Increasing attention is being directed at mapping the fractional woody cover of savannahs using Earth-observation data. In this study, we test the utility of Landsat TM/ ETM-based spectral-temporal variability metrics for mapping regional-scale woody cover in the Limpopo Province of South Africa, for 2010. We employ a machine learning framework to compare the accuracies of Random Forest models derived using metrics calculated from different seasons. We compare these results to those from fused Landsat-PALSAR data to establish if seasonal metrics can compensate for structural information from the PALSAR signal. Furthermore, we test the applicability of a statistical variable selection method, the recursive feature elimination (RFE), in the automation of the model building process in order to reduce model complexity and processing time. All of our tests were repeated at four scales (30, 60, 90, and 120 m-pixels) to investigate the role of spatial resolution on modelled accuracies. Our results show that multi-seasonal composites combining imagery from both the dry and wet seasons produced the highest accuracies (R2 = 0.77, RMSE = 9.4, at the 120 m scale). When using a single season of observations, dry season imagery performed best (R2 = 0.74, RMSE = 9.9, at the 120 m resolution). Combining Landsat and radar imagery was only marginally beneficial, offering a mean relative improvement of 1% in accuracy at the 120 m scale. However, this improvement was concentrated in areas with lower densities of woody coverage (<30%), which are areas of concern for environmental monitoring. At finer spatial resolutions, the inclusion of SAR data actually reduced accuracies. Overall, the RFE was able to produce the most accurate model (R2 = 0.8, RMSE = 8.9, at the 120 m pixel scale). For mapping savannah woody cover at the 30 m pixel scale, we suggest that monitoring methodologies continue to exploit the Landsat archive, but should aim to use multi-seasonal derived information. When the coarser 120 m pixel scale is adequate, integration of Landsat and SAR data should be considered, especially in areas with lower woody cover densities. The use of multiple seasonal compositing periods offers promise for large-area mapping of savannahs, even in regions with a limited historical Landsat coverage.