Three-dimensional evolution of Saharan dust transport towards Europe based on a 9-year EARLINET-optimized CALIPSO dataset

In this study we use a new dust product developed using CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation) observations and EARLINET (European Aerosol Research Lidar Network) measurements and methods to provide a 3-D multiyear analysis on the evolution of Saharan dust over North Africa and Europe. The product uses a CALIPSO L2 backscatter product corrected with a depolarization-based method to separate pure dust in external aerosol mixtures and a Saharan dust lidar ratio (LR) based on long-term EARLINET measurements to calculate the dust extinction profiles. The methodology is applied on a 9-year CALIPSO dataset (2007–2015) and the results are analyzed here to reveal for the first time the 3-D dust evolution and the seasonal patterns of dust over its transportation paths from the Sahara towards the Mediterranean and Continental Europe. During spring, the spatial distribution of dust shows a uniform pattern over the Sahara desert. The dust transport over the Mediterranean Sea results in mean dust optical depth (DOD) values up to 0.1. During summer, the dust activity is mostly shifted to the western part of the desert where mean DOD near the source is up to 0.6. Elevated dust plumes with mean extinction values between 10 and 75 Mm−1 are observed throughout the year at various heights between 2 and 6 km, extending up to latitudes of 40 N. Dust advection is identified even at latitudes of about 60 N, but this is due to rare events of episodic nature. Dust plumes of high DOD are also observed above the Balkans during the winter period and above northwest Europe during autumn at heights between 2 and 4 km, reaching mean extinction values up to 50 Mm−1. The dataset is considered unique with respect to its potential applications, including the evaluation of dust transport models and the estimation of cloud condensation nuclei (CCN) and ice nuclei (IN) concentration profiles. Finally, the product can be used to study dust dynamics during transportation, since it is capable of revealing even fine dynamical features such as the particle uplifting and deposition on European mountainous ridges such as the Alps and Carpathian Mountains.


Introduction
Mineral dust is ubiquitous in the atmosphere and one of the main contributors to the global aerosol load (Zender et al., 2004;Textor et al., 2006), with almost half of the global dust emissions generated in Africa (Huneuus et al., 2011).This has large consequences for air quality downwind (Viana et al., 2002;Gobbi et al., 2007), for the radiative budget due Published by Copernicus Publications on behalf of the European Geosciences Union.
to scattering, absorption, and emission of solar and terrestrial radiation (Balkanski et al., 2007), as well as for cloud formation and lifetime (e.g., DeMott et al., 2003;Levin et al., 2005;Koren et al., 2010).These effects depend strongly on the vertical distribution of dust.For example, dust particles will have a stronger impact on shortwave radiation absorption when they are located above bright clouds (Yorks et al., 2009;Winker et al., 2013).Moreover, dust atmospheric lifetime is much longer in the free troposphere than in the planetary boundary layer, and, upon entering the free troposphere, dust particles can be transported across vast areas, altering the geographic pattern of their impacts (Prospero and Lamb, 2003;Levin et al., 2007;Ridley et al., 2012).Finally, the dust vertical distribution is crucial for dust-cloud interactions (e.g., Mamouri and Ansmann, 2016;Nickovic et al., 2016).Therefore, observing, monitoring and quantifying the atmospheric dust burden and especially its vertical distribution is an important step towards understanding the climatic role of dust (IPCC, 2013, WG1, chap. 5, 7 and 9).
Lidar is the most prominent tool for aerosol profiling and has largely contributed to our knowledge of the vertical distribution of the dust optical properties (e.g., Liu et al., 2002;Ansmann et al., 2003;Balis et al., 2004;Papayannis et al., 2008;Mona et al., 2012;Granados-Muñoz et al., 2016;Bovchaliuk et al., 2016).Polarization lidar observations greatly expand the capabilities for dust detection, as non-spherical dust particles have a distinct signature on the particle depolarization ratio (e.g., Z. Liu et al., 2008;Tesche et al., 2009).In Europe, the European Aerosol Research Lidar Network (EARLINET; Pappalardo et al., 2014) operates advanced lidar systems employing depolarization techniques that have been invaluable for dust research.Moreover, sophisticated methodologies developed in EARLINET allow the complete characterization of different aerosol types including dust (e.g., Papayannis et al., 2008) as well as the dust contribution to the total aerosol load (Tesche et al., 2009).
The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) mission equipped with the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument has been delivering aerosol and cloud profiles across the globe for more than 10 years (Winker et al., 2009).This dataset offers the possibility to characterize the three-dimensional spatial distribution of aerosol as well as its temporal variation.CALIPSO is established as an accurate and robust means for mineral dust identification from space (Liu et al., 2008;Omar et al., 2009).The application of EARLINET methodologies on CALIPSO observations can improve the observations for mineral dust research, as already suggested and applied in Amiridis et al. (2013).Specifically, this study retrieves the extinction of pure dust from CALIPSO with high accuracy, applying the depolarization-based separation method introduced by Tesche et al. (2009), coupled with a regionally uniform climatological LR (lidar ratio) for calculating dust extinction.The latter is estimated from long-term EARLINET measure-ments (Wandinger et al., 2010;Baars et al., 2016).It has been shown that the EARLINET-optimized CALIPSO dust product presented in Amiridis et al. (2013) is in better agreement with Aerosol Robotic Network (AERONET) collocated measurements over the Sahara and Europe and with Moderate Resolution Imaging Spectroradiometer (MODIS) measurements over the Mediterranean for collocated cells with low cloudiness.This product is considered as the first accurate dust retrieval from space, since dust discrimination methods applied on passive sensors are based on the separation of the fine from coarse particle mode (e.g., Kaufman et al., 2005), delivering mostly biased DODs (dust optical depths) over the oceans due to the contamination of the coarse mode by sea salt particles (Su et al., 2013).Another advantage of the EARLINET-optimized CALIPSO dust product is its capability to provide accurate dust retrievals over all surface types, since CALIOP uses its own light source, overcoming the surface reflectance limitations of passive sensors (e.g., Hsu et al., 2004;Sayer et al., 2012).
Many studies have used satellite observations to derive dust properties over the Mediterranean during the last 15 years.Most of them focus on the horizontal distribution of dust using passive remote sensing techniques.Antoine and Nobileau (2006) used SeaWIFS (Sea-viewing Wide Field-of-view Sensor) observations to study the seasonal evolution and variability of dust aerosols over the broader Mediterranean Sea during the period 1998-2004.Alpert and Ganor (2001) and Israelevich et al. (2002) used the Total Ozone Mapping Spectrometer (TOMS) aerosol index (AI) product to study the concentration of dust over the Middle East and the dust sources of northern Africa, respectively.The MODIS instrument, onboard both Terra and Aqua satellites, has been extensively used in studies of airborne mineral dust over the Mediterranean Basin: Barnaba and Gobbi (2004) analyzed 1-year (2001) MODIS-Terra AOD (aerosol optical depth) at 550 nm observations and reported on the spatial distribution and seasonal variability of aerosols, including dust, over southern Europe, with a focus over the Mediterranean region; Papayannis et al. (2005) used MODIS-Terra data synergistically with lidar measurements and dust model simulations and investigated the vertical distribution of aerosols during dust outbreaks over Greece; Kosmopoulos et al. (2008) and Papadimas et al. (2008) used MODIS-Terra and MODIS-Aqua to investigate the seasonal and interannual variability of AOD at 550 nm over Athens (Greece) and over the broader Mediterranean Sea, respectively; Marey et al. (2011) analyzed 10 years of MODIS data synergistically with MISR (Multi-angle Imaging SpectroRadiometer) and OMI (Ozone Monitoring Instrument), and they produced a monthly climatology of aerosols over a domain covering the Nile Delta and northeast Africa.With respect to CALIPSO, the 3-D distribution of dust and its optical properties have been studied for specific cases (e.g., Amiridis et al., 2009;Mamouri et al., 2009;Marey et al., 2011;de Meij et al., 2012;Nabat et al., 2012Nabat et al., , 2013;;Mamouri and Ansmann, 2015).Moreover, Winker et al. (2013) provided a 3-D global aerosol climatology from 5-year CALIPSO data, along with the global distribution of mineral dust, derived using the ratio of columnar dust AOD to total AOD.Other studies offering a global view of desert dust using CALIPSO are provided in D. Liu et al. (2008), Z. Liu et al. (2008), Adams et al. (2012), Yang et al. (2012), Tsamalis et al. (2013), Huang et al. (2015a, b) and Gkikas et al. (2016).In particular, the studies of Z. Liu et al. (2008), Yang et al. (2012) and Tsamalis et al. (2013) examined the transatlantic Saharan dust transport, focusing on the optical properties of dust, the influence of nearby clouds and the vertical distribution of the Saharan Air Layer, respectively.Huang et al. (2015a) assessed the inferred most probable heights of global dust and introduced a separation method (Huang et al., 2015b) of anthropogenic dust (produced by human activities on disturbed soils) and free-tropospheric dust using CALIPSO and MODIS products.D. Liu et al. (2008), Adams et al. (2012) and Gkikas et al. (2016) used CALIPSO observations in order to demonstrate the vertical structure of dust globally and/or above the Mediterranean.All the aforementioned studies are based on standard CALIPSO products with known limitations in accurately typing and quantifying the optical properties of pure dust (Wandinger et al., 2010;Tesche et al., 2013).The EARLINET-optimized dust CALIPSO product presented herein was used in Georgoulias et al. (2016a) to apply aerosol typing on MODIS and derive an aerosol climatology over the eastern Mediterranean.
To our knowledge, this is the first time that a 3-D puredust dataset is statistically analyzed over the area of North Africa and Europe in order to provide not only the horizontal but also the vertical patterns of Saharan dust intrusion in the Mediterranean.The study domain is from 20 to 60 • N and from 20 • W to 30 • E.More specifically, we investigate the 3-D inter-seasonal variation and intensity of dust transport patterns along with the interannual variations of DOD above this region.The paper is organized as follows.In Sect. 2 the CALIPSO lidar data are briefly introduced and the puredust retrieval scheme is described in detail.In Sect.3, the main findings are presented and discussed; initially, the interseasonal variation and intensity of dust transport patterns (e.g., DOD, dust layer heights) are presented (Sect.3.1-3.3),and the representative extinction coefficient values inside the dust plumes are derived (Sect.3.4).In Sect.3.5, the interannual variation of dust is examined, while our summary and concluding remarks are given in Sect. 4.
2 Data and methodology 2.1 CALIPSO product CALIOP, flying onboard the joint NASA-CNES CALIPSO satellite, has delivered global aerosol and cloud profiles since June 2006 (Winker et al., 2009).CALIOP measures aerosol backscatter profiles at 532 and 1064 nm, including parallel and perpendicular polarized components at 532 nm, at high horizontal and vertical resolution.The data are processed to Level 2 (L2) products, providing aerosol and cloud backscatter and extinction coefficients at 532 and 1064 nm as well as the linear particle depolarization ratio at 532 nm (Winker et al., 2009).First, the processing algorithm separates the atmospheric scene in distinct atmospheric layers (i.e., aerosol, cloud and surface returns; Vaughan et al., 2009).Then, for each aerosol layer the algorithm determines an aerosol subtype (i.e., dust, polluted dust, clean continental, polluted continental, marine and smoke) based on a combination of information, such as the surface type, the layer integrated attenuated backscatter, the depolarization ratio at 532 nm and the aerosol layer height (Omar et al., 2009).The inferred subtype is used to derive the appropriate lidar ratio, which is a crucial input for the subsequent aerosol extinction retrieval (Young and Vaughan, 2009).Burton et al. (2013) showed an 80 % successful detection of dust from CALIPSO, upon comparison to underflights with the HSRL (High Spectral Resolution Lidar) system of NASA.This score is considered very high for aerosol typing purposes and is attributed to the depolarization measurement capability of the CALIOP sensor.Finally, the L2 products are aggregated to a gridded monthly mean Level 3 (L3) product, providing mean profiles of extinction at 532 nm and mean AOD at a 2 • × 5 • spatial grid resolution (Winker et al., 2013).The most recent version of the L3 product (version 3), released in October 2015, includes the correction of the AOD of cloudy scenes, the improved averaging of individual types as proposed by Amiridis et al. (2013) and D. Liu et al. (2008), and corrections of signal artifacts responsible for high and low biases as also observed in Papagiannopoulos et al. (2016).

EARLINET-optimized CALIPSO product
In this study, we make use of the EARLINET-optimized pure-dust extinction product, monthly averaged at a horizontal resolution of 1 • × 1 • , based on the methodology described in Amiridis et al. (2013).This product is a prominent outcome from the EARLINET-ESA collaboration for the LIVAS database (LIdar climatology of Vertical Aerosol Structure for space-based lidar simulation studies; Amiridis et al., 2015).Unlike the original CALIPSO L3 product of 2 • × 5 • resolution, the 1 • resolution of LIVAS has been proven very useful in supporting studies of the same spatial resolution, specifically for the retrievals from passive satellite sensors and model evaluation studies (e.g., Popp et al., 2016;Georgoulias et al., 2016b;Tsikerdekis et al., 2017).In our methodology, the pure-dust backscatter coefficient (β d ) is decoupled from the total aerosol backscatter (β p ) based on depolarization measurements (δ p ), assuming a particle depolarization ratio value for pure dust (δ d ) equal to 0.31 (Tesche et al., 2009) value, showing little variation independent of the source region (e.g., Sakai et al., 2000;Z. Liu et al., 2008;Freudenthaler et al., 2009;Burton et al., 2013;Groß et al., 2011Groß et al., , 2013Groß et al., , 2015;;Illingworth et al., 2015).During the SAMUM (Saharan Mineral Dust Experiment) 1 and 2 campaigns, Saharan dust δ nd values varied between 0.27 and 0.35 at 532 nm (Ansmann et al., 2011), introducing a 4 % error in our calculations for the dust separated backscatter values.Using this separation technique, we avoid relying on the polluted dust and dust aerosol types used in CALIPSO and, thus, eliminate possible misclassifications found in the CALIPSO L2 product (Burton et al., 2013).A final correction is related to the particle linear depolarization ratio, which is recalculated from L2 perpendicular and total backscatter profiles, to improve the accuracy compared to the original CALIPSO L2 version 3 product, affected by a known bug (Tesche et al., 2013;Amiridis et al., 2013).
The quality control procedures and filtering criteria applied in the dataset are summarized in Table 1.In brief, CALIPSO L3 version 3 screening procedure is followed (Winker et al., 2013;CALIPSO L3-V3, 2015), and additional filters are incorporated to ensure the use of only cloud-free profiles.The additional methodology is as follows: a.We remove all profiles with cloud features anywhere in the column.
b.We remove all profiles which fulfil the L3 CALIPSO "CAD score" or "Cirrus fringes" filters (see also Table 1).
The pure-dust extinction coefficient is computed using a lidar ratio of 55 sr instead of the 40 sr used in the CALIPSO product (Omar et al., 2009;Lopes et al., 2013).This value is representative of dust over Europe, mainly originating from northwest Africa, as measured in coordinated CALIPSO-EARLINET measurements (Pappalardo et al., 2010;Wandinger et al., 2011), and is in excellent agreement with recent studies of dust measurements both near the source (Tesche et al., 2009;Veselovskii et al., 2016) and during long-range transport (Preißler et al., 2012;Kanitz et al., 2013;Groß et al., 2015;Baars et al., 2016;Papagiannopoulos et al., 2016).The individual backscatter coefficient profiles at 532 nm are aggregated at a horizontal spatial resolution of 1 • × 1 • and a vertical resolution of 60 m from −0.5 to 20.2 km and 180 m from 20.2 to 30.1 km.This product height is referenced to above sea level (a.s.l.) altitude.

Climatological vs. Conditional dust product
In this study, we calculate two separate dust products -the climatological and the conditional.
The climatological dust product is based on Amiridis et al. (2013), with a value of 0 km −1 assigned to the non-dust aerosol types when averaging within a cell.This product, hereinafter, is referred to as the Climatological Dust Extinction (Clim-DE), and the corresponding AOD as Dust AOD (DOD), and is presented and discussed in Sect.3.1-3.3.As already discussed in the introduction, this product has been evaluated against AERONET data and is in very good agreement with collocated measurements over the Sahara and Europe (Amiridis et al., 2013).The averaging methodology has been adapted by the L3-V3 CALIPSO product.
The conditional dust product is derived from averaging the CALIPSO dust extinction coefficients where dust is present, ignoring non-dust observations in the area.In particular, the clear air and non-dust aerosol types detected in the cell are ignored (set as NaN values when averaging).This product is referred to as the Conditional Dust Extinction coefficient product (Con-DE) and is presented and discussed in Sect.3.4.
The two products can be used for different applications.For example, Clim-DE is representative of the dust contribution to the total aerosol load and can be valuable in climatological studies.Moreover, the near-surface DOD helps to estimate the natural aerosol contribution in the total aerosol load close to the surface for air-quality applications.The Con-DE product, on the other hand, provides a measure of the intensity of the dust plumes.

Dust product uncertainties
The sources of uncertainties for the pure-dust product are discussed in this section.CALIOP is able to detect aerosol layers with AOD > 0.005 and β > 0.25 Mm −1 sr −1 (Winker et al., 2009).The uncertainty estimation of particulate backscatter, extinction and AOD retrievals reported in the CALIPSO Level 2, version 3 data release, are based on the simplified assumption that all the uncertainties are random, uncorrelated and produce no biases (Young, 2010).More specifically, ignoring multiple scattering, the errors in the layer optical depth calculations typically arise from three main sources: (a) signal-to-noise ratio within a layer, (b) calibration accuracy and (c) the accuracy of the lidar ratio used for the extinction retrieval.The lidar ratio uncertainty is the dominant contributor to the total uncertainties, and the relative error in the layer optical depth is always at least as large as the relative error in the lidar ratio of the layer and grows as the solution propagates through the layer (CALIPSO L2-V3, 2010).In our dataset the typical uncertainties in the CALIPSO Level 2 version 3 product are between 30 and 100 % for the AOD, between 30 and 160 % for the aerosol backscatter and extinction coefficient, and > 100 % for the particle depolarization ratio.
Several studies report that CALIPSO underestimates the columnar AOD due to undetected aerosol in the free atmosphere.For instance, Rogers et al. (2014) report a ∼ 0.02 AOD CALIPSO underestimation, when compared to collocated airborne HSRL measurements over the North American and Caribbean regions at night.In their data, the dust layers were primarily non-opaque with extinction less than 1 km −1 , so there were negligible multiple scattering effects.
Table 1.Quality control procedures and filtering applied in CALIPSO data.
1 Screen out all features that are not aerosols 2 Set all clear air profile measurements to 0.0 km −1 3 Samples below opaque cloud and aerosol layers are removed 4 Clear-sky mode: only measurements in which no clouds are in the column are considered 5 Large negative near-surface extinction filter: all level 2 aerosol extinction samples adjacent to the surface having a value less than −0.2 km −1 are ignored 6 Samples where aerosol extinction uncertainty is less than 99.9 km −1 are allowed 7 CAD score: only features having cloud-aerosol discrimination (CAD) scores between −100 and −20 are used 8 Only features having extinction QC flag values of 0, 1, 16 or 18 are allowed 9 Cirrus fringes: misclassified cirrus in the upper troposphere, coming from CAD artifacts, are removed 10 Remove measurements which are contaminated by surface values: extinction values near the surface less than −0.2 km −1 are ignored 11 Undetected surface-attached aerosol low bias filter (changed between CALIPSO L3 version 1 and version 3): samples classified as "clear air" lying beneath the lowest quality screened aerosol layer whose base is below 250 m from the local surface are ignored 12 Negative signal anomaly mitigation strategy: all level 2 aerosol extinction coefficients within 60 m of the planetary surface are excluded from level 3 calculations (new in L3 version 3) 13 All non-dust aerosol types detected in the cell are assigned with a value of 0.0 km −1 Extra filters with more strict cloud screening 14 All profiles having cloud features anywhere in the column are removed 15 All profiles which fulfil the L3 CALIPSO "CAD score" or "Cirrus fringes" filters are removed These detection limits and uncertainties of CALIPSO products are propagated to the dust product presented here.
As already described, the EARLINET-optimized CALIPSO dust product is derived using the depolarizationbased separation method, coupled with the selection of a uniform climatological LR value.These steps introduce uncertainties in the pure-dust product.In particular, the uncertainty in the selection of the representative LR (55 ± 11) is 20 % for the study area (e.g., Wandinger et al., 2010;Baars et al., 2016, and references within).This uncertainty in LR is less than half of the uncertainty of the generic LR in the CALIPSO version 3 product (40 ± 20 for dust layers and 55 ± 22 for polluted dust layers).As already addressed in several studies (e.g., Wandinger et al., 2010;Schuster et al., 2012;Amiridis et al., 2013), the CALIPSO V3 dust extinction coefficient and AOD values are about 30 % lower than those obtained from collocated ground-based Raman lidar retrievals due to the low LR used in the CALIPSO aerosol retrievals.Amiridis et al. (2013) applied the EAR-LINET LR for the pure-dust CALIPSO cases above North Africa and Europe and compared with synchronous and collocated AERONET measurements.The results showed an absolute bias on the AOD of the order of −0.03, improving on the statistically significant biases of the order of −0.10 reported in the literature for the original CALIPSO product.The bias of −0.03 is similar to the low bias of CALIPSO's column AOD due to undetected aerosol layers.In Kim et al. (2017), they found a global mean undetected layer AOD of 0.0031 ± 0.052 by comparing 2 years of CALIPSO (L1-V4) and MODIS AODs.
Regarding the error induced from the application of the dust separation method, this might be due to the selection of the particle depolarization ratio of dust and the other aerosol types (marine, anthropogenic or smoke).Tesche et al. (2009, 2011), and Ansmann et al. (2012) estimated that the uncertainty in dust related backscatter coefficients is 15-20 % in well-detected dessert dust layers and 20-30 % in less pronounces aerosol layers.Moreover, we have calculated that the uncertainty of the dust occurrences presented in Sect.3.1 ("% dust/used overpasses"), might be up to 8 % in latitude away from the sources, which is induced from the error in the selection of the δ nd value (0.03 ± 0.04).Finally, an uncertainty induced in the dust product presented in this work originates from the CALIPSO subtype selection algorithm.In this version of our product, both dust and polluted dust observations are considered polluted dust and the pure-dust component is separated using the dust separation method.The other aerosol layers, which are characterized as clean marine (CM), smoke (S), polluted continental (PC) or clean continental (CC), are considered to be cases clear of dust and are not tested for a dust component.This introduces negligible error in our analysis and is expected to induce a negative bias in the parameter % dust/used overpasses less than 8 %, mainly in areas above sea.In general, for Clim-DE and Cond-DE products, the uncertainty of the dust extinction values close to the surface and at high latitudes is < 54 %.At high altitudes and for latitudes up to 45 • N, the uncertainty of the values is < 20 %.Nevertheless, the standard deviation of the climatological products, coming from the natural variability of the dust events, may exceed to a large extent the uncertainty of the retrieval, reaching values as high as 100 and 200 %.
In the latest release of the CALIPSO Level 2 version 4 product (CALIPSO L2-V4, 2016), based on the CALIPSO team announcement, the accuracy of the original CALIPSO product is increased and the uncertainty is reduced.This version is based on a revised calibration approach which leads to an increase in the total attenuated backscatter coefficients by ∼ 3 % overall as compared to the version 3 values (CALIPSO L1-V4, 2016).Several bugs are fixed and a major overhaul of the aerosol subtyping algorithms along with revisions on the lidar ratio selections is applied.

Additional satellite and model dataset
s The sixth version level-3 MODIS-Terra is a 1 • × 1 • gridded aerosol dataset that is acquired from the NASA Giovanni system (https://giovanni.sci.gsfc.nasa.gov/giovanni/).In the current study the MODIS combined dataset of aerosol optical depth (that takes into account the dark target, dark surface, and deep blue, bright surface, measurements) was used for the period from 2007 to 2015 (Sayer et al., 2014).Over the Mediterranean, MODIS-Terra v6 was evaluated against 23 AERONET stations and was proven to score better than its predecessor MODIS-Terra v5 (Georgoulias et al., 2016b).
The MACC global dataset is a reanalysis product based on the Integrated Forecast System (IFS) of the European Centre for Medium-Range Weather Forecast (ECMWF) coupled with the chemistry transport model MOZART-3 (Kinnison et al., 2007).The horizontal resolution of the model is 80 km, and it uses 60 vertical levels from the surface up to 0.1 hPa.MACC has been used in numerous gas phase and particulate matter studies (Innes et al., 2013;Katragkou et., 2015;Eskes et al., 2015;Flemming et al., 2015;Cuevas et al., 2015;Georgoulias et al., 2017).The dust optical depth data used in this study covers the period 2007-2012, and all MACC data are open to the public (http://apps.ecmwf.int/datasets/data/macc-reanalysis/levtype=sfc/).RegCM4 is an open source, area-limited, sigma-p vertical coordinated regional climate model (Giorgi et al., 2012) based on the hydrostatic core of the Penn State University/National Center for Atmospheric Research (PSU/NCAR) mesoscale model (MM5; Grell et al., 1994).The simulation used in the current study is part of a previous research where the dust optical depth of the model was evaluated against the dust climatological product of this work, after it was fully spatially and temporally collocated with the exact flyby of CALIPSO (Tsikerdekis et al., 2017).The simulation covers the period from 2007 to 2014 with a horizontal resolution of 50 km and 18 vertical sigma-p levels.

Results and discussion
In Sect.3.1-3.4,we examine the inter-seasonal variation and intensity of dust transport patterns, from 2007 to 2015, for the domain 20 • W to 30 • E and 20 to 60 • N. In Sect.3.1 we provide the average climatological state of the seasonal dust distribution at a spatial resolution of 1 • × 1 • .In Sect.3.2 we give information on dust layer heights.In Sect.3.3, we illustrate the mean climatological vertical structure of dust reaching Europe.To achieve that, the area of study is separated into five longitudinal zones with a step of 10 • .In Sect.3.4, we illustrate the vertical intensity of the dust plumes, using again longitudinal zone maps.Finally, in Sect.3.5, we examine the interannual variation of dust.

Horizontal dust distribution
In this section, we provide the average climatological state of the seasonal horizontal dust distribution derived from the CALIPSO dust product at a spatial resolution of 1 • × 1 • for the domain of North Africa and Europe.The seasonal grouping used in this study is as follows: from January to March (JFM), from April to June (AMJ), from July to September (JAS) and from October to December (OND).In our study region, March and October are considered transition months for Saharan dust advection (e.g., Ganor, 1994;Guirado et al., 2014).This grouping is based on the dominant patterns revealed from the maps of monthly mean DODs (not shown): the events during February-March and October-November, although rarer, are usually more intense than those of the other months.This is further supported from a 10-year (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011) analysis of African dust outbreak PM 10 (particulate matter with aerodynamic diameter lower than 10 µm) observations over the Mediterranean Basin (Pey et al., 2013).
Figure 1 shows the geographical distribution of dust occurrences (Fig. 1a, c, e, g) and the corresponding mean DOD values for each season (Fig. 1b, d, f, h).In order to provide a more quantitative representation of the dataset, the domain is aggregated in six areas over the study region.The main results and statistical parameters are provided in Table 2, and a map with the domains is shown in Fig. 2. In particular, the information provided is the mean and standard deviation of the DOD, the maximum values along with the 95th percentile in parenthesis, the layer's center of mass and top height along with their standard deviations (these parameters are discussed in the next section), the percentage of the observations with DOD greater than zero in the cloud-free observations, the percentage of the cloud-free occurrences in the total observations provided by the CALIPSO product (with 100 % as unity) and each domain's geographic extent.
Table 2 shows the impact of cloud contamination in our dataset.During AMJ, JAS and OND, more than 80 % of the total observations are cloud free above North Africa.The percentage is the same above the central-eastern Mediterranean  ) is approximately 60-80 %.With increasing latitude, the cloud-free sampling is reduced to percentages of ∼ 40-60 % in latitudes greater than 45 • N.During JFM, cloudy conditions restrict our dataset in the greatest extent.During the same period, the cloud-free cases used represent ∼ 80 % of the total observations above North Africa, approximately 60-70 % above the Mediterranean and ∼ 30 % in the domain between 45 and 60 • N. In the areas (and seasons) where clouds do not dominate (e.g., 70 % clearsky conditions), our cloud-free product is considered representative of the dust distribution.In areas where cloudy skies dominate (e.g., 30 % clear-sky conditions), the clearsky CALIPSO profiles cannot be considered as representative of all meteorological conditions, so the results should be used with caution.
Based on Fig. 1 and Table 2, the overall percentages of dust occurrences and mean DOD values are greater during summer and spring months.During autumn and winter the emission and transport of dust towards Europe is suppressed due to the more effective removal processes and due to the atmospheric dynamics favoring the transport of dust towards the Atlantic (e.g., Israelevich et al., 2002;Schepanski et al., 2009).More specifically, during JFM (Fig. 1a, b) limited dust activity is observed almost uniformly over the Sahara desert.The DOD remains below 0.13 roughly over the entire study domain, with 75 % of the observations having DODs < 0.17, 95 % of the observation having DODs < 0.5 and extreme values with DODs > 2. The dust occurrences decrease with latitude and the presence of dust is approximately 70 % over Africa and the Mediterranean region and decreases to lower than 50 % over northern Europe.The most affected area during these months is eastern Mediterranean.The cyclone formation over the central Mediterranean, which is affected by mid-latitude depressions generated either in the Atlantic Ocean or in northwestern Europe (e.g., Trigo et al., 1999;Maheras et al., 2001), results in the transportation of dust from the Libyan Desert towards the Balkans, leading to dust occurrences up to 70 % (Fig. 1a) along with mean seasonal DODs of 0.1-0.2 (Fig. 1b).In the domains between 10 and 30 • E and 30-40 • N, 5 % of dust events are observed with DODs > 0.41, 1 % with DODs > 0.95 and extreme observations with DODs are up to 1.6.Similar mean values have been reported in the literature for this period, along with  2 and 3. extreme events characterized by AOD values higher than 1 (Gerasopoulos et al., 2011).Moving northward, mean DOD tends to decrease due to the increasing distance from the major dust sources and also due to higher precipitation at the northern parts of the study region that efficiently removes dust from the atmosphere (e.g., Moulin et al., 1998;Marrioti et al., 2002).
During AMJ (Fig. 1c, d) dust production occurs over the entire Saharan desert, with mean DOD values of 0.26 ± 0.26 and occurrences of 86 %, uniformly at latitudes between 20 • N and 30 • N. The activated dust sources are located in the broad "dust belt" and are usually associated with topographical lows in the arid regions and with the intermountain basins (Prospero et al., 2002).The arrival of mid-latitude extratropical cyclone systems from the Atlantic Ocean as well as cyclogenesis at the Gulf of Genoa and/or at the northern African coast favors dust transport over the central and eastern Mediterranean.Mean DOD over these areas reaches values of 0.12 ± 0.20 (Fig. 1d) and extreme observations observed with DODs up to 2.74.Dust is also present over central and northern Europe with mean DOD up to 0.033 ± 0.062 and occurrence percentages up to 61 % (Fig. 1c; Table 2), revealing that dust particles can be transported far away from their sources under favorable meteorological conditions.
During JAS (Fig. 1e, f), intense dust activity is prominently shifted to the western part of the Sahara where dust occurrences are > 90 % and mean DOD near the sources is up to 0.6 (Fig. 1f).In the domain between 10 • W-00 • and 20-35 • N, the mean DOD is 0.43, with 25 % of the dust observations having DODs > 0.69, 5 % > 1.2 and the extreme DODs up to 3 (Table 2).The migration of the ITCZ (Intertropical Convergence Zone) towards higher latitudes and the dominance of trade wind patterns (easterlies) benefit the trans-portation of dust towards the Atlantic Ocean as seen also by the westward plumes in Fig. 1e and f.In the same period, increased dust occurrences (83 %) are also found over the western Mediterranean and South Italy.In the domain between 10 • W-00 • and 35-45 • N, the mean DODs are 0.09 ± 0.14 with 5 % of the dust observations having DODs > 0.55 and extreme DODs up to 2.3.
During OND dust activity is significantly suppressed (Fig. 1g) except from the southwest desert areas close to the Sahel where mean DOD lies in the range 0.2-0.3 (Fig. 1h).In the domain between 10 • W-00 • and 20-30 • N, the mean DODs are 0.43 ± 0.39 and extreme DODs up to 3 (Table 2).
In order to provide a more informative representation of the dust product, we performed a comparison with MODIS AOD for the same period, the dust optical depth of the MACC reanalysis for the period 2007-2012 and a RegCM4 simulation for the period 2007-2014 (Fig. 3).MODIS provides the AOD for all natural and anthropogenic aerosol types.As a result the MODIS average value for the whole period and domain (0.267) is 281 % bigger than our product (0.095 ± 0.04).It is noted though that the values between the two satellite products are very similar over the Sahara desert.On the contrary, the corresponding average dust optical depth values of MACC (0.100) and RegCM4 simulations (0.104) consider only dust and are in better agreement with our product, with lower values by 5 and 8.6 %, respectively.The 95 % confidence interval of the mean for MACC is between 0.092 and 0.108, and for RegCM4 it is between 0.099 and 0.108.Considering these ranges, the discrepancies between the CALIPSO dust product and the two models are within the combined uncertainty.
Dust optical depth is overestimated over Europe and the Mediterranean by MACC and RegCM4 simulations compared to our product, in all seasons and especially in the hot periods of AMJ and JJA.The reasons of these discrepancies have to be further studied.

Vertical dust distribution
CALIPSO offers the ability to assess the vertical distribution of dust from space.To facilitate the investigation of the vertical characteristics of dust, two parameters are introduced the dust top height (TH) and the dust center of mass height (CoM) (Mona et al., 2006(Mona et al., , 2014;;Binietoglou et al., 2015).TH is defined as the height corresponding to the altitude with 98 % of the dust extinction below it.CoM is the extinctionweighted altitude given by the formula where z b and z t are the base and top altitude of the dust feature, respectively, and α denotes the dust extinction coeffi- cient at altitude z.CoM provides a measure of the altitude where most of the dust load is located.This parameter is considered ideal for comparisons with aerosol layer height retrievals from passive remote sensing (e.g., IASI, GOME-2A, Sentinel5P and the future Sentinel-4 and Sentinel-5 missions; Ingmann et al., 2012), since these retrievals are sensitive to the location of the dust mass maximum within the layer (e.g., TROPOMI Aerosol Layer Height product; Sanders et al., 2015).
Figure 4 shows the spatial distribution of TH and CoM for the four seasons.In Table 2, the TH and CoM values above surface elevation (a.s.e.) are accompanied with their standard deviations providing an indication of the variability of the dust heights in the atmosphere of the study area.During JFM dust resides in general below 3 km a.s.e. over land, with CoM at about 1.3 ± 1.6 km a.s.e. (Fig. 4a, b).Over the sea, several transport paths are discernible especially over the eastern Mediterranean, with dust tops traveling at 2.3 ± 1.9 km a.s.e.Over the eastern parts of the Sahara during AMJ, TH and CoM are up to 4.2 ± 1.7 km and 2.4 ± 1.1 km a.s.e., respectively.Over the Mediterranean Sea and southern Europe, dust tops extend at ∼ 2-3.5 km and CoM is ∼ 1-2 km a.s.e., with the central and eastern Mediterranean having the most elevated plumes (Fig. 4c, d).The latitudinal slope of CoM denotes the latitudinal transport of dust during AMJ from south to north.The highest TH values (> 4.5 km) are found during the warm period (JAS) over northwestern Africa and over the adjacent Atlantic Ocean region (Fig. 4e, f).This is most likely attributed to the intrusion of the lower tropospheric Atlantic monsoon, south of the ITCZ, and the development of mesoscale convective systems that favor the elevation of dust at this area (Bou Karam et al., 2008).The dust height decreases towards the eastern part of the study region.In the interim, the dominance of the strong Saharan high enables the mobilization of dust from the western part of the Sahara towards the western Mediterranean and Europe.This pattern leads to elevated dust at 3.0 ± 1.7 km a.s.e. and CoM at 1.6 ± 1.1 km a.s.e. over southern European countries and the Balkans.During OND the horizontal pattern is similar to JJA, however, with much lower heights (Fig. 4g, h).
In general, our results are in agreement with lidar-based studies which have been performed in several European sites.Papayannis et al. (2008) performed an exhaustive analysis on Saharan dust particles over Europe using EARLINET lidar profiles.They found that the dust layer center of mass extends from 3.0 to 3.8 km and the thickness ranges from 0.7 to 3.4 km.Moreover, Balis (2012) calculated the mean base and top of dust layers in the eastern Mediterranean, Thessaloniki, to be around 2.5 ± 0.9 km and 4.2 ± 1.5 km, respectively.More recently, Mona et al. (2014) analyzed a long dataset of Saharan dust intrusions over Potenza, Italy, and found a mean layer center of mass of 3.5 ± 1.5 km.

Climatological dust cross sections
To further illustrate the vertical dynamics of dust reaching Europe, the area of study between 20 • W and 30 • E is separated into five longitudinal zones of 10 • , covering latitudes from 20 to 60 • N, and the results are presented as latitudeheight cross-section plots in Fig. 5, with the respective statistics in Table 3.The vertical structure of the averaged Climatological Dust Extinction coefficient (Clim-DE) for each of these five longitudinal zones reveals several dust layers and strong seasonal variations.The two dashed lines drawn in the cross-section plots show the number of dust observations averaged for the extinction retrievals.The extinction values below the higher dashed line correspond to > 18 dust observations (two dust overpasses per season and year), whereas below the lower dashed line they correspond to > 54 (two dust overpasses per month and year).The median surface elevation is depicted with black color (and is labeled as NaN) in the plots.
Figure 5 shows that dust is always ubiquitous at heights close to the surface throughout the year.The lower layers are representative of near-source dust activity and boundary layer processes.The spring and summer peaks indicate the increased activity of Saharan dust sources (Moulin et al., 1998;Schepanski et al., 2007).More specifically, for the area between 10 and 20 • W over the Atlantic, extending from Africa to the west of the Iberian Peninsula and Ireland (Fig. 5a-d), the presence of elevated dust plumes is evident mainly during summer and for latitudes up to 30 • N.During JFM the plume is located below 2 km height above sea level (a.s.l.), while from spring to autumn the plume reaches a height of 5 km a.s.l. and yields high values of extinction coefficient (∼ 75 Mm −1 ) over Africa.Over the area from 0 to 10 • W, extending from western Algeria, Morocco, the Iberian Peninsula and the British Isles, we found Clim-DE values inside the Africa mixing layer greater than 60 Mm −1 for all seasons.Maximum values of extinction are observed during summer months when dust is elevated up to 6 km with Clim-DE values around 120 ± 140 Mm −1 above N. Africa and mean values exceeding 200 Mm −1 above the Algerian Desert (Fig. 5g).These findings are in good agreement with more than 2 years of AERONET observations in the Tamanrasset site, a strategic site for dust research located in the heart of the Sahara (Guirado et al., 2014).A steep decrease in extinction values is observed along the African coastline, with values of 20 Mm −1 above the southern part of the Iberian Peninsula (38-42 • N) where dust is trapped by the Pyrenees.The distinct decrease of extinction values across the African coastline is an indication that dust is always present inside the rather deep Saharan boundary layer while it is only occasionally transferred towards the Mediterranean when atmospheric dynamics favor this kind of flow.At higher latitudes, the CALIPSO dust extinction is drastically reduced but still observed at 1-2 km a.s.l., with mean Clim-DE values of 5 Mm −1 .As discussed in detail in Sect.2.4, the uncertainty of the dust extinction values close to the surface and at high latitudes is < 54 %, with the higher uncertainty in this region mainly originating from the selection of the δ nd value during the dust separation step.Moreover, the standard deviation, coming from the natural variability of the dust events, is an order of magnitude higher than the mean values (Table 3).Moving eastwards (0-10 • E), elevated dust is trapped topographically by the Alps (47-52 • N), with values > 10 Mm −1 .As the dust-laden air-masses approach the mountains, they decelerate and their dust concentrations increase (Israelevich et al., 2012).Maximum values of extinction (> 50 Mm −1 ) are observed over northern Africa during summer (Fig. 5k).Close to the Algerian sources, south of the Atlas Mountains (∼ 30 • N), the extinction coefficient is greater than 200 Mm −1 close to the surface (Fig. 4k).The area south of the Atlas Mountains (Fig. 5e, f, g, h) is characterized by haboob activity (Knippertz et al., 2009;Solomos et al., 2012).These systems are generated from convective outflows and contribute to the interannual burden of dust at this area.As dust extends to higher latitudes (30-40 • N) Clim-DE decreases (< 75 Mm −1 ).Over the area between 10 and 20 • E (Fig. 5m-p), similar patterns are observed.This region includes the dust sources of Libya and the central Sahara, the central Mediterranean, the eastern Alps and part of northern Europe.It is evident from this figure that dust extinction over the central Mediterranean (35-45 • N) is around 25 Mm −1 throughout the year.As in the previous western zonal section, the same pattern over the Alps is encountered.Moving further eastwards, maximum values of Clim-DE are found during spring.At the most eastern part of the study area (20-30 • E; Fig. 5q-t), dust is trapped by the Carpathian Mountains (45-49 • N) especially during winter, highlighting once more the role of topography.Significant dust presence is evident all over the zonal section (until 60 • N) and is mostly attributable to elevated dust traveling along with the westerlies from western and central parts of Europe towards the east.Above the Balkans and during JFM, values of 29 ± 65 Mm −1 are observed in the first 1.5 km and 10 ± 30 Mm −1 between 2.5 and 3.5 km.In AMJ and JAS, respectively, mean values of ∼ 16 ± 40 Mm −1 and ∼ 9 ± 20 Mm −1 are observed at altitudes between 1.5 and 5 km.Over Africa during winter and spring, the values of Clim-DE are higher (> 45 Mm −1 ) compared to the ones observed during the other two seasons (< 45 Mm −1 ), reaching high altitudes (5-6 km a.s.l.) during spring and summer.In summary, the obtained cross sections for the five longitudinal zones indicate that higher extinction coefficient values are observed near the source and at low altitudes, where dust particles are efficiently deposited.
The above results are representative of the spatial distribution of dust load as this is approximated by the aerosol extinction coefficient.In order to provide the dust load in units that are more relevant for modeling studies, we estimate here the dust mass concentration.The mass concentration can be obtained from the optical properties of dust with an uncertainty of 20-30 % (Ansmann et al., 2012;Mamouri and Ansmann, 2014).For example, the Clim-DE values correspond to dust mass concentrations > 75 µg m −3 above Africa  -p) and 20 to 30 • E (q-t) for the latitudinal regions from 10 to 60 • N as illustrated by domain maps for the 3-month averages: January-March (a, e, i, m, q), April-June (b, f, j, n, r), July-September (c, g, k, o, s) and October-November (d, h, l, p, t).The median surface elevation is depicted with black color.The decreasing intensity with height and latitude of Clim-DE is representative of the average dust distribution over the area.However, this behavior is not representative of the distribution during dust episodes over Europe.This is because the extinction coefficient values presented in Fig. 5 for the Clim-DE product are produced by averaging both par-tially and fully dominated dust cases.Conversely, the Con-DE product presented in the next section describes the spatial patterns and the intensity of the dust plumes during the dust episodes only.

Conditional dust cross sections
The two dashed lines in Fig. 6 correspond to > 18 dust observations (two dust overpasses per season and year) and > 54 dust observations (two dust overpasses per month and year) for the lower and higher dashed line, respectively.Con-DE values derived from less than 4 dust observations (dO) in each cell are masked with grey color (and are labeled as < 4dO) in the plots.The median surface elevation is depicted with black color (same as in Fig. 5).Con-DE values are significantly different from the Clim-DE, as seen in Fig. 5.Although Con-DE has similar values to Clim-DE near the sources, where dust is always present, above the Atlantic and the Mediterranean Con-DE is characterized by significantly higher values.This is expected, since the two products differ mostly over areas which are not dominated by dust.
In the vertical cross-section plots of Fig. 6 the patterns of Con-DE show two distinct dust features: over the longitudinal zone from 20 to 30 • E during summer (Fig. 6o) a distinctive feature is seen above North Africa extending from the surface to ∼ 5 km a.s.l. and a second feature is seen above the Mediterranean between 3 and 6 km a.s.l.The two distinct layers are also identified in other regions and in other seasons (e.g., Fig. 6a, l, p, s, t).These populations are linked to two different processes: the near-surface dust at the southern parts of the study region is connected to fresh emissions from the dust sources, while the elevated plumes that extend north until 40 • N are due to the advection of dust, associated to the seasonality of the long-range transport paths (Lelieveld et al., 2002;Israelevich et al., 2012;Huneeus et al., 2016).This separation is enhanced as one moves from the west to the east sectors.At the western part of the domain (10-20 • W) the near-surface and elevated dust probably originates from the same sources.Similar double layer patterns are found in all seasons and over all areas with various characteristics.For example, during JAS at the region extending from 0 to 10 • W (Fig. 6g), the generation of dust from the source region is much more intense than the transportation of dust.For the same period, in the area 0 to 10 • E, the dust transportation above the Mediterranean between 3 and 6 km height, originating from the intense source regions, becomes much more important than the source emissions.
Moreover, the vertical cross-section plots in Fig. 6 show the rare but very intense elevated dust plumes during JFM (Fig. 6a, e, i, m).During that period, dust is advected between 1.5 and 4 km height a.s.l. with Con-DE values > 45 Mm −1 , equivalent to dust mass concentrations > 75 µg m −3 .The intensity of the JFM dust episodes above the Balkans is also depicted (Fig. 6q): the Con-DE value is similar to other regions, but the dust plumes can be thicker, extending from the ground until 4 km a.s.l.The trapping of Saharan dust from the mountainous ridges of Europe (located between 40 and 50 • N, e.g., the Alps 45-48 • N) is also evident (e.g., Fig. 6i,  m).The deceleration of the transport air masses along the mountain ridges results in the accumulation of dust at the windward slopes.Dry deposition of dust at these areas also result in the formation of "brown snow" and albedo reduction, with profound climatological implications (e.g., Fujita, 2007;Shahgedanova et al., 2013).This phenomenon is more intense during the JFM period due to the advection of dust at lower heights.
During AMJ (Fig. 5b, f, j, n, r) and JAS (Fig. 5c, g, k, o, s) the elevated dust above the Mediterranean presents Con-DE values of 35-50 Mm −1 (58-83 µg m −3 ), in heights between 2 and 6 km and up to latitudes of 40 • N. The transport of dust during AMJ is mostly due to the eastward propagation of N. Africa-Mediterranean low pressure systems (Sharav cyclones).Dust is embedded in the cyclonic circulation and the penetration to latitudes higher than 40 • N is limited.For latitudes 40 and 50 • N during the warm seasons (AMJ and JAS), the Con-DE values inside the transported dust plumes are between 20 and 40 Mm −1 (33-67 µg m −3 ).Rare events, characterized by relatively higher Con-DE (> 35 Mm −1 and > 58 µg m −3 ) between 2 and 5 km a.s.l., are observed over the British Isles and Germany during OND (Fig. 6h, l).These events, caused by the propagating low pressure systems over the east Atlantic, have been documented in detail from the EARLINET community reporting extinction coefficient values up to 200 Mm −1 inside dust plumes (Ansmann et al., 2003;Müller et al., 2003).In the vertical cross-section plots of Fig. 6, it is evident that dust reaches the upper levels of the troposphere (> 8 km a.s.l.), with Con-DE values of ∼ 10 Mm −1 in all longitudinal zones and during all seasons.Dust occurrence is very low, close to zero for heights greater than 8 km a.s.l.during spring and summer and for heights greater than 6 km a.s.l.during autumn and winter.A quantitative representation of the Clim-DE and Con-DE products is provided in Table 3.In this, regional statistics of the two products, along with their standard deviation, are provided for three altitudinal ranges (0-2, 2-4 and 4-6 km a.s.l.).

Interannual variability of dust
In this section we present the CALIPSO-derived monthly mean DOD values, for the total column and for five individual layers (0.18-0.5, 0.5-1, 1-2, 2-4 and 4-8 km), in order to study their interannual variability during the 9-year period between 2007 and 2015.The selected layers are representative for both near-surface and long-range transported dust plumes.The data are aggregated on a 10 • × 10 • cell over the study region.Using a first-order autoregressive linear regression model on the deseasonalized monthly DOD values (108 in total) as described in Zanis et al. (2006), temporal trends of DOD were calculated.We note that 9 years are considered a small period for a robust trend calculation  -p) and 20 to 30 • E (q-t) for the latitudinal regions from 10 • N to 60 • N as illustrated by domain maps for the 3-month averages: January-March (a, e, i, m, q), April-June (b, f, j, n, r), July-September (c, g, k, o, s) and October-November (d, h, l, p, t).The median terrain elevation is depicted with black color.and it would be interesting to extend this analysis with future measurements.Figure 7   the trends are not statistically significant.The small negative DOD trends (< 0.002 yr −1 corresponding to < 5 % yr −1 ) are in good agreement with the global decrease of dust estimated from a 161-year time series of dust from 1851 to 2011, created by projecting wind field patterns onto surface winds from a historical reanalysis in Evan et al. (2016).The agreement is also good with the global mean near-surface dust concentration decrease of 1.2 % yr −1 reported in Shao et al.'s (2013) (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), with the trends being statistically significant at the 95 % confidence level in both cases.A possible increase is only found for the western Sahara areas (10-0 • W, 20-30 • N).However, the results for this cell are not statisti-cally significant.Figure 8 shows the DOD interannual variability of the 20 individual areas, as it is calculated from the monthly mean DODs.It is evident from this figure that the DOD values in 2008 are relatively higher than the other years and in almost all the domains below 40 • N. Similarly, relatively high values are observed in some of these areas for the year 2010.Since these years are at the beginning of our study period, they have a significant contribution on the negative trends observed during the examined period.

Summary and conclusions
An optimized CALIPSO dust product was recently developed by Amiridis et al. ( 2013) using a regional correction for the Saharan dust lidar ratio derived from EARLINET measurements.The same product is used here to provide the three-dimensional dust distribution and its transport pathways across northern Dust transport mechanisms are more efficient during summer when dust is often lifted up to 6 km, coinciding with the deepest dust layer.Localized regions of increased extinction coefficient values over mountains (the Alps, the Pyrenees and the Carpathian Mountains) trace the aerosol transport routes that decelerate in front of the mountain ranges.Rare and intense events are observed above the Balkans during the winter period and above northwest Europe during autumn.The interannual analysis revealed that DOD trends during the study period are of the order of 0.001 yr −1 for southern Europe, showing constant decrease throughout the different layers.
The dust climatology presented here is of paramount importance in understanding the three-dimensional production and transport of Saharan dust, providing valuable information for better estimations of the dust climatic impacts.The climatological and conditional products presented here describe both the dust contribution to the total aerosol load over our domain as well as the Saharan dust events recorded in the region, respectively.Future work includes (i) the optimization of CALIPSO dust retrievals based on measured dust LR from ground-based lidars and particle depolarization ratio over extended regions of deserts in the Middle East and China, to obtain a robust global climatology of dust; and (ii) the calculation of cloud condensation nuclei (CCN) and ice nuclei (IN) concentrations from polarization lidar measurements as suggested by Mamouri and Ansmann (2016), to provide a quantification of the climatic effect of dust on cloud formation.
Competing interests.The authors declare that they have no conflict of interest.

Figure 1 .
Figure 1.Geographical distribution of the seasonal dust occurrences (a, c, e, d) and the mean DOD values (b, d, f, h) for the 3-month averages: January-March (a, b), April-June (c, d), July-September (e, f), and October-November (g, h), and the domain between 20 • W-30 • E and 20-60 • N for the period 2007-2015, measured with the CALIPSO climatological dust product.

Figure 2 .
Figure2.The six domains of the regional statistics provided on Tables2 and 3.

Figure 3 .
Figure 3.Comparison of the seasonal spatial distribution of the optical depth as received by (first column) pure-dust CALIPSO DOD product, (second column) MODIS AOD product, (third column) MACC reanalysis DOD product, (fourth column) RegCM4-simulated DOD product.

Figure 4 .
Figure 4. Geographical distribution of the dust top height (a-d) and the center of mass (e-h) in km a.s.e.measured with CALIPSO dust product for the 3-month averages: January-March (a, e), April-June (b, f), July-September (c, g), and October-November (d, h), and the domain between 20 • W-30 • E and 20-60 • N for the period 2007-2015.

Figure 5 .
Figure 5. Geographical zonal distribution of the climatological dust extinction coefficient values (Mm −1 ) measured by the CALIPSO dust product for the regions 10 to 20 • W (a-d), 0 to 10 • W (e-h), 0 to 10 • E (i-l), 10 to 20 • E (m-p) and 20 to 30 • E (q-t) for the latitudinal regions from 10 to 60 • N as illustrated by domain maps for the 3-month averages: January-March (a, e, i, m, q), April-June (b, f, j, n, r), July-September (c, g, k, o, s) and October-November (d, h, l, p, t).The median surface elevation is depicted with black color.
throughout the year and > 125 µg m −3 above West Africa during JAS.In southern Europe and the Mediterranean, the corresponding values are > 17 µg m −3 in the first 2 km a.s.l. and ∼ 50 µg m −3 close to the surface.For latitudes greater than 45 • N, values of ∼ 8 µg m −3 are the most common.

Figure 6 .
Figure 6.Geographical zonal distribution of the conditional dust extinction coefficient values (Mm −1 ) measured by CALIPSO dust product for the regions 10 to 20 • W (a-d), 0 to 10 • W (e-h), 0 to 10 • E (i-l), 10 to 20 • E (m-p) and 20 to 30 • E (q-t) for the latitudinal regions from 10 • N to 60 • N as illustrated by domain maps for the 3-month averages: January-March (a, e, i, m, q), April-June (b, f, j, n, r), July-September (c, g, k, o, s) and October-November (d, h, l, p, t).The median terrain elevation is depicted with black color.
shows the geographical distribution of deseasonalized trends (yr −1 ) for the columnar DOD (a) and for the five individual layers (b-f).Hatched filled grid cells depict the statistical significance trends with 99 % confidence.A decrease of ∼ 0.001 yr −1 (∼ 4 % yr −1 ) is evident for the southern European cells (0-30 • E, 40-50 • N) (with these values being > 95 % statistically significant).Examination of the five vertical layers shows a similar decreasing pattern.The negative trends observed in the area (mainly above North Africa and the Mediterranean) show a constant decrease throughout the layers as well, although

Figure 7 .
Figure 7. Geographical distribution of the deseasonalized trends (yr −1 ) derived from monthly columnar DOD (a) and for five individual layers (b-f), for the period 2007-2015, aggregated over 10 • × 10 • grid cells.Hatched filled grid cells depict the statistical significance trends with 99 % confidence.
Africa and Europe from 2007 to 2015.The study of the mean state climatology shows strong seasonal shifts in dust source regions and transportation pathways.The seasonal cycle of the dust transport is well captured with the lowest values of DOD in winter and the highest values in spring and summer.During summer and autumn, dust aerosols are mostly confined to the source region, while during spring dust aerosols from the Sahara are extended uniformly over the northern Sahara and are transported over the Mediterranean and Europe.The dust extinction coefficient, CoM and TH parameters are used to quantitatively describe the 3-D evolution of dust and its seasonal variations.Over the source region of the Sahara Desert, dust CoM and TH are higher during spring and summer and lower during winter.
. Typical dust δ d values measured with lidars in field campaigns around the globe are generally consistent with this

Table 2 .
Regional statistics on mean dust optical depth, max values, dust layer center of mass (CoM) and top height (TH) (a.s.l.), ratio of dust observations to cloud-free observations, ratio of cloud-free observations to total observations and domain boundaries.DOD Mean ±SD DOD Max Vals.(Perc.95 %) CoM ±SD Top Height ±SD Nr Dst in Nr cl-free Nr cl-free in Nr obs.Domain (C-E Med.), whereas above the central-western Mediterranean (C-W Med.

Table 3 .
Regional statistics on the dust extinction coefficient for altitudes between 0 and 2 km, 2 and 4 km and 4 and 6 km (a.s.l.).