You are on page 1of 264

INFORMATION TO USERS

This manuscript has been reproduced from the microfilm master. UMI films
the text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality illustrations
and photographs, print blood through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and continuing
from left to right in equal sections with small overlaps.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6* x 9" black and white
photographic prints are available for any photographs or illustrations appearing
in this copy for an additional charge. Contact UMI directly to order.
Bell & Howell Information and Learning
300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA
800-521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
QUANTIFYING THE UNCERTAINTY IN SEISMIC RISK AND LOSS
ESTIMATION
Patricia Grossi
A DISSERTATION
in
Systems Engineering
Presented to the Faculties of the University of Pennsylvania in Partial Fulfillment of
the Requirements for the Degree of Doctor of Philosophy
2000
Supervisor of Dissertation
Graduate Group Chairperson
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number 9976429
Copyright 2000 by
Grossi, Patricia
All rights reserved.
_ ___ _
UMI
UMI Microform9976429
Copyright 2000 by Bell & Howell Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
Bell & Howell Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
COPYRIGHT
Patricia Grossi
2000
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DEDICATION
To my parents
in
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGMENTS
The support of the National Science Foundation through grant CMS97-14401 and the
Lillian Beck Fund in the School of Engineering and Applied Science at the University of
Pennsylvania are both gratefully acknowledged. Additionally, this work has been partially
supported through the NEHRP Graduate Fellowship in Earthquake Hazard Reduction awarded by
the Earthquake Engineering Research Institute under a cooperative program funded by the
Federal Emergency Management Agency.
The support o f numerous individuals is also gratefully acknowledged. This work could
not have been completed without the members of my dissertation committee, others in industry
working with me as part of the National Science Foundation project, and the respondents to my
survey. First, I would like to thank the members of my dissertation committee from the
University of Pennsylvania: Paul Kleindorfer and Howard Kunreuther from the Department of
Operations and Information Management at the Wharton School and G. Anandal ingam and John
Lepore from the Department of Systems Engineering. I would especially like to thank Robert
Whitman from the Massachusetts Institute of Technology, who so graciously took the time to
advise me on various aspects of the HAZUS methodology. Also, Weimin Dong of Risk
Management Solutions, Inc. and Scott Lawson of Durham Technologies, Inc., as part of the
National Science Foundation project, provided the necessary material and support for the creation
of the Scenario Builder and EP Maker software modules used in this work. Finally, the
anonymous structural engineers and contractors in California who took the time to respond to my
expert opinion survey provided me with crucial data for this work.
I would also like to acknowledge the support of those in the Wharton School and in
industry with whom I have worked on the Managing Catastrophic Risks project, especially those
on the Technical Advisory Committee. From the Wharton School, I would like to thank Steve
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Levy and Jaideep Hebbar for their support. From industry, I would like to acknowledge Karen
Clark and Nozar Kishi of Applied Insurance Research, Inc., Dennis Kuzak and Tom Larsen of
EQE, and Hemant Shah and Don Windeler of Risk Management Solutions, Inc. for their
dedication to the project.
Of course, I must thank my friends, family, and colleagues who have supported me
throughout the years. The help of Shelley Brown and Denice Gorte in the Department of Systems
Engineering and the support of my extended relatives in the Grossi and Quinn families, has not
gone unnoticed. Most importantly, the members of my immediate family, Dad, Mom, Jim,
Teresa, Matt, Julia, and Briana, as well as Mohan and Lara, are my inspiration.
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
QUANTIFYING THE UNCERTAINTY IN SEISMIC RISK AND LOSS ESTIMATION
Patricia Grossi
Howard Kunreuther
In this work, the role of uncertainty in a probabilistic earthquake loss estimation is
studied. In order to quantify the uncertainty associated with earthquake loss estimation (ELE), a
sensitivity' analysis is completed, assuming alternative estimates for different parameters and
models in the ELE process. The parametric and modeling estimates are varied one-by-one and the
effects on the calculations of average annual loss (AAL) and worst case loss (WCL) are analyzed.
These losses are generated via a loss exceedance probability (EP) curve. This study is unique in
that it uses a regional loss estimation model, HAZUS, with pre-processing and post-processing
software modules to estimate direct economic losses to homeowners and the insurance industry in
the Oakland, California region. A probabilistic seismic hazard analysis (PSHA) is mimicked
through the use of the pre-processor (Scenario Builder) and post-processor (EP Maker) in the
study. Within this sensitivity analysis, annual recurrence of earthquakes, attenuation models for
ground motion, soil mapping schemes, and exposure and vulnerability parameters for residential
structures are considered. Additionally, techniques to incorporate expert opinion on the
vulnerability of structures, the benefits of structural mitigation, and the costs of mitigation into
the probabilistic seismic risk assessment are introduced. Conclusions are three-fold. First, the
earthquake loss estimation process is very uncertain, producing estimates of direct economic loss
that are most sensitive to the ground motion attenuation in the Oakland, California region.
Second, the residential structural mitigation studied, bolting a low-rise wood frame structure to its
foundation and bracing its cripple wall, is extremely worthwhile for a homeowner to complete,
using a straightforward cost-benefit approach. Finally, the homeowner will have to cover the
vi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
majority of the loss on an average annual basis under various insurance deductible and limit level
schemes.
vii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
COPYRIGHT.............................................................. ii
DEDICATION.................................................................................................................................. iii
ACKNOWLEDGMENTS.............................................................................................................. iv
ABSTRACT____________________________________ vi
LIST OF TABLES.......................................................................................................................... xii
LIST OF ILLUSTRATIONS____________________________________________________xv
1 INTRODUCTION........................................................................................................................ 1
1.1 Overview................................................................................................................................... I
1.2 Scope and Objective.................................................................................................................. 4
1.3 Organization of Dissertation....................................................................................................13
2 UNCERTAINTY IN EARTHQUAKE LOSS ESTIMATION........................................... 15
2.1 Aleatory versus Epistemic Uncertainty.................................................................................. 15
2.2 Seismic Hazard Model.......................................................................................................... 17
2.2.1 Previous Studies on PSHA Uncertainty........................................................................ 18
2.2.2 Probabilistic Seismic Hazard Analysis......................................................................... 22
2.2.2.1 Seismic Source Determination.................................................................................. 24
2.2.2.2 Recurrence Relationship........................................................................................... 27
2.2.2.3 Estimation of Ground Motion................................................................................... 30
2.2.2.4 Probability of Exceedance.........................................................................................33
2.3 Previous Studies on Earthquake Loss Estimation................................................................. 35
2.4 Inventory Exposure Characteristics........................................................................................ 38
2.5 Damage Estimation................................................................................................................. 40
2.6 Loss Estimation........................................................................................................................43
viii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.7 Summary................................................................................................................................. 45
3 HAZUS METHODOLOGY AND SOFTWARE..................................................................46
3.1 Scenario Builder......................................................................................................................49
3.1.1 Seismic Source Determination......................................................................................50
3.1.2 Recurrence Relationship................................................................................................ 53
3.1.3 Estimation of Ground Motion........................................................................................54
3.1.4 Scenario Builder Input File............................................................................................ 57
3.2 HAZUS.................................................................................................................................... 59
3.2.1 Ground Motion Methodology........................................................................................59
3.2.1.1 Standardized Response Spectrum............................................................................ 60
3.2.1.2 Soil Amplification.....................................................................................................63
3.2.2 Inventory Exposure Characteristics.............................................................................. 66
3.2.2.1 Building Capacity Curve.......................................................................................... 68
3.2.3 Direct Physical Damage................................................................................................ 70
3.2.3.1 Building Fragility Curve.......................................................................................... 72
3.2.4 Direct Economic Losses................................................................................................ 76
3.3 EP Maker.................................................................................................................................83
4 MODEL CITY AND MITIGATION..................................................................................... 88
4.1 Demographics and Seismic Sources......................................................................................89
4.2 Residential Building Stock and Mitigation Technique........................................................92
4.3 Expert Opinion Incorporation................................................................................................ 97
4.3.1 Related Research on Vulnerability and Mitigation...................................................... 98
4.3.2 Expert Opinion Surveys............................................................................................... 101
4.3.3 Response Rate and Comments..................................................................................... 104
ix
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3.4 Mean Damage and Cost................................................................................................107
4.3.5 Distribution Fit.............................................................................................................. 115
4.3.6 Structural Fragility Curve Development..................................................................... 121
4.3.6.1 Damage State Probability Matrices........................................................................ 123
4.3.6.2 Cumulative Lognormal Distributions................................................................... 125
4.3.6.3 Conversion to Peak Ground Acceleration.............................................................. 129
4.3.6.4 Intersection of Demand Spectrum and Capacity Curve........................................130
4.3.7 Discussion.................................................................................................................... 139
5 PARAMETERS IN SENSITIVITY ANALYSIS............................................................... 145
5.1 Seismic Hazard Parameters..................................................................................................147
5.1.1 Earthquake Recurrence................................................................................................ 149
5.1.2 Ground Motion Attenuation....................................................................................... 155
5.1.3 Soil Mapping Schemes............................................................................................... 156
5.2 Inventory, Damage, and Loss Parameters............................................................................ 160
5.2.1 Inventory Exposure...................................................................................................... 161
5.2.2 Fragility Curves............................................................................................................162
6 RESULTS................................................................................................................................. 168
6.1 Range of Uncertainty............................................................................................................ 168
6.2 Expert Opinion Incorporation for Mitigation....................................................................182
6.3 Cost-Effectiveness of Mitigation......................................................................................... 187
6.4 Earthquake Insurance........................................................................................................... 192
7 CONCLUSIONS..................................................................................................................... 202
7.1 Summary...............................................................................................................................202
7.2 Future Work......................................................................................................................... 206
x
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8 Appendices................................................................................................................................. 207
8.1 Earthquake Measures............................................................................................................ 207
8.2 Attenuation Relationships.....................................................................................................210
8.3 EP Maker Code......................................................................................................................213
8.4 Scenario Builder Input File................................................................................................. 219
8.5 Benefits Questionnaire.........................................................................................................221
8.6 Costs Questionnaire............................................................................................................... 224
8.7 Benefits Data..........................................................................................................................227
8.8 Costs Data..............................................................................................................................231
8.9 Distributions...........................................................................................................................232
8.10 Recurrence Data................................................................................................................... 234
8.11 Soils Data............................................................................................................................. 235
8.12 Exposure Data...................................................................................................................... 238
8.13 Summary of Average Annual Loss.................................................................................... 241
9 Bibliography............................................................................................................................... 242
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
Table 1.1 Homeowner Losses as a Function of Mitigation and Insurance............................. 12
Table 1.2 Insurer Losses as a Function of Structural Mitigation. ____........___........___ 13
Table 3.1 Event Loss Table.............................................................................................................. 47
Table 3.2 Fault Segment and Property Information.................................................................... 51
Table 3 3 Scenario Builder Input File............___......................................... 58
Table 3.4 Regression Coefficients for Fault Rupture................................................................... 59
Table 3.5 NEHRP Site Classes......................................................................................................... 64
Table 3.6 NEHRP Soil Amplification Factors............................................................................... 65
Table 3.7 Occupancy Classes and Model Building Types............................................................67
Table 3.8 Drift Ratios and Spectral Displacement for Structural Damage.............................. 73
Table 3.9 Structural Fragility Curves.............................................................................................75
Table 3.10 Structure Occupancy Mapping.................................................................................... 76
Table 3.11 Types of Direct Economic Loss to Residential Structures.......................................77
Table 3.12 Time Independent Direct Economic Loss Parameters............................................. 80
Table 3.13 Time Dependent Direct Economic Loss Parameters................................................ 82
Table 3.14 Aggregation of Direct Economic Losses..................................................................... 84
Table 4.1 Oakland, California Region Demographics................................................................. 90
Table 4.2 Default RES1 Occupancy Scheme.................................................................................. 93
Table 4.3 Updated RES1 Occupancy Scheme................................................................................ 96
Table 4.4(a) RES1 Occupancy Scheme for Analysis (Before Mitigation)................................ 96
Table 4.4(b) RES1 Occupancy Scheme for Analysis (After Mitigation).................................. 96
Table 4.5 Mean Damage Factors.................................................................................................... 100
Table 4.6 Costs of Mitigation.......................................................................................................... 101
xii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.7 Survey Respondents' Statistics.....................................................................................104
Table 4.8 Cases for Mathematical Aggregation.......................................................................... 108
Table 4.9 Benefits Survey Statistics of MDF.............................................................................. I l l
Table 4.10 Costs Survey Statistics................................................................................................. 114
Table 4.11 Beta Distributions Parameters for Damage Data................................................... 117
Table 4.12 Best Distribution Fit Parameters for Damage Data............................................... 119
Table 4.13 Best Distribution Fit Parameters for Cost Data..................................................... 120
Table 4.14 Damage Factor State Limits.......................................................................................124
Table 4.15 Damage State Probability Matrices.......................................................................... 125
Table 4.16 Cumulative Damage State Probability Matrices.................................................... 126
Table 4.17 Statistics of Cumulative Lognormal Distributions (MMI)....................................128
Table 4.18 MMI to PGA Conversion Table................................................................................ 130
Table 4.19 Statistics of Cumulative Lognormal Distributions (PGA).................................... 130
Table 4.20 Response Spectrum Demand Parameter Values..................................................... 134
Table 4.21 Capacity Curve Parameter Values............................................................................ 135
Table 4.22 Capacity Curve Parameter Values............................................................................ 136
Table 4.23 Structural Fragility Curve Parameters.................................................................... 138
Table 5.1 Seismic Fault Source Parameters................................................................................ 150
Table 5.2 Characteristic Recurrence............................................................................................ 151
Table 53 Exponential Recurrence................................................................................................ 152
Table 5.4 Structural Fragility Curve Parameters...................................................................... 164
Table 5.5 Nonstructural Fragility Curve Parameters................................................................ 166
Table 6.1 Cases for Range of Uncertainty Calculations.............................................................170
Table 6.2 Notation for Range of Uncertainty Calculations...................................................... 171
xiii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.3 Summary of Loss Estimates for Case 2 .......... 175
Table 6.4 Summary of Losses for Case 3........................................... 178
Table 6.5 Breakdown of Average Annual Loss........................................................................ 180
Table 6.6 Best Distribution Fit Parameters for Cost Data........................................................ 183
Table 6.7 Effects of Mitigation.................................................................................................... 184
Table 6.8 Cost-Benefit Analysis of Mitigation...... .......___...___ .... ----------- 189
Table 6.9 Simulation Results........................................................................................................... 191
Table 6.10 Homeowner Loss as a Function of Mitigation and Insurance...............................192
Table 6.11 Insurer Loss as a Function of Structural Mitigation............................................... 193
Table 6.12 Parameters for Analysis............................................................................................... 194
Table 6.13 Homeowner Average Annual Loss without Insurance............................................195
Table 6.14 Homeowner Average Annual Loss with Insurance and No Mitigation......... 197
Table 6.15 Homeowner Average Annual Loss with Insurance and Mitigation.....................198
Table 6.16 Insurer Losses................................................................................................................200
Table 6.17 Percentage Reduction in Insurer Losses................................................................... 201
Table 7.1 Ranking of Parameters and Models by Range of Uncertainty............................... 203
xiv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1 INTRODUCTION
1.1 Overview
In 1989, two events occurred that had a drastic impact on both the insurance industry and
the political climate in the United States. Late on September 21, 1989, Hurricane Hugo hit the
coast of South Carolina, devastating the towns of Charleston and Myrtle Beach. Insured loss
estimates totaled $4.2 billion before the storm moved through North Carolina the next day
(Insurance Information Institute 2000). Less than a month later, on October 17, 1989, the Loma
Prieta Earthquake hit with a Richter Magnitude of 7.1 near the town of Santa Cruz, California.
Property damage was estimated at $6 billion to the surrounding Bay Area (Stover and Coffman
1993). In order to remain solvent, the insurance industry realized that it needed a better way to
estimate and manage the losses associated with such natural disasters. Moreover, the Federal
Emergency Management Agency (FEMA) recognized a need for better catastrophic loss
estimates for mitigation and emergency planning purposes.
As a result, over the course of the past decade, advanced tools have emerged that allow
insurance companies and agencies of the government to more accurately assess their catastrophic
risk exposures. These software tools utilize advances in Information Technology (IT) and
Geographic Information Systems (GIS). With the ability to store and manage vast amounts of
spatially referenced information, GIS is an ideal environment for conducting catastrophic hazard
and loss studies for large regions. While it is true that catastrophe loss studies were done for
twenty years prior to this time (eg. earthquake loss estimates in Steinbrugge 1982), this
advancement in computer technology has created an easier, more cost-effective way to perform
these studies.
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
These software tools, known as catastrophic risk models, were developed in two separate
arenas (Figure 1.1). First, private companies developed models for insurance companies to
estimate their portfolio losses and individuals to estimate their site-specific losses from either a
probabilistic seismic hazard analysis (PSHA) or a deterministic earthquake scenario. Second, the
federal government developed a regional loss estimation model (HAZUS) to estimate monetary
losses as well as other types of losses (eg. casualties and shelter requirements) from an earthquake
event.
Geographic
Information
Systems
HAZUS
Information
Technology
AIR
EQECAT
RMS
Private Companies
Catastrophic Risk Modeling
US Government
Figure 1.1 Catastrophic Risk Modeling Development
A few private companies have emerged as leaders in the field of catastrophic risk
modeling. Those familiar to the author through joint work with the Wharton School include
Applied Insurance Research, Inc. (AIR), EQECAT, and Risk Management Solutions, Inc. (RMS).
Each firm has its own software package, analyzing the economic effects from both earthquakes
and hurricanes in the United States for insurance and reinsurance companies. Furthermore, in
1997, FEMA released the first version of its own software estimating potential earthquake losses
in the United States, called HAZUS (Hazards U.S.)- Current development is underway to
estimate wind (eg. hurricane) and flood hazards using HAZUS.
2
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
While the private industrys software and the HAZUS software are intended for different
audiences, each utilizes the same general methodology to analyze catastrophic earthquake losses.
The earthquake loss estimation (ELE) methodology is comprised of four basic stages: (1) define
the earthquake hazard, (2) define the inventory characteristics, (3) estimate the inventory damage,
and (4) calculate the economic losses (Figure 1.2).
Calculate Economic Loss
(Expected Loss or WCL
to Insurer and Owner)
Define the Earthquake Hazard
(Seismic Sources, Recurrence,
Attenuation, Soils)
Define the Inventory Characteristics
(Structure Location, Value, Year Built,
Construction Class, etc.)
Estimate Inventory Damage
(Through Historical Loss Data,
Engineering Data Expert
Opinion)
Figure 1.2 Steps of Earthquake Loss Estimation
There are numerous instances in the earthquake loss estimation procedure where
uncertainty plays a role. Uncertainty is classified by leaders in the field of seismic hazard analysis
as either aleatory (i.e. randomness) or epistemic (i.e. lack of knowledge) in nature (Budnitz et al.
1997). Furthermore, uncertainty has been categorized into modeling uncertainty, due to
differences between the process being modeled and the simplified model used for analysis, and
parametric uncertainty, due to differences between actual values of parameters and estimates for
analysis. For the four-stage earthquake loss estimation process, there is uncertainty: in the
seismologicai data used and its interpretation to define the earthquake hazard, in the exposure
data and vulnerability functions used for damage estimation, and in the costs used in determining
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
losses. In general, limited scientific information, lower quality data, or limited engineering
information results in greater variability of expected losses.
Understanding the uncertainty associated with earthquake losses is very important.
Similar to other low-probability, high-consequence (LPHC) events, the problem arises when
decisions need to be made. Decision-making should not be based solely on one average annual
estimate or expected value of loss, but should consider the variability associated with the
estimate. For example, standard deviations, variances or other statistical measures of spread
that loss can be from the average are important in decision-making. How widely dispersed the
losses are in a distribution is a key ingredient to making a well-informed, correct decision. If the
decision-maker does not appreciate the complexity of the problem and chooses to ignore the
uncertainties involved, decisions can be made based only on the expected value of loss and
difficulties can result.
For example, an emergency planner in the city of Oakland, California can use the
HAZUS software to estimate the number of displaced households in the city for post-earthquake
event planning. If he bases his decision only the best or mean estimate of red tagged (i.e. unsafe)
homes, though, a number of residents could remain homeless for longer than necessary after a
severe earthquake. In addition, an insurer can use one of the private industrys software packages
to analyze his insured losses in a significant earthquake event. If he thinks in terms of the annual
expected loss only, he could stand to lose his business through insolvency.
1.2 Scope and Objective
In this work, the role of uncertainty in a probabilistic earthquake loss estimation
methodology is studied. In order to quantify the uncertainty associated with earthquake loss
estimation (ELE), a sensitivity analysis is completed, assuming alternative estimates for different
parameters and models in the ELE process. The parametric and modeling estimates are varied
4
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
one-by-one and the effects on the calculations of expected and worst case losses are analyzed.
These losses are generated via a loss exceedance probability curve. A loss exceedance probability
curve, EP(L), is a graphical representation of the probability that a certain level of loss will be
exceeded on an annual basis (Equation 1.1 and Figure 1.3).
EP(L) = P(Loss > Z,) = 1F(L) (1.1)
Probability of
Exceedance
EP(L)
AAL
probability = 1%
Loss (in Dollars) WCL
Figure 1J Loss Exceedance Probability Curve
In Equation 1.1, F(L) denotes the cumulative probability function for the loss or
P(Loss < L) and the loss exceedance probability curve follows simply as 1 - F(L). In Figure 1.3,
it is clear that as the probability of exceedance increases, expected losses will decrease.
Additionally, two other terms are graphically represented in Figure 1.3: AAL and WCL. AAL is
the expected loss on an annual basis or Average Annual Loss, and it is the area under the loss
exceedance probability curve. WCL is the Worst Case Loss, defined as the loss expected one
percent (1%) of the time on an annual basis. It is the loss toward the tail end of the curve.1
1Note that the term worst case loss, rather than probable maximum loss (PML), is used. The term PML is a
well-defined concept in the literature (Steinbrugge 1982) and this distinction is made to avoid confusion.
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The goal of this work is to ascertain the Range of Uncertainty of Loss, L, over the
parametric and modeling assumptions, X, denoted RU[L(X)]. L is generalized as a real-value
performance function defined on X. Further, X is a range of information states over which
parametric and modeling values vary. In its generalized form, the Range of Uncertainty of/, over
A'is defined in Equation 1.2.
/?[/[L(Jr)]=[MaxZ(x)| x e X ] - [Min L(x)| x e X ] (1.2)
In this work, however, the information states, x e X, are limited to two states, defined as
default and updated. The default state is one in which the HAZUS default value is used in the
analysis. The updated state is one in which a new parameter or model value is incorporated into
the analysis (See Chapter Five). Also, L is in terms of monetary losses. In its simplest form, in
which A'has two information states, {x<*x/}, Equation 1.3 is used.
RU[L(X)] = | L(x0 - L(x0) | (1.3)
In other words, RU[LPQ] is defined as the absolute difference between monetary loss
with new knowledge, x/, at the time of the decision and the monetary loss without such
knowledge, x0For each stakeholder in question, it is the absolute value of loss using updated
information minus the same losses using default data. It is important to take the absolute value
because, with certain updated knowledge, the losses will reduce. A significant change in either
direction from the loss is important to understand the uncertainty involved in the earthquake loss
estimation process. In this study, monetary loss is in terms of average annual loss (AAL) or worst
case loss (WCL). These two measures take into account different aspects of the loss EP curve: the
area under the curve (AAL) and the tail of the curve (WCL).
Additionally, the WCL estimate varies from user to user. In the case o f an insurer, it often reflects the loss
at a pre-specified level o f probability of insolvency.
6
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
This study is unique in that it will use a regional loss estimation model, HAZUS, with
pre-processing and post-processing software modules to estimate direct economic losses to
homeowners and the insurance industry in the Oakland, California region. A probabilistic seismic
hazard analysis (PSHA) is mimicked through the use of the pre-processor (Scenario Builder) and
post-processor (EP Maker) in the study. Due to the proprietary nature of the private companies
software packages, the public domain HAZUS software and methodology is utilized.
So, taking into consideration a clearly defined list of parametric and modeling
uncertainties, in this research, there are four separate questions being addressed.
1. Which parameters and models in the earthquake loss estimation process give rise to the
most uncertainty?
First and most importantly, the primary goal is to discover the sensitivity of the economic
loss to the parametric and modeling uncertainties in the earthquake loss estimation (ELE) process.
If one can rank these uncertainties by their influence on the loss, the most appropriate areas of
future research and data collection can be recommended to reduce the uncertainty in ELE. In
other words, the sensitivity analysis completed will offer insight into the magnitude of losses to
changes in the values of uncertain models and parameters. Moreover, some insights into how
various parameters and models affect each other will be gained.
The largest absolute difference or RU value will define the parametric or modeling
uncertainty most influential in the earthquake loss estimation process and will be ranked first. The
parametric and modeling uncertainties considered in the analysis include seismic hazard
estimation parameters and models (earthquake recurrence, ground motion attenuation, soil
mapping schemes) and inventory, damage, and loss estimation parameters and models (inventory
exposure of the residential building stock and fragility of certain residential structures before and
after mitigation).
7
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The Range of Uncertainty, in its barest form, considers losses to society for a static
building stock. This study, however, incorporates earthquake mitigation in the form of structural
retrofit of residential buildings and residential earthquake insurance. Therefore, the RU values can
be dependent on the stakeholders, the building fragility state, and the use of insurance. Thus,
Equation 1.3 is restated in Equation 1.4 to reflect these dependencies.
RU[L{X))k = | L{xx(/, j ) ) - L(x0 (/, J)) | (1.4)
Specifically stakeholders, k, denote either the homeowner {HO) or Insurer (/); fragility
states, /, denote no structural mitigation {0) or structural mitigation (/); and insurance use,./,
indicates no insurance is in place (0) or insurance is utilized (/). In this way, each RU value
reflects the incorporation of mitigation and insurance {eg. 22 = 4 combinations for each
stakeholder). This leads to question two.
2. How does one define a parameter in the earthquake loss estimation process so it reflects all
the information provided by experts? How does one define a parameter in a cost-benefit
analysis so it reflects all the information provided by experts?
Expert opinion incorporation is extremely important in the earthquake loss estimation
process and subsequent cost-benefit analyses of the stakeholders in the analysis. In determining
the updated parameters in the fragility {eg. damage) curves utilized in the HAZUS
methodology, the incorporation of expert opinion into the software is addressed. Specifically, the
two parameters that define a cumulative lognormal structural fragility curve in HAZUS are
estimated using the expert opinion of structural engineers. These are S D,ds <which is the median
value of spectral displacement at which building reaches threshold of structural damage stated,
and PDfc >{he standard deviation of the normal logarithm of spectral displacement of damage
state ds (See Section 3.2.3.1 for more details).
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
These parameters are defined for pre-1940 low-rise wood frame residential structures
before and after structural mitigation. Structural mitigation is comprised of bolting the iow-rise
wood frame structures to their foundation and bracing their cripple walls. The five-step process to
incorporate expert opinion on the fragility of these older wood frame homes and the benefits of
this structural mitigation technique (i.e. reduction in damage) into HAZUS is presented.
The change in the annual benefits of mitigation, AB, due to the incorporation of new
knowledge o f the fragility, x h is the difference between the absolute value of the annual benefits
of mitigation in this updated information state and those in the default state, x0. In Equation 1.5,
considering benefits to society and no insurance (j = 0), i = 0 reflects no mitigation and / = 1
reflects structural mitigation (i.e. a Boolean indicator of zero or one).
A = \(AAL(xi(0.0))-AAL(xi(l.0))l - \ (AAL(x0(0,0)) - AAL(x0(l.O)) | (1.5)
Furthermore, the expert opinion of contractors experienced in this particular residential
earthquake retrofit technique (i.e. bracing and bolting) is incorporated into the homeowner cost-
benefit analysis. Specifically, the estimated cost for the contractor to retrofit an older wood frame
home, Crotai, or the homeowner to retrofit the structure himself, CKlalenai, is presented and utilized
in the cost-benefit analysis. These costs, in terms of upfront dollar values, have sample mean
values, with a distribution surrounding the sample mean for the cost-benefit analysis (See Section
4.3.4 and 4.3.5). This leads to question three.
3. Is it beneficial for a homeowner in the Oakland, California to mitigate fo r earthquake
hazard?
In performing a sensitivity analysis on the parameters and models used in the earthquake
loss estimation process, the homeowners losses in all cases can be determined. Considering the
case where insurance is not utilized (J = 0), the RU value for the homeowner is different,
9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
depending on whether structural mitigation is in place (Equations 1.6(a) and 1.6(b)). In the
equation, / = 0 indicates no mitigation and / = 1 indicates structural mitigation. One of these
scenarios is chosen for calculation of the RU.
RU[L{X)} ho = | L(xx( 0.0)) - L(x0( 0,0)) | (1.6 (a))
RU[L(X)] h o = IL(x\(\.0)) - L(x q ( \ ,0J) | ( 1 .6 (b))
Further, the point at which it is beneficial for the homeowner to mitigate can be
established using a net present value (NPV) calculation (Equation 1.7(a)). In this formulation, the
upfront cost of mitigation, C, is compared with the annual expected benefits of mitigation, 4 7 ,
over the lifetime of the structure, T.
NPV = - C 0 + (1.7(a))
t=\ ( 1 + r)
The upfront cost of mitigation, CQ, is either Crotai or CUalenal, as previously defined. The
annual expected benefits of mitigation, BAh are simply the average annual loss (AAL) difference
between the unmitigated and mitigated losses using the updated information state, Jt/, or the
default information state, xo (Equation 1.7(b)).
BtM = [AAL(xx( 0,0)) -AAL(xx(\,0))]ho or [AAL(x0(0.0)) - AAL(x0(\,0))JHO (1.7(b))
Finally, r is the discount rate and T is the lifetime of the structure (i.e. the time horizon
ranges from / = 1 to 7). Whenever the NPV in Equation 1.7 is greater than or equal to zero, it will
be beneficial to the homeowner to mitigate.
In the HAZUS methodology, direct benefits are defined as the reduction in direct
economic losses (eg. repair and replacement costs, building content losses, relocation expenses).
While indirect benefits can also be defined (i.e. reduction in induced damage), direct benefits are
the focus of this study. Framing the analysis in this way, this work assesses the robustness of a
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
specific residential structural mitigation technique (i.e. bolting and bracing), and the cases in
which mitigation is monetarily beneficial are assessed. Ideally, all cases will be beneficial to the
homeowner for a reasonable life of the structure.
4. Considering residential earthquake insurance, what are the expected losses to the
homeowner and the residential insurer in the Oakland, California area?
Finally, in performing a sensitivity analysis on the parameters and models used in the loss
estimation process, the homeowners losses and primary insurers losses when insurance is
purchased (j = I) can be determined. The RU value for the primary insurer will be different when
the structures are mitigated (/ = I) and when they are unmitigated (/ = 0), as in the case of the
homeowner (Equations 1.8(a) and 1.8(b)).
RU[L(X)] i = | L(x\(0,l)) - L(xq(Q,\))\ (1.8(a))
RU[L(X)] i = | L(x\(\,\)) - L(x0(\.\))\ (1.8(b))
Considering various deductible and limit level schemes, the expected losses to
homeowners and the insurance industry in the region are calculated. The reduction in losses due
to mitigation is incorporated in the analysis to ascertain the interaction effects of the structural
mitigation technique and residential earthquake insurance with the uncertainty in the ELE
process.
In Table 1.1, the annual losses to the homeowner are laid out in terms of the expected
losses, AAL(x\(iJ))ffQ or AAL(xft(iJ))h q , the insurance premiums paid by the homeowner,
(1 + / y ) - AAL(x\(i,j)) or (1 + /y ) AAL(xq('i, j)~), and the cost of mitigation, a y -C0 . Expected
losses are calculated using the updated information state,x h or default information state, x0. with
mitigation (i.e. i = / ) or without mitigation (i.e. i = 0), and with insurance (i.e.j = 1) or without
insurance (i.e.j = 0).
II
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Moreover, premiums paid by the homeowner for residential earthquake insurance are
proportional to the Annual Average Loss (AAL) for the property covered and then multiplied by
a loading factor, If, to reflect the administrative costs associated with insurance marketing and
claims settlement. In this case, the loading factor,//, is taken as 0.5. In other words, premiums are
1.5 times the total AAL to the residence. Finally, the costs of mitigation,C0, are multiplied by a
factor, a/, to convert the upfront costs to annualized costs over the lifetime of the structure
T
{a f = ^ +^ ). For example, if T= 30 years and r = 10%, a/= 0.106.
(1 + r) -1
Table 1.1 Homeowner Losses as a Function of Mitigation and Insurance
Losses to
Homeowner
Insurance
(/ = /)
No Insurance
<j = 0)
Mitigation
(/ = /)
AAL{xx(\,\))ho + ( ! + / / ) AAL(Xla,l)) + af -C0
AAL{xQ(\,\))HO+ ( \ + l f ) AAL{x0(\,\)) + af C0
AAL{x\ (\,0y) ho + af Co
AAL(xq (1,0))//o + a f Cq
No Mitigation
(/ = 0)
AAL{x\ (0,1 ))HO +(! + / / ) AAL{xx(0,1))
AAL(xq0,\))ho +(! + / / ) AAL(xq (0,1))
AAL(x\ (0,0 ))f/o
AAL(xq(0,0)) f / o
Similarly, the losses to the insurer can be calculated as seen in Table 1.2. As expected,
the calculations for the insurer are not as complicated as the calculations for the homeowner. This
is primarily because the insurer is not bearing any of the cost to mitigate. Moreover, consideration
of the primary insurer losses as a function of asset level, reinsurance, and other financial
instruments is out of the scope of this work. The losses calculated here are simply the average
annual losses paid by the insurer above the deductible loss to the homeowner,
AAL(x\(i,j))f or AAZ(xo(/,y))/ (See Section 3.3).
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 1.2 Insurer Losses as a Function of Structural Mitigation
Losses to Insurer Insurance
(/ = /)
No Insurance
<j = 0)
Mitigation
(/ = /)
AAL(x i (1,1))/
AAL(x0(l,l)) i
N/A
No Mitigation
(/ = 0)
AAL(xt (0,1))/
AAL(x0(0,\))l
N/A
1.3 Organization of Dissertation
This dissertation presents a systematic approach to quantifying uncertainty in the
earthquake loss estimation (ELE) process. Chapter Two begins the dissertation with a general
discussion of the ELE process and the sources of uncertainty that arise from it, including the
seismic hazard model, database accuracy and interpretation, the fragility of structures, and loss
estimation and insurance parameters. Previous research in this area is also reviewed.
Chapter Three is an overview of the earthquake loss estimation methodology and
software used in this analysis. The methodology of the pre-processor (Scenario Builder),
HAZUS, and post-processor (EP Maker) software modules are discussed in detail. Important
assumptions in the hazard and loss estimation process are noted and the development of loss
exceedance probability (EP) curves using an Event Loss Table (ELT) approach is discussed.
Chapter Four elaborates upon the study region of Oakland, California, including the
seismic sources, residential building stock and structure type considered for mitigation.
Moreover, the process to incorporate the expert opinion of individuals proficient in residential
earthquake mitigation is introduced. The procedure for conversion of this data from mean damage
to a fragility curve for use in HAZUS is a data-intensive, five-step process. This process
combines techniques from previous studies on earthquake damage estimation, and it is given in
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
detail in this chapter. The updated structural fragility curves developed for use in HAZUS are
presented.
Chapter Five summarizes the parameters used in this sensitivity analysis. Details on the
default and updated information states of earthquake recurrence, attenuation relationships,
soil databases, building stock exposure, and fragility of structures before and after mitigation are
all clearly defined. In all, sixty-four (26= 64) HAZUS runs are completed for input into the EP
Maker and loss exceedance probability curves are generated under the different states of
information.
Chapter Six presents the results of this analysis. The impact of uncertainty, mitigation,
and deductible and limit levels on the homeowner and the residential insurer are presented. The
average annual loss (AAL) and worst case loss (WCL) are emphasized. Moreover, the ranking of
the importance of the parameters and models in this analysis is given.
Finally, Chapter Seven summarizes the conclusions and suggests future developments of
this work.
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2 UNCERTAINTY IN EARTHQUAKE LOSS ESTIMATION
The earthquake loss estimation process can be delineated in four basic stages: define the
earthquake hazard, define the inventory characteristics, estimate the inventory damage, and
calculate the economic loss. Each step gives rise to much uncertainty. Completed research into
the uncertainty in this area is broken down into two separate sections: estimating the uncertainty
in probabilistic seismic hazard analysis (PSHA) and estimating the uncertainty in earthquake loss
estimation (ELE). A distinction between the two should be emphasized. Probabilistic seismic
hazard analysis is the determination of the likelihood that a defined level o f ground motion will
be exceeded at a site in a certain time period. Earthquake loss estimation is the determination of
the losses for a region or a site in question for a certain time period using either a deterministic or
probabilistic seismic hazard analysis. Much has been published on the uncertainty in the
parameters and models used in seismic hazard analysis, but little has been published on the
uncertainty in the rest of the earthquake loss estimation process. This chapter reviews the
earthquake loss estimation process in detail and the uncertainty involved in the methodological
process. First, a general discussion of aleatory and epistemic uncertainty is given.
2.1 Aleatory versus Epistemic Uncertainty
In a report published by leaders in the field of probabilistic seismic hazard analysis
(Budnitz et al. 1997), uncertainty is defined as either aleatory or epistemic. Aleatory uncertainty
is "uncertainty that is inherent to the unpredictable nature of future events. It represents unique
details of source, path, and site response that cannot be quantified before the earthquake occurs.
Given a model, one cannot reduce the aleatory uncertainty by collection of additional
information. One may be able, however, to quantify the aleatory uncertainty better by using
additional data." Epistemic uncertainty is uncertainty that is due to incomplete knowledge and
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
data about the physics of the earthquake process. In principle, epistemic uncertainty can be
reduced by the collection of additional information. To extend these definitions to the rest of the
earthquake loss estimation process: there are unique details in the vulnerability functions used for
damage estimation that cannot be quantified before an earthquake occurs (aleatory uncertainty).
Additionally, there is incomplete knowledge about the inventory site characteristics or exposure
data used in the analysis (epistemic uncertainty).
Besides this distinction between aleatory and epistemic uncertainty, a differentiation
between modeling uncertainty and parametric uncertainty in PSHA is made in Budnitz (1997).
Modeling uncertainty represents differences between the actual physical process that generates
the strong earthquake ground motions and the simplified model used to predict ground motions.
Modeling uncertainty is estimated by comparing model predictions to actual, observed ground
motions. Parametric uncertainty represents uncertainty in the values of model parameters in
future earthquakes. Parametric uncertainty is quantified by observing the variation in parameters
inferred for several earthquakes and/or several recordings. Again, extending these definitions to
the rest of earthquake loss estimation: there are differences between the actual physical damage to
structures and the simplified model used to predict damage (modeling uncertainty), as well as
uncertainty in the values of parameters estimating repair and replacement cost from future
earthquakes (parametric uncertainty). Both modeling and parametric uncertainties contain
aleatory and epistemic uncertainty.
While the advantage of differentiating between aleatory and epistemic uncertainty in an
analysis is clear {i.e. only epistemic uncertainty can be reduced), the necessity of distinguishing
between aleatory and epistemic uncertainty is not. As Thomas Hanks and C. Allin Cornell state
(Hanks and Cornell, 1994):
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
...epistemic and aleatory uncertainties are fixed neither in space (across a range
of models existing in 1997, say) nor in time. What is aleatory uncertainty in one
model can be epistemic uncertainty in another model, at least in part. And what
appears to be aleatory uncertainty at the present time may be cast, at least in part,
into epistemic uncertainty at a later date. As a matter of practical reality, the trick
is to make sure that uncertainties are neither ignored nor double counted. The
possibilities of doing so with parametrically complex models are large.
Agreeing with this reasoning, the analysis presented herein is careful not to ignore
uncertainties. The handling of the sources of uncertainty is clearly outlined in a logic tree
framework. Moreover, the framing of the problem to incorporate updated information states is
implicitly considering epistemic uncertainty, while aleatory uncertainty is ingrained in the
software methodology itself. With this clear, the next step is to determine which parameters and
models used in earthquake loss estimation give rise to uncertainty. This begins with a look at the
seismic hazard model.
2.2 Seismic Hazard Model
The first step in the earthquake loss estimation process is the determination of the seismic
hazard. The hazard can either be defined as deterministic or probabilistic. Deterministic
earthquake hazards are defined primarily by their magnitude and epicenter location. These
hazards include both historical events and user-defined events {i.e. scenario-based events). An
historical event is a chronicled earthquake occurrence, such as the Northridge earthquake of
January 17, 1994. A user-defined event is a hypothetical event chosen by the user based on an
arbitrary choice of earthquake epicenter along a known fault or in an area source.
Alternatively, a probabilistic hazard is considered. In a probabilistic seismic hazard
analysis (PSHA), all possible seismic sources locations and geometries are determined, the
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
maximum magnitude expected from each source is estimated, and the recurrence model or
frequency of earthquake events for each source is obtained. In this study, a series of carefully
selected scenario-based events on established sources are used to mimic a full-blown PSHA (See
Chapter Three).
Whether the hazard is defined as deterministic or probabilistic, the last portion needed to
define the hazard is the attenuation relationship or ground motion expected at certain distances
from the source for various soil types. In a probabilistic analysis, a further step is to generate a
ground motion exceedance probability curve, designating the probability of exceeding a particular
ground motion level at a certain site.
There are numerous instances in the seismic hazard estimation stage where uncertainty is
prevalent. The estimation of parameters and choice of models to utilize are based on the expert
opinions of geologists and seismologists, the manipulation of historical earthquake catalogue
data, and assumptions in empirically based attenuation and recurrence relationships. Each
assumption plays a role in the overall uncertainty in the process.
2.2.1 Previous Studies on PSHA Uncertainty
In the I980s, two probabilistic seismic hazard analysis (PSHA) studies were performed,
estimating the seismic hazard for nuclear power plant sites in the central and eastern United
States (McGuire et al. 1989; Bemreuter et al. 1989). Both documents produced similar hazard
curves for the relative seismic hazard, but had drastically different estimates of absolute hazard
levels for several plant sites. Because of the discrepancy, much controversy has surrounded
estimates of seismic hazard. Since this time, the controversy has extended to determining seismic
hazard through PSHA not only for hazardous facilities in the central and eastern United States,
but for other structures across the country. Studies completed on the uncertainty encountered in a
probabilistic seismic hazard analysis have been either qualitative or quantitative in nature.
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Qualitatively, in April of 1997, a study was published addressing the debate over the
appropriate way in which to conduct a probabilistic seismic hazard analysis (Budnitz et al. 1997).
It was a comprehensive study by a group of seven individuals, known as the Senior Seismic
Hazard Analysis Committee (SSHAC), for the U.S. Nuclear Regulatory Commission, the U.S.
Department of Energy, and the Electric Power Research Institute. Within this document, the issue
of uncertainty is addressed in terms of a PSHA, and a distinction is made between aleatory (i.e.
random) and epistemic (i.e. lack of knowledge) uncertainty, as summarized in Section 2.1.
The SSHAC define the primary objectives of a well-done PSHA as: proper and full
incorporation of uncertainties, inclusion of a range of diverse technical interpretations, and
consideration of site-specific knowledge and data sets. Further, complete documentation of the
process and results, clear responsibility for the conduct of the study, and a proper peer review are
all necessary. They recommend that the inputs into a PSHA be derived either using a Technical
Integrator (TI) approach or a Technical Facilitator/Integrator (TFI) approach. These approaches
assert that one individual, a TI or TFI, is responsible for incorporating and representing the views
of the entire scientific community on the technical issue of interest. They give examples of a
number of techniques that could be utilized in aggregating expert opinions for seismic source
determination or ground motion estimates. In essence, the SSHAC develop a systematic way in
which to incorporate subjective information into a model of seismic hazard. In all aspects of this
dissertation, every effort is made to follow the guidelines of the SSHAC report. More
specifically, the author acts as the Technical Integrator in all aspects of the study.
As for quantitative analyses completed on the uncertainty in PSHA, the first notable
study was completed in the early 1980s on the uncertainties associated with the seismic hazard in
the Northern California Bay Area region (McGuire and Shedlock 1981). Five parameters were
varied, using a discrete logic tree. The logic tree included: two interpretations of the mean rupture
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
length, three equations for estimating the mean acceleration (ground motion), three estimates of
the Richter b-value (see Equation 2.4 in Section 2.2.2.2), three estimates of the maximum
magnitude, and three estimates of the expected events per year. These parameters were varied
over a fault system in the Bay Area that included nineteen faults. The authors conclude that the
coefficient of variation of the 500-year acceleration ranges from 0.2 to 0.4 in the Northern
California region. Additionally, they note that determining the largest sources of uncertainty in
seismic hazard analysis should be done. In this way, the most appropriate areas for future
research and data collection can be recommended to reduce the uncertainty associated with
seismic hazard analysis. They do not, however, indicate their estimates of the most influential
parameters in the uncertainty analysis.
More recent publications on quantifying the uncertainty in a probabilistic seismic hazard
analysis are a series of papers in the Bulletin o f the Seismological Society of America by members
of the California Department of Conservation's Division of Mines and Geology (CDMG). In
these papers, they assess the uncertainty of ground shaking in the Los Angeles, Ventura, and
Orange Counties following the Northridge earthquake (Peterson et al. 1996a; Cao et al. 1996;
Cramer et al. 1996).
In the first of these publications (Peterson et al. 1996a), representatives of the CDMG
update the earthquake source model of the Southern California Earthquake Center (SCEC) with
new seismological information and present probabilistic seismic hazard maps (10% probability of
exceedance in 50 years) at two locations, Los Angeles and Northridge. These hazard maps
represent the ground motion at the two sites, incorporating the statistical uncertainty of a number
of parameters in the analysis. Specifically, they assumed uncertainty in the fault length, fault
width, slip rate, shear modulus, and recurrence b-value, as well as the uncertainty associated with
the moment-magnitude, magnitude-rupture, attenuation and magnitude distribution relationships.
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In their analysis, they assessed the impact of each of these parameters on the final ground
motion estimate, asserting that the slip rate, moment-magnitude relationship, magnitude
distribution, magnitude-rupture length relationship, and attenuation relationship contribute most
to the overall uncertainty at the Los Angeles and Northridge sites. Furthermore, they state that, at
these two locations, earthquake magnitudes between 5.0 and 8.0 and epicenter distances of less
than 60 km contribute the most to the earthquake hazard. While this final conclusion may be an
obvious one (Le. epicenters closer to the site will be more damaging), this is an important point to
keep in mind when developing the earthquake event catalogue for our probabilistic seismic
hazard analysis.
In the second of the CDMG publications (Cao et al. 1996), the seismologists developed
estimates of background seismicity in Southern California, producing a ground motion map for
10% probability of exceedance in fifty years for the area from background events with magnitude
between 5.0 and 6.5. A distinction between random (aleatory) and modeling (epistemic)
uncertainties is made, noting that parameters used in the recurrence relationship (b-value)
incorporate random uncertainties, while the choice of lower (m0) and upper (m) magnitude
events introduces modeling uncertainties into the analysis. In the end, they present the hazard
map, again utilizing a Monte Carlo approach to estimating the uncertainty associated with the
background seismicity of Southern California.
In the final paper of this series of publications (Cramer et al. 1996), those involved in the
analysis used a logic tree approach coupled with Monte Carlo simulation techniques to perform
the uncertainty analysis of the seismic hazard in Southern California. They considered nine
separate parameters in the analysis, three having a discrete uncertainty distribution and six having
a continuous distribution. The continuous distributions are assumed to be normal and the discrete
distributions assume equal weights among all the outcomes. The parameters include fault length,
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
fault width, fault slip rate, attenuation relationship, magnitude-frequency distribution,
incorporation of blind faults, maximum magnitude, b-value, and the shear modulus o f the earths
crust. From the final analysis, the seismologists at CDMG assert that the maximum magnitude,
the choice of attenuation relationship, the magnitude-frequency distribution, and the slip rate are
the most influential in estimating the uncertainty in seismic hazard.
With an understanding of the sensitivity of certain parameters in a PSHA, an in-depth
look into the process itself is needed.
2.2.2 Probabilistic Seismic Hazard Analysis
The beginnings of Probabilistic Seismic Hazard Analysis (PSHA) are generally attributed
to C. Allin Cornell (Cornell 1968). PSHA is an analytical methodology that determines the
likelihood that a specific level of earthquake-induced ground motion will be exceeded at a given
location during a future time period (Equation 2.1).
jV mu ao
N{Z)= A-i ! f f i ( m)fi (r)P(Z > z\m,r)drdm (2.1)
'=1 mQr=0
In equation 2.1, N(z) designates the expected number of times ground motion exceeds
level Z during time period, /; At is the mean rate of occurrence of earthquakes between lower and
upper bound magnitudes being considered in the i h source; f(m) is the probability density
distribution of magnitude within source/; and f(r) is the probability density distribution of source
distance between the various locations within source / and the site for which the hazard is being
estimated. Finally, P(Z > z | m, r) is the probability that a given earthquake of magnitude m and
epicentral distance r will exceed ground motion level z.
PSHA is fundamentally different from Deterministic Seismic Hazard Analysis (DSHA)
in that it carries units of time (Hanks and Cornell 1994). And, the appropriateness of PSHA
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
versus DSHA has been debated for the last thirty years. Some seismologists believe that in
determining the mean rates of occurrence of certain magnitude events from a seismic source, the
uncertainties are so great that it is best to perform a DSHA. They believe that with estimates of
ground shaking for a certain magnitude event from a certain distance, no units o f time (i.e. annual
probability of recurrence) need to be carried in order to get a picture of the likely hazard. The
design earthquake or maximum likely earthquake should be the focus of an earthquake hazard
evaluation, especially in siting hazardous facilities (eg. nuclear power plants).
The steps involved in a probabilistic seismic hazard analysis have been classified
different ways, but one way is to describe the process in four steps. These include: seismic source
determination, recurrence relationship for sources, estimation of ground motion at a site due to
seismic sources, and probability of exceedance of a certain level of ground motion for the site
(Figure 2 after Reiter 1990).
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A t e i
Sauice
S u p L
SOURCES
S
I
i
s
j
s
d
Z
b
X
Step 2
RECURRENCE
^ ^ Uncertainly in AUetma&ian
Distance
Step 3
GROUND MOTION
Step 4
PROBABILITY OF EXCEEDANCE
Figure 2.1 Probabilistic Seismic Hazard Analysis
2.2.2.1 Seismic Source Determination
In seismic source determination, all known faults and tectonic regions are catalogued and
a seismic source model is defined. This model is a combination of faults, represented as line or
plane sources, and area sources. Their locations and geometries are identified and the seismicity
associated with each source is determined. This step usually involves the aggregation of historical
earthquake occurrence data and it is sorted according to seismic source zone. It is a rather
laborious and subjective operation. The less amount of historical accounts available, the more
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
subjective the source determination. Whenever possible, though, paleoseismic data {i.e. the
location, timing, and size of prehistoric earthquakes) is used to establish seismic sources.
The problems associated with source modeling arise from the fact that for many regions,
there are errors in epicentral locations, especially with older earthquake events. This is primarily
due to the accuracy, or rather inaccuracy, of older seismic instruments. Before the establishment
of the National Earthquake Information Center (NEIC) as a part of the United States Geological
Survey (USGS) in the early 1970s, the collection of seismological data from seismographs
located across the country and around the world was not well coordinated. Today, the NEIC
collects seismological data on a 24-hour-a-day basis, monitoring events of body wave magnitude
or mb > 2.5 in the United States (For types of Earthquake Measures, see Appendix 8.1).
As already mentioned, in 1997, the SSHAC issued a report on the use of expert opinion
and uncertainty in PSHA. In this report, as with other reports on seismicity in the continental
United States (Peterson et al. 1996b; Frankel et al. 1996; NIBS 1997), there is a distinction made
between earthquake activity in the Western United States (WUS) and the Eastern United States
(EUS). In the WUS, line sources or faults are the primary seismic source type, and in the EUS,
area sources are the principal seismic source types.
If the seismic source is a fault line or plane, as in the WUS, a number of parameters may
be used to define the seismicity of the source. These include the length, width, endpoints, and dip
angle of the fault to define its orientation and the style, slip rate, and average displacement on the
fault (for characteristic faults) to define fault movement. All of the parameters are based on
paleoseismic data, historical earthquake records, and/or expert opinion. Paleoseismic data is the
most reliable source of information, and as previously noted, it is used to define the fault source
parameters whenever possible. Additionally, data from satellite measurements of the earths
surface are now influencing choices of creep (slip) rates along established fault lines (Figure 2.2).
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
With these parameters, then, the maximum earthquake magnitude can be defined for each
fault source. The maximum magnitude is very important to the seismic hazard calculation
(Cramer et al. 1996). In general, estimating maximum magnitudes can be accomplished through
empirical relationships from analysis of historical data, slip rate data, or geologic and geodetic
data. Past research relates fault rupture parameters to maximum earthquake magnitudes. These
parameters include rupture length and area, maximum surface displacement, and average surface
displacement (eg: Bonilla etal. 1984; Wells and Coppersmith 1994).
Historical Expej t Paleoseismic
Opinion
Earthquake
Records '
Data
Geologic and
Geodetic Data
Satellite
Imaging
Area Source Fault Source
Maximum
Magnitude
Mm.X
Figure 2.2 Seismic Source Determination
The most popular measure of maximum magnitude is the maximum moment magnitude
earthquake (Mwor simply M). It is the maximum earthquake that can occur on the fault plane,
based on measurement of the seismic moment, M0The seismic moment is defined as the seismic
release on a fault, given in Equation 2.2.
M0 =\ i - ~DA (2.2)
The parameter fj, is the modulus of rigidity of the earths crust (normally taken as 3 x 10"
dyne/cm2s); D is the average displacement on the fault; and A is the rupture area (i.e. L x IV or
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Rupture Length times Width of the fault). Using Equation 2.2, the moment magnitude is
generated using Hanks and Kanamori (1979) in Equation 2.3:
M = ylogio Mq - 10.7 (2.3)
Alternatively, if the seismic source is an area source, things become more complicated.
Area sources can either be a concentrated zone with a definitive boundary, a regional source with
a fuzzy boundary, or a background zone, designating changes in the character of seismicity
between zones (Budnitz et al. 1997). Area sources were developed for use in a PSHA in the
eastern United States although area sources do exist in the WUS to a limited degree. They are
simplified representations of the tectonic history of a region that is not readily comprehended or
accepted by the experts. Subsequently, parameters used to define a seismic area source are not
easily listed, as in the case of a seismic fault source.
Area sources and their boundaries are based on historical earthquake records, recorded
changes in regional stress, and expert opinion (Figure 2.2). In determining the maximum
earthquake magnitude generated from an area source in the EUS, the historical seismicity record
is very important. This record, along with analogies to similar seismic sources, is the primary
basis for the maximum magnitude event. In contrast to a fault source, where the determination of
Mmax is more rigorous {i.e. some estimated empirical relationships do exist), the determination
of the maximum earthquake magnitude for an area source is more an art than a science
(Budnitz et al. 1997).
2.2.2.2 Recurrence Relationship
Once the seismic source model is generated, with an estimation of the maximum
magnitude expected from each source, the next step is the establishment of the recurrence of
earthquakes for each source based on the given data. Earthquake recurrence relationships estimate
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the frequency of occurrence of magnitude events up to the maximum magnitude. Various
methods for estimating recurrence relationships of earthquake events exist, and they are typically
different for fault sources and area sources.
For fault sources, recurrence relationships are generally established from historical
earthquake records and geologic data (Figure 2.3). Historical records are used to establish a
recurrence curve for smaller magnitude events, while geologic data is used to estimate the
recurrence of larger events (i.e. characteristic events). In general, the slip rate and the mean return
period are two important parameters that are used to generate a recurrence relationship. The slip
rate, in millimeters per year, is the rate at which each side of the fault plane moves relative to the
other. In this way, fault movement is modeled as a continual process and the rate at which the
fault displaces relates how quickly stress is building up in the fault.
The mean return period, in years, is the average recurrence rate or interval of earthquake
events, based on a certain magnitude event. When performing a PSHA, mean return periods need
to be generated for all magnitudes considered in the analysis (i.e. f j (m) in Equation 2.1).
Establishing a recurrence relationship is typically done using one or more of competing models:
the Gutenberg-Richter relationship, the bounded exponential model, or the characteristic
earthquake model (Youngs and Coppersmith 1985).
The most popular recurrence model is the Gutenberg-Richter magnitude-distribution
relationship. In the mid-I950s, Gutenberg and Richter discovered that the number of earthquake
events occurring in a region is often log-linearly related to the magnitude of earthquakes (i.e.
f ( m) ~ I0~bm). The unbounded form of the equation (Richter 1958) is given in Equation 2.4.
N(m) = \ 0a' bM (2.4)
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In this equation, for a given time period, N(m) is the cumulative number of earthquakes
of magnitude m or greater, where M designates moment magnitude, and a and b are constants.
The parameter a describes the severity of seismic activity; large values for a indicate a large
number of events over time. The parameter b (i.e. Richter b-value as mentioned in Section 2.2.1)
describes the relative frequency of smaller events to larger ones; large values for b imply that
small earthquakes occur much more frequently than large ones. Using historical earthquake data,
methods for determining the values for a and b include the method of least squares fit or the
maximum likelihood method (Weichert 1980).
Alternatively, the bounded, or truncated, exponential model explicitly assumes a
maximum magnitude event in its calculation. In this model, the recurrence relationship considers
an upper bound on the magnitude, m, as well as a lower bound, m0- The common assumption is
that earthquake events occur on a fault source within a specified time period at a relative
frequency, f ( m ) , which is of the form:
/ (m) - for mQ< m < mu (2.5)
One form of the model, first proposed by Cornell and Van Marke (1969), is a truncated,
shifted exponential distribution, which is renormalized (Equation 2.6). The cumulative number of
earthquakes of magnitude m or greater, N(m), in a given time period is limited between mu and
ma.
- P ( m - m 0 ) _ - P ( m u - m 0 )
N{m) = N{m0 ) -------------- - (2.6)
| _ e P \ mu ~ mo )
In contrast to the exponential recurrence relationship, the characteristic earthquake model
assumes a characteristic or same-size earthquake frequency density, usually banded at or near
the maximum earthquake magnitude, mu. Historically, this type of model is used less frequently,
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
but it is actually more appropriate to use for certain individual faults, particularly in California
(Youngs and Coppersmith 1985). Furthermore, recent work done in estimating ground shaking
probabilities uses a combination of the two recurrence models for earthquake magnitude-
frequency distributions of certain fault segments (Peterson et al. 1996b).
Historical
Earthquake
Records
Geologic
Data
Expert
Opinion
Area Source Fault Source
Recurrence
f(m)
Figure 23 Earthquake Recurrence
As in the case of determining maximum magnitude events for seismic area sources, the
recurrence relationships derived for area sources are complicated. They rely heavily on historical
earthquake data and expert opinion (Figure 2.3). With an expert reviewing these historical
records, seismicity rates can be derived for each magnitude level of the area sources. In the
SSHAC report (Budnitz et al. 1997), it is suggested that a truncated exponential magnitude-
distribution relationship is valid, and the values for a and b should be determined using the
maximum likelihood method.
2.2.2.3 Estimation of Ground Motion
Now, once the recurrence relationship is established for the various sources, a ground
motion attenuation function is used to describe the decay of seismic waves from the source with
distance for a given earthquake magnitude. Seismic waves decrease in amplitude and change
frequency content as they travel away from their source due to various geological reasons {eg.
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
soil type, existence of land basins, etc). With this geological information, the ground motion as a
function of the distance from the earthquake source and the magnitude of the event is calculated
( f i ( m)and f j ( r) in Equation 2.1). But, different ground motion measures can be utilized,
including peak ground acceleration {PGA), peak ground velocity {PGV) and peak ground
displacement (PGD), or spectral acceleration (SA), spectral velocity (.SV), and spectral
displacement (SD). Notably, SA, Sy, and Sn are simply related to each other in terms of the period,
T, or frequency, at, of a single-degree-of-freedom system (For more details, see Section 3.2.1.1
and Equations 3.7 and 3.8).
While PSHA studies can characterize ground motion at a site using a few different
measures, the most popular method is peak ground acceleration with a few response spectra
ordinates. PGA is defined as the maximum absolute magnitude of a ground acceleration time
series, while a response spectrum describes the maximum response of a building as a function of
period for a given level of damping. The SSHAC report (Budnitz et al. 1997) debates the
appropriateness of using PGA, as well as ordinates of response spectra. They note that the
response spectrum has been accepted as the standard in defining earthquake motions, citing the
Uniform Building Codes use of standardized response spectra shapes for structural design. But,
they do state the following:
It is recommended that the representation of seismic hazard as a function of
structural frequency be obtained directly through attenuation functions (or equal
formulations) that predict spectral acceleration as a function of structural
frequency, rather than using a fixed spectral shape anchored to a value of PGA
Ground motion is characterized by the magnitude of the earthquake event, the distance
from source to site, and the local site characteristics (Equation 2.7 and Figure 2.4). In this
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
equation, z is the ground motion measure (eg. PGA or SA), m is the magnitude, r designates the
distance from source to site, and s is a local soil condition parameter.
z = f ( m, r , s ) (2.7)
As in the case of determining seismic sources and their recurrence relationships, a
distinction is made between ground motion prediction in the WUS (which contains primarily fault
sources) and the EUS (which contains primarily area sources). The magnitude measures used in
historical earthquake records are different in each region. Historically, the earthquake magnitude
scale used in the WUS is the moment magnitude, M, while the common recorded earthquake
magnitude scale in the EUS is body wave magnitude, mb (For comparison of Earthquake
Measures, see Appendix 8.1).
Distance Local Soil
Magnitude (Source to Site) Conditions
Area Source Fault Source
Ground Motion
z
Figure 2.4 Ground Motion Estimation
For the distance from source to site, the distinction between the hypocentral (focal)
distance and the epicentral distance is important (Figure 2.5). The hypocentral distance is the
distance from the site to the hypocenter of the earthquake, the actual point of energy release
below the earths surface. The epicentral distance is the distance from the site to the epicenter, the
point vertically above the hypocenter. Clearly, the epicenter is the projection of the hypocenter on
the ground surface.
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For the local site conditions, the ground motion estimated may be amplified by the
properties of the underlying soil. For example, if the soil properties at a site exhibit a high
potential for liquefaction, landslide, or surface fault rupture, permanent ground deformation can
result and the structure will have more damage. Various empirical relationships are used for
ground motion attenuation curves in the EUS and the WUS and for both shallow and deep
earthquakes. Details on the WUS attenuation relationships for shallow earthquake used in this
study are given in Appendix 8.2.
epicentral
distance
epicenter
hypocentral
distance
hypocenter
Figure 2.5 Hypocentral versus Epicentral Distance
2.2.2.4 Probability of Exceedance
In the last step, the recurrence and ground motion are combined in a probabilistic model
to estimate the site hazard (Figure 2.6). From Equation 2.1, the site hazard is defined as the
expected number of times that the ground motion will exceed a certain level, Z, during time
period, t. It is based on the mean recurrence of each source ( A, in Equation 2.1) and the ground
motion (which uses /}()and /}(r) in Equation 2.1). Seismologists take all the information
gathered in the first three steps of the process and systematically develop these probabilistic
hazard curves (See Figure 2.1).
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Specifically, using weighting schemes for alternative source models, maximum
magnitude events, and ground motion attenuation, various estimates o f ground shaking are
established for each mean recurrence interval. These are statistically combined to establish the
median {i.e. 50th percentile) hazard curve. Typically, other percentile curves are established to
reflect lower and upper bounds on the site hazard, such as the 5th and 95th or 15th and 85th
percentiles.
Expert Opinion Ground Motion Recurrence
Area Source Fault Source
Exceedance
Probability
P [ Z > z ]
Figure 2.6 Ground Motion Exceedance Probability
In this step, event occurrences are generally assumed to be independent in time and
equally likely in space, so a Poisson model is used (Equation 2.8). In this equation. X(m) denotes
the rate of occurrence of events with magnitude m, n is the number of events and t is the time
period.2 While earthquake occurrences from seismic sources are generally agreed not to occur
equally in time and space, in practice this type of model is used for simplicity.
2In determining A/m) for use in PSHA, where A.{m) = or the number of magnitude m events per
unit time, it is important to designate the type of source. For a fault source, the recurrence rate is the
number of events per unit time per unit length, L. For an area source, the recurrence rate is the number of
events per unit time per unit area, A. The assumption is that the earthquakes occur with equal likelihood
over the length L or over the area A in these calculations.
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
P[n\ A(m\ (0, t )] = e ~ ^ m)t for A(m) > 0 and n = 0,1,2,... (2.8)
n!
Since debate over the use of this model is prevalent, renewal models to predict
earthquake recurrence have been developed over the past two decades or so (eg. Dieterich et al.
1990). These models are generally believed to be more accurate for event occurrences. In these
models, a conditional probability is developed for earthquake recurrence on a fault source. Based
on the build up of stress and strain in the fault plane, once a certain level of moment is achieved, a
release of energy is needed and an earthquake occurs. Furthermore, earthquakes have been
suggested as one example of a self-organized critical (SOC) phenomenon (Bak and Tang 1989).
The earth is described as being in a critical state everywhere, where the stress is approaching
failure, and the earthquakes that occur represent small disturbances on this critical state. If this
theory is accepted as true, it is in direct conflict with the Poisson model used to predict future
earthquake occurrences.
For site hazard generation, using the Poisson model, the probability of ground shaking to
exceed a certain rate is mathematically given as:
P[Z>z;(0, / )] = / - e - ' i(z)t (2.9)
A(z) designates the number of events per unit time with Z> z at the site, where z is the ground
motion measure. For small values of ?.(z), it has been shown that I ~ A(z)t (Benjamin
and Cornell, 1970).
2.3 Previous Studies on Earthquake Loss Estimation
With this thorough understanding of all the assumptions in a probabilistic seismic hazard
analysis, attention turns to the rest of the earthquake loss estimation process. In contrast with a
number of publications on the uncertainty associated with probabilistic seismic hazard analysis,
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
there is no one publication that attempts to quantify the uncertainty in the rest of the earthquake
loss estimation process in a similar systematic way. The main problem is that once the analysis is
expanded to include earthquake losses, a whole new set of issues arises. Specifically, the
techniques used to estimate losses from earthquakes have historically been less studied than
techniques to estimate seismic hazard. Additionally, depending on the stakeholders perspective,
losses are defined differently (eg. economic losses versus loss of lifeline functions).
In the remaining parts of the earthquake loss estimation process, there is uncertainty in
estimating the inventory exposure characteristics, the inventory damage, and the loss estimation
parameters. The inventory exposure is estimated using tax assessors data, insurance company
data, or the census. The damage from ground shaking during an earthquake is estimated from
historical mean damage data, expert opinion, and experimental testing. The losses or costs from
earthquake damage are estimated using historical loss data and expert opinion. For each of these
three main areas of uncertainty (inventory exposure, damage and loss), publications do exist that
discuss developing these estimates.
Specifically, there are published articles on the use of expert opinion in probabilistic risk
assessment (Chhibber and Apostolakis 1993; Clemen and Winkler 1992; Cooke 1991).
Mathematical aggregation schemes in these publications can be utilized for the development of
mean damage parameters and the subsequent sensitivity analysis (See Section 4.3.4).
Additionally, there are publications on the regional loss estimation on which HAZUS is based
(EERI 1997), and current studies into designing catastrophic risk portfolios for insurance
companies (Ermoliev 1998). These discuss uncertainty in inventory exposure and loss. But, none
of these publications quantitatively compare the uncertainty in earthquake loss estimates from
these different sources.
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
There has been much discussion recently, though, on this issue. Tom Holzer of the
United States Geological Survey (USGS) in Menlo Park, California, gave a talk at the University
of Pennsylvania in the fall of 1998, specifically talking about the uncertainty in losses generated
using a deterministic seismic hazard analysis (DSHA). He noted that the current loss models are
crude but helpful to insurance companies that want to diversify their risk. He described the
earthquake loss estimation process as a four-step process: earthquake source modeling,
earthquake shaking, damage estimation, and loss estimation. He looked at a simple, logic tree
analysis, discretely varying five parameters in estimating losses from an event on the Hayward
fault in the Northern California Bay Area. This sensitivity analysis included three estimates of
maximum magnitude (7.5, 7.3, 6.5), two attenuation relationships (Boore et al. 1994; Joyner and
Boore 1981), and two soil amplification factors (rock versus mud). Additionally, two intensity
relationships (i.e. in estimating peak ground acceleration from Modified Mercalli Intensity) and
the vulnerability of two types of structures (wood versus brick) were considered. He concluded
that the most important parameters influencing the uncertainty in loss estimation are the
earthquake sources, the soils, and the building inventory.
Furthermore, recent work done at the Wharton School on the uncertainty in earthquake
loss estimation indicates similar results. In the fall of 1999, the four-county region surrounding
Charleston, South Carolina was used to test certain assumptions in the HAZUS earthquake loss
estimation methodology for the eastern United States (EUS). A scenario-based event was chosen,
with moment magnitude (M) 7.3 in the center of the region. The susceptibility of losses to the
scenario earthquakes duration (See Section 3.1.4) and the attenuation relationship in the EUS
was tested. Also, the inventorys vulnerability was studied, assuming the entire building stock in
the region was unreinforced masonry structures (most vulnerable) and then assuming the region
was composed of wood frame structures (least vulnerable). Results reinforce the conclusion that
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
both the hazard and database assumptions are extremely influential in estimating losses from a
significant earthquake event in the Eastern United States. Loss estimates to the region varied from
$9 Billion to $18 Billion (Grossi and Windeler 2000).
2.4 Inventory Exposure Characteristics
Looking more closely at the rest of the ELE process, once the hazard is defined, the
exposure inventory is characterized. In this step, the private industrys software and the HAZUS
software differ. In the private industries models, the structures considered for analysis include
the general building stock of an area, including the residential, commercial, and industrial
buildings {i.e. insurable structures). A pre-defined construction class, such as wood frame,
reinforced concrete shear wall, or steel moment-resisting frame, typically classifies the structures.
In the HAZUS software, the general building stock is considered, but transportation
systems, utility systems, and hazardous material facilities can also be analyzed. In this way, the
HAZUS model serves as a regional loss estimation tool, analyzing the building stock of the
region and the networks serving the community. Since the private industrys software is intended
to analyze losses to portfolios of structures for an insurance company, this additional information
is not needed. In keeping with the scope of this study, which analyzes losses to residential
buildings, the information required to define the general building stock is given here. The
information needed to define networks {eg. transportation and utility systems) is quite different
and, arguably, more complicated. Therefore, networks will not be considered as part o f the study.
The first piece of information needed to characterize the inventory exposure is the
location of the various structures or properties to be assessed. Typically, a process called
geocoding is used to map locations into a latitude and longitude pair on the earths surface. Next,
the parameters of the structures affecting both building damage and loss are required (Figure 2.7).
These include such things as structure value, contents value, square footage, occupancy type,
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
construction class, year built, and the number of stories in the structure. The enormity of loss is
directly proportional to both the value and the square footage of a structure, so these parameters
are quite important in estimating losses. Also, the construction class and the year the structure
was built are used to estimate the capacity (i.e. resistance to ground motion) of the structure, so
these parameters are quite important in damage estimation.
The structures characteristics can be aggregated into a portfolio line of business (LOB),
a zip code, or census tract for damage and loss estimation. The private companies software
packages aggregate structures characteristics by LOB or zip code. HAZUS aggregates this
information by census tract level. Specifically, in HAZUS, the total square footage of the wood
frame residential building stock in a census tract is used when calculating the expected loss to this
sector. In the private companies models, the details of the location of structures and their
associated characteristics are kept throughout the loss estimation process as much as possible for
accuracy. But, once damage is calculated, loss estimates are typically grouped by a certain class
of structure (LOB) or in a certain location (zip code).
Occupancy and Structures
Construction Year Built
Structure and Class and Height Location and
Contents Local Soil
Value Conditions
General
Building Stock
Inventory
Exposure (exp)
Square Footage
or Value
Figure 2.7 Inventory Exposure
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Besides the aforementioned structural characteristics, more specific site characteristics,
such as liquefaction and landslide potential of the underlying soil, are sought to estimate damage
and loss (Figure 2.7). In general, the more detailed the information about the inventorys structure
and site characteristics, the fewer assumptions that need to be made in the analysis. This will lead,
in turn, to better {i.e. more accurate) damage and loss estimates.
Unfortunately, problems often arise in the classification of buildings and the underlying
soil structure. As previously mentioned, the inventory exposure is generally estimated using tax
assessors data, insurance company data, or the census. Within this limited information,
sometimes only an occupancy type {eg. single family dwelling) classifies a structure, rather than a
structure type {eg. wood frame). The structure type is inferred from the occupancy classification,
leading often to mismatched data. Similarly, fire ratings are often used to classify a structures
construction type, if the construction type is not known. This is a typical practice when structures
are classified using insurance company data.
Problems also arise with assumptions in the underlying soil characteristics at the different
locations. Either the type of soil underlying the site is not known or this information is not
aggregated correctly. The best information on the characteristics of the soil is obtained through
soil boring logs {i.e. a soil description of the geologic strata encountered during drilling down to a
certain depth to determine the nature and consistency of the soil). Unfortunately, this type of
detailed information is not available for most sites.
2.5 Damage Estimation
From all this information (ground motion, location and exposure of inventory), the
potential damage to the building stock considered in the ELE is calculated (Figure 2.8). The
damage or vulnerability functions for each type of structure is estimated through the use of
historical loss data, engineering data, and expert opinion of structural engineers proficient in post-
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
earthquake damage assessment. Focusing on the state of California, historical loss data includes
damage estimates from recent earthquakes, such as the Loma Prieta earthquake in 1989 or the
Northridge earthquake in 1994. Engineering data refers to experimental testing in laboratories on
the strength o f materials and construction methods, such as the current CUREe-Caltech
Woodffame Project (Seible et al. 1999). Expert opinion of damage estimates for structures in
California include data compiled and published by the Applied Technology Council (ATC-13
1985).
Structures Local Soil Location of _________
Structures Characteristics Conditions
Ground
Motion
L General
Building Stock
Inventory
Damage
Distribution
f(yJ
Figure 2.8 Damage Estimation
In the ATC-13 document, Earthquake Damage Evaluation Data for California, the
Federal Emergency Management Agency (FEMA) and the Applied Technology Council (ATC)
worked with researchers to ascertain expected damage for seventy-eight (78) different types of
structures. These structures, called Earthquake Engineering Facility Classes, included buildings,
bridges, pipelines, dams, tunnels, storage tanks, roadways and pavements, high industrial
chimneys, cranes, conveyor systems, on-shore and off-shore towers, canals, earth retaining
structures, waterfront structures, and equipment. The thirteen members of the Project Engineering
Panel (PEP), as well as 58 additional experts in earthquake engineering, were asked their opinion
on the low, best, and high estimates o f the damage factor to each of the facility classes for
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Modified Mercalli Intensity (MMI) levels VI through XII. The damage factor is defined as the
ratio of dollar loss to replacement value. Additionally, each expert evaluated his own level of
experience and his own degree of certainty associated with each damage estimate for each facility
class.
The procedure for estimating direct physical damage from ground shaking for the
engineering facility classes was completed using the Delphi Technique. The Delphi Technique is
an iterative procedure to obtain feedback from experts whose responses are anonymous (Dalkey
1969). It was developed in the 1950s by the Rand Corporation. Further, in estimating damage,
the experts were asked to base their answers on two main assumptions. First, the damage is based
on ground shaking only. The underlying soils, potential fault rupture, inundation, and fire
following do not aggravate damage. Second, the design and construction quality is standard. In
other words, the design is based on the current model building code and no specially designed
structures, such as hospitals or schools, are considered.
Beta distributions of damage were statistically compiled based on the experts responses
(Equation 2.10). In this equation,^ denotes the damage factor and a / and a2 are the parameters
of the distribution, which must be greater than zero. The beta distribution is used because it is a
very "flexible distribution, allowing for a wide range of distribution shapes. Additionally,^ is
limited to values between zero and one, which are the limits of the damage factor (i.e. no damage
= 0 and total loss = 1).
f ( y m)
0
In the current catastrophic risk models, however, more advanced tools are used to
estimate damage (See Section 3.2.3 for HAZUS methodology of damage). Further, damage is
42
yma ' - l 0- ym) a2'
(2. 10)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
broken down into direct damage or induced damage. Direct damage is structural and
nonstructural damage, as well as damage to the contents of a building, due to ground motion and
ground failure {i.e. liquefaction, landslide, and surface fault rupture). For a deterministic
earthquake scenario, the level of ground motion estimated in step one of the ELE process is used
to estimate direct damage. Typically, direct damage is increased based on the probabilities of
ground failure.
Induced damage includes such things as damage due to inundation {i.e. dam or levee
failure, tsunami or seiche) or fire following an earthquake. Hazardous material leaks or debris
generation can also be considered in the HAZUS methodology. The methods used to estimate
these occurrences are based on a few episodes of induced damage in more recent earthquake
events. With each type of direct and induced damage, a corresponding distribution of damage is
used to estimate the losses.
In the HAZUS software, probabilities of damage to a structure are calculated in two steps
(See Section 3.2 for more details). These damage estimates were developed using published
methodologies, reports on historical losses, engineering judgement, and expert opinion. In the
private companies software models, however, damage estimates are validated using proprietary
data from recent historical earthquake events, supplied to the modelers by insurance companies
claims. Therefore, the private companies models have damage curves that are likely to be
different from the HAZUS methodology.
2.6 Loss Estimation
With the hazard defined and the location and vulnerability of the structures in the
inventory characterized, the loss can now be determined. But, what does one mean by loss? There
are numerous types of loss. When estimating earthquake losses using one of the software models,
they are typically broken down into direct economic loss, indirect economic loss, and social loss.
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The private companies software estimates direct and indirect economic loss, while HAZUS
estimates all three types.
Direct economic losses are either time independent or time dependent. Time independent
losses are the costs associated with the repair and replacement to the building stock, as well as
content and inventory loss costs. These are usually calculated in terms of dollars per square foot
or a percentage of the structure or inventory value that is damaged. Time dependent losses are
those associated with the time it takes to repair a building and the loss of function during this
period (eg. business interruption). A monetary value is estimated for this loss of function.
Indirect economic losses include the losses from supply and demand disruptions
following an earthquake (eg. to the regional economy). Damages calculated in the previous step
are used to estimate these losses. Finally, social losses include casualties following an event, to
which a monetary value cannot easily be associated.
Depending on the stakeholders, different losses are applicable. For example, residential
insurers tend to focus on their portfolio loss, to which they must compensate the owners after an
earthquake (i.e. time independent direct economic loss). A commercial business owner is very
concerned about business interruption losses, defined as the lost inventory and sales due to the
disruption of business (i.e. time dependent direct economic loss). Also, government agencies are
very concerned about the expected number of casualties from a significant earthquake event and
the expected number of displaced households for which shelter must be provided (i.e. social loss).
These losses are important to the emergency response and recovery of a region after a significant
earthquake event.
In the case of analyzing a portfolio of structures for an insurance company, the
supplementary information of insurance coverage limits, deductibles, and any reinsurance in
place is required for calculating the stakeholders losses (Figure 2.9). Typically, the worst case
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
loss (WCL) and the average annual loss (AAL) are derived for the owner and insurer, given these
parameters. Integrating under the loss exceedance probability curve, as defined in Chapter One,
generates an AAL estimate. A WCL estimate is the loss expected at a set exceedance probability
level or recurrence interval (i.e. \% probability of exceedance/100 year recurrence interval or
0.4% probability of exceedance/2S0 year recurrence interval).
Induced
Physical
Damage
Estimates
Damage
Estimates
Insurance
Coverage
General
Building Stock
Inventory
Economic Losses
Social Losses
Figure 2.9 Loss Calculation
2.7 Summary
The simplification of the earthquake loss estimation process, first defined as a four-stage
process, is misleading. In defining the earthquake hazard and the inventory exposure to estimate
earthquake damage and loss, there are numerous parameters and models (and stakeholders) to
consider in an analysis. Earthquake loss estimation is a very complex, subjective process.
Depending on the parameters and models chosen for the analysis and the level at which the
damage and loss estimates are aggregated, results can be quite different.
Careful consideration of all the assumptions in an analysis is very important in order to
create valid and useful final loss estimates. Generalizations in the methodology must not be done
lightly and a thorough understanding of the sources of uncertainty in the process is a must. Only
with this knowledge can one truly gain insight into the earthquake loss estimation process and its
results.
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3 HAZUS METHODOLOGY AND SOFTWARE
In order to complete the sensitivity study of the earthquake loss estimation process, a
catastrophic risk model is needed. Due to the proprietary nature of the private companies
software models, the HAZUS model is employed. More specifically, the HAZUS model is
combined with two other software components to estimate the damage and loss to a region from a
series of significant earthquake events. The methodological framework of these three software
modules is explained in detail in this chapter. These modules include a pre-processor to HAZUS
called Scenario Builder, the applicable portions of the HAZUS software, and a post-processor to
HAZUS entitled EP Maker.
In this chapter, the softwares methodology is described as part of the four-step
earthquake loss estimation process delineated in Chapter Two (i.e. defining the earthquake
hazard, defining the inventory characteristics, estimating the inventory damage, and calculating
the economic loss). The earthquake hazard is defined in the Scenario Builder. The inventory
characteristics are defined, the inventory damage is estimated, and the economic losses are
calculated in HAZUS. Finally, the economic losses are aggregated and assigned to each of the
stakeholders in the insurance framework of the EP Maker.
The two additional software components, Scenario Builder and EP Maker, were
necessitated by the shortcomings in HAZUS. In its current form, the HAZUS software is a
scenario-based software tool. It cannot perform a full-fledged probabilistic seismic hazard
analysis (PSHA), generating an exceedance probability curve for ground shaking at a certain site.
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In HAZUS, a probabilistic hazard is one in which the probabilistic seismic hazard contour
maps developed by the United States Geological Survey (Frankel et al. 1996) are used.3
The preprocessor (Scenario Builder) and the postprocessor (EP Maker), used in
conjunction with HAZUS, mimic a probabilistic seismic hazard analysis, estimate damage to the
regions building stock, and develop loss exceedance probability curves for the stakeholders in
the analysis. The approach is event-based. In other words, a catalogue of earthquake magnitude
events and their associated recurrence (i.e. annual probabilities of occurrence) are carefully
chosen to reflect the overall seismic risk of the region. With this catalogue, an event loss table
(ELT) is generated, which summarizes the catastrophic loss information (Table 3.1).
Table 3.1 Event Loss Table
J
EventID TotalExp
($ Thousands)
L)
($ Thousands)
Pi
COVj
1 SF2 $16,439,620 $27,612 0.06862 0.1758
2 SF2 $16,439,620 $106,377 0.01220 0.2642
3 SF2 $16,439,620 $335,041 0.00217 0.3524
4 SF2 $16,439,620 $1,135,081 0.00039 0.4307
. . . . . . . . . . . . . . . . . .
N SA9 $16,439,620 $894,290 0.0025 0.3799
In Table 3.1, the ELT lists in tabular form theN scenario events (j = considered
in the analysis, with the corresponding seismic fault source (EventID) and total building exposure
value (TotalExp). For example, an EventID of SF2 indicates the South Hayward fault segment
and an EventID of SA9 indicates the event is generated from the Peninsula segment of the San
Andreas fault (See Appendix 8.10). Also, the TotalExp is the same in each case; it is the building
exposure value for the entire region.
3Note that HAZUS97 is being used here. HAZUS99 software was released during the time of this research.
Methodological improvements in HAZUS99 include more sophisticated network and fault analyses, as well
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For each event, the analysis generates losses to the building stock, Z.y, the annual
probability of this loss occurrence, pj, and the coefficient of variation of the loss, cov, (See
Section 3.3 for this calculation). The probability of the loss occurrence, is equivalent to the
annual earthquake recurrence discussed in Chapter Two. The average annual loss to the region or
AAL is simply the sum of the losses times their respective annual probabilities (Equation 3.1).
N
AAL = X L j - p j (3.1)
j = 1
Furthermore, by ordering the losses in the ELT from maximum to minimum and
computing the cumulative probabilities, Pepq>, at each loss level (eg.
PEP(j) P j ) ' 0 ~ PEP(j-1))]) loss exceedance probability (EP) curve is
developed. The WCL is simply the point on the loss EP curve at the one percent (1%) probability
of exceedance level.
WCL = Lj where PEP(j) = 0-01 (3.2)
The flow chart of the earthquake loss estimation methodology is given in Figure 3.1
(Partial NIBS 1997 Figure 2.1). Scenario Builder is used to input a series of events (i.e. while j <
N in Figure 3.1) into the HAZUS software. The methodology computes the direct physical
damage, induced physical damage and direct economic and social losses. The direct economic
losses from HAZUS are input into EP Maker for a further insurance analysis. Note in the
flowchart that induced physical damage (in the form of debris generation) and social losses (in
the form of casualties and shelter requirements) are generated within this framework. Since these
damage and loss estimates are not used as input to EP Maker, they are not elaborated upon here.
The reader is referred to NIBS (1997) for the methodological framework.
as updated databases.
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
With this overview of how the three modules work together, the next step is to take a
closer look at defining the seismic hazard in the Scenario Builder {i.e. fault sources and locations,
the recurrence relationship assumed, and the ground motion expected).
/ Scenario Builder
(EQ Event j: M, Epicenter,
Duration, Recurrence)/
Initialize
j = l
HAZUS
I Potential Earth Science Hazards I
Ground Motion Ground Failure
Direct Physical
Damage
General
Building
Stock
Induced Physical
Damage
Direct Economic/
Social Losses
Debris Casualties Economic Shelter
Yes
j < N
EP Maker
No
Client Loss
Primary Insurer Loss
Over Limit Loss
ELT &
AEP
Deductible &
Limit Levels
Figure 3.1 Flowchart of Earthquake Loss Estimation Methodology
3.1 Scenario Builder
In Scenario Builder, an earthquake catalogue of events is defined that will affect the
region in question. In defining these earthquake hazard events, the seismic source determination,
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the associated annual earthquake recurrence, and the choice of ground motion attenuation
relationship are established using this pre-processor. The seismic sources and attenuation
equations are limited to those defined in HAZUS; the annual probabilities (recurrence) of events
in the catalogue are completely at the discretion of the user. Since the focus of this study is the
Western United States (WUS), the stream of scenario-based earthquake events is established
using the fault segments and parameters defined within the HAZUS methodology. Seismic area
sources, which are used primarily in the Eastern United States (EUS), are not utilized and,
therefore, will not be discussed here.
3.1.1 Seismic Source Determination
Fault segments and their associated parameters are outlined in Table 3.2 (Partial NIBS
1997 Appendix 4A). As discussed in Chapter Two, these include the length, width, endpoints,
and dip angle of the fault to define its geometry and the style, slip rate, and average displacement
on the fault to define fault movement. An expected maximum Moment Magnitude, Mmax. and
mean return period of this maximum magnitude event, T, are estimated based on this information.
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.2 Fault Segment and Property Information
Field Entry Description
Fault Name Name typically attributed to fault plane
Length, L Length of the fault segment, in kilometers
Width, W Width of fault segment, in kilometers
( D r o p , D bottom) Minimum and Maximum Depth of fault segment, in kilometers
(LAT 1, LONG 1) Latitude and Longitude of northern-most end point of fault segment
(LAT 2, LONG 2) Latitude and Longitude of southern-most end point of fault segment
Dip Angle Angle of the fault relative to the ground, in degrees
Style Fault Segment Type/Orientation, designating fault movement
Slip Rate, S Geologically determined rate of displacement of fault plane, in mm/yr
Displacement, D
Average displacement or coseismic slip for characteristic fault rupture,
in meters
Mmax Maximum Moment Magnitude (M) of the fault segment
T Mean Return Period of Mmax, in years
First, the assumed geometry of the fault plane is defined. One of the most important
geometric parameters is the dip angle. The dip angle is the angle that the fault lies relative to the
ground surface. In HAZUS, dipping fault segments {i.e. fault segments with a dip angle* 90) are
more complex than vertical fault segments (i.e. those with dip angles = 90). The assumed
configuration is given in Figure 3.3 (NIBS 1997 Figure 4.3), with the assumed surface projection
of the fault segment clearly shown. The top and bottom of the fault segment, as well as the width
and the dip angle of the fault segment, define the top and bottom depth of the fault source.
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
km
km
Surface Projection o f
Fault Segment
Top of Fault
Segment
Hanging Wall
Dip Angle
Width
Bottom of Fault
Segment
Foot Wall
Figure 3J Geometry of Dipping Fault Segment
Next, the style of the fault plane is designated. Fault segments are defined as strike-slip,
reverse, normal, or oblique in type and either left-lateral or right-lateral in strike-slip. Using
Figure 3.3, with designations of the foot wall and the hanging wall, a normal fault is one in which
the hanging wall moves down relative to the foot wall. A reverse fault is one in which the hanging
wall moves up relative to the foot wall. A strike-slip fault is one in which the two rock planes on
either side of the fault move sideways past each other (i.e. horizontal component of slip). An
oblique fault is a combination of strike-strip and normal or reverse slip. All of the fault sources
used in this work are right lateral strike-slip faults, which are characteristic to the Northern
California Bay Area region.
Finally, the slip rate information and average displacement on the fault are assessed. Slip
rate is defined as the annual rate the fault displaces, in millimeters per year. Slip rates are ranked
as either: well constrained, if direction, amount, and timing of displacement along the fault plane
has been shown, or poorly constrained, if there is uncertainty in amount and timing of
displacement. In California, fault planes are both well and poorly constrained, depending on the
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
amount of existing geologic and paleoseismic data. The average displacement on the fault is the
slip on the fault during a characteristic earthquake, also known as the coseismic slip.
3.1.2 Recurrence Relationship
So, with the geometry and fault movement specified, the maximum magnitude event,
Mmax, and associated mean return period, T, are estimated. The maximum magnitude event is a
moment magnitude characteristic event, as defined by the USGS documentation used in the
development in HAZUS (Petersen et al 1996b). Using Equation 2.2 in Chapter Two, which
defines Seismic Moment, M q >and Wesnousky (1986), the mean return period is developed.
The mean return period is the ratio of the seismic moment of the characteristic
earthquake, Mq , to the rate that the fault accumulates moment, . In Equation 3.3, this
simply boils down to the ratio of the average displacement, D, in meters, to the slip rate, S,
converted to meters per year.
T = ^ - = iL' P ' A = (3.3)
M S p - 5 / 1 8
For other moment magnitude events considered on the fault up to this maximum
magnitude event, recurrence can be estimated using an exponential relationship (Petersen et ai.
1996b). The exponential distribution is utilized to partition the moment rate of the fault sources
into events between the minimum and maximum magnitude. The relationship between the
moment rate and the exponential recurrence is given below.
mu
= j'N (m )M 0 dm (3.4(a))
m0
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
mu
A / g = j \ O a ~ b m \ 0 d + c m d m
(3.4(b))
Mg = I0a+d [ l 0 ( c - 6 K , _ 10( c - 6) mo ]
(c - 6 ) In 10
(3.4(c))
In Equation 3.4(a), N(m) is the annual number of events of magnitude M and M q is the
seismic moment of the events. In Equation 3.4(b), N ( m ) is defined in terms of the constants a and
b, previously defined in Equation 2.4 of Chapter Two. Again, these are the incremental rate of
earthquakes with magnitude M and the slope of the exponential distribution, respectively. Also,
M q is redefined in terms of the constants c and d , estimated (from historical data) as 1.5 and
16.1, respectively, by Hanks and Kanamori (1979).
Integrating between maand mu, Equation 3.4(c) follows. This equation is used to develop
an incremental a-value, given in Equation 3.3. The incremental a-value defines the frequency of
incremental magnitude earthquake events between the lower and upper bounds. For example, for
the Rodgers Creek fault with a moment rate ( M q ) equal to 170 -10*5 ^ m , and using b = 0.9, c
yr
= 1.5, d = 16.1, ma = 6.5 and mu = 7.0, the incremental a-value is 2.37 (See Section 5.1.1 for use
of this equation).
3.1.3 Estimation of Ground Motion
The last piece of information needed in developing a catalogue of events for input into
HAZUS is the choice of ground motion attenuation. In the Western United States (WUS),
attenuation is designated in HAZUS by one of four empirical relationships. These relationships
a = log(--------------------- )
10^ ( 1 0 ( c ^) /wu _ i q ( c ~ b ) m 0 ^
(c-b)M% In(IO)
(3.5)
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
are for ground motion from shallow crustal earthquakes (i.e. earthquakes that occur on faults
within the upper 20 to 25 kilometers of continental crust). They include: the Boore, Joyner and
Fumal relationship (1993, 1994); the Sadigh, Chang, Abrahamson, Chio, and Power relationship
(1993); the Campbell and Bozorgnia relationship (1994); and the Project97 relationship, a linear
combination of the first three attenuation relationships?
First, the Boore, Joyner, and Fumal (BJF) relationship (1993, 1994) predicts peak ground
acceleration (PGA) and spectral acceleration (SA) in terms of moment magnitude, distance, and
various site conditions for strike-slip, reverse-slip, or an unspecified faulting mechanism. The
BJF relationship is limited to Moment Magnitudes (M) between 5.5 and 7.7 and distances within
100 kilometers. Due to the scarcity of data (i.e. ground motion recordings), there is much
uncertainty in estimating ground motion for large distances. Therefore, the authors recommend
that their relationship not be used for distances greater than 100 kilometers.
Next, the Sadigh, Chang, Abrahamson, Chio, and Power (SCACP) relationship (1993)
predicts PGA and 5%-damped SA for rock sites and deep soil sites from shallow crustal
earthquakes. These relationships were developed from ground motion data of historical California
earthquakes. For various distances and geologic conditions, relationships for strike-slip and
reverse-slip faults are given. The SCACP relationship is limited to moment magnitude, M,
between 4.0 and 8.0 and distances up to 100 kilometers. For magnitudes greater than 8.0, the
ground motion is predicted using M = 8.0. Furthermore, there is a transition at moment
magnitude (M) of 6.5. Different coefficients are developed in the empirical relationships for
events larger and smaller that this level of moment magnitude to account for near-field saturation
effects.
4 More information on these attenuation relationships is given in Appendix 8.2.
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The final unique attenuation relationship used in HAZUS for the western United States
(WUS) is the Campbell and Bozorgnia (CB) relationship (1994). Unlike the other two empirical
equations, it estimates seismic demand in terms of horizontal PGA only. No spectral acceleration
values were developed. The authors recommend that this relationship be limited to distances
within 60 kilometers and moment magnitudes greater than S.O.
These three, unique ground motion attenuation relationships are combined to form a
fourth choice in the HAZUS methodology: the Project97 relationship. It is an equally-weighted,
linear combination of the BJF, SCACP, and CB functions. For estimates of peak ground
acceleration, the attenuation relationship is one-third of the BJF relationship, one-third of the
SCACP relationship, and one-third of the CB relationship. For distances larger than 60
kilometers, the CB relationship is not used, and for moment magnitudes greater than 7.7, the BJF
relationship is not used.
For estimates of 5%-damped spectral acceleration at 0.3 and 1.0 seconds, the expected
ground motion is one-half of the BJF relationship and one-half of the SCACP relationship. For
moment magnitudes greater than 7.7, the BJF relationship is not utilized. A comparison ofPGA
from a Magnitude 7.0 event using these four attenuation relationships is shown (as a function of
distance from the source) in Figure 3.4. Note that the CB relationship gives the highest peak
ground acceleration values for close distances to the source; the BJF curve estimates the lowest
PGA values at close distances.
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
distance, km
1 10 100
1 . . . . --------
0.1
BJF
SCACP
CB
Project97!
M = 7.0
0.01
Figure 3.4 Ground Motion Attenuation Relationships
3.1.4 Scenario Builder Input File
With the choice of ground motion attenuation, the final piece of information is in place to
define the catalogue of events for input into HAZUS. These scenario-based earthquake events are
defined primarily by epicenter location on a seismic fault source {i.e. latitude and longitude,
depth), Moment Magnitude (M), and the associated rupture length of the fault (surface and
subsurface). Additionally, the annual recurrence of the event and the ground motion attenuation
are chosen (Table 3.3).
In Table 3.3, a few field entries in the input file require explanation: EQDURATION,
SSRLEN, and SRLEN. First, the duration of an earthquake event is designated as Short,
Moderate, or Long. The duration of the event is used to estimate the damage to the building stock
due to ground motion. Assuming that longer events will do more damage than shorter ones, the
response spectrum demand curve for ground motion is shifted to the left to reflect a shorter
duration of ground shaking {i.e. demand is less). Alternatively, it is shifted to the right to reflect a
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
longer duration of ground motion (i.e. demand is higher). (See Section 3.2.1 for further details on
response spectrum demand curve).
Table 3.3 Scenario Builder Input File
Field Entry Definition
SCENID Scenario event identification number (j = 1,2,...,N)
SCENDESC HAZUS identification of defined seismic fault source
LATITUDE Epicenter Latitude
LONGITUDE Epicenter Longitude
EPIDEPTH Epicenter depth, in kilometers
SSRLEN Subsurface rupture length, in kilometers
SRLEN Surface rupture length, in kilometers
RORN Rupture orientation, in degrees (in plane of dip angle)
FAULTTYPE Fault type or style of movement (eg. strike-slip)
MAGNITUDE Moment magnitude, M, of scenario event
EQDURATION Earthquake duration of ground shaking
ATTENUATN Attenuation equation choice, designated by number flag
PROBAB Annual recurrence of scenario event
Additionally, the HAZUS methodology computes the expected value of fault surface and
subsurface rupture length based on the empirical relationship developed by Wells and
Coppersmith (1994).
logjQ L = a + b M (3.6)
In Equation 3.6, L designates the rupture length in kilometers andM is the moment magnitude of
the earthquake. The coefficients, a and b, vary according to the type of fault segment being
considered, such as strike-slip or reverse. A listing is given in Table 3.4 (NIBS 1997 Table 4.1).
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.4 Regression Coefficients for Fault Rupture
Rupture Type Fault Type a b
Surface Strike Slip
Reverse
All
-3.55
-2.86
-3.22
0.74
0.63
0.69
Subsurface Strike Slip
Reverse
All
-2.57
-2.42
-2.44
0.62
0.58
0.59
3.2 HAZUS
With the earthquake catalogue developed, input into HAZUS is the next step. Estimation
of ground motion, defining the building stock characteristics of the region, and generating the
damage and loss from the catalogue of events are all executed using the HAZUS software. As
mentioned in the beginning of this chapter, only portions of the HAZUS software is used in this
study. Specifically, the methodology to develop the direct physical damage estimates and direct
economic losses to the residential building stock are studied. Portions not utilized include the
direct physical damage to other types of structures and systems (eg. essential and high potential
loss facilities, transportation systems, utility systems), induced physical damage (eg. inundation,
fire, hazardous material release), and indirect economic losses (eg. supply shortages in regional
economy).
3.2.1 Ground Motion Methodology
The HAZUS methodology generates damage and loss from what is referred to as
Potential Earth Science Hazards or PESH. These potential earth science hazards include ground
motion and ground failure (See Figure 3.1 Flowchart). Ground motion includes estimates of PGA
and SA defined in the previous section. Ground failure includes such things as liquefaction,
landslide, or surface fault ruptures, which amplifies the damage (and subsequent losses) to
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
structures due to permanent ground deformation. The methodology used to estimate ground
motion, one of the most important steps in the ELE process, is discussed here. Note that ground
failure, in the form of liquefaction, landslide, and surface fault rupture, is not detailed. The
amplification of damage due to these modes of failure (i.e. permanent ground deformation) is
discussed in NIBS (1997).
3.2.1.1 Standardized Response Spectrum
The HAZUS methodology characterizes ground motion using a standardized response
spectrum shape. A response spectrum is defined as a graphical relationship of the maximum
response of a single-degree-of-freedom elastic system with damping to dynamic motion or
forces (Newmark and Hall 1982). It is used to estimate the demand of the earthquake ground
motion on structures in the region. Typically, the response spectrum is a plot of the expected
ground motion acceleration versus the period of a structure.
The HAZUS approach is to plot the spectral acceleration as a function of spectral
displacement rather than period. This is the format necessary to evaluate building damage using
the methodology in the software, which is similar to the capacity spectrum method (CSM)
(Mahaney et al. 1993). In the HAZUS methodology, the intersection of the earthquake ground
motion demand (in terms of spectral displacement, 5b) and the capacity of the building (in terms
of allowable SD) is used to estimate damage. To convert spectral displacement,# (in inches), to a
period, T (in seconds), for given value of spectral acceleration, SA(in gs), Equation 3.7 is used.
From Equation 3.7, conversion of 5^ to So for a given period, T, can be calculated as in Equation
3.8.
(3.7)
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S d =9 . S S a T 2 (3.8)
A standardized response spectrum has four parts: peak ground acceleration, PGA, a
region of constant spectral acceleration, SA, a region of constant spectral velocity Sy, and a region
of constant spectral displacement, SD(Figure 3.5 from NIBS 1997 Figure 4.2).
Spectral
Acceleration
(gs)
Spectral Displacement (inches)
Constant
Constant
Constant
VD
PGA
1.0 second
0.3 second
SA(Velocity Domain) ~ 1/T
Figure 3.5 Standardized Response Spectrum Shape
Looking closely at this standard shape of the response spectnim, the region of constant
spectral acceleration is between periods zero and TAy (in seconds) and is defined by the spectral
acceleration at a 0.3-second period (a short-period). TAV is the transition from constant spectral
acceleration to constant spectral velocity {i.e. intersection of the two regions). It is established
based on the structures underlying soils. Using the NEHRP soil site classes and amplification
factors (FEMA 1997), it is the ratio of the 1.0-second period (a long period) spectral
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
acceleration, SAt, to the short-period spectral acceleration, S^s, modified for site class (Equation
3.9, Tables 3.5 and 3.6).
At periods from TAy to Tvo seconds, the region of constant velocity has spectral
acceleration proportional to ^ and it is anchored at the 1.0-second period. At 7V, there is a
transition from constant spectral velocity to constant spectral displacement. Note the usage of 0.3
and 1.0-second spectral acceleration to define/anchor the different regions of the standardized
response spectrum. These periods are the ones used to define spectral acceleration, SA, in the
attenuation relationships discussed in Section 3.1.3 (See Tables in Appendix 8.2).
The period 7>d, where constant spectral velocity transitions to constant spectral
displacement, is defined using the relationship developed Joyner and Boore (1988). In Equation
3.10, this transition period is defined as a function of the moment magnitude, M, of the event.
When M is not known, the period TyDis assumed to be ten (10) seconds.
The standard response spectrum shape is used to simplify calculations of building
response needed to estimate damage and loss in the HAZUS methodology. It represents the
building response at the 5%-damped level at a given site. In reality, the shape of the spectrum will
vary by earthquake magnitude, earthquake source-to-site distance, the underlying soils, and the
damping of the structure, which is usually greater than five percent. For instance, at yield
capacity, wood frame structures have damping values between ten and twenty percent (Newmark
(3.9)
( MS )
Tvd =10 2 (3.10)
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and Hall 1982).5So, in the HAZUS methodology, the standard response spectrum shape is shifted
downwards to reflect the actual response spectrum of demand of different structures for damage
analysis (See Section 3.2.3).
Since the HAZUS methodology is a regional loss estimation model, simplifications are
made to these variables (eg. distance, damping of structures). Specifically, structure types are
grouped by census tract and response spectra are developed for the structure types with source-to-
site distances from the earthquake epicenter to the centroid of the census tract. In other words,
distances from earthquake epicenter to each individual structure within the tract are not
calculated. Furthermore, damping of structures is assumed to be the same across similar types of
structure. As for the underlying soils, the next section discusses the amplification in the response
spectra coordinates due to local soil conditions.
3.2.1.2 Soil Amplification
In HAZUS, amplification of ground motion due to underlying soil conditions is based on
the soil site classes and soil amplification factors developed in the NEHRP (National Earthquake
Hazard Reduction Program) Provisions (FEMA 1997). Soil site classes are developed using the
average shear wave velocity of the upper thirty meters of the local site geology (Table 3.5). Site
class A is described as hard rock. Site class F has either underlying soil susceptible to potential
failure under seismic loading (eg. liquefiable soils) or the soil properties are not sufficiently
known. A site-specific evaluation is needed for these areas. Site classes B (rock) through E (soft
soils) vary between these two extremes. Ground motion will be less for structures on hard rock
and more amplified for structures on soft and liquefiable soils.
5 For this work, a damping value of 15% is used.
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the NEHRP Provisions, soil amplification factors are developed for the soil site classes
A through E at 0.3-second (short-period) and 1.0-second (long-period) spectral acceleration, SA.
Since site class F needs a site-specific evaluation, amplification factors are not developed for
these areas. Moreover, the NEHRP Provisions do not provide soil amplification factors for peak
ground acceleration (PGA). The factors used for 0.3-second spectral acceleration are utilized to
amplify PGA in the HAZUS methodology.
Table 3.5 NEHRP Site Classes
Site Class Site Class Description
Shear Wave Velocity
(m/sec)
Minimum Maximum
A Hard Rock 1500
B Rock 760 1500
C Very Dense Soil and Soft Rock 360 760
D Stiff Soils 180 360
E Soft Soils 180
F Soils Requiring Site-Specific Evaluations
In the HAZUS methodology, one of the attenuation relationships defined in Section 3.1.3
is used to estimate ground motion for soil site class B. Then, equations 3.11(a) through 3.11(c)
amplify peak ground acceleration and spectral acceleration for site classes other than site class B.
PGAj = PGA -FAi (3.11 (a))
S ASi = s AS FAi (3H (b
SAH = SAI ' Fvi (3.11(c))
In these equations, PGA, S^, and SAi designate the peak ground acceleration, 0.3-second
spectral acceleration, and 1.0-second spectral acceleration for site class B, respectively. Similarly,
PGA,, Sas and SAI, are the peak ground acceleration, 0.3-second spectral acceleration, and 1.0-
second spectral acceleration for site class i, respectively. Finally, FA, and Fy, are the 0.3-second
period and 1.0-second period spectral acceleration amplification factors for site class /,
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
respectively. These last two factors are given in Table 3.6 for soil site classes A through E. The
factors are highest for soil site class E and least for soil site class A.
Table 3.6 NEHRP Soil Amplification Factors
Site Class B
Spectral Acceleration
Site Class
A B C D E
Short-Period, Sas (g) 0.3-Second Period Amplification Factor, FA
<0.25 0.8 1.0 1.2 1.6 2.5
0.50 0.8 1.0 1.2 1.4 1.7
0.75 0.8 1.0 1.1 1.2 1.2
1.0 0.8 1.0 1.0 1.1 0.9
> 1.25 0.8 1.0 1.0 1.0 0.8
1-Second-Period, SAi (g) 1.0-Second Period Amplification Factor, Fv
<0.1 0.8 1.0 1.7 2.4 3.5
0.2 0.8 1.0 1.6 2.0 3.2
0.3 0.8 1.0 1.5 1.8 2.8
0.4 0.8 1.0 1.4 1.6 2.4
>0.5 0.8 1.0 1.3 1.5 2.0
Figure 3.6 (NIBS 1997 Figure 4.6) is an illustrative example of how the standardized
response spectrum shown in Figure 3.5 will shift upwards from site class B (rock) to site class D
(stiff soils). The shift will be even greater for soil site class E (not shown). This increase in
ground motion demand can be significant when changing from firm underlying soils to softer
soils.
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Site Class B
Site Class D
Spectral
Acceleration
(gs)
'/ I S
At
Spectral Displacement (inches)
Figure 3.6 Amplification of Ground Shaking
3.2.2 Inventory Exposure Characteristics
Once the ground motion is estimated, the inventory in the region is defined. In keeping
with the scope of this work, the information needed to define the building stock characteristics in
HAZUS is offered. In HAZUS, buildings are classified according to their occupancy class and
structure type. Structure types are referred to as model building types within the methodology.
In all, there are thirty-six (36) model building types and twenty-eight (28) occupancy classes
defined in the HAZUS methodology. Examples o f the general occupancy classes and model
building types are given in Table 3.7.
This classification scheme (of a structures occupancy and model building type) is used
to group together buildings with similar damage and loss patterns for analysis. As previously
mentioned in Chapter Two (Section 2.4), the total square footage of an occupancy class of
structures within a census tract is used to estimate the building stock exposure for loss estimation.
The modeling building types of this residential exposure are used to estimate the capacity of the
building stock to withstand earthquake damage. In other words, damage and loss is not estimated
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
per structure; it is estimated per group of structure and occupancy class square footage within a
census tract.
Table 3.7 Occupancy Classes and Model Building Types
Occupancy Classes Model Building Types
Residential Wood Frame
Commercial Steel Moment Frame
Industrial Concrete Moment Frame
Agriculture Precast Concrete Tilt-Up
Religion/Non-Profit Reinforced Masonry
Government Unreinforced Masonry
Education Mobile Homes
The typical square footage per structure unit and associated repair and replacement costs
per square foot are both based on an occupancy classification. For example, residential structures
classified as single family dwellings {i.e. RES1 in HAZUS) are 1,500 square feet on average.
Additionally, they cost approximately ninety-five dollars ($95) per square foot to replace in the
Oakland, California region, based on 1999 Means Square Foot Cost data (Means 1999).
Furthermore, model building types designate the typical building story height and other
parameters needed to develop the capacity of the building stock to withstand seismic demand. For
example, a low-rise wood frame structure (i.e. W1 in HAZUS) is typically a one- or two-story
building with an average height of fourteen (14) feet. Moreover, low-rise wood frame structures
have an elastic fundamental-mode period of 0.35 seconds when subject to ground motion. This
estimation of earthquake resistance for each model building type is developed in the form of a
capacity curve for damage estimation. Specifically, the standardized response spectrum of ground
motion demand, plotted as spectral acceleration (SA) versus spectral displacement (SD), is
compared to the building capacity curve (also plotted SA versus SD)- The intersection of these
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
curves is used to estimate damage state probabilities (See Section 3.2.3). This leads to a
discussion of the capacity curve within HAZUS.
3.2.2.1 Building Capacity Curve
In the HAZUS methodology, the building capacity curve is a plot of a buildings lateral
load resistance as a function of its lateral displacement. In order to facilitate comparison to the
standardized response spectrum, the load is converted to spectral acceleration,^, and the lateral
displacement is converted to spectral displacement, SD.
Two points control the shape of the building capacity curve: its yield capacity and its
ultimate capacity. The yield capacity represents the true lateral strength of the building (NIBS
1997) and the point at which elastic state of the structure is breached (i.e. at (Ay, Dy) in Figure
3.7). In other words, it is the point at which the structure no longer can respond and recover to its
full strength (i.e. in a plastic mode). Up until this point, the buildings strength is based on its
true elastic fundamental-mode period, Te (in seconds), and the displacement is linear. The
ultimate capacity, ((Am in Figure 3.7), represents the maximum strength of the building with
fully plastic deformation. Between the yield and ultimate capacity, the building assumes to
deform without collapse.
The building capacity curve is constructed based on a number of engineering design
parameters that affect the yield and ultimate capacity of a building. First, the design strength
coefficient, Cs, is defined as the fraction of the buildings weight. It is approximated from the
lateral-force design requirements of the current seismic building codes (e.g. Uniform Building
Code, NEHRP Provisions, etc.). at defines the fraction of the building weight effective in push
over mode. Overstrength factors /and k relate true yield strength to design strength (i.e. nominal
building strength required by current seismic codes) and ultimate strength to yield strength,
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
respectively. Finally, n is the ductility factor relating ultimate displacement to yield displacement.
Equations 3.12(a) and 3.12(b) define the yield capacity, (Ay, Dy). Equations 3.13(a) and 3.13(b)
define ultimate capacity, (Au, Du). Figure 3.7 is a graphical representation of a typical building
capacity curve, relating these two control points through the overstrength and ductility factors.
c sy
Ay =
I
(3.12(a))
Dy = 9.8Ay T
Ay AAy
(3.12(b))
(3.13(a))
Du =XpDy (3.13(b))
The yield and ultimate points for a capacity curve are established assuming the building
was designed to one of four seismic design levels: High-Code, Moderate-Code, Low-Code, or
Pre-Code. High-Code implies that the building was designed to a more stringent design code (i.e.
UBC Seismic Zone 4). Pre-Code designates buildings that were either built before the
development of model building codes or designed without using any seismic provisions.
Spectral
Acceleration
(gs)
Spectral Displacement (inches)
Figure 3.7 Building Capacity Curve
69
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the HAZUS methodology, the design strength coefficient, C and the ductility factor,
// (both defined above), are dependent on the design level but the rest of the terms (Tt, a t, y, and
A . ) are independent of it. With the yield capacity (A>, Dy) and ultimate capacity (Aw Du) both
dependent on these two terms, the change in a model building capacity curve is solely dependent
on the design strength coefficient and the ductility factor. Within the HAZUS methodology, the
Cvand // terms range from 0.017 and 2.5 (eg. for high-rise concrete moment frame structures
designed to Pre-Code) to 0.267 and 8.0 (eg. for low-rise reinforced masonry bearing wall
structures with wood or metal deck diaphragms designed to High-Code), respectively.
3.2.3 Direct Physical Damage
With the building stock characteristics (eg. square footage and occupancy classifications)
and structure capacity for each modeling building type established, the direct physical damage to
the building stock is estimated. The peak building response, or point at which a threshold level of
damage is reached, is the intersection of the building capacity curve and the standardized
response spectrum of demand. The 5%-damped response spectrum previously defined is the
reduced (i.e. shifted downwards) for higher levels of effective structural damping to estimate the
demand.
In the HAZUS methodology, this estimate of direct physical damage is described as
being in one of five ranges of damage states. These damage states include: None, Slight,
Moderate, Extensive, or Complete. Furthermore, damage states are broken down into structural
damage states and nonstructural damage states. Each structural damage state is described
qualitatively for the structural systems of the thirty-six model building types in the HAZUS
methodology (See NIBS 1997). It is noted in the methodology that estimates of collapse fractions
for Complete damage are based on judgment and limited earthquake data considering the
70
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
material and construction of different model building types. Also, it is acknowledged that there
are numerous instances where damage is not directly visible due to architectural finishing and the
damage state may be hard to estimate.
Nonstructural damage is considered independent of the structure or model building type;
it is assumed to be dependent on a buildings occupancy type. Further, nonstructural damage is
described as either drift-sensitive or acceleration-sensitive. Drift-sensitive damage is a function of
interstory drift (So) and acceleration-sensitive damage is a function of floor acceleration (SA).
Drift-sensitive components include partition wails and architectural finishes. Acceleration-
sensitive components include mechanical and electrical equipment and building contents such as
file cabinets, bookcases, and computer equipment.
Using the intersection of response spectrum demand and building capacity, building
fragility curves are developed for structural damage, nonstructural drift-sensitive damage, and
nonstructural acceleration-sensitive damage (Figure 3.8 from NIBS 1997 Figure 5.2). Curves for
these three types of damage are developed for the Slight, Moderate, Extensive, and Complete
damage states.6 So, for any given occupancy class, i, and model building type,y, there are twelve
building fragility curves given to estimate damage to the structure (i.e. four damage states, ds,
time three damage types, k, as defined in Section 3.3).
6 The None damage state implies no damage at all.
71
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Standardized Response
Spectrum (5% damping)
Spectral
Acceleration
(gs)
Capacity Curve
Demand Spectrum
(Damping > 5%)
Spectral Displacement (inches)
Figure 3.8 Intersection of Building Capacity Curve and Demand Spectrum
3.23.1 Building Fragility Curve
Building damage functions are in the form of cumulative lognormal fragility curves that
relate the probability of being in, or exceeding, a damage state of interest for a given Potential
Earth Science Hazard (PESH) demand parameter (i.e. spectral displacement, So, or spectral
acceleration, S^). Mathematically, this is given as:
P(ds\SD] = <t>[L - l n ( = ^ - ) ] (3.14)
PSds S D,ds
In Equation 3.14, SD,ds is the median value of spectral displacement at which the
building reaches the threshold of damage state, ds; P s ^ *s ^ standard deviation o f the natural
logarithm of spectral displacement of damage state, ds-, and <!>is the standard normal cumulative
distribution function. These cumulative lognormal damage functions are developed in terms of
spectral displacement to estimate structural damage and nonstructural drift-sensitive damage to
72
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the building stock. Additionally, these functions are developed in terms of spectral acceleration to
estimate nonstructural acceleration-sensitive damage.
The S D,ds term or median value of spectral displacement is developed for each damage
state based on damage-state drift ratios (Equation 3.IS). Damage-state drift ratios are based on
numerous published sources of drift/damage data (See listing in NIBS 1997). In equation 3.15,
^R'Sds *s the drift ratio at the threshold of the damage state, a2 is the fraction of the building
weight at the location of push-over mode displacement, and h is the typical roof height in inches.
For the case of low-rise wood ftame structures, a2 is 0.75 and A is 168 inches {i.e. 14 feet 12
inches per foot).
SD,ds = ^ , 5 ^ a 2 h (3.15)
To determine the median value of spectral displacement for this type of building, built to
Moderate-Code and having Slight structural damage, a drift ratio of 0.004 yields SD,ds = 0.50
inches. The values for damage state drift ratios, , and the corresponding median values of
spectral displacement, SD,ds, in inches, are given in Table 3.8 for structural damage to low-rise
wood frame structures for the four levels of seismic design.
Table 3.8 Drift Ratios and Spectral Displacement for Structural Damage
Low-Rise Wood
Frame Structures
Structural Damage Parameters
Sli|ght Moderate Extensive Complete
Seismic Design Level
sR.Sds S D.ds
(in)
sR.Sds S D.ds
(in)
sR.sds S D . d s
(in)
s*.Sds S D . ds
(in)
High-Code 0.004 0.50 0.012 1.51 0.040 5.04 0.100 12.60
Moderate-Code 0.004 0.50 0.010 1.25 0.031 3.86 0.075 9.45
Low-Code 0.004 0.50 0.010 1.25 0.031 3.86 0.075 9.45
Pre-Code 0.003 0.40 0.008 1.00 0.025 3.09 0.060 7.56
73
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The fiSrfs term in equation 3.14, or standard deviation of the natural logarithm of
spectral displacement, is developed for each damage state based on the variability of the capacity
curve, the demand spectrum, and the threshold of damage. In equation 3.16, (3Cis the lognormal
standard deviation parameter that describes the uncertainty of the capacity curve and pD is the
lognormal standard deviation parameter that describes the uncertainty of the demand spectrum.
Pm (Srfs) ' s the lognormal standard deviation parameter that describes the uncertainty o f the
median value of threshold damage. These three terms are all considered to be lognormally
distributed random variables.
P S * = ^CONV[pc ,f}D, S D,Sds ])2 +{fiM{Sd s ) ) 2 (3-16)
In this equation, note that the variability in building capacity and earthquake demand are
interrelated (i.e. in the form of building response via a convolution process), but the variability of
the median value of structural damage is assumed to be independent of the capacity and demand.
Therefore, the overall variability in P s ^ ' s ^ square-root-sum-of-the-squares (SRSS) of the
variability of building response and threshold of structural damage.
As is the case in estimating median values of spectral displacement, the standard
deviation of the natural logarithm of spectral displacement is developed for each modeling
building type. In the HAZUS methodology, pc equals 0.25 for High- Moderate- and Low-Code
buildings and 0.30 for Pre-Code buildings. Pm ^S^)equals 0.40 for all structural damage states
and model building types. Therefore, the differences in the overall variability term, P s ^ >fr
different levels of design will be determined by the variability in the response spectrum of
demand, Pn- The parameters defining structural fragility for low-rise wood frame structures are
given in Table 3.9 and graphically represented in Figure 3.9 for the Moderate-Code buildings.
74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.9 Structural Fragility Curves
Low-Rise Wood
Frame Structures
Structural Damage
Slijght Moderate Extensive Complete
Seismic Design Level
S D , d s
P S d s
S D.ds
P S d s
S D.ds
P S *
S D.ds
Ps*
High-Code 0.50 0.80 1.51 0.81 5.04 0.85 12.60 0.97
Moderate-Code 0.50 0.84 1.25 0.86 3.86 0.89 9.45 1.04
Low-Code 0.50 0.93 1.25 0.98 3.86 1.02 9.45 0.99
Pre-Code 0.40 1.01 1.00 1.05 3.15 1.07 7.56 1.06
0.8
0.6
Slight Damage
Moderate Damage
Extensive Damage
Complete Damage
0.2
Moderate Code
0 5 10 15 20 25
Spectral Displacement
Figure 3.9 Structural Fragility Curve
For the mapping of the building stock to a region, the HAZUS methodology allows a user
to designate both the design and performance levels of the model building types for damage
estimation. Design levels, or seismic zone levels to which the building is designed, are three of
the four levels previously mentioned: Low-Code, Moderate-Code, or High-Code. Performance
levels, described as the expected ductility and strength of the structure, are designated as Inferior,
Code, and Superior. Therefore, for each occupancy class of buildings, an associated model
building or structure type is mapped into one of nine bins to estimate its damage (i.e. three-by-
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
three matrix of seismic design and performance level). Table 3.10 shows this structure
occupancy mapping scheme.
Table 3.10 Structure Occupancy Mapping
Seismic Design
Level
Performance Level
Superior (S) Code(C) Inferior (I)
High (H) HS HC HI
Moderate (M) MS MC MI
Low (L) LS LC LI
Two clarifications to this mapping scheme description are needed, however. First, for
structural damage to low-rise wood frames structures (which are the focus of this study), values
of SD,ds are quite similar across the three seismic design levels for a given performance level
(See Table 3.9 and NIBS 1997). A difference in the structural fragility curve materializes through
the functions P s ^ term.
Additionally, for the general building stock, at most four (4) of the bins of this mapping
scheme are filled at any one time. By default, in the Oakland, California region, the building
stock is mapped to the High-Code (HC), Moderate-Code (MC), Low-Code (LC), and Low-
Inferior (LI) bins. This is due to the fact that the Low-Inferior (LI) Bin is the Pre-Code level of
design described earlier.
3.2.4 Direct Economic Losses
The last step in the HAZUS methodology used in this work is the calculation of direct
economic losses. In the previous step, the probabilities of a model building type being in one of
the four damage states of interest for structural and nonstructural acceleration-sensitive and drift-
sensitive damage were estimated. These probabilities are used to calculate direct economic losses.
Keeping once again with the focus of the work, only the direct economic losses to residential
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
structures are given, which are input into EP Maker for further analysis. Other types of genera!
occupancy classifications, as listed in Table 3.7, incorporate other types of direct economic
losses.
Direct economic losses for residential buildings include six components, listed in Table
3.11. For the residential building stock, the total cost of building damage is simply the sum of the
cost of structural damage (CS), non-structural acceleration-sensitive damage (CNSA), and non
structural drift-sensitive damage (CNSD). These costs are in terms of repair and replacement costs
per square foot, and they are a function of the damage state, ds, and the occupancy class, /. The
structural damage is also dependent on the model building type,y. Costs per square foot for these
types of damage are developed from the Means Square Foot Costs publication (Means 1999).
For this work, these three types of loss, CS, CNSA, and CNSD, comprise, or. average,
approximately seventy-five percent (75%) of total loss to the residential building stock. The other
types of direct economic losses, including contents loss (CCD), relocation expenses (REL), and
rental income loss (RY), comprise the other twenty-five (25%) of direct economic loss (See
Section 4.3.7).
Table 3.11 Types of Direct Economic Loss to Residential Structures
Type of Loss Description
Structural Loss (CS) Repair and replacement cost for damage to structural
components
Non-Structural Acceleration-
Sensitive Loss (CNSA)
Repair and replacement cost for damage to
acceleration-sensitive components (ceilings,
equipment, etc.)
Non-Structural Drift-Sensitive Loss
(CNSD)
Repair and replacement cost for damage to drift-
sensitive components (partitions, exterior walls, etc.)
Contents Loss (CCD) Cost for damage to contents (furniture, computers, etc.)
Relocation Loss (REL) Cost of relocation expenses during building repair
Rental Income Loss (RY) Loss of income to owner when renter is displaced
during building repair
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For direct economic loss due to structural damage (Equation 3.17), FA,j represents the
floor area of model building type j in occupancy class / and PMBTSTR&j represents the
probability of model building type j being in damage state ds. Also, RCSu,.,j is the structural
repair and replacement cost per square foot for model building type j in occupancy class / being
in damage state ds. Finally, a regional cost index multiplier, C/, is used to take into account the
increased cost of repairing and replacing a structure in the current California economy.
5 36
CS,= (Cl Z FAU PMBTSTR&.J R C S ^ j j ) (3.17)
ds=2 j=l
Note that the None damage state (i.e. ds = 1) does not contribute to loss and that all
thirty-six (36) model building types, j , are listed in the equation. In this work, only one model
building type is analyzed (J = 1 or low-rise wood frame/Wl) for one occupancy class (/ = 1 or
single family dwellings/RESl). Furthermore, for this analysis, the cost index multiplier, Cl, is
updated from the 1994 Means Square Foot Cost multiplier (that is default in the 1997 version of
HAZUS) to a 1999 Means Square Foot Cost multiplier (Means 1999). Since the modifier for
Oakland, California is not given in the publication, it is the average of the cost modifier for the
national thirty-city average and San Francisco. Also, the floor area, FAtJ, for a low-rise wood
frame residential building is taken as 1,500 square feet. See Table 3.12 for complete listing of the
direct economic loss parameters.
Loss from nonstructural damage is broken down into loss from damage to acceleration-
sensitive and drift-sensitive building components (Equation 3.18(a) and Equation 3.18(b)). Unlike
structural damage, these types of losses are dependent on the occupancy class only. In the
equations below, FA, represents the floor area of occupancy class / in square feet. PONSAlh , and
PONSDj,.! represent the probability of occupancy class / being in non-structural acceleration-
sensitive damage state ds and non-structural drift-sensitive damage state ds, respectively. RCAd,,
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and RCDjsj are the nonstructural repair and replacement cost per square foot for occupancy class i
being in acceleration-sensitive and drift-sensitive damage state ds, respectively. The regional cost
index multiplier, Cl, is the same as above. See Table 3.12 for values of RCA* , and RCD*, for
single family dwellings (i.e. RES1 in HAZUS).
5
CNSAds,i = Z c / ' FAi PONSA& j RCA&i (3.18(a))
ds=2
5
C N S D , = ]TC/- FAi PONSD& , RCD&j (3.18(b))
ds=2
Based on the breakdown given in the Means Square Foot Costs publication, thirty-five
percent (35%) of the nonstructural components in this studys buildings (i.e. RES1 in HAZUS)
are considered acceleration-sensitive and sixty-five percent (65%) of the components are drift-
sensitive. Also, the derivation of repair and replacement costs (for structural and nonstructural
damage) for this study follow the methodology described in Appendix 15C of the HAZUS
methodology (NIBS 1997), with an update from the 1994 dollars that is default in HAZUS to
1999 dollars.
Finally, the cost of damage for each structural and nonstructural damage state is taken as
a percentage of the cost to replace the structure (i.e. the Complete damage state). Slight damage is
two percent (2%), Moderate damage is ten percent (10%), and Extensive damage is fifty percent
(50%) of the Complete damage. Both the 1994 default HAZUS dollar values and 1999 dollar
values used in this study for the four damage states are listed in Table 3.12.
79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.12 Time Independent Direct Economic Loss Parameters
Parameter 1994 Values
HAZUS
1999 Values
Used in Study
Regional Cost Index Multiplier (Cl) 116.9 130.4
Square Footage (FAtJor FAi) 1,500 sq. ft. 1,500 sq. ft.
(1994 S) (1999 $)
Structural Repair and Replacement Costs (RCS^ij)
Slight Damage $0.30 $0.36
Moderate Damage $1.50 $1.80
Extensive Damage $7.50 $9.01
Complete Damage $15.00 J $18.02
Non-Structural Acceleration-Sensitive Repair and Replacement Costs (RCA^i)
Slight Damage $0.30 $0.40
Moderate Damage $1.70 $2.47
Extensive Damage $5.10 $12.34
Complete Damage $17.00 $24.68
Non-Structural Drift-Sensitive Repair and Replacement Costs (RCD&j)
Slight Damage $0.60 $0.92
Moderate Damage $3.20 $4.58
Extensive Damage $16.00 $22.92
Complete Damage $32.00 $45.84
Contents Repair and Replacement Costs (RCD4J)
Slight Damage $0.32 $0.44
Moderate Damage $1.60 $2.21
Extensive Damage $8.00 $11.07
Complete Damage $16.00 $22.14
The fourth type of direct economic loss is building contents loss (CCD), defined as the
losses associated with contents such as furniture and equipment that is not integral to the structure
(Equation 3.19(a)). In the HAZUS methodology, the contents value is a percentage of the
replacement value of the structure. For single family dwellings (i.e. RES1 in HAZUS), the
contents value is fifty percent (50%) of the buildings replacement value.
In equation 3.19(a), the regional cost index multiplier, Cl, and the contents value for
occupancy /, CV are multiplied by the percent contents damage for occupancy / in damage state
ds (CDd,,) and the replacement costs for occupancy / in damage state ds (RCdy,). The total
replacement cost for the structure, is the sum to replace the three components of the
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
structure: the structural elements, the nonstructural acceleration-sensitive elements, and the
nonstructural drift-sensitive elements (Equation 3.19(b)).
5
CCDi =CI -CVj 'LCDfc, -RC& i (3.19(a))
d s = 2
36
RCds i = Z PMBTNSA& j FAj j (RCA5 / +RCD5 ; + RCMBT5 /y ) (3.19(b))
j=l
In equation 3.19(b), the replacement costs, RCj,.,, are in terms of the probability of model
building type j being in nonstructural acceleration-sensitive damage state ds (PMBTNSA&j) and
the floor area o f model building type j in occupancy / (FAjj). Using this probability term, it is
assumed that most contents damage is a function of building accelerations. More specifically, the
percent contents damage is based on the acceleration-sensitive nonstructural damage state. Slight
damage is one percent (1 %), Moderate damage is five percent (5%), Extensive damage is twenty-
five percent (25%), and Complete damage is fifty percent (50%) of acceleration-sensitive
nonstructural damage state.
These two terms (PMBTNSAa, , and FAU) are multiplied by the sum of the repair costs per
square foot for Complete damage to the acceleration-sensitive building components (RCAjj),
drift-sensitive building components (RCD5J), and the structure itself (RCMBTs.,j). Contents loss
per square foot for each o f the four damage states of interest is listed in Table 3.12. Note their
relation to the structural and nonstructural repair and replacement costs.7
The two final types of direct economic loss are losses due to relocation expenses, REL,,
and rental income losses, RYt. Both losses are time dependent losses, dependent on four main
terms. These include the floor area of occupancy /, FA the percentage of structures that are
7Note that these costs are rounded to two significant figures (after multiplication).
81
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
owner-occupied, OO,, the rental cost for occupancy /, RENT,, in dollars per square foot per day,
and the damage state repair time, RT& (Table 3.13).
In equation 3.20, the relocation expenses are a linear combination of the expenses borne
by the owner (%OOt) and the renter ( l-%OOt) for a given occupancy class, i. DC, denotes the
disruption cost for occupancy /' in dollars per square foot and POSTR&, is the probability of
occupancy class / being in structural damage stated. In equation 3.21, the rental income losses
are simply the unpaid rent to the owner due to displacement. It is assumed that a renter will pay
full rent if the building is in the None or Slight damage state (i.e.ds = 1 or 2). The default values
in the HAZUS methodology (NIBS 1997) are used for calculation of these two time dependent
losses. Notably, a good deal of time was spent looking at the data to update these terms, but the
change in resulting loss was not substantial. Furthermore, these losses comprise approximately
eight-percent (8%) of the total loss to the residential structures.
5
REL, = FAt -[(/ - %OOi) - POSTRjs i DC' +
ds=3 (3.20)
% 0 0 , j r POSTRjs i ( DCi + RENT, + RT& )]
ds=3
5
RY, = ( / - % 0 0 , ) - FA, RENT, POSTR& , RT& (3.21)
ds=3
Table 3.13 Time Dependent Direct Economic Loss Parameters
Parameter 1994 Values Used in Study
Square Footage (FAi) 1,500 sq. ft.
Percent Owner Occupied (%OOi) 75%
Rental Cost in S/ff2per day (RENT,) $0.02 /ft2 /day
Disruption Cost in S/ft2 (DC) $0.60 /ft2
Repair Time in days (RT&)
Slight Damage 5 days
Moderate Damage 120 days
Extensive Damage 360 days
Complete Damage 720 days
82
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.3 EP Maker
These six types of direct economic losses are used for input into the EP Maker, which
performs four functions. First, it aggregates the direct economic dollar loss (by type) to different
occupancy classes of buildings in each census tract. Second, with insurance deductibles and limits
defined by the user, direct economic losses are broken down into client (eg. homeowner) loss,
primary insurer loss, and over limit loss. Third, these three types of losses can be either taken
separately or consolidated to form an event loss table (ELT), associating the scenario-based
earthquake event with an estimated loss (See Table 3.1). Finally, based on the ELT, an annual
loss exceedance probability (EP) curve is generated, representing the annual probability that the
loss will exceed a threshold level amount. In this way, the area under the EP curve is equivalent
to the average annual loss (AAL) expected for the building stock.
First, the aggregation of the types of direct economic loss to various occupancy classes is
completed. This loss for the residential building stock is a linear combination of the six terms
described in the previous section. These include structural loss (CS), nonstructural acceleration-
sensitive loss (CNSA), nonstructural drift-sensitive loss (CNSD), contents loss (CCD), relocation
cost (REL), and rental income loss (RY). From the database file generated from HAZUS
(decolosoc.dbf), these loss terms are consolidated in tabular form (See AggregateLosses code in
Appendix 8.3). An example is given in Table 3.14. Note in this table, the nonstructural
acceleration-sensitive losses and nonstructural drift-sensitive losses are combined into
nonstructural losses (i.e. CNS = CNSA + CNSD).
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.14 Aggregation of Direct Economic Losses
Types of Direct Economic Losses (S Thousands)
j
CS CNS CCD REL RY Lj (Total Loss)
1 $3,918 $14,851 $6,809 $1,627 $407 $27,612
2 $14,147 $59,880 $23,079 $7,375 $1,896 $106,377
3 $44,573 $184,871 $71,243 $27,175 $7,179 $335,041
4 $151,766 $665,706 $196,774 $94,875 $25,960 $1,135,081
. . . . . . . . . . . . . . . . . . . . .
46 $125,695 $520,356 $150,743 $76,325 $21,171 $894,290
Then, these direct economic losses are distributed to the insurer and the homeowner
through user-input deductible and coverage limit levels. Deductible levels, %d, and coverage
limit levels, %/, are in terms of a percentage of the total building replacement value per
occupancy class within each census tract. Redefining the total replacement value of the structure
as V, it is the sum of the dollar loss at each loss type for complete damage (i.e.
V = RCA5 i + RCD5i + RCMBT5 i j ).
In equations 3.22(a) and 3.22(b), losses to the homeowner, L/HO), and the primary
insurer, L/l), are distributed based on the terms, %d V and %/ V , which are the percentage of
the loss that must be covered by the homeowner before primary insurance takes effect and the
limit of insurance, respectively. Note that the over limit loss (i.e. loss above the coverage limit) is
added to the homeowner loss in equation 3.22(a) (i.e. L j - % I V).
Lj if Lj < %d V
Lj(HO) = %d V if %d -V <Lj < %/ V (3.22(a))
%d V + (Lj - %/ V) if Lj > %l V
0 if L j < %d V
Lj (I) = ( L j - %d V) if %d V < L j < %/ - V (3.22(b))
(%l -%d)-V if Lj > %/ V
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The third function of EP Maker is the event loss table generation. As explained in the
introduction to this chapter, the event loss table (ELT) generated in EP Maker is simply a listing
of all the events that affect the region, the total regions exposure value, and the associated losses,
Lj, probabilities, Pj, and coefficients of variation on the losses, coy,. A partial example was given
in Table 3.1. In EP Maker, two separate macros are used to generate the ELT (See
GetLossesProbsforELT and CalculateCOV in Appendix 8.3). While the consolidation of losses
and probabilities is rather straightforward {i.e. just pull this information from the database), the
calculation of cov is more complicated.
In order to calculate the coefficient of variation for each scenario earthquake loss
estimate, the total mean loss for the scenario event and associated standard deviation are required.
Specifically, the structural loss, nonstructural acceleration-sensitive loss, and the nonstructural
drift-sensitive loss are loss types, k, considered below for an estimation of total mean loss for
each event. In equation 3.23(a), the mean loss to any one of these loss types, fj(k), is a function of
the dollar loss for loss type k in damage state ds, denoted RC(k,ds), and the probability that loss
type k is in damage state ds, denoted P(k,ds). Note the redefinition of terms here from the
HAZUS methodology. The standard deviation, cr(k), for each type of loss follows, based on
sample estimation (Equation 3.23(b)).
Then, with the total replacement value of the structure, V, as previously defined, the mean
damage ratio, //,, is the ratio of the sum of the mean losses for each loss type, k, over the
//(*)= ' RC(k,ds) P(k,ds)
ds2
(3.23(a))
(3.23(b))
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
replacement value, V (Equation 3.24(a)). Similarly, the overall standard deviation o f the mean
damage ratio, oj, is the ratio of the sum of the standard deviations for each loss type, k, over
replacement value (Equation 3.24(b)). And, the coefficient of variation,covj, is simply calculated
So, with this coefficient of variation term defined for each scenario event, cov,, the final
function of EP Maker can be performed. This entails developing an annual exceedance
probability curve. Based on the event loss table (eg. Table 3.1), a step exceedance probability
(EP) curve is generated (from cumulative probabilities), representing the annual probability that
the loss will exceed a threshold level amount. From this exceedance probability curve, denoted
EP(Step) in Figure 3.10, the average annual loss (AAL) and worst case loss (WCL) can be
estimated (See Equations 3.1 and 3.2).
In most instances, however, the tail end of the exceedance probability curve, which
contains information about the WCL, only has a few points. A smooth annual exceedance
probability curve is needed to get a better understanding and more accurate value of the worst
case loss at the one-percent probability of exceedance. Denoted EP(Smooth) in Figure 3.10, this
annual EP curve interpolates between the points of the EP(Step) curve and extrapolates beyond
the highest scenario-based loss. In order to generate this smooth EP curve, uncertainty is assumed
fj.;
as the proportion of the mean damage ratio over the standard deviation (cov, = ).
3
2>oo
(3.24(a))
(3.24(b))
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
in the loss estimates, Lp and the probability estimates, pj, used to develop the loss exceedance
probability curve. The overall variability in the average annuai loss is a function of the mean
probability and loss at each point and the standard deviation of these estimates (Equation 3.26)?
Var(AAL ; ) = n 2 . a 2 + p 2 cr2 + a 2 a 2 (3.26)
Pj J J Pj Pj J
0.4
i s 0 3
EP(Smooth)
EP(Step) e
t 0-2
W
a
o
Qt
S 0.1
Ead
SO $6,000,000 $4,000,000
Loss
Figure 3.10 Exceedance Probability Curve
8More information on these calculations is not given here due to the proprietary nature of the methodology.
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4 MODEL CITY AND MITIGATION
With a thorough understanding of the earthquake loss estimation (ELE) methodology and
the three software modules being used in this analysis, attention now turns to the region specific
analysis. Two important aspects of the region specific analysis are covered in this chapter. The
first is the Oakland, California case study region. More specifically, the first part o f this chapter is
an overview of the demographics, surrounding fault sources, and residential building stock of the
Oakland, California region. The structures in the residential building stock are described as either
those eligible for mitigation or those not eligible for mitigation. Eligible structures are low-rise
wood frame structures built before 1940.
The second part of this chapter is an overview of the expert opinion analysis used in this
study. In order to incorporate mitigation in the study and perform a sensitivity analysis of the
HAZUS fragility curves, the expert opinions of two groups proficient in residential earthquake
mitigation were gathered. With a residential structural mitigation technique of bolting a structure
to its foundation and bracing its cripple wall, first, structural engineers were surveyed to
determine the reduction in damage due to this technique. Second, contractors with experience
installing bolts and bracing cripple walls were questioned about the cost to do so.
From the information gathered from the structural engineers, new structural fragility
curves for low-rise wood frame structures are developed for use in HAZUS. From the
information gathered from the contractors, new estimates of the costs of this mitigation measure
are calculated for the homeowner cost-benefit analysis. In this way, the eligible structures in the
case study region are subject to structural fragility curves before mitigation and after
mitigation within the HAZUS software. Then, direct economic losses generated from HAZUS
88
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for each scenario-based earthquake event are combined with the costs of mitigation to perform
subsequent homeowner cost-benefit analyses.
The procedure to develop these structural fragility curves before mitigation and after
mitigation is a rather subjective, data-intensive process, however. Therefore, the series of five (5)
steps required to develop these curves are explained in detail in this chapter. In the end, the
fragility curves reflecting damage before mitigation and after mitigation are presented and
compared to the HAZUS default fragility curves. But, first, the Oakland, California case study
region is discussed.
4.1 Demographics and Seismic Sources
Due to the uniqueness of seismic hazard from region to region on the earth's surface, a
case study region must be chosen for this hazard and loss estimation sensitivity analysis. So, for
this work, the Oakland, California region was selected. The choice was clear for a number of
reasons. First, previous research was completed in this area on the effects of insurance,
mitigation, and uncertainty in earthquake loss estimation (Grossi et al. 1999). Therefore, the
author is intimately familiar with the residential building stock.
Second, there is concurrent research being done as part of the Federal Emergency
Management Agencys Project Impact. Entitled the Bay Area HAZUS Risk Assessment
Capabilities Project, this work encourages the use of HAZUS by local Bay Area governments to
assess their earthquake risk (See http://www.hazus.org). Finally, new United States Geological
Survey (USGS) studies indicated that there is an increased seismic potential on the Hayward fault
(Lienkaemper and Galehouse 1998). According to the latest report by the Working Group on
California Earthquake Probabilities (Schwartz et al. 1999), there is a thirty-two percent (32%)
probability that there will be a Magnitude 6.7 or greater in the next thirty (30) years on the
Hayward-Rodgers Creek fault segments.
89
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The study region is comprised of one hundred and eight (108) census tracts, including the
city of Piedmont, California, situated in the center of the region (Figure 4.1). The region is
bounded to the west by the San Francisco Bay and extends eastward from there. Additionally, the
Hayward fault bisects the eastern part of the region. The historical record of earthquake events
includes a moment magnitude (M) 6.8 quake on October 21, 1868 on the Hayward fault. The
estimated loss was $350,000 in 1868 dollars with thirty (30) deaths. If a similar magnitude
earthquake occurred today on the Hayward fault, the losses would be catastrophic. For example,
current loss estimates to residential structures (for structural damage only) are approximately $10
Billion for a moment magnitude (M) 7.0 on the Hayward fault (EERI 1996).
The region size of 108 census tracts was chosen keeping in mind that it must big enough
to be considered a regional analysis yet small enough for the software to handle the data. From
the demographic data included in the HAZUS software, the residential population is
approximately 393,000 individuals, with a total of 153,000 households.9 Moreover, the
population is compromised of 89,000 people under the age of sixteen, 256,000 people between
the ages of sixteen and sixty-five, and 48,000 people over the age of sixty-five (Table 4.1).
Table 4.1 Oakland, California Region Demographics
Population Total
Population Age < 16 89,000
Population Age between 16 and 65 256,000
Population Age > 65 48,000
9This is based on the 1990 census data.
90
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.1 Oakland, California Region
In this study, seismic fault sources include seven right-lateral strike-slip faults in the
Northern California region. These include the North Hayward, South Hayward, San Andreas
Peninsula, Rodgers Creek, Calaveras (north of Calaveras Reservoir), Concord-Green Valley, and
Greenville faults (Figure 4.2). There are a total of forty-six (46) events considered on these fault
sources between moment magnitude (M) of 5.0 and 7.5 at a depth of 5 kilometers (See Scenario
Builder Input File in Appendix 8.4).
Moreover, the San Andreas, Hayward, and Rodgers Creek faults are designated as class
A faults. Class A faults are those with slip rates greater than five (5) millimeters per year and
possessing well-constrained paleoseismic data (Peterson et al 1996b). On the other hand, the
91
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Calaveras, Concord-Green Valley, and Greenville faults are class B faults, which lack the data
necessary to constrain recurrence intervals of large (i.e. characteristic) events.10
Concard-Greei
Valley
North
hiyward Oakland
California
Region
North
Calaveras
an eas
'emnsOJa
SouthN
[ayward
1 in = 25.3 km
Figure 4.2 Seismic Sources
4.2 Residential Building Stock and Mitigation Technique
As previously mentioned, in the HAZUS software, a regions building stock fragility is
mapped into one of nine bins describing its design and performance level (See Table 3.10). The
capacity and fragility curves described in Chapter Three for High-Code, Moderate-Code, and
Low-Code buildings are based on model building codes. Pre-Code buildings are those built
10 Designation of class A and B faults is important in earthquake recurrence development (See Section
5.1.1).
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
before building codes were enforced or even in existence (LI or Low-Inferior in Table 3.10). In
HAZUS, there are default occupancy mapping schemes that designate the percentage of
structures in each occupancy class that are built to each of these design levels. These building
stock schemes are based on the ATC-13 document (1985).
Specifically, the HAZUS default data for the Oakland, California region (Table 4.2)
estimates that eighty-three percent (83%) of single family dwellings (i.e. RES1 in HAZUS) are
low-rise wood frame structures {i.e. W1 in HAZUS). Furthermore, the majority of these wood
frame structures are built to Moderate-Code requirements {i.e. buildings built between 1940 and
1973).
Table 4.2 Default RESI Occupancy Scheme
Seismic Design and
Performance Level
DEFAULT SCHEME"
Total {% and
1000s sq. ft.)
W1 (% and
1000s sq. ft.)
S3
(%)
S4L
(%)
RM1L
(%)
High-Code (HC) 25% 20,630 20% 16,504 2% 0% 3%
Moderate-Code (MC) 72% 59,416 63% 51,989 2% 3% 4%
Low-Inferior (LI) 3% 2,476 0% 0 1% 0% 2%
Total 100% 82,522 83% 68,493 5% 3% 9%
'W1 = wood light frame; S3 = steel light frame; S4L = low-rise steel frame with cast-in-place
concrete shear walls; and RM1L = low-rise reinforced masonry bearing walls with wood or metal
deck diaphragms
An update to this information is necessary to accurately estimate the total number of
wood frame residential structures in the study region and the number eligible for mitigation. The
structure type chosen for mitigation was predetermined to be pre-1940 wood-fiame single family
residences.11 These homes ordinarily have cripple walls {i.e. the walls between the top of the
foundation and the first floor diaphragm).
11 Earlier analyses on earthquake mitigation for the Managing Catastrophic Risks project at the Wharton
School used this type o f structure (Grossi et al. 1999).
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Typically, there are two structural deficiencies of these homes. First, there is usually a
lack of connection between the wood frame bearing walls and the foundation. Second, there is a
lack of shear bracing at the cripple wall level. Therefore, the structural mitigation for pre-1940
homes in California requires bolting the structure to its foundation and bracing its cripple wall
(Figure 4.3).
Framkia Anchors
2 x 4 blocking
for p a n e l
nailing
Figure 4 J Cripple Wall Configuration and Mitigation
In California, there is evidence of local governments encouraging this type of structural
mitigation (EERJ 2000). Specifically, the city of San Leandro has set priorities to retrofit older
wood-frame homes. The Home Earthquake Strengthening Program is a comprehensive,
residential seismic strengthening program that provides homeowners with simple and cost-
effective methods for strengthening their wood-frame houses for earthquake survival. The
program includes earthquake-strengthening workshops for residents, a list of earthquake
contractors to do the work, as well as a tool-lending library for the homeowners should they wish
to do the work themselves. Furthermore, the Association of Bay Area Governments (ABAG) has
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
an on-line listing of contractors proficient in this type of earthquake retrofit work
(http://www.abag.ca.gov) and there is current research being conducted on the earthquake hazard
mitigation of woodframe construction in the state of California, namely the CUREe-Caltech
Woodframe Project (Seible et al. 1999).
In order to accurately assess the total number of wood frame structures and a breakdown
of when these structures were built, a review of the 1990 census data was completed.
Additionally, the Oakland building department, the Alameda County tax assessors office, and a
few seismic retrofit companies in the Bay Area were contacted. Through communications with
the Oakland building department, it was estimated that virtually all housing stock in Oakland is
wood frame construction. Through a review of the census data, it was estimated that thirty-eight
percent (38%) of the wood frame construction in Oakland was built before 1940. Both the
Oakland building department and a seismic retrofit company in the Oakland area confirmed this
statistic.
Based on this information, it was estimated that ninety-five percent (95%) of the total
residential building stock is comprised of low-rise wood frame single family dwellings (Table
4.3). With the increase in percentage of wood frame structures from 83% to 95% (from Table
4.2), the percentage of structures considered steel light frame (S3), low-rise steel frame (S4L),
and low-rise reinforced masonry bearing walls (RM1L) are scale down accordingly. Of the total
number of wood frame structures, seventy-two percent (72%) are designed to Moderate-Code and
twenty-three percent (23%) are designed to High-Code {i.e. built after 1973).
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 43 Updated RES1 Occupancy Scheme
Seismic Design Level Total (% and
1000s sq. ft.)
W1 (% and
1000s sq. ft.)
S3
(%)
S4L
(%)
RM1L
(%)
High-Code (HC) 25% 20,630 23% 18,980 1% 0% 1%
Moderate-Code (MC) 72% 59,416 72% 59,416 0% 0% 0%
Low-Inferior (LI) 3% 2,476 0% 0 1% 0% 2%
Total 100% 82,522 95% 78,396 2% 0% 3%
The assumption that thirty-eight percent (38%) of the low-rise wood frame structures are
built prior to World War II needs to be incorporated in the analysis. In other words, an update to
the occupancy mapping scheme is required. Therefore, the low-rise wood frame structures
considered for mitigation (i.e. the ones built before 1940) are placed in one seismic design level
before mitigation and in another seismic design level after mitigation. Specifically, the structures
are placed in the Low-Inferior (LI) bin before mitigation (Table 4.4(a)), and they are placed in the
Moderate-Code (MC) bin after mitigation (Table 4.4(b)).12
Table 4.4(a) RE SI Occupancy Scheme for Analysis (Before Mitigation)
Seismic Design Level Total (% and
1000s sq. ft.)
W1 (% and
1000s sq. ft.)
S3
(%)
S4L
(%)
RM1L
(%)
High-Code (HC) 25% 20,630 23% 18,980 1% 0% 1%
Moderate-Code (MC) 34% 28,058 34% 28,058 0% 0% 0%
Low-Inferior (LI) 41% 2,476 38% 31,358 1% 0% 2%
Total 100% 82,522 95% 78,396 2% 0% 3%
Table 4.4(b) RES1 Occupancy Scheme for Analysis (After Mitigation)
Seismic Design Level Total (% and
1000s sq. ft.)
W1 (% and
1000s sq. ft.)
S3
(%)
S4L
(%)
RM1L
(%)
High-Code (HC) 25% 20,630 23% 18,980 1% 0% 1%
Moderate-Code (MC) 72% 59,416 72% 59,416 0% 0% 0%
Low-Inferior (LI) 3% 2,476 0% 0 1% 0% 2%
Total 100% 82,522 95% 78,396 2% 0% 3%
12To be completely accurate, when using the updated structural fragility curves in the sensitivity analysis,
the structures are placed in the Low-Code (LC) bin for the after mitigation analysis (See Section 5.2.2).
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
With the occupancy mapping scheme defined, the discussion turns to the incorporation of
expert opinion to reflect structural mitigation in the analysis.
4.3 Expert Opinion Incorporation
One of the most important parts of this work is the inclusion of data collected from
experts on the benefits and costs of earthquake mitigation (Figure 4.4). When one considers
mitigating a structure for earthquake risk, the assumed benefits and costs of the mitigation
technique complicate an earthquake loss estimation analysis. To date, limited published data
exists on the vulnerability of structures for various levels of ground shaking, and quantitative
benefit-cost studies on earthquake mitigation are minimal. Therefore, surveys were conducted to
gather information from experts on structural vulnerability and the benefits and costs of
retrofitting a wood frame structure for an earthquake hazard.
The rest of this chapter is broken down into the following six sections. First, a review of
related research on structural vulnerability, the benefits of structural earthquake mitigation, and
the costs of mitigation is briefly presented (Section 4.3.1). Next, the expert opinion surveys sent
to structural engineers and contractors proficient in residential earthquake mitigation are
described (Section 4.3.2). Third, the response rate and comments from the respondents are
summarized (Section 4.3.3).
Following this, the statistics and qualitative information on the mean damage o f loss from
the structural engineers and mean cost of retrofit from contractors are proffered (Section 4.3.4).
The approach used to fit this damage and cost data to reasonable distributions is given (Section
4.3.5). Finally, the remaining steps to achieve the before mitigation and after mitigation
structural fragility curves in HAZUS are detailed (Section 4.3.6).
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In Figure 4.4, this framework is shown with the ultimate goal of performing the Range of
Uncertainty analysis and cost-benefit analysis defined in Chapter One. But, first, a quick review
of structural vulnerability and mitigation is presented.
Structural
Vulnerability
Structural Engineers
....
Contractors
Benefits of
Mitigation
Costs of
Mitigation
Before Mitigation
/ After Mitigation
P[ds | S D ] = <D[ - J ln(=^)]
Before Mitigation
I
L(x {(0,0))
L(x0(0,0))
L(x0(l.O))
L(xq(\,0))
After Mitigation
1r
( >
C o C jowl
or
Co ^-'malcnal
^ J
Range o f Uncertainty Analysis
Homeowner Cost-Benefit Analysis
RU[L(X)]= \L(x\(i,j)) - L(xQ(i,j))
NPV = - C a + BtM
r=l (1 + r ) r
Figure 4.4 Expert Opinion Incorporation
4.3.1 Related Research on Vulnerability and Mitigation
The seminal report on earthquake damage to buildings in California was published in
1985 by the Applied Technology Council. Known as the ATC-13 guidelines, it was a gathering of
expert opinion on expected mean damage to many different types of structures (See discussion in
Section 2.5).
In the ATC-13 guidelines (1985), it is noted that typical wood frame residential
construction in California is expected to perform well during an earthquake event. But, this is not
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
true in the case of pre-1940 buildings. "These structures normally are supported on cripple studs
and generally do not have a perimeter foundation that extends up to the floor (ATC-13 1985). In
the guidelines, these structures are so-called nonstandard construction and are more susceptible
to damage in an earthquake event. The way in which design and construction quality is dealt
within the study is to shift the probabilities of a given damage state intensity downward from
standard construction to nonstandard construction.
Furthermore, the reduction in damage to pre-1940 residential structures with this type of
mitigation (i.e. bolting the structure and bracing the cripple wall) is not well documented.
Although there exists a number of qualitative sources of information on the benefits of bolting
residential structures to their foundations and bracing cripple walls of older residential
construction, there is limited information on quantifying the benefits of such techniques. The only
two notable sources of information (that the author found) on quantifying the benefits of
mitigation are the FEMA 227/228 report (FEMA 1992) and published data on experimental
testing of retrofitted cripple walls after the Loma Prieta earthquake (Shepherd and Delos-Santos
1991).
The FEMA 227/228 series, A Benefit-Cost Model f o r the Seismic Rehabilitation o f
Buildings, provide percentage reductions in mean damage to various classes of structures for
levels of ground shaking ranging from MMI VI to XII. These new mean damage ratios for after
mitigation effects assume that the earthquake retrofit technique used is a structural strengthening
based primarily on life safety. As in ATC-13, these estimates were based on the engineering
experience and judgment of those involved in compiling the report. The authors of this document
were quick to note that effectiveness of retrofit will vary depending on the rehabilitation
techniques used, on the standard, code, or safety level to which seismic rehabilitation is carried
out, and on the design, construction, and condition of the building before rehabilitation. In Table
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.5, the ATC-13 MDF before mitigation for standard and nonstandard (eg. pre-1940) construction
is compared to the ATC-13 MDF after mitigation using the FEMA guidelines.
Table 4.5 Mean Damage Factors
MMI
ATC-
Before
13 MDF
Mitigation
FEMA 227/228
Percentage
Reduction
ATC-13 MDF
After Mitigation
Standard Nonstandard Nonstandard
VI 0.8 4.7 50% 2.7
VII 1.5 9.2 50% 6.0
VIII 4.7 19.8 43% 14.3
IX 9.2 24.4 35% 19.5
X 19.8 37.3 28% 29.8
XI 24.4 -- 20%
XII 37.3 20%
As for the experimental testing on cripple walls (Shepherd and Delos-Santos 1991), the
testing was done at the University of California, Irvine, in 1990. Test specimens included non
retrofitted panels and those retrofitted with diagonal bracing, Simpson strap ties, and plywood
sheathing. Panels retrofitted with plywood showed a 90% increase in shear load capacity (in
pounds per linear foot). Although this testing indicates the increase in strength and stiffness of
bracing a cripple wall, it does not include the increased benefit of bolting the structure to its
foundation.
Finally, the costs associated with retrofitting an existing structure for earthquake effects
can be much more than those associated with a new building, and they can be very uncertain.
During a search on current reports estimating the costs associated with seismic rehabilitation of
buildings, publications by the California Seismic Safety Commission (CSSC 1994) and R. P.
Gallagher Associates, Inc. (1990) were uncovered. The CSSC ranges are based on 1991 prices
with the lower end price for material only and the upper end price for a professional doing the
work. The Gallagher estimates are based on 1990 mitigation costs for San Francisco and Los
100
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Angeles areas. Converted to 1999 dollars (at a six-percent rate of inflation), cost estimates range
in price from under $2,000 to over $ 11,000 (Table 4.6).
Table 4.6 Costs of Mitigation
CSSC R. P. Gallagher Associates
Bolting foundation: $400 -$7970
Bracing cripple wall: $800-$3990
Total: $1,200-$11,960
New cripple wall sheathing
and foundation anchor bolts:
$1,690-$3,380
Now, with this brief overview of structural earthquake damage and the benefits and costs
of structural mitigation, attention turns to the incorporation of expert opinion on these concepts.
4.3.2 Expert Opinion Surveys
The purpose of the survey is to determine, through expert opinion, the expected benefits
and costs associated with bracing the cripple wall and bolting the structure to its foundation for
pre-1940 low-rise wood frame residential structures in the cities of Oakland and Long Beach,
California. Long Beach was included in the original work completed for the Managing
Catastrophic Risks Project at the Wharton School (Grossi 1998 and Grossi et al. 1999).13
Estimated ranges of benefits and costs are based on responses from structural engineers
on the benefits of mitigation and contractors on the costs of mitigation. Five hundred surveys
were sent to structural engineers in northern California and aouthem California (i.e. 250 to
structural engineers in northern California and 250 to structural engineers in southern California).
The mailing list was obtained from the Structural Engineers Association of California (SEAOC)
roster, of which the author is a member. Nearly all of those queried were registered structural
engineers in the state of California.
13Notably, the tax assessors office supplied the inventory of the residences in these cities.
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Further, one hundred and sixty (160) surveys were sent to contractors in Northern and
Southern California {i.e. 80 to contractors in northern California and 80 contractors in southern
California). Mailing lists were obtained from the Association of Bay Area Governments
(ABAG) website for the northern Californian contractors and EQEs website for the southern
Californian contractors. All contractors had previously completed training sessions on retrofitting
wood frame homes.
The mailings were done in the spring and summer months of 1998 (See surveys in
Appendices 8.5 and 8.6). All respondents were informed as to the purpose of the questionnaire
and assured that their specific response information would be anonymous. Typically, each was
given a month to reply, and reminders were sent to those who had not responded by the reply
date. Both the structural engineers and the contractors were given the following building
specifications:
Consider a 2,200 square-foot wood frame residential building in the city of
Oakland, CA [or Long Beach, CA].14It is a pre-1940 two-story structure with a
market value of $130,000. It has 24 unbraced cripple walls and it is
inadequately bolted to its foundation.
The structural engineers were asked to rank their experience in post-earthquake damage
evaluation of residential structures on a scale of 0 (no experience) to 10 (extensive experience).
Based on their expert knowledge, they were queried as to the expected mean damage factor
(MDF) to this structure before mitigation and after mitigation for levels of MMI ranging from VI
to XII (For explanation of MMI, see Appendix 8.1). MDF is defined as the expected ratio of
14If the structural engineer or contractor lived in Northern California, the city of Oakland was used. If he
lived in Southern California, Long Beach was used.
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
dollar loss to replacement value for the structure, where 0 is no damage and 1 is complete
damage. The mitigation technique was broken down into bracing the cripple wall, bolting the
structure to its foundation, and bracing and bolting. The engineers were asked for estimates of
direct physical damage from ground shaking only. As in ATC-13, no collateral hazards such as
ground failure, fault rupture, inundation, or fire following are considered.
Furthermore, the structural engineers were asked to rank their confidence in their
estimates of damage for the various levels of ground shaking, a 0 designating not confident at
all and a 10 designating very confident. Their belief in the determining factors in evaluation
of damage was inquired, with a ranking of 0 implying not important at all and a 10
implying very important. These factors include, but are not limited to, square footage of the
structure, age of the structure, number o f stories of the structure, value of the structure, height of
the cripple wall, and inadequacy of the bolting to the foundation. Finally, their estimates on the
percentage of structures in the model city which would have more than S1,000 worth of damage
for MMI levels ranging from VI to XII were obtained.
Similar to the questionnaire sent to the structural engineers, the contractors were asked
about their experience with retrofitting o f wood frame residential construction {i.e. 0 designates
no experience and 10 designates extensive experience). Based on this experience, they
were asked cost estimates on bracing the cripple wall, bolting the structure to its foundation,
and bracing and bolting. Cost estimate ranges given were {$1,000-53,000, $3,000-55,000,
$5,000-57,000, $7,000-$9,000, or >$9,000}. Within these ranges, estimates of median values
were requested, as well estimates of the cost of material and labor in terms of a percentage of this
value.
The contractors were inquired about their confidence in their cost estimate and the
importance of several factors in determining the cost of the retrofit. Similar to the benefits survey,
103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
a 0 designated not important at all and a 10 designated very important. The factors
include, but are not limited to, square footage of the structure, age of the structure, number of
stories of the structure, value of the structure, height of the cripple wall, and accessibility to the
foundation.
4.3.3 Response Rate and Comments
First, the relevant statistics on the background of those who completed a survey are
presented (Table 4.7), as well as the comments from those responding. The return rate is based on
the number of surveys sent out less the number that were returned undeliverable. Notably,
twenty-seven percent (27%) of the structural engineers and twenty-four percent (24%) of the
contractors who responded declared their experience with damage assessment and retrofitting of
residential wood frame construction, respectively, as extensive (i.e. 10 on a scale of 0 to 10).
Further, as seen in Table 4.7, only a portion of the surveys returned were used for statistical
analysis (i.e. only surveys completed in their entirety were used). Finally, the majority of
structural engineers surveys utilized for the statistical analysis were from those who either work
for or own a structural engineering consulting firm (76.8%).
Table 4.7 Survey Respondents Statistics
Benefits of Mitigation Costs of Mitigation
Total Number of Respondents to Survey 119 46
Return Rate of Survey 25% 33%
Average Respondents Range of Age 45-55 years old 35-45 years old
Average Respondents Years in Practice 20-25 years in practice 15-20 years in practice
Total Number of Respondents Surveys
used in Statistical Analysis
69 35
Although a pilot test of both surveys was conducted, some errors in the data can still
arise. For both groups of respondents, there were two issues they had with the survey: the limited
information given on the characteristics of the structure and the separation of bracing the cripple
104
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
wall and bolting the structure to its foundation as two separate retrofit techniques. One
engineer stated that there was too little information...How well is the building tied together?
How strong is the shear resisting material? How much weight exists above the first floor - at the
2nd floor, at the roof? What kind of soil conditions exists under the building? How long does the
severe shaking last? A contractor also noted that this survey is too limited. There are more
factors relevant to the cost of retrofitting: condition of existing concrete or brick foundation; level
lot or hillside lot; width and length of building; presence of large openings in walls.
These respondents have a valid point. A limited amount of information was given; this is
purposeful. Most of the software models estimating damage calculate this damage based on a few
structural characteristics (i.e. construction type, occupancy class, number of stories, year built).
There is a need to ascertain how experts differ in their views on damage and retrofit cost based on
this limited information, and what additional information on the structure would be most useful in
predicting a better estimate of the benefits and costs. In general, most respondents indicated that
the quality and type of construction is very important in predicting earthquake damage and the
cost to retrofit.
As for the partitioning of bracing and bolting as two separate techniques to mitigate
for an earthquake hazard, engineers were quick to note that doing just one or the other...makes
no sense. Both should be done. Bolting in absence of bracing can actually increase the hazard
as the energy absorbed by sliding will not happen and the walls are more likely to overturn out-
of-plane since the bolting pins them at the base. This notion was confirmed by comments from
contractors, including: It would be poor practice to apply shear panel without bolts; both must be
done together. Again, this is a valid point. Bolting and bracing must be done together to be
effective. The purpose of the separation of the mitigation techniques was to get the respondents to
ponder their answers.
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The structural engineers had two other broad comments on the design of the surveys: one
relating to monetary values and one relating to the MMI scale. First, the use of $130,000 for
market value of the structure seemed low. One engineer noted that: value of such a property
would be nearer to $300,000 than $ 130,000. Caution must be used if you are going on the basis of
assessed value of structures {i.e. California Proposition 13, passed in the late 1970s, caps the
valuation increase at 2% per year unless a property is sold). Most respondents realized that there
is a distinction between market value and replacement value of a structure. The value of the
structure used in the survey is based on an average of the structure replacement value in databases
from the tax assessors offices in the cities of Oakland and Long Beach.
The use of the Modified Mercalli Intensity (MMI) scale to estimate mean damage was
not, in general, well received by the engineers responding to the survey. Not surprisingly, one
engineer noted that engineering seminars almost never refer to MMI, but to Richter or peak
ground acceleration. Further, the use of MMI levels above X is questioned. Most communities
that have experienced earthquakes with the type of construction described have not been
subjected to ground shaking over great areas of MMIs greater than VIII or IX. Once you get
above a X, then the superstructure will probably fail and you are back to having a lot of damage
even though the cripple wall was braced and anchor bolts added. MMI XI and XII relate to
geological damage, so the bolting/bracing retrofit measures not applicable. This survey is only
reliable for MMI up to X.
This is a well-founded point and one to keep in mind. MMI is not the ideal measure of
ground motion to use and extreme values of ground shaking intensity {i.e. MMI XII or total
damage) are not used in most probabilistic analyses. The choice was made to be consistent with
previous work done in this area {i.e. ATC-13). On the other hand, it has been viewed by some as
appropriate if characterizing damage on a regional level {eg. the HAZUS methodology). One
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
engineer summed it up nicely by commenting: MDF or PML [Probable Maximum Loss]
is...only appropriate for institutions that own many buildings or governments seeking to lessen
impacts for whole cities or regions.
4.3.4 Mean Damage and Cost
The aggregation of this expert opinion data is divided into information obtained from
structural engineers on structural vulnerability and the benefits of mitigation and from contractors
on the costs of mitigation. First, though, a note of clarification is required. While much has been
published on how to elicit opinions from experts in probabilistic risk assessment (Budnitz et al.
1997, Hora and Imam 1989), this is not the focus of this work. The focus is on how to aggregate
the data already gathered through the surveys. In other words, mathematical schemes rather than
behavioral schemes to aggregate expert opinion data are investigated within the scope of this
work.
A number of mathematical schemes utilizing Bayesian theory to aggregate expert opinion
data have been published, taking into consideration dependence, among other things (Genest and
Zidek 1986; Cooke 1991; Clemen and Winkler 1992). In its simplest form, the problem is
formulated as a Bayesian Normal model for independent experts who provide a point of scalar
quantity. Opinions are aggregated using a weighting scheme (Winkler 1968), and various
weighting schemes are used to estimate the precision of the experts forecasts.
With sixty-nine (69) complete surveys from the structural engineers, a Bayesian normal
model for independent experts is utilized to mathematically aggregate the information. The
results presented consider four weighting schemes or cases for the sixty-nine (69) experts (Table
4.8). Case A considers a simple average, where each experts mean damage factor is weighted
equally {i.e. no weighting scheme). Case B is a weighted average using the experts confidence in
their estimates, denoted C, m. Case C is a weighted average using the experts rankings of their
107
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
experience, denoted ,. Finally, case D uses both confidence and experience rankings to calculate
weights. With consideration of different combinations of confidence and experience weighting,
these four cases serve as a sensitivity analysis on the estimates of mean damage.
Table 4.8 Cases for Mathematical Aggregation
Case Description
Case A Simple Average
Case B Confidence (C,m) Weighted Average
Case C Experience (Ei) Weighted Average
Case D Confidence (C,.) and Experience (Ei) Weighted Average
In each of the four cases, the rankings are scaled downward to lie between zero and one.
In other words, a confidence or experience ranking of 10 is taken as 1 and a ranking of 0 is
kept at 0 {i.e. each ranking is multiplied by 0.1). From the structural engineers response^5, C/.m
is the confidence of expert / in his estimate of mean damage at MMI level m. Further, E, is the
experience of expert / in post-earthquake damage assessment of residential structures. Equations
4.1(a) and 4.1(b) are used to scale these rankings to the appropriate confidence weight, and
experience weight, ,.
0.1 CLn
l o . i c ^
i=i
v
0.1 -Ei
(4.1(a))
(4.1(b))
1.0.1-Ei
i=l
These four cases (Table 4.8) are presented mathematically in equations 4.2(a), 4.2(b), and
4.2(c) and 4.2(d). In each of these equations, y ijn designates the estimate of mean damage factor
from expert i for MMI level m. mis the confidence weight of expert / at MMI level m and e, is
15See Benefits Data in Appendix 8.7.
108
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the experience weight of each expert /, as given in equations 4.1(a) and 4.1(b). In Equation 4.2(d),
co,M is a combination of the experts confidence in his estimate of mean damage and his
experience: coi m =,
N
y m = ^ (4.2(a))
.V
T . C i ,m y i ,m
y m ---------- (4.2(b))
2 ^ / . *
i=l
.V
I * , yi.m
y m =M -Xr (4.2(c))
1*1
i=I
;V
X ^i,m ' yi,m
y n , = M ------------ (4 .2 (d))
i=l
Estimates of the mean damage factors for the various levels of ground shaking using the
four weighting schemes, as given in the above equations, are shown graphically in Figure 4.5. For
each level of MMI considered in the survey, denoted 6 through 12 on the x-axis of the graph,
estimates of mean damage factors before mitigation {i.e. the first four bars on the graph at each
MMI level) and after mitigation {i.e. the second four bars on the graph at each MMI level) are
given. These estimates range from 0 to 70 on the y-axis. Note that the highest mean damage
factor is actually 100 {i.e. total damage), but the highest weighted estimate is approximately 68.5.
109
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A (Before) HB (Before) C (Before) HD (Before)
A (After) DB (After) C (After) D (After)
6 7 8 9 10 11 12
Modified Mercalli Intensity (MMI)
Figure 4.5 Comparison of Weighting Schemes
From this graph, it is clear that there is no substantial difference in the estimates of mean
damage for the different weighting schemes either before or after mitigation.16 While this seems
misleading at first, it is actually the expected result. There are a large number of experts
estimates used in the calculations (namely sixty-nine). Previous research indicates that as the
number of experts increases, the less likely that different weighting schemes will have an impact
on the expected value calculation (Clemen 1989).
Additionally, standard deviations, <xm, on the mean damage factor estimates are
calculated, assuming a normal distribution of data. While the best distribution fit to the data in
most cases is not a normal distribution (See Section 4.3.5), some insight into the data may be
16 Performing tests of statistical significance is not completed due to the small number of data points {i.e.
four). Alternatively, it is taken as given from the results in Table 4.9 that the resulting y mvalues in each of
the four weighting schemes are close enough to consider the differences negligible {i.e. 1% to 3%
difference).
110
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
gained (i.e. the expected spread in the data). For each of the four cases, these are calculated using
Equation 4.3, which is based on Winklers method (Winkler 1968).
In equation 4.3, the term weight stands for the confidence weight, the experience
weight, or the weighting term which is a combination of both, co,M. The estimated means, y,
and associated standard deviations, crm, for cases A through D are given in Table 4.9.
Additionally, the minimum and maximum mean damage factors estimated by the experts are
listed. From Table 4.9, there is a significant spread in most of the mean estimates of damage.
'V 2
weighti m ( y i m y m )
' =/ (4.3)
jV
I. weight
i=l
Table 4.9 Benefits Survey Statistics of MDF
MMI
I Before Mitigation
Case A Case B Case C Case D
Min Max
ym Cm ym Om ym Om ym om
VI 2.27 2.95 2.35 3.09 2.29 2.34 2.38 1.84 0 15
VII 6.19 7.03 6.07 7.67 6.24 6.06 6.16 5.63 0 35
VIII 15.36 15.00 15.66 14.94 15.29 15.64 15.46 14.56 2 100
IX 27.32 20.21 27.75 21.15 27.51 27.73 27.67 20.91 9 100
X 43.91 25.78 44.41 25.69 43.55 44.39 44.01 37.02 15 100
XI 55.74 27.14 57.32 27.61 55.87 50.83 57.64 49.36 20 100
XII 66.62 24.51 68.09 25.00 66.71 61.34 68.54 60.35 25 100
MMI
After Mitigation
Case A Case B Case C Case D
Min Max
ym Om ym om ym om ym om
VI 1.00 1.53 1.05 1.72 1.08 1.03 1.13 0.85 0 9
VII 2.67 3.63 2.73 3.85 2.82 2.71 2.87 2.45 0 21
VIII 6.26 6.44 6.33 6.32 6.49 6.31 6.52 5.98 0 28.5
IX 12.05 10.73 12.03 9.95 12.50 12.02 12.44 6.27 0 58.5
X 22.10 16.17 22.14 15.59 22.56 22.12 22.65 16.44 0 80.75
XI 33.95 23.31 34.91 22.48 34.18 28.09 35.29 27.61 2.45 100
XII 46.19 29.70 48.52 28.93 46.55 42.09 48.98 40.41 5.61 100
1 1 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
While these calculations are very important to this study and they are the first step to
incorporating new fragility curves in HAZUS, there was other equally important qualitative
information gathered through these surveys. Specifically, three additional factors to consider in
post-earthquake damage evaluation emerged from the structural engineers surveys. They are (1)
the importance of the quality of construction; (2) the importance of the weight of the structure and
shear capacity above ground level; and (3) the significance of nonstructural damage.
First and foremost, the quality of construction is critical. One engineer noted that current
quality is so bad that older homes if they are bolted down and have cripple walls braced will
have less damage than new ones. Newer homes (i.e. many wall openings, discontinuous floors
and roofs, cathedral ceilings, etc.) make structural design critical and most architects and builders
will not pay for it. Current code requirements are not adequate for this, and the present code is not
understood and ignored too often.
Additionally, another engineer noted that in your questions, you have ignored the
conditions above the cripple walls. Though foundation rehabilitations are a cost effective
method for hazard mitigation of single family dwellings, I feel that upgrading of stucco only
shear walls to 54 minimum plywood shear walls would go a long way to improving the
performance of pre-1940 houses. Also, one engineer remarked that he had observed that the
houses which had the original roofing replaced with a tile roof sustained much more damage than
those which still had a light weight shingle roof. Houses which have heavier roofs installed
should also be evaluated for lateral (seismic) forces.
Finally, the importance of nonstructural damage should not be underestimated. The
1994 Northridge earthquake exposed a flaw in our philosophy concerning seismic resistant design
of residential buildings. It is not enough to bolt down and brace the cripple walls of raised
foundation houses. The damage to so-called nonstructural plaster and stucco walls often caused
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the repair of residences damaged by the Northridge earthquake to cost in excess of 35% o f the
pre-earthquake value of the residence. So although bolting down and bracing the cripple walls
prevent total loss, it does not prevent severe economic hardship and loss of use of the residence
after a moderate to severe earthquake.
Turning to the costs of mitigation, the analysis of the data collected from the contractors
considered the same four weighting schemes for compilation (Table 4.8). As expected, the
calculations are easier with less experts opinions included in the analysis {i.e. 35 contractors
versus 69 structural engineers). But, the different weighting schemes are no more influential in
the mean estimates of cost. In calculating the costs of mitigation, both total cost estimates and the
cost to do-it-yourself (i.e. cost of material only) were calculated. As presented in Chapter One,
the notation for mitigation cost, Ca, is either CTolai for total cost or CMalcnat for material cost,
depending on the decision processes of the homeowner (Equation 4.4).
CTmat if the homeowner hires a contractor
Ca = T (4.4)
CMatenai if the homeowner does the work
Similar to the damage estimates, case A considers a simple average, where each
contractors median cost estimate is weighted equally. Cases B and C are the weighted averages
using the experts confidence in their cost estimates, C and their experience with earthquake
retrofitting, E respectively. Finally, case D uses both confidence and experience estimates to
calculate weights. Equations 4.1(a) through 4.3 are used for these calculations.17As in the case of
damage estimates, these four cases serve as a sensitivity analysis on the estimates of the costs of
mitigation (See Costs Data in Appendix 8.8).
17 There is a slight difference in notation. Specifically, in the case o f cost estimates, confidence is not
dependent on level o f damage. Therefore, C, replaces C, min the applicable equations.
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
For the four weighting schemes, the estimated means for CTo,ai and Cmmc*, with
associated standard deviations, o ^ a n d crSktttnai, are presented in Table 4.10. The expected costs
under each of the weighting schemes are not substantially different.18Also, looking closely at the
data in Appendix 8.8, in general, those with more experience estimated a lower cost of material.
Finally, considering ail the respondents, the total cost ranged from $2,000 at the minimum to
$10,000 at the maximum (See Appendix 8.8).
Table 4.10 Costs Survey Statistics
Total Cost Cost of Material
cTotal
Ofotal (-'Material ^Material
Case A $5,583 $2,211 $1,277 $837
Case B $5,461 $2,130 $1,323 $766
Case C $5,651 $2,210 $1,323 $815
Case D $5,535 $2,160 $1,277 $725
As in the case of the structural engineers surveys, additional information was gathered
from the contractors on the costs of mitigation. Specifically, although accessibility to the
foundation is the number one important factor, additional determining factors in the cost of
structural mitigation were obtained. These include the size of the company doing the work {i.e. a
larger company has a bigger overhead cost), local codes, the structures configuration, and the
existing conditions of the foundation. One contractor declared: Existing conditions are of
extreme importance. Often foundations of older homes have significant settlement cracks, may
have deteriorated concrete and/or the cripple walls and mudsills may be dryrotted or termite
infested. Significant additional costs of repair can result. There are also many different foundation
designs. Some dwellings have cripple walls going through the center of the crawl space in
addition to the perimeter. Foundation/cripple wall design determines the type and cost of
18As in the case o f the damage data, performing tests o f statistical significance is not completed due to the
small number o f data points (i.e. four).
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hardware and methods employed. These comments explain why there was a large disparity
among the contractors estimates of cost.
4.3.5 Distribution Fit
Gathering this data from experts on the benefits and costs of mitigation is an important
step in quantifying the vulnerability of structures before and after mitigation and estimating the
costs of mitigation. Specifically, the mean damage and cost estimates can be utilized for a more
accurate assessment of cost-benefit ratios in the homeowner analysis. One more step, however, is
necessary to get the most accurate estimate of damage at each MMI level and cost to retrofit the
structure. This entails fitting distributions to the data.
A large amount of time was spent fitting the damage and cost data to the best distribution
possible. In all, fourteen distributions of damage {i.e. for 7 levels of MMI, levels VI through XII,
and 2 states of fragility, before and after mitigation) were fit. Furthermore, two distributions of
costs {i.e. for material cost and total cost) were fit. Each of these distributions is presented here,
with an explanation of any notable intermediate steps to arrive at the best fit distribution of data.
The BestFit software was used to fit these distributions. This commercial software
package, uses maximum-likelihood estimators to fit distributions to a set of sample data. A
maximum-likelihood estimator or MLE of a distribution is the parameter of that distribution that
maximizes the likelihood function, given a set of observations (Larsen and Marx 1986). In all,
twenty-one (21) types of distributions are available within the software to fit to a data sample.
Three goodness-of-fit tests, defined as the probability that a given distribution function
produced the sample data, are used to check the accuracy of each of these fits. The BestFit
software provides goodness-of-fit statistics for the Chi-Square Test, the Kolmogorov-Smimov
115
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Test, and the Anderson-Darling Test. For more information on these tests, see Larsen and Marx
(1986).
First, the fitting of the mean damage estimates from the structural engineers is presented.
With no substantial difference between the weighting schemes of the four cases defined in Table
4.8, the raw data {i.e. no weighting schemes used) of mean damage versus MMI level from the
sixty-nine experts was used for curve fitting. This result was fortunate, given the fact that
including a weighting scheme would complicate the calculations.
The first iteration of fitting curves to the mean damage data considered the beta
distribution only. This distribution, utilized in the ATC-13 study (1985) and given in Equation
4.5, was tested first because of its flexibility and its variable limitation between zero and one {i.e.
0 < y m< 100 is scaled downward to 0 < MDF < 1). However, as seen in Table 4.11 and Figure
4.6, the beta distribution proved to be too limiting in some cases. Specifically, Figure 4.6 shows
the beta distribution for the mean damage factor at MMI level IX before mitigation and after
mitigation. These curves have spikes in them {i.e. they are not smooth distributions) and prove
not well suited for the data. Moreover, at other MMI levels, the data failed all three goodness-of-
fits tests for the beta distribution.
In Table 4.11, the parameters at and a2 are given, as well as the mean damage factor
(y d 100) and the standard deviation of the mean damage factor {aJlQOi).
f {ym^ y f ^ ^ L (4.5)
jtl
0
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.11 Beta Distributions Parameters for Damage Data
Before Mitigation
MMI
a; yjioo <TJ100
VI 0.55 23.87 0.02 0.03
VII 0.67 10.10 0.06 0.07
VIII 0.56 3.61 0.15 0.15
IX 0.49 2.18 0.27 0.20
X 0.62 1.49 0.44 0.26
XI 0.73 1.09 0.60 0.29
XII 1.33 1.41 0.73 0.26
After Mitigation
MMI
a2
yjioo CTh/ 1 0 0
VI 0.41 39.84 0.01 0.02
VII 0.49 17.57 0.03 0.04
VIII 0.84 10.76 0.07 0.07
IX 1.01 6.37 0.14 0.12
X 1.26 4.19 0.23 0.17
XI 0.94 2.04 0.34 0.23
XII 0.73 0.91 0.50 0.31
c
S
c
a
5#5
at
<>
o
l.
u

12
10
IX Before
IX After
8
6
4
2
0
0.1 0.2 OJ 0.4
Mean Damage Factor (ym)/ 100
0.5
Figure 4.6 Beta Distribution Fit to Sample Damage Data (MMI IX)
117
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The next iteration expanded to fitting the gamma distribution to the sample data. This
distribution was chosen because it is quite similar to the beta distribution, with no upper
limitation on the domain value (i.e. to limit spikes in the distribution). Upon inspection, however,
fitting the mean damage factor sample data to this one distribution for all fourteen cases was not
feasible. More specifically, in subsequent manipulation of the data to develop HAZUS fragility
curves, the gamma distribution curves also proved limiting. Other distributions fit the data much
better.
After much time and consideration, a set of distributions was settled upon that fit the data
best (Table 4.12). There are seven different distributions considered, with each distribution
passing at least two of the goodness-of-fit tests. A listing of these distributions density functions
is given in Appendix 8.9. In Figure 4.7, an example of the mean damage factor distribution for
MMI IX before and after mitigation is given. Comparing this to Figure 4.6 and considering the
confidence in the goodness-of-fit tests, it is clear that this is a much better fit of the sample data.
For example, in the case of MMI IX before mitigation, the Inverse Gaussian distribution passes
the Anderson-Darling test at the 0.15 level of significance. After mitigation, the Gamma
distribution passed the Anderson-Darling test at the 0.05 level of significance.
One final note on these best distribution fits is required. In some cases, these distributions
allow values over one. In other words, the mean damage factor is above 100, which is not feasible
or even reasonable. At most, there can be 100% damage. This discrepancy, however, does not
pose a problem in further analysis of the data. In the worst case, integration under the distribution
equals 0.98 and it is scaled up to 1.0 (See Section 4.3.6.1).
118
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.12 Best Distribution Fit Parameters for Damage Data
Before Mitigation
MMI Type
a, a2 as
yjioo
a ./100
VI Gamma 0.590 0.038 0.023 0.030
VII PearsonVI 1.541 2.894 0.080 0.065 0.102
VIII InverseGaussian 0.164 0.134 - - 0.164 0.181
IX InverseGaussian 0.273 0.477 - - 0.273 0.207
X InverseGaussian 0.439 1.231 - - 0.439 0.262
XI Rayleigh 0.438 0.549 0.287
XII Weibuil 3.034 0.749 0.669 0.241
After Mitigation
MMI Type
at a2 aj
yjioo o^/100
VI Gamma 0.424 0.024 0.010 0.015
VII Gamma 0.543 0.049 0.027 0.036
VIII Erlang 1.000 0.066 0.066 0.066
IX Gamma 1.538 0.087 0.133 0.108
X PearsonVI 4.599 3.239 0.122 0.251 0.275
XI Lognormal 0.356 0.289 0.356 0.289
XII Erlang 2.000 0.233 0.465 0.329
IX Before
e
v
a
I/!
a
o
u
u
.2
s
Z
5 IX After
4
3
2
0
0.1 0.2 0.3 0.4 0.5 0
Mean Damage Factor (ym)/100
Figure 4.7 Best Distribution Fit to Sample Damage Data (MMI IX)
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In the case of costs of mitigation, the fitting of the estimates from the contractors proved
much easier. Only one iteration was needed to fit this data to its best distribution. With no
substantial difference between the four weighting schemes in Table 4.8, the cost data for total
cost of mitigation, Ctoioi, and cost of material, CK(altnai, was utilized for curve fitting. Results for
the first two distributions that best fit the data for each of these costs are given in Table 4.13.
From this analysis, it was discovered that the total cost of mitigation is closely
symmetrical in distribution {i.e. only slightly skewed), as seen in Figure 4.8. This is a graphical
representation of the weibull distribution fit to the total cost data. The symmetry of the
distribution is reinforced by the normal distribution, which was the third best distribution to fit the
sample data set.
On the other hand, the cost of material for mitigation is skewed to the left. More
contractors tended toward the lower end of material cost, as a percentage of total cost. And, with
these best distributions fit, the cost of mitigation is in its final form for the homeowner cost-
benefit analysis described in Chapter One. This analysis will be completed in Section 6.3.
Table 4.13 Best Distribution Fit Parameters for Cost Data
Type
Total Cost
a,
a2
CTotal 07*910/
(1) Weibull 2.77 6275.94 5586.02 2181.99
(2) Gamma 6.17 904.12 5582.86 2246.68
Type
Cost of Vlaterial
a2 Materia! (^Material
(1) Lognormal 1383.78 976.40 1383.78 976.40
(2) Gamma 2.87 473.18 1359.14 801.94
120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.2
ii 0.4
s
3
X 0.6
<>
e
e
u
0 .. . . ---------------------------------------
$2,000 $4,000 $6,000 $8,000 $10,000 $12,000
1
Total Cost of Mitigation
Figure 4.8 Best Distribution Fit for Sample Cost Data
4.3.6 Structural Fragility Curve Development
While the cost of mitigation data is in its final form for subsequent analyses, more needs
after mitigation. Specifically, the mean damage factor data before and after mitigation must be
converted into HAZUS fragility curves.
To reflect mitigation in the HAZUS software (i.e. bolting the structure to its foundation
and bracing its cripple wall), a change in the intersection of the structures capacity curve and the
response spectrum of demand for the four applicable structural damage states is calculated. The
applicable damage states include Slight, Moderate, Extensive, and Complete, as described in
Section 3.2.3. Specifically, SD,ds, the median value of spectral displacement at which the
building reaches the threshold of damage state, ds, is shifted to reflect mitigation (i.e. to a higher
allowable drift ratio). Additionally, Pd^ >the standard deviation of the natural logarithm of
to be done with the distribution of structural damage to low-rise wood frame homes before and
121
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
spectral displacement of damage state, ds, is changed to reflect the spread in the expert opinion
data.
The methodology to achieve reasonable estimates for the So,ds and Pd^ terms in the
structural fragility curves for low-rise wood frame structures in Oakland before and after
mitigation involves five (5) functional steps to achieve (Figure 4.9). Notably, this approach builds
on previous work on the vulnerability of structures subject to earthquake ground shaking (eg.
ATC-13 1985; Anagnos et al. 1995; Newmark and Hall 1982).
Step one entails aggregating the expert opinion data on mean damage factors (MDF)
versus Modified Mercalli Intensity (MMI) into an analytical distribution at each level of ground
shaking. This was done in Section 4.3.5. Second, these BestFit15 distributions are utilized to
develop damage state probability matrices (DSPM), designating the probability that damage is
within one of the four damage states of interest.
Third, once these probabilities are developed, cumulative probabilities, estimating the
probability that a given level of damage will be met or exceeded, are developed. A cumulative
lognormal distribution function is fit to these cumulative DSPM data points. Fourth, theMMI
measure of ground shaking is converted to peak ground acceleration (PGA). Finally, the PGA is
converted to an equivalent spectral acceleration (Sx) and spectral displacement (SD) pair based on
a reference demand spectrum. In other words, a demand spectrum is developed, which intersects
the building capacity curve at the median value of spectral displacement for each pre-defined
damage state. This defines the SD,ds term, and the Pd^ term follows from the estimate of
spread in the peak ground acceleration value.
The next four sections delve into the details of steps two through five only, since step one
was covered in the previous section.
122
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
MDF vs. MMI
Sample Data
Step 1: Fit Mean Damage Factor
(MDF) Sample Data to
Distribution f(y,,in) for each MMI
level (Section 4.3.5)
I
Step 2: Develop Damage State
Probability Matrices for each
damage state ds
(Section 4.3.6.1)
Convert MMI to In(MMI)
Step 3: Fit Cumulative
Lognormal Distributions of
damage for each damage state ds
(Section 4.3.6.2)
I
Step 4: Convert Cumulative
Lognormal Distributions from
MMI to PGA
(Section 4.3.6.3)
I
Step 5: Construct Response
Spectrum of Demand based on
PGA and intersect with Capacity
Curve (Section 4.3.6.4)
P[ds\SD] = <t>[ ln(=
S D
P s j s S D,ds
-)]
Figure 4.9 Development of Structural Fragility Curves
4.3.6.1 Damage State Probability Matrices
With the expert opinion survey data fit to the sample curves (Table 4.12), the second step
is to develop damage state probability matrices (DSPM). These DSPM reflect the probability of
being in one of the four damage states of interest, as defined in the HAZUS software. While the
123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
damage states characterized as Slight, Moderate, Extensive, and Complete are considered, the
fifth damage state is not considered here. This state, None, refers to no damage at ail.
Based on the qualitative descriptions o f structural damage states given in the HAZUS
methodology (NIBS 1997) and previous estimates published on mean damage factor (MDF)
versus MMI (ATC-13 1985), the lower and upper limits of damage for the four structural damage
states are chosen accordingly. For example, a mean damage factor of 0.01 is the lower limit on
Slight structural damage and 0.60 is the upper limit on Extensive structural damage. A full listing
of these limits is given in Table 4.14.
Table 4.14 Damage Factor State Limits
Damage State Lower Limit
(lower)
Upper Limit
(upper)
Slight 0.01 0.10
Moderate 0.10 0.30
Extensive 0.30 0.60
Complete 0.60 1.00
The damage state probability matrices are then developed by integrating under the
fourteen curves fit in step one (Section 4.3.5) and assuming bounds of integration between the
pre-defined limits given in Table 4.14. The Mathcad software, developed by MathSoft, Inc., is
used to integrate these functions. For example, in the case of MMI IX, a Gamma distribution is fit
to the damage data after mitigation (Equation 4.6).
ym=uPPer a2~ai Xmai_1 exP<_ )
P[lower <y m < upper] = f -------------------- ---------- ~ d y m (4.6)
; r ( a , )
y m =lower
In Equation 4.6, ym designates the damage, a/ and or? are the parameters of the
distribution given in Table 4.12, and lower and upper terms are limits of integration defined in
Table 4.14. This integration is performed for each structural damage state, both before and after
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
mitigation. In all, eight sets ot' probabilities are developed (i.e. four damage states and two states
of fragility). These DSPM are checked by adding the probability of damage state None to the
set and confirming that the sum of the integration for each MMI level is one (Table 4.15). A full
listing of the damage state probabilities is given in Table 4.15.
Table 4.15 Damage State Probability Matrices
Before Mitigation
MMI None Slight Moderate Extensive Complete
VI 0.461 0.510 0.029 0 0
VII 0.127 0.699 0.151 0.018 0.004
VIII 0 0.486 0.375 0.106 0.033
IX 0 0.130 0.553 0.246 0.071
X 0 0.006 0.358 0.465 0.171
XI 0 0.028 0.198 0.432 0.343
XII 0 0.002 0.064 0.373 0.561
After Mitigation
MMI None Slight Moderate Extensive Complete
VI 0.697 0.300 0.003 0 0
VII 0.442 0.510 0.048 0 0
VIII 0.141 0.639 0.209 0.011 0
IX 0.025 0.448 0.447 0.076 0.004
X 0 0.216 0.533 0.186 0.065
XI 0 0.077 0.469 0.316 0.138
XII 0 0.075 0.323 0.387 0.215
43.6.2 Cumulative Lognormal Distributions
From these damage state probabilities, cumulative damage state probability matrices
(DSPM) are developed. The cumulative DSPM reflect the probability that the structural damage
meets or exceeds the various damage state levels. Specifically, the probability that the structural
damage will meet or exceed the Slight damage state is the sum of the damage state probabilities
for Slight, Moderate, Extensive, and Complete damage. The probability that the structural
damage will meet or exceed the Moderate damage state is the sum of the damage state
probabilities for Moderate, Extensive, and Complete damage. The probability that the structural
125
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
damage will meet or exceed the Extensive damage state is the sum o f the damage state
probabilities for Extensive and Complete damage. Finally, the probability that the structural
damage will meet or exceed the Complete damage state is the damage state probability for
Complete damage.
Put simply, the limits of integration {lower, upper) will change from (0.01, 0.1) to (0.01,
1.0) to calculate cumulative Slight damage, from (0.1, 0.3) to (0.1, 1.0) to calculate cumulative
Moderate damage, and from (0.3, 0.6) to (0.3, 1.0) to calculate cumulative Extensive damage.
The limits of integration for the Complete damage state remain the same. The probabilities for the
cumulative DSPM are given in Table 4.16. These probabilities reveal that the surveyed structural
engineers believe there is a large reduction in structural damage at the Extensive and Complete
damage states for this mitigation measure {i.e. bolting the structure to its foundation and bracing
its cripple wall).
Table 4.16 Cumulative Damage State Probability Matrices
DSPM Values Before Mitigation
MMI Slight Moderate Extensive Complete
VI 0.539 0.029 0 0
VII 0.873 0.174 0.022 0.004
VIII 1 0.514 0.139 0.033
IX 1 0.870 0.317 0.071
X 1 0.972 0.636 0.171
XI 1 0.994 0.774 0.343
XII 1 0.998 0.934 0.561
DSPM Values After Mitigation
MMI Slight Moderate Extensive Complete
VI 0.303 0.003 0 0
VII 0.558 0.049 0 0
VIII 0.860 0.220 0.011 0
IX 0.975 0.527 0.080 0.004
X 1 0.784 0.251 0.065
XI 1 0.923 0.453 0.138
XII 1 0.925 0.602 0.215
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
But, a lognormal curve still needs to be fit to these data points in order to develop the
final fragility curves. The choice of the cumulative lognormal distribution function is predicated
by the choice of this function for the fragility curves in HAZUS. With a cumulative lognormal
function, the computations are quite efficient and curves can be combined easily. And while the
lognormal function is not the best fit in all damage states, it fits the discrete data points well in
most cases. First, however, the MMI values are converted to In(MMI) for ease of data
manipulation, as suggested by Anagnos and others (1995).19
Mathematically, the cumulative lognormal probability function reflects the probability of
being in or exceeding each structural damage state as a function of In(MMI). Given in equation
4.7, it is the integral of the lognormal density function (See Appendix 8.9). More specifically, it is
the probability that the median value of structural damage meets or exceeds a pre-defined level of
damage. Note the use of median value here. This is consistent with the definition of fragility in
HAZUS (i.e. SD,ds is the median value of spectral displacement at which the building reaches
the threshold of damage state ds).
y
P{Y <y) = F( y) = J - L ^ exp(- n y 1-f - -)dy (4.7)
0 yy2nr-<jx~ 2- <xx
Looking more closely at equation 4.7, y represents MMI and x represents In(MMI)
(Anagnos et al. 1995). The parameters for these curves, /jx and <rx, denote the median value of
In(MMI) and the standard deviation of In(MMI), respectively. These parameters are given in
Table 4.17, and they are consistent with the earlier notion of reduction in damage after mitigation.
In other words, a higher In(MMI) value is needed to reach the threshold of structural damage after
mitigation. Note, too, in Table 4.17, the standard deviation ofIn(MMI) reduces after mitigation
19Note that this step is not defined as a separate step, being simplistic in nature.
127
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
for the Extensive and Complete damage states. This is anticipated, due to the nature of the
cumulative probabilities given in Table 4.16 {i.e. the data points are zero at the lower MMI
levels).
Table 4.17 Statistics of Cumulative Lognormal Distributions (MMI)
Damage
State
Before Mitigation After Mitigation
M l m W M I ) O'Im(MStl)
Slight 1.81 0.17 1.90 0.17
Moderate 2.06 0.14 2.19 0.15
Extensive 2.26 0.16 2.40 0.12
Complete 2.40 0.13 2.46 0.07
In the end, eight (8) cumulative lognormal distribution functions are calculated, which
estimate the median value of ln(MMI) for the threshold of Slight, Moderate, Extensive, and
Complete structural damage states before and after mitigation. An example of the fitted
cumulative lognormal functions for the Moderate structural damage state, before and after
mitigation, is given in Figures 4.10. At this damage state level, the curves show quite a good fit.
Both before and after mitigation, the lognormal distribution passes the Kolmogorov-Smimov test
at the 0.15 confidence level.
DSPM Value (Before)
1 Moderate (Before)
DSPM Value (After)
- - Moderate (Alter)
0.8 -
0.6
0.4
1.6 1.8 2 2.2 2.4 2.6
In(MMI)
Figure 4.10 Cumulative Lognormal Distribution (Moderate Damage)
128
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4J.6J Conversion to Peak Ground Acceleration
With the cumulative lognormal distribution functions fit to the data points, the next-to-
last step is to convert the In(MMI) values to equivalent peak ground acceleration (PGA) values.
While this step seems straightforward, the process is rather subjective in nature and, therefore,
hard to achieve.
In essence, the use of Modified Mercalli Intensity (MAO) or another intensity level to
assess ground shaking is not an ideal measure (See discussion in Section 4.3.3). Since wide
statistical scatter exists in the data measuring intensity, conversion to peak ground acceleration is
a tricky matter. There are numerous empirical relationships that can be utilized for this
conversion. Some early relationships include ones developed by Trifunac and Brady (1975),
Ambraseys (1974), and Gutenberg and Richter (1956). These relationships are graphically
represented in Figure 4.11, along with a portion of the empirical relationship used in the HAZUS
software (i.e. MMI levels VI through X).
10.00 r
1.00
s
<
O
- 0.10 r
MMI
0.01
--Gutenberg-Rich ter -n-Trifunac-Brady
--Ambraseys HAZUS
Figure 4.11 Empirical Relationships of MMI versus PGA
129
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
These empirical relationships and others were used to test the sensitivity of converting
the In(MMI) values to PGA. In the end, the HAZUS conversion of MMI to PGA was used. These
values, given in Table 4.18 (NIBS 1997 Table 10.1), result in equivalent median PGA values for
threshold of structural damage in the Slight, Moderate, Extensive, and Complete damage states.
Table 4.19 is a listing of these equivalent median PGA values, /jPCa>and the standard deviation of
these values, erPOA, for the four structural damage states of interest. These median PGA values are
used to construct the reference demand spectrum (and find the final SD,ds values). The standard
deviations of the PGA values are used to develop final values for the term in the HAZUS
structural fragility curve.
Table 4.18 MMI to PGA Conversion Table
MMI VI VII VIII IX X XI XII
PGA 0.12 0.21 0.36 0.53 0.71 0.86 1.15
Table 4.19 Statistics of Cumulative Lognormal Distributions (PGA)
Damage
State
Before Mitigation After Mitigation
U pga OPGA P pga <*KA
Slight 0.13 0.09 0.18 0.12
Moderate 0.32 0.16 0.49 0.24
Extensive 0.60 0.30 0.91 0.31
Complete 0.92 0.34 1.09 0.22
4.3.6.4 Intersection of Demand Spectrum and Capacity Curve
Finally, in the last step, the median PGA values to reach the threshold of the various
structural damage states are transformed into an equivalent spectral displacement, SD, ds,
through an assumed demand spectrum shape. In this way, the assumed demand spectrum shape
130
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
relates spectral response to the PGA values. The approach taken is the inverse of the approach to
develop equivalent PGA fragility curves for lifeline loss in HAZUS (Section 5.4.4 in NIBS 1997).
The elastic (5%-damping) reference spectrum shape used represents ground shaking of a
large-magnitude Western United States (WUS) earthquake for NEHRP soil site class D (See
Section 3.2.1 on response spectrum and soil site classes). More specifically, a moment magnitude,
M, earthquake of 7.0 is the event used to create the reference spectrum and the spectrum is scaled
upward to reflect increase in ground motion from rock (site class B) to stiff soils (site class D).
Moreover, site-to-source distances of fifteen (15) kilometers or greater for buildings are
considered. It should be emphasized that the resulting reference spectrum is very sensitive to
these values. Therefore, the choice in magnitude event and source-to-site distance was not done
lightly. It is believed that these choices bode well for a future expected moment magnitude event
(7.0) at this distance (15 km) from the Oakland region.
Since it is not feasible to create a whole set of response spectrum shapes based on all
possible combinations of magnitude and distance values, this one elastic reference spectrum is
used for analysis. Specifically, this elastic spectrum is scaled downward, as detailed in Chapter
Three, to an equivalent response spectrum of demand for the structures in the residential building
stock. This demand reflects 15% damping of low-rise wood frame structures {i.e. W1 in HAZUS)
and the hysteretic energy dissipated by these structures pushed beyond their elastic limits"
(NIBS 1997).
As detailed in Section 3.2.3, the point at which the reference demand spectrum intersects
the building capacity curve (before and after mitigation) is the median value of spectral
displacement at which the building reaches the threshold of the structural damage state of interest.
The details to arrive at this point, however, are rather cumbersome. The eight different reference
demand spectra, reflecting the earthquake demand for the four structural damage states of interest
131
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and the two fragility or mitigation states, must be drawn. This set of curves is based on many
different assumptions. Then, the two capacity curves, reflecting the capacity of the structure
before and after mitigation, must be drawn on the same graph of spectral displacement, S&, versus
spectral acceleration, SA. These curves are also based on many different assumptions. Finally, the
point at which they intersect is found through an algorithm supplied to the author by the
developers of the HAZUS earthquake loss estimation methodology. This point is the spectral
acceleration and spectral displacement pair (SA, SD), which defines the earthquake demand at the
threshold of the structural damage state of interest.
While the details to arrive at this point are copious, it is important to given an overview
of the approach and the assumptions in the development of the demand and capacity curves for
completion. The reader is referred to NIBS (1997) for an exhaustive account of the methodology.
So, first, the details on the development of the demand spectrum are given. Then, the assumptions
of the parameters to define the capacity curves before and after mitigation are proposed. Finally,
the parameters defining the structural fragility curve in HAZUS, SD,ds and P s ^ are
presented.
As indicated in Chapter Three, a standardized response spectrum has four parts: peak
ground acceleration, PGA, a region of constant spectral acceleration, SA, a region of constant
spectral velocity Sv, and a region of constant spectral displacement,^ (See Figure 3.5). The PGA
values for the first part of the curve are those developed from the structural engineers opinions.
Specifically, these are given in Table 4.19 and they are labeled /Jpga- The regions of constant
spectral acceleration, constant spectral velocity, and constant spectral displacement are developed
using Equations 4.8 through 4.10, respectively.
132
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(4.8)
s A\i
By ( Bef f )
TAVp < T ^ T vd
(4.9)
S AM TVP
R y ( B r y n )
(4.10)
In this set of equations, SA[7] is the demand spectral acceleration. The demand spectra
curves are developed for spectral acceleration values and the spectral displacement values are
calculated using equation 3.8 in Chapter Three (i.e. SD[T] = 9.8- SA/ 7 ] T2). Additionally, the
pertinent parameters used in equations 4.8 through 4.10 (RA, Ry, and Bcg) are mathematically and
qualitatively described in Table 4.20.
In the first equation (4.8), the region of constant spectral acceleration is calculated based
on the 0.3-second spectral acceleration of the elastic spectrum for site class/, S^,, and modified
for effective damping, Brff. This part of the curve is limited to periods less than the TAip, which is
the transition period from constant spectral acceleration to constant spectral velocity. TAyp is
calculated using equation 4.11. In equation 4.11, the transition period is a function of the effective
damping at that period, B TA\p- Recall that TAy which is the transition period between 5%-damped
constant spectral acceleration and 5%-damped constant spectral velocity for site class/, was first
introduced in Chapter Three (Equation 3.9).
133
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.20 Response Spectrum Demand Parameter Values
Parameter Description Equation (see NIBS 1997)
Effective damping, defined as
the sum of elastic damping
and hysteretic damping
D , Area .
Be +K- (------------ )
I tc- D A
R a
Acceleration-domain (short-
period) reduction factor
2.12
3.21- 0.68 ln(fl)
Ry
Velocity-domain (1.0-second
spectral acceleration)
reduction factor
1.65
2.31- 0.41 ln(fl)
Ka ( B t a v / 3 ) s ..........
TAVp = AVi ( ; r ) (4-11)
Ryi Bj AVp)
In the second equation of the set (4.9), the region of constant spectral velocity is
calculated. It is a function of the elastic 1.0-second spectral acceleration for site class/, SAi
modified for effective damping, Btg. This part of the curve is limited to the period TiD, where
constant spectral velocity transitions to constant spectral displacement. The period 7V/j was
defined in Chapter Three (Equation 3.10) and it is independent of effective damping.
Finally, the third equation of the set (4.10) defines the region of constant spectral
displacement. Similar to the constant spectral velocity region, it is a function of the elastic 1.0-
second spectral acceleration for site class / \ Sah- But, it is modified for the damping, Bud, which
is the effective damping at the transition period, TyD(usually 10 seconds).
The response spectra of demand developed for this work, for the four damage states of
interest (Slight, Moderate, Extensive, and Complete), both before and after mitigation, are
graphically displayed in Figure 4.12.
134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Complete (After)
1.4
Complete (Before)
BC
1.2 y
e
o
Extensive (After)
Extensive (Before)
08 /ySEiJPK.
2 0 . 6 ^ / M o d t m t f ( B r f B ^ '
W
u
&
Vi
0.4 / SligfcKAfte?
Slight (Befot?J'
m i i w u w . .
5 0 10 15
Spectral Displacement, inches
Figure 4.12 Response Spectra Demand Curves
Now, with these demand spectra for the four damage states of interest, building capacity
curves must be developed. The parameters used to create the capacity curve for a low-rise wood
frame structure (i.e. W1 in HAZUS) before and after mitigation are given in Table 4.21. These
values are consistent with those used for the capacity curves designed to the Low-Inferior (LI) or
Pre-Code level before mitigation and designed to Moderate-Code (MC) after mitigation (NIBS
1997 Tables 5.4 through 5.7(d)). With these parameter values, the points on the curve at yield
capacity, (A}>D}), and ultimate capacity, (A* Du), are defined (See Section 3.2.2.1). These are
also given in Table 4.21; a graph of the capacity curves before and after mitigation is shown in
Figure 4.13.
Table 4.21 Capacity Curve Parameter Values
Fragility
State
c , Te
(seconds)
Height
(feet)
0t|
Y
k
(Ay, Dy) (A, D)
Before Mitigation 0.1 0.35 14 0.75 1.5 3 6 (0.2, 0.24) (0.6, 4.32)
After Mitigation 0.15 0.35 14 0.75 1.5 3 6 (0.3, 0.36) (0.9, 6.48)
135
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1 4 '_______________________________________ Capacity(Before).
Capacity(After)
1.2
1 0.8
<
0.6
I 0.4
Vi
0.2
5 10 15 0
Spectral Displacement, inches
Figure 4.13 Capacity Curves
Finally, the building capacity curves are overlaid on the same graph as the response
spectra demand curves. The points of intersection on these curves are given in Table 4.22 for the
four damage states, both before and after mitigation. An example of the intersection for the case
before mitigation is graphically shown in Figure 4.14.
Table 4.22 Capacity Curve Parameter Values
Slight Moderate Extensive Complete
(SA, Sq) Before (0.29, 0.43) (0.52, 2.06) (0.60, 4.15) (0.60, 9.19)
(SA, So) After (0.37, 0.46) (0.71,2.27) (0.90,5.91) (0.90, 8.64)
136
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1.6
-Extensive
-Complete
1.4 -
Before Mitigation
Capacity
- - Slight
Moderate
0 5 10 15
Spectral Displacement, inches
Figure 4.14 Intersection of Capacity and Response Spectra Demand Curves
This intersection, shown above, gives the median value of spectral displacement at which
the low-rise wood frame structures reaches the threshold of damage state ds, designated S D,ds
But, one last piece of information is needed to complete the HAZUS structural fragility curves:
the beta term, P s ^
Recall the formulation of the structural fragility curves in HAZUS. First given in
equation 3.14, it is rewritten in equation 4.12 to emphasize its importance. This formulation
defines the conditional probability o f being in, or exceeding, a particular damage state,ds. given
the spectral displacement, SD. The beta term, which is the standard deviation of the natural
logarithm of spectral displacement for damage state, ds, must be derived.
/ W S D] = <D[ (4.12)
137
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The PSfc term ' n equation 4.12 is a function of the uncertainty in the capacity curve,(3C,
the uncertainty of the demand spectrum, /?, and the uncertainty in the median value of threshold
damage, Pm (Sc{s ) (Equation 4.13 rewritten from Equation 3.16).
fiSjs =J(COW[/?c ,/1d J D,5(& ])2 +(PM( S d s ) ) 2 (4.13)
Using default HAZUS values for P s ^ ^ Pm (Sjs ) die CONV (convolution) term in
equation 4.13 is calculated for the Low-Inferior designed buildings and the Moderate-Code
designed buildings (See Section 3.2.3.1). Then, with Pm (s c/s )^ describing the uncertainty in the
median value of structural damage from the survey data (i.e. crPCA in Table 4.19), new overall
uncertainty values for P s ^ are developed. The final parameter values used for SD,ds and
Ps ds ^or fur damage states, before and after mitigation, are given in Table 4.23.
Furthermore, an illustration of the after mitigation curves are presented in Figure 4.15.
Table 4.23 Structural Fragility Curve Parameters
Parameters Slight Moderate Extensive Complete
S D , d s (Before)
0.43 2.06 4.15 9.19
P S d s (Before)
0.84 0.90 1.00 0.97
S D.ds (After)
0.46 2.27 5.91 8.64
/ ?s<&(After)
0.85 0.92 1.00 0.93
138
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.8
O
C/5
- 0.6
O'
</]
A 0.4
Slight Damage
Moderate Damage .
Extensive Damage
Complete Damage
Q.
0.2
After Mitigation
5 15 0 10 20 25
Spectral Displacement
Figure 4.15 Structural Fragility Curves
4.3.7 Discussion
Incorporating expert opinion on the benefits of mitigation into the HAZUS earthquake
loss estimation model is a very data-intensive process. In the approach presented in this chapter,
many assumptions were made to achieve it. For example, a relationship between Modified
Mercalli Intensity (MMI) and peak ground acceleration (PGA) is assumed (Figure 4.11 and Table
4.18), and an elastic reference spectrum shape is chosen for construction of the response spectra
demand curves. Therefore, the process, though reasonable and logical, is not exact. And, in this
first application, it produced some results that were different than anticipated.
Specifically, note that in Table 4.23, at the Complete damage state, the threshold of
spectral displacement after mitigation, SD,ds (After), is actually slightly less than the threshold of
spectral displacement before mitigation, So,ds (Before) (i.e. 8.64 inches versus 9.19 inches). At
first glance, this seems counter intuitive. The threshold of spectral displacement should increase
139
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
after mitigation. Therefore, an explanation is warranted and a defense of the validity of the
approach is needed.
In this discussion section, three separate issues are raised in regard to this approach and
result. First, the validity of the threshold values and their associated standard deviations or beta
terms, ' s argued. Second, the sensitivity of the threshold of spectral displacement, S D.ds,
to the capacity and reference spectrum parameters is discussed. Finally, the overall impacts of
these parameter assumptions {i.e. So,ds and P s ^ ) on ^ e direct economic losses are foretold.
First, using this method, the spectral displacement values at all of the damage state levels
are generally consistent with HAZUS. In particular, the thresholds for the Slight and Compete
damage states are comparable. Recall that the thresholds of spectral displacement at the Slight
and Complete damage states for buildings designed to Moderate-Code are 0.50 inches and 9.45
inches, respectively (Table 3.9). Using the approach presented here, the results indicate SD,ds~
0.43 and 0.46 inches for Slight damage (before and after mitigation, respectively) and SD,ds~
9.19 and 8.64 for Complete damage (before and after mitigation, respectively).
Furthermore, this particular result {i.e. So,ds = 9.19 inches at the Complete damage state
before mitigation and 8.64 inches at the Complete damage state after mitigation) is not
completely unreasonable. One could argue that the same range of spectral displacement {i.e.
around 9 inches for Complete damage noted above) is needed to reach the Complete damage state
either because the cripple wall collapses (before mitigation) or because a structure with adequate
bracing at the cripple wall level is simply racked too much in a big event (after mitigation).
Others would most likely agree with this statement. Recall from an earlier comment cited from a
respondent to the expert opinion survey:
140
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Once you get above a [MMI] X, then the superstructure will probably fail and
you are back to having a lot of damage even though the cripple wall was braced
and anchor bolts added.
Additionally, the original result for the other parameter in the structural fragility curve,
PSds 's a*so a reasonable one. Recall that P s ^ ' s die standard deviation of the natural
logarithm o f spectral displacement. It is an indication of the spread of the lognormal distribution
from the median. If a higher value is used for this beta term, it indicates that the probability of
achieving the damage state at values of spectral displacement less than the lognormal curves
median value is greater than if a smaller beta term was used. This is why, in general, higher beta
terms are used for buildings designed to Pre-Code (Low-Inferior) levels. In the structural fragility
curves developed here, P s ^ = 0-97 before mitigation and P s ^ ~ 0 93 after mitigation at
Complete damage.
Second, recall that two different capacity curves were used to intersect with two sets of
response spectra demand curves: one before mitigation, reflecting Low-Inferior (LI) seismic
design, and one after mitigation, reflecting Moderate-Code (MC) seismic design. This choice was
made to be consistent with the assumptions in the HAZUS default fragility curves before and
after mitigation {i.e. as they are described in the methodology). Moreover, it is intuitive to
increase the capacity of the structure after mitigation {i.e. after bolting the structure to its
foundation and bracing its cripple wall).
Therefore, the capacity of the structure is increased from Dy = 0.24 to Dy = 0.36 inches (at
yield) and from Du = 4.32 to Du= 6.48 inches (at ultimate) after the structure is mitigated. And,
this after mitigation capacity curve will intersect the same demand spectrum at a lower value of
spectral displacement than the before mitigation capacity curve.
141
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Moreover, two sets of demand spectra were developed from the same elastic reference
spectrum shape, representing ground motion of a moment magnitude 7.0 earthquake for soil site
class D and site-to-source distances of fifteen kilometers (at the closest distance). Using these
assumptions, the spectral acceleration response factor used, denoted S^PGA in the HAZUS
methodology, was 2.1 (NIBS 1997 Table 4.2). And, as previously mentioned, the reference
spectrum shape is quite sensitive to the assumptions in magnitude, soil, and site-to-source
distances, and hence, this term also.
Therefore, if different parameters were chosen to construct the reference spectrum (i.e.
different magnitude or distance), the points at which the capacity curves and the demand spectra
intersect would be different. And, the threshold of spectral displacement at the Complete damage
state could increase after mitigation. For example, if the assumption of site-to-source distances
was changed to forty kilometers, the spectral acceleration response factor, Sa^PGA, noted above
would change to 2.2. And, the intersection of the capacity curves with the new reference demand
spectra results in So,ds = 9.52 inches before mitigation and SD,ds ~ 9.54 inches after
mitigation for the Complete damage state.
But, finally, the real issue is how the threshold of spectral displacement in the Complete
damage state impacts the final results on the benefits of mitigation (i.e. the direct economic
losses). It is not too significant. First, despite the lack of increase in the threshold of spectral
displacement in the Complete damage state, the computed average annual loss (AAL) estimate
decreased significantly after mitigation. Looking closely at the structural losses per scenario
event, the structural losses do increase slightly in some of the events. But, this negative effect
comprises less than 1% of the total structural loss (in absolute terms).
142
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Additionally, recall that the fragility curves developed in this section are for structural
damage to residential structures only. Two other types of fragility curves exist in the HAZUS
methodology to estimate damage to the nonstructural drift-sensitive components and damage to
the nonstructural acceleration-sensitive components (See Section 3.2.3.1).
A decision was made early on in this work to incorporate the results of the expert opinion
surveys {i.e. MDF before and after mitigation) as a reduction in structural damage. The
nonstructural fragility curves were not changed due to the assumptions in the HAZUS
methodology. Specifically, the nonstructural damage drift ratios for nonstructural drift-sensitive
damage are assumed to be the same across all model building types and seismic design levels.
Furthermore, the nonstructural damage acceleration values for nonstructural acceleration-
sensitive damage are assumed to be the same across all modeling building types.
Looking ahead at this studys results (in Chapter Six), the breakdown of losses to
residential structures indicate that the damage to nonstructural components of the building {i.e.
CNS in Figure 4.16) comprise over sixty-percent (61.8%) of the total loss, on average. Damage to
the structure itself {i.e. CS in Figure 4.16) is approximately twelve-percent (11.5%) of the total
loss, on average. Therefore, the impact of structural damage is much less in comparison to
nonstructural damage. This will have implications for the homeowners cost-benefit analysis for
structural mitigation (Section 6.2).
143
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
70%
; 50%
2
I 40%
30%

5 20%
Type of Lou
Figure 4.16 Breakdown of Residential Losses
144
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5 PARAMETERS IN SENSITIVITY ANALYSIS
With the structural fragility curves developed in Chapter Four, this chapter outlines the
default and updated parameters and models used in the sensitivity analysis of the earthquake
loss estimation (ELE) methodology. Arguably the most vital portion of this study, it is important
to outline all of the assumptions in order to avoid misunderstanding. As indicated in Chapter
Two, an ELE analysis can become very complicated, so all the details are needed to completely
understand the results of the analysis. Further, while the HAZUS software is utilized, with a well-
documented earthquake loss methodology (as outlined in Chapter Three), the additional
information used to define and update parameters and models are critical. Therefore, all the
specifics of the assumptions are given here.
Overall, there are six areas of uncertainty considered in the analysis. Three areas of
uncertainty are part of the seismic hazard estimation portion of the earthquake loss estimation
methodology. The other three areas of uncertainty are part of the inventory, damage, and loss
portion of the ELE methodology. Specifically, a number of parameters and models in both
portions of the methodology are updated from their default values in HAZUS as part of this
sensitivity analysis. The ultimate goal is to ascertain the Range of Uncertainty of Loss, L, over the
parametric and modeling assumptions, jf, denoted RU[L(X)]. As indicated in Chapter One, the
two information states, {xo, */}, denote the default state of information and the updated state
of information for each area of uncertainty considered in the analysis (Equation 5.1 rewritten
from Equation 1.3).
RV[L(X)] = \L(xx) - L(x0)\ (5.1)
The default and updated information states are grouped into choices of seismic hazard
estimation parameters and inventory, damage, and loss estimation parameters in order to
145
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
simplify the description of the process (Figure 5.1). For seismic hazard estimation, the parameters
and models include the annual recurrence of earthquake events, attenuation relationship, and soil
mapping schemes. For inventory, damage, and loss estimation parameters, the parameters and
models include the building inventory exposure and fragility of low-rise wood frame residential
structures before mitigation and after mitigation.
Earthquake
Recurrence
Inventory
Exposure
(Square
Footage)
SoQ
Mapping
Schemes
~ Fragility r
(Before and
After
.MhigationVj
Calculate Economic Loss
(Expected Loss or WCL
to Insurer and Owner)
Define the Earthquake Hazard
(Seismic Sources, Recurrence,
Attenuation, Soils)
Estimate Inventory Damage
(Through Historical Loss Data,
Engineering Data, Expert
Opinion)
Define the Inventory Characteristics
(Structure Location, Value, Year Built,
Construction Class, etc.)
Seismic Hazard Estimation Inventory, Damage and Loss
Parameters Estimation Parameters
Figure 5.1 Scope of Sensitivity Analysis
In this study, the assumption is that the areas of uncertainty considered in the sensitivity
analysis are independent of one another. Specifically, the choice of seismic hazard estimation
parameters and models are independent of one other. The recurrence of earthquake events on
seismic fault sources is independent of the ground motion attenuation relationship chosen. And,
these are independent of the soil mapping schemes used. Similarly, the choice of inventory,
damage, and loss estimation parameters and models are independent of one another. The
exposure inventory of the residential building stock, in terms of the total square footage, is
independent of the vulnerability of the low-rise wood frame residential structures before
146
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
mitigation and after mitigation. Finally, the seismic hazard estimation parameters and models are
independent of the inventory, damage, and loss estimation parameters and models.
So, the sensitivity analysis completed assumes independence and utilizes a logic tree
approach to keep track of the assumptions in the study. This approach is similar to previous
studies completed on the uncertainty in seismic hazard analysis (Coppersmith and Youngs 1986),
as well as the California Department of Conservation's Division of Mines and Geology (CDMG)
studies summarized in Chapter Two (Peterson et al. 1996a; Cao et al. 1996; Cramer et al. 1996).
As others have discovered, it is appropriate and intuitive to try to estimate the most influential
parameters in the process in this way. Furthermore, with the sensitivity analysis combining both
parametric and modeling uncertainties, a logic tree approach will easily keep track of the process.
The remaining portions of this chapter outlines the default and updated information states
on the six areas of uncertainty considered in the analysis, shown graphically through logic trees.
First, the default and updated states of seismic hazard estimation parameters and models are
discussed. Then, the default and updated states of inventory, damage, and loss estimation
parameters and models are summarized. In all, there are sixty-four branches on the logic tree {i.e.
[2J for seismic hazard estimation]- [23for inventory, damage, and loss estimation] = 64).
5.1 Seismic Hazard Parameters
First, one logic tree is constructed that keeps track of the default and updated choices in
earthquake recurrence, attenuation curves, and soil mapping schemes in the sensitivity analysis of
the seismic hazard estimation parameters and models (Figure 5.2). Specifically, for the seven
strike-slip fault sources in the Oakland California region (outlined in Chapter Four), default and
updated magnitude-frequency relationships are tested. The default recurrence parameters are
calculated from published United States Geological Survey (USGS) data (Peterson et al. 1996b),
147
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and the updated recurrence parameters are obtained from data supplied by Risk Management
Solutions, Inc. (RMS). Then, the forty-six events in the earthquake catalogue on the seven strike-
slip faults are run through HAZUS assuming two different ground motion attenuation
relationships in the Western United States, discussed in Section 3.1.3. The default relationship is
the Boore, Joyner, Fumal (BJF) relationship (1993, 1994) for the Western United States (WUS).
The updated equation is the Project97 relationship, outlined in NIBS (1997).
Finally, the last seismic hazard estimation parameter considered is the soil mapping
scheme in the Oakland, California region. The default scheme is the soil site classifications, as
well as the liquefaction and landslide potential, assumed in the default mode of the HAZUS
software. The updated scheme, provided by Risk Management Solutions, Inc., incorporates new
information on the soil site classes and their liquefaction and landslide potential. Considering
these states, there are eight resulting branches of the tree (Figure 5.2).
<
Default Soils Map
<
BJF Attenuation
More Detailed
Soils Map
Default Soils Map
Earthquake Recurrence
based on USGS Data^
Fault |
Source
<
<
Project97 Attenuation
BJF Attenuation
More Detailed
Soils Map
Default Soils Map
More Detailed
Soils Map
Default Soils Map
Earthquake Recurrence
based on RMS Data ^
Project97 Attenuation
More Detailed
Soils Map
Figure 5.2 Logic Tree for Seismic Hazard Parameters
148
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.1.1 Earthquake Recurrence
In varying the recurrence of certain magnitude events on the faults considered in the
seismic hazard analysis, the uncertainty in the time between significant earthquake events is being
considered. This recurrence relationship, denoted f(m) in Chapter Two, contains both aleatory
{i.e. random) and epistemic {i.e. lack of knowledge) uncertainty. Due to the epistemic uncertainty,
it is assumed that this distribution can be updated with new information.
For the seven strike-slip faults considered in this analysis, the development of the default
magnitude-frequency relationship is a combination of an exponential relationship and a
characteristic earthquake model. This approach is similar (although not exact) to previous studies
estimating the probabilistic seismic hazard in California published by the USGS (Peterson et al.
1996b; Frankel et al. 1996). The exponential magnitude distribution partitions the rate of seismic
release on the fault into events between a minimum magnitude,m0, and maximum magnitude, mu,
considered damaging to the building stock of the region. The characteristic earthquake model
assumes a characteristic or same-size earthquake frequency density, banded at or near the
maximum magnitude event considered on the fault source. Using both models allows for the
consideration of larger earthquakes than predicted by the exponential distribution and different
(smaller) earthquakes than predicted by the characteristic model. This fully characterizes the
behavior of the fault sources.
First, though, the fault segment and property information needed to develop the
recurrence parameters is outlined (Table 5.1). These include the fault name, the maximum
moment magnitude of the fault segment, Mmax, and the associated mean return period, T, along
with the length, L, width, W, and slip rate, 6 of the fault.20
20For further explanation of these parameters, see Section 3.1.1.
149
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 5.1 Seismic Fault Source Parameters
Fault
Name
T
(years)
L
(km)
W
(km)
8
(mm/yr)
M$x 10IS
(Nm/yr)
South Hayward 6.9 167 43 12 9 140
North Hayward 6.9 167 43 12 9 140
Rodgers Creek 7.0 222 63 10 9 170
San Andreas Peninsula 7.1 400 88 14 17 630
Calaveras 6.8 146 52 13 6 120
Concord-Green Valley 6.9 176 66 12 6 140
Greenville 6.9 521 73 11 2 48
The seven faults in the table are listed according to their category. The class A faults are
listed first (eg. South Hayward, North Hayward, Rodgers Creek, San Andreas Peninsula). The
class B faults are listed next (eg. Calaveras, Concord-Green Valley, Greenville). Class A faults
have well-constrained slip rates and class B faults are poorly constrained (See Section 3.1.1).
Furthermore, from the information in Table 5.1, the rate at which the fault accumulates moment,
Mq , is calculated (Equation 5.2). In equation 5.2 (rewritten from equation 2.2),// represents the
modulus of rigidity of the earths crust (3 x 10 dyne/cm2s), 8 is the slip rate of the fault, and/i is
the rupture area. The rupture area is simply rewritten as the rupture length,/., times the width of
the fault, W.
= \ l - 8 -A = \ i - 8 - L- W (5.2)
Now, recurrence for the characteristic event on each of the faults and the exponential
recurrence for magnitude events less than the characteristic event can be calculated. First, the
characteristic rate of earthquake recurrence is determined. For the seven fault sources listed
above, the characteristic rate is simply the reciprocal of the mean return period, T. Note here that
no difference between class A faults and class B faults is made. While class A faults are assumed
to rupture in a characteristic manner, class B faults do not. The possibility that they can rupture
150
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
their full length, however, is assumed. A listing of the characteristic recurrence for each seismic
fault source is given in Table 5.2.
Table 5.2 Characteristic Recurrence
Fault
Name
T
(years)
1 IT
(annual)
South Hayward 167 0.00599
North Hayward 167 0.00599
Rodgers Creek 222 0.00450
San Andreas Peninsula 400 0.00250
Calaveras 146 0.00685
Concord-Green Valley 176 0.00568
Greenville 521 0.00192
The exponential rate of earthquake recurrence for magnitude events less than the
characteristic event is a bit more complicated. In this work, the exponential distribution estimates
the number of events between a minimum and maximum moment magnitude considered
damaging to the low-rise wood frame residential structures. As in the case of characteristic
recurrence, all fault sources assume this model of recurrence below the characteristic magnitude.
In other words, the class B faults are assumed to rupture either their full length or a portion of
their length equally. And, the class A faults can rupture only part of their length.
A note should be made here about the USGS publication noted earlier: the 1996 USGS
National Seismic Hazard maps(Frankel et al. 1996). Within this documentation for the maps, the
class A faults and class B faults are treated separately for magnitude events less than the
characteristic magnitude. Class A faults use the characteristic magnitude event for rupture and
class B faults are a combination of a characteristic maximum magnitude event with a truncated
Gutenberg-Richter recurrence for magnitudes between M = 6.5 and the maximum magnitude
event. These recurrence events are combined with gridded seismicity data of events less than a
moment magnitude of 7.0 in order to develop the seismic hazard maps in the Western United
151
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
States. The reader is referred to Frankel and others (1996) or the online documentation for details
on how this data were developed.
For the fault sources in this work, however, a different approach is taken.21 An
exponential distribution of recurrence is used for ail the faults, with a minimum moment
magnitude of 4.75 and a maximum moment magnitude of 6.75. Then, an incremental rate of
earthquakes in 0.5 increments can be developed, centered at 5.0, 5.5, 6.0, and 6.5. The equation to
determine this incremental a-vaiue for each fault is given in Equation 5.3 (rewritten from
Equation 3.5). As indicated in Section 3.1.2, the 6-value is taken as 0.9, c is 1.5 and d is 16.1.
( c - b ) Af Z In(10)
^ 10*(l0(c- fe),w io^c - ^ ,Wo )^
With this calculated a-value and estimation of the b-value (0.9), the frequency of
earthquake events is estimated using the Gutenberg-Richter relationship (Equation 5.4). A full
listing of the annual earthquake recurrence for the four magnitudes considered is shown in Table
5.3. Additionally, a full listing of the default frequency data (characteristic and exponential) is
given in Appendix 8.10.
N(m) = \ 0 a-bM (5.4)
Table 5J Exponential Recurrence
Fault Name M = 5.0 M = 5.5 M = 6.0
1
2
i
i
9
s
I
A
South Hayward 0.06862 0.01220 0.00217 0.00039
North Hayward 0.06862 0.01220 0.00217 0.00039
Rodgers Creek 0.08378 0.01490 0.00265 0.00047
San Andreas Peninsula 0.30946 0.05503 0.00979 0.00174
Calaveras 0.05993 0.01066 0.00190 0.00034
Concord-Green Valley 0.07021 0.01249 0.00222 0.00039
Greenville 0.02373 0.00422 0.00075 0.00013
21The author does not claim to be a seismologist and believes that this is a reasonable approach.
152
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The updated earthquake recurrence is developed from proprietary data supplied by Risk
Management Solutions, Inc. Specifically, a catalogue of surface wave magnitude events, Ms, with
their corresponding annual recurrence was used to develop the recurrence parameters. The
conversion of Ms to moment magnitude, M, was completed using the relationship graphically
shown in Appendix 8.1. Implementation of this update is accomplished through changing the
annual recurrence values in the output file from HAZUS before inputting the data into the EP
Maker.
A representation of the annual recurrence for the South Hayward and the Peninsula
segment of the San Andreas fault is shown in Figures 5.3 and 5.4, respectively. Notably, the
recurrence of M = 6.5 events on the Hayward fault is conspicuously less in the default scheme
developed from the USGS data. This is due to the assumptions on the parameters in the
exponential distribution. Additionally, this is important to note because this is the case in the
majority of the faults (eg. South Hayward, North Hayward, Rodgers Creek, Calaveras. Concord-
Green Valley, and Greenville). Its effect on the final estimates of loss may or may not be
significant.
As for the characteristic recurrence, the default and updated schemes match quite well at
the Mmax value in Table 5.1 (i.e. M = 6.9 on South Hayward and M = 7.1 on San Andreas
Peninsula). Notably, this characteristic recurrence is used at the Mmaxvalue up to M = 7.5 in the
default scheme. In this way, there is a band of uncertainty above the A/muc magnitude (See listing
of events in the Scenario Builder Input File in Appendix 8.4).
153
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Moment Magnitude
0.1
6.0 7.0
0.01
1 0.001
Default (Exponential)
Default (Characteristic)j
Updated
0.0001
Figure S3 Magnitude-Frequency Relationship for South Hayward Fault
Moment Magnitude
0.01
7!5
Default (Exponential)
Default (Characteristic):
Updated
(J
e

3
3*
*
b
m
3
e
e
<
0.001 J
Figure 5.4 Magnitude-Frequency Relationship for San Andreas Peninsula Fault
154
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.1.2 Ground Motion Attenuation
In defining the seismic hazard, the second key ingredient tested in this study is the
assumed relationship between site to source ground motion. Known as the attenuation
relationship, it is the estimation of how the seismic wave propagates from its source. As
previously discussed in Section 3.1.3, there are four choices of attenuation relations for shallow
crustal earthquakes in the western United States (WUS) in the HAZUS software. Since the list of
earthquake events in the Scenario Builder Input File (in Appendix 8.4) are shallow (i.e. depth = 5
km), these relationships are chosen.
Of the four relationships, three are unique relationships. These are the Boore, Joyner,
Fumal (BJF) relationship (1993, 1994), the Sadigh, Chang, Abrahamson, Chio, and Power
(SCACP) relationship (1993), and the Campbell and Bozorgnia (CB) relationship (1994). The
fourth one, labeled the Project97 relationship (NIBS 1997), is a linear combination of the
previous three, where applicable (See Section 3.1.3).
For this work, the default attenuation relationship is the Boore, Joyner, and Fumal (BJF)
relationship. This relationship is chosen for two reasons. First, it empirically estimates both peak
ground acceleration (PGA) and spectral acceleration (5^). For this study, this is a requirement for
the choice in attenuation. For example, the Campbell and Bozorgnia (CB) relationship is not
chosen because it only estimates PGA.
Second, the BJF relationship is the lower bound on ground motion estimated within small
distances from the source in the majority of the cases. Specifically, it is the lower bound for
estimates of PGA at close distances from the source, and in most cases, it is the lower bound for
estimates of 0.3-second and 1.0-second spectral acceleration at close distances from the source.
Additionally, between the BJF relationship and SCACP relationship, it is the better choice in
terms of ground motion difference from the updated choice of relationship.
155
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The Project97 relationship is the updated choice. In updating the model, it is a linear
combination of the other three unique relationships to estimate peak ground acceleration, PGA,
and a linear combination of the BJF and SCACP relationship to estimate spectral acceleration,^.
The 0.3-second spectral acceleration as a function of distance from the source for both the default
(BJF) and updated (Project97) choices in relationships is given in Figure 5.5. As mentioned, in
the figure, the updated" Project97 attenuation curve estimates a higher ground motion at
distances close to the source.
distance, km
1 10 100
1 . ______
0.1
BJF
Project97
0 . 0 1 -----------------------------------------------------------------------------------------------
Figure 5.5 Attenuation Relationships
5.1.3 Soil Mapping Schemes
The final area of uncertainty considered in the seismic hazard estimation portion of the
logic tree is the uncertainty in the soils. In the HAZUS software, the default soils mapping
scheme is the same in all regions of the Eastern and Western United States (EUS and WUS). Soil
site class D, as categorized by the NEHRP guidelines (FEMA 1997), is assumed. Additionally, no
liquefaction or landslide potential is assumed anywhere in the region. So, for this study, the
156
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
default mapping scheme is this same scheme {i.e. soil site class D, no liquefaction or landslide)
for the 108 census tracts in the Oakland, California region (Figure 5.6 Default). In the HAZUS
software, *4 denotes soil site class D and O denotes no liquefaction or landslide potential (See
Appendix 8.11).
The updated scheme is described as a more detailed soils map (Figure 5.6 Updated). As
in the case of the recurrence parameters, Risk Management Solutions, Inc. supplied this
information. This updated information is more detailed because there is a range of soil site classes
considered in the region, as well as a range of liquefaction and landslide potentials considered.
Specifically, the updated scheme has soil site classes B, D, and E. As expected, soil site
class E, which indicates soft soils, borders the San Francisco Bay (See Figure 4.1 in Chapter
Four). Soil site class D is within the center of the region, indicating stiff soils, and soil site class B
is assumed throughout the eastern (inland) part of the region, indicating rock conditions. Within
the HAZUS methodology, 2 denotes soil site class B \ 4 ! denotes soil site class D and 5
denotes soil site class E (See Appendix 8.11).
In the updated scheme, liquefaction and landslide potential also have a range of values.
Liquefaction potential is in one of four categories in the region: very low, low, medium, or
very high. As expected, the very high potential for liquefaction is near the San Francisco Bay
on soil site class E. The very low potential for liquefaction is in the eastern part of the region on
soil site class B. The low and medium potential for liquefaction is in between these two areas
(Figure 5.7). In the HAZUS methodology, M indicates very low liquefaction potential, 2
indicates low liquefaction potential, 3 indicates medium liquefaction potential, and 5
indicates very high liquefaction potential (See Appendix 8.11).
157
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Default
Updated
Figure 5.6 Soil Mapping Schemes
Figure 5.7 Updated Liquefaction Potential
Finally, in the updated scheme, landslide potential is considered. As defined in the
HAZUS methodology, the potential for landslide has a wider range of values (eg. eleven
categories versus six categories for liquefaction potential). In this scheme, there is a potential for
susceptibility category II, IV, VI, and VIII (Figure 5.8). The lower categories have less
158
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
potential for landslides in the region. As expected, landslide susceptibility category VIII is in
the northeastern portion of the region and landslide susceptibility category II borders the San
Francisco Bay. In Appendix 8.11, 2 denotes landslide susceptibility category II, 4 denotes
landslide susceptibility category IV, 6 denotes landslide susceptibility category VI, and 8
denotes landslide susceptibility category VIII.
Figure 5.8 Updated Landslide Potential
With these updated schemes of some liquefaction and landslide potential everywhere in
the Oakland, California region, the probability of damage to the residential building stock is
increased. Specifically, ground failure, in terms of liquefaction, landsliding, or surface fault
rupture, is quantified by permanent ground deformation (PGD) in the HAZUS methodology. In
essence, the probability of damage due to permanent ground deformation is added to the
probability of damage due to ground shaking for calculations o f loss (See NIBS 1997 for more
details).
159
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2 Inventory, Damage, and Loss Parameters
The second logic tree constructed for this work includes the parameters and models tested
in the inventory, damage, and loss portion of the HAZUS earthquake loss estimation
methodology. As in the case of the seismic hazard estimation parameters, this logic tree approach
keeps good track of the default and updated states of inventory exposure, building fragility, and
the condition of mitigation in the sensitivity analysis (Figure 5.9).
For the residential building stock in the Oakland, California region, the first part of this
logic tree is the inventory exposure file, in terms of the assumed square footage of the low-rise
wood frame residential structures. The default state of information is based on the 1990 census
data and the updated state of information is developed from the Alameda County tax assessors
data.
Figure 5.9 Logic Tree for Inventory, Damage, and Loss Estimation Parameters
The second and third parts of the inventory, damage, and loss portion of the HAZUS
methodology being tested are the fragility curves and structural mitigation. In the default state,
Fragility Curves
(HAZUS)
After Mitigation
Before Mitigation
Fragility Curves
(Survey)
After Mitigation
Before Mitigation
Exposure File based
on 1990 Census
(publicly available)
Residential
Inventory
Fragility Curve:
(HAZUS)
Before Mitigation
Exposure File based
on Tax Assessor Data
(BAHUG)
After Mitigation
Fragility Curve
(Survey)
After Mitigation
Before Mitigation
160
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the fragility curves before and after mitigation are based on the HAZUS default fragility curves.
In the updated state, the fragility curves before and after mitigation incorporate the new structural
fragility curves developed in Chapter Four. Mitigation entails bolting the structure to its
foundation and bracing its cripple wall, and it is applicable to pre-1940 low-rise wood frame
structures.
5.2.1 Inventory Exposure
There is a great deal of epistemic (i.e. lack of knowledge) uncertainty associated with the
exposure inventory of structures in the earthquake loss estimation process. Uncertainty in the
exposure value of residential structures is incorporated to reflect the tendency of the insurance
industry to misclassify and/or undervalue structures (See Section 2.4). With policy deductibles
and fixed coverage limits, if the building is undervalued, there is a higher chance the economic
loss will exceed the deductible, resulting in a higher loss to the homeowner. Additionally, from
the tax assessors data, or even from insurance data, assumptions are routinely made about the
structure type from occupancy class information, the replacement cost of a structure from
outdated material and labor cost information, and value of a structure from assessment or
coverage limit information.
In HAZUS, the inventory exposure is in terms of total square footage per census tract for
each occupancy class. Recall that this sensitivity analysis considers the uncertainty in the square
footage of residential structures classified as single family residences in the HAZUS methodology
(i.e. RES1 in HAZUS). As in the case of the soils mapping scheme, the default exposure is
taken from the same data in all regions of the eastern and western United States: the 1990 Census.
From the 1990 Census, there is a total of 123,783 thousand square feet of single family
residences in the region and 31,358 out of 82,522 structures can be mitigated (i.e. 38% of total
number of structures, where each structure is 1,500 square feet). The updated exposure is based
161
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
on the Alameda County tax assessors data supplied by Scott McAfee of the California Office of
Emergency Services. From this information, there is a total of 110,899 thousand square feet of
single family residences in the region and 28,094 out o f 73,933 structures can be mitigated.
Focusing on the density of residential square footage per census tract (Figure 5.10 and
Appendix 8.12), two observations should be noted. First, if the square footage is zero (0) in a
census tract in the 1990 census data, the square footage will most likely be zero (or ju s t slightly
higher) in the tax assessors data. In other words, in areas where there are no residential
structures, this is usually consistent across the two sets of data. Second, the data from the
Alameda County tax assessors office, on average, estimates a lower square footage per census
tract than the 1990 census data. This can be seen in the overall estimate of square footage for the
region (i.e. 110,899 versus 123,783 thousand square feet), as well as the breakdown o f square
footage per census tract (See Appendix 8.12).
Figure 5.10 Residential Exposure
5.2.2 Fragility Curves
Finally, the last two areas of uncertainty in the analysis are grouped together under the
heading of fragility curves. In performing a sensitivity analysis on the damage and loss portion of
162
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the earthquake loss estimation process, the vulnerability functions for the residential structures
under consideration for mitigation (i.e. pre-1940 low-rise wood frame single family residences)
are studied only. The remaining fragility curves for the other residential structures in the
inventory are not modified. The residential structural fragility curves can be modified in HAZUS
by manually changing the form of the cumulative lognormal fragility curves in the software.
Uncertainty in the damage to the residential building stock is reflected in an update in the
HAZUS fragility curves for structural damage to low-rise wood frame structures {i.e. W1 in
HAZUS). Specifically, So,ds. the median value of spectral displacement at which the building
reaches the threshold of damage state ds, and P s ^ >the standard deviation of the natural
logarithm of spectral displacement in damage state ds, are updated with new knowledge of
fragility. In other words, the structural fragility curves developed from the structural engineers
expert opinions are incorporated, which were presented in Section 4.3.6.4.
The default structural fragility curves are the ones for low-rise wood frame structures
currently in HAZUS. Curves to reflect damage before mitigation are those designed for a Low-
Inferior (LI) seismic hazard; curves to reflect damage after mitigation are those designed for a
Moderate-Code seismic hazard (See discussion in Section 4.3.7). The parameters, SD.ds and
PSds fr both the default (HAZUS) and updated (Survey) structural fragility curves, are given
in Table 5.4.
163
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 5.4 Structural Fragility Curve Parameters
Parameters Slight Moderate Extensive Complete
( S D,ds j fiSds )
HAZUS (Before)
(0.40, 1.01) (1.00, 1.05) (3.09, 1.07) (7.56, 1.06)
( S D,ds, 0 S& )
HAZUS (After)
(0.50, 0.84) (1.25, 0.86) (3.86, 0.89) (9.45, 1.04)
(SD,ds, P s ^ )
Survey (Before)
(0.43, 0.84) (2.06, 0.90) (4.15, 1.00) (9.19, 0.97)
(SD,ds, 0 S& )
Survey (After)
(0.46, 0.85) (2.27, 0.92) (5.91, 1.00) (8.64, 0.93)
Two additional notes are made here about the incorporation of the new structural fragility
curves and the assumptions about the nonstructural fragility curves. First, in order to incorporate
the updated structural fragility curves before mitigation, 38% of the low-rise wood frame
structures are placed in the Low-Inferior (LI) bin of the structure occupancy mapping scheme
(See Table 3.10 and Table 4.4(a)). And, the parameters that define the structural fragility are
updated. For example, ( SD,ds, P s ^ ) ls updated from (0.40, 1.01) for Slight damage in the
default state to (0.43, 0.84) for Slight damage in the updated state.
In order to incorporate the updated structural fragility curves after mitigation, the same
38% should be placed in the Moderate-Code (MC) bin of the structure occupancy mapping
scheme (See Table 4.4(b)). But, this is infeasible. The other 34 % (i.e. 72% - 38%) of the low-
rise wood frame structures that are not eligible for mitigation are already in the Moderate-Code
bin. And, their fragility will be the same both before and after mitigation (i.e. they are not
mitigated in the analysis).
Therefore, the 38% of low-rise wood frame structures eligible for mitigation are placed in
the Low-Code (LC) bin for analysis purposes. Since the default mapping scheme does not map
164
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
any other occupancy class of structures into the Low-Code design level for low-rise wood frame
{i.e. W1 in HAZUS), this is the obvious choice for the analysis. Additionally, the capacity curve
for low-rise wood frame structures designed to Moderate-Code is the same for the Low-Code
design level. In this way, there are now four mappings for low-rise wood frame residential
structures rather than three. And, the fragility curves for the Low-Code design level of the low-
rise wood frame structures can be manipulated to reflect mitigation {i.e. the S o . d s and
PSck terms for structural fragility can be changed to reflect the parameter values in Table 5.4).
Additionally, it should be reemphasized that the fragility curves for nonstructural drift-
sensitive damage and nonstructural acceleration-sensitive damage are not updated here. In both
the default and updated state of knowledge, the { S D,ds P s ^ ) parameters for drift-sensitive
damage and the ( S 4 ^ , P s ^ ) parameters for acceleration-sensitive damage are the same. Before
mitigation, the nonstructural fragility curves for buildings designed to the Low-Inferior (LI)
design and performance levels are used. After mitigation, the nonstructural fragility curves for
buildings designed to the Moderate-Code (MC) design and performance levels are used.
The parameter values to estimate nonstructural damage are contained in Table 5.5.
Additionally, Figures 5.11 and 5.12 are a graphical representation of the comparison of the
default and updated structural fragility curves for the low-rise wood frame structures in the
Oakland, California region. Notably, in the Moderate and Extensive damage states, these
structures are less susceptible to damage with the updated information.
165
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 5.5 Nonstructural Fragility Curve Parameters
Parameters Slight Moderate Extensive Complete
( SD,ds , flSds )
Drift-Sensitive (Before)
(0.50, 1.07) (1.01, 1.11) (3.15, 1.11) (6.30, 1.15)
(SA,ds , PSfc )
Acceleration-Sensitive (Before)
(0.20, 0.72) (0.40, 0.70) (0.80, 0.66) (1.60, 0.66)
(SD,ds, PSfc )
Drift-Sensitive (After)
(0.50, 0.89) (1.01,0.91) (3.15, 0.90) (6.30, 1.04)
{ S A,ds >PSrfs )
Acceleration-Sensitive (After)
(0.25, 0.73) (0.50, 0.68) (1.00, 0.67) (2.00, 0.64)
Slight (HAZUS)
Slight (Survey)
Moderate (HAZUS)
Moderate (Survey)
Extensive (HAZUS)
Extensive (Survey)
Complete (HAZUS)
Complete (Survey)
Before Mitigation
10 15 20 25 30
Spectral Displacement (inches)
35 40
Figure 5.11 Structural Fragility Curves (Before Mitigation)
166
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1
0.8
e
0.6
v
II
C, 0.4
2,
a.
0.2
0
0 5 10 15 20 25 30 35 40
Spectral Displacement (inches)
Figure 5.12 Structural Fragility Curves (After Mitigation)
167
Slight (HAZUS)
Slight (Survey)
Moderate (HAZUS)
Moderate (Survey)
Extensive (HAZUS)
Extensive (Survey)
Complete (HAZUS)
Complete (Survey)
After Mitigation
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6 RESULTS
This research is aimed to help advance the state-of-the-art in catastrophic risk modeling
and encourage the use of earthquake mitigation to reduce the impact of earthquake losses. Both
risk modelers and policy makers in the realm of earthquake engineering can use these results. The
results include the loss exceedance probability (EP) curves, average annual loss (AAL), and worst
case loss (WCL) estimates to the region under the various assumptions outlined in Chapter Five.
The conditions under which it is beneficial for the homeowner to mitigate and/or insure for
earthquake hazard in Oakland, California are established. Most importantly, the ranking of the
importance of various parameters and models on the uncertainty in the earthquake loss estimation
(ELE) in the region is given. The results in this chapter are presented by revisiting the four
questions outlined in Chapter One.
6.1 Range of Uncertainty
Which parameters/models give rise to the most uncertainty?
The parametric and modeling uncertainties considered in the analysis include seismic
hazard estimation parameters (recurrence of earthquake events, ground motion attenuation, soil
mapping schemes) and inventory, damage, and loss estimation parameters (exposure of
residential buildings and fragility of pre-1940 low-rise wood frame residential structures before
and after mitigation).
In order to rank the parameters and models by their influence on the ELE process, the
results presented are the loss exceedance probability (EP) curves for the entire low-rise wood
frame residential building stock in the Oakland, California region. Other occupancy classes are
not considered; other model building types are not considered. It should be noted that the
168
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
rankings presented here may well be different for other regions, other occupancy classes and
other modeling building types.
In this summary of results, the total average annual loss (AAL) and worst case loss
(WCL) are emphasized. Note that the WCL is the loss at the one-percent (1%) probability of
exceedance level on the loss EP curve, as defined in Chapter One. Also, thirty-eight percent
(38%) of the low-rise wood frame residential structures in the Oakland, California region are
mitigated (See Section 4.2). The effects of mitigation are presented in Section 6.2.
Recall from Chapter One that the Range of Uncertainty of Loss is defined in terms of the
absolute difference between the monetary loss with new knowledge, X/, at the time of the decision
and the monetary loss without such knowledge, x0 (Equation 6.1 rewritten from equation 1.4).
RU[L(X)]k = | L ( x\ ( i , j ) ) - L(x0( i , j ) ) | (6.1)
In this equation, the stakeholders, k, denote either the homeowner (MO) or Insurer (I);
fragility states, /, denote no structural mitigation (0 ) or structural mitigation (/); and insurance
use, j , indicated no insurance is in place (0) or insurance is utilized (/). In this way, each RU
value reflects the incorporation of mitigation and insurance (22 = 4 combinations for each
stakeholder). In this section, / = 0 (no structural mitigation) and j = 0 (no insurance).
Four different cases are considered and the Range of Uncertainty (RU) is calculated in
each case (Table 6.1). Case 1 compares the use of all default parameters and models with all
updated parameters and models. In other words, the losses in the default state for the recurrence,
attenuation, soils, exposure, and fragility before mitigation is compared to the losses in the
updated state for recurrence, attenuation, soils, exposure, and fragility before mitigation. Note
that the fragility before mitigation is used here, keeping consistency with the assertion above (i.e.
/ = 0).
169
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Case 2 is the comparison of the default and updated states for the seismic hazard
estimation parameters and models. Keeping all other parameters and models in their default state,
the losses using recurrence in the default state is compared to the losses using recurrence in the
updated state. Then, keeping all other parameters and models in their default state, the losses are
compared for the default and updated states o f attenuation. Finally, the same is done for the
default and updated soil mapping schemes.
Case 3 is the comparison of the default and updated states for the inventory, damage, and
loss estimation parameters and models. The losses using the default exposure database are
compared to the losses using the updated exposure database, keeping all other parameters and
models in their default state. Then, the losses are compared for the default and updated states of
fragility, keeping all other parameters and models in their default state. In this way, only one
parameter or model is updated at a time.
Table 6.1 Cases for Range of Uncertainty Calculations
Case Description
Case 1 All Default versus All Updated States
Case 2 Updated States of Seismic Hazard Estimation
(A) Recurrence
(B) Attenuation
(C) Soils Mapping Scheme
Case 3 Updated States of Inventory, Damage, and Loss Estimation
(A) Exposure
(B) Fragility before Mitigation
(C) Fragility after Mitigation
Case 4 Overall Maximum and Minimum Losses
Finally, case 4 compares the largest losses to the smallest losses. In this way, the upper
and lower bounds on the loss exceedance probability curves and the values of average annual loss
(AAL) and worst case loss (WCL) are established. This case is unique in that the state of each of
the uncertain parameters and models tested in the analysis is not known beforehand.
170
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
First, though, a comment is made about notation. In Table 6.2, a listing of the six areas of
uncertainty are given, along with the abbreviation used to designate the parameter or model in the
sensitivity analysis. These abbreviations were chosen to keep consistency throughout the
document. For example, in earlier chapters, recurrence is given as a function of magnitude (f(m)),
attenuation is described in terms of ground shaking, z, and the fragility of the building stock is
developed in terms of mean damage, y.
Table 6.2 Notation for Range of Uncertainty Calculations
Updated Parameter/Model Abbreviation
Earthquake recurrence f(m)
Attenuation relationship z
Soils mapping scheme soil
Inventory exposure exp
Fragility before mitigation ym(before)
Fragility after mitigation yjafter)
The loss exceedance probability curve for case I is shown in Figure 6.1. In this case, the
default parameters and models in the study are run using the HAZUS methodology and then the
updated parameters and models are run for a comparison of loss. As the graph shows, the loss at
every exceedance probability level is less in the case of the updated state of information.
Specifically, the average annual loss (AAL) for the default case is SI82.8 Million with a worst
case loss (WCL) of $3,149 Million. In the updated state, the AAL is $130.4 Million with a WCL
of $2,642 Million. For these loss estimates, the Range of Uncertainty or RU value is $52.4
Million and $507.4 Million for the AAL and WCL, respectively.
171
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
> 0.08
0.06
All Default
All Undated
0.04
0.02
S2,642 M S3,149 M
SO $2,000,000 $4,000,000
Loss
Figure 6.1 Loss Exceedance Probability Curve for Case 1
Next, case 2 calculates the Range of Uncertainty, RU, using updated knowledge of the
seismic hazard parameters: recurrence, f(m), attenuation, z, and soils mapping schemes, soil.
Graphical representations of the loss EP curves for these three areas of uncertainty are given in
Figures 6.2 through 6.4.
Figure 6.2 compares the loss exceedance probability curve with the default estimates of
earthquake recurrence based on the USGS data and the updated estimates of earthquake
recurrence based on the RMS data. With the updated frequency information, losses are lower.
This is an interesting result. Recall from Chapter Five that the estimates of earthquake recurrence
at the M = 6.5 event were significantly lower for the default case. In overall estimates of loss, in
terms of exceedance probability loss, average annual loss, and worst case loss, this significance is
lost. Specifically, the average annual loss (AAL) for the default case is $182.8 Million with a
worst case loss (WCL) of $3,149 Million. And, the AAL for updated state of information is
172
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
$149.6 Million and the WCL is $2,769 Million. The RU is calculated as $33.2 Million for AAL
and $380 for WCL (See Table 6.3).
0.1
0.08
0.06
Default Recurrence
Updated Recurrence;
0.04
0.02
$2,769 M $3,149 M
$2,000,000 $4,000,000
Loss
Figure 6.2 Loss Exceedance Probability Curve (Recurrence)
The next seismic hazard estimation model tested is the attenuation relationship. Recall
that the default relationship is the Boore, Joyner, Fumal (BJF) relationship (1993, 1994) and the
updated relationship is the Project97 relationship (NIBS 1997). Also, recall from Chapter Five
that the default BJF equation was chosen because it was the lower bound on ground motion at
close distances to the seismic source in most cases.
As expected, this translates into a lower loss for the default case (Figure 6.3). In other
words, the updated ground motion attenuation results in higher expected losses. Specifically, the
Project97 relationship results in an average annual loss of $219.1 Million and a worst case loss of
$3,810 Million. This translates into an RU value of $36.3 Million for AAL and $661 Million for
WCL (See Table 6.3). Notably, this portion of the seismic hazard estimation process corresponds
173
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
to a higher Range of Uncertainty in expected loss than earthquake recurrence (and it will be the
highest Range of Uncertainty overall).
0.1
^ 0.08
a
m
o 0.06

4)
O
5 0.04
V
u
0.02
~~ Default Attenuation
Updated Attenuation
WCL @ 1%
$3,149 M $3,810 M
S2,000,000
Loss
$4,000,000
Figure 63 Loss Exceedance Probability Curve (Attenuation)
Finally, the last portion of the seismic hazard estimation process is tested: the soils
mapping scheme. The default state of information is a map with the same soil site classification
(i.e. site class D) everywhere in the region and no liquefaction or landslide potential anywhere.
The updated state of information is a map with a range of soil site classes and liquefaction and
landslide potentials (See Section 5.1.3).
The results for this area of uncertainty are quite interesting. There is little difference in
the overall loss exceedance probability curve (Figure 6.4). Looking closer at the output file from
HAZUS, for the forty-six events considered in the earthquake catalogue, thirty-nine of the events
(i.e. 39/46 = 85%) estimate lower expected losses with the updated state of information.
Basically, in the updated mapping scheme, the number of census tracts in the region with soil of
site class B is much larger than the number of census tracts with soil site class E (i.e. 24 versus 7).
174
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
And, the loss to the structures in site class B is less in the updated scheme and loss to the
structures in site class E is greater in the updated scheme.
In the updated state of soils information, the average annual loss is estimated at SI71.3
Million and the worst case loss is estimated at $2,914 Million. Therefore, the RU with the
updated soil mapping scheme is approximately $11.5 Million for AAL and $235 Million for
WCL (See Table 6.3). This is the lowest Range of Uncertainty value calculated thus far.
Attention now turns to the inventory, damage, and loss estimation parameters and models.
Table 63 Summary of Loss Estimates for Case 2
Parameter/Model Losses i[AAL, WCL) in $ Millions
Defaulter#) Updated(xi) RU
f(m) ($182.8, $3,149) ($149.6, $2,769) ($33.2, $380)
* ($182.8, $3,149) ($219.1, $3,810) ($36.3, $661)
soil ($182.8, $3,149) ($171.3, $2,914) ($11.5, $235)
0.08 r
0.06
Default Soils
Updated Soils
a.
3 0.04
$2,914 M $3,149 M
$2,000,000 $4,000,000
Loss
Figure 6.4 Loss Exceedance Probability Curve (Soils)
175
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Case 3 calculates the RU with updated knowledge of the inventory, damage and loss
estimation parameters and models. First, the losses are tested for their sensitivity to the residential
inventory exposure database. Recall from Chapter Five that the default exposure square footage is
compiled from the 1990 Census data and the updated exposure square footage is developed from
the Alameda County tax assessors data. As in the case of the earthquake recurrence and soils
mapping scheme, the expected losses are lower with the updated exposure (Figure 6.5). This is
primarily due to the decrease in total residential square footage in the region in the updated
scheme. Specifically, the AAL is $166.6 Million and the WCL is estimated to be approximately
$2,762 Million. This results in a Range o f Uncertainty of $16.2 Million for average annual loss
and $387 Million for worst case loss (Table 6.4).
0.1
0.08
0.06
Default Exposure
Updated Exposure
0.04
0.02
WCL@ 1%
$2,762 M $3,149 M
$0 $2,000,000 $4,000,000
Loss
Figure 6.5 Loss Exceedance Probability Curve (Exposure)
The other area of uncertainty in case 3 is the general area of vulnerability of the
residential building stock. Recall in Chapter Five that the sensitivity testing is on the low-rise
wood frame structures {i.e. W1 in HAZUS) only. Testing the losses for sensitivity to the fragility
176
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
curves in the HAZUS software (default) and those developed in this work (updated in Chapter
Four) is broken down into the before mitigation state of fragility and the after mitigation state
of fragility. Since the loss exceedance probability curves presented thus far have not considered
the losses after mitigation {i.e. i = 1), a look at the loss EP curve for the case of ail default
parameters and models with mitigation is warranted. Additionally, this will serve as an
introduction to the effects o f mitigation, discussed further in Section 6.2.
In the HAZUS software, the low-rise wood frame residential structures are mitigated by
changing the occupancy scheme (See Section 4.2). In the default mode, the structure occupancy
mapping of thirty-eight percent (38%) of the structures are moved from the Low-Inferior seismic
design level bin to the Moderate-Code seismic design level bin (See Section 3.2.3.1). Performing
this task, while using the default recurrence, attenuation, soils, exposure, and fragility curves, the
losses are expected to reduce accordingly. In the updated mode, the structures are moved from the
Low-Inferior seismic design level bin to the Low-Code seismic design level bin and the fragility
curve parameters are changed (See Section 5.2.2).
First, the case of the fragility before mitigation is tested. In the default state, the same loss
EP curve utilized up to this point is once again used. The default case results in an AAL of $182.8
Million and a WCL of $3,149 Million. With the updated fragility information, losses are slightly
lower. The average annual loss is $169.8 Million, with a worst case loss of $2,979 Million
(Figure 6.6). This corresponds to a RU value of $13 Million for the AAL and $170 Million for the
WCL.
177
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.1
| Default Fragility (Before)
Updated Fragility (Before)
$2,979 1M $3,149 M
$2,000,000 $4,000,000
Loss
Figure 6.6 Loss Exceedance Probability Curve (Fragility Before Mitigation)
For the case of losses expected using the fragility curves after mitigation, the losses are
slightly lower (although just barely) in the updated case. The default loss exceedance probability
curve has an average annual loss of $127.8 Million and a worst case loss at the 1% exceedance
probability level of $2,268 Million. In the updated state, the AAL is $124.7 Million and the WCL
is $2,199 Million (Figure 6.7). Table 6.4 lists the results for case 3: the updated inventory
exposure, exp, the updated structural fragility curves before mitigation, y m(before), and the
updated structural fragility curves after mitigation, ym(after).
Table 6.4 Summary of Losses for Case 3
Parameter/Model Losses [AAL. WCL) in $ Millions
Defaulter*) Updated(x/) RU
exp ($182.8, $3,149) ($166.6, $2762) ($16.2, $387)
ym(before) ($182.8, $3,149) ($169.8, $2,979) ($13.0, $170)
yjafter) ($127.8, $2,268) ($124.7, $2,199) ($3.1, $69)
178
WCL @ 1%
$0
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.02 ;
jj 0.04
o 0.06 t
a.
0
Default Fragility (After)
Updated Fragility (After)
WCL@ 1%
S2,199M $2,268 M
$2,000,000
Loss
SO $4,000,000)
Figure 6.7 Loss Exceedance Probability Curve (Fragility After Mitigation)
Recall in Chapter Five, in most damage states, the low-rise wood frame structures were
less susceptible to structural damage with the updated information. In fact, there was a significant
reduction in the vulnerability for structural damage in some cases. But, as a percentage of the
total damage, on average, structural damage is only 11.5% (Figure 4.16). Therefore, the reduction
in loss is expected for the case of updated fragility curves (both before and after mitigation). And,
the small amount of reduction is also expected.
It is worth taking a closer look at the structural losses and the RU values of the different
types of losses in this case, though. Recall that in HAZUS, losses to low-rise wood frame
residential structures are a summation of losses to the structure (CS), nonstructural components
(CNS), and the contents (CCD), as well as the costs associated with relocation expenses (REL)
and rental income (RY) (Section 3.2.4). Also, the shift in fragility from more susceptible to less
susceptible is done for structural damage only. Nonstructural damage to acceleration-sensitive
179
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and drift-sensitive components of the structure and damage to the contents of the structure are the
same (Section 4.3.7).
Therefore, as expected, in Table 6.5, the difference between the nonstructural losses and
contents loss from the default case of structural fragility to the updated case of structural fragility
remains the same. Additionally, the structural loss is between (approximately) eleven- and
fourteen-percent of the total loss. This difference in loss (i.e. RU value) from the default case to
the updated case for both fragility before mitigation and fragility after mitigation is shown
graphically in Figure 6.8. The greater difference in loss before mitigation, as seen in the earlier
loss exceedance probability curves (Figures 6.6 and 6.7), is confirmed in the breakdown of the
losses.
Table 6.5 Breakdown of Average Annual Loss
AAL
(in S Millions)
CS CNS CCD REL RY Total Loss
Defaultfbefore) $26.0 $106.8 $30.6 $15.2 $4.3 $182.8
Default(after) $14.7 $77.7 $25.0 $8.1 $2.2 $127.8
y m(before) $19.6 $106.8 $30.6 $10.0 $2.8 $169.8
yjafter) $13.4 $77.7 $25.0 $6.7 $1.8 $124.7
180
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
S14
BEFORE MITIGATION
S12 ^-----------------i AFTER MITIGATION
CS CNS CCD REL RY TOTAL
Type of Loss
RY TOTAL
Figure 6.8 Difference in Breakdown of Residential Losses
Finally, the last case (case 4), compares the largest loss to the smallest loss. In this way,
the bounds on the loss exceedance probability curve and the range of AAL and WCL values are
established. In this study of the Oakland, California region, the maximum losses are obtained
using default earthquake recurrence, soil mapping scheme, residential inventory exposure data,
and fragility curves before mitigation with the updated attenuation relationship. The minimum
losses use default attenuation relationship with updated earthquake recurrence, soil mapping
scheme, inventory exposure data, and fragility curves after mitigation.
In the first combination of default and updated states, the average annual loss is $219.1
Million, with a worst case loss of $3,810 Million. In the second combination of states, the AAL is
only $80.4 Million and the WCL is $1,570 Million. The/?t/ value in this case is $138.7 Million
for AAL and $2,240 Million for WCL. This result is shown via the loss exceedance probability
curve in Figure 6.9. At the 0.01 (1%) exceedance probability level, there is a factor of 2.43
difference in loss!
181
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.1
Lower Bound1
Upper Bound ?
0.02
$1570 M $3,810 M
$4,000,000 $2,000,000
Loss
Figure 6.9 Loss Exceedance Probability Curve (Bounds)
6.2 Expert Opinion Incorporation for Mitigation
How does one define a parameter in the earthquake toss estimation process so it reflects all the
information provided by experts? How does one define a parameter in a cost-benefit analysis
so it reflects all the information provided by experts?
Expert opinion incorporation on the fragility of pre-1940 low-rise wood frame structures
before and after mitigation and the costs of mitigation has been completed in this study. The
results of this incorporation are given in detail in Chapter Four. In this section, the costs of
mitigation are summarized, the effects of mitigation are evaluated, and the expected change in
benefits with updated information is explored.
First, the expected cost of mitigation, Ce, is summarized. This cost is defined in terms of
the cost to hire a contractor to do the work (Crowd or the homeowner to retrofit the structure
himself (CAto(ra/), depending on the homeowners decision process (See Equation 4.4). In this
estimate, there is no RU value. Instead, the costs estimated based on the contractor survey data
182
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
are incorporated into the homeowner cost-benefit analysis. In Chapter Four, this survey data was
fit to a number of different curves. The best distribution fit for Croiai and CKuacrxai are summarized
in Table 6.6. In this analysis, Ctoiat = $5,586 and Csimenai= $1,384.
Table 6.6 Best Distribution Fit Parameters for Cost Data
Type (ah az) ('Total
Ofotal
Weibull (2.77, 6275.9) $5586 $2182
Type (ah ad
( 'M a te r ia l o,Material
Lognormal (1383.7, 976.4) $1384 $976
Next, attention turns the incorporation of the structural fragility of the residential building
stock and the benefits of mitigation. The change in the annual benefits of mitigation,zh?, due to
the incorporation of new knowledge of the fragility, x h is the difference between the absolute
value of the annual benefits of mitigation in this updated information state and those in the default
state, x0. In Equation 6.2 (rewritten from equation 1.5), considering benefits to society and no
insurance (j = 0), / = 0 reflects no mitigation and / = / reflects structural mitigation.
AB = | {AAL(xx(0.0)) - AAI(;c, (1,0)) | - | (AAL(x0 (0,0)) - AAL(x0 (1,0)) | (6.2)
The change in the annual benefits of mitigation, AB, is important in understanding the
impact of incorporating the new structural fragility curves in the software. A lot of time and effort
was spent developing these curves. A significant difference from the loss using the default
HAZUS curves must be ascertained. First, though, the expected benefits of mitigation in both
cases (default and updated) are given. Then, this change is discovered.
The effects of mitigation are calculated using the default state of information in all areas
of uncertainty. In other words, the default earthquake recurrence, attenuation relationship, soils
mapping scheme, and residential inventory exposure are used to calculate the losses. Using the
default HAZUS curves, the average annual loss is $182.8 Million before mitigation and $127.8
183
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Million after mitigation (Figure 6.10). Using the updated fragility curves, the average annual loss
is $169.8 Million before mitigation and $124.7 Million after mitigation (Figure 6.11). This
translates into a savings of $55 Million in the default case and a savings of $45.1 Million in the
updated case to the entire residential building stock of the region. Therefore, mitigation is slightly
more effective in the default state of information than the updated state of information.
Furthermore, the change in annual benefits of mitigation is negative and its just under $10
Million (Table 6.7).
Looking closely at the losses before mitigation, the $13.1 Million difference between the
default and updated state consists of $6.5 Million in structural loss, $5.2 Million in relocation
expenses, and $1.4 Million in rental income losses. After mitigation, the $3.2 Million difference
between the default and updated state consists of $1.3 Million in structural loss, $1.4 Million in
relocation expenses, and $0.4 Million in rental income losses (See Table 6.5). Before mitigation,
the larger difference in structural loss ($6.5 Million) is primarily due to the large difference in
threshold of spectral displacement at the Moderate and Extreme damage states from the default to
the updated state of information (See Table 5.4).
Table 6.7 Effects of Mitigation
Losses (AAL)
in $ Millions
Before
Mitigation
After
Mitigation
Benefits of
Mitigation
Default(r) $182.8 $127.8 $55.0
Updated(x/) $169.8 $124.7 $45.1
Change in annual benefits o f mitigation - AB -$9.9
184
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.08 T
0.06 *
Default (Before Mitigation)
Default (Alter Mitigation)
i
0.04 T
0.02 r
S2,268 M S3,149 M
52,000,000 SO 54,000,000
Loss
Figure 6.10 Loss Exceedance Probability Curve (Default Mitigation)
0.1
0.08
0.06
Updated (Before Mitigation)
i Updated (After Mitigation)
0.04
0.02
WCL@ 1%
$2,199 M S2,979 M
SO $2,000,000 $4,000,000
Loss
Figure 6.11 Loss Exceedance Probability Curve (Updated Mitigation)
While this analysis is a good estimate of the annual benefits of mitigation and the change
in annual benefits of mitigation with the updated fragility curves, there are sixty (60) other
185
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
branches on the logic tree to consider here. Or, more specifically, thirty (30) other estimates of
the annual benefits of mitigation and fifteen (15) other estimates of the change in annual benefits
of mitigation. The lower and upper bounds on the annual benefits of mitigation need to be
determined. Taking into consideration the lower bound and upper bound on the loss exceedance
probability curve developed in case 4 of Section 6.1, the lower and upper bounds on the benefits
of mitigation are determined.
Specifically, the lower bound on the annual benefits of mitigation is $23.3 Million using
the default attenuation relationship with the updated earthquake recurrence, soils mapping
scheme, inventory exposure, and fragility curves. The upper bound on the annual benefits of
mitigation is approximately $61 Million using the updated attenuation relationship with the
default recurrence, soils mapping scheme, inventory exposure, and fragility curves. The average
annual benefits of mitigation across all thirty-two cases are approximately $39.9 Million (Figure
6. 12).
<S25 S25-S30 S30-S35 S35-S40 S40-S45 S45-S50 S50-S55 S55-S60 >S60
Benefits (S Millions)
Figure 6.12 Annual Benefits of Mitigation
186
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Finally, the change in the annual benefits of mitigation is determined. Using the same
cases outlined above, the lower bound on the change in annual benefits is -$11.0 Million and the
upper bound is -$5.5 Million. The average change in annual benefits of mitigation (considering
all sixteen cases) is a decrease of approximately $8.1 Million (Figure 6.13). This information is
important, because it implies that incorporating the new structural fragility curves decreases the
annual benefits of mitigation, on average, by $8.1 Million. This corresponds to a decrease of $260
per homeowner (i.e. for those who mitigate only). Note that a complete listing of the AAL
estimates for the sixty-four branches of the logic tree is given in Appendix 8.13.
10
$8.1 M
41
41
e
! 41
1 w
! =
w
41
: o
>>
u
c
o
o
Im
fa
<-SU -(S9-S11) (-S7-S9) -(S5-S7)
Change in Annual Benefits (S Millions)
Figure 6.13 Change in Annual Benefits of Mitigation
6.3 Cost-EfTectiveness o f Mitigation
Is it beneficialfor a homeowner in Oakland to mitigate fo r earthquake hazard?
Now, attention turns to the cost-effectiveness of mitigation. Using the RU values
calculated in Section 6.1, a cost-benefit analysis on the benefits of a homeowner mitigating for
earthquake hazard is completed. The consideration of the cost-effectiveness of mitigation is done
187
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
without the additional effects of insurance. In other words, the assumption is that the homeowner
has no insurance. The effects of the mitigation and insurance together are considered in the next
section.
The point at which it is beneficial for the homeowner to mitigate is established using a
net present value (NPV) calculation (Equation 6.3).
NPV = -Ca + - 'M- (6 3)
1=1 (1 + r)'
In this equation, the upfront cost of mitigation, C, is either CTomi or CMaleai, as
summarized in the preceding section. The annual benefits of mitigation, Bm, are simply the
difference between the annual unmitigated and mitigated losses using the updated information
state, X/, or the default information state, x0. This was also summarized in the previous section.
Finally, r is the discount rate and T is the lifetime of the structure.
Whenever the NPV is greater than or equal to zero, it will be beneficial to the homeowner
to mitigate. Framing the analysis in this way, this work assesses the robustness of a specific
residential structural mitigation technique {i.e. bolting the structure to its foundation and bracing
its cripple wall), and the cases in which mitigation is monetarily beneficial are assessed.
First, the annual benefits of mitigation are summarized for the six areas of uncertainty in
Table 6.8, and the annual benefits per structure (or homeowner) are assessed. The assumption is
that the annual benefits of mitigation are seen by those who mitigate only {i.e. 31,358 residential
structures in the default state and 28,094 residences in the updated state).
Then, the discount rate, r, and life of the structure, T, are varied to determine the point at
which it becomes beneficial for the homeowner to mitigate. Two different techniques are used to
test the effects of r and T on the net present value: two-way data tables and Monte Carlo
simulation.
188
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.8 Cost-Benefit Analysis of Mitigation
Parameter/Model Before
Mitigation
After
Mitigation
Btu
(all structures)
Bm
(per structure)
f(m) $149.6 M $105.6 M $44 M $1,402
z $219.1 M $158.1 M $61 M $1,947
soil $171.2 M $125.1 M $46.1 M $1,471
exp $166.6 M $116.6 M $50 M $1,779
y*
$169.8 M $124.7 M $45.1 M $1,439
Default $182.8 M $127.8 M $55 M $1,755
First, a two-way data table is constructed using the six cases of annual benefits of
mitigation given in Table 6.8. A two-way data table is a technique within the Microsoft Excel
environment that allows the user to perform sensitivity analyses. The two-way data table can
show how two different input values can affect one output value. In this case, the input
values are the discount rate, r, and the life of the structure, T. The output value is the net present
value, NPV.
In constructing the two-way data tables (for the six cases in Table 6.8), the discount rate
ranges from five- to twenty-percent (5% to 20%) and the life of the structure is varied from one
(1) to thirty (30) years. In the majority of the cases, if the homeowner pays a contractor to retrofit
his home {i.e. using Croiai $5,586), within five years the homeowner will recoup the cost of
mitigation. In Figure 6.14, this example of the NPV calculation as a function ofr and T is shown,
using all default states o f information. Additionally, in all the cases, if the homeowner chooses to
retrofit the structure himself {i.e. using C.uatenai = $1,384), he will recoup the cost of mitigation in
those same five years.
189
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
$20,000
5% discount rate
10% discount rate
15% discount rate -
20% discount rate
$15,000
> $10,000
$5,000
25 30 20
($5,000) .
Years
Figure 6.14 Net Present Value (Total Cost)
Also, a Monte Carlo simulation is run using the Crystal Ball software by
Decisioneering, Inc. The input parameters include the discount rate, r, the life of the structure,
T, and the cost of mitigation, Cro,ai or CMalena/. In this way, the uncertainty in the cost of mitigation
is introduced into the analysis. The discount rate, r, is set at ten-percent (10%). The life of the
structure, T.. is taken as a uniform distribution, ranging from one (1) to thirty (30) years. The
distributions on the total cost of mitigation, CToiai, and the material cost to mitigation, CMalenat, are
those from Table 6.6 in Section 6.2 (i.e. a Weibull distribution for Croiai and a lognormal
distribution for C^atcnot). One thousand trials were run, and the certainty that the net present value
(NPV) is greater than zero was discovered.
For the same six cases used in the data-table construction, it was confirmed that in most
cases, it is beneficial for the homeowner to pay someone to retrofit his home. Specifically, with a
certainty of approximately eight-five percent (85%), if the homeowner has a contractor bolt his
structure to its foundation and brace its cripple wall, the homeowner will have a positive NPV
(Table 6.9). With a certainty of approximately ninety-eight percent (98%), if the homeowner
190
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
retrofits the structure himself, he will have a positive NPV (Table 6.9). Therefore, it is almost
certain that this type of mitigation technique, as modeled here, is a cost-effective means of retrofit
for an earthquake hazard. An example of a simulation runs output file (for the Default case using
Total Cost, CTomi) is given in Figure 6.15.
Table 6.9 Simulation Results
Parameter/Model Certainty of Positive NPV
(using Ctmi)
Certainty of Positive NPV
(using Cuauriab
f(m)
85.20% 98.60%
z 91.90% 99.50%
soil 86.90% 98.60%
exp 88.90% 99.40%
ym 86.60% 98.70%
Default 89.10% 99.10%
Forecast: NPV Default with Total Cost
1,000 Trials Frequency Chart 0 Outliers
63
i 47.25
n
31.5 JO
e
r
S
15.75 ^
S12.500.00
Certainty is 89.10% from $0.00 to Infinity US S
Figure 6.15 Simulation of Net Present Value (Total Cost)
191
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6.4 Earthquake Insurance
Considering residential earthquake insurance, what are the expected losses to the homeowner
and the residential insurer in the Oakland, California area?
Finally, the last result presented is the effects of earthquake insurance. Considering a few
different deductible and limit level schemes, the expected losses to homeowners and the
insurance industry in the region are calculated. In Table 6.10 (rewritten from Table 1.1), the
annual losses to the homeowner are laid out in terms of the expected losses, AAL(x\(ij)) hq or
AAL{xo(iJ))ffQ, the insurance premiums paid by the homeowner, (1 +l f ) - AAL(x\(i, j))or
(1 + l f ) - AAL(xo(i,j) ) , and the cost of mitigation, a f - C Q (See Section 1.2).
Table 6.10 Homeowner Loss as a Function of Mitigation and Insurance
Losses to
Homeowner
Insurance
0 = 0
No Insurance
0 = 0 )
Mitigation
(; = I)
AAL{xx(\ , \ ))h o +(! + / / ) AAL(xx( U )) + af C0
AAL(xq(\,\))h o + (1 + / / ) AAL(xq(l.U) + af CQ
AAL(xi(l,0))ho + af c o
AAL(xq (\,Q))hO + af c o
No Mitigation
(/ = 0)
AAL(x\ (0,l))HO + ( ! + / / ) AAL(Xl (0,1))
AAL(xq0,1)) ffo +( l +l f ) AAL( xo( 0, l ) )
AAL(xj (0,0)) ho
AAL(xq (0,0)) ffQ
Similarly, the losses to the insurer are laid out in Table 6.11 (rewritten from Table 1.2).
Recall that the consideration of the primary insurer loss as a function of asset level, reinsurance,
and other financial instruments is out of the scope of this work. The losses calculated here are
simply the losses paid by the insurer above the deductible level and below the limit level loss to
the homeowner.
192
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.11 Insurer Loss as a Function of Structural Mitigation
Losses to Insurer Insurance
(/ = /)
No Insurance
0 = 0)
Mitigation
(/ = / )
A4Z(x,(U))/
/L4L(x0(l,l))/
N/A
No Mitigation
(/ = 0)
AAL(xt ( 0,1))/
AAL(x0 (0,l))j
N/A
There are three deductible and limit level schemes considered here. They are: a 10%
deductible scheme with 100% limit, a 15% deductible with a 100% limit, and a 10% deductible
with a 25% limit. This last case of a low (25%) limit level is added to look at a possible wrap
around coverage based on the California Earthquake Authority (CEA) policy (EERI 2000).
First, the homeowner losses are presented. Based on the results of the previous section,
the discount rate, r, is set at 10% and the life of the structure, T, is taken as 30 years. Therefore, in
this analysis, ay, = 0.106, which converts the upfront cost of mitigation to an annualized cost over
the lifetime of the structure and discounted for interest (See Section 1.2). Looking at the two by
two matrix of Table 6.10, first, the loss with no insurance (/ = 0) is presented (i.e. the last
column). Then, the loss with insurance (/= 1) and no mitigation (/' = 0) is shown. Finally, the
most complicated calculation of losses, with insurance (/'= 1) and mitigation in place (/ = 1), is
summarized. A summary of the parameters used in the analysis is given in Table 6.12. Note that
the cost of retrofitting the structure (either CTo,ai or CMalenai) is taken at its mean value for this
analysis.
193
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.12 Parameters for Analysis
Parameter Value
Discount rate, r 10%
Life of the structure, T 30 years
Insurance loading factor, lf 0.5
Annualized factor for mitigation, a{ 0.106
Total Cost of mitigation, CTotai $5,584
Material Cost for mitigation, $1,384
Total number of structures (Default state) 82,522
Total number of structures (Updated state) 73,933
Number of structures eligible to mitigate (Default state) 31,358
Number of structures eligible to mitigate (Updated state) 28,094
First, the average annual loss to all of the homeowners in the Oakland, California region,
if no insurance is in place, is summarized in Table 6.13. In the case of no mitigation, the losses to
the entire region range from under $130 Million to just under $220 Million, as already discovered
in Section 6.1. These total AAL estimates correspond to an average loss between approximately
$ 1,800 and $2,700 per homeowner (HO).
In the case of structural mitigation, the losses reduce, as expected (Figure 6.16). Also, the
average annual loss (AAL) per homeowner is the least for the case of the homeowner retrofitting
the structure himself {i.e. using Material Cost). Note that, in the case of mitigation, the average
annual loss per homeowner (HO) is for the homeowner who mitigates. It is assumed that the
homeowners that can not mitigate {i.e. who own a structure built after 1940) will incur the same
AAL estimate. Losses for an individual homeowner who hires a contractor to perform the retrofit
range from $1,000 to $1,300 on an average annual basis. Losses for an individual homeowner
who performs the retrofit himself range from $560 to $860 on an average annual basis.
194
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.13 Homeowner Average Annual Loss without Insurance
Parameter/
Model
No
Mitigation
Mitigation
(using Total Cost,
C tmmi)
Mitigation
(using Material Cost,
Material)
Total Per HO Total Per HO Total Per HO
f(m) $149.6 M $1,813 $124.2 M $1,003 $110.3 M $558
r
$219.2 M $2,656 $176.7 M $1,301 $162.7 M $856
soil $171.3 M $2,075 $143.7 M $1,196 $129.7 M $751
exp $166.6 M $2,253 $133.3 M $1,067 $120.8 M $621
ym $169.8 M $2,057 $143.2 M $1,211 $129.3 M $765
Default $182.8 M $2,216 $146.4 M $1,053 $132.4 M $608
k.
s
e
*

4>
E
e
X
o
J
u
o*
a
u
<
S3,000
INo Mitigation Total Cost Material Cost
S2,500
S2,000
SI,500
SI,000
toil exp ym
Type of Uncertainty
Default
Figure 6.16 Homeowner Loss without Insurance
Next, the effects of insurance on the homeowner losses, without any mitigation in place,
are surmised. In this case, the homeowner is expected to pay the deductible loss and an annual
premium for insurance, based on a loading factor times the average annual loss. Recall from
Chapter Three (Section 3.3), that the losses to the homeowner for each earthquake event,L/HO),
195
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
are based on the deductible and limit level scheme of the residential earthquake insurance
coverage (Equation 6.4). If the loss is below the deductible level, defined in terms of a percentage
of the value of the structure, %dV, the homeowner pays for the loss out of his own pocket. If the
loss is above the deductible level but below the limit level, %l- V, he pays only the deductible.
Finally, if the loss is above the limit level, the homeowner pays both the deductible and the loss
over the limit (i.e. Lj - %d- V).
L j \ i L j < %d V
L} ( H O ) = %d V if %d -V <Lj < %/ V (6.4)
%d V + ( Lj -% I V ) if L j > %/ V
Similarly, the loss to the insurer, L/I), is zero, if the loss is below the deductible level,
%dV. If the loss is above the deductible level but below the limit level, the insurer pays the
difference. Finally, if the loss is above the limit level, he only pays the portion of the loss between
the deductible and limit level (i.e. (%l - %d) V).
0 if L j < %d V
Lj (I) = (Lj - %d -V) if %d V <Lj < %l V (6.5)
(%l %d) V if Lj >%l-V
As already mentioned, three cases of deductible and limit level schemes are considered
and the results are presented in Table 6.14. The first case, of a ten-percent deductible and full
coverage, the homeowner will have the least out-of-pocket expenses. These AAL estimates, per
homeowner, range from approximately $3,900 to $5,700. The second and third cases, of a fifteen-
percent deductible with full coverage and a ten-percent deductible with only a twenty-five percent
coverage limit, result in larger losses to the homeowner. These estimates, per homeowner, range
from under $4,100 to over $6,000.
196
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.14 Homeowner Average Annual Loss with Insurance and No Mitigation
Parameter/
Model
10% Deductible,
100% Limit
15% Deductible,
100% Limit
10% Deductible,
25% Limit
Total Per HO Total Per HO Total Per HO
f(m) $321.5 M $3,896 $337.5 M $4,090 $339.3 M $4,112
z $467.2 M $5,661 $490.1 M $5,939 $496.0 M $6,011
soil $364.4 M $4,415 $381.1 M $4,618 $390.0 M $4,726
exp $358.9 M $4,854 $374.4 M $5,064 $382.4 M $5,172
y(before) $364.8 M $4,421 $380.9 M $4,616 $388.3 M $4,705
Default $392.4 M $4,755 $410.0 M $4,968 $418.1 M $5,067
Finally, the effects of mitigation and insurance together on the homeowner losses are
summarized (Table 6.15). This is the most interesting case. The results are broken down into the
losses to the homeowners who can not mitigate because their homes were built after 1940 and the
losses to the homeowners who do mitigate. Furthermore, those who do mitigate have an option of
either retrofitting their home themselves {i.e. using Material Cost,CKta)enal) or hire a contractor to
do the work for them {i.e. using Total Cost, C Total)-
As expected, the average annual losses to the homeowner are the least in the case of a
ten-percent deductible and full coverage. Furthermore, mitigation reduces the average annual loss
estimate per homeowner, on average, about fourteen-percent (14%) when the homeowner hires a
contractor to retrofit his home and about twenty-three percent (23%) when the homeowner
retrofits the structure himself. Notably, the premium also decreases with mitigation due to the
reduction in average annual loss {i.e. because premiums are simply 1.5 times average annual
loss).
While, on an average annual basis, the homeowner losses at the 10% deductible and 25%
coverage limit are slightly less than the losses at the 15% deductible and 100% coverage limit for
those who mitigate, the extra coverage will make a difference in a catastrophic scenario event.
For example, for a M = 7.5 on the North Hayward fault within the Oakland, California region, the
197
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
homeowner will have a higher cost with only 25% coverage. In Figure 6.19, the homeowner loss
(with mitigation) for the entire Oakland, California region is shown for this catastrophic event.
Table 6.1S Homeowner Average Annual Loss with Insurance and Mitigation
10% Deductible,
100% Limit
Homeowner
can not
mitigate
HO mitigates
(using Total Cost,
Croud
HO mitigates
(using Material
Cost, CjUaUrimt)
Total Per HO Total Per HO Total Per HO
f(m) $199.3 M $3,896 $109.4 M $3,490 $95.5 M $3,044
z $289.6 M $5,661 $152.8 M $4,873 $138.8 M $4,427
soil $225.9 M $4,415 $124.0 M $3,955 $110.0 M $3,509
exp $222.5 M $4,854 $117.0 M $4,163 $104.4 M $3,717
ym
$226.2 M $4,421 $125.9 M $4,016 $112.0 M $3,570
Default $243.3 M $4,755 $128.2 M $4,089 $114.2 M $3,643
15% Deductible,
100% Limit
Homeowner
can not
mitigate
HO mitigates
(using Total Cost,
Croui)
HO mitigates
(using Material
Cost, Cnimlerimi)
Total Per HO Total Per HO Total Per HO
f(m) $209.3 M $4,090 $113.7 M $3,626 $99.7 M $3,180
z $303.9 M $5,939 $159.5 M $5,086 $145.5 M $4,640
soil $236.3 M $4,618 $128.8 M $4,109 $114.9 M $3,663
exp $232.1 M $5,064 $121.0 M $4,306 $108.5 M $3,861
ym
$236.2 M $4,616 $130.4 M $4,159 $116.5 M $3,714
Default $254.2 M $4,968 $132.9 M $4,237 $118.9 M $3,792
10% Deductible,
25% Limit
Homeowner
can not
mitigate
HO mi
(using Tc
C tl
tigates
ital Cost,
u l)
HO mitigates
(using Material
Cost, CMaterial)
Total Per HO Total Per HO Total Per HO
f(m) $210.4 M $4,112 $110.9 M $3,537 $96.9 M $3,091
z $307.5 M $6,011 $155.5 M $4,959 $141.5 M $4,513
soil $241.8 M $4,726 $127.4 M $4,063 $113.4 M $3,618
exp $237.1 M $5,172 $119.0 M $4,237 $106.5 M $3,792
ym
$240.7 M $4,705 $127.8 M $4,076 $113.9 M $3,631
Default $259.2 M $5,067 $130.4 M $4,157 $116.4 M $3,712
198
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(10%, 100%) B(15%, 100%) (!<)%, 25%)
S3,000
g $2,000
1
(A
SI,000
so
Mm
Default z soil exp ym
Type of Uncertainty
Figure 6.19 Homeowner Loss (M = 7.5 on the North Hayward fault)
Finally, the estimates of loss to the insurer are summarized (Table 6.16 and Figure 6.20).
This calculation is rather straightforward. The insurer losses are simply the losses that the insurer
must cover above the deductible level and below the limit level. As expected, the insurer only
pays for the extreme magnitude events. For lower magnitude events, with losses below the
deductible level, the insurer pays nothing.
For the direct economic losses from the forty-six events in the Event Loss Table (ELT),
the insurer only pays out during a portion of them. Specifically, in the case of a fifteen-percent
deductible and full coverage, the insurer pays out, on average, in eleven of the forty-six events.
For the ten-percent deductible and full coverage, the insurer pays out, on average, for sixteen of
the forty-six events. In other words, when the deductible level decreases, the insurer will have to
cover more of the cost.
199
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6.16 Insurer Losses
Before Mitigation
Parameter/
Model
(10% Deductible,
100% Limit)
(15% Deductible,
100% Limit)
(10% Deductible,
25% Limit)
f(m) $52.6 M $36.5 M $34.8 M
2 $80.8 M $57.8 M $51.9 M
soil $63.8 M $47.1 M $38.2 M
exp $57.6 M $42.1 M $34.1 M
y m(before) $59.6 M $43.5 M $36.1 M
Default $64.7 M $47.1 M $39.0 M
After Mitigation
Parameter/
Model
(10% Deductible,
100% Limit)
(15% Deductible,
100% Limit)
(10% Deductible,
25% Limit)
f(m) $25.0 M $13.8 M $21.1 M
z $42.1 M $24.5 M $35.0 M
soil $35.4 M $22.7 M $26.4 M
exp $27.6 M $17.0 M $22.1 M
yja fie r) $29.1 M $17.3 M $24.1 M
Default $31.0 M $18.7 M $25.3 M
(10%, 100%) Before Mitigation (10%, 100%) After Mitigation
(15%, 100%) Before Mitigation B(15%, 100%) After Mitigation
S90 - ------------- -------------
| S60 r
1
<*>
| $30 i-
-J
SO -
f(a) z soil exp ym Default
Type of Uncertainty
Figure 6.20 Insurer Losses
200
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Additionally, the effects of mitigation on the insurer losses are significant. In the case of
a 10% deductible and a 25% limit level, there is a 30.8% reduction in loss with the updated soils
mapping scheme, soil. On the other extreme, in the case of a 15% deductible and full coverage,
the losses reduce approximately 62.2% with the updated information on earthquake recurrence,
f(m) (Table 6.17). These differences in reduction in the various areas of uncertainty are a function
of the total loss that the insurer will cover between the deductible and limit level.
Specifically, in the case of the 10% deductible and 25% coverage limit, the insurer (on
average) has a reduction of 34.4%. And, in the case of the 10% deductible and full coverage, the
insurer has an average reduction of 50.0%. In the case of the 15% deductible and 100% coverage,
the insurer has an average reduction of 58.6% (Table 6.17).
Table 6.17 Percentage Reduction in Insurer Losses
Parameter/
Model
(10% Deductible,
100% Limit)
(15% Deductible,
100% Limit)
(10% Deductible,
25% Limit)
f(m) 52.4% 62.2% 39.2%
z 47.9% 57.7% 32.6%
soil 44.6% 51.9% 30.8%
exp 52.1% 59.6% 35.3%
y m(before) 51.1% 60.3% 33.2%
Default 52.1% 60.2% 35.0%
Average r 5 0 .0 % 58.6% 34.4%
201
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7 CONCLUSIONS
This final chapter summarizes the results presented in Chapter Six and suggests future
developments in this work on uncertainty in earthquake loss estimation.
7.1 Summary
The summary of results can be broken down into three main areas: the effects of
uncertainty on the earthquake loss estimation process, the effects of mitigation on catastrophic
earthquake loss estimates, and the effects of residential earthquake insurance on a homeowners
expected loss.
First, the earthquake loss estimation process is, with no surprise, very uncertain. More
specifically, final estimates of direct economic loss are quite sensitive to the choices of
parameters and models chosen for use in the earthquake loss estimation process. Depending on
which parameters and models are chosen for use in an analysis, there can be a large difference in
estimates of loss. In the case of the Oakland, California region, direct economic loss estimates at
the one-percent probability of exceedance level range from $1,570 Million to $3,810 Million (See
Figure 6.9).
Furthermore, using the Range of Uncertainty calculated in Chapter Six, the parameters
and models tested within this study can be ranked according to their influence on the final
estimate of direct economic loss in the Oakland, California region. On average, the parameters
and models in the seismic hazard estimation portion of the ELE process are more influential on
the estimates of economic loss than the parameters and models in the inventory, damage, and loss
estimation portion of the process. Furthermore, the choice of empirical relationship for ground
motion attenuation is most influential in the earthquake loss estimation process. Alternative
202
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
estimates of the attenuation of ground motion from source-to-site give a ten-percent to thirty-
percent difference in direct economic loss, depending on the exceedance probability level.
Based on the largest Range of Uncertainty or RU value calculated in Chapter Six, the
ranking of the parameters and models according to their influence on average annual loss (AAL)
and worst case loss (WCL) is given in Table 7.1. It is clear that the direct economic loss estimates
calculated in this work are least sensitive to the incorporation of new structural fragility curves
after mitigation. Furthermore, in the case of the Oakland, California region and the HAZUS
methodology, the ranking of the parameters and models is (1) ground motion attenuation, (2)
earthquake recurrence, (3) residential inventory exposure, (4) soils mapping scheme/structural
fragility before mitigation, and (5) structural fragility after mitigation.
Table 7.1 Ranking of Parameters and Models by Range of Uncertainty
Parameter/
Model
RU based on
AAL Estimate
Parameter/
Model
RU based on
WCL Estimate
Attenuation (z) $36.3 M Attenuation (z) $661 M
Recurrence (f(m)) $33.2 M Exposure (exp) $387 M
Exposure (exp) $16.2 M Recurrence (f(m)) $380 M
yjbefore) $13.0 M Soils (soil) $235 M
Soils (soil) $11.5 M yjbefore) $170 M
ym(after) $3.1 M yjafter) $69 M
In the listing above, note that the soils mapping scheme and the structural fragility before
mitigation are grouped together. Looking closely at Table 7.1, the structural fragility before
mitigation is ranked higher that the soils mapping scheme based on the AAL estimate (i.e. $13.0
Million for ym(before) versus $11.5 Million for soils). On the other hand, if the Range of
Uncertainty is based on the worst case loss estimate (WCL), the soils mapping scheme is ranked
higher than the structural fragility before mitigation (i.e. $235 Million for soils versus $170
Million for ym(before)). Therefore, there is no distinction made between the sensitivity of the loss
203
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
estimates to these two areas of uncertainty. Supporting this choice is the fact that the loss
exceedance probability (EP) curves are too similar to make a clear distinction between these two
areas (See Figures 6.4 and 6.6).
Note also, in the listing above, that the earthquake recurrence is taken as more influential
than the inventory exposure, even though the exposure has a larger RU value for worst case loss
(WCL) in Table 7.1 (i.e. $387 Million for exp versus $380 Million for f(m)). The Range of
Uncertainty for the average annual loss (AAL) is significantly greater for the earthquake
recurrence than the inventory exposure (i.e. $33.2 Million for f(m) versus $16.2 for exp).
Additionally, the difference between these two areas of uncertainty is clear at the higher
exceedance probability levels in the loss EP curves presented in Chapter Six (Figures 6.2 and
6.5).
The second conclusion drawn from this work is the effects of mitigation. The mitigation
technique chosen for analysis requires bolting low-rise wood frame residential structures to their
foundation and bracing their cripple walls. With the use of this technique, applied to homes built
before 1940 in the Oakland, California region, the reduction in direct economic loss estimates is
quite substantial.
The breakdown of these losses indicates that the majority of the loss is due to damage to
the nonstructural elements (CNS in Figure 7.1). Comparing the default HAZUS structural
fragility curves to the updated structural fragility curves (with all other areas of uncertainty in the
default state), there is a higher percentage of difference in the damage and loss to the structure in
the default mode before and after mitigation (i.e. CS has 2.7% difference in the default state
versus 0.8% difference in the updated state). Therefore, this translates into the result that the
benefits of mitigation are slightly higher using the default HAZUS structural fragility curves than
204
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
using the updated ones. Of course, one would still want to use the updated structural fragility
curves because it is assumed that they are more accurate.
In the default state of information, the average annual benefits of mitigation are
approximately $55 Million to 31,358 homes (i.e. $1,755 per homeowner). In the updated state of
information, the average annual benefits of mitigation are approximately $45.1 Million to 31,358
homes (i.e. $1,439 per homeowner). Considering all the cases of uncertainty, on average, the
annual benefits of mitigation is $39.9 Million (i.e. $1,273 per homeowner).
$200 ----------------------------------------------------------------------------------------
$160
$120
1 Default (Before) i
_ Default (After)
, Updated (Before)!
i Updated (After)
CS CNS CCD REL
Type of Loss
RY TOTAL
Figure 7.1 Breakdown of Residential Losses
An additional result is the cost-effectiveness of this particular structural mitigation
technique. Using upfront costs of mitigation, C, calculated in terms of the cost for a contractor to
perform the retrofit (CToiai) or the homeowner to retrofit the structure himself (CMaleriai), and a net
present value (NPV) calculation, in the majority of the cases, it is beneficial for the homeowner to
mitigate. Furthermore, within five years, the homeowner will recoup the cost of mitigation with
an 85% certainty, based on Monte Carlo simulation results (Section 6.3).
205
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Finally, in terms of insurance parameters, deductible levels considered here play a minor
role in which stakeholder bears the brunt of the direct economic loss on an average annual basis.
In the three cases of deductible and limit level schemes (eg. (10% deductible, 100% limit), (15%
deductible, 100% limit), (10% deductible, 25% limit)), the homeowner will have to cover the
majority o f the average annual loss (AAL).
Also, considering average annual loss to homeowners and the insurance industry in the
region, the homeowner has the smallest loss with a 10% deductible level and total coverage. The
insurer will only have to cover the losses resulting from catastrophic events. Not surprisingly,
then, mitigation reduces the insurers losses significantly.
7.2 Future Work
There is much that can be done to build on the results of this work. More time can be
spent on the earthquake recurrence portion of the model. Time dependent models can be used to
more accurately assess the magnitude-frequency distribution. Additionally, based on the result
that ground motion attenuation is most influential in the estimation of direct economic loss, the
focus of future research and data collection should be on this area of uncertainty. Finally, a more
robust calculation of insurance industry losses as a function of asset level, reinsurance, and other
financial instruments can be developed. The annual average loss estimates to the insurance
industry in the Oakland, California region developed here can be used as a foundation for the
work.
As for the incorporation of structural fragility curves into HAZUS, a great deal of time
and effort went into this development. The HAZUS software (as-is) is not well suited for easily
incorporating the effects of structural mitigation techniques. But, as this analysis shows, it is
possible and should be done. The key is to develop a way to make the incorporation much more
user friendly and much less time consuming.
206
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8 Appendices
8.1 Earthquake Measures
1. Modified Mercalli Intensity
One of the earliest scales, developed in the 1930s, the Modified Mercalli scale is a subjective
scale. Determined from surveys following an earthquake, isoseismals are usually drawn to show
the distribution of damage and ground shaking intensity. The Modified Mercalli Intensity scale
ranges from I to XII (Steinbrugge 1982):
I Not felt except by a very few under especially favorable circumstances.
II Felt only by a few persons at rest, especially on upper floors of buildings. Delicately
suspended objects may swing.
III Felt noticeably indoors, especially on upper floors of buildings, but many people do not
recognize it as an earthquake. Standing motor cars may rock slightly. Vibration like passing of
truck. Duration estimated.
IV During the day felt by many, felt outdoors by few. At night some awakened. Dishes,
windows, doors disturbed; walls make creaking sound. Sensation like heavy truck striking
building. Standing motorcars rocked noticeably.
V Felt by nearly everyone; many awakened. Some dishes, windows, etc. broken; a few
instances of cracked plaster; unstable objects overturned. Disturbance of trees, poles, and other
tall objects sometimes noticed. Pendulum clocks may stop.
VI Felt by all; many frightened and run outdoors. Some heavy furniture moved; a few
instances of fallen plaster or damaged chimneys. Damage slight.
VII Everybody runs outdoors. Damage negligible in buildings of good design and construction'
slight to moderate in well-built ordinary structures; considerable in poorly built or badly designed
structures; some chimneys broken. Noticed by persons driving motor cars.
VIII Damage slight in specially designed structures; considerable in ordinary substantial
buildings with partial collapse; great in poorly built structures. Panel walls thrown out of frame
structures. Fall of chimney, factory stacks, columns, monuments, and walls. Heavy furniture
overturned. Sand and mud ejected in small amounts. Changes in well water. Disturbs persons
driving motor cars.
IX Damage considerable in specially designed structures; well-designed frame structures
thrown out of plumb; damage great in substantial buildings, with partial collapse. Buildings
shifted off foundations. Ground cracked conspicuously, underground pipes broken.
207
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
X Some well-built wood structures destroyed; most masonry and frame structures destroyed
with foundations; ground badly cracked. Rails bent. Landslides considerable from river banks
and steep slopes. Shifted sand and mud. Water splashed (slopped) over banks.
XI Few, if any, (masonry) structures remain standing. Bridges destroyed. Broad fissures in
ground. Underground pipe lines completely out of service. Earth slumps and land slips in soft
ground. Rails bent greatly.
XII Damage total. Waves seen on ground surfaces. Lines of sight and level are distorted.
Objects thrown upward into the air.
2. Richter Magnitude (or Local Magnitude), ML
The Local Magnitude measures the logarithm of the ratio of the maximum displacement to a
standard amplitude at some distance A on a Wood-Anderson seismograph with period 0.8,
damping of 80% of critical, static magnification of 2800. It is valid primarily for earthquakes
with shallow focus.
ML = log A log Ag
A is the maximum amplitude of seismogram from Wood-Anderson instrument measured in
micrometers (1 p.m = 10"* cm) and/f is the amplitude of Ml = 0 earthquake at 1O0 km.
3. Surface Wave Magnitude, Ms
The Surface Wave Magnitude is evaluated from teleseismic instruments, measuring the amplitude
of a 20 second period wave. It is applicable to shallow focus earthquakes primarily.
Ms = log A + a log A + fi
The parameters a and p are representative of the Rayleigh waves from shallow earthquakes.
4. Body Wave Magnitude, mb
The Body Wave Magnitude is also obtained from measurement of teleseismic instruments, but it
measures the magnitude of deep focus earthquakes.
mb =log(y-) + (2(/j,A)
T is the measured wave period and Q is the empirical function of focal depth, h, and the distance,
A. This magnitude is used mostly for intraplate events in the continental United States.
208
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3 ^ -----------:---------------------------- ----------------------------- -------------- -
3 4 5 6 7 8 9
Moment Magnitude, M
Ms M mb ML
Figure 8.1.1 Comparison of Earthquake Measures (from Kanamori 1983)
209
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.2 Attenuation Relationships
1. Boore, Joyner, and Fumal (BJF) relationship (1993,1994)
In equation 8.2.1 below, SD is the mean of seismic demand {PGA or SA, in units o f g) and
M is the moment magnitude of the earthquake. The variable r is the closest horizontal distance, in
kilometers, from the recording station to a point on the surface projection of fault rupture (See
Figure 3.3 in Chapter Three for surface projection).
logi0(SZ)) = ^ &4 +<*ssG s s + <*RSGRS + b{ M- 6 ) + C { M- 6 ) 2 + d ^ r 2 +d 2
I------------- (8.2.1)
+ elogio(Vr2 + d 2 ) + f { 2.881 - l o g 10 VB)
&SA' aRS' c e' / are aNcoefficients. is a factor that converts spectral
velocity (Sv) to spectral acceleration (SA). ag and are coefficients for strike-slip and
reverse-slip faults. GSS and Gftg are the strike-slip and reverse-slip fault-type flags {i.e. 1or 0)
and h is the effective hypocentral depth. Finally, Vg is the effective shear wave velocity for WUS
rock sites, designated as site class B in the 1997 National Earthquake Hazard Reduction Program
(NEHRP) Provisions (FEMA 1997). The values for the terms in Equation 8.2.1 for peak ground
acceleration (PGA) and 5%-damped spectral acceleration (5^) at periods of 0.3 seconds and 1.0
seconds are given in Table 8.2.1 (NIBS 1997 Table 4.4). The standard shape of the response
spectra used in HAZUS predetermines the choice of SAat 0.3 and 1.0 second periods. For more
discussion on this, as well as soil site classes, see Section 3.2.1.
Table 8.2.1 Coefficients for BJF Attenuation
Demand Bsa **ss Ok s
b c d e
/
h vB
PGA 0 -0.136 -0.051 0.229 0 0 -0.778 -0.371 5.57 1400
SA(0.3) -1.67 1.93 2.019 0.334 -0.07 0 -0.893 -0.401 5.94 2130
Sa (I-O) -2.193 1.701 1.755 0.45 -0.014 0 -0.798 -0.698 2.9 1410
210
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2. Sadigh, Chang, Abrahamson, Chio, and Power (SCACP) relationship (1993)
Equation 8.2.2(a) is used for M < 6.5, and Equation 8.2.2(b) is used for M > 6.5.
ln(5Z>) = assGss + orsGrs + 1,0M + 6(8.5 - A/)2 5
+ c In(/2 + <?129649+0-25^ ) + j |n(/? + 2)
In(5D) = assGss +orsGdc + I.1A/ +A(8.5 - A/)2'5
y > ss ss j M > 6.5 (8.2.2(b))
+ cInO? + e 0-4845,+0-524^ ) + /ln(/? + 2)
Similar to the BJF relationship, SD is the mean of seismic demand and M is the moment
magnitude of the earthquake. R is the minimum distance, in kilometers, to the rupture surface. It
is the square-root-sum-of-the-squares of the closest distance to the surface projection of the fault
rupture and the rupture depth at this distance. Moreover, and are coefficients for strike-
slip and reverse-slip/thrust faults, with GSS and G jfg as the fault-type flags. Finally, b,c, and / a r e
coefficients developed for peak ground acceleration and 5%-damped spectral acceleration for
rock sites (i.e. site class B). These coefficients, as well as the values for the c o e f f i c i e n t s a n d
aRS' are 'ven ,n Table 8.2.2.
Table 8.2.2 Coefficients for SCACP Attenuation
Magnitude Demand U.V.V a*s
b c
/
For M < 6.5
PGA -0.624 -0.442 0 -2.1 0
SA(0.3)
-0.057 0.125 -0.017 -2.028 0
Sa(1.0)
-1.705 -1.523 -0.055 -1.8 0
For M > 6.5
PGA -1.274 -1.092 0 -2.1 0
SA(0.3)
-0.707 -0.525 -0.017 -2.028 0
SA(10)
-2.355 -2.173 -0.055 -1.8 0
211
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3. Campbell and Bozorgnia (CB) relationship (1994)
In equation 8.2.3 below, SD is the mean value of peak ground acceleration in gs, and M
is the moment magnitude of the earthquake. R is the distance, in kilometers, to the seismogenic
rupture on the fault. It is the square-root-sum-of-the-squares of the closest distance to the surface
projection of the fault rupture and the depth to the seismogenic zone. The seismogenic zone is the
portion of the seismic fault source (or seismogenic source) that ruptures. Finally, Grs is a fault-
type flag, equal to I for reverse-slip faults and 0 otherwise.
In(5D) = -3.512 + 0.904A/-1.3281n//?2 +0.149e(0647A/)2 (8.2.3 )
+ [1.125-0.112 ln(/?) - 0.0957A/]Gfts + 0.440 - 0.171 ln(/?)
212
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.3 EP Maker Code
1. Aggregation of Residential Losses
Sub AggregateLosses()
Dim i As Integer
Dim j As Integer
Dim numEvents As Integer
Dim numScenarios As Integer
Dim EventNo As String
Dim ScenarioNo As String
Worksheets("Scenarios").Activate
Range("A 1")-Select
Selection.CurrentRegion.Select
NumScenarios = Selection.Count
For i = 1To numScenarios
ScenarioNo = Cel ls(i, 1). Value
ChDir "D:\Hazus\HAZl" & "\" & ScenarioNo
Sheets.Add After:=Worksheets(Worksheets.Count)
Sheets(Worksheets.Count).Select
ActiveSheet.Name = ScenarioNo
Worksheets("Events").Activate
Range("A 1").Select
Selection.CurrentRegion.Select
numEvents = Selection.Count
For j = 1To numEvents
EventNo = Cells(j, 1). Value
ChDir "D:\Hazus\HAZl" & "\" & ScenarioNo & "\" & EventNo
Worksheets("Losses").Activate
With ActiveSheet.QueryTables.Add(Connection:=Array(Array( _
"ODBC;CollatingSequence=ASCII;DBQ=;Deleted=l;Driver={_
Microsoft dBase Driver (*.dbf)}; _
Destination :=Range(" A1"))
.Sql = Array(SELECT DECOLSOC.CS, DECOLSOC.CNS, DECOLSOC.CCD,
DECOLSOC.REL, DECOLSOC.RY, DECOLSOC.TOTAL" & Chr(13) & "" & Chr(10)
& "FROM DECOLSOC DECOLSOC" & Chr(13) & "" & Chr(10) & "WHERE
(DECOLSOC.OCCURES I')" )
.FieldNames = False
.RefreshStyle = xlInsertDeleteCells
.RowNumbers = False
.FillAdjacentFormuIas = False
.RefreshOnFileOpen = False
.HasAutoFormat = True
2 1 3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
.BackgroundQuery = True
.TablesOnlyFromHTML = True
.Refresh BackgroundQuery:=False
.SavePassword = True
.SaveData = True
End With
ActiveWorkbook.Names.Add Name:="Summation", RefersToRICl:=_
'-Losses! R109C1:R109C6"
Application.Goto Reference:="Summation"
Selection.Copy
Sheets(ScenarioNo).Select
Cells(j, l).Select
Selection.PasteSpecial Paste:=xlValues, Operation:=xlNone, SkipBlanks
False, Transpose:=FaIse
Worksheets("Losses").Activate
Cells(l, l).Se!ect
Selection.CurrentRegion.Select
Selection.Clear
Worksheets("Events").Activate
Next j
Worksheets("Scenarios").Activate
Next i
End Sub
2. Residential Losses and Probabilities for Event Loss Table
Sub GetLossesProbsforELT()
Dim i As Integer
Dim j As Integer
Dim numEvents As Integer
Dim numScenarios As Integer
Dim EventNo As String
Dim ScenarioNo As String
Dim Entries As Integer
Dim myRange As Range
Dim Loss As Range
Dim TotalLoss As Long
Worksheets("Scenarios").Activate
Range("A 1").Select
Selection.CurrentRegion.Select
numScenarios = Selection.Count
For i = 1To numScenarios
ScenarioNo = Cel Is(i, 1). Value
ChDir "D:\Hazus\HAZ 1" & "\" & ScenarioNo
214
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Sheets.Add After:=Worksheets(Worksheets.Count)
Sheets(Worksheets.Count).Select
ActiveSheet.Name = ScenarioNo
Worksheets("Events").Activate
Range(" A1").Select
Selection.CurrentRegion.Select
numEvents = Selection.Count / 2
For j = 1To numEvents
EventNo = Cells(j, 1). Value
ChDir "D:\Hazus\HAZl" & "\" & ScenarioNo & "\" & EventNo
Worksheets("Losses").Activate
With ActiveSheet.QueryTables.Add(Connection:=Array(AiTay( "ODBC; _
ConatingSequence=ASCII;DBQ=;DeIeted=l;Driver={Microsoft dBase Driver (*.dbf)}; _
DriverId=277;FIL=dBase HI")), Destination:=Range("Al"))
Sql = ArrayCSELECT DECOLSOC.TRACT, DECOLSOC.OCCU, DECOLSOC.CS,
DECOLSOC.CNS, DECOLSOC.CBD, DECOLSOC.CCD, DECOLSOC.REL, DECOLSOC.RY,
DECOLSOC.TOTAL" & Chr(I3) & "" & Chr(10) & "FROM DECOLSOC DECOLSOC" &
Chr(13) & "" & Chr(10) & "WHERE (DECOLSOC.OCCU-RES I')" )
.FieldNames = False
.RefreshStyle = xlInsertDeleteCells
.RowNumbers = False
.FillAdjacentFormulas = False
.RefreshOnFileOpen = False
.HasAutoFormat = True
.BackgroundQuery = True
.TablesOnlyFromHTML = True
.Refresh BackgroundQuery:=False
.SavePassword = True
.SaveData = True
End With
Range("A 1").Select
Selection.CurrentRegion.Select
Entries = Selection.Count / 9
Range(" A1").Select
Set myRange = ActiveCell.CurrentRegion
Set Loss = myRange.Offset(myRange.Rows.Count).Rows(l)
Loss.Formula = "=SUM(" & myRange.Columns(l).Address(False, False) & ")"
Cells(Entries + 1, 9).Select
TotalLoss = Selection.Value
Selection.Copy
Sheets(ScenarioNo).Select
Cells(j, l).Select
Selection.PasteSpecial _
Paste:=xlValues, Operation:=xlNone, SkipBlanks:= False, Transpose:=False
Worksheets("Events"). Activate
Cells(j, 2).Select
215
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Selection.Copy
Sheets(ScenarioNo).Select
Cells(j, 2).Select
Selection.PasteSpecial _
Paste:=xlValues, Operation:=xlNone, SkipBlanks:= False, Transpose:=False
WorksheetsC'Losses").Activate
CeIIs(l, I).Select
Selection.CurrentRegion.Select
Selection.Clear
Worksheets("Events").Activate
Next j
Worksheets("Scenarios"). Activate
Next i
End Sub
3. Coefficients of Variation for Event Loss Table
Sub CalculateCOV()
Dim i As Integer
Dim j As Integer
Dim numEvents As Integer
Dim numScenarios As Integer
Dim EventNo As String
Dim ScenarioNo As String
Dim Tracts As Integer
Dim my Range As Range
Dim COV As Range
Dim TotalCOV As Long
Worksheets("Scenarios").Activate
Range("A 1").Select
Selection.CurrentRegion.Select
numScenarios = Selection.Count
For i = I To numScenarios
ScenarioNo = Cells(i, 1). Value
ChDir "D:\Hazus\HAZl" & "V & ScenarioNo
Sheets.Add After:=Worksheets(Worksheets.Count)
Sheets(Worksheets.Count).Select
ActiveSheet.Name = ScenarioNo
Worksheets("Events").Activate
Range( A1").Select
Selection.CurrentRegion.Select
numEvents = Selection.Count
For j = 1 To numEvents
216
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
EventNo = Cel ls(j, 1). Value
ChDir "D:\Hazus\HAZ 1" & "V & ScenarioNo & "\" & EventNo
Worksheets("COV").Activate
With ActiveSheet.QueryTables.Add(Connection:=Array(Array( "ODBC;CollatingSequence= _
ASCII; DBQ=;Deleted=l; Driver={Microsoft dBase Driver (*.dbf)} ;DriverId=277;FIL=dBase
III")), Destination :=Range("A1"))
Sql = ArrayC'SELECT ctgost-SLIGHT, ctgost.MODERATE, ctgost-EXTENSIV,
ctgost.COMPLETE, ctgonsa.SLIGHT, ctgonsa.MODERATE, ctgonsa-EXTENSIV,
ctgonsa.COMPLETE, ctgonsd.SLIGHT, ctgonsd.MODERATE, ctgonsd.EXTENSIV,
ctgonsd.COMPLETE" & Chr(13) & "" & Chr(10) & "FROM ctgonsa ctgonsa, ctgonsd ctgonsd,
ctgost ctgost" & Chr(13) & "" & Chr(10) & "WHERE ctgonsa.OCCU = ctgonsd.OCCU AND
ctgonsa.OCCU = ctgost.OCCU AND ctgonsd.OCCU = ctgost.OCCU AND ctgonsa.TRACT =
ctgost.TRACT AND ctgonsd.TRACT = ctgost-TRACT AND ctgonsd.TRACT = ctgonsa.TRACT
AND ((ctgost.OCCU-RES') AND (ctgonsa.OCCU=RES') AND (ctgonsd.OCCU=RES))" )
.FieldNames = False
.RefreshStyle = xlInsertDeleteCells
.RowNumbers = False
.FillAdjacentFormulas = False
.RefreshOnFileOpen = False
.HasAutoFormat = True
.BackgroundQuery = True
.TablesOnlyFromHTML = True
.Refresh BackgroundQuery:=False
.SavePassword = True
SaveData = True
End With
Range(" A1").Select
Selection.CurrentRegion.Select
Tracts = Selection.Count / 12
Range("M 1").Select
ActiveCell.FormulaRIC 1="=RC[-12]*0.36+RC[-11 ]* 1,8+RC[-10]*9.0 l+RC[-9]* 18.02"
Range("N 1").Select
ActiveCell.FormulaRIC 1= =RC[-9]*0.49+RC[-8]*2.47+RC[-7]* 12.34+RC[-6]*24.68"
RangeC'01").Select
ActiveCell.FormulaRIC 1= "=RC[-6]*0.92+RC[-5]*4.58+RC[-4]*22.92+RC[-3]*45.84"
Range("P 1").Select
ActiveCell.FormulaRIC 1="=SQRT((0.36A2*RC[-15]+1.8A2*RC[-14]+9.01A2* _
RC[-13]+18.02A2*RC[-12])-RC[-3]A2)"
Range("Q 1").Select
ActiveCell.FormulaRIC 1="=SQRT((0.49A2*RC[-12]+2.47A2*RC[-11]+I2.34A2*_
RC[-10]+24.68A2*RC[-9]>RC[-3]A2)"
Range("Rl").Select
ActiveCell.FormulaRIC 1= "=SQRT((0.92A2*RC[-9]+4.58A2*RC[-8]+22.92A2* _
RC[-7]+4S.84A2*RC[-6]>RC[-3]A2)"
Range("S 1").Select
ActiveCell.FormulaRIC 1 = "=SUM(RC[-6]:RC[-4])/( 18.02+24.68+45.84)"
Range("Tl ").Select
217
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ActiveCell.FormulaRIC 1= "=SUM(RC[-4]:RC[-2])/( 18.02+24.68+45.84)"
Range("U I ").Select
ActiveCell.FormulaRl C1 = "=IF(RC[-l]=0,"""",RC[-2]/RC[-l])"
Range("MI :U 1").Select
Application.CutCopyMode = False
Selection.Copy
Range("M2:U l08").Select
ActiveSheet.Paste
Range("A 1").Select
Set myRange = ActiveCell.CurrentRegion
Set COV = myRange.OfFset(myRange.Rows.Count).Rows(l)
COV.Formula = "=AVERAGE(" & myRange.Columns( 1).Address(False, False) & ")"
CelIs(Tracts + 1, 21).Select
TotalCOV = Selection.Value
Selection.Copy
Sheets(ScenarioNo).Select
CellsG, l).SeIect
Selection.PasteSpecial Paste:=xlValues, Operation:=xINone, SkipBlanks:= False, _
T ranspose:=False
WorksheetsC'COV"). Activate
Cells(l, l).Select
Selection.CurrentRegion.Select
Selection.Clear
Worksheets("Events").Activate
Next j
Worksheets("Scenarios").Activate
Next i
End Sub
218
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.4 Scenario Builder Input File
SCENID SCEN
DESC
ATTEN MAG SSRLEN SRLEN RORN LAT LONG DEPTH FAUL
TTYPE
EQDUR
ATION
PROBAB
CA00001 SF2 10 5.0 1.41 3.39 0 37.66 -122.05 5 0 2 0.06862
CA00002 SF2 10 5.5 3.31 6.92 0 37.66 -122.05 5 0 2 0.01220
CA00003 SF2 10 6.0 7.76 14.13 0 37.66 -122.05 5 0 2 0.00217
CA00004 SF2 10 6.5 18.20 28.84 0 37.66 -122.05 5 0 2 0.00039
CA00005 SF2 10 6.9 35.97 51.05 0 37.59 -121.97 5 0 2 0.00599
CA00006 SF2 10 7.0 42.66 58.88 0 37.59 -121.97 5 0 2 0.00599
CA00007 SF2 10 7.5 100.00 120.23 0 37.59 -121.97 5 0 3 0.00599
CA00008 SF3 10 5 1.41 3.39 0 37.83 -122.21 5 0 1 0.06862
CA00009 SF3 10 5 1.41 3.39 0 37.90 -122.27 5 0 1 0.06862
CA00010 SF3 10 5 1.41 3.39 0 37.96 -122.33 5 0 1 0.06862
CA00011 SF3 10 5.5 3.31 6.92 0 37.76 -122.16 5 0 1 0.01220
CA00012 SF3 10 5.5 3.31 6.92 0 37.83 -122.21 5 0 1 0.01220
CA00013 SF3 10 5.5 3.31 6.92 0 37.9 -122.27 5 0 1 0.01220
CA00014 SF3 10 5.5 3.31 6.92 0 37.96 -122.33 5 0 1 0.01220
CA00015 SF3 10 5.5 3.31 6.92 0 38.02 -122.38 5 0 1 0.01220
CA000I6 SF3 10 6 7.76 14.13 0 37.76 -122.16 5 0 I 0.00217
CA00017 SF3 10 6 7.76 14.13 0 37.83 -122.21 5 0 1 0.00217
CA00018 SF3 10 6 7.76 14.13 0 37.90 -12227 5 0 1 0.00217
CA00019 SF3 10 6 7.76 14.13 0 37.96 -122.33 5 0 1 0.00217
CA00020 SF3 10 6 7.76 14.13 0 38.02 -122.38 5 0 1 0.00217
CA00021 SF3 10 6.5 18.20 28.84 0 37.97 -122.34 5 0 2 0.00039
CA00022 SF3 10 6.5 18.20 28.84 0 37.81 -122.2 5 0 2 0.00039
CA00023 SF3 10 6.9 35.97 51.05 0 37.89 -12227 5 0 2 0.00599
CA00024 SF3 10 7.5 100.00 120.23 0 37.89 -12227 5 0 j 0.00599
CA00025 SF4 10 6.0 7.76 14.13 0 38.19 -122.43 5 0
2 0.00265
CA00026 SF4 10 6.5 18.20 28.84 0 38.19 -122.43 5 0 2 0.00047
CA00027 SF4 10 7.0 42.66 58.88 0 38.31 -122.55 5 0 3 0.00450
CA00028 SF6 10 6.0 7.76 14.13 0 37.75 -121.97 5 0 2 0.00190
CA00029 SF6 10 6.5 18.20 28.84 0 37.75 -121.97 5 0 2 0.00034
CA00030 SF6 10 6.8 30.34 44.26 0 37.65 -121.92 5 0 2 0.00685
CA00031 SF6 10 7.0 42.66 58.88 0 37.65 -121.92 5 0 2 0.00685
CA00032 SF6 10 7.5 100.00 120.23 0 37.65 -121.92 5 0 3 0.00685
CA00033 SF7 10 6.0 7.76 14.13 0 38.04 -122.04 5 0 2 0.00222
CA00034 SF7 10 6.5 18.20 28.84 0 38.04 -122.04 5 0 2 0.00039
CA00035 SF7 10 6.9 35.97 51.05 0 38.17 -122.09 5 0 3 0.00568
CA00036 SFI5 10 6.0 7.76 14.13 0 37.83 -121.82 5 0 2 0.00075
CA00037 SF15 10 6.5 18.20 28.84 0 37.83 -121.82 5 0 2 0.00013
CA00038 SF15 10 6.9 35.97 51.05 0 37.7 -121.72 5 0 3 0.00192
CA00039 SA9 10 6.0 7.76 14.13 0 37.7 -122.5 5 0 2 0.00979
CA00040 SA9 10 6.0 7.76 14.13 0 37.5 -122.3 5 0 2 0.00979
219
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CA00041 SA9 10 6.5 18.20 28.84 0 37.5 -122.3 5 0 2
0.00174
CA00042 SA9 10 6.5 18.20 28.84 0 37.7 -122.5 5 0 2 0.00174
CA00043 SA9 10 7.0 42.66 58.88 0 37.5 -122.3 5 0 2 0.00250
CA00044 SA9 10 7.0 42.66 58.88 0 37.7 -122.5 5 0 2
0.00250
CA00045 SA9 10 7.1 50.58 67.92 0 37.7 -122.5 5 0 3 0.00250
CA00046 SA9 10 7.5 100.00 120.23 0 37.5 -122.3 5 0 3 0.00250
220
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.5 Benefits Questionnaire
Age: <25 yrs. 25-35 yrs. 35-45 yrs. 45-55 yrs.
Yrs. in practice: <5 yrs. 5-10 yrs. 10-15 yrs. 15-20 yrs.
55-65 yrs.
20-25 yrs.
>65 yrs.
>25 yrs.
Your employer: Structural Engineering Consulting or Design Firm
Government (City, State, Federal)
University
Self-employed
Other ______________
Please rate your experience with post-earthquake investigation of residential wood frame construction:
No Experience Extensive
at all Experience
0 1 2 3 4 5 6 7 8 9 10
In considering how to mitigate a structure to reduce the damage incurred during an earthquake, there is a
need to relate the cost o f the mitigation technique to its expected benefits. According to the ATC-13
guidelines, the mean damage factors (MDF or expected ratio of dollar loss to replacement value) caused by
ground shaking for a residential wood frame building of standard construction for a given Modified
Mercalli Intensity are given below. (For example, a MDF of 25 is defined as an expected dollar loss o f
532,500 for a structure that is worth 5130,000.) Consider a 2,200 square-foot wood frame residential
building in the city of Oakland, CA. It is a pre-1940 two-story structure with a market value of
5130,000. It has 24 unbraced cripple walls and it is inadequately bolted to its foundation. Do you
agree with the values given in the ATC-13 guidelines? If not, please indicate your estimate for the mean
damage factors for ground shaking only (i.e. no collateral hazards such as ground failure, fault rapture,
inundation, or fire following). How do you feel about this estimate?
YOUR
MMI ATC-
VI 0.8
VII 1.5
VIII 4.7
X 9.2
X 19.8
XI 24.4
XII 37.3
Not Confident
at all
0 I 2 3 4 5 6 8 9
Very
Confident
10
0 1 2
*>
J 4 5 6 8 9 10
0 1 2 3 4 5 6 8 9 10
0 1 2 3 4 5 6 8 9 10
0 I
2
3 4 5 6 8 9 10
0 1
2
3 4 5 6 8 9 10
0 1 2 3 4 5 6 8 9 10
What do you believe are the determining factors in the evaluation of this MDF?
Not Important
at all
Very
Important
Square footage o f structure 0 1 2 3 4 5 6 8 9 10
Age of structure 0 1
2
3 4 5 6 8 9 10
Number o f stories o f structure 0 1
2
3 4 5 6 8 9 10
Value o f structure 0 1 2 3 4 5 6 8 9 10
Height of cripple wall 0 I 2 3 4 5 6 8 9 10
Inadequate bolt to foundation
Any other factors?
0 1 2 3 4 5 6 8 9 10
221
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0 1 2 3 4 5 6 7 8 9 ! 0
0 1 2 3 4 5 6 7 8 9 10
Consider the same 2,200 square-foot pre-1940 two-story wood frame residential building in the city of
Oakland, CA, with a market value of $ 130,000. One way to reinforce the structure is to brace its cripple
wall. Please estimate how much the mean damage factors (MDF) would be reduced with this type of
retrofit for an earthquake with the following Modified Mercalli Intensities. How do you feel about this
estimate?
% REDUCTION Not Confident Very
MMI AFTER MITIGATION at all Confident
VI 0 1 2 3 4 5 6 8 9 10
VII 0 1 2 3 4 5 6 8 9 10
VIII 0 I 2 3 4 5 6 8 9 10
X 0 I 2 3 4 5 6 8 9 10
X 0 1 2 3 4 5 6 8 9 10
XI 0 I 2 3 4 5 6 8 9 10
XII 0 1 2 3 4 5 6 8 9 10
Consider the same 2,200 square-foot pre-1940 two-story wood frame residential building in the city of
Oakland, CA, with a market value of 5130,000. Another way to reinforce it is to bolt the structure to its
foundation. Please estimate how much the mean damage factors (MDF) would be reduced with this type of
retrofit for an earthquake with the following Modified Mercalli Intensities. How do you feel about this
estimate?
% REDUCTION Not Confident Very
MMI AFTER MITIGATION at all Confident
VI 0 1 2 3 4 5 6 8 9 10
VII 0 1 2 3 4 5 6 8 9 10
VIII 0 1 2 3 4 5 6 8 9 10
X 0 1 2 3 4 5 6 8 9 10
X 0 1 2 3 4 5 6 8 9 10
XI 0 1 2 3 4 5 6 8 9 10
XII 0 1 2 3 4 5 6 8 9 10
Now, consider a combination o f these two mitigation techniques: bracing the structures cripple wall and
bolting the structure to its foundation. This is a multi-level" approach to reduce the potential damage from
an earthquake. Please estimate how much the mean damage factors (MDF) would be reduced with this type
of retrofit for an earthquake with the following Modified Mercalli Intensities. How do you feel about this
estimate?
MMI
VI
% REDUCTION
AFTER MITIGATION
Not Confident
at all
0 1 2 3 4 5 6 8 9
Very
Confident
10
VII 0 1 2 3 4 5 6 8 9 10
VIII 0 1 2 3 4 5 6 8 9 10
X 0 1 2 3 4 5 6 8 9 10
X 0 1 2 3 4 5 6 8 9 10
XI 0 1 2 3 4 5 6 8 9 10
XII 0 1 2 3 4 5 6 8 9 10
222
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Finally, consider the TOTAL number of residential wood frame buildings in the city o f Oakland, CA,
which is approximately 90,000 (including both pre-1940 and post-1940 structures). Please estimate the
number of these structures that are damaged requiring more than $1,000 to fix for an earthquake with the
following Modified Mercalli Intensities. How do you feel about this estimate?
NUMBER OF Not Confident Very
MMI STRUCTURES
DAMAGED
at all Confident
VI 0 1 2 3 5 6 8 9 10
VII 0 1 2 j 5 6 8 9 10
VIII 0 1 2 3 5 6 8 9 10
X 0 1 2 3 5 6 8 9 10
X 0 1 2 3 5 6 8 9 10
XI 0 I 2 3 5 6 8 9 10
XII 0 1 2
3 5 6 8 9 10
COMMENTS:
223
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8 . 6 Costs Questionnaire
Age: <25 yrs. 25-35 yrs. 35-45 yrs. 45-55 yrs. 55-65 yrs. >65 yrs.
Yrs. in practice: <5 yrs. 5-10 yrs. 10-15 yrs. 15-20 yrs. 20-25 yrs. >25 yrs.
Please rate your experience with retrofitting o f residential wood frame construction:
No Experience Extensive
at all Experience
0 1 2 3 4 5 6 7 8 9 10
In considering how to mitigate a structure to reduce the damage incurred during an earthquake, there is a
need to relate the cost o f the mitigation technique to its expected benefits. Consider a 2,200 square-foot
wood frame residential building in the city of Oakland, CA. It is a pre-1940 two-story structure with
a market value of $130,000. It has 24 unbraced cripple walls and it is inadequately bolted to its
foundation.
Assuming adequate access to the foundation, one way to reinforce the structure is to brace its cripple wall.
How much do you estimate it would cost to brace the structures cripple wall?
$1,000-53,000 $3,000-55,000 $5,000-57,000 $7,000-59,000 >$9,000
Within this range o f cost, what is the median value o f expected cost? ____________________
What percentage o f the cost is the estimated cost of the material?
<10% 10% 20% 30% 40% 50% 60% 70% 80% 90% >90%
What percentage of the cost is the estimated cost for labor?
<10% 10% 20% 30% 40% 50% 60% 70% 80% 90% >90%
How do you feel about this estimate?
Not Confident
at all
Very
Confident
0 1 2 3 4 5 6 7 8 9 10
What do you believe are the determining factors in the cost of this retrofit?
Not Important
at all
Square footage of structure 0 1 2 3 4 5 6 7 8 9
Very
Important
10
Age of structure 0 1 2 3 4 5 6 7 8 9 10
Number of stories of structure 0 I 2
3 4 5 6 7 8 9 10
Value o f structure 0 1 2 3 4 5 6 7 8 9 10
Height o f cripple wall 0 1 2 3 4 5 6 7 8 9 10
Accessibility to foundation 0 1 2 3 4 5 6 7 8 9 10
Any other factors?
0 1 2 3 4 5 6 7 8 9 10
0 1 2 3 4 5 6 7 8 9 10
224
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Consider the same 2,200 square-foot pre-1940 two-story wood frame residential building in the city of
Oakland, CA, with a market value o f $130,000. Assuming adequate access to the foundation, another way
to reinforce it is to bolt the structure to its foundation.
How much do you estimate it would cost to bolt the structure to its foundation?
$1,000-53,000 $3,000-55,000 $5,000-57,000 $7,000-59,000
Within this range o f cost, what is the median value of expected cost? _____________
What percentage of the cost is the estimated cost o f the material?
>$9,000
<10% 10% 20% 30% 40% 50% 60% 70% 80% 90% >90%
What percentage o f the cost is the estimated cost for labor?
<10% 10% 20% 30% 40% 50% 60% 70% 80% 90% >90%
How do you feel about this estimate?
Not Confident Very
at all Confident
0 1 2 3 4 5 6 7 8 9 10
What do you believe are the determining factors in the cost of this retrofit?
Not Important Very
at all Important
Square footage o f structure 0 I 2 3 4 5 6 8 9 10
Age of structure 0 1 2 3 4 5 6 8 9 10
Number o f stories o f structure 0 1 2 3 4 5 6 8 9 10
Value o f structure 0 I 2 3 4 5 6 8 9 10
Height o f cripple wall 0 1 2 3 4 5 6 8 9 10
Accessibility to foundation 0 1 2 3 4 5 6 8 9 10
Any other factors?
0 1 2 3 4 5 6 8 9 10
0 1 2 J 4 5 6 8 9 10
Now, consider a combination o f these two mitigation techniques: bracing the structure's cripple wall and
bolting the structure to its foundation. This is a multi-level approach to reduce the potential damage from
an earthquake.
How much do you estimate this multi-level mitigation technique would cost?
$l,000-$3,000 $3,000-55,000 $5,000-57,000 $7,000-59,000 >$9,000
Within this range o f cost, what is the median value of expected cost? ____________________
What percentage o f the cost is the estimated cost of the material?
<10% 10% 20% 30% 40% 50% 60% 70% 80% 90% >90%
225
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
What percentage o f the cost is the estimated cost for labor?
<10% 10% 20% 30% 40% 50% 60% 70% 80% 90% >90%
How do you feel about this estimate?
Not Confident Very
at all Confident
0 1 2 3 4 5 6 7 8 9 10
What do you believe are the determining factors in the cost of this retrofit?
Not Important
at all
Square footage o f structure 0 1 2 3 4 5 6 7 8 9
Very
Important
10
Age of structure 0 1 2 3 4 5 6 7 8 9 10
Number of stories o f structure 0 1 2 3 4 5 6 7 8 9 10
Value o f structure 0 1 2 3 4 5 6 7 8 9 10
Height of cripple wall 0 1 2 j 4 5 6 7 8 9 10
Accessibility to foundation 0 I 2 3 4 5 6 7 8 9 10
Any other factors?
0 1 2 3 4 5 6 7 8 9 10
0 1 2 3 4 5 6 7 8 9 10
COMMENTS:
226
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.7 Benefits Data
Before Mitigation
i
E,
yi.it Qw yi.ni Ci.ni yi.nu Ci.nu yi.K Cox yuc Cue yua
Cija
-V/JKH Cijai
1 6 l 5 2 5 5 5 10 5 15 5 20 5 25 5
2 8 0 3 5 3 10 5 10 8 20 8 25 8 30 8
3 8 0.8 7 2.1 7 6.3 7 1 1 6 23 4 30 3 34 2
4 10 5 7 7 7 7 7 9 7 15 7 30 7 35 7
5 3 0.8 3 1.5 3 4.7 3 9.2 3 19.8 3 24.4 3 37.3 3
6 3 0 4 0 4 2 4 9.2 10 19.8 10 24.4 10 37.3 10
7 4 0.8 8 1.5 8 4.7 8 9.2 8 19.8 7 24.4 7 37.3 6
8 5 0.8 8 1.5 8 4.7 8 9.2 8 19.8 8 24.4 8 37.3 8
9 6 0.8 10 1.5 10 5 8 12 7 22 8 28 9 37.3 10
10 7 0.8 5 1.5 5 4.7 5 9.2 5 19.8 5 24.4 5 37.3 5
11 8 0.8 5 1.5 6 4.7 6 9.2 6 19.8 7 24.4 7 37.3 7
12 10 0.8 10 1.5 10 4.7 10 9.2 10 19.8 10 24.4 10 37.3 10
13 10 0.8 8 1.5 8 4.7 8 92 8 19.8 3 24.4 2 37.3 2
14 10 0.8 9 1.5 9 4.7 9 9.2 9 19.8 9 24.4 9 37.3 9
15 10 0.8 8 1.5 8 4.7 8 9.2 8 19.8 8 24.4 8 37.3 8
16 10 0 8 3.85 8 7.69 8 38.46 9 38.46 9 38.46 8 38.46 7
17 8 1 4 1.5 5 10 7 20 7 27 5 35 4 40 3
18 2 0.96 3 1.8 3 5.64 3 11.04 3 23.76 3 29.28 3 44.76 3
19 6 1 5 2 5 6 5 15 5 20 5 30 5 45 5
20 10 2 7 3 7 7 6 12 6 25 5 30 5 45 5
21 10 1 8 2 8 5 7 10 7 25 6 35 6 45 6
22 4.5 0 9 0 9 5 7 10 5 20 5 30 5 50 5
23 5 1 5 3 5 6 5 12 5 25 5 35 5 50 6
24 5 5 8 5 8 10 8 15 7 30 7 40 7 50 7
25 7 2 6 15 6 20 6 35 6 40 7 45 7 50 8
26 7 0 6 1 6 5 6 15 6 25 6 30 6 50 6
27 5 1.5 7 4 7 7 7 19 7 25 7 46 7 60 7
28 7 0.8 8 1.5 8 4.7 8 9.4 5 19.8 5 40 3 60 3
29 7 1 8 1.5 8 5 8 20 5 40 4 50 4 60 3
30 8 2 5 5 5 10 5 15 5 25 5 40 5 60 5
31 9 5 7 10 7 15 7 20 7 50 7 50 7 60 7
32 9 5 6 10 6 20 6 30 6 40 6 50 6 60 6
33 9 0 3 2 3 10 3 15 6 30 6 50 6 60 6
34 10 2 7 5 7 10 7 15 7 25 7 40 7 60 7
35 5 1 6 3 6 10 6 25 6 35 6 50 6 65 6
36 2 5 5 5 5 30 5 30 5 50 5 70 5 70 5
37 6 0.8 8 1.5 8 20 8 40 8 70 6 70 9 70 9
38 10 1 10 1.5 10 5 10 10 10 25 9 40 9 70 9
227
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
39 1 0 5 8 1 0 8 30 8 40 8 50 8 60 8 70 8
40 5 0 . 8 7 1.5 6 5 5 1 0 4 25 4 40 4 75 4
41 7 1.5 9 2.5 8 5 7 2 0 0 40 0 60 2 75 2
42 7 1 0 7 25 7 37 7 75 7 75 7 75 7 75 7
43 8 1 3 1 0 4 15 5 25 6 40 6 50 6 75 6
44 1 0 1 8 3 8 8 8 2 0 8 30 8 75 6 75 6
45 1 0 1.5 6 4.7 3 34.6 7 53.8 6 65.4 6 76.9 6 76.9 6
46 3 1 3 1 0 3 2 0 3 30 3 50 3 65 3 80 3
47 6 1 5 1 0 5 2 0 5 35 5 50 5 65 5 80 5
48 7 0.5 5 5 5 30 5 50 5 60 5 70 5 80 5
49 7 2 3 5 3 1 0 4 2 0 5 30 6 50 7 80 8
50 8 15 7 35 7 40 7 50 7 60 7 70 7 85 7
51 7 5 7 1 0 7 2 0 7 30 7 50 7 70 7 90 7
52 8 5 6 15 6 30 7 45 7 60 8 75 8 90 9
53 9 1.9 6 2.7 6 7.7 6 26.9 7 53.8 8 73.1 8 92.3 7
54 1 5 5 15 5 25 5 50 5 1 0 0 5 1 0 0 5 1 0 0 5
55 5 1.5 7 5 6 1 0 5 25 4 60 4 90 4 1 0 0 4
56 5 0 5 1 0 5 30 5 65 5 80 5 95 5 1 0 0 5
57 5 1 0 5 2 0 5 1 0 0 8 1 0 0 9 1 0 0 1 0 1 0 0 1 0 1 0 0 1 0
58 6 0 8 0.769 8 7.69 8 38.46 8 1 0 0 8 1 0 0 8 1 0 0 8
59 6 3 7 6 6 1 2 5 30 4 1 0 0 3 1 0 0 3 1 0 0
3
60 6 1 3 15 4 30 5 50 7 90 8 1 0 0 9 1 0 0 9
61 7 5 5 1 0 5 25 5 50 5 75 5 90 5 1 0 0 5
62 8 2 1 1 0 4 2 0 5 30 5 50 5 1 0 0 1 0 1 0 0 1 0
63 8 1 0 9 15 9 2 0 9 25 8 50 8 75 8 1 0 0 8
64 9 0 8 6.15 5 15.39 7 30.77 6 61.54 5 1 0 0 1 0 1 0 0 1 0
65 9.5 0 3 5 3 25 3 50 3 70 3 1 0 0 3 1 0 0 3
6 6 1 0 0 . 8 1 0 1.5 8 2 0 6 50 5 1 0 0 6 1 0 0 8 1 0 0 1 0
67 1 0 0.154 9 0.769 9 46.15 9 69.23 9 92.31 9 1 0 0 9 1 0 0 9
6 8 1 0 1 0 8 33 9 40 9 85 8 95 9 1 0 0 1 0 1 0 0 1 0
69 1 0 2
8 5 8 1 0 8 2 0 8 40 8 70 8 1 0 0 8
After Mitigation
/
E, yi,yi Qn yi.m Ci.ni yi.nn
Ci'Vni
yi.rx Cijx yuc
Cue yua Cva yuat
Cuai
1 6 1 5 1 5 2 5 5 5 8 5 15 5 2 0 5
2 8 0 8 0 8 4 8 4 8 1 2 8 15 5 18 3
3 8 0.04 2 0 . 2 1 5 1.323 7 6.49 5 11.96 3 15 1 18.02 1
4 1 0 4.5 9 6.3 9 6.3 9 7.2 5 1 2 5 24 5 28 5
5 3 0 . 8 5 1.5 5 4.7 5 7.36 5 7.92 5 7.32 5 7.476 5
6 3 0 8 0 7 0 6 0 6 1.98 6 2.448 6 5.607 6
7 4 0.704 8 1.305 8 3.854 7 7.176 7 12.87 6 13.42 6 11.19 5
8 5 0.4 7 0.75 7 2.3 7 4.5 7 1 0 7 1 2 7 18 7
9 6 0 1 0 0 1 0 1 8 4.8 8 17.6 9 28 1 0 37.3 1 0
228
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10 7 0.76 4 1.395 4 4.324 4 8.372 4 17.82 4 21.47 4 31.71 4
11 8 0.4 8 0.75 8 2.35 8 4.6 8 9.9 8 12.2 8 18.65 8
12 10 0.8 9 1.5 9 4.7 9 9.2 9 18.41 9 20.98 9 29.47 9
13 10 0.4 4 0.9 4 3.29 4 6.44 3 13.86 3 17.08 3 26.11 3
14 10 0 8 0 8 0 8 8.28 8 17.82 8 23.18 8 37.3 8
15 10 0.2 6 0.375 6 1.175^ 6 2.3 6 4.95 6 6.1 6 9.325 6
16 10 1.54 9 2.31 9 6.15 9 15.38 9 23.08 8 30.77 7 38.46 7
17 8 1 4 1.25 5 6 7 10 7 15 5 25 4 30 3
18 2 0.64 3 1.2 3 3.76 3 7.36 3 15.84 3 19.52 3 29.84 3
19 6 1 5 2 5 5.1 5 12.75 5 17 5 22.5 5 33.75 5
20 10 0.4 8 0.4 7 1.5 6 2.5 6 4 6 8 6 10 6
21 10 0.4 9 0.8 8 2 8 4 7 10 7 14 6 18 6
22 4.5 0 5 0 5 5 5 8 5 14 5 21 5 40 5
23 5 1 5 3 5 4 5 6 5 8 5 12 5 15 5
24 5 4.25 7 4.25 7 7 7 10.5 7 18 7 20 7 20 7
25 7 0.1 2 0.1 2 0.5 3 5 5 15 4 18 5 25 6
26 7 0 5 0 5 3 5 8 5 18 5 25 5 30 5
27 5 0.15 6 0.8 6 2.1 5 5.7 5 10 5 27.6 5 36 5
28 7 0.08 7 0.15 7 0.47 7 0 7 0 7 40 7 60 7
29 7 1 9 1.5 9 2.5 9 8 9 8 7 10 7 12 7
30 8 1.6 5 3.5 5 6 5 7.5 5 10 5 12 5 18 5
31 9 0.5 6 1 6 1.5 6 4 6 15 6 20 6 24 6
32 9 1.5 7 4 7 10 7 18 7 28 7 40 7 54 7
33 9 0 3 2 3 10 3 12 5 15 5 25 5 35 5
34 10 0.4 6 1.5 6 4 6 7.5 6 15 6 28 6 48 6
35 5 1 6 2 6 4 6 7 6 10 6 15 6 20 6
36 2 0.5 5 0.5 5 3 5 3 5 5 5 7 5 7 5
37 6 0 8 0 8 0 4 10 4 17.5 4 70 8 70 8
38 10 0.85 10 1.275 10 4.25 10 8.5 10 21.25 9 34 9 59.5 9
39 10 5 7 10 7 24 7 28 7 31.5 7 34.2 7 35 7
40 5 0.16 7 0.45 7 2.5 6 7 6 21.25 5 36 5 71.25 4
41 7 0.6 4 1.375 4 3.25 4 15 4 34 4 54 4 75 4
42 7 1 7 2.5 7 3.7 7 15 7 45 7 37.5 7 75 7
43 8 1 5 9 6 11.25 6 15 7 22 7 25 7 26.25 7
44 10 1 8 3 8 7 8 10 8 20 8 25 8 35 8
45 10 1.5 6 4.7 6 5 7 11.5 6 15.4 4 38.5 4 65.4 3
46 3 0.5 3 1 3 2 3 5 3 10 3 15 3 20 3
47 6 1 7 9 5 17 5 26.25 5 37.5 5 55.25 5 80 5
48 7 0.125 9 1.25 9 7.5 9 12.5 9 15 9 17.5 9 20 9
49 7 0.5 3 1 3 2 3 7 4 12 4 30 5 45 6
50 8 9 9 21 9 24 9 30 9 36 9 42 9 51 9
51 7 0 7 0 7 5 7 15 7 30 7 45 7 60 7
52 8 2.5 6 7.5 6 15 6 22.5 6 30 6 37.5 6 45 6
229
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
53 9 1.33 4 1.89 4 5.39 4 9.415 5 18.83 5 43.86 6 73.84 6
54 1 0 5 0 5 0 5 0 5 50 5 100 5 100 5
55 5 0.6 7 3 7 7.5 6 21.25 6 54 5 85.5 5 100 4
56 5 0 4 9.5 4 28.5 4 58.5 5 68 6 76 6.5 75 7
57 5 0 8 0 8 10 7 20 7 30 7 50 5 100 8
58 6 0 8 0.769 8 3.85 8 3.85 8 7.69 8 38.46 8 100 8
59 6 1.5 9 2.4 8 3.6 7 6 6 40 5 60 5 80 5
60 6 0.99 7 13.05 8 22.5 8 30 8 27 9 60 9 100 10
61 7 4.25 2 8.5 2 21.25 2 42.5 2 63.75 2 76.5 2 85 2
62 8 1 1 5 4 10 5 15 5 25 5 40 4 100 3
63 8 5 8 5 8 10 8 15 8 20 8 50 8 75 8
64 9 0 10 2.308 8 7.692 6 16.92 5 34.62 6 69.23 7 100 10
65 9.5 0 8 5 8 22.5 8 35 8 42 8 40 8 40 8
66 10 0.8 10 1.5 10 5 8 12.5 5 50 5 100 8 100 10
67 10 0 7 0 7 0 7 6.923 7 46.15 7 60 7 70 7
68 10 0 10 1.65 8 6 7 42.5 6 80.75 7 95 8 100 10
69 10 1.6 8 3.5 8 6 8 10 8 18 8 28 8 35 8
230
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.8 Costs Data
/
E ,
Ci y,. total yi.material
1 3 8 3500 700
2 3 4 6000 3000
3 5 5 8500 1700
4 5 4 4500 900
5 5 9 7000 2800
6 5 10 3500 1050
7 5 5 4000 800
8 6 7 2900 580
9 6 8 4500 1800
10 7 9 7000 2100
11 7 9 5500 2200
12 7 4 10000 4000
13 7 8 4500 1800
14 8 8 3500 700
15 8 8 6500 1300
16 8 8 5000 500
17 8 3 7500 1500
18 8 9 9000 1800
19 8 8 3800 760
20 8 7 2500 250
21 8 9 6000 1200
22 8.5 8 2000 200
23 9 9 6000 1200
24 9 8 9500 2850
25 9 9 8000 1200
26 9 9 2300 920
27 10 8 5000 1000
28 10 8 6000 1200
29 10 5 6500 650
30 10 8 3600 1080
31 10 6 7000 1400
32 10 8 7300 730
33 10 6 4500 900
34 10 5 9500 1900
35 10 6 3000 900
2 3 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.9 Distributions
1. Erlang
o j "' - ^ '-'expC-^!-)
/(>)=------------- ------- 22_ (a, as an integer)
r ( a , )
2. Gamma
o 2 - . ^ ' e x p ) - )
/ ( . I ' m ) = -----------------------------P J : -------------
n i )
3. InverseGaussian
v2
4. Lognormal
2 ^ - ^ m 3 2 a , 2_ym
/ ( * , ) = - - exp(J ln^ - ^ 2 )
y mS 2 n - c r 2 a
where a = lln( + l ) and / / =
12 V " ? ~ 2 + 1
5. PearsonVI
^g| - I
/(.Pm) = j--------------- ----------------------------
a3 . J, 1-1 .(1_,)2 "I d m + rSL)a] +a2
0 3
6. Rayleigh
/ C P * ) = ^ - e x p ( - ^ )
car, 2 a , 2
232
(8.9. 1)
(8.9.2)
(8.9.3)
(8.9.4)
(8.9.5)
(8.9.6)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7. Weibull
/ O m ) = <*l2_ a , >'mai_1 e x p t f - - ^ - ) 0'! ) (8 9 7)
233
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.10 Recurrence Data
EventID SCEN
DESC
Name Magnitude Default Frequency
24 SF2 South Hayward 5.0 0.06862
24 SF2 South Hayward 5.5 0.01220
24 SF2 South Hayward 6.0 0.00217
24 SF2 South Hayward 6.5 0.00039
24 SF2 South Hayward 6.9 0.00599
24 SF2 South Hayward 7.0 0.00599
24 SF2 South Hayward 7.5 0.00599
25 SF3 North Hayward 5.0 0.06862
25 SF3 North Hayward 5.5 0.01220
25 SF3 North Hayward 6.0 0.00217
25 SF3 North Hayward 6.5 0.00039
25 SF3 North Hayward 6.9 0.00599
25 SF3 North Hayward 7.5 0.00599
26 SF4 Rodgers Creek 6.0 0.00265
26 SF4 Rodgers Creek 6.5 0.00047
26 SF4 Rodgers Creek 7.0 0.00450
28 SF6 Calaveras 6.0 0.00190
28 SF6 Calaveras 6.5 0.00034
28 SF6 Calaveras 6.8 0.00685
28 SF6 Calaveras 7.0 0.00685
28 SF6 Calaveras 7.5 0.00685
29 SF7 Concord-Green Valley 6.0 0.00222
29 SF7 Concord-Green Valley 6.5 0.00039
29 SF7 Concord-Green Valley 6.9 0.00568
37 SF15 Greenville 6.0 0.00075
37 SF15 Greenville 6.5 0.00013
37 SF15 Greenville 6.9 0.00192
117 SA9 San Andreas (Peninsula) 6.0 0.00979
117 SA9 San Andreas (Peninsula) 6.5 0.00174
117 SA9 San Andreas (Peninsula) 7.0 0.00250
117 SA9 San Andreas (Peninsula) 7.1 0.00250
117 SA9 San Andreas (Peninsula) 7.5 0.00250
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.11 Soils Data
Census
Tract
Default
Soils
Class
Default
Landslide
Potential
Default
Liquefaction
Potential
New
Soils
Class
New
Landslide
Potential
New
Liquefaction
Potential
6001409000 4 0 0 5 2 5
6001428300 4 0 0 5 2 5
6001402000 4 0 0 4
2
5
6001401800 4 0 0 4 2 2
6001401700 4 0 0 5
2
5
6001401900 4 0 0 5
2
5
6001402100 4 0 0 4 2 2
6001402200 4 0 0 4
2
2
6001402300 4 0 0 4 2 2
6001402500 4 0 0 4 2 2
6001401500 4 0 0 4 2 2
6001400700 4 0 0 4 2 2
6001400900 4 0 0 4 2 2
6001401000 4 0 0 4
2
2
6001403500 4 0 0 4 2 2
6001402800 4 0 0 4 2 2
6001401400 4 0 0 4
2 2
6001401300 4 0 0 4
2 2
6001401600 4 0 0 4 2 2
6001402400 4 0 0 4 2 2
6001402700 4 0 0 4
2
2
6001403600 4 0 0 4 4 2
6001403700 4 0 0 4 2 2
6001404000 4 0 0 4 4 2
6001403300 4 0 0 4 2 5
6001402600 4 0 0 4 2 2
6001402900 4 0 0 4 2 2
6001403000 4 0 0 4 2 2
6001403100 4 0 0 4
2
2
6001403200 4 0 0 4 2 3
6001403400 4 0 0 5
2
2
6001405300 4 0 0 5 2 2
6001405900 4 0 0 4 2 2
6001405400 4 0 0 4 2 2
6001405500 4 0 0 4
2
2
6001405600 4 0 0 4
2
2
6001405800 4 0 0 4 4 2
235
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6001406000 4 0 0 4
2
5
6001406100 4 0 0 4 2 I
6001400200 4 0 0 4
2
2
6001400300 4 0 0 4 2 2
6001400400 4 0 0 4 2 2
6001400500 4 0 0 4 2 2
6001400600 4 0 0 4 2 2
6001401100 4 0 0 4 2 2
6001401200 4 0 0 4 2 2
6001404100 4 0 0 4 2 2
6001404200 4 0 0 2 6 2
6001404300 4 0 0 2 8 1
6001426200 4 0 0 2 4 2
6001405100 4 0 0 4 6 2
6001403800 4 0 0 4 6 2
6001403900 4 0 0 4 6 2
6001405000 4 0 0 2 4 2
6001405200 4 0 0 4 6 2
6001404600 4 0 0 2 8 1
6001404400 4 0 0 2 8 1
6001404501 4 0 0 2 8 1
6001404502 4 0 0 2 8 1
6001404700 4 0 0 2 8 1
6001404800 4 0 0 2 4 2
6001404900 4 0 0 4 4 2
6001406700 4 0 0 2 4 2
6001406800 4 0 0 2 4 2
6001406900 4 0 0 2 4 2
6001407900 4 0 0 2 6 2
6001408000 4 0 0 2
6 1
6001426100 4 0 0 2 8 1
6001407300 4 0 0 5 2 5
6001407400 4 0 0 4 2 2
6001407500 4 0 0 4 2 2
6001408800 4 0 0 4 2 2
6001409100 4 0 0 4 2 2
6001409200 4 0 0 4 2 2
6001409500 4 0 0 4 2 2
6001407600 4 0 0 4 4 2
6001407000 4 0 0 4 2 2
6001405700 4 0 0 4 6 2
6001406200 4
. 0
0 4 2 2
236
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6001406300 4 0 0 4 2
2
6001406400 4 0 0 4 6 2
6001406500 4 0 0 4 2 2
6001406600 4 0 0 4 2 2
6001407100 4 0 0 4 4 2
6001407200 4 0 0 4 2 2
6001408100 4 0 0 2 6 1
6001408900 4 0 0 4 2 2
6001409300 4 0 0 4 2 2
6001409400 4 0 0 4 2 2
6001409600 4 0 0 4 2 2
6001409700 4 0 0 4 2 2
6001410300 4 0 0 4 2
2
6001409800 4 0 0
2
6 2
6001407700 4 0 0 4 4 2
6001407800 4 0 0 2 6 2
6001408200 4 0 0 4 8 2
6001408300 4 0 0 4 6 2
6001408400 4 0 0 4 2
2
6001408500 4 0 0 4 2 2
6001408600 4 0 0 4 2 2
6001408700 4 0 0 4 2
2
6001410200 4 0 0 2 2
2
6001410400 4 0 0 4 2
2
6001409900 4 0 0 2 6 1
6001410000 4 0 0
2
6 I
6001410100 4 0 0
2
6 2
6001400800 4 0 0 4 2
2
6001400100 4 0 0 2 8 1
237
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.12 Exposure Data
Census
Tract
Default
Square Footage
(1000s of Square Feet)
Updated
Square Footage
(1000s of Square Feet)
6001409000 1419.0 829.0
6001428300 6498.0 1823.7
6001402000 0.0 5.2
6001401800 372.0 193.0
6001401700 589.5 203.0
6001401900 276.0 44.9
6001402100 78.0 51.8
6001402200 367.5 179.9
6001402300 106.5 81.9
6001402500 193.5 43.2
6001401500 538.5 314.6
6001400700 1402.5 941.6
6001400900 862.5 413.0
6001401000 1768.5 950.9
6001403500 477.0 348.0
6001402800 52.5 11.4
6001401400 784.5 376.8
6001401300 192.0 58.1
6001401600 433.5 180.5
6001402400 328.5 185.5
6001402700 322.5 119.9
6001403600 327.0 277.3
6001403700 40.5 111.9
6001404000 357.0 273.4
6001403300 210.0 62.6
6001402600 97.5 25.7
6001402900 0.0 0.0
6001403000 52.5 5.6
6001403100 30.0 3.2
6001403200 43.5 2.5
6001403400 0.0 0.0
6001405300 208.5 160.9
6001405900 1410.0 674.0
6001405400 634.5 395.8
6001405500 787.5 419.4
6001405600 592.5 458.3
238
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6001405800 1122.0 677.3
6001406000 235.5 120.5
6001406100 706.5 280.5
6001400200 883.5 938.9
6001400300 1959.0 1543.6
6001400400 1342.5 1030.0
6001400500 1107.0 757.7
6001400600 591.0 411.8
6001401100 699.0 414.2
6001401200 880.5 494.1
6001404100 1023.0 768.5
6001404200 1713.0 2033.3
6001404300 1767.0 2222.4
6001426200 2401.5 3160.8
6001405100 2343.0 3355.3
6001403800 849.0 1064.1
6001403900 457.5 509.0
6001405000 1401.0 1486.7
6001405200 936.0 1064.2
6001404600 2487.0 3455.2
6001404400 2466.0 5478.3
6001404501 940.5 1454.1
6001404502 3159.0 3908.9
6001404700 1107.0 1484.4
6001404800 1095.0 873.7
6001404900 1747.5 1518.7
6001406700 2550.0 2106.7
6001406800 1725.0 1456.9
6001406900 1468.5 1282.4
6001407900 1507.5 1246.4
6001408000 1189.5 1381.0
6001426100 3232.5 6093.0
6001407300 579.0 295.6
6001407400 904.5 499.8
6001407500 793.5 526.2
6001408800 1224.0 649.4
6001409100 883.5 552.8
6001409200 888.0 645.6
6001409500 991.5 443.8
6001407600 1989.0 1390.5
6001407000 1764.0 946.7
6001405700 732.0 492.4
239
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6001406200 1132.5 697.0
6001406300 912.0 626.3
6001406400 630.0 541.8
6001406500 1267.5 785.1
6001406600 1749.0 1158.2
6001407100 2040.0 1407.4
6001407200 1125.0 579.1
6001408100 2614.5 3661.4
6001408900 786.0 426.2
6001409300 1480.5 956.4
6001409400 1231.5 622.2
6001409600 1410.0 849.5
6001409700 1291.5 724.7
6001410300 870.0 538.6
6001409800 1341.0 1358.4
6001407700 2449.5 1923.3
6001407800 921.0 799.3
6001408200 1423.5 1132.7
6001408300 2233.5 1761.1
6001408400 1293.0 778.6
6001408500 1710.0 1110.8
6001408600 1935.0 1305.0
6001408700 2272.5 1575.3
6001410200 1135.5 687.1
6001410400 1347.0 964.3
6001409900 2010.0 2329.2
6001410000 1617.0 2025.4
6001410100 1110.0 818.3
6001400800 997.5 499.4
6001400100 1752.0 7515.5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.13 Summary of Average Annual Loss
Abbreviation for cases is as follows:
For the cases in column (1) BEFORE MITIGATION, AAL are in (2):
ATTENUATION SOILS EXPOSURE FRAGILITY
DEFAULT 12 D D DUN
UPDATED 10 N N NUN
For the cases in column (4) AFTER MITIGATION, AAL are in (5):
ATTENUATION SOILS EXPOSURE FRAGILITY
DEFAULT 12 D D DM
UPDATED 10 N N NM
The cases above are combined with the updated recurrence in columns (3) and (6).
0 )
Case
(2)
AAL
(SThousands)
(3) AAL
(SThousands)
Updated
Recurrence
(4)
Case
(5)
AAL
(SThousands)
(6) AAL
(SThousands)
Updated
Recurrence
10DDDUN $219,164 $179,615 I0DDDM $158,108 $130,383
I2DDDUN $182,835 $149,621 12DDDM $127,811 $105,649
10NDDUN $205,628 $169,111 I0NDDM $156,428 $129,637
12NDDUN $171,270 $139,780 12NDDM S125,128 $103,025
10DNDUN $199,801 $163,675 10DNDM $144,586 $119,067
12DNDUN $166,599 $136,410 12DNDM $116,628 $96,430
10DDNUN $204,010 $167,357 10DDNM $154,009 $126,885
12DDNUN $169,769 $138,935 12DDNM $124,657 $102,911
10NNDUN $166,219 $138,659 10NNDM $128,537 $108,433
12NNDUN $135,717 $111,121 12NNDM $99,554 $82,348
10NDNUN $192,843 $158,687 10NDNM $152,907 $126,613
12NDNUN $159,901 $130,443 12NDNM $122,374 $100,582
10DNNUN $185,908 $152,479 10DNNM $140,880 $115,898
12DNNUN $154,691 $126,692 12DNNM $113,767 $93,945
10NNNUN $156,186 $130,421 I0NNNM $125,776 $106,052
12NNNUN $126,768 $103,726 12NNNM $97,410 $80,430
241
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
9 Bibliography
Ambraseys, N., 1974. The Correlation of Intensity with Ground Motion, Proceedings o f the
l 4 h Conference o f the European Seismology Committee, Trieste, Italy.
Anagnos, T., Rojahn, C., and A. S. Kiremidjian, 1995. NCEER-ATC Joint Study on Fragility o f
Buildings, NCEER-95-0003, Applied Technology Council, Redwood City, California.
ATC-13, 1985. Earthquake Damage Evaluation Data for California, Applied Technology
Council, Redwood City, CA.
Bak, P. and Tang, C., 1989. Earthquakes as Self-Organized Critical Phenomena, Journal o f
Geophysics Research, Vol. 94, pp. 15635-16337.
Benjamin, J.R. and C. A. Cornell. Probability, Statistics, and Decision fo r Civil Engineers.
McGraw-Hill: New York, New York.
Bemreuter, D.L., Savy, J.B., Mensing, R.W., and J.C. Chen, 1989. Seismic Hazard
Characterization o f 69 Nuclear Plant Sites East o f the Rocky Mountains, NUREG/CR-5250,
Vols. 1-8, Nuclear Regulatory Commission, Washington, D.C.
Bonilla, M.G., Mark, R.K., and J.J. Lienkaemper, 1984. Statistical Relations Among Earthquake
Magnitude, Surface Rupture Length, and Surface Fault Displacement. Bulletin o f the
Seismological Society o f America, Vol. 74 (6), pp. 2379-2411.
Boore, D.M., Joyner, W.B., and Fumal, T.E., 1993. Estimation o f Response Spectra and Peak
Acceleration fo r Western North American Earthquakes: An Interim Report, USGS Open-File
Report 93-509, United States Geological Survey, Menlo Park, California.
Boore, D.M., Joyner, W.B. and T. E. Fumal, 1994. Estimation o f Response Spectra and Peak
Acceleration from Western North American Earthquakes: An Interim Report, Part 2, USGS
Open-File Report 94-127, United States Geological Survey, Menlo Park, California.
Budnitz, R.J., Apostolakis, G., Boore, D.M., Cluff, L.S., Coppersmith, K.J., Cornell, C.A., and
P.A. Morris, 1997. Recommendations fo r Probabilistic Seismic Hazard Analysis: Guidance on
Uncertainty and Use o f Experts, NUREG/CR-6372, U.S. Nuclear Regulatory Commission,
Washington, D.C.
Campbell, K.W. and Bozorgnia, Y., 1994. Near-Source Attenuation of Peak Horizontal
Acceleration from Worldwide Accelerograms Recorded from 1957 to 1993, Proceedings o f the
Fifth U.S. National Conference on Earthquake Engineering, Vol. Ill, pp. 283-292.
Cao, T., Peterson, M.D. and M. S. Reichle, 1996. Seismic Hazard Estimate for Background
Seismicity in Southern California, Bulletin o f the Seismological Society o f America, Vol. 86(5),
pp. 1372-1381.
242
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Chhibber, S. and G. Apostolakis, 1993. Some Approximations Useful to the Use of Dependent
Information Source, Reliability Engineering and System Safety, Vol. 42, pp. 67-86.
Clemen, R.T. and R. L. Winkler, 1992. Sensitivity of Weights in Combining Forecasts,
Operations Research, Vol. 40(3), pp. 609-614.
Clemen, R.T., 1989. Combining forecasts: A review and annotated bibliography, International
Journal o f Forecasting, Vol. 5, pp. 559-583.
Cooke, R.M., 1991. Experts in Uncertainty: Expert Opinion and Subjective Probability in
Science, Oxford University Press: New York, New York.
Coppersmith, K..J. and R. R. Youngs, 1986. Capturing Uncertainty in Probabilistic Seismic
Hazard Assessments within Interplate Tectonic Environments, Proceedings o f the Third U.S.
National Conference on Earthquake Engineering, Earthquake Engineering Research Institute,
Vol. l,pp. 301-312.
Cornell, C. Allin, 1968. Engineering Seismic Risk Analysis, Bulletin o f the Seismological
Society o f America, Vol. 58 (5), pp. 1583-1606.
Cornell, C. Allin, and E.H. Van Marke, 1969. The Mayor Influences on Seismic Risk,
Proceedings o f the Third World Conference on Earthquake Engineering, Vol. A-l, pp. 69-93.
Cramer, C.H., Peterson, M.D. and M. S. Reichle, 1996. A Monte Carlo Approach in Estimating
Uncertainty for a Seismic Hazard Assessment of Los Angeles, Ventura, and Orange Counties,
California, Bulletin o f the Seismological Society o f America, Vol. 86(6), pp. 1681-1691.
CSSC, 1994. The Homeowners Guide to Earthquake Safety, SSC No. 92-02, California Seismic
Safety Commission, Sacramento, CA.
Dalkey, N.C., 1969. The Delphi Method. Rand Corporation: Santa Monica, California.
Dieterich, J.H. et al., 1990. Probabilities o f Large Earthquakes in the San Francisco Bay Region,
California, USGS Circular 1053, United States Geological Survey, Menlo Park, California.
EERI, 1996. Scenario for a Magnitude 7.0 Earthquake on the Hayward Fault, EERI Report
Number HF-96: Earthquake Engineering Research Institute, Oakland, California.
EERI, 1997. Theme Issue: Loss Estimation, Earthquake Spectra, Vol. 13(4).
EERI, 2000. Financial Management o f Earthquake Risk, EERI White Paper, Earthquake
Engineering Research Institute, Oakland, California.
Ermoliev, Y.M., Ermolieva, T.Y., MacDonald, G. and V. I. Norkin, 1998. On the Design o f
Catastrophic Risk Portfolios, Interim Report IR-98-056, International Institute for Applied
Systems Analysis, Luxemburg, Austria.
243
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
FEMA, 1992. A Benefit-Cost Model fo r the Seismic Rehabilitation o f Buildings, FEMA 227:
Volume 1 - A Users Manual and FEMA 228:Volume 2 - Supporting Documentation, Federal
Emergency Management Agency, Washington, D.C.
FEMA, 1997. NEHRP Recommended Provisions for Seismic Regulations fo r New Buildings and
Other Structures, FEMA 302: Part 1Provisions and FEMA 303: Part 2 Commentary, Federal
Emergency Management Agency, Washington, D.C.
Frankel, A., Mueller, C., Bamhard, T., Perkins,D., Leyendecker, E.V., Dickman, N., Hanson, S.,
and M. Hopper, 1996. National Seismic Hazards Maps: Documentation, June 1996, USGS Open-
File Report 96-532: United States Geological Survey, Golden, Colorado.
R. P. Gallagher Associates, Inc., 1990. Mitigation o f Principal Earthquake Hazards to Wood
Frame Dwellings and Mobile Homes, Report for the State of California Department of Insurance,
San Francisco, California.
Genest, C. and J. V. Zidek, 1986. Combining Probability Distributions: A Critique and an
Annotated Bibliography, Statistical Science, Vol. 1(1), pp. 114-148.
Grossi, P., 1998. Assessing the Benefits and Costs o f Earthquake Mitigation, Working Paper 98-
11-11, Risk Management and Decisions Processes Center, The Wharton School, Philadelphia,
Pennsylvania.
Grossi, P., Kleindorfer, P., and H. Kunreuther, 1999. The Impact o f Uncertainty in Managing
Seismic Risk: The Case o f Earthquake Frequency and Structural Vulnerability, Working Paper
99-03-26, Risk Management and Decision Processes Center, The Wharton School, Philadelphia,
Pennsylvania.
Grossi, P. and D. Windeler, 2000. Sensitivity Analysis of Earthquake Risk in the Charleston,
South Carolina Region, Proceedings o f the Sixth International Conference on Seismic Zonation,
Earthquake Engineering Research Institute, (publication pending).
Gutenberg, B. and C.F. Richter, 1956. Earthquake Magnitude, Intensity, Energy, and
Acceleration, Bulletin o f the Seismological Society o f America, Vol. 46, pp. 105-145.
Hanks, T.C. and H. Kanamori, 1979. A Moment Magnitude Scale. Journal o f Geophysical
Research, Vol. 84, pp. 2348-2350.
Hanks, T.C. and C. A. Cornell, 1994. Probabilistic Seismic Hazard Analysis: A Beginners
Guide. Proceedings o f the Fifth Symposium on Current Issues Related to Nuclear Power Plant
Structures, Equipment and Piping, 1/1-1 to 1/1-17, North Carolina State University, Raleigh,
North Carolina.
Hora, S.C. and R. L. I man, 1989. Expert Opinion in Risk Analysis: The NUREG-1150
Methodology, Nuclear Science and Engineering, Vol. 102, pp. 323-331.
244
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Insurance Information Institute, 2000.Catastrophes [online]. The Insurance Information Institute
[cited 10 June 2000]. Available from World Wide Web: www.iii.org/media/issues/catastrophcs.
Joyner, W.B. and D. M. Boore, 1981. Peak Horizontal Acceleration and Velocity from Strong
Motion Records Including Records from the 1979 Imperial Valley, California, Earthquake,
Bulletin o f the Seismological Society o f America, Vol. 71, pp. 2011-2038.
Joyner, W.B. and D. M. Boore, 1988. Measurement, Characterization, and Prediction of Strong
Ground Motion, Proceedings o f Earthquake Engineering and Soil Dynamics II, New York,
Geotechnical Division of the American Society of Civil Engineers, pp. 43-102.
Kanamori, H. Magnitude Scale and Quantification of Earthquakes, Tectonophysics, Vol. 93, pp.
185-199.
Larsen, R.J. and M.L. Marx, 1986. An Introduction to Mathematical Statistics and its
Applications. Prentice-Hall Company: Englewood Cliffs, New Jersey.
Lienkaemper, J.J. and Galehouse, J.S., 1998. New Evidence Doubles the Seismic Potential of
the Hayward Fault, Seismological Research Letters, Vol. 69(6), pp. 519-523.
Mahaney, J.A., Paret, T.E., Kehoe, B.E., Freeman, S.A., 1993. The Capacity Spectrum Method
for Evaluating Structural Response during the Loma Prieta Earthquake, Proceedings o f the 1993
United States National Earthquake Conference, Vol. 2, pp. 501-510.
McGuire, R.K. and K. M. Shedlock, 1981. Statistical Uncertainties in Seismic Hazard
Evaluations in the United States, Bulletin o f the Seismological Society o f America, Vol. 71(4),
pp. 1287-1308.
McGuire, R.K., Toro, G.R., Jacobson, J.P., O'Hara, T.F., and W.J. Silva, 1989. Probabilistic
Seismic Hazard Evaluations at Nuclear Power Plant Sites in the Central and Eastern United
States: Resolution o f the Charleston Earthquake Issue, EPRI Report NP-6395-D, Electric Power
Research Institute, Palo Alto, California.
Means, 1999. Means Square Foot Costs. R.S. Means Company, Inc.: Kingston, Massachusetts.
Newmark, N.M., and Hall, W.J., 1982. Earthquake Spectra and Design, Earthquake Engineering
Research Institute: Oakland, California.
NIBS, 1997. HAZUS: Hazards U.S.: Earthquake Loss Estimation Methodology. NIBS Document
Number 5200, National Institute of Building Sciences, Washington, D.C.
Peterson, M. D., Cramer, C. H., Bryant, W. A., Reichle, M. S. and T.R. Toppozada, 1996a.
Preliminary Seismic Hazard Assessment for Los Angeles, Ventura, and Orange Counties,
California, Affected by the 17 January 1994 Northridge Earthquake, Bulletin o f the
Seismological Society o f America, Vol. 86(1B), pp. S247-S261.
245
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Peterson, M.D., Bryant, W.A., Cramer, C.H., Cao, T., Reichle, M.S., Frankel, A.D.,
Lienkaemper, J.L., McCrory, P.A., and D.P. Schwartz, 1996b. Probabilistic Seismic Hazard
Assessment fo r the State o f California, USGS Open-File Report 96-706: United States Geological
Survey, Menlo Park, California.
Reiter, Leon, 1990. Earthquake Hazard Analysis: Issues and Insights. Columbia University
Press: New York, New York.
Richter, C.F., 1958. Elementary Seismology. W.H. Freeman and Company: San Francisco,
California.
Sadigh, K. et a!., 1993. Specification of Long-Period Ground Motions: Updated Attenuation
Relationships for Rock Site Conditions and Adjustment Factors for Near-Fault Effects,
Proceedings o f ATC 17-1 Seminar on Seismic Isolation, Passive Energy Dissipation, and Active
Control, Applied Technology Council, pp. 59-70.
Schwartz, D. et al., 1999. Earthquake Probabilities in the San Francisco Bay Region: 2000 to
2030 - A Summary o f Findings, USGS Open-File Report 99-517, Prepared by the Working
Group on California Earthquake Probabilities for the United States Geological Survey, Menlo
Park, California.
Seible, F., Filiatrauit, A. and C.-M. Uang, 1999. Proceedings o f the Invitational Workshop on
Seismic Testing, Analysis and Design o f Woodframe Construction, CUREe Publication No. W-
01: Division of Structural Engineering, University of California, San Diego.
Shepherd, R. and E.O. Delos-Santos, 1991. An Experimental Investigation of Retrofitted Cripple
Walls, Bulletin o f the Seismological Society o f America, Vol. 81(5), pp. 2111-2126.
Steinbrugge, K. V., 1982. Earthquakes, Volcanoes, and Tstmamis: An Anatomy o f Hazards.
Skandia America Group: New York, New York.
Stover, C.W. and J.L. Coffman, 1993. Seismicity o f the United States, 1568-1989. U.S.
Geological Survey Professional Paper 1527, United States Government Printing Office,
Washington, D.C.
Trifunac, M.D. and A.G. Brady, 1975. On the Correlation of Seismic Intensity with Peaks of
Recorded Strong Ground Motion, Bulletin o f the Seismological Society o f America, Vol. 65 (1),
pp. 139-162.
Weichert, D.H., 1980. Estimation of the Earthquake Recurrence Parameters for Unequal
Observation Periods for Different Magnitudes. Bulletin o f the Seismological Society o f America,
Vol. 70, pp. 1337-1346.
Wells, D.L., and K.J. Coppersmith, 1994. New Empirical Relationships among Magnitude,
Rupture Length, Rupture Width, Rupture Area, and Surface Displacement. Bulletin o f the
Seismological Society o f America, Vol. 84 (4), pp. 974-1002.
246
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Wesnousky, S.G., 1986. Earthquakes, Quaternary faults, and Seismic Hazard in California,
Journal o f Geophysical Research, Vol. 91(B12), pp. 12,587-12,631.
Winkler, R.L, 1968. The Consensus of Subjective Probability Distributions, Management
Science, Vol. 15(2), pp. B61-B75.
Youngs, R.R., and K.J. Coppersmith, 1985. Implications of Fault Slip Rates and Earthquake
Recurrence Models to Probabilistic Seismic Hazard Estimates. Bulletin o f the Seismological
Society o f America, Vol. 75 (4), pp. 939-964.
247
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

You might also like