| Summary: | Over recent years Greenland Ice Sheet (GrIS) has experienced increased mass loss due to reductions in albedo. South West GrIS (SW GrIS) has experienced the largest declines in albedo with a dark ice band appearing every melt season over recent years. Past studies have used Moderate-Resolution Imaging Spectroradiometer (MODIS) imagery to quantify dark ice spatial changes on SW GrIS. However, there is still a limited understanding of dark ice dynamics and the contribution dark ice can have on lowering GrIS albedo. This study aims to quantify dark ice spatial changes on SW GrIS using Sentinel-2’s high spatial resolution imagery and an unsupervised machine learning classification method for the first time. This study uses Sentinel-2 and MODIS satellite imagery for June, July and August (JJA) of 2016/2017 as well as meteorological and field spectra data. Two classification methods were used: the traditional reflectance threshold method and an unsupervised machine learning method. This study shows that Sentinel-2 is more accurate than MODIS at quantifying dark ice throughout the melt seasons at regional and local scales, with significant correlations found between Sentinel-2 dark ice reflectance and field spectra light (R2=0.49, P<0.01) and heavy algal blooms (R2=0.45, P<0.01). Additionally, both classification methods produced similar results with overall accuracies more than 70% between Sentinel-2 and MODIS dark ice pixels. Machine learning has the potential to classify different sources of dark ice: however, this needs to be explored further.
|
|---|