Synthetic Aperture Radar (SAR) satellite imagery is often used to discriminate ice types and identify glacial ice hazards to marine operations. This 'data-mining' study examined 123 polarimetric SAR variables in 70 RADARSAT-2 images to assess their utility in separating ice islands (large tabular icebergs) from other ocean covers. Difference of means tests identified five SAR variables to separate each ocean cover. Further SAR variables were found using forward-selection in Redundancy Analysis (RDA). RDA also identified incidence angle, air temperature and wind speed as primary confounding factors for all assessed ocean covers. Support Vector Machine classification using five SAR variables was used to develop an ocean cover classification model. This technique competently distinguished open water, first- and multi-year ice, but could not discriminate ice islands. The selected SAR variables, many not previously investigated, warrant further study and the RDA approach showed promise to guide future development of remote sensing classifiers.