The influence of volcanic ash on heterogeneous ice nucleation in tropospheric clouds is investigated on the basis of 90 observed cloud cases. The clouds were observed with polarization lidars at the two central-European EARLINET stations Leipzig (51.3°N, 12.4°E) and Maisach (48.2°N, 11.3°E, 25 km northwest of Munich), Germany, in volcanic aerosol layers which originated from the strong eruptions of the Icelandic Eyjafjallajökull volcano in April 2010. Case studies of evolving boundary layer cumuli and long-lasting free tropospheric cloud events with unusual behavior (mixed-phase cloud complex, cirrus deck) are discussed. A clear impact of ash is observed. The ice nuclei concentration derived from the lidar observations has been estimated to range from 2–20 per liter in the boundary layer and from 100–300 per liter at cirrus level. The statistical analysis based on the 90 evaluated cloud cases revealed that all observed cloud layers with cloud top temperatures of below −15°C contained ice. Typically (under non-volcanic aerosol conditions) such a high fraction of ice-containing clouds is not reached before temperatures decrease below −25°C over central Europe.
 Dense volcanic ash clouds were advected towards central Europe after the strong eruptions of the Eyjafjallajökull volcano in southern Iceland in April 2010 [Ansmann et al., 2010, 2011; Flentje et al., 2010; Schumann et al., 2011]. This volcanic episode provided an unprecedented opportunity to investigate the impact of volcanic aerosols on meteorological processes. The European Aerosol Research Lidar Network (EARLINET) continuously monitored the dispersion of the ash plumes over Europe. The lidars detected the ash layers mostly above the boundary layer up to 5–6 km height. However, volcanic aerosol was also found in the boundary layer as well as in the upper troposphere. The volcanic plumes caused an almost complete shutdown of the European airspace for more than one week. As a consequence, contrails were absent during this period. In the ash layers, liquid-water clouds developed and partly glaciated. Cirrus clouds in the contrail-free upper troposphere formed and showed unusual behavior.
 The importance of ash-related cloud research arises from the fact that volcanic eruptions and thus the emission of ash particles occur almost every day somewhere around the globe. Volcanoes are permanent sources of IN. Investigations of the impact of volcanic aerosols on cloud and precipitation processes are thus of high relevance for weather and climate research. However, in contrast to ground-based, airborne, and laboratory IN studies, field observations of the evolution of the ice phase in mid-tropospheric clouds in volcanic ash layers have not been reported in the literature, but are essential to improve our knowledge of the role of fresh and aged volcanic ash plumes on the glaciation of clouds in the context of given meteorological conditions.
 The process of heterogeneous ice nucleation as a whole is not well understood because of the complexity of involved mechanisms and atmospheric influences. It thus deserves more research [Cantrell and Heymsfield, 2005; Seifert et al., 2010; DeMott et al., 2011]. At least four ice nucleation mechanisms (contact freezing, condensation freezing, immersion freezing, deposition freezing) can be distinguished [Pruppacher and Klett, 1997]. Heterogeneous ice nucleation depends in a complex way on meteorological conditions (ambient air temperature, relative humidity, occurrence of updrafts and downdrafts, turbulent mixing and entrainment of dry air into the cloudy environment), cloud properties (drop size distribution, fraction of large drops), and aerosol characteristics (aerosol type, mixture, aging and coating, size distribution, morphological properties). Field observations of cloud evolution (e.g., with focus on the timing of the development of liquid water, mixed-phase, and ice layers) as presented here are required to guide the development of more realistic parameterizations of this important cloud ice formation process in atmospheric models [Morrison et al., 2005; Fridlind et al., 2007; de Boer et al., 2011].
 The paper is organized as follows: In section 2, we briefly outline the instruments and techniques applied, and discuss the relationship of measured volume extinction coefficient, aerosol particle concentrations for particles with radii >250 nm, and ice nuclei concentrations and provide a short literature overview with focus on volcanic ash. In section 3, after an overview of the entire set of observed cloud layers embedded in the volcanic aerosol layers, case studies are presented. We start with two cases of ice formation in boundary layer clouds, followed by two cases with unusual cloud evolution in a mid-level mixed-phase cloud complex and an upper tropospheric subvisible cirrus deck. The statistical results based on the observations of 90 cloud layers are discussed in section 4. A summary is given in section 5.
2. Instrumentation and Methodology
2.1. EARLINET Lidars
 The EARLINET aerosol Raman/polarization lidars used in this study are operated at Leipzig (51.3°N, 12.4°E, 125 m height asl) [Mattis et al., 2004; Seifert et al., 2010], and Maisach (48.2°N, 11.3°E, 515 m height asl, 25 km northwest of Munich), Germany [Freudenthaler et al., 2009; Groß et al., 2011]. For cloud studies it is important to mention that the Leipzig lidar is pointing to the zenith, the laser beam is expanded (15 times), and the receiver field-of-view (RFOV) is 0.4 mrad. The Maisach lidar is pointing off-zenith, the laser beam is not expanded, and the RFOV is 2 mrad.
 The lidars provide height profiles of particle backscatter and extinction coefficients (by using the Raman lidar method [Ansmann et al., 1992]), and volume and particle depolarization ratios (linear depolarization ratio) at the wavelength of 532 nm. During daytime, the elastic backscatter lidar method [Fernald, 1984] is predominantly used to determine particle backscatter coefficients.
 The particle depolarization ratio allows us to discriminate layers dominated by backscattering of spheres (e.g., liquid water droplets) and of non-spherical particles such as volcanic ash particles and ice crystals [Sassen, 2005]. Spherical particles cause a rather low particle depolarization ratio (close to zero, in the absence of multiple scattering), whereas the particle depolarization ratio is in the range of 0.35–0.37 for volcanic ash [Ansmann et al., 2010; Groß et al., 2010] and 0.3–0.7 for ice crystals [Sassen and Benson, 2001]. Because of the laser beam pointing to the zenith, the Leipzig lidar observations of depolarization ratio and thus of the cloud phase may be affected by specular reflection of horizontally aligned ice crystals [Seifert et al., 2008], whereas multiple scattering may affect the depolarization observations in the case of the Maisach lidar [Bissonnette, 2005] as a consequence of the large RFOV. These aspects and the respective consequences in the interpretation of the findings are discussed in section 3.
2.2. Leipzig AERONET Photometer
 At Leipzig, the EARLINET lidar and a Sun photometer of the Aerosol Robotic Network (AERONET) [Holben et al., 1998] are run side by side. The AERONET photometer measures the aerosol optical thickness (AOT) from 340–1640 nm in eight channels. The observations were extensively used to describe the optical and microphysical properties of the volcanic plumes in 2010 [Ansmann et al., 2011].
 A detailed description of the cloud analysis scheme is given by Seifert et al. . Cloud top temperatures are required because first heterogeneous nucleation of ice crystals typically begins at the coldest part of a liquid-water cloud (cloud top) [Rauber and Tokay, 1991; Harrington et al., 1999; Lebo et al., 2008]. After nucleation at cloud top, ice crystals grow fast and begin to fall through the cloud layer and finally form fall streaks, so-called virgae, below the cloud layer. By using the polarization lidar technique these virgae allow us to identify the clouds as mixed-phase, or more generally, as ice-containing cloud [Ansmann et al., 2008, 2009; Seifert et al., 2010]. The meteorological data (profiles of temperature and relative humidity) for Leipzig were taken from the GDAS1 data archive of the U.S. National Weather Service's National Center of Environmental Prediction (NCEP). This data archive contains the assimilated observational data for the initialization of weather forecast models. It is based on the global data assimilation system GDAS (Global Data Analysis System, http://ready.arl.noaa.gov/gdas1.php) [Kanamitsu, 1989] that stores the assimilated data fields, including ground-based observations as well as radiosonde and satellite-based data. The nearest grid points of the GDAS1 data archive to the lidar stations are 25 km south of Leipzig and 30 km southwest of Maisach. GDAS1 data is available every 3 hours. The uncertainty in the cloud top temperatures (deviation of the actual temperature from the GDAS1 temperature) is on the order of 1 K [Seifert et al., 2010]. The model-derived humidity profiles (also shown in the case studies in the next section) have to be treated with care. Atmospheric models are unable to simulate detailed and accurate humidity conditions in shallow heterogeneous cloud fields. For Maisach, atmospheric soundings launched by the German Meteorological Service (DWD) at Munich-Oberschleißheim, 25 km east of Maisach, are available twice a day (00, and 12 UTC). Motivated by the need for accurate information about the atmospheric state during the episode of the volcanic eruption additional sondes were launched by DWD at 06 and 18 UTC between 19 and 24 April.
2.4. Regional Transport Model COSMO-MUSCAT
 The weather prediction model COSMO (Consortium for Small-scale Modelling) is a regional-scale weather forecast model of the German Meteorological Service. The model version COSMO-MUSCAT (Multi-scale atmospheric transport modelling) was developed and used for simulations of the spatio-temporal distribution and radiative effects of Saharan dust [Heinold et al., 2007, 2009]. In this study the model is used to estimate the ash mass concentration and ash particle number (APC) concentration throughout the troposphere.
 The model setup that was applied to perform the transport modeling of the volcanic ash is explained in detail by Heinold et al.  and is here only briefly described. The simulations are performed with 28 km horizontal grid spacing and 40 vertical layers. The advection time step is 45 s. Volcanic plume input parameters are the plume height from the London Volcanic Ash Advisory Centre (VAAC, http://www.metoffice.gov.uk/aviation/vaac/vaacuk_vag.html) reports, fractional vertically resolved emission rates derived from Multi-angle Imaging Spectroradiometer (MISR) plume height observations (http://www-misr.jpl.nasa.gov/getData/accessData/MisrMinxPlumes/), and emission strengths provided from the unified EMEP (European Monitoring and Evaluation Programme) model conducted by the Norwegian Meteorological Institute (https://wiki.met.no/emep/emep_volcano_plume). In the simulations only the transport of primary ash is considered. Secondary formation of sulfate particles within aged volcanic ash plumes is not taken into account. The calculations are performed for five particle radius classes, determining transport and sedimentation characteristics of the ash (100–300 nm, 300–900 nm, 900 nm–2.6 μm, 2.6–8.4 μm, and 8.4–24 μm). Uncertainties in the simulated ash profiles result from uncertainties in the volcanic plume height, the vertical distribution of emitted ash and gases, timing of the emissions, and meteorological wind fields (see Heinold et al.  for more details).
2.5. Relationship Between Aerosol Extinction, Large Particle Fraction, and Ice Nuclei Concentration
 COSMO-MUSCAT allows us to compute profiles of the ash particle number concentration (APC) considering large particles with radii >300 nm only. Studies of DeMott et al. [2006, 2009], Richardson et al. , and Hoose et al.  show that the IN concentration (INC) is correlated with the concentration of larger aerosol particles so that INC can be estimated from the modeled APC values.
Ansmann et al.  also discussed the potential to estimate APC from measurements of the particle extinction coefficient (EXT) by means of combined lidar (extinction profiling) and Sun photometer observations (volume size distribution retrieval) in the case of Saharan dust so that INC can finally be estimated from the measured extinction coefficients by using the APC/EXT relationship and literature values of the APC/INC ratio. Ansmann et al.  derived an APC/EXT ratio of 0.5–1 cm−3/Mm−1 for particles with radii >250 nm (see Figure 3 of Ansmann et al. ). The studies of Ansmann et al. [2010, 2011], and Schumann et al.  indicate that the ash and dust size distributions were similar.
 However, the APC/EXT relationship is restricted to environments dominated by coarse mode particles, i.e., in the absence of natural or anthropogenic fine mode particles. Fine-mode particles strongly influence APC in the radius range from 250 or 300 nm to 1 μm, but have a comparably moderate influence on the volume extinction coefficient so that a clear relationship between the extinction coefficient and the APC cannot be given. A lidar/photometer study similar to the one of Ansmann et al.  performed for Saharan dust was applied to the Leipzig observations from 16–25 April 2010. From this analysis we conclude that the same APC/EXT relationship found for desert dust holds for volcanic particles only during periods in which volcanic ash (coarse mode) dominated the aerosol optical depth. However, during most situations the observations at Leipzig were hampered by the strong impact of freshly formed volcanic sulfate particles and anthropogenic pollution on the optical measurements [Ansmann et al., 2011].
 The suitability to estimate APC from the combined lidar/photometer method under volcanic aerosol conditions was checked for the 17 April 2011. On this day the APC levels in the main ash layer, reaching from the ground up to 2 km height, were of the order of 10–20 cm−3 according to the Leipzig lidar observations of extinction coefficients around 20 Mm−1. For the same day, Bingemer et al.  reported APC and INC of volcanic aerosol that was observed in situ at the Taunus Observatory mountain research station (825 m asl) located close to Frankfurt, Germany, about 300 km west of Leipzig. AERONET photometer observations at Mainz (25 km southwest of the Taunus Observatory) and Leipzig indicated similar ash conditions with coarse-mode optical depths ranging from 0.04 to 0.05 at both stations on 17 April 2010. H. Bingemer (personal communication, 2011) measured APC levels ranging from 5–15 cm−3 on 17 April 2010. Assuming that the depth of the main ash layer at Taunus Observatory was similar to the one above Leipzig, the APC derived for both stations are similar and a APC/EXT conversion factor of 0.5–1 cm−3/Mm−1 should be suitable. In order to derive the INC from the lidar/photometer observations, the APC/INC ratio must be selected carefully. The APC/INC ratio can vary strongly with temperature and with chemical composition and size distribution of the aerosol as studies of DeMott et al. [2006, 2009], Richardson et al. , and Hoose et al.  show. For Saharan dust, APC/INC ranges from 100–500 for temperatures around −20°C. The ratio may increase by a factor of 100 for temperatures as high as −10°C [Hoose et al., 2010].
 From recent studies of Belosi et al.  and Bingemer et al. , we conclude that the APC/INC ratio for volcanic ash is lower than for Saharan dust, i.e., volcanic ash particles are the more efficient ice nuclei. Belosi et al.  report in situ observations of the ratio of APC (particle radius >150 nm) to INC (measured at −17°C and ice supersaturation of 10%–20%) at the urban background station of Bologna in northern Italy and found APC/INC values of about 100 when a strong Eyjafjallajökull ash plume reached the field site on 20 April 2010. By keeping in mind that urban particles strongly contributed to the measurement of the APC, when considering particles with radii >150 nm, but the efficiency of urban particles to serve as ice nuclei may be likewise low (at least much lower than the one for dust or ash), the APC/INC ratio is certainly lower (and thus probably in the range of 50 or even lower) in an environment of almost pure volcanic ash and when particles with radii >250 to 300 nm are considered only in the computation of APC.
Bingemer et al.  reported surface observations of INC values of typically 50–100 per liter after the Eyjafjallajökull event and peak values exceeding 600 IN per liter (measured at a temperature of −18°) on 17–18 April 2010. The APC/INC ratio was in the range of 200–400 and was as low as 50 during IN peak periods. These values are similar to the ones inferred from the AERONET photometer observations and indicate APC/INC ratios of the order of 60–150 with peak values of around 20. Bingemer et al.  further state that the ice nucleation efficiency of ash was found to be almost a factor of 2 larger than during a strong Saharan dust event.
 The lowest limit for the APC/INC ratio may be given by laboratory studies of Isono et al. , who found for an APC of 5–10 cm−3 an APC/INC ratio of 10000 yielding 0.5–1 IN per liter at temperatures of −12°C. Fornea et al.  mentioned that even relatively small populations of IN on the order of 1 liter−1 can have a substantial impact on the overall development of an ice cloud.
 Taking into account the strong variation in the derived APC/INC ratios between the different operating temperatures of the INC measurement systems, we assume in the following discussions of the lidar observations a volcanic-ash-related APC/INC ratio of 1000, 100, 25 for the temperatures around −10, −20, and <−30°C. Figure 1 illustrates the assumed relationship between the extinction coefficient EXT and INC at these three temperatures by assuming an APC/EXT ratio of 0.5–1.0 cm−3/Mm−1.
Figure 2 gives an overview of the cloud observations with the two lidars from 16 April to 25 April 2010. COSMO-MUSCAT simulations of ash mass concentration for the two measurement sites of Leipzig and Maisach are shown in addition to provide an impression to what extent the atmosphere and especially cloud formation was disturbed by the presence of volcanic ash. During the full observational period 67 cloud layers in the lower troposphere (cloud top height below 3 km above ground level, agl), 43 mid-level cloud systems with top heights from 3 to 8 km, and 37 cirrus layers were analyzed. The majority of the cloud fields developed in an environment that was predicted by COSMO-MUSCAT to be ash rich. From the total of 147 observed cloud layers, 90 layers had cloud top temperatures in the range from −40 to 0°C. At these temperatures heterogeneous ice nucleation alone is responsible for ice formation and cloud glaciation. The freezing behavior of this subset of clouds is statistically evaluated in section 4.
Figure 3 shows two measurement examples that confirm that traces of ash typically reached the tropopause as predicted by COSMO-MUSCAT. Cirrus clouds as well as boundary layer clouds (at heights below 2 km) formed in ash layers on 20 and 22 April 2010. On 20 April the formation of cirrus began at 8 km height, about 2 km below the tropopause (Figures 3a–3c). This is an unusual behavior and not in accordance with our long-term cloud observations at Leipzig [Seifert et al., 2010]. Also Sassen and Campbell  report in a 10-year climatology of midlatitude cirrus a mean distance between cirrus cloud top and the tropopause of less than 1 km in winter and spring.
 The occurrence of cirrus significantly below the tropopause is probably related to the fact that rather high relative humidities (related to ice) as usually required for homogeneous freezing are not needed when a large number of IN is present. The tropopause region is the coldest part of the troposphere and shows the highest levels of relative humidity in case of a moist upper troposphere and thus usually provides the most favorable conditions for the generation of first cirrus cells. At volcanically disturbed conditions, however, ice formation can obviously take place at any height with low enough temperatures whenever the ice saturation level is slightly exceeded. This point is further discussed in section 3.4.
 On 22 April 2010 contrail cirrus formed below 9 km height (−49°C, Figures 3d–3f) after the resumption of air traffic which had been shut down from 16 to 21 April 2010. Aircraft which took off upwind of Leipzig at the airport of Frankfurt am Main, Germany (50.0°N, 8.5°E), caused during their ascent (before reaching the final flight level at 10–10.5 km height) the contrail-induced cirrus structures with the development of strong virgae immediately after forming the main contrail ice cloud.
 The profiles of the 532 nm particle backscatter coefficient in Figures 3c and 3f show the cloud layers and several volcanic particle layers (ash plus sulfate) up to heights of 10 km. The tropopause was above 10 km and thus more than 2 km above the cirrus layer around 1300 UTC on 20 April 2010. The ice clouds on 22 April 2010, around 1500 UTC were found 1 km below the tropopause.
 The backscatter coefficients measured in the volcanic aerosol layers above the PBL (red areas below 2–3 km height) indicated volume extinction coefficients of 100–200 Mm−1 in the PBL on both days, 20–50 (partly) 80 Mm−1 in the pronounced lofted layers just above the PBL, of 1–10 Mm−1 in the free troposphere above the lofted layers up to about 6 km, and 0.1–1 Mm−1 at cirrus level. In the backscatter-to-extinction conversion a lidar ratio (extinction-to-backscatter ratio) of 50 sr is assumed as given by Ansmann et al. . By using the sulfate-ash discrimination technique described by Ansmann et al. , the ash-related extinction coefficients in the PBL ranged from 5–20 Mm−1 on both days shown in Figure 3, and the free-tropospheric ash extinction coefficients were about 10–50 Mm−1 in the pronounced volcanic layers between 2–3 km height, 1–5 Mm−1 up to 6 km height, and 0.1–1 Mm−1 above.
 According to the discussion in section 2.5 the INC may be inferred from the measured extinction as illustrated in Figure 1. The extinction coefficients correspond to APC values (for ash particles with radii >250 nm) of 2–20 cm−3 in the PBL, 5–50 cm−3 in the lofted, pronounced ash layers between 2–3 km height, 0.5–5 cm−3 from 3–6 km, and 0.05–1 cm−3 at cirrus level in the upper troposphere. These APC values, in turn, may roughly indicate IN concentrations of the order of 2–20 IN per liter (APC/INC = 1000 for temperatures around −10°C) at the top of the PBL (see discussion in the next section on ice formation in cumulus clouds), 10–50 IN per liter from 2–3 km height (same APC/INC ratio of 1000 is assumed), 10–100 IN per liter in the volcanic layers up to 6 km height (APC/INC = 100 for temperatures around −20°C), and 2–40 IN per liter (APC/INC = 25 for temperatures around <−30°C) above 6 km. At cirrus level with temperatures below −40°C, the APC/INC for volcanic ash may be about 10 or even lower, so that an INC of >100 liter−1 may have occurred in the upper troposphere where cirrus layers developed.
3.2. Boundary Layer Clouds
 A convectively active planetary boundary layer (PBL) up to 1.6 km was observed on 20 April 2010 until 1300 UTC (Figure 4) at Leipzig. The PBL contained aged ash and sulfate particles of volcanic origin [Ansmann et al., 2011]. Above the PBL another volcanic aerosol layer is visible before 1300 UTC between 2 and 3 km height. The volume depolarization ratio was low in the PBL (Figure 4d) and indicates that non-depolarizing spherical sulfate particles dominate backscattering of laser light. Obviously as a result of a sudden air mass change between 1300 and 1400 UTC, the PBL depth increased from 1.6 to 3.2 km height. The isolated volcanic layer around 2.6 km is no longer visible. Enhanced volume depolarization ratios (light blue) in the PBL indicate the presence of volcanic ash particles even at low heights. Cumulus clouds developed after 1420 UTC in the upper part of the moist boundary layer. Cloud top temperature was around −12°C. In the optically densest clouds ice crystals formed and fell out of the cloud layer around 1530 UTC which is indicated by the sharp red spots in the volume depolarization plot in Figure 4 (1520–1600 UTC, 1700 UTC).
 In Figure 4c, a profile of the ash particle number concentration (APC) with diameters larger than 600 nm simulated with COSMO-MUSCAT is presented. Particles with diameters larger than 600 nm (corresponding to size bins 2 to 5 in the model) may best represent the reservoir of potential IN.
 According to the discussion in the foregoing section, the backscatter coefficients in Figure 3c (red curve for the period from 1445–1503 UTC), indicating total extinction coefficients of 100–200 Mm−1, of which 5–20 Mm−1 is related to volcanic ash, point to APC values of the order of 2–20 cm−3 (in good agreement with the COSMO-MUSCAT simulations in Figure 4c) and 2–20 IN per liter at cloud level.
 According to Fornea et al.  contact freezing is most likely responsible for heterogeneous ice formation in this case of ice nucleation in an ash-laden PBL cumulus cloud layer. Fornea et al.  found a five-degree spread of freezing temperatures (−8.3°C to −12.8°C) in the case of contact freezing by ash particles with an average freezing temperature of −11.0°C. For immersion freezing, the mean freezing temperature was −18.3°C (range from −14.5°C to −22.2°C). These findings are in general agreement with earlier studies by Mason and Maybank  and Isono and Ikebe , who observed freezing in the range from −10 to −18°C and −13 to −20°C on various volcanic ejecta.
 A case of cloud glaciation at the top of the residual layer was observed at Maisach in the early morning of 20 April 2010 (Figure 5). The radiosonde profile of the relative humidity indicates a well-mixed residual layer up to 2.6 km height and that the water saturation level was almost reached at the top of the moist layer. Many wave-like structures are visible in the vertically stable residual layer. The up- and downward motions may have contributed to the formation and dissolution of the cumuli fields. Between 0400 and 0540 UTC clouds formed between 2.2 km height and the top of the residual layer, indicating that water saturation was occasionally reached. Until 0520 UTC the clouds consisted of supercooled liquid water droplets indicated by strong light attenuation (optical depth >2 caused the dark blue columns above the clouds in Figure 5b) as well as by the absence of ice virgae. Because of the large receiver field of view of the Maisach lidar (2 mrad, compared to 0.4 mrad of the Leipzig lidar) multiple scattering sensitively influences water cloud measurements (with number concentration of scatterers of the order of 100 drops per cm3) so that they appear mostly red in the depolarization plot. After 0520 UTC virgae falling out of the liquid-water clouds unambiguously indicate ice crystals. Here, the number density of scatterers is of the order of 0.01–0.1 crystals per cm3 so that multiple scattering effects can not mask the discrimination of drops and ice crystals.
 From the backscatter coefficients in Figure 5, we estimated the volume extinction coefficients of all (sulfate plus ash) particles by using a lidar ratio of 50 sr. The ash-related extinction coefficients were then about 30–60 Mm−1 within the residual layer and 10–15 Mm−1 in the lofted layer around 3 km height (at 0400 UTC). As mentioned above, the separation of ash and sulfate extinction is based on the polarization-lidar-based discrimination technique [Ansmann et al., 2011]. For these ash extinction coefficients, APC is in the range of 15–60 cm−3 (residual layer) and 5–15 cm−3 (lofted layer above), respectively, and INC estimates range from 15–60 IN per liter and 5–15 IN per liter, respectively. The COSMO-MUSCAT model indicates APC values of 5–22 cm−3 in Figure 5c.
 According to Fornea et al. , again contact freezing is the only possible freezing mode to take place at −9°C and almost instantaneously converted the liquid water droplets into ice particles. Evaporation of drops in downdraft zones support the growth of existing ice crystals and, as a consequence the formation of extended virgae.
3.3. Midlevel Cloud Layer
 Over the Munich area, extended cirrus fields developed in the upper troposphere below 10 km height throughout the whole day until about 1800 UTC on 22 April 2010. Figure 6 shows the lidar measurement period from 1030–1830 UTC. During this time the height range above 7 km was cloud-free. According to the AERONET photometer observation at Munich University (48.1°N, 11.6°E) and the lidar observations at Maisach, the cloud optical depth ranged from 0.1 to 0.15 at 500 nm for the time period shown in the figure. The height-time display of the depolarization ratio indicates mostly cloud layers containing ice crystals. In the late afternoon (after 1530 UTC), however, layers with rather low depolarization ratio dominate. Apparently the phase of the cloud changed from a pure ice cloud to a mixed-phase cloud. This unusual aspect is discussed in more detail below.
 The particle backscatter coefficient profile in Figure 6d, measured from 1200–1220 UTC, indicates volume extinction coefficients (backscatter coefficients multiplied by 50 sr) of 150 Mm−1 in the PBL. The separation of ash and sulfate contributions yield PBL ash extinction coefficients from 10–20 Mm−1. The ash-related extinction coefficients were also 10–20 Mm−1 in the pronounced volcanic layer between 2 and 3 km height, and mostly 1–5 Mm−1 outside this layer. Above 3 km height, the extinction values were close to zero with most values from 0.1–0.5 Mm−1. However, traces of volcanic particles are visible in Figure 6b above 3 km as coherent structures, even at heights of 5.5 km after 1700 UTC.
 The ash extinction coefficients point to APC values (ash particles with diameters larger than 500 nm) of 5–20 cm−3 in the PBL. This estimation agrees well with the COSMO-MUSCAT simulation in Figure 6d. COSMO-MUSCAT overestimates the APC in the free troposphere. According to the lidar retrieval, APC was on the order of 5–20 cm−3 in the pronounced volcanic layer, and 0.5–5 cm−3 outside this layer. Above 3 km height, APC values from 0.05–0.5 cm−3 may characterize the ash conditions in the middle troposphere including the cloud environment from 4–7 km height on that day. Such low APC values roughly indicate INC values of 0.5–20 liter−1 when assuming ash APC/INC ratios ranging from 25–100 for temperatures from −20 to −30°C at which most cloud cells formed.
 As shown in Figure 6b, cloud top temperature increased from −34°C (7 km) between 1230 and 1330 UTC to values around −26°C (6 km height) after 1630 UTC. At temperatures below about −30°C ideal conditions for deposition freezing [Pruppacher and Klett, 1997] are given, i.e., direct deposition of water vapor on the ash IN when ice saturation is slightly exceeded. However such ideal conditions are seldom as our 11 year cloud study [Seifert et al., 2010] and the cloud observations over Cape Verde showed [Ansmann et al., 2009]. With few exceptions, the liquid phase forms first before ice crystals are nucleated via heterogeneous freezing.
 The relative humidity profile measured with the 1200 UTC radiosonde (see Figure 6a) indicates that ice saturation occurred from 5.5–7 km before new layers with ice crystals appear in the lidar beam around 1300 UTC. As mentioned above, these cirrus layers were optically thin with optical depth mostly around 0.1. Theoretically, condensation freezing could have occurred in addition (not distinguishable from deposition freezing effects with lidar), but this freezing mode needs water saturation which was most probably not given, at least around 1300 UTC.
 Later on (after 1530 UTC), layers with dominating drop backscattering appeared (blue areas in Figure 6e), obviously at times when the cloud top temperature was above −26°C. Droplet formation requires water saturation conditions. Unfortunately, the 1800 UTC sonde was launched after the dissolution of the cloud deck, but still indicates a very moist middle troposphere (even higher when taking a possible dry bias effect of the order of 10% into account [Cady-Pereira et al., 2008; Agustí-Panareda et al., 2009]).
 What may have caused such a seldom observable, sudden transformation of an ice cloud into a mixed-phase cloud? This is even more surprising when taking into account that at −20°C to −26°C, rather favorable conditions for immersion and contact freezing are given in the presence of volcanic ash particles. Several hypothesis are consistent with the observation. Among these, the most convincing argument is that (1) temperatures became too high for efficient deposition freezing, and (2) relative humidity increased (by advection) in addition, and when water saturation was exceeded liquid drops formed, and this predominantly on the volcanic sulfate particles. The presence of large sulfate particles (after water uptake at high relative humidity) acting as condensation nuclei for supercooled droplets can cause a freezing-point depression (homogeneous freezing) as was also observed in cirrus clouds after the eruption of the Pinatubo volcano [Sassen, 1992]. Cloud drops which formed on the volcanic sulfate particles are also not available for immersion freezing.
 Contact freezing may have been inefficient, too, because of the low number of available dry ash particles. In addition, drops may have been rather small and drop concentration very low (consistent with the low optical depth of 0.1) so that the number of collisions between drops and remaining dry ash particles was rather low.
 It is also possible that the depolarization ratio dropped because the ice crystals became spherical. However as discussed by Lawson et al. , observations show that even frozen droplets have a slight but noticeable deviation from a perfect spherical shape so that the depolarization ratio of ice particles will always be significantly higher than the one for pure liquid drops. On the other hand, it remains an open question how ice crystals suddenly change their morphological properties and this over a large height range simultaneously.
 There may be many other theories including the development of optically thick haze layers (water uptake by sulfate particles) or an air mass change (advection of a new moist air mass associated with a new volcanic plume with a changed sulfate/ash particle ratio) occurred. However, further potential processes will not be discussed here because additional information about meteorological and cloud microphysical properties that would help to limit the number of possible processes are not available.
 In conclusion, the observation at least shows that complex ice-containing cloud systems can develop in the presence of volcanic aerosol mixtures of sulfate and ash particles. Regarding ice nucleation, deposition freezing comes into play besides immersion and contact freezing. It remains an open question, whether the observed long-lasting cloud deck would have evolved (at least as long as the water saturation level was not reached) without the presence of volcanic aerosols.
 As mentioned in section 3.1, most of the cirrus layers which developed within the week after the approach of first volcanic layers on 16 April 2010 appeared at heights significantly below the tropopause, and obviously formed on the ash particles via deposition freezing at ice supersaturation levels of 10% or lower. In this section we describe a cirrus event that has never been observed with lidar over Leipzig during the last 11 years. For more then ten hours a coherent subvisible cirrus deck developed (optical depth <0.03).
 We begin with the explanation of the basic lidar and photometer observations performed on Sunday, 18 April 2010. Figure 7 shows this long lasting cirrus event that occurred at temperatures from −41 to −54°C. During the first hour, the cloud was completely invisible to the naked eye. The cirrus optical depth was 0.006 according to the lidar and Sun photometer (1640 nm channel) observations. There was no sharp and large increase in the signal strength at cirrus height level as is usually the case. Pronounced virgae did not develop immediately after the generation of the first crystal nucleation cells. The optical depth remained below 0.03 during the entire day. At the manned meteorological stations in the area of Leipzig (WMO station codes 10469 and 10471) no cloud observations were reported during the whole day. All this is very unusual for a midlatitude cirrus, and it is the first time (within 15 years of cirrus observations at Leipzig) that a cirrus detected by lidar, and even when classified as subvisible (optical depth below 0.03), remained almost invisible over the whole time period of occurrence (over several hours). Instead of pronounced virgae, many laminar coherent structures lasting several hours were visible in the lidar observations. Such structures are usually seen in aerosol layers only.
 The discrimination of cirrus particles from ash particles cannot be based on the measured depolarization ratio because both, ash and ice particles cause strong depolarization of laser radiation. However, a first indication for the occurrence of cirrus formation are the inhomogeneous backscatter structures. Significant changes occur within minutes and several tens of height meters as typical for cirrus. In contrast, all volcanic ash layers observed after 16 April 2008 (first strong volcanic ash front [Ansmann et al., 2010]) showed a smoother behavior of the backscatter characteristics in time and space. The second and more unambiguous indication of the presence of an ice cloud is the extinction-to-backscatter ratio. This lidar ratio was estimated from the cirrus optical depth derived from the AERONET sun photometer observation at 1640 nm (see discussion on this retrieval below), and the cirrus-layer-integrated backscatter coefficient obtained from the lidar observations. This approach revealed typical cirrus extinction-to-backscatter ratios of 10–25 sr [Ansmann et al., 1992]. In contrast the lidar ratio of volcanic ash is about 40–60 sr [Ansmann et al., 2010, 2011].
 Time series of AERONET 1640 nm AOT were used to describe the evolution of the cirrus deck in terms of particle optical depth. As mentioned in section 2.2, the 1640 nm channel is rather sensitive to the presence of large particles (cirrus, ash), but almost insensitive to changes in the boundary layer optical depth caused by an increasing or decreasing impact of fine-mode urban haze. By assuming that the 1640 nm AOT values at 1000–1100 UTC (clear skies) represent the ash-related AOT and that the volcanic AOT was constant over the day, the increasing 1640 nm AOT was interpreted as cirrus AOT. This increase indicated a cirrus-related AOT of 0.006 before 1300 UTC, 0.02–0.03 from 1300–1600 UTC, and of 0.01 from 1600 UTC up to the end of the AERONET observations (around 1715 UTC). The lidar observations of the cirrus backscatter coefficients are in full agreement with the photometer-derived AOT values throughout the day when assuming a typical cirrus lidar ratio of around 20 sr at 1064 nm [Vaughan et al., 2010].
 Because the geometrical depth of the ice cloud was 1 km during the first hours and roughly 1–2 km afterwards, the mean cirrus extinction coefficient was rather low with values of 5–10 Mm−1 in the beginning and 10–15 Mm−1 later on, and thus just a factor of 2–3 larger than the ash-related extinction coefficients of about 5 Mm−1 below and above the cirrus. The ash backscatter coefficients shown in Figure 7c were computed from the 532–nm lidar signal and for an ash extinction-to-backscatter ratio of 50 sr [Ansmann et al., 2010, 2011]). For comparison, typical cirrus extinction coefficients are an order of magnitude larger with values of 100–300 Mm−1 [Seifert et al., 2007].
 The bluish colors in the depolarization plot also reflect the small ice crystal extinction coefficients. The volume depolarization ratio is influenced by particle and Rayleigh backscattering [Cairo et al., 1999]. Ice crystals cause a depolarization ratio of 0.3 to 0.6, whereas the Rayleigh depolarization ratio is less than 0.015. Rayleigh backscattering dominated the volume depolarization ratio in this optically thin cirrus deck. Because the Leipzig lidar is pointing to the zenith, specular reflection by some larger, horizontally well-aligned ice crystals cannot be excluded. This effect also leads to a decrease of the depolarization ratio [Seifert et al., 2010].
 We estimated the IN concentration from the photometer observation in the way described by Ansmann et al.  for Saharan dust in southern Morocco. The coarse-mode-related 500 nm particle optical depth was 0.03 at 1110 UTC (before the cirrus development started). The lidar observation indicated a volcanic ash layer from the PBL top to about 11 km with almost height-independent backscattering by ash particles. Thus the ash optical depth of 0.03 for the 10 km deep ash layer roughly indicates an ash extinction coefficient of 3 Mm−1. The slightly larger lidar-derived extinction coefficients on the order of 5 Mm−1 may result from the fact that not only ash but also freshly formed sulfate particles (fine-mode particles) contributed to total particle backscattering and extinction of laser light in the free troposphere [Ansmann et al., 2011]. The inversion of the photometer observations yield a column-integrated bimodal volume size distribution of the particles. If we simply assume that all particles per size class have the same radius ri,c (interval center radius) we can estimate the number concentration per size class (22 classes in total) by assuming that all ash particles are spheres with radius ri,c. In this way we found for 18 April 2010, 1100 UTC, that, on average, about 3–5 particles per cm3 (APC) with radii >250 nm were present. If we further keep in mind that a part of the particles with radii >250 nm are non-ash particles (anthropogenic pollution, volcanic sulfate particles) the ash-related extinction coefficients of roughly 3 Mm−1 indicate the particle concentration of coarse mode ash particles of about 3 cm−1 as outlined in section 2.5.
 If we further assume that 10% of these large particles acted as IN (APC/INC = 10 at temperatures around −50°C, see section 2.5), the IN concentration was about 300 per liter, a rather high number. This estimation explains why we found these unusual cirrus characteristics with almost no virgae formed by large crystals. A typical cirrus shows ice crystal number concentrations of 1–30 ice crystals per liter. If we now have 300 ice crystals per liter or even more (for the same ice water content) the crystals must have been small and unable to fall with significant terminal velocity in well organized fall strikes. Such conditions were already described in Mason's classical text [Mason, 1971], as commented by Durant et al. : Overseeding with massive doses of nuclei will result in large concentrations of crystals that will be unable to grow sufficiently large to fall out of the cloud layer (and reach the ground).
 The simulated values of APC = 0.4 cm−3 in Figure 7c are too low, and do not agree with the lidar measurements of the optical properties around the cirrus layer in the upper troposphere. The main reason is most probably that the plume height (input parameter in the simulations) estimated from spaceborne remote sensing is too low, because the spaceborne sensor is insensitive to traces of ash in the upper troposphere and does only detect the densest part of the emitted ash plumes.
 Heterogeneous ice formation via the deposition freezing mode is expected to occur at temperatures below −45°C as soon as the relative humidity reaches and exceeds the ice saturation level by a few percent. The hypothesis that water saturation was not reached on 18 April 2010 is supported by the GDAS1 relative humidity profiles in Figure 7a.
 We investigated the water vapor conditions in more detail in Figure 8. Radiosonde profiles (Vaisala RS 92 sondes) of Meiningen, 170 km southwest of Leipzig, and Lindenberg, 150 km northeast of Leipzig, are shown. Even at temperatures down to −50°C to −60°C the dry bias in the radiosonde (VAISALA RS 92) observations is small (5%–10% for humidities >50% [Miloshevich et al., 2006]). However, a solar heating effect of the humidity sensor box may cause another dry bias of 5%–10% [Cady-Pereira et al., 2008; Agustí-Panareda et al., 2009]. The synoptic analysis reveals that the 1200 UTC Meiningen profiles best describe the atmospheric humidity conditions at Leipzig (advection and lifting of humid air masses from southeast just before first cirrus clouds formed). The Meiningen profile corroborates the hypothesis that only ice saturation was reached. The humid layer obviously reached Lindenberg in the late afternoon. However, the relative humidity values remained clearly far below the water saturation level.
4. Statistical Results
Figure 9 shows the increase of the occurrence frequency of ice-containing clouds with decreasing temperature based on all cloud layers observed during the volcanic ash period in the second half of April 2010. For comparison the cloud statistics for clean conditions (also mostly occurring during northwesterly airflows) and situations with desert dust advection as shown by Seifert et al.  are added. Both curves for northwesterly flows (clean conditions, volcanic cases) are very similar for the temperature intervals down to −15°C. Error bars were calculated according to equation (1) from Seifert et al.  and express the statistical significance. A steep increase of the curve from almost 0% (−5 to −10°C interval) to 100% (−15 to −20°C interval) is found in the case of the volcanically influenced clouds. This reflects the major role of immersion freezing. This heterogeneous ice nucleation mode is obviously already fully active at temperatures of −15 to −16°C. In contrast to the volcanic aerosol curve, the desert dust curve is broader, ice nucleation starts at higher temperature, but the 100% value is not reached before −25°C (as is the case for clean conditions). The volcanic curve is in good agreement with the laboratory studies [Mason and Maybank, 1958; Isono and Ikebe, 1960; Fornea et al., 2009] in which first activation was found around −10°C (contact freezing) and from −17°C to −20°C (immersion freezing).
5. Summary and Conclusions
 The opportunity of the Eyjafjallajökull volcanic eruptions were used to study the impact of volcanic ash on heterogeneous ice formation. The investigation was performed in the same way as presented by Seifert et al.  for the influence of Saharan dust on ice nucleation. About 90 cloud cases embedded in volcanic aerosol were observed over Leipzig and Maisach (close to Munich) from 16 to 25 April 2010. Unusual case studies of cloud developments were discussed.
 The presented case studies clearly show the effectiveness of ash particles to serve as IN. The statistics were found to be in full agreement with laboratory studies. The analysis revealed that all observed cloud layers with cloud top temperatures of below −15°C contained ice. Typically (under non-volcanic aerosol conditions) such a high fraction of ice-containing clouds is not reached before temperatures decrease below −25°C over central Europe.
 It would be interesting to perform such a study in the tropics, where volcanic eruptions frequently occur. The development of deep cumulus convection, within a volcanically disturbed free troposphere, and the evolution of precipitation in the presence of ash may be rather different from situations without a volcanic impact.
 We thank the AERONET team headed by Brent Holben for their service (Sun photometer calibration, data analysis, quality assurance), and for the high quality measurements during the volcanic period. The financial support for EARLINET by the European Commission under grant RICA-025991 is gratefully acknowledged, too.