Multi-source satellite observations of the column-averaged dry-air mole fraction of carbon dioxide (XCO₂) are limited by sparse orbital sampling, cross-platform footprint mismatch, and heterogeneous retrieval uncertainties, leading to substantial spatial discontinuities and uncertainty propagation errors in regional XCO₂mapping. To address these limitations, we developed an adaptive fusion framework integrating physics-based Gaussian oversampling and Bayesian maximum entropy (PGO-BME) for uncertainty-aware monthly XCO₂reconstruction over China. Monthly observations from DQ-1, OCO-2, OCO-3, and GOSAT collected between June 2022 and December 2024 were fused within the framework. In hold-out validation, the framework achieved an overall R² of 0.90 and an RMSE of 0.63 ppm, substantially outperforming conventional pixel-center and area-weighting approaches. Independent validation against the Total Carbon Column Observing Network (TCCON) further showed R²values of 0.93 at both the Hefei and Xianghe stations. The reconstructed dataset preserved both large-scale atmospheric transport patterns and localized anthropogenic enhancements, while posterior uncertainties were consistent with the corresponding spatial prediction errors. These results demonstrate that physically constrained, uncertainty-aware fusion can improve the spatial continuity and reliability of satellite XCO₂fields while retaining atmospheric CO₂variability across spatial scales.