainland China. From the vertical component, there were obvious descending trends during the past 72 hr of these episodes, and the results show that the transboundary air pollution altitudes in the PBL varied among episodes.

With respect to relative contributions among the different emission source regions, Hou et al. (2018) defined two classifications of transboundary air pollution in the PRD regions as follows: regional transboundary air pollution (between cities in the PRD region) and superregional transboundary air pollution (from outside of the PRD region). In this study, superregional (regional) transboundary air pollution was identified when air mass in Hong Kong in one episode came from outside (within) the PRD region within 72 hr based on the backward trajectories. Notably, our use of backward trajectories was to identify possible sources of the heavy air pollution episodes in Hong Kong. We would like to clarify that our episode naming convention was designed to highlight particular features of the episodes. For example, regional transbounary EP means regional sources had remarkably contributed to the episode, whereas local contribution is not necessary to be insignificant. The name was chosen to highlight the regional transboundary air pollution feature in the episode. It should also be noted that a trajectory from superregion to Hong Kong must pass through the PRD region. As mentioned above, when the 72-hr backward trajectory indicated a superregional contribution, the regional and local contributions could be also important. Generally, it can be clearly seen from Figure 7 that EP3, EP5, EP6, and EP9 were related to regional transboundary air pollution for Hong Kong, while the others (EP1, EP2, EP4, EP7, and EP8) were associated with superregional transboundary air pollution. Particularly, EP4 was a very long distance transboundary air pollution event related to a dust storm event in northern China, in which the strong wind brought the dust belt of high PM₁₀ concentrations with a large ratio of PM₁₀ and PM_2.5 from northern China to southern China during 03–07 April 2018 (Figures S5 and S6). As a result, when the dust storm and accompanying strong northerly winds in the whole PBL (Figures 6 and S3) reached Hong Kong on 6 and 7 April, the surface PM₁₀ increased abruptly and exhibited peaks, while PM_2.5 had no peak in our observations (Figure 2).

3.3 Relationships Between Surface Particulate Matters and Vertical Wind Profiles

To assess the observed transboundary air pollution path and height related to wind profiles in Hong Kong, Figure 8 shows the correlation coefficients between hourly wind speed in each layer and hourly surface PM₁₀ during 72/48/24 hr (peak day) for nine episodes, and 1-hr-lag hourly surface PM₁₀ during 24 hr (peak day), indicating the various relationships between surface particulate matter and vertical wind profiles.

In detail, the relationships were significantly positive at different PBL heights in most episodes (EP1, EP4, EP5, EP7, EP8, and EP9), suggesting that the higher the wind speed was, the heavier the pollution was. It indicates that the wind speeds in these PBL heights show positive contributions to the pollutant concentration on the ground, which were different from the negative statistical relationships between surface wind speed and surface PM at peak time for these nine episodes (Figure 3a). Our results show that the large wind speed caused pollutants from mainland China to be transported to Hong Kong before or at peak time, which can be clearly observed by higher MODIS-based aerosol optical depth or higher ground-based PM concentrations in the upstream mainland China regions (see Figures S3, S5, and S6). In addition, the significant-correlation heights in Figure 8 ranging from 200 to 2,000 m can be considered as the transport path and layer of regional or superregional pollutants during these episodes. In contrast, the relationships between hourly wind speed and hourly surface PM₁₀ were significantly negative at most heights of the PBL in EP2, EP3, and EP6, suggesting that the lower the wind speed was, the heavier the surface PM pollution was, because the wind speed in the whole PBL were relatively weak (Figure 6). This is similar to the negative statistical relationships between surface wind speed and surface PM at peak time in these nine episodes (Figure 3a).

Wind shear is a source of turbulent mixing. At peak time in the nine episodes, the averaged wind shears in PBL were positively related to mixing layer heights and negatively related to surface PM₁₀ (Figures 2e and 2f), respectively, implying that the larger the averaged wind shear was, the stronger turbulent mixing was and thus the lower surface PM level was. However, the variability in the correlation coefficients between hourly vertical wind shear and hourly surface PM₁₀ shows that turbulence mixing process at different heights in the PBL influenced surface PM₁₀ (Figure 9). Compared with vertical wind speed and vertical velocity variance (Figures 5 and 6), in EP2, EP3, EP6, and EP9, wind speeds in the whole PBL were relatively weak, and the wind shear and mixing process were small. The relationship between hourly vertical wind shear and hourly surface PM₁₀ mainly shows negative correlations at certain heights in the PBL. In EP1, EP4, EP5, EP7, and EP8, wind speeds in the whole PBL were relatively strong, and the wind shear and mixing process were strong. The relationship between hourly vertical wind shear and hourly surface PM₁₀ generally exhibited positive correlations at certain heights in the PBL, especially near surface. The high transboundary PM was due to the upper level strong wind transporting PM to Hong Kong, and the PM then impacted surface PM concentrations at the ground through the vertical turbulent exchange in the PBL (Figure 5) to abruptly cause a significant increase in surface PM, which can be demonstrated by that the thicker layer of high-backscatter usually appeared around peak time due to the mixing of local PM and transboundary PM (Figure 4).