Mapping key browse resources in a heterogeneous agricultural landscape

Abstract

We investigated application of MaxEnt, a one-class classifier, in mapping the spatial distribution of Colophospermum mopane, Dichrostachys cinerea and Salvadora persica using drainage, elevation, slope, soil and Normalised Difference Vegetation Index as environmental variables. Model performance was evaluated based on the area under the ROC curve (AUC), Kappa and Total Skills Statistic. The AUC results demonstrated the high predictive power of MaxEnt as test data values of all species, respectively, were 0.694, 0.754 and 0.998 (p < 0.05). Elevation contributed the most in explaining spatial distributions of all species. Results also indicated that several one-class species maps can be integrated into one species distribution map. We showed that C. mopane is likely to co-occur with D. cinerea and S. persica, whereas S. persica is likely to co-occur with D. cinerea. However, there was no habitat suitable for co-occurrence of all species. One-class species mapping can therefore be successful in heterogeneous agricultural landscapes.