ALGORITHMS FOR REGIONAL STRUCTURE - PROGRESS AND APPLICATION R. B. Herrmann¹, C. J. Ammon², Kiehwa Lee³ and H. J. Yoo³ Saint Louis University¹, Pennsylvania State University², Seoul National University³ ABSTRACT Significant progress has been made in the development and documentation of standardized algorithms for analysis of regional earth structure and sources. Computer Programs in Seismology is officially released. This distribution contains corrections to the surf96, rftn96 and joint96 released in February, new codes for wavenumber integration synthetics in TI media, and as well as codes for inversion of regional surface-wave spectral amplitude patterns and complete waveforms for source parameters. Receiver functions were determined for all current broadband station location in Korea except for INCN (IRIS) and KSAR. The use of the time-domain iterative deconvolution tool of Ligorria and Ammon (1999) provides stable, causal and consistent receiver functions for inversion. Except for the stations on two islands, the receiver functions are very similar, attesting to the uniformity of the crust on the southern part of the Korean peninsula. The receiver functions require a uniform velocity model for the crust with a sharp, within resolution, crust-mantle transition. A test of the new programs rftn96 and joint96 was made using the data set at the Seoul National University station – the initial inversions highlight the non-uniqueness of receiver function inversion unless a priori crustal model information or other independent data, such as surface-wave dispersion, are introduced. Fig 1. Location of broadband stations used in this study Fig. 2 Stacked receiver functions for the two filter values. The number of traces used is indicated at he end of each trace. The island stations SOG/SGP and ULL have different receiver functions from the others. Note that the receiver function at the KIGAM station BRD is similar many of those from stations on the peninsula. Fig. 3. Inversion fit (red) to observed receiver functions (black) Fig. 4. Joint inversion models starting from halfspace Fig. 7. CHU model predicted Rayleigh wave fundamental mode phase and group velocities shown with long-period constraints on the model. The short period dispersion is from an event in North Korea ( ). The distances to CHU, INJ and WON are 68, 59 and 129 km, respectively. Given location accuracy, there could be 10% errors in group velocity at the nearest stations. The low group velocities at 1.0 sec support the low surface velocities from the joint inversion. Fig.6. Multiple filter analysis of WON ELZ km Fig. 5. P-wave first arrival predictions for a surface focus Computer Programs in Seismology 3.20 Now available for download: ftp://ftp.eas.slu.edu/pub/rbh/PROGRAMS.320 UNIX/LINUX ftp://ftp.eas.slu.edu/pub/rbh/PROGRAMS.320.PC WIN32 What’s new? Wavenumber integration synthetics for TI media (Jun, 2002) First arrival times for isotropic TI media (Jun, 2002) surf96, rftn96, joint96 inversion code (Feb, 2002) Regional source mechanism programs (Aug, 2002) srfgrd96 - surface wave amplitude radiation pattern wvfgrd96 wvfmtd96 wvfmt96 - regional waveform inversion Three tutorials in PDF and PostScript. The Source Inversion tutorial discusses instrument deconvolution from SEED and AUTODRM sources, directory layout for storage of Green’s functions, program use, and an application to the June 18, 2002 Evansville, IN earthquake. Assessment of current package 114 programs - always a learning curve problem Powerful tool set. Surface wave radiation pattern fit -> source mechanism -> revised earth model -> new Green’s functions What’s next? Regional surface waves for TI media (Jan, 2003) CALPLOT tutorial (Jan, 2003) Generalized ray for TI media (Feb, 2003?) More case studies - Yunnan? Extend inversion codes - update tutorials Korea This study focuses on an area of interest to the program. The senior PI has spent a total of 5 weeks during the past three summers establishing working relationships with researchers at Seoul National University, KMA and KIGAM. The visits followed a major upgrade in seismic monitoring in the Republic of Korea in the last five years, and begins to address the question about what can be learned from new data sets in a region of low seismicity. Initial research focuses on earth structure from surface-wave and receive function observations. Figure 1 shows the locations of the broadband stations used. Teleseismic P-wave data were acquired for receiver function analysis. The source regions were Indonesia/Philippines, India/Afghanistan and the Aleutians, with most data from the south and west. Waveforms were inspected, rotated and radial component receiver functions computed using the Ligorria and Ammon (1999) iterative time-domain deconvolution technique with filter parameter = 1.0 and 2.5. The receiver functions are simple with a sharp converted phase. Inversions using rftn96 yielded a simple crustal model A test was made using the SNU data whether the models from a simultaneous inversion of many receiver functions or a stack, using high quality deconvolutions, would yield the same model. The answer was yes! The stacked values were used to speed the inversion and because the ray parameters did no vary mch in the data set. Figure 2 shows the stacked receiver functions for each station an filter parameter. A joint inversion of receiver functions and surface wave dispersion was performed using the program joint96 for each station. The surface-wave dispersion used was the same for each station, which is not appropriate for ULL, and consisted of Stevens (2001) group velocities in the period range of sec, and some phase velocities derived from a phase-velocity stack of a few teleseisms crossing the peninsula. The starting model is a uniform halfspace with Vp-8.0 and Vs=4.7 km/sec. The inversion kept Poisson’s ratio fixed. Figure 3 shows the model fit to the observations. The fit is excellent except at SOG/SGP and ULL. Figure 4 presents the models. The variability of models is real and reflects the fine detail in the = 2.5 receiver functions. KWJ and SEO both have large numbers of traces used in the stack, but have different Moho’s. DAG, SEO, SES, SNU, SOS, TEJ, TJN and ULJ seem to have sharp Moho discontinuites while other stations have a more transitional Moho. There is always the problem of uniqueness. Figure 5 shows the model predicted P-wave first arrival times in red. The blue lines are from a study by Song and Lee (2001) who used the VELEST program to analyze a small data set of arrival times from the old KMA analog network. The red curves parallel their blue curve except at short distances, which are sensitive to the shallow velocity structure. To find some independent evidence of shallow structure, one blast was identified in the KMA data set. The do_mft analysis is shown in Figure 6. The short period dispersion is compared to the CHU model prediction in Figure 7. There seems to be support for lower crustal velocities near the surface in this region of Archaen and Proterozoic surface geology. Korea Summary The results of the receive function analysis are similar to those of Kim and Lee (2001) who used many recordings in common with this study. The differences are that we used a time-domain deconvolution while they used a frequency domain water-level deconvolution, that we started with an unbiased halfspace model to permit the data to define the Moho location and degree of sharpness, and we used a joint inversion. Do we believe the velocity models - not yet! We hope to do the following Apply the VELEST analysis to the modern KMA observations from the past 4 years. Even though the annual umber of earthquakes is small, the large number of broadband, short-period and strong motion stations should yield a significantly better data set than that used by Song and Lee (2001). Use the new VELEST models to predict surface-focus first arrival times which would then be used in joint96-3. Waveform model any local or regional earthquake with M > 4 to test the models, These occur every years. Collect KMA and KIGAM data to provide better long period phase velocities for the peninsula. Learn more about the data set and confidence in the application o the Stevens (2001) dispersion predictions for the peninsula.