Advancing air pollution forecasting: a review of physical, statistical, and machine learning methods

Air pollution is one of the important environmental and public health hazards that requires a combination of assessments, monitoring, and mitigation approaches. The current review is an attempt to review the tools for monitoring and modeling air pollution data focusing on predictive forecasting. The review involves different modeling techniques, including deterministic models, statistical methods, and machine learning and deep learning techniques employed to process pollutant concentration, considering data quality, uncertainty assessment, model interpretability, and temporal and spatial scales. An emphasis is laid on hybrid and ensemble models integrating physics-based and data-driven approaches to enhance the prediction accuracy and robustness. Finally, the review explores emerging directions, such as physics-informed machine learning and edge-computing models, and identifies opportunities for open data platforms and collaborative research to promote fair and meaningful air quality management in a global context.