Constraining Lithospheric Deformation Mechanisms. Using Teleseismic Conversions. Charles K. Wilson. A thesis submitted to the

Transcription

1 Constraining Lithospheric Deformation Mechanisms Using Teleseismic Conversions by Charles K. Wilson B.S., University of Arizona, l998 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirement for the degree of Doctor of Philosophy Department of Geological Sciences 2000

2 This thesis entitled: Constraining Lithospheric Deformation Mechanisms Using Teleseismic Conversions written by Charles K. Wilson has been approved by the Department of Geological Sciences Craig H. Jones Anne Sheehan Date The final copy of this thesis has been examined by the signatories, and we find that both the content and the form meet acceptable presentation standards of scholarly work in the above mentioned discipline. HRC protocol #

3 Acknowledgements I would like to thank my adviser Craig Jones for giving me a chance despite my less that stellar academic history and for helping me to learn to think like a scientist. Thanks also to Anne Sheehan who looked out for me when my committee was rough on me. And for being rough on me, I would like to thank Peter Molnar because without rough times now I would probably have many more tough times ahead. I would finally like to thanks my family. Without them I would not have had the resources and support to accomplish what I have.

4 1-2 Abstract I seek to address outstanding questions about continental tectonics such as lithospheric strength and regional-scale strain accommodation mechanisms in areas undergoing extension or transpression. Variations in seismic properties correlate with changes in rock type or rock properties. Mapping the variations through the lithosphere allows us to understand the structure of the lithosphere and develop conceptual models of lithospheric deformation processes. This body of work describes the use of teleseismic P to S converted waves for constraining local and regional scale lithospheric structure. I infer large-scale structure by depth migrating the converted energy and stacking to create a lithospheric volume of converted wave amplitude. I use back-azimuthal patterns and timing and amplitude of particular conversions from a single station or a group of nearby stations to determine the character of the conversion interface (e.g. dipping or anisotropic interface). I show application of this technique to two different areas: one undergoing extension in the presence of volcanism and the other transpression. Results from a recent passive seismic study of the Coso geothermal area near Ridgecrest, California suggest heat for the geothermal system comes from a single shallow magma reservoir (~5 km below sea level) that also plays a crucial role in the local change in deformation style from areas to the north and west. Converted phases from teleseisms recorded by a seismic array spanning the Marlborough strike-slip fault system, South Island, New Zealand show a continuous, unbroken Moho beneath the two northernmost faults of the fault system, which suggests that distributed, ductile deformation, not slip on a narrow vertical fault, accommodates most of the lower crustal strain.

5 1-3 Table of Contents 1. Using teleseismic scattered waves to determine crust and upper mantle structure Scientific Motivation Introduction Convolutional model and receiver function calculation Effects of anisotropy and dipping interfaces Use of arrays in receiver function analysis Common conversion point imaging in practice Pre-stack versus Post-stack depth migration Effect of receiver function quality on the final image A Single-chamber Silicic Magma System Inferred From Shear-wave Discontinuities of the Crust and Uppermost Mantle, Coso Geothermal Area, California Introduction Geologic and Tectonic Setting Data and Method Array Design and Placement Event Selection Data preparation Receiver Function Calculation and Compositing Common Conversion Point Stacking Seismological Observations Upper Crustal Negative Location and Character Complications of the Moho and Mid Crustal Positive Arrivals Post-UCN Arrivals Discussion Geometry and Inference of an Upper Crustal Magma Chamber Absence of Lower Crustal Magma Bodies Melt in the Upper Crustal Magma Chamber

6 Relationship to Regional Tectonics Conclusions Accommodating Crustal Strain within a Volcanic Area: Seismological Observations and Numerical Simulations from Coso Geothermal Area, Ridgecrest, California Introduction Seismological Observations Table 3.1 Best synthetic model for the JSH array Table 3.2 Best synthetic model for the BPT array Numerical Simulation Initial and Boundary Conditions Results Conclusions Evidence for Distributed Lower Crustal Deformation Within a Continental Strike-Slip Fault Zone: Marlborough Fault System, South Island, New Zealand Introduction Geologic and Tectonic Setting Table 4.1 Table summarizing total displacement [Little and Jones, 1998] and current slip rates [Bourne et al., 1998] on major faults of the Marlborough Fault System Data and Method Array Design and Placement Event Selection and Data Preparation Receiver Function Calculation and Compositing Common Conversion Point Stacking Table 4.2 Table summarizing wavespeed models used for receiver function depth migration Seismological Observations Mean receiver functions and back-azimuthal stacks Northwest of the Wairau Fault

7 1-5 Table 4.3 Parameters used in synthetic receiver function modeling for rays piercing the base of the crust north of the surface trace of the Wairau fault Between the Wairau and Awatere faults Table 4.4 Parameters used in synthetic receiver function modeling for rays piercing the base of the crust between the Wairau and Awatere faults Southeast of the Awatere fault Common conversion point stacks Discussion Moho topography and resolution of teleseismic imaging Mid crustal conversions Implications for strain distribution mechanisms and the rheology of the lower crust Reconciliation of CCP image with gravity observations Conclusions References Cited

8 Using teleseismic scattered waves to determine crust and upper mantle structure 1.1. Scientific Motivation Uncertainties remain about the response of continental lithosphere undergoing extension or transpression primarily because we have poor constraints on both the strength of and deformations mechanisms within the lower crust and mantle lithosphere. Difficulties arise because we are only able to observe lithospheric processes from the surface, forcing the use of less direct methods of observation. Observational seismology provides a tool to infer the character of the lithosphere indirectly. Variations in seismic properties correlate with changes in rock type or rock properties. Mapping the variations through the lithosphere allows us to understand the structure of the lithosphere and develop conceptual models of lithospheric deformations processes. This body of work will describe the use of teleseismic P-to-S converted waves for constraining local and regional scale lithospheric structure. I infer large-scale structure by observing regional variations in converted interfaces. I use back-azimuthal patterns and timing and amplitude of particular conversions from a single station or a group of nearby stations to determine the character of observed conversion interfaces (e.g. dipping or anisotropic interface). The results from application of the technique to the Coso geothermal area near Ridgecrest, California (Chapter 2 and 3) and the Marlborough Fault Zone, New Zealand (chapter 4) are shown here.

9 Introduction A radially polarized seismic wave impinging on an interface, produces transmitted, reflected, and converted phases that propagate as both P and S waves. The amplitude of each of the resulting phases is a function of the incidence angle of the impinging wave and the impedance contrast across the interface. If the scattered phases can be identified in the seismic wave train then the locus of phase generation point or seismic interface can be estimated with the knowledge of the local seismic wavespeed structure. For over 30 years, P-to-S conversions from teleseismic earthquakes Figure 1.1 Ray paths of forward and backscattered phases generated by an impinging teleseismic P wave front after Clouser and Langston [1995]. Solid lines mark P ray paths with S rays marked by dashed lines. Uppercase letters denote downgoing phases with lowercase letters symbolizing upgoing phases. The free surface multiples returning to the surface as S (e.g. PpSs) will be contained in the receiver function. The idealized vertical and radial seismograms indicate the approximate arrival and amplitude relationships for some of the arrivals in the ray diagram above. ( 30 degrees epicentral distance) have been used to determine the crust and mantle structure beneath three-component seismic stations in a technique referred to as receiver function analysis [Burdick and Langston, 1977; Langston, 1977; 1979; Phinney, 1964]. An impinging P wave from a distant earthquake generates a converted S wave at lithospheric boundaries such as the Moho (figure 1.1). The incoming P wave continues to the surface followed by the converted shear wave with the delay time between the direct P arrival and the converted S wave related to the depth to the converter and the

10 1-8 Vp/Vs ratio. The near vertical incidence of the teleseismic P wave allows the natural separation of the impinging P wave and the converted S wave because the orthogonal particle motions insures the projection of the P-wave energy onto the vertical component and the S-wave energy onto the radial component. Because the pulse shape of the converted phase is similar to the impinging teleseismic P wavelet, the P wave record on the vertical component can be used to design a filter that enhances converted arrivals on the radial component through a process called deconvolution. Knowledge of the seismic wavespeed beneath the station allows estimation of the depth to the interface by converting the observed Ps-P delay time to a depth Convolutional model and receiver function calculation The traditional model describing the relationship between the input source wavelet and the resulting seismogram has been stated as a convolution between the source wavelet and the earth s seismic response (equation 1.1) where convolution in the time domain is described by equation 1.2. (1.1) R = S *e (1.2) R(t) = e(z)s(t - z)dz Ú - Simply stated the convolution operation (indicated by the star in equation (1.1) takes an input signal, in this case represented by S, and performs a shift, multiplicity, and add operation using a second stationary signal represented by e [Ammon, 1991; 1992; Bostock and Sacchi, 1997; Gurrola et al., 1995; Phinney, 1964]. The result, R, can be thought of as a combination of both the transient input signal and the stationary signal.

11 1-9 In receiver function analysis, we approximate the input wavelet or source term using the vertical component seismogram. We assume that the input wavelet recorded on the vertical component convolved with the earth s response produces the radial component seismogram. By removing the vertical component seismogram from the radial component seismogram through deconvolution, we recover the receiver function or earth s response (equation 1.1). Deconvolution in receiver function calculations can be performed in both the frequency domain [Ammon, 1992; Park and Levin, 2000] and time domain [Ammon, 1991; 1992; Bostock and Sacchi, 1997; Gurrola et al., 1995; Sheehan et al., 1995] and is the focus of much debate among practitioners of receiver function methodology. Recently, improvements in the deconvolution process has increased the usefulness of some lesser quality datasets [Ligorria and Ammon, 1999; Park and Levin, 2000; 2001], but the best deconvolution procedure still needs to be chosen according to the station recording parameters and the experiment design. Unfortunately, the simple convolutional model described in equation 1.1 and 1.2 is only partially correct. If it were exact, the choice of deconvolution would be trivial because the frequency spectra of both the source and the convolution of source and earth s response would be accurately known. In fact, the representation of the source by the vertical component is demonstrably inexact. The approximation is not so poor that meaningful results cannot be recovered from the procedure, but it does significantly hinder the quality of the results. The vertical and radial components actually contain ground motion due to scattering from topography and short wavelength heterogeneities in the near surface (n s ). This signal-generated noise enters equation 1.1 as a convolutional noise term. Also, not accounted for in equation 1.1 is the static background noise that

12 1-10 projects differently on the vertical and radial components and therefore is not always easily identified as either signal or noise (n b ) [Ammon, 1992]. (1.3) R = S *e * n s + n b Equation 1.3 gives us a more realistic equation to describe the relationship between signal, noise, and the recorded seismogram. Because it remains difficult to estimate the noise terms (n s and n b ), we assume that they are absent, thereby guaranteeing errors in the deconvolution. Inaccurate amplitude recovery on the receiver function result from deconvolution in the presence of noise 1.4. Effects of anisotropy and dipping interfaces To this point, we have implicitly assumed an isotropic seismic structure of the crust and upper mantle comprised of flat lying planar interfaces. Learning more about the lithosphere and upper mantle, we realize that these assumptions are not accurate [Ammon, 1992; Bank and Bostock, 2003; Bostock, 2003; Levin and Park, 1997; 1998; Savage, 1998; Vinnik and Montagner, 1996]. Predictable back-azimuthal variations in timing and amplitude of teleseismic-converted phases allow recognition of more complicated features such as dipping interfaces or anisotropic fabrics. A dipping interface refracts P wave energy out of the radial plane for all azimuths of ray propagation not parallel to the dip direction of the interface. The loss of energy out of the radial plane affects the amplitude of the P to S v conversions, which are assumed to be radially polarized and recorded primarily on the radial component (figure 1.2a). The

13 1-11 transverse component records energy projected out of the radial plane. We calculate the transverse component receiver function in the same manner described previously to search for deviations from the assumed simple isotropic structure such as a dipping interface. The amplitude of the converted arrival on the transverse component should show similar systematic amplitude variations including a polarity flip in directions perpendicular to the dip direction. The observed amplitude and arrival time patterns complete a full cycle of variations over a full range of back-azimuths following the predicted 1-q pattern.

14 Figure 1.2 Synthetic receiver radial and transverse receiver functions sorted by back-azimuth for three different earth models [Frederiksen and Bostock, 2000]. (a) Represents a conversion generated by a dipping interface. The amplitude and arrival time variations follow a 1-q pattern over back-azimuth. Row (b) represents a conversion generated by an interface beneath a 10 km thick anisotropic layer. The amplitude and arrival time variations follow a 2q pattern over backazimuth. While, (c) shows the conversion from the base of a thin anisotropic layer with a plunging axis of symmetry. The 1-q pattern from the dipping interface and the 2-q pattern from the anisotropic interface appear superimposed. 1-12

15 1-13 A P wave impinging on the base of an anisotropic medium with horizontal symmetry axis will also produce a Ps converted wave with predictable back-azimuthal variations in arrival time and amplitude. By definition the elastic properties of an anisotropic medium vary with orientation [Crampin, 1977]. Therefore, the amplitude of the converted wave on the radial component varies back-azimuthally because the impedance contrast across the interface differs with azimuth. A shear wave propagating through an anisotropic medium, becomes separated into two separate quasi-s phases (qs 1 and qs 2 ) traveling at two different wavespeeds with the polarization of the faster wave orthogonal to that of the slower S wave [Levin and Park, 1997; 1998; Savage, 1999]. The effect is similar to optical birefringence. An incident P wave, therefore converts into two orthogonally polarized S phases both of which, in general, will be recorded on the radial and transverse components. The arrival time and amplitude of the quasi-s phases depends on the azimuth of the incident P wave, magnitude of anisotropy within the layer, and thickness of the anisotropic layer. The orientation of the anisotropy symmetry axis governs the back-azimuthal patterns. For sufficiently long transit times within the anisotropic medium, a clear time separation between the two S phases will be observable. However, for most crustal conversions traversing anisotropic media, the travel time within the anisotropic media will not be enough to separate the phases. Radial and transverse seismograms recording split crustal conversions show a difficult to interpret combination of the two S phases (figure1.2b and 1.2c).

16 Use of arrays in receiver function analysis To reduce contamination by the signal generated noise and to map regional scale lithospheric structure, passive seismic arrays stretching 10 s to 1000 s of kilometers have become the foundation of many modern receiver function studies [Dueker and Sheehan, 1997; 1998; Gilbert et al., 2003; Gilbert et al., 2001; Jones and Phinney, 1998; Rondenay et al., 2000; Rondenay et al., 2001; Sheehan et al., 1995; Sheehan et al., 1997; Sheehan et al., 2000; Shen et al., 1998; Wilson et al., 2003; Zhu and Kanamori, 2000; Zhu et al., 1995; Zurek and Dueker, in press]. Seismic arrays recording converted phases from the same locus of conversion from many teleseismic events allows the employment of techniques such as common conversion point stacking or seismic migration [Bank and Bostock, 2003; Bostock, 2003; Bostock and Rondenay, 1999; Bostock et al., 2001; Dueker and Sheehan, 1997; Gilbert et al., 2003; Morozov and Dueker, 2003; Neal and Pavlis, 1999; Poppeliers and Pavlis, 2003; 2003; Rondenay et al., 2001; Sheehan et al., 2000; Shragge et al., 2001; Wilson et al., 2003; Zurek and Dueker, in press]. Seismic imaging approaches seek to focus energy at the locus of conversion with the use of an assumed seismic wavespeed structure allowing rough imaging of the subsurface by converting Ps- P delay time to a depth of conversion (depth migration). For all seismic imaging schemes, multiple samples of the same conversion point from different directions are required to recover the amplitude of the original conversion. In theory, the amplitude of the converted phase should provide information about the variation in seismic properties at a conversion interface. Estimation of seismic property variations from imaged converted phase amplitude variations remains a difficult problem due to the incomplete ray parameter coverage observed by teleseismic experiments.

17 1-15 Difficulties arise when multiply scattered free-surface reverberations interfere with deeper primary conversions [Wilson et al., 2003] while other work has shown the identification of multiples to be useful in determining interval Vp/Vs [Zandt and Ammon, 1995], the depth to an interface (assuming accuracy of independently derived wavespeed models) [Wilson et al., 2003], or both with sufficient ray parameter sampling and Ps delay time [Gurrola and Minster, 1998; Gurrola et al., 1994; Zhu and Kanamori, 2000]. Recent incorporation of multiply scattered arrivals into forward scattering imaging schemes has allowed the location of perturbations in P wave parameters, particularly the lame s parameter (l) [Bostock and Rondenay, 1999; Bostock et al., 2001]. However, incomplete knowledge of lithospheric wavespeed structure and interface topography (e.g. a step or dip in the Moho) coupled with poor ray parameter and back-azimuth sampling make use of reverberations in seismic imaging schemes difficult Common conversion point imaging in practice The majority of the observations in this dissertation result from images created from common conversion point stacking techniques as applied to high-density passive seismic arrays. These high-density arrays allow some flexibility with processing techniques not previously afforded by sparser arrays. In this section, I will explain general observations from common conversion point imaging with high-density array data that have arisen from this body of research

18 Pre-stack versus Post-stack depth migration For several decades, exploration seismology has pondered whether to depth-migrate data before or after stacking individual traces. Deciding requires knowledge of the quality of each data trace, which may be difficult to estimate. In post-stack depth migration, the quality of each individual trace is assumed to be poor. To enhance arrivals coherent over spatial wavelengths similar to the lateral extent of the recording array, traces recorded on adjacent stations from a single event are stacked along the predicted moveout curve for first P arrivals from that event. Stacking creates a single trace of improved signal-to-noise relative to single traces for each component assuming the recorded noise becomes decorrelated over the distance between receivers. Post-stack depth migration leads to loss of spatial resolution by smearing arrivals of unexpected ray parameter and in some implementations by reducing the total number of traces in the final processing sequence. Conversely, pre-stack depth migration assumes each individual trace is of sufficient quality to clearly observe converted arrivals without further enhancement through stacking. Deconvolution is performed on individual, unstacked traces, reserving depth migration and trace stacking for the final step. Unaccounted for signal-generated noise found on single station seismograms become amplified during the numerically unstable processing step of deconvolution. In theory, this noise should be removed in the final common conversion point stacked cross-section because data is stacked along predicted plane-wave moveout. With a sufficient number of ray paths from different direction any energy not traveling with predicted plane-wave moveout should be suppressed. Often the

19 1-17 dataset is not sufficient to overcome noise introduced from previous steps especially in the presence of short wavelength interface topography. To demonstrate the effect of pre- and post-stack depth migration on the final common conversion point image we show a single cross-section from a network of dense seismic arrays deployed in the Marlborough fault zone, South Island, New Zealand (see chapter 4 for more detail on this experiment). The cross-sections are created using both (1) single station receiver functions (Fig. 1.3b) and (2) receiver functions calculated after stacking raw, unprocessed seismograms from each component from a set of nearby stations (~500 m apart) for each component (e.g. beamforming; Fig. 1.3a). For the purposes of this demonstration, we focus our attention on the arrival that begins at ~24 km depth at the north end of the cross-section, dips south for approximately 30 km and then flattens at a depth of 34 km. When made with single station receiver functions, we see the arrival dims and becomes blurred as it approaches the dipping section until reaching the flat portion to the south (figure 1.3a). Alternatively, amplitudes of the conversions observed on cross-section (1.3a) remain brighter and crisper through the dipping region than on cross-section (1.3b). This example indicates that artifacts introduced by pre-stack depth migration cannot be accounted for completely by common conversion point stacking degrading image clarity suggesting a post-stack depth migration (beamforming -> stacking) processing flow is optimal for this type of high-density array passive seismic imaging experiment.

20 Figure 1.3 Cross-sections demonstrating the effect of using post-stack depth migrated data (a) and pre-stack depth migrated data (b) on common conversion point images. The clarity and resolution is much better on image (a) especially in the area between 15 and -15 km north of the Wairau fault. 1-18

21 Effect of receiver function quality on the final image Calculating the variance of the radial component and the convolution of the vertical with the receiver function allows the estimation of receiver function quality. Theoretically there should be no difference between the two and the variance would be zero. Errors introduced in the deconvolution process or signal-generated noise projecting differently onto the vertical and horizontal components will increase the variance of the calculated receiver function. Some have suggested signal remains on each receiver function regardless of quality and noise will be uncorrelated for large groups of traces, therefore most of the traces should be used in the final image [Dueker and Sheehan, 1997]. Others have indicated data quality does make a difference and a minimum quality should be imposed that would exclude large portions of the dataset (>70 % excluded) [Beucler et al., 1999]. The importance of data quantity with respect to data quality will be addressed here by reproducing the same image from New Zealand seen in the previous section using processed data of different quality. Figure 1.4 contains four cross-sections with four different minimum variance reduction cutoffs where variance is defined as the radial subtracted from the convolution of the calculated receiver function and vertical component seismogram normalized by the autocorrelation of the radial seismogram. All calculated receiver functions with variance reduction higher than the cutoff are incorporated into the common conversion point image. From the top image to the bottom image, the cutoff increases by 10% from 50% to 80%. Again, we focus on the dipping near 30 km depth conversion interface, as it is the clearest, most consistent feature in the image. On image (1.4a) using a 50% cutoff, the dipping conversion interface discussed

22 1-20 Figure 1.4 Cross-sections demonstrating the effect of using different minimum variance reduction estimates to determine the rayset used in common conversion point imaging. The images represent change in the minimum in increments of 10%, with (a) showing a minimum of 50% and (d) showing an image made with an 80% minimum. in the previous sections does not appear to dip but instead stretches across the Wairau,

23 1-21 Figure 1.4 (continued) Cross-sections demonstrating the effect of using different minimum variance reduction estimates to determine the rayset used in common conversion point imaging. The images represent change in the minimum in increments of 10%, with (a) showing a minimum of 50% and (d) showing an image made with an 80% minimum. continuing at a depth of 26 km. On (1.4b) and (1.4c) the conversion interface returns to the dipping geometry observed in the previous section, appearing smoother and brighter on (1.4c). Cross-section (1.4d) begins to look like (1.4a) again.

24 1-22 A significant reduction in number of useable rays occurs for a variance reduction cutoff of 80% (1.4d; ~60 rays) compared to a cutoff of 70%(1.4c; ~130 rays). Many of these rays also impinge nearly perpendicularly on the dipping section of the interface, significantly reducing the amplitude of the converted phase because rays traveling perpendicular to an interface do not generate converted energy. Absence of oblique ray paths does not explain the loss of the Moho dip on the 50% image where the ~360 rays are almost 3 times those used in image (1.4c) and twice those used in image (1.4b). Instead, the loss of resolution results from the addition of noisy traces. The number of traces and additional directions introduced by lowering the cutoff to 50% does not overcome the additional noise introduced by the lower quality traces. Using traces with variance reduction as low as 50% results in loss of resolution especially in areas where converted wave amplitude depends strongly on propagation direction such as in the section with the dipping interface. It is important to understand that in any teleseismic imaging experiment the minimum variance reduction cutoff should be an empirically derived value based primarily on observable changes in the image quality and coherence and some common sense about target structure geometry. Images from data sets processed for the studies described in chapters 2 and 4 use a minimum variance reduction cutoff of 70% although other practitioners of receiver function analysis have indicated that they often do not use data with variance reduction estimates lower than 80% [Charles Ammon, Craig Jones, personal communication].

25 A Single-chamber Silicic Magma System Inferred From Shear-wave Discontinuities of the Crust and Uppermost Mantle, Coso Geothermal Area, California Abstract. Results from a recent passive seismic study of the Coso geothermal area near Ridgecrest, California suggest heat for the geothermal system comes from a single shallow magma reservoir (~5 km below sea level) that also plays a crucial role in the local change in deformation style from areas to the north and west. The character of the shallow feature and the absence of a lower crustal magma reservoir is inferred from three crustal P-to-S conversions observed using receiver function analysis: (1) A high amplitude shallow negative arrival, Ps-P time of seconds (4-5 km below sea level), (2) a moderate amplitude positive conversion, Ps-P time of seconds (15-18 km below sea level), and (3) the Moho conversion, Ps-P time of seconds (30-32 km below sea level). Observations of Moho converted arrivals indicate the interface is flat and uncomplicated throughout the study area while the mid-crustal conversion is inconsistent and shows considerable topography. The absence of the large negative amplitude conversion on waveforms recorded at stations outside the geothermal area strongly suggests the feature lies only underneath the modern geothermal area. In addition, rays sampling the shallow converter also contain later arrivals with retrograde moveout consistent with an origin as reverberations above the conversion. Synthetic receiver function modeling using a single isotropic layer over a half space indicates that

26 2-24 the shear velocity decreases by 30% across the interface (VS1=2.6 km/s; VS2=1.8 km/s; layer one thickness=4.9 km) further supporting the presence of shallow magma Introduction Volcanic plumbing in the lower crust remains poorly understood. End-member hypotheses include both single chamber magma systems from which a full spectrum of melts are produced through magma mixing and gravitationally induced stratification and multichamber systems that allow melt compositional changes to occur through differentiation within several crustal reservoirs. Geophysical studies of volcanic systems have provided some insight into this problem [Lutter et al., 1995; Masturyono et al., 2001; Plouff and Isherwood, 1980; Reasenberg et al., 1980; Weiland et al., 1995], but often results are ambiguous because of vertical tradeoffs in structures derived from gravity and teleseismic travel time measurements, inherent difficulties produced by lowvelocity zones, and the absence of deep crustal earthquakes in the volcanic centers. Geochemical efforts also suffer from ambiguities, which make it difficult to discern between the various models of reservoir development [Bacon et al., 1980; Bacon et al., 1981; Duffield et al., 1980; Manley and Bacon, 2000; Verplanck et al., 1999] One of the largest geothermal fields related to young silicic volcanic centers in the Basin and Range, the Coso geothermal field has been extensively studied geochemically [Adams et al., 2000; Bacon, 1982; Bacon et al., 1980; Bacon et al., 1981; Duffield et al., 1980; Manley and Bacon, 2000] structurally [Duffield et al., 1980; Roquemore, 1980; Whitmarsh, 1998], gravimetrically [Plouff and Isherwood, 1980], and seismologically [Lees, 2001; Lees and Wu, 1999; 2000; Reasenberg et al., 1980; Wu and Lees, 1999;

27 2-25 Young and Ward, 1980] Geochemical studies have established possible compositions and source depths for the extruded melts while the geophysical studies have been helpful in outlining the depth and character of the top of the upper crustal reservoir and geothermal production area. However, like other recently studied volcanic systems, the connection of the upper crustal magma reservoir to its deeper origins remains poorly resolved. To explore the whole volcanic plumbing under the Coso geothermal area and its possible relationship to regional tectonics, the lithospheric structure of the area was studied using receiver functions computed from seismograms recorded by a group of high-density seismic arrays deployed between November 1998 and May Geologic and Tectonic Setting The Coso geothermal field near Ridgecrest, California, occupies the corner between three tectonic regions: the central Basin and Range, the Sierra Nevada block, and the Mojave Desert. The geothermal field is in the western part of the Eastern California Shear Zone [Dokka and Travis, 1990]. Large-scale west directed normal faulting characterizes this part of the Basin and Range [e.g., [Wernicke, 1985] and may root into a low-angle deformation zone that extends under the Sierra. This shear zone extends under the Sierra and Basin and Range to the north [Jones and Phinney, 1998] and possibly northeast [Zhou and Phinney, 2000] of the Coso region. [Jones, 1987] suggested that the Coso area acts as an accommodation terrane, absorbing differences between northwest directed extension to the north and west-southwest oriented strike-slip faulting along the Garlock Fault to the south. This supports a change in deformation style from that inferred

28 2-26 to the north by [Jones and Phinney, 1998] to something different near the Coso geothermal area. Two distinct episodes characterize magmatic production in the Coso area over the past four million years. The first episode erupted a broad spectrum of volcanic rocks (basalt, andesite, dacite, and rhyolite) from 4 Ma to 2.5 Ma. This was followed by a period of intense bimodal magmatism (figure 2.1) from 1 Ma nearly to the present that produced 39 high-silica rhyolite domes with basalt flows along the periphery of the rhyolitic volcanism [Bacon et al., 1981; Duffield et al., 1980]. Sugarloaf Mountain, emplaced ~40,000 years ago, is the youngest dated dome and is the latest surface expression of a partially molten silicic magmatic reservoir inferred to be present between 5 to 20 km depth beneath the modern geothermal field [Adams et al., 2000; Bacon et al., 1981; Duffield et al., 1980; Reasenberg et al., 1980; Young and Ward, 1980]. Surface mapping of hydrothermally altered deposits supports the existence of the modern geothermal field for at least the last 300,000 years [Adams et al., 2000]. [Duffield et al., 1980] suggested that the heat for this system has been recharged through mantle derived basaltic magmas typical of recent Basin and Range volcanic development.

29 2-27 Figure 2.1 Shaded relief map with geology centered on Coso geothermal area shown with a western United States tectonic map to the right. The red box on the tectonic map indicates the location of the study area. The upper few kilometers of the magmatic system contains dramatic variations in seismic properties [Lees and Wu, 1999; 2000; Malin, 1994; Wu and Lees, 1999] Below it lies a sharp, shallow brittle-ductile transition [Lees, 2001] and reversed polarity reflection about 5 km below the surface [C.W. Caruso et al., A seismic transect of the western Coso Range, eastern central California, unpublished manuscript, 1994] Unfortunately, these recent studies do not penetrate the top of the magmatic system; deeper levels were

30 2-28 sampled using teleseismic P wave tomography by [Reasenberg et al., 1980], who found low velocities in the crust but had limited resolution due to station source geometries available. [Young and Ward, 1980] found attenuation of teleseismic arrivals near the geothermal field. Other surveys using local earthquakes were focused to the south and found little variation in this region, though a paucity of ray paths could explain the absence of identified anomalies [Sanders et al., 1988; Walck, 1988; Walck and Clayton, 1987]. Thus questions remain about the nature of the discontinuity near 5 km depth (is it purely hydrothermal or does it represent the top of a magma chamber?) and the underlying magmatic system Data and Method We recorded ~220 Gb of 40 samples per second, three-component seismograms in and around the Coso Geothermal Area from November 1998 to May 2000; most of the data is available from the IRIS Data Management Center With over 150 sites occupying an area of >2000 km 2, this is one of the densest portable, passive seismic deployments to date (figure 2.1). In an effort to support further use of high-density seismic arrays for lithospheric study, we give a detailed explanation of the experiment design and data processing methods used in this study Array Design and Placement The employment of arrays to improve crustal imaging has been widely used with much success in multichannel reflection processing [Yilmaz, 1997]. Array processing

31 2-29 enhances the desired signal while suppressing scattered energy from short wavelength topographic and subsurface variations [Abers, 1998; Jones and Phinney, 1998; McNamara and Owens, 1993]. Our arrays consisted of 5 8 short-period sensors (1 or 2 Hz free period: Mark Products L22 or L4c, and/or Teledyne Geotech S-13) spaced 500 m apart and arranged into two orthogonal lines with, when available, 1 3 broadband sensors (Guralp CMG3-ESP, CMG-40T). Ground velocity from each station was recorded by a Reftek 72A-07 or 72A-08 data acquisition system (DAS) at 40 samples per second and 32 bits per sample. Timing was corrected from DAS-recorded GPS log files using PASSCAL time correction software. We prefer an orthogonal arrangement of stations because it allows easy visual inspection of arrival moveout and rapid deployment on preexisting road networks (figure 2.1). The station density chosen for this experiment enhances higher frequencies of the teleseismic wave field relative to other passive seismic experiments, improving the overall crustal imaging capabilities of the techniques used here. Because of the heterogeneity of the geology in this region, we deployed arrays in a two-dimensional network (figure 2.1). This permits us to examine the subsurface geology in three dimensions with minimal bias. Arrays were placed on bedrock insofar as was possible; in most cases, the arrays are on Mesozoic plutonic rocks or a thin (<100 m) veneer of Quaternary sediments over Mesozoic bedrock. The exceptions are REN and WHA, which are on the edges of the Airport Lake basin, and POR, which, though on the footwall of the Sierra Nevada frontal fault, is over an unknown and variable amount of sediment.

32 Event Selection An empirical relationship involving magnitude (W, maximum of mb and Ms), depth (G, km), and distance (D, in degrees) developed by [Jones and Phinney, 1998] was used to select teleseismic events for analysis (equation 2.1): (2.1) W * D *G events, ranging from 95 to 29 degrees distant, were chosen according to signal-tonoise ratio estimates for the P arrival with consideration given to the overall backazimuthal distribution of events (figure 2.2). Traces with spurious noise from equipment failures or other causes were dropped from processing Data preparation

33 2-31 Figure 2.2 Map showing the 244 events used in the Coso Geothermal Area receiver function study. From these events, over 476 receiver functions were calculated. The map was generated using the Antelope dbmapevents utility ( Careful visual checking of each event-array pair insured proper timing and quality of traces on all components. Shifting and stacking the recorded waveforms according to predicted moveout of the direct P phase helped in checking station timing for a given event. In a very few cases, a station obviously misstacked due to timing errors (>0.1 s) was removed for that event. The theoretical instrument response was removed from each trace and replaced with the theoretical response of a Mark Products L4c seismometer (free period is 1 s, damping parameter is 0.7). After response correction, beams were

34 2-32 formed for each event-array pair for the vertical, north, and east components by time shifting seismograms to remove plane wave move out of the P phase [Abers, 1998; Langston and Hammer, 2001; McNamara and Owens, 1993]. We calculated the slowness for each event using the IASPEI91 velocity model [Kennett and Engdahl, 1991] and locations reported in the Preliminary Determination of Epicenters (PDE catalog) (as archived at A surface P velocity of 5.5 km/s was used to correct the theoretical move out for topography. Other near-surface velocities were tested but the resulting beams showed little variation probably due to the small incidence angle of the incoming arrivals. Rotations of horizontal component beamed seismograms into the radial and transverse system followed beam forming because both horizontal components weren t always available at each site. Before further processing, the beams were band-pass filtered in the frequency range of Hz to remove microseismic noise and high-frequency background noise present at some arrays Receiver Function Calculation and Compositing The goal of receiver function analysis is to enhance the arrivals of receiver side P- to-s converted phases (figure 2.3) [Burdick and Langston, 1977]. This is done through deconvolution of the vertical from the radial component seismogram. The result is the receiver function, which represents arrivals of various Pds conversions after the direct P wave. In this paper we follow the convention that all letters following the initial phase are upper case if downgoing and lowercase if upgoing and the subscripted letters denote the new phase initiation point with the letter d symbolizing some arbitrary velocity discontinuity (figure 2.3). Thus a Moho conversion is PpMs and a reverberation of an

35 2-33 upgoing P to a downgoing P to an S reflected from the Moho would be PpPMs. The amplitude of the converted arrival is a function of the impedance contrast across an interface and the incidence angle of the incoming wave [Langston, 1977]. A conversion from the top of a low velocity body will produce a conversion reversed in polarity from the impinging wavelet. Particle motion of Pds conversions from flat lying planar isotropic structures will be radially polarized, therefore a conversion from a discontinuity should be consistent for all back azimuths for the same incidence angle. We calculated radial Figure 2.3 Ray paths of forward and backscattered phases generated by an impinging teleseismic P-wave front. Solid lines mark P ray paths with S rays marked by dashed lines. Uppercase letters denote downgoing phases with lowercase letters symbolizing upgoing phases. Teleseismic phases are assumed to be initially downgoing. receiver functions from the filtered, beamed seismograms. Several techniques yielded similar results: a standard frequency domain Oldenburg deconvolution [Oldenburg, 1981]a least squares time domain deconvolution [Bostock and Sacchi, 1997; Gurrola et al., 1995], and an iterative cross-correlation deconvolution [Ligorria and Ammon, 1999]. Some variations between the results aremost likely related to the damping parameters chosen for the particular deconvolution method [Ammon, 1992]. We use an iterative method relying on a cross correlation of the radial and vertical component seismograms [Ligorria and Ammon, 1999]. A spike on the evolving receiver function is inserted at the

36 2-34 location of the largest amplitude of the cross correlation with the amplitude of the spike scaled by the inverse of the vertical autocorrelation. After each iteration, the receiver function is convolved with the vertical component seismogram. The result is subtracted from the radial to remove radial energy already accounted for by the receiver function. The modified radial is then used for subsequent Figure 2.4 Velocity models used for common conversion point stacking (solid lines). Also, shown is the velocity model used for moveout analysis and receiver function modeling discussed later (dashed line). The UCN and MCP boundaries represented by the horizontal lines indicate the locations of conversion interfaces observed on the receiver functions from this study. iterations. The convolution of the receiver function with the vertical seismogram is compared to the appropriate filtered radial seismogram to calculate the variance reduction. Once the variance reduction improvement between iterations falls below a threshold or the number of iterations reaches a predetermined limit, the procedure is terminated. Receiver functions reducing the variance by greater than 70% were used in the final analysis. To compare and stack events from different distances and back-azimuths, we trace Pds rays under the region to 200 km depth using the same velocity structure as [Jones and Phinney, 1998] (figure 2.4). A thick gradational Moho instead of a sharp boundary between crust and mantle velocities reduces errors in amplitude and depth from

37 2-35 improper placement of the Moho in an a priori model. Ray tracing allows the projection of the conversions to a pseudodepth, making single array back-azimuthal stacks and common conversion point stacks possible. In addition, we correct receiver function amplitudes to those of a fixed incidence angle [Jones and Phinney, 1998]. Rescaling of conversion amplitudes prevents bias when stacking over varying incidence angles and permits interpretation of arrival amplitude as directly proportional to the S wave impedance contrast across an interface Common Conversion Point Stacking To study the lateral variations in subsurface features and further mitigate effects from unwanted energy, we stack the receiver functions (e.g., figure 2.5) into common conversion point (CCP) bins (figure 2.6). Significant amounts of energy on receiver functions can result from signal-generated noise: for instance, scattering from surface basins [Levander and Hill, 1985; 1985] and nonplanar interfaces [Clouser and Langston, 1995]. Stacking data from multiple arrays together further diminishes signal-generated noise and scattered energy that may

38 2-36 be coherent at one array, but commonly stacks down when combining data from multiple arrays. Figure 2.5 Back-azimuthally binned receiver functions from four arrays shown on the map in figure 2.1. The receiver functions have been projected to depth using the CCP stacking velocity model shown in figure 2.4. The depth is plotted on the horizontal axis in kilometers below sea level. The back-azimuth bin is shown on the left vertical axis. The right vertical axis indicates the number of receiver functions in each bin. The attribute is also indicated by the change in saturation of the fill for each binned receiver function. These plots demonstrate the rapid change of the UCN depth within individual arrays. Step like east-west variations can be seen in the left column of plots from arrays CGT and CDR while smooth north-south variations of the UCN depth can be seen on the right two plots at arrays JSH and SSL.

39 2-37 Figure 2.6 Shaded relief map centered on the Coso geothermal area demonstrating several aspects of the CCP processing. The lines marked 1 and 2 indicate the location of CCP profiles shown later in figure 2.7 Each profile averages a swath 9 km centered on the gray transparent stripes. Each CCP voxel is an average of the Ps conversions from a depth range of.25 km over a 9 km X 9 km grid as illustrated by the grey grid appearing nearing the southern end of cross-section 2. The geographic stacking procedures used here largely follow those of [Dueker and Sheehan, 1997; 1998] with modifications. P d s arrival piercing points, which have been found using the ray tracing techniques described above, are binned using a 3-D grid of sample points spaced 3 km apart horizontally and 0.25 km apart in depth within a 47

40 2-38 (east-west) by 43 (north-south) by 65 (vertical) km volume. The piercing points within a 9x9 km2 area centered on each sample point are stacked at that point (figure 2.6). Array elevation corrections were made before common conversion point stacking based on the predicted angle of incidence, the station elevation, and near-surface velocity information from recent active source studies [Pullammanappallil et al., 2001; Unruh et al., 2001; Unruh et al., 2001; Unruh et al., 2000]. Changing the size of the bins, or weighting rays differently depending on their distance from the bin center point, does not greatly influence our results. Bootstrap resampling with replacement within individual CCP bins, performed in order to estimate the error associated with that trace [Efron and Tibshirani, 1986], helps assure that a few errant receiver functions do not control the results. We calculated one hundred bootstrap realizations of each stack and present their mean and standard deviation for each bin to give an idea of the reliability of the results for each bin. Images produced by common conversion point stacking may suffer from effects of velocity variations that could influence interpretation if unrecognized. All the depths presented are calculated by tracing rays through a one-dimensional structure as discussed previously. Lateral variations in seismic velocities will cause features to be mapped to erroneously deep depths below low S velocity bodies and shallow depths beneath high S velocity bodies. In extreme cases, conversions from a single point to different arrays will not stack at the same depth, blurring the feature. Recognition of this effect can identify substantial variations in velocity, as we discuss below, but unless quite large over a short distance, the effect on interpretation is minor. Lateral velocity effects could be corrected by tracing rays through a three-dimensional structure, but the absence of a crustal-scale three-dimensional structure precludes such analysis.

41 Seismological Observations Analysis of the receiver functions reveals a wealth of converted energy from the crust under this region. We have identified three principal discontinuities that help to illuminate the tectonic features of the crust in this region: Upper Crustal Negative (UCN) - Shallow negative conversion characterized by a high amplitude swing at 4-5 km below sea level in the vicinity of the geothermal field. The discontinuity appears to dip to the north along its eastern edge. The depth and amplitude of this converter varies over short horizontal distances to the south and west of the geothermal field. Moho - A prominent positive arrival that runs throughout the study area at ~31 km below sea level. Previous studies from this area detected an interface believed to be the Moho at similar depths [Fliedner et al., 2000; Jones and Phinney, 1998]. Mid Crustal Positive (MCP) A positive discontinuity characterized by a high amplitude pulse near 15 km below sea level. Under the upper crustal negative, the amplitude of the arrival decreases as the depth increases 2-3 km.

42 2-40 The Figure 2.7 East-west (1) and north-south cross-sections (2) located on the map in figure 2.6 stacked using all rays (top row), no rays sampling the UCN (middle row), and only rays sampling the UCN (bottom row). The vertical axis is marked in kilometers below sea level while the horizontal axis marks the distance from the coordinate center at Coso Hot Springs, CHS. The cross-sections illustrate the disrupted Moho arrival from near 31 km depth and the high-amplitude shallow negative arrival near 5 km depth (UCN). The mid crustal positive is also visible near 16 km depth below sea level (row 1 and 2). However, it appears disrupted and depressed in depth beneath high-amplitude portions of the upper crustal negative arrival. The cross-sections with the UCN rays excluded demonstrate a surprisingly consistent depth and amplitude in the Moho converted arrival. The MCP arrival also appears to have less distortion. following discussion details important changes in these conversions found on CCP

43 2-41 stacked cross sections as well as single array back-azimuth stacks. We provide the CCP sections and back-azimuth stacks pertinent for the following observations Upper Crustal Negative Location and Character The shallow negative arrival observed on the individual array back-azimuth stacks (figure 2.5) and the cross sections through the center of the geothermal field (figure 2.7; cross sections located on figure 2.6), appears strongest in the region between 4 km east to 10 km west and 5 km north to 10 km south of the coordinate center at Coso Hot Springs (figure 2.1). The converter resides near 3 to 4 km depth below sea level in this area (figure 2.7). The large amplitude of the UCN arrival suggests a dramatic decrease in shear velocity below the interface. The depth of this low-velocity zone is consistent with previous geophysical studies from the Coso geothermal area [Lees and Wu, 1999; 2000; Malin, 1994; Pullammanappallil et al., 2001; Unruh et al., 2001; Unruh et al., 2001; Unruh et al., 2000; Wu and Lees, 1999]. The depth of the UCN varies over short distances at the southern and western edges of its core area (figure 2.8). The single array back-azimuth stacks from JSH, SSL, CDR, and CGTshow rapid variations in the depth of the UCN with back-azimuth (figure 2.5). JSH shows a deepening to the north and west from depths of 7 km, while a few kilometers to the north, SSL s arrivals shallow to 5 km.

44 2-42 Figure 2.8 Character and extent of the UCN conversion from CCP stacks. Colored blocks are the amplitude of the conversion; converter depth is contoured in kilometers below sea level for the regional velocity structure (depths shallow by 1.5 km using the UCN velocity structure). Rays penetrating the UCN (for purposes of stacking, e.g. figure 2.9) are in yellow; otherwise are blue and are plotted at 5 km depth. Note increasing depth and increasing amplitude to the south and west. Green line separates rhyolite-dominant interior from basalt-dominant exterior. Coso Hot Springs is at the corner of the L of the CHS array. Assuming these shallow arrivals indicate the location of the shallow discontinuity beneath each array, the depth variation between arrays may be explained by a grossly east-west trending fault with 2 km of apparent vertical offset. The deeper arrivals to the north and west observed at JSH can be explained as a diffraction tail generated by arrivals encountering the fault from different directions. The deepest arrivals at SSL are sampling the offset UCN from beneath the JSH array and to the east of JSH. This is

45 2-43 probably the high-angle fault inferred from geologic mapping to trend about northwest between SSL and JSH [Bacon et al., 1980; Roquemore, 1980; Whitmarsh, 1998]. To the west, CGT and CDR image a dip in the western boundary of the converter. For eastern back-azimuths, both arrays show a negative arrival from a depth near 6 km. UCN arrivals from the west indicate a deeper origin near 9 to 10 km depth. This could point to the presence of a north-south trending fault with a 3 to 4 kilometer offset or a strong dip of the conversion to the west. This edge of Rose Valley is the site of significant normal faulting [Bacon et al., 1980; Roquemore, 1980; Whitmarsh, 1998], and this is plausibly a subsurface extension of this faulting Complications of the Moho and Mid Crustal Positive Arrivals Detailed study of the CCP stacked sections reveals large short-wavelength variations in amplitudes of the Moho and the MCP conversions. In cross-section 1 (figure 2.7), the location of low-amplitude and distorted Moho arrivals is under the stronger parts of the shallow negative anomaly. The MCP arrival shows similar amplitude variations, which is particularly clear in cross-section 1 (figure 2.7, top). In addition, the MCP appears to be depressed by several kilometers under the UCN. The presence of a strong UCN conversion above lateral changes in the deeper converters suggests either that the UCN causes these changes by attenuating and/or delaying arrivals from greater depth, or that the UCN is part of a larger feature extending to the Moho. The former seems most plausible: an area of high shear wave attenuation associated with the shallow negative conversion could attenuate converted rays from deeper discontinuities such as the Moho or the MCP. Converted energy from these

46 2-44 discontinuities traversing the low-velocity region would appear to be from greater depths due to conversion misprojection related to the low-velocity zone. However, these surmises do not inform us of the depth extent of anomalously slow crust. To consider the effect of upper crustal elastic parameter variations on deeper conversions, the complete CCP sections were remade excluding the rays piercing the high amplitude sections of the UCN (figure 2.7, middle row). In this image, the Moho arrival appears consistent and at a nearly constant depth across the images. In addition, the MCP arrival has less topography and stronger amplitude beneath the UCN in the new images. The sections stacked only with rays piercing the UCN (figure 2.7, bottom row) show several high-amplitude midcrustal arrivals that will be discussed in detail in the next section, but the Moho and MCP arrivals are weak or even nonexistent at the depths seen in figure 2.7 (middle row). These deeper conversions might be present at greater model depths (e.g., the Moho might be the positive anomaly ~42 44 km model depth); this would be consistent with large velocity pull down from a major low velocity zone. These observations relate the amplitude and topography of deeper conversions to low velocities and moderate attenuation of the region at and immediately beneath the UCN Post-UCN Arrivals The restacked CCP images demonstrate the presence of a high-amplitude negative arrival from near 23 km depth and a low-amplitude positive arrival from ~29 km depth on the cross sections using only UCN rays (figure 2.7, bottom row). The arrivals are noticeably absent on the cross sections where UCN rays were excluded (figure 2.7, middle row). An important attribute of these arrivals is their confinement beneath the

47 2-45 high-amplitude portions of the UCN. To determine the nature of these localized post-ucn arrivals, the receiver functions from rays sampling the UCN and rays not sampling the UCN were stacked in ray parameter bins to examine variations in arrival time with changing event distance (figure 2.9). Commonly known as moveout or tau-p analysis, plotting receiver functions in this manner allows the easy identification of reflected phases because of their reverse or retrograde moveout [Bacon et al., 1980; Gurrola and Minster, 1998; Gurrola et al., 1994; Roquemore, 1980; Whitmarsh, 1998]. The solid lines indicate the theoretical moveout curves for a conversion from a discontinuity shown at the depth indicated to the right at the top of the curve. The most prominent arrivals are the high-amplitude, shallow, negative arrivals from 4.9 km below the surface on the UCN receiver function plot and the strong, consistent Moho on the plot with rays not sampling the UCN. Other arrivals include the PpPUCNs reverberation and the PsPUCNs + PpSUCNs reverberation off a discontinuity at 4.9 km below the surface marked by the long dashed and short dashed curves, respectively (figure 2.9; see figure 2.3 for illustration of the various backscattered and forward scattered phases). The retrograde moveout demonstrated by the theoretical reverberation curves make the reflected arrivals easy to distinguish from direct phases. Both the negative arrival at 23 km depth on the CCP stacks and the positive arrival near 29 km depth (figure 2.7, bottom) match predicted times, polarity, and retrograde moveout of reverberations from the top of the low-velocity zone at the UCN.

48 2-46 Figure 2.9 Receiver Functions stacked according to ray parameter (with averaging over s/km) for rays sampling the UCN (top) and rays not sampling the UCN (bottom). The horizontal axis represents time after the P arrival. The right vertical axis represents the number of receiver functions used within each ray parameter bin. The solid curves show calculated moveout curves of converted arrivals from depths plotted at the top of the curves. The long dashed line and short dashed line represent PpPs reverberation and PsPs+PpSs reverberation fro 4.9 km depth, respectively. The curves in the top plot were calculated using the velocity model shown as a dashed line in figure 2.4. The curves of the bottom plot were calculated using the standard CCP stacking velocity (figure 2.4). Receiver function modeling using synthetic data generated by the technique of [Keith and Crampin, 1977] helped to further constrain the shallow velocity structure within the geothermal field. All receiver functions from rays piercing the UCN region

49 2-47 described in the beginning of section 4.1 were stacked to create a mean representative receiver function for the UCN region. Receiver functions calculated from synthetic data were compared to the mean UCN receiver function until the best fit was found. The bestfit criteria were arrival time, Figure 2.10 Synthetic receiver function calculated from model 1 (table 1) plotted with the mean receiver function for the rays sampling the UCN anomaly. The dashed lines represent the standard deviation associated with the mean receiver function. The Pds and PpPs arrivals are fit nicely but the synthetic amplitude exceed the observed amplitude for the PpSs+PsPs arrival. The diagram included in the waveform window demonstrates one contribution to the observed decrease in amplitude. The loss of expected energy is due to the limited area where the PsPs ray can be generated. amplitude, and pulse shape. The model space was thoroughly explored, with particular attention paid to the layer 1 P and S velocities, the S velocity contrast across the UCN interface, and the density contrast across the interface (Table 1 indicates possible variations of these model parameters). To fit the amplitude of the various conversions and reverberations, the two-layer isotropic velocity model requires a decrease in shear wave velocity across the UCN of 30% (+13/-3%) (figure 2.10 and table 2.1). The PpPUCNs arrival is sensitive to density contrasts making it possible to constrain the density contrast across the UCN to be g/cm (-0.05/+0.15 g/cm). The resulting synthetic receiver function fits the observed mean receiver function for the primary P-to- S conversion as well as the PpPUCNs arrival but the amplitude for the PsPUCNs + PpSUCNs is

50 2-48 larger than that observed. The loss in amplitude is most likely due to the limited lateral extent of the UCN and its effect on the PsPUCNs + PpSUCNs arrival (figure 2.10). Reflectivity based synthetic modeling [Kennett, 1983], which permits variations in Q, produces slightly better pulse shape matching of the forward scattered converted arrival, but the required density contrast necessary to fit the PpPs reverberation is unrealistic. The predicted amplitude of the PsPs + PpSs arrival is still a factor of 2 too large. Therefore we believe that although attenuation in the region above the UCN may be important, it may be difficult to distinguish from other effects such as destructive interference of multiply scattered waves generated within the low-velocity region. The absence of significant pull-up or push-down on the Moho and possibly MCP conversions permits us to address the thickness of the low-velocity zone beneath the UCN conversion. A simple geologic structure consistent with previous teleseismic imaging [Reasenberg et al., 1980] would extend the low-velocity zone down to 15 km bsl without change in velocity. The base of this zone would then be a positive increase compatible with the presence of the MCP but probably demanding much greater amplitude. It would also move the expected Moho conversion arrival time to almost 7.5 s after the direct P arrival, where no arrival is seen (figure 2.9). Finally, enough of the rays not penetrating the UCN do extend under it enough to have their Moho PMs conversion delayed were this structure correct; the Moho does not show such pullup (figure 2.7, middle). Unless a compensatory high-velocity body directly underlies this low-velocity zone with a geometry preventing detection, these observations limit the vertical extent of a constant velocity low-velocity zone to about 10 km. A more likely velocity structure under the UCN is found by incorporating a likely

51 2-49 Moho arrival demonstrating prograde moveout at ~5.3 s (figure 2.9). Moho depth seen on rays under but not penetrating the UCN is 31.5 km bsl (figure 2.7, middle row). Applying a simple velocity model (figure 2.4) that linearly increases velocities to normal lower crustal velocities from about 5 km bsl to near the depth of the MCP (~15 km bsl; figure 2.4) places the 5.3 s conversion near 31.5 km bsl figure 2.9 shows the fit on both moveout plots for a Moho depth of 32.9 km below the surface (31.5 km bsl) using the appropriate 1-D structure. Although the details of velocity within the low-velocity zone might be noticeably different, the overall delay from this zone is well fit with this model. Additionally, decreasing the velocity anomaly with depth makes it unlikely to have influenced rays not penetrating the UCN Discussion Geometry and Inference of an Upper Crustal Magma Chamber The UCN conversion is the strongest crustal conversion in the area and is tightly tied to the geothermal field and recent silicic volcanism. Thus this converter is a product of Quaternary volcanism and provides important clues to the structure of the Coso volcanic center. The discontinuity is highly localized, with a center just south and west of the Coso Hot Springs area (figure 2.8). The amplitude is greatest within a 14 km by 15 km square area (~210 km 2 ) where the converter is about 5 km below the surface when using the reverberation velocity structure of figure 2.4 (figures 2.5, 2.7, 2.9, and 2.10). The boundaries on this region are only defined within 3 5 km because of spatial averaging of the stacking techniques. Ray coverage is particularly poor to the southeast (figure 2.8), and the largest errors in the UCN boundaries may be found in this region

52 2-50 (figure 2.6). The core UCN region is bounded on the south and west by sharp increases in the converter depth, which we earlier related to faulting (section 2.4.1, figure 2.5). These terminations are not far from the edges of the recognized geothermal field. Structural termination of the UCN strongly indicates that upper crustal geothermal systems (and possibly magmatic systems) can be controlled by tectonic boundaries and do not merely taper out laterally. These observations may also support previous assertions that the buoyancy of the magma body may be responsible for arcuate structural features observed in the area [Duffield et al., 1980]. As noted before, existing geophysical work provides few constraints on the magmatic system beneath the UCN. The low-velocity body could either be a hydrothermal system, possibly highly overpressured, well above magma, or it could be magma itself. The rhyolitic volcanism and petrologic relationships in the volcanic field require a crustal magma chamber [Manley and Bacon, 2000], so if there is not another candidate magma chamber, we can proceed to interpret the characteristics of the upper crustal low-velocity zone for magma present. We first consider the possibility of deeper magma chambers beneath the observed low-velocity zone, and then use geophysical observations to constrain the bulk properties of the crust under the geothermal area. Finally, we address the implications of our inferred structure for deformation of the crust Absence of Lower Crustal Magma Bodies Some geophysical studies of other volcanic systems support the presence of a lower crustal reservoir [Lutter et al., 1995; Steck et al., 1998; Weiland et al., 1995] or

53 2-51 melt filled basaltic feeder system extending to a mantle source [Masturyono et al., 2001]. Any candidate magma bodies within the crust should have low S velocities and thus delay any deeper Ps conversions and generate Ps conversions of their own, particularly at the top where a velocity decrease with depth should exist. A lower crustal magma pool would certainly produce large negative conversions comparable to the UCN on both UCN-penetrating rays and rays from adjacent arrays that undershoot the UCN. The only plausible candidate is between CHS and BPT at a depth of 25 km (cross-section 1, figure 2.7, middle rowf); the amplitude of this anomaly is a quarter of the UCN, and the anomaly is only observed from a single array (BPT), although it might be delayed and present in the CHS gathers. We view this as a very weak candidate for a lower crustal magma chamber. Otherwise, the absence of strong conversions and the complications in Moho depth such a velocity perturbation should introduce to the CCP sections make such a body unlikely under Coso. A pervasive, meltfilled feeder dike system might not produce strong and coherent Ps conversions, but the absence of velocity pullup or pushdown artifacts on the Moho image make this an unlikely scenario. Instead, the upper crustal reservoir may be fed either through discrete injections of basaltic melt from the mantle or by small dikes. Basaltic injections through these mechanisms would provide heat and volatiles to the previously stratified reservoir while leaving the seismic velocities in the lower crust sufficiently high that artifacts would not be produced on our images of the Moho. Input of volatiles and heat to the upper crustal reservoir eventually forces eruption of high silica magma from the top of the reservoir, producing one of the many high silica rhyolite domes. Injections not trapped by the upper crustal reservoir would be extruded outside the area of rhyolite

54 2-52 domes, producing the bimodal magmatism and allowing the main geothermal field to remain mostly free of basalt [Manley and Bacon, 2000]. To establish the limits of the width of melt-filled basaltic dikes feeding the upper crustal magma system without significantly reducing lower crustal velocities, we consider vertical dikes (or vertically elongated magma chambers) of varying widths and their impact on the timing of the Moho Ps arrival. Assuming seismic waves are sensitive only to structural features longer than a quarter wavelength, and noting that teleseismic energy recorded by our short-period instruments is peaked near 1 Hz, a dike 500 m wide could remain undetected by incoming waves traveling at velocities as low as 2 km/s (a more normal lower crustal P velocity of 6 km/s might not detect a dike 1.5 km wide). In addition, if this main dike were divided into smaller dikes and distributed throughout the lower crust beneath the UCN then the effect on individual teleseismic arrivals would be even less. Thus we suggest that any melt-filled dikes currently present under the upper crustal magma body can be no more than a few hundred meters wide and must be distributed irregularly through the area. Attenuation in the crust can also be used to explore for melts and fluids. We observe attenuation of Ps conversions traversing the UCN: Moho and MCP converters are noticeably lower in amplitude for these rays (figures 2.7 and 2.9). Previously, [Young and Ward, 1980] and [Lees, 2001] inferred the presence of a high-attenuation region in the shallow crust above the depth of the UCN (<5 km depth), presumably a reflection of severe hydrothermal activity. [Young and Ward, 1980] also inferred some attenuation from the mid to lower crust (12 20 km depth) within the Coso area. If most of the attenuation is from the top 5 km, as [Young and Ward, 1980] suggest, then the UCN is an

55 2-53 even greater velocity contrast than in our velocity structures, and the presence of the strong PpPs reverberation above the UCN would be difficult to explain. In contrast, an absence of attenuation within the low-velocity zone below the UCN should produce large amplitude reverberations from within that layer. Attempts to model such reverberations with synthetic seismograms lacking attenuation produced large amplitude arrivals not visible on our seismograms. We suggest that the bulk of the attenuation we and [Young and Ward, 1980] observe is from between about 5 and 15 km depth and that this is probably compatible with the limited vertical resolution of the teleseismic observations used by [Young and Ward, 1980] Melt in the Upper Crustal Magma Chamber From our analysis above, it seems likely that the body under the UCN is magma, since there is no deeper candidate within the crust; some amount of hydrothermal fluids immediately above the UCN must exist to drive the geothermal field. [Nakajima et al., 2001] showed a relationship between VP/VS and VP for crustal rocks with water and melt filled cracks of different aspect ratios. A VP/VS and VP similar to our modeled values for layer 1 within the geothermal field (VP/VS = 1.94; VP = 4.9 km/s; figure 2.4) would predict a system with a crack aspect ratio between 0.01 and and ~1 2% hydrothermal fluid. Estimates of the layer 2 P velocity are ambiguous because of low sensitivity of the arrival amplitudes to the layer 2 P velocity. Our only control on sub-ucn P velocities is the forward modeling used in matching the Moho arrival on the moveout plots for rays sampling the UCN; this is consistent with previous P wave tomographic studies in this region [Reasenberg et al., 1980]. Velocities directly beneath the UCN (VP/VS = 2.5; VP =

56 ) would indicate 1.5 5% melt for a crack aspect ratio between 0.01 and If, in fact, the P velocity is also much lower just beneath the UCN, both the percentage of melt and aspect ratio could be much higher. The gravity signature of the magma reservoir further constrains the character of the magma chamber. [Plouff and Isherwood, 1980] previously noted the ~10 mgal anomaly associated with the geothermal field. Approximating a magma body as an infinite Bouguer slab, we can determine the minimum thickness of a melt column contained within the region beneath the UCN. Assuming a density of 2700 kg/m3 for the average crustal density, crystal free rhyolite melt would create a density contrast of 400 kg/m3; a crystal free basalt melt would have little or no density [Bacon et al., 1980; Bergantz and Dawes, 1994]. Thus a column of rhyolitic magma represents the minimum amount of magma capable of producing the observed gravity anomaly. The minimum thickness of melt is 0.6 km. If the melt percentage beneath the UCN is no higher than 5%, as suggested by velocities, a column of 5% melt stretching to 17.5 km bsl would produce the observed gravity anomaly. The column would stretch even farther into the lower crust if the melt percentage decays with depth or is compositionally zoned to more mafic compositions with increasing depth; it would be less if a considerable part of the gravity anomaly is caused by hydrothermal alteration of granite in the top few kilometers. Larger melt percentages may be present for more mafic melts. In contrast, if the density contrast of 0.15 Mg/m3 inferred from fitting of reverberations is used, the melt percentage just under the UCN is 30 35%, and such a region of melt could be as thin as 2 km and produce the observed gravity anomaly. A structure with ~5 20% rhyolite melt at the top of the magma chamber,

57 2-55 decreasing with depth to be nearly melt free at a depth of about 15 km is most compatible with our observations. This decrease with depth could be more irregular than the smooth change in the simple velocity model used for the moveout analysis (figure 2.4). The positive arrival shortly after the UCN could reflect the base of a fluid-rich part of the magma system, and some of the deeper positive arrivals could be from progressively more melt-poor parts of the crust (figure 2.7). A general decrease in melt percentage with depth is most consistent with the amplitudes of the reverberations above the UCN, the estimates of P wave velocities of other studies, attenuation of Moho Ps conversions on rays penetrating the UCN, the gravity anomaly, and the absence of velocity push-down features of the Moho Ps conversion for rays undershooting the UCN. The mantle reservoir feeding the crustal reservoir is possibly the negative anomaly observed in the upper mantle on cross section 2 near 33 km depth (figure 2.7). The arrival is present on the CCP stacks using all rays as well as the stacks made with rays not sampling the UCN. Its presence on the CCP sections not including the UCN rays indicates that it is not a reverberation resulting from the low-velocity zone even though it partially resides beneath it. Instead, we interpret the anomaly as the seismic expression of an upper mantle magmatic staging area instrumental in the recharging of the Coso volcanic system Relationship to Regional Tectonics Previously, [Jones and Phinney, 1998] inferred the presence of a midcrustal shear zone slightly to the north of Coso based on observations of a midcrustal anisotropic converter that was dipping down beneath the Sierra Nevada to the west. The

58 2-56 structure was associated with major shear zones extending down from the large normal faults in the region as drawn by [Wernicke, 1992], supporting earlier assertions that the area west of Death Valley was deforming under simple shear [Jones, 1987; Wernicke, 1985]. The converter has since been observed farther to the north and east and appears to underlie a large part of eastern California [Jones and Phinney, 1998; Zhou and Phinney, 2000]. Notably, this shear zone underlies a large region lacking major silicic magmatism. Observations of lithospheric discontinuities in the Coso area provide a unique insight into the interplay of magmatism and tectonics. The anisotropic conversion seen by [Jones and Phinney, 1998] appears to be absent beneath Coso except for a small region at the north end of the study area. This suggests that this master structure accommodating extension to the north, and the mode of deformation it represents, is not at work within the Coso geothermal area. Either this structure has been hidden by igneous reworking of the crust, or it is absent. Assuming the latter, a different style of crustal extension is required in the Coso region. The deformation regime must satisfy observations that shallow faults seem to root into a decollement on top of the magma chamber (J. R. Unruh, personal communication, 2000) and that the Moho seems nearly planar. We propose that extension in the uppermost crust occurring on mapped faults is accommodated within the magma body at midcrustal levels (figure 2.11). This model requires surface faults to root into the magma body as has been inferred from unpublished active seismic profiles [Jones, 1987; Pullammanappallil et al., 2001; Unruh et al., 2001; Unruh et al., 2001; Unruh et al., 2000; Wernicke, 1985]. Extension beneath the magma is unlikely to be localized directly under the magma body: such extension would raise the Moho many kilometers depending on the amount of extension accommodated in this

59 2-57 area. Instead, flow within the lower crust into the region under the magma body could combine with magmatic additions to accommodate extension and preserve the nearly flat Moho observed by this study, much as envisioned by [Gans, 1987] for the northern Basin and Range. As for the upper crust, widespread extension in the lower crust must root upward into the narrower deforming region of the magma chamber. The broadly deforming lower crust will be separated from the near rigid midcrust to the sides of the magma chamber by a zone with substantial shear. Displacement across this zone would increase toward the magma body. More work needs to be done on the nature of the MCP conversion (anisotropy; density and velocity contrast across the interface), but this conversion seems a possible candidate for such a midcrustal decollement (figure 2.11). In particular, its presence near the inferred base of the magma chamber and increase in amplitude approaching the magma chamber seems a good fit to the character expected of such a decollement.

60 2-58 [Jones, 1987] suggested that the Coso area absorbs differences in tectonic strains Figure 2.11 Cartoon showing structure of the Coso geothermal field and the possible relationship between the shallow magma body and regional tectonic features. A new CCP-stack along profile 1 combines non-ucn rays using the original velocity model with UCN rays migrated using the UCN structure of fig. 4. The most important observations is the lack ofa lower crustal magma reservoir as determined by the moveout analysis. The magma body may act as a strain guide in the upper crust. The strain guide coupled with lower crustal flow signals a change in deformation mechanisms in the Coso area compared to mechanisms believed to exist to the north and possibly the east [Jones and Phinney, 1998]. between the area of westnorthwest extension to the north, left-lateral strike-slip motion to the south along the Garlock fault, right-lateral bending of the Garlock related to the Eastern California shear zone, and the rapidly departing Sierra Nevada block to the west. Such a regime, which is more nearly plane strain than the region to the north and east, might provide a more favorable environment for volcanic processes than surrounding areas. Localization of strain that initiated magmatism could well have reflected the extensional step-over geometries of strike-slip faults crossing the region [Unruh et al.,

61 ]. Magmatic injections in the Coso area may have originally resulted from a brittle initial response of the crust to the complex tectonic environment. The injections along with lithospheric thinning under the Sierra would serve to the heat lower crust throughout the accommodation terrain, perhaps making this more pure-shear model of deformation possible Conclusions The presence of an upper crustal magma reservoir situated 5 km below the center of the modern Coso geothermal field has been confirmed using receiver function analysis. This reservoir is between 2 and 15 km thick with >5% rhyolitic melt. Thinner or more mafic reservoirs require higher melt percentages to satisfy our observations. Receiver function modeling combined with moveout analysis has shown that a lower crustal magma reservoir is unlikely to underlie the Coso geothermal area. A possible candidate for an upper mantle reservoir has been detected near 35 km depth. This mantle reservoir probably feeds the crustal magma body with periodic injections or continuous flow in dikes (<1 km width). Strain localization in the shallow magma reservoir probably causes the Coso area to extend differently than extensional terranes to the north. Acknowledgements. This research was conducted with the support of contract number N C-0234 from the Geothermal Programs Office of the U.S. Navy. Acquisition of the seismograms would have been impossible without the diligent help of many members of that office, most especially Mike Hastings, Rick Webber, Bob Johnson, and the continued enthusiastic support of Frank Monastero. Brian Zurek, Seth Mueller,

62 2-60 Shelly Bolus, Damon Lytle, and Bob Phinney provided needed assistance in the field. Scott Whitehead devised a preamp that greatly improved our data quality. Randy Keller graciously provided a number of seismometers. The Ridgecrest Search and Rescue team gleefully assisted in driving a snowcat to recover seismometers buried in a late season snowfall. Receiver function modeling would have been considerably more time consuming without the assistance of Anne Sheehan. Discussions with Lang Farmer helped us better understand silicic magma systems, but any errors we retained are in spite of his efforts. Discussions with Jeff Unruh and Bob Phinney improved our analysis and understanding, as have the discussions accompanying the technical symposia run by the Geothermal Programs Office.

63 Accommodating Crustal Strain within a Volcanic Area: Seismological Observations and Numerical Simulations from Coso Geothermal Area, Ridgecrest, California Abstract. Numerical simulations based on seismic observations made in the Coso geothermal area show that a low viscosity body in the upper crust promotes strain accommodation on sub-horizontal shear zones in the middle crust. Wilson et al (2003) interpreted a low velocity body in the upper crust as partial melt that is underlain by a flat Moho. We suggest that the zone of partial melt accommodates horizontal extensional strain variations between the upper and lower crust. Plane-strain numerical simulations of a low viscosity zone embedded in the upper crust that overlies a less viscous lower crust and mantle show that strain localized within the low viscosity zone of partial melt induces vertical gradients in horizontal strain. The strain gradients produce shear zones that extend horizontally from the top and the base of the low viscosity zone. The shear zones may connect across the low viscosity body, producing a uniformly dipping shear zone that spans the entire crust. Thus, magma induced rheology variations provide a new mechanism to induce low-angle normal faulting in the upper crust and sub-horizontal shear zones in the middle crust.

64 Introduction For almost three decades, a debate has raged about the character of core complex development [Abers et al., 2002; Baldwin et al., 1993; Block and Royden, 1990; Buck, 1991; 1993; Buck et al., 1988; Chery, 2001; Coleman and Walker, 1994; Coney and Harms, 1984; Crittenden et al., 1978; Davis and Coney, 1979; Lavier and Buck, 2002; Lavier et al., 1999; McCarthy et al., 1991; Wills and Buck, 1997]. Some have proposed that shallow dipping fault surfaces observed within exhumed core complexes have simply rotated from their steep initial dip after tectonic unroofing [Axen and Bartley, 1997; Buck, 1988; Lavier et al., 1999]. Others propose principal stress orientations change in the upper ~5 km of the crust surface, allowing initiation of seismogenic slip on low angle normal faults [Wernicke, 1981; 1985; 1995; Wernicke and Burchfiel, 1982]. Recent evidence of seismogenic slip on low-angle fault surfaces has provided new support for the latter low angle fault model [Collettini and Barchi, 2002; Jackson and White, 1989; Wernicke, 1995]. It remains difficult to reconcile mechanics arguments with seismotectonic and structural field observations. Several mechanisms for stress field rotation within the shallow crust have been proposed to explain slip on low angle normal faults. Parsons and Thompson [1993] suggested dike tip propagation would locally reorient the principal stresses and therefore allow low angle normal faults to initiate. Chery [2001] proposed that elevated pore pressures on initially steep fault planes would reorient the local stresses and lead to a rotation of the slip surface. Others have invoked either rheology differences between the upper and lower crust [Melosh, 1990; Wells, 2001] as a mechanism to rotate principal stresses in the shallow subsurface within extending regions.

65 3-63 Regardless of which theory is followed, most workers agree that core complexes develop as local strain rates and magmatism increase [Coney and Harms, 1984; Gans, 1987; Lister and Baldwin, 1993; Parsons and Thompson, 1993; Ruppel, 1995; Wells, 2001]. The two are most likely related, though whether magmatism is cause or effect for increased extension remains unclear: rapid unroofing could trigger decompression melting [Hill et al., 1995] or magmatism could localize strain within the thermally weakened area [Buck, 1991; Ruppel, 1995]. Dikes frequently pre- and post-date normal faults in core complexes, demonstrating the simultaneity of extension and magmatism within these structures [Hill et al., 1995; Holm, 1995; Lister and Baldwin, 1993; Walker et al., 1995]. Although a clear connection exists between magmatism and low angle normal faulting, the kinematic and temporal relationship between the two remains unclear. We seek to address some of these unanswered questions surrounding core complex development and its association to magmatism using results from a passive seismic imaging experiment within the Coso geothermal area near Ridgecrest, California [Wilson et al., 2003]. Using constraints on crustal structure provided by this experiment, we design numerical simulations of lithospheric extension to explore how upper crustal strain localized within pockets of partial melt may lead to mid crustal sub-horizontal shear zones and shallow low angle normal faults. The partial melt within the crust clearly plays a role in determining fault geometry in the Coso area, and the results may have implications for other areas with large rheology variations undergoing rapid extension.

66 Seismological Observations We recorded ~220 Gb of 40 samples per second, three-component seismograms in and around the Coso Geothermal Area from November 1998 to May 2000 [Wilson et al., 2003]; most of the data is available from the IRIS Data Management Center ( With over 150 sites within an area of ~700 km 2, this is one of the densest portable, passive seismic deployments to date (figure 3.1). We examined over 220 high quality teleseismic events to exploit P-to-S conversions within the shallow lithosphere [Burdick and Langston, 1977; Phinney, 1964] in an effort to map crustal variations in seismic impedance. Assuming a seismic wavespeed Figure 3.1 Shaded relief map with geology centered on Coso geothermal area shown with a southwestern United States tectonic sketch to the right. The red box on the tectonic map indicates the location of the study area. The black lines on the map represent the location of three component seismometer arrays used in the seismic imaging study. The transparent red circle shows the horizontal extent of the imaged partial melt region while the blue circles show the location of observed mid crustal seismic anisotropy inferred to be the locations of a sub-horizontal shear zone accommodating strain variations between the upper an lower crust. structure for the crust and upper mantle within the study area, we can project the recorded converted waves back to their loci of generation to create a three dimensional volume of converted wave amplitude that is related to shear modulus variations. This tool is very effective at mapping discontinuities in material properties such as those found at the Moho or between normal crust and regions of partial melt.

67 3-65 An important arrival observed by Wilson et al. [2003] was a high-amplitude negative polarity arrival from near 5 km depth generated from the top of a low wavespeed zone. The low wavespeed zone appears to be confined below the recent rhyolite domes within the modern geothermal area, which is ringed by basaltic flows. Based on forward modeling with synthetic seismograms, the magnitude of the velocity decrease was shown to be ~30%. A velocity contrast of this magnitude suggests the presence of partial melt (figure 3.2) possibly as high as 30% [Wilson et al., 2003]. In addition to the region of shallow partial melt, we also recorded arrivals from the Moho and a mid-crustal interface. The Moho appeared to be flat and near 30 km depth below sea level throughout the study area (figure 3.2) as has been reported for much of the Basin and Range province [e.g. as summarized by Gans, 1987]. Another positive arrival generated near 16 km depth shows considerably more variability. The amplitude decreases and depth increases for this conversion beneath the shallow low wavespeed zone. It appears to be strongest, and clearest just off the edge of the partial melt region (figure 3.2). The close association of this mid-crustal feature with the magma body suggests that they are related. These could be basaltic sills, which would produce seismically isotropic conversions. Alternatively, the mid-crustal converter could be a shear zone, with the magnitude of shear decreasing away from the magma body. We expect a conversion interface at a mid crustal shear zone to be anisotropic. A region of partial melt imbedded in the upper crust would lead to large local viscosity variations, which we expect would localize upper crustal strain within the low viscosity region. The strain localization leads to an increase in horizontal velocities relative to the lower crust

68 3-66 adjacent to the low viscosity region. If the vertical strain gradient and total strain are large enough, accommodation of the relative motion will require the development of a mid crustal shear zone. Shear zones provide a very effective means to create seismic anisotropy within crustal rocks [Godfrey et al., 2000]. Several studies have demonstrated the usefulness of Figure 3.2 Cartoon showing structure of the Coso geothermal field and the possible relationship between the shallow magma body and regional tectonic features. A common conversion point stacked image made from teleseismic earthquake rays illustrates the Moho, a mid crustal discontinuity, and the shallow magma body. The magma body may act as a strain guide for deformation in the mid to upper crust. Coupled with lower crustal flow this area may accommodate a significant amount of strain. crustal conversions in locating crustal seismic anisotropy [Jones and Phinney, 1998; Levin and Park, 1997; McNamara and Owens, 1993]. Anisotropic material above a conversion interface produces variations with back-azimuth of arrival time and conversion amplitude resulting from the directional dependence of elastic properties within anisotropic material. We use synthetic seismograms to demonstrate the presence of seismic anisotropy of the mid-crustal converter to the east and south of the Coso partial melt zone (see figure 3.1 for array location; figure 3.3). Using seismograms from arrays JSH and BPT (Fig. 3.1), we apply a forward modeling approach to match the observed radial and transverse receiver functions in an attempt to determine the seismic structure beneath the arrays. The negative polarity

69 3-67 conversion observed on the JSH radial receiver function just after 1 s indicates the presence of a rapid decrease in wavespeed with depth (low wavespeed zone; Fig. 3.3c; Table 3.1). The associated energy near the same time on the transverse receiver functions indicates the interface dips to the north-northeast towards the region of partial melt (Fig. 3.3c; Table 3.1). The arrival from near 3 s shows large variations in arrival time with back-azimuth. We find the pattern of arrival time variations difficult to match without the dipping interface located at the base of the low speed zone (Fig. 3.3a) or a plunging axis of symmetry within a mid-crustal anisotropic region. Using the same approach for observed receiver functions from array BPT, we see that a combination of a dipping interface and mid-crustal anisotropy (Fig. 3.3d; Table 3.2) provides a satisfactory fit.

70 3-68 Figure 3.3a Comparison of several synthetic models to observed data collected just to the southwest of the partial melt region (JSH on figure 3.1). The left hand column shows radial receiver functions plotted in time and stacked by back-azimuth. The color plot behind shows calculated synthetics for three models (a,b, and c) shown next. The right hand column shows the associated recorded and calculated synthetic transverse receiver functions. Model (a) has a shallow low wavespeed zone overlying a mid crustal layer of anisotropy. Model (b) has no anisotropy but adds a dipping interface on the top of the low velocity zone. The best-fit model, (c), uses both the dipping low velocity layer and mid crustal anisotropy to explain the observations.

71 3-69 Figure 3.3b Comparison of several synthetic models to observed data collected just to the southwest (JSH on figure 3.1) and the east (BPT on figure 3.1) of the partial melt region. Model (d) used to fit the observations from BPT contains a thin layer of anisotropy capped by a dipping interface.

72 3-70 The results from the synthetic modeling show that areas to the southwest and east of the shallow partial melt region contain thin regions of mid-crustal seismic anisotropy (see table 3.1 and 3.2). The anisotropy is inconsistent with a mid-crustal collection of basaltic sills. Instead, a shear zone is our preferred interpretation. We have previously speculated on the kinematics that might cause such a shear zone [Wilson et al., 2003; Fig. 3.2]. Simply put, the shear zone could accommodate the difference between upper crustal extension being focused in the magma chamber and lower crustal thinning being distributed more broadly. Such a difference in strain seems to be required by the absence of much topography on the Moho. In the next section we will test this model by simulating extensional strain in a region with a shallow low viscosity body (partial melt zone) with the goal of reproducing the flat Moho and sub-horizontal shear zone observed seismologically. thickness(m) Density Vp Vs % Vp %Vs Trend Plunge Strike Dip anis. anis Table 3.1 Best synthetic model for the JSH array. Results are shown in figure 3.3. thickness(m) Density Vp Vs % Vp %Vs Trend Plunge Strike Dip anis. anis Table 3.2 Best synthetic model for the BPT array. Results are shown in figure 3.3.Numerical Simulation

73 3-71 We used the BASIL numerical simulation program [England et al., 1985; Houseman and England, 1986] to demonstrate that a crustal viscosity structure exists that would deform in manner that is consistent with our seismic interpretation Initial and Boundary Conditions We choose the simulation parameters to best replicate constraints on the Coso system we have from seismological and other geophysical observations. Wanting the y or vertical axis to be near unity for simplicity, we choose a lithospheric thickness (L 0 =100 km) as the length normalization. The Sierra block moves north-northwest relative to the Basin and Range province at a rate of 12 mm/yr [Dixon et al., 2000; Dixon et al., 1995]. Currently, the Coso area does not accommodate all the relative motion, but we also recognize that extension rates have decreased over the last few million years [Snow and Wernicke, 2000]. Therefore we use a slightly reduced rate of current relative motion to describe the average rate of motion over the last several million years (10 mm/yr). To improve the efficiency of the calculations without sacrificing resolution, we will take advantage of the symmetry of the model and bisect the study area. Thus, the divergence rate of interest becomes the half rate (u 0 = 5 mm/yr;). The remaining normalizations, time and strain rate, are ratios of the length and velocity scales (equation 3.1 and 3.2). Ê (3.1) t = L ˆ 0 Á Ë u t 0 Ê (3.2) e = u ˆ 0 Á Ë L e 0

74 3-72 We choose the boundary conditions so as to best simulate the system of interest while maintaining a numerically stable problem (figure 3.4). In this case we are interested in observing the development of surface and Moho topography so we impose a free-slip condition on the base of the model (our simulated mantle) and the model symmetry axis that bisects the low viscosity crustal region with a stress-free boundary condition on the y=1 plane (free-surface). On the x=0 plane we impose a moving freeslip boundary condition with u x =0. Figure 3.4 Schematic representation of the boundary and initial conditions for the numerical simulations. The cartoon shows the boundary conditions and the location of the low viscosity zone within the model. The viscosity profile for the reference crustal profile (profile 1 in the cartoon) adjacent to the low viscosity zone is shown in green with the red line representing the viscosity profile through the low viscosity zone (profile 2). We use a stratified non- Newtonian viscosity crust with stress exponent (n) equal to 3 with an embedded low viscosity region. Figure 3.5 Plots showing Horizontal and Vertical rate of increase in the non-dimensional model width at 80 ky. Dimensional units are in mm/yr. The low viscosity zone (marked by the transparent white box) perturbs the velocity field by concentrating strain thereby creating a vertical gradient in horizontal velocity. Strain localization also induces relative vertical motion as material from above and below the low viscosity zone gets pulled in to replace the material removed through the strain localization process. The relative motion is indicated by the perturbation in the vertical velocity field beneath the low viscosity region. The vertical gradient in horizontal velocity may lead to initiation of a mid-

75 3-73 These viscosity profiles were based on previous work on lower crustal viscosity within the Basin and Range specifically beneath metamorphic core complexes [Block and Royden, 1990]. Assuming the maximum shear strain will occur at the depth where the largest viscosity contrasts occur we suggest the base of the low viscosity zone should be at the depth of the observed seismic anisotropy. Therefore, the low viscosity region is 10 km (0.1 dimensionless distance) thick Results The horizontal velocity field (figure 3.5) shows a greater horizontal velocity adjacent to the low viscosity zone than above or below; vertical gradients in horizontal velocity decrease with distance from the low-viscosity body. The entire area subsides, as is expected of an area undergoing extension. The largest gradient in vertical velocity occurs adjacent to the low viscosity region, but slightly above and below, indicating the variations could induce topography on both the surface and the Moho. Lateral changes in the vertical component of velocity require one section to be moving up relative to another. An increased extension rates above the low viscosity region is plausible considering the strain localization occurring immediately below. Also, we should expect injection of magma into the crust throughout this process. The additional material will help to minimize topography created on either the Moho or surface due to extension. Decreasing the viscosity of the lower crust would also reduce the amount of Moho topography because it would be easier for surrounding crust to replace any material drawn into the low viscosity region because of the strain localization.

76 3-74 The location of maximum shear strain in the simulation corresponds with the maximum gradients in vertical and horizontal velocity. In the model, the locations of maximum shear strain extend outward from the upper and lower corners of the low viscosity region. The extension of this region of high shear strain would enhance the development of sub-horizontal shear zones in the middle crust. The seismological observations support the presence of a shear zone near 16 km depth adjacent to the upper crustal low velocity body. We observed no perturbation in Moho depth for rays that pierced the lower crust beneath the zone of shallow partial melt. If the partial melt By analogy with our numerical experiments, we expect that the shear zone should correspond with the base of the low-viscosity body. A low viscosity zone greater than 10 km thickness would force the maximum shear strain into the lower crust. We also discard models with a thin low viscosity body because the magnitude of shear strain produced by the model (assuming viscosity contrasts we use are accurate to within several orders of magnitude) would be smaller than that of the 10 km thick low-viscosity body and may not be of sufficient magnitude to require a shear zone.

77 3-75 Figure 3.6 Velocity arrows plotted over sub-horizontal shear strain for a model with a 10 km thick low viscosity zone overlying a low viscosity lower crust. The velocity arrows show the relative motion of the lower crust as it fills in for material removed by the increased extension rates above within the low viscosity region. Shear strain concentrates on the corners of the upper crustal low viscosity region as well as within the low viscosity lower crust. If the low viscosity regions within the upper and lower crust were to become connected by increasing the thickness of the either body the areas of high shear strain would become connected and a crustal scale low angle shear zone could develop. Positive and negative shear strain indicate top to the right and top to the left relative motion, respectively Conclusions A model with a 10 km thick low viscosity zone embedded in an exponentially decreasing viscosity crust is consistent with the seismic observations. This model predicts peak shear strains in the middle crust at then same point where we observe a significant magnitude of seismic anisotropy. Shear stress induced by this model may also initiate sub-horizontal slip on extensional faults leading to core complex development.

78 3-76 The results indicate extension of a local low viscosity upper crust underlain by a regionally low viscosity lower crust could lead to a crustal scale shallow dipping shear zone. From these observations, we introduce a new mechanism that may initiate core complex development in the presence of shallow, localized partial melt. In this model, motion in the upper and lower crust become decoupled across a mid-crustal shear zone with maximum displacement at the top and bottom of the low viscosity zone and decreasing with distance from the body. Normal faults dipping towards the low viscosity body at angles as low as 20 accommodate extension above the low viscosity body in the upper 5 km of the crust. A local increase in extension rates in the upper crust would lead to rapid unroofing and cooling of the partial melt, reducing the viscosity contrast and possibly returning fault dips to higher angles. Further intrusion of mantle magmas plays two important roles. First, multiple intrusions replace mass removed by the extension process, helping to maintain a nearly flat Moho deflected by the extension above. Second, continued intrusion maintains the viscosity contrast necessary to produce low angle normal faults and allow continued core complex development.

79 Evidence for Distributed Lower Crustal Deformation Within a Continental Strike-Slip Fault Zone: Marlborough Fault System, South Island, New Zealand Abstract. Converted phases from teleseisms recorded by a passive seismic array spanning the Marlborough fault system, South Island, New Zealand show a continuous, unbroken Moho beneath the two northernmost faults of the fault system, which suggests that distributed, ductile deformation, not slip on a narrow vertical fault, accommodates lower crustal strain. Beneath the northernmost fault, the Wairau fault (~450 km offset), the Moho dips between 25 to 30 degrees to the southeast from a depth of ~26 km northwest of the fault to a depth of~34 km southeast of the Wairau fault. Farther to the southeast, the Ps conversions from the Moho continue under the Awatere fault with a constant amplitude and depth of ~34 km. Mid-crustal conversions extend beneath the Awatere fault at 10, 16, and 26 km depth, but their continuation beneath the Wairau is not clear. Images derived using common conversion point stacking schemes lose coherence and resolution in the presence of either large, lateral variations in seismic wavespeed or interface topography with a characteristic wavelength similar to the dimensions of the smallest bin size of the stacking algorithm. By constructing synthetic converted-wave images from models with either a step in the Moho or a dip, using approximately the same station-event geometry as the processed data set, we test the possibility that our image is consistent with an offset of the Moho. The synthetic image produced indicates station-event geometries similar to ours would recover either the offset or dipping Moho.

80 4-78 The dipping Moho model explains our results, but that with a step does not. We also investigate the effect of differences in seismic wavespeed structure across the Wairau Fault and find reasonable wavespeed variations do not affect the lateral continuity of the Moho. The observation of a continuous but dipping Moho under the Wairau Fault and its ~400 km of displacement implies that surface slip is distributed over a broad (60 km wide) region of weak, ductile lower crust Introduction Shortly after the introduction of plate tectonics, its limitations within continental settings became apparent. Deformation between oceanic plates remains confined to areas immediately adjacent to the boundary but deformation observed at continental boundaries may stretch across regions hundreds and even thousand of kilometers wide [England and Jackson, 1989; Isacks et al., 1968]. Some suggest that in continents a weak continental crust only blurs coherent block deformation occurring in the mantle [Replumaz and Tapponnier, 2003; Tapponnier et al., 1982]. If instead, the mantle were to deform in a distributed, ductile manner the change in strain distribution between the upper crust and mantle would require distributed deformation within the lower crust to accommodate the relative motion. The distinction in differing views remains the style by which the lower crust and mantle deform. Is deformation in the lower crust and upper mantle concentrated along narrow shear zones or distributed over broad areas? We can discriminate between opposing theories of lower crustal and upper mantle deformation by observing Moho topography beneath large offset strike-slip faults. If the lower crust and upper mantle were to deform as blocks, an offset in the Moho beneath the fault would be expected as

81 4-79 two different crustal blocks of differing thickness become juxtaposed. Conversely, if the lower crust deforms in a distributed fashion, Moho topography should be continuous with large variations in crustal thickness accommodated by a dip not a step in the Moho beneath the fault. To address this question, we recorded P-to-S converted phases from teleseismic earthquakes to image the lithospheric structure of the Marlborough strike-slip fault system located at the north end of the South Island of New Zealand. Using seismic images of the lithospheric structure across the northern two faults, we test for Moho offsets beneath the strike-slip faults and look for indications of pervasive shear based on the presence of seismic anisotropy. Converted waves recorded by large arrays of seismometers have been used to map regional scale changes in crustal thickness [Crosswhite and Humphreys, 2003; Sheehan et al., 1995; Zandt and Furlong, 1982; Zhu, 2000; 2002; Zhu and Kanamori, 2000; Zhu et al., 1995] and the location of downgoing oceanic slabs beneath modern subduction zones [Bostock et al., 2002; Gilbert et al., 2001]. Our work follows that of Wilson et al. [2003] from the Coso geothermal area, where it was demonstrated that a network of small highdensity seismic arrays allows greater resolution of crustal discontinuities than was previously possible with standard passive seismic array deployments. Here, we employ a similar array-network geometry to determine the location of lithospheric seismic discontinuities in the Marlborough fault zone Geologic and Tectonic Setting

82 4-80 The islands of New Zealand mark the emergent area of a low-lying continental landmass between the Pacific and Australian plates (figure 4.1). Slip on the Alpine fault along with subduction along the Hikurangi trench and Puseygur trench accommodate most of the relative plate motion. The transition from subduction to transcurrent motion at the northeastern end of the South Island is accommodated by a system of four major strike-slip faults known as the Marlborough fault system. At ~45 Ma deformation within modern New Zealand began [King, 2000; Sutherland, 1994; Figure 4.1 Image showing seafloor bathymetry surrounding New Zealand which straddles the plate boundary between the Pacific and Australian plates (bathymetry plot created by New Zealand National Institute of Water and Atmospheric Research ). A shallow sea bounds the North and South Islands of New Zealand to the east and northwest due to the buoyant continental landmass on which New Zealand resides. Two subduction zones of opposing polarity lie to the north and south of the South Island. The arrow represents the directions of relative motions between the two plates. Walcott, 1998]. Subduction along some parts of the plate boundary [Sutherland, 1995] and transcurrent motion along the modern Alpine fault began at ~22-23 Ma [King, 2000;

83 4-81 Walcott, 1998]. Restoration of the Junction magnetic anomaly from a location just north of a northern extension of the Alpine fault, the Wairau fault in Marlborough, to a point southeast of the Alpine fault indicates ~460 km displacement has occurred along the Alpine fault since transcurrent motion began [Sutherland, 1999]. Between ~6-7 Ma, relative motion changed, so that convergence across the modern Alpine fault increased from about 1 mm/yr to 10 mm/yr [Sutherland, 1999; Walcott, 1998]. The change in relative plate Figure 4.2 Detailed map of the northern end of the South Island of New Zealand showing the topography and fault traces associated with the Marlborough fault zone. The dots show the location of highdensity arrays of short period seismometers and stars single broadband stations used to fill sensor gaps between high-density arrays. Assuming major variations in lithospheric stratigraphy occur in directions perpendicular to the strike of the Wairau fault, we rotate the common conversion point stacking grid into a fault parallel fault perpendicular system centered on the dark line. Rays with piercing points outside of the grid are projected onto the stacking grid boundary. motion between ~6-7 Ma slightly postdated the initiation of the Marlborough fault

84 4-82 system, but most likely they are linked [Little and Jones, 1998]. Initially, much of the Pacific-Australian relative motion accommodated within the Marlborough fault system was absorbed on the Wairau fault, which saw ~120 km of displacement in the last 7 My [Little and Jones, 1998]. To the south of the Wairau, the other faults have slipped substantially less, with 7-16 km of motion on the Awatere, <18 km on the Clarence, and ~20 km on the Hope fault [Little and Jones, 1998]. Fault Total Displacement (km) Current Slip Rate (mm/yr) Wairau ~400 (~120 since 7 Ma) 3-5 Awatere ~ Clarence ~ Hope ~ Table 4.1 Table summarizing total displacement [Little and Jones, 1998] and current slip rates [Bourne et al., 1998] on major faults of the Marlborough Fault System. Regional scale geology within our study area is simple. The Permian to early Cretaceous Torlesse terrane, a sequence of continental shelf and slope sediments underlies almost a third of the South Island east of the Alpine fault and its northeast continuation, the Wairau fault. Elevated lower crustal seismic wavespeeds [Davey et al., 1998; Eberhart-Phillips and Reyners, 1997] and the occurrence of lower crustal earthquakes beneath an aseismic middle crust [Reyners and Cowan, 1993] are interpreted as evidence for the presence of oceanic crust underlying much and possibly all of the Torlesse on the South Island. To the northwest of the Wairau fault sits a portion of a Permian age ophiolite suite (Dun Mountain-Maitai Terrane) and a volcaniclastic section (Brook Street Terrane). A high-wavespeed middle to lower crust north of the Wairau fault inferred from local earthquake wavespeed tomography suggests the ophiolite suite observed on the surface extends into at least the middle crust [Eberhart-Phillips and Reyners, 1997]. In contrast to these rocks, to the west of the Brook Street Terrane, lies a

85 4-83 belt of Cretaceous and Devonian-Carboniferous granite indicated by a clear decrease in seismic wavespeed [Eberhart-Phillips and Reyners, 1997]. The clearly very different lithology observed at the surface across the Wairau fault should remain so if the faults penetrate into the mantle Data and Method Array Design and Placement Between December 2000 and May 2002, a network of 5 seismic arrays, each consisting of 8-9 short period 3-component seismometers (2 Hz free period: Mark Products L22) and, when available, 1 broad-band sensor (Guralp CMG3-ESP) recorded nearly 1 terabyte of data (figure 4.2). (Note: We use the term seismic array to mean a group of regularly spaced seismometers while seismic network indicates a group of irregularly spaced seismometers. Therefore, a network of seismic arrays indicates a set of irregularly spaced groups of evenly spaced seismometers.) The data was archived at the IRIS data management center using Antelope software. Within each array the sensors were spaced 500 m apart and arranged into two orthogonal lines. Other broadband sensors were distributed between short period arrays to fill gaps where no arrays could be deployed because of logistical difficulties. To exploit roads available in the Rainbow ski area 10 km to the southeast of the Wairau, the RSF array occupied an unevenly spaced grid instead of orthogonal legs. A Reftek 72A-07 or 72A-08 data acquisition system (DAS) recorded ground velocity from each station at 40 samples per second and 32 bits per sample. Timing was corrected from DAS-recorded GPS log files using PASSCAL time correction software.

86 4-84 Because our goal was to image a large portion of the crust in the Marlborough fault system and because we assumed that variations were minimal parallel to the mean trend of the Wairau fault, we deployed arrays in a roughly linear network trending almost north to south. Arrays were placed on bedrock insofar as was possible while other arrays were deployed over an unknown and variable amount of sediment within basins or river valleys Event Selection and Data Preparation We chose 142 teleseismic events, ranging from 29 to 95 degrees distant, based on visual inspection of the coherence of the direct P arrival on all three components with some consideration given to the overall back-azimuthal Figure 4.3 Map of the 142 teleseismic earthquakes used in this study. The blue squares show the earthquake location with the red dot representing the location of the study. The red contours are spaced at 30 degree distance intervals from the study area. distribution of events (figure 4.3). Shifting and stacking the recorded waveforms according to predicted P phase velocity helped to expose timing errors. A station obviously shifted incorrectly after moveout correction due to waveform timing errors (> 0.1 s) was removed for that event. To include broadband traces within the final

87 4-85 processing, we deconvolved the theoretical instrument response from each broadband trace and convolved the theoretical response of a Mark Products L-22 seismometer (free period is 2 s, damping parameter is 0.7). We then formed beams of ground velocity for each event-array pair by time shifting seismograms by the predicted P-wave delay relative to the predicted P arrival time at a reference station location and stacked them [Abers, 1998; Langston and Hammer, 2001; McNamara and Owens, 1993]. We calculated the slowness for each event using the IASPEI91 wavespeed model [Kennett and Engdahl, 1991] and locations reported in the Preliminary Determination of Epicenters (PDE catalog) (as archived at A surface P wavespeed of 5.5 km/s was used to correct the predicted arrival times for delays due to variations in topography between stations. Because both horizontal components were not always available for all events and stations, we stacked horizontal components separately to form beams before rotating in to the radial-transverse system. Before further processing, the beams were band-pass filtered in the frequency range of Hz to remove some Figure 4.4 Diagram showing impinging teleseismic P wave and examples of the forward and multiply scattered conversions produced. The cartoon seismograms show relative arrival time for some of the arrivals above.

88 4-86 portions of recorded microseismic noise and high-frequency background noise present at some arrays Receiver Function Calculation and Compositing Receiver function analysis seeks to enhance Ps converted phases, S waves converted at interfaces beneath a receiver from incident P phases (figure 4.4) [Burdick and Langston, 1977]. We assume the vertical component approximates the earthquake source function and the radial component is a convolution between the source function and the earth s response (discontinuities in seismic wavespeed such as the Moho). Deconvolution of the vertical from the radial component seismogram allows calculation of the receiver function, which represents arrivals of various P d s conversions after the direct P wave. The ratio of the amplitude of the converted arrival to that of the incident P wave depends on the impedance contrast across an interface and the incidence angle of the incoming wave [Langston, 1977]. Particle motion of P d s conversions from flat lying interfaces separating layers with isotropic wavespeeds will be radially polarized; therefore the amplitude of a converted phase from such a discontinuity should be constant for all back-azimuths with the same incidence angle. If a plane wave encounters an anisotropic medium, the impedance contrast and elastic properties of the medium will vary with back-azimuth and incidence angle and therefore the amplitude and arrival time of the converted wave should vary with vector slowness. For all azimuths not aligned with the symmetry axes of the anisotropic medium the converted shear wave will be split into orthogonally polarized quasi-s waves traveling at two different wavespeeds. The particle motions of the converted shear waves

89 4-87 will be aligned with the symmetry axes of the anisotropic medium. Therefore, both the transverse and radial components will both contain ground motion related to the split converted S waves. The arrival time and amplitude of the quasi-s phases present on the receiver function depends on the azimuth of the incident P wave, magnitude of anisotropy within the layer, and thickness of the anisotropic layer. The orientation of the anisotropy symmetry axis governs the back-azimuthal variations observed on processed receiver functions. For sufficiently long transit times within the anisotropic medium, a clear time separation between the two S phases will be observable. However, for most crustal conversions traversing anisotropic media, the travel time within the anisotropic media will not be enough to separate the phases. Radial and transverse seismograms recording split crustal conversions often show a difficult to interpret combination of the two quasi-s phases. Signal-generated noise from topographic scattering and short wavelength subsurface heterogeneities lowers the signal to noise ratio for teleseismic arrivals with periods of ~1s more than for longer periods. Recognizing this fact, most receiver function studies low pass filter the recorded seismograms prior to processing in an attempt to remove unwanted scattered energy, but often do so at the expense of spatial resolution. Not wanting to degrade spatial resolution and assuming the noise is not coherent from stations to station, we attempt to remove unwanted energy not by filtering but by stacking data recorded at the high-density arrays before processing. Coherent noise traveling with slowness similar to that of the direct arrival (e.g. some types of topographic scattering or scattering from near vertical planes) may still contaminate the signal, destabilizing the deconvolution, and undermining the reliability of the receiver function amplitudes. In

90 4-88 our experience, only in extreme cases [Jones and Phinney, 1998] does the anisotropy reach sufficient magnitude to create clear back-azimuthal variations in the amplitude of the radial conversion in the presence of noise; more often, anisotropy can be more reliably identified based on back-azimuthal variations in converted phase arrival time on radial receiver functions and back-azimuthal patterns of polarity reversals on transverse receiver functions. We deconvolved stacked vertical components seismograms from the stacked radial and transverse components seismograms to produces radial and transverse receiver functions. We use an iterative deconvolution method relying on a cross correlation of the radial and vertical component seismograms [Ligorria and Ammon, 1999]. The convolution of the receiver function with the vertical seismogram is compared to the appropriate filtered radial seismogram to calculate the variance. We use receiver functions that reduced the variance by greater than 70% in the final analysis. Allowing fewer rays by increasing the variance reduction cutoff to 80% significantly reduced the number of useable rays and degraded the final image. Lowering the cutoff below 60% allowed the incorporation of too many noisy traces, which smeared important arrivals and also degraded the image resolution (Chapter 1). To compare and stack events from different distances and back-azimuths, we project converted energy along the predicted P d s raypaths to 200 km depth using a wavespeed structure derived from local seismic tomography [Eberhart-Phillips and Reyners, 1997]. Ray tracing allows the projection of the conversions to a pseudodepth, allowing us to stack conversions from the same locus of conversion recorded at different arrays. Large vertical gradients in the 1-D seismic wavespeed structure used in depth

91 4-89 migration may cause erroneous projection of conversions to depths at the point of the largest gradient. To minimize the possibility that the assumed wavespeed structure could produce erroneous Moho topography, we use a thick, gradational Moho instead of a sharp boundary to decrease the possibility that the imaged Moho depth results from the placement of the Moho boundary in an a priori wavespeed model. In addition to depth migration, we correct receiver function amplitudes to those of a fixed incidence angle to reduce the dependence of conversion amplitude on incidence angle [Jones and Phinney, 1998]. Rescaling of conversion amplitudes prevents bias towards larger incidence angles when stacking over a range of incidence angles and permits interpretation of arrival amplitude as directly proportional to the S-wave impedance contrast across an interface Common Conversion Point Stacking To study the lateral variations in subsurface features and further mitigate contamination from and topographic and near surface scattering, we stack the receiver functions into common conversion point bins (analogous to common midpoint stacking in reflection seismology) after depth migration and amplitude correction. The geographic stacking procedures used here follow those of Dueker and Sheehan [1998] with modifications. Before common conversion point stacking, array elevation corrections were made before common conversion point stacking based on the predicted angle of incidence, the reference station elevation, and an assumed surface wavespeed of 5.5 km/s. P d s arrival piercing points, determined using the ray tracing techniques described above, are binned using a 3-D grid of sample points spaced 5 km apart in a Wairau fault perpendicular direction, 40 km apart in a Wairau fault parallel direction, and 0.25 km

92 4-90 apart in depth within a volume 40 km (along strike), 100 km (perpendicular to strike), and 120 km (vertical). The piercing points within a 40 x 5 km 2 area centered on each sample point are stacked at that point. Changing the size of the bins significantly influenced our results with the chosen bin size chosen qualitatively based on image coherence. Bootstrap resampling of receiver function rays traversing a common conversion point bin with replacement within individual common conversion point bins, performed in order to estimate the error associated with that stacking bin [Efron and Tibshirani, 1986], helps assure that a few aberrant receiver functions do not control the results. We calculated one hundred bootstrap realizations of each stack and present their mean and standard deviation for each bin on common conversion point stacked images to demonstrate the reliability of the results for each bin. Images produced by common conversion point stacking may suffer from effects of wavespeed variations that could influence interpretation if unrecognized. All depth projections of receiver functions are calculated by tracing predicted rays through a onedimensional wavespeed structure as discussed previously. Lateral variations in seismic wavespeed will cause features to be mapped erroneously deep below low S-wavespeed bodies and too shallow beneath high S-wavespeed bodies. In some cases, converted phases from a single locus of conversion recorded at different arrays will not stack at the same depth, blurring the feature on the common conversion point stacked image. Recognition of this effect can assist in identification of substantial wavespeed variations. But unless the variations are quite large over a short distance, the effect on interpretation is minor.

93 4-91 Depth (km) 1-D speed model (km/s) Coarse 2-D speed model (km/s) South North Vp Vs Vp Vs Vp Vs Table 4.2 Table summarizing wavespeed models used for receiver function depth migration. For the coarse two-dimensional model we mapped all portions of each ray path with the appropriate wavespeed model depending on where the ray path was in relation to the Wairau fault. All back-azimuth and mean receiver function stacks were depth migrated using the coarse two-dimensional model Seismological Observations Mean receiver functions and back-azimuthal stacks To demonstrate the major crustal features of the Marlborough fault system, we show Figure 4.5 Waveforms from a teleseismic event recorded on an array just to the North of the Wairau Fault. The darker traces towards the top show the vertical components from each station. The lighter traces towards the bottom show the beamed seismograms from all three components.

94 4-92 the mean and back-azimuthally binned depth projected radial and transverse receiver functions from three regions within the study area: northwest of the Wairau fault, between the Awatere and Wairau fault, and southeast of the Awatere fault. Stacking all depth projected receiver functions from each region allows us to remove short wavelength variability contained in the data and illuminate the large-scale, continuous features found within each individual region. Viewing the mean receiver function in combination with the back-azimuthally binned receiver functions permits the identification of important arrivals and the nature of these conversions Northwest of the Wairau Fault The mean receiver function from the area northwest of the Wairau fault contains a series of closely spaced, positive converted phases from 12 and 16 km depth and a separate arrival from ~24 km depth (figure 4.6a). Based primarily on its amplitude on the mean receiver function and its observed consistency on individual receiver functions, we interpret the conversion from near 24 km depth as the Moho conversion. The amplitude of the positive polarity converted phase from near 16 km depth appears to vary with back-azimuth but arrival time variations are < 0.1 s, suggesting the interface is not dipping but the material surrounding the interface may be seismically anisotropic (figure 4.6b). The 12 km arrival apparent on the mean receiver function shows considerable variability in amplitude on the back-azimuth stacks, including a reversal in polarity for on rays from eastern back-azimuths. A single reversal in polarity of a radial conversion over a full range of back-azimuths can be generated by a steeply dipping interface (>~25 degrees), by a point scatterer. A large variation in arrival time with back-azimuth would

95 4-93 be expected for a point scatterer, therefore it seems more plausible that the energy is from a dipping interface considering the relatively small arrival time variation on the radial and transverse. The transverse receiver function stacks help us to constrain interpretations of the radial receiver functions (Figure Figure 4.6 Mean radial receiver function (a) with depth projected back-azimuth stacks of (b) radial and (c) transverse receiver functions from rays rays piercing 30 km depth north of the surface trace of the Wairau fault. The Moho conversion is the prominent arrival from 24 km depth. The large arrival from 16 km depth appears to be generated at an anisotropic interface based on amplitude and arrival time patterns on the radial and transverse back-azimuth stacks. The number on the right vertical axis indicates the number of receiver functions stacked within each back-azimuth bin. 4.6c). We expect to observe predictable polarity reversals on the transverse receiver function with changing back-azimuth if the shallow dipping interface and lower crustal seismic anisotropy are indeed present. The polarity of the 12 km depth arrival on the transverse receiver function varies with a pattern consistent with a slightly dipping interface (not the large dip we expected). This observation is puzzling considering the large arrival time variations expected from a steeply dipping interface.

96 4-94 To test the plausibility of the interpretations of the back-azimuth stacks (figure 4.6), we calculate reflectivity based synthetic receiver functions for a model with a shallow steeply dipping interface overlying a hexagonally symmetric anisotropic layer within the lower crust (figure 4.7; table 4.3) [Frederiksen and Bostock, 2000]. Although our constraints on magnitude and direction of anisotropy within the lower crust are weak, the results of the forward modeling procedure require some Figure 4.7 Synthetic radial (a) and transverse (b) receiver functions plotted behind recorded receiver functions for rays arriving to the northwest of the Wairau fault. The variations observed in the radial and transverse receiver functions are best modelled with an anisotropic lower crust beginning near 15 km depth (see table 4.2). amount of anisotropy in the bottom 13 km of the crust to produce the arrival time variations seen on the radial Moho arrival and the rapid polarity reversals over varying back-azimuth seen on the transverse arrivals from both 16 and 24 km depth (figure 4.7; table 4.3). A simple dipping interface or a combination of multiple dipping interfaces will not produce the multiple unevenly distributed polarity reversals observed on the

97 4-95 transverse receiver functions for differing back-azimuths. Shear wave splitting observations indicate the upper mantle fast direction beneath the Marlborough fault system follows the mean trend of the Wairau fault (50 ), which is roughly coincident with our fast axis orientation of 70 [Audoine et al., 2000; Klosko et al., 1999; Molnar et al., 1999]. The synthetic receiver functions are relatively insensitive to the trend of anisotropy, with only slight variations in the character of the synthetic seismograms for trends between 10 and 100. A plunge is not required but does improve the fit to the observations by minimizing the amplitude of later transverse arrivals generated near 20 km depth. thickness(m) Density Vp Vs %P %S Anis. Anis. Strike Dip (kg/m 3 ) (m/s) (m/s) anis. anis. Trend Plunge Table 4.3 Parameters used in synthetic receiver function modeling for rays piercing the base of the crust north of the surface trace of the Wairau fault Between the Wairau and Awatere faults To the southeast of the Wairau fault, the Moho conversion arrives later than to the northwest, indicating a depth of conversion near ~ 30 km, based primarily on common conversion point images (next section) and the mean receiver function (figure 4.8). Assuming no lateral variations in seismic wavespeed, such a difference in Moho depth across the Wairau fault suggests the presence of a step or a steep dip (~25 degrees) in the Moho. The Moho arrival on the back-azimuth stacks shows considerable variation in amplitude and arrival time, but the arrivals do not appear to follow a clear, systematic

98 4-96 pattern. The large stacking area (>30 km normal to the Wairau fault) and the considerable interface topography (~10 km) probably contributes to the incoherence of the Moho arrival on the backazimuth stacks. A large negative polarity arrival underlying the Moho for northern back-azimuths is also consistent with a large southward dip on the Moho. The clearest crustal arrival on the mean receiver function comes from a depth near 20 km. Arrival time Figure 4.8 Mean radial receiver function (a) with backazimuthal depth projected stacks of radial (b) and transverse (c) receiver functions from rays piercing 30 km depth between the surface traces of the Awatere and the Wairau faults. The Moho conversion is the doublet arrival from 30 km depth on the mean radial receiver function. The large arrival from 20 km depth on the mean receiver function shows considerable variation in arrival time, as does the Moho arrival. variations on the back-azimuth stacks indicate this interface dips to the south near 16 km depth and may be smeared to 20 km depth on the mean receiver function stack. Unfortunately, there are no clear patterns in the variation of the converter s radial amplitude that might confirm a dipping interface. As in the last section, we seek to constrain the dip and anisotropy of converters responsible for the amplitude and arrival times observed on the radial receiver functions

99 4-97 using synthetic receiver functions (figure 4.9). Energy on transverse receiver functions from 16 km depth is best fit by a steeply dipping interface (~25 dip). A dip alone cannot explain the total arrival time variation observed on the radial receiver function for both the arrival from 16 km depth and the Moho arrival. A steep dip on the 16 km deep converter would perturb the arrival times enough, but it would also produce a negative polarity conversion not observed on the radial receiver functions. Therefore, combining anisotropy with a dipping interface may best account for the variations in arrival time while maintaining the correct polarity on the radial arrivals (figure 4.9). Sensitivity analysis of the parameters used to calculate the synthetic seismograms indicate Figure 4.9 Synthetic radial (a) and transverse (b) receiver functions plotted behind recorded receiver functions from rays piercing 30 km depth between the surface traces of the Wairau and Awatere fault. The variations observed in the radial and transverse receiver functions are best modelled with a seismically anisotropic lower crust with a plunging axis of symmetry overlain by a dipping interface. The combination of the two significantly reduces the amplitude of the Moho conversion. little variation occurs in the features of the synthetic seismograms for large variations in the magnitude of anisotropy. The strike ( ) and dip (>15 ) of the interface near

100 km depth are reasonably well constrained as is the geometry of the Moho interface (strike: 0 to 50 ; dip angle: >20 ). Z (m) Density Vp Vs %P %S Anis. Anis. Strike Dip (kg/m 3 ) (m/s) (m/s) anis. anis. Trend Plunge Table 4.4 Parameters used in synthetic receiver function modeling for rays piercing the base of the crust between the Wairau and Awatere faults Southeast of the Awatere fault Southeast of the Awatere fault, the character of the receiver functions differs from those observed to the north (figure 4.10). They appear to be highly periodic ( ringy ). The ringy character suggests that the Figure 4.10 Mean radial receiver function (a) with backazimuthal depth projected stacks of radial (b) and transverse (c) receiver functions from rays piercing 30 km depth south of the Awatere fault. The Moho conversion arrives from 34 km depth on the mean radial receiver function. The large arrival from 26 km depth on the mean receiver function and backazimuth stacks is similar in amplitude to the Moho indicating a large lower crustal impedance contrast. The number along the right axis indicates the number of traces within each bin. converted wave energy is reverberating within the upper crust and may be overprinting deeper arrivals. The mean

101 4-99 receiver function shows high amplitude arrivals near 9, 17, 26, and 34 km depth. It is difficult to interpret the arrivals observed in the back-azimuth stacks in terms of dip or anisotropy considering the few directions sampled and the relatively incoherent patterns observed on the transverse back-azimuthal stacks. Although we cannot model dips and anisotropy for this reason, we are reluctant to exclude the possibility of anisotropy south of the Awatere fault. We stack the data according to ray parameter (angle of incidence) to determine if the arrivals are forward scattered conversions or reverberations, as suggested by the character of the receiver functions (figure 4.11). A forward scattered conversion follows a line of increasing Ps-P time for increasing ray parameter (prograde moveout). All of the high-amplitude arrivals fit the predicted moveout of forward scattered arrivals indicating that they are crustal interfaces, not freesurface reverberations (figure 4.11). We interpret the arrival from 34 km depth as the Moho conversion. Figure 4.11 Radial receiver functions with Moho piercing points South of the Awatere stacked by ray parameter. The red lines indicate predicted moveout curves for forward scattered P-to-S conversions from interfaces shown to the right of the line near the top of the plot. All arrivals with considerable amplitude before 4 seconds appear to follow moveout curves of forward scattered arrivals despite the ringy character of the receiver functions. The numbers to the right axis indicate the mean distance for the earthquakes within each stacking bin.

102 Common conversion point stacks The mean receiver function and back-azimuth stacks indicate the importance of a mid crustal arrival as well as significant Moho topography across the Wairau fault. We use the information provided by the back-azimuth stacks to interpret common conversion point stacked images which help in determining the lateral extent and topography of crustal interfaces (figure 4.12). The Moho arrival begins at 24 km depth northwest of the Wairau fault, dips at >25 degrees across the Wairau fault, and becomes flat again 15 km southeast of the fault. Unmistakably, the Moho remains flat at least until it crosses the Awatere fault. At that point, it is not clear if Figure 4.12 Common conversion point stacked cross-sections made with the 133 radial receiver functions with variance reduction greater than 70 %. The top cross section (a) uses velocities from Eberhart-Phillips and Reyners [1998]. In practice, this means ray portions north and south of the Wairau are mapped with different velocity models. To explore the effect of using spatially varying velocity models on observed interface topography we restack the data in (b) using a mean of the velocity models used in (a). There is little change on the geometry of the interfaces. the Moho remains at 34 km depth or it steps to 26 km depth. A low amplitude arrival

103 4-101 from near 34 km depth continues to the southeast while an arrival everywhere equal and sometimes greater in amplitude comes from a depth near 26 km depth. The similar amplitude of the 34 km and the 26 km depth arrival is puzzling considering the large impedance contrast required to equal the contrast between crust and mantle material. It is unlikely that a contrast of this magnitude could reside beneath the crust, in the upper mantle. Therefore, we suggest the 34 km depth arrival to be from the Moho and another strong impedance contrast lies several kilometers above the base of the crust. The arrival from 26 km depth continues across the Awatere fault to the northwest although at reduced amplitude. The arrival then shallows and fades as it reaches the Wairau fault. It is possible the arrival could be related to the same interface at 25 km depth northwest of the Wairau fault but the connection is not clear in the vicinity of the fault. The drastic difference in rock type across the Wairau (ophiolite suite vs. continental shelf/slope sediments) suggests a difference in seismic wavespeed might affect the coherence converted phases across the image. To explore the possibility we restack the data using a mean of the [Eberhart-Phillips and Reyners, 1997] wavespeed model that covers our study area (figure 4.12). If wavespeed variations play a huge role in image coherence, then using the same wavespeed model for all ray paths would lead to image degradation due to misprojection of conversion points. The images demonstrate that the use of plausible but different wavespeed models for ray tracing does not destroy the major features of the image. However, the mean wavespeed model does yield a different absolute depth of some conversion interfaces. Most notably, it flattens the dip on the

104 4-102 Moho because of higher Vp/Vs ratios north of the Wairau fault than south of it [Eberhart-Phillips and Reyners, 1997] Discussion Moho topography and resolution of teleseismic imaging Undoubtedly, the most important observation from the results shown here is evidence for a continuous, unbroken Moho dipping beneath the Wairau fault. The observed Moho geometry precludes the possibility that the Wairau fault extends into the upper mantle as a discrete fault because, although the crust on either side of the Wairau fault is different in character Figure 4.13 Synthetic CCP stacks from seismograms calculated for a model with a dipping interface (a and b) and a discontinuous interface (c and d). The top diagrams from each panel (a and c) shows the CCP cross-section for all plane waves from both directions with angles of incidence spaced at 3 increments from 9 to 30. Stations are spaced 500 m apart. The bottom plots on each panel (b and d) shows the results using the station-event geometry from our study for each of the dip and step model. and thickness, the two sides are still connected at the base of the crust. This implies that

105 4-103 somewhere between the surface trace of the Wairau fault and ~30 km beneath it, the deformation accommodated by the fault becomes distributed over a wider region. The presence of short wavelength heterogeneity (e.g., velocity contrasts across the Wairau fault) or interface topography (e.g. a Moho step) may contaminate the image in due to the coarse nature of the stacking procedure coupled with the limited ray parameter and back-azimuth coverage. To address uncertainty about the resolution of our common conversion point stacking scheme using the recorded data set, we calculate synthetic receiver functions with a finite difference algorithm for two models of Moho topography, one containing a dipping Moho (figure 4.13) like the one interpreted from Figure 4.14 Synthetic CCP stacks from seismograms calculated for a model with a step in the interface (Moho). our image (figure 4.12) and the other with an offset in the Moho across the Wairau fault (figure 4.14). The station-event geometry chosen for the finite-difference modeling mimics that recorded by the experiment.

106 4-104 The results show a step of the magnitude indicated by the Moho offset observed in the back-azimuth stacks should be clearly imaged, without ambiguity by a ray set with geometry similar to our recorded dataset. For the image of the dipping Moho model, the lateral averaging blurs the image as the dipping section becomes broadened and appears as medium-amplitude steps centered on the dipping section. The blurring observed on the synthetic image resembles that on the common conversion point image created from actual recorded data (figure 4.12) further supporting the presence of a step beneath the Wairau. Results from finite difference modeling indicate the interpretation of a continuous, dipping Moho based on the common conversion point image is valid. This precludes the presence of a vertical fault extending through the crust into the mantle and supports accommodation of Wairau motion through distributed lower crustal deformation Mid crustal conversions Implications for strain distribution mechanisms and the rheology of the lower crust If the Wairau and Awatere faults are not discrete faults through the crust, we reexamine features within the crust to understand the transition from brittle deformation in the upper crust to distributed deformation below. Where does the transition from localized (brittle) to distributed (ductile) deformation occur?

107 4-105 The presence of >10 km thick sections of anisotropic lower crust adjacent to the Wairau and Awatere fault requires distributed deformation within the lower crust. Straining crustal rocks is a very efficient mechanism to induce seismic anisotropy [Godfrey et al., 2000; 2002]. Therefore the presence or absence of seismic anisotropy within the crust places constraints on the strain history of that crustal section (figure 4.15). Based on the assumption that the presence of anisotropy within a section of crust indicates that material has been strained, we propose that the transition from concentrated to distributed deformation occurs at ~12-16 km depth northwest of the Wairau fault and ~16 km depth southeast of the Wairau fault. Lack of data to the southeast and northwest does not allow us to constrain the lateral extent of the crustal anisotropy. Figure 4.15 Cross sections illustrating possible vertical variation in displacement (d) across strike-slip fault systems with total displacement D. The top panel (A) shows distributed deformation in the mantle, while the bottom panel (B) shows a model with highly localized deformation in the mantle. Areas with predominantly subhorizontal schistosity (and presumably sub-horizontal planar seismic anisotropy) indicated by horizontal ruling; in all cases, for high strains lineations in strained rocks will be normal to the section. Thinner lines follow surfaces of constant displacement (cartoon created by C. Jones). If the Wairau fault were to continue through the crust as a discrete fault zone, the crust surrounding the discrete fault zone would not be strained and no seismic anisotropy would be induced (although the crust may have acquired anisotropy from prior strain). Instead, all seismic anisotropy would be concentrated within the vertical shear zone and

108 4-106 would not be detectable with the low frequency teleseismic energy used in our experiment. We observe seismic anisotropy of similar magnitude on both sides of the Wairau indicating that strain accommodation in the lower crust occurs over a broad region Reconciliation of CCP image with gravity observations For Airy isostasy, an 8 km difference in crustal thickness across the Wairau Figure 4.16 Bouguer gravity anomaly with calculated profile. The gradient across the Wairau fault requires the presence of a shallow high-density body to the northeast of the Wairau fault. Based on the variation in crustal thickness across the Wairau fault, a ~160 mgal anomaly should be expected. We suggest the presence of high-density partially eclogitized lower crust to the southeast of the profile minimizes the amplitude of the Bouguer anomaly. fault implies Bouguer gravity anomaly possibly as high as 160 mgal. The observed Bouguer anomaly is actually much smaller (~45 mgal; figure 4.16). A Bouguer anomaly smaller in magnitude than that predicted based on Moho topography indicates either a mass excess southeast of the Wairau or mass deficit to the northwest of the Wairau, or both. Based on variations in surface geology across the Wairau fault, differences in crustal density should be expected. However, the high-density mafic to ultramafic material found northwest of the Wairau fault makes a mass deficit in that area unlikely, unless there was wholesale serpintinization of the crust.

109 4-107 As stated before, two lines of evidence have been used to infer the presence of a body beneath the Torlesse on the South Island with elevated seismic wavespeed and strength. If the crust inferred to reside beneath the Torlesse to the southeast also underlies the Torlesse here, this may provide the extra mass necessary to reduce the magnitude of the Bouguer anomaly to the south. Assuming the conversion interface at 26 km depth from the common conversion point cross sections represents the top of an 8 km thick high density body, we show a gravity model that fits the observed Bouguer anomaly. A body of this thickness including uncertainties would require a density decrease of /-0.03 g/cc relative to the adjacent mantle north of the Wairau fault to reduce the size of the predicted anomaly (figure 4.16). If the body were mafic to ultramafic in composition we should expect the impedance contrast between the middle crust and the anomalous body to be similar in magnitude to the impedance contrast between the oceanic crust and mantle. The conversions from both the top and the base of the proposed oceanic crustal block are indeed similar in magnitude as seen on both the back-azimuth stacks as well as the common conversion point stacked sections. We explain the near mantle densities of the lower crustal block required by the gravity modeling by a reaction of basalt to eclogite within the block. Unfortunately, the gravity profile is insensitive to the dip on the Moho. A shallow high-density body is needed to fit the gradient across the Wairau fault (figure 4.16). The high-density ophiolite complex found near the Wairau fault would most likely alter the local gradient of the Bouguer anomaly. There are other reasonable models, but we offer the presence of the mafic to ultramafic material north of the Wairau fault because it crops out in this region.

110 Conclusions We address the nature of continental deformation and the strength of the continental lithosphere using observations from a teleseismic imaging experiment. The lower crust in the northern part of the Marlborough fault zone, on the South Island of New Zealand, deforms in a distributed fashion based on observations of a clear continuous, undisrupted Moho throughout the study area despite the substantial topography (~8 km) observed at the interface (figure 4.17). In addition, we suggest the middle crust on either side of the Wairau has a thick lower crustal layer (10-13 km) of seismic anisotropy with a substantial magnitude (~15-20%). This may be the top of the transition to distributed deformation below and most likely represents a thin, sub-horizontal shear zone that redistributes strain from the vertical strike-slip fault to the surrounding crust. The 8 km change in crustal thickness and ~50 mgal Bouguer anomaly across the Wairau fault, requires the presence of a high-density lower crust south of the Awatere consistent with an underthrust sliver of oceanic crust that has partially metamorphosed to eclogite. The material property contrast between continental shelf/slope metasediments above and the oceanic crust below could allow the concentration of shear at the 26 km interface, making seismic anisotropy undetectable even though the interface may accommodate a substantial amount of strain. In summary, we suggest a model of a sub horizontal shear zone overlying a ductilely deforming lower crust accommodating relative motion across Wairau and Awatere faults. The presence of the ductilely deforming crust suggests a decrease in strength in the lower crust in the Marlborough fault system, supporting lithospheric strength profiles

111 4-109 previously suggested by others and precluding the possibility that thin vertical faults extend into the mantle in this portion of the Marlborough fault system. Figure 4.17 Interpreted common conversion point image showing the location of important features. Gravity modeling and a high amplitude lower crustal conversion in the southeast support The presence of a high-density lower crustal block. The observation of a dipping Moho connecting crustal columns of different character and thickness with the overlying lower crustal seismic anisotropy indicate strain becomes distributed below ~16 km depth in the central and northwestern sections of the study area. The small white circles represent local earthquakes located by the New Zealand Institute of Geological and Nuclear Sciences. Acknowledgements. This research was conducted with the support of NSF grant to C. H. Jones and A. F. Sheehan and to P. Molnar. Acquisition of the seismograms would have been impossible without help from the New Zealand Department of Conservation, Allistair, Mike and Jock Nichols, Allistair Martin, Weyerhauser, and Molesworth station. Shaun Bilham, Noah Daniels, Chas Offutt, Odette Singleton, Nick Cozens, Suzannah Toulumin, and Michelle Salmon provided needed assistance in the field. Thanks to Oliver Boyd for use of and help with the finite difference modeling programs. Thanks to all St. Arnaud residents that dealt with the periodic American invasion into their quiet village.