2009 PSHA Interactive Deagregation
Introduction
This is a revised version of the 2008 NSHMP PSHA Interactive Deaggregation web site. In this second release, the 2008–update source and attenuation models of the NSHMP (Petersen and others, 2008) are used with just one exception. For the New Madrid Seismic Zone (NMSZ), the deaggregation source model is set up for the “unclustered” event branches only. These unclustered New Madrid sources are given full weight (90% weight to the 500 year mean recurrence models; 10% weight to the 1000–year mean recurrence models) whereas in the 2008 NSHMP PSHA they are only given 50% weight. Clustered–source models receive the other 50% weight in 2008 NSHMP PSHA. This is a temporary difference. The interactive deaggregation will include the NMSZ clustered–source models when a few software checkups are completed.
Seismic–hazard deaggregations are available for the following spectral periods anywhere in the conterminous U.S: 0.0 s (PGA), 0.1 s., 0.2 s, 0.3 s, 0.5 s, 1.0 s, and 2.0 s. This is the same set of periods that has been available at the USGS interactive deaggregation web sites since 1996 (for sites in the conterminous United States).
In the western US, long–period seismic–hazard deaggregations at 3.0 s, 4.0 s, and 5.0 s are also available at this web site.
Other New Features in 2008
Vs30
Site condition is now selectable as the average shear–wave velocity in the top 30 meters, Vs30. Previous NSHMP interactive deaggregations were confined to the NEHRP BC–rock site condition. The 2008 deaggregation allows the user to select the site condition by inputting Vs30. For sites in the western US, Vs30 in the range 180 m/s to 1300 m/s may be input, and the Next Generation of Attenuation models (NGA) will base the response on the input Vs30. All users who try the soil–site deaggregation feature should read the soil–response parts of the NGA reports in Earthquake Spectra, v. 24, no. 1. Some of the NGA relations have more input parameters than the user can select. Default settings are used for the depth of rock having 1.0 km/s and 2.5 km/s Vs, respectively. The 1.0 km/s depth is defined in the software to a suggested value (Chiou and Youngs, 2008).
Z1.0 = e[28.5 – (3.82 / 8) ln(Vs308 + 378.88)]
units meters, except where Vs30 is 760 m/s, in which case z1.0= 40 m. The depth at which rock Vs30 is 2.5 km/s, or z2.5, is defined as 2.0 km. These rock depths cannot be adjusted by the user in this preliminary release. Also, the Chiou–Youngs relation has coefficient sensitivity to whether Vs30 is measured or inferred. We use the inferred–value coefficient at this web site, and this cannot be adjusted by the user. Some WUS source attenuation models have less up–to–date soil–response models than those of NGA. Where Cascadia subduction sources and/or deep intraplate (Benioff) sources are important to the hazard, soil effects are based on models that are defined with respect to NEHRP site classes rather than Vs30. This web site cannot be used to deaggregate seismic hazard at sites on soft soil with Vs30<180 m/s (NEHRP E and F site classes).
For sites in the central and eastern US, deaggregation corresponding to only two Vs30 site classes is available. These are the BC (760 m/s) and A (2000 m/s Vs30) site classes. BC is sometimes designated “firm rock” and A is designated “hard rock.” The attenuation models used to compute CEUS seismic hazard do not exhibit sensitivity to the z1 or z2.5 depths discussed earlier.
There is a transition region between east and west where BC is the only site class in common in the software used for deaggregation. This is a region that is roughly defined on the west by the eastern range front of the Rocky Mountains (see fig. 1), and a corresponding edge 200 km east of there. In New Mexico and west Texas, the western source region extends through the Rio Grande rift zone. Cities like El Paso, TX, and Albuquerque, NM, are mostly affected by WUS sources and soil–site Vs30 is explicitly modeled for those sources. The CEUS contribution at those cities, if any, will by computed using a BC–rock site condition, which is the only available model other than hard–rock in the NSHMP CEUS attenuation software.
Long–Period Analysis
For seismic engineering purposes, any ground–motion period above about 1 s is often considered long period, or LP. At WUS sites, deaggregations for the 3– to 5–s spectral acceleration are available for the first time at this USGS interactive deaggregation Web site. Here, WUS means sites west of the Rocky Mountain range front. Figure 1 shows the region where soil–site and LP SA analysis are possible. West of the wavy gray line or band in fig. 1, the deaggregated hazard will be predominantly from WUS sources, and in that region the analysis should be valid. Please determine that CEUS sources contribute negligibly to the hazard at your site before using the soil–site and/or long–period analysis provided at this Web site.
There are some limitations in the availability of longer–period attenuation models for Cascadia and deep–intraplate sources that prevent analysis at spectral periods longer than 5.0 s. Several CEUS attenuation models are limited to 2.0 seconds at the long–period end of the ground–motion spectrum.
Many More Return Times in 2008
Return times are now selected using two menu drop downs. One of these menu items is a probability of exceedance, or PE (in percent), and the other is the exposure time (number of years). Table 1 shows the range of these parameters. Any combination of entries from the two columns may be selected.
| Probability (%) | Number of Years |
|---|---|
| 1 | 10 |
| 2 | 30 |
| 5 | 50 |
| 10 | 75 |
| 20 | 100 |
| 50 | 200 |
Different building codes’ seismic provisions specify a variety of return times. We wish to offer sufficient flexibility to allow users to deaggregate seismic hazard for all of these. Let us know if your requirements are not covered by the above table.
Site Location
There are now two distinct ways to enter the site location, (1) as a latitude–longitude pair, or (2) as an address. You can toggle between these alternatives. You may now input your site’s coordinates with as much as four decimal–place accuracy (e.g., 34.1234, –111.2345). All of this site location information will be retained for the analysis but the location fields in output files may be rounded to 3 decimal places (e.g., 34.123, –111.235). Previously the input site–location coordinates were limited to 3 decimal places. Alternate ways to input the site include a city, state abbreviation, such as Boise, ID, with more address information if available. A map showing the vicinity of the site can be requested, and the red USGS icon on that map can be dragged and dropped to neighboring locations for further analysis.
2009 Addition
You can output a text file that has deaggregations associated with each ground–motion prediction equation as well as the deaggregation of the mean hazard. If you choose this option, the text file first shows the mean–hazard deaggregation, followed by all of the individual GMPE deaggregations. The plot, however, corresponds to the mean–hazard deaggregation only.
Probabilistic seismic–hazard calculations generally use many input models to capture uncertainty in both the source and the propagation of seismic waves to obtain the mean hazard. You can find out how each model used to model uncertainty in the propagation model (and some source properties) contributes to the mean hazard, by answering this question “Yes”. The mean hazard deaggregation will be followed by deaggregations for each of the ground motion prediction equations, or GMPEs.
Different GMPEs are used in different parts of the country. For example, there are eight GMPES in the eastern US 2008 NSHMP model, three in southern California, and six in the Pacific Northwest. In a narrow transition region between eastern and western USA, there can be more than eight GMPEs in the model. The names of researchers associated with each of the GMPEs are listed in the output file. For all sites in the western US, you will always see deaggregation tables that include the three Next Generation of Attenuation models that NSHMP used, namely Boore–Atkinson 2008, Campbell–Bozorgnia 2008, and Chiou–Youngs 2008.
If you answer “No”, only the mean hazard is deaggregated and output, as in previous editions of the National Seismic Hazard Mapping Project deaggregations.
By examining this detailed output file, you can find out which GMPEs tend to dominate the hazard, in the sense of producing the most ground–motion exceedances, and how these models vary in their estimates of the relative “threat” of various distance, magnitude combinations. All of these outputs are weighed by the subjectively determined GMPE model weights discussed in the 2008 seismic–hazard model documentation, OFR 2008–1128. The individual GMPEs are deaggregated at the same ground motion that produces the specified mean hazard probability of exceedance, for example, 2% in 50 years. The contributions in percent of the individual GMPEs should sum to 100.
The Output Files
In 2008, two output files are generated. These are a text file (.txt) which is a table of binned hazard contributions, with binning in magnitude, distance, and ground motion uncertainty, or ε, and a graph of these data in a format suitable for viewing. Both of these files correspond to one spectral period, specified by the user. The range of binned distances is the same as that of the 2008 PSHA model. These distances are 0 to 200 km for WUS crustal sources, 0 to 1000 km for Cascadia sources, and 0 to 1000 km for all CEUS sources. The distance metric displayed is Rcd (Rcd = closest distance to fault or hypocentral distance to point source) and units are km. Moment magnitude is displayed for all sources. The colors on the deaggregation plots are retained from the corresponding 2002 Web site. You should read the documentation at that Web site for more details on what the colors mean. Both text and graphic output prominently display important statistical summary information, such as mean and modal values of R,M and ε. Epsilon is binned in 1–σ increments. The gray bars on the front of the graph columns are placed at the bin edges, at 2 σ, 1 σ, 0 σ, … The uppermost colored band of each column represents the contribution from ground motions greater than 2 σ above the mean. The text table exhibits the binned hazard data as percent contributions. At the end of this file are lists showing two other binning schemes used to display or deaggregate source contributions. The first is a set of broad–scale bins, such as Cascadia megathrust source contributions and California a–faults. The second is a set of fine–scale bins, which correspond to individual fault contributions and epistemic and aleatory branches for those faults (Gutenberg Richter source distribution, characteristic events, dip uncertainty, and more). Fault dip uncertainty is a new feature for the 2008 PSHA update. The deaggregation output shows the contributions from the 3 dip–uncertainty branches, including those branch weights (60% weight to the 50–degree dip branch, and 20% to the 40– and 60–degree dip branches, respectively) for Basin and Range normal–slip faults. Dip uncertainty is not part of the hazard model for California faults, reverse–slip faults in Washington and Oregon, or steeply dipping faults (near 90 degrees) anywhere in the U.S.
Interesting and Important Details
Graphic output distance bins can be either 10 km or 25 km wide for nearby sources. The graphic output has broader distance bins for more distant sources. The largest distance bin (for CEUS sites only) corresponds to the annulus from 500 to 1000 km from the site. The text–file distance bins are 10 km wide. Magnitude binning is uniform, 0.2=dM. Epsilon bins are also uniform, with width 1 σ. Users should be aware that estimates of the modal event depend on binning details. Within a given bin, the mean source contribution is computed. The binned column position is always determined by the mean source distance and magnitude (M) for all sources within that bin. For example, if a nearby fault contributes 90% of the hazard in a bin, the plotted column center will be very close to the Rcd and M for that fault, and its color to that fault’s ε. This is the reason that the binned hazard columns do not plot along ramrod straight lines. Our choice of plotting scheme should help users quickly identify bins where significant fault hazard contributes to the seismic hazard at their site. Figure 2 illustrates the effect of centering the column at the mean hazard (R,M) within that bin. In fig. 2, which corresponds to a 10–hz seismic hazard deaggregation at McAffee stadium in Oakland, CA, the dominant hazard comes from the South Hayward fault (with 150–year rupture anniversary in 2009). The reason that the modal hazard plots at about 5 km distance is not because 5 km is the center of the 0–to–10 km bin. Rather, the reason is that the distance of the South Hayward fault to the stadium is about 5 km. Similarly, the San Andreas fault contribution is plotted at about 25 km distance because the fault is 25 km from the stadium, not because 25 is halfway between 20 and 30. Note from fig. 2, some considered fault–rupture scenarios are further from McAffee stadium, with Rcd of about 33 km. Figure 2 shows that no source with contribution greater than 0.05% has distance greater than about 34 km from the stadium, given the return time (2500 years), spectral frequency (10 hz) and geotechnical site condition (Vs30 of 280 m/s). This figure is for illustration purposes only and is not intended to be an actual hazard analysis.
This early release does not yet have a geographic deaggregation plot option. The geographic deaggregation allows the user to see the contributions from faults and other sources projected onto a 3–dimensional map. This option has been a popular tool in past releases, helpful for sharing PSHA hazard ideas with a wider audience. We will include geographic deaggregation in a later 2008 update.
The 2008 update is not yet ready to generate synthetic seismograms. We need to obtain or develop appropriate input files for soil sites, and our input files for rock sites are mostly restricted to the BC–rock site condition. With respect to rock–site seismogram generation, the 1996, 2002 and other earlier versions of USGS interactive deaggregations contained a synthetic seismogram option. We served up selected seismograms computed from a stochastic point–source model called smsim, developed by David Boore (USGS). The engineering community has asked for more variety and flexibility for generating or collecting appropriate seismograms for modal–event or other deaggregation sources. To date this is an open item.
A USGS seismic–hazard update for Alaska was completed in 2008 but the corresponding deaggregation web tool has not yet been put together for the updated Alaska PSHA.
Quality Assurance Checks
The 2008 source and attenuation models are more complex than those of previous editions. All software has been revised for the 2008 seismic–hazard model. Several cross–checks have been performed on codes by comparing output data at many spectral periods. Cross checks include computing the hazard from independently written software packages, such as Open SHA and some proprietary codes used for the California Earthquake Authority. Distance–computing algorithms for irregular–shaped faults have been revised and checked against known solutions.
Most of the cross checks have been performed for the BC–rock site condition. Some checks have been performed for D Soil and for California sites and source models. The relevant PSHA software has been distributed and critically studied by several users. Software QA is a never–ending process, and all concerns and questions should be brought to our attention.
Speed and Performance
The deaggregations at this web site must complete in a short time window (about 30 seconds) available to the application on the Web server computer. This stringent time limit requires a consolidation of the input files and mathematical calculations compared to those of the national seismic–hazard maps, and revision of the computer software to increase speed. Results based on these changes have been checked for consistency with the original model output. You can also examine the consistency of the deaggregation calculations with those of the national map (Petersen and others, 2008) by computing the deaggregation at “grid points” for the BC–rock site condition. The 2008 published value of SA is used at the sampled grid points of the national maps. If the computed rate of exceedance is within a few percent of the expected value (2.107*10–3 or 4.04*10–4), the deaggregation model is thereby shown to be consistent with the national map model. Sampled grid points start at 125° west longitude and 50° north latitude, and continue east and south every 0.05 degrees. Many other consistency checks have been performed. Please keep in mind (from the introductory remarks) that the NMSZ source model is simplified in this preliminary release, and consistency will not be achieved at sites where the clustered–source hazard is significantly different from the non–clustered source hazard.
For sites off the pre–computed map grid, and for all sites with Vs30 different from the initial values (these are 259 m/s, 537 m/s, 760 m/s, and 2000 m/s), the solution is iterated until the computed rate of exceedance is close to the requested value (e.g., 4.04*10–4 for the 2% in 50 year PE).
Some Caveats
For the most part, the Cascadia subduction and deep–interface soil response models are of an earlier vintage than the NGA soil–response models. The Cascadia subduction and deep–interface ground–motion prediction models have broad site categories, such as NEHRP site classes or even a “rock versus soil” dichotomy, rather than continuous variation with Vs30. Thus, you need to be aware that sites whose seismic hazard is controlled by these sources have less sophisticated models of soil nonlinearity and other site amplification features than those of the NGA models (which were designed to model only crustal–earthquake ground motion).
Because this Web site is accepting a much broader set of return times and site Vs30, there are many poorly or even untested combinations of input parameters. It is too early to say how well some of these will work out. Please proceed with patience and with a cautious or skeptical attitude.
The soil–amplification models that are used in the 2008 interactive deaggregations correspond to a generic soil column without significant resonance at any period. Some soil columns are known to exhibit band–limited amplification, with primary and secondary resonant periods, and it is possible that one or more of these could correspond to a building’s fundamental vibration mode, T1, or higher mode. In these instances, site–specific geotechnical analysis is needed to characterize expected soil behavior and possible soil–structure interaction.
At intermediate to long periods (T > 1 s), the effect of sedimentary basins on the seismic wave field and spectral acceleration is at best only broadly modeled by the NGA soil–amplification models. Many effects of source directivity and its interaction with basin geometry and other basin properties have been shown theoretically to produce strong surface waves that propagate far into deep sedimentary basins. Some of these effects have been documented at sites in the Los Angeles and San Fernando basins in Day and others (2008) and many related publications.
The bulk of soil–site strong–motion data that was used by NGA developers corresponds to relatively deep basins. Sites on shallow–sediment basins may experience less long–period soil amplification than is predicted by the generic model. One of the NGA models, that of Campbell and Bozorgnia, has a basin–depth term, but this is not currently varied in the beta version of the web site. Instead, we assume a uniform depth of 2 km for the basin.
Concluding Remarks
We welcome your questions and comments about this new web site. We sincerely hope it will help meet the growing demand for information about seismic–hazard issues in the United States.
References
- Campbell, K. and J. Bozorgnia, 2008. NGA ground motion model for the geometric mean horizontal component of PGA, PGV, PGD, and 5% damped linear elastic response spectra for periods ranging from 0.01 to 10 s, Earthquake Spectra, v 24. pp. 139–172.
- Chiou, B. and R. Youngs, 2008. An NGA model for the average horizontal component of peak ground motion and response spectra, Earthquake Spectra, v 24. pp. 173–216.
- Day, S., R. Graves, J. Bielak, D. Dreger, S. Larsen, K. Olsen, A. Pitarka, and L. Ramirez–Guzman (2008). Model for Basin Effects on Long–Period Response Spectra in Southern California. Eq. Spectra, v. 24, pp. 257–277.
- Petersen, M. and others, 2008. Documentation for the 2008 update of the national seismic hazard maps, USGS OFR 08–1128. Available on the web at http://pubs.usgs.gov/of/2008/1128/.
