Termite mounds contribute to the spatial heterogeneity of ecological processes in many savannas, but the underlying patterns and determinants of mound distributions remain poorly understood. Using the Carnegie Airborne Observatory (CAO), we mapped the distribution of termite mounds across a rainfall gradient within a river catchment (similar to 27 000 ha) of the Kruger National Park, South Africa. We assessed how different factors were associated with the distribution and height of termite mounds at three spatial scales: the entire catchment, among three broad vegetation types, and on individual hillslope crests. Abiotic factors such as the underlying geology and mean annual precipitation shaped mound densities at broad scales, while local hillslope morphology strongly influenced mound distribution at finer scales, emphasising the importance of spatial scale when assessing mound densities. Fire return period had no apparent association with mound densities or height. Mound density averaged 0.46 mounds ha(-1), and exhibited a clustered pattern throughout the landscape, occurring at relatively high densities (up to 2 mounds ha(-1)) on crests, which are nutrient-poor elements of the landscape. Mounds exhibited significant over-dispersion (even spacing) at scales below 60 m so that evenly spaced aggregations of termite mounds are embedded within a landscape of varying mound densities. The tallest mounds were found in dry savanna (500 mm yr(-1)) and were positively correlated with mound density, suggesting that dry granitic savannas are ideal habitat for mound-building termites. Mound activity status also varied significantly across the rainfall gradient, with a higher proportion of active (live) mounds in the drier sites. The differential spacing of mounds across landscapes provides essential nutrient hotspots in crest locations, potentially sustaining species that would otherwise not persist. The contribution to biodiversity and ecosystem functioning that mounds provide is not uniform throughout landscapes, but varies considerably with spatial scale and context.