Discrimination of blackberry ( Rubus fruticosus sp. agg.) using hyperspectral imagery in Kosciuszko National Park,NSW, Australia

Abstract

Blackberry, Rubus fruticosus sp. agg., is a perennial, semi-deciduous shrub which forms dense thickets that infest approximately 8.8 million ha of land in Australia. It is highly invasive with a high potential for spread, and causes significant negative economic and environmental impacts. During 2004, HyMap hyperspectral imagery was acquired across the foreshores of Blowering Dam in Kosciuszko National Park, NSW, Australia to evaluate its utility for mapping the distribution of blackberry. Strategies for mapping image-derived blackberry spectra using hyperspectral imagery were assessed using Spectral Angle Mapper, Spectral Feature Fitting, Matched Filter, and Mixture-Tuned Matched Filter mapping algorithms. A Mixture-Tuned Matched Filter (MTMF) approach using the blackberry spectrum with a restricted wavelength range was adopted. MTMF distribution maps showed the highest agreement to the distribution of blackberry as assessed using an error matrix and independent groundtruth data. The MTMF distribution map generated a producer's accuracy of 91%, user's accuracy of 81%, overall accuracy of 92% and a Kappa coefficient of 0.715. This study has demonstrated that hyperspectral imagery can effectively quantify the distribution of blackberry in open canopies.