Multitemporal Sentinel-2 Patches (64×64) for Flood Mapping - Training and Validation Set, Jacuí River Basin, Rio Grande do Sul, Brazil

The acquisitions used are: three pre-disaster images (December 17, 2023; January 10, 2024; February 21, 2024) and one post-disaster image (May 6, 2024).It is designed for training and validation of deep learning models for flood semantic segmentation.  4-time setup: 3 pre-flood + 1 post-flood acquisitionIn this dataset, the post-flood acquisition and flood mask correspond to the same reference date (May 6, 2024).Data Structure  images/:  1,598 patches, shape (64, 64), each with 12 channels (4 time steps × 3 bands: Green, Red, NIR)  masks/:  1,598 corresponding flood masks, shape (64, 64), values: 0 = non-flooded, 1 = floodedAll filenames are synchronized between images and masks.Patch counts:| Total | 1,598 |Spatial Reference  CRS: EPSG:32722 (WGS 84 / UTM zone 22S)  Resolution: 10 m/pixel  Location: Jacuí River Basin, Rio Grande do Sul, Brazil