Compliance Methodology — EUDR and ISCC.
How AGROGIS produces compliance reports for the EU Deforestation Regulation and ISCC sustainability certification. The same satellite fusion approach applies to both — what differs is the cutoff date and the indicators screened.
What EUDR actually asks for
The EU Deforestation Regulation (EU) 2023/1115 forbids placing on the EU market goods produced on land deforested or degraded after 31 December 2020. Seven commodities are in scope: cattle, cocoa, coffee, oil palm, rubber, soya and wood, plus their derivatives (leather, paper, chocolate, tyres, etc.).
Article 9(1)(d) requires every operator to supply geolocation evidence of each plot of land where the relevant commodity was produced. For plots above 4 ha that means a polygon; below 4 ha a point is allowed. AGROGIS always operates on polygons.
Article 8 places responsibility for verifying that evidence on the operator. A report we produce supports the Due Diligence Statement (DDS) you file in TRACES NT — it is evidence, not a submission.
Article 12 requires operators to retain due-diligence evidence for five years. The reports AGROGIS produces include the report UUID, methodology version, and Hansen GFC version on every page so the evidence is dateable and reproducible later.
Three public datasets, one polygon
AGROGIS combines three independent, public remote-sensing sources: Hansen Global Forest Change for the historical baseline, and Copernicus Sentinel-2 (optical) and Sentinel-1 (radar) for the years after it. All are global, 30 m or finer, and freely available. No commercial imagery is used in this analysis, so an auditor can reproduce the evidence from open inputs.
Hansen Global Forest Change
Version GFC-2025-v1.13, distributed by the University of Maryland / Google Earth Engine partner programme. 30 m resolution. Two bands matter for us:
- treecover2000 — percent canopy cover in 2000 (0-100).
- lossyear — 0 if no loss observed; 1-25 = the year loss occurred (1 = 2001, 25 = 2025).
Copernicus Sentinel-2 L2A
ESA Copernicus surface reflectance, served via Sentinel Hub by Sinergise. 10 m visible/NIR resolution, 5-day revisit at the equator. We compute:
- NDVI = (B08 - B04) / (B08 + B04)
- SCL band 12 used to mask clouds, cloud shadow, snow.
- Scene-level cloud cover capped at 60 percent.
Why both? Hansen GFC is the gold standard for long historical series (2001 onward) but its latest year always lags one calendar year (the v1.13 release covers up to 2025). Sentinel-2 has the spatial detail and the near-real-time cadence to fill the gap, but it cannot reproduce the Hansen multi-decade pixel-history. We use each where it is strongest.
How each pixel gets a verdict
For every Hansen pixel whose centre falls inside the parcel polygon, the analysis runs the following deterministic decision:
A. Was the pixel forest at the 2020 cutoff?
Yes if
treecover2000 >= 10percent (the FAO + EUDR canopy threshold) ANDlossyear = 0ORlossyear > 20(no loss observed in Hansen, or lost after 2020).B. Was it lost between 2021 and 2025 (Hansen window)?
Hansen GFC-2025 records annual tree-cover loss through 2025. A pixel that was forest at the 2020 cutoff and whose Hansen lossyear falls in 2021-2025 is counted as post-cutoff loss, dated to that exact year.
C. Was it lost after 2025? (Sentinel-1 + Sentinel-2 fusion)
For the years after the Hansen record (2026 onward) a live Sentinel-1 + Sentinel-2 fusion runs on pixels still forested at the end of that record. Phenology-matched quarter pairs (a quarter vs the same calendar quarter of the last Hansen year) are scored by a per-pixel 2-of-3 vote across NDVI, NBR and Sentinel-1 VH backscatter, with an NDVI∧NBR fallback where radar is unavailable. A pixel detected in ≥2 quarter pairs is confirmed; one detected only in the latest pair is marked provisional.
D. Otherwise: forest intact, or insufficient data.
A pixel that cleared step A and was not lost in step B or C is intact, but only if the step C check actually ran for its parcel. When the check has no result — the time limit of one analysis (about 150 seconds), which the check shares across every parcel with forest in the report; a Sentinel Hub service error; no quarter pair with enough cloud-free coverage; or the per-report Sentinel Hub call budget — the pixel counts as not evaluated rather than "no loss". A parcel with more than 20 percent of its 2020 forest not evaluated is at least
data_insufficient. Never assume absence of evidence is evidence of absence.
Baseline robustness. The fusion baseline is phenology-matched: each post-Hansen quarter is compared against the same calendar quarter of the latest Hansen year, which removes seasonal variation by construction. Combining Sentinel-1 radar with Sentinel-2 optical indices keeps detection robust to cloud cover and single-index false positives.
From pixels to a parcel-level verdict
For each parcel the analysis takes the union of Hansen-detected and Sentinel-1 + Sentinel-2-detected loss pixels, divides by the 2020 forest pixel count, and maps the resulting percentage to one of six verdicts:
no_riskNo forest detected at the 2020 cutoff. Goods originating here cannot have caused post-cutoff deforestation by definition.
negligible_riskForest existed in 2020 and zero post-cutoff loss was detected across the Hansen + Sentinel-1/Sentinel-2 chain.
low_riskBelow 0.1 percent of the 2020 baseline. Document the supply chain and consider mitigation before relying on this verdict.
review_requiredBetween 0.1 percent and 1.0 percent. Manual review of flagged pixels is required before placing goods on the EU market.
risk_detectedAt or above 1.0 percent of the 2020 forest baseline. Goods sourced from this parcel are presumed to carry deforestation risk under EUDR Article 9(1).
data_insufficientMore than 20 percent of the parcel's 2020 forest has no result from the Sentinel-1 + Sentinel-2 check of the years after 2025, or the parcel is too small to contain the centre of any Hansen pixel. The usual cause is the time limit of one analysis (about 150 seconds) when several parcels with forest share a report; splitting them into smaller reports usually completes the check. A service error or no usable cloud-free quarter pair can also leave it without a result. Missing data never lowers a low-risk, review or risk verdict.
Report-level aggregation. When a report covers multiple parcels, the report-level verdict is the worst per-parcel verdict, using the severity ordering: no_risk < negligible_risk < data_insufficient < low_risk < review_required < risk_detected.
WGS84 to six decimal places
All coordinates in our reports — both parcel centroids and the vertex list — are WGS84 (EPSG:4326) to six decimal places. Six decimals at the equator is approximately 0.11 m, well inside any field instrument's precision and well inside the Hansen pixel grid (30 m).
Hansen GFC and Sentinel-2 L2A are both natively in WGS84, so no reprojection happens inside the analysis pipeline. Areas reported in hectares use a latitude-corrected pixel side: a Hansen pixel is 1 arc-second on a side, which is approximately 30 m north-south everywhere but shrinks east-west with cos(latitude). At 46 degrees north (typical Romania) one pixel is approximately 0.066 ha; at 24 degrees south (Chaco region) it is approximately 0.082 ha.
What this analysis does NOT detect
Honesty about scope is part of due-diligence. The fusion pipeline above is deliberately conservative — it answers a single binary question per pixel ("was forest, then cleared after 2020?"). It does not:
- Detect forest gains. Afforestation and natural regrowth are visible in Hansen v1.7+ "gain" layers but the EUDR is concerned with losses, not gains, so we do not factor them in.
- Catch sub-pixel clearings. A clearing smaller than one Hansen pixel (approximately 0.07 ha) will not register unless it pushes the pixel's remaining canopy below 10 percent. Operators with smallholdings should document on-the-ground evidence.
- Identify selective logging. If selective logging removes some trees but the remaining canopy stays above 10 percent, the pixel remains "forest" under the FAO definition that EUDR adopts.
- Submit a DDS to TRACES NT. That remains your responsibility under EUDR Article 4. Our report is evidence you attach.
- Verify your supply-chain geolocation. We trust the polygon you supply. If the polygon is wrong the report is wrong.
ISCC EU and ISCC PLUS — the same methodology, different cutoff and indicators
ISCC certification has been part of the EU biofuels regulatory framework since 2011. Unlike EUDR — which applies to a broad list of commodities placed on the EU market and uses a December 2020 cutoff for deforestation — ISCC focuses on biofuel and bioenergy supply chains and uses a January 2008 cutoff for permissible land-use change.
The technical core of an AGROGIS ISCC compliance report is the same as the EUDR report described above: Hansen Global Forest Change as the historical baseline, with the same Sentinel-1 + Sentinel-2 fusion covering the years after the Hansen record: clearing it detects there lifts an ISCC parcel to at least review required. What differs is:
- Cutoff date. 1 January 2008 (ISCC) versus 31 December 2020 (EUDR).
- Additional layers. Forest loss is reported at two canopy thresholds, 10% and 30%; wetland and mangrove land cover is counted inside the parcel; and the parcel is screened against international and national protected-area registers. High Conservation Value, peatland and grassland layers are not part of the analysis.
- Output framing. ISCC certification bodies care about the evidence for the certification process. We present the analysis as a per-parcel verdict with the supporting data trail, not as a final certification. The certification itself is conducted by accredited bodies (SGS, Bureau Veritas, DNV, others) — AGROGIS provides only the spatial evidence layer they review.
For details on the Hansen layer, the fusion logic, the verdict thresholds and the worked examples, see the EUDR sections above — the same logic and the same numbers apply, only the cutoff year shifts.
What an AGROGIS compliance report is — and what it is not
AGROGIS compliance reports are designed to surface spatial risk signals from satellite data. They are a useful evidence layer in any compliance workflow. They are not:
- Not a certification. AGROGIS is not an accredited certification body under either EUDR or ISCC. A report from AGROGIS does not exempt the operator from due-diligence obligations or from working with an accredited auditor.
- Not a legal opinion. Whether a given parcel meets a given regulatory requirement is ultimately a question for the operator's legal counsel and the relevant authority — not for a remote-sensing report.
- Not exhaustive in scope. Satellite data has known limitations: persistent cloud cover, very small change events below sensor resolution, and land-use changes that do not produce a visible canopy signal (for example, soil disturbance under closed canopy). We document the confidence level of each per-parcel verdict accordingly.
- Not a substitute for on-site verification. Where the spatial signals are inconclusive or contested, on-site verification by an accredited third party is the appropriate next step. The AGROGIS report is designed to make such verification more targeted, not to replace it.
These limitations are stated explicitly in every report PDF and they are stated here so that anyone arriving at this page directly — whether from the marketing site, from a printed report cover, or from a procurement team's review — has the same understanding of scope.
What to read next
- Hansen, M. C., et al. (2013). "High-resolution global maps of 21st-century forest cover change." Science 342 (6160), 850-853.
- Copernicus Sentinel-2 L2A documentation: dataspace.copernicus.eu
- Regulation (EU) 2023/1115 on deforestation-free products: eur-lex.europa.eu
- FAO Global Forest Resources Assessment 2020 (the canopy / forest definition EUDR adopts).
Need a report for your parcels?
EUDR reports are bought as credits — one credit, one report, covering up to 50 parcels. €99 for a single report, down to €79.20 each in a pack of 20. On any paid plan, buy them in the app and run the analysis yourself; the PDF evidence report lands in your dashboard.