THE NEAR REAL-TIME SYSTEM FOR ESTIMATING THE SEISMIC

DAMAGE IN ROMANIA (SEISDARO). RECENT UPGRADES AND

RESULTS.

Dragos TOMA-DANILA1,2, Carmen Ortanza CIOFLAN3, Constantin IONESCU4, Alexandru

TIGANESCU5

ABSTRACT

Romania is one of Europe’s countries with the greatest seismic hazard, mainly due to major earthquakes
originating in the subcrustal Vrancea Source. A future major earthquake could be a shock, therefore it is vital to
estimate in actual context the size and distribution of losses, based on scenarios (leading to seismic mitigation),
but also in real-time (with impact on effective emergency response). Since 2011, the National Institute for Earth
Physics engaged in this effort and created SeisDaRo – a system dedicated to the estimation of seismic damage in
Romania, in terms of building damage and fatalities. In 2016, a 3rd version, described in this paper, was released,
providing significant upgrades: a custom ShakeMap module for computing faster and more representative to
Vrancea Source intensities and accelerations, and a module representing an adaptation to Romania of the
USGS’s PAGER methodology. The other version’s module relying on the SELENA software is also included,
using the Improved Displacement Coefficient Method for estimating building failure probability and HAZUS
method for fatality estimation. As building and population database, SeisDaRo now uses newer data from the
2011 National Census. The system is functional in real-time, collecting input now also from the NIEP’s Early
Warning System; therefore, SeisDaRo 3 can generate a range for fatality estimates within 35 seconds after the
alarm is issued. All modules provide results in less than 6 minutes. In order to test the system, we used 5 relevant
Vrancea earthquake scenarios (with magnitudes from 6.4 to 7.7). Results show that although estimations using
the two different modules do not match exactly (SELENA apparently slightly overestimates), representative
insights regarding the potential damage scale of Vrancea earthquakes can be extracted.

Keywords: Seismic loss estimation; Real-time; Earthquake damage; Vrancea Source; Romania

1. INTRODUCTION

Immediately after an earthquake, an efficient management of emergency response is an essential

component of earthquake risk reduction. It was observed that poor emergency response can multiply
the death toll of an earthquake by a factor of 10 (Coburn and Spence, 2002). To help decision-makers
and planners to recommend post-earthquake actions quickly and knowingly, Earthquake Loss
Estimation (ELE) tools have been developed over the past few decades. These link, more or less
empirically, ground shaking intensity (provided by systems such as ShakeMap, or estimated directly in
ELE through prediction equations) to information regarding the physical (buildings) and socio-
economic vulnerability and distribution. However, ELE still involve a high degree of uncertainty.
They can be generated by other uncertainties in earthquake’s source parameters and by our lack of
knowledge on built environment, its fragility characteristics, and of the survival rates in an earthquake
(Erdik et al. 2011). Since ELE results need to be represented generally on maps with good visual

1
Researcher, National Institute for Earth Physics, Magurele, Romania, toma@infp.ro
2
PhD student, Faculty of Geography, University of Bucharest, Romania
3
Researcher, National Institute for Earth Physics, Magurele, Romania, cioflan@infp.ro
4
Researcher, General Manager, National Institute for Earth Physics, Magurele, Romania, viorel@infp.ro
5
Research assistant, National Institute for Earth Physics, Magurele, Romania, alexandru.tiganescu@infp.ro
impact, but reflecting also uncertainties, Geographic Information Systems (GIS) are often used for
final processing. The state of the art of ELE methodologies and software have been previously
assessed by Daniell (2011); there are more than 18 packages available (many being free and open-
source). The research in this field has started more extensively in the United States and Japan, with the
development of the HAZUS multi-hazard software (FEMA, 2014). Several local systems capable of
computing damage and casualties in near real time already exist in several cities of the world, such as
Yokohama, Tokyo, Istanbul, Taiwan, Bucharest and Naples (Erdik et al. 2011).
The science of seismic risk estimation is one that always has to evolve, which is essential, since on one
side we are still far from modelling complex processes that lead to real damage recorded few years
ago, and on the other side keep up with the new technological and territorial developments. Each
destructive earthquake contributed and will contribute to adjustments in loss estimation parameters.
Romania has one of the greatest seismic hazard levels in Europe, mainly due to earthquakes occurring
in the intermediate-depth Vrancea Source (at depths regularly between 70 and 180 km), located in the
Eastern Carpathian Belt, near the center of the country (Figure 1). Statistically, 2 or 3 earthquakes with
moment-magnitude (Mw) greater than 7 occur here per century. The last major event (on March 4,
1977, with Mw of 7.4, at 94 km depth) was the most destructive, resulting in the death of 1578 people
and the collapse of tens of thousands of buildings, not just in the epicentral area, but also hundreds of
km apart (in cities like Bucharest, Iasi, Craiova or Chisinau). It is clear that a major earthquake will
strike again, therefore we must use the opportunity of recent scientific knowledge as an essential tool
for mitigating the potential damage and creating effective preparedness solutions.

Figure 1. Map of Romania, showing earthquakes with Mw ≥ 4 and seismic sources according to the BIGSEES
Catalog (NIEP, 2017), results of the probabilistic seismic hazard analysis of the RO-RISK Project and seismic
stations with real-time transmission at the Romanian Seismic Network (RSN)

The expansion of the Romanian Seismic Network and the implementation of ShakeMap at the
National Institute for Earth Physics (NIEP) provide the means for quantifying the seismic ground
motion, although there will be a never-ending effort to improve the coverage of seismic stations and to
2
better understand local attenuation. The feedback that we’ve received from authorities and other actors
with responsibilities in emergency response shows that information provided by ShakeMaps is not
relevant enough. Therefore, a new system responding to their necessities (need of knowing potentially
affected areas, considering also exposure and vulnerability) was created. The system is called
SeisDaRo (The near real-time system for estimating the Seismic damage in Romania), and
automatically produces maps and reports regarding easy to grasp aspects such as estimated number of
collapsed buildings and fatalities, minutes after an earthquake with local magnitude ≥ 4. Since its
implementation in 2011, SeisDaRo evolved, and is since December 2017 at the 3rd version. In this
paper, we present its technical characteristics and test it for 5 representative Vrancea scenarios, with
the goal of providing insights on how an ELE system like this works in real time, which optimizations
can be performed and what types of useful representations can be generated. Nevertheless, scenario
results are of great interest, since they show the actual potential seismic risk of Romania.

2. SYSTEM CHARACTERISTICS

In order to reflect the SeisDaRo progress, we must briefly mention the characteristics of previous
versions:
- Version 1, implemented in 2011, used only SELENA (Seismic Loss Estimation using a logic tree
approach), a highly appreciated (Stafford et al. 2007) open-source software developed by
NORSAR (Molina et al. 2010). The input for SELENA came from the USGS ShakeMap as
implemented at NIEP (Erduran et al. 2012), and a building and population dataset from 1999 or
2002, for 7 Romanian counties (at city/commune level, or smallest Local Administrative Units –
LAU2 - in Romania), and from 2009, for 9 northern Bulgaria municipalities.
- In Version 2, implemented in 2013, the database from 1999 was further processed and the analysis
was extended for Bucharest and 19 Romanian counties (located in the most exposed area to
Vrancea earthquakes). The method for estimating fatalities was changed in SELENA from Simple
(use of only one general population to building typology profile) to HAZUS (individual profiles
for each LAU2), and a new representation and reporting code was developed.
As shown in Figure 2, SeisDaRo 3 has two independent modules based on different methodologies
that estimate seismic losses:
- PAGER-RO module, which is an adaptation of the USGS PAGER methodology (Jaiswal et al.
2009); this module expresses earthquake fatality rate (the ratio of the total number of shaking-
related fatalities to the total population exposed at a given shaking intensity) in terms of a two-
parameter lognormal cumulative distribution function. The module currently uses the PAGER
coefficients for Romania (θ=17.5, β=0.24 and stdev = 2.6), applied to the equation in Jaiswal et al.
(2009), together with intensity data and a refined population distribution dataset at the level of
2011, at LAU2 level. PAGER-RO generates rough fatality estimates at national level (the mean
result is not as relevant as the ranges of fatality and their probabilities are), however in a very short
amount of time (35 seconds), providing the first scale of the impact.
- SELENA module, in which the methodology relies on the Improved Displacement Coefficient
Analytical Method (IDCM, desribed in Molina et al. 2010) along with capacity and fragility
curves for determining probabilities of general building typologies collapse. Further on, for
describing the impact of building collapse in terms of human casualties, the HAZUS approach is
applied. The response spectrum is calculated using the IBC-2006 methodology. Building and
population data at LAU2 level, from the Romanian National Census in 2011, are used. Economic
losses are not estimated at the moment.

3
 Figure 2. Implementation scheme for SeisDaRo 3

The Custom ShakeMap module provides intensity and acceleration data for the PAGER-RO and
SELENA loss estimation modules. This component can mix real-recorded data from the seismic
stations of the Romanian Seismic Network with estimates obtained with Ground Motion Prediction
Equations (GMPE). Earthquake parameters produced automatically by the Romanian Early Warning
System (REWS) or by the Antelope software (and potentially SEISCOMP, through which only data
acquisition is run at the moment) are also highly important for intensity and acceleration prediction
equations.
As input for PAGER-RO, the Custom ShakeMap module calculates for the Vrancea intermediate-
depth source the mean between two specific Intensity Prediction Equations (IPE): Sorensen IPE
(Sorensen et al. 2007) and Sokolov IPE (Sokolov et al. 2008).
As input for SELENA, peak ground acceleration (PGA) and spectral acceleration (SA) values at
periods of 0.3 and 1 second are provided. As previously observed (Sokolov et al. 2008), the complex
subcrustal nature of the Vrancea source leads to difficult to predict ground motions, as major
earthquakes tend to show different patterns. The current GMPEs developed specifically for this source
have difficulties in reflecting local effects and have high error ranges for magnitudes greater than 7.
That is why we chose to add a Custom ShakeMap module, in which we incorporated a new capability
(Toma-Danila and Cioflan, 2017): when real recordings are available, for a specific azimuth (of 450 to
be representative for the Vrancea patterns) it is checked which GMPE fits best with recorded values
and is selected. At the moment, two GMPEs can be selected: Sokolov GMPE (Sokolov et al. 2008)
and Vacareanu et al. (2015). Overall, the advantage of this approach lies in the fact that it gives more
credit to real recorded values (nevertheless checking for their correctness), it can dynamically select
and use a GMPE for a specific azimuth, it is easier to be parameterized and takes only a few seconds
to run. Statistics will show in time which combination or weighting scheme is more relevant for major
earthquakes. A comparison between results obtained with the classical and custom procedure can be
found in subchapter 3.3.
To each of the above modules, a special code section was added to generate output data in GIS format
(shapefiles, geoJSON or kml), maps and graphs easy to understand. Disclaimers are an important
aspect to be incorporated in these high sensitivity products.

4
3. DATA

3.1 Data about residential buildings

The upgrades that we mentioned refer first of all to the statistical data that we used; the old dataset
version of SeisDaRo 2 was limited to 19 Romanian Counties and Bucharest (1400 analysis points in
total), and dated from 1999 or 2002. In the new version, we implemented data from the 2011 National
census, for all Romanian counties (41 and Bucharest), at administrative-unit level: city or commune
(Figure 3). In total now there are 3186 analysis points, which leads to 1 minute of extra-time for
SELENA computation. Data was obtained within the framework of the Ro-Risk national project. The
characteristics of the new dataset imposed a different selection of vulnerability curves for buildings,
compared to the arrangement for the previous data.

Figure 3. The distribution of residential buildings and inhabitants per city/commune, according to the 2011
database used in SeisDaRo 3

Table 1. SeisDaRo 3 building typologies

Construction Typology + Number of Typology + Construction Typology

material floors period
Reinforced RC Ground floor, Low ≤ 1945, 1946- PreCode
concrete or precast 1-2 floors 1960
concrete
Masonry, stone or M3_1 (with 3-5 floors Medium 1961-1977 LowCode
substitutes, and HighCode –
wooden floors M4)
Masonry, stone or M3_2 (with ≥ 6 floors High 1978-1992 ModerateCode
substitutes, and HighCode –
concrete floors M4)
Adobe M2 Not available Not 1993-2002, HighCode
information considered 2003-2006,
2006-2011
Wood W Not available Not considered
information
Mixt materials, Not
other materials considered
Not available Not
information considered

5
 Figure 4. Total number of residential buildings from Romania in each typology presented in Table 1

Newer data structure (Table 1) did not allow keeping three sub-typologies for reinforced concrete;
after multiple tests it was considered that vulnerability curves for RC2, as in the old typologies (Toma-
Danila and Armas, 2017), fits best for a generalized quantification of damage for the respective
construction material. Unfortunately, in the 2011 census, there were partially or completely missing
information for 9.79% of the total number of buildings in Romania (1,041,716 out of 10,638,078)
(Figure 4). 5.52% of the buildings had the construction material reference “mixt materials, other
materials” or “not available information”, and could not be considered for the loss estimation,
although having info about number of floors and construction period. The rest of 4.27% had even
fewer information. These percentages have to be considered when concluding upon the results, since
loss estimates for all buildings could have values 10% or more than produced at the moment, due to
the aforementioned issue. We estimate that at least 500,000 buildings built in the PreCode or
LowCode period are not considered in SeisDaRo 3, due to lack of sufficient data for considering them
reasonably.
Most of the capacity curves for typologies presented in Table 1 were associated with curves from the
Hazus-MH database (FEMA, 2014), after expert analysis of their representativeness and comparison
testing with other regional curves. Capacity and fragility curves for the M4 typology were selected
from the Reinforced Masonry Bearing Walls (RM2) Hazus-MH (FEMA 2014) typology, in order for
loss estimates to become more consistent with values for M3_1 and M3_2 typologies. Fragility curves
for RC are adapted to the works of the Technical University of Civil Engineering in Bucharest, for the
DACEA Project.

3.2 Population

Compared to the previous SeisDaRo dataset from 1999, which had associated for each building
typology a number of corresponding inhabitants, individually for each city/commune, for 2011 we
only had general information regarding population – total number of inhabitants per city/commune,
with no info about their distribution per building typology. In order to overcome this limitation, we
developed an empirical methodology similar to the ELER approach (KOERI, 2010) which enabled us
to make assumptions (with high degree of uncertainty) upon the degree of population distribution per
building typology, based on the 1999 population distribution profiles. First, we calculated the average
number of inhabitants per building typology, from the 1999 data. This number was multiplied with the
2011 number of buildings in each city/commune with correspondent in 1999, and a simulated number
of inhabitants was obtained. A ratio between this value and the actual number of inhabitants in 2011
was calculated, and applied to the values for each typology in order to reach the 2011 number of
inhabitants. This ratio was between 0.42 and 1.91, with an average of 0.92 and a standard deviation of
0.12, showing that the estimates fit better than expected with the data from 2011. When a direct
correspondent between 1999 and 2011 geounits could not be determined (half of Romania), average
profiles for two types – for cities or communes – were applied, leading to the ratios with an average of
1.23, but a considerable standard deviation of 1.09. The main reason for this are the regional
differences between areas for which we extracted inhabitant distribution profiles (mainly Moldova and
Muntenia) and areas for which we applied these profiles (Transilvania, Crisana, Oltenia); it is clear
that the number of inhabitants per building typology varies from a region to another, still at this point
we could not account for this aspect.

6
3.3 Earthquake scenarios

In order to test the validity of loss estimations it is extremely useful to test them, based on relevant
scenarios which can be linked to past damage or other considerations regarding the damage degree
they might produce. In Romania there was not a severe enough recent earthquake enabling us to
compare our loss estimation results with real damage, in the actual context of development, however
the 31 August 1986 (Mw 7.1) and 30 May 1990 (Mw 6.9) Vrancea earthquakes produced low damage
(each less than 10 deaths in Romania, some of them related to panic). Although they did not equal the
size of the 10 November 1940 (Mw 7.7) or 4 March 1977 events (Mw 7.4), they can be considered a
useful limit for understanding the potential starting point of damage due to Vrancea intermediate-
depth earthquakes. Important is also that they were recorded by 30 seismometers in Romania each, so
real recordings can be used. We relate in this study to the limit represented by these earthquakes (and
also to the event of 31 May 1990 with Mw 6.4), to identify potential over or under estimates, since
newer building stock can generally be considered more resistant. We also test the 1940 and 1977
scenarios, for which a direct comparison with past losses is not entirely relevant, still insightful. All
scenario parameters can be found in Table 2. At the moment, we focused on testing only Vrancea
subcrustal scenarios, since they have the most destructive potential in Romania and can be compared
with previous 20th century damage. A more extended analysis including test crustal scenarios for
sources such as Banat, Fagaras-Campulung or Shabla will soon be published. For crustal sources,
other representative IPEs and GMPEs need to be determined and incorporated in the Custom
ShakeMap.

Table 2. Earthquake scenarios used for testing the results of SeisDaRo 3

Scenario Earthquake Lat Lon Mw Depth Real Intensity data Acceleration

ID correspondent (km) number of used (for data used (for
(local time) fatalities PAGER-RO SELENA
in module) module)
Romania
1940 Vrancea 10 45.80 26.70 7.7 150 593 Interpolated Mean between
November 1940 observed Sokolov and
(03:39) intensities from Vacareanu
Kronrod et al. GMPE
(2013) and
mean between
Sokolov and
Sorensen IPE
1977 Vrancea 4 45.77 26.76 7.4 94 1578 As for 1940 As for 1940
March 1977
(21:22)
1986 Vrancea 31 45.52 26.49 7.1 131 8 As for 1940 Custom
August 1986 ShakeMap,
(00:28) using 30 real
recordings
1990(1) Vrancea 30 May 45.83 26.89 6.9 91 9 As for 1940 Custom
1990 (12:40) ShakeMap,
using 30 real
recordings
1990(2) Vrancea 31 May 45.84 26.91 6.4 87 0 As for 1940 Custom
1990 (02:17) ShakeMap,
using 22 real
recordings

After running the Custom ShakeMap module or using other types of data specified in Table 2,
intensity and acceleration maps as in Figures 2 and 3 were produced and used in the PAGER-RO and
SELENA modules. As Figures 5 and 6 show, the influence of putting more emphasis on real-recorded
data and the capability of reducing uncertainties to azimuthal selection of GMPEs leads to a different

7
representation of ground motion, which we believe to be closer to reality, when also comparing with
patterns reflected by macroseismic intensity maps.

Figure 5. Results of classical ShakeMap methodology (left) and Custom ShakeMap module (right) for
the 1986 scenario

Figure 6. Results of classical ShakeMap methodology (left) and Custom ShakeMap module (right) for
the 1990(1) scenario

Figure 7. Intensity maps for the 1940 earthquake: interpolated observed intensities from Kronrod et al. 2013
(left) and mean of two IPEs, used within the SeisDaRo PAGER-RO module (right)

4. RESULTS

Since SeisDaRo 3 estimates fatalities with two different methodologies we chose to present results as
a mix. SELENA methodology can be considered more complex since it uses structural considerations
and connects more detailed vulnerability parameters with instrumental ground motion parameters.
However, more detailed is not always the best, since uncertainties will pile up along the way; an
artifact due to data quality or method inadequacy at an early stage can drastically modify final results.
It is needed to be aware of this aspect, especially in risk estimation. Our future work will focus on a

8
better understanding of error ranges for the SELENA module. In this paper we only present median
results, mentioning that in the way SELENA computes 16% and 84% fractiles of damage, we do not
notice major result differences.

Figure 8. Fatality estimates obtained with PAGER-RO and SELENA modules, for earthquake scenarios in Table
2. PAGER-RO module used mean intensities between Sokolov and Sorensen IPE, as it would in real-time

Figure 9. Fatality estimates obtained with PAGER-RO and SELENA modules, for earthquake scenarios in Table
2. PAGER-RO module used interpolated observed intensities from Kronrod et al. (2013)

Figures 8 and 9 are meant also to show the differences between using observed intensity (reevaluated
by Kronrod et al. 2013) and IPEs, for PAGER-RO. As results for the 1940 and 1977 scenarios show
(and also a direct comparison between intensity maps), IPEs developed for Vrancea underestimate
intensity values at higher magnitudes, these aspect propagating also in PAGER-RO estimates. In the
case of the 1940 scenario, intensity values of IX and even X are reported in Kronrod et al. (2013) in
LAU2 located SW and E from the epicenter, but these are not reproduced by the IPEs (Figure 7). In
the case of other scenarios, observed and estimated intensities are much more related. Still, as
expected, PAGER-RO shows clear and valid distinctions between fatality ranges of Vrancea
earthquakes with Mw>7.1, earthquakes between 6.9 and 7.1 Mw and smaller earthquakes. For all
scenarios, it is notable that SeisDaRo 3 provides lower estimates than SeisDaRo 2, with the SELENA
module. SeisDaRo 3’s SELENA results are in general not within the most probable fatality range
provided by PAGER-RO, but they mostly fit in the second most probable range. SELENA module
seems to overestimate fatalities, especially for the 1986, 1990(1) and 1990(2) scenarios, when the
building stock was more representative to actual times and for which there were no considerable
human losses. The 1986 earthquake happened soon after midnight, so most people were inside, and
still no fatal losses occurred, because most buildings withstand the shock. Since also for the 1990(2)
scenario, with a much smaller magnitude (Mw 6.4), the SELENA fatality estimation is high, we will
reconsider how the methodology is applied, by firstly trying to see the implications of selected

9
building vulnerability curves and then by modifying collapse and casualty rates.

Figure 9. SeisDaRo 2 loss estimations for the 1940 (left) and 1977 (right) scenarios, in terms of percentage of
completely damaged buildings from total per city/commune

Figure 10. SeisDaRo 3 loss estimations for the 1940 (left) and 1977 (right) scenarios, in terms of percentage of
completely damaged buildings from total per city/commune

Figures 9 and 10 show comparable results for the 1940 and 1977 scenarios, between SeisDaRo 2 and
3. SeisDaRo 3, which uses a building dataset from 2011 with a slightly different structure (missing
however about 10% of the total buildings), estimates a lower probability of damage. This might reflect
the influence of newer building stock (SeisDaRo 2 uses data from 1999 or 2002). In the simulations,
the same analytical method (IDCM) and ground motions were used, so the difference comes from
building numbers and their classification.
Figures 11 and 12 reveal the distribution of fatalities, for the 1940, 1977 and 1986 scenarios. As
expected, cities with high number of inhabitants relatively close to the epicenter are evidenced:
Bucharest (which is analyzed at the level of its six LAU2 sectors), Ploiesti, Buzau, Braila, Galati or
Bacau. Interesting is that Focsani is not highlighted as hotspot for the 1940 scenario, but it is for the
1977 scenario (which is consistent with observed damage distribution). Iasi, another major city which
experienced significant losses especially during the 1940 earthquake, does not appear evidenced,
which might be a sign that major building improvements since 1940 occurred. The 1986 maps show
probable overestimations for Bucharest and Focsani, but otherwise values indicate potential but not
certain fatalities, which seems credible. When summed up, these values give an overall high estimate
(2621), however when summing up only more certain ranges (> 5 fatalities per city/commune), the
sum becomes 1159, which is still high, but more relevant to consider as estimate. The 1990(1) results
show the same characteristics.

10
Figure 11. SeisDaRo 3 loss estimations for the 1940 (left) and 1977 (right) scenarios, in terms of fatalities for the
worst-case scenario (all people inside)

Figure 12. SeisDaRo 3 loss estimations for the 1986 scenario, in terms of percentage of completely damaged
buildings from total per city/commune (left) and fatalities for the worst-case scenario (all people inside) (right)

5. CONCLUSIONS

In this paper we presented the characteristics and results of the new version of SeisDaRo, a system
dedicated to the estimation of seismic damage in Romania in near real-time, in terms of building
damage and fatalities. The new version now incorporates a custom ShakeMap like module dedicated
to the evaluation of intensity and acceleration, using specific IPEs and GMPEs for the Vrancea
intermediate-depth source and a new methodology which allows a more appropriate representation of
ground motion when real data from seismic stations is available. To SeisDaRo 3 was also added a fast
fatality estimation module (PAGER-RO). Preliminary results of this module show that, although
estimates and probabilities for meeting certain fatality ranges are rough, it can give a credible
indication regarding the severity of an earthquake which happened only seconds before. By using
color-coded representations for showing uncertainty ranges, they can tell much more than magnitude
does (especially for subcrustal earthquakes, for which also depth is important).
In SeisDaRo 3, also the SELENA module was enhanced: new data from the National Census in 2011
was incorporated, replacing previous data from 1999 and 2002, available only for half of Romania.
The new data structure had some limitations (10% of building information couldn’t be used and also
population was not distributed per building typology), but these were overcome by applying statistical
analysis. Results for test scenarios show that new estimates are much comparable with what SeisDaRo
2 produced, but still they are a bit above credible fatality losses for 1977, 1986, 1990(1) and 1990(2)
scenarios; future investigations will give a complex diagnostic to the causes of overestimation.
For the next versions of SeisDaRo we shall consider:
- adding new vulnerability curves developed for representative building typologies in Romania;
- more complex error calculations for SELENA module results;

11
- performing real-time estimation for Bucharest, for the 128 distinct areas analyzed in Toma-Danila
and Armas (2017);
- implementing new IPEs and GMPEs for the Custom ShakeMap module;
- computing economic loss estimates;
- developing new coefficients for the PAGER-RO module, to constrain better fatalities in ranges.
SeisDaRo 3, implemented at NIEP, is at the moment one of Europe’s few functional seismic loss
estimation systems running in real-time. It contributes effectively to both seismic risk mitigation plans
(through scenario runs) and quick emergency response (in less than 6 minutes all modules run),
delivering high quality representations of the losses (as presented in the paper).

6. ACKNOWLEDGMENTS

This work was supported through the NUCLEU Programme financed by ANCSI (Romania) and by
the Seismology and Earthquake Engineering Research Infrastructure Alliance for Europe (SERA)
Project. Some of the data was provided by the Evaluation of disaster risks at national level (Ro-Risk)
Project and the Danube Cross-border system for Earthquakes Alert (DACEA) Project.

7. REFERENCES

Coburn A, Spence E (2002). Earthquake protection. Wiley, Hoboken.

Daniell JE (2011) Open source procedure for assessment of loss using global earthquake modelling software
(OPAL). Nat Hazards Earth Syst Sci. 11:1885–1900.
Erdik M, Sesetyan K, Demircioglu MB, Hancılar U, Zulfikar C (2011). Rapid earthquake loss assessment after
damaging earthquakes. Soil Dyn Earthq Eng. 31(2):247–266.
Erduran E, Lang DH, Lindholm CD, Toma-Danila D, Balan SF, Ionescu C, Aldea A, Vacareanu R, Neagu C
(2012) Real time earthquake damage assessment in Romanian-Bulgarian border. In: Proceedings of the 15th
World Conference on Earthquake Engineering, Lisbon, Portugal.
FEMA - Federal Emergency Management Agency (2014) HAZUS-MH: multi-hazard loss estimation
methodology. http://www.fema.gov/hazus.
Jaiswal K., Wald D.J., Hearne M. (2009). Estimating Casualties for Large Earthquakes Worldwide Using an
Empirical Approach. U.S. Geological Survey Open-File Report OF 2009–1136, 78 p.
KOERI (2010). Earthquake loss estimation routine (ELER 3), technical manual and users guide.
http://www.koeri.boun.edu.tr/ Haberler/NERIES%20ELER%20V3.1_6_176.depmuh
Kronrod T., Radulian M., Panza G., Popa M. et al. (2013). Integrated transnational macroseismic data set for the
strongest earthquakes of Vrancea (Romania). Tectonophysics. 590:1-23.
Molina S, Lang DH, Lindholm CD (2010) SELENA - An open-source tool for seismic risk and loss assessment
using a logic tree computation procedure. Comput Geosci. 36(3):257–269.
NIEP—National Institute for Earth Physics (2017). Bigsees earthquake catalog.
http://bigsees.infp.ro/Results.html
Sokolov V, Bonjer KP, Wenzel F, Grecu B, Radulian M (2008) Ground-motion prediction equations for the
intermediate depth Vrancea (Romania) earthquakes. B Earthq Eng. 6(3):367–388.
Sorensen M.B., Stromeyer D., Grunthal G. (2007). Deliverable 4.2: Development of magnitude calibrated
intensity attenuation relations for target areas considering area-typical faulting characteristics and microzonation
effects. SAFER (Seismic eArly warning For EuRope) Project.
Strasser FO, Bommer JJ, Sesetyan K, Erdik M, Cagnan Z, Irizarry J, Goula X, Lucantoni A, Sabetta F, Bal IE,
Crowley H, Lindholm C (2008) A comparative study of European earthquake loss estimation tools for a scenario
in Istanbul. J Earthq Eng. 12(2):246–256.
Toma-Danila D., Armas I. (2017). Insights into the possible seismic damage of residential buildings in
Bucharest, Romania, at neighborhood resolution. B Earthq Eng. 15(3):1161-1184.
Vacareanu R., Radulian M., Iancovici M., Pavel F., Neagu C. (2015). Fore-arc and back-arc ground motion
prediction model for Vrancea intermediate depth seismic source. J Earthq Eng. 19(3):535-562.

12