Accuracy Assessment of Four Land Cover Datasets at Urban, Rural and Metropolitan Area Level

Assessing land use/land cover changes currently represents an important avenue for achieving a better understanding of the urbanization phenomenon. Various free datasets based on satellite imagery are available, but the user should decide which one is the most suitable for their study area. The aim of the present paper is to perform an accuracy assessment of built-up areas using four datasets: Corine Land Cover Backbone (CLC Backbone), High Resolution Layers (HRL)–Imperviousness, Esri Land Cover and Dynamic World. The study case is represented by 12 major metropolitan areas (MAs) in Romania which have the most dynamic economic development and urban expansion. Confusion matrices were created, and the following metrics have been computed: overall accuracy (OA), kappa coefficient (k) and user accuracy (UA). The analysis was performed on three levels: for the entire surface of the MAs and separately for the urban and rural sides. The results at the metropolitan level show that even though CLC Backbone 2018 is the most suitable for extracting the built areas (0.85 overall accuracy), HRL and Esri Land Cover could also be used, as they share the same overall accuracy values (0.67). Significant differences exist between the urban and rural areas. CLC Backbone performed better in the rural areas (0.87) than in the urban areas (0.84). The other three datasets recorded major variations in the overall accuracy for the urban and rural areas. Esri Land Cover has the second greatest overall accuracy for the urban areas (0.81), while HRL is the second most accurate, after CLC Backbone, for assessing the rural areas (0.67). In conclusion, CLC Backbone has the best accuracy performance for all three levels of analysis. The significance of the study lies in the accuracy assessment results on the four datasets, performed at urban and rural levels. This paper aims to help researchers and decision makers choose the best dataset for assessing land use changes. Additionally, having a reliable dataset may help compute the indicators used to monitor the Sustainable Development Goals (SDGs).