You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Remote sensing for resources development and environmental management
Damen, M. C. J.

Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
LANDSAT TM band combinations for crop discrimination
Sherry Chou Chen, Getulio Teixeira Batista & Antonio Tebaldi Tardin
Departamento de Sensoriamento Remoto, Instituto de Pesquisas Espaciáis (INPE), Sao José dos Campos, SP, Brasil
data for the
used for
e been
he radio-
of the
ween the
igital words
the NLR for
facility has
ric and geo-
titude data
h position
R sensor
by an inter-
ty tape to
AR data are
.11 radio-
up tables
f each
n software
ied in order
¡eometry of
:ent removed
m model
atform data.
:a from
¡nt dates of
l registra-
ie performed
•ol points,
is overlays
i programme
close co-
>rogramme is
L be investi-
/ailable high
ige approach.
ABSTRACT: LANDSAT Thematic mapper provides not only more spectral bands but also improved spatial resolutions
in the visible and infrared wavelenghts as compared to the MSS data. However, problems are encountered by
analysts in working with the increased number of wavelength bands. In order to learn how to analyze TM data
for agriculture studies, LANDSAT data of a 15x15m area in Parana State, Brazil, were acquired on Jan. 19,
1985. The predominant crops in the study area were soybeans, com and sugarcane. To choose the best
combination of three TM bands, which represents most information of the agricultural scene the entropy
criterion was used. Once the triplet bands were chosen, the color of green, red and blue were associated to
them according to the magnitudes of their variances to form the color composite. Interpretability of these color
images were evaluated visually. For digital analyses the criterion of Jeffreys-Matusita distance was applied
to verify the best band combination if 2,3,4 or 5 TM bands were used. A classification algorithm based on
the maximum likelihood decision rule was then employed to classify the study area using the designated TM
bands. Classification performances were compared pixel-by-pixel on alphanumeric printouts, the computer
time consumed, the classification matrice and the upper bounds of the probability of error. After these
analyses, the TM bands which should be used for an effective digital analysis of this agricultural scene
were decided.
RÉSUMÉ: Le LANDSAT TM fournit davantage de bandes spectrales et une plus grande résolution spatiale que
le MSS. Cependant, il existe des problèmes qui apparaissent lors de l'utilisation des sept bandes spectrales.
Pour tester les données TM dans les applications agricoles, on a utilisé des données TM LANDSAT dans un
carré de 15x15km dans l'État du Paraná (Brésil) durant le passage du LANDSAT, le 19 janvier 1985. Les
principales cultures de la région étudiées étaient: soja maïs et canne a sucre. Pour choisir la meilleure
combinaison des trois bandes TM qui comportent le plus d'informations intéressantes, on a utilisé le critère
de l'entropie. Après le choix des trois bandes, les couleurs verte, rouge et bleue ont été associées à ces
bandes selon les grandeurs de leur variance pour former une composition colorée. Lès résultats de L ' interpreta tin
visualle des images produites ont été compares. Pour vérifier les meilleurs combinaisons de
bandes pour la classification par ordinateur, on a utilisé le critère de la distance de Jeffreys-Matusita.
Ensuite, on a utilisé un algorithme de classification qui utilise les meilleures combinaisons choisies
pour 3,4 ou 5 bandes du TM. Des analyses ont été faites avec quatre types différents de présentation des
résultats: a) les sorties alphanumériques; b) les matrices de classification; c) les limites supérieures
de la probabilité d'erreur, et d) le temps d'ordinateur utilisé. D'après ces comparaisons la meilleure
combinaison de bandes TM pour la classification des cultures a été determinee.
Since 1982 a new sensor, called the Thematic Mapper
(TM) , was mounted on the Land Observation Satellite
(LANDSAT) together with the Multispectral Scanner
System (MSS). The TM sensor provides data frcm
seven better selected spectral regions. There are
three bands frcm the visible, one iron the near
infrared (NIR) , two frcm the middle infrared and
one iron the thermal infrared spectrum region. The
reflected energy frcm the Earth surface is encoded
into 8 bits per band, with an improved spatial
resolution of 30m instead of the 6 bits data and
80m resolution provided by MSS. In short, the TM
sensor has considerably better spectral, spatial,
and radiometric resolution than the MSS system;
consequently a superior data quality and a much
larger data volume are obtained. This new sensor
design was mainly for vegetation discrimination
considering the charecteristic spectral response
of vegetation of the selected TM bands (Solomonson
et al. 1980). Thus, TM data are expected to
improve crop identification and area estimation
accuracy in Parana State, where the problem of
strip fields is presented. However, one of the
problem encountered in the analysis of TM data is
to decide how to handle this huge data quantity
efficiently. Experiences of the passed decade
demonstrated that for MSS data bands 4,5 and 7 are
used to form false color composite, and for digital
analysis, normally all four bands are employed,
event though information contents of the two visible
bands (band 4 and 5) or infrared bands (band 6 and
7) are intercorrelated. Now, for the seven TM bands,
questions arise about which three bands should be
used for color image production, and how to reduce
the dimensionality of TM data for digital analysis
in order to achieve cost-effective results
considering the crop identification and area
estimation accuracy and computer time consumed. The
knowledgement of how to produce color image using
TM data for visual interpretation is especially
important for developing countries, where the
lacking of computer facility and properly trained
analyzer are limitations for the implementation of
digital analysis at local government agencies or
research institutes. On the other hand, in many
application areas, information contents can only
be extracted by digital analysis. Thus, there is an
urgent need for exploiting how to handle and analyze
TM data both visually and digitally.
In this study TM band combinations for visual and
digital analyses in an agricultural scene were
investigated and the best band combination for crop
discrimination was selected. Note that band 6 was
not included in this study due to its low spatial
resolution (120 m).