 The study investigated the potential of hyperspectral remote sensing to estimate above-ground biomass over Brazilian savannas using a Hyperion Earth observing one image and five machine learning models. The results showed that the random forest model with vegetation indices had the lowest root mean square error and was the most stable predictive performer. The five most important ranked vegetation indices were normalized difference vegetation index pigment specific simple ratio enhanced vegetation index red edge normalized difference vegetation index and structure insensitive pigment index. The study set a baseline for AGB modeling of savannas during the transition from current sampling type hyperspectral missions to large coverage hyperspectral satellites. This article was authored by Alain Daniel-Jaykin, Líneo Suarez Galvão, Ricardo Dalliniel, and others.