Investigating the potential of a classification algorithm to identify black wattle (Acacia mearnsii De Wild.) tress using imaging spectroscopy.
In South Africa, invasive black wattle trees (Acacia mearnsii D. Wild) are a major threat to ecosystem functionality causing widespread social, economic and environmental degradation. It is important that environmental managers are provided with rapid, regular and accurate information on the location of invasive black wattle trees to coordinate removal efforts. This study investigated the potential of an automated image classification algorithm to accurately identify black wattle (A. mearnsii De Wild.) trees using imaging spectroscopy. Hyperspectral data acquired by the EO-1 Hyperion sensor was used to identify black wattle trees in two study areas near Greytown, KwaZulu-Natal, South Africa. Image classifications were performed by the classification algorithm to identify black wattle trees using general and age specific spectral signatures (three to five years, seven to nine years, eleven to thirteen years). Results showed that using the general spectral signature an overall accuracy of 86.25% (user’s accuracy: 72.50%) and 84.50% (user’s accuracy: 69%) was achieved for study area one and study area two respectively. Using age specific spectral signatures, black wattle trees between three to five years of age were mapped with an overall accuracy of 62% (user’s accuracy: 24%) and 74.50% (user’s accuracy: 49%) for study area one and study area two respectively. The low user’s accuracies for the age specific classifications could be attributed to the use of relatively low resolution satellite imagery and not the efficacy of the classification algorithm. It was concluded that the classification algorithm could be used to identify black wattle trees using imaging spectroscopy with a high degree of accuracy.