New mapping system ends farm mislabeling, protecting coffee and cacao trade

Sample Earth, launched by the Alliance of Bioversity International and CIAT, helps mapmakers build accurate, inclusive maps to prevent smallholder farmers from being wrongly classified as producing major commodities in forested areas. Misclassification risks excluding compliant producers from markets enforcing deforestation-free rules, particularly the European Union’s new regulation (EUDR). The initiative is the result of a collaboration between Alliance researchers, tech companies (including Google), and the World Cocoa Foundation. Researchers call on private-sector mapmakers to adopt their model to harden their supply chains against disruption. 

Producers of coffee and cacao, and the companies that buy their products, could soon lose access to the world’s second-largest economy. The European Union, at the end of next year, will phase in the long-delayed EUDR legislation that requires many agricultural commodities to be certified deforestation-free. Unfortunately, hundreds of thousands of producers will face considerable hurdles, and not because they produce on land that hasn’t been deforested since 2020 (the EU’s cutoff date): It’s due to maps that wrongly classify their farmland as forest. 

For example, the EU’s main reference map, published in 2025, misclassifies more than half the coffee production zones in Colombia, China, Guatemala and Mexico as forest, according to research by the Alliance of Bioversity International and CIAT. Similar reference maps have the same shortcomings. This is because these maps are 'trained' on land-cover datasets that largely exclude remote areas cultivated by smallholders. 

Improving these maps is urgent. To spark the creation of better maps, the Alliance recently launched Sample Earth, a trusted and inclusive global benchmark and reference dataset that accurately represents remote smallholder farms. The initial data tranche includes approximately 100,000 open-access, time-stamped geolocation points in Ghana and Vietnam. The countries are the second-largest producers of cacao and coffee, respectively. 

“Maps are needed for due diligence, and buyers will likely steer clear of areas misclassified as ‘high risk’ for deforestation,” said Louis Reymondin, a data scientist at the Alliance. “With Sample Earth, we invite governments, companies, NGOs and research institutions to invest in expanding this inclusive, high-quality land-cover reference to preserve livelihoods and incentivize environmental protection.” 

Smallholders produce an estimated 60% of the world’s coffee and 90% of its cacao. If maps used for compliance are inaccurate, buyers may decline purchases from entire regions rather than risk penalties for non-compliance, effectively shutting smallholders out of major markets.

“Most maps are not accurate at local scales because the data is biased toward regions with a lot of training data,” said Thibaud Vantalon, a scientist at the Alliance’s Digital Inclusion research area. “Remote regions are very poorly mapped. Sample Earth means to fill this gap in training data for smallholders.” 

Making map-making better 

Sample Earth is designed to improve map accuracy and to streamline the map-making workflow. Data scientists, the people who make maps with satellite imagery, spend an estimated 80% of their time collecting, cleaning and organizing training data. Sample Earth provides reference samples to reduce that burden and speed up the creation of accurate land-cover maps for compliance. 

“High-quality data and data-based action are the foundation for compliance with deforestation-free rules and net-zero carbon emission targets,” said Michael Matarasso, the Impact Director and Head of North America at the World Cocoa Foundation (WCF), a partner in Sample Earth. “However, highly accurate public data is rare… This poses a significant risk to all stakeholders involved. A standard to deliver highly accurate and transparent data in partnership with governments and farmers is of critical importance more than ever.” 

Sample Earth aims to set a new transparency and quality benchmark for map-based compliance tools. Currently, no universal standard exists for third-party accuracy assessments of maps used in deforestation due diligence. Sample Earth plans to include a built-in improvement mechanism that allows mapmakers to access confidential land-use reference data to validate and refine their maps without exposing individual farmers’ locations. 

"Global forest maps have advanced, but without open, standardized reference data, progress in disambiguating forest land use from other land use like cacao and coffee agroforestry remains limited” said Rémi d'Annunzio, Forestry Officer at FAO and product manager of Whisp. “Today, initiatives like the Forest Data Partnership and DIASCA are putting efforts such as Sample Earth high on the global agenda as we work to define and standardize guidelines for open reference data collection.” 

Sample Earth builds on nearly two decades of Alliance research using satellite imagery to monitor land-cover changes across the Global South. The team plans to expand the dataset within Vietnam and Ghana and add other countries with high rates of misclassified smallholder farms, including Colombia and Honduras, along with coffee- and cacao-producing nations across Africa and Asia. 

Seeking modern cartographers 

Sample Earth’s roster of collaborators includes the United Nations’ Food and Agriculture Organization, Germany’s international development agency (GIZ), Google, Satelligence and WCF. The Alliance is actively seeking more collaborators and investors. 

“For EUDR to succeed, we need to lower the burden of monitoring and reporting, and we need to ensure that longstanding smallholder farms can be reliably reported as non-deforested areas,” said Dan Morris, a researcher at Google AI for Nature and Society. “AI combined with satellite imagery is a powerful tool that can help address these challenges, but AI systems are only as good as their training and validation data.” 

Inaction could disrupt supply chains and consumer markets, and not just in the EU; other jurisdictions are following suit in building similar legislation that will apply to most agricultural commodities. Supply constraints are feasible if maps do not quickly improve, which could push up prices. It’s bad news across supply chains, from vulnerable smallholders who already face myriad challenges to food-inflation-weary consumers worldwide. 

Sample Earth’s proposition is straightforward: better, inclusive training datasets will yield more accurate maps, protect compliant farmers from unwarranted exclusion, and give buyers and governments transparent tools to verify deforestation-free claims. By filling the data gaps that leave smallholder landscapes underrepresented, Sample Earth aims to make compliance affordable and fair, while supporting conservation and sustainable livelihoods in the tropics.