AbstractThe aim of this thesis was to investigate various methods of mapping land cover types in the savannas of northern Australia by using remotely sensed imageries and GIS tools. The questions were about the type and resolution of imageries as well as the digital image analysis methodology that would best map the land cover types of the region.
Several image classification techniques in combination with various resolution imageries were used to classify land cover types in semi arid savannas at various study sites in the Victoria River District of the Northern Territory of Australia. The efficacy of the data sets and the techniques was tested at several study sites, most notably Kidman Springs Research Station and Mathison Station.
Various combinations of imagery and digital image analysis methods were used at the Kidman Springs site. These were then used to predictively classify land cover types at the Mathison study area. This was done to test the validity of the image analysis techniques used at the Kidman Springs site in a wetter and more densely vegetated area as well as to test the influence of these environmental parameters on the classification results.
It was consistently found that the finer resolution imagery (SPOT XS) combined with an unsupervised classification technique best performed in differentiating the major land cover types in both the study sites. Overall classification accuracy was in the vicinity of 65%-80%. The classification results at the Mathison area was poorer than the Kidman Springs site largely because of the influence of the more rugged and varied terrain at the later region. A GIS analysis technique using a digital elevation model (DEM), however, enhanced the classification accuracy considerably by reducing the ambiguity in the classified imageries.
A method of spectral signature extrapolation was developed from the Kidman Springs data set, and applied as a classification mask to Mathison study site and to a larger surrounding region. This worked reasonably well in mapping the major land cover types of the wider region, although there were some limitations.
As a whole, the problems of land cover classification in the region were not entirely methodological. Rather, the problem of digital image classification for land cover mapping in the savannas of the region lies largely in the definition and delineation of the land cover classes of the region. The classification scheme available for defining and delineating the land cover class boundaries in the region was based on the spatial characteristics of the land cover types. It is hypothesised that a redefinition of the land cover types of the region on the basis of the spectral characteristics of the components of the land covers will produce improved classification accuracy in the region.
|Date of Award||2001|
|Supervisor||Waqar Ahmad (Supervisor)|