Robin Young, Srinivasan Keshav
The study demonstrates that Earth embeddings—compressed representations of annual satellite data—encode specific ecological phenomena such as fire events, deforestation, and crop types. By analyzing two major embedding products, researchers show these vectors can detect abrupt canopy loss within the first growing season after disturbance and distinguish between prescribed and wildfire, with differences in sensitivity to small, fast-recovering fires and crop classification accuracy.
Earth embeddings are increasingly used in environmental monitoring and policy-making, so understanding their internal structure is critical for trust and interpretability. This work enables better use of these models by revealing what ecological signals they capture, improving applications in carbon accounting, biodiversity tracking, and land-use change detection.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.