89
Lstics of the
tion and range
rded under the
the training
1 B7+B4
B7-B4
56
39.39
59.24
57
5.29
8.14
21
32-48
50-70
29
84.31
11.26
56
2.23
3.97
2
76-92
6-16
19
66.58
22.69
51
2.23
3.66
1
63-69
19-26
58
79.57
30.OC
47
4.65
3.53
4
72-87
24-35
32
36.41
15.35
28
4.19
4.65
7
31-41
10-24
41
78.00
18.16
71
4.89
3.39
3
72-85
13-21
60
8.90
17.50
48
2.66
4.76
50
6-14
8-24
idsat MSS band
ieir values
50-70)
ess biased by
nined using the
of 3 bands,
also examined
>hich were used
were: addition
7 with band 4
>and 7 (B7-B5),
getation index
rd deviations
;d for each of
ling to each of
e 2 shows the
classes.
ower Jurassics
bands 5 and 4,
in the middle
ion and shadow
le histogram,
i. Subtraction
4 from band 7
aestone in the
e histogram
tures tend to
(b)
Figure 1 Vegetation discrimination. a)Subtraction
of (B7-B5) of the September image- White and yellow
colours representing vegetated areas- b) Vegetation
Index (B7-B5)/(B7+B5) of the September image- The
lowest values represent sunfacing unvegetated areas,
while high values correspond to shadow and
vegetation.
(b)
Figure 2 subtractions of bands 4 and 5 from band 7
in the September and July images, a) subtraction of
band 4 from band 7 of the September image (high
values, white and yellow, correspond to vegetation),
b) subtraction of band 4 from band 7 of the July
image ( also high values represent vegetation).
Addition of band 7 to band 5 reveals more detail and
permits shadow and coarse lava to be classified
across the range of the histogram. However, other
features still tend to coincide on the upper and
middle part of the histogram- As was pointed out
previously, shadow and vegetation overlap in the
vegetation index for most of the upper range of the
histogram, whilst landslide and scree ( usually
associated with each other on the ground) can be
delineated from the lower part of the data range.
The vegetaion index therefore appears to be less
suitable for classifying vegetation in a
mountainous region (such as the study area) than for
a plain terrain where there is such little or no
shadow. Table 3 shows the results achieved using
the approach described above. With all seven
classes classified using three spectral bands
together with the classifications obtained by the
five combination of the three bands.
A detailed study of each of the features listed in
Tables 2 and 3 provides a useful insight into the
nature of image processing. For example one can see
how, for all three bands, low pixel values in
shadowed areas produce a very high range of values
on the vegetation index. While in the vegetated
areas a moderate response of the spectral signatures
in the three bands gives moderately high values
using the same combination.
Taken as a package, the various combinations of
bands used in this research provide a comprehensive
and concise method of classification. Each
band/combination of bands yields additional, unique
information on at least one of the seven features.
The number of bands/combinations of bands used
represents the smallest number that will allow all
the features to be properly classified. The
classification procedure is illustrated
diagrammatically in Figure 4.
As a general point, this section has demonstrated
that without an understanding of how an image
processing system works, the use of sophisticated
computer equipment to analyse images will produce
spurious results. A thorough understanding of the
principles underlying image processing is therefore
essential, in order to both choose the appropriate
computer routines and to usefully interpret the
results.
3 LIMITATIONS
During this study the main aim was to search for a
method to locate unstable areas in mountainous
region. Several limitations of MSS data for this
kind of discrimination were revealed. The main
shortcoming was the size of the ground cell
resolution (79x79 m). A second limitation in this
particular area was the shadow effects of the
dramatic relief. The other limitations can be listed
as vegetation cover, and atmospheric effects.
4.1 Ground Cell Resolution
The smallest picture element (pixel) in MSS imagery
is 79 x 79 m. Due to the technical characteristics
of the equipment on boardthe Landsat satellite, an
overlap occurs along the latitudinal scan lines.
This makes each pixel 79 m long in N - S direction
and 57 m in E - W direction and prevents features
smaller than 79 m from being distinguished (except
for very highly reflective features such as water
bodies and disturbed construction sites).
Despite the narrowness of the roads and channels,
these can be detected on the MSS images. However,
the high reflectivity of small areas of disturbed
land, combined with the reflectivity of surrounding
features produces a mixed pixel response and makes
interpretation difficult.