The ability to capture 3D point clouds from LiDAR sensors and the advancement in algorithms has enabled the explicit analysis of vegetation architecture, branching characteristics and crown structure for the accurate estimation of Above Ground Biomass (AGB). The ability to have geometrically accurate 3D volume of vegetation reduce the uncertainty in AGB estimation without destructive sampling through the application of volume reconstructions algorithms on high-resolution point clouds from Terrestrial Laser Scanning (TLS). These methods, however, have been developed and tested on temperate and boreal vegetation with very little emphasis on the savanna vegetation. Here, we test the reconstruction algorithms for the estimation of AGB in a savanna ecosystem characterised by a dense shrub understory and irregular multi-stemmed trees. Leaf off multi scan TLS point clouds were acquired during the dry season in 2015 around the Skukuza flux tower in Kruger National Park, South Africa. From the multi scan TLS point clouds, we extracted individual tree and shrub point clouds. Tree Quantitative Structure Models (TreeQSMs) were used to reconstruct tree woody volume whilst voxel approaches were used to reconstruct shrub volume. The AGB was estimated using the derived woody volume and wood specific gravity. To validate our method, we compared the TLS derived AGB with allometric equations. TreeQSMs predicted AGB with a high concordance correlation coefficient (CCC) compared to the allometry reference, although tree crown biomass was overestimated especially for the large trees. The biomass of the shrub understory was described with reasonable accuracy using the voxel approach. These findings indicate that the application of 3D reconstruction algorithms improve the estimation of savanna vegetation AGB as compared to allometry references and combined tree and shrub woody biomass estimates of the savanna allow for calibration and validation for accurate monitoring and mapping at large spatial scales.