Flood Mapping Training Samples from Sentinel-1 and HAND

This dataset contains pixel-level training samples used for developing and validating a deep learning (MLP) model for flood inundation mapping. Samples were derived from two sources: (1) 466 manually labeled image chips from the Sen1Floods11 dataset and (2) 1,624 image chips from an in-house dataset of 104 flood events across the continental United States (CONUS). Each sample represents one pixel, with four key variables: Sentinel-1 VV backscatter, Sentinel-1 VH backscatter, Height Above Nearest Drainage (HAND), and flood status label (0 = non-flooded, 1 = flooded), as well as several auxiliary variables: Country and Chip ID for Sen1Flood11 samples while Case ID and Clip ID for In-House samples.