Comparative validation of recent 10m resolution global land cover maps

Accurate and high-resolution land cover (LC) information is vital for addressing contemporary environmental challenges. With the advancement of satellite data acquisition, cloud-based processing, and artificial intelligence, high-resolution (10~30 m) Global Land Cover (GLC) map production has become increasingly feasible. With a growing number of available GLC maps, a comprehensive evaluation and comparison is necessary to assess their accuracy and suitability for diverse uses; in particular since several maps are not provided with statistically robust assessment or transparent information on validation. This study conducts a comparative validation of recent high-resolution (10~30m) GLC maps, namely ESRI Land Use/Land Cover (LULC), ESA WorldCover, Google and World Resources Institute’s Dynamic World, and GLAD GLC map, examining their spatial detail representation and thematic accuracy. Since high-resolution map validation can be impacted by reference data uncertainties related to geolocation and labelling errors, five validation approaches dealing with reference data uncertainty were evaluated. Results suggest that validation methods considering alternative labels, i.e., majority class, from the surrounding areas (3×3 kernel) produced more reasonable (not too optimistic nor too pessimistic) validation results. The four GLC maps varied in map accuracy across different LC classes and continents. Overall accuracies of the maps ranged between 73.4% ± 0.7% to 83.8% ± 0.4% with WorldCover having the highest global overall accuracy followed by GLAD GLC. The maps’ spatial detail representation was assessed at homogeneity levels within a 3×3 kernel. Although all maps are considered as “high-resolution”, the study highlighted that ESRI LULC and the Google-WRI Dynamic World had less spatial detail, while in comparison, WorldCover showed greater spatial detail at 10 m-resolution and GLAD GLC achieved a similar level for representing the land cover spatial detail and heterogeneity at 30 m-resolution. These important differences in thematic accuracy and spatial detail representation should help users to select a suitable GLC product for their application and offer improvement insights for future GLC mapping efforts, advocating for customizing LC class definitions, enhancing LC characterization to reflect both thematic and spatial details, and addressing validation efforts at regional/national levels as well as LC changes; all contributing to the advancement of high-resolution GLC mapping and its applications.