Sentinel-1 SAR based sea ice classification for winter season

Abstract

Sentinel-1A and 1B operate in Extra Wide swath dual-polarization mode over the Arctic Seas, and the two-satellite constellation provides the most frequent SAR observation of the Arctic sea ice ever. The cross-polarization is known to be more sensitive to difference in the scatter of sea ice and open water than the co-polarization, and the combination of HH- and HV-polarizations has been widely used for ice edge detection and ice type classification. However, the majority of the ice classification algorithms were developed using RADARSAT-2 ScanSAR images which has different sensor characteristics from Sentinel-1 TOPSAR, and the use of Sentinel-1 for the same purpose is very limited in literature. The main drawback of applying existing algorithms to Sentiel-1 data is the relatively higher level of thermal noise contamination and its propagation to image textures. We developed an efficient noise correction method by adopting a novel concept of noise equivalent standard deviation and compensating for the multiplicative textural noise along with the additive noise equivalent sigma nought. The denoised images are processed into Haralick texture features and a machine learning-based classifier is trained by feeding digitized ice charts. The use of ice charts rather than manually classified images makes enable an automated training which minimizes the effects from biased human decision. The resulting classifier is tested over the Fram Strait area for an extensive dataset of Sentinel-1 acquired from October 2017 to May 2018. The classification results will be shown in comparison with both the ice charts and manually classified images, and the feasibility of the ice chart-feeded automated classifier will be discussed.