Sar Remote Sensing of Sea Ice: Towards automatic extraction of geophysical information

Synthetic Aperture Radar (SAR) images from the ERS-1 satellite have great potential for sea ice remote sensing due to its independence of daylight and clouds. The very high data rate of SAR (over 100 million bits per second) results in an enormous volume of data that threatens the ability of human interpreters. This thesis aims at developing advanced techniques for automatic extracting sea ice information from SAR images, which has been considered as a vital task to realistically benefit from the powerful SAR technique. Two major processing applications, sea ice classification, and sea ice motion detection are developed in the present thesis in order to retrieve the steady and dynamic sea ice information. Based on this information various geophysical parameters of sea ice can be determined. A problem common to all SAR images is the grainy appearance due to noise known as speckle. It is different from the noise in optical images so that methods for processing SAR images have to pay special concern to speckle. For sea ice classification two approaches, pixel-based and segment-based, are investigated. The segment-based approach includes an additional processing, image segmentation, in order to isolate homogenous regions that offer statistical estimations based on a relative large number of image pixels. Accordingly, it is superior in reduction of the speckle noise to the pixel-based approach. The derivation of dynamic information of sea ice is more complex in the sense that the input is a time sequence of SAR images. A great effort has been made here to deal with rotational and deformational ice motion, and to save computation time. The algorithm is based on the optical flow calculation. The results show that the optical flow method has larger capacity to cope with the rotation and deformation of an ice cover, and requires less computing time than the area correlation method that is commonly used for ice motion detection. For better representation of the discontinuities of a motion field, a segment-oriented method has been developed with the aid of image segmentation to refine the raw estimates of motion vectors. In addition a fast correlation method based on FFT transformation is investigated to detect the first order motion information, i.e. mean rotation and translation. This method can be used alone to get a quick estimate of an ice motion field, or be used in combination with the optical flow method to simplify and speed up the whole process of motion retrieval. For applications, both methods, the optical flow and the fast correlation, can provide very useful information on ice dynamics for sea ice modelling which is important to characterise sea ice condition for environmental/climate aspects as well as for ship navigation applications. The information obtained by the methods have also shown a potential to unravel some oceanographic phenomena