2.1 Lidar system

The multiwavelength Mie–Raman lidar LILAS (LIlle Lidar AtmosphereS) is based on a tripled Nd:YAG laser with a 20 Hz repetition rate and pulse energy of 70 mJ at 355 nm. Backscattered light is collected by a 40 cm aperture Newtonian telescope, and the lidar signals are digitized with Licel transient recorders with a 7.5 m range resolution, allowing simultaneous detection in the analog and photon counting mode. The system is designed for the detection of elastic and Raman backscattering, allowing the so-called $\mathrm{3}\mathit{\beta }+\mathrm{2}\mathit{\alpha }+\mathrm{3}\mathit{\delta }$ data configuration, including three particle backscattering values (β₃₅₅, β₅₃₂, β₁₀₆₄) and two extinction (α₃₅₅, α₅₃₂) coefficients, along with three particle depolarization ratios (δ₃₅₅, δ₅₃₂, δ₁₀₆₄). The particle depolarization ratio, determined as a ratio of cross- and co-polarized components of the particle backscattering coefficient, was calculated and calibrated in the same way as described in Freudenthaler et al. (2009). Many calibration and operation procedures have been automated for the LILAS system to improve the overall performance of the lidar in terms of observation frequency and data quality. The aerosol extinction and backscattering coefficients at 355 and 532 nm were calculated from Mie–Raman observations (Ansmann et al., 1992), while β₁₀₆₄ was derived by using the Klett (1985) method. The full geometrical overlap was achieved at approximately 750 m range. For the calculation of α and β at 532 nm, we use the rotational Raman scattering instead of the vibrational one (Veselovskii et al., 2015), which allows us to increase the power of Raman backscatter and to decrease separation between the wavelengths of elastic and Raman components. Additional information about atmospheric parameters was available from radiosonde measurements performed at Herstmonceux (UK) and Beauvechain (Belgium) stations, located 160 and 80 km away from the observation site, respectively.

The LILAS system can also profile the laser-induced fluorescence of aerosol particles. A part of the fluorescence spectrum is selected by a wideband interference filter of 44 nm width centered at 466 nm. The strong sunlight background during the daytime restricts the fluorescence observations to nighttime hours. The fluorescence backscattering coefficient, β_F, is calculated from the ratio of fluorescence and nitrogen Raman backscattering signal, as described in Veselovskii et al. (2020b). This approach allows us to evaluate the absolute values of β_F, if the relative sensitivity of the channels is calibrated and the nitrogen Raman scattering differential cross section is known. All β_F profiles presented in this work were smoothed with the Savitzky–Golay method, using second-order polynomials with 21 points in the window. For the calculation of the fluorescence capacity G_F, in principle, backscattering coefficients at any laser wavelength can be used. In our study, we always used β₅₃₂ because it is calculated with the use of a rotational Raman component and is considered to be the most reliable; thus, the fluorescence capacity is calculated as $G_{\mathrm{F}}=\frac{{\mathit{\beta }}_{\mathrm{F}}}{{\mathit{\beta }}_{\mathrm{532}}}$.

2.2 Calculation of the particle backscattering coefficient from Mie–Raman measurements

Mie–Raman lidar measurements allow the independent evaluation of aerosol extinction and backscattering coefficients. A commonly used approach for β calculation was formulated in the paper of Ansmann et al. (1992). This approach includes the choice of a reference height, where the scattering is purely molecular. However, such a height range is not always available, for example, in the presence of the low-level clouds. Moreover, when long-term spatiotemporal variations in backscattering coefficients are analyzed, then the uncertainty in the choice of the reference height leads to oscillations in β profiles. To resolve this issue, we modified the Raman method as described below.

In an elastic channel, the backscattered radiative power P_L, at wavelength λ₀ and distance z, is described by the lidar equation as follows:

while in a Raman channel, it can be written as follows:

Here O(z) is the geometrical overlap factor, which is assumed to be the same for elastic and Raman channels. C_L and C_R are the range independent constants, including the efficiency of the detection channel. T_L and T_R are one-way transmissions which describe light losses on the way from the lidar to distance z at laser λ_L and Raman λ_R wavelengths. Backscattering and extinction coefficients contain the following aerosol and molecular contributions: ${\mathit{\beta }}_{\mathrm{L}}^{\mathrm{a}}+{\mathit{\beta }}_{\mathrm{L}}^{\mathrm{m}}$ and ${\mathit{\alpha }}_{\mathrm{L}}^{\mathrm{a}}+{\mathit{\alpha }}_{\mathrm{L}}^{\mathrm{m}}$, where the superscripts a and m indicate aerosol and molecular scattering, respectively. The Raman backscattering coefficient is as follows:

where N is the number of Raman scatters (per unit of volume), and σ_R is the Raman differential scattering cross section in the backward direction.

Dividing Eq. (1) into Eq. (2), we obtain the following:

The backscattering coefficient is calculated from Eqs. (3) and (4) as follows:

The differential transmission $\frac{T_{\mathrm{L}}}{T_{\mathrm{R}}}$ can be calculated in the same way as it is done for the water vapor measurements (Whiteman, 2003). For a rotational Raman signal, which we use in our 532 nm channel (Veselovskii et al., 2015), λ_L≈λ_R, so that $\frac{T_{\mathrm{L}}}{T_{\mathrm{R}}}$ = 1.

The calibration constant $K=\frac{C_{\mathrm{R}}}{C_{\mathrm{L}}}{\mathit{\sigma }}_{\mathrm{R}}$ can be found by comparing ${\mathit{\beta }}_{\mathrm{L}}^{\mathrm{a}}$ in Eq. (5) with the backscattering coefficient ${\stackrel{\mathrm{̃}}{\mathit{\beta }}}_{\mathrm{L}}^{\mathrm{a}}$ computed with the traditional Raman method and using the reference height (Ansmann et al., 1992).

For simplicity, hereinafter we will use the notation β_L instead of ${\mathit{\beta }}_{\mathrm{L}}^{\mathrm{a}}$. Thus, if during the measurement session we have a temporal interval where the reference height is available, then we can determine the calibration constant K and use it for β_L calculations from Eq. (5), assuming that the relative sensitivity of the channels during the session is not changed. Even if cloud layers occur during the whole session, we can use K from the previous cloud-free profiles (assuming, again, that the relative sensitivity of the channels is the same). We will call this approach for β calculation the modified Raman method to distinguish it from the traditional one (Ansmann et al., 1992).

To estimate variations in the relative sensitivity of the channels, we analyzed long-term cloudless measurements when the reference height was available for every individual profile. The results demonstrate that variations in the calibration constant during the session (about 8 h) were below 3 %. Figures 1 and 2 present the application of this modified Raman method to the measurements on 2 March 2021. The dust layer extended from 2 to 8 km height, and inside this layer, the ice and liquid clouds were formed during the 00:00–05:00 UTC interval; thus, β₅₃₂ could not be calculated with the traditional Raman technique. The temporal interval of 19:00–20:00 UTC was used to find calibration constant K. Figure 1 shows the vertical profiles of backscattering coefficient ${\stackrel{\mathrm{̃}}{\mathit{\beta }}}_{\mathrm{532}}$, calculated with the traditional Raman method (with reference height), and β₅₃₂ ,calculated with the modified method (with the calibration constant). Profiles of ${\stackrel{\mathrm{̃}}{\mathit{\beta }}}_{\mathrm{532}}$ and β₅₃₂ coincide for the whole height range. The calibration constant K, shown on the same plot, does not demonstrate height dependence, though oscillations around the mean value increase with height. For computations, we choose the value of K at low altitudes averaged inside some height interval.

Figure 1Backscattering coefficients at 532 nm for the period 19:00–20:00 UTC on 2 March 2021, calculated from Mie–Raman observations, using the same reference height as Ansmann et al. (1992; green) or the calibration constant as in Eq (5) (magenta). The profile of the calibration constant K is shown with a red line.

Download

Figure 2Spatiotemporal distributions of the backscattering coefficient β₅₃₂, the particle depolarization ratio δ₅₃₂, and the fluorescence capacity G_F on the night of 2–3 March 2021. The backscattering coefficient β₅₃₂ is calculated with the modified Raman method. The values of δ₅₃₂ and G_F are shown for β₅₃₂ > 0.2 Mm⁻¹ sr⁻¹.

Download

Figure 2 provides spatiotemporal variations in β₅₃₂, particle depolarization δ₅₃₂, and the fluorescence capacity G_F. Depolarization measurements reveal the presence of dust (δ₅₃₂≈30 %) and the ice cloud above 4 km (δ₅₃₂ > 40 %). The liquid cloud below 4 km after midnight can be identified by a low depolarization ratio δ₅₃₂ < 3 %. The fluorescence capacity of dust is low, at about 0.2 × 10⁻⁴. However, below 2 km, G_F is significantly higher, up to 1.2 × 10⁻⁴. In combination with a high depolarization ratio (up to 20 %), it can indicate the presence of pollen at low altitudes. On the fluorescence capacity panel, we can see that, after 01:00 UTC, the dust and pollen layers are mixed below 2 km, resulting in a value of G_F of about 0.5 × 10⁻⁴. The fluorescence capacity inside ice and liquid clouds is below 0.01 × 10⁻⁴. Figure 2 clearly demonstrates the advantage of simultaneous depolarization and fluorescence measurements for the study of cloud formation in the presence of aerosol. All spatiotemporal distributions of β₅₃₂ presented in this paper were calculated from Eq. (5) with a modified Raman method.

3.1 Approach for aerosol classification

As was discussed in our recent publication (Veselovskii et al., 2021), the δ-G_F diagram allows us to separate several aerosol types, including smoke, dust, pollen, urban, and ice and liquid water particles. Smoke and urban aerosols both have a small depolarization ratio, but the fluorescence capacity of smoke is almost 1 order of magnitude higher, so these particles can be separated. Dust and pollen both have a high depolarization ratio (up to 30 %), but G_F of dust is significantly lower, which again provides the basis for discrimination. The depolarization ratio of some aerosol types is characterized by a strong spectral dependence. For example, the depolarization ratio of aged smoke decreases with wavelength. It is below 5 % at 1064 nm, but at 355 nm in upper troposphere, it may exceed 20 % (Burton et al., 2015; Haarig et al., 2018; Hu et al., 2019; Veselovskii et al., 2022), which complicates the smoke and dust separation. For pollen, on the contrary, the depolarization ratio at 1064 nm can be the highest (Veselovskii et al., 2021). Thus, the choice of δ₁₀₆₄ for the δ-G_F diagram could be advantageous. However, as already mentioned, the backscattering coefficient at 1064 nm is calculated with the Klett (1985) method, which, besides the assumptions about lidar ratio, needs a reference height and cannot be used in cloudy situations. This is why, in our study, we used the δ₅₃₂-G_F diagram.

In our present work, we consider a simple classification scheme since we use only two intensive parameters, G_F and δ₅₃₂. Our goal is to demonstrate that, in the δ₅₃₂-G_F diagram, our lidar observations form clusters and characteristic patterns which can be attributed to different aerosol types or their mixtures. We consider four aerosol types, i.e., dust, smoke, pollen and urban, and two cloud types, i.e., liquid and ice clouds. Dust and pollen are large particles of a complicated shape, characterized by high depolarization ratio, while smoke and urban pollution are small particles with low depolarization. In our classification, urban aerosol includes continental aerosol, sulfates, and soot. At this stage, we do not yet consider absorption to discriminate the particles.

The choice of the range of particle properties variations for each aerosol type is an important aspect of the approach. Typical ranges of G_F and δ₅₃₂ variations used in our classification scheme are given in Table 1 and are shown in Fig. 3. These ranges are based on results obtained in LOA (Laboratoire d'Optique Atmosphérique) and on results presented in aerosol classification studies (Burton et al., 2012, 2013; Nicolae et al., 2018; Papagiannopoulos et al., 2018; Mylonaki et al., 2021).

• Dust. The depolarization ratio, δ_532, of Saharan dust near the source regions is up to 35 % (Veselovskii et al., 2020a). However, after transportation and mixing with local aerosol, δ₅₃₂ can be as low as 20 % (Rittmeister et al., 2017). In many studies, the dust events having a smaller depolarization ratio are classified as polluted dust (e.g., Burton et al., 2012, 2013). At the moment, we do not introduce the discrimination between the two subtypes and mark as dust the particles with 20 % < δ₅₃₂ < 35 % and 0.1 × 10⁻⁴ < G_F < 0.5 × 10⁻⁴.

• Smoke. In 2021–2022, we regularly observed, over the ATOLL platform, smoke layers originating from Californian and Canadian forest fires (Hu et al., 2022). The particle depolarization and fluorescence capacity of this transported smoke varied from episode to episode, and for classification, we selected the ranges 2 % < δ₅₃₂ < 10 % and 2 × 10⁻⁴ < G_F < 6 × 10⁻⁴. At this stage, we do not discriminate between fresh and aged smoke, and the range of the δ₅₃₂ variation is similar to the one used in the classification of Burton et al. (2012).

• Pollen. The pollen over northern France is usually mixed with other aerosol, and the particles which we mark as pollen are actually the mixtures. The depolarization ratio of clean pollen varies strongly for different taxa. For birch pollen, Cao et al. (2010) reported δ₅₃₂ = 33 %, and in the measurements over Finland during birch pollination, Bohlmann et al. (2019) observed values of δ₅₃₂ up to 26 %. The observations over Lille during the pollen season (Veselovskii et al., 2021) rarely revealed values δ₅₃₂ exceeding 20 %. Based on those observations, we classify as pollen the particles mixtures with 15 % < δ₅₃₂ < 30 % and 0.8 × 10⁻⁴ < G_F < 3.0 × 10⁻⁴.

• Urban. This type of aerosol includes a variety of particle types (e.g., sulfates and soot), and its properties may depend on the relative humidity. Based on our measurements inside the boundary layer, for classification, we choose the ranges 1 % < δ₅₃₂ < 10 % and 0.1 × 10⁻⁴ < G_F < 1.0 × 10⁻⁴. A similar range for δ₅₃₂ is used in the classification of Burton et al. (2013). Urban and smoke particles both have a low depolarization, but the smoke fluorescence capacity can be up to 1 order of magnitude higher, so these particles can be reliably discriminated.

• Ice and water clouds. Both cloud types have low fluorescence capacity G_F < 0.01 × 10⁻⁴. However, the ice clouds are usually observed at the heights where the fluorescence signal is low and cannot be used for classification. Thus, above ∼ 8 km, the ice cloud are identified by a high depolarization ratio δ₅₃₂ > 40 %. The depolarization ratio of the liquid water clouds is usually affected by the effects of the multiple scattering, so for their identification, we use δ₅₃₂ < 5 %.

Table 1Ranges of particle depolarization δ₅₃₂ and fluorescence capacity G_F, which were used for aerosol classification.

Download Print Version | Download XLSX

Figure 3Aerosol type with a δ₅₃₂-G_F diagram. The ranges of the particle parameter variations for dust, pollen, smoke, and urban aerosol are given by rectangles. The symbols show the results of simulation performed for pollen and urban (circles) and smoke and dust (stars) mixtures. The relative contribution of pollen (smoke) to the total backscattering β₅₃₂ varied in the 0–1.0 range, with steps of 0.1. The particle parameters used in the calculations are given in the text.

Download

The analysis of aerosol mixtures is an important subject, and the possibility to separate the mixture components based on lidar measurements was discussed in publications of Sugimoto and Lee (2006), Groß et al. (2011), Gasteiger and Freudenthaler (2014), Tesche et al. (2009), and Burton et al. (2014). The information about the mixture composition can be also revealed in the δ₅₃₂–G_F diagram. For example, pollen can be mixed with urban particles. At different heights, the pollen contributes differently to β₅₃₂, so in the δ₅₃₂–G_F diagram, the data points will form the pattern, which extends from the location attributed to pure urban aerosol to the location attributed to pure pollen. To estimate how such a pattern looks, simplified modeling for fixed particle parameters was performed. The corresponding results are shown in Fig. 3 by the symbols (circles). The particle depolarization ratio δ of the mixture, containing urban aerosol (u) and pollen (p), with the depolarization ratios δ^u and δ^p, can be calculated as follows:

The fluorescence capacity of the mixture is given by the following:

Here, the total backscattering $\mathit{\beta }={\mathit{\beta }}^{\mathrm{u}}+{\mathit{\beta }}^{\mathrm{p}}$.

The computations in Fig. 3 were performed for values of the pollen contribution $\frac{{\mathit{\beta }}_{\mathrm{532}}^{\mathrm{p}}}{{\mathit{\beta }}_{\mathrm{532}}}$ in the 0–1.0 range, with steps of 0.1. We assume that the depolarization ratios of pollen and urban aerosol are ${\mathit{\delta }}_{\mathrm{532}}^{\mathrm{p}}$ = 30 % and ${\mathit{\delta }}_{\mathrm{532}}^{\mathrm{u}}$ = 3 %, while the fluorescence capacities are $G_{\mathrm{F}}^{\mathrm{u}}$ = 0.2 × 10⁻⁴ and $G_{\mathrm{F}}^{\mathrm{p}}$ = 2.5 × 10⁻⁴. We remind the reader that the fluorescence capacities are calculated at 532 nm wavelength. In the δ₅₃₂–G_F diagram, the computed points provide a characteristic curve, which in the next section will be compared with experimental results. The same computations were performed for a smoke (s) and dust (d) mixture, assuming ${\mathit{\delta }}_{\mathrm{532}}^{\mathrm{d}}$ = 30 %, ${\mathit{\delta }}_{\mathrm{532}}^{\mathrm{s}}$ = 3 %, $G_{\mathrm{F}}^{\mathrm{d}}$ = 0.2 × 10⁻⁴, and $G_{\mathrm{F}}^{\mathrm{s}}$ = 4.0 × 10${}^{-\mathrm{4}.}$. The corresponding results are shown in Fig. 3 with stars. In a similar way, the characteristic curves for other mixtures can also be represented.

We are also able to identify liquid water and ice layers. Liquid water cloud layers have low fluorescence capacity (G_F < 0.01 × 10⁻⁴) and δ₅₃₂ < 3 %. Ice particles also have low G_F, but at heights where ice clouds are usually observed, the signal of the fluorescence backscattering is noisy. Thus, at high altitudes, ice particles are discriminated by a high depolarization ratio δ₅₃₂ > 40 %.

Figure 4Spatiotemporal distributions of the backscattering coefficient β₅₃₂, the fluorescence backscattering coefficient β_F (in 10⁻⁴ Mm⁻¹ sr⁻¹), the particle depolarization ratio δ₅₃₂, and the fluorescence capacity G_F on the night of 12–13 September 2020. The calculation of δ₅₃₂ and G_F was not performed for β₅₃₂ < 0.2 Mm⁻¹ sr⁻¹. The values of the backscattering coefficient and the depolarization ratio of ice clouds are high (above 20 Mm⁻¹ sr⁻¹ and 40 %, respectively) and are off the scale for the maps presented.

Download

Figure 5(a) The δ₅₃₂-G_F diagram for the data from Fig. 4 in the 500–6000 m height range. Red crosses show the uncertainty of the measurements. (b) Spatiotemporal distribution of aerosol types on the night of 12–13 September 2020. The gray coloring shows undefined aerosol types, while measurements with β₅₃₂ < 0.2 Mm⁻¹ sr⁻¹ are marked in black.

Download

3.2 Classification of spatiotemporal observations

The input parameters in our classification scheme are the spatiotemporal distributions of β₅₃₂, δ₅₃₂, and G_F, which are presented as matrices ${\mathit{\beta }}_{\mathrm{532}}^{i,j}$, ${\mathit{\delta }}_{\mathrm{532}}^{i,j}$, and $G_{\mathrm{F}}^{i,j}$, where $i=\mathrm{1},\mathrm{\dots },N_{\mathrm{T}}$; $j=\mathrm{1},\mathrm{\dots },N_{\mathrm{H}}$. Values N_T and N_H are the numbers of temporal and height intervals in the analyzed dataset. In a single measurement, we accumulate 2 × 10³ laser pulses, so the temporal resolution of the measurements is about 100 s, while the height resolution is 7.5 m.

The particle intensive parameters cannot be evaluated reliably when the backscattering coefficient is low. Thus, we set a threshold value for β₅₃₂ (normally 0.2 Mm⁻¹ sr⁻¹), namely when ${\mathit{\beta }}_{\mathrm{532}}^{i,j}$ < 0.2 Mm⁻¹ sr⁻¹, then the elements of the matrices ${\mathit{\delta }}_{\mathrm{532}}^{i,j}$ and $G_{\mathrm{F}}^{i,j}$ are classified as low signal and ignored. For the remaining elements, we determine the aerosol type, using our approach. A primary classification is being made for each point (i,j) separately, in accordance with parameter ranges given in Table 1. The elements, which are out of all these ranges, are marked as undefined. We consider six types of the particles, i.e., dust, smoke, pollen, urban, ice crystals, and water droplets, respectively. Moreover, there can be two additional results of primary classification, namely undefined and low signal. Thus, there are altogether eight possible results of primary classification. For every aerosol type, a N_T×N_H dimension matrix is constructed. If at this first stage of classification some single pixel point (i,j) is classified as, e.g., dust, then the corresponding value in the dust matrix is set to 1; otherwise, it is set to 0.

Figure 6Spatiotemporal distributions of the backscattering coefficient β₅₃₂, the fluorescence backscattering coefficient β_F (in 10⁻⁴ Mm⁻¹ sr⁻¹), the particle depolarization ratio δ₅₃₂, and the fluorescence capacity G_F on the night of 30–31 May 2020. The values of δ₅₃₂ and G_F are shown for β₅₃₂ > 0.2 Mm⁻¹ sr⁻¹.

Download

Figure 7(a) The δ₅₃₂-G_F diagram for observations in the 500–2500 m height range and (b) spatiotemporal distribution of aerosol types on the night of 30–31 May 2020. The gray coloring shows an undefined aerosol type, which is a mixture of urban and pollen for this case. Measurements with β₅₃₂ < 0.2 Mm⁻¹ sr⁻¹ are marked in black.

Download

The single pixel particle parameters contain statistical noise, which influences the results of the primary classification, thus producing high-frequency oscillations of nonphysical character. From a physical point of view, the aerosol single-type areas should form smooth regions, so a special smoothing procedure (stage 2 of our algorithm) was developed to remove the oscillations. The smoothing procedure is based on a convolution with a Gaussian kernel, as follows:

where t and h are temporal and height coordinates. The resolution of the classification is being controlled by the parameters s_T and s_H, which are set as the number of temporal and height bins.

In the second stage of classification, each of these matrices is separately convoluted with the Gauss kernel Z. After the convolution, the values for each pixel (i,j) are compared. If, e.g., the dust matrix contains a maximal value at the pixel (i,j) in respect to all other matrices, then the pixel (i,j) is finally classified as dust. The choice of smoothing parameters depends on the aerosol loading and aerosol type. For the measurements inside the boundary layer, in many cases the single pixel classification (s_T = 1, s_H = 1) is possible, while for analysis of the weak elevated layers, the smoothing should be applied. All results presented in this study were obtained for s_T = 3 and s_H = 5; thus, the temporal and range resolutions of our classification procedure are estimated to be about 8 min and 60 m, respectively.

4.1 12 September 2020: wildfire smoke

Figure 4 presents the spatiotemporal variations in aerosol and fluorescence backscattering coefficients (β₅₃₂ and β_F), together with the particle depolarization ratio δ₅₃₂ and the fluorescence capacity G_F, during the smoke episode on the night of 12–13 September 2020. The smoke layer extends from approximately 2 to 5 km in height, and it is characterized by a high fluorescence capacity G_F > 3.0 × 10⁻⁴ and low depolarization ratio δ₅₃₂ < 7 %. The cirrus clouds occurred above 11 km height during the whole night. The smoke layer was transported from North America; detailed analysis of the layer origin and transportation is given in a recent publication of Hu et al. (2022). The results of the aerosol classification for this episode are shown in Fig. 5. On the δ₅₃₂-G_F diagram, these data form two clusters. The first cluster includes points in the range 2.0 × 10⁻⁴ < G_F < 6.0 × 10⁻⁴ and 2 % < δ₅₃₂ < 7 %, and such a high fluorescence and low depolarization should be attributed to smoke particles. The second cluster consists of points localized inside 0.1 × 10⁻⁴ < G_F < 0.8 × 10⁻⁴ and 1 % < δ₅₃₂ < 3 % intervals, which corresponds to the urban particles in Table 1. After cluster localization, the observations can be plotted as aerosol types, using the parameters in Table 1 and the approach described in Sect. 3.2. The aerosol types in Fig. 5b are spatially separated and contain no high-frequency oscillations. Urban particles are localized at low heights, below 1 km. We would like to remind the reader that, under the condition of high relative humidity (RH), the fluorescence capacity can decrease due to the particle's hygroscopic growth. The water uptake increases the particle backscattering but does not change the fluorescence. As a result, the fluorescence capacity decreases (Veselovskii et al., 2020b). In accordance with radiosonde data, the relative humidity below 1 km was quite high (about 70 % at 500 m) and decreased with height, which can explain the wide range in the G_F variation observed for urban particles in Fig. 5a.

Table 2Intensive particle parameters, such as the lidar ratios (S₃₅₅, S₅₃₂), particle depolarization ratios (δ₃₅₅, δ₅₃₂, δ₁₀₆₄), extinction ($A_{\mathrm{355}/\mathrm{532}}^{\mathit{\alpha }}$), and backscattering ($A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$, $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$) Ångström exponents for six episodes are analyzed in this work. The parameters are given for chosen height–temporal intervals, and the types of aerosol are determined from fluorescence measurements.

Download Print Version | Download XLSX

The particle intensive properties, such as the lidar ratios at 355 nm and 532 nm wavelengths (S₃₅₅, S₅₃₂), the particle depolarization ratios (δ₃₅₅, δ₅₃₂, δ₁₀₆₄), the extinction ($A_{\mathrm{355}/\mathrm{532}}^{\mathit{\alpha }}$), and the backscattering ($A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$, $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$) Ångström exponents for the episodes analyzed in this study, are summarized in Table 2. For this measurement session, in the smoke layer, the lidar ratio at 532 nm significantly exceeds the corresponding value at 355 nm (S₅₃₂ = 80 ± 12 sr and S₃₅₅ = 50 ± 7 sr). The particle depolarization ratio decreases with wavelength from 4.5 % at 355 nm to 2 % at 1064 nm. Such a spectral dependence of the lidar ratio and depolarization ratio for the aged smoke is in agreement with previous studies (e.g., Haarig et al., 2018; Hu et al., 2022, and references therein).

Figure 10Vertical profiles of (a) backscattering coefficient β₅₃₂ and particle depolarization ratios δ₅₃₂ and δ₁₀₆₄. (b) Fluorescence backscattering β_F, fluorescence capacity G_F, and backscattering Ångström exponent $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ on 15 September 2020 for the period 00:00–04:00 UTC.

Download

4.2 30 May 2020: urban vs. pollen

Pollen grains represent a significant fraction of primary biological materials in the troposphere, and a fluorescence-induced emission provides an opportunity for their identification. Figure 6 presents the spatiotemporal variations of β₅₃₂, β_F, δ₅₃₂, and G_F during the pollen season on the night of 30–31 May 2020. The presence of different types of pollen over Lille in spring–summer 2020 was discussed in our recent publication (Veselovskii et al., 2021). In particular, on 30 May 2020, the in situ measurements on the roof of the building demonstrate the presence of a significant amount of grass pollen. The transport of pollen can be analyzed with a global scale to mesoscale dispersion model, SILAM (Sofiev et al., 2015). In the Appendix, we show the maps of the pollen index for four sessions from this study at 22:00 UTC. On 30 May, the pollen index in the Lille region is about 5.0, indicating a high content of pollen.

Figure 11Spatiotemporal distributions of the backscattering coefficient β₅₃₂, the fluorescence backscattering coefficient β_F (in 10⁻⁴ Mm⁻¹ sr⁻¹), the particle depolarization ratio δ₅₃₂, and the fluorescence capacity G_F on the night of 10–11 April 2020. Measurements are performed at an angle of 45^∘ to the horizon. The values of δ₅₃₂ and G_F are shown for β₅₃₂ > 0.2 Mm⁻¹ sr⁻¹.

Download

Figure 12(a) The δ₅₃₂-G_F diagram for observations at 350–1500 m (blue symbols) and 1500–2500 m (pink symbols) height ranges. (b) Spatiotemporal distribution of aerosol types on the night of 10–11 April 2020. The gray coloring shows an undefined aerosol type, which is a mixture of urban and pollen for this case. Measurements with β₅₃₂ < 0.2 Mm⁻¹ sr⁻¹ are marked in black.

Download

The aerosol is located inside the planetary boundary layer (PBL) below 2.5 km. At altitudes below 1 km, the depolarization ratio δ₅₃₂ after 23:00 UTC increases up to ∼ 15 % simultaneously, with an increase in the fluorescence capacity up to 2.0 × 10⁻⁴, which can be an indication of pollen presence. In the δ₅₃₂-G_F diagram in Fig. 7a, the single pixel data points spread from the values typical for the urban particles to the values typical for the pollen. The contribution of pollen to the total backscattering changes with height, and the points form the pattern, similar to a characteristic curve, calculated for urban–pollen mixture in Fig. 3. In accordance with radiosonde data from the Herstmonceux station, the RH at midnight was about 40 % at 500 m, and it increased up to 70 % at 2000 m; thus, the spatiotemporal variations in RH could influence the observed values of the backscattering coefficient and depolarization ratio. In particular, the hygroscopic growth can decrease the values of both δ₅₃₂ andG_F. However, the value of the fluorescence capacity in Fig. 7a changes for almost 1 order of magnitude, and such a strong change in G_F cannot be explained by the particle hygroscopic growth only. For example, from the recent publication of Sicard et al. (2022), an increase of β₅₃₂ in urban aerosol for this range of RH is below a factor of 1.5. Thus, we suppose that the pattern in Fig. 7a is due to the mixing of the urban and pollen particles. The spatiotemporal distribution of aerosol types is shown in Fig. 7b. The urban particles (brown) are predominant, while pollen (yellow) is localized below 1 km in height. The gray color corresponds to an unidentified aerosol type which, in our case, is the mixture of urban particles and pollen.

An indicator of pollen presence in an aerosol mixture, along with a high depolarization ratio, can be a higher value of δ₁₀₆₄ with respect to δ₅₃₂ or δ₃₅₅ (Cao et al., 2010; Veselovskii et al., 2021). Vertical profiles of the particle depolarization ratio at all three wavelengths for this episode are given in Fig. 8c of Veselovskii et al. (2021). At 0.75 km height, where δ₁₀₆₄ is about 15 %, the ratio $\frac{{\mathit{\delta }}_{\mathrm{1064}}}{{\mathit{\delta }}_{\mathrm{532}}}$ is 1.5, which corroborates the suggestions about pollen presence. For urban aerosol, the depolarization spectral ratio $\frac{{\mathit{\delta }}_{\mathrm{1064}}}{{\mathit{\delta }}_{\mathrm{532}}}$ can also be above 1.0 (Burton et al., 2013), but the absolute values of depolarization are significantly lower than for pollen particles (below 10 %).

4.3 14 September 2020: wildfire smoke vs. pollen mixture

Another strong smoke episode occurred on the night of 14–15 September 2020, and the corresponding distributions of β₅₃₂, β_F, δ₅₃₂, and G_F are shown Fig. 8. The elevated smoke layer with a low depolarization ratio (δ₅₃₂ < 5 %) and high fluorescence capacity (up to 4.0 × 10⁻⁴) was observed at approximately 6 km height during the whole night. Inside the boundary layer, the depolarization ratio is higher, up to 15 %, while the fluorescence capacity is lower (about 1.0 × 10⁻⁴) compared to the elevated layer. In the δ₅₃₂-G_F diagram in Fig. 9a, we can see the cluster of data points corresponding to the smoke. At the same time, a part of the points are inside the range of parameters attributed to the pollen (Table 1). The remaining points should be attributed to the mixture of pollen, smoke, and urban aerosol. The distribution of the particle types (Fig. 9b) of this mixture is marked with gray. The pollen particles are localized below 1 km. The presence of pollen over Lille in September is not common, but it can be transported from other regions. The SILAM pollen index in Fig. A1 for this date demonstrates the transport of pollen to northern France from the southeast of France and the eastern Mediterranean.

Figure 10a presents the profiles of δ₅₃₂ and δ₁₀₆₄, together with β₅₃₂, for the temporal interval of 00:00–04:00 UTC. The relative humidity, in accordance with radiosonde data from Herstmonceux station, did not exceed 50 % below 1.7 km. Above that height, RH increased up to 75 % at 2.5 km; thus, the observed increase of β₅₃₂ above 1.5 km can be partly related to RH growth. The relative humidity inside the smoke layer did not exceed 10 %. Similar to Fig. 8, δ₁₀₆₄ exceeds δ₅₃₂ at low heights. The ratio $\frac{{\mathit{\delta }}_{\mathrm{1064}}}{{\mathit{\delta }}_{\mathrm{532}}}$ is about 1.5 at 1 km and inside the smoke layer $\frac{{\mathit{\delta }}_{\mathrm{1064}}}{{\mathit{\delta }}_{\mathrm{532}}}$ 0.4. Higher values of the depolarization ratio at 532 nm compared to 1064 nm are reported for aged smoke by Haarig at al. (2018) and Hu et al. (2019, 2022). The BAE does not present significant height variations because $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ is about 1.0 inside the PBL, and it increases to 1.25 inside the smoke layer (Fig. 10b). Simultaneously, the fluorescence capacity in the smoke layer increases about a factor of 4, compared to the PBL, which demonstrates the efficiency of the fluorescence technique for discriminating the smoke from other aerosol types.

4.4 10 April 2020: urban vs. pollen

At the beginning of April, we experienced several atmospheric situations for which elevated layers were classified as urban aerosols. One of these cases, on the night of 10–11 April 2020, is shown in Fig. 11. Lidar observations were performed at an angle of 45^∘ to the horizon, so the minimum height reachable in the analysis is 350 m. The relative humidity, in accordance with radiosonde data from Herstmonceux station, increased with height from 54 % at 1.0 km to 65 % at 2.2 km. The layer with depolarization ratio δ₅₃₂ below 5 % was observed at about 2 km height during the night. The fluorescence capacity in the layer is low (below 0.5 × 10⁻⁴), so it is identified as urban aerosol. HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory model) backward trajectories (not shown) indicate that the air masses at 750 and 2000 m heights were transported from England (HYSPLIT, 2022). For the period 21:00–23:00 UTC, the depolarization ratio below 500 m increased simultaneously with the fluorescence capacity, which can be an indication of pollen presence.

In the δ₅₃₂–G_F diagram (Fig. 12a), the single pixel measurements at 350–1500 and 1500–2500 m height ranges are shown by different colors. The data points related to the upper layer are within the range of parameters expected for urban aerosol. The points in the lower layer (below 1500 m) are partly out of this range, so the aerosol type for these points is undefined. We assume that this is the mixture of urban and pollen particles because we observe particles with high depolarization (δ₅₃₂ > 15 %) and fluorescence capacity up to 0.7 × 10⁻⁴. This mixture is marked in gray on the aerosol mask in Fig. 12b. The pollen index provided by SILAM over Lille at midnight is above 4.0, so the presence of pollen particles is expected.

Figure 13Vertical profiles of the (a) backscattering coefficient β₅₃₂ and particle depolarization ratios δ₅₃₂ and δ₁₀₆₄. (b) Fluorescence backscattering β_F, fluorescence capacity G_F, and backscattering Ångström exponent $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ on 10 April 2020 for the period 21:00–23:00 UTC.

Download

The presence of pollen is also supported by the profiles of δ₅₃₂ and δ₁₀₆₄, as shown in Fig. 13. At low heights, δ₁₀₆₄ exceeds δ₅₃₂, and the ratio $\frac{{\mathit{\delta }}_{\mathrm{1064}}}{{\mathit{\delta }}_{\mathrm{532}}}$ is about 1.4 at 0.5 km. However, inside the elevated layer, this ratio decreases and becomes about 0.8 at 2.25 km, which indicates that the mixture composition has changed. For the same height range, the fluorescence capacity decreases from 0.6 × 10⁻⁴ to 0.3 × 10⁻⁴, while $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ gradually increases from 0.75 to 1.25, which can be due to a decrease in pollen contribution.

Figure 14Spatiotemporal distributions of the backscattering coefficient β₅₃₂, the fluorescence backscattering coefficient β_F (in 10⁻⁴ Mm⁻¹ sr⁻¹), the particle depolarization ratio δ₅₃₂, and the fluorescence capacity G_F on the night of 11–12 August 2021. The values of δ₅₃₂ and G_F are shown for β₅₃₂ > 0.3 Mm⁻¹ sr⁻¹.

Download

Figure 15(a) The δ₅₃₂-G_F diagram for observations in the 500–1400 m (blue symbols) and 1400–2500 m (pink symbols) height ranges. (b) Spatiotemporal distribution of aerosol types on the night of 11–12 August 2021. Measurements with β₅₃₂ < 0.3 Mm⁻¹ sr⁻¹ are marked in black.

Download

As follows from Table 2, in the lower layer, the values of S₃₅₅ and S₅₃₂ are close (about 48 ± 7 sr). However, in the elevated layer, S₅₃₂ increases to 70 ± 7 sr, while S₃₅₅ remains the same. Higher values of S₅₃₂, with respect to S₃₅₅, are typical for aged smoke (e.g., Müller et al., 2005; Hu et al., 2022). Moreover, $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ significantly exceeds $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\alpha }}$, which was also reported for aged smoke. Thus, based on intensive properties only, we could classify this layer as smoke. However, due to a low fluorescence capacity, in our approach, we identify it as urban.

4.5 11 August 2021: contacting layers of smoke and urban aerosol

Separation of smoke and urban particles is a challenging task for Mie–Raman lidar because both types have a small effective radius and similar depolarization ratios δ₅₃₂. However, the fluorescence capacity of smoke is about a factor of 4–5 higher than that of urban aerosol, which allows their reliable separation. The analyses of the measurements on the night of 11–12 August 2021 are shown in Fig. 14. The RH decreases with height from 70 % to 40 % inside the 500–2250 m range. The main part of aerosol is concentrated below 2500 m, and two height intervals can be distinguished. Above approximately 1500 m, the layer with high fluorescence capacity (up to 3.0 × 10⁻⁴) is observed, while in the layer below 1500 m, the G_F is low (below 0.8 × 10⁻⁴). HYSPLIT backward trajectories (not shown) indicate that the air masses at 1800 m heights were transported from North America, so these may contain wildfire smoke.

In the δ₅₃₂-G_F diagram (Fig. 15a), the single pixel measurements in the 500–1400 and 1400–2500 m height ranges are shown by different colors. The cluster of points, corresponding to the upper layer, is localized mainly inside the interval 1.8 × 10⁻⁴ < G_F < 4.0 × 10⁻⁴ and 4 % < δ₅₃₂ < 10 % and can be attributed to smoke. The points corresponding to the lower layer are partly identified as urban particles, but a part of the points is out of the range and forms a pattern typical for urban–pollen mixture. The SILAM pollen index in Fig. A1 is above 5.0, so the contribution of pollen can be noticeable. The smoke and urban layers are in contact and the particle mixing occurs, which increases dispersion within the clusters.

Vertical profiles of δ₅₃₂ and $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ in Fig. 16 do not demonstrate significant differences for the upper and lower layers. Meanwhile, the fluorescence capacity increases by factor of 4. The lidar ratios S₃₅₅ and S₅₃₂ in the upper layer, as follows from Table 2, are 45 ± 7 and 72 ± 11 sr, respectively. The $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ significantly exceeds $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\alpha }}$ (2.2 ± 0.2 and 1.0 ± 0.2, respectively), so, based on intensive parameters, the upper layer can be also identified as smoke.

Figure 16Vertical profiles of (a) backscattering coefficient β₅₃₂ and particle depolarization ratios δ₅₃₂ and δ₁₀₆₄. (b) Fluorescence backscattering β_F, fluorescence capacity G_F, and backscattering Ångström exponent $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ on 12 August 2021 for the period 00:00–03:30 UTC.

Download

4.6 1 April 2021: dust

Dust layers transported from Africa are regularly observed over northern France. One such dust episode took place on the night of 1–2 April 2021, and the corresponding spatiotemporal variations of β₅₃₂, β_F, δ₅₃₂, and G_F are shown in Fig. 17. The dust layer, with a depolarization ratio exceeding 30 % and low fluorescence, extends from approximately 1.0 to 5.0 km in height. The fluorescence capacity varied inside the layer. In the center, it was the lowest (about 0.1 × 10⁻⁴), but at the bottom of the layer and near the top, G_F increased up to (0.2–0.3) × 10⁻⁴. In Fig. 18a (δ₅₃₂-G_F diagram), we observe a cluster of particles which can be identified as dust. There is also a second small cluster attributed to urban aerosols. In the distribution of particle types in Fig. 18b, the urban aerosol occurs below 800 m after 23:00 UTC.

Figure 17Height–temporal distributions of the backscattering coefficient at 532 nm β₅₃₂, the fluorescence backscattering coefficient β_F (in 10⁻⁴ Mm⁻¹ sr⁻¹), the particle depolarization ratio at 532 nm δ₅₃₂, and the fluorescence capacity G_F on the night of 1–2 April 2021. The values of δ₅₃₂ and G_F are shown for β₅₃₂ > 0.3 Mm⁻¹ sr⁻¹.

Download

Figure 18(a) The δ₅₃₂-G_F diagram for observations in the 500–5000 m height range and (b) spatiotemporal distribution of aerosol types on the night of 1–2 April 2021. Measurements with β₅₃₂ < 0.3 Mm⁻¹ sr⁻¹ are marked in black.

Download

Figure 19 provides vertical profiles of β₅₃₂, δ₅₃₂, δ₃₅₅, β_F, G_F, and $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$. Measurements at 1064 nm were not available for this episode. Depolarization ratios at 355 and 532 nm are close to 30 % through the layer, though, at heights below 1.5 km, there is a small enhancement of δ₅₃₂ up to 34 %. The fluorescence capacity is about 0.4 × 10⁻⁴ at 1.5 km, and it decreases with height to 0.1 × 10⁻⁴ at 2.5 km. However, this decrease is not accompanied by changes in the depolarization ratio. The backscattering Ångström exponent $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ is sensitive to the enhancement of dust absorption in UV and can be negative (Veselovskii et al., 2020a). For this episode, $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ decreases with height (together with G_F) to −0.3 at 2.5 km. Similar values of $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ were observed during SHADOW campaign in the Western Sahara (Veselovskii et al., 2020a). Above 3.75 km, both $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ and G_F start to increase. Hence, dust properties change with height, and this change is not revealed in the δ₅₃₂ profile. We should mention that, in the publication of Veselovskii et al. (2020a), the increase in the dust imaginary part in the UV range also did not lead to changes in δ₅₃₂.

Figure 19Vertical profiles of (a) backscattering coefficient β₅₃₂ and particle depolarization ratios δ₅₃₂ and δ₃₅₅. (b) Fluorescence backscattering β_F, fluorescence capacity G_F, and backscattering Ångström exponent $A_{\mathrm{532}/\mathrm{1064}}^{\mathit{\beta }}$ on 1 April 2021 for the period 19:00–20:40 UTC.

Download

The application of our new fluorescence–depolarization-based approach to the six episodes considered in this section demonstrates its ability to discriminate between several aerosol types. The first step in the validation of the results presented could be a comparison of the particle properties for obtained aerosol types with the corresponding values used in the existing classification algorithms. Table 3 provides the range of the variation in particle intensive properties from publications of Burton et al. (2013), synthetic values used in the NATALI algorithm (Nicolae et al., 2018), and parameters used in the algorithm of Papagiannopoulos et al. (2018) for the urban, smoke, and dust particles. The table contains also the range of properties variations for the episodes considered in the current study for the same aerosol types. The parameters chosen in different algorithms, even for the same aerosol type, vary in a wide range, and the values observed in this study mainly match this range of variation. We observe higher values of $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ for urban and smoke particles, and, for dust, $A_{\mathrm{355}/\mathrm{532}}^{\mathit{\beta }}$ could be negative. Still, the values obtained in this study and the values used by other algorithms are in reasonable agreement.

Table 3Intensive particle parameters from publications of Burton et al. (2013), Nicolae et al. (2018), and Papagiannopoulos et al. (2018), together with values observed in current study for the urban, smoke, and dust particles.

Download Print Version | Download XLSX