Aerosol concentrations determine the height of warm rain and ice initiation in convective clouds over the Amazon basin

We have investigated how pollution aerosols affect the height above cloud base of rain and ice hydrometeor initiation and the subsequent vertical evolution of cloud droplet size and number concentrations in growing convective cumulus. For this purpose we used in-situ data of hydrometeor size distributions measured with instruments mounted on HALO (High Altitude and Long Range Research Aircraft) during the ACRIDICON-CHUVA campaign over the Amazon during September 2014. The results show that the height of rain initiation by collision and coalescence processes ( D r , in units of meters above cloud base) is linearly correlated with the number concentration of droplets ( N d in cm −3 ) nucleated at cloud base ( D r  ≈ 5  N d ). When N d exceeded values of about 1000 cm −3 , D r became greater than 5000 m, and particles of precipitation size were initiated as ice hydrometeors. Therefore, no liquid water raindrops were observed within growing convective cumulus during polluted conditions. Furthermore, also the formation of ice particles took place at higher altitudes in the clouds in polluted conditions, because the resulting smaller cloud droplets froze at colder temperatures compared to the larger drops in the unpolluted cases. The measured vertical profiles of droplet effective radius ( r e ) were close to those estimated by assuming adiabatic conditions ( r ea ), supporting the hypothesis that the entrainment and mixing of air into convective clouds is almost completely inhomogeneous. Secondary nucleation of droplets on aerosol particles from biomass burning and air pollution reduced re below r ea , which further inhibited the formation of raindrops and ice particles and resulted in even higher altitudes for rain and ice initiation.


Introduction
Understanding cloud and precipitation forming processes and their impacts on the global energy budget and water cycle is crucial for meteorological modeling. Therefore, many studies have focused on improving cloud parameterization in numerical weather and climate models (e.g., Frey et al., 2011;Khain et al., 2005Khain et al., , 2000Klein et al., 2009;Lee et al., 50 2007;Machado et al., 2014). These parameterizations need to represent in simplified form the complex chain of events that occur in clouds.
Cloud droplets form when humid air rises and becomes supersaturated with respect to liquid water. Then water vapor condenses onto surfaces provided by pre-existing cloud condensation nuclei (CCN, a list of abbreviations and symbols is given in Table 1) aerosols. For ice formation, the ambient temperatures must reach values lower than 0 °C. At 55 temperatures between 0 ºC and -36 ºC, ice in convective clouds mostly forms inhomogeneously on ice nuclei (IN) aerosols, often when they interact with supercooled liquid water droplets (Pruppacher et al., 1998). At colder temperatures (less than -36 ºC), cloud particles freeze due to homogeneous ice nucleation (Rosenfeld and Woodley, 2000).
A cloud predominantly consists of droplets with diameters larger than about 3 μm, except for transient smaller sizes 60 right at cloud base. The number concentration of cloud droplets (N d in cm -3 ) at cloud base mainly depends on the conditions below cloud base, i.e., the updraft wind speed (W) and the supersaturation (S) activation spectra of cloud condensation nuclei [CCN(S)] (Twomey, 1959). In very clean conditions, values of N d near cloud base are in the range of ~50-100 cm -3 , while in polluted condition N d may reach values between 1000-2000 cm -3 (Andreae, 2009;Rosenfeld et al., 2008Rosenfeld et al., , 2014a. The effects of aerosol particles on clouds and precipitation have been studied in different parts of the globe (e.g., Fan et al., 2014;Li et al., 2011;Ramanathan et al., 2001;Rosenfeld and Woodley, 2000;Rosenfeld et al., 2014a;Tao et al., 2012;Voigt et al., 2016;Wendisch et al., 2016). A particularly interesting region is the Amazon basin, which presents contrasting environments of aerosol particle concentration between dry and wet seasons as well as steep aerosol concentration gradients within regions with near-constant thermodynamic conditions (Andreae et al., 2004;Artaxo et 75 al., 2013). The background number concentrations of aerosol particles and CCN over the pristine parts of the Amazon region are about a factor of 10 times lower than those of polluted continental regions, including polluted conditions over the Amazon (Martin et al., 2016). During the dry-to-wet transition season in the Amazon region, total aerosol number concentrations reach values up to 10,000 cm -3 , mostly due to forest fires (Andreae, 2009;Andreae et al., 2012;Artaxo et al., 2002). On the other hand, in the rainy season aerosol number concentrations are about 500-1000 cm -3 with CCN 80 concentrations on the order of 200-300 cm -3 for 1 % supersaturation, mainly consisting of forest biogenic aerosol particles (Artaxo, 2002;Martin et al., 2016;Pöhlker et al., 2016;Pöschl et al., 2010). Additionally, Manaus city, which is located at the central Amazon basin, releases significant concentrations of urban pollution aerosol particles (e.g., due to traffic, combustion-derived particles, or different types of industrial activities). This increases CCN concentrations by up to one order of magnitude (for 0.6% supersaturation) from the wet (Green Ocean) to the dry season (Kuhn et al., 85 2010). Rosenfeld et al. (2012b) showed that by estimating the adiabatic number of droplets nucleated at cloud base (N a ), the height above cloud base at which the first raindrops evolve can be parameterized. This approach is based on the assumption that the entrainment and mixing of air into convective clouds is almost completely inhomogeneous (Beals et al., 2015;Burnet and Brenguier, 2007;Freud et al., 2011;Paluch, 1979). This implies that the vertical profile of the 90 actual cloud droplet effective radius behaves nearly as in an idealized adiabatic cloud. This connects uniquely the adiabatic drop number concentration, which is approximated by N a at cloud base, with the adiabatic droplet effective radius (r ea ), based on an adiabatic parcel model for which droplet growth is dominated by condensation (Freud and Rosenfeld, 2012;Pinsky and Khain, 2002). This parameterization can be applied to estimate the height above cloud base at which raindrops start to form, when r ea reaches 13 μm (D 13 ) [Freud and Rosenfeld, 2012;Konwar et al., 2012; 95 2012b]. Braga et al. (2016) applied the methodology described by Freud and Rosenfeld (2011) to calculate N a at the base of growing convective cumulus clouds for the Amazon region during the ACRIDICON-CHUVA (Aerosol, Cloud, Precipitation, and Radiation Interactions and Dynamics of Convective Cloud Systems)-CHUVA (Cloud processes of tHe main precipitation systems in Brazil: A contribUtion to cloud resolVing modeling and to the GPM [Global Precipitation 100 Measurements]) campaign (Wendisch et al., 2016). The calculated N a based on the measured vertical profile of r e agreed well (within 20-30 %) with the actual measurements of cloud droplet number concentrations at cloud base. This approach provides the opportunity to test the agreement between estimated r ea and the height above cloud base of warm rain initiation (D r ) within clouds for the Amazon region. In addition, measurements of the height above cloud base of ice initiation (D i ) in convective clouds are also available from flights that include cloud penetrations at ambient temper-105 atures as low as -60 ºC with the High Altitude and Long Range Research Aircraft (HALO) aircraft (Wendisch et al., 2016).
This study analyzes the vertical development of cloud and precipitation particles (water drops and ice crystals) in growing convective cumulus over the Amazon, based on measurements of cloud microphysical properties from instruments mounted on HALO during ACRIDICON-CHUVA (Wendisch et al., 2016). The vertical profile of r ea is used 110 to estimate the depth above cloud base at which warm rain initiation occurs. The dominance of inhomogeneous mixing causes the r e profile to behave almost as in adiabatic clouds, constrained by N d at cloud base (Burnet and Brenguier, 2007;Freud et al., 2011). This means that the height above cloud base for reaching r e of 13-14 μm, which is required for rain initiation, is also determined by cloud base N d (Freud and Rosenfeld, 2012). Rain initiation depends strongly on r e because the rain production rate by collision and coalescence is proportional to ~r e 5 (Freud and Rosenfeld, 2012). Here 115 we test and quantify these relationships for the measurements conducted with HALO during ACRIDICON-CHUVA.
The HALO flights during the ACRIDICON-CHUVA campaign were performed over the Amazon region under various conditions of aerosol concentrations and land cover (Wendisch et al., 2016). Figure 1a shows the flight tracks during which cloud profile sampling in growing convective cumulus was performed. Figure 1b shows a schematic sketch of the flight pattern while sampling cloud clusters (the locations in three dimensions of each flight are available at Figure 1 120 on supplementary material). The aircraft obtained a composite vertical profile by penetrating young and rising convective elements, typically some 100-300 m below their tops.
Each DSD spectrum represents 1 s of flight path (covering ~150 m of horizontal distance for a typical aircraft speed).

125
The value of r e was calculated for each 1-s DSD. The two probes (CAS-DPOL and CCP-CDP) were mounted on opposite wings of HALO (horizontal distance of ~15 m). Similar values of N d and derived r e were measured by CAS-DPOL and CCP-CDP (they agree within 30 %), even though they were mounted on different wings. A previous study (Braga et al, 2016) showed that both probes were in agreement within the measurement uncertainties with respect to the measured cloud droplet number concentrations at cloud base and in accordance with the expected values for different 130 conditions of CCN concentration and updraft wind speed below cloud base. In addition, the cloud water content (CWC) calculated from the measured DSDs shows similar values to those measured with a hot wire device for different heights above cloud base [the probes' measurements agree within their uncertainty range (16% for probe DSDs and 30% for hot-wire device)] (Braga et al., 2016).
The determination of the height of rain initiation is based on the drizzle water content (DWC) calculation from the 135 CCP-CIP probe (Brenguier et al., 2013). The DWC is defined as the mass of the drops integrated over the diameter range of 75-250 µm (Freud and Rosenfeld, 2012). This size range includes only drops with terminal fall speed of 1 m s -1 or less, which maximizes the chance that the drizzle was formed in situ and did not fall a large distance from above.
Rainwater content (RWC) is defined as the CCP-CIP integrated liquid water mass of droplets with diameters between 250 and 960 μm. The CCP-CIP images were used to distinguish raindrops and ice particles during cloud passes. The 140 hydrometeor type is identified visually by their shapes. The phase of the smaller CCP-CIP particles cannot be distinguished. Therefore, the precipitation is considered as mixed phase when ice particles are identified, and the combined DWC and RWC are redefined as mixed phase water content (MPWC). Table 2 summarizes the calculated cloud microphysical properties with respect to the instrumentation used and its size ranges.

Cloud particle measurements
The instrumentation used to measure cloud particles and rain or ice formation consists of three cloud probes: CAS-DPOL, CCP-CDP and CCP-CIP (Brenguier et al., 2013). In this study, cloud particle counts are accumulated for bin diameters larger than 3 µm from the CCP-CDP and CAS-DPOL; the lower size bins from these probes overlap with 150 haze particles. Nucleated cloud drops in convective clouds grow quickly beyond 3 μm. Details about the cloud probe Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1155 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 February 2017 c Author(s) 2017. CC-BY 3.0 License. measurements characteristics are described in the following sub-sections and in Braga et al. (2016).

CCP-CDP and CCP-CIP measurements
The Cloud Combination Probe (CCP) combines two detectors, the Cloud Droplet Probe (CDP) and the greyscale Cloud 155 Imaging Probe (CIPgs). The CDP detects forward scattered laser light of cloud particles when penetrating the CDP detection area (Lance et al., 2010). The CIP records 2-D shadow cast images of cloud elements. In this study, we deduced the existence of ice from the occurrence of visually non-spherical shapes of the shadows. The particle detection size range is 2 μm to 960 μm when measuring with the CCP at 1 Hz frequency (Wendisch et al., 2016). The combination of CCP-CDP and CCP-CIP information provides the ability to measure cloud droplets and raindrops within 160 clouds for nearly the same air sample volume. The maximum number of particles measured by CCP-CDP and CCP-CIP are about 2,000 and 500 cm -3 for 1 Hz cloud pass, respectively.

CAS-DPOL measurements
The CAS-DPOL measures particle size distributions between 0.5 and 50 µm at 1Hz time resolution (Baumgardner et 165 al., 2011;Voigt et al., 2010;Voigt et al., 2011). Number concentrations are derived using the probe air speed measured at the instrument. Particle inter-arrival time analysis did not show influences of coincidence (Lance, 2012). The data analysis and uncertainties are described in detail in Braga et al. (2016). Braga et al. (2016) have shown sufficient agreement between both CAS-DPOL and CCP-CDP measurements of cloud droplet number concentration to distinguish convective clouds that develop above clean vs. polluted regions during the 170 ACRIDICON-CHUVA campaign. In addition, the CWC estimated by integration of the DSDs measured with both probes showed good agreement with hot wire CWC measurements (Braga et al., 2016).

Meteorological data
The HALO aircraft was equipped with a meteorological sensor system (BAsic HALO Measurement And Sensor System 175 -BAHAMAS) located at the nose of the aircraft (Wendisch et al., 2016). The uncertainties for measurements of temperature, relative humidity and vertical wind speed are 0.5 K, 5 % and 0.3 m s -1 , respectively (Mallaun et al., 2015).

Aerosol measurements
Aerosol particle measurements were performed using the Passive Cavity Aerosol Spectrometer Probe 100X (PCASP-180 100X), which is an airborne optical spectrometer that measures aerosol particles in the 0.1 μm to 3 μm diameter range (Liu et al., 1992). The maximum number of particles measured by PCASP is about 3,000 cm -3 for 1 Hz cloud pass.

Methods
The analyses are performed along the following general steps: b) The N a at cloud base is estimated through the vertical profile of r e .
c) The height of rain initiation based on the modelled adiabatic growth of r e with height is estimated for different 190 aerosol condition as a function of estimated N a . The value of D 13 is estimated as the cloud depth for which the Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1155 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 February 2017 c Author(s) 2017. CC-BY 3.0 License.
d) The extent of agreement between the directly measured D r within convective clouds and the estimated D 13 based on the assumption of adiabatic r e growth and on the measured r e is discussed.

Estimation of r e , rain and ice initiation
Rain is initiated during the warm phase of growing convective cumulus by intensification of the collision and coalescence (coagulation) processes with height. The efficiency of the process of droplet coalescence is determined by the collection kernel (K) of the droplets and their concentrations (Pruppacher et al., 1998). Freud and Rosenfeld (2012) have shown through model simulations and aircraft measurements that K ∝ r v 4.8 , where r v is the mean volume radius 200 obtained from the cloud probe DSDs in the absence of ice. r v is defined as follows:

( )
where ρ is the water density (1 g cm -3 ), CWC is in g m -3 , and N d is in cm -3 . The values are obtained from the 1-Hz data of droplet size distributions from the cloud probes. The calculation of CWC is performed separately with CAS-DPOL 205 and CCP-CDP probe droplet concentrations as follows: where N is the droplet concentration and r the droplet radius. The calculations of DWC, RWC, and MPWC are done in 210 similar fashion to CWC but with different cloud probes and particle size ranges (see Table 2).
The definition of r e is: ∫ ∫ 2 3 Freud and Rosenfeld (2012) showed that r v 1.08·r e , depending on the droplet size distribution. Using this relationship, they derived r e from r v and showed that warm rain initiates within clouds when r e is about 13-14 µm (Klein et al., 2009; 215 Rosenfeld and Gutman, 1994;Rosenfeld and Lensky, 1998;Rosenfeld et al., 2012aRosenfeld et al., , 2014b. Only measurements with CWC larger than 25% of the adiabatic water content are considered in order to exclude convectively diluted or dissipating clouds. It is assumed that rain (or ice) formation starts when calculated DWC exceeds 0.01 g m -3 (Freud and Rosenfeld, 2012). The small terminal fall speed of the drizzle drops (≤ 1 m s -1 ) allows to focus on in-situ rain (or ice) initiation while minimizing the amount of DSDs affected by rain drops fallen from above into the 220 region of measurements. In addition, cloud passes with rain were eliminated when cloud tops were visibly much higher than the penetration level (> ~1000 m), based on the videos recorded by the HALO's cockpit forward-looking camera.
However, cloud tops higher than few hundred meters above the penetration level occurred only rarely. Table 3 shows the cloud depth above cloud base at which warm rain initiation occurs (D r ) (i.e., DWC > 0.01 g m -3 ) for all flights as a function of estimated N a . The D r is taken as the cloud depth for ice initiation (D i ) if ice particles are evi-225 dent in the CCP-CIP images.
The N a for the convective clusters is estimated based on the slope between the calculated CWC and the mean volume droplet (M v ) for 1-s DSD measurements of CAS-DPOL and CCP-CDP for non-precipitating cloud passes (Braga et al., 230 2016). Braga et al. (2016) have shown that this estimated N a was in a reasonably good agreement with the directly measured cloud base droplet number concentration, N d , as obtained from the CCP-CDP and CAS-DPOL during ACRIDICON-CHUVA. Once N a is estimated, the adiabatic r e (r ea ) can be calculated based on a simple adiabatic parcel model where droplet growth is dominated by condensation (Pinsky and Khain, 2002).
The N a calculated for cloud base was used to classify clouds as having developed in clean, polluted, or very polluted 235 regions. A clean cloud case was defined as N a < 500 cm -3 , polluted for 500 cm -3 < N a < 900 cm -3 , and very polluted for N a > 900 cm -3 . During ACRIDICON-CHUVA, a flight in clean clouds (AC19) was performed over the Atlantic Ocean.
Clouds observed during flights over the northern Amazon were classified as polluted, mainly due to diluted smoke from biomass burning advected by long-range transport. This region represents the Amazon background condition for aerosol concentration during the dry season. Very polluted conditions were met over the Central Amazon, which was affected 240 strongly by biomass burning over the Amazonian deforestation arc (southern Amazon).

Threshold of r e for warm rain initiation
The values of r e derived from integrating the cloud probe DSDs were used to identify rain initiation. Some caution is 245 required to eliminate possible bias resulting from peculiar shapes of the drop size spectrum. An r e value of 13-14 μm represents the rain initiation threshold for growing convective cumulus observed at different locations in the world, as long as there is no significant influence from giant CCN (GCCN; dry soluble diameter > 1 µm) (Freud and Rosenfeld, 2012).The presence of GCCN during cloud droplet formation at cloud base can lead to a faster formation of raindrops due to both, the rain embryo effect and the competition effect that reduces cloud base maximum supersaturation and 250 consequently reduces N d (Rosenfeld, 2000;Segal et al., 2007). Such cases are very common over the ocean due to sea spray aerosols; there, the values of r e at which raindrops start to form are commonly smaller than the usual threshold of 13-14 µm (Freud and Rosenfeld, 2012). In our study the DSDs from flight AC19 performed over the Atlantic Ocean did not show a large drop tail near cloud base (see Figure  expected to occur at r e > 13 µm, whilst for CAS-DPOL the rain initiation threshold is r e > 12 µm. Difference of the two instruments in the r e range below ~7 µm and above ~11 µm have been discussed in Braga et al. (2016). For r e < 7 µm, they are related to a higher sensitivity of the CAS-DPOL for small cloud and aerosol particles, whereas for r e > 11 µm CAS-DPOL has lower sensitivity to large particles than CCP-CDP; however the differences are not significant within 265 the uncertainties of the measurements. According to Figure 3, the r e calculated with the DSD measured with the CCP-CDP is about ~7% higher than the r e calculated from CAS-DPOL data.

Comparing estimated r ea with measured r e
The comparison between the values of r ea (calculated from the estimated N a at cloud base described in Section 3.2) with the measured r e is the basis for analyzing the evolution of cloud particle size until rain or glaciation initiation occurs within the cloud. Rosenfeld et al. (2012b) showed that a tight relationship between the N a calculated for cloud base and the evolution of r ea with height (r ea -D c ) provides a useful proxy of the depth in convective clouds at which raindrops 280 start to form.

Case study: Flight AC07 over the Amazon deforestation arc
Flight AC07 was performed over the deforestation arc (see Figure 1a). Figure 5 shows the number of droplets measured at different heights in the convective clouds. Droplet concentrations reaching ~2000 cm -3 were measured at cloud base, 285 which is characteristic for very polluted clouds. The cloud base was located at about 1900 m above sea level, with ambient air temperature at about 16ºC. Figure 6a shows the mean DSD for a cloud penetration at cloud base. It emphasizes the higher number concentration of small droplets (< 10 µm) that are observed in convective clouds forming in polluted environments. Figure 6b shows the evolution of r e measurements and estimated r ea as a function of temperature. The figure also shows that the values of r e do not exceed the 13 µm threshold at warm temperatures. These results suggest 290 that cloud droplets formed at cloud base grow mainly via condensation and no raindrops were formed during the warm phase of convective cloud development. However, to rule out coalescence processes as a possible reason for droplet growth, further analysis using CCP-CIP images was done.
Figures 7a-c show the evolution of DSD and CWC (mean values) as a function of height above cloud base and the cloud particle images from the CCP-CIP. Figure 7a  was smaller than 1 drop m -3 . This is a negligible rain rate, and supports the notion of practical shut of coalescence.
Furthermore, the CCP-CIP did not detect any raindrops at lower levels (warm temperatures) for a cumulative sample The mean DSD and CIP images shown in Figure 7c result from a passage through a convective cloud with lightning activity. Figure 8 shows a photo of the cloud taken from the HALO cockpit just before the cloud penetration. The CCP-CIP has imaged graupel in this case. The presence of these type of ice particles within convective clouds is very common in thunderstorms, and previous studies highlight the large frequency of lightning occurrence during the dry-to-wet 315 season over the deforestation arc region of the Amazon (Albrecht et al., 2011;Williams et al., 2002). These results also highlight the role of aerosols from biomass burning on warm rain inhibition and on the aerosol invigoration effect due to the generation of large ice particles and lightning (Rosenfeld et al., 2008).
Regarding the values of r e as a function of D c , Figure 9a shows This region was classified as clean because N a is about 300 cm -3 (see Table 3). The cloud base was located at 600 m above sea level at a temperature of 23 ºC. Given the clean conditions over the ocean, the high relative humidity at surface level and the low concentration of CCN lead to the formation of large droplets already close to cloud base. Figure   10b shows the estimated r ea and the measured r e as a function of D c . Several cloud passes showed large droplets with r e 330 ~ 13 µm at only 1660 m above cloud base. Figures 11a-b show the DSDs and CCP-CIP images for the cloud passes at the height where rain starts to form and at the greatest height measured above cloud base, respectively. Figure 11a shows that rain is initiated (DWC > 0.01 g m -3 ) already when the droplets become larger than about r e > 12 µm. This is probably due to the presence of GCCN over this maritime region. Figure 12 shows the mean aerosol particle size distribution (PSD), as measured by the PCASP, just below cloud base for 335 clean, polluted, and very polluted regions. The mean total number concentration of aerosol particles with sizes larger than 0.1 μm is about 1000 cm -3 over the Atlantic Ocean, whilst for polluted (very polluted) case this value is about three (ten) times larger. This figure indicates the presence of large aerosols particles with sizes greater than 1 µm (possibly GCCN) over the ocean. When it nucleates droplets, this type of aerosol accelerates the growth of droplets during the warm phase leading to a faster formation of raindrops than predicted by the adiabatic parcel model. About 3500 m 340 above cloud base, large raindrops are observed in the CCP-CIP images (see Figure 11b). The low CWC indicates that most of it was already converted into raindrops. These results highlight that under clean conditions, raindrops were formed mainly by warm phase processes of cloud development. Even if the convective clouds reach colder temperatures, the low remaining amount of cloud water suppresses the development of cloud electrification.
Before raindrops start to form (Dc ~1,660 m) updrafts were observed with most values < 4 m s -1 , and when rain starts 345 downdrafts starts to be evident (see Figure 3g at supplementary material). The values of vertical velocities measured at flight AC19 (clean region) were smaller than measured for flight AC07 (very polluted region), but the vertical velocities patterns of updrafts been observed when particles are growing via condensation, and downdrafts been observed after larger particles appear (as raindrop or ice form). However, for polluted case strong updrafts (~10 m s -1 ) are also observed after ice starts to form, probably due to the latent heat was released during freezing processes.

-Polluted regions
The flights AC09 and AC18 were classified as polluted (see Table 3). These flights were performed over the northern Amazon region (see Figure 1a). Figure 13a shows the measured N d from flight AC09. The cloud base was located about 1200 m above sea level at a temperature of 19.5 ºC. Figure 13b shows and ice particles were observed on flight AC09 (see Figure 14b). Larger raindrops and a high amount of DWC were observed on AC09 for warmer temperatures than on flight AC18 (not shown). These results show that differences in cloud particle formation are associated with the D c at which convective clouds start to form raindrops or ice, defined 365 earlier as D r and D i . Flight AC18 has a droplet concentration, N d , of up to 100 cm -3 greater than the measurements during AC09. With higher N d at cloud base, droplet growth via condensation in convective clouds is a less pronounced function of height due to the water vapor competition between droplets. Under these conditions, the collision and coalescence process and freezing of droplets are initiated at higher D c (Freud and Rosenfeld, 2012;Rosenfeld et al., 2008).

-Very polluted regions
Five flights were classified as very polluted (see Table 3): AC07, AC08, AC12, AC13, and AC20. The microphysical analysis of the measurements collected in growing convective cumulus during flight AC07 was already presented in Section 4.2.1. Figure 15a show the measured N d from flight AC13, which was made in the same region as flight AC07.
The figure shows that the values of N d near cloud base on flight AC13 reach 2000 cm -3 , similar to AC07. However, the 380 rate of decrease of N d with height above cloud base is much smaller in AC13 compared to AC07. A possible explanation for this relative increase of N d above cloud base is the occurrence of secondary nucleation when the high concentrations of aerosol particles extend well above cloud base. This is supported by the fact that the observed r e are smaller than the calculated r ea , as shown in Figure 15b. Only values below 13 µm are observed (maximum of 12 µm), indicating the suppression of raindrop formation. Indeed, no raindrops were observed in the CCP-CIP images from growing convec-385 tive cumulus passes on this flight, and only cloud droplets and ice particles were detected. Figure 16 shows the DSD and CCP-CIP images at the start of glaciation (D i ~4800 m). These results highlight the role of aerosols in inhibition of raindrop formation due to suppression of collision and coalescence processes in very polluted regions. In addition, D i is about 300 m greater in convective clouds from flight AC13 than in the case of flight AC07.
The measured N d during flights AC08, AC12, and AC20 was greater above cloud base than at cloud base on several 390 cloud passes (especially in flights AC08 and AC20; see  [Pruppacher et al., 1998]. This was previously documented by satellite retrievals , where glaciation temperatures of convective clouds were strongly dependent on r e at the -5 °C isotherm, where smaller r e were correlated with lower glaciation temperatures.

405
The results from cloud probe measurements under clean, polluted, and very polluted conditions highlight the role of aerosol particles in rain and ice formation for growing convective cumulus. Figure   AC09 and AC18. The r e for rain initiation is slightly smaller (12 μm) on AC19; probably due to the sea spray giant CCN, which accelerate the coalescence for a given r e . Mixed phase precipitation was initiated on flights AC07 and AC13, well below the height of D 13 at an r e of 12 and 10 μm, respectively. Ice starts to form at lower temperatures when the cloud droplets are smaller, as manifested by D i of -9 and -14 °C for flights AC07 and AC13, respectively. The re-430 maining flights did not reach the height for rain initiation (AC08, AC12, and AC20).
It is evident that raindrops form faster via collision and coalescence process in a cleaner atmosphere. For the polluted cases, raindrops form at colder temperatures (~0 ºC and colder) via collision and coalescence than for clean conditions.
Rain can initiate at supercooled temperatures, e,g., -5 °C on AC18. The raindrops were documented to start freezing at -9 °C in AC09. In very polluted conditions, only cloud droplets, but no raindrops were observed at D c < 4000 m. In these 435 cases, precipitation was initiated as ice particles at D c > 4000 m. These flights with completely suppressed warm rain were performed over the smoky deforestation arc. Measurements of aerosol concentrations and N d above cloud base indicate new nucleation of cloud droplets for flight AC13 (not observed at AC07) in the course of the development of convective cumulus. This secondary nucleation leads to smaller r e . For flights where secondary nucleation was significant, the differences between the estimated r ea and the r e measurements at same height are larger, because the adiabatic 440 estimation does not consider the secondary nucleation of droplets above cloud base and thus overestimates the observed size.

Conclusions
This study focused on the effects of aerosol particle number concentration on the initiation of rain drops and ice hydro-445 meteors in growing convective cumulus over the Amazon. Data from aerosol and cloud probes on board of the HALO aircraft were used in the analysis. The values of the estimated N a at cloud base were applied to classify the atmospheric conditions where convective clouds developed as a function of aerosol particle number concentration (i.e., clean, polluted, and very polluted regions). From the estimated N a , the evolution of r ea , the theoretical r e assuming adiabatic growth of droplets, with cloud depth above cloud base (D c ) were compared with the observed r e at the various heights. A DWC 450 value of 0.01 g m -3 was used as a threshold for rain initiation or glaciation within clouds. Images from the CCP-CIP probe were used to detect the presence of raindrops and/or ice hydrometeors. The results support the use of r e ~ 13-14 µm as a threshold for rain initiation in convective clouds. The evolution of the directly observed r e follows that of the calculated r ea with slight differences as a function of aerosol particle size distributions. Rain initiation occurred higher in more polluted clouds, as manifested by higher D c . Rain was initiated at supercooled levels in moderately polluted 455 clouds. In very polluted conditions, warm rain was suppressed completely. This was exacerbated by the occurrence of secondary nucleation above cloud base, which further reduced r e compared to r ea . The initiation of ice hydrometeors is also delayed to greater D c in more polluted clouds, because smaller drops freeze at colder temperatures. Ice was initiated mostly by freezing raindrops in cases when warm rain formation was not completely suppressed. Both the D 13 and D r increased linearly with N a , in agreement with the theoretical considerations of Freud and Rosenfeld (2012). The results 460 suggest also that, in the absence of new droplet nucleation above cloud base, D 13 is very similar to D i under very polluted conditions, where raindrops are not formed at warmer temperatures.
These results show that even moderate amounts of smoke, which fill most of the Amazon basin during the drier season, are sufficient to suppress warm rain and elevate its initiation to above the 0 ºC isotherm level. This results in a suppression of rain from small clouds and an invigoration in the deep clouds, as hypothesized by Rosenfeld et al. (2008). While 465 the net effect on rainfall amount is unknown, the redistribution of rain intensities and the resulting vertical latent heating profiles are likely to affect the regional hydrological cycle in ways that need to be studied further.
for each location is denoted by "n" in the legend. The r e and r v mean linear relationship for all flights is shown below 720 the linear relationship for each flight in black color.          Table 1. List of abbreviations and symbols. Table 2. Description of cloud probes, size range intervals and hydrometeor shapes observed on CCP-CIP images used to calculate CWC, DWC, RWC and MPWC.

μm
Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-1155 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 9 February 2017 c Author(s) 2017. CC-BY 3.0 License. Figure 8 Image taken from the HALO cockpit just before the aircraft penetration of a convective cloud with lightning activity during flight AC07. In this case, the cloud pass height was 9,022 m (temperature ~ -25 ºC) and the maximum CWC measured was 0.55 g m -3 .