EcoDataCube for Europe: combining spatiotemporal ML and open EO data cubes

EcoDataCube for Europe comprises multiple land cover, terrain, climatic, soil and vegetation ARCO (Analysis-Ready Cloud-Optimised) layers for continental EU at relatively fine spatial resolutions (30-m to 1-km) for a period of 20 years (2000—2020). In general, the 30-m datasets are based on harmonised time-series of Landsat and Sentinel-2 imagery (GLAD Landsat ARD), multi-source Earth Observation data (e.g. ERA-5, Chelsa climate, MOD11A2, Suomi NPP/VIIRS), and a spatiotemporal ensemble machine learning approach capable of producing per-pixel probabilities and uncertainty outputs. This mapping framework was used to predict 43 land cover classes (compatible with CORINE Land Cover-CLC), 16 forest tree species (potential and realized distributions), and 5 soil properties for 4 depths (i.e. organic carbon, pH, bulk density, clay and sand content). All ARCO layers are publicly accessible as proper Open Data (CC BY) through Web Viewer (https://ecodatacube.eu) and SpatialTemporal Asset Catalog (STAC - https://stac.ecodatacube.eu).