03 Jul 2023
 | 03 Jul 2023
Status: this preprint is currently under review for the journal TC.

SAR Deep Learning Sea Ice Retrieval Trained with Airborne Laser Scanner Measurements from the MOSAiC Expedition

Karl Kortum, Suman Singha, Gunnar Spreen, Nils Hutter, Arttu Jutila, and Christian Haas

Abstract. Automated sea ice charting from Synthetic Aperture Radar (SAR) has been researched for more than a decade and still, we are not close to unlocking the full potential of automated solutions in terms of resolution and accuracy. The central complications arise from ground truth data not being readily available in the polar regions. In this paper, we build a dataset from 20 near coincident X-Band SAR acquisitions and as many Airborne Laser Scanner (ALS) measurements from the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC), between October and May. This dataset is then used to assess the accuracy and robustness of five machine learning based approaches, by deriving classes from the freeboard, surface roughness (standard deviation at 0.5 m correlation length) and reflectance. It is shown that there is only a weak correllation of the radar backscatter and the sea ice topography. Accuracies between 40 % and 69 % percent and robustnesses between 68 % and 85 % give a realistic insight into modern classifiers' performance across a range of ice conditions over 8 months. It also marks the first time algorithms are trained entirely with labels from coincident measurements, allowing for a probabilistic class retrieval. The results show that segmentation models able to learn from the class distribution significantly perform pixel-wise classification approaches.

Karl Kortum et al.

Status: open (extended)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on tc-2023-72', Anonymous Referee #1, 16 Aug 2023 reply

Karl Kortum et al.

Karl Kortum et al.


Total article views: 326 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
220 95 11 326 9 7
  • HTML: 220
  • PDF: 95
  • XML: 11
  • Total: 326
  • BibTeX: 9
  • EndNote: 7
Views and downloads (calculated since 03 Jul 2023)
Cumulative views and downloads (calculated since 03 Jul 2023)

Viewed (geographical distribution)

Total article views: 303 (including HTML, PDF, and XML) Thereof 303 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 28 Sep 2023
Short summary
A dataset of 20 radar satellite acquisitions and near simultaneous helicopter-based measurements of the ice topography during an expedition is constructed and used to train a variety of deep learning algorithms. The results show, that the ice types derived directly from the helicopter measurement are harder to retrieve than those from human annotations. Models that can learn from the spatial distribution of measured sea ice classes are shown to have a clear advantage over those that cannot.