Biodiverse coffee at risk from zero-deforestation regulation
In this opinion piece, Dr. Christian Bunn invites you to a coffee break where AI-technology, smallholder farmers, forests, and EU policy meet.
Everybody knows coffee isn’t black and latte, but comes in shades of chocolate. Did you know that the same holds true for forests? They aren’t necessarily green or gone — there can be a lot of shaded coffee in between. In this post, we will discuss the disruptive power of AI vision for smallholder livelihoods, and why it is such a challenge to support high carbon biodiverse shade coffee while stopping deforestation.
The challenging path towards zero-deforestation coffee supply chains
Coffee is often promoted based on its sustainability attributes: Commercials show intact landscapes, striving smallholder farmers, and use certification generously to differentiate brands. Shade grown coffee — especially from systems in which coffee plants grow in the understory of diverse, native trees — has been shown to benefit everyone.
“Shade grown coffee systems sustain higher biodiversity and carbon stocks, deliver better-quality coffee, and greater resilience to climate hazards. But, the issue is that in a commodity market impoverished smallholders often don’t benefit economically from these opportunities.”
Coffee farmers often face challenges such as food insecurity, lack of access to healthcare, and educational opportunities (Camayo et al., 2022). The reality of coffee production often deviates from advertisers’ romanticized claims: Mechanized mega estates in Brazil, over-exploited ground water in Vietnam or deforestation in Honduras are commonplace but rarely seen on TV screen. According to recent research, coffee trade from Honduras is the second largest source of imported attributable deforestation in Germany after palm oil from Indonesia (Pendrill, Persson, and Kastner 2020).
In response, the groundbreaking EU regulation on deforestation free supply chains (EUDR) (European Union 2023) stipulates that any product which comes from a plot of land with recent deforestation will be prohibited from entry into the European market. The EUDR entered into force on June 29, 2023, and 18 months onward from this date all firms selling coffee, soy, beef, palm oil, wood, cocoa, and rubber in the EU must have credible solutions in place to trace these commodities, geolocate their production plots, and conduct rigorous deforestation analysis.
“The coffee sector is in turmoil over the EUDR. Everybody from origin to retail is concerned about the impacts. But these concerns are not related to coffee-attributable deforestation at all.”
Coffee is a perennial, and the productive life cycle starts at least three years after planting. This makes it virtually impossible for coffee from plots of land which were deforested after the cutoff date of December 31, 2020, to enter the EU by the time EUDR compliance is required. It is not impossible, however, for importers to have coffee containers rejected at the port of entry.
Understanding the EUDR compliance and monitoring conundrum
If importers struggle to demonstrate compliance — how would this be a problem for sustainable coffee production and smallholders? Because their coffee incurs the highest due diligence costs and most likely gets wrongly classified as incompliant.
“The EUDR and its proponents suggest an all-too-simple solution to monitor zero-deforestation compliance: All coffee to be placed on the European market must have the plot of land where it was produced documented.”
Remote sensing systems can then be used to check whether deforestation has occurred on that plot of land. In the words of an EU delegate: ‘Just use your phone to check your coordinates and write them down.’
Unfortunately, this conception of the functioning of the coffee sector, remote sensing, and the ecology of coffee production systems couldn’t be further detached from reality:
“The coffee value chain is highly decentralized at its origin. The majority of producers are smallholders, who typically interact with multiple intermediaries. Not all farmers own a smartphone, and satellite vision can’t distinguish their shade-coffee plots from natural forest.”
It goes without saying that it is more costly to procure coffee and collect geo-coordinates from a large number of sometimes illiterate producers which are difficult to integrate into information systems, than working with large estates under university-trained management. The regulation itself calls for efforts to ensure the inclusion of indigenous and smallholder producers but remains unspecific.
How AI algorithms see coffee below the canopy
The interaction between satellite observation and coffee ecology requires a deeper dive: Are you familiar with satellite data in map platforms — such as Google Earth? On high resolution satellite data, even relatively small objects — like the lane markings on streets are recognizable.
AI algorithms are more powerful than human vision when we augment the satellite imagery by using time series, pixel noise, or individual color bands including infrared. Trained on a few hundred ground-validated sample locations, the AI can then recognize similar locations at large spatial scales. For coffee, the challenges for the AI are in the production system and the climatic conditions at coffee locations.
Coffee is produced in a broad range of production systems. Often, the ‘good stuff’ will be advertised as ‘forest coffee’, ‘shade grown’ or ‘strictly high grown’. The ‘not-so-good stuff’ has different properties: it’s likely been grown at low altitude — without shade trees. The first type of coffee is often cultivated in remnant forest, for instance, because the coffee trees naturally reproduced or were planted between old growth trees. People managing such stands can live within the forest and have an income without destroying it. But the vast majority of coffee is produced in high-input systems under full-sun (Jha et al 2014).
Brazil alone supplies over 30% of the world’s coffee often on large mechanized operations. On satellite imagery, such estates under full sun can easily be distinguished from forests, while smallholder plots are difficult to identify.
Our remote sensing team is working hard to overcome these challenges: Time series data, infrared and additional covariates can be used, and the visible canopy of a managed plantation — even under shade — has a different spectral fingerprint than a forest. Introduced species among the shade trees, and intense white flowering of the coffee plants, may provide valuable clues to identify shade coffee plots.
“Yet, the closer the coffee system resembles natural vegetation — for example because diverse native trees are used to provide shade, without fertilizer or irrigation inputs — the closer the coffee canopy fingerprint will be to forest.”
To add further complexity, primary forest often remains at high altitudes with frequent cloud cover — so frequent, that meaningful satellite data of the earth’s surface can’t be collected. Yet, premium coffee needs such conditions to develop its nuanced flavor. Thus, cutting-edge AI algorithms perform especially poor when faced with the best that coffee growers can offer: High quality coffee from low emission, biodiverse production systems.
Unclear impact off-farm, collateral damage on farm
Critiques of the EUDR expect coffee importers to minimize their risks by preferentially sourcing from regions with low forest cover, close value chains and large producers. To illustrate the conundrum, consider two coffee production extremes: On the one hand, indigenous producers in the Central American highlands organize informal trade routes to central collection stations because of their remote location. Their coffee grows in biodiverse systems with old native shade trees — it’s the good stuff. Yet, importers will be hard pressed to show where the coffee comes from, and our smartphone equipped EU bureaucrat detect native forest, even when coordinates for coffee stands can be provided. This coffee faces great risks of being rejected when reaching the EU.
On the other hand, coffee from a Brazilian estate — grown under full sun, lots of in fertilizers, irrigated and mechanically harvested — will easily comply, putting the EUDR broader intention to foster sustainable production to the test. While it remains unclear whether the EUDR will actuall stop deforestation off-farm, its binary view on forests is bound to translate into disincentives for sustainable practices on-farm.
“A downward spiral could be triggered in a worst-case scenario: Smallholders which were excluded from the value chain abandon their coffee and start producing for local value chains, e.g. by shifting to livestock.”
Contrary to the stated objectives, the EUDR implementation may cause reduced carbon stocks and biodiversity. To stop deforestation, reduce emissions, protect biodiversity and improve livelihoods, we need a smallholder-inclusive implementation of the EUDR.
Coming up
In Dr. Christian Bunn’s next blogpost, we’ll go back to AI vision for sustainable coffee farming, but this time as a close up. “Proximal AI vision — a technofix to include smallholders in coffee value chains, or a real solution?”
Further Reading
European Union. 2023. Regulation (EU) 2023/1115 of the European Parliament and of the Council of 31 May 2023 on the Making Available on the Union Market and the Export from the Union of Certain Commodities and Products Associated with Deforestation and Forest Degradation and Repealing Regulation (EU) No 995/2010. https://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX:32023R1115.
Hidalgo, Francisco, Xiomara F. Quiñones-Ruiz, Athena Birkenberg, Thomas Daum, Christine Bosch, Patrick Hirsch, and Regina Birner. 2023. “Digitalization, Sustainability, and Coffee. Opportunities and Challenges for Agricultural Development.” Agricultural Systems 208 (May): 103660. https://doi.org/10.1016/j.agsy.2023.103660.
Jha, S., C. M. Bacon, S. M. Philpott, V. Ernesto Mendez, P. Laderach, and R. A. Rice. 2014. “Shade Coffee: Update on a Disappearing Refuge for Biodiversity.” BioScience 64 (5): 416–28. https://doi.org/10.1093/biosci/biu038.
Pendrill, Florence, U. Martin Persson, and Thomas Kastner. 2020. “Deforestation Risk Embodied in Production and Consumption of Agricultural and Forestry Commodities 2005–2017.” Zenodo. https://doi.org/10.5281/ZENODO.4250532.
