Accuracy Performance of Three 10-m Global Land Cover Products Around 2020 in an Arid Region of Northwestern China

Abstract

High-resolution and timely global land cover (GLC) products provide valuable information to understand the status of ecosystems and associated trends, pressures on biodiversity, climate change, as well as natural and $\backslash $ or anthropogenic processes. The 10-m resolution GLC products known as FROM GLC, ESA WC, and ESRI LC have not been comprehensively assessed yet, as they have been only recently released. In this study, a spatial consistency analysis and accuracy assessment for these products was conducted in an arid region of northwestern China. Constituent similarity and the spatial consistency were analyzed using the spatial overlay method. In addition, the quantitative accuracy was evaluated based on an error matrix. The results revealed that: 1) based on the comparison of spatial consistencies, the consistent area of the three GLC products accounted for 46.44%, FROM GLC and ESA WC were relatively consistent, and the consistency ratio reached 81.84%; 2) the inconsistency mainly occurred in heterogenous regions, such as transitional zones with mixed land cover types and the edges of small land cover patches; and 3) based on the accuracy measures, FROM GLC had the highest overall accuracy of 77.83%, followed by ESA WC (69.64%) and ESRI LC (62.54%). The ESRI product had the lowest overall accuracy in the arid region examined, which was primarily due to misclassification between shrubland and bare land. Overall, FROM GLC and ESA WC showed the highest consistency and had a good classification accuracy, while ESRI LC had a good performance in the identification of built-up and wetland areas in the arid region examined. The GLC users and producers who are interested in these regions should pay particular attention for improvement of classification and selection of optimal maps.

Keywords

• 10-m resolution
• accuracy performance
• land cover
• arid region