Seismic Wavefield Calibration of the Korean Peninsula Topic 2 – Seismic Calibration and Ground Truth Collection Technical Proposal

HerrmannDepartment of Earth and Atmospheric Sciences Saint Louis University 1.Summary This proposal addresses the calibration of seismic wave propagation in the Korean peninsula to impro

Seismic Wavefield Calibration of the Korean Peninsula Topic 2 – Seismic Calibration and Ground Truth Collection

Technical Proposal

R B HerrmannDepartment of Earth and Atmospheric Sciences

Saint Louis University

1.Summary

This proposal addresses the calibration of seismic wave propagation in the Korean peninsula to improve confidence in locations, to determine source mechanisms of seismic events through waveform inversion, to describe the high frequency attenuation of regional phases and to provide

a catalog of calibrated events for other studies This effort is possible because of cooperation with Dr Kiehwa Lee of Seoul National University and Dr Duk Kee Lee of the Korean

Meteorological Research Institute to address their research on seismic hazard This cooperative effort provides access to KMA (Korea Meteorological Administration) and KIGAM (Korean Institute of Geology and Mining) digital data sets The connection with Korea is furthered by the participation of Young-Soo Jeon as a post-doc for the duration of the effort and Hyun-jae Yoo as

a visiting researcher for the first 6 months of the proposal

The scientific problems to be addressed are the suitability of joint inversion of surface-wave dispersion and receiver functions for the determination of a crustal model adequate for waveformmodeling, the effect of adding travel time constraints to that inversion, the fine tuning of the crustal model for source parameter determination, and, when possible, the collection of ground truth In spite of the low seismicity rate, the existing data sets will permit the identification of GT595%

Seismic Wavefield of the Korean Peninsula Topic 2 – Seismic Calibration and Ground Truth Collection

R B HerrmannDepartment of Earth and Atmospheric Sciences

Saint Louis University

W Walters and M Pasyanos (Team members)Lawrence Livermore National Laboratory

2 Narrative

Seismic calibration is a complicated broad task, incorporating source location, identification and quantification, each of which has its own difficulties The Sino-Korean para-platform has been and is a long-term region of interest for such calibration Much work done over the broad region

of north China which must to be synthesized and tested An initial focus on seismic events within or near the Republic of Korea provides this opportunity by critically evaluating the

contribution of different data sets to calibration The discussion within this section focuses upon what can be accomplished through the application of cutting edge modeling techniques to both refine crustal velocity models and define ground truth location information at the GT595% level The study will use KMA/KIGAM waveform data and draw upon LLNL surface-wave

tomography products

2.1 Korean Framework

The Korean peninsula occupies the southeastern part of the North China Block or Sino-Korean craton (Fitches et al, 1991) in the Eurasian plate It is an important tectonic link between easternChina and the Japanese Islands The peninsula represents a denudation remnant of deformed basement rocks and sedimentary successions as well as granitic intrusions and volcanics,

concealing a long history of basin formation and crustal deformation The

peninsula has three major Precambrian massifs, viz., Nangrim, Kyonggi, and Yongnam massifs

in the north, central and southern part of the peninsula The massifs are associated with the higher elevations In the southeast, the Cretaceous Kyongsang Basin has gently eastward-

dipping successions of nonmarine sediments (Chough et al., 2000) Figure 1 presents a surface geology map of the peninsula The Bouguer gravity map, Figure 2, has negative residuals that correlate with the massifs Other than the southeast, the peninsula is characterized by old rocks

Analysis of earthquake data to improve crustal structure models and to define earthquake source parameters has progressed slowly because of the low rate of seismic activity ( ~ 50 earthquakes located annually in the south) and the lack of quality data The second issue was recently

addressed by the installation of modern digital seismograph systems by KMA and KIGAM Newly instrumented sites have a 3-component strong motion sensors combined with short period velocity sensors broadband sensors The data sets on local earthquakes are slowly

growing in size Selected teleseisms are achived Stations of the KMA network were initially deployed at local KMA offices in cities, which were noisy In addition some of the KMA sites were near KIGAM stations Subsequently KMA stations were redeployed to provide broader

national coverage from quieter sites.

Moment tensor source inversion has been performed by Kim and Kraeva (1999) and Kim et al (2000) (I note that one of the two Korean events studied has a seismic moment 4 times too large because of the use of an incorrect gain for the INCH LH channels)

Using C J Ammon's codes, Kim and Lee (2001), Kim et al (1998) and Yoo (2001) used the teleseismic P-wave receiver function technique to estimate crustal structure variations within the peninsula The Kim and Lee (2001) study is quite extensive but can be extended using

additional waveform data, domain, rather than water level, deconvolution techniques and by the addition of other data, such as surface-wave dispersion or body-wave travel times

Travel time studies have been performed using limited data sets References to many of these models are found in Kim and Lee (2001) Song and Lee (2001) used the VELEST program to

Figure 1 Simplified geology map of Korea

set of 178 travel times from 29 earthquakes A plot of first arrival times for an initial location based on the Kim and Kim (1983) velocity model, showed a simple linear trends corresponding

to velocities of 6.3 and 8.0 km/sec from which an average crustal thickness of 35 km is inferred

Figure 2 Bouguer gravity map of Korea and the neighboring region Note the very negative

anomalies in the northern part of the country.

2.2 Source parameter determination

Dr Duk Kee Lee of KMRI visited Saint Louis University at the end of November, 2002, and brought event recordings from 8 events made by the KMA network These events had local magnitudes in the range of 3.4-4.1 and were among the larger events recorded over a two year period Table one gives the preliminary event locations

Table 1 Recent local event locations

combined KMA and KIGAM seismic networks Figure 3 shows the locations of the earthquake and the broadband stations that recorded the 21 NOV 2001 event Data are also available from the accelerometers and short period sensors at these and other locations which are not discussed here The broadband stations are at distances of 83 – 205 km from the event

Waveform inversion was initially performed using only the traces at SEO, ULJ and TAG which had clean records The Central U S (CUS) model was used because the Green’s functions were

at hand and since they matched the P – Surface wave interval time better than the Song and Lee

(2001) model The program search program, wvfgrd96, described in Computer Programs in Seismology 3.20 – Source Inversion (2002), was used Figure 4 compares the observed and

predicted waveforms for this event for all the stations shown in Figure 3 The ground velocity traces are bandpass filtered in the 0.02 – 0.10 Hz band The source depth used was 13 km, the

Mw = 3.44 and the mechanism has a strike, dip and rake angles of 15, 65 and 150 degrees, respectively The fits are quite good but indicate a tendency for the synthetic surface-wave arrival

to occur slightly later than the observed Part of this is due to the discrete distances at which the Green's functions were computed SNU and SEO are 143 and 148 km from the source,

respectively, and the program used Green's functions at 145 and 150 km, respectively This concern over seemingly small time shifts is critical if high frequencies must be used in the inversion, which is necessary for even smaller events

Figure 5 compares the observed and predicted waveforms for the bandpass filter range of 0.02 –

1.0 Hz (SAC command bp c 0.02 1.0 np 2 ) The difference between the observed and predicted

traces is less than a factor of 2 for some of the traces, which is surprising given the simplicity of the crustal model used for the Green's functions

A detailed discussion of this event and that of 24 NOV 2001 is given at

http://www.eas.slu.edu/People/RBHerrmann/KOREA.2003/

The interesting fact is that waveform inversion was successfully applied to 2 of the 43

earthquakes located during 2001 Having demonstrated the ability to obtain source parameters the challenge is to extend this to events which have lower signal-to-noise ratios because of smaller size or increased seasonal noise

Figure 3 Location of broadband stations used for determining source parameters of the

21 NOV 2001 earthquake.

Figure 4 Comparison of observed (light gray) and predicted (dark) traces at each station The trace pairs for a given component are plotted with the same scale and the peak amplitudes are indicated for each Each trace is 80 sec long and starts at a time r/8.0 -5.0 sec after the origin Signals are bandpass

filtered between 0.02 and 0.10 Hz.

The primary result of this preliminary study is that determination of source mechanism, seismic

moment and event depth is possible, even for M =3.4 earthquakes if a reference earth model is

known In addition, recognition of phases, such as sP (S up from the source, refracted as P along the surface) and P in the 100-200 km distance range can provide source depths within a few km Precise source depths are a component of discrimination

Figure 5 Comparison of observed and predicted ground velocity traces at frequencies < 1.0 Hz.

2.3 Structure Inversion

Julia et al (2000, 2003) implemented the technique of joint inversion of surface-waves

and receiver functions for crustal structure beneath a station The many broadband stations in the Republic of Korea operated by KMA and KIGAM permit the application of such an inversion technique Application of this technique requires quality receiver functions, good dispersion and

a starting model that does not bias the results

Mr Hyun-Jae Yoo of Seoul National University collected teleseisms recorded at 25 locations which had KMA and KIGAM instruments Most of the events were from Indonesia, with a few from India/Afghanistan and Alaska All waveforms were examined for a P-wave arrival with good signal-to-noise and the better ones were processed using the time-domain deconvolution technique of Ligorria and Ammon (1999) The implementation places a goodness of fit

parameter in the SAC header of the receiver function to indicate the ability of the receiver

function to predict the filtered radial component, with 100% being a perfect prediction All receiver functions with at least an 80% goodness of fit were identified Because of the small variation in ray parameter, a presumed simple structure beneath Korea and the lack of significantazimuthal coverage, the individual traces for Gaussian filter parameters ALPHA = 1.0 and 2.5, corresponding to low pass corner frequencies of about 0.3 and 0.8 Hz, respectively, were stacked

to create a data set two receiver functions for each station Figures 6 and 7 show the station locations and the stacked receiver functions Inversions with the stacked data were quicker than with using the many individual traces, but the resulting model did not differ significantly

The shaded area display of the receiver function stacks in Figure 7 was organized to receiver function similarity using a cluster analysis If the receiver functions are controlled by crustal structure, then geographically adjacent stations should have similarly shaped receiver functions with locations grouped together

The Rayleigh-wave dispersion data available are currently sparse A single dispersion curve was used for all stations, even for the island stations of ULL, SOG and SGP The group velocities were taken from Stevens and Adams (2000) by asking the program for the dispersion between two points 1 degree apart in latitude in the peninsula In addition, a few phase velocity dispersionpoints were obtained from a p-omega stack of teleseisms propagating across the array of

broadband stations

Stable inversion requires constraints and a conscious decision to prevent persistence of initial model detail in the final inversion results The same starting model and processing scripts were used for each of the 25 stations so that resultant models could be compared The starting model was based on AK135 (Kennett et al, 1995) with the upper 50 km having the velocities fixed at their 50 km values In this case, the receiver function and dispersion data are required to define the crustal structure and the sharpness of the Moho Layering consists of twenty-five 2 km thick, followed by ten 5 km thick and finally ten 10 km thick layers to yield a 200 km thick model The halfspace velocities were fixed, and the model was constrained to be very smooth beneath a

50 km depth, permitting minor departure from the AK135 model The fit to the receiver functionsand surface wave dispersion was such that 96% of the signal power in the receiver functions and99% if the signal power in the dispersion data were fit

Figure 6 Location of broadband stations used for receiver function analysis

Figures 8 and 9 compare the 25 models and also the model predicted P-wave first arrival times for a surface source depth Most inversions share the same features, as expected from the

similarity of all receiver functions The exceptions are ULL, SOG and SGP The first arrival timepredictions for each model, including ULL, SOG and SGP, are similar and in qualitative

agreement with the Song and Lee (2001) simplified crustal velocity structure

As an independent test on the models, waveform integration synthetics were computed in an attempt to use a Korean velocity model for waveform inversion of the 21 NOV 2001 earthquake instead of the Central U S model (CUS) This was not successful since these models predicted later surface-wave arrivals than observed Since the inversion derived the P-wave arrival times from the shear-wave velocities based on the initial Vp/Vs ratios, inversions were rerun using different values of these ratios This was not sufficient to improve the waveform modeling of the regional events

To force a better fit to the surface-wave arrivals the original dispersion set was augmented by the theoretical CUZ model Love- and Rayleigh-wave fundamental mode phase and group velocity dispersion between 4 and 30 seconds, the bandwidth of the surface-waves shown in Figures 4 and 5 Figures 10 and 11 compare the models and the predicted P-wave travel times Evidently the additional surface-wave data provide a very strong constraint on the upper crustal velocities such that the predicted arrival times vary little

Figure 7 Receiver function stacks for the two Gaussian filter parameters used The number adjacent

to each receiver function indicates the number of individual receiver function in the stack.

Figure 8 Joint inversion model for each station The solid black line is the mean model of all stations

except for the stations on the two islands.