88 
Table 1 Quantification of vegetation and landslide- 
A comparison of unsupervised classification of MSS 
data with ground truth. 
area 
veg.l 
plain 
veg.2 sunfacing 
veg.3 poorly 
veg.4 poorly 
surfacing 
Alfalfa 
farms/orchards 
i 1 luminateci 
Illuminated 
Landslide 
farms 
orchards 
wood lands 
CO 
S3 
v £ 
v £ 
C *-> 
0) 
+J u 
4. £ 
C 4-» 
V 
+J L- 
C 4-> 
g ^ 
1 g 
h ® 
■ match 
¡ 8 
U 0) 
§ rt 
I 
£ 
h 
is £ 
M 3 
x: 
o 
£ 
p 
1 \ 
x: 
0 
4-> 
1 
1 £ 
*9 £ 
a § 
JZ 
o 
+J 
g 
O 0 
£ s 
i S' 
* 
£ s 
i S 
b* 
a u 
b* 
map 
18750 
134375 
306250 
16000 
53000 
Sept. 
~1572 
B7-B4 
22500 
120 
150000 
112 
230000 
75 
20000 
125 
B7-B5 
20000 
107 
145000 
108 
185000 
60 
7500 
47 
VI 
B7tB4 
12500 
67 
142500 
106 
485000 
158 
30000 
187 
72500 
137 
B7+B5 
90000 
170 
July 
T577 
B7-BS 
7500 
40 
182500 
136 
445000 
145 
30000 
187 
VI 
2500 
13 
135000 
100 
440000 
144 
25000 
156 
Note - In the vegetation index of the September image 
185,000 square metre of shadowed areas were 
classified as vegetation and in the vegetation index 
of the July image 145,000 square metre of shadow 
were misclassified as vegetation (see Fig 1 for 
details of selected features). 
2 CLASSIFICATION METHOD FOR THE LANDSAT MSS DATA 
The results of the evaluation of the MSS data ( 
shown in Table 1) were not very satisfactory. 
Various manipulation techniques were explored in 
order to find an appropriate method of classifying 
the MSS data. Several combinations of techniques 
were applied and the results examined. The method 
which yielded optimum results is described below. 
Table 2 Pixel values and characteristics of the 
training areas Mean standard deviation and range 
of each band or combination are recorded under the 
name of each selected features of the training 
areas 
T77] 
B 5 
B 4 1 B7+B5 
VEGETATION 
B7-B5 
VI 
B7+B4 
B7-B4 
mean 
48.30 
20.29 
22.07 
39.78 
61.01 
15.66 
39.39 
59.24 
SD 
6.51 
5.88 
5.87 
5.57 
7.89 
2.67 
5.29 
8.14 
meant,- 
39-58 
11-29 
13-31 
34-52 
50-70 
11-21 
32-48 
50-70 
1.5 SD 
SCREE 
mean 
64.41 
85.97 
86.16 
87.16 
11.45 
1.29 
84.31 
11.26 
SD 
4.34 
5.76 
6.15 
5.67 
3.03 
0.56 
2.23 
3.97 
meant,- 
1.5 SD 
58-71 
77-95 
77-95 
76-96 
7-14 
1-2 
76-92 
6-16 
COARSE LAVA 
mean 
54.23 
65.37 
64.54 
69.41 
21.87 
3.19 
66.58 
22.69 
SD 
2.30 
3.00 
3.08 
2.68 
2.83 
0.51 
2.23 
3.66 
Meant,- 
] .5 SD 
51-58 
61-70 
60-69 
66-74 
18-27 
3-4 
63-69 
19-26 
LIMESTONE 
mean 
69.53 
72.19 
72.47 
82.22 
30.34 
3.68 
79.57 
30.00 
so 
4.53 
3.91 
4.40 
4.62 
2.71 
0.47 
4.65 
3.53 
meant,- 
1.5 ED 
63-76 
66-78 
66-79 
76-91 
27-36 
3-4 
72-87 
24-35 
LOWER JURASSICS 
mean 
23.71 
40.05 
41.36 
36.96 
16.66 
4.62 
36.41 
15.35 
SD 
3.92 
4.54 
4.80 
4.45 
3.74 
1.28 
4.19 
4.65 
meant,- 
1-5 SD 
18-30 
33-47 
34-49 
31-45 
12-24 
3-7 
31-41 
10-24 
LANDSLIDE 
mean 
62.28 
75.91 
77.13 
80.16 
19.38 
2.41 
78.00 
18.16 
SD 
4.56 
6.44 
4.76 
6.03 
4.00 
0.71 
4.89 
3.39 
mean+ 
1.5 SD 
55-69 
66-86 
70-84 
72-87 
15-24 
2-3 
72-85 
13-21 
SHADOW 
r— 
mean 1 0.00 
14.31 
15.50 
8.31 
18.67 
27.60 
8.90 
17.50 
SD 1 0.00 
4.37 
4.76 
2.52 
4.37 
19.48 
2.66 
4.76 
meant,- 0.00 
0-18 
0-19 
5-13 
12-26 
10-50 
6-14 
8-24 
1.5 SD 1 
Table 3 Feature separation using Landsat MSS band 
combinations• 
2.1 Methodology 
A sample of 35 subscenes along the Haraz road was 
used to develop a classification method. The most 
dominant features observed across a majority of the 
subscenes were chosen for classification. The 
features comprised: 
1- (VEG) Vegetation: orchards and farms 
2- (SC) Scree: sunfacing steep slopes, usually 
slumps resulting from landslides. 
3- (CL) Coarse lava: poorly vegetated quaternary 
volcanic conglomerate mixed with 
volcanic tuffs. 
4- (LIME) Light limestone 
5- (SHEM) Shemshak formation: dark coaly lower 
Jurassics 
6- (LANDS) Landslide: unstable grounds 
7- (SHAD) Shadow 
By combining data acquired by field work with 
existing maps, a geological map of the Haraz Valley 
was compiled at a scale of 1:23,000. Subscenes were 
then plotted at the same scale and superimposed on 
the map. Several appropriate training areas, 
containing the seven features listed above, were 
selected from the map and were superimposed on the 
plots of the subscenes. The pixel values of each of 
the selected areas were transferred to separate data 
files. The files contained bands 4, 5 and 7 of the 
two images for the seven selected categories. 
To simplify the training area data sets, the mean 
and standard deviation of the data were calculated, 
and only pixels falling within the range of three 
standard deviations ( mean, plus and minus 1.5 
standard deviations) were reselected. The pixels 
selected by this sieving procedure were more 
MSS band canbination 
seperated features (and their values 
band 7 
band 5 
band 4 
band 7 t band 5 
band 7 - band 5 
Vgetation Index 
band 7 t band 4 
band 7 - band 4 
SHAD (0-1) and SHEM (18-30) 
SHEM (33-47) 
SHEM (34-49) 
SHAD (5-13) and CL(66-74) 
SC (7-14), LIME (27-36) and VEG (50-70) 
SC (1-2) and LANDS (2-3) 
SHAD (6-14) and CL (63-69) 
LIME (24-35) and VEG (50-70) 
homogenous than before, and were less biased by 
outlying values (Table 2). 
Images of the training areas were examined using the 
simplified sets of data for each of 3 bands. 
Various combinations of bands were also examined 
employing the same gains and offsets which were used 
for the subscene. The combinations were: addition 
of band 7 with band 5 (B7 + B5), band 7 with band 4 
(B7+B4), subtraction of band 5 from band 7 (B7-B5), 
band 4 from band 7 (B7-B4) and the vegetation index 
(B7-B5)/(B7+B5). Means and standard deviations 
were calculated and histograms plotted for each of 
the newly derived data sets corresponding to each of 
the 8 combinations of bands. Table 2 shows the 
range of pixel values for each seven classes. 
According to Table 3 dark shadow and lower Jurassics 
can be distinguished in band 7. In bands 5 and 4, 
lower Jurassics can also be delineated in the middle 
part of the histogram, while vegetation and shadow 
coincide in the lower part of the histogram, 
preventing definitive classification. Subtraction 
of band 5 from band 7 (B7-B5) and band 4 from band 7 
(B7-B4) separates vegetation and limestone in the 
upper and middle paart of the histogram 
respectively, while the other features tend to 
overlap. 
Figure 1 
of (B7-B5) 
colours re 
Index (B7- 
lowest va] 
while hi 
vegetatio 
Addition c 
permits s 
across th 
features 
middle pa 
previousl 
vegetation 
histogram 
associate 
delineate 
The veget 
suitable 
mountainoo 
a plain ti 
shadow. 
the appro 
classes 
together \ 
five combi 
A detailec 
Tables 2 < 
nature of 
how, for 
shadowed ; 
on the ve 
areas a mo 
in the th 
using the 
Taken as 
bands used 
and cone: 
band/combi 
informati 
The numbe 
represents 
the featu 
classifi 
diagrammat