Modelling lantana camara invasion in the inkomati catchment in Mpumalanga, South Africa

Abstract

Lantana Camara ranks among the most notorious and hazardous invasive species on the planet. There has been extensive research focusing on the remote sensing of Lantana Camara (L. Camara). However, the factors that facilitate its spread, especially in savanna rangelands, are not yet fully comprehended. The present research investigated the recent spatial distribution of L. Camara in the Inkomati catchment, Mpumalanga, South Africa. The research used the Random Forest classification algorithm, with Sentinel-2 remotely sensed data within Google Earth Engine. The study also modelled areas vulnerable to L. Camara invasion using the Maxent species distribution model. The study found that L. Camara covered approximately 34.86% of the entire study area, with a user's and a producer's accuracies of 91% and 84%, respectively. Additionally, elevation was identified as the most influential topographic variable on the species spatial distribution and invasion, while Topographic Wetness Index had the least influence. The model developed using topographic variables had the highest accuracy of 0.88. On the other hand, the predictive model that utilised Sentinel 2 bands had the least accuracy of 0.81. The red edge and NIR bands with Random Forest allowed for an accurate assessment of L. Camara distribution. The study therefore provides the basis for the control or further expansion of L. Camara species in the province.