The following subsections discuss the datasets implemented in the framework of the study to facilitate the realization of its overarching objectives. More specifically, Sect. 2.1.1 provides an overview of the CALIPSO–CALIOP mission and products, the cornerstone of the near-global fine-mode and coarse-mode pure-dust climate data record. In addition, towards establishing the accuracy of the products and consistency checks, the study utilizes ISS-CATS (International Space Station, Cloud–Aerosol Transport System) optical products (Sect. 2.1.2) and AERONET retrievals (Sect. 2.1.3).

The present work aims to decouple the fine-mode (particles with diameter less than 1 µm) and coarse-mode (particles with diameter greater than 1 µm) components of atmospheric pure dust, which is in turn is a component of the total aerosol mixture, at a near-global scale. The decoupling methodology follows the series of well-established polarization-based algorithms for decoupling the atmospheric pure-dust component from the total aerosol load, initially established by Shimizu et al. (2004) and accordingly expanded through the family of the POLIPHON algorithms. More specifically, the study is based on the one-step POLIPHON (Tesche et al., 2009) methodology for decoupling the pure-dust component from the total aerosol load (Shimizu et al., 2004) and accordingly on the conceptual approach of the two-step POLIPHON (Mamouri and Ansmann, 2014) for extracting the coarse-mode pure-dust component from the total aerosol load. Finally, in the framework of the present study, the submicrometer (fine-mode) component of pure dust is extracted as the residual between the total pure dust (one-step POLIPHON) and the supermicrometer (coarse-mode) component of pure dust (first step of the two-step POLIPHON). The retrieval scheme is applicable to single-wavelength lidar observations, as long as profiling of calibrated linear-polarization is included. The methodology is applied to CALIPSO backscatter coefficient and particulate depolarization ratio profiles at 532 nm (Sect. 2.1.1), with the overarching objective to provide the fine-mode and coarse-mode pure-dust atmospheric components at near-global scale for the temporal period extending between June 2006 and December 2021. Section 2.2.1 presents the decoupling methodology of the fine-mode and coarse-mode components of pure dust in terms of backscatter coefficient at 532 nm, extinction coefficient at 532 nm, and mass concentration, while Sect. 2.2.3 discusses the uncertainties of the established products.

This section aims to evaluate the CALIPSO-based DODcoarse and DODfine at 532 nm products through the extensive implementation of AERONET AOTcoarse and AOTfine retrievals at 532 nm, taking into consideration the unique characteristics of CALIOP and sun-photometer measurements and products, the quality assurance criteria, and the synchronization and collocation requirements. More specifically, CALIOP L2 reports aerosol and cloud measurements and products in near-vertical “sheets” of near-zero swath (∼ 100 m footprint) on the Earth's surface along the CALIPSO orbit track with 5 km horizontal resolution. On the contrary, AERONET instruments are characterized by an approximately 1.2° full-angle field of view, resulting in columnar pencil-like multiwavelength measurements of aerosol optical thickness between the solar disc and the sun–sky photometers (Holben et al., 1998; Omar et al., 2009). Consequently, given the natural atmospheric inhomogeneity of the aerosol fields in both time and space, CALIOP and AERONET rarely probe the same air volumes of the atmosphere. Hence, to extensively compare CALIOP and AERONET columnar observations, implementation of a set of constraints and criteria is a prerequisite to ensure a robust comparison and thus to establish the quality of the CALIPSO-based DODcoarse and DODfine at 532 nm products against AERONET AOTcoarse and AOTfine retrievals at 532 nm.

One of the major factors driving comparison discrepancies in the correlative observational dataset relates to the high spatial and temporal inhomogeneity of the aerosol fields. In this study, we adopt the CALIPSO–AERONET collocation criteria established in satellite-based lidar studies, though the collocation criteria vary from study to study depending on the scope and requirements (Amiridis et al., 2013; Anderson et al., 2003; Omar et al., 2013; Pappalardo et al., 2010; Proestakis et al., 2019; Schuster et al., 2012). The spatial and temporal collocation criteria applied in the framework of the present study require CALIPSO maximum overpass distance less than 80 km from the AERONET site and AERONET AOD acquisition measurements within ±60 min of the CALIPSO overpass. The selection of the spatial matching distance of CALIPSO L2 profiles within an 80 km radius of the AERONET sites is justified in terms of the CALIPSO feature detection algorithms. The CALIOP L2 Selective Iterated BoundarY Locator (SIBYL) processing module detects atmospheric features with L1 profile horizontal averaging of 5, 20, and 80 km along the CALIPSO orbit track (Vaughan et al., 2009). Thus, the maximum 80 km along-track averaging resolution in the detection of atmospheric layers performed by SIBYL prohibits observations of the same atmospheric air masses by CALIOP and AERONET photometers, even in cases of the CALIPSO orbit track directly over the AERONET stations (Schuster et al., 2012). Consideration of CALIPSO distance closer than 80 km, although reported to result in slight improvement in CALIOP–AERONET absolute biases, would be performed at the expense of larger errors due to reduction of the sample size of synchronized measurements. Accordingly, the temporal collocation criterion between CALIOP and AERONET observations of 60 min (Δt ≤ 60 min) is justified in terms of the mesoscale natural variability of aerosol in the lower troposphere (Anderson et al., 2003; Pappalardo et al., 2010).

CALIPSO AODs at 532 nm are computed as vertical integration of the mean CALIOP extinction coefficient at 532 nm profile with respect to height between TOA and the elevation height of AERONET sun photometers, thus not considering atmospheric aerosol fields possibly accumulating below the AERONET sites (Amiridis et al., 2013; Schuster et al., 2012). Regarding the negative effects of clouds in the comparison cases, the use of CALIOP L2 profiles of cloud optical depth equal to zero (0) and atmospheric features of CAD score lower than -20 ensures the lowest possible contamination of CALIPSO AODs by cloud features. It must be noted that the CALIPSO–AERONET overpasses may result in non-representative CALIPSO AODs in cases of extensive presence of clouds due to reduction of the considered CALIPSO L2 profiles over the area (Sect. 2.1.1). To minimize the negative impact of clouds in the analysis, cases with extensive presence of clouds detected in CALIPSO L2 measurements are not considered in the comparison by applying a lower acceptable threshold of at least eight CALIPSO L2 5 km cloud-free profiles (∼ 80 km) within the 80 km distance from the AERONET stations. Accordingly, at least two AERONET L2 AOD measurements are required to have been performed in the CALIPSO overpass ±60 min temporal window for accepting an AERONET case.

The evaluation of the CALIPSO-based DODcoarse and DODfine products against AERONET AOTcoarse and AOTfine products should be performed in cases characterized by dust presence. The AERONET Ångström exponent (AE) cannot be used because the implementation of low AE to establish correlative CALIPSO–AERONET cases dominated by larger particles would drive the comparison towards excluding from the analysis submicrometer-dominated pure-dust cases. The CALIPSO aerosol subtype cannot be used either as an appropriate dust identifier, since it provides only qualitative information on the presence of dust. More specifically, implementation of CALIPSO aerosol subtyping as a proxy for high dust presence would result in underestimations of CALIPSO DODcoarse (DODfine) extracted from layers classified as dusty marine (polluted dust) upon comparison against AERONET AOTcoarse (AOTfine), since the retrievals would be based on columnar measurements of both dust and marine (smoke) aerosols. Finally, the CALIOP particulate depolarization ratio at 532 nm cannot be used as a proxy for the identification of dust cases, since this optical parameter is implemented in the process of extracting the CALIPSO-based fine-mode and coarse-mode pure-dust components, and thus the parameter does not constitute an independent identifier. Thus, to facilitate the evaluation of the CALIPSO-based DODcoarse and DODfine with AERONET AOTcoarse and AOTfine, only for cases of dust presence in the atmosphere, we implement the pure-dust product developed in the framework of the ESA-LIVAS activity (Amiridis et al., 2013, 2015) under the quantitative assumption that at least 50 % of the observed AERONET AOT is related to the presence of dust, as provided by the LIVAS DOD at 532 nm product (|Rel.Dif.CALIOP_DOD,  AERONET_AOT| ≤ 50 %). Finally, the comparison between the CALIPSO-based DODcoarse and DODfine at 532 nm products and AERONET AOTcoarse and AOTfine at 532 nm retrievals is performed only for the cases of both CALIPSO AOD at 532 nm and AERONET AOT at 532 nm at least equal to 0.01.

Figure 5 provides an overview of the comparison for the case of a Saharan dust outbreak reaching Dakar on 20 April 2017 (Fig. 5a). FLEXPART v10.4 (FLEXible PARTicle) Lagrangian dispersion model (Stohl et al., 2005; Pisso et al., 2019) 6 d air masses back trajectories at 2 km in the area of interest (lat: 14.39°, long: -16.95°) are performed. The initial and boundary conditions for the FLEXPART runs are produced with 3-hourly meteorological data from the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) provided at 0.5° × 0.5° spatial resolution and 41 model pressure levels. FLEXPART has been used in a large number of similar studies on long-range atmospheric transport (Stohl et al., 2005; Solomos et al., 2019; Kampouri et al., 2021). According to the back trajectories, the advected air masses were of north-northwestern Saharan origin (Fig. 5a – yellow line). According to the CALIPSO overpass (Fig. 5a – red line) in the vicinity of the AERONET Dakar sun photometer (Fig. 5c), the observed aerosol layer extended vertically between 1 and 4 km a.m.s.l. and was dominated by the presence of Saharan dust, as corroborated by both the cross-section of backscatter coefficient at 532 nm (Fig. 5b) and the mean particulate depolarization ratio at 532 nm profile (Fig. 5e). Figure 5d provides the pure-dust coarse-mode (red shaded area) and fine-mode (blue shaded area) components, following decoupling from the total aerosol load, including the non-dust aerosol component (gray shaded area) in terms of profiles of backscatter coefficient at 532 nm. Columnar-integrated pure-dust coarse-mode (red shaded area) and fine-mode (blue shaded area) extinction coefficient at 532 nm profiles yield CALIPSO-based DODcoarse and DODfine at 532 nm equal to 0.29 ± 0.21 and 0.11 ± 0.1, respectively, and optical depth values in good agreement with AERONET Dakar AOTcoarse and AOTfine of 0.36 and 0.12, respectively (Fig. 5f).

Enforcing the collocation criteria and constraints described above (Table 5) to account for the spatiotemporal atmospheric homogeneity and sampling differences between the two systems yields 1737 CALIPSO–AERONET coincidences fulfilling the requirements over the globe for a 14-year period (June 2006–May 2019). The CALIPSO–AERONET coincidences, extracted under the condition of dust presence in the atmosphere, mainly populate the arid areas of the globe and downwind of the source region areas. In this section we provide an assessment of the derived CALIPSO-based fine-mode and coarse-mode DODs at 532 nm against the corresponding AERONET fine-mode and coarse-mode AOTs at 532 nm, retrieved as discussed in Sect. 2.2.2–2.2.3 and as presented above, for CALIPSO–AERONET in Dakar on 20 April 2017 (Fig. 5).

Figure 6 evaluates the total dataset of correlative CALIPSO-based fine-mode and coarse-mode DODs and AERONET-based fine-mode and coarse-mode AOTs in the form of 2D histogram density scatterplots with a bin step equal to 0.01, while the overall evaluation intercomparison metrics are summarized in Table 6. Data pairs delineated by the black one-to-one dashed line correspond to perfect optical depth agreement between the CALIPSO-based fine and coarse DOD modes and the respective AERONET fine and coarse AOT modes. Linear regression (solid black line) for the submicrometer category reveals good agreement between the two datasets, with slope 0.652, offset 0.018, and Pearson's correlation coefficient of the order of 0.692, though there is a tendency to deviate from the one-to-one line with increasing aerosol load (i.e., underestimation). With respect to the coarse-mode category the agreement between the two datasets improves substantially, with linear fit of slope 0.779, intercept close to -0.002, and Pearson's correlation coefficient equal to 0.916. An apparent feature revealed through the evaluation of the CALIPSO-based fine-mode and coarse-mode DODs at 532 nm products against the AERONET retrievals in fine-mode and coarse-mode AOTs at 532 nm is the increase of the degree of dispersion with increasing AERONET AOT values (Fig. 6). More specifically, the general tendency is for data pairs to be closer to the AERONET axis, resulting in an overall CALIPSO fine and coarse DOD underestimation compared to the respective AERONET fine and coarse AOT modes.

Figure 7 quantitatively provides the CALIPSO-based fine-mode and coarse-mode DOD at 532 nm absolute biases (Fig. 7a, b), relative biases (Fig. 7c, d), and root mean square errors (RMSEs) (Fig. 7e, f) compared to the AERONET fine-mode and coarse-mode AOT for the 532 nm class of equal increment steps of 0.05 for the AOT at 532 nm range extending between 0 and 0.5. Overall, with respect to the CALIPSO-based fine-mode DOD at 532 nm, the evaluation reveals negative mean absolute biases (-0.012), mean relative biases (-7.38 %), and RMSE (0.059). Overall, with respect to the CALIPSO-based coarse-mode DOD at 532 nm, the evaluation reveals negative mean absolute biases (-0.036), positive mean relative biases (9.55 %), and RMSE (0.076). With respect to the CALIPSO-based coarse-mode DOD at 532 nm it must be noted though that the mean relative bias is driven positive by the observed patterns in the range of small DOD values (0–0.05), while for EARLINET AOT classes larger than 0.05 negative relative biases are observed (Fig. 7d). Moreover, the analysis shows that the CALIPSO-based fine-mode and coarse-mode DOD absolute biases (Fig. 7a, b) are more clearly affected for larger AOD values (increasing), while mean relative absolute biases demonstrate significantly less variability, remaining relatively constant (Fig. 7c, d).

It must be noted that the revealed CALIPSO fine- and coarse-mode DOD at 532 nm underestimation constitutes an expected feature. More specifically, it is well-documented that a discrepancy – underestimation – exists in CALIPSO extinction coefficient profiles and AODs at 532 nm, as established against correlative observations by Aqua-MODIS AODs (e.g., Kacenelenbogen et al., 2011; Kittaka et al., 2011; Redemann et al., 2012; Kim et al., 2013; Ma et al., 2013), AERONET-derived AOTs (Schuster et al., 2012; Omar et al., 2013; Amiridis et al., 2013; Toth et al., 2018), and airborne lidar observations (Rogers et al., 2014). These studies attribute the apparent CALIPSO AOD at 532 nm underestimation to a number of factors, including – among others – misclassification of cloud layers as aerosol (and vice versa), erroneous subclassification of the classified atmospheric layers as aerosol, incorrect selection of the aerosol subtype lidar ratios, limited or restricted penetration and/or attenuation of the CALIOP beam within thick aerosol layers, and profiles frequently populated with retrieval fill values (RFVs) due to failure to detect diffuse and tenuous aerosol layers of SNR below CALIOP minimum detection thresholds. The CALIPSO fine- and coarse-mode DOD at 532 nm underestimation with respect to AERONET AOT at 532 nm is further enhanced by the fine-mode, coarse-mode, and total pure-dust decoupling algorithm (Sect. 2.2.1). The apparent underestimation features relate to an extent to the evaluation of DODfine with AOTf, to an AERONET retrieval including non-dust fine-mode aerosol subtypes (e.g., biomass burning and urban haze), and to the evaluation of DODcoarse with AOTc, an AERONET product including non-dust coarse-mode aerosol subtypes (e.g., marine, pollen, volcanic ash).

Overall, the CALIPSO-based DOD–AERONET AOT evaluation comparison corroborates on the good performance of the lidar-based algorithms developed with the objective of decoupling the fine-mode, coarse-mode, and total pure-dust components of the total aerosol load (Shimizu et al., 2004; Mamouri and Ansmann, 2014, 2017; Tesche et al., 2009), the high quality of the already established ESA-LIVAS pure-dust database that is the cornerstone of the present study (Amiridis et al., 2013, 2015; Marinou et al., 2017; Proestakis et al., 2018), and the quality of the established CALIPSO-based products of fine-mode and coarse-mode pure-dust atmospheric components in terms of extinction coefficient profiles and DODs at 532 nm.

In August 2015 a large-scale collaborative field campaign, the “Ice in Clouds Experiment – Dust” (ICE–D), including a subcomponent, the “AERosol properties – Dust” (AER–D) experiment, was conducted over the eastern tropical Atlantic Ocean in the broader region extending between Cabo Verde and the Canary Islands. The main science objective of the AER–D campaign evolved around a better characterization of airborne mineral dust optical and microphysical properties (Marenco et al., 2018), while ICE–D aimed at studying dust–cloud interactions and the efficiency of airborne mineral dust to act as CCN and IN (Liu et al., 2018). In the framework of AER-D and ICE-D 16 research flights were carried out with the Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 research aircraft. Research flight b920 conducted on 7 August 2015 aimed to perform highly spatially and temporarily coordinated measurements with CATS on board the ISS, targeting the thorough validation of CATS products (McGill et al., 2015; Pauly et al., 2019; Yorks et al., 2016).

AER–D and ICE–D FAAM flights deployed remote sensing and in situ instrumentation. Remote sensing measurements were conducted, amongst others, by a Leosphere ALS450 elastic backscatter lidar system, designed to acquire vertical profiles of aerosol and clouds at 355 nm in a nadir-viewing geometry. The system specifications are provided by Marenco et al. (2011) and Chazette et al. (2012), and the processing algorithms, starting from the level of the lidar beam returns to the determination of the lidar ratio, the derivation of the extinction coefficient profiles, and the corresponding AOD values, are described in Marenco (2013), Marenco et al. (2014), and O'Sullivan et al. (2020). In situ measurements are described in Ryder et al. (2018) and include aerosol size distribution measurements using a combination of a wing-mounted PCASP (passive cavity aerosol spectrometer probe; Rosenberg et al., 2012), a CDP (cloud droplet probe; Lance et al., 2010), and a 2DS (two-dimensional stereo probe), sampling a combined size distribution from 0.1 to 100 µm in diameter. Here we use airborne lidar measurements carried out during straight FAAM flights performed at a constant altitude between two locations (“runs” – R) and in situ aerosol size distribution measurements from aircraft ascents and descents (“profiles” – P).

Figure 8a shows the flight track of b920 (black line), demonstrating the both spatially and temporarily highly coordinated flight trajectory to the ISS orbit track. The flight performed both airborne lidar measurements and air mass sampling at different altitudes. In addition, Fig. 8a provides the structure of the atmospheric aerosol scene in the broader area extending between Cabo Verde and the Canary Islands on 7 August 2015, as probed by ISS-CATS and based on L2 5 km backscatter coefficient at 1064 nm profiles (17:44–17:50 UTC). With respect to the vertical structure of the aerosol layers, the profiles of extinction coefficient obtained by the airborne lidar at 355 nm and by CATS at 1064 nm, in terms extinction coefficient, are in good agreement. Both extinction coefficient profiles report values in the range between 0.03 and 0.08 km-1, with FAAM lidar values well within the variability of the atmospheric scene as observed by ISS-CATS. The aerosol observed by both the airborne and spaceborne lidar systems was dominated by dust, extending between 2 and 5 km a.m.s.l. in the Saharan Air Layer (SAL), originating from convective activity over central Algeria, while a mixture of dust and marine and broken clouds was present within the marine boundary layer (MBL). The CATS mean particulate depolarization ratio at 1064 nm lies in the range between 0.3 and 0.4, confirming through the highly depolarizing optical properties of the non-spherical dust particles the presence of dust aerosols in the atmosphere (Fig. 8c). Thus, it is evident that the FAAM b920 and ISS-CATS collocated measurements were performed in the presence of a deep dust layer within the Saharan Air Layer (SAL), transported westwards by the predominantly easterly winds. The FAAM flight on 7 August 2015 therefore provides an ideal case for establishing the quality of the satellite-based lidar fine-mode and coarse-mode pure-dust mass concentration products due to (1) the implementation of both remote sensing and in situ airborne measurements, (2) the highly collocated FAAM underflight of the ISS orbit track, and (3) the significant presence of atmospheric dust within the SAL.

Towards the accurate derivation of the fine-mode and coarse-mode pure-dust mass concentration profiles, based on the CATS backscatter coefficient and particulate depolarization ratio at 1064 nm, the implementation of suitable decoupling parameters (Eqs. 5 and 7) and conversion factors (Eqs. 9 and 10) at the wavelength of 1064 nm is required. Towards this objective, the pure-dust δd, coarse-mode pure-dust δcd, and non-dust δnd particulate depolarization ratio values at 1064 nm are set equal to 0.27, 0.28, and 0.05, respectively (Freudenthaler et al., 2009; Haarig et al., 2017b; Mamouri and Ansmann, 2017). CATS L2 5 km quality-assured profiles between latitudes 14.61 and 17.92° N, following the latitudinal geographical coverage of the FAAM b920 flight, were used. Accordingly, height profiles of the pure-dust and coarse-mode pure-dust extinction coefficient at 1064 nm are obtained (Eq. 9a–c) by multiplying the pure-dust and coarse-mode pure-dust backscatter coefficient profiles at 1064 nm with a dust lidar ratio of 67 sr (Mamouri and Ansmann, 2017). To convert the obtained profiles of pure-dust and coarse-mode pure-dust extinction coefficient at 1064 nm into mass concentration profiles, conversion factors cv,d and cv,dc equal to 0.73 and 0.72 are utilized, as calculated for Cabo Verde and Barbados in Mamouri and Ansmann (2017). As a final step, the profiles of fine-mode pure-dust mass concentration are estimated as the residual between the profiles of total pure-dust mass concentration and the coarse-mode pure-dust mass concentration (Eq. 10a–c). Accordingly, the airborne in situ mean aerosol size distribution in terms of mass concentration for aerosol particles with diameter between 0.1 and 100 µm is averaged based on P2 and P7 of the FAAM b920 flight, selected on the basis of the vertical extent, covering the full range of the SAL.

Here, the fine-mode and coarse-mode pure-dust mass concentration profiles obtained based on the ISS-CATS backscatter coefficient and particulate depolarization ratio at 1064 nm profiles through the separation technique (Sect. 2.2) are validated against the FAAM b920 airborne in situ full aerosol size distribution (Fig. 9). From the airborne in situ measurements it is evident that the dust layer is dominated by particles of supermicrometer diameter, consisting of ∼ 99.73 % by mass (Fig. 9a). Moreover, it is evident that CATS total pure-dust mass concentration and the airborne in situ mass concentration profiles are in good agreement, both in the vertical extent and quantitatively, with values extending between 150 and 170 µg m-3 within the SAL and reaching as high as 250-300 µg m-3 within the MBL (Fig. 9b). The comparison is allowed even though the airborne in situ measurements consist of both dust and non-dust aerosols due to the dominant dust presence, as is confirmed by the CATS particulate depolarization ratio at 1064 nm observations (Fig. 8c). The high agreement between the CATS total pure-dust mass concentration and the airborne in situ mass concentration profiles is crucial, since discrepancies would not allow further intercomparison between the airborne fine-mode and coarse-mode profiles of mass concentration and the corresponding satellite-derived fine-mode pure-dust and coarse-mode pure-dust profiles of mass concentration (Fig. 8c–d).

The CATS-estimated coarse-mode pure-dust concentration (163.3 ± 31.8 µg m-3; averaged over the altitude range 1.5 to 5 km a.m.s.l.) and the in situ supermicrometer mass concentration (149.4 ± 55.2 µg m-3) are within 10 % (Fig. 9c), and very close to the total in situ mass concentration measured (154.8 ± 57.1 µg m-3). However, the CATS fine-mode pure-dust concentration is underestimated (2.3 ± 0.4 µg m-3; averaged over the altitude range 0 to 1.5 km a.m.s.l.) at ∼ 58 % lower than the airborne in situ fine-mode mass concentration (5.4 ± 2.1 µg m-3) in SAL, although the performance increases significantly in the MBL region (Fig. 9d). According to Ryder et al. (2018) and the provided size-resolved composition in both the SAL and the MBL for b920, no dust at sizes d < 0.5 µm was present in the atmosphere between Cabo Verde and the Canary Islands on 7 August 2015. However, the fine-mode and coarse-mode dust decoupling algorithm is established on the basis of a boundary diameter separating the two modes of 1 µm; thus, the low dust fine-mode concentrations apparent in both the SAL and the MBL for b920 most probably correspond to dust of diameter between 0.5 and 1 µm. Overall, CATS fine-mode pure-dust mass concentration underestimation may be partially attributed to the natural atmospheric inhomogeneity of the aerosol fields and partially to CATS deficiency in detecting tenuous aerosol layers during sunlight illumination conditions due to a low signal-to-noise ratio (CATS 7.2 daytime minimum detectable backscatter 1064 nm: 1.30 × 10-3 ± 0.24 × 10-3 km-1 sr-1 for cirrus clouds; Yorks et al., 2016), introducing negative biases of the order of 20 %–30 % for daytime observations (Proestakis et al., 2019). However, the overall good performance of the fine-mode and coarse-mode pure-dust decoupling methodology is corroborated by the fine-mode to total mass concentration fractions, being in good agreement and of the same order of magnitude (CATS: ∼ 1.4 %, b920 in situ: ∼ 3.5 %). Finally, both the spaceborne-lidar-derived and the airborne fine and coarse modes are in good agreement with respect to the vertical extent of the probed dust layer, characterized by similar profile characteristics in the SAL and MBL.

With respect to the horizontal distribution of aerosol, the optical depth is the parameter most frequently used by spaceborne passive sensors to quantify the columnar aerosol load under cloud-free sky conditions. Since CALIOP is not a passive sensor but an active system, in order to process the L2 extinction coefficient profiles into optical depth, specific steps have to be taken. More specifically, initially and for the needs of the present study, a near-global uniform grid of 1° latitude by 1° longitude is established (between 70° S and 70° N). Accordingly, the L3 processing algorithm iterates through all the CALIPSO L2 overpass granules within each grid and for the period between June 2006 and December 2021. As a next step, temporal averaging is accomplished by averaging all the grid-aggregated quality-screened L2 extinction coefficient at 532 nm profiles within each grid of spatial resolution 1° × 1° over a time period of interest (e.g., of annual, seasonal, or monthly temporal resolution). Finally, the mean AOD, DOD, DODfine-mode, and DODcoarse-mode at 532 nm are computed by vertical integration of the gridded mean total aerosol, pure-dust, fine-mode pure-dust, and coarse-mode pure-dust extinction coefficient at 532 nm profiles, respectively. Figure 10 demonstrates the outcomes of these processing steps, more specifically the near-global annual mean horizontal distribution of the CALIPSO-based AOD at 532 nm at 1° × 1° spatial resolution (Fig. 10a), the ESA-LIVAS DOD at 532 nm product (Amiridis et al., 2013, 2015; Marinou et al., 2017; Proestakis et al., 2018; Fig. 10b), and the coarse-mode (Fig. 10c) and fine-mode (Fig. 10d) components of DOD at 532 nm. Moreover, Fig. 10 provides the coarse-mode pure-dust fraction (CMF; coarse-mode pure-dust optical depth to total pure-dust optical depth fraction; Fig. 10e) and the fine-mode pure-dust fraction (FMF; fine-mode pure-dust optical depth to total pure-dust optical depth fraction; Fig. 10f) for areas of annual mean DOD at 532 nm greater than 0.01.

With respect to the optical depth outcomes, we begin with an examination of the near-global long-term horizontal distribution of CALIOP AOD at 532 nm to initially provide the generic characteristics of the net effect of dust and non-dust aerosol subtypes in the atmosphere (Fig. 10a). According to the observed AOD features, significant spatial variability is evident, especially over land and the Northern Hemisphere, where major sources of natural aerosol and major sources of anthropogenic aerosol are present (Winker et al., 2013). More specifically, mineral dust particles emitted from the vast arid and semi-arid regions of the planet (i.e., Saharan, Arabian, Taklimakan, Gobi deserts), biomass burning aerosols from tropical grasslands and forests (i.e., Amazonian region, sub-Sahel, Gulf of Guinea), volcanic emissions, sea salt marine particles, and aerosol emissions related to anthropogenic activity, especially over major urban, densely populated, and heavily industrialized areas of the planet (i.e., East Asia, Indo-Gangetic Plain), result in high annual mean AOD values frequently exceeding 0.6, apparent not only over the source regions, but also over distances of thousands of kilometers downwind.

The effect of dust aerosol on optical depth, decoupled and isolated from the effect of all non-dust aerosol subtypes, is provided by the well-established ESA-LIVAS pure-dust product (Fig. 10b). Overall, the pure-dust product of the ESA-LIVAS database has verified and demonstrated capacity to fully reveal the sources of dust, to quantitatively describe the three-dimensional evolution of dust in the atmosphere from source to sink, and to efficiently provide the seasonality of activation of dust source regions and the four-dimensional transition of dust transport pathways (Amiridis et al., 2013; Marinou et al., 2017; Proestakis et al., 2018; Aslanoğlu et al., 2022).

Towards these objectives, a generic apparent feature is that the horizontal distributions of both the DODcoarse-mode and DODfine-mode are capable of efficiently delineating both the sources of dust and the dust transport pathways (Fig. 10c, d). Overall, well-documented geographical patterns are evident in all four seasons, and although between the fine-mode and coarse-mode DODs few differences in terms of spatial distribution are revealed, the observed features significantly vary in magnitude. In general, the most intense values of the coarse-mode and fine-mode components of DOD at 532 nm are observed over the desert areas of the planet and especially over the dust belt, though the extent of the observed features highly varies with regional meteorology and seasonality (Fig. 11).

The effects of the coarse-mode and fine-mode components of pure dust on optical depth, decoupled and isolated from the effect of all non-dust aerosol subtypes, are illustrated in Fig. 10c and d, respectively. In addition, Fig. 11 provides the seasonal coarse-mode (left column) and fine-mode (right column) components of DOD at 532 nm, grouped for December–January–February (DJF) (Fig. 11a, b), March–April–May (MAM) (Fig. 11c, d), June–July–August (JJA) (Fig. 11e, f), and September–October–November (SON) (Fig. 11g, h). The objective here is to provide and discuss, through the coarse-mode and fine-mode pure-dust products, characteristics of dust sources and transport, reporting at the same time noticeable features of the datasets.

With respect to the desert areas, the Sahara is characterized by particularly intense loads of dust persistently present in the atmosphere throughout the year, though of remarkable spatial and intra-annual variability. The expected west-to-east DOD gradient is apparent in both the DODcoarse-mode and DODfine-mode products, with higher values evident to the west and lower values to the east of the domain. More specifically, over the central and western parts of the desert substantially high annual mean DODcoarse-mode (∼ 0.3) and DODfine-mode (∼ 0.12) values are observed, while over the central-eastern parts of the desert significantly lower annual mean DODcoarse-mode (∼ 0.14) and DODfine-mode (∼ 0.06) values are evident (Fig. 10c, d). With respect to seasonality, remarkably intense loads of dust are observed primarily during JJA, as high as DODcoarse-mode ∼ 0.6 and DODfine-mode ∼ 0.25 (Fig. 11e, f), and secondarily during MAM, as high as DODcoarse-mode ∼ 0.39 and DODfine-mode ∼ 0.17 (Fig. 11c, d). In contrast, during DJF (Fig. 11a, b) and SON (Fig. 11g, h) remarkably less dust activity is evident, almost uniformly over the entire Saharan desert, with DODcoarse-mode and DODfine-mode values not exceeding ∼ 0.25 and ∼ 0.15, respectively. The particularly intense loads of dust observed over the Sahara, especially between March and August, are mainly attributed to the Saharan heat low and the West African monsoon system in spring and summer (Schepanski et al., 2017). At the same time, the latitudinal migration of the Intertropical Convergence Zone (ITCZ; Schneider et al., 2014) and the intensification of the trade winds (easterlies) favor the long-distance transport of massive loads of mineral dust aerosol across the tropical Atlantic Ocean within the SAL (Kanitz et al., 2014) as far as the Caribbean Sea (Prospero, 1999), as evident by both DODcoarse-mode and DODfine-mode products. Short-distance transport of fine-mode and coarse-mode dust layers originating from the Saharan desert is also evident, especially over the Mediterranean Sea (Gkikas et al., 2013, 2015, 2016, 2022; Marinou et al., 2017; Aslanoğlu et al., 2022) under the impact of cyclone formation at the Gulf of Genoa–North African coast (e.g., Trigo et al., 1999; Maheras et al., 2001), over the Gulf of Guinea in the south (Ben-Ami et al., 2009), and over the Red Sea in the east (Banks et al., 2017; Li et al., 2018).

With respect to the Middle East and central Asia, dust activity is more pronounced in the dust-belt region extending between the eastern parts of the Sahara and the Himalaya orographic barrier, a zone encompassing several major natural sources of dust, including the great Arabian Desert, the Thar Desert, and the arid and semi-arid regions of Ethiopia, Somalia, Iran, Iraq, and Afghanistan (Ginoux et al., 2012; Hamidi et al., 2013; Proestakis et al., 2018). These desert areas are clearly mapped through systematically high annual mean DODcoarse-mode (∼ 0.16) and DODfine-mode (∼ 0.11) values (Fig. 10c, d), though the observed features are subject to high spatial and intra-annual variability (Fig. 11). With respect to seasonality, remarkably intense loads of dust are observed primarily during JJA, as high as DOD coarse mode at ∼ 0.35 and DOD fine mode at ∼ 0.23 (Fig. 11e, f), and secondarily during MAM, as high as DOD coarse mode at ∼ 0.3 and DOD fine mode at ∼ 0.15 (Fig. 11c, d), although the activation mechanisms of the desert regions in the area differ (Prospero et al., 2002). More specifically, dust emission and transport over the Arabian Desert, Thar Desert, and the deserts of Ethiopia and Somalia are mainly attributed to west Indian monsoon activity (Vinoj et al., 2014), while the relatively limited emission and transport of dust layers over Dasht-e Lut Desert and the arid areas of Iran, Iraq, and Afghanistan are mainly attributed to convective episodes (Karami et al., 2017). With respect to the spatial variability, this zone is characterized by extensive inhomogeneities in the observed fine-mode and coarse-mode dust aerosol loads. Higher annual mean values are evident over the Arabian Peninsula and the Middle East mainland (DODcoarse-mode ∼ 0.15 and DODfine-mode ∼ 0.09), with lower values over the mainland of Southeast Asia and the broader Bay of Bengal (DODcoarse-mode ∼ 0.03 and DODfine-mode ∼ 0.04). In addition, the expected and evident west-to-east gradient in the pure-dust products is attributed primarily to increased wet deposition of dust aerosol particles during the monsoon season of the year and secondarily to gravitational settling (Lau et al., 1988). The long-range transport of dust is more evident during MAM (Fig. 11c, d) and JJA (Fig. 11e, f) pre-monsoon seasons (Dey et al., 2004), when dust plumes emitted into the atmosphere are carried as far as the Indo-Gangetic Plain and exceptionally as far as the Bay of Bengal (Mao et al., 2011; Proestakis et al., 2018; Ramaswamy et al., 2018).

To the east of the Himalaya orographic barrier, the dominant natural sources of mineral dust delineated by the fine-mode and coarse-mode pure-dust products are the very arid Taklimakan Desert in northwestern China (Liu et al., 2008; Ge et al., 2014) and the vast semi-arid Gobi Desert located eastwards in the same latitudinal band across northern China and Mongolia (Prospero et al., 2002; Gong et al., 2006; Proestakis et al., 2018). Over this zone of the dust belt, the favorable topographical features of the Tarim Basin and the strong midlatitude cyclonic systems which develop over the Mongolian Plateau (Bory et al., 2003; Yu et al., 2008) mobilize dust emission sources, resulting in uplift and formation of elevated dust layers, especially during the spring and summer seasons (Proestakis et al., 2018), evident also by the high values of DODcoarse-mode (∼ 0.3) and DODfine-mode (∼ 0.1). The dust layers suspended in the atmosphere are accordingly captured by the strong westerly jet in the upper troposphere and transported over the eastern Asia region (Zhang et al., 2003; Yu et al., 2019), across the broader northern Pacific Ocean (Duce et al., 1980; Shaw, 1980; Yu et al., 2008), and exceptionally as far as the western coast of the United States (Husar et al., 2001). In contrast, during DJF (Fig. 11a, b) and SON (Fig. 11g, h) remarkably more limited dust activity is evident, almost uniformly over the entire East Asia region, as also shown by the pure-dust products, with DODcoarse-mode and DODfine-mode values not exceeding ∼ 0.25 and ∼ 0.14, respectively.

In addition to the dust-belt arid and semi-arid areas stretching from the Sahara through the Middle East and central Asia to the Taklimakan and Gobi deserts of East Asia, several other natural sources of dust are delineated and revealed by the pure-dust products, though of lower dust emission strength comparable to the major deserts of the planet. Such areas include the Mojave and Sonoran deserts in the southwestern United States–northwestern Mexico (Hand et al., 2017), and the Australian (Prospero et al., 2002), as well as southern Africa (Bryant et al., 2007) and South America (Gassó and Torres, 2019) desert areas located in the Southern Hemisphere. In contrast to the major sources of dust, over these areas remarkably more limited dust load is observed, with annual mean DODcoarse-mode and DODfine-mode values not exceeding ∼ 0.03 and ∼ 0.02, respectively.

Finally, two more noticeable features are worth mentioning. The first one relates to the preferential removal of larger dust particles during long-range transport, evident in both the CMF (Fig. 10e) and the FMF (Fig. 10f). In terms of optical depth, close to the major natural dust sources, the supermicrometer component contributes ∼ 70 % of the total dust load, falling to ∼ 40 % following removal processes during long-range transport of dust layers over distances of thousands of kilometers. This apparent feature corroborates outcomes of the SALTRACE campaign, reporting that the fraction of removed dust particles is size-dependent, shape-dependent, and increases with increasing particle size (Weinzierl et al., 2016). The second one relates to the sharp decrease (increase) in the CMF (FMF) over densely populated and heavily industrialized areas of the globe (i.e., Indian Peninsula, China, and the broader South and East Asia region). Although subject to extended scientific discussion and hard to quantify, the current consensus is that atmospheric dust is composed of natural dust (∼ 75 %) and anthropogenic dust (∼ 25 %), a categorization established on the basis of dust sources and emission mechanisms (Penner et al., 1994; Tegen and Fung, 1995; Ginoux et al., 2012). More specifically, anthropogenic activity over erodible soils or areas of little vegetation results in locally emitted anthropogenic dust into the atmosphere, mainly confined within the PBL. The contribution of anthropogenic dust over areas of major anthropogenic activity results in perturbations of atmospheric dust accumulation load compared to the surrounding areas affected mainly by long-range transport of dust, a feature apparent in the modification of the fine-mode and coarse-mode to pure-dust fractions over the regions of extensive anthropogenic activity (Fig. 10e, f).

Emphasis in this climatology of the established products, however, is not only on the horizontal distribution of the fine-mode and coarse-mode components of pure dust, as provided in terms of DODfine-mode and DODcoarse-mode at 532 nm, respectively (Figs. 10 and 11). The most unique information provided by CALIOP is the vertical structure of atmospheric constituents detected along the CALIPSO orbit path. Vertically averaging all the CALIPSO-based L2 grid-aggregated quality-screened backscatter coefficients at 532 nm, extinction coefficients at 532 nm, and mass concentration profiles of the two modes within each grid of spatial resolution 1° × 1° over a time period of interest (e.g., of annual, seasonal, or monthly temporal resolution) provides, in addition to the horizontal distribution, the vertical distribution and temporal evolution of the two modes and allows the datasets to be established under a four-dimensional database. This capacity of the established products is illustrated in Fig. 12 based on the case study of the broader Indian subcontinent and the New Delhi megacity area. More specifically, Fig. 12 provides the vertical structure of the coarse-mode (left column) and fine-mode (right column) components of atmospheric pure dust through long-term (15.5 years) annual mean mass concentration cross-sections over the Indian Peninsula (Fig. 12a, b) and time series of the temporal evolution of the coarse-mode (Fig. 12c) and fine-mode (Fig. 12d) pure-dust components in terms of seasonal mean mass concentration profiles over the New Delhi megacity area for the period June 2006–December 2021. Based on the four-dimensional (4D) reconstruction of the atmosphere in terms of fine-mode and coarse-mode pure-dust components, as illustrated in Fig. 12, several interesting characteristics of atmospheric dust are revealed, especially with respect to emission sources and transport in both the FT and the PBL.

With respect to dust layers in the free troposphere, the Indian subcontinent is largely affected by the major sources of dust located to the west of the peninsula, including, among others, the great Arabian, Dasht-e Lut, and Thar deserts (Hamidi et al., 2013; Proestakis et al., 2018; Gkikas et al., 2022). The layers of dust in the free troposphere are mainly observed during the pre-monsoon season (Dey et al., 2004), more specifically during MAM (Fig. 11c, d) and JJA (Fig. 11e, f), mainly attributed to west Indian pre-monsoon activity (Vinoj et al., 2014) when dust plumes emitted into the atmosphere are carried as far as the Indo-Gangetic Plain and exceptionally as far as the Bay of Bengal (Mao et al., 2011; Proestakis et al., 2018; Ramaswamy et al., 2018). This seasonality of long-range transport of dust layers over India, originating from the major deserts located to the west of the peninsula, is revealed by both the supermicrometer and the supermicrometer modes of atmospheric pure dust in terms of seasonal mean mass concentration profiles over the New Delhi megacity area, with higher values observed during the pre-monsoon period (Fig. 12c, d). However, in the case of the coarse-mode pure-dust product, an almost clear wave-like seasonal variation is revealed (Fig. 12c). Moreover, due to the proximity of the New Delhi area to the major desert dust sources to the west, primarily the Thar Desert and secondarily Dasht-e Lut, dust aerosol transport over this area is revealed to take place not only in the free troposphere but also in the PBL. In addition, although the long-range transport of dust layers of natural origin occurs mainly in the free troposphere, a fraction of this dust due to gravitational settling is expected to enter the PBL. Overall, the coarse-mode pure-dust product yields an annual mean mass concentration of 53.94 ± 16.98 µg m-3 for the New Delhi megacity area, as computed based on the lower 1500 m of the atmosphere. Similarly computed, the coarse-mode pure-dust mass concentration over the New Delhi area is estimated for DJF to be equal to 19.08 ± 5.36 µg m-3, for MAM 86.88 ± 18.72 µg m-3, for JJA 114.11 ± 11.35 µg m-3, and for SON 47.44 ± 10.15 µg m-3 (Table 7).

With respect to dust layers in the PBL, however, atmospheric dust is the net contribution of dust particles emitted and transported from areas outside the region of interest and dust particles emitted from within the region of interest, resulting in increased complexity with respect to atmospheric aerosol features. More specifically, the current consensus is that atmospheric dust is categorized into natural dust and anthropogenic dust, a classification established on the basis of dust sources and emission mechanisms (Penner et al., 1994; Tegen and Fung, 1995). Natural dust is suspended in the atmosphere by mechanisms that involve dust devils (Koch and Renno, 2005), haboobs (Knippertz et al., 2007), pressure gradients (Klose et al., 2010), and low-level jets (LLJs; Fiedler et al., 2013), developed over dry and unvegetated desert regions (Prospero et al., 2002). Anthropogenic dust is related directly and/or indirectly to human activity, involving, among others, transportation, infrastructure, building and road construction (Moulin and Chiapello, 2006; Chen et al., 2018), deterioration of extended soil surfaces, and changes in land use due to deforestation, grazing (Ginoux et al., 2012), urbanization, and agriculture (Tegen et al., 1996). According to studies, the magnitude of atmospheric dust classified as anthropogenic is significant. Ginoux et al. (2012) reported that on a global scale ∼ 25 % of the total dust load is of anthropogenic origin. Recent studies report that over densely populated and heavily industrialized urban areas of developing countries anthropogenic dust accounts for as much as ∼ 70 % of the total dust load (Huang et al., 2015; Chen et al., 2019). Over the Indian Peninsula, the presence of anthropogenic dust is corroborated by long-term near-surface in situ measurements of PM2.5 conducted in the framework of the Surface PARTiculate mAtter Network (SPARTAN; https://www.spartan-network.org/; last access: 14 August 2023). More specifically, SPARTAN measurements over several megacities in the broader region of the Indian Peninsula (Delhi and Kanpur, India; Dhaka, Bangladesh) report crustal material of high zinc (Zn) to aluminum (Al) ratios present in PM2.5 in situ measurements (Snider et al., 2015, 2016). Zn is mainly of anthropogenic origin (Councell et al., 2004; Begum et al., 2010), whereas Al is mainly of natural origin (Zhang et al., 2006). As such, the high Zn : Al ratios detected in crustal materials are a clear indicator that a high percentage of fine-mode dust suspended within the PBL is of anthropogenic origin (Snider et al., 2015, 2016).

Overall, the fine-mode pure-dust product yields an annual mean mass concentration of 76.55 ± 34.16 µg m-3, for the New Delhi megacity area, as computed based on the lower 1500 m of the atmosphere and based on more than 15 years of Earth observations. Similarly computed, the fine-mode pure-dust mass concentration over the New Delhi area is estimated for DJF to be equal to 50.46 ± 33.76 µg m-3, for MAM 79.27 ± 30.05 µg m-3, for JJA 83.99 ± 23.24 µg m-3, and for SON 96.95 ± 42.75 µg m-3 (Table 7). Moreover, it is estimated that in the New Delhi area the fine-mode pure-dust mass concentration values – approximately 2 d out of 3 – exceed the World Health Organization​​​​​​​ (WHO) Air Quality Guidelines (AQGs) for the PM2.5 24 h mean (https://www.who.int/publications-detail-redirect/WHO-SDE-PHE-OEH-06-02, last access: 23 August 2023).

The reported two-mode pure-dust mass concentration product tackles several challenges, especially with respect to potential applications in addressing the impact of dust on human health. To date, studies aiming to explore the negative effects of dust on air quality and human health frequently rely on passive remote sensing (De Longueville et al., 2010; Deroubaix et al., 2013; Katra et al., 2014; Prospero et al., 2014). However, it is considered particularly challenging to retrieve the elevation and extent of dust aerosol layers residing in the atmosphere by means of passive remote sensing, hampering the ability to resolve the dust load within the PBL with high accuracy, where the main anthropogenic activity takes place. Even more significant is the fact that the majority of dust-related health disorders and their intensity depend primarily on dust PSD and secondarily on the total load of dust. More specifically, only the fine mode of dust particles is inhalable deep into the lungs and alveoli, increasing morbidity and mortality rates among populations (Brook et al., 2010; Martinelli et al., 2013; Goudie, 2014). Recent studies reveal that the size distribution of airborne dust particles ranges from 0.1 to more than 100 µm in diameter (Weinzierl et al., 2016; Ryder et al., 2018), also reflecting the limitations of addressing the impact of dust on induced health disorders by means of integrated parameters corresponding to the entire PSD, since only the inhalable component of dust is of high significance for air quality. Finally, atmospheric aerosol models extensively used to provide supporting spatiotemporal information on dust emission, transport, deposition, and vertical structure (Textor et al., 2006; Astitha et al., 2012; Randles et al., 2017; Inness et al., 2019) usually employ static land cover types to classify arid and semi-arid regions as dust sources (Ginoux et al., 2001). However, the implementation of static emission inventories leads to large uncertainties, especially with regard to non-considered anthropogenic dust emission fluxes (Ginoux et al., 2012) over highly industrialized and densely populated regions of the Earth, leading to underestimations of the risk factor of dust impact on human health. Thus, the fine-mode pure-dust product established here is considered of paramount importance, since it offers pure observational evidence on the fine-mode component of dust residing within the PBL and over extended geographical regions and temporal periods, thus enhancing our EO-based capacity to assess and understand the complex role of inhalable dust particles in induced disorders on human health.