Editor's note
This week’s stories share a quieter theme: revisiting foundations.
From growth and yield models being rebuilt, to canopy reconstruction with standard cameras, to consolidated protection maps, several examples point to underlying systems being updated rather than entirely new layers being added.
As AI tools become more capable, they rely increasingly on the quality of the data and models beneath them. Improving those foundations may not be the most visible form of innovation, but it tends to have long-term impact.
This week is less about disruption, and more about maintenance—of models, interfaces, and shared information structures that forestry depends on.
Axel
WHAT GOT ME THINKING
Canada Invests $5.9M to Rebuild Forest Growth Models from Scratch
University of Alberta professor Robert Froese is leading an eight-year project to replace Alberta's two aging forest growth and yield models—GYPSY and the Mixedwood Growth Model—with a modern generation built on remote sensing data. Funded by industry giants including Canfor, West Fraser, Weyerhaeuser, and Tolko, the new models will handle long rotations, genetic improvements, carbon storage, and climate change effects that the old tools weren't designed for.
Axel’s notes:
What stood out here is the decision to modernize something that most people rarely talk about: growth and yield models.
These models sit quietly underneath planning systems, harvest forecasts, carbon calculations, and investment decisions. Many were developed decades ago for different climatic assumptions and rotation patterns. Updating them is not headline-grabbing work, but it directly affects the reliability of everything built on top.
As AI tools become better at handling complex reasoning and running large-scale scenarios, the importance of reviewing base models increases. Smarter systems amplify whatever logic they are given. That makes foundational research like this worth paying attention to.
Swedish Startup EcoVibes Turns Forest Sounds into Biodiversity Data
SLU doctoral students Christian Höök and Mathias Kristoferqvist launched EcoVibes from Örnsköldsvik, Sweden. Their AI software identifies species—capercaillie, three-toed woodpecker, and others—by analyzing audio from sensors placed in forests, no field visits required. Having already processed 2,700+ hours of recordings, the startup has joined BizMaker's incubator and is running pilots with forestry companies and government agencies.
DeepForest: Standard Drone Cameras Now See Through the Canopy
Researchers from Johannes Kepler University, Helmholtz Center, and Leipzig University published DeepForest—an AI system using 3D convolutional neural networks and synthetic-aperture focal stacking to reconstruct spectral data through entire forest volumes using standard multispectral drone cameras. Accuracy improved 2–12x over conventional imaging, with field tests achieving MSE = 0.05, enabling full 3D NDVI mapping. The low hardware cost makes it a scalable alternative to LiDAR for carbon accounting and biodiversity monitoring.
Skogforsk: Boom-Tip Control Makes Novice Harvester Operators 9.5% Faster
A Skogforsk field experiment with 18 student operators (Komatsu 911 equipped with Smart Crane) found that boom-tip control—where the operator steers the crane tip directly rather than each individual joint—cut crane task time by 9.5% in thinning-like conditions. Since crane work accounts for up to 90% of harvester operating time in thinning, the authors recommend boom-tip control be made standard on new machines and integrated into all operator training programs.
Sweden's Forest Industries Update Unified Protected Area Map
Triona has updated the data behind their web map consolidating all of Sweden's formally protected and voluntarily set-aside forest areas — national parks, nature reserves, biotope protections, and industry-voluntary areas — into one publicly accessible platform. Commissioned by Sweden's largest forest owners, the tool gives politicians, researchers, and the public a complete picture of where roughly 25% of Swedish forests are protected.