Between 2022 and 2023, the EU’s forest area grew a miniscule 0.1%. What’s more, this figure does not capture the quality of those forests, including their age, diversity and ability to store carbon and support wildlife. For initiatives like the EU’s Biodiversity Strategy for 2030 to successfully protect and restore forests, governments need an accurate picture of the trees that already exist.

That’s where machine learning can help. Satellite imagery can show where land is green but does not always indicate what that green represents, whether it’s a complex forest ecosystem or agricultural tree crops. In practice, that distinction can hinge on the height of the trees, since many international standards require a minimum canopy height of five metres to classify an area as forest.

Meta’s open-source Canopy Height Maps, built in partnership with the World Resources Institute, use the company’s DINO self-supervised model, a computer vision training programme, to estimate tree height from satellite imagery. The model, built on more than a decade of the best available satellite imagery, can capture the subtle, unlabelled visual cues that indicate tree height, such as shadows, textures and crown shapes without requiring millions of manually labelled examples.

As the European Commission’s Joint Research Centre (JRC) developed its 2020 Global Forest Cover map – a spatially explicit baseline of forest presence and absence at ten-metre resolution – it needed a tool that could capture tree height. By integrating the first version of Meta’s maps into its workflow, canopy height data improved the JRC’s ability to apply the five-metre threshold and better distinguish forests from other tree-covered land. It also supported efforts to filter out tree crops that can exceed five metres but do not qualify as forest under internationally agreed definitions. The resulting product, available through the EU Forest Observatory, provides one of the most comprehensive open views of global forest cover.

Going forward, research organisations hope to use Meta’s Canopy Heights Maps for other tree monitoring efforts, including the 3 Billion Trees initiative, a commitment to plant at least 3 billion biodiverse trees across the EU by 2030. This partnership demonstrates how artificial intelligence tools can support national- and international-scale forest inventory and help track progress toward climate commitments.