Chancellor's Professor University of Tennessee-Knoxville, Tennessee
Accurate weather and air quality forecasts are vital for protecting public health, shaping environmental policy, and understanding climate systems. Traditional chemical transport models (CTMs), while detailed, are computationally intensive and impractical for real-time use. To address this, the AI application, DeepCTM4D framework, uses deep learning to emulate CTM outputs, significantly improving computational efficiency while maintaining scientific credibility. Trained on emissions, meteorology, and chemical data, DeepCTM4D captures key atmospheric processes and enables rapid, interpretable predictions. It supports real-time applications and policy analysis, offering a powerful, scalable tool for air quality and weather forecasting. An application is also discussed in the measurement-model fusion processes. Complementing this, the DeepMMF model introduces a physics-constrained deep learning approach to improve NO₂ estimates over the U.S. It combines CTM outputs with satellite and ground observations, addresses sample imbalance issues, and enhances spatial-temporal accuracy. DeepMMF achieves superior performance compared to traditional methods and aligns emission adjustments with observed trends, highlighting its potential for air pollution exposure assessment and policy support.
