Machine Learning Ready Dataset for Glacier Crevasse Detection in Mountain Regions

This dataset provides a manually annotated training and test set for mountain glacier crevasse detection. It is based on RGB orthophotos (8-bit) from Land Tirol, acquired over the Ötztal and Stubai Alps in 2019/2020 at a spatial resolution of 0.2 m. The training and test regions were selected to capture as wide a range of glacier and crevasse morphologies as possible, spanning different surface types (bare ice, snow, and debris). The imagery was masked using glacier boundaries from the Austrian Federal Office for Metrology and Surveying (BEV, Bundesamt für Eich- und Vermessungswesen) so that only glaciated areas are retained. Crevasses were manually annotated as line shapefiles tracing the outer crevasse boundary. These shapefiles were rasterized into single-pixel-width masks aligned to the original orthophotos (EPSG:31254). Please note that crevasse mapping and annotation are subjective tasks, and annotations may differ between this and other datasets. Images and masks vary in size and must be tiled before use; we used 512 × 512 pixel tiles in our study. The data is structured as follows: ├── train/ (14 scenes)│ ├── image/│ └── mask/├── test/ (6 scenes)│ ├── image/│ └── mask/ When using this dataset, please cite the dataset itself and one of the following studies for which this dataset was created: Baumhoer, C.A., Leibrock, S., Zapf, C., Beer, W., Kuenzer, C., 2025. Automated crevasse mapping for Alpine glaciers: A multitask deep neural network approach. International Journal of Applied Earth Observation and Geoinformation 139, 104495. https://doi.org/10.1016/j.jag.2025.104495