A machine learning paradigm for necessary observations to reduce uncertainties in aerosol climate forcing

Uncertainties in estimating climate cooling by anthropogenic aerosols remain significant. This study presents a machine learning (ML) paradigm to derive higher-level aerosol properties (aerosol light absorption and cloud condensation nuclei concentrations) using only lidar observations and reanalysis data. High-accuracy suborbital lidar and in situ measurements train and test two neural network algorithms. The algorithms accurately predict these properties, significantly improving upon conventional aerosol retrievals. This paradigm holds great potential for constraining Earth System Models and reducing uncertainties in aerosol climate forcing estimates.

Jens Redemann, Lan Gao