This article was downloaded by: [Moskow State Univ Bibliote]

On: 27 November 2013, At: 21:44

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Remote Sensing

Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tres20

The selection of the best possible Landsat TM band

combination for delineating salt-affected soils
a a
R. S. DWIVEDI & B. R. M. RAO
a
National Remote Sensing Agency , Balanagar, Hyderabad, 500 037, India
Published online: 25 Jun 2007.

To cite this article: R. S. DWIVEDI & B. R. M. RAO (1992) The selection of the best possible Landsat TM band combination for
delineating salt-affected soils, International Journal of Remote Sensing, 13:11, 2051-2058, DOI: 10.1080/01431169208904252

To link to this article: http://dx.doi.org/10.1080/01431169208904252

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 The selection of the best possible Landsat TM band combination for
delineating salt-affected soils

R. S. DWlVEDl and B. R. M. RAO

National Remote Sensing Agency, Balanagar, Hyderabad-SO0 037 India

(Received 8 January 1990: in j n a l form 8 November 1990)

Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

Abstract. A quantitative approach was pursued for identifying the most appro-
priate three-band combination of Landsat Thematic Mapper (TM) reflective
bands data for delineating salt-affected soils of the Indo-Gangetic alluvial plain.
The standard deviation and correlation coefficients values of the TM data were
used for computing a statistical parameter called the 'Optimum Index Factor'
.~ . that is indicative of the information (variance) content of the data.
(OIF)
Amongsl all the 20 three-band comb~nalionsconsdered, the hand combmatton
1 . 3 and 5 was found to hc the hest I n terms of ~nlurmationcontent. The norrndlv
used band combination 2, 3 and 4 ranked relatively very low. The validation df
this conclusion with the accuracy estimates of the delineation of salt-alfected soils
using the same data revealed a mixed relation between the ranking oblained from
the OIF values and the accuracy estimates, thereby pointing towards lurther
investigation in other areas with similar terrain conditions.

I. Introduction
Spaceborne multi-spectral data have been available to the user community since
the launch of the Earth Resources Technology Satellite, ERS-I, in 1972. For
deriving information on natural resources multi-spectral data in the form of
standard false colour composite (FCC) image made from green, red and near-infra-
red bands are used; and digital data are also widely used. Special products are also,
however, generated from any three spectral bands from the set of available bands.
For instance, 20 combination of three bands are possible from Landsat Thematic
Mapper (TM) reflective bands.
In order to make the most efficient use of multi-spectral data it is essential to
identify the best possible three-band combination that could provide the desired
information on natural resources. This can be achieved by visual comparison of
FCC images. However, deciding the best possible combination based on visual
interpretation is relatively difficult, subjective and time-consuming. Digital tech-
niques make use of such a parameter as the Optimum Index Factor (OIF) and
selective principal component analysis involving principal component analysis of a
highly correlated sub-set or band pairs (Chavez er a / . 1982, Chavez 1984). Besides
these approaches, scattergram analysis (Chavez and Berlin 1984) and first-order
differencing have also been advocated for information content analysis of remote
sensing data.
The mapping and monitoring of salt-affected soils which cover an estimated area
of about 7 million hectares of India's geographical area is o f paramount importance
in the large-scale reclamation programme. Spaceborne multi-spectral, especially
014>1161/92 S3.m 0 1992 Taylor & Fmncis Ltd
 2052 R. S. Dwiuedi and B. R. M . Rao

Landsat-TM, data offer a great potential for mapping and monitoring these soils.
Hence, the selection of the best three-band combination for either generating FCC
images for visual interpretation or for computer-aided digital analysis assumes
greater significance. In the study reported here, an attempt has been made to
evaluate quantitatively the amount and distribution of information contained in
different three TM-band combinations using an O I F approach in part of the Indo-
Gangetic alluvial plains of northern India.

2. The OIF approach

The OIF value (Chavez er a / . 1982) is based on the variance and the correlation
among the different bands. I t weighs the variance of the individual bands by using
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

their standard deviations and the correlation between the bands by their correlation
coefficients. The O I F is computed for each of the band combinations by dividing the
sum or the standard deviations of each of the three-band components by the sum of
the absolute value of the correlation coefficients computed for the same three bands
taken two at a time.

where SD,is the standard deviation for band i, and ICCjI is the absolute value of the
correlation coefficient between two of the three bands.
The three-band combination having the largest O I F is then selected for colour
compositing because it should display the most information with the least amount of
duplication. Often, three-band combinations that are within two to three rankings of
each other appear similar in colour composite form because there is little difference
in their total information content (Chevez er a/. 1984).

3. Test site description

Covering a geographical area of 78.38 km2, the area is defined by geo-
coordinates 76'14'-77"13'E and 2g00'-2Y59'N and forms part of Karnal district,
Haryana State (figure I). Geologically, it consists of alluvium from the Himalayas
transported by the river Yamuna and its tributories. Physiographically, it is a part of
the lndo-Gangetic alluvial plain. The climate is semi-arid and sub-tropical with a
mean annual precipitation of 700mm and a mean annual temperature of 24.3"C.
Soil temperature and moisture regimes may be classified as ustic and hyperthermic,
respectively according to Soil Taxonomy (USDA 1975).

4. Methodology
For determining the amount and distribution of information (variance) con-
tained in each T M three-band set, a small area (714 pixels by 714 scanlines)
representing ample variation in salinity/alkalinity was chosen. The mean, standard
deviation and correlation coefficient for the area were generated on a MicroVAX-
based digital pixel (DIPIX) system using digital data in the form of computer
compatible tape. Subsequently, the O I F values were computed using the equation
(I). For example, for the band combination 1, 3 and 5 the SD,values for bands 1, 3
and 5 are 9.74, 8.62 and 17.30, respectively (table I). The correlation coefficient
values for band I and 3, 3 and 5, and 5 and I are 0,227, 0.438 and 0.801,
 TM bands for delineation of salt-affected soils 2053

respectively. Substituting the values of SDi and CCj in the formula, the OIF value
could be calculated as:

Similarly. O I F values for other three-band combinations in the reflective bands of

TM were computed and arranged in descending order (table 2).
In order to validate the ranking of three-band combinations in terms of
information content, all 20 three-band combinations were displayed as colour
composites on the colour monitor of the MicroVAX-based DIPIX system and
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

Figure 1. Location map of the study area.

 2054 R. S. Dwiuedi and El. R. M. Rao

evaluated qualitatively based on ground truth collected by the authors. In addition,

an attempt was also made to estimate the accuracy of the delineation of salt-affected
soils in the area. Being a small area and displaying only two categories, namely salt-
affected soils and normal soils, the accuracy estimation was accomplished by
dividing the area into I mm by I mm cells and subsequently determining the number
of cells that were mapped as salt-affected and normal soils. This exercise was done
for those three-band combinations that displayed higher, middle and lower ranks in
terms of descending order of O I F values.

5. Results and discussion

The objective of the study was to find the best three-band combination of
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

Landsat-TM data. The ranking of the 20 possible three-band combinations was

done both by estimating the O I F value that ranks the multi-spectral data in terms of
scene variance; and by accuracy estimation of visually interpreted maps derived
from three-band combinations.

5.1. Information conrenr a s measured by OIF ualues

The mean, standard deviation and correlation coeficient values for reflective
bands of T M data are given in table 1 and the 0 1 F values computed using these
parameters in table 2. It is evident from table 2 that the band combination 1, 3 and 5
ranks first in terms of O I F value followed by 3, 5 and 7 and then by 3 , 4 and 5; with
an O I F value of 24.32, 23.20 and 22.49, respectively.
Since there is very little difference in the O I F values of these three-band
combinations (OIF values being 24.32, 23.20 and 22.49), the FCC print generated
from them look apparently almost similar (figures 2(a), (b) (c)). Therefore, any one
of these three-band combinations provides the maximum information and may be
used for digital analysis o r visual interpretation for deriving information on natural
resources. The observations made here are similar t o those made by Lauer and
Waltz (1983) and Hass and Waltz (1983) wherein band 3, 4 and 5 was preferred over
the normally used band 2, 3 and 4 combination. However, in contrast to the

Table I. TM six-band correlation matrix for the study area.'

Bands I 2 3 4 5 7
I 1.ooo - - - -
2 0.982 1.000 - - -
3 0.227 0.422 1.000 - -
4 0.994 0.912 0,357 1.000 -
5 0.801 0.771 0.438 0.809 1.000
7 0.903 0.809 0.360 0.850 0.919
Mean 73.17 35.52 48.09 40.71 72.10
Standard
deviation 9.74 6.57 8.62 10.16 17.30
'TM 6 thermal band not included.
 TM band for delineation of salt-affecred soils 2055

Table 2. Optimum index factor values for 20 possible band combinations using the six
reflective TM bands.

SI.
No.
Band
combination"
L+ lccjlb
j= I
imic
i = ~ OIF
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

"Six bands combined three at a time produce 20 possible combinations.

bCC;=Absolute value of the correlation coefficient between anv two of the three bands
being ded.
'SD,=Standard deviation of band i

observations made by Chavez er al. (1984) who had observed the band combination
1, 3 and 5 to rank second in terms of O I F values, in our study it ranks first.
Interestingly, the commonly used standard false colour composite made from
TM bands 2, 3 and 4 ranks twelfth with an O I F value of 14.99 indicating relatively
very low information (variance) content as compared to band combinations 1.3 and
5, 3, 5 and 7 and 3.4 and 5. The I, 2 and 4 combination with an O I F value of 9.17
ranks last indicating extremely low information (variance) content.

5.2. Discernibility of salt-affected soils

In order to evaluate the discernibility of salt-affected soils the accuracy estimates
were made for selected three-band combinations. Salient physico-chemical charac-
teristics of typical variants of salt-affected soils are given in tables 3 and 4. It is quite
interesting to note that the interpretation accuracy of salt-affected soils was found to
be maximum a t 97.6% in the case of the band combination 1.3 and 5 that happens
to rank first in terms of O I F values. Contrastingly, this trend does not hold good for
the band combination 1, 2 and 7 that ranks last but one in terms of 0 1 F value while
it provided fairly accurate mapping of salt-affected soils with an interpretation
accuracy of 96.2%.
In another extreme case, the band combination 3,4 and 5 that ranked third was
found to provide very little contrast between salt-affected and normal soils, thereby
 2056 R. S. Dwiuedi and B. R. M. Rao
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

Figure 2. FCC prints from (a) bands 3, 5, 7, (b) bands 1, 3 and 5, ( c ) bands 3. 4 and 5,
(d)bands 4, 3 and 2, (e) bands 1 , 2 and 7, all in red, green and blue, respectively.
 TM bands for delbreation of solr-affected soils 2057

it could afford very poor accuracy, i.e. 65.6%. Further, it is surprising to note that
the normally used band combination 2, 3 a n d 4 ranked lowest with a n interpretation
accuracy of 55.8%. The band combination 3, 5 and 7 that ranked second in terms of
OIF value, however, is very close to 1 , 3 and 5 with a n interpretation accuracy value
of 96.1%.

Table 3. Morphological and physical properties of typical salt-affected soils.

Depth Sand Silt Clay Clay Reaction

Horizon (cm) Colour (X) (%) (%) Texture cutans with HCI

I . Flne loomv, strongly sodic. Tvpic NorrusroIJs

All 0-7 2.5Y 5,512.5 80.0 12.0 08.0 - Strong
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

- Strong
Patchy, thin Moderate
Thin. Strong
continuous
Patchy, thin Strong
-
Strong
Strong
2. Coarse loonry, moderorely sodic, Typic NarrusIuIJs
All 0-2 2.5Y 5-512.5 79.4 12.6 08-0 - Strong
- Strong
Patchy. thin Strong
I Patchy, thin Strong
SI Patchy, thin Strong
sil - Violent
Is - Violent

Table 4. Chemical properties of the lypical pcdons.

pH EC
(1:2 soil il:2 soil
Depth 0.C water water CaC03
Horizon (cm) ratio) ratio) (x) ESP SA R

I . Fine loon^?. srrongly sodic, v p i c NorrusraIJs

All 0-7 0.14 . 10.05 5.1

2. Fine loamy, nroderorely sodic, Qpic Nurrusrolfs

All 0-2 0. i 09.8 11-4
A12 2-392 0.1 09 .95 1.62
0211 32-46 0.08 9-85 1.26
B22t 46-75 0.09 9.80 4.5
 2058 TM hands for ~ l ~ l i n e a r i oofsalt-uflected
n soils

6. Conclusion
The band combination 1, 3 and 5 was found to be the best amongst all the 20
three-band combinations of the reflective T M bands for delineating salt-affected
soils. This combination ranked first in terms of O I F values a s well a s the accuracy of
mapping salt-affected soils. T h e other band combinations, however, did not show
any consistency with respect to the correspondence between image variance a s
measured by O I F value and the mapping accuracy. As there is mixed correspon-
dence between the ranking of different three-band combinations in terms of OIF
value and the accuracy estimates obtained from a n objective comparison of
delineated salt-affected category and ground truth measurements, there is a need to
extend such study in other areas having salt-affected soils and also normal soils with
different lithologic, physiographic and vegetation cover conditions.
Downloaded by [Moskow State Univ Bibliote] at 21:44 27 November 2013

Acknowledgments
T h e authors are indebted to Prof. B. L. Deekshatulu, Director, a n d D r D. P.
Rao, Mission Director a n d G r o u p Head, Applications-I, National Remote
Sensing Agency, Hyderabad for providing necessary facilities and evincing keen
interest during the course of this investigation. Thanks are due to Shri M. A. Fyzee
for his contribution to the computation of O I F values and accuracy estimation, and
to Shri K . Anjaneyulu for the neat typing of the manuscript.

References
CHAVEZ, P. S., Jr, BBRLIN, G. L.. and SOWERS,L. B., 1982, Statistical methods for selecting
Landsat-MSS ratios. Journal of Applied Pltorogranmterric Engineering, 8 (I), 23-30.
CHAvEz, P. S., Jr, and BERLIN, G. L., 1984, Digital processing of SPOT simulator and
Landsat-TM data for the SP mountain region, Arizona. Proceedings S P O T Sym-
posium, May 1984, Scousdale, Arizona.
CHAVEZ, P. S. Jr, 1984, Digital processing techniques for image mapping with Landal-TM
and SPOT simulator data. Presented at the Eighteenth International Symposium on
Remote Sensing of Environment, Paris, France, 1-5 Ocl. 1984. (Ann Arbor, Michigan,
USA: Environmental Research Institute).
CHAVEZ, P. S. Ir, GUPTILL,C., and BOWEL,J. A,, 1984, Image processing technique for
Thematic Mapper data. Proceedings Spring 1984 Anmricun Sociery of Phorogrammerry
Conference, Washington. D C , pp 728-752
HAS, R. H., and WALTZ,F. A., 1983, Evaluation of Thematic Mapper data for natural
resources assessment. Proceedings Eighrh Pecora Symposiun~, Sioux Falls, South
Dakota, 1983.
LAUER,D. T., and WALTZ,F. A., 1983, Landsat-4 investigation of Thematic Mapper and
Multispectral Scanner applications. Proceedings of rhe Symposium on Landsar-4
Scientific Cl~oracrerisarion-Early resulrs. Greenbelt, Maryland, 1983.
USDA,1975, Soil Taxonomy. A Comprehensive System for Making and Interpreting Soil
Surveys. (Agriculture Handbook No 436), (Washington, D.C.: Government Printing
Office).