Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

Porto, Portugal, 30 June - 2 July 2014

A. Cunha, E. Caetano, P. Ribeiro, G. Müller (eds.)
ISSN: 2311-9020; ISBN: 978-972-752-165-4

Earthquake risk analysis of structures

Johnn P. Judd1, Finley A. Charney2
1
Graduate Research Assistant, Department of Civil and Environmental, Virginia Tech, Blacksburg, Virginia, USA 24061
2
Professor, Department of Civil and Environmental, Virginia Tech, Blacksburg, Virginia, USA 24061

ABSTRACT: This paper first discusses the effect of the methods used to account for ground motion direction and soil
amplification on the prediction of structural vulnerability of buildings, and the vulnerability of various structural systems based
on studies employing the FEMA P-695 methodology. The paper also examines the effect of the method used to calculate
collapse risk. It is recommended that the calculation of risk uses the derivative of the fragility curve. Using the derivative of the
seismic hazard curve to calculate risk is more computationally expensive, may be unstable, and, depending on the method used,
may lead to inaccurate estimates of risk.

KEY WORDS: Risk analysis; Collapse assessment; Earthquake engineering.

1 INTRODUCTION probability in 50 years [4, 7]. Another example is the ASCE

Earthquake risk analysis may be thought of as encompassing 43-05 approach for designing nuclear power plants [8].
four basic elements: hazard, exposure, vulnerability, and The difference between uniform-hazard and uniform-risk
consequence [1]. Hazard refers to the likelihood of a natural maps depends on the convolution of two factors: the shape of
phenomenon that can produce damaging impacts [2], but it the ground motion versus annual frequency of exceedence
may also be expanded to include the potential for blasts and curve, termed the “hazard curve”, and the structural sensitivity
technological threats (terrorism). In earthquake engineering, (vulnerability of collapse) to ground motion, termed the
hazard is the potential for severe ground motions. “fragility curve.” For new construction it is implied that the
Exposure refers to people and assets subject to the design map leads to structures with fragility curves where the
damaging impacts from a hazard [2]. In structural probability of collapse is consistent at a given hazard level.
engineering, the focus is on the people and infrastructure that Thus, the difference between uniform-hazard and uniform risk
are exposed. The degree of exposure involves the type, design maps is primarily a reflection of the underlying
quantity, and location of assets and the distribution (in terms differences in seismic hazard.
of both space and time) of occupants. Seismic hazard varies within many regions of the world. In
Vulnerability refers to the sensitivity of buildings, bridges, the US, for instance, differences between hazard curves in the
and other structures to hazard [2]. Structural vulnerability is a western US (tectonic plate boundary area) and hazard curves
function of the structural design and construction methods. in the central and eastern US (intra-plate area) are well
Vulnerability is described in terms of one or more documented [4, 9, 10]. Similar (but less dramatic) differences
consequences. Consequences refer to the impact of a hazard, exist in Italy (tectonic plate boundary area) [11]. Of course,
in terms of loss of life, downtime, and monetary damages. within some regions—especially smaller regions—the seismic
The most negative consequences arise from structural hazard curves do not vary greatly. In France, for example, the
collapse. Monetary loss can be direct (building damage and shapes of the seismic hazard curves are similar [12], and thus
repair costs) as well as indirect (downtime and losses due to the distinction between a uniform-hazard and uniform-risk
unsafe placarding). Ultimately, it is knowledge of the likely map is less pronounced.
consequences that assists in making design decisions. 1.2 Objectives
1.1 Risk-targeted ground motions This paper examines the hazard and vulnerability elements of
Ground motion maps for seismic design have traditionally risk and the method used to quantify risk. The effect of
been defined to provide a uniform hazard. Typically the directionality and site response on uniform-hazard ground
ground motions have a 2% to 10% probability of being motion hazard is first presented. Second, the collapse
exceeded in 50 years, depending on occupancy and use of the vulnerability of various modern structural systems is
structure, and the document and region involved [3, 4, 5]. summarized. Third, three common methods used to quantify
However, it is the probability of structural collapse that is of risk are compared. Two examples are included to illustrate
paramount concern [4, 6], and, as a result, the goal of a the application of earthquake risk analysis (a risk assessment
uniform risk of collapse has replaced uniform hazard in recent of partial restraint composite connection moment frame
maps. The chief example of this is the current ASCE 7-10 buildings and a risk map of a non-ductile moment frame
ground motion maps for the United States (US), which are building). Although the US is emphasized in this paper, the
intended to provide a uniform risk of collapse of 1% concepts presented are applicable in Europe and elsewhere.

2929
Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

2 UNIFORM-HAZARD GROUND MOTIONS 1.6 X-Component

The US Geological Survey (USGS) provides ground motion Y-Component
1.4
values in the US and corresponding maps for various Geomean

Pseudo-acceleration (g)
probability levels. These values are based on the ground 1.2 SRSS
shaking from potential earthquakes identified in workshops. Maximum Direction
The recommendations from workshops are then peer- 1
reviewed by several science organizations and two expert 0.8
panels. The USGS updates the ground motion values
periodically, with the latest version—the 2008 United States 0.6
National Seismic Hazard Maps—adopted by ASCE 7-10 and 0.4
the 2009 NEHRP provisions (next version of ground motion
values is planned for 2014). The maps provide 0.2-second 0.2
and 1.0-second spectral accelerations, respectively, that have a
0
2% probability of being exceeded in 50 years—corresponding 0.5 1 1.5 2 2.5 3 3.5 4
to the so-called Maximum Considered Earthquake (MCE) Period (s)
ground motion. Spectral accelerations for other periods and
Figure 1. Spectral acceleration using different definitions.
recurrence intervals are provided (but not mapped).
In Europe, the Swiss Seismological Service coordinates a compared to two-dimensional analyses, all other factors being
collaborative project—called Seismic Hazard Harmonization equal. To remove this unconservative bias in collapse
in Europe (SHARE)—that provides ground motion values capacity, the ASCE 7-10 ground motion map adjusts
similar to those provided by the USGS. The SHARE project (increases) the USGS geomean uniform-hazard spectral
involves a core team of over 50 scientists from institutions accelerations to approximate maximum-direction demand
across Europe, North Africa and Turkey, along with [19]. The ASCE 7-10 (NEHRP) approximation is based on
additional experts participating in workshops [13]. The the ratio of maximum-direction to geomean spectral demand
European map provides peak ground accelerations that have a for near-field ground motions in the western US [16].
2% probability of being exceeded in 50 years. Ground motion Table 1 gives the ratio of maximum-direction to geomean
values for other probability levels can be accessed through the spectral demand for sets of ground motions used in FEMA P-
European Facility for Earthquake Hazard and Risk (EFEHR) 695 and in the development of ASCE 7-10.
[14]. The ratios reflect the fact that the maximum-direction
spectral acceleration depends on the ground motions. The
2.1 Directional Effects
ratios for far-field ground motion sets and for ground motion
Ground motion is a three-dimensional phenomenon that can sets intended for the central and eastern US can be quite
be described in terms of three components: one component in different from each other. For far-field ground motions in the
the vertical direction and two orthogonal horizontal western US, the median ratio is 1.20 and 1.27, for 0.2-second
components. Relative to a structure, the direction of ground and 1.0-second spectral accelerations, respectively. In the
motion and spectral acceleration may be based on the central and eastern US, these ratios are 1.28 and 1.35.
geometric mean (“geomean”) of two horizontal components. Figure 2 compares the ratio of maximum direction to
The USGS ground motions are based on the geomean because geomean spectral acceleration for the FEMA P-695 Far-Field
that description is embedded in the ground motion attenuation ground motion set with the NEHRP approximation. Note that
models [4, 15]. Alternatively, the direction may be based on the ratios using the FEMA P-695 method will differ for these
the square root of the sum of the squares (SRSS) of the two periods compared to the NEHRP approximation.
horizontal components, or the “maximum-direction” of the Based on data from Huang et al. [17] for the US, maximum-
two components. Maximum-direction spectral acceleration is direction to geomean spectral acceleration ratios of 1.2 and
determined by calculating the maximum spectral acceleration 1.3 are appropriate in the western US, whereas ratios of 1.3
based on both horizontal components for an entire orbit and 1.4 may be more appropriate in the central and eastern
(orientations from 0° to 360°). The maximum direction is the US, for 0.2-second and 1.0-second spectral accelerations,
orientation with the maximum spectral demand (identified by respectively.
the point on the orbit farthest from the origin) [16, 17].
To illustrate, Figure 1 compares the geomean, SRSS, and Table 1. Ratio of Max. Demand to Geomean Demand.
maximum-direction spectra for the 1999 Kocaeli, Turkey P-695 Huang et al. (2010)
earthquake ground motion recorded at the Duzce station Near-Field Western US Central
and
(FEMA P-695 Far-Field No. 9 [18]). The maximum-direction Far- No Far- Near- Eastern
spectrum envelopes the component and geomean spectra, but T (s) Field Pulse Pulse Field Field US NEHRP
is less than the SRSS spectrum. 0 1.2 1.21 1.25 1.18 1.21 1.23 1.1
The direction of ground motion effects structural response. 0.2 1.29 1.24 1.31 1.2 1.22 1.28 1.1
Three-dimensional analyses of buildings, subjected to two 0.5 1.32 1.26 1.31 1.23 1.29 1.32 1.2

orthogonal components of ground motion simultaneously, 1 1.3 1.27 1.37 1.27 1.31 1.35 1.3

show that structural collapse is predominately in the direction 2 1.39 1.41 1.4 1.27 1.31 1.39 1.3

of the stronger component [4]. As a consequence, the overall 3 1.34 1.33 1.31 1.27 1.31 1.35 1.35
4 1.27 1.48 1.35 1.31 1.37 1.4 1.5
structural failure rate is higher for three-dimensional analyses

2930
 Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

Figure 3 shows a Vs30 map of the US based on topographic

2
FEMA P-695 Far-Field data. An important consideration using the topographic proxy
1.9 Ground Motion Set method is that the effect of sediment thickness is not included
1.8 (among other effects), and that the proxy method tends to
1.7 under-predict Vs30 for hard rock sites [28].
Average In the ASCE 7-10 approach to risk, site amplification of
1.6
ground motion is not considered until after calculation of the
Ratio

1.5
risk-targeted motions. Site amplification is dependent on the
1.4 spectral acceleration, however, so it varies for different values
1.3 of hazard. Clearly, a more accurate risk analysis requires a
1.2
geotechnical investigation (where possible), or an estimate of
site response using proxy data, to obtain the soil-amplified
1.1 NEHRP / ASCE 7-10
hazard curve before integrating hazard and vulnerability.
1
0 0.5 1 1.5 2 2.5 3 3.5 4
Spectral period, Tn (s) 3 STRUCTURAL VULNERABILITY

Figure 2. Ratio of maximum-direction to geomean Sa. Structural vulnerability can be assessed by using empirical
data (e.g. [34]) or by employing analytical models. The
2.2 Site Response uniform-risk ground motion maps for the US are based on the
The ASCE 7-10 (NEHRP) procedure to account for site latter, and incorporate two important assumptions regarding
response is founded on correlating shear wave velocity in the structural vulnerability:
top 30 meters, Vs30 with ground motion amplification. The • The fragility curve (mathematically, the probability of
correlation is based on ground motion records in the San collapse versus spectral acceleration) is described using a
Francisco bay area [20], and response resolution limited to cumulative distribution function (CDF). Prior analytical
two spectral period bands—short periods (near 0.2 seconds) studies suggest that this distribution is appropriate [35,
and medium range periods (near 1.0 seconds)—because the 36, 37, 38]. It follows therefore that the fragility curve
design spectrum is defined by two periods. Although a may be defined by two points on the curve, or a point on
procedure using a broader range of spectral periods and Next the curve and the slope of the curve (the logarithmic
Generation Attenuation (NGA) relationships for the western standard deviation, or dispersion, β).
US [21] and similar refinements [22] have been proposed, the • The fragility curve is defined by assuming that the
method appears to work well in the western US [23]. structure is designed using ASCE 7-10, and that this
The Eurocode 8 procedure is similar to the NEHRP implies two conditions [39]: first, there is a 10%
procedure, except that site response factors are applied to the probability of collapse under the MCE-level spectral
entire spectrum (instead of period bands), and that there are demand for risk categories I and II; second, the
some differences in defining soil classifications. dispersion, β is estimated to be 0.8 including all
Improvements have been suggested for both the classifications uncertainties. Analytical studies performed during the
and the amplification factors [24], and for incorporating some development of FEMA P-695 and “other past research”
knowledge of the type and depth of the soil column [25]. [19, 39] form the basis for these conditions.
Nevertheless, site response is complex. Ground motion The application of the latter assumption to the risk analysis
amplification depends on the shear wave velocity profile and of new structures in the US is examined in this section by
thickness of the soil column, bedrock characteristics, the reviewing FEMA P-695 evaluations reported to-date in the
magnitude and depth of the soil-to-bedrock impedance literature. Included are steel moment frame and braced-frame,
contrast, and topographic effects, among other factors [26]. reinforced concrete moment-frame and shear wall, reinforced
As a consequence, various alternative site response masonry shear wall, and light-frame wood shear wall
procedures have been proposed [27, 25, 28]. structures.
The USGS uniform-hazard ground motions reference a
shear wave velocity, Vs = 760 m/s. This corresponds to the
boundary of the ASCE 7-10 soil classification of “rock” (site
class B) and “very dense soil and soft rock” (site class C).
Although the actual bedrock shear wave velocity in the central
and eastern US is much higher (three times or more) [29], site
class B was selected because it represented an average of the
rock sites used in the western US attenuation relationships
[30].
Where it is not possible to know site conditions with
certainty (such as when generating a ground motion map for
code adoption), a coarse estimate of ground motion
amplification can be correlated using geological data [28],
topographic data as a proxy for Vs30 [31, 32], or a combination
thereof [33]. Ground motion maps along these lines are Figure 3. Vs30 map of US based on topographic data.
proposed for Italy [11]. Vs30 values range from 180 m/s (blue) to 760 m/s (red).

2931
Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

In a FEMA P-695 evaluation, a structure is deemed to Table 3. Steel braced-frame results.

“pass” if (1) the average conditional collapse probability for a Accept.
LFRS β PG No. Ω ACMR
10%
group of similar archetypical structures does not exceed 10%, 1 1.42 1.63
and if (2) the conditional collapse probability for any Steel 2 1.67 3.37
individual structure does not exceed 20%. FEMA P-695 Special
0.525 3 1.9 2.93 1.96
CBF (R=6,
includes the effect of modeling and design uncertainties in the Ω=2) 4 1.87 4.73
evaluation, and it incorporates an estimate of the effect of Ave 1.9 3.17
spectral shape on collapse capacity [18]. 1 1.4 3.13
Steel BRB 2 1.21 4.12
3.1 Steel Moment Frames (R=8, 0.525 3 1.77 2.86 1.96
Ω=2.5) 4 1.29 4.19
The vulnerability of steel moment frame structures is
Ave 1.77 3.58
summarized in Table 2. Three types of steel moment
structures are examined: special steel moment frames (SMF) 3.3 Reinforced Concrete Moment Frames
[40], partially-restrained composite connection (PR-CC) steel
moment frames [41, 42], and cold-formed steel bolted The vulnerability of reinforced concrete moment frame
moment frames [43]. For each Performance Group (PG) the structures is summarized in Table 4. Two types of moment-
frames are included: special (SMF) and ordinary (OMF) based
dispersion, overstrength, Ω, Adjusted Collapse Margin Ratio
on the FEMA P-695 supporting study [18].
(ACMR) accounting for the effect of spectral shape, and the
Most concrete frame structures pass FEMA P-695
acceptable ACMR corresponding to 10% probability of
requirements. However, SMF frames taller than 4 stories (the
collapse are given.
long-period perimeter frame performance groups) do not. The
All steel SMF systems are adequate except one performance
long-period performance group for OMF frames also fails to
group: the steel SMF long-period SDC Dmax performance
pass. Accordingly, the study suggested that a height limit is
group does not pass (but is not far off).
needed for SMF frames and that OMF frames should not be
allowed.
Table 2. Steel moment-frame results.
Accept. The reinforced concrete moment frame results and
LFRS β PG No. Ω ACMR
10% recommendations underscore a weakness of the assumptions
0.5 1 4.16 2.97 1.9 for the current uniform risk-targeted ground motions:
0.45 2 3.67 1.94 1.75 structures designed using ASCE 7-10 do not necessarily have
0.475 3 2.55 2.53 1.81
less than a 10% probability of collapse given the MCE.
Steel 0.45 4 4.27 3.13 1.77
special Ave 4.27 2.64
moment
0.5 1 4.7 2.91 1.9
Table 4. Reinforced concrete moment-frame results.
frame (R=8, Accept.
Ω=3) 0.475 2 2.61 1.76 1.81 LFRS β PG No. Ω ACMR
10%
0.45 3 2.84 2.72 1.78 1 1.7 1.9
0.475 4 2.93 3.28 1.84 Reinforced 2 1.9 1.84
Ave 4.7 2.67 concrete
0.5 3 3.5 2.66 1.9
SMF (R=8,
1 1.39 2.19
Ω=3) 4 2.8 2.36
2 1.57 2.89 Ave 3.5 2.19
Steel PR- 3 1.58 3.99 1 2.2 3.86
2.16
CC frame 0.6 4 1.59 3.96 2 6 4.58
(R=6, Ω=3) 5 1.78 4.74 3 1.6 2.2
6 1.63 4.25 4 3.2 2.44
Reinforced
Ave 1.78 3.67 concrete Ave 6 3.27
0.575 2.09
Cold-form 1 1.9 2.2 OMF (R=3, 1 1.6 2.2
steel 2 2.09 2.16 Ω=3) 2 3.2 2.44
Special
0.5 3 3.14 3.66 1.9 3 1.5 1.48
bolted MF
(R=3.5, 4 5.59 4.98 4 2.4 2.12
Ω=3) Ave 5.59 3.25 Ave 3.2 2.06

3.2 Steel Braced Frames 3.4 Reinforced Concrete Shear Walls

The vulnerability of steel braced-frame structures is The vulnerability of reinforced concrete shear wall structures
summarized in Table 3. Two types of steel braced-frames are is summarized in Table 5. Two types of shear walls are
considered [40]: steel special concentrically braced frames considered: special and ordinary [40].
(CBF), and buckling restrained braced (BRB) frames. (For The results indicate that long-period performance groups
further consideration of BRB frames, see also [44].) pass, but short-period performance groups do not. One and
All steel braced-frame structures satisfy the FEMA P-695 two story structures collapsed due to shear failures at low drift
requirements, except one performance group: the CBF short- levels (about 1.5%). The study notes that in past earthquakes,
period SDC Dmax performance group did not pass. However, collapse of low-rise concrete shear wall buildings has only
a recent study of special CBF buildings [45] indicates that 3- been observed in precast parking type structures where the
story and 20-story buildings also do not pass. In fact, short- diaphragm failed, yet “insufficient information exists to
period systems often do not pass the FEMA P-695 criterion, establish more liberal failure criteria.”
regardless of the type of structure [46, 47].

2932
 Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

Table 5. Reinforced concrete shear-wall results. margins of wood buildings in the western US are significantly
Accept. lower than those in the central and eastern US due to regional
LFRS β PG No. Ω ACMR
10%
1 1.98 1.5
construction practices [48]. In this manner, it is possible that
2 1.54 3.18 wood structures may not necessarily have the performance
3 2.41 1.5 implied by conformance with ASCE 7-10.
Reinforced
concrete 4 3.88 5.71
1.96
Special SW 0.525 5 2.08 1.75 Table 7. Wood (light-framed) shear-wall results.
(R=5, 6 1.7 3.2 Accept.
Ω=2.5) LFRS β PG No. Ω ACMR
7 2.41 1.5 10%
0.5 1 2.2 1.89 1.9
8 3.52 4.65 Wood light-
frame wood 0.675 2 3.4 2.48 2.38
Ave 3.88 3.05
panels 0.675 3 4.1 2.91 2.38
1 2.85 1.77
(R=2, 0.675 4 3.6 3.18 2.38
2 3.43 5.8 Ω=2.5)
Ave 4.1 2.62
3 3.25 2.28
Reinforced
concrete 4 5.2 8.23
1.96 3.7 Discussion
Ordinary 0.525 5 2.85 1.77
SW (R=4, 6 2.37 6.42 To summarize this section, the available P-695 evaluations
Ω=2.5)
7 3.25 2.28 indicate that, on average, the assumed ASCE 7-10 fragility
8 3.37 9.38 curve is conservative in terms of the variability in response, at
Ave 5.2 5.06 least for structures that have been detailed for seismic
resistance. Yet, the conditional probability of collapse may
3.5 Reinforced Masonry Shear Walls
exceed 10%, even for structures designed using ASCE 7-10.
The vulnerability of reinforced masonry shear wall structures
is summarized in Table 6. Again, two types of shear walls are 4 COLLAPSE RISK
considered: special and ordinary [40]. Calculation of risk requires the integration of seismic hazard
The results indicate that many reinforced masonry curves and structural vulnerability (fragility) curves. The
structures do not pass the FEMA P-695 criteria. 1-, 2-, and 4- mean hazard exceedence frequency corresponding to an
story OMF shear wall buildings with low gravity loads pass, intensity measure (demand parameter) is shown in Equation 1.
but high gravity loads do not. Part of the poor performance is This equation means that hazard may be expressed as a
attributed to the low ductility capacity of short shear walls. probability that the intensity of an earthquake, A will exceed
Regardless of the reason for performance, the results again some level, a in a given year:
demonstrate that structures designed using ASCE 7-10 do not =  |  1 (1)
necessarily meet the 10% probability of collapse assumption. For seismic hazard analysis, a commonly used intensity
measure is spectral acceleration. As noted earlier, one
Table 6. Reinforced masonry shear-wall results. example of a fragility curve is simply the probability of
Accept.
LFRS β PG No. Ω ACMR
10% collapse or failure, F conditioned on the occurrence of a
1 2 1.41 ground motion hazard value. The vulnerability of the
2 1.75 2.3 structure to the hazard (conditional probability of failure) is
Reinforced
3 2.12 1.69 calculated using Equation 2, where f(t) is the probability
masonry 4 1.75 2.25 density function of spectral acceleration:
1.96
Special SW 0.525 5 1.8 1.71
(R=5, 6 1.53 2.33
| (2)
Ω=2.5) Integrating hazard and fragility depends on the overall
7 1.71 1.95
8 1.5 2.28 approach. In a “deterministic” analysis, each fragility
Ave 2.12 1.99 ordinate is multiplied by the corresponding hazard probability
1 1.89 1.51 2.23 [49]. In probabilistic seismic hazard analysis (PSHA), there
2 2.08 1.54 2.38 are three analytically equivalent methods to integrate hazard
3 1.41 1.46 2.16
Reinforced and fragility.
masonry 4 1.99 1.96 2.38
Ordinary 0.525 5 1.91 1.98 2.23 4.1 Method 1: Numerical Derivative of Hazard Curve
SW (R=2, 6 1.67 2.69 2.38
Ω=2.5)
7 1.37 1.94 2.16
The integration of hazard and fragility is shown in the
8 1.64 3.19 2.38
following equation and is the so-called “risk integral” which
Ave 1.99 2.20 calculates an annual rate of failure, λF:
  | (3)
3.6 Wood (Light-Framed) Shear Walls The failure rate usually refers to the collapse rate, λc which
can also be used to calculate the risk of collapse in an
The vulnerability of light-frame wood shear wall structures
exposure time (such as calculating the probability of collapse
(i.e. residential and small commercial buildings) is
in 50 years).
summarized in Table 7. The table is based on the results from
The minus sign accounts for the negative slope of the
the P-695 supporting study [18]. All performance groups
hazard curve (in PSHA the hazard curve is the probability of
pass, except the short-period low-aspect-ratio performance
exceeding a value so it decreases as the demand parameter
group. However a recent study shows that the collapse

2933
Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

increases). An equivalent version of the risk integral would 4.4 Discussion

be to use an absolute value instead of a minus sign. This section discusses the methods to calculate risk using an
The risk integral equation represents the concept that the example: the collapse risk of a 4-story steel special moment
probability of failure is the probability of the ground motion frame building in Los Angeles. For purposes of comparison,
times the probability of building failure given that level, the ground motion hazard is determined using the 2002 USGS
integrated over all possible levels of hazard [50]. The hazard data (fundamental building period, T = 1.33 seconds)
equation assumes ergodicity, meaning hazard and fragility and prior fragility data [35]. Using a 4-degree polynomial
(conditional failure) are independent and “the structure does curve fit, the collapse rate (number of predicted collapses per
not deteriorate” and is “instantaneously restored to its original year) λc = 3.51×10-4 for Los Angeles. A comparison between
state after each damaging earthquake.” This approximation is risk calculated using numerical derivative (Method 1) and
only accurate as long as the failure rate is small. For curve fitting (Method 2) is shown in Figure 4.
probabilities of failure greater than 1%, the approximation
0
overestimates risk [51]. 10

The hazard curve is not commonly represented using a

continuous function, but instead a discrete data set is used, so 10
−1

in practice the derivative is approximated numerically [52].

Mean annual rate of exceedance

−2
10
4.2 Method 2: Derivative of Fitted Hazard Curve
An alternative approach to using the discrete set of hazard a) 10
−3

data points is to fit the hazard curve with a function (usually a MCE
polynomial) and then take the derivative of that function when 10
−4

calculating Equation 3 [35].

4.3 Method 3: Derivative of Fragility Curve 10
−5

Since each of the previous approaches leads to some −6

10
approximation, an alternate form of the risk integral that does 0 0.5 1 1.5 2
Spectral acceleration (g)
not require differentiating the hazard curve is attractive. This
alternate method is used by the USGS to create risk-targeted x 10
−4

6
ground motion maps [53]. The alternate form is derived as
follows. Numerical Derivative
Given the risk integral: 5 of Hazard Curve

  |
4
Let Derivative
Failure rate

  | b) 3
of Fitted
| Hazard
  Curves
which is the same as 2

   
and let 1

 
0
  0 0.5 1 1.5 2
Spectral acceleration (g)
Invoking definite integration by parts:
  0.02

  |   0.018

0.016
  |      
0.014
  |      
Probability of failure

0.012
c)
  1 0 0 1     0.01

0.008

     (4) 0.006

Note that this form of the risk integral (Equation 4) requires 0.004

the derivative of the fragility curve—the CDF, which yields 0.002

the probability density function (PDF). This is a more stable 0
approach than taking the derivative of the hazard curve. 0 0.5 1 1.5 2 2.5 3 3.5 4
Spectral acceleration (g)
It is important to note that there is not a closed form
solution to the risk integral regardless of the method used, and Figure 4. Hazard, deaggregation of risk, and cumulative risk.
in practice the integral is computed numerically.

2934
 Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

For Method 2, various degree polynomials are fit to the Besides, using a curve-fit (Method 2) is not necessarily
hazard curve. The effect of the methods to determine the slope more accurate, because the derivative of the curve is unstable.
of the hazard curve is shown in Figure 4a. For the hazard The deaggregation of the risk is also different in each
curve, the black line is linearly interpolated in log-log space. method because the numerical integration of incremental
For the risk deaggregation curves (Figure 4b) the gray line is hazard (da) differs for each method. Thus, Method 1 and
based on the incremental slope, and colors represent a 3- Method 3 produce the same value of risk, but differ in terms
degree (cyan), 4-degree (blue), 5-degree (magenta), and 6- of hazard contributions, which is easily visualized using
degree (green) polynomial curve fit for the hazard curve. deaggregation. Deaggregation of collapse risk is analogous to
A 4-degree or more polynomial curve fit converges to the using deaggregation of ground motion to ascertain magnitude
same cumulative risk (Figure 4c). For very small spectral and distance pairs that are primary contributors to hazard [54].
accelerations (spectral acceleration less than 0.1g), a 5-degree Using Method 2, risk may be over- or under-estimated,
polynomial is a very bad fit (this effect is not visible in Figure depending on the shape of the hazard curve relative to the
4a). Using the derivative of the fitted polynomial hazard fragility curve. Hazard curve points are derived from
curve gives λc = 3.51×10-4 and risk = 1.74% (green line). interpolated data, so the validity of using a fitted curve is
Using the incremental derivative of hazard curve gives λc = questionable. Furthermore, fitted curves are sensitive at low
3.97×10-4 and risk = 1.97%. values of spectral accelerations. This may be an additional
Figure 5 shows a comparison between Method 1, Method 2 concern for ascertaining non-performance at lower hazard
using a 4-degree polynomial fit, and Method 3 (integration by levels (i.e. serviceability and immediate occupancy levels).
parts and the derivative of the fragility curve). Using Method
3 gives λc = 3.97×10-4 and risk = 1.97% (red line). 5 EXAMPLE APPLICATIONS
The method used to calculate risk may significantly affect Two example applications of earthquake risk are provided.
results. Both the contribution of different hazards to this risk First, the collapse risk of buildings that use partial restraint
(the deaggregation shown in the Figure 5a) and the cumulative composite connection (PR-CC) moment frames is assessed.
value of risk calculated (Figure 5b) may vary. The derivative Second, a collapse risk map of the US for a building using
of the fragility curve (Method 3) is a more stable approach non-ductile steel moment frames is produced.
compared to using the other methods, because the derivative
5.1 Example: Risk analysis of PR-CC buildings
of the hazard curve is avoided.
This example contrasts risk in a location with frequent
x 10
−3 seismicity, in the San Francisco bay area (38.0° N, 121.7° W),
1.2 and a location with infrequent seismicity, in the Memphis
MCE metropolitan area (35.2°, N, 89.9° W).
Collapse rate, P(C|Sa)*|dλ/dSa| (1/g/year)

1 The deaggregation of seismic hazard (spectral acceleration)

corresponding to a long-period structure (T = 1.0 seconds)
0.8 indicates that for San Francisco, the dominant contribution to
Derivative of Fragility Curve hazard is along one fault line producing magnitude 5.0 to 7.0
a) 0.6
earthquakes. For Memphis the dominant hazard is one fault
(New Madrid) about 60 km away. Compared to San
Numerical Derivative Francisco, however, in Memphis there are large uncertainties
0.4 of Hazard Curve
with respect to source and size of potential earthquakes.
Derivative of Fitted
Structural vulnerability results for PR-CC moment frame
0.2
Hazard Curve buildings (Table 2) indicate the lateral system over-strength is
somewhat less than anticipated, keeping in mind that Ω in
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 ASCE 7-10 is a conservative upper bound intended to protect
Spectral acceleration, Sa(T=1.33s) (g) vulnerable components. The primary collapse mechanism of
the buildings is generally column hinging of the lower story
0.02
columns. The low collapse margin ratios adjusted for spectral
0.018 shape and period elongation (ACMR) for the 4 story buildings
0.016 (performance groups 1 and 2) suggest the mid-rise buildings
0.014
are more sensitive to collapse than higher-rise buildings.
Probability of collapse, P

Table 8 gives the collapse risk for each building

0.012
performance group in Table 2. Site response is included
b) 0.01 based on the NEHRP relationship and topographic data.
0.008 Risk is calculated using the derivative of the fragility curve
0.006
and is based on achieving the ACMR values in Table 2. It is
important to note that this definition of fragility (achieving the
0.004
ACMR) represents the predicted sensitivity of code-
0.002
conforming PR-CC buildings in general, not the sensitivity of
0
0 0.5 1 1.5 2 2.5 3 3.5 4
the specific buildings proportioned in the analytical study. In
Spectral acceleration, Sa(T=1.33s) (g) this way, building fragility is not constant (it increases or
decreases geographically) and has a consistent safety margin.
Figure 5. Deaggregation of risk and cumulative risk.

2935
Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

Table 8. Collapse risk analysis of PR-CC buildings. Site response based

Performance Group Summary
Risk analysis (Prob. on B/C boundary
collapse in 50 years)
No. Stories Hx (ft) Te (s) ACMR SF Memphis
1 4 12 0.87 2.19 1.30% 0.76%
2 4 14 0.98 2.89 0.45% 0.39%
3 6 12 1.2 3.99 0.12% 0.17%
4 6 14 1.36 3.96 0.12% 0.17%
5 8 12 1.51 4.74 0.05% 0.10%
6 8 14 1.71 4.25 0.08% 0.14%

The analysis is fairly insensitive to the Vs30, but highly

sensitive to the method used to calculate the risk (λc =
2.57×10-4 collapses per year using the derivate of the fragility
curve, compared to λc = 7.44×10-4 collapses per year using the
derivative of the fitted hazard curve). Site response based on
NGA relationships and
All the PR-CC buildings meet the acceptable risk objective topographic data
of the building code (collapse risk not greater than 1% in 50
years), except for 4-story buildings with a 12-foot story height
in the San Francisco location. Surprisingly, for 6- to 8-story
buildings, the risk of collapse in the Memphis location is
greater than in the San Francisco location. Importantly, a
deaggregation of the risk (not shown in Table 8) reveals that
the contribution to risk shifts from likely events (earthquakes
with return periods less than 2,475 years) to unlikely events
(earthquakes with return periods greater than 2,475 years).
The outcomes of this example serve to illustrate that hazard
(aleatoric) uncertainty and analysis (epistemic) uncertainty
lead to rare-event dispersion.
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1
5.2 Example: Risk map for a non-ductile SMF building 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

This example shows cumulative risk maps (risk of collapse in Figure 6. 50-year collapse risk for 4-story non-ductile building
50 years) for a 4-story non-ductile building in the US. The Assessment of collapse risk via the integration of hazard
risk maps (Figure 6) are based on the 2008 USGS hazard data and fragility is more stable when calculated using the
and structural vulnerability [55], considering only the moment derivative of the fragility curve. Using the derivative of the
frame (ignoring the gravity frame contribution). Risk is hazard curve to calculate risk is more computationally
calculated for every 0.5-degree latitude and longitude. Figure expensive, unstable, inaccurately determines the contribution
6a is based on assuming the B/C site class boundary (Vs30 = of hazard, and, depending on the method used, may lead to
760 m/s). Site response is included in Figure 6b based on grossly inaccurate values of risk.
NGA relationships and topographic data (shown in Figure 3). Of course, earthquake risk analysis should be viewed in
The comparison demonstrates the critical effect of site proper perspective. Interaction of hazards and vulnerability
response in earthquake risk analysis. Without incorporating and the corresponding calculated risk may be amplified or
site response, the map would infer that non-ductile structures attenuated by public response. In other words, in addition to
are only prone to collapse in high-seismic zones. calculating risk, an overall assessment should include
political, psychological, sociological, and cultural factors [56].
6 CONCLUSIONS
Political factors especially amplify risk [57]. Additionally,
Ground motions for risk analysis should reflect regional and risk may be compounded in “megacities,” where there is a
local differences in seismic hazard. For example, in the concentration of populations and infrastructure [58, 59, 60].
western US, the ratio of maximum-direction to geomean Society’s acceptance of risk is also complex and changing.
spectral acceleration is approximately 1.2 and 1.3, for 0.2- Voluntary risks, such as driving automobiles and air travel,
second and 1.0-seconds, respectively, whereas in the central are tolerated more compared to involuntary risks associated
and eastern US ratios of 1.3 and 1.4 are more appropriate. with earthquakes [61]. Yet even with that distinction, the
Risk targeted motions can benefit from topographic data in degree of risk tolerated by society is a moving target.
order to predict soil amplification for design when site Earthquake risk analysis is limited by poor resolution of site
investigations have not been performed a priori. Adjusting for response, aleatoric uncertainty (although this is partially
soil site conditions later in the design process may accounted for), epistemic uncertainty [62], and rare-event
substantially under- or over-estimate risk. dispersion (uncertainty tends to dominate for extremely rare
Risk analysis is improved through fragility curves that events). Instead of “predicting” collapse, earthquake risk
recognize system-dependent vulnerability. Structures, short- analysis is more appropriately used as a tool to help allocate
period structures in particular, may not provide a 10% or less resources and determine policies that reduce loss of life,
conditional collapse probability. downtime, and monetary damages in the community [2].

2936
 Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

ACKNOWLEDGMENTS [22] E. Seyhan, J. P. Stewart. Site response in NEHRP provisions and NGA
models. Geotechnical Engineering State of the Art and Practice,
This research was supported by the U.S. National Institute of Proceedings, Geo-Congress, March 25-29, Oakland, California, 2010.
Standards and Technology grant No. 60ANB10D107. The [23] R. D. Borcherdt. Empirical evidence for site coefficients in building
views presented in this paper are solely those of the authors. code provisions. Earthquake Spectra 2002, 18(2): 189–217.
[24] K. Pitilakis, E. Riga, A. Anastasiadis. Design spectra and amplification
factors for Eurocode 8. Bulletin of Earthquake Engineering 2012, 10(5):
REFERENCES 1377–1400.
[1] World Bank. Natural hazards, unnatural disasters: the economics of [25] K. Pitilakis, E. Riga, A. Anastasiadis. New code site classification,
effective prevention. International Bank for Reconstruction and amplification factors and normalized response spectra based on a
Development (World Bank), Washington, D.C., 2010. worldwide ground-motion database. Bulletin of Earthquake Engineering
[2] National Academies Press (NAP). Disaster resilience: a national 2013, 11(4): 925–966.
imperative. The National Academies, Washington, D.C., 2012. [26] S. L. Kramer. Geotechnical earthquake engineering. Prentice-Hall, Inc.,
[3] American Society of Civil Engineers (ASCE). Minimum design loads Upper Saddle River, New Jersey, 1996.
for buildings and other structures. ASCE/SEI 7-05, American Society [27] A. Rodriguez-Marek, J. D. Bray, N. A. Abrahamson. An empirical
of Civil Engineers, Reston, Virginia, 2005. geotechnical seismic site response procedure. Earthquake Spectra 2001,
[4] Federal Emergency Management Agency (FEMA). NEHRP 17(1):65–87.
Recommended provisions for seismic design of new buildings and other [28] A. R. Kottke, Y. M-A. Hashash, J. P. Stewart, C. J. Moss, S. Nikolaou,
structures. FEMA P-750, Federal Emergency Management Agency, E. M. Rathje, W. J. Silva, K. W. Campbell. Development of geologic
Washington, D.C., 2009. site classes for seismic site amplification for central and eastern North
[5] J. J. Bommer, R. Pinho. Adapting earthquake actions in Eurocode 8 for America. Proceedings, 15th World Conference on Earthquake
performance-based seismic design. Earthquake Engineering and Engineering, Lisbon, Portugal, 2012.
Structural Dynamics 2006, 35(1): 39–55. [29] S. Nikolaou, J. E. Go, C. Z Beyzaei, C. Moss, P. W. Deming. Geo-
[6] Applied Technology Council (ATC). Tentative provisions for the seismic design in the eastern United States: state of practice.
development of seismic regulations for buildings. ATC-3-06, Applied Proceedings, Geotechnical Engineering State of the Art and Practice,
Technology Council, Redwood City, California, 1978. Geo-Congress, Oakland, California, 2012.
[7] American Society of Civil Engineers (ASCE). Minimum design loads [30] Federal Emergency Management Agency (FEMA). NEHRP
for buildings and other structures. ASCE/SEI 7-10, American Society recommended provisions and commentary for seismic regulations for
of Civil Engineers, Reston, Virginia, 2010. new buildings and other structures. FEMA P-450, Federal Emergency
[8] R. P. Kennedy. Performance-goal based (risk informed) approach for Management Agency, Washington, D.C., 2003.
establishing the SSE site specific response spectrum for future nuclear [31] D. J. Wald, T. I. Allen. Topographic slope as a proxy for seismic site
power plants. Nuclear Engineering and Design 2011, 241(3): 648-656. conditions and amplification. Bulletin of the Seismological Society of
[9] E. V. Leyendecker, R. J. Hunt, A. D. Frankel, K. S. Rukstales. America 2007, 97(5): 1379–1395.
Development of maximum considered earthquake ground motion maps. [32] T. I. Allen, D. J. Wald. On the use of high-resolution topographic data
Earthquake Spectra 2000, 16(1): 21-40. as a proxy for seismic site conditions (Vs30). Bulletin of the
[10] J. P. Judd, F. A. Charney. Performance-based design in the central and Seismological Society of America 2009, 99(2A): 935–943.
eastern United States. Proceedings, Structures Congress, Boston, [33] H. Magistrale, Y. Rong, W. Silva, E. Thompson. A site response map of
Massachusetts, 2014. the continental U.S. Proceedings, 15th World Conference on
[11] H. Crowley, M. Colombi, B. Borzi, M. Faravelli, M. Onida, M. Lopez, Earthquake Engineering, 24–28 September, Lisbon, Portugal, 2012.
D. Polli, F. Meroni, R. Pinho. A comparison of seismic risk maps for [34] T. Rossetto, A. Elnashai. Derivation of vulnerability functions for
Italy. Bulletin of Earthquake Engineering 2009, 7(1): 149–180. European-type RC structures based on observational data. Engineering
[12] J. Douglas, T. Ulrich, C. Negulescu. Risk-targeted seismic design maps Structures 2003, 25(10): 1241–1263.
for mainland France. Natural Hazards 2013, 65(3): 1999–2013. [35] L. Eads, E. Miranda, H. Krawinkler, D. G. Lignos. An efficient method
[13] D. Giardini, J. Woessner, L. Danciu, H. Crowley, F. Cotton, G. for estimating the collapse risk of structures in seismic regions.
Grünthal, R. Pinho, G. Valensise and the SHARE consortium. SHARE Earthquake Engineering and Structural Dynamics 2013, 42(1): 25–41.
European seismic hazard map for peak ground acceleration, 10% [36] L. F. Ibarra, H. Krawinkler. Global collapse of frame structures under
exceedance probabilities in 50 years. http://www.share-eu.org, 2013. seismic excitations. Report No. 152, The John A. Blume Earthquake
[14] D. Giardini, J. Woessner, L. Danciu, H. Crowley, F. Cotton, G. Engineering Center, Department of Civil and Environmental
Grünthal, R. Pinho, G. Valensise, S. Akkar, R. Arvidsson, R. Basili, T. Engineering, Stanford University, Stanford, California, 2005.
Cameelbeeck, A. Campos-Costa, J. Douglas, M. B. Demircioglu, M. [37] B. A. Bradley, R. P. Dhakal. Error estimation of closed-form solution
Erdik, J. Fonseca, B. Glavatovic, C. Lindholm, K. Makropoulos, C. for annual rate of structural collapse. Earthquake Engineering and
Meletti, R. Musson, K. Pitilakis, K. Sesetyan, D. Stromeyer, M. Structural Dynamics 2008, 37(15): 1721–1737.
Stucchi, A. Rovida, Seismic Hazard Harmonization in Europe [38] M. Ghafory-Ashtiany, M. Mousavi, A. Azarbakht. Strong ground
(SHARE): Online Data Resource. http://www.efehr.org, 2013. motion record selection for the reliable prediction of the mean seismic
[15] N. Abrahamson, K. M. Shedlock. Overview: special issue on ground collapse capacity of a structure group. Earthquake Engineering and
motion attenuation relations. Seismological Research Letters 1997, Structural Dynamics 2011, 40(6): 691–708.
68(1): 9–23. [39] N. Luco, B. R. Ellingwood, R. O. Hamburger, J. D. Hooper, J. K.
[16] Y.-N. Huang, A. S. Whittaker, N. Luco. Maximum spectral demands in Kimball, C. A. Kircher. Risk-targeted versus current seismic design
the near-fault region. Earthquake Spectra 2008, 24(1): 319-341. maps for the conterminous United States. Proceedings, SEAOC 76th
[17] Y.-N Huang, A. S. Whittaker, N. Luco. Establishing maximum spectral Annual Convention. Structural Engineers Association of California,
demand for performance-based earthquake engineering: collaborative Sacramento, California, 2007.
research with the University at Buffalo and the USGS. Technical [40] National Institute of Standards and Technology (NIST). Evaluation of
Report, United States Geological Survey, Reston, Virginia, 2010. the FEMA P-695 methodology for quantification of building seismic
[18] Federal Emergency Management Agency (FEMA). Quantification of performance factors. NIST GCR 10-917-8, National Institute of
building seismic performance factors. FEMA P-695, Federal Standards and Technology, Gaithersburg, Maryland, 2010.
Emergency Management Agency, Washington, D.C, 2009. [41] A. Bozorgmehr. Collapse assessment of PR composite connections.
[19] Federal Emergency Management Agency (FEMA). NEHRP 2009 M.S. thesis, Civil and Environmental Engineering, Chalmers University
NEHRP recommended seismic provisions: design examples. FEMA P- of Technology, Göteborg, Sweden, 2012.
751, Federal Emergency Management Agency, Washington, D.C., 2012. [42] A. Bozorgmehr, R. T. Leon. Collapse assessment of partial restraint
[20] R. D. Borcherdt. Estimates of site-dependent response spectra for design composite connection (PR-CC) moment resisting frames using FEMA
(methodology and justification). Earthquake Spectra 1994, 10(4): 617– P695 (ATC-63) methodology. Proceedings, 15th World Conference on
653. Earthquake Engineering, 24-28 September, Lisbon, Portugal, 2012.
[21] Y.-N. Huang, A. S. Whittaker, N. Luco. NEHRP site amplification [43] A. Sato, C.-M. Uang. A FEMA P695 study for the proposed seismic
factors and the NGA relationships. Earthquake Spectra 2010, 26(2): performance factors for cold-formed steel special bolted moment
583–593. frames. Earthquake Spectra 2013, 29(1): 259–282.

2937
Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014

[44] C. Ariyaratana, L. A. Fahnestock. Evaluation of buckling-restrained

braced frame seismic performance considering reserve strength.
Engineering Structures 2011, 33: 77–89.
[45] P.-C. Hsiao, D. E. Lehman, C. W. Roeder. Evaluation of the response
modification coefficient and collapse potential of special concentrically
braced frames. Earthquake Engineering and Structural Dynamics 2013,
42(10): 1547–1564.
[46] F. A. Charney, S. Darling, M. Eatherton. Seismic performance of very
short period buildings. Proceedings, 15th World Conference on
Earthquake Engineering, Lisbon, Portugal, 2012.
[47] National Institute of Standards and Technology (NIST). Tentative
framework for development of advanced seismic design criteria for new
buildings. NIST GCR 12-917-20, National Institute of Standards and
Technology, Gaithersburg, Maryland, 2012.
[48] Y. Li, Y. Yin, B. R. Ellingwood, W. M. Bulleit. Uniform hazard versus
uniform risk bases for performance-based earthquake engineering of
light-frame wood construction. Earthquake Engineering and Structural
Dynamics 2010, 39(11): 1199–1217.
[49] Z. Wang. Seismic hazard assessment: issues and alternatives. Pure and
Applied Geophysics 2011, 168(1-2): 11–25.
[50] R. P. Kennedy. Performance-goal based (risk informed) approach for
establishing the SSE site specific response spectrum for future nuclear
power plants. Nuclear Engineering and Design 2011, 241: 648–656.
[51] A. Der Kiureghian. Non-ergodicity and PEER’s framework formula.
Earthquake Engineering and Structural Dynamics 2005, 34:1643–1652.
[52] Federal Emergency Management Agency (FEMA). Seismic
performance assessment of buildings. FEMA P-58, Federal Emergency
Management Agency, Washington, D.C, 2012.
[53] N. Luco. Towards risk-targeted ground motion maps. USGS central and
eastern United States Workshop, May 9, Boston, Massachusetts, 2006.
[54] M. C. Chapman. A probabilistic approach to ground-motion selection
for engineering design. Bulletin of the Seismological Society of America
1995, 85(3): 937–942.
[55] J. P. Judd, F. A. Charney. Seismic collapse prevention system.
Proceedings, 10th National Conference in Earthquake Engineering,
Earthquake Engineering Research Institute, Anchorage, Alaska, 2014.
[56] R. E. Kasperson. A perspective on the social amplification of risk. The
Bridge 2012, 42(3): 23–27.
[57] D, Metlay, D. Sarewitz. Decision strategies for addressing complex,
‘messy’ problems. The Bridge 2012, 42(3): 6–16.
[58] R. Aster. Expecting the unexpected: black swans and seismology.
Seismological Research Letters 2012, 83(1): 5–6.
[59] R. Bilham. The seismic future of cities. Bulletin of Earthquake
Engineering 2009, 7(4): 839–887.
[60] T. L. Holzer, J. C. Savage. Global earthquake fatalities and population.
Earthquake Spectra 2013, 29(1): 155–175.
[61] Applied Technology Council (ATC). Tentative provisions for the
development of seismic regulations for buildings. ATC-3-06 Amended,
Redwood City, California, 1984.
[62] S. Stein, J. L. Stein. Shallow versus deep uncertainties in natural hazard
assessments. Eos 2013, 94(14):133–140.

2938