Assessing the impact of Syrian refugees on earthquake fatality estimations in southeast Turkey

The influx of millions of Syrian refugees into Turkey has rapidly changed the population distribution along the Dead Sea Rift and East Anatolian fault zones. In contrast to other countries in the Middle East where refugees are accommodated in camp environments, the majority of displaced individuals in Turkey are integrated into local cities, towns, and villages – placing stress on urban settings and increasing potential exposure to strong earthquake shaking. Yet displaced populations are often unaccounted for in the census-based population models used in earthquake fatality estimations. This study creates a minimally modeled refugee gridded population model and analyzes its impact on semi-empirical fatality estimations across southeast Turkey. Daytime and nighttime fatality estimates were produced for five fault segments at earthquake magnitudes 5.8, 6.4, and 7.0. Baseline fatality estimates calculated from census-based population estimates for the study area varied in scale from tens to thousands of fatalities, with higher death totals in nighttime scenarios. Refugee fatality estimations were analyzed across 500 semi-random building occupancy distributions. Median fatality estimates for refugee populations added non-negligible contributions to earthquake fatalities at four of five fault locations, increasing total fatality estimates by 7–27 %. These findings communicate the necessity of incorporating refugee statistics into earthquake fatality estimations in southeast Turkey and the ongoing importance of placing environmental hazards in their appropriate regional and temporal context.

Under either explanation, refugee settlement in southeastern Turkey represents a migration away from a stable tectonic setting into an area characterized by frequent earthquake activity.

Historical seismicity
There is a wealth of information detailing the long history of earthquake activity on the East Anatolian and Dead Sea Rift fault systems (Ambraseys, 2009;Sbeinati et al., 2005;Barka and Kadinsky-Cade, 1988;Garfunkel et al., 1981). Ambraseys 25 (2009) provides a detailed overview of historical seismicity in the region, with Sbeinati et al. (2005) providing additional information on Syrian earthquakes. Both the East Anatolian and Dead Sea Rift fault systems have seen a recent quiescence in seismic activity, but paleoseismic evidence indicates a consistent long term pattern of infrequent large earthquakes (Ambraseys, 1989;Meghraoui et al., 2003). Large earthquakes in southern Turkey first appear in the historical records in 148 B.C.E. in the writings of John Malalas, who chronicles the destruction of the city of Antioch due to the 'Wrath of God', a phrase often used 30 to describe earthquake events (Ambraseys, 2009). The city of Antioch alone, located in the modern day Hatay province, is   Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-69, 2017 Manuscript under review for journal Nat. Hazards Earth Syst. Sci. Discussion started: 13 March 2017 c Author(s) 2017. CC-BY 3.0 License. Figure 3. The distribution of earthquake shaking as gathered from historical documents and modern seismic networks, compiled in (Sesetyan et al., 2013). shaken over forty times before 1900 C.E.. Figure 3 plots seismic activity greater than magnitude 5.0 across the study area over the last millennia, showing an fairly even distribution across the length of the fault zones.
Historical records also provide insight into the human impact of several notable earthquakes.The earthquakes that destroy Antioch in 115 C.E. and 526 C.E. are estimated to have killed 250,000 or more individuals each. If these numbers are correct, both earthquakes fall into the top ten most deadly earthquakes of all time (Musson, 2001) (the death estimates may be exagger- 5 ated, but are generally considered to be plausible (Ambraseys, 2009)). The 526 C.E. earthquake is particularly notable, striking on the 29th of May, Ascension Day. Ambraseys (2009) notes that the high death totals for this earthquake (250,000-300,000) were likely amplified by the influx of visitors into the city. The last Turkish census was completed in 2011 before the onset of Syrian mass migration. Therefore, most population models built from census-based sources do not account for the presences of Syrian refugees. This is not an intentional error (The
Gridded Population of the World, version 4 dataset (GPWv4) (Doxsey-Whitfield et al., 2015) explicitly states this particular shortcoming), but rather a systematic problem associated with infrequent data collection. Any forward modeled population dataset for Turkey based on pre-2011 data will mischaracterize true populations in high migration areas unless refugees are explicitly included. Furthermore, the uncertainty associated with the sub-province level position of refugees complicates their inclusion in disaggregated grid-based datasets. Population models that include Syrian migration do exist, most notably Oak As a framework for modifying regional census data for inter-period migration events, a geographic information systems (GIS) workflow was utilized to construct a regional refugee inclusive gridded population model using freely available data from Turkey's Address Based Population Registration System (ABPRS) and the Turkish Directorate General of Migration 10 Management. This model, like the GPWv4, employs a minimally modeled aeral distribution process that disaggregates administrative population counts into cells of equal population. Turkish district level boundaries from the GADM database of Global Administrative Areas (GAA, 2015), clipped to the area of interest, were first converted into three kilometer grid cells and equally distributed 2015 ABPRS populations according to the proportional number of cells in each district.
Refugee migration data is monitored at the province level, one administrative boundary larger than the ABPRS estimates. As 15 mentioned above, the exact position of non-camped Syrian refugees within their respective provinces is unknown. Accordingly, the existing district level population distribution was used as a proxy for refugee settlement patterns. The non-camped refugee population was distributed according to each district's relative percentage of its corresponding province province percentage.
Known camped refugee populations were assigned to their residing district and removed from the populations otherwise distributed. The model was finalized by repeating the process used above for distributing ABPRS populations to allocate refugees 20 into equally populated grid cells. The resulting gridded population model ( Fig. 4) is spatially consistent and has discrete values for base population and migrated population.

Advantages and drawbacks
This population model has a number of advantages and drawbacks over other freely available population models. The primary advantage being that, in contrast to other aeral grid models, this model explicitly accounts for the spatial distribution refugee  ulations exhibit different spatial clustering patterns, but this information is difficult to model without additional constraining information. The Turkish government has started tracking province level registration and implemented freedom of movement restrictions as of August, 2015, but these regulations still allow for movement for refugees within their registered provinces (Çorabatır, 2016). Assuming an equivalent population distribution to that of existing populations maintains the regional urbanrural breakdown-an important element in building-type assignment for loss estimation. Outside of specific refugee camp lo-5 cations (which have been accounted for), there is not clear evidence for assigning an alternative distribution pattern. Secondly, refugee populations were aerally distributed equally into district level grid cells-the same assumption made for Turkish populations. This assumption retains consistency between population types in the absence of sufficient reason to minimally model refugees differently than that of existing population.

Earthquake scenarios
10 Earthquake scenarios are an important tool for emergency management planning. Tools like the USGS' Prompt Assessment of Global Earthquake Risk (PAGER) system and FEMA's HAZUS software have been used in the U.S. for emergency planning and both the national and state level (FEMA, 2008;Chen et al., 2016;EERI, 2015). As part of the earthquake loss estimation process, synthetic ground motion fields were produced for a series of earthquake ruptures spanning five faults across southeastern Turkey. For each fault, moment magnitude 5.8, 6.4 and 7.0 earthquakes were simulated. This spread of earthquake 15 magnitudes reflects moderate to major earthquakes within the magnitude range of historical earthquakes in the area as seen in earthquake catalogs covering Turkey (Zare et al., 2014;Woessner et al., 2015). Five earthquake epicentral locations were   Table 1. 5 The Global Earthquake Model's OpenQuake software platform was utilized to produce ground motion fields for each earthquake scenario. OpenQuake's scenario-based hazard assessment implements ground motion prediction equations to estimate the geographic distribution of shaking intensity for a user-specified fault rupture (GEM, 2016). All of our scenarios utilized  Akkar and Bommer (2010), relevant to earthquakes in Europe and the Middle East. We accounted for seismic site amplification using Vs30 estimates from the USGS Global Vs30 Map server, which estimates Vs30 from topographic slope (Wald and Allen, 2007). OpenQuake implements site parameters by assigning each observation grid cell the site parameters of the nearest measurement in the Vs30 grid (GEM, 2016). For each earthquake scenario, we produced ten ground-motion fields-each resampling the aleatory uncertainty in the calculation. can be thought of as simplified analytical approaches, estimating casualties using empirical estimates of structural parameters.
The choice of methodology depends primarily on data availability and the scale of analysis (Jaiswal et al., 2011b).
This analysis seeks to assess the impact of refugee migration on the magnitude of earthquake casualty estimates. Fully empirical approaches that rely on fatalities in past earthquake events are poorly suited to analyzing short term variability in loss estimations, a fundamental portion of this analysis. A fully analytical approach would be preferable, but the structural engi-15 neering and building occupancy data does not exist to support such calculations. Even if structural information was available at the individual buildings level, improved refugee settlement data would be necessary for disaggregating province level refugee populations into specific building occupancy. Therefore, this study instead employs the hybrid, semi-empirical loss estimation approach detailed in (Jaiswal and Wald, 2010), given by Eq.
(1). 20 This approach estimates fatalities given a series of n grid cells and m structural types. Each grid cell's population P i is first broken out into a fractional percentage for a given structural type f ij . Fatalities are then calculated based on of the collapse rate of structural type j (CR j (S i )) at macroseismic intensity (S i ), and the fatality rate F R j of structure type j under collapse (Jaiswal and Wald, 2010).
Empirical data from the World Housing Encyclopedia (WHE)-PAGER project, phase I, was used to constrain collapse 25 rates. Jaiswal and Wald (2009b) provides estimates of the building stock distribution under the PAGER taxonomy along with estimated collapse percentages. It is noted that several of the collapse probabilities for Turkey are listed higher than those generalized for European macroseismic scale intensities across the entire WHE-PAGER phase I dataset (Jaiswal et al., 2011a).
Accordingly, when available, collapse rates are calculated using generalized fragility coefficients using Eq.  Turkish values from Porter et al. (2008) in their absence.
The casualty estimation process was implemented by loading the average peak ground acceleration (PGA) values for each scenario into GIS software and spatially joining them to the gridded population model. The grid raster for each scenario was exported to a CSV file containing each cell's district identifier code, PGA value, pre-migration population estimate, and 5 migrated population estimate. These CSV files were combined with the fragility information provided in Table 2 and population distribution information from Table 3 to implement Eq. (1) in R. For each grid cell, the PGA value was converted to Modified Mercalli Intensity values using the relationship of Wald et al. (1999) and the gridded cell populations were fractionally divided into building types according to the percentages in Table 2. While it is probable that the structural occupancy of refugees is different than that of local populations (approximately 25% of refugees live in makeshift or rubble housing (3RP, 2015)), 10 estimates for this distribution were not sufficiently known and both population sources were distributed equivalently. Any cell with a population density greater than 150 persons per kilometer was assigned an urban distribution while the rest were assigned a rural distribution (OECD, 1994). Each scenario's gridded population was distributed for both daytime and nighttime percentages using the corresponding information in Table 3.

Results and discussion
The total number of projected casualties in a particular earthquake scenario depends on the spatial overlap between population, shaking intensity, and structural type distribution. Adjustments in any of these parameters affects the number of projected casualties. This study estimated casualties for fifteen earthquake scenarios representing three earthquake magnitudes at five geographically distributed fault zones. Casualties were calculated for both daytime and nighttime hours and refugee inclusive 5 and refugee exclusive population scenarios, producing a total of four casualty estimates for each earthquake scenario. The results indicate intra-fault, inter-fault, and temporal differences in earthquake casualty projections across the study area, as well as notable casualty increases after refugee inclusion.
Tables 4 and 5 present the casualty numbers for each of the twelve earthquake scenarios. Table 4 provides baseline casualty estimates produced using the gridded population model before Syrian refugee adjustment, while Table 5 updates the estimations 10 using the refugee inclusive population model. It is important to highlight that the values presented in Tables 4 and 5 are not specific casualty predictions for future events, but representations of the order of magnitude that could be expected in events of varying size and location. Therefore, any conclusions drawn henceforth are scenario specific-and should only be generalized to other scenarios with appropriate caution.

Consistent trends 15
Refugees and local populations were disaggregated using a consistent methodology across population scenarios. Accordingly, several of the trends drawn from Table 4   tion. At all five fault locations, increasing earthquake magnitude from 5.8 to 6.4 resulted in a larger casualty increase (241% average) than the subsequent 6.4 to 7.0 magnitude increase (175% average). This reflects the non-linear relationship between earthquake magnitude, intensity, and building collapse rates. Increases in earthquake ground motions, not magnitude, are the driving force behind increase earthquake casualties. These results are consistent with general magnitude-intensity relationships. Areas simulated with PGA values high enough to produce building collapse are highly localized for magnitude 5.8 5 scenarios compared to larger magnitudes. However, local site conditions or poorly constructed buildings can amplify casualties even in moderate magnitude earthquakes-the 1960 Agadir earthquake in western Morocco resulted in 15,000 casualties despite a moment magnitude of 5.7 (Paradise, 2005).
Nighttime casualties were forecasted consistently higher than daytime casualties in all fault locations (an average of 160%).
This indicates that the building stock distribution occupied during working hours is less susceptible to collapse than the build-10 ing stock distribution occupied during living hours. These results follow out of Table 3 which shows populations generally transitioning from vulnerable masonry buildings to concrete structures during working hours. These findings add to the growing volume of research stressing the importance of including temporal elements into natural hazards studies (Chen et al., 2004;Ara, 2014;Aubrecht et al., 2012;Freire and Aubrecht, 2012;Guha-Sapir and Vos, 2011).

12
Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-69, 2017 Manuscript under review for journal Nat. Hazards Earth Syst. Sci. Discussion started: 13 March 2017 c Author(s) 2017. CC-BY 3.0 License. Inter-fault casualty differences reflect the location of particular faults to population centers. The two highest casualty locations, the Kırıkhan and Göksun fault segments, extend over several of the highest population districts in all of southeast Turkey.
With casualties estimates ranging from hundreds at magnitude 5.8 to thousands at magnitudes 6.4 and 7.0, these scenarios indicate serious consequences in the event of similar earthquake ruptures. The proximity of major cities and other population dense areas to faults strongly contributes to increased earthquake casualty estimates in this region. This has been historically true as 5 well, with earthquakes repeatedly destroying ancient cities near the modern day provinces of Hatay and Adana (Ambraseys, 2009).

Refugee related trends
Examining the new casualty and percent difference statistics in Table 5 shows that the presence of refugees has a non-trivial impact on earthquake impacts. Consistent increases in simulated casualties reflects the widespread presence of refugees amongst Understanding these variations explains the differences in the earthquake casualties by location. It is noted that our simulations find the refugee related casualty increases at the event level to be close to an average of the province level refugee populations within affected provinces. This implies that, in the absence of additional data, applying a 20 static severity increase in proportion to province migration statistics may be sufficient as a minimum estimate.

Conclusions
This study assessed the impact of Syrian refugee migration on earthquake casualty estimations in southeastern Turkey using a semi-empirical loss estimation technique on minimally modeled gridded population datasets created from refugee statistics and Turkish ABPRS district level population data. It was shown that using the refugee adjusted population model in the earthquake 25 fatality estimation process increased casualties in proportion to migrated population exposure. Earthquake scenarios on four of the five fault zones included in this study produced tens to hundreds of additional casualties after the inclusion of refugee data, with a maximum of 1579 extra casualties and a minimum of one extra casualty. While it naturally follows that places with population increases will sustain additional casualties in earthquakes, it had not yet been shown the degree to which current refugee populations impact loss estimations.

30
Characterizing the expected casualty increases related to refugee migration is an important step in loss estimation-even if a basic province-level correction is a sufficient adjustment. Disaster scale underestimations have the potential to greatly complicate the work of local governments and aid agencies working to respond to earthquake disasters (Jaiswal et al., 2011b).