Accuracy assessment of global and regional land cover maps: a case study in Northern Greece

Accurate and up-to-date land cover information is essential for natural ecosystem monitoring, land management and planning, decision making and climate modeling. Recent advancements in remote sensing technology, combined with the increasing availability of high-resolution satellite imagery, have significantly improved land cover mapping capabilities. Modern machine learning algorithms and cloud computing platforms have enabled the systematic generation of higher resolution global and regional land cover products. While these products offer consistent land cover information at large scales, their accuracy can vary across different regions due to differences in classification methodologies, training data availability, and landscape complexity. Therefore, independent validation studies or intercomparison among products are essential steps to assess their reliability and guide users in selecting the most appropriate product for specific applications. This study evaluates the accuracy of two global products, ESA WorldCover and ESRI LULC, and of the pan-European Copernicus CLC+ Backbone product over a large region in Northern Greece. Validation was performed using an independent reference dataset for the year 2021. All maps achieved comparable overall accuracies (73-74.75%), with CLC+ achieving the highest. Notable variations in class-specific accuracies were observed, with the most homogeneous classes such as trees and water being mapped with high accuracy by all products, while mixed vegetation classes exhibited larger errors. It is worth mentioning that CLC+ displayed a more balanced performance in most classes. Nevertheless, additional comparative and independent validations at regional and national level are important for providing further insight into the usefulness of these new fine resolution maps.