Content uploaded by Mohsen Rahnama
Author content
All content in this area was uploaded by Mohsen Rahnama on Jun 03, 2015
Content may be subject to copyright.
13th World Conference on Earthquake Engineering
Vancouver, B.C., Canada
August 1-6, 2004
Paper No. 1072
EARTHQUAKE RISK ESTIMATES FOR RESIDENTIAL
CONSTRUCTION IN THE WESTERN UNITED STATES
Don WINDELER1, Guy MORROW2, Chesley R. WILLIAMS3, Mohsen RAHNAMA4, Gilbert
MOLAS5, Adolfo PEÑA6, and Jason BRYNGELSON7
SUMMARY
This study presents relative seismic risk estimates for thirteen states: eleven coterminous west of the
Rockies, Alaska and Hawaii. We focus on residential construction, considering both economic and
insured exposure.
The loss estimation system uses a seismic source model based on the USGS 2002 National Seismic
Hazard Mapping project. Building damage is estimated via spectral response-based vulnerability
functions. This model incorporates variations in site conditions, construction design levels, building
inventory, and insured value.
The thirteen western states are ranked in terms of their relative earthquake risk, with risk per state
compared on the basis of both economic and insured loss cost. The consideration of insurance has a
distinct effect on relativities due to differences in penetration rates and prevailing policy structures. This
is particularly true for California; currently, the high deductibles and prices charged have driven down
the purchase of earthquake policies to an extent that the insured proportion of potential earthquake losses
is significantly less than it was at the time of the Northridge earthquake.
INTRODUCTION
Catastrophe modeling brings together a range of technical disciplines to estimate future losses from
natural disasters, rather than relying only on a potentially incomplete historic record. For the earthquake
peril these include seismology, civil and geotechnical engineering, economics, and actuarial science. Loss
1 Chief Geologist, Earthquake Hazards Practice, RMS, Inc, Newark, California, USA:
Don.Windeler@RMS.com
2 V.P. of Model Development, RMS, Inc, Newark, California, USA: Guy.Morrow@RMS.com
3 Lead Engineer/Geologist, RMS, Inc, Newark, California, USA: Chesley.Williams@RMS.com
4 V.P. Engineering & Model Dev., RMS, Inc, Newark, California, USA: Mohsen.Rahnama@RMS.com
5 Lead Engineer, RMS, Inc, Newark, California, USA: Gilbert.Molas@RMS.com
6 Technical Marketing Manager, RMS, Inc, Newark, California, USA: Adolfo.Pena@RMS.com
7 Engineer, RMS Inc., Newark, California, USA: Jason.Bryngelson@RMS.com
estimation tools influence public policy, mitigation decisions, local planning, insurance and reinsurance
purchasing and pricing.
In this study, we present relative risk estimates for the thirteen states west of the Rocky Mountains. This
analysis is noteworthy for incorporation of source modeling from the USGS 2002 National Seismic
Hazard Maps [1], a new spectral response-based approach to building vulnerability [2, 3], and NEHRP-
classified site condition data for all thirteen states. Risk estimates are presented in terms of average
annual loss cost for economic and insured exposure. We focus on residential construction in this study,
i.e. homeowners and renters only.
MODEL DESCRIPTION
Results provided in this analysis were generated from RiskLink, a proprietary insurance loss-estimation
tool. It applies an event-based approach in which a set of stochastic events with corresponding rates has
been defined, portfolio loss and variability are generated for each event, and exceedance probabilities of
portfolio losses are calculated for various economic or insurance perspectives. The inputs to the RiskLink
model are based primarily on publicly-available data and are summarized below.
Exposure and analysis resolution
Three exposure data sets were analyzed for this study. The first two comprise residential value for the
western U.S. by ZIP Code, one total economic value and one insured value. The insured portfolio
incorporates the local penetration rate of earthquake insurance purchase, policy conditions on deductibles
and limits, and coverages for structures, contents, and additional living expenses. These were estimated
from a value of public and private data sources, including insurance companies, state insurance regulators,
the California Earthquake Authority, the U.S. Census, gross domestic product, Dun & Bradstreet square
footage, Means construction costs, and other statistical factors.
The third was used to make a relative risk map and comprised a portfolio spaced on a variable grid for the
thirteen western states considered. Grid cells were of approximately 1-, 5-, or 10-km size, with finer
resolution used in areas of high exposure and/or hazard. Each cell contained a uniform value, split 65%
structure / 35% contents, with the default inventory of residential building stock for the state.
Seismic source model
The fundamental inputs for the seismic source model are the documentation and parameters developed for
the 2002 USGS National Seismic Hazard Maps. These are described for the lower 48 states, Hawaii, and
Alaska by Frankel [1], Klein [4], and Wesson [5] respectively. For the purpose of this study, the most
significant source modeling differences are the inclusion of “cascade” scenarios and/or time dependent
recurrence on selected faults.
Cascade events refer to the potential for an earthquake to “jump” between fault segments during the
rupture process. Recent examples include the 1992 Landers and 1999 Hector Mine events in the Mojave
Desert, each of which ruptured smaller faults that had previously been considered separate sources.
Incorporation of these events in the stochastic set generates earthquakes that are larger than would be
possible on any of the constituent faults. There is a balancing effect on the model, however, as allowing
the occurrence of cascades reduces the seismic moment available for smaller events and overall rates for
the fault system will decrease. The model used in this study includes the cascade scenarios detailed in the
2002 USGS maps, including those defined for the Bay Area (WGCEP [6, 7]), as well as events on seven
additional fault systems in California. Rate calculations for these follow the moment-balancing approach
of Field [8] with values weighting three different probabilities of multi-segment rupture for each system.
The USGS hazard maps assume a Poisson (time-independent) process for estimating event probability
within a given time window. A different approach to representing probability of event occurrence is the
time-dependent model. Time-dependence explicitly recognizes the time interval since the last occurrence
of an event associated with a given source and incorporates that information in estimating the probability
of a future event on that same source. As the time since the last event increases, the probability of the
event occurring in the near future will generally increase depending on the distribution used (cf Matthews
[9]). Time-dependence is used for major fault systems in the state, including the San Andreas, Hayward-
Rodgers Creek, San Jacinto, and Whittier-Elsinore fault. Key references include WGCEP [6, 7].
Ground motion attenuations vary by source type. Abrahamson [10], Boore [11], Campbell [12], and
Sadigh [13] are considered for thrust and strike-slip crustal faults, with Spudich [14] included for
extensional events. Subduction interface events combine Youngs [15], Sadigh [13], and Atkinson [16],
while intraslab ground motions are modeled with the appropriate formulations of Youngs [15] and
Atkinson [16]. Ground motion is calculated in terms of spectral acceleration from periods 0 to 4 seconds;
the value experienced for a given location is a function of the building’s predominant period.
Geotechnical site conditions
Digital geologic maps were assigned NEHRP site classes on the basis of published or inferred 30-m shear
wave velocity. The approach follows the scheme and data of Wills [17]. Over 200 map coverages were
incorporated into the site conditions dataset for the thirteen states in the study area. Resolution varies with
data availability and exposure density. Small-scale geologic maps at 1:500,000 or 1:750,000 resolution
were used for regional coverage. California was an exception, with 1:250,000 scale data used as the
lowest resolution input. For primary urban areas, the input maps used were typically 1:100,000 scale or
better.
For the purpose of analysis these data were aggregated to the ZIP Code and grid resolutions described
above. Grid resolutions for the soil data are limited by the scale of the input map, so as not to exceed the
applicability intended by the map authors. In both cases, aggregate values have been exposure-weighted
with land use / land cover attribute data.
Liquefaction and landslide susceptibilities are also incorporated into the site inputs. Liquefaction
susceptibilities were either aggregated from published maps (e.g. Knudsen [18]) or estimated from
geology using the schema of Youd [19]. Landslide susceptibilities were developed following results of
seismic hazard zoning studies by the California Geological Survey (e.g. [20]). These studies used a
Newmark approach to define a matrix relating material properties and slope to susceptibility; these
matrices were used directly where available and generalized for geologic materials elsewhere in
California, Oregon, Washington, and Utah.
Vulnerability
The vulnerability module generates an estimate of the damage to exposure at a specific location as a
function of the ground shaking for an event. The damage is expressed in terms of a mean damage ratio and
a coefficient of variation around the mean. The RiskLink model uses separate vulnerability functions for
building, contents, and time element losses.
Development of the damage functions followed the framework recently developed at the Pacific
Earthquake Engineering Research Center (PEER), first described by Cornell [2]. The PEER approach
considers both the entire spectrum of earthquake ground motion characteristics at a site and a building’s
response to that motion, and is thus referred to as spectral response-based vulnerability. See also
Rahnama [3] for additional discussion of this implementation.
Contents modeling considers both damage from the intensity of ground shaking and from distress to the
building itself. The former is more important at low shaking levels and is relatively independent of
structure type, whereas at higher shaking the structural damage will contribute to the contents loss.
Users without data on the construction class for locations in their portfolio rely on inventory curves that
store relative proportions of building types for different lines of business. For this study, the inventory
data vary by state.
RESULTS
Model results are presented in terms of annualized loss cost, which is the modeled average annual loss
normalized by the structure replacement value. Loss costs are given in units of $/$1,000 (per mille), a
metric commonly used to quote insurance coverage premium, or have been normalized to the value for all
of California.
Average annual loss (AAL) is the expected value of an exceedance probability loss distribution. It can be
thought of as the product of the loss for a given event with its probability, summed over all events in the
stochastic set. Normalizing the AAL by exposure facilitates comparison of relative risk, as it reduces
potential differences in how the total building stock value is modeled in other studies. Population
estimates from the US Census Bureau [21] are used as a proxy here for ordering of states by residential
value.
Regional Loss Metrics
Figure 1 shows normalized residential loss costs for the thirteen western States considered in this study.
This is similar to a seismic hazard map, but incorporates differences in construction inventory and
damageability. It spans four orders of magnitude, providing a synoptic view of the relative seismic risk
and context for the state and county level loss costs.
The highest point values are along the creeping section of the San Andreas fault system in central
California. Average annual loss is strongly affected by event rates and the high relative rates of moderate
earthquakes along this segment of the fault drive up the loss costs. Other notable areas include the
southeast side of the island of Hawaii, with risk due to the active volcanic flank, and the sharp NW-SE
discontinuity in eastern California. The latter case highlights the boundary between the Central Valley
sediments to the west and Sierra Nevadan batholith to the east. Most of the risk in the Central Valley is
from distant San Andreas events, the effects of which are filtered out by the predominantly hard igneous
rocks of the Sierra Nevada.
Statewide loss costs are summarized in Figure 2, with the values normalized to the loss cost for California.
These incorporate the relative distribution of population and residential construction within each state.
California has by far the highest risk, at more than three times the relative risk on a statewide basis than
Washington, the second highest. When the relative exposure is considered, California has about twenty
times the annualized economic loss from earthquake as Washington and ten times that of the rest of the
western states combined.
Figure 1. Average annual loss costs for residential construction in the western United States.
Figure 2. Statewide risk relative to California and volatility estimates.
The impact of exposure location relative to the hazard is evident. The states of Idaho, Montana, Utah, and
Wyoming share the Intermountain Seismic Belt, the roughly N-S, arcuate band of hazard illustrated in the
center of Figure 1. While the relative risk within this belt is similar, the only major city to fall in is Salt
Lake City in Utah. Consequently, the statewide loss cost is 5-10 times greater in Utah than the other three
states.
Figure 2 also includes a measure of the volatility of the annual loss. This is the coefficient of variation on
the loss cost, reflecting the range in possible losses for any given year rather than uncertainty in the actual
loss estimate. A comparison between Washington and Utah provides an illustrative example. The lost
costs are similar between the two, but the volatility in Utah is much higher because the main contributors
are rare but large losses from the Wasatch fault system. Washington, on the other hand, has a large
contribution from the more frequent intraslab earthquakes (e.g. 1949, 1965, 2001), which historically have
caused moderate damage. The high rates of occurrence on the island of Hawaii result in a relatively low
loss volatility for the state, but the overall loss cost is moderated because over 70% of the state’s populace
live on the seismically quiet island of Oahu.
Impact of Insurance
Insurance is a mechanism for risk transfer in which the property owner pays a premium to another party
for protection against some loss-causing event. The insured loss relative to economic is most significantly
affected by penetration rate and predominant policy structures. Penetration rate refers to the proportion of
property owners who choose to purchase insurance. The insurance policy is a contract defining the
conditions for payout to the insured. It typically includes deductibles (proportion of loss the insured must
absorb before the policy pays a claim) and limits (maximum amounts payable by the insurer). These may
be defined for individual coverage types, the policy as a whole, or both.
The 1994 Northridge earthquake and 1992 Hurricane Andrew events provide contrasting examples of
insured : economic loss relativities. The Northridge earthquake caused an estimated $42 bn in economic
loss, $15 of which was insured [22, 23]. In contrast, $16 bn of Hurricane Andrew’s estimated $30 bn
economic loss was insured ([24, 25] adjusted to 2001$ using [26]). The relative proportion was higher for
Andrew for reasons of both penetration and policies. Homeowners are usually required by lenders to have
fire insurance, and wind was covered as part of the standard policy at that time. Coverage for earthquake,
State Abbrev Population
(2002 est)
California CA 35,001,986
Washington WA 6,067,060
Utah UT 2,318,789
Alaska AK 641,482
Nevada NV 2,167,455
Hawaii HI 1,240,663
Oregon OR 3,520,355
Montana MT 910,372
New Mexico NM 1,852,044
Idaho ID 1,343,124
Wyoming W Y 498,830
Arizona AZ 5,441,125
Colorado CO 4,501,051
CA
WA
UT
AK
NV
HI
OR
MT
NM
ID
WY
AZ
CO
0.0
10.0
20.0
30.0
40.0
50.0
60.0
0.001 0.01 0.1 1
loss cost rel. to California
volatility
on the other hand, must be purchased separately and thus a smaller fraction of structures were insured for
Northridge. Earthquake deductibles are also much higher for earthquake, around 10% of the structure
value at the time, while a typical wind deductible in Florida was $500.
Figure 3 provides an estimate of the fraction of the total residential loss cost that would be covered by
insurance. (Note that this is the percentage of the average annual loss over all events, not the relativity
one should expect for a single event.) The ratios of insured to economic loss are calculated from absolute
average annual loss values and thus have a greater dependence on the exposure assumptions than the loss
costs; relativities between states are less uncertain than the actual percentages. The per-event correlation
in loss becomes more important once financial structures such as deductibles are considered. An
earthquake with severe localized loss might have the same economic loss as an event with lesser losses
spread out over a large area but, assuming constant deductibles, the latter event would generally have less
insured loss because more policy deductibles would have to be exceeded.
ID
NM
AZ
CO
WY MT
NV
AK
UT
HI
OR WA
CA
1%
10%
100%
0.001 0.01 0.1 1
loss cost, normalized to California
%of AAL covered by insurance
Figure 3. Modeled percentage of residential average annual loss per state covered by insurance.
Abbreviations as per Figure 2, small ‘x’ is California commercial.
In general, percentage increases with the loss cost as there is some expectation that the rate of earthquake
insurance takeup will increase in areas of higher risk. Deductibles tend to be higher as well, but the
relativities are more sensitive to assumptions in penetration. The outlier in this analysis is California,
where the current cost of homeowners’ insurance relative to coverage has greatly reduced the fraction of
earthquake losses borne by insurers. This is considered further in the Discussion below.
DISCUSSION
Factors influencing loss costs
Comparison of Figure 1 with USGS hazard maps of the area [1] supports an obvious observation: a
primary source of local variations in the relative loss costs is the source model modified by local site
conditions. At this scale, variations in the vulnerability data are largely overwhelmed by the hazard
changes. The vulnerability would show much greater differentiation if additional construction or
occupancy types were included for comparison, particularly for buildings of different heights. The
performance-based approach used for damage calculation considers the ground motion spectrum, period-
dependent site amplification, and structure period.
When considered in the aggregated context of a portfolio, the distribution of exposure relative to the
hazard becomes more crucial in determining the loss cost. Los Angeles County has both high hazard from
numerous active faults and absolute risk due to its large population, but its relative loss cost ranks lower
because the population is split into several different urban areas [27, 28].
Average annual loss or loss cost is a useful metric for comparing risk but is a collapsed version of the full
loss exceedance curve; in isolation it lacks detail on what kind of losses comprise the total. Environments
with very frequent small losses versus rare catastrophic losses could generate the same AAL, but the latter
presents a more problematic case for a homeowner or insurer.
The volatility measure shown in Figure 2 provides some idea of where a given state falls in the spectrum,
with the previous example of Washington and Utah illustrating this point. Both have similar loss costs,
with Utah showing a higher volatility. Consider three annual probability ranges, <0.2%, 0.2-1%, and
>1%, nominally equivalent to return periods of >500 years, 100-500 years, and <100 years. The average
annual loss for Washington splits approximately 25-35-40 across these ranges, reflecting the high
contribution by frequent intraslab events. Utah, in contrast, splits 50-35-15 due to rare but catastrophic
losses from Wasatch fault events.
A related point is that frequent events will contribute greatly to average annual loss, even if they do not
generate large losses. California is the extreme case for short return period losses, with over 80% derived
from the >1% probability portion of the loss curve.
Residential insurance in California
Neglecting epistemic changes from modeling, the relative loss costs discussed above are fairly constant
risk metrics. The local seismic hazard may be impacted by time-dependent recurrence after a large event,
but overall is driven by long-term tectonics and seismic moment budgets. Building vulnerability is
gradually affected by code changes and new construction, but is another factor that changes slowly. What
does change are insured loss estimates, as prevailing market conditions will drive changes to the insured
proportion of value at risk. The 1994 Northridge earthquake was a seminal event in its influence on the
US property insurance industry; much of the commentary in this section follows a recent report on the
event [23].
Following the devastating losses incurred in the Northridge earthquake, insurers moved to better
understand their risk from earthquakes and develop pricing and policy terms that reflected this risk.
Insurance availability gradually returned to equilibrium on the commercial side, but residential lines
experienced a crisis in the years following. Personal lines insurers are required to offer the option of
earthquake insurance when selling homeowners’ policies, and many insurers stopped writing insurance
due to concerns that they would not be able to remain solvent if another similar event occurred. The
compromise was the CEA (California Earthquake Authority), a government entity funded by insurers.
The CEA offers a “mini-policy” with a high deductible (15%) and limited coverage for contents and
additional living expenses. Because it is perceived to be expensive relative to the coverage provided,
many homeowners elected not to purchase a earthquake rider on their policy. Only about 17% of
homeowners currently have earthquake insurance, down from 30-40% at the time of Northridge. The
combination of higher deductibles and lower limits with fewer policyholders has reduced the proportion
of losses that will be paid out by insurers. RMS model results suggest that insured residential losses from
a repeat of the Northridge earthquake today would be 70% lower than those incurred in 1994.
The commercial market in California is less regulated than for personal lines and has continued to grow
since 1994. An equivalent analysis of the insured annual loss for commercial lines yields almost twice the
percentage covered as for residential lines; this commercial result is shown on Figure 3 with an ‘x’. The
residential markets are gradually adapting to fill the demand for earthquake coverage, as evidenced by the
recent expansion of the CEA’s product line to include policies with lower deductibles and increased
coverage for contents and ALE.
CONCLUSIONS
We have presented relative seismic risk for thirteen western states on the basis of modeled results for
residential exposure, illustrated in terms of average annual loss cost. Natural breaks in per-state results
suggest four groups. California stands far above the rest in risk, a result borne out by historical experience
and common sense. Ranked in order of decreasing risk, Washington, Utah, Alaska and Nevada comprise
the next tier. Each have significant exposure close to active seismic sources. Hawaii and Oregon follow
closely behind; both have locally high hazard, but the loci of population and exposure are not in the
highest risk zones. There is a wide range in the last group, ordered from Montana, New Mexico, Idaho,
Wyoming, Arizona, to Colorado.
Relativities in these loss cost results are similar to the absolute economic risk, but have been normalized
to exposure and thus do not provide actual dollar losses. Washington is second in absolute loss to
California, followed by Utah, Oregon, and Nevada. The remaining eight states in rank order are Hawaii,
Alaska, Arizona, New Mexico, Montana, Idaho, Colorado, and Wyoming.
Insurance coverage of residential losses generally increases with the risk for the state, with California
currently a notable exception. The impact of the Northridge earthquake is still being felt, but conditions
are evolving in the insurance market that will eventually lessen the direct impact that would be borne by
homeowners in the next major earthquake.
REFERENCES
1. Frankel A, Petersen M, Mueller C, Haller K, Wheeler R, Leyendecker EV, Wesson R, Harmsen S,
Cramer C, Perkins D, Rukstales, K. “Documentation for the 2002 National Seismic Hazard Maps.”
US Geol Survey 2002; Open-File Rpt. 02-420.
2. Cornell CA, Krawinkler H. “Progress and challenges in seismic performance assessment.” PEER
Center News 2000; 3: 2.
3. Rahnama M, Seneviratna P, Morrow GC, Rodriguez A. “Seismic performance-based loss
assessment.” Proceedings of the 13th World Conference on Earthquake Engineering, Vancouver,
Canada. Paper no. 1050. Oxford: Pergamon, 2004.
4. Klein FW, Mueller CS, Frankel AD, Wesson RL, Okubo PG. “Seismic hazard in Hawaii: high rate
of large earthquakes and probabilistic ground motion maps.” Bull Seis Soc Amer 2001; 91: 479-
498.
5 Wesson RL, Frankel AD, Mueller CS, Harmsen SC. “Probabilistic seismic hazard maps of Alaska.”
US Geol Survey 1999; Open-File Rpt 99-36.
6. Working Group on California Earthquake Probabilities. “Earthquake probabilities in the San
Francisco Bay region: 2000-2030 – a summary of findings.” US Geol Survey 1999; OFR 99-517.
7. Working Group on California Earthquake Probabilities. “Earthquake probabilities in the San
Francisco Bay region: 2002-2031.” US Geol Survey 2003; Open-File Rpt. 03-214.
8. Field EH, Jackson DD, Dolan J. “A mutually consistent seismic hazard source model for Southern
California.” Bull. Seis. Soc. Amer. 1999; 89(3): 559-578.
9. Matthews MV, Ellsworth WL, Reasenberg PA. “A Brownian model for recurrence earthquakes.”
Bull. Seis. Soc. Amer. 2002; 92(6): 2233-2250.
10. Abrahamson NA, Silva WJ, “Empirical response spectral attenuation relations for shallow crustal
earthquakes,” Seis Research Ltrs 1997; 68(1): 94-127.
11. Boore DM, Joyner WB, Fumal TE, “Equations for estimating horizontal response spectra and peak
acceleration from western North American earthquakes: a summary of recent work,” Seis Research
Ltrs 1997; 68(1): 128-153.
12. Campbell KW, Bozorgnia Y, “Updated near-source ground motion (attenuation) relations for the
horizontal and vertical components of peak ground acceleration and acceleration response spectra,”
Bulletin of the Seismological Society 2003; 93(1):314-331.
13. Sadigh K, Chang CY, Egan J, Makdisi F, Youngs R, “Attenuation relationships for shallow crustal
earthquakes based on California strong motion data.” Seis Research Ltrs 1997; 68(1): 180-189.
14. Spudich P, Joyner WB, Lindh AG, Boore DM, Margaris BM, Fletcher JB. “SEA99: A revised
ground motion prediction relation for use in extensional tectonic regimes.” Bull. Seism. Soc. Am.
1999; 89: 1156-1170.
15. Youngs RR, Chiou S-J, Silva WJ, Humphrey JR. “Strong ground motion attenuation relationships
for subduction zone earthquakes.” Seismological Research Letters 1997; 68(1): 58-73.
16. Atkinson G, Boore D. “Preliminary empirical ground motion relations for subduction zone
earthquakes.” preprint 2001.
17. Wills CJ, Petersen M, Bryant WA, Reichle M, Saucedo GJ, Tan S, Taylor G, Treiman J. “A site
condition map for California based on geology and shear wave velocity.” Bull. Seis. Soc. Amer.
2000; 90: S187-S208.
18. Knudsen KL, Sowers JM, Witter RC, Wentworth CM, Helley EJ, Nicholson RS, Wright HM,
Brown KH. “Preliminary maps of Quaternary deposits and liquefaction susceptibility, nine-county
San Francisco Bay region, California: a digital database.” US Geol Surv 2000; OFR 00-444.
19. Youd TL, Perkins DM. “Mapping of liquefaction induced ground failure potential.” Proc. of the
ASCE, Journal of the Geotechnical Engineering Division 1978; 104 (GT4): 433-446.
20. Wilson RI, Wiegers MO, McCrink TP. “Earthquake-induced landslide zones in the City and
County of San Francisco, California.” Seismic Hazard Evaluation of the City and County of San
Francisco, California. California Division of Mines & Geology 2000; Open-File Rpt. 2000-009.
21. US Census Bureau, Population Div. “Annual Estimates of the Population for the United States and
States, and for Puerto Rico: April 1, 2000 to July 1, 2003”. Table NST-EST2003-01, 2003.
22. Petak WJ, Elahi S. “The Northridge earthquake, USA, and its economic and social impacts.”
EuroConference on Global Change and Catastrophe Risk Management, 2001.
http://www.iiasa.ac.at/Research/RMS/july2000/ Papers/Northridge_0401.pdf
23. Risk Management Solutions. “The Northridge Earthquake: RMS 10-yr Retrospective.” 2004.
http://www.rms.com/Publications/NorthridgeEQ_Retro.pdf
24. Pielke, Jr. RA, Landsea CW. “Normalized Hurricane Damages in the United States: 1925-1995.”
Weather and Forecasting 1998; 13: 621-631.
25. Property Claims Service division of Insurance Services Office
26. Sahr R. “Inflation Conversion Factors for Dollars 1665 to Estimated 2013.” Oregon State
University 2002; http://oregonstate.edu/Dept/pol_sci/fac/sahr/sahr.htm
27. FEMA. “HAZUS 99 Estimated Annualized Earthquake Losses for the United States.” FEMA 366:
Washington, DC. 2001.
28. Rowshandel B, Reichle M, Wills C, Cao T, Petersen M, Branum D, and Davis J. “Estimation of
Future Earthquake Losses in California.” California Geol. Survey 2003;
ftp://ftp.consrv.ca.gov/pub/dmg/rgmp/CA-Loss-Paper.pdf