%0 Dataset
%T 10 m Sentinel-2 Semantic Segmentation Dataset for Thermokarst-Associated Lake and Pond Water Surfaces in Arctic Permafrost Lowlands
%J National Cryosphere Desert Data Center
%I National Cryosphere Desert Data Center(www.ncdc.ac.cn)
%U http://www.ncdc.ac.cn/portal/metadata/af618102-22ef-4618-b1a8-66d55a360fd9
%W NCDC
%R 10.12072/ncdc.permafrost.db7645.2026
%A Zhang Ze
%A guo zi you
%A yan qing kai
%A zhai jin bang
%A Leonid Gagarin 
%A Nikolai Torgovkin
%A Andrei Zhang
%K Keywords permafrost;hot melt lake and pond;semantic segmentation
%X This dataset supports supervised semantic segmentation of visible lake and pond water surfaces in thermokarst-affected Arctic permafrost lowlands. It covers 23 selected study areas across Alaska, northwestern Canada, European Russia, Yamal, and Siberia and is derived from summer 2023 Sentinel-2 L2A surface-reflectance median composites. Each sample contains a five-band image, a binary water-surface label, and a QA mask on a 10 m analysis grid with a tile size of 256 by 256 pixels. The dataset contains 5513 image-label-QA triplets, including 3982 training, 739 validation, and 792 test samples. Positive pixels account for 8.07 % of all pixels and correspond to a pixel-count-equivalent area of approximately 2914.2 km². The supervised target is visible summer lake and pond water, while thermokarst relevance and river, channel, or coastal-water risk are recorded separately as tile-level geomorphic interpretation fields. The package provides fixed benchmark splits, geographic holdout definitions, quality-review samples, a hard-negative subset, temporal-source and sampling-bias summaries, processing configurations, baseline results, and SHA256 checksums. It supports model training, geographic-transfer testing, error analysis, and reproducible algorithm comparison.