SEA ICE LEAD DETECTION USING DEEP LEARNING
Sea ice leads, recurrent and elongated patches of open water, are among the most distinguishing characteristics of sea ice cover. Leads act as a medium between the atmosphere above and seawater below, making it paramount to detect and understand them. Using the U-Net deep neural network, we propose a deep learning-based approach for sea ice lead detection from visible-band WorldView and C-band Synthetic Aperture Radar Sentinel images. We detect the boundaries of these leads due to their clear visual distinction from the neighboring sea ice. Furthermore, we show that our approach detects leads from multiple modalities using small amounts of training data.