What you have to show about the cut-off date
Regulation (EU) 2023/1115 requires the commodity to come from land that has not been deforested after 31 December 2020. Deforestation is the conversion of forest to agricultural use (Art. 2(3)), and forest is land of more than 0.5 ha with trees higher than 5 m and a canopy cover of more than 10%, excluding land that is predominantly under agricultural or urban use (Art. 2(4)).
The operator has to collect conclusive and verifiable information that the product is deforestation-free (Art. 9(1)(g)). The regulation does not prescribe a dataset, but authorities may use satellite imagery in their checks (Art. 18). Overlaying each geolocated plot with public data is the most direct way to know where to look before you submit.
The three layers and what each one shows
Each layer answers a different question, and none is enough on its own:
| Layer | What it shows | Resolution | Updates |
|---|---|---|---|
| JRC Global Forest Cover 2020 (GFC2020) | Forest or non-forest in 2020 | 10 m | Occasional revisions (V3 Nov 2025, V4 in 2026) |
| Tree cover loss (UMD, Hansen) | Year in which tree cover was lost | 30 m (Landsat) | Once a year |
| GFW integrated alerts | Recent disturbances, with date and confidence | 10 m grid | Daily |
- The JRC map is made by the Commission's Joint Research Centre for the EUDR. The JRC itself says its use is not mandatory, not exclusive and not legally binding. On Global Forest Watch (GFW) it is listed as version v2020.3; the JRC has already published a version 4.
- The University of Maryland loss data comes out once a year. Version v1.13, with data through 2025, was released on 29 April 2026. Earlier releases came out between April and June, so the 2026 data is expected in spring 2027; check GFW.
- GFW integrated alerts combine several systems (GLAD, RADD and DIST-ALERT), cover the whole world and tag each pixel with a confidence level: nominal, high or highest. They show what happened after the last year of yearly data, but GFW warns they do not measure area.
Why is loss not the same as deforestation?
The Maryland data measures tree cover, not land use. When an acacia or eucalyptus plantation is harvested for pulp, the satellite sees trees disappear and records loss, even if the plot is replanted and stays what it was in 2020. The same happens with a fire, a storm or the renewal of an old plantation. DIST-ALERT alerts also flag harvests.
It cuts the other way too. The JRC map excludes agricultural tree crops such as oil palm, rubber or cocoa, and may classify some plantation forests as non-forest. A palm oil plot planted before 2020 will usually come out as no forest in 2020, which is what you would expect. That is why loss should be restricted to 2020 forest rather than read over the whole plot.
For timber you also need to look at forest degradation, which the regulation treats separately. Harvesting a planted forest is not deforestation, but you cannot accept it without the supplier's documents either.
How to read a plot's result
A useful result never says "deforested" or "clean": it says what the data saw and what to do next. Sourevo uses these verdicts:
- No loss: there was forest in 2020 and no significant loss is detected inside it after 2020.
- Loss, to review: there is loss inside 2020 forest from 2021 onwards. Ask the supplier for an explanation (plantation rotation, fire, felling licence) and decide on evidence.
- No forest in 2020: the JRC map sees no forest on the plot. This is the usual case for long-established farmland.
- Inconclusive: the area is too small to measure, or there are recent high-confidence alerts after the yearly data. Look at imagery or wait for the next year of data.
To avoid noise at the edges, Sourevo ignores loss under 0.1 ha or under 1% of the area, unless it reaches 1 ha. A plot declared as a point (up to 4 ha) is analysed as a 4 ha circle around the point, about 113 m in radius, and recorded as an approximation: with the real polygon the result is more reliable.
What to keep as evidence
Due diligence information is kept for 5 years (Art. 12). For a satellite check that means keeping not only the verdict but the data behind it: dataset, exact version (for example UMD v1.13 with data through 31-12-2025), the date the alerts are current to, the date of the check, the geometry analysed and the loss per year.
A result ages. When Maryland publishes a new year or the JRC changes the map, an old check no longer reflects the current data. Sourevo treats a result as current for 30 days for the same geometry and flags it as outdated when a new data version comes out. When the DDS is submitted, it freezes each plot's result with its versions and dates in the record. How that fits into the risk assessment and a valid DDS is covered in those guides.
What the satellite cannot see
The check is a filter, not the whole due diligence:
- Clouds: Landsat is optical and in cloudy tropical areas it can take time to see a change. Radar alerts (RADD) help but do not cover everything.
- Resolution: a Landsat pixel is about 0.09 ha. Very small clearings or selective logging under the canopy may not show.
- Geometry: if the supplier's GeoJSON is badly drawn or shifted, you are checking the wrong place.
- Legality: land titles, permits, labour rights or community consent cannot be seen from space. Art. 9(1)(h) asks for separate information.
- Low-risk countries: under simplified due diligence (Art. 13) no risk assessment is required, but geolocation is still mandatory.
Frequently asked questions
- Do I have to use the JRC forest map?
- No. The JRC says its use is not mandatory, not exclusive and not legally binding. You can use other data or imagery, as long as you document what you used and why you concluded the risk is negligible.
- The satellite shows loss in 2023 on a eucalyptus plantation. Can I submit?
- Only if, after review, the risk is negligible: for example, with evidence that the plantation already existed in 2020 and was harvested in rotation. Keep that evidence with the check. If there is none, leave the plot out of the batch.
- How often should the check be repeated?
- The regulation sets no frequency. It makes sense to repeat it when the geometry changes, when a new version of the yearly data comes out, and before each DDS if the last result is weeks old. Daily alerts are for monitoring in between.
- Why does my plot show "no forest in 2020" when it has trees?
- Because the JRC map excludes agricultural tree crops (palm, rubber, cocoa, coffee) and may not count some plantation forests as forest. For the EUDR that is usually a good sign, but confirm that the plantation already existed in 2020.