| Summary: | Antarctica’s coastline is constantly changing due to fluctuations in glacier and ice shelf front positions. Changes in ice front position can influence ice sheet discharge as retreating fronts may enhance ice sheet flow through reduced buttressing effects. To investigate changes in the ice front position it is important to continuously map the Antarctic coastline. The traditional approach of manual delineation is too time-consuming to cope with the temporal and spatial abundance of satellite data nowadays. Automation of ice front delineation is a very challenging task as conventional edge detection methods fail due to the very low contrast between shelf and sea ice. Therefore, we apply novel imaging processing techniques by using the neural network architecture ‘U-Net’. In combination with pre- and post-processing techniques we are able to automatically extract glacier and ice shelf fronts from Sentinel-1 data and map changes in their positions. The workflow includes (1) pre-processing of Seninel-1 data, (2) adding elevation information with the TanDEM-X digital elevation model, (3) normalization & tiling, (4) classification into ocean and land ice via deep learning and (5) a final post-processing step to generate shapefiles form the classification results. Our approach generates good classification results throughout the year for our training and test areas. Only summer months with extensive surface melt can be challenging. Compared to a manually delineated coastline the average accuracy can vary between 2-3 pixels in stable and clear coastline areas and up to 7 pixels in more difficult regions (e.g. mélange, melt, fast ice). With our SAR and DEM-based approach, we show that it is not only possible to map single glacier fronts but to automatically map entire coastal sections of Antarctica during various seasons. The created time-series of calving front changes may allow future studies on the relationship between glacier flow and buttressing effects of ice shelves and glacier tongues.
|
|---|