Enhanced Particulate Matter (PM2.5) Forecasting via Spatiotemporal Graph Neural Networks with Adaptive Kernel Fusion (S-GNN-AKF)

**Abstract:** Ambient particulate matter (PM2.5) pollution poses a significant threat to public health. Accurate forecasting of PM2.5 concentrations is crucial for implementing effective mitigation strategies and protecting vulnerable populations. Existing models often struggle to capture complex spatiotemporal dependencies and non-linear relationships within atmospheric dynamics. This paper introduces a novel forecasting framework, Spatiotemporal Graph Neural Networks with Adaptive Kernel Fusion (S-GNN-AKF), which leverages graph neural networks (GNNs) to model spatial correlations and incorporates adaptive kernel fusion (AKF) to non-linearly combine diverse meteorological and environmental features. Demonstrating substantial improvements over traditional and existing machine learning methods, S-GNN-AKF enables more accurate and robust PM2.5 forecasting, facilitating proactive air quality management.