We analyze the importance of cloud top temperature, dust aerosol, sea salt aerosol, and sea ice cover for the thermodynamic phase of low-level, mid-level, and mid to low-level clouds observed by CloudSat/CALIPSO over the Arctic and the Southern Ocean using an explainable machine learning technique.
As expected, the cloud top temperature is found to be the most important parameter for determining cloud phase. The results show also a predictive power of sea salt and sea ice on the phase of low-level clouds, while in mid-level clouds dust shows predictive power. Over the Southern Ocean, strong zonal winds coincide with the aerosol distribution. While they can produce high mixing ratios of sea spray at lower levels, the strong zonal winds may prevent the pole-ward transport of dust. Sea ice may prevent the release of sea salt aerosols and marine organic aerosols leading to higher liquid fractions in clouds over sea ice.