SMASH seminar

Abstract

Machine learning (ML) is emerging as a powerful tool in atmospheric sciences, enabling novel insights into complex, nonlinear processes that govern aerosol behavior and climate interactions. In this work, we present three independent studies showcasing the application of ML to key challenges in aerosol science. First, we develop a supervised ML algorithm to predict the optical properties of black carbon particles from their physical and chemical characteristics, improving the representation of their radiative effects in climate models. Second, we apply ML techniques to parameterize aerosol hygroscopicity using κ (kappa) measurements from a global dataset spanning diverse environmental conditions. This approach captures spatial and compositional variability in κ-values, aiding improved prediction of cloud condensation nuclei activity. Third, we use multivariate ML models to explore the relationships between aerosol size distributions and a suite of environmental drivers, including meteorological conditions, gas-phase precursors, and chemical composition. Together, these studies demonstrate the versatility of ML in addressing different aspects of aerosol behavior and its potential to enhance atmospheric modeling and understanding.