A new free map from Google DeepMind and Google Research, built with the World Resources Institute and IIASA, sorts all land into six types: primary forest, forest that grew back on its own, planted forest, plantation, tree crops such as oil palm, rubber and cocoa, and everything else. The 2020 map is the baseline. An early yearly series from 2017 to 2025 shows how the land changed, including natural forest replaced by plantations or farms.
A green patch can hide a swap
From orbit, an old rainforest, a eucalyptus plantation and an oil palm estate are all green. A tree-cover map asks one question, which is whether trees are standing, and for all three the answer is yes. When a natural forest is logged and replanted as a plantation, the answer goes to no for a few years and then back to yes. The loss is recorded, but the map cannot say what the land became, and the replanting can hide it entirely.
The new map tells them apart by how the land behaves over time. Plantations are planted in patterns and harvested on a rhythm, and natural forest has neither. The model reads years of satellite records for each 10-metre square, and it learned the differences from about 1.7 million labelled locations. Checked against more than 9,000 random sites, it named the right forest type 90% of the time. Last year's Natural Forests of the World map could only say whether a patch was natural forest in 2020. This one says what it was, and what it turned into later.
What a buyer can do with it
Take a chocolate maker in Europe that buys cocoa from a cooperative with 400 farms. The law asks for the coordinates of every plot and proof that none of them was forest in 2020 and cleared since. Coordinates only say where the farm is. Until now, finding out what the land used to be meant paying an analyst, or relying on a tree-cover map that cannot separate a cocoa farm under shade trees from a patch of natural forest.
With this map, someone who can use Google Earth Engine can check all 400 plots in an afternoon. Plots that were farmland or tree crops in 2020 and still are can go through. A plot that was primary or regrown forest in 2020 and is a tree crop now gets stopped, and someone asks the cooperative what happened. The map gives each class a probability, so the buyer chooses how strict to be. A cautious importer can accept only the plots the model is very sure about.
Where it will bite
A map that is right nine times in ten is wrong about one patch in ten. That makes it a way to decide which plots need questions, not a verdict. Some clean plots will be flagged, and some problem plots will pass, so the farmer who is wrongly flagged will need photos and records to answer. Suppliers will want to look at their own plots on the map before their buyers do. The yearly series is an early release and will keep changing.
The law is due to apply to large companies from 30 December 2026, so the first real arguments will be about thresholds, meaning how sure a buyer must be, not about whether anyone can see the land. The same map works for forest finance. A project that is paid to plant trees can be checked for whether it planted on open land or on natural forest it replaced.
Once the buyer, the supplier, the auditor and the regulator can all open the same free map, nobody can say they did not know what a green patch used to be.