A New Framework for Medium- to Long-Term PM_2.5 Predictions Using AI-Based High-Resolution Meteorological Forecasts

Accurate forecasting of PM_2.5 concentrations is vital for effective air quality management and public health protection, particularly in the context of medium- to long-term planning. Such forecasts are also essential for guiding emission control and timely responses in urban and regional environments. Reliable prediction over extended horizons critically depends on access to accurate, high-resolution meteorological fields. To address limitations in traditional approaches that rely on short-term site-level observations or retrospective meteorological fields, this study proposes a deep learning framework that couples AI-based gridded meteorological forecasts with a dynamic graph-based architecture for PM_2.5 prediction. The model captures evolving spatiotemporal patterns of pollutant transport and enables extended-range forecasts. By leveraging gridded meteorological forecasts, the proposed framework reduces 3-day RMSE and MAE of the prediction by 12% compared to models without the gridded inputs. For seasonal 10-day forecasts, errors can be decreased by approximately 20% relative to the physics-based Weather Research and Forecasting (WRF)-Community Multiscale Air Quality (CMAQ) system. Furthermore, interpretability analysis reveals the importance of wind-driven spatial asymmetries and supports the design of boundary-aware graph construction. Additional evaluations using Integrated Forecasting System (IFS) and PanGu-Weather forecasts further demonstrate that the model maintains high performance across different gridded data sources without retraining, highlighting its flexibility and potential to incorporate other AI-based meteorological models for adaptable and effective air quality forecasting. These findings underscore the potential of the proposed framework to serve as a computationally efficient and operationally feasible alternative to conventional chemical transport models in future air quality management systems, particularly for early warning systems and targeted emission control strategies.