We introduce the Sentinel-2 Irish Vegetation Edge (SIVE) Dataset. Its purpose is to aid the development of deep learning methods for detecting the vegetation line in sandy beach environments using Sentinel-2 imagery. It provides a benchmark for evaluating the model on both locations used and not used in training, and to assess its ability to calculate erosion rates. To understand more about how this dataset was created, please see this paper: O’Sullivan, C., [coming soon] training.zip: 8,626 training crops (144 × 144 pixels) with the following channels:1–4: Sentinel-2 spectral bands (Red, Green, Blue, NIR)5: Guidance band (derived)6: Ground-truth vegetation line mask test_seen_location.zip: Uncropped Sentinel-2 scenes from beaches included in the training set, but at different times. Used to evaluate within-site generalisation. Channels:1–4: Sentinel-2 spectral bands (Red, Green, Blue, NIR)5: Guidance band (derived)6: Ground-truth vegetation line mask test_unseen_location.zip: Uncropped Sentinel-2 scenes from beaches not included in the training set. Used to evaluate cross-site generalisation. Channels:1–4: Sentinel-2 spectral bands (Red, Green, Blue, NIR)5: Guidance band (derived)6: Ground-truth vegetation line mask test_erosion.zip: Sentinel-2 scenes from Bull Island (2017, 2019, 2021), selected to allow direct comparison between model-derived vegetation lines and manually annotated lines for estimating erosion rates. Both .npy and .tif files are included for gelocation information. Channels:1–4: Sentinel-2 spectral bands (Red, Green, Blue, NIR)5: Guidance band (derived)6: Ground-truth vegetation line mask