{"success":true,"data":{"author":"Image and Signal Processing group, University of Valencia (ESA OpenSR)","bucketAccessType":"public","bucketName":"isp-uv-es/SEN2NEON","bucketPrefix":"","category":"Earth Observation & Geospatial","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:15.828Z","description":"Benchmark of paired low- and high-resolution multispectral tiles for validating super-resolution models on Sentinel-2. The low-resolution input is observed Sentinel-2 Level-2A surface reflectance at 10 m (12 bands, 256 x 256 pixels); the high-resolution references are 12-band products derived from NEON AVIRIS-NG airborne hyperspectral imagery and linearised to the Sentinel-2 bands, at 2.5 m (the canonical 4x reference evaluated in the paper, 1024 x 1024) and a supplementary 1 m (10x). It contains 2,269 aligned tile ids over NEON field sites in the United States, in projected UTM zones, as a single validation split rather than a training corpus. The listing (measured 2026-11-14) shows s2_l2a_10m/, neon_2.5m_linearized/ and neon_1m_linearized/ GeoTIFF folders with metadata.parquet, .jsonl and .csv indexes, a SHA-256 manifest and release notes; the repository uses about 386 GB. Load the index with Hugging Face datasets or pandas and open tiles with rasterio; validation code is in the ESAOpenSR/SEN2NEON GitHub repository. Distributed under CC-BY-4.0.","displayTitle":"SEN2NEON: Sentinel-2 to NEON Hyperspectral Multispectral Super-Resolution Benchmark","id":"656f0e02-a478-502f-b777-6a9524980854","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/isp-uv-es/SEN2NEON","objectCountEstimate":7607,"probeStatus":"ok","referenceUrl":"https://github.com/ESAOpenSR/SEN2NEON","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"isp-uv-es/SEN2NEON\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"isp-uv-es/SEN2NEON\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=isp-uv-es, license:cc-by-4.0, size_categories:1K<n<10K, format:parquet, downloads=294"],"format":"mixed","generatedAt":"2026-11-14T14:15:18.021Z","layout":{"example":"metadata.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"id,name,split,lr,hr,hr_2_5m_path,hr_1m_path,hr_available_resolutions_m,hr_canonical_resolution_m,lon,lat,LC_detail_id,LC_detail_text,LC_superclass_id,LC_superclass_text,bands,native_band_resolution_m,stored_dtype,reflectance_scale,lr_source,lr_collection,lr_grid,lr_native_resolution,lr_width,lr_height,lr_pixel_size_m,lr_nodata,hr_source,hr_width,hr_height,hr_pixel_size_m,hr_nodata,crs,lr_transform,hr_transform,centroid_lon,centroid_lat,land_cover,land_cover_detail_id,land_cover_detail,land_cover_superclass_","encoding":"utf8","key":"metadata.csv","truncatedAt":512}],"sampleObjects":["DATASET_RELEASE_NOTES.md","assets/sen2neon_banner.png","metadata.csv","metadata.jsonl","metadata.parquet"],"sidecars":[{"bytes":"---\npretty_name: SEN2NEON\nlicense: cc-by-4.0\ntags:\n  - remote-sensing\n  - geospatial\n  - sentinel-2\n  - super-resolution\n  - multispectral\n  - earth-observation\ntask_categories:\n  - image-to-image\ntask_ids:\n  - super-resolution\nannotations_creators:\n  - no-annotation\nlanguage_creators:\n  - other\nlanguage: []\nmultilinguality:\n  - other\nsize_categories:\n  - 1K<n<10K\nconfigs:\n  - config_name: default\n    default: true\n    data_files:\n      - split: validation\n        path: metadata.parquet\n---\n\n# SEN2NEON\nLR/HR Multispectral Super-Resolution Dataset for benchmarking and research.  \n![Example](assets/sen2neon_banner.png)\n\n---\n\n## Dataset Summary\nSEN2NEON provides paired **low-resolution (LR, 10 m)** and **high-resolution (HR)** GeoTIFF tiles for validating super-resolution (SR) models in remote sensing. The LR product contains observed Sentinel-2 Level-2A reflectances; the HR references are derived from NEON AVIRIS-NG hyperspectral acquisitions. Each product shares the same spatial footprint and is pixel-aligned. Companion Parquet/JSONL/CSV indexes provide paths for both HR resolutions plus per-tile provenance, acquisition timing, geometry, cloud/nodata statistics, and land-cover labels for **stratified evaluation** (e.g., Forest vs Built-up).\n\nThe code, examples, and validation workflows can be found at [ESAOpenSR/SEN2NEON](https://github.com/ESAOpenSR/SEN2NEON).\n\n> **Release correction — 20 August 2026:** A previous dataset upload mistakenly provided a linearized 10 m NEON product as `lr`. This was corrected on 20 August 2026. The canonical LR files are now in `s2_l2a_10m/` and contain the original Sentinel-2 observations exported through Google Earth Engine. The incorrect product is excluded from this release. The quantitative LR/HR consistency values reported in the paper were calculated using the real Sentinel-2 values.\n\n> **HR resolution:** The **2.5 m HR product is the canonical SEN2NEON reference described and evaluated in the paper**. It remains the default in the metadata and companion code. We additionally provide aligned 1 m HR tiles because they are produced by our processing workflow and may be useful to others. The 1 m product is supplementary and does not redefine the published benchmark or its reported results.\n\n- **Modality:** 12-band multispectral GeoTIFFs\n- **Tasks:** Super-resolution (image-to-image), benchmarking, and analysis\n- **Scale:** 2,269 aligned tile IDs with one LR and two HR products\n- **Alignment:** Canonical 2.5 m HR is 4× LR; supplementary 1 m HR is 10× LR; both are integer, isotropic, and pixel-aligned\n- **Geo:** Projected UTM zones per tile; WGS84 centroids are included in the metadata\n- **Split:** One `validation` split; this is a benchmark, not a globally representative training corpus\n\n---\n\n## Repository Layout\n```text\n.\n├── README.md\n├── DATASET_RELEASE_NOTES.md\n├── assets/\n│   └── sen2neon_banner.png\n├── metadata.parquet             # default HF Dataset/Viewer index\n├── metadata.jsonl               # equivalent line-delimited JSON index\n├── metadata.csv                 # equivalent tabular index\n├── s2_l2a_10m.sha256            # SHA-256 manifest for LR tiles\n├── s2_l2a_10m/                  # observed Sentinel-2 LR, 12×256×256\n├── neon_2.5m_linearized/        # canonical paper HR, 12×1024×1024\n└── neon_1m_linearized/          # supplementary workflow HR, 12×2560×2560\n```\n\nRelative paths in all metadata indexes match the on-Hub layout. The Parquet index is the default source for the Hugging Face Dataset Viewer and `load_dataset`; JSONL and CSV are retained as portable equivalents.\n\n---\n\n## Record Schema\nEach record describes one aligned tile and its LR/canonical/supplementary HR paths. Core fields are:\n\n```json\n{\n  \"id\": \"<stem>\",\n  \"name\": \"<filename.tif>\",\n  \"split\": \"validation\",\n  \"lr\": \"s2_l2a_10m/<file>.tif\",\n  \"hr\": \"neon_2.5m_linearized/<file>.tif\",\n  \"hr_2_5m_path\": \"neon_2.5m_linearized/<file>.tif\",\n  \"hr_1m_path\": \"neon_1m_linearized/<file>.tif\",\n  \"hr_available_resolutions_m\": [2.5, 1.0],\n  \"hr_","encoding":"utf8","key":"README.md","name":"README.md","size":11429,"truncated":true}],"sourceObjectKey":"metadata.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/isp-uv-es/SEN2NEON/metadata.csv\")\ndf.head()","starterSamplePath":"assets/sen2neon_banner.png","totalObjects":7607},"summary":"Benchmark of paired low- and high-resolution multispectral tiles for validating super-resolution models on Sentinel-2.","title":"SEN2NEON: Sentinel-2 to NEON Hyperspectral Multispectral Super-Resolution Benchmark","updatedAt":"2026-11-14T17:41:11.624Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/isp-uv-es/SEN2NEON"}}