Aeolus was launched on 22 August 2018 from French Guiana. It was placed into a Sun-synchronous orbit at an altitude of 320 km and inclination angle of 97∘. It moves at 7.71 km s-1 and completes an orbit in 90 min (16 orbits per day) (Reitebuch et al., 2018; Flamant et al., 2020). The revisit time is 7 d (Flamant et al., 2020). The orbit setting of Aeolus allows it to overpass the Equator at 06:00 and 18:00 LST (local solar time) during the ascending and descending modes, respectively, in order to minimize the background noise caused by solar radiation, as it avoids solar zenith.

The Aeolus satellite is equipped with a single instrument: the ALADIN lidar. The instrument, based on the Doppler wind lidar technique, acquires profiles of wind speed and particle optical properties in the troposphere and low stratosphere up to 30 km (Ansmann et al., 2007; Flamant et al., 2008). The ALADIN lidar points towards the Earth's surface at an angle of 35∘ compared to nadir, although (due to the planet's curvature) this angle changes to 37.6∘ at surface level (Reitebuch et al., 2018). Due to the orbit configuration and the instrument design, ALADIN only retrieves the projection of the horizontal wind speed over the horizontal line of sight (HLOS), a variable sufficient for the characterization of the wind field (ESA, 2008). This variable is known as a Level 2B (L2B) product. As spin-off products, ALADIN also provides particle optical properties, namely the particle backscatter and extinction coefficients (among others), known as Level 2A (L2A) products, which are separately retrieved through the standard correct algorithm (SCA), standard correct algorithm middle bin (SCAmb), iterative correct algorithm (ICA) and Mie channel algorithm (MCA) (Flamant et al., 2020; Flament et al., 2021). The retrieval employs the High Spectral Resolution Lidar (HSRL) technique (Wandinger, 1998). Additionally, Level 2C (L2C) products consist of wind fields after the assimilation of L2B profiles by the forecast models of the European Center for Medium-Range Weather Forecasts (ECMWF) (Ingmann and Straume, 2016). On 12 May 2020, L2B products were made available to the general public (https://aeolus-ds.eo.esa.int, last access: 7 January 2022) after going through bias correction procedures. Recently, L2A products began to be produced under a new processor version (baseline 12) and are openly published alongside L2B and L2C products. At the time of writing this article, the longest, fully homogeneous product dataset has been reprocessed in baseline 10 (B10). In this study, we evaluated Aeolus B10 optical products with a thorough analysis of Aeolus co-polar backscatter coefficients using SCA and SCAmb.

The instrument emits ultraviolet (UV) radiation at 355 nm and acquires the backscattered radiation through a dual receiver, consisting of two spectrometers, with Rayleigh and Mie channels, for molecule and particle backscattering, respectively (Ingmann and Straume, 2016). Consequently, two independent wind profiles can be retrieved. However, a single measurement of aerosol optical products is obtained from the combination of both signals. The emitted radiation is circularly polarized, whilst the receiver only detects the parallel (co-polar) component, resulting in underestimated particle backscatter coefficients and overestimated extinction coefficients (Flamant et al., 2020), especially under conditions with highly depolarizing particles. ALADIN emits light pulses at a repetition frequency of 50.5 Hz. A single observation, i.e., profile, is generated by averaging shots over a 12 s period corresponding to a horizontal resolution of 87 km. Each profile is divided into 24 vertical bins. The vertical resolution of each bin depends on the altitude: 500 m between 0 and 2 km (roughly the atmospheric boundary layer), 1 km between 2 and 16 km (roughly free troposphere), and 2 km between 16 and 30 km (roughly the lowermost stratosphere) (Ingmann and Straume, 2016).

EARLINET (European Aerosol Research Lidar Network; Pappalardo et al., 2014), in the framework of ACTRIS (Aerosols, Clouds, and Trace gases Research Infrastructure Network; https://www.actris.eu/, last access: 7 January 2022), aims to generate a vast database of quality-assured vertical profiles of aerosol measurements under homogeneous standards around Europe. Thanks to its large spatial coverage of the European continent, ACTRIS/EARLINET has actively participated and participates in the calibration and validation of satellite mission measurements (Pappalardo et al., 2010; Amiridis et al., 2015; Papagiannopoulos et al., 2016; Proestakis et al., 2019). In this study, three ACTRIS/EARLINET lidar stations from the Iberian Peninsula, namely Granada, Évora and Barcelona, are considered (details in Table 1).

The ACTRIS/EARLINET Granada station (37.164∘ N, 3.605∘ W; 680 m a.s.l.) is located in the southeastern part of Spain, in a fairly populated region. The city lies in a geographic depression, at the foot of Sierra Nevada, with altitudes of up to 3479 m a.s.l. to the east of the station. Aerosol particles of anthropogenic origin, mainly released by fossil fuel burning, can be detected during the whole year (Lyamani et al., 2006, 2010, 2012). Due to the station proximity to the north of Africa, mineral dust intrusions from the Sahara Desert are often detected during the year, mainly in the summer season (Guerrero-Rascado et al., 2008, 2009; Bravo-Aranda et al., 2015; Granados-Muñoz et al., 2016; Mandija et al., 2016), although winter dust intrusions have become more frequent over the last few years (Cazorla et al., 2017; Fernández et al., 2019). The concentration of continental aerosols from the European continent are also significant throughout the year (Lyamani et al., 2010). Notable events of local wildfire smoke (Alados-Arboledas et al., 2011) and long-range-transported smoke from North America (Ortiz-Amezcua et al., 2014, 2017; Sicard et al., 2019) are usual. Bioaerosol particle concentrations, especially pollen grains, are significant over the city during specific periods of the year (Cariñanos et al., 2021). The station is equipped with a multispectral Raman lidar system, MULHACEN (LR331D400, Raymetrics S.A.), operated by the Atmospheric Physics Research Group at the Andalusian Institute for Earth System Research (IISTA-CEAMA), University of Granada.

The ACTRIS/EARLINET Évora station (38.568∘ N, 7.912∘ W; 293 m a.s.l.) is located in the southern part of Portugal, around 100 km from Lisbon and the border with Spain, in a mainly flat and rural region with relatively low industrialization and low concentrations of anthropogenic aerosol (Pereira et al., 2009; Preißler et al., 2013). Smoke particles from nearby wildfires or industrialized regions are regularly transported over the station (Preißler et al., 2013; Pereira et al., 2014), and it also receives long-range-transported smoke from North America (Sicard et al., 2019; Baars et al., 2019). Furthermore, due to the station's proximity to the Sahara Desert, mineral dust layers are frequent over the city (Pereira et al., 2009; Preißler et al., 2011, 2013) with many extreme events occurring (Preißler et al., 2011; Valenzuela et al., 2017; Couto et al., 2021). Rare events of simultaneous biomass burning and mineral dust intrusions have been characterized (Salgueiro et al., 2021). Significant bioaerosol concentration events can also occur at specific stages of the year (Galveias et al., 2021). The lidar station is operated by the Institute of Earth Sciences, associated with the University of Évora. Among other atmospheric research instruments, the station holds a multispectral Raman lidar of the POLLYXT type (Baars et al., 2016; Engelmann et al., 2016) named PAOLI.

The ACTRIS/EARLINET Barcelona station (41.393∘ N, 2.120∘ E; 115 m a.s.l.) is located on the northeastern coast of Spain, in a highly populated and industrialized region. Due to its location, the different types of aerosols present are significantly diverse. The background aerosol load, mostly made of urban, traffic-related particles and marine aerosols, is located in the lowermost part of the boundary layer (Sicard et al., 2011). Saharan mineral dust intrusions are frequent throughout the year due to the station's relative proximity to North Africa (Pérez et al., 2006). The variability of the aerosol optical properties in Barcelona in the atmospheric column has been studied by Sicard et al. (2011). The Barcelona lidar system is developed and operated by the Remote Sensing Laboratory of the Department of Signal Theory and Communications at the Universitat Politècnica de Catalunya.

The variety of locations, surroundings and orography of the three stations enables the study to explore coastal and inland, rural and urban, and flat and mountainous effects on the quality of the comparisons between the ground station and the satellite. Aeolus overpasses Évora during an ascending mode (south to north) at around 52 km east every Friday at 18:17 UTC. In the case of Barcelona, the satellite overpasses the station during an ascending mode at 26 km west every Tuesday at 17:39 UTC. The station in Granada lies at the intersection of two Aeolus overpasses: every Thursday at 06:24 UTC (ascending orbit) and 18:04 UTC (descending orbit), both at about 24 km west. All three stations fulfill the ESA requirement that only satellite overpasses with a ground-track distance of less than 100 km should be considered (Straume et al., 2019). Figure 1a shows Aeolus overpasses over a wide part of Europe. Figure 1b is a zoomed in view of the Iberian Peninsula, where all three stations are located, as well as the associated Aeolus overpasses of the case studies analyzed in Sect. 4.2.

For the intercomparison between ground-based measurements and satellite products, a series of spatiotemporal criteria was established taking the location of each station and the overpasses times into account. On the one hand, the Aeolus observation closest in distance to the station was chosen. For the location of the Aeolus overpass, the middle point of the 87 km horizontal average forming each single profile was considered. On the other hand, the temporal co-location was established according to the measurement protocols at each station. For Granada, a 1.5 h interval containing the morning overpass time (i.e., 05:30–07:00 UTC) and a 1 h interval containing the evening overpass time (i.e., 17:30–18:30 UTC) were chosen. For Barcelona, a 1 h range centered at the overpass time (i.e., 17:09–18:09 UTC) was considered. For Évora, a 1.5 h interval containing the overpass time (i.e., 17:30–19:00 UTC) was considered to take the larger distance between the Aeolus ground track and the lidar site into account. The temporal evolution of the meteorological conditions and layers has been checked individually so as to ensure that the selection of 1 or 1.5 h as average interval for the lidar signals does not introduce any artifact in the comparison.

Aeolus products are automatically processed by ESA. They are generated under specific data processing algorithms, called baselines, which are constantly being improved and updated. In October 2020 Aeolus products from July 2019 to December 2019 and from 20 April 2020 to 6 October 2020 were reprocessed under a single baseline, baseline 10 (B10), aiming to homogenize the processing of the products and encourage Aeolus Cal/Val teams to proceed with B10 products. The Aeolus database considered in this work is exclusively composed of B10 Aeolus products and covers different seasons and atmospheric conditions.

In the current study, only aerosol products (L2A) are considered, in particular particle backscatter coefficients derived from the standard correct algorithm (SCA) and standard correct algorithm middle bin (SCAmb). SCA and SCAmb employ the information from the Rayleigh and Mie channels, and the derived profiles are divided into a series of vertical bins. The difference between them is that SCAmb bins (or middle bins) are obtained from two halves of adjacent original SCA bins, aiming to reduce noise in the products. These products come with quality flags that individually mark the validity of each bin measurement. The quality flags assess the signal-to-noise ratio of both Mie and Rayleigh channels as well as retrieval uncertainties (known as error estimates). A full description of L2A products and their implemented algorithms is given by Flamant et al. (2020), and a more concise description is given by Flament et al. (2021).

The current study is developed under cloud-free scenarios. Cloud screening was performed by visual inspection of the ground-based profiles. Aeolus observations have been carefully and individually checked, and atmospheric conditions have been studied along each overpass to ensure cloud-free conditions.

The ground-based measurements are processed by the Single Calculus Chain (SCC) (D'Amico et al., 2015, 2016; Mattis et al., 2016), the standardized tool that allows one to automatically process the lidar data acquired at each station within EARLINET. Very few measurements of the Barcelona station were not inverted successfully by the SCC. In those cases, after checking the cloud-free condition, the measurements were inverted manually with an algorithm validated in previous algorithm intercomparisons at the network level (Böckmann et al., 2004; Pappalardo et al., 2004; Sicard et al., 2009). All data used in this work are Level 1.0 and Level 2.0 data from the EARLINET/ACTRIS database (http://www.actris.nilu.no, last access: 7 January 2022). The Raman-derived extinction profiles retrieved at ground level could not be used in the study due to the time of the satellite overpass, during which the signal-to-noise ratio is not good enough for these channels. Therefore, Aeolus extinction coefficients are not exploited in the study.

Aeolus underestimates the particle backscatter coefficient, as the receiver only detects the co-polar component of circular polarized backscattered radiation at 355 nm (Flamant et al., 2020), whereas ground-based lidars retrieve particle backscatter coefficients with the total backscattered radiation at 355 or 532 nm. For comparison purposes, the co-polar component of the ground-based observations at 355 nm must be extracted from the total particle backscatter coefficient. This is done using the following expression: 1βco,355part=βtotal,355part1+δcirc,355part, where βco,355part is the co-polar component of the particle backscatter coefficient at 355 nm (henceforth labeled as Aeolus-like coefficient, βAeoluslike,355part); βtotal,355part is the total component of the particle backscatter coefficient at 355 nm derived from the ground-based lidar; and δcirc,355part is the circular particle depolarization ratio at 355 nm, which is not directly measured by the considered ground-based lidars. The linear particle depolarization ratio can be easily converted into the circular particle depolarization ratio from Mishchenko and Hovenier (1995) as follows: 2δcirc,355part=2δlinear,355part1-δlinear,355part, where δlinear,355part is the linear particle depolarization ratio at 355 nm. The stations in Évora and Granada do not possess a depolarization channel at 355 nm and lack the possibility to measure δlinear,355part. A conversion of δlinear,532part into δlinear,355part is proposed in the form of 3δlinear,355part=Kδ⋅δlinear,532part, where Kδ is the spectral conversion factor. Thus, a thorough bibliographic review of previous multispectral depolarization studies has been conducted and discussed in Sect. 4.1 to estimate such a conversion factor. The third station, Barcelona, does measure both depolarization ratios; however, for the sake of consistency of the data processing, Barcelona δlinear,355part was calculated in the same way as the other two stations. In Sect. 4.1, the measurements of δlinear,355part and δlinear,532part in Barcelona are superimposed onto the literature results in order to validate our methodology.

A key point of the intercomparison is the vertical resolution of each profile. Aeolus divides each profile into a set of 24 vertical bins that are not homogeneously distributed. The resolution of these bins depends on the altitude range: 500 m between 0 and 2 km a.s.l., 1 km between 2 and 16 km a.s.l., and 2 km between 16 and 30 km a.s.l. (Ingmann and Straume, 2016). Because the ground-based lidars present a much finer resolution, of the order of a few meters, the resolution of each ground-based profile has been degraded to the Aeolus vertical resolution. Thus, the different ground-based vertical values within a given Aeolus bin are averaged into a single value, permitting a bin-to-bin intercomparison. This degradation process is performed on the ground-based Aeolus-like profiles in the last stage, prior to the statistical analysis calculations. Ground-based vertical coverage depends on the station and on each particular case, typically up to 10 km or up to the top of the highest aerosol layer, whereas Aeolus profiles extend way beyond 10 km. Hence, only statistical comparisons below 10 km, where most of the aerosols are, are presented.

The statistical results are presented in vertical ranges of 1 km. A pair of Aeolus and Aeolus-like values will fall into a given 1 km range if the middle point of the bin lies within the vertical range (for instance, if a bin ranges from 1900 to 2400 m, its middle altitude, 2150 m, lies within the 2 km vertical range). Three statistical parameters are calculated to assess the intercomparison results: difference (Δ), root-mean-square error (RMSE) and linear fit. Δ profiles are calculated as follows: 4Δr=1N∑βAeolus,355partz-βAeoluslike,355partz, where r is the vertical range, z is the middle altitude of the bin's range that lies within the r vertical range and N is the number of pairs of Aeolus and Aeolus-like values whose z lies within r. This parameter indicates if Aeolus underestimates (Δ(r)<0) or overestimates (Δ(r)>0) the co-polar particle backscatter coefficient in each region r. The RMSE profile is also obtained as follow: 5RMSEr=1N∑βAeolus,355partz-βAeoluslike,355partz-Δ(r)2. Finally, the linear regression of βAeoluslike,355part(z) against βAeolus,355part(z) is performed under a series of different settings, in order to test if they lie close to the 1:1 relation. The Pearson correlation coefficient, R, is calculated in all cases.

For the calculation of the ground-based Aeolus-like profile (Eq. 1), δlinear,355part is needed (Eq. 2). PAOLI (Évora) and MULHACEN (Granada) hold only one depolarization channel at 532 nm, whereas the lidar system at Barcelona retrieves depolarization information at 355 and 532 nm. Therefore, the estimation of δlinear,355part from δlinear,532part is required. Pairs of (δlinear,355part, δlinear,532part) obtained from a thorough review of the literature and for different aerosol types are listed in Table 2. The literature provides a modest but significant dataset for different well-characterized aerosol types, including mineral dust (fresh, aged, mixed), marine and mixed anthropogenic aerosol. These three aerosol types are the predominant aerosol types in Barcelona. Although not listed in Table 2, the literature also offers data for volcanic, bioaerosol and biomass burning particles.

A linear fit has been applied to the pairs (δlinear,355part, δlinear,532part) in order to set up a simple relationship to estimate δlinear,355part from δlinear,532part (Eq. 3) through the so-called depolarization spectral conversion factor (Kδ). Figure 2 shows the scatterplot of (δlinear,355part, δlinear,532part) for dust, marine and mixed anthropogenic aerosol from the literature, which are the aerosol types present in the cases used for the intercomparison. The best linear fit for these types together is obtained for Kδ=0.82±0.02, with a fairly acceptable statistical significance (Pearson correlation coefficient R= 0.99). A similar analysis for the other aerosol types (volcanic, bioaerosol and biomass) is performed with a linear fit calculated for each aerosol type separately. For biomass burning particles (Groß et al., 2011a; Burton et al., 2015; Janicka et al., 2017; Haarig et al., 2018; Hu et al., 2019; Stachlewska et al., 2018 and Ohneiser et al., 2020), δlinear,355part is higher than δlinear,532part and Kδ=1.34±0.07, with a reliable statistical significance (R=0.98). For bioaerosols (Cao et al., 2010 and Shang et al., 2020), Kδ is significantly smaller (0.46±0.04) and less significant (R= 0.91). For volcanic particles (Groß et al., 2012), Kδ=0.82±0.13 (and R= 0.98). The large variability in biomass burning depolarization is related to the aging of smoke particles, whereas it comes from the wide variety of bioaerosols for bioaerosol particles (see, e.g., Cao et al., 2010, where most of the bioaerosol data come from). Multispectral studies for volcanic particles are currently scarce in the literature. Dust and non-dust (generally a mixture of marine and anthropogenic particles) dual-polarization measurements from the system in Barcelona are consistent with the literature. Thus, the literature-derived relation for dust, marine and mixed anthropogenic is corroborated by the experimental values acquired over the Iberian Peninsula.

The relationships obtained between δlinear,355part and δlinear,532part (dust plus non-dust, biomass burning, volcanic and bioaerosols; see Fig. 2) aim to serve as a lookup table for any station where only the depolarization channel at 532 nm is available, which is a frequent handicap for many lidar systems worldwide. In the case of the Aeolus overpasses considered in our study, only dust, marine and anthropogenic particles have been identified, with no evidence of biomass, volcanic or bioaerosol particles in significant concentrations. For dust and non-dust types, Kδ equals 0.82±0.02, and this value is implemented from now on in the calculation of the Aeolus-like profile. The literature-derived factor is implemented in order to base the results in previous measurements reported for different environmental conditions and aerosol types, which are subsequently used as a reference in other studies.

A set of case studies is given for the different stations under relevant atmospheric conditions. These case studies illustrate the intercomparison process and serve as graphic examples of the Aeolus performance. Sun-photometer measurements are taken into account for the sake of completeness, through the study of the aerosol optical depth (AOD) at 440 nm (AOD440), the AOD-related Ångström exponent (AE) calculated between the channels at 440 and 870 nm (AOD - AE440–870), the fine-mode fraction at 500 nm (FMF500), the particle size distribution, and the single-scattering albedo (SSA) at 440 and 1020 nm (SSA440 and SSA1020) (e.g., Dubovik et al., 2002; Gobbi et al., 2007; Lee et al., 2010; Foyo-Moreno et al., 2019; Shin et al., 2019). Aerosol Robotic Network (AERONET) Level 1.5 or Level 2.0 products, depending on availability, computed from the version 3 algorithm (Giles et al., 2019) are used.

The location of the stations is highly interesting due to their proximity to the Sahara Desert and mainland Europe, so frequent events of mineral dust and anthropogenic particles could be detected by the satellite. In addition, Barcelona lies just on the coastline, and both Barcelona and Granada present high concentrations of anthropogenic aerosol, whereas Évora aerosol concentrations could be classified as rural. Thus, Aeolus operation can be tested under a complete set of atmospheric scenarios.

This section assesses the intercomparison of Aeolus SCA and SCA middle bin products with ground-based measurements from a statistical point of view. The process is performed considering Aeolus quality flags to achieve a further understanding of the products. Moreover, ground-based measurements are cloud screened (Granada case study for 5 September 2019, see details in Sect. 4.2.1, is removed as well). Taking the requirements and considerations presented in Sect. 3 into account, the initial database is largely reduced, in order to ensure the reliability of the intercomparison. From the initially available measurements, 101 B10 overpasses for Granada, 51 for Évora and 52 for Barcelona were considered, and after applying the set of requirements, the intercomparison was performed with 21 cases for Granada, 15 cases for Évora and 16 cases for Barcelona, leading to a wide dataset of cases.

First, we address the general performance of the satellite, with emphasis on the domain of Aeolus co-polar particle backscatter coefficient retrievals. On the one hand, Fig. 13a shows that Aeolus SCA retrievals range from approximately -2 Mm-1 sr-1 to large and unrealistic values that are associated with the surface-related effect shown by the satellite. On the other hand, Aeolus SCAmb retrievals range from 0 Mm-1 sr-1 to similarly large and unrealistic values (Fig. 13b). The mentioned large values (86 Mm-1 sr-1 in the case of SCA and 68 Mm-1 sr-1 in the case of SCAmb) can be considered as unrealistic when compared to particle backscatter coefficients detected during extreme events on the Iberian Peninsula (e.g., Guerrero-Rascado et al., 2009; Preißler et al., 2011; Cazorla et al., 2017; Fernández et al., 2019). With the implementation of the quality flags (Fig. 13c, d), any of the sets present negative backscatter coefficients. In fact, a little more than one out of every three SCA values (35.5 %) are negative. Additionally, unrealistic values are still flagged as valid after the application of the quality flags, although the amount is slightly reduced in absolute terms but not in relative terms. Actually, the maximum values mentioned are still flagged as valid, 86 and 68 Mm-1 sr-1 in the case of SCA and SCAmb, respectively. In the case of the dataset without the quality flag filtering, SCA presents 28 measurements above 7 Mm-1 sr-1 (used here to delimit unrealistic values) (6.6 % of all 425 measurements) whereas SCAmb presents 23 (6.6 % of all 346 measurements). After the quality flag filtering of the dataset, 27 values (28.1 % of all 96 measurements) above 7 Mm-1 sr-1 can be found for SCA and 20 (26.6 % of all 75 measurements) are found for SCAmb. Regarding the unfiltered dataset, the number of values above 7 Mm-1 sr-1 is negligible, although their effects on the statistical results of the lowermost regions can be observed, as will be addressed further on. On the contrary, the amount of values above 7 Mm-1 sr-1 is substantial and relevant for the filtered dataset, as the number of values below 7 Mm-1 sr-1 is dramatically reduced.

Figure 14a and b present βAeoluslike,355part and βAeolusSCApart values, which do not fit any linear or nonlinear relation. The linear fit of each dataset (red lines in Fig. 14) is not good (R smaller than 0.4 in all cases). Several types of relationship were tested, and no valid model was found. Quality flagged data (Fig. 14c, d) worsen the linear relationship (R smaller than 0.25 in both cases). The same analysis has been performed for each station separately (not presented here) in order to search potential particularities of each site. Analogous and unsatisfactory results were found with the dataset of each station.