Nutzungsbedingungen: Lizenz, https://creativecommons.org/licenses/by/4.0 / terms of use: https://creativecommons.org/licenses/by/4.0/

{"id": "cc-by/4.0",

"name": "Creative Commons Namensnennung – 4.0 International (CC BY 4.0)",

"url": " http://dcat-ap.de/def/licenses/cc-by/4.0 ",

"quelle": "Copyright DLR (year of production)"}

Created from Sentinel-1 Copernicus data provided by the Alaska Satellite Facility (ASF) https://vertex.daac.asf.alaska.edu/ and the European Space Agency (ESA) https://scihub.copernicus.eu/ .

Processing steps include the pre-processing of Sentinel-1 dual and single polarized GRD scenes with the ESA SNAP toolbox. The fronts are extracted with a deep learning approach described in:

Heidler, K., Mou, L., Baumhoer, C., Dietz, A., and Zhu, X. X.: HED-UNet: Combined Segmentation and Edge Detection for Monitoring the Antarctic Coastline, IEEE Trans. Geosci. Remote Sensing, 1–14, https://doi.org/10.1109/TGRS.2021.3064606 , 2021.

Baumhoer, C. A., Dietz, A. J., Kneisel, C., and Kuenzer, C.: Automated Extraction of Antarctic Glacier and Ice Shelf Fronts from Sentinel-1 Imagery Using Deep Learning, Remote Sensing, 11, 2529, https://doi.org/10.3390/rs11212529 , 2019.

IceLines Processing:

Processing:

Calving front locations are derived from single- and dual-polarized Sentinel-1 EW data which were pre-processed with the ESA SNAP Toolbox 8.0. Front extraction is performed with the deep learning architecture “HED-UNet” and several post-processing steps including elevation thresholding, morphological filtering and vectorization. On average, the true mean distance error for fronts extracted from dual-pol imagery is 210±11 m (5.2 pixel) with 95% confidence and 354±17 m (8.8 pixel) for single-polarized imagery depending on ice shelf front morphology, occurrence of surface melt and calving pattern.

Quality Assurance:

The daily product of IceLines has been validated against 94 manually delineated fronts. For each ice shelf, two manually delineated front positions were created for random dates outside the training data period. On average, the true mean distance error for fronts extracted from dual-pol imagery is 210±11 m (5.2 pixel) with 95% confidence and 354±17 m (8.8 pixel) for single-polarized imagery. Frontal movement along centerlines can be accessed with an accuracy of 85±75 m (2.1 pixel) with 95% confidence.