Accuracy assessment of the global forest cover map for the year 2020 : assessment protocol and analysis

This technical report presents the assessment protocol for forest/non-forest interpretation from high spatial resolution images that was used to estimate the accuracy of the second version of the Global Forest Cover map for the year 2020 (GFC2020). GFC2020 is a novel spatial representation of forest presence or absence at 10 m spatial resolution. This map is designed to support the risk assessment in the implementation of the EU Regulation on deforestation-free supply chains (EUDR, Regulation (EU) 2023/1115), but its use is not mandatory, not exclusive and not legally binding. A group of international experts from within and outside the JRC interpreted 21,752 sample units. Results show an overall accuracy of 91% with a commission error for forest (areas erroneously mapped as forest) of 18% and an omission error (areas missed to be classified as forest) of 8%. Detailed analysis reveals the reasons for misclassification, e.g. the presence of other wooded land with very sparse tree cover or a mix of shrubs and bushes that can be easily confused with forest or ambiguity in the geolocation precision of the sample unit or the forest boundary. The spatial analysis shows clusters of misclassified sample units, and thus lower accuracies in regions of open dry forests, e.g. in Eastern Brazil or Central Australia.