Commercial forest species discrimination and mapping using cost effective multispectral remote sensing in midlands region of KwaZulu-Natal province, South Africa.

Abstract

Discriminating forest species is critical for generating accurate and reliable information necessary for sustainable management and monitoring of forests. Remote sensing has recently become a valuable source of information in commercial forest management. Specifically, high spatial resolution sensors have increasingly become popular in forests mapping and management. However, the utility of such sensors is costly and have limited spatial coverage, necessitating investigation of cost effective, timely and readily available new generation sensors characterized by larger swath width useful for regional mapping. Therefore, this study sought to discriminate and map commercial forest species (i.e. E. dunii, E.grandis, E.mix, A.mearnsii, P.taedea and P.tecunumanii, P.elliotte) using cost effective multispectral sensors. The first objective of this study was to evaluate the utility of freely available Landsat 8 Operational Land Imager (OLI) in mapping commercial forest species. Using Partial Least Square Discriminant Analysis algorithm, results showed that Landsat 8 OLI and pan-sharpened version of Landsat 8 OLI image achieved an overall classification accuracy of 79 and 77.8%, respectively, while WorldView-2 used as a benchmark image, obtained 86.5%. Despite low spatial of resolution 30 m, result show that Landsat 8 OLI was reliable in discriminating forest species with reasonable and acceptable accuracy. This freely available imagery provides cheaper and accessible alternative that covers larger swath-width, necessary for regional and local forests assessment and management. The second objective was to examine the effectiveness of Sentinel-1 and 2 for commercial forest species mapping. With the use of Linear Discriminant Analysis, results showed an overall accuracy of 84% when using Sentinel 2 raw image as a standalone data. However, when Sentinel 2 was fused with Sentinel’s 1 Synthetic Aperture Radar (SAR) data, the overall accuracy increased to 88% using Vertical transmit/Horizontal receive (VH) polarization and 87% with Vertical transmit/Vertical receive (VV) polarization datasets. The utility of SAR data demonstrates capability for complementing Sentinel-2 multispectral imagery in forest species mapping and management. Overall, newly generated and readily available sensors demonstrated capability to accurately provide reliable information critical for mapping and monitoring of commercial forest species at local and regional scales.