Abstract
Air quality forecasting (AQF) is critical for public health and environmentalmanagement, yet remains challenging due to the complex interplay of emissions,meteorology, and chemical transformations. Traditional numerical models, suchas CMAQ and WRF-Chem, provide physically grounded simulations but arecomputationally expensive and rely on uncertain emission inventories. Deeplearning models, while computationally efficient, often struggle withgeneralization due to their lack of physical constraints. To bridge this gap,we propose PCDCNet, a surrogate model that integrates numerical modelingprinciples with deep learning. PCDCNet explicitly incorporates emissions,meteorological influences, and domain-informed constraints to model pollutantformation, transport, and dissipation. By combining graph-based spatialtransport modeling, recurrent structures for temporal accumulation, andrepresentation enhancement for local interactions, PCDCNet achievesstate-of-the-art (SOTA) performance in 72-hour station-level PM2.5 and O3forecasting while significantly reducing computational costs. Furthermore, ourmodel is deployed in an online platform, providing free, real-time air qualityforecasts, demonstrating its scalability and societal impact. By aligning deeplearning with physical consistency, PCDCNet offers a practical andinterpretable solution for AQF, enabling informed decision-making for bothpersonal and regulatory applications.