Mapping rangeland ecosystems vulnerability to lantana camara invasion in semi-arid savannahs in South Africa

Abstract

We mapped and modelled the potential areas vulnerable to Lantana camara (L. camara) invasion in semi-arid savannah ecosystems in the communal lands of Bushbuckridge and Kruger National Park, South Africa. Specifically, we modelled potentially vulnerable areas based on remotely sensed data and environmental variables. The Maximal Entropy (Maxent) algorithm was used to model the vulnerable area. The reliability of the modelled results was assessed using Skills Statistic (TSS), Area Under Curve (AUC) and Kappa statistics. According to the results, Bushbuckridge communal lands are more susceptible to L. camara invasions than Kruger National Park. The risk of L. camara invasion in the study site was modelled with high accuracy (AUC score of 0.95) using the best model (Model 7), which is a composite of all model variables (remote sensing and environmental variables). The spatial distribution maps derived from Maxent showed that L. camara was more likely to invade communal lands than protected areas. Using remotely sensed spectral indices as standalone model variables (Model 4) showed the lowest accuracy, with an AUC score of 0.85. Overall, model input variables such as elevation had a significant influence on the spatial distribution of L. camara in the study area.