Editor's note
Nobody measured anything new this week.
I have argued here before that the value sits in data already collected for other reasons, and two of this week's stories are that argument again. Lantmäteriet, Sweden's national mapping agency, publishes its maps and laser scans free, and a model trained on them now reads land cover and canopy height across the country. In Arkansas, a team went after the merchantable weight of standing oaks without felling any. They gave a model only diameter and height, the two numbers every inventory already writes down, and it landed closer to the true weight than the equations foresters normally use.
Reading an existing record has become the cheap part of the job, and that changes where the next budget goes. Before commissioning a flight campaign or a sensor network, the question worth asking is what the records already hold and who could be paid to read them properly.
Axel
The Intelligence Company is not a forestry company
The ingredients were all sitting in the open: Meta's DINOv3 backbone, pretrained on satellite imagery, and the maps, orthophotos and airborne laser scans Lantmäteriet publishes under a free licence. The Intelligence Company, an AI firm, ran 287,612 tiles from 17 counties through them and published the result on Hugging Face, ten ground classes and a canopy height for every square metre. By its own count one graphics card covers 2,800 square kilometres an hour. The model card names the catch: land use, not tree cover, so a fresh clearfelling still reads as forest.
A rulebook from 1998 starts taking questions
Ask what the rules require on a given forestry job and the honest answer usually depends on how the firm has written its own routines. Regelrätt skogsbruk has held both halves since Skogforsk built it: Swedish environmental and work-safety law, and each customer's own working methods entered alongside it. The rebuilt version adds an AI function that, the institute says, answers from that combination rather than the statute alone, so two firms asking the same thing get different replies.
Palantir's chief executive buys 15,000 hectares of Härjedalen
SCA had judged the ground not optimal for forestry. Alex Karp paid 235 million kronor for it anyway, through Cladonia Skog AB, a company registered for the occasion. The purchase agreement says Cladonia is there to own and manage the property and conduct forestry. Its chairman says the land offers habitat for wildlife and room for recreation. What the ground ends up being used for is the part worth watching.
Dryad's fire network told the real ignitions from the decoys
XPRIZE ran the final of its wildfire contest in the Alaskan bush in June: live fires, plus lures built to trigger a false alarm. Berlin's Dryad, last here in #042 with 1,100 units on Colorado aspen, fielded 170 solar-powered gas sensors and flagged the real ones inside twelve minutes. Silvaguard drones then flew out, confirmed by infrared camera and dropped up to 100 litres of suppressant. That won second prize on 23 September, 800,000 dollars against Anduril's 2.2 million.
Two ways to guess an oak's weight, tested on the same 1,175 trees
Every estimate of what a standing oak weighs traces back to oaks that were cut down and put on a scale. A team led by the Arkansas Forest Resources Center set the equations built that way against a machine-learning model given the same tape-measure inputs, diameter and height, on 34 bottomland hardwood plots in the Arkansas Delta measured between 2015 and 2023. The equations landed within 17.2 percent of the measured merchantable weight on average, the model within 12.4. Both are good enough to plan with, the authors say, and neither needs a tree on the ground.