Geo Analysis in Detail

This page explains step by step how Polygon One checks a plot. It is written for compliance officers. You do not need any background in satellite data. For an overview of all data sources and risk flags, see Risk Assessment.

​

What the analysis answers

The analysis answers one question: Was forest removed or damaged on this land after 31 December 2020? That is the EUDR cut-off date (Art. 2(13)). The regulation distinguishes two cases:

• Deforestation (Art. 2(3)): forest is converted to agricultural use, for example cropland, pasture or an oil palm plantation.
• Forest degradation (Art. 2(7)): primary forest or naturally regenerating forest is converted into plantation forest or other wooded land. The analysis checks this question for wood only.

What happened before the cut-off date does not matter for the EUDR. Land that was not forest in 2020 cannot be deforested under the regulation.

​

The checks in order

The checks always run in the same order. As soon as one step settles the result, the analysis stops.

​

1. Is the geometry plausible?

First, the system checks whether the area can be a production area at all. It uses the ESA WorldCover land-cover map. The plot fails if:

• more than 10 % of the area is built up (buildings, roads, settlement),
• more than 10 % of the area is water, or
• there is no land-cover data for the geometry, usually because it lies in the sea.

If the plot fails, the result is “Invalid”. Details and solutions are under When the geometry is not plausible.

​

2. Was there forest on 31.12.2020?

Next, the system reads the JRC Global Forest Cover 2020 map (on the map: “JRC 2020 forest baseline”). It shows where forest stood on the cut-off date.

• If forest covers at most 1 % of the plot, there was no forest at the cut-off date. The result is “Low Risk”.
• Exception: if other datasets still show clear tree-cover loss after 2020 (Hansen: more than 1 % of the plot) or a high-confidence disturbance alert after 2020 (GFW), the data disagree. The result is then “Moderate Risk” with the flag “Datasets disagree about forest presence at the EUDR cutoff”. This never leads to high risk.

​

3. Which forest type was there? (wood only)

For wood, the system also reads the JRC Global Forest Types 2020 map (GFT). It splits the 2020 forest into Primary Forest, Natural Forest and Planted Forest. This decides whether a later harvest counts as degradation. The forest types are explained below. For other commodities (for example cocoa, coffee, soy), the forest type does not matter. Only the conversion of forest into agriculture counts there.

​

4. Did anything change after 2020?

Now the system looks for changes after the cut-off date. It uses several independent datasets:

• Hansen Global Forest Change: tree-cover loss measured each year.
• JRC TMF (Tropical Moist Forest): deforestation and degradation in tropical moist forests, with a time series.
• GFW disturbance alerts (GFW/DIST): near-real-time alerts about vegetation disturbance.
• PALSAR radar: confirms forest independently of clouds and detects shade crops under tree canopy.
• ESA WorldCover: land cover, for example cropland, forest or agroforestry.

​

5. How the signals become a risk level

No single dataset decides on its own. These rules apply: Measured tree-cover loss (Hansen)

• “High Risk” arises when more than 5 % of the 2020 forest area was lost after 2020. In addition, the 2020 forest area must be at least 0.5 ha and the lost area at least 0.18 ha (two satellite pixels).
• Smaller but detectable loss leads to “Moderate Risk” with the note “Minor post-2020 tree cover change detected”. For wood, a loss of up to 1 % of the forest area is treated as too small; the other datasets then decide.
• Loss smaller than one pixel (0.09 ha) is not treated as deforestation. Signals that small cannot be told apart from edge effects at the plot boundary.

Disturbance alerts (GFW)

• Alerts alone lead to “Moderate Risk” at most.
• “High Risk” only arises when a second dataset confirms the loss (Hansen or TMF), or when high-confidence alerts occur in at least two different years after 2020. In the second case, the 2020 forest area must be at least 0.5 ha and the alert area at least 0.18 ha.
• Low-confidence alerts only lead to “Moderate Risk”.

Tropical moist forest (TMF)

• If TMF shows deforestation that the time series dates before 2020, it counts as historical and does not raise the risk.
• A TMF signal after 2020 on its own leads to “Moderate Risk” at most. For “High Risk”, a second dataset must confirm the change. Exception: for wood on primary or natural forest, TMF deforestation after 2020 counts as evidence of degradation (see the next paragraph).

Forest degradation for wood (GFT)

• If primary forest (more than 1 % of the plot) or natural forest (more than 5 % of the plot) stood on the land in 2020, the system checks for degradation.
• “High Risk” only arises when a change after 2020 is confirmed: by Hansen or TMF, and for primary forest also by alerts in at least two years. For natural forest alone, alerts are never enough.
• For primary forest, in addition: if the measured change lies demonstrably on the planted part only, the result stays at “Moderate Risk”.
• Without a confirmed change, the result stays at “Moderate Risk” (“Primary or naturally regenerating forest at the 2020 baseline, but no post-2020 conversion evidence. Verification recommended.”).

Low risk

• “Low Risk” arises, for example, when Hansen shows no loss and GFW covers the plot and reports no alerts.
• Or, for wood, when the land is a confirmed plantation (see Forest types).

Missing data never lowers the risk

• If a dataset is missing, the system does not treat the plot as clean. Typical results are then “Moderate Risk”, for example when GFW data is missing, when no change dataset covered the plot after 2020, or when GFT data is missing for wood (“Forest detected but GFT degradation data unavailable”).
• Forest presence alone is not evidence of deforestation either. If forest stands on an agricultural plot and no dataset shows a change, the result stays at “Moderate Risk” for verification.

​

Forest types

The forest types come from the JRC Global Forest Types 2020 map. They describe the state on the cut-off date. On the map, the layers are called “GFT natural or primary forest” and “GFT planted forest”.

​

What this means for the EUDR

• Harvesting a plantation is neither deforestation nor degradation. When a plantation is harvested and replanted, it stays forest of the same type.
• Degradation means primary or natural forest becomes plantation forest or other wooded land (Art. 2(7)). That is why the 2020 forest type matters for wood.
• Deforestation means forest of any type becomes agricultural land (Art. 2(3)).

​

When land counts as a confirmed plantation

The system treats wood land as a confirmed plantation when all three conditions hold:

1. primary forest covers at most 1 % of the plot,
2. natural forest covers at most 5 % of the plot, and
3. more than half of the 2020 forest is planted forest.

On a confirmed plantation, tree-cover loss and disturbance alerts after 2020 count as harvest. The result is then “Low Risk”. The signals stay visible, and the land continues to be monitored for a change in land use. The harvest attribution no longer applies if:

• TMF measures deforestation after 2020, meaning the land does not become forest again, or
• the disturbances lie on the natural or primary part of the land and affect at least 5 ha and at least 10 % of that part.

​

Datasets and their limits

​

Risk levels

​

When the geometry is not plausible

If the result is “Invalid”, the plot failed the plausibility check. The plot details then show “Sanity Check Failed” and the reason. There are three reasons: The article then shows the status “Invalid”. It blocks until the plot is resolved. How this affects the article status is described in the Findings Catalogue.

​

Typical causes

• Company or warehouse address instead of the growing area. The coordinates point to an office, a sawmill or a warehouse. That lies in a settlement, hence the overlap with urban area.
• Latitude and longitude swapped. The point lands somewhere completely different, often in the sea. Then there is no land-cover data.
• Point instead of polygon. For the analysis, a point is expanded into a circle the size of the declared area. If the point sits at the farmstead or in the village, the circle falls on buildings and roads.
• Boundary drawn into a lake or river. The polygon reaches past the shore into the water.

​

What you do

The geometry is wrong: correct it, or ask the supplier to correct it. As the customer, you can correct or delete any invalid plot. A supplier can only change the plots it created itself. After the correction, the plot is analysed again. The geometry is correct: sometimes the area is captured correctly and still overlaps, for example a plantation right on a lakeshore. In that case, contact support.

​

What to do with moderate or high risk

1. Open the plot and read the Risk Flags. They name the dataset and the reason.
2. On the map, use “Layers” to see where the change lies, for example “Hansen loss on planted forest” or “GFW alerts on natural or primary forest”.
3. Request suitable evidence from the supplier, such as a planting record or a management plan. You do this in the compliance gaps view.

​

Frequently asked questions