Machine learning algorithms for mapping <i>Prosopis glandulosa</i> and land cover change using multi-temporal Landsat products: a case study of Prieska in the Northern Cape Province, South Africa

Abstract

Invasive alien plants (IAPs) are responsible for loss in biodiversity and the depletion of water resources in natural ecosystems. Prosopis species are IAPs previously introduced by farmers to provide shade and fodder for livestock. In the Northern Cape, Prosopis spp. invasions are associated with the loss of native species resulting in overgrazing and degrading rangelands. Mapping Prosopis glandulosa is essential for management initiatives to assist the government in minimising the spread and impact of IAPs. This study aims to evaluate the performance of two machine learning algorithms i.e., Support Vector Machine (SVM) and Random Forest (RF) to map the spatial dynamics of P. glandulosa in Prieska. The spatial invasion extent of P. glandulosa was mapped using multitemporal Landsat data spanning the period from 1990 to 2018. Validation of the results was done through an estimated error matrix with the use of the proportion of area and the estimates of overall accuracy, user’s accuracy and producer’s accuracy with a 95% confidence interval. The performance of the SVM and RF classifiers showed similar results in the overall accuracy and Kappa statistics throughout the years. These methods showed an overall increase of at least 3.3% of the area invaded by P. glandulosa from 1990 to 2018. The study indicates the importance of Landsat imagery for mapping historical and current land cover change of IAPs. The spread of P. glandulosa was confirmed by an increase in the total area of invasion, which enables decision-makers to improve monitoring and eradication initiatives.