A wealth of research has focused on elucidating the key controls on mass loss from the Greenland and Antarctic ice sheets in response to climate forcing, specifically in relation to the drivers of marine-terminating outlet glacier change. The manual methods traditionally used to monitor change in satellite imagery of marine-terminating outlet glaciers are time-consuming and can be subjective, especially where mélange exists at the terminus. Recent advances in deep learning applied to image processing have created a new frontier in the field of automated delineation of glacier calving fronts. However, there remains a paucity of research on the use of deep learning for pixel-level semantic image classification of outlet glacier environments. Here, we apply and test a two-phase deep learning approach based on a well-established convolutional neural network (CNN) for automated classification of Sentinel-2 satellite imagery. The novel workflow, termed CNN-Supervised Classification (CSC) is adapted to produce multi-class outputs for unseen test imagery of glacial environments containing marine-terminating outlet glaciers in Greenland. Different CNN input parameters and training techniques are tested, with overall F1 scores for resulting classifications reaching up to 94 % for in-sample test data (Helheim Glacier) and 96 % for out-of-sample test data (Jakobshavn Isbrae and Store Glacier), establishing a state of the art in classification of marine-terminating glaciers in Greenland. Predicted calving fronts derived using optimal CSC input parameters have a mean deviation of 56.17 m (5.6 px) and median deviation of 24.7 m (2.5 px) from manually digitised fronts. This demonstrates the transferability and robustness of the deep learning workflow despite complex and seasonally variable imagery. Future research could focus on the integration of deep learning classification workflows with free cloud-based platforms, to efficiently classify imagery and produce datasets for a range of glacial applications without the need for substantial prior experience in coding or deep learning.

An estimated 76% of global stream area is occupied by channels with widths above 30 m. Sentinel‐2 imagery with resolutions of 10 m could supply information about the composition of river corridors at national and global scales. Fuzzy classification models that infer sub‐pixel composition could further be used to compensate for small channel widths imaged at 10 m of spatial resolution. A major challenge to this approach is the acquisition of suitable training data useable in machine learning models that can predict land‐cover type information from image radiance values. In this contribution, we present a method which combines unmanned aerial vehicles (UAVs) and Sentinel‐2 imagery in order to develop a fuzzy classification approach capable of large‐scale investigations. Our approach uses hyperspatial UAV imagery in order to derive high‐resolution class information that can be used to train fuzzy classification models for Sentinel‐2 data where all bands are super‐resolved to a spatial resolution of 10 m. We use a multi‐temporal UAV dataset covering an area of 5.25 km². Using a novel convolutional neural network (CNN) classifier, we predict sub‐pixel membership for Sentinel‐2 pixels in the fluvial corridor as divided into classes of water, vegetation and dry sediment. Our CNN model can predict fuzzy class memberships with median errors from −5% to +3% and mean absolute errors from 10% to 20%. We also show that our CNN fuzzy predictor can be used to predict crisp classes with accuracies from 95.5% to 99.9%. Finally, we use an example to show how a fuzzy CNN model trained with localized UAV data can be applied to longer channel reaches and detect new vegetation growth. We therefore argue that the novel use of UAVs as field validation tools for freely available satellite data can bridge the scale gap between local and regional fluvial studies. © 2020 The Authors. Earth Surface Processes and Landforms published by John Wiley & Sons Ltd