By Beverley Adams, Charles K. Huyck and Ron
Eguchi, ImageCat Inc. (www.imagecatinc.com), Long Beach, Calif.;
Fumio Yamazaki and Miguel Estrada, Institute of Industrial Science,
University of Tokyo (www.u-tokyo.ac.jp); and Chuck Herring,
DigitalGlobe (www.digitalglobe.com), Longmont, Colo.
For several years, researchers at ImageCat Inc.
have investigated how remote sensing technologies can improve
response-and-recovery activities after major earthquakes. The
company’s latest study—exploring the use of satellite imagery for
post-earthquake analysis in Algeria—represents a milestone in the
field of earthquake research.
Disaster Strikes
The seismic event that prompted the research was a massive
earthquake that struck Algeria on May 21, 2003, measuring 6.8 on the
Richter scale. Its destructive forces were most intense in the
densely populated towns of Rouiba, Boumerdes and Thenia east of the
country’s capital, Algiers. The first priority was to assess and
limit human injuries and fatalities, a monumental effort considering
the total death toll reached 2,287, with more than 11,000 injured. A
second priority was to assess structural damage and recovery.
Throughout the region the earthquake damaged about 182,000
residential buildings and 6,200 public structures, including schools
and hospitals.
The study—the most comprehensive research on
the subject to date—was conducted jointly with the Multidisciplinary
Center for Earthquake Engineering Research (MCEER), headquartered at
the University of Buffalo, with support from Oakland, Calif.-based
Earthquake Engineering Research Institute (EERI). MCEER funded the
project as part of its mission to improve community resilience in
times of disaster. EERI provided DigitalGlobe QuickBird imagery
through its “Learning from Earthquakes” program. The high-resolution
imagery was analyzed by researchers at ImageCat, the University of
Tokyo and several other research organizations around the world
involved in the use of remote sensing for disaster response.
To evaluate the potential of satellite imagery
for assisting in damage assessment and coordinating site visits and
relief efforts, the researchers contacted DigitalGlobe to obtain
“before” and “after” QuickBird imagery of the earthquake region.
DigitalGlobe provided QuickBird data from its archive collected on
April 22, 2002—approximately one year prior to the earthquake—and
May 23, 2003—two days after the earthquake. QuickBird imagery
collected on June 18, 2003, allowed researchers to further monitor
recovery efforts.
Analyzing the Imagery
During the evaluation phase, the project researchers created
automated change detection algorithms that offered a “quick-look”
damage assessment and provided the focus for more detailed building
inspections using visualization techniques. A visual comparison then
was drawn between enlarged views of the “before” and “after”
QuickBird images, which were displayed side by side within Research
Systems’ ENVI image processing environment. The QuickBird imagery’s
detailed representation enabled researchers to readily identify
severely damaged structures. In addition to urban damage, the images
also showed the location of temporary tent camps that housed
displaced residents. Researchers hope to use such imagery in future
events to provide “real-time” assessments that will help guide the
work of field reconnaissance teams.
he research team concluded that
high-resolution satellite imagery is a highly effective and valuable
tool in the response-and-recovery phases of the emergency management
cycle for reconnaissance and monitoring recovery operations. The
following excerpts from an EERI report of the research team’s
findings focus on the analysis of “before” and “after” imagery of
Boumerdes, east of Algiers.
A Tiered Reconnaissance System The flow chart in Figure 1 illustrates how
satellite imagery can be used during response-and-recovery phases of
the emergency management cycle.
In the immediate aftermath of an earthquake,
satellite imagery presents a regional overview of damage sustained.
The location and extent of the damage can rapidly be determined to
help emergency workers scale and prioritize relief efforts. This
reconnaissance process can be undertaken using tiered methodology.
First, automated change detection algorithms
offer a “quick-look” damage assessment. In simple terms, these
algorithms compare images taken before and after the earthquake.
Damage is detected by comparing changes between the images. Change
detection algorithms have been used to successfully evaluate damage
that resulted from the 2001 Gujurat, 1999 Turkey, 1993 Hokkaido and
1995 Kobe earthquakes.
Figure 2 shows the spatial distribution of
severely damaged and collapsed buildings in Boumerdes, which were
identified in the pan-sharpened QuickBird coverage using
change-detection algorithms. The areas highlighted in red and yellow
correspond with concentrated building damage. The scenes acquired
before and soon after the earthquake were analyzed using ENVI image
processing software. A 9x9-pixel Laplacian edge detection filter was
initially applied, followed by a 25x25-pixel dissimilarity texture
measure. The resulting images were differenced, and the mean
standard deviation was plotted within a 200x200-pixel window.
Within the Tiered Reconnaissance System, this
quick-look assessment allowed for a more detailed inspection of
building damage using visualization techniques. A visual comparison
was drawn between enlarged views of the “before” and “after”
pan-sharpened images, which were displayed side by side within the
ENVI image processing environment. Due to the detailed
representation offered by QuickBird satellite imagery, severely
damaged structures were readily identified. Figure 3 maps the damage
(in blue). In addition to urban damage, Figure 3 also shows the
location of temporary tent camps (in green) that housed displaced
residents. Extraneous areas of change are attributable to isolated
cloud cover in the “before” scene (in yellow) and changing
conditions within the coastal waters.
Generally speaking, correspondence is high
between the damage map in Figure 2 and visually determined building
collapse. Such close agreement is a reflection of the distinctive
characteristics of severe structural damage in high-resolution
satellite coverage. Figure 4 shows these definitive characteristics
in greater detail. Collapsed apartment blocks are readily
distinguished by the bright yet chaotic appearance of debris and
piles of rubble. Changes in shape and position are evident where
buildings have “pancaked” or toppled sideways.
Having performed the initial reconnaissance of
damage location and extent, remote sensing imagery has a further
role to play in monitoring clean-up operations. The acquisition of
extended temporal coverage permits debris clearance and
reconstruction to be monitored. Figure 5 shows the full temporal
sequence for an area of apartment blocks in western Boumerdes. The
first image illustrates the buildings prior to the earthquake. The
second shows their collapsed state, surrounded by debris. The third
scene tracks recovery efforts, indicating that the site has been
mostly cleared.
To build on the study’s success, MCEER
continues to fund research on automated ways to detect building
damage with high-resolution satellite imagery. Such efforts are
expected to lead to real-time damage assessments for emergency
responders.
