Technical Updates: 2012, Third Quarter

Data Sets (Task A)

  • The database QA/QC is continuing. A conference call was held on September 6 2012 to clearly define the data processing assessment. As part of the QA/QC process, Prof. Cramer (CERI, University of Memphis) has assigned static identification numbers for earthquake events and stations. He has also provide Dr. Goulet (PEER) with all of the instrument-corrected, unfiltered time series for the independent evaluation of filtering bandwidth limits (e.g. corner frequencies).A meeting is scheduled for the end of November to make final decisions on the post-processing required before the release of the database.
  • Dr. Boore (USGS) started revisions of his program smc2psa_rot_gmrot for computing the measure of ground-motion intensity to be used in the ground-motion prediction equations (RotD50). The revision incorporates suggestions from Dr. Abrahamson (U.C. Berkeley/PG&E) to speed up the calculations and to produce more accurate measures of response spectra for short oscillator periods when the time spacing dt is greater than Tosc/10. This later revision makes use of an interpolation of the input time series that improves on the previsouly-assumed straightline interpolation; the new interpolation is consistent with the Nyquist frequency of 0.5/dt (in essence, the interpolation assumes a sinc function in the time domain).

Reference rock and site amplification models (Task B)

  • The Geotechnical Working Group continues to meet regularly, making progress on many parallel tasks (see 2012 Q2 summary for more details). The group is putting the final touch to their reports on reference rock and kappa for CENA (task B.1).
  • In parallel to this work, Dr. Boore is continuing his investigation of the conversion factors from very hard rock to NEHRP BC sites (Vs30=760 m/s) in CENA. He has been collecting measured shear-wave profiles for CENA sites for which Vs30~760 m/s, looking for estimates of kappa for surrogate sites in other parts of the world for which the velocity profiles are similar to those in CENA, and he is comparing the oft-used square-root-impedance method for computing site amplifications with transfer functions that account for all reverberations within the velocity model.

Regionalization (Task C)

Activities below from Dr. Mooney and his team (USGS):

  • Activities in July-September have been focused on: (1) continuing data compilation of field measurements of seismic crustal structure of the Central and Eastern US; (2) calculating average crustal models for P- and S-waves; (3) computing synthetic seismograms for the evaluation of regional differences in wave propagation.
  • A teleconference was held on July 17, and it was agreed that the main priority was to examine possible differences in the regionalization of crustal structure for the Central and Eastern USA (Figure 1). Any proposed regional differences in crustal models (Figures 2 and 3) must lead to significant differences in simulated ground motions. Thus, Dr. Mooney and the research team have begun to calculate synthetic seismograms to evaluate such differences (Figure 4).
  • Another in-person meeting took place on August 7, at the USGS in Menlo Park. The meeting was attended Drs. Abrahamson and Goulet, as well as the following USGS staff: Drs. Boore, Mooney, and Chulick, with a video and telephone hookup with Prof. Chapman (Virginia Tech.). Prior to the August 7 meeting, data compilations of published crustal models were prepared by Dr. Mooney’s USGS junior staff consisting of Neil Fenning, Alex Ferguson, John Brockman, Kealani Kitaura, Amy Radakovich and Chase Glennster. At this meeting, Dr. Mooney presented a comprehensive overview of the regional crustal models, the methodology for calculating average values of the crustal structure (for a group of crustal models), and examples of synthetic seismograms. Drs. Abrahamson, Goulet and Boore offered suggestions for future work, such as: (1) eliminating near-surface sediments with S-wave velocities less than 3.0 km/s from crustal models (to be consistent with the reference rock defined in task B.1) used for synthetic seismogram calculations; (2) calculating 10 hz and higher-frequency synthetics.

Figure 1. Tentative regionalization of crustal provinces for CENAs (exact boundaries not definitive). These four provinces are: the Eastern Coastal Plain, Appalachians, Central US, and Gulf Coast Province. The Eastern Coastal Plain is distinguished by its rifted continental structure and sedimentary accumulations; the Appalachians have a crustal thickness that is about 10 km greater than the Eastern Coastal Plain; The Central US consists of a Precambrian crystalline crust with a classic stable continental crustal structure; the Gulf Coast has unusually thick sediments (up to 15 km) and thin (15-20 km) crystalline basement. The research team is in the process of calculating average/representative seismic properties for each of these crustal provinces and comparing their seismic response.

Figure 2. Location map for crustal data used to make the initial average models for the four regions depicted in Figure 1. Key: NC-SC, North Carolina-South Carolina; NY-PA, New York-Pennsylvania; IL-IN, Illinois-Indiana; WGCo, Western Gulf Coast. Black dots are locations of deep crustal measurements of seismic velocity.

Figure 3. Shear-wave velocity depth function for the Illinois-Indiana region (Figure 2) determined at 2 km depth intervals. Four crustal models (N=4) were used to calculate this initial average model.

Figure 4. Synthetic seismogram for simple three layer crustal model: 6.0 km/s (13 km thick), 6.4 km/s (13 km thick) 7.0 km/s (13 km thick). Method is fk (frequency-wavenumber) integration of Lupei Zhu (code from his web site at St. Louis University). Plot is up-side-down from some similar displays: first arrival Pn phase is masked in red. High-amplitude wide angle reflections marked in green. Blue box indicates far-offset (greater than 300 km) where seismic phases are well separated. This result illustrates our success in calculating full-wavefield synthetic seismograms for this sub-Task. (Abscissa scale shows offset in km.)

Regionalization & Source/Path Studies (Tasks C and D)

  • Dr. Boatwright (USGS) has incorporated five new small earthquakes into his data set of moderate earthquakes in Northeastern North America. The spectrum from the shallowest earthquake is fit with a corner frequency of fc = 11.8 Hz and a Brune stress drop of 36 bars, while the spectrum of the deepest earthquake is fit with a corner frequency fc ≥ 35 Hz and a Brune stress drop ≥ 555 bars, which exceeds the stress drop of the 1989 Saguenay earthquake. This comparison reaffirms the dependence of stress drop of source depth already determined for these moderate earthquakes. Unfortunately, the small size of the earthquakes and the amplitude of the microseismic noise makes the determination of seismic moment uncertain at 10 s. Moreover, the attenuation cannot be resolved from the regional broadband recordings, so he has used the average attenuation Q = 410 f 0.5 that Boatwright and Seekins (2011) determined previously for the northeastern part of the Appalachian Province.
  • Dr. Atkinson (U. of Western Ontario) has submitted a draft report on the work she recently completed with Dr. Boore on a new set of CENA models for geometrical spreading and Q. Key features of this work is a new bilinear geometrical model with an initial slope of R^-1.3 and a frequency-dependent distance to transition to the R^-1 slope. The analysis allows for event-dependent Q. An example of the model for two CENA event is shown in Figure 5. The work was completed for Canadian and north-eastern US events; future work involve the verification of validity for central US events/sites.

Figure 5. Example plot showing the new Atkinson and Boore bilinear model fit for two CENA earthquakes.

Finite Fault Simulations (Task E)

  • Dr. Frankel (USGS) is further validating his method of making broadband synthetics for the CEUS by comparing spectral response values of the synthetics with observations from the Saguenay and Riviere-du-Loup earthquakes. The crustal model used to make the long-period synthetics was from Hartzell et al. (1994). The observed waveforms were supplied by the NGA-East project. The average slip velocity in the long-period calculation was determined using the 2.7 m/s slip velocity found from comparison with NGA West data (100 bar stress drop) and applying a factor to account for the stress drop for the CEUS earthquakes. A cross over frequency of 3 Hz was used in the matched filter to merge the low and high frequency seismograms, for both earthquakes. A stress drop of 500 bars was used for the Saguenay earthquake and 200 bars for the Riviere-du-Loup earthquake. For the Saguenay earthquake, the synthetics had similar spectral accelerations as the data for distances less than 90 km but underestimated most of the observed values from 90-200 km (Figure 6, Top). The lack of decay of the observed spectral accelerations with distance from 40-200 km is striking and is likely caused by waves critically reflected by the Moho. While the long-period synthetics contain this Moho reflection, they still underestimate its amplitude. The spectral accelerations of the synthetics for the Riviere-du-Loup earthquake were similar to those of the data (Figure 6, Bottom).

Figure 6. Comparison of simulated (synthetic) and recorded ground motions using Frankel’s model.

  • Dr. Frankel is continuing the validation and calibration of his method of making broadband synthetics for the CEUS by comparing spectral response values of the synthetics with observations from the Mineral, Virginia earthquake. He is making progress on the extension of the comparisons for larger distances for these two events. Dr. Frankel is also investigating the geometrical spreading to about 100 km distance by analyzing waveforms and spectra from this earthquake. A coda normalization method is being applied to remove site amplification from the S-wave spectral amplitudes.
  • Dr. Graves (USGS) has performed a suite of low frequency (f < 1 Hz) ground motion simulations for the 2011 Mw 5.74 Mineral, Virginia earthquake with the goal of calibrating a rock-site 1D crustal velocity and Q structure model for central and eastern US (CEUS).
    • The observed ground motions for this earthquake were simulated at 69 sites extending out to about 850 km from the epicenter. The 1D velocity model has been previously tested using recordings from the 2008 Mt. Carmel earthquake where it was found that high Q values (Qs=5000) are needed in the lower crust to match the observed distance decay. The source is represented as a point double couple (strike=26, dip=55, rake=108, Mo=4.60×1024 dyne-cm) at a depth of 6 km. The model does well at reproducing the median level of observed response pseudo-spectral acceleration (PSA) for most sites at periods 2 to 5 sec.
    • Modified Mercalli Intensity (MMI) estimates have been computed from the simulated ground motions using the ground motion-intensity conversion equations (GMICEs) of Atkinson and Kaka (2007; AK2007) and Dangkua and Cramer (2011; DC2001-ENA) for comparison against the observed “Did You Feel It” (DYFI) intensity estimates (Figure 7, Right). Given the bandwidth limitations of the simulations, conversion relations for 2 sec SA are used (Figure 7, Left). The MMI values obtained from DC2011-ENA are systematical higher than the AK2007 values for all distances and over predict the median trend of the observed values by roughly 1 MMI unit, whereas the AK2007 values provide a better match (slight under prediction between 50 – 250 km).

Figure 7. Left panel compares observed (circles) and simulated (crosses) spectral acceleration levels at 2 second period for the Mineral, VA earthquake. The solid line is the median rock-site prediction from Atkinson and Boore (2006) with one-sigma levels indicated by the dashed lines. Simulated values match the observations well. Right panel compares observed (DYFI) intensities with those derived from the simulations using the Atkinson and Kaka (2007) and Dangkua and Cramer (2011) conversion relations. The Atkinson and Kaka (2007) relation provides a better overall match to the observations.

  • Participants and modelers from the Simulations Working Group are continuing their work for the validation exercises of the SCEC broadband ground motion simulation codes. The initial work has been done using the Loma Prieta event to set-up the workflows with the existing methods on the platform (Graves and Pitarka, Olsen, Archuleta). The intent is that once the workflows and initial input files are set-up for all the methods, then the remaining validations can be run by the SCEC IT group or by independent researchers without detailed interaction with the model developers. Members of the Simulations Working Group and additional collaborators have weekly teleconference calls to discuss the ongoing work that benefits many entities including NGA-East.

Sigma (Standard Deviation) (Task J)

Uncertainty from simulated ground motions (J.1):

  • Dr. Spudich (USGS) is continuing previous work he has been doing in collaboration with Dr. Cirella (INGV). They have been investigating the cross-validation method of Custodio, Liu, and Archuleta (Geophys. Res. Let. , 32, L23312, 2005) to determine whether their technique is a useful way to quantify relative errors in a kinematic source model. In this work Custodio et al.  inverted ground motions observed during the 2004 Parkfield earthquake. From the actual recording of the earthquake, they drew 12 subsets of stations and inverted them to produce 12 rupture models. The scatter in the rupture models gives some evidence of the error in the models, but there remain two unanswered questions: first, how much would the scatter be reduced if the whole complement of stations were inverted, and second, even if they fit the data perfectly there would be nonuniqueness of models, and how can this nonuniqueness be quantified?
  • In a meeting with Drs. Goulet and Abrahamson in August, Dr. Spudich reviewed the results of Spudich and Cirella’s previous work and they proposed a redirection of future work to determine the correlation of kinematic rupture model parameters inferrable from observed large-earthquake ground motions. This work would complement parallel work inferring these same correlations from dynamic rupture calculations. Dr. Spudich spent most of the the month of September at the Istituto Nazionale di Geofisica e Vulcanologia (INGV) in Rome, Italy, working with Dr. Cirella on two topics. The first topic was the quantitative inclusion of error bounds in ground motion inversions to determine source rupture behavior. They have been working to estimate the Green’s function errors in the simulated annealing inversion algorithm, which looks like it will be possible and will enable them to conclusively accept or reject rupture models based on estimates of Green’s function errors. The second topic they are addressing is the question of determining the correlation between rupture model parameters, such as peak slip velocity and rupture velocity, in models of real earthquakes. They are determining ways to modify the inversion’s “cost” function to find rupture models given the degree of correlation between pairs of rupture parameters.