Utilizing the dynamic behavior of sea ice for determining ice thickness distributions in SAR imagery

Addresses the issue of combining time sequences of low-resolution ERS-1 SAR data and tracking techniques to produce sea ice thickness maps having high certainty and low error rates. Dynamic, local thresholding and expert systems can be used to estimate initial sea ice thicknesses from a single SAR image. Next, significant features such as floes and leads in the image which have an initial ice thickness estimate of low certainty can be identified. Given these features, the authors can use neural networks to track them through a temporal sequence of SAR imagery. By observing the changes in these features over time and combining these observations with other information (qualitative models of sea ice behavior, temperature and wind information, ice growth models, etc.), they can determine the correct ice thicknesses for all features. Because they concentrate on specific observable sea ice features (floes, leads, etc.), and on specific observable behaviors of these features, they believe they should be able to study directly surface melt, ridging, lead freeze and melt, polynya formation, average size and perimeter of ice floes, and other phenomena on a small scale, in addition to generating estimations of ice thickness distributions of high certainty