ne of the 
ments in 
auses the 
remotely 
ve of this 
| different 
| change 
bul, was 
leveloped 
jricultural 
racteristic 
red rapid 
tion axes 
d rapid 
the rural 
services. 
| of the 
gion was 
5. Today, 
accepted 
protection 
sion and 
nplanned 
eptember 
)SAT TM 
the study 
nosaic of 
The key 
‘able 1. 
LE 
  
  
  
  
(a) 
2.1.2. Image registration 
As a first step all image data set were 
registered geometrically to each other. A 
topographic map (1:25 000 scale) is used as a 
reference grid in the registration process. 
Linear fit polynomial and cubic convolution 
interpolation techniques were used to produce 
the registered output image with a 0.5 pixel 
registration accuracy. 
2.1.3. Change detection techniques 
A change in reflectance often indicates a 
physical change on the ground. The changes 
in reflectance registered from one area 
between two points in time provide a key 
information on land use/cover changes. There 
are many digital change-detection techniques. 
The most common used are ; image overlay, 
image differencing, principal component 
analysis, and classification comparisons. 
1. /mage overlay : The simplest way to 
produce a change image is a photographic 
comparison of a single band of data from 
the two (or more) dates. The image is 
prepared by making a photographic three- 
color composite showing the three dates in 
separate color overlays. The colors in the 
resulting image indicate the changes in 
reflectance values between these dates 
[Virag et a/, 1987]. 
2. Image difference : Another procedure is to 
prepare temporal difference image by 
subtracting the DN(digital number)'s for 
one date from those of the other. The 
difference in the areas of no change will be 
very small and areas of change will reveal 
larger positive or negative values [Lillesand 
et al, 1987]. 
679 
  
| (b) 
Figure 1. Study area ; (a) 1984 Landsat image (band 3/2/1). (b) Aerial photograph (1994). 
3. Principal component analysis : Principal 
Component Analysis can be used to detect 
and identify temporal change when 
registered Landsat TM images are merged 
and treated as a single data set. 
[Ingebritsen et al, 1985] By this method, a 
new set of coordinate axes was fitted to the 
image data, choosing as the first new axis 
or component would account for maximum 
variance. Subsequent axes (components) 
would account for smaller portions of the 
remaining variance. Changes to be 
anticipated were of two types: (/) those that 
would extend over a substantial part of the 
scene, such as changes in atmospheric 
transmission and soil water status; (i) 
those that were restricted to parts of the 
scene, such as construction of roads, 
destruction of green areas. 
4. Classification comparisons : This method 
involves independently classifying each 
image, registering the results and locating 
those pixels that have changed their 
land cover classification between dates. 
Successful application of this method 
requires accurate classifications of both 
scenes, so that differences between two 
dates represent true differences in land 
use rather than differences in classification 
accuracy [Campbell, 1987]. 
3. Results 
Image overlay In the simplest change 
detection procedure, the single band change 
image was prepared by color coding TM band 
3 from the 1992 data as red and from the 
1984 data as green from the 1990 data as 
blue (Figure 2). Band 3 of the Landsat TM 
image was selected because it provides the 
best visual discrimination of rural-to-urban 
land conversion among the land cover groups 
in the study area. The industrial complex (a - 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996