Sentinel-1 SAR Coastal Water/Land Segmentation Dataset

This dataset contains 792 co-registered Sentinel-1 Ground Range Detected synthetic aperture radar samples for coastal water/land segmentation. Each sample includes VV and VH backscatter GeoTIFFs and a corresponding per-acquisition water/land label on the same raster grid. The dataset contains 679 training samples from 39 sites and 113 validation samples from 22 geographically disjoint sites. Initial reference masks were generated from ESA WorldCover 2021 v200 permanent-water class 80. Site-level Random Forest classifiers used the reference masks and multiscale features from VV, VH, and VV−VH composites to generate acquisition-specific labels, followed by manual correction of problematic labels. The release includes the original WorldCover-derived site masks, labels, tifs, per-scene metadata, manifests, processing documentation, and the scripts used to construct the labels.