Abstract. By utilising imagery from overlapping orbits, the Sentinel-2 programme offers high-frequency observations of high-latitude environments well in excess of its 5 d repeat rate, which is valuable for obtaining large-scale records of rapid environmental change. However, the production of glacier velocity datasets from optical feature tracking of Sentinel-2 imagery is limited by the orthorectification error in ESA products, which introduces significant systematic errors (on the order of

more »

... s of metres) into displacement fields produced from cross-track image pairs. As a result, most standard processing chains ignore cross-track pairs, which limits the opportunity to fully benefit from Sentinel-2's high-frequency observations during periods of intermittent coverage or for rapid dynamic events. Here, we use temporally complete glacier velocity datasets to empirically reconstruct systematic error, allowing for the corrected velocity datasets to be produced for four key fast-flowing marine-terminating outlets across the Greenland Ice Sheet between 2017–2021. We show that corrected data agree well with comparison velocity datasets derived from optical (Landsat 8) and synthetic aperture radar (Sentinel-1) data. The density of available velocity pairs produces a noisier dataset than for these comparative records, but a best-fit velocity reconstructed by time-series modelling can identify periods of rapid change (e.g. summer slowdowns), even where gaps exist in other datasets. We use the empirical error maps to identify that the commercial DEM used to orthorectify Sentinel-2 scenes over Greenland between 2017–2021 likely shares data sources with freely available public DEMs, opening avenues for the analytical correction of Sentinel-2 glacier velocity fields in the future.