Spectral Harmonization Landsat-8 and Sentinel-2A: The Matching Bands Adjustment Before the Missing Bands Prediction Method

The Sentinel-2 MSI and Landsat-8 OLI are often used as data sources to form temporally dense harmonization observations for accurate quantitative studies. However, challenges remain in resolving surface reflectance differences between both sensors due to inconsistencies in spectral band positions, widths, and spectral response functions. In this study, we developed the matching bands adjustment before missing bands prediction method (MA-MP) for performing spectral harmonization between Landsat-8 and Sentinel-2A to reduce such surface reflectance differences. This method uses the global representative samples from the harmonized Landsat Sentinel-2 (HLS) product to build the spectral harmonization model of Landsat-8 and Sentinel-2A at the cluster level for band-pass adjustment in the bands where Landsat-8 and Sentinel-2A are matched, and for prediction in the band which is unilaterally missing bands in Landsat-8 image relative to Sentinel-2A image. Experimental results show robust spectral harmonization in four selected study areas with a variety of land cover types, and the spectral harmonization results are significantly better than those of current mainstream methods. Overall, the proposed MA-MP can adjust the observations of Landsat-8 and Sentinel-2A in matching bands and predict the observations of Landsat-8 in a unilaterally missing band, which has the potential to be used flexibly for building the spectral harmonization model between any two different sensor images.