Predictions of the\nOptical Properties of Brown Carbon\nAerosol by Machine Learning with Typical Chromophores

The linkages between BrC optical properties and chemical\ncomposition\nremain inadequately understood, with quantified chromophores explaining\nless than 25% of ambient aerosol light absorption. This study characterized\n38 typical chromophores in aerosols collected in Xi’an, with\nlight absorption contributions to BrC ranging from 1.6 ± 0.3\nto 5.8 ± 2.6% at 365 nm. Based on these quantified chromophores,\nan interpretable machine learning model and the Shapley Additive Explanation\n(SHAP) method were employed to explore the relationships between BrC\noptical properties and chemical composition. The model attained high\naccuracy with Pearson correlation coefficients (<i>r</i>) exceeding 0.93 for the absorption coefficient (Abs<sub>λ</sub>) and surpassing 0.57 for mass absorption efficiency (MAE<sub>λ</sub>) of BrC. It explains more than 80% of the variance in Abs and over\n50% in MAE, significantly improving the understanding of BrC light\nabsorption. Polycyclic aromatic hydrocarbons (PAHs) and oxygenated\nPAHs (OPAHs) with four and five rings exhibit significant positive\neffects on Abs<sub>λ</sub>, suggesting that similar unidentified\nchromophores may also notably impact BrC optical characteristics.\nThe model based on chromophore mass concentrations further simplifies\nstudying BrC optical characteristics. This study advances understanding\nof the relationship between BrC composition and optical properties\nand guides the investigation of unrecognized chromophores.