Professional Documents
Culture Documents
Validating Assessments of Seismic Damage Made From Remote Sensing
Validating Assessments of Seismic Damage Made From Remote Sensing
Validating Assessments of Seismic Damage Made From Remote Sensing
INTRODUCTION
Innovative and unprecedented use was made of remotely sensed images to provide
assessments of damage caused by the January 2010 Haitian earthquake. Remotely sensed
images proved to be of immense value in planning and coordinating the massive interna-
tional relief effort required after the disaster. This use is discussed elsewhere (Ghosh et al.
2011); this paper describes the validation of damage assessments made using three distinct
sets of observations using increasingly detailed data. The first dataset, comprising assess-
ments of 107,000 buildings in the Port-au-Prince urban area, was based on images with a
15–25 cm spatial resolution, taken in a vertical direction by high-altitude aircraft within a
few days after the earthquake and assessed in the Global Earth Observation-Catastrophe
Assessment Network II (GEO-CAN II) exercise. This exercise was carried out in the days
following the earthquake by around 600 engineers and scientists around the world, linked
by the Internet.
The second dataset, comprising a subset of 1,241 buildings from the GEO-CAN II
data, used Pictometry images. These are multiangle (both vertical and oblique) images
with a resolution of better than 25 cm. The images were taken in February 2010 and the
a)
Edmund Booth Consulting Engineer, 2 Miswell Cottages, Tring HP23 4JU, UK
b)
Willis Research Fellow, Cambridge Architectural Research Ltd, 25 Gwydir St No. 6, Cambridge CB1 2LG, UK
c)
Cambridge Architectural Research Ltd, 25 Gwydir St No. 6, Cambridge CB1 2LG, UK
d)
Department of Engineering Soils Group, Trumpington Street, Cambridge CB2 1PZ, UK
e)
ImageCat, Inc., 400 Oceangate, Suite 1050, Long Beach, CA 90802, USA
S157
Earthquake Spectra, Volume 27, No. S1, pages S157–S177, October 2011; V
C 2011, Earthquake Engineering Research Institute
S158 BOOTH ET AL.
Figure 1. Part of the GEO-CAN II survey area in Port-au-Prince, with the EEFIT ground obser-
vation locations and the ImageCat land use classes shown.
Table 1a. A total of 107,000 buildings were assessed in the Port-au-Prince area. Figure 2
shows a typical image used in the assessment.
The Pictometry assessment was made of 1,241 buildings, consisting of about 20 build-
ings at each of 60 locations within the GEO-CAN II area, all of which were identified as
having also been assessed by the GEO-CAN II exercise. Images with a spatial resolution
of better than 15 cm were used, taken during February 2010 from aircraft flying at a height
of around 1 km with cameras taking orthorectified (i.e., corrected to give a true horizontal
plane) oblique angle images in four orthogonal directions and vertical images. Damage
assessments were made by the staff of Cambridge Architectural Research Ltd. (CAR),
using the five-point damage scale shown in Table 1b. The 60 locations were planned to be
reasonably distributed between four land-use classes (commercial=downtown, residential
high-density, residential low-density, and slums or shantytowns) as identified on a pre-
event land-use map produced by ImageCat using pre- and post-earthquake imagery of
Haiti (Spence and Saito 2010, see Figure 1). Commercial and downtown, identified sepa-
rately in the PDNA document, were here amalgamated as the building types and collapse
rates seemed similar. Figure 3 shows four orthogonal images at one of the locations
assessed.
The third dataset, comprising a subset of 142 buildings from the Pictometry data at the
nine locations shown in Figure 1, was collected in and around Port-au-Prince by a three-
S160 BOOTH ET AL.
Table 1. Damage descriptions: (a) GEO-CAN; (b) Pictometry and EEFIT ground observation
person team from EEFIT from 7–11 April 2010. Their full report and photos of the survey
buildings, which is freely downloadable, has been published on the EEFIT website
(www.EEFIT.org.uk). The surveys were conducted on foot and 15% of the buildings (other
than those which had been destroyed) were inspected internally as well as externally. Basic
survey methods were used, comprising a detailed photographic record, handwritten notes,
basic dimensional measurements, and checks with a spirit (carpenter’s) level. Interviews
with a number of building occupants and owners were carried out.
Figure 2. Typical detail of a vertical angle aerial image used by GEO-CAN II.
the damage allocation made using the Pictometry data and ground observations by EEFIT
for all buildings in the common dataset of 142 buildings, while Tables 3b and 3c present
similar comparisons of GEO-CAN assessments with the ground observation and Pictom-
etry data, respectively. Table 4 presents a comparison between the Pictometry data and
ground observations, disaggregating the information of Table 3a into the nine survey
locations.
Figure 4. Comparison between the assessments by the three methods aggregated over 142
buildings. (NB: GEO-CAN assessments gave only nvd, D4, or D5—see Table 1a—so no GEO-
CAN II values are shown for D2 or D3.).
omission errors (number omitted from a damage state by the less accurate method, as a per-
centage of the total assigned to that state by the more accurate method) are shown in Table
3. It is striking that both commission and omission errors are particularly high for D4 in all
three parts of the table, suggesting that the assessment of D4 is most in need of improve-
ment for remote damage assessments. This may be partly associated with the difficulty of
distinguishing between D4 and D5, discussed in the section below on damage state defini-
tions, and suggests that D4 is very difficult to assess from vertical images alone. The Kappa
index, which measures the statistical correlation between the two assessments being com-
pared, is very low for the comparison of ground observations with GEO-CAN II assess-
ments, increasing to 0.32 for ground observations with Pictometry data; this value is still
low, however. The higher Kappa value of 0.42 for the comparison of the Pictometry data
with the GEO-CAN II assessments reflects the fact that both assessments have wrongly
attributed many damaged buildings to the lowest damage state.
A detailed examination was made of the buildings where discrepancies arose. It identified
the following main reasons for the divergence in buildings assigned a damage grade of D5 by
ground observation but a lower damage grade by the Pictometry data (Spence and Saito 2010):
1) The difficulty of spotting soft-story collapses, even using Pictometry data (Figure
5a and 5b).
2) The problem of damage being obscured by buildings and trees (Figure 5c).
Similar reasons are considered to apply in some of the cases of the buildings classified
as D4 or D3 by ground observation, but classified as at least two damage states lower by
Pictometry data. However, Pictometry cannot be expected to be a reliable method to
observe damage such as major cracking of walls, failure of beam-column joints, etc, which
are properly classified as damage level D3.
S164 BOOTH ET AL.
Table 3. Contingency tables showing the agreement between the damage assessment and
interpretation results from ground observation by EEFIT and Pictometry data, aggregated over
142 buildings. (a) Pictometry data compared to ground observations; (b) GEO-CAN II assess-
ments compared to ground observations; (c) GEO-CAN II assessments compared to Pictome-
try data
VALIDATING ASSESSMENTS OF SEISMIC DAMAGE MADE FROM REMOTE SENSING S165
Table 3. Continued
Notes to Table 3:
1) Matrices show numbers of buildings assigned to the damage levels of two surveys.
2) When a building was assigned a damage level that was between two states, it appears under both headings as
0.5 (e.g., a building with damage between states D2 and D3 appears as 0.5 at D2 and 0.5 at D3).
3) Damage states are defined in Table 1.
4) Using the symbols shown as a, b, and c for damage ratio D2 in Table 3a as an example:
commission error is derived as: (1 – a=b),
omission error is derived as: (1 – a=c).
5) The Kappa index is a well-established discrete multivariate technique used for the accuracy assessment of clas-
sification results (Congalton 1991). The Kappa index has a range of values between [0,1] with 1 meaning perfect
agreement and 0 indicating a pattern resulting from chance. Kappa index Khat is derived using the following
formula:
P
r P
r
N xii ðxiþ xþi Þ
i¼1 i¼1
Khat ¼ P
r
N2 ðxiþ xþi Þ
i¼1
where r is the number of rows in the matrix, xii is the number of observations in row i and column i, and xiþ and
xþi are the marginal totals for row i and column i (the total rows and columns), respectively. N is the total number
of observations (Jensen 1996).
Remarkably, Table 3a shows that 14 buildings were assigned a higher damage state
using Pictometry data than ground observation, including five assigned as having severe
damage (D4) found in the field to be only moderately or lightly damaged (D2 or D1). The
reasons were found to be misattributing damage from a neighboring collapsed building and
mistaking construction materials stored on roofs as earthquake damage.
S166 BOOTH ET AL.
Table 4. Comparison between the assessments using ground observation and Pictometry data
classified by the nine locations shown in Figure 1: (a) L1, (b) L3, (c) L9, (d) L17, (e) L22, (f)
L33, (g) L41, (h) L55, and (i) L61 (see Table 3 for notes)
VALIDATING ASSESSMENTS OF SEISMIC DAMAGE MADE FROM REMOTE SENSING S167
Table 4. Continued
The data collected by the EEFIT ground observations reported here indicate that dam-
age assessments from remote observations, made not only from vertical images but also
from more detailed oblique angle photographs (Pictometry), can have significant errors,
when compared with ground observations. However, in the aftermath of a very large catas-
trophe such as that which struck Haiti in January 2010, it is unlikely that comprehensive
damage assessments can be made using ground observations, or even using Pictometry, suf-
ficiently rapidly to suit the urgent needs of a postdisaster needs assessment, and remote
sensing is likely to continue to play a vital role, as it did in Haiti. Therefore, ways of
improving assessments made by remote sensing are discussed below.
Three possibilities suggest themselves. First, it may be that the underestimate of high
damage using remote sensing using a particular method (e.g., vertical aerial images or Pic-
tometry) may be corrected, using standard correction factors which are a function of varia-
bles such as instrumentally determined intensity, proportion of buildings assessed as having
damage class D5, type of construction, intensity of ground motion and local soil, topograph-
ical conditions, and so on. If such standard correction factors could be established, they
could immediately be applied to the remotely assessed damage proportions without any
fieldwork. However, much more data would be needed than that collected by EEFIT to
S168 BOOTH ET AL.
Figure 5. Examples of underestimated damage by Pictometry. (a) Soft-story collapse not evi-
dent from Pictometry: ground photos; (b) soft-story collapse not evident from Pictometry: Pic-
tometry image; (c) damage obscured by trees: ground photos.
establish the values of such factors with sufficient reliability to use after an earthquake. The
widespread availability of high-resolution satellite images, and the possibility of pooling
damage descriptions and photographs after earthquakes via the Internet make this a feasible
task for the future, and one which would be very valuable.
The second possibility for improving assessments is to subject a small sample of the
total number of assessed buildings to an assessment using more detailed methods,
VALIDATING ASSESSMENTS OF SEISMIC DAMAGE MADE FROM REMOTE SENSING S169
Figure 6. Prior distributions of heavy damage or collapse (D4 þ D5) probability for the four
land-use classes from GEO-CAN II.
Pictometry
(a) Commercial=downtown
mean 51% 31% 40%
qþr 6.00 7.86 13.86
q 3.06 2.47 5.53
r 2.94 5.39 8.33
r2 0.036 0.024 0.016
(b) Residential: high-density
mean 19% 19% 19%
qþr 6.00 6.38 12.38
q 1.14 1.19 2.33
r 4.86 5.19 10.05
r2 0.022 0.021 0.011
(c) Residential: low-density
mean 14% 19% 16%
qþr 6.00 6.38 12.38
q 0.84 1.19 2.03
r 5.16 5.19 10.35
r2 0.017 0.021 0.010
(d) Shantytown
mean 25% 22% 24%
qþr 6.00 3.94 9.94
q 1.50 0.87 2.37
r 4.50 3.07 7.57
r2 0.027 0.035 0.017
following the earthquake. For this analysis the prior assumed damage distribution is based
on all 107,000 buildings in the GEO-CAN II survey, which assessed rapidly and remotely
the proportion of collapsed structures across the entire area of Port-au-Prince. The results
were published online on 11 March 2010 (European Commission et al. 2010), and gave
mean proportions of collapsed and heavily damaged (D4 þ D5) buildings for each of four
separate land-use classes. Following earlier work on earthquake vulnerability distributions
(Spence et al. 2008), it can be assumed that the prior distribution follows a beta distribution
of the form f(phd) ¼ b(phd, q, r, a, b),1) where phd is the probability of heavy damage or col-
lapse, and a and b are the limits of the distribution, here taken as 0 and 1. Spence et al.
(2008) propose that where data quality is poor, a beta distribution for which q þ r ¼ 6 is
appropriate, and this is considered suitable for a prior distribution. The resulting prior distri-
butions are shown in Figure 6 for the four land-use classes, for all 107,000 buildings in the
GEO-CAN II survey. The distribution shows the probability density function for the aver-
age ratio of buildings with damage at D4 and D5 levels combined in an affected zone. Dif-
ferent distributions would apply for other damage levels.
Updating these distributions using the observed data from the Pictometry survey is now
possible. The data for all 1,241 buildings from the Pictometry assessment was also divided
into the same land-use classes, as identified in Table 5, although for the Pictometry analysis,
the high- and low-density residential land-use data were amalgamated. For each land-use class
there were a number of separate survey locations, and the mean and variance of the proportion
of collapsed buildings in each location was used to find the likelihood function for the distri-
bution of the observed data (also shown in Table 5), assuming this is also a beta distribution.
A v2 test was used to check the goodness of fit of the data points to the beta distribution, and
the fit was found to be acceptable with an average significance of about 90% at a 10% level.
With this prior and likelihood function defined, it is possible to define the posterior col-
lapse probability distribution. Where both prior and likelihood functions have the form of a
beta distribution, the posterior function also has a beta distribution, whose parameters, q
and r, turn out to be the arithmetic sum of those for the prior and likelihood function
(Jaiswal et al. 2011). Table 5 shows the detail of the posterior functions for the four land-
use classes, and the corresponding mean and variance; these are presented graphically in
Figure 7 for the commercial=downtown land-use class.
In principle, a further update is possible using the posterior determined above as a new
prior and the ground observation data as a new set of observations. However, the small
1)
The density function of the beta distribution bðx; q; r; a; bÞ is
1 ðx aÞq1 ðb xÞr1
fxðxÞ ¼ ; a < x < b;
Bðq; rÞ ðb aÞqþr1
in which q and r are parameters of the distribution and Bðq; rÞ is the beta function:
CðqÞCðrÞ
Bðq; rÞ ¼ and C is the gamma function ðAng and Tang 1975Þ:
Cðq þ rÞ
The parameters q and r of the beta distribution given depend on the mean lx , and variance r2x of the variable,
such that
lx ¼ q=ðq þ rÞ; r2x ¼ qr=ððq þ rÞ2 ðq þ r þ 1ÞÞ
S172 BOOTH ET AL.
amount of EEFIT ground observation data available was not well enough distributed across
the four land-use classes and damage states for a separate likelihood function to be deter-
mined for each land-use class. Instead, as a demonstration of the method, a single likelihood
function for collapse probability aggregated across all land-use classes was estimated with
parameters, as shown in Table 6; this is compared (Figure 8) with the equivalent prior and
posterior data. Table 6 and Figure 8 show that the spatial variability of the (D4 þ D5) dam-
age ratio between the nine survey locations of the ground observation data is much greater
than that between the 60 survey locations of the Pictometry data, so in this case the further
update yields results of limited value. However, the further update would be potentially use-
ful if more ground observation data were available. This is because the likelihood values of
variance shown in Table 6 for the Pictometry and ground observation data are based on spa-
tial variability alone. While that is reasonable for the ground observation data, it neglects
the epistemological (knowledge) uncertainty attached to the Pictometry data, which Figure
4 suggests is large; hence the variance is underestimated. A Bayesian updating process
using ground observation data has the potential to remove this difficulty because the dam-
age values are established with near certainty. The GEO-CAN prior estimate of uncertainty
in Table 6 (derived by assuming q þ r ¼ 6) is associated solely with epistemological
Figure 8. All land use areas aggregated: Prior distribution from GEO-CAN-II, likelihood func-
tion based on Pictometry data, first posterior, likelihood function based on ground observation,
and posterior 2 distribution for probability of D4 þ D5.
1) Most of the buildings inspected utilized neither unreinforced masonry nor concrete-
frames with masonry infill, but were a form of lightly confined masonry in which
vertical and (sometimes) horizontal reinforced concrete strips were added after con-
struction of the masonry walls. In such structures, the distinction made in EMS-98
between “structural” damage to beams and columns and “nonstructural” damage
(assumed greater) to infill masonry was therefore hard to apply, since the masonry
and reinforced elements worked intimately together.
VALIDATING ASSESSMENTS OF SEISMIC DAMAGE MADE FROM REMOTE SENSING S175
2) It was sometimes difficult to make distinctions between damage states D4 and D5.
The short definition of D5 found in EMS-98 is, “Destruction - total or near total
collapse,” and the icon-like sketches show ruined buildings that are little more than a
heap of rubble. D4 is described as “…heavy structural damage…partial collapse…”
However, Figure A-18 in Grünthal et al. (1998) suggests that the collapse of part of
the wing of a building may constitute level D5, even though the rest of the building
shown appears much less damaged, and Figure A-19 shows the loss of an upper
floor defined as D5, even though much lower damage is apparent on lower floors,.
3) In EMS-98 descriptions are only given for unreinforced masonry buildings or
concrete-frame buildings with masonry infill. There is a need for definitions appli-
cable to construction in confined or reinforced masonry, timber, steel, and precast
concrete. No distinction is made between concrete-frame and concrete shear wall
construction, let alone the various intermediate possibilities.
The goal of producing an internationally agreed-upon set of definitions, suitable for
postdisaster needs assessments, as well as for other uses, is proposed as an important
task which could form part of the development of the Global Earthquake Model (www.
globalquakemodel.org), perhaps in association the World Housing Encyclopedia
(www.world-housing.net) by the Earthquake Engineering Research Institute (EERI) and the
International Association for Earthquake Engineering (IAEE), particularly since it would
need to reflect regional differences. The task would include:
1) A review of the needs for the damage definitions for various applications, including
those in postdisaster needs assessments, the immediate civic response to an earth-
quake (habitability, repairability, danger to adjacent areas from aftershocks), and
loss estimates in future earthquakes for civic planning or insurance purposes.
2) A detailed set of damage descriptions for a comprehensive range of common build-
ing construction types. The special needs of assessments made by remote sensing
as well as by ground observation should be taken into account. For example, since
aerial surveys seldom give reliable information on construction types, it would also
be necessary to propose damage rates for the mixed building stocks found in partic-
ular areas for the particular needs of remote assessments.
3) A Web-based set of photos of damaged buildings with damage assignments, anno-
tations, and classifications. Remote vertical and oblique photos would be needed, as
well as those taken from the ground internally and externally.
CONCLUSIONS
1) Building damage assessments made from ground observations carried out for 142
buildings in Port-au-Prince following the 2010 Haiti earthquake show significant
discrepancies from those made remotely using high-quality vertical and oblique aer-
ial images. High damage was missed in the remote assessments, mainly because
damage was obscured by vegetation and other buildings, or because soft-story col-
lapses were not evident when the upper stories were largely intact. More rarely,
remote sensing overestimated damage for reasons including cases where remote
assessments grouped undamaged buildings with adjacent damaged ones, and where
materials stored on roofs were mistaken for damage to structures.
S176 BOOTH ET AL.
2) While there is no doubt that there is a general tendency in remote sensing to underes-
timate damage, there is a need to carry out more detailed “ground truthing” exercises
to establish how typical the degree of error found in Port-au-Prince was. The under-
estimation of high damage states (D4 þ D5) between estimates from oblique aerial
images (Pictometry) and ground observation was found to be a factor of 1.5, rising
to over 2 for vertical aerial images (GEO-CAN II), but this was based on a sample of
only 142 buildings and more data is urgently needed. It is important that the ratios
found in this limited exercise are not taken as typical and adopted for use elsewhere.
3) Rapid damage assessments made using remote sensing played a vital role in the
relief effort in Haiti, and will undoubtedly continue to do so in future disasters. It is
therefore important to develop methods by which such assessments can be
improved. Some possibilities are as follows.
4) Where damage assessments are made solely on the basis of remote sensing, it would
be very useful to have standard correction factors, which might depend on factors
such as instrumentally determined intensity, proportion of buildings assessed as
having collapsed, type of construction, intensity of ground motion, local soil and to-
pographical conditions, and so on. To establish such correction factors, an extensive
comparison between assessments made from ground observations and remote sens-
ing would be needed for a wide range of future earthquakes. Given the possibility
of pooling damage data via the Internet, this should be a feasible task, but would
need leadership from an organization such as the IAEE.
5) A method is described by which damage assessments made using vertical imagery are
improved using a much smaller sample of assessment from oblique images (Pictome-
try) or ground observations, using Bayesian methods. The method is demonstrated for
the Port-au-Prince data, and appears promising enough to test for other cases.
6) While heavy damage can be observed in vertical images, they do not reveal the pro-
portion of buildings that are only moderately or lightly damaged. A preliminary
attempt was made to determine whether extrapolation from proportions of buildings
suffering heavy damage to proportions suffering lower damage levels could be
made using an assumed binomial distribution. The investigation showed that, for
the data available, this particular distribution was not appropriate. Investigating
other possible distributions would be worthwhile.
7) The goal of producing an internationally agreed-upon set of damage definitions is
proposed as an important task. The set would need to be suitable for postdisaster
needs assessments as well as for other uses, including estimates of future losses,
and for damage assessments made using both ground observation and remote sens-
ing. This task could be undertaken as a contribution to the Global Earthquake
Model (www.globalquakemodel.org).
ACKNOWLEDGEMENTS
The authors acknowledge the funding provided by the UK government’s Engineering
and Physical Sciences Research Council (EPSRC) which met the travel expenses of all three
members of the EEFIT team. Thorough and thought-provoking reviews of the manuscript
were provided by three anonymous reviewers, to whom the authors are most grateful. The
VALIDATING ASSESSMENTS OF SEISMIC DAMAGE MADE FROM REMOTE SENSING S177
authors are grateful to many colleagues for constructive comments on earlier versions, espe-
cially Dr. Kishor Jaiswal and Fiona McRobie.
REFERENCES
Ang, A. H-S, and Tang, W. H., 1975. Probability Concepts in Engineering Planning and
Design, Vol 1, John Wiley & Sons, Hoboken, NJ, pp. 420.
Applied Technology Council (ATC), 1989. Procedures for the Postearthquake Safety Evalua-
tion of Buildings, Report No. ATC-20, Redwood City, CA.
Booth, E., Wilkinson, S., Spence, R., Free, M., and Rossetto, T., 2011. Proceedings of the Insti-
tution of Civil Engineers - Forensic Engineering, vol. 164, issue 3, U.K. Earthquake Engi-
neering Field Investigation Team (EEFIT), London, England, in press.
Braga, F., Dolce, M., and Liberatore, D., 1982. A statistical study on damaged buildings and an
ensuing review of the MSK 1976 scale, in The Southern Italy 23.11.1980 Earthquake,
National Research Council (CNR) Publication No. 503, Rome, Italy.
Congalton, R., 1991. A review of assessing the accuracy of classifications of remotely sensed
data, Remote Sensing of Environment 37, 35–46.
Dolce, M., Masi, A., and Goretti, A., 1999. Damage to buildings due to the 1997 Umbria-
Marche Earthquake, in Seismic Damage to Masonry Buildings (A. Bernardini, ed.) Balkema,
Rotterdam, Holland, pp. 243.
European Commission Joint Research Centre (JRC), United Nations Institute for Training and
Research Operational Satellite Applications Programme (UNOSAT), World Bank Global Fa-
cility for Disaster Reduction and Recovery (GFDRR), and Centre National d’Information
Géo-Spatial (CNIGS), 2010. Building Damage Assessment Report, Haiti Earthquake, 12
January 2010, Report to the Haitian Government, http://www.eqclearinghouse.org/
20100112-haiti/wp-content/uploads/2010/02/PDNA_damage_assessment_report_v03-1.pdf.
Ghosh, S., Huyck, C. K., Greene, M., Gill, S. P., Bevington, J., Svekla, W., DesRoches, R., and
Eguchi, R. T., 2011. Crowdsourcing for rapid damage assessment: The Global Earth Obser-
vation Catastrophe Assessment Network (GEO-CAN), Earthquake Spectra 27.
Government of Haiti (GoH), 2010. Haiti Earthquake PDNA: Assessment of damage, losses,
general and sectoral needs, Annex to the Action Plan for National Recovery and Develop-
ment of Haiti, http://www.refondation.ht/resources/PDNA_Working_Document.pdf.
Grünthal, G. (ed.), 1998. European Macroseismic Scale, Cahiers du Centre Européen de Géody-
namique et de Séismolgie, Vol. 7, European Seismological Commission, Luxembourg.
Jaiswal, K., Wald, D., and D’Ayala, D., 2011. Developing empirical collapse functions for
global building types, Earthquake Spectra, this issue.
Jensen, J. R., 1996. Introductory Digital Image Processing: A Remote Sensing Perspective,
Prentice-Hall, Englewood Cliffs, NJ, pp. 316.
Spence, R., So, E., Jenny, S., Castella, H., Ewald, M., and Booth, E., 2008. The global earth-
quake vulnerability estimations system: An approach for earthquake risk assessment for in-
surance applications, Bulletin of Earthquake Engineering 6, 463–483.
Spence, R., and Saito, K., 2010. Port-au-Prince Earthquake Damage Assessment using Pictom-
etry, A Report for ImageCat, Inc., June 2010, Cambridge Architectural Research Ltd., Cam-
bridge, England, www.carltd.com/downloads/Haiti Pictometry.pdf.
(Received 9 November 2010; accepted 1 June 2011)