In East Tyrol, a large contingent of foresters is currently on the move to search 7,000 hectares of forest for trees that are infested with the bark beetle. In the course of the inspections, each individual infested tree is marked. 1,700 infested areas and 67,000 affected trees have been counted so far. The still relatively young Styrian company Festmeter takes a different approach and localizes infested trees on aerial photos, which are analyzed using artificial intelligence. Laura Anninger met two experts from the start-up for an interview.
Mr. Wöls, your company Festmeter supports foresters with smart applications. What exactly can your technology do?
Kurt Wöls (KW): Our goal is simple: we want to recognize when the forest is changing – as early as possible. With a drone, light aircraft or satellite we fly over the forest and take pictures. We then evaluate these with the help of image algorithms. Artificial intelligence (AI) helps us find conspicuous trees. We do this in order to be able to take forestry measures – for example to fell trees.
Where did the idea for this come from?
K. W.: I took over my parents’ farm in Etmißl in Upper Styria 17 years ago. There are also 45 hectares of forest. I manage it with the help of my neighbors. Every Saturday I walked through the forest and found trees that the bark beetle had infested. They have to go immediately so that the beetle doesn’t infest other trees. It takes eight weeks until it has multiplied and flies out. If you do nothing, it becomes exponential and can endanger the entire forest. The sooner you find infested trees, the better. One weekend in 2013, I went on a trip with my family. I was tense because I should have been in the forest to find infested trees. Next to us on the meadow, a child was piloting his drone. Then my wife said: Why don’t you fly it over the forest. That was the impulse to build a prototype, still with the help of friends. It worked fine for my purposes. And I thought to myself: This could also help others.
What were the steps from prototype to today’s technology?
Andreas Steiner (A. S.): I joined Festmeter in 2016, two years after it was founded. At that time there was still a lot of research work to be done with the drone. Big topics were the exposure, weather influences, light and shadows on the pictures. We only got the problems under control with time. Another big point was to improve positional accuracy. So to be able to show where the infested tree is. We did that with georeferencing. Then the interest of forest owners increased and with it the areas. The drone cannot map very large areas. After mapping 40 acres, it has to fly back and the battery needs to be replaced.
How did you solve that?
A.S.: In the search for alternatives, we came across the ultralight aircraft. That accelerated the process a lot. Finally, the last development step was to access satellites. There are many providers, which made it easier. But satellites also bring problems. Clouds and the shadows they cast naturally play a much bigger role in satellite imagery than in drone images. Initially, we used images from the EU satellites Sentinel-2A and Senitnel-2B. However, these have a resolution of ten by ten meters. It’s very difficult to see a treetop there. That’s why we bought access to images from a total of five satellites with better resolution. For example from the satellite “Planet”.
But you still use all three methods – why?
A.S.: With the drone we fly about a hundred meters above the treetop. The images are accurate, the AI can identify diseased trees very quickly. The satellite is 500 kilometers away. But: he comes by the spot once a month. This means we can use the satellite imagery to see how the forest is changing over time. It also lowers the price for forest owners. For areas under 100 hectares we now use drones, for larger areas microlights. The satellite is ideal for areas over 2,500 hectares.
What do you see in the pictures?
K. W.: Healthy plants reflect a lot of electromagnetic radiation in the near infrared, i.e. in a range of around 840 nanometers wavelength. In this area they “glow” very strongly. If the tree gets sick, they do it less. So you can see quickly in the near-infrared range when the tree is getting sick. Our cameras and image sensors pick up this light. Healthy trees are reddish in the pictures. Green spots show us that they are going to break or already are. That means we don’t know exactly when the bark beetle digs in. But we sort of see the symptoms of that about four weeks after that. Then the resin flow in the tree has changed and it carries out much less photosynthesis. I always explain it like this: The tree has chills.
And the AI recognizes that?
K.W.: Our AI scans the images and constantly decides: Is this tree healthy or not? She currently finds about 80 out of 100 damaged trees. The AI is a learning algorithm. Their nature is that they get better the more decisions and data they get. We continuously train our AI to make it even more accurate. It gets better with every forest area that it analyzes. As a result, it can also adapt more and more easily to the various forest areas. Finally, the forest owner receives an aerial photo on which the AI has identified damaged trees in the forest – and the GPS data of these trees.
Festmeter has received a grant from the EU because you use a “trustworthy AI”. How must an AI be programmed that can be trusted?
A.S.: We map surfaces from above – be it with drones, light aircraft or satellites. We focus on the forest. But theoretically, our AI could also find other things in the images – such as houses, vehicles or people. We form a so-called “Area of Interest” around the forest areas. That means our AI doesn’t look at areas that aren’t forest. This rules out her learning to recognize people, for example.
Is the bark beetle plague so bad that all the effort is worth it?
K. W.: I grew up on the farm. We had a forest and a sawmill. When I was a child, more than 40 years ago, my father left infested trees standing. He only fetched a bark beetle tree when someone needed dry wood, for example for the roof structure. Today I can no longer leave the forest unobserved. If I did that, eight more trees would be dead in eight weeks.
What has changed since you were a child?
K.W.: The forest is the same. What has changed is the weather. There is now heavy rain and thunderstorms, it is getting hotter. The hotter the summer, the more active the bark beetle becomes. This is not good for the trees, especially the spruce. The problem is: we have many spruce monocultures in Austria. Every infested tree means a risk for the forest. I know that there is a lot of drama in East Tyrol. This is where climate change and the consequences of the storms of recent years come together. It is difficult to remove fallen trees on very steep surfaces.
A. S.: All of Austria is in a bark beetle crisis, and has been for several years. I remember the summer of 2019 when the bark beetle destroyed hectares of forest in the Waldviertel.
Can your AI also recognize other diseased trees?
K. W.: Yes. Our AI always recognizes in the crown when the vitality of the tree changes. She finds trees that produce less chlorophyll. But we really can’t say why the tree is doing this. The bark beetle is one of the more aggressive pests in the forest. But the ash trees also have a fungus, for example, which causes their roots to die off. The ash tree then suddenly falls over, without warning. We can train our AI to find ash trees that have lost their vitality. But this is only possible in spring and summer when the ash tree has leaves.
Are forest owners already open to digital solutions like yours?
KW: In Osttirol there are a lot of people out and about, walking in the woods. You follow the bark beetle on foot. A lot of money is invested in this. We could support monitoring from above. We are currently in contact with the authorities about this. There are people who are open to digitization in the forest. We have implemented many projects with the Bavarian Federal Forests. The Iseltal Forest Cooperative in East Tyrol is also a project partner. We were allowed to map their areas with drones and satellites. But actually, digitization has not really arrived in forestry. The information we provide must also be translated into action. To do this, it must be as easy as possible for the forester to access the data in the forest. We are currently working on a joint project with a large local forestry company where we want to find out how we can make it easier for foresters. It’s about questions like: How do I actually find the tree? How up-to-date is the data? Am I actually connected to the cloud in the forest? Can I voice control the app if I don’t have both hands free?
A. S.: Large-scale monitoring will become increasingly important. One cannot assume that there is a fleet of forest rangers that can pass every tree. We don’t want to replace the foresters. We want to help identify bark beetle infestations so that forest owners can react.
What is your motivation to continue with digitization in the forest?
A. S.: When I started at Festmeter, I was an intern for two months. I immediately gave up my studies in geosciences at Montanuniversität and started full-time. I find it exciting to be part of new developments. I want to be able to say: I worked on this project and we achieved something.
KW (nods): That’s correct. My personal motivation is my farm. We have five buildings here, 60 hectares of land and a stone house that is almost 200 years old. Eighty years ago, ten people could live here self-sufficiently, only from the products they produced. Today, not even I can live from what we produce here. But I don’t want to give up the farm, I’m rooted here. My children should also have the opportunity to live here in the country. In order for them to be able to do that, I have to be a “data manager” as well as a farmer. With algorithms and data science, I also want to enable perspectives on land – always in combination with the forest and nature. To do this, I have to provide technology that helps people and nature and is also used.
