A multiscale remote sensing method to measure aboveground woody biomass in savanna ecosystems

Background Woody aboveground biomass (AGB) stores and releases carbon in savannas, with fire as a key driver. Aims Savanna fire management (SFM) programs reduce emissions from AGB burning but do not incorporate live tree carbon sequestration. Assessing the impact of SFM on woody AGB carbon sequestration requires precise measurement and modelling. Methods We developed a multiscale remote sensing method for woody AGB estimation and applied it across ~105,000 ha of tropical savanna. A novel metric (shade volume) bridged the gap between terrestrial lidar-derived woody AGB and a convolutional neural network (CNN) model trained on airborne lidar and satellite imagery. Using the method, we estimated savanna woody AGB and quantified AGB prediction error. Key results CNN-predicted shade volume had 5.5% mean absolute error and −2.1% bias. Validation against independent 1 ha woody AGB measurements (n = 7) showed 7.9% mean error. In 40.1% of the study region, woody AGB predictions exceeded maximum potential biomass estimated by Australia’s national carbon accounting model. Conclusions This methodology improves carbon estimation accuracy over large areas, enabling fine-scale monitoring of woody AGB under varied SFM strategies. Implications Enhancing SFM carbon credit integrity requires direct measurement and transparency in woody AGB quantification, both achievable with this method.