Classification comparisons : The range of 
cover types and absence of efficient ground 
data (especially for 1984) caused to use the 
classification method based on an interactive 
ISODATA (ERDAS Ver.7.5) approach. From 
an examination of the data with ancillary data 
(aerial photographs, topographic maps and 
merged Landsat TM + Spot P image (Figure 
5)), both spectrally and spatially, 35 clusters (5 
main land use category; urban area, lake, 
unproductive area, green area, industrial- 
commercial structures and road/construction) 
were separated initially. Statistics from the 
clustering were than used as input to the 
Maximum likelihood classification algorithm. 
The following assessments were obtained 
from the classified images : 
e Changes in land development (both in 
urban areas and industrial/commercial 
structures) were easily separated 
e Green areas (destructed and/or changed 
due to atmospheric effects) were also 
mapped 
e Open space was led to confusion due to 
insufficient ground data 
(c) 
Figure 6. Classification results (a) 1984 Landsat TM (b) 1990 Landsat TM (c) 1992 Landsat TM. 
  
681 
Changes in lake and environments were 
readily seen. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996 
Es