National tree-planting pledges have surged since the Paris Agreement, but governments lack reliable, spatially explicit baselines of existing trees. A study led by the University of Copenhagen, published (DOI: 10.34133/remotesensing.0970) on August 18, 2026, in the Journal of Remote Sensing, addresses this gap with a deep-learning framework that estimates tree abundance across entire countries using multisensor satellite imagery.
The team applied the framework to Rwanda and Tanzania, estimating approximately 5.25 billion trees in total—roughly 213 million in Rwanda and 5.04 billion in Tanzania after calibration. Only 48.8% of these trees fell within the WorldCover “tree cover” class; the remaining 51.2% occurred in shrublands, grasslands, croplands, built-up areas, wetlands, sparse vegetation, and mangroves. This finding challenges conventional forest-focused inventories, which often overlook trees outside forests, and underscores the need for methods that capture heterogeneous landscapes.
The framework combines PlanetScope, Sentinel-1, and Sentinel-2 imagery into a 20-band composite resampled to a 3-meter grid. Instead of detecting individual crowns, a hybrid UNet-ResNet50 architecture predicts one tree count per 96 × 96 meter patch. In comparisons, UNet-R50 achieved a mean absolute error of 17.32 and an r² 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 r² of 0.77, comparable to the three-sensor composite, suggesting that scalable, cost-effective monitoring may not require all sensors. Training used about 12 million tree-crown centers from diverse landscapes, and predictions were bias-corrected with independent National Forest Inventory data, yielding median densities of 37 trees per hectare in Rwanda and 42 in Tanzania.
The implications for restoration policy are significant. The researchers emphasize that the framework “can guide the planning and optimization of resource allocation for tree restoration efforts at national scales.” Repeatable satellite-based baselines could help governments track restoration outcomes, identify appropriate planting areas, and account for trees beyond conventional forest boundaries. The approach offers a direct, spatially explicit alternative to costly field inventories or coarse statistical extrapolations, which often smooth local variation and underrepresent trees in nonforest systems.
Limitations remain in dense closed-canopy forests, where crown overlap and signal saturation reduce sensitivity. Future work aims to incorporate understory and smaller woody plants, integrate very-high-resolution data, and explore attention-based sensor fusion and transfer learning to extend the method to new regions and sensors. The full study is available at https://spj.science.org/doi/10.34133/remotesensing.0970. As countries strive to meet ambitious restoration targets, such observation-based tools could become essential for verifying progress and ensuring that planting efforts are directed where they are most needed.


