100m Resolution SDGSAT-1 PM2.5 Concentration Maps

This dataset contains PM2.5 concentration estimates at 100-meter resolution across the North China Plain and Loess Plateau for the year 2022,derived from SDGSAT-1 imagery using the LAD-GAT model—a novel deep learning framework proposed for fine-scale air quality mapping. The LAD-GAT architecture constructs a high-dimensional scene attribute space, which integrates multi-source variables to form a graph structure and models relationships between in-situ PM2.5 observations and estimated pixel. Meanwhile, it introduces a LAD module to extract latent aerosol signals. The released dataset includes prediction results from three representative scenes, serving as demonstration cases of the model’s performance.