The Absorbing Aerosol Index (AAI) indicates the presence of elevated numbers of absorbing aerosols in the Earth's atmosphere. It is a unitless index and separates the spectral contrast at two UV wavelengths (340 and 380 nm) caused by aerosol scattering and absorption from other effects, including molecular Rayleigh scattering, surface reflection and gaseous absorption (Torres et al., 1998). The aerosol types that are mostly seen in the AAI are desert dust and biomass burning aerosols. AAI is a unitless parameter, with higher values indicating an elevated number of aerosols present in the atmosphere. Negative values are caused by the presence of clouds and/or scattering aerosol in the scene. However, a positive value for the AAI can only be explained by the presence of absorbing aerosols. The paper of de Graaf et al. (2005) provides several sensitivity analyses that detail the importance of the aerosol height for the interpretation of the AAI. The AAI from GOME-2 is produced by the Royal Netherlands Meteorological Institute (KNMI) – within the framework of the AC SAF. The GOME-2 AAI products are calculated for all three satellite instruments (MetOp-A, MetOp-B and MetOp-C), and data are available starting from January 2007, December 2012 and January 2019, respectively (AC SAF: https://acsaf.org/datarecord_access.php, last access: 12 February 2021; KNMI: http://www.temis.nl/airpollution/absaai/, last access: 12 February 2021).

The AAH is a new operational AC SAF EUMETSAT product for aerosol layer height detection, developed by KNMI within the AC SAF. It uses the AAI as an indicator to derive the actual height of the absorbing aerosol layer in the O2-A band using the Fast Retrieval Scheme for Clouds from the Oxygen A band (FRESCO) algorithm (Wang et al., 2008, 2012; Tilstra et al., 2010, 2012). The retrieved aerosol height varies from the bottom to the top of the aerosol layer, depending on the aerosol optical thickness (AOT), solar zenith angle (SZA) and actual aerosol layer top height (Wang et al., 2008). The AAH product can be used to monitor volcanic eruptions globally and provide the height of the ash layers (Balis et al., 2016). The AAH is very sensitive to cloud contamination. However, aerosols and clouds can prove difficult to distinguish, and AAH is computed for different FRESCO cloud fractions. FRESCO is able to determine the height of an absorbing aerosol layer not only in the absence of clouds but under certain conditions also in the presence of clouds. Further details and more information associated with the AAH product are available in the product user manual (PUM) and algorithm theoretical basis document (ATBD; Tilstra et al. 2019, PUM; Tilstra et al., 2020). The product is available openly from the AC SAF repository (https://acsaf.org/offline_access.php, last access: 12 February 2021) and has been officially validated (De Bock, et al., 2020). As discussed in the ATBD, observation pixels with AAI values below 2.0 correspond to scenes with too-low levels of aerosol to result in a reliable AAH retrieval. Also, for AAI values larger than 2.0 but smaller than 4.0, the aerosol layer is not in all cases thick enough for a reliable retrieval. However, most of our aerosol cases correspond to AAI values below the 4.0 level. The AAH product is provided, among others, with the related standard deviation value. In summary, the AAH algorithm retrieves, from the GOME-2 level-1b product, the following parameters: CF (effective aerosol/cloud fraction), CH (aerosol or cloud height), SA (scene albedo), SH (scene height). Two different aerosol/cloud layer heights (CH and SH) are determined by the AAH algorithm. It is up to the algorithm to decide which of the two is the best candidate to represent the actual AAH level. According to Wang et al. (2012), in order to distinguish whether the contribution of clouds is crucial, three situations about the reliability of the AAH product are used and the effective CF is used to check in which of these regimes is the better solution (A: high reliability, B: medium reliability, C: low reliability). In more detail:

Regime A (CF ≤0.25) refers to the situation in which there is either only a low degree of cloud cover or the aerosol optical depth is sufficiently large to compensate the presence of a cloud layer below the aerosol layer. Exceptions are cases with low aerosol numbers, but these scenes were filtered out beforehand by demanding that the AAI must be higher than a threshold AAI value.

In Sect. 3.2, a pie chart (see Fig. 6) with the distribution of reliability category (regimes) of collocated observations is presented, including the contribution of clouds.

The EARLINET database represents the largest collection of ground-based data of the vertical aerosol distribution on a continental scale. EARLINET members, as well as external users, get access to the database through a web interface (https://www.earlinet.org, last access: 23 April 2020). Additionally, EARLINET data are permanently indexed and published at World Data Center for Climate (WDCC) (https://www.earlinet.org/index.php?id=247, last access: 12 February 2021). The main information stored in the files of the EARLINET database is the vertical distribution of aerosol backscatter and extinction coefficients. Additionally, there are more optional variables included in the files, such as the lidar ratio, the particle linear depolarization ratio and the water vapour mixing ratio profiles. In this study, we use the backscatter profiles for aerosol layer height retrieval. The backscatter files contain at least a profile of the aerosol backscatter coefficient (m-1 sr-1) derived from the elastic backscatter signal and may be accompanied by an extinction coefficient profile. Here, we use the vertical information of backscatter profiles (at 1064 and 532 nm in some cases) for selected EARLINET stations. Quality-assurance (QA) tests have been established and software intercomparison campaigns (Wandinger et al., 2016; Freudenthaler et al., 2018) have been organized in the framework of EARLINET in order to assure the homogeneity of the data despite the differences in the lidar systems of the stations. A list of the EARLINET stations used for the validation of GOME-2 AAH and their geographical coordinates is given in Table 2 and presented in Fig. 1. The stations are located such that four European regions are covered: central Europe, western Mediterranean, central Mediterranean and eastern Mediterranean. In this way, a large variety of aerosol optical and geometrical characteristics can be investigated.

In this section, we analyse the algorithmic processes that are required to extract geometrical features from lidar signals employed in this work. The aerosol geometrical properties carry information about the structure of lidar profiles, such as the boundary layer height and the features of the lofted aerosol layers, and can be obtained from any lidar profile. In this study, a full lidar dataset from 13 EARLINET stations has been used for the calculations. Some lidar optical products, however, are more reliable to use than others. For example, the longer wavelengths typically magnify the differences in the vertical distribution of the aerosol load, resulting in layers that are easier to identify. Furthermore, the Raman inversion always results in profiles that are less structured for the extinction coefficients than the backscatter coefficients. This is the reason why we prioritize them so as to produce geometrical properties (Baars et al., 2008; Siomos et al., 2017). The product with the highest potential to magnify the aerosol layer structure available is selected for each measurement. More specifically, the backscatter products are prioritized over the extinction products and the longer wavelengths over the shorter ones. For this study, backscatter profiles at 1064 nm have been chosen primarily and in some cases backscatter profiles at 532 nm have been chosen.

The wavelet covariance transform (WCT) technique has proved to be one of the most reliable methods for the planetary boundary layer (PBL) top detection. Many methods have been proposed for the calculation of the PBL height from lidar data (e.g. Flamant et al., 1997; Brooks, 2003; Banks et al., 2016; Kokkalis et al., 2020). Our analysis is based on the method of Baars et al. (2008) that applies the WCT to the raw lidar data in order to extract geometrical features such as the PBL height, aerosol and cloud boundaries. The WCT transformation has also been applied successfully in the past on other lidar products (e.g. Kalman, 1960; Rocadenbosch et al., 1999). Siomos et al. (2017), for example, use an adaptation of the WCT method to calculate the geometrical features from the aerosol concentration profiles. The wavelet covariance transform was defined as a means of detecting step changes in a signal. It is based upon a compound step function, the Haar function h, defined as shown in Eq. (1): 1hz-ba=+1:b-a2⩽z⩽b-1:b⩽z⩽b+a20:elsewhere. Here, h[(z-b)/a] is the Haar function, a is the dilation of the Haar function indicating the size of the window (or dilation), b is the centre of the Haar function (or the translation), and z is the altitude range. The covariance transform of the Haar function, Wfa,b, is defined as shown in Eq. (2): 2Wfa,b=a-1∫z0z1fzhz-badz, where fz is the backscatter lidar signal, z0 and z1 are the lowest altitude and the highest altitudes of possible layers heights. The Wfa,b is referred to as the wavelet coefficient. These variables define the window function. Based on the defined lower and upper limits, the Haar transform is calculated. The obtained Haar values are subjected to the covariance transform, and the maximum negative value of the covariance transform provides the aerosol layer top. The key issues of performing the WCT are the determination of the dilation value of the Haar function. As with previous studies (Brooks et al., 2003; Baars et al., 2008), the dilation factor a affects the number of covariance wavelet transform coefficient local minima. Larger values of dilation factor reveal a few large local minima at the height of the biggest aerosol loading in the aerosol backscatter profile. In addition, lower dilation values create local minima at heights of smaller aerosol loads in the profiles. A dilation of 0.5 km is used in this study for the lofted aerosol layer height calculations. An example of a lidar backscatter profile with resulting WCT profile from the Barcelona lidar station (Universitat Politechnica de Catalunya, Barcelona – UPC) on 29 June 2019 is given in Fig. 2. This figure reasonably shows the ability of the lidar to detect multiple layers. The blue lines refer to the S–G (Savitzky–Golay smoothed signal) and the yellow one to the noisy backscatter lidar signal. The horizontal dashed red line represents the detected aerosol layer top applying the WCT methodology, and three aerosol layers are detected, according the methodology that we follow. Applying the WCT, we can check if there are strong variations in the backscatter coefficient profile within an aerosol layer, which may lead to a classification of a separate layer. The coloured “star” symbols represent the local maxima (purple) and minima (red) of wavelet transform signal.

This Évora station is located about 100 km eastward from the western Atlantic Ocean. Due to its geographical location, Évora is influenced by different aerosol types, namely urban, as well as mineral and forest fire aerosol particles. The lidar system installed here (Portable Aerosol and Cloud Lidar; PAOLI) is a multi-wavelength Raman lidar belonging to the PollyXT family (Baars et al., 2016) with high temporal and spatial resolution, operating since September 2009. It is installed at the Évora Atmospheric Sciences Observatory (EVASO) and operated by the University of Évora (UE) and the Institute of Earth Sciences (ICT) (38.56∘ N, -7.91∘ E; 293 m a.s.l.). The instrument features three elastic channels in the UV–VIS–IR range (355, 532 and 1064 nm), two inelastic (Raman) channels (387 and 607 nm) and a polarization channel which detects the cross-polarized signal at 532 nm. PAOLI is participating both in the EARLINET and the Spanish and Portuguese Aerosol Lidar Network (SPALINET) (Sicard et al., 2009, 2011). The Évora lidar system, part of EARLINET, has been quality assured through direct intercomparisons, both at hardware and algorithm levels (Pappalardo et al., 2004). During daytime, data provided by the Klett technique (Klett, 1981, 1985) use a constant lidar ratio value as input to retrieve the backscatter coefficient values with an average uncertainty of the order of 20 %–30 % (Pappalardo et al., 2014).

In February 2017, an exceptionally extreme event affected the whole Iberian Peninsula, as examined with the Aerosol Robotic Network (AERONET), EARLINET lidars and passive satellite observations (Fernández et al., 2019). MetOp overpasses close to the EARLINET station of Évora are analysed here to demonstrate the performance of the GOME-2 instrument under the intense Saharan dust outbreak (see Fig. 13). This typical case concerns an intense Saharan dust outbreak, which lasted for 3 d (21–23 February 2017) and was successfully followed during these 3 d by the Évora lidar station. A combined use of lidar profiles, back-trajectory analysis, dust models and satellite observations allows the identification of Saharan dust cases. Figure 10 shows the temporal evolution of the total attenuated aerosol backscatter coefficient at 1064 nm (m-1 sr-1) over Évora on 21–23 February.

In order to verify the origin of the aerosol layers, observed by the ground-based lidar and GOME-2/MetOp satellite, we calculated backward air-mass trajectories by using the HYSPLIT model (Hybrid Single-Particle Lagrangian Integrated Trajectory; available online at http://ready.arl.noaa.gov/HYSPLIT.php, last access: 12 February 2021) through the READY system at the site of Air Resources Laboratory (ARL) of NOAA (National Oceanic and Atmospheric Administration) in the US (Stein et al., 2015; Rolph et al., 2017). GDAS (Global Data Analysis System) meteorological files with a spatial resolution of 1∘×1∘ every 3 h, generated and maintained by ARL, are used as data input. The calculations of backward air-mass trajectories show the provenance of the air mass traversed for a chosen time period before arriving at Évora at 10:00 UTC. The temporal evolution of 5 d backward trajectories, from 21–23 February 2017 for arrival heights 1000 (red), 2000 (blue) and 3500 m (green) to cover the height range of the observed layers that we recognize in structures of height–time displays of the range-corrected lidar signal, is shown in Fig. 11. The trajectory analysis reveals that the origin of aerosol air masses is indeed the Sahara.

In Fig. 12, satellite maps from the Moderate Resolution Imaging Spectroradiometer (MODIS; Levy et al., 2013), an instrument aboard the Terra satellite, show the dust being transported by air masses over the Atlantic before returning towards Portugal and Spain on 21 (Fig. 12a), 22 (Fig. 12b) and 23 February 2017 (Fig. 12c). To illustrate the evaluation methodology for the GOME-2 level-2 AAH, a pair of collocated and concurrent GOME-2 and EARLINET lidar observations is shown in Fig. 13. We apply the proposed methodology in the measurement performed on the morning of 23 February 2017. The case study was selected as a large set of GOME-2 AAH retrieved pixels is available, and extremely high values of AAI are observed, indicating the large aerosol dust load during this day. The retrieved AAH pixels are shown in Fig. 13b, d and the retrieved AAI in Fig. 13a, c. Data gaps in the maps represent screened-out bright pixels due to either cloud or pixels affected by the sun glint effect; recall that AAH retrievals are only available when AAI is ≥2. We will examine this date in particular later on, as the extremely high AAI values, as well as the direct temporal morning collocations, give us confidence in the resulting comparisons.

As previously mentioned, both ground- and satellite-based instruments followed this major dust event for 3 d in February 2017. An example of the equivalent backscatter profiles observed by the EARLINET station and the information about coincidence of AAH measured by GOME-2 are reported in Fig. 14. The horizontal dashed blue lines in the left plots indicate the AAH value derived from the centred GOME-2 pixel. Additional information, such as the AAH, aerosol height error, AAI, CF and distance of collocated centred GOME-2 pixels from EARLINET station, is displayed in the legend. On 21 February, a well-defined aerosol layer is picked up by the lidar at 10:01:23 UTC (Fig. 14a), spanning between 1.5 and 3 km. The collocated GOME-2B observation between 09:59 and 10:30 UTC, at a distance of 62.7 km from the ground station, has an associated AAI value of 2.65, cloud fraction of 10 % and an AAH estimate at 2.07 km (dashed blue line), well within the range seen by the lidar at the surface. For the case of 22 February, the aerosol layer appears to split into two separate plumes (Fig. 14c), with GOME-2A reporting an AAI value of 2.07, i.e. quite close to the threshold value of 2.0. Even though the cloud fraction remains low (∼ 10 %), the satellite AAH estimate is quite low (0.8 km). On 23 February (Fig. 14e, f), GOME-2B reports a pixel quite close to the station, at 25 km, and even though the reported AAH of 2.8 km (dashed blue line) is well within the range of the aerosol layer height reported by the lidar, the high cloud fraction of 45 % and associated extreme AAI value of 5.75 make it difficult to draw further conclusions.

In Fig. 15, we show the comparisons for all GOME-2 pixels against the simultaneous lidar observation for 23 February over the Évora station. The collocated points are colour coded by their associated AAI value. In this way, we can assess whether the general agreement shown by the collocations of Fig. 13 can be turned into a generalized comment on the behaviour of the GOME-2 AAH algorithm for cases of high AAI and good temporal collocations. Due to the sufficient number of collocations in this case study, only observations with AAI larger than 4 are shown. The spread of the satellite estimates is within ± 1 km from the lidar observations (dashed red and green lines) for the vast majority of the cases shown, for all spatial distances between ground and satellite pixel. The results of this study case could be also interpreted by taking into account the representativeness study done using EARLINET and CALIPSO data (Pappalardo et al., 2010) during an intense dust case on 27–30 May 2008. The agreement seems to decrease with larger distances, and this follows the loss of correlation between observations when the distance from the station increases. Additionally, in the same study, Pappalardo et al. (2010) demonstrate that at 100 km maximum horizontal distance, the variability is already strong with time differences larger than 1 h, so this is probably the reason for the observed differences between satellite and ground-based observations. These results further strengthen our original assessment that the satellite algorithm is mature enough to observe stable and homogeneously distributed aerosol layers in the troposphere.