Sentinel-2 offers the capabilities to observe Arctic sea ice features with high spatial and temporal resolution. Arctic sea ice drift, however, exacerbates observing the temporal evolution of floes by means of time series analysis. We therefore developed a novel rotation-invariant ice floe descriptor based on ice floe geometry and an image-processing workflow consisting of three main steps: (i) ice floe extraction, (ii) ice floe description and (iii) ice floe matching. We tested the methodology on Sentinel-2A images from 10 June and 3 July, 2017, and selected five floes present in both images. We further added ten “false samples” in the second image. All floes from the first image were correctly identified and matched with the floes from the second image. The methodology enables the identification of individual ice floes and determination of their relative rotation from multiple Sentinel-2 images.
|