EUDR Assessment Methodology

How TreeSight uses satellite imagery to support EU Deforestation Regulation due diligence

1. Regulatory Context

The EU Deforestation Regulation (Regulation 2023/1115, "EUDR") requires operators placing certain commodities on the EU market to demonstrate that those products were not produced on land subject to deforestation or forest degradation after 31 December 2020.

Article 10 of the EUDR requires a due-diligence statement with geolocation data confirming the origin of products. TreeSight provides satellite-based evidence to support — but not replace — that due-diligence obligation.

Key EUDR Dates

  • Cutoff date: 31 December 2020 — land must have been non-deforested after this date
  • Large operators: Obligations apply from 30 December 2025
  • SMEs: Extended to 30 June 2026

2. Data Sources

TreeSight analyses multiple Earth observation datasets to build a comprehensive vegetation history for each parcel:

Source Resolution Coverage Role in Assessment
Sentinel-2 L2A 10 m/px Global, 5-day revisit (since 2017) Primary NDVI time series for vegetation change detection
NAIP 0.6 m/px Continental US only High-resolution visual confirmation (summer seasons)
ESA WorldCover 10 m/px Global (2020 & 2021 epochs) Land-cover classification baseline — confirms tree cover at cutoff date
Open-Meteo ~25 km Global Weather context — distinguishes drought stress from land-use change
FIRMS / MODIS 1 km Global Fire hotspot detection — identifies burn events
WDPA Vector Global Protected area overlap — additional risk indicator

3. Analysis Method

3.1 Seasonal Frame Plan

For EUDR assessments, TreeSight builds a time series starting from January 2021 (immediately after the EUDR cutoff). Four seasonal observations per year (winter, spring, summer, autumn) capture the vegetation cycle. This avoids false positives from natural seasonal leaf-fall.

3.2 NDVI Computation

The Normalised Difference Vegetation Index (NDVI) is computed for each seasonal frame using Sentinel-2 bands B08 (near-infrared) and B04 (red):

NDVI = (B08 − B04) / (B08 + B04)

NDVI ranges from −1 to +1. Healthy vegetation typically yields 0.3–0.8. Values below 0.2 may indicate bare soil, impervious surfaces, or water.

3.3 Year-over-Year Comparison

TreeSight compares the same season across different years to detect genuine vegetation loss. A decline of >0.15 in same-season NDVI between consecutive years is flagged as a significant event requiring attention. This approach avoids false alarms from comparing winter (low NDVI) with summer (high NDVI).

3.4 Change Detection

Where NDVI rasters are available, pixel-level change detection identifies the spatial extent and magnitude of vegetation changes. The system reports:

  • Mean NDVI delta — average change across the AOI
  • Trajectory — Improving, Stable, or Declining
  • Significant events — individual year-over-year declines above the threshold

3.5 AI-Assisted Interpretation

An AI model (Azure AI Foundry) analyses the aggregated data and produces a structured assessment with a clear deforestation-free conclusion, confidence level, supporting evidence, and identified risk factors.

4. Assessment Output

The EUDR assessment produces:

  1. Binary conclusion: "Deforestation-free" (yes/no) with confidence level (high/medium/low)
  2. Supporting evidence: Specific NDVI values, trends, and observations
  3. Risk factors: Any caveats (data gaps, border seasons, cloud cover)
  4. Contextual data: Weather patterns, fire events, flood events, land-cover class, protected-area status
  5. Exportable report: GeoJSON, CSV, or PDF for audit records

Confidence Levels

  • High: 8+ seasonal observations, all showing stable or improving vegetation, no significant decline events
  • Medium: 4–7 observations, minor fluctuations within normal ranges
  • Low: Fewer than 4 observations, or one or more significant decline events requiring investigation

5. Limitations & Caveats

  • Resolution: Sentinel-2 at 10m cannot distinguish individual trees — suitable for parcel-level assessment, not single-tree monitoring
  • Cloud cover: Persistent cloud cover in tropical regions may reduce the number of valid observations
  • Degradation vs. deforestation: Gradual degradation may not trigger NDVI thresholds in a single season comparison
  • Time lag: Sentinel-2 has a 5-day revisit but usable cloud-free data may be less frequent
  • Baseline assumption: The analysis compares post-2020 trends — it does not independently verify land cover as of 31 December 2020 (ESA WorldCover 2020/21 provides the closest proxy)

Important Disclaimer

TreeSight's satellite analysis provides supporting evidence for EUDR due diligence but does not constitute a complete due-diligence assessment under Regulation 2023/1115. Operators remain responsible for fulfilling all EUDR obligations, including in-country verification, supply-chain traceability, and declaration submission. This tool is intended to augment — not replace — professional judgement and on-the-ground verification.

6. References