A ER a SR 
band 7 value. With increasing vegetation canopy cover the soil 
surface is masked and a high reflectance from the dry soil is 
reduced. Reflectance from vegetation with a canopy cover of 100 
percent at point A on Figure 2, and a radiance value of 27 in 
band 5 and 25 in band 7, is lower than reflectance from 100 
percent bare soil at point B, but is a significant component of a 
combined soil and vegetation spectral response. The intensity of 
this response varies with the varying proportions of soil and 
vegetation within a LANDSAT pixel of 56m x 79m, such that a high 
vegetation canopy cover gives a relatively low reflectance, and a 
high proportion of soil produces a high reflectance. 
Observations both on the ground and from the air, have 
suggested that although the spectral response of the soil is 
generally uniform, the vegetation with its own distinct spectral 
response and cover of more than one species, is unlikely to give 
a similarly uniform reflectance. The triangular shape of the data 
cluster enclosed by the points marked ABC on Figure 2, suggests 
that the influence of variations in species composition produces 
a change in the spectral content of ihe vegetation response. 
Further statistical evidence will be required to show that the 
observed spread of points is significant and is the result of 
real variations in the species composition of the vegetation 
cover and not scatter due to other factors. If this can be shown 
to be true then the data might be more accurately described by a 
three component system represented by a bare soil point at B, and 
two points A and B each representing 100 percent cover for two 
species of vegetation, in this case with dwarf shrub at C and 
shrub at A. It should then be possible to divide the feature 
space of this three component system into a number of classes 
and determine the statistical probability of a particular class 
having a radiance value representing the combined reflectance 
from three variables; soil, vegetation canopy cover, and species 
composition. 
4 .DISCUSSION 
The development of a digital image processing technique 
which successfully classifies the vegetation cover, depends to a 
large extent upon gaining a quantititative knowledge of the 
relationship between vegetation and radiance values. Supervised 
multispectral classifiers rely upon defining homogeneous 
training areas with a distinct spectral response to classify an 
image. In this area of semi-arid vegetation it has been shown 
that a weaker response from the vegetation cover modifies a 
dominant reflectance from the soil to give a combined response 
from both soil and vegetation. This results in a high correlation 
between bands and the absence of distinct vegetation clusters in 
the feature space. Preliminary results from supervised 
multispectral classification techniques have produced 
misclassification and it is clear that in this area other image 
processing techniques will have to be employed. From the data it 
appears that a density-slice of a band 5 and band 7 composite 
would take into account both variations in canopy cover and in 
species composition, in order to classify the image. 
384 
554 
504 
radiance in MSS 7 
40- 
304 
204 
10 -