On Dec. 7, 2004, the Malaysian grain cargo
freighter M/V Selendang Ayu lost power and went adrift off Unalaska
Island in the Aleutians. After efforts to tow the vessel failed, it went
aground and broke apart between Skan Bay and Spray Cape the following
day. It was a disaster in the making. Onboard were nearly 425,000
gallons of intermediate fuel oil and 21,000 gallons of marine diesel
fuel. To make matters worse, nearby was a wildlife refuge—home to sea
otters, harbor seals, stellar sea lions, halibut, tanner crabs and
numerous seabird colonies, among other species.
Greenpeace, an international environmental organization that was among
the responders to the spill, offered its wildlife and oil spill
experience. To document the accident, the group turned to Applied
Analysis Inc. (AAI) for help. AAI acquired high-resolution QuickBird
satellite imagery of the accident site from DigitalGlobe (www.digitalglobe.com)
and processed one of the scenes from Dec. 13 shortly after the spill
began.
A Dark Challenge
Extracting useful information from the image was a challenge because of
the low light levels encountered that far north in December. AAI,
creator of the Subpixel Classifier module in ERDAS IMAGINE image
processing software from Leica Geosystems (www.gis.leica-geosystems.com),
used another one of its modules, Image Calibrator, to automatically
calibrate the dark four-band multispectral image to reflectance units.
This removed the degrading effects of haze and atmospherically
attenuated incident and emergent solar radiance, resulting in a “cleaned
up” calibrated reflectance image. Subpixel Classifier then could derive
a spectral signature of the oil from the calibrated reflectance image,
and classify the image to quickly size up the spill.
The spectral signature of the oil was derived from the image using
Subpixel Classifier’s Signature Derivation module. The training set
comprised 54 image pixels from an obvious streamer of oil that hadn’t
yet dispersed. The reflectance image then was classified using Subpixel
Classifier to determine the oil’s characteristics and spatial extent.
Recognizing that the oil was likely to be in a
variety of physical states across the scene, Subpixel Classifier had
to accommodate significant variation in the oil’s signature
properties. Using the same signature, the image was classified six
times, each using a different classification tolerance. Each
tolerance allowed progressively greater spectral deviation from the
signature characteristics of the streamer of oil. This approach
yielded three general classes of oil-spill material.
Final Results
The Subpixel Classifier results clarified the spill’s initial
impact. The results confirmed officials’ initial suspicions that
most of the oil that emerged from the freighter did so during the
first five or six days after grounding. Although the crews had a
hard time seeing oil on the dark, rough waters from aircraft and
helicopter over-flights, officials knew it was there; they could see
the effects along the shoreline. Contaminated areas along a large
expanse of shoreline extended well beyond the local bays and
inlets—occurrences of black oil (on and off the beach), dull sheen,
silver sheen, tar patties and mousse (a water-in-oil emulsion). Some
of the oil along the shoreline had been beaten to froth by the heavy
surf.
Subpixel Classifier filled in more details, particularly in the open
waters. Fortunately, over-flights were made on Dec. 12 and 13—close
to the time of the satellite image acquisition—and numerous sites
with observed contamination correlated with Subpixel Classifier
detections. This allowed researchers to tentatively identify the
three classes of materials. The Subpixel Classifier results
confirmed what the over-flights already had discovered, and they
showed oil that was still in the middle of the bay affecting the
fishing areas and heading toward sensitive shoreline areas.
As detailed in the image, streamers of black oil and mousse emerging
from the broken hull of the freighter were detected as the class
shown in magenta. Also evident was a streamer of the soybean cargo
extending from the broken hull to the shoreline. Subpixel Classifier
showed there was little mixing of the oil with the soybeans, which
was consistent with ground observations. The streamers of oil were
consistent with past observed spill characteristics of heavy refined
products. They tend to be persistent and slowly weather to mousse.
After several days they begin to break up into pancakes (patches of
oil ranging from meters to hundreds of meters across) and tar balls
(patches of oil less than a meter in size). Because of their
persistence, the tar balls can have impacts on shorelines a long
distance away. They also tend to show up without warning, because
they are difficult to see and track in the water.
Dominating the bay were extensive sheens (expansive layers of oil). One
class (shown in orange) was correlated with field-reported occurrences
of dull sheens with mousse, and the other (shown in gold) with silver
sheens. The dull sheens generally comprise “brown oil,” i.e., layers of
water-in-oil emulsion, frequently embedded with tar balls and mousse.
They are common derivatives of intermediate fuel oil spills, harmful to
the environment and the prime targets for skimming operations. The
silver sheens generally are associated with thinner layers. Although
some of this latter class of detected material may have been silver
sheen in the form of thin layers, a disturbing but real possibility was
that much of it was simply a more dispersed version of the menacing dull
sheens. It’s common under low light conditions for certain dull sheens
to be misidentified in the field as silver sheens.
Subpixel detections of black oil/mousse were used to distinguish between
the two types of sheen. The area to the south of the accident site,
between the soybeans and oil streamers, contained few subpixel
detections of the black oil/mousse. These were likely silver sheens of
concentrated diesel fuel. However, extensive subpixel detections of
black oil/mousse in the silver sheen areas in the bay to the east of the
accident site indicated that these sheens were apparently not thin-layer
silver sheens. Instead, they were simply more dispersed versions of the
dull (more compressed, thicker) sheens. This creation of dispersed and
concentrated layers of “brown oil” is a frequent consequence of wind
drift and surface currents as a spill progresses, and it produces the
kind of spatial distribution of sheens seen in the classification
results (more dispersed sheens trailing more compressed sheens). The
findings were alarming. There was an imminent threat to commercial and
Alaska native fishing grounds and large expanses of wildlife habitat.
By February, the Alaska Department of Environmental Conservation
reported widespread contamination. Shoreline clean-up operations in
Portage, Makushin, Skan, Humpback, Anderson and Cannery bays alone
retrieved more than 38,300 bags (nearly 640 cubic yards) of oil.
Although clean-up crews were able to capture 29 birds, and successfully
clean and release 10 of them, they discovered more than 1,600 birds and
six mammal carcasses. Officials continue to assess the impact on
fisheries.
A Silver Lining
Although there is little to cheer about following such a tragic event,
there is a silver lining. The quick response by DigitalGlobe to acquire
high-resolution QuickBird imagery only a few days after the accident,
combined with the creative quick turn-around processing by AAI, provided
a remarkable demonstration of a new level of situation awareness now
possible for oil spill and other disaster response planning. The
accomplishment was particularly notable considering the poor sea state
and low light level conditions that foil conventional disaster-response
surveillance and reconnaissance means for sizing up spills. Particularly
intriguing was the level of information produced about the spill from
this one image alone. Other images have been acquired, and the combined
information promises to add to researchers’ understanding of these kinds
of events, and how to more effectively respond to them in the future.
