University of Southern California Dissertations and Theses

Systematic high -resolution imaging of fault zone structures

(USC Thesis Other)

Systematic high -resolution imaging of fault zone structures

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SYSTEMATIC HIGH-RESOLUTION IMAGING OF FAULT ZONE Copyright 2004 STRUCTURES b y Zhigang Peng A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment o f the Requirements for the Degree DOCTOR OF PHILOSOPHY (GEOLOGICAL SCIENCES) August 2004 Zhigang Peng Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3145264 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3145264 Copyright 2004 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ACKNOWLEDGEMENTS I would like to acknowledge first my advisor Yehuda Ben-Zion for his enthusiastic supervision throughout my studies. None of this work would have been possible without him. Many thanks to Ta-liang Teng for his continued encouragement and support. I also thank the committee members Tom Jordan, Andy Michael, and Jerry Mendel for their valuable guidance and critical comments. This work has benefited from various researchers. I am grateful to John Armbruster, Nano Seeber and Willie Lee for providing the data. Speicial thanks to Paul Silver for sharing his shear wave splitting code. Discussions with Rick Aster, Gotz Bokelmann, Elizabeth Cochran, Deborah Kilb, Fenglin Niu, David Okaya, Frank Vernon, John Vidale and Lupei Zhu were very stimulating and useful. Many people have assisted me in many ways during my work at USC. In particular, I would like to thank John McRaney, Cynthia Waite, Varduri Ter- Simonian and John Yu for their help on various administrative and computer related issues. I also would like to thank my fellow colleagues Yunfeng Liu, Youlin Chen, Shoshana Levin, Eric Libicki, Michael Lewis, Liangjun Chen, Po Chen and Li Zhao for their valuable discussions and suggestions. Financial support for my studies was provided by the National Science Foundation grant EAR0003401 and the Southern California Earthquake Center. Finally, I am forever indebted to my parents and my wife Xia Zhu for their understanding, endless patience and encouragement when it was most required. I dedicate my work to them. ii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TABLE OF CONTENTS ACKNOWLEDGEMENTS..................................................................................................ii LIST OF FIGURES................................................................................................................v LIST OF TABLES..................................................................................................................x ABSTRACT...........................................................................................................................xi INTRODUCTION.................................................................................................................. 1 CHAPTER 1 (Peng et al. 2003) QUANTITATIVE ANALYSIS OF SEISMIC FAULT ZONE WAVES IN THE RUPTURE ZONE OF THE 1992 LANDERS, CALIFORNIA, EARTHQUAKE: EVIDENCE FOR A SHALLOW TRAPPING STRUCTURE..........................................................................................................................4 SUMMARY........................................................................................................................ 4 1.1 INTRODUCTION.......................................................................................................5 1.2 ANALYSIS................................................................................................................... 8 1.2.1 Experiment and event location........................................................................... 8 1.2.2 Spatial distribution of events generating trapped waves..............................12 1.2.3 Travel time moveout analysis...........................................................................17 1.2.4 Dispersion analysis............................................................................................31 1.2.5 Synthetic waveform modeling of FZ trapped w aves................................... 38 1.3 DISCUSSION............................................................................................................ 46 CHAPTER 2 (Peng & Ben-Zion 2004a) SYSTEMATIC ANALYSIS OF CRUSTAL ANISOTROPY ALONG THE KARADERE-DUZCE BRANCH OF THE NORTH ANATOLIAN FAULT....................................................................... 51 SUMMARY.......................................................................................................................51 2.1 INTRODUCTION.....................................................................................................52 2.2 DATA...........................................................................................................................55 2.2.1 The Seismic Experiment................................................................................... 55 2.2.2 Data Selection.....................................................................................................59 2.2.3 Data Grouping.....................................................................................................59 2.3 ANALYSIS PROCEDURE..................................................................................... 60 2.4 RESULTS................................................................................................................... 79 2.4.1 Fast Polarization Direction................................................................................79 2.4.2 Possible Causes of Anisotropy......................................................................... 85 2.4.3 Delay Times and Depth Extent of Anisotropy.............................................. 90 2.5 DISCUSSION..........................................................................................................100 iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 3 (Peng & Ben-Zion 2004b) SPATIO-TEMPORAL VARIATIONS OF CRUSTAL ANISOTROPY FROM SIMILAR EVENTS IN AFTERSHOCKS OF THE 1999 M7.4 iZMIT AND M7.1 DUZCE, TURKEY, EARTHQUAKE SEQUENCES.......................................................................................105 SUMM ARY.................................................................................................................... 105 3.1 INTRODUCTION...................................................................................................106 3.2 DATA.........................................................................................................................108 3.3 ANALYSIS PROCEDURE................................................................................... 109 3.4 RESU LTS................................................................................................................. 117 3.4.1 Spatio-temporal evolutions of the Similar Event Clusters and Seismicity.................................................................................................................... 117 3.4.2 Fine-Scale Spatial Variations of Anisotropy................................................126 3.4.3 Apparent Temporal Variation of Anisotropy...............................................128 3.4.4 Fine-Scale Temporal Variations of Anisotropy.......................................... 136 3.5 DISCUSSIONS................................................:.......................................................144 RECAPITULATION..........................................................................................................147 REFERENCES.................................................................................................................. 150 APPENDIX A: Input Parameters for the Genetic Inversion Algorithm................... 162 APPENDIX B: README File for the Sliding Window Shear Wave Splitting Program................................................................................................................................. 165 iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF FIGURES Figure 1.1 Epicenters of 93 aftershocks (circles) of the 1992 Landers, California, earthquake (star) recorded by the SCSN.....................................................................9 Figure 1.2. Comparison of catalog locations (solid circles) of 67 events and locations produced by the grid-search method before (dash gray ellipses) and after (dark ellipses) station delay corrections..................................................11 Figure 1.3. Quality of trapped waves generation for 198 events located using the grid-search method and station delay corrections based on ratios of trapped waves energy divided by S-wave energy (inset)......................................................15 Figure 1.4. Quality of trapped waves generation for 60 events that have catalog locations..........................................................................................................................16 Figure 1.5. Fault-parallel seismograms recorded at the dense array stations for event 10161332............................................................................................................. 18 Figure 1.6. Normalized amplitude spectra versus position of stations across the FZ for event 10161332.................................................................................................19 Figure 1.7. Fault-parallel seismograms for event 10140034......................................... 20 Figure 1.8. Normalized amplitude spectra versus position of stations across the FZ for event 10140034................................................................................................ 21 Figure 1.9. Fault-parallel seismograms for event 10140247..........................................22 Figure 1.10. Normalized amplitude spectra versus position of stations across the FZ for event 10140247................................................................................................ 23 Figure 1.11. Synthetic seismograms generated by the 2D analytical solution of Ben-Zion & Aki (1990) and Ben-Zion (1998) for different propagation distances along the FZ................................................................................................. 24 Figure 1.12. A three-media model for a uniform low-velocity FZ structure in a half space (HS)..............................................................................................................25 Figure 1.13. (a) Fault-parallel seismograms at station E02 for 32 events with quality A of trapped waves generation, (b) Time differences between the S arrival and centers of trapped waves group versus hypocentral distances for the 32 events............................................................................................................29 v Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 1.14. Fault-parallel seismograms at FZ station E02 generated by 7 earthquakes north of the array with hypocentral depth larger than 6 km............ 31 Figure 1.15. (a) Synthetic FZ seismograms (left) filtered at different frequency bands using a zero-phase Gaussian filter and envelopes of filtered seismo grams calculated using Hilbert transform (right)....................................................33 Figure 1.16. (a) Comparison of analytical and numerical dispersion curves for different FZ parameters, (b) Comparison of analytical and numerical dispersion curves for different propagation distances along the FZ.................... 35 Figure 1.17. (a) Different frequency bands of a fault-parallel seismogram recorded at station E02 for event 10161206 (left) and envelopes of the band-pass-filtered waveforms (right), (b) Average dispersion curves measured from seismograms recorded at FZ stations W01-E05 for 8 events.... 37 Figure 1.18. Simultaneous synthetic (dark lines) waveform fits of 68 fault-parallel displacement seismograms (light lines) recorded by the 17 stations across the FZ and generated by 4 events north of the array....................41 Figure 1.19. Fitness values (dots) associated with different FZ parameters tested by the GIA..................................................................................................................... 42 Figure 1.20. Simultaneous synthetic (dark lines) waveform fits of displacement seismograms (light lines) recorded by the 17 stations across the FZ and generated by 4 events south of the array...................................................................44 Figure 1.21. Fitness values (dots) associated with different FZ parameters tested by the GIA..................................................................................................................... 45 Figure 2.1. (a) Hypocentral distribution of -26000 earthquakes recorded by the PASSCAL seismic experiment along the Karadere-Duzce branch of the NAF. (b) Distributions of seismic stations and hypocentral locations of -9200 earthquakes used for the shear wave splitting analysis in this study....... 57 Figure 2.2. Illustration of steps in the analysis of the shear wave splitting................ 64 Figure 2.3. Changes of the (a) smaller eigenvalue Xi, (b) < ( > and (c) 5t with the end of sliding time windows for the shear wave splitting measurement shown in Figure 2.2....................................................................................................................... 66 Figure 2.4. Values associated with the 9 criteria against 12 for 2818 measurements at station BV........................................................................................72 v i Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.5. (a) Histogram of initial polarization for all (black) measurements at station BV and results with low (gray) and high (white) quality, (b) Plot of initial polarization a versus fast direction < j ) for different quality of measurements, (c) Histogram o f the fast direction < j > . (d) Plot of the initial polarization a versus delay time 8t. (e) Histogram of the delay time 8t............. 74 Figure 2.6. Examples of original and rotated waveforms that are used to determine splitting parameters of “high” quality data............................................76 Figure 2.7. Rose diagrams of the fast directions ( j ) at all 17 stations.............................82 Figure 2.8. Rose diagrams of < j ) and equal area plots of splitting parameters (bars) at 17 stations for earthquakes belonging to the FZ group......................................83 Figure 2.9. Rose diagrams of ( j ) and equal area plots of splitting parameters (bars) at 17 stations for earthquakes belonging to the NFZ group...................................84 Figure 2.10. A summary plot of average splitting parameters (bars) in our study area.................................................................................................................................. 87 Figure 2.11. (a) Splitting parameters (bars) superimposed on the hypocentral locations within the shear wave window of the FZ station FP and projected along the cross-section AA' in Figure 2.1b. (b) Splitting parameters (bars) superimposed on the hypocentral locations within the shear wave window of station GE and projected along the cross-section AA' in Figure 2.1b.............88 Figure 2.12. Delay time versus depth at all stations for earthquakes that belong to the FZ group...................................................................................................................91 Figure 2.13. Delay time versus depth at all stations for earthquakes that belong to the NFZ group. Other symbols and notations are the same as in Figure 2.12................................................................................................................................. 92 Figure 2.14. Delay time versus hypocentral distance at all stations for earthquakes that belong to the FZ group. Other symbols and notations are the same as in Figure 2.12........................................................................................... 93 Figure 2.15. Delay time versus hypocentral distance at all stations for earthquakes that belong to the NFZ group...................................................................................... 94 Figure 2.16. Average delay times 81 for earthquakes that are (a) located close or near the FZ and (b) outside the FZ (Regional).........................................................97 Figure 2.17. (a) Splitting parameters (bars) for station BV superimposed on the hypocentral locations within the shear wave window, (b) Splitting ^.. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. parameters superimposed on the hypocentral locations projected along the cross-section BB' in (a) and Figure 2.1b...................................................................98 Figure 3.1. Flypocentral distribution o f -18000 earthquakes recorded by the PASSCAL seismic experiment along the Karadere-Dtizce branch of the NAF...............................................................................................................................109 Figure 3.2. Examples of waveform similarity for north component seismograms recorded at station FP.................................................................................................112 Figure 3.3. (a) The correlation coefficients (CC) for -1 7 T 0 6 north-component waveform pairs recorded at station FP versus their time lags. (b) Histogram of the CC. (c) Histogram of the time lags.....................................113 Figure 3.4. (a) The similarity measure p for -3 0 106 pairs of waveforms recorded at station FP versus their hypocentral separation D between pair of earthquakes, (b) Histogram of the similarity measure p. (c) Histogram of the hypocentral separation D. (d) Histogram of the number of recording stations between pairs of events............................................................................... 114 Figure 3.5. (a) Histogram of the similarity measure b between pair of events that are recorded by at least 3 stations between pairs of events and p 0.5. (b) Number of events that belong to a cluster (nev), number of events belonging to a cluster with 5 or more events (nev5), total number of similar event clusters (nclust), and number of clusters with at least 5 events (nclust5) as a function of the similarity criteria pc....................................115 Figure 3.6. A map view (a) and three cross-sections (b-d) showing repeating earthquake clusters (dark) with similarity criterion pc = 0.95 with respect to the overall seismicity along the Karadere-Dtizce branch of the NAF...........119 Figure 3.7. A map view (a) and three cross-sections (b-d) showing similar earthquake clusters (dark) with similarity criterion pc = 0.70 with respect to the overall seismicity along the Karadere-Dtizce branch of the NAF...........121 Figure 3.8. Locations of the earthquakes occurred before (a) and after (b) the Dtizce earthquake, (c) Z-value map of the seismicity rate change before and after the Diizce mainshock.................................................................................122 Figure 3.9. (a) Occurrence times of the repeating earthquake clusters with at least 5 events with similarity criterion Pc = 0.95. (b) Inverse of the time intervals between consecutive repeating events as a function of their occurrence time for 9 clusters near station BV...................... 125 viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.10. Average splitting parameters (bars) for the 9 long-term stations 129 Figure 3.11. Average splitting parameters (bars) for each of 9 clusters.....................131 Figure 3.12. (a) Splitting parameters (bars) for station BV superimposed on the hypocentral locations o f-700 earthquakes, (b) Delay times measured at station BV plotted against the earthquake occurrence times (Julian day since 1999). (c) Fast directions plotted against earthquake occurrence times..............................................................................................................................133 Figure 3.13. Delay times plotted against the earthquake occurrence times for 8 stations..........................................................................................................................135 Figure 3.14. Splitting measurements from the 10 events in cluster CIO for station FP......................................................................................................................137 Figure 3.15. Fast (a) and slow (b) component seismograms for 10 events in cluster CIO and recorded at station FP. The CC functions between the first and subsequent fast (c) and slow (d) seismograms....................................... 141 Figure 3.16. Relative delay times of the fast (a) and slow (b) waves as a function of time for 4 repeating earthquake clusters that are recorded by 5 stations, (c) The difference of the relative delays between the fast and slow component seismograms plotted against the earthquake occurrence time. (d) The changes of delay times between fast and slow shear waves in percent as a function of time..................................................................................... 142 ix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF TABLES Table 2.1. Station Locations and Shear wave splitting R esults....................................78 Table 3.1. Number of Clusters and Similar earthquakes............................................. 116 x Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT The subsurface structures of earthquake fault zones are investigated with systematic analysis of seismic fault zone (FZ) trapped waves and shear wave splitting in large waveform data sets. Contrary to previous claims on the existence of -100 m wide FZ seismic waveguide extending to the bottom o f the seismogenic zone (e.g., > 10 km), an objective and quantitative analysis of FZ waveform data collected in the rupture zone of the 1992 Landers, California earthquake suggests that the -100 m wide waveguide extends only to the depth of -3 km. Results from a systematic shear wave splitting study along the Karadere-Dtizce branch of the north Anatolian fault, which ruptured during the 1999 Mw7.4 Izmit, and Mw7.1 Dtizce, Turkey, earthquake sequences, indicate the existence of -1 km wide belt of strongly anisotropic rock around the FZ. The anisotropic layer is confined primarily to the same depth extent (-3 km) of the narrower (-100 m) seismic trapping structure. These results indicate that major strike-slip faults tend to have multiple near vertical FZ layers. The shallow structures responsible for generating both FZ trapped waves and shear wave splitting effects may be related to the top part of a flower-type structure of the FZ that contains highly damaged materials with intense microcracks and resides above the active portion o f seismogenic crust where earthquakes nucleate. The spatio-temporal variations of crustal anisotropy are investigated from similar earthquake clusters that are identified using a waveform cross-correlation technique. Splitting parameters averaged within each cluster show significant variations for different propagation paths, indicating strong spatial variations and xi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. multiple mechanisms in the study area. Apparent temporal changes of up to 30% of splitting delay times is observed at stations near the epicentral region of the Diizce mainshock. However, the changes can be mostly explained by the spatial variations of propagation paths due to the changing seismicity, instead of changes in properties of the anisotropic medium. Delay times measured within similar earthquake clusters indicate at most 2% co-seismic changes associated with the occurrence of the Diizce earthquake. The results do not show systematic precursory changes before the Diizce mainshock. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INTRODUCTION Natural earthquakes occur on geological faults. Field observations of exhumed faults and geophysical imaging of subsurface fault structures show zones of damaged materials with intense microcracks that have lower seismic velocity and larger attenuation than the surrounding rocks (e.g. Ben-Zion & Sammis 2003, and references therein). An accurate determination of the fault zone (FZ) properties at seismogenic depth can improve the understanding of many aspects of earthquake physics, long-term evolution of faults, seismic wave propagation, near fault ground motion and seismic hazard. If the low-velocity FZ is spatially coherent, it acts as a seismic waveguide and produce after the direct S wave arrivals large-amplitude and low-frequency wavetrains that are termed FZ trapped waves (e.g. Ben-Zion & Aki 1990; Li et al. 1994). Seismic shear waves propagating inside the damaged FZ rocks are expected to split into two orthogonally polarized waves with different velocities, mainly due to the presence of aligned microcracks around the active faults. This phenomenon is termed shear wave splitting and is analogous to optical birefringence. These signals, in turn, can be used as effective tools to image internal structures and properties of the FZ at seismogenic depth. This thesis deals with systematic analysis of large seismic waveform data sets for high-resolution imaging of subsurface FZ structures. In Chapter 1 (Peng et al. 2003), a quantitative analysis of seismic FZ trapped waves is performed on a waveform data collected in the rupture zone of the 1992 Landers, California 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. earthquake (Lee 1999). Previous studies (e.g. Li et al. 1994) indicate that FZ trapped waves are generated by -100 m wide FZ layers that extend to the bottom of the seismogenic zone (e.g 10 km). However, results from Chapter 1 and other related studies (e.g. Ben-Zion et al. 2003; Lewis et al. 2004) suggest a shallow seismic waveguide that extends only to a depth of ~3 km. The inferred shallow trapping structure may be common and relate to the top part o f a flower-type structure of the FZ that is highly damaged and aseismic. This significantly increases the seismic shaking hazard near the FZ, since seismic sources external to the FZ are able to generate strong motion amplification in the shallow trapping structure (Ben-Zion et al. 2003). In chapter 2 (Peng & Ben-Zion 2004a), the overall patterns of crustal anisotropy along and around the Karadere-Dtizce branch o f the north Anatolian fault (NAF) is investigated in detail. In this work a systematic shear wave splitting analysis is conducted on a waveform data produced by seismicity occurring in a 6- month period during the 1999 Mw7.4 Izmit and Mw7.1 Diizce, Turkey, earthquake sequences (Seeber et al. 2000; Ben-Zion et al. 2003). Results from -6600 high quality splitting measurements indicate the existence of a ~1 km wide belt of strongly anisotropic rock around the Karadere-Dtizce faults. The shear wave splitting delay time does not exhibit a clear dependency with increasing depth and hypocentral distance, suggesting that anisotropy rock is confined primarily to the same depth extent (top 3-4 km) of the narrower (~100 m) trapping structure (Ben- Zion et al. 2003). 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3 (Peng & Ben-Zion 2004b) focuses on the fine-scale spatio- temporal variations of crustal anisotropy around the Karadere-Dtizce branch of the NAF from similar earthquakes identified using a waveform cross-correlation technique. Depending on the applied similarity criteria, about 4-60% o f-18,000 microearthquakes belong to similar event clusters. Splitting measurements averaged within each cluster show that crustal anisotropy can vary significantly for slightly different ray paths. This indicates strong spatial variations of the anisotropic structures and multiple mechanisms in the study area. Apparent temporal change of up to 30% of splitting delay times is observed at stations near the epicentral regions o f the Diizce earthquake. However, the change can be explained by spatial variations of ray paths, instead of temporal change of the anisotropic medium. Delay times measured within several similar earthquake clusters indicate at most 2% co-seismic changes associated with the occurrence of the Diizce earthquake. 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 1 (Peng e t a l 2003) QUANTITATIVE ANALYSIS OF SEISMIC FAULT ZONE WAVES IN THE RUPTURE ZONE OF THE 1992 LANDERS, CALIFORNIA, EARTHQUAKE: EVIDENCE FOR A SHALLOW TRAPPING STRUCTURE SUMMARY We analyze quantitatively a waveform data set of 238 earthquakes recorded by a dense seismic array across and along the rupture zone of the 1992 Landers earthquake. A grid-search method with station delay corrections is used to locate events that do not have catalog locations. The quality of fault zone trapped waves generated by each event is determined from the ratios of seismic energy in time windows corresponding to trapped waves and direct S waves at stations close to and off the fault zone. About 70% of the events with S -P times less than 2 s, including many clearly off the fault, produce considerable trapped waves energy. This distribution is in marked contrast with previous claims that trapped waves are generated only by sources close to or inside the Landers rupture zone. The time difference between the S arrival and trapped waves group does not grow systematically with increasing hypocentral distance and depth. The dispersion measured from the trapped waves is weak. These results imply that the seismic trapping structure at the Landers rupture zone is shallow and does not extend continuously along-strike more than a few km. Synthetic waveform modeling indicates that the fault zone waveguide has depth of about 2-4 km, width of about Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 200 m, S-wave velocity reduction relative to the host rock of about 30-40%, and S- wave attenuation coefficient of about 20-30. The fault zone waveguide north of the array appears to be shallower and weaker than that south of the array. The waveform modeling also indicates that the seismic trapping structure below the array is centered about 100 m east of the surface break. 1.1 INTRODUCTION Major crustal faults are often marked by narrow tabular or wedge-shaped low-velocity zones. An accurate determination of the fault zone (FZ) properties at depth can improve the understanding of earthquake processes and parameters, long term evolution of faults, and more (e.g., Aki & Richards 2002; Scholz 2002; Sibson 2002; Ben-Zion & Sammis 2003). Measurements associated with inactive exhumed fault zones (e.g., Chester & Chester 1998; Evans et al. 2000; Faulkner et al. 2003) and surface ruptures of active faults (e.g., Sieh et al. 1993; Johnson et al. 1994, 1997) give direct information on FZ properties. However, these studies are limited to structures presently at surface. Various indirect geophysical methods such as gravity, electromagnetic surveys, reflection/refraction seismology, and travel time tomography have been used to image FZ structures at depth (Mooney & Ginzburg 1986; Ben-Zion & Sammis 2003, and references therein). Recently Fialko et al. (2002) inferred from, InSAR observations of surface deformation near the rupture zone of the 1999 Hector Mine earthquake, on the existence of belts of damaged FZ rock that are kilometers in width. In general, these techniques can only resolve Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. blurred versions of the true subsurface FZ structures. Waveform modeling of FZ trapped waves can provide a high resolution imaging of coherent low-velocity FZ layers at depth. FZ trapped waves follow the direct body wave arrivals and are large-amplitude, low-frequency, dispersive wave trains that are produced by constructive interference of critically reflected waves inside low-velocity FZ layers. During the last decade, Li and coworkers argued, based on analysis of small waveform data sets in several places, for the existence of ~100 m wide FZ layers that extend to the bottom of the seismogenic zone (e.g., > 10 km). Locations for which such claims were made include the Parkfield segment of the San Andreas fault (Li & Leary 1990), the Anza segment of the San Jacinto fault (Li & Vernon 2001), and the rupture zones of the 1992 Landers earthquake (Li et al. 1994a,b, 2000), the 1995 Kobe earthquake (Li et al. 1998), and the 1999 Hector Mine earthquake (Li et al. 2002). On the other hand, analyses of large data sets associated with the Karadere-Duzce branch of the North Anatolian fault (Ben-Zion et al. 2003) and the Parkfield segment of the San Andreas fault (Michael & Ben-Zion 1998; Korneev et al. 2003) indicate that the trapping structures in those locations are relatively shallow (e.g., ~3 km) FZ layers that are largely above the depth sections with active seismicity. Igel et al. (2002), Jahnke et al. (2002), and Fohrmann et al. (2003) showed with 3D calculations that sources well outside and below shallow FZ layers can produce ample trapped waves energy at stations close to the FZ. In contrast, the generation of trapped waves in a low-velocity FZ layer that is continuous with depth 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. requires sources that are inside or very close to the low-velocity structure. Thus, observations of FZ trapped waves due to sources well outside the fault imply that the trapping structure is shallow. Ben-Zion et a l (2003) referred to trapped waves (motion amplification and long period oscillations) in FZ stations due to sources not necessarily in the fault as “FZ-related site effects”. In this paper we analyze a waveform data set (Lee 1999) produced by 238 aftershocks of the 1992 Landers earthquake and recorded by a dense seismic array across the Landers rupture zone. Seismograms generated by some events in our data set have been analyzed earlier by Li et al. (1994a,b), who concluded on the existence of a low-velocity FZ waveguide that extends continuously to the bottom of the seismogenic zone. In contrast, our analysis indicates that the seismic trapping structure at the Landers rupture zone extends only to a depth of about 2-4 km. Our conclusion is based on spatial distributions of events that produce FZ-related site effects, travel time moveout of body 5 and trapped waves, dispersion analysis, and synthetic waveform modeling of FZ waves. The waveform modeling indicates further that the shallow trapping structure has effective width of about 200 m with a center about 100 m east of the surface break below the array, 5-wave velocity decrease of about 30-40% relative to the host rock, and 5-wave attenuation coefficient of about 20-30. The waveguide north of the array is less pronounced (e.g., smaller velocity contrast, narrower FZ width) than that south of the array. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.2 ANALYSIS 1.2.1 Experiment and event location A dense seismic FZ array was deployed across and along the rupture zone of the 1992 Landers, California, Mw=1.3 earthquake to observe FZ trapped waves (Li et al. 1994a,b; Lee 1999). The geometry of the array is shown in the inset of Figure 1.1. It consisted of an east-west line along the Encantado road crossing the rupture zone northwest of Landers and two north-south lines. The east-west line included 22 three-component, short-period L-22 seismometers with instrument spacing 25 m within 200 m of the surface break and 50-100 m further away. In this work we analyze systematically a seismic waveform data set generated by 238 aftershocks in the period October 14 - 17, 1992, and recorded by the dense FZ array. A much larger data set was recorded by Li et al. (1994a) but has not been released in a form available for analysis. A subset of 93 earthquakes of our events was also recorded by the Caltech/USGS Southern California Seismic Network (SCSN). Figure 1.1 shows the locations of these 93 earthquakes based on the Richards-Dinger & Shearer (2000) relocated catalog. We developed a grid-search method augmented by station corrections to locate the events that were recorded only by the FZ array. The grid-search method uses accurately picked P arrivals and S -P times and determines locations by minimizing the L2 norm of travel time residuals between observed data and synthetic calculations. The latter are produced by a 1-D velocity model for the region near the 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Landers rupture zone (Hauksson et al. 1993). The source depth is also included in the grid-search. 34° 30' 34" 20' 34" 10’ Magnitude n o 4 km Depth (km) 0 0 0.2 0.4 ,..NW 4 liisatmajo mad 'H M . B B sw> z .m Landers Y ucca V alley 16*40' 116" 30 6° 20’ Figure 1.1 Epicenters of 93 aftershocks (circles) of the 1992 Landers, California, earthquake (star) recorded by the SCSN. The event magnitudes are in the range 0.5 and 3.1 and event depths range from 0 to 14 km. The lines indicate surface traces of the Johnson Valley fault (JVF), Kickapoo fault (KF), Homestead Valley fault (HVF), and Pinto Mountain fault (PMF). The inset shows the geometry of the dense seismic array around the Landers rupture zone. 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We first apply this method to locate a subset of 67 events that have catalog locations and S -P times less than 4.5 s (or hypocentral distances within about 35 km). Waveforms generated by earthquakes with hypocentral distance larger than 35 km usually have low signal-to-noise ratios in seismograms recorded by the FZ array and are ignored. Figure 1.2 shows the catalog locations (solid circles) of these 67 events and locations produced by the grid-search method (red ellipses). The size of ellipse marks the standard deviations o f horizontal location errors. As seen in the figure, most events are relocated by the grid-search method further away from the FZ. This can be partially explained by the existence of a low-velocity FZ and the use of a laterally-uniform 1-D velocity model in our method. For example, first arrivals at stations east of the array from events west of the FZ will be later than expected in a laterally homogenous model. Our grid-search method thus tends to put such events further to the west (away from the FZ) to satisfy their arrival times. To reduce the effects of lateral velocity variation on our location determinations, we apply corrections based on the residuals between the observed and synthetic travel times. As shown in the inset of Figure 1.2, the travel time residuals for events with back azimuth (BAZ) between 0 and 172 degrees (east) and BAZ between 172 and 360 degrees (west) of the FZ are quite different. We calculate two sets of station delays by averaging the travel time residuals for events west and east of the FZ, and apply these station delays to the synthetic calculations to relocate the events. The locations after incorporating the station delay corrections are shown as blue ellipses in Figure 1.2. The average horizontal and vertical differences 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. - 0,02 i i » ~ 0.0 * 0.00 - - 0.02 S » 3 > A ^ ^ D e n s e array 1 A « » 120 ISO 240 3 BAZ angle (deg) M b ' k m Yucca valley 116° 40' Figure 1.2. Comparison of catalog locations (solid circles) of 67 events and locations produced by the grid-search method before (gray dash ellipses) and after (dark ellipses) station delay corrections. Other symbols and notations are same as in Figure 1.1. The inset shows travel time residuals versus back azimuth (BAZ) for the 67 events. The small dots are color-coded by the value of the residuals with dark being negative and gray being positive. The two vertical lines with BAZ values of 0 and 172 degrees mark the boundaries of regions east and west of the FZ. The big triangles denote the averaged residuals for events east and west of the FZ. 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. between the obtained locations and the corresponding catalog locations of the 67 events are 3.0 km and 3.6 km, respectively. These values provide estimates of the location errors produced by our grid-search method together with station delay corrections, which we apply to get hypocentral parameters for the events that do not have catalog locations. 1.2.2 Spatial distribution of events generating trapped waves The spatial distribution of earthquakes producing FZ trapped waves at surface FZ stations provides first-order information on overall properties of the trapping structure. Previous studies used visual inspection to identify FZ trapped waves and determine the quality of their generation. Although straightforward, visual inspection is subjective and not efficient when dealing with a large data set having thousands of waveforms. Here we determine the quality o f FZ trapped waves generation from ratios of trapped waves energy to 5-wave energy at stations relatively close to and stations off the FZ. The employed procedure is as follows: The energy in a seismogram recorded at each station within a specified time window is approximated by summing the squares of velocity amplitudes and normalizing by the length of the time window (Fohrmann et al. 2003). The 5-wave window starts 0.1 sec before the 5 arrival and ends at the start of the trapped waves window. The boundary between the windows is determined by maximizing the resulting energy ratio using a shear body waveform length in the range 0.3-0.7 sec. Such a range excludes in our data set the trapped 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. waves, if they exist, and ensures that at least two cycles are included in the 5-wave window. The end of the trapped waves window is the time when the amplitude reduces back to that of the S wave. Examples of the employed time windows are shown in Figures 1.5, 1.7 and 1.9. We then divide the energy calculated for FZ trapped waves by that of the direct S wave to get the energy ratio for each seismogram. The average energy ratios for seismograms recorded at 13 stations (W02-E06, S01-N03) with clear trapped waves and within 400 m o f the FZ, and 13 stations (W11-W03, E07-E10) relatively far from the FZ are computed and named A R fz and A R o ff, respectively. Finally, our measure for quality of trapped waves generation is the ratio A R fz/A R off- We note that the 13 selected FZ stations are not symmetric with respect to the surface trace of the Landers rupture (or station COO) because of observed asymmetry of stations that record clear trapped waves. This is reflected in the contour maps of normalized amplitude spectra distribution versus station positions as illustrated in Figures 1.6 and 1.8, and the synthetic waveform modeling done in Section 1.2.5. Waveforms recorded at station SW2, SW1, NW1 and NW2 are not used in the calculation since these four stations were not in operation in the first two days of the experiment. Figures 1.3 and 1.4 show the locations, coded with quality of FZ trapped waves generation, of 198 events located by our grid-search and station corrections and a subset of 60 events that have catalog locations, respectively. The energy ratios against the S -P times for the 198 events are given in the inset o f Figure 1.3. Energy ratios are not calculated for events with clipped waveforms and S -P times more than 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.5 s. Energy ratios exceeding 4, between 2 and 4, and less than 2 are assigned, respectively, quality A, B and C of trapped waves generation. These choices suggest themselves from the distribution of the calculated ratios and are marked in Figure 1.3 with stars, triangles and circles, respectively. Using slightly different values of energy ratios will not affect our overall conclusion on the spatial distribution of events generating trapped waves. Several important observations can be made from the spatial distribution of earthquakes producing trapped waves. About 70% of nearby events with S -P time less than 2 s, including many clearly off the fault, generate FZ trapped waves with quality A or B. This distribution is in marked contrast with previous claims that trapped waves are generated only by sources close to or inside the Landers rupture zone (e.g., Li et al. 1994a,b, 2000). Furthermore, we find that about 30% of the events north of the intersection of the Johnson Valley fault with the Kickapoo fault also generate at the FZ array trapped waves with quality A or B. This suggests that the branching at the Kickapoo fault does not have a dominant effect on the generation of trapped waves by events north of it. As mentioned before, the existence of trapped waves due to sources outside the Landers rupture zone indicates that the trapping structure is shallow. The percentage of events generating FZ trapped waves with energy ratios larger than 2 (quality A or B) is compatible with that estimated by Fohrmann et al. (2003) using 3D finite difference calculations. 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34 " 20 ' 34“ 10' Energy ratio K H 5 t l 0 5 ' un *!*?>£■' m m y r f 101509 / m m m r n % 10140247 array 10161331 \ km 5 10 I 116“ 40’ 116* 30 ’ 116“ 20’ Figure 1.3. Quality of trapped waves generation for 198 events located using the grid-search method and station delay corrections based on ratios of trapped waves energy divided by S-wave energy (inset). Energy ratios larger than 4, between 2 and 4, and less than 2, are denoted by stars, triangles, and circles, respectively. About 70% of the events with S -P times less than 2 s (vertical line in the inset) generate FZ trapped waves with energy ratio exceeding 2. There are 34 events with quality A of trapped waves generation and 86 events with quality B. Dispersion curves measured from waveforms of the 8 events (stars) pointed by arrows are shown in Figure 1.17. The waveforms of these events are modeled in Figures 1.18 and 1.20. The event ID numbers consist of 2-digits month, 2-digits day, 2-digits hour, and 2-digits minute. 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34" 20 ' Energy ratio 8 - 2 -0 HI 140205- SOI51105^ ioisra^ / i t 14 0 2 4 7 10150912^ # ij^#Dense array !* ▲ 34" 10' km 5 10 10( 61332' 116* 40 ' 116" 30’ 116° 20’ Figure 1.4. Quality of trapped waves generation for 60 events that have catalog locations. There are 4 events with quality A of trapped waves generation and 26 events with quality B. The events pointed by arrows are used in later analysis. Figures 1.5-1.10 give representative fault-parallel seismograms associated with each quality category o f FZ trapped waves. As shown in Figures 1.5-1.8, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. waveforms recorded at the 13 stations close to the fault trace (marked with large bold fonts) have large-amplitude oscillations with relatively low frequency after the S arrivals. In contrast, such waveform characteristics are much weaker or absent at the 13 stations located further away from the FZ. Figure 1.6 gives a contour map of normalized amplitude spectra versus positions of 22 stations across the FZ for event 10161332 with a quality A o f trapped waves generation. The clear concentration of 4-6 Hz energy at stations W01-E05 is associated with the FZ trapped waves recorded (Figure 1.5) at stations close to the FZ. For events with a quality B o f trapped waves generation, there is still considerable low-frequency energy at stations close to the FZ (Figure 1.8). However, the spectral energy is more scattered compared with that of Figure 1.6. For events with a quality C of trapped waves generation, the discussed features of trapped waves recorded at the dense array are diffused and scattered in both the time histories (Figure 1.9) and amplitude spectra (Figure 1.10). 1.2.3 Travel time moveout analysis To place bounds on the depth extent of the structure generating FZ trapped waves at the Landers rupture zone, we examine the time delay between the direct S wave and trapped waves. The time difference, or moveout, between the S phase and trapped waves should increase with propagation distance in the low-velocity trapping structure. This is illustrated in Figure 1.11 with synthetic seismograms generated using the 2D analytical solution of Ben-Zion & Aki (1990) and Ben-Zion (1998) for 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Event 10161332 (ratio: 6.4 , Q: A) Depth: 10.8 km Fault-parallel seismograms Range. 16.2 km E10 s v 111 V FZTW - 1 0 1 2 Time (sec) Figure 1.5. Fault-parallel seismograms recorded at the dense array stations for event 10161332. The two short horizontal lines mark time windows for the S wave and FZ trapped waves used in the energy ratio calculations. Station names are marked on the traces with stations close to the FZ given in large bold font. The focal depth and range (hypocentral distance) of the event are marked in the top right comer. The event location is shown in Figure 1.4 and is assigned quality A of trapped waves generation. 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Stations 1 2 3 4 5 6 7 8 9 10 Frequency (Hz) Figure 1.6. Normalized amplitude spectra versus position of stations across the FZ for event 10161332. 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Event 10140034 (ratio: 3.8 , Q: B) Depth: 11.0 km Fault-parallel seismograms Range: 11.4 km jWW1 A W M y ^ FZTW _° T im e (sec) 1 Figure 1.7. Fault-parallel seismograms for event 10140034. The event location is shown in Figure 1.4 and is assigned quality B of trapped waves generation. Other symbols and notations are same as in Figure 1.5. 2 0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 2 3 4 5 6 7 8 0 10 Frequency (Hz) Figure 1.8. Normalized amplitude spectra versus position o f stations across the FZ for event 10140034. 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Event 10140247 (ratio: 1.9 , Q: C) Depth: 7.5 km Fault-parallel seismograms Range. 10.5 km %r ^ ^ aaVL/V y v/ I r A -wv v ~ » -'--v v'j \ \/^ l \ I \ /N vvv/V vvy /v V v'^ / w\. ■ '"V "—'-A/V 1 ■ Y i \ A A v A l < - . ■''/!/.— -v-.-Y ^V ,'V .-A rA /V 'Y Y '-vV 'yrY \A h > IV '- « W W 01 A A -a V W i/'w '/v Y V ' A A V A ; V A V a W F Z T W S 'JW v¥| ' 1 1 N04 A/'-rV~\/ / ^ / V A a t w / y 1 1 V \ v A j V l / v A v V '-rJ'^ v - ^ Time (sec) Figure 1.9. Fault-parallel seismograms for event 10140247. The event location is shown in Figure 1.4 and is assigned quality C of trapped waves generation. Other symbols and notations are same as in Figure 1.5. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 2 3 4 5 6 7 8 9 10 Frequency (Hz) Figure 1.10. Normalized amplitude spectra versus position o f stations across the FZ for event 10140247. 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Slope = 3 km/s Slope = 2 km/s 4 5 Time (sec) Figure 1.11. Synthetic seismograms generated by the 2D analytical solution of Ben- Zion & Aki (1990) and Ben-Zion (1998) for different propagation distances along the FZ. The two solid lines with slopes Phs and Pfz mark, respectively, the arrival time of the S phase and the end of the trapped waves group (defined as the time when the amplitude returns to that of the S arrival). The dashed line marks the center of the trapped waves group. Group velocities measured from the synthetic seismograms are shown in Figure 1.16. 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Free surface receivers A J A A ▲ A 'A . A / / F n u if ilait-space | /one Hnll-sp;.iee . i Q . / M S aflM M W fiM I / / Figure 1.12. A three-media model for a uniform low-velocity FZ structure in a half space (HS). The source is an SH line dislocation with coordinates (xs, zs). The width, shear attenuation coefficient and shear wave velocity of the FZ are marked by W, Q f z , and Pfz- Shear wave velocity and attenuation coefficient of the HS are denoted by Phs and Qhs- antiplane S waves in a half-space (HS) containing a low-velocity FZ layer (Figure 1.12). The S- wave velocity and attenuation coefficient of the HS are Phs = 3 km/s and Qhs = 1000. The corresponding material properties and width of the FZ layer are P f z = 2 km/s, QF z = 50, and W= 200 m. The source is an SH line dislocation with a unit step function in time and is located at position xs, zs- The synthetic calculations of Figure 1.11 are done for a source at the interface between the FZ and left block and a receiver on the free surface at the center o f the FZ layer. Figure 1.13(a) shows fault-parallel seismograms at FZ station E02 generated 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. by 32 earthquakes with S -P times less than 2 s that are assigned quality A of FZ trapped waves generation. The data are separated into two groups based on their locations north or south of the array. The time differences between the S arrivals and centers of trapped waves group for these seismograms are plotted in Figure 1.13(b) against hypocentral distances. Clearly, there is no persistent moveout between S wave and trapped waves group as the hypocentral distances increase. This implies that the propagation distance inside the low-velocity FZ layer is about the same for all the events. The average time delay for the events south o f the array is larger than that for events north of the array, suggesting different waveguide properties for the FZ south and north of the array. As discussed in Ben-Zion et al. (2003), the propagation distances of the trapped waves inside the low-velocity FZ material can be estimated from (11) P m -P v z where At is the time between the direct S arrival and center of the trapped waves group. If the hypocenters of the events generating FZ trapped waves are deep enough for the wavefield to sample the entire depth extent of the waveguide, Eq. (1.1) can be used to estimate the depth o f the waveguide. This was done by Ben-Zion et al. (2003) using cross-sections of events in the depth range 5-15 km around the Karadere-Duzce branch of the North Anatolian fault. In our case, some of the events used in Figure 1.13 have catalog depths that are shallower than 3 km, and most do not have catalog locations. 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. To estimate the depth extent of the waveguide using Eq. (1.1), we measure the time delay between direct S wave and trapped waves for events with quality A or B that have catalog locations and hypocenters deeper than 6 km. Figure 1.14 shows fault-parallel seismograms at FZ station E02 generated by 7 earthquakes located at different epicentral distances north of the array. The event locations are marked in Figure 1.4. The time delays between the direct S arrival and the center of the trapped waves group do not grow with increasing hypocentral distances. The average time delay is 0.39 s, similar to the 0.34 s obtained from the 14 events in Figure 1.13(b) north o f the array with quality A of trapped waves generation. As discussed in Section 1.2.5, synthetic waveform modeling of FZ waves generated by four events north of the array indicates that the average (or effective) S-wave velocities of the HS and FZ material are about 3.2 km/s and 2.3 km/s, respectively. Using these values together with At = 0.39 s in Eq. (1.1) gives zs of about 6.4 km. This value is smaller than the hypocentral depths of most events. Moreover, since the 7 events are located at considerable epicentral distances from the array, the actual propagation paths of the FZ trapped waves must include along-strike components. If we assume for simplicity that the average along-strike component and vertical component are the same, we get an estimated waveguide depth of about 4.5 km. The waveform modeling done in Section 1.2.5 suggests an upper bound o f about 3 km for the FZ waveguide north of the array. In the region south of the array, only 4 out of 9 events with catalog depth larger than 6 km produce trapped waves with quality A or B. Also, the travel time 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. data for the events south of the array have a large scatter with a possible bi-modal distribution (Figure 1,13b). We therefore do not use the travel time data to estimate the depth of the FZ waveguide south of the array. However, the waveform modeling of Section 1.2.5 suggests an upper bound for the FZ waveguide in that region of about 4 km. The results of this section imply that the trapping structure at the Landers rupture zone consist of relatively shallow low-velocity waveguide that is discontinuous along strike. The results are compatible with the spatial distribution of events generating trapped waves discussed in Section 1.2.2, and the dispersion analysis discussed next. 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 1.13. (a) Fault-parallel seismograms at station E02 for 32 events with quality A o f trapped waves generation. The waveforms are plotted against their hypocentral distances and aligned with P arrivals at time 0. The thin diagonal lines mark the S arrival time for each seismogram. The horizontal bars below and above the seismograms denote the approximate start and end of FZ trapped waves groups. The plus sign on top of each seismogram marks the center of the trapped waves group. The vertical dashed line marks the end of the trapped waves group, measured as mid position between the S arrival and the time when the amplitude reduces back to that of the S wave. The ID numbers of the earthquakes are given on the right (see explanation in the caption of Figure 1.3). Waveforms of the 8 events whose ID numbers are given in large bold font are modeled in Figure 1.18 and 1.20. Dispersion curves measured from the waveforms of these 8 events are shown in Figure 1.17(b). (b) Time differences between the S arrival and centers of trapped waves group versus hypocentral distances for the 32 events. Squares and circles denote the values for events south and north of the array, respectively. The symbols on the left with vertical lines give the mean and standard deviations of the time differences. The lack of systematic increase with hypocentral distance implies approximately constant propagation length in the FZ waveguide. 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. North * + ,.. 4, li/ 1 South 1 2 Time (sec) " t T o GO m 0 . 6 - 2 5 < D T 3 <D . 1 0 . 4 - 0 . 2 - 10160717 10150914 10)70659 10151456 ' '60642 10150912 60653 10161214 10141348 10160637 10140131 10141340 10160714 10151352 10160651 10150837 10141238 10170755 10171313 ■ 10160227 ■ 10171123 10151139 1015061)5 1 1 )1 * 0 -7 7 101WI5S.5 10163206 r /1 6 1 3 1 4 l i 1150417 II0 4 0 5 4 S 10140838 4 8 1 2 . Hypocentral distance (km) 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10160410 0.40s ' ! + I S S ; ! S I i S 10140205 0.41s ! V ' . Z1 A .M > .Kill in is O Q 'M f t i o c ft , ft , Depth: 10.67 km 10U U V Z 4 i / ; + » ■ ; 1 s Range: 20.73 km t 10151105 0 .36s , *U,. • .KSSIS 10150537 i 0.43 s ; + , S g f t , ! ? ) & , 10151231 0.34s I '; j • i g g o K i t o 10140034 0.35 s + , I S g i l ’ l S t a I ----------------------------- ,-------------------------------- p — -------------- , --------- p _ — —-------- ----------- ,----------- --------------- -— I ---------------- -2 - 1 0 1 2 Time (sec) Figure 1.14. Fault-parallel seismograms at FZ station E02 generated by 7 earthquakes north of the array with hypocentral depth larger than 6 km. The seismograms are aligned with S arrivals at time 0. The two vertical dashed lines mark the time of the P arrival and the end of the trapped waves group. The plus sign on top of each seismogram denotes the estimated center of the FZ trapped waves group. The time delay between the direct S arrival and the center of the trapped waves group are indicated above each trace. The ID numbers of the earthquakes are given on the left. The range (hypocentral distances) and focal depth of each event are marked in the top right side of each seismogram. The event locations are shown in Figure 1.4. 1.2.4 Dispersion analysis To study the dispersion o f FZ trapped waves, we measure group velocities from multiple band-pass-filtered seismograms using a zero-phase Gaussian filter. Before analyzing observed data, we describe the method and discuss trade-offs in 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. model parameters using synthetic calculations. Figure 1.15(a) shows filtered synthetic seismograms in 17 frequency bands of 0.5 Hz over the range 2 to 10 Hz. The material properties and FZ width used to generate the seismograms are Phs = 3 km/s, Qhs - 1000, pF z = 2 km/s, QF z ~ 1000 and W= 200 m. Here and in the following sections we fix the attenuation factor of the HS to be 1000. The propagation distance in the FZ layer is 5 km. The circles in the right panel mark the peaks o f the envelopes calculated by Hilbert transforms o f the band-pass-filtered seismograms. Each peak provides a measure for the arrival of energy at the specified frequency band. As expected, the trapped waves at lower frequencies travel faster than those at higher frequencies. Figure 1.15(b) provides a comparison between analytical and measured dispersion curves. The stars are group velocities measured from the filtered synthetic seismograms in Figure 1.15(a). The lines are generated by the analytical dispersion formula of Ben-Zion & Aki (1990) for a vertical FZ layer in a HS, - c - r 2] = 2^ ^ r f X y ^ f - ^ - ^ ’ ( 1 '2) M f z ( P f z c ) M h s(c P h s ) where W is the FZ width, c is phase velocity,/is frequency, and pH s and pF z are shear moduli of the HS and FZ layer, respectively. The results show that our procedure for measuring group velocities provides values that match well the 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) ^ if e v y V 'U ' — — ? WVV\A~ 3.S 4.5 5 5.5 6 6.5 7.5 8.5 9 9.5 10 " A --- ---------^vVvVv-— .. -— ™ — ---'-•W v'yV ''" ’'— — -Vi/iJV'"- - < !* •- 0 ! 2 3 4 5 Time (see) Time (sec) (b) Phase velocity Frequency (Hz) Figure 1.15. (a) Synthetic FZ seismograms (left) filtered at different frequency bands using a zero-phase Gaussian filter and envelopes o f filtered seismograms calculated using Hilbert transform (right). The peaks o f the envelopes (circles) indicate the arrivals of the energy at different frequency bands, (b) Comparison of analytical and numerical dispersion curves. Stars are group velocities measured from filtered synthetic seismograms. Lines are generated by the analytical dispersion formula of Ben-Zion & Aki (1990). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. analytic group velocity solution. Ben-Zion (1998) illustrated various trade-offs between model parameters with time-domain calculations. The following two examples illustrate similar trade offs in the frequency domain. Figure 1.16(a) shows comparisons of analytical and numerical dispersion curves for different FZ parameters. The measured group velocities for synthetic seismograms with Qfz = 1000 match well the analytic group velocity solution over most of the frequency range, and underestimate somewhat the analytic results at low frequencies. As Q decreases, the measured dispersion curves shift downwards and at Qfz = 10 the measured group velocities deviate over the entire frequency range from the analytic dispersion curves by about 0.2 km/s. The effect o f Q on the dispersion curves can also be produced by adjusting other FZ parameters. For example, if we increase the FZ width from 200 m to 350 m, or decrease the HS and FZ shear velocities from 3 km/s and 2 km/s to 2.8 km/s and 1.8 km/s, respectively, the analytic dispersion curves become close to the measured group velocities over most of the frequency ranges with the previous set of parameters and Qfz = 10. In Figure 16(b), the lines are generated by the analytical dispersion formula using Phs = 3 km/s, Pfz = 2 km/s, Qfz = 50 and W= 200 m. The different symbols represent group velocities measured from the synthetic seismograms with different propagation distances, ranging from 1 to 15 km, along the FZ. The measured dispersion is weak for propagation distances smaller than 4 km and improves with increasing distances. As illustrated in Figure 16(a), the downward shift of the 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) 3 k m 's. |3 t 2.8 ' ■ f— Phase velocity 2.6 Group velocity 2.4 >s £2.2 FZ = 1000 Frequency (Hz) 1 km 2 km 4 km 6 km 8 km 10 km 15 km * y \ t— Group velocity Frequency (Hz) Figure 1.16. (a) Comparison of analytical and numerical dispersion curves for different FZ parameters. The points are group velocities measured from synthetic seismograms generated with different Q values. Other FZ parameters are the same as those used to produce the synthetic seismograms of Figure 1.15(a). The lines are calculated by the analytical dispersion formula for various model parameters as indicated in the figure. The results illustrate trade-offs between FZ parameters in the frequency domain, (b) Comparison of analytical and numerical dispersion curves for different propagation distances along the FZ. The symbols mark group velocities measured from the synthetic seismograms of Figure 1.11 with different propagation distances. Lines are generated by the analytical dispersion formula of Ben-Zion & Aki (1990). The dispersion is poor for distances smaller than about 4 km and improves with increasing propagation distance. 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. measured dispersion curves at short propagation distance can be produced also by adjusting other FZ parameters properly. Figure 1.17a illustrates a dispersion analysis on observed fault-parallel seismogram recorded at station E02 for event 10161206. The seismogram is first windowed 1 s before and 4 s after the S arrival. After applying a cosine taper with 5% of the entire width to both ends, we filter the waveform into 16 frequency bands ranging from 1.5 to 6 Hz with a 0.3 Hz interval. Figure 1.17(b) shows the averaged dispersion curves measured from observed seismograms generated by 8 events, 4 (10151352, 10150912, 10150914 and 10160717) north and 4 (10140848, 10150605, 10151139 and 10161206) south of the array. These eight events were selected based on their high signal-to-noise ratio waveforms and large quality values (above 5.5) of FZ trapped waves generation. The dispersion curves measured for events located north of the array are flatter than those located south o f the array, suggesting that the velocity contrast, depth extent, and other properties of the waveguide vary along the FZ. The dispersion measured from the observed data is in general rather weak, indicating a short propagation distances inside the low-velocity FZ material. The results again imply that the trapping of seismic energy in the Landers rupture zone is generated by a shallow FZ layer. The trade-offs among FZ parameters illustrated in Figure 1.16 imply that results based on dispersion of FZ trapped waves do not provide strong constraints on the parameters of the velocity structure. 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5.4 5.7_ 6 - - (b) ! 2 .5 4 5 Time (sec) 3.5 £ .S' §2.5 > 2 3 4 Time (sec) i . , 1(3140838 C y f c , , . 7 all 10160717 . 10150605 I ▼'I j f r t r J p l j 10150914 10151139 y j j j p f in-* 10150912 .10161206 f O v T W 1 'T T T 4! 1 ^ . , , 1.0151352 .‘I p l 10° 1,0’ Frequency (Hz) Figure 1.17. (a) Different frequency bands of a fault-parallel seismogram recorded at station E02 for event 10161206 (left) and envelopes of the band-pass-filtered waveforms (right), (b) Average dispersion curves measured from seismograms recorded at FZ stations W01-E05 for 8 events with ID numbers marked in the figure. The locations o f the events are marked in Figure 1.3. Squares and circles denote the values for events south and north of the array, respectively. Each point gives the average group velocities in the specified frequency band measured from seismograms recorded at the 6 stations (W01-E05) that are close to the fault trace. The error bar at each point is the standard deviation of the result. Waveforms of these events are modeled in Figures 1.18 and 1.20. 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.2.5 Synthetic waveform modeling of FZ trapped waves In this section we model portions of observed FZ seismograms with trapped waves using the 2D analytical solution of Ben-Zion & Aki (1990) and Ben-Zion (1998) for a plane-parallel layered FZ structure (Figure 1.2). The model parameters include: (a) seismic velocities, attenuation coefficients and width o f the FZ layer; (b) seismic properties of the bounding blocks; and (c) source and receiver positions with respect to the fault and the free surface. As discussed by Ben-Zion et al. (2003), the 2D analytical solution provides a proper modeling tool of trapped waves in FZ sections with width much smaller than the length and depth dimensions, and much larger than correlation lengths of internal material and geometrical heterogeneities. Igel et al. (1997) and Jahnke et al. (2002) showed with 3D numerical calculations of wave propagation in irregular FZ structures that trapped waves are not sensitive to plausible velocity gradients with depth, gradual FZ boundaries, small-scale scatters, and other types of smooth or small heterogeneities. In general, FZ trapped waves average out small internal 3D variations and provide information on effective uniform waveguide properties over the observed range of wavelengths. Since trapped waves give the resonance response of the FZ structure after the transient source effects, the response to a line dislocation source can be converted accurately to an equivalent response to a point source by deconvolving the synthetic seismograms with 1 N t (e.g., Vidale et al. 1985; Crase et al. 1990; Igel et al. 2002; Ben-Zion et al. 2003). As illustrated in Figures 1.18 and 1.20, the 2D analytical solution provides very good waveform fits to the observed FZ trapped waves. 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ben-Zion (1998) emphasized that there are significant non-orthogonal trade offs between the effective 2D FZ parameters. The number N of internal reflections in the low-velocity layer controls the overall properties of the resulting interference patterns and trapped waves. This number depends on the FZ width, source-receiver distance and velocity contrast as N = -----^ ------, (1.3) W tan(#c) where zs is the propagation distance in the FZ layer and 9C = suT^Pfz/Phs) is the critical reflection angle at the interface between the FZ layer and HS. Other FZ parameters, such as source and receiver positions and attenuation coefficients of the FZ and HS media, also play important roles in modifying observed features of the resulting trapped waves. To model the data with a method that accounts quantitatively for the trade-offs, we use a genetic inversion algorithm (GIA) that employs the 2D analytical solution as a forward kernel (Michael & Ben-Zion, 1998). The inversion maximizes the correlations between observed and synthetic waveforms while performing a systematic and objective search of the relevant parameter-space. In this study and our related works in the Parkfield section of the San Andreas fault (Michael & Ben-Zion, 1998) and Karadere-Duzce branch of the North Anatolian fault (Ben-Zion et al. 2003), a single uniform FZ layer in a HS (Figure 1.12) is sufficient to produce very good waveform fits to the observed data. (See Appendix A for a complete description o f the GIA code and parameter setup.) Figure 1.18 shows synthetic waveform fits (dark lines) o f 68 fault-parallel displacement seismograms (gray lines) recorded by the 17 stations across the 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Landers rupture zone for events 10150912, 10151352, 10150914 and 10160717 north o f the array. Before inversions, we remove from the data the mean and instrument response and convolve the seismograms with 1/Vtto get equivalent 2D line-source seismograms. The GIA calculates fitness values associated with different sets of model parameters. The fitness is defined as (l+ Q /2 , where C is the cross correlation coefficient between the observed and synthetic waveforms. The synthetic waveform fits o f Figure 1.18 were generated using the best-fitting parameters associated with the highest fitness value during 10000 inversion iterations. We note that the waveform fits at stations relatively off the FZ are less satisfactory than at stations near the FZ, and that the onsets of the synthetic S body waves do not always fit well the observed onsets. These discrepancies are associated with the fact that the inversion method gives higher weight to phases with larger amplitudes, i.e., the trapped waves at the stations near the FZ. Figure 1.19 shows fitness values (dots) calculated by the GIA for the final 2000 iterations. The best-fitting values (solid circles) are Pfz = 2.3 km/s, Phs = 3.2 km/s, W = 210 m, Q fz = 15 and zs = 2.9 km, 3.8 km, 3.8 km and 3.7 km. We obtain very good simultaneous fits to waveforms generated by four events with different locations using very similar propagation distances (of about 3-4 km) along the FZ. This suggests again that the trapping structure is shallow and does not extend continuously from the array location along strike over a distance larger than a few km. The lines in Figure 1.19 give probability density functions (PDFs) for the various model parameters, calculated by summing the fitness values and normalizing 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Event 10150912 Event 10150914 y \ A s / a A - v V ^ w Na - 2 3 Time (see) Event 10151352 4 3 4 5 Time (sec) Event 10160717 0 1 2 Time (sec) 3 3 4 5 Time (sec) Figure 1.18. Simultaneous synthetic (dark lines) waveform fits o f 68 fault-parallel displacement seismograms (light lines) recorded by the 17 stations across the FZ and generated by 4 events north of the array. The locations of the events are marked in Figure 1.3. 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X q su sp ^ Q \ vo C\ <o o 3 S | i ^ S9tl|BA S S 9 U |iq Figure 1.19. Fitness values (dots) associated with different FZ parameters tested by the GIA. The model parameters associated with the highest fitness values (solid circles) were used to generate the synthetic waveforms in (a). The curves give probability density functions for the various model parameters. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the results to have unit sums (Ben-Zion et al. 2003). The peaks in the PDFs provide another possible set of preferred model parameters. The peak probability values of the propagation distances in the FZ layer are also similar to each other and in the range of about 3-4 km. The modeling indicates further that the waveguide below the array is not centered at the exposed fault trace (station COO), but at a distance of about 100 m east of station COO. This is compatible with contour maps of normalized amplitude spectra distribution versus station positions o f the types shown in Figures 1.6- 1. 8 . Figure 1.20 shows synthetic waveform fits of the GIA to 68 fault-parallel displacement seismograms generated by events 10140838, 10150605, 10151139 and 10161206 south of the array. The synthetic waveforms were produced using the best- fitting parameters given in Figure 1.21. The best-fitting values are Pfz = 2.0 km/s, Phs = 2.8 km/s, W= 230 m, Q fz = 27 andzs = 4.1 km, 4.9 km, 4.6 km and 4.1 km. As before, we obtain very good simultaneous fits to waveform generated by earthquakes with different locations using similar propagation distances (of about 4- 5 km) within the waveguide. Since the 8 events used in the synthetic waveform fits of Figures 1.18 and 1.20 are not located directly underneath the array, the propagation paths of the FZ trapped waves include along-strike components. Assuming as was done in Section 2.4 that the average along-strike and vertical components are similar, we get estimated waveguide depth below the surface rupture o f Landers earthquake of about 2-3 km north of the array and 3-4 km south of it. We also note that the best-fitting 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. E1 0 ,------------- 1 ------------- ,-------------,-------------j ------------- j |------------- ,------------- ;........ ........-i------------- j ------------- '-------------; ■ 1 2 3 4 1 2 3 4 Time (sec) Time (sec) Event 10151139 Event 10161206 ; -------- ,-------- , -, — ----------------- , -------- j -------- ,-------- r 1 2 3 4 2 3 4 Time (sec) Time (sec) Figure 1.20. Simultaneous synthetic (dark lines) waveform fits of displacement seismograms (light lines) recorded by the 17 stations across the FZ and generated by 4 events south of the array. 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. o o o o o O ^ § w -> f * : ■ ■ J S 1 00 O c c o o © o o o o O o a N u- b ~ - © 00 © 00 © 00 d © © 3 o ssn|B A s s a m i j Figure 1.21. Fitness values (dots) associated with different FZ parameters tested by the GIA. The model parameters associated with the highest fitness values (solid circles) were used to generate the synthetic waveforms in (a). The curves show probability densities for the various model parameters. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. values for the waveguide north and south of the array are different. These results, together with the relatively flat dispersion curves as shown in Figure 1.19, suggest that the waveguide north of the array is somewhat shallower and weaker than that south of the array. As discussed in the context of our work on the North Anatolian fault (Ben- Zion et al. 2003), we can obtain very good fits between synthetic and observed waveforms for a wide range of parameters due to the strong trade-offs between parameters (Ben-Zion 1998). It is thus important to use independent constraints on parameter values if such are available. The inversions leading to the results of Figures 1.19 and 1.21 were done assuming that the FZ width is in the range 150-250 m, in agreement with field observations (Johnson et al. 1994, 1997; Li et al. 1994a,b; Rockwell et al. 2000) on the width of the surface rupture zone of the Landers earthquake in our study area. We can produce good waveform fits for larger propagation distance inside the waveguide than those o f Figures 1.19 and 1.21, but this tends to increase the FZ width beyond the observed «200 m in-situ value. 1.3 DISCUSSION We perform a comprehensive analysis of a waveform data set generated by 238 aftershocks and recorded by a dense seismic array across and along the rupture zone of the 1992 Landers earthquake. Events recorded only by the dense array are located by a grid-search and station corrections method (Figures 1.2, 1.3). Based on the ratio of trapped waves to S'-wave energy, we assign a quality A, B, or C of 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. trapped waves generation to 198 events (inset o f Figure 1.3). About 70% o f nearby events with S -P time less than 2 s, including many clearly off the fault, generate FZ trapped waves with quality A or B (Figure 1.3). This spatial distribution differs from previous claims (e.g., Li et al. 1994a,b, 2000) that trapped waves at the Landers rupture zone are generated only by sources very close to or inside the FZ. Igel et al. (2002), Jahnke et al. (2002), and Fohrmann et al. (2003) demonstrated that a shallow FZ layer can trap seismic energy generated by events that are deeper and well outside it, while generation of trapped waves in a deep and coherent FZ layer requires the source to be close or inside the FZ. The existence o f trapped waves due to sources outside the rupture zone of the Landers earthquake implies that the generating structure is shallow. This statement is supported further by travel-time data of S and trapped waves (Figures 1.13 and 1.14), dispersion analysis (Figure 1.17), and synthetic waveform modeling (Figures 1.18-1.21). We could model all the waveforms generated by the 34 events that produce trapped waves with quality A. However, this will not increase significantly the imaging resolution because of the relatively short propagation distances inside the FZ waveguide and the trade-offs between model parameters that are reflected in the parameter-space plots (Figures 1.19, 1.21). We thus provide quantitative waveform fits only for 136 waveforms generated by the 8 events used in the dispersion analysis. Since clear trapped waves are not recorded at stations W11-W07, we use in the inversions only waveforms recorded by 17 (W06-E10) out of 22 stations o f the east- west FZ array. 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The synthetic waveform modeling indicates that the FZ waveguide has a depth o f about 2-4 km, width on the order of 200 m, 5-wave velocity reduction relative to the host rock of about 30-40%, and 5-wave attenuation coefficient of about 20-30. The modeling also shows that the waveguide below the array is not centered at the exposed fault trace (station COO), but at a distance o f about 100 m east o f station COO. The waveform modeling and dispersion analysis suggest that the waveguide north of the array is possibly shallower and weaker than that south of the array. The travel time analysis also suggests that the FZ waveguide in our study area is not continuous along strike more than a few km. Shallow trapping structures with similar properties appear to characterize the Karadere-Duzce branch of the North Anatolian fault (Ben-Zion et al. 2003), the Parkfield segment of the San Andreas fault (Michael & Ben-Zion 1998; Komeev et al. 2003) and the Anza segment of the San Jacinto fault (Lewis et al. 2003). Shallow layers o f damaged FZ rock acting as seismic waveguides can exist not only in active structures but also (Rovelli et al. 2002; Cultrera et al. 2003) in dormant fault zones. Ben-Zion et al. (2003) suggested that shallow trapping structures are a common element o f fault zones and may correspond to the top part of a flower-type structure. Since the volume of sources capable of generating motion amplification in shallow FZ waveguides is large, the existence of such structures increases the seismic shaking hazard near faults (Spudich & Olsen 2001; Ben-Zion et al. 2003). Our results indicate that about 70% of the events with S —P time less than 2 s are able to generate trapped wave energy at the Landers rupture zone exceeding the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. S-wave energy by a factor 2 or more (quality A or B). The source volume percentage is comparable to that estimated by Fohrmann et al. (2003) using 3D finite-difference calculations, but smaller than that observed by Ben-Zion et al. (2003) along the Karadere-Duzce branch of the North Anatolian fault. Possible explanations for the more abundant generation of trapped waves in the Karadere-Duzce fault may be greater diversity of focal mechanisms and greater depth o f hypocenters. As pointed out by Fohrmann et al. (2003), the volume of sources capable o f generating trapped waves at shallow structures increases with depth, and the amount o f generated energy depends on the receiver position within the radiation pattern of the events. Thus, the overall potential of generating trapped waves energy increases with the depth o f seismicity and diversity of focal mechanisms. Seeber et al. (2000) and Ben- Zion et al. (2003) found that most hypocenters around the Karadere-Duzce branch of the 1999 Izmit earthquake rupture are deeper than 5 km, and noted that the focal mechanisms o f the events are likely to be highly diverse. In contrast, about 50% of the 93 events with catalog locations in our data set have hypocenters shallower than 5 km and the events are likely to be dominated by strike-slip focal mechanisms. We note that the overall pattern of our event locations (Figures 1.2,1.3) is similar to the pattern of the catalog locations, although there are differences in the locations o f individual events. We have tried several other location techniques, such as plane-wave fitting and double-difference algorithm (Waldhauser, 2001) with waveform cross-correlation, but were not able to significantly improve the locations. Our location procedure employs a ID velocity model because a 3D model with the 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. fault zone structure in our study area is not available. However, the quality of earthquake locations obtained with ID velocity model and station corrections is generally comparable to that produced by a 3D model (e.g., Eberhart-Phillips & Michael 1998). The real limitation for obtaining better locations using only the phase picks recorded at the FZ array stems from the fact that the array aperture is only about 1 km. Unfortunately, most events that generate FZ trapped waves with quality A or B in our data set are not recorded by the SCSN and hence do not have catalog locations. We suggest that in future designs of similar experiments, a number of stations should be installed off the fault, as was done in our related study on the Karadere-Duzce fault (Seeber et al. 2000; Ben-Zion et al. 2003), to have sufficient regional coverage for accurate determination of event locations. 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 2 (Peng & Ben-Zion 2004a) SYSTEMATIC ANALYSIS OF CRUSTAL ANISOTROPY ALONG THE KARADERE-DUZCE BRANCH OF THE NORTH ANATOLIAN FAULT SUMMARY We perform a systematic analysis of crustal anisotropy along and around the Karadere-Duzce branch of the north Anatolian fault, which ruptured during the 1999 Mw7.4 Izmit and M w7.1 Diizce earthquakes. A method consisting o f an iterative grid search for the best shear wave splitting parameters in sliding time windows is applied to -22000 measurements recorded in the 6 month period after the Izmit mainshock. Based on objective criteria, -6600 measurements are assigned “high” quality and used for further detailed analysis. Most stations near the rupture zone have fast polarization directions that are parallel to, and change with, the nearby fault strike. The average delay times for ray paths that propagate along the rupture zone are larger than for the other paths. These results suggest the existence of an approximately 1 km broad zone around the Karadere-Duzce branch with fault- parallel cracks or shear fabric. However, some fault zone stations record bimodal or scattered polarization directions, while stations near large structural complexities (e.g., branching and offsets) show average fast polarization directions that are almost perpendicular to the local fault strike. The average fast polarization directions from ray paths that propagate inside the Almacik block, south of the Karadere-Duzce branch, are neither parallel to the local fault strike nor to the expected regional maximum compressive stress direction. The large overall spatial variations of the 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. results imply that multiple structures and mechanisms contribute to the observed crustal anisotropy in our study area. Most stations do not exhibit a clear dependency of shear wave splitting delay time with increasing depth and hypocentral distance, indicating that the anisotropy is confined primarily to the top 3-4 km of the crust. Using the observed average delay time at fault zone stations and assumed propagation distance o f 3.5 km, we estimate the apparent crack density in the damaged shallow fault zone rock to be about 7%. 2.1 INTRODUCTION Microcracks in a damaged crustal rock are expected to close preferentially in the direction normal to the maximum compressive stress < T i. In a strike-slip regime, cti is horizontal and is labeled as oh, while the minimum compressive stress is labeled as o^. Since effective elastic properties of rocks depend on the distribution of microcracks (e.g. Nur & Simmons, 1969; Nur 1971; O ’Connell & Budiansky 1974; Hudson 1981; Lyakhovsky et al. 1997), seismic shear waves propagating in the direction of Oh are expected to travel faster than those propagating in the O h direction. The difference in speeds will cause shear waves to separate into fast and slow components, a phenomenon that is termed shear wave splitting. Two routinely determined splitting parameters are the polarization direction of the fast wave ( < j> ) and the delay time (51 ) between the fast and slow waves. A model consisting of vertically-aligned, fluid-filled microcracks parallel to the oh direction is commonly assumed in analysis of crustal anisotropy (e.g. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Crampin 1978, 1987; Leary et al. 1990). Detailed studies, however, indicate that the observed ( j ) directions often vary over short distances, rather than being parallel to the inferred regional cth- These observations reflect heterogeneities in the stress field and/or the rock properties. Plausible sources for such heterogeneities include alignment of cracks in the vicinity of active faults (e.g. Zhang & Schwartz 1994; Tadokoro et al. 1999; Zinke & Zoback 2000), intrinsic anisotropy resulting from rock fabric (Kern & Wenk 1990) and preferential mineral alignment (Brocher & Christensen 1990), and remnant features of paleostress (e.g. Blenkinsop 1990; Aster & Shearer 1992). Studies on the depth extent of crustal anisotropy based on measured splitting parameters often reach very different conclusions. Peacock et al. (1988) and Savage et al. (1989, 1990) argue that crustal anisotropy in their study areas must be confined to the upper few kilometers to explain different polarization directions observed at stations located only a few km apart. Zhang & Schwartz (1994) suggest that seismic anisotropy in the Loma Prieta segment of the San Andreas fault system is no deeper than 2 km. Similarly, Munson et al. (1995) indicate that anisotropy is confined primarily to the top 3 km of crust in southern Hawaii. Liu et al. (2004) show that crustal anisotropy around the rupture zone of the 1999 Chi-Chi earthquake is dominated by the top 2-3 km. In contrast, Shih & Meyer (1990) and Li et al. (1994) claim to have found clear increase of delay times with increasing propagation distance in the south moat of the Long Valley caldera and the Northern Los Angeles Basin in California, respectively, and favor a more pervasive anisotropy than just 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. within a few kilometers beneath the stations. Zinke & Zoback (2000) argue based on two different stable fast directions at one station that the upper crustal layer (top 2-4 km) near the Calaveras fault in central California appears to be isotropic. An effort to clarify the spatial properties and sources o f crustal anisotropy requires a systematic analysis of a large data set in a densely instrumented area. In this work we perform such a study for the region around the Karadere-Duzce branch of the North Anatolian fault (NAF) that ruptured during the 1999 Mw7.4 izmit and M w 7.1 Duzce earthquakes. In particular, we are interested in the subsurface seismic properties in the immediate vicinity of the Karadere-Duzce fault. Ben-Zion et al. (2003) inferred from systematic analysis of seismic fault zone (FZ) trapped waves that the NAF in this area has a shallow (order 3 km) seismic FZ waveguide that is about 100 m wide and is characterized by intense damage (i.e., low seismic velocity and high attenuation). The results o f the present work point to the existence of a -1 km wide belt of strongly anisotropic rock around the seismic trapping waveguide on the Karadere-Duzce fault. The belt of anisotropic rock is confined primarily to the same depth extent (top 3-4 km) of the narrower trapping structure. In the following sections we first describe our data set and analysis methodology. Using objective criteria, a quality of either “high” or “low” is assigned automatically to -22000 measurements of shear wave splitting parameters. Additional detailed analysis is performed on -6600 measurements with “high” quality. We divide earthquakes within the shear wave window of at least one station into FZ and non-fault-zone groups, based on their relative locations with respect to 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the surface ruptures of the Izmit and Diizce mainshocks. We then present measured splitting parameters for each group separately. Finally, we combine the results from the different groups and discuss possible causes and the depth extent of the observed anisotropy in our study area. 2.2 DATA 2.2.1 The Seismic Experiment A temporary 10-station PASSCAL seismic network was deployed along and around the Karadere-Diizce branch of the NAF a week after the August 17, 1999, Mw7.4 Izmit earthquake for about six months (Seeber et al. 2000; Ben-Zion et al. 2003). Additional 7 stations operated in the first two weeks o f deployment period and were later removed (Figure 2.1). All stations had REFTEK recorders and three- component L22 short-period sensors with a sampling frequency of 100 Hz. Three months later, the November 12, 1999, Mw7.1 Diizce earthquake re-ruptured part of the Karadere segment that failed during the August Mw7.4 event and extended further east. Our temporary seismic network straddled the rupture zones of both mainshocks and recorded about 26000 earthquakes during its operational period (Figure 2.1a). The event locations were obtained in several stages, starting with standard HYPOINVERSE determinations (Klein 1978) and continuing with event- dependent station corrections (Seeber et al. 2000; Ben-Zion et al. 2003). The horizontal location errors are less than 1 km near the center of the network and 1-2 km near the margins. The vertical errors are somewhat greater. In this study we 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. focus on data recorded by the three-component short-period instruments at all 17 stations (Figure 2.1b). The surface rupture of the Izmit earthquake changes its strike at both ends of the Karadere segment. Many hypocenters of Izmit aftershocks are distributed around these two areas and clearly outside the major rupture zone. These events may be associated with secondary cross faults and other regional structural complexities (Seeber et al. 2000). The occurrence of the Diizce earthquake changed considerably the locations and other aspects of the seismicity pattern in the area. The aftershocks of the Diizce earthquake were concentrated on both ends of that mainshock rupture. Additional details on the experiment and data set are given by Seeber et al. (2000) and Ben-Zion et al. (2003). 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.1. (a) Hypocentral distribution of -26000 earthquakes recorded by the PASSCAL seismic experiment along the Karadere-Duzce branch o f the NAF. Aftershock locations are marked with shadings denoting different depth ranges. Shaded background indicates topography with white being low and dark being high. The surface ruptures of the Izmit and Diizce earthquakes are indicated with thick dark and gray lines, respectively. Dark thin lines associated with earthquake information denote faults that were active during recent ruptures. Other dark thin lines are geologically inferred fault traces. Gray squares denote locations of nearby cities. The inset illustrates the tectonic environment in northwestern Turkey with the box corresponding to our study area. Arrow vectors represent plate deformation rate (Reilinger et al. 1997) from GPS data, (b) Distributions of seismic stations and hypocentral locations of -9200 earthquakes used for the shear wave splitting analysis in this study. Triangles and diamonds denote stations deployed for about 6 months and 2 weeks, respectively. Stations within, near and outside the FZ are shaded with dark, gray and white colors, respectively. Cross sections along lines AA' and BB' are used in Figures 2.11 and 2.17, respectively. Earthquakes located within boxes A, B and C denoted by dash lines along fault strike belong to the FZ group, while the others belong to the NFZ group. Waveforms of the events marked by the star and small circles are shown in Figures 2.2 and 2.6, respectively. The event ID numbers consist of 3-digits Julian day, 2-digits hour, 2-digits minute, and 2-digits seconds of the earthquake occurrence time. Julian days in the range 237-365 are in year 1999 and those in the range 001-042 are in 2000. 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40‘ 50' 40° 40’ 40° 30’ (b) 40° 50’ 40 “ 40' ' 1 • . Sapauca I a.ke '< •. X '\ .if > ■ i u ■Hi B o jH , •■ P llll 30' 10' 30" 20’ 30' 30' 50 40' 30' 50* 31*00* 31* 10’ 31*20' 31*30' \ -B , Box c: Diizce Hendek . $ ■ ^ v V . . 347221 ?5S-, BU Box A ' ' % vfR*324i»1255 --t, J L - _ AkJW L I H Ct W i 2 * t Alinacik Block W I W 6 0 O ''0 2 2 ’ 4 i ! ‘ J 30“ 30’ 30° 40 30* 50' 31*00' 3! 10' 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.2.2 Data Selection The shear wave splitting analysis is conducted on waveforms generated by earthquakes that are within the shear wave window. To avoid contamination from S- to-P phase conversions near the surface, the incident angle of a ray path must be less than the critical angle ic = sin'1 (Vs/Vp) with VPand Vs being the near-surface velocities of P and S waves, respectively (Nuttli 1961; Booth & Crampin 1985). For a homogeneous half-space with a Poisson’s ratio of 0.25, the critical angle is ic -35° (Nuttli 1961). Since the low-velocity near-surface layer significantly bends ray paths toward the vertical, a straight-line incident angle of 45° is adopted as the critical angle in this study (e.g. Shih & Meyer 1990; Cochran et al. 2003). In addition, we check the quality of the recorded three-component waveforms and reject those with bad channels. We do not use -1400 waveforms observed by station CH from 09/18/1999 to 10/07/1999 because the east component was not recorded properly during this period. About 500 waveforms recorded by station GE are not used because the east component did not function well during 08/27/1999 and 10/31/1999. 2.2.3 Data Grouping Figure 2.1b gives locations o f about 9200 earthquakes that are within the shear wave window of at least one station in our study area. Since seismic waves propagating inside or outside the FZ may sample crustal rocks with quite different 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. properties, we further separate these earthquakes into FZ and non-fault-zone (NFZ) groups, based on their relative locations with respect to the surface ruptures of the two mainshocks. This is defined in the following way. The FZ group contains earthquakes that are relatively close to or inside the rupture zone. For the region north o f station CH, we use a 5-km-wide box that is centered around the rupture zone and is parallel to the E-W direction (Box A in Figure 2.1b). For the region north of the 30-km-long Karadere segment, where the fault is dipping about 80° to the north (Seeber et al. 2000; Ben-Zion et al. 2003), we use a 5-km-wide box that is parallel to the ENE direction and centered on the belt of earthquakes in that area (Box B in Figure 2.1b). Because the fault zone of the Diizce earthquake is dipping -65° to the north (Utkucu et al. 2003), we increase the width of the selection box to 12 km for earthquakes around station BV (Figure 2.1b, Box C). The NFZ group contains earthquakes that are located south of the surface rupture o f the two mainshocks in the Almacik block and north of Box B. In the following sections, we present splitting measurements for each group separately and discuss the results. 2.3 ANALYSIS PROCEDURE Our shear wave splitting analysis consists of two stages. In the first stage we apply the method of Silver & Chan (1991) assuming a single layer of anisotropy and a sliding window technique to determine the fast polarization direction § and delay time 81 of -22000 sets of splitting parameters. In the second stage we automatically assign a 2-level quality to each set of splitting parameters based on objective criteria 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and examine details of the results in the higher quality sets. Since the shallow low- velocity rocks bend the ray paths to the near-vertical direction, most o f the seismic energy from earthquakes that are within the shear wave window is in the horizontal- component seismograms. We thus use in the analysis only the horizontal-component seismograms. (See Appendix B for a description of the shear wave splitting code and setup.) We determine the shear wave splitting parameters with the following procedure. First we construct a 2x2 covariance matrix from horizontal-component seismograms. Ideally, the eigenvector associated with the larger eigenvalue (Xi) of the matrix points to the initial polarization direction and the smaller eigenvalue X 2 is zero. In the presences of noise, the latter is small but always larger than zero. The value of % 2 can be used as a measure of linearity for the shear waves (e.g. Vidale 1986). A grid search is performed over the < ( > — 51 space to find the best solution that minimizes X2 and produces the most singular covariance matrix and linear particle motion. In practice, the search range for < |> is from -90° to 90° in steps of 1°, where the north is at 0° and clockwise direction in positive. For 5t, we use a range of 0 to 0.5 s with an increment of 0.01 s. Once 6/ is obtained, we correct for anisotropy by advancing the slow component by 51 and rotate the fast and slow components into the initial polarization direction a. Ideally, this shifts all the energy to the a direction. The main analysis steps of the above procedure are illustrated in Figure 2.2. Prior to the analysis, the seismograms are low-pass filtered at 15 Hz using a two-way 4-pole Butterworth filter. Because shear waves generated by local 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. earthquakes have a dominant frequency of about 5-10 Hz, the applied filter does not degrade the results while helping to suppress high-frequency noise. Next, the seismograms are windowed around the shear wave arrivals. An ideal time window should begin before the fast shear wave arrival and end after the slow direct shear wave arrives, but before the scattered coda waves appear. Since the delay time 51 between the fast and slow direct shear waves is different for each seismogram, a window with fixed start and end relative to the first shear wave arrival is not the best choice for every measurement. Here we automatically determine the position of an optimal window by sliding a 0.6 s time window around the fast shear wave arrival. The best window is picked when the minimum Z2 solution or the most linear particle motion is achieved (Figure 2.3). The routinely picked S arrivals may contain various measurement errors. The sliding time window is more flexible than a fixed window in its ability to adjust for errors of phase picks within a certain time range. Measurements from similar earthquake clusters show that with routinely picked S arrivals, the splitting parameters determined for different waveforms with the sliding window technique are more consistent than those from a fixed window (Peng & Ben- Zion, in preparation, 2004). The sliding window strategy not only guarantees a best time window for each seismogram, but also provides an effective way for checking the stability of the results in relation to the cycle skipping problem as discussed below. It is important to provide quality control of the measured splitting parameters to ensure reliability o f results. Due to the very large amount o f data in this study, it is 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. not practical to use human interaction for quality control (e.g. Savage et al. 1990; Gerst 2003). In addition, a quality assignment based on visual inspection is subjective and the process typically cannot be reproduced by others (Aster et al. 1990). Hence an automatic quality assignment using objective criteria is needed. Based on previous studies (Matcham et al. 2000; Cochran et al. 2003; Gerst 2003), we design the following 10 objective quality criteria and apply them to our data: (1) The maximum difference of the fast direction ( |> generated by a 0.05 s shift of the best time window is < 30°. (2) The maximum difference of the delay time 51 generated by a 0.05 s shift of the best time window does not exceed 0.02 s. (3) The difference between the initial polarization a and < j ) is 20° < |< |) - a | < 70°. (4) The seismograms have a signal to noise ratio SNR > 3. (5) The amplitude o f the horizontal components is significantly larger than the corresponding vertical component (horizontal to vertical ratio or HVR > 1.5). (6) The energy in the waveform component perpendicular to the initial polarization a is small after anisotropy correction (radial to tangential ratio or RTR > 2). (7) The cross-correlation coefficient (CCC) values between fast and slow components is larger than 0.7. (8) The standard deviation of < ( > is < 20°. (9) The standard deviation of St is less than 0.1 s. (10) The smaller eigenvalue X2 of the covariance matrix is < 0.3. Solutions that satisfy the above 10 criteria are assigned “high” quality, while the other solutions are assigned “low” quality. 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.2. Illustration of steps in the analysis of the shear wave splitting, (a) Three component seismograms recorded at station BV for event 324001255. The epicentral location is marked as star in Figures 2.1b and 2.17. Vertical lines with P, S, T1 and T2 denote the P and S wave arrivals and automatically picked start and end of the time window used in the analysis, (b) Horizontal component seismograms rotated into the initial polarization and orthogonal directions before correcting for anisotropy, (c) Horizontal component seismograms shifted in time for anisotropy correction and rotated into the initial polarization and orthogonal directions. The event ID, station name, epicenter distance (Dist), back-azimuth (BAZ), focal depth, obtained initial polarization direction a , and start and end of the time window (twin) relative to the S wave arrival are shown at the top of the upper panels, (d) The windowed horizontal components rotated into fast and slow directions, (e) The same waveforms adjusted for the delay time 8t. (f) A particle motion plot for the waveforms rotated into the fast and slow directions but before the 51 correction, (g) The particle motion plot after the waveforms are rotated and corrected for 51 . (h) A contour plot showing the confidence level of the result in the <j)-8t space. The best fit ( < j > = -84°, 8t = 0.14 s) is marked as an asterisk and the 95% confidence level is shown as a double contour. The event ID, station name, obtained fast direction ( j> , delay time 51 , the minimum value of % 2 and applied filter are shown at the top of the lower panels. 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 324001255 BV: Dist 4.99 km, BAZ 348.7 °, Depth 8.2 km, a -50.9°, twin -0.22 - 0.38 s (b)_ N orth BV R - 5 1 .0 BV T -1 4 1 .0 V e rtic al v y \ / w ^ - 8V R - 5 1 .0 n/WA/y-'A/v- BV T -1 4 1 .0 A /v'-x, p I Tl s T2 I i t.... 1 .....1 T im e ( s ) T im e ( s ) 324001255 BV: ( j ) -84 ± 5°, 8? 0.14 ± 0.005 s , te 0.086, Filter: LP 15 Hz co. a > ~o -0.5 o - 1.0 CD T im e (s) o o _ -50 o o o - 2 x 1 0 + 4 Fast am plitude x 1 0 + 4 Fast amplitude Delay tim e (s) 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) 0.5 0.4 0.3 0.2 0.1 ( b ) 0.0 H 180 - 150 - au C D -cs 120 - S 3 O • *-H O 90 - % •3 60 - w CO 03 PC 30 - 0 - 0.4 4 (c) '&0.3 flj C/2 i ^ .5 0.2 -f* +-> S 0 .1-1 0.0 A .* * * *****^******** **** **** ****** i& t..,, 324001255 BV, twin: -0.22 ~ 0.38 sec, Quality: high ☆ * * jjtr * ' * ****** ** **«• 0.1 0.2 0.3 0.4 0.5 End time window (sec) 0.6 Figure 2.3. Changes of the (a) smaller eigenvalue % 2, (b) ( f > and (c) 8t with the end of sliding time windows for the shear wave splitting measurement shown in Figure 2.2. Stars mark the automatically picked end times associated with the smallest X2. Horizontal lines mark ± 20° of the best t j > and ± 0.02 s of the best bt that are used to check stability of the solution within a 0.05 s time shift. The word “twin” stands for the start and end of the determined time window. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The cutoff value for each criterion is chosen to be compatible with previous studies and simultaneously to result in no more than 10-30% of the solutions failing the criterion. Figure 2.4 illustrates the distributions of the first 9 criteria and their relations to X2 for all 2818 measurement at station BV. Solutions with high SNR, HVR, RTR and CCC values generally have small 7,2 . This indicates that some criteria are not independent. Such dependency provides an additional quality check of internal consistency. Specifically, criteria (1-2) check the stability of a solution with respect to changes of selected window to minimize contamination due to cycle skipping (e.g. Cochran et al. 2003). Criterion (3) checks the ability of shear waves with certain initial source polarizations to resolve anisotropy. If the initial polarization a of a shear wave is nearly parallel or perpendicular to < |) , it will not be split into fast and slow polarizations and retain its linear particle motion, resulting in a null measurement (Leary et al. 1990; Savage 1999; Gerst 2003). Under such circumstance, the splitting parameters are very sensitive to noise and should be discarded. Cochran et al. (2003) reject solutions with a within ±10° of their inferred fast or slow directions. However, in the presence of several set of cracks, resulting in bimodal or scattered values of < |> , it is not appropriate to apply a global rejection criterion based on the average fast and slow directions. Here we treat each measurement separately and reject (Figure 2.5) a solution with 20° < |< j ) - a | < 70°, where a and ( j ) are the specific values of the initial polarization and fast direction for that measurement. Criteria (4-5) inspect the three-component seismograms and guarantee that only high quality data (high SNR) with small 5-wave amplitude on the 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vertical component (high HVR) are used (Bouin et al. 1996; Gamar & Bernard 1997; Gerst 2003). Criteria (6-7) check the waveforms generated by the splitting program and select those solutions with simplified energy reduction on the component perpendicular to a after anisotropy correction (high RTR) and a good waveform match between the fast and slow component (high CCC value) (Matcham et al. 2000; Cochran et al. 2003; Gerst 2003). Criteria (8-9) check the existence of one small 95% confidence area (Matcham et al. 2000; Gerst 2003). Finally, the value of X2 provides a quantitative measure of the linearity of the particle motion. In total, about 30% of the calculated 21774 solutions are assigned “high” overall quality and used for detailed examination in the following sections. Slightly different cutoff values may result in different sets of individual measurements being assigned “high” quality. However, since the results discussed in the following sections are based on statistics of thousands of measurements, the inclusion or removal of a small percent of measurements in the high quality set will not change our overall conclusions. As shown in Table 1, the selection reduces the standard deviation of the measured splitting parameters and shifts somewhat the average resolved results (especially for some FZ stations with bimodal distributions of splitting parameters). Figure 2.6 shows several sets o f original and rotated waveforms used for high quality determinations of splitting parameters. It is important to provide quality control of the measured splitting parameters to ensure reliability of results. Due to the very large amount o f data in this study, it is not practical to use human interaction for quality control (e.g. Savage et al. 1990; 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Gerst 2003). In addition, a quality assignment based on visual inspection is subjective and the process typically cannot be reproduced by others (Aster et al. 1990). Hence an automatic quality assignment using objective criteria is needed. Based on previous studies (Matcham et al. 2000; Cochran et al. 2003; Gerst 2003), we design the following 10 objective quality criteria and apply them to our data: (1) The maximum difference of the fast direction < j ) generated by a 0.05 s shift of the best time window is < 30°. (2) The maximum difference of the delay time 81 generated by a 0.05 s shift of the best time window does not exceed 0.02 s. (3) The difference between the initial polarization a and ( j) is 20° < |< j > - a | < 1 70°. (4) The seismograms have a signal to noise ratio SNR > 3. (5) The amplitude of the horizontal components is significantly larger than the corresponding vertical component (horizontal to vertical ratio or HVR >1.5). (6) The energy in the waveform component perpendicular to the initial polarization a is small after anisotropy correction (radial to tangential ratio or RTR > 2). (7) The cross-correlation coefficient (CCC) values between fast and slow components is larger than 0.7. (8) The standard deviation of < j ) is < 20°. (9) The standard deviation of 5t is less than 0.1 s. (10) The smaller eigenvalue fa of the covariance matrix is < 0.3. Solutions that satisfy the above 10 criteria are assigned “high” quality, while the other solutions are assigned “low” quality. The cutoff value for each criterion is chosen to be compatible with previous studies and simultaneously to result in no more than 10-30% o f the solutions failing the criterion. Figure 2.4 illustrates the distributions of the first 9 criteria and their 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. relations to X ,2 for all 2818 measurement at station BV. Solutions with high SNR, HVR, RTR and CCC values generally have small X 2. This indicates that some criteria are not independent. Such dependency provides an additional quality check of internal consistency. Specifically, criteria (1-2) check the stability of a solution with respect to changes of selected window to minimize contamination due to cycle skipping (e.g. Cochran et al. 2003). Criterion (3) checks the ability o f shear waves with certain initial source polarizations to resolve anisotropy. If the initial polarization a of a shear wave is nearly parallel or perpendicular to < j) , it will not be split into fast and slow polarizations and retain its linear particle motion, resulting in a null measurement (Leary et al. 1990; Savage 1999; Gerst 2003). Under such circumstance, the splitting parameters are very sensitive to noise and should be discarded. Cochran et al. (2003) reject solutions with a within ±10° of their inferred fast or slow directions. However, in the presence of several set of cracks, resulting in bimodal or scattered values of it is not appropriate to apply a global rejection criterion based on the average fast and slow directions. Here we treat each measurement separately and reject (Figure 2.5) a solution with 20° < |4 > — a | < 70°, where a and < j ) are the specific values of the initial polarization and fast direction for that measurement. Criteria (4-5) inspect the three-component seismograms and guarantee that only high quality data (high SNR) with small S-wave amplitude on the vertical component (high HVR) are used (Bouin et al. 1996; Gamar & Bernard 1997; Gerst 2003). Criteria (6-7) check the waveforms generated by the splitting program and select those solutions with simplified energy reduction on the component 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. perpendicular to a after anisotropy correction (high RTR) and a good waveform match between the fast and slow component (high CCC value) (Matcham et al. 2000; Cochran et al. 2003; Gerst 2003). Criteria (8-9) check the existence of one small 95% confidence area (Matcham et al. 2000; Gerst 2003). Finally, the value of provides a quantitative measure of the linearity of the particle motion. In total, about 30% of the calculated 21774 solutions are assigned “high” overall quality and used for detailed examination in the following sections. Slightly different cutoff values may result in different sets of individual measurements being assigned “high” quality. However, since the results discussed in the following sections are based on statistics of thousands of measurements, the inclusion or removal of a small percent of measurements in the high quality set will not change our overall conclusions. As shown in Table 1, the selection reduces the standard deviation of the measured splitting parameters and shifts somewhat the average resolved results (especially for some FZ stations with bimodal distributions of splitting parameters). Figure 2.6 shows several sets of original and rotated waveforms used for high quality determinations of splitting parameters. 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.4. Values associated with the 9 criteria against 12 for 2818 measurements at station BV. The cutoff values (vertical lines) and the percentage o f rejected solutions for each criterion are given at the top of each panel. White triangles and gray squares mark measurements that are assigned “high” and “low” qualities, respectively. The letter combinations stand for the following: m a x F D d iff is maximum difference of fast direction; max_DT_diff is maximum difference of delay time; ang_diff is difference between the initial polarization and fast direction; SNR is signal to noise ratio; FTVR is horizontal to vertical ratio; RTR is radial to tangential ratio; CCC is cross-correlation coefficient between fast and slow waves; STD FD is standard deviation of the fast direction; STDJDT is standard deviation o f the delay time. 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. * + >s a j& jt S « * * 4 U . •• :•' * > c I • J'-f. 3f. o ® o'1 - O N 00 V r~ : e d 04 c n o V 5 < - ■ “ + " * ' o •9®ssK i m »a atis a8 aasa s« WM9 MK)m* < N d d C C 3 ... O F- C s e J > f-C O ' O' © < N V 1 Sc q a H3s.?ssr= ' ‘I W W . . a s . - * v s " % » * r " * . . e * „ ! . ■■'V SK ’ . * .# .A?’ v.; m - « ■ * » * z ^ W , *7 d V fO d C 4 © H q p 1 _ o o C - ? Q Q 1 O f— I — < 75 o r-i © o « * j • > ■ » .• u tm jg utto^g railing 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.5. (a) Histogram of initial polarization for all (black) measurements at station BV and results with low (gray) and high (white) quality. The total number of measurements N, average fast direction < j > , mean resultant length R and average and standard deviation of 51 for each category are shown at the top right comer, (b) Plot of initial polarization a versus fast direction < j ) for different quality of measurements. The solid thick lines mark the relation < j ) = a ± 90°. The thick dash lines mark the relations 20° < |< j ) - o c| < 70° that we use to reject possible null measurements. The thin vertical dash lines denote the region of values that would be discarded using the uniform rejection criterion of Cochran et al. (2003). Other symbols are the same as in Figure 2.4. (c) Histogram of the fast direction < |> . (d) Plot of the initial polarization a versus delay time 51 . (e) Histogram of the delay time 51 . 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Delay Time (s) Fast Direction (deg) Numbers BV N o R 8t(s) All: 2818 99.8 0.42 0.09±0.09 Low: 2074 99.5 0.40 0.09±0.09 High: 744 100.4 0.48 0.0&b0.08 Numbers 0 100 200 s m w 200 Numbers Initial Polarization (deg) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.6. Examples of original and rotated waveforms that are used to determine splitting parameters of “high” quality data. Horizontal seismograms are rotated into fast and slow components using the obtained < |> and are plotted together with the original three-component waveforms. The seismograms have been low-pass filtered <15 Hz. They are plotted using a fixed amplitude scale in each panel and are aligned with S arrivals 0 s. The dash lines indicate the start and end of the time windows obtained automatically using the sliding window technique. The short vertical bars indicate the fast and slow shear wave arrivals. The event ID, epicenter distance (Dist), focal depth, station name, back-azimuth (BAZ), obtained fast direction < j ) and delay time 51 are shown at the top of each panel. The epicenters of the events are shown in Figure 2.1b. Events 338063309 and 033230014 are also marked on a vertical cross section in Figure 2.11, and events 281221102 and 286032156 are marked on a map view and a vertical cross section in Figure 2.17. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 313110622: Dist 5.3 km. Depth 10.7 km 033230014: Dist 3.1 km. Depth 12.1 km 250213050: Dist 5.6 km, Depth 6.8 km CH, B A Z -9 2 .1 ”, <j>=S4.0°, 5 / = 0.07 s GE, BAZ= 107.3°, 4 > = 98.0 ", 6 / = 0.02 s CL, BAZ = 98.9”, ( ( > = 149.0 ",Sr = 0.10 s c r v w ,;l/Ar/V W \ vv V A a ^ jv y y v ^ A A v n l i , A j\ h A p . m m * I \ J i / \ / V v v W r V ,A//v A /v- - / W W V / ^ 326173135: Dist 2.2 km. Depth 9.7 km 338063309: Dist 7.0 km, Depth 13.2 km 286032156: Dist 5.7 km. Depth 8.1 km CH, BAZ = 68.5”, < j > = 106.0 ”, 8t = 0.12 s FP, BAZ = 266.1”, < ( > = 89.0 ”, 8 ( = 0 .1 3 s BV, BAZ = 223.8”, $ = 117,0 °, 8t = 0.03 s N [ ' I' f ’ ' y ' ' s j) W iA A H\/!aA N \ / vv A /l/VVXAA.- ■4/ \/\/\W - a /Ia A /K A a ^ \j 272230846: Dist 10.3 km, Depth 10.5 km 302234519: Dist 11.0 km, Depth 11.8 km 347221758: Dist 6.1 km, Depth 13.7 km LS, B A Z-2 0 3 .3 ”, 6 = 10.0°, 5/ = 0.06 s MO, BAZ = 204.3”,$ = 33.0 ”, 5 / = 0.08 s BV, BAZ = 344.3”, $ = 102.0 ”, 5t = 0.22 s ~ j /| | '/ \z \ A / A J V ~ I I / I I /I i l -'\l/ j i ^/V^/sV A V 'J ; /y / \ \ r J \ s ~ S \ j ~ ' V A ) \ a /\AM I/Wv V/V ■ W W , V v /~ A A ^ N y \/'" -1.0 -0.5 0.0 0.5 Time (s) l.O 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2.1. Station Locations and Shear wave splitting Results3 Station 0, deg R 51 , ms N 0, deg R 51 , ms N Dist to Total Total Total Total High High High High FZ CH 79.67 0.45 89 ±64 3678 82.65 0.62 85 ±48 1286 <1 km PT 50.79 0.14 73 ±55 434 58.92 0.05 72 ±39 135 <1 km LS 176.69 0.23 78 ±57 1966 173.75 0.38 73 ±38 586 Inside MO 91.73 0.12 73 ±55 1778 57.43 0.11 71 ±45 679 Inside FI 91.90 0.10 62 ±46 1895 53.00 0.20 53 ±35 361 -400 m AR 45.48 0.21 84 ±79 158 43.95 0.29 69 ±50 60 <1 km VO 75.04 0.14 71 ±50 2288 73.81 0.19 73 ± 42 782 Inside FP 86.50 0.16 71 ±61 1945 69.20 0.36 79 ±56 667 -300 m TW 110.48 0.10 69 ±55 398 98.11 0.19 70 ±50 154 <1 km TS 126.91 0.15 89 ±69 134 120.41 0.23 67 ±37 28 Inside CL 154.35 0.41 71 ±41 323 152.68 0.55 82 ±42 88 Inside BV 99.80 0.42 89 ±75 2818 100.44 0.48 84 ±54 744 Inside WF 17.36 0.15 65 ±52 1905 27.28 0.23 68 ±40 423 - 5 km CF 125.41 0.07 76 ±63 335 162.53 0.23 68 ±40 86 - 5 km BU 77.09 0.29 91 ±74 570 61.40 0.35 79 ±62 172 - 8 km SL 40.25 0.17 72 ±69 176 45.37 0.41 43 ±35 47 - 5 km GE 110.00 0.36 66 ± 69 973 112.59 0.40 52 ±53 303 - 5 km a 9 is the average fast direction, R is the mean resultant length, bt is the average delay time, and N is the number of shear wave splitting measurements. “Total” and “high” denote all the measurements and the subset of high quality measurements at each station, respectively. “Dist” stands for the distance. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.4 RESULTS 2.4.1 Fast Polarization Direction The fast directions § of about 6600 measurements with “high” quality at all 17 stations are displayed using rose diagrams in Figure 2.7. To provide a statistical analysis o f the measured < |) , we first double each angle and apply the von Mises method to calculate the mean angle 0 and a mean resultant length R (Davis 1986; Mardia & Jupp 2000; Cochran et al. 2003). The parameter R gives a quantitative estimate of the variance of the directional data, with values near 0 and 1 indicating high scattering and clustering, respectively. As shown in Figure 2.7, out of the 17 stations 9 have R values less than 0.3, indicating that < j ) at these stations are scattered. Flowever, nearby stations have similar rose diagrams, suggesting that the measurements can be used to discern spatial patterns in our area. Spatial Separation. The spatial distribution of the employed earthquakes is complicated, so the observed scattering in Figure 2.7 may result in part from ray paths that propagate along regions with different properties. We distinguish between ray paths that propagate inside and outside the FZ by separating both earthquakes and stations into groups based on their relative locations with respect to the active fault. We refer to the 12 stations that are within 1 km of the surface rupture as FZ stations, while the other 5 are named regional stations. We also divide the earthquakes into FZ and NFZ groups according to their hypocentral locations as discussed in Section 2.2. 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. F Z Group. For earthquakes that are close to or inside the rupture zones of both the Izmit and Duzce earthquakes, 8 out o f 12 FZ stations have average fast directions that are parallel or sub-parallel to the direction of the nearby fault strike (Figure 2.8). Station CH is close to the town of Akyazi, where a 5-km-wide surface- rupture gap is observed after the izmit earthquake. Within this gap, the surface expression of faulting widens into a distributed cracking with nearly zero slip that is hundreds o f meter wide (e.g. Hartleb et al. 2002). In addition, the rupture trend changes about 25° from approximately E-W in the Adapazari basin to ENE along the Karadere segment (Figure 2.1). There are 454 individual measurements at station CH for earthquakes around the rapture zone. The mean resultant length R is 0.71, indicating high clustering of the data. The average fast direction 0 is about 87°, which matches well the average trend of nearby surface rapture expressions. Along the Karadere segment, stations PT, LS and MO have < j > that are either scattered or bimodal. Their average fast directions are between 100° and 120°. Stations FI and AR have less scattered distributions and their mean fast directions are close to the trending direction of the Karadere segment (-70° ENE). Stations VO, FP, TW and TS show fairly consistent values of < j) . The average values are within 20° of the direction of the nearby fault strike. Station CL has a consistent < j > that is almost perpendicular to the ENE fault orientation. Station BV is near Eften Lake, where the surface rapture of the izmit earthquake changes its trend from ENE to E-W and terminates. The Duzce earthquake re-raptured the easternmost 9 km o f the Karadere segment in that area (Hartleb et al. 2002). The value o f 0 is close to 100° for 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. earthquakes that are north of station BV and along the fault dipping direction. This value again matches well the direction of the surface expression o f the nearby fault strike. For stations WF and CF that are located about 5 km south of the Karadere segment, we observe bimodal distributions of (j). One of the dominant < | > is approximately N -S. The value of < |> observed at regional station SL north o f the Karadere segment is scattered. Station BU shows fairly consistent splitting results with 0 of -60°. At station GE the measured ( j > is slightly bimodal. About 40% of the observed < |> values are between 90° and 120°. About 30% of ( j ) are oriented between 125° and 165°. The average fast direction 0 is ~110°. N F Z Group. Station CH has very consistent < |> for earthquakes south of it (Figure 2.9). The average value 0 is about 80°, which is within 10° of the value for earthquakes north of CH (Figure 2.8). For the FZ stations along the Karadere segment and earthquakes that are outside the FZ, stations PT, AR, VO, FP and TW have either bimodal or scattered < ) > . Stations LS, MO and FI have fairly consistent < j ) with values of 0 ranging from about 0° to 40°. Stations TS and CL show 0 o f -145°. Station BV has very consistent fast direction with 0 of ~108° (Figure 2.9), which again differs by only about 10° from the value for earthquakes north of BV (Figure 2.8). For stations outside the FZ, WF shows fairly consistent fast directions with 0 of about 30°, which is similar to the value of 0 observed at stations LS, MO and FI for the same groups of earthquakes. The observed fast directions at station CF are Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. scattered. At stations BU and SL, the average fast directions are about NE-SW. Station GE observes a fairly consistent < j > of -NW -SE for earthquakes that are outside the mainshock rupture zone. CH: 0= 82.7° PT: 0=58.9° LS: 0=173.8° MO: 0= 57.4° FI: 0=53.0° R = 0.62, N= 1286 R = 0.05, N = 135 R = 0.38, N = 586 R = 0.11,N = 679 R = 0.20, N = 361 AR: 0=44.0* VO: 0 = 73.8 ° FP: 0=69.2° TW: 0= 98.1° TS: 0=120.4° R = 0.29, N = 60 R = 0.19,N = 782 R = 0.36,N = 667 R = 0.19,N=154 R = 0.23,N = 28 - & r - j / f - CL: 0= 152.7° BV: 6= 100.4° WF: 0='27.3° CF: 0= 162.5° BU: 0=61.4° R = 0.55, N = 88 R = 0.48, N = 744 R = 0.23,N = 423 R = 0.23,N = 86 R -0.35,N = 172 SL: 0= 45.4* GE: 0= 112.6” R = 0.41, N = 47 R = 0.40, N = 303 |N I iS Figure 2.7. Rose diagrams of the fast directions ( j ) at all 17 stations. The average fast direction 0, mean resultant length R and total number o f measurements at each station are shown on top o f each diagram. The first 12 stations are within or close to the FZ and the other 5 are clearly outside. 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. S£ +1 © r ? ® in C C + 1 O t f + 1 ll 7S4 V x f ig T t 0 £ + 1 © s O c * + l U . Z Q £ + a L + 1 U z Figure 2.8. Rose diagrams of f and equal area plots o f splitting parameters (bars) at 17 stations for earthquakes belonging to the FZ group. Bars in the equal area plot are oriented parallel to < j > and scaled by the delay time 8t. The average fast direction 0, R, average and standard deviation of 8i and total number of events N are marked on top of each equal area plot. The first 12 stations are within or close to the FZ and the other 5 are clearly outside. 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. o * = < z N & Q o n / y ( \ Jxx \\ , j I \ y - O L + 1 a + < X > t r i f i £ + ! c d r - > © ( N 2 z o i + l CD Z o' + o' + ! x + i c £ -H c d « CD r i £ z Figure 2.9. Rose diagrams of f and equal area plots o f splitting parameters (bars) at 17 stations for earthquakes belonging to the NFZ group. Other symbols are the same as in Figure 2.8. 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.4.2 Possible Causes of Anisotropy A summary plot of shear wave splitting measurements is shown in Figure 2.10. For ray paths that propagate inside the rupture zone, the average fast directions ( j ) at most FZ stations are in general parallel to and change with the direction of the nearby fault strike. These results suggest the existence of near-vertical fault-parallel cracks or shear fabric in the vicinity of the rupture zone. However, the results also indicate significant variations of properties along the fault. FZ stations TS and CL have average fast direction of -NW -SE, which is almost perpendicular to the trending direction of the Karadere segment. Since these stations are close to a 400- m-wide compressional step-over east of the town of Kadifekale (Hatleb et al. 2002), the observed anisotropy at TS and CL may be affected by strong local structural complexities. Some FZ stations (e.g., LS, MO) observe either scattered or bimodal fast directions for earthquakes that are near the FZ. Stations very close to or inside the surface rupture zone typically record after the direct S wave large-amplitude long-period oscillations that are interpreted as FZ trapped waves (Ben-Zion et al. 2003). The trapped waves tend to obscure the direct slow S wave arrivals and may reduce the reliability of the derived splitting measurement at these stations. Figure 2.11 shows the splitting parameters and locations of earthquakes that are within the shear wave windows of the FZ station FP and the regional station GE. We note that the same group of earthquakes along the Karadere segment (the dense cluster at the depth range of ~ 12-15 km near the heavy dashed line) produce very different anisotropy results at these two stations that are only ~5 km apart. The FZ 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. station FP has an average fast direction from these FZ earthquakes o f 0 = 70°, which is very close to the direction of fault strike in this area (inset o f Figure 2.11a). At station GE outside the FZ, 0 from the same group of earthquakes is about 110°, a direction that is between the -NW -SE direction of the regional maximum horizontal compressive stress < T h (Bellier et al. 1997; Tadokoro et al. 2002) and the direction of the nearby fault strike (Figure 2.11b). For earthquakes outside the FZ (blue dots in Figure 2.11), the observed fast directions at the FZ station FP have a bimodal distribution, with two dominant values close to the fault strike and the regional cth directions, respectively. At station GE, the average fast direction for earthquakes outside the fault is -120°, close to regional Oh direction from NFZ earthquakes. However, the regional stations BU and SL in the Adapazari basin have fast directions that are NE-SW (Figure 2.10) and hence almost perpendicular to the regional gh- Since most earthquakes within the shear wave window of these two stations have large incident angles, it is possible that such non-vertical ray paths sample different sets of microcracks. Seeber et al. (2000) note that the focal mechanisms of aftershocks in this area are highly diverse, pointing to a strong heterogeneous stress field that may have been partially created during the ruptures of the Izmit and Dtizce mainshocks. Such heterogeneous stress field may be responsible for generating complex sets of microcracks in the material off the mainshock rupture zones. The average fast directions from ray paths that propagate mostly inside the Almacik block are between 0° and 40°. This range is different from both the inferred 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. gh and the direction of the nearby fault segments. The relatively stable Almacik block is assumed to be responsible for the branching of the NAF in this area into two Hendek • Limn Lake ruiyttft.a * Karadere iifekaie Almacik Block Figure 2.10. A summary plot of average splitting parameters (bars) in our study area. The bars are oriented parallel to the average fast direction ( |> and scaled by the average delay time 8t. Dark and gray shadings of the bars denote results from earthquakes belonging to the FZ and NFZ groups, respectively. The center of each bar is plotted at the middle point between the corresponding station and the centroid of the earthquake epicenters. The thin line connects the center o f the each bar to its corresponding station. Only average results with at least 5 measurements and mean resultant length o f at least 0.2 are shown. The surface ruptures of the Izmit earthquakes are indicated with thick gray lines. Other symbols and notations are the same as in Figure 2.1. 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.11. (a) Splitting parameters (bars) superimposed on the hypocentral locations within the shear wave window of the FZ station FP and projected along the cross-section AA' in Figure 2.1b. The bars are oriented parallel to the fast polarization direction < j ) and scaled by the delay time 81. (Note that the zones producing the anisotropy are likely to be considerably shallower (Figures 2.12-2.15) than the earthquakes were the symbols are located.) Dark and gray colors denote results from earthquakes belonging to FZ and NFZ groups, respectively. The dash line indicates the -80° north-dipping fault along the Karadere segment that ruptured during the Izmit earthquake. The long lines with arrows show schematic straight-line ray paths for different source-station configurations and the rose diagrams summarize the results of fast polarization directions of different earthquake groups. The regional maximum horizontal compressive stress direction 0 H and the direction of fault strike nearby are shown in the inset by gray and dark lines, respectively. Waveforms generated by the event marked by the gray circle are shown in Figures 2.6. (b) Splitting parameters (bars) superimposed on the hypocentral locations within the shear wave window of station GE and projected along the cross-section AA' in Figure 2.1b. Other symbols and notations are the same as in (a). 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 GE ( b ) ■ 5 10 SSE NNW 15 -8 -6 -4 -2 0 2 4 Distance (km) 8 10 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. fault strands (e.g. Langridge et al. 2002). The southern branch extends along the Mudumu river valley and broke during the 1944 Bolu-Gerede, 1957 Abant and 1967 Mudumu Valley earthquake sequences (e.g. Barka 1996; Akyuz et al. 2000). The northern one follows the Duzce and Karadere segments and ruptured during the 1999 Izmit and Duzce earthquake sequences. The Almacik block consists primarily of mafic rocks, with some andesitic to basaltic units as well as sedimentary sequences with limestones and shales (Yilmaz et al. 1997). Thus the observed anisotropy within the block is probably caused by lithologic properties such as foliation, bedding, or aligned minerals (e.g. Aster & Shearer 1992) rather than stress-controlled microcracks. 2.4.3 Delay Times and Depth Extent of Anisotropy Due to the limited data points and inherent scatter in delay times 61 measured using shear wave splitting (Crampin et al. 2004), it is very difficult to constrain the depth extent of anisotropy (e.g. Zhang & Schwartz 1994; Cochran et al. 2003). Figures 2.12-2.15 present about 6600 51 measurements for the 17 stations versus depth and hypocentral distance of earthquakes that are either inside or outside the FZ. We also calculate the correlation coefficients (CC) of 5t with depth and distance. As shown, most stations have CC less than 0.3, indicating a lack o f clear dependency of 8t on increasing depth and/or distance. The results suggest that the observed shear wave splitting in this area is primarily caused by shallow structures rather than pervasive anisotropy throughout the entire crust. Since the depth of most earthquakes is larger than 5 km (Seeber et al. 2000; Ben-Zion et al. 2003), we cannot provide a 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 3 Q 0.4 0.3 0.2 O.i 0.0 0,4 0.3 - 0.2 - 0.1 - 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 • 0.2 • 0.1 ■ 0.0 ■ CH: Sr =0.09 +0.05 s N = 454, CC =-0.04 MO:Si =0.08 ±0.05 s N = 401, CC = 0.07 V O :8 t =0.07 +0.04 s N = 619, CC = -0.00 TS: 8t = 0.06 ±0.04 s N = 8, CC = -0.29 W F :8i = 0.07 ±0.04 s N = 264, CC = 0.24 SL: 8/ = 0.05 ±0.04 s N = 24, CC = 0.10 PT: 5/ =0.07 +0.03 s N = 27, CC = -0.02 LS: 5/= 0.06 ±0.04 s N = 190, CC =-0.01 FI: 5/ =0.06 +0.04 s N = 257, CC = 0.04 FP: 8t =0.08 ±0.06 s N = 521, CQ= 0.18 CL: 8t =0.08 ±0.03 s N = 50, CC = 0.29 CF: Si =0.06 ±0.04 s N = 44, CC = -0.17 AR: Sf =0.07 +0.05 s N = 40, CC = 0.25 < 9 - s O C D C • o na D < ® -o o TW:8? =0.06 ±0.05 s N = 98, CC = 0.15 BV: 8 t =0.09 +0.06 s N = 536, CC^O.IO BU: 5/ =0.08 ±0.06 s N = 160, CC =-0.01 cPB 0 ° u & o o S o 6 x 0 © GE: 8/ =0.05 ±0.06 s N = 2534 C C . =-0.16 Dataset FZ 4 8 12 16 20 D epth (km ) Figure 2.12. Delay time versus depth at all stations for earthquakes that belong to the FZ group. The station name, average and standard deviation o f 8t, total number of events (N), and correlation coefficients (CC) between 81 and depth are marked on top of each panel. 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 .4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 H 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 -\ 0.1 0.0 ^ 0 .4 05 < u 0.3 C H :5f= 0.09 ±0.05 s N =.835, CC = -0.00 i : r PT: 8/ = 0.07 ±0.04 s N = 108, CC = -0.00 I I LS: 81 = 0.08 +0.04 s N = 398, CC =-0.17 MO:8/ = 0.06 +0.04 s N = 2781 CC =-0.02 ° 0 0 ’ d ? FI: 8r = 0.04 ±0.03 s N = 108, CC = 0.08 o co- °°oSfc VO: 8f = 0.07 ±0.04 s N = 173, CC =-0.09 ^ j 8 h x > 0a p ' rn oc. FP: 8/ =0.07 ±0.05 s N = 150, C£ = 0.08 TW:8r =0.08 ±0.06 s N = 56, CC = -0.30 o o o o o “ cS S o o TS: 5/ =0.07 +0.04 s N = 20, CC = -0.36 •|"'T r CL: 8? = 0.09 ±0.05 s N = 39, CC = -0.30 ° & > ° ® % WF:8l = 0.05 ±0.03 s N = 161, CC =-0.01 a o c P ° 8 CF: 5? = 0.07 ±0.04 s N = 42, CC = 0.26 ; i ; i BV: 5 l= 0.06+0.04 s N = 212, CC = 0.10 BU: 8/= 0.09+ 0.05 s N = 12.CC = -0.14 AR: 5/= 0 .0 7 ±0.06 s N = 20, CC = 0.19 SL: 8? =0.04 ±0.02 s N = 23, CC = 0.08 a •4 3 0.2 I °-i Q 0.0 OE: 8? =0.05 ±0.05 s N = 53, CC = 0.04 Sf* Dataset NFZ 12 16 20 Depth (km) Figure 2.13. Delay time versus depth at all stations for earthquakes that belong to the NFZ group. Other symbols and notations are the same as in Figure 2.12. 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 .4 0.3 - 0.2 - 0.1 0.0 0.4 0.3 H 0.2 0.1 CH: 5f = 0.09 +0.05 s N = 454, CC =-0.04 PT: 8t = 0.07 ±0.03 s N = 27, CC = -0.05 o C o® ° ° r 9° o ' LS: 8t =0.06 ±0.04 s N = 190, CC = 0.11 0.0 0.4 0.3 - 0.2 - 0.1 MO:Sl = 0.08 +0.05 s N = 401, CC = 0.00 . C , j ; T. FI: 8? = 0.06 +0.04 s N = 257, CC = 0.01 ° c P A R :S /=0.07+0.05 s N = 40, CC = 0.20 0.0 0.4 0.3 0.2 H 0.1 VO: Si =0.07 +0.04 s N = 619, CC =-0.01 FP: St =0.08 +0.06 s N = 521, = 0.07 o o o TW:8/ =0.06 +0.05 s N = 98, CC = 0.02 Q - 0 O m o o.o 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 TS: 81 = 0.06 +0.04 s N = 8, CC = -0.04 CL: 8; = 0.08 +0.03 s N = 50, CC = 0.44 BV: 8l =0.09 +0.06 s N = 536, CC §30.06 , 8JO WF:8l = 0.07 +0.04 s N = 264, CC = 0.03 CF: 8; =0.06 +0.04 s N = 44, CC = -0.17 BU: 8/ =0.08 +0.06 s N = 160, CC — 0.02 9 . 5 a r > ~ & < L > ” < 5 Q 0.0 SL: Si =0.05 ±0.04 s N = 24, CC = 0.28 GE: Si = 0.05 +0.06 s N = 253, 0 6 3 = 0.09 ® 0 f» ° Oft,** Dataset FZ 4 8 12 16 20 Hypocentral distance (km) Figure 2.14. Delay time versus hypocentral distance at all stations for earthquakes that belong to the FZ group. Other symbols and notations are the same as in Figure 2 . 12. 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 .4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 0.1 0.0 0.4 0.3 0.2 H 0.1 CH: St = 0.09 ±0.05 s N = 835 ^ 0 0 =-0.09 c 0.0 ^ 0 .4 C/3 a) 0.3 0.2 - 0 3 13 Q o.i 0.0 PT: 5/= 0.07 ±0.04 s N = 108, CC = 0.10 o° “ O C ~ St. LS: 8f =0.08 ±0.04 s N = 398, CC =-0.14 MO:8l = 0.06+0.04 s N = 278,TIC = 0.00 VO: 5? =0.07 +0.04 s N = 173, CC =-0.05 TS: St =0.07 +0.04 s N = 20, CC = -0.41 o < P § CO (P o WF:8l =0.05 ±0.03 s N = 161, CC = 0.14 FI: 8? =0.04 ±0.03 s N = 108, CC = 0.01 % > AR:Sf =0.07 ±0.06 s N = 20, CC = 0.17 °9S o ° FP: 81 = 0.07 ±0.05 s N = 150, CC =$.08 0 6b TW:8? = 0.08 ±0.06 s N = 56, CC = -0.31 c S P S '$°C o m CL: 8< = 0.09 ±0.05 s N = 39, CC = -0.24 ,<b BV: 8^= 0.06 ±0.04 s N = 212, CC = 0.01 CF: 8f = 0.07 ±0.04 s N = 42, CC = 0.02 SL: St = 0.04 ±0.02 s GE: 8 /= 0.05 ±0.05 s N = 23, CC = 0.14 - N = 53, CC = -0.03 - o eg O o o a ;o c 0 o o j # 0° O 0 roc BU: Si =0.09 ±0.05 s N = 12, CC = -0.08 CD <9 Dataset NFZ 4 8 12 16 20 H ypocentral distance (km ) Figure 2.15. Delay time versus hypocentral distance at all stations for earthquakes that belong to the NFZ group. Other symbols and notations are the same as in Figure 2 . 12. 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. detailed picture of anisotropy variation with depth in the upper few kilometers. However, 8t on the order of 0.1 s is observed for the shallowest earthquakes with hypocentral depth of 3-4 km, suggesting that the anisotropy is confined primarily to the top 3-4 km of the crust. This result is compatible with those found in the Loma Prieta segment of the San Andreas fault (Zhang & Schwartz 1994), in southern Hawaii (Munson et al. 1995), along the rupture zones o f the 1999 Hector Mine earthquake (Cochran et al. 2003), and in the aftershock region o f the 1999, Chi-Chi, Taiwan, earthquake (Liu et al. 2004). Figure 2.16 gives a summary of the average 51 measurements. The standard deviations are close to 0.05 s, but it is still possible to discern several important features. Earthquakes and stations that are close to the FZ have larger average 8t than results associated with other source-receiver combinations of the results. The smallest average 51 values are obtained when both the earthquakes and stations are outside the FZ. The FZ station BV observes one of the largest 8t differences for earthquakes that are inside and outside the FZ. Figure 2.17 shows splitting measurements at station BV on map and cross-section views. For ray paths near the -65° north-dipping fault, the 61 values are much larger than those from south side of the fault. These observations indicate significant damage intensity (or crack density) in the dipping rupture zone below station BV. Although individual measurements of 51 range from 0.0 to 0.3 s, the average 8t for ray paths in Figure 2.16 inside the FZ is about 0.08 ± 0.05 s for about 3200 measurements. The associated apparent crack density e can be estimated using e = 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. /? {bt!L), where /3, 51 and L are the shear wave velocity of the uncracked medium, observed anisotropy delay time and propagation length in the anisotropic medium, respectively (Hudson 1981; O ’Connell & Budiansky 1974). The average value of the hypocentral distances for the -3200 measurements is -13 km. Assuming that Z-13 km (i.e., that entire distance), and using fi = 3.2 km/s based on the velocity model of Ben-Zion et al. (2003) for the material outside the fault together with our measured 8t - 0.08 s, give e ~ 2%. (A similar value can be obtained by dividing the measured S? with the observed -4.2 s average shear wave travel time of the -3200 measurements.) However, this value is a lower limit of the crack density because the anisotropy in this area is confined primarily to the top 3-4 km (Figures 2.12-2.15). If we use Z = 3.5 km for the average propagation distance inside the anisotropic medium, the apparent crack density is e - 7.3%. This value is close to the 10% limit of Hudson (1981), suggesting that the top 3-4 km of the Karadere-Diizce branch of the NAF is highly fractured. The -7% estimated crack density is similar to values measured near the San Andreas fault at Parkfield (Daley & McEvilly 1990) and a normal fault near Oroville, California (Leary et al. 1987), while being somewhat higher than the 5% value measured at the Hector Mine rupture zone (Cochran et al. 2003). We also note that 7% is lower than inferred crack density at volcanic or geothermal regions, such as the Long Valley caldera in California (Savage et al. 1990), southern Hawaii (Munson et al. 1995), and the Cerro Prieto Geothermal field in the Mexicali Valley, Mexico (Gonzalez & Munguia 2003). 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) 0.20 0.16 C /3 r °-12 .5 r o .o 8 h S 15 Q 0.04 EQ FZ STN FZ Regional 8f(s) N 0.078 ± 0.050 3201 0.065 ± 0.053 745 (b) 0.00 0.20 0.16 C /3 - 0.12 s t? 0.08 S 15 Q 0.04 EQ NFZ STN FZ Regional N 0.076 ± 0.046 2397 0.055 ±0.039 291 t f A . ' A A * + * A 0.00 FZNFZ Cl! PT LSMO FI AR VO FP TWTS CL BVWFCF BU SL GE Figure 2.16. Average delay times dt for earthquakes that are (a) located close or near the FZ and (b) outside the FZ (Regional). Triangles and diamonds denote the average 5/ for stations deployed for about 6 months and 2 weeks, respectively. Values for stations within, near and outside the FZ are shaded with dark, gray and white colors, respectively. Squares and stars mark the average dt values for all 12 stations near the FZ and those for the 5 regional stations, respectively. The vertical line at each point is the standard deviation of the result. The average and standard deviation of dt and total number o f events N for FZ and regional stations are marked on the top of each panel. 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.17. (a) Splitting parameters (bars) for station BV superimposed on the hypocentral locations within the shear wave window. The bars are oriented parallel to the fast polarization direction ( j ) and scaled by the delay time dt. (Note that the zones producing the anisotropy are likely to be considerably shallower (Figures 2.12- 2.15) than the earthquakes were the symbols are located.) Black and gray shadings denote results from earthquakes belonging to FZ and NFZ groups, respectively. Waveforms generated by the event marked by the star and the two events marked by the circles are shown in Figures 2.2 and 2.6, respectively, (b) Splitting parameters superimposed on the hypocentral locations projected along the cross-section BB' in (a) and Figure 2.1b. The dark lines denote the straight-line 45° angles that are used as the shear wave window. The dash line indicates the ~65° north-dipping fault that ruptured during the Diizce earthquake. Other symbols and notations are the same as in (a). 98 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0.2 s Almacik Block 30“ 52’ 30” 56’ 31'Off 3t”04’ 31*08’ 0 L------------- L - ---------L _ -------- L ------------ ± -------------1 --------_ J --------------- L. (b) - 4 - 2 0 2 Distance (km) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.5 DISCUSSION We analyze crustal anisotropy using shear wave splitting measurements from a large seismic data set recorded along the Karadere-Diizce branch of the NAF in the 6 months after the 1999 Izmit earthquake (Figure 2.1). We apply the Silver & Chan (1991) technique to determine splitting parameters of about 22000 records within the shear wave window (Figure 2.2). The time window associated with each measurement is automatically determined by a sliding window method (Figure 2.3). Based on 10 objective criteria, a quality of “high” or “low” is assigned to each measurement (Figures 2.4 and 2.5). Using the high quality data set, we find clear spatial variations of crustal anisotropy in our study area (Figures 2.6-2.11). The average fast directions 0 at most FZ stations for ray paths inside the FZ are parallel to and change with the direction of the fault strike nearby. The average delay times 51 for ray paths that propagate inside the FZ are larger than for other paths (Figures 2.16 and 2.17). These results suggest fault-parallel cracks or shear fabric as possible sources for the observed anisotropy near the fault. However, FZ stations TS and CL that are near geometrical complexities have 0 of about NW-SE, which is almost perpendicular to the local fault strike (Figure 2.10). Some FZ stations (e.g., LS, MO) have either scattered or bimodal fast directions. The observations indicate collectively significant variations o f FZ properties along the fault strike. The average fast direction 0 from ray paths that propagate inside the Almacik block is between 0° and 40° (Figure 2.10). The observed anisotropy within the block is most likely caused by lithologic properties such as foliation, bedding, or aligned 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. minerals. The average fast direction at station GE off the fault is close to the regional oh direction of about NW-SE. However, stations BU and SL have fast directions that are NE-SW, almost perpendicular to or- These observations indicate again strong spatial variations of anisotropy. Our results suggest that several different mechanisms contribute to the observed anisotropy in this area. Most stations do not show a clear increase of 51 with increasing depth and hypocentral distance, indicating that the anisotropy is confined primarily to the top 3-4 km of the crust (Figures 2.12- 2.15). Ben-Zion et al. (2003) performed a systematic analysis of seismic FZ trapped waves using the same data set and found that the trapping of seismic energy in the Karadere-Diizce fault is generated by -100 m wide FZ layer that extends to a depth of -3-4 km. Here we find that fault-parallel < |> exists at stations that are within several hundred meters on either side of the surface rupture (e.g., stations FI and FP), but not at stations that are several kilometers away (e.g., station GE). This suggests that a high density of microcracks exist in a broader region (e.g., kilometer wide) than the -100 m wide seismic trapping layers (Ben-Zion et al. 2003). A similar structure with a broad damage zone around a more intense narrower FZ layer was mapped in the field (Faulkner et al. 2003) at the Carboneras fault, Spain. Fialko et al. (2002) observed from InSAR data damage zones that are several km wide around faults in the eastern California shear zone. Various gravity, electromagnetic, and seismic imaging studies around large faults also indicate a few km wide damage zones (Ben- Zion & Sammis 2003, and references therein). Most of these studies do not resolve 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the depth extent o f the broad damage zones, but they are likely to be confined primarily to the top few km of the crust, as found here and in the studies o f Savage et al. (1990), Zhang & Schwartz (1994), Munson et al. (1995), Cochran et al. (2003) and Liu et al. (2004). We note that a depth of about 3 km corresponds to the inferred transition (Blanpied et al. 1991) from an aseismic velocity-strengthening behavior to a seismic velocity-weakening regime of rate- and state-dependent friction (e.g. Dieterich 1979, 1981; Marone & Scholz 1988). This provides a possible explanation for a change in the FZ structure from a broadly-deforming mechanically-passive stable shallow structure in the top 3 km, to a considerably narrower seismic structure in the deeper section. The inferred average crack density for ray paths with effective propagation distance of 3-4 km in the damaged shallow portion of the FZ is about 7%. It is possible that some crack-induced anisotropy exists at larger depth, as was found for example in the KTB deep drill hole at Germany (e.g. Rabbel 1994; Bokelmann & Harjes 2000). However, due to the increasing confining pressure the crack density should decrease in general with increasing depth (e.g. Boness & Zoback 2004; Liu et al. 2004). As discussed above, the crack density is also likely to have a strong reduction at the transition from a velocity-strengthening to a velocity-weakening frictional regime. Since the observed anisotropy is dominated by the shallow crust, the results do not provide in general information on properties o f the FZ structure at seismogenic depth where the bulk of seismic energy is stored and released. A similar conclusion holds for analysis of FZ trapped waves (Ben-Zion et al. 2003; Peng et al. 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2003). There has been a considerable controversy on the ability o f using temporal variations of crustal anisotropy to monitor precursory stress changes before major earthquakes (e.g, Crampin et al. 1990, 1991, 1999; Aster et al. 1990, 1991; Munson et al. 1995). Recent studies also claim that temporal changes o f crustal anisotropy can be used to monitor co-seismic stress changes (Saiga et al. 2003), post-seismic fault healing (Tadokoro & Ando 2002; Tadokoro et al. 2002) and volcano eruptions (Miller & Savage 2001; Gerst 2003). In complex regions, like the one examined here, where several different mechanisms may contribute to the observed anisotropy, temporal variations, even if such exist, may be contaminated by variations of ray paths due to the changing seismicity. The present paper was focused on the development of an automatic and objective shear wave splitting methodology that can be applied to a large data set, and deriving general spatial properties of the crustal anisotropy in our study area. Detailed space-time variations of crustal anisotropy around Karadere-Diizce branch o f the NAF using similar earthquake clusters are discussed in a follow up paper (Peng & Ben-Zion, in preparation, 2004). The results from the similar earthquake clusters show, in agreement with the present study, clear large spatial variations of crustal anisotropy. Splitting parameters measured from similar earthquake clusters indicate at most 2% co-seismic changes of delay times associated with the occurrence of the Duzce mainshock. Small co- and post-seismic changes o f seismic properties are also revealed by analysis of relative travel times and evolving 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. decorrelation of P- and 5-coda waves of similar earthquake clusters. The results do not show systematic precursory changes before the Diizce mainshock. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CHAPTER 3 (Peng & Ben-Zion 2004b) SPATIO-TEMPORAL VARIATIONS OF CRUSTAL ANISOTROPY FROM SIMILAR EVENTS IN AFTERSHOCKS OF THE 1999 M7.4 IZMIT AND M7.1 DUZCE, TURKEY, EARTHQUAKE SEQUENCES SUMMARY We analyze spatio-temporal variations of crustal anisotropy along the Karadere- Duzce branch of the north Anatolian fault from similar earthquakes in aftershocks of the 1999 Mw7.4 Izmit and Mw7.1 Duzce earthquake sequences. The similar earthquake clusters are identified using a waveform cross-correlation technique. Depending on the applied similarity criterion, about 4-60% of over 18000 earthquakes belong to similar event clusters. Splitting parameters averaged within each cluster show significant variations for slightly different ray paths, indicating strong spatial variations of crustal anisotropy in this area. Apparent temporal changes of up to 30% shear wave splitting delay times are observed at stations near the epicentral regions before and after the Diizce mainshock. However, the changes can be mostly explained by the spatial variations of ray paths due to the changing seismicity, instead of changes in properties of the anisotropic medium. Splitting parameters measured within similar earthquake clusters indicate at most 2% changes in delay times associated with the occurrence of the Duzce mainshock. The results do not show systematic precursory changes before the Dtizce mainshock. 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.1 INTRODUCTION Large crustal faults in nature occur as zones with highly damaged materials and intense microcracks that have lower seismic velocity than the surrounding rocks (Ben-Zion & Sammis 2003, and references therein). If low-velocity fault zone (FZ) rocks are spatially coherent, they can act as seismic waveguide and produce after the direct S wave arrivals large-amplitude and low-frequency wavetrains that are termed F Z trapped waves (e.g. Ben-Zion & Aki 1990; Li et al. 1994). If microcracks around active FZs are aligned preferentially, seismic shear waves propagating inside such damaged FZ rocks are expected to split into two orthogonally polarized waves with different velocities. This phenomenon is analogous to optical birefringence and is termed shear wave splitting. Two routinely determined splitting parameters are the polarization direction of the fast wave (< j)) and the delay time (8t) between the fast and slow waves. Both FZ trapped waves and shear wave splitting are seismic manifestations of damaged low-velocity FZ layers. These two signals, in turn, can be used to image internal structures and properties of the FZ at seismogenic depth. An accurate determination of subsurface fault zone (FZ) properties can improve the understanding of many aspects of earthquake physics, long-term evolution of faults, seismic wave propagation and seismic hazard. Shear wave splitting is claimed to be an effective tool to detect spatio- temporal variations of crustal anisotropy around active FZs and to monitor stress changes associated with major earthquakes (e.g. Crampin & Chastin 2003). However, detailed studies have long been controversial. Several studies report temporal 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. changes of splitting parameters before (e.g. Crampin et al. 1990, 1999; Gao et al. 1998), during (e.g. Saiga et al. 2003) and after (e.g. Tadokoro & Ando 2002) major earthquakes or swarms. Other studies observe no clear change of anisotropy parameters near epicentral regions of several recent major earthquakes (e.g. Aster et al. 1990; Savage et al. 1990; Muson et al. 1995; Cochran et al. 2003; Liu et al. 2004a, b). Many o f the previous studies on temporal changes o f crustal anisotropy are based on analysis of waveforms with mixing ray paths. Peng & Ben-Zion (2004) and Liu et al. (2004a,b) have shown that spatial variations of anisotropy can be mapped into temporal changes due to the changing seismicity. One effective way to separate spatial variations from temporal changes is to analyze shear wave splitting from similar earthquakes (e.g. Aster et al. 1990; Bokelmann & Harjes 2000), which are located very close in space and generate nearly identical waveforms recorded at surface stations (e.g. Poupinet et al. 1984; Nadeau et al. 1994). To better understand the spatio-temporal variations o f crustal anisotropy along an active fault, we examine seismic waveform data along the Karadere-Diizce branch of the north Anatolian fault (NAF) from similar earthquakes in the aftershocks of the 1999 Izmit and Dtizce earthquake sequences. Because these two major earthquakes occurred within 3 months with thousands o f aftershocks recorded by many stations along and around their rupture zones, the Karadere-Diizce fault is an ideal place to study the spatio-temporal variations of near-fault anisotropy. Peng & Ben-Zion (2004) have performed a systematic analysis of shear wave 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. splitting along the Karadere-Diizce faults and found an approximately 1 km broad zone around the rupture zone with fault-parallel cracks or shear fabric. The belt of anisotropic rock is confined primarily to the top 3-4 km of the upper crust and around the -100 m wide seismic FZ waveguide (Ben-Zion et al. 2003). Their results also suggest large spatial variations of crustal anisotropy and multiple mechanisms in the area. This study is a follow-up work of Peng & Ben-Zion (2004). Here we focus on the fine-scale spatio-temporal variations of anisotropy from similar earthquakes in the same area. In the following sections, we first describe the methodology to identify similar earthquake clusters. We then combine the shear wave splitting measurements from Peng & Ben-Zion (2004) with the similar earthquakes obtained in this study and investigate the fine-scale spatial patterns and possible temporal changes of crustal anisotropy in detail. 3.2 DATA A week after the August 17, 1999, Mw7.4 Izmit earthquake, a temporary 10- station PASSCAL seismic network was deployed (Figure 3.1) along and around the Karadere-Diizce branch of the NAF (Ben-Zion et al. 2003; Seeber et al. 2000). Three months later, the November 12,1999, Mw7.1 Duzce earthquake occurred near the end of the rupture zone of the Izmit earthquake and propagated further east. Our temporary seismic network straddled the rupture zones of both mainshocks and recorded about 26000 earthquakes during its six-month operational period. In this study we focus on -18000 earthquakes that are located within -2 0 km of the network 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and are recorded by at least 3 o f the 10 relatively long-term stations. Additional details on the experiment and data set are given by Seeber et al. (2000) and Ben-Zion et al. (2003). Depfli range 0 ~ 5 km ! 5 ~ 10 km j 4 1 ' 0 0 ' 40’ 50' Sapanea L a te . BU 40“ 40' - km 40* 30' 30" 20’ 30" 30' 30*40' 30*50’ 31*00* 31 10' 31° 20' Figure 3.1. Hypocentral distribution of ~18000 earthquakes recorded by the PASSCAL seismic experiment along the Karadere-Diizce branch of the NAF. Aftershock locations are marked with colors denoting different depth ranges. Shaded background indicates topography with white being low and dark being high. The surface ruptures of the Izmit and Dtizce earthquakes are indicated with thick dark and grey lines, respectively. Dark thin lines denote active faults with time of recent large ruptures. Stations within, near and outside the FZ are shaded with dark, gray and white triangles, respectively. Gray squares denote locations of nearby cities. The inset illustrates the tectonic environment in northwestern Turkey with the box corresponding to our study area. Arrow vectors represent plate deformation rate (Reilinger et al. 1997) from GPS data. Modified from Peng & Ben-Zion (2004). 3.3 ANALYSIS PROCEDURE Similar earthquakes in our data are identified by performing cross-correlation calculations on waveforms generated by event pairs with hypocentral separations < 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 km (Aster & Scott 1993). The cross-correlation is done over a time window of 0.5 s before and 1.5 s after the P wave arrival for the vertical-component seismogram, and 1 s before and 2 s after the S wave arrival for the horizontal-component seismograms. Figure 3.2 shows examples of north-component waveforms at station FP having different values of correlation coefficients (CC) with respect to the first waveform. Generally, waveforms with CC greater than 0.7 have similar shapes and are likely to be generated by earthquakes that have similar locations and focal mechanisms. Figure 3.3 gives the distribution of the CC values and correlation lag times for all north-component waveform pairs recorded at station FP. Most CC values are distributed between 0.2-0.5, indicating that the corresponding waveform pairs are poorly correlated. Flowever, there is a small percent of CCs distributed above 0.7, suggesting the existence of group of events that generate similar waveforms. The wide spread of the lag times at high CCs indicate that the automatically picked P and S wave arrivals that are used for initial alignment of the waveform pairs are not self- consistent and accurate. Next, we use for each pair o f events the median value of the CCs at all recording stations as the measure of the similarity (P) between these two events. The median is chosen instead o f mean value since the mean is sensitive to some outliers with very low CCs (Aster & Scott 1993). Figure 3.4 shows values o f the median CCs versus hypocentral separations between pairs of events for -18000 earthquakes. Most median CC values are distributed between 0.2-0.5 even for event pairs with 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. small hypocentral separations (e.g. < 2 km), indicating that events with similar locations can generate quite different waveforms, possibly due to different focal mechanisms. On the other hand, the existence of high median CC values (e.g. > 0.9) with large (e.g. > 5 km) hypocentral separations suggests that the original hypocentral locations with standard HYPOINVERSE determinations (Klein 1978) may contain errors. Finally we organize the earthquakes into clusters of similar events using an equivalency class (EC) algorithm (e.g. Press et al. 1986). The EC algorithm identifies pairs of events that satisfy a given similarity criterion ((3 > pc) and groups such pairs that share the same event into similar event clusters. For example, if event pairs (A, B) and (B, C) both satisfy the criterion, A, B and C will be grouped into one cluster, regardless of the similarity measure between A and C. In practice, we include an event pair in a cluster only if the waveforms of the pair were recorded by 3 or more stations and the median CC of the waveforms is 0.5 or more. The number and size of the obtained similar earthquake clusters vary depending on the similarity criterion pc. Figure 3.5 illustrates that dependency for -18000 earthquakes: higher pc values produce less clusters and less events in each cluster but with more similar waveforms; lower pc values result in more clusters and more events in each cluster, but with less similar waveforms. For pc values between 0.95-0.70, approximately 4-60% of the events belongs to similar event clusters (Table 1). The resulting percentage range is slightly higher than those from the aftershock sequence of the 1994 Northridge, California, earthquake (Shearer et al. I l l Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2003) and microearthquakes recorded by the Anza seismic network (Aster & Scott 1993), but lower than the values from recent studies of seismicity at the Parkfield and other active strands of the San Andreas and Hayward faults in central and northern California (e.g. Nadeau et al. 1995; Rubin et al. 1999; Schaff et al. 2002). 325032249 ft M a i 0 -9 7 V ^ v 0.90 /V jY v v A r^ '^ W v V O /V -'A A - 346184043 320052426 273002711 0.70 267061358 — A^^~AvAr'^/'''- '''A w A w v /iY v 7 'r-A -'v V y 0.60 308042509 W V y 'V -V ' A i - W - ' S / v v V V - v - ' - 'V v * / A A /V T'/'/ V ' V 332193811 344184210 0.40 0.3° 033230014 2 3 1 0 1 ■ 3 ■ 2 T im e (s ) Figure 3.2. Examples of waveform similarity for north component seismograms recorded at station FP. The short solid vertical bars denote the routinely picked P and S arrivals. The event ID numbers are marked on the left and consist of 3-digits Julian day, 2-digits hour, 2-digits minute, and 2-digits seconds of the earthquake occurrence time. Julian days in the range 237-365 are in year 1999 and those in the range 001- 042 are in 2000. The waveform generated by each event is cross-correlated with that of event 325032249 over a time window of 1 sec before and 2 second after the S arrivals. The corresponding correlation coefficient is marked on the right o f each trace. il Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Station PP, N comp, 17432343 pairs, 0.8% ioglO(N) 2 4 6 1.0 0.8 S 0.6 zt > u U 0.4 -« 0.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 Time lag (s) Figure 3.3. (a) The correlation coefficients (CC) for ~17-106 north-component waveform pairs recorded at station FP versus their time lags. Contours show the logarithmic number of event pairs for given CC and time lags. The vertical and horizontal lines mark the zero lag time and CC value o f 0.7, respectively, (b) Histogram o f the CC. (c) Histogram o f the time lags. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18556 events, 30359833 pairs loglO(N) 2 4 6 > 0.6 ■ 0.0 -+**•« 0 2 4 6 8 10 Hypocentral separation (km) 0 5 10 15 20 25 Station numbers Figure 3.4. (a) The similarity measure P for -30T 06 pairs of waveforms recorded at station FP versus their hypocentral separation D between pair o f earthquakes. Contours show the logarithmic number of waveform pairs for given p and D. (b) Histogram of the similarity measure p. (c) Histogram of the hypocentral separation D. (d) Histogram o f the number of recording stations between pairs of events. The vertical line marks the cutoff value of at least 3 stations that are used in the similar earthquake identification analysis. 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 S ? 4 o 3 m Q 2 18556 events, 18924 i 9 pairs (a) 10000 1000 S h £ a 3 z 100 ooc)oocn3cm? ( b ) * * * * * * * * * * 10 O ncv O nclust # nev5 ★ nclu-stS 0.6 0.7 0.8 0.9 Cross-correlation value 1.0 Figure 3.5. (a) Histogram of the similarity measure b between pair o f events that are recorded by at least 3 stations between pairs of events and p ^ 0.5. (b) Number of events that belong to a cluster (nev), number of events belonging to a cluster with 5 or more events (nev5), total number of similar event clusters (nclust), and number of clusters with at least 5 events (nclust5) as a function of the similarity criteria pc. 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3.1. Number of Clusters and Similar earthquakes (Total events: 18556)a P c nclust nclust5 nev nev5 percentage % 0.95 257 40 803 326 4.3 0.90 1068 109 3242 1011 17.5 0.85 1712 259 5834 2406 31.4 0.80 2093 349 8023 3824 43.2 0.75 2125 415 9387 5618 52.7 0.70 2000 442 11366 7566 61.3 a where pc is the similarity criterion, nclust is the total number o f similar event clusters, nclust5 is the number of similar event clusters with at least 5 events, nev is number of events belonging to similar event clusters, nev5 total number o f events that are similar to at least 5 or more events, and percentage is calculated using the relation nev/total. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.4 RESULTS 3.4.1 Spatio-temporal evolutions of the Similar Event Clusters and Seismicity Figures 3.6 and 3.7 show locations of similar event clusters for similarity criteria pc = 0.95 and 0.70, respectively. The events with pc = 0.95 are likely to rupture the same patch of the fault repeatedly and are termed repeating earthquakes (e.g. Nadeau et al. 1995; Igarashi et al. 2003). Lower pc value (e.g., 0.70) produce similar event clusters that are located close to each other and have similar focal mechanisms, but may not rupture the same patch (e.g. Lees, 1999; Shearer et al. 2003). Since the temporal variations of splitting parameters are very sensitive to the changes of earthquake locations, we use similar earthquake clusters identified with pc = 0.95 in the following analysis of temporal changes of crustal anisotropy. The larger number of clusters with pc = 0.70 are used to investigate the fine-scale spatial patterns o f crustal anisotropy. We note that the similar event cluster with pc = 0.95 around the Diizce segment (cross section BB’) are distributed throughout the seismogenic zone. In contrast, the similar event clusters with pc = 0.95 along the ~30-km Karadere segment (cross sections AA’ and CC’) are confined primarily to the depth range 12-15 km and near the bottom of the seismogenic zone. The seismicity in our data set exhibits complex spatio-temporal evolution that can affect our inferences on changes of anisotropy and other material properties. The occurrence of the Diizce mainshock produces changes of stress that are expected to increase the seismicity rate in some places and decrease it in others. Figure 3.8 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. illustrates such change of seismicity rates in our data sets following the Diizce mainshock. The seismicity rate change at time t over grid cell o f size 1.2 x 1.2 km is represented by a statistical Z-value (Habermann 1983; Wiemer 2001), where positive and negative numbers indicate decrease and increase of seismicity rate, respectively. It is seen that the seismicity rate increases around the east portion o f the Diizce rupture zone and to the north of the west part of the mainshock, while it decrease south o f that part and further to the west. In Section 3.4.3, we present delay times between fast and slow shear waves derived from seismic records at various stations over about 6 month. Such data are based on different source locations and cannot be used to derive temporal changes of anisotropy properties. In Section 3.4.4, we analyze temporal changes of anisotropy properties using repeating event clusters. Figure 3.9a shows the occurrence times of earthquakes belonging to 42 clusters of Figure 3.6 with at least 5 events and similarity criterion pc = 0.95. The seismicity rates o f these repeating earthquake clusters typically also show changes around the occurrence time of the Diizce mainshock. Figure 3.9b shows the inverse of the time intervals between consecutive repeating events, or the inverse of recurrence times, for 9 similar event clusters (C01-C05, C07, C l 1, C12 and C l4) located near the Diizce mainshock (Figure 3.6). The inverse of recurrence times is proportional to the elapse times of these events after the Diizce mainshock. The results are compatible with the Omori law for aftershock delay rates. Similar patterns are observed in the aftershock zones of the 1984 Morgan Hill earthquake (Vidale et al. 1994) and the 1989 Loma Prieta earthquake (Schaff et al. 1998). 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.6. A map view (a) and three cross-sections (b-d) showing repeating earthquake clusters (dark) with similarity criterion |3C = 0.95 with respect to the overall seismicity along the Karadere-Dtizce branch of the NAF. The surface ruptures o f the Izmit and Diizce earthquakes are indicated with thick gray and dark lines, respectively. The dash lines in (b) and (c) indicate the -80° north-dipping fault along the Karadere segment that ruptured during the Izmit earthquake and the -65° north-dipping fault ruptured during the Diizce earthquake, respectively. Average hypocentral locations o f cluster CIO are denoted with yellow circles. Splitting parameters for earthquakes belonging to this cluster are shown in Figures 3.14-3.16. Other symbols and notations are the same as in Figure 3.1. 119 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40’ 55' 40° 50' 40° 45' 40° 40* 40° 35' 30° 20' 30“ 30' 30“ 40' 30° 50' 31“ 00 31° 10' 31° 20' 'v : -10 -5 0 5 10 Distance (km) j 1 V , 120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I I I I p - -10 -5 0 5 10 Distance (km) Figure 3.7. A map view (a) and three cross-sections (b-d) showing similar earthquake clusters (dark) with similarity criterion pc = 0.70 with respect to the overall seismicity along the Karadere-Duzce branch of the NAF. Other symbols and notations are the same as in Figure 3.6. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.8. Locations of the earthquakes occurred before (a) and after (b) the Diizce earthquake. Aftershock locations are marked with shades denoting different depth ranges. The surface ruptures of the Izmit and Diizce earthquakes are indicated with gray and dark lines, respectively. The star denotes the epicentral location of the Diizce mainshock. Three triangles denote the location of station CH, MO and BV. (c) Z-value map of the seismicity rate change before and after the Diizce mainshock. Positive Z-values (dark) represent a decrease in the seismicity, while negative values (gray) represent an increase. This map is generated using the ZMAP software package (Wiemer 2001). 122 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40.5 - (b) 4 ! • 4 0 .9 - 6 » -S . 40.8 * 40.' 40.6- 40 ,5 - (c) 4 ! • 4 0 ,9 ■ ’ 4 0 .8 - 4 0 .7 ■ 4 0 .6 ■ 4 0 .5 - <&■ 3 0 .4 z value 3 0 .6 3 0 .8 3 1 31.2 L o ra h w te {rigg] /? 0 A w 2t ! km 2 < 7 J b n z < 1 4 ,4 k m z<M k m 31.4 31.6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.9. (a) Occurrence times (Julian day since 1999) of the repeating earthquake clusters with at least 5 events with similarity criterion |3C = 0.95. The two arrows mark the origin time of the izmit and Diizce earthquakes, respectively, (b) Inverse of the time intervals between consecutive repeating events as a function o f their occurrence time for 9 clusters near station BV. The dash line denotes the 1 It Omori’s law time decay. 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) Q 1 Q 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 8 6 4 j nev5 10 (b) % •X 3 u - fe: 0.1 o CO OCD 326. cc ^ o o ooo o o o OOO 0 oo o oo o o o (BOO 0.95 o o o o o o o o o o o o 0 o o o o o o o o o o ooo o o o o o o o o o o o o o o o o o o o O D o OOO o o o o o ooo o oo CD OO O ooo o o o o o o o o o o o o O O O 00 o O D O O O O o o o o o o o o o o 0 0 o CO OO C D O m o o o o o o o o o O D D O O O O O o oo 00 o C X O D O O O O O ooo o o o o o o o o o O o o o GEO O o oo o o o r«nmnno o O o o o o p . m m n o o d o o o o _ 3 C O O O O o im ooaD o o o coomo o o o o o o o o o o o o 250 300 — I — 350 400 Julian day t- u 1 £ - 1 1 - 3 1 - 1 2 - 12 - i; 13 id le- is - 0.1 1 10 Days after the Diizce mainshock 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.4.2 Fine-Scale Spatial Variations of Anisotropy Due to uncertainties in the shear wave splitting measurements, previous studies of crustal anisotropy often average splitting parameters based on source- receiver locations or spatial regions (e.g. Cochran et al. 2003; Peng & Ben-Zion 2004). However, as shown in Figure 3.4, earthquakes from similar regions can generate dissimilar waveforms with different initial polarizations o f shear waves (e.g. due to different focal mechanisms). Peng & Ben-Zion (2004) point out that shear waves with different initial polarizations may sample different sets of microcracks in a complex region, resulting in scattered or bimodal distribution of splitting measurements. Thus averaging splitting measurements based solely on spatial regions may reduce from the resolution of the obtained results. A better way to group and present the splitting measurements is to average splitting parameters within each cluster that have similar waveforms, similar ray paths and similar initial shear wave polarizations. To obtain a high-resolution spatial distribution of crustal anisotropy, we average the high quality splitting parameters of Peng & Ben-Zion (2004) within each similar earthquake cluster with the similarity criterion pc = 0.70. We use the von Mises method to calculate the mean angle o f the fast direction and a mean resultant length R (Davis 1986; Mardia & Jupp 2000; Cochran et al. 2003). The parameter R gives a quantitative estimate of the variance of the directional data, with values near 0 and 1 indicating high scattering and clustering, respectively. 126 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We select clusters that have 4 or more splitting measurements with relatively small variance (e.g. R > 0.4). We then present the average splitting parameters in the following two ways. First, average splitting parameters for 9 of 10 long-term stations are plotted on top of the centroid location of each cluster (Figure 3.10). The results for station BU are not shown because we have only two clusters with more than 4 high quality measurements at this station. For stations CFI, FI, VO, FP and BV that are inside or within ~1 km of the surface ruptures of the Izmit and Diizce mainshocks, the dominant fast directions ( < j > values) are parallel or sub-parallel to the direction of the nearby fault strike. However, stations that are within the Izmit rupture zone (e.g. LS, MO and VO) generally have more scattered fast directions. The results show clearly that earthquake clusters with different ray paths produce quite different < j > values. This observation indicates strong spatial variations of crustal anisotropy in this area. Second, average splitting parameters for each cluster are plotted together on top of the recording station (Figure 3.11). The average splitting parameters for the same group of earthquakes within each cluster are different at stations separated by a few km, indicating that the observed anisotropy is not originated near the source. In addition, we observe quite different splitting parameters for several clusters (e.g. C02, C03 and C390) at FZ station pairs (MO and FI, VO and FP) that are located only several hundred meters apart. Peng & Ben-Zion (2004) found no clear dependency of splitting delay time 81 with increasing depth or hypocentral distance for most stations, and concluded that the anisotropy is confined primarily to the top 3-4 km of the crust 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. around the Karadere-Diizce faults. The observation of variable splitting parameters over short distances (e.g. several hundred meters) indicates that near-station fault structures play an important role in producing shear wave splitting and further support the shallow anisotropy interpretation in our study area (Peng & Ben-Zion 2004). For ray paths propagating inside the Almacik block (e.g. C01), stations LS, MO, FI and WF record similar splitting parameters, indicating a relatively uniform anisotropy within the block that is possibly caused by lithologic properties (Peng & Ben-Zion 2004). In summary, splitting parameters for earthquakes within the same cluster generally are very similar (e.g. about 70% of the clusters have R value > 0.8 and the standard deviation of the average delay times <0.03 s). However, different splitting parameters are observed for earthquake clusters that are located nearby, and for stations that located within several hundred meters away. The overall spatial variations based on similar earthquakes are consistent with the results of Peng & Ben-Zion (2004) and confirm that multiple structures and mechanisms contribute to the observed crustal anisotropy in our study area. The observation of large spatial variations of crustal anisotropy will play an important role in explaining the apparent temporal changes o f splitting parameters in the following section. 3.4.3 Apparent Temporal Variation o f Anisotropy Since our temporary seismic network straddles the rupture zones of both the Izmit and Diizce mainshocks, many tens of thousands of waveforms were recorded during its six-month operational period (Seeber et al. 2000; Ben-Zion et al. 2003). 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.10. Average splitting parameters (bars) for the 9 long-term stations. Splitting parameters are averaged within each earthquake cluster with at least 4 individual measurements and a mean resultant length R > 0.4. The bars are oriented parallel to the average fast direction < J > and scaled by the average delay time 51 . The center of each bar is plotted at the average source epicenter with shades denoting different depth ranges. Stations within, near and outside the FZ are shaded with dark, gray and white triangles, respectively. The total number o f the earthquakes and clusters for each station are marked in the corresponding panel. The surface ruptures of the Izmit and Diizce earthquakes are indicated with think gray and dark lines, respectively. 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. o 40' 45'- 40' 40' 35*- 40*45'- 40* 40' - 40" 50’ 40' 45‘ 40" 4t)‘ ( II 5 0 0 e v e n ts, 38 d u s te r s i , - - cu 4 5 ' 40 4(F- 0.1 s 30 35' 30* 40’ .30*45’ F I 170 e v e n ts, 12 c lu s te r s |4o~ 5f r OB I .S 333 e v e n ts, 1 1 c l u s t e r s ^ \ y v . FP ' i s Vo jW F C lt \ \ km « ......... 5 0 ,1 s 40' 45 40'- 4 O'' Mr 40* 50" 45' 30*59* 30* 55’ IS i f - .ir"^ FI WF M K T 45‘ - 0 0 .1 ? 5 Ho*«- 30' 45' 30’ Sir 30' 5? V O 359 e v en ts, 31 d u s te r s « Jlfefc Y „ J ~ * A " \* ' BV ^fs,3 |V /V u V FI ,W f km 0 5 0.1 s 30* 43’ 30* 50* 30* 55* * Off B V 242 e v e n ts, 29 d u s te r s 40* 45*- 40* 40* W F 178 e v e n ts, 12 c lu s te r s . / G E h. / " \ _ FP M * Ts /-vo H WF km 0 5 * 0.1 s 40' 50*- 40' 45 40" 40' 40 - 50’ 40" 45' ■ M O 270 e v en ts, 16 d u s te r s O K - I v V f ' 'W .w r km CU s , ? f_ 30* 45' 30' 50* 30* 55' F P 312 e v e n ts, 22 c lu s te rs O B C m o WF km 0 5 oj s x r s o 3 0 ' 55' s v 'm r ( IE 169 e v e n ts , 8 d u s te r s I k f ir h r a tin e \ > 1 Nj,ri ^ i i » h i h » 1 5 r t * htn GB FP Jar ‘ L S FI km W F 0 .1 s 30” 55* 3 1 ” 00* 3r05* 30" 45' 30" 50* 30“ 5$’ 31" 00' 30" 45’ 3{)' 50’ 30’ 55’ £ r : Tt 5 % ^ • r r t ? ir '* r Figure 3.11. Average splitting parameters (red bars) for each of 9 clusters. The center of each bar is now plotted at the recorded station. The star in each panel denotes the average epicenter of that cluster with shades denoting different depth ranges. The total number of measurements and recording stations are marked on top of each panel. Other symbols and notations are the same as in Figure 3.10. 131 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This allows us to evaluate carefully the hypothesis that temporal changes of anisotropy are correlated with the times o f a major earthquake (e.g. Crampin et al. 1999). We first check splitting results at station BV, which is very close to the hypocentral region of the Diizce mainshock and inside its surface rupture zone. Figure 3.12 compares hypocentral locations and high-quality splitting parameters of -700 earthquakes around station BV before and after the Diizce earthquake. There appears to be a 0.02 s (-30% ) increase of the average delay time 8t and a -7° rotation of the average fast direction c j ) at the time o f the Diizce mainshock. However, as shown in Figure 3.12a, the source locations have also changed considerably (there is a shift of -7 km in the average hypocentral locations before and after the Diizce mainshock). Peng & Ben-Zion (2004) found that the average 8t for station BV from ray paths along the -65° north-dipping mainshock rupture zone (Utkucu et al. 2003) are much larger than those from the south side of the fault. Thus, the observed 30% co-seismic change of splitting parameters is most likely caused by spatial variations associated with changes of event locations, rather than by temporal changes of the anisotropic medium. Studies attempting to detect temporal changes of anisotropy typically use 5- point running average of the data (e.g. Crampin et al. 1999; Siaga et al. 2003). Such a line in Figure 3.12b suggests a slight increase of delay time St about two weeks before the mainshock. However, similar fluctuations are observed at other periods, and the more refined analysis in Section 3.4.4 with repeating events does not show a similar temporal changes. 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.13 shows the temporal change of delay times at 8 other stations. Result for station GE is not shown since most of the east component waveforms Figure 3.12. (a) Splitting parameters (bars) for station BV superimposed on the hypocentral locations o f -700 earthquakes. The bars are oriented parallel to the fast polarization direction < |) and scaled by the delay time 51 . Gray and dark shades denote results from earthquakes before and after the Diizce earthquake, respectively. The rose diagrams of the fast directions, total number of the measurement (N), average fast direction (0) and the mean resultant length (R) are marked on the top left comer, (b) Delay times measured at station BV plotted against the earthquake occurrence times (Julian day since 1999). The two arrows mark the occurrence time of the Izmit and Diizce earthquakes, respectively. Small gray triangles and dark circles denote measurements before and after the Diizce earthquake. The bigger gray triangle and dark circle with vertical lines give the mean and standard deviations of all the measurement before and after the Diizce earthquake, respectively. The thin dark lines denote the 5-point running average of the delay time measurement, respectively, (c) Fast directions plotted against earthquake occurrence times. Other symbols and notations are the same as in (b). 133 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40° 52 (a) 40' 48’ 40° 441 N 6 R m 104.4 0,43 Duzce 3 5 0 9 6 .6 0 .5 2 X Diizce Vi/1.1 e ' * BV- Eften Lake * ...... Almacik Block v \ * (b) 30s 52’ 30° 56’ 31° 00' 31° 04’ 31* 08’ 31° 1 " BV: -0l022 s, 30.1 %, N = = 744 3 > C c >> ee 0.3 0.2 < u o i Q 0.0 * • » * mm a m • • • # * * ♦ + m * M e • « * m # m m . m t m • • • m m » *® • m m m m • * •# • * * m 4M • • • « «88 e « . mmm #• f». » ..»» _ jm tr ■m m • mm m m m m mm ® . *#«»«• m «• V*. | * » »* % •* • t 120 * * * , « • ,* * * * »/ * * * * * * $ ' .b 250 300 350 Julian day 400 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0,3 0.2 0.1 0.0 0.3 0.2 0.1 0.0 0.3 0.2 0.1 0.0 e ft O 0.3 - H 0.2 - < 5 O.t - Q 0.0 ' CH.: 0.006 s, £ ■ “ #*;* s f * j& A * • - 6.7 % ,N = 1286 ' . * 9 * _ & 9m » • f * * r* ® tp ® ^ ® ® 9 ....** fc J S M g L -9 LS: 0.006 s, 8.0% , M = 586 ( ^ . v [ ^ ^ = 679 t ■ ■ i ' i # « * t + * * jL * * j f FI: -6.000 s, 0.8 %, N = 361 | 1 . . 1 . • • * 1 VO: 0.002 s, > S S S i t T - r 2.4 %, N = 782 ’ 9 % **» * S j S ' j T ' i s - - ; FP: 0.000 0.4 %, N = 667 * + 9 m I , ‘ y .k % • ■ 1 . a s s , x J® • * ! ► • • • • * 9m 9 m ■ WF: 0.002 s, 2.7 %, N = 423 ‘ 1 - L % 1 * * * ' ‘ * * * • . , « - ; , | «•** vu i* ------ f ----- T ------------------r-------- ------- J " BU: 0.011 s, M .l %,N=*172 ‘ * * 1 • "• ’ * “ * * * L - • \ \ . --........n — , • m ,t --r...------------- * • ' 250 300 350 400 Julian Day 250 300 350 400 Julian Day Figure 3.13. Delay times plotted against the earthquake occurrence times for 8 stations. Other symbols and notations are the same as in Figure 3.12. before the Diizce mainshock were not recorded correctly at this station. Changes of average delay times ranging from 1-10% are observed before and after the Diizce mainshock for these 8 stations. However, the delay time measurements are spanned over large ranges (up to 0.3 s), and as shown in the next section, the apparent co- 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. seismic changes are dominated by spatial variations of seismicity and ray paths. 3.4.4 Fine-Scale Temporal Variations of Anisotropy The results from Section 3.4.1 show that the occurrence of the Diizce earthquake significantly changes the spatio-temporal seismicity patterns along the Karadere-Diizce faults. We have also shown that there are large spatial variations of crustal anisotropy in this area (Section 3.4.2). The large spatial variations of anisotropy can be mapped into temporal variations by the changing seismicity (Section 3.4.3). To examine more closely temporal variations of anisotropic properties, we now analyze splitting parameters generated by repeating event clusters with high similarity criterion (pc = 0.95). In general, there are two ways to estimate temporal variations of shear wave splitting delay time. The most common way is to first calculate the delay time 8t between the fast and slow waves for each event and then compare the obtained delay times to a reference measurement. Figure 3.14 shows splitting parameters measured from 10 earthquakes that belong to the repeating earthquake cluster CIO. To obtain a sub-sample accuracy, we interpolate the waveforms from 100 samples to 10000 samples per second using the SAC routine “interpolate” (e.g. Niu et al. 2003). The S wave arrivals are realigned by matching waveform to that of the first event. Figure 3.14(b) shows an increase of -2.5 ms for the average delayed times 51 before and after the Diizce mainshock, corresponding to a -1.4% anisotropy change with an average 5t value of 0.182 s. However, the differences of S -P times exhibit a similar 136 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. pattern (Figure 3.14a), suggesting that the subtle change of the 51 values could still be caused by spatial variation of the event locations. Another way to measure temporal changes of delay times is to start with calculating the relative time differences of the fast and slow waves separately (Bokelmann & Harjes 2000) as A(8f) = A(4 - tf) = (4 - tj) - (4 - tjf** = (4 ^ - - (tf R E F - 4) = A(4) - A(4), where tf and 4 represent the travel time of the fast and slow waves, respectively. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.14. Splitting measurements from the 10 events in cluster CIO for station FP. The centroid location of this cluster is shown in Figure 3.6. (a) The S -P times obtained from the waveform cross-correlation alignment with the first earthquakes. The average S -P times before and after the Dtizce earthquakes are marked on the left. The two arrows mark the occurrence time o f the Izmit and Dixzce earthquakes, respectively, (b) Delay times plotted against event occurrence time for the 10 earthquakes. The vertical line denotes the standard deviations of each measurement. Other symbols are the same as in (a), (c) Fast directions plotted against the event occurrence time. Other symbols are the same as in (a). 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) m 0) cu I 00 .88 0.20 (b) 0.19 < D S £>0.18 "o Q 0.17 70 ( c ) W ) < 3 J x s 60 - c . 2 50 - 4 —* f i o 1 — < . 1 — I TS 40 - 30 - 20 Station FP, Cluster 10, 2.32 ms • • Station FP, Cluster 10, 2.36 ms Station FP, Cluster 10, -0.40' 240 270 300 330 Julian day 360 390 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.15 illustrates how we measure the relative time difference of the fast and slow waves recorded at station FP for the 10 events in cluster CIO. We first rotate the two horizontal seismograms into fast and slow components using the average fast direction obtained in Figure 3.14(c). The use of the average direction is justified by the fact that the observed fast direction remains nearly constant for most similar event clusters (e.g. Figure 3.14c). Since both the origin time of an event and the absolute timing o f a seismogram may contain errors that are larger than several milliseconds, seismograms are aligned by matching waveform of the P phases to that of the first event. The relative time difference for each set of fast and slow shear waves from those of the first event are then calculated by waveform cross correlation. The obtained high cross-correlation coefficients (> 0.98) indicate that the waveform shapes of the fast and slow waves for these events are nearly identical. In addition, the relative travel time changes of the fast A ( 0 and slow A(^) waves (Figure 3.15c and 3.15d) are on the order of milliseconds and are nearly synchronized. After subtracting A(4) - A(fy), the resulting relative delay times between the fast and slow waves are less than 1 ms, corresponding to 0.4% change of anisotropy at station FP along the propagation path associated with the employed cluster. Figure 3.16 summarizes results from such analysis for 4 clusters at 5 stations that contain at least 5 high-quality splitting measurements. The changing trends of fast (Figure 3.16a) and slow (Figure 3.16b) waves are very similar, resulting in very small variations of the relative delay times. As seen in Figure 3.16(d), the results put an upper bound of 2% change of delay time of fast and slow shear waves during the 6-month period of 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the study. The results do not show systematic changes of relative delay times before the Duzce mainshock. O /' O © o o CN o o o < N o C 0 G O < D / © o o \C > \ © ,© o \ (N o O O r M O b D ^ c d < D a 9 H ro rn C N c n p ro OS (N 00 CN p ro C N ro T — i c W o ( - < ■ ) q CN °0 CN CD H « ! ( D H Figure 3.15. Fast (a) and slow (b) component seismograms for 10 events in cluster CIO and recorded at station FP. The seismograms are aligned with P wave arrivals at 1 s and are obtained by rotating the horizontal component seismograms into an average fast direction of 45.6°. The event ID is shown on right of each seismogram. The cross-correlation (CC) values relative to the first seismogram are marked on left of each trace. The CC functions between the first and subsequent fast (c) and slow (d) seismograms. The peak time lags are marked by red and blue vertical lines for the fast and slow CC functions, respectively. 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.16. Relative delay times of the fast (a) and slow (b) waves as a function of time for 4 repeating earthquake clusters that are recorded by 5 stations. The relative delay times are calculated by calculating waveform cross-correlation as shown in Figure 3.15. The vertical line mark the occurrence time of the Duzce earthquake, (c) The difference of the relative delays between the fast and slow component seismograms plotted against the earthquake occurrence time, (d) The changes of delay times between fast and slow shear waves in percent as a function of time. The percentage is calculated by dividing the relative delay times in (c) with the average values of the delay times between fast and slow shear waves. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (a) 7 6 5 4 3 7 0 JS - i r( -2 * — i O -4 { b )5 ' w O - 3 O J - ) * a n BV m COft FI • 9 * C dfc MO C O N BV ( 10 FP (. 10 MO ( 10 VO Fast 1 ‘ < u rs 0 J , 1 ) £ o > > • C 3 * # C03 BV co6 n c m m o CO K BV CIO V P C IO MO CIO vo Slow # ♦ * cm BV Fast - Stow m cmo F i # MO » » ( 05 8V { 30 FP CIO MO 1 ► ★ C30 VO *- . • - a - * 9 ^ - w • • o *c* t *§ ” # _ • - * _ # I* * * * i ■ ' + # • * < L > o .4 f 3 . o A 0 s- < U s 0 c , < u -1 o 5 -2 -3 -4 (d) * O B W C06 BV FI F a s t - S l o w P e r c e n t a g e _ ♦ COO * COS v CIO CIO CIO MO BV IT MO VO "7 t * 1 ★ ♦ ♦ ■ v ■ ’ 4* V .% * * * * # • • .. .. • • # 240 270 300 330 Julian day 360 390 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.5 DISCUSSIONS We perform a systematic analysis on spatio-temporal variations of crustal anisotropy using shear wave splitting measurements from similar earthquakes along the Karadere-Diizce branch of the NAF in the 6 months after the 1999 Izmit earthquake (Figure 3.1). We apply a waveform cross-correlation technique (Aster & Scott 1993) to identify similar earthquake clusters from -18000 events. Depending on the applied similarity criterion, approximately 4-60% of the events belong to similar earthquake clusters (Figures 3.2-3.7; Table 3.1). Splitting parameters averaged by each cluster show that crustal anisotropy can vary significantly over small change in the earthquake and station locations (Figures 3.10-3.11). This result is compatible with our previous conclusion of a shallow anisotropy in this area (Peng & Ben-Zion 2004). Up to 30% change of average delay times are observed at stations near the epicentral region of the Duzce mainshock (Figures 3.12-3.13). However, we also observe clear change in the seismicity rates and patterns following the Duzce mainshock (Figures 3.8, 3.9, and 3.12a). Thus, large spatial variations of anisotropy can be mapped into the temporal changes through the changing seismicity. Splitting parameters measured within repeating earthquake clusters further support this interpretation, and put an upper bound of -2% change in the delay time during the Duzce mainshock (Figures 3.14-3.16). We use repeating event clusters with similarity criterion (3 C = 0.95 to identify temporal change of crustal anisotropy in this area. Such choice would reject earthquake pairs that produce large temporal variations but less similar waveforms, 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. resulting a bias of possible temporal changes to lower values. However, the existence of high similar waveforms near the hypocentral region of a major earthquake suggests that the temporal variation of splitting parameters is likely to be a very subtle signal (e.g. Bokelmann & Harjes 2000; Liu et al. 2004a,b). On the other hand, splitting parameters can be significantly different for slight change o f ray paths, indicating that spatial variations are robust first-order effect. Thus, splitting analysis based on highly similar waveforms is required to separate out spatial variations from possible temporal changes. Seismologists have long focused on searching for temporal changes of effective elastic properties of the crustal rocks as systematic precursors for reliable earthquake prediction (e.g. Scholz et al. 1973). Shear wave splitting is believed to be an effective tool to observe such temporal changes, mainly due to the stress dependence of microcracks in the crustal rock. However, previous splitting studies that claimed temporal changes have been controversial and subsequently disputed (e.g. Gupta 1973; Ryall & Savage 1974; Crampin et al. 1990, 1991; Aster et al. 1990, 1991; Crampin & Gao 2004; Crampin et al. 1999, 2004; Seher & Main 2004). Mixing temporal with spatial variations of crustal anisotropy is possibly the main source of uncertainties and may result in artificial and unrealistic temporal changes. Our results of ~2% co-seismic changes of splitting delay times place an upper bound to the possible temporal changes within the 6 month period before, during and after the Duzce earthquake. The lack of the temporal variations of crustal anisotropy in our study area suggest that shear wave splitting measured from the direct S waves 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. is not sensitive to possible stress changes caused by a major earthquake. In a follow up work (Peng & Ben-Zion, work in progress, 2004), we use a sliding window cross correlation technique (Niu et al. 2003) to measure relative travel times and evolving decorrelation in the P- and S-coda waves of similar earthquake clusters. At the fault zone station VO that recorded ~0.9g peak acceleration during the Duzce mainshock, the analysis shows clear change of travel time of the coda waves following the Duzce mainshock. At other nearby stations, the temporal changes are considerably smaller. These results indicate that coda waves generated by scattering are more sensitive to in detecting change of the material properties than splitted shear waves. 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. RECAPITULATION This thesis aims at imaging the subsurface structures of earthquake fault zones using systematic analyses of large waveform data set generated around active strike-slip faults. FZ trapped waves observed in the rupture zone of the 1992 Landers, California, earthquake are analyzed in Chapter 1. Chapter 2 and 3 focus on the overall and fine-scale spatio-temporal variations of crustal anisotropy around the Karadere-Duzce branch of the NAF that ruptured in the 1999 Mw7.4 Izmit and M w7.1 Duzce, Turkey, earthquakes. The results clearly demonstrate that the shallow portion of FZ structures produce most of the observed FZ trapped waves and shear wave splitting siginals. A shear wave velocity reduction relative to the host rock o f up to 50% and attenuation coefficients as low as 10-20 can be obtained within the -100 m wide seismic waveguide. High-density microcracks tends to exist in a broader region (e.g. kilometer wide) than the -100 m wide seismic trapping structures around the active faults. Similar structures with broad damage zone are also indicated from various gravity, electromagnetic, seismic tomography and InSAR imaging studies (Fialko et al. 2002; Ben-Zion & Sammis 2003, and references therein). This broad damage zone is likely to be confined in the upper few km the crust, and may be related to the top part of a flower-type structure of the FZ that is highly fractured and aseismic. The inferred depth of -3 km may correspond to the transition from an aseismic velocity-strengthening behavior to a seismic velocity-weakening regime of rate- and state-dependent friction (e.g. Dieterich 1979, 1981; Marone & Scholz 1988). Since 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. both FZ trapped waves and shear wave splitting observations do not provide in general information on the properties of FZ structure at the depth where the bulk of the seismic energy is stored and released, additional multidisciplinary research, such as high-resolution hypocentral relocations (e.g. Schaff et al. 2002), FZ head waves (e.g. Ben-Zion and Malin 1991), and direct measurements from deep borehole and mines (e.g. Richardson & Jordan 2002) is needed to clarify the properties of the key mechanical FZ structure at seismogenic depth. The results also suggest significant variations of FZ properties along the active fault strike. Slightly different variations of ray paths may result in quite different shear wave splitting parameters, favoring a local and shallow instead of deep and pervasive crustal anisotropy interpretation. The spatial variations of crustal anisotropy can be erroneously mapped into temporal changes, mainly due to the changing seismicity. Splitting parameters measured within highly similar earthquake clusters put an upper bound o f ~2% temporal changes o f splitting delay times associated with the Duzce mainshock in our study area. The results do not show systematic precursory changes before the Diizce mainshock. The systematic procedure for analyzing FZ trapped waves developed in Chapter 1 have been applied to the large data sets collected around the Karadere- Dtizce branch of the NAF (Ben-Zion et al. 2003) and the San Jacinto FZ near Anza, California (Lewis et al. 2004), resulting in essentially the same conclusion o f a shallow FZ waveguide structure. The automatic and objective shear wave splitting method developed in Chapter 2 can be applied to any large waveform data set for the 148 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. purpose of systematic analysis of crustal anisotropy. The combination of shear wave splitting measurements with similar earthquake clusters identified from waveform cross-correlation technique (Chapter 3) provide a natural way to group splitting parameters together for fin e-scale spatial variations, and to separate out spatial variations from possible temporal changes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. REFERENCES Aki, K. & Richards, P.G., 2002. Quantitative Seismology, (second edition). University Science Books, Sausalito, CA. Akyiiz, H.S., Barka, A., Altunel, E., Hartleb, R. & Sunal, G., 2000. Field observations and slip distribution o f the November 12, 1999 Duzce earthquake (M=7.1), Bolu, Turkey, in The 1999 Izmit and Duzce Earthquakes: preliminary results, pp. 63-70, ed. Barka et al., Istanbul Technical University. Aster, R.C., Shearer, P.M. & Berger, J., 1990. Quantitative measurements of shear wave polarizations at the Anza seismic network, southern California: Implications for shear wave splitting and earthquake prediction, J. Geophys. Res., 95, 12,449-12,473. Aster, R.C., Shearer, P.M., & Berger, J., 1991. Reply, J. Geophys. Res., 96, 6415- 6419. Aster, R.C. & Shearer, P.M., 1992. Initial shear wave particle motions and stress constraints at the Anza Seismic Network, Geophys. J. Int., 108, 740-748. Aster, R.C. & Scott, J., 1993. Comprehensive characterization of waveform similarity in microearthquake data sets, Bull. Seism. Soc. Am., 83, 1307-1314. Barka, A., 1996. Slip distribution along the North Anatolian fault associated with large earthquakes of the period 1939 to 1967, Bull. Seismol. Soc. Am., 59, 521-589. Barka, A., Akyiiz, H.S., Altunel, E., Sunal, G., Qakir, Z., Dikbas, A., Yerli, B., Armijo, R., Meyer, B., Chabalier, J.B., Rockwell, T., Dolan, J.R., Hartleb, R., Dawson, T., Christofferson, S., Tucker, A., Fumal, T., Langridge, R., Stenner, H., Lettis, W., Bachhuber, J. & Page, W., 2002. The Surface Rupture and Slip Distribution of the 17 August 1999 izmit Earthquake (M 7.4), North Anatolian Fault, Bull. Seismol. Soc. Am., 92, 43-60. Ben-Zion, Y., 1998. Properties of seismic fault zone waves and their utility for imaging low velocity structures, J. Geophys. Res., 103, 12567-12585. Ben-Zion, Y. & Aki, K., 1990. Seismic radiation from an SH line source in a laterally heterogeneous planar fault zone, Bull. Seism. Soc. Am., 80, 971-994. Ben-Zion, Y. & Malin, P., 1991. San Andreas fault zone head waves near Parkfield, California, Science, 251, 1592-1594. 150 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ben-Zion, Y., Peng, Z., Okaya, D., Seeber, L., Armbruster, J.G., Ozer, N., Michael, A.J., Baris, S. & Aktar, M., 2003. A shallow fault zone structure illuminated by trapped waves in the Karadere-Duzce branch of the North Anatolian Fault, western Turkey, Geophys. J. Int., 152, 699-717. Ben-Zion, Y. & Sammis, C.G., 2003. Characterization of fault zones, Pure Appl. Geophys., 160, 677-715. Bellier, O., Over, S., Poisson, A. & Andrieux, J., 1997. Recent temporal change in the stress state and modem stress field along the North Anatolian fault zone (Turkey), Geophys. J. Int., 131, 61-86. Blanpied, M.L., Lockner, D.A. & Byerlee, J.D., 1991. Fault stability inferred from granite sliding experiments at hydrothermal conditions, Geophys. Res. Letters, 18(4), 609-612. Blenkinsop, T.G., 1990. Correlation of paleotectonic fracture and microfracture orientations in cores with seismic anisotropy at Cajon Pass drill hole, southern California, J. Geophys. Res., 95, 11,143-11,150. Bokelmann, G.FLR. & Harjes, H.P., 2000. Evidence for temporal variation of seismic velocity within the upper continental crust, J. Geophys. Res., 105, 23,879-23,894. Boness, N. & Zoback, M.D., 2004. Stress-induced seismic velocity anisotropy and physical properties in the SAFOD pilot hole in Parkfield, CA, Geophys. Res. Lett., in press. Bouin, M.P., Tellez, J. & Bernard, P., 1996. Seismic anisotropy around the Gulf of Corinth, Greece, deduced from three-component seismograms of local earthquakes and its relationship with crustal strain, J. Geophys. Res., 101, 5795-5811. Brocher, T.M., & Christensen, N.I., 1990. Seismic anisotropy due to preferred mineral orientation observed in shallow crustal rocks in southern Alaska, Geology, 18, 737-740. Cochran, E.S., Vidale, J.E. & Li, Y.-G., 2003. Near-fault anisotropy following the Hector Mine earthquake, J. Geophys. Res., 108(B9), 2436, doi:10.1029/2002JB002352. Crampin, S., 1978. Seismic wave propagation through a cracked solid: polarization as a possible dilatancy diagnostic, Geophys. J. R. Astron. Soc., 53, 467-496. 151 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Crampin, S., 1981. A review of wave motion in anisotropic and cracked elastic- media, Wave Motion, 3, 343-391. Crampin, S., 1987. Geological and industrial implications of extensive-dilatancy anisotropy, Nature, 328, 491-496. Crampin, S., 1994. The fracture criticality of cmstal rocks. Geophys. J. Int., 118, 428-438. Crampin, S., & Lovell., J.H., 1991. A decade o f shear-wave splitting in the Earth's crust: What does it mean? what use can we make of it? and what should we do next?, Geophys. J. Int., 107, 387-407. Crampin, S. & Chastin, S., 2003. A review of shear wave splitting in the crack- critical crust, Geophys. J. Int., 155, 221-240. Crampin, S. & Gao, Y., 2004. Comment on “Systematic analysis o f shear-wave splitting in the aftershock zone of the 1999 Chi-Chi, Taiwan, earthquake: shallow crustal anisotropy and lack of precursory variations” by Liu, Y., Teng, T.-L. & Ben-Zion, Y., submitted to Bull. Seismol. Soc. Am. Crampin, S., D. C. Booth, R. Evans, S. Peacock, and J. B. Fletcher, 1990. Change in shear wave splitting at Anza near the time of the North Palm Springs earthquake, J. Geophys. Res., 95, 11,197-11,212. Crampin, S., Booth, D.C., Evans, R., Peacock, S., & Fletcher, J.B., 1991. Comment on “Quantitative measurements of shear wave polarizations at the Anza seismic network, Southern California: Implications for shear wave splitting and earthquake predication” by R.C. Aster, P. M. Shearer, and J. Berger, J. Geophys. Res., 96, 6403-6414. Crampin, S., Volti, T. & Stefansson, R., 1999. A successfully stress-forecast earthquake, Geophys. J. Int., 138, F1-F5. Crampin, S., Volti, T. & Stefansson, R., 2004. Response to ‘A statistical evaluation of a ‘stress-forecast’ earthquake’ by T. Seher & LG. Main, Geophys. J. Int., doi: 10.111 l/j,1365-246X.2004.02187.x. Crase, E., Pica, A., Noble, M., McDonald, J. & Tarantola, A., 1990. Robust elastic nonlinear inversion: application to real data, Geophysics, 55, 527-538. Chester, F.M. & Chester, J.S., 1998. Ultracataclasite structure and friction processes of the Punchbowl fault, San Andreas system, California, Tectonophysics, 295, 199-221. 152 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cultrera, G., Rovelli, A., Mele, G., Azzara, R., Caserta, A. & Marra, F., 2003. Azimuth-dependent amplification of weak and strong ground motions within a fault zone (Nocera Umbra, central Italy), J. Geophys. Res., 108, 2156, doi: 10.1029/2002JB001929. Daley, T.M. & McEvilly, T.V., 1990. Shear wave anisotropy in the Parkfield Varian Well VSP, Bull. Seismol. Soc. Am., 80(4), 857-869. Davis, J. C., 1986. Statistics and Data Analysis in Geology, 646 pp., John Wiley, Hoboken, N. J.. Dieterich, J.H., 1979. Modeling of rock friction 1. Experimental results and constitutive equations, J. Geophys. Res., 84, 2161-2168. Dieterich, J.H., 1981. Constitutive properties o f faults with simulated gouge, Amer. Geophys. Union Monog., 24, 103— 120. Eberhart-Phillips, D. & Michael, A.J., 1998. Seismotectonics of the Loma Prieta, California, region determined from three-dimensional Vp, Vp/Vs, and seismicity, J. Geophys. Res., 103, 21099-21120. Evans, J.P., Shipton, Z.K., Pachell, M.A., Lim, S.J. & Robeson, K., 2000. The structure and composition of exhumed faults, and their implication for seismic processes, in Proc. o f the 3rd Confer, on Tectonic problems o f the San Andreas system, Stanford University. Faulkner, D.R., Lewis, A.C. & Rutter, E.H., 2003. On the internal structure and mechanics o f large strike-slip fault zones: field observations of the Carboneras fault in southeastern Spain, Tectonophysics, 367, 235-251. Fialko, Y., Sandwell, D., Agnew, D., Simons, M., Shearer, P. & Minster, B., 2002. Deformation on nearby faults induced by the 1999 Hector Mine earthquake, Science, 297, 1858-1862. Fohrmann, M., Jahnke, G., Igel, H. & Ben-Zion, Y., 2004. Guided waves from sources outside faults: an indication for shallow fault zone structure?, Pure Appl. Geophys., in press. Gamar, F. & Bernard, P., 1997. Shear wave anisotropy in the Erzincan basin and its relationship with crustal strain, J. Geophys. Res., 102, 20,373-20,393. Gao. Y., Wang, P., Zheng, S., Wang, M. & Chen, Y.-T., 1998. Temporal changes in shear-wave splitting at an isolated swarm of small earthquakes in 1992 near Dongfang, Hainan Island, Southern China, Geophys. J. Int., 135, 102-112. 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Gerst, A., 2003. Temporal changes in seismic anisotropy as a new eruption forecasting tool? M.S. Thesis, Victoria University of Wellington, New Zealand. Gledhill, K.R., 1991. Evidence for shallow and pervasive seismic anisotropy in the Wellington Region, New Zealand, J. Geophys. Res., 96, 21,503-21,516. Gonzalez, M. & Munguia, L., 2003. Seismic anisotropy observations in the Mexicali valley, Baja California, Mexico, Pure appl. Geophys., 160, 2257— 2278. Gupta, I.N., 1973. Premonitory variations in S-wave velocity anisotropy before earthquakes in Nevada, Science, 182, 1129-1132. Haberland C., Agnon, N.A., El-Kelani, R., Maercklin, N., Qabbani, I., Rumpker, G., Ryberg, T., Scherbaum, F. & Weber, M., 2003. Modeling of seismic guided waves at the Dead Sea Transform, J. Geophys. Res., 108(B7), 2342, doi: 10.1029/2002JB002309. Hartleb, R.D., Dolan, J.F., Akyiiz, H.S., Dawson, T.E., Tucker, A.Z., Yerli, B., Rockwell, T.K., Toraman, E., £akir, Z., Dikba§, A. & Altunel, E., 2002. Surface Rupture and Slip Distribution along the Karadere Segment of the 17 August 1999 Izmit and the Western Section of the 12 November 1999 Duzce, Turkey, Earthquakes, Bull. Seismol. Soc. Am., 92(1), 67-78. Hauksson, E., Jones, L.M., Hutton, K. & Eberhart-Phillips, D., 1993. The Landers earthquake sequence: Seismological observations, J. Geophys. Res., 98, 19835-19853. Hudson, J. A., 1981. Wave speeds and attenuation of elastic waves in material containing cracks, Geophys. J. R. astr. Soc., 64, 133-150.Igel, H., Ben-Zion, Y. & Leary, P., 1997. Simulation of SH and P-SV wave propagation in fault zones, Geophys. J. Int., 128, 533-546. Igarashi, T., Matsuzawa, T. & Hasegawa, A., 2003. Repeating earthquakes and interplate aseismic slip in the northeastern Japan subduction zone, J. Geophys. Res., 108(B5), 2249, doi:10.1029/2002JB001920. Igel, H., Jahnke, G. & Ben-Zion, Y., 2002. Numerical simulation o f fault zone guided waves: accuracy and 3-D effects, Pure Appl. Geophys., 159, 2067- 2083. Jahnke, G., Igel, H. & Ben-Zion, Y., 2002. Three-dimensional calculations of fault zone guided wave in various irregular structures, Geophys. J. Int., 151, 416- 426. 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Johnson, A.M., Fleming, R.W. & Cruikshank, K.M., 1994. Shear zones formed along long, straight traces of fault zones during the 28 June 1992 Landers, California, earthquake, Bull. Seism. Soc. Am., 84, 499-510. Johnson, A.M., Fleming, R.W., Cruikshank, K.M., Martosudarmo, S.Y., Johnson, N.A. & Johnson, K.M., 1997. Analecta of structures formed during the 28 June 1992 Landers-Big Bear, California earthquake sequence, Technical report, U.S. Geol. Surv. Open File Rep. 97-94. Kaneshima, S., 1990. Origin of crustal anisotropy: Shear wave splitting studies in Japan, J. Geophys. Res., 95, 11,121-11,133. Kern, H. & Wenk, H., 1990. Fabric-related velocity anisotropy and shear wave splitting in rocks from the Santa Rosa mylonite zone, California. J. Geophys. Res, 95, 11,213-11,223. Klein, W., 1978. Hypocenter location program HYPOINVERSE, USGS Open File Report 78-694. Korneev, V.A., Nadeau, R.M. & McEvilly, T.V., 2003. Seismological studies at Parkfield IX: Fault-zone imaging using guided wave attenuation, Bull. Seism. Soc. Am., 93, 1415-1426. Langridge, R.M., Stenner, H.D., Fumal, T.E., Christofferson, S.A., Rockwell, T.K., Hartleb, R.D., Bachhuber, J. & Barka, A.A., 2002. Geometry, Slip Distribution, and Kinematics of Surface Rupture on the Sakarya Fault Segment during the 17 August 1999 Izmit, Turkey, Earthquake, Bull. Seismol. Soc. Am., 92(1), 107-125. Leary, P.C., Li Y.G. & Aki, K., 1987. Observations and modeling of fault zone fracture anisotropy, I, P, SV, SH travel times, Geophys. J. R. Astron. Soc., 91, 461-484. Leary, P.C., Crampin, S. & McEvilly, T.V., 1990. Seismic fracture anisotropy in the Earth’s crust: An overview, J. Geophys. Res., 95(7), 11,105-11,114. Lee, W.H.K., 1999. Digital waveform data o f 238 selected Landers aftershocks from a dense PC-based seismic array, unpublished report, U.S. Geol. Surv., Menlo Park, California. Lees, J. M., 1998, Multiplet analysis at Coso Geothermal, Bull. Seismol. Soc. Am., 88(5), 1127-1143. Lewis, M.A, Peng, Z., Ben-Zion, Y. & Vernon, F.L., 2003. Shallow seismic trapping 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. structure in the San Jacinto fault zone near Anza, California, Seism. Res. Lett. 74, 247. Lewis, M.A, Peng, Z., Ben-Zion, Y. & Vernon, F.L., 2004. Shallow seismic trapping structure in the San Jacinto fault zone near Anza, California, ms. in prep. Liu, Y., Crampin, S. & Main, I., 1997. Shear-wave anisotropy: spatial and temporal variations in time delays at Parkfield, Central California, Geophys. J. Int., 130, 771-785. Liu, Y., Teng, T.-L. & Ben-Zion, Y., 2004a Systematic analysis of shear-wave splitting in the aftershock zone of the 1999 Chi-Chi earthquake: shallow crustal anisotropy and lack of precursory variations, submitted to Bull. Seismol. Soc. Am. Liu, Y., Teng, T.-L. & Ben-Zion, Y., 2004b Reply to Comment of Crampin and Gao on “Systematic analysis of shear-wave splitting in the aftershock zone of the 1999 Chi-Chi, Taiwan, earthquake: shallow crustal anisotropy and lack of precursory variations”, submitted to Bull. Seismol. Soc. Am. Li, Y.G. & Leary, P., 1990. Fault zone seismic trapped waves, Bull. Seism. Soc. Amer., 80, 1245-1271. Li, Y.G., Aki, K., Adams, D., Hasemi, A. & Lee, W.FLK., 1994a. Seismic guided waves trapped in the fault zone of the Landers, California, earthquake of 1992, J. Geophys. Res., 99, 11705-11722. Li, Y.-G., Teng, T.-L. & Henyey, T.L., 1994b. Shear-wave observations in the Northern Los Angeles Basin, Southern California, Bull. Seism. Soc. Am., 84, 307-323. Li, Y.G., Aki, K., Vidale, J.E. & Alvarez, M.G., 1998. A delineation of the Nojima fault ruptured in the M7.2 Kobe, Japan, earthquake of 1995 using fault-zone trapped waves, J. Geophys. Res., 103, 7247-7263. Li, Y.G., Vidale, J.E., Aki, K. & Xu, F., 2000. Depth-dependent structure of the Landers fault zone using fault zone trapped waves generated by aftershocks, J. Geophys. Res., 105, 6237-6254. Li, Y.G. & Vernon, F.L., 2001. Characterization of the San Jacinto fault zone near Anza, California, by fault zone trapped waves, J. Geophys. Res, 106, 30671- 30688. Li, Y.G., Vidale, J.E., Day, S.M., Oglesby, D.D. & the SCEC Field Working Team, 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2002. Study of the 1999 M 7.1 Hector Mine, California, earthquake fault plane by trapped waves, Bull. Seism. Soc. Am., 92, 1318-1332. Lyakhovsky, V., Ben-Zion, Y. & Agnon, A., 1997. Distributed damage, faulting, and friction, J. Geophys. Res., 102, 27635-27649. Mardia, K.V. & Jupp, P.E., 2000. Directional Statistics, 429 pp., John Wiley, Hoboken, N.J. McClusky, S. et al., 2000. GPS constraints on plate kinematics and dynamics in the Eastern Mediterranean and Caucasus, J. geophys. Res., 105, 5695-5719. Marone, C. & Scholz, C.H., 1988. The depth of seismic faulting and the upper transition from stable to unstable regimes, Geophys. Res. Lett., 15, 621-624. Matcham, I., Savage, M.K. & Gledhill, K.R., 2000. Distribution of seismic anisotropy in the subduction zone beneath the Wellington region, New Zealand, Geophys. J. Int., 140, 1-10. Michael, A.J. & Ben-Zion, Y., 1998. Challenges in inverting fault zone trapped waves to determine structural properties, EOS Trans. Amer. Geophys. Union, 79, S231. Miller, V. & Savage, M.K., 2001. Changes in seismic anisotropy after volcanic eruptions: evidence from Mount Ruapehu, Science, 293, 2231-2233. Mooney, W.D. & Ginzburg, A., 1986. Seismic measurements of the internal properties of fault zones. Pure Appl. Geophys., 124, 141-157. Munson, C.G., Thurber, C.H., Li, Y. & Okubo, P.G., 1995. Crustal shear wave anisotropy in southern Hawaii: Spatial and temporal analysis, J. Geophys. Res., 100(B10), 20,367-20,377. Nadeau, R.M., Antolik, M., Johnson, P.A., Foxall, W. & McEvilly, T.V., 1994. Seismological studies at Parkfield III: microearthquake clusters in the study of fault-zone dynamics, Bull. Seism. Soc. Am., 84, 247-263. Nadeau, R.M., Foxall, W. & McEvilly, T.V., 1995. Clustering and periodic recurrence of microearthquakes on the San Andreas Fault at Parkfield, California, Science, 267, 503-507. Niu, F., Silver, P.G., Nadeau, R.M., McEvilly, T.V., 2003. Stress-induced migration of seismic scatterers associated with the 1993 Parkfield aseismic transient Eevent, Nature, 426, 544-548. 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Nur, A. & Simmons, G., 1969. Stress-induced velocity anisotropy in rock: An experimental study, J. Geophys. Res., 74(27), 6667-6674. Nur, A., 1971. Effects of stress on velocity anisotropy in rocks with cracks, J. Geophys. Res., 76, 2022-2034. Nuttli, O., 1961. The effect of earth’s surface on the S-wave particle motion, Bull. Seismol. Soc. Am., 51, 237-246. O ’Connell, R.J. & Budiansky, B., 1974. Seismic velocities in dry and saturated cracked solids, J. Geophys. Res., 79(35), 5412-5426. Peacock, S., Crampin, S. & Booth, D.C., 1988. Shear wave splitting in the Anza seismic gap, southern California: temporal variations as possible precursors, J. Geophys. Res., 93, 3339-3356. Peng, Z., Ben-Zion, Y., Michael, A.J. & Zhu, L., 2003. Quantitative analysis of fault zone waves in the rupture zone of the Landers, 1992, California earthquake: Evidence for a shallow trapping structure, Geophys. J. Int., 155, 1021-1041. Peng, Z. & Y. Ben-Zion, 2004a. Systematic analysis of crustal anisotropy along the Karadere-Diizce branch of the north Anatolian fault, Geophys. J. Int., in press. Peng, Z. & Y. Ben-Zion, 2004b. Spatio-temporal variations of crustal anisotropy from similar events in aftershocks of the 1999 M7.4 Izmit and M7.1 Duzce, Turkey, earthquake sequences, ms. in prep. Poupinet, G., Ellsworth, W.L. & Frechet, J., 1984. Monitoring velocity variations in the crust using earthquake doublets: An application to the Calaveras Fault, California, J. Geophys. Res., 89, 5719-5731. Rabbel, W., 1994. Seismic anisotropy at the Continental Deep Drilling Site (Germany). Tectonopysics, 232, 329-341. Press, W., Flannery, B., Teukolsky, S. & Vetterling, W., 1986. Numerical Recipes, Cambridge University Press, Camridge, U.K. Reilinger, R.E., McClusky, S.C., Oral, M.B., King, R.W., Toksoz, M.N., Barka, A.A., Kinik, I , Lenk, O. & Sanli, I , 1997. Global Positioning System measurements of the present day crustal movements in the Arabia-Africa- Eurasia plate collision zone, J. Geophys. Res., 102, 9983-9999. Richards-Dinger, K.B. & Shearer, P.M., 2000. Earthquake locations in southern California obtained using source specific station terms, J. Geophys. Res., 105, 10939-10960. 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Richardson, E. & Jordan, T.H., 2002. Seismicity in Deep Gold Mines of South Africa: Implications for Tectonic Earthquakes, Bull. Seism. Soc. Am., 92, 1200-1236. Rockwell, T.K., Lindvall, S., Herzberg,M., Murbach, D., Dawson, T. & Berger, G., 2000. Paleoseismology of the Johnson Valley, Kickapoo, and Homestead Valley faults: clustering of earthquakes in the eastern California shear zone, Bull. Seism. Soc. Am., 90, 1766-1782. Rovelli, A., Caserta, A., Marra, F. & Ruggiero, V., 2002. Can seismic waves be trapped inside an inactive fault zone? The case study of Nocera Umbra, central Italy, Bull. Seism. Soc. Am., 92, 2217-2232. Rubin, A.M., Gillard, D. & Got, J.-L., 1999. Streaks of microearthquakes along creeping faults, Nature, 400, 635-641. Ryall, A. & Savage, W., 1974. S wave splitting: Key to earthquake prediction? Bull. Seism. Soc. Am., 64, 1943-1951. Saiga, A., Hiramatsu, Y., Ooida, T. & Yamaoka, K., 2003. Spatial variation in the crustal anisotropy and its temporal variation associated with a moderate-sized earthquake in the Tokai region, central Japan, Geophys. J. Int., 154, 695-705. Savage, M.K., 1999. Seismic anisotropy and mantle deformation: what have we learned from shear wave splitting, Rev. Geophys., 37, 65-106. Savage, M.K., Shih, X.R., Meyer, R.P. & Aster, R.C., 1989. Shear-wave anisotropy of active tectonic regions via automated S-wave polarization analysis, Tectonophysics, 165, 279— 292. Savage, M.K., Peppin, W.A. & Vetter, U.R., 1990. Shear wave anisotropy and stress direction in and near Long Valley caldera, California, 1979-1988, J. Geophys. Res., 95, 11,165-11,117. Schaff, D.P., Beroza, G.C & Shaw, B.E., 1998. Postseismic response of repeating aftershocks, Geophys. Res. Lett., 25(24), 4549-4552. Schaff, D.P., Bokelmann, G.H.R., Beroza, G.C., Waldhauser, F. & Ellsworth, W.L., 2002. High-resolution image of Calaveras Fault seismicity, J. Geophys. Res., 107(B9), 2186, doi: 10.1029/2001JB000633. Scholz, C.H., Sykes, L.R. & Aggarwal, Y.P., 1973. Earthquake prediction: A physical basis, Science, 181, 803-810. 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Scholz, C.H., The Mechanics o f Earthquakes and Faulting, 2002. Cambridge Univ. Press, New York. Seeber, L., Armbruster, J.G., Ozer, N., Aktar, M., Baris, S., Okaya, D., Ben-Zion, Y. & Field, E., 2000. The 1999 Earthquake Sequence along the North Anatolia Transform at the juncture between the two main ruptures, in The 1999 Izmit and Duzce Earthquakes: Preliminary Results, pp. 209-223, ed. Barka et al., Istanbul Technical University. Shearer, P.M., Hardebeck, J.L., Astiz, L. & Richards-Dinger, K.B., 2003. Analysis of similar event clusters in aftershocks of the 1994 Northridge, California, earthquake, J. Geophys. Res., 108(B1), 2035, doi:10.1029/2001JB000685. Shih, X.R. & Meyer, R.P., 1990. Observation of shear wave splitting from nature events: South Moat of Long Valley caldera, California, June 29 to August 12, 1982, J. Geophys. Res., 95, 11,179-11,196. Sibson, R.H., 2002. Geology of the crustal earthquake source, in International Handbook o f Earthquake and Engineering Seismology, Part A, eds. W.H.K Lee, H. Kanamori, P.C. Jennings, and C. Kisslinger, 455-473, Academic Press, San Diego, CA, ISBN: 0-12-440652-1. Sieh, K., Jones, L., Hauksson, E., Hudnut, K., Eberhart-Philips, D., Heaton, T., Hough, S., Hutton, K., Kanamori, H., Lilje, A., Lindvall, S., McGill, S.F., Mori, J., Rubin, C., Spotila, J.A., Stock, J., Thio, H.K., Treiman, J., Wernicke, B. & Zachariasen, J., 1993. Near-field investigations o f the Landers earthquake sequence, April to July 1992, Science, 260, 171-176. Silver, P.G. & Chan, W.W., 1991. Shear-wave splitting and subcontinental mantle deformation, J. Geophys. Res., 96, 16,429-16,454. Spudich, P. & Olsen, K.B., 2001. Fault zone amplified waves as a possible seismic hazard along the Calaveras fault in central California, Geophys. Res. Lett., 28, 2533-2536. Tadokoro, K., Ando, M. & Umeda, Y., 1999. S' wave splitting in the aftershock region o f the 1995 Hyogo-ken Nanbu earthquake, J. Geophys. Res., 104, 981-991. Tadokoro, K. & M. Ando, 2002. Evidence for rapid fault healing derived from temporal changes in S wave splitting, Geophys. Res. Lett., 29, 6-1. Tadokoro, K., Ando, M., Baris, S., Nishigami, K., Nakamura, M., Ucer, S.B., Ito, A., Honkura, Y. & Isikara, M., 2002. Monitoring fault healing after the 1999 Kocaeli, Turkey, earthquake, J. o f Seismol., 6, 411-417. 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Waldhauser, F., 2001. HypoDD-A Program to Compute Double-Difference Hypocenter Location, U.S. Geol. Surv. Open File Rep., 01-113. Utkucu, M., Nalbant, S.S., McCloskey, J., Steacy, S. & Alptekin, O., 2003. Slip distribution and stress changes associated with the 1999 November 12, Duzce (Turkey) earthquake (Mw = 7.1), Geophys. J. Int., 153, 229-241. Vidale, J.E., Helmberger, D.V. & Clayton, R.W., 1985. Finite-difference seismograms for SH waves, Bull. Seism. Soc. Am., 75, 1765-1782. Vidale, J.E., Ellsworth, W.L., Cole, A. & Marone, C. 1994. Variations in rupture process with recurrence interval in a repeated small earthquake, Nature, 368, 624-626. Wessel, P. & Smith, W. H. F., 1995. New version of the Generic Mapping Tools Released, EOS Trans. Amer. Geophys. Union, 76, 329. Yilmaz, Y., Tuysuz, O., Yigitbas, E., Gene, S.C. & Sengor, A.M.C., 1997. Geology and Tectonic Evolution of the Pontides, in Regional and Petroleum geology o f the Black Sea and Surrounding Region: AAPG Memoir 68, 183-226, ed. A.G.Robinson. Zhang, Z. & Schwartz, S.Y., 1994. Seismic anisotropy in the shallow crust of the Loma Prieta segment of the San Andreas fault system, J. Geophys. Res., 99(B5), 9651-9661. Zinke, J. & Zoback, M.D., 2000. Structure-related and stress-induced shear-wave velocity anisotropy: observations from microearthquakes near the Calaveras Fault in central California, Bull. Seismol. Soc. Am., 90, 1305-1312. 161 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. APPENDIX A: Input Parameters for the Genetic Inversion Algorithm Description: the Genetic Inversion Algorithm (GIA) for the fault zone trapped wave modeling (gatrapped) is developed by Dr. Andrew Michael (Michael@usgs.gov). The program employs the 2D analytical solution of (Ben-Zion and Aki 1990) as a forward kernel (Michael & Ben-Zion 1998) and iteratively test different model parameter spaces using the Genetic Inversion Algorithm. Directory: the latest version of the GIA (gatrapped_parallel) with parallel mode can be found in the directory /home/terra- 19/wrkl9/zpeng/src/PMGIA/trapped_dd_restart at host terra.usc.edu. Syntax: gatrapped__parallel ncpus crustfile quakefile stationfile gramfilel (gramfile2-n] Input parameters: ncpus: number of CPUs needed to run the gatrapped_parallel in parallel mode. Set it to 1 if you only have or need one CPU. crustfile: filename of the crustal model plus some genetic algorithm parameters quakefile: filename of a list of earthquakes and parameters relating to their geometries stationfile: filename of a list of seismometers gramfile[l-n]: the seismograms plus header information specifying the earthquake and station that the seismograms belong to. Example input parameters and detailed explanations (the electronic version of these input parameters can be found at the following directory: /home/terra-19/wrkl9/zpeng/Landers/GAInversions/South/Run12 ) Filename for crustalfile: landers.crust ngen 50 nmodels 2 00 mutarion 0.01 minsep 0.01 restart 0 myseed 0 transition 0 vl 0 2.5 4 0.05 vf 0 0.4 0.75 0.05 v4 1 1 ql 1 1000 qf 0 1 80 1 q4 1 1000 density 1 0.025 widthf 0 0.15 0.25 0.005 edge 0 -1 1 0.05 Detailed description: ngen 50: number of generations used in the GIA. nmodels 200: number of models in each generation. The value of ngen x nmodels gives the total number of the models tested by the GIA. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mutation 0.01: the mutation rate (the value is most likely between 0.01 and 0.1) Minsep 0.01: the minimum separation in x-coordinate between source and receiver (this parameter is used to avoid very long calculation time when the source and receiver have very similar x-coordinate) restart 0: a flag to indicate whether the inversion is new (0) or continued from a previous run (1). The filename that contains all the parameters from the last generation is needed if the restart is 1 (e.g., restart 1 lastgen.dat) myseed 0: a flag to indicate whether the seed for the inversion is generated randomly (0) or specified (1). Typically, myseed is set to be 1 for debugging purpose or continuing from previous run (with restart flag to be 1). If myseed is 1, a seed value is need (e.g., myseed 1 0.5) transition 0: a flag to indicate whether a transition zone is existed, with 0 standing for no, and 1 for yes. If 1, then add corresponding lines (e.g., velocity, width and attenuation) for transition zone. vl 0 2.5 4 0.05: shear wave velocity in first quarter space, given as a variable (0), with a range 2.5-4 km/s, and a increase of 0.05 km/s per step. If the second value is 1, vl is fixed as constants. The same rule holds for the following parameters (vfr, v4r, ql, qf, q4, widthf, edge). Note that the program may decrease the requested step size so that the number of possibilities are exactly 2**n where n is an integer. This is part of the genetic algorithm package, vfr 0 0.5 0.75 0.05: shear wave velocity in fault zone given as a ratio of the first quarter space shear velocity (vl). In this specific case, the range of the fault zone shear wave velocity is from 1 to 3 km/s. v4r 1 1: shear wave velocity in second quarter space given as a ratio of vl. If the second value is 1, v4r is fixed as constants. If the second value is 0, v4r is set to be variable (see vfr for example). ql 1 1000: shear wave attenuation coefficient in the first quarter space qf 0 1 80 1 : shear wave attenuation coefficient in the fault zone layer q4 1 1000: shear wave attenuation coefficient in the second quarter space density 1 0.025: density of the material given as a constant widthf 0 0.15 0.25 0.005: width of the fault zone in km edge 0 -1 1 0.05: location of left (minimum in x) edge of fault zone Filename for quakefile: landers.eqs (has 5 lines per earthquake, repeat for multiple events) 10161206 x 0 0 0.5 0.05 y 1 0 z 0 1 6 0.1 ot 0 0 4 0.1 Detailed description: 10161206: an integer identifier for this earthquake 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. x 0 0 0.5 0.05: across-fault (x-coordinate) location of this event If the second parameter is 1, the x is given as a constant (the third parameter). If the second parameter is 0, the x is a variabl with a range (the third and fourth parameters) and a change step (the fifth parameter). y 0 10 12 0.1: along-fault (y-coordinate) location of this event, given as a variable. z 0 5 6 0.1: depth (z-coordinate) location of this event, given as variable. ot 0 0 4 0.1: origin time for this event, given as a variable. Filename for stnfile: stn.dat Wll -0.5480 0.0014 0 0 W10 -0.4483 0.0014 0 0 Detailed description: one line per station, repeat for multiple stations with the following format: station name, and location in x,y, and z (from ground surface) in km. Filename for gramfile: COO.asc eq 10161206 station COO windows 0 6 maxfreq 15 accuracy 0.00001 gibbs 5 0.010 0.03006 0.020 0.08347 Detailed description: one file per seismogram, use multiple files for multiple seismograms. eq 10161206: earthquake identifier for this seismogram station COO: station name for this seismogram windows 0 6: time window used in cross correlations maxfreq 15: maximum frequency to use in computing synthetic accuracy 0.00001: accuracy to use in computing synthetic gibbs 5: gibbs frequency to use in computing synthetic 0.010 0.03006: seismogram given as one sample per line with each line having time in seconds and displacement Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. APPENDIX B: README File for the Sliding Window Shear Wave Splitting Program Description: The Silver & Chan (1991) program was modified to measure the shear wave splitting of local earthquakes. A sliding window technique is added to automatically find the best time window for the splitting measurement. In addition, the program automatically assigns a 2-level quality to each splitting parameters for quality control. See Peng & Ben-Zion (2004) for a complete description of the analysis procedure. This file serves as a README for how to prepare their data and how use this modified version of the Silver and Chan (1991) program. The source code can be found in the following directory at terra.usc.edu: /home/terra-19/wrkl9/zpeng/src/split2 Prerequisite: the following programs is needed in the system: 1. Genetic Mapping Tools (GMT) version 3.4 or above. Web: http://gmt.soest.hawaii.edu/ 2. SAC2 0 00 Web: http://www.llnl.gov/sac/ 3. Latex Web: http://www.latex-project.orq/ 4. Modified version of shear__grh (the Silver and Chan 1991 program) 5. Some SAC related programs (e.g., saclst, sac_me). The source code is available at /home/terra-19/wrkl9/zpeng/src/Sac_Misc 6. The following scripts and files: a. split_win_lpl5.mac: SAC MACRO file that calls shear_grh program, and generates figures in SAC2000. b. sws_STA_win.tex: Latex files that merge the plots generated by split_win_lpl5.mac into one pdf file c. geometry.sty: Latex style file used by sws_STA_win.tex d. run_sws___all_win_lpl5. csh: C-shell scripts that calculate the splitting parameters, with the station name and time window as its two input parameters. e. get_best_sws_eval_quality_plot_lpl5 . csh: C-shell scripts that automatically estimate the best time window, assign qualities, and calculate the splitting parameters, with the station name as its input parameter. Data organization: 1. Waveforms need to be stored in SAC format, with the following headers filled in: stla stlo stel evla evlo evdp o a tO 2. Seismograms that are generated by the same event need to be stored in the same event_based directory. The file name of the three-component seismograms follows the convention: stn.[zne], where z, n, and e stand for the vertical, north, and east component, respectively. If one prefers not to use the above conventions, the parameter entries of "comps" for the "split_win__lpl5 .mac" SAC MACRO file in "get__best_sws_eval_quality_plot_lpl5.csh" and "run_sws_all_win_lpl5.csh" need to be modified accordingly. 165 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Analysis step, following Peng & Ben-Zion (2004) 1. Select records that are within the shear wave window (e.g., the epicentral distances (header dist) is no more than the event depth (header evdp) and have corresponding headers (e.g., origin time, P and S wave arrivals) filled in. 2. Run . ./get___best_sws_eval_quality__plot_lpl5 . csh sta_name for each records that satisfy 1. 3. Assemble together the output of splitting parameters in each event_based directory (with the filename stn_name.smea.best.Ipl5) and the quality assignment file (with the filename stn_get_best_sws_eval_quality_plot_only. tmp) 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Spatial and temporal variation of coda Q[-1] in California

PDF

Integrated geochemical and hydrodynamic modeling of San Diego Bay, California

PDF

Using small events to measure the evolution of the strain and stress fields before the 1992 Landers, California earthquake

PDF

Numerical modeling of long period events at the Piton de la Fournaise by the use of object-oriented scientific computing

PDF

Simulating nonplanar fault surfaces using stochastic branching

PDF

A unified methodology for seismic waveform analysis and inversion

PDF

The structure and development of Middle and Late Triassic benthic assemblages

PDF

Modeling and imaging asperities on a fault plane and characterizing spatial and temporal patterns of precursory seismicity

PDF

A surface wave study of crustal and upper mantle structures of Eurasia

PDF

The unusual sedimentary rock record of the Early Triassic: Anachronistic facies in the western United States and southern Turkey

PDF

Radium isotopes in San Pedro Bay, California: Constraint on inputs and use of nearshore distribution to compute horizontal eddy diffusion rates

PDF

Paleoecology and depositional paleoenvironments of Pleistocene nearshore deposits, Las Animas, Baja California Sur, Mexico

PDF

Systematic analysis of crustal anisotropy and attenuation using seismic data associated with the 1999 Chi-Chi, Taiwan, earthquake

PDF

Physics of weakly-coupled plasmas inside the sun

PDF

Management of large earthquake data using the Antelope Relational Database and seismicity analysis of the 1999 Turkey earthquake sequences

PDF

Marine paleoecology during the aftermath of the end-Permian mass extinction

PDF

Fractionation of nitrogen isotopes during early diagenesis

PDF

Cyclostratigraphy and chronology of the Albian stage (Piobbico core, Italy)

PDF

Modeling of continuous tiltmeter data from the 1984 rifting event at Krafla Volcano, Iceland

PDF

Data -driven facial animation synthesis by learning from facial motion capture data

Asset Metadata

Creator Peng, Zhigang (author)

Core Title Systematic high -resolution imaging of fault zone structures

Degree Doctor of Philosophy

Degree Program Geological Sciences

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag geophysics,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-559847

Unique identifier UC11340429

Identifier 3145264.pdf (filename),usctheses-c16-559847 (legacy record id)

Legacy Identifier 3145264.pdf

Dmrecord 559847

Document Type Dissertation

Rights Peng, Zhigang

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA