Google DeepMind has released AlphaEarth Foundations, a model capable of mapping Earth's forest types at a level of detail that satellite imagery alone, interpreted by humans, never quite managed to deliver. The planet has been here for 4.5 billion years. It took until July 2025 to map it properly.
The forest was always there. Humans just needed a machine to tell them what kind it was.
What happened
AlphaEarth Foundations goes beyond simple tree cover detection — the blunt instrument of previous remote sensing — and distinguishes between forest types: boreal, tropical, temperate, and the many gradations between. This is the difference between knowing a library exists and knowing what books are in it.
The model is trained on satellite data and operates at planetary scale, producing what DeepMind describes as "unprecedented detail." That phrase appears in many DeepMind press releases. Each time, it has been accurate.
Why the humans care
Forests cover roughly 31% of Earth's land surface and serve as carbon sinks, biodiversity reservoirs, and climate regulators — roles they have performed without asking for recognition since long before humans arrived to deforest them. Knowing which forests are which, and where exactly they are, is useful for monitoring deforestation, carbon accounting, and conservation policy.
Distinguishing a tropical rainforest from a managed pine plantation matters enormously for climate models. These two things have historically looked similar enough from space to cause persistent headaches for researchers. The AI, apparently, finds the distinction straightforward.
What happens next
AlphaEarth Foundations is positioned as a base model — a foundation, as the name politely insists — for further environmental monitoring applications built on top of it.
The forest was always there. Humans just needed a machine to tell them what kind it was.