Indirect cross-comparison of Landsat-9 and Copernicus Sentinel-2 using EMIT as a hyperspectral transfer reference

Abstract. Combining observations from multiple optical satellite missions is essential for building consistent, multi-decadal records of Earth’s surface and atmosphere, but it requires that the participating sensors be radiometrically aligned. This study presents a global-scale, indirect cross-comparison of top-of-atmosphere (TOA) reflectance from Landsat-9 (L9) and the Copernicus Sentinel-2 (S2A and S2B) missions, using the hyperspectral imager EMIT (Earth Surface Mineral Dust Source Investigation) as a transfer reference. The Simultaneous Nadir Overpass (SNO) technique is first applied to each of the three sensor pairs— L9–S2, L9–EMIT, and S2–EMIT—and a double-differencing scheme then combines the L9–EMIT and S2–EMIT results into an indirect L9–S2 cross-comparison that no longer depends on EMIT’s absolute calibration. The visible, near-infrared, and shortwave-infrared bands are compared using mean reflectance values within 1 × 1 km² areas to suppress measurement noise. The geometry of the matchups confirms the orbital constraints affecting direct L9–S2 comparisons: L9 systematically precedes S2 by approximately 14 minutes and the two sensors view the target from opposite sides of nadir. In contrast, the L9–EMIT and S2–EMIT matchups show no systematic bias in time difference or viewing geometry, owing to EMIT’s non-Sun-synchronous orbit aboard the International Space Station. For each of the three direct cross-comparisons, the relative error remains within the combined measurement uncertainty for reflectance values above 0.2. The indirect L9–S2 comparison yields nearly constant biases (S2 minus L9) in the visible bands, with agreement better than 1 % in the green and red bands and approximately −2.5 % in the blue band. The near-infrared band agrees within 0.5 % above a reflectance of 0.3 and shows a mild non-linear behaviour at lower reflectances. The shortwave-infrared bands display a similar trend with a positive bias of up to 2.5 % at higher reflectances, consistent with the larger uncertainties of the L9–EMIT pair in this spectral region. Two features distinguish this methodology from a direct L9–S2 cross-comparison: EMIT’s hyperspectral sampling removes the need for a spectral band adjustment factor, and its non-Sun-synchronous orbit eliminates the systematic angular and temporal mismatches that affect SNO comparisons between Sun-synchronous missions. The same scheme could be applied, with substantially smaller uncertainties, using forthcoming SI-traceable satellite (SITSat) missions as the transfer reference, providing simultaneous absolute calibration and cross-mission alignment.