Avoiding Access Inequity Due to classification errors in zero-deforestation value chains: Coffee and the European union deforestation regulation

European Union's Regulation 2023/115, commonly known as the European Union Deforestation Regulation (EUDR), promises to be a watershed event in global deforestation governance. A significant example of the hardening of soft law, spurred by major corporations committing to zero-deforestation supply chains, the EUDR is also a substantial wager on the efficacy of satellite-based remote sensing technologies for effective global forest governance. As remote sensing becomes more deeply embedded into global environmental governance, it is necessary to pay attention to the possibility that misclassification errors - mistaking one type of land cover for another - could become institutional errors with real consequences for those targeted by these initiatives. If compliant producers were to be excluded from zero-deforestation markets due to uncertainties resulting from misclassification errors, this would raise questions about the initiative's access equity. To develop recommendations for a strategy for avoiding this eventuality, we examine how classification errors could shape the EUDR's effects in the coffee sector. Coffee, a commodity predominantly cultivated for export by smallholders under tree shade, faces heightened susceptibility to the legislation, given the European market's significant influence on global consumption. Using ground-truth points collected in coffee-growing regions in Ethiopia and Tanzania, combined with other open datasets, we assess the rate at which five global land cover datasets identify coffee production as forest, finding high rates of misclassification in some geographies, particularly for shade-grown and agroforestry cultivation. Then, following a systematic review of remote sensing studies designed to detect the presence of coffee, we use quantile regression analysis to identify strategies that could be used to reduce classification accuracy for coffee to unproblematic rates. Based on these assessments, we argue that, even in a hard case like coffee, access inequities due to misclassification errors could be mitigated substantially by starting with a global dataset and then building regional, commodity-specific datasets. We suggest that finding ways to compensate and include smallholders, cooperatives, and other producer groups in a project of building monitoring datasets as a public good may be an appropriate strategy for the EUDR and similar zero-deforestation initiatives.