Glacial-Lake-Bench: A Global Multi-Sensor Benchmark Dataset for Evaluating Deep Learning Models for Glacial Lake Mapping

This dataset (Glacial Lake-Bench) consists of multi-source remote sensing image–mask pairs collected globally. Each image contains 11 channels: Blue, Green, Red, NIR, SWIR1, SWIR2, NDWI, Slope, Elevation, VV, and VH. The corresponding mask uses pixel values of 0 for background and 1 for lake pixels. This dataset consists of two components: Glacial-Lake-Bench (GLB):A globally sampled dataset designed to train and evaluate deep learning models for glacial lake mapping. The primary goal is to establish a globally scalable approach. We strongly encourage using leave-one-region-out cross-validation to assess generalization across diverse geographic regions.Naming Convention: Naming Convention:The first initials represent the RGI regions. For example: AC = Arctic Canada ACS = Arctic Canada South AK = Alaska CA = Central Asia GL = Greenland RA = Russian Arctic SA = Southern Andes SAW = South Asia West SC = Scandinavia SV = Svalbard WC = Western Canada CE = Central Europe LL = Lower Latitudes ME = Caucasus and Middle East NA = North Asia NZ = New Zealand IS = Iceland SEE = South East Asia Glacial-Lake-Challenge (GLC):A curated dataset intended to benchmark and test advanced deep learning algorithms under challenging conditions such as cloud cover, shadows, small lakes, frozen lakes, and other complex scenarios.Please cite our work:Saurabh Kaushik, Beth Tellman, Ian Howat, Umesh Haritashya, Lalit Maurya. “Glacial-Lake-Bench: A Global Multi-Sensor Benchmark Dataset for Evaluating Deep Learning Models for Glacial Lake Mapping”https://zenodo.org/records/17917359