Aerosol-cloud interactions in cirrus clouds based on global-scale airborne observations and machine learning models

Cirrus cloud formation and evolution are subject to the influences of thermodynamic and dynamic conditions and aerosols. This study developed near global-scale in situ aircraft observational datasets based on 12 field campaigns that spanned from the polar regions to the tropics from 2008 to 2016. Cirrus cloud microphysical properties were investigated at temperatures ≤ −40 °C, including ice water content (IWC), ice crystal number concentration (Ni​​​​​​​), and number-weighted mean diameter (Di). Positive correlations were found between the fluctuations of these ice microphysical properties and the fluctuations of aerosol number concentrations for larger (> 500 nm) and smaller (> 100 nm) aerosols (i.e. Na500 and Na100, respectively). Steeper linear regression slopes were seen for large aerosols compared with smaller aerosols. Machine learning (ML) models showed that using relative humidity with respect to ice (RHi) as a predictor significantly increased the accuracy of predicting cirrus occurrences compared with temperature, vertical velocity (w), and aerosol number concentrations. The ML predictions of IWC fluctuations showed higher accuracies when larger aerosols were used as a predictor compared with smaller aerosols, even though their effects were similar when predicting cirrus occurrences. To predict IWC magnitudes accurately, aerosol concentrations were particularly important at 50 to 250 s scales (i.e. 10–50 km) and showed increasing effects at low temperatures, small ice supersaturation, and strong updraughts/downdraughts. These results improve the understanding of aerosol–cloud interactions and can be used to evaluate model parameterizations of cirrus cloud properties and processes.