A Multi-Label Sentinel-2 Dataset for Urban Green and Blue Space Classification

This dataset contains 10,000 Sentinel-2 true-color image chips for multi-label classification of green and blue spaces across Germany. Each image is annotated independently with four binary labels: urban park, urban forest, urban water, and allotment gardens. These labels result in 16 possible combinations, including images containing none of the four categories. The dataset is fully balanced and comprises 625 image chips for each label combination. Each image is provided as a 224 × 224 pixel RGB PNG file and represents a 512 m × 512 m geographic footprint. Label annotations were derived from OpenStreetMap polygon geometries and recomputed for the complete geographic footprint of every image. The urban-park category includes publicly accessible areas tagged leisure=park. Parks tagged access=private or access=no were excluded. Individual playgrounds, cemeteries, and traffic greenery were not included in the park definition. The urban-forest category includes woodland polygons tagged natural=wood or landuse=forest. Individual trees, tree rows, and small woodland groups were excluded from this category. The urban-water category includes polygons tagged natural=water without a more specific water value or in combination with water=lake, water=pond, water=reservoir, water=river, or water=canal. Polygonal water features tagged waterway=river, waterway=canal, or waterway=riverbank were also included. Swimming pools, fountains, basins, wastewater or sewage features, technical pools, wastewater facilities, and storage tanks were excluded. Linear or culverted waterways without a mapped surface polygon were likewise not included. The allotment-gardens category is based on polygons tagged landuse=allotments. Private residential gardens, residential areas containing gardens, farmland, and plant nurseries were excluded from this label definition. The four labels are not mutually exclusive. A single image may contain any combination of parks, forests, surface water, and allotment gardens and receives all applicable labels. Positive samples were selected separately for each of the 15 non-empty label combinations. Source locations were sampled from retained OpenStreetMap polygons, spatially distributed across Germany using a round-robin procedure, and labeled according to the features occurring within their complete 512 m × 512 m footprints. Images without any of the four labels were generated from neighboring locations in the surroundings of selected positive samples. These images did not receive any of the four labels under the dataset definition. The absence of a label denotes the absence of the corresponding retained OpenStreetMap category; it does not necessarily imply the complete absence of vegetation, water, gardens, or other open spaces. The term urban green-blue infrastructure describes the intended mapping and classification context. Sample locations are distributed across Germany, and no separate administrative or morphological urban-area mask was imposed. The labels therefore represent the defined OpenStreetMap feature categories and do not guarantee that every image lies within a formally delineated urban area. Image chips were generated from Sentinel-2 Level-2A imagery using bands B04, B03, and B02 as red, green, and blue channels. Imagery was retrieved through the Copernicus Data Space Ecosystem Sentinel Hub Process API for the period from July 13, 2025, to July 12, 2026. The collection was generated using a maximum cloud-cover threshold of 10 percent and least-cloud-cover mosaicking. The repository includes the complete image collection and a compact CSV table containing image filenames and the four binary labels. The dataset can be used for supervised multi-label image classification, benchmarking of computer vision models, representation learning, urban green-space and surface-water mapping, environmental monitoring, and research on green-blue infrastructure.