Lesiv et al. (2025) A global reference data set for land cover mapping at 10 m resolution

Identification

• Journal: Earth system science data
• Year: 2025
• Date: 2025-11-17
• Authors: Myroslava Lesiv, S. Fritz, Martina Dürauer, Ivelina Georgieva, Marcel Buchhorn, Luc Bertels, Nandin‐Erdene Tsendbazar, Ruben Van De Kerchove, Daniele Zanaga, Dmitry Schepaschenko, Linda See, Martin Herold, Bruno Smets, Michael Cherlet, Andreas Brink, Ian McCallum
• DOI: 10.5194/essd-17-6149-2025

Research Groups

• NODES, ASA, International Institute for Applied Systems Analysis (IIASA), 2361 Laxenburg, Austria
• VITO Remote Sensing, 2400 Mol, Belgium
• Laboratory of Geo-information Science and Remote Sensing, Wageningen University, 6700 HB Wageningen, the Netherlands
• Remote Sensing and Geoinformatics section, GFZ Helmholtz Centre for Geosciences, 14473 Potsdam, Germany
• European Commission-Joint Research Center, 1050 Brussels, Belgium

Short Summary

This paper presents a unique global reference land cover dataset for 2015, featuring over 16.5 million 10 m resolution records across 12 land cover classes, collected through expert visual interpretation to support high-resolution land cover mapping.

Objective

• To present a unique global reference data set for land cover mapping at a 10 m resolution, aligned with Sentinel-2 imagery for the year 2015, developed to support the production of the Copernicus Global Land Service (CGLS-LC100) annual global land cover maps.

Study Configuration

• Spatial Scale: Global coverage, with data records at 10 m resolution (collected within 100 m pixels).
• Temporal Scale: Data collected for the year 2015, over a 5-year period of expert annotation.

Methodology and Data

• Models used: Geo-Wiki platform (for data collection and visual interpretation).
• Data sources: Very high-resolution imagery (Google Maps, Microsoft Bing, ESRI World, MAXAR imagery), regional land cover maps (e.g., CORINE), street view imagery (Mapillary, Google Street View), Normalized Difference Vegetation Index (NDVI) time series (Landsat Collection 2 Tier 1 Level 2 32-Day NDVI Composite, MOD13Q1.061 Terra Vegetation Indices 16-Day Global 250 m) derived from Google Earth Engine, Sentinel-2 image time series from Sentinel hub, geo-tagged photographs, Google Earth Pro historical imagery.

Main Results

• A global reference land cover dataset was created, containing over 16.5 million data records at a 10 m resolution (or 165,000 records at 100 m resolution).
• The dataset includes information on 12 different land cover classes, plus a "Not sure" class, defined using the UN FAO Land Cover Classification System (LCCS).
• Data collection involved a two-stage systematic sampling design, initially at approximately 35 km intervals, with additional sites added in areas of low classification accuracy.
• A rigorous quality assurance process, including regular expert coordination meetings, individual quality checks (90%-95% accuracy requirement), and comparison with regional land cover maps, ensured high data quality.
• The overall accuracy of the reference data set is estimated at 90%-95%.
• The dataset is openly available for download at https://doi.org/10.5281/zenodo.14871659.

Contributions

• Provides a unique, high-quality global reference land cover dataset at 10 m resolution, independent from existing land cover products, filling a critical gap for training and validation of high-resolution land cover maps (e.g., CGLS-LC100).
• Offers an extensive collection of over 16.5 million expert-interpreted records for 2015, enabling various applications such as land cover analysis, ecosystem mapping and modeling, and biodiversity studies.
• Complements other recently released high-resolution datasets by providing specific spatial coverage, resolution, and quality, significantly benefiting the land cover mapping community.

Funding

• JRC Framework Service Contract no. 199494 (Copernicus Global Land Operations)
• ESA WorldCover
• JRC Framework service contract no. 945329-IPR-2023 (Copernicus Global Land Cover and Tropical Forest Mapping and Monitoring Service)

Citation

@article{Lesiv2025global,
  author = {Lesiv, Myroslava and Fritz, S. and Dürauer, Martina and Georgieva, Ivelina and Buchhorn, Marcel and Bertels, Luc and Tsendbazar, Nandin‐Erdene and Kerchove, Ruben Van De and Zanaga, Daniele and Schepaschenko, Dmitry and See, Linda and Herold, Martin and Smets, Bruno and Cherlet, Michael and Brink, Andreas and McCallum, Ian},
  title = {A global reference data set for land cover mapping at 10 m resolution},
  journal = {Earth system science data},
  year = {2025},
  doi = {10.5194/essd-17-6149-2025},
  url = {https://doi.org/10.5194/essd-17-6149-2025}
}