Discrimination of arid vegetation with airborne multispectral scanner hyperspectral imagery

Abstract

Hyperspectral imagery from the airborne multispectral scanner was evaluated for discrimination and mapping of vegetation components in a semi-arid rangeland environment in Southern Australia. Automated unmixing of two image strips with 5-m resolution revealed several vegetation endmembers in the visible, near infrared (near-IR), and short-wave infrared portions of the imagery. Identity of the endmembers was determined through examination of their short-wave infrared and full-wavelength spectra, and their mapped distributions and correlation with percent cover of vegetation species were measured in sample plots. In addition, to assist interpretation of the image signatures, short-wave infrared reflectance spectra for the dominant vegetation components at the study site were collected with a portable infrared mineral analyzer (PIMA) spectrometer. Endmembers separately mapped included Eucalyptus and other trees such as sugarwood, understorey chenopod shrubs, dry plant litter, and soil surface cryptogamic crust. Several endmembers were significantly positively correlated with field measurements of plant cover. Most of the tree canopy endmembers showed broad cellulose–lignin absorption features in the short-wave infrared (SWIR), and narrower absorptions caused by plant waxes and oils. The field spectra confirmed that the sclerophyll and xerophytic plants show identifiable cellulose, lignin, and plant wax absorption features, even when live and actively photosynthesising. This spectral expression of biochemical constituents in live plants points to the benefit of using the whole spectral range from visible to short-wave infrared in vegetation studies.