Heavy aerosol loads have been observed to suppress warm rain by reducing cloud drop size and slowing drop coalescence. The ice forming nuclei (IFN) activity of the same aerosols glaciate the clouds and create ice precipitation instead of the suppressed warm rain. Satellite observations show that desert dust and heavy air pollution over East Asia have similar ability to glaciate the tops of growing convective clouds at glaciation temperature of Tg < ∼ −20°C, whereas similarly heavy smoke from forest fires in Siberia without dust or industrial pollution glaciated clouds at Tg ≤ −33°C. The observation that both smoke and air pollution have same effect on reducing cloud drop size implies that the difference in Tg is due to the IFN activity. This dependence of Tg on aerosol types appears only for clouds with re-5 < 12 μm (re-5 is the cloud drop effective radius at the −5°C isotherm, above which ice rarely forms in cloud tops). For the rest of the clouds the glaciation temperature increases strongly with re-5 with little relation to the aerosol types, reaching Tg> ∼ −15°C for the largest re-5, which are typical to marine clouds in pristine atmosphere.
 The heavy air pollution, known as the Atmospheric Brown Cloud (ABC), has been researched mainly for its radiative effects and cloud condensation nuclei (CCN) activity that act to nucleate larger number of smaller cloud drops, and hence suppress coalescence and warm rain-forming processes [Ramanathan et al., 2001]. Precipitation in convective clouds with aerosol-induced suppressed coalescence is initiated by nucleation of ice crystals that form ice hydrometeors that collect the supercooled cloud water. These ice particles are nucleated by ice forming nuclei (IFN) aerosols, which can also serve as CCN that nucleate cloud drops before freezing them. Greater IFN activity of the ABC aerosols means formation of more ice hydrometeors at higher temperatures lower in the clouds, potentially compensating to some extent for the suppression of precipitation by the CCN activity of the ABC.
 Little is known about the IFN activity of the ABC. Documentation of the IFN activity of the ABC is challenging because of the large diversity of sources for the aerosols composing the ABC, which include industrial emissions, burning of low grade fuels and agricultural fires. Air pollution aerosols have been generally considered as having less IFN activity than desert dust aerosols, which are known to be major source of IFN. A review paper [Szyrmer and Zawadzki, 1997, pg. 212] stated that “certainly, natural mineral particles dominate the atmospheric nucleation of ice in the conditions of low temperatures of −12° to −15°C or less.”
 For testing whether this is really the case we focus our study over China and the vicinity, where both pollution and desert dust aerosols abound. Comparative measurements of IFN activity of heavy pollution haze and dust storms were conducted previously in China using the Bigg cloud chamber [Bigg, 1957]. The observations in Beijing [You and Shi, 1964] showed that average IFN concentrations in pollution haze exceed by up to a factor of two the average IFN in dust, especially at the higher temperatures (Figure 1). The same observational procedure and method were used in the same season and location in 1995. The IFN for heaviest dust and air pollution with visibilities ∼1km in 1995 showed much greater overall concentrations, but air pollution and heavy desert dust had still comparable IFN activity [You et al., 2002], with desert dust exceeding the pollution activity by a factor of 2 at −15°C. The reported concentrations of IFN in China increased by a factor of 5 to 10 between 1980 and 2000 [Yin et al., 2011]. This cannot be explained by natural causes, leaving the increasing levels of air pollution as the main suspected cause.
 These measurements contradict the common belief that desert dust dominates the IFN activity of aerosols whereas air pollution, at least in Beijing, mainly act as CCN, except for high supercooling, e.g., at T < −30°C. This common belief is re-examined here using satellite measurements of cloud and aerosol properties over East Asia.
 Glaciation temperature (Tg) of growing convective cloud towers was obtained by analyzing the vertical evolution of cloud top particle effective radius (re) as a function of cloud top temperature (T), using the methodology developed by Rosenfeld and Lensky  and further refined by Yuan et al. . This method examines the statistical relation between the satellite-retrieved pixel values of T and re at the tops of a cloud ensemble in an area that contains the growing clouds of a convective cluster. The 30th percentile of re is taken for all the 1-km pixels having T of 1°C interval. The ensemble approach is equivalent to tracking the evolution of cloud top microstructure as it grows to lower T [Lensky and Rosenfeld, 2006]. The re increases with height (i.e., with decreasing T), until full glaciation is reached. At greater heights and lower T re remains constant or decreases.
 The methodology was applied to the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the AQUA satellite. We selected 140 cases with convective cloud clusters for which coinciding measurements of aerosol classification were available from the Cloud-Aerosol Lidar with Orthogonal Polarisation (CALIOP) space borne instrument [Omar et al., 2009], and for which MODIS measurements of aerosol optical thickness (AOT) were available. Seven cases with visibly obvious heavy dust storms that did not coincide with the CALIOP were added, for better representation of such situation.
 The aerosol types investigated were obtained by the CALIOP classification [Omar et al., 2009], and consisted of Pristine maritime, Desert dust, Polluted dust and Pollution. The latter contains the CALIOP classifications of Polluted continental, clean continental and Smoke. The CALIOP distinction between Desert dust, Polluted dust and Polluted continental is based mainly on the lidar depolarization ratio. Polluted dust can be misclassified as Polluted continental if heavily coated with pollution. Another type of pollution-free smoke from forest fires in remote areas of Siberia was added, defined as visibly thick smoke emanating from fires seen as “hot spots” in the satellite imagery. The MODIS “Deep Blue” AOT was used for the dust and polluted-dust aerosol types, and the 550 nm AOT was used for the other aerosol types. Back trajectory analysis of the air masses between 0 and 1000 m above ground was done using HYSPLIT for identifying cases that have air mass in source regions of desert dust during the past 96 hours. This was used for identifying the Polluted continental cases that might contain some desert dust. A similar analysis was done for identifying marine cases that might contain some aerosols emitted form land or island areas. The back trajectory analysis did not show any apparent difference between the cases that had suspected other sources and the pure cases (see Figures 4b and 5), giving more credibility for the CALIOP aerosol classification.
 It was shown that much more ice forms in clouds when the largest drops exceed radius of 12 μm already at modest super cooling of −7°C without an apparent relation to the IFN concentrations [Hobbs and Rangno, 1985]. This was ascribed mostly to the ice multiplication mechanism shown operative in the laboratory in clouds containing drops >12 μm in radius and at temperatures between −3 and −8°C [Hallett and Mossop, 1974; Mossop, 1976]. Another mechanism for enhancing glaciation of large cloud drops is their larger chance to contain or contact IFN [Pruppacher and Klett, 2004]. These mechanisms do not explain fully the observed fast glaciation in deep convective clouds with large drops, which is typical for tropical maritime conditions. There are still wide gaps in our understanding of the processes that glaciate clouds. According to our present knowledge, cloud drop size appears to dominate Tg at re-5 > 12 μm, where re-5 is the cloud drop effective radius at the −5°C isotherm.
 In order to minimize the role of cloud drop size in masking the role of aerosol composition in glaciating the clouds, the data was divided based on re-5, in accordance with the conditions for onset of ice multiplication [Hallett and Mossop, 1974]. Thirty eight cases had re-5 ≥ 12 μm. The remaining 109 cases were defined as re-5 < 12 μm. They include 15 cases with highest observed cloud base temperature <−5°C. The geographic distribution and the types of cases are given in Figure 2.
 The IFN activities of the different aerosol types were compared using cumulative probability functions of Tg, shown in Figure 3. According to Figure 3a, marine clouds glaciate at the highest T (median of −16°C), followed by polluted and/or dusty continental clouds (−22° to −25°C), and pure smoke from forest fire (−35°C). When excluding the clouds with drops that are large enough to make ice multiplication efficient (Figure 3b), all the marine clouds are excluded, but most of the continental cases remain, as shown in Figure 2.
 The large difference in Tg between the maritime and continental aerosols highlights the importance of cloud drop size and its relevance to the determination of Tg. This was looked into by relating Tg to re-5, which captures the cloud drop size just below the isotherm where cloud drops can start glaciating. Indeed Tg is correlated positively with re-5. Furthermore, the range of Tg narrows for larger re-5 (Figure 4a). The clouds with the largest drops always glaciate at Tg > −20°C, whereas clouds with the smallest drops can glaciate at any temperature between −10° and −38°C. This means that there is much more room for IFN to determine Tg in clouds with smaller drops, whereas the IFN activity in clouds with large drops is amplified greatly by secondary ice forming processes, causing Tg to increase with re. Observations of clouds with extremely low Tg = −37.5°C [Rosenfeld and Woodley, 2000] and the simulation of the sensitivity of their Tg to CCN concentrations [Khain et al., 2001] show that high CCN concentrations are necessary for allowing the very low Tg. The CCN nucleate larger concentrations of smaller drops that have reduced probability of freezing because they have less chance to contain or contact IFN [Pruppacher and Klett, 2004] and do not support secondary ice forming processes [Mossop, 1976]. Therefore, occurrence of high Tg in clouds with small drops can be explained by large concentrations of IFN that overcome the tendency of these small drops to freeze at low temperatures.
 This conclusion is put to test in Figure 4b, which is the same as Figure 4a, but with distinction between the aerosol types. At the small end of re-5 there is a wide range of the Tg for desert dust, air pollution and polluted dust. The smoke from Siberian forest fires has distinctly very low Tg ≤ −33°C. The high AOT of the Siberia smoke (Figure 5a). The Tg of the Siberian smoke was statistically significantly (using Wilcoxon test) colder than the Polluted continental. This indicates that its low Tg is due to the combination of very high concentration of small droplets with lack of high concentrations of IFN. This means that the smoke particles serve as good CCN and as poor IFN at T > −33°C. This is in agreement with laboratory experiments for black carbon IFN activity [Möhler et al., 2005]. Ash from pyro-Cb was observed to produce ice at −15°C [Sassen and Khvorostyanov, 2008], but it was in layered flying ash that stayed for long time at the −15°C isotherm. This underlines the fact that the calculated Tg in this study pertains to the tops of growing convective clouds.
 In searching for the cause of this variability, it is expected that IFN concentrations would increase and induce glaciation at higher temperatures with greater values of AOT. On the other hand, it is also expected that higher AOT would be associated with greater CCN concentrations [Andreae, 2009] that reduce cloud drop size and hence lower Tg. In attempt to disentangle these opposite effects, the dependence of Tg on AOT is shown in Figure 5 separately for smaller and larger drops with respect to re-5 = 12 μm. In agreement with these considerations, Tg is observed to increase with AOT for clouds with small drops (Figure 5a) and decreases with AOT for the clouds with large drops (Figure 5b). The significance level of the slope for all the points combined in Figure 5b is P = 0.07.
 Remarkably, the air pollution and desert dust over East Asia have similar Tg, often >−20°C for the heaviest aerosol loads. Is high Tg caused by the black carbon (BC) or other constituents of the air pollution? This question is addressed by comparing the polluted clouds to clouds that ingest pure smoke from forest fires in eastern Siberia, in an air mass that is mostly free from other pollution sources. The smoke had very low Tg ≤ −33°C, whereas polluted clouds with similarly high AOT and small cloud drops glaciated at much higher temperatures. Smoke particles from flaming forest fires that are rich in BC would nucleate the cloud drops and then would freeze them at the BC ice nucleation activation temperature. The abundance of BC in the cloud drops means that it does not act as IFN at T > −33°C. This implies that BC is not likely to contribute substantially to the high IFN activity in the polluted clouds. The observation that polluted dust or pure dust does not have higher Tg than pollution alone means that incorporation of some desert dust in the pollution cannot explain its high Tg, but rather other unknown constituents, probably of industrial origin, as was shown by Ebert et al. . Furthermore, the polluted dust glaciates clouds at somewhat lower temperatures than dust alone, in agreement with the hypothesis that air pollution deactivates IFN activity of mineral dust [Möhler et al., 2005; Braham and Spyers-Duran, 1974; DeMott et al., 1999]. According to Figure 5a the pure dust has narrower range of Tg values compared with pollution or polluted dust, probably because their composition is more variable than the composition of the pure dust.
 For clouds with large drops, where re-5 > 12 μm (Figure 5b), the Tg decreases with increasing AOT, reflecting the dominant role of the aerosols impact on cloud drop size over their IFN activity. Evidently, making a cloud composed of larger drops is a more effective way to glaciate it than adding high concentrations of IFN to it. The causes for that are not yet fully understood.
 Until now, the smoke and pollution aerosol effects on precipitation have been recognized mainly through their impact on cloud drop size and suppression of drop coalescence. Here we show that, in heavy aerosol loads, the suppression of warm rain processes allows the IFN activity to become dominant in the precipitation forming processes. Furthermore, we show that the industrial air pollution over East Asia has comparable IFN activity to that of desert dust there, in agreement with previous IFN measurements in China. This means that the IFN activity of air pollution might restore at least some of the precipitation that is suppressed by the CCN activity of the same aerosols in deep convective clouds. Adding IFN for restoring suppressed precipitation is in fact the objective of glaciogenic cloud seeding for rain enhancement. This means that glaciogenic seeding is superfluous in an atmosphere that contains large amounts of industrial pollution and/or desert dust aerosols, as was already previously suspected [Rosenfeld and Farbstein, 1992]. On the other hand, clouds ingesting pristine air with relatively small aerosol concentrations develop large drops that allow efficient warm rain processes and high Tg even without large IFN concentrations. This leaves only lightly polluted clouds as having the potential for very low Tg, where glaciogenic seeding has the potential to increase Tg and possibly enhance precipitation or suppress hail.
 More generally, the results show that IFN have a major role in potential buffering of the rain suppression effects of aerosols. This study provides unique measurements that are necessary for quantifying the IFN effects of various aerosol types in glaciating convective clouds at different cloud drop sizes. This kind of quantification can be used in parameterizing these aerosol effects in cloud and climate models.
 This research was supported by National Natural Science Foundation of China (grant 40975087). We are grateful to the supports from Shaanxi Administration of Foreign Expert Affairs. The authors thank A. Khain for discussing the interpretation of the results.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.