433
200
250
nd 7
200
250
and 8
uce and black
en healthy and
dependent on
is of green,
aduced varying
Figure 5. Scattergram outlines of MEIS-II bands 5 and 7
for five classes at 10 m resolution.
results. The unsupervised maximum likelihood classi
fication gave very poor results and therefore cannot
be used to classify budworm infestation. The super
vised maximum likelihood classification gave slightly
better results, but the results were dependent on the
purity of the training areas and bands used. The
same bands used to visually separate the healthy and
budworm infested spruce again gave the better
classification results. Confusion occurred between
healthy spruce, budworm infested spruce, black
spruce, and treed bogs (Tables 4 and 5). Edge pixels
around the bogs created the greatest confusion. This
classifier was found to be suitable for mapping bud
worm infestation but had to be restricted to the
drier areas where the least confusion occurred.
Table 4. Classification similarities of 5 classes
using 3
MEIS-II
bands at
5.5 m
resolution.
Class
BUDW
HEAL
ASPN
BLSP
WATR
% scene
BUDW
322
5
30
5
1
41.3
HEAL
19
107
0
64
1
6.9
ASPN
14
0
1023
0
0
27.5
BLSP
0
18
0
353
0
5.8
WATR
0
0
0
0
2308
1.3
Total %
90
82
97
83
98
Table
using
5. Classification similarities of 5
7 MEIS-II bands at 5.5 m resolution.
classes
Class
BUDW
HEAL
ASPN
BLSP
WATR
% scene
BUDW
342
4
0
5
0
24.8
HEAL
13
116
0
28
0
2.7
ASPN
0
0
1047
0
0
13.6
BLSP
0
10
0
389
0
3.9
WATR
0
0
0
0
2300
1.2
Total
% 96
89
99
92
98
The third, real-time parallelipiped classifier gave
the best results with the least confusion in the
drier areas of the scene. However confusion still
occurred between the budworm infested spruce, black
spruce and treed bogs. Because individual pixels can
be used for training areas, specific reflectance
values can be chosen for classification. Again, the
number and choice of bands determined the amount of
confusion between classes. Two appropriate bands
quite often gave a good classification with the least
confusion between classes. By adding a third band
confusion between classes increased. This classifier
can therefore be used to classify and separate
healthy white spruce from budworm infested white
spruce at the severe infestation level. There were
some indications that a second level of infestation
could be classified but with little accuracy and more
confusion with other vegetation types.
TM data
A combination of band 1 (red), band 2 (green), and
band 3 (blue) which gave a natural colour image with
spruce budworm infested areas appearing reddish-brown
was useful for visual analysis. Some confusion
occurred between budworm infested spruce, black
spruce, treed bogs, and healthy white spruce. A
second combination which provided good visual sépara
tion was band 3 (red) , band 4 (green) and band 5
(blue). However, some confusion occurred between the
budworm infested areas, wetland areas, black spruce,
and healthy white spruce (Figure 6). Visual analysis
of enhanced TM data can therefore be used for
detecting spruce budworm infestation but with greater
difficulty.
Principal component and Martin Taylor enhancements
did not improve upon the contrast stretches and were
of little use.
The unsupervised maximum likelihood classification
gave very poor results and can therefore be discarded
as a way of mapping budworm infestation. The super
vised maximum likelihood classifier gave better
results depending on the purity of the training areas
used. The main areas of confusion were between bud
worm infestation and the wetland areas. There was
also confusion with black spruce (Table 6). In drier
areas a greater accuracy of classification relative
to the air photos was obtained because there is less
confusion with other vegetation types.
The real-time parallelipiped classifier gave the
best results with less confusion between vegetation
types. At the same time there was confusion between
the budworm infested spruce, the wetland areas, black
spruce and some healthy white spruce but only the
severe infestation could be classified. To some
extent, this method could therefore be used to map
budworm infestation, but should be limited to drier
areas where the least confusion occurred.
Biomass indices obtained for a number of band com
binations were displayed on the colour monitor for
visual interpretation. The best combination was the
biomass index of bands 7 and 3 (red), bands 5 and 3
(green), and bands 4 and 1 (blue). Various combina
tions of biomass indices and contrast stretched bands
were also displayed, but gave no improvements in
separation over straight biomass indices or contrast
stretches. Separation of white spruce, both healthy
and budworm infested, from other vegetation types was
easily determined. Separation between healthy white
spruce and budworm infested white spruce, on the
other hand, was difficult.
Table 6. Classification similarities of 5 classes
using 3 TM bands at 10 m resolution.
Class
BUDW
HEAL
ASPN
BLSP
WATR
% scene
BUDW
719
1
0
19
0
25.8
HEAL
0
766
0
0
0
9.0
ASPN
0
0
1108
1
0
25.6
BLSP
82
0
0
219
4
7.9
WATR
0
0
0
0
918
1.5
Total %
89
98
100
91
99