Polarimetric Decomposition Analysis of Sea Ice Data

This paper explores polarimetric properties of Arctic sea ice. The analysis is performed in several steps. First, the SAR image is segmented into distinct classes using an unsupervised classification algorithm, which incorporates both polarimetric and statistical signal features. The algorithm has built in contextual smoothing through Markov Random Field modeling. The image segments are subsequently analyzed by sea ice experts, which based on the SAR data and all available in-situ information gives their opinion about ice type and ice properties. Next, a polarimetric analysis is performed to characterize the segments in terms of polarimetric properties. The analysis clearly shows that areas of thin ice and open water are easily detected. The polarimetry also enables discrimination between smooth surfaces and more deformed ice.