Modeling and explaining the distribution of Lantana camara in South African savanna ecosystems.

Abstract

Globally, the Invasive Alien Plant (IAPs) species pose a great threat to global biodiversity, agro-ecological systems and socio-economic development. In particular, Lantana camara (L. camara) is amongst the most notorious and problematic of all invasive plants globally. Its threats and effects are undeniably recognizable and it is ranked amongst the world’s ten worst weeds. As a result, it is one of the most documented weeds in the world. Most studies have focused mainly on detecting and mapping the spatial distribution of L. camara. Although its spatial distribution remains rudimentary, the mechanisms driving its distribution are not yet fully understood, especially in savanna rangelands. This study aimed at modelling and explaining the distribution of L. camara in South African savanna ecosystems (the Kruger National Park and Bushbuckridge communal lands). Specifically, the study sought to identify the environmental factors influencing the spatial distribution of L. camara in savanna ecosystems using the Maximum Entropy (Maxent) algorithm, coupled with remotely-sensed derivatives from Sentinel-2 satellite data. The performance of the model was assessed by using the Area Under Curve (AUC), the True Skills Statistic (TSS) and the Kappa Statistic. From the findings, the Bushbuckridge communal lands had the highest L. camara infestations, with the weed covering an area of 10%, when compared to the Kruger National Park, which had an estimated coverage of 7%. The derived spatial distribution maps from Maxent revealed that communal lands of Bushbuckridge are more vulnerable to L. camara invasion than the protected area. The study also demonstrates that bioclimatic factors influence the occurrence, spread and infestation of this invasive weed species. Comparatively-speaking, elevation was found to have the greatest influence on the infestation and spatial distribution of L. camara. The model that was derived from a composite of all the variables yielded the highest AUC score of 0.96. Subsequently, the model based on indices alone (Model 4) achieved the lowest accuracies, with an AUC score of 0.85. This work is critical for providing the necessary information to assist in effective management and clearing practices by informing the strategic planning, control and rehabilitation of the affected areas.