Artificial Intelligence in Atmospheric Composition Studies for Sustainable Air Quality Management: Spatiotemporal Concentration Forecasting and Emission Inference from Mobile and Point Sources

Air pollution remains a major challenge for sustainable development because of its impacts on human health, ecosystems, and climate. At the same time, the rapid growth of environmental data and advances in artificial intelligence (AI) have created new opportunities for atmospheric composition research and air-quality management. This review examines AI applications in atmospheric composition studies, focusing on two related but distinct tasks: (i) spatiotemporal forecasting of pollutant concentrations and (ii) emission inference from mobile and point sources. It emphasizes the fundamental differences between these tasks in terms of data requirements, model design, and physical interpretability. A synthesis of representative studies published between 2018 and 2025 is provided, covering machine learning and deep learning approaches for air-quality prediction and emission characterization. Recent foundation-style architectures and global AI weather models introduced in late 2025 and early 2026 further demonstrate the growing role of large-scale spatiotemporal learning in atmospheric and environmental prediction. Particular attention is given to hybrid and physics-informed models that aim to connect data-driven methods with atmospheric processes. The review also discusses major methodological challenges, including data representativeness, sensor uncertainty, spatial transferability, and model generalization under nonstationary conditions. It highlights the importance of leakage-resistant evaluation, appropriate temporal and spatial splitting strategies, and the roles of interpretability and uncertainty quantification in physically meaningful atmospheric modelling. From a sustainability perspective, these AI approaches can support more reliable monitoring, improved emission assessment, and better-informed strategies for air-pollution mitigation.