EGLC: Ensemble Global Land Cover Reference Dataset (2000-2022)

EGLC: Ensemble Global Land Cover Reference Dataset (2000-2022) The Ensemble Global Land Cover (EGLC) dataset is a harmonized global compilation of land cover reference samples from 2000 to 2022. It integrates and standardizes data from multiple authoritative sources to support global-scale land cover classification and validation. Reference points datasets eglc_c_go_epsg.4326.parquet (~72M points)Compiled dataset with all original points (before class harmonization). eglc_harmonized_c_go_epsg.4326.parquet (~22M points)Reference points selected per class and dataset from compiled points (above) and harmonized land cover classes (see details below). samples_eglc_c_us_epsg.4326.parquet (~2.4M points)Pilot points for US (CONUS). Source datasets and class harmonization GLanCE (Global Land Cover Estimation)Data Range: 2000-2021Original Classes Used: Beach/Sand, Rock, Soil, Developed, Shrub, Deciduous, Evergreen, Mixed, WaterHarmonized Classes: Shifting Sand, Rock Outcrops, Bare Soil, Developed/Urban, Shrubland, Deciduous Forest, Evergreen Forest, Mixed Forest, WaterReference: Friedl et al. (2022) Dynamic WorldData Range: 2017-2019Original Classes Used: Flooded Vegetation, Snow/Ice, WaterHarmonized Classes: Wetland, Permanent Snow/Ice, WaterReference: Tait et al. (2021) GPW (Global Pasture Watch)Data Range: 2000-2022Original Classes Used: Cultivated grassland, Natural/Semi-natural grasslandHarmonized Classes: Cultivated Grassland/Pasture, Natural Semi-Natural GrasslandReference: Parente et al. (2024) MangrovesDB (Temporally augmented)Data Range: 2000-2021 (Visual inspected for 2021. Augmented points backwards in time from imagery year to 2000, assuming temporal persistence).Compiled points:Original Class Used: MangroveHarmonized Class: MangroveReference: Maxwell et al. (2023) Compiled WorldCerealData Range: 2016-2022Original Classes Used: Annual cropland, Cropland, Deciduous forest, Evergreen forest, Mixed/unknown forest, Shrubland, Snow / ice, WaterHarmonized Classes: Cropland, Deciduous Forest, Evergreen Forest, Mixed Forest, Shrubland, Permanent Snow/Ice, WaterReference: Oliveira et al. (2024) Data processing and integration Data Ingestion: Each dataset was loaded and converted into a common tabular-geospatial format with a consistent CRS (EPSG:4326). Specific Adjustments: MangrovesDB samples were manually relocated and verified using VHR imagery before temporal expansion. Schema Standardization: All datasets were standardized to use the following fields: dataset, id, original_class_1, original_class_2, class, year, lon, lat, tile, geometry. Final harmonized land cover classes Bare Soil Cropland Cultivated Grassland/Pasture Deciduous Forest Developed/Urban Evergreen Forest Mangrove Mixed Forest Natural Semi-Natural Grassland Permanent Snow/Ice Rock Outcrops Shifting Sand Shrubland Water Wetland Technical details Spatial Coverage: Global (excluding Antarctica)Temporal Range: 2000-2022Coordinate Reference System: EPSG:4326Data Format: Parquet Dataset structure dataset: Identifier of the original source (e.g., GLanCE-5, GPW-calibration, WorldCereal). id: Identifier (source-specific information). original_class_1: First-level land cover label from the original dataset. original_class_2: Second-level label when applicable (e.g., GLanCE level 2 class). class: Final harmonized land cover class mapped to a unified 15-class legend. year: Observation year. lon: Longitude in WGS84 coordinates (EPSG:4326). lat: Latitude in WGS84 coordinates (EPSG:4326). tile: ID of the spatial tile intersected by the sample (used for spatial indexing). geometry: Point geometry in WKT or binary format (CRS: EPSG:4326). How to Load the Data In Python (DuckDB): import duckdb con = duckdb.connect() df = con.execute("SELECT * FROM '{parquet_path}'").fetchdf() In Python (GeoPandas): import geopandas as gpd gdf = gpd.read_parquet("{parquet_path}") In R: library(arrow) df <- read_parquet("{parquet_path}")