Three years ago, the European Union (EU) passed the most ambitious anti-deforestation legislation in human history, attempting to pull the handbrake on market pressures that are deforesting the equivalent of Greece each year for farmland. The European Union Deforestation Regulation (EUDR) requires producers of major commodities, like coffee, palm oil, and beef, to certify their products were not produced on deforested land.
For developing countries whose economies depend on agricultural exports, the scale of the compliance challenge is enormous. Take just one commodity: coffee. Some 12.5 million coffee farms worldwide, 85% of which are “smallholders” tending plots of land no larger than several football fields, will need to certify that their land was not deforested in order to sell their coffee to the EU.
Verifying this involves a mix of farmer-reported information and data from GPS recordings and satellite imagery, with an ecosystem of providers offering solutions for integrating this data and detecting deforestation. But these solutions vary widely in their approaches and associated costs. “Can we identify cost-effective ways to achieve EUDR compliance, and ensure compliance requirements are not excluding smallholder farmers from EU markets?” asked Dr. Seth Goodman, a research scientist at AidData. “That’s the central challenge we’re trying to solve.”
Supported by the Gates Foundation, Dr. Goodman and fellow AidData researchers have partnered with the Committee on Sustainability Assessment (COSA) to assess the accuracy, cost-effectiveness, and inclusiveness of using Earth Observation (EO) data to reach EUDR compliance for coffee producers in Central America, one of the world’s largest coffee-producing regions.
To that end, the research team pulled together and crunched vast amounts of data: geospatial boundaries of coffee plots recorded with high precision GPS devices, very high-resolution satellite imagery, drone-based validation, and a nationally representative survey of Honduran coffee farmers. William & Mary’s High-Performance Computing clusters, a supercomputing network, was leveraged to store and process some of this imagery and data.
“While our initial study is focused on Central America and Honduras in particular, we think it holds valuable and actionable insights not just for the broader coffee sector, but also for other commodities covered by the EUDR around the world,” said Jacob Hall, a data analyst at AidData and a co-author of the study. Hall will meet with coffee producers and exporters this week in Addis Ababa at the African Fine Coffee Association (AFCA) conference to preview key findings and hear from government and private sector actors.
The resulting report, to be published later this month, is “a data-driven and fact-driven analysis of what works, what doesn’t, and realistic approaches and options to solving these problems,” said Dr. Goodman. “No solution is going to be perfect. Even with the extremely high-quality data we collected for this study—including additional field work, custom tasking of satellites, and drone imagery—which would not be practical at scale, the deforestation detection solutions and the humans doing manual verification could not always tell when areas were deforested or not. The real question is, how do we help farmers get to ‘good enough’ and advocate for compliance solutions that don’t leave smallholders responsible for addressing their limitations?”
The researchers uncovered some unexpected findings. Collecting individual plot locations for every farmer contributing to EU-bound coffee shipments is a massive undertaking. How precisely the locations or boundaries of plots are measured can significantly impact the time it takes to collect the data, and what level of precision is needed for accurate deforestation assessment has been an open question. But the researchers found that boundary precision has a relatively limited influence on the ability of a model or algorithm to detect deforestation, compared to other factors. “This suggests that lower-cost, lower-precision tools are workable, which would decrease costs and the barriers to participation,” Hall said.
The study also revealed that the six deforestation detection solutions from various providers tested performed well overall and were in full agreement for over 80% of sites, where no deforestation was found. However, for the remaining 20% of sites, where at least one provider detected deforestation, there were zero locations where all providers agreed on the deforestation status and agreement between multiple providers was generally low.
This may be due to how the nature of coffee farming itself interacts with EUDR rules. Coffee is typically shade-grown beneath some amount of tree canopy, and the definition of “deforestation” used by the EUDR does not include provisions for tree removal where agroforestry systems previously existed—making actual deforestation detection particularly challenging. The results suggest that incorporating validation procedures to allow farmers to address “false positives” will be critical to ensuring they are not unfairly penalized. As a next step, the team will share provider-specific results with deforestation detection providers to help them improve their approaches.
Goodman, Hall, and their collaborators are turning now to explore the options and challenges for EUDR compliance faced by another major coffee producer: Ethiopia. And they are interested in expanding this research down the line to other commodities as well—EUDR will eventually apply to some $110 billion of goods imported by the EU annually, including cacao, cattle, palm oil, rubber, soy, and wood, in addition to coffee. The EU has delayed enforcement until the end of 2026 for large producers and mid-2027 for small producers, to give global supply chains more time to adapt. While the delay provides some breathing room for compliance efforts, the new deadline still represents a tight schedule to collect locations of millions of farmer plots worldwide in the coffee sector alone, as well as setting up and testing the broader compliance workflows.
