Seamless Global AOD Reconstruction Based on Transformer Model and Fusion of Multisource Satellite Products

Aerosol optical depth (AOD), as a key parameter reflecting the atmospheric aerosol load, plays a crucial role in the study of climate change, air quality, and the Earth’s radiation budget. However, current satellite-retrieved AOD products suffer from high missing rates, spatial discontinuities, and inter-product inconsistencies, which significantly limit their global applications. In this study, we propose a seamless reconstruction framework based on a Cross-Dimensional Self-Attention Transformer, by combining optimal interpolation fusion of multiple satellite products with Transformer-based imputation to generate globally continuous AOD fields with high accuracy. First, eight AOD products from MODIS and VIIRS sensors are optimally fused, which substantially improves both coverage and accuracy across the entire sample range. Subsequently, a Transformer imputation model that integrates temporal, spatial, and meteorological variables is employed to seamlessly recover missing regions through cross-dimensional attention mechanisms, further enhancing data completeness and reliability. Experimental results indicate that: (1) compared with individual satellite products, the optimal interpolation fusion significantly reduces errors and improves correlations with AERONET observations, doubling the overall coverage; (2) the Transformer-based imputation achieves 100% coverage while further increasing the coefficient of determination (R²) and reducing the root mean square error (RMSE), outperforming the reanalysis product MERRA-2; (3) at the regional scale, the imputed results accurately capture the spatiotemporal evolution of typical pollution episodes. The final global product reaches an overall accuracy of R² = 0.701 and RMSE = 0.073. This work provides a feasible pathway for generating seamless global AOD datasets, offering essential support for climate model validation, air pollution monitoring, and environmental policy-making.