Estimated desert-dust ice nuclei profiles from polarization lidar: Methodology and case studies

A lidar method is presented that permits the estimation of height profiles of ice nuclei concentrations INC in desert dust layers. The polarization lidar technique is applied to separate dust and non-dust backscatter and extinction coefficients. The desert dust extinction coefficients 5 σd are then converted to aerosol particle number concentrations APC280 which consider particles with radius >280 nm only. By using profiles of APC280 and ambient temperature T along the laser beam, the profile of INC can be estimated within a factor of 3 by means of APC-T -INC parameteriza10 tions from the literature. The close relationship between σd at 500 nm and APC280 is of key importance for a successful INC retrieval. We studied this link by means of AERONET (Aerosol Robotic Network) sun/sky photometer observations at Morocco, Cape Verde, Barbados, and Cyprus during desert 15 dust outbreaks. The new INC retrieval method is applied to lidar observations of dust layers with the spaceborne lidar CALIOP (Cloud Aerosol Lidar with Orthogonal Polarization) during two overpasses over the EARLINET (European Aerosol Research Lidar Network) lidar site of the Cyprus 20 University of Technology (CUT), Limassol (34.7◦ N, 33◦ E), Cyprus. The good agreement between the CALIOP and CUT lidar retrievals of σd, APC280, and INC profiles corroborates the potential of CALIOP to provide 3-D global desert dust APC280 and INC data sets. 25


Introduction
The evolution of the ice phase in initially liquid-water clouds is still poorly understood and thus not well considered in climate models.Trustworthy predictions of the overall indirect aerosol effect on climate are impossible as long as the important branch of heterogeneous ice formation, the subsequent production of rain, and the associated removal of water from Correspondence to: R. E. Mamouri (rodanthi.mamouri@cut.ac.cy) the atmosphere is not properly described in atmospheric circulation models.Aircraft-based field campaigns in cloudy environments are performed to improve the basic process un-35 derstanding on heterogeneous ice formation at given meteorological and aerosol conditions.Laboratory studies and in situ aerosol characterization provide important knowledge on the influence of a variety of natural and anthropogenic aerosol types on cloud glaciation.Lidar and radar-based re-40 mote sensing allow a continuous monitoring of co-occurring aerosol and cloud fields in their natural environment and thus a detailed study of the evolution of the ice phase in liquid and mixed-phase cloud layers and the impact of aerosol layers on these processes.

45
Ground-based active remote sensing is, in general, of great importance for observational studies of aerosol-cloud interactions because of its unique potential to observe aerosol layers and clouds (from base to top) with high vertical and temporal resolution and this over long time periods (Ansmann 50 et al., 2005;Ansmann et al., 2009;Illingworth et al., 2007;Seifert et al., 2010;Martucci and O'Dowd, 2011;de Boer et al., 2011;Kanitz et al., 2011;Wandinger et al., 2012;Bühl et al., 2013;Schmidt et al., 2014).Continuously running stations can provide dense aerosol-cloud data sets for statistical 55 analysis for all seasons of the year.Organized in networks, regional aspects regarding aerosol conditions (varying mixtures of urban haze, biogenic aerosols, smoke, marine and dust particles), and orographic and meteorological influences can be studied in large detail.60 However, further efforts are required to improve the retrieval capabilities of lidar-radar supersites.Regarding heterogeneous ice formation it is of interest to explore the potential of polarization lidar to deliver height profiles of ice nuclei concentration (INC) up to cloud base as well as around 65 and above cloud layers.Such an INC profiling would open new ways to explore the evolution of mixed-phase clouds and the role of aerosols in this context.Applied to spaceborne CALIOP (Cloud Aerosol Lidar with Orthogonal Polar-help to establish global 3-D maps of INC.CALIOP is part of the CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation) mission (Winker et al., 2009).
A first attempt to use particle optical properties measured with lidar for an INC estimation was undertaken during the Saharan Mineral Dust Experiment (SAMUM) in May-June 2006 (Ansmann et al., 2008).It was demonstrated that the retrieval of the particle number concentration APC 250 of large desert dust particles with radius r >250 nm from lidar profiles of the dust extinction coefficient σ d measured at 532 nm 80 wavelength is possible with good accuracy.This is an essential prerequisite for the next step, the estimation of INC.The large particle fraction is assumed to be the reservoir for most favorable ice nuclei (DeMott et al., 2006(DeMott et al., , 2010)).However, to estimate INC from APC 250 in observed dust layers, a rather crude assumption on the ratio of APC 250 to INC of 30±20 was used in this first, preliminary approach of Ansmann et al. (2008).The study of the correlation between particle extinction and APC 250 was based on the SAMUM Aerosol Robotic Network (AERONET) photometer obser-90 vations of the 500 nm aerosol optical thickness (AOT) and the corresponding column-integrated particle size distribution from which the column APC 250 values were derived.
After the eruption of the Icelandic volcano Ejyafjallajökull in April 2010, Seifert et al. (2011) investigated the role of 95 volcanic particles to serve as ice nuclei and continued the discussion regarding the relationship between the particle extinction coefficient, number concentration APC 250 and INC in the case of volcanic particles, which were found to be very efficient ice nuclei.The authors used the same parameteriza-100 tion to obtain APC 250 from the lidar-derived σ d values as Ansmann et al. (2008) for desert dust and also applied very rough assumptions on the ratio of APC 250 to INC to provide estimates for INC in the case of volcanic aerosol layers.
The recently published INC-APC 250 parameterization 105 schemes (DeMott et al., 2010(DeMott et al., , 2015) ) motivated us now to intensify our effort to develop a quantitative lidar-based method for INC profiling.These parameterizations are developed for immersion freezing which is the most important heterogeneous ice nucleation mechanism (Ansmann et al., 110 2008;de Boer et al., 2011;Murray et al., 2012).We start with an INC retrieval scheme for desert dust.Mineral dust from desert areas belongs to the most important aerosol components in the atmosphere regarding heterogeneous ice formation (Richardson et al., 2007;Kamphus et al., INC retrieval a clear relationship between the lidar-derived extinction coefficient σ d and the particle number concentration APC 250 is of key importance.Such a close correlation is given for desert dust because desert dust particles always shows a pronounced bi-modal volume size distribution con-125 sisting of a fine mode (particles with r <500 nm) and a pronounced coarse mode (particles with r >500 nm).This feature significantly facilitates the APC 250 retrieval and subsequent INC estimation as will be discussed in Sect 3. The correlation study in Sect. 3 is based on several combined li-130 dar and AERONET photometer data sets collected during the two desert dust SAMUM campaigns in Morocco (SAMUM-1, 2006, Heintzenberg, 2009) and Cape Verde (SAMUM-2, 2008, Ansmann et al., 2011) and the Saharan Aerosol Longrange Transport and Aerosol-Cloud Interaction Experiment 135 (SALTRACE, Ansmann et al., 2014) conducted in Barbados in the summer of 2013.Furthermore, we include a four-year AERONET data set collected at Limassol, Cyprus, in our investigation of the relationship between dust extinction and the number concentration of large dust particles.Cyprus is 140 frequently crossed by major desert dust outbreaks from the Sahara and the Middle East deserts (Nisantzi et al., 2015).For the sake of simplicity (see Sect. 3) we will use APC 280 (number concentration of particles with r >280 nm) in our study instead of APC 250 which was introduced by DeMott The lidar station of the Cyprus University of Technology (CUT) at Limassol (34.7 • N, 33 • E, 50 ma.s.l.) is located about 150 km south of Turkey and 250 km west of Syria and belongs to EARLINET (Pappalardo et al., 2014).The lidar is described by Mamouri et al. (2013) and enables us to de-165 termine height profiles of the particle backscatter coefficient and particle linear depolarization ratio at 532 nm.
In the analysis of the CUT lidar data as described in Sect.3.1 actual height profiles of temperature and pressure profiles are required for the Limassol region.The operational systems GDAS (Global Data Assimilation System) of the National Weather Service's National Centers for Environmental Prediction (NCEP) provide these meteorological parameters.NOAA's Air Resources Laboratory (ARL), (https://www.ready.noaa.gov/gdas1.php)NCEP model GDAS output archives contain basic meteorological fields for the horizontal wind components, temperature, and humidity for specific pressure levels.

CALIOP
The spaceborne lidar CALIOP is described by Winker et al. (2009).Numerous validation projects have been carried out (e.g., Mamouri et al., 2009) to demonstrate the capability of this lidar to provide accurate aerosol backscatter profiles throughout the troposphere.The spaceborne aerosol/cloud lidar measures polarization sensitive backscatter signals at 532 nm.CALIOP aerosol products include height profiles of the 532 nm particle backscatter coefficient, extinction coefficient, particle backscatter coefficient determined from the cross-polarized 532 nm signal channel, and the particle linear depolarization ratio.We use the CALIOP level 2 version 3 aerosol profile products.Besides the available profile of the particle depolarization ratio, we calculated this quantity in addition from the individual profiles of the crosspolarized and total 532 nm particle backscatter coefficients after smoothing of these individual profiles as suggested by Tesche et al. (2013).
For the CALIPSO level 2 aerosol profiles product, vertical profiles of the mean values of meteorological parameters along the flight track are provided in addition.These data are given for the midpoint of each range bin of the CALIOP profile.The meteorological parameters such as temperature, pressure, and relative humidity are derived from the God-  Draxler and Hess (1997, 1998), and Draxler (1999).

Method
In this section, we present the INC retrieval scheme.The data 245 analysis consists of three parts.In the first part (Sect.(2010,2015), which are based on comprehensive laboratory studies and field campaigns con-260 ducted during the last 14 years.Table 1 provides an overview of all steps of the method.The different steps will now be explained in the following subsections.CUT lidar at Limassol.The retrieval of the shown particle backscatter coefficient β p at 532 nm (step 1 in Table 1) and particle linear depolarization ratio is discussed in detail by Mamouri et al. (2013) and Mamouri and Ansmann (2014).Fine-mode soil dust dominated particle backscattering on 270 27 September 2011 in Fig. 1, while a major dust outbreak from the Middle East deserts was observed on 29 September 2011.The particle linear depolarization ratio at 532 nm is used to separate the non-dust backscatter coefficient from the dust backscatter coefficient β d (step 2 in Table 1).As explained in Mamouri and Ansmann (2014), two methods are available for the identification and quantification of the dust contribution to total particle backscattering and extinction.The so-called one-step method can be used to separate non-dust particle backscattering from total (fine and coarse) dust backscattering, whereas the two-step method allows to distinguish backscattering from non-dust aerosol, fine-mode dust, and coarse-mode dust particles.In the case of desert dust observation on 29 September 2011, however, both methods lead to almost the same profile for the total (fine plus coarse-mode) dust backscatter coefficient and the respective total dust extinction coefficient in the pronounced dust layer above 500 m height (above the polluted boundary layer).The one-step and two-step solutions for the dust extinction coefficient are shown in the right panel of Fig. 1.On 27 September 2011, we observed a smoke-dust plume over Limassol with dominating fine-mode smoke as well as fine-mode soil dust (Mamouri and Ansmann, 2014;Nisantzi et al., 2014).
A pronounced coarse mode was missing.At these conditions, only the two-step method delivers correct results.The one-step and two-step retrieval profiles do not match in this case.To avoid such complicated situation in this first paper on lidar-derived INC retrievals, we concentrate on desert dust plumes only, and use the well-established one-step method in the case studies presented in Sect. 4. The source region of the observed dust can easily be identified by means of backward trajectory analysis.
The dust extinction coefficient σ d at 532 nm (step 3 in Table 1) shown in Fig. 1 is obtained from the dust backscatter coefficient β d by multiplying β d with the dust lidar ratio which was obtained before as the optimum dust lidar ratio from the particle backscatter retrieval after Mamouri et al. (2013).The dust lidar ratio is 43 sr for the 27 September soil-dust case, and 39 sr for the desert dust observation on 29 September 2011.Both dust lidar ratios are characteristic for the Middle East region (Schuster et al., 2012;Mamouri et al., 2013).The uncertainty in the desert dust extinction coefficients is of the order of 20 % (Tesche et al., 2009a;Mamouri and Ansmann, 2014) with a range from 10 % for strong dust outbreak and 30 % for moderate to background dust conditions.

Part 2: From dust extinction to APC 280
In the next step (step 4 in Table 1), we estimate the number concentration of large particles, APC 280 (denoted as n d,280 in Eq. ( 1)) from the 532 nm dust extinction coefficient σ d in Mm −1 by means of with the conversion factor c d,280 = 0.673 in Mm/cm 3 .This conversion factor is obtained from a comprehensive analysis of AERONET sun/sky photometer observations during strong Saharan dust outbreaks reaching Ouarzazate, Morocco, Praia, Cape Verde, Barbados, and Limassol, Cyprus.
An overview of the AERONET measurements considered in our study is shown in Fig. 2. The column-integrated value of the particle number concentration APC 280 is plotted against
In the following we assume equal dust extinction at the laser wavelength of 532 nm and the AERONET photometer wavelength of 500 nm, i.e, we ignore a weak wavelength dependence of dust backscattering and extinction in the 500- given radius interval (or for the discrete radius point r i ) by the volume of a single particle with radius r i .
The column APC 280 value is then simply given by the sum of all particles of the radius classes with radii r i ≥ 330 nm and includes therefore all particles with roughly r >280 nm 350 (the part of the size spectrum on the right side of the dashed vertical line in Fig. 3b), when taking the width of the radius interval around 330 nm into account.The radius interval for r 8 = 330 nm roughly represents the radius interval from 280 to 380 nm.As mentioned before, we use APC 280 instead of 355 APC 250 as originally suggested by DeMott et al. (2010) for the sake of simplicity.We simply add the particle number concentrations of the radius classes 8-22 and avoid to analyze the radius class 7 (200-280 nm radius interval) for the contribution of 250-280 nm particles to the class-7 particle 360 number concentration.APC 250 is about 10-15% larger than APC 280 .
As can be seen in Fig. 2, a good correlation between the 500 nm AOT and the column-integrated APC 280 was found for all dust observations in Morocco, Cape Verde, and Bar-365 bados.For dust outbreaks towards Limassol, Cyprus, the correlation was found to be comparably low because of the contribution of omnipresent anthropogenic aerosol pollution to the total AOT in the eastern Mediterranean.The correlation between the coarse-mode AOT and column APC 280 is 370 much better.The coarse-mode AOT is widely determined by light extinction by dust particles.From all the AERONET data in Fig. 2 we conclude that a clear and close relationship between desert-dust APC 280 and dust-related extinction coefficient exists.According to the regression line in Fig. 2 375 the column APC 280 is approximately given by AOT(500 nm) multiplied by a factor of 6.85 × 10 11 m −2 .
To translate the column-related findings in Fig. 2 into scales of particle extinction coefficient (measurable with lidar) and respective particle number concentration, we simply used the layer depth information from the lidar observations in Morocco, Cape Verde, and Barbados.The dust layers were typically well mixed and reached from the surface to 5-6 km height (Morocco, summer 2006), from the surface to 1.0-1.5 km height (Cape Verde, winter 2008), and from 1.0 to 4-6 km height (Barbados, summer 2013).We divided the individual AOT and column APC 280 pairs by the respective layer depths and obtained in this way the correlation shown in Fig. 4. We ignore here a small contribution of marine particles (<20% contribution to the 500 nm AOT during the major dust outbreak situations) to the dust observations at Cape Verde and Barbados.The linear regression yield a clear relationship between the dust extinction coefficient and APC 280 (Eq.( 1), step 4 in Table 1).The correlation coefficient is 0.91.The slope of the regression line in Fig. 2 of 0.685 in units of 10 12 m −2 is slightly steeper than the respective one in Fig. 4 of 0.673 in Mm/cm 3 because the regression in Fig. 2 includes the Cyprus data (coarse-mode AOT values).
The overall uncertainty in the lidar-derived APC 280 values is estimated to be of the order of 30 %, keeping the 20 % uncertainty in the determination of the dust particle extinction coefficient σ d in mind and assuming a further uncertainty of the order of 20 % in the conversion of σ d into APC 280 values.This 20% uncertainty value includes a potential error of 10-15% of the AERONET-derived particle volume size distributions obtained by applying an inversion algorithm to the basic spectral AOT and sky radiance observations.From studies of Toledano et al. (2011) and Müller et al. (2012), which compared AERONET size distributions with respective airborne in situ observations during the SAMUM campaigns, we conclude that the uncertainty in terms of APC 280 obtained from the AERONET data is of the order of <20% or less.There may be situations with giant dust particles (>15 µm in radius), for example during dust storms close to the Saharan source region.These giant dust particles are not considered in the AERONET retrieval of the volume size distribution.However, the impact of these missing giant particles on the AERONET results seems to be small as the harmonic overall correlation between APC 280 and σ d in Fig. 4 suggested which includes SAMUM-1 results (Morocco, close to the dust source) as well as SALTRACE observations (Barbados, 5000-8000 km west of the main dust sources).
It should also be clearly mentioned in this context that there is no real alternative to AERONET observations shown in Figs. 2 and 4. Alternative measurements could be airborne insitu observations of aerosol microphysical and optical properties.But in situ observations always include significant manipulations of the probed aerosols and airborne observations are expensive and thus rare from the statistical point of view.Only AERONET can provide statictically dense, high-quality data sets of optical and microphysical properties of aerosol particles for the same air column at undisturbed ambient conditions.Exactly those data are needed for our correlation study.Well-established and approved methods are available to derive particle size distribu-435 tions with high accuracy and uncertainties below 10-20 % (Dubovik and King, 2000;Dubovik et al., 2006).Complementary methods can be used in addition to check the quality of the microphysical products (O'Neill et al., 2003) and the consistency between the retrieved optical and microphysical   3)).In the result section (Sect.4), we use these immersion-freezing-based parameterizations even for higher as well as lower temperatures.According to Wex et al. (2014) ice nucleation for coated mineral dust particles (coated with natural and/or anthropogenic soluble material) can be described as immersion freezing as well.Above the deliquescence relative humidity, additional water is added to the coating and a solution shell forms around the particles, causing them to nucleate ice from concentrated solutions via the immersion freezing pathway, taking a freezing point depression into account.
Regarding uncertainties in the INC estimation, DeMott et al. ( 2010) pointed out that Eq. ( 2) allows a prediction of INC within an uncertainty range (standard deviation) of less than a factor of 5, with the remaining variability apparently due to variations in aerosol chemical composition or other factors.By means of Eq. ( 2), 62 % of the observational data collected during nine field studies could be predicted within a factor of 2. These field studies were performed at a variety of locations around the globe over a 14-year period.DeMott et al. ( 2010) further pointed out that an INC uncertainty of an order of magnitude is still acceptable for cloud process modeling.The uncertainties (standard errors) are lower and within a factor of 2 when using Eq. ( 3) (DeMott et al., 2015).

500
Regarding the overall uncertainty of our lidar-based INC retrieval we can summarize that the profile of APC 280 can be derived from the dust extinction coefficients with a relative error of the order of 30% and that the estimation of INC profiles is therefore possible within an overall factor of 3 by 505 applying Eq. (3).
Figure 5 provides an overview of the retrieval approach (steps 3-5 in Table 1) for the dust outbreak on 29 September 2011 shown in Fig. 1.Dust extinction values range from 30-400 Mm −1 .The red APC 280 curve in Fig. 5 3) and by a factor of 20 when using the more general aerosol parameterization (Eq.( 2)).Thus, a ten-degree decrease in temperature (equivalent to about 1000-1500 m height change in the free 520 troposphere) during lifting of air particles in a convective cloud tower leads to an enormous increase of the potential of a given dust load to initiate ice nucleation via immersion freezing.

525
We applied our INC retrieval scheme to two CALIOP observations in the eastern Mediterranean close to Limassol, Cyprus.One of these overpasses took place during a strong Saharan dust outbreak on 1-2 June 2013.Traces of dust reached cirrus level (8-10 km height).During the second overpass, mineral dust was advected from the deserts in the Middle East on 1-2 November 2013.This second case can be regarded as representative for typical dust outbreaks with dust layers mainly in the lower free troposphere at heights between 1 and 5 km (Papayannis et al., 2008(Papayannis et al., , 2009 Figure 8 compares the basic optical properties derived from the ground-based CUT lidar and the CALIOP obser-550 vations.The nearest horizontal distance of the CALIOP laser foot print to Limassol was 45 km.We used HYSPLIT forward and backward trajectory analysis to identify the air mass which was seen by both lidars.This air mass crossed Limassol about one hour before reaching the CALIPSO flight 555 track.Accordingly we selected the CUT lidar signal averaging period from 22:28-23:28 UTC.CALIPSO crossed the area at 23:53 UTC.The small deviations between the optical properties derived from the CALIOP and CUT lidar observations are mainly caused by different data analysis schemes, 560 assumptions on input parameters, and the different signal averaging periods (one hour in the case of the CUT lidar, a few seconds in the case of CALIOP) (Mamouri et al., 2009).Different aerosol conditions in the lowest 500 m over the city of Limassol and over the open Mediterranean Sea (CALIOP) 565 may widely explain the differences at heights below 500 m.
The CALIOP products (backscatter and extinction coefficients) are determined by using lidar-ratio look-up tables (Omar et al., 2009;Kanitz et al., 2014).For desert dust scenarios, the lidar ratio is set to 40sr.Our own measurement 570 yield a particle (dust plus non-dust) lidar ratio of 55 sr for the total tropospheric column following the complex data analysis procedure described by Mamouri et al. (2013).The lidar ratio of 55 sr for this desert-dust-dominating scenario is in full agreement with respective SAMUM lidar-ratio obserser-575 vations (Tesche et al., 2009b).
For the INC retrieval, we smoothed the height profile of the particle backscatter coefficient measured with CALIOP with a vertical smoothing length of 600 m.To reduce the noise in the CALIOP depolarization ratios, we also smoothed 580 the basic cross-polarized and total particle backscatter coefficient profiles with 600 m vertical window length before we calculated the volume depolarization ratios and then the particle depolarization ratios.This procedure was recommended by Tesche et al. (2013).The left panel in Fig. 9 shows the 585 smoothed basic CALIOP products (backscatter coefficient, depolarization ratio).The right panel presents the extinction profiles for desert dust particles and for remaining non-dust aerosol (marine and anthropogenic) particles calculated from the smoothed CALIOP profiles.The separation of the dust and non-dust optical properties was explained in Sect.3.1 In the conversion of the backscatter coefficients to the respective dust and non-dust extinction coefficients we used lidar ratios of 50 sr for non-dust particles and 55 sr for Saharan dust particles.(Ansmann et al., 2008;Murray et al., 2012).Consequently, for temperatures from −5 to −15 • C the INC estimates are slightly higher by using Eq. ( 2) compared to the INC values from Eq. (3).For higher temperatures (0 to −5 • C) the INC values are at all not trustworthy because these temperatures are far outside the temperature range for which the parameterizations (Eqs.( 2) and ( 3)) were developed.Differences between the INC profiles derived from the CUT lidar and CALIOP observations are likewise small and mainly caused by differences in the temperature profiles over Limassol and above the CALIOP laser foot print.
This unique Saharan dust outbreak with dust traces up to 10 km provides a favorable opportunity to continue the discussion on the rather strong temperature influence on INC and the consequences for cloud glaciation.As can be seen, although the dust number concentration APC 280 is almost constant with height in the layer from 6 to 8.5 km height, the INC values increase by a factor of 1000, from 6 km (−12 • C) to 8.5 km (−30 • C) when the mineral dust INC parameterization (Eq.( 3)) is applied.This means that even traces of desert dust occurring at the base of an evolving convective cumulus tower can develop an enormous potential to glaciate the cloud system when lifted by updrafts over several kilometers.

CALIOP overpass during a dust outbreak from the Middle East in November 2013
A Middle East desert dust outbreak was observed on 1-2 November 2013. Figure 11 provides an overview of the dust situation in terms of CALIOP attenuated backscatter observations.Figure 12 presents the respective HYSPLIT backward trajectories arriving at Limassol on 2 November 2013, 00:00 UTC and shows the source regions of the dust (deserts in Saudi Arabia, Iraq, and Syria).In contrast to the foregoing case study, dust was detected at heights below 4 km only.
The CALIOP and CUT data analysis was performed in the same way as described in Sect.4.1.The noisy CALIOP data 645 profiles, averaged over 45 km horizontal length, had to be smoothed with 600 m vertical window length.We generally used particle lidar ratios around 40 sr (for CALIOP as well as for the CUT lidar observations) in this case of a major Middle East desert dust outbreak.These low lidar ratios around 40 sr 650 are representative for Midde East desert dust (Schuster et al., 2012;Mamouri et al., 2013).
Figure 13 compares the CUT and CALIOP lidar findings for this dust outbreak in terms of dust extinction coefficient σ d , APC 280 , and INC.Because the CALIOP laser foot print 655 was 180 km east of Limassol and both (CALIOP and CUT) observations were performed within a relative small time window of less than two hours (23:45 UTC on 1 November to 01:00 UTC on 2 November), different air masses were definitely observed (Pappalardo et al., 2010).This ex-660 plains the differences between the two observations in terms of σ d and APC 280 .Regarding INC, temperatures along the CALIPSO flight track were up to 2.5 • C lower in the free troposphere above 600 m height compared to Limassol temperatures so that the CALIOP-derived INC values were con-665 siderably higher because of the lower temperatures and the, on average, higher APC 280 values.
However, the 0 • C level was above 3 km and the −5 • C level was reached at 4 km height so that only a few INC estimates for the uppermost part of the dust layer could be cal-670 culated.Such conditions were already observed during the SAMUM-1 campaign in southern Morocco (Ansmann et al., 2008).Ice formation in altocumulus layers developing at the top of such dust layers was found to be almost impossible at all because of the high temperatures throughout the dust 675 layers.Only when cumulus convection was strong enough so that cloud parcels could penetrated deeply into the free troposphere, ice formation was observed.
Figure 13 contains further INC profiles.We shifted the GDAS and GMAO temperature profiles by 22-23 • C so that 680 the surface temperature was 0 • C at both sites.We simulated these profiles to visualize the consequences (in terms of INC increase) of a horizontal transport of air masses towards colder areas, i.e., when such a warm dust plume is, e.g., advected to the north (towards Turkey, Black Sea, Rus-685 sia, and Scandinavia) and gets colder by radiative cooling and mixing with colder air during the long-range transport.As be seen, the potential of a given dust layer to initiate ice formation in water clouds steadily and strongly increases.As discussed already, the ice nucleation efficiency increases by 690 three to four orders of magnitude if a dusty air mass is cooled by about 20 • C.These INC levels are further increased by the fact that air masses are usually also lifted by several kilometers during long-range transport over thousands of kilometers (Mattis et al., 2002;Ansmann et al., 2003).
A method has been introduced that permits the estimation of desert-dust-related INC profiles from polarization lidar measurements at 532 nm wavelength.Of key importance for a successful INC retrieval is a close relationship between the lidar-derived dust extinction coefficient and the number concentration of large particles with r >280 nm, APC 280 .Based on unique desert dust field observations and long-term studies with AERONET photometers we demonstrated that this close link is given for desert dust.The uncertainties for the different lidar products are of the order of 20 % for the derived dust extinction coefficients, 30 % for APC 280 , and within a factor of 3 for INC when using the recently developed dust INC parameterization of DeMott et al. (2015).
The approach paves the ground for an INC vertical profiling as a support to ground-based and airborne in situ IN characterization and to conduct a global, vertically resolved mapping of dust-related INC in the framework of spaceborne lidar missions such as CALIPSO (NASA) or Earth-CARE (European Space Agency, Illingworth et al., 2015).This was demonstrated by two case studies of spaceborne CALIOP and ground-based CUT lidar measurements during overpasses of CALIOP over the eastern Mediterranean.Because there are already several dust-related global studies based on CALIOP observations (Liu et al., 2008;Tsamalis et al., 2013;Amiridis et al., 2013;Luo et al., 2015) with focus on geometrical and optical properties of desert dust, it should be a comparably easy effort to do the next step towards the characterization of the found aerosol conditions in terms of APC 280 and INC.Such a global INC characterization may allow an improved consideration of heterogeneous ice formation in atmospheric circulation models.However, we recommend to establish a global aerosol data set in terms of APC 280 rather than of INC, and to combine the APC 280 data set with actual temperature fields from numerical weather prediction models.This is a more flexible approach to account for the rather large influence of ambient temperature conditions on the efficiency of any given aerosol layer to initiate heterogeneous ice nucleation.Furthermore, the accuracy of desert-dust APC 280 data sets is comparably high, in contrast to the uncertainties in the INC estimates.
As an outlook, we need to study to what extent and with what uncertainty the method presented here can also be used for an INC profiling during situations with dominating finemode aerosol such as urban haze, biomass burning smoke, or even soil dust injected into the atmosphere during fire events (Nisantzi et al., 2014), i.e., when a pronounced coarse mode is missing.At such conditions, a good and clear correlation of the particle extinction coefficient and APC 280 may be no longer given so that a good INC estimation is difficult.We also may test alternative INC retrieval approach which relate INC to particle surface area rather than to APC (Niemand et al., 2012;Steinke et al., 2014;Hande et al., 2014).
Acknowledgements.The authors thank the CUT Remote Sensing Laboratory for their support.We thank Patric Seifert (TRO-750 POS) for fruitful comments, and the SAMUM and SALTRACE lidar/photometer groups for taking care of all lidar and photometer observations.We are grateful to AERONET for highquality sun/sky photometer measurements in Cyprus, Morocco, Cape Verde, and Barbados.The authors gratefully acknowledge 755 the NOAA Air Resources Laboratory (ARL) for the provision of the HYSPLIT transport and dispersion model and/or READY website (http://www.ready.noaa.gov)used in this publication.We thank the NASA Langley Research Center and the CALIPSO science team for the constant effort and improvement of the CALIPSO 760 data.The research leading to these results has also received scientific support from the European Union Seventh Framework Programme (FP7/2011-2015) under grant agreement no.262254 (AC-TRIS project).We acknowledge funding from the EU FP7-ENV-2013 programme "impact of Biogenic vs. Anthropogenic emissions 765 on Clouds and Climate: towards a Holistic UnderStanding" (BAC-CHUS), project no.603445.bein, H., Kollias, P., Kubota, T., Nakajima, T., Nakajima, T. Y., Nishizawa, T., Ohno, Y., Okamoto, H., Oki, R., Sato, K., Satoh, M., Shephard, M., Wandinger, U., Wehr, T., and Zadelhoff, G.-J.: THE EARTHCARE SATELLITE: The next step forward in global measurements of clouds, aerosols, pre-Table 1.The five steps required to obtain the dust-related INC profile from the profile of the particle backscatter coefficient βp measured with polarization lidar.The determination of βd and σd is explained in Sect.3.1.In Sect.3.2, the retrieval of the number concentration of large particles APC280 is described, and the INC estimation is finally outlined in Sect.3.3.★ ▲ ▲ ▲ ▲ ▲ ▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ★                              ★ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ★ ★ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ★ introduction into the instrumentation, field activities and auxiliary meteorological data in Sect.2, the INC retrieval method is explained in Sect.3. Sect. 4 contains applications of the new INC retrieval scheme to CALIOP ob-150 servations performed during overpasses over the European Aerosol Research Lidar Network (EARLINET) station of Limassol, Cyprus.The CALIOP and Limassol lidar results (optical properties, APC 280 , INC) are compared for a major and unique Saharan dust outbreak with dust layers up to 155 10 km height and for another dust outbreak from the Middle East deserts.Summarizing and concluding remarks are given in Sect. 5. 2 Instrumentation and auxiliary meteorological data 2.1 CUT lidar 160

3. 1
Figure 1 presents two examples of dust profiling with the 265 range.The determination of the column APC 280 values from the basic AERONET information (column particle volume size distribution) is explained in Fig. 3. Particle volume size distributions measured at Limassol during two dust outbreaks 335 from the Middle East (1 November 2013 and 29 September 2011) and two outbreaks from the Sahara (11 March and 2 June 2013) are shown in Fig. 3a.The volume particle size distribution is retrieved for 22 logarithmically equidistant discrete radius points r i with index ifrom 1-22.The 340 particle radius range from r 1 = 0.05 µm to r 22 = 15 µm is covered.Each radius r i represents a radius interval of logarithmically equal width of about 0.27.To obtain the particle number concentration for each individual radius class shown in Fig. 3b, we simply divided the volume concentration of a 345 3: Estimation of INC from APC 280 The retrieval of APC 280 from σ d is the basic lidar contribution to the estimation of the INC profiles.Part 3 now provides the link to the published INC parameterizations (step 5 in Ta-445 ble 1) gained from comprehensive INC laboratory and field studies.The INC parameterizations introduced by DeMott et al. (2010, 2015) hold for standard (std) pressure (p 0 = 1013 hPa) and temperature (T 0 = 273.16K) conditions so that we have 450 to convert each profile value APC 280 (p z , T z ) given for ambient pressure p z and temperature T z at height z to APC 280 (p 0 , T 0 ) by using the factor (T z p 0 )/(T 0 p z ).DeMott et al. (2010) introduced a so-called global (aerosol-type-independent) INC parameterization: 455 n IN (p 0 ,T 0 ,T z ) = a(273.16− T z ) b × n a,280 (p 0 ,T 0 ) [c(273.16−Tz)+d] (2) with n a,280 in stdcm −3 representing APC 280 , n IN in stdL −1 representing INC, a = 0.0000594, b = 3.33, c = 0.0265, d = 0.0033, and temperature T (z) in K (and < 273.16 K).As 460 mentioned, we use n d,280 instead of n d,250 as given in the original formula of DeMott et al. (2010).DeMott et al. (2015) recently introduced another INC parameterization scheme which is explicitly developed for mineral dust: ) with the so-called atmospheric correction factor f d =3, a d = 0.074, b d = 3.8, c d = 0.414, and d d = −9.671.Finally, we transfer the obtained INC values n IN (p 0 ,T 0 ,T z ) to 470 the ones for ambient pressure and temperature conditions, n IN (p z ,T z ), by multiplying n IN (p 0 ,T 0 ,T z ) with the factor (T 0 p z )/(T z p 0 ).According to DeMott et al. (2010, 2015), Eqs.(2) and (3) can be used to estimate INC for immersion freezing pro-475 cesses and are applicable to the temperature range from −9 to −35 • C (Eq. (2)) and −21 to −35 • C (Eq. ( describes the linear increase of the large particle number concentration with increasing dust particle extinction coefficient σ d .To illustrate the large influence of the ambient temperature T z on INC, INC curves for −15 • C, −25 • C, and −35 • C are plotted in Fig. 5.As can be seen, a temperature decrease by 515 10 degrees causes an increase in the INC concentration by two orders of magnitude when using Eq. ( and CUT lidar observations during a Saharan dust outbreak in June 2013 An overview of the dust and cloud observations of the spaceborne CALIOP lidar in the night of 1-2 June 2013 is shown 540 in Fig. 6.The spaceborne lidar crossed eastern Ukraine (52-56 • N), the Black Sea area (44-52 • N), Turkey (36-42 • N), the eastern Mediterranean Sea (30-36 • N), Egypt (22-32 • N), and Sudan (< 22 • N) within 12.5 min (corresponding to a distance of about 5000 km).The backward trajectory anal-545 ysis in Fig. 7 indicates the southern parts of the Sahara as sources for the dust observed in the middle and upper troposphere over the eastern Mediterranean.

Figure 10
Figure 10 presents the results for this CALIOP overpass case in terms of dust extinction coefficient σ d , APC 280 , and INC.The respective products derived from the ground-based CUT lidar observations are shown in addition.Because of the high temperatures over the eastern Mediterranean in the beginning of June 2013 with surface temperatures close to 30 • C, INC values are only given for the upper part of the Saharan dust layer where temperatures <0 • C where given.Significant differences are found for the different INC parameterizations.Compared to the INC profile after Eq. (2) (global aerosol INC parameterization), significantly higher INC values are obtained with Eq. (3) (mineral dust parameterization) for temperatures < −20 • C.This reflects that desert dust particles are known to be very efficient ice nuclei at temperatures < −20 • C, but not at temperatures > −15 • C

Fig. 4 .
Fig. 4. Relationship between dust layer mean 500 nm extinction coefficient (EC) and dust layer mean APC280 for observations taken during three desert dust field campaigns at Morocco, Cape Verde, and Barbados.The linear regression yields c d,280 = 0.673±0.07Mm cm −3 .The correlation coefficient is 0.915.c d,280 is used in Eq. (1).

Fig. 6 .
Fig. 6.CALIOP measurement (height versus latitude/longitude display) of the attenuated 532 nm particle backscatter coefficient during an overpass 45 km to the east of Limassol on 1 June 2013, 23:47 UTC, to 2 June 2013, 00:01 UTC.Desert dust layers are given in green to yellow colors and reach up to 4-10 km height.The inserted height-time display shows the CUT lidar observation of the range-corrected cross-polarized 532 nm backscatter lidar signal on 1 June 2013, 21:33-23:53 UTC (height range from 250-10000 m a.s.l.).Dust (green, yellow, and light blue layers) is observed up to 9-10 km height.The vertical black line indicates the closest position of CALIOP (laser foot print) to the ground-based CUT lidar at Limassol, Cyprus.

Fig. 9 .Fig. 10 .
Fig. 9. CALIOP data analysis products based on the 135 signal profile averages shown in Fig.8: (Left) Vertical profiles of 532 nm particle backscatter coefficient (green) and particle linear depolarization ratio (black), and (right) derived particle extinction coefficients for non-dust (blue) and desert dust particles (red).The particle backscatter coefficients (left, green) are taken from the CALIOP data base and smoothed with 600 m gliding window length.The particle linear depolarization ratio is computed from the crosspolarized and total backscatter coefficient profiles after smoothing the profiles with 600 m vertical window length.Lidar ratios used in the backscatter-to-extinction conversion are 50 sr for the non-dust particles and 55 sr for Saharan dust.

Fig. 11 .
Fig. 11.CALIOP measurement (height versus latitude/longitude display) of the attenuated 532 nm particle backscatter coefficient during an overpass 180 km to the east of Limassol on 1 November 2013, 23:43-23:57 UTC.Desert dust layers are given in green to yellow colors and reach up to about 4 km height.The inserted height-time display shows the CUT lidar observation of the cross-polarized range-corrected 532 nm backscatter signal up to 5 km height on 1 November 2013, 22:30 to 2 November 2013, 02:04 UTC.The vertical black line indicates the closest position of CALIOP (laser foot print) to the ground-based CUT lidar at Limassol, Cyprus.

Fig. 13 .
Fig. 13.Same as Fig. 10, except for a Middle East desert dust outbreak on 1-2 November 2013.In the case of CALIOP profiles (180 km east of Limassol), again 135 signal profiles (45 km horizontal resolution) are averaged, collected during seconds 8-14 of 23:47 UTC on 1 November 2013.The CUT lidar profiles show average values for the time period from 00:00-00:59 UTC on 2 November 2013.The original GDAS and GMAO temperature and pressure profiles were used in the computation of the INC profiles indicated by T surface = 23 • C. The GDAS and GMAO temperature profiles were shifted by 22 K (Limassol) and 23 K (CALIOP) in the case of the INC curves indicated by T surface = 0 • C. Uncertainties are of the order of 20 % (for σ d ), 30 % (for APC280), and within a factor of 3 (for INC) when using the dust INC parameterization after DeMott et al. (2015).