The Global PM2.5 Concentration Product (LGHAP V2 PM2.5) features seamless coverage of global land areas at 1 km resolution and provides continuous daily PM2.5 concentration distributions from 2000 to 2023. This product, with its high accuracy and extensive coverage, effectively supports air pollution management, haze trend analysis, and health risk assessment, while also providing essential data support for climate research and environmental policy optimization. The Global PM2.5 Concentration Product (LGHAP V2 PM2.5) generates seamless global PM2.5 concentrations using multi-source geospatial big data and an advanced deep learning technique centered on a Scene-Perceptive Integrated Learning Graph Attention Network (SCAGAT). By integrating a dynamic graph attention mechanism and multi-scene adaptive learning, this method overcomes the limitation of insufficient training samples in areas with sparse monitoring stations. It effectively fills data gaps by incorporating spatiotemporal analysis, multi-scale feature extraction, and self-correcting iterative optimization, ensuring high-precision, continuous global coverage.