Available via license: CC BY 4.0
Content may be subject to copyright.
Production Editor:
Gareth Funning
Handling Editor:
Pablo Heresi
Copy & Layout Editor:
Hannah F. Mark
Received:
May 6, 2024
Accepted:
October 11, 2024
Published:
October 28, 2024
doi:10.26443/seismica.v3i2.1374
Population displacement aer earthquakes:
benchmarking predictions based on housing damage
Nicole Paul �∗1, Carmine Galasso �2, Vitor Silva �3,4, Jack Baker �5
1PhD Candidate, University College London, London, UK, 2Professor, University College London, London, UK, 3Head of Risk Engineering, Global
Earthquake Model Foundation, Pavia, Italy, 4Researcher, University of Aveiro, Aveiro, Portugal, 5Professor, Stanford University, Stanford, USA
Author contributions: Conceptualization: All authors. Soware: Nicole Paul, Vitor Silva. Validation: Carmine Galasso, Vitor Silva, Jack Baker. Formal Analysis:
Nicole Paul. Writing - Original dra: Nicole Paul. Writing - Review & Editing: All authors. Visualization: Nicole Paul. Supervision: Carmine Galasso, Vitor Silva,
Jack Baker. Funding acquisition: Carmine Galasso, Nicole Paul.
Abstract In the aermath of an earthquake, the number of residents whose housing was destroyed is of-
ten used to approximate the number of people displaced (i.e., rendered homeless) aer the event. While this
metric can provide rapid situational awareness regarding potential long-term housing needs, more recent re-
search highlights the importance of additional factors beyond housing damage within the scope of household
displacement and return (e.g., utility disruption, tenure, place attachment). This study benchmarks popula-
tion displacement estimates using this simplified conventional approach (i.e., only considering housing de-
struction) through three scenario models for recent earthquakes in Haiti, Japan, and Nepal. These model
predictions are compared with oicially reported values and alternate mobile location data-based estimates
from the literature. The results highlight the promise of scenario models to realistically estimate population
displacement and potential long-term housing needs aer earthquakes, but also highlight a large range of
uncertainty in the predicted values. Furthermore, purely basing displacement estimates on housing damage
oers no view on how the displaced population counts vary with time as compared to more comprehensive
models that include other factors influencing population return or alternative approaches, such as using mo-
bile location data.
1 Introduction
An average of 24 million annual displacements were
triggered by disasters between 2008 and 2018, approx-
imately three times greater than those triggered by con-
ict and violence (IDMC,2019). The number of peo-
ple displaced annually is likely to increase under ongo-
ing trends, driven by poorly managed urban growth in
hazard-prone areas, and potentially exacerbated by cli-
mate change eects. Despite this scale of human im-
pact, disaster risk models have primarily focused on
quantifying economic losses due to direct physical dam-
age.
In earthquakes, the conventional practice for cal-
culating the population displaced assumes that direct
physical damage can render housing uninhabitable,
thereby dislocating residents (see Fig. 1). Although
housing damage has oen been considered a primary
driver of both initial displacement and potential long-
term housing needs, more recent studies have high-
lighted the importance of additional factors beyond
housing damage that inuence displacement duration
and population return. These additional factors span
across the categories of physical damage to the built
environment (e.g., utility disruption, reconstruction
time), psychological and social phenomena (e.g., place
attachment, social capital), household demographics
∗Corresponding author: nicole.paul.22@ucl.ac.uk
(e.g., tenure, socioeconomic status), and pre- and post-
disaster policies (e.g., permanent or temporary housing
reconstruction programs, rental subsidies; Paul et al.,
2024). Recent studies of population displacement af-
ter earthquakes have begun to incorporate these addi-
tional factors and explicitly capture population return
(e.g., Burton et al.,2019;Bhattacharya and Kato,2021;
Grinberger and Felsenstein,2016;Costa et al.,2022). Al-
though these modeling improvements are promising,
validation of the conventional approach (i.e., estimating
population displacement based on housing destruction
alone) has yet to be performed.
Figure 1 An illustration of the conventional practice for
estimating population displacement aer disaster events.
This study aims to benchmark the conventional prac-
tice within earthquake risk models of using housing
destruction as the sole driver of population displace-
ment. We compare these model-based estimates with
ocially reported statistics and alternative estimates
1SEISMICA | ISSN 2816-9387 | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
derived from mobile location data, allowing us to un-
derstand this conventional approach’s prediction po-
tential and uncertainty range. Additionally, this study
provides an initial attempt to validate risk model out-
puts (i.e., housing damage and population displacement
estimates), a broader challenge within disaster risk as-
sessment (e.g., Beguería,2006;Ward et al.,2020;Crow-
ley et al.,2020). Despite the importance of population
return and the duration of displacement, this bench-
marking study is limited to single snapshot values of dis-
placed population counts that represent potential long-
term housing needs.
2 Quantifying disaster displacement
2.1 Population displacement metrics
Researchers have highlighted a lack of consistent termi-
nology regarding population displacement in disasters
(e.g., Esnard and Sapat,2014;Greer,2015;Paul et al.,
2024). This inconsistent use of terminology complicates
eorts to quantify and interpret displacement metrics.
The Internal Displacement Monitoring Centre (IDMC)
denes displacement as “involuntary or forced move-
ments […] of individuals or groups of people from their
habitual places of residence”that can be triggered by
disasters or other causes such as conict and violence
or development projects (IDMC,2020). As a part of
their Global Internal Displacement Database (GIDD) ini-
tiative, the IDMC gathers information on various met-
rics associated with displacement aer disaster events
(IDMC), including:
•Evacuations: People leaving their habitual resi-
dence in advance of or during the onset of a haz-
ard. These estimates are based on the population
covered by mandatory or advisory evacuation or-
ders, which are triangulated with evacuation cen-
tre headcounts (IDMC,2018). Evacuations are typi-
cally assumed to be relatively short-term, however,
there is ample evidence that not all households
that evacuate are able to return in a timely manner
(McAdam,2022).
•Sheltered populations: People accommodated in
shelters or relief camps provided by national au-
thorities or other organizations. These estimates
are more typically available using headcounts from
the relevant authorities or organizations.
•Population rendered homeless: People that are
displaced due to disaster-induced housing destruc-
tion. The IDMC typically estimates this value
by multiplying the reported housing destruction
counts from government agencies, UN organiza-
tions, or local authorities by the average household
size. This metric is most similar to past attempts
to quantify displacement within the earthquake en-
gineering discipline (also known as “dislocation”;
Lin,2009). In contrast to evacuations, this metric is
more representative of potential long-term housing
needs (Guadagno and Yonetani,2023). This is the
conventional approach within disaster risk models
noted herein.
In general, each of these metrics of population dis-
placement is reported as single snapshot values rather
than as a time series. While each of these metrics are
recorded by IDMC to inform their internal triangulation
and quality assurance processes, the specic metric re-
ported and underlying data source are not made pub-
licly available in the GIDD.
2.2 Proxies for estimating displacement
It is dicult 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 seek public shelter. While headcounts
of sheltered populations can be relatively straightfor-
ward, oen only a small subset of the displaced popula-
tion seeks public shelter (Quarantelli,1982,1995;IDMC,
2022a,b), and data regarding those that seek accommo-
dation elsewhere is dicult to ascertain. As such, a va-
riety of approaches have been used to estimate popula-
tion displacement following disasters. Some of the com-
mon proxies used to estimate population displacement
aer disaster events are shown in Fig. 2. Further infor-
mation about each proxy is provided in this section.
Figure 2 Common proxies for measuring population dis-
placement aer disaster events.
Household surveys and interviews have long been
employed to understand disaster impacts on individual
households. These can broadly be categorized as cross-
sectional or longitudinal studies. Cross-sectional stud-
ies gather observations at a single point in time, provid-
ing a snapshot at that moment. In some cases, there
are multiple observation windows, but each uses a dif-
ferent sample population. Longitudinal studies draw
repeated observations over time from the same sam-
ple of households, tracking changes over time amongst
that sample population. The vast majority of household
displacement surveys in the disaster literature take a
cross-sectional approach (e.g., Kolbe et al.,2010;Mayer
et al.,2020;Cong et al.,2018;Lee et al.,2017;Elliott and
Pais,2006;Groen and Polivka,2010). These studies pro-
vide rich information about that snapshot in time (e.g.,
one month aer the disaster, two years aer the disas-
ter) and for that specic disaster and sample popula-
tion. However, extending ndings to other time win-
dows for the same event is dicult. Another concern
with this approach is the sample’s representativeness,
as displaced households are hard to identify (e.g., they
are inaccessible in door-to-door visits, and mail may
not be forwarded to their current address). Specic
sampling methods have been employed to mitigate this
2SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
challenge, such as surveying at local community events
or at community shelters (e.g., Binder et al.,2015;Nejat
and Ghosh,2016) and through snowball sampling (i.e.,
asking participants to refer other participants who meet
the study criteria; see Nejat and Ghosh,2016, for an
example of this approach). Some recent studies have
taken a longitudinal approach (e.g., The Asia Founda-
tion,2019;Lines et al.,2022;Van De Lindt et al.,2020),
which provides a richer context for those communities
and disaster events. However, longitudinal surveys suf-
fer attrition challenges (i.e., the loss of participants over
time, which can aect the sample’s representativeness)
and costliness, making it dicult to scale across several
communities and disaster events for a broader under-
standing of disaster recovery.
In recent years, mobile location data has been ex-
plored as a proxy for population displacement and re-
turn. In one of the rst studies, Lu et al. (2012) used
call detail records to estimate that the population of
Port-au-Prince, Haiti, decreased by a maximum of 23%
aer the 2010 Haiti earthquake and that the destina-
tions of households were highly correlated with the lo-
cations where they had social bonds (i.e., where they
spent their time during holidays such as Christmas and
New Year’s). Mobile location data seems promising,
as it can continuously capture how populations move
across both time and space, potentially informing post-
disaster planning needs in real-time and enabling data-
driven retrospective studies. The validity of this proxy
relies on the movements of those with mobile phones
being similar to the movements of the overall disaster-
aected population. Concerns regarding sample rep-
resentativeness may be higher in countries where mo-
bile phone ownership is limited or potentially corre-
lated with factors such as housing quality, homeowner-
ship, or income level. Yabe et al. (2022) reviewed the
potential use of mobile location data for capturing dis-
aster impacts, classifying three main sources: call de-
tail records, smartphone GPS data from location intelli-
gence rms, and smartphone GPS data from major tech
rms.
•Call detail records (CDRs): In contrast to the other
two categories, this category is not reliant on smart-
phone ownership but on broader mobile phone
ownership, covering a more substantial subset of
the population. These records include the location
of nearby cellphone towers when users call or send
text messages. As a result, there is a lower spatial
and temporal resolution than with the GPS datasets
from smartphones. Some example studies using
CDRs to track population displacement aer disas-
ters include Bengtsson et al. (2011), Lu et al. (2012),
and Wilson et al. (2016).
•Smartphone GPS data from location intelligence
rms: Several location intelligence rms, which
collect and aggregate data from various third-party
smartphone applications, have emerged in recent
years. Precise location information could theoret-
ically be available, but some form of spatial aggre-
gation is typically required to alleviate data securi-
ty/privacy concerns. Additionally, data may not be
available over the full disaster recovery timeline, as
many location intelligence rms shue unique de-
vice identiers aer a set period of time (e.g., ev-
ery few months, every year) to preserve data con-
tributors’ privacy. There is oen limited trans-
parency in the data generation process, and these
rms cover fewer countries. Some example studies
using smartphone GPS data from location intelli-
gence rms include Yabe et al. (2021,2020) and Lee
et al. (2022).
•Smartphone GPS data from major tech rms: Ma-
jor tech rms can collect GPS location data from
their users directly rather than rely on third-party
services. Under specic agreements, these rms
may provide processed forms of this data, aggre-
gated in both time and space to address data secu-
rity/privacy concerns. However, this data is oen
only tracked for more limited periods and may not
cover the entire recovery timeline (e.g., Meta Data
for Good records data for up to three months aer
a disaster event). These outputs are generally re-
stricted to select products produced by each tech
rm, with a limited ability to modify those selected
metrics. An example study using smartphone GPS
data from a major tech rm includes Yabe et al.
(2019).
Data on mailing address changes of households in
disaster-aected communities has been explored as a
proxy for understanding disaster migration, such as
from postal redirection records, voter registration data,
or consumer credit reports (e.g., Plyer et al.,2010;De-
Waard et al.,2019,2020;Hinojosa,2018;Price,2011).
Postal redirection data can be a useful way to under-
stand the destination communities of displaced house-
holds (e.g., what proportion of households redirect to
their origin community versus an alternate commu-
nity). However, it seems unlikely that households that
are displaced for short periods (e.g., evacuate for less
than a week) would either voluntarily submit a change
of address (Plyer et al.,2010) or be automatically de-
tected by algorithms used to determine an individual’s
most-likely mailing address (such as those used by con-
sumer credit reports). For example, DeWaard et al.
(2019) found the proportion of displaced households
that returned within 12 months of Hurricanes Katrina,
Harvey, and Maria in the United States ranged from 12%
to 38% using data from the Consumer Credit Panel. This
would imply that over half of the households did not re-
turn within a year of each event, indicating a sample
that was heavily aected or that otherwise anticipated
more permanent moves. The quarterly sampling fre-
quency likely inuences these results, as displacement
durations of less than three months seem unlikely to be
counted in this approach.
Some studies have explored using school enrollment
data to understand migration and return aer disas-
ters (e.g., Sharygin,2021;Hinojosa and Meléndez,2018;
Newell et al.,2012). An implicit assumption in the rep-
resentativeness of this proxy is that the movement of
households with children enrolled in public schools re-
sembles that of the broader aected population. The
3SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
viability of these datasets in tracking longitudinal dis-
placement will vary depending on the sampling fre-
quency of the relevant administrative area. For exam-
ple, many dierent states within the United States the-
oretically capture continuous data on student transfers,
but in practice, the accuracy is limited beyond the of-
cial school census dates (e.g., early October for Cali-
fornia). In contrast, monthly estimates were available
to track student movement in New Zealand aer the
Christchurch earthquake on February 22, 2011 (Newell
et al.,2012).
Passenger trac data can have higher sampling fre-
quencies. For example, daily arrival and departure
cards for those on international ights before and af-
ter the 2010-11 Canterbury earthquake sequence in New
Zealand (Newell et al.,2012) or monthly net movement
of air travel passengers data before and aer Hurricane
Maria in Puerto Rico (Hinojosa and Meléndez,2018).
However, these datasets have limited relevance beyond
movements out of island communities or across inter-
national boundaries, as a consistent observation aer
past events is that displaced households usually move
short distances (e.g., Nawrotzki et al.,2014;Sharygin,
2021;Love,2011).
Since mobile location data appears to be an emerging
and promising source of population displacement esti-
mates, this study includes benchmarks from available
literature using such approaches (FlowMinder,2021;
Yabe et al.,2020;Wilson et al.,2016). However, it is ac-
knowledged that no standard approach was undertaken
in these studies. Dierences among the considered
mobile location data studies include the primary data
source used (i.e., CDRs versus smartphone GPS data),
the data provider and their associated sampling rate,
the displacement criteria established, and the analysis
methodology employed.
3 Past earthquake scenario models
3.1 Overview of the scenario models
In this section, the term “scenario model” is used rather
than the more general term “disaster risk model” to
distinguish the fact that a single earthquake rupture is
modeled rather than a set of rupture events (Silva,2016,
2018). This study benchmarks a conventional scenario
model-based approach against other available estimates
for recent earthquake events. These estimates include
reported impacts from ocial statistics or the IDMC and
mobile location data-based estimates published in the
literature. While the scenario model-based estimates in
this study typically follow the same underlying assump-
tion as the reported gures from ocial statistics or the
IDMC (i.e., housing destruction displaces residents), the
ground shaking is simulated based on the earthquake
rupture characteristics, resulting ground shaking local
intensity estimates, and any available seismic station
data in the study area rather than assumed from o-
cial reports. Additionally, the distribution of occupants
is more rened (i.e., dierent building types have dif-
ferent numbers of occupants rather than using a single
average household size) and the damage assessment is
performed using analytical fragility models (i.e., based
on simulated damage rather than using observed empir-
ical damage). As such, the results from the benchmark-
ing study allow us to evaluate the prediction potential
and uncertainty range of earthquake scenario models.
Such models might be used to assess disaster risk po-
tential in terms of population displacement for future
events and evaluate the cost-benet of potential mitiga-
tion strategies (alongside other risk metrics or decision
variables; e.g. Liel and Deierlein,2013;Cremen et al.,
2022;Hoyos and Silva,2022).
3.2 Selection of past earthquake events
Three recent earthquakes were selected, as summa-
rized in Tab. 1. These events were selected based on the
following criteria:
•Recency: The exposure model used herein (de-
scribed in the next section; Yepes-Estrada et al.,
2023) is representative of the year 2021. Therefore,
the modeled populations may not represent past
decades, particularly if there has been signicant
population growth or decline in recent years.
•Availability of mobile location data studies: Many
approaches to estimating population displacement
assume housing destruction as the primary driver;
thus, studies using mobile location data were tar-
geted 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 en-
tail dierent tectonic regions, standard construc-
tion practices (and associated physical vulnerabil-
ity of the building stock), and levels of data avail-
ability.
Country Name/Location MWDate
Haiti Nippes 7.2 August 14, 2021
Japan Kumamoto 7.0 April 16, 2016
Nepal Gorkha 7.8 April 25, 2015
Table 1 Selected earthquake scenarios for the bench-
marking study.
3.3 Data collection and input models
Two primary data sources were used to derive the sce-
nario 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 (GEM 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 de-
nitions (i.e., magnitude, geometry, mechanism), candi-
date ground motion models (GMMs), and impact data
(e.g., reported deaths, injuries, damages). This repos-
itory is publicly available at: https://github.com/gem/
earthquake-scenarios. For this study, ground shaking
4SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
Earthquake Seismic stations Rupture model Selected GMM
MW7.2 Nippes USGS1(us6000f65h) USGS finite fault model
(us6000f65h)
Akkar et al. (2014)
MW7.0 Kumamoto USGS1(us20005iis)
NIED2
USGS fault rupture model
(us20005iis)
Chiou and Youngs (2014)
MW7.8 Gorkha USGS1(us20002926)
CESMD3
Bhattarai et al. (2015)
Hayes et al. (2015)Atkinson and Boore (2003)
1US Geological Survey (USGS) ShakeMap’s station list (https://earthquake.usgs.gov/ data/shakemap/ )
2National Research Institute for Earth Science and Disaster Prevention (NIED)’s strong motion seismograph networks (https:
//www.kyoshin.bosai.go.jp/)
3Center for Engineering Strong Motion Data (CESDMD)’s archive (https://www.strongmotioncenter .org/)
Table 2 Summary of key inputs to the scenario hazard model component.
estimates from seismic stations, rupture model deni-
tions, and candidate GMMs were taken from this repos-
itory to develop the hazard model component. Tab. 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 combina-
tion was chosen for each earthquake scenario based on
the consistency of the simulated median ground motion
elds with the observations from seismic stations. Ad-
ditionally, the soil conditions (i.e., shear wave velocity
in the upper 30 meters; VS,30) at each site were derived
using the global hybridVS,30 map from the United States
Geological Survey (Heath et al.,2020).
This benchmarking study also uses model compo-
nents from version 2023.0.0 of GEM’s Global Risk Model
(Silva et al.,2020). In particular, the residential ex-
posure models for Haiti, Japan, and Nepal from the
Global Exposure Model (Yepes-Estrada et al.,2023) and
the structural fragility functions from the Global Vul-
nerability Model (Martins and Silva,2021). The expo-
sure models include building counts, the number of oc-
cupants, and building typologies, which are based pri-
marily on national statistics, further adjusted to rep-
resent the year 2021 (i.e., to account for population
growth or decline in each administrative area). The
structural fragility models are denedfor each building
class within the exposure model for four discrete dam-
age states: slight, moderate, extensive, and complete
damage. These four damage states roughly correspond
to the ve damage states in the European macroseis-
mic scale (EMS-98; Grünthal,1998), except that the com-
plete damage state in OQ encompasses both the fourth
damage state and the h damage state in EMS-98 (i.e.,
heavy structural damage with very heavy nonstructural
damage and very heavy structural damage). Further
documentation on the fragility derivation process can
be found at: https://docs.openquake.org/vulnerability/.
For this benchmarking study, it is assumed that
all occupants within extensively and completely dam-
aged buildings would be rendered homeless. That is,
dwellings in the extensive damage state (i.e., moderate
structural damage and heavy nonstructural damage) or
complete damage state (i.e., heavy structural damage or
beyond) are assumed to be “uninhabitable,” thereby dis-
placing their occupants. Although this assumption is
held constant for each of the three earthquake scenar-
ios, it is possible that dierent countries or communi-
ties would exhibit dierent behaviors or relationships
between housing damage and dislocation. For exam-
ple, some areas could require building inspection prior
to re-occupancy even at more moderate levels of dam-
age or mandate evacuations in light of potential aer-
shocks. Further, dierent dwelling types could have
stricter requirements for re-occupancy, such as requir-
ing water and power availability for re-safety in multi-
storey apartment buildings. However, the assumption
taken herein seems most consistent with IDMC, one of
the key benchmarks included in this study.
3.4 Scenario analysis methodology
The scenario analyses are performed using the Open-
Quake Engine (OQ), an open-source seismic hazard and
risk analysis soware (Silva et al.,2014). To leverage ob-
served data from recording stations, the scenario calcu-
lator within OQ has been extended to condition ground
motion elds using data from seismic stations following
the procedure proposed in Appendix B by Engler et al.
(2022). Uncertainties in the hazard component (i.e.,
source model and GMMs) have previously been iden-
tied to dominate the uncertainty in regional risk pre-
dictions (Kalakonas et al.,2020) and there is evidence
that incorporating observational data from recording
stations improves the accuracy and precision of sce-
nario loss estimates (Silva and Horspool,2019).
For this study, 1,000 Monte Carlo samples of cross-
spatially correlated ground motions conditioned on
available seismic station data are generated for each
event. The median estimates across all 1,000 realiza-
tions for each scenario are visualized in Fig. 3.
For each simulated ground motion eld, a damage
state is sampled for each asset in the exposure model us-
ing the associated fragility curves for that asset (based
on the building typology) and the corresponding ground
motion intensity measure (from the simulated ground
motion eld). The 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.
5SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
Figure 3 Median peak ground acceleration (PGA) for each scenario earthquake, from le to right: 2021 MW7.2 Nippes in
Haiti, 2016 MW7.0 Kumamoto in Japan, and 2015 MW7.8 Gorkha in Nepal. Available recording station PGA values are shown
as triangles, which were used to condition the simulated ground motion fields.
Country Source Damaged housing Destroyed housing
All This study (OQ) Slight
Moderate
Extensive
Complete
Haiti CDEMA (2021) Damaged Destroyed
Japan JCO (2017) Partially damaged (一部破損) Partially destroyed (半壊)
Completely destroyed (全壊)
Nepal ICIMOD (2015) Partially damaged Fully damaged
Table 3 Mapping of reported damage states to aggregate housing damage and destruction.
4 Benchmarking study results
4.1 Selected metrics for comparison
The metrics for this benchmarking study include hous-
ing damage and destruction counts, as well as any avail-
able displacement gures (i.e., sheltered population,
population rendered homeless, and the number of evac-
uations).
As discussed above, four damage states are included
in the OQ scenario models (i.e., slight, moderate, ex-
tensive, and complete). However, dierent entities
may dene damage states dierently. For example,
the Japanese Cabinet Oce (JCO) identies the follow-
ing building damage states: partially damaged (一部
破損), partially destroyed (半壊), and completely de-
stroyed (全壊;JCO,2017). To facilitate comparison,
the dierent reported damage states are summed into
the categories “damaged”and “destroyed,”where de-
stroyed dwellings are considered uninhabitable and
damaged dwellings suered some damage (but are not
destroyed). The assumed mapping is shown in Tab. 3.
Similarly, dierent sources report displacement g-
ures using a dierent basis for the metric (i.e., rendered
homeless, sheltered, evacuated). Unlike damage, it is
unrealistic to sum the various metrics to get an aggre-
gate value, as there may be considerable overlap be-
tween individuals who evacuate, are rendered home-
less, or are accommodated in shelters. Thus, the maxi-
mum estimate is used if a source reports multiple met-
rics.
The criteria used to estimate displacement using mo-
bile location data can also vary and is summarized be-
low for the referenced studies:
•Haiti’s 2021 MW7.2 Nippes earthquake: Based
on CDRs where mobile users “moved from
their pre-earthquake usual locations” within the
Grand’Anse, Sud, and Nippes departments during
the rst week aer the earthquake (FlowMinder,
2021).
•Japan’s 2016 MW7.0 Kumamoto earthquake: “The
rate of aected users who stayed outside their
home [shichoson (cities/wards)] out of all aected
users” on the day of the earthquake using smart-
phone GPS data (Yabe et al.,2020).
•Nepal’s 2015 MW7.8 Gorkha earthquake: The
“people above normal levels [that] had le the
[Kathmandu] valley”‘ in the rst three weeks aer
the earthquake, per CDRs (Wilson et al.,2016). This
refers to the number of post-earthquake outows
in excess of the pre-earthquake outows during the
benchmark period (January 1, 2015 through April 7,
2015).
4.2 Haiti’s 2021 MW7.2 Nippes earthquake
A comparison of the results for the 2021 Nippes earth-
quake is shown in Tab. 4and Fig. 4. For this event,
the IDMC based the displaced estimate on the reported
housing destruction count from CDEMA (2021) and mul-
tiplied that by an average household size of 4.08.
In this case, the scenario model predicted similar
average damage estimates (and therefore similar aver-
age displacement estimates) to ocial reports and the
6SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
Scenario model Reported Mobile data
This study (OQ) CDEMA (2021)IDMC 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 4 Comparison of results for the 2021 MW7.2 Nippes earthquake in Haiti; “n. r.” indicates the value was not reported in
that source.
IDMC. In contrast, the mobile location data-based esti-
mate predicted approximately half the number of dis-
placements.
Notably, the criteria used for the mobile location data-
based estimate was described as “moved from their pre-
earthquake usual locations” in the rst week aer the
earthquake. However, the spatial resolution used in
their assessment was unspecied; therefore, it is pos-
sible that a signicant population remained near their
usual location but remained outside their habitual resi-
dence (e.g., stayed outside or in a tent due to fear of af-
tershocks and/or to protect their property). This high-
lights a potential challenge of using mobile location
data as a proxy for population displacement in disasters:
physical return to a ‘home’ location does not necessarily
signify that a durable solution (e.g., stable housing) has
been found.
Additionally, the mobile location data-based esti-
mates assume that the movement of the sample popula-
tion (i.e., those 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 uni-
form across population subgroups. There were approx-
imately 64 mobile cellular subscriptions per 100 people
in Haiti in the year of this earthquake, which is notably
lower than the global average of 108 (World Bank Group,
2021). Past studies investigating the use of mobile loca-
tion data in low- to middle-income countries havefound
that mobile phone owners tend to be wealthier and
more highly educated (Blumenstock and Eagle,2010;
Wesolowski et al.,2012;Frias-Martinez and Virseda,
2012), and that higher income groups tend to travel fur-
ther and more frequently in baseline conditions (e.g.,
Wesolowski et al.,2012). On one hand, baseline mobil-
ity estimates (i.e., irrespective of disasters) using CDRs
have been found to be surprisingly robust despite these
dierences in mobile phone ownership (Wesolowski
et al.,2013), especially as compared with smartphone-
based estimates (Milusheva et al.,2021). On the other
hand, these studies do not consider sudden-onset haz-
ards and the potential role of damage in forcing move-
ment away from habitual dwelling units (i.e., disaster
displacement). There is also evidence that households
with lower socioeconomic status tend to experience
more damage in disasters (e.g., Hallegatte et al.,2020).
If those with mobile phones are less likely to experience
signicant damage than those without mobile phones,
estimates using this approach are likely to be biased.
Lastly, although all estimates are within the modeled
distribution, the range of values is signicant (112k to
306k displaced for ±one standard deviation).
Figure 4 The modeled distribution of population dis-
placed in this study (OQ) relative to other benchmarks for
the 2021 MW7.2 Nippes earthquake in Haiti.
4.3 Japan’s 2016 MW7.0 Kumamoto earth-
quake
The comparison of results for the 2016 Kumamoto
earthquake is shown in Tab. 5and Fig. 5. In this case,
the IDMC directly adopted the sheltered estimates from
the Japanese Cabinet Oce.
For this event, the scenario model again predicted
average damage and displacement estimates similar to
those of the reported data. Despite the similarity be-
tween the average scenario model estimates and the re-
ported values, there is a notable discrepancy between
the average buildings estimated in complete damage in
OQ and reported as completely destroyed (全壊) by the
ocial statistics, which could be in part due to varying
damage state denitions. The Japan Cabinet Oce re-
ports standard statistics aer earthquake events, includ-
ing the number sheltered and the number under evacu-
ation orders. Interestingly, the number sheltered in this
earthquake greatly exceeds those under evacuation or-
ders or advisories. This contradicts ndings from disas-
ters in the United States and the Pacic Islands, whereby
7SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
Scenario model Reported Mobile data
Yabe et al. (2020)*
This study (OQ) JCO (2017)IDMC Day 0 Day 160
Damaged houses 150,072 155,902 n. r. n. r. n. r.
Slight 104,502 n. r. n. r. n. r. n. r.
Moderate 45,570 n. r. n. r. n. r. n. r.
Destroyed houses 65,066 42,716 n. r. n. r. n. r.
Extensive 23,911 34,037 n. r. n. r. n. r.
Complete 41,155 8,679 n. r. n. r. n. r.
Displaced 218,708 196,325 196,300 308,422 154,816
Sheltered n. r. 196,325 196,300 n. r. n. r.
Evacuated n. r. 1,224 n. r. 308,422 154,816
Homeless 218,708 n. r. n. r. n. r. n. r.
*The displacement estimates in Yabe et al. (2020) arereported as rates (25.5% on the day of the mainshock and 12.8% 160 days aer the mainshock);
to convert the rate into an absolute value, the rate is multiplied by the estimated population in the 33 aected districts considered within that study.
Table 5 Comparison of results for the 2016 MW7.0 Kumamoto earthquake in Japan; “n. r.” indicates the value was not re-
ported in that source.
Prefecture Damaged houses Destroyed houses Displaced persons
This study Reported (JCO) This study Reported (JCO) This study* Reported (JCO)**
Kumamoto 112,959 147,563 60,730 42,497 204,721 183,882
Oita 9,076 8,062 2,331 231 6,520 12,443
Fukuoka 17,335 251 1,281 4 5,076 n. r.
Miyazaki 4,086 21 335 2 973 n. r.
Yamaguchi 710 3 16 n. r. 64 n. r.
Saga 3,317 1 266 n. r. 982 n. r.
Nagasaki 2,046 1 97 n. r. 334 n. r.
All other areas 543 n. r. 10 n. r. 38 n. r.
*The displaced persons estimates from the scenario models in this study (OQ) represent the population rendered homeless due to housing destruc-
tion
**The displaced persons estimates from the reported source (JCO,2017) is based on the max sheltered population counts
Table 6 Comparison of subnational results for the 2016 MW7.0 Kumamoto earthquake in Japan; “n. r.” indicates the value
was not reported in that source.
residents who evacuate seek public shelter only as a last
resort (Quarantelli,1982,1995;IDMC,2022a,b).
In this case, the initial mobile location data-based es-
timate exceeds the modeled and reported estimates but
is of a similar magnitude. The estimated displacement
rate over time is also provided by Yabe et al. (2020) as
shown in Fig. 6, which shows reasonable consistency
between the scenario model and the estimated displace-
ment rate from mobile location data at 160 days aer the
earthquake. This supports the assumption that housing
destruction-based displacement estimates might rea-
sonably estimate long-term housing needs. However,
the scenario model provides no view on the time pe-
riod of the displaced estimate. Notably, there were 131
mobile cellular subscriptions per 100 people in Japan
the year of this earthquake (World Bank Group,2021).
However, smartphone penetration during this time was
lower (reported as 50.1%; Newzoo,2017).
Since the JCO (2017) reports data at the subnational
level, the damage and displacement estimates can be
compared across prefectures as shown in Tab. 6. More
variability across the estimated damage and displace-
ment is evident at the prefecture-level, with the sce-
nario model (OQ) notably overestimating impacts in
Fukuoka relative to what was reported. This could
be partially due to the conditioning of ground motion
elds, where stations near to cities outlying the heavily-
populated Fukuoka city recorded higher values of PGA
than expected for the GMM employed. Since the ground
motion elds are conditioned on the recording station
data, the model correspondingly adjusts the inter-event
term (bias) in the GMM for all sites and reduces the
intra-event term at other sites inversely proportional to
distance (using a cross-spatial correlation model). This
can be observed in Fig. 3: while the estimated ground
shaking mostly attenuates with distance from the rup-
ture, there are some “islands” of relatively higher (or
lower) ground shaking near to station observations. In
the case of Fukuoka city, the station recorded relatively
higher shaking, leading to higher damage counts than
would have predicted without conditioning. While con-
ditioning the ground motion elds on available observa-
tional data is appealing, there could also be situations
where highly localized site conditions or station mal-
function yield unrealistic predictions at the station site
and neighboring sites (due to cross-spatial correlation).
A higher density of stations, especially near heavily pop-
ulated areas, could help mitigate this issue. More gener-
ally, the estimated damage and displacement is less con-
centrated in the scenario model as compared with what
was reported. This may underscore the importance
of incorporating on-the-ground observations to better-
8SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
constrain engineering forecasts (Loos et al.,2020).
All population displacement estimates are well within
the range of the modeled distribution. The range of
values predicted by the model (130k to 308k displaced
for ±one standard deviation) has a similar but slightly
smaller range than in the 2021 Nippes earthquake in
Haiti.
Figure 5 The modeled distribution of population dis-
placed in this study (OQ) relative to other benchmarks for
the 2016 MW7.0 Kumamoto earthquake in Japan.
Figure 6 The estimated population displacement rate
over time from mobile location data in Yabe et al. (2020) ver-
sus the estimated population displaced in this study (OQ)
for the 2016 MW7.0 Kumamoto earthquake in Japan.
4.4 Nepal’s 2015 MW7.8 Gorkha earthquake
The comparison of results for the 2015 Gorkha earth-
quake in Nepal is shown in Tab. 7and Fig. 7. For this
event, the IDMC estimated displacement based on the
number of households identied as eligible for receiv-
ing the housing reconstruction grant per Nepal’s Hous-
ing Recovery and Reconstruction Platform (HRRP),
multiplied by an average household size of 4.3.
Although the average estimates of any level of dam-
age (i.e., damaged plus destroyed) are similar between
the model and the ocial statistics, the breakdown
by severity (i.e., damaged versus destroyed) is notably
dierent. These discrepancies could be exacerbated
by the inapplicability of existing GMMs for this sce-
nario: this was a continent-continent subduction zone
earthquake, whereas existing subduction GMMs are pri-
marily derived from data in ocean-continent or ocean-
ocean subduction zones (Rajaure et al.,2017). However,
past studies have indicated that Atkinson and Boore
(2003) explains the available recorded PGA values well
(Chadha et al.,2015;Hough et al.,2016). Beyond the
hazard component, these discrepancies could also be
driven by potential inaccuracies in the exposure model
or associated fragility functions used for Nepal. Due to
the discrepancy in damage estimates, the average dis-
placed estimates are more markedly dierent than the
other two earthquake scenarios.
The mobile location data-based estimate is signi-
cantly lower than the modeled and reported estimates,
although this could be due to the criteria employed
within that study (“people above normal levels [that]
had le the [Kathmandu] valley” in the rst few weeks
aer the earthquake). Under that criterion, individu-
als who may have le their habitual residence but re-
mained in the Kathmandu Valley would not be counted,
nor would individuals normally residing outside the
Kathmandu Valley in the rst place. According to data
during the year of the earthquake, there were 100 mo-
bile cellular subscriptions per 100 people in Nepal,
much higher than in Haiti (World Bank Group,2021).
Once again, all population estimates lie within the
modeled distribution. However, the range of predicted
values (1,012k to 2,592k displaced for ±one standard de-
viation) is signicant 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 elds
and the higher sigma within the selected GMM for this
combination of magnitude and source-to-site distances.
Figure 7 The modeled distribution of population dis-
placed in this study (OQ) relative to other benchmarks for
the 2021 M7.8 Gorkha earthquake in Nepal.
5 Conclusions
This benchmarking study compares population dis-
placement estimates for recent earthquake events in
Haiti, Japan, and Nepal, which is summarized in Fig. 8.
9SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
Scenario model Reported Mobile data
This study (OQ) ICIMOD (2015)IDMC 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.
Figure 8 Summary of the key benchmarking results for housing damage, housing destruction and population displacement
in the following earthquakes: 2021 MW7.2 Nippes in Haiti, 2016 MW7.0 Kumamoto in Japan, and 2015 MW7.8 Gorkha in Nepal.
The conventional practice in earthquake risk assess-
ment is to consider housing destruction as the sole
driver of population displacement, which is imple-
mented in the three scenario models herein. This con-
ventional approach oers a way to estimate potential
long-term housing needs, which can provide useful
rapid situational awareness and inform early recovery
decisions. The results of this simplied approach are
compared with ocially reported statistics and alterna-
tive mobile location data-based estimates.
The scenario model estimates are largely consistent
with what was ocially reported for these earthquake
events, albeit with a large range of uncertainty. How-
ever, the ocial statistics are oen underpinned by the
same fundamental assumption (i.e., housing destruc-
tion leads to displacement). Thus, a fully indepen-
dent comparison is not possible to validate the mod-
els. Additionally, scenario models require several as-
sumptions across the rupture characterization, ground
motion model selection, building and population ex-
posure derivation, and fragility function assignment.
Each of these model inputs inuences the resulting risk
estimates, and this epistemic uncertainty complicates
comparisons. Various observational data could be used
to better-constrain model predictions and reduce un-
certainty: while this study only incorporated recorded
ground shaking from seismic stations, other relevant
sources of observational data, such as eld surveys or
remote sensing-derived damage data, could also poten-
tially be incorporated (Loos et al.,2023). Validation is
further complicated by the use of many dierent met-
rics to quantify displaced populations (i.e., rendered
homeless, sheltered, evacuated). Moreover, neither the
scenario models nor the ocial reports oered a view
on population return or the duration of displacement.
Mobile location data could theoretically close the data
gap on displacement duration and return, but those es-
timates are less consistent with the scenario model es-
timates and the ocially reported data. In particular,
the mobile location data-based estimates for the Nepal
and Haiti earthquakes are much lower than the scenario
model estimates and the ocially reported estimates.
These discrepancies could result from the criteria used
to dene displacement in these data-driven approaches
or the un-representativeness of the data sample. In
some cases, discrepancies may exist because the con-
sidered population is restricted to specic areas (e.g.,
within the Kathmandu Valley) or that there is an insuf-
cient spatial resolution used in the displacement cri-
teria (i.e., neglecting those who le their habitual res-
idence but migrated short distances). In other cases,
discrepancies could exist because the movements of
the sample population (i.e., those with mobile phones)
are not fully representative of the aected population
10 SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
(e.g., lower income populations may be less likely to
have mobile phones and may experience disproportion-
ate damage, elderly populations may be less likely to
carry phones and may also inhabit older buildings more
prone to damage). This data representativeness issue
is likely even more prevalent in countries or communi-
ties with lower rates of mobile phone ownership. Fur-
ther evaluation of both the displacement criteria and
the sample population’s representativeness may be re-
quired to instill condence in the use of mobile loca-
tion data to estimate population displacement aer dis-
asters.
The results from this benchmarking study demon-
strate the potential use of disaster risk models to eval-
uate population displacement and potential long-term
housing needs with minimal information. However, by
only considering housing destruction, additional fac-
tors known to inuence household displacement dura-
tion and return into the recovery phase (e.g., home own-
ership, place attachment, social capital) are neglected.
Moreover, critical factors inuencing shelter-seeking
behavior (e.g., utility disruption, weather) are ignored.
Thus, the standard practice of only considering housing
destruction can provide useful, rapid situational aware-
ness, but fails to capture a more holistic view of popula-
tion displacement aer disaster.
Ultimately, various metrics of population displace-
ment (e.g., population rendered homeless due to hous-
ing destruction, evacuations in the emergency phase,
shelter needs, return rates) can help expand the metrics
quantied within “what-if” scenarios and inform cost-
benet studies to capture more equitable and people-
centered metrics beyond economic loss.
Data and code availability
The GEM Foundation’s Earthquake Scenario Database
(ESD) is publicly available at https://github.com/gem/
earthquake-scenarios.
The GEM Foundation’s Global Exposure Model is
publicly available at the rst administrative level
at https://github.com/gem/global_exposure_model. For
ner resolutions, please send a request at https://
www.globalquakemodel.org/products.
The soware used to conduct the scenario analyses,
OpenQuake, is open source and publicly available at
https://github.com/gem/oq-engine. Training materials
to learn how to use the OpenQuake Engine are freely
available at https://www.training.openquake.org/.
Seismic station intensity estimates were downloaded
from multiple sources, and combined into a consistent
format for analysis.
• US Geological Survey (USGS) ShakeMap’s station
list (https://earthquake.usgs.gov/data/shakemap/)
• National Research Institute for Earth Sci-
ence and Disaster Prevention (NIED)’s
strong motion seismograph networks (https:
//www.kyoshin.bosai.go.jp/)
• Center for Engineering Strong Mo-
tion Data (CESDMD)’s archive (https:
//www.strongmotioncenter.org/)
Competing interests
The authors state that no competing interests inu-
enced this study.
Acknowledgements
This work was partly funded by the University College
London Overseas Research Scholarship (ORS) and the
Willis Towers Watson Research Network.
Additionally, the authors thank the Global Earth-
quake Model (GEM) Foundation for providing access to
the exposure and fragility models used in this study and
the Internal Displacement Monitoring Centre (IDMC)
for providing their triangulation data, which included
further details on the displacement metric type and data
source published within the Global Internal Displace-
ment Database (GIDD).
References
Akkar, S., Sandıkkaya, M. A., and Bommer, J. J. Empirical ground-
motion models for point-and extended-source crustal earth-
quake scenarios in Europe and the Middle East. Bulletin of earth-
quake engineering, 12:359–387, 2014.
Atkinson, G. M. and Boore, D. M. Empirical ground-motion rela-
tions for subduction-zone earthquakes and their application to
Cascadia and other regions. Bulletin of the Seismological Society
of America, 93(4):1703–1729, 2003.
Beguería, S. Validation and Evaluation of Predictive Models in Haz-
ard Assessment and Risk Management. Natural Hazards, 37(3):
315–329, Mar. 2006. doi: 10.1007/s11069-005-5182-6.
Bengtsson, L., Lu, X., Thorson, A., Garfield, R., and Schreeb, J.v. Im-
proved Response to Disasters and Outbreaks by Tracking Pop-
ulation Movements with Mobile Phone Network Data: A Post-
Earthquake Geospatial Study in Haiti. PLOS Medicine, 8(8):
e1001083, Aug. 2011. doi: 10.1371/journal.pmed.1001083.
Bhattacharya, Y. and Kato, T. Development of an Agent-Based
Model on the Decision-Making of Dislocated People Aer Dis-
asters. In Geertman, S. C. M., Pettit, C., Goodspeed, R., and
Staans, A., editors, Urban Informatics and Future Cities, The
Urban Book Series, pages 387–406. Springer International Pub-
lishing, Cham, 2021. doi: 10.1007/978-3-030-76059-5_20.
Bhattarai, M., Adhikari, L. B., Gautam, U. P., Laurendeau, A.,
Labonne, C., Hoste-Colomer, R., Sèbe, O., and Hernandez, B.
Overview of the large 25 April 2015 Gorkha, Nepal, earthquake
from accelerometric perspectives. Seismological Research Let-
ters, 86(6):1540–1548, 2015.
Binder, S. B., Baker, C. K., and Barile, J. P. Rebuild or Relo-
cate? Resilience and Postdisaster Decision-Making Aer Hurri-
cane Sandy. American Journal of Community Psychology, 56(1):
180–196, Sept. 2015. doi: 10.1007/s10464-015-9727-x.
Blumenstock, J. and Eagle, N. Mobile divides: gender, socioe-
conomic status, and mobile phone use in Rwanda. In Pro-
ceedings of the 4th ACM/IEEE International Conference on In-
formation and Communication Technologies and Development,
pages 1–10, London United Kingdom, Dec. 2010. ACM. doi:
10.1145/2369220.2369225.
Burton, H., Kang, H., Miles, S., Nejat, A., and Yi, Z. A framework and
case study for integrating household decision-making into post-
11 SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
earthquake recovery models. International Journal of Disaster
Risk Reduction, 37:101167, 2019.
CDEMA. Haiti Earthquake: Final Situation Report #12.
Technical report, St. Michael, Barbados, Sept. 2021.
https://www.cdema.org/images/2021/09/FINAL_CDEMA_
Situation_Report_12_Haiti_Earthquake_14Sep2021.pdf.
Chadha, R. K., Srinagesh, D., Srinivas, D., Suresh, G., Sateesh, A.,
Singh, S. K., Pérez-Campos, X., Suresh, G., Koketsu, K., Masuda,
T., Domen, K., and Ito, T. CIGN, A Strong-Motion Seismic Net-
work in Central Indo-Gangetic Plains, Foothills of Himalayas:
First Results. Seismological Research Letters, 87(1):37–46, Dec.
2015. doi: 10.1785/0220150106.
Chiou, B. S.-J. and Youngs, R. R. Update of the Chiou and
Youngs NGA model for the average horizontal component of
peak ground motion and response spectra. Earthquake Spec-
tra, 30(3):1117–1153, 2014.
Cong, Z., Nejat, A., Liang, D., Pei, Y., and Javid, R. J. Individual relo-
cation decisions aer tornadoes: a multi-level analysis. Disas-
ters, 42(2):233–250, 2018. doi: 10.1111/disa.12241.
Costa, R., Haukaas, T., and Chang, S. E. Predicting popu-
lation displacements aer earthquakes. Sustainable
and Resilient Infrastructure, 7(4):253–271, July 2022. doi:
10.1080/23789689.2020.1746047.
Cremen, G., Galasso, C., and McCloskey, J. A Simulation-Based
Framework for Earthquake Risk-Informed and People-Centered
Decision Making on Future Urban Planning. Earth’s Future, 10
(1):e2021EF002388, Jan. 2022. doi: 10.1029/2021EF002388.
Crowley, H., Silva, V., Kalakonas, P., Martins, L., Weatherill, G., Piti-
lakis, K., Riga, E., Borzi, B., and Faravelli, M. Verification of
the European seismic risk model (ESRM20). In Proceedings of
the 17th world conference on earthquake engineering, Sendai,
Japan, volume 27, 2020. https://wcee.nicee.org/wcee/article/
17WCEE/8b-0045.pdf.
DeWaard, J., Johnson, J.,and Whitaker, S. Internal migration inthe
United States: A comprehensive comparative assessment of the
Consumer Credit Panel. Demographic Research, 41:953–1006,
Oct. 2019. doi: 10.4054/DemRes.2019.41.33.
DeWaard, J., Johnson, J.E., and Whitaker, S. D. Out-migration from
and return migration to Puerto Rico aer Hurricane Maria: evi-
dence from the consumer credit panel. Population and Environ-
ment, 42(1):28–42, Sept. 2020. doi: 10.1007/s11111-020-00339-
5.
Elliott, J. R. and Pais, J. Race, class, and Hurricane Katrina: So-
cial dierences in human responses to disaster. Social Sci-
ence Research, 35(2):295–321, June 2006. doi: 10.1016/j.ssre-
search.2006.02.003.
Engler, D. T., Worden, C. B., Thompson, E. M., and Jaiswal, K. S.
Partitioning Ground Motion Uncertainty When Conditioned on
Station Data. Bulletin of the Seismological Society of America,
112(2):1060–1079, Jan. 2022. doi: 10.1785/0120210177.
Esnard, A.-M. and Sapat, A. Displaced by Disaster: Recovery and Re-
silience in a Globalizing World. Routledge, New York, July 2014.
doi: 10.4324/9780203728291.
FlowMinder. Haiti: Earthquake on 14 August 2021 (Version 1.2).
Technical report, Aug. 2021. https://www.flowminder.org/
media/dpxfefl4/haitiearthquake_report_27-aug_report-2_
eng_v1-2_final.pdf.
Frias-Martinez, V. and Virseda, J. On the relationship between
socio-economic factors and cell phone usage. In Proceedings
of the Fih International Conference on Information and Commu-
nication Technologies and Development, ICTD ’12, pages 76–84,
New York, NY, USA, Mar. 2012. Association for Computing Ma-
chinery. doi: 10.1145/2160673.2160684.
Greer, A. Household residential decision-making in the wake of
disaster: cases from Hurricane Sandy. PhD thesis, University
of Delaware, 2015. https://udspace.udel.edu/handle/19716/
31364.
Grinberger, A. Y. and Felsenstein, D. Dynamic agent based sim-
ulation of welfare eects of urban disasters. Computers, En-
vironment and Urban Systems, 59:129–141, Sept. 2016. doi:
10.1016/j.compenvurbsys.2016.06.005.
Groen, J. A. and Polivka, A. E. Going home aer Hurricane Katrina:
Determinants of return migration and changes in aected areas.
Demography, 47(4):821–844, 2010.
Grünthal, G. European macroseismic scale 1998 (EMS-98). 1998.
https://gfzpublic.gfz-potsdam.de/rest/items/item_227033_2/
component/file_227032/content.
Guadagno, L. and Yonetani, M. Displacement risk: Unpacking a
problematic concept for disaster risk reduction. International
Migration, 61(5):13–28, 2023. doi: 10.1111/imig.13004.
Hallegatte, S., Vogt-Schilb, A., Rozenberg, J., Bangalore, M., and
Beaudet, C. From Poverty to Disaster and Back: a Review of
the Literature. Economics of Disasters and Climate Change, 4(1):
223–247, Apr. 2020. doi: 10.1007/s41885-020-00060-5.
Hayes, G. P., Briggs, R. W., Barnhart, W. D., Yeck, W. L., McNamara,
D. E., Wald, D. J., Nealy, J. L., Benz, H. M., Gold, R. D., and Jaiswal,
K. S. Rapid characterization of the 2015 M w 7.8 Gorkha, Nepal,
earthquake sequence and its seismotectonic context. Seismo-
logical Research Letters, 86(6):1557–1567, 2015.
Heath, D. C., Wald, D. J., Worden, C. B., Thompson, E. M., and
Smoczyk, G. M. A global hybrid VS 30 map with a topographic
slope –based default and regional map insets. Earthquake
Spectra, 36(3):1570–1584, 2020.
Hinojosa, J. Two Sides of the Coin of Puerto Rican Migra-
tion: Depopulation in Puerto Rico and the Redefinition
of the Diaspora. Centro Journal, 30(3), 2018. https:
//www.academia.edu/download/59765896/J.HINOJOSA_
CENTROJOURNAL-FALL2018.pdf.
Hinojosa, J. and Meléndez, E. Puerto Rican Exodus: One Year Since
Hurricane Maria. Technical Report Centro RB2018-05, Centro
Library and Archives, New York, NY, USA, Sept. 2018.
Hough, S. E., Martin, S. S., Gahalaut, V., Joshi, A., Landes, M., and
Bossu, R. A comparison of observed and predicted ground mo-
tions from the 2015 MW7.8 Gorkha, Nepal, earthquake. Natural
Hazards, 84(3):1661–1684, Dec. 2016. doi: 10.1007/s11069-016-
2505-8.
Hoyos, M. C. and Silva, V. Exploring benefit cost analysis to sup-
port earthquake risk mitigation in Central America. Interna-
tional Journal of Disaster Risk Reduction, 80:103162, 2022. doi:
https://doi.org/10.1016/j.ijdrr.2022.103162.
ICIMOD. Lessons from Nepal’s Gorkha earthquake 2015. Technical
report, Kathmandu, Nepal, 2015.
IDMC. Global Internal Displacement Database. https://
www.internal-displacement.org/database.
IDMC. GRID Methodological Annex. Technical report, 2018. https:
//www.internal-displacement.org/global-report/grid2018/
downloads/report/2018-GRID-methodological-annex.pdf.
IDMC. Disaster Displacement - A global review, 2008-2018. Tech-
nical report, 2019. https://www.internal-displacement.org/
publications/disaster-displacement-a-global-review.
IDMC. GRID Methodology. Technical report, 2020.
https://www.internal-displacement.org/global-report/
grid2020/downloads/2020-IDMC-GRID-methodology.pdf.
IDMC. Urban case study: Ba Town, Fiji. Technical report, July
2022a. https://www.internal-displacement.org/publications/
pacific-response-to-disaster-displacement-urban-case-study-
12 SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
ba-town-fiji/.
IDMC. Urban case study: Port Vila, Vanuatu. Technical report, July
2022b. https://www.internal-displacement.org/publications/
pacific-response-to-disaster-displacement-urban-case-study-
port-vila-vanuatu/.
JCO. Disaster Report for 2016 Kumamoto earthquake. Tech-
nical report, Apr. 2017. https://www.bousai.go.jp/updates/
h280414jishin/pdf/h280414jishin_39.pdf.
Kalakonas, P., Silva, V., Mouyiannou, A., and Rao, A. Exploring the
impact of epistemic uncertainty on a regional probabilistic seis-
mic risk assessment model. Natural Hazards, 104(1):997–1020,
2020. doi: https://doi.org/10.1007/s11069-020-04201-7.
Kolbe, A. R., Hutson, R. A., Shannon, H., Trzcinski, E., Miles, B.,
Levitz, N., Puccio, M., James, L., Noel, J. R., and Muggah, R.
Mortality, crime and access to basic needs before and aer the
Haiti earthquake: a random survey of Port-au-Prince house-
holds. Medicine, Conflict and Survival, 26(4):281–297, Oct. 2010.
doi: 10.1080/13623699.2010.535279.
Lee, C.-C., Chou, C., and Mostafavi, A. Specifying Evacuation Return
and Home-switch Stability During Short-term Disaster Recov-
ery Using Location-based Data. Scientific Reports, 12(1):15987,
Sept. 2022. doi: 10.1038/s41598-022-20384-4.
Lee, Y.-J., Sugiura, H., and Gečienė, I. Stay or Relocate: The
Roles of Networks Aer the Great East Japan Earthquake. In
Jones, E. C. and Faas, A. J., editors, Social Network Analysis of
Disaster Response, Recovery, and Adaptation, pages 223–238.
Butterworth-Heinemann, Jan. 2017. doi: 10.1016/B978-0-12-
805196-2.00015-7.
Liel, A. B. and Deierlein, G. G. Cost-Benefit Evaluation of Seis-
mic Risk Mitigation Alternatives for Older Concrete Frame Build-
ings. Earthquake Spectra, 29(4):1391–1411, Nov. 2013. doi:
10.1193/030911EQS040M.
Lin, Y.-S. Development of algorithms to estimate post-disaster popu-
lation dislocation—a research-based approach. Texas A&M Uni-
versity, 2009.
Lines, R., Faure Walker, J. P., and Yore, R. Progression through
emergency and temporary shelter, transitional housing and
permanent housing: A longitudinal case study from the 2018
Lombok earthquake, Indonesia. International Journal of Dis-
aster Risk Reduction, 75:102959, June 2022. doi: 10.1016/j.ij-
drr.2022.102959.
Loos, S., Lallemant, D., Baker, J., McCaughey, J., Yun, S.-H., Bud-
hathoki, N., Khan, F., and Singh, R. G-DIF: A geospatial data in-
tegration framework to rapidly estimate post-earthquake dam-
age. Earthquake Spectra, 36(4):1695–1718, Nov. 2020. doi:
10.1177/8755293020926190.
Loos, S., Lallemant, D., Khan, F., McCaughey, J. W., Banick, R., Bud-
hathoki, N., and Baker, J. W. A data-driven approach to rapidly
estimate recovery potential to go beyond building damage aer
disasters. Communications Earth & Environment, 4(1):1–12, Feb.
2023. doi: 10.1038/s43247-023-00699-4.
Love, T. Population movement aer natural disasters: a literature
review and assessment of Christchurch data. Technical report,
Sapere Research Group, Wellington, New Zealand, 2011.
Lu, X., Bengtsson, L., and Holme, P. Predictability of population
displacement aer the 2010 Haiti earthquake. Proceedings of
the National Academy of Sciences, 109(29):11576–11581, July
2012. doi: 10.1073/pnas.1203882109.
Martins, L. and Silva, V. Development of a fragility and vulnerability
model for global seismic risk analyses. Bulletin of Earthquake
Engineering, 19(15):6719–6745, Dec. 2021. doi: 10.1007/s10518-
020-00885-1.
Mayer, J., Moradi, S., Nejat, A., Ghosh, S., Cong, Z., and Liang, D.
Drivers of post-disaster relocations: The case of Moore and Hat-
tiesburg tornados. International Journal of Disaster Risk Reduc-
tion, 49:101643, Oct. 2020. doi: 10.1016/j.ijdrr.2020.101643.
McAdam, J. Evacuations: a form of disaster displace-
ment? Forced Migration Review, (69):56–57, 2022.
https://www.proquest.com/docview/2647725690/abstract/
DB0755D2F79B4311PQ/1.
Milusheva, S., Bjorkegren, D., and Viotti, L. Assessing Bias in Smart-
phone Mobility Estimates in Low Income Countries. In ACM SIG-
CAS Conference on Computing and Sustainable Societies (COM-
PASS), pages 364–378, Virtual Event Australia, June 2021. ACM.
doi: 10.1145/3460112.3471968.
Nawrotzki, R. J., Brenkert-Smith, H., Hunter, L. M., and Champ, P. A.
Wildfire-Migration Dynamics: Lessons from Colorado’s Four-
mile Canyon Fire. Society & Natural Resources, 27(2):215–225,
Feb. 2014. doi: 10.1080/08941920.2013.842275.
Nejat, A. and Ghosh, S. LASSO Model of Postdisaster Housing Re-
covery: Case Study of Hurricane Sandy. Natural Hazards Re-
view, 17(3):04016007, Aug. 2016. doi: 10.1061/(ASCE)NH.1527-
6996.0000223.
Newell, J., Beaven, S., and Johnston, D. M. Population move-
ments following the 2010-2011 Canterbury Earthquakes: Sum-
mary of research workshops November 2011 and current evi-
dence. Technical report, 2012.
Newzoo. Global Mobile Market Report 2017. Technical re-
port, Apr. 2017. https://newzoo.com/resources/trend-reports/
global-mobile-market-report-light-2017.
Paul, N., Galasso, C., and Baker, J. Household Displacement and
Return in Disasters: A Review. Natural Hazards Review, 25(1):
03123006, Feb. 2024. doi: 10.1061/NHREFO.NHENG-1930.
Plyer, A., Bonaguro,J., and Hodges, K. Using administrative data to
estimate population displacement and resettlement following
a catastrophic U.S. disaster. Population and Environment, 31(1):
150–175, Jan. 2010. doi: 10.1007/s11111-009-0091-3.
Price, D. Population and household trends in Christchurch post
February 22 earthquake. In Population and Employment Eects
of the Christchurch Earthquakes workshop, 2011.
Quarantelli, E. L. General and particular observations on shelter-
ing and housing in American disasters. Disasters, 6(4):277–281,
1982. doi: 10.1111/j.1467-7717.1982.tb00550.x.
Quarantelli, E. L. Patternsof sheltering and housing in US disasters.
Disaster Prevention and Management: An International Journal,
4(3):43–53, Jan. 1995. doi: 10.1108/09653569510088069.
Rajaure, S., Asimaki, D., Thompson, E. M., Hough, S., Martin, S.,
Ampuero, J. P., Dhital, M. R., Inbal, A., Takai, N., Shigefuji, M., Bi-
jukchhen, S., Ichiyanagi, M., Sasatani, T., and Paudel, L. Char-
acterizing the Kathmandu Valley sediment response through
strong motion recordings of the 2015 Gorkha earthquake se-
quence. Tectonophysics, 714-715:146–157, Sept. 2017. doi:
10.1016/j.tecto.2016.09.030.
Sharygin, E. Estimating Migration Impacts of Wildfire: California’
s 2017 North Bay Fires. In Karácsonyi, D., Taylor, A., and Bird,
D., editors, The Demography of Disasters: Impacts for Popula-
tion and Place, pages 49–70. Springer International Publishing,
Cham, 2021. doi: 10.1007/978-3-030-49920-4_3.
Silva, V. Critical Issues in Earthquake ScenarioLoss Modeling. Jour-
nal of Earthquake Engineering, 20(8):1322–1341, Nov. 2016. doi:
10.1080/13632469.2016.1138172.
Silva, V. Critical Issues on Probabilistic Earthquake Loss Assess-
ment. Journal of Earthquake Engineering, 22(9):1683–1709, Oct.
2018. doi: 10.1080/13632469.2017.1297264.
Silva, V. and Horspool, N. Combining USGS ShakeMaps and the
OpenQuake-engine for damage and loss assessment. Earth-
13 SEISMICA | volume 3.2 | 2024
SEISMICA |RESEARCH ARTICLE | Population displacement aer earthquakes
quake Engineering & Structural Dynamics, 48(6):634–652, 2019.
doi: 10.1002/eqe.3154.
Silva, V., Crowley, H., Pagani, M., Monelli, D., and Pinho, R. Develop-
ment of the OpenQuake engine, the Global Earthquake Model’s
open-source soware for seismic risk assessment. Natural Haz-
ards, 72:1409–1427, 2014.
Silva, V., Amo-Oduro, D., Calderon, A., Costa, C., Dabbeek, J.,
Despotaki, V., Martins, L., Pagani, M., Rao, A., and Simionato, M.
Development of a global seismic risk model. Earthquake Spec-
tra, 36(1_suppl):372–394, 2020.
The Asia Foundation. Independent Impacts and Re-
covery Monitoring Phase Five. Technical report,
The Asia Foundation, San Francisco, CA, USA, 2019.
https://asiafoundation.org/wp-content/uploads/2021/03/
IRM-Nepal_Aid-and-Recovery-in-Post-Earthquake-Nepal-
Qualititative-Field-MonitoringNovember-2019_EN.pdf.
Van De Lindt, J. W., Peacock, W. G., Mitrani-Reiser, J., Rosenheim,
N., Deniz, D., Dillard, M., Tomiczek, T., Koliou, M., Graettinger, A.,
Crawford, P. S., Harrison, K., Barbosa, A., Tobin, J., Helgeson, J.,
Peek, L., Memari, M., Sutley, E. J.,Hamideh, S., Gu, D., Cauman,
S., and Fung, J. Community Resilience-Focused TechnicalInves-
tigation of the 2016 Lumberton, North Carolina, Flood: An Inter-
disciplinary Approach. Natural Hazards Review, 21(3):04020029,
Aug. 2020. doi: 10.1061/(ASCE)NH.1527-6996.0000387.
Ward, P. J., Blauhut, V., Bloemendaal, N., Daniell, J. E., de Ruiter,
M. C., Duncan, M. J., Emberson, R., Jenkins, S. F., Kirschbaum,
D., Kunz, M., Mohr, S., Muis, S., Riddell, G.A., Schäfer, A., Stanley,
T., Veldkamp, T. I. E., and Winsemius, H. C. Review article: Nat-
ural hazard risk assessments at the global scale. Natural Haz-
ards and Earth System Sciences, 20(4):1069–1096, Apr. 2020. doi:
10.5194/nhess-20-1069-2020.
Wesolowski, A., Eagle, N., Noor, A. M., Snow, R. W., and Buc-
kee, C. O. Heterogeneous Mobile Phone Ownership and Us-
age Patterns in Kenya. PLoS ONE, 7(4):e35319, Apr. 2012. doi:
10.1371/journal.pone.0035319.
Wesolowski, A., Eagle, N., Noor, A. M., Snow, R. W., and Buckee,
C. O. The impact of biases in mobile phone ownership on esti-
mates of human mobility. Journal of The Royal Society Interface,
10(81):20120986, Apr. 2013. doi: 10.1098/rsif.2012.0986.
Wilson, R., zu Erbach-Schoenberg, E., Albert, M., Power, D.,
Tudge, S., Gonzalez, M., Guthrie, S., Chamberlain, H., Brooks,
C., Hughes, C., Pitonakova, L., Buckee, C., Lu, X., Wetter, E.,
Tatem, A., and Bengtsson, L. Rapid and Near Real-Time Assess-
ments of Population Displacement Using Mobile Phone Data
Following Disasters: The 2015 Nepal Earthquake. PLoS Cur-
rents, 8:ecurrents.dis.d073fbece328e4c39087bc086d694b5c,
Feb. 2016. doi: 10.1371/cur-
rents.dis.d073fbece328e4c39087bc086d694b5c.
World Bank Group. Mobile cellular subscriptions (per 100 people),
2021. https://data.worldbank.org.
Yabe, T., Sekimoto, Y., Tsubouchi, K., and Ikemoto, S. Cross-
comparative analysis of evacuation behavior aer earthquakes
using mobile phone data. PLOS ONE, 14(2):e0211375, Feb. 2019.
doi: 10.1371/journal.pone.0211375.
Yabe, T., Tsubouchi, K., Fujiwara, N., Sekimoto, Y., and Ukkusuri,
S. V. Understanding post-disaster population recovery patterns.
Journal of The Royal Society Interface, 17(163):20190532, Feb.
2020. doi: 10.1098/rsif.2019.0532.
Yabe, T., Jones, N. K. W., Lozano-Gracia, N., Khan, M. F., Ukkusuri,
S. V., Fraiberger, S., and Montfort, A. Location Data Reveals Dis-
proportionate Disaster Impact Amongst the Poor: A Case Study
of the 2017 Puebla Earthquake Using Mobilkit, July 2021. doi:
10.48550/arXiv.2107.13590.
Yabe, T., Jones, N. K. W., Rao, P. S. C., Gonzalez, M. C., and Ukkusuri,
S. V. Mobile phone location data for disasters: A review from
natural hazards and epidemics. Computers, Environment and
Urban Systems, 94:101777, June 2022. doi: 10.1016/j.compen-
vurbsys.2022.101777.
Yepes-Estrada, C., Calderon, A., Costa, C., Crowley, H., Dabbeek,
J., Hoyos, M. C., Martins, L., Paul, N., Rao, A., and Silva, V.
Global building exposure model for earthquake risk assess-
ment. Earthquake Spectra, 39(4):2212–2235, Nov. 2023. doi:
10.1177/87552930231194048.
The article Population displacement aer earthquakes:
benchmarking predictions based on housing damage ©
2024 by Nicole Paul is licensed under CC BY 4.0.
14 SEISMICA | volume 3.2 | 2024
