Escolar Documentos
Profissional Documentos
Cultura Documentos
Determining the Extent of the Oil Spill in the 1991 Persian Gulf War in Ad Daffi Bay of the Persian Gulf
Using Density Slicing and Masks
Tori Niewohner
Abstract
From January 1991 through May 1991 a minimum amount of 4.1 million barrels of oil leaked
into the Persian Gulf during the Persian Gulf War (Baumann 2001). In this lab, I used a Landsat image
taken on February 16, 1991 to analyze the extent of the oil spill. After creating a density slice of TM
Band 5 to distinguish between water, light oil, heavy oil, and land, I realized that much of the land was
being misclassified as oil. To correct for this I created a land/water mask of TM Band 4 to distinguish
between land and water. I multiplied the mask by TM Band 5 and overlaid the density slice to get a
pseudo-color image which more accurately represents the extent of the oil spill. Although some areas of
land are displayed as oil, this can be explained by the counterclockwise current of the Bay and the
mangrove forests along the coasts. Although the creation of the density slice and land/water mask was
somewhat subjective and I am not sure how quantitatively accurate the estimates in Table 1 are, the
Methods
All of the methods that follow and the image used were provided by Dr. Wallin on his website
(Wallin 2016). Additional background information and an example of how the lab was to be carried out
were described in Paul R. Baumanns Environmental Warfare: 1991 Persian Gulf War" (Baumann 2001).
The image I used in this lab was a Landsat image from the Persian Gulf in the 1991 Gulf War
focused on Ad Daffi Bay taken on February 16, 1991. The pixel size of the image is 30 meters by 30
I began by opening the image in ENVI and displaying it in true color by loading TM Bands 3, 2,
and 1. After switching the image to grey-scale, I viewed each TM band and settled on TM Band 5 as it
2
gave the best distinction between oil, water, and land. I looked at the histogram for TM Band 5 which
displayed digital numbers (DN) representing the brightness of each pixel on the x-axis and frequency on
the y-axis. Using the Cursor Location Value tool and the histogram which displayed three distinct
bumps, I created ranges of DN values for water (0-25), light oil (26-50), heavy oil (51-75), and land (75-
182). I made a density slice of the image with the ranges I decided on and assigned a color to each
TM Band 4 was the best to differentiate between water and land, so I determined that a DN of
22 was a rough division point between land and water. Then I built a mask where all the water had a DN
of 1 and all the land had a DN of 0. Now that the land and the water could be appropriately
distinguished through the mask, I multiplied TM Band 5 by the mask. I then overlaid the result (TM Band
5 with the mask) with the previous density slice. However, I had to alter the density slice because water
was listed as having a DN of 0, but now it was land which had a value of 0. After applying the altered
density slice, I arrived at a final image. The last step I took was to calculate the area of each cover type
Results
The initial true-color image of Ad Daffi Bay made it difficult to distinguish between land, water,
light oil, and heavy oil (Figure 1). Once I switched to grey scale and clicked through each TM Band, I
found TM Band 5 to display the best distinction between land, water, and oil because the oil was bright
enough to clearly see (Figure 2). The density slice that I created using my DN ranges made a pseudo-
color image that, while distinguishing between the cover types decently, seemed incorrect because it
misclassified much of the area which I knew to be land, as oil (Figure 3). Since land was not being
displayed accurately, I viewed TM Band 4 which displayed a clear distinction between land and water
(Figure 4). The mask I created using the cut-off between land and water ended up with a DN of 0 for
land and 1 for water (Figure 5). Multiplying TM Band 5 (Figure 2) and the land/water mask (Figure 5) by
3
one another resulted in an image which, after having a density slice applied to it, appeared more
accurate as it did not classify as much of the land as oil. (Figure 6).
In TM Band 5, 58% of the image was water, 6% light oil, 15% heavy oil, and 21% land (Table 1).
In Masked TM Band 5, 54% of the image was water, 2.5% was light oil, less than 0.5% was heavy oil, and
43% was land (Table 1). The area displayed as oil decreased by about 18% between TM Band 5 and
Masked TM Band 5, which was what I was attempting to do (Figure 7). The land subsequently increased
by 22%, mostly from the 18% decrease in oil and slightly from the 4% decrease in water (Table 1).
Figure 1. A true color Landsat image of Ad Daffi Bay during the Persian Gulf War taken on February 16,
1991 displaying the oil spill.
4
Figure 2. TM Band 5 of a Landsat image of Ad Daffi Bay during the Persian Gulf War taken on February
16, 1991 displaying the oil spill. This band allowed the clearest distinction between oil and water.
Water: DN 0-25
Light Oil: DN 26-50
Heavy Oil: DN 51-75
Land: DN 75-180
Figure 3. A pseudo-color Landsat image of Ad Daffi Bay taken during the Persian Gulf War on February
16, 1991. It displays the oil spill and was created by overlaying a density slice on top of the image.
5
Figure 4. TM Band 4 of a Landsat image of Ad Daffi Bay during the Persian Gulf War taken on February
16, 1991 displaying the oil spill. This band showed the sharpest distinction between land and water.
Figure 5. A land/water mask created using TM Band 4 (Figure 5) from a Landsat image of Ad Daffi Bay
taken during the Persian Gulf War on February 16, 1991. This mask displays all water with a DN of 1 and
all land with a DN of 0.
6
Water: DN 1-25
Light Oil: DN 26-50
Heavy Oil: DN 51-75
Land: DN 0 and 76-180
Figure 6. A second pseudo-color Landsat image of Ad Daffi Bay taken during the Persian Gulf War on
February 16, 1991. It displays the oil spill and was created by multiplying the land/water mask (Figure 6)
by TM Band 5 (Figure 2) and then overlaying a density slice. This image was the final image created and
represents the oil spill the most accurately.
20
Percent (%)
15
10
0
TM Band 5 Masked TM Band 5
Figure 7. A graph showing the percentage of the image which was displayed as oil in the Landsat image
of Ad Daffi Bay taken during the Persian Gulf War on February 16, 1991 displaying the oil spill.
7
Table 1. A table comparing TM Band 5 and Masked TM Band 5 of a Landsat image of Ad Daffi Bay taken
during the Persian Gulf War on February 16, 1991. Each cover type has information on the number of
pixels it takes up, the area, and the percent of the total image.
TM Band 5 Masked TM Band 5
Cover Area Percent Cover Area Percent
Pixels Pixels
Type (hectares) (%) Type (hectares) (%)
Water 151,311 13,618 58 Water 142,503 12,825 54
Light Light
16,853 1,517 6 5,822 525 2.5
oil oil
Heavy Heavy
39,292 3,536 15 1,484 133 <0.5
oil oil
Land 54,688 4,922 21 Land 112,335 10,110 43
Discussion
The image I began with (Figure 1), was surprisingly difficult to distinguish. Without the
information that an oil spill had occurred, I would not have known that it had. Remote sensing can be
used to analyze changes in the land and water over time, but for many images, it clearly requires
analysis and some manipulation to view things clearly. Once the image was converted to grey-scale,
things became easier to distinguish, but it wasnt until I looked at TM Band 5 that I could really see the
oil. I assume that this means that the components of many Landsat images will not be apparent until
manipulated.
TM Band 5 was what I used to create a density slice because of the brightness of the oil. None of
the other bands showed the oil that well, so it was clearly the right band to choose. But, what about
other images? Will it always be simple to pick which band to use or are there times in which the bands
look similar? The density slice creation was the one part of the process that seemed a bit too subjective.
For example, when I was trying to look at the histogram as well as the values from the cursor, I wasnt
really sure how to determine the cutoff. The DN ranges that I chose worked out well, but I feel like the
results could vary widely from person to person doing this analysis. This could be a problem if you were
trying to determine the extent of the oil spill very accurately. The distinction between light oil and heavy
oil was also a problem as it was difficult to tell which was which on the Landsat image. The subjectivity
8
of those steps is what caused my results to be different than Baumanns. Had we both arrived at the
The first pseudo-color image (Figure 3) clearly was misclassifying land as oil as it makes it appear
as if there is oil covering large quantities of land. In fact, the oil was taking up 21% of the image! This
was the same amount as was being taken up by what was being quantified as land. In this case it is fairly
obvious that Figure 3 is not quite right. There can certainly be some oil on land, but it is unlikely it would
be so far inland. This raises a question however. What if this was a different situation, maybe one not
involving oil, where the result was inaccurate, but the viewer was unable to tell because it wasnt
something as obvious? Does this mean that this form of remote sensing analysis has a high level of
The creation of the land/water mask and the band math which multiplied the mask by TM Band
5 were steps that I initially was confused by, but now understand why they occurred. Since the initial
pseudo-color image displayed land as oil, I just had to more clearly define land, which explained the
creation of the mask. The second pseudo-color image (Figure 6) is much more reasonable. However,
there is still some oil on land! This can be explained by the current of the Persian Bay. The current
travels counterclockwise and thus would be entering this image from the top and traveling down
through the image and gradually going slightly from left to right. The locations with oil on land are
where the current would be pushing against, thus explaining why oil could have been pushed inland. Ad
Daffi Bay also has many mangrove forests which exist in shallow water, but are most likely still classified
as land in this image. So some of the oil on land may actually be in the shallow water of a mangrove
forest.
The Persian Gulf Oil Spill was very large, dumping a minimum of 4.1 million barrels of oil into the
Gulf (Baumann). The initial pseudo-color image created from TM Band 5 made it appear as if Ad Daffi
9
had tremendous amounts of oil in it as light oil was taking up 6% of the image and heavy oil was taking
up 15% (Table 1). The second pseudo-color image created from Masked TM Band 5 put light oil at 2.5%
of the image and heavy oil at less than 0.5% (Table 1). While this is still a problem, as any oil in an
ecosystem can have consequences, it is much less alarming than the percentages from the original
pseudo-color image. If this analysis was done in 1991, the results could have been used to know where
to focus clean-up efforts. They also could have been used to predict the location where oil will end up in
future oil spills and thus be prepared for where clean-up will need to occur.
The technique used in this lab is an important one as it could be applied to other oil spills that
may arise. Although it is at times subjective and I am unsure as to its quantitative accuracy, it is
Literature Cited
Baumann, P.R. 2001. Geo/SAT 2: Environmental Warfare: 1991 Persian Gulf War.
http://employees.oneonta.edu/baumanpr/geosat2/Environmental_Warfare/ENVIRONMENTAL_WARFA
RE.htm