Search

The continuous and precise mapping of glacier calving fronts is essential for monitoring and understanding rapid glacier changes in Antarctica and Greenland, which have the potential for significant sea level rise within the current century. This effort has been mostly restricted to the slow and painstaking manual digitalization of the calving front positions in thousands of satellite imagery products. Here, we have developed a machine learning toolkit to automatically detect glacier calving front margins in satellite imagery. The toolkit is based on semantic image segmentation using Convolutional Neural Networks (CNN) with a modified U-Net architecture to isolate the calving fronts from satellite images after having been trained with a dataset of images and their corresponding manually-determined calving fronts. As a case study we train our neural network on a varied set of Landsat images with lowered resolutions from Jakobshavn, Sverdrup, and Kangerlussuaq glaciers, Greenland and test the results on images from Helheim glacier, Greenland to evaluate the performance of the approach. The neural network is able to identify the calving front in new images with a mean deviation of 96.3 m from the true fronts, equivalent to 1.97 pixels on average, while the corresponding error for manually-determined fronts on the same resolution images is 92.5 m (1.89 pixels). We find that the trained neural network significantly outperforms common edge detection techniques, and can be used to continuously map out calving-ice fronts with a variety of data products.

Since 2002, global snow water equivalent (SWE) estimates have been generated using Advanced Microwave Scanning Radiometer (AMSR-E)/Aqua data. Accurate estimates of SWE are important to improve monitoring and managing of water resources in specific regions. SWE and snow map product accuracy are functions of topography and of land cover type because landscape characteristics have a strong influence on redistribution and physical properties of snow cover, and influence the microwave properties of the surface. Here we evaluate the AMSR-E SWE and derived snow map products in the Mackenzie River Basin (MRB), Canada, which is characterized by complex topography and varying land cover types from tundra to boreal forest. We compare in situ snow depth observations and Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover maps from January 2003 to December 2007 with passive microwave remotely sensed SWE from AMSR-E and derived snow cover maps. In the MRB the mean absolute error ranges from 12 mm in the early winter season to 50 mm in the late winter season and overestimations of snow cover maps based on a 1 mm threshold of AMSR-E SWE varies from 4% to 8%. The optimal threshold for AMSR-E SWE to classify the pixels as snow ranges from 6 mm to 9 mm. The overall accuracy of new snow cover maps from AMSR-E varies from 91% to 94% in different sub-basins in the MRB.