We created a new global map of forests and tree crops for 2020 at 10 meter resolution using satellite embeddings. After testing several machine learning methods, we found that simple linear models performed as well as or better than more complex ones. The map identifies forests with 91 % accuracy and separates tree crops with low confusion with forests. This shows that satellite embeddings can support reliable and efficient global forest monitoring and inform international and national policies.
Introducing GEM-Forest, a 10m Global EMbedding-based #forest and tree crop map for 2020 that combines linear SVM and Google DeepMind's AEF satellite #embeddings.
Read the preprint: doi.org/10.5194/egus...
@valeriomarsocci.bsky.social @adamhastie.bsky.social #philab
#GISChat #GeoSky #EOChat
