Method for detection of leads from Sentinel-1 SAR images

• Published in: Annals of glaciology Vol. 59; no. 76pt2; pp. 124 - 136
• Main Authors: Murashkin, Dmitrii, Spreen, Gunnar, Huntemann, Marcus, Dierking, Wolfgang
• Format: Journal Article
• Language: English
• Published: Cambridge, UK Cambridge University Press 01.07.2018; Cambridge University Press (CUP)
• Subjects: Algorithms; Arctic sea ice; Classification; Detection; Earth science; Fluxes; Freezing; Geofag: 450; Geosciences: 450; Glaciers; Growth rate; Ice; Ice cover; Ice environments; ice/atmosphere interactions; ice/ocean interactions; Image detection; Kvartærgeologi, glasiologi: 465; Matematikk og Naturvitenskap: 400; Mathematical models; Mathematics and natural science: 400; Methods; Neural networks; Open access; Quality; Quaternary geology, glaciology: 465; Radar; Radar imaging; Radar polarimetry; remote sensing; SAR (radar); Satellite data; Satellites; Sea ice; sea-ice dynamics; Studies; Synthetic aperture radar; VDP; Arctic Ocean; Arctic region; ice/atmosphere interactions; remote sensing; sea-ice dynamics; ice/ocean interactions; sea ice
• ISSN: 0260-3055; 1727-5644; 1727-5644
• DOI: 10.1017/aog.2018.6

The presence of leads with open water or thin ice is an important feature of the Arctic sea ice cover. Leads regulate the heat, gas and moisture fluxes between the ocean and atmosphere and are areas of high ice growth rates during periods of freezing conditions. Here, an algorithm providing an automatic lead detection based on synthetic aperture radar images is described that can be applied to a wide range of Sentinel-1 scenes. By using both the HH and the HV channels instead of single co-polarised observations the algorithm is able to classify more leads correctly. The lead classification algorithm is based on polarimetric features and textural features derived from the grey-level co-occurrence matrix. The Random Forest classifier is used to investigate the importance of the individual features for lead detection. The precision–recall curve representing the quality of the classification is used to define threshold for a binary lead/sea ice classification. The algorithm is able to produce a lead classification with more that 90% precision with 60% of all leads classified. The precision can be increased by the cost of the amount of leads detected. Results are evaluated based on comparisons with Sentinel-2 optical satellite data.