Satellite Data and EUDR Deforestation Evidence

Satellite observations support an assessment, not automatic proof

Satellite data can help establish historical land cover, detect changes and investigate whether production land underwent post-cut-off forest conversion. A satellite alert is not itself conclusive proof of EUDR deforestation. The legal assessment must establish qualifying earlier forest, later agricultural use and the relevant timing.

Likewise, a clean map result is not a legal guarantee. Resolution, observation gaps, classification methods and boundary quality can affect what the dataset detects. Use suitable data as evidence within a reasoned assessment of the actual source land and goods.

The deforestation-free requirement provides the legal question. The July 2026 Commission guidance says the EU Observatory is non-exclusive, non-mandatory and without legal value; the Regulation remains the binding source.

Choose data that match the question

Different products show imagery, tree-cover estimates, forest classes, land-use classes or disturbance alerts. Check what the layer measures and the period it represents before interpreting a colour as compliance. A tree-cover-loss layer does not necessarily distinguish a timber harvest from conversion to pasture.

Record the dataset, provider, version, acquisition or composite dates, resolution and relevant limitations. A high-level country map can inform risk context without being suitable to classify a small production plot. A recent image may establish current conditions but cannot alone show the land's status around the cut-off.

The question should guide the selection. Historical baseline, conversion timing, current agricultural use and wood-specific forest transitions may need different evidence. Do not assume one platform output answers all of them simply because it generates a single risk score.

Validate the production boundary first

Satellite analysis depends on the source location. Check coordinate order, projection, polygon shape and alignment with identifiable landscape features. A shifted boundary can create an apparent alert on neighbouring land or miss a conversion inside the real production area.

Confirm what the polygon represents: an entire property, a production area or a particular subarea. The guidance distinguishes forest and agricultural sections within a mixed plot. A broad property boundary may need additional information to explain which section produced the commodity and which section changed.

Keep the link to the actual goods. Accurate imagery for an approved supplier plot is not enough where a batch can contain material from other unrecorded plots. Purchasing and production records should support the source list analysed for the relevant products.

Investigate alerts with a dated sequence

When a layer flags change, compare suitable earlier and later observations. Establish what the earlier land was, what happened and what use followed. Look for corroborating production, management or land-use records where imagery alone leaves uncertainty.

Cloud cover, seasonal variation, shadows, fires and harvesting can complicate interpretation. A disturbance can be real without proving agricultural conversion. Conversely, replacement by an agricultural tree crop can retain considerable canopy and require land-use evidence beyond a simple loss threshold.

Record why an alert was accepted, rejected as irrelevant or left unresolved. Do not remove an inconvenient alert from the file without explanation. A documented review is more useful than a final map with no trace of the signals that prompted investigation.

Use complementary evidence and review conflicts

Land registers, dated farm records, forest inventories, management plans and field information can clarify imagery. Evaluate their authorship, dates, spatial coverage and consistency. An official agricultural label may describe current use rather than establish that no forest existed before conversion.

Where sources conflict, identify the reason. They may use different definitions, boundaries or observation periods, or one may be inaccurate. The tree-cover and forest distinction is especially important when a map's categories differ from the Regulation's agricultural-use exclusion.

For wood, satellite data may help identify harvesting and regrowth but may not establish every characteristic of primary, naturally regenerating or planted forest. Evidence about the defined forest-type transition should match the wood degradation question rather than reuse an agricultural-conversion score.

For example, an alert near a coffee plot might initially appear to overlap the source boundary. Checking coordinate order and higher-resolution context may show that the polygon was shifted onto a neighbouring forest. Preserve the original error, corrected coordinates and evidence explaining the correction, then rerun the assessment on the real source area. The correction resolves a location problem; it should not be presented as proof that all other land-history questions are answered. Conversely, a corrected boundary can reveal previously missed conversion and must be reviewed with equal care.

Preserve a reproducible evidence record

Save the source boundaries, relevant observations, dates, layer versions, annotations and reviewer reasoning. A screenshot without scale, coordinates or image date can be difficult to interpret later. Where access depends on a platform, retain enough information to explain the assessment if the service changes.

The conclusion should distinguish observation, legal classification and risk decision. If evidence remains insufficient, record the further work or mitigation needed before an ordinary operator can reach no or negligible risk. A low automated score cannot replace the operator's responsibility.

Review the file when boundaries, sources, imagery or credible new information changes. The aim is a supportable assessment of the goods' production land, not a permanent green badge based on one dataset run at supplier onboarding.