Cross-sensor harmonization and calibration transfer

Cross-sensor harmonization answers the question: can observations from two different satellite instruments be treated as interchangeable in a single time-series? Instruments differ in their spectral response functions, view geometries, and absolute calibration, so combining them without adjustment introduces artefacts. The method corrects for these differences by adjusting all inputs onto a common radiometric and geometric reference.

The Harmonized Landsat and Sentinel-2 (HLS) product applies a four-step pipeline.^[1] First, LaSRC atmospheric correction retrieves surface reflectance from top-of-atmosphere radiance using aerosol optical thickness estimated directly from the imagery. Second, cloud and cloud shadow are masked. Third, BRDF normalization via the c-factor technique rescales each observation to a nadir-normalized surface reflectance using per-pixel BRDF model parameters from the MODIS MCD43A1 product, removing view-angle effects from off-nadir acquisitions. Fourth, a band-pass adjustment transforms Sentinel-2 MSI reflectance to the Landsat OLI spectral reference using linear slope-and-intercept coefficients derived from co-located hyperspectral simulations.^[1] The inter-sensor residual after adjustment is below 4.2% for red, NIR, SWIR-1, and SWIR-2 bands.^[2]

HLS V2.0 combines Landsat 8, Landsat 9, Sentinel-2A, Sentinel-2B, and Sentinel-2C to achieve a nominal 1.6-day global revisit at 30m resolution with 2-3 day latency.^[3] The choice of reference sensor varies by implementation: HLS designates Landsat OLI as the reference, while other harmonization products use different anchors. The method fails where the BRDF model is poorly constrained (sparse observations at high latitudes in winter) and where surface types are not represented in the hyperspectral training data used to derive band-pass coefficients.