.png){kind=logo}
Every year, illegal loggers destroy over 10 million hectares of forest worldwide. Most of them get away with it because the damage is detected long after the chainsaws are turned off. And that is the main problem, we can document the losses but not prevent them. The fix already exists.
Neural networks trained on a historic satellite images archive can compare incoming visual data from orbit with existing imagery and notice the exact moment when trees start disappearing. High-resolution photos are sharp enough to detect individual tree crowns, and the AI model can submit exact coordinates to rangers within days. This combination closes the gap between the start of illegal logging and the response to it, giving even more details with high-resolution satellite images.
Global Forest Watch, run by the World Resources Institute, reported that the tropics lost 3.7 million hectares of primary forest in 2023. The pace is 10 football fields a minute. Forest protection authorities just couldn’t register that destruction due to the large area of illegal logging. That's the structural flaw in traditional remote sensing: you trade detail for coverage. But that’s not all. Traditional methods fail for other reasons as well:
Resolution is too coarse to see the crime begin. The Landsat program by NASA and USGS makes images of the planet with a 30-meter resolution. One pixel covers an area larger than a basketball court. At that scale, a single felled tree, a narrow logging road, or a small dump simply doesn't register — it's smaller than the smallest unit the sensor can record. You can confirm a clear-cut once it sprawls across dozens of pixels, but the first cut is invisible.
Revisit times leave long blind windows. A given satellite only passes over the same spot every several days or weeks, and cloud cover routinely wipes out usable images in tropical regions — exactly where most illegal logging happens. Weeks can pass between one clear shot and the next, and a crew can move in and out of an area entirely within that gap.
Human analysis can't keep up with the volume. Traditional monitoring relies on people scanning frames. No analyst team can review the planet's worth of imagery frequently enough to catch small events in time.
The solution to that issue is that high-res satellite images change what's detectable in the first place — and neural networks change how fast we act on what they reveal.
Start with the imagery. Commercial providers like Maxar or Planet Labs now operate high-resolution satellites that watch our planet with a resolution of 30 centimeters per pixel. They offer imagery where individual tree crowns, vehicles, and debris piles become distinct objects. With such resolution, it is possible to see the difference between knowing a forest patch changed and seeing the specific trees that were taken. The same capabilities power mine monitoring using satellite imagery, letting inspectors track illegal pits, waste piles, and access routes before they expand beyond control.
Now add the algorithm. A neural network trained on high-quality satellite images learns the visual signature of disturbance and scans incoming data continuously, flagging deviations a human analyst would need weeks to find. This isn't theoretical. Researchers have used deep learning on very high-resolution satellite imagery to map individual trees across the West African Sahara and Sahel — a 2020 study published in Nature, led by Martin Brandt and Compton Tucker, counted over 1.8 billion individual trees using sub-meter imagery and a trained model. If you can count single trees across a continent, you can detect when specific ones disappear.
Why this works in the real world, not just in research papers:
It's fast. A model can process enormous amounts of imagery in days instead of the months a human team would need.
It's exact. Ultra-high-resolution satellite imagery lets the model tell you where the damage is and how much of it there is. Rangers get map coordinates, not a hunch.
It doesn't slip. A person scanning thousands of images gets tired and starts skimming. An algorithm checks every frame the same way, every time.
Viktoriia Troian, Product Manager of LandViewer, counts forestry agencies among the platform's core users: they "monitor changes in land cover over time, as this information helps them detect illegal logging, monitor deforestation, plan for reforestation or afforestation efforts, or even determine forest health." But that capability isn't reaching only the agencies with budgets for it. Asked where the platform had mattered most recently, Troian points to Suriname: "the Saamaka community turned to us for custom research on tracking illegal logging activities. We used LandViewer to find the images and analyze them."
None of this works without quality source data, and this is where good intentions collide with reality. A model is only as accurate as the images it learns from. Train it on coarse data, and it finds coarse problems.
The math is unforgiving:
Feed an algorithm low-res tiles and it recognizes clear-cuts.
Feed it hires satellite images, and it identifies every felled tree.
Feed it the highest quality satellite imagery available, and it isolates an individual bag of trash.
Detection accuracy depends on input resolution. That's why the push toward high-res satellite sourcing matters more than the choice of architecture.
So why aren't we already doing this everywhere? It's not the technology — that part's solved. It's the money, and more honestly, the priorities. Real-time high-resolution satellite images and live high-resolution satellite imagery still cost a lot, and the free archives you can pull for nothing rarely come close to the sharpness of commercial feeds.
Aggregation takes some of the edge off that cost. "Those companies sell data from their own satellites, while EOSDA combines data from different sources in one place, which gives more flexibility to users in choosing their source," says Viktoriia Troian from EOSDA LandViewer. In practice, that means a monitoring program can lean on free archive coverage for baseline change detection and pay for high-resolution only over the parcels that flag.
But these same agencies pay for helicopters, fuel, and boots on the ground without a second thought. Even if the smallest part of the budget spent on helicopters, fuel, and boots on the ground is used to pay for high-resolution satellite imagery and user-friendly software programs, with trained models onboard, it will catch the cut while it's happening.
