Predicting air pollution events in the low atmosphere over megacities requires a thorough understanding of the tropospheric dynamics and chemical processes, involving, notably, continuous and accurate determination of the boundary layer height (BLH). Through intensive observations experimented over Beijing (China) and an exhaustive evaluation of existing algorithms applied to the BLH determination, persistent critical limitations are noticed, in particular during polluted episodes. Basically, under weak thermal convection with high aerosol loading, none of the retrieval algorithms is able to fully capture the diurnal cycle of the BLH due to insufficient vertical mixing of pollutants in the boundary layer associated with the impact of gravity waves on the tropospheric structure. Consequently, a new approach based on gravity wave theory (the cubic root gradient method: CRGM) is developed to overcome such weakness and accurately reproduce the fluctuations of the BLH under various atmospheric pollution conditions. Comprehensive evaluation of CRGM highlights its high performance in determining BLH from lidar. In comparison with the existing retrieval algorithms, CRGM potentially reduces related computational uncertainties and errors from BLH determination (strong increase of correlation coefficient from 0.44 to 0.91 and significant decreases of the root mean square error from 643 to 142 m). Such a newly developed technique is undoubtedly expected to contribute to improving the accuracy of air quality modeling and forecasting systems.
The boundary layer height (BLH) illustrates the relationships between air pollution intensity, duration, and scope; it constitutes an important factor influencing the diffusion of pollutants in the low atmosphere (Tie et al., 2007; Quan et al., 2013). An increase of air pollutants is often associated with a shallow BLH, while a decrease of pollutants is accompanied by obvious uplift of the BLH. Besides the physical effects, BLH can also affect the precursor particles' concentration and distribution, which might affect the chemical transformation of fine particulate matter (Ansari and Pandis, 1998). BLH is also a key parameter for air pollution models; it determines the volume available for the dispersion of pollutants and is involved in many predictive and diagnostic methods and/or models that assess pollutant concentrations (Seibert et al., 2000). The bias of the BLH between the air quality model and observation is a potential cause of model's difficulties to accurately forecast air pollution episodes (Dabberdt et al., 2004). Therefore, accurately acquiring the BLH, especially during polluted episodes, is of great significance to investigating air pollution issues.
Many techniques have been developed to determine the BLH, for example, through radiosonde measurements (Stull, 1988), remote sensing (Emeis et al., 2007), laboratory experiments (Park et al., 2001), and model simulations (Dandou et al., 2009). The high spatiotemporal resolutions make aerosol lidar techniques (light detection and ranging) one of the most suitable systems for analyzing the boundary layer structure and determining the BLH (Flamant et al., 1997). Due to the complex vertical structure of boundary layer, numerous methods have been proposed to accurately retrieve the BLH from lidar, such as the maximum variance method (Hooper and Eloranta, 1986), fitting idealized profile method (Steyn et al., 1999), first point method (Boers and Melfi, 1987), threshold method (Dupont et al., 1994), wavelet transform method (Davis et al., 2000; Baars et al., 2008), first gradient method (Flamant et al., 1997), logarithm gradient method (Senff et al., 1996), and normalized gradient method (He et al., 2006). However, so far most of the algorithms have been tested and validated only over relatively unperturbed homogeneous terrain, for example, oceans (Melfi et al., 1985; Flamant et al., 1997), rural areas, and clean meteorological conditions (Piironen and Eloranta, 1995). Limited evaluations of the algorithms have been carried out in polluted megacities in developing countries, associated with a high density of buildings and heavy anthropogenic pollutants. Nevertheless, the surface roughness and high aerosol loading in the boundary layer result in a more complex structure and increase the difficulty of BLH retrieval based on these algorithms.
As one of the largest megacities in Asia affected by heavy pollution, Beijing provides a particular challenge to resolving the BLH determination. Specifically, the movement of the atmosphere can affect the distribution of pollutant concentrations; moreover, vertically propagating gravity waves influence the structure of the atmosphere and cause some of the spatiotemporal variability (Fritts and Alexander, 2003). Gravity waves thus provide new theoretical insights for the development of a new algorithm in determining BLH by taking into account a probably insufficient vertical mixing of pollutants under weak thermal convection and pollutant accumulation at high altitudes due to long-range transport processes. Beijing, often governed by stagnant meteorological conditions, is surrounded by mountains to the west, north, and northeast, and characterized by favorable conditions to generate and maintain gravity waves. Such specific atmospheric conditions provide the opportunity to obtain insights into the difficulties related to the BLH retrieval based on existing algorithms and to evaluate the performance of a new approach that considers the impact of gravity waves. Based on an intensive observation campaign over Beijing, this paper aims at delving into the limitations of current retrieval algorithms employed for BLH determination from lidar during a polluted period and at coming up with the development of a new algorithm compatible with all atmospheric pollution conditions. This work therefore provides, for the first time, a prototype of how to integrate into the BLH retrieval process gravity waves and the resulting complexity of the low-troposphere structure under conditions of heavy aerosol loading. Section 2 presents a detailed description of the lidar observational experiment setting over Beijing and discusses the limitations of current algorithms for BLH retrieval; Sect. 3 discusses the development of a new algorithm; Sect. 4 presents the comprehensive evaluation of the new retrieval algorithm and comparative analysis with existing methods; and conclusion and environmental implications are given in Sect. 5.
Daily variation of the PM
Beijing, the capital of China, is located at 39
An unprecedented 78-day intensive radiosonde campaign was conducted over the
Institute of Atmospheric Physics site (four times per day: 02:00, 08:00,
14:00, and 20:00 local standard time) in line with the lidar campaign at a
radiosonde observatory located in southern Beijing (39
In normal conditions of an aerosol-laden boundary layer and clean overlying
free atmosphere, the gradient of the range-squared-corrected signal (RSCS)
exhibits a strong negative peak at the transition between the boundary layer
and free atmosphere. Based on this principle, gradient algorithms were
proposed and had become the most widely used ones. In this paper, we focus
on the three most popular gradient methods, including the first gradient
method (GM), first logarithm gradient method (LGM), and first normalized gradient method (NGM). The optical power measured by lidar is proportional
to the signal backscattered of particles and molecules present in the
atmosphere. The lidar signal can be expressed by Eq. (1) below:
The RSCS is then defined in Eq. (2) by
As a key parameter for air pollution forecasting models, BLH can determine the volume available for the dispersion of pollutants (Seibert et al., 2000). Accurate retrieval of the BLH by automatic algorithms not only allows making insights into its diurnal fluctuations during pollution episodes, but it also contributes to validating modeling results and improving prediction performance.
Prior to the calculation of the gradient with the current three BLH retrieval algorithms, a moving average of 30 m in height was assumed in the stored lidar profiles in accordance with the study of Pal et al. (2010), who previously reported that a height difference of 30 m was the most appropriate for identifying the minimum of the gradient. Typical gradient profiles of the RSCS and retrieved BLH from various algorithms with corresponding radiosonde profiles of the potential temperature and relative humidity are illustrated in Fig. 2b and c. Strong negative peaks were detected in the profiles for each algorithm to define the BLH (Fig. 2b). As illustrated in Fig. 2b, at 20:00 on 27 July, the BLH retrieved by GM is 480 m versus about 1590m retrieved by LGM and NGM. Determining the BLH from radiosonde measurements based on the potential temperature sharply increasing with altitude and decreasing relative humidity is the classic and most accurate approach usually applied to evaluate lidar retrieval results (Seibert et al., 2000). At 20:00 on 27 July, the radiosonde identified a region at 1350 m, considered as the actual BLH (Fig. 2c). Thus, GM significantly underestimated the BLH by approximately 870 m, while LGM and NGM overestimated the BLH by about 240 m. The diurnal cycle of the BLH retrieved by these algorithms is illustrated in Fig. 2a in comparison with the four radiosonde measurements (02:00, 08:00, 14:00, 20:00). The results demonstrated that none of the algorithms was able to fully capture the diurnal cycle of the BLH. The average underestimation was 500–600 m for the GM algorithm (strongly supporting previous findings of He et al., 2006), as opposed to an overestimation of 400–500 m for the LGM and NGM algorithms on 27 July, in agreement with the profile analyses (Fig. 2b). In addition, the performance of the retrieval algorithms on 24 and 28 July (Figs. S2 and S3) strongly correlated with that found on 27 July. This highlights the critical bias and limitations of these algorithms in accurately determining the BLH under heavy aerosol loading.
The top of the boundary layer is often associated with strong gradients in the aerosol content, so a simple negative gradient peak seems suitable to determine the BLH. However, data interpretation from aerosol lidar is often not straightforward. Aerosol loading in the low troposphere mainly originates from the ground level. Thus, under stable conditions, large negative gradient peaks possibly exist near ground level (even larger than that of the BLH) due to insufficient vertical mixing of the pollutants in the boundary layer. Thus, the BLH might be wrongly determined by the GM based on these negative gradient peaks with critical underestimation. On the other hand, both LGM and NGM originally developed to filter out the influence of aerosols near the surface and to improve the original GM (Sicard et al., 2006; Emeis et al., 2007) result in an overestimation of the BLH. LGM is normally supposed to filter out the negative gradient peak near the ground to a certain extent, producing a higher BLH than GM (He et al., 2006). Such overestimation is probably induced by accumulation of aerosol at higher altitude due to adventive chemical transport (Stettler and Hoyningen-Huene, 1996), undetectable by the retrieval algorithms due to the impact of gravity waves on the atmosphere structure (Gardner, 1996), which inhibits the filtration skills of LGM and NGM. It is clear that the accuracy of current retrieval algorithms in determining the BLH from lidar is limited by conditions of heavy aerosol loading (with insufficient vertical mixing in the boundary layer) associated with the impact of vertically propagating gravity waves.
The canonical gravity wave vertical wave number spectrum of horizontal wind fluctuations. From J. Atmos. Terr. Phys., 58, 1577, 1996.
As evoked in previous sections, heavy pollution and propagation of gravity
waves critically limit the accuracy of current retrieval algorithms in
determining the BLH from lidar. Beijing is characterized by favorable
conditions to generate and maintain gravity waves in particular due to the
presence of Qinghai–Tibet Plateau in the west, which is considered as a
potential source of gravity waves in Beijing (Gong et al., 2013). In fact,
during a campaign of more than 2 years (from April 2010 to September 2011),
daily and seasonal vertical mixing of wavelengths and phase velocities of 162
quasi-monochromatic gravity waves were observed over Beijing from lidar (Gong
et al., 2013). Moreover, statistical analysis of the captioned campaign
revealed that gravity waves were maximal in summer (June–August),
corresponding practically to the discussed observation period of the present
study (1 July–16 September). It is clear that this finding serves as potential
observational evidence of gravity wave and strong support of the present
study. According to the research of the Global Atmospheric Sampling Program, the
gravity waves generated by the mountains are
The linear instability theory (LIT) of gravity waves (Dewan and Good, 1986)
is illustrated in Fig. 3.
The motion of aerosol in the boundary layer is determined by the background
atmosphere (the aerosol particles move with the background atmosphere).
Thus, the aerosols and the background share the same fractional fluctuation.
The cubic root gradient method (CRGM), a new algorithm for BLH determination,
is thus defined by
Statistical parameters for each lidar retrieval algorithm compared with radiosonde measurements.
Figure 4b (similar to Fig. 2b) shows the BLH retrieved by CRGM as a red dotted line. Strong negative peaks were detected in the profiles for each algorithm to define the BLH (Fig. 4b). At 20:00 on 27 July, the BLH retrieved by CRGM was 1350 m, in perfect agreement with the actual BLH determined by radiosonde (1350 m), as opposed to 480 and 1590 m determined by LGM and NGM, respectively. The diurnal cycles of the BLH retrieved by CRGM presented in Fig. 4a show CRGM's good capture of the unimodal diurnal cycle of the BLH, presenting a peak at 14:00–15:00 and a valley at 07:00–08:00, induced by the thermal activity of the ground. In comparison with CRGM, the BLH determined by LGM and NGM did not present unimodal diurnal cycles. On the other hand, although the GM-retrieved BLH showed a unimodal diurnal cycle, the amplitudes of the valley and peak were lower. Comparing the four-moment radiosonde-retrieved BLH (02:00, 08:00, 14:00, 20:00) with the algorithms' results highlights that CRGM presents the least bias, while GM shows an average underestimation of 500–600 m, and LGM and NGM result in an average overestimation of 400–500 m. To enrich our analysis, a comparison of CRGM with the other most frequently employed methods for BLH retrieval, such as the ideal curve fit (Steyn et al., 1999) and wavelet method (Davis et al., 2000), is provided in the Supplement (Figs. S4–S6). The result illustrates that fitting curve and wavelet methods also significantly underestimate BLH by approximately 600 and 800 m in maximum, respectively, during heavily polluted episodes in line with several other previous studies (Sawyer and Li, 2013; Wang et al., 2012; Su et al., 2017).
Histograms of the differences in BLH detected by
radiosonde and by CRGM, NGM, LGM, and GM. The
In order to further compare the performance of CRGM with the current
algorithms in heavily polluted episodes, the period of daily PM
Comparison of
To investigate the performance of the algorithms under clean meteorological
conditions, a comparison between the new and current algorithms is performed on
9 August 2008 with low PM
Figure 6b shows strong negative peaks in the profiles for each algorithm to determine the BLH. At 14:00 on 9 August the retrieved BLH for all algorithms is 1680 m, in perfect agreement with the actual BLH determined by radiosonde (1680 m). All diurnal cycle results converge at 14:00 on 9 August, demonstrating that all retrieval algorithms capture the overall diurnal cycle of the actual BLH. Comparison of CRGM with existing ideal curve fit and wavelet methods also confirms such performance (Fig. S7). Such good performance of all the algorithms under clean meteorological conditions is a result of the homogenous vertical distribution of aerosols, since under clean conditions, mixing of the aerosols by strong thermal convection is more sufficient due to weak pollutant loading. In addition, there is no obvious large negative gradient peak to disturb the determination of the BLH.
Root mean square error (RMSE) for each lidar retrieval method compared with radiosonde measurements and sample size in each comparison level.
Under various air pollution conditions (all pollution levels), a total of
298 radiosondes measurements are analyzed to estimate the BLH with
comparison to retrieval algorithms. Cases of nocturnal BLH below the useful
lidar signal (before the overlap reaches 1) or thin cumulus cloud formations
at the upper boundary layer (resulting in large error in the retrieval) are
neglected. The cloud and rain detection follows the methods employed by the
Asian Dust and Aerosol Lidar Observation Network (AD-net) in East Asia,
supported by the World Meteorological Organization (WMO) Global Atmosphere Watch
(GAW) program. Rain was detected by color ratio (
Lidar is an appropriate instrument with which to determine the boundary layer height with high temporal and vertical resolution. In this paper, an intensive lidar observation campaign was conducted in Beijing to thoroughly evaluate the limitations of the current method for boundary layer height determination and develop an algorithm suitable to all pollution conditions. Incontestably, current commonly employed retrieval algorithms (first gradient method, logarithm gradient method, and normalized gradient method) are unable to determine the boundary layer height during heavily polluted episodes due to inhomogeneous vertical distribution of aerosols under stable meteorological conditions associated with the impact of vertically propagating gravity waves on the tropospheric structure. The gradient algorithm critically underestimates the boundary layer height by 30–1140 m, with an underestimation higher than 300 m occurring 70 % of the time. The logarithm and normalized gradient methods overestimate the boundary layer height, exceeding 300 m more than 85 % of the time.
The newly developed method (the cubic root gradient) considers the linear
instability theory of gravity waves to determine the boundary layer height
by capturing the vertical movement of aerosol at the transition between
waves and turbulence. As a result, the cubic root gradient method describes
the fluctuation of the boundary layer with the best correlation (
In terms of environmental implication, such innovation would technically contribute to improving the accuracy of regionally spatiotemporal distribution models and forecasts of aerosol loadings for an effective pollution control measure, in particular over a number of megacities in China, since accurately determining the boundary layer is one of the important factors of uncertainties and bias reduction for reasonable air pollution modeling and forecasts. However, further development and expansion of lidar observation system are needed notably under cloudy and rainy conditions in order to provide a greater benefit to pollution control management.
Contact Ting Yang (tingyang@mail.iap.ac.cn) for data requests.
The authors declare that they have no conflict of interest.
This work was supported by the Natural National Science Foundation of China (NSFC) (41305115, 41225019), Program 973 (2014CB447900), the Commonweal Project of the Ministry of Environmental Protection (201409001), and Program 863 (2014AA06AA06A512). Ting Yang is grateful for the invaluable emotional support received from her family over the years to overcome all the periods of darkness, and the endless happiness and courage received from her baby daughter. Edited by: F. Yu Reviewed by: three anonymous referees