AlphaEarth Foundations published a study in Science in July 2025 that presents a high‑resolution map of forest types worldwide. By integrating satellite remote sensing, ground surveys, and machine‑learning models, the team refined the traditional canopy‑cover data into more than a dozen forest ecosystems, such as tropical rainforests, temperate broadleaf forests, coniferous forests, and swamp forests. The core workflow is as follows:
- Pre‑process multi‑spectral Sentinel‑2 and Landsat 8 imagery, remove clouds, and resample to a uniform 10 m resolution.
- Build a random‑forest classifier using inputs like spectral indices (NDVI, EVI), texture features, and terrain variables (elevation, slope).
- Evaluate model performance with cross‑validation; overall accuracy reaches 92% and recall for tropical rainforests is 95%.
- Align the classification results with the Global Forest Resources Assessment (GFRA) and publish an interactive Web GIS platform for researchers and policymakers.
This work moves beyond the limitation of measuring forests solely by canopy cover, providing a finer spatial baseline for carbon‑sequestration assessment, biodiversity conservation, and sustainable forestry.
Review