Impact of satellite image pairs selection when deriving ice velocities trends

When assessing ice velocity trends in Greenland, optical feature-tracking has previously been used to derive one-year velocity averages. Indeed, this technique requires pairs of images separated by approximately one year, and usually all possible pairs are used in order to achieve the best spatial coverage for every year. But this implies averaging pairs that start at different time in the year, and it is common to also use pair of images that are separated by shorter or longer time (ranging from 336 days up to 400 days between images).

Since ice velocities display strong seasonal variations, we argue that combining all pairs may impact the yearly ice velocities estimations by sampling differently summer and winter velocities, and therefore impacting the long-term trends.

Here we assess this impact by reproducing the work done from previous studies (Tedstone et al. 2015, Williams et al. 2020) using optical feature-tracking on Landsat-5, 7 and 8 as well as Sentinel-2 constellation, focusing on land-terminating parts of the Southwest of Greenland Ice sheet, and by comparing obtained velocity trend maps with trends of the same area obtained when operating a precise selection of the data.