related soil moisture stress with temperature differen 
ces in cotton and potatoes. Wear (1966) found an in-’ — 
crease in temperature in forest trees with roots dama 
ged by insects. Myers and Allen (1968) realted soil — 
salinity with high cotton leaf temperatures. 
The Corn Blight Watch Experiment, demonstrated that 
use of infrared remote sensing has possitive effects 
in stress levels determinations (MacDonald, et al., 
1972; Kumar and Silva, 1973). 
2.3 Airbone multispectral scanning thermography 
Myers, et al. (1966) made use of pictorial and thermal 
infrared data to determine differences in the tempera 
ture of plants as an indicator of the relative subsur 
face salinity and moisture conditions affecting crop - 
production. They stated that the temperature contrasts 
between salt affected and unaffected cotton plants are 
likely to be greater than the temperature contrasts 
between moisture stressed and unstressed cotton. 
Wiegand, et al. (1968), using the UNiversity of Mi 
chigan airborne thermal scanner in Texas, studied the 
thermal behavior of several variables such as crop 
species, plant spacing, tillage, irrigation regime and 
special features, such as highways and water reservoirs. 
They found that irrigated crops tend to be cooler that 
non irrigated at midday conditions, but the opposite 
results were obtained at early morning hours. Thermal 
differences related to tillage were minimal. 
The feasibility of using thermal imagery for land use 
land cover studies has been demonstrated. Brown and 
Holz (1976) following Anderson^s classification system 
(Anderson, et al. ,1976), produced a land use/land cover 
map of Oak Creek Lake, West Texas. 
2.4 Thermal band of Landsat 3 
The Landsat 3 MSS characteristics are in sense the same 
as those of the previous Landsats, except that Landsat 
3 acquired additional data in the thermal infrared por 
tion of the spectrum (10.4 to 12.6 um) with a ground 
resolution of 237 m. As a result, a single thermal 
band measurements corresponds to an area represented 
by nine measurements in each of the four reflective 
spectral bands, a 9 to 1 ratio (Price, 1981). 
The Landsat 3 thermal band did not function properly 
due to several unexpected causes. The problems asso 
ciated with the thermal sensing system were reflected 
in the quality of the imagery. Both thermal and spa 
tial resolution were affected and the thermal imaging 
system was eventually turned off in the spring of 1979 
(Price,1981; Lougeay,1982). 
Despite the problems associated with the thermal band, 
some analysis was performed to evaluate the contribu 
tion and usefullness of this band. Price (1981), using 
Principal Components analysis, assessed the statistical 
correlation between the emissive band, and the four 
reflective bands. He found that the thermal data ei 
ther were not useful or were associated with a physical 
parameter that is not directly related to surface type. 
He found that thermal data made a limited contribution 
to multispectral classifications. He concluded that its 
use for classification is subject to ambiguities and 
prone to error: "...an indiscrimante use of the thermal 
data appears to be undesirable because of many possi 
bilities for misinterpretation and the fact that the 
thermal ’signature' is not a direct indicator of sur 
face type." 
Lougeay (1982) compared the Landsat 3 MSS band 5 
(0.6 to 0.7 um) and the thermal MSS band 8 (10.4 to 
12.6 um). He found the thermal imagery of MSS band 8 
to be of limited use by itself due to its coarse spa 
tial and thermal resolution. However it did provide 
a rendition of gross topographic structure which was 
not readily available from the other MSS spectral 
bands. 
2.5 Classifiaction and data compression techniqes 
If the use of all available channels was not possible, 
data compression techniqes have been used to represent 
the large content of data into fewer components. 
Principal Components or Karhunem - Loeve transforma 
tion is an orthogonal linear transformation that com 
presses multidimensional data into fewer dimensions 
without significant loss of information content. This 
transformation assigns the random variance or noise to 
eigenvectors with lowest variance (Bartolucci, et al., 
1983). 
Data compression is one result of the generation of 
principal components. It is possible to describe the 
relative influence or "pull" of the original ban-s on 
each of the new components. This procedure allows us 
to evaluate which of the original bands contains most 
of the significant variance or information content for 
a particular data set (ANuta, et al. , 1984) 
3 METHODOLOGY 
3.1 Landsat TM characteristics 
The TM data utilized to carry out the present project 
were gathered by Landsat 4 on 3 September 1982 over 
the central Iowa. The NASA scene number is 40049-16264 
accesion 182, path 27, row 31. The TM data used was 
radiometrically and geometrically corrected, i.e. , 
P-tape or fully processed tape, and consisted of 5,965 
scan lines with 6,976 pixels per line. The geomtric 
correction of the TM thermal data requires special 
consideration, since the spatial resolution of thermal 
data is 129 m compared to 30 m for the other TM bands. 
One image sample or pixel of raw thermal data repre 
sents an area equivalent to 16 area units from any of 
the reflective bands. The coarse resolution of the 
thermal data is resaimpled to forma a registered grid 
of 28.5 m by 28.5 m pixels. Thus all bands of the geo 
metrically corrected TM data contain the same number 
of pixels per unit area. 
3.2 Description of the study area 
A study area of 10 by 10 sections (approximatelly 
26,000 hectares), was selected as representative of • 
a great diversity of land use/land cover features. 
This area is located in Polk County which is in south 
central Iowa. 
The area lies between latitudes 41°37'45" N and 
41°46’15" N, and from longitude 93°37’ W to 93°45_' W. 
The general topography is nearly level to undulating 
with some steep areas along the streams and rivers. 
The geology of the area consists mainly of a Wiscon- 
sonian glacial till. The entire area is underlain by 
a shale bedrock of the Des Moines Group. 
The native vegetation of Polk County was praire gra 
sses and hardwood forests. The forests grew along the 
major streams, particularly along the Des Moines River. 
The cover types in this area are water bodies, agri 
cultural fields, urban areas (new and old developments) 
industrial and commercial parks, and a dense road net 
work (from gravel roads to four lane highways). 
The Agricultural Stabilization and Conservation Ser 
vice (ASCS) of the US Department of Agriculture in 
Polk County collected 35 mm color aerial slides for 
the entire county in August 1982. Each slide covers 
two sections (approximately 520 ha) on the ground. 
These slides were used in conjunction with aerial 
infrared slides obtained by the Laboratory for Appli 
cations of Remote Sensing (LARS) of Purdue University 
in May 1983 over selected sites in the county as re 
ference data. 
The hardware and software used for the present re- 
searcg resided at LARS/Purdue U. The software system 
for digital analysis of multispectral data is LARSYS 
(Phillips, 1973) and LARSYSDV (Mrcoczynski,1980). "