Deep Spatio-Temporal Graph Neural Networks for Satellite-Calibrated Urban Air Quality Forecasting
Presents an end-to-end spatio-temporal framework coupling Sentinel-5P TROPOMI satellite imagery with sparse ground IoT air sensors across central Indian urban corridors. Formulates dynamic adjacency matrices driven by boundary-layer wind advection, demonstrating a 23.4% reduction in root mean square error during extreme winter thermal inversion events.