Sensitivity Analysis of FEMA HAZUS Earthquake Model: Case Study from King County, Washington

Hazards U.S. Multi-Hazard Maintenance Release 4 (HAZUS-MH MR4) is damageand loss-estimation software developed by FEMA to estimate potential losses from natural disasters. Federal, state, regional, and local governments use theHAZUS earthquake model for earthquake risk mitigation, preparedness, response, and recovery planning. This paper examines earthquake model input parameters for earthquake source, including epicenter location, hypocentral depth, magnitude, and fault-plane dimensions, orientation, and dip, as well as geologic site conditions, to show how modifying the user-supplied settings affect ground-motion analysis, seismic risk assessment, and earthquake loss estimates.HAZUS calculates groundmotion and resulting ground failure to estimate direct physical damage for general building stock, essential facilities, and lifelines, including transportation systems and utility systems. Earthquake losses in HAZUS are expressed in building-damage, economic, and social terms; this paper focuses on monetary building damages, which are predicted by building type and occupancy classification (building use). This analysis centers on both shallow crustal and deep intraslab events that affect King County, Washington, in the Pacific Northwest; however, themethods and results of this paper may help to assess the accuracy ofHAZUS estimates more generally for seismically active regions. The results show that the estimated economic building damage varies by a factor of 14, on average, when usingmore accurate user-supplied source and site parameters rather than default values. In extreme cases, the estimated economic building damage varies by a factor ofmore than 500. The results also show thatHAZUS scenarios forKingCounty aremore sensitive to changes in source parameters than site conditions. The considerable variability in the estimated economic building damage can have a dramatic impact on both hazard-mitigation plans and initial postevent assessments used by emergency managers. DOI: 10.1061/(ASCE)NH.1527-6996.0000089. © 2013 American Society of Civil Engineers. CE Database subject headings: Seismic effects; Risk management; Sensitivity analysis; Earthquakes; Case studies; Washington. Author keywords: Seismic hazard; Risk assessment; Loss estimation;HAZUS; Sensitivity analysis; Earthquake; Modeling; Seismic effects. Introduction to HAZUS HAZUS is widely used loss-estimation software that provides a standardized methodology for assessing potential losses from earthquakes, floods, and hurricanes. HAZUS software operates on geographic information systems (GIS) technology to calculate potential physical, economic, and social impacts of disasters. Primary users ofHAZUS, such as government officials, GIS specialists, and emergency managers and responders, use the software to develop mitigation and recovery plans as well as preparedness and response procedures for a suite of natural hazards. FEMA asserts that for state, Indian tribal, and local governments to receive nonemergency disaster assistance, the governing body must have a standing hazard-mitigation plan (FEMA 2010). In addition to estimating losses prior to a natural hazard, governments run HAZUS scenarios after big events to estimate the degree of damage and to decide if federal recovery efforts may be needed. Given the potential enormity of the losses associated with natural hazards, it is imperative to understand the intrinsic limitations of running natural-hazard scenarios within the HAZUS framework. Many options are available to construct an earthquake scenario within HAZUS, including input parameters describing both the earthquake source and the geologic site conditions, where manipulation of these parameters may result in widely varying outputs. Using embedded databases, such as demographic aspects of the population, building counts and construction types, and utility and transportation lines,HAZUS users can carry out general loss estimations for a study region as a result of an earthquake (Whitman et al. 1997; FEMA 2000, 2003a, b, 2008, 2010). These loss estimates are presented in terms of economic and social losses, such as calculated direct losses for buildings (occupancy class), casualties, and shelter requirements owing to damage to building stock, essential facilities, and transportation and utility systems. Generally, model accuracy can be determined by comparison with real data; however, there have been few large earthquakes near U.S. urban centers in the last decade, yielding few rigorous tests of the HAZUS earthquake model. FEMA notes that uncertainties in the HAZUS loss-estimation model arise from incomplete scientific knowledge concerning earthquakes and their effects on buildings and facilities (FEMA 2003a). Limitations also arise from incomplete or inaccurate inventories of the built Graduate Student Researcher, Dept. of Earth Sciences, Univ. of California, Riverside, CA 92521 (corresponding author). E-mail: corrie. [email protected] Research Geophysicist, USGS, 525 South Wilson Ave., Pasadena, CA 91106; Adjunct Professor, Dept. of Earth Sciences, Univ. of California, Riverside, CA 92521. E-mail: [email protected] Senior Geographic Information Systems (GIS) Analyst, King County GIS Center, 201 South Jackson St., Suite 706, Seattle, WA 98104. E-mail: [email protected] Graduate Student Researcher, Dept. of PublicHealth, GeologicHazards and Disasters Research Group, Univ. of California, Irvine, CA 92697. E-mail: [email protected] Note. This manuscript was submitted on May 17, 2011; approved on July 23, 2012; published online on July 26, 2012. Discussion period open until October 1, 2013; separate discussions must be submitted for individual papers. This paper is part of the Natural Hazards Review, Vol. 14, No. 2, May 1, 2013. ©ASCE, ISSN 1527-6988/2013/2-134–146/$25.00. 134 / NATURAL HAZARDS REVIEW © ASCE / MAY 2013 Nat. Hazards Rev. 2013.14:134-146. D ow nl oa de d fr om a sc el ib ra ry .o rg b y U S G eo lo gi ca l S ur ve y L ib ra ry o n 05 /0 1/ 13 . C op yr ig ht A SC E . F or p er so na l u se o nl y; a ll ri gh ts r es er ve d. environment, demographics, and economic parameters within the embedded database. FEMA notes that these factors can result in a range of uncertainty in loss estimates produced by the HAZUS earthquake model, possibly at best a factor of 2 or more (FEMA 2003b). The HAZUS User’s Manual explicitly states that running a scenario using default data can result in incomplete estimated losses and recommends including at least some user-supplied input; however, the expected range over which the loss estimates may vary is not provided (FEMA 2003c). Al-Momani and Harrald (2003) and Kircher et al. (2006a) compared damage and loss estimated for the 1994 Mw 6.7 Northridge earthquake with actual damage and loss as a result of the earthquake and found that variations ranged between two and four for social losses (i.e., injured, killed, and displaced people). For the 1989 Mw 7.1 Loma Prieta earthquake, Al-Momani and Harrald (2003) found that social and economic losses varied within a factor of 3. Price et al. (2010) conducted a source-sensitivity study in Nevada and found that uncertainty in source characteristics can result in output variations often within a factor of 5 (i.e., total economic loss and number of buildings with major damage) but can be up to a factor of 13 (i.e., estimated fatalities). As is inherent to any model, there are many uncertainties owing to approximations and simplifications that can have a negative impact on model results. Previous studies considered the effects of changing earthquake-source and ground-motion functions on HAZUS outputs, and some compared their results with measured outcomes after an earthquake (Bendimerad 2001; Comartin-Reis 2001; Al-Momani and Harrald 2003; Eguchi and Seligson 2008). To assess ground-motion accuracy in HAZUS, Kircher et al. (2006a) compared HAZUS-modeled scenarios for the 1994 Northridge earthquake and found thatHAZUS (including site/soil amplification effects) underestimated the ground motion observed during the earthquake. Price et al. (2010) compared the impact of varying such inputs as epicenter location, hypocentral depth, magnitude, and fault-plane dip for selected cities in Nevada. Previous studies found that the variation across model scenarios depends on the HAZUS output examined; for example, there is more variation in estimated number of casualties than in building damage or total economic loss (Kircher et al. 2006a, b; Price et al. 2010; Shoaf and Seligson 2011). This variability has been attributed to building damage-state probabilities, such that casualties are more sensitive to the degree of damage, whereas all damage states contribute to the total economic loss (Kircher et al. 2006a). In this study, inputs are varied for the earthquake source, such as epicenter location, hypocentral depth, magnitude, fault-plane dimensions, and fault-plane orientation and dip, as well as geologic site conditions, for seven historic earthquakes affecting King County, Washington, and examine resulting variations in peak ground acceleration (PGA) and economic building damage (EBD). While variations in some source and ground-motion parameters have been investigated and quantified by other studies (e.g., Al-Momani and Harrald 2003; Price et al. 2010), this study employs a sensitivity analysis to answer the question: How much variation in output is there between a HAZUS catalog or default scenario and a scenario with realistic geophysical input parameters? Andmore specifically, how sensitive isHAZUS to variations in source parameters and site inputs? The study then further explores the spatial variation in output across all scenarios to allow a better understanding of where (and, similarly, why) the variation occurs. This study aims to provide quantitative estimates of the variability in PGA and EBD owing to commonly used implementations of source and site parameters in HAZUS scenarios. Introduction to the Study Region The study area is King County, WA, which has a population of over 2 million people and is the most populated county in the state. The county includes the Seattle-Tacoma-Bellevue region, which FEMA (2008) ranks as fifth in the nation for annualized earthquake loss. The region has a high potential seismic hazard as a result of three types of earthquake events: (1) ruptures along the length of the Cascadia subduction zone capable of producing up to Mw 9 earthquakes, (2) ruptures occurring deep in the subducting oceanic slab resulting inMw 6–7 events, and (3) ruptures along numerous shallow faults in the region (Fig. 1) (Atwater and Hemphill-Haley 1997; Goldfinger et al. 2003; Nelson et al. 2006; Bucknam et al. 1992; Atwater and Moore 1992). In the past century, this area has experienced four deep (∼55 km) plate-boundary intraslab earthquake events: the 1949 Mw 6.8 Olympia, 1965Mw 6.5Seattle-Tacoma, 1999Mw 5.8Satsop, and2001 Mw 6.8 Nisqually events. There also have been earthquakes on active shallow crustal faults within the Puget Sound area, the Mount St. Helens volcanic complex to the south, and othermajor fault systems in the region. Inparticular, the Seattle fault zone is a 4to 7-km-wide zone of blind shallow thrust faults with slip rates estimated to be 0.7–1.1 mm/year in the Puget Sound and Seattle urban center (Danes et al. 1965; Johnson et al. 1999; Blakely et al. 2002; ten Brink et al. 2002). Because the Cascadia subduction zone earthquake is an extreme, infrequent event, the study instead focuses on historic deep intraslab events and shallow crustal earthquakes. The largest shallow crustal events to affect the region in the last three decades are the 1981Mw 5.3 Elk Lake, the 1995Mw 5.0 Robinson Point, and 1996Mw 5.1 Duvall events. In total, seven historic earthquakes are analyzed to determine Fig. 1. (a) Tectonic setting of King County (outlined in dark gray) in Washington State relative to the Cascadia subduction zone (base map courtesy of USGS National Elevation Dataset); (b) locations of modeled earthquakes are shown as black stars on the cross-sectional schematic (deep earthquakes occur in the down-going Juan de Fuca oceanic plate, and shallow earthquakes occur in the North American continental crust) NATURAL HAZARDS REVIEW © ASCE / MAY 2013 / 135 Nat. Hazards Rev. 2013.14:134-146. D ow nl oa de d fr om a sc el ib ra ry .o rg b y U S G eo lo gi ca l S ur ve y L ib ra ry o n 05 /0 1/ 13 . C op yr ig ht A SC E . F or p er so na l u se o nl y; a ll ri gh ts r es er ve d. the variability in HAZUS ground-motion and building-loss estimates. By constructing scenarios for seven historic earthquakes, this paper allows for a quantitative look at variation across a study region and for a better understanding of the range of HAZUS model results. Ground-Motion Simulation