Assessing the radiometric consistency of Gaofen-6 WFV and Sentinel-2 MSI multispectral reflectance for harmonization

Harmonizing various satellite imagery could enhance Earth observation capabilities by facilitating the generation of dense time series of fine-resolution data, which is crucial for the monitoring of large-scale environmental changes. The Chinese Gaofen-6 Wide Field View (GF-6 WFV) and Sentinel-2 Multispectral Instrument (S2 MSI) are particularly suitable for such harmonization due to their similar spectral and spatial characteristics. Assessing the radiometric consistency of these instruments under various observation conditions is crucial for successful data integration. The objective of this study is to conduct a comprehensive cross-comparison analysis of WFV and MSI. In this paper, we assessed radiometric consistency across top-of-atmosphere (TOA) reflectance, bottom-of-atmosphere (BOA) reflectance, and nadir reflectance adjusted for bidirectional reflectance distribution function (BRDF) effects (BRDF-corrected) in near-contemporaneous observations for six bands between the MSI and WFV imagery. We evaluated differences using absolute difference metrics and reduced major axis regression (RMA) analysis between 16 image pairs across China in 2020, under various observation conditions, including different illumination angles, atmospheric conditions, and land cover types. RMA regression results indicate that simulated WFV and MSI surface reflectance derived using the sensor spectral response functions (SRFs) and laboratory spectra showed high radiometric agreement (R2 > 0.9). However, the observed reflectance revealed significant discrepancies, primarily attributed to wider field-of-view of GF-6 WFV. Incorporating the pixel-level view zenith angle during atmospheric correction significantly improved the radiometric consistency of both products. Root-mean-square deviation (RMSD) values ranging from 0.0199 to 0.0390 were obtained between MSI and WFV TOA reflectance, and RMSD values ranging from 0.0202 to 0.0369 were obtained from BOA reflectance. The improvement was particularly pronounced in the red edge 2 band, evidenced by a 22% reduction in the RMSD and a 65% decrease in bias between the two instruments. This study also identifies the necessity of BRDF correction for wide field-of-view sensors like GF-6 WFV.