Landscape fires are substantial sources of (greenhouse) gases and aerosols. Fires in savanna landscapes represent more than half of global fire carbon emissions. Quantifying emissions from fires relies on accurate burned area, fuel load and burning efficiency data. Of these, fuel load remains the source of the largest uncertainty. In this study, we used high spatial resolution images from an Unmanned Aircraft System (UAS) mounted multispectral camera, in combination with meteorological data from the ERA-5 land dataset, to model instantaneous pre-fire above-ground biomass. We constrained our model with ground measurements taken in two locations in savanna-dominated regions in Southern Africa, one low-rainfall region (660 mm year−1) in the North-West District (Ngamiland), Botswana, and one high-rainfall region (940 mm year−1) in Niassa Province (northern Mozambique). We found that for fine surface fuel classes (live grass and dead plant litter), the model was able to reproduce measured Above-Ground Biomass (AGB) (R2 of 0.91 and 0.77 for live grass and total fine fuel, respectively) across both low and high rainfall areas. The model was less successful in representing other classes, e.g., woody debris, but in the regions considered, these are less relevant to biomass burning and make smaller contributions to total AGB.