Every tree has a fingerprint: combating illegal timber trade
The trade of illegal timber is growing worldwide: timber with unknown origins increasingly finds its way to the market through shady routes. To combat this, Thomas Mortier (UGent) and Victor Deklerck (World Forest ID, Plantentuin Meise) developed an AI model that can trace the origin of timber via a unique ‘chemical fingerprint’. Their research was recently published in the prestigious journal ‘Nature Plants’.
15 to 30% of global timber trade is illegal
Illegal logging is one of the most profitable forms of environmental crime, accounting for 15 to 30% of the global timber trade. The damage is severe, both environmentally and socio-economically.
Russia’s invasion of Ukraine shows how geopolitical events exacerbate this problem. The EU, US and UK decided to punish the Russian invasion with (among other things) an import ban on wood products from Russia and Belarus. A number of labels such as the well-known Forest Stewardship Council (FSC) label also has decided to exclude Russian timber from now on.
But these sanctions obviously do not reduce the demand for timber, quite the contrary: criminals find ways to circumvent the sanctions and import timber through illegal routes, masking the timber’s origins.
“ Thanks to our new method, FPS Environment inspectors found evidence of Russian timber being imported into Belgium; a clear violation of embargo. ”
The chemical fingerprint of wood
To stop the illegal trade, we investigated a way to prove where wood really comes from. We found a solution in the form of ‘chemical fingerprints’.
Prior to our research, World Forest ID and Preferred by Nature first collected thousands of wood samples from 24 tree species from 12 countries. The samples were analysed for:
- stable isotopes, chemical elements that tell something about the soil and climate in which the tree grew;
- concentrations of trace elements, unique chemical elements linked to growth location.
This gave each piece of wood its own ‘chemical fingerprint’, unique to the place where the tree grew. We combined that fingerprint with information on the distribution of tree species. Using AI – through Gaussian processes – we built a model that predicts where a piece of wood comes from based on its fingerprint:
- the model compares the chemical fingerprint to the wood samples it already knows;
- it calculates the most likely origin of the tree;
- it gives an estimate of the uncertainty of the prediction.
Gaussian model
Gaussian models are powerful statistical models that recognise and predict patterns in data. They are widely used in AI: for time series analysis, classification and regression problems.
A Gaussian process is a probabilistic model consisting of a collection of random points (e.g. chemical fingerprints of wood samples with a GPS location). Each subset of the points is assumed to be normally distributed. The relationship between the points is modelled using so-called kernel functions. The functions can then be used to determine how to make a prediction for unknown points.
As such, a Gaussian model can describe data (chemical fingerprints) and predict data in a flexible and powerful way. Moreover, the model can also estimate the uncertainty of the prediction, allowing for better interpretation of the results and better-informed decisions.
There's proof: Russian timber on the Belgian market despite embargo
The method has already proven its worth: inspectors from the Belgian Federal Public Service (FPS) Environment have already discovered Russian timber in Belgium, despite the embargo. This is undoubtedly a first discovery; trade from other high-risk areas such as Indonesia, Brazil and Congo can also be better controlled from now on.
Flanders AI Research Programme
This research was developed as part of the Flanders AI Research Programme.
Thomas Mortier
After obtaining a master degree in Computer Science and a master degree in Statistical Data Analysis at Ghent University in 2017, Thomas Mortier started as an assistant at the KERMIT research group at the Faculty of Bioengineering Sciences. In addition to obtaining a PhD, he provided support for courses on statistics and machine learning and supervised master and PhD students. He currently works as a researcher, focusing on uncertainty in machine learning with applications in applied biological and climate sciences. He also has experience as a data scientist and is involved in teaching various courses on machine learning at the Ghent University Academy for Engineers (UGain).
Expertise: machine learning, artificial intelligence, data science, statistical data analysis
Victor Deklerck
Victor Deklerck obtained his PhD in Bioengineering Sciences: Natural Resources at Ghent University, in collaboration with the Museum for Central Africa. He is a research associate at the Plantentuin Meise and scientific director of the World Forest ID not-for-profit organisation. He leads the team in the development of chemical and genetic techniques, combined with geospatial machine learning models, to trace wood and forest risk commodities.
