Physics-Guided Inductive Spatiotemporal Kriging for PM2.5 with Satellite Gradient Constraints
The Spatiotemporal Physics-Guided Inference Network (SPIN) reconstructs high-resolution PM2.5 pollution maps using sparse ground data and satellite observations. It integrates physical laws like wind advection and diffusion into a deep learning framework, allowing it to generate continuous pollution fields even in areas with no sensors, overcoming data gaps caused by clouds or nighttime.