Imaging the Upper Crust of the Korean Peninsula by ... - CiteSeerX

Bulletin of the Seismological Society of America, Vol. 97, No. 1B, pp. 198–207, February 2007, doi: 10.1785/0120060096

Imaging the Upper Crust of the Korean Peninsula by Surface-Wave Tomography by K. H. Cho, R. B. Herrmann, C. J. Ammon, and K. Lee

Abstract Cross correlation of seismic-background motions (Campillo and Paul, 2003; Shapiro et al., 2005) is applied to observations from the Korean Meteorological Administration seismic network to estimate the short-period Rayleigh and Love wave dispersion characteristics of the region. Standard processing procedures are applied to the cross correlation, except that signal whitening is used in place of one-bit sampling to equalize power in signals from different times. Multiple-filter analysis is used to extract the group velocities from the estimated Green’s functions, which are then used to image the spatially varying dispersion at periods between 0.5 and 20 sec. The tomographic inversion technique used inverts all periods simultaneously to provide a smooth dispersion curve as a function of period in addition to the usual smooth spatial image for a given period. The Gyeongsang Basin in the southeastern part of the peninsula is clearly resolved with lower group velocities. Introduction The estimation and validation of a seismic model of the crust is necessary to reliably estimate earthquake source parameters in the region of the Korean Peninsula. Previous geophysical studies of the deep crustal structure of the Korean Peninsula have been attempted using seismic traveltime observations (Lee, 1979; Kim and Kim, 1983; Kim, 1995; Song and Lee, 2001). Crustal refraction profiles (Cho et al., 2006) and receiver function studies (Lee and Kim, 1998; Yoo and Lee, 2001; Chang et al., 2004; Chang and Baag, 2005) have also been used to constrain South Korea’s crustal structure. Because of limited spatial sampling, these studies only provided a rough estimate of the three-dimensional crustal structure beneath Korea, especially for the shallow crust. Recent work on the extraction of interstation Green’s functions through the cross correlation of ambient ground motion provides a valuable passive technique for investigating earth structure. This approach was introduced to seismology by Campillo and Paul (2003), who studied the scattered waves following a regional earthquake in Mexico. Subsequent work by Shapiro and Campillo (2004) and Shapiro et al. (2005) showed that the technique need not be restricted to the scattered wave field in an earthquake coda, but could also be applied to ground noise in general. A theoretical basis for the approach was derived by Wapenaar (2004) through the representation theorem of elastic wave theory. The essence of this technique is that cross correlation of ground noise at two stations yields a time series that is equivalent to the signal observed at one of the locations due to a surface point source acting at the other location. This signal has a spectral-amplitude shape that combines the ef-

fects of the ambient ground-motion spectra and the Green’s function excitation, but the phase information reflects the elastic structure between the stations. We focus on wave propagation within the southern Korean Peninsula because of the availability of high-quality digital seismic data from a network with over 80 stations. The station density produces approximately 3000 interstation pairs for the cross-correlation analysis, which permits a high-resolution study of short-period surface-wave dispersion in the southern peninsula.

Data Processing Our noise data were obtained from continuous 20-Hz data streams recorded at broadband velocity and accelerometer stations in South Korea during 2004 and 2005, operated by the Korea Meteorological Administration (KMA). The instrumentation at the accelerometer sites consists of Kinemetric Episensors and Quanterra digitizers. Some locations also have colocated STS-2 broadband velocity sensors. All acquired data are forwarded in real time to KMA in Seoul. Figure 1 is a map of the locations of both types of sensors for this study; the possibility of greater spatial resolution due to the use of accelerometer data is immediately obvious. We used digital data from 2004/349 (year/day) through 2005/167 for the broadband velocity channels and 2005/201 through 2005/253 for the accelerometer channels. Corrections for instrument response were unnecessary because we cross-correlated waveforms with the matched instrumental responses. Cross correlation then removes the common instrumental phase response, but squares the effect 198

199

Imaging the Upper Crust of the Korean Peninsula by Surface-Wave Tomography

Figure 1. Maps of the study area showing the locations of the broadband (a) and the acceleration sensors (b). The wide gray line from station SES in the northwest through DAG in the southeast defines a profile that will be used to illustrate the tomography results. The dotted lines separate tectonic provinces: gm, Gyeonggi massif; of, Okcheon fold belt; ym, Yeongnam massif; gb, Gyeongsang Basin (Chough et al., 2000; Cho et al., 2006). Station BRD is on BaegRyeong-do Island and station JJU is on Jeju Island. of the instrumental amplitude response. This amplitude distortion is removed by a “whitening step” in the processing. Because noise observations are available from threecomponent seismograms, empirical Green’s functions for both Rayleigh and Love waves can be obtained from the noise cross-correlation method. The data were inspected for glitches and other abnormal signal properties and then rotated into vertical, radial, and transverse components for each station pair with suitable day-long time windows. Cross correlations were computed between the respective vertical, radial, and transverse components to yield two Rayleigh and one Love wave data set for each station pair. Campillo and Paul (2003) and Shapiro et al. (2005) applied a one-bit transform to the traces prior to performing a cross correlation to reduce the effect of a strong coherent signal, such as a teleseismic or regional seismic events. Although this operation may succeed in this task, the spectrum of the resultant cross-correlation function is often strongly peaked, making the extraction of the interstation groupvelocity dispersion more difficult. We apply an alternative nonlinear procedure. Day traces are processed in 23 onehour segments starting at 00:30 and ending at 23:30 to avoid possible data loss at the beginning and end of the day due to the start and end time of the original raw data. The onehour segments are spectrally whitened to produce a flatamplitude spectrum in the 0.05–5 Hz band, tapering the spectrum to zero at 0.02 and 8 Hz. Whitening preserves the signal phase in the passband for propagation over short distances to 500 km. The whitened traces are then crosscorrelated. Because we cross-correlate only acceleration data with one other, and broadband velocity data with each other,

the instrument phase is eliminated. All 23 one-hour cross correlations are stacked with their time-reversed cross correlations to create a symmetric day cross correlation. Finally all available day stacks for a given station pair are stacked to produce the empirical Green’s functions. For the most complete data we were able to stack 88 days from the broadband velocity sensors and 36 days from the acceleration sensors. Figure 2 presents a sample of the 2069 available cross-correlation Green’s functions between the vertical components of the acceleration channels. These signals are the direct product of the processing and have not been filtered for presentation; most of the traces exhibited this excellent signal-to-noise ratio. The short-period dispersion is obvious, justifying the effort to use the accelerometer data. We used a similar visual display for quality control; for surface-wave arrivals that had unexpected propagation speeds, we checked the correctness of the assumed station coordinates, in one case detecting incorrect coordinates, and component orientation with KMA personnel. Because of the large number of interstation pairs considered in this study, we did not examine the symmetry of the stacks, which would be indicative of the isotropic nature of the noise field. We just used the symmetric portion of the stack. Because the interstation paths sampled all azimuths uniformly and the derived group velocities were very similar, we believe that the assumptions were satisfied.

Group-Velocity Analysis We measured group velocities using a multiple-filter technique (Dziewonski et al., 1969; Herrmann, 1973). The

200

K. H. Cho, R. B. Herrmann, C. J. Ammon, and K. Lee

Figure 2.

Examples of computed vertical-component interstation cross correlations with respect to the accelerometer station AND. The distances between AND and the other stations are 24.6 (EUS), 36.8 (YOJ), 50.1 (MUG), 60.8 (KUM), and 62.9 (YOD) km in order from top to bottom. All traces start at zero lag and are unfiltered. Group velocities of arrivals are indicated by the blue ticks.

waveforms were narrow bandpass filtered using the operator exp[ⳮ␣( f ⳮ fc )2/f c2 ] where fc is the filter center frequency. As is known from Dziewonski et al. (1969) there is a tradeoff between resolution in the time and frequency domains with such filtering. A large ␣ provides better frequencydomain resolution at the expense of reduced resolution in the time domain. This is a concern for our study because our cross correlation is constructed to be symmetric to match the expectations of distributed random noise sources (Wapenaar, 2004). In addition to standard signal-processing noise, we must be careful to avoid interference of the forward- and reverse-correlation pulses, which have smaller time separations as interstation distance decreases, thus limiting the maximum period for which dispersion can be resolved from the Green’s functions. Before using multiple-filter analysis, we must define values of the optimal ␣-filter parameter as a function of distance and the range of filter center frequen-

cies, or periods, that can be trusted. We addressed this problem using synthetic seismograms computed for shallow sources in Herrmann et al.’s (2005) Korea velocity model. Because the model and theoretical dispersion are known, a direct comparison with the group velocities derived by multiple-filter analysis provides guidelines for acceptance. The resulting filter parameters and guidelines are listed in Table 1. In practice we only used dispersion estimates for interstation distances greater than 20 km. To gain further confidence on the dispersion results, we processed some station pairs using both the broadband velocity and accelerometer data channels. The low-gain accelerometers are much less sensitive at lower frequencies than the velocity sensors and thus questions of sensor self-noise arise. Using the BRD-DGY and the BRD-JJU station pairs, which have both acceleration and broadband velocity sensors, we found that the accelerometers had limited ability to


Table 1 Selection of Period Range for ␣-Filter Parameter Distance (km)

␣ ⳱ 6.25 (sec)

␣ ⳱ 12.5 (sec)

␣ ⳱ 25 (sec)

10 20 30 40 50 70 100 200

No 1–3 1–5 1–7 1–7 1–10 1–20 1–20

No No 1–4 1–4 1–4 1–7 1–10 1–20

No No 1–3 1–3 1–3 1–5 1–10 1–20

provide dispersion for periods greater than 10 sec, but that the dispersion in the 2- to 10-sec period range agree between the two instrument systems. Another test investigated the DAG-SEO path comparing the similarity of the interstation Rayleigh wave dispersion using 10, 20, 50, 100, and 150 days of stacks. We found that 20 days of data begins to provide a stable dispersion curve. To be safe, we required at least 30 days of cross correlations for our dispersion values. We obtained acceptable cross correlations for all paths between stations, except for those to the station ULL on Ulleung-do Island, off the east coast of Korea. The failure along this path may be due to inefficient short-period surface-wave propagation along a mixed continent–ocean path. We also noticed low group velocities for some paths from the Korean Peninsula to Jeju Island located to the south along the west coast. Figure 3 is a comparison of the observed mean dispersion, together with standard deviations, to that predicted by a model used for moment-tensor inversion for Korea earthquakes (Herrmann et al., 2005). Their model was obtained from the joint inversion of receiver functions at the Korea Institute of Geoscience and Mineral Resources broadband station on the campus of Seoul National University, Korea average Rayleigh wave phase velocities for periods of 10–150 sec, and some short-period Rayleigh wave group velocity for four paths in the northern part of the country. The dispersion shown in Figure 3 highlights several important points. First, the dispersion obtained from the cross correlation of ground noise is similar to that predicted from independent data. The smaller standard deviations near the 4-sec period reflect the greater number of dispersion measurements near that period range. The greater scatter at longer periods is indicative of the difficulty of the technique to give good estimates over these relatively short interstation paths.

Surface-Wave Tomography To map first-order variations in dispersion and shearvelocity across the Korean Peninsula, we used a simple great-circle-based ray tomography approach with a uniformslowness cell width of 12.5 km and smoothing, which allows us to relate the tomography results with surface geology. The

201

method is similar to that used by Mokhtar et al. (2001), but we extended the approach to allow smoothing between periods, which are often analyzed separately in such analyses (e.g., Pasyanos and Walter, 2002). The extension allows us to ensure smooth dispersion variations from period to period. We used standard checkerboard resolution tests to investigate the resolution and stability of the inversion. We created a numerical test data set using a grid pattern with 8 ⳯ 8 cell groupings (100 km ⳯ 100 km), with velocities alternating between 2.85 and 3.15 km/sec (3.00 km/sec Ⳳ 10%) and inverted the target-model group delays with the same smoothing and damping parameters used for the observations. Figure 4 shows the test pattern and the results of the inversion for the same paths as for the observed Rayleigh waves at periods of 1, 4, and 20 sec. Several features are apparent from these figures. First, the ability to resolve the velocities is related to the total number of paths available for the inversion. Second, the smoothing operator inhibits resolution of the sharp step discontinuities of the target model. Third, the limited number of ray paths in the sea to the west precludes resolution in this region except for a few select locations. Finally, when the grid cells contain few crossing rays, such as in the northwest corner of the study region, the limited data, and the minimum length and smoothing constraints result in ray-parallel streaking. The variance reduction was 91%, 95%, and 75% at the periods of 1, 4, and 20 sec, respectively, with the poorer fits due to the smaller number of paths at 20 sec. If we had used only the data from the few broadband stations, it would not have been possible to perform such a high-resolution tomographic study. At the shorter periods, we believe that we have selected reasonable smoothing parameters to characterize spatial variations in dispersion for most of the study area. Of course, artifacts arise along the periphery of the study region because of insufficient sampling. Figure 5 presents the Rayleigh wave group-velocity inversion results at periods of 1, 2, 3, and 6 sec. To the extent that surface-wave dispersion is primarily affected by the velocity structure to a depth of one-third wavelength, these figures can be interpreted as providing average structure over depths of 1, 2, 3, and 6 km, respectively. The 1-sec dispersion in Figure 5a is constrained by 1589 path values with a mean velocity of 2.80 km/sec. The 2-, 3-, and 6-sec dispersion images are the result of 3187, 3376, and 2837 observations and have mean velocities of 2.86, 2.94, and 3.07 km/ sec, respectively. The variance reduction at these four periods was 40%, 60%, 70%, and 33%, respectively. To assist in interpretation, Figure 5 also shows major geological province boundaries. The most striking features are the low velocities on interstation paths to BaegRyeong-do Island (BRD) in the northwest and to Jeju Island in the south. Because of the limited number of crossing paths, these anomalies may be the result of a few bad observations. On the other hand, Jeju Island (JJU) is volcanic and the shallow structure could be composed of low shear-velocity material. Focusing on the

202


Figure 3. Mean dispersion curves for all peninsular paths for Love (left) and Rayleigh (right) waves. Each observation period is plotted together with the standard deviation bars. The wide solid curves are theoretical predictions based on the crustal velocity model given in Table 2.

peninsula, we see that there is no direct relation between the velocities and the massifs and fold belt. The southwest part of the peninsula appears to be a few percent faster than structures to the northeast. The Gyeongsang Basin, located in the southeast, consistently has velocities lower than the other tectonic units by about 10%, 15%, 8%, and 4% at periods of 1, 2, 3, and 6 sec, respectively. Love wave dispersion maps for the same periods are given in Figure 6. The number of observations at each period are roughly half those for the Rayleigh wave study, because the latter had dispersion estimates from both the verticalvertical and radial-radial cross correlations. The variance reduction at the periods of 1, 2, 3, and 6 sec is 44%, 47%, 56%, and 30%, respectively. The Love wave results show low velocities on paths to BRD and to Jeju Island. The patterns of relative high and low group velocities are similar to those for Rayleigh waves. The pattern of the relatively lower velocities in the Gyeongsang basin agrees with that seen in the Rayleigh wave tomography.

Discussion Although this effort was directed toward creating tomographic maps of short-period group-velocity dispersion for the Korean Peninsula, inversion of the spatially varying dispersion for structure was not done systematically for several reasons. First, surface-wave dispersion, even with if higher-mode dispersion can be determined, only provides a smooth velocity model because of limited resolution, unless some a priori assumptions about layering are made, which we do not have for Korea. On the other hand, the lower velocities inferred for the Gyeongsang Basin are intriguing

enough to attempt a preliminary inversion. Cho et al. (2006) reported on the first-crustal scale controlled source experiment in the southern Korean Peninsula. Their profile is essentially the west-northwest–east-southeast diagonal line in Figure 1a. They summarized their interpretation in a 2D crustal section for the 300-km-long profile. A striking feature of their interpretation was the Gyeongsang Basin, which occupied the second half of their profile. This basin has lowvelocity materials to a depth of 5 km. Their model had limited resolution at these shallow depths and in this region, however, because of shotpoint locations. We used the Rayleigh wave tomography results along a 300-km path shown in Figure 1a. Group-velocity dispersion in the 0.5- to 10.0-sec period range was extracted from the tomography results for nine locations along this profile. These data were augmented with the Korea 10- to 150-sec phase-velocity dispersion used to define the velocity model in Table 2. The starting model for the surface-wave dispersion was that of Table 2, except that the upper 4 km was replaced by 0.5-km-thick layers with velocities increasing stepwise from 2.9 to 3.6 km/sec by 0.5 km/sec. Including the phase-velocity dispersion and starting with a regional crustal–upper-mantle model produces a model that fits the new short-period dispersion observations but does not yield a completely erroneous model at depth. We used the program surf 96 (Herrmann and Ammon, 2004) to invert the observed dispersion for the layer thickness of the top eight layers, while keeping other layer thicknesses fixed. Working with layer thicknesses is more convenient for creating a 2D cross section of the uppermost crust along the profile, which we show in Figure 7. The cross section in Figure 7 indicates a narrow de-


203

Figure 4. Resolution pattern test for the Rayleigh wave data set. (a) target-model velocity distribution. Tomographic inversions use observed Rayleigh-wave paths at 1.0 s (b), 4.0 s (c), 20.0 s (d). There were 852, 1853 and 40 inter-station paths at periods of 1, 4, and 20 seconds, respectively. pression, about 75 km wide near the south-southeast end of the profile. The location of this feature is controlled by the 1.0-sec dispersion in Figure 5a. The lower velocities penetrate to about a 3.0-km depth, slightly different than the model presented by Cho et al. (2006), who showed a uni-

formly deep basin for the right third of Figure 7. In addition, their assumed P-wave velocity of 4.71 km/sec would correspond to much lower S-wave velocities than we obtained. We suggest that our data have better resolution of the shallow velocities than theirs. Although the Gyeongsang Basin

204


Figure 5. Rayleigh wave group-velocity tomography results for periods of 1.0 (a), 2.0 (b), 3.0 (c), and 6.0 (d) sec. The boundary lines on the peninsula separate the major tectonic units: Gyeonggi massif, Okcheon fold belt, Yeongnam massif, and Gyeongsang Basin (Chough et al., 2000; Cho et al., 2006).

is Cretaceous in age, the basin consists of different units, having a roughly north-northeast strike, and we may be imaging those units in our Figure 7. We were pleased that we were able to successfully obtain interstation Green’s functions for Korea through the

cross correlation of ground noise, especially from the acceleration time histories. One reason for our success is that the accelerometers are sensitive, with a clip level of less than 1g. The accelerometer data were essential for our study because this permitted the dense spatial sampling. We were


205

Figure 6. Love wave group-velocity tomography results for periods of 1.0 (a), 2.0 (b), 3.0 (c), and 6.0 (d) sec. The boundary lines on the peninsula separate the major tectonic units: Gyeonggi massif, Okcheon fold belt, Yeongnam massif, and Gyeongsang Basin (Chough et al., 2000; Cho et al., 2006).

careful in extracting the group velocities, taking into account filter parameters, distance ranges, and the limitations of the multiple-filter analysis technique. The Rayleigh wave tomography is more believable because of the smoother dispersion maps (Fig. 5), which we believe reflects a more con-

sistent set of dispersion observations than the Love waves. Our efforts are another step in defining a 3D crustal model for the Korean Peninsula, and our focus in this instance was on the variability of velocities in the upper few kilometers of the crust. The power of the short-period dispersion in

206


Figure 7. Inverted-velocity structure along the 300-km path shown in Figure 1. The left of the figure is is at 36.7⬚ N, 126.5⬚ E and the right is at 35.7⬚ N and 129.5⬚ E. S-wave velocities in the layers are noted.

Table 2

Defense DTRA01-02-C-0038 and Department of Energy DE-FC5204NA25539.

Joint Inversion Velocity Model for Korea (Herrmann et al., 2005) H (km)

VP (km/sec)

VS (m/sec)

Density (g/cm3)

1 1 1 3 5 5 4 5 2 2 2 2 5 5 —

5.38 5.81 6.17 6.29 6.32 6.42 6.56 6.64 6.65 7.10 7.92 7.89 7.87 7.57 7.75

3.00 3.24 3.44 3.51 3.53 3.58 3.66 3.70 3.71 3.96 4.41 4.40 4.39 4.22 4.33

2.58 2.66 2.75 2.79 2.80 2.82 2.87 2.89 2.89 3.01 3.28 3.27 3.26 3.16 3.23

defining shallow structure shown in Figure 7 can be combined with receive-function analysis at many of the same accelerometer sites used for this investigation. Such a joint inversion provides stronger constraints on the crustal velocity model than are possible from the very-short-period dispersion derived in this study and is the subject of the article by Yoo et al. (2007). Perhaps such a study can also resolve some questions we have about the velocity structure near the island stations.

Acknowledgments This study would not have been possible without the digital data sets provided by the Korea Meteorological Administration. Research at Seoul National University was supported under the BK21 project. The effort in the United States was supported in part by Contracts U.S. Department of

References Campillo, M., and A. Paul (2003). Long range correlations in the diffuse seismic Coda, Science 299, 547–549. Chang, S.-J., and C.-E. Baag (2005). Crustal structure in Southern Korea from joint analysis of teleseismic receiver functions and surface wave dispersion, Bull. Seism. Soc. Am. 95, 1516–1534. Chang, S.-J., C.-E. Baag, and C. A. Langston (2004). Joint analysis of teleseismic receiver functions and surface wave dispersion using the genetic algorithm, Bull. Seism. Soc. Am. 94, 691–704. Cho, H.-M., C.-E. Baag, J. M. Lee, W. M. Moon, H. Jung, K. Y. Kim, and I. Asudeh (2006). Crustal velocity structure across the southern Korean Peninsula from seismic refraction survey, Geophys. Res. Lett. 33, L06307, doi 10.1029/2005/GL025145. Chough, S. K., S.-T. Kwon, J.-H. Ree, and D. K. Choi (2000). Tectonic and sedimentary evolution of the Korean Peninsula: a review and new view, Earth-Sci. Rev. 52, 175–235. Dziewonski, A., S. Bloch, and M. Landisman (1969). A technique for the analysis of transient seismic signals, Bull. Seism. Soc. Am. 59, 427– 444. Herrmann, R. B. (1973). Some aspects of band-pass filtering of surface waves, Bull. Seism. Soc. Am. 63, 663–671. Herrmann, R. B., and C. J. Ammon (2004). Surface waves, receiver functions and crustal structure, in Computer Programs in Seismology, Version 3.30, Saint Louis University. http://www.eas.slu.edu/People/ RBHerrmann/CPS330.html (last accessed December 12, 2006). Herrmann, R. B., Y.-S. Jeon, and H.-J. Yoo (2005). Broadband source inversion using digital data from Korea seismic networks, in Proc. 4th International Seminar on Seismic Tomography of Far-East Asia and Related Works, 2 December 2005, Korea Institute of Geoscience and Mineral Resources, Daejeon, Korea. Kim, S. J., and S. G. Kim (1983). A study of the crustal structure of South Korea by using seismic waves, J. Korean Inst. Mining Geol. 16, no. 1, 51–61 (in Korean). Kim, S. K. (1995). A study on the crustal structure of the Korean Peninsula, J. Geol. Soc. Korea 31, no. 4, 393–403 (in Korean). Lee, K. (1979). On crustal structure of the Korean peninsula, J. Geol. Soc. Korea 15, 253–258. Lee, S. K., and S. G. Kim (1998). Crustal structure study and characteristics

207

Imaging the Upper Crust of the Korean Peninsula by Surface-Wave Tomography of moho discontinuities beneath the Seoul and Inchon stations using teleseismic receiver functions, Econ. Environ. Geol. Korea 31, 339– 347 (in Korean). Mokhtar, T., C. J. Ammon, R. B. Herrmann, and H. A. A. Ghalib (2001). Surface wave velocities across Arabia, Pure Appl. Geophys. 158, 1425–1444. Pasyanos, M. E., and W. R. Walter (2002). Crust and upper-mantle structure of North Africa, Europe and the Middle East from inversion of surface waves, Geophys. J. Int. 149, 463–481. Shapiro, N. M., and M. Campillo (2004). Emergence of broadband Rayleigh waves from correlations of the ambient seismic noise, Geophys. Res. Lett. 31, L07614, doi 10.1029/2004GL019491. Shapiro, N. M., M. Campillo, L. Stehly, and M. Ritzwoller (2005). Highresolution surface-wave tomography from ambient seismic noise, Science 307, no. 5715, 1615–1618. Song, S. G., and K. Lee (2001). Crustal structure of the Korean Peninsula by travel time inversion of local earthquakes, J. Geophys. Soc. Korea 4, 21–33. Wapenaar, C. P. A. (2004). Retrieving the elastodynamic Green’s function of an arbitrary inhomogeneous medium by cross correlation, Phys. Rev. Lett. 95, 254–301. Yoo, H., and K. Lee (2001). Crustal structure under the Taejon (TJN) station by receiver function methods, J. Geophys. Soc. Korea 4, 35–46.

Yoo, H. J., R. B. Herrmann, K. H. Cho, and K. Lee (2007). Imaging the three-dimensional curst of the Korean peninsula by joint inversion of surface-wave dispersion and teleseismic receiver functions, Bull. Seism. Soc. Am. (in press). School of Earth and Environmental Sciences Seoul National University Seoul 151-742, South Korea (K.H.C, K.L) Department of Earth and Atmospheric Sciences Saint Louis University, 203 O’Neil Hall 3642 Lindell Boulevard St. Louis, Missouri 63108 (R.B.H.) Department of Geosciences Penn State University, Deike Bldg. University Park, Pennsylvania 16802 (C.J.A) Manuscript received 28 April 2006.