Abstract. In-situ measurements in convective clouds (up to the freezing level) over the Amazon basin show that smoke from deforestation fires prevents clouds from precipitating until they acquire a vertical development of at least 4 km, compared to only 1–2 km in clean clouds. The average cloud depth required for the onset of warm rain increased by ~350 m for each additional 100 cloud condensation nuclei per cm³ at a super-saturation of 0.5% (CCN_0.5%). In polluted clouds, the diameter of modal liquid water content grows much slower with cloud depth (at least by a factor of ~2), due to the large number of droplets that compete for available water and to the suppressed coalescence processes. Contrary to what other studies have suggested, we did not observe this effect to reach saturation at 3000 or more accumulation mode particles per cm³. The CCN_0.5% concentration was found to be a very good predictor for the cloud depth required for the onset of warm precipitation and other microphysical factors, leaving only a secondary role for the updraft velocities in determining the cloud drop size distributions.

The effective radius of the cloud droplets (r_e) was found to be a quite robust parameter for a given environment and cloud depth, showing only a small effect of partial droplet evaporation from the cloud's mixing with its drier environment. This supports one of the basic assumptions of satellite analysis of cloud microphysical processes: the ability to look at different cloud top heights in the same region and regard their r_e as if they had been measured inside one well developed cloud. The dependence of r_e on the adiabatic fraction decreased higher in the clouds, especially for cleaner conditions, and disappeared at r_e≥~10 μm. We propose that droplet coalescence, which is at its peak when warm rain is formed in the cloud at r_e=~10 μm, continues to be significant during the cloud's mixing with the entrained air, cancelling out the decrease in r_e due to evaporation.