Abstract
ABSTRACTDespite their importance for mass-balance estimates and the progress in techniques based on optical and thermal satellite imagery, the mapping of debris-covered glacier boundaries remains a challenging task. Manual corrections hamper regular updates. In this study, we present an automatic approach to delineate glacier outlines using interferometrically derived synthetic aperture radar (InSAR) coherence, slope and morphological operations. InSAR coherence detects the temporally decorrelated surface (e.g. glacial extent) irrespective of its surface type and separates it from the highly coherent surrounding areas. We tested the impact of different processing settings, for example resolution, coherence window size and topographic phase removal, on the quality of the generated outlines. We found minor influence of the topographic phase, but a combination of strong multi-looking during interferogram generation and additional averaging during coherence estimation strongly deteriorated the coherence at the glacier edges. We analysed the performance of X-, C- and L- band radar data. The C-band Sentinel-1 data outlined the glacier boundary with the least misclassifications and a type II error of 0.47% compared with Global Land Ice Measurements from Space inventory data. Our study shows the potential of the Sentinel-1 mission together with our automatic processing chain to provide regular updates for land-terminating glaciers on a large scale.
Highlights
The accurate knowledge of glacier extent is a prerequisite to many glaciological studies,for example volume change estimates (Kääb and others, 2012; Gardelle and others, 2013) and ice dynamic modeling (Huss and Farinotti, 2012; Fürst and others, 2016)
We kept the coherence window size at 15 × 15 pixels and varied the multilooking factor (MLF) during interferogram generation resulting in a different spatial resolution of the output binary mask
A strict dependency on wavelength is not explained by the results derived from Sentinel-1 in SNAP, as the C-band radar frequency lies in the middle of X- and L-band radar data
Summary
The accurate knowledge of glacier extent is a prerequisite to many glaciological studies,for example volume change estimates (Kääb and others, 2012; Gardelle and others, 2013) and ice dynamic modeling (Huss and Farinotti, 2012; Fürst and others, 2016). Projects like the Global Land Ice Measurements from Space (GLIMS) or the Randolph Glacier Inventory (RGI), which aim to continuously provide updated glacier outlines in digital vector format, are of high importance (Ranzi and others, 2004; Pfeffer and others, 2014; Paul and others, 2015) These global datasets are often not consistent over debris-covered glaciers, as observed in this pilot study over Yazgyl Glacier in the Karakoram (Fig. 1), since they have been generated by different operators or methods. Smith and others (2015) improved the algorithm of Paul and others (2004) and coupled multispectral image classification using Landsat ETM+ and OLI data with elevation, slope and velocity thresholds to map debris-covered glacier outlines Their reported misclassification ranged between 2% and 10% of the glacier area. Classifications which are largely based on topography are limited by the availability of highresolution DEM
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
