Determining the reflectance offset per tile in a Sentinel-2 archive of mixed processing baselines: data, scripts and verification

Data, scripts and figures behind the article "Determining the reflectance offset per tile in a Sentinel-2 archive of mixed processing baselines" (Computer Optics, submitted 2026). Since Sentinel-2 processing baseline 04.00, surface reflectance is stored with an additive offset: physical reflectance is (DN - 1000) / 10000 rather than DN / 10000. A corpus assembled over several years therefore mixes two scales, and the mixture is silent: nothing in the raster says which convention a given tile follows. This archive contains the determination of the processing epoch for 2022 Major TOM Core-S2L2A tiles, the correction of the additive offset, and everything required to check both. The processing epoch is established by three routes: pairing with same-scene Landsat Collection 2 acquisitions (1388 tiles), reading the processing baseline from the product identifier (349 tiles), and propagation within a single Sentinel-2 product (237 tiles, plus 2 reassigned by product majority). Coverage reaches 97.7 per cent; 46 tiles are left undetermined and are reported as such rather than guessed. The accuracy of the reference route is measured independently of its construction, by the rate at which reference tiles disagree with their own product: 4 of 954, or 0.42 per cent. For tiles carrying the offset, the correction shifts the salinity index by a median of -0.060 (interquartile range -0.077 to -0.047), and 95.2 per cent of them shift by more than 0.02. For tiles not carrying it the median shift is zero. The archive includes the indices both before and after the correction, so that this comparison can be reproduced rather than taken on trust. Every number printed in the article is recomputed from these files by verify_paper.py, which prints each published figure next to the value it computes and marks the line as matching or not. Code: MIT. Data tables: CC BY 4.0. Copernicus Sentinel data, processed by ESA. Landsat data courtesy of the U.S. Geological Survey, accessed through Microsoft Planetary Computer.