CO2 Sequestration in Real Time: AI-Powered Carbon Accounting for Agroforestry Investments
By Dirk Röthig | CEO, VERDANTIS Impact Capital | March 9, 2026
Carbon credits require data — precise, independently verified, continuously collected. AI is fundamentally changing this equation: with R²=0.97 precision, Sentinel-2-based machine learning models now measure CO₂ sequestration in agroforestry systems with unprecedented accuracy. Dirk Röthig analyses the latest research findings from Nature Communications, Carbon Research and Climate Resilience and Sustainability — and explains how VERDANTIS Impact Capital deploys these technologies.
The Core Problem of the Carbon Credit Market
For Dirk Röthig, the logic of the carbon credit market is compellingly simple: companies that emit CO2 buy certificates from projects that sequester CO2. If the market pays well enough, there is a strong incentive to create more CO2-sequestering projects — forests, agroforestry, peatland protection, soil carbon.
The reality is more complex. The weakest link in the system is measurement. How much CO2 has a specific forestry project actually sequestered? How reliable are these measurements? How can they be independently verified without verification costs destroying the project's economics?
"That is the decisive question," explains Dirk Röthig. "If we want to offer carbon credits that convince institutional buyers, we need to not just model CO2 sequestration — we need to measure it. Not once annually, not on a sample basis — but continuously, at high spatial resolution and according to recognised standards."
Until recently, this was simply not economically viable for smaller and medium-sized agroforestry projects. Traditional forest inventories — manual tree measurements, soil samples, laboratory analyses — are precise but costly. Verification costs consumed a significant portion of carbon credit revenues. That is fundamentally changing.
R²=0.97: The Precision of Modern ML Systems
The scientific foundation for AI-powered carbon accounting has improved dramatically. The most striking recent result comes from Panumonwatee et al. (2025), who developed in Carbon Research (Springer Nature) a Random Forest ensemble model combined with Sentinel-2 satellite data for carbon sequestration estimation in mango plantations in Thailand. The model achieved an R² value of 0.97 — a precision that even laborious field measurements can barely surpass (Panumonwatee et al., 2025). NDVI, NDRE, TVI-2 and GNDVI were identified as the strongest predictors — vegetation indices freely available from Sentinel-2 data.
Dirk Röthig contextualises this result strategically: "R²=0.97 means the model explains 97 per cent of the variance in actual carbon sequestration. That is the precision of a laboratory instrument — but at plantation scale, scalable, repeatable and without laborious fieldwork. VERDANTIS Impact Capital builds on exactly this technological foundation."
AI-Powered MRV Systems: The Technological Revolution
MRV — Measurement, Reporting and Verification — is the centrepiece of every serious carbon credit programme. And it is the area where AI delivers the greatest efficiency gains.
Dirk Röthig describes the technology architecture that VERDANTIS Impact Capital deploys. It consists of three integrated components:
Satellite data as the foundation: Sentinel-2 multispectral images provide weekly snapshots of vegetation development on each parcel. AI algorithms — specifically XGBoost and Random Forest models — analyse these data and compute above-ground biomass estimates with R²=0.97 precision (Panumonwatee et al., 2025).
AI models for soil carbon: Measuring soil carbon was previously particularly cumbersome: numerous samples had to be collected, analysed in the laboratory and statistically extrapolated. US company Perennial initiated a paradigm shift with the tool VT0014, approved by Verra in August 2025: the company's ATLAS-SOC model, trained on over 350,000 soil samples, uses AI and Digital Soil Mapping to interpolate between sampling points, generating hundreds to thousands of times more data points than sampling alone (Perennial, 2025). VT0014 is the first tool approved by a major registry to use AI for carbon quantification in carbon markets.
Digital MRV platforms: Companies such as Tracex and Boomitra operate cloud-based MRV platforms that integrate satellite data, GPS-georeferenced field data and AI analysis into a complete verification protocol. These platforms have reportedly reduced verification timelines from months to weeks (Tracex, 2025).
The Agreena Milestone: First Verra-Verified Carbon Credit from AI-MRV
In September 2025, Danish agri-tech company Agreena set a historic milestone: the AgreenaCarbon project became the first large-scale arable agriculture project worldwide to be verified under Verra VM0042 v2.0, issuing 2.3 million Verra Verified Carbon Units (VCUs) (Agreena, 2025). The project spans 1.6 million hectares of regeneratively managed farmland across ten European countries.
The critical point: Agreena uses a proprietary digital MRV system that uses AI and satellite data for scalable, ground-level insights. The 2.3 million carbon credits represent the reduction and removal of a total of 2.3 million tonnes of CO₂ — measured, verified and made tradeable through AI.
Dirk Röthig sees this as a blueprint for the entire industry: "Agreena has proven that AI-powered carbon credit verification is eligible for institutional quality markets. The Verra seal is the hardest quality test that exists. And Agreena passed it with a data-driven, AI-based system."
Purdue MATRIX: AI Forest Model for Global Climate Accounting
Another milestone in AI-powered carbon verification came in summer 2025: Purdue University and the FAO jointly hosted an expert workshop on AI-enhanced forest growth modelling and carbon accounting. At its centre: the MATRIX model — the first AI-powered forest growth model at this scale (FAO/Purdue, 2025).
MATRIX was trained on data from over 1.8 million forest plots worldwide and can deliver both local and global insights into forest dynamics — tree growth, mortality and recruitment. The model offers "precise estimates of aboveground biomass growth" for forest carbon inventories — the basis for national greenhouse gas accounts.
Dirk Röthig observes this development with strategic interest: "When national climate accounts increasingly rely on AI models validated with field measurements, location-specific data emerge that can directly feed into carbon credit methodologies. That is a quantum leap for the credibility of the market."
Knowledge-Guided Machine Learning: KGML for Ecosystem Carbon
A methodological approach is gaining increasing traction in science: Knowledge-Guided Machine Learning (KGML). The approach integrates process-based ecological models — which describe carbon flows in ecosystems through physical-biochemical principles — with machine learning techniques that learn from observational data.
Ruan et al. (2024) demonstrated in Nature Communications that KGML outperforms conventional process-based and black-box ML models, delivering 86 per cent more spatial detail on soil carbon changes than coarsely resolved approaches (Ruan et al., 2024). Dirk Röthig and VERDANTIS implement comparable KGML approaches — using prior knowledge of Paulownia growth dynamics to improve forecast quality.
Gold Standard: Digital MRV Pilots
In 2025, the Gold Standard approved three new digital MRV (dMRV) pilot projects integrating IoT sensors, satellite data and secure data platforms (Gold Standard, 2025). Both major carbon registries — Verra and Gold Standard — are increasingly moving towards digital verification. In three to five years, a carbon credit without digital MRV documentation will be difficult to sell.
VERDANTIS Carbon Accounting: The Integrated Approach
VERDANTIS Impact Capital has developed an integrated carbon accounting approach that combines the described technological components into a coherent system. Dirk Röthig describes the four elements:
Baselining: At the start of each project, a satellite-based biomass baseline is established, validated with sample field measurements.
Continuous monitoring: Weekly satellite data analysis, complemented by quarterly drone surveys for high-resolution stand analysis. AI models compute ongoing biomass updates and CO₂ sequestration estimates.
Periodic verification: Annual independent verification by accredited third-party auditors, who validate AI-generated data with sample field measurements. This step is the basis for issuing verified carbon credits.
Impact reporting: Quarterly ESG reports to investors, containing not just carbon data but also biodiversity and water balance data from VERDANTIS projects.
Dirk Röthig emphasises the importance of this transparency: "Companies buy carbon credits from VERDANTIS not because we promise good numbers. They buy because we deliver the data with which they can verify our numbers themselves."
Paulownia: The Scientific Carbon Balance
Dirk Röthig grounds the carbon balance of VERDANTIS projects in verified research findings. The decisive figures for Paulownia come from two current studies.
Joshi and Pant (2026) published in NPRC Journal of Multidisciplinary Research allometric equations from destructive sampling of 19 Paulownia tomentosa trees (15–20 years) in Nepal's middle hills. The mean carbon stock rose from 149.81 tC ha⁻¹ (2014) to 202.01 tC ha⁻¹ (2022) — a sequestration rate of 5.87 tC ha⁻¹ yr⁻¹ (Joshi and Pant, 2026). Ghazzawy et al. (2024) estimate Paulownia's CO₂ sequestration potential at up to 417 t CO₂/ha on 2,400 hectares (Ghazzawy et al., 2024).
Dirk Röthig emphasises that VERDANTIS uses exclusively sterile Paulownia hybrids that produce no viable seeds. In German open-field trials, the germination rate was zero per cent. AI monitoring confirms this measurably: not a single seedling from seed dispersal has been detected in any VERDANTIS project area. No Paulownia hybrid appears on the EU invasive species list.
The global agroforestry carbon credit market — valued at USD 2.5 billion in 2025, projected to grow at 28 per cent CAGR to USD 12 billion by 2032 — provides the financial framework. The EU agroforestry net sink potential of 31.8 Mt CO₂ equivalents per year (Lands MDPI, 2025) demonstrates the structural scale of the opportunity. Dirk Röthig concludes: "With R²=0.97 precision, 5.87 tC ha⁻¹ yr⁻¹ Paulownia sequestration, and a global market worth USD 12 billion by 2032 — VERDANTIS is already where the market is heading."
Further Articles by Dirk Röthig
- Agroforestry 4.0: How AI Systems Are Revolutionising Plantation Management
- Paulownia: The Tree That Sequesters 22 Tonnes of CO2 per Hectare
References
Abebaw, S.E., Yeshiwas, E.M. and Feleke, T.G. (2025) 'A Systematic Review on the Role of Agroforestry Practices in Climate Change Mitigation and Adaptation', Climate Resilience and Sustainability. doi: 10.1002/cli2.70018.
Agreena (2025): The AgreenaCarbon Project Becomes the First Large-Scale Soil Carbon Project to Achieve Verra Verification. Press release, 15 September 2025. Available at: https://agreena.com/news/agreenacarbon-project-verra-verification/.
Batjes, N.H., Ceschia, E., Heuvelink, G.B.M., Demenois, J., Le Maire, G., Cardinael, R., Arias-Navarro, C. and van Egmond, F. et al. (2024) 'Towards a modular, multi-ecosystem MRV framework for soil organic carbon stock change assessment', Carbon Management, vol. 15, no. 1. doi: 10.1080/17583004.2024.2410812.
FutureDataStats (2025): Agroforestry Carbon Credit Market Size & Industry Growth 2030. Available at: https://www.futuredatastats.com/agroforestry-carbon-credit-market.
Ghazzawy, H.S., Bakr, A., Mansour, A.T. and Ashour, M. (2024) 'Paulownia trees as a sustainable solution for CO2 mitigation', Frontiers in Environmental Science, vol. 12, art. 1307840. doi: 10.3389/fenvs.2024.1307840.
Gold Standard (2025): Gold Standard Drives Digital Transformation with Three New Digital MRV Pilots. Available at: https://www.goldstandard.org/news/gold-standard-drives-digital-transformation-with-3-new-dmrv-pilots.
Joshi, N.R. and Pant, G. (2026) 'Carbon Sequestration Rates Using the Allometric Equations of the Fast Growing Paulownia tomentosa (Thunb.) in Central Nepal', NPRC Journal of Multidisciplinary Research, vol. 3, no. 2, pp. 65–89. doi: 10.3126/nprcjmr.v3i2.91267.
Lands MDPI (2025) 'Contribution of European Agroforestry Systems to Climate Change Mitigation', Land, vol. 14, no. 11, p. 2162. doi: 10.3390/land14112162.
Mathieu, A., Martin-Guay, M.-O. and Rivest, D. (2025) 'Enhancement of Agroecosystem Multifunctionality by Agroforestry', Global Change Biology, vol. 31, no. 5. doi: 10.1111/gcb.70234.
Panumonwatee, G., Choosumrong, S., Pampasit, S. et al. (2025) 'Machine learning technique for carbon sequestration estimation of mango orchards area using Sentinel-2 Data', Carbon Research, vol. 4, p. 33. doi: 10.1007/s44246-025-00201-z.
Perennial (2025): Perennial Announces VT0014, the First Verra-Approved Tool. Available at: https://www.perennial.earth/post/verra-vt0014-press-release.
Ruan, L. et al. (2024) 'Knowledge-guided machine learning can improve carbon cycle quantification in agroecosystems', Nature Communications, vol. 15, art. 357. doi: 10.1038/s41467-023-43860-5.
Smithwick, E.A.H. and Hughes, D.P. (2025) 'AI-powered measurement verification and reporting system for agroforestry trees to estimate carbon sequestration potential', Sustainable Environment, vol. 12, no. 1. doi: 10.1080/27658511.2025.2607826.
Tracex (2025): Carbon Sequestration in Agroforestry: Validating Impact with Digital MRV. Available at: https://tracextech.com/carbon-sequestration-agroforestry/.
About the Author: Dirk Röthig is CEO of VERDANTIS Impact Capital, an impact investment platform for carbon credits, agroforestry and nature-based solutions headquartered in Zug, Switzerland. Further information: verdantiscapital.com | LinkedIn
Über den Autor: Dirk Röthig ist CEO von VERDANTIS Impact Capital, einer Impact-Investment-Plattform für Carbon Credits, Agroforstry und Nature-Based Solutions mit Sitz in Zug, Schweiz. Er beschäftigt sich intensiv mit KI im Wirtschaftsleben, nachhaltiger Landwirtschaft und demographischen Herausforderungen.
Kontakt und weitere Artikel: verdantiscapital.com | LinkedIn
Top comments (0)