Detecting and mapping invasive Parthenium hysterophorus L. along the northern coastal belt of KwaZulu-Natal, South Africa using image texture

Abstract

Parthenium hysterophorus L. (parthenium) is a devastating weed that is spreading rapidly throughout three continents, including Africa. Traditionally, information on plant distribution was collected using spatially restrictive, time-consuming and expensive methods. Remote sensing has revolutionized the collection of landscape feature data in recent years, making plant distribution studies more efficient. This study investigates the use of freely accessible SPOT-6 and Sentinenl-2 imagery to map parthenium and associated land cover. Specifically, it utilized texture analysis to compare the mapping capability of a 1.5 m SPOT-6 panchromatic band, 6 m multispectral SPOT-6 image and 10 m Sentinel-2 image. The Partial Least Squares-Discriminant Analysis (PLS-DA) was used to classify the images and the variable importance in the projection (VIP) score was used to determine the significant predictor variables. Although all images adequately detected parthenium and surrounding land cover classes, the panchromatic band achieved the highest user's and producer's accuracies for parthenium and a higher overall classification accuracy (77%). The most significant texture parameters computed by the SPOT-6 panchromatic band and selected by VIP were mean, correlation and homogeneity. Overall, this study shows the potential of image texture integrated with PLS-DA to effectively detect and map parthenium and surrounding land cover classes.