Editor's note
Two labs, two sensors, the same number.
Finland measured nine species from a plane and got 92 percent. In Galicia they measured seven with a handheld scanner on the ground and got 93. Different forests, different species, different hardware, and the accuracy lands in the same place. Species has stopped being the thing you send a crew out to confirm.
What replaces it is duller and more useful. What does a hectare cost to cover, how often can you repeat it, and who owns the sensor. Those are procurement questions rather than research questions, and they answer differently depending on whether you fly, walk, or wait for a satellite pass.
Which is why the companies worth watching now are the ones deciding where to put the sensor, not how to sharpen it.
Axel
Finland's FGI identifies nine tree species from the air at 92 percent accuracy
The sensor is a multispectral airborne laser scanner, and the species list runs well past the easy three: aspen, rowan, alder, oak, linden and maple sit alongside pine, spruce and birch. Markus Holopainen's group at the University of Helsinki puts the average across all nine at 92 percent, measured on extremely dense 3D point clouds. His own framing is that a single multispectral scan now comes close to what laser and aerial imagery deliver together.
Universidade de Vigo tells seven commercial species apart at 93 percent from the ground
Same question from the other end of the sensor stack, and from the ground. Laura Alonso's team at Universidade de Vigo walked the stands with a GeoSLAM ZEB Horizon collecting 300,000 points a second, then flattened the point clouds into 2D images and ran a convolutional net over them. That beat the Random Forest baseline on the same data, 93 percent against 72. The species are the ones Galicia actually sells: two eucalypts, two pines, chestnut, oak and Lawson cypress.
Mai Intelligence moves from renting satellite eyes to building its own
Mai Intelligence AB (m-ai.earth) watches Swedish private forest on Copernicus Sentinel-2 today, around 5,400 hectares by its own count, backed by Almi and Västra Götalandsregionen. The company says phase two takes it down the stack into its own sensing hardware: founder Selvin Jayakumar has brought in Simon Håkansson to build an edge-AI CubeSat, with a prototype going up over Värmland soon. The target is spruce bark beetle caught while the crown is still green, and that first flight goes to the stratosphere rather than orbit.
Fraunhofer IIS wires 200 square kilometres of Bavarian forest with mioty sensors
The radio is mioty rather than LoRaWAN, picked to hold a signal under canopy. Ground sensors read air temperature, wind, soil moisture and CO2, and AI cameras at three sites watch for smoke. Tennenloher Forst has run one camera since September 2025, with the full build due late summer across roughly 200 square kilometres. Bundesforst, Bayerische Staatsforsten and the Erlangen fire inspectorate are all in it.
WoodComp opens a free database of the world's pulp mills and sawmills
Disclosure: I built this one. Judge it accordingly.
WoodComp (woodcomp.world) has been rebuilt from the ground up. Every pulp mill, industrial sawmill and biorefinery in the set carries verified coordinates, owner, capacity, raw material and operating status, and 1,146 of the 1,236 facilities are running today across 52 countries. It is CC BY-SA 4.0, free to use and free to check, with a knowledge bank of conversion factors behind it. A fibre balance view comes next.