Results and Discussion
The complete fire boundary was not clearly defined on either the
enhanced imagery or on the digital classifications since areas of slightly
damaged forest were indistinguishable from unburnt forest. This parallels
the experience of the photointerpreters in preparing the map from the
colour photography where the same intensity of fire was not clearly re-
corded on the photographs. Myers (1978) maintains that much larger scales
are necessary to identify those areas in a prescribed burn which have
experienced a low intensity fire. Much of the slightly damaged forest at
Mt Buffalo was burnt by low intensity fires resulting from backburning
operations during suppression activities. The plotting of the fire
boundary shown on the photointerpreted map required additional information
by way of ground checking and details supplied by fire fighting personnel.
It was possible by the digital approach to define ten classes. The
signatures for the unnormalised data are set out in Table 1 and illustrated
in Figure l. There were five classes of forest cover (unburnt), two
classes of leaf scorch, one for areas with the crowns removed, one for
cleared agricultural land and one for water bodies.
On the enhanced visual image the boundaries of the severely and
moderately damaged areas were improved in clarity particularly on the band
7 image. A colour composite has not yet been prepared and an accuracy
evaluation has yet to be performed quantitatively.
Leaving aside the unburnt forest classes, which also include the
slightly damaged type, the patterns of the moderately and severely damaged
classes on the computer classification line printer output appear to
visually match the patterns on the photointerpreted map reasonably well.
When all classes were assigned specific colours and were displayed on the
COMTAL colour image display device, the screen photographed on 35 mm colour
film and the resulting slide projected onto the photointerpreted map, the
match of the overall patterns was assessed as satisfactory. The normalised
version of the classification, in which the effect of variation in il-
lumination with slope and aspect was minimised, was judged to give a cleaner
result and provided a better visual match with the photointerpreted map.
Figure 2 illustrates the geometrically corrected classification as
graphically displayed on the III COMp 80 device. For the sake of clarity,
grouping of the unburnt forest classes into one class and also the two
moderately damaged classes into one has been carried out in producing this
result.
Efforts devoted to digitally overlaying the classification and the
photointerpreted map have been unsuccessful to date mainly because of the
difficulty in identifying adequate control points to link the two digital
images. The nomination of control points was attempted on the line printer
output however when compared for position after warping functions were
performed the resultant mean square deviations were too great (some as high
as 7 pixels in both the x and y directions).
These difficulties have rendered it impossible at this stage to provide
a statement of the accuracy of the classification in the accepted confusion
table manner in which percentage of agreement and errors of omission and
commission are portrayed.