Sea Ice Classification using Synthetic Aperture Radar data

Satellite radar systems have an important ability to observe the earth's surface, independently from cloud and light condition, which is very useful in Arctic during polar night and severe weather condition. For navigation support, industrial activities o n the shelf and other operations in the ice covered seas, it is necessary to give the operational information in the convenient form for users - ice charts. Synthetic Aperture Radar (SAR) images are used for sea ice classification. For fast and objective reconstruction of sea ice types from SAR images it is necessary to develop methods of automated processing and classification. In this project we aim at using a multilayer feed forward Neural Network (NN) algorithm for the Barents Sea and Fram Strait area . The algorithm uses backscatter and extracted SAR image texture features in order to identify several sea ice classes. To develop the NN algorithms for automatic derivation of the main sea ice parameters (ice type and concentration) we used ice visual in terpretation of SAR image by ice experts, for training. The optimal neural network's architecture is determined by using a set of textural characteristics for a predefined number of classes (sea ice types and sea ice/open water). The additional informatio n from AMSR-E is used for automated distinguishing of sea ice and open water. The opportunity to extend the number of sea ice classes by using the Alternating Polarization Mode data will also be investigated. The classification results are validated using ground truth data from field observations and ice charts produced by experts.