Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Assessment of TM thermal infrared band contribution in land 
cover/land use multispectral classification 
José A.Valdes Altamira 
ICI, Mexico 
Marion F. Baumgardner 
Purdue University, Laboratory for Applications of Remote Sensing, West Lafayette, Ind, USA 
Carlos R.Valenzuela 
ITC, Enschede, Netherlands 
ABSTRACT :. Thermal data from Landsat 4 TM were used in conjunction with the six reflective TM bands to assess 
the contribution of the thermal band in eight multispectral classifications using four different data sets. 
Despite its coarse resolution and differences in radiometric measurements, the thermal data provided an 
additional informational plane in the generation of Principal Components. This informational plane did not 
appear when the thermal band was excluded from the linear transformation. The use of all seven TM bands for 
cluster statistics generation provided greater statistically separability between pairs of spectral classes 
than when only reflective bands were used. Classification with subsets of selected bands gave better results 
than classification performed without the use of the thermal band for statistics generation. Classifications 
with Principal Components reduced the number of spectrally separable classes, but with a significant reduction 
in computer time. 
The present paper is an abbreviated version of a Master of Science thesis (Valdés, 1984) , as part of the LIQDA 
NASA contract NAS5-26859 conducted by LARS/Purdue University. 
1 INTRODUCTION 
Thematic Mapper sensor era started with the launch of 
Landsat 4, the first of the second generation of Land) 
sat satellites. This sensor has better spatial reso-~ 
lution than the earlier Multispectral Scanner onboard 
Landsats 1,2 £ 3 (30 m -vs- 80m), seven spectral 
bands instead of four, and four the number of Quanti 
zation levels (256 -vs- 64). 
The T.M. also has a band in the thermal infrared re 
gion of the spectrum, this band differs from the re 
flective bands in its spatial resolution (120 m) and 
the type of electromagnetic measurements. This band 
has not been used often by the scientific community 
either in the experiments with T.M. simulators or in 
the first analysis conducted by NASA on the Landsat 
Image Data Quality Analysis. 
The hypothesis of this study is that the use of the 
T.M. thermal infreres band in conjunction with the 
six reflective bands will provide better discrimina 
tion of agricultural and urban features than does 
classifications with the six reflective bands only. 
The hypothesis can be expressed as: 
Ho = P(7 TM bands) ^P(6 TM reflective bands) 
HI = P(7 TM bands) ^rP(6 TM reflective bands) 
Where P = goodness of classification. 
Principal Components analysis (data compression tech 
nique) was also performed to evaluate the contribu- ~ 
tion of each band to the informational content of the 
T.M. data. 
2 LITERATURE REVIEW 
2.1 Agricultural mapping with remote sensing data 
The specialized literaure in remote sensing contains 
many examples of the detection and quantification of 
crops using techniques of digital analysis. Many of 
these applications are considered either experimental 
systems (Bauer', et al. ,1971 ;Bauer, 1977; Valdes ,1981) 
or quasi-operational systems (McDonald and Hall,1978) 
The results of some of these experiments show diffe 
rent degrees of accuracy in the identification and 
quantification of crop resources. However, all these 
results demonstrate a great potential for surveying 
crops due to the characteristics of.the data obtained 
by the Landsat sensors, and the computer processing, 
for monitoring the vegetative resources in large geo 
.graphic areas. 
There is a great amount of documentation available 
related to the physiological, physical and spectral 
behavior of vegetation. These must be considered in 
understanding how solar energy interacts with the ve 
getation and in order to interpret data from multi 
spectral sensors. 
In 1963 , Hoffer and Johannsen working with different 
vegetative species (com, soybeans and 3 timber spe 
cies) , found that the spectral response of all those 
species have the same typical vegetation curve. They 
also found significant differences in the response — 
at certain wavelengths, mainly in the visible and 
near infrared portions of the spectrum. 
To discriminate crop species by means of remote sen 
sing, several factors related to the cultural practi_ 
ces for each crop must be considered, such as plant 
and row spacing, geometric arrangement of the plants, 
fertilization and irrigation practices, and growth 
cycles. The differences in reflectance wich allows us 
to discriminate between vegetative species, are due 
to the characteristics of the leaves and canopies of 
different species. All these internal and external 
factors influence the optical properties of the leves 
and canopies. The spectral patterns sensed by the scan 
ners represent the integration of all of them. 
2.2 Thermal and environmental effects of incoming 
solar energy 
In order to interpret remote. sensing data of vegeta 
tion, it is important to comprehend the interaction 
of the plant with its environment. A plant is exposed 
to electromagnetic radiation from its surroundings, 
such as soil, rocks, plants, sun, sky, clouds and 
atmosphere. All objects above.absolute zero radite 
energy by virtue of their tempreature and emittance. 
At temperatures normally exhibited by objects at or 
near the earths surface, this radiation is almost en 
tirely in the infrared wavelength region from 4 urn to 
100 urn approximately (Swain and Davis, 1978), 
Plants in stress caused by insects, diseases, physio 
logical disorders, nutrient deficiency and adverse en 
vironmental effects suffer detectable temperature and 
or emittance changes (Kumar and Silva, 1973). 
Several authors have presented the potential use of 
thermal change detection on plants in order to evalua 
te stress causal agents. Clum (1926) and Curtis (19357)