Tracking Seamless All-Hour PM_2.5 in China Using a Gridded Surface Visibility-Based Transformer Model

High-resolution spatiotemporal monitoring of PM_2.5 (particulate matter less than 2.5 μm in diameter) is essential for accurately assessing its environmental and health impacts. However, passive satellite-based techniques for estimating PM_2.5 concentrations are hindered by persistent cloud cover and the lack of nighttime aerosol optical depth (AOD) data, resulting in spatially and temporally fragmented retrievals that are restricted exclusively to daylight hours or daily averaged values. To address this, we introduce surface visibility (SV) as a robust alternative to AOD and develop a gridded SV-based transformer model (GSVTM). By integrating multisource meteorological and environmental data through multihead attention mechanisms and residual networks, the GSVTM effectively decouples the complex nonlinear relationship between SV and PM_2.5, enabling seamless all-hour PM_2.5 tracking at 6.25 km and hourly resolution across China. Extensive cross-validation demonstrates that at the hourly scale, GSVTM achieves an R² of 0.80 and an RMSE of 15.14 μg m^-3. At the daily scale, its accuracy (R² = 0.89 and RMSE = 9.73 μg m^-3) is comparable to that of existing satellite-based PM_2.5 products. Applied to a large-scale trans-regional pollution event, GSVTM successfully captures the complete dynamics of PM_2.5 transport and evolution. This study marks a critical advance toward seamless all-day PM_2.5 monitoring, effectively addressing the critical nighttime data gap in satellite-based PM_2.5 products and delivering reliable, real-time data essential for understanding diurnal patterns and multiscale impacts of PM_2.5 at national and urban scales.