Advances in air quality modeling through artificial intelligence, machine learning, and deep learning: A...

Area of Science:

• Environmental Science
• Computer Science
• Atmospheric Chemistry

Background:

• Traditional air quality models (e.g., CTMs) have limitations in accuracy and computational cost.
• Machine learning (ML) and deep learning (DL) offer novel approaches to overcome these limitations.

Purpose of the Study:

• To systematically review the evolution of ML/DL in air quality modeling.
• To categorize current ML/DL approaches and identify challenges and future directions.

Main Methods:

• Categorization of 112 publications into data-driven and ML-assisted models.
• Analysis of ML applications in pollutant concentration estimation, CTM refinement, and emulation.

Main Results:

• Data-driven models significantly improve forecast accuracy for PM2.5 and ozone.
• ML-assisted methods enhance traditional modeling through bias correction and emulation, reducing computational costs.
• Challenges include model transparency, uncertainty quantification, data scarcity, and integrating physical laws.

Conclusions:

• ML/DL are crucial for next-generation air quality modeling.
• eXplainable AI (XAI) and Physics-Informed Neural Networks (PINN) are key to developing faster, reliable, and interpretable systems.
• These advancements will improve pollution impact assessment and mitigation strategies.