Measurement report: Balloon-borne in-situ profiling of Saharan dust over Cyprus with the UCASS optical particle counter

This paper presents measurements of mineral dust concentration in the diameter range from 0.4 to 14.0 μm with a novel balloon-borne optical particle counter, the Universal Cloud and Aerosol Sounding System (UCASS). The balloon launches were coordinated with ground-based active and passive remote-sensing observations and airborne in-situ measurements with a research aircraft during a Saharan dust outbreak over Cyprus from 20 to 23 April 2017. The aerosol optical depth at 500 nm reached values up to 0.5 during that event over Cyprus and particle number concentrations were as high as 50 cm−3 5 for the diameter range between 0.8 and 13.9μm. Comparisons of the total particle number concentration and the particle size distribution from two cases of balloon-borne measurements with aircraft observations show reasonable agreement in magnitude and shape despite slight mismatches in time and space. While column-integrated size distributions from balloon-borne measurements and ground-based remote sensing show similar coarse-mode peak concentrations and diameters, they illustrate the ambiguity related to the missing vertical information in passive sun photometer observations. Extinction coefficient inferred 10 from the balloon-borne measurements agrees with those derived from coinciding Raman lidar observations at height levels with particle number concentrations smaller than 10 cm−3 for the diameter range from 0.8 to 13.9μm. An overestimation of the extinction coefficient of a factor of two was found for layers with particle number concentrations that exceed 25 cm−3. This is likely the result of a variation in the refractive index, the shapeand size-dependency of the extinction efficiency of dust particles along the UCASS measurements. 15 1 https://doi.org/10.5194/acp-2020-977 Preprint. Discussion started: 6 November 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Atmospheric aerosols are of significant importance for the Earth's radiative budget. They have a direct impact on climate by scattering and absorbing solar radiation. They can also act as ice nucleating particles and cloud condensation nuclei, and thus, influence not only the formation and evolution of clouds but also the hydrological cycle (Stocker et al., 2013). Aerosols and their precursors originate from natural and anthropogenic sources. Natural sources include emissions from the ocean, soils, 20 volcanoes, and vegetation, whereas anthropogenic sources include emissions from the combustion of fossil fuels or the result of changes in land use (Boucher, 2015). For instance, sulphates and soot can be of both natural and anthropogenic origin, while mineral dust and marine aerosols originate predominantly from natural sources (Stocker et al., 2013). The latter two aerosol types are abundant in the atmosphere and particularly mineral dust can be transported over intercontinental distance from its source regions (Prospero, 1999;Weinzierl et al., 2017). 25 Over the past 15 years, several measurement campaigns have focused on gaining deeper insight into the role of mineral dust on the Earth's system. An overview of several mineral dust field campaigns is given in Weinzierl et al. (2017). These experiments generally featured comprehensive remote-sensing instrumentation, detailed monitoring of chemical, microphysical, and optical properties of aerosols at the surface as well as airborne in-situ observations with research aircraft. Such observations have been performed, for instance, during the two Saharan Mineral Dust Experiments (SAMUM, Weinzierl et al. 2009; Ans-30 mann et al. 2011), Fennec (Ryder et al., 2013), the Saharan Aerosol Long-Range Transport and Aerosol-Cloud-Interaction Experiment (SALTRACE, Weinzierl et al. 2017), and the CHemistry and AeRosols Mediterranean EXperiments (CHArMEx, Renard et al. 2018). Recently, the focus of such activities has extended towards the eastern Mediterranean as this region is on the cross road of aerosol transport of mineral dust from Sahara and Middle East, continental outflow from Europe, as well as biomass-burning smoke from eastern Europe and central Asia (Georgoulias et al., 2016). The majority of dust storms over 35 the eastern Mediterranean basin occurs between December and April with maximum dust load during April (Israelevich et al., 2002). The main zones of cyclogenesis in the Mediterranean Sea determine dust uplift and transport in the region (Alpert et al., 1990). Heavy dust periods over the eastern Mediterranean are frequently associated with the so-called Cyprus Low (Katsnelson, 1970;Dayan et al., 2008) as well as the Sharav cyclone (Alpert and Ziv, 1989) which transport dust from the Arabian deserts and northern Sahara into the eastern Mediterranean basin where they are frequently observed over Cyprus. 40 Statistical information on the size distributions of atmospheric aerosols, cloud droplets, and ice crystals is of vital importance for identifying and evaluating the physical processes governing aerosol-cloud interactions and their climate effects which currently contribute considerable uncertainty to our understanding of current and future climate change (Stocker et al., 2013) as well as to the performance of Numerical Weather Prediction models (Baldauf et al., 2011). The majority of the data assimilated into models and used for model verification comes from remote-sensing observations (Lahoz and Schneider, 2014).  logical soundings in combination with an optical particle counter (OPC) can provide time series of aerosol size distribution profiles that have the potential to complement the data for assimilation in and verification of atmospheric models. The purpose of this paper is to present results of in-situ measurements of mineral dust particles over the eastern Mediterranean with a novel disposable balloon-borne OPC and to assess the quality of the collected data based on independent observations. size bins together with time of flight data for quality assurance. Other sondes employing the XDATA protocol can be used as well. The UCASS-radiosonde payload can be used to obtain aerosol and cloud profiles from the surface up to the tropopause within about 60 min from launch. The flow speed through the UCASS' open detection path is determined by the ascent rate u of the meteorological balloon. Effects of a tilt of the instrument on the flow rate are discussed in Smith et al. (2019). During the launch preparation, the balloon is filled to a size that translates into an ascent rate of about 5 ms −1 to guarantee optimum 95 measurement performance of the UCASS. In the data analysis, the ascent rate is calculated from the change in height h with time t by u = ∆h/∆t. The ascent rate is used to calculate the volume v of sampled air by v = Aut with the UCASS sample area of A = 5.0 × 10 −7 m 2 which is specified as a section of the laser beam (Smith et al., 2019). The device electronics can measure up to 104 particles per second and can operate in air flow speeds between 2 and 15,ms −1 , with the standard firmware.

100
The raw particle counts C per size bin i are used to calculate the particle number concentration per size bin n i = C i /V as number of particles per unit volume over the covered size range. Summation of n i over all size bins leads to the total number concentration N . The particle number size distribution is determined by with the assumption of spherical particles. While mineral dust particles are non-spherical, the shape effect on the scattering 105 phase function with respect to spherical particles is less pronounced within the angular range exploited in the UCASS setup (forward to sideward scattering) compared to scattering in the backward direction. Hence, the use of Mie scattering has a small effect on the calculated size distributions even in the presence of non-spherical particles (Johnson and Osborne, 2011;Lacis and Mishchenko, 1995).
The column-integrated volume size distribution for comparison to the normalised volume size distributions provided from 110 remote-sensing retrievals is calculated using the sum of the number concentration for each bin over the entire ascent together with the bin centre (D i+1 + D i )/2 and width D i+1 − D i by The effective diameter is defined by Hansen (1971) as the ratio of the volume to the surface-area concentration by 115 UCASS measurements can be used to calculate the aerosol extinction coefficient. This can then be compared to the extinction coefficient profile derived from collocated lidar measurements. For a particle with diameter D and known refractive index, the size-dependent extinction efficiency Q ext (D) (unitless) can be derived from Mie-scattering calculations. Here, we use a refractive index of 1.52+0.002i. Then, the extinction cross section of the particle (in m 2 ) is calculated by C ext = (πD 2 /4)Q ext (D).
Using the measured number concentration, the extinction coefficient (in m −1 ) is derived by

A-LIFE Instrumentation
In order to demonstrate the UCASS' capability for profiling of aerosol number concentrations and size distributions, the quality of its observations needs to be evaluated with the help of independent data. To meet optimum conditions for comparison, UCASS launches during A-LIFE were coordinated with ground-based remote sensing (ensuring also the temporal collocation 125 of active and passive instruments) and the flight schedule of the DLR-Falcon research aircraft.
A Polly XT multiwavelength aerosol Raman lidar (Engelmann et al., 2016)  concentration, size, and type (Engelmann et al., 2016). Near real-time data from the Polly XT website (http://polly.tropos.de) were consulted to schedule UCASS launches for dust observations.
Aerosol Robotic Network (AERONET) sun photometer measurements (Holben et al., 1998) during A-LIFE were performed at Paphos and Limassol. These measurements provide information on the optical and microphysical properties of the bulk aerosol in the atmospheric column. AERONET sun photometers perform spectrally resolved measurements of aerosol optical 135 depth (AOD) at 340, 440, 675, 870, 1020, and 1640 nm and of sky radiances at several almucantar angles at 440, 675, 870, and 1020 nm (Holben et al., 1998). In this work, only AERONET version 3 level 2.0 data are considered.
The DLR-Falcon research aircraft was equipped with an extensive in-situ aerosol payload including total aerosol concentration measurements (0.005 -930 µm), highly resolved size distribution measurements in the range between 0.25 and 930 µm particle diameter, a wind lidar and meteorological sensors. Furthermore, aerosol optical properties were determined, and parti-140 cles were collected for offline chemical analyses. The setup was similar to earlier campaigns that also focused on mineral dust (Weinzierl et al., 2009(Weinzierl et al., , 2017. Local column closure flights were performed at the sites of Paphos airport and the Limassol lidar station. In this paper, UCASS measurements are compared to data collected with a second generation Cloud, Aerosol, and Precipitation Spectrometer (CAPS, Spanu et al. 2020) that was mounted at the aircraft wing. The CAPS instrument consists of a Cloud and Aerosol Spectrometer with Depolarization Detection (CAS), and a Cloud Imaging Probe (CIP). Furthermore,

Remote sensing retrievals
UCASS in-situ measurements of the particle size distribution and the subsequently derived extinction coefficient are also evaluated with the findings from remote-sensing observations. For this purpose, lidar and sun-photometer data are used as 155 input to the Generalised Aerosol Retrieval from Radiometer and Lidar Combined data algorithm (GARRLiC, Lopatin et al. 2013) and the AERONET (Dubovik et al., 2006) and ERS/SKYNET-SKYRAD (Campanelli et al., 2007) inversions. The use of in-situ data from OPC measurements as a benchmark allows for an assessment of the reliability of the different methods (Tsekeri et al., 2017) in the presence of coarse-mode dominated aerosols.

160
The AERONET inversion employs measurements of direct and diffuse radiation with sun and sky radiometers to retrieve aerosol optical and microphysical particle properties that are representative of the total atmospheric column (Dubovik et al., 2000(Dubovik et al., , 2006. The AERONET algorithm assumes a vertically homogeneous atmosphere and a mono-component aerosol with a single complex refractive index. AERONET inversion products include the particle size distribution, the complex refractive index, the scattering phase function, the single-scattering albedo, and spectral and broad-band fluxes. Size distributions obtained 165 from AERONET measurements of mineral dust have shown a dominant mode at around 4 to 5 µm in diameter (Müller et al., 2012;Marenco et al., 2018). The AERONET retrieval forces the particle size distribution to zero at 30 µm in diameter. This constraint may therefore lead to an underestimation of the concentration of large particles by AERONET (Ryder et al., 2019).
The reported uncertainties for the AERONET size distribution retrievals in the range from 0.1 to 7.0 µm in radius are given as 10% to 35%, while for larger sizes, uncertainties rise up to 80% to 100% (Dubovik et al., 2000(Dubovik et al., , 2002.

ESR/SKYNET
SKYNET is an international research network of users of the PREDE Co. Ltd POM sky radiometer with a growing number of instruments now exceeding 100 units. Currently, SKYNET uses two versions (4.2 and 5) of the inversion algorithm SKYRAD to analyse the radiance measurements of the PREDE POM sky radiometers, although other versions are being developed and currently tested. In order to benefit the international community of users, a re-organisation of the network structure has been 175 initiated (Nakajima et al., 2020).
Although the International SKYNET Data Center (ISDC) has already started data collection and analysis, different regional sub-networks are well established, and develop new research products and test new methodologies (Nakajima et al., 2020). In Europe, the regional sub-network is called the European SKYNET Radiometers network (ESR). In ESR, versions of SKYRAD software have been adapted to analyse data from CIMEL sun-sky photometers (Estellés et al., 2012). In this analysis the current 180 SKYRAD version 4.2 is used and the corresponding inversions will be called SKYRAD retrievals.
As for the AERONET inversion, the SKYRAD algorithm estimates the size distribution, phase function and surface albedo of aerosols from measurements of diffuse sky radiance (Campanelli et al., 2007). A notable difference to AERONET is, however, that the SKYRAD retrieval does not prescribe an upper boundary for particle size (Estellés et al., 2012).

185
The GARRLiC retrieval is a synergistic algorithm that combines quasi-simultaneous passive sky-radiance measurements with active lidar measurements during cloud-free conditions. The required input from sun photometer observations includes the total AOD and radiances at 440, 670, 870, and 1020 nm. Concurrently, lidar measurements of the elastic backscatter signals at 355, 532, and 1064 nm are used as input for GARRLiC (Lopatin et al., 2013). The output of the retrieval provides profiles of aerosol volume concentration together with a column-mean aerosol volume size distribution, spectral refractive index, and 190 spherical particle fraction. GARRLiC also retrieves aerosol optical properties such as the single-scattering albedo, backscatter and extinction coefficients, and the aerosol lidar ratio.
The lidar input enables GARRLiC to account for variations in aerosol stratification. Due to the wider set of input parameters, the GARRLiC retrieval requires fewer assumptions than other algorithms (Bovchaliuk et al., 2016;Tsekeri et al., 2017). In case of a bi-modal aerosol distribution, GARRLiC provides the flexibility to use a bi-component aerosol model that may have 195 different refractive indices in the fine and coarse mode. In the presence of mixtures of aerosol types with multiple contributions to the fine and coarse mode (e.g. mixture of marine and dust particles, Tsekeri et al. 2017), the algorithm provides an average estimation similar to the AERONET retrieval. We constrain the investigation in this study to one dust mode because the UCASS observations at Cyprus show a dominance of coarse-mode dust particles throughout the atmospheric column.  Table 1 provides an overview of the UCASS launch times 215 together with the time periods and locations of the remote-sensing and airborne measurements used for comparison. The first airborne mission over this period was performed on 19 April; the leading edge of the dust plume was found to be over Malta and moved eastwards across the Mediterranean in the following days.

HYSPLIT backward trajectories
The AERONET measurements in Figure 1a show  launches. They reveal that these air parcels were lifted from dust source regions in North Africa, crossed the Mediterranean, and reached Cyprus within three days. The air parcels arriving at 2 and 3 km height originated from northern Libya while those arriving at 5 and 7 km height originated from Algeria, Morocco, and Mauritania. The difference in source region and transport time for air arriving at different altitudes might lead to differences in the observed particle size distributions at those heights (Weinzierl et al., 2009Ryder et al., 2013. The inspection of dust composites derived from measurements with 250 the Spinning Enhanced Visible and Infra-red Imager (SEVIRI) on the Meteosat Second Generation satellite (Schepanski et al. 2007, not shown) shows that dust was mobilised in the northern part of Cyrenaica (i.e. north eastern Libya) about 24 h before the observations of the second UCASS launch and transported directly to Cyprus.  Looking at the size distribution of SALTRACE, the highest particle number concentrations are found between 0.5 and 1 µm.

Number concentration profiles
The particle number concentration decreases towards the larger sizes. A closer look at the size range between 0.4 and 20.0 µm, which is closer to the measurement capability of UCASS, reveals number concentrations spanning from 0.01 to 100 cm −3 . Note that most of the observations listed above have been conducted much closer to dust sources compared to the measurements at 295 Cyprus presented here. Hence, it can be concluded that the UCASS observations give values that are in line with data from airborne campaigns.

Layer-averaged number size distributions
A closer look at layer-mean particle size distributions from the measurements with the UCASS and the research aircraft on 20 and 21 April 2017 is provided in Figure 5. The extent of the considered height layers is marked in Figure 4. The particle This is likely due to the fact that the UCASS as deployed during A-LIFE measured only up to particle diameters of 14 µm while its nominal detection range extends to 17 µm. In any case, a comparison of the effective diameters from different measurements 320 may not be comprehensive due to the different source regions and travelled distances of the observed dust particles from the different observations.

Columnar size distributions
A comparison of the columnar aerosol volume size distribution from the GARRLiC, AERONET, and SKYRAD retrievals and the first, and third UCASS launches is presented in Figure 6. Figure 6a also includes the airborne in-situ observations with the 325 Falcon aircraft. All distributions in Figures 6a and b show a predominance of coarse-mode particles with comparable volume concentrations of the different coarse-mode peaks. During Launch 1, the UCASS and GARRLiC retrieval show a different shape of the volume size distribution compared to the ones retrieved by the sun photometer inversions. Both AERONET and SKYRAD show a single coarse mode that peaks between 3 and 5 µm, while GARRLiC and UCASS give a coarse mode with two peaks. AERONET observed the highest concentration of 0.12 µm 3 / µm 2 between 3.4 and 4.5 µm diameter, whereas 330 SKYRAD's size distribution peaks at 3.4 µm with a concentration of 0.13 µm 3 / µm 2 . The UCASS observed its highest concentration of about 0.1 µm 3 / µm 2 at 5.5 µm in diameter and a second mode at 2.8 µm. The two coarse modes retrieved by GARRLiC are at 2.0 and 7.7 µm in diameter. It is noteworthy to mention that the first UCASS unit was launched about 2 hours 40 min after the considered sun-photometer measurement as outlined in Table 1. The first UCASS launch shown in Figure 6a is also the only case for which a column-integrated volume size distribution is available from the observations aboard the re-335 search aircraft. These independent airborne in-situ measurements also find a bi-modal coarse mode which supports the results of the UCASS measurements and suggests that the findings of the GARRLiC retrieval are closer to reality than those from AERONET and SKYRAD.
Post-processing was applied to the UCASS data. In addition, further laboratory measurements with a set-up comparable to the conditions encountered during the launches on Cyprus were performed to examine if the observed bi-modal size distribu-tions in Figure 6 could be the result of an instrumental artefact. Mono-modal sample materials were used in these laboratory tests (Smith et al., 2019). The corresponding UCASS measurements also showed only mono-modal size distributions. This lead us to reject the idea of a systematic instrumental error. Hence, the bi-modal coarse mode might be a special characteristic of the origin of the observed air masses. The observed bi-modal peak may be caused by the following reasons: (i) the diversity of sources across the African basin whose mineralogy can lead to intrinsic differences in the properties of the emitted parti-345 cles (e.g. size distributions, chemical composition) (Engelstaedter et al., 2006;Coz et al., 2009), (ii) cloud processing during transport which could cause aggregation of particles that were collected by droplets that evaporated at a later stage or wash-out of larger particles (Matsuki et al., 2010), (iii) gravitational settling of particles for longer transport times compared to freshly emitted dust after about one day of transport might lead to the systematic removal of large particles, particularly in the upper part of dust plumes (Ellis and Merrill, 1995;Maring et al., 2003), or (iv) dust electrification that could counteract gravitational 350 settling by creating an electric field within the dust layer (Nicoll, 2012). A similar bi-modal coarse-size distribution was also observed during the Puerto Rico Dust Experiment (PRIDE, Reid et al. 2003) and Fennec SAL (Song et al., 2018). However, neither study provides further discussion of these observations.
The second UCASS launch on 21 April 2017 was performed about 3.0 h and 1.5 h before the first sun photometer and lidar measurements, respectively (Table 1). This is likely to have an effect on the UCASS comparison in Figure 6b as the aerosol 355 conditions varied strongly during that period (Figure 1). The UCASS size distribution peaks at 5.8 µm with a concentration of 0.13 µm 3 / µm 2 . This peak is also resolved by the GARRLiC-derived size distribution, though it is located at 2.6 and 5.9 µm with concentrations of 0.24 and 0.18 µm 3 / µm 2 , respectively. The AERONET-derived size distribution shows a peak concentration of 0.22 µm 3 / µm 2 between 3.4 and 4.5 µm particle diameter. The SKYRAD retrieval gives a peak concentration of 0.21 µm 3 / µm 2 at a coarse-mode diameter of 3.4 µm which is the smallest compared to the other observations. Although the 360 sun photometer inversions rely on the same input data sets, it is found that the SKYRAD size distribution is shifted to smaller sizes compared to AERONET. This is surprising as SKYRAD does not force the size distribution to zero at larger particle diameters (Campanelli et al., 2007) and, in principle, would enable the retrieval of size distributions with larger coarse-mode diameters than AERONET. This particular property of the AERONET retrieval is likely to produce the artificial fine-mode peak at around 0.15 µm that is absent in the SKYRAD size-distributions (Dubovik et al., 2006).

365
The peaks of volume size distribution from the sun photometer inversions are found to be systematically at smaller particle sizes than the observations from GARRLiC and the UCASS. A similar shift towards larger particle size was also observed from in-situ measurements aboard the Falcon compared to AERONET-derived size distributions during SAMUM (Müller et al., 2012). A similar tendency between AERONET size distributions and in-situ measurements was observed during DABEX in the Sahelian west Africa basin (Osborne et al., 2008). During the SAVEX-D/AER-D campaign at Cape Verde, the AERONET 370 retrievals also showed a similar single coarse mode shifted towards a smaller radius compared to in-situ measurements from aircraft (Estelles et al., 2018). Simultaneous retrievals from SKYRAD (performed on Prede POM radiometers) also determined a slight shift of the coarse mode to a smaller radius, although in this case the coarse mode was broader or even bimodal, depending on the SKYRAD version used (Nakajima et al., 2020). The coarse mode retrieved by GARRLiC shows a consistent shift towards larger sizes when compared to the AERONET output (Benavent-Oltra et al., 2017;Lopatin et al., 2013;Bovchaliuk 12 https://doi.org/10.5194/acp-2020-977 Preprint. Discussion started: 6 November 2020 c Author(s) 2020. CC BY 4.0 License. et al., 2016). This feature is generally attributed to the additional information from the backscatter lidar profiles that provides GARRLiC with extra information on the particle size. In addition, the restriction of the AERONET (and hence also the GAR-RLiC) data inversion scheme to a particle diameter smaller than 30 µm may lead to an underestimation of the concentration of coarse-mode particles (Müller et al., 2012).
A closer look at the UCASS measurements during the second launch is provided in Figure 7 in terms of the profile of 380 total number concentration and volume size distributions averaged over four height layers. Figure 7c shows that the altitude range between 2.8 and 3.1 km is dominated by particles with a mode diameter of 8.4 µm. In contrast, all other layers show an up to an order of magnitude smaller concentration of particles with such large diameters. Hence, the thin filament of dust particles observed in the morning of 21 April 2017 is the major contributor to the coarse-mode in the columnar size distribution in Figure 6b. This structure was confined to a very small height range and only lasted for about 12 h. It had already 385 disappeared from the ground-based remote-sensing sites during the time of the DLR-Falcon research flight that day. Longer transport times translate to a longer time period during which large particles are exposed to gravitational settling. This effect is most pronounced at higher altitudes where no particles can settle into the layer from above. Figure 3 indicates that the aerosols observed at 5 and 7 km height have been transported over longer distances than those at lower altitudes. As stated before, MSG-SEVIRI imagery shows active dust sources in north eastern Libya about 24 h before the observations at Limassol. Backward 390 trajectories corroborate that dust emitted from these sources would was transported directly to Cyprus. It is likely that this is the origin of the thin filament observed in the morning of 21 April 2017. It is noteworthy to state that the first sun photometer observations (used for AERONET, SKYRAD, and GARRLiC retrievals) took place after sunrise (0427 UTC), when the dense aerosol filament over Limassol had changed its appearance and extended in depth.
The overarching message of Figures 6 and 7 is twofold. Firstly, reasonable agreement can be found between the UCASS 395 measurements and data from remote-sensing observations in case of homogeneous dust properties and optimum temporal matching of the observations (Figure 6a). Under such conditions, the more complex GARRLiC retrieval which is based on a larger set of input data is capable of better resolving the features of the UCASS in-situ measurements, i.e. the doublepeak in the coarse mode. Secondly, the requirement for homogeneous aerosol conditions is vital, if observations at different times are compared or used as combined input to a retrieval. In that context, Figure 6b provides some insight into the actual 400 spread of findings that can result from extreme variations in the aerosol situation such as changes in total aerosol load or the vertical distribution of the particles. This is particularly important when using passive remote-sensing data for the validation of vertically-resolved measurements as they provide no information on aerosol stratification.  Table 1. Lidar profiles of the extinction coefficient measured by the Polly XT at Limassol were derived using two methods. Firstly, the extinction coefficient was obtained without assumptions using the Raman method (Ansmann and Müller, 2005). Secondly, the likely range of extinction coefficients was estimated by multiplying the particle backscatter coefficient obtained using Klett's method with the lower and upper limits of reasonable dust lidar ratios for Cyprus of 40 and 60 sr, lowermost layers are in a reasonable agreement with values below 100 Mm −1 . Discrepancies are more pronounced for the observations within the elevated layers.

Extinction coefficient profiles
The extinction coefficient profiles were integrated with height to obtain an estimate of AOT that can be compared to the sun photometer measurements. This comparison is shown in Figure 1a and in Table 2. The lower lidar-estimate of AOT derived using Klett's method and a lidar ratio of 40 sr shows the best agreement with the independent sun photometer observations at that are well above the remote-sensing estimates that range from 0.32 to 0.58. Figure 8 confirms that this is due to the elevated layers characterised by peak particle concentrations. Particularly, this occurs when UCASS-derived extinction coefficients are as high as 300 Mm −1 . Much lower extinction coefficients of 70 to 150 Mm −1 are found from the different analyses of the lidar measurements (Klett, Raman). As these layers are characterised by an increased concentration of larger particles, there is reason to believe that the current UCASS extinction conversion is not universally applicable to different aerosol conditions.

425
The refractive index and the size-dependent extinction efficiency are the main factors in the retrieval of extinction coefficients from the UCASS measurements. The large particles in the elevated layers might therefore be of different chemical composition compared to those at lower layers. This is supported by the backward trajectories in Figure 3 which indicate different source regions for air arriving at different height levels. Alternatively, the extinction efficiency used in the current conversion might be representative only for situations dominated by smaller particles for which the effect of particle non-sphericity is less 430 pronounced. Comparisons with extinction coefficient, which is a secondary-order parameter derived from UCASS data, are therefore questionable and require further investigation that is beyond the scope of this study.

Summary and conclusions
We have presented findings from balloon-borne UCASS optical particle counter measurements of mineral dust conducted over Cyprus in April 2017 during the A-LIFE experiment. The UCASS launches were embedded in research activities that 435 included airborne in-situ measurements with the DLR-Falcon research aircraft as well as ground-based remote sensing with advanced aerosol lidars and sun photometers. This setup allows for a comprehensive evaluation of the quality of the UCASS measurements as well as an assessment of a variety of remote-sensing retrievals.
The highest particle number concentration observed by the UCASS was found during the first launch, with values of up to 50 cm −3 within a layer from 3 to 5 km height. Aircraft observations gave slightly lower values with a maximum of 40 cm −3 .

440
The UCASS profile of number concentration during the third launch resembles the aircraft observations in the vertical structure as well as the dust load.
During the first launch, layer-averaged particle size distribution from the UCASS measurements resembles those measured by the instruments aboard the aircraft in the size range between 2 to 5 µm. Larger differences in the particle number concentration are found for particle sizes larger than 5 µm, though these are still within the error bar of the measurements. During 445 the second launch conducted from Limassol, a thin filament of dust was observed between 2.8 and 3.1 km height. This feature revealed a dominance of very large particles with average mode diameter of 8.4 µm. In contrast, the layers below 2.2 km and above 4.8 km height were dominated by lower concentration of coarse-mode particles. Furthermore, it was found that the number concentration of large particles decreased with altitude. This is likely to be a result of gravitational settling.
Column-integrated particle volume size distribution were calculated from the UCASS measurements for a comparison to the 450 findings of remote-sensing retrievals. For the first launch, results from the GARRLiC retrieval are the only ones that reproduce the bi-modal coarse mode detected by UCASS while the AERONET and SKYRAD inversions give a single coarse mode.
Nevertheless, the AERONET-derived size distribution particle concentrations are similar to the UCASS measurements in the size range from 1 to 6 µm. In contrast, the SKYRAD-derived size distribution shows a coarse-mode peak at 2.0 to 3.0 µm which is in line with the first coarse-mode peak of the GARRLiC-derived size distribution but smaller then the other inferred coarse-455 mode peak diameters. During the second launch, the volume size distributions obtained by the UCASS and GARRLiC peak at a particle diameter of around 6.0 µm. However, GARRLiC also gives a second pronounced peak at around 3.0 µm that is hardly visible in the UCASS measurement. In addition, large discrepancies on the shape and maximum of the volume size distribution were observed between the UCASS and the retrievals obtained by sun photometer data alone, i.e. the AERONET and SKYRAD inversions. This is attributed to the temporal difference between the observations of 3 h and the strongly heterogeneous dust 460 layering: the sun photometer observations were performed when the thin dust filament observed by the UCASS and lidar had dissipated into the layer below the filament. . Overall, UCASS measurements of particle concentrations and size distributions are found to be reasonably in line with coincident observations with research aircraft and remote-sensing instruments. The low-cost and disposable nature of the instrument therefore makes it a attractive tool for the in-situ profiling of atmospheric particle concentrations in the framework of field experiments and long-term observations.

465
The comparison with secondary-order parameters, i.e. extinction coefficient, is yet questionable and requires further investigation beyond the scope of this study. Given the relatively low cost of the disposable UCASS, it provides a promising new opportunity for in-situ measurements of particle size distributions within different aerosol layers and for validation studies between remote-sensing and in-situ observations (Sawamura et al., 2017), not only for optical data but also microphysical data.   Table 1. Dates and times (UTC) of UCASS launches, aircraft profiles (together with location of observation), and measurements with lidar and sun photometer (SPM) used in this study. UCASS units were launched from Paphos except for the launch at 0134 UTC on 21 April 2017 which was performed next to the lidar site at Limassol. Lidar 1 refers to the time period used for the comparison of AOD and extinction coefficients in Figures 1 and 8, respectively. Lidar 2 marks the time period used for the combined lidar-SPM retrievals with GARRLiC in Figure 6.  Table 2. Column AOD derived from the integration of the extinction coefficient profiles in Figure 8 for the times of the UCASS launches (see Table 1). AODs are also shown in Figure