Satellite Deforestation Monitoring for EUDR: Tools and Plot Data

Satellite deforestation detection gets bought as a technology decision. Under the EU Deforestation Regulation it is a documentation decision: before a consignment moves, you have to show plot by plot that the land was not cleared after 31 December 2020.

What EUDR actually requires before a shipment moves

Regulation (EU) 2023/1115 covers cattle, cocoa, coffee, oil palm, rubber, soya and wood plus their derived products. For every consignment the operator has to hold three things: the geolocation of every plot where the commodity was produced, as coordinates or as a polygon for larger plots; a risk assessment that the goods are deforestation-free and legal under the producer country's law; and a due diligence statement filed before the goods reach the EU market.

The obligation applies from 30 December 2026 for medium and large operators and from 30 June 2027 for micro and small enterprises. The six-month sanction grace period that circulated in earlier drafts was removed by the legislators.

Note what is missing from that chain: the regulation never asks for a satellite. It asks for evidence that one piece of land did not change between the cut-off and the harvest. Monitoring produces part of that evidence cheaply, it is not a requirement in itself, and that distinction shapes how you build a compliant supply chain.

How satellite and AI deforestation detection works

Detection rests on two data families. Optical sensors such as the Copernicus Sentinel-2 mission record reflected light in several spectral bands: good on vegetation type, blind under cloud, which is a real constraint over tropical regions in the wet season. Radar sensors such as Sentinel-1 emit their own C-band signal and see through cloud and at night, at the cost of imagery that is harder to interpret.

A monitoring system builds a baseline of the plot as it looked at the cut-off, then compares each new acquisition against it. Where reflectance or backscatter shifts beyond a threshold, it raises a change alert. Artificial intelligence enters at the classification step: a trained model decides whether a change is clearance, selective logging, seasonal leaf loss or cloud shadow, and attaches a confidence score. That classification varies most between providers and is the part you should test rather than trust. What an auditor wants is not the map but a dated, georeferenced record of the plot state at the cut-off and at delivery.

Choosing and testing a deforestation monitoring tool

Start with the evidence file the tool exports, not the alert feed, and work backwards to the data behind it. Four data classes are on offer, and most operators combine two.

What you can obtain without supplier questionnaires: land cover change on a known coordinate, from any of the four rows. What it cannot give you is the coordinate itself, the harvest date and the legal status of the clearing. Screening goes questionnaire-free, the statement does not.

In the file itself, the polygon does the work the questionnaire used to do. A plot geometry with a dated state at the cut-off and a change history since then replaces the supplier's self-declaration that no clearing took place: instead of a signature on a form you hold a georeferenced record an auditor can re-check against the same archive. What a polygon cannot replace is the supplier's statement of which plots fed which consignment, the volumes and the legality evidence under national law. Collect the geometry once, keep the declaration narrow, and the questionnaire shrinks from a survey to a mapping exercise.

Before signing, pilot the tool on plots whose history you know: two cleared after the cut-off, two not. A tool that flags all four, or none, is not calibrated for your commodity. Buying or assembling that capability is the usual build, buy or automate decision. Per plot, the evidence file should carry:

• plot identifier and geometry, with the coordinate reference system stated
• plot state at 31 December 2020, with source scene and date
• change history between cut-off and harvest, each alert with its disposition
• an export format you can still hand an auditor after the contract ends

Where EUDR due diligence actually fails: plot data, not satellites

In an Article 8/9 classification I supported for a venture fund, the do-no-significant-harm thresholds did not break on strategy. They broke on the data-procurement routine: nobody owned the job of asking the same question, the same way, every quarter. EUDR breaks one tier further out for the same reason.

The satellite layer is the solved part. The unsolved part is a defensible coordinate for every plot when the last mile is thousands of smallholders selling into a cooperative, or a trader who treats farm identity as confidential. An importer with a few hundred suppliers can buy monitoring in an afternoon and still spend a year collecting geolocation, a data problem, not an imaging one.

Three things satellite evidence cannot prove, whatever the resolution:

• Legality. Clearing can be lawful nationally and still count as deforestation here.
• Ownership and identity. Imagery shows that a plot changed, not whose beans reached the mill.
• Degradation under canopy. Selective logging often leaves the canopy signal largely intact.

Plan the data-procurement routine first and the monitoring subscription second.

Risk classification and what non-compliance costs

The Commission benchmarks producer countries as low, standard or high risk, and the class sets the workload. Low-risk countries qualify for simplified due diligence and fewer authority checks; standard and high-risk sourcing carries the full obligation, which is where deforestation exposure meets the rest of your hidden supply chain risk.

Non-compliance is priced against turnover: fines whose maximum must be set at no less than 4 percent of annual EU-wide turnover, confiscation of the goods and the revenue from them, exclusion from public procurement. The earlier exposure is commercial: a consignment held at the border is a service failure with a customer, not a fine.

The plot evidence is reusable: the same land cover history feeds ESRS E4 disclosures on land use change, so biodiversity reporting gets cheaper once EUDR forces the data into existence. It is climate change data as well, because clearing a plot releases stored carbon, and the E1 chapter asks about that in the same reporting year.

Frequently Asked Questions