Black carbon (BC) aerosol absorbs solar radiation and can act as cloud condensation nucleus and ice formation nucleus. The current generation of climate models have difficulty in accurately predicting global-scale BC concentrations. Previously, an ensemble of such models was compared to measurements, revealing model biases in the tropical troposphere and in the polar troposphere. Here global aerosol distributions are simulated using different parameterizations of wet removal, and model results are compared to BC profiles observed in the remote atmosphere to explore the possible sources of these biases. The model-data comparison suggests a slow removal of BC aerosol during transport to the Arctic in winter and spring, because ice crystal growth causes evaporation of liquid cloud via the Bergeron process and, hence, release of BC aerosol back to ambient air. By contrast, more efficient model wet removal is needed in the cold upper troposphere over the tropical Pacific. Parcel model simulations with detailed droplet and ice nucleation and growth processes suggest that ice formation in this region may be suppressed due to a lack of ice nuclei (mainly insoluble dust particles) in the remote atmosphere, allowing liquid and mixed-phase clouds to persist under freezing temperatures, and forming liquid precipitation capable of removing aerosol incorporated in cloud water. Falling ice crystals can scavenge droplets in lower clouds, which also results in efficient removal of cloud condensation nuclei. The combination of models with global-scale BC measurements in this study has provided new, latitude-dependent information on ice formation processes in the atmosphere, and highlights the importance of a consistent treatment of aerosol and moist physics in climate models.