SSL4EO-L Benchmark: Landsat Land Cover (NLCD) and Crop Type (CDL) Segmentation
- Provider
- Stewart, Lehmann, Corley, Wang et al. (TorchGeo; UIUC, TUM, Microsoft AI for Good)
- License
- CC0-1.0 (source)
- Size
- 13 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/torchgeo/ssl4eo_l_benchmark)
- Access
- Free and public. No account with the data provider is needed.
- DOI
- 10.48550/arXiv.2306.09424
- Category
- Earth Observation & Geospatial
- Source
- https://huggingface.co/datasets/torchgeo/ssl4eo_l_benchmark
About SSL4EO-L Benchmark: Landsat Land Cover (NLCD) and Crop Type (CDL) Segmentation
Semantic segmentation benchmark built for the SSL4EO-L paper (NeurIPS 2023) so that Landsat foundation models can be compared across five sensor products, including historical TM and ETM+ surface reflectance for which no labelled benchmark existed. Locations are sampled uniformly across the conterminous United States in a 60-day window centred on 1 August, when crops are easiest to tell apart, about 25,000 264x264 pixel patches per product at 30 m. Each patch is paired with a USDA Cropland Data Layer (CDL, 134 classes) mask and a USGS National Land Cover Database (NLCD, 17 classes) mask. Files are tar.gz archives per product and label set: ssl4eo_l_{tm_toa, etm_toa, etm_sr, oli_tirs_toa, oli_sr}_benchmark for imagery and ssl4eo_l_{tm, etm, oli}_{cdl, nlcd} for masks; about 48 GB. Nodata masks from ETM+ SLC-off imagery are copied into the labels. Load with torchgeo.datasets.SSL4EOLBenchmark. Released under CC0 1.0 per the paper's licensing appendix; cite Stewart et al. 2023.
How to load this dataset in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="torchgeo/ssl4eo_l_benchmark", filename="ssl4eo_l_etm_cdl.tar.gz", repo_type="dataset")
print(path)Data structure
Files
Example file: ssl4eo_l_etm_cdl.tar.gz
Analyze this dataset without downloading it
On Scigantic, SSL4EO-L Benchmark: Landsat Land Cover (NLCD) and Crop Type (CDL) Segmentation opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Earth Observation & Geospatial datasets
- New Zealand National Coastal LiDAR 1 m Digital Elevation Model (Topo-Bathymetric DEM), LINZ
- Digital Earth Africa - Copernicus Global Land Service - Lake Water Quality
- GWL_FCS30D: Global 30 m Annual Wetland Maps, eight wetland classes, 2000-2022 (AIRCAS, Source Cooperative)
- BigEarthNet v2.0 (reBEN) Sentinel-1/Sentinel-2 Multi-Label Benchmark, LMDB Format
- Digital Earth Africa Cropland Extent Map (2019)
- Satellite - Ocean Colour - MODIS - 1 day - Chlorophyll-a concentration (Carder model)
- LandCoverNet Europe: Annual Land Cover Labels on Sentinel-1, Sentinel-2 and Landsat-8 Time Series (2018)
- LandCoverNet Asia: Annual Land Cover Labels on Sentinel-1, Sentinel-2 and Landsat-8 Time Series (2018)