Vertical profiles of size-resolved particle concentrations were measured over the Tibetan Plateau by using balloon-borne optical particle counters. The measurement on 17 August 1999 showed that there was a thin layer of particles with diameters of >3.6 μm at concentrations of >0.01 cm−3 around 150 hPa. These particles are much larger than the background aerosols and most likely composed of ice crystals. Within the layer, a large number of submicron aerosols still coexisted with the ice particles, indicating that not all of the background aerosols were removed by ice nucleation and scavenging. Considering recent model and laboratory studies, these results suggest an occurrence of selective ice nucleation involving a fraction of the background aerosols (i.e., effective ice nuclei), which is associated with dynamical and constituent fields in the upper troposphere over the Tibetan Plateau.
 The indirect effects of aerosol particles on global climate are a subject of current scientific debate due in part to an insufficient understanding of cirrus cloud formation mechanisms [Lohmann and Feichter, 2005, and references therein]. As an example, homogeneous ice nucleation in supercooled solution droplets is believed to be a key process for cirrus cloud formation. This nucleation process depends closely upon the water activity of supercooled liquid solutions and occurs at 140–170% relative humidity with respect to ice (RHice) in the upper troposphere/lower stratosphere [Koop et al., 2000; Murphy and Koop, 2005]. In addition, if available ice nuclei at lower ice supersaturations exist, heterogeneous ice nucleation is expected to play an important role in cirrus cloud formation. This nucleation process has the potential to form cirrus clouds with small number concentrations of ice crystals (i.e., subvisible or thin cirrus clouds) compared to the case where only supercooled solution droplets exist [Jensen and Toon, 1997; Gierens, 2003].
 A number of satellite observations have shown that cirrus clouds associated with deep convection frequently occur over the Tibetan Plateau [Li et al., 2005; Fu et al., 2006; Jin, 2006]. Recent chemical-transport model studies have indicated that convectively lifted aerosols from South Asia are widely distributed in the upper troposphere during the Asian summer monsoon period [Li et al., 2005]. Since deep convection is regarded as a major pathway for upward material transport from the troposphere into the stratosphere [Gettelman et al., 2004; Fu et al., 2006], the indirect effects of aerosols on cirrus clouds are of particular interest. Multiple ice nucleation processes (i.e., both homogeneous and heterogeneous freezing) have the potential to occur under such convective conditions. However, knowledge of the size, phase, shape, concentration, and formation mechanisms of particles in cirrus clouds over the Tibetan Plateau is very limited, because few measurements on this subject have been conducted in the real atmosphere.
 Measurements of aerosol/cloud particles and meteorological parameters by balloon-borne instruments were carried out in August-October 1999 at Lhasa (29.7°N, 91.1°E, 3650 m a.s.l.), located in the southern part of the Tibetan Plateau. One of the balloon flights encountered a thin cloud layer in which homogeneous ice nucleation could not be expected. In this paper, we present the particle number-size distributions within the cloud layer, and discuss the mechanisms potentially responsible for the continental cloud formation.
2. Particle Measurements
 Vertical profiles of size-resolved particle concentrations were obtained using balloon-borne optical particle counters (OPCs). The OPCs measure particle number concentrations by detecting forward scattering of laser beams at 780 or 810 nm at a sample flow rate of 3 L min−1. The instruments count particles by equivalent optical diameters (Dp). The available size ranges are Dp > 0.3, 0.5, 0.8, 1.2, and 3.6 μm. The counting uncertainties for the concentrations of 0.1, 0.01, and 0.001 cm−3 are about ±10, ±32, and ±100%, respectively. The detailed characteristics of the OPCs had been described by Hayashi et al. . In addition to the particle number-size distributions, ambient temperatures and pressures were measured by on-board Vaisala radiosondes.
 The balloon-borne observations at Lhasa were carried out on 17 August, 19 September, and 11 October 1999. The vertical profiles of each parameter were measured during the balloon ascent with a vertical resolution of about 100–110 m (every 20-second) from ground (3.65 km or ∼650 hPa) to an altitude of ∼35 km. The observations were made in the daytime while avoiding local convective clouds or rainfalls. An overview of the observations has been compiled by Tobo et al. .
Figure 1 shows the vertical profiles of number concentrations of particles with Dp > 3.6 μm, together with those of ambient temperatures, on 17 August, 19 September, and 11 October 1999. The 17 August flight showed the existence of some distinct layers between about 300 and 130 hPa, in contrast to the other two flights. In particular, relatively high concentrations (>0.01 cm−3) of the particles were observed at ∼150 hPa. Such heights are typical of the top of cirrus clouds over the Tibetan Plateau during the Asian summer monsoon period [Li et al., 2005; Fu et al., 2006].
Figure 2a shows the vertical profiles of the total concentrations of particles with Dp > 0.3 μm. The total particle concentrations were closely dependent on the concentrations of aerosols between 0.3 and 1.2 μm (Figure 2b). The concentrations were a little lower around 170 hPa and higher at <130 hPa. In the cloud layer at ∼150 hPa, the concentrations of submicron aerosols were relatively low; however, there were still a large amount of submicron aerosols compared to particles with Dp > 3.6 μm.
 Previous aircraft measurements in continental cumulonimbus anvils (about 12–13 km) over the Rocky Mountains also detected the particles with Dp > 3 μm, which coexisted with a large amount of submicron aerosols [Knollenberg et al., 1993]. The continental clouds detected by them were similar in temperature range as well as in number-size distribution to those over the Tibetan Plateau presented in this study. As a contrastive case, Knollenberg et al.  reported very high number concentrations (about 10–100 cm−3) of the particles with Dp > 3 μm in tropical cumulonimbus anvils, which they attributed to homogeneous freezing of sulfate droplets.
4.1. Properties of Larger Particles in the Upper Troposphere
 In this section, we will discuss the particles with Dp > 3.6 μm measured in the upper troposphere over the Tibetan Plateau during the 17 August flight. The Moderate Resolution Imaging Spectroradiometer (MODIS) data have shown that the mean top pressure of cirrus clouds over the Tibetan Plateau is ∼150 hPa [Fu et al., 2006], suggesting that the condition around this altitude is favorable for ice nucleation. The satellite data indicated that ice crystals in the clouds were in the size range of about 10–90 μm in diameter. Unfortunately, the OPCs used in this study cannot display information on size-resolved particle concentrations in the size range of ice crystals because the largest integration limit is set to 3.6 μm. However, it is expected that background aerosols in the upper troposphere rarely contained particles with Dp > 3.6 μm because these larger particles were rarely detected at <400 hPa (Figure 1), with the exception of 17 August 1999. This means that the background aerosols are considerably smaller than 3.6 μm. In fact, it is well known that the background aerosols are usually submicron aerosols, consisting of solutions (mixtures of H2SO4, HNO3, NH3, H2O, or organics), followed by carbon, mineral dust, fly ash, etc. [e.g., Murphy et al., 1998]. For these reasons, we consider that the particles above the 3.6-μm detection threshold were predominantly ice crystals.
 To further investigate the possibility of ice formation around the cloud layer, we referred to the RHice values estimated from the radiosonde temperature and the specific humidity from the European Centre for Medium-range Weather Forecasts (ECMWF) data, which has a temporal resolution of 6 hours, horizontal resolution of 2.5° and vertical resolution of 23 pressure levels from 1000 to 1 hPa (not shown here). According to this result, relatively high water vapor and supersaturated or subsaturated conditions with respect to ice were reproduced from about 250 to 150 hPa over the Tibetan Plateau. However, the ECMWF meteorological fields are quite problematic and in-situ measurements of water vapor are needed in the future work.
4.2. Processes Responsible for Cloud Formation
Figure 2b shows the number fractions of submicron aerosols and particles with Dp > 3.6 μm (i.e., ice crystals)' to the total number concentrations of particles with Dp > 0.3 μm. The observed total particles were dominated by submicron aerosols and the percentage of ice crystals was less than that of submicron aerosols even at ∼150 hPa. This implies that not all particles in the cloud layer took part in ice nucleation, and most of the background aerosols were not removed by ice nucleation and scavenging.
 The selectivity of the ice nucleation process may depend closely on aerosol size, because larger droplets freeze prior to smaller ones [Koop et al., 2000]. Here we estimated the relative humidity required for homogeneous freezing of supercooled solution droplets between 0.3 and 1.2 μm in diameter by using the parameterization given by Koop et al. . As shown in Figure 3, around 150 hPa or 210 K, the relative humidity required for homogeneous freezing is about 150–160% RHice, and make little difference in the size of the droplets. Also, given that the relative humidity is much lower than the liquid water saturation, submicron solution droplets in the background aerosols probably do not show exponential growth via adsorption as well as ice nucleation until the critical RHice is reached [e.g., Möhler et al., 2003]. For this reason, we consider that ice nucleation processes other than homogeneous ice nucleation (i.e., heterogeneous ice nucleation) are required to explain the coexistence of submicron aerosols and ice crystals in cirrus clouds.
 The particles with Dp > 3.6 μm in the continental cirrus clouds observed over the Tibetan Plateau may reflect natural and anthropogenic inputs of ice nuclei into the upper troposphere via deep convection, orographic lifting or advection. During the observing period, a synoptic scale monsoon anticyclone (i.e., Tibetan anticyclone) dominated the large-scale circulation in the upper troposphere over the Tibetan Plateau [Tobo et al., 2007]. The temperature profiles from the 17 August flight indicated a warm core up to ∼150 hPa compared to those from the other two flights (Figure 1) and the zonal mean temperature profiles from the ECMWF data (not shown here). It is well known that the Tibetan anticyclone occurs primarily as a response to diabatic heating associated with deep convection over the Tibetan Plateau [Randel and Park, 2006]. Recent model studies by Li et al.  suggested that anthropogenic aerosols, such as sulfate, black carbon, and organic carbon, emitted from South Asia extend to the upper troposphere and then are entrained by the Tibetan anticyclone for several days in deep convective systems. These points raise the possibility of the widespread existence of continental aerosols in the upper troposphere.
 In addition, we speculate that a temperature-dependence of effective ice nuclei is kind of responsible for the cirrus formation. For example, Rangno and Hobbs  indicated extremely low concentrations of effective ice nuclei at warmer temperatures (warmer than 254 K), which are several orders of magnitude less than the concentrations of the particles with Dp > 3.6 μm at cirrus temperatures (∼210 K) presented in this study. On the other hand, the concentrations of the cloud particles with Dp > 3.6 μm are similar to the concentrations (∼0.01 cm−3) of effective ice nuclei at lower temperatures (less than 235 K), which have been measured in the free troposphere over the Rocky Mountains [DeMott et al., 2003a].
 As illustrated in Figure 3, recent laboratory experiments with the Aerosol Interaction and Dynamics in the Atmosphere (AIDA) cloud chamber have suggested that deposition nucleation (the formation of ice from a supersaturated gas environment directly on a solid surface) or immersion freezing (the formation of ice at the surface of a nucleus suspended in a supercooled liquid) involving mineral dust [Möhler et al., 2006], soot [Möhler et al., 2005] or solid-phased soluble species (solid ammonium sulfate) [Abbatt et al., 2006] is efficient at temperatures of about 210–220 K, corresponding approximately to the altitude ranges from 150 to 200 hPa over Lhasa. It seems that the critical RHice required for heterogeneous ice nucleation on these continental aerosols is approximately consistent with that required for orographic cirrus formation indicated by Heymsfield and Miloshevich . The temperature-dependent changes in the ice nucleation threshold are still largely uncertain, but suggest that these aerosols act as efficient ice nuclei at cirrus temperatures.
 The balloon-borne measurement on 17 August 1999 showed that there was a thin layer of particles with Dp > 3.6 μm at concentrations of >0.01 cm−3 around 150 hPa over the Tibetan Plateau. These particles are much larger than the background aerosols and most likely composed of ice crystals. Within the layer, a large number of submicron aerosols coexisted with the ice, indicating that not all of the background aerosols were removed by ice nucleation and scavenging. While processes for the continental cirrus formation are still speculative, we have suggested that selective ice nucleation on heterogeneous ice nuclei, which have been studied in recent laboratory experiments [e.g., Möhler et al., 2005, 2006; Abbatt et al., 2006], might be more efficient at cirrus temperatures. These conclusions must be evaluated carefully with more comprehensive studies to determine the key process responsible for ice nucleation in the real atmosphere over the Tibetan Plateau. In particular, information on relative humidity as well as ice nucleus composition and concentration will become a critical parameter that needs to be measured in the future studies.
 We thank D. J. SuoLang and the staff members of the Meteorological Bureau of Tibet Autonomous Region for their technical supports in the fieldwork. We also thank K. Tamura, T. Ohashi, M. Nagatani, and H. Nakata for their efforts in assembling the sonde data. This research was funded by the Japan Ministry of Education, Culture, Sports, Science and Technology (PI: Y. Iwasaka, 10144104 and 10041115), National Natural Science Foundation of China (PI: G.-Y. Shi, 49775275), and 21st-Century COE program of Kanazawa University (PI: K. Hayakawa).