s. Equally
5 with low
ebruary 28,
| the desert
cally active
d principal
vegetation
ime time of
irms where
vegetation
> to erosion
Vegetation,
vironment-
soils and
n arid an
itive to use
at satellite
vegetation
vegetation
ly warning
te sensing
yore useful
lites with
96).
ra] Landsat
in the arid
acted from
, that were
28, 1993,
le principal
shrubs and
vegetation
) automatic
differences
f the State
wait-Saudi
an Shield,
ment. with
s very hot,
especially July and August, with a mean temperature of 37.4?C
and maximum mean temperature of 45°C. The average total
precipitation is approximately 100 mm/y and the rate of
evaporation is 16.6 mm/d. The winds in the area are from the
northwest and, to a lesser extent, from the southeast, and have
a pronounced influence on the oceanographic and
sedimentological processes. Dust and dust storms, locally
known at 'toze', occur in the Kuwait region throughout the year
but are more frequent during the spring and summer months,
je, March to August. Kuwait is a low-relief desert country
with a maximum relief of approximately 125 m. The land
surface slopes gradually northeastwards with an average -
gradient of approximately 2 m/km. Kuwait’s desert can be
divided into four main provinces, namely: (1) Al-Dibdibba
gravely plain; (2) southern desert flat; (3) coastal flat; and (4)
coastal hills (Khalaf et al., 1984). The surface is overlain by
several recent sediment deposits that include, eolian, residual,
playa, desert plain, slope, and coastal deposits (Figure 1).
Eolian deposits are the most predominant and account for 50%
of the surface deposits. Observed surface outcrops consist of
clastic deposits which are locally called the Kuwaiti Group,
and range in age from Miocene to recent.
2. METHODOLOGY
The Landsat TM digital images were geometrically registered
and radiometrically calibrated to each other to facilitate their
comparison. The NDVI images for the two dates were used as
input to an automatic change detection procedure using
selective principal component analysis. Landsat TM band 3
images were further enhanced to improve the overall
information content in the low frequency end of the images.
Image processing was performed on a UNIX-based SUN
workstation with PCI’s EASI/PACE image processing software
at Kuwait Institute for Scientific Research’s (KISR) Remote
Sensing Laboratory.
Radiometric calibration of satellite imagery is critical in
multitemporal and change detection studies due to the
degradation by haze. Haze, caused by scattering of
electromagnetic waves, increases the overall radiance of an
image thereby reducing the image contrast as well as degrading
the spatial resolution of the sensor. The haze effect is much
more severe on the shorter than the longer wavelength. The
true reflectance, which is characteristic of a target, is modified
by the atmosphere through, (1) atmospheric scattering; (2)
attenuation, i.e., absorbing of the energy reflected by the Earth;
and (3) ground scattering. For quantitative analysis or
comparison of multitemporal images, it is imperative that the
gray level reflect the true spectral reflectance of the target area,
Le, the elimination of the atmospheric influence.
The atmospheric correction program, ATCOR (Richter, 1991),
that is part of PCI software package was used to derive the true
Spectral image of the TM bands used in this study. The
Program employs the LOWTRAN-7 atmospheric model code
and adjusts each pixel to account for atmospheric influences
such as albedo, optical depth and aerosol concentrations. The
Program incorporates a catalogue of atmospheric functions to
calculate ground reflectance values for cloud-free images. The
catalog compiled for Landsat multispectral scanner (MSS),
; and SPOT consists of aerosol types and concentration,
Zenith angles, sensor view angles, and ground altitudes for
different standard atmospheres. The program specifically
employs the following three steps;
ATCORO ......... Determines ground visibility.
ex AICORI >... Calculates reflectance image with no
adjacency effect.
9e ATCOR2........... Calculates reflectance image with
adjacency effect.
Several change detection technique applications to satellite
digital data have been reported in the literature. Some of these
include image difference, ratioing, principal component
analysis, and selective principal component analysis (Jensen
and Toll, 1982; Singh, 1989; Chavez and Kwarteng, 1989;
Chavez and MacKinnon, 1994). The applied radiometric
correction can either be absolute or relative depending upon
the intended use. In absolute calibration, the satellite digital
number (DN) is converted to ground reflectance, whereas in
the relative sense, the same DN in two images represent the
same reflectance. A third type of calibration is a hybrid
between the absolute and the relative methods. Application of
relative calibration is only meaningful if the DN changes
between two images is statistically small and do not alter the
overall dynamic range of the images. Such conditions are
usually observed in arid and semi-arid environments (Chavez
and MacKinnon, 1994).
In this study, we applied selective principal component in the
same manner as used by Chavez and Kwarteng (1989), where
only two bands from the same image are used as input to
principal component analysis (PCA). Principal component
analysis is a statistical technique that rotates the axes of a
multi-dimensional image space in the direction of maximum
variance. The generated components or axes, that are simple
linear combinations of the original image data, are orthogonal
to each other and, thus, have no further mathematical relations.
Eigenvectors are used as multiplication coefficients or loadings
in the PCA for each pair of input bands. By being selective
and using two images or bands as input to PCA, information
that is common to the two images/bands, typically topographic,
albedo or reflectance, is mapped to the first component (PC1),
whereas information that is unique to either of the input
images is mapped to the second component (PC2) (Chavez and
Kwarteng, 1989). Consequently, PC2 maps the spectral
contrast of two bands from the same image, or the temporal
contrast when the images were taken at different times. In
transforming the data into a new coordinate system, the
selective principal component technique performs a first-order
relative image to image calibration and, therefore,
automatically eliminates most low-frequency noise between the
two images, which invariably includes atmospheric and solar
effects. Such noise reduction capabilities enhance the quality
of images generated. A major advantage of selective principal
components over the traditional PCA analysis, with several
bands or images as input into PCA, is that the analysis and
interpretation of the results are relatively easy and straight-
forward (Chavez and Kwarteng, 1989).
3. DISCUSSION OF RESULTS
Optical satellite images of desert environments can often be
dull with little contrast because of the overshadowing of the
high frequency information in the dominant low frequency
399
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
