Supraglacial Lakes on Debris-Covered Glaciers in the Karakoram: A Reference Dataset for Deep Learning Models

This is a supraglacial lake dataset based on Sentinel-2 imagery for the Karakoram region. Specifically, the dataset contains supraglacial lake labels for the fall and spring seasons for both Baltoro Glacier (10,584 labels) for the years 2017-2022 and Hispar Glacier (1,821 labels) for the year 2024. These labels were created with the intent to train a deep learning model to segment supraglacial lakes from Sentinel-2 images. The motivation for creating this dataset was due to a lack of openly available datasets to train machine learning models for the specific purpose of monitoring supraglacial lakes in mountainous environments. Therefore, the provided dataset can either be used to train a machine learning model to identify supraglacial lakes from Sentinel-2 imagery or it can be used as a reference dataset to evaluate models trained on other glacial lake datasets. The supraglacial lake labels are provided as shapefiles which can be adjusted to the individual user's needs and then further processed into binary masks and NumPy arrays that can be used as input to a machine learning model. However, the specific training, validation, and test dataset split that were used to train and evlauate this study's model are also provided. This includes 256x256 raster binary mask patches for training, validation, and test datasets. The NumPy arrays (masks and images) for each of these datasets are also provided for immediate use in a machine learning model. More detailed information on the dataset and its contents as well as the Sentinel-2 data used to create the labels can be found in the metadata file.