Yellow pixels in the places where (illegal) felling will take place

Yellow pixels in the places where (illegal) felling will take place

In the middle of the forest in Zeist looms the shiny, oval building of the World Wildlife Fund (WWF-NL). Inside, ecologist Merijn van Leeuwen looks at a satellite image of the Amazon rainforest. There are small yellow pixels over that satellite image. Each small pixel is actually an area of ​​about four hundred by four hundred meters. These areas are currently still full of trees, but Van Leeuwen knows – even before the first tree has been cut down – that those yellow pixels will probably be felled (illegally) within a few months. This was predicted by an artificial intelligence (AI) model developed in the Netherlands: Forest Foresight.

Google Earth clearly shows how gigantic the Amazon rainforest is. The largest rainforest in the world extends over nine countries in South America, including Brazil and Colombia. With an area of ​​approximately 6.7 million km2 the land area of ​​the Netherlands fits in there about 160 times.

It is therefore impossible to patrol the entire Amazon region. But anyone who knows in advance where logging is likely to take place can send agents and forest rangers there to stop deforestation.

How does Forest Foresight know in advance where logging will take place? And can this AI model reduce deforestation in the Amazon?

When too much forest disappears, a catastrophic change threatens

Van Leeuwen proudly shows a photo of a sloth on his laptop that he took during field work in the Amazon rainforest. “It’s magical,” he says. “You hear animals all the time. During the day, when it is warm, you can hear the dee-dee-dee sound of the cicadas. At night the crickets take over.” Nowhere else in the world are there as many animal and plant species as in the Amazon rainforest. In addition, the 390 billion trees in the Amazon help limit global warming.

The rainforest also determines the local climate. Through evaporation of water, the forests provide rainfall in the region. According to Van Leeuwen, approximately ten bathtubs worth of water evaporate every day from each tree in the Amazon. That water vapor blows to places in the region, and then rains out again. This makes agriculture possible in places that would otherwise be too dry.

Furniture and minerals

In the meantime, the area is being destroyed by illegal mining, agriculture and infrastructure construction. Wood is cut down to use for furniture and farmers make room for agriculture. Large areas of forest are also cut down to extract minerals such as iron and gold from the ground. In 2022 alone, two million hectares of forest will disappear: that is more than five times the area of ​​all the forest in the Netherlands. About 17 percent of the Amazon has already been deforested, according to Van Leeuwen.

And when too much forest disappears, a catastrophic change threatens. If more and more trees disappear, there will also be less and less evaporation. This results in less moisture in the air and less rain in the region. Drought will cause even more trees to disappear. That is a negative spiral. How far we are from that tipping point is still a mystery, but it is clear that we must stay as far away from it as possible.

The AI ​​model Forest Foresight should therefore help prevent even more forests from disappearing by calculating where there is a high risk of deforestation. Previous satellite images showed where felling took place, so when the damage had already been done. To predict where felling will take place, preferably months before the saw is switched on, the AI ​​model first looks at where and when felling took place in the Amazon in the past.

To this end, the model compares afforestation at different times that are far apart using old radar images. These radar images come from Europe’s Sentinel-1 satellites. “The AI ​​model looks for properties that the previously deforested areas have in common,” says radar specialist Johannes Reiche by telephone from Wageningen University. Forest Foresight uses satellite images and other data about, for example, the soil. “For example, we see that felling was done less often on steep or wet forest land. Building roads is often a precursor to large-scale felling.”

In the future, Reiche also wants to use mobile phone data to find out where in the forest there is suddenly a lot of human activity. This is already being done to predict illegal fishing. Forest Foresight looks for forested areas that have the same characteristics as places that have been logged in the past. Those areas are the yellow pixels, the hotspots.

Yellow ‘risk pixels’ on a map of part of Borneo.
Image WWF

How do those yellow pixels from the forests in Zeist ultimately reach the enforcers in the Amazon region? The Dutch WWF office in Zeist sends the maps showing the hotspots to local authorities, via a local WWF office in the relevant Amazon country, says project manager Jorn Dallinga of Forest Foresight by telephone from an AI meeting in Germany. “These local authorities then tell enforcers, for example people who monitor protected nature reserves or who combat illegal mining, where to go.” That’s how it goes now. “In a few months we will be training local authorities so that they can work with the model themselves.”

Illegal gold traders

Van Leeuwen talks about the first results, which are promising. In the African country of Gabon, local authorities have sent groups of enforcement officers to hotspots found by Forest Foresight 34 times. Illegal gold traders were arrested. Thirty hectares of forest were spared.

But in Gabon the government is making efforts to combat deforestation. That is not the case everywhere. “That is why we check in advance which countries have the political will and are able to use Forest Foresight,” says Dallinga. “In Suriname we have stopped applying the AI ​​model. It turned out that local government authorities, for example those for mining and those for forests, do not communicate well with each other. Then there is no point in getting involved with our model.”

Also read
Gabon wants something in return for the CO2 that its jungle stores. Only: who wants to pay for that?

Elephants in Lopé National Park in Gabon.


The WWF does have confidence in Bolivia, Peru and Colombia. Colombian President Gustavo Petro “takes nature conservation seriously,” says Dallinga. In 2023 there will be more than 50 percent less deforestation than in 2022. That is a good basis.”

Although Forest Foresight already appears to work well, the model will also miss areas, according to Reiche. Van Leeuwen: “It is also about the signal that governments send out. That they monitor deforestation more efficiently.”

Ecologist Roderick Zagt of the Tropenbos International foundation sees potential in Forest Foresight. The Dutch foundation supports local organizations in countries to protect tropical forests and is not itself involved in Forest Foresight. But, says Zagt, there must be people who have an interest in stopping deforestation. “The local forest service, for example, is often not interested in stopping deforestation because they can make money from deforestation. It helps when you strengthen the rights of the local population over an area, as it is in their interest that no logging takes place in their backyard. There are projects in West Africa where local residents are given tools, such as a GPS and camera, to collect data from the ground about deforestation in their area. Forest Foresight can be an addition to this. With the model you can see at a glance where deforestation is taking place in a large area. Forest Foresight could help to direct local residents who have an interest in ensuring that their area is not deforested to a place where illegal activities are likely.”