Brazilian scientists have mapped the oxygen locked inside Amazon wood closely enough to exclude 92 per cent of the forest as the origin of any seized log. That is according to Luiz Antonio Martinelli, a professor at the Center for Nuclear Energy in Agriculture in Piracicaba, whose team published the reference map in the journal Molecules.
“You can’t falsify stable isotopes,” Martinelli said of a method built for Brazil’s Federal Police, whose National Institute of Criminalistics contributed co-author Fabio José Viana Costa to a paper carrying 17 names.
Rain moving inland from the Atlantic sheds its heavier oxygen as it falls, so wood from the western Amazon holds more oxygen-16 and less oxygen-18 than timber cut in the east. Lead author Ana Claudia Gama Batista traced that gradient southeast to northwest through 387 trees at 25 sites, from Atalaia do Norte on the Peruvian border to Bragança on the Atlantic coast.
Both models produced the same picture, with a climate regression explaining 70 per cent of the variation in the wood and a machine-learning version reaching 67 per cent. Analyses were conducted across laboratories at the University of California, Davis, and São Paulo State University at Botucatu, with precision better than 0.26 parts per thousand.
Batista drew those figures from site averages, and running the same models on single trees yields 42 per cent for the climate regression and 44 per cent for the machine-learning version. A seized log is one tree, and never a site means, so the lower pair governs a criminal case.
Resolution is not what holds the method back; the map is drawn at 18 square kilometres per cell, and the chemistry setting the real limit, because oxygen varies by just 3 parts per thousand across the whole basin, and the researchers put the model’s own error at 0.7. That margin allows confident calls between the far western and far eastern Amazon and much weaker ones through the middle.
The researchers collected the fewest samples inside the arc of deforestation, the frontier band the paper names as a zone of intense land-use change and continuing illegal harvesting, and the two models diverge most sharply across exactly that ground.
Every reference tree was cut legally, with samples taken from trunks felled under approved management plans at a minimum diameter of 50 centimetres, in line with Brazil’s CONAMA Resolution 406 of 2009. Sampling was conducted through the legal trade itself, with RRX Timber facilitating collection under its concession in the Amapá National Forest and Forest Ark assisting in the Jamari National Forest in Rondônia, meaning the library that will test suspect wood was built within the industry it is meant to police.
It comes as the European Commission adopted two measures on 13 July setting the product scope and the due diligence system for the EU Deforestation Regulation, which starts applying to large and medium operators on 30 December 2026. That standard asks for the geolocation of the single plot where a consignment was grown, a precision no isoscape can yet deliver.
Martinelli’s group is expanding the model before it reaches casework and now holds samples from 800 trees across 63 sites. Carbon and nitrogen are being mapped through a doctoral project run by Isabela Maria Souza-Silva, with deuterium and strontium named in the paper as the tracers to follow.
Origin testing has been moving from the laboratory towards the border for several years, with oxygen ratios already used to date and place the Forbidden City’s roof timbers. Adelaide University has built a genetic and chemical library to expose Russian wood entering Australia, and testing commissioned by Canberra found half of all sampled timber imports were unable to prove where they grew.
Martinelli’s strongest model clears 3 million square kilometres of forest, and still leaves 240,000 square kilometres where a seized log could have grown, an area the size of the United Kingdom.
For more information: Batista, A.C.G., Araújo, M.G.S., Souza-Silva, I.M. et al. Isoscape of oxygen stable isotopes in woods of the Amazon. Molecules 31, 1542 (2026). https://doi.org/10.3390/molecules31091542
