This project extends the successful methodology of the iHorse initiative (NAISS 2024/22-1450), which developed high-resolution air quality forecasting for the Stockholm region using interpretable deep learning. While the previous phase successfully validated a physics-guided spatiotemporal decoupling framework for urban-scale pollution, air pollution is fundamentally a trans-boundary phenomenon driven by large-scale meteorological transport. Therefore, the primary objective of this renewal is to scale our modeling efforts from the Stockholm municipality to a pan-Nordic domain, covering Sweden, Finland, Norway, and Denmark. We aim to develop a unified, physics-informed spatiotemporal forecasting system by integrating data from over 200 air quality monitoring stations, sourced from EU and national agencies, alongside meteorological forecasts and deterministic model outputs (e.g., CAMS/MetCoOp). The core innovation involves extending our "Physics-Guided Spatiotemporal Decoupling" architecture to model long-range pollutant transport and advection-diffusion processes across the Nordic region. Unlike standard data-driven approaches, this framework embeds physical constraints (wind vectors and geographical topology) directly into attention mechanisms to ensure physical consistency. We will rigorously benchmark this approach against state-of-the-art baselines, including AirFormer and AirPhyNet, to demonstrate superior predictive accuracy and interpretability for regional environmental monitoring.