Disaster Advances Vol. 6 (6) June 2013
(37)
Building Damage Assessment with VHR Images and
Comparative Analysis for Yushu Earthquake,China
Lu L.1*, Guo H.1 and Corbane C.2
1. Center for Earth Observation and Digital Earth, Chinese Academy of Sciences, No.9 Dengzhuang South Road, Haidian District, Beijing, CHINA
2. Institute for the Protection and Security of the Citizen, European, Commission, Joint Research Centre, TP 267, 21027 Ispra, ITALY
*lllu@ceode.ac.cn
Abstract
In the aftermath of a disaster, one urgent need is to
estimate the degree of structural damage to physical
infrastructure with adequate reliability and rapidity.
The new generation of very high-resolution imagery
(VHR, with a spatial resolution of 1 m or less), if
acquired after an earthquake, can be used to assess
structural damage at the building level. This paper
presents the experience of post-earthquake building
damage assessment for the Yushu earthquake using
VHR imagery by three organizations: the Center for
Earth Observation and Digital Earth, Chinese
Academy of Sciences (CEODE), the German
Aerospace Center (DLR) and the Joint Research
Center, European Commission (JRC).
The data sources and methodologies employed for
building damage evaluation are presented. To assess
the reliability of these sources and methodologies,
damage assessment maps produced are compared.
The intercomparision results indicate that damage
assessment maps based on satellite VHR data are
capable of capturing the damage distribution.
However, these maps are not suitable for providing
accurate information on the damage intensity as the
results tend to be underestimated in areas where the
damage is moderate and overestimated in areas with
a high level of damage and many collapsed structures.
Keywords: Building damage assessment, very high
resolution (VHR) image, earthquake, comparative analysis.
Introduction
In the aftermath of disasters, one of the most urgent needs
is to estimate the number of people affected and also the
degree of damage done to the physical infrastructure with
sufficient reliability and rapidity20. Remote sensing (i.e.
satellite and aerial imagery) technology has proved to be a
valuable tool for post-disaster damage assessment8.
Following the launch of commercial satellite systems (e.g.
IKONOS, Quick Bird, EROS-B, WorldView-1 and the
recently launched Geoeye), a new generation of very high-
resolution imagery (VHR) with a spatial resolution of 1 m
or less has become available. VHR images acquired after
an earthquake can be used to assess structural damage at
the building level.
The location, extent and severity of building damage
following a disaster can be quantified with VHR imagery.
Post-disaster satellite VHR imagery has been manually
interpreted in damage assessment in a number of natural
disasters6,21. To ensure the timeliness of information
provision during the rapid emergency response situation
after the occurrence of an earthquake, automatic algorithms
have been developed and applied to VHR images. Pesaresi
et al14 developed an automatic detection method to detect
damaged built-up structures after the earthquake and
subsequent tsunami (tidal wave) in South East Asia in
December, 2004. Chini et al2 detected the collapse of
individual buildings using an unsupervised classifier from
QuickBird panchromatic images.
Besides satellite images, post-event aerial VHR imagery is
also an optimal data source for rapid building damage
assessment11,18,19. The resolution increase from 41 cm for
satellite data to 15 cm for the aerial imagery was found to
have improved the accuracy of building damage mapping
greatly in the Haiti earthquake13. Despite the widespread
use of post-disaster satellite VHR images, damage maps
produced were found to underestimate the extent and
severity of the devastation compared with ground survey12.
The low consistency of satellite-derived products with
respect to the aerial-derived assessment has been reported
by Corbane et al5 in the analysis of building damage for the
Haiti 2010 earthquake.
Another strategy for identifying the building damage using
VHR data is based on change detection with multi-temporal
images2,15,16,17. The damage to buildings can be detected by
comparing the major changes in building features using co-
registered images taken before and after the earthquake.
Considering that there may be large differences between
pre- and post-disaster images in terms of the resolution,
solar illumination and viewing geometry that can prevent
the use of automatic change detection procedures, visual
image examination is still the most widely used method for
producing reliable damage assessment3.
On 14 April 2010, an earthquake with a magnitude of 7.1
hit the Yushu prefecture in Qinghai Province, China. Many
hundreds of people were reported dead, injured or missing.
Buildings were severely damaged. The road network and
communication lines were interrupted. Airborne images
were collected immediately after the earthquake by the
Center for Earth Observation and Digital Earth, Chinese
Disaster Advances Vol. 6 (6) June 2013
(
38
)
Academy of Sciences (CEODE). Damage assessment
reports produced by CEODE using the remote sensing
dataset played an important role in the post-disaster
response and relief. The other rapid damage assessment
map was also derived using a joint analysis of high
resolution satellite data by the German Aerospace Center
(DLR) and the Joint Research Center, European
Commission (JRC). In October 2010, the DLR rapid
damage assessment product was assessed through a
cooperated disaster mitigation activity involving CEODE
and JRC 10.
Fig.1: Study area of the Yushu earthquake
Study area and remote sensing data
On 14 April 2010, an Ms 7.1 earthquake occurred in Yushu
County of the Yushu Tibetan Autonomous Prefecture,
Qinghai Province, China (Fig. 1). The epicenter was
located at 33°12' N, 96°36' E and the focal depth was 14
km as reported by the China Earthquake Networks Center.
The seismogenic structure of the Yushu earthquake was the
left-lateral GanziYushu fault which runs in a
northwesterly direction for a length of nearly 500 km.
The earthquake caused huge loss of life as well as serious
damage to buildings and the urban infrastructure.
According to the investigation of the earthquake
mechanism and rupture process made by Chen et al1,
Yushu County is located near a seismic surface rupture
zone and thus was severely affected in this disaster. The
town of Gyegu, which has a population of 23000, is located
at the center of Yushu County.
In this study, very high resolution images covering Gyegu,
taken from both satellites and aircraft, were acquired to
assess the damage to buildings caused by the earthquake.
Four types of optical VHR images were used in the post-
event damage assessment as shown in table 1.
Three of the data sets were obtained using optical VHR
satellite sensors which hold the potential for post-disaster
damage assessment. Data covering Yushu prefecture and
the surrounding area was also acquired using an airborne
sensor from CEODE for the period 14th April to 16th April
that is, after the earthquake. The UltraCamXp (UCXp)
digital aerial camera system was carried on a B4101
Citation II/Model 550 aircraft. After about twelve hours’
flight comprising three sorties, 260 GB of airborne remote
sensing data with 0.33 m spatial resolution covering an area
of 2000 km2 was obtained.9
Building Damage assessment
Airborne imagery-based rapid damage assessment:
After the UCxp airborne images were received at CEODE,
a comprehensive building-by-building damage assessment
strategy was performed. An automatic method for detecting
earthquake-induced building collapse was applied to derive
damage indicators and highlight damaged areas. This
method was able to give a prompt overview of the extent of
the damage over the large earthquake-hit area and provided
reference information for the subsequent visual interpreta-
tion. The automatic detection method was an improvement
of a novel image processing technology employed in the
Wenchuan earthquake7. Using this method, damage
indicators such as potential debris and the rubble of
collapsed buildings are generated and settlement areas
which suffered severe damage are targeted. However, the
final output of the automatic detection algorithm across the
whole airborne scene showed a lot of variability. In particu-
lar, the spectral inconsistency between the different
airborne frames impeded the use of automatic methods.
Additionally, small buildings and cars were confused with
collapsed buildings, as they have similar textural features.
Partly damaged buildings without visible debris cannot be
targeted. After the automatic detection algorithm, a visual
interpretation process was applied to obtain immediate
information about damaged buildings.
Tens of image analysts manually delineated individual
buildings in different areas of Yushu County using ArcGIS
9.3 software. The ground photographs of collapsed
buildings taken by disaster relief organizations and the
media were referenced to facilitate the image interpretation
work. By analyzing the spatial distribution of the collapsed
houses, it was found that with an average collapse rate of
82%, low-rise buildings in the western and southern urban
areas were most vulnerable to the impact of the earthquake.
Thanks to the reinforced concrete construction of the
buildings, the collapse rate in the eastern and central areas
was much lower. The damage assessment map produced
was delivered to the Chinese disaster relief organizations
on 16th April.
Disaster Advances Vol. 6 (6) June 2013
(
39
)
Fig. 2: Damage assessment map produced by the CEODE and post-earthquake airborne images
Satellite imagery-based rapid damage assessment: The
mapping of the Yushu earthquake damage assessment was
performed by the Center for Satellite Based Crisis
Information (ZKI/DLR). ZKI provides a service for the
rapid provision, processing and analysis of satellite imagery
during natural and environmental disasters to assist
emergency response activities worldwide. ZKI’s typical
rapid mapping workflow is triggered when a crisis occurs.
The final results are integrated into map products and
transferred to users consisting of civil protection offices,
situation centers and decision makers.
The damage assessment performed at ZKI/DLR was based
on a joint visual interpretation and analysis of post-
earthquake Geoeye-1 and QuickBird-2 images, both
acquired on 15 April 2010. According to DLR, two classes
of areas with collapsed buildings were defined: vast
damage and moderate damage. The ‘vast damage’ level
includes areas in which most buildings have collapsed
(totally or partially) whereas ‘moderate damage’ represents
areas in which houses have not been damaged (not
collapsed).
The two categories refer to the predominant damage extent
within the defined polygon. The damage assessment results
are shown together with the Geoeye-1 image in fig. 3(a).
Thick clouds covered part of the image and made the
evaluation of the extent of building damage in these areas
impossible. A supplementary data source used in this case
was 61cm resolution QuickBird-2 pan-sharpened imagery.
In the context of the GMES Emergency Response Project
“SAFER - Services and Applications for Emergency
Response”, funded by the European Community's Seventh
Framework Programme, the ZKI damage assessment
product for Yushu earthquake was delivered to the end
users on 16 April and updated on 22 April 2010.
Multi-temporal satellite imagery-based damage
assessment : To evaluate the thematic consistency of the
map in detail, the damage assessment performed at JRC
was based on the visual interpretation of pre- and post-
earthquake satellite images. These images consisted of
IKONOS data from 22 November 2007 and Geoeye
imagery from 15 April 2010 respectively. Through the
photo-interpretation of multi-temporal images, polygons of
vast and medium damage degree were generated. Three
classes were considered consisting of polygons with “no
visible damage”, “medium damage” and “vast damage”.
The same classes and definitions used by the DLR team
were used by the JRC team in order to allow a consistent
inter-comparison of the two products. The assessment
result was shown in fig. 3(b) together with the Geoeye-1
image.
Disaster Advances Vol. 6 (6) June 2013
(
40
)
(a)
(b)
Fig.3: Damage assessment map produced by DLR(a) and JRC(b) underlain by the post-earthquake Geoeye-1 image.
Disaster Advances Vol. 6 (6) June 2013
(
41
)
Fig.4: IKONOS satellite imagery of Yushu, acquired before the earthquake and examples of building
damage as seen in pre and post-earthquake satellite VHR images
Table 1
VHR imagery used for building damage assessment for the Yushu earthquake
Sensor Type Acquisition date Spatial
resolution at
Nadir (m)
Spectral range(nm) Organization
Geoeye-1 Spaceborne 15th April,2010 0.41
1.65 450-900 (pan)
450-520 (blue)
520-600 (green)
625-695(red)
760-900 (NIR)
DLR
IKONOS Spaceborne 22nd
November,2007 0.8
4 450-900 (pan)
445-516 (blue)
506-595 (green)
632-698 (red)
757-853 (NIR)
JRC
QuickBird
-
2
Spaceborne
15th April,2010
0.61
2.4
450
-
900 (pan)
450-520 (blue)
520-600 (green)
630-690 (red)
760-900 (NIR)
DLR
UCXp Airborne 14th-16th
April,2010 0.33 450-520 (blue)
520-600 (green)
630-690 (red)
CEODE
Disaster Advances Vol. 6 (6) June 2013
(42)
Table 2
Quantitative comparison of the three damage assessment products
Classes CEODE DLR JRC
Vast damage Areas (km²) 1.5496 1.8417 2.3552
No. of buildings/polygons 1857 31 46
Medium damage
Areas(km²)
0
2.5166
3.0508
No. of buildings/polygons
14
81
No visible damage Areas(km²) 1.2628 0 0.1493
No. of buildings/polygons 3278 3
Total Areas(km²) 2.8124 4.3583 5.5554
No. of buildings/polygons 5133 45 148
The pre-earthquake IKONOS image and examples of
building damage are shown in fig.4. For the high-rise
buildings in the central part of the county, the extent of the
damage can be easily visualized and interpreted using the
post-disaster image. For low-rise buildings with partial
structural damage, the amount of damage to the buildings is
hard to identify and thus comparison of the pre- and post-
event images is helpful. For totally collapsed buildings and
for areas covered by alluvial fan deposits, the pre-
earthquake image is necessary for the enumeration of
collapsed buildings.
Intercomparison of damage assessment products
The timely collection of VHR satellite images was crucial
for rapid damage assessment by the international
community after the Yushu earthquake. Post-earthquake
satellite images with a spatial resolution of 1m or better are
capable of capturing the general spatial pattern of building
damage in earthquake-affected areas. Images from
alternative sensors can provide supplementary information,
if one sensor suffers from quality deficits such as cloud/
haze contamination or poor off-nadir angles. The DLR map
was the first rapid damage assessment product released to
the general public on 16 April.
The Geoeye-1 images, which had a 50-cm spatial
resolution and were collected on 15 April 2010,
complemented by the QuickBird-2 images, laid the
foundation for its production. The carrying out of the
flights and the subsequent intensive work at CEODE
enabled the rapid production of the damage assessment
map and its dissemination to national disaster relief
organizations on 16th April. The aerial photographs
acquired from 14 April to 16 April at a spatial resolution of
33 cm allowed analysts to extract information about
individual buildings. However, the deployment of flights
over a disaster area in the first few days after a catastrophe
is in many cases impossible due to the weather or because
of the economic situation. It is also possible that the
airborne images are inaccessible to the global rapid
mapping community for political reasons. The pre-
earthquake VHR images (IKONOS) were helpful for
detecting structural damage during the JRC damage
assessment while these images were acquired after the
rapid response phase and processed in October 2010.
An automatic collapsed building detection technique was
applied to the post-earthquake airborne images to highlight
damaged areas during the production of the CEODE
damage assessment map. Due to the spectral and spatial
variability in high resolution images, the detection results
were only used as reference data for subsequent visual
analysis. Compared with automatic methods, computer-
aided visual interpretation by image analysts combined
with the assignment of damage categories is a more reliable
approach. Manual interpretation by image analysts was
carried out for all three products.
It took hours to days to complete the interpretation,
especially for the building-level damage assessment
products and the product accuracy depended greatly on the
experience of the image analysts. Taking into account the
time-consuming nature of manual interpretation, a practical
strategy is to produce an aggregated rapid damage
assessment map first to ensure that information about the
disaster is provided in a timely way. Detailed assessment
products that include the level of damage to buildings can
be generated in an updated release.
Comparing the spatial distribution of the features in the
three damage assessment products visually (Fig.2 and 3), it
seemed that all three products had some degree of spatial
consistency over most of the study area. The western and
southern areas suffered the most severe damage and were
dominated by areas of ‘vast damage’. However, it can be
seen that the DLR map omitted several areas of medium
damage on the periphery of Gyegu town. Several of these
areas showed medium damage in the airborne imagery.
Errors of omission existed due to the spatial resolution and
viewing geometry of the satellite sensor as well as the
complex settlement patterns in suburban areas. The aerial
photographs used at CEODE, which have a spatial
resolution of 33 cm, allow for the tasks including detection,
enumeration and damage level assignment of buildings to
be carried out more accurately.
Minor structural damages were less discernible in the
pansharpened VHR satellite data having a resolution of
about 50 cm. Dwellings are often small and informal in the
suburban areas of Gyegu town. The distortion of roof
structures was hard to indentify in these districts. Also,
piles of debris could be confused with other features, such
Disaster Advances Vol. 6 (6) June 2013
(43)
as small cars and rubbish heaps that have similar textures in
these images. The comparison of pre- and post-event VHR
images can lead to a better understanding of the amount of
damage in affected areas, as we can see in the JRC damage
assessment product.
However, omission errors also occurred: the ‘no visible
damage’ polygon on the eastern periphery was shown to be
the site of tens of collapsed buildings using the CEODE
product. Other evident disagreements on damage degrees
are mainly distributed in the suburbs of Geygu town.
Damage assessment using post-earthquake satellite VHR
images overestimated the amount of damage in the western
periphery and the change detection result overestimated the
eastern, where the damage level should both be medium
according to the proportion of collapsed buildings in
CEODE product. The geometric distortion of the high
resolution satellite images hindered the accurate
recognition of damage information.
The damage assessment results are compared quantitatively
in table 2. The JRC damage assessment products have a
larger spatial coverage (5.5554 km²) than the DLR product
(4.3583 km²). Although the area assessed by the CEODE is
the smallest (2.8124 km²), this area is the sum of individual
buildings extending over a vast area (Fig. 2). A total of
1857 of all the delineated buildings were considered to be
collapsed and these occupied 55% of the total assessed
area. Due to the different rules applied in the damage
assessment processes, the geometry and degree of damage
for the polygons in the JRC and DLR maps are
inconsistent. More polygons were assessed in the JRC map
(148), which provides more thematic detail than the DLR
map (45). Polygons with no visible damage were not
included in the DLR map.
The airborne imagery-based product which evaluated more
than 5000 buildings, provided the most detailed
information for the end-users. For polygon-level damage
assessment, the standards and rules for polygon generation
and damage category assignment are still unavailable at
present. The increase in the number of polygons and
damage types can enhance the value of damage assessment
maps. Other types of damage assessment maps can be used
to provide more detailed information. For example, refined
results derived from remote sensing imagery were
represented by 250m*250m grid cells with different
damage categories in the 2006 Indonesia 12 and 2010 Haiti
earthquakes 21. Finally, quantitative validation of the
damage assessment products requires ground surveys,
including photographing and recording attributes of
buildings, to be conducted in the earthquake-affected areas.
The accuracy of the remote sensing results can be
computed by comparing them with actual ground
observation records.
Conclusion
The joint efforts made in relation to the Yushu earthquake
damage assessment proved the potential of very high
resolution remote sensing images for rapid post-disaster
damage mapping. Based on the experience of rapid
mapping activities after the Yushu earthquake, several
problems should be considered to optimize the efficiency of
the operational emergency response after future
earthquakes. The acquisition and preparedness of remote
sensing datasets are crucial in rapid damage assessment
after extreme events. Coordinated collaboration should be
built up between international organizations to facilitate the
sharing of available data.
Manual interpretation of remote sensing images is the most
reliable method of damage assessment but is time-
consuming. For near-real-time earthquake damage
assessment, robust and consistent automatic image
processing methods should be developed to derive damage
indicators from VHR images. The VHR satellite imagery-
based results tend to underestimate the amount of damage
in areas with a medium level of damage and to
overestimate it in areas where the level of damage is high.
Mapping standards and accuracy assessment methodologies
for building damage assessment products should be
explored in future studies.
Acknowledgement
The authors appreciate the organizations and individuals
that contributed to this joint damage assessment mission for
the Yushu earthquake. This research is supported by the
Major International Cooperation and Exchange Project of
the National Natural Science Foundation of China
“Comparative study on global environmental change using
remote sensing technology” under grant NO. 41120114001
and the 973 project 2009CB723906 “Earth observation for
sensitive factors of global change: mechanisms and
methodologies” of the Ministry of Science and Technology
of the People´s Republic of China.
References
1. Chen Y.T., Epicenter mechanism and break process
(http://www.csi.ac.cn/manage/html/) (in Chinese) (2010)
2. Chini M., Pierdicca N. and Emery W.J., Exploiting SAR and
VHR optical images to quantify damage caused by the 2003 Bam
earthquake, IEEE Transaction on Geoscience and Remote
Sensing, 47(1),145–152(2009)
3. Chini M., Earthquake damage mapping techniques using SAR
and optical remote sensing satellite data, In Jedlovec G., editor,
Advances in geoscience and remote sensing, InTech -Open
Access publisher, 269–278(2009)
4. Congalton R.G., A review of assessing the accuracy of
classifications of remotely sensed data, Remote Sensing of
Environment, 37, 35–46 (1991)
5. Corbane C., Saito K., Dell’Oro L., Gill S., Piard B., Huyck C.,
Kemper T., Lemoine G., Spence R., Krishnan R., Bjorgo E.,
Senegas O., Ghesquiere F., Lallemant D., Evans G., Gartley R.,
Toro J., Ghosh S., Svekla W., Adams B. and Eguchi R., A
Disaster Advances Vol. 6 (6) June 2013
(44)
comprehensive analysis of building damage in the January 12,
2010 Mw7 Haiti earthquake using high-resolution satellite and
aerial imagery, Photogrammetric Engineering and Remote
Sensing, 77, 997–1009 (2011)
6. Ehrlich D., Guo H.D., Molch K., Ma J.W. and Pesaresi M.,
Identifying damage caused by the 2008 Wenchuan earthquake
from VHR remote sensing data, International Journal of Digital
Earth, 2(4), 309-326 (2009)
7. Guo H.D., Lu L.L., Ma J.W., Pesaresi M. and Yuan F., An
improved automatic detection method for earthquake-collapsed
buildings from ADS40 image, Chinese Science Bulletin, 54,
3303-3307 (2009)
8. Guo H.D., Understanding global natural disasters and the role
of earth observation, International Journal of Digital Earth, 3 (3),
221-230 (2010)
9. Guo H.D., Zhang B., Lei L.P., Zhang L. and Chen Y., Spatial
distribution and inducement of collapsed buildings in Yushu
earthquake based on remote sensing analysis, Science in China
Series D: Earth Sciences, 53(6), 794- 796 (2010)
10. J.R.C., Emergency response core services validation report
(end of year 2 operations), SAFER-D20700-3 – PREP – JRC
(Restricted), Joint Research Center, European Commission, 54-69
(2010)
11. Kerle N., Heue S.l. and Pfeifer N., Real-time data collection
and information generation using airborne sensors, In Zlatanova
S. and Li J., editors, Geospatial information technology for
emergency response, Taylor & Francis, London, 43-74 (2008)
12. Kerle N., Satellite-based damage mapping following the 2006
Indonesia earthquake – how accurate was it?, International
Journal of Applied Earth Observation and Geoinformation, 12(6),
466–476 (2010)
13. Lemoine G., Validation of building damage assessments
based on post-Haiti 2010 earthquake imagery using multi-source
reference data, proceeding of the 2nd International workshop on
validation of geo-information products for crisis management, 11-
13 October 2010, Ispra, Italy, 33-34 (2010)
14. Pesaresi M., Gerhardinger A. and Haag F., Rapid damage
assessment of built-up structures using VHR satellite data in
tsunami-affected areas, International Journal of Remote Sensing,
28(13), 3013-3036 (2007)
15. Rathje E.M., Crawford M., Woo K. and Neuenschwander A.,
Damage patterns from satellite images of the 2003 Bam, Iran,
Earthquake, Earthquake Spectra, 21(S1), 295-307 (2005)
16. Saito K., Spence R. J.S., Going C. and Markus M., Using
high-resolution satellite images for post-earthquake building
damage assessment: a study following the 26 January 2001
Gujarat earthquake, Earthquake Spectra, 20, 145-169 (2004)
17. Tiede D., Lang S., Füreder P., Hölbling D., Hoffmann C. and
Zei P., Automated damage indication for rapid geospatia
reporting, Photogrammetric Engineering and Remote Sensing, 77,
933–942 (2011)
18. Turker M. and San B. T., Detection of collapsed buildings
caused by the 1999 Izmit, Turkey earthquake through digital
analysis of post-event aerial photographs, International Journal
of Remote Sensing, 25(21), 4701–4714 (2004)
19. Turker M. and Sumer E., Building-based damage detection
due to earthquake using the watershed segmentation of the post-
event aerial images, International Journal of Remote Sensing,
29(11), 3073–3089 (2008)
20. Voigt S., Kemper T., Riedlinger T., Kiefl R., Scholte K. and
Mehl H., Satellite image analysis for disaster and crisis-
management support, IEEE Transaction on Geoscience and
Remote Sensing, 45(6), 1520-1528(2007)
21. Voigt S., Schneiderhan T., Twele A., Gahler M., Stein E. and
Mehl H, Rapid damage assessment and situation mapping,
learning from the 2010 Haiti earthquake, Photogrammetric
Engineering and Remote Sensing, 77, 923-931(2011)
22. Yamazaki F., Yano Y. and Matsuoka M., Visual damage
interpretation of buildings in Bam city using QuickBird images
following the 2003 Bam, Iran earthquake, Earthquake Spectra,
21, 329-336(2005).
(Received 11th January 2013, accepted 25th April 2013)