Soya under the EUDR

Soya is the commodity where the gap between what the EUDR does and what people assume it does is widest. It has four entries in Annex I, no ambiguous chemical derivatives, no species field, and a geometry requirement that is uniform rather than conditional. Read as a compliance problem it is one of the more tractable of the seven.

Read as an environmental instrument it is the one that will disappoint most, and it is worth being straightforward about why: the biome most associated with soya expansion in the public debate largely does not meet the regulation's definition of forest. That is not a loophole somebody exploited. It is what the text says, and a due diligence programme built on the assumption that it says something else will produce results that satisfy nobody.

What is in scope

The narrowing to 1201 90 00 removes soya beans for sowing. It is a small, sensible change — seed is a specialised trade in small volumes, and requiring plot geolocation for a bag of certified seed was disproportionate. It affects seed companies and essentially nobody else.

The entry that carries the volume is 2304, soya bean meal. Europe imports soya overwhelmingly as animal feed, and the businesses most exposed to this regulation through soya are compound feed manufacturers, integrators and livestock producers rather than anyone in the food trade. If your organisation buys soya meal to feed animals, you are handling a relevant product and the obligation is yours as directly as it is a chocolate manufacturer's.

The Cerrado is mostly not forest

The EUDR defines forest by reference to the FAO threshold that the regulation retains: an area over half a hectare with trees taller than five metres at maturity and canopy cover of at least ten per cent. Deforestation is the conversion of forest, so defined, to agricultural use after 31 December 2020.

The Brazilian Cerrado is a mosaic of grassland, shrubland and open woodland. Parts of it — gallery forest, the denser cerradão — clear the threshold. A great deal of it does not, and land with five to ten per cent canopy or predominantly shrub cover falls into the category of other wooded land, which the regulation does not cover. The Amazon, by contrast, clears the definition comfortably, and conversion there after the cut-off is squarely caught.

The consequence is a two-tier result within one country. Soya expanding onto Amazonian forest cleared after 2020 produces a non-compliant consignment. Soya expanding onto Cerrado savanna cleared last year may produce a perfectly compliant one. Both are conversion of native vegetation; only one is deforestation as this regulation defines it.

Argentina, Paraguay and Uruguay carry their own versions of this. The Gran Chaco is dry forest that in many places does clear the canopy threshold, so Argentine and Paraguayan expansion is more likely to be caught than Cerrado expansion is. Reading South American soya as one risk landscape is the mistake; the answer changes by biome, not by border.

Polygons, universally

Soya is the only commodity where the four-hectare question does not really arise. A commercial soya field is tens to thousands of hectares; the point option exists in law and will almost never be available in practice. That sounds like a burden and is close to the opposite, because soya farms are large, mechanised, cadastrally registered businesses that frequently already hold field boundaries as GIS files for agronomic reasons.

In Brazil the Cadastro Ambiental Rural gives most rural properties a registered perimeter, and precision agriculture has meant field-level boundaries have existed on farm for years. The collection exercise is therefore a data-transfer and reconciliation problem rather than a survey. What it does generate is scale of a different kind: polygon files for large properties are geometrically complex, and a consignment drawn from many farms produces a submission payload measured in megabytes.

The reconciliation is where the real work sits. A property boundary is not a plot of land in the regulation's sense. The regulation's unit is land homogeneous enough that deforestation risk across it can be assessed as one thing, and a four-thousand-hectare property containing soya, pasture, a legal reserve and a block converted in 2022 is emphatically not that. Filing the property perimeter is the single most common soya geometry error, and it fails in the worst way — by asserting a clean result over land that includes the part you should have excluded.

Where the chain mixes

Grain is fungible, which is the whole design of the trade. Soya is delivered to an elevator, pooled with other deliveries of the same grade, moved to a crusher or a port terminal, and loaded as a vessel lot. Field identity survives to the elevator scale ticket and no further.

Compounding this, a large share of volume — commonly put at more than a third — reaches exporters through intermediaries rather than through direct farm relationships. A trader with excellent records for its direct suppliers can still have a substantial fraction of its book arriving from resellers who bought from farms the trader has never dealt with. This indirect share is where soya due diligence programmes tend to stall, and it is not a data-capture problem so much as a commercial one: nobody in the middle has an incentive to disclose their sources to a buyer who might then go around them.

The practical response mirrors palm's. Segregated flows with a known farm list are achievable at a cost, and mass-balance flows are not compliant merely because they are audited. Where a buyer needs a defensible statement, the volume has to be handled as an identity-preserved or segregated stream from the elevator forward, and that decision is made in a procurement contract long before it is made in a compliance system.

Risk assessment

Brazil, Argentina and Paraguay are all classified standard risk, so full risk assessment and mitigation apply across effectively the whole South American supply base. The United States and Canada are low risk, which matters more than it used to: as European buyers look for supply with a cleaner compliance profile, North American soya becomes commercially more attractive for reasons that have nothing to do with agronomy.

That substitution effect is worth naming honestly, because it is the most likely real-world outcome of the regulation in this commodity. Shifting European purchasing towards low-risk origins satisfies the regulation cleanly. Whether it reduces conversion in the Cerrado, or simply redirects that soya to buyers who are not asking, is a question the EUDR does not attempt to answer and a due diligence programme cannot resolve.

Filing the statement

The importer or crusher first placing soya on the EU market files. Feed manufacturers buying imported meal from a European trader are downstream and retain the reference number; feed manufacturers importing meal directly file themselves, which catches a fair number of businesses that assumed their trader was handling it.

Payload size is the recurring practical constraint. Large, vertex-heavy polygons for thousands of fields will test submission limits, and the grouping mechanism introduced by the implementing regulation in July 2026 exists precisely for this. Simplifying geometry to fit is tempting and should be resisted: a generalised boundary that shaves a field edge is a different claim about the land than the one your analysis was run on.

How ERWAY handles soya

• GeoJSON upload is the primary path for soya, because the geometry usually already exists. Files are accepted as FeatureCollection, Feature, Polygon or MultiPolygon in WGS-84, and validated before anything is scored.
• Thirteen named checks run on every upload. Two matter especially here: plot-to-plot overlap detection, which catches the same field submitted under two farm records, and an area cross-check that flags geometry whose computed area does not match what was declared — the usual signature of a property perimeter filed where a field boundary was needed.
• Each plot is scored against the JRC Global Forest Cover 2020 layer, so the forest question is asked against a baseline built to the regulation's own forest definition rather than against a generic tree-cover product. That distinction is what keeps Cerrado savanna from being scored as though it were forest.
• Hansen annual loss, GFW Integrated Alerts and OPERA DIST provide change detection since the cut-off, with alerts intersected against the 2020 forest mask before they are counted.
• Sentinel-2 imagery per year from 2020 lets a flagged block be inspected directly, which on large fields usually resolves whether a change is conversion, a legal reserve boundary or a cropping-pattern artefact.
• Where a property contains a block that has to come out, mitigation is recorded against that polygon with the revised boundary and supporting documents attached, and the excluded area is carried in the audit record rather than quietly dropped.

Common questions

Is soya biodiesel covered the same way as soya oil?

Not obviously. Heading 1507 covers soya-bean oil and its fractions, refined or not, but not chemically modified — and biodiesel is produced by transesterification, which is a chemical modification. Finished soya-based biodiesel sits outside the wording of the soya entry in a way that crude or refined but unmodified soya oil does not, which is a real difference from palm's entry, where several chemically modified derivatives were deliberately added in July 2026. If you handle both the oil and a biodiesel product made from it, check each against its own classification rather than assuming one answer covers both.

Is soya lecithin in scope?

No — lecithin is classified under its own heading outside the four codes Annex I lists for soya (beans, flour and meal, oil, and oilcake), so a lecithin ingredient on its own is not currently a relevant product under this commodity, even though it is extracted from the same beans.

What about soya protein isolate or concentrate used in plant-based meat?

Further-processed protein products are not among the four listed soya codes either, so as a finished ingredient they generally sit outside Annex I's soya entry. The obligation still exists further up that same chain — whoever imports the raw beans, meal or oil used to make the isolate is handling a relevant product at that point, even if the isolate itself, once made, is not separately covered.

If I buy soya meal from an EU-based crusher rather than importing it myself, do I have an obligation?

If the crusher imported the beans and filed the statement, you are a downstream operator retaining its reference number rather than filing your own — the same rule that applies to any relevant product bought after someone else placed it on the EU market. Your obligation is to keep that reference number on file, not to reconstruct the crusher's plot data yourself.

How is quantity handled for a vessel cargo blended from several countries' soya?

Annex II allows the quantity in a due diligence statement to be expressed with a percentage estimate or deviation where a precise figure is not practicable, which is written specifically for cases like a blended bulk cargo. It does not remove the underlying duty to geolocate the plots behind each origin represented in the blend — it only accommodates how precisely you can state how much came from where.

Are tofu, soy milk and other foods for direct human consumption covered?

Generally not under the current four soya entries, which stop at beans, flour and meal, oil, and oilcake — further-processed consumer foods are classified elsewhere. As with protein isolates, the obligation sits with whoever imported the raw soya used to make them, not with the finished consumer product.

Do I still need geolocation for soya from the United States or Canada, given their low-risk status?

Yes. Article 9's collection duty is not conditional on the origin's risk tier — it applies to every plot regardless of where it sits on the benchmarking list. What a low-risk origin removes is the risk assessment and mitigation steps in Articles 10 and 11, absent a substantiated concern. As European buyers shift purchasing toward North American origins partly for this reason, it is worth being clear internally that the shift reduces analysis, not data collection.

Does the regulation distinguish GM from non-GM soya?

No — the EUDR's test is about the land the soya was grown on, not about the variety or whether it is genetically modified. GM status is a separate regulatory question handled under different EU rules and has no bearing on whether a shipment needs plot geolocation.

What if soya from a large industrial farm is blended with soya from smaller neighbouring farms at the same silo?

Each contributing farm is still a separate plot requiring its own geolocation, even where the grain is physically blended immediately on arrival. Silo-level blending is a traceability challenge for reconstructing which farms fed which outbound shipment, not a reason fewer plots need to be geolocated in the first place.

Is soya grown for animal grazing rather than harvested beans covered?

The Annex I entries name soya beans, flour and meal, oil, and oilcake — products derived from harvesting and processing the bean. Standing forage grazed directly by livestock rather than harvested and traded as one of those products is not itself one of the listed goods, though beans from the same farm that are harvested and sold would be.

What to do before December

1. 1

   Check whether you import meal directly

   Feed businesses are the most under-scoped group in this commodity. Buying soya meal from a European trader makes you downstream; importing it yourself makes you the operator who files.

2. 2

   Reconcile property boundaries down to plots

   A CAR perimeter is not a plot of land. Split registered properties into homogeneous units before scoring, or you will file a clean result over land that includes the block you needed to exclude.

3. 3

   Decide the segregation question in procurement

   Mass-balance flows will not produce a plot list. If a volume needs a defensible statement, it has to be handled as a segregated stream from the elevator forward, and that is a contract decision made months ahead of a filing.

4. 4

   Separate your EUDR finding from your conversion-free claim

   The Cerrado question is real and this regulation does not ask it. Run the forest test for the EUDR and keep native-vegetation conversion in its own reporting line, so neither result gets used to answer the other.

Soya is administratively the most straightforward of the seven and politically the most awkward. The geometry exists, the scope is short, the origins are professionalised, and a competent programme can produce defensible statements at scale. What it cannot produce is the reassurance people expect from it, because the regulation asks about forest and the debate is about savanna. Being clear about that inside your own organisation is worth more than any amount of additional analysis.

Published August 17, 2026 · Last reviewed August 17, 2026 against the sources listed above.