Conference PaperPDF Available

Benchmarking housing damage as a driver of population displacement following earthquakes

Authors:
  • GEM, EUCENTRE, University of Aveiro

Abstract

In the aftermath of an earthquake, the number of occupants within destroyed housing is often used to approximate the number of people rendered homeless after the event. While this metric can provide rapid situational awareness, more recent research highlights the importance of additional factors beyond housing damage within the scope of household displacement (e.g., utility disruption, housing tenure, place attachment). This study models three recent earthquakes from different geographies (Haiti, Japan, and Nepal) to benchmark housing damage as a driver of population displacement against reported values and mobile location data-based estimates. The results highlight the promise of risk models to realistically estimate population displacement after earthquakes in the emergency phase as compared with official reports, but also indicate a large range of uncertainty in the predicted values. Furthermore, purely basing displacement estimates on housing damage may limit the ability of models to capture protracted displacement compared to more comprehensive models that include other factors influencing population return or alternative approaches such as using mobile location data. Although mobile location data offers potential to quantify displacement duration, the results of this study indicate further need to benchmark and validate such approaches.
BENCHMARKING HOUSING DAMAGE AS A DRIVER OF
POPULATION DISPLACEMENT FOLLOWING EARTHQUAKES
Nicole PAUL
1
,
2
, Carmine GALASSO
3
, Vitor SILVA
4
,
5
& Jack BAKER
6
Abstract: In the aftermath of an earthquake, the number of occupants within destroyed housing
is often used to approximate the number of people rendered homeless after the event. While this
metric can provide rapid situational awareness, more recent research highlights the importance
of additional factors beyond housing damage within the scope of household displacement (e.g.,
utility disruption, housing tenure, place attachment). This study models three recent earthquakes
from different geographies (Haiti, Japan, and Nepal) to benchmark housing damage as a driver
of population displacement against reported values and mobile location data-based estimates.
The results highlight the promise of risk models to realistically estimate population displacement
after earthquakes in the emergency phase as compared with official reports, but also indicate a
large range of uncertainty in the predicted values. Furthermore, purely basing displacement
estimates on housing damage may limit the ability of models to capture protracted displacement
compared to more comprehensive models that include other factors influencing population return
or alternative approaches such as using mobile location data. Although mobile location data offers
potential to quantify displacement duration, the results of this study indicate further need to
benchmark and validate such approaches.
Introduction
An average of 24 million annual displacements were triggered by disasters between 2008 and
2018, approximately three times greater than those triggered by conflict and violence (IDMC,
2019). The number of people displaced annually is likely to increase under ongoing trends, driven
by poorly managed urban growth in hazard-prone areas and potentially exacerbated by climate
change. Despite this scale of human impact, most disaster risk assessments have focused on
economic losses and casualties. More recent studies have aimed to quantify population
displacement following earthquakes (Grinberger and Felsenstein, 2016; Burton et al., 2019;
Bhattacharya and Kato, 2021; Costa, Haukaas and Chang, 2022), identifying a range of
influencing factors. Numerous potential determinants of population displacement have been
identified (e.g., homeownership, place attachment, utility disruption); yet, standard practice is
simply to multiply the number of destroyed (i.e., uninhabitable) housing units by the average
household size. Regardless of the selected risk metric, an issue that plagues disaster risk
assessment is the need for more benchmarking or validation studies to ensure that risk models
reasonably predict observed values. This study aims to benchmark the standard practice of using
housing destruction as a driver of displacement against official statistics and alternative estimates
using mobile location data, allowing us to understand the prediction potential and uncertainty
range of this simplified approach.
Approaches to quantifying population displacement
Defining population displacement
Past researchers have highlighted a lack of consistent terminology regarding population
displacement in the disaster context (e.g., Mitchell, Esnard and Sapat, 2012; Esnard and Sapat,
2014; Greer, 2015), which has complicated efforts to quantify and interpret displacement metrics.
The Internal Displacement Monitoring Centre (IDMC) defines displacement as “involuntary or
forced movementsof individuals or groups of people from their habitual places of residence”
that can be triggered by disasters or other causes such as conflict and violence or development
1
PhD Candidate, University College London, London, UK, nicole.paul.22@ucl.ac.uk
2
Risk Analyst, Global Earthquake Model Foundation, Pavia, Italy
3
Professor, University College London, London, UK
4
Risk Coordinator, Global Earthquake Model Foundation, Pavia, Italy
5
Professor, University Fernando Pessoa, Porto, Portugal
6
Professor, Stanford University, Stanford, USA
SECED 2023 Conference Paul et al.
2
projects (IDMC, 2020). As a part of their Global Internal Displacement Database (GIDD) initiative,
the IDMC gathers information on metrics associated with displacement after disaster events,
including evacuations (i.e., people leaving their habitual residence in advance of or during the
onset of a hazard), sheltered populations (i.e., people accommodated in shelters or relief camps
provided by national authorities or other organizations), and the population rendered homeless
(i.e., due to housing destruction; IDMC, no date). As discussed in IDMC (2018), evacuation
estimates are based on the population covered by mandatory evacuation orders and/or the
population in shelters. In contrast, estimates of the homeless population are primarily based on
housing destruction estimates, typically multiplied by the average household size. This metric is
most similar to the majority of past attempts to quantify displacement (or “dislocation”; Lin, 2009)
within the earthquake engineering discipline; that is, damage incurred by an earthquake render
dwellings uninhabitable, thereby displacing residents.
Although physical damage to housing has often been considered a primary driver of initial
displacement (i.e., in the emergency phases), more recent studies have highlighted the
importance of additional factors beyond housing damage (e.g., Henry, 2013; Costa, Haukaas and
Chang, 2022). In particular, household decisions to permanently (and voluntarily) relocate (i.e.,
resettle) after disasters may be affected by factors such as place attachment (e.g., Costa, Wang
and Baker, 2022), social networks or social capital (e.g., Nejat and Damnjanovic, 2012; Nejat,
Cong and Liang, 2016; Lee, Sugiura and Gečienė, 2017), and home ownership (e.g., Kim and
Oh, 2014; Mayer et al., 2020). Despite the importance of population return, the benchmarking
study presented in this paper is limited to initial displacement estimates, which may inform shelter
needs in the emergency phase.
Approaches to quantifying population displacement
It is difficult to get reliable estimates of population movements following disaster events.
Households that evacuate or dislocate may stay with family and friends, stay in hotels or rentals,
remain outdoors (e.g., in tents or their car), or may seek public shelter. While headcounts of
sheltered populations can be relatively straightforward, evidence from past events indicates that
only a small subset of the displaced population seeks public shelters (Quarantelli, 1982), and data
regarding those that seek accommodation elsewhere is difficult to ascertain. As such, a variety of
approaches have been undertaken to estimate population displacement following disasters:
Based on housing destruction estimates: Reported or modelled estimates of housing
destruction are multiplied by the household size to determine the population rendered
homeless. This is the standard practice used by the IDMC to determine many of its
displacement estimates (IDMC, 2018).
Based on household surveys: A sample of households that habitually resided in an affected
area can be surveyed to understand the proportion that continues to be away from home,
ever evacuated, or ever sought public shelter. However, it can be challenging to contact
displaced populations. As an example of this approach, Kolbe et al. (2010) estimated that
1,269,110 people were still displaced 1-2 months after the 2010 Haiti earthquake, and
79,213 people sought shelter.
Manual counting of movements: Population movements can be estimated by tracking data
such as bus and ship movements out of an affected area, as the Haitian National Civil
Protection Agency (NPCA) performed following the 2010 Haiti earthquake. According to
their estimates, 511,405 people left Port-au-Prince about three weeks after the earthquake
(Bengtsson et al., 2011).
Mobile location data-based estimates: Call detail records (CDRs) or smartphone GPS
location data can be used to track population movements following disaster events (Yabe
et al., 2022). For example, Bengtsson et al. (2011) used CDRs to estimate that 580,000
people left Port-au-Prince about three weeks after the 2010 Haiti earthquake.
In this study, a model-based approach using housing destruction estimates is benchmarked
against other available estimates for recent earthquake events. For the considered events, only
estimates based on reported housing destruction estimates (from official statistics or IDMC) and
estimates from mobile location data (from the literature) were available. While the model-based
estimates in this study follow the same underlying assumption as the reported figures from official
statistics or the IDMC (i.e., based on housing destruction), the housing damage is simulated
based on the earthquake rupture characteristics, resulting ground shaking local intensity
estimates, and any available seismic station data rather than assumed from official reports.
Additionally, the distribution of occupants is more refined (i.e., different building types have
SECED 2023 Conference Paul et al.
3
different numbers of occupants rather than using a single average household size). As such, the
results from the benchmarking study allow us to evaluate the prediction potential and uncertainty
range of earthquake risk models. Such models might be used to assess disaster risk potential in
terms of population displacement (together with other risk metrics) for future events and evaluate
the cost-benefit of potential mitigation strategies.
Past earthquake scenario risk models
Selected scenarios
Three recent earthquakes were selected for study, as summarized in Table 1. These events were
selected based on the following criteria:
Recency: The exposure model used herein is representative of the year 2021. Therefore,
the modelled populations may not represent past decades, particularly if there has been
significant population growth or decline in recent years.
Availability of mobile location data-based estimates: Many approaches to estimating
population displacement assume housing destruction as the primary driver; thus, studies
using mobile location data were targeted to include an estimate that is not reliant on the
same assumption.
Geographic coverage: The events were selected to cover a range of geographic locations,
which entail different tectonic regions, standard construction practices (and associated
physical vulnerability of the building stock), and levels of data availability.
Earthquake
Date
Country
2021 MW7.2 Nippes
2021 August 14
Haiti
2016 MW7.0 Kumamoto
2016 April 16
Japan
2015 MW7.8 Gorkha
2015 April 25
Nepal
Table 1. Selected earthquake scenarios for the benchmarking study.
Data collection and input models
Two primary data sources were used to derive the scenario risk models discussed herein, both
courtesy of the Global Earthquake Model (GEM) Foundation. These data sources are described
further in this section.
The GEM Earthquake Scenario Database (ESD) is an ongoing initiative within the GEM
Foundation to collect information about past earthquake events, including ground shaking from
seismic stations and macroseismic intensity estimates, rupture model definitions (i.e., magnitude,
geometry, mechanism), candidate ground motion models (GMMs), and impact data (e.g.,
reported deaths, injuries, damages). This repository is available online at:
https://github.com/gem/earthquake-scenarios. For this study, ground shaking estimates from
seismic stations, rupture model definitions, and candidate GMMs were taken from this repository
to develop the hazard model component. Table 2 presents a summary of the primary sources of
data used. Although multiple rupture models and candidate GMMs are available in the GEM ESD,
a single combination was chosen for each earthquake scenario based on the consistency of the
simulated ground motion fields with the observations from seismic stations. Additionally, the soil
conditions (i.e., shear wave velocity in the upper 30 meters; VS,30) at each site were derived using
the global hybrid VS,30 map from the United States Geological Survey (Heath et al., 2020).
Earthquake
Seismic stations
Rupture model
Selected GMM
2021 MW7.2 Nippes
USGS1 (us6000f65h)
USGS finite fault
model (us6000f65h)
Akkar,
Sandıkkaya and
Bommer, (2014)
2016 MW7.0 Kumamoto
USGS1 (us20005iis)
NIED2
USGS fault rupture
model (us20005iis)
Chiou and
Youngs, (2014)
2015 MW7.8 Gorkha
USGS1 (us20002926)
CESMD3
Bhattarai et al. (2015)
Hayes et al., (2015)
Atkinson and
Boore, (2003)
Table 2. Summary of key inputs to the scenario hazard model component.
SECED 2023 Conference Paul et al.
4
This benchmarking study also used model components from GEM’s current Global Risk Model
(Silva et al., 2020). In particular, the residential exposure models for Haiti, Japan, and Nepal from
the Global Exposure Model (Yepes-Estrada et al., 2023) and the structural fragility functions from
the Global Vulnerability Model (Martins and Silva, 2021) were directly used. The exposure models
include building counts, the number of occupants, and building typologies, which are based
primarily on national statistics but are further adjusted to represent the year 2021 (i.e., to account
for population growth or decline in each administrative area). The structural fragility models are
defined for each building class within the exposure model for four different damage states: slight,
moderate, extensive, and complete damage. Further documentation on the fragility derivation
process can be found at: https://docs.openquake.org/vulnerability/. For this benchmarking study,
it was assumed that all occupants within extensively and completely damaged buildings would be
rendered homeless. That is, dwellings in the extensive or complete damage state were assumed
to be “uninhabitable,” thereby displacing their occupants.
Scenario risk analysis methodology
The scenario risk analyses were performed using the OpenQuake Engine (OQ; Silva et al., 2014),
an open-source seismic hazard and risk analysis software. Recently, the scenario calculator
within OQ has been extended to condition ground motion fields using data from seismic stations
following the procedure proposed in Appendix B by Engler et al. (2022). For this study, 1,000
Monte Carlo samples of cross-spatially correlated ground motions conditioned on available
seismic station data were generated for each event. For each simulated ground motion field, a
damage state is sampled for each asset in the exposure model using the associated fragility
curves for that asset (based on the building typology) and the corresponding ground motion
intensity measure (from the simulated ground motion field). The realized damage state for each
asset in each realization is then directly mapped to the displacement consequence (i.e., 100%
displaced in the complete and extensive damage state, 0% otherwise) and multiplied by the
number of occupants in that asset.
Benchmarking results
Selected metrics for comparison
The metrics for this benchmarking study include housing damage counts, housing destruction
counts, and multiple displacement figures (i.e., sheltered population, population rendered
homeless, and the number of evacuations).
As discussed above, four damage states are included in the OQ scenario models (i.e., slight,
moderate, extensive, and complete). However, different entities may define damage states
differently. For example, the Japanese Cabinet Office identifies the following building damage
states: partially damaged (一部破損), partially destroyed (半壊), and completely destroyed (全壊
). To facilitate comparison, the different reported damage states were summed into the categories
“damaged” and “destroyed,” where destroyed dwellings are considered uninhabitable and
damaged buildings suffered some damage (but are not destroyed). The assumed mapping is
shown in Table 3.
Source
Damaged housing
Destroyed housing
This study (OQ)
Slight
Moderate
Extensive
Complete
Caribbean Disaster
Emergency
Management
Agency (2021)
Damaged
Destroyed
Japan Cabinet Office
(2017)
Partially damaged (一部破損)
Partially destroyed (半壊)
Completely destroyed (全壊)
International Centre
for Integrated
Mountain
Development (2015)
Partially damaged
Fully damaged
Table 3. Mapping of reported damage states to aggregate housing damage and destruction.
SECED 2023 Conference Paul et al.
5
Similarly, different sources report displacement figures using a different basis for the metric (i.e.,
rendered homeless, sheltered, evacuated). Unlike damage, it is not realistic to sum the various
metrics to get an aggregate metric, as there may be considerable overlap between individuals
who evacuate, are rendered homeless, or are accommodated in shelters. Thus, the maximum
estimate is used if a source reports multiple metrics.
The criteria used to estimate displacement can also vary for mobile location data-based
displacement estimates, which is summarized in Table 4 for the referenced studies.
Source
Criteria for displacement
FlowMinder (2021)
moved from their pre-earthquake usual locationswithin the
Grand’Anse, Sud, and Nippes departments during the first
week after the earthquake
Yabe et al. (2020)
the rate of affected users who stayed outside their home
[shichoson (cities/wards)] out of all affected users on that
day” on the day of the earthquake
Wilson et al. (2016)
people above normal levels had left the [Kathmandu] valley
in the first three weeks after the earthquake
Table 4. Criteria used to estimate displacement based on mobile location data.
Haiti’s 2021 MW 7.2 Nippes earthquake
A comparison of the results for the 2021 Nippes earthquake is shown in Table 5 and Figure 1.
For this event, the scenario model predicted similar average damage estimates (and therefore
similar average displacement estimates) to official reports and the IDMC. In contrast, the mobile
location data-based estimate predicted approximately half the number of displacements. Notably,
the criteria used for the mobile location data-based estimate was described as “moved from their
pre-earthquake usual locations” in the first week after the earthquake. However, the spatial
resolution used in their assessment was unspecified; therefore, it is possible that a significant
population remained near their usual location but remained outside their habitual residence (e.g.,
stayed outside or in a tent due to fear of aftershocks). Additionally, the mobile location data-based
estimates assume that movement of the sample population (i.e., with SIM cards) is representative
of the overall population, which may not be the case if phone ownership and/or the damage
experienced is not uniform across population subgroups. Although all estimates are within the
modelled distribution, the range of values is significant (~100,000 to ~350,000 displaced).
Risk model
Reported
Mobile data
This study (OQ)
Caribbean
Disaster
Emergency
Management
Agency (2021)
IDMC (no date)
FlowMinder
(2021)
Damaged
houses
115,747
83,770
n. r.
n. r.
Slight
88,692
n. r.
n. r.
n. r.
Moderate
27,056
n. r.
n. r.
n. r.
Destroyed
houses
48,913
53,815
n. r.
n. r.
Extensive
13,302
n. r.
n. r.
n. r.
Complete
35,611
n. r.
n. r.
n. r.
Displaced
209,059
n. r.
220,000
90,000
Sheltered
n. r.
n. r.
n. r.
n. r.
Evacuated
n. r.
n. r.
n. r.
90,000
Homeless
209,059
n. r.
220,000
n. r.
Table 5. Comparison of results for the 2021 MW7.2 Nippes earthquake in Haiti;
“n. r.” indicates the value was not reported in that source.
SECED 2023 Conference Paul et al.
6
Figure 1. The modelled distribution of population displaced in this study (OQ) relative to other
benchmarks for the 2021 MW7.2 Nippes earthquake in Haiti.
Japan’s 2016 MW 7.0 Kumamoto earthquake
The comparison of results for the 2016 Kumamoto earthquake is shown in Table 6 and Figure 2.
For this event, the scenario model again predicted similar average damage and displacement
estimates to the reported data. However, there was a notable discrepancy between the average
buildings estimated in complete damage in OQ and reported as completely destroyed by the
official statistics. The Japan Cabinet Office reports standard statistics after earthquake events,
including the number sheltered and the number under evacuation orders. Interestingly, the
number sheltered in this earthquake greatly exceeds those under evacuation orders or advisories.
This contradicts findings from disasters in the United States, whereby residents who evacuate
seek public shelter only as a last resort (Quarantelli, 1982). In this case, the mobile location data-
based estimate exceeds the modelled and reported estimates but is of a similar magnitude. All
estimates are well within the range of the modelled distribution. The range of values predicted by
the model (~100,000 to ~300,000) has a similar but slightly smaller range than in the 2021 Nippes
earthquake in Haiti.
Risk model
Reported
Mobile data
This study (OQ)
Japan Cabinet
Office (2017)
IDMC (no date)
Yabe et al.
(2020)*
Damaged
houses
150,072
155,902
n. r.
n. r.
Slight
104,502
n. r.
n. r.
n. r.
Moderate
45,570
n. r.
n. r.
n. r.
Destroyed
houses
65,066
42,716
n. r.
n. r.
Extensive
23,911
34,037
n. r.
n. r.
Complete
41,155
8,679
n. r.
n. r.
Displaced
218,708
196,325
196,300
308,422
Sheltered
n. r.
196,325
196,300
n. r.
Evacuated
n. r.
1,224
n. r.
308,422
Homeless
218,708
n. r.
n. r.
n. r.
*The displacement estimates in Yabe et al., (2020) are reported as rates (25.5% on the day of the earthquake); to
convert the rate into an absolute value, the rate is multiplied by the estimated population in the 33 affected districts
considered within that study.
Table 6. Comparison of results for the 2016 MW7.0 Kumamoto earthquake in Japan;
“n. r.” indicates the value was not reported in that source.
SECED 2023 Conference Paul et al.
7
Figure 2. The modelled distribution of population displaced in this study (OQ) relative to other
benchmarks for the 2016 MW7.0 Kumamoto earthquake in Japan.
Nepal’s 2015 MW 7.8 Gorkha earthquake
The comparison of results for the 2015 Gorkha earthquake in Nepal is shown in Table 7 and
Figure 3. Although the average estimates of any level of damage (i.e., damaged plus destroyed)
are similar between the model and the official statistics, the breakdown by severity (i.e., damaged
versus destroyed) is notably different. For this reason, the average displaced estimates are more
markedly different than the other two earthquake scenarios. The mobile location data-based
estimate is significantly lower than the modelled and reported estimates, although this could be
due to the criteria employed within that study (“people above normal levels had left the
[Kathmandu] valley” in the first few weeks after the earthquake). Under that criterion, individuals
that may have left their habitual residence but remained in the Kathmandu Valley would not be
counted, nor would individuals normally residing outside the Kathmandu Valley in the first place.
Once again, all estimates lie within the modelled distribution. However, the range of predicted
values (~800,000 to ~3,000,000) is significant and notably larger than the other two scenarios.
This is likely due to a combination of the limited number of seismic stations (as compared with
Japan) to properly condition the ground motion fields and the higher sigma within the selected
GMM.
Risk model
Reported
Mobile data
This study (OQ)
International Centre
for Integrated
Mountain
Development (2015)
IDMC (no
date)
Wilson et al.
(2016)
Damaged
houses
810,176
282,300
n. r.
n. r.
Slight
599,480
n. r.
n. r.
n. r.
Moderate
210,696
n. r.
n. r.
n. r.
Destroyed
houses
284,604
508,215
n. r.
n. r.
Extensive
98,961
n. r.
n. r.
n. r.
Complete
185,643
n. r.
n. r.
n. r.
Displaced
1,802,535
2,860,000
2,623,000
390,000
Sheltered
n. r.
n. r.
n. r.
n. r.
Evacuated
n. r.
n. r.
n. r.
390,000
Homeless
1,802,535
2,860,000
2,623,000
n. r.
Table 7. Comparison of results for the 2015 MW7.8 Gorkha earthquake in Nepal;
“n. r.” indicates the value was not reported in that source.
SECED 2023 Conference Paul et al.
8
Figure 3. The modelled distribution of population displaced in this study (OQ) relative to other
benchmarks for the 2015 MW7.8 Gorkha earthquake in Nepal.
Conclusion
This study compared displacement predictions based on residential damage estimates from
earthquake risk models against official statistics and mobile location data-based estimates. The
model results seem promising as they are all broadly consistent with alternative estimates, but
there is a notable uncertainty range in all considered earthquake scenarios. Additionally, the
official statistics typically are underpinned by the same fundamental assumption (i.e., housing
destruction leads to displacement). Thus, a fully independent comparison is not possible to
validate the models. Additionally, validation is complicated by the use of many different metrics
to quantify displaced populations (i.e., rendered homeless, sheltered, evacuated).
The mobile location data-based estimates offer an interesting comparison, but further evaluation
of the displacement criteria used and the representativeness of the sample population may be
required. In the case of the Haiti and Nepal earthquakes, the mobile location data-based
estimates were notably lower than the mobel-based estimates and official reports. In some cases,
this may be because the considered population was restricted to specific areas (e.g., within the
Kathmandu Valley) or that there was an insufficient spatial resolution used in the displacement
criteria (i.e., neglecting those who left their habitual residence but migrated short distances).
Another issue could be that the movements of the sample population (i.e., those with phones) are
not fully representative of the affected population (e.g., elderly populations may be less likely to
carry phones and may also inhabit older buildings more prone to damage). Although more study
is needed, mobile location data offers the potential to explicitly capture the space and time
components of displacement.
In this study, the housing destruction-based estimates yielded reasonable estimates as compared
with the official reports. While this is a promising result for rapid estimates using the standard
practice, some critical factors that influence population displacement and shelter-seeking
behavior are neglected (e.g., utility disruption, weather). Moreover, quantification of the duration
of displacement remains a challenge as critical factors influencing population return in the
recovery phase (e.g., home ownership, place attachment, social networks) are not considered.
The results from this benchmarking study demonstrate the potential use of disaster risk models
to evaluate population displacement in the emergency phase, which can be useful for real-time
predictions to rapidly estimate shelter needs or can help expand the metrics quantified within
“what-if” scenarios and cost-benefit studies to capture more equitable and people-centered
metrics beyond economic loss and casualties.
SECED 2023 Conference Paul et al.
9
References
Akkar, S., Sandıkkaya, M.A. and Bommer, J.J. (2014) ‘Empirical ground-motion models for point-and
extended-source crustal earthquake scenarios in Europe and the Middle East’, Bulletin of earthquake
engineering, 12, pp. 359387.
Atkinson, G.M. and Boore, D.M. (2003) ‘Empirical ground-motion relations for subduction-zone
earthquakes and their application to Cascadia and other regions’, Bulletin of the Seismological
Society of America, 93(4), pp. 17031729.
Bengtsson, L. et al. (2011) ‘Improved Response to Disasters and Outbreaks by Tracking Population
Movements with Mobile Phone Network Data: A Post-Earthquake Geospatial Study in Haiti’, PLOS
Medicine, 8(8), p. e1001083. Available at: https://doi.org/10.1371/journal.pmed.1001083.
Bhattacharya, Y. and Kato, T. (2021) ‘Development of an Agent-Based Model on the Decision-Making of
Dislocated People After Disasters’, in S.C.M. Geertman et al. (eds) Urban Informatics and Future
Cities. Cham: Springer International Publishing (The Urban Book Series), pp. 387406. Available at:
https://doi.org/10.1007/978-3-030-76059-5_20.
Bhattarai, M. et al. (2015) ‘Overview of the large 25 April 2015 Gorkha, Nepal, earthquake from
accelerometric perspectives’, Seismological Research Letters, 86(6), pp. 15401548.
Burton, H. et al. (2019) ‘A framework and case study for integrating household decision-making into post-
earthquake recovery models’, International Journal of Disaster Risk Reduction, 37, p. 101167.
Caribbean Disaster Emergency Management Agency (2021) Haiti Earthquake: Final Situation Report #12.
St. Michael, Barbados. Available at:
https://www.cdema.org/images/2021/09/FINAL_CDEMA_Situation_Report_12_Haiti_Earthquake_14
Sep2021.pdf (Accessed: 27 April 2023).
Chiou, B.S.-J. and Youngs, R.R. (2014) ‘Update of the Chiou and Youngs NGA model for the average
horizontal component of peak ground motion and response spectra’, Earthquake Spectra, 30(3), pp.
11171153.
Costa, R., Haukaas, T. and Chang, S.E. (2022) ‘Predicting population displacements after earthquakes’,
Sustainable and Resilient Infrastructure, 7(4), pp. 253271. Available at:
https://doi.org/10.1080/23789689.2020.1746047.
Costa, R., Wang, C. and Baker, J.W. (2022) ‘Integrating Place Attachment into Housing Recovery
Simulations to Estimate Population Losses’, Natural Hazards Review, 23(4), p. 04022021. Available
at: https://doi.org/10.1061/(ASCE)NH.1527-6996.0000571.
Engler, D.T. et al. (2022) ‘Partitioning Ground Motion Uncertainty When Conditioned on Station Data’,
Bulletin of the Seismological Society of America, 112(2), pp. 10601079. Available at:
https://doi.org/10.1785/0120210177.
Esnard, A.-M. and Sapat, A. (2014) Displaced by Disaster: Recovery and Resilience in a Globalizing
World. New York: Routledge. Available at: https://doi.org/10.4324/9780203728291.
FlowMinder (2021) Haiti: Earthquake on 14 August 2021 (Version 1.2). Available at:
https://www.flowminder.org/media/dpxfefl4/haitiearthquake_report_27-aug_report-2_eng_v1-
2_final.pdf (Accessed: 27 April 2023).
Greer, A. (2015) Household residential decision-making in the wake of disaster: cases from Hurricane
Sandy. University of Delaware. Available at: https://udspace.udel.edu/handle/19716/31364
(Accessed: 25 November 2022).
Grinberger, A.Y. and Felsenstein, D. (2016) ‘Dynamic agent based simulation of welfare effects of urban
disasters’, Computers, Environment and Urban Systems, 59, pp. 129141. Available at:
https://doi.org/10.1016/j.compenvurbsys.2016.06.005.
Hayes, G.P. et al. (2015) ‘Rapid characterization of the 2015 M w 7.8 Gorkha, Nepal, earthquake
sequence and its seismotectonic context’, Seismological Research Letters, 86(6), pp. 15571567.
Heath, D.C. et al. (2020) ‘A global hybrid VS 30 map with a topographic slopebased default and regional
map insets’, Earthquake Spectra, 36(3), pp. 15701584.
Henry, J. (2013) ‘Return or relocate? An inductive analysis of decision-making in a disaster’, Disasters,
37(2), pp. 293316. Available at: https://doi.org/10.1111/j.1467-7717.2012.01303.x.
IDMC (2018) GRID Methodological Annex. Available at: https://www.internal-displacement.org/global-
report/grid2018/downloads/report/2018-GRID-methodological-annex.pdf (Accessed: 10 January
2023).
IDMC (2019) Disaster Displacement - A global review, 2008-2018. Available at: https://www.internal-
displacement.org/publications/disaster-displacement-a-global-review (Accessed: 29 March 2023).
IDMC (2020) GRID Methodology. Available at: https://www.internal-displacement.org/global-
report/grid2020/downloads/2020-IDMC-GRID-methodology.pdf (Accessed: 10 January 2023).
IDMC (no date) ‘Global Internal Displacement Database’. Available at: https://www.internal-
displacement.org/database (Accessed: 27 April 2023).
SECED 2023 Conference Paul et al.
10
International Centre for Integrated Mountain Development (2015) Lessons from Nepal’s Gorkha
earthquake 2015. Kathmandu, Nepal.
Japan Cabinet Office (2017) Disaster Report for 2016 Kumamoto earthquake. Available at:
https://www.bousai.go.jp/updates/h280414jishin/pdf/h280414jishin_39.pdf (Accessed: 27 April 2023).
Kim, J. and Oh, S.S. (2014) ‘The virtuous circle in disaster recovery: who returns and stays in town after
disaster evacuation?’, Journal of Risk Research, 17(5), pp. 665682. Available at:
https://doi.org/10.1080/13669877.2013.822917.
Kolbe, A.R. et al. (2010) ‘Mortality, crime and access to basic needs before and after the Haiti earthquake:
a random survey of Port-au-Prince households’, Medicine, Conflict and Survival, 26(4), pp. 281–297.
Available at: https://doi.org/10.1080/13623699.2010.535279.
Lee, Y.-J., Sugiura, H. and Gečienė, I. (2017) ‘Chapter 15 - Stay or Relocate: The Roles of Networks After
the Great East Japan Earthquake∗∗This chapter is a modified and extended version of Lee and
Sugiura (2014) in association with Ingrida Gečienė.’, in E.C. Jones and A.J. Faas (eds) Social
Network Analysis of Disaster Response, Recovery, and Adaptation. Butterworth-Heinemann, pp.
223238. Available at: https://doi.org/10.1016/B978-0-12-805196-2.00015-7.
Lin, Y.-S. (2009) Development of algorithms to estimate post-disaster population dislocationa research-
based approach. Texas A&M University.
Martins, L. and Silva, V. (2021) ‘Development of a fragility and vulnerability model for global seismic risk
analyses’, Bulletin of Earthquake Engineering, 19(15), pp. 67196745. Available at:
https://doi.org/10.1007/s10518-020-00885-1.
Mayer, J. et al. (2020) ‘Drivers of post-disaster relocations: The case of Moore and Hattiesburg tornados’,
International Journal of Disaster Risk Reduction, 49, p. 101643. Available at:
https://doi.org/10.1016/j.ijdrr.2020.101643.
Mitchell, C.M., Esnard, A.-M. and Sapat, A. (2012) ‘Hurricane Events, Population Displacement, and
Sheltering Provision in the United States’, Natural Hazards Review, 13(2), pp. 150161. Available at:
https://doi.org/10.1061/(ASCE)NH.1527-6996.0000064.
Nejat, A., Cong, Z. and Liang, D. (2016) ‘Family Structures, Relationships, and Housing Recovery
Decisions after Hurricane Sandy’, Buildings, 6(2), p. 14. Available at:
https://doi.org/10.3390/buildings6020014.
Nejat, A. and Damnjanovic, I. (2012) ‘Agent-Based Modeling of Behavioral Housing Recovery Following
Disasters’, Computer-Aided Civil and Infrastructure Engineering, 27(10), pp. 748763. Available at:
https://doi.org/10.1111/j.1467-8667.2012.00787.x.
Quarantelli, E.L. (1982) ‘General and particular observations on sheltering and housing in American
disasters’, Disasters, 6(4), pp. 277281. Available at: https://doi.org/10.1111/j.1467-
7717.1982.tb00550.x.
Silva, V. et al. (2014) ‘Development of the OpenQuake engine, the Global Earthquake Model’s open-
source software for seismic risk assessment’, Natural Hazards, 72, pp. 14091427.
Wilson, R. et al. (2016) ‘Rapid and Near Real-Time Assessments of Population Displacement Using
Mobile Phone Data Following Disasters: The 2015 Nepal Earthquake’, PLoS Currents, 8, p.
ecurrents.dis.d073fbece328e4c39087bc086d694b5c. Available at:
https://doi.org/10.1371/currents.dis.d073fbece328e4c39087bc086d694b5c.
Yabe, T. et al. (2020) ‘Understanding post-disaster population recovery patterns’, Journal of The Royal
Society Interface, 17(163), p. 20190532. Available at: https://doi.org/10.1098/rsif.2019.0532.
Yabe, T. et al. (2022) ‘Mobile phone location data for disasters: A review from natural hazards and
epidemics’, Computers, Environment and Urban Systems, 94, p. 101777. Available at:
https://doi.org/10.1016/j.compenvurbsys.2022.101777.
Yepes-Estrada, C. et al. (2023) ‘Global Exposure Modeling for Earthquake Risk Assessment', Earthquake
Spectra [In review].

Supplementary resource (1)

ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Seismic fragility and vulnerability assessment is an essential step in the evaluation of probabilistic seismic risk. Ideally, models developed and calibrated for the building portfolio of interest would be readily available. However, the lack of damage data and insufficient analytical studies lead to a paucity of fragility and vulnerability models, in particular in the developing world. This study describes the development of an analytical fragility and vulnerability model covering the most common building classes at the global scale. Nearly five hundred functions were developed to cover the majority of combinations of construction material, height, lateral load resisting system and seismic design level. The fragility and vulnerability were derived using nonlinear time-history analyses on equivalent single-degree-of-freedom oscillators and a large set of ground motion records representing several tectonic environments. The resulting fragility and vulnerability functions were validated through a series of tests which include the calculation of the average annual loss ratio for a number of locations, the comparison of probabilities of collapse across all building classes, and the repetition of past seismic events. The set of vulnerability functions was used for the assessment of economic losses due to earthquakes as part of the global seismic risk model supported by the Global Earthquake Model Foundation. (full-text viewer available at: https://rdcu.be/b4Kal)
Article
Full-text available
Despite the rising importance of enhancing community resilience to disasters, our understandings on when, how and why communities are able to recover from such extreme events are limited. Here, we study the macroscopic population recovery patterns in disaster affected regions, by observing human mobility trajectories of over 1.9 million mobile phone users across three countries before, during and after five major disasters. We find that, despite the diversity in socio-economic characteristics among the affected regions and the types of hazards, population recovery trends after significant displacement resemble similar patterns after all five disasters. Moreover, the heterogeneity in initial and long-term displacement rates across communities in the three countries were explained by a set of key common factors, including the community’s median income level, population, housing damage rates and the connectedness to other cities. Such insights discovered from large-scale empirical data could assist policymaking in various disciplines for developing community resilience to disasters.
Article
The global building exposure model is a mosaic of local and regional models with information regarding the residential, commercial, and industrial building stock at the smallest available administrative division of each country and includes details about the number of buildings, number of occupants, vulnerability characteristics, average built-up area, and average replacement cost. We aimed for a bottom-up approach at the global scale, using national statistics, socio-economic data, and local datasets. This model allows the identification of the most common types of construction worldwide, regions with large fractions of informal construction, and areas prone to earthquakes with a high concentration of population and building stock. The mosaic of exposure models presented herein can be used for the assessment of probabilistic seismic risk and earthquake scenarios. Information at the global, regional, and national levels is available through a public repository ( https://github.com/gem/global_exposure_model ), which will be used to maintain, update and improve the models.
Article
Rapid urbanization and climate change trends, intertwined with complex interactions of various social, economic, and political factors, have resulted in an increase in the frequency and intensity of disaster events. While regions around the world face urgent demands to prepare for, respond to, and to recover from such disasters, large-scale location data collected from mobile phone devices have opened up novel approaches to tackle these challenges. Mobile phone location data have enabled us to observe, estimate, and model human mobility dynamics at an unprecedented spatio-temporal granularity and scale. The COVID-19 pandemic, in particular, has spurred the use of mobile phone location data for pandemic and disaster management. However, there is a lack of a comprehensive review that synthesizes the last decade of work and case studies leveraging mobile phone location data for response to and recovery from natural hazards and epidemics. We address this gap by summarizing the existing work, and point to promising areas and future challenges for using mobile phone location data to support disaster response and recovery.
Article
Rapid estimation of earthquake ground shaking and proper accounting of associated uncertainties in such estimates when conditioned on strong-motion station data or macroseismic intensity observations are crucial for downstream applications such as ground failure and loss estimation. The U.S. Geological Survey ShakeMap system is called upon to fulfill this objective in light of increased near-real-time access to strong-motion records from around the world. Although the station data provide a direct constraint on shaking estimates at specific locations, these data also heavily influence the uncertainty quantification at other locations. This investigation demonstrates methods to partition the within- (phi) and between-event (tau) uncertainty estimates under the observational constraints, especially when between-event uncertainties are heteroscedastic. The procedure allows the end users of ShakeMap to create separate between- and within-event realizations of ground-motion fields for downstream loss modeling applications in a manner that preserves the structure of the underlying random spatial processes.
Chapter
The process of disaster recovery involves the dynamic interaction of multiple factors which often remain hidden from the surface. Furthermore, due to empirical limitations, recovery outcomes are often superficially understood. This research introduces the conceptual framework of an agent-based model of disaster recovery which allows clarification of inter-related aspects such as: pre-disaster community characteristics, damage type, and the recovery policies applied. The proposed model is applied to simulate separate recovery scenarios with emphasis on different policy parameters. The emergent outcomes are consistent with documented recovery phenomena in the existing literature, indicating the potential use of such models for understanding recovery mechanisms and pre-evaluating recovery policies.
Article
Time-averaged shear wave velocity over the upper 30 m of the earth’s surface ( V S30 ) is a key parameter for estimating ground motion amplification as both a predictive and a diagnostic tool for earthquake hazards. The first-order approximation of V S30 is commonly obtained through a topographic slope–based or terrain proxy due to the widely available nature of digital elevation models. However, better-constrained V S30 maps have been developed in many regions. Such maps preferentially employ various combinations of V S30 measurements, higher-resolution elevation models, lithologic, geologic, geomorphic, and other proxies and often utilize refined interpolation schemes. We develop a new hybrid global V S30 map database that defaults to the global slope-based V S30 map, but smoothly inserts regional V S30 maps where available. In addition, we present comparisons of the default slope-based proxy maps against the new hybrid version in terms of V S30 and amplification ratio maps, and uncertainties in assigned V S30 values.
Article
Post-disaster relocation may occur in the aftermath of major disasters, either unplanned or through buyout programs. There is an extensive research on recovery of affected population. However, research on the factors which cause the households to decide in favor of relocation or reconstruction is still in its infancy. This paper suggests a framework for classification and prediction of households' recovery decisions based on their attributes. To achieve this objective, data was collected through a telephone survey of residents affected by the 2013 Moore, Oklahoma and Hattiesburg, Mississippi tornados. The analysis was implemented using conditional inference trees and revealed the significance of level of damage, pre- and post-disaster homeownership, and housing type on the households' relocation decisions. The model can help with devising data-driven recovery policies tailored to a community's characteristics.
Article
An agent-based object-oriented model for household displacements is presented and used to analyze household decision-making after a hypothetical earthquake in the City of Vancouver, Canada. Temporary displacements and permanent relocation are accounted for. The model for households include considerations of socioeconomic demographics, social networks, and disaster preparedness. The analysis results indicate that nearly 70,000 persons are expected to be displaced by the earthquake. Of those, close to 19,000 will need public sheltering. In addition, nearly 40,000 persons are expected to relocate in the years following the earthquake. Among the displaced persons, occupants of multi-family pre-code and low-code buildings are over-represented. Among those needing public shelter or relocation, there is a disproportionately high number of renters and low-income households. The models in this paper can help the development of pre-disaster plans by suggesting optimal location of public shelters, and by identifying decisions that reduce the number of households relocating.