AI applications in air pollution domain: A systematic review from flue gas treatment to air quality management and carbon capture

This study conducted a systematic bibliometric analysis and comprehensive literature review of 906 publications from 2016 to 2025. The review categorizes AI applications into three core pillars: industrial flue gas treatment, air quality management, and CO₂ capture. This analysis identifies a significant paradigm shift from traditional statistical monitoring toward sophisticated AI-driven surrogate modeling, generative material discovery. In industrial flue gas treatment domain, AI-based surrogates have enhanced multi-pollutant prediction accuracy (e.g., R²>0.94). In outdoor and indoor air quality management, hybrid AI and Large Language Model (LLM) frameworks have improved spatiotemporal forecasting and monitoring reliability by up to 94.46% alarm judgement accuracy under data-constrained conditions. Furthermore, in CO₂ capture, the integration of generative AI and agentic AI systems has revolutionized adsorbent discovery, enabling rapid inverse design of metal-organic framework (MOF). Despite these advancements, the air pollution domain-AI faces challenges, including data scarcity in transient process, limited model interpretability, and the gap between computational design and experimental validation. Future research should prioritize Physics-Informed Neural Networks for high-fidelity modeling, Reinforcement Learning for adaptive control, and Agentic AI for End-to-End workflow. These emerging AI technologies are essential for developing scalable, high-precision air pollution management systems capable of meeting future environmental challenges.