The European Union Deforestation Regulation requires that commodities placed on the Union market be deforestation-free with reference to 31 December 2020, and that operators document how that was established for every plot of production. The obligation reaches large and medium operators from 30 December 2026 and micro and small enterprises from 30 June 2027. Penalties run to four per cent of Union wide annual turnover.
The difficulty is not the imagery. It is the reference maps.
A plantation that reads as forest
The Union's own global forest baseline classifies 31 per cent of rubber as forest, the highest overlap of any regulated commodity, against 7 per cent for oil palm and 6 per cent for cocoa. Its peer review reports an 18 per cent commission error and states plainly that co-location with mapped forest does not by itself indicate non-compliance. An assessment of 21 global forest maps found only two meeting all of the Regulation's indicators, and in one Mexican test 75 per cent of 600 smallholder coffee plots established decades before the cut-off were incorrectly flagged, because agroforestry looks like forest from above.
A mature plantation is tree cover in every global layer, and its normal life produces exactly the signals a tree-cover-loss alert exists to catch. Rubber is felled and replanted on a 25 to 35 year cycle, and it drops its leaves every dry season. Neither is deforestation under the Regulation, which excludes land predominantly under agricultural use from its definition of forest. Both are flagged, and a flagged plot today is resolved by sending someone to the field with a camera.
The Regulation itself never names satellite imagery. Its standard is adequately conclusive and verifiable information, and the Commission describes its own map as non-exclusive, non-mandatory and carrying no legal value. Satellite evidence is therefore not required, in practice unavoidable, and entirely unspecified, while the burden of proof sits with the operator. That gap is what plot level evidence fills.
What the paper covers
The paper sets out the method in full: how a plot's state at the cut-off is established from the record that existed at the cut-off, how change since is measured against the noise floor of its own scene pair, and how spectral fractions say what replaced a canopy rather than only that one was lost. Felled stems and slash on the ground read differently from bare soil, which is the difference between a working estate on its rotation and a clearance.
It works through a rubber estate on the Ivorian coast, chosen because a method that calls its replanting deforestation fails the case the market actually has. In the three years before the cut-off, change over the estate barely exceeds what noise alone produces at the same threshold, which settles the question of conversion. In the five years since, roughly a tenth of the estate has been opened, and the fractions show felled stems and slash far more than bare soil. On a 25 to 35 year cycle, a tenth in five years is the cycle.
It also documents a monitoring season in which a naive rule failed. Read consecutively, the readings fell steadily from December to March and flagged two months as change. Nothing had been felled. Rubber drops its leaves every dry season, and the comparison that survives is against the same season a year earlier, not against last month.
What it does not claim
Axine holds no rubber specific error rate and does not quote one. The fraction and index layers ship as screening, because the method is published and reproduced from a public implementation but has not yet been scored against field data on rubber. A trial on a client's own plots, with the planting years they hold, is what produces that number under a pre-registered protocol.
The precedent for how that is reported is on record. On the lane where this discipline is longest established, burn extent and severity were scored against a federal reference across thirty fires with the sample fixed before scoring. Extent passed its gate. The severity class failed, and is published as failed with its diagnosed cause.
Read the paper
Plantation Evidence Under EUDR is 26 pages and covers the regulatory background, the method, the worked example, the validation posture, the operational limits and the platform. It is available on its paper page.