Cost Effective Approach for Mapping Prosopis Invasion in Arid South Africa Using SPOT-6 Imagery and Two Machine Learning Classifiers

Abstract

This study evaluates the use of SPOT-6 data in conjunction with two machine learning classifiers, namely, Random Forest (RF) and Support Vector Machines (SVM) to map Prosopis glandulosa, its co-existing acacia species and other land-cover types in an arid South African environment. This highly invasive species has been difficult to control using physical, chemical and biological methods because of insufficient knowledge of the species dynamic and lack of spatial data. Results show that it is possible to distinguish Prosopis glandulosa from coexisting Acacia karoo and Acacia mellifera as well as other general land cover types. Classification using SVM obtained a higher overall accuracy of 78.66% (Kappa of 0.7428) whilst RF obtained a lower classification accuracy of 69.93% (Kappa of 0.6331). The high accuracies obtained show the potential to map the invasive species spread on a large scale. This can assist monitoring and planning against future invasions.