535 
jri techniqes 
3.3 Study data sets 
3.4 Spectral analysis procedure 
is not possible, 
;ed to represent 
jnrponents. 
ive transforma- 
ition that corn 
er dimensions 
1 content. This 
ance or noise to 
rolucci, et al., 
generation of 
a describe the 
iginal ban-s on 
lure allows us 
3 contains most 
:ion content for 
L984) 
Four data sets of the classified area were used to 
evaluate the contribution of the thermal data in the 
multispectral classifiaction. The first data set is 
the original seven TM bands. The second set is compo 
sed of the same original TM bands excluding the ther 
mal band. The third data set is formed by Principal 
Components loaded from the original seven TM bands. 
The fourthdata set is also Princiapl Components, but 
generated from the second data set, i.e. , only the 
reflective bands. 
The satistics used in calculating the Principal Com 
ponents were generated from data samples of the ori 
ginal TM data set using every fifth line and fifth 
column. 
Tables 1 and 2 shpw the statistics for both Principal 
Components data sets. Tables 3 and 4 list the eigen- 
vales and the corresponding amount of data variance 
taht is accounted for by their respective eigenvectors 
for both data sets. 
A non-supervised approach (Clustering) was selected 
to generate the training statistics. This approach 
groups spectrally similar pixels regardles of their 
spatial position (Tilton and Bartolucci, 1982). and 
extracts the maximum quantity of information availa 
ble in the TM data. 
Eight classifications were carried out in this study. 
Only four spectral analysis were conducted, one for 
each of the data sets, the classifications are results 
of different channel combinations selected after the 
analysis procedure (Table 5). 
To avoid analysis bias in the generation of training 
statistics, the same eight training areas and number 
of cluster classes were requested for each of the 
four data sets. 
The analysis was performed utilizing a defined thres 
hold of 1850 for the transformed divergence distance 
(D.T.),(Swain and Davis, 1982). 
present project 
:>er 1982 over 
? is 40049-16264 
lata used was 
icted, i.e., 
isisted of 5,965 
The geomtric 
ires special 
ition of thermal 
other TM bands, 
al data repre- 
Lts from any of 
ition of the 
agistered grid 
ands of the geo- 
le same number 
Table 1. Eigenvector values for by their respective TM band for the 
original seven TM bands (Data set C). 
Wavelength Principal Component (Karhunen 6 Loeve) Eigenvector 
Band 
1 
2 
3 
4 
5 
6 
7 
1 
0.0376 
0.4331 
0.5665 
-0.1086 
-0.1359 
-0.6781 
-0.0092 
2 
2 
0.0377 
0.2641 
0.2770 
-0.0547 
-0.1632 
0.4311 
0.7988 
3 
3 
0.0293 
0.4032 
0.3564 
-0'0806 
-0.0598 
0.5898 
-0.5930 
4 
4 
0.8109 
-0.4312 
0.3666 
0.0817 
0.1167 
0.0396 
-0.0163 
5 
5 
0.5574 
0.4391 
-0'5642 
-0.0719 
-0.4097 
-0.0659 
-0.0275 
7 
7 
0.1670 
0.4115 
-0.1578 
-0.0465 
0.8770 
-0.0210. 
0.0961 
6 
6 
-0.0101 
0.1830 
0.0285 
0.9822 
-0.0272 
-0.0116 
-0.0013 
roximatelly 
ssentative of ■ 
ar features. 
i.ch is in south 
'45" N and 
W to 93°45' W. 
to undulating 
s and rivers. 
7 of a Wiscon- 
3 underlain by 
P- 
was praire gra- 
grew along the 
Des Moines River, 
bodies, agri- 
old developments) 
dense road net- 
ighways). 
reservation Ser- 
riculture in 
al slides for 
h slide covers 
the ground, 
with aerial 
tory for Appli- 
rdue University 
county as re- 
e present re- 
software system 
data is LARSYS 
nski,1980). 
Table 2. Eigenvector values for by their respective TM band for 
the six reflective TM bands (Data set D). 
Wavelength Principal Component (Karhunen S LOeve) Eigenvector 
1 
2 
3 
4 
5 
6 
1 
0.0393 
0.4400 
0.5694 
-0.1389 
-0.6792 
-0.0092 
2 
0.0388 
0.2654 
0.2787 
-0.1646 
0.4307 
0.7986 
3 
0.0309 
0.4096 
0.3590 
-0.0619 
0.5888 
-0.5932 
4 
0.8093 
-0.4434 
0.3638 
0.1190 
0.0405 
-0.0162 
5 
0.5591 
0.4440 
-0.5619 
-0.4115 
-0.0666 
-0.0275 
7 
0.1686 
0.4177 
-0.1454 
0.8754 
-0.0220 
0.0960 
Table 3. Eigenvalue and the corresponding amount of 
variance that is accounted for by their respective 
eigenvector for the data set C. 
Eigenvector 
Eigenvalue 
Percent 
Variance 
Cumulative 
Percent 
Variance 
1 
795.642 
54.449 
54.449 
2 
554.802 
37.967 
92.416 
3 
81.346 
5.567 
97.983 
4 
14.888 
1.019 
99.002 
5 
10.281 
0.704 
99.706 
6 
2.818 
0.193 
99.899 
7 
1.482 
0.101 
100.000 
Table 4. Eigenvalue and the corresponding amount of 
variance that is accounted for by their respective 
eigenvector for the data set D. 
Eigenvector 
Eigenvalue 
Percent 
Variance 
Cumulative 
Percent 
Variance 
1 
795.569 
55.706 
55.706 
2 
536.714 
37.581 
93.287 
3 
81.290 
5.692 
98.979 
4 
10.285 
0.720 
99.699 
5 
2.820 
0.197 
99.896 
6 
1.482 
0.104 
100.000