ForTy maps forest types beyond tree cover

Google’s October 22 announcement introduces ForTy 2020, distinguishing natural and planted forests, plantations, tree crops and other land. AlphaEarth’s compact satellite summaries help the model identify planting patterns and harvest cycles. An early annual series covering 2017–2025 also tracks what replaces lost forest. This matters for conservation because land can become green again after natural forest is replaced by a plantation.

The ForTy catalog describes a 2020 baseline, rather than a map of today’s conditions. Users can assign the highest-scoring class or choose their own probability thresholds. Those scores have not received subsequent calibration; the documentation suggests considering regional recalibration with independent reference data for quantitative risk assessment. Land-management history is not directly visible from satellites. For example, cocoa or coffee growing beneath native canopy can still be mistaken for natural forest in some regions.

The satellite-embedding documentation explains this type of AI input. A learned representation packages observations from multiple instruments across one calendar year into 64 values for each ten-meter pixel. These values summarize surface conditions and their changes during the year. They are not separate physical measurements, such as individual sensor readings: the documentation says to use all components together for analysis, because they cannot be interpreted independently.

Jonathan Kemper’s October 31 report links the Google DeepMind announcement and the research paper as its sources.