Russian AI Will Monitor Carbon Stocks in Forests
Researchers from Skoltech, Irkutsk Polytechnic, and the AIRI Institute have developed a neural-network tool that can effectively see through forest canopies to estimate carbon stocks in forests. Just as importantly, it can identify the limits of its own confidence, highlighting areas where its predictions should be verified.
Existing algorithms that process satellite imagery often produce little more than numerical estimates of timber volume. But what happens when the model gets it wrong? Verifying thousands of hectares through field surveys is practically impossible. The new system takes a different approach. It could become a key tool for achieving carbon neutrality – an ambitious goal that Sakhalin has already begun to realize.
An Honest Algorithm
The new solution, built on the XGBoost machine-learning framework, identifies tree species with 83% accuracy, estimates tree age with 70% accuracy, and, most importantly, calculates carbon stocks. The latter remains the most challenging parameter for remote sensing, and current accuracy ranges from 53% to 63%.
The real breakthrough, however, lies in a shift of methodology. The system generates a "confidence map." Areas where calculations are considered reliable are marked in green, while zones requiring additional verification appear in red. To train the model, researchers created a digital representation of the unique ecosystems found in three districts of Sakhalin – Korsakovsky, Nevelsky, and Kholmsky. The system was also fed Sentinel-2 satellite imagery, topographic information, and field-collected data.
A Climate Experiment
The Sakhalin Climate Experiment began in 2022. The region became a testing ground for carbon-regulation mechanisms. Russia's Ministry of Economic Development has recognized Sakhalin Region as the first Russian region to achieve carbon neutrality ahead of schedule. Greenhouse gas absorption now exceeds emissions.
"Thanks to the experiment, Sakhalin Region has become a dynamically developing territory both economically and environmentally. It offers clean air and unique natural ecosystems. It is a region where people want to live, work, and spend their leisure time," Sakhalin Governor Valery Limarenko said.
Russia is home to more than 20% of the world's forests. According to the Russian Science Foundation, this national asset stores approximately 55.8 billion metric tons of carbon. Yet managing that resource effectively is impossible without an accurate digital twin of the forest. In 2023, Russian President Vladimir Putin approved an updated Climate Doctrine that established a national target of achieving carbon neutrality by 2060. That same year, the Space Research Institute of the Russian Academy of Sciences launched Carbon-E (Uglevod-E), an information and analytical platform for monitoring carbon stocks in Russia's forests and other ecosystems.
Toward Smarter Data
In the past, AI primarily collected statistics. Now it is beginning to evaluate the reliability of its own conclusions. That marks a transition from big data to smart data. More precise monitoring could accelerate the detection of illegal logging, forest disease outbreaks, and windthrow damage. Trustworthy AI may also help companies audit forest-climate projects without relying on costly field expeditions.
"Climate policy remains a long-term priority for the government. It affects the competitiveness of our products in international markets and Russia's participation in emerging industries where we have the expertise to become a leader. These include low-carbon products and sectors such as aluminum production, fertilizers, petrochemicals, lithium, nuclear power, hydropower, and renewable energy," Russia's Minister of Economic Development Maxim Reshetnikov said.
The researchers say they plan to scale the system significantly. The algorithm is expected to learn how to distinguish biodiversity patterns across the entire country. The technology could also attract interest from nations with vast and difficult-to-access forest resources.
