The IASBS scanning polarization lidar (ISPL) is a single-wavelength (532 nm) linear polarization lidar that is designed and constructed at the Institute for Advanced Studies in Basic Sciences (IASBS) remote sensing laboratory. The ISPL is installed on the southwest coast of Urmia Lake, at 37.34° N, 45.29° E, and 1270 m above mean sea level (a.m.s.l.) (see Fig. ). The system operates in two modes: zenith-aiming or horizontal azimuthal scanning . For this study, horizontally scanned data were used. The ISPL transmitter consists of a frequency-doubled, 10 ns pulsed Nd-YAG laser and a 10× beam expander. The output energy per pulse at 532 nm is approximately 100 mJ at a repetition rate of 2 Hz. The full overlap range of the ISPL is at a distance of ∼3.0 km from the lidar. On the receiver side, a Cassegrain telescope with an 8-inch primary mirror and focal length of 1950 mm collects the backscattered light. Signals at 532 nm are recorded in both parallel and perpendicular polarization channels relative to the transmitted pulse. Data acquisition is performed using a four-channel, 350 MHz Tektronix digital oscilloscope. The system performs azimuthal scanning from φ=-50 to +50°, where φ=0° corresponds to the east direction in Fig. . Azimuthal scans performed in steps of 0.1° of the arc . Further details on the ISPL optical setup can be found in . Each complete azimuthal scan takes ∼45 min. To minimize background noise, all lidar measurements are conducted during the night (see Fig. ).

Using polarization lidar recordings, key aerosol parameters can be derived, including the backscattering coefficient and the linear depolarization ratio. These parameters are essential for particle classification . The well-known Klett-Fernald algorithm has been used to retrieve the particle backscatter coefficient, βp. To apply this technique, a reference extinction coefficient (αref) is required for each recorded lidar signal. The slope method is commonly used to determine αref when the lidar is operating horizontally . But when aerosol plumes are present within the lidar operating range, especially when they spread over the sampling region, the technique incurs considerable errors due to difficulty distinguishing the plume from the background in the lidar signals. modified the slope method by dividing the lidar signal into multi-sections and investigating the statistics of the slopes of fitted lines to each section. As a result, they found an αref that, to a good extent, is not affected by noise and aerosol emissions. They called the technique the multi-section method (MSM), and we used it to retrieve a proper αref for each recorded azimuthal lidar signal. The obtained αref is fed into the Klett-Fernald algorithm to obtain βp for each lidar signal. In retrieving the βp, the aerosol lidar ratio is taken as 32 sr, which is the average value of that of dust, salt-dust, and wet salt (see Table ). Once βp is determined, the particle linear depolarization ratio can be calculated from the method introduced by . 1δp=βm(δv-δm)+βpδv(1+δm)βm(δm-δv)+βp(1+δm), where the subscripts p and m refer to particle and molecular components, respectively. δv is the volume linear depolarization ratio (i.e., the combined depolarization of molecules and particles), and the molecular depolarization ratio is taken as δm=0.0036. Even though δm is a theoretical value, considering 1 nm bandwidth of the 532 nm interference filter (IF) in the ISPL setup and the operation temperature of the lidar (∼20 °C), it is very close with δm≃0.0042 reported by . Such a small difference in δm affects the retrieved δp (for the ISPL horizontal scans), in orders of 1.0×10-3, which is too small to affect the other retrieved parameters.

It should be mentioned that the lidar polarization channels are calibrated by the technique introduced by . There is a subtle point on the impacts of different optical elements of the ISPL on the polarization state of the passing lights through them and eventually their effects on the measured volume depolarization. The beam expander in the transmitter part of the ISPL should be investigated for such effects. , in a valuable work, presented the details of such effects on the measured volume depolarization in a lidar system. We used the framework introduced by and simulated the passage of a linearly polarized 532 nm laser light through the 10× ISPL beam expander in Zemax. In result, we found that the total amount of depolarization of the light after its passage through the beam expander was just ∼2.3×10-5%. Such small values of depolarization can be neglected quite easily, and we may neglect the depolarization effects of the beam expander.

Now βp and δp can be fed into the POLIPHON algorithm to classify the particles .

introduced the two-step POLIPHON method for classifying atmospheric particles into coarse dust, fine dust, and non-dust particles. Since the ISPL lidar is installed just at the lake's coast and scans the atmosphere azimuthally a few tens of meters above the lake bed, it is reasonable to consider that the atmospheric aerosols are either dust (d), salt-dust (sd), or wet-salt (ws). To classify atmospheric particles into these three types, the POLIPHON method should be modified. Figure  is a schematic diagram of the modified two-step POLIPHON algorithm for categorization of d, sd, and ws particles. In this algorithm, δ and β denote the linear depolarization ratio and the backscattering coefficient, respectively, while the subscript p refers to particles. βp is obtained from the Klett-Fernald algorithm and δp driven by the method introduced by . The POLIPHON algorithm is sensitive to the presumed boundaries on particle linear depolarization. As Fig.  shows, in the first step of the algorithm, when δp rises beyond 0.31 , particles are considered as pure dust. Knowing that the depolarization ratio of dry salt particles is ∼0.15±0.03 , and for dusty mix or polluted dust, δp is mainly greater than 0.12 . Then, on the particle depolarization line, δws+sd,e=0.12 is defined as a boundary between the mixture of ws + sd particles and the region that includes all particle types (Fig. ). Outputs of the first step of the algorithm are βd and βws+sd. After that, the algorithm goes to its second step. In this step, first, the depolarization ratio of the mixture (δws+sd) is retrieved. Then, it is considered if δws+sd≤0.05, particles are just wet-salt (Fig. ), otherwise they are a mixture of wet-salt and salt-dust particles, and their backscatter coefficients (βws and βsd) are obtained from the last stage of step 2 (Fig. ).

Having β values for different types of particles, their mass concentrations, MCi, can be obtained from Eq. () . 2MCi=ρicv,iβiSi, where i denotes the particle type (i.e., d, sd, and ws). The parameters ρi, cv,i, and Si represent the particle density, the extinction-to-volume conversion factor, and the lidar ratio of the corresponding particle type, respectively . The values of these parameters are presented in Table . It should be mentioned that we put the parameters ρi and cv,i of the wet-salt particles similar to the marine particles in Table . On the other hand, since the composition of salt-dust particles isn't well known, we considered that ρi, cv,i, and Si parameters for these particles should be something between those of wet-salt and dust particles.

Since we are making measurements just above the lake, we would like to know about the contributions of different types of particles inside a detected plume. Therefore, a plume detection algorithm should be applied to recordings of the lidar azimuthal scans. Several algorithms have been proposed for cloud- or aerosol-layer detection . introduced a multiscale detection (MSD) algorithm to identify cloud and aerosol layers. Their method was originally designed to detect cloud boundaries where the lidar measures in the zenith direction. The method also works for dense aerosol layers. Here, we applied this method to the ISPL azimuthal scans to detect lofted plumes over the Urmia Lake dried bed. Figure  is a schematic range-corrected lidar signal (RCS) which also includes an aerosol layer (the red region).

In the MSD algorithm, the layer has three characteristic points: base (B), peak (P), and top (T). The layer is divided into two parts: base-to-peak region (BPR) and peak-to-top region (PTR). Once the BPR and PTR are identified, the rest of the recorded data can be masked out, allowing one to deal only with the aerosol layer. To identify the aerosol layer, four steps should be followed in the MSD algorithm: (1) determining the trend index function, (2) multiscale detection, (3) overdetection rejection, and (4) base, peak, and top detection. In the first two steps, BPRs are identified using the slope sign of the range-corrected signal. The slope sign at different points of the RCS is determined by sliding windows of various sizes along the signal. Larger windows characterize the overall behavior of the signal, but smaller ones are needed to obtain information about finer features of the signal. In aerosol-free regions dominated by Rayleigh scattering, the RCS slope is generally negative. A positive slope indicates the presence of an aerosol layer and corresponds to a BPR, where the first bin can be taken as the base and the last bin as the peak of the detected layer. The third step removes spurious detections: if the difference between base and peak in the raw lidar signal is below a pre-assumed threshold, the BPR is discarded. The threshold is assumed as six times the standard deviation of the signal's far-end (tail) region. In the final step, the top of each layer is determined as the first range beyond the peak where the RCS value falls below that of the base. The range between the base and top is called the base-to-top region (BTR). After computing the BTR, the aerosol layer can be masked from its surroundings. For each detected plume in all azimuthal lidar recording profiles, we construct a plume mask. Applying these masks to the lidar signal profiles isolates the aerosol plumes, enabling targeted analysis of plume characteristics. Details of the MSD algorithm appear in the work by .

Figure a is an azimuthal scan of the lidar recorded RCS's on 20 September 2022, 07:43–08:28 local time (LT; UTC +03:30). The lidar Range-Angle (RA) series covers up to a distance of ∼15 km and an angular span of 100°. The Met Office reported mean temperature and surface wind speed are 22°C and 2ms-1, respectively, with a relative humidity of 27% and no cloud coverage. It should be mentioned that September is one of the driest months of the year in the Urmia Lake region (Fig. ). The presence of an aerosol plume in Fig. a is evident, mainly indicated by brownish-red and yellow colors. MODIS imagery together with the Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) back-trajectory analysis reject any possibility for trans-regional aerosol plume transport into the area (not shown here).

For the classification analysis of the aerosols included in the plume, the signals between the two red dashed lines in Fig. a were averaged and further examined. The obtained results are presented in Fig. b–d. Figure b displays the volume (v) and particle (p) linear depolarization ratios, and the particle backscatter coefficient (βp), which is obtained from the Klett-Fernald Algorithm, is depicted in Fig. c for the selected region. Two peaks on δp (∼0.30–0.35) within the first ∼5 km, followed by a decrease (δp∼0.10) between ∼5 and ∼10 km, indicate the possible presence of dust and salt particles, respectively. Feeding the obtained βp and δp into the modified 2-step POLIPHON (Figs.  and ), the backscatter coefficients for dust, salt-dust, and wet-salt can be retrieved (Fig. c). Putting the obtained βd, βsd, βws and their corresponding parameters of the three particle types from Table  into Eq. (), the mass concentrations of particles, MC_d, MC_sd, and MC_ws will be in reach (Fig. d). Figure c and d show a dust layer that starts at a distance of ∼1 km, extends up to ∼7 km from the station. Meanwhile, the second part of the plume, which started at a range of ∼4.5 km and spread up to ∼10 km, contains both salt-dust and wet-salt particles. Though βsd and βws are almost the same over the second part of the plume (Fig. c), their mass concentrations are quite different; MC_sd reaches up to ∼70 µg m⁻³, but the maximum value of MC_ws is ∼30 µg m⁻³. The mass concentration of dust particles is quite high at ∼1.4 km from the station, which MC_d is getting close to 95 µg m⁻³.

The same classification procedure was applied to all scanning directions. The detected aerosol plumes were then selected by using the MSD algorithm and removing the plume-free regions from the RA series of the lidar RCSs. Figure a shows the obtained mask when the MSD has been applied to the lidar signals, and Fig. b–f are the corresponding spatial profiles of the optical (δp and βp) and physical (MC_d, MC_sd, and MC_ws) parameters of the plume. Figure b shows that particles in higher ranges (≳4 km) have a higher backscattering with respect to those close to the lidar station (ranges ≲5 km) that show higher values of δp (Fig. c). Corresponding mass concentrations of different types of particles in Fig. d–f indicate that dust particles are mostly close to the station (Fig. d), but salt-dust and wet-salt particles mostly are at ranges beyond ∼4 km. Comparison of Fig. e and f indicate that salt-dust particles are dominant in the plume with respect to the wet-salt ones. Taking the MC_is averages over the plume leads to MC‾d=22±27, MC‾sd=43±26, and MC‾ws=13±11 µg m⁻³. Based on the obtained MC_is, the contributions of the three types of aerosols in the plume are 28%, 55%, and 17% for d, sd, and ws particles, respectively. Such average values are calculated for 64 plume-included cases (see Sect. ).

During the measurement campaign, 119 scans were performed. Among them, 64 cases contained aerosol plumes. As mentioned in the Sect. , the physical and optical properties of atmospheric aerosols for each recorded scan that contained a plume are retrieved using the modified 2-step POLIPHON and Eq. (). Then the MSD algorithm has been applied to all of the 64 cases to isolate the aerosol plume from the background and the obtained parameters: δp, βi, and MC_i are averaged over the plume to find δ‾p, β‾i, and MC‾i. These plume-averaged parameters are depicted in Fig. .

As Fig. a shows, δ‾p values for all detected aerosol plumes are higher than 0.10. Therefore, all of the plumes should contain dust or salt-dust particles. Some plumes, like those in cases: No. 1, 18, 26, and 27 contained considerable amounts of pure dust particles which δ‾p is higher than 0.28. Referring to Fig. b and c, β‾sd, β‾ws, MC‾sd, and MC‾ws are quite small for such cases. On the other hand, for cases like No. 17 and 29, where the plumes contained considerable amounts of salt particles (salt-dust and wet-salt), δ‾p decreased to ∼0.12, β‾sd, β‾ws, MC‾sd, and MC‾ws considerably increased, in contrast, β‾d and MC‾d are quite low. Figure b and c also show how the salt content of aerosols influences the backscatter coefficient and eventually the albedo of the plume. For example in case 17, MC‾ws≃ 40 µg m⁻³ and β‾ws≃2.5Mm-1sr-1 but for case 39, when MC‾d≃82 µg m⁻³ the β‾d is only ∼1.0Mm-1sr-1. This is in complete agreement with the lower lidar ratios obtained for salt particles (∼25 sr) compared to dust particles (Table ).

We also calculated the plume averaged extinction coefficients (α‾i), using the retrieved β‾i and Si values in Table , then to obtain an overall view on the contribution of different types of particles to the formation of aerosol plumes over the lake, averaged on β‾i, α‾i and MC‾i across all 64 cases and for all types of pre-assumed particles, including: dust (d), salt-dust (sd), wet-salt (ws), and a combination of salt-dust and wet-salt (sd+ws). The results are presented in Table , where the 〈〉 denotes the average over all 64 cases. Table  shows that even though the average values of particle backscatter coefficient for mixtures of wet-salt and salt-dust particles, 〈β‾sd+ws〉 is almost twice of 〈β‾d〉, their averaged mass concentration, 〈MC‾sd+ws〉, is ∼2/3 of 〈MC‾d〉. In other words, the salt particles made the main contribution to the formation of aerosol plumes over the lake, but their lower share in the mass concentration is mainly due to the difference in the mass densities and lidar ratios of the salt and dust particles (Table ). Table  also shows that salt particles mostly appear as salt-dust aerosols. To categorize aerosol plumes, we define three new parameters, σd and σsd+ws as the standard deviations of 〈β‾d〉 and 〈β‾sd+ws〉, respectively, and their sum, σ=σd+σsd+ws. Then we categorize the plumes into three modes: dust-dominant, salt-dominant, and mixed mode, as Table . The number of cases observed in each mode and the probability distribution function (pdf) of each mode are shown in Table  columns 3 and 4, respectively. It shows that 47% of the 64 recorded cases were salt-dominant, while 25% of them were dust-dominant.

Solid colored circles in Fig.  are the plume-averaged mass concentrations (MC‾i) of different types of aerosols (d-dark red, sd-green, ws-blue, and sd+ws-orange), as functions of the plume-averaged particle linear depolarization ratio (δ‾p) and the backscattering coefficients (β‾i) for each recorded event. Figure  shows clearly that as MC‾d increases, δ‾p increases too, but δ‾p has a reverse relation with variations of salt particle mass concentrations (sd, ws, and sd+ws). Interestingly, salt particles have a significantly greater influence on particle backscatter coefficients. For instance, Fig.  shows the backscattering coefficient for MC‾ws≃40 µg m⁻³ is about 2.5 Mm-1sr-1 while similar mass concentrations of dust particles have β‾d<0.5 Mm-1sr-1.

The black and red solid lines are fits to variations of β‾sd+ws+σ4 and β‾d+σ4 vs δ‾p, respectively. As Table  indicates, these two lines are used to categorize the aerosol plumes. This categorization somehow determines borders on δ‾p for each category of the aerosol plumes. Two black vertical dashed lines in Fig.  show these borders. Using these borders, it can be found that the average values of δ‾p for the salt-dominant, dust-dominant, and mixed mode aerosol plumes are 0.16±0.02, 0.26±0.02, and 0.22±0.01, respectively.