Spices – Sea Ice Edge Maps From The Fram Strait

Sea ice edge maps derived from Sentinel-1 SAR dual polarisation EW images using a Support Vector Machine (SVM) algorithm. Thisalgorithm is based on a SVM approach, and in addition uses texture calculation and principal component analysis (PCA) to classify sea ice types (Korosov et al., 2016). The main steps of the algorithms include:<br> (1) pre-processing of the raw SAR data, <br> (2) calculation of texture features, <br> (3) unsupervised pre-classification of the image using PCA and k-means cluster analysis to reduce the number of ice classes, <br> (4) expert re-classification of the image into the pre-calculated classes, <br> (5) training of the SVM using input from the previous step, and <br> (6) classifying the full image into the reduced number of classes using the trained SVM. <br> To generate an ice edge product, the SVM algorithm is used with only two classes: sea ice and open water. Korosov, A., N. Zakhvatkina, A. Vesman, A. Mushta, and S. Muckenhuber, Sea ice classification algorithm for Sentinel-1 images, Poster at ESA Living Planet Symposium 2016, Prague, Czech Republic, 9-13 may, 2016.