A satellite study counted 5.25 billion trees in Rwanda and Tanzania, in 2026 finding more than half of these growing outside forests.
But previously in 2015, a team writing in Nature estimated about 19.9 billion trees in Tanzania, nearly four times this count.
The estimates are not directly comparable: the earlier study included small trees beneath the canopy, while the new study also counted trees outside forests.
The full study was published in the Journal of Remote Sensing. A Remote Sensing-Based Solution for National Scale Tree Counting.
Most trees, it seems, crop up from local village farms as well as in towns and cities.
Count a country’s forests and you’ve counted its trees, or so the thinking goes.
However, in Rwanda and Tanzania, most of the trees aren’t in the forests at all.
Researchers counted about 5.25 billion trees across the two countries from 2019 satellite images, and more than half stood on farms, in grassland and shrubland, as well as in towns.
Countries and organizations have pledged to plant more than a trillion trees worldwide since the Paris climate agreement of 2015.
A pledge needs a count of what is already there, and that count has mostly come from forest surveys that often exclude farmland.
Maurice Mugabowindekwe and Martin Brandt of the University of Copenhagen led the team, with colleagues at the Rwanda Space Agency, the University of Rwanda, and the Tanzania Forest Services Agency (TFS).
Their count covers only trees big enough to see from space. That means a crown of at least 32 square feet (3 square meters), or a height of at least 10 feet (3 meters).
Most trees grow outside forests
The mainland Tanzania has approximately 48 million hectares of forest, equivalent to 55 percent of the total land area of 88.6 million hectares. Zanzibar, on the other hand, has an estimated forest area of 106,500 hectares, equivalent to 40 percent of the Islands’ total land area.
The team sorted its count by the land cover map that the European Space Agency’s WorldCover project publishes.
Only 48.8 percent of the trees fell inside the class that map calls tree cover, which is roughly forest and woodland.
The other 51.2 percent were spread across everything else: shrubland held 22.1 percent, grassland 16.4 percent, and cropland 12 percent.
The rest, under 1 percent, stood in towns, wetlands, bare ground and mangroves. Counting only trees in the tree-cover category would leave out more than half the total.
Tanzania accounted for about 5.04 billion of the trees and Rwanda for about 213 million.
The typical acre in Tanzania carried 17 trees (42 per hectare) and the typical acre in Rwanda 15 (37 per hectare).
A typical acre of forest carried about 25 (61 per hectare).
Forest holds trees more thickly, but there is far more land outside it.
Counting trees by the hectare
Older tree maps worked from pictures sharp enough to outline each crown, taken by drones, aircraft or satellites. Those images cost money, and clouds spoil many.
The Planet satellites the team used have a 10-foot (3-meter) pixel.
The freely available Sentinel-2 optical images and Sentinel-1 radar data have 33-foot (10-meter) pixels.
So, the team stopped trying to outline crowns.
They stacked 20 bands of data from the three satellite systems into one image and cut it into squares 315 feet (96 meters) on a side.
Then they trained a deep learning model to estimate how many trees each square held, from the patterns of color, texture and radar echo inside it.
The model learned from about 12 million tree crowns in 2,832 sample areas across nine African countries and India.
People had marked those crowns by hand or pulled them from laser-scanner height maps.
The team also found that dropping the sharper Planet images and using only the two Sentinel satellites gave nearly the same accuracy.
Field plots raised the count
Rwanda and Tanzania both run national forest inventories, in which crews measure the trees in fixed plots.
The team compared its predictions with 196 of those plots in Rwanda and 19,664 in Tanzania.
It counted only trees at least 8 inches (20 centimeters) thick at chest height, the size most likely to show from above.
The model undercounted in most landscapes. Its raw estimate was about 179 million trees for Rwanda and 3.79 billion for Tanzania.
Matching the field plots raised those figures by about a fifth and a third, respectively, to the 213 million and 5.04 billion reported.
The authors’ explanation is that overlapping crowns in dense forest blur into one signal at a 10-foot pixel.
Built-up areas and mangroves in Tanzania were overcounted instead, and the authors suspect buildings and wetland plants were being read as trees.
Even after correction, the Tanzania figure could be anywhere from 4.6 to 5.6 billion.
Farms have room for billions more
A survey of East African farms found an average of 79 trees per acre (195 per hectare).
At that density, the two countries’ 37.6 million acres (15.2 million hectares) of cropland could take about 2.3 billion more trees.
That would be nearly half as many again as the two countries hold now, and trees among crops can raise harvests as well, the authors added.
Tanzania’s own census counts far more cropland than the satellite map does, about 99 million acres (40 million hectares), half of it on smallholder farms.
On that figure, the authors wrote, smallholders alone could add around 3.4 billion trees “if they access adequate support and incentives.”
That is more trees than the count found in all of Tanzania’s forests.
The authors’ point is that a baseline like this one can guide where a government puts its restoration money, and where planting new trees would crowd out naturally open land.
Restoration programs across Africa, from national pledges to the Great Green Wall, have mostly had no such baseline, because operational tree-count systems remain largely unavailable, the authors wrote.
What satellites still cannot see
The satellites can’t see under a dense canopy, so the authors treat their own estimates for closed forest with caution.
The count also leaves out every tree too small to show at a 10-foot pixel, so it can’t yet tell whether newly planted seedlings survive.
The correction factors came from Rwandan and Tanzanian field plots, so they may not apply elsewhere.
To use this method elsewhere, researchers would need local field measurements to check and adjust the satellite estimates.
The data and code are publicly available.
The team plans to incorporate field measurements into the model to better account for trees hidden beneath the canopy.
The researchers also hope to test the method in other regions and with data from other satellites.




