A Dual-Wavelength Lidar Boundary Layer Height Detection Fusion Method and Case Analysis

Abstract

Accurate detection of the atmospheric boundary layer (ABL) is important for weather forecasting, urban air quality monitoring, and agricultural and ecological protection. In this study, we propose a new method for enhancing ABL height detection accuracy by integrating multi-channel polarized lidar signals at 355 nm and 532 nm wavelengths. Radiosonde observations and ERA5 reanalysis are used to validate the lidar-derived results. By calculating the gradients of signals of different wavelengths and weighted fusion, the position of the top of the boundary layer is identified, and corresponding weights are assigned to signals of different wavelengths according to the signal-to-noise ratio of the signals to obtain a more accurate atmospheric boundary layer height. This method can effectively mitigate the influence of noise and provides more stable and accurate ABL height estimates, particularly under complex aerosol conditions. Three case studies of ABL height detection over the Beijing region demonstrate the effectiveness and reliability of the proposed method. The fused ABLHs were found to be consistent with the sounding data and ERA5. This research offers a robust approach to enhancing ABL height detection and provides valuable data support for meteorological studies, pollution monitoring, and environmental protection.

1. Introduction

The atmospheric boundary layer (ABL) typically extends from the surface up to approximately 1–2 km in altitude under average conditions, although its height can vary from several hundred meters to over 3 km depending on local meteorology and time of day [1,2,3]. The ABL plays a crucial role in the exchange of energy and matter between the Earth’s surface and the free atmosphere [4,5,6]. Meteorological conditions are affected by surface characteristics such as topography, vegetation, and urban development, as well as by external factors, including solar radiation, ground thermal properties, and wind dynamics. These influences often manifest as significant gradient changes in elements such as temperature, humidity, and wind speed [7,8]. Based on the thermal stability of the lower atmosphere, the ABL can be classified into three types: the stable boundary layer, the neutral boundary layer, and the convective boundary layer [1,9]. There are various observation methods for the atmospheric boundary layer height (ABLH), mainly including traditional means such as radiosonde, satellite remote sensing, lidar detection, and machine learning methods [10,11,12]. Radiosonde measurements analyze the thermodynamic structure of the atmosphere by vertically profiling temperature, humidity, and wind speed, although their spatial and temporal coverage is limited, typically providing only two observations per day [13,14]. Satellite remote sensing technology monitors the ABL by using radiation data of different bands (such as infrared, microwave, visible light, etc.), and can provide multi-dimensional information such as temperature, humidity, aerosols, clouds, and wind fields [15,16,17]. With broad spatial coverage and high temporal resolution, it serves as a vital tool for weather forecasting, climate change research, and air quality monitoring. Nevertheless, satellite remote sensing faces challenges related to spatial resolution and atmospheric interference, and its data processing and interpretation still require further improvement [18,19]. Numerical simulation methods predict the behavior of the ABL by modeling atmospheric physical processes through computer algorithms, combined with observational data. They are well-suited for large-scale, continuous forecasting and are particularly advantageous under complex climatic conditions or extreme weather events [20,21,22]. However, the accuracy of numerical simulations depends on the quality of input data and the underlying model assumptions, and their high computational demands pose limitations for rapid-response applications [13,17,18].

The lidar detection method employed in this study represents a key application within the field of photonics. As the science of light generation, propagation, detection, and interaction, photonics provides both the theoretical foundation and technological support necessary for the high-precision observation of the ABL. Lidar, by emitting laser pulses and analyzing the returned signals, enables precise detection of airflow, turbulence, wind speed variations, and water vapor distribution within the ABL, thereby providing real-time dynamic information [23,24,25]. This capability makes it particularly effective under complex weather conditions. Among various types, aerosol lidar is the most commonly used method for detecting the atmospheric boundary layer height. The underlying principle is that aerosol particles are primarily concentrated within the ABL, with their concentrations significantly higher than those in the free atmosphere. In the transitional region between the ABL and the free atmosphere, the entrainment process causes the upper dry and clean air to descend into the ABLH [8]. The gradient method determines the ABLH by identifying the extrema in the aerosol concentration gradients of lidar return signals. Variants of this approach include the first-order gradient method, inflection point method, and logarithmic gradient method [26]. The threshold method defines the ABLH as the altitude where the lidar signal or its gradient exceeds a predefined threshold, though this typically yields relatively coarse results. The CWT method detects the ABLH by comparing the similarity between the lidar signal and wavelet functions, such as the Haar or Mexican Hat wavelets [27]. This technique is widely applied in the ABLH detection due to its ability to suppress noise and its suitability for processing complex aerosol structures and non-stationary signals. Despite its relatively high computational complexity, the continuous wavelet transform method offers significant advantages in practical applications, particularly for large-scale data analysis and high-precision extraction of the ABLH [14,18]. In addition, more methods have been proposed to handle complex aerosol structures, such as multi-layer recognition methods and image processing methods, to analyze the height of ABLH by analyzing the extreme values of backscattered images [10,14,18]. However, in the existing literature, most of these methods focus solely on the backscatter signal at a single wavelength, with other lidar detection parameters rarely considered [1]. The volume linear depolarization ratio (VLDR) is a key parameter for distinguishing aerosol types, as it reflects the morphology of atmospheric particles. It is defined as the ratio of the perpendicular to the parallel component of the lidar return signal [26]. A lower VLDR indicates more regular or spherical particles, whereas a higher VLDR suggests more irregular or elongated shapes [28,29]. Determining particle shape is highly relevant when probing the structure and extent of the atmospheric boundary layer (ABL) because aerosol shape directly influences optical properties and atmospheric behavior. Spherical and regular-shaped particles, such as liquid droplets, tend to occur above the ABL, while irregular or mixed aerosols, like dust or pollution, are more common within the ABLH. As a result, the vertical profile of depolarization provides valuable information for identifying the ABL height and transitions [6,18]. Therefore, it is essential to incorporate polarization measurements at 532 nm along with backscatter signals at 355 nm and 1064 nm when probing the structure and extent of the ABL. In this study, a two-wavelength signal-to-noise ratio weighted fusion algorithm is proposed for retrieving atmospheric boundary layer heights in Beijing. The accuracy of the method is analyzed through the application of three lidar observations at Huairou Station from 6 to 11 November 2014, in comparison with co-located soundings and ERA5 reanalysis. The method fills the gap of the single-wavelength method, which suffers from a lack of stability during soundings.

2. Data and Methods

2.1. Observation Site

This study primarily used ground-based meteorological and lidar data from Huairou, Beijing. The Huairou Meteorological Station, located in the Huairou District of Beijing, serves as a key site for meteorological research, climate monitoring, and early disaster warning. It provides essential atmospheric parameters, including temperature, humidity, wind speed, pressure, and precipitation, supporting comprehensive environmental and boundary layer studies. The station is situated in a suburban area northeast of central Beijing, characterized by relatively low building density, and surrounded by mixed terrain, including gentle hills, farmland, and some scattered residential and industrial developments. Compared to urban core sites, Huairou experiences fewer anthropogenic emissions and less urban heat island influence, making it suitable for studying natural boundary layer processes with minimal direct urban disturbance. During the period from 6 November to 10 November 2014, measurements were conducted twice daily at 07:15 and 19:15 Beijing time, at 12 h intervals.

2.2. Instruments

2.2.1. Lidar

Dual-wavelength polarization lidar offers the capability to directly observe the vertical structure of the atmosphere. It not only provides high resolution in the lower troposphere but also enables effective identification of the ABL top. The ground-based lidar system used in this study was developed by the Key Laboratory of Atmospheric Optics, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences. It is a multi-wavelength aerosol lidar installed at the Yanqi Lake Campus of the University of Chinese Academy of Sciences (40.41° N, 116.68° E). The lidar is capable of stable, continuous 24 h operation.

To analyze the vertical distribution of aerosols, lidar experiments were conducted from 6 November to 10 November 2014. In this study, to regulate the operating condition of the laser, the lidar system was set to pause for 4 min after every 15 min of continuous operation. Measurements were suspended during rain or snow events. Due to the overlap between the emitted laser pulse and the return signal, which affects the accuracy of lidar measurements in the near-ground region, the lidar signal analysis in this study begins at an altitude of 225 m above ground level [16]. The specific parameters of the lidar system are listed in

Table 1. Detailed information on the multi-wavelength meter scattering lidar.

Technical Indicator	Parameter
Laser type	Nd:YAG
Laser wavelength/nm	355, 532
Receive channel	355 nm, 532 nm ||, 532 nm ⊥
Pulse energy/mJ	50,90
Pulse width/ns	20
Repetition rate/Hz	20
Receiving Telescope/mm	400 diameter, Cassegrain
Vertical resolution/m	7.5

, and the system configuration is illustrated in Figure 1. The lidar system includes three receiving channels: a 355 nm channel, a 532 nm channel for parallel polarization (‖), and a 532 nm channel for perpendicular polarization (⊥). These channels are respectively used to measure aerosol scattering properties, backscattered signal intensity, and depolarization ratio.

2.2.2. Radiosonde

The radiosonde data used in this study were obtained from the Huairou region in Beijing. As part of Beijing’s meteorological observation network, the Huairou sounding station plays a significant role in both meteorological monitoring and scientific research. The data collected were used to analyze the ABLH. In this study, the data were compiled and averaged with a vertical resolution of 10 m.

2.2.3. ERA5

The ERA5 dataset, provided by the European Centre for Medium-Range Weather Forecasts (ECMWF), is a global atmospheric reanalysis product designed to support climate research, weather forecasting, and a broad range of environmental sciences with high-precision historical meteorological data [30]. Released in 2017, ERA5 is a next-generation global reanalysis dataset covering atmospheric variables from 1950 to the present, with hourly temporal resolution. It offers a wide array of meteorological parameters, including temperature, humidity, wind speed, precipitation, and atmospheric pressure. ERA5 data are generated through the integration of numerical weather prediction models and observational data using advanced data assimilation techniques and complex numerical computations [31,32]. The dataset provides continuous meteorological fields and is widely applied in studies on climate change, weather forecasting, disaster monitoring, and model validation. Importantly, ERA5 not only delivers global spatiotemporal meteorological information but also allows users to download data at various spatial and temporal resolutions, making it highly valuable for both large-scale climate research and localized weather event analysis.

2.3. Different ABLH Algorithms

2.3.1. Traditional Lidar ABL Algorithm

The working principle of lidar involves emitting laser beams from a laser source and detecting the light scattered back by the atmosphere. Aerosol particles in the atmosphere play a key role in attenuating the laser signal due to their scattering effect, making them significant extinction factors in atmospheric laser transmission. In this study, the mixing layer refers to the convective portion of the ABL characterized by strong vertical turbulence and aerosol mixing. This study defines the mixing layer as the turbulent part of the ABL, where surface-driven convection leads to vigorous vertical mixing. It generally corresponds to the convective boundary layer during the daytime. Compared to the free atmosphere above, the mixing layer contains a higher concentration of aerosols, which enhances the scattering of the laser beam and contributes to stronger lidar backscatter signals. As a result, the ABLH can be readily identified by analyzing the gradient of the backscattered signal. The traditional lidar-ABLH consists of a gradient method, a threshold method, and a continuous wavelet variation method [26]. In the gradient method, the vertical gradient of the echo signal (or its logarithm) is first calculated, and the top of the ABL is located at the extreme value of the gradient plunge; in the threshold method, the signal intensity is compared with the empirical threshold of the free atmospheric background, and the first time that the echo is lower than this threshold is the ABLH, which is a simple method, but is more sensitive to the change in aerosol loading; the continuous wavelet transform (CWT) is used to convolve the logarithmic echo contour with the mother wavelet and search for mutant layers indicated by the pair of positive–negative extremes, and can automatically identify multi-layer structures, which is suitable for complex boundary layer scenarios. The CWT convolves the logarithmic echo profile with the mother wavelet to find the mutant layer labeled by the positive–negative extreme value pairs.

The lidar equation is typically expressed as follows [33]:

$$P(Z)={\displaystyle \frac{C}{Z^2}}\beta (Z)\exp \left[-2{\displaystyle {\int }_0^Z\alpha }\left(Z^{\prime }\right)d{\mathrm{Z}}^{\prime }\right]$$

(1)

$$\mathrm{R}(Z)=P(Z)Z^2=C\beta (Z)\exp \left[-2{\displaystyle {\int }_0^Z\alpha }\left(Z^{\prime }\right)d{Z}^{\prime }\right]$$

(2)

Here, $C$ is the lidar constant, typically obtained through calibration. The backscatter coefficient $\beta (Z)$ represents the intensity of laser light scattered back toward the lidar per unit volume and per unit solid angle at a given height $Z$. It accounts for the combined contribution of atmospheric molecules (Rayleigh scattering) and aerosol particles (Mie scattering). In the lidar equation, $\beta (Z)$ serves as the primary source of the backscattered signal received by the detector, determining the strength of the return signal from the atmosphere. $\alpha (Z)$ represent the total backscatter coefficient and total extinction coefficient at height $Z$; the signal strength decays exponentially with distance during the laser propagation process. Therefore, $\alpha (Z)$ is the key parameter that determines the signal strength.

Since $\alpha (Z)$ and $\beta (Z)$ coexist and are coupled with each other in the lidar equation, it is impossible to directly separate the two based solely on single-channel echoes. Therefore, the following assumptions are often adopted in the inversion process:

$$S={\displaystyle \frac{\alpha \left(Z\right)}{\beta (Z)}}$$

(3)

This formula reflects the ratio between the scattering and extinction of atmospheric components. Different types of aerosols have different Lidar ratios, and the general value is 50 [10,13].

2.3.2. New Fusion Algorithm

According to Mie scattering theory, the sensitivity of a laser to particles of a specific size can be explained using the size parameter, which is defined as [13,33]:

$$x={\displaystyle \frac{2\pi r}{\lambda }}$$

(4)

where $r$ is the particle radius and $\lambda$ is the laser wavelength. According to Mie scattering theory, when the particle size is comparable to the wavelength of the incident light ($x$ ≈ 1), the scattering efficiency is significantly enhanced. Fine particulate matter with radii in the range of 100–200 nm falls within this sensitive range when irradiated with a laser at 355 nm. This enables the lidar system to effectively detect high aerosol concentrations in the lower atmosphere during polluted conditions, thereby facilitating accurate retrieval of the ABLH.

Polluted weather refers to conditions with high concentrations of aerosols and pollutants trapped within the atmospheric boundary layer due to weak winds, stable stratification, or temperature inversion. These conditions lead to strong near-surface lidar backscatter and can affect the detection of boundary layer height. High concentrations of aerosols, especially fine particulate matter such as PM2.5 and PM10, influence the dynamics of the ABL mainly through their interaction with solar radiation [10,14,16]. These aerosols scatter and absorb incoming solar radiation, reducing surface heating. This suppresses turbulent mixing in the ABL and lowers its height, thereby modifying its dynamic and thermodynamic structure. Under polluted weather conditions, the concentration of aerosols in the atmosphere increases significantly. This can directly affect the dynamic characteristics of the atmospheric boundary layer. Traditional single-wavelength lidar cannot accurately measure the height of the ABL, and multi-wavelength lidar provides an important means. The parallel and perpendicular components at 532 nm enable the characterization of scattering properties of various aerosol types, facilitating more accurate retrievals of particle type, concentration, and distribution. This capability is particularly valuable during dust transport events, where detailed vertical profiling is essential for identifying subtle variations in boundary layer structure. In urban aerosol detection, where particle radii are typically small, the 355 nm wavelength provides enhanced sensitivity to fine particles. Additionally, due to the presence of external aerosol sources such as dust in urban environments, the 532 nm channel complements the detection of coarser particles. This dual-wavelength approach improves responsiveness to high aerosol concentrations and haze conditions, strengthens boundary layer identification precision, and increases adaptability to various aerosol types and pollutant conditions. Therefore, this study proposes a novel method that combines the 355 nm and 532 nm parallel and polarization channels to improve the accuracy of ABLH detection.

First, the lidar signal undergoes range correction to eliminate attenuation caused by varying propagation distances. The signal is denoised and normalized to ensure consistency and comparability among different wavelength channels. Normalization maps the signal onto a unified scale, facilitating comparison and integration of different wavelengths. Next, the gradients of the 355 nm and 532 nm signals are computed to identify regions where the rate of change is minimal. These gradient minima typically correspond to the top of the ABL, characterized by sharp transitions in aerosol concentration. By calculating the gradient minima separately for the 355 nm and 532 nm channels, two independent estimates of the ABLH are obtained. Specifically, weights are assigned based on the SNR of each signal, with higher SNR channels contributing more to the final estimation. The fused ABLH, thus, represents an optimal integration of both signals. This approach not only effectively reduces noise and improves signal quality but also leverages the complementary information from multiple wavelengths to enhance the precision of boundary layer height detection. The fused ABLH is calculated as follows:

$${\mathrm{ABLH}}_{\mathrm{fused}}={\displaystyle \frac{{\omega }_{355}\cdot {\mathrm{ABLH}}_{355}+{\omega }_{532}\cdot {\mathrm{ABLH}}_{532}}{{\omega }_{355}+{\omega }_{532}}}$$

(5)

$${\omega }_{\lambda }={\displaystyle \frac{{\mathrm{SNR}}_{\lambda }}{{\mathrm{SNR}}_{355}+{\mathrm{SNR}}_{532}}}$$

(6)

where ${\omega }_{\lambda }$ is the weight of different wavelengths, ${\mathrm{ABLH}}_{\mathrm{fused}}$ is the height of the fused boundary layer, and ${\mathrm{SNR}}_{\lambda }$ is the signal-to-noise ratio. Here, the SNR for each channel is defined as the ratio of the mean signal intensity to its standard deviation. By calculating the gradient minima separately for the 355 nm and 532 nm channels, two independent estimates of the ABLH are obtained. Given the differing signal-to-noise ratios (SNRs) of each channel, a weighted fusion approach is applied to improve the reliability of the result. Specifically, weights are assigned based on the SNR of each signal, with higher SNR channels contributing more to the final estimation. The fused ABLH thus represents an optimal integration of both signals. This approach not only effectively reduces noise and improves signal quality but also leverages the complementary information from multiple wavelengths to enhance the precision of boundary layer height detection. The corresponding flowchart above is shown in Figure 2.

Compared with traditional approaches—the gradient method, which is prone to signal-to-noise fluctuations and spurious extrema; the threshold method, which relies on empirical thresholds and is highly sensitive to aerosol loading; and the continuous wavelet transform (CWT), which requires manual selection of mother-wavelet type and scale and involves substantial computation—the proposed dual-wavelength, SNR-weighted fusion algorithm combines the complementary gradient information from the 355 nm and 532 nm lidar channels. This dual-channel synergy suppresses noise and thin-cloud contamination, yielding stable ABLH estimates under both pristine and heavily polluted conditions. Because the scheme dispenses with external thresholds and wavelet parameters, it remains computationally light and suitable for real-time deployment while still automatically resolving multilayer boundary structures. Consequently, it offers a more reliable and generalizable determination of atmospheric boundary layer height in complex environments.

Figure 3 shows the lidar intensities and boundary layer heights at different time points calculated by the new method. In each pair of panels, the left panel shows the backscatter intensity profiles (black for 532 nm, green for 355 nm), and the right panel shows the corresponding vertical gradient profiles. Horizontal dashed lines indicate the ABLH estimates based on 532 nm (red), 355 nm (blue), and the final combined result (black). The signal intensities (a), (c), (e), and (g) on 4, 6, 21 and 22 November are selected to represent the signal intensities of different time periods, respectively, and (b), (d), (f), and (h) represent the corresponding gradient profiles after differentiation. As can be seen from the figure, the height of the boundary layer is higher during the day and lower at night.

2.3.3. Richardson Number Method

The Richardson number method is a widely used approach for analyzing the ABLH. This method determines the ABLH by calculating the Richardson number (Ri), which represents the ratio between atmospheric static stability and vertical wind shear. It provides insight into the dynamic and thermal structure of the atmosphere. The bulk Richardson number is typically calculated using the following formula [34]:

$$Ri(z)={\displaystyle \frac{\frac{\mathrm{g}}{{\theta }_{vs}}({\theta }_{vz}-{\theta }_{vs})(z-z_s)}{{(u_z-u_s)}^2+{(v_z-v_s)}^2+b{u}_{*}^2}}$$

(7)

Here, z represents the height above the ground, and s represents surface-level quantities. ${\theta }_v$ is the virtual potential temperature, while $u$ and $v$ represent the horizontal wind components. $u_{*}$ represents the surface friction velocity and b is a coefficient. In the calculation process, since the magnitudes of the wind components $u$ and $v$ are generally much larger than the surface friction velocity $b{u}_{*}$, the $b{u}_{*}$ term is typically neglected.

According to Seidel et al. [35], the surface wind speed is set to zero in the Richardson number calculation. By integrating multiple meteorological factors such as temperature, humidity, pressure, and wind speed, the Richardson number method identifies the ABLH based on the thermodynamic characteristics of the atmosphere. One of the key advantages of this method is its comprehensive consideration of various atmospheric variables, which enables a relatively accurate estimation of ABLH. Comparative studies using critical Richardson number values of 0.20, 0.25, and 0.303 have shown that 0.25 provides the most reliable estimate for the ABLH [36,37]. This threshold reflects a balance between dynamic and thermal effects, making it particularly suitable for identifying the ABL top under varying atmospheric conditions. Therefore, in this study, a critical Ri value of 0.25 is adopted for retrieving the ABLH from the radiosonde. Figure 4 shows the ABLH obtained by calculating the temperature from the radiosonde data. It is acknowledged that the retrieved ABLH values, particularly in the afternoon profiles, may appear lower than expected. These results are dependent on the local meteorological conditions and the sensitivity of the Ri-based retrieval method. Due to the lack of direct validation data, the ABLH has not been verified at this stage.

3. Results

Figure 5 shows the intensity of the lidar backscatter signal, with red color indicating stronger backscatter, illustrating the temporal variation in the ABL. Panels (a) and (b) display the vertical distributions of the depolarization ratio at 532 nm and the backscatter coefficient at 355 nm, respectively. The black and red dashed lines represent the ABLH obtained by the gradient method at 355 nm and 532 nm, respectively; the yellow curves represent the final fused ABLH results, and the pink dots correspond to the ABLHs derived from the radiosonde.

In the early morning, due to stable stratification and residual aerosol accumulation overnight, the ABLH remained around 1.5–2 km. Around midday, enhanced solar radiation induced surface heating and strong convective mixing, causing the ABLH to rapidly rise and peak at 2.5–3 km between 13:00 and 15:00. During this period, lidar occasionally detected thin cloud layers at altitudes of 6–7 km. Toward evening, the weakening of solar radiation and decay of turbulence led to a gradual decline in the ABLH. After sunset, rapid surface cooling and stabilized stratification caused the ABLH to drop to around 0.8 km, which was consistent with radiosonde observations.

The 532 nm retrieval generally overestimated the ABLH, while the 355 nm-derived values were lower before 21:00. Between 21:00 and 24:00, the 355 nm estimate exceeded that of 532 nm, indicating a higher concentration of fine aerosol particles in the atmosphere at that time.

Regarding surface meteorological conditions: at 06:00, temperature was as low as 4 °C, wind speed about 2 m/s, and relative humidity between 45 and 60%, resulting in a shallow ABLH due to the lack of thermal and mechanical turbulence. After sunrise, the temperature rose quickly to 7–8 °C, reaching approximately 10 °C around 14:00, accompanied by an increase in wind speed to 3.5 m/s and a sharp rise in humidity to near saturation. These factors together intensified thermomechanical mixing and lifted the ABLH to its daytime peak. In the afternoon, both temperature and wind speed declined while humidity remained high, weakening surface fluxes and convection, leading to a gradual drop in the ABLH. By evening and into the night, temperature fell rapidly to 2–3 °C, wind became nearly calm, and humidity increased again, stabilizing the stratification and suppressing turbulence, causing the ABLH to collapse to approximately 0.8 km, consistent with radiosonde measurements. The observed abrupt variations in the ABLH—reaching up to approximately 1.5 km within a short time—are primarily caused by a combination of physical processes and lidar retrieval characteristics. During the daytime, intense solar heating of the surface leads to the rapid development of turbulence, allowing the mixed layer to rise from around 1 km to 2.5–3 km within 1–2 h.

Figure 6 illustrates the variations in the ABLH and meteorological parameters between 6:00 and 24:00 from 9 to 10 November 2014. In the early morning, the atmosphere was stably stratified, and residual aerosols supported a shallow boundary layer at approximately 1 km. After 08:00, surface heating due to solar radiation initiated thermal convection, and the ABLH rose and stabilized around 2.0–2.3 km. During this period, the 532 nm lidar channel detected aerosol tops reaching 3.5–4 km, indicating well-developed convective mixing, while the 355 nm-derived ABLH remained lower, underestimating the actual boundary layer height.

Around 16:00, a sharp drop in relative humidity combined with weak wind speed reduced surface heat flux, resulting in a declining trend in the ABLH. Following sunset, the rapid decay of turbulence caused the ABLH to collapse to about 0.5 km. After 18:00, a stable nocturnal layer formed, and the ABLH remained in the range of 0.6–0.9 km for an extended period. The ABLH retrieved by the fused algorithm closely matched those derived from radiosonde observations and proved to be more accurate than either the 532 nm or 355 nm channels alone.

Meteorological conditions support these findings: in the early morning, the temperature was only 0–1 °C, relative humidity ranged from 75 to 85%, and wind speed was around 2 m/s, maintaining atmospheric stability and suppressing the development of thermomechanical turbulence. After 09:00, the temperature rose rapidly to around 10 °C, wind speed initially increased to 2.5 m/s, then gradually weakened, and relative humidity generally stayed within 70–90%. Two abrupt drops in humidity (to 50% at 09:00 and to 10% at 16:00) indicate intrusions of dry, warm air aloft into the ABL. These conditions, combined with elevated temperatures and moderate wind, led to ABLH enhancement. However, in the late afternoon, the weakening of wind speed and the reduction in temperature and humidity gradients led to the simultaneous decline of both thermal and mechanical turbulence, resulting in a sharp drop in the ABLH. At night, temperatures fell to 1–3 °C, winds were nearly calm, and humidity increased to 40–50%, reinforcing stable stratification and trapping aerosols near the surface, thereby maintaining a low ABLH.

Figure 7 illustrates the evolution of the ABLH and meteorological parameters between 06:00 and 24:00 from 10 to 11 November 2014. During the early morning hours (06:00–08:00), a nocturnal inversion persisted, limiting vertical mixing, and residual aerosols kept the ABLH at approximately 1 km. As solar radiation intensified, convective development rapidly progressed between 09:00 and 15:00, lifting the ABLH and stabilizing it around 2.0–2.3 km. During this period, aerosol layers were mixed up to 3–3.5 km. Between 16:00 and 17:00, weakening surface heat flux and the intrusion of descending dry, warm air led to a marked decline in the ABLH, which fell below 1.5 km by sunset. After sunset, radiative surface cooling re-established a stable stratification, and the ABLH remained low, between 0.6 and 0.9 km, for the rest of the evening. The fused ABLH estimates more accurately captured the structure and evolution of the ABL than those derived individually from the 355 nm and 532 nm channels using the gradient method, showing close agreement with radiosonde observations.

From a meteorological perspective, the low ABLH during early morning was caused by the combined effects of a surface-based inversion and only moderate mechanical turbulence, which restricted mixing. After sunrise, rapid surface warming overcame the brief weakening of wind speed, and strong buoyancy-driven convection lifted the ABLH. Around midday, peak temperatures coupled with moderate wind speeds sustained efficient thermo-mechanical mixing, maintaining a high-level ABLH plateau. In the afternoon, the weakening of surface heat flux and the intrusion of dry, subsiding air caused a sharp drop in humidity and a reduction in mixing efficiency, initiating the decline in the ABLH. From evening into night, surface radiative cooling, near-calm winds, and rising humidity reestablished a stable stratification. As turbulence dissipated, the ABLH contracted sharply and remained at a low level.

Figure 8 compares the ABLH retrieved from the ERA5 reanalysis dataset with that obtained using the proposed fusion algorithm. To minimize the influence of outliers, extreme values were excluded during the analysis, and a linear fit was applied to the remaining data. The figure shows a clear linear trend: as the ERA5-derived ABLH increases, the ABLH estimated by the fusion algorithm also exhibits a strong upward trend. The correlation coefficient (R) of the fit is 0.79, indicating a moderate positive correlation between the two datasets. The root mean square error (RMSE) is 422.97, and the mean deviation (MD) is 290.17. The observed deviations can be attributed to both the limitations of the ERA5 dataset and the characteristics of the fusion algorithm. ERA5 is widely used for atmospheric reanalysis due to its comprehensive temporal and spatial coverage. It features a horizontal resolution of approximately 31 km. Vertically, it is structured with 137 model levels, with layer spacing near the surface ranging from tens to hundreds of meters. These spatial constraints hinder its ability to capture rapid variations in the local ABLH. In contrast, lidar measurements offer a much finer vertical resolution (e.g., 7.5 m), making them highly suitable for monitoring boundary layer structures. The coarser resolution of ERA5 data may thus fail to resolve localized meteorological processes, leading to significant discrepancies. The coarser resolution of ERA5 data may thus fail to resolve localized meteorological processes, leading to significant discrepancies. This is particularly the case if the fusion model is prone to overfitting or lacks robustness under specific boundary conditions. Meanwhile, ERA5, being based on a general circulation model, may not respond quickly to such localized changes, thereby further amplifying the deviation between datasets.

4. Conclusions

This study proposes a novel method for detecting the ABLH based on multi-channel lidar observations. By combining dual-wavelength lidar signals at 355 nm and 532 nm, the method significantly improves the accuracy and reliability of ABLH retrieval. Through a weighted fusion approach integrated with the gradient method, the algorithm enables precise identification of the ABL top, particularly under complex meteorological conditions and during episodes of severe air pollution.

The proposed method integrates lidar signals at 355 nm and 532 nm using a weighted fusion algorithm, yielding enhanced accuracy in boundary layer detection. Experimental results demonstrate that the fused ABLH estimates align well with radiosonde observations and ERA5 reanalysis data, showing strong adaptability and precision. Analysis of data from 6 November to 10 November 2014 reveals typical diurnal variations in the ABLH over Beijing. The ABLH is lowest in the early morning, rises rapidly with increasing solar radiation, and reaches its peak around midday. As solar heating weakens in the late afternoon, the ABLH gradually declines. The study also shows that the 532 nm channel tends to overestimate the ABLH, while the 355 nm channel underestimates it during certain periods—reflecting the influence of fine particulate matter on boundary layer structure. Compared with traditional single-wavelength lidar techniques, the dual-wavelength fusion method enables more accurate ABLH retrieval, especially in complex atmospheric environments. Experimental results across different time periods confirm that the fusion algorithm enhances detection accuracy and closely matches radiosonde-derived heights. When compared with ERA5 reanalysis data, the fused ABLH shows a correlation coefficient of 0.79, further validating the method’s reliability.

Despite the effectiveness of the proposed multi-wavelength lidar fusion approach in improving ABLH detection under complex weather conditions, some limitations remain. For example, signal quality may be affected under cloudy conditions. Future work will focus on further refining the processing algorithm to enhance performance in the presence of complex aerosol structures. Incorporating additional meteorological variables and high-resolution datasets will also be important for validating the algorithm’s robustness and general applicability.