Binary classification experiments to interpret Google AlphaEarth Foundation embeddings across ESA WorldCover land cover classes. Part of the study "What on Earth is AlphaEarth?" — 130,000+ experiments using Random Forest, XGBoost, LightGBM and progressive ablation across 64 embedding dimensions.
machine-learning embeddings remote-sensing earth-observation interpretability explainable-ai land-cover-classification foundation-models geospatial-ai alphaearth esa-worldcover gogle-earth-engine
-
Updated
Mar 7, 2026 - Jupyter Notebook