Skip to content
Daviah IO
Daviah IO
PlatformsPricingDesign PartnersKnowledge BaseBlogAbout
Daviah IO
Daviah IO

  • Platforms
  • Pricing
  • Design Partners
  • Knowledge Base
  • Blog
  • About
All posts
eudr
map accuracy
jrc
agroforestry
smallholders
deforestation regulation

A Green Pixel Is a Policy

Daviah
June 15, 2026
A Green Pixel Is a Policy — Daviah IO
Daviah. Field Notes
Vol. 01 · No. 12
Investigative Essay

A Green Pixel Is a Policy

EUDR enters enforcement on December 30, 2026. The peer-reviewed evidence shows that the satellite maps it relies on misclassify agroforestry at rates above 60%. The regulator built enforcement on a foundation it would not accept from any operator.

The EU Deforestation Regulation enters enforcement on December 30, 2026. It is one of the most ambitious supply chain laws ever written. It is also built on a foundation that researchers and producing countries are quietly demonstrating cannot hold the weight the regulation places on it.

The regulation requires operators to provide the geolocation of every production plot tied to relevant commodities — coffee, cocoa, rubber, palm, soy, cattle, timber. It accepts the European Commission's Joint Research Centre map of global forest cover for the year 2020 as the reference layer against which those plots are judged. If the plot intersects with the JRC's 2020 forest, the assumption is deforestation. The dossier is rejected. The shipment does not enter the market.

The architecture is clean. The architecture's foundation is not.

What the maps actually say

On Sources The argument that follows draws heavily on van Noordwijk et al. (2025), "Beyond imperfect maps: Evidence for EUDR-compliant agroforestry," published in People and Nature. The authors are working scientists, including Meine van Noordwijk, whose career on tropical agroforestry spans decades at the World Agroforestry Centre. Their paper is the most detailed peer-reviewed assessment to date of whether the maps the EUDR relies on can do what the regulation asks them to do. The conclusion they reach is that the maps cannot, and that operator-side evidence built on ground projects is required to supplement them. What follows applies their argument at the regulatory architecture layer.

An accuracy assessment of version 2 of the EU-JRC global forest cover map for 2020 — the reference layer EUDR uses — found an 18% commission error for forest. That is, 18% of land the map labels as forest is not actually forest by the definitions the map claims to apply. The omission error was 8%. Map accuracy decreases for dry and open forests, and for landscapes with small forest/non-forest mosaics.

The EU-JRC version 2 maps 4.56 billion hectares as forest globally. The FAO Global Forest Resources Assessment, compiled from national reporting using definitions that include institutional criteria beyond tree cover, maps 4.08 billion hectares. The two disagree by 12%. The disagreement varies sharply by country:

1.6%
Brazil JRC / FAO map disagreement on forest area. Low — the easy case.
26.9%
Indonesia JRC / FAO disagreement — the country where most agroforestry-based EUDR exposure lives.
28.3%
Democratic Republic of Congo The largest disagreement among major forested nations.

A cross-tabulation of Indonesia's REALU land cover map — built from extensive ground-truthing — against the EU-JRC 2020 forest map found that 18.7 million hectares of Indonesian agroforestry and tree-crop monoculture were classified as forest by the JRC. The misclassification rates are not edge cases:

65%
Rubber Agroforest (Indonesia) Probability of being misclassified as forest by the EU-JRC layer.
59%
Other Agroforestry Systems Mixed cultivation under tree canopy — traditional smallholder practice across Southeast Asia.
50%
Rubber Monoculture Single-species plantation — not forest, classified as forest half the time.
63%
All Agroforests Combined The aggregate misclassification rate. Sixty-three percent.

In Mexico, researchers studied 600 coffee plots, each 1\u20135 hectares, many cultivated for decades. They could not have been deforested since 2020. The maps suggested three-quarters of those plots were non-compliant.

There is more. Reymondin et al. compared 19 map sources for Ghana with a focus on cocoa, and found precision scores of 83% and 79% for the two best-performing maps — with many lower scores across the others. All claimed to apply the FAO forest definition.

The maps are not error-free. They are not close to error-free. They are operating at a level of accuracy that, under the regulation's own standard of "adequately conclusive and verifiable evidence," would not survive scrutiny.

The asymmetry of error

Van Noordwijk and colleagues make the statistical structure of EUDR enforcement explicit in a way the regulation itself does not. Two kinds of error are possible when a binary classification gets applied to a population of plots. A Type I error wrongly accepts a non-compliant shipment as compliant. A Type II error wrongly rejects a compliant shipment as non-compliant. The two errors are inversely related. Reducing one increases the other.

EU policy claims zero tolerance of aggregated Type I error: any location in a shipment that has recently been deforested triggers rejection. To enforce zero tolerance on Type I error, the regulation accepts very large Type II error. The math is unforgiving.

A shipment of ten independent locations, each with 85% chance of being correctly represented on the map, has an 80.3% chance of being erroneously rejected. A single due diligence statement can contain a thousand locations.

In other words: the regulation is constructed so that even when the underlying maps are 85% accurate — a level the JRC map has not reached, by independent assessment — the cumulative effect of mapping errors across a shipment makes erroneous rejection almost certain.

The Type II error is the producer's burden to bear. The smallholder coffee farmer in Mexico, the cocoa cooperative in Ghana, the rubber agroforest community in Indonesia — they pay the cost of a system that has prioritized fraud prevention over the risk of collateral damage. They cannot see the map. They cannot dispute the map. They cannot appeal the verdict the map renders against their land. They simply lose access to the European market.

Van Noordwijk et al. put the consequences plainly: recent scandals in several European countries followed policies where fraud prevention was prioritized and risks of collateral damage downplayed. EUDR is being built on the same asymmetry. Type I errors are unacceptable; Type II errors are the producer's burden to bear.

Pixels are policy

A pixel on a satellite-derived forest map is not a description of the land. It is a policy decision that gets enforced against the people who work the land. And when the pixels are wrong, the policy harms in both directions.

A green pixel can mask deforestation. The imagery may not have caught the clearing yet. The canopy may still appear continuous from above. Exploitation may be happening under cover that has not yet been removed. A green pixel does not certify that the land is intact. It only certifies what the imagery has registered, at the resolution and timing of its capture.

A red pixel can punish producers whose practice has been entirely consistent with the regulation's intent. The trees grew shorter than the classifier expected. The shade canopy in an agroforest looked like cleared ground to a binary model. The plot was misregistered against a coordinate system that drifts at the scale of small plots. A red pixel does not prove deforestation. It only signals what the imagery has flagged.

Vietnamese officials have already raised the specific case. Coffee farms that integrate shade trees into their cultivation, a practice that has been part of those landscapes for generations, can be misclassified as forest by satellite imagery. The producer's compliant plot ends up flagged for non-compliance under a regulation it has never violated.

This is the structural problem. The regulation has taken a tool — satellite forest cover classification — and elevated it to the level of legal evidence. The tool was not designed for that role. It is useful as a screening layer, as a way of generating leads for ground investigation. It was never meant to render judgment on producer livelihoods at scale, with no mechanism for dispute.

— The Counter-Example

What accountability for a forest actually looks like

Vietnam is one country that has noticed.

The Vietnamese Ministry of Agriculture and Environment, in cooperation with the Forestry EUDR Network, is building its own 2020 forest boundary map. The work combines provincial forest-status data from 2020, multi-temporal satellite imagery, ground verification, and cross-referencing against international layers including the JRC and UMD/GLAD. Nineteen of thirty-four provinces and cities have submitted complete datasets. The remaining localities are continuing verification.

The methodology categorizes plots into three groups. Green areas are locations where all data sources agree. Yellow areas indicate discrepancies that require further verification. Red areas identify high-risk zones directly linked to EUDR-sensitive commodities such as coffee, rubber, and timber. Local authorities have conducted field verification on every single red-zone location.

The result is striking. Across the thirteen provinces that have completed review:

54.2%
Green
All sources agree
45.0%
Yellow
Required ground verification
0.8%
Red
Genuine high-risk zones

Forty-five percent of land required further verification beyond the international datasets. Eight tenths of one percent represented genuine high-risk areas — verified on the ground.

Pham Ngoc Hai, from Vietnam's Forest Inventory and Planning Institute, put the point with the clarity it deserves: international datasets such as Global Forest Change and JRC products should be treated as reference layers rather than legally decisive evidence.

That is the structural correction. The maps are reference. The ground is truth. The country that knows its land builds its own dataset and holds itself accountable for the verification, rather than accepting the verdict of a satellite classifier built ten thousand kilometers away by people who do not work that ground.

Vietnam is not doing this for compliance theater. It is doing this so its coffee, rubber, and timber producers do not get erroneously rejected by an imagery layer that does not understand the texture of Vietnamese agroforestry. It is doing this because the country has decided that protecting its producers is part of what it means to be responsible for its forest.

Vietnam's approach is, structurally, what van Noordwijk et al. argue is required. The maps are insufficient on their own. Ground projects, producer-side evidence, multi-source verification, and local data treated as the determining factor rather than as a supplement — these are the conditions under which an honest verification architecture can be built. Vietnam is one country building those conditions in advance of the deadline. The question is whether the regulation itself will be flexible enough to credit that work, or whether it will continue to treat the satellite layer as legally decisive against producers who can demonstrate, with their own data, that the layer is wrong.

What the regulation did not do

The regulation requires operators to provide adequately conclusive and verifiable evidence for every claim. It expects operators to assess the credibility of their certification schemes, to document the methods their service providers use, to verify the data they import from national systems. It expects operators to do the epistemic work of confirming that the evidence supporting their claims is robust enough to defend.

The regulation has not held itself to that same standard.

The foundational map against which every producer's plot is judged has not been validated to the level the regulation demands of operator evidence. There is no public methodology for resolving false positives on the JRC layer. There is no documented dispute mechanism. There is no acknowledgment in the regulatory text that the map has known structural error in exactly the agroforestry contexts where smallholders predominantly grow EUDR commodities.

Van Noordwijk and colleagues do not call this irresponsibility. They are scientists, and they write in the careful tone scientific work requires. But the structural fact their paper documents is plain. The regulator built enforcement architecture on imagery that has not been ground-validated to the level the regulator requires of every operator's evidence. The asymmetry is the problem. It is irresponsible at the level of design.

The fix is the kind of work Vietnam is doing.

The part worth saying plainly

We are months away.

The maps are not ready. The people being judged by the maps cannot dispute them. The regulation built its enforcement architecture on imagery that has not been validated to the standard the regulation itself demands of operators.

EUDR asks operators to demonstrate adequately conclusive and verifiable evidence for every claim. If preventing deforestation is the goal, the regulation has to respond to the evidence researchers and producing countries are providing — and credit the ground projects that close the gap between what the imagery sees and what the land actually is.

A green pixel is not a description of the land. It is a policy. And when the policy is wrong, the people who pay are the ones who have been working the land for generations, under regulations they have never violated, judged by maps they have never seen.

That is the part worth saying plainly, before December 30.

· · ·
References & Further Reading
Sources cited in this essay
  1. Bourgoin, C., et al. (2024). Global mapping of forest cover for the year 2020. EU Joint Research Centre.
  2. Colditz, R., et al. (2025). Accuracy assessment of the EU-JRC global forest cover map for 2020.
  3. Mongabay (2025). Researchers find concerning gaps in global maps used for EUDR compliance. news.mongabay.com
  4. Reymondin, L., et al. (2025). Comparative analysis of forest cover maps for Ghana's cocoa-growing regions.
  5. van Noordwijk, M., et al. (2025). Beyond imperfect maps: Evidence for EUDR-compliant agroforestry. People and Nature. besjournals.onlinelibrary.wiley.com
  6. Vietnam+ (2026). Vietnam speeds up 2020 forest map to support compliance with EUDR. en.vietnamplus.vn
Daviah IO

Daviah

Regulator-ready due diligence for companies whose supply chains carry regulatory exposure.

admin@daviah.io

Platform

ProductPricingDesign PartnersKnowledge BaseBlogAboutSecurityContact
Admin

© 2026 Daviah IO. All rights reserved.

Enterprise SaaS. Purpose-built.

Data resident in your elected region (US / EU) · SOX-grade audit logs · Per-tenant isolation · 30-day post-termination export window · Privacy policy · DPA