Soil data are largely absent for most of Africa. For landscapes with recognizable catenary elements, this data gap can be filled by mapping the catenary units and assigning them with known soil properties. An example is the landscape map for a region with dambos in central Uganda, which shows the four catenary units in order from well-drained to seasonal wetland: uplands, margins, floors, and bottoms. However, this map was created using optical data, which are cost prohibitive and are also limited by cloud cover. We evaluated the potential of freely available aerial gamma-ray spectrometry (AGRS) data as an alternative source of classification inputs. Analysis of variance based upon field data for a region with dambos in central Uganda showed gamma activity to differ along the catenary sequence, with landscape position explaining an appreciable proportion of variation of potassium (28%), thorium (27%), and uranium (46%). Using the three gamma channels, together with terrain indices derived from the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) as inputs, three classifiers were evaluated – conditional inference trees (CITs), random forests (RF), and multinomial-iterative self-organizing data analysis (ISODATA). While untransformed terrain and gamma predictors were used for the first two methods, we applied the ISODATA classification to landscape unit probability maps generated using multinomial principal components regression. For the CIT classification, all decision rules were based on terrain data, which might explain why the map was slightly less accurate (unweighted kappa = 0.61, linear weighted kappa = 0.73) than the map created using a RF classifier (unweighted kappa = 0.63, linear weighted kappa = 0.74), where both terrain and gamma predictors were used. But the existence of artefacts of margins within uplands in the map based on CIT modelling, and not that created using RF, is because the former missed the smoothing effect of gamma, attributed to zonal differences in activity of all three gamma channels. The multinomial-ISODATA predictions were poor (unweighted kappa = 0.56, linear weighted kappa = 0.67), partly because the regression model could not adequately resolve differences between bottoms and floors. However, we did find the probability maps generated using multinomial regression to be useful end products that capture the continuous nature of landscape unit transitions. It is important to note that in this study we used 90 m grid resolution gamma and terrain data to predict features that transition over distances of less than 10 m, so better results might be possible with finer-resolution gamma and terrain data.