A Bayesian framework to make GF-1/6 WFV and Sentinel-2 MSI reflectance consistent with MODIS NBAR

A major challenge for direct multi-satellite data fusion is the substantial bias introduced by significant differences in hardware configurations, spectral response functions (SRFs), and observation conditions among satellites. Therefore, ensuring reflectance consistency is an essential prerequisite for multi-satellite synergistic application. This study employs MODIS NBAR as the benchmark dataset and adopts a Bayesian framework to harmonize GaoFen-1/6 (GF) wide field view (WFV) and Sentinel-2 multi-spectral instrument (MSI) reflectance with MODIS NBAR. Quantitative evaluation using R 2 , RMSE, rRMSE and bias demonstrates that the Bayesian framework significantly improves consistency. Independent validation using the HLS dataset shows that the post-consistency reflectance of GF-1/6 WFV achieves R 2 of 0.83–0.94, RMSE of 0.014–0.075, rRMSE of 6.5%–37.1% and bias of −0.0726–0.0187 across all bands with p-values < 0.001. For Sentinel-2 MSI, the HLS validation also yields statistically significant correlations (p-values < 0.001), with R 2 ranging from 0.33 to 0.66. Time-series analysis of 57 multi-satellite NDVI observations indicates that reflectance consistency improves temporal coherence and reduces anomalous fluctuations, providing a more robust data foundation for crop monitoring.