Ice nucleating particles over the Eastern Mediterranean measured by unmanned aircraft systems

. During an intensive ﬁeld campaign on aerosol, clouds and ice nucleation in the Eastern Mediterranean in April 2016, we have measured the abundance of ice nucleating particles (INP) in the lower troposphere from unmanned aircraft systems (UAS). Aerosol samples were collected by miniaturized electrostatic precipitators onboard the UAS at altitudes up to 2.5 km . The number of INP in these samples, which are active in the deposition and condensation modes at temperatures from − 20 5 to − 30 ◦ C , were analyzed immediately after collection on site using the ice nucleus counter FRIDGE. During the one month campaign we encountered a series of Saharan dust plumes that traveled at several kilometers altitude. Here we present INP data from 42 individual ﬂights, together with aerosol number concentrations, observations of lidar backscattering, dust concentrations derived by the dust transport model DREAM (Dust Regional Atmospheric Model), and results from scanning electron microscopy. The effect of the dust plumes is reﬂected by the coincidence of INP with the particulate mass (PM), the lidar 10 signal and with the predicted dust mass of the model. This suggests that mineral dust or a constituent related to dust was a major contributor to the ice nucleating properties of the aerosol. Peak concentrations of above 100 INP std . l -1 were measured at − 30 ◦ C . The INP concentration in elevated plumes was on average a factor of 10 higher than at ground level. Since desert dust is transported for long distances over wide areas of the globe predominantly at several km altitude we conclude that INP measurements at ground level may be of limited signiﬁcance for the situation at the level of cloud formation. search Unit FOR 1525 (INUIT). This campaign has been performed at the Cyprus Atmospheric Observatory (CAO) which is part of the ACTRIS2 project that has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 654109. The support by the international Research Institute for Climate and Society, Columbia University, Palisades, N.Y. with meteorological data and software is gratefully acknowledged. We acknowledge support from the DFG-Research Center / Cluster of Excel-lence “The Ocean in the Earth System-MARUM”. We thank SNO PHOTONS/AERONET from INSU/CNRS/University of Lille, France, 5 AERONET-Europe/ACTRIS Calibration Center and AERONET Team at GSFC for their kind cooperation. The Department of Labour Inspection (DLI, Ministry of Labour, Welfare, and Social Insurance) is thanked for the provisions of ground-based weather and PM data at CAO. The provision of the HYSPLIT transport and dispersion model from the NOAA Air Resources Laboratory is gratefully acknowledged.

below −20 • C. Many of the earlier field studies conclude the ice-nucleating properties of dust by circumstantial reasoning, i.e. from correlation of dust parameters with INP abundance. The advanced coupling of INP measurements to electron microscopy or to mass spectrometry, which allows identification of single nucleating particles, again demonstrated that mineral dust is a major constituent of INP, but that a significant biological INP-compound can be present too (Pratt et al., 2009;Prenni et al., 2009b). INP concentrations measured within a dust layer can reach up to 1000 l -1 (DeMott et al., 2003a). Once mobilized by 5 high surface winds in arid and semi-arid regions of the world, mineral dust particles can travel up to several thousands of kilometers (Prospero, 1999). Hence, regions that are far away from the desert can still be influenced by mineral dust due to efficient long distance transport. Liu et al. (2008) have shown the first height-resolved global distribution of dust aerosol based on lidar measurements of CALIPSO. They found northern hemispheric spring to be the most active dust season, with 12 % of the areas between 0°and 60°N to be influenced at least half the time. In general the vertical extent of the mineral dust was 10 found to be strongly dependent on season and source region. Peak dust layers were found to be between 2-3 km in summer and 1-2 km in the other seasons. The regions of North Africa and the Arabian Peninsula were found to be the most persistent sources of mineral dust. A transatlantic transport of African dust was seen all year, with a significant amount of dust transported in the free troposphere in layers above 2 km in summer. In winter most dust was transported below 2 km. In summary, existing climatologies demonstrate that mineral dust is mainly transported in the lowest few kilometers of the atmosphere. Nonetheless, 15 to this date most INP measurements are made at ground level and not at elevations where clouds actually are formed.
These findings strongly emphasize the need for more measurements of INP above ground level. Yet, achieving INP measurements in free tropospheric air masses is challenging and usually requires substantial effort. For over a decade large research aircraft equipped with continuous-flow diffusion chambers (CFDC) have been used to measure INP concentration and composition in the free troposphere (e.g. Rogers et al., 2001a, b;DeMott et al., 2003a;Cziczo et al., 2004;Prenni et al., 2009a). So potential of UAS-operated observations, especially for atmospheric sciences. However, to our knowledge, no measurements of INP based on UAS have been performed or published to this day.
The method we present here to measure INP from the UAS is based on the offline ice nucleus counter FRIDGE (Schrod et al., 2016).

Meteorological conditions
During the campaign the Eastern Mediterranean was mainly under westerly flow, as indicated by the contours of the monthly mean stream function at 500 hPa (Fig. 1a). This flow meandered according to the eastward propagation of troughs and ridges with periods of several days. At sea surface level, pressure gradients over the Eastern Mediterranean were mostly weak ( Fig. 1b). The days of March 27 to 30 in Cyprus were characterized by cold, cyclonic conditions, and April 1 to 8 by warm, 5 anticyclonic conditions. A low pressure system that was cut off the major trough over Spain on April 6 traveled slowly eastward along the North African coastline and the Eastern Mediterranean Sea and over Cyprus (April 12 to 13) towards Syria, where it dissipated. On April 14 to 20 anticyclonic conditions prevailed in Cyprus with westerly flow of warm air, on April 21 to 27 again a cyclonic pattern with predominantly warm southwesterly air was present. Supplement S1 gives a more detailed view on atmospheric transport as seen by the DREAM model (cf. sections 2.8 and 3.1), using the dust load as a tracer for 10 atmospheric motion during the campaign. Figure 2 shows the meteorological conditions at CAO during the campaign. During the first two weeks of the campaign the daily maximum temperature was increasing and relative humidity was decreasing. This was followed by a strong increase of humidity on April 9 and a period of high humidity and lower temperatures with a few millimeters of rain on April 12. Days with rising temperatures and decreasing humidity conditions then followed. The local wind direction at CAO was nearly 70 % of the time from the western sector with wind speeds usually below 5 m s -1 (Figs. 2c 15 and 3). Figure 4 shows the 10-day backward trajectories ending at 1000 m above ground at Orounda. The model used here is the Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT, Stein et al., 2015;Rolph, 2016). Trajectories were calculated at 03:00, 06:00 and 09:00 UTC, in phase with most UAS flights which took place from 04:00 to 12:00 UTC.
The vast majority of air masses that reached the UAS site were transported from the west over the Mediterranean Sea. More than 10 % of the trajectories touched Central and Western Europe, northern Africa and the Atlantic. Additionally, more than 20 5 % of the trajectories showed paths over Northern Europe and the northern Atlantic or the African continent. Since many trajectories originated from the Sahara or adjacent regions, mineral dust particles were episodically transported to Cyprus during the campaign. Fig. 5 displays the aerosol mass concentration (particulate matter, PM) during the campaign. In desert dust that was transported over thousands of kilometers the particles usually still have a diameter of a few microns (e.g. Prospero, 1999) and thus are larger than most other aerosol species. As these large particles make up the bulk of aerosol mass, PM can be 25 considered as a good proxy for mineral dust in the air. By far the highest concentration of PM occured on April 9.

Unmanned Aircraft Systems
Two different types of UAS were used for INP sampling in this campaign. They are described below.
The Cruiser (Fig. 6) is a fixed-wing, medium-size UAS (3.8 m wingspan) with a two-stroke engine and a maximum take-off weight of 40 kg that can carry a payload of up to 10 kg for a maximum flight duration of 3 hours. During the campaign the 30 maximum altitude never exceeded 2.5 km above ground level (ca. 2850 m asl) due to flight plan restrictions. Flight duration was approximately 1.5 hours.
The Skywalker X8 (Fig. 7) has a wingspan of 2.1 m, an electric engine and a maximum take-off weight of 5 kg. It can fly up to 3 km altitude with an endurance of about 1 hour. The maximum altitude reached in this campaign was 2.5 km above ground level with a maximum flight duration of about 1 hour. Compared to the Cruiser it is a much more flexible system as it does not require a runway for take-off and landing. Indeed, it can take off from almost anywhere using a bungee-launching catapult system and it lands on its belly. On the other hand, it is limited to a payload of ca. 2 kg. 5 The flights were operated from a mobile ground control station, which is equipped with state-of-the-art technology to establish a stable communication link with the UAS for reliable data streaming. The UAS airfield site consists of a private paved runway of 12 × 200 m and overhead airspace, which is approved by the Civil Aviation Authority of Cyprus. Both UAS run on the same autopilot system, which enables the UAS to fly automatically in a pre-programmed flight plan.  Both UAS were equipped with a customized inlet system connected to an aerosol sampling unit. A custom build, light weight version (600 g) of a single-sampling PEAC was integrated in the Skywalker X8 UAS. The Cruiser UAS had a 2.5 kg multisampling PEAC installed, which enabled the sampling of up to seven substrates in one flight. Thereby an altitude profile could 20 be sampled from a single flight. The PEAC uses the principle of electrostatic precipitation to ensure that aerosol particles are homogeneously distributed on the silicon sample substrate. In the sampling process, a pump is used to establish a constant air flow (5 l min -1 ), a high voltage source generates an electric field and the aerosol particles are charged negatively by collisions with electrons from a corona discharge. The charged aerosol particles then precipitate on the grounded silicon wafer. The sampling process was controlled remotely from the ground control station and was started as soon as the UAS reached the 25 desired altitude and flight conditions were stable. Sampling length was selected according to criteria like the presence of dust, aerosol concentration, weather conditions, etc., and usually ranged between 6 and 20 minutes (30 to 100 l of sampled air).

Measurements of ice nucleating particles: FRIDGE
During the campaign a total of 42 flights were performed, which generated a total of 52 samples over 19 different days (Cruiser: 7 flights with a total of 17 samples over 6 days, Skywalker: 35 flights with a total of 35 samples over 16 days).
After the flights the samples were analyzed in the isostatic diffusion chamber FRIDGE. A sample substrate is placed on the 30 cold table inside the sample cell. The evacuated cell is then inflated with water vapor. The combination of desired substrate temperature and ice supersaturation defines the exact pressure of water vapor that is inserted into the measurement cell. The water vapor rapidly activates the INP, and ice crystals begin to grow on the surface of these aerosol particles. A CCD camera monitors growth of ice (usually for 100 s) and a LabView controlled software automatically detects changes in the brightness of the images of the emerging ice crystals. For this purpose the real time picture is compared to a reference picture taken prior to the measurement. After a measurement is completed the sample cell is evacuated again and the ice crystals evaporate completely. Then the temperature and relative humidity can be set to new conditions for the next activation.
During this campaign, samples were usually analyzed at −20 • C, −25 • C and −30 • C and relative humidity of 95 %, 97 %, 99 % and 101 % with respect to water, or equivalently 115 to 135 % with respect to ice (see Tab. 1). 5 After the analysis, selected samples were examined by scanning electron microscopy (SEM) to gain information on the chemical composition and morphology.
For a detailed description of the sampling procedure and FRIDGE's measurement principle as well as its limitations and possible caveats, see Schrod et al. (2016). 10 Size and elemental composition of individual particles of selected samples were investigated by scanning electron microscopy using a field emission gun instrument (FEI ESEM Quanta 200 FEG, Eindhoven, The Netherlands), equipped with an energydispersive X-ray microanalysis system (EDX). Almost all ambient particle types are detectable by SEM/EDX analysis, but as SEM is a high vacuum method very volatile organic compounds (VVOC) will be lost. The samples were analyzed automatically by the software-controlled electron microscope (software EDAX/AMETEK GENESIS 5.231). Since the substrates cause a 15 high silicon signal, silicon could not be used for classification of desert dust particles. Instead, aluminum was used for the identification of the Saharan components (in addition with: Mg, K, Ca, Ti and Fe). Besides the main alumosilicate group, Ti-rich alumosilicates and Ca-rich particles (either Ca (Mg) carbonates or mixtures of Ca carbonates with alumosilicates) were classified separately.

20
The Cruiser UAS was equipped with an optical particle counter (OPC, Met One Instruments, Model 212 Profiler) that measured the size distribution of airborne aerosol as a function of their optical diameter. The OPC reports aerosol particle number concentration with 1 Hz resolution in eight different channels ranging from 0.3 to 10 µm. The inlet of the OPC was preheated to keep relative humidity below 50 % to minimize the influence of water absorption onto particles.

25
An automated multiwavelength PollyXT Raman polarization lidar with near-range capabilities (Althausen et al., 2009;Engelmann et al., 2016) was operated at Nicosia. This system emits linearly polarized light at 355, 532, and 1064 nm and has 10 receiver channels. The system is part of PollyNET (Baars et al., 2016) and was measuring around-the-clock autonomously.
EARLINET quality standards were applied, e.g. the depolarization signal was calibrated automatically three times a day. The  Products from the lidar used in this study are time-height-plots of the attenuated backscatter coefficient at 1064 nm and the volume depolarization ratio at 532 nm. All heights are above ground level. The attenuated backscatter coefficient is the calibrated range-corrected lidar signal (see e.g. Wandinger, 2012), i.e. it contains information on the backscattering by molecules and particles and is attenuated by extinction of these two types of scatterers. However, at 1064 nm, molecular scattering and total extinction are very low, so that the attenuated backscatter coefficient at 1064 nm is nearly equal to the real particle 5 backscatter coefficient. Thus, this product gives an indication about the amount of particles in the atmosphere. The lidar signal is calibrated using 2 hour mean profiles of extinction and backscatter coefficient obtained with the Raman method (Ansmann et al., 1992). The volume depolarization ratio is defined as the ratio of the backscattered light in orthogonal and parallel polarization plane with respect to the plane of polarization of the emitted light. It contains information from the whole volume, i.e. molecules and particles. It is a measure of the non-sphericity of the observed scatterers, i.e. the higher the value the more non-10 spherical particles (e.g., dust) are present. At 532 nm Saharan dust typically yields a particle depolarization ratio of about 0.3, whereas spherical scatterers are considered to show a particle depolarization ratio of less than 0.05 (Mamouri and Ansmann, 2014, and references therein). Depolarization from molecular scattering is between 0.005 and 0.006.
Using the particle backscatter coefficient and the particle depolarization ratio at 532 nm, the number concentration of particles with a diameter > 0.5 µm can be estimated with an uncertainty of ± 30 % from lidar observations following the method-15 ology of Mamouri and Ansmann (2016). A dust lidar ratio of 40 sr and a extinction-to-number conversion factor of 0.2 have been considered for that methodology according to the values provided by Mamouri and Ansmann (2016).

Dust transport model DREAM
In this study we used the Dust Regional Atmospheric Model -DREAM (Nickovic et al., 2001;Nickovic, 2005;Pejanovic et al., 2011) driven by the National Centers for Environmental Predictions Nonhydrostatic Multiscale atmospheric Model -NMME 20 (Janjic et al., 2001;Janjic, 2003;Janjic et al., 2011). The NMME-DREAM coupled modeling system has been developed to predict the atmospheric dust process, including dust emission from desert surfaces, horizontal and vertical turbulent mixing, long-range transport and deposition. It solves the Euler-type partial differential non-linear equation for dust mass continuity.
Dust concentration is composed of eight bins with radii ranging from 0.15 to 7.1 µm. Dust emission in the model is proportional to the intensity of the turbulent vertical mixing regimes (laminar, transient and turbulent mixing) near the surface. Specification 25 of dust sources is based on the mapping of the areas that are dust-productive under favorable weather conditions. The USGS land cover data combined with the preferential sources of dust originating from the sediments in paleo-lake and riverine beds (Ginoux et al., 2001) have been used to define barren and arid soils as dust-productive areas. The North Africa -Middle East -Europe domain of the model has a horizontal resolution of 25 km; in the vertical, the model has 28 layers ranging from the surface to 100 hPa. The initial and boundary atmospheric conditions for the NMME model have been updated every 24 hours 30 using the ECMWF 0.5°analysis data. The concentration was set to zero at the "cold start" of DREAM, launched 4 days before April 1, thus permitting a 4-day spin up time to develop a meaningful concentration field at the date considered as an effective model start. After that time, 24-hour dust concentration forecasts from the previous-day runs have been declared as initial states for the next-day run of DREAM.  In the first days of the campaign (March 27 to April 7) no significant dust was observed throughout the lowest 10 kilometers of the atmosphere over Cyprus. On April 2 and 3 a very strong dust outbreak from sources in the western Sahara carried dust far north across Europe, but did not affect the Eastern Mediterranean. On April 7 dust was mobilized from Central Saharan 25 sources (ranging diagonally from northern Niger at 20°N, 10°E to northeastern Libya at 31°N, 23°E, dark brown ellipse in Fig 1b) and was advected by southwesterly flow directly towards Cyprus, causing the major dust event of the campaign on April 8 and 9. This plume impacted Cyprus with a layer of dust arriving at 3 to 4 km altitude above ground early on April 8.
The concentration intensified until it peaked at 450 µg m -3 at 2 to 3 km altitude in the night of April 8 to 9. The highest dust load reached Cyprus at 6 UTC on April 9, when a homogeneous dust layer of 350 µg m -3 between 1 and 6 km altitude swept across  Fig 1b). After this dust layer passed over Cyprus, concentrations decreased while the dust sedimented on April 17.
At about this date dust was mobilized in Central North Africa once more (yellow rectangle in Fig 1b). The corresponding intermediate dust event reached Cyprus on April 21 and showed peak concentrations above 150 µg m -3 at 2 to 5 km altitude.
In the last days of the campaign two minor dust plumes (April 24 and 26) traveled to Cyprus, both coming from sources in northeastern Libya (light brown rectangle in Fig 1b).

Ice nucleating particles
The concentrations of INP measured from the UAS range over five orders of magnitude, when the full range of analyzing conditions is considered. Figure 9 shows the INP concentrations (color coded) as a function of relative humidity with respect to ice (RH ice ). At the lowest ice supersaturations and highest temperature tested (i.e. at −20 • C), concentrations were typically around 0.1 std.l -1 or below. Concentrations increased exponentially with ice supersaturation. At the highest relative humidity and April 16 to 20 due to unfavorable meteorological conditions and maintenance on the UAS.
The contribution of dust to local INP is reflected by the good agreement between INP and specific dust parameters, such as particulate matter (PM), aerosol optical thickness (AOT) and the modeled dust mass concentration. Figure 10 presents the INP concentration at T = −30 • C and RH ice = 135.4 % (i.e. the top row of Fig. 9) from UAS together with coarse mode PM 20 (PM 10 -PM 2.5 ), AOT at 1020 nm measured at CAO and the dust mass concentrations calculated by the DREAM model for the sampling altitude. The mean UAS sampling altitude is given by the color coding of the symbols.
The mean vertical INP profile as derived from averaging INP concentrations in the height bins 0.5 to 1 km, 1 to 1.5 km, 1.5 to 2 km, 2 to 2.5 km is given in Fig. 11. The INP data in each altitude bin show considerable spread. However, the median vertical profile shows the lowest INP concentration close to the surface and a gradual increase towards the top layer. The low 25 INP concentration of around 1 l -1 measured from FRIDGE samples at ground level at CAO is consistent with the observations made by the UAS. This reflects the overall situation during the campaign, in which layers of dust were frequently advected over the island at 2 to 3 km altitude. The lower parts of these layers were sampled by the UAS. The vertical profiles of dust calculated by the DREAM model (right panel of Supplement S1) support this view. The dominance of large scale dust advection can be seen from the correlation between the levels of INP aloft and at the ground (Tab. 2). The highest correlation is found between are correlated to coarse mode PM (R = 0.59, n = 49) measured at CAO at ground level and to the vertically integrated AOT (R = 0.31, n = 49). Furthermore, the peaks of INP and the mineral dust parameters coincide (Fig. 10).

Parameterizations
The establishment of robust empirical correlations between ice nucleating properties and physical characteristics (size spectra) of atmospheric aerosol has been a challenge for decades (Georgii & Kleinjung, 1976), since it might allow to predict INP 5 from aerosol data, which are much easier available than INP measurements. As presented above, we found that the INP concentration from UAS is highly correlated to the number of large particles measured simultaneously by OPC onboard.
Thus, in the the following we will compare our INP measurements to recent particle-based empirical parameterizations of INP (n INP (T k )). However, we have to bear in mind that the nucleation modes addressed by the different methods on which these parameterizations build do not perfectly overlap. From aircraft measurements in various different locations DeMott et al. 10 (2010) derived the relationship (hereafter called D10): The parameters α = 0, β = 1.25, γ = 0.46 and δ = −11.6 were empirically fitted by DeMott et al. (2015). The calibration factor cf has been introduced to separately account for instrument-specific calibration and was set by default to cf = 1, but in special cases it shows a better fit when using cf = 3. 20 Figure 12 compares the INP concentration that we measured over Cyprus (for RH water = 101 %, T k = 253 K, 248 K and 243 K) to the INP concentration that is predicted for the same temperature (but somewhat higher RH) on the basis of the D10 and D15 parameterizations and the n a>0.5 that was measured onboard the UAS. D10 is shown in Fig. 12a and D15 in Fig. 12b.
The range of confidence intervals shown in red (1σ) and gray (2σ) are a little smaller for D10 than for D15. D10 predicts the high concentrations better than D15, but the lower concentrations are predicted poorly. The dust specific parameterization D15 25 on the other hand seems to be better suited to predict the observed measurements. While the slope of Fig. 12b is close to unity, Germany (Schrod et al., 2016) we have recently observed a slope close to unity between FRIDGE and D15, but with a 40 % underestimate (T k = 257 K to 249 K) of the INP observed by FRIDGE as compared to the D15 parameterization.
However, in the case presented here, we do see a better fit between observed FRIDGE INP concentrations and D15 predicted immersion mode INP concentration when empirically setting cf = 0.086 (Fig. 12c), thus lowering the predicted values by a factor of 11. With this alteration the confidence intervals are narrowed markedly. Now 70 % of the data are within a factor of 3.28 and 97 % are within a factor of 14.37 around the 1:1 line. Even when allowing all the parameters to change freely we only achieve a slightly better agreement (Fig. 12d). In this case the parameter γ was set to γ = 0.472 while the cf = 0.086 and the 5 other parameters were kept fixed. Then 70 % of the data are located within a factor of 2.63 and 97 % within a factor of 10.65 around the 1:1 line. Note that the variation of these constants (other than the prefactor cf ) between this best fit case and the values given in D15 is very small.

Ice active fraction and active site density
The nucleating properties of the aerosol encountered over Cyprus may be characterized by its activated fraction (AF ) as well as 10 by the active site density (n s ). Both parameters are a measure of how well the aerosol acted as a seed surface for ice nucleation.
AF stands for the fraction of INP out of the total aerosol particle number (Eq. 3), i.e. it indicates how many particles are needed in total to encounter one active ice nucleus: The n s parameter is an estimate of how many active sites are present upon the total aerosol surface (s a>0.5 ) (Eq.4): 15 n s ≈ n INP s a>0.5 (4) Fig. 13 depicts AF (a) and n s (b) as functions of RH ice in a box plot. Both parameters increase towards higher RH ice in a similar manner, because they originate from the same aerosol and INP values. An exponential increase with RH ice is discerned, which is characteristic for deposition freezing (Meyers et al., 1992). Some measurements are made nearly at the same supersaturation, but at different temperatures (see the set of freezing conditions in Tab. 1). This yields an interesting result, 20 that is somewhat difficult to interpret. It appears that in a sample that is analyzed at roughly the same RH ice but at a different temperature, the higher INP counts are found at the higher temperature, which is opposite to the expectation. Since the number of measurements are relatively small, we cannot say with certainty whether this observation is a real effect or not. Possibly, the higher values at higher temperatures could be explained by condensation freezing starting to matter at RH close to water saturation, whereas at the lower temperatures only the deposition mode is dominating the freezing process. 25 The values of AF and n s compare reasonably well with published measurements performed in regions influenced by mineral dust. E.g. Boose et al. (2016) found the active site density of two month measurement data in the summers of 2013 and 2014 at Izaña in Tenerife to range between 7 × 10 7 -3 × 10 8 m -2 at T = −25 • C, RH ice = 130 % (this study: 2 × 10 7 -7 × 10 8 m -2 at T = −25 • C, RH ice = 129 %) and between 2×10 8 -2×10 9 m -2 at T = −33 • C, RH ice = 135 % (this study: 7×10 7 -2×10 9 m -2 at T = −30 • C, RH ice = 135 %). that a background of mineral dust of fairly uniform nucleating properties was present over the Mediterranean Sea at all times and it was only the burden of dust that affected the INP concentration that we measured.
3.3 Case study: Major dust event of April 9 On April 9 the highest concentration of both dust and INP were observed during the campaign (Fig. 10). Therefore, we present an in-depth analysis of this day. that no measurement could be conducted at −30 • C. This is rather surprising since the sampling altitude was seemingly well above the maximum of the actual dust layer (500 to 1000 m). The depolarization signal (Fig. 14b), on the other hand, showed still a signal of medium strength for a broad range of altitudes up to 3 km, suggesting that a considerable amount of mineral dust might have been collected. Furthermore, we cannot rule out that the observed differences might have been caused by a 20 heterogeneity in the dust spatial distribution between the two different operational sites.
The following paragraphs focus on the first Cruiser flight (red rectangle of Fig These findings agree with the LIDAR measurements discussed above, and are supported by the high concentration of large particles measured by OPC, which was the highest of all flights. The latter translates also into the highest INP concentrations predicted (see Fig. 12), when the parameterizations are applied to these high n a>0.5 values. The vertical profile and size spectra of n a>0.5 during the flight are shown in Figs. 16 and 17. In Fig. 16 the three-dimensional flight track is plotted along with the color coded aerosol number concentration measured with the OPC. It illustrates a typical flight routine.
Right after the take-off the Cruiser UAS was set to ascent mode. While spiraling up the UAS accelerated up to its maximum speed. The aerosol concentration for particles d > 0.5 µm increased from about 15 cm -3 at ground level to 50 cm -3 at 1 km and 5 remained more or less constant until Cruiser reached the maximum altitude (Fig. 17c). These measurements agree well with the aerosol particle concentration retrieved from lidar in Nicosia by the method introduced in Mamouri and Ansmann ( where it decreased linearly with altitude at a rate of −8 • C km -1 (Fig. 17b). The relative humidity showed a dry layer between 0.2 and 1 km and a more humid layer above (Fig. 17a). As soon as the desired elevation was hit (1800 m agl), the electrostatic sampling process started automatically for ten minutes. The UAS took an oval shaped course (Fig. 16), while maintaining the 15 altitude until the sampling ended. After sampling was completed the UAS descended to the second pre-set height (1000 m agl) where the sampling process was repeated.

Electron microscopy of aerosol particles
After the ice nucleation analysis in FRIDGE two selected Si-wafers (samples No. 25  INP concentrations were an order of magnitude lower than aloft, pointing to relatively weak local marine and of terrestrial sources from Cyprus. From these pronounced vertical profiles we conclude that in atmospheric environments that are affected by the large dust sources of the globe, INP measurements performed at ground level will be only of limited significance for the situation several kilometers aloft at cloud level, for which this information is needed. Although the situation encountered by us over the Mediterranean Sea must not be generalized, it is well known as a climatological feature that desert dust regularly 10 travels over distances of thousands of kilometers in this altitude range (Prospero, 1999;Liu et al., 2008).
Several events of long-range transport of Saharan mineral dust with varying intensity were registered during the campaign.
The INP concentration followed various dust proxy parameters, and correlated well with the number of large aerosol particles (d > 0.5 µm) measured in-flight as well as with the dust mass modeled by DREAM. The ice-active site density of the aerosol encountered during the flights compared reasonably well with published data from the Sahara (Boose et al., 2016). SEM 15 analysis of samples taken during the two strongest dust events of the campaign identified about 99 % of the individual aerosol particles with diameters above 400 nm as dust from the North Sahara. Concomitant with the strongest dust event, the INP concentrations reached a peak value of more than 100 active ice nuclei per liter air. The measurements allow no conclusion whether it is dust that actively nucleates ice or particles that are admixed and travel with the dust.
These are the first INP measurements obtained with the technology of unmanned aircraft systems. The combination of an 20 UAS and an offline sampling system with subsequent laboratory analysis of INP is novel and represents a promising alternative to measurements on a research aircraft. Aside from the simplicity as compared to a conventional aircraft mission, the main advantage of the combination of UAS and INP sampling device is its versatility. We were able to adapt rapidly to the current meteorological situation, thereby scheduling flights to accurately target specific small-to-medium scale phenomena such as dense layers of dust. However, admittedly, the small to medium sized UAS have certain limitations in terms of maximum 25 payload, top elevation, flight time, spatial coverage and meteorological conditions (wind speed / precipitation), as well as flight restrictions due to safety of air traffic.
Nevertheless, we encourage other groups to consider UAS as an option to carry out measurements of ice nucleating particles, whether with a similar setup as presented here or in a any different configuration. This tool could broaden the few existing sets of non-surface based INP observations significantly for regions all over the world.