One deforestation is not the other. Large -scale wood production, transformation into agricultural land, the construction of roads, a new gold mine, small -scale hood, trees that blunt or go up in flames during a forest fire: all kinds of variants are conceivable. But when it comes to disappearing forest, then all those nuances will soon be swept in a lot, says PhD student Bart Slagter. So in recent years, at Wageningen University, he has changed exactly that, with the development of a new, self -learning computer model for tropical forests. “Not all forest losses is permanent. So to know how far -reaching the consequences are, for example in terms of co2-Moet, you need accurate data. “”
For his research, Slagter uses various satellite images, such as radar and infrared, to measure the changes as accurately as possible. “During the rainy season in the tropics it is cloudy for a long time, but with Radar you can look through the clouds.”
New roads can be recognized on the satellite images as line -shaped changes. But how do you recognize the difference between a gold mine and agricultural land from the air? “Differences in reflection play a role in this: fierce reflection indicates a lot of bare soil and often on mining. Building, form and especially context are also important. If you detect forest loss without a road in the neighborhood, then there is a good chance that it is natural.” The increasingly higher resolution of the images also helps. “With pixels of 10 by 10 meters you can distinguish a lot of details. Ten years ago the resolution was much lower. Moreover, satellites now supply new images on a weekly basis.”
The ‘training’ of the algorithm is done with the help of artificial intelligence. “You enter a lot of data and based on that, image recognition is getting better. Compare it with other forms of Deep Learning: Just as you can learn to distinguish a computer from a cat from a cat, that is also possible with different forms of deforestation. ”
The second-big rain forest
Slagter is certainly not the first to deal with deforestation monitoring via satellites, he emphasizes. “Only: often the models are behind the facts, while you prefer to map small -scale changes as quickly as possible, so that you can still intervene. Roads are too often underestimated in the past, because they are relatively small changes, seen from space. They are often built to facilitate selective logging, but can also often form the prelude to much greater destination.”
Precisely thanks to the self -learning algorithm, changes will soon be able to enter real time are displayed so that it can be taken in time with illegal or non-durable activity. “We are aiming for mapping them on a weekly basis.”
As an example, he mentions a recent publication about the road network in Congries, African. “That is the largest rainforest on earth after the Amazon, but relatively little is known about the developments in the forest there. We have looked at the speed with which the roads expand. Based on that, you can calculate the amount of logging, and thus how much biomass is lost. But the impact goes much further: the roads also determine for example where poaching is taking place.”
In the future, the results could also be used to check companies, says Slagter. “So that you can see if they are not cutting down at illegal locations, or continuing for too long. If roads are laid out outside of an area for which a concession has been issued, you know that it is wrong. Certainly if you want to spend certificates for responsible timber hood, such information is essential.”
The European Coverage Act – which was postponed by the European Parliament last December – also prescribes that you can only market products that do not provide deforestation. “That cannot be checked without monitoring. So our model could help with that.”
Learning curve
The love for satellite images is deep at Slagter. “I love the technology behind it. When I came from high school, I started with little enthusiasm on a study planning in Groningen. Only when I learned about GIS – Geographic Information Systems, with which you can analyze geographical data – I became enthusiastic. That technology fascinates me, you can do it so infinitely with it. And I think it is great that the field is beautiful that the field is beautiful that the field is beautiful that the field is beautiful that the field is beautiful that the field is beautiful that the field that is the field that the field is beautiful” ”” ”” ”” ”” ”” ”
Algorithms also have a learning curve. That is why the model always neatly indicates how accurate the interpretation made is. “If you certainly have a gold mine for 80 percent, then you dare to count. But in the case of probably 60 percent we are a lot more careful. It sometimes remains guessing, even with GIS.” To test the accuracy of the model in the field, he will go to Suriname with colleagues next month with colleagues.
It is double, says Slagter at the end, after he has shown dozens of satellite images full of fire. “I find my work methodologically interesting, I can really become enthusiastic if we can map those roads, gold mines and logging in super accurate. But you are also looking live at the foresting of forest’s destruction. That is why I really hope that this method can contribute to better managed forests.”