Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
87 
Classification of land features, using Landsat MSS data 
in a mountainous terrain 
H.Taherkia & W.G.Collins 
Remote Sensing Unit, Civil Eng. Dept., Aston University, UK 
ABSTRACT - This paper evaluates Landsat MSS data in a hostile terrain located in central Alborz, north of Iran. 
Correlation of the Three hands of the image were evaluated, and separation of differnt ground features was 
examined. Training areas of different categories were located on the image using field work. Data from the 
training areas then were manipulated to eliminate unwanted pixels. 
Mean and standard deviation of the categories were calculated and data which fell in the range of three 
standard deviations was selected. For the execution of the training areas a selected subscene was chosen and 
then classified. 
1. INTRODUCTION 
The study area is located in central Alborz, north 
of Iran in a very hostile terrain, where the Haraz 
Road traverses through. Two Landsat MSS secenes 
recorded at 4th September 1972 and 13th July 1977 
were used for this study. 
After applying geometric and radiometric 
corrections, resampling at 50x50 metres, registerig 
two multitemporal images together on the DIPIX, 35 
subscenes (e each of 40x40 pixels) were chosen along 
the road for further investigation. 
A FORTRAN program was developed to enable plotting 
of the subscenes in eight colours using a Tekronix 
ink jet colour plotter. The package developed can 
perform mathematical manipulation of the data to 
produce a variety of combinations viz: additions, 
subtractions, ratios, and vegetation index/formulae 
etc. and plots the output data in the form of a 
histogram. These statistical analyses have been 
applied on a subscene which consists of landslide, 
coarse lava, scree, vegetation cover and limestone. 
In processing the subscene, plots of single bands in 
eight colours (from black to white) were found to be 
most acceptable to the human eye. However, due to 
low values of vegetation in band 5 and higher values 
of them in band 7, different combinations of the 
bands showed further improvement in the 
identification of ground features. Subtraction of 
the two bands (B7-B5) showed clearly the boundaries 
of ground vegetation (Figure la). The high 
reflectivities of screes and bare soil in both bands 
7 and 5 when they were combined in vegetation index 
(B7-B5)/(B7+B5) caused low values in the resultant 
image (Figure lb). Hence all unvegetated areas 
appeared dark in the vegetation index. 
The results of the different combinations were 
compared against the ground truths at same the scale 
(1:23,000). The outcome is shown in Table 1 
comparing results of different combinations of 
different bands and the ground truths derived from 
existing maps of the study area. 
As can be seen from the eight different combinations 
in Table 1, the best results are obtained after 
subtraction which allows discrimination of 
vegetation covered areas (units 1, 2, 3 and 4). The 
worst combination appears to result from addition of 
the bands, which provides no basis for significant 
discrimination of the vegetated areas. 
The vegetation index, shown in Table 1 and Figure 2 
highlights vegetated areas, but some areas of dark 
shadow are also classified in the vegetation class. 
Hence the use of the vegetation index in unknown 
areas produces imperfect results. Despite the 
recommendation of many Remote Sensing texts that a 
vegetation index should be used as a tool to 
distinguish vegetated areas in the Landsat MSS 
images, it seems, on the basis of this research that 
the best results are obtained by subtracting band 5 
from band 7 (B7-B5). 
Subtraction of band 4 from band 7 (B7-B4) in both 
September and July images (figure 2) resulted in 
over- classification in three vegetated areas and 
under-classification in one area located on the west 
facing slope of the Valley (Table 1). The shadow is 
responsible for misclassification in the number 3 
vegetated area, whereas mixed pixels of high 
reflection of vegetation in boundaries are 
responsible for over-classification. 
Where band 5 was subtracted from band 7 (B7-B5), two 
sunfaced vegetated areas (units 1 and 2) were found 
to match the base map (Figures 1 and 4) whereas two 
low illuminated areas (units 3 and 4) mismatched 
(Table 1 and Figure 3). 
Additions of different bands in each of the 
September and July images revealed no more 
information in this case. Only by the additions of 
both band 5 and band 7 (B7+B5) and band 7 and band 4 
(B7+B4) does there seem to be slight improvement in 
landslide discrimination. It should be mentioned 
here that addition of two bands in image processing 
systems in a vast area showed a better result in 
linear features recognition. 
In addition to conventional image procesing 
techniques, the use of either a vegetation index or 
subtraction can assist the interpreter in the 
identification of unstable areas between bare soil 
and in the classification of vegetated areas. Both 
procedures should be performed after applying 
radiometric correction to the MSS image. Without 
haze removal, any mathemtical manipulation on the 
data will not produce a reliable result.