Final Report, 2001

Analysis of Shallow Site Response to LARSE-2 Blasts at Precarious Rock Sites Near the San Andreas Fault

Robert E. Abbott

Seismological Laboratory (174), University of Nevada, Reno NV 89557-0141

rabbott@seismo.unr.edu

Principal Investigators:

John N. Louie, James N. Brune, and Rasool Anooshehpoor

Seismological Laboratory (174), University of Nevada, Reno NV 89557-0141

louie@seismo.unr.edu brune@seismo.unr.edu rasool@seismo.unr.edu

INTRODUCTION

The geographical distribution of fields of precariously balanced rocks is gaining increased acceptance as a ground motion constraint in seismic hazard models. Precarious rocks (defined as rocks with toppling accelerations of 0.1 to 0.3 g) and semiprecarious rocks (0.3 to 0.5 g) are located as close as 14 km to the Mojave section of the San Andreas fault [Brune, 1999]. The continued existence of such rocks in their current configuration for many thousands of years [Bell et al., 1998] is incompatible with most ground motion prediction models [Brune, 1996], given that the rocks have remained untoppled for many great earthquake cycles.

There are several theories to explain the discrepancy. Among them is that great earthquakes on the San Andreas Fault have "characteristic" spectra, repeated each earthquake cycle, with ground motions below the toppling acceleration of the rocks at the key frequencies [Brune, 1999]. Building upon this, Anderson and Brune [1999a] discuss a thought experiment created by limiting or eliminating the statistical variation of repeated large earthquakes.

Anderson and Brune [1999b] find that seismic hazard maps that limit earthquakes to mapped faults (i.e. no "area" sources) are compatible in most cases for precarious rocks in Nevada. Incomplete knowledge of active fault distribution, as evidenced by the Landers, Hector Mine, and Northridge earthquakes, suggests that this particular result of Anderson and Brune [1999b] would not be appropriate for all situations. It would be most appropriate for situations for areas where the hazard is dominated by a single, well characterized, fault, as is the case for the precarious rock sites in this study.

This project aims to test a third hypothesis. Namely, we aim to test whether or not de-amplification of ground motions at these sites relative to "representative" sites is sufficient to decrease the level of shaking to below the rocks’ toppling accelerations. It is currently not known if these precarious rocks experience the same ground motions as the surrounding regions they are located in, or if they define small "islands of stability" in a general ground motion field.

It is widely recognized that deep basin geometry can be a dominant factor in determining amplification at sites within sedimentary basins [e.g Gao et al., 1996, Hartzell et al., 1997; and Davis et al., 2000]. Subsurface focusing of seismic energy propagating through velocity contrasts at basin margins is thought a mechanism producing the amplifications. If this is true, than there must be areas of de-focusing and de-amplification.

Another mechanism by which amplitudes can be amplified is by conservation of energy. The energy in a seismic shear-wave is proportional to where r is density and b is shear-wave velocity (Aki and Richards, 1980). Generally, both density and shear-wave velocity decrease as one nears the surface (due to weathering, decreased overburden, etc.). Shear-wave velocity and density values at the precarious rock sites differing significantly from those assumed when calculating hazard (see Frankel et al., 1996, Appendix A) would skew ground motion predictions at those sites.

This project was funded to characterize the site effects at precarious rock sites to better understand the nature of the difference between the toppling accelerations of the rocks and the accelerations predicted by current probabilistic seismic hazard curves. We accomplish this by computing single- and multi-station spectral ratios of LARSE-II blasts and (fortuitously) Hector Mine aftershocks. We also measure the shallow subsurface shear-wave velocity at two of the precarious rock sites using seismic refraction and Louie’s [2001] refraction-microtremor techniques. We then make use of the technique of Boore and Brown [1998] to test how our results compare to those expected at the velocities used by Frankel et al. [1996] to compute the National Seismic Hazard Maps. We also compare our results to those generated by using the "generic rock" velocities from Boore and Joyner [1997].

Our results show, in general, that there is insufficient de-amplification at the precarious rock sites relative to the regional approximations used in the seismic hazard calculations and that site effects alone cannot explain the continued existence of the rocks. Indeed, our results indicate possible amplification above 4 Hz relative to assumptions used to create the national seismic hazard maps.

DATA

Earthquake Data

In October 1999, we deployed matched, calibrated, 3-componant digital stations with 1 Hz L4 seismometers at 6 sites near the San Andreas fault in preparation to record LARSE-II blasts (Figure 1, Table 1). Four of the sites (Lovejoy Buttes, Piute Butte, Alpine Butte, and near Black Butte) are located at or near fields of precarious or semi-precarious rocks. We also installed a station (Mill Creek Summit) near the USC strong motion site MCS, a site that is listed as being on NEHRP Engineering Class A rock [http://smdb.crustal.ucsb.edu/ows-bin/owa/summary3.main?SiteName=MCS].

An additional station was located on a rock site near the town of Llano, just a few kilometers from the fault. All of these sites except Llano were active the first two nights of LARSE-II blasting. Numerous aftershocks of the October 16, 1999, Hector Mine earthquake were recorded along with the blasts.

Figures 2 and 3 show sample seismograms from a blast and an aftershock. In general, the blast recordings had disappointing signal-to-noise ratios such that only seven blasts were deemed suitable for analysis. Curiously, the large blasts north of the San Andreas fault, closest to our array, were among those deemed unsuitable. Meanwhile, some smaller blasts much farther away showed improved S/N ratios. We feel this discrepancy is a function of the seismic coupling efficiency at each site rather than being caused by variable attenuation. Given that the Hector Mine aftershocks were more numerous and of much higher quality, they form the backbone of our analysis. We split the events into earthquakes and blasts prior to analysis, and we only present earthquake data here.

The parameters of the 61 earthquakes in the analysis are presented in Table 2. Local noise or multiple events sometimes contaminated the data at one or more locations, so Table 2 also lists which events were recorded for each site. The event magnitudes ranged from 1.8 to 5.1 M_L. We tried to add events that have different back azimuths to create, as much as possible, a path-independent amplification scheme. Due to our short occupation time at the sites, however, the sources are predominately near the Hector Mine rupture area.

Velocity Data

We also measured shallow shear-wave velocities at three of the sites for which we have 3-componant seismograms (PB, MCS, and LJB). The data was acquired using a combination of standard refraction and refraction-microtremor techniques. The methodology for the acquisition is given in Louie [2001]. Louie [2001] demonstrates that the resulting shear-wave velocity depth profiles (Figure 4) agree with borehole measurements to within 20%. Standard seismic refraction equipment was used with 12-24 8-Hz vertical sensors and spread lengths of 96 to 192 meters.

METHODS

Earthquake Data Preparation

Data recorded by the stations were segregated into event catalogs using the SCEC Data Center preliminary Hector Mine earthquake catalog and the LARSE-II blasting schedule. Each event was examined and only events with good signal-to-noise ratio on the majority of the stations was accepted for further analysis. S-wave arrivals were picked for all the events.

Spectra Calculation

We computed the S-wave spectra of each component for each record. The Fourier spectra were calculated for a 5 second time window, starting one-half second before S-wave arrival. This value was chosen to best contain most of the high amplitude direct S-wave energy. As pointed out in Bonilla et al. (1997) using longer times results in better spectral resolution at the cost of contaminating the spectra with scattered and reflected energy, as well as surface waves. However, Bonilla et al. (1997) and Field and Jacob (1995) find no statistical variations in site response computed with spectra of different time window lengths.

For each 5-second seismogram, the mean value was removed, and a 5% Hanning taper was applied. We account for attenuation using the expression:

where A is amplitude, r is epicentral distance, f is frequency, v is the seismic shear-wave velocity, and Q is the unitless quality factor. After trying several values, Q value was chosen to be 1000, based on when the high frequency spectra flattened out. This regional Q is similar to that found in Adams and Abercrombie [1998].

After FFT and Q removal, the two horizontal spectra for each record were averaged to form one horizontal spectrum. The resulting vertical and horizontal spectra were smoothed with a three point moving average and then logarithmically stacked.

Two Ways to Compute Site Response from Spectra

Multi-Station Spectral Ratios

One of the most common methods of estimating site response is the spectral ratio technique introduced by Borcherdt [1970]. It is computed by dividing the Fourier spectrum of one station by the Fourier spectrum of another. A record of ground motion can be considered a convolution of source, path, site, and instrument response effects. The spectral ratio technique assumes that site effects can be deconvolved from the others if: a) the instrument response is known for each station; b) the same sources are used; and c) the path from source to station is the same. By using the same set of earthquakes, condition b) is satisfied. If the hypocentral distance is much greater than the distance between the stations, differences in path effects will most likely be minimal. Local site conditions, therefore, cause the remaining differences in the spectra.

In general, spectral ratios are computed by dividing the horizontal spectra of a station for which you wish to know the site response, by the horizontal spectra of a nearby reference station on rock. The reference station, by virtue of being on rock, is assumed to have minimal site effects, although there is evidence that this is not necessarily true [Steidl et al, 1996]. Horizontal spectra are used because the horizontal spectra contain the P-S conversions caused by impedance discontinuities near the surface. Vertical spectra are thought to be largely uninfluenced by local structure. As is discussed later, this forms the key assumption of single-station, "receiver function-like" spectral ratios.

For this study, we simply divide the previously computed horizontal spectra for each station by the horizontal spectra at Piute Butte. We chose Piute Butte as our reference station because it had the highest average shear-wave velocity (most "rock-like") of our three characterized sites (Figure 4). Also, as can be seen in Table 2, Piute Butte recorded all 61 earthquakes used in this study. We originally wanted to use MCS for this purpose, but the average shear-wave velocity of 462 m/s puts the site well below any rock site classification, even though it is listed to be NEHRP Class A. This may be a widespread problem, as 4 of the 6 "rock" sites analyzed in Nigbor et al.’s [1998] ROSRINE preliminary results study fell below the 760 m/s average shear-velocity standard for southern California soft rock.

Single-Station Spectral Ratios

In Nakamura [1989], it was proposed that the vertical spectrum of ground motion, being uninfluenced by sedimentary layers and therefore containing uncontaminated information about the source, could be used to deconvolve source effects from site effects. In the frequency domain, this is accomplished by simply dividing the horizontal spectrum of a site by the vertical spectrum of the same site (H/V). The technique has proved useful in predicting resonance frequencies of sediments [e.g. Lermo and Chavez-Garcia, 1994; Field et al. 1990; and Lachet et al., 1996] using ambient noise (microtremor) as a source.

Lermo and Chavez-Garcia [1993], in a modification of Nakamura’s [1989] technique, showed that it is possible to use S-wave spectra to estimate site response. They find "very good" agreement between the H/V technique and standard spectral ratios (SR). Others, [e.g. Field and Jacob, 1995; Bonilla et al., 1997; Coutel and Mora, 1998; and Lachet et al., 1996] find that the H/V method compares favorably to the standard spectral ratio technique when determining the fundamental frequency of the sediments, but has significantly less success in determining the amplitudes, especially at higher frequencies. We attempt to compare the H/V and SR methods for our sites to see if the results converge given the extremely limited stack of sediments. In other words, we seek to test whether or not the magnitude of the disconnect is related to the thickness of the sedimentary stack under the site. If we could prove that the two techniques provide adequately consistent results, it would make future rock site characterizations simpler and cheaper.

Site Effect Estimation Using Shear-Wave Velocity Data

Comparisons using Quarter-Wavelength Approximations

The methodology of Boore and Brown [1998] allows direct comparison of theoretical amplifications between two sites given the shallow shear-wave velocity structures. The method uses the ¼ wavelength amplification approximation [Joyner et al., 1981] of each site. After converting the velocity profiles to travel time profiles, it is easy to determine which frequencies correspond to each depth interval. We then compute the ratio of the amplifications at each frequency. Like Boore and Brown [1998], we assume vertical angle of incidence at all sites, identical density profiles, and convergence of velocity profiles at some reference depth. Since we use Piute Butte as our denominator site, we have a direct analogy with our standard spectral ratios.

RESULTS

Horizontal and Vertical Spectra

Velocity spectra for both the vertical and averaged horizontal components are presented in Figure 5. As can be seen the horizontal spectrum is almost always greater than or equal to the vertical spectrum. This is especially true at MCS and LLA, where we know or suspect the subsurface velocity to be lower than at the other rock sites. Also notice that for almost all the stations, the maximum separation of the two components happens around 4-7 Hz. Steidl et al. [1996], noted that spectral ratios using rock sites were often underestimated as the rock sites had their own maximum site effect in this approximate frequency range.

Standard Spectral Ratios

Traditional spectral ratios, with Piute Butte as the reference site, show spectral ratios that are generally flat, with moderate amplification (Figure 6). The spectral ratio for MCS (Figure 6d) is an exception to this generalization. The spectra has a "step-like" geometry with higher than average amplification from 0.4 to 4 Hz, and less than average amplifications from 4 to 20 Hz. Although we have no direct evidence, we suspect that the amplitude spectrum is strongly influenced by topographic effects. MCS is situated on an eroding ridge, with relatively steep slopes on two sides. Although the co-located strong motion site (USC’s MCS) is classified as being on NEHRP Class A rock [http://smdb.crustal.ucsb.edu/ows-bin/owa/summary3.main?SiteName=MCS], we would re-classify the site as being Class C, based on our velocity measurements (Figure 4).

Single-Station H/V Spectral Ratios

H/V spectral ratios are presented in Figure 7. The H/V ratios show much the same character as the standard spectral ratios. A comparison of the methods is plotted in Figure 8. As can be seen, the two methods give generally consistent results for most stations, with the most notable exception being station LJB above 3 Hz. Otherwise, the H/V amplifications hover near the standard spectral ratio amplifications and no systematic over predictions or under predictions are noted. The frequency dependence of amplification is reasonably predicted, as well.

These results are in contrast to the results of Field and Jacob [1995] and Lachet et al. [1996], who find that that the H/V method is able to find the fundamental frequency of amplification, but systematically under predict the amplification of higher modes. Castro et al. [1997] speculate that the H/V method fails when the vertical component is amplified, as would be the case when surface waves contaminate the sampling interval. Our use of a short 5-second time window and smaller earthquakes, with less well-developed surface waves, may explain our more successful comparison. Alternatively, the under prediction might be proportional to the overall magnitude of amplification. If true, this would mean the more economical (in terms of time and equipment) and convenient (no need to find a suitable reference site) H/V method would be sufficient in characterizing the site effect at rock sites.

An additional benefit is demonstrated by comparing the spectral ratio methods at LJB (Figure 8a). As noted before, this is the station with the most disagreement between the two methods. PB and LJB have very similar local geology (discussed below). If two stations have similar seismic hazard, than a ratio between the two would always be near unity. If one were to assume a flat amplification at the reference site, than the standard spectral ratio would under predict the hazard at both sites. In this case, it would be wise to compute H/V ratios as a check of the standard method. Since the computational cost to do so is negligible, we view this as a wise course of action.

Quarter Wavelength Amplifications

Predicted ¼ wavelength amplifications, using the methodology of Boore and Brown [1998] are presented in Figure 9. The limited depth of our velocity information limits our analysis to 3.5 Hz and higher. We compare the predicted amplifications of our sites relative to the predicted amplifications of three shear-wave velocity profiles. One of the shear-wave velocity profiles is that measured at Piute Butte in this study. We thus have a direct analogue to our standard spectral ratio. The other two shear-wave profiles are for the NEHRP B-C boundary provided by Frankel et al., [1996], and the "generic rock" site profile provided by Boore and Joyner [1997]. The first 100 m of all profiles are plotted in Figure 4. Comparisons with the B-C boundary are of interest as the probabilistic seismic hazard maps of Frankel et al. [1996] are computed assuming that site condition. The "generic rock" profile in Boore and Joyner [1997] is an average of over 250 borehole velocity profiles. As such it allows us to test if amplification at rock sites is significantly different than at other representative rock sites. All three sites for which we have velocity information indicate that our sites are amplified relative to both the B-C boundary and "generic rock" (Figure 4).

The extremely low velocities seen at the surface at our sites (Figure 4) is the likely cause of high frequency amplification. As discussed in Boore and Joyner [1997], travel time in sedimentary layers, as opposed to 30-m average velocity , is the critical factor in determining amplification. Although both PB and LJB have higher V₃₀ than either the B-C Boundary or "generic rock" sites, the travel time near the surface is much longer. Both PB and LJB are at eroding granite outcrops in an arid environment. The outcrops are ringed by an extremely dry sand, which probably explains the low near surface velocity. The ¼ wavelength approximation, being insensitive to strong impedance contrasts, and instead presenting a smoothed version of the response [Boore and Joyner, 1997], smears the increased hazard over frequencies above 4 Hz.

It is unknown to what extent we have successfully quantified the nature of the sand-rock interface. Tests are underway to acquire more data at the sites in order to determine: (1) the lateral variability of depth-to-bedrock surrounding the outcrops, and (2) the repeatability and sensitivity to near surface layers of Louie’s [2001] shear-wave velocity measurement technique.

CONCLUSIONS

1.  
2. The site effects at precarious rock sites are consistent in magnitude and spectral shape with site effects from "ordinary" rock sites from other studies [Steidl et al., 1996; Boore and Joyner, 1997]. Site effects alone are insufficient to explain the continued existence of precarious rocks.
3.  
4. We achieve good agreement between H/V and standard spectral ratios. This may be because the overall site-effect is small at rock sites. Alternatively, it may be because of our short time window and limited surface waves.
5.  
6. Velocity characterizations at two precarious rock sites (PB and LJB) show 30-meter average shear-wave velocities well into the "rock site" range. Unusually low near surface velocities at each site couple with high velocities below 7 meters to create the high 30-m average. This is in contrast to the shear-wave velocity profiles of Frankel et al [1996] and Boore and Joyner [1997], which show more smoothed velocities relative to the 30-m average. As a result comparisons of ¼ wavelength amplifications predict amplification at the precarious rock sites above 4 Hz relative to the B-C boundary and "generic rock" sites.
7.  
8. Further verification is needed for the shear-wave velocity characterization technique of Louie [2001]. Specifically, the error at the crucial interface between low velocity sand and high velocity granite needs to be quantified.

REFRENCES