In this subsection, we present the inversion results from the cloud radar and the lidar measurements separately, and then combined. Afterwards, we show the results to provide specific examples.
3.2.1. Cloud Radar Inversion Results
Next, the inversion algorithm performance is evaluated by comparing the measured reflectivity and LDR to their estimated values for different scenarios, depending on the LDR pixel values compared to the sensitivity threshold and the maximum LDR value present in the scattering database (referred to in this section as LDRmax).
In
Figure 9, this comparison is displayed for Case B (19 June 2013), where
Original corresponds to the measured values and Outputto the inversion methodology retrieval. The rest of the categories are presented to illustrate the different retrieval scenarios depending on the LDR correction values applied. In
Figure 9a, the estimated reflectivity (Output, blue line) reproduces well the measured radar reflectivity (Original, purple line), suggesting a good performance of our inversion methodology. The yellow line (Output (null LDR)) corresponds to the pixels for which LDR is null or equal to 0. The proportion of pixels for which this condition is met increases with lower reflectivity values due to the instrument sensitivity. The teal line (Output (0 < LDR < LDR
max)) corresponds to the pixels where the LDR is not corrected. The green line (Output (LDR < LDR
max)) shows the estimated results for all the pixels below the maximum value in the scattering database (LDR
max), including those with null LDR values, and it closely follows the original values. Finally, the red line (Output (LDR > LDR
max)) presents the results for the LDR pixel values that were corrected for high values. After adding the green and red lines (Output, blue line), the values are closer to the input than if we just consider the pixels with an LDR value lower than LDR
max (Output (LDR < LDR
max), green line), and therefore we decided to correct the high values of LDR instead of keeping them out of the inversion procedure.
The agreement is also satisfactory regarding the performance of the LDR inversion (
Figure 9b). The values measured by the radar are depicted with the purple line (Original), spanning from −30 to 0 dB. The LDR values over the scattering database maximum LDR value (LDR
max) were, in this case, corrected with the mean LDR of the layer (∼−13 dB). The dark blue line (Corrected) presents the original values with this correction applied. The retrieval (Output, light blue line) is very close to the used input (Corrected, dark blue line). The output for the LDR pixels to which no correction was applied (Output (LDR < LDR
max), green line) is in quite a good agreement with the measured values (Original, purple line) up to almost −10 dB. The retrieval for the corrected pixels (Output (LDR > LDR
max), red line) is centred at the mean LDR of the layer, which is the value that these higher LDR values were corrected with. It becomes clear that our retrieval algorithm underestimates the LDR values, and therefore, the scattering database should be enlarged to include higher LDR values (e.g., considering different particles’ orientation).
The reflectivity and LDR differences between the original and the retrieved values for the same case are presented in
Table 4. The first row refers to all the layer points, including the corrected LDR values, and the difference is −44.5 dB for reflectivity and −43.4 dB for LDR. The lowest differences correspond to the pixels for which the LDR was null, which indicates either that the particles were spherical or that the LDR values were below the sensitivity threshold. The retrieval, therefore, relies mainly on the reflectivity values, which explains the lower difference. The highest differences correspond to the pixels in which the LDR values were not corrected (0 < LDR ≤ LDR
max), which are associated with larger pixel LDR variability. Considering all the pixels with null LDR and the non-corrected (LDR ≤ LDR
max), the discrepancies are very close to those considering the whole layer, which is also the case for the pixels in which the LDR was corrected for being too high (LDR > LDR
max).
The described inversion method was applied to all the cloud radar observations of giant aerosol lofted layers. The performance of the method in terms of difference is presented in
Table 5 for both reflectivity and LDR. The difference averaged for all the cases is
for the reflectivity and
for the LDR. The difference is one order of magnitude higher for the LDR, which can be attributed to the corrections to the pixels that had values measured above the maximum value in the scattering database. The worst performance has discrepancies that are two orders of magnitude higher for both parameters (
and
, respectively). The best performance for the reflectivity has a difference of
and of
for the LDR.
Figure 10 shows the frequency distributions of reflectivity and LDR for all the aerosol layers as measured by the radar (purple line) and as retrieved by our method (blue line). The reflectivity plot (
Figure 10a) shows a fair agreement between the two distributions, where the centers of the distributions (between −40 and −30 dB) coincide. The inversion method, though, generally overestimates reflectivity values over −30 dB and underestimates them below −40 dB. Accordingly, the lower discrepancies from
Table 5 correspond to layers with mean reflectivity values in the −40–−30 dB range, and the highest ones to layers at the edges of the distribution. The reason for these discrepancies is probably linked to the LDR pixels’ correction.
In the LDR plot (
Figure 10b), the effect of the correction of pixel values over the maximum LDR value in the scattering database is obvious: the frequency’s distribution is shifted towards lower LDR values for the retrieval. This impacts the retrieval, and especially the axis ratio.
The reflectivity and LDR differences between the original and the retrieved values that we have presented for Case B (
Figure 9 and in
Table 4) are quite similar to the mean situation when considering all the giant aerosol layers (
Table 5).
The effective radius resulting from the inversion for all the layers (
Figure 11a) ranges between 1.5 and 7 µm and has a bimodal distribution. The first maximum is around 3 µm and the second between 4 and 5 µm. The corresponding size parameter is between
and
, in agreement with the Rayleigh regime assumption embedded in the retrieval method. The number concentration (
Figure 11b) is found to be between 0.10 and 1 cm
−3, with maximum values in the region 0.4–0.8 cm
−3. The mean geometric radius of the aerosol distribution (
Figure 11c) extends between 1 and 12 µm, and the maximum radii are found in the range 1–3 µm. The frequency decreases with aerosol size, which suggests that the smallest aerosols reach the observational site more frequently compared to bigger ones, as one would expect according to their respective settling velocities.
The aerosol particle sizes and number concentration results retrieved by our algorithm are in line with the observational measures by Exton et al. [
60], who found total particle concentrations of about 0.1–0.5
in the 5–150 µm range, even though in the frequency distribution, the portion of particles over 5 µm is low. They also fall in the range of the measurements by Lasher-Trapp and Stachnik [
52], who detected giant to ultragiant aerosol particles spanning approximately from 5 µm to 50 µm in diameter (2.5–25 µm in radius) in concentrations between 10
−3 and 10
by using airborne probes. Instead, the retrieved number concentrations are higher than those measured during the First Aerosol Characterization Experiment (ACE-1), in which researchers measured particles with dry radii in the order of 6 to 12 µm in concentrations between 10
−4 and 10
−2 cm
−3 [
12].
The axis ratio (horizontal/vertical axis) frequency distribution (
Figure 11d) indicates that more than 90% of the aerosol particles are prolate, with axis ratios between 0.7 and 0.8. A small portion of the aerosols are oblate, with axis ratios around 2, and almost none are spherical. These findings are consistent with our hypothesis of giant particles, which mainly comprise dust, pollen, and volcanic materials; hence, there are irregular shaped aerosol types [
61,
62,
63]. Considering that our retrieval algorithm underestimates LDR values, though, it is likely that the aerosol particles’ irregularity is underestimated.
The refractive index depends on the aerosols’ composition and size distribution and on the incident wavelength. The real part accounts for refraction, while the imaginary part handles the attenuation. The real part of the refractive index (
Figure 11e) oscillates, mainly, between 2.43 and 2.47, and for a few cases, it is between 2.62 and 2.65. The imaginary part of the refractive index (
Figure 11f) fluctuates around 0.4, and for a few cases, it is between 0.57 and 0.6. This suggests, that, in general, the observed aerosol particles are less absorbing than the volcanic particles studied by Adams et al. [
51]. Given the observational site’s location and climatology, this is in line with what we would expect, as volcanic aerosols are generally more absorbing with respect to desert dust, which leads to higher SSA. The values of SSA depend on the aerosol’s composition, size, and wavelength. Reported values at 440 nm, for example, are
for volcanic particles [
64] and
for dust particles [
65]. The estimation of the refractive index of aerosols is important given that it has an impact on the SSA and, therefore, can impact the radiative budget.
3.2.2. Lidar Inversion Results
The inversion code by Veselovskii et al. [
47] uses three aerosol backscatter coefficients and two aerosol extinction coefficients as input, and therefore it can be only applied to the night-time measurements, when the aerosol particle light extinction coefficient (hereinafter extinction) profiles are available (they cannot be retrieved in an independent way during day-time because the Raman backscattered signal cannot be measured accurately due to solar background radiation). All the input optical properties were available for 11 out of the 25 night-time measurements, for which we performed the inversion. Not all the optical properties were available for all the cases, which may be due to low SNR values or to quality assurance of the estimated properties.
When we classify the cases according to the cloud radar observed target, five correspond to aerosol and six to insect observations. We decided to perform the inversion for all the lidar cases independently of the cloud radar-observed target in order to increase the number of cases for which we could retrieve the microphysical properties and to infer if there are any differences depending on the presence or not of giant aerosols in the atmosphere.
Figure 12 reports the number, surface, and volume distribution for each individual case (a, b and c respectively) and the mean distribution of all the cases according to the target observed by the radar (d, e and f). The number distribution describes how many particles exist per unit volume of air for each particle size, and it is dominated by small particles. The surface distribution represents the total surface area of particles per unit volume of air for each size, and it typically peaks at intermediate sizes. The volume distribution describes the total volume of particles per unit volume of air for each size, and it is usually dominated by larger particles (coarse mode).
In
Figure 12a, the number distribution for each case shows that most of the particles fall in the size range 0.06–1 µm. The number concentration reaches up to 670 cm
−3 but for most cases is lower than 150 cm
−3. In
Figure 12b, the surface distributions reveal the existence of a minor mode in the region 1–6 µm. This indicates that even in low number concentration, some coarse particles are detected by the lidar. The surface concentration is, in general, lower than 50 µm
2 cm
−3 but in some cases reaches up to 100 µm
2 cm
−3. The volume concentration in
Figure 12c shows that not all the distributions are bimodal. Some distributions are trimodal, and their corresponding radii ranges are from 0.1 to 0.3 µm, between 0.3 and 1 µm, and from 1 to 6 µm radius. No differences in the distributions of all the individual cases could be observed depending on the detection of aerosols or insects (hereinafter target) by the cloud radar.
To infer whether differences exist in the number, surface area, and volume distributions when aerosols or insects are detected by the cloud radar, the distributions were averaged based on the corresponding radar targets.
Figure 12d shows the number distribution for all the lidar aerosol cases averaged depending on the cloud radar-detected target. In this case, the distributions are slightly different: the aerosol radii when insects are observed by the cloud radar (between 0.06 and 0.13 µm) is lower than when aerosols are also observed (between 0.09 and 0.14 µm). The number concentration in both cases, though, is similar.
Figure 12e presents the surface distribution in the same way. For the cloud radar insect observations, two modes are observed, the second from 0.3 to 1 µm, while for radar aerosol observations, there is one main mode between 0.10 and 1 µm. The surface concentration is slightly higher for the aerosol cases.
Figure 12f reports the volume distribution, which has three modes both for cloud radar aerosols and insects. In this case, the distributions are relatively closer than before, and the volume concentration is also quite similar.
In conclusion, in the studied cases, it was possible to observe particles with lidar with radii between 0.1 and 6 µm, with mean radii being below 1 µm. To investigate whether the lidar retrievals capture a coarse aerosol fraction closer to the larger particles detected by the cloud radar, the lidar retrievals were grouped according to the radar target classification. This analysis revealed some differences in the aerosol size distributions (number, surface, and volume). For instance, the aerosol mean radius was larger in cases where the cloud radar detected giant aerosol particles. This aligns with expectations, suggesting that lidar measurements can retrieve the microphysical properties of the smallest and likely more numerous fraction of giant aerosols. However, due to the limited number of cases available for this analysis, these findings should be viewed as indicative rather than definitive.
Case B: 19 June 2013
As we described previously, smoke aerosol particles were observed on 19 June 2013.
Figure 13 shows the number, surface, and volume distributions retrieved with the lidar and radar inversions both together and separately. The inversions were performed for the 2–2.5 km a.s.l. layer measured by the lidar and for the 3.2–4 km a.s.l. layer measured by the radar, both from 19:27 to 19:57 UTC. The number distribution (
Figure 13a) shows that the number of particles below 0.4 µm is predominant, whilst the number concentration of particles with radii over 1 µm is negligible. In the surface distribution graph (
Figure 13b), particles below 0.4 µm still predominate, although coarser particles generate three secondary maxima. The first and second secondary maxima are seen by the lidar (∼1 µm and ∼3 µm) and the third by the radar (∼7 µm). A partial overlap between the third lidar mode and the radar mode is present. The volume distribution (
Figure 13c) shows a similar behaviour to that of surface distribution, with the first lidar mode being less intense and the radar mode more intense and centred at ∼9 µm instead of at ∼7 µm. In this case, the partial overlap between the third lidar mode and the radar mode is more evident, between 2 and 6 µm approximately. This indicates that the lidar detected a small fraction of the coarser particles seen by radar.
This case is an example of how to obtain an enlarged aerosol size distribution by combining Raman lidar and cloud radar information. The Raman lidar can be used to retrieve the size distribution for aerosols in the Aitken and in the accumulation mode, as well as for a small part of the coarse mode. The cloud radar can be used to retrieve the size distribution for coarse-mode aerosols.
The mean microphysical properties retrieved from the radar and lidar measurements for this case can be found in
Table 6. The effective and mean radius are larger for the aerosols detected by the radar, and the number concentration is much higher for the lidar-detected aerosol particles. The complex refractive index is different for both instruments, and the different input values for the two inversion methods bias the retrieved mean values, as this parameter depends both on the particle type and the wavelength. The axis ratio for the radar shows that the particles were aspherical (specifically prolate).
The discrepancies in aerosol microphysical properties retrieved from radar and lidar measurements are substantial and can be primarily attributed to the differing operating wavelengths and sensitivities of the two instruments. In terms of particle size, these differences arise from the relative scale of the aerosol particle size compared to each instrument’s wavelength; radar typically detects larger aerosols, whereas lidar is more sensitive to smaller ones. Consequently, the number concentration retrieved by lidar tends to be higher, which is consistent with its sensitivity to smaller, more numerous particles. Finally, the observed differences in complex refractive index can be linked to its wavelength dependence and to variations in aerosol particle composition [
66].