March 16, 2026
Leaf it to Meta
Canopy Height Maps v2
Meta says it can see every tree—nerds cheer, skeptics squint
TLDR: Meta launched an open-source forest map that estimates tree height worldwide with a major accuracy boost, aiming to help climate and planning efforts. The community split fast: tinkerers celebrated hands-on mapping wins, while skeptics questioned how fresh the data is and why Meta is doing this at all.
Meta and the World Resources Institute just dropped Canopy Height Maps v2, an open-source AI model and world-scale map that claims to see forest height with much sharper detail. The headline stat? An accuracy jump from R² 0.53 to 0.86—a big leap that has map nerds buzzing and privacy hawks peeking through the shrubs. The tech behind it, DINOv3 (an AI trained on tons of satellite photos), promises to spot tree height from shadows and textures without armies of human labelers. Governments in the UK, EU, and US are already lining up to plug this into climate and city planning tools.
But the comments are where the real canopy chaos lives. One hands-on tinkerer flexed a DIY version using USGS 3DEP laser scans plus an AI assistant to count the trees in their own yard, which had the thread yelling “open data for the win.” Meanwhile, the mood swings to “hmm” as others ask the blunt questions: How fresh is this satellite data? and Why is Meta mapping tree heights, exactly? Cue the memes—“tree-veillance,” “Meta but make it leaf”—as fans celebrate the open release while skeptics side-eye corporate motives. For now, the vibe is hopeful, curious, and just a little spicy—like a nature doc narrated by Reddit.
Key Points
- •Meta and the World Resources Institute released CHMv2, an open-source model with global canopy height maps for forest monitoring and management.
- •CHMv2 replaces DINOv2 with DINOv3, pre-trained on the SAT-493M satellite imagery dataset, improving global accuracy and detail.
- •Model performance improved significantly, with R² increasing from 0.53 to 0.86 and reduced bias for tall trees.
- •Training data and methods were enhanced via more diverse lidar examples, automated satellite–lidar matching, and a specialized loss function.
- •CHMv1 has been adopted by public bodies in the UK, EU, and US; CHMv2 is expected to support initiatives including the EU’s 3 Billion Tree Initiative and US city planning efforts.