Classification of sea ice types for the East part of Greenland waters using SENTINEL 1 data

• Main Author: Gkanatsios, Ioannis
• Other Authors: Gourmelen, Noel, Woodhouse, Iain
• Format: Master Thesis
• Language: English
• Published: The University of Edinburgh 2018
• Subjects: sea ice; SAR; texture analysis; SVM classifier; Greenland; Sea ice
• Online Access: http://hdl.handle.net/1842/35521

Ships navigate in Greenland waters all year round. Cruises to Greenland due to tourism and educational purposes have increased the last decade. Hence, it is essential for ships that navigate through Sea Ice in winter to use reliable and accurate information on sea ice conditions. An accurate classification of Sea Ice types is an ongoing problem. Many classification algorithms depend only on the SAR image intensity for discriminating the sea ice types. Different Sea Ice types exhibit similar backscatter signature which makes the algorithm unable to correctly classify them. In this study, two dual-polarization SENTINEL-1 images with a spatial resolution of 40 x 40m acquired over the East part of Greenland in February and May of 2016. Support Vector Machine (SVM) classifier was used to perform the classification. In order to improve the discrimination of ice types, texture analysis was performed using Grey Level Co-occurrence Matrix (GLCM) algorithm. Ten GLCM texture features were calculated. The analysis revealed that the most informative texture features for the sea ice classification are Energy, mean, dissimilarity for HV polarization and Angular Second Moment, variance and energy for HH polarization. The classification results for the SAR images acquired during winter and spring period were compared against the sea ice charts produced by DMI. A good agreement between the classification results and validation data is shown. The results show that the overall classification accuracy for both SAR images amount to 91%. The most hazardous for ships navigation sea ice types (old ice and deformed first year ice) have been successfully discriminated.