Earthquake Damage Detection Using High-resolution

Satellite Images

Fumio Yamazaki Ken’ichi Kouchi, Masayuki Kohiyama, Masashi Matsuoka

Department of Urban Environment Nanae Muraoka Earthquake Disaster Mitigation
Systems, Chiba University Institute of Industrial Science Research Center, NIED,
Chiba, Japan The University of Tokyo Kobe, Japan.
[email protected] Tokyo, Japan [email protected]

Abstract— QuickBird observed the city of Zemmouri, Algeria, QuickBird, a high-resolution commercial satellite with the
before and after the May 21, 2003 Algeria earthquake. Using the maximum spatial resolution of 0.6 m, launched successfully on
pre-event and post-event pan-sharpened images, visual inspection October 18, 2001 and it acquires optical images of urban areas,
of building damage was carried out by the five authors of this in which individual buildings can be identified. Hence, these
paper individualy. A total 1,399 buildings were classified into five images can be used to detect damages of individual buildings
damage levels of European Micro-seismic Scale. The results from and infrastructures after natural disasters. Using the images
the different interpreters were reasonably close for collapsed obtained by QuickBird before and after the 21 May, 2003
buildings but the difference becomes larger for smaller damage Algeria earthquake, this paper presents the results of visual
levels. The locations of refugee tents in the two post-event images
damage detection for Zemmouri City to demonstrate the
were also identified. These observations indicate that high-
resolution satellite images can provide quite useful information to
capability of high-resolution optical satellite images.
emergency management after natural disasters.
II. THE 2003 ALGERIA EARTHQUAKE AND QUICKBIRD
Keywords-QuickBird; the 2003 Algeria earthquake; building IMAGES
damage; visual inspection; Zemmouri City
A strong earthquake of magnitude 6.8 struck the
Mediterranean coast of Algeria on May 21, 2003. The
I. INTRODUCTION epicenter was located at 36.90N, 3.71E (USGS), offshore of
Recent advancements in remote sensing and its application the province of Boumerdes, about 50 km east of the capital
technologies made it possible to use remotely sensed imagery city, Algiers. According to the last official report from National
for assessing vulnerability of urban areas and for capturing Earthquake Engineering Center of Algeria, 2,278 people were
damage distribution due to natural disasters [1]. Especially it is killed, more than 10,000 were injured and about 180,000
important for emergency management and recovery works to people were made homeless. Zemmouri City is one of the most
capture damage distribution immediately after an earthquake or heavily damaged areas due to the earthquake.
other disasters. First, three pan-sharpened images (one pre-event on May
13, 2003, and two post-event images on May 23 and June 13,
2003) were produced (Fig. 1) and they were used in visual
inspection of building damage. Figure 2 shows the urbanized

Figure 1. Pan-sharpened QuickBird image of Zemmouri on May 23, 2003 Figure 2. Central part of Zemmouri City on May 23, 2003

0-7803-8742-2/04/$20.00 (c) 2004 IEEE 2280

area of Zemmouri on May 23, two days after the earthquake, in after this procedure, the damage grade of some buildings
which many collapsed buildings are observed. cannot be determined. If the result by the five persons includes
Grade 3 to 5, the damage level is classified as Grade 4.
TABLE I. CLASSIFICATION OF DAMAGE TO REINFORCED CONCRETE Otherwise the damage level is classified as the severer level.
BUILDINGS (EMS, 1998).
Damage Pattern Damage Level IV. VISUAL DAMAGE DETECTION OF ZEMMOURI CITY
Grade 1: Negligible to slight damage The detection results of the five interpreters and the
(no structural damage, slight non- majority grade using both the pre- and post-event images are
structural damage) shown in Table II. The numbers of buildings classified to the
majority damage levels were 35, 62, and 174 for Grades 3, 4,
Grade 2: Moderate damage
(slight structural damage, moderate and 5, respectively, based on the pre- and post- event images.
non-structural damage) The other 1,128 buildings were classified as Grade 1 or 2 out
of a total of 1,399 buildings. The interpreters #1 and #2 tend to
Grade 3: Substantial to heavy damage judge the damage to severer levels than the others.
(moderate structural damage, heavy
non-structural damage)
Consequently, through the first majority decision rule, the
interpreters #3, #4 and #5 became the majority in many cases.
Grade 4: Very heavy damage (heavy
structural damage, very heavy non- The differences of the numbers of classified levels between
structural damage) using only the post-event image and using both the pre- and
post-event images are investigated. For Grades 3, 4 and 5, the
Grade 5: Destruction difference among the interpreters is seen to be larger.
(very heavy structural damage) Especially it may be difficult to identify buildings as Grade 3
by using only a post-event image.
The average number of interpreters who classify the
TABLE II. COMPARISON OF THE DETECTION RESULTS OF THE FIVE damage level same as the majority level for Grade 5 is larger
INTERPRETERS
than those for Grades 3 and 4. The identification of “collapse
Damage / Interpreter No. #1 #2 #3 #4 #5 Majority
(Grade 5)” buildings does not vary much although it is
Grade 1 or 2 923 973 1168 1122 1120 1128 necessary to evaluate its accuracy based on ground truth data.
Grade 3 200 107 4 73 65 35 On the other hand, it is difficult to reach a consensus for
Grade 4 96 128 78 56 32 62 identifications of “moderate damage (Grade 3)” or “partially
Grade 5 180 191 149 148 182 174 collapsed (Grade 4)” cases.
Total 1399 1399 1399 1399 1399 1399
The average numbers who reached the same damage grades
Using one
Detection image 2.5 4 3 4 2.5 - based on the two images inspection are larger than those based
time
(hours) Using two
5 3.5 4.5 5 2.5 -
on the one image inspection. Hence the detection results based
images
on the two images can be said more stable than those based on
the one image.

V. COMPARISON WITH GROUND TRUTH IMAGES AND

III. DAMAGE CLASSIFICATION AND DAMAGE DETECTION DISCUSSION
Visual inspection of building damage was conducted based Figure 3 shows comparison between the satellite image and
on the classification in the European Macroseismic Scale [2], the on-site video pictures. In the satellite image, a circle
shown in Table I. Using only the post-event (May 23, 2003) symbol means “Grade 1 or 2”, a triangle symbol “Grade 3”, a
image and using both the pre- and post-event images, buildings diamond symbol “Grade 4”, and a star symbol “Grade 5”,
surrounded by debris (Grade 3), partially collapsed buildings respectively. The lower pictures show damaged/collapsed tall
(Grade 4) and totally collapsed buildings (Grade 5) were buildings (A) and a damaged low-rise house (B), which are
identified. For the purpose to obtain more confidence in the examples that the detection results using satellite images were
result of visual detection, five persons (actually the authors of verified by ground photographs.
this paper, who are researchers and graduate students in the
A total of 1,399 buildings in the central part of Zemmouri
fields of structural engineering) conducted visual inspection
City were classified based on their damage grades as shown in
and the differences among their results were investigated.
Figure 4. The damage ratios of buildings in each city block (a
From the results of individual damage detection, “the total 15 blocks) were calculated. Figure 5 shows the damage
majority damage level” was determined. Comparing the ratio of Grade 5 for each city block. It is seen that the heavy
number of persons who classified a building as “no damage” damage zone is concentrated in the central part of the urban
(Grade 1 or 2) with that as “damaged” (a total of Grades 3 to 5), area. The distribution of tents can clearly be observed in
if the former is majority, the damage level of the building is QuickBird images of Zemmouri. The locations of tents in the
determined as “Grade 1 or 2”. If not, the damage level is two post-event images were plotted in Fig. 6.
determined in the next stage, by comparing between the
numbers of persons who classified it as Grade 3, 4 or 5. Even

0-7803-8742-2/04/$20.00 (c) 2004 IEEE 2281

 Figure 4. 1,399 buildings classified based on their damage grades

Figure 3. Comparison between the satellite images and on-site video (by
courtesy of National Earthquake Engineering Center of Algeria).

The damage map in Zemmouri evaluated by the United Figure 5. Ratio of Grade5 buildings in each city block
States Government is available on Internet (UN-OCHA [3]).
Figure 7 compares the damage map in Zemmouri evaluated by
the United States Government and the result of this study. We
contacted UN-OCHA about the data source, but only reply we
received is “It used a variety of classified and unclassified
aerial and satellite remote sensing images.” Although the map
by US Government is very close to our interpretation, there is
possibility that they also used QuickBird images as a part of
data source.
Among the five interpreters, the interpreter #1 also
conducted visual detection for Boumerdes City [4] using
QuickBird images. It should be pointed out that the damage
ratio of buildings was different by the one-image and two-
image interpretations. For Boumerdes, the difference in the
ratios is 17.1% (of which the buildings identified as “Unclear”
based on one image was 15.6%) and for Zemmouri 23.2% (of
which 17.5% unclear). Hence, it is more difficult to classify
Figure 6. Location of tents in the two post-event images
damaged buildings in the image of Zemmouri than that of
Boumerdes. The satellite images of Zemmouri indicate that
there are more low-rise buildings placed close together. This A future research is suggested on the relationship between
fact explains the difference of difficulty in image interpretation. the damage ratio and building type, the accuracy of
interpretation, and the application of automated damage
detection [5].

0-7803-8742-2/04/$20.00 (c) 2004 IEEE 2282

 were reasonably close for collapsed buildings, but the
difference among the interpreters becomes larger for smaller
damage levels.

ACKNOWLEDGMENT
The QuickBird images used in this study are owned by
DigitalGlobe, Inc. The field survey data were provided by Dr.
Mohamed Belazougui, the Director of National Earthquake
Engineering Center of Algeria.

REFERENCES
[1] F. Yamazaki, “Applications of remote sensing and GIS for damage
assessment,” Proceedings of the 8th International Conference on
Structural Safety and Reliability, CD-ROM,12p, 2001.
[2] European Seismological Commission. “European Seismic Scale 1998.”
1998.
[3] The United Nations, Office for the Coordination of Humanitarian
Affairs. “Algeria: Damage from Earthquake - Zemmouri (May 2003).”
(a) Damage map for Zemmouri evaluated by the United States http://www.reliefweb.int/w/map.nsf/home. 2003.
Government (UN-OCHA, 2003) [4] K. Kouchi, F. Yamazaki, M. Kohiyama, M. Matsuoka, N. Muraoka,
“Damage detection from QuickBird high-resolution satellite images for
the 2003 Boumerdes, Algeria Earthquake,” Proceedings of the Asia
Conference on Earthquake Engineering, CD-ROM, pp.215-226, 2004.
[5] H. Mitomi, F. Yamazaki and M. Matsuoka, “Development of automated
extraction method for building damage area based on maximum
likelihood classifier,” Proceedings of the 8th International Conference
on Structural Safety and Reliability, CD-ROM, 8p, 2001.

(b) Visual damage detection by this study

Figure 7. Comparison of damage map made by US Government and

this study

VI. CONCLUSIONS
Using the high-resolution satellite images of Zemmouri,
Algeria, acquired by QuickBird before and after the 21 May,
2003 Algeria earthquake, visual interpretation of building
damage was conducted by five interpreters. Using only the
post-event pan-sharpened image, buildings surrounded by
debris (Grade 3), partially collapsed buildings (Grade 4), and
totally collapsed buildings (Grade 5) were identified. Some
buildings were difficult to judge their damage levels, and thus,
the pre-event image was also employed as a reference to judge
the damage levels. A total of 1,399 buildings were classified
and the locations of refugee tents in the two post-event images
were also detected. The results from the different interpreters

0-7803-8742-2/04/$20.00 (c) 2004 IEEE 2283