EUDR Compliance Gets a Free 10 m Forest Map: What ForTy Changes

Google DeepMind and the World Resources Institute have released ForTy, a free global map that labels every 10-by-10-meter patch of land as one of six forest and land-use types for the year 2020. That year is no accident. It's the reference point for the EU Deforestation Regulation, and for any company working on EUDR compliance, a 2020 baseline was the hardest piece of evidence to get.

If you ship coffee, cocoa, palm oil, soy, rubber, cattle or wood products into Europe, the large-company deadline is 30 December 2026. This post covers what ForTy actually gives you, where it falls short, and how to turn it into a working screening pipeline instead of another spreadsheet.

TL;DR#

• ForTy is an open 10 m map (CC BY 4.0) that separates primary, naturally regenerating, planted and plantation forest from tree crops and other land, with a probability score per class.
• It is very good at "forest or not" (about 95%) and much weaker on managed classes like plantations and tree crops, so treat it as a screening layer, not a verdict.
• The value comes from wiring it into supplier data, thresholds and human review, with a clean evidence trail for each due diligence statement.

What DeepMind and WRI released#

According to the Google DeepMind announcement, Forest Typology (ForTy) 2020 classifies the planet's land into six types based on FAO definitions:

1. Primary forest, with no visible signs of human disturbance
2. Naturally regenerating forest, which grew back after logging or fire
3. Planted forest, managed over long cycles
4. Plantation forest, meaning fast-growing tree farms
5. Tree crops and agroforestry, such as oil palm, rubber, cocoa and fruit
6. Other land, from cropland to cities

Each pixel carries a probability for every class, so you choose how strict to be. There's also an early-release annual time series for 2017 to 2025 that shows change year by year, including natural forest being replaced by farms.

The map is free on Google Earth Engine and as a download, and The Decoder reports the map, classification layer and test data are all licensed CC BY 4.0. No application, no waiting list.

Why it matters for EUDR compliance#

The EUDR says covered products must not come from land deforested after 31 December 2020. For coffee, cocoa and palm oil, that means land that was forest in 2020 can't have been converted to farmland. For wood, the regulation cares about whether it came from converted primary or naturally regenerating forest. You need to know not just whether trees were there in 2020, but what kind.

That's exactly the gap ForTy fills, because older maps mostly said "tree cover yes or no".

The timing is tight. EY's summary of the latest amendment puts the application date at 30 December 2026 for large companies, with most micro and small enterprises given until 30 June 2027. Operators must collect geolocation for the plots their commodities came from and file a due diligence statement in the EU's TRACES system.

For US businesses this isn't somebody else's problem. If you export covered commodities or finished goods made from them (furniture or chocolate, for example) into the EU, your European buyers will ask you for plot data and evidence. Many already are.

The technical breakdown and its trade-offs#

How ForTy was built#

ForTy takes AlphaEarth Foundations embeddings as a key input. These are compact summaries of years of satellite observations for each 10 m pixel, combining optical imagery, radar and other sources. On top of that, DeepMind trained a custom model on about 1.7 million labelled locations, then ran several rounds of active learning that added roughly 43,000 expert corrections where the model struggled.

The lesson for anyone building geospatial AI: the heavy lifting was done by a foundation model's embeddings. The task-specific part was the labels and the error correction loop.

Where it's strong and where it isn't#

Figures are from The Decoder's analysis of the paper. Overall accuracy is 90.2%, but the authors note the easy "other land" class flatters that number, and shaded cocoa or coffee under natural canopy can be labelled as forest.

That's the honest trade-off. ForTy is excellent at telling you where to look, and a poor substitute for a person looking.

What this means for businesses#

The cost of a first-pass deforestation screen has just dropped close to zero. Until now, getting a credible 2020 baseline meant paying a specialist vendor or commissioning custom satellite imagery analysis. Now the base layer is free and open.

That doesn't make compliance free. The work moves to three places:

• Data quality upstream. Supplier plot coordinates are often wrong, duplicated or recorded as a single point for a large farm. Garbage polygons give you garbage risk scores.
• Decision rules. Someone has to decide what probability of "primary forest" triggers a hold, and document why.
• Review and evidence. Flagged plots need a human decision, supporting documents and an audit trail that maps cleanly onto each due diligence statement.

This is where most companies are actually stuck. The map isn't the bottleneck. The bottleneck is spreadsheets of supplier data that nobody trusts, emailed back and forth with buyers.

How to act on it: a practical checklist#

Here's the pipeline we'd build for a mid-sized exporter:

Supplier portal / ERP
        |
        v
Plot geometry check  -> reject: bad coords, overlaps, missing polygons
        |
        v
ForTy 2020 lookup (Earth Engine) + 2017-2025 change series
        |
        v
Risk rules (your thresholds) -> low risk: auto-clear, log evidence
        |
        v
High or uncertain -> human review queue -> decision + documents
        |
        v
Evidence pack per shipment -> due diligence statement

Steps to get there:

1. Inventory exposure. List which SKUs contain covered commodities and which EU customers buy them.
2. Collect plot data properly. Use a structured intake form, not email. Validate coordinates and geometry on entry.
3. Run a pilot screen. Pull ForTy probabilities for a few hundred plots and compare against what your field teams know.
4. Set and document thresholds. Start strict on primary forest and conservative on tree crops, where the map is weakest.
5. Add the time series. Flag plots where the class changed after 2020, not just plots that were forest in 2020.
6. Keep humans in the loop. Every flag gets a named reviewer, a decision and attached evidence.
7. Version everything. Record which map version and thresholds were used for each decision, because both will change.

How MagicMakers Lab approaches this#

We treat compliance tools like any other workflow: start from how your team actually sources, sells and documents, then automate the dull parts and keep people on the judgement calls. We've done this kind of work before. For GeoVerdant, we cut satellite-data analysis for agriculture from weeks to minutes. A ForTy-based screening pipeline wired into your ERP and supplier portal is a good fit for a fixed-scope Rapid MVP Sprint, so you're running real plots well before the December deadline.

Key takeaways#

• ForTy gives every company a free, open 2020 forest-type baseline at 10 m, which is the reference year EUDR cares about.
• It's strong on forest vs non-forest and primary forest, and weak on plantations and tree crops.
• Use it as a screening layer with documented thresholds and human review, never as an automatic pass or fail.
• The real work is clean plot data, decision rules and an evidence trail tied to each due diligence statement.
• Large companies face a 30 December 2026 application date, so pilots should start now.

FAQ#

What is EUDR compliance?#

EUDR compliance means proving that covered products sold into or exported from the EU (cattle, cocoa, coffee, oil palm, rubber, soy, wood and many derived goods) don't come from land deforested after 31 December 2020 and were produced legally. In practice it means collecting plot geolocation, assessing risk, mitigating it and filing a due diligence statement before goods reach the market.

When does the EUDR apply to my company?#

Under the amended regulation, large and medium operators must comply from 30 December 2026. Most micro and small enterprises have until 30 June 2027, though those already covered by the old EU Timber Regulation follow the December 2026 date. US exporters are affected indirectly too, since EU importers will require their plot data and evidence.

Can satellite imagery analysis prove a plot is deforestation-free?#

Not on its own. Satellite imagery analysis, including maps like ForTy, is very good at showing whether land was forest in 2020 and whether that changed. It's weaker at telling managed tree crops from natural forest. Combine it with supplier documents, field knowledge and a documented human review for anything flagged as high risk or uncertain.

Is the ForTy map free for commercial use?#

Yes. DeepMind says ForTy is freely available on Google Earth Engine under an open license, and the published data is under CC BY 4.0, which allows commercial use with attribution. You still pay for your own compute, storage and the engineering to connect it to supplier data, so budget for the pipeline rather than the map.

If EUDR has your team juggling supplier spreadsheets and buyer questionnaires, let's look at what a screening pipeline would take for your actual data and systems. Book a free audit.

Sources#

• Not All Forests Are Equal: Mapping the World's Forests (Google DeepMind)
• Google DeepMind creates a global map of forests with satellite data and AI (The Decoder)
• EU Deforestation Regulation application postponed to 30 December 2026 (EY)
• EU Regulation on Deforestation-free products (Wikipedia)