Decoding the forest: Dr. Jari Vauhkonen on AI, remote sensing, and the future of forestry

If you open Google Earth and zoom in over the boreal forests of Canada or Finland, you’ll find your screen filled with mottled green. From this satellite view, a forest looks deceptively simple. If you were to analyze the forest from above, you might assign each pixel a numerical value describing how green it is, which would tell you about the amount of healthy, light‑absorbing vegetation in that spot. Looking at the greenness values for all the pixels on your screen, you could begin to infer the overall health and density of that patch of forest. You might start to see some patterns emerging. Going a step further, you could recruit the help of artificial intelligence. AI programs can ingest torrents of this kind of data and spot patterns no human eye could track.

Yet, for researchers like Dr. Jari Vauhkonen, a professor of forest planning and an adjunct professor of forest remote sensing at the University of Eastern Finland, the promises of technological precision belie the stubborn truth that forests are much more complex than satellite imagery alone would lead you to believe. A deep-learning algorithm can parse millions of these values, but it can’t always explain why a stand is shifting or describe what’s going on under that green forest canopy. Forests, of course, are not just a green carpet but a multi-story ecosystem, and that depth can’t be captured through simple modelling. Even the most advanced remote-sensing tools, Vauhkonen notes, must be tested against proven methods grounded in decades of observation and practical use. More data isn’t always better data.

The newest Co-Editor-in-Chief of the Canadian Journal of Forest Research, Vauhkonen has built his career at the intersection of sillviculture and technology, where decisions must be made even when the data isn’t perfect. In this conversation, we explore the possibilities and limitations of AI, the value of honest critique, and the need for reliable measurements, especially as forests shift on a warming planet.

What led you to become a researcher in this field?

I have always been drawn to the intersection of forestry and computer science. I worked on individual tree detection from remote sensing data during my master’s and doctoral studies, but I became frustrated with the limitations of those methods. The field was already very crowded, and the same issues persist today. At the same time, I wanted to use modern forest inventory data in ways that actually supported decision-making. I was lucky to end up in positions related to forest management planning, and that really shaped my path into forest planning research, where the data and the decisions optimally meet.

Your work has explored how deep learning and remote sensing can transform our understanding of forests. What can these kinds of techniques reveal that more traditional forms of data collection can’t?

Actually, I tend to approach new techniques with healthy skepticism. I always like to compare them to well-established benchmarks. Often, when a traditional method has been well-defined, the new ones perform similarly. If traditional methods provide clearer causal understanding, they can be the better option.

However, deep learning and related methods do open new opportunities. They allow us to extract information from large and complex data that are difficult to model explicitly. They can help with scaling—applications that require local to regional assessments. But I’d say we are still in the early stages with these methods. They are promising, but we should use them carefully and transparently.

Smartphones and AI-powered tools are starting to appear in the forester’s toolkit. As co-Editor-in-Chief of the Canadian Journal of Forest Research, how do you think forestry will change over the next ten years?

Some ten years ago, I would have answered this confidently, but after recent global events, it’s clear how quickly circumstances can change.

Yes, we’ll see more applications of AI and digital tools in forestry. There is real potential there. But the bigger changes may come from shifting objectives. Geopolitics, climate change, and biodiversity concerns—these factors will probably reshape forestry much more dramatically than any single technology. In fact, AI may also bring risks by introducing new biases. So, forestry will have to adapt faster than before, but one thing that won’t change is the need for objective and accurate information.

You’ve written several commentaries engaging critically with other researchers’ work. Why is this kind of scholarly dialogue important in forest science?

Nice that you noticed this! For example, I’ve commented on studies of bird responses to forest management where conclusions were drawn from models used outside their valid assumptions. These are not just academic arguments; they would provide an entirely wrong picture for decision-making and policy-making if based on individual papers.

I find it a bit contradictory that there is increasing pressure for science-based decision-making, but at the same time, we are aware of reproducibility issues and various biases in research. Nevertheless, critical dialogue helps clarify assumptions, expose errors, and improve the reliability of the knowledge that ultimately informs decision-making. Honest critique should help strengthen the science that society relies on.

If you could redesign the entire forest monitoring system for Finland or Canada from scratch, what principles would guide you?

Big question! It is difficult to redesign something that has evolved over more than a century.

But in principle, I would build on the solid statistical inference of earlier systems. I would establish permanent sample plots, as many as resources would allow. These should be measured as consistently and as frequently as possible, and as many variables as possible should be tracked. I’ve noticed we often realize only later which variables would have been important to track.

Then, I’d use auxiliary information, from remote sensing, management records, or other sources, but keep it auxiliary. The core should be reliable and representative field data that can anchor everything else.

Meet Dr. Nelson Thiffault and Dr. Lisa Venier, who serve alongside Dr. Jari Vauhkonen as Co-Editors-in-Chief of the Canadian Journal of Forest Research.