LUCAS land cover representative areas

Spatially representative areas have been computed around LUCAS points in 2018 using an original shape-constrained region-growing algorithm. This algorithm was specifically designed to enhance the accuracy and consistency of machine learning approaches for land cover and land use mapping by ensuring that the delineated areas accurately represent the surrounding landscape. The algorithm takes as input the OSM/CLCplus land cover product (Zenodo), which provides a harmonized and detailed land cover dataset suitable for large-scale analysis.

To facilitate transparency and reproducibility, the tool developed for this purpose has been made openly available through a GitHub repository the MIT license, enabling researchers and practitioners to implement and refine the approach for various applications in land cover and environmental monitoring. The methodology is described in Brodský, L., Landa, M., Bouček, T., Pešek, O., & Halounová, L. (2026). The LUCAS dataset revisited: enhancing spatial representativeness for machine learning land cover mapping. International Journal of Digital Earth, 19(1). https://doi.org/10.1080/17538947.2026.2644671.

Content:

• lucas_points_rg_repre_areas_2018_v1.2.0.gpkg.7z - GPKG dataset with representative areas computed around LUCAS points in 2018 (EU coverage), see attr. desc
• lucas_rg_repre_areas_lvl1.qml - QGIS style for representative areas
• lucas_representativeness_sentinel2_cz_ee_gr_ie_pt.h5.7z - HDF5 dataset with Sentinel-2 time-series image patches (Ireland, Portugal, Greece, Estonia, and the Czech Republic)

LUCAS land cover representative areas can also be downloaded using the Python API or the QGIS plugin; more information is available on the ST_LUCAS project website.