IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India, 2002 
  
  
by satellite over a period of time. This may also happens due to 
inaccuracy of registration of multi-temporal data. Thus accurate 
registration is one of the important factor for creating correct 
change detection (difference area) maps. The pixels which are 
misclassified either due to software limitations or due to 
satellite data sets are kept in misclassified area. In the present 
study, threshold for mixed pixel area for changing from one 
category to another is taken as 0.05 sq. km. It is found that 
dense forest area has been changed to sparse or degraded forest 
on almost all the islands measuring a total of 4.7 sq. km (- 30 
% of total changed area) and near Wandoor village about LI 
sq. km area of barren land is changed to forest probably due to 
coconut plantation. 
2. STUDY AREA 
The study area covers almost the entire Wandoor Marine 
National Park of Andaman Islands lying between 92 ° 30" - 
92° 37° 30” E longitude and 11° 30” - 11? 37' 30" N 
latitudes. This is located in the South Western coast of South 
Andaman, in the Bay of Bengal covering a total area of 188.54 
sq.km. This includes islands namely Tarmugli, Alexandra, 
Redskin, Boat, Malay, Grub, Chester, Belle, Snob, Hobday 
and part of Jolly Boys. 
3. DATA USED 
i) IRS !C LISS-III data of January 18, 1996 & 
ii) IRS-1D LISS-IH data of March 9, 2000. 
4. METHODOLOGY 
i) Two dates data covering the study area was extracted 
using ERDAS Imagine software. 
ii) The two data sets were registered with each other 
using 17 GCPs distributed uniformly over the study 
area. 
iii) A common area of interest (A.I.O) was taken up for 
both the data sets. 
iv) Unsupervised classification was performed taking 
convergence threshold as 0.98, number of iteration 15 
and 60 classes. 
v) The thematic categories were re-coded to delineate 
appropriate land use classes. 
vi) The classified images were smoothed using 3X3 
median filter. 
vii) The images were converted to grid form and then into 
polygon coverages in GIS. 
viii) The query shell was used to generate change 
detection maps. 
5. RESULTS AND DISCUSSION 
IRS 1C LISS-III data of January 18, 1996 and IRS 1D LISS-III 
data of March 9, 2000 were extracted covering the study area 
having scan lines and pixels (538 X 547) and (534 X S15) 
respectively. Band 5 of IRS 1C data was not properly registered 
with other bands 2,3,4 so it was first registered taking FCC of 
remaining bands 2, 3 and 4 as the reference. The uncorrected 
band 5 was replaced by registered band 5. The following 
analysis was then performed using two data sets: 
566 
Registration : 
The two data sets were registered taking 1996 data as the 
reference using ERDAS Imagine software. A total of 17 GCPs 
distributed uniformly throughout the study area were taken for 
registration. The over all accuracy of the registration was less 
than one pixel. It was checked by swapping 2000 data over 
1996 data. 
Classification : 
Unsupervised classification was performed for both data sets. A 
total of 60 classes were taken with convergence threshold 0.98 
and number of iterations 15 for both data sets. The different 
classes were then grouped into proper land use classes using 
ground truth information of the study area. A total of 11 land 
use classes were delineated namely water body, reef flat, sand 
over reef, sand, mudflat, dense and sparse mangroves, dense, 
sparse and degraded forest and barren land as shown in Figure 
l. 
Change Detection Map Generation & Query Shell : 
The classified images were smoothed by using 3X3 median 
filter and then converted in vector form in GIS. The present 
query shell was then used for extracting same area of interest 
(A.O.I and map generation. To take into consideration of 
various types of data sets ( raster and vector), the following 
Customized Packages (C.P) are required 
i) C.P for raster and vector data integration 
and map generation 
ii) C.P for map generation using registered 
raster — raster data sets. 
iii) C.P for map generation using registered 
vector-vector data sets. 
For first case, user friendly customized package has been 
developed by Gupta et, al. (2001). The present query shell 
developed using ARC/INFO GIS can be used for other two 
cases. Raster data sets used in the present study are converted 
into polygon vector form in GIS. The query shell has the 
following capabilities (sub-shells): 
€ Concurrent display of Coverages & existing map 
compositions — For coverages, it displays one coverage 
with polygon shades and other as the line coverage for 
checking the registration. For map compositions, it 
displays the list of all the available map compositions 
(.com) on a popup menu and display the selected map 
composition by clicking it in the menu with any page size. 
By default, it displays A3 size (16"X10.5" ) map 
composition. 
€ Selection of common Area of Interest (A.O.I) - It displays 
two registered data sets, one with polygon shades and 
other with line option. It then prompts for any user specific 
area by choosing a box using two corner points. The area 
can be selected more accurately by using zooming option 
of the query shell. It then automatically extract the selected 
area from both data sets and display it on the screen. It 
then asks user's option to select the area again if required 
so that finally a optimally desired area can be selected. 
+ Automatic Change Detection Maps Generation — It 
generates a map having two data sets on the same scale, 
with automatic legend generation and areas of land use 
categories (Gupta et, al. 2001) 
+ Automatic Constant and Difference Area Maps Generation 
— It automatically generates two maps, one showing the