High spatial resolution and hyperspectral remote sensing for mapping vegetation species in tropical rainforest

Abstract

The focus of this study is on vegetation species mapping using high spatial resolution IKONOS-2 and digital Color Infrared (CIR) Aerial Photos (spatial resolution 4 m for IKONOS-2 and 20 cm for CIR) and Hyperion Hyperspectral data (spectral resolution 10 nm) in Pasoh Forest Reserve, Negeri Sembilan. Spatial and spectral separability in distinguishing vegetation species were investigated prior to vegetation species mapping to provide optimal vegetation species discrimination. A total of 88 selected vegetation species and common timber groups of the dominant family Dipterocarpaceae with diameter at breast height more than 30 cm were used in this study, where trees spectra were collected by both in situ and laboratory measurements of foliar samples. The trees spectra were analysed using first and second order derivative analysis together with scatter matrix plot based on multiobjective optimization algorithm to identify the best separability and sensitive wavelength portions for vegetation species mapping. In high spatial resolution data mapping, both IKONOS-2 and CIR data were classified by supervised classification approach using maximum likelihood and neural network classifiers, while the Hyperion data was classified by spectral angle mapper and linear mixture modeling. Results of this study indicate that only a total of ten common timber group of dominant Dipterocarpaceae genus were able to be recognized at significant divergence. Both high spatial resolution data (IKONOS-2 and CIR) gave very good classification accuracy of more than 83%. The classified hyperspectral data at 30 m spatial resolution gave a classification accuracy of 65%, hence confirming that spatial resolution is more sensitive in identification of tree genus. However, for species mapping, both high spatial and spectral remotely sensed data used are marginally less sensitive than at genus level.