This study assesses the utility of modelling approaches to predict vegetation distribution in agricultural landscapes of southwestern Australia. Climate surfaces, hydrologic and erosion process models are used to link vegetation to environmental variables. Generalized additive models (GAM) are derived for presence/absence data of mapped vegetation types. Vegetation distribution shows significant responses to rainfall and subsequent water redistribution due to the relief; however, these variables are insufficient to effectively explain vegetation patterns at the local scale. Accordingly, prediction accuracy remains low (k‐values below 0.5). The striking unpredictability of the local distribution of the vegetation in the Wheatbelt is discussed with regard to the performance of topographically driven processes in subdued landscapes and with regard to geological, historical and biological factors determining the southwestern Australian plant species distribution.

Nomenclature: Chapman et al. (in prep.).

Abbreviations: AVHGT = average altitudinal height of the upslope area; AVSLP = average slope of the upslope area; CURVPL = plan curvature; CUPL500 = plan curvature in a 1.5 km window; CURVPR = profile curvature; DEM = Digital Elevation Model; DIRIDGE = distance from ridges; DROUGHT = drought index; EROS = sediment transport index; GAM = Generalized Additive Model; HABOVE = height above streamlines; RAIN = annual rainfall; SLOPE = maximum slope of the surface; STRP = stream power index; WET = wetness index.