Training, testing and intercomparison data for the segmentation of supraglacial lakes in Antarctica

This dataset complements the publication "Advancing Supraglacial Lake Mapping in Antarctica: A Deep Learning Approach Using Sentinel-2 Imagery" by Koehler et al., submitted. The dataset contains three files: SGL-S2-train-test-data.zip: Manually segmented supraglacial lake extents for deep learning based segmentation. The labels are split into training and test data as described in the publication. Description of features within the labels and the corresponding Sentinel-2 scenes can be found in the Excel sheet. SGL-intercomparison.zip: This folder contains the maximum extent of supraglacial lakes over the Amery and Bach ice shelves in January 2017. These extents were derived using a U-Net (Koehler et al.) and a Random Forest (Dirscherl et al. 2020) approach, and were then used to conduct a relative intercomparison of established supraglacial lake extent products. SGL_annotation_rules.pdf: This file contains the rules according to which the training and test labels were annotated.