In this study, the aerosol fluorescence backscatter coefficient and fluorescence capacity were retrieved using the method described by Veselovskii et al. (2020). The method relies on the fluorescence signal PF and the nitrogen Raman signal PR, which are described by the following equations: 1PFz=CFβFzTL(z)TF(z)2PRz=CRβRzTL(z)TR(z), where C represents the calibration constants, β the backscatter coefficients and T the atmospheric transmission terms. The subscripts F, R and L denote fluorescence, nitrogen Raman and elastic scattering, respectively.

The ratio of the Eq. (1) to the Eq. (2) is used to obtain the aerosol fluorescence backscatter coefficient βF(z): 3βF(z)=CRCFPFzPRzNR(z)σRTR(z)TF(z), where σR=2.744×10-30 cm² sr⁻¹ (Venable et al., 2011) is the vibrational-rotational Raman differential backscattering cross section of molecular nitrogen at 355 nm and the number density of nitrogen is given by the relation: 4NR(z)=cN2Nmol(z)=0.78Nmol, where cN2 is the volume fraction of nitrogen in air and Nmol is the molecular number density of air for a given temperature and pressure, considering the compressibility of air. The ratio TR(z)/TF(z) of the atmospheric transmissions at the Raman and fluorescence wavelengths is calculated from the extinction coefficients at the corresponding wavelengths applying the related extinction Ångström exponent correlation, EAE355/532 (Ansmann et al., 1990). The calibration constants CF and CR are range-independent system constants accounting for the overall optical and electronic response of the fluorescence and Raman detection channels, respectively. The calibration factors ratio CR/CF therefore represents the relative calibration factor between the Raman and fluorescence channels; it corrects for their different instrumental efficiencies and is calculated following the calibration procedure described by Veselovskii et al. (2020) and analysed by Gast et al. (2025). This procedure is based on the ratio of the electronic gains of the PMTs and considers each component in the respective detection path of the two channels, as described above, thus the calibration factor is given by the relationship: 5CRCF=gRgFTIF,RTIF,F. The electronic gain (g) was obtained by exchanging the PMTs of the two channels without changing the IF and calculating the ratio of the mean signals measured by both PMTs for the fluorescence or the Raman channel. The test produced a PMT gain ratio (gR/gF) of 1.63. Considering the transmission of the different IFs of the channels (TIF,R/TIF,F∼0.75) one obtains a value of ∼1.59 for CR/CF, the ratio of the calibration factors.

Once the fluorescence backscatter coefficient is derived, the fluorescence capacity GF can be calculated using the ratio of the fluorescence backscatter to the aerosol backscatter coefficient at 532 nm retrieved with the Raman backscatter method (Ferrare et al., 1998): 6GF=βFβ532. Although fluorescence is excited at 355 nm and it would therefore be preferable to use the aerosol backscatter coefficient at the same wavelength (β355), the elastic aerosol backscatter coefficient at 532 nm is used here in order to enable comparison with existing literature.

For improved intercomparison with other fluorescence lidar configurations, particularly lidar systems employing broadband fluorescence channels with varying filter widths or spectrometers, we convert the retrieved fluorescence capacity to a spectral quantity. The spectral fluorescence capacity GF355 is calculated according to Gast et al. (2025) as 7GF355=βFβ355dIF, with dIF representing the effective bandwidth of the fluorescence IF. Both spectral fluorescence capacity (GF355) and fluorescence capacity (GF) are presented in the results below to allow comparison with the existing literature. In the text, GF is given in parentheses.

In lidar instruments that deploy broadband fluorescence channels, the characteristics of the IFs that are being used, such as the central wavelength and the FWHM, are selected according to the fluorescence emission spectrum of the fluorescing particles of interest. The particle fluorescence spectrum depends both on the excitation laser line (here 355 nm) and on the particle chemical composition and mixing state. For an excitation with the 355 (more precisely 354.71 nm) line, an IF with central wavelength between 466 to 470 nm and bandwidths ranging up to 100 nm has been preferred for broadband fluorescence applications. Although not necessarily located at the maximum of the fluorescence spectrum, the spectral region 415–525 nm avoids contamination of other major Nd:Yag-related and Raman lines such as the water vapor Raman line at 408 nm and the 2nd Nd:Yag laser harmonic at 532 nm and its Anti-Stokes purely rotational Raman band. In THELISYS, an IF centered at 469.55 nm (approximately 470 nm) with 100 nm FWHM was initially deployed to isolate part of the fluorescence spectrum. However, differences in the IF transmission hindered direct comparison with fluorescence measurements performed in Lille (Veselovskii et al., 2021) and Leipzig (Gast et al., 2025). In addition, Reichardt et al. (2023) and Veselovskii et al. (2023) reported that the 424.4 nm second-order vibrational Raman line of N₂ can introduce biases to the fluorescence spectrum (Reichardt et al., 2023). To improve the consistency of THELISYS lidar with the lidars in Lille and Leipzig, the IF at 470 nm was replaced by an IF with transmission centered at 465.73 nm (approximately 466 nm) and with 44 nm FWHM. We will refer to those filters as IF_470,100 and IF_466,44 respectively.

The mean GF355 calculated using such a broad IF will differ from the actual GF355 at the IF center wavelengths. Using the GF355 spectrum measured by Reichardt et al. (2025) as a reference (Fig. 1) we can estimate the systematic relative bias σGF,IF466 and σGF,IF470 of GF,466355 and GF,470355 which are introduced when using IF_466,44 and IF_470,100, respectively, as 8σGF,IFλ0=GF,λIF355‾-GF,λ0355GF,λ0355, where GF,λ0355 the spectral fluorescence capacity at λ0=466 or 470 nm and GF,λIF355‾ is the spectral fluorescence weighted mean signal. The latter is calculated using the filter transmission as weighting function: 9GF,λIF355‾=∫GF355(λ)×TIF(λ)dλ∫TIF(λ)dλ.

The values are summarized in Table 2. The lower and upper systematic relative biases correspond to the most negative and positive bias calculated from each spectrum of Fig. 1 (including both spectra from biomass burning aerosol and dust). This ensures that the error bars are conservative and cover the whole uncertainty range of the dataset displayed in Fig. 1. We see that IF_470,100 leads to generally larger biases, even at its central wavelength, with a tendency to underestimate the actual GF355 value. We further compare the biased GF,470355‾ and GF,466355‾ values which occur after transmission through IF₄₇₀ and IF₄₆₆, respectively. The relative difference between GF,470355‾ and GF,466355‾with respect to GF,466355‾ is defined as 10σΔIF=GF,470355‾-GF,466355‾GF,466355‾‾, and can range from -5.8 % to 0.6 % for the different spectra of Fig. 1.

In order to compare the spectral fluorescence capacity considering the characteristics of IF_466,44 and IF_470,100 we use Eq. (7). The IF comparison test was performed on seven separate days in 2025. Here, we present two representative cases: one dominated by dust aerosols and another characterized by urban and smoke aerosols, used to compare the two IFs. Table 3 lists the dates and times of each measurement.

Across the GF355, small differences are observed when comparing the IF_466,44 and the IF_470,100 over Thessaloniki (Fig. 2). For instance, on 13 March 2025, an extreme dust event passed over Thessaloniki, providing an opportunity to evaluate the performance of the two IFs (Fig. 2a). The aerosol layer extended from 0.65 to 3.2 km.

Applying Eq. (7), we calculated GF355 using IF_466,44 and GF355 using IF_470,100, as shown in Table 4. These results indicate that the GF355 retrieved using the two different IFs is comparable and consistent with recent literature (Reichardt et al., 2025) and that Eq. (7) provides a useful approach for comparing GF355 values obtained between different lidar systems. As introduced by Reichardt et al. (2025), GF355 varies with aerosol type, therefore, a second case characterized by urban and smoke aerosols is presented in Fig. 2b and the calculated GF355 are again shown in Table 4.

Here, the observed differences between GF,466355 and GF,470355 (Fig. 2b, c) are much bigger than the systematic error estimation due to the IF transmission band especially for shallow layers, and also according to the error bars the noise is too low to explain the differences, which leads us to the conclusion that the variations were probably related to atmospheric changes between the two measurements as they were not performed simultaneously or iteratively.

These findings highlight the importance of standardizing spectral fluorescence capacity within EARLINET to enhance the comparability of fluorescence measurements and to improve aerosol characterization and typing schemes.

The complete THELISYS dataset for the period between April 2024 and December 2025 is analysed as the cases above. From a total of 50 measurements, 108 layers are identified by the WCT method. All layers are located above the full-overlap region (i.e., 0.35 km), extending to altitudes exceeding 10 km, indicating the presence of both boundary-layer aerosols and elevated layers in the free troposphere.

Figure 7a illustrates the individual layers, with colours representing their corresponding GF355 values. The observed GF355 values span several orders of magnitude, from approximately 10-7 to 10-5. As shown, higher values (GF355>3×10-6 nm⁻¹) are observed during the summer months at mid to high altitudes, particularly between 3 and 8 km, suggesting the presence of strongly fluorescent aerosols, commonly associated with biomass-burning smoke transported over long distances. Throughout the year, GF355 varies around 1×10-6 nm⁻¹ for aerosol layers residing mainly at low altitudes (<4 km), while lower values (GF355<1×10-6 nm⁻¹) are observed for some specific dust events, totally clear days.

The classification of all layers was primarily based on the spectral fluorescence capacity and the particle depolarization ratio as shown in Table 8. For cases where the particle type was not clearly identified from these two parameters, the lidar ratio and backscatter Ångström exponent were also considered. In addition, the layer altitude was considered, since at different heights certain aerosols may exhibit similar spectral fluorescence capacity. For example, layers with biomass burning aerosols are typically observed above 4 km (Michailidis et al., 2024), whereas transported pollen layers are observed below 2.5 km (Damialis et al., 2017). Furthermore, the maximum layer value of the spectral fluorescence capacity was used to discriminate among clean continental (0.6×10-6 nm-1<GF,max⁡355<1.8×10-6 nm⁻¹), polluted continental (1.4×10-6 nm-1<GF,max⁡355<3.6×10-6 nm⁻¹) and mixed pollen (2.3×10-6 nm-1<GF,max⁡355<6.4×10-6 nm⁻¹) aerosol. The presence of smoke and dust is further confirmed by the HYSPLIT trajectory model, other forecasting models, and satellite observations, as discussed in the two corresponding cases in Sect. 3.1 and 3.2. Table 8 summarizes the thresholds considered for the characterization of the different aerosol layers.

Based on the above criteria, Fig. 7b shows the classified aerosol types; smoke (green), dust (orange), continental (brown), clean continental (red), mixed pollen (blue). Overall, increased GF355 values are observed during the period from March to October, indicating enhanced concentrations of fluorescent aerosol components. This seasonal increase is attributed to the presence of aerosol mixtures influenced by both pollen emissions and biomass-burning aerosols. Especially, during the summer months (i.e., June–August), higher values in GF355 are evident, consistent with the frequent occurrence of smoke aerosols originating from regional wildfires in Greece or the Balkans as well as long-range transported biomass-burning plumes from North America (e.g., Canada), Russia, Eastern Europe (Amiridis et al., 2009). These smoke plumes are detected not only within the boundary layer but also at higher altitudes.

Based on the analysis of 50 cases of fluorescence lidar measurements obtained by THELISYS, Fig. 8 and Table 9, summarize the aerosol typing framework adopted in this study that relies on the combined use of spectral fluorescence capacity, with IF center wavelengths of either 466 or 470 nm, and the particle linear depolarization ratio at 532 nm. Dust particles, including mixed dust in this category, are characterized by low spectral fluorescence capacity and high depolarization ratios, which is consistent with their non-spherical mineral nature. Smoke aerosols exhibit significantly enhanced spectral fluorescence capacity combined with low depolarization values, reflecting the presence of biomass-burning particles at high altitudes during the summer months. Pollen particles exhibit a fairly high spectral fluorescence capacity with relatively low depolarization, supporting their identification primarily at lower to mid-tropospheric altitudes during the pollen season. It is important to note that their characterization is complicated by the fact that pollen particles frequently occur as part of mixed aerosol layers, often combined with urban, smoke or other aerosol types. Moreover, the particle linear depolarization ratio of pollen strongly depends on the dominant pollen taxa, which varies with geographic region and season, leading to additional variability in the observed optical properties. Continental aerosols show lower spectral fluorescence capacity and depolarization ratios compared to smoke and dust, indicating weaklier fluorescent and small particles, which are typically confined to the lower atmospheric layers.

The results obtained from the analysed THELISYS measurement cases (colored boxes) show some differences from the values reported by Miri et al. (2025) (empty boxes) in Fig. 8b. For dust and smoke, consistent typing is observed, taking into account that the measurements were performed at different locations and during different periods. For urban aerosol, higher fluorescence capacity values are observed in the THELISYS measurements, indicating a contribution from fluorescent particles, that may be related to the influence of locally emitted smoke. More pronounced differences are observed for the mixed pollen layers, particularly in the particle depolarization ratio. As discussed in Sect. 3.3, these differences are likely related to the mixing of pollen with urban aerosols observed at the Thessaloniki lidar station or to differences in the dominant pollen taxa. Pollen grains from different taxa can exhibit different sizes, shapes, and morphological characteristics, which may consequently affect their optical properties, and particularly the particle depolarization ratio. It should also be noted that the proposed typing scheme may be influenced by particle hygroscopic growth, particularly under conditions of high relative humidity, as water uptake can modify particle size and consequently their optical properties, including the elastic backscatter coefficient and particle depolarization ratio (Miri et al., 2024).

The THELISYS lidar sub-dataset covering March 2024 to April 2025 was analysed for aerosol typing using both the EMD and the fluorescence capacity classification. For the 16 cases, 31 layers are identified by the WCT and all of them are above 350 m, as previously. EMD classified 29 layers (93 %), and 2 layers (7 %) were not assigned to any cluster showing an overall higher identification rate. Figure 9 presents the vertical distribution and temporal evolution of the aerosol layer types detected over Thessaloniki area. Across this period, layers are observed from near the surface up to approximately 8 km, indicating the presence of both boundary layer aerosols and elevated free-tropospheric layers. The dominant aerosol categories include dust and mixed/polluted dust (orange and dark orange), smoke (green), pollen (blue), clean continental (red), polluted continental (brown), and a small number of unclassified or unknown layers (grey), as inferred from the application of both EMD and fluorescence based aerosol typing.

From this dataset, it is evident that altitude layers below 2 km are dominated by clean and polluted continental aerosols, reflecting local and regional sources, while higher altitude layers are more frequently classified as dust, mixed dust, or smoke, indicating long-range transport processes. The highest altitude layers, occasionally extending beyond 8–9 km, are predominantly classified as polluted continental or smoke, especially in the EMD classification.

Figure 10 illustrates the overall differences in aerosol typing within the THELISYS dataset that emerge from the comparison of the two classification approaches. We considered the EMD output as the reference in our study because EMD has previously been applied in EARLINET dataset and been validated as a reliable method for typing (Voudouri et al., 2019; Mylonaki et al., 2021). Thus, each row is normalized by the total number of components identified by the EMD algorithm. Dust and mixed-dust layers (summarized as dust in Fig. 10) are consistently detected by both methods (75 % agreement), as the particle depolarization ratio plays a critical role in the separation of dust aerosols. A similarly high agreement is observed for smoke aerosols, with all EMD classified smoke layers assigned to the smoke fluorescence category. However, the number of smoke cases in the dataset is limited. For polluted continental aerosols, a partial agreement (53.3 %) is found between the two methods. The EMD classification overlaps with multiple fluorescence categories (e.g. smoke 20 %) suggesting that the spectral fluorescence capacity is a very useful parameter especially for the separation between smoke and polluted continental aerosols, as smoke is a highly fluorescing aerosol compared to others. Clean continental aerosols exhibit a lower agreement (50 %), with difficulty in separation of polluted and clean continental aerosols. This separation based on their fluorescence capacity values is at an initial stage of analysis and remains under investigation. The same applies to pollen detection. In the present dataset, only two mixed pollen layers were identified based on high fluorescence capacity values. Of these, one was classified as polluted continental by the EMD method, while the other remained unclassified. Mixed pollen layers are also under investigation and represent a promising opportunity for further development of aerosol typing algorithms.

We conclude that aerosol typing using fluorescence channel capabilities, in combination with existing aerosol properties (e.g., particle depolarization ratio, Ångström exponent, and lidar ratio), could help in clearer separation between dust, smoke and mixed pollen layers. Overall, the comparison makes their combined use valuable for a more accurate characterization of aerosol vertical structure and source attribution.