A Lagrangian Time‐Series Machine Learning Framework for Predicting Concentrations and Exploring Drivers of Atmospheric Aerosols: Model Development and Application to Cloud Condensation Nuclei in Marine Boundary Layer

Abstract

Atmospheric aerosols play critical roles in climate and air quality. Accurately assessing their environmental impacts requires understanding aerosol abundance, distribution, and the key factors and processes that control them. Although machine learning is increasingly used in predicting aerosol concentrations, its application in exploring underlying processes remains limited, and existing studies often fail to capture processes occurring during preceding days that substantially shape aerosol populations. Here we develop a Lagrangian time‐series machine learning framework that better represents those dynamic processes. Three‐dimensional back‐trajectories of airmasses arriving at the receptor site are derived using a physics‐driven Lagrangian transport model, and environmental variables along these trajectories are included as time‐series input features for a sequence model to predict aerosol concentrations. We apply this framework to multiyear cloud condensation nuclei (CCN) observations from the eastern North Atlantic (ENA). The resulting model exhibits strong predictive skill and outperforms conventional machine learning approaches that rely on static, non‐sequential input features. Moreover, the model trained solely on the ENA data successfully reproduces a large fraction of CCN variability at a site in the South Atlantic, indicating that it indeed captures the general processes driving marine CCN variability. Model interpretation reveals that the seasonal variation of CCN over ENA is primarily controlled by aerosol production associated with short‐wave radiation, whereas the variation on a shorter timescale is dominated by wet removal. This Lagrangian time‐series framework, which is more physically consistent with actual atmospheric processes, offers a powerful tool for predicting and understanding the mechanisms of other atmospheric components.