The development of remote sensing techniques for green vegetation cover discrimination in sub-tropical, semi-arid and arid zones of Northern Territory

  • Wilma Matheson

    Student thesis: Masters by Research - CDU


    An attempt was made to establish procedures whereby green vegetation cover and where possible species composition were identified using Landsat Multispectral Scanner (MSS) imagery over three regions in Northern Territory. To provide a comparison of vegetation spectral response under different climatic conditions three Study Regions, each 1120 km2 were chosen to represent sub-tropical, semi- arid and arid zones . Digital data were acquired on 2 .5.91 for King River (subtropical), on 29.12.90 for Barkly (semi-arid) and on 29 .12.90 for Simpson (arid). Field work took place at all locations co-incidentally with the satellite overpass. Twenty field sites, each 25600 m2 were chosen in each region. Percentage green vegetation cover, live and dead herbaceous cover , dead wood and litter were obtained from each site. Antecedent rainfall and time of year constraints were such that the subtropical region involved the spectra l separation of green vegetation cover from dead herbaceous cover, soils and plant litter components. Background components were mainly soils in the semi- arid and arid zones. Green vegetation cover ranged from 31-39% in King River, from 23-28% in Barkly and from 3-9% in Simpson. King River region had the highest plant cover (0.21) and species richness (0.54) indices per unit area with Eucalyptus spp. and Erythrophleum chlorostachys being the most prevalent species. The Barkly region had intermediate plant count cover (0.13) and species richness (0.44) indices with Acacia spp. and Ventilago viminalis as the most prevalent species. Simpson region had the lowest plant count (0.11) and species richness (0.17) indices with Acacia spp. and Sclerolaena spp. most prevalent. 

    Hand held radiometer results showed similar vegetation reflectance characteristics for each Study Region. In all regions little direct association was found between single band values (and band transforms) and the proportion of green vegetation cover. The ability of MSS data transforms to predict green vegetation cover decreased as vegetation cover decreased. The measurement of incremental vegetation cover over different background components (combined spectra) resulted in the grouping of vegetation cover types into Type 1 darkening, Type 2 partial darkening and Type 3 highly reflective NIR vegetation groups. The latter were most prevalent in King River over soils and senesced vegetation backgrounds. Darkening and partial darkening types were prevalent in all study Regions but mostly in the arid regions. Throughout all regions specific reflectance relationships were soil type and species type dependent. 

    Spectral plots of digital data over field sites enhanced characteristics inferred from radiometer results. The relatively low visible and high near infrared (NIR) reflectance of King River resulted from the higher proportion of green vegetation cover. Curves for Barkly and Simpson showed a closer approximation to soil curves with darkening evident in the NIR, especially in Barkly. Two- dimensional plots showed that key darkening species from monospecific stands could be identified. These were Acacia shirleyi in King River, Acacia georginae in Barkly and Acacia aneura in Simpson. Green vegetation cover was separable at high NIR reflectance values when the total green cover exceeded 45%. Below 45% cover darkening species were evident. A series of band transforms and Principal Components Analyses (PCA) were run on the three regional data sets prior to channel selection for classification. In King River a file containing twenty four channels was created containing single bands, transforms and PCA results. These visually showed good green vegetation separation. The four channels finally selected for classification input showed highest correlations using analysis of variance of regression. Fifteen channels were visually chosen in Barkly and five were statistically selected for classification input. All channels (37) were chosen in Simpson due to problems with visual separation of processed data. However, dark pixel corrected inputs were used for final classification of Simpson due to problems in channel selection. Classification algorithms were run on all selected data sets. Maximum likelihood classifiers produced the most spatially acceptable green vegetation cover data compared to aerial photograph and random field check data. 

    The classified maps showed nine vegetation classes in King River including mainly NIR reflective but also darkening classes. Nine vegetation classes were also identified in Barkly, including mainly darkening but also NIR reflective classes. Four vegetation classes were mapped. While species were scattered throughout the classes specific darkening and NIR reflective species were identified in each class area. High classification accuracies were ascribed mainly to the statistical channel selection techniques applied prior to classification.
    Date of AwardDec 1991
    Original languageEnglish
    SupervisorSusan Ringrose (Supervisor), Gordon Duff (Supervisor) & James Mitroy (Supervisor)

    Cite this

    The development of remote sensing techniques for green vegetation cover discrimination in sub-tropical, semi-arid and arid zones of Northern Territory
    Matheson, W. (Author). Dec 1991

    Student thesis: Masters by Research - CDU