2 SELENA – An open-source tool for seismic risk and loss 3 assessment using a logic tree computation procedure 4
1,32,32,35 S. Molina , D.H. Lang , and C.D. Lindholm
(1)7 Dpto. de Ciencias del Mar y Biología Aplicada (Área de Física de la Tierra), Facultad de Ciencias, Fase II, 8 Universidad de Alicante, Campus de S. Vicente del Raspeig s/n, 03690 Alicante, Spain
(2)9 NORSAR, P.O. Box 53, 2027 Kjeller, Norway
(3)10 International Centre for Geohazards (ICG), 0806 Oslo, Norway
The era of earthquake risk and loss estimation basically began with the seminal paper on 15
16 hazard by Allin Cornell in the year 1968 (Cornell, 1968). Following the 1971 San Fernando 17 earthquake, the first studies placed strong emphasis on the prediction of human losses 18 (number of casualties and injured) and to foresee the needs in terms of health care and 19 shelters in the immediate aftermath of a strong event. In contrast to the early risk modeling 20 efforts, the later studies have focused on the disruption of the serviceability of roads, 21 telecommunications and other important lifeline systems. In the 1990‟s the National
?22 Institute of Building Sciences (NIBS) developed a tool (HAZUS99) for the Federal
23 Emergency Management Agency (FEMA), where the goal was to incorporate the best 24 quantitative methodology in the earthquake loss estimates.
25 Herein, the current version of the open-source risk and loss estimation software SELENA 26 v4.1 is presented. While using the spectral displacement-based approach (capacity spectrum 27 method), this fully self-contained tool analytically computes the degree of damage on specific 28 building typologies as well as the connected economic losses and number of casualties. The 29 earthquake ground shaking estimates for SELENA v4.1 can be calculated or provided on
1 three different ways: deterministic, probabilistic or based on near-real-time data. The main 2 distinguishing feature of SELENA compared to other risk estimation software tools is that it 3 is implemented in a „Logic Tree‟ computation scheme which accounts for uncertainties of
4 any input (e.g., scenario earthquake parameters, ground-motion prediction equations, soil 5 models) or inventory data (e.g., building typology, capacity curves and fragility functions). 6 The data used in the analysis is assigned with a decimal weighting factor defining the weight 7 of the respective branch of the Logic Tree. The weighting of the input parameters accounts 8 for the epistemic and aleatoric uncertainties that will always follow the necessary 9 parameterization of the different types of input data.
10 Like previous SELENA versions, SELENA v4.1 is coded in MATLAB which allows an easy 11 dissemination among the scientific-technical community. Furthermore, any user has access 12 to the source code in order to adapt, improve or refine the tool according to his particular 13 needs. The handling of SELENA‟s current version and the provision of input data is 14 customized to an academic environment which might support decision-makers of local, state 15 and regional governmental agencies in estimating possible losses from future earthquakes. 16
17 1. Introduction
19 Seismic risk estimation is based on the need to quantify the expectations of ground shaking 20 and the corresponding performance of structures. Based on such investigations and efforts 21 construction techniques have improved over time, and appropriate countermeasures can be 22 taken. The scientific field of seismic risk and loss assessment is a growing research area 23 which traditionally has been either based on macroseismic intensity or peak ground 24 acceleration (PGA). In recent years different risk assessment methodologies have been 25 developed which are incorporated in a considerable number of different software (Crowley 26 et al., 2004; McGuire, 2004; Oliveira et al., 2006).
27 Unfortunately, it is a large damaging earthquake of all which is able to verify or refute the 28 estimated seismic scenario, the chosen methodology and the defined assumptions. But still 29 this creates a fruitful situation when we are able to calibrate our models and input
1 parameters on this experience and results of different risk estimation methods can be 2 compared with each other. In general nearly all available risk and loss assessment studies 3 defined individual scenario earthquakes as a main basis for planning (ATC, 1996a; 4 Algermissen et al., 1988; CDMG, 1982a, 1982b, 1987, 1988, 1990, 1995; CUSEPP, 1985; 5 Dames and Moore, 1996; EQE, 1997; Harlan and Lindbergh, 1988; NOAA, 1973). All 6 these studies use already existing knowledge of regional geology and seismic activity to 7 generate maps with estimated intensities I or ground motion accelerations a. This, in
8 combination with other types of input data (e.g., building stock, population density), is used 9 to calculate the extent of damages to structures and life-lines as well as the impacts on 10 population. Some of these studies additionally address potential secondary hazards such as 11 fire, flood, and hazardous materials release. Earthquake scenarios of this type have been 12 employed by governmental institutions and public utilities to prepare for and to mitigate 13 the degree of damage from future events. Thus it appears that the typical loss study has 14 been focused on a single event, applied in the long-term pre-event period, and utilized 15 primarily by those concerned with seismic safety planning and disaster management. 16 In this respect, the Federal Emergency Management Agency (FEMA 366, 2001, 2008) 17 initiated a study on seismic risk estimation for all regions of the United States using the
??18 national loss estimation tool HAZUS99 and HAZUSMH, respectively. The study‟s main
19 task was to analyze and compare the seismic risk across regions in the U.S. which have 20 different hazard levels, characterized by different population density or physical building 21 vulnerability.
22 The advent of high-speed computing, satellite telemetry and Geographic Information 23 Systems (GIS) made it possible to electronically generate loss estimates for multiple 24 earthquake scenarios, to provide a nearly unlimited mapping capability, and perhaps most 25 important, to develop estimates for a current earthquake event in near real-time given its 26 source parameters, i.e. magnitude and location, are provided.
1 Currently, a number of different computer tools able to estimate the seismic risk using 2 different methodologies are available. Table 1 lists some of them and briefly describes their 3 main principles and outputs.
4 A very powerful approach being attractive from a scientific-technical perspective is the 5 HAZUS software (HAZUS?97, HAZUS?99, HAZUS?99-SR1, HAZUS?99-SR2,
6 HAZUS?MH, HAZUS?MH MR1, HAZUS?MH MR2, HAZUS?MH MR3) which was
7 developed by the National Institute of Building Sciences (NIBS) for the Federal 8 Emergency Management Agency (FEMA, 1997, 1999, 2001, 2002, 2004, 2005a, 2006, 9 2007). The HAZUS tool is built upon the integrated geographic information system 10 platform ArcGIS (ESRI, 2004) and can be considered as a software extension to ArcGIS. 11 HAZUS is directly integrated with the national and regional databases on building stock 12 and demography data of the United States (FEMA 366, 2008). This enables any larger 13 community in the United States to simulate earthquake risk scenarios with a minimum 14 effort since most of the necessary data is already prepared. The basic methodology behind 15 HAZUS represented the starting point for the development of alternative tools (see Table 16 1) in order to compute seismic risk and loss estimates as well as initiated numerous 17 application studies (i.e. Kircher, 2003; Wang et al., 2005; Kircher et al., 2006; Nielson and 18 DesRoches, 2007; Rojas et al., 2007). The fact that HAZUS is tailored so intimately to U.S. 19 situations which makes it very difficult to be applied to other environments or geographical 20 regions, has recently activated the development of GIS-based methodologies which 21 facilitate the application of HAZUS to other parts of the world (Hansen and Bausch, 2006). 22 Aware of the importance of a proper seismic risk estimation, the International Centre for 23 Geohazards ICG, through NORSAR (Norway) and the University of Alicante (Spain), has 24 developed a software tool running under MATLAB (The MathWorks, Inc.) in order to 25 compute seismic risk of urban areas using the capacity-spectrum method. The tool named 26 SELENA (SEimic Loss EstimatioN using a logic tree Approach) is open for any user-
1 defined data and thus can be applied to any part of the world. The user has to supply a 2 number of input files which contain the necessary input data (e.g., building inventory data, 3 demographical data, definition of seismic scenario etc.) in a simple ASCII format. A more 4 detailed description of the type of input data will be given below.
5 It should be explicitly stated that the core of the HAZUS methodology (FEMA, 1999, 2003) 6 was adopted for SELENA. However, one of the main differences between both tools is 7 that SELENA works independent of any Geographic Information System, while HAZUS
8 is connected to the ArcGISsoftware (ESRI, Inc.). In addition, a Logic Tree-computation 9 scheme has been implemented in SELENA which allows the user to define weighted input 10 parameters and thus being able to properly account for epistemic uncertainties. 11 Consequently, any type of output is provided with corresponding confidence levels. 12
13 2. Technical components of the SELENA-tool
15 2.1 Basic procedure
16 The current version of SELENA (Molina et al., 2009) allows for three analysis types which 17 are differing in the way the seismic impact is described. In general, spectral ordinates of 18 seismic ground motion at different reference periods have to be provided for each 19 geographical unit (i.e. census tract), in order to allow the construction of a design spectra 20 following a selectable seismic code provision, i.e. spectral ordinates (PGA, S) at reference a
21 periods T = 0.01, 0.3 and 1.0 s for IBC-2006 (International Code Council, 2006), and T =
22 0.01 s (PGA) for Eurocode 8 (CEN, 2002) as well as for Indian seismic building code IS 23 1893 (Part 1) : 2002 (BIS, 2002).
24 Once the seismic ground motion in each geographical unit is defined, the computation of 25 physical damage to the building stock, the total economic loss related to these damages, 26 and the number of casualties, i.e. the number of injured people and fatalities, is conducted.
1 Damage results are given in terms of cumulative probabilities P of being in, or exceeding
2 one particular damage state ds following the classification scheme given by HAZUS
3 (FEMA, 1999) into none, slight, moderate, extensive and complete damage. Absolute numbers of
24 damaged buildings or the damaged building (floor) area in [m] are calculated by combining
5 damage probabilities for each building type with the inventory data of the building stock. 6 Figure 1 exemplarily illustrates the sequence of SELENA for a deterministic analysis and 7 shows which input layers are required for the different program outputs. 8
9 Since the described methodology of risk assessment is based on statistical data, the level of 10 resolution is of utmost importance. Since a resolution of the damage outputs on the level 11 of individual buildings would require huge computation efforts, SELENA as most other 12 risk estimation software tools considers the geographical unit (GEOUNIT) as the smallest 13 area unit. In practice, this unit is related to building blocks or city districts. The decision on 14 the extent of each geographical unit has to be made considering different aspects such as 15 having equal soil conditions, constant surface topography or a homogeneous level of 16 building quality within the demarcated area.
19 2.2 Specification of the seismic input (demand)
21 A key point in any seismic risk assessment is the provision of seismic ground motion. In 22 SELENA v4.1 this can be done by:
23 - the provision of spectral ordinates (e.g., PGA, S at 0.3 and 1.0 seconds) taken out from a
24 probabilistic shaking maps) and assigned to the geographical units (probabilistic analysis),
1 - the definition of deterministic scenario earthquakes (e.g. historical or user-defined events) 2 and adequate/suitable ground-motion prediction equations in order to compute the 3 spectral ordinates in each geographical unit (deterministic analysis),
4 - provision of spectral amplitudes of recorded ground motion at the locations of seismic 5 (strong-motion) stations (analysis with near-real-time data).
6 The first results of each analysis type consist in the provision of seismic ground motion 7 amplitudes at the center of each geographical unit. Thereby it has to be considered that 8 these amplitudes can either represent the motion on rock conditions (deterministic and 9 probabilistic analysis) or already include amplification effects of local (near-surface) subsoil 10 conditions (analysis with near-real-time data). Based on these acceleration values SELENA
generates an elastic response spectrum (damping factor 11 ， = 5 %) following a selectable
12 seismic code provision. Currently, the provisions of the U.S. seismic code IBC-2006 13 (International Code Council, 2006), Eurocode 8 – Type 1 and Type 2 (CEN, 2002) and
14 Indian seismic building code IS 1893 (Part 1) : 2002 (BIS, 2002) with their soil 15 amplification factors are incorporated. Figure 2 schematically depicts the shape of a design 16 response spectra in the T-S-domain. a
18 2.3 Structural performance under seismic action
20 The core methodology of SELENA was adopted from the HAZUS-MH software whose 21 basic approach in order to identify the level of structural damage under a given seismic 22 impact is the capacity-spectrum method (FEMA, 1996). During the last years, several 23 modifications and extensions of this powerful and efficient technique have been developed. 24 Two of these procedures are incorporated in SELENA v4.1:
25 - Procedure 1: The conventional Capacity Spectrum Method as proposed in ATC-40 (Applied
26 Technology Council ATC, 1996b)
1 - Procedure 2: The Modified Capacity Spectrum Method (MADRS) as proposed in FEMA 440
2 (FEMA, 2005b)
4 To apply any of the available capacity spectrum methods, both seismic demand and the 5 capacity curve have to be transformed into the spectral acceleration-spectral displacement 6 (S-S) domain (Figure 3). Thereby, seismic demand is represented by the elastic response ad
7 spectrum while the capacity curve reflects the building‟s lateral displacement ！ as a function
8 of a horizontal force V applied to the structure. Beside a number of factors, building 9 capacity curves mainly depend on the building type (working materials and construction), 10 number of stories (height), and also from its region reflecting local building regulations as 11 well as local construction practice and quality.
12 The main task of the capacity-spectrum method is to find that point on the capacity curve 13 consistent with the seismic demand being reduced for nonlinear effects. Since each point 14 on the capacity curve represents a certain state of structural damage and thus reflects an 15 increase in structural damping as the damages accumulate, the performance point can only 16 be found iteratively. As Figure 3 illustrates, the performance point finally is characterized 17 by a spectral acceleration S and spectral displacement S (and establishing the basis for ad
18 assigning discrete damage probabilities P.)
19 Once the performance point and its corresponding spectral displacement Sare found, d
20 structural vulnerability (fragility) functions for each damage state ds are required to assign
21 damage probabilities P (Figure 4). These represent cumulative probabilities of a certain 22 building type of being in or exceeding one of the different damage states ds dependent on
23 spectral displacement S. Figure 4 shows a set of fragility functions for a random model d
24 building type as described by HAZUS (FEMA, 1999). The differences between the 25 intersection points of two neighboring fragility functions for a given spectral displacement 26 are discrete damage probabilities (Figure 5) which establish the basis for the calculation of
1 absolute damage values. Generally, fragility functions are provided in dependence on 2 building typology and level of seismic code design reflecting the general quality of 3 construction practice. The fragility functions given by HAZUS (FEMA, 1999) are 4 described by the following form:
~)，?S1d：?5 (1) ;；((PdsS？？ln??d：?，Sdsd,ds??；(?，
6 in which:
8 - median value of spectral displacement at which the building reaches the Sd,ds
9 threshold of damage state ds,
10 ， - standard deviation of the natural logarithm of spectral displacement for damage ds
11 state ds,
12 ； - standard normal cumulative distribution function.
14 2.4 Risk and loss assessment based on statistical inventory data
16 Seismic risk results are essentially represented by the total extent of physical damage an 17 earthquake scenario likely to occur may produce to the building stock. Within SELENA 18 v4.1 the extent of structural damage can either be quantified in the number of buildings or 19 building floor area affected by a certain damage state ds.
20 Based on the damage results, both economic losses and number of casualties are calculated. 21 The first requires a suitable economic model, which provides realistic costs for the repair or 22 replacement of damaged or collapsed buildings and which then allows the appraisal of the 23 total amount of loss in each geographical unit.
1 The computation of economic losses caused by direct structural damage and which are 2 required for building repair and in case of complete damage or collapse for replacement is 3 done in the following way adopting the methodology described by FEMA (2003):
4 (2) Loss;CI(A(P(C,,,,；；；ijjkijk;;;111ijk
5 in which:
6 CI - regional cost multiplier accounting for the geographic cost variations 7 between the different geographical units,
28 A - built area of occupancy type (oct) i and model building type (mbt) j (in [m]) i, j
9 P - damage probability for model building type j to experience structural j, k
10 damage of damage state ds k (slight, moderate, extensive or complete),
11 C - cost of repair or replacement for each occupancy type i and model building i, j, k
12 type j suffering structural damage state k in input currency per floor area,
2e.g. [$/m13 ].
15 To determine the estimated number of casualties (i.e. injured people and fatalities) which 16 are mainly caused by the total or partial collapse of buildings, reliable statistical data on the 17 study area‟s demography is required. This does not only consist of population statistics, like 18 e.g. total number of inhabitants per district, but also in average numbers of people staying 19 in buildings of different type or occupancy and percentages of people staying outdoors or 20 indoors at different times of the day.
21 SELENA‟s current version 4.1 facilitates the computation of casualty numbers using two
22 different methodologies. During the preparation of input files, the user can explicitly 23 choose whether the HAZUS approach (FEMA, 2003) or the basic approach following 24 Coburn and Spence (2002) will be applied.