National pledges to plant and restore trees have multiplied since the Paris Agreement, but governments have lacked a reliable way to establish how many trees already exist and where they stand. A study led by the University of Copenhagen, published (DOI: 10.34133/remotesensing.0970) on August 18, 2026, in the Journal of Remote Sensing, offers a scalable answer by combining multisensor satellite imagery with a deep-learning model that estimates tree abundance without mapping individual crowns.
The framework counted roughly 5.25 billion trees across Rwanda and Tanzania, about 213 million in Rwanda and 5.04 billion in Tanzania after calibration. Perhaps the most consequential finding is where those trees are: only 48.8% fell within the WorldCover "tree cover" class, while 51.2% occurred in shrublands, grasslands, croplands, built-up areas, wetlands, sparse vegetation, and mangroves. The implication is that conventional forest inventories and coarse statistical extrapolations substantially undercount trees in nonforest landscapes, which are precisely the areas many restoration programs target.
The method analyzed 96 x 96 meter image patches drawn from a 20-band composite of four Planet optical bands, ten Sentinel-2 multispectral bands, and six Sentinel-1 radar polarization bands, all resampled to a common 3-meter grid. A hybrid UNet-ResNet50 architecture predicted one tree-count value per patch, achieving a mean absolute error of 17.32 and an r-squared of 0.73, outperforming ResNet50, ResNeXt50, SwinT, and ViT. Sensor-ablation tests showed that Sentinel-1 plus Sentinel-2 alone reached an MAE of 18.42 and an r-squared of 0.77, comparable with the three-sensor composite, suggesting that a freer-flowing radar and optical data stream could support monitoring without costly submeter imagery.
Training relied on about 12 million tree-crown centers assembled from manual interpretation of submeter imagery and semi-automated extraction from airborne and UAV LiDAR canopy-height models, covering 2,832 sample areas across roughly 608,483 hectares. National predictions were then bias-corrected using independent National Forest Inventory measurements with country- and land-cover-specific bootstrap calibration, yielding median densities of 37 trees per hectare in Rwanda and 42 in Tanzania. The full study is available at https://spj.science.org/doi/10.34133/remotesensing.0970.
The researchers argue the framework's practical value lies in repeatability. They note it "can guide the planning and optimization of resource allocation for tree restoration efforts at national scales," and that satellite-based baselines could help governments track restoration outcomes, identify appropriate planting areas, and account for trees growing beyond conventional forest boundaries. For accountability, that matters: a repeatable baseline lets gains and losses be measured consistently rather than estimated from sparse field plots. The authors acknowledge limits, including reduced sensitivity in dense closed-canopy forests where crowns overlap and signals saturate, and the exclusion of understory and smaller woody plants. Future work is proposed to integrate forest inventories and very-high-resolution data, and to explore attention-based sensor fusion, domain adaptation, and transfer learning so the approach can extend to new regions, ecosystems, and sensors at larger scales.


