TY  - Data
T1  - Sentinel-2 River-Ice Segmentation Dataset for the Full Lena River Main Stem
A1  - Zhang Ze
A1  - guo zi you
A1  - Leonid Gagarin 
A1  - yan qing kai
A1  - jiang xin yue
A1  - ju meng yao
A1  - Andrei Zhang
A1  - Nikolai Torgovkin
DO  - 10.12072/ncdc.hydrology.db7722.2026
PY  - 2026
DA  - 2026-08-26
PB  - National Cryosphere Desert Data Center
AB  - River ice is an important part of the seasonal hydrological process of rivers in cold regions, and high spatial resolution and reusable pixel-level supervision data are still lacking. This dataset is aimed at remote sensing identification of Lena River ice, machine learning training and ice-condition analysis from source to delta. It forms a related product consisting of manual supervision tags, model and reproduction codes, whole river model mapping, and continuous trunk status.The data source is Sentinel-2 Level-2A surface reflectance images from July to November 2024. The images are retrieved through the Microsoft Planetary Computer STAC catalog; river network and trunk spatial indexes are built based on OpenStreetMap data. The processing process includes image retrieval and screening, band alignment, 512 pixel segmentation, manually supervised water/river ice pixel labeling, data segmentation for river section isolation, U-Net training and verification set threshold selection, river-wide reasoning, supplementary image processing, along route inspection and continuous product construction. All 1,000 final supervision masks were confirmed by manual supervision, and category values of 0, 1 and 2 represent ignore/background, water and river ice respectively.The supervision data includes 1,000 512×512 pixel samples from 656 river sections with a spatial resolution of 10 meters. Among them, 640, 160 and 200 training, verification and river isolation test samples are respectively, containing a total of 19,444,786 valid labeled pixels. The reference line from the source to the delta is 4,440.422 kilometers long and has a spatial range of 104.960°-130.167°E and 53.706°-73.158°N. The entire river product covers 3,709 spatial units and 4 observation time windows: open water surface in summer, early freezing, middle freezing, and late freezing. It contains 14,836 unit-time window records, corresponding geographical registration classification grids and river ice fraction gr
DB  - NCDC
UR  - http://www.ncdc.ac.cn/portal/metadata/c574d198-9db0-4155-9ffe-4bec105a2ff8
ER  -