Satellite-derived forest datasets are widely used for forest monitoring, especially where there is a lack of high-quality national land-cover data. However, different datasets can vary markedly in terms of their spatial resolution and classification methods, which could potentially affect the conclusions of studies that rely on these data. Writing in One Earth, Sarah Castle and colleagues compare the area-based spatial agreement among 10 different, widely used global forest cover and change datasets, and find only 26% spatial congruence among them at their native resolutions. Harmonizing to a standard spatial resolution of 500 m increased the dataset agreement by only 1%. Among different biomes, there was highest agreement in tropical and subtropical moist broadleaf forests; agreement was lowest in tropical and subtropical dry broadleaf forests, and temperate and boreal biomes had moderate spatial agreement of between 23 and 33%. In general, areas with lower tree cover had worse spatial agreement as compared to where canopy cover is high. To demonstrate the potential implications of these classification mismatches on sustainability measures, the authors outline two case studies that reveal an order-of-magnitude difference in forest carbon estimates in Kenya as well as estimates of forest-proximate people living in poverty in India. In a third case study they also show that, across 6 years, there was only a 37% agreement in the estimated change in suitable habitat for the endangered white-cheeked spider monkey in Brazil. The authors conclude that researchers and decision-makers need to evaluate carefully whether the global forest dataset they are using is appropriate, and they further provide a decision support tool to help to guide users in selecting the right dataset for their specific application.
Original reference: One Earth https://doi.org/10.1016/j.oneear.2025.101558 (2025)
This is a preview of subscription content, access via your institution
