Discriminating commercial forest species using image texture computed from a WorldView-2 pan-sharpened image and partial least squares discriminant analysis

Abstract

Discriminating forest species is crucial for effective management of commercial forest plantations. Whereas several variants of vegetation indices have commonly been used to delineate forest species, their efficacy is impeded by saturation at high biomass levels. However, image texture does not suffer from saturation at high biomass levels, hence offers a unique opportunity to reliably map commercial forest species typically characterized by dense canopies. In this study, we integrated image texture computed from a 0.5 m WorldView-2 pan-sharpened image with partial least squares discriminant analysis (PLS-DA) to detect and map commercial forest species. The results illustrated that the image texture (overall accuracy (OA) = 77%, kappa = 0.69) outcompeted both the vegetation indices (OA = 69%, kappa = 0.59) and raw spectral bands models (OA = 64%, kappa = 0.52). PLS-DA together with variable importance in the projection selected homogeneity, correlation and mean as the most significant texture parameters, which were predominantly computed from the red-edge, near infrared (NIR) 1 and 2 bands. Furthermore, PLS-DA model commonly selected the 7 × 7 than the 3 × 3 and 5 × 5 moving windows. Overall, this study demonstrates the ability of image texture in discriminating commercial forest species.