Environmental numerical models for predicting chemical transport and fate: From multimedia mass balance to machine...

The increasing release of persistent and emerging chemicals into the environment has intensified the need for robust frameworks capable of predicting their transport, fate, and environmental behavior across multiple spatial and temporal scales. Over the past decades, environmental multimedia mass-balance models have played a key role in mechanistic simulations of chemical fate. Rapid advances in machine learning (ML) have introduced data-driven alternatives with enhanced capacity for handling high-dimensional, nonlinear datasets. The relative strengths, limitations, and appropriate application domains of these two paradigms remain insufficiently discussed. We examine classical and emerging numerical environmental models for chemical transport and fate, with a focus on multimedia mass-balance models and machine learning. We review the theoretical foundations, structures, and applications of widely used mass-balance models at local, regional, and global scales. We evaluate recent ML applications in parameter estimation, concentration prediction, source identification, and degradation modeling. We synthesize these models according to their thermodynamic basis, spatial segmentation, temporal treatment, and chemical applicability. We highlight the differences in model interpretability, data dependence, uncertainty propagation, and policy relevance. We identify key limitations that affect model reliability, including limited availability and uncertainty of parameters, incomplete characterization and propagation of uncertainty, and restricted generalization beyond the model assumptions or training data. We further discuss emerging hybrid strategies that integrate machine learning with mechanistic mass-balance models. When appropriately designed, independently validated, and accompanied by uncertainty analysis, these approaches can potentially improve parameterization, reduce systematic bias, and support scenario-based projections; however, they may also introduce additional uncertainty associated with ML predictions and model coupling. This review provides a methodological reference for future environmental fate modeling, supporting the development of more reliable tools for predicting the environmental fate of chemicals and for decision-making.