SEARCH

SEARCH BY CITATION

Keywords:

  • heterogeneous ice nucleation;
  • Saharan dust;
  • lidar

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[1] More than 2300 observed cloud layers were analyzed to investigate the impact of aged Saharan dust on heterogeneous ice formation. The observations were performed with a polarization/Raman lidar at the European Aerosol Research Lidar Network site of Leipzig, Germany (51.3°N, 12.4°E) from February 1997 to June 2008. The statistical analysis is based on lidar-derived information on cloud phase (liquid water, mixed phase, ice cloud) and cloud top height, cloud top temperature, and vertical profiles of dust mass concentration calculated with the Dust Regional Atmospheric Modeling system. Compared to dust-free air masses, a significantly higher amount of ice-containing clouds (25%–30% more) was observed for cloud top temperatures from −10°C to −20°C in air masses that contained mineral dust. The midlatitude lidar study is compared with our SAMUM lidar study of tropical stratiform clouds at Cape Verde in the winter of 2008. The comparison reveals that heterogeneous ice formation is much stronger over central Europe and starts at higher temperatures than over the tropical station. Possible reasons for the large difference are discussed.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[2] Laboratory studies since the 1950s indicate that mineral dust particles are favorable ice nuclei (IN) [Pruppacher and Klett, 1997], and may have a significant impact on cloud and precipitation processes [Levin et al., 1996, 2005]. Stimulated by the potentially important role of desert dust for weather and climate we conducted several dust-related cloud studies based on lidar observations [Ansmann et al., 2005, 2008, 2009]. We analyzed several hundred altocumulus layers observed over Morocco (31 cases), close to the Saharan dust source, and over Cape Verde (227 cases), in an area with continental aerosol outflow of aged dust and biomass burning smoke from western and tropical Central Africa. The Morocco and Cape Verde observations were performed in the framework of the Saharan Mineral Dust Experiment (SAMUM) project [Heintzenberg, 2009]. The main result was that significant cloud glaciation does not occur as long as the cloud top temperature is above −20°C, even in the case of a rather high dust particle load and thus high ice nuclei concentrations of 1–20 cm−3 [Ansmann et al., 2008]. The findings were in good agreement with laboratory studies [Field et al., 2006; Connolly et al., 2009].

[3] Concurrently to our activities at the subtropical and tropical sites, the DRIFT (Dust-related Ice Formation in the Troposphere) campaign at the central European EARLINET (European Aerosol Research Lidar Network, http://www.earlinet.org [Wandinger et al., 2004]) site of Leipzig was conducted from February 2006 to June 2008 with focus on heterogeneous ice formation in stratiform clouds in the free troposphere. As part of the DRIFT study, we reanalyzed our 1997–2006 EARLINET data set with special emphasis on cloud glaciation and ice formation in liquid water and mixed phase clouds. More than 2300 cloud layers were analyzed.

[4] When measuring at a central European site, the unique opportunity is given to contrast cloud observations in dust-free air masses from the North Atlantic and observations in air containing Saharan dust. Ansmann et al. [2003] presented a case study of a major Saharan dust outbreak toward northern Europe and later discussed the role of aged dust regarding ice formation [Ansmann et al., 2005]. Such a contrasting study (dust versus dust-free) as presented below is almost impossible in regions like Africa or Asia where dust is omnipresent throughout the troposphere.

[5] In section 2, we briefly describe the lidar systems and provide information on the atmospheric temperature profiles required in the analysis. The Dust Regional Atmospheric Modeling (DREAM) system that was used for the determination of the Saharan dust load is introduced. Section 3 deals with fundamental aspects regarding the discrimination of liquid water, mixed phase, and ice clouds by means of a vertically pointing polarization lidar. Based on a measurement with the vertically pointing EARLINET lidar and a simultaneously operated scanning Doppler lidar [Seifert et al., 2008], our approach to identify ice-containing cloud layers with the EARLINET lidar is explained. The statistical results of the 1997–2008 cloud data set are presented in section 4. Two case studies are discussed in addition. In section 5, we compare the DRIFT results with the observations taken over Cape Verde in the tropics during SAMUM in 2008 [Ansmann et al., 2009]. Large differences regarding the occurrence of ice-containing cloud layers as a function of cloud top temperature are found. The comparison of the tropical and midlatitude data sets provide an opportunity to discuss the meteorological and aerosol-related aspects of heterogeneous ice formation in a broader context. A summary and concluding remarks are given in section 6.

2. Instrumentation and Data Analysis

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

2.1. MARTHA

[6] The data were collected with the stationary Raman lidar MARTHA (Multiwavelength Atmospheric Raman lidar for Temperature, Humidity, and Aerosol profiling) [Mattis et al., 2002, 2004, 2008] at Leipzig (51.3°N, 12.4°E), Germany. Regular measurements were carried out three times a week, i.e., Monday afternoon, and Monday and Thursday after sunset. From April 1997 to June 2008 almost 900 measurement sessions (regular and special observations, 2344 measurement hours) were performed.

[7] The Nd:YAG laser of MARTHA transmits laser pulses at 355, 532, and 1064 nm wavelength at an elevation angle of 90°. The lidar return signals were stored with 60 m vertical and 30 s temporal resolution. For the transmitted wavelengths, vertical profiles of particle backscatter and extinction coefficients are determined [Ansmann et al., 1992; Mattis et al., 2004; Ansmann and Müller, 2005; Seifert et al., 2007; Ansmann et al., 2009].

[8] All lidar sessions lasting 30 minutes to several hours were documented as height-time displays of range-corrected signal and volume depolarization ratio. The volume depolarization ratio is defined as the ratio of the cross-polarized to the parallel-polarized signal component (with respect to the transmitted linearly polarized laser light at 532 nm). In the scope of DRIFT, cloud layers were classified in the same way as described by Seifert et al. [2007]. A cloud layer was counted as a single, isolated cloud when it was separated by more than 500 m in the vertical or by 5 minutes (temporally, horizontally) from neighboring cloud layers.

[9] The detection of ice crystals is based on the measurement of the volume depolarization ratio. Backscattering (single scattering at exactly 180°) by spheres (liquid drops) does not produce any depolarization. Backscattering by nonspherical crystals introduces considerable depolarization (via the process of many internal reflections). Thus, the volume depolarization ratio allows us to discriminate cloud layers consisting of water drops and cloud layers in which backscattering by ice crystals dominate [Sassen, 2005; Ansmann et al., 2009]. It should however be clearly mentioned that lidar is not sensitive enough to detect ice crystals with typical concentrations of below 50 per liter in the center and top regions of a cloud layer with drop concentrations of 50–100 cm−3. Thus we had to develop an alternative approach to identify the ice-containing clouds. This method is explained in section 3.

[10] Specular reflection by aligned falling ice crystals complicates the discrimination between liquid droplets and ice crystals when the laser beam is pointed to the zenith. This aspect is also discussed in section 3 based on a measurement example with the scanning wind Doppler lidar (WiLi) in combination with MARTHA.

2.2. WiLi

[11] A coherent zenith-pointing Doppler lidar enables us to measure the vertical motion of aerosol particles, cloud droplets, and ice crystals [Engelmann et al., 2008; Ansmann et al., 2009]. The transmitted wavelength is 2.022 μm, the vertical and temporal resolution is 75 m and 5 s, respectively. The laser beam can be pointed to any zenith angle with a resolution of 0.1°. This potential is used as described in section 3.

2.3. Temperature and Humidity Profiles

[12] Cloud top temperatures are required in the cloud analysis. Heterogeneous nucleation of ice crystals typically starts at the coldest part of the cloud (cloud top) [Rauber and Tokay, 1991; Lebo et al., 2008]. The meteorological data (temperature and relative humidity) were taken from different sources. For our study we archived the data with 6 h resolution (for the 11 year period). The nearest radiosonde station of the German Meteorological Service was Oppin, about 30 km northwest of Leipzig. However, radiosondes were only launched during the period from March 2000 to September 2006. To cover the missing periods before and after the Oppin radiosonde time period, as well as the time intervals between the soundings (two sondes per day were launched most of the time), data of the U.S. National Weather Service's National Center of Environmental Prediction (NCEP) were used. NCEP maintains a data archive that contains the assimilated observational data for the initialization of weather forecast models. This archive is based on the global data assimilation system GDAS (Global Data Analysis System, http://www.arl.noaa.gov/gdas.php) [Kanamitsu, 1989] that stores the assimilated data fields, including ground-based observations as well as radiosonde and satellite-based data. Between 1997 and November 2004 the assimilated data were archived into the FNL (FiNaL, http://www.arl.noaa.gov/fnl.php) with 6 h resolution. The FNL data set contains the assimilated data that were used to initialize the final (FNL) run of the global model of NCEP. The nearest grid point of both data archives (GDAS, FNL) is 25 km south of Leipzig.

[13] In order to check the uncertainties in the used temperature profiles, the difference between the radiosonde-derived and model-derived temperatures were analyzed for all available radiosonde profiles (2000–2006). The temperature difference was found to be small for the height range of interest from 2 to 10 km height with a standard deviation of 0.7 K. Thus, the uncertainties in the cloud top temperatures used in the data analysis are on the order of 1 K.

[14] The model-derived humidity profiles (shown in the case studies in the next section) have to be exercised with care. Atmospheric models are unable to simulate detailed and accurate humidity conditions in shallow heterogeneous cloud fields.

2.4. Regional Dust Model DREAM

[15] We used the column dust load and vertical profiles of the dust mass concentration computed with the regional dust model DREAM [Nickovic et al., 2001] to separate cloud observations under dust and dust-free conditions. DREAM delivers operational dust forecasts for North Africa, the Middle East, and Europe that are verified on a daily base with observational data. It was selected to deliver dust forecasts for the SAMUM field campaign and has been intensively tested and validated using available observational data [Pérez et al., 2006a, 2006b; Haustein et al., 2009].

[16] Model results are available for the grid point Leipzig with 12 h resolution for the entire lidar observation period from 1997 to 2008. Before 2007, data were taken from a 48 year reanalysis (1958–2006) at 0.3° × 0.3° resolution [Pérez et al., 2007]. In this simulation, meteorological fields were initialized every 24 h and boundary conditions updated every 6 h with the NCEP/NCAR I global reanalysis (2.5° × 2.5° resolution).

[17] We assigned the respective profile of dust concentration given in μg m−3obtained from DREAM calculations for grid point Leipzig to each cloud layer with cloud top temperature from 0 to −40°C. In section 4.2, we compare cloud statistics (relative number of ice-containing clouds versus cloud top temperature) for dust-free or dust-laden air masses.

3. Identification of Ice-Containing Clouds

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[18] MARTHA is mounted in the uppermost floor of the main building of our institute and points exactly vertically (0° zenith angle). Two aspects (multiple scattering, specular reflection) complicate the detection of ice. Strong multiple scattering by liquid, spherical drops introduces depolarization [Bissonnette, 2005; Ansmann et al., 2009] that may be misinterpreted as backscattering from ice crystals. Fortunately, the multiple scattering effect shows a clear signature in water and mixed phase clouds and thus can easily be identified [Ansmann et al., 2009]. The volume depolarization ratio increases monotonically with increasing laser light penetration into the cloud, from values close to zero at cloud base (weak multiple scattering impact) to values of 0.1–0.15 at cloud top in the case of MARTHA (receiver field of view of 0.4 mrad). Therefore, if a cloud layer causes rather low volume depolarization ratios at cloud base and does not produce any virga of ice (enhanced depolarization), we count this cloud as a pure liquid cloud.

[19] The second effect can lead to rather complex features in the measured depolarization ratio profiles. Falling ice crystals (plates and columns) with plate diameters or column lengths of roughly above 100 μm assume an almost perfect horizontal orientation (longest axis perpendicular to the falling direction) and then cause strong specular reflection [Platt, 1978; Sassen, 1980; Thomas et al., 1990; Seifert et al., 2008]. When pointing to the zenith, specular reflection leads to volume depolarization ratios very close to zero. These low values cannot be distinguished from depolarization ratios resulting from backscattering by spherical liquid droplets. Based on cloud chamber studies and modeling Amsler et al. [2009] showed that randomly oriented plates and columns with rather large aspect ratios (maximum crystal length or width divided by crystal thickness) of more than 100 may also cause depolarization ratios below 0.05 which are hard to be distinguished from liquid drop depolarization behavior. However, our extended cirrus data set obtained at a zenith angle of 5° during SAMUM in 2006 (Morocco) and 2008 (Cape Verde) never showed layers in cirrus clouds with such low depolarization ratios so that we conclude that ice crystals with aspect ratios above 100 are rare and do not affect ice detection.

[20] By considering these disturbing effects (multiple scattering, specular reflection), we identify ice-containing cloud layers as follows. Clouds are classified as ice-containing if their depolarization ratio varies strongly with height or shows high values throughout the cloud layer. In ambiguous cases, we analyze the depolarization measurement at cloud base and below the main cloud (virga region) in detail. Furthermore, a cloud is classified as ice-containing if it produces a virga which contains ice crystals and/or if it shows an increased volume depolarization at cloud base. We assume that all cloud drops evaporate and the falling large crystals become quickly smaller at the lower cloud edge in the region with relative humidity over water of below 100%. When the length of the longest axis of the crystals is less than 100 μm, the crystals are no longer able to attain their horizontal orientation. This behavior was also found in a recent study of Westbrook et al. [2010]. Even large crystals may already be randomly oriented caused by turbulence at cloud base. Ice breakup effects and changes in the shape characteristics may further favor a random distribution of evaporating crystals. These remaining ice particles produce a significantly enhanced depolarization signal. This is a typical feature of mixed phase clouds observed in our 11 year data record.

[21] Our approach of ice detection is explained in Figure 1 which shows a measurement example. Simultaneous observations with the scanning Doppler lidar WiLi and the EARLINET polarization lidar MARTHA are shown. We conducted many observations of this kind from November 2006 to March 2007 to investigate the impact of specular reflection on lidar cloud observations [Seifert et al., 2008]. Such combined lidar experiments provide a unique insight into the influence of falling ice crystals on the volume depolarization ratio. In these measurements, WiLi was pointing to the zenith in the beginning. Afterward, the laser beam was tilted, step by step (with 0.2°–0.5° resolution), to zenith angles of up to 3°. This scan pattern (from 3° over 0° to −3° and back) was then repeated many times. When pointing to the zenith, a sharp increase in the backscatter signal strength occurs in the case of specular reflection, while the observed (negative or positive) strength of the line-of-sight velocity is no longer affected by horizontal wind components and thus takes its minimum (close to 0 m/s).

image

Figure 1. Influence of specular reflection by horizontally aligned ice crystals on measurements with a vertically pointing polarization lidar (MARTHA) and a continuously scanning lidar (WiLi). The measurement was taken at Leipzig, Germany, on 16 January 2007. Range-corrected 1064 nm signal and 532 nm volume depolarization ratio were measured with MARTHA. Signal-to-noise ratio (expressing the strength of particle backscattering) and line-of-sight (LOS) velocity are derived from WiLi measurements. The laser beam of WiLi is step by step pointed to zenith angles from −3° over 0° to 3° and back and forth. The zebra-stripe pattern in the height-time display of the WiLi signal-to-noise ratio indicates horizontally aligned ice crystals (backscatter maximum at 0°). Zebra patterns in the LOS velocity are caused by the increasing contribution of the horizontal wind speed to the LOS velocity with increasing off-zenith angle.

Download figure to PowerPoint

[22] The measurements shown in Figure 1 indicate low-level water clouds (below 2 km height), midlevel mixed phase altocumulus (2–6.5 km height) and cirrus (above 6.5 km height). The extended ice cloud at 6 km height (best seen in the MARTHA 1064 nm signal plot) causes a zebra pattern (stripes) in the plot of the WiLi backscatter signal-to-noise ratio (SNR) in the upper part of the cloud layer (with maximum values at 0° when the velocity component is close to 0 m/s). An almost scan-angle-independent backscattering was observed in the lower part of this cloud layer. Thus, specular reflection occurs in the upper part of the cloud, but not in the lower part. Correspondingly, the volume depolarization ratio is very low in the upper part of the cloud layer and high in the lower part as a result of backscattering by randomly oriented ice crystals. Further ice-containing clouds occur at 3 km and 4.0–4.5 km before 1115 UTC. The volume depolarization ratio is weakly enhanced at the base of these clouds and indicates the presence of ice crystals.

[23] Zebra patterns are absent in the water clouds below 2.5 km height and in the aerosol layers below 1.5 km height. Also, the cirrus cloud detected after 1150 UTC (upper right corner in the color plots) does not indicate specular reflection so that the volume depolarization ratio is high in this cloud. Depolarization is generally low in water clouds and aerosol layers (except in the presence of desert dust) in the lower troposphere.

[24] Figure 1 corroborates our basic hypothesis that a pure liquid water cloud shows no enhanced volume depolarization ratio at cloud base, opposed to ice-containing clouds. Ice virgae may remain as the most critical cloud element regarding the determination of the phase of the cloud because specular reflection can occur from virga base to top during the short time period during which the virga crosses the laser beam. However, ice virgae are usually linked to large and vertically extended cirrus fields. They can easily be defined due to their descending behavior.

4. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

4.1. Overview

[25] An overview of all 2319 cloud observations taken from April 1997 to June 2008 is given in Table 1. Most of the analyzed clouds are stratiform and occurred during the approach of frontal systems from the North Atlantic. The statistical analysis presented below is based on the well-defined cloud cases only for which cloud phase and cloud top height could be clearly identified. Unclear cases (20% out of all 2319 cases) include 230 low-quality depolarization measurements caused by frequent changes of the receiver unit during lidar upgrading periods so that a proper alignment was not always possible. Further 190 cases are related to opaque clouds for which the cloud top could not be identified.

Table 1. Cloud Statistics From the 1997–2008 Lidar Data Recorda
April 1997 to June 2008CasesPercentage (%)
  • a

    In 82% of the observed 2319 cases, the cloud layers were well defined in terms of cloud phase and top height; wd, well defined.

Observed cloud layers (all cases)2319 
Observed cloud layers (wd)1899100
Liquid water clouds (wd)79042
Mixed phase clouds (wd)23512
Ice clouds (wd)87446
Clouds with top temperatures < −40°C (wd)75340
Clouds with top temperatures > −40°C, <0°C (wd)78942
Clouds with top temperatures > 0°C (wd)35719

[26] Disregarding the ambiguity in the depolarization observation (see section 3), we made an attempt to classify all cloud layers. As shown in Table 1, 46% out of all well-defined cloud cases (1899 cases) are categorized as cirrus layers (mostly with cloud top temperatures of below −40°C). In 54% out of all well-defined cases, we observed liquid water clouds (42%) and mixed phase clouds (12%). These clouds are mostly altocumuli, altostrati, and stratocumuli. The uncertainty in this classification is about 5%–10%.

[27] Figure 2 shows the distribution of cloud top heights and cloud depths of the identified well-defined liquid water, mixed phase, and ice clouds. Histograms for 500 m height and depths intervals are presented. As can be seen, pure liquid water clouds with cloud top heights of above 6.5 km are rare. Most of the well-defined liquid clouds are geometrically thin with depths lower than 500 m. The lidar was not operated during overcast situations or when the probability for the occurrence of precipitation was given. Therefore deep cumuliform clouds are missing in the statistics. Most of the mixed phase clouds are found between 2.0 and 8.5 km height. The geometrical depths of these clouds are less than 2 km in 80% out of all cases. Cloud depths include the virga layer in the case of the mixed phase and cirrus clouds in Figure 2. A broad distribution of cirrus depths is observed.

image

Figure 2. Distribution of (a) cloud top heights and (b) cloud depths, separately for liquid water clouds (black bars, 790 cases), mixed phase clouds (shaded bars, 235 cases), and ice clouds (white bars, 874 cases). The statistics is based on the 1899 well-defined cloud cases of the 1997–2008 cloud data set.

Download figure to PowerPoint

[28] Figure 3 shows the ratio of ice-containing clouds to the total number of observed cloud layers for eight 5 K temperature intervals from 0 to −40°C. We counted 789 cloud layers within this temperature range (see Table 1). A strong increase of the relative number of ice-containing clouds is observed in the temperature range from −5 to −25°C. The error bars express the statistical significance. The corresponding standard error σ depends on the number of observed cloud events n and the fraction f of ice-containing clouds according to

  • equation image
image

Figure 3. Frequency of occurrence of ice-containing clouds as a function of cloud top temperature (for 5 K intervals, solid line). Error bars (after equation (1)) indicate the statistical uncertainty. The dashed curve is obtained after correction of the estimated natural cloud seeding effect. Total number of observed clouds for each temperature interval is given at the top.

Download figure to PowerPoint

[29] In Figure 3, the impact of natural cloud seeding [Rutledge and Hobbs, 1983; Ansmann et al., 2009] on the fraction of ice-containing clouds is estimated in addition. In the seeding process ice particles from a seeder cloud grow as they fall through a lower-level feeder cloud. The ice particles serve as centers upon which the available water in the feeder zone can be collected and converted to precipitation. New ice particles are thus not formed in the lower cloud layer. Cloud seeding effects could be easily detected in the tropics during the SAMUM 2008 campaign and were found to significantly influence the height distribution of ice in the free troposphere [Ansmann et al., 2009]. Therefore, we made also an attempt to roughly estimate the potential impact of cloud seeding on midlatitude clouds. However, an estimation of the seeding effect is much more complicated at midlatitudes because of the complex occurrence of cirrus decks, fragments, remnants, and virgae, layered mixed phase clouds and virgae originating from the mixed phase clouds. As a consequence, we inspected any observed cloud scenario in the middle to upper troposphere at temperatures below −25°C for potential cloud seeding effects by eye, and counted an ice-containing cloud as a liquid cloud in all cases in which a higher cloud layer or an ice virga was sufficiently close to the lower mixed phase or ice cloud layer of interest. We simply assume in the estimation that the ice in the lower cloud is caused by seeding from above. For higher temperatures above −25°C, a more objective method can be applied. Here, we count any mixed phase cloud layer occurring below another ice-containing cloud layer (cirrus or mixed phase cloud) within one kilometer vertical distance between the cloud base of the higher cloud and the cloud top of the lower cloud as a liquid water cloud. The result of these corrections in the cloud statistics is shown in Figure 3 (dashed line). The correction leads to a reduction in the fraction of ice-containing clouds by, on average, 10%. Because the effect is small and the uncertainty of this rough estimation is in the same order of magnitude, we ignore cloud seeding effects on the DRIFT data set in the further discussion.

[30] In Figure 4, we compare our cloud statistics with airborne in situ observations conducted in Germany (P40, 180 km northeast of Leipzig), in the former USSR (B63), Canada (I79), and at different sites in North America (42°N to 76°N) from about 1930 to 2000. The curves P40 and B63 (9000 aircraft samples) in Figure 4 are based on impactor samplings. All samples that showed no ice crystals were counted as liquid clouds. The bulk of the K03 data was sampled in stratiform clouds. The I79 data were recorded in convective clouds.

image

Figure 4. Comparison of the DRIFT observations with airborne in situ measurements at different midlatitudinal sites (data are taken from Figure 14 of Korolev et al. [2003]). Cloud observations of Peppler [1940] (P40), Borovikov et al. [1963] (B63), Isaac and Schemenauer [1979] (I79), and [Korolev et al., 2003] (K03) are considered. The airborne observations are related to temperatures at flight level, whereas the lidar results are given as a function of cloud top temperature.

Download figure to PowerPoint

[31] The comparison is performed to check the reliability of the lidar data set, i.e., whether the lidar data provide a biased picture of the height distribution of ice-containing clouds because of the limitations of the lidar method discussed in sections 2 and 3. As mentioned, the polarization lidar is not sensitive enough to identify ice crystals in a cloud layer that mostly consists of water drops. So, we generally underestimate the number of mixed phase clouds. However, Figure 4 indicates that the lidar data well reflect the frequency distribution of ice-containing clouds in the vertical column of the troposphere at midlatitudes. The systematic deviation of the lidar data from the P40 curve (obtained at Lindenberg, close to Leipzig) can widely be explained by the fact that the in situ measurements are related to ambient temperatures (at flight level), but the lidar results are related to cloud top temperatures. The cloud top temperature can be about 2–6 K lower for clouds with depths of 200–1000 m, and even lower for thicker cumulus clouds. By shifting the lidar-derived curve in Figure 4 by 3–5 K (toward higher temperatures) a good agreement to the P40 observations is obtained.

4.2. Clouds in Dust-Laden and Dust-Free Air

[32] As described in section 2.4, we combined DREAM-derived vertical profiles of dust mass concentration with the cloud phase information obtained from the polarization lidar observations. Dust-free and dust-laden cloud cases were identified by examining the maximum dust mass concentration within 500 m of the cloud boundaries. Out of the total number of 789 cloud cases with cloud top temperatures from 0 to −40°C, 218 clouds formed in air masses with significant Saharan dust mass concentration within 500 m of the cloud boundaries of more than 2 μg m−3, and 213 cloud layers were observed in dust-free air which corresponds to a dust mass concentration within 500 m of the cloud boundaries of below 0.001 μg m−3. A dust mass concentration of 0.001 μg m−3 is the lowest output level of DREAM above 0 μg m−3. Two cases representing very different dust conditions are shown in Figures 5 and 6. DREAM results for these two cases are given in Figures 7 and 8.

image

Figure 5. Ice-precipitating altocumulus clouds between 3 and 6 km height on 20 June 2007. The clouds develop in the upper part of a Saharan dust layer. Range-corrected 1064 nm signal (top, arbitrary units) and volume depolarization ratio at 532 nm are shown. Green to red column-like features in the depolarization plot from 3 to 5 km height indicate virga-containing ice crystals. Height profiles of the simulated GDAS temperature and humidity profiles for the Leipzig grid point at 1200 UTC, particle extinction coefficient at 532 nm, and estimated IN concentration are given. Aerosol properties are computed from the mean signal profiles observed from 0950 to 1040 UTC (indicated by a white frame). Temperature levels (horizontal lines) are taken from the temperature profile. Backward trajectories ending at Leipzig at 2, 4, and 6 km height are computed with HYSPLIT [Draxler and Rolph, 2003; Draxler et al., 2009].

Download figure to PowerPoint

image

Figure 6. Same as Figure 5, except for 11 July 2006. Liquid water clouds (altocumulus) occur at 5, 6, and 6.5 km height. The liquid phase is indicated by the low depolarization. The cloud layers developed at aerosol background conditions according to the low particle extinction coefficients, estimated IN concentrations, and low depolarization ratios.

Download figure to PowerPoint

image

Figure 7. Vertical profiles of dust mass concentration calculated with the DREAM model for grid point Leipzig, Germany, on 11 July 2006 (solid, see Figure 8, left) and 20 June 2007, 1200 UTC (dashed, see Figure 8, right).

Download figure to PowerPoint

image

Figure 8. Column mass concentration of desert dust particles calculated with the DREAM model for (left) 11 July 2006 and (right) 20 June 2007, 1200 UTC. The EARLINET station Leipzig (LE) and the SAMUM 2008 lidar site at Cape Verde (CV) are indicated in addition.

Download figure to PowerPoint

[33] A moderate Saharan dust outbreak was observed on 20 June 2007 (see Figure 5). The dust layer extends from the top of the boundary layer up to 5–6 km height. Traces of dust are detected up to 8 km height. According to the HYSLPIT backward trajectories dust is advected from Northern Africa at heights from 2 to 5 km and traveled roughly 3–5 days before arriving at Leipzig. Large dust particles with radii above 5 μm may have been removed after 3 days of long-range transport [Tegen and Fung, 1994; Tegen and Lacis, 1996].

[34] The particle extinction coefficient σp(z) derived from the lidar observations ranges from 20 to 60 Mm−1 in the main dust layer from 3.5 to 5.0 km height. Dust concentrations above 100 μg m−3 occur around 4 km (Figure 7, dashed line). Figure 8 (right) gives an overview of the Saharan dust outbreak toward northern Europe on 20 June 2007 in terms of column dust load.

[35] To provide a very rough estimate (order of magnitude) of the ice nuclei concentration INC at such dust conditions, we computed INC(z) from the extinction coefficient σp(z) after INC(z) = cpσp(z) with cp = 0.006 Mm cm−3 [Ansmann et al., 2008, 2009]. As can be seen in Figure 5, INC values are on the order of 10−1 cm−3 in the dust layer at heights below 5 km height. Free tropospheric INC background values are on the order of 10−2 to 10−3 [Richardson et al., 2007].

[36] The range-corrected signal in Figure 5 indicates that shallow altocumuli form at the top of the dust layer at temperatures above −13°C. Volume depolarization ratios of more than 0.3 (yellow to red) are observed in several virgae below the main cloud layer and indicate ice formation. Further mixed phase and ice clouds are present from 7 to 11.5 km height.

[37] The shown modeled humidity profiles (Figure 5) must be interpreted with care. However, the profile indicates both moist layers in which the altocumulus and cirrus clouds develop.

[38] An example for dust-free conditions at cloud level is shown in Figure 6. The backward trajectories indicate an air mass transport from North America. Tropospheric aerosol background conditions prevail as indicated by the low extinction coefficients and INC estimates. Three liquid water cloud layers occur close to 5, 6, and 6.5 km height. The depolarization ratio is at all lower than 0.05 in the clouds that formed at temperatures from −6 to −16°C. According to Figure 7 (solid line), dust concentrations at cloud level are below 1 μg m−3. Again, the modeled humidity profile roughly indicates the moist layer in which the clouds evolve.

[39] Figure 9 shows the temperature dependence of the occurrence of ice-containing clouds separately for dust-free and dust-containing air masses. A significantly increased fraction of ice-containing clouds is observed in Saharan air masses in the temperature range from −10 to −20°C. The fraction increases from 38% to 53% (−10 to −15°C) and from 50% to 81% (−15 to −20°C) and thus by a factor of 1.6. The impact of dust outside the temperature range from −10 to −20°C seems to be weak. The comparison suggests that dust particles are not activated as IN at temperatures above −5°C, and, at colder temperatures, there are generally enough IN (dust and nondust nuclei) to initiate heterogeneous freezing. But in the temperature range from −5 to −20°C, where mixed phase clouds dominate, dust is obviously playing a major role.

image

Figure 9. Frequency of occurrence of ice-containing clouds as a function of cloud top temperature (5 K intervals) for clouds which formed in dust-free air (dashed, open circles) and in dust-containing air (solid, closed stars). The gray line shows the curve for the full cloud data set (as shown in Figure 3). All clouds from 0 to −40°C (from near the ground up to about 8.5 km height) are included in the analysis. Error bars (after equation (1)) indicate the statistical uncertainty. Total numbers of observed clouds for each 5 K interval and separately for each statistics (218 dust cases, 213 dust-free observations) are given at the top.

Download figure to PowerPoint

[40] Different meteorological conditions (strength of updrafts, downdrafts, turbulent mixing, entrainment of dry air into the cloudy environment, drop evaporation, drop size broadening) in the different air masses influence the results in Figure 9. However, because of similar cloud characteristics (horizontal extent of cloud fields, base and top height distributions, geometrical depths, extinction coefficients, optical depths) for the two cloud data sets, we assume that the meteorological conditions were quite similar so that the observed differences are related to dust. Meteorological aspects are further discussed in the next section.

5. Midlatitudinal Versus Tropical Observations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[41] The comparison of the Cape Verde observations during the SAMUM campaign and the DRIFT results provide an opportunity to discuss the aerosol-related influence on heterogeneous ice formation in a broader context and to include meteorological and geographical aspects in the discussion. The comparison is shown in Figure 10. The Cape Verde (CV) lidar site is indicated in Figure 8. The two SAMUM curves in Figure 10 are taken from Figure 15 of Ansmann et al. [2009]. A strong difference between the midlatitudinal and tropical cloud statistics is found. The onset temperature for significant ice formation over Europe is in the range from −5 to −10°C (even if we consider the cloud seeding effect, see Figure 3, or the dust-free cases only in Figure 9). In contrast, significant ice formation was not observed for temperatures above −10 to −15°C or even −20°C (CV, cloud seeding–corrected statistics) over Cape Verde. A similar result (onset temperature below −20°C) was found during SAMUM 2006 in southern Morocco, close to the source of mineral dust [Ansmann et al., 2008].

image

Figure 10. Comparison of the temperature dependence of occurrence of ice-containing cloud over the tropical lidar site at Cape Verde (CV, thick solid, thick dashed) [Ansmann et al., 2009] and the central European lidar site (LE, thin solid). All clouds observed over Cape Verde and over Leipzig (in dust-free and dust-containing air masses, from 3 to 8.5 km height, see Figure 3) are considered. Cloud seeding effects are corrected for (dashed curves) and neglected (solid curves) in the case of the tropical data. Error bars are computed after equation (1).

Download figure to PowerPoint

[42] What causes the surprisingly large difference in the observations in Europe and Africa? A simple explanation is not available. The discussion is complicated because of the multitude of effects that may be active and probably even linked together. The presence of high dust concentrations alone cannot fully explain the enhanced nucleation at the midlatitude site. As mentioned, even the curve in Figure 9, which considers the dust-free cases only, shows this enhancement. Over the European continent, there seem to be other IN (in addition to mineral dust particles) which are active at higher temperatures. From laboratory studies we know that biological particles (bacteria, pollen, decayed organic material) have been identified as favorable IN which may be active already in the temperature range from −2 to −10°C [Szyrmer and Zawadzki, 1997; Diehl et al., 2002; Möhler et al., 2007]. There are many other aerosol constituents (e.g., lead, metals, salts) over polluted, industrialized continents that show favorable IN properties [Szyrmer and Zawadzki, 1997; Stith et al., 2009; Cziczo et al., 2009b; Pratt et al., 2009; Kamphus et al., 2009]. Air masses in the free troposphere over central Europe, in which the analyzed clouds formed, are mainly advected from the Atlantic Ocean (from southwesterly to northwesterly directions), and the free tropospheric aerosol may frequently contain a complex mixture of traces of maritime, biogenic, and anthropogenic particles, mineral dust, and aged forest fire smoke from Canada or Siberia [Mattis et al., 2008].

[43] Aerosol outbreaks from Africa, transported from east to west in the lowermost 5 km of the atmosphere, predominantly contain Saharan dust and, in the winter season, biomass burning smoke in addition. Most of the clouds considered in Figure 10 occurred however above 5 km height. Aerosols in the free troposphere above 5 km height (transported from west to east) mostly originate from remote, less polluted tropical maritime regions (Caribbean, northern South America, Pacific). However, they may contain traces of aged, cloud-processed African smoke and dust which are injected into the free troposphere by deep convection around the Cape Verde Islands [Dunion and Velden, 2004; Ansmann et al., 2009; Wiacek and Peter, 2009]. The differences in the sources of IN and correspondingly in the IN characteristics over central Europe and western Africa may consequently best explain the different onset temperatures found at which significant ice formation starts.

[44] There are many other influences that may have contributed to the observed contrasts. Chemical processing (e.g., leading to a coating of dust with hygroscopic material), which may reduce the ability of dust particles to serve as ice nuclei [Wurzler et al., 2000; Möhler et al., 2008; Cziczo et al., 2009a], must be taken into account. Chemical aging (coating) during cloud processes as well as removal of large dust particles (most favorable IN) by washout may have a stronger impact on the IN concentration in the tropics than in midlatitudes. Most of the long-range transports toward central Europe occur during high-pressure situations during which cloud development (and thus dust contamination and removal) is widely suppressed [Ansmann et al., 2003]. On the other hand, cloud processes can also lead to preactivation of dust particles [Knopf and Koop, 2006; Vali, 2008] that leads to increased ice nucleation efficiency. Thus simple conclusions on the role of cloud processing cannot be drawn here.

[45] Another complicating issue is ice enhancement by the riming-splintering mechanism [Hallet and Mossop, 1974], leading to an increase in ice particle concentration at temperatures from about -5 to −10°C, and other secondary ice production mechanisms discussed by Cantrell and Heymsfield [2005]. These effects may influence ice production over Europe and western Africa in different ways.

[46] Differences in the meteorological and orographic conditions may also have contributed to the found differences in the cloud processes and observed characteristics in the tropics and at the midlatitude site. Large-scale lifting in front of warm fronts and complex wave activity [Ansmann et al., 2005], partly caused by the pronounced European orography upwind of Leipzig during southwesterly airflows, may often lead to vertical wind characteristics in the free troposphere over central Europe that differ significantly from those over Cape Verde. Strong vertical winds provide favorable conditions for heterogeneous ice formation. Frontal activity is absent in the tropics and orographic influences are expected to be less relevant at islands in the tropical North Atlantic although the small islands have mountains with heights of 1.5–3 km.

[47] At the end it is noteworthy to mention that, according to Hobbs and Rangno [1985], the onset of the occurrence of ice particles in stratiform clouds appear to coincide with the development of a broad particle size distribution with a relatively large amount of large drops. Mature and aging cumuliform clouds provide best conditions to develop broad drop size spectra. Hobbs and Rangno [1985] further reported that the altocumulus layers that produced no ice were found to be generally thin, stratiform and often newly formed. Those clouds that contained ice particles were often thicker and persistent. Contact nucleation is more likely in thick, long-lived, clouds after Hobbs and Rangno [1985]. The results of Hobbs and Rangno [1985] are in good agreement with many other midlatitude cloud studies [Gultepe et al., 2001; Fleishauer et al., 2002; Field et al., 2004; Korolev et al., 2003] and also with our observations.

[48] Most clouds observed with lidar over Leipzig are midtropospheric stratiform clouds and are, on average, geometrically and optically thicker than the ones over Cape Verde. During the SAMUM 2008 campaign most altocumulus clouds were short-lived and optically thin. Even persistent altocumulus layers were often remarkably thin over hours [Ansmann et al., 2009]. In these cases, the development of a broad liquid drop spectrum seems to be unlikely so that heterogeneous ice formation is suppressed after Hobbs and Rangno [1985].

6. Summary and Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[49] A statistical analysis of an 11 year lidar-derived cloud data set observed at a central European EARLINET site from February 1997 to June 2008 was analyzed with focus on the specific role of aged dust in processes of heterogeneous ice formation. The lidar data (cloud top and depth, cloud phase) and DREAM results (dust mass concentration) were combined for the study. Compared to air masses that were advected from dust-free regions of the North Atlantic, a significantly higher amount of ice-containing clouds (25%–30% more) was observed for cloud top temperatures from −10 to −20°C in air masses that contained Saharan dust. The comparison with our lidar study of tropical stratiform clouds at Cape Verde during the Saharan Mineral Dust Experiment (SAMUM) 2008 reveals that ice formation is considerably enhanced and starts at higher temperatures at the midlatitudinal site than over the tropical station. Both aerosol-related as well as meteorological aspects may have contributed to the strong contrast in the height distribution of ice-containing clouds over central Europe and tropical western Africa.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

[50] The DRIFT study was funded by the Deutsche Forschungsgemeinschaft (DFG) under grant AN 258/10. We thank the reviewers for their fruitful suggestions.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Instrumentation and Data Analysis
  5. 3. Identification of Ice-Containing Clouds
  6. 4. Results
  7. 5. Midlatitudinal Versus Tropical Observations
  8. 6. Summary and Conclusions
  9. Acknowledgments
  10. References
  11. Supporting Information
FilenameFormatSizeDescription
jgrd16108-sup-0001-t01.txtplain text document1KTab-delimited Table 1.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.