Why shade-grown coffee and cocoa get false-flagged, and how a type-aware read tells them apart.

Last updated: 22 August 2026

The trap: management looks like loss

Think of Ethiopian garden and forest coffee grown under native shade, cocoa farmed beneath a canopy across Côte d’Ivoire and Ghana, or the shade coffee of Colombia, Peru and Central America. These are agroforestry systems: the crop and the trees share the same ground, on purpose. Farmers manage that canopy constantly: pruning, coppicing, thinning, replacing ageing shade trees, taking the occasional timber tree. To the farm, this is routine husbandry. To a satellite that measures tree cover year on year, each of those events shows up as tree-cover loss.

So the raw signal a naïve check keys on (“tree cover went down here after 2020”) fires on a legitimately managed shade farm just as readily as on a patch of forest being bulldozed for pasture. Left there, the check condemns exactly the farms the regulation was designed to leave standing.

Why one map isn’t enough

No single dataset settles the question, and none is legally binding on its own. Each of the major sources answers a different question, and used alone each one mis-reads agroforestry in its own way:

• A baseline forest map tells you what counted as forest in 2020, but in dense shade systems it can over-read a canopy-covered farm as forest, so the very starting point can be wrong for these crops.
• A tree-cover-loss map tells you where canopy was disturbed after 2020, but it measures loss, not land-use change. It cannot, by itself, tell a pruned shade tree from a cleared forest.

Overlay just those two and flag every post-2020 loss pixel that sits on 2020 forest, and you get a check that is confidently, systematically wrong about shade-grown coffee and cocoa. The fix is not a better single map. It is reading several authoritative sources against each other, in a way that is aware of what kind of land it is looking at: a type-aware, multi-source read.

What EUDR actually prohibits

The distinction the maps have to respect is written into the regulation. Under Regulation (EU) 2023/1115, Article 2, deforestation is the conversion of forest to agricultural use: forest becoming farmland. That is the thing that must not have happened after 31 December 2020 for coffee or cocoa to be placed on the EU market.

Deforestation is forest turned into farmland, not a tree-crop farm being worked. That narrower question is the only one that matters.

Two consequences follow, and both are easy to miss. First, a shade-coffee or shade-cocoa plot is, in legal terms, already agricultural land. So managing its canopy is not converting forest to agriculture. Second, a reduction in tree cover is not automatically deforestation: forest that stays forest, or a tree-crop system being worked, is a different event from forest turned into a field. An assessment that flags on tree-cover loss alone is answering the wrong question. The right question is narrower: was forest converted to agriculture here after 2020?

The three layers, and what each one tells you

A defensible read for coffee and cocoa draws on three authoritative, openly documented datasets, each contributing the one thing it is good at:

• GFC2020: the Copernicus/JRC Global Forest Cover map for 2020. It establishes the baseline: what was forest at the cut-off line.
• Hansen Global Forest Change: the University of Maryland annual tree-cover-loss record. It shows where and when canopy was disturbed after 2020. Crucially, this is generic loss: harvest, fire, pruning and clearing all land in the same layer.
• JRC Tropical Moist Forest (TMF): the Joint Research Centre’s long-run classification of tropical moist forest, which distinguishes deforestation (conversion) from degradation (a disturbance the forest or tree system is not converted by). It is the layer that can speak to what kind of change occurred, not just that change occurred.

None of these is a compliance verdict, and none is mandated as the single source of truth. Their value is in combination: one fixes the baseline, one dates the disturbance, and one speaks to whether that disturbance was a conversion at all.

How they combine into a determination

Read together, the layers let an assessment do what no single overlay can. A post-2020 disturbance on 2020 forest is not treated as guilt on sight; it is checked against the classification layer:

• Where the classification confirms conversion of forest to agriculture after 2020, the plot is flagged: a genuine EUDR concern, surfaced clearly.
• Where the signal is consistent with shade-tree management or degradation without conversion (the agroforestry case), it is held for human review, not silently failed and not silently passed. A person looks at the evidence and decides.

That is the whole point of the type-aware read: a shade-grown plot isn’t condemned by a single loss pixel. The ambiguous middle, which is where agroforestry lives, is put in front of a human with the evidence assembled, rather than resolved by a threshold. The mechanism is described end to end on our method page, and the same reasoning runs through the full importer guide.

Why this matters to you

For a shade-grown importer, both errors cost you: a false flag can stall a clean shipment; a missed conversion is the one that comes back to you.

If you import shade-grown coffee or cocoa, a false flag can stall a shipment or sink you into disputing a determination about a farm that never did anything wrong. A false clear (a real conversion missed because a single map couldn’t tell) is a liability that stays with you, the operator.

What a careful, type-aware read buys you is not a guarantee; it is a determination you can stand behind: one that treated your agroforestry supply for what it is, weighed the right sources, and wrote down the reasoning. If an authority or a customer asks how you concluded a plot was deforestation-free, that reasoning is already on paper.