We have developed a novel and hybrid method for sea ice motion tracking that is both dense, providing a drift field with a resolution equal to that of the image, and highly accurate, with a displacement error within 2 pixels, of 74 m.
Analysis of sea ice motion can vary in spatial scale from meters to thousands of kilometers, with each scale useful for some type of analysis. But currently, there is a gap in the understanding of how sea ice motion relates to small-scale processes occurring within the ice field. We try to provide analysis that will allow research to bridge this gap, by providing high resolution, dense, and accurate motion estimates of the sea ice field.
Our algorithm consists of a combined feature tracking and Optical Flow approach, exploiting feature tracking for accuracy and Optical Flow for density.
We have used a sequence of 59 RADARSAT-2 images at 50 m resolution, covering a 40 km x 40 km area, to test our algorithm and 1606 GPS buoy tracks for validation.