Methodology
How Earthline turns satellite imagery and supply-chain records into auditable, plot-level evidence for EUDR due diligence — and what that evidence is not.
Earthline’s role
Earthline is a geospatial evidence provider. Legal responsibility for due diligence, for filing the Due Diligence Statement (DDS) and for placing product on the EU market stays with the operator. Every assessment follows a written, versioned procedure with separate roles for mapping, monitoring, risk analysis and an independent QA/QC reviewer; the person who signs off a result is never the analyst who drew the geometry.
The plot verdict
Each plot receives a binary verdict — Compliant or Non-compliant. It is cumulative: a plot is compliant only if it passes both tests.
Test A — deforestation-free
No tree-cover loss strictly after 31 December 2020 (Regulation (EU) 2023/1115; the cutoff is unchanged by the 2025/2650 amendment). Raw alerts are never a verdict on their own.
Test B — legality
Produced in accordance with the relevant legislation of the country of production (Article 2(40)), including the legal function of the land — Indonesia’s kawasan hutan.
Village risk
Separately, the village around each plot is scored Low, Medium or High with a multi-criteria index. The variables are:
- Post-2020 deforestation dynamics (validated alerts)
- Legal forest function — overlap outside APL
- Supply-chain traceability
- Cadastral maturity (ATR/BPN)
- Customary and social-forestry presence
- Agricultural matrix (oil-palm cover)
- Jurisdictional governance
- Baseline forest cover remaining
Weights are fixed at the start of each assignment. Beneficial-ownership and sanctions screening is a separate flag, not part of the score. Article 10(2) concerns that can’t be scored per village — corruption, grievances, human-rights allegations — are kept as sourced qualitative flags; no flag is not proof of no risk.
Mitigation categories A–F
The plot verdict is crossed with the village risk tier. Every plot lands in exactly one class:
| Class | Plot and village risk | Outcome |
|---|---|---|
| A | Compliant — Low-Risk Area | Approved for Submission |
| B | Compliant — Medium-Risk Area | Approved with Routine Monitoring |
| C | Compliant — High-Risk Area | Approved with Enhanced Monitoring |
| D | Non-Compliant — Low-Risk Area | Pending Geospatial Review |
| E | Non-Compliant — Medium-Risk Area | Mitigation Required Before Submission |
| F | Non-Compliant — High-Risk Area | Not Approved for Submission |
The wording is exactly what each plot carries in its data. A–C are compliant plots, approved for submission with monitoring that grows with village risk; D–F are non-compliant. Mitigation has to be proportionate, documented and effective (Article 11) before residual risk is treated as negligible.
Baseline: what counted as forest in 2020
A harmonised existing-forest layer is resampled to a 10 m grid aligned with Sentinel-2. Areas are always computed in an equal-area projection, never Web Mercator.
| Layer | Source | Role |
|---|---|---|
| Global Forest Cover 2020 | European Commission JRC | Primary regulatory baseline (FAO forest: > 0.5 ha, trees > 5 m, canopy > 10%). Non-binding. |
| SBTN Natural Cover | Science Based Targets Network | Wider ecological baseline — non-forest ecosystems, peat, wetland — for NDPE |
| HCSA | High Carbon Stock Approach | Separates secondary forest from degraded land |
| Forest cover | KLHK | Host-country context for classification differences |
Detecting and validating loss
Alert feeds run continuously and are triaged. Every alert that is used must pass spectral validation that separates permanent clearing from legal replanting.
| Sensor / product | Resolution | Use |
|---|---|---|
| RADD (Sentinel-1 radar) | ~10 m | All-weather change; closes cloud and smoke gaps |
| GLAD (Landsat) | ~30 m | Optical tree-cover loss under clear sky |
| GLAD (Sentinel-2) | ~10 m | Optical loss on the 10 m analysis grid |
| NDVI / NBR time series | Sensor-native | Before / at / after check — clearing vs replanting |
| PlanetScope composites | ~3 m | Cloud-free dating and boundary support |
| Maxar / Pléiades | Sub-metre | Micro-plots and ambiguous boundaries |
- Validated loss is overlaid on the 2020 baseline.
- A minimum mapping unit removes slivers and sub-pixel noise.
- Legal overlays use a small buffer so cadastral shift can’t create false flags; only core overlap is a hard legal flag.
The most precise geometry available
For each supplier the most precise spatial model the data supports is used — never a mill buffer when plot or estate geometry exists.
| Preference | Model | Used when |
|---|---|---|
| 1 — highest | Estate polygon | A surveyed estate, concession or HGU/IUP boundary exists |
| 2 | Cadastral parcel | An ATR/BPN parcel exists, clipped to planted oil palm |
| 3 | Village boundary | Only the village or sub-district is known (typical smallholder) |
| 4 — lowest | Mill catchment buffer | Only the mill’s location is known — 20 km by default, adjusted for roads and terrain |
Geometry rules
- Plots over 4 ha are submitted as a polygon of the actual production area; smaller plots may be a point, but a polygon is preferred.
- One feature is one plot. No overlapping, floating or self-intersecting polygons; rings are closed.
- Mixed estates are segmented first (roads, canals, cadastral limits, planting cohort), and each segment is assessed on its own.
- Every split keeps its lineage — parent, segment, reason and snapshot date — so it can be audited back to the source.
Accuracy checks
| Control | What it does |
|---|---|
| Rubber-sheeting | Aligns legacy cadastral boundaries to a WGS 84 orthomosaic |
| Buffer on legal overlays | Keeps boundary noise from creating legality flags |
| Minimum mapping unit | Removes sliver and sub-pixel artefacts |
| Projection discipline | Geometry in WGS 84; areas in an equal-area projection |
| Precision | Latitude and longitude to at least six decimal places |
| Independent review | Sample geometry signed off by someone who didn’t draw it |
| Viewer test | Files must open correctly in a GIS before they count as ready |
Delivery formats
The supply-base geolocation layer is delivered in standard formats:
- For the EU Information System: GeoJSON, WGS 84 (EPSG:4326), [longitude, latitude], at least six decimals; Point, Polygon or MultiPolygon; at most 25 MB per file.
- For GIS and risk-system integration: Shapefile, KML or KMZ on request.
- Maps are built per assignment from the operator’s own sourcing records. There is no public all-client map.
Monitoring and re-assessment
Monitoring follows the A–F class. A plot is re-assessed — against a newly frozen, versioned data snapshot — when:
- new post-2020 loss is validated;
- the EU’s country benchmarking changes;
- a gazette or boundary is revised, or a legality instrument expires;
- the supply base changes;
- a substantiated concern is raised (Article 31).
What you see in PalmTrace
The workspace shows the result of this procedure: each plot’s verdict, risk, category and the reason for any non-compliance, alongside JRC 2020 forest cover (version V3) and dated Sentinel-2 imagery from 2020 and 2026 as context layers. Context layers are shown next to the verdict, never blended into it.
What this is not
- It is not legal advice, and it does not replace the operator’s decision or a competent authority’s.
- It does not claim every chain has polygons for every mill — coverage is shown next to every figure.
- It does not publish client data. Nothing on this page comes from a client’s supply chain.
Regulatory basis: Regulation (EU) 2023/1115 as amended by Regulation (EU) 2025/2650. Cutoff 31 December 2020. Indonesia is currently treated as standard risk under Article 29, so full due diligence applies. Application dates, benchmarking and Information System requirements are checked at the start of each assignment.