Extracting Sea Ice Data from Satellite SAR Imagery

Abstract

With the prospect of operational satellite SAR's by the end of the decade, there is a clear need to develop automated algorithms for the extraction of geophysical data about sea ice from highresolution radar imagery. To this end, we have developed techniques for distinguishing ice from open water and for resolving the details of deformation within areas 100 km square imaged by Seasat SAR. The classification of ice and open water is based on the creation of a second band of image data consisting of the local variance of the original brightness, the first band being a local average brightness. In the space of these two variables, ice and open water are separated into two distinct clusters. The deformation is found on a 3.4-km mesh by local cross correlations of the brightness. The strategy is to find a coarse displacement field with a highly averaged image, and to proceed through a hierarchy of images with increasing resolution, improving the accuracy of the displacements at each step. Comparison with manually determined displacement shows room for improvement in regions of high deformation by using smaller areas for cross correlation. The concentration and deformation data are used together to determine localized regions of the scene where open water is produced or lost.