The 2018 update of the US National Seismic Hazard Model ......2018 NSHM update, comparisons with the 2014 NSHM, final products (e.g. maps, hazard curves, deaggregations), and general

Article

Earthquake Spectra

1–37

� The Author(s) 2019

Article reuse guidelines:

sagepub.com/journals-permissions

DOI: 10.1177/8755293019878199

journals.sagepub.com/home/eqs

The 2018 update of the USNational Seismic HazardModel: Overview of modeland implications

Mark D Petersen, M.EERI1, Allison M Shumway1,Peter M Powers, M.EERI1, Charles S Mueller1,Morgan P Moschetti, M.EERI1, Arthur D Frankel, M.EERI2,Sanaz Rezaeian1, Daniel E McNamara1, Nico Luco, M.EERI1,Oliver S Boyd, M.EERI1, Kenneth S Rukstales1,Kishor S Jaiswal, M.EERI1, Eric M Thompson1,Susan M Hoover1, Brandon S Clayton1,Edward H Field, M.EERI1, and Yuehua Zeng1

AbstractDuring 2017–2018, the National Seismic Hazard Model for the conterminous United States was updatedas follows: (1) an updated seismicity catalog was incorporated, which includes new earthquakes thatoccurred from 2013 to 2017; (2) in the central and eastern United States (CEUS), new ground motionmodels were updated that incorporate updated median estimates, modified assessments of the associ-ated epistemic uncertainties and aleatory variabilities, and new soil amplification factors; (3) in the west-ern United States (WUS), amplified shaking estimates of long-period ground motions at sites overlyingdeep sedimentary basins in the Los Angeles, San Francisco, Seattle, and Salt Lake City areas were incor-porated; and (4) in the conterminous United States, seismic hazard is calculated for 22 periods (from0.01 to 10 s) and 8 uniform VS30 maps (ranging from 1500 to 150 m/s). We also include a description ofupdated computer codes and modeling details. Results show increased ground shaking in many (but notall) locations across the CEUS (up to ~30%), as well as near the four urban areas overlying deep sedi-mentary basins in the WUS (up to ~50%). Due to population growth and these increased hazard esti-mates, more people live or work in areas of high or moderate seismic hazard than ever before, leadingto higher risk of undesirable consequences from forecasted future ground shaking.

Keywordsseismic hazard, seismic risk, earthquake rupture-forecast, ground motion model

Date received: 19 December 2018; accepted: 16 August 2019

1US Geological Survey, Denver Federal Center, Denver, CO, USA2US Geological Survey, University of Washington, Seattle, WA, USA

Corresponding author:

Mark D Petersen, US Geological Survey, Denver Federal Center, P.O. Box 25046, MS 966, Denver, CO 80225, USA.

Email: [email protected]

https://uk.sagepub.com/en-gb/journals-permissions

https://doi.org/10.1177/8755293019878199

journals.sagepub.com/home/eqs

http://crossmark.crossref.org/dialog/?doi=10.1177%2F8755293019878199&domain=pdf&date_stamp=2019-11-28

Introduction

Over the past four decades, the US National Seismic Hazard Model (NSHM) Project of theUS Geological Survey (USGS; for example, Algermissen and Perkins, 1976; Frankel et al.,1996, 2002; Petersen et al., 1996, 2008, 2014, 2015) has provided science-based hazard infor-mation for use in seismic provisions of US codes for buildings, bridges, railways, and defensefacilities, among other structures; risk assessments for insurance and disaster managementplanning; and federal, state, and local governmental policy and planning documents. Theseprobabilistic seismic hazard models define the potential for earthquake shaking by integrat-ing two fundamental inputs (Cornell, 1968): (1) earthquake rupture forecast models, whichdefine a potential range of earthquakes that could strike at any location across the UnitedStates and (2) ground motion models (GMMs), which provide estimates of the distributionof ground shaking from each event. Seismic hazard analyses show where and how oftenfuture earthquakes may occur and the predicted strength of the ground shaking. Seismicforecasts may be employed to encourage safer building designs in areas with high shakingpotential while saving resources in areas with lower shaking hazard. Hazard informationcoupled with prudent mitigation strategies can facilitate more resilient communities andenable building functionality following an earthquake-related disaster.

The NSHM was updated during 2017–2018 for the conterminous United States; Alaskaand Hawaii models are updated separately. Figure 1 shows a 2018 chance-of-damage map for100 years with population density superimposed. Many populated centers are coincident withareas of medium to high ground shaking hazard, not only across the western United States(WUS) where most earthquakes occur but also within the central and eastern United States

Figure 1. Map showing the chance of minor damaging earthquake shaking in 100 years from the 2018NSHM. The shaking is equivalent to Modified Mercalli Intensity VI and is based on the average 1-s horizontalspectral response acceleration. Ground motions are amplified using topographic-based VS30 values (Waldand Allen, 2007). Population density (from LandScan with 1 km 3 1 km resolution Dobson et al., 2000) issuperimposed on the map.

2 Earthquake Spectra 00(0)

(CEUS) where earthquakes are less common. In this article, we provide an overview of the2018 NSHM update, comparisons with the 2014 NSHM, final products (e.g. maps, hazardcurves, deaggregations), and general implications of the model for seismic hazard and risk.The model is built on the methodologies and parameters defined in previous versions of theNSHMs. We describe published information that was submitted before March 2017 and dis-cussed at our 2018 NSHM Update Workshop. Changes to the model fall into these primaryareas: (1) new earthquake catalogs and forecasts that account for locations of future earth-quakes based on past seismicity, (2) new models for assessing earthquake shaking within theCEUS, (3) new models for assessing ground motions over deep sedimentary basins underlyingWUS urban areas, and (4) new suites of maps for 22 periods and eight uniform VS30 mapsacross the conterminous United States. We also include a description of updated computercodes and modeling details.

Results of our updated analysis show that earthquake risk is quite high in many regions ofthe country which can be mitigated through improved engineering analyses. For example, theNSHMs are an important consideration in developing building codes, with the goal of moreresilient and safer communities. Engineers have responsibility for safety of construction inhigh hazard areas, so they are a key end user of the NSHMs and provide requests to USGSon the types of hazard products needed for building code improvements. The BuildingSeismic Safety Council (BSSC, 2019) Provisions Update Committee and Project 17 workedwith the USGS to improve building code procedures by connecting the best available scienceand data with sound engineering principles. In past versions of their seismic provisions, theBSSC recommended development of peak horizontal ground acceleration (PGA) as well as0.2- and 1-s spectral accelerations (SA) that define simplified response spectral shapes with afirm-rock site condition of 760 m/s VS30 (time-averaged shear-wave velocity in the upper30 m of the crust). The new building provisions allow for increased seismic safety of buildingsin areas of high seismic risk and this is one of the most important applications of these mod-els. The new model provides more complete ground shaking information for engineeringpurposes.

In addition to the building code committees, the insurance industry, government agencies,risk modelers, and policy makers are also important end users of the NSHMs and use thisinformation to evaluate earthquake risk and mitigation priorities. The insurance and nuclearindustries typically apply hazard and risk models generated by third-party risk modelinggroups who use information from the NSHMs that are needed for estimating the impacts ofground shaking peril on losses. Government officials, emergency management, and publicpolicy makers use the NSHMs to address local efforts and to understand and reduce seismicrisk. For example, Federal Emergency Management Agency (FEMA) incorporates theUSGS hazard models directly in their computer programs to assess earthquake risk to thenation and help communities prepare for earthquake shaking (e.g. FEMA 366 and HAZUSsoftware; Jaiswal et al., 2015, 2017; http://www.fema.gov/hazus/). Other examples of endusers are the government agencies such as the US Army Corps of Engineers, Bureau ofReclamation, and the Department of Interior who use the models to help evaluate earth-quake risk for dams and other important facilities as well as geological surveys and otherstate and local governments who have responsibility for regional and local seismic hazardand risk planning. The NSHMs can be useful for prioritizing retrofit schedules, reassessingvulnerable structures, or developing risk mitigation strategies. This document providesimportant information on the likelihood and consequences of earthquakes which benefitsthese important communities. An updated assessment of the population that is exposed atalternative risk levels is presented in this publication.

Petersen et al. 3

One of the primary improvements to the 2018 model is incorporating new GMMs and siteamplification factors that increase the number of period and site classes compared to GMMsapplied in earlier NSHMs. New ground shaking information recently became availablethrough a decade-long ground motion research study that was released as the NextGeneration Attenuation Relationships for the central and eastern North America Region(NGA-East, Goulet et al., 2017, 2018) and corresponding site amplification factors (Hashashet al., 2019; Stewart et al., 2019b). Previous versions of the CEUS GMMs did not allow forcalculations of many periods and site classes that were available in the WUS GMMs.However, these new NGA-East equations enable the USGS to directly calculate the groundshaking at similar periods and site classes as in the WUS GMMs (i.e. multi-period responsespectra: https://www.nibs.org/page/bssc_P17C). These more detailed products wererequested by the BSSC and serve to improve building code design procedures. In the 2018NSHM update, we calculate shaking across the United Sates for more than double the peri-ods and site classes than for the 2014 NSHM (Shumway et al., 2018).

The NSHMs consider both short and long-period ground shaking levels because buildingdesign codes require short-period (high frequency) ground motions for designing smaller(one- or two-story) structures and long-period ground motions for designing tall buildingsand long-span bridges. In this update, we concentrate on better defining the long periods[T] ø 1 s in the WUS by using supplemental information on depths of sedimentary basinsand basin amplification factors provided in the GMMs. It is clear from ground motion obser-vations and three-dimensional (3D) simulations (e.g. Aagaard et al., 2008; Frankel et al.,2018; Graves et al., 2011; Moschetti et al., 2017) that deep sedimentary basins amplify long-period ground motions (defined in this article as periods [T] ø 1 s) due to reflecting, refract-ing, and focusing of waves that are trapped within the basin structure. We account for modi-fied shaking using the NGA-West2 GMMs (Abrahamson et al., 2014; Boore et al., 2014;Campbell and Bozorgnia, 2014; Chiou and Youngs, 2014). Considering both the seismic velo-city models for urban regions and GMMs with deep basin effects allows us to improve thecalculations of long-period ground shaking potential for sites overlying sedimentary basins.This requires that new hazard models incorporate detailed information about the geometryand velocities of basin structure. In previous NSHMs, we did not consider basin-depth termsbecause (1) the building code committees were not requesting long-period ground motionmaps, (2) we wanted to construct uniform maps across the United Sates, and (3) detailed 3Dvelocity models were not uniformly available at a scale useful for hazard evaluations. Duringthis 2018 update, Project 17 requested long-period information for consideration in the designprocedures, and we decided that it would be preferable to provide information on a regionalbasis instead of waiting for this uniform velocity information to be developed across thecountry. In this update, we are considering, for the first time, seismic velocities and depths ofsedimentary basins that can be explicitly used to assess the long-period shaking for four areasthat have been characterized at a sufficiently detailed scale: Seattle, San Francisco Bay area,greater Los Angeles region, and areas surrounding Salt Lake City. In other areas, we useaverage models that implicitly account for basin shaking conditioned on the shallower siteparameter, VS30.

In developing updates to the 2018 NSHM, the USGS makes every reasonable effort toinclude the best science, data, models, and methods available at the time of publication forthese important applications (e.g. building codes, risk analyses, public policy). Collaboratorsprovide technical advice, review model details, and assess the reasonability of the models;they include (1) nine members of the NSHM Steering Committee who serve as active partici-patory experts in the development and review of the model; (2) more than 100 technicalexperts from the USGS and other federal, state, and local government agencies as well as


from academia and industry who provide essential input information and review; and (3) thepublic who attended our 2018 NSHM workshop and/or submitted comments during ourreview period. The public workshop was held from 7 to 8 March 2018, in Newark, CA, tobuild community consensus on the best methods, data, and models to apply in the assessmentand was attended in person and online by about 140 science and engineering experts acrossthe country. Meeting presentations are posted at our website: https://earthquake.usgs.gov/hazards/workshops/update2018.php. The model was also open for comments by any inter-ested members of the public from 7 November to 7 December 2018. Additional figures andsupporting information can be found at the USGS ScienceBase Catalog (Rukstales andPetersen, 2019). The new model will be considered for incorporation in the 2020 NationalEarthquake Hazards Reduction Program (NEHRP) Provisions, 2022 ASCE7 Standard, 2024IBC, as well as other building code, risk, and policy implementations.

Update of earthquake catalog and rate models

We use fault models from the 2014 NSHM (Boyd et al., 2015; Frankel et al., 2015; Halleret al., 2015; Mueller et al., 2015; Petersen et al., 2014, 2015; Powers and Field, 2015) to modellarge earthquakes, and apply gridded, smoothed seismicity models from an earthquake cata-log to account for smaller earthquakes on and off the faults. To facilitate this update, wedeveloped new seismicity catalogs for the CEUS and WUS including earthquakes from 2013through 2017 that occurred since the last model was constructed. Between 2013, when thecatalog was last updated, and 2018, strongly felt earthquakes (magnitude 4+) occurred inalmost half of the states in the United States (e.g. Alaska, Arizona, California, Colorado,Delaware, Hawaii, Idaho, Kansas, Michigan, Montana, Nebraska, Nevada, New Mexico,Oklahoma, Oregon, South Carolina, Tennessee, Texas, Utah, Washington, and Wyoming;Figure 2). In this section, we describe the earthquake catalog development and the gridded(smoothed) seismicity model.

Earthquake catalog development

A critical component of the NSHMs is an earthquake catalog that can be used to assess thesizes and locations of past earthquakes and guide us in our assessment of the location andrate of future earthquakes. The catalog used for the 2014 NSHM starts in 1700 and ends inDecember 2012. To update the catalogs for both the CEUS and WUS through 2017, weincorporated new data from the USGS, the Geological Survey of Canada, and Saint LouisUniversity. We used the methodology described by Mueller (2019) to develop catalogs forseismic hazard analyses: converting original magnitudes to uniform moment magnitudes,deleting duplicate events, deleting non-tectonic events, and declustering. We use the notationM to represent moment magnitude in the rest of the article. We exclude induced earthquakesin the CEUS following the method described in Petersen et al. (2016a: Table 1). In severalcases, such as Oklahoma, southern Kansas, and north-central Arkansas, the polygon bound-aries used for the 2018 NSHM have been updated from 2014. As shown by Petersen et al.(2016a), new zones have also been added (e.g. near Alice, Texas, and Perry, Ohio). Theinduced earthquakes were removed because they are ephemeral features, change quite rapidlyover short times, and are not appropriate for consideration in a long-term hazard model. Wehave considered these earthquakes in several short-term (1-year) forecasts for the CEUS(Petersen et al., 2016a, 2016b, 2017, 2018). The catalog and earthquake sources for Californiaare not updated for the 2018 NSHM because the California earthquake rates are interwoveninto the Uniform California Earthquake Rupture Forecast model (UCERF3; Field et al.,

Petersen et al. 5

2014). We examined the rate grids for California using an updated catalog through 2017to compare with the 2012 catalog applied in UCERF3 and found that differences were typi-cally less than 20%, very localized, and often in areas of active faults which dominate thehazard.

There are several challenges in developing the NSHM catalogs and seismicity-based statis-tical models that rely on an independent set of tectonic earthquakes. One challenge is to con-sistently and uniformly estimate M for small earthquakes and for some older earthquakes.One way to lessen the impact of these uncertainties is to raise the minimum magnitude ofcompleteness used in the seismicity recurrence analysis, for example, from M2.7 to M3.0. Inthe future, more refined classifications of the data by source catalog, era, or region might alsolead to improved procedures for estimating M and for identifying duplicates. Another chal-lenge is to develop declustered models that remove dependent foreshocks and aftershocks.For this analysis, the NSHM catalogs are declustered using the methodology of Gardner andKnopoff (1974), which is based on California data. Assessment of various declustering methodssuggest that this method of declustering is reasonable for tectonic earthquakes across the UnitedStates (e.g. CEUS–SSCn, 2012). A third challenge is to identify explosions, induced earthquakes,and mining-related seismic events using the limited available information (Mueller, 2019).

Figure 2. The 2018 NSHM declustered seismicity catalogs: (M ø 2.7) for the central and eastern UnitedStates (CEUS) (red) and (M ø 4.0) for the western United States (blue). New CEUS magnitudecompleteness zones shown as green polygons. CEUS completeness years (M2.7+, M3.7+, M4.7+) for sevencompleteness zones for the 2018 NSHM: C1 (1970, 1880, 1800), C2 (1930, 1750, 1700), C3 (1970, 1880,1800), C4 (1970, 1880, 1800), C5 (1980, 1920, 1850), C6 (1980, 1920, 1850), and C7 (1980, 1920, 1850).


Gridded or smoothed seismicity-based rate models

As in the previous NSHM, gridded and smoothed earthquake rate forecasts were developedseparately for the CEUS and WUS (outside of California) using the updated catalogs, assum-ing that future earthquakes will occur near the locations of past events. To accomplish this,earthquakes are counted and rates are computed on a 0.1�-latitude-by-0.1�-longitude grid foreach region, accounting for catalog completeness. Completeness varies by region, reflectingpatterns of written history and seismic network coverage.

The models assume that earthquakes follow an exponential magnitude-frequency distribu-tion as in previous models (e.g. Petersen et al., 2014). Regional b values used in the analysisare unchanged from the 2014 NSHM: 1.0 for the CEUS (the Charlevoix, Canada source wasb = 0.9 in the 2014 NSHM but was updated to 1.0 after a sensitivity study indicated no sig-nificant changes in US hazard) and 0.8 for the WUS. To smooth the gridded rates, we applyfixed-kernel and adaptive-kernel (nearest-neighbor) smoothing methods (Frankel, 1995;Helmstetter et al., 2007; Moschetti, 2015; Petersen et al., 2014). As in the 2014 logic tree, thefixed-kernel model is assigned three-fifth (0.6) weight and the adaptive-kernel model two-fifth (0.4) weight. Parameters such as minimum magnitudes of completeness, b values, andsmoothing distances differ for the fixed- and adaptive-smoothing models. The two modelsrepresent an important epistemic uncertainty in the NSHM source model (Moschetti et al.,2015). Both models consider statistical-based counts of earthquakes in a particular grid cell(10a values; Frankel, 1995). The 2014 version of the UCERF3 smoothed seismicity model(Field et al., 2014; Powers and Field, 2015) is used for California, so the seismicity rate fore-casts have not changed for the 2018 update.

Table 1. Logic tree for Updated Seed GMMs

Geometric spreading(\50–70 km)

Model type Model Model weight

R–1 (0.33) Point Sourcea (0.67) B_bca10db (0.3) 0.06633B_ab95b (0.1) 0.02211B_bs11b (0.1) 0.022112CCSPb (0.25) 0.0552752CVSPb (0.25) 0.055275

Traditional Empiricala,c (0.33) Graizer16d (0.5) 0.05445Graizer17e (0.5) 0.05445

R–1.3 (0.33) Hybrid Empirical-Stochastica (0.33) PZCT15_M1SSb (0.5) 0.05445PZCT15_M2ESb (0.5) 0.05445

Hybrid Empirical-Broadbanda (0.33) SP16f (1.0) 0.1089Stochastic Equivalent PointSourcea (0.34)

YA15b (1.0) 0.1122

Otherg (0.34) Other (1.0) HA15b,h (0.33) 0.1122Frankelb,i (0.33) 0.1122PEER_GPFrankelb,i (0.34) 0.1156

GMM: ground motion model.aDetails are in Goulet et al. (2017, 2018).bAcronyms for NGA-East seed models from Table 1.2 of Pacific Earthquake Engineering Research Center (2015).cSpectral shapes developed for the western United States empirically adjusted to central and eastern United States.dGraizer16 is an update of the NGA-East seed model Graizer.eGraizer17 is an alternative to the Graizer16 model.fSP16 is an update of the NGA-East seed model SP15.gGeometric spreading different from R–1 or R–1.3.hReference Empirical.iSimulation-based Hybrid.

Petersen et al. 7

The fixed-kernel smoothing applies a two-dimensional (2D) Gaussian operator with 50 or75 km correlation length (Frankel, 1995). This model uses a fixed value to assign reasonabledistances between future and past earthquakes. The WUS completeness model for areas out-side of California is unchanged from previous NSHMs:M4+ since 1963, M5+ since 1930, andM6+ since 1850. The CEUS completeness zonation consists of seven zones (Figure 2), and itis modified in two ways from the 2014 NSHM, which was roughly based on CEUS–SSCn(2012): (1) the zone encompassing the Washington–New York–Boston corridor is extendedsouthwestward into central Virginia, and (2) the westernmost zone is divided at –100� longi-tude and extended westward to the NSHM CEUS boundary. We determine completenessyears for magnitudes 2.7+, 3.7+, and 4.7+ for each of the seven completeness zones by exam-ining numbers of earthquakes in magnitude and time bins; results are similar to the 2014NSHM. Figure 2 shows the 2018 CEUS earthquake catalog and completeness zonation thatwe apply for the fixed-kernel smoothing model.

The adaptive-kernel smoothing employs smoothing distances that are based on a nearest-neighbor parameter (Helmstetter et al., 2007). This model has the effect of spatially concen-trating seismicity rates in regions of high seismicity and of diffusing seismicity rates in regionsof low seismicity, relative to fixed-kernel smoothing. The method uses event-specific smooth-ing parameters that are identified by optimizing a likelihood parameter (Helmstetter et al.,2007; Werner et al., 2011), which is separately computed for the WUS (Petersen et al., 2014)and CEUS (Moschetti, 2015). For the 2018 NSHM, we used the smoothing parametersderived in these previous studies, but with updated earthquake catalog durations. Thesmoothing parameters for the adaptive-kernel method correspond to (1) specific catalogdurations and (2) minimum magnitudes of completeness. Therefore, we use the same mini-mum magnitudes of completeness from the 2014 NSHM. To maintain the catalog duration,we advance the completeness year of the earthquake catalog by five additional years toaccount for the increased duration in the 2018 earthquake catalog. The reason we need tomaintain the catalog duration is that if we apply the constant nearest-neighbor parameter toa longer catalog, this would have the effect of spatially concentrating the seismicity morethan is supported by the likelihood testing. The effect on the earthquake rates and hazards ofthis choice in catalog duration is minimal, less than 0.05 g differences everywhere with twoexceptions (slightly larger differences appear over very small areas in eastern Utah andsouthern Montana). For the adaptive-kernel smoothing models, we used neighbor num-bers of 3 for the WUS (Petersen et al., 2014) and 4 for the CEUS (Moschetti, 2015), withthe updated earthquake catalogs. For the WUS, the minimum magnitude of completenesswas M = 4 since 1963, with a b value of 0.8 applied across the entire region. For theCEUS, the minimum magnitude of completeness was M = 3.2 with a b value of 1.0applied uniformly across this CEUS region, and with spatially variable completeness timesdefined using the zones described by Petersen et al. (2014) and completeness values com-puted by Moschetti (2015).

We tested the sensitivity of the seismic hazard to raising the magnitude threshold (Mmin) inthe fixed-smoothed seismicity recurrence calculation from Mmin = 2.7 to Mmin = 3.0 in theCEUS. The result was sensitive to single earthquakes in areas of sparse seismicity, was notrobust, and resulted in minimal changes to the hazard over small areas. We have not appliedthis alternative in the 2018 update.

Update of GMMs in the CEUS

In developing the 2014 NSHM, we applied GMMs for the CEUS that were developed byindividual teams of ground motion experts that applied their independent opinions of the best


science and physical principles (Rezaeian et al., 2015). However, the modeling decisions andoutput parameters (e.g. site conditions, magnitude, and distance ranges) differed betweenteams making it difficult to compare alternative models and apply these in an NSHM. One ofthe purposes of the NGA-East models was to overcome these inconsistencies by standardizingthe input data and output parameters through a collaborative process (Goulet et al., 2017). Indeveloping the 2018 NSHM, we use the new GMMs developed by the NGA-East Project(Goulet et al., 2017), new estimates of aleatory variability (sigma; Stewart el al., 2019a), andnew estimates of site amplification factors (Hashash et al., 2019; Stewart et al., 2019b). Thesemodels result in significant changes to ground shaking hazard.

CEUS median GMMs and weights

We used two sets of GMMs that were developed during the NGA-East Project: (1) 19adjusted seed GMMs developed by independent modelers (Pacific Earthquake EngineeringResearch Center (PEER), 2015) and (2) 17 updated ‘‘NGA-East for USGS’’ GMMs devel-oped by an integrator team that account for epistemic uncertainty in the median groundmotions as a function of magnitude, distance, and oscillator period (Goulet et al., 2017, andaddendum). The adjusted seed GMMs were developed by independent modelers using a con-sistent set of data and simulations with their own interpretation of the best science. Two ofthe seed GMMs were updated following the 2015 PEER report to account for new informa-tion and enhanced modeling techniques: Graizer (2016) and an alternative Graizer (2017),and Shahjouei and Pezeshk (2016). Based on expert recommendations, we reduced the num-ber of seed GMMs from 19 models to 14 models, which we refer to as the Updated SeedGMMs. Following some iteration with the NGA-East Technical Integration (TI) team(Goulet et al., 2017, with addendum), the 17 NGA-East for USGS models ended up beingidentical to the 17 final NGA-East models for nuclear facilities, which were released after this2018 NSHM model was completed (Goulet et al., 2018).

The new CEUS’s GMMs used in the 2018 NSHM are based on two sets of GMMs(Updated Seed GMMs and NGA-East for USGS GMMs) and on a reference site conditiondescribed by VS30 of 3000 m/s and kappa (high-frequency decay of Fourier Spectra) of0.006 s (Hashash et al., 2014), magnitudes from M4 to 8, distances of 0 out to 1500 km (weonly apply the models out to 1000 km), and median (RotD50) 5%-damped SAs for periodsfrom 0.01 to 10 s. Logic tree weights assigned to the Updated Seed GMMs are based on the2014 NSHM methodology that groups GMMs based on their model types and geometricspreading features, as well as some judgment on our confidence in their performance basedon extrapolations and available residual analysis (Table 1; McNamara et al., 2018; Rezaeianet al., 2015). Weights for the NGA-East for USGS GMMs are provided by the NGA-EastTI team and are based on Sammon’s mapping methodology (Sherbaum et al., 2010). Overall,the Updated Seed GMMs receive one-third (0.333) weight to account for alternative modelsthat are developed by independent modelers and may account for physics-based alternativeswhile the NGA-East for USGS GMMs receive the other two-third (0.667) weight to accountfor the additional effort by the developers to quantify the epistemic uncertainty more fully,as discussed at the 2018 NSHM Update Workshop. The new 2018 NSHM CEUS GMMsrepresent a significant advancement in the forecasting of ground shaking in the CEUS formany new periods based on a new uniformly developed strong-motion database and earth-quake simulations.

Figure 3 shows the mean of the weighted medians for each set of CEUS GMMs consid-ered in the 2014 and 2018 NSHMs for a variety of oscillator periods and for M7 and M5earthquakes on hard rock (VS30 = 3000 m/s) at short distances (10 km) and longer distances

Petersen et al. 9

(50 km). The new NGA-East for USGS and Updated Seed averaged GMMs are similar tothe averaged value of the group of GMMs applied in the 2014 NSHM, with the exception ofoscillator periods lower than about 0.1 s. Most of the ground motion values are similar forperiods longer than about 0.1 s SA, varying by less than 10% from one another; differencesare also quite low, typically less than 0.03 g.

Aleatory variability (sigma)

The new aleatory variability (i.e. sigma) model for the CEUS GMMs is based on a logic treewith two branches: (1) an updated version of Electrical Power Research Institute (EPRI,2013) recommended by the NGA-East for USGS report (Goulet et al., 2017), and (2) a work-ing group model that accounts for additional aleatory within-event site-to-site variability(fS2S, Stewart et al., 2019a). Figure 4a shows the two branches and final model. We onlyapply the central branch of the NGA-East for USGS sigma model because a sensitivity studyshowed application of additional branches did not cause significant differences in meanground shaking hazard.

(a) (b)

(c) (d)

Figure 3. Median ground motion versus period plots showing comparisons of central and eastern UnitedStates (CEUS) ground motion models (M = 5, 7) for (a) VS30 = 3000 m/s, distance (R) = 10 km, (b)VS30 = 3000 m/s, R = 50 km, (c) VS30 = 760 m/s, R = 10 km, and (d) VS30 = 760 m/s, R = 50 km. Note thatthe 2014 NSHM ground motions were not calculated between peak ground acceleration and 0.1 s spectralacceleration.


The NGA-East Project developers originally recommended that the USGS apply an ergodicsigma model that is an updated version of the EPRI (2013) model (Al Atik, 2015; Table 5.5 ofGoulet et al., 2017) for VS30 = 3000 m/s. This model was vetted in a Senior Seismic HazardAnalysis Committee (SSHAC) process and is largely based on analysis of ground motion varia-bility in active regions like California. During the 2018 NSHM Update Workshop, concernswere raised regarding potential differences in ground motion variability in the CEUS relative toactive regions. It was concluded that an additional model was needed in the aleatory variabilitylogic tree that would account for additional site-to-site variability arising from lack of knowledgeof site conditions in the CEUS and from VS30 being a less effective proxy for site effects in theCEUS relative to the WUS (Parker et al., 2018). We convened a working group,

1

which ulti-mately developed a model that increases aleatory variability for shorter periods while leaving the

(a)

(b)

(c)

Figure 4. Comparison of the (a) updated EPRI, Working Group, and final weighted 2018 CEUS sigmamodels for M = 7.5, VS30 < 1200 m/s. Comparison of the sigma models for the central and eastern UnitedStates (CEUS) and western United States (WUS) for (b) VS30 < 1200 m/s and (c) VS30 ø 1500 m/s.

Petersen et al. 11

longer periods similar but lower than the original model recommendation and the 2014 CEUSandWUS crustal GMM sigmas (EPRI, 2013; Goulet et al., 2017, Figure 4).

Our final, total aleatory variability model gives the updated EPRI model four-fifth (0.8)weight and the working group model (Stewart et al., 2019a) one-fifth (0.2) weight. The lattermodel is weighted lower because even though workshop participants felt strongly about theneed for inclusion of site-to-site variability, the model is based on smaller magnitude earth-quakes in the CEUS (no records from large events are available) and this leads to largeuncertainties in the analysis. The CEUS typically has a broader range of stress drops thanearthquakes in the WUS, so we expect higher sigmas in the CEUS; more research is neededto better ascertain these uncertainties.

Figure 4b and c shows comparisons of the sigma models and illustrates that the 2018CEUS sigma model is higher than the 2014 CEUS model for periods up to about 1-s SA,regardless of site class. As discussed earlier, the 2018 model sigmas for the CEUS should notdrop significantly, if at all, below the sigmas applied in the WUS. Because of the weighting,the final sigmas are very similar to the EPRI and WUS sigmas for long periods. Figure 4ashows comparisons of the working group model, updated EPRI and 2018 final models, anddemonstrates that the working group model is higher than both the EPRI and 2018 modelsup to about 1-s period. It is interesting to note that the sigmas for M5 are considerably largerthan those applied for M7 earthquakes.

Site amplification factors

The NGA-East results only account for ground shaking on hard-rock site condition ofVS30 = 3000 m/s. Therefore, the USGS funded a CEUS amplification working group

2

toconsider how to account for site amplification across the East as a function of VS30. Thisgroup developed new CEUS site amplification factors for the 2018 NSHM that includedboth linear and nonlinear terms (Hashash et al., 2019; Stewart et al., 2019b). These factorsconvert the NGA-East hard-rock site condition (VS30 = 3000 m/s) to a firm-rock site condi-tion (VS30 = 760 m/s) and then to the specific site condition defined by VS30 between2000 m/s and 200 m/s. Although the model as published is recommended for use down toVS30 = 200 m/s, the VS30 term is essentially flat at VS30 less than about 250 m/s and capturesincreasing uncertainty as VS30 decreases. Based on discussions with the modelers, we applythe model down to VS30 = 150 m/s to compute ground motions for NEHRP site class E. InFigure 5, we compare the new models for VS30 = 760 m/s to the Frankel et al. (1996) andAtkinson and Boore (2011) models applied in the 2014 NSHM. For periods greater thanabout 0.2 s, the new firm-rock amplification models are lower than previous models. Figure 5also shows linear, nonlinear, and total amplification factors at (Figure 5b) VS30 = 2000 m/sand (Figure 5c) VS30 = 260 m/s.

The final CEUS model was developed by the working group in coordination with theNSHM Project and Steering Committee. The final model includes VS30-dependent weightingof site profiles that were more gradual (gradient model) and profiles characterized by sharpvelocity contrasts (strong impedance model; Stewart et al., 2019b). The impedance model isstrongly peaked between 0.075 and 0.15 s and, at the recommendation of our SteeringCommittee, we smoothed the impedance model term and associated uncertainty at 0.1 s viainterpolation. The modified term at 0.1 s is well within 61s uncertainty envelope of theimpedance model. Key elements of the amplification model include (1) flatter VS30 scalingand (2) alternative models for VS30 = 760 m/s amplification relative to the hard-rockVS30 = 3000 m/s amplification (the F760 term in Stewart et al., 2019b). The interpolatedimpedance model is available as part of our NGA-East for USGS implementation innshmp-haz (Powers, 2017).


The updated CEUS amplifications differ in several ways from WUS amplification models:(1) the CEUS model for stiff site conditions is peaked at 0.1 s due to the presence of strongimpedance (shallow soil and weathered rock over hard rock) compared to WUS amplifica-tions that are generally peaked at 0.2–0.3 s; (2) the shape of the CEUS model rises and fallsmuch more rapidly and generally has higher peaks for short periods than WUS crustal mod-els; (3) spectra for different site classes are not as evenly spaced; and (4) the CEUS modelshows larger differences between VS30 of 2000 and 760 m/s compared to models in the WUSGMMs (Figure 6).

GMMs in the WUS

For the 2018 NSHM, we implemented the same WUS GMMs (crustal GMMs and subduc-tion interface and intraslab GMMs) that were used to calculate the most recent set of mapsfor the 2014 NSHM (Shumway et al., 2018). This set excludes the Atkinson and Boore (2003,

(a)

(b) (c)

Figure 5. (a) Comparison of central and eastern United States (CEUS) amplification factors (M7,distance = 10 km) from hard rock to VS30 = 760 m/s used in the 2014 NSHM (Atkinson and Boore, 2011;Frankel et al., 1996) and the 2018 NSHM. Linear, nonlinear, and total amplification factors at (b)VS30 = 2000 m/s and (c) VS30 = 260 m/s.

Petersen et al. 13

2008) subduction GMMs and Idriss (2014) NGA-West2 crustal GMM that were included inthe original 2014 update, because these models do not have the required features to allow forcalculation of hazard at additional periods and site classes. The Atkinson and Boore (2003,2008) GMMs did not include periods beyond 3 s, and the Idriss (2014) GMM does not sup-port soil site conditions below 450 m/s. For the 2018 NSHM, we required that GMMs bevalid (or reasonably extrapolated) for a broad range of spectral periods (i.e. 0.01–10 s) andsite conditions (i.e. VS30 values between 1500 and 150 m/s), consistent with new building coderequirements. The exclusions do not reflect the quality of the models, only their limited rangeof applicability; they may be appropriate for other applications.

Basin amplification of ground shaking in the WUS

The 2014 NSHM did not consider basin depths directly in the hazard calculations for theWUS. The model did, however, apply VS30 site terms as a proxy for basin depth, given thatsofter soils are more likely to occur in deeper basins and commonly result in larger long-period ground motions (this proxy assumption is not as useful for the Puget Sound area inthe Pacific Northwest region). For the 2018 NSHM, we identified four urban regions in theWUS for which geophysical information is available to characterize depths to various shear-

Figure 6. Ground motion versus period plots showing comparisons of central and eastern United States(CEUS) and western United States (WUS) median ground motions (M7, distances (R) of 10 and 50 km) atvarious NEHRP site classes.


wave velocity horizons and which should improve the longer-period calculations of the 2018NSHM. This basin-specific depth information is input into the NGA-West2 GMMs toaccount for enhanced shaking from sedimentary basin deposits. In this section, we discuss (1)acquisition of velocity information, (2) application of the basin factors using the NGA-West2 equations, and (3) modification of subduction GMMs to include basin terms for theSeattle region.

Velocity information

To update the NSHM, we explored published scientific literature to find urban regions builton sedimentary basins for which 3D shear-wave velocity models are available. Basin amplifi-cation and deamplification in the NGA-West2 GMMs relies on depths to a shear-wave velo-city of 2.5 km/s (Z2.5, Campbell and Bozorgnia, 2014, referred to hereafter as CB14) or1.0 km/s (Z1.0, Abrahamson et al., 2014, referred to hereafter as ASK14; Chiou and Youngs,2014, referred to hereafter as CY14; and Boore et al., 2014, referred to hereafter as BSSA14).While tomographic velocity models are available that span the conterminous United States,we concluded that these large scale models often have resolutions that are several tens ofkilometers between points and would not characterize details of the basins sufficiently for theNSHM. We instead identified published community velocity models for four WUS urbanregions that we felt had sufficient resolution for purposes of identifying basin velocities: LosAngeles (Lee et al., 2014), San Francisco Bay area (Aagaard et al., 2008), Seattle(Stephenson, 2007 and Stephenson et al., 2017), and Salt Lake City (Magistrale et al., 2008).Some of the local velocity models spanned a broader geographic area than the four limitedregions we defined. For example, the Aagaard et al. (2008) velocity model includes portionsof the Central Valley basin, which we exclude because including (1) portions of a basin intro-duces spatial discontinuities in the hazard results and (2) ground motion simulations for theCentral Valley, unlike for the other basins we include, are not consistent with the NGA-West2 basin terms. Other basins may be considered in future updates when more analysesare completed.

Figure 7 shows polygons around the four urban areas in the WUS in which local seismicvelocity model data are considered in the 2018 NSHM. Models show Seattle, Los Angeles,and Ventura basins have maximum Z2.5 depths of 7–8 km (Lee et al., 2014; Stephenson,2007). The Livermore basin and other basins in the San Francisco Bay area have Z2.5 valuesdown to about 4 km (Aagaard et al., 2008). Based on the Magistrale et al. (2008) model, themaximum Z2.5 depths in the Salt Lake City basin are slightly greater than 3 km only in a fewsmall areas north and south of Salt Lake City. These velocity models are incorporated in the2018 NSHM only when they primarily influence long-period ground shaking.

Application of basin factors in GMMs

We apply NGA-West2 GMMs in our hazard calculations for crustal earthquakes in activetectonic regions which include factors that account for basin amplification and deamplifica-tion based on Z1.0 and Z2.5. The four NGA-West2 GMMs in the 2018 NSHM account forbasin depth directly (CB14) or a differential basin depth, defined as the difference betweendepth and the mean depth conditioned on VS30 (referred to here as ‘‘default basin depth’’;ASK14, BSSA14, and CY14). Table 2 shows the default basin depths used in the NGA-West2 GMMs conditioned on VS30 (VS30 values used to define site class and site class bound-aries are from Shumway et al., 2018). While basin effects are well known to be complex andinfluenced by many factors besides the depth of the basin at the location of the station, the

Petersen et al. 15

Figure 7. Map of the local seismic velocity models used in the 2018 NSHM, showing the outlines of thebasin areas where ground motions are amplified when Zx terms from the local velocity models are greaterthan the default terms calculated from the NGA-West2 ground motion models. The local seismic velocitymodels used in the 2018 NSHM are Los Angeles (Lee et al., 2014), the San Francisco Bay Area (Aagaardet al., 2008), Salt Lake City (Magistrale et al., 2008), and Seattle (Stephenson, 2007; Stephenson et al., 2017).

Table 2. NGA-West2 default basin depths (km)

Site class VS30 (m/s)a ASK14 (Z1.0)b BSSA14 (Z1.0)

c CB14 (Z2.5)d CY14 (Z1.0)

e

A 2000 0 0 0.201 0A/B 1500 0 0.001 0.279 0.001B 1080 0.005 0.005 0.406 0.005B/C 760 0.048 0.041 0.607 0.041C 530 0.213 0.194 0.917 0.194C/D 365 0.401 0.397 1.4 0.4D 260 0.475 0.486 2.07 0.485D/E 185 0.497 0.513 3.06 0.513E 150 0.502 0.519 3.88 0.519

NGA: Next Generation Attenuation.aVS30 is time-averaged shear-wave velocity in the upper 30 m of the crust. Values used to define VS30 from Shumway et al.

(2018).bASK14 (Abrahamson et al., 2014).cBSSA14 (Boore et al., 2014).dCB14 (Campbell and Bozorgnia, 2014).eCY14 (Chiou and Youngs, 2014).


GMM developers included this predictor variable because it explained the average trends inground motions from the NGA-West2 database with basin depth, particularly at longer peri-ods. Figure 8 shows the basin amplification and deamplification factors at 5-s SA as a func-tion of velocity horizon (Z1.0 or Z2.5) for each of the NGA-West2 GMMs, given a VS30. Thehorizontal lines give the default basin depths for NEHRP site class boundary B/C(VS30 = 760 m/s; dashed black line) and NEHRP site class D (VS30 = 260 m/s; dashed grayline). For the GMMs using Z1.0, there is no modification to ground motion at the defaultbasin depth. If the basin depth is greater than the default, the basin factors amplify theground motion and if the basin depth is less, they deamplify the ground motion. For CB14,basin effects are not purely dependent on VS30, and for default depths between 1 and 3 km,the CB14 amplification factor is 1.0.

Many scientists at the March workshop supported the use of Z1.0 and Z2.5 values derivedfrom local 3D seismic velocity models in densely populated regions in the update of theNSHM. However, wholesale use of these factors results in either amplification or

Figure 8. Five-second amplification factors for NGA-West2 ground motion models. ASK14 (Abrahamsonet al., 2014), CY14 (Chiou and Youngs, 2014), and BSSA14 (Boore et al., 2014) use Z1.0, and CB14(Campbell and Bozorgnia, 2014) use Z2.5 to calculate default basin depths. Note that for the first three, ifZ1.0 is equal to the default value, the amplification factor is 1. For CB14, the amplification factor is 1, for Z2.5

between 1- and 3-km depth.

Petersen et al. 17

deamplification of ground motions, depending on whether the local model indicated basindepths deeper or shallower, respectively, than the default basin depths conditioned on VS30.This is consistent with how the GMMs were developed. Other participants preferred that weonly apply the basin models where they produce amplification of ground motions (deamplifi-cation effects excluded) at long periods. For shallower sites, the default basin depths wouldbe used, which effectively ‘‘turns off’’ the basin models.

For this 2018 NSHM update, we apply basin factors at sites over deep basins for periodsø 1 s. We define deep basins using the following two criteria: (1) Z2.5 ø 3.0 km when apply-ing CB14 with a depth taper between 1.0 and 3.0 km and (2) Z1.0 ø 0.5 km when applyingASK14, BSSA14, and CY14 with a depth taper between 0.3 and 0.5 km. The Z1.0 deep basindefinition is based on regression between Z2.5 and Z1.0 depth from the NGA-West2 database(described below). For Z1.0 depths less than 0.3 km and Z2.5 less than 1.0 km, we applydefault ground motions. This application primarily amplifies ground motions except at softsoil sites where the default basin depth may be deeper than that of the velocity model, result-ing in slight deamplification of ground motions over and just below the depth taper range.To maintain a smooth spectral shape, we also taper basin effects at 0.75-s SA via log-periodinterpolation.

We only consider basin effects over the deepest portion of the sedimentary basins for this2018 update. The primary consideration here is that there have been numerous observationsthat the edges of sedimentary basins, which are shallow, can focus S-waves and producebasin-edge generated surface waves. These increased ground shaking levels observed in thedata and simulations are not well understood and are arguably not parameterized by simpledepth constraints applied in the NGA-West2 basin factors. In addition, there are severalexamples of amplified shaking near the basin edge. For example, the Santa Monica area nearthe northern edge of the Los Angeles basin had amplification and enhanced damage duringthe 1994 M6.7 Northridge earthquake, caused by S-wave focusing and/or basin-edge-generated surface waves (Alex and Olsen, 1998; Davis et al., 2000; Graves et al., 1998). Theincreased ground shaking and damage from the Nisqually earthquake observed in westSeattle, located on shallow bedrock near the edge of the Seattle basin, were also attributed toS-wave focusing from the southern edge of the Seattle basin (Frankel et al., 2009; Stephensonet al., 2006). This basin-edge effect was also recognized in damage patterns of the 1995 Kobeearthquake (Kawase, 1996). In addition, preliminary findings of recent research using datafrom southern California basins indicates that deamplification effects in current basin modelsare mostly from broad, relatively shallow basin structures and not from the edges of deepbasins (Nweke et al., 2018). All of these factors can explain why basin-edge sites tend to havelarger aleatory variability than mid-basin sites and may require additional considerations inthe calculations of ground shaking.

Given all of these uncertainties, and the fact that directivity effects are also not explicitlymodeled in the NSHMs, which contribute to the large uncertainties, we felt it would not beprudent to lower ground motions with respect to defaults in areas of shallow basin edges thatare often characterized by complex shaking. Even though opinions are mixed about how toincorporate basin ground shaking, most workshop participants agreed that the amplificationsrelative to the default for sites that overlie the deepest portion of the basins are reliable.

Special considerations for Seattle

In 2013 and 2018, workshops of engineers and seismologists were convened by the city ofSeattle and the USGS to discuss incorporation of basin amplification effects into the Risk-Targeted Maximum Considered Earthquake (MCER) ground motions used in the design of


high-rise buildings in Seattle (Chang et al., 2014; Wirth et al., 2018a, 2018b). An importantrecommendation from the 2013 workshop was to use Z2.5 rather than Z1.0 to characterizebasin amplification in Seattle. Because much of Seattle is situated on stiff glacial sediments,for which the Z1.0 values are relatively shallow (often less than 500 m; Stephenson et al.,2017), the Z1.0-based GMMs predict no-to-weak amplification from the Seattle basin.However, these sites are observed to have amplification of factors of two to three relative torock sites with similar VS30 values south of the Seattle basin, for crustal and intraslab earth-quakes, and in this region, Z2.5-based models of basin amplification better reproduceobserved ground motion trends with basin depth (e.g. Chang et al., 2014; Frankel et al.,2009). Furthermore, the two workshop reports pointed out that the NGA-West2 data setincluded few strong-motion recordings from sites with conditions similar to what is observedin the Seattle basin (deep basins with VS30 values of 400–700 m/s), and the NGA-West2 mod-els of basin amplification were largely based on data from California and Japan, and did notconsider data from the Seattle region.

For the reasons discussed above, we preferred the use of the Z2.5 values, rather than Z1.0

values, from the Seattle basin to characterize the basin depths for ground motion amplifica-tions from crustal earthquakes. As we want to use all of the NGA-West2 equations, weneeded Z1.0 values that are compatible with the Z2.5 values of the Seattle basin. To this end,we calculated a Z1.0,eff parameter, which is the effective value obtained from Z2.5 using anequation similar to that derived by Campbell and Bozorgnia (2007). Use of the regressed Z1.0

values ensures consistency of basin amplifications between the Z1.0- and Z2.5-based GMMs.We regressed Z1.0 as a function of Z2.5 using two databases: (1) the NGA-West2 site database(Ancheta et al., 2013) and (2) basin depths from the most recent southern California seismicvelocity model (Lee et al., 2014). Use of the two databases represents an epistemic uncertaintyin computing the basin depth parameters for the Seattle basin. We only considered values forsites with appreciable sediment thicknesses (Z1.0 . 0.05 km and Z2.5 . 0.5 km). The ortho-gonal regression assumes a linear relation and minimizes the Euclidean distance between theline and all points, through the sum of squared distances. Equal weights were assigned toeach data point. Regression equations derived from the two databases are as follows:

Z1:0, eff = 0:1146Z2:5 + 0:2826 ð1Þ

Z1:0, eff = 0:0933Z2:5 + 0:1444 ð2Þ

where Z1.0,eff and Z2.5 are depths in kilometers. Standard deviations of the residuals (Z1.0,eff–Z1.0) are high from both of the two regressions to both databases, 0.41 and 0.17, respectively.These large uncertainties are likely due to regional variations in seismic velocity structure,uncertainties in the basin depth parameters, and differing methods for characterizing Z1.0

and Z2.5. We apply equal weight to the two regressions.The Z1.0,eff represents the effective depth derived from Z2.5, which we apply to the NGA-

West2 equations to calculate ground shaking for crustal earthquakes in the Seattle region.The Z1.0,eff derived from these equations are typically larger than the Z1.0 values derived fromthe velocity model. For example, the result of applying these models in the Seattle basin is toincrease the Z1.0 values by as much as 1 km compared to the Z1.0 values from the basin velo-city model.

Modification of subduction GMMs with basin terms for the Seattle region

We applied the CB14 basin term based on Z2.5 to modify the existing subduction GMMs usedin the 2014 NSHMs for earthquakes in the Cascadia subduction zone. We acknowledge that

Petersen et al. 19

there are substantial epistemic uncertainties with the application of this and with other NGA-West2 basin models. Decomposing the VS30 and basin-depth site terms from the subductionGMMs is not straightforward because, among other reasons, these equations are not basedon the same reference rock velocity definitions, and there are significant incompatibilitiesbetween VS30 terms in the different subduction GMMs. There are several alternative ways toaccount for basin amplification and deamplification effects in the NSHMs, and we consideredseveral methods. For the final hazard assessment, we ended up simply adding the basin depthterm from CB14 to the ground motions obtained from the VS30 amplified GMMs as done byChang et al. (2014). This is a simple method that allows for inclusion of the native VS30 termswhile still including basin effects using the CB14 GMM which does not depend directly onthe VS30 term. Generally, we only modify ground motions within the defined regional poly-gons and for the deepest portion of the basin. Consistent with the application of basin effectsin the crustal GMMs (described above), the CB14 basin term is included in the subductionGMMs only at periods ø 1 s (with a period taper at 0.75-s SA) and where basin depth values.3 km (with a depth taper between 1 and 3 km).

Recent 3D simulations for Cascadia M9 earthquakes show higher amplifications of long-period (1–10 s) motions in the Seattle basin (factors of 2–3) than predicted from the basinterms in Campbell and Bozorgnia (2014; factor of ;1.5) and Chiou and Youngs (2014; fac-tor of ;1.3) derived from crustal earthquake data (Figure 8; assuming Z1.0 = 0.9 km andZ2.5 = 6.7 km for a site in Seattle, WA). The higher amplifications of the simulations wereconsistent with observed amplifications from an earthquake located just below the plateinterface (Frankel et al., 2018). Therefore, it is justified to apply the higher amplification fac-tors in this region for subduction zone GMMs based on the CB14 terms, rather than usingthe lower factors from some of the other NGA-West2 relations.

Effects of considering basin effects on hazard

The hazard has primarily increased across the four basin areas compared to the default/aver-age model that is a function of VS30 only (Figure 9). These basin effect models, which are partof the NGA-West2 equations, increase ground motions between 0.75- and 10-s oscillator peri-ods. Ground motions are most influenced for the rock results (from 2000 m/s down to about760 m/s) at long periods, because the default basin depths are shallowest for rock VS30 values.Ground motions calculated for other site classes generally show increases of up to 60% atlong periods greater than about 1 s.

The 2% probability of exceedance in 50 years ground shaking for 5-s SA at a Seattle site(VS30 = 500 m/s) has a default ground motion of about 0.10 g and a basin-amplified motionof 0.15 g (50% increase). The ground shaking for 1-s SA at the same site increases by 39%.M9 simulation results are more than a factor of 2 higher than the NGA-West2 equations,which depend on basin depth, for periods greater than about 2 s (Frankel et al., 2018).

Similar ratios between basin-amplified and default ground motions for a VS30 of 260 m/sare also predicted in the deepest portions of the Los Angeles area basins (including theVentura Basin), San Francisco Bay area basins, and Salt Lake City area basins (Figure 9).Relative to the 2014 NSHM, ground shaking hazard for the regions near Los Angeles, SanFrancisco, and Salt Lake City have increased or remained the same due to the assumptionsdiscussed above. There are essentially no changes (compared to the 2014 NSHM) to groundmotion hazard in the city centers of Los Angeles and San Francisco, where basin depths areshallow. Other locations within their metropolitan areas have significantly increased groundshaking due to basin amplification (e.g. Long Beach and San Jose). Salt Lake City is in a


generally shallower basin, and only a small area to the northwest of Salt Lake City showsamplified ground motions.

Final seismic ground motion maps

In this section, we discuss seismic ground motion maps produced at a 2% probability ofexceedance in 50 years for PGA, 0.2- and 1-s SA on a firm-rock site condition

Figure 9. Ratio maps showing the difference in 5-s spectral acceleration total mean hazard for (a) Seattle,(b) Salt Lake City, (c) the San Francisco Bay area, and (d) Los Angeles when basin depths from local seismicvelocity models are used versus when default basin depths calculated from NGA-West2 ground motionmodels are used. The plots are for 2% probability of exceedance in 50 years and for NEHRP site class D(VS30 = 260 m/s).

Petersen et al. 21

(VS30 = 760 m/s), as well as 5-s SA produced on a stiff soil site condition (VS30 = 260 m/s).Seismic hazard maps for additional periods (i.e. PGA, 0.2 s, 1 s, 5 s for VS30 = 760 m/s andVS30 = 260 m/s) and probabilities of exceedance (5% and 10% in 50 years) are available atthe USGS ScienceBase Catalog (Rukstales and Petersen, 2019). Figure 10 shows meanhazard maps for (Figure 10a) 0.2 s at VS30 = 760 m/s and (Figure 10b) 5 s atVS30 = 260 m/s. Hazard is highest for regions that contain active sources capable of generat-ing future earthquakes (e.g. the San Andreas fault system and Cascadia subduction zone,and the Intermountain West seismic belt that is centered on the Wasatch and Teton faults)and regions that have experienced large earthquakes in the past (e.g. historical earthquakesin the New Madrid Seismic Zone (NMSZ) and Charleston, SC, occurred more than a cen-tury ago).

Figure 11 shows difference and ratio maps comparing 0.2-s SA ground shaking on a uni-form firm-rock site of VS30 = 760 m/s and illustrates where ground motions have changedfrom 2014 to 2018. The ground shaking has increased across a broad region within 1000 kmof the NMSZ (e.g. up 16% at Memphis, TN, 25% at Chicago, IL, and 30% at St. Louis,MO) mostly due to changes in the GMM. Decreases in seismic hazard are located in thenortheast United States (down 12% at New York City, NY), the Eastern Tennessee seismiczone, and across the Intermountain West which are mostly due to catalog changes. We findthat ground motions increased in areas where new earthquakes occurred (e.g. Delaware,Ohio, Kansas, Alabama, and the NMSZ in the CEUS, and Nevada and Idaho in the WUS)and decreased in areas where seismicity abated (e.g. northeast portion of the United States).Table 3 presents a summary of changes from 2014 to 2018 for PGA, and 0.2- and 1-s SA for10 sites across the United States. Hazard is up at Chicago, St. Louis, and Memphis at allthree periods, but hazard changes are larger at shorter periods (PGA and 0.2 s) than at lon-ger periods (1 s). Hazard is down at New York City at all three periods. Hazard is up atPGA and 0.2 s at Charleston, but down at 1 s. In addition, changes in hazard for 5-s SA(VS30 = 260 m/s) for five WUS sites are shown. Ground shaking at 5-s SA has increased infour WUS urban areas (Los Angeles, San Francisco Bay area, Seattle, and Salt Lake City)by up to 50% compared to the 2014 NSHM maps that used default basin depths. Theseincreases are due to the influence of deep sedimentary basin amplification of ground shaking.At short periods, the small changes in hazard (i.e. 1%–2%) seen at WUS sites can be contrib-uted to increased near-field discretization of gridded seismicity sources (see ‘‘ComputerCodes and Implementation Details’’ section below), and/or updates to the seismicity catalog(in sites outside California).

Figure 12a shows 2018 and 2014 NSHM 0.2 s total mean hazard curves (VS30 = 760 m/s), highlighting various customary exceedance levels, for the same sites as in Table 3. Hazardis highest for sites in Los Angeles, San Francisco, and Seattle in the WUS, and Charlestonand Memphis in the CEUS. Ground motions are an order of magnitude or more lower inChicago and New York City than those found in the more active regions. Figure 12b shows5 s total mean hazard curves (VS30 = 260 m/s) for ten sites in the WUS. At longer periodsand softer soil sites, the effect of basin amplification (relative to default basin depths) is evi-dent in Long Beach, Ventura, San Jose, Vallejo, and Seattle. At longer periods and softer siteconditions, the effect of the modification of the subduction GMMs can be seen at Seattle.

Additional periods and site classes

For the 2014 NSHM, we computed ground motions for three periods and one reference siteclass applied by the building code procedures (Petersen et al., 2014, 2015). However, we alsocalculated additional period and site class maps later for the WUS (11 periods and 8 site


(a)

(b)

Figure 10. Total mean hazard maps for 2% probability of exceedance in 50 years calculated for the 2018NSHM at: (a) 0.2-s spectral acceleration, NEHRP site class boundary B/C (VS30 = 760 m/s) and (b) 5-sspectral acceleration, NEHRP site class D (VS30 = 260 m/s), for the conterminous United States.

Petersen et al. 23

(a)

(b)

Figure 11. (a) Difference and (b) ratio maps showing comparisons between total mean hazard fromthe 2018 NSHM and the 2014 NSHM for the conterminous United States. Maps are for 0.2-s spectralacceleration, 2% probability of exceedance in 50 years, and NEHRP site class boundary B/C(VS30 = 760 m/s).


classes) and CEUS (7 periods and 2 site classes) (Shumway et al., 2018). Not all of the 2014CEUS GMMs contained factors for various periods and soils so we could only calculateground motions for limited periods and site classes. With the implementation of the 2018NGA-East GMMs and CEUS amplification model, we are now able to compute 22 periods(from 0.01 to 10 s) and 8 uniform VS30 maps (ranging from 1500 m/s to 150 m/s) for the con-terminous United States (Shumway et al., 2019).

Ground motions calculated for soils with VS30 less than ;200 m/s and greater than1500 m/s have high uncertainties and in some cases are extrapolated beyond the limits rec-ommended by some modelers. Two of the NGA-West2 models (BSSA14 and CB14) are validfor VS30 of 150 m/s but the other two (ASK14 and CY14) are only valid for higher VS30 val-ues of 180 m/s. Most of the developers felt that extrapolation down to 150 m/s was reason-able; however, concern was expressed that the ASK14 equation would benefit from furtherconsideration of nonlinearity effects of the soft soil sites. We compared each of the individualGMMs to assess the behavior of the NGA-West2 equations. Our comparison of the modelsfor M between 6.5 and 7.5 and at distances ranging from 10 to 50 km shows similar shakinglevels for ASK14 compared to the other models and especially the CB14 and BSS14 modelsthat were developed for VS30 down to 150 m/s. The NGA-West2 GMMs follow similar pat-terns and generally differ by about 50% or less at periods greater than 0.2 s, which is wellwithin the uncertainties of the data. We also compared the GMMs for hard-rock site condi-tions. The ASK14 GMM was not developed for VS30 higher than 1000 m/s but extrapolationto higher values is very similar to other models that were developed for the values up to1500 m/s. The developers felt that additional research should be conducted if we were torelease VS30 of 2000 m/s or greater maps and that in future versions of the models, it may beadvantageous to more completely consider the influence of lower kappa values and largershort-period ground shaking for these hard-rock sites (Abrahamson, 2019, written communi-cation). After our assessment of the extrapolations for higher or lower VS30 values, we felt

(a) (b)

Figure 12. (a) Comparison of total mean hazard curves for ten sites across the United States is shown forthe 2018 NSHM (solid lines) versus the 2014 NSHM (dashed lines). Hazard curves are plotted for 0.2-sspectral acceleration and NEHRP site class boundary B/C (VS30 = 760 m/s). (b). Comparison of total meanhazard curves for ten sites in the western United States are shown for the 2018 NSHM (solid lines) versusthe 2014 NSHM (dashed lines). Hazard curves are plotted for 5-s spectral acceleration and NEHRP siteclass D (VS30 = 260 m/s).

Petersen et al. 25

Tab

le3.

Com

par

ison

ofch

ange

sin

tota

lmea

nhaz

ard

(diff

eren

ces

and

ratios)

ofth

e2018

and

2014

NSH

Ms

for

PG

A,0

.2-,

1-,

and

5-s

SAat

2%

pro

bab

ility

of

exce

edan

cein

50

year

s

Site

Longi

tude

Latitu

de

PG

A,V

S30

=760

m/s

a0.2

s,V

S30

=760

m/s

a1

s,V

S30

=760

m/s

a5

s,V

S30

=260

m/s

b

Diff

eren

ce(R

atio

)D

iffer

ence

(Rat

io)

Diff

eren

ce(R

atio

)D

iffer

ence

(Rat

io)

Chic

ago,IL

–87.7

41.9

0.0

13

g(2

3%

)0.0

28

g(2

5%

)0.0

11

g(2

0%

)N

ASt

.Louis

,M

O–90.2

38.6

0.0

70

g(2

6%

)0.1

36

g(3

0%

)0.0

26

g(1

9%

)N

ewYo

rkC

ity,

NY

–74

40.8

–0.0

04

g(–

2%

)–0.0

34

g(–

12%

)–0.0

04g

(–8%

)C

har

lest

on,SC

–80

32.8

0.1

43

g(1

6%

)0.0

65

g(3

%)

–0.0

34g

(–7%

)M

emphi

s,T

N–90.1

35.2

0.1

29

g(1

9%

)0.1

90

g(1

6%

)0.0

14

g(4

%)

Los

Ange

les,

CA

–118.3

34.1

–0.0

18

g(–

2%

)–0.0

22

(–1%

)0.0

07

g(1

%)

0.0

00

g(0

%)

San

Fran

cisc

o,C

A–122.4

37.8

–0.0

08

g(–

1%

)–0.0

21

g(–

1%

)–0.0

03g

(0%

)0.0

00

g(0

%)

Port

land,O

R–122.7

45.5

–0.0

26

g(–

6%

)–0.0

63

g(–

7%

)–0.0

05g

(–1%

)–0.0

01

g(–

1%

)Se

attle,

WA

–122.3

47.6

–0.0

15

g(–

2%

)–0.0

31

g(–

2%

)0.1

75

g(4

3%

)0.0

79

g(5

0%

)Sa

ltLa

keC

ity,

UT

–111.9

40.8

–0.0

02

g(–

2%

)–0.0

07

g(0

%)

0.0

14

g(3

%)

–0.0

01

g(–

2%

)

NSH

M:U

SN

atio

nal

Seis

mic

Haz

ard

Model

;PG

A:p

eak

hori

zonta

lgr

ound

acce

lera

tion;SA

:spec

tral

acce

lera

tions

.a 2

%in

50

year

spro

bab

ility

ofex

ceed

ance

,N

atio

nal

Ear

thquak

eH

azar

ds

Red

uction

Pro

gram

site

clas

sboun

dar

yB/C

(VS30

=760

m/s

).b2%

in50

year

spro

bab

ility

ofex

ceed

ance

,N

atio

nal

Ear

thquak

eH

azar

ds

Red

uct

ion

Pro

gram

site

clas

sD

(VS30

=260

m/s

).


that in spite of the deficiencies, these models represent the best available science and thatreleasing them as well as short-period maps would be preferable to not having this informa-tion available. Nevertheless, while we are releasing VS30 maps between 150 m/s and 1500 m/sin this version of the NSHM (we have excluded softer soils and firmer rock classes), we rec-ommend: (1) caution in applying these end-member GMMs and (2) future improvements tothe hard rock and soft soil models be made.

With the availability of the new CEUS GMMs, Project 17 recommended transitioningfrom a simplified spectrum defined at two spectral periods (i.e. 0.2 and 1 s) to a more repre-sentative spectrum defined at multiple periods. They are also moving away from the use ofsite factor tables, toward using the GMMs to compute ground motions for soil sites directly(BSSC, 2019). The 2018 NSHM additional period and site class maps are currently beingconsidered by the building code update committees.

Figure 13 shows response spectra for additional periods and site classes at selected sitesacross the United States. WUS sites include Los Angeles, San Francisco, and Seattle. Thespectra at periods greater than about 0.2-s SA for different VS30 values are evenly spaced, andaccelerations increase with decreasing VS30. The CEUS sites include Charleston, Memphis,and New York City. The spectra for the CEUS sites look different than spectra for the WUSin that they (1) have a sharper peak at 0.1-s SA; (2) are unevenly spaced with varying VS30;and (3) exhibit decreased amplitudes between about 0.2- and 1-s oscillator periods caused bythe nonlinear effects (Hashash et al., 2019). The spectra at Charleston, SC, and Memphis,TN, have similar amplitudes at 0.1 s compared to the sites in the WUS.

Changes in earthquake shaking, exposure, and risk

In this section, we discuss the shaking hazard and exposure analyses performed using themost recent version of high-resolution LandScan population exposure data set produced byOak Ridge National Laboratory (LandScan� Database, 2017). In this, we consider the prob-abilistic PGA hazard curve at each site, assuming a reference site condition to be NEHRP siteclass boundary B/C, which refers to VS30 = 760 m/s, and convert it into a Modified MercalliIntensity (MMI, relates observed shaking to observed effects; Wood and Neumann, 1931)hazard curve using the Worden et al. (2012) intensity conversion relationship. Using log inter-polation, we develop MMI hazard maps for 50% (likely), 10% (infrequent), and 2% (rarebut possible) intensities with probabilities of exceedance in 50 years as shown in Figure 14.The 50% in 50-year (likely) hazard map may require consideration of the full rather thandeclustered catalog to account for the total strong ground shaking. This map may underesti-mate shaking in areas where large earthquakes with associated aftershocks occur. The MMIvalues are obtained after rounding to an integer value, for example, the numeric value ofMMI 5.50 and 6.49 are rounded to intensity level of VI. Analogous to the Jaiswal et al.(2015) study, we compared the MMI hazard and population exposure using the most perti-nent exposure data sets and provided an estimate of population exposure based on 1996,2002, 2008, 2014, and 2018 hazard models. Changes in population exposure reflect both thegeospatial change in population due to population growth as well as the spatial variation ofchanges in the estimates of seismic hazard, both of which have changed over time. Such anal-yses are not only helpful for discussing the changes in hazards (defined in terms of MMI,which are relatively easier to communicate) between different versions of hazard models butalso for understanding their impact in terms of change in the number of people (a proxy forbuilt environment) exposed to various levels of shaking hazards.

Petersen et al. 27

The analyses show that earthquake shaking, and its impact, is not limited to Californiansor the larger West Coast population but is spread over a much wider geographic region.Clearly, more Americans are at risk to damaging levels of earthquake shaking than everbefore (Table 4). As many as 34 million people (or 1 in 9) are expected to experience a stronglevel of shaking at least once in their lifetime. Considering the ground shaking intensities withaverage rate of 1/475 per year (i.e. 10% probability of exceedance in 50 years), there havebeen significant increases in the number of people exposed to MMI VIII and above, that is,from 28.0 million for the 2014 model to 32.2 million for the 2018 model. While the populationgrowth from 2013 to 2017 is modest (;3%), this increase of 14% in exposure is mainly attrib-uted to the relatively larger area, mainly in the CEUS, estimated to experience damagingground motions. Similarly, when considering low probability ground motions (2% probabilityof exceedance in 50 years), the exposure analysis shows one in three Americans may beexposed to MMI VII and above, which reflects an increase of ;10% in human exposurebetween the 2014 and 2018 hazard models. Such a level of increase is also evident for virtuallyall levels of shaking intensities, indicating that the hazard has increased considerably betweenthe two versions of the maps.

Figure 13. Uniform hazard response spectra are plotted for six sites in the United States for a 2%probability of exceedance in 50 years. Sites are calculated for NEHRP site classes A (VS30 = 2000 m/s), B(VS30 = 1080 m/s), C (VS30 = 530 m/s), D (VS30 = 260 m/s), and E (VS30 = 150 m/s) and site classboundaries A/B (VS30 = 1500 m/s), B/C (VS30 = 760 m/s), C/D (VS30 = 365 m/s), and D/E (VS30 = 185 m/s).


Figure 14. The 2018 Modified Mercalli Intensity (MMI) hazard maps of the conterminous United Statesshowing estimates of earthquake shaking in terms of (a) 50% probability of exceedance (PE) in 50 years(likely), (b) 10% PE in 50 years (infrequent), and (c) 2% PE in 50 years (rare).

Petersen et al. 29

Computer codes and implementation details

For past NSHMs, we calculated hazard using a computer code written in Fortran. For the2014 NSHM, the California model (UCERF3; Field et al., 2014; Powers and Field, 2015) andhazard calculations were implemented in OpenSHA (Field et al., 2003). Following publica-tion of the 2014 NSHM, we developed nshmp-haz (Powers, 2017), a Java platform derivedfrom OpenSHA, that uses a well-defined XML source model format. This developmentallows us to support hazard and related calculations (e.g. deaggregation) for both current andpast NSHMs with a single codebase and makes it easier to develop and deploy web servicesthat support our online web application, the Unified Hazard Tool (UHT; https://earthqua-ke.usgs.gov/hazards/interactive). The UHT consolidates legacy web applications that wereusually developed on a per-NSHM basis into a single application supporting multiple modeleditions. Key modifications and improvement to nshmp-haz to support the 2018 NSHMinclude the following:

� For the CEUS, implemented (1) the NGA-East for USGS median, hard-rock GMMand associated aleatory variability logic tree (Goulet et al., 2017; Stewart et al., 2019a)and (2) the CEUS ergodic site amplification model (Hashash et al., 2019; Stewartet al., 2019b) leading to support for 22 spectral periods and site classes spanning150 < VS30 < 3000 m/s. We smoothed the impedance model term and associateduncertainty at 0.1 s via interpolation.

Table 4. Estimated counts (in thousands) of population exposurea at various Modified Mercalli Intensity(MMI) shaking thresholds during different national probabilistic seismic hazard map cycles

Map year Total population estimates at MMI

øV(Moderate shaking)

øVI(Strong shaking)

øVII (Verystrong shaking)

øVIII(Severe shaking)

øIX(Violent shaking)

50% PE in 50 years (~72 years return period, Poisson probability converted to annual rate of exceedance)1996 41,844 31,945 19,827 1935 192002 45,375 35,313 21,103 1388 –2008 45,045 34,001 16,490 97 –2014 43,376 32,294 9198 93 –2018 45,806 34,461 11,946 119 –

10% PE in 50 years (~475 years return period)1996 104,715 54,268 39,642 26,381 10,5062002 104,591 59,463 44,109 29,967 93362008 82,286 56,632 45,240 27,922 50642014 97,597 62,685 47,259 28,061 39692018 107,580 65,460 48,324 31,357 3878

2% PE in 50 years (~2475 years return period)1996 235,236 164,769 95,654 45,393 30,0512002 247,753 169,004 95,179 49,208 33,0522008 241,898 155,909 84,193 49,558 32,0212014 242,069 163,748 97,364 52,632 34,7612018 267,586 183,411 106,368 57,520 38,481

PE: probability of exceedance.aPopulation counts are estimates based on LandScan data sets for the conterminous United States and are rounded up to

the nearest 1000. The total population counts for the conterminous United States according to LandScan data sets are:

LandScan 1998, 268,411,000; LandScan 2003, 288,883,000; LandScan 2009, 304,937,000; LandScan 2013, 313,940,000; and

LandScan 2017, 323,535,000.


� For the CEUS, eliminated the clamping of median and truncation of high groundmotions introduced in previous versions of the NSHMs.

� Updated WUS GMMs to support 22 spectral periods via interpolation and reasonableextrapolation.

� Updated NGA-West2 GMMs to support basin effects at spectral periods ø 1 s andmodified WUS Subduction GMMs to use the CB14 basin term at spectral periodsø 1 s, when considering basin depths within the defined basin polygons.

� Updated the 2012 BC Hydro GMM (Addo et al., 2012; used in 2014 NSHM) to beconsistent with Abrahamson et al. (2016).

� Added support for multi-branch GMMs. Multi-branch GMMs are composed ofunderlying logic trees of median ground motion and aleatory variability (sigma) mod-els. Such GMMs include NGA-East for USGS GMM, the Updated Seed GMM, andthe NGA-West2 GMMs with additional epistemic uncertainty (Rezaeian et al., 2015).Multi-branch GMM support also includes correct calculation of deaggregation, andhazard from cluster models (e.g. in the NMSZ and along the Wasatch Fault in thevicinity of Salt Lake City).

� Added deaggregation by period-specific intensity measure level to support epsilon cap-ping as required by the BSSC and Project 17 (BSSC, 2019).

� Increased near-field discretization of gridded seismicity sources, which removes arti-facts in hazard that result from the relatively coarse (0.1�) distribution of grid sources.

� Updated the southern border of the US hazard map region to include Key West, FL.

Nshmp-haz is available publicly on GitHub (https://github.com/usgs/nshmp-haz) alongwith the 2018 NSHM (https://github.com/usgs/nshm-cous-2018). The version of the codeused for these hazard calculations is v1.2.0 (https://github.com/usgs/nshmp-haz/releases/tag/v1.2.0). The version of the source model used for these hazard calculations is v5.0.0 (https://github.com/usgs/nshm-cous-2018/releases/tag/v5.0.0). Bug fixes and other modifications willincrement these values. Older source models (i.e. 2014 NSHM and the 2008 NSHM) are alsoavailable on GitHub and can be run with the new code. Converting previously releasedsource models (e.g. Guam and the Northern Mariana Islands) to work with nshmp-haz,which will make future updates to these models more straightforward. The 2018 NSHMrelease will be version 5.0.0 (bug fixes and other modifications will increment this value) andwill be deployed to the UHT shortly after publication of the model.

Conclusion

We have updated the NSHM for the conterminous United States by applying an updatedearthquake catalog, new GMMs for the CEUS, basin terms for long periods in the WUS,and by updating our seismic hazard computer code. We have calculated the hazard for sev-eral additional periods and site classes across the entire WUS than in previous models. Theground motions are mostly higher across the CEUS due to new GMMs having higher sigmaand amplification at periods less than about 2 s. In the WUS, long-period shaking for soils ishigher over the deepest portions of the sedimentary basins. This article summarizes the data,methods, and models applied in the 2018 NSHM update. Additional period and site classmaps, hazard curves, and methods to calculate deaggregations are available at the USGSScienceBase Catalog (Rukstales and Petersen, 2019; Shumway et al., 2019) and on GitHub(https://github.com/usgs/nshmp-haz).

This update required several different types of inputs: (1) introduction of alternative mod-els that were widely agreed on in the workshop and review process (e.g. CEUS GMMs and

Petersen et al. 31

basin amplification models), (2) development of models that were required for implementa-tion of multi-period and site classes but that did not exist (e.g. subduction zone site amplifi-cation models), and (3) changes that were implemented in one way but that could have beenimplemented in a logic tree containing alternative models (CEUS shallow soil amplificationfactors). As the quality assurance process continues and additional improvements are made,subsequent NSHMs will be revised to bring in new science and more robust and efficientmethods and data for calculating the hazard.

While the models we have included are reasonable and consistent with the available data,future updates and follow-on studies may consider important improvements that were notmade because either they were not vetted at the time of this publication or due to time con-straints that limited the content of this update. Updates that we are considering for futureNSHMs include reevaluation of the acceptance criteria of GMMs based on period and siteclass and associated weighting schemes for assessing epistemic (Sammon’s map) and fullyergodic and non-ergodic aleatory uncertainties, more complete calculation of uncertainties(fractiles) in the hazard models, assessment of earthquake magnitude-frequency distributionson faults and especially those with unknown or poorly defined slip rate data, regionalizationof input models and implementation of non-ergodic inputs, appraisal of time-dependent fore-casts of large earthquakes on faults, improvements of fault slip rate information and the asso-ciated joint inversion of geologic and geodetic models, development and implementation ofphysics-based source and ground motion computer simulation models, consideration of alter-native decluster models (e.g. for induced and swarm-like earthquakes), improvements of shal-low site amplification models (especially in the CEUS), enhancements of sedimentary basinmodels and better models for anticipating long-period basin effects over shallow basins andbasin edges, simplification of complex model inputs (e.g. UCERF3), and generation of effi-cient and effective computer algorithms and hazard outputs. We welcome discussions of theinputs and methodologies that will improve future versions of the NSHM.

Acknowledgment

Many scientists and organizations (e.g. Pacific Earthquake Engineering Research Center (PEER) andSouthern California Earthquake Center (SCEC)) contributed to the development of the local velocitymodels, earthquake catalogs, seismicity models, ground motion models (GMMs), standard deviation(sigma) models, and soil amplification models. We thank the many people who significantly improvedthe manuscript through reviews and discussions including Jill McCarthy, Bruce Worden, Grace Parker,Bill Stephenson, Gail Atkinson, Jon Ake, Janet Slate, working groups that produced important inputsthat are acknowledged within the paper, participants at the 2018 US National Seismic Hazard Model(NSHM) Update workshop, the NSHM Steering Committee, US Geological Survey editors andreviewers, Earthquake Spectra editors and reviewers, and the public who participated in the open com-ment period. Jon Stewart and an anonymous reviewer both made important comments in theEarthquake Spectra review that substantially improved the manuscript. Any use of trade, firm, or prod-uct names is for descriptive purposes only and does not imply endorsement by the US Government.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/orpublication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/orpublication of this article: The U.S. Geological Survey’s Earthquake Hazards Program provided fund-ing for this research and publication.


Notes

1. Aleatory sigma working group composed of experts Jonathan P Stewart, Grace A Parker, Linda AlAtik, Gail M Atkinson, and Christine A Goulet.

2. Working group on amplification models for central and eastern United States composed of expertsJonathan P Stewart, Youssef MA Hashash, Gail M Atkinson, David M Boore, Robert B Darragh,and Walter J Silva.

References

Aagaard BT, Brocher TM, Dolenc D, et al. (2008) Ground-motion modeling of the 1906 San Francisco

earthquake, part II: Ground-motion estimates for the 1906 earthquake and scenario events. Bulletin

of the Seismological Society of America 98: 1012–1046.Abrahamson NA, Gregor N and Addo K (2016) BC Hydro ground motion prediction equations for

subduction earthquakes. Earthquake Spectra 32(1): 23–44.Abrahamson NA, Silva WJ and Kamai R (2014) Summary of the ASK14 ground motion relation for

active crustal regions. Earthquake Spectra 30(3): 1025–1055.Addo K, Abrahamson N and Youngs R (2012) Probabilistic seismic hazard analysis (PSHA) model,

volume 3: Ground motion characterization (GMC) model. Engineering report no. E658, Vol. 3.

Vancouver, BC, Canada: BC Hydro.Al Atik L (2015) NGA-East: Ground-motion standard deviation models for central and eastern North

America. Report no. 2015/07. Berkeley, CA: PEER, 217 pp.Alex CM and Olsen KB (1998) Lens-effect in Santa Monica? Geophysical Research Letters 25:

3441–3444.Algermissen ST and Perkins DM (1976) A probabilistic estimate of the maximum acceleration in rock in

the contiguous United States. Open-File report no. 76-416, 2 plates, scale 1:7,500,000. Reston, VA:

US Geological Survey, 45 pp.Ancheta TD, Darragh RB, Stewart JP, et al. (2013) PEER NGA-West2 database. Report no. 2013/03.

Berkeley, CA: PEER, 134 pp.Atkinson GM and Boore DM (2003) Empirical ground-motion relations for subduction-zone

earthquakes and their application to Cascadia and other regions. Bulletin of the Seismological

Society of America 93(4): 1703–1729.Atkinson GM and Boore DM (2008) Erratum to empirical ground-motion relations for subduction

zone earthquakes and their application to Cascadia and other regions. Bulletin of the Seismological

Society of America 98(5): 2567–2569.Atkinson GM and Boore DM (2011) Modifications to existing ground-motion prediction equations in

light of new data. Bulletin of the Seismological Society of America 101(3): 1121–1135.Boore DM, Stewart JP, Seyhan E, et al. (2014) NGA-West2 equations for predicting PGA, PGV, and

5% damped PSA for shallow crustal earthquakes. Earthquake Spectra 30(3): 1057–1085.Boyd OS, Haller K, Luco N, et al. (2015) Seismic hazard in the nation’s breadbasket. Earthquake

Spectra 31(S1): S109–S130.Building Seismic Safety Council (BSSC) (2019) BSSC Project 17 Final Report: Development of Next

Generation of Seismic Design Value Maps for the 2020 NEHRP Provisions. Washington, DC:

National Institute of Building Sciences.Campbell KW and Bozorgnia Y (2007) Campbell-Bozorgnia NGA ground motion relations for the

geometric mean and horizontal component of peak and spectral ground motion parameters. Report no.

2007/02. Berkeley, CA: PEER, 240 pp.Campbell KW and Bozorgnia Y (2014) NGA-West2 ground motion model for the average horizontal

components of PGA, PGV, and 5% damped linear acceleration response spectra. Earthquake

Spectra 30(3): 1087–1115.CEUS–SSCn (2012) Central and Eastern United States Seismic Source Characterization for Nuclear

Facilities. Palo Alto, CA: EPRI, US DOE, and US NRC.

Petersen et al. 33

Chang SW, Frankel AD and Weaver CS (2014) Report on workshop to incorporate basin response in the

design of tall buildings in the Puget Sound region, Washington. Open-File report no. 2014-1196.

Reston, VA: US Geological Survey, 28 pp.Chiou BS-J and Youngs RR (2014) Update of the Chiou and Youngs NGA model for the average

horizontal component of peak ground motion and response spectra. Earthquake Spectra 30(3):

1117–1153.Cornell CA (1968) Engineering seismic risk analysis. Bulletin of the Seismological Society of America 58:

1583–1606.Davis PM, Rubinstein JL, Liu KH, et al. (2000) Northridge earthquake damage caused by geologic

focusing of seismic waves. Science 289(5485): 1746–1750.Dobson JE, Bright EA, Coleman PR, et al. (2000) LandScan: A global population database for

estimating populations at risk. Photogrammetric Engineering & Remote Sensing 66: 849–857.Electrical Power Research Institute (EPRI) (2013) EPRI (2004, 2006) Ground-Motion Model (GMM)

review project. Report no. 3002000717. Palo Alto, CA: EPRI, 67 pp.Field EH, Arrowsmith RJ, Biasi GP, et al. (2014) Uniform California earthquake rupture forecast,

version 3 (UCERF3)—The time-independent model. Bulletin of the Seismological Society of America

104(3): 1122–1180.Field EH, Jordan TH and Cornell CA (2003) OpenSHA: A developing community-modeling

environment for seismic hazard analysis. Seismological Research Letters 74(4): 406–419.Frankel A (1995) Mapping seismic hazard in the central and eastern United States. Seismological

Research Letters 66(4): 8–21.Frankel A, Chen R, Petersen M, et al. (2015) 2014 update of the Pacific Northwest portion of the U.S.

National Seismic Hazard Maps. Earthquake Spectra 31(S1): S131–S148.Frankel A, Mueller C, Barnhard T, et al. (1996) National seismic hazard maps—Documentation June

1996. Open-File report no. 96-532. Reston, VA: US Geological Survey, 110 pp.Frankel A, Stephenson W and Carver D (2009) Sedimentary basin effects in Seattle, Washington:

Ground-motion observations and 3D simulations. Bulletin of the Seismological Society of America

99: 1579–1611.Frankel A, Wirth E, Marafi N, et al. (2018) Broadband synthetic seismograms for magnitude 9

earthquakes on the Cascadia megathrust based on 3D simulations and stochastic synthetics, Part 1:

Methodology and overall results. Bulletin of the Seismological Society of America 108(5A): 2347–2369.

Frankel AD, Petersen MD, Mueller CS, et al. (2002) Documentation for the 2002 update of the national

seismic hazard maps. Open-File report no. 02-420. Reston, VA: US Geological Survey, 33 pp.Gardner JK and Knopoff L (1974) Is the sequence of earthquakes in southern California, with

aftershocks removed, Poissonian? Bulletin of the Seismological Society of America 64(5): 1363–1367.

Goulet CA, Bozorgnia Y, Abrahamson N, et al. (2018) Central and eastern North America ground—

motion characterization—NGA-East final report. Report no. 2018/08. Berkeley, CA: PEER, 817 pp.Goulet CA, Bozorgnia Y, Kuehn N, et al. (2017) NGA-East ground-motion models for the US Geological

Survey National Seismic Hazard Maps. Report no. 2017/03. Berkeley, CA: PEER, 207 pp., and 12

pp. addendum.Graizer V (2016) Ground-motion prediction equations for central and eastern North America. Bulletin

of the Seismological Society of America 106: 1600–1612.Graizer V (2017) Alternative (G-16v2) ground-motion prediction equations for central and eastern

North America. Bulletin of the Seismological Society of America 107(2): 869–886.Graves R, Jordan TH, Callaghan S, et al. (2011) Cybershake: A physics-based seismic hazard model for

southern California. Pure and Applied Geophysics 168(3–4): 367–381.Graves RW, Pitarka A and Sommerville PG (1998) Ground-motion amplification in the Santa Monica

area: Effects of shallow basin structure. Bulletin of the Seismological Society of America 88:

1224–1242.Haller KM, Moschetti MP, Mueller CS, et al. (2015) Seismic hazard in the Intermountain West.

Earthquake Spectra 31(S1): S149–S176.Hashash YMA, Ilhan O, Harmon JA, et al. (2019) Nonlinear site amplification model for ergodic

seismic hazard analysis in central and eastern North America. Earthquake Spectra. DOI: 10.1177/

8755293019878193


Hashash YMA, Kottke AR, Stewart JP, et al. (2014) Reference rock site condition for central and

eastern North America. Bulletin of the Seismological Society of America 104(2): 684–701.Helmstetter A, Kagan YY and Jackson DD (2007) High-resolution time-independent grid-based

forecast for M ø 5 earthquakes in California. Seismological Research Letters 78(1): 78–86.Idriss IM (2014) An NGA-West2 empirical model for estimating the horizontal spectral values

generated by shallow crustal earthquakes. Earthquake Spectra 30(3): 1155–1177.Jaiswal K, Bausch D, Rozelle J, et al. (2017) Hazus� estimated annualized earthquake losses for the

United States. Report no. P-366. Washington, DC: Federal Emergency Management Agency, 78 pp.Jaiswal KS, Petersen MD, Rukstales K, et al. (2015) Earthquake shaking hazard estimates and exposure

changes in the conterminous United States. Earthquake Spectra 31(S1): S201–S220.Kawase H (1996) The cause of damage belt in Kobe: ‘‘The basin-edge effect,’’ constructive interference

of the direct S-wave with the basin-induced diffracted/Rayleigh waves. Seismological Research

Letters 67(5): 25–34.LandScan� Database (2017) High resolution global population data set [LS1998, LS2003, LS2009,

LS2013, LS2017]. UT-Battelle, LLC, at Oak Ridge National Laboratory. Available at: http://

landscan.oml.gov/ (accessed 18 November 2019).Lee E-J, Chen P, Jordan TH, et al. (2014) Full-3-D tomography for crustal structure in southern

California based on the scattering-integral and the adjoint-wavefield methods. Journal of Geophysical

Research 119: 6421–6451.McNamara DE, Petersen MD, Thompson EM, et al. (2018) Evaluation of ground motion models for

USGS seismic hazard forecasts: Induced and tectonic earthquakes in the Central and Eastern U.S.

Bulletin of the Seismological Society of America 109(1): 322–335.Magistrale H, Olsen KB and Pechmann JC (2008) Construction and verification of a Wasatch Front

community velocity model. Technical report no. HQGR.060012. Reston, VA: US Geological Survey.Moschetti MP (2015) A long-term earthquake rate model for the central and eastern United States from

smoothed seismicity. Bulletin of the Seismological Society of America 105(6): 2928–2941.Moschetti MP, Hartzell S, Ramırez-Guzman L, et al. (2017) 3D Ground-motion simulations of Mw 7

earthquakes on the Salt Lake City segment of the Wasatch fault zone: Variability of long-period

(Tø 1 s) ground motions and sensitivity to kinematic rupture parameters. Bulletin of the

Seismological Society of America 107(4): 1704–1723.Moschetti MP, Powers PM, Petersen MD, et al. (2015) Seismic source characterization for the 2014

update of the U.S. National Seismic Hazard Model. Earthquake Spectra 31(S1): S31–S57.Mueller CS (2019) Earthquake catalogs for the USGS National Seismic Hazard Maps. Seismological

Research Letters 90(1): 251–261.Mueller CS, Boyd OS, Petersen MD, et al. (2015) Seismic hazard in the eastern United States.

Earthquake Spectra 31(S1): S85–S107.Nweke CC, Wang P, Brandenberg SJ, et al. (2018) Reconsidering basin effects in ergodic site response

models. UCLA. Available at: https://escholarship.org/uc/item/6048v74k (accessed 18 November

2019).Pacific Earthquake Engineering Research Center (PEER) (2015) NGA-East: Adjustments to median

ground-motion models for central and eastern North America. Report no. 2015/08. Berkeley, CA:

PEER, 129 pp.Parker GA, Stewart JP, Hashash YMA, et al. (2018) Empirical linear seismic site amplification in

central and eastern North America. Earthquake Spectra 35(2): 849–881.Petersen MD, Bryant WA, Cramer CH, et al. (1996) Probabilistic seismic hazard assessment for the

state of California. Open-File report no. 96-706. Reston, VA: US Geological Survey, 56 pp.Petersen MD, Frankel AD, Harmsen SC, et al. (2008) Documentation for the 2008 update of the United

States National Seismic Hazard Maps. Open-File report no. 2008-1128. Reston, VA: US Geological

Survey, 61 pp.Petersen MD, Moschetti MP, Powers PM, et al. (2014) Documentation for the 2014 update of the United

States National Seismic Hazard Maps. Open-File report no. 2014-1091. Reston, VA: US Geological

Survey, 243 pp.Petersen MD, Moschetti MP, Powers PM, et al. (2015) The 2014 United States National Seismic

Hazard Model. Earthquake Spectra 31(S1): S1–S30.

Petersen et al. 35

Petersen MD, Mueller CS, Moschetti MP, et al. (2016a) 2016 one-year seismic hazard forecast for the

central and eastern United States from induced and natural earthquakes. Open-File report no. 2016-

1035. Reston, VA: US Geological Survey, 52 pp.Petersen MD, Mueller CS, Moschetti MP, et al. (2016b) Seismic-hazard forecast for 2016 including

induced and natural earthquakes in the central and eastern United States. Seismological Research

Letters 87(6): 1327–1341.Petersen MD, Mueller CS, Moschetti MP, et al. (2017) One-year seismic-hazard forecast for the central

and eastern United States from induced and natural earthquakes. Seismological Research Letters

88(3): 772–783.Petersen MD, Mueller CS, Moschetti MP, et al. (2018) 2018 one-year seismic hazard forecast for the

central and eastern United States from induced and natural earthquakes. Seismological Research

Letters 89(3): 1049–1061.Powers PM (2017) National Seismic Hazard Model Project computer code (nshmp-haz), software.

Available at: https://doi.org/10.5066/F7ZW1K31 (accessed 18 November 2019).Powers PM and Field EH (2015) The 2014 update to the National Seismic Hazard Model in California.

Earthquake Spectra 31(S1): S177–S200.Rezaeian S, Petersen MD and Moschetti MP (2015) Ground motion models used in the 2014 U.S.

National Seismic Hazard Maps. Earthquake Spectra 31(S1): S59–S84.Rukstales KS and Petersen MD (2019) Data release for 2018 update of the U.S. National Seismic Hazard

Model: Overview of model and implications. US Geological Survey data release. Available at: https://

doi.org/10.5066/P9WT5OVBShahjouei A and Pezeshk S (2016) Alternative hybrid empirical ground-motion model for central and

eastern North America using hybrid simulations and NGA-West2 models. Bulletin of the

Seismological Society of America 106(2): 734–754.Sherbaum F, Kuehn NM, Ohrnberger M, et al. (2010) Exploring the proximity of ground-motion

models using high-dimensional visualization techniques. Earthquake Spectra 26(4): 1117–1138.Shumway AM, Clayton BS and Rukstales KS (2019) Data release for additional period and site class

maps for the 2018 National Seismic Hazard Model for the conterminous United States. US

Geological Survey data release. Available at: https://doi.org/10.5066/P9RQMREV (accessed 18

November 2019).Shumway AM, Petersen MD, Powers PM, et al. (2018) Additional period and site class maps for the 2014

National Seismic Hazard Model for the conterminous United States. Open-File report no. 2018-1111.

Reston, VA: US Geological Survey, 46 pp.Stephenson WJ (2007) Velocity and density models incorporating the Cascadia subduction zone for 3D

earthquake ground motion simulations. Open-File report no. 2007-1348. Reston, VA: US Geological

Survey, 24 pp.Stephenson WJ, Frankel AD, Odum JK, et al. (2006) Toward resolving an earthquake ground motion

mystery in West Seattle, Washington State: Shallow seismic focusing may cause anomalous chimney

damage. Geophysical Research Letters 33: L06316.Stephenson WJ, Reitmann NG and Angster SJ (2017) P- and S-wave velocity models incorporating the

Cascadia subduction zone for 3D earthquake ground motion simulations—Update for OFR 2007-1348.

Open-File report no. 2017-1152. Reston, VA: US Geological Survey, 40 pp.Stewart JP, Parker GA, Al Atik L, et al. (2019a) Site-to-site standard deviation model for central and

eastern North America. UC E-Scholarship publication. Available at: https://escholarship.org/uc/

item/2sc5g220 (accessed 18 November 2019).Stewart JP, Parker GA, Atkinson GM, et al. (2019b) Ergodic site amplification model for central and

eastern North America. Earthquake Spectra. DOI: 10.1177/8755293019878185Wald DJ and Allen TI (2007) Topographic slope as a proxy for seismic site conditions and

amplification. Bulletin of the Seismological Society of America 97(5): 1379–1395.Werner MJ, Helmstetter A, Jackson DD, et al. (2011) High-resolution long-term and short-term

earthquake forecasts for California. Bulletin of the Seismological Society of America 101(4):

1630–1648.


Wirth EA, Chang SW and Frankel AD (2018a) 2018 report on incorporating sedimentary basin response

into the design of tall buildings in Seattle, Washington. Open-File report no. 2018-1149. Reston, VA:

US Geological Survey, 19 pp.Wirth EA, Frankel AD, Marafi N, et al. (2018b) Broadband synthetic seismograms for magnitude 9

earthquakes on the Cascadia megathrust based on 3D simulations and stochastic synthetics, Part 2:

Rupture parameters and variability. Bulletin of the Seismological Society of America 108(5A):

2370–2388.Wood HO and Neumann F (1931) Modified Mercalli Intensity scale of 1931. Bulletin of the

Seismological Society of America 2: 277–203.Worden CB, Gerstenberger MC, Rhoades DA, et al. (2012) Probabilistic relationships between ground

motion parameters and Modified Mercalli Intensity in California. Bulletin of the Seismological

Society of America 102(1): 204–221.

Petersen et al. 37

Documents

The 2018 update of the US National Seismic Hazard Model ......2018 NSHM update, comparisons with the 2014 NSHM, final products (e.g. maps, hazard curves, deaggregations), and general