Mapping and modelling spatial distribution of prosopis juliflora invasion in the Northern Cape Province of South Africa

Abstract

The negative effects of Prosopis on the ecological infrastructure and economies of invaded areas is well documented. One of the factors hindering invasion management of eradication is lack of reliable spatial data. Policy makers, ecologists and biodiversity planners are seeking efficient and cost friendly methods of evaluating spatial dynamics such as invasion rate and hot spots. Inventory methods such as field surveys are expensive and labour-intensive thus unsustainable while Remote Sensing (RS) and Geographic Information Systems (GIS) methods are efficient and cost effective. Moreover, field surveys are often limited to small geographic extents while RS and GIS techniques have the ability to analyse vast geographic extents. The aim of this study was to evaluate potential of mapping and modelling Prosopis Juliflora using Sentinel – 2A and Species Distribution Modelling (SDM) in arid regions of the Northern Cape Province of South Africa. It was achieved by three specific objectives; using Random Forest (RF) and Support Vector Machine (SVM) classifiers to map Prosopis Juliflora and other land cover; evaluation of the distribution of Prosopis Juliflora using spatial point pattern analysis; predicting areas susceptible to invasion using binomial logistic regression. Results obtained from the first objective illustrated the potential of 10 m resolution Sentinel – 2A in classifying Prosopis Juliflora from other land cover, RF and SVM reported overall accuracies of 88.86 % and 91.42 % respectively. Spatial point pattern analysis illustrated that Prosopis Juliflora exhibited clustering which was evaluated with respect to various geophysical factors such as geology, soil, elevation and slope. The binomial regression successfully predicted areas that are susceptible to invasion. The final model had a pseudo R2 of 0.78 and Akaike Information Criterion (AIC) of 223.35.