The Potential of GRASP/GARRLiC Retrievals for Dust Aerosol Model Evaluation: Case Study during the PreTECT Campaign

: We use the Generalized Retrieval of Aerosol Surface Properties algorithm (GRASP) to compare with dust concentration proﬁles derived from the NMME-DREAM model for a speciﬁc dust episode. The GRASP algorithm provides the possibility of deriving columnar and vertically-resolved aerosol properties from a combination of lidar and sun-photometer observations. Herein, we apply GRASP for analysis of a Saharan dust outburst observed during the “PREparatory: does dust TriboElectriﬁcation affect our ClimaTe” campaign (PreTECT) that took place at the North coast of Crete, at the Finokalia ACTRIS station. GRASP provides column-averaged and vertically resolved microphysical and optical properties of the particles. The retrieved dust concentration proﬁles are compared with modeled concentration proﬁles derived from the NMME-DREAM dust model. To strengthen the results, we use dust concentration proﬁles from the POlarization-LIdar PHOtometer Networking method (POLIPHON). A strong underestimation of the maximum dust concentration is observed from the NMME-DREAM model. The reported differences between the retrievals and the model indicate a high potential of the GRASP algorithm for future studies of dust model evaluation.


Introduction
Desert dust is one of the most dominant aerosol types on Earth, produced in arid regions. Windblown dust particles can have a significant climate impact since they can cover great distances from from their sources, directly affecting the planetary radiative balance through their interaction with solar and terrestrial radiation [1] and indirectly influencing cloud formation and cloud optical properties [2]. Dust also affects the prevailing atmospheric conditions and air quality in the transported areas (i.e., the Mediterranean Sea and Europe), thus affecting the health of large populations, especially when synoptic conditions favor its advection within the boundary layer [3][4][5][6].
One of the largest sources of dust particles is the Sahara region, which transports dust throughout the year towards the Mediterranean Sea. The dust aerosols strongly

The PreTECT Campaign and the Site of Finokalia
The PreTECT experimental campaign (http://pre-tect.space.noa.gr/, accessed on 29 December 2020) was held at the northern coast of Crete, Greece, at the Finokalia research station (35 • 20 N, 25 • 40 E), during April 2017. A large consortium of European research institutes and universities participated in this combined effort, was organized by the National Observatory of Athens (NOA). The Finokalia station is located at the top of a hilly elevation (252m above sea level (a.s.l.)) facing the sea, far from human activities (more than 20 km distance). The site is representative of the regional conditions of the eastern Mediterranean [32] and is very often affected by long-range transported mineral dust from north Africa. Other aerosol types that are frequently observed are marine and smoke particles [33]. The PreTECT campaign is mainly focused on desert dust characterization, uses remote sensing measurements and applies advanced inversion techniques to derive the microphysical and optical properties of the particles.

Instruments and Tools
During the PreTECT campaign a large suite of remote sensing sensors was deployed at Finokalia, along with the in-situ instrumentation that regularly operates at the station. Unfortunately, for the case study presented here, no in-situ data were available. The instruments and tools used herein are described in detail below.

PollyXT Lidar
The PollyXT lidar [34] of NOA is a multi-wavelength system that operated on an all day and night monitoring basis during the campaign, providing vertically resolved aerosol optical properties. PollyXT-NOA employs an Nd:YAG laser emitting linearly-polarized light at the primary wavelength of 1064 nm. The system is equipped with three elastic backscatter channels (at 355, 532 and 1064 nm), two rotational-vibrational Raman extinction channels (at 387 and 607 nm), two linear depolarization channels (at 355 and 532 nm), and one water vapor detection channel (at 407 nm). The receiver part of PollyXT-NOA is composed of a far-range and a near-range receiver with a full overlap height at~200 m above ground, decreasing the minimal measurement height in the planetary boundary layer. The PollyXT-NOA lidar is part of the European Aerosol Research LIdar NETwork (EARLINET) [35], and the PollyNET network (portable Raman lidar systems) [36]. The derived products meet the EARLINET guidelines for data quality assurance [37].

CIMEL Sun-Photometer
The CIMEL sun-photometer installed at the Finokalia station during the campaign, is part of the Aerosol Robotic Network (AERONET) [38] and was operated by NOA. The direct sun and sky measurements provide the aerosol optical depth (AOD) at eight wavelengths from 340 to 1640 nm, along with the products derived from the AERONET inversion algorithm [11,39]: the columnar volume size distribution, complex refractive index, single scattering albedo and the scattering phase function.

GRASP/GARRLiC Retrieval Algorithm
As briefly mentioned in the introduction, the main retrieval algorithm used in this study, the GRASP/GARRLiC inversion algorithm [18], combines lidar with sun-photometer measurements and provides the column-integrated and vertically resolved optical and microphysical properties of the aerosol particles in the atmosphere. GRASP/GARRLiC input consists of the sun-photometer sun and sky measurements and the elastic backscatter lidar signals, while different combinations of inputs/outputs are supported (e.g., [20,40]). As shown in Figure 1, in the current study we use as input the lidar measurements of elastic backscatter signals at 355, 532 and 1064 nm, and the sun-photometer measurements of AOD and of diffused light (up to 35 scattering angles) at 440, 670, 870 and 1020 nm. To extend the lidar profiles as close to the ground as possible, we combined the information from lidar far-range and near-range receivers. Specifically, a gluing technique [41] on the lidar signal level was applied, extending the profiles at ∼200 m above the ground.
we combined the information from lidar far-range and near-range receivers. Specifically, a gluing technique [41] on the lidar signal level was applied, extending the profiles at ~200 m above the ground.
The acquired optical and microphysical properties are derived for both fine and coarse mode particles. These are the aerosol volume concentration profiles and columnar microphysical and optical properties: the single scattering albedo (SSA), the particle size distribution, the real and imaginary part of refractive index (RRI and IRI, respectively) and the particle spherical fraction. GRASP/GARRLiC provides the possibility of a singlemode inversion as well for events dominated by a certain type of aerosol, as is the case in the present study. Several conditions are required for the successful application of the GRASP/GARRLiC retrieval algorithm. The daytime sun-photometer data should be obtained in cloud-free atmospheric conditions. Moreover, it is preferable to use early-day measurements in order to avoid the sun background in lidar measurements. Moreover, high aerosol loading (AOD at 440 nm of 0.4 or greater) is desirable for reliable retrievals of aerosol absorption. Moreover, the retrieval uses the assumption of a plane-parallel atmosphere, and therefore only one averaged dataset of lidar profiles should be used. Furthermore, in the situations when one type of aerosol dominates the measurements, the default retrieval of two aerosol modes can be ambiguous and an assumption of a single aerosol mode is more appropriate [23].

POLIPHON Retrieval Algorithm
POLIPHON [42] uses single-wavelength lidar observations with one polarizationsensitive channel, and as such the method is simple and robust. POLIPHON performs the vertical separation of two to three particle components in different aerosol mixtures, deriving their optical properties and mass concentrations. The POLIPHON approach is designed explicitly to avoid the use of a particle shape model and a strong dependence on photometer observations, providing, however, a less complete suite of optical and microphysical products than GRASP/GARRLiC. The retrieval uses the particle linear depolarization ratio that separates the non-spherical particle contribution to the particle backscat- The acquired optical and microphysical properties are derived for both fine and coarse mode particles. These are the aerosol volume concentration profiles and columnar microphysical and optical properties: the single scattering albedo (SSA), the particle size distribution, the real and imaginary part of refractive index (RRI and IRI, respectively) and the particle spherical fraction. GRASP/GARRLiC provides the possibility of a single-mode inversion as well for events dominated by a certain type of aerosol, as is the case in the present study.
Several conditions are required for the successful application of the GRASP/GARRLiC retrieval algorithm. The daytime sun-photometer data should be obtained in cloud-free atmospheric conditions. Moreover, it is preferable to use early-day measurements in order to avoid the sun background in lidar measurements. Moreover, high aerosol loading (AOD at 440 nm of 0.4 or greater) is desirable for reliable retrievals of aerosol absorption. Moreover, the retrieval uses the assumption of a plane-parallel atmosphere, and therefore only one averaged dataset of lidar profiles should be used. Furthermore, in the situations when one type of aerosol dominates the measurements, the default retrieval of two aerosol modes can be ambiguous and an assumption of a single aerosol mode is more appropriate [23].

POLIPHON Retrieval Algorithm
POLIPHON [42] uses single-wavelength lidar observations with one polarizationsensitive channel, and as such the method is simple and robust. POLIPHON performs the vertical separation of two to three particle components in different aerosol mixtures, deriving their optical properties and mass concentrations. The POLIPHON approach is designed explicitly to avoid the use of a particle shape model and a strong dependence on photometer observations, providing, however, a less complete suite of optical and microphysical products than GRASP/GARRLiC. The retrieval uses the particle linear depolarization ratio that separates the non-spherical particle contribution to the particle backscatter coefficient for a given wavelength. The AERONET observations are only used to provide with the aerosol-type-specific "mass-specific extinction coefficients" used in POLIPHON for the calculation of the mass concentration profiles.
For the dust case, the discrimination is performed for three components (POL-2 version) [14]: dust coarse, dust fine and non-dusty aerosols. For the case study herein, the latter component is expected to include marine and anthropogenic pollution.

NMME-DREAM Dust Model
NMME-DREAM is the dust model evaluated in this study. It consists of the meteorological core NCEP-NMME (Nonhydrostatic Mesoscale Model on E-grid) atmospheric model [43]. DREAM [44][45][46] is a numerical model developed to simulate and predict the atmospheric life cycle of mineral dust including dust emission, dust horizontal and vertical turbulent mixing, long-range transport and dust deposition, using an Euler-type nonlinear partial differential equation for dust mass continuity. The model is configured at 0.2 • × 0.2 • resolution and includes 8 dust size bins with effective radius of 0.15, 0.25, 0.45, 0.78, 1.3, 2.2, 3.8 and 7.1 µm respectively. In the present study, 3 hourly instantaneous model outputs are used.

Characterization of the Dust Event of 14 May 2017
The dust event that took place at Finokalia on 14 May 2017 is shown in Figure 2a presents the total attenuated backscatter coefficient from the lidar measurements at 1064 nm, depicting the vertical extend of the plume that reaches up to ∼5 km, while there is also a thin layer at ∼6-7 km. The volume linear depolarization ratio (VLDR) values at 532 nm exceed 15%, which indicates the presence of non-spherical particles from ∼0.5 up to 5 km. Lower VLDR values below 500 m can be attributed to a possible mixture of dust with marine particles. The AERONET product also indicates the presence of dust particles, with a high aerosol load (AOD ≈ 0.7 at 440 nm) and an Angstrom exponent at 440/870 nm of ∼0.14, with the latter within the range of the climatological values for desert dust [47].
For the dust case, the discrimination is performed for three components (POL-2 version) [14]: dust coarse, dust fine and non-dusty aerosols. For the case study herein, the latter component is expected to include marine and anthropogenic pollution.

NMME-DREAM Dust Model
NMME-DREAM is the dust model evaluated in this study. It consists of the meteorological core NCEP-NMME (Nonhydrostatic Mesoscale Model on E-grid) atmospheric model [43]. DREAM [44][45][46] is a numerical model developed to simulate and predict the atmospheric life cycle of mineral dust including dust emission, dust horizontal and vertical turbulent mixing, long-range transport and dust deposition, using an Euler-type nonlinear partial differential equation for dust mass continuity. The model is configured at 0.2° × 0.2° resolution and includes 8 dust size bins with effective radius of 0.15, 0.25, 0.45, 0.78, 1.3, 2.2, 3.8 and 7.1 μm respectively. In the present study, 3 hourly instantaneous model outputs are used.

Characterization of the Dust Event of 14 May 2017
The dust event that took place at Finokalia on 14 May 2017 is shown in Figure 2a presents the total attenuated backscatter coefficient from the lidar measurements at 1064 nm, depicting the vertical extend of the plume that reaches up to ~5 km, while there is also a thin layer at ~6-7 km. The volume linear depolarization ratio (VLDR) values at 532 nm exceed 15%, which indicates the presence of non-spherical particles from ~0.5 up to 5 km. Lower VLDR values below 500 m can be attributed to a possible mixture of dust with marine particles. The AERONET product also indicates the presence of dust particles, with a high aerosol load (AOD ≈ 0.7 at 440 nm) and an Angstrom exponent at 440/870 nm of ~0.14, with the latter within the range of the climatological values for desert dust [47]. To determine the origin of the particles, we perform back-trajectory analysis using the atmospheric dispersion model FLEXPART-WRF [48]. The 24-h backtrajectories ( Figure  3a) show that the air masses arriving at 3 to 10 km above Finokalia, have followed a westerly flow originating from the Sahara region (Tunisia and Algeria), while in the origins for the air-masses arriving below 2 km, sources include the Ionian Sea and the northern part of Greece, suggesting possible contributions from marine and anthropogenic aerosols. The forecast of the NMME-DREAM dust model on 12 May 2017 (Figure 3b) confirms the origin of mineral dust from Algeria and Tunisia and the transport path through the Mediterranean.
(a) To determine the origin of the particles, we perform back-trajectory analysis using the atmospheric dispersion model FLEXPART-WRF [48]. The 24-h backtrajectories (Figure 3a) show that the air masses arriving at 3 to 10 km above Finokalia, have followed a westerly flow originating from the Sahara region (Tunisia and Algeria), while in the origins for the air-masses arriving below 2 km, sources include the Ionian Sea and the northern part of Greece, suggesting possible contributions from marine and anthropogenic aerosols. The forecast of the NMME-DREAM dust model on 12 May 2017 (Figure 3b) confirms the origin of mineral dust from Algeria and Tunisia and the transport path through the Mediterranean.

Retrieval of Dust Properties
We characterized the dust plume on 14 May 2017 with the GRASP/GARRLiC retrieval, using the lidar measurements at 05:00-06:00 UTC (red box in Figure 2), and the sun-photometer measurements at 05:52 UTC. The AERONET volume size distribution (black line in Figure 4) exhibits a predominant coarse mode, and thus the single-mode inversion (i.e., without retrieving separated values for fine and coarse modes) is performed.
Blue lines in Figures 4-6 present the retrieved aerosol columnar microphysical properties from GRASP/GARRLiC (i.e., volume size distribution, refractive index and SSA), all in comparison with the AERONET retrievals (black lines) for this case study. The error bars in the plots denote the variability of the GRASP/GARRLiC retrieval for different input settings, and not the retrieval errors. The GRASP/GARRLiC and AERONET retrievals show good agreement for the size distribution of large dust particles (Figure 4), but not for the smaller ones, which are almost absent in the AERONET retrieval.

Retrieval of Dust Properties
We characterized the dust plume on 14 May 2017 with the GRASP/GARRLiC retrieval, using the lidar measurements at 05:00-06:00 UTC (red box in Figure 2), and the sunphotometer measurements at 05:52 UTC. The AERONET volume size distribution (black line in Figure 4) exhibits a predominant coarse mode, and thus the single-mode inversion (i.e., without retrieving separated values for fine and coarse modes) is performed.
Remote Sens. 2021, 13, x FOR PEER REVIEW sun-photometer measurements at 05:52 UTC. The AERONET volume size dist (black line in Figure 4) exhibits a predominant coarse mode, and thus the sing inversion (i.e., without retrieving separated values for fine and coarse modes formed. Blue lines in Figures 4-6 present the retrieved aerosol columnar microphysic erties from GRASP/GARRLiC (i.e., volume size distribution, refractive index and in comparison with the AERONET retrievals (black lines) for this case study. T bars in the plots denote the variability of the GRASP/GARRLiC retrieval for diffe put settings, and not the retrieval errors. The GRASP/GARRLiC and AERONET re show good agreement for the size distribution of large dust particles (Figure 4), for the smaller ones, which are almost absent in the AERONET retrieval. The real part of the refractive index (RRI, Figure 5a) provided by GRASP/GA agrees well with the AERONET values for wavelengths >670 nm, while for smalle lengths the first is much lower than the latter. The range of the RRI is ~1.4-1.45, at the low end of the climatological values for dust [47]. These values indicate tha probably mixed with marine particles, with a typical value for RRI of ~1.36, as d also in [23]. The same holds true for the values of the imaginary part of refractiv in the plots denote the variability of the GRASP/GARRLiC retrieval for different input settings, and not the retrieval errors. The GRASP/GARRLiC and AERONET retrievals show good agreement for the size distribution of large dust particles (Figure 4), but not for the smaller ones, which are almost absent in the AERONET retrieval. The real part of the refractive index (RRI, Figure 5a) provided by GRASP/GARR agrees well with the AERONET values for wavelengths >670 nm, while for smaller wa lengths the first is much lower than the latter. The range of the RRI is ~1.4-1.45, whic at the low end of the climatological values for dust [47]. These values indicate that du probably mixed with marine particles, with a typical value for RRI of ~1.36, as discus also in [23]. The same holds true for the values of the imaginary part of refractive in (CRI, Figure 5b), which ranges from 0.003 to 0.002 from the UV to the Near-IR. The SSA increases with increasing wavelength, from 0.92 at 440 nm to 0.98 at 1 nm, and is in good agreement with the AERONET retrieval ( Figure 6). These values also indicative of the presence of dust particles [47,49].  Figure 7 shows the vertical profiles of the particle backscatter and ext cients at 355 and 532 nm retrieved from lidar measurements with the Klet (blue and green dashed lines for 355 and 532 nm, respectively), a GRASP/GARRLiC algorithm (blue and green solid lines for 355 and 532 nm, For the Klett retrievals, we used a lidar ratio of 67.5 sr and 56 sr at 355 and 53 tively, so that the integral of the retrieved extinction coefficient profiles clos the AOD measured by the sun-photometer at 340 and 500 nm. Moreover glued signals from lidar far-range and near-range receivers. The resulting p tion and backscatter coefficient profiles retrieved with the Klett GRASP/GARRLiC are found to be in reasonable agreement, with the GRA values being slightly smaller, in the order of ~3% compared to the values r The real part of the refractive index (RRI, Figure 5a) provided by GRASP/GARRLiC agrees well with the AERONET values for wavelengths >670 nm, while for smaller wavelengths the first is much lower than the latter. The range of the RRI is ∼1.4-1.45, which is at the low end of the climatological values for dust [47]. These values indicate that dust is probably mixed with marine particles, with a typical value for RRI of ∼1.36, as discussed also in [23]. The same holds true for the values of the imaginary part of refractive index (CRI, Figure 5b), which ranges from 0.003 to 0.002 from the UV to the Near-IR.
The SSA increases with increasing wavelength, from 0.92 at 440 nm to 0.98 at 1020 nm, and is in good agreement with the AERONET retrieval ( Figure 6). These values are also indicative of the presence of dust particles [47,49]. Figure 7 shows the vertical profiles of the particle backscatter and extinction coefficients at 355 and 532 nm retrieved from lidar measurements with the Klett method [50] (blue and green dashed lines for 355 and 532 nm, respectively), and with the GRASP/GARRLiC algorithm (blue and green solid lines for 355 and 532 nm, respectively). For the Klett retrievals, we used a lidar ratio of 67.5 sr and 56 sr at 355 and 532 nm, respectively, so that the integral of the retrieved extinction coefficient profiles closely reproduce the AOD measured by the sun-photometer at 340 and 500 nm. Moreover, we used the glued signals from lidar far-range and near-range receivers. The resulting particle extinction and backscatter coefficient profiles retrieved with the Klett method and GRASP/GARRLiC are found to be in reasonable agreement, with the GRASP/GARRLiC values being slightly smaller, in the order of ∼3% compared to the values retrieved with the Klett method. These differences may arise from the different assumptions of the profile behavior near the ground for the two methods: GRASP/GARRLiC assumes it to be constant at the lowest 200 m, which is not the case for the Klett method. The GRASP/GARRLiC particle extinction and backscatter coefficient profiles are much smoother due to the reduced resolution of the lidar signals, which is suggested for reducing lidar signal noise levels and the complexity of the retrieval [18].

Comparison between Retreived and Simulated Dust Concentration Profiles
In Figure 8 the aerosol mass concentration profile retrieved with GRASP/GARRLiC (blue line) is compared with the corresponding profile simulated with the NMME-DREAM model (black line). For the conversion of the volume concentration profile provided by GRASP/GARRLiC, to the corresponding mass concentration profile, we consider that all particles have a density of 2.35 g/cm 3 , which is the mean of the 2.1-2.6 g/cm 3 density range provided in the literature for dust [51]. The error bars in the GRASP/GARRLiC mass concentration profile in Figure 8 denote this density range.

Discussion
The comparison between GRASP/GARRLiC retrievals and the NMME-DRE model dust concentration profiles suggests an under-prediction of the actual dust con trations in the model for the dust episode under study. The poor representation o boundary layer of the dust event is probably due to the low resolution of the model The NMME-DREAM model predicts two concentration peaks in the dust plume: one at 3.5 km with a concentration of ∼130 µg/m 3 and one at 2 km with a concentration of ∼250 µg/m 3 , revealing the models' biases when compared with the GRASP/GARRLiC retrievals for this specific episode. The mass concentration values retrieved by GRASP/ GARLLiC from the surface up to 2 km are strongly underestimated by the model; GRASP/ GARRLiC retrieval presents a maximum concentration of 300 µg/m 3 at ∼1.5 km, while the NMME-DREAM model profile quickly declines from the max concentration of ∼250 µg/m 3 at 2 km to low concentration values of 50 µg/m 3 closer to the surface. The model's behavior of underestimating dust concentrations at low altitudes is consistent with the findings of previous studies [10,23,30]. At higher layers (above 2 km) dust concentration decreases with height up to 5km, while an optically thin layer located between 5.5 and 7 km is also depicted by GRASP/GARRLiC retrieval. The model overestimates concentrations between 1.5 to 3 km and underestimates them between~3 to 5 km height.

Discussion
The comparison between GRASP/GARRLiC retrievals and the NMME-DREAM model dust concentration profiles suggests an under-prediction of the actual dust concentrations in the model for the dust episode under study. The poor representation of the boundary layer of the dust event is probably due to the low resolution of the model that makes it difficult to capture the topography in detail with the continental mountainous ranges. These features may play an important role in the model's dynamics and the downward winds that have a very local nature and can cause mixing of dust concentrations induced by the complex topography [10].
To strengthen the above results, the mass concentration profiles simulated with a NMME-DREAM model and retrieved with GRASP/GARRLiC are compared with the mass concentration profiles retrieved with POLIPHON ( Figure 9) for continental (green dashed line), dust (magenta dashed line) and marine particles (blue dashed line). For the POLIPHON retrieval we used only the far range signals, since there is no near-range cross- polarized measurements. Figure 9 shows that the POLIPHON algorithm also indicates the limited presence of other aerosols types compared to dust, at heights above ∼1 km and finds the maximum peak for dust particles at 2 km at high concentration values of 350 µg/m 3 , thus enhancing GRASP/GARRLiC retrievals.
induced by the complex topography [10].
To strengthen the above results, the mass concentration profiles simulated w NMME-DREAM model and retrieved with GRASP/GARRLiC are compared with mass concentration profiles retrieved with POLIPHON ( Figure 9) for continental (g dashed line), dust (magenta dashed line) and marine particles (blue dashed line). Fo POLIPHON retrieval we used only the far range signals, since there is no near-range c polarized measurements. Figure 9 shows that the POLIPHON algorithm also indicate limited presence of other aerosols types compared to dust, at heights above ~1 km finds the maximum peak for dust particles at 2 km at high concentration values o μg/m 3 , thus enhancing GRASP/GARRLiC retrievals.

Conclusions
This study applies the GRASP/GARRLiC inversion algorithm to coincident E LINET lidar and AERONET sun-photometer observations performed at the Finokali tion in Crete during the PreTECT campaign. The algorithm was applied during an in Saharan dust episode on 14 May 2017 reaching 256 μg/m 3 at 250 m and derived colu dust properties, in terms of the particle size distribution, spectral complex refractive i and SSA, as well as the concentration profiles. The GRASP/GARRLiC retrieval agrees with the AERONET product for the larger particles, and with climatological valu dust.
A second objective of this study was to demonstrate a high potential value of concentration profiles retrieved by GRASP/GARRLiC for evaluation of the dust m simulations (NMME-DREAM is used for this case study). Additionally, the retri from the POLIPHON technique were used to estimate the mass concentration profile enhance the driven conclusions. The GRASP/GARRLiC retrievals agree reasonably with the model simulations at high altitudes for this dust episode, with both capturin maximum dust concentrations at 2 km (250 μg/m 3 for GRASP/GARRLiC and 300 μ for the model) but at the lower layers (~1-2 km) the model strongly underestimates concentrations. This underestimation at low levels is in accordance with POLIPHO trieved mass concentrations that further confirm that there is no other dominant ty aerosols but dust at these heights. Below 1 km, we expect the contribution of the m

Conclusions
This study applies the GRASP/GARRLiC inversion algorithm to coincident EAR-LINET lidar and AERONET sun-photometer observations performed at the Finokalia station in Crete during the PreTECT campaign. The algorithm was applied during an intense Saharan dust episode on 14 May 2017 reaching 256 µg/m 3 at 250 m and derived columnar dust properties, in terms of the particle size distribution, spectral complex refractive index and SSA, as well as the concentration profiles. The GRASP/GARRLiC retrieval agrees well with the AERONET product for the larger particles, and with climatological values of dust.
A second objective of this study was to demonstrate a high potential value of dust concentration profiles retrieved by GRASP/GARRLiC for evaluation of the dust model simulations (NMME-DREAM is used for this case study). Additionally, the retrievals from the POLIPHON technique were used to estimate the mass concentration profile and enhance the driven conclusions. The GRASP/GARRLiC retrievals agree reasonably well with the model simulations at high altitudes for this dust episode, with both capturing the maximum dust concentrations at 2 km (250 µg/m 3 for GRASP/GARRLiC and 300 µg/m 3 for the model) but at the lower layers (∼1-2 km) the model strongly underestimates dust concentrations. This underestimation at low levels is in accordance with POLIPHON retrieved mass concentrations that further confirm that there is no other dominant type of aerosols but dust at these heights. Below 1 km, we expect the contribution of the marine particles, but the lack of near-range cross-polarized measurements does not facilitate their retrieval with POLIPHON.
The selected event of 14 May 2017 is characterized by a two-peak vertical structure that is adequately reproduced by the model, but also by concentrations up to 300 µg/m 3 inside the PBL that are not reproduced. Similar limitations have been previously reported for mesoscale simulations at the station of Finokalia in Crete, due to the inaccurate repre-