AutoCoast Baltic Sea Data

Dataset Description Title: AutoCoast - Coastal Change Detection: Sentinel-2 Baltic Sea Image Pairs, Predictions, and Transect Analyses Description: This dataset contains processed Sentinel-2 satellite image pairs, semantic segmentation predictions, certainty maps, and transect-based change analyses for the Baltic Sea coastline. The data was generated as part of a study on automated coastal change detection using deep learning and multi-temporal satellite imagery. Contents: 2017_2018/ and 2023_2024/: RGB/: True-color composite images for each location and year/month. Prediction/: Semantic segmentation results (class predictions) for each image. Certainty/: Per-pixel certainty/confidence maps for each prediction. Change/ (2023_2024 only): Overlay images highlighting detected changes between time periods. meta/: JSON files containing the geographic bounding box (corner coordinates) for each image location. transects/: JSON files with per-transect class counts and certainty values for each time period, as well as computed changes along each transect. File Naming Convention: Files are named as: bs_<longitude>_<latitude>_<year>_<month>.png or .json (e.g., bs_18.56857913358233_54.48658554015609_2023_06.png) Data Structure: Each image location is represented by a pair of time points (e.g., June 2018 and June 2024). For each location and time, the dataset includes the original RGB image, the model’s prediction, and a certainty map. Change overlays and transect analyses provide spatially explicit information on coastal change. Intended Use: This dataset is intended for research on coastal change detection, remote sensing, semantic segmentation, and uncertainty quantification. It can be used for benchmarking, training, or validating machine learning models for environmental monitoring. Format: Images: PNG Metadata and transect results: JSON Contact: For questions or further information, please contact the dataset authors.