Landsat-8 land-cover benchmark for semi-supervised collaborative clustering: Hanoi, Thanh Hoa, Ho Chi Minh City, Hai Phong (Vietnam) and Valencia, Alicante (Spain)

Six Landsat-8 Collection 2 Level-2 surface-reflectance composites (median of cloud-masked scenes, 2020–2023, Google Earth Engine, nominal 50 m grid, EPSG:4326) with independent land-cover reference maps derived from ESA WorldCover 2021 v200, prepared for benchmarking semi-supervised and collaborative (multi-site) clustering and classification. Set 1 — Vietnam: Hanoi, Thanh Hoa, Ho Chi Minh City, Hai Phong (added in v1.1.0); bands B2–B5; five classes (water, built-up, agriculture, perennial vegetation, forest); 4,707,442 pixels. Set 2 — Spain: Valencia, Alicante; bands B2–B7 (incl. SWIR); six WorldCover classes (tree cover, shrubland, grassland, cropland, built-up, water); 2,406,804 pixels. Labelled regions and spatial block hold-outs are regenerated from the reference maps (protocol in README.md). Landsat-8 imagery courtesy of the U.S. Geological Survey; reference maps contain modified Copernicus Sentinel data (2021) processed by the ESA WorldCover consortium (CC BY 4.0).