International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
Table —1-area of different classes 
  
  
  
  
  
  
  
  
  
  
Class Area(ha) 
Forest 44952.4 
Rangeland 12236.7 
Rice field 3951.2 
Urban area 2599.6 
Dry Farm 514 
Water body 16.5 
Wheat field 670.5 
Shadow 1356.6 
Cloud 369.2 
  
  
  
  
To determine the best band set OIF and correlation method was 
used. The OIF results are given in Table (2). 
Table —2-results of OIF 
  
  
  
  
  
  
  
Category Band set OIF 
2,4,6 46697.89 
2 3.4, 7 20121.29 
3 1,3,4 2411.28 
4 1,4,6 2358.39 
5 2,3,4 2331.73 
6 1,4,7 1614.67 
  
  
  
  
  
Correlation among bands demonstrates that the bands 1, 5 and 6 
are a suitable set for land classification. The classification 
accuracies by different band sets are presented in Table (3). 
Table (3) Classification accuracies using different band sets 
  
  
  
  
  
  
  
  
Band set Accuracy Band set Accuracy 
15,6 85.8 1,4,6 75.4 
1,4,6,5 83.6 4, 3,2,5 73.2 
All bands 81.2 1,47 70.2 
4,3,2,5,.0 80.4 4,3,2 67.8 
  
  
  
  
According to above table the highest accuracy is allied to band 
set: TM 1. TM5 and TM6. Where the band sets does not include 
the thermal band (TM6), spectral similarity between forests and 
rice fields are understandable, such as band set of TM4, TM3 
and TM2 (Fig. 1) which these two land use can not be 
separated. During 18^ July 2002 (The used image date), the rice 
fields has achieved their high biomass peak and the rice plants 
have covered the field completely, therefore spectral similarity 
occurs between forests and rice fields on visible bands such as 
TM2, TM3 and near infrared TM4. On the other hand, water 
can be detected with TMI easily because it has a high 
reflectance on this band, whereas the forest absorbs the blue 
spectrum for photosynthesis. 
392 
  
FCC 156 FCC 432 
Figure — 1- comparison between 432 and 156 false 
color composites 
Also the water on the rice field causes a temperature reduction 
by evaporation which can be detected with TM6. But the forest 
canopy has only a common transpiration. So, due to this 
different water evaporation and water transpiration above them, 
various thermal characteristics of forest and rice field occur. 
Because of water accumulation on rice fields it is cooler than 
the forests. The temperature values of these two land use were 
calculated and few degrees of centigrade was recorded for forest 
canopy and rice field. The outcome results of this study confirm 
our claim about presence of thermal differences between forest 
and rice fields and it proves the other studies as well. 
Malingreau suggests that thermal band can show the thermal 
variation of various crown cover of the forests (Fallah Shamsi, 
1997). Also Zobairi and Majd (1996) explain that one can 
separate two phenomenon using thermal bands when their 
reflectance is similar in visible bands. Also Alavipanah (2004) 
emphasizes on use of thermal band in earth sciences studies. 
According to Table (3) the accuracy of classification is high 
when one uses TM1, TM5 and TM6 bands set, instead of using 
all bands for this kind of studies. It can be concluded that this is 
due to the high correlation among bands. Sabins (1996) states 
that high correlation among bands means replication of data. 
Maximum likelihood method has a good result on out coming 
maps. The overall accuracy and Kappa coefficient of this 
method is quite higher than the other algorithms such as: 
Minimum distance and parallelepiped classifiers (Table 4). 
According to this table the highest accuracy is related to 
Maximum likelihood classifier. Booth and Oldfield (1989), and 
Alvipanah et al. (2001) have emphasized on the priority of 
Maximum likelihood algorithm in compared to Minimum 
distance and parallelepiped classifiers. 
Table (4) Accuracy of classification in different 
classification methods and algorithms 
  
  
  
  
  
  
  
  
  
Classification Method Overall a Kappa 
Accuracy Coefficient 
2 Parallelepiped 34.27 19.03 
5 Minimum Distance 73.77 47.12 
- 
2 Maximum Likelihood 85.83 62.81 
e —— ——— 
Hierarchical 94 84.89 
  
  
According to Table (4) highest accuracy related to hierarchical 
method. Also Stanz (1987) used hierarchical classification and 
Darvishsefat (1997) used this method for forest and forest- 
stands classification. 
This study showed that there is spectral similarity between 
rangelands and wheat fields. Sparse vegetation of rangelands 
Interne 
causes 
of veg 
time t 
vegeta 
confus 
]-Alav 
using I 
Journal 
Univer 
8, No. 
2-Alav 
therma 
ess of 
sensing 
applica 
Vallod 
3-Apan 
rehabili 
Vol. 18 
4-Bootl 
classifi