ule lag, 3 =
tive sand
5, 11 =
%
ance
9.82
bright and dark frequency areas (Kwarteng and Chavez, in
ress). 10 maximize the spatial information in the TM data
used in this study, a high pass filter with a relatively large
kernel size of 201 by 201 pixels and a 50% addback option
was used to enhance the high frequency information in both the
bright desert and other relatively dark areas. The filtered
results were edged enhanced with a 7 by 7 filter to sharpen the
Jocal/textural information or very high frequency features
(Kwarteng and Chavez, in press). The vast improvement over
the non-processed image is more spectacular when printed on a
large scale, and also in color composites, such as TM bands 2,
4, and 7. However, due to publication restrictions, only black
and white images are shown in this paper. Additionally, image
interpretation was originally done on a scale of 1:100,000, but
reduced considerably for publication. The enhanced TM band
3 image acquired on February 28, 1993, is shown in Figure 2.
The Arabian Gulf, shown as the black, was masked and
excluded from further processing. The north-south-trending
dark area in the middle of the image is the scar from the
burned oil wells and the subsequent cleanup activities at the
Greater Burgan oil field. The Greater Burgan, consisting of
Burgan, Magwa, and Ahmadi oil fields, is the second largest
oil field in the world. During the 1991 Gulf War, the Iraqi
forces detonated several pounds of explosives laid against all
the active 810 oil wells in Kuwait. In the ensuing
environmental disaster unequaled in the world’s history, 656
oil wells were set ablaze while 74 others gushed uncontrollably
from the damaged well heads (Petroleum Economist, 1992).
During the peak of the fires, 365 burning wells were observed
at the Greater Burgan oil field (Kwarteng and Bader, 1993).
The most commonly used technique for vegetation analysis is
the NDVI, a parameter derived from the red and near-infrared
channels. The ratio is computed from TM bands as follows:
NDVI = (TM4-TM3)/(TM4 + TM3) (1)
where TM3 and TM4 are DN values in the red (0.63-0.69 um)
and near-infrared (1.55-1.75 pm) bands, respectively. The
ratio is a measure of the deviations between a vegetation
spectrum’s chlorophyll absorption minimum and the infrared
plateau and, thus, a direct indication of the amount of
photosynthetically active green biomass (Tucker and Sellers,
1986). The NDVI image computed for the two dates are
presented in Figures 3 and 4. The images were linearly
stretched to occupy the dynamic range 0-255. A histogram
matching procedure from PCI software was used to create a
lookup table matching the NDVI image of February 4, 1987, to
that of February 28, 1993. Consequently, the same DN values
in both Figures 3 and 4 represent the same reflectance values.
The gray color denotes areas with insignificant vegetation or
NDVI values. The darker than gray areas which include
standing water, coastal sabkhas, soot from oil fires, and man-
made structures (i.e., tar roads, airport runways and buildings),
have no relations with NDVI. In Figure 3, such areas are
observed mainly within the Kuwait City limits, north and
Southeast of Kuwait International Airport. The dark tones
along the coast south of Ras Al-Qulaiah represent sabkhas. The
black spot at the Wafra oil field represent an oil spill that
occurred before February 1987. For the 1993 image, such
areas include the oil lakes, soot, and tarmats found at the
Burgan oil field and, to a lesser extent, at the Wafra oil field.
Winds in the area are predominantly from the northwest and,
fo a lesser extent, from the southeast. Therefore, most of the
soot is observed southeast of the Burgan oil field. The amount
of vegetation that normally would have been found in the oil
fields and areas downwind had been reduced drastically due to
the negative effect of the burning of the oil wells. A
comparison of Figures 3 and 4 show that vegetation within the
coastal wetlands east of Wafra farms had been adversely
affected by the burning oil wells.
The lighter than gray tones represent the distribution of
vegetation or photosynthetically active areas that include both
natural vegetation and cultivated crops. The degree of
whiteness is a measure of the vegetation vigor. The extreme
white areas represent cultivated farms. In both images,
vegetation is observed mostly to the north and east of the
Burgan oil field, within the Kuwait City limits and at the
Wafra farms. The suburb of Ahmadi with several trees show
up as white on both images. The 1987 image (Figure 3) shows
that the majority of the desert lands had less vegetation
compared with the February 1993 image that exhibits increase
in greenness/biomass in both the desert and city areas.
Variation in vegetation distribution between the two images
was primarily due to climate and, more importantly, rainfall.
Rainfall in Kuwait is scanty, erratic, and fluctuates from year
to year with the main rainfall season occurring between
November to April. The total precipitation recorded at the
Kuwait International Airport Observatory from November 1986
to February 1987 was 44.9 mm. For the same period between
November 1992 to February 1993, the amount of rain recorded
was 150.2 mm. Lack of rainfall intensifies aridity and causes
degradation of natural vegetation. For most regions, shifting
sand dunes and sand sheets are incapable of sustaining plant
life. Kuwait's vegetation consists of undershrubs, perennial
herbs and spring ephemerals. The vegetation types are
controlled by four major ecosystems, i.e., sand dunes, desert
plain, desert plateau, and salt marsh and saline depressions
(Halwagy and Halwagy, 1974). The major plant communities
are: (a) Cyperus conglomeratus, (b) Rhanterium epapposum;
and (c) Hammada salicornica, with the first two being the
most predominant in the study area. Comparison of both
images show that the Wafra farms were extended further to the
east from 1987 to 1993. Furthermore, the 1993 image shows
higher plant vigor/biomass and most likely yielded abundant
crops compared with 1987.
The NDVI images of the two dates were used as input to the
selective PCA technique. The resulting image statistical
variance (Table 1) is related to the surface spectral responses
such as vegetation and soil. PCI, which represents albedo,
accounts for 80.18% of the total scene variance (excluding the
Arabian Gulf) and is composed of negative weighting for the
input bands. PC2 that maps vegetation related differences
between the two dates is 19.82% of the scene variance (Table
1). The ordering of the PCA (i.e., PC1 and PC2) is influenced
by both the image statistics and spatial abundance of surface
materials. Because the spectral property mapped into each
NDVI image are related to biomass/greenness, any changes
between the image statistics is associated with temporal
vegetation variations. From Table 1, this can be interpreted as
an increase in vegetation of 19.82% from 1987 to 1993 and,
conversely, a 19.82% decrease in biomass from 1993 to 1987.
The weighting mapped into PC1 and PC2 is influenced by the
magnitude of the standard deviation (SD) and statistical
dimensionality of the images that are related to sensor gain,
offsets, and spectral differences (Loughlin, 1991). Even
401
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
