We analyze and compare the different sensitivities of aerosol/cloud lidar and 35-GHz cloud radar to detect ice formation in midlevel clouds in order to harmonize mixed phase cloud observations performed with lidar and radar. We found good agreement between spaceborne Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO)/CloudSat and ground-based lidar/radar observations at Leipzig, Germany. However, large differences were found to a previous study with an 11-year cloud statistics solely based on lidar observations which is caused by significantly higher sensitivity of the cloud radar to detect ice crystals. By introducing a lidar detection threshold for the ice water content of 10−6kgm−3, we find that lidar and radar cloud statistics become increasingly similar.
 Recently, Zhang et al.  presented latitude-dependent statistics on mixed phase cloud occurrence retrieved from combined Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) lidar and CloudSat radar observations. The results for the latitude belt from 30°N to 60°N were found to be in contradiction with respective lidar observations performed at Leipzig, Germany (51°N, 21°E) from 1997 to 2008 [Seifert et al., 2010]. Compared to CALIPSO/CloudSat, these lidar measurements show a much lower mixed phase cloud fraction for the temperature range between −10°C and −25°C. Motivated by this discrepancy and the fact of the increasing number of lidar, radar, and lidar/radar-based cloud statistics [Seifert et al., 2010; Kanitz et al., 2011; Riihimaki et al., 2012], we analyzed our combined ground-based lidar/radar data set covering the time period from August 2011 to January 2013. We tried to answer the question whether it is possible to harmonize lidar and radar data regarding mixed phase cloud observations. This would increase the comparability and therefore the scientific value of past studies. The main question is: What are the threshold values of particle backscatter and extinction above which lidar can detect a cloud being mixed phase? To quantify at which levels of ice formation lidar and radar observations deviate from each other, we take into account the observed ice water path (IWP) and liquid water path (LWP) of the clouds under study. The mass ratio of ice and liquid water within a cloud parcel is decisive for the cloud microphysics and life cycle [Korolev and Mazin, 2003]. We can estimate it by the ratio
with CWP=IWP+LWP being the condensed water path. This ratio is a robust upper estimate for the mass ratio of ice and water in the predominantly liquid top layer of the clouds investigated in the scope of this study. Column integration of the ice water content (IWC) allows us to determine the maximum ice mass that was produced at the cloud top, because water vapor deposition below the liquid layer may contribute to the CWP by an unknown extent.
 In section 2, the ground-based lidar/radar cloud data set is described. section 3 presents the approach to determine the temperature dependence of ice formation efficiency from the considered cloud data sets of spaceborne and ground-based observations. The discrepancies between lidar/radar and lidar data sets are explored in section 4.
2 Data Set
 The data set presented in this work was acquired between August 2011 and January 2013. The observations were conducted with the Leipzig Aerosol and Cloud Remote Observations System (LACROS) [Wandinger, 2012] at Leipzig, Germany. LACROS is part of Cloudnet [Illingworth et al., 2007] and includes the European Aerosol Research Lidar Network (EARLINET) lidar Multiwavelength Atmospheric Raman Lidar for Temperature, Humidity, and Aerosol Profiling (MARTHA) and an Aerosol Robotic Network Sun-photometer (AERONET) as well. Continuous measurements with a 35GHz Mira35 cloud radar, a Humidity and Temperature Profiler (HATPRO) microwave radiometer and a CHM15KX-Ceilometer are performed. The measurements were accompanied by about 3300 h of targeted measurements with Doppler wind lidar “WILI” [Bühl et al., 2012] and 6 months of continuous observations with a PollyXT Raman polarization lidar [Althausen et al., 2009] pointing to an off-zenith angle of 5° so that specular reflections [Westbrook et al., 2010] of falling ice crystals do not affect the detection of ice features.
 Figure 1 presents a case study of an altocumulus layer which is precipitating ice particles. The depolarization ratio in Figure 1b monotonically increases from low values at cloud base to high values around 0.3 at cloud top due to strong multiple scattering of drops in the liquid water layer. Enhanced depolarization ratios below the cloud base, however, indicate the presence of ice crystals down to about 1000m below the base of the liquid water cloud. The lidar hardly detects the virga particles around 6.6km (07:00 universal time coordinated (UTC)) or at 6.3km (07:24 UTC) but clearly sees the liquid cloud top. The radar in turn only detects the precipitating ice particles. Within the ice layer between 6 and 7km height, we derive lidar extinction values at 532 and 1064nm wavelength of 50±20Mm−1(see Figure 1a). Alternatively, the method of Hogan et al.  allows us to calculate the visible extinction from the radar reflectivity (Figure 1c). There were two parameterizations derived by Hogan et al. : One for the visible extinction coefficient and one for the IWC:
 Here Z is the radar reflectivity in dBZ and T the temperature in degrees Celsius. The relations presented here are valid for a 35-GHz radar. For the sake of continuity, the equations are left in the original state as in Hogan et al.  which implies that their units are [m−1] (extinction) and [gm−3] (IWC). It is interesting to note that IWC and the visible extinction from equations (2) and (3) do not depend on particle size. Hogan et al.  use power laws to calculate particle mass m(D) and area A(D) from the measured maximum particle diameter D and parameterize number concentration with a gamma distribution n(T,D). Hence, for radar reflectivity Z(n(D),m(D)2), and visible extinction α(n(D),A(D)), particle size can be eliminated.
 In our example case study (Figure 1), the mean radar reflectivity of falling ice particles is −27.5dBZ so the parametrization yields a mean visible extinction coefficient α of about 40Mm−1. Cloudnet additionally provides us with the liquid water content (LWC) of the cloud profiles. The LWC values are computed by the linear-scaled adiabatic approach of Cloudnet which is assisted by the HATPRO microwave radiometer. The algorithm attributes the liquid water path (LWP) measured by the HATPRO radiometer among all detected liquid sublayers. Together with the parameterizations of Hogan et al. , we can derive the IWP, LWP, and CWP of the cloud under study (Figure 1d).
3 Adaption to CALIPSO/CloudSat Observation Strategy
Zhang et al.  focused on so-called “mid-level liquid-layer topped stratiform clouds” with vertical extent of the liquid layer <500m. The CALIPSO and CloudSat satellites fly in the so-called “A-Train” formation with only 15 s separation. The CALIPSO lidar delivers information about cloud top height whereas the CloudSat radar measures the properties of the precipitating particles below the cloud with a horizontal resolution of about 1400 m. Five CALIPSO profiles are averaged over the 1400m wide footprint of the CloudSat radar. Zhang et al.  analyzed each snapshot-like profile individually and considered it as “mixed phase” if a radar signal was present below the cloud top. Otherwise, it was considered “liquid only”.
 From our data set, we selected 352 cases of midlevel layered clouds (like the one in Figure 1). The number of cloud cases naturally decreases with decreasing cloud top temperature. However, in the interval between −25°C and 0°C (where deviations between lidar and radar are strongest), the cloud numbers are sufficient to provide statistical errors around ±10%. Our study is restricted to mixed phase clouds. Pure ice clouds (e.g., cirrus) are left out intentionally. The selection criteria comprise the absence of low-level cloud cover, only small height variation in cloud base and a maximum vertical extent of the liquid cloud layer of about 500m. Potential precipitation had to be clearly linked to the cloud under investigation, seeding from higher-level clouds for example could be identified by the radar. Additional liquid layers below the cloud could be discriminated from precipitation because of the strong lidar attenuation and/or by the vertical wind velocity characteristic within these layers (alternating up- and downdrafts). Drizzle and rain are identified by terminal fall velocity and lidar/radar depolarization ratio.
 To be able to compare our measurements to those of CALIPSO and CloudSat, we thus try to do a similar evaluation process with our ground-based data which is recorded with 30s averaging time. Assuming a horizontal wind speed of 15m s−1, this would yield a horizontal extent of about 450m for each profile. Equal to the satellite evaluation method, we consider a profile (or a set of profiles) to be mixed phase if falling ice particles are detected with radar below a liquid cloud top. If there is no ice precipitation detected or the precipitation is found to be drizzle, the profile is declared liquid only. The LACROS precipitation classification scheme is presented in Figure 2. At first, the liquid top is detected by lidar. It can be easily recognized by its high backscatter signal and low volume depolarization ratio. The layer is further classified as being mixed phase if the radar detects a melting layer or shows a depolarization ratio greater than −25dBZ. If the radar signal is inconclusive, we search in the lidar signals for the presence of oriented ice crystals or a depolarization ratio greater than 0.25. For each profile, the cloud top height is estimated and the corresponding cloud top temperature and advection speed are derived from the Global Data Assimilation System (GDAS1) reanalysis data set. With help of the advection speed, the geometrical length of each profile can be determined by multiplying its time duration with the advection speed. From the single profiles, the mixed phase cloud fraction is calculated: For each 5-K interval of cloud top temperature, the total length of the ice-containing profiles is divided by the length of all profiles.
 This profile-based evaluation method is different to previous lidar-based studies [e.g., Seifert et al., 2010; Kanitz et al., 2011] where always complete cloud cases (lasting from minutes to several hours) were classified to be “ice-containing” if there were ice particles detected anywhere below the liquid cloud base.
 The mixed phase cloud fraction per temperature interval is shown in Figure 3 where we compare the results of this study (red curve) with the study of Seifert et al.  (black curve) and the satellite-based study of Zhang et al.  (green curve). The results of this study compare well with the satellite measurements over the whole temperature range. However, they deviate strongly from the results of Seifert et al. .
 There is obviously a difference in the ability of lidar and radar to detect falling ice particles. We will therefore investigate the conditions (α, IWC, and IWP/CWP) at which both lidar and radar can equally detect ice formation. In this way, we try to obtain a better agreement between the lidar-based and radar-based mixed phase cloud statistics. We eventually may be able to define the threshold conditions when significant ice formation is taking place within a mixed phase cloud.
 In Figure 4, the mean visible extinction α of ice particles within the cloud virga and the corresponding mean IWC is shown for each cloud case retrieved with the parameterizations of Hogan et al.  from the mean radar reflectivity (Z) of the ice particles falling below the liquid cloud top. Assuming a lidar detection threshold of αT=30Mm−1 for falling ice particles, we can reproduce the graph of Seifert et al.  from the data of this study by defining all cases which lie below this threshold as liquid only. The resulting curve of mixed phase fraction is shown in Figure 3(dashed purple). This corresponds to the introduction of an IWC detection threshold of 10−6kgm−3. In the context of the parameterizations used here, both values correspond to a radar reflectivity factor of −27dBZ for the temperature interval between −40°C and −10°C.
 The threshold proposed here should not be confused with the signal-to-noise ratio of a lidar system. The PollyXT and the MARTHA lidars are capable of detecting extinction values well below 30Mm−1. However, assuming this threshold, we can derive the temperature dependence of ice formation previously published in Seifert et al.  and also from our present study. The detailed reason for the different ice detection capabilities of lidar and radar might be more difficult to describe because of the large technical differences (e.g., emitting wavelength, integration time, beam divergence, and signal processing) between the two systems.
 Up to this point, only the presence of ice at different temperature levels has been studied. To further quantify the process of ice formation, we compute the liquid and ice water paths (IWP and LWP) of the clouds under study by column integration of the IWC and LWC values of each profile. Both IWC and LWC are quality-assured products of Cloudnet, but we want to note again that all cloud cases in this study were checked for elements which could potentially compromise this approach. For example, in the presence of a melting layer, the IWC integration is always started above it. The calculation of the LWP is restricted to the liquid cloud top only. For each vertical profile, we also compute the ratio of the IWP to the condensed water path (CWP). It is worth noting that in the case of a mixed phase cloud, the IWP/CWP ratio is averaged only over profiles where ice was detected making the IWP/CWP values an upper estimate for each single cloud case.
 The average IWP, LWP, and IWP/CWP ratio for all cloud cases are shown in Figures 5 and 6. The case study of Figure 1 is marked with a black star (at temperature T=−24.5°C); all cases in Figures 5 and 6 which fall below the lidar IWC detection threshold are marked with red squares. It is visible that the example case from Figure 1 shows relatively low values of both IWP=5.3·10−4kgm−3 and IWP/CWP=3.4·10−2compared to other cases in its temperature interval and lies close to the lidar detection threshold. Since this threshold is derived in a purely statistical manner, it is interesting to see from Figure 1 that the lidar signals within the ice layer are indeed very low.
 Ninety percent of the cases that fall below the lidar IWC detection threshold show an IWP smaller than 10−3kgm−2. 85% of the IWP/CWP ratios are below 10−1; 60% are below 10−2. The IWP/CWP ratio indicates the maximum mass ratio between ice and liquid water within the mixed phase cloud top.
 The temperature at which half of the cloud profiles contain ice is −10°C in this study (see red curve in Figure 3). Yet it cannot be directly concluded from these measurements that at this temperature level, ice production plays a major role for the further development of the cloud layer toward glaciation or precipitation formation. The high measurement sensitivity of powerful radar systems like the Mira35 or the Cloud Profiling Radar on CloudSat could therefore be misleading. Hence, the ability to measure and quantify the efficiency of heterogeneous ice production from ground can significantly improve our understanding of heterogeneous ice formation. In this context, the IWP/CWP ratio seems to be an interesting measurement quantity since the ratio of ice to liquid water strongly influences the microphysics of a mixed phase cloud. It may be necessary to take into account the level of cloud glaciation when investigating heterogeneous ice formation because the mass ratio between ice and water was found to deviate by several orders of magnitude. It could distort our picture of heterogeneous ice formation significantly if we put a cloud with an IWP/CWP ratio of 10−4 into the same category with a cloud which is just about to glaciate completely and therefore has an IWP/CWP ratio approaching 1.0.
 The red and the green curves in Figure 3 basically tell us that in the atmosphere, ice formation is occurring at any temperature level below 0°C. Hence, we have to quantify the amount of ice and water to learn where ice formation plays a major role for cloud microphysics. Hill et al.  for example show that it is still difficult to assess at which levels of ice formation a mixed phase cloud will turn into an ice cloud. We show that ground-based remote sensing can give us additional information about such cloud systems allowing to estimate the ratio of ice and liquid water. However, we have to keep in mind that our IWC (and therefore also the IWP measurements) are estimations and to a certain amount reflect the parametrization method they were retrieved with. The method of Hogan et al.  gives us a good start, but in future, more direct ways to measure the IWC by synergistic combination of active and passive remote sensing and in situ observations must be found.
 Our study confirms previous satellite-based measurements of mixed phase cloud occurrence over the temperature range between −30°C and 0°C. We explain discrepancies to previous own measurements by taking into account the IWC, IWP, and LWP of the clouds and calculate an approximate IWC detection threshold of 10−6kgm−3 for the PollyXT and MARTHA lidars used by Seifert et al.  and in this study.
 The Editor thanks two anonymous reviewers for their assistance evaluating this manuscript.