J. Agric. Sci. Technol. (2004) Vol.

6: 31-40

Behavior of Landsat Thematic Mapper Wavebands on Lut

Desert (Iran)

S. K. Alavi Panah1 , Ch. B. Komaki2 and M. Ghadiri Masoum3

ABSTRACT

To study the capabilities of remotely sensed data in discriminating land cover types and
geomorphologic features of desert, the southeast of Lut Desert was selected. In this study,
seven bands of Landsat Thematic Mapper (TM), data and other sources of information
such as topographic maps (1:50,000), aerial photos (1:20,000) and field data were used.
The visual image interpretation was performed based on photomorphic unit analysis and
interpretation keys. To classify the images, 10 training classes were used for maximum
likelihood classification algorithm. Then the classified images were assessed with the test
areas and the overall accuracy was about 92 percent. Based on the obtained results it may
be concluded that Landsat TM thermal and reflective bands could be useful for studying
Lut Desert conditions especially in the Yardang and sand dune regions. From the results
obtained we may also conclude that such a hyper arid climatic and bare land are suitable
conditions for a better understanding of the behavior of TM wavebands on different
surfaces and soil conditions.
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

Key words: Desert, Image classification, Lut Desert, Sand dune, TM wavebands, Yardang.

INTRODUCTION the basis of remote sensing. Understanding

weathered material and soil spectra princi-
Recently, considerable effort has been put ples is crucial to the application of remote
into the development, operation, and use of sensing procedures for soil and geomorpho-
spaceborne image spectrometers. These ad- logic studies. Therefore, in this study we
vances in technology can provide a near- will make our attempt to study behavior of
laboratory-quality spectrum of every pixel in TM reflective and thermal bands on Lut De-
the image and very soon will permit remote sert features.
sensing of soil and weathered material in Lut Desert with an extent of about 80000
deserts to be improved. Experience has Km2 is the largest playa in Iran and includes
shown that many earth surface features of great diversity of hydroaeolian processes
deserts can be identified, mapped and stud- with a very interesting pattern of landforms,
ied on the basis of their spectral characteris- such as Yardangs which are wind-abraded
tics of the particular features under investi- ridges of cohesive materials. Yardangs are
gations (Alavi Panah, 1996; Alavi Panah, et most commonly developed in soft lithology,
al. 2001). Therefore a better understanding such as lacustrine sediments. The Lut mega
of the behavior of different wavelength re- Yardangs have been cut recently in pleisto-
gions on different materials and surface cene lake deposits (Bobek, 1969). To the
authors knowledge not much research about
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

conditions in deserts may increase the effi-

ciency of the study of land cover types on the striking features of Lut Desert has been
performed. Application of remotely sensed
_____________________________________________________________________________________
1
College of Geography, University of Tehran, Islamic Republic of Iran.
e-mail: Salavipa@chamran.ut.ac.ir
2
Desert Zone Management, Islamic Republic of Iran.
3
Collage of Geography, University of Tehran, Tehran, Islamic Republic of Iran.

31
 _______________________________________________ Alavi Panah, Komaki and Ghadiri Masum

data for geomorphologic mapping in such ized by an extremely arid climate, with ex-
regions requires a knowledge of the influ- cessive summer heat, and winter tempera-
ence of some land cover types, such as de- tures below freezing point and an annual
sert varnish on the surface reflectance in or- rainfall of less than 50 mm. Alavi Panah, et
der to yield more reliable results. Spectral al. (2002) has shown that some part of Yar-
analysis shows the paucity and Mn/Fe pro- dangs contain a great amount of gypsum and
portion in composition, and the wavelength salts on the surface or very close to the
under consideration control the effect of de- eroded surfaces. The study area includes two
sert varnish on Landsat TM data (Keneam, main regions of Yardang and sand dune
2001). To assist in management of the de- which are rare and striking features (Figure
sert, information is required about land 2). The hyper-arid south-east of the Yardang
cover types especially eroded land, sand region consists of large gravel plain and de-
sheet, sand dune and salt affected land. sert pavement, sand dune, salt affected wet
Drake (1995) found that remote sensing may lands. Sand regions in the east of Lut Desert
be a useful tool for locating economic form the largest body of sand in Iran with
evaporative deposits. Remote sensing has probably the greatest diversity to be found.
shown its greatest value where fieldwork has
been difficult. Therefore, in this study it was
MATERIALS AND METHODS
attempted to study the eastern part of the
poorly accessible Lut Desert using remotely
sensed data. In fact, the main purpose of this In this study, the available seven Landsat The-
study is to evaluate the capability of Landsat matic Mapper bands dated June 1989 were used.
The number of row and column are 1960 and
TM data for studying land cover types and
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

2028 pixels respectively. For visual image inter-

the geomorphology of deserts. pretation of the whole part of Yardang area,
Landsat TM bands 7-4-2 combination (TM7-red,
STUDY AREA TM4-green, TM2-blue) dated 1987 and 1988
were used for land cover discrimination.
The spatial resolution of all bands (except
The study area that covers about 25439 band 6) is 30 m, therefore TM6 was only
square Km was located in Lut Desert (Fig- used for visual interpretation by using
ure 1). The Lut Desert, particularly its cen- diffent false colure composites. This study
tral part and Yardang region, is character-
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

Figure 1. Location map of the study area

32
 Thematic Mapper Wavebands on Lut Desert (Iran) ________________________________

(a)

(b)
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

Figure 2. The terrestrial photographs of two land cover types in the area, a) Yardangs
with gully erosion and b) mobile sand in between the broad ridges of Yardang.

was concentrated on using Landsat TM data binations. The FCC,s were evaluated visu-
because ground resolution of 30 m seems ally based on the information content on
suitable for studying large areas in this field work and image interpretation.
study. Other information sources, such as Based on our experience, the land cover
topographic maps (at the scale of 1/50,000), types were defined (Figure 2) and the train-
geologic map (at the scale of 1/250,000),) ing sets were identified and spectral patterns
and aerial photos (at the scales of 1/50,000 of different types of land covers were gener-
and 1/20,000) were applied. The fieldwork ated. Different aeolian features such as Yar-
as one of the most important steps was car- dangs, desert pavements, sand dunes, sand
ried out in autumn season of 2000. The sheets and erg were defined (Cooke and
methodology used comprises of 1) correla- Goudi, 1993) as indicators of the wind effect
tion between TM bands, 2) image processing on the morphology of the study areas. Yar-
techniques, 3) two dimensional feature space dangs a term used for abraded ridges of co-
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

(FS) analysis and 4) image classification and hesive materials that occur in unidirectional
accuracy assessment. In this study, the three- regime were also defined. Desert pavements,
band combinations were ranked based on soils containing coarse fragments on the sur-
Optimum Index Factor (OIF) and the differ- face and salt affected lands with two differ-
ent False Color Composites (FCC, s) were ent surface conditions of salinity and salt
created. The application of the OIF criteria crust, were defined. Wet lands the places
to seven TM bands resulted in 35 band com- where water table lies at shallow depth were

33
 _______________________________________________ Alavi Panah, Komaki and Ghadiri Masum

also defined. The class separability analysis Table 1. The rank of 35 possible band com-
was carried out by computing the statistics binations of the TM bands based on their
(mean and standard deviation). Based on the OIF values.
results obtained, TM reflective bands 1, 4, 5, Rank Band combination OIF value
7 were selected as the most informative. TM 1 1-5-6 24.12
reflective bands and then all data sets were 2 1-6-7 21.60
classified using a per pixel maximum likeli- 3 3-5-6 21.50
hood classification algorithm and finally the 4 5-6-7 21.25
5 4-5-6 21.09
land cover map was produced. 6 1-5-7 20.80
7 1-3-6 20.53
8 1-4-6 20.09
RESULTS AND DISCUSSION 9 1-4-5 19.96
10 2-5-6 19.76
11 3-6-7 19.33
Color Composites 12 1-2-5 18.79
13 1-3-7 18.48
14 3-5-7 18.46
In order to select the most suitable three
15 4-6-7 18.34
band combinations for creating color com- 16 3-4-6 18.28
posites, the OIF criteria was applied to 7 TM 17 3-4-5 17.92
bands (Table 1). It is evident from Table 1 18 1-4-7 17.87
that the band combinations 1, 5, 6 and 1, 6, 7 19 1-3-5 17.79
20 1-2-6 17.73
ranks first in terms of the OIF value. In this 21 4-5-7 17.59
study, the color composites were created and 22 2-3-5 17.32
were displayed on monitor and their capa- 23 2-5-7 17.28
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

bilities were compared for discrimination of 24 2-6-7 17.14

the land cover types. Two color composites 25 1-3-4 17.13
26 1-2-7 16.81
of Landsat TM scene over a portion of Lut 27 2-3-6 16.70
Desert are shown in Figure 3. In image a, 28 2-4-5 16.69
TM band 6 is displayed in red, band 4 in 29 2-4-6 16.07
green and band 2 in blue. In contrast, image 30 1-2-3 15.86
31 3-4-7 15.84
b shows the same region, but uses TM 32 1-2-4 15.50
band 7 in red, band 4 in green and band 2 in 33 2-3-7 15.28
blue. Comparison between these color com- 34 2-4-7 14.61
posites reveals that color composite a exhib- 35 2-3-4 14.26
its wet land, sand dune and sand sheet better
than color composites 7, 4, 2 (a) and the re-
gions rich in sand sheets exhibit the distinc- found between the TM3 and TM4 which
tive color and wet lands between the Yar- forms a soil line and a very low negative
dang and sand dune regions is shown with a correlation was obtained between TM red
high reflectance in blue and appears blue. and TM thermal band which forms a clus-
Results from visual interpretation of various tered shape. Due to the lack of vegetation
combinations of bands (FCC,s) improve our (zero vegetation cover) in the study area, the
knowledge of the spectral properties of ma- strong relationship between TM3 and TM4
terials or may provide an opportunity for leads to a very high correlation coefficient of
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

assessment of the land cover units in a re- 0.996 (Table 2). The result obtained from
gion. the high correlation coefficient between
TM3 and TM4 (r = 0.996) reveals a lack of
Soil Line vegetation that normally exhibits low reflec-
tion value in near infrared band and high
reflectance value in red band.
The highest correlation of coefficients was

34
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ] [ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

Thematic Mapper Wavebands on Lut Desert (Iran) ________________________________

Figure 3. Comparison of color composites from Landsat TM for the eastern part of Lut Desert.

35
 _______________________________________________ Alavi Panah, Komaki and Ghadiri Masum

Table 2. Correlation matrix of TM bands of southeast of Lut Desert.

Bands TM1 TM2 TM3 TM4 TM5 TM6 TM7
TM1 1.00
TM2 0.98 1.00
TM3 0.95 0.98 1.00
TM4 0.91 0.96 0.99 1.00
TM5 0.85 0.92 0.95 0.96 1.00
TM6 -0.56 -0.58 -0.55 -0.53 -0.57 1.00
TM7 0.82 0.89 0.93 0.95 0.99 -0.53 1.00

Spectral Separability fied map, randomly selected samples repre-

senting different land cover types were used
The validity of training data was evaluated randomly. Many sample points SD (from 63
from visual examination of FS and quantita- to 317) representing all types of investigated
tive characterization (Table 3). Table 3 land cover were identified on the FCC for
shows the mean and standard deviation of accuracy estimation. A one to one compari-
the training classes. Among the 10 training son was made between the ground truth data
classes, classes no. 10 (wet land) shows the and the information available in reports,
highest SD (SD = 17.8) for TM 1 which is fieldwork and maps. The overall accuracy
due to the heterogeneity of the soil surface, was calculated after generating a confusion
moisture and salinity variation and litholo- matrix. Table 4 indicates the error matrix
gies. This training class shows the lowest resulting from the maximum likelihood clas-
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

SD (SD = 6.6) in TM reflective band 7. sification. In this study, areas different from
Spectral separability helps to refine the training samples are used for classification
digital classification and it has long been accuracy. As Table 4 indicates the pixels
used as an index of the efficiency of re- that are properly classified are located along
motely sensed data to distinguish various the major diagonal of the error matrix. The
terrain features and phenomena. Figure 4 result of percentage accuracy per category
shows the feature space between TM bands. from class no.1 (Yardang) to class no.10
A close look at this FS indicates that the be- (wet lands) shows that the accuracy of class
havior of TM bands is different from the no.1(Yardang) and class no.2 (salt effected
view-point of spectral separability. For ex- land) is 85.04 and 80.13 percent respectively
ample, FS bands1 and 3 indicate a fairly which is relatively low. That is due to great
good spectral separability of training classes, confusion between these two classes. The
while FS bands 1 and 2 show confusion be- reason for confusion between Yardang and
tween the training classes. It means that the salt affected land may be due to the very
overall separability of classes is poor in similar surface condition of some Yardangs
TM1 and TM2 data set. with a high amount of salt and gypsum.

Image Classification and Accuracy CONCLUSION

Assessment
This study has demonstrated the utility of
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

The training samples which are used to es- TM data for mapping Lut Desert features
timate the statistical characteristics of the including Yardang, sand dune land cover
classes were used to classify the TM scene. types with relatively high spatial resolution
The result of the maximum likelihood classi- satellite data. Lut Desert land cover types,
fication is shown in Figure 5. such as Yardangs, sand dunes and salt
For the quantitative estimates of the classi-

36
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ] [ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

37
Thematic Mapper Wavebands on Lut Desert (Iran) ________________________________
 _______________________________________________ Alavi Panah, Komaki and Ghadiri Masum
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

Figure 4. The feature space between TM bands.

38
 Thematic Mapper Wavebands on Lut Desert (Iran) ________________________________

Figure 5. Land cover classification image of the southeast of Lut Desert based on the
TM band combination 1, 4, 5 and 7.

affected lands carry unique and important and Yardangs and therefore it may provide
information about many desert characteris- an opportunity for delineating a land cover
tics. This study also showed that, in some unit. Based on the results obtained from FS
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

cases, only a slight spectral difference could analysis, we may conclude that hyper arid
be effective in discriminating desert land climatic conditions are ideal for a detailed
cover types on the basis of their spectral sig- mapping of land cover types and a better
natures. To achieve a high classification ac- understanding of the behavior of TM bands.
curacy, extensive fieldwork and aerial pho-
tos would be very effective. The result of
ACKNOWLEDGMENT
visual interpretation of various band combi-
nations may improve our knowledge of the
spectral reflectance of wet lands, sand dunes The authors thank the Vice Chancellor for

Table 4. Error matrix resulting from maximum likelihood classification using TM reflective bands
1,4,5,7.
no. Y SL1 DP SL2 W1 SSH E1 SD E2 W2 Total Accu
1 Y 108 17 2 127 85.04
2 SL1 57 254 1 4 1 317 80.13
3 DP 167 1 4 172 97.9
4 SL2 1 217 218 99.54
5 W1 46 2 4 52 88.46
6 SSH 197 9 206 95.63
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

7 E1 176 1 177 99.44

8 SD 7 226 15 248 91.13
9 E2 1 1 1 224 227 96.68
10 W2 1 1 61 63 96.83
Total 173 271 168 223 47 200 177 243 240 65 1807 93.20
*Accu 64.43 93.73 99.40 97.31 97.87 98.50 99.46 93.00 93.33 93.85 92.89

*Accu =Accuracy

39
 _______________________________________________ Alavi Panah, Komaki and Ghadiri Masum

Research, University of Tehran for partial 5. Cooke, R. U., Warrant, A. and Goudi, A. S.
financial support of this research. 1993. Desert Geomorphology, London Press
Limited.
6. Drake, N. A. 1995. Reflectance Spectra of
REFERENCES Evaporate Minerals; Applications for Re-
mote Sensing. Int. J. Remote Sens, 16(14):
1. Alavi Panah, S. K., Barzegar F., Ahmadi, H. 2555-2571.
and Komaki, C. B. 2002. Study of Yardang 7. Hunt, G. R. and Salisbury, J. W. 1976. Mid-
Lut Desert Based on Image Processing, Uni- infrared Spectral Behavior of Metamorphic
versity of Tehran, 150 p. Rocks. Environmental Research Paper 543-
AFCRL-TR 76-0003, Hanslom Air Force
2. Alavi Panah, S. K. 1996." Study of soil sa-
linity based upon field observations, remote Cambridge Research Laboratories, USA.
sensing and GIS", Ph.D. Thesis, Gent Uni- 8. Keneam, N. H. 2001. Influence of Desert
versity. Varnish on the Reflectances of Gossans in
3. Alavi Panah, S. K., De Dapper, M., the Context of Landsat TM data, Southern
Red Sea Hillsm Sudan, Int. J. Remote Sens.
Goossens, R. and Massoudi, M. 2001, The
2(5): 1869-1888.
use of thermal band for land cover/ land use
mapping in two different environmental 9. Millington, A.C. et al. 1989. Monitoring Salt
conditions of Iran. J. Agric. Sci. Technol, 3: Playa Dynamic Using Thematic Mapper
27 -36. Data. I. E. E. E. Trans. Geosci. Remote
4. Bobek, H. 1969. Zur Kenntis der sudlichen Sens., 27: 745-761.
Lut. Mitteilungen der Ostereichinhen
geograpischen Gesselschaft, Wein 3: 155-92.
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]

()      TM     

&'() * + .&  $%# ." . , ! , .

 $,

$$  #   ! "           

'6 5  TM $ 4 ,  3  .$'1 ,/0 ,. ) +' ,- ,) % &' (%
 F  (A: E=/===)  CD9 ,(A :?=/===) &'1  78 $   .9:;
$O PNG  L M'   'K I7J 'D .4&'1 '+ 0   'GH
S  $8R; ' ; M:F A= ,'K $8R; 4C- .4&'1 Q 'D $NF  "&  0&
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

Z0  $'1   7 X Y;  $ $8R; 78 VW .$ 0&'1 F  TI0O 'UF$O
) $ X 4$ Z0 M' .$ $H  [E $O $8R; NF 4+ F  )7 PHO
 ^  ]  . ) \'   ' TM M   'O $ F I '10

40
 Thematic Mapper Wavebands on Lut Desert (Iran) ________________________________

#   "7a  `& \' 3J F  '10 3_I .$ $  W
.$ dN0/ \'    ' TM ; c$ 0& b  ' R \G 4/
[ Downloaded from jast.modares.ac.ir on 2024-05-03 ]
[ DOR: 20.1001.1.16807073.2004.6.1.7.1 ]

41

Powered by TCPDF (www.tcpdf.org)