Abstract Factors controlling savanna woody vegetation structure vary at multiple spatial and temporal scales, and as a consequence, unraveling their combined effects has proven to be a classic challenge in savanna ecology. We used airborne LiDAR (light detection and ranging) to map three-dimensional woody vegetation structure throughout four savanna watersheds, each contrasting in geologic substrate and climate, in Kruger National Park, South Africa. By comparison of the four watersheds, we found that geologic substrate had a stronger effect than climate in determining watershed-scale differences in vegetation structural properties, including cover, height and crown density. Yet on a given geologic substrate, variable rates of weathering over time result in changing edaphic conditions [ 10 , 11 ]. For example, catenas are a common feature in African savanna landscapes, resulting from differential hydrologic conditions along hillslopes [ 12 ]. Gradients of vegetation structure are thus often found along catenas [ 13 , 14 ]. Within a given substrate and climatic regime, other factors significantly affect vegetation structure. The relative abundance of grass and woody plants in a savanna system results from the differential response of these two growth forms to the environment and to one another [ 15 ]. Remote sensing is currently the best way to collect such a dataset. To this end, Bucini et. al. [ 29 ] combined optical and radar satellite data to gain insight into vegetation cover response to many of the above factors in the Kruger National Park, South Africa One key finding was that inclusion of spatial coordinates into the model more than doubled the variance explained, indicating that much of the variation occurs at landscape scales than is often considered. To more fully understand drivers of savanna vegetation structure, it may be advantageous to look more closely at landscape-scale patterns. Medium- and large-sized tree vegetation was nearly non-existent in the dry (northern) basalt watershed, while at the other extreme nearly 5% of the vegetation cover in the wet (southern) granite watershed was made up of these two tree classes. Additionally, total annual rainfall had minimal effect on the total amount of shrub vegetation, with drier landscapes having 2—\3% more cover than wetter landscapes on the same geology. For all trees ≥2.5 m cover was significantly greater in wetter landscapes than in the paired geological sites of lower rainfall. However, for this mechanism to cause the consistent spatial patterns found here, it would have to be indifferent to the changes in climate, substrate, terrain and species composition represented in our study area. Vegetation height was the only metric for which spatially-aligned trends were tied with substrate and climate. Because height is controlled in a large part by water availability and transport limitations [ 51 ], topo-edaphic factors that affect water availability should exert the most control over vegetation height. In the variograms of vegetation height ( Fig 4 ), it is the granite watersheds that have the shortest range. These five factors, in addition to other factors, may mediate vegetation structure at various spatial scales, showing up in our models as spatially correlated errors. Conclusion We combined LiDAR-derived maps of vegetation and topography with local information of four entire African savanna watersheds to examine how factors influencing vegetation structure work at scales ranging from fine (< 50 m) to geologic (> 100 km). Substrate was clearly more influential than climate, yet fewer clear patterns emerged beyond theseregional controls. While models incorporating topography, hydrology and fire history explained a significant portion of the structural in some watersheds, unknown factors lumped into a single autocovariate term with a range of 50–450 m were often as or more important. This suggests that the strength and range of influential environmental factors differ from site to site. Further study is needed to better understand how multiple environmental factors combine at different scales to determine vegetation structure and how vegetation may respond to future changes in environment. The ACV excelled at removing any remaining spatial autocorrelation from the models. Before adding the ACV term, most remaining spatial correlation was limited to the first 100 to 200 m. Proportion vegetation is defined as the proportion of area in a 16.8 x 16.8 m window of the vegetation height map that is in the given height class: 0.5 to 2.5 m (shrub), 2.5 to 5.0 m (small tree), 5.0 to 10.0 m (medium tree), and >10.0 m (large tree). (TIF) Click here for additional data file. S2 Fig Importance of individual environmental factors to the models. Factor importance, quantified using the difference in R 2 values between the full model and the model without the indicated variable (and its associated interaction terms), for each watershed and response variable. Listed variables are: slope (SLP), aspect (ASP), relative slope position (RSP), distance to order 1,2 and 3 stream (DO1, DO2, and DO3), fire return interval (FRI), fire wait time (FWT), topographic wetness (WET), and soil type (STP, available only on the southern granite site). S2 Fig Importance of individual environmental factors to the models. Factor importance, quantified using the