Intraplate Earthquakes, Regional Stress, and Fault Mechanics in the Central and Eastern Us and Southeastern Canada

Home , Charlevoix Seismic Zone, Intraplate earthquake

GEOMECHANICAL ANALYSIS OF INTRAPLATE EARTHQUAKES AND EARTHQUAKES INDUCED DURING STIMULATION OF LOW PERMEABILITY GAS RESERVOIRS

SRB Volume 130

A DISSERTATION SUBMITTED TO THE DEPARTMENT OF GEOPHYSICS AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Owen Hurd August 2012

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

______

Mark D. Zoback (Principal Advisor)

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

______

Norman Sleep

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

______

Gregory Beroza

I certify that I have read this dissertation and that in my opinion it is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.

______

Anthony Kovscek

Approved for the University Committee on Graduate Studies.

iii

ABSTRACT

This thesis applies the principals of geomechanics to address a breadth of fundamental scientific questions regarding rock properties near major active faults, physical mechanisms driving intraplate earthquake activity, and how induced earthquakes in low-permeability natural gas reservoirs provide insight into reservoir deformation during hydraulic fracture stimulations. Through the chapters within, the thesis demonstrates that geomechanical principles can answer fundamental questions regarding the above study areas, the conclusions of which are relevant and applicable to a variety of current scientific and industrial endeavors.

The first part of this thesis applies a shear-wave splitting analysis to examine physical controls on crustal anisotropy mechanisms on and near the North Anatolian Fault in northwest Turkey. The analysis indicates that stress-controlled mechanisms appear to prevail at distances greater than 1 km from the main fault trace while structure-controlled mechanisms are observed within 1 km of the main fault trace. In addition, significant anisotropic intervals appear to be confined to the upper 8km of the crust. The shear-wave splitting analysis indicates heterogeneous physical properties between the main fault zone and surrounding crust, and that it is possible to distinguish between the controlling anisotropy mechanisms with a shear-wave splitting analysis.

The second part of this thesis applies fundamental geomechanical principals to examine the compatibility of shear slip on intraplate faults in the central and eastern United States and southeastern Canada with frictional faulting theory in the context of the regional tectonic stress field. The application demonstrates that maximum horizontal stress (SHmax) orientations consistently trend NE-SW across the study area, although local variations are also present. In addition, horizontal principal stress magnitudes become increasingly compressive with respect to the vertical stress moving from the central U.S. toward the northeastern U.S. and southeastern Canada, which is manifested by predominantly strike-slip/normal and thrust faulting earthquake focal mechanisms in the respective regions. Finally, the application demonstrates that shear slip on the vast majority of intraplate fault planes is well- described by the Mohr-Coulomb failure criterion in the current stress field assuming hydrostatic pore pressure in the upper crust and laboratory-determined coefficients of fault friction. In particular, shear slip on many fault planes within the New Madrid seismic zone is highly consistent with a uniform, ENE-WSW stress field and does not require anomalous pore pressure or fault strength. These observations provide fundamental insight into the nature of intraplate earthquake activity within the central and eastern U.S. and southeastern Canada, and could be of practical use in evaluating earthquake hazards in these regions.

The final part of this thesis applies an integrated, geomechanical and seismological analysis to evaluate the response of low permeability natural gas reservoirs to hydraulic fracturing stimulations in two case studies in western Canada. The reservoirs in both case studies are characterized by a strike-slip stress regime with

SHmax oriented approximately NE-SW. b-values calculated from induced microseismic event populations range between 1.0 and 2.5, illustrating a relative abundance of small to large magnitude events compared to naturally-occurring earthquake populations (b ≈ 1). In addition, highly similar microseismic event seismograms are observed in event populations, suggesting some events may represent repeated slip on small faults during hydraulic fracture stimulation. The double-difference relative earthquake location technique is applied in both case studies to improve microseismic event

vi hypocenter locations. While the technique does appear to reduce scatter in hypocenter locations, it is prone to producing hypocenter location artifacts in single- or dual- monitoring array deployments. However, the technique does appear to yield accurate event hypocenter locations when three or more well-distributed monitoring arrays are available, suggesting the technique could be a viable tool to improve microseismic event hypocenter locations when suitable monitoring array configurations exist.

vii

ACKNOWLEDGEMENTS

I am deeply indebted to a great many people during my graduate studies at Stanford, and could readily fill an inordinate amount of space on deserving acknowledgements alone. However, for the sake of brevity, and the fact 200 pages of detailed science lay ahead, I will try my best to extend my gratitude in reasonable space. I must begin with my advisor, Mark Zoback, who provided me the incredible opportunity to study Geophysics at Stanford. In addition to the immeasurable quantity (and quality) of science I learned from Mark, he taught me what it takes to be, and how to carry myself as, a successful research scientist. Mark, thank you for your tremendous patience, confidence, guidance, and friendship over the last five years as well as your unheralded understanding that we are humans first and graduate students second. It was very much noted and appreciated!

I also thank the remaining members of my dissertation committee, Professors Greg Beroza, Tony Kovscek, and Norm Sleep, for their continual support, enlightening discussions, and encouragement. I will greatly miss being able to walk into your offices whenever I have a question. Thanks also to Associate Professor Tapan Mukerji for serving as my committee chair and as a valuable source of knowledge and assistance during my time at Stanford.

viii

Much gratitude is also due to my undergraduate advisors, Professor (Emeritus) William (Bill) Dickinson and Professor George Zandt, at the University of Arizona who provided incredible research opportunities and served as unparalleled role models for a young scientist just entering the world of research. Without your influence, patience, and guidance, I would have never made it this far, or had anywhere near as much fun along the way. Thanks, Bill and George, for the opportunities, and for convincing me that Stanford was a dream within reach.

I also wish to thank Professor Marco Bohnhoff for serving as an unofficial second advisor and good friend during my first two years at Stanford. I often wondered if Marco knew the potential ramifications of casually suggesting I write a key computing script for our analysis given my woeful programming skills early on, but I’m incredibly grateful that he asked regardless. My first two years at Stanford were trying for many reasons, and I’ve thought many times about how incredibly fortunate I was to have been able to work with you during that time. Thanks, Marco!

I’d certainly be remiss if I didn’t thank my fellow officemates and research group members over the years for their support, friendship, and humor: Katie Boyle, Jeremy Brown, Chandong Chang, Laura Chiaramonte, Indrajit Das, Paul Hagin, Rob Heller, Sander Hol, Madhur Johri, Arjun Kohli, Amie Lucier, Pijush Paul, Hiroki Sone, John Vermylen, Rall Walsh, Randi Walters, Ali Yaghoubi, and Alec Yang. No matter what the day brought, be it working into the small hours of the morning on research or building giant paper airplanes out of leftover poster paper, sharing the experience and office space with you all was immensely rewarding, and is something I’ll cherish for the rest of my days.

I must also thank the incredible staff of the Geophysics Department, most notably Susan Phillips-Moskowitz, Tara Ilich, and Lauren Nelson for their tremendous efforts and genuine concern for everyone’s well being. As Susan and Tara can likely attest, leaving us students (and sometimes faculty!) to our own devices when attempting such simple tasks as arranging a rental car or submitting a form in timely

ix fashion can sometimes have disastrous consequences, so all your support and assistance was very much appreciated!

I’m generally not a fan of bland, cover-all thank you statements, but I must apply a communal thank you to all my friends in the Geophysics department and beyond over the last five years, too large in number to name specifically, for their support, humor, understanding, and concern. Grad school has been quite the voyage indeed, and as historian Henry Adams said, it’s always more fun and enlightening to travel with someone else. Thanks for being around on the trip.

I thank the Stanford Rock and Borehole Geophysics (SRB) consortium for financial support during my studies at Stanford as well as BP Canada, Apache Canada, and EnCana corporations for providing datasets I analyzed for portions of this thesis (and permission to publish the results).

I would also like to thank the ARCS foundation for the tremendous financial support during much of my graduate studies at Stanford. ARCS is a wonderful foundation with admirable and timely ideals, and I look forward to contributing to its success in the future.

I must reserve my most heartfelt acknowledgements to my parents, Chris and Dia, and my sister, Emily. I am privileged to have received your unwavering love and support over the last five years (and my whole life!) during the brightest of days and darkest of nights. Simple words cannot properly convey my gratitude, but I thank you most for the confidence you’ve always had in me, which was often times much more than I had in myself. Combined though, we had enough to make it, and while this dissertation will only list my name on the front, it is as much all of yours as it is mine. It is my pleasure to dedicate it to all of you.

TABLE OF CONTENTS

ABSTRACT ...... V

ACKNOWLEDGEMENTS ...... VIII

TABLE OF CONTENTS ...... XI

LIST OF TABLES ...... XIV

LIST OF FIGURES ...... XV

CHAPTER 1 - INTRODUCTION ...... 1 1.1 OVERVIEW AND MOTIVATION ...... 1 1.2 THESIS OUTLINE ...... 4 CHAPTER 2 - STRESS AND STRUCTURE-INDUCED SHEAR-WAVE ANISOTROPY ALONG THE 1999 IZMIT RUPTURE, NORTHWEST TURKEY ...... 8 2.1 INTRODUCTION ...... 9 2.2 SEISMIC NETWORK AND DATA ...... 13 2.3 SHEAR-WAVE SPLITTING ANALYSIS PROCEDURE ...... 13 2.4 DELAY TIMES AND FAST SHEAR-WAVE POLARIZATIONS ...... 16 2.4.1 Station CAY ...... 18 2.4.2 Station DOK ...... 21 2.4.3 Station HEN ...... 23 2.4.4 Station BAL...... 26 2.4.5 Station CND ...... 28 2.5 DISCUSSION ...... 30 2.6 CONCLUSIONS ...... 36 2.7 REFERENCES ...... 36 CHAPTER 3 - INTRAPLATE EARTHQUAKES, REGIONAL STRESS, AND FAULT MECHANICS IN THE CENTRAL AND EASTERN U.S. AND SOUTHEASTERN CANADA .. 40 3.1 INTRODUCTION ...... 41 3.2 DATA COLLECTION ...... 43 3.3 DEFINING STRESS ORIENTATIONS AND RELATIVE STRESS MAGNITUDES ...... 44 3.3.1 Stress orientations ...... 44 3.3.2 Relative stress magnitudes ...... 46

3.4 SLIP COMPATIBILITY IN THE REGIONAL STRESS FIELD ...... 49 3.5 DISCUSSION...... 53 3.5.1 The stress field in the central and eastern U.S...... 53 3.5.2 Relative stress magnitudes and faulting styles ...... 53 3.5.3 Slip compatibility and fault friction ...... 55 3.6 CONCLUSIONS ...... 57 3.7 REFERENCES ...... 57 APPENDIX 3A: NEW FOCAL PLANE MECHANISMS COMPILED IN THIS STUDY ...... 60 APPENDIX 3B: NEW FORMAL STRESS INVERSIONS COMPILED IN THIS STUDY ...... 61 APPENDIX 3C: FOCAL MECHANISMS FROM M.L. ZOBACK (1992) ...... 62 CHAPTER 4 - REGIONAL STRESS ORIENTATIONS AND SLIP COMPATIBILITY OF EARTHQUAKE FOCAL PLANES IN THE NEW MADRID SEISMIC ZONE ...... 63 4.1 INTRODUCTION ...... 64 4.2 DATA COLLECTION ...... 66 4.3 UPDATING STRESS ORIENTATIONS ...... 68 4.4 CONSTRAINTS ON RELATIVE PRINCIPAL STRESS MAGNITUDES...... 69 4.5 SLIP COMPATIBILITY ...... 71 4.6 DISCUSSION...... 75 4.7 CONCLUSIONS ...... 78 4.8 REFERENCES ...... 79 CHAPTER 5 - APPLICATION OF AN INTEGRATED, GEOMECHANICS-BASED ANALYSIS TO IDENTIFY FLUID PATHWAYS IN A TIGHT-GAS RESERVOIR ...... 82 5.1 INTRODUCTION ...... 83 5.2 STUDY AREA AND GEOLOGY ...... 85 5.3 DEVELOPING A GEOMECHANICAL MODEL ...... 87 5.3.1 Pore pressure ...... 87 5.3.2 Compressive rock strength ...... 87 5.3.3 Vertical stress (SV) ...... 89 5.3.4 Shmin magnitude ...... 89 5.3.5 SHmax orientations and magnitudes from wellbore failure observations ...... 90 5.3.6 Summary ...... 97 5.4 FRACTURE CHARACTERIZATION AND SLIP ANALYSIS ...... 98 5.4.1 Fracture characterization ...... 99 5.4.2 Critically-stressed fractures ...... 101 5.4.3 Shear-stimulated fractures and wellbore production ...... 104 5.4.4 Summary ...... 105 5.5 PRECISE MICROSEISMIC EVENT LOCATIONS ...... 106 5.5.1 The double-difference technique ...... 108 5.5.2 Stimulation procedure, monitoring array configuration, and microseismic data...... 111 5.5.3 Seismic data processing ...... 114 5.5.3.1 Step 1: Event grouping ...... 115 5.5.3.2 Step 2: Seismogram stacking ...... 115 5.5.3.3 Step 3: Picking absolute P- and S-wave arrivals ...... 117 5.5.3.4 Step 4: Calculating cross-correlation differential travel times ...... 119 5.5.4 Double-difference relocations ...... 120 5.5.4.1 Event relocations ...... 121 5.5.4.2 Event hypocenter resolution ...... 128 5.5.5 Summary ...... 130 5.6 DISCUSSION...... 131 5.7 CONCLUSIONS ...... 133 5.8 REFERENCES ...... 135 APPENDIX 5A: FRACTURE CONSTRAINTS FROM AN AVAZ ANALYSIS ...... 139 APPENDIX 5B: POST-HYDRAULIC FRACTURE WELLBORE IMAGE LOG ...... 143

xii

CHAPTER 6 - A GEOMECHANICAL AND MICROSEISMIC STUDY OF A GAS SHALE DEVELOPMENT IN THE HORN RIVER BASIN ...... 147 6.1 INTRODUCTION ...... 148 6.1.1 Geological setting and reservoir properties ...... 149 6.1.2 Production pad, hydraulic fracturing program, and microseismic monitoring ...... 151 6.2 GEOMECHANICAL AND MICROSEISMIC DATA ...... 153 6.3 GEOMECHANICAL ANALYSIS ...... 154 6.3.1 Stress, pore pressure, and rock strength constraints ...... 154 6.3.2 Shmin variations ...... 156 6.4 MICROSEISMIC ANALYSES ...... 159 6.4.1 Initial microseismic event hypocenter locations ...... 160 6.4.2 Microseismic magnitude distributions ...... 163 6.4.3 Precise microseismic event relocations ...... 167 6.4.3.1 Stage selection, monitoring configuration, and seismic data processing ...... 167 6.4.3.2 Double-difference relocations ...... 169 6.4.3.3 Synthetic location tests: 2 arrays ...... 172 6.4.3.4 Synthetic location tests: 3 arrays and recommendations ...... 175 6.4.4 Waveform similarity analysis ...... 178 6.5 CONCLUSIONS ...... 181 6.6 REFERENCES ...... 183 APPENDIX 6A: GENERAL HYPODD SYNTHETIC TESTS ...... 186

xiii

LIST OF TABLES

Table 2.1: Average shear-wave delay times ...... 18 Table 2.2: Summary of preferred fast shear-wave polarizations...... 31 Table 4.1: New and revised focal plane mechanisms compiled for this study ...... 68 Table 4.2: New Madrid fault planes examined for frictional slip compatibility ...... 68 Table 5.1: Final geomechanical parameters...... 97 Table 5.2: TOMODD weighting scheme for optimal event hypocenter relocations ...... 123 Table 6.1: Reservoir properties of Fort Simpson Formation members ...... 151

xiv

LIST OF FIGURES

Figure 2.1: Izmit region of the North Anatolian fault zone (NAFZ) illustrating major fault trends, tectonic features, surface rupture of the 1999 Izmit mainshock, focal mechanisms of significant earthquakes, seismic stations used in this study, and aftershocks from the 1999 Izmit earthquake ...... 11 Figure 2.2: Example of a shear-wave splitting analysis at station CAY...... 16 Figure 2.3: Graphs of shear-wave splitting delay time vs. hypocenteral depth for all five stations examined in the IZMIT SWS study ...... 18 Figure 2.4: Aftershock hypocenter locations and shear-wave splitting measurements for station CAY ...... 20 Figure 2.5: Aftershock hypocenter locations and shear-wave splitting measurements for station DOK ...... 22 Figure 2.6: Aftershock hypocenter locations and shear-wave splitting measurements for station HEN...... 25 Figure 2.7: Aftershock hypocenter locations and shear-wave splitting measurements for station BAL...... 27 Figure 2.8: Aftershock hypocenter locations and shear-wave splitting measurements for station CND...... 29 Figure 3.1: Stress indicators in the CEUS and SE Canada...... 45 Figure 3.2: Spatial variation of the AΦ parameter across the central and eastern U.S...... 49 Figure 3.3: Example fault friction (μ) map for a single focal mechanism ...... 51 Figure 3.4: Histograms showing the mis-fit in strike and dip between (a) the preferred and (b) the conjugate nodal planes and the theoretically optimal (μ = 0.6) fault plane for all 75 focal plane mechanisms...... 52

Figure 3.5: Mis-fit in strike and dip between the preferred nodal planes and the nearest nodal planes that fail with μ = 0.2, μ = 0.6, and μ = 0.8 ...... 52 Figure 4.1: Major tectonic structures, seismicity, and stress measurements in the New Madrid seismic zone ...... 66 Figure 4.2: Frictional slip compatibility results for a thrust (a) and strike-slip (b) focal mechanism analyzed in the New Madrid seismic zone...... 72 Figure 4.3: Histograms displaying the misfit in strike and dip between the preferred (a) and the conjugate (b) focal mechanism nodal planes and the nearest well-oriented test plane that slips with μ = 0.6 ...... 73 Figure 4.4: (a) Mohr diagram illustrating frictional slip compatibility of focal mechanism nodal planes (circles) and large-scale NMSZ faults (squares) assuming a uniform, transpressional stress field and hydrostatic pore pressure. (b) Large-scale NMSZ faults examined for frictional slip compatibility in the current stress field ...... 74 Figure 5.1: Depositional setting of the NC formation ...... 86 Figure 5.2: Example breakouts observed in the NC formation in two separate wellbore image logs. (a) Well A; (b) Well B ...... 92 Figure 5.3: Stress polygons illustrating the allowable stress states (red polygons) based on borehole breakout widths in (a) Well A and (b) Well B ...... 95 Figure 5.4: Combined stress polygon for Wells A and B ...... 96 Figure 5.5: Orientation of fractures and bedding planes observed in pre-frac wellbore image logs penetrating the NC formation ...... 101 Figure 5.6: Stereonet plots of fracture orientation in the pre- and post-frac wellbore image logs ...... 101 Figure 5.7: Stereonet and Mohr diagrams indicating the orientation and tendency for shear failure of preexisting fractures in the current stress field under (a) unpeturbed reservoir pressure and (b) a 25 MPa increased reservoir pressure ...... 103 Figure 5.8: Shear-stimulated fractures (white poles in stereonet plots) and gas production for three NC formation wells in the WC field ...... 105 Figure 5.9: Map view (left) and west-east cross section (right) illustrating the treatment well orientation and microseismic monitoring array configuration ...... 112 Figure 5.10: Initial event hypocenter locations (710 events) reported by the microseismic service company ...... 113 Figure 5.11: 1-D velocity model for double-difference relocations...... 114 Figure 5.12: Example master trace recorded at Array B ...... 117 Figure 5.13: Initial event hypocenter starting locations (red circles) for the TOMODD double- difference relocations ...... 122 Figure 5.14: TOMODD relocation iterations...... 125

xvi

Figure 5.15: (a) Final microseismic event hypocenter relocations using the TOMODD program (394 events). (b) Original event hypocenter locations from the microseismic service company (710 events) ...... 127 Figure 5.16: EW, NS, and Z differential hypocenter relocations between original event hypocenter relocations and the relocated test sets ...... 129 Figure 5A1: Attribute maps from an AVAZ analysis over the WC field showing (a) P-wave

anisotropy (Vfast – Vslow) and (b) fast P-wave azimuth ...... 140 Figure 5A2: Comparison between fracture orientations observed from wellbore image logs (top row of stereonet diagrams) and fast P-wave azimuths from AVAZ analysis (bottom row of fast P-wave orientations) ...... 142 Figure 5B1: Common fracture characteristics observed in the post-stimulation image log. (a) clustered fractures; (b) fracture coalescing; (c) notch-like features ...... 144 Figure 5B2: (Left) Predicted initiation points of tensile fractures (hydraulic fractures) on the wellbore wall. (Right) Predicted tensile fracture initiation points overlain on actual fractures recorded in the post-frac wellbore image log ...... 145 Figure 6.1: Location of Horn River Basin in northeastern British Columbia ...... 150 Figure 6.2: Production pad well configuration and hydraulic fracturing sequence ...... 152 Figure 6.3: First-year production by well ...... 152

Figure 6.4: Shmin gradient and Shmin/SV ratio for all stages...... 157 Figure 6.5: ISIP difference versus time and spatial difference between consecutively-stimulated stages across the entire production pad...... 158 Figure 6.6: Initial microseismic event hypocenter locations for 15,000 events reported by the microseismic service company ...... 161

Figure 6.7: Rose diagrams illustrating SHmax orientations inferred from microseismic event hypocenter trends ...... 162 Figure 6.8: (a) Distribution of b-values calculated from events recorded in individual hydraulic

fracture stages. (b) The b-value calculated using all MW ≥ -2.0 events in the production pad (4618 events)...... 164 Figure 6.9: b-values and corresponding standard deviations calculated from events falling within 250m x 250m grid blocks ...... 166

Figure 6.10: Hypocenter locations of all events with MW > -1.5 compared with Shmin gradient and b-value distributions ...... 167 Figure 6.11: Initial microseismic event hypocenter locations from the Well 1 Stage 2 stimulation (212 events) ...... 168 Figure 6.12: P-wave velocity log over microseismic hypocenter depths ...... 170 Figure 6.13: Double-difference microseismic event hypocenter relocations of events recorded during the Well 1 Stage 2 stimulation...... 172 xvii

Figure 6.14: Synthetic test of the double-difference technique in the Well 1 Stage 2 monitoring configuration ...... 174 Figure 6.15: Synthetic test of the double-difference technique with a hypothetical third array (Array 3) in the Well 1 Stage 2 monitoring configuration...... 176 Figure 6.16: Example event doublet recorded during the Well 1 Stage 2 stimulation ...... 180 Figure 6.17: Hypocenter locations of events within the two multiplets identified from the Well 1 Stage 2 waveform similarity analysis ...... 180 Figure 6A1: Array and hypocenter configurations for the general HYPODD synthetic relocation test...... 186 Figure 6A2: Synthetic relocation test results for a general monitoring configuration...... 188

xviii

Chapter 1

INTRODUCTION

1.1 Overview and Motivation

This thesis applies the principles of geomechanics to analyze a variety of topics relating to rock properties near major active fault zones, intraplate earthquakes occurring far away from major tectonic plate-bounding faults, and earthquakes associated with hydrocarbon exploitation in low permeability natural gas reservoirs. The thesis contains five independent applications organized into five chapters. One chapter (Chapter 2) examines fundamental controls on rock velocity anisotropy near the North Anatolian Fault in northwestern Turkey. Two more chapters (Chapters 3 and 4) explore the compatibility of shear slip on intraplate faults in the central and eastern United States with frictional faulting theory in the context of the regional tectonic stress field. Finally, the two final chapters (Chapters 5 and 6) integrate geomechanical and seismological techniques to examine the response of low permeability natural gas reservoirs to hydraulic fracture stimulations and how stimulations create fluid pathways within the reservoir. Through these five applications, this thesis will demonstrate the importance of geomechanical principals

in better understanding the crustal conditions controlling intraplate seismicity and how hydraulic fracturing stimulates low permeability hydrocarbon reservoirs.

Accurately constraining the in-situ stresses, fluid pressures, and rock properties within Earth’s upper crust is essential to addressing a broad variety of scientific, engineering, and industrial problems over a range of scales. Fortunately, it is possible to constrain these properties using a multitude of techniques developed over the last several decades. For example, stress magnitudes can be estimated from pressure tests conducted in wells drilled several kilometers into the upper crust, while advanced laboratory tests on rock samples can constrain basic rock properties such as strength, porosity, and permeability. I will refer to the process of constraining the stresses, fluid pressures, and rock properties within a specific region as developing a geomechanical model, and will provide a detailed methodology in Chapter 5. Constraining an accurate geomechanical model is the basis of numerous chapters in this thesis, and provides the opportunity to apply advanced analyses that would otherwise be impossible without accurate knowledge of in-situ stresses, fluid pressures, and rock properties.

The second and third chapters in this thesis examine intraplate earthquake activity in the central and eastern United States (CEUS) in the context of the regional tectonic stress field. Intraplate earthquakes are simply those occurring away from major tectonic plate boundaries, and are ubiquitous throughout the world. Intraplate earthquakes are particularly common within the CEUS, the most famous instance being the devastating 1811-1812 New Madrid earthquake sequence in western Tennessee and northeastern Arkansas. Fundamentally, intraplate earthquakes are the same as their interplate counterparts in that they represent energy released during rapid shear deformation episodes along a fault surface, but their location away from major plate-bounding faults poses several problems for earth scientists. Unlike earthquakes clearly associated with deformation on major tectonic plate-bounding faults, intraplate earthquakes occur on faults rarely breaking Earth’s surface that are often unidentified prior to hosting a significant earthquake. Identifying faults that could host significant

intraplate earthquakes has therefore become a primary objective in the CEUS from a hazard mitigation perspective. In addition, although well established that intraplate earthquakes commonly occur on relict tectonic structures within Earth’s upper crust which are reactivated in the contemporary tectonic stress field, it remains unclear why certain intraplate faults are seismically active while others appear dormant. Therefore, it is also imperative from a fault mechanics perspective to identify what conditions (fault strength, fluid pressure) are necessary to initiate shear slip on intraplate faults. To address these outstanding problems, in Chapters 3 and 4 I evaluate the compatibility of shear slip on intraplate faults both in the broader CEUS and specifically in the New Madrid seismic zone with frictional faulting theory in the context of the contemporary tectonic stress field. The work in these chapters builds upon the work of M.L. Zoback (1992), which originally addressed many of the above problems using earthquake focal mechanisms from eastern North America.

The two primary chapters in this thesis, Chapters 5 and 6, address fundamental questions regarding physical mechanisms driving hydraulic fracturing stimulations in low permeability natural gas reservoirs. Although tremendous natural gas resources have long been known to exist in shale formations throughout the world, the characteristically low permeability of shales, and similar low permeability ‘tight’ sandstones, has prevented feasible exploitation for many decades. However, with the advent of horizontal drilling technology, stage-by-stage completion strategies, and water-based (slickwater) hydraulic fracturing techniques over the last couple decades, gas shale formations have since become economically and technologically feasible exploitation targets. While feasibility has led to increased production, the physical mechanisms driving enhanced productivity remain unclear. Simply put, understanding why hydraulic fracture stimulations are successful in low permeability gas reservoirs remains a fundamental objective within the petroleum industry.

Since hydraulic fracturing stimulations directly fracture rock to create permeable fluid pathways into hydrocarbon reservoirs, understanding the physical mechanisms driving successful hydraulic fracturing stimulations is intrinsically linked

to constraining the in-situ stresses, fluid pressures, and reservoir rock strength. In addition to constraining these properties, small earthquakes induced during hydraulic fracture stimulations, hereafter referred to as microseismic events, can also yield valuable insight into physical deformation mechanisms associated with hydraulic fracturing. Much of this physical insight, however, has come from mapping spatial distributions of microseismic event hypocenters, thereby making our improved understanding contingent upon hypocenter location accuracy. Unfortunately, hypocenter location errors up to 100s of meters are commonly encountered due to limited seismic monitoring array configurations (as well as poorly constrained velocity models and inaccurate P- and S-wave phase picks), highlighting the need to greatly improve hypocenter location accuracy to best understand physical deformation mechanisms. In response to these issues, Chapters 5 and 6 present an integrated, geomechanical and seismological analysis to better ascertain how low permeability natural gas reservoirs respond to hydraulic fracture stimulations. Specifically, Chapter 5 combines geomechanical modeling, fracture characterization, and advanced earthquake location techniques to identify fluid pathways stimulated during hydraulic fracturing in a tight-gas sandstone reservoir. Chapter 6 also combines geomechanical modeling with advanced earthquake location techniques, microseismic event magnitude scaling, and waveform similarity analysis to better evaluate reservoir response to hydraulic fracturing in a gas shale production pad.

1.2 Thesis Outline

Chapter 2 – Stress- and Structure-Induced Shear-Wave Anisotropy Along the 1999 Izmit Rupture, Northwest Turkey

In Chapter 2, I perform a shear-wave splitting analysis to delineate between stress and structure-controlled anisotropy mechanisms in the upper crust beneath the North Anatolian fault zone (NAFZ) in northwestern Turkey. Using seismograms of aftershocks recorded by five stations following the 1999 MW = 7.4 Izmit earthquake, I perform a cross-correlation shear-wave splitting analysis and discover evidence for both stress and structure-controlled anisotropy mechanisms near the NAFZ. In 4

general, stations located greater than 1 km from the NAFZ surface trace exhibit preferred fast shear-wave polarizations parallel to the regional strike of the maximum horizontal stress (SHmax) while stations within 1 km of the NAFZ generally exhibit preferred fast shear-wave polarizations parallel to the local NAFZ strike. Delay times between fast and slow shear-wave arrivals suggest that anisotropic intervals are largely constrained within the upper 8 km of the crust. These analyses indicates highly variable physical properties between the primary fault zone and the surrounding crust, and that it is possible to utilize shear-wave splitting to spatially map these variations with a careful data selection and quality control procedure.

Chapter 3 – Intraplate Earthquakes, Regional Stress, and Fault Mechanics in the Central and Eastern U.S. and Southeastern Canada

In Chapter 3, I revisit the question of slip compatibility of intraplate faults within the contemporary stress field in eastern North America originally examined by M.L. Zoback (1992). I compile well-constrained earthquake focal plane mechanisms and formal stress inversions from the CEUS and southeastern Canada over the past 20 years and utilize these data to investigate the consistency of regional stress orientations, to map faulting styles and relative stress magnitudes, and to investigate the likelihood of shear failure on the more well-oriented nodal planes in the local stress field in the context of frictional faulting theory. The results indicate that horizontal stresses become increasingly more compressive with respect to the vertical stress moving from the south-central United States toward the northeastern U.S. and southeastern Canada, and that nearly all earthquakes in the study area slip in a manner compatible with shear failure on pre-existing faults in the local stress field. That is, slip on intraplate faults can generally be described with a well established fault failure criterion assuming normal hydrostatic fluid pressures and laboratory-determined fault friction coefficients in the upper crust.

Chapter 4 – Regional Stress Orientations and Slip Compatibility of Earthquake Focal Planes in the New Madrid Seismic Zone

Chapter 4 extends the slip-compatibility analysis in Chapter 3 to the New Madrid seismic zone (NMSZ) in western Tennessee and northeastern Arkansas using NMSZ earthquake focal plane mechanisms. In addition, slip compatibility of the three primary faults characterizing the NMSZ is directly examined. The analysis demonstrates that New Madrid fault planes are well oriented for shear failure in the regional stress field assuming hydrostatic pore pressure in the brittle crust and fault strengths consistent with laboratory observations. In other words, I demonstrate that active faults in the NMSZ do not require elevated pore pressure, local stress sources, or anomalous fault strength to enable slip in the current ENE-WSW compressional stress field. This is an important result in light of previous interpretations that NMSZ earthquake activity may be driven by anomalous rock properties and fluid pressures within the upper crust.

Chapter 5 – Application of an Integrated, Geomechanics-Based Analysis to Identifying Fluid Pathways in a Tight-Gas Reservoir

In Chapter 5, an integrated, geomechanical and seismological analysis is performed to identify flow pathways created during a hydraulic fracture stimulation of a tight-gas sandstone reservoir. The analysis begins by developing a geomechanical model for the study area, which indicates a strike-slip stress regime with SHmax oriented NNE-SSW. Next, preexisting natural fractures within the sandstone formation are characterized with wellbore image logs, and I show that many preexisting fractures are stimulated in shear under elevated fluid pressures associated with hydraulic fracturing. Moreover, wellbores intersecting more shear-stimulated fractures tend to have better production, suggesting the stimulated preexisting natural fracture network serves as a fluid pathway into the wellbore. Finally, I also present a novel application of the double-difference earthquake relocation technique to improve hypocenter locations of microseismic events induced during a hydraulic fracturing stimulation. While the double-difference technique significantly reduces scatter in microseismic event hypocenter locations, its ability to delineate fluid pathways is significantly diminished by artifacts arising from the limited monitoring array configurations.

Chapter 6 – A Geomechanical and Microseismic Study of a Gas Shale Development in the Horn River Basin

Chapter 6 reapplies the integrated methodology from Chapter 5 to evaluate reservoir response of a large gas shale production pad in the Horn River Basin in northeastern British Columbia, Canada to hydraulic fracture stimulations. Capitalizing on an extraordinarily large geomechanical and microseismic data set, I am able to denote stress variations over the production pad and speculate on their potential origins. I also examine microseismic event magnitude-scaling relationships to evaluate seismic deformation mechanisms during hydraulic fracturing, and discover evidence for a variety of deformation mechanisms throughout the production pad. Applying the double-difference relocation technique to improve microseismic event hypocenter locations, I find that, similar to Chapter 5, the limited monitoring array configuration appears to introduce significant artifacts into the relocated hypocenters. I am able to demonstrate the artifacts with simple synthetic tests, and make valuable recommendations for optimizing future monitoring array configurations for the double-difference relocation technique. Finally, I perform a waveform similarity analysis between event seismograms and find evidence for repeated slip on small faults within the reservoir. Collectively, the observations suggest that reservoir response to hydraulic fracturing is generally characterized by seismic deformation occurring on small faults reactivated during hydraulic fracturing stimulation, although there is also some evidence for seismic deformation occurring on larger scale localized structures within the reservoir. The chapter demonstrates the advantages of an integrated analysis to understanding physical deformation mechanisms of hydraulic fracture stimulations in low permeability gas reservoirs.

Chapter 2

STRESS AND STRUCTURE-INDUCED SHEAR-WAVE ANISOTROPY ALONG THE 1999 IZMIT RUPTURE, NORTHWEST TURKEY1

Abstract

We perform a shear-wave splitting analysis using seismograms of aftershocks following the 1999 Mw = 7.4 Izmit earthquake to delineate between stress- and structure-controlled anisotropic mechanisms in the upper crust beneath the northwest North Anatolian fault zone (NAFZ). The splitting analysis is performed on direct shear-wave arrivals using a cross-correlation technique to calculate the fast shear-wave polarization and the delay time between fast and slow shear-wave arrivals. Five stations located both within (< 1 km) and near (> 1 km) the NAFZ yield a combined 333 quality shear-wave splitting measurements. The shear-wave splitting

1 This chapter is published in: Hurd, O., and Bohnhoff, M., 2012, Stress and Structural-Induced Shear-Wave Anisotropy Along the 1999 Izmit Rupture, Northwest Turkey, Bulletin of the Seismological Society of America, in press.

measurements indicate that it is occasionally possible to delineate between the two anisotropic mechanisms. Stations greater than 1 km from the NAFZ tend to have dominant fast shear-wave polarizations parallel or subparallel to the regional strike of the maximum horizontal compressive stress, which suggests a stress-controlled anisotropic mechanism. The single station within 1 km of the NAFZ yields a bimodal dominant fast shear-wave polarization that is both parallel and perpendicular to the strike of the nearby Mudurnu fault. We interpret the fault-parallel mode as representing a structure-controlled anisotropic mechanism from the Mudurnu fault. The average delay time for all earthquakes is 0.038 ± 0.031 seconds, and anisotropic intervals do not appear to extend below 8 km. The fast shear-wave polarizations and delay times suggest a heterogeneous anisotropic distribution within the recording array and that the physical properties of the upper crust vary between the NAFZ and surrounding area.

2.1 Introduction

Shear-wave velocity anisotropy, or shear-wave splitting (SWS), occurs when a shear-wave (S-wave) enters an anisotropic medium and is split into approximately orthogonal fast and slow polarizations. Within the upper crust, the most common explanation for this phenomenon is known as extensive dilatancy anisotropy in which the anisotropy is controlled by vertical, fluid-filled microcracks preferentially aligned by the regional stress field (e.g. Crampin, 1991). Other potential controls on anisotropy include alignment of macroscopic features such as fault fabrics or cracks near active faults (Zhang and Schwartz, 1994; Zinke and Zoback, 2000; Tadokoro et al., 2002), preferential closure of fractures in an anisotropic stress field (Boness and Zoback, 2004), sedimentary bedding planes and rock fabrics (Alford, 1986; Kern and Wenk, 1990; Leary et al., 1990) and preferred mineral alignment (Brocher and Christensen, 1990; Sayers, 1994). These mechanisms can be divided into stress- controlled and structure-controlled categories. The stress-controlled mechanisms are those in which the maximum horizontal compressive stress (SHmax) controls the opening or closing of pre-existing microcracks and fractures, producing a fast shear- 9

wave polarization parallel to SHmax. Structure-controlled mechanisms include bedding planes, rock fabrics, preferred mineral alignment and fault-aligned macroscopic features, all of which produce a fast shear-wave polarization parallel to the plane of the structure. Previous SWS studies at the San Andreas fault system in California (e.g. Zhang and Schwartz, 1994; Zinke and Zoback, 2000; Boness and Zoback, 2006) have demonstrated that, with well-located earthquakes, careful quality control procedures and knowledge of local geologic and tectonic structures, it is possible to delineate between the two mechanisms. SWS measurements can also provide an independent means by which to record stress orientations in the upper crust and may be a useful tool for obtaining stress information in the absence of other stress measurements.

The North Anatolian fault zone (NAFZ) in northwestern Turkey trends approximately east-west along a 1,600 km boundary separating the Anatolian plate from Eurasia. Within the current series of large (Mw > 7) earthquakes along the NAFZ that started in Erzincan (east Anatolia) in 1939 and initiated a westward migration of earthquakes, the last two major earthquakes occurred in northwest Turkey in 1999.

Focal mechanisms of the Aug. 17, 1999 Mw = 7.4 Izmit mainshock, its major aftershocks and the Nov. 12 Mw = 7.2 Düzce mainshock all indicate right-lateral, strike-slip faulting with nodal planes striking dominantly east-west and dipping steeply to the north (Örgülü and Aktar, 2001; Tibi et al., 2001; Bohnhoff et al., 2006) (Figure 2.1). The western termination of the Izmit rupture is located offshore beneath the eastern Sea of Marmara (Bouchon et al., 2002; Bulut et al., 2007). Eastward rupture propagation of the Izmit earthquake extended to the Karadere-Düzce area where the Düzce earthquake extended the rupture by another 40 km to the east 87 days later (Barka et al., 1999). The inferred motion from focal mechanism solutions is in good agreement with the regional GPS-derived velocity field (McClusky et al., 2000).

The North Anatolian Fault Zone (NAFZ) in northwestern Turkey is characterized by right-lateral transform faulting. It trends approximately east-west along 1,600 km separating the Anatolian plate from Eurasia. Within the current series of large (MW > 7) earthquakes along the NAFZ that started in Erzincan (east Anatolia)

in 1939 and initiated a westward migration of earthquakes, the last two major earthquakes along the NAFZ occurred in northwest Turkey in 1999. Focal mechanisms of the Aug. 17, 1999 MW = 7.4 Izmit mainshock, its major subevents and the Nov. 12 MW = 7.2 Düzce mainshock all indicate right-lateral strike-slip faulting with nodal planes striking dominantly east-west and dipping steeply to the north (Örgülü and Aktar, 2001; Tibi et al., 2001; Bohnhoff et al., 2006) (Figure 2.1). The western termination of the Izmit rupture is located offshore beneath the eastern Sea of Marmara (Bouchon et al., 2002; Bulut et al., 2007). Eastward rupture propagation of the Izmit earthquake extended to the Karadere-Düzce area where the Düzce earthquake extended the rupture by another 40 km to the east 87 days later (Barka et al., 1999). The inferred motion from focal mechanism solutions is in good agreement with the regional GPS-derived velocity field (McClusky et al., 2000).

The regional stress field across the NAFZ lacks many in-situ measurements, but earthquake focal mechanism inversions from NW Turkey indicate that SHmax along the westernmost NAFZ is oriented NW-SE, roughly 35° clockwise from the average

east-west fault trend (Kiratzi, 2002). Focal mechanism stress indicators in the World

Stress Map database (Heidbach et al., 2008) also indicate a regional NW-SE SHmax orientation striking between 120° and 160°. Local stress field analyses based on aftershock focal mechanisms of the 1999 Izmit earthquake suggested local SHmax variations along the mainshock rupture, although the SHmax orientations were generally consistent with the NW-SE trend (Bohnhoff et al., 2006; Görgün et al., 2010). The stress analyses as well as analyses of surface rupture, teleseismic, strong motion and geodetic data all indicate a separation of the Izmit earthquake into distinct faulting styles on specific segments (Reilinger et al., 2000; Tibi et al., 2001; Barka et al., 2002; Delouis et al., 2002) (Figure 2.1). The 40-50° difference in orientation between the strike of the NAFZ in the study area and the regional SHmax trend make this region an ideal location to distinguish between stress and structure-controlled anisotropy.

Previous SWS studies examining the Izmit rupture area identified both stress- and structure-controlled anisotropic mechanisms (Tadokoro et al., 2002; Peng and Ben-Zion, 2004). However, the dominant fast shear-wave polarizations varied considerably, and were not consistently identifiable with a specific anisotropic mechanism. The studies also recorded numerous dominant fast shear-wave polarizations that did not reflect the regional SHmax trend or the strike of any known tectonic structure. To examine these issues in greater detail, we perform a shear-wave splitting analysis using seismograms of aftershocks following the 1999 Izmit rupture to calculate the fast shear-wave polarization and the delay time between the fast and slow shear-wave arrivals. The primary objective is to distinguish between stress- and structure-controlled shear-wave velocity anisotropy at distinct locations both within (< 1 km) and near (> 1 km) the NAFZ. We will also interpret the SWS measurements in the context of the local tectonic and stress setting. Our study expands upon the SWS study by Peng and Ben-Zion (2004) in the Izmit region by using a broader recording array and carefully analyzing spatial variations in fast shear-wave polarization and delay time. We hope that the analysis will lend insight into the complex local controls on anisotropy and the distribution of anisotropic intervals within the upper crust.

2.2 Seismic Network and Data

We analyzed seismograms recorded by two seismic networks deployed across the western NAFZ both preceding and following the Izmit rupture. The first network consisted of 15 short-period stations that have been in operation since 1996 (SABONET, Milkereit et al., 2000). To monitor the Izmit aftershock activity at low magnitude-detection threshold, the German Task Force for Earthquakes (GTF) of the Helmholtz-Centre Potsdam GFZ deployed a second network consisting of 21 short- period stations in the area within four days after the Izmit earthquake (Baumbach et al., 2003). The combined deployment therefore contained 36 three-component stations and was in operation for a period of 60 days following the Izmit earthquake. Only earthquakes recorded during this time period were used in the analysis. The average station spacing was 15 km, and seismograms were sampled at 100 samples per second. The aftershock catalog contained over 10,000 earthquakes, ~4,700 of which were relocated using the double-difference technique to an accuracy of ~400 meters (Bulut et al., 2007). Only this subset of relocated earthquakes was considered in this study. Peng and Ben-Zion (2004) used a separate recording array with stations generally much closer to the surface strike of the NAFZ and largely restricted to the Karadere- Düzce segment. Due to the brief recording period, we did not conduct a temporal SWS analysis, but note that Peng and Ben-Zion (2005) examined spatiotemporal variations in SWS parameters following the Izmit and Düzce earthquakes.

2.3 Shear-Wave Splitting Analysis Procedure

The analysis procedure consisted of three main steps: earthquake preselection, the splitting analysis and quality control. During preselection, all earthquakes in the relocated hypocenter catalog passed through a screening process on a station-by- station basis that selected earthquakes meeting the following two criteria. (1) The earthquake fell within a cone beneath the respective station whose edges were defined by a straight-line incidence angle of 45° from the vertical. Assuming a homogeneous half-space with a Poisson’s ratio of 0.25, Nuttli (1961) demonstrated that incident S-

waves arriving at the free surface at angles < 35° from the vertical will not experience any contaminated particle motions from S-to-P phase conversions. However, as ray paths are bent significantly toward the vertical upon entering the low-velocity, near- surface layer, we extended the acceptable straight-line incidence angle to < 45° for this study. (2) The S-wave arrival was clearly identifiable on the seismogram. After the preselection, five stations (CAY, DOK, HEN, BAL and CND (Figure 2.1)) each containing more than 50 earthquakes were selected for further analysis. Stations with fewer than 50 earthquakes were not considered in order to achieve statistically significant results. Of the 2,176 total aftershocks falling within the cones beneath the five stations, 737 were selected for the shear-wave splitting analysis.

We used a semi-automated procedure to perform the splitting analysis and developed a set of criteria, which included a visual check of each individual analysis result, to ensure quality SWS measurements. The advantage of this procedure comes from utilizing the efficiency of automated calculation without being limited by the potential oversights of a fully automated quality control procedure or the inherent subjectivity of a purely visual quality control procedure. The analysis technique used a cross-correlation between the two S-wave arrivals on horizontal-component seismograms to determine both the fast shear-wave polarization and the delay time between the fast and slow shear-wave arrivals that best correct the shear-wave velocity anisotropy (Bowman and Ando, 1987). For all earthquakes, we assumed the shear- wave splitting results from a single anisotropic layer.

The splitting analysis was performed separately for each station on an earthquake-by-earthquake basis. We used only the horizontal (radial and transverse) seismograms since high raypath incidence angles place the vast majority of the S-wave energy on these components (Figure 2.2a). All seismograms were bandpass filtered between 3 and 15 Hz, which preserved the ~7 Hz S-wave dominant frequency while limiting both high- and low-frequency noise. We performed the analysis by overlaying the horizontal seismograms and moving a 0.3 second time window across the S-wave arrival (Figure 2.2b). A 0.3 second time window provided the optimal

window about the S-wave that minimized contamination with converted phases and the onset of the coda. The window started 0.07 seconds before the S-wave arrival and moved in 0.02 second increments, eventually moving through ten windows altogether. For each increment, we rotated the seismograms within the window through potential fast shear-wave polarizations from -90º (west) to 90º (east) in 1º increments and through potential delay times from 0 to 0.3 seconds in steps of 0.01 seconds. At each individual polarization and delay time pairing, a cross-correlation was performed between the two windowed S-wave seismograms and a correlation coefficient, which is a measure of the similarity between the signals (0 representing no correlation, 1 being the exact same waveform), was calculated. A grid search was then performed over all windows and all possible azimuths and delay times to find the largest correlation coefficient. The corresponding azimuth and delay time represented the polarization of the fast shear-wave and the delay time between the arrivals of the split shear-waves, respectively.

Lastly, we utilized five criteria to ensure a quality splitting measurement. (1) Earthquakes having corrected seismograms with a correlation coefficient less than 0.8 were discarded. (2) The corrected seismograms passed a visual inspection to ensure a quality delay time correction of the split shear-wave and that cycle-skipping did not occur (Figure 2.2c). (3) The particle motion of the corrected shear-wave was examined to ensure that it was resolved into a linear pattern after correction (Figure 2.2d). Earthquakes having corrected shear-waves with a rectilinearity coefficient less than 0.8 were discarded. (4) The standard deviation of the fast shear-wave polarization and delay time over a ± 0.04 second time shift surrounding the best time window was less than 30° and 0.02 seconds, respectively. This ensured a stable solution for each analyzed earthquake. (5) Earthquakes having fast and slow shear- wave polarizations within ±10° of the initial shear-wave polarization were discarded to prevent analysis of null measurements which are unable to unambiguously resolve fast shear-wave polarizations (Leary et al., 1990; Savage, 1999; Wüstefeld and Bökelmann, 2007). A total of 333 of the 737 (45%) earthquakes between the five stations passed all the quality control criteria.

Figure 2.2: Example of a shear-wave splitting analysis at station CAY. (a) Bandpass (3 – 15 Hz) filtered vertical, transverse and radial seismograms prior to anisotropy correction with P- and S- wave arrivals denoted. (b) Magnification of a 0.3-second time window around the S-wave arrival on the transverse (solid line) and radial (dashed line) components illustrating shear-wave splitting. (c) The same seismograms after being corrected for anisotropy. (d) Particle motion plot of windowed seismograms before (dashed line) and after (solid line) anisotropy correction. Note the relatively linear particle motion of the S-wave arrival after correction. The final fast shear-wave azimuth (Φ), delay time (∆t), cross-correlation coefficient (CC) and rectilinearity coefficient (Rect) are listed below window (d). See text for complete discussion of splitting parameters.

2.4 Delay Times and Fast Shear-Wave Polarizations

The shear-wave splitting analysis yielded 333 quality measurements recorded by five stations (CAY, DOK, HEN, BAL and CND) located along the NAFZ. These five stations were located in distinctly different local tectonic settings and at varying

distances from the surface rupture of the Izmit earthquake. Prior to presenting the SWS measurements, it is important to recall two inherent limitations of a SWS analysis. First, a SWS analysis alone only serves as an indicator for the presence or absence of anisotropy along the raypath between the earthquake and recording station. It cannot usually resolve precisely where the anisotropy is encountered along the raypath. For crustal SWS studies, a broad depth distribution of earthquake hypocenters or local geological information may be used to further constrain the position of anisotropic intervals.

Second, while the delay time is influenced by the amount of anisotropy accumulated along the entire raypath, the fast shear-wave polarization reflects the last interval of anisotropy encountered along the raypath (Yardley and Crampin, 1991). That is, if two anisotropic intervals with different properties are encountered, the delay time would reflect the cumulative anisotropy from both intervals while the recorded fast shear-wave polarization would only reflect the anisotropy in the shallower interval closer to the station. As mentioned, we assumed only a single horizontal layer of anisotropy in our interpretations, although we discuss potential evidence for geologic heterogeneity.

The average delay time for all 333 earthquakes is 0.038 ± 0.031 seconds (Table 2.1), with a maximum delay time of 0.16 seconds. The vast majority (95%) of earthquakes have a delay time less than or equal to 0.1 seconds, and the average error for all individual delay time measurements is 0.0014 seconds. Assuming an average delay time of 0.05 seconds, a shear-wave velocity of 3 km/s and a hypocenter distance between 5 and 15 km from the station, this represents approximately 1 – 3% anisotropy in the upper crust. There does not appear to be any clear correlation between hypocentral depth and shear-wave delay time for any of the stations (Figure 2.3), which suggests that the anisotropy is generally constrained to the upper several kilometers of the crust and is largely independent of hypocentral distance from the station. However, detailed analyses of SWS measurements from multiple stations suggest some anisotropic intervals may be located earlier along the raypaths and thus

deeper in the crust. The dominant fast shear-wave polarization is variable between the five stations analyzed in this study, but is generally quite consistent at each specific station. The fast shear-wave polarizations are well constrained, and the average error for all individual measurements is 3.5° ± 6.2°. The fast shear-wave polarizations and delay times are presented station-by-station below.

Table 2.1: Average shear-wave delay times.

Station Average delay time (sec.) CAY 0.034 ± 0.032 DOK 0.049 ± 0.039 HEN 0.036 ± 0.022 BAL 0.059 ± 0.033 CND 0.036 ± 0.031 All Stations 0.038 ± 0.031

Figure 2.3: Graphs of shear-wave splitting delay time versus hypocentral depth for all five stations examined in the study (see Figure 2.1 for station locations). Average error in each delay time measurement is 0.0014 seconds.

2.4.1 Station CAY Station CAY is located ESE of Lake Sapanca at the western flank of the Adapazari Basin that was identified as a small-scale, pull-apart basin along the NAFZ within an east-west extensional normal faulting regime (Bohnhoff et al., 2006) (Figures 2.1 & 2.4). It is approximately 5 km south of the Izmit surface rupture at a location where almost no coseismic slip was observed. A total of 94 quality SWS measurements were obtained for this station, and earthquakes had nearly full backazimuthal coverage. Interestingly, only an extremely small number of aftershocks fall on the north-dipping NAFZ plane that ruptured during the Izmit earthquake. The

vast majority of aftershocks occur several kilometers south of the surface rupture on a system of NW-SE trending normal faults that were reactivated by supershear behavior during the Izmit rupture (short black lines, Figure 2.4) (Bouchon and Karabulut, 2008). A large number of aftershocks are located beneath the station resulting in nearly vertical incidence angles. Aftershocks recorded at station CAY tend to form a NW-SE trending cloud of activity. The dominant fast shear-wave polarization is approximately 40º-50º, although nearly all polarizations fall within a larger range of 30º-90º. The fast shear-wave polarization does not correspond to the regional orientation of SHmax, the local strike of the NAFZ or any mapped secondary fault system. A small group of 10-15 earthquakes indicates a fast shear-wave polarization trending ~315º, which is parallel to the strike of both the regional SHmax and short normal faults near the station.

The observed fast shear-wave polarizations do not appear to correlate with either hypocenter backazimuth or hypocenter distance from the station (Figure 2.4). SWS measurements from stations BU and SL in Peng and Ben-Zion (2004), which were also located in the Adapazari Basin, also indicated an unusual NE-SW dominant fast shear-wave polarization. There is a slight gap in the aftershock distribution between 30.48º and 30.5º longitude that breaks the seismicity into two clusters (Figure 2.4). Interestingly, the average delay times for the distal cluster of aftershocks with respect to the station are smaller than those for the nearer cluster (0.025 ± 0.027 seconds vs. 0.043 ± 0.033 seconds, respectively), albeit with overlapping standard deviations (Figure 2.4). This could reflect a heterogeneous anisotropic distribution near the station such that raypaths from the further cluster spend less time sampling the anisotropic interval and accumulate a smaller delay time. Since the fast shear- wave polarization is highly consistent and largely independent of hypocenter backazimuth and distance, the raypaths from both clusters appear to encounter a common anisotropic mechanism before reaching the station. As there is no clear correlation between earthquake depth and delay time (Figure 2.3), we conclude that the anisotropic interval likely does not extend below the shallowest earthquakes (~8 km).

Figure 2.4: Aftershock hypocenter locations and shear-wave splitting measurements for station CAY. Red triangles represent station locations, solid black lines represent active faults, and dashed red lines indicate the Izmit surface rupture. Earthquakes are color-coded by fast shear-wave orientation (-90˚ = west, 90˚ = east) and scaled in size by delay time. All earthquakes are collapsed onto an E-W line for the depth section. The regional SHmax trend is inferred from stress measurements in the WSM database. Major faults and tectonic features are labeled on maps. Fault trends are from Bouchon and Karabulut (2008) and Pucci et al. (2007). Shaded relief maps are from Ryan et al. (2009).

2.4.2 Station DOK Station DOK is located near the Mudurnu Fault, which is the southern of the two NAFZ fault branches bounding the elevated crustal Almacik block towards the eastern part of the Izmit rupture (Figures 2.1 & 2.5). The Mudurnu fault hosted a MW = 7.1 earthquake in 1967 (Ambraseys and Zatopek, 1969) and was not activated during the 1999 Izmit rupture. Since not a single Izmit aftershock is located on the Mudurnu Fault, this fault segment does not seem to have been substantially re-loaded since 1967, and the stress redistributions of the nearby Izmit earthquake were not sufficient to trigger any seismic activity. Aftershocks have narrow backazimuthal coverage (310º-360º) and are mostly located at epicentral distances greater than 8 km from the station. The SWS analysis indicates two dominant fast shear-wave polarizations. The first is oriented east-west, roughly parallel to the strike of the nearby Mudurnu Fault, while the second exhibits a nearly perpendicular north-south polarization. We interpret the east-west polarization as reflecting a structural source of anisotropy from the Mudurnu fault, although the source of the north-south polarization is unclear. It may reflect structural fabrics of local north-south trending normal faults related to extension of the Akyazi basin (Bulut et al., 2007). Ben-Zion et al. (2003) observed that stations falling on or near the Izmit surface rupture often recorded fault zone trapped waves, which can obscure S-wave arrivals and affect the splitting analysis. However, the waveforms recorded at DOK did not generally appear to exhibit any long-period oscillatory motions characteristic of fault zone trapped waves in the region (Ben-Zion et al., 2003).

The bimodal fast shear-wave polarization is not clearly reflected in the aftershock distribution (Figure 2.5). Delay times from all aftershocks are very similar and all aftershock raypaths pass through the Mudurnu fault zone upon reaching the station. Since the raypaths traverse a similar volume beneath the station yet yield two very different fast shear-wave polarizations, we suggest that the anisotropic region(s) may exist closer to the hypocenters rather than directly beneath the station. The lack

of correlation between delay time and earthquake depth (Figure 2.3b) implies that little anisotropy exists below the shallowest earthquake depth (~12 km).

Figure 2.5: Aftershock hypocenter locations and shear-wave splitting measurements for station DOK. Red triangles represent station locations, solid black lines represent active faults, and dashed red lines indicate the Izmit surface rupture. Earthquakes are color-coded by fast shear-wave orientation (-90˚ = west, 90˚ = east) and scaled in size by delay time. All earthquakes are collapsed onto an E-W line for the depth section. The regional SHmax trend is inferred from stress measurements in the WSM database. Major faults and tectonic features are labeled on maps. Fault trends are from Bouchon and Karabulut (2008) and Pucci et al. (2007). Shaded relief maps are from Ryan et al. (2009).

2.4.3 Station HEN Station HEN is located towards the eastern part of the Izmit rupture and close to the Karadere Fault that hosted up to 1.5 m of right-lateral coseismic slip during the Izmit earthquake (Hartleb et al., 2002) (Figures 2.1 and 2.6). We obtained quality SWS measurements from 91 earthquakes that originated on or near the downdip extension of the Karadere fault that dips at ~67º to the NNW (Bulut et al., 2007). Backazimuthal coverage is largely from 100º-240º, and aside from several earthquakes falling very near the station, all epicentral distances exceed 5 km. The dominant fast shear-wave polarization from the splitting analysis is WNW-ESE (270º-310º) while a less pronounced fast shear-wave polarization is observed at ~60º. The WNW-ESE polarization might reflect a stress-controlled anisotropic mechanism from a slightly

(20-30°) counterclockwise-rotated SHmax, which would be consistent with the local trend of the Karadere fault in this region being rotated ~25° counterclockwise with respect to the regional east-west NAFZ trend. The ~60º fast shear-wave polarization matches the local Karadere fault trend.

There appears to be a slight correlation between hypocenter location and the corresponding fast shear-wave polarization. Earthquakes falling on the S and SW sides of the aftershock cloud, specifically along what appears to be an ENE-WSW linear band of seismicity on the downdip portion of the Karadere fault, tend to yield the dominant WNW-ESE polarization. In contrast, NE aftershocks occurring at the transition to the Düzce Basin tend to yield the secondary NE-SW fast shear-wave polarization. We suggest the primary polarization indicates stress-controlled anisotropy while the secondary polarization indicates a slight component of structure- controlled anisotropy from the Karadere fault.

The SW aftershocks appear to exhibit a slightly smaller average delay time than the NE aftershocks (Figure 2.6), which might be due to their overall shorter distance from the station. Although the correlation between hypocenter location and fast shear-wave polarization suggests that the anisotropic interval may lie closer to the hypocenters, the anisotropic interval appears shallower than 10 km based on the lack 23

of correlation between delay time and earthquake depth (Figure 2.3c). We note that several of the same aftershocks recorded by station HEN were also recorded by station CAY, although the earthquakes do not yield similar fast shear-wave polarizations. Station GE in Peng and Ben-Zion (2004) was located near HEN and it also revealed a dominant WNW-ESE fast shear-wave polarization.

Figure 2.6: Aftershock hypocenter locations and shear-wave splitting measurements for station HEN. Red triangles represent station locations, solid black lines represent active faults, and dashed red lines indicate the Izmit surface rupture. Earthquakes are color-coded by fast shear-wave orientation (-90˚ = west, 90˚ = east) and scaled in size by delay time. All earthquakes are collapsed onto an E-W line for the depth section. The regional SHmax trend is inferred from stress measurements in the WSM database. Major faults and tectonic features are labeled on maps. Fault trends are from Bouchon and Karabulut (2008) and Pucci et al. (2007). Shaded relief maps are from Ryan et al. (2009). 25

2.4.4 Station BAL Station BAL is located at the center of the Almacik Block that represents an elevated crustal block between the two major NAFZ branches, the Karadere-Düzce in the north and the Mudurnu in the south (Figures 2.1 and 2.7). The station records aftershock activity from the easternmost part of the Izmit rupture, the Düzce Basin, at about 10 km epicentral distance from the fault. Backazimuthal coverage varies from 320˚ to 40˚. Fourteen earthquakes passed the preselection and quality control criteria and yielded reliable measurements. The dominant fast shear-wave polarization trends

NW-SE (310º-320º), which corresponds to the regional trend of SHmax. We therefore conclude that the dominant fast shear-wave polarization reflects a stress-controlled anisotropic source, although a series of en-echelon Pliocene-Quaternary normal faults in the region share a similar orientation (Pucci et al., 2007) (short black lines, Figure 2.7).

Earthquake raypaths recorded at station BAL coming from aftershocks north of the NAFZ surface trace must pass through the fault zone before reaching the station while those on the south side do not travel through the fault zone (Figure 2.7). Most of the aftershocks yielding the dominant NW-SE fast shear-wave polarization appear to be located on the north side of the NAFZ, which suggests that either the fault zone is not a source of anisotropy or that anisotropic crust beneath BAL overwrites the fault zone effects on the fast shear-wave polarization. However, since the average delay time from northern aftershocks is generally larger than from the southern aftershocks (Figure 2.7), it appears that the fault zone is a source of anisotropy. We conclude that the anisotropic interval may lie closer to the hypocenters rather than directly underneath station BAL, although the dominant NW-SE fast shear-wave polarization indicates that the dominant mechanism is stress-controlled. There does not appear to be a correlation between delay time and depth (Figure 2.3d), suggesting that very little anisotropy exists below the shallowest earthquakes (~10 km).

Figure 2.7: Aftershock hypocenter locations and shear-wave splitting measurements for station BAL. Red triangles represent station locations, solid black lines represent active faults, and dashed red lines indicate the Izmit surface rupture. Earthquakes are color-coded by fast shear-wave orientation (-90˚ = west, 90˚ = east) and scaled in size by delay time. All earthquakes are collapsed onto an E-W line for the depth section. The regional SHmax trend is inferred from stress measurements in the WSM database. Major faults and tectonic features are labeled on maps. Fault trends are from Bouchon and Karabulut (2008) and Pucci et al. (2007). Shaded relief maps are from Ryan et al. (2009). 27

2.4.5 Station CND Station CND is located in the vicinity of the Düzce fault at the easternmost tip of the aftershock activity where no coseismic slip during the Izmit earthquake was observed at the surface (Figures 2.1 and 2.8). A total of 91 quality SWS measurements were obtained for this station. The horizontal seismograms recorded at CND usually displayed a low signal-to-noise ratio compared to those of other stations. Nearly all aftershocks are located more than 5 km to the NNW of the station, and backazimuthal coverage is largely from 260º-40º. The dominant fast shear-wave polarization is between 310º-340º, which corresponds to the regional orientation of

SHmax and some local Pliocene-Quaternary faults. A much smaller number of earthquakes display an east-west fast polarization that reflects the trend of the Düzce fault. Consequently, we conclude the dominant anisotropic mechanism is stress- controlled while the secondary fast polarization appears to reflect a slight component of structure-controlled anisotropy.

Raypaths from nearly all of the analyzed aftershocks pass through the Düzce fault zone, although delay times from all earthquakes display little variation with respect to station-hypocenter distance or backazimuth (Figure 2.8). The fast shear- wave polarization also generally appears to be independent of station-hypocenter distance and backazimuth, which suggests a source of anisotropy in the upper crust directly beneath the station. Note that several aftershocks on the west side of the cluster share a similar spatial location and fast shear-wave polarization to several aftershocks recorded at station BAL. A lack of correlation between shear-wave delay time and hypocenter depth implies that the anisotropy is insignificant below a depth of 8 km (Figure 2.3e).

Figure 2.8: Aftershock hypocenter locations and shear-wave splitting measurements for station CND. Red triangles represent station locations, solid black lines represent active faults, and dashed red lines indicate the Izmit surface rupture. Earthquakes are color-coded by fast shear-wave orientation (-90˚ = west, 90˚ = east) and scaled in size by delay time. All earthquakes are collapsed onto an E-W line for the depth section. The regional SHmax trend is inferred from stress measurements in the WSM database. Major faults and tectonic features are labeled on maps. Fault trends are from Bouchon and Karabulut (2008) and Pucci et al. (2007). Shaded relief maps are from Ryan et al. (2009).

2.5 Discussion

The objective of this study was to delineate between stress- and structure- controlled anisotropic mechanisms along the North Anatolian fault zone by performing a shear-wave splitting analysis on seismograms of aftershocks following the 1999 Izmit earthquake. Our results indicate that it is occasionally possible to delineate between the two mechanisms, and that there appears to be a slight correlation between the controlling mechanism and station proximity to the NAFZ or Izmit surface rupture (Table 2.2). Of the five stations examined in our study, only one (DOK) is located within 1 km of the NAFZ or Izmit surface rupture. Station DOK exhibits a bimodal dominant fast shear-wave polarization that appears to partially reflect a significant structure-controlled source of anisotropy, possibly related to aligned macroscopic features associated with the Mudurnu fault. Three of the four stations located greater than 1 km from either the main NAFZ fault trace or the Izmit surface rupture (HEN, BAL and CND) exhibit a dominant fast shear-wave polarization that is parallel or sub-parallel to the regional SHmax trend, which we interpret as representing a stress-controlled anisotropic mechanism such as extensive dilatancy anisotropy. The final station located further than 1 km from the mapped NAFZ (CAY) exhibits a NE-SW fast shear-wave polarization that does not reflect the regional SHmax orientation or the strike of any nearby mapped faults.

Table 2.2: Summary of preferred fast shear-wave polarizations.

Station Station-NAFZ Preferred Anisotropic mechanism Remarks distance (km) polarization(s) (Stress- or structure- controlled) CAY 2-3 NE-SW Unclear Fast polarization not parallel to SHmax or any mapped faults DOK < 1 N-S and E-W Structural (for E-W fast Bimodal preferred fast polarization). Unclear for N-S polarization. E-W fast polarization polarization reflects source from Mudurnu fault HEN > 5 WNW-ESE Stress Might reflect a slightly rotated SHmax BAL > 10 NW-SE Stress Parallel to regional SHmax CND 3-4 NW-SE Stress Parallel to regional SHmax

The difficult and often ambiguous nature of interpreting crustal SWS measurements has been noted by many authors (see review by Crampin and Peacock, 2008). Interpretation difficulties can result from shortcomings of the analysis procedure, poorly-constrained geological characterizations, and the complex behavior of the shear-wave itself. Despite these uncertainties, SWS measurements contain important information on the anisotropic properties of the upper crust, and can also be useful for constraining local stress orientations. Keeping these uncertainties in mind, we use the well constrained SWS measurements in this study to further examine and discuss the first-order distribution of and controls on anisotropy in the Izmit region.

To illuminate the spatial distribution of anisotropy along the NAFZ, we examine the relationship between fast shear-wave polarizations and earthquake hypocenter locations. The relationships suggest that both near station and near source anisotropic intervals may influence the fast shear-wave polarizations. For example, station CND yielded a highly-consistent dominant fast shear-wave polarization despite recording aftershocks distributed over a wide backazimuth (Figure 2.8). This suggests that the anisotropic interval influencing the fast shear-wave polarization may lie closer to the station since all raypaths must pass through a common crustal volume as they bend toward vertical when approaching the station. In contrast, station DOK exhibits

a bimodal dominant fast shear-wave polarization, possibly suggesting near-source anisotropy (Figure 2.5). That is, the presence of two dominant fast shear-wave polarizations may suggest that the crustal volume directly beneath the station traversed by most incident raypaths has a minimal influence on the fast shear-wave polarizations. Still other stations might suggest a combination of near-source and near-station anisotropy. For example, station HEN has a dominant WNW-ESE fast shear-wave polarization, but appears to illustrate a slight correlation between hypocenter location and fast shear-wave polarization (Figure 2.6).

The dominant NE-SW fast shear-wave polarization recorded at station CAY is notable since it does not reflect the regional SHmax orientation or the trend of any significant faults in the area. Interestingly, the NAFZ fault segment near station CAY, which extends from the Izmit epicenter eastward to the western tip of the Almacik block, ruptured at supershear speeds during the Izmit earthquake (Bouchon et al., 2002). In addition, the majority of aftershocks occurring along the supershear portion were not located on the surface rupture itself but rather off the fault, presumably on secondary structures activated during the rupture (Bouchon and Karabulut, 2008). Most of the relocated aftershocks from Bulut et al. (2007) are located several kilometers off the Izmit surface rupture, possibly on a series of short, NW-SE striking normal faults resulting from the splaying of the NAFZ into two segments at this location (Figure 2.5) (Bouchon and Karabulut, 2008). While supershear rupture may have influenced the aftershock activity near CAY, it is not clear what effect, if any, it had on the fast shear-wave polarizations. We are currently unable to explain the dominant fast shear-wave polarization at CAY within the context of local tectonic structures or the contemporary stress field. Based on the fast shear-wave polarizations across all stations, we conclude that there are distinctly separate and spatially variable anisotropic intervals within and near the NAFZ.

In addition to fast shear-wave polarizations, delay times between the fast and slow shear-wave arrivals are also used to examine spatial anisotropic distributions. However, since delay times may accumulate within multiple anisotropic intervals

along a raypath, they are commonly more difficult to interpret and often less insightful than fast shear-wave polarizations in constraining locations of anisotropic intervals (e.g. Crampin and Booth, 1985). The most insightful delay time measurements may be from station CAY where hypocenters located further away from CAY tended to yield smaller delay times than those located near CAY (Figure 2.4). This may indicate a heterogeneous anisotropic distribution such that raypaths from hypocenters directly beneath the station spend a longer time in the anisotropic interval than those coming from greater distances, although the consistent fast shear-wave polarization indicates that a common anisotropic mechanism is influencing most raypaths prior to reaching the station. Considering the variable delay times measured over all stations, we conclude that the spatial distribution of anisotropy is locally variable over the study area, which is consistent with previous interpretations of delay time measurements from both the North Anatolian (Peng and Ben-Zion, 2004) and San Andreas (e.g. Cochran et al., 2003) fault systems.

Our SWS measurements are consistent with previous crustal SWS studies having indicated primary sources of anisotropy lying both within the upper few kilometers of the crust (e.g. Liu et al., 2004 and Peng and Ben-Zion, 2004) and deeper in the crust (e.g. Li et al., 1994; Zinke and Zoback, 2000). While the limited depth distribution of hypocenters examined in this study inhibits precise resolution of anisotropic intervals, we can conclude that the depth of anisotropy is generally above 8-10 kilometers based on the lack of correlation between hypocenter depth and delay time for any station (Figure 2.3). Delay times measured in this study indicate 1-3% anisotropy in the upper crust, which is in good agreement with values observed in other crustal SWS studies (e.g. Li et al., 1994, Boness and Zoback, 2006). We do not observe an increased average delay time for stations located on the fault zone compared to stations located outside the fault zone, as was reported by both Boness and Zoback (2006) and Peng and Ben-Zion (2004) and interpreted as an increased fracture density within the fault zone.

Fast shear-wave polarizations and delay times measured at a single station can be highly sensitive to specific tectonic structures traversed by raypaths in the Izmit region. Significant structures such as the ~3 km deep pull-apart sedimentary basin below the Akyazi Plain, the elevated crustal Almacik Block and the Izmit rupture zone itself likely contribute to the variable anisotropic distributions across the NAFZ. Variations in SWS measurements may also result from dipping anisotropic symmetry axes or multiple layers of anisotropy (e.g. Crampin and Booth, 1985; Savage and Silver, 1993), although we do not examine these possibilities in this study. It is also important to note that the earthquakes used in this study and the study of Peng and Ben-Zion (2004) were aftershocks following a major rupture. That is, the aftershock distributions, and perhaps local stress fields, were likely perturbed by the mainshock, as evident by the activation of secondary faults following the Izmit rupture (e.g. Seeber et al., 2000). Although not discussed in this study, we refer the reader to Bohnhoff et al. (2006) and Görgün et al. (2010) for discussions regarding the potential effects of the Izmit earthquake on the local stress field and aftershock characteristics.

Crustal anisotropy along the Karadere-Düzce segment of the NAFZ was previously examined by Peng and Ben-Zion (2004) using seismic data from 17 densely spaced stations recording aftershocks following the 1999 Izmit earthquake. They divided both their stations and earthquakes into fault zone (FZ) and non-fault zone (non-FZ) subgroups based on distance from the Izmit surface rupture and reported their shear-wave splitting measurements based on these groupings. Their results generally indicated that stations within the FZ subgroup exhibited a dominant fast shear-wave polarization parallel to and changing with the local fault strike while stations in the non-FZ subgroup tended to exhibit more complex sets of dominant fast shear-wave polarizations. They concluded that different mechanisms and tectonic structures likely influence anisotropy measurements in the region. While different station locations commonly prevent comparison between the results of our study and those of Peng and Ben-Zion (2004), two regions are directly comparable. First, all three stations within the Adapazari Basin (CAY in this study, BU and SL in Peng and Ben-Zion) exhibit a dominant NE-SW fast shear-wave polarization that does not

reflect the strike of the local NAFZ or the regional SHmax trend. Second, stations within the Almacik Block (BAL this study, CF and WF in Peng and Ben-Zion) yield variable dominant fast shear-wave polarizations. Station BAL exhibits a dominant NW-SE fast shear-wave polarization while CF and WF display significantly more complex fast polarizations including a dominant N-S trend. To summarize, although the results between the two studies sometimes differ on local scales, both suggest that there are a variety of mechanisms controlling the dominant fast shear-wave polarizations throughout the Izmit region, and that SWS measurements from stations near the NAFZ tend to indicate a structure-controlled anisotropic mechanism.

Crustal SWS studies on the San Andreas fault system in western California have indicated both a fairly clear (Boness and Zoback, 2006; Zhang and Schwartz, 1994) and a less distinct (Cochran et al., 2006; Zhang et al., 2007; Liu et al., 2008) correlation between dominant fast shear-wave polarization and station-fault distance. This likely reflects the complex tectonic and geologic settings near and within the San Andreas fault zone. Furthermore, several studies have noted variations in both fast shear-wave polarization and delay time with hypocenter location, which also implies heterogeneous physical properties between the fault zones and surrounding areas (Zinke and Zoback, 2000). Spatial variations in fast shear-wave polarizations and delay times near the San Andreas fault zone have been mapped in detail by back- projecting SWS measurements along ray paths (Liu et al., 2008) and using shear-wave splitting tomography (Zhang et al., 2007). A similar analysis was not conducted here given the limited hypocenter depth distribution and poor raypath coverage between the five stations. The shear-wave splitting measurements from this study, Peng and Ben- Zion (2004) and numerous studies from the San Andreas fault system indicate spatially-variable anisotropic distributions, suggesting that the physical properties of the upper crust near major fault zones are different from those within major fault zones.

2.6 Conclusions

We perform a shear-wave splitting analysis using seismograms of aftershocks following the 1999 Izmit MW 7.4 earthquake to delineate between stress- and structure-controlled anisotropic mechanisms in the upper crust beneath the western North Anatolian fault zone. The aftershocks were recorded by five seismic stations located within different geological and tectonic settings along the Izmit surface rupture. Our results indicate it is occasionally possible to delineate between stress- and structure-controlled mechanisms, and that there is a slight correlation between the dominant fast shear-wave polarization and station-fault distance. Three of the four stations located greater than 1 km from the Izmit surface rupture exhibit dominant fast shear-wave polarizations parallel or subparallel to the local trend of SHmax, which we interpret as representing a stress-controlled anisotropic mechanism such as extensive dilatancy anisotropy. The only station within 1 km of the NAFZ has a bimodal dominant fast shear-wave polarization reflecting both fault-parallel and fault- perpendicular modes. We conclude that the fault-parallel mode reflects a structure- controlled anisotropic mechanism such as fault-aligned macroscopic features associated with the Mudurnu fault. The lack of correlation between shear-wave delay times and hypocenter depths indicates that the anisotropic intervals are generally constrained to the upper 8 km of the crust. The fast shear-wave polarizations and delay times suggest a heterogeneous anisotropic distribution within the recording array and that the physical properties of the upper crust vary considerably between the NAFZ and surrounding area. Although beyond the scope of this study, performing a similar shear-wave splitting analysis using contemporary seismicity may lend insight into the physical properties of the upper crust both during quiescent intervals between major ruptures and immediately following a major rupture.

2.7 References

Alford, R.M., 1986, Shear data in the presence of azimuthal anisotropy: Dilley, Texas: Annual International Meeting of the Society of Exploration Geophysicists, Expanded Abstracts 56 476- 479.

Ambraseys, N.N., and Zatopek, A., 1969, The Mudurnu Valley, west Anatolia, Turkey earthquake of 22 July 1967: Bulletin of the Seismological Society of America, 69, 521-589. Barka, A., Akyüz, H.S., Altunel, E., Sunal, G., Çakir, Z., Dikbas, A., Yerli, B., Rockwell, T., Dolan, J.R., Dawson, T., Hartleb, R., Tucker, A., Fumal, T., Langridge, R., Stenner, H., Christofferson, S., Armijo, R., Meyer, B., and de Chabalier, J.B., 1999, 17 August 1999 Izmit earthquake, northwestern Turkey: EOS, 80, F647. Barka, A., Akyüz, H.S., Altunel, E., Sunal, G., Çakir, Z., Dikbas, A., Yerli, B., Armijo, R., Meyer, B., de Chabalier, J.B., Rockwell, T., Dolan, J.R., Hartleb, R., Dawson, T., Christofferson, S., Tucker, A., Fumal, T., Langridge, R., Stenner, H., Lettis, W., Bachhuber, J., and Page, W., 2002, The surface rupture and slip distribution of the 17 August 1999 Izmit earthquake (M 7.4), North Anatolian Fault: Bulletin of the Seismological Society of America, 92, 43-60. Baumbach, M., Bindi, D., Grosser, H., Milkereit, C., Parolai, S., Wang, R., Karakisa, S., Zünbül, S., and Zschau, J., 2003, Calibration of an ML scale in northwestern Turkey from 1999 Izmit aftershocks: Bulletin of the Seismological Society of America, 93, 2289-2295. Ben-Zion, Y., Peng, Z., Okaya, D., Seeber, L., Armbruster, J.G., Ozer, N., Michael, A.J., Baris, S., and Aktar, M., 2003, A shallow fault-zone structure illuminated by trapped waves in the Karadere- Duzce branch of the North Anatolian Fault, western Turkey: Geophysical Journal International, 152, 699-717. Bohnhoff, M., Grosser, H., and Dresen, G., 2006, Strain partitioning and stress rotation at the North Anatolian Fault Zone from aftershock focal mechanisms of the 1999 Izmit Mw=7.4 earthquake: Geophysical Journal International, 166, 373-385. Boness N.L., and Zoback, M.D., 2004, Stress-induced seismic velocity anisotropy and physical properties in the SAFOD Pilot Hole in Parkfield, CA: Geophysical Research Letters 31, L15S17. Boness, N.L., and Zoback, M.D., 2006, Mapping stress and structurally controlled crustal shear velocity anisotropy in California: Geology, 34, 825-828. Bouchon, M., Toksöz, M.N., Karabulut, H., Bouin, M.-P., Dietrich, M., Aktar, M., and Edie, M., 2002, Space and time evolution of rupture and faulting during the 1999 Izmit (Turkey) earthquake: Bulletin of the Seismological Society of America, 92, 256-266. Bouchon, M., and Karabulut, H., 2008, The aftershock signature of supershear earthquakes: Science, 320, 1323-1325. Bowman, J.R., and Ando, M., 1987, Shear-wave splitting in the upper-mantle wedge above the Tonga subduction zone: Geophysical Journal International, 88, 25-41. Brocher, T.M., and Christensen, N.I., 1990, Seismic anisotropy due to preferred mineral orientation observed in shallow crustal rocks in southern Alaska: Geology, 18, 737-740. Bulut, F., Bohnhoff, M., Aktar, M., and Dresen, G., 2007, Characterization of aftershock-fault plane orientations of the 1999 Izmit (Turkey) earthquake using high-resolution aftershock locations: Geophysical Research Letters, 34, L20306. Cochran, E.S., Vidale, J.E., and Yang, Y.-G., 2003, Near-fault anisotropy following the Hector Mine earthquake: Journal of Geophysical Research, 108, B9. Cochran, E.S., Li, Y.-G., and Vidale, J.E., 2006, Anisotropy in the shallow crust observed around the San Andreas Fault before and after the 2004 M 6.0 Parkfield earthquake: Bulletin of the Seismological Society of America, 96, S364-S375. Crampin, S., 1991, Wave propagation through fluid-filled inclusions of various shapes: Interpretation of extensive-dilatancy anisotropy: Geophysical Journal International, 104, 611-623. Crampin, S., and Booth, D.C., 1985, Shear-wave polarizations near the North Anatolian Fault – II. Interpretation in terms of crack-induced anisotropy: Geophysical Journal of the Royal Astronomical Society of London, 83, 75-92. Crampin, S., and Peacock, S., 2008, A review of the current understanding of seismic shear-wave splitting in the Earth’s crust and common fallacies in interpretation: Wave Motion, 45, 675-722. Delouis, B., Giardini, D., Lundgren, P., and Salichon, J., 2002, Joint inversion of InSAR, GPS, teleseismic, and strong-motion data for the spatial and temporal distribution of earthquake slip: Application to the 1999 Izmit mainshock: Bulletin of the Seismological Society of America, 92, 278-299.

Görgün, E., Bohnhoff, M., Bulut, F., and Dresen, G., 2010, Seismotectonic setting of the Karadere- Düzce branch of the North Anatolian Fault Zone between the 1999 Izmit and Düzce ruptures from analysis of Izmit aftershock focal mechanisms: Tectonophysics, 482, 170-181. Hartleb, R.D., Dolan, J.F., Akyüz, H.S., Dawson, T.E., Tucker, A.Z., Yerli, B., Rockwell, T.K., Toraman, E., Çakir, Z., Dikbaş, A., and Altunel, E., 2002, Surface rupture and slip distribution along the Karadere segment of the 17 August 1999 Izmit and the western section of the 12 November 1999 Düzce, Turkey, earthquakes: Bulletin of the Seismological Society of America, 92, 67-78. Heidbach, O., Tingay, M., Barth, A., Reinecker, J., Kurfeß, D., and Müller, B., 2008, The world stress map database release 2008. http://www.world-stress-map.org. Last accessed August 2010. Kern, H., and Wenk, H.-R., 1990, Fabric-related velocity anisotropy and shear wave splitting in rocks from the Santa Rosa mylonite zone, California: Journal of Geophysical Research, 95, 11213-11223. Kiratzi, A.A., 2002, Stress tensor inversions along the westernmost North Anatolian Fault Zone and its continuation into the north Aegean Sea: Geophysical Journal International, 151, 360-376. Leary, P.C., Crampin, S., and McEvilly, T.V., 1990, Seismic fracture anisotropy in the earth’s crust: An overview: Journal of Geophysical Research, 95, 11105-11114. Li, Y.-G., Teng, T.-L., and Henyey, T.L., 1994, Shear-wave splitting observations in the northern Los Angeles Basin, southern California: Bulletin of the Seismological Society of America, 84, 307-323. Liu, Y., Teng, T.-L., and Ben-Zion, Y., 2004, Systematic analysis of shear-wave splitting in the aftershock zone of the 1999 Chi-Chi, Taiwan, earthquake: Shallow crustal anisotropy and lack of precursory variations: Bulletin of the Seismological Society of America, 94, 2330-2347. Liu, Y., Zhang, H., Thurber, C., and Roecker, S., 2008, Shear wave anisotropy in the crust around the San Andreas Fault near Parkfield: Spatial and temporal analysis: Geophysical Journal International, 172, 957-970. McClusky, S., Balassanian, S., Barka, A., Demir, C., Ergintav, S., Georgiev, I., Gurkan, O., Hamburger, M., Hurst, K., Kahle, H., Kastens, K., Kekelidze, G., King, R., Kotzev, V., Lenk, O., Mahmoud, S., Mishin, A., Nadariya, M., Ouzounis, A., Paradissis, D., Peter, Y., Prilepin, M., Reilinger, R., Sanli, I., Seeger, H., Tealeb, A., Toksöz, M.N., and Veis, G., 2000, Global Positioning System constraints on plate kinematics and dynamics in the eastern Mediterranean and Caucasus: Journal of Geophysical Research, 105, 5695-5719. Milkereit, C., Zünbül, S., Karakisa, S., Iravul, Y., Zschau, J., Baumbach, M., Grosser, H., Günther, E., Umutlu, N., Kuru, T., Erkul, E., Klinge, K., von Seht, M., and Karahan, A., 2000, Preliminary aftershock analysis of Mw = 7.4 Izmit and Mw = 7.1 Düzce earthquake in western Turkey, In: Barka, A., Kozaci, Ö., Akyüz, S., Altunel, E., eds., The 1999 Izmit and Düzce Earthquakes: Preliminary results, Istanbul Technical University, Istanbul, 179-187. Nuttli, O., 1961, The effect of the earth’s surface on the S-wave particle motion: Bulletin of the Seismological Society of America, 51, 237-246. Örgülü, G., and Aktar, M., 2001, Regional Moment Tensor inversion for strong aftershocks of the August 17, 1999 Izmit earthquake (Mw = 7.4): Geophysical Research Letters, 28, 371-374. Parsons, T., 2004, Recalculated probability of M >= 7 earthquakes beneath the Sea of Marmara, Turkey: Journal of Geophysical Research, 109, B05304. Peng, Z., and Ben-Zion, Y., 2004, Systematic analysis of crustal anisotropy along the Karadere-Düzce branch of the North Anatolian fault: Geophysical Journal International, 159, 253-274. Peng, Z., and Ben-Zion, Y., 2004, Spatiotemporal variations of the crustal anisotropy from similar events in aftershocks of the 1999 M7.4 Izmit and M7.1 Düzce, Turkey, earthquake sequences: Geophysical Journal International, 160, 1027-1043. Pucci, S., Pantosti, D., Barchi, M.R., and Palyvos, N., 2007, A complex seismogenic shear zone: the Düzce segment of the North Anatolian Fault (Turkey): Earth and Planetary Science Letters, 262, 185-203. Reilinger, R.E., Ergintav, S., Bürgmann, R., McClusky, S., Lenk, O., Barka, A., Gurkan, O., Hearn, L., Feigl, K.L., Cakmak, R., Aktug, B., Ozener, H., and Töksoz, M.N., 2000, Coseismic and postseismic fault slip for the 17 August 1999, M = 7.5, Izmit, Turkey earthquake: Science, 289, 1519-1524.

Ryan, W.B.F., Carbotte, S.M., Coplan, J., O’Hara, S., Melkonian, A., Arko, R., Weissel, R.A., Ferrini, V., Goodwillie, A., Nitsche, F., Bonczkowski, J., and Zemsky, R., 2009, Global Multi-Resolution Topography (GMRT) synthesis data set: Geochemistry Geophysics Geosystems, 10, Q03014. Savage, M.K., 1999, Seismic anisotropy and mantle deformation: what have we learned from shear wave splitting?: Reviews in Geophysics, 37, 65-106. Savage, M. K., and P. G. Silver (1993), Mantle deformation and tectonics: constraints from seismic anisotropy in the western United States, Phys. Earth Planet. In. 78 207-227. Sayers, C.M., 1994, The elastic anisotropy of shales: Journal of Geophysical Research, 99, 767-774. Seeber, L., Armbruster, J.G., Ozer, N., Aktar, M., Baris, S., Okaya, D., Ben-Zion, Y., and Field, N., 2000, The 1999 earthquake sequence along the North Anatolian transform at the juncture between the two main ruptures, In: Barka, A., Kozaci, Ö., Akyüz, S., Altunel, E., eds., The 1999 Izmit and Düzce Earthquakes: Preliminary results, Istanbul Technical University, Istanbul, 209-223. Tadokoro, K., Ando, M., Bariş, S., Nishigami, K., Nakamura, M., Üçer, S.B., Ito, A., Honkura, Y., and Işikara, A.M., 2002, Monitoring of fault healing after the 1999 Kocaeli, Turkey, earthquake: Journal of Seismology, 6, 411-417. Tibi, R., Bock, G., Xia, Y., Baumbach, M., Grosser, H., Milkereit, C., Karakisa, S., Zünbül, S., Kind, R., and Zschau, J., 2001, Rupture processes of the 1999 August 17 Izmit and November 12 Düzce (Turkey) earthquakes: Geophysical Journal International, 144, F1-F7. Wüstefeld, A., and Bökelmann, G., 2007, Null detection in shear-wave splitting measurements: Bulletin of the Seismological Society of America, 97, 1204-1211. Yardley, G.S., and Crampin, S., 1991, Extensive-dilatancy anisotropy: Relative information in VSPs and reflection surveys: Geophysical Prospecting, 39, 337-355. Zhang, H., Liu, Y., Thurber, C., and Roecker, S., 2007, Three-dimensional shear-wave splitting tomography in the Parkfield, California, region: Geophysical Research Letters, 34, L24308. Zhang Z., and Schwartz, S.Y., 1994, Seismic anisotropy in the shallow crust of the Loma Prieta segment of the San Andreas fault system: Journal of Geophysical Research, 99, 9651-9661. Zinke, J.C., and Zoback, M.D., 2000, Structure-related and stress-induced shear-wave velocity anisotropy: Observations from microearthquakes near the Calaveras Fault in central California: Bulletin of the Seismological Society of America, 90, 1305-1312.

Chapter 3

INTRAPLATE EARTHQUAKES, REGIONAL STRESS, AND FAULT MECHANICS IN THE CENTRAL AND EASTERN U.S. AND SOUTHEASTERN CANADA2

Abstract

Utilizing 75 high quality individual earthquake focal plane mechanisms and 10 formal stress inversions we investigate the consistency of regional stress orientations in the central and eastern United States and southeastern Canada, the variation of relative stress magnitudes across the region, and the compatibility of slip on preferred nodal planes with frictional faulting theory. To map faulting styles and relative stress magnitudes across the region of study, we utilize the high quality focal plane mechanisms to calculate the A parameter (following Angelier, 1979; Simpson, 1997) that ranges from 0 (uniform horizontal extension with SV >> SHmax = Shmin) to 1.5

2 This chapter is published in: Hurd, O., and Zoback, M.D., 2012, Intraplate Earthquakes, Regional Stress and Fault Mechanics in the Central and Eastern U.S. and Southeastern Canada: Tectonophysics, doi: 10.1016/j.tecto.2012.04.002 40

(strike-slip faulting with SHmax > SV > Shmin) to 3 (uniform horizontal compression with

SHmax = Shmin > SV). We find that horizontal stresses become increasingly more compressive with respect to the vertical stress from the south-central United States (characterized predominantly by strike-slip focal mechanisms) toward the northeastern U.S. and southeastern Canada (predominantly thrust mechanisms). In a manner similar to the study by M.L. Zoback (1992a), which used a much smaller data set, we utilize the Mohr-Coulomb criterion to calculate the difference in orientation between the theoretically-optimal orientation of a fault plane (for various coefficients of friction, ) and the focal mechanism nodal planes assuming that pore pressure in the brittle crust is hydrostatic. For the 75 focal plane mechanisms utilized in our study, the preferred (better fitting) nodal planes deviate on average only 7° in strike and dip from the theoretically-optimal planes for µ = 0.6. As such minor differences could represent small variations in the stress field (or uncertainties in the focal plane mechanisms), we conclude that nearly all earthquakes in the study region slip in a manner compatible with shear failure on pre-existing faults in the local stress field.

3.1 Introduction

Significant amounts of seismicity occur in intraplate regions throughout the world, often on tectonic structures such as pre-existing fault zones, sometimes associated with failed rifts, and ancient suture zones (e.g. Sykes, 1978). Intraplate seismicity in North America is frequently correlated with pre-existing faults which are optimally-oriented for reactivation in the current stress field (e.g., Zoback and Zoback, 1981; Zoback, 1992a). The stress field in the central and eastern United States (CEUS) and southeastern Canada is remarkably consistent on the lateral scale of 100s of kilometers and is generally characterized by a horizontal, compressive, NE-SW trending maximum horizontal stress (e.g. Sbar and Sykes, 1973; Zoback and Zoback, 1980, 1991) thought to derive from buoyancy-driven forces such as ridge push (see Zoback and Zoback (2007) for review) or from geoid perturbations and mantle thermal anomalies (Davies, 1999).

Second order stress fields, some of which may deviate from the large-scale regional field described above, are also observed across the CEUS. These stresses are generally driven by more localized buoyancy forces related to processes such as sediment loading and deglaciation or the presence of lateral lithospheric heterogeneities (e.g., Zoback and Mooney, 2003). The stresses generated by these processes may also contribute to the nucleation of intraplate seismicity in the CEUS and southeastern Canada. Since earthquakes are a direct result of stresses acting within the crust, analyzing seismicity in intraplate regions may yield valuable information regarding the current state of stress and physical conditions of the upper crust (pore pressure, fault friction) that is often unavailable from other sources. This information is essential to addressing potential seismic hazards in intraplate regions.

Earthquake focal plane mechanisms are often used to estimate the orientation of the three principal stresses (vertical stress, SV, maximum horizontal stress, SHmax, and minimum horizontal stress, Shmin) in the crust. The P-axis of the focal mechanism, which is defined as the bisector of the dilatational quadrants, is generally taken to represent the approximate orientation of SHmax, although it could significantly deviate from the true SHmax orientation in the absence of friction (McKenzie, 1969). In contrast to SHmax orientations estimated from individual focal mechanisms, a formal stress inversion of multiple earthquake focal mechanisms directly estimates the orientation of the three principal stresses and provides a more accurate SHmax orientation than the P-axis of an individual focal mechanism (Angelier, 1979; Gephart and Forsyth, 1984; Michael, 1984). The inversion procedure assumes a uniform stress field over the crustal volume containing all focal mechanisms used in the inversion and that shear slip occurs in the direction of maximum resolved shear stress (Bott, 1959).

Since earthquake focal mechanisms are the basis of this analysis, it is crucial that they be well-constrained. In general, earthquake focal plane mechanisms are obtained from body-wave first-motions and polarizations (e.g. Khattri, 1973), body- wave amplitude ratios (e.g. Kisslinger et al., 1981), waveform modeling (e.g. Nábělek,

1984) or a combination of these methods. While the quality of an individual focal mechanism depends on the recording array geometry, seismogram signal-to-noise ratio and the accuracy of the earth velocity model, certain constraints generally yield higher quality and more reliable solutions. For example, because waveform modeling uses body-wave amplitude information and searches over a broader coverage of the focal sphere for a solution, it is often more powerful for constraining fault orientations than a focal mechanism created solely from P-wave polarities (e.g. Lay and Wallace, 1995). Solutions constrained by only P-wave polarities, for instance, may have several distinctly different nodal plane pairs (and slip configurations) that fit the data equally well and are highly dependent on recording array geometry. Consequently, we only consider high quality individual focal mechanisms constrained by waveform modeling in this study.

We compile well-constrained focal mechanisms and formal stress inversions from the CEUS and southeastern Canada over the past ~20 years. We utilize these data to investigate the consistency of regional stress orientations, to map faulting styles and relative stress magnitudes across the region and to investigate the likelihood of shear failure on the more well-oriented nodal planes in the local stress field in the context of frictional faulting theory, in a manner analogous to M.L. Zoback (1992a) who worked with a much smaller data set.

3.2 Data Collection

All individual focal mechanisms and focal mechanism inversions are compiled from publications and earthquake catalogs over the past ~20 years. Since the individual focal mechanisms will directly be used to calculate relative stress magnitudes and examine slip compatibility in our analysis, it is crucial that the mechanisms be well constrained. To ensure such quality, we only select mechanisms constrained by waveform modeling. Again, waveform modeling techniques provide a better constraint on fault orientations because they use a broader coverage of the focal sphere along with relative body-wave amplitudes to constrain solutions.

The study area includes the CEUS, with the western boundary corresponding roughly to the 105˚W line of longitude (to exclude extensional stress regimes such as the Rio Grande Rift and Basin and Range province), and southeastern Canada. A total of 52 individual focal mechanisms and 10 stress inversions (from Mazzotti and Townend, 2010) are compiled (Appendices 3A and 3B, respectively). Of the 52 new focal mechanisms, 24 indicate thrust faulting, 25 are strike-slip and 3 represent normal faulting regimes, as determined using criteria in Zoback, 1992b. All focal mechanisms have magnitudes greater than MW = 3.1 with the maximum magnitude being MW = 5.2. The Canadian earthquakes range in depth from 2 to 25 km with an average depth of 14.1 km compared to a depth range of 2 to 18 km with an average of 8.0 km for the CEUS earthquakes. We also include 23 of the focal mechanisms analyzed by Zoback (1992a) within this study area (Appendix 3C). In instances where a precise latitude and longitude location are not available for a data point, a location is estimated using the original data source.

3.3 Defining Stress Orientations and Relative Stress Magnitudes

3.3.1 Stress orientations The first objective in our analysis is to investigate the consistency of the maximum horizontal principal stress (SHmax) orientation throughout the study area as inferred from formal stress inversions and P-axes of newly compiled individual focal mechanisms. Figure 3.1 illustrates the new data points overlain on the 2008 World Stress Map (WSM) database (Heidbach et al., 2008), which is essentially identical to the database used by Zoback (1992b). In general, SHmax orientations inferred from the new focal mechanisms (shown by blue bars on black and white mechanisms) as well as the stress inversions (dark green circles with dark green bars) are consistent with the

NE-SW SHmax orientation seen over much of the CEUS and southeastern Canada. Moreover, the new data points are locally consistent with pre-existing data which often show slight variations from the regional stress orientation.

Figure 3.1: Stress indicators in the CEUS and SE Canada. Map includes the 52 newly-compiled focal mechanisms (black and white mechanisms with blue bars), 10 stress inversions (dark green circles with dark green bars) and 23 focal mechanisms from Zoback (1992a) (gray mechanisms) overlain on the 2008 World Stress Map. Bars on focal mechanisms and stress inversions represent the approximate and estimated orientation of SHmax, respectively.

In contrast to the broadly homogeneous SHmax orientation, several focal mechanisms and stress inversions appear to indicate locally variable SHmax orientations. For example, the stress inversion in central Virginia yields a SHmax orientation of 90˚, which is a roughly 45˚ clockwise rotation from stress indicators just to the west (Figure 3.1). Similarly, the six individual focal mechanisms in the Wabash Valley seismic zone in southern Illinois have an average P-axis orientation of 77˚, which is relatively consistent with the regional SHmax direction, but differs from the local E-W SHmax orientation indicated by nearby breakout stress indicators in western Kentucky and the focal mechanism inversion in the New Madrid seismic zone in NE

Arkansas. Finally, four of the five new data points in the Charlevoix seismic zone and both new focal mechanisms (and the stress inversion) in the St. Lawrence seismic zone also display a significant clockwise SHmax rotation from the regional trend as inferred from nearby borehole breakout measurements.

3.3.2 Relative stress magnitudes The second objective of this study is to estimate the relative magnitudes of the three principal stresses at hypocentral depths. First, we estimate the local SHmax orientation near each earthquake from independent stress measurements in the WSM database. SHmax is estimated by averaging the SHmax orientation from the three nearest data points in the WSM, regardless of type. If the standard deviation of the average is greater than 25°, the average of the two nearest ‘A’ quality stress measurements is used. For the 52 earthquakes, the two nearest ‘A’ quality stress measurements are usually from either borehole breakouts or hydraulic fractures. Next, to constrain the orientations of the remaining principal stresses Shmin and SV, we assume that the three principal stresses are perpendicular to one other and oriented horizontally and vertically (Zoback and Zoback, 1980). In Figure 2 of Mazzotti and Townend (2010), it is clear that one principal stress is near vertical in each of the ten areas in the CEUS where focal mechanism inversions were carried out, suggesting that this is a valid assumption.

With the stress orientations constrained, the relative magnitudes of the three principal stresses are then calculated. For SV, we assume a regional lithostatic gradient of 25 MPa/km, which corresponds to an overburden density of 2,500 kg/m3. Although rock densities increase with depth, and a higher gradient (27-28 MPa/km) may be more appropriate for the earthquakes of greater depth, we use the 25 MPa/km gradient since the majority of earthquakes examined in this study fall within the upper crust. More importantly, since only relative principal stress magnitudes are calculated, changing the overburden gradient does not affect the calculations. The remaining principal stresses are then solved for using two physical constraints. First, the relationship 46

S  S   2 3 S  S 1 3 (3.1)

(after Angelier, 1979), where S1, S2, and S3 represent the three principal stresses in order of decreasing magnitude, places constraints on the potential orientation of slip vectors on the nodal planes. If slip on a nodal plane is geometrically compatible with the local stress field, Φ must fall between 0 and 1 for a given faulting regime. Following the technique of Gephart (1985), Φ is calculated from the orientations of the two focal mechanism nodal planes and the three principal stresses using the following relationship

    1    13 23   33 23     12 22 32 22 (3.2)

where ßij corresponds to a matrix of angle cosines relating the principal stress and focal mechanism coordinate systems.

A second physical constraint on relative stress magnitudes after Jaeger and Cook, (1979) is

S  P 1 P  [( 2 1)1/ 2  ]2 S  P 3 P (3.3)

where PP is the pore pressure and μ is the coefficient of fault friction. For given values of PP and μ, the differential stress magnitudes cannot exceed the stress required to cause shear failure on pre-existing, optimally-oriented faults in the brittle crust. This constraint will be utilized in the next section to evaluate the consistency of each of the focal mechanism nodal planes with frictional faulting theory for reasonable values of

PP and µ.

Since Φ provides a measure of the magnitude of S2 relative to the maximum

(S1) and minimum (S3) principal stresses, it can be used to map relative stress magnitudes, and therefore faulting styles, across the study area. Following Simpson

(1997), we use the Φ values and faulting regimes of each focal mechanism to calculate the AΦ parameter, which scales relative stress magnitudes from 0 to 3 based on faulting style. The relationship is given by:

n A  n  0.5  (1)   0.5 (3.4) where Φ is calculated in (3.2) and n = 0, 1, and 2 for normal, strike-slip, and reverse faulting types, respectively.

A total of 85 AΦ data points were determined; 52 from focal mechanisms in this study, 10 from stress inversions in this study, and 23 from focal mechanisms in Zoback (1992a). The AΦ values are shown spatially in Figure 3.2. Physically, an AΦ value of 0 represents uniform horizontal extension (SV >> SHmax = Shmin), 1.5 represents strike-slip faulting (SHmax > SV > Shmin) and 3 indicates uniform horizontal compression (SHmax = Shmin >> SV). The AΦ distributions illustrate that the horizontal principal stresses become increasingly compressive with respect to the vertical stress moving from the south-central U.S. to the northeastern U.S. and southeastern Canada.

Figure 3.2: Spatial variation of the A parameter across the study area. Horizontal stresses become increasingly compressive (A becomes larger in value) with respect to the vertical stress moving from the south-central U.S. to the northeastern U.S. and southeastern Canada. Values are interpolated using a bilinear interpolation scheme, and extrapolated linearly to the boundaries of the map. Background seismicity (black dots) is from the USGS/NEIC catalog 1973-2010.

3.4 Slip Compatibility in the Regional Stress Field

Our final objective in analyzing the newly-compiled data set is to assess the proximity of each nodal plane in orientation to that expected for shear failure in the local stress field in the context of Mohr-Coulomb failure criterion. We assume PP is hydrostatic in the brittle crust (following Zoback and Townend, 2001) and μ is

consistent with laboratory values determined by Byerlee (1978), who demonstrated that a wide variety of rock types exhibit a coefficient of friction between 0.6 and 1.0 over a wide range of confining pressures. However, to include the possibility that some intraplate faults might have unusually low frictional strength, we evaluate the consistency of slip with the theoretically-optimal planes for values of µ as low as 0.2.

Thus, for a given stress orientation and value of Φ, µ, and PP, we determine which of the two nodal planes is more optimally-oriented for shear failure. In other words, our goal is to determine which focal mechanism nodal plane in each pair is closest to the theoretically-optimal orientation for failure assuming hydrostatic PP and µ consistent with laboratory-derived friction values from Byerlee (1978).

A grid search method is utilized to find the most optimally-oriented planes in the local stress field. For each focal mechanism nodal plane pair, the strike on both planes is simultaneously varied from the observed strike by up to  45°. The nodal plane dips are also varied from the observed dip by up to 45° while applying the constraint that dips must remain in the range 0-90°. At each strike and dip iteration, the value of μ to fit the observed slip is calculated using the Mohr-Coulomb failure criterion assuming hydrostatic PP. Figure 3.3 illustrates an example μ map for one of the analyzed earthquakes. The black dots represent the orientations of the preferred (left) and auxiliary (right) nodal planes which were identified based on which plane best fits the assumption of Mohr-Coulomb failure for values of friction between 0.6 and 0.8. Test plane configurations where slip is frictionally impossible in the current stress field are indicated by hatched areas. In the example shown in Figure 3.3, the preferred nodal plane is essentially perfectly oriented for shear failure for a coefficient of friction of about 0.6-0.7. The auxiliary plane would have to be rotated by about 15- 20° in strike to be consistent with laboratory-derived friction values.

Figure 3.3: Example fault friction (μ) map for a single focal mechanism. Black dots represent the orientations of the two nodal planes. Color indicates the μ value required to cause shear failure on a plane with the corresponding strike and dip in the local stress field assuming hydrostatic Pp. Test plane configurations where slip is frictionally impossible in the current stress field are indicated by hatched areas. Plane 1 is the preferred nodal plane as it is closer to the theoretically-optimal orientation for μ = 0.6. Event location: NW Texas, 35.49° N, 102.65°W Date: 2/10/2010 Depth: 13 km Regime: strike-slip SHmax: N 109° E.

Figure 3.4a displays histograms of the difference in strike and dip between the preferred nodal plane orientation and nearest theoretically-expected nodal plane orientation for μ = 0.6 for all 75 earthquake focal mechanisms considered in this study. The results indicate that overall the orientation of the preferred nodal planes is quite consistent with the theoretically-optimal orientation for μ = 0.6. The mean mis-fit is only ~7° in strike and dip, which is well within the range of uncertainty associated with the stress orientations and nodal plane determinations. Figure 3.4b shows the orientation difference for the conjugate nodal plane for all events, which fit much more poorly. Finally, Figure 3.5 shows the mis-fit of the preferred plane in strike and dip with a theoretically-optimal plane for assumed friction values of 0.2, 0.6 and 0.8. Note that a coefficient of friction of 0.6 is much more consistent with the observations than either the higher or lower friction values.

Figure 3.4: Histograms showing the mis-fit in strike and dip between (a) the preferred and (b) the conjugate nodal planes and the theoretically optimal (μ = 0.6) fault plane for all 75 focal plane mechanisms. Most preferred nodal planes strike and dip within 10° of the theoretically optimally- oriented fault plane.

Figure 3.5: Mis-fit in strike and dip between the preferred nodal planes and the nearest nodal planes that fail with μ = 0.2, μ = 0.6, and μ = 0.8. A coefficient of friction of 0.6 is much more consistent with the observations than higher or lower friction values.

3.5 Discussion

3.5.1 The stress field in the central and eastern U.S. In agreement with previous observations, the newly compiled focal mechanisms and stress inversions suggest a highly consistent NE-SW SHmax orientation throughout the CEUS and southeastern Canada (Figure 3.1). Such large- scale uniform stress fields are typically thought to result from buoyancy-driven forces such as ridge push and internal density heterogeneities in the lithosphere (Zoback and Zoback, 2007) or from geoid perturbations and mantle thermal anomalies (Davies, 1999). The central Virginia, Charlevoix, St. Lawrence, and New Madrid seismic zones all contain evidence for local rotations of SHmax from this general trend. Note that the stress rotations within these seismic zones are frequently supported by numerous individual focal mechanism stress indicators occurring on different faults over a variety of depths. Stresses in the New Madrid seismic zone are addressed in detail in Chapter 4.

Many of these second-order stress orientations have been observed for several decades, and the physical processes generating such seismicity may include buoyancy- driven forces from deglaciation or sediment loading and lower crustal heterogeneities. Baird et al. (2010), in using 3D numerical modeling techniques to predict spatial locations of seismicity in the Charlevoix seismic zone, illustrated the importance of a detailed structural understanding of ancient fault zones and how slip on pre-existing structures may potentially modify the local stresses, and therefore the seismicity distribution and faulting type (see also Mazzotti and Townend, 2010).

3.5.2 Relative stress magnitudes and faulting styles The AΦ parameter is used to map relative stress magnitudes and faulting styles across the study area. AΦ distributions indicate a clear contrast between primarily thrust faulting mechanisms in southeastern Canada and the northeastern United States

and dominantly strike-slip faulting mechanisms moving toward the south-central United States (Figures 3.1 and 3.2). In other words, the horizontal stresses become increasingly compressive with respect to the vertical stress moving from the south- central to the northeast U.S. and southeastern Canada. One proposed mechanism for relative principal stress contrasts has been the superposition of stresses in relation to unloading of a massive Pleistocene ice sheet (e.g. Clark, 1982; James, 1991; James and Bent, 1994; Stein et al., 1979; Wu and Hasegawa, 1996; Wu and Johnson, 2000; Wu and Mazzotti, 2007). These models typically assumed a disk-shaped load applied on a layered earth model with either elastic or viscous lithosphere properties and generally matched the contrast in relative stress magnitudes in a qualitative sense. However, as Zoback (1992a) noted, glacial rebound models are often inconsistent with the observed sense of relative stress contrasts between southeastern Canada and the eastern United States and produce stress perturbations that are too small to account for the observed stress change at seismogenic depths when superimposed on the ambient stress field. Zoback and Mooney (2003) discussed the possibility that relatively high compression in the northeastern U.S. and southeastern Canada might be related to negative buoyancy effects associated with relatively high density in the mantle lithosphere which “pulls down” on the crust and increases compression.

Baird et al. (2010) noted that faulting in southeastern Canada may be related to the orientation of paleotectonic rift structures with respect to the modern day regional stress field. For example, many seismic zones in southeastern Canada fall along preexisting NW-SE trending structures, such as the Ottawa and Saguenay grabens, which are perpendicular to the orientation of SHmax and thus more likely for reactivation through thrust faulting. Conversely, strike-slip faulting in the CEUS may result from a general NE-SW trend of ancient rift structures combined with a slightly rotated ENE-

WSW SHmax orientation, which makes the structures more favorable for reactivation in a strike-slip sense.

The analysis used to examine relative stress magnitudes and faulting styles in this study could be extended to other continental regions where a relatively small set

(20-40) of well-constrained and well-distributed focal mechanisms is available. Western Europe, China, Central Asia, and NW South America all represent regions of extensive seismic activity, and would perhaps be the most feasible candidates for a similar study. The AΦ parameter in particular may help illuminate spatial transitions between a range of faulting types in structurally and tectonically complex regions.

3.5.3 Slip compatibility and fault friction For each nodal plane pair for all 75 earthquakes, we select one nodal plane as being preferentially-oriented for shear failure in the local stress field on the basis of its proximity to the nearest plane compatible with Mohr-Coulomb failure with μ = 0.6. The vast majority of these preferred nodal planes are within 7° in strike and dip from a fault plane that fails with μ = 0.6 (Figure 3.4a), and we interpret these planes to be generally compatible with shear failure in the local stress field. We interpret the results in terms of a rotated nodal plane pair about a stationary stress tensor, although since we assume the three principal stresses lie in vertical and horizontal planes the analysis is equivalent to rotating a stress tensor about fixed nodal planes. Regardless of reference frame, only small perturbations are required for the preferred nodal planes to be optimally-oriented for shear failure in the local stress field.

We consider coefficients of friction (μ) between 0.6 and 0.8 for our analysis based on several lines of evidence. First, Byerlee (1978) demonstrated from laboratory experiments on a wide variety of rock types over a range of confining pressures that μ generally takes a value between 0.6 and 1.0, although it may be lower in shaley rocks, which is not relevant to the earthquakes studied here. Second, in-situ stress measurements extending to as deep as ~9 km in the upper crust are regularly consistent with predicted stress magnitudes using Coulomb frictional-failure theory with 0.6 ≤ μ ≤ 1.0 (e.g. Figure 3.1 in Townend and Zoback, 2000). Thirdly, Sibson and Xie (1998) and Collettini and Sibson (2001) demonstrated using the Coulomb failure criterion that the dip range of active thrust and normal faults is consistent with fault reactivation assuming 0.6 ≤ μ ≤ 0.85 and principal stresses lying in horizontal and vertical planes. While their study considered only fault planes which produced 55

moderate to large earthquakes (MW > 5.5), which is notably larger than the majority of earthquakes examined in this study, they also support our prescribing laboratory- consistent friction coefficients to seismogenic faults in the upper crust.

Gudmundsson et al. (2010) demonstrated that variable physical properties within major fault zones, specifically within the damage zone and fault core, can affect local stress orientations and magnitudes, which may subsequently influence fracture propagation behavior. Our analysis directly examines whether or not local stress perturbations, anomalous fault friction, or elevated pore pressures are required to explain the observed slip on intraplate faults in a relatively uniformly oriented regional stress field. Specifically, we consider the slip compatibility of focal mechanism nodal planes with friction coefficients as low as 0.2 and as high as 0.8. The results demonstrate that slip on the vast majority of nodal planes is consistent with laboratory- derived friction coefficients assuming hydrostatic pore pressure in the brittle crust.

Our assumption of hydrostatic pore pressure is based on widespread observations of hydrostatic pore pressure persisting to as deep as 12 km in the upper crust (Table 1 in Townend and Zoback, 2000) and the consistency of hydrostatic pore pressure in the upper crust with maintaining observed lithospheric deformation rates in force-limited stress models (Zoback and Townend, 2001). While we acknowledge that faults can be conduits for fluid flow and elevated pore pressures, our results suggest that, in terms of the regional stress field, there is generally no reason to call on elevated pore pressure to explain the occurrence of intraplate earthquakes. Our slip compatibility results are consistent with the analysis of Zoback (1992a), and are in agreement with the hypothesis that the brittle crust is generally in a state of frictional failure equilibrium due to regional plate driving forces (Zoback et al, 2002) and local perturbations associated with variations of lithospheric density (Zoback and Mooney, 2003).

3.6 Conclusions

(1) Newly compiled stress data including 75 earthquake focal plane mechanisms and 10 formal stress inversions from the central and eastern United States and southeastern Canada indicate a highly consistent, compressional, NE-SW-oriented maximum horizontal stress across much of intraplate North America. The new data are consistent with many pre-existing stress measurements from a wide variety of stress indictors.

(2) Using the AΦ parameter calculated from the orientation of the focal mechanism nodal planes and the stress tensor at each earthquake location, we investigate the variation in relative stress magnitudes and faulting type across the study area. There is a clear transition from predominantly strike-slip faulting in the south-central U.S. to predominantly thrust faulting in the northeastern U.S. and southeastern Canada which reflects increasingly compressive (higher AΦ values) horizontal stresses with respect to the vertical stress moving from central to northeastern North America.

(3) Using Mohr-Coulomb failure criterion and assuming hydrostatic pore pressure, we find the vast majority of preferred focal mechanism nodal planes are consistent in orientation with optimally-oriented planes (µ = 0.6) for shear failure in the local stress field. This suggests that shear failure on the preferred nodal planes generally do not require reduced fault friction or elevated pore pressures.

3.7 References

Ammon, C.J., Herrmann, R.B., Langston, C.A., and Benz, H.M., 1998, Faulting parameters of the January 16, 1994 Wyomissing Hills, Pennsylvania earthquakes: Seismological Research Letters, 69, 261-269. Angelier, J., 1979, Determination of the mean principal directions of stresses for a given fault population: Tectonophysics, 56, T17-T26. Baird, A.F., McKinnon, S.D., and Godin, L., 2010, Relationship between structures, stress and seismicity in the Charlevoix seismic zone revealed by 3-D geomechanical models: Implications for the seismotectonics of continental interiors: Journal of Geophysical Research, 115, B11402. Bott, M.H.P., 1959, The mechanics of oblique slip faulting: Geological Magazine, 96, 109-117. Byerlee, J.D., 1978, Friction of rock: Pure and Applied Geophysics, 116, 615-626.

Clark, J.A., 1982, Glacial loading: A cause of natural fracturing and a control of the present stress state in regions of high Devonian shale gas: SPE paper 10798 presented at the SPE/DOE Unconventional Gas Recovery Symposium of the Society of Petroleum Engineers in Pittsburgh, PA., May 16-18, 1982. Collettini, C., and Sibson, R.H., 2001, Normal faults, normal friction?: Geology, 29, 927-930. Davies, G.F., 1999, Dynamic earth plates, plumes and mantle convection, Cambridge, UK, Cambridge University Press, 458 pp. Du, W.-X., Kim, W.-Y., and Sykes, L.R., 2003, Earthquake source parameters and state of stress for the northeastern United States and southeastern Canada from analysis of regional seismograms: Bulletin of the Seismological Society of America, 93, 1633-1648. Gephart, J.W., 1985, Principal stress directions and the ambiguity in fault plane identification from focal mechanisms: Bulletin of the Seismological Society of America, 75, 621-625. Gephart, J.W., and Forsyth, D.D., 1984, An improved method for determining the regional stress tensor using earthquake focal mechanism data: Application to the San Fernando earthquake sequence: Journal of Geophysical Research, 89, 9305-9320. Gudmundsson, A., Simmenes, T.H., Larsen, B., and Philipp S.L., 2010, Effects of internal structure and local stresses on fracture propagation, deflection, and arrest in fault zones: Journal of Structural Geology, 32, 1643-1655. Heidbach, O., Tingay, M., Barth, A., Reinecker, J., Kurfeß, D., and Müller, B., 2008, The world stress map database release 2008. http://www.world-stress-map.org, last accessed August 2010. Jaeger, J.C., and Cook, N.G.W., 1979, Fundamentals of Rock Mechanics, London, UK, Chapman and Hall, 475 pp. James, T.S., 1991, Post-glacial deformation, Ph.D. thesis. Princeton University Press, Princeton, N.J. 190 pp. James, T.S., and Bent, A.L., 1994, A comparison of eastern North American seismic strain-rates to glacial rebound strain-rates: Geophysical Research Letters, 21, 2127-2130. Khattri, K., 1973, Earthquake focal mechanism studies – A review: Earth-Science Reviews, 9, 19-63. Kim, W.-Y., 2003, The 18 June 2002, Caborn, Indiana, earthquake: Reactivation of ancient rift in the Wabash Valley seismic zone?: Bulletin of the Seismological Society of America, 93, 2201-2211. Kim, W.-Y., Dineva, S., Ma, S., and Eaton, D., 2006, The 4 August 2004, Lake Ontario, earthquake: Seismological Research Letters, 77, 65-73. Kisslinger, C., Bowman, J.R., and Koch, K., 1981, Procedures for computing focal mechanisms from local (SV/P) data: Bulletin of the Seismological Society of America, 71, 1179-1729. Lay, T., and Wallace, T.C., 1995, Modern Global Seismology, San Diego, Academic Press, 521 pp. Ma, S., Eaton, D.W., and Adams, J., 2008. Intraplate seismicity of a recently deglaciated shield terrane: A case study from northern Ontario, Canada: Bulletin of the Seismological Society of America, 98, 2828-2848. Mazzotti, S., and Townend, J., 2010, State of stress in central and eastern North American seismic zones: Lithosphere, 2, 76-83. McKenzie, D.P., 1969, The relation between fault plane solutions for earthquakes and the directions of the principal stresses: Bulletin of the Seismological Society of America, 59, 591-601. Michael, A.J., 1984, Determination of stress from slip data: Faults and folds: Journal of Geophysical Research, 89, 11517-11526. Nábělek, J.L., 1984, Determination of earthquake source parameters from inversion of body waves. Ph.D. Thesis, Massachusetts Institute of Technology, Cambridge, MA, 361 pp. Sbar, M.L., and Sykes, L.R., 1973, Contemporary compressive stress and seismicity in eastern North America: An example of intra-plate tectonics: Geological Society of America Bulletin, 84, 1861- 1882. Sibson, R.H., and Xie, G., 1998, Dip range for intracontinental reverse fault ruptures: Truth not stranger than friction?: Bulletin of the Seismological Society of America, 88, 1014-1022. Simpson, R.W., 1997, Quantifying Anderson’s fault types: Journal of Geophysical Research, 102, 17909-17919. Saint Louis University Earthquake Center, 2010, http://www.eas.slu.edu/Earthquake_Center/MECH.NA/index.html, Last accessed September 2010.

Stein, S., Sleep, N.H., Geller, R.J., Wang, S.-C., and Kroeger, G.C., 1979, Earthquakes along the passive margin of eastern Canada: Geophysical Research Letters, 6, 537-540. Street, R., Taylor, K., Jones, D., Harris, J., Steiner, G., Zekulin, A., and Zhang, D., 1993, The 4.6 mbLg northeastern Kentucky earthquake of September 7, 1988: Seismological Research Letters, 64, 187- 199. Sykes, L.R., 1978, Intraplate seismicity, reactivation of preexisting zones of weakness, alkaline magmatism, and other tectonism postdating continental fragmentation: Reviews of Geophysics and Space Physics, 16, 621-688. Townend, J., and Zoback, M.D., 2000, How faulting keeps the crust strong: Geology, 28, 399-402. Wu, P., and Hasegawa, H.S., 1996, Induced stresses and fault potential in eastern Canada due to a disc load: A preliminary analysis: Geophysical Journal International, 125, 415-430. Wu, P., and Johnston, P., 2000, Can deglaciation trigger earthquakes in N. America?: Geophysical Research Letters, 27, 1323-1326. Wu, P., and Mazzotti, S., 2007, Effects of a lithospheric weak zone on postglacial seismotectonics in eastern Canada and northeastern USA, In: Stein, S., Mazzotti, S., eds., Continental intraplate earthquakes: Science, hazard and policy issues, Geological Society of America Special Paper 425, 113-128. Zoback, M.L., 1992a, Stress field constraints on intraplate seismicity in eastern North America: Journal of Geophysical Research, 97, 11761-11782. Zoback, M.L., 1992b, First- and second-order patterns of stress in the lithosphere: The World Stress Map Project: Journal of Geophysical Research, 97, 11703-11728. Zoback, M.L, and Mooney, W.D., 2003, Lithospheric buoyancy and continental intraplate stress: International Geology Review, 45, 95-118. Zoback, M.D., and Townend, J., 2001. Implications of hydrostatic pore pressures and high crustal strength for the deformation of intraplate lithosphere: Tectonophysics, 336, 19-30. Zoback, M.D., Townend, J., and Grollimund, B., 2002, Steady-state failure equilibrium and deformation of intraplate lithosphere: International Geology Review, 44, 383-401. Zoback, M.D., and Zoback, M.L., 1981, State of stress and intraplate earthquakes in the United States: Science, 213, 96-104. Zoback, M.D., and Zoback, M.L., 1991, Tectonic stress field of North America and relative plate motion, In: Slemmon, B., et al., eds., The geology of North America, Decade Map Volume 1, Neotectonics of North America, Geological Society of America, Boulder, Colorado, 339-366. Zoback, M.L., and Zoback, M.D., 1980, State of stress of the conterminous United States: Journal of Geophysical Research, 85, 6113-6156. Zoback, M.L., and Zoback, M.D., 2007, Lithosphere Stress and Deformation, In: Watts, A., Schubert, G., eds., Earthquake Seismology – Treatise on Geophysics Vol. 6, Elsevier Ltd., Amsterdam, 253- 274.

Appendix 3A: New Focal Plane Mechanisms Compiled in this Study

Type = faulting regime (based on criteria from M.L. Zoback (1992b)): N = normal, SS = strike-slip, and T = thrust. Preferred nodal plane is indicated in bold. Δstr and Δdip are mis-fits between the preferred nodal plane and the theoretically optimally-oriented nodal plane (μ = 0.6). P and T-axes plunges are measured from horizontal. *mN. References: (1) Ammon et al., 1998; (2) Du et al., 2003; (3) Street et al., 1993; (4) Kim, 2003; (5) Kim et al., 2006; (6) St. Louis University Earthquake Center; (7) Ma et al., 2008.

Appendix 3B: New Formal Stress Inversions Compiled in this Study

Type = faulting regime: SS = strike-slip, T = thrust. All data are from Mazzotti and Townend (2010).

Lat Long SHmax Type N Ф Seismic Zone (˚N) (˚W) Azi (N˚E) 49.35 66.59 104 T 12 0.6 Lower St. Lawrence (Canada)

47.9 69.67 86 T 60 0.7 Charlevoix (Canada)

48.68 75.23 38 T 19 0.2 Gatineau (Canada)

46.03 77.40 78 T 8 0.7 Ottawa (Canada)

45.13 74.09 58 T 21 0.6 Montreal (Canada)

46.76 66.55 70 T 12 0.4 North Appalachian (Canada)

37.78 78.23 90 T 13 0.3 Central Virginia

35.27 84.60 54 SS 26 0.8 East Tennessee

36.12 89.67 82 SS 18 0.1 New Madrid (Missouri)

32.92 80.47 64 SS 11 0 Charleston (South Carolina)

Appendix 3C: Focal mechanisms from M.L. Zoback (1992)

Type = faulting regime: N = normal, SS = strike-slip, TS = transpressive and T = thrust. Preferred nodal plane is indicated in bold. Δstr and Δdip are mis-fits between the preferred nodal plane and the theoretically optimally-oriented nodal plane (μ = 0.6). P and T-axes plunges are measured from horizontal. See Zoback (1992b) for citations.

Chapter 4

REGIONAL STRESS ORIENTATIONS AND SLIP COMPATIBILITY OF EARTHQUAKE FOCAL PLANES IN THE NEW MADRID SEISMIC ZONE3

Abstract

We revisit the question of slip on faults in the New Madrid seismic zone in the context of the regional stress field. Specifically, we utilize newly available data on stress orientations and relative stress magnitudes to investigate whether fault slip is compatible with Coulomb faulting theory assuming laboratory-determined coefficients of fault friction (µ). We demonstrate that New Madrid fault planes are well-oriented for shear failure in the regional stress field assuming hydrostatic pore pressure in the brittle crust and coefficients of friction between 0.6 and 0.8. In other words, active New Madrid faults do not require elevated pore pressure, local stress sources, or

3 This chapter is published in Hurd, O., and Zoback, M.D., 2012, Regional Stress Orientations and Slip Compatibility of Earthquake Focal Planes in the New Madrid Seismic Zone: Seismological Research Letters, 83, 672-679.

anomalous fault strength to enable slip in the current ENE-WSW compressional stress field.

4.1 Introduction

The New Madrid seismic zone (NMSZ) in the central United States is one of the most active regions of intraplate seismicity in North America and site of the devastating 1811-1812 earthquake sequence (Nuttli, 1973; Johnston and Schweig, 1996). The NMSZ lies within the NE-SW-trending Reelfoot rift (Figure 4.1), which represents a failed rift arm that developed during the Late Proterozoic and Early Cambrian opening of the Iapetus Ocean on the southeast margin of early North America (Ervin and McGinnis, 1975). Fault offsets inferred from seismic reflection and trench data suggest that current seismicity levels likely initiated during the Holocene (Pratt, 1994; Schweig and Ellis, 1994; Van Arsdale, 2000) when optimally- oriented Proterozoic and Cambrian faults were reactivated in the contemporary stress field (Zoback et al., 1980; Braile et al., 1986; Dart and Swolfs, 1998). The reoccurrence of large and potentially damaging earthquakes in late Holocene time (Tuttle et al., 2005) may be a response to Pleistocene deglaciation (Grollimund and Zoback, 2001; Calais et al., 2010).

Contemporary seismicity illuminates a complex network of fault trends and deformation styles within the NMSZ (Stauder et al., 1976; Andrews et al., 1985; Himes et al., 1988; Chiu et al., 1992; Liu, 1997; Pujol et al., 1997; Mueller and Pujol, 2001; Dunn et al., 2010), although three large-scale faults characterize the seismic zone; the southern, NE-striking Axial fault, the central, NNW-striking Reelfoot fault, and the northern, NE-striking New Madrid North fault (Johnston and Schweig, 1996; Baldwin et al., 2005; Csontos and Van Arsdale, 2008). The Axial and New Madrid North faults exhibit strike-slip motion on near-vertical fault planes while thrust motion characterizes the southwest-dipping Reelfoot fault (Liu, 1997; St Louis University Earthquake Center). The fault patterns and senses of offset, which have been interpreted as a dextral strike-slip fault system with a left-stepping crossover (Russ,

1982), are geometrically consistent with those expected for active faults in an ENE- WSW compressive stress field (Zoback and Zoback, 1981). What remains to be examined in detail, however, is whether or not the faults and senses of offset are frictionally consistent with slip in the current stress field.

In this paper, we investigate whether or not local stress sources or anomalous fault strengths are required to explain active NMSZ faulting. Twelve well-constrained individual earthquake focal plane mechanisms augment previously available stress information. Following the methodology described in M.L. Zoback (1992) (and in Chapter 3), the new data are used to update the stress map, evaluate the consistency of maximum horizontal compressive stress (SHmax) orientations, and investigate fault stability using the Mohr-Coulomb failure criterion.

Figure 4.1: Major tectonic structures, seismicity, and stress measurements in the New Madrid seismic zone. The black crosses represent the twelve earthquakes examined in this study while the black stars represent approximate locations of the three mainshocks in the 1811-1812 earthquake sequence (Tuttle et al., 2005). Other earthquakes (gray crosses) are from the USGS/NEIC catalog (1973-2010). Stress measurements are from focal mechanisms compiled in this study, the 2008 World Stress Map database, and the St. Louis University Earthquake Center (NF = normal faulting, SS = strike-slip faulting, TF = thrust faulting). Red and blue dashed lines indicate boundaries of the Reelfoot rift and Mississippi embayment, respectively.

4.2 Data Collection

Since we use earthquake focal mechanisms to infer stress orientations and calculate relative stress magnitudes, it is crucial that the selected mechanisms are well- constrained. To ensure sufficient quality, all focal mechanisms are from MW > 2.5 earthquakes and are constrained by waveform modeling. This modeling technique commonly uses relative body-wave amplitudes combined with a broader search range over the focal sphere to constrain solutions, and often provides more reliable focal

mechanisms than those derived solely from P-wave first-motion polarities (e.g. Lay and Wallace, 1995).

Twelve new focal mechanisms are compiled that do not overlap with the Zoback (1992) dataset. Ten mechanisms are from the St. Louis University Earthquake Center North America moment tensor catalog or the Lamont-Doherty Cooperative Seismograph Network catalog (Table 4.1). The remaining two focal mechanisms were originally published in Herrmann (1979) and were designated slip-incompatible in the current stress field by Zoback (1992). However, Herrmann and Ammon (1997) revised the two focal mechanisms using improved waveform modeling techniques, and we reexamine both updated mechanisms in this study.

To examine slip compatibility of NMSZ faults, we utilize the individual focal mechanism nodal planes as well as large-scale fault planes delineated from hypocenter distributions within the NMSZ. Specifically, we consider the fault geometries from Csontos and Van Arsdale (2008) representing the New Madrid North, Reelfoot (northern and southern segments), and Axial faults (Table 4.2).

Table 4.1: New and revised focal plane mechanisms compiled for this study. The preferred nodal plane is indicated in bold. Δstr and Δdip are strike and dip misfits between the preferred nodal plane and the nearest test plane that slips with µ = 0.6. SS = strike-slip, T = thrust. References: (1) Herrmann and Ammon (1997); (2) St. Louis University Earthquake Center; (3) Lamont-Doherty Cooperative Seismograph Network.

Date Lat Long Z MW Strike Dip Rake Type SH Azi Ф Δstr Δdip Ref yyyy/mm/dd (˚N) (˚W) (km) (˚) (˚) (˚) (N°E) (˚) (˚) 1963/03/03 36.64 90.05 15 4.7 304 78 -28 SS 83 0.73 3 2 1 40 63 15 1967/07/21 37.44 90.44 15 4.0 350 60 135 T 82 0.31 3 2 1 116 52 40 1991/05/04 36.56 89.83 8 4.1 90 67 20 T 83 0.09 16 10 1 352 72 158 1996/11/29 35.97 90 11 3.8 120 65 15 T 75 0.20 4 1 2 24 76 164 2003/06/06 36.89 89.01 0.5 4.0 250 80 155 SS 96 0.71 21 2 3 345 65 11 2004/06/15 36.73 89.68 4.5 3.5 175 55 70 T 83 0.36 2 9 2 27 40 116 2004/07/16 36.86 89.17 4 3.5 43 71 159 SS 88 0.1 2 12 2 140 70 20 2005/02/10 35.75 90.23 14 4.1 55 80 -165 SS 75 0.59 7 4 2 322 76 170 2005/05/01 35.83 90.15 8 4.2 315 60 20 SS 75 0.45 9 4 2 214 74 148 2005/06/02 36.14 89.46 15 3.9 155 65 70 T 88 0.57 5 18 2 18 32 128 2005/06/20 36.95 88.96 9 3.7 315 80 10 SS 96 0.09 0 20 2 223 81 179 2010/03/02 36.79 89.36 5 3.4 211 86 135 SS 83 0.38 4 1 2 305 45 5

Table 4.2: New Madrid fault planes examined for frictional slip compatibility. The friction coefficient (µ) required for Coulomb failure on the fault plane considers a uniform, transpressional stress field. All fault planes are from Csontos and Van Arsdale (2008).

Strike Dip μ Description (N˚E) (˚) 46 90 0.7562 Axial fault (AF) 29 72 0.5879 New Madrid North fault (NMNF) 167 30 0.7858 Reelfoot fault, northern segment (RFN) 150 44 0.68 Reelfoot fault, southern segment (RFS)

4.3 Updating Stress Orientations

Using the twelve new focal mechanism P-axes, we first update the stress map in the New Madrid seismic zone and evaluate the consistency of SHmax orientations. Although a P-axis may deviate substantially from the maximum principal stress orientation in the absence of friction (McKenzie, 1969), experience has shown that P- axes in intraplate regions are consistent with maximum principal stress orientations

from other stress indicators (Zoback and Zoback, 1980). The average P-axis trend of the twelve focal mechanisms is N84˚E ± 7.4˚. This ENE-WSW trend is consistent with nearby stress measurements in the 2008 World Stress Map (WSM) database (Heidbach et al., 2008) from a variety of different stress indicators (Figure 4.1). In particular, we note the consistency with SHmax orientations inferred from borehole breakouts within regions of active faulting in northeastern Arkansas. The ENE-WSW trend represents a slight, clockwise rotation from the NE-SW SHmax orientation found throughout much of central and eastern North America.

4.4 Constraints on Relative Principal Stress Magnitudes

To evaluate slip compatibility on New Madrid fault planes, we follow the approach in Zoback (1992), which was previously described in Chapter 3, to calculate relative principal stress magnitudes from the focal mechanism nodal planes and the local stress tensor geometry using independent stress observations. The relative magnitudes are constrained with two physical relationships. The first relationship limits the allowable orientation of nodal plane slip vectors in the current stress field as defined by the magnitude of the intermediate principal stress (Angelier, 1979):

S  S   2 3 S  S 1 3 (4.1)

S1, S2, and S3 represent the three principal stresses in order of decreasing magnitude. Thus, Φ must fall between zero and one. Physically, this indicates that slip in the direction of the nodal plane slip vector is possible within the given stress configuration. Following Gephart (1985), Φ is calculated from the geometries of the two focal mechanism nodal planes and the three principal stresses:

      13 23 1  33 23 1     12 22 32 22 (4.2)

ßij corresponds to a matrix of angle cosines relating the principal stress and focal mechanism coordinate systems. Note that (4.2) yields a Φ value for both nodal planes, and therefore identifies which of the two nodal planes, if either, is geometrically consistent with slip in the current stress field (satisfies 0 ≤ Φ ≤ 1). For each of the twelve focal mechanisms, we find that only one of the two nodal planes is geometrically consistent. We will examine the frictional slip consistency of the nodal planes in the next stage of the analysis.

To independently estimate the stress tensor geometry near each earthquake, we calculate the average SHmax orientation of the three nearest A or B quality stress measurements in the WSM database (see Zoback and Zoback (1989) for details on the quality ranking system). In the NMSZ, all of the A and B quality stress measurements are from borehole breakouts. Thus, SHmax is inferred independent of focal mechanism data. For each of the twelve SHmax estimates, the standard deviation about the mean is less than 15˚. The remaining principal stress orientations are constrained assuming that the three principal stresses lie perpendicular to one another in vertical and horizontal planes (Zoback and Zoback, 1980). In this area, this assumption is validated by the NMSZ stress inversion in Mazzotti and Townend (2010) that illustrates nearly vertical and horizontal principal stresses.

The second physical relationship constrains relative principal stress magnitudes based on the frictional strength of optimally-oriented faults in the crust:

S  P 1 P  [( 2 1)1/ 2  ]2 S  P 3 P (4.3)

PP is the pore pressure and μ is the coefficient of fault friction (Jaeger and Cook,

1979). For given values of PP and μ, the differential stress magnitudes cannot exceed the stress required to cause shear failure on preexisting, optimally-oriented faults in the brittle crust. Therefore, (4.3) provides an upper bound on the ratio of the maximum and minimum effective stresses. The relative principal stress magnitudes are calculated using (4.1) – (4.3) assuming hydrostatic PP in the brittle crust (Zoback and

Townend, 2001), μ = 0.8 (Byerlee, 1978), and SV = 25 MPa/km (overburden density = 2500 kg/m3).

4.5 Slip Compatibility

The final analysis step uses three separate approaches to examine the frictional slip compatibility of NMSZ fault planes in the contemporary stress field using the Mohr-Coulomb failure criterion. The Mohr-Coulomb criterion for slip on a preexisting, cohesionless fault is given by:

  (S  P ) N P (4.4)

τ and SN are the shear and normal stresses acting on the fault plane, respectively. Slip occurs when the shear stress exceeds the frictional strength of a fault and the effective normal stress acting on a fault. The slip compatibility analysis will quantitatively determine the friction coefficients required for shear failure on NMSZ fault planes.

In the first approach, we begin by calculating the relative principal stress magnitudes near each earthquake. Since the magnitudes are calculated directly from the nodal plane and local SHmax orientations, they vary from earthquake to earthquake. Next, a set of test planes are created for each focal mechanism by perturbing the actual nodal plane strikes and dips up to ± 45° in 1° increments. For each test plane, we then compute the friction coefficient required for failure using the Mohr-Coulomb criterion assuming hydrostatic pore pressure. The results of a friction coefficient calculation for a thrust (4.2a) and strike-slip (4.2b) focal mechanism are shown in Figure 4.2. The strike and dip of the actual focal mechanism nodal planes are denoted by black dots. Test plane configurations where slip is frictionally impossible in the current stress field are indicated by hatched areas. All twelve focal mechanisms have one nodal plane on which shear slip is frictionally possible. That is, for one of the two nodal planes, the black dot falls within a colored region. We denote this nodal plane as the preferred nodal plane for each focal mechanism (Table 4.1).

Figure 4.2: Frictional slip compatibility results for a thrust (a) and strike-slip (b) focal mechanism analyzed in this study. Focal mechanism nodal plane orientations are indicated by black dots. The friction coefficients (μ) required for Coulomb failure on the test planes in the local stress field are scaled by color. Test plane configurations where slip is frictionally impossible in the current stress field are indicated by hatched areas. Only one nodal plane from each focal mechanism is frictionally consistent with slip in the current stress field (black dot falls in a colored region) and is identified as the preferred nodal plane.

For each focal mechanism, we then select the test planes that would slip with μ = 0.8 and μ = 0.6 which are most similar in orientation to the focal mechanism nodal planes. These test planes are identified as well-oriented for failure in the current stress field. Next, we calculate the misfit in strike and dip between the well-oriented test planes and each of the focal mechanism nodal planes. Histograms of the misfits for the preferred (4.3a) and conjugate (4.3b) nodal planes for all twelve earthquakes are shown in Figure 4.3. The preferred nodal plane deviates on average only ~6° in strike

and ~7° in dip from the nearest well-oriented test plane that slips with μ = 0.6. Note that the conjugate nodal planes have significantly higher misfits of 25° in strike and 14° in dip. For test planes that slip with μ = 0.8, the average misfit from the preferred nodal plane is less than 20° in strike and dip (not shown). Since the preferred nodal plane misfits are within the range of uncertainty associated with the stress measurements and focal mechanism solutions, we conclude that the preferred nodal planes are slip-compatible in the current stress field.

Figure 4.3: Histograms displaying the misfit in strike and dip between the preferred (a) and the conjugate (b) focal mechanism nodal planes and the nearest well-oriented test plane that slips with μ = 0.6. The preferred nodal planes deviate on average less than 8° in strike and dip from the nearest well-oriented test plane. Note the different horizontal scales between preferred and conjugate nodal plane histograms.

A second approach for testing frictional slip compatibility utilizes a simplified version of the first approach. Instead of calculating relative principal stress magnitudes from each focal mechanism, a uniform stress field where Φ ≈ 0 and SHmax is oriented N80°E is prescribed across the entire NMSZ. The Φ value corresponds to a transpressional stress regime where SHmax > SV ≈ Shmin. A transpressional stress

regime is selected based on the average Φ value of the twelve focal mechanisms (0.38 ± 0.2) and the prevalence of strike-slip and thrust focal mechanisms across the NMSZ.

The N80°E SHmax orientation is chosen based on the dominant ENE-WSW SHmax trend within the study area. Assuming the uniform stress field and hydrostatic pore pressure, the Mohr-Coulomb criterion is again used to calculate the friction coefficient required for slip on all focal mechanism nodal planes. Ten of the twelve focal mechanisms have at least one nodal plane that slips with 0.6 < µ < 0.8. In other words, the nodal planes slip with friction coefficients consistent with those derived from laboratory experiments (Byerlee, 1978). The frictional slip compatibility is shown graphically on a Mohr diagram normalized by the vertical stress in Figure 4.4a. The nodal planes are represented by the gray circles. The remaining two focal mechanisms would only require minimal nodal plane rotations (10˚ in strike, for example) to have one nodal plane become slip compatible (0.6 < µ < 0.8).

Figure 4.4: (a) Mohr diagram illustrating frictional slip compatibility of focal mechanism nodal planes (circles) and large-scale NMSZ faults (squares) assuming a uniform, transpressional stress field and hydrostatic pore pressure. Both axes are normalized by the vertical stress (SV). Ten of the twelve focal mechanisms (gray circles) have at least one nodal plane that is frictionally consistent with slip (0.6 < μ < 0.8). All four large-scale NMSZ fault segments (gray squares) are also frictionally consistent with slip. AF = Axial fault, RFS = Reelfoot fault, southern segment, RFN = Reelfoot fault, northern segment, NMNF = New Madrid north fault. (b) Large-scale NMSZ faults examined for frictional slip compatibility in the current stress field. Slip compatibility results are shown in Table 4.2 and Figure 4.4a. The large, black arrows represent the SHmax orientation inferred from focal mechanism P-axes in this study. Fault trends are modified from Csontos and Van Arsdale (2008).

A third approach examines the frictional slip compatibility of large-scale fault planes delineated from seismicity clusters within the NMSZ. Four fault planes are tested from Csontos and Van Arsdale (2008) representing the New Madrid North, Reelfoot (northern and southern segments), and Axial faults (Figure 4.4b and Table 4.2). Using the Mohr-Coulomb criterion with the uniform, transpressional stress field and hydrostatic pore pressure, we find all four fault planes are compatible with slip in the current stress field with 0.58 < µ < 0.8 (Table 4.2). The slip compatibility of each plane is shown on a Mohr diagram in Figure 4.4a (gray squares).

4.6 Discussion

The twelve individual focal mechanism P-axes indicate a consistent ENE- WSW maximum horizontal compressive stress across the NMSZ. This represents a

20-30˚ clockwise rotation from the dominant NE-SW SHmax trend observed throughout most of the eastern United States and southeastern Canada thought to derive from buoyancy-driven forces (ridge push) (Zoback and Zoback, 1980, 1989, 2007). While this minor rotation is similar in scale to the uncertainties in many stress measurements, our careful data selection procedure and the consistency of SHmax orientations with different types of stress indicators in the NMSZ implies this is a robust observation.

Several studies have suggested that SHmax orientations also perturb locally within the NMSZ (O’Connell et al., 1982; Liu 1997), possibly reflecting complex fault interactions (Russ, 1982). While transient local stress states may exist near major fault intersections, the stress measurements inferred in this study as well as the vast majority of measurements in the WSM database indicate a consistent SHmax orientation throughout the NMSZ.

The slip-compatibility results indicate that NMSZ faults are well-oriented for shear slip in the contemporary stress field and generally do not require elevated pore pressure, local stress sources, or anomalous fault friction to enable shear failure (they are slip compatible). The results also compliment the conclusions of Zoback (1992)

and Hurd and Zoback (2012) which demonstrated general slip compatibility of intraplate faults in the current stress field using larger focal mechanism data sets spanning the central and eastern United States and eastern Canada. Furthermore, the findings corroborate previous qualitative assessments that the fault patterns and deformation styles inferred from NMSZ focal mechanisms are consistent with an

ENE-WSW SHmax orientation (Zoback and Zoback, 1981, 1989). While we acknowledge that fault interactions and igneous intrusions may locally affect stress distributions within the NMSZ (O’Connell et al., 1982; Andrews et al., 1985; Ellis, 1994; Hildenbrand and Hendricks, 1995; Liu, 1997; Hildenbrand et al., 2001), our results quantitatively demonstrate that local stress perturbations are not required to explain slip on most NMSZ faults.

Friction coefficients (μ) between 0.6 and 0.8 are considered in our analysis based on multiple lines of evidence. First, Byerlee (1978) demonstrated from laboratory experiments that μ generally falls between 0.6 and 1.0 for a wide variety of rock types over a range of confining pressures, although it may be lower in shaley rocks, which is not relevant to the NMSZ earthquakes. Second, in-situ stress measurements as deep as ~8 km in the upper crust are regularly consistent with predicted stress magnitudes using Coulomb frictional-failure theory with 0.6 ≤ μ ≤ 1.0 (Figure 1 in Townend and Zoback, 2000). Lastly, Sibson and Xie (1998) and Collettini and Sibson (2001) demonstrated using the Coulomb failure criterion that dips of active thrust and normal faults associated with relatively large earthquakes (M > 5.5) are consistent with fault reactivation assuming 0.6 ≤ μ ≤ 0.85 and principal stresses lying in horizontal and vertical planes.

Exhumed major fault zones often illustrate highly variable physical properties between the host rock, damage zone and fault core. These variable physical properties can affect local stress orientations and magnitudes, which may subsequently influence fracture propagation behavior (Gudmundsson et al., 2010). Our analysis directly examines the potential for variable friction coefficients or local stress perturbations to enable slip on NMSZ faults. In particular, a uniform stress field is applied over the

NMSZ, and the friction coefficient required for shear failure is calculated for each fault plane. Even in this homogeneous stress configuration, friction coefficients consistent with empirical observations can explain slip on most NMSZ faults assuming hydrostatic pore pressure in the brittle crust. Thus, the scale of perturbation considered by Gudmundsson et al. (2010) (and many others) appears to be smaller than the scale primarily driving large scale earthquake rupture.

Previous researchers have suggested that NMSZ faulting may be controlled in part by elevated fluid pressures in the upper crust (Al-Shukri and Mitchell, 1988; McKeown and Diehl, 1994; Powell et al., 2010). The hydrostatic pore pressure assumption in this study is based on widespread observations of hydrostatic pore pressure persisting to as deep as 12 km in the upper crust (Table 1 in Townend and Zoback, 2000) and the consistency of hydrostatic pore pressure in the upper crust in maintaining observed lithospheric deformation rates in force-limited stress models (Zoback and Townend, 2001). While active faults can often be conduits for fluid flow, our results and those of Zoback (1992) indicate that it is generally not necessary to prescribe substantially elevated fluid pressure in the intraplate crust to enable fault slip. Moreover, they further support the hypothesis that the brittle crust is generally in a state of frictional failure equilibrium due to regional plate driving forces (Zoback et al, 2002) with regional variations reflecting changes of lithospheric density (Zoback and Mooney, 2003).

The slip incompatibility of the two original focal mechanisms from Herrmann (1979) and compatibility of the revised mechanisms from Herrmann and Ammon (1997) highlights the importance of having well-constrained stress data. This is particularly significant for focal mechanisms, which are a primary source of stress information in the NMSZ and consequently the basis of many tectonic interpretations. Quality guidelines for stress measurements were presented by Zoback and Zoback (1989), although the inherent subjectivity of the ranking criteria requires that careful examination and selection is always necessary. Newer focal mechanisms constrained by waveform modeling, such as those from the St. Louis University Earthquake

Center, are more reliable and often present much more self-consistent stress information with nearby data points than mechanisms constrained with less stringent techniques. High resolution seismic monitoring will be available over the entire NMSZ and Mississippi embayment with the repositioning of EarthScope’s U.S. Transportable Array of broadband seismometers over the central U.S. during 2011. This will hopefully lead to improved hypocenter locations and focal mechanisms, enhanced images of seismogenic fault planes, and improved tectonic models of the NMSZ.

4.7 Conclusions

In this study, we revisited the question of fault slip in the New Madrid seismic zone within the current stress field originally addressed by M.L. Zoback (1992). We compiled a set of twelve well-constrained focal mechanisms from earthquakes occurring in the NMSZ and four large-scale NFSZ fault planes delineated from seismicity clusters. Using this data, we updated the local stress map, examined the consistency of SHmax orientations across the NMSZ, and investigated fault stability.

The twelve focal mechanism P-axes indicate a consistent ENE-WSW SHmax orientation, which is in agreement with preexisting stress measurements in the World Stress Map database. The ENE-WSW trend represents a slight clockwise rotation from the NE-SW SHmax orientation observed over much of the central and eastern United States.

To examine fault stability, we used the Mohr-Coulomb failure criterion to calculate the friction coefficients required for shear failure on the focal mechanism nodal planes and large-scale fault planes assuming a uniform, transpressional stress field and hydrostatic pore pressure. Nearly all fault planes are compatible with shear slip with laboratory-determined coefficients of friction (0.6 < µ < 0.8) within the uniform stress field. We therefore conclude that, in general, elevated pore pressures, local stress perturbations or anomalous fault strength are not necessary to explain the faulting within the NMSZ. Our conclusions support the findings of Zoback (1992) in

the central and eastern United States as well as previous qualitative arguments by Zoback and Zoback (1981, 1989) that NMSZ faults are well-oriented for slip in an ENE-WSW compressive stress field.

4.8 References

Al-Shukri, H.J., and Mitchell, B.J., 1988, Reduced seismic velocities in the source zone of New Madrid earthquakes: Bulletin of the Seismological Society of America, 78, 1491-1509. Andrews, M.C., Mooney, W.D., and Meyer, R.P., 1985, The relocation of microearthquakes in the northern Mississippi embayment: Journal of Geophysical Research, 90, 10223-10236. Angelier, J., 1979, Determination of the mean principal directions of stresses for a given fault population: Tectonophysics, 56, T17-T26. Baldwin, J.N., Harris, J.B., Van Arsdale, R.B., Givler, R., Kelson, K.I., Sexton, J.L., and Lake, M., 2005, Constraints on the location of the late Quaternary Reelfoot and New Madrid north faults in the northern New Madrid seismic zone, central United States: Seismological Research Letters, 76, 772-789. Braile, L.W., Hinze, W.J., Keller, G.R., Lidiak, E.G., and Sexton, J.L., 1986, Tectonic development of the New Madrid rift complex, Mississippi embayment, North America: Tectonophysics, 131, 1-21. Byerlee, J.D., 1978, Friction of rock: Pure and Applied Geophysics, 116, 615-626. Calais, E., Freed, A.M., Van Arsdale, R., and Stein, S., 2010, Triggering of New Madrid seismicity by late-Pleistocene erosion: Nature, 466, 608-612. Chiu, J.M., Johnston, A., and Yang, Y., 1992, Imaging the active faults of the central New Madrid seismic zone using PANDA array data: Seismological Research Letters, 63, 375-393. Collettini, C., and Sibson, R.H., 2001, Normal faults, normal friction?: Geology, 29, 927-930. Csontos, R., and Van Arsdale, R., 2008, New Madrid seismic zone fault geometry: Geosphere, 4, 802- 813. Dart, R.L., and Swolfs, H.S., 1998, Contour mapping of relic structures in the Precambrian basement of the Reelfoot rift, North American midcontinent: Tectonics, 17, 235-249. Dunn, M., Horton, S., DeShon, H., and Powell, C., 2010, High-resolution earthquake relocation in the New Madrid Seismic Zone: Seismological Research Letters, 81, 406-413. Ellis, W., 1994, Summary and discussion of crustal stress data in the region of the New Madrid seismic zone: United States Geological Survey Open File Report 1538A-C, B1-B13. Ervin, C.P., and McGinnis, L.D., 1975, Reelfoot rift: Reactivated precursor to the Mississippi embayment: Bulletin of the Geological Society of America, 86, 1287-1295. Gephart, J.W., 1985, Principal stress directions and the ambiguity in fault plane identification from focal mechanisms: Bulletin of the Seismological Society of America, 75, 621-625. Grollimund, B., and Zoback, M.D., 2001, Did deglaciation trigger intraplate seismicity in the New Madrid seismic zone?: Geology, 29, 175-178. Gudmundsson, A., Simmenes, T.H., Larsen, B., and Philipp, S.L., 2010, Effects of internal structure and local stresses on fracture propagation, deflection, and arrest in fault zones: Journal of Structural Geology, 32, 1643-1655. Heidbach, O., Tingay, M., Barth, A., Reinecker, J., Kurfeß, D., and Müller, B., 2008, The world stress map database release 2008. http://www.world-stress-map.org. Last accessed August 2010. Herrmann, R.B, 1979, Surface wave focal mechanisms for eastern North American earthquakes with tectonic implications: Journal of Geophysical Research, 84, 3543-3552. Herrmann, R.B., and Ammon, C.J., 1997, Faulting parameters of earthquakes in the New Madrid, Missouri, region: Engineering Geology, 46, 299-311. Hildenbrand, T.G., and Hendricks, J.D., 1995, Geophysical setting of the Reelfoot rift and relations between rift structures and the New Madrid seismic zone: United States Geological Survey Professional Paper 1528E, 30 pp.

Hildenbrand, T.G., Stuart, W.D., and Talwani, P., 2001, Geologic structures related to New Madrid earthquakes near Memphis, Tennessee, based on gravity and magnetic interpretations: Engineering Geology, 62, 105-121. Himes, L., Stauder, W., and Herrmann, R.B., 1988, Indication of active faults in the New Madrid seismic zone from precise location of hypocenters: Seismological Research Letters, 59, 121-131. Hurd, O., and Zoback, M.D., 2012, Intraplate earthquakes, regional stress and fault mechanics in the central and eastern U.S. and southeastern Canada: Tectonophysics (in press) Jaeger, J.C., and Cook, N.G.W., 1979, Fundamentals of Rock Mechanics, Chapman and Hall, London, UK. Johnston, A.C., and Schweig, E.S., 1996, The enigma of the New Madrid earthquakes of 1811-1812: Annual Reviews of Earth and Planetary Science, 24, 339-384. Lay, T., and Wallace, T.C., 1995, Modern Global Seismology, Academic Press, San Diego. Lamont-Doherty Cooperative Seismograph Network, http://www.ldeo.columbia.edu/LCSN/, Last accessed August 2011. Liu, Z., 1997, Earthquake modeling and active faulting in the New Madrid seismic zone, Ph.D. Dissertation, Saint Louis University. Mazzotti, S., and Townend, J., 2010, State of stress in central and eastern North American seismic zones: Lithosphere, 2, 76-83. McKenzie, D.P., 1969, The relation between fault plane solutions for earthquakes and the directions of the principal stresses: Bulletin of the Seismological Society of America, 59, 591-601. McKeown, F.A. and Diehl, S.F., 1994, Evidence of contemporary and ancient excess fluid pressure in the New Madrid seismic zone of the Reelfoot Rift, central United States, In: Shedlock, K.M., Johnston, A.C., eds., Investigations of the New Madrid Seismic Zone, United States Geological Survey Professional Paper 1538N, 24 pp. Mueller, K., and Pujol, J., 2001, Three-dimensional geometry of the Reelfoot blind thrust: Implications for moment release and earthquake magnitude in the New Madrid seismic zone: Bulletin of the Seismological Society of America, 91, 1563-1573. Nuttli, O.W, 1973, The Mississippi Valley earthquakes of 1811-1812: Intensities, ground motion and magnitudes: Bulletin of the Seismological Society of America, 63, 227-248. O’Connell, D.R., Bufe, C.G., and Zoback, M.D., 1982, Microearthquakes and faulting in the area of New Madrid, Missouri-Reelfoot Lake, Tennessee, In: McKeown, F.A., Pakiser, L.C., eds., Investigations of the New Madrid, Missouri, Earthquake Region, United States Geological Survey Professional Paper 1236, 31-38. Powell, C.A., Withers, M.M., DeShon, H.R., and Dunn, M.M., 2010, Intrusions and anomalous Vp/Vs ratios associated with the New Madrid seismic zone: Journal of Geophysical Research, 115, B08311. Pratt, T.L., 1994, How old is the New Madrid seismic zone?: Seismological Research Letters, 65, 172- 179. Pujol, J., Johnston, A., Chiu, J., and Yang, Y., 1997, Refinement of thrust faulting models for the central New Madrid seismic zone: Engineering Geology, 46, 281-298. Russ, D.P., 1982, Style and significance of surface deformation in the vicinity of New Madrid, Missouri, In: McKeown, F.A., Pakiser, L.C., Investigations of the New Madrid, Missouri, Earthquake Region, United States Geological Survey Professional Paper 1236, 95-114. Saint Louis University Earthquake Center, http://www.eas.slu.edu/Earthquake_Center/MECH.NA/index.html, last accessed July 2011. Schweig, E.S., and Ellis, M.A., 1994, Reconciling short recurrence intervals with minor deformation in the New Madrid seismic zone: Science, 264, 1308-1311. Sibson, R.H., and Xie, G., 1998, Dip range for intracontinental reverse fault ruptures: Truth not stranger than friction?: Bulletin of the Seismological Society of America, 88, 1014-1022. Stauder, W., Kramer, M., Fischer, G., Schaefer, S., and Morrissey, S.T., 1976, Seismic characteristics of southeast Missouri as indicated by a regional telemetered micro-earthquake array: Bulletin of the Seismological Society of America, 66, 1953-1964. Townend, J., and Zoback, M.D., 2000, How faulting keeps the crust strong: Geology, 28, 399-402.

Tuttle, M.P., Schweig III, E.S., Campbell, J., Thomas, P.M., Sims, J.D., and Lafferty III, R.H., 2005, Evidence for New Madrid earthquakes in A.D. 300 and 2350 B.C.: Seismological Research Letters, 76, 489-501. Van Arsdale, R., 2000, Displacement history and slip rate on the Reelfoot fault of the New Madrid seismic zone: Engineering Geology, 55, 219-226. Zoback, M.L., 1992, Stress field constraints on intraplate seismicity in eastern North America, Journal of Geophysical Research, 97, 11761-11782. Zoback, M.D., Hamilton, R.M., Crone, A.J., Russ, D.P., McKeown, F.A., and Brockman, S.R., 1980, Recurrent intraplate tectonism in the New Madrid seismic zone: Science, 209, 971-976. Zoback, M.L, and Mooney, W.D., 2003, Lithospheric buoyancy and continental intraplate stress: International Geology Review, 45, 95-118. Zoback, M.D., and Townend, J., 2001, Implications of hydrostatic pore pressures and high crustal strength for the deformation of intraplate lithosphere: Tectonophysics, 336, 19-30. Zoback, M.D., Townend, J., and Grollimund, B., 2002, Steady-state failure equilibrium and deformation of intraplate lithosphere: International Geology Review, 44, 383-401. Zoback, M.L., Zoback, M.D., 1980, State of stress of the conterminous United States: Journal of Geophysical Research, 85, 6113-6156. Zoback, M.D., and Zoback, M.L., 1981, State of stress and intraplate earthquakes in the United States: Science, 213, 96-104. Zoback, M.L., Zoback, M.D., 1989, Tectonic stress field of the conterminous United States: Memoirs of the Geological Society of America, 172, 523-539. Zoback, M.L., and Zoback, M.D., 2007, Lithosphere Stress and Deformation, In: Watts, A., Schubert, G., eds., Earthquake Seismology – Treatise on Geophysics Vol. 6, Elsevier Ltd., Amsterdam, 253- 274.

Chapter 5

APPLICATION OF AN INTEGRATED, GEOMECHANICS-BASED ANALYSIS TO IDENTIFY FLUID PATHWAYS IN A TIGHT-GAS RESERVOIR

Abstract

An integrated geomechanical and seismological analysis was applied to identify flow pathways created during a hydraulic fracture stimulation of a tight-gas sandstone reservoir. Geophysical log data, rock strength tests, leak-off tests, and borehole image logs indicate the study area is characterized by a strike-slip stress regime with SHmax oriented NNE-SSW, and the reservoir contains abundant preexisting natural fractures. At near peak injection pressures during hydraulic fracturing, we find that many preexisting natural fractures intersecting the wellbore would be stimulated in shear. Wellbores intersecting more of these shear-stimulated fractures generally have higher production over an initial two-year period, suggesting the stimulated preexisting natural fracture network serves as a flow pathway into the reservoir. Using the double-difference technique, we relocate ~400 microseismic events induced during a hydraulic fracture stimulation in an attempt to identify flow paths extending into the reservoir. The double-difference relocated hypocenters resolve into two vertical, N-S

trends which parallel the optimal orientation of strike-slip faults in the regional stress field, suggesting potential fault reactivation during stimulation. However, resolution tests indicate that hypocenter locations are poorly resolved in the N-S direction due to the limited monitoring array configuration, unfortunately precluding definitive interpretation of the origin of the N-S hypocenter trend. Thus, while the double- difference technique does significantly reduce scatter in the microseismic event locations, the limited monitoring array geometry severely inhibits resolution of reservoir structures serving as potential flow pathways during hydraulic fracturing stimulations.

5.1 Introduction

Low permeability sandstone formations host significant hydrocarbon reserves in sedimentary basins throughout the world. Although these potential formations, known as tight-gas reservoirs, have been appreciated for several decades, exploitation has proven difficult due to the characteristically low rock permeabilities and porosities. However, tight-gas reservoirs have recently become feasible exploitation targets with the advent of horizontal and directional drilling techniques and water- based (slickwater) hydraulic fracturing stimulations over the last two decades. Despite generally successful exploitation, questions remain regarding how hydraulic fractures are placed during stimulations, the influence of preexisting natural fractures on hydrocarbon production, and how both serve as fluid pathways through the reservoir. Addressing these questions could provide valuable insight into optimizing future hydraulic fracture design and implementation in tight-gas formations.

Accurately constraining in-situ stresses and fluid pressures within a hydrocarbon reservoir is crucial to addressing a variety of drilling issues including casing design (Moos et al., 2003), well trajectory (Zoback et al., 2003), and mud weight selection (Ito et al., 2001). In addition, the in-situ stress field and fluid pressures also fundamentally affect the behavior of preexisting and placed fracture networks during hydraulic fracturing stimulations. For example, Mode I hydraulic

fractures will always propagate in a plane perpendicular to the least principal stress (e.g. Hubbert and Willis, 1957), and preexisting natural fractures optimally-oriented for shear slip in the current stress field can serve as preferential fluid pathways during hydraulic fracture stimulations (Barton et al., 1995). Additionally, slow-slip on poorly-oriented natural fractures also appears to be a significant stimulation mechanism during hydraulic fracturing (Zoback et al., 2012). Since the stimulated fracture network serves as the primary fluid pathway into the hydrocarbon reservoir, constraining in-situ stresses and fluid pressures is essential to identifying and optimizing flow pathways created during hydraulic fracture stimulations.

The preexisting and stimulated fracture networks are typically characterized using resistivity wellbore image logs which provide an unwrapped view of the wellbore wall that allows for mapping fractures and faults. More recently, fracture characteristics have been inferred from hypocenter distributions of microseismic events induced during hydraulic fracture stimulations (e.g. Fehler et al., 1987; Phillips et al., 1996; Rutledge et al., 1998; Maxwell et al., 2002; Williams-Stroud, 2008). This fracture characterization technique, known as microseismic fracture mapping, has been used to evaluate stimulated reservoir volumes as well as hydraulic fracture effectiveness, and can provide key insights into fluid pathways extending deep into reservoirs. However, fracture characterization using this technique is sensitive to microseismic event hypocenter location errors, which can be up to 100s of meters when using conventional location techniques in sub-optimal (e.g. single or dual) monitoring array configurations.

In this chapter, we apply an integrated analysis to identify fluid pathways being stimulated in a tight-gas reservoir during hydraulic fracturing stimulations. The analysis consists of three primary steps. First, using wireline, pressure test, and wellbore image log data, we develop a geomechanical model to constrain in-situ stresses, rock strengths, and pore pressures within the tight-gas reservoir. In the second step, we characterize the preexisting natural fracture network within the tight- gas reservoir using wellbore image logs and evaluate the tendency for shear slip on

preexisting natural fractures in the current stress field. The final analysis step utilizes advanced earthquake location techniques to obtain precise hypocenter relocations of microseismic events induced during a hydraulic fracture stimulation in the tight-gas reservoir. We hope the analysis will provide enhanced resolution of fluid pathways stimulated during hydraulic fracture stimulations in tight-gas reservoirs while simultaneously demonstrating the advantages of a highly-integrated analysis to addressing questions regarding fluid flow in low-permeability hydrocarbon reservoirs.

5.2 Study Area and Geology

The study area (hereafter named the WC field) lies in eastern British Columbia, Canada near the eastern border with Alberta. The WC field actively produces from several hydrocarbon reservoirs within the Western Canada Sedimentary Basin (WCSB), a prolific hydrocarbon-bearing basin formed as part of a Jurassic-Paleocene foreland basin system on the western margin of North America (e.g. Porter et al., 1982; Creaney and Allan, 1990). The WC field is seismically quiet with very few naturally-occurring earthquakes and no major tectonic structures or faults. The region is characterized by a NE-SW-striking maximum horizontal stress (SHmax), which is consistent with the general orientation of SHmax throughout much of North America (Zoback and Zoback, 1980). The primary target reservoir within the WC field is the gas-bearing Cretaceous NC formation, a low permeability, low porosity reservoir comprising mostly sandstones and conglomerates.

The Lower Cretaceous NC formation was deposited within a massive foreland basin fluvial environment. NC formation sediments being drained off high peaks west and southwest of the foreland basin were deposited in large alluvial fans which gradually spread out into a broad fluvial plain. At the eastern edge of the fluvial plains, a NW-flowing braided river system, termed the Spirit River by McLean (1977), reworked the deposited sediments and transported them northward (Figure 5.1). The NC formation is therefore characterized by two primary facies, an alluvial plain facies consisting of low-permeability sandy conglomerates with chert and quartzite pebbles

within a dark, fine-grained chert and quartz matrix and a braided river facies consisting of more permeable sandstones and sandy conglomerates within a lighter, coarse- grained, and well-sorted matrix (McLean, 1977; Gies, 1984). In the alluvial plain facies, permeability is commonly less than 0.01 mD and porosity is ~3%.

Figure 5.1: Depositional setting of the NC formation (from Gies, 1984).

The NC formation at the WC field lies within the downdip, gas-saturated region of the WCSB. Natural gas is presently being generated in source rocks adjacent to the NC formation and continually injected vertically into the downdip portion of the reservoir (Gies, 1984). The NC stratagraphic interval is dominated by Type III kerogens, and mean vitrinite reflectance values are generally greater than 0.8% and possibly as high as 1.8%, indicating a highly mature gas. NC formation depths are around 2,500 meters, and thickness ranges from 5-40 meters. The NC formation is also significantly underpressured within the deeper gas-bearing portions of the deep basin, and contains a well-developed natural fracture system (Gies, 1984). The NC

formation at the WC field has produced more than 100 MMcfd of natural gas through 2010 (Nordheimer and Jones, 2010).

5.3 Developing a Geomechanical Model

Developing a well-constrained geomechanical model is critical to understanding the in-situ stresses and fluid pressures that act on and within a hydrocarbon reservoir. A geomechanical model constrains the magnitudes and orientations of the three principal stresses (vertical, SV, minimum horizontal, Shmin, and maximum horizontal, SHmax), the reservoir pore pressure, and the reservoir rock strength. The geomechanical model parameters are discussed in detail below, and are constrained following the methodologies described in Zoback et al. (2003). The completed geomechanical model will serve as a basis for the fracture slip analysis presented in Section 5.4.

5.3.1 Pore pressure Reservoir pore pressures can be measured by a variety of means including drill stem tests, repeat formation tests, geophysical wireline logs, pressure while drilling recordings, seismic reflection data, and drilling mud weights. Unperturbed NC reservoir pore pressures were available for many wells in WC field, and ranged between 14 and 21 MPa with an average of 18 MPa. Assuming an average reservoir depth around 2,500m, this corresponds to a significantly underpressured gradient (7.2 ± 0.6 MPa/km), which is characteristic of the NC formation (Gies, 1984; Kovalsky and Kelly, 2008). The drilling mud weights exceeded the unperturbed reservoir pressures by 2-7 MPa when drilling through the NC formation.

5.3.2 Compressive rock strength Compressive rock strength (UCS) estimates, as well as static and dynamic physical rock properties, were constrained from uniaxial and triaxial laboratory tests conducted by an independent service company. Rock core samples of the NC

formation from both surface outcrops and reservoir depths were tested. The first set of core samples consisted of 4”, full-diameter samples and 1.5” core plugs taken from a single well near the WC field over depths from 2,400 to 2,421 meters. The samples comprised both the alluvial plain and braided river facies of the NC formation. Unfortunately, the first set of core material encountered a variety of mechanical issues during testing, which yielded spurious and inconsistent strength and rock property measurements. The test results from the first core set were therefore not considered in this study.

In response to the problems encountered while testing the first set of cores, a second set of smaller-diameter core samples was also collected and tested by the service company. The second core set consisted of five 1” x 3” core plug samples from an NC formation outcrop, three of which were vertical orientation (perpendicular to bedding) and two horizontal. The second core set sampled the alluvial plain facies, and is representative of the reservoir rocks at depth. The core samples were subject to both uniaxial (1 sample) and triaxial (four samples) tests. The triaxial test for each sample consisted of two loading stages at a single confining pressure: the first using ultrasonic end caps where the sample was brought near failure and then unloaded, and the second where the sample was refitted with steel end caps and taken completely to failure. Confining pressures were 0 (UCS), 22, 33, 44 and 66 MPa, and the tests were performed on brine-imbibed samples at room temperature with pore pressure drained. The second core set yields a UCS measurement of 110 ± 25 MPa.

In addition to rock strength constraints from uniaxial and triaxial laboratory tests on core samples, we also estimate rock strength from wireline log data using an empirical relationship for conglomerates derived by Moos et al. (1999). The average rock strength at NC reservoir depths using wireline log data from eight wells is 89 ± 23 MPa, which is within the range of strength values from the core analyses. We will adopt a UCS value of 100 ± 10 MPa for our geomechanical model.

5.3.3 Vertical stress (SV) The vertical stress represents the stress imparted by the overlying column of rock at a given point in the earth, and is typically the most straightforward of the three principal stress magnitudes to estimate. The SV magnitude can be calculated from wireline density logs with the following formula

Z S  (z)gdz  gz V  0 (5.1) where ρ(z) is density as a function of depth,  is the mean overburden density and g is gravitational acceleration. The average SV gradient calculated from wireline density logs from seven wells in the WC field assuming a gravitational constant of 9.81 m/s2 and a depth of 2,500m is 25.2 ± 0.64 MPa/km. This SV gradient is consistent with a commonly assumed lithostatic gradient of 25-26 MPa/km for the upper crust, which corresponds to a rock density of 2,600 kg/m3. Following Zoback and Zoback (1980), we assume the three principal stresses are oriented in vertical and horizontal planes and are orthogonal to one another. Therefore, assuming SV is approximately vertical, knowing the orientation of either Shmin or SHmax constrains the orientation of the three principal stresses.

5.3.4 Shmin magnitude

Shmin magnitudes are commonly estimated from leak off or extended leak off tests, mini-frac tests, and/or lost circulation and wellbore ballooning incidents observed during drilling. In this study, Shmin magnitudes are estimated from instantaneous shut-in pressures (ISIPs) at the end of seven hydraulic fracturing stimulations. The ISIPs yield an average Shmin gradient of 19.3 ± 2.4 MPa/km, which is roughly 5 MPa/km less than the overburden gradient. No leak off tests, lost circulation, or ballooning incidents were recorded in any of the drilling reports.

5.3.5 SHmax orientations and magnitudes from wellbore failure observations As rock is removed when drilling a well, stresses once supported by the rock become concentrated on the wellbore wall. It is possible for these stresses to exceed the tensional and/or compressive strength of the surrounding rock, leading to formation of drilling-induced tensile fractures (DITFs) and wellbore breakouts, respectively. The presence (and orientation) of wellbore breakouts was originally inferred from wireline caliper logs, although the advent of ultrasonic resistivity and acoustic imaging tools including FMI, OBMI, and UBI microimagers (hereafter collectively denoted wellbore image logs) has provided significantly enhanced breakout resolution. These wellbore image logs display a continuous, unwrapped image of the wellbore wall (i.e. Figure 5.2) and also provide opportunities to visualize DITFs, faults, and fractures intersecting the wellbore wall.

The observation of breakouts and/or DITFs is beneficial from a geomechanical modeling perspective in two ways. First, either breakouts or DITFs can constrain the orientation of the maximum and minimum horizontal stresses acting on the wellbore wall (e.g. Bell and Gough, 1979). In a vertical well with the principal stresses lying in vertical and horizontal planes, breakouts will form in the direction of Shmin while

DITFs will form in the direction of SHmax. Secondly, wellbore breakouts and DITFs can constrain the magnitude of SHmax, which is often the most difficult principal stress magnitude to constrain. Breakouts and DITFs can therefore provide extremely valuable information on the magnitude and orientation of far-field stresses that is otherwise difficult to obtain. In this section, we examine several wellbore image logs penetrating the NC formation in the WC field for evidence of wellbore failure to constrain both the orientation and magnitude of SHmax.

Wellbore breakouts were observed in two wells in the WC field; one vertical (Well A) and one horizontal (Well B) (Figure 5.2). Wells A and B are located less than 2 km apart, and both penetrate the NC formation at about 2,500 meters TVD. Wellbore breakouts and DITFs in Well A were difficult to positively identify given the limited pad coverage and poor OBMI image log resolution (i.e. Figure 5.2a).

However, numerous wellbore enlargements were inferred from the wireline caliper log from Well A. The enlargements were typically located at high angles (> 50°) on the wellbore wall from the slight wellbore deviation direction, suggesting that they are not likely keyseats, which are grooves formed by contact between the rotating drill string and the wellbore wall that can be mistaken for breakouts. Based on this evidence, the enlargements are interpreted as wellbore breakouts, which led to identification of 14 breakouts between 2486 and 2833 meters (TVD) with an average breakout length of 7.6 meters. An example breakout from Well A within the NC formation is shown in Figure 5.2a. Note that the caliper arm orientations at the point of greatest caliper difference correlate with a blurry region in the image log, which likely reflects poor contact between the OBMI tool pad and the fractured rock associated with the breakout. Since no DITFs were identifiable in the OBMI image log and DITFs cannot be resolved from caliper logs, we assume there are no DITFs present in Well A.

The breakouts observed in the vertical Well A can be used to constrain the orientation of SHmax at reservoir depths. Since breakouts form in the direction of Shmin on the wellbore wall, 90° must be added to the observed breakout position on the wellbore wall to estimate the SHmax orientation. Of the 14 breakouts recorded in Well

A, only one was within the NC formation, and it indicated a SHmax orientation of

N24°E, which is consistent with SHmax orientations inferred from wellbore breakouts in nearby wells in the World Stress Map database (Heidbach et al., 2008). The consistency between the SHmax orientation inferred from wellbore enlargements in

Well A and the preexisting regional SHmax measurements corroborates our interpretation of the enlargements as wellbore breakouts.

Figure 5.2: Example breakouts observed in the NC formation in two separate wellbore image logs. (a) Well A; (b) Well B.

Wellbore breakouts were also observed in Well B, a horizontal well drilled in the direction of Shmin. Breakouts were identifiable in the UBI image log as dark regions of low acoustic impedance spaced 180° degrees apart on the top and bottom of the wellbore (Figure 5.2b). Since Well B is a horizontal well, particular care must be taken not to misinterpret potential keyseats as wellbore breakouts. We interpret the dark, low impedance regions in Well B as breakouts since keyseats usually exhibit a smooth and channeled appearance in wellbore image logs, which contrasts with the coarse and uneven appearance of the low impedance features in Well B (Figure 5.2b). In addition, the low impedance regions are systematically located 180° apart on the wellbore wall, which contrasts with the commonly (but not always) asymmetric wellbore enlargement characteristic of keyseats. Ten breakouts with an average length

of two meters and width of 44° were recorded over a 600 meter interval. No DITFs were observed in Well B.

In addition to constraining SHmax orientation, the observation of wellbore breakouts can also be used to constrain the SHmax magnitude within the NC reservoir.

To constrain the SHmax magnitude, we use the widths of the wellbore breakouts around the wellbore wall to model the allowable SHmax magnitudes as constrained by compressive rock strength. For example, in a vertical well, the magnitude of SHmax can be constrained from breakout widths using the following relationship (after Barton et al., 1988)

(C  2Pp  P  T )  S (1 2cos 2 ) S  0 h min b H max 1 2cos 2 b (5.2)

where C0 is the initial rock strength, PP is the pore pressure, ∆P is the drilling mud weight minus the pore pressure, σ∆t is the thermal stress imparted by the mud and formation temperature differences, and θb = π – breakout width. The possible range of

SHmax magnitudes for Well A assuming a breakout width of 40°, the PP, Shmin, and SV values discussed above, and no thermal stresses (temperature data was unavailable) is shown in a stress polygon in Figure 5.3a. Since Well B is a horizontal well, the formulations for SHmax as a function of wellbore breakout width in deviated wells presented in Peska and Zoback (1995) are utilized to constrain SHmax. The possible range of SHmax magnitudes for Well B assuming a breakout width of 44° is shown in a stress polygon in Figure 5.3b.

The dark black outer lines of the stress polygons in Figure 5.3 correspond to stress configurations where shear failure would occur on optimally-oriented faults within the crust, and therefore constrain the allowable stress state to within the polygon. The red lines indicate isobars of compressive rock strength (UCS) in MPa and blue lines represent values of tensile strength (in MPa). Since no DITFs were observed in either Well A or Well B, the allowable stress state must fall to the right of the ‘0’ blue line contour and cannot exceed the maximum UCS estimate. Using these

constraints along with the Shmin estimates discussed above, the allowable stress states are confined within the red polygons shown in Figure 5.3. The allowable stress states for both Well A and Well B indicate a strike-slip faulting regime (SHmax > SV > Shmin). In fact, the allowable stress states actually overlap between the two wells, as is illustrated by the red shaded region in the combined stress polygon in Figure 5.4. The red shaded region provides a precise SHmax estimate between 79.4 and 83.1 MPa (31 and 32.4 MPa/km). The SHmax magnitude represents the final piece of the geomechanical model, the final model parameters of which are summarized in Table 5.1.

Figure 5.3: Stress polygons illustrating the allowable stress states (red polygons) based on borehole breakout widths in (a) Well A and (b) Well B. RF = reverse faulting, SS = strike-slip faulting, and NF = normal faulting. Input parameters are listed next to each polygon. Biot = Biot’s coefficient, Poisson = Poisson’s ratio, Well azi. = wellbore deviation direction, Well dev. = wellbore deviation angle (from vertical), wBO = wellbore breakout width, ΔP = mud weight – pore pressure, µi = coefficient of internal friction, µs = coefficient of sliding friction, TVD = true vertical depth.

Figure 5.4: Combined stress polygon for Wells A and B. The allowable stress state as constrained by wellbore breakout widths in both wells is indicated in red. Stresses are calculated at a depth of 2,560m.

Table 5.1: Final geomechanical model parameters.

5.3.6 Summary A variety of wireline, pressure test, laboratory, and wellbore image log data are used to constrain the orientation and magnitude of the three principal stresses in the WC field as well as the NC formation pore pressure and compressive rock strength. These data indicate a significantly underpressured reservoir within a strike-slip stress regime with SHmax oriented NE-SW. It is worth emphasizing that two of the more commonly difficult geomechanical model parameters to constrain, compressive rock strength and SHmax magnitude, were particularly well constrained in this analysis. Specifically, triaxial laboratory tests and empirical relationships independently yielded consistent compressional rock strength estimates, and consistent SHmax magnitudes were resolved using two independent wellbore breakout observations. This well constrained geomechanical model will be crucial to evaluating the preexisting natural fracture network as a potential flow pathway within the NC formation in the following section.

5.4 Fracture Characterization and Slip Analysis

The primary fractures placed during hydraulic fracture stimulations are Mode I (tensile opening) hydraulic fractures that open in a plane perpendicular to the least compressive stress and serve as a conduit for fluids entering and exiting the reservoir via the stimulated natural fracture network. While Mode I fracture placement and evolution has been reasonably well understood and extensively modeled for decades, much less attention has been given to the influence of the preexisting natural fracture network on permeability enhancement and fluid flow during hydraulic fracture stimulations. The preexisting fracture network is a particularly important flow pathway in slickwater hydraulic fracturing procedures frequently employed in low- permeability reservoir stimulations. This is because slickwater stimulations, which typically utilize water based fracturing fluids instead of highly viscous gels, generally increase fracture and fault permeability by means of shear failure on preexisting fractures and faults within the reservoir (Pearson, 1981; Pine and Batchelor, 1984; Tezuka and Niitsuma, 2000). The preexisting fracture network serving as a potential fluid pathway is physically evident from the spatial distribution of microseismic event hypocenters induced during hydraulic fracture stimulations, which often delineate a complex stimulated fracture network (e.g. Phillips et al., 1996; Rutledge et al., 1998; Rutledge and Phillips, 2003; Maxwell et al., 2002).

While the preexisting fracture network appears to serve as a significant fluid pathway during hydraulic fracture stimulations, a given rock volume may contain fractures (and faults) of widely varying size and orientation. Therefore, it is highly desirable to be able to identify which fractures within a diverse population will serve as the primary fluid pathways. Using a variety of in-situ stress measurements and fracture characterization techniques, numerous studies have demonstrated that preexisting fractures which are optimally-oriented for shear failure in the current stress field (critically-stressed) tend to be those which are conduits for fluid flow (Barton et al., 1995; Finkbeiner et al., 1997; Rogers, 2003). In other words, fracture planes having high ratios of shear to effective normal stress due to increased fluid pressure,

high differential stresses, or low friction tend to be preferential fluid pathways within a reservoir.

This concept, known as the critically-stressed fault hypothesis (Barton et al., 1995; Zoback, 2007), is likely related to permeability enhancement resulting from brecchiation and subsequent dilation due to shear slip on preexisting fractures (e.g. Lee and Cho, 2002). Moreover, the enhanced permeability of critically-stressed fractures has also been shown to influence hydrocarbon production. For example, wellbores intersecting larger numbers of critically-stressed fractures have been correlated with higher production in tight-gas sandstone reservoirs (Tezuka et al., 2002).

The above discussion presents three primary concepts: (1) Preexisting natural fractures can serve as significant to fluid pathways during slickwater hydraulic fracture stimulations; (2) preexisting natural fractures which are critically-stressed in the current stress field tend to be those which are hydraulically conductive; (3) wellbores intersecting more preexisting, critically-stressed natural fractures tend to have higher production compared to wellbores intersecting fewer critically-stressed fractures. With these concepts in mind, the objectives of this section are to characterize the preexisting natural fracture network within the NC formation in the WC field, identify which fractures, if any, are critically-stressed in the current stress field described in Section 3, and evaluate the potential contribution of critically-stressed natural fractures to gas production.

5.4.1 Fracture characterization Wellbore image logs from horizontal sections of five separate wells penetrating the NC formation within the WC field are examined to constrain characteristics (density and orientation) of the preexisting fracture and fault network in contact with the wellbore wall. Four of the wellbore image logs were recorded prior to hydraulic fracturing stimulation (pre-frac) and one was record following a hydraulic fracturing stimulation (post-frac). The wellbore image log data consisted of FMI, OBMI, and

UBI formats, and were provided with lists containing strikes and dips of previously- interpreted fractures, faults, and bedding planes. Although unable to re-pick fractures and bedding planes (due to file format incompatibilities with the picking software), we quality-controlled the previously interpreted fracture and bedding picks by visually inspecting the image logs and discarding any questionable picks. To supplement the wellbore image logs, P-wave anisotropy attributes from a wide-azimuth 3D seismic survey completed over the WC field were available to constrain far-field fracture density and orientation, and are discussed in Appendix 5A.

The orientations of all clearly discernible bedding and fracture planes in the four pre-frac image logs are shown in Figure 5.5. All fractures were visually identified as either “open” or “partially open” since very few appeared definitively closed (due to mineralization, for example). All fracture orientations for each of the pre-frac and the single post-frac well are shown in Figure 5.6. In general, very few fractures were observed in the alluvial plain facies of the NC formation. The vast majority of fractures in the four pre-frac image logs were observed in thin, sandy intervals interspersed within the sandy conglomerate. No dominant fracture sets were identifiable in the pre-frac fracture populations (Figure 5.6). However, the post-frac image log contained a dominant population of steeply-dipping, large-aperture fractures striking parallel to SHmax (Figure 5.6), which is the orientation in which Mode I fractures would be expected to propagate away from the wellbore in the current strike- slip stress field. These fractures may serve as significant fluid pathways into the reservoir, and are discussed in detail in Appendix 5B. Since fractures were inferred from deviated (horizontal) wellbores, it is important to correct for potential sampling biases in the fracture populations (e.g. Terzaghi, 1965, Barton and Zoback, 1992). However, the fracture bias correction only minimally alters the fracture distribution in most wells, and hence only the original uncorrected fracture distributions are reported and analyzed in this study. Only one small fault was recorded in the pre-frac wellbore image logs, and was distinguished by its unusually-large aperture compared to nearby natural fractures. Bedding planes were ubiquitous in all wellbore image logs, and generally indicate sub-horizontal bedding (Figure 5.5).

100

Figure 5.5: Orientation of fractures and bedding planes observed in pre-frac wellbore image logs penetrating the NC formation. Orientations are shown as poles to planes on stereonet diagrams. Blue dots represent wellbore deviation direction.

Figure 5.6: Stereonet plots of fracture orientation in the pre- and post-frac wellbore image logs. Dots represent poles to planes of fractures.

5.4.2 Critically-stressed fractures Since preexisting fractures which are optimally-oriented for shear slip in the current stress field (critically-stressed) tend to be conduits for fluid flow, ascertaining

101

whether or not the preexisting fractures observed in the wellbore image logs are critically-stressed could yield valuable insight into potential flow pathways within the reservoir. This analysis is particularly relevant in the WC field since all wells were completed with open-hole hydraulic fracturing techniques. That is, the entire uncased horizontal wellbore sections were pressurized simultaneously, which allowed elevated injection fluid pressures to contact and potentially stimulate the entire preexisting fracture network intersecting the wellbore wall.

Using the stress magnitudes from the geomechanical model (the center of the red-shaded area in Figure 5.4) along with the Mohr-Coulomb failure criterion, we calculate the pore pressure required to initiate shear failure on all open and partially- open fractures identified in the wellbore image logs. The Mohr-Coulomb failure criterion, assuming no cohesion, is given by:

τ = μ(SN – PP) (5.3) where τ is the shear-stress acting on the fracture plane, SN is the normal stress acting on the fracture plane, μ is the coefficient of fault friction, and PP is the pore pressure. We assume a coefficient of fault friction of 0.6 based on empirical observations from Byerlee (1978) which demonstrated that μ generally falls between 0.6 and 1.0 for a wide variety of rock types over a range of confining pressures. Shaley rocks may exhibit lower friction values, but these are not relevant for the NC formation.

The pore pressure required to initiate shear failure on all open and partially- open fractures referenced to unperturbed reservoir pressures is shown in Figure 5.7a. Note that the slight variation in unperturbed reservoir pressure from well to well is accounted for in Figure 5.7a. None of the preexisting fractures are critically-stressed in the ambient stress field. In fact, even the fractures most well-oriented for shear failure in a strike-slip stress regime (those dipping vertically and striking ± 30˚ from

SHmax) would still require a significantly elevated fluid pressure to experience shear failure. Note that fractures would be closer to shear failure if Shmin were lower, for example (e.g. the left side of the red shaded region in Figure 5.4 closer to the black

102

line bounding the stress polygon). Although not shown in Figure 5.7a, bedding planes recorded from the wellbore image logs would also require a substantially elevated pore pressure to fail in shear.

Figure 5.7: Stereonet and Mohr diagrams indicating the orientation and tendency for shear failure of preexisting fractures in the current stress field under (a) unperturbed reservoir pressure and (b) a 25 MPa increased reservoir pressure. Fractures stimulated in shear are indicated by white dots on the stereonet plots and red dots on the Mohr diagrams. Black arrows on the stereonet diagrams indicate the SHmax orientation (N24°E).

Although none of the preexisting natural fractures are critically-stressed in the ambient stress field, we can investigate their tendency for shear slip under the influence of the elevated fluid pressures attained during hydraulic fracture stimulation. The tendency for shear failure on the preexisting natural fractures under an elevated fluid pressure of 25 MPa is shown in Figure 5.7b. Assuming pore pressures at reservoir depth of about 18 MPa, a 25 MPa pressure increase corresponds to a total pressure of 43 MPa, which is approximately equal to both the fracture gradient and maximum injection pressures (41-48 MPa at 2500 meters). It is clear that nearly all of the preexisting fractures would fail in shear under the increased fluid pressure, and

103

therefore could potentially serve as fluid pathways into the reservoir. As reservoir pressure continues to increase, more fractures would become stimulated in shear. However, as elevated fluid pressures dissipate away from the wellbore (and the placed Mode I hydraulic fractures), only more well-oriented fractures would be stimulated.

5.4.3 Shear-stimulated fractures and wellbore production Although none of the preexisting natural fractures observed in wellbore image logs are critically stressed in the ambient stress field, many preexisting fractures would be activated in shear under the elevated fluid pressures associated with hydraulic fracturing. Since the shear slip can increase fracture permeability, we examine the correlation between shear-stimulated fracture density and cumulative gas production for three wells from which both wellbore image logs and production data were available to ascertain the extent to which the stimulated preexisting natural fracture network serves as a fluid pathway in the reservoir.

Cumulative, two-year production volumes for each of the three wells are compared with the number of shear-stimulated fractures (assuming an elevated reservoir pressure of 25 MPa) intersecting the wellbores (Figure 5.8). The production volumes were normalized to the longest well for which data was available (~2000 meters). The highest production volumes belong to the two wells intersecting the greatest number of shear-stimulated fractures (Wells 1 and 3), suggesting that the shear-stimulated fracture network may serve as a pathway for fluid flow into the wellbore. However, Well 2 is surprisingly productive despite intersecting relatively few shear-stimulated fractures. Note the well numbers correspond with the well numbers shown in Figures 5.5 and 5.6. Since there are no reliable constraints on the far-field fracture network in the WC field, we are assuming the fracture orientations and densities observed in the wellbore image logs are representative of the far-field fracture network, which may not necessarily be true. Additionally, we also assume that the injected fluid volume along the entire injection well is sufficient such that the maximum number of preexisting natural fractures would be stimulated. In reality, once fluid begins to propagate into placed hydraulic fractures, it is unlikely sufficient 104

injected fluid volumes would exist to simulate fractures along the entire wellbore, which would affect the number of preexisting natural fractures actually being stimulated.

Figure 5.8: Shear-stimulated fractures (white poles in stereonet plots) and gas production for three NC formation wells in the WC field. Black triangles in stereonet diagrams indicate wellbore trajectory (all wells are horizontal).

5.4.4 Summary The preexisting natural fracture network within the NC formation is characterized by examining wellbore image logs from five horizontal wells penetrating the NC formation in the WC field. Preexisting natural fractures are observed in all four image logs recorded prior to hydraulic fracture stimulation, although fracture orientations and densities vary considerably both within and between wells. The fracture orientations observed in the single image log recorded following a hydraulic fracture stimulation are consistent with the expected orientation of Mode I hydraulic

105

fractures in the current stress field. Under unperturbed reservoir pressures, none of the preexisting natural fractures are critically-stressed in the current stress field, but many would be stimulated in shear when subject to elevated fluid pressures during hydraulic fracturing. Wellbores intersecting greater numbers of shear-stimulated fractures at elevated injection pressures are generally more productive over an initial two-year period, although wellbores intersecting few shear-stimulated fractures are also remarkably productive. We conclude that the preexisting fracture network in the NC formation appears to serve as a fluid pathway within the reservoir, although to what extent is currently unknown.

5.5 Precise Microseismic Event Locations

As fluid is injected at high pressures into rock formations during hydraulic fracturing stimulations, fractures open and propagate from the wellbore wall into the surrounding formation. In addition, elevated injection pressures can initiate shear slip on preexisting fractures within the reservoir, thereby increasing their permeability and providing potential conduits for fluid flow. The energy released during this deformation is manifested, at least in part, by the occurrence of extremely small earthquakes (hereafter called microseismic events) that can be recorded using downhole sensors positioned near injection wells, although there is also increasing evidence for a significant slow-slip component of seismic reservoir deformation (Das and Zoback, 2011). The spatial distribution of induced microseismic event hypocenters are routinely used to delineate the hydraulic fracture geometry, the extent of the stimulated reservoir volume (e.g. Warpinski et al., 2005), map the geometry of stimulated natural fracture networks (e.g. Phillips et al., 1996; Rutledge et al., 1998), and monitor potential caprock or seal failure.

As microseismic event hypocenter location errors directly affect the accuracy of the above evaluations, it is crucial to obtain the most accurate hypocenter locations possible. Unfortunately, event hypocenter locations resolved using conventional location techniques based on Geiger’s method (Geiger, 1910, 1912) often exhibit

106

errors ranging from 10s to 100s of meters due to poor P- and S-wave phase picks, inadequate or unfavorable monitoring array configurations, and/or poorly constrained velocity models. In naturally-occurring earthquake location studies, the effect of many of these uncertainties has been minimized through the use of waveform cross- correlation to improve phase pick accuracy and/or relative event location techniques (e.g. Poupinet et al., 1984; Frémont and Malone, 1987; Deichmann and Garcia- Fernandez, 1992; Got et al., 1994; Dodge et al., 1995; Waldhauser and Ellsworth, 2000; Rowe et al., 2002; Schaff et al., 2004). The double-difference relocation technique developed by Waldhauser and Ellsworth (2000) combines relative event location with high-precision, cross-correlation differential travel times to reduce location uncertainty arising from unmodeled earth structures. This technique, which is described in detail in the following section, is capable of reducing hypocenter location errors by and order of magnitude and resolving previously undetected tectonic structures (e.g. Waldhauser and Ellsworth, 2000; Schaff et al., 2002; Prejean et al., 2002, 2003).

Despite the well-documented success of the double-difference technique in naturally-occurring earthquake studies, it has rarely been applied to locating microseismic events induced during fluid injection stimulations (Kumano et al., 2006; Miyazawa et al., 2008; Zhou et al. 2010). To examine this potential application in more detail, we perform a microseismic relocation analysis utilizing the double- difference technique to improve relative microseismic event hypocenter locations reported during hydraulic fracture stimulation in the NC formation. The objective is to better resolve potential seismogenic structures serving as flow pathways within the reservoir. Our application will also test the viability of the double-difference event relocation technique in limited monitoring array configurations (e.g. 1 or 2 monitoring arrays) typical of hydraulic fracturing stimulations, which could provide valuable insight into future monitoring array designs.

107

5.5.1 The double-difference technique Traditional earthquake location techniques utilize absolute P- and S-wave arrival times picked from seismograms in conjunction with a layered earth velocity model to calculate earthquake hypocenter locations (Geiger, 1910, 1912). In these techniques, each earthquake is located independently, which makes hypocenter location errors particularly sensitive to unmodeled velocity structures. To minimize this source of error, relative earthquake location techniques have been developed that exploit the clustered nature of earthquake hypocenter populations. Specifically, relative event location techniques assume that the offset between a pair of earthquakes is much smaller than the distance from the earthquakes to a common recording sensor such that the two earthquakes share nearly identical raypaths to the recording sensor. Under such conditions, the P- and S-wave travel time differences between two earthquakes recorded at the same sensor can be largely attributed to the spatial offset between the two earthquakes and not the unmodeled velocity structure between the earthquakes and the sensor. Therefore, by utilizing P- and S-wave differential travel times between earthquake pairs instead of absolute travel times from individual earthquakes, it is possible to better locate the earthquakes in the pair with respect to one another. Furthermore, by allowing individual earthquakes to pair with multiple other earthquakes (e.g. a multiplet), a chain of linked earthquakes can be created, providing the opportunity to improve relative earthquake locations over large regions.

The above principals are the basis of the double-difference earthquake relocation program developed by Waldhauser and Ellsworth (HYPODD, 2000) as well as the double-difference seismic tomography program from Zhang and Thurber (TOMODD, 2003). Both programs have been applied in many naturally-occurring earthquake environments, often yielding greatly improved earthquake hypocenter locations and resolution of tectonic and velocity structures. The latter program will be used in this study, although only in hypocenter relocation mode (see Section 5.5.4). Note that HYPODD and TOMODD both require that initial event hypocenter locations and origin times are known and that absolute P- and S-wave arrival times are available.

108

To illustrate the principles of the double-difference technique mathematically, we begin with the original earthquake location problem. The hypocenter location of an earthquake, i, can be calculated from P- and S-wave travel time data using a set of equations of the form (Geiger, 1910, 1912)

t i t i t i r i  k x i  k y i  k z i  t k x y z o (5.4)

i obs cal i where rk = (t t ) k is the residual between the observed and calculated P- or S-

t i t i t i wave travel time for the kth station, k , k , and k are the partial derivatives of x y z the travel time to the kth station with respect to the earthquake hypocenter coordinates, i i i i and Δx , Δy , Δz , and Δto are the necessary perturbations to the earthquake hypocenter

i coordinates and origin time to minimize rk . Equation (5.4) can be rewritten as

i i t k i rk  m m (5.5)

i i i i i where Δm = [Δx Δy Δz Δto ]. The problem is to iteratively solve for the earthquake hypocenter position and origin time which produces the smallest travel time residual . The formulation presented in equation (5.5) is relevant for absolute P- and S-wave arrival times for a single earthquake recorded at multiple stations.

The double-difference technique utilizes differential travel times between a pair of earthquakes to improve relative hypocenter locations. In other words, the technique computes the difference in equation (5.5) for a pair of events, i and j, such that (5.5) becomes (Fréchet, 1985; Waldhauser and Ellsworth, 2000)

i j ij ij tk i tk j tk ij drk  m - m  m m m m (5.6)

ij ij ij ij ij where Δm = [Δdx Δdy Δdz Δdto ] and

109

dr ij  (t i t j )obs (t i t j )cal k k k k k (5.7)

The partial derivatives with respect to space (x,y,z) and time (t0) are calculated from the initial earthquake hypocenter locations. As in equation (5.5), the solution is obtained by finding the Δdx, Δdy, Δdz, and Δdto that best minimize the residual in equation (5.7). It is important to note that equation (5.6) is applicable for both absolute differential travel times (such as obtained from P- and S-wave phase picks in earthquake catalogs) and relative travel time differences (such as derived from waveform cross-correlation). For a group of earthquake pairs recorded by an array of stations, the equations in (5.9) can be written as a system of linear equations

WGm=Wd (5.8) where G is the matrix of partial derivatives, m is a vector containing the unknown space and time model parameters, d is a vector of the differential travel times between earthquake pairs (5.4), and W is a diagonal weighting matrix. The solution to (5.8) can be found iteratively using either conjugate gradient (LSQR, Paige and Saunders, 1982) or singular value decomposition (SVD) algorithms (Waldhauser and Ellsworth, 2000).

To reiterate, the double-difference technique accepts two types of data to improve relative earthquake hypocenter locations. The first data are the absolute P- and S-wave arrival times, such as would be available from an earthquake catalog. The absolute P- and S-wave arrival time data are then used to calculate absolute differential P- and S-wave travel times between pairs of events (note the differential travel times are computed internally within HYPODD; the user only supplies the absolute P- and S-wave travel times). This first type of data will hereafter be referred to as the “catalog” differential travel times. The second data are the high precision relative differential P- and S-wave travel times calculated from waveform cross-correlation. This data is calculated by the user outside of the HYPODD program, and will hereafter be referred to as the “cross-correlation” differential travel times. Note that the double-

110

difference program can still be executed if only catalog differential travel times are available.

A significant advantage of the double-difference technique developed by Waldhauser and Ellsworth (2000) is the ability to dynamically control earthquake pairing and data weighting between iterations when solving equation (5.8). For example, to take advantage of the cross-correlation differential travel times to finely adjust relative hypocenter locations, the catalog differential travel time data can be downweighted relative to cross-correlation differential travel times, or excluded completely. In addition, earthquake pairing can be adjusted between iterations such that two earthquakes exceeding a specified inter-earthquake distance lose their pairing for the next iteration (and hence are no longer directly located relative to one another). This dynamic pairing and reweighting will be discussed in detail in later sections. The following section describes the monitoring array configuration and microseismic event population on which the double-difference technique was applied in this study. The processing methodology is described in Section 5.5.3.

5.5.2 Stimulation procedure, monitoring array configuration, and microseismic data Several wells within the WC field were microseismically monitored during hydraulic fracture stimulation. The stimulated well considered in this analysis was monitored by three downhole geophone arrays, which was the most optimal array configuration of all the monitored stimulations. The horizontal segment of the injection well extends roughly 1,300m through the NC formation at a depth of 2,400 meters (Figure 5.9). The entire horizontal portion of the uncased injection well was pressurized for approximately six hours during the hydraulic fracture stimulation. The treatment was pumped at an average pressure of 44 MPa and rate of 16 m3/min, and paced 96 tonnes of sand.

The hydraulic fracture stimulation was monitored by three downhole arrays of 3-component geophones located in vertical wellbores (Arrays A, B, and C, Figure 5.9).

111

Arrays A and C both contained 24 geophones while Array B contained 21 geophones, providing a maximum of 69 geophones recording a given microseismic event. Geophone spacing within the monitoring wells varied between 1 and 11 meters, and the arrays covered an aperture of 100-200 meters in depth. The monitoring arrays were located between 200m and 900m from the injection well (Figure 5.9). The monitoring arrays were only in use during the hydraulic fracture stimulation.

Figure 5.9: Map view (left) and west-east cross section (right) illustrating the treatment well orientation and microseismic monitoring array configuration.

The microseismic data was originally processed by a microseismic service company, and their reported event hypocenters locations are shown in Figure 5.10. Event hypocenter locations were reportedly determined using P- and S-wave phase picks from all three recording arrays. Event hypocenters appear to fall into three main clusters aligned in NNW-SSE trends. These hypocenter trends are unusual as they do not reflect the NE-SW SHmax direction in which Mode I hydraulic fractures are expected to propagate in the current strike-slip stress field. Event hypocenters are typically centered about the wellbore, and have reported location errors less than 50 meters. Given the hypocenter clustering and the unusual hypocenter trend, the microseismic dataset provides a good opportunity to apply the double-difference relocation technique to better resolve the ‘cloud-like’ hypocenter distributions and potentially illuminate any reservoir structures seismically activated during hydraulic

112

fracture stimulation. Unfortunately, the initial event hypocenter locations were not provided to us; only an illustration was available for reference (Figure 5.10).

Figure 5.10: Initial event hypocenter locations (710 events) reported by the microseismic service company.

The seismic data available to us for this study included raw velocity seismograms, P-wave polarization data (azimuth and angle of incidence), and absolute P- and S-wave phase pick times for 710 microseismic events recorded during the stimulation. While the 710 events were reportedly recorded on all three monitoring arrays, we discovered that all seismograms from Array C, the furthest array from the injection well, were extremely noisy and lacked identifiable phase arrivals. Consequently, only P- and S-wave phase picks from seismograms recorded by Arrays A and B were used in the relocation analysis. Seismograms were sampled at 4000 Hz, and reported microseismic event magnitudes ranged between MW = -4 and MW = -2.

A 12-layer, 1-D velocity model extending from 2,100 to 2,600 meters TVD which was used by the microseismic service company to obtain the initial hypocenter locations was also provided with the seismic data. The velocity model was derived from wireline sonic logs from the Array A monitoring well and was improved with a 113

joint hypocenter-velocity inversion performed by the microseismic service company. Since the improved velocity model contained several thin layers (< 20 meters), we further refine the model by using the λ/4 criterion to eliminate potential irresolvable layers (Sheriff and Geldart, 1995). Using a dominant P-and S-phase frequency of 50 Hz estimated from event seismograms, we conclude that layers thinner than ~23 meters would be irresolvable, and refine the velocity model by removing or consolidating layers below this thickness. The updated velocity model used for the relocation is shown in Figure 5.11.

Figure 5.11: 1-D velocity model for double-difference relocations. The VP and VS models are represented by red and yellow lines, respectively. The VP sonic log from the Array A well is shown in blue. The depths of the NC formation and recording arrays are also indicated.

5.5.3 Seismic data processing The fundamental input for the HYPODD (and TOMODD) program is absolute P- and S-wave earthquake arrival times (phase picks). Although P-and S-wave phase picks were provided with the microseismic data, it was not revealed to us how they

114

were obtained or quality-controlled. More importantly, some provided picks were inconsistent with the observed P- and S-wave arrivals on event seismograms. The provided P- and S-wave phase picks are therefore discarded, and a new set is selected for the relocation analysis. This section describes how new P- and S-wave phase arrivals are picked as well as how cross-correlation differential travel times are calculated.

The data processing consists of four steps. First, all event seismograms are assigned to groups based on waveform similarity. Second, event seismograms within each group are stacked to form a master trace for each group on which the P- and S- wave arrivals are picked. Third, the P- and S-wave picks on the master trace from each group are used to select absolute P- and S-wave arrivals on the individual event seismograms within each group using waveform cross-correlation. Finally, relative P- and S-wave differential travel times between event pairs within the same group are calculated using waveform cross-correlation.

5.5.3.1 Step 1: Event grouping All 710 microseismic event seismograms are visually sorted into groups based on waveform similarity, yielding 15 distinct groups. The 15 groups are reduced to six groups containing 635 events after discarding groups containing fewer than five events. Since events exhibiting similar waveforms have been shown to share similar hypocenter locations and source mechanisms (e.g. Geller and Mueller, 1980; Rutledge and Phillips, 2003), the distribution of events into distinct groups based on waveform similarity suggests that multiple reservoir structures were seismically deforming during stimulation. With the events distributed into groups, the next step is to stack individual event seismograms to create master traces that will be used to calculate precise absolute P- and S-wave arrival times for individual events within each group.

5.5.3.2 Step 2: Seismogram stacking The stacking procedure is performed on a group-by group basis. For each group defined in Step 1, we first separate seismograms recorded by Array A from

115

seismograms recorded by Array B since seismograms will be stacked separately for each array. Next, we identify which sensor component (east-west (E-W), north-south (N-S), or vertical (Z)) consistently exhibits the clearest P-wave arrivals. Then, for each event, P-wave arrivals are manually picked on individual event seismograms (of the pre-selected component) on as many sensors as possible. If a clear P-wave arrival is identifiable, the seismogram (and seismograms of the remaining two sensor components) is cut into a window spanning 0.05 seconds before and 0.2 seconds after the P-wave pick that also includes the S-wave arrival. Individual seismograms without a clear P-wave arrival are discarded. All the individual seismograms are then aligned on the P-wave arrivals and stacked to obtain a master trace containing the P- and S- wave arrival for each sensor component (Figure 5.12). Finally, the P- and S-wave arrivals are manually picked on each master trace. It is important to emphasize that stacking removes much of the random noise within each individual seismogram, and allowed identification of P- and S-wave arrivals much more accurately than would have been possible on most individual seismograms.

To summarize, the stacking procedure yields a master trace for both Array A and Array B for each of the six groups. Since the master trace for each group is a synthesis of the individual event seismograms within the group, the P- and S-wave arrivals between the master trace and individual seismograms are typically quite similar. This waveform similarity will be capitalized upon in the following step to pick highly-accurate absolute P- and S-wave arrival times for the individual events in each group.

116

Figure 5.12: Example master trace recorded at Array B. The P- and S-wave arrivals are noted. Note the clearer, more impulsive P-wave arrival compared to the S-wave arrival typical of most master traces.

5.5.3.3 Step 3: Picking absolute P- and S-wave arrivals Since the P- and S-wave arrivals on master trace seismograms are similar to P- and S-wave arrivals on individual event seismograms within each group, we can utilize P- and S-waves picked on the master trace to calculate precise P- and S-wave arrival times on the individual event seismograms with waveform cross-correlation. This pick selection procedure is analogous to that used in master-event approaches (e.g. Frémont and Malone, 1987; Deichmann and Garcia-Fernandez, 1992; Reyes- Montes et al., 2009; Kapetanidis and Papadimitriou, 2011; Plenkers et al., 2012), but with two key differences. First, the master traces created in Step 2 are a synthesis of up to hundreds of individual event seismograms. This is in contrast to traditional master-event approaches which typically use a single seismogram, usually from the 117

largest magnitude (and hence best locatable) event, as the master trace. Second, although P- and S-wave arrivals are selected on the master traces in this study, they are not used to directly locate events within the reservoir, and are only used to select precise P- and S-wave phase picks on individual event seismograms. In other words, the cross-correlation lag times we calculate between the master trace and individual event waveforms in this step are only used for picking phase arrivals and not for hypocenter relocation. In contrast, traditional master-event approaches first locate the master event and then locate all subsequent events relative to the master event by calculating differential P- and S-wave travel times. The decision to not choose a single event master trace in this study is appropriate in our limited monitoring array configuration since there is no guarantee that even the largest magnitude event will be well-locatable, or even that P- and S-wave arrivals will be clearly identifiable. In addition, P- and S-wave phase picks can be made with more accuracy on the stacked master traces, which is crucial for accurately picking phase arrivals on individual event seismograms.

To calculate absolute P- and S-wave arrival times on individual event seismograms within each group defined in Step 1, we perform a time-domain waveform cross-correlation between the P- and S-phase arrivals on the master trace and the P- and S-phase arrivals on each individual event seismogram within the group (Schaff et al., 2004). Again, the analysis is performed separately for Array A and B master traces. First, the P- and S-wave arrivals on both the master trace and individual event seismogram are cut into separate 350-sample windows containing the full phase arrival. Then, the window containing the phase arrival from the master trace is cross- correlated with the window containing the phase arrival from the individual event seismogram. The resulting cross-correlation function contains a series of coefficients ranging between 0 and 1 representing the degree of correlation between the two windows for different time shifts (0 indicating no correlation, 1 indicating identical waveforms).

118

If the cross-correlation function maximum is greater than 0.7, we perform a two-step approach to calculate the precise phase arrival time on the individual event seismogram. In the first step, the two seismograms are lined up to the nearest sample according to the time shift which produced the maximum correlation coefficient. In the second step, we calculate the time shift to sub-sample precision by fitting a parabola to the peak of the cross-correlation function (e.g. Schaff et al., 2004). One the total time shift of the master trace relative to the event seismogram is calculated, the phase pick from the master trace is transferred by the appropriate time shift to the individual event seismogram. The new phase pick is visually confirmed on each individual event seismogram to prevent potential cycle skipping during cross- correlation. If the cross-correlation function maximum is below 0.7, the phase was not picked on the individual event seismogram. This cross-correlation analysis was repeated between the master trace and all individual event seismograms within a group for both P- and S-wave arrivals. A total of 394 of the 635 events (62%) have a precise absolute P- or S-wave phase pick on at least one sensor in both Array A and Array B and are used in the double-difference relocation.

In summary, the purpose of this step was to capitalize on the waveform similarity between P- and S-phase arrivals on the master traces and individual event seismograms within a group to precisely pick the absolute P- and S-wave arrivals on individual event seismograms. These absolute P- and S-wave arrival times will be the fundamental input into the HYPODD double-difference relocation program, where they will be used to calculate absolute differential P- and S-wave travel times between event pairs. However, since the double-difference technique can also use cross- correlation differential P- and S-wave travel times, we will use the new absolute P-and S-wave phase picks to compute cross-correlation differential travel times between event pairs in the next step.

5.5.3.4 Step 4: Calculating cross-correlation differential travel times The final data processing step is to calculate relative P- and S-wave differential travel times between event pairs recorded by the same sensor for all individual events

119

within a group. This is accomplished by applying the same time-domain cross- correlation technique described in Step 3. However, instead of calculating differential travel times between the master trace and each individual event trace, relative differential travel times are calculated between all event pairs within a group on as many sensors as possible. Relative differential travel times are discarded if the cross- correlation function maximum is less than 0.8. All cross-correlation results are visually inspected to ensure a quality differential travel time measurement. It is important to reiterate that cross-correlation differential travel times are only calculated between event pairs within the same group and not between event pairs in different groups. This ensures that cross-correlation differential travel times only exist for events having similar waveforms (and therefore similar hypocenter locations).

5.5.4 Double-difference relocations With the 1-D velocity model, absolute P- and S-wave arrival times, and cross- correlation differential travel times available, the microseismic events are relocated using the double-difference technique. The double-difference hypocenter relocations are calculated using the TOMODD software program (Zhang and Thurber, 2003), which is an extension of the original HYPODD double-difference program. In addition to inverting differential travel times to relocate event hypocenters, TOMODD can jointly invert differential travel times to improve the earth velocity model. However, for this study, TOMODD will only be used in hypocenter inversion mode given the limited azimuthal raypath coverage across the stimulated region that is unsuitable for a tomographic study.

The TOMODD software program is selected over the original HYPODD program for two primary reasons. First, the TOMODD program allows for a subsurface monitoring array, and more importantly, for event hypocenters to be located above the monitoring array, which is essential given the hypocenter locations in this study. Second, TOMODD can accommodate a velocity model containing interbedded high and low velocity layers, which is necessary for the WC field (Figure

120

5.11). This section describes running the TOMODD program to obtain precise event hypocenter relocations.

5.5.4.1 Event relocations Initial event hypocenter locations and origin times must be available to relocate event hypocenters with TOMODD. Since initial hypocenter locations were not available from the microseismic service company, initial hypocenter locations were assigned to a synthetic starting grid. Initial event hypocenters within the grid are spaced 50 meters apart in easting and 10 meters in northing and depth. The starting grid is shown in Figure 5.13, and is intended to mimic the hypocenter distribution observed in Figure 5.10. Initial origin times (t0) are estimated from absolute P- and S- wave arrival times from each individual event using the following relationship

    s  p 1 t  p   time time   O time  1 1 V    P  V V  P S  (5.9)

where ptime and stime are the absolute P- and S-wave arrival times and VP and VS are the

P- and S-wave velocities, respectively (VP = 4,850 m/s and VS = 2,950 m/s). Although initial event hypocenter locations and origin times are estimated for this analysis, if the events hypocenters are well-locatable, the double-difference technique should accurately resolve event hypocenters and origin times independent of their initial values. Both P- and S-wave arrival times are used in the relocation.

121

Figure 5.13: Initial event hypocenter starting locations (red circles) for the TOMODD double- difference relocations.

As previously mentioned, dynamic event pairing and data weighting are extremely important aspects of the double-difference technique. Identifying which events should be paired and how the catalog and cross-correlation differential travel times should be weighted during each iteration is crucial to obtaining accurate double- difference relocations, and takes considerable trial and error. All 394 events are initially allowed to pair with each other event to calculate catalog differential travel times. This is simply to ensure that catalog differential travel times for all potential event pairs are available, and does not mean that all events will be paired throughout all iterations. As described in Section 5.5.3.4, cross-correlation differential arrival times are only calculated between event pairs within the same group. Therefore, prior to the first iteration, catalog differential travel times exist between all potential event pairs across all groups, but cross-correlation differential travel times only exist for event pairs within the same group.

The ideal event pairing and data weighting should reduce hypocenter location and origin time residuals after each iteration while simultaneously using as much arrival time information as possible. The optimal data weighting and pairing scheme for the events relocated in this study is shown in Table 5.2. The double-difference

122

equations were solved with the conjugate gradient algorithm (LSQR, Paige and Saunders, 1982).

Table 5.2: TOMODD weighting scheme for optimal event hypocenter relocations. Values of 1.0 and 0 indicate full and minimal weight, respectively. A value of -9 indicates no weighting was specified.

Cross-correlation data Absolute (catalog) data Iterations A priori A priori Distance A priori A priori Distance P-wave S-wave weight P-wave S-wave weight weight weight (separation weight weight (separation in km) in km) 1-9 0.01 0.01 -9 1.0 0.8 0.5 10-19 1.0 0.8 0.5 0.01 0.01 -9

The optimal relocation consisted of 19 total iterations. The first nine iterations downweighted the cross-correlation P- and S-wave differential travel time data and emphasized the catalog differential travel time data. This allows event hypocenters to spread out more during early iterations to better restore absolute event locations. In addition, a distance weight of 0.5 km is applied to the catalog differential travel time data during the first nine iterations, which means that event pairs having an inter-event distance greater than 0.5 km lose their pairing. When this occurs, the catalog differential travel times between the two events are not used during the next iteration. This ensures that events which are poorly locatable (due to inaccurate P- and S-wave picks, for example) do not adversely influence hypocenter location accuracy of better- constrained events through pairing. During iterations 10-19, the catalog differential travel times are downweighted compared to the cross-correlation differential travel times. This allows for fine-scale hypocenter adjustments using the precise cross- correlation differential travel times. Again, event pairs exceeding an inter-event distance of 0.5 km are de-paired in following iterations. Note that events are not re- paired between iterations.

To illustrate the pairing and weighting effects, hypocenter relocations following several iterations are shown in Figure 5.14. After the first iteration, which again is largely controlled by the catalog differential travel times, the event hypocenters fall into two clusters centered about the injection well. Note the

123

hypocenter distribution is similar in appearance to the initial microseismic service company locations in Figure 5.10. Through iterations one to eight, event hypocenter distributions tighten notably within the clusters. After iteration 10, event hypocenters dramatically collapse into two north-south-trending clusters, which reflect emphasizing the high-precision cross-correlation differential travel times during the iteration. The dramatic collapsing of event hypocenters also illustrates the quality of the cross-correlation differential travel times calculated in Section 5.5.3.4 as well as appropriate event pairing and weighting. Through the last iterations (13-19), event hypocenters continually adjust on a fine scale driven by cross-correlation differential travel times. The final relocations are shown in iteration 19 in Figure 5.14 and in Figure 5.15a. None of the 394 events were removed (due to de-pairing from all other events) during the relocation, which also indicates an appropriate weighting and pairing scheme was applied.

124

Figure 5.14: TOMODD relocation iterations.

125

Figure 5.14 (cont)

126

The final microseismic event hypocenter relocations generally cluster about the injection well in map view and in cross-section (Figure 5.15a). The hypocenters generally fall within ±50 meters of the injection well in depth, suggesting that the stimulation is contained within the target reservoir. The hypocenter relocations also delineate an unusual approximately N-S trend in map view (Figure 5.15a) that does not reflect the orientation in which Mode I hydraulic fractures would be expected to propagate in the current stress field (Figure 5.6, Appendix 5B). However, interestingly, the vertically-oriented, N-S hypocenter trends do reflect the optimal orientation of faults in the current strike-slip stress field (~30° from SHmax, red dashed lines in Figure 5.15a). The hypocenter trends are similar to those in the initial event locations from the microseismic service company, although the service company locations are typically less tightly clustered and locate below the target reservoir

127

(Figure 5.15b). It is also important to note that the N-S hypocenter trend is perpendicular to a plane connecting Arrays A and B, and is therefore constrained by few travel times in this direction. It is therefore crucial to evaluate the resolution of the N-S trend before further speculating on its potential origins.

5.5.4.2 Event hypocenter resolution A simple test is performed to evaluate the relocated event hypocenter resolution. In this test, all initial event starting locations are randomly reassigned within the starting grid and the relocation analysis is repeated. That is, each event still occupies a point in the grid, but the point it occupies is different from the one occupied in the original relocation. The test examines the hypocenter relocation sensitivity to different initial starting locations. A similar test performed by Waldhauser and Ellsworth (2000) demonstrated the robustness of the double-difference technique in accurately resolving event hypocenter trends given a variety of different starting locations (see Figure 7 in Waldhauser and Ellsworth, 2000). Thus, if the relocations are well-constrained, the individual events should relocate to approximately the same hypocenter locations independent of the initial starting location. This test is repeated ten times with ten different sets of starting locations. The test uses the same arrival times, velocity model, and weighting scheme as the original relocations. After all ten relocation sets were obtained, the difference in E-W (x), N-S (y), and depth (z) between each original event hypocenter relocation and the same event in the test relocation sets was calculated. The combined x, y, and z differences between the original relocations and all relocated test sets are shown in Figure 5.16. Note the plots are arranged to illustrate the x, y, and z differences in map view and cross-section, just as with the original relocations.

128

Figure 5.16: EW, NS, and Z differential hypocenter relocations between original event hypocenter relocations and the relocated test sets.

Ideally, the blue dots in Figure 5.16 would all plot at (0, 0). This would indicate that the event hypocenters resolve to the same exact location regardless of their initial starting location. However, the actual hypocenter relocation differences between the original event relocations and relocated test sets illustrate more complex results. Most notably, event hypocenter relocations in the N-S direction appear highly sensitive to the initial event starting location, with some hypocenter differences exceeding 100 meters. Excluding hypocenter differences more the 500 meters, the average hypocenter shift for each event in the N-S direction is 10.8 ± 36.8 meters. The lack of resolution suggests that the double-difference technique cannot precisely constrain event hypocenter locations in the N-S direction with the existing monitoring array configuration. A similar lack of resolution is observed in depth, with an average shift of 5.6 ± 23.6 meters (excluding differences > 500 meters). Conversely, most hypocenter relocations are reasonably well-resolved in the E-W direction (average shift: 2.8 ± 12.5 meters), which is expected given that most travel times are constrained in this direction.

The resolution test demonstrates a notable lack of resolution in the N-S direction, leaving the origin of the N-S relocated event hypocenter trends unclear. The N-S hypocenter trends could reflect seismic deformation on well-oriented strike-slip faults in the current stress field, severely limited resolution from unfavorable

129

monitoring array configuration, or a combination thereof. If the hypocenter trend reflected seismic deformation on a fault, the events falling along the trend might have higher magnitudes than events outside the hypocenter trend. Unfortunately, event magnitudes were not provided to us to examine this possibility. The hypocenter trends could also be validated by constructing earthquake focal plane mechanisms of individual events, although the prospects of obtaining a set of well constrained focal plane mechanisms in the limited monitoring array configuration are quite small. In addition, it may also be possible to validate the hypocenter trends based on the moveout of phases observed on seismograms across the two monitoring arrays. While the third monitoring array (Array C, Figure 5.9) would have provided an additional set of travel time constraints had the seismograms been of higher signal quality, its position would not greatly improve azimuthal coverage, possibly limiting its effectiveness in improving the double-difference relocations.

We therefore conclude that the origin of the N-S hypocenter trend is irresolvable without additional information or analyses, but appears to be at least partially related to an unfavorable monitoring array configuration. Consequently, we cannot delineate any definitive fault trends (and hence flow paths) from the precise microseismic event hypocenter relocations.

5.5.5 Summary We utilize the double-difference relocation technique to improve hypocenter locations of microseismic events induced during a hydraulic fracturing stimulation in the WC field. Seismograms from three downhole monitoring arrays are available for 710 microseismic events, although our analysis only considered seismograms from the nearer two arrays due to the low signal-to-noise ratio of seismograms from the third, most distal array. All 710 events were visually sorted into groups based on event waveform similarity, which yielded six groups containing 635 events. All the seismograms in each group were then stacked to create a single master trace for each group which was used to calculate precise absolute P- and S-wave arrival times on individual event seismograms within each group using waveform cross-correlation. 130

Next, waveform cross-correlation was used to calculate high-precision relative P- and S-wave differential travel times between pairs of events within the same group recorded by the same sensor. The absolute P- and S-wave arrival times as well as the precise cross-correlation differential travel times were then used as input into the double-difference program to calculate precise hypocenter relocations.

The precise microseismic event hypocenter relocations were performed using the TOMODD software program with a 1-D velocity model. Both catalog- and cross- correlation differential travel times were used during the relocation. The double- difference technique significantly tightens the cluster of initial hypocenter locations into two N-S hypocenter trends centered about the wellbore and extending up to hundreds of meters into the reservoir. While the N-S hypocenter trends are inconsistent with the expected Mode I hydraulic fracture propagation direction, they do reflect the orientation of well-oriented faults in the current strike-slip stress regime in the WC field, suggesting that faults may be reactivated during hydraulic fracturing. However, hypocenter relocation resolution tests indicate event hypocenters are poorly resolved in the N-S direction, suggesting the N-S trends are at least in part artifacts due to limited monitoring array configurations and not actual reservoir structures stimulated during hydraulic fracturing. Thus, while the double-difference technique does greatly reduce the scatter in the event hypocenter relocations, hypocenter resolution is severely limited by an unfavorable monitoring array configuration, which precludes any definitive interpretation of reservoir structures or flow pathways into the reservoir.

5.6 Discussion

The geomechanical model developed in Section 5.3 indicates a strike-slip stress regime with SHmax oriented N24°E. The geomechanical model is well-constrained by wellbore failure observations, instantaneous shut-in pressures (ISIPs) from hydraulic fracturing stimulations, and laboratory rock strength measurements. It is worth reiterating that the magnitude of SHmax, which is typically the most difficult

131

geomechanical model parameter to constrain, is particularly well-constrained from independent wellbore breakout observations in two separate wellbores. In addition, the orientation of SHmax is directly constrained from a wellbore breakout observed within the NC formation, and is consistent with SHmax orientations inferred from wellbore breakouts in nearby wells.

The microseismic event hypocenter relocations in Section 5.5.4.1 delineate vertical, approximately N-S trends extending away from the wellbore. This hypocenter trend is similar to that observed in the original microseismic service company hypocenter locations, but is significantly tighter due to the application of the double-difference relocation technique and the quality catalog and cross-correlation differential travel times used during the relocation. In addition, the significant hypocenter tightening between iterations (with no events needing to be removed during all iterations) demonstrates that an appropriate weighting and pairing scheme was used during relocation.

As mentioned, the relocated microseismic event hypocenter trends strike approximately 30° from the SHmax, which corresponds to the orientation of well- oriented faults in a strike-slip stress regime. This is a particularly enticing observation since the WC field is characterized by a strike-slip stress regime, and suggests that the hypocenter trends could indicate deformation on strike-slip faults reactivated during hydraulic fracture stimulation. However, no faults are known to exist in the WC field, and hypocenter resolution tests indicate poor resolution in the N-S direction due to the limited monitoring array configuration. Therefore, it is unclear whether the trends indicate seismic deformation along a fault or relocation artifacts due to a limited monitoring array configuration. This observation demonstrates that hypocenter trends must always be interpreted with caution, even if they appear to indicate geologically- reasonable features.

Although not examined in this study, we note there is evidence for long-period, long-duration seismic deformation occurring within the tight-gas reservoir, although its location and prevalence is unknown (I. Das, personal communication). It is unclear

132

to what extent this deformation may contribute to fluid flow in the reservoir, although it is expected that slow-slip could be a significant stimulation mechanism during hydraulic fracturing (Zoback et al., 2012).

5.7 Conclusions

In this chapter we applied an integrated, geomechanics-based analysis to identify flow pathways stimulated during a hydraulic fracture stimulation in a low- permeability tight-gas sandstone reservoir. The analysis consisted of three primary steps. First, we developed a geomechanical model to constrain the in-situ stresses, pore pressures, and rock strength within the target reservoir. Next, we utilized wellbore image logs to characterize the preexisting natural fracture network within the reservoir and evaluated whether or not the fractures were critically-stressed in the current stress field and therefore potential conduits for fluid flow. Finally, we employed the double-difference earthquake relocation technique to obtain precise hypocenter relocations of microseismic events induced during a hydraulic fracturing stimulation in order to delineate potential reservoir structures serving as flow pathways into the reservoir. The primary results and conclusions are summarized in the following points.

(1) The principal stress magnitudes and orientations are well-constrained in the study reservoir through wireline logs, hydraulic fracturing shut-in pressure measurements, breakout observations in wellbore image logs, and laboratory rock strength data. The reservoir is significantly underpressured (~7 MPa/km) and is characterized by a strike- slip stress regime (SHmax > SV > Shmin) with SHmax oriented N24°E.

(2) Wellbore image logs indicate that preexisting natural fractures are present within the target reservoir, although their orientation and density is highly variable between wells. In addition, a post-hydraulic fracture wellbore image log indicates an abundance of placed hydraulic fractures striking parallel to the direction of SHmax and dipping nearly vertical, consistent with expected characteristics of Mode I hydraulic

133

fractures in the current stress field. None of the preexisting natural fractures are critically-stressed in the ambient stress field, although many preexisting natural fractures would be stimulated in shear with increased fluid pressures during hydraulic fracturing stimulation. Wellbores intersecting a greater number of shear-stimulated fractures tend to yield higher production, suggesting that the stimulated fracture network may serve as a fluid pathway into the reservoir.

(3) The double-difference earthquake relocation technique significantly reduces scatter in hypocenter locations of induced microseismic events in the target reservoir. In particular, the use of high-precision, relative differential P- and S-wave travel times significantly collapses event hypocenters into distinct N-S trends extending hundreds of meters into the reservoir. Unfortunately, resolution tests indicate that hypocenters are poorly constrained in the N-S direction due to the limited monitoring array configuration, suggesting the trends may not reflect actual reservoir structures stimulated during hydraulic fracturing. We therefore cannot definitively identify any fluid pathways from the microseismic event hypocenter relocations in this study. However, we conclude that the double-difference technique appears to be able to significantly reduce hypocenter scatter in downhole monitoring configurations, and could be a highly viable technique to identify flow pathways provided a well- distributed monitoring array is available.

(4) Our analysis suggests that both stimulated preexisting natural fractures and Mode I hydraulic fractures serve as potential fluid pathways into the tight-gas reservoir, although the relative flow contribution from each is unclear.

(5) Since all wells within the study area were open-hole hydraulic fracture stimulations, it is unfortunately impossible to evaluate the potential effectiveness of an open hole versus a traditional staged stimulation in the study area. However, given the influence the preexisting natural fracture network appears to have on production, uncased, open-hole completions may be the optimal design as they provide the maximum contact between the wellbore and the preexisting natural fracture network

134

within the reservoir and at a lesser cost than traditional, staged hydraulic fracture stimulations.

5.8 References

Abass, H.H, Hedayati, S., and Meadows, D.L., 1996, Nonplanar fracture propagation from a horizontal wellbore: Experimental study: Society of Petroleum Engineers Production & Facilities, 11, 133- 137. Barton, C.A., and Zoback, M.D., 1988, In-situ stress orientation and magnitude at the Fenton Geothermal Site, New Mexico, determined from wellbore breakouts: Geophysical Research Letters, 15, 467-470. Barton, C.A., and Zoback, M.D., 1992, Self-similar distribution and properties of macroscopic fractures at depth in crystalline rock in the Cajon Pass scientific drill hole: Journal of Geophysical Research, 97, 5181-5200. Barton, C.A., Zoback, M.D., and Moos, D., 1995, Fluid flow along potentially active faults in crystalline rock: Geology, 23, 683-686. Bell, J.S., and Gough, D.I., 1979, Northeast-southwest compressive stress in Alberta evidence from oil wells: Earth and Planetary Science Letters, 45, 475-482. Byerlee, J.D., 1978, Friction of rock: Pure and Applied Geophysics, 116, 615-626. Crampin, S., 1991, Wave propagation through fluid-filled inclusions of various shapes: Interpretation of extensive-dilatancy anisotropy: Geophysical Journal International, 104, 611-623. Creaney, S., and Allan, J., 1990, Hydrocarbon generation and migration in the Western Canada sedimentary basin, In: Brooks, J., ed., Classic Petroleum Provinces, Geological Society of London Special Publication No 50, 189-202. Daneshy, A.A., 1973, A study of inclined hydraulic fractures: Society of Petroleum Engineering Journal, 13, 61-68. Das, I., and Zoback, M.D., 2011, Long-period, long-duration seismic events during hydraulic fracture stimulation of a shale gas reservoir: The Leading Edge, 30, 778-786. Deichmann, N., and Garcia-Fernandez, M., 1992, Rupture geometry from high-precision relative hypocentre locations of microearthquake clusters: Geophysical Journal International, 110, 501-517. Dodge, D.A., Beroza, G.C., and Ellsworth, W.L., 1995, Foreshock sequence of the 1992 Landers, California, earthquake and its implications for earthquake nucleation: Journal of Geophysical Research, 100, 9865-9880. Fehler, M., House, L., and Kaieda, H., 1987, Determining planes along which earthquakes occur: Method and application to earthquakes accompanying hydraulic fracturing: Journal of Geophysical Research, 92, 9407-9414. Finkbeiner, T., Barton, C.A., and Zoback, M.D., 1997, Relationships among in-situ stress, fractures and faults, and fluid flow: Monterey Formation, Santa Maria Basin, California: American Association of Petroleum Geologists Bulletin, 81, 1975-1999. Fréchet, J., 1985, Sismogenèse et doublets sismiques, Thèse d’Etat, Université Scientifique et Médicale de Grenoble, 206 pp. Frémont, M.-J., and Malone, S.D., 1987, High precision relative locations of earthquakes at Mount St. Helens, Washington: Journal of Geophysical Research, 92, 10223-10236. Geiger, L., 1910, Herdbestimming bei Erdbeben aus den Ankunftszeiten: K. Gessell. Wiss. Goett., 4, 331-349. Geiger, L., 1912, Probability method for the determination of earthquake epicenters from the arrival time only: Bulletin of Saint Louis University, 8, 60-71. Geller, R.J., and Mueller, C.S., 1980, Four similar earthquakes in central California: Geophysical Research Letters, 7, 821-824.

135

Gies, R.M., 1984, Case history for a major Alberta Deep Basin gas trap: The Cadomin Formation, In: Masters, J.A., ed., Elmworth – Case Study of a Deep Basin Gas Field, American Association of Petroleum Geologists Memoirs, 38, 115-140. Got, J.-L., Fréchet, J., and Klein, F.W., 1994, Deep fault plane geometry inferred from multiplet relative relocation beneath the south flank of Kilauea: Journal of Geophysical Research, 99, 15375-15386. Gray, D., 2008, Fracture detection using 3D seismic azimuthal AVO: Canadian Society of Exploration Geophysicists Recorder, 33, 38-49. Heidbach, O., Tingay, M., Barth, A., Reinecker, J., Kurfeß, D., and Müller, B., 2008, The world stress map database release 2008, www.world-stress-map.org, last accessed Jan. 2010. Hubbert, M.K., and Willis, D.G., 1957, Mechanics of hydraulic fracturing: Petroleum Transactions AIME, 210, 153-163. Ito, T., Zoback, M.D., and Peska, P., 2001, Utilization of mud weights in excess of the least principal stress to stabilize wellbores: Theory and practical examples: Society of Petroleum Engineers Drilling and Completions, 16, 221-229. Kapetanidis, V., and Papadimitriou, P., 2011, Estimation of arrival-times in intense seismic sequences using a master-events methodology based on waveform similarity: Geophysical Journal International, 187, 889-917. Kovalsky, K., and Kelly, P., 2008, Advances in tight-gas completions – Noel tight-gas project: SPE paper 118350 presented at the 2008 Abu Dhabi International Petroleum Exhibition and Conference held in Abu Dhabi, UAE, 3-6 November 2008. Kumano, Y., Asanuma, H., Niitsuma, H., Tezuka, K., and Kamitsuji, R., 2006, Reservoir structure at Yufutsu gas field, Japan determined by analysis of induced microseismic multiplets: Abstract presented at the 2006 SEG Annual Meeting in New Orleans, LA, USA. Lee, H.S., and Cho, T.F., 2002, Hydraulic characteristics of rough fractures in linear flow under normal and shear load: Rock Mechanics and Rock Engineering, 35, 299–318. Lynn, H.B., Simon, K.M., Bates, C.R., and Van Dok, R., Azimuthal anisotropy in P-wave 3-D (multiazimuth) data: The Leading Edge, 15, 923-928. McLean, J.R., 1977, The Cadomin Formation; stratigraphy, sedimentology and tectonic implications: Bulletin of Canadian Petroleum Geology, 25, 792-827. Maxwell S., Urbancic, T.I., Steinsberger, N., and Zinno, R., 2002, Microseismic imaging of hydraulic fracture complexity in the Barnett Shale: SPE paper 77440 presented at the SPE Annual Technical Conference and Exhibition held in San Antonio, Texas, 29 September-2 October 2002. Miyazawa, M., Venkataraman, A., Snieder, R., and Payne, M.A., 2008, Analysis of microearthquake data at Cold Lake and its applications to reservoir monitoring: Geophysics, 73, O15-O21. Moos, D., Zoback, M.D., and Bailey, L., 1999, Feasibility study of the stability of openhole multilaterals, Cook Inlet, Alaska: SPE paper 52186 presented at the 1999 SPE Mid-Continent Operations Symposium held in Oklahoma City, Oklahoma, 28-31 March 1999. Moos, D., Peska, P., Finkbeiner, T., and Zoback, M.D., 2003, Comprehensive wellbore stability analysis using quantitative risk assessment: Journal of Petroleum Science and Engineering, 38, 97- 109. Nordheimer, D., and Jones, W., 2011, Extracting more from less – The development of the Noel Cadomin resource play: Paper presented at the 2011 Canadian Society of Petroleum Geologists, Canadian Society of Exploration Geophysicists, and Canadian Well Logging Society Annual Convention, May 9-13, Calgary, Alberta, Canada. Paige, C.C., and Saunders, M.A., 1982, LSQR: Sparse linear equations and least squares problems: ACM Transactions on Mathematical Software, 8, 195-209. Pearson, C., 1981, The relationship between microseismicity and high pore pressures during hydraulic stimulation experiments in low permeability granitic rocks: Journal of Geophysical Research, 86, 7855-7864. Peska, P. and Zoback, M.D., 1995, Compressive and tensile failure of inclined well bores and determination of in situ stress and rock strength: Journal of Geophysical Research, 100, 12791- 12811. Phillips, W.S., Rutledge, J.T., Fairbanks, T.D., Gardner, T.L., Miller, M.E., and Schuessler, B.K., 1996, Reservoir fracture mapping using microearthquakes: Austin chalk, Giddings Field, TX and 76

136

Field, Clinton Co., KY: SPE paper 36651 presented at the 1996 SPE Annual Technical Conference and Exhibition held in Denver, Colorado, USA, 6-9 October 1996. Pine, R.J., and Batchelor, A.S., 1984, Downward migration of shearing in jointed rock during hydraulic injections: International Journal of Rock Mechanics, Mining Sciences & Geomechanics Abstracts, 21, 249-263. Plenkers, K., Ritter, J.R.R., and Schindler, M., 2012, Low signal-to-noise event detection based on waveform stacking and cross-correlation: application to a stimulation experiment: Journal of Seismology, in press. Porter, J.W., Price, R.A., and McCrossan, R.G., 1982, The Western Canada sedimentary basin: Philosophical Transactions of the Royal Society of London. Series A, Mathematical and Physical Sciences, 305, 169-192. Poupinet, G., Ellsworth, W.L., and Fréchet, J., Monitoring velocity variations in the crust using earthquake doublets: An application to the Calaveras Fault, California: Journal of Geophysical Research, 89, 5719-5731. Prejean, S., Ellsworth, W., Zoback, M.D., and Waldhauser, F., 2002, Fault structure and kinematics of the Long Valley Caldera region, California, revealed by high-accuracy earthquake hypocenters and focal mechanism stress inversions: Journal of Geophysical Research, 107, B12. Prejean, S., Stork, A., Ellsworth, W., Hill, D., and Julian, B., 2003, High precision earthquake locations reveal seismogenic structure beneath Mammoth Mountain, California: Geophysical Research Letters, 30. Reyes-Montes, J.M., Pettitt, W.S., Hemmings, B., Haycox, J.R., Andrews, J.R., and Young, R.P., 2009, Application of relative location techniques to induced microseismicity from hydraulic fracturing: SPE paper 124620 presented at the SPE Annual Technical Conference and Exhibition in New Orleans, Louisiana, USA, 4-7 October 2009. Rogers, S.F., 2003, Critical stress-related permeability in fractured rocks, In: Ameen, M., ed., Fracture and in-situ stress characterization of hydrocarbon reservoirs, Geological Society of London Special Publication 209, 7-16. Rowe, C.A., Aster, R.C., Phillips, W.S., Jones, R.H., Borchers, B., and Fehler, M.C., 2002, Using automated, high-precision repacking to improve delineation of microseismic structures at the Soultz Geothermal Reservoir: Pure and Applied Geophysics, 159, 563-596. Rüger, A., and Tsvankin, I., 1997, Using AVO for fracture detection: Analytic basis and practical solutions: The Leading Edge, 16, 1429-1434. Rutledge, J.T., Phillips, W.S., and Schuessler, B.K., 1998, Reservoir characterization using oil- production-induced microseismicity, Clinton County, Kentucky: Tectonophysics, 289, 129-152. Rutledge, J.T., and Phillips, W.S., 2003, Hydraulic stimulation of natural fractures as revealed by induced microearthquakes, Carthage Cotton Valley gas field, east Texas: Geophysics, 68, 441-452. Schaff, D.P., Bokelmann, G.H.R., Beroza, G.C., Waldhauser, F., and Ellsworth, W.L., 2002, High- resolution imaging of Calaveras Fault seismicity: Journal of Geophysical Research, 107, B9. Schaff, D.P., Bokelmann, G.H.R., Ellsworth, W.L., Zanzerkia, E., Waldhauser, F., and Beroza, G.C., 2004, Optimizing correlation techniques for improved earthquake location: Bulletin of the Seismological Society of America, 94, 705-721. Sheriff, R.E., and Geldart, L.P., 1995, Exploration Seismology, 2nd edition, New York, Cambridge University Press. Terzaghi, R., 1965, Sources of error in joint surveys: Geotechnique, 15, 287-304. Tezuka, K., and Niitsuma, H., 2000, Stress estimated using microseismic clusters and its relationship to the fracture system of the Hijiori hot dry rock reservoir: Engineering Geology, 56, 47-62. Tezuka, K., Namikawa, T., Tamagawa, T., Day-Lewis, A., and Barton, C., 2002, Roles of the fracture system and state of stress for gas production from the basement reservoir in Hokkaido, Japan: SPE paper 75074 presented at the SPE Gas Technology Symposium held in Calgary, Alberta, Canada, 30 April – 2 May 2002. Waldhauser, F. and Ellsworth, W., 2000, A double-difference earthquake location algorithm: Method and application to the northern Hayward Fault, California: Bulletin of the Seismological Society of America, 90, 1353-1368.

137

Warpinski, N.R., Kramm, R.C., Heinze, J.R., and Waltman, C.K., 2005, Comparison of single- and dual-array microseismic mapping techniques in the Barnett Shale: SPE paper 95568 presented at the SPE Annual Technical Conference and Exhibition in Dallas, Texas, USA, 9-12 October 2005. Williams-Stroud, S., 2008, Using microseismic events to constrain fracture network models and implications for generating fracture flow properties for reservoir simulation: SPE paper 119895 presented at the 2008 SPE Shale Gas Production Conference held in Fort Worth, Texas, USA, 16- 18 November 2008. Yew, C.H., and Li, Y., 1988, Fracturing of a deviated well: Society of Petroleum Engineers Production Engineering, 3, 429-437. Zhang, H., and Thurber, C.H., 2003, Double-difference tomography: The method and its application to the Hayward Fault, California: Bulletin of the Seismological Society of America, 93, 1875-1889. Zhou, R., Huang, L., and Rutledge, J., 2010, Microseismic event location for monitoring CO2 injection using double-difference tomography: The Leading Edge, 29, 208-214. Zoback, M.D., 2007, Reservoir Geomechanics, Cambridge University Press, Cambridge, United Kingdom. Zoback, M.L., and Zoback, M.D., 1980, State of stress in the conterminous United States: Journal of Geophysical Research, 85, 6113-6156. Zoback, M.D., Barton, C.A., Brudy, M., Castillo, D.A., Finkbeiner, T., Grollimund, B.R., Moos, D.B., Peska, P., Ward, C.D., and Wiprut, D.J., 2003, Determination of stress orientation and magnitude in deep wells: International Journal of Rock Mechanics & Mining Sciences, 40, 1049-1076. Zoback, M.D., Kohli, A., Das, I., and McClure, M., 2012, The importance of slow-slip on faults during hydraulic fracturing stimulation of shale gas reservoirs: SPE paper 155476 presented at the Americas Unconventional Resources Conference in Pittsburg, Pennsylvania, USA, 5-7 June 2012.

138

Appendix 5A: Fracture Constraints From an AVAZ Analysis

Over the last couple decades, many studies have attempted to extract hydrocarbon reservoir fracture characteristics from wide-azimuth 3D reflection seismic surveys using the amplitude versus angle and azimuth (AVAZ) technique (e.g. Lynn et al., 1996, Ruger and Tsvankin, 1997). This technique assumes that a hydrocarbon reservoir is characterized by a set of vertical, open, fluid-filled fractures (horizontal transverse anisotropy, HTI) and exploits the azimuthal dependency of P- wave amplitudes to infer fracture characteristics. Specifically, the technique uses the orientation of the fast P-wave and the velocity difference between the fast and slow P- waves to infer fracture orientation and fracture density, respectively. An overview of the AVAZ method, its assumptions, and several case studies demonstrating the validity of fracture characteristics inferred from the technique is given in Gray (2008).

In this section, we examine P-wave fast azimuths and fast and slow P-wave velocity differences reported from an AVAZ analysis conducted over the WC field prior to hydraulic fracture stimulation in the NC formation. The analysis will hopefully provide an opportunity to constrain fracture characteristics (orientation and density) away from wellbores in the WC field to supplement the near-wellbore fracture network characteristics inferred from the wellbore image logs. The P-wave anisotropy magnitude (Vfast – Vslow in milliseconds) and fast P-wave azimuth (0-360) measured at the top of the NC formation is shown in Figure 5A1a-b. P-wave anisotropy and fast azimuths are calculated at common depth points spaced every 10 meters in Northing and 30 meters in Easting. P-wave fast azimuths at common depth points with Vfast – Vslow < 100 ms are discarded to ensure only reliable anisotropy measurements are shown.

139

Figure 5A1: Attribute maps from an AVAZ analysis over the WC field showing (a) P-wave anisotropy (Vfast – Vslow) and (b) fast P-wave azimuth. Only gather points with Vfast – Vslow > 100 ms are shown in the Vfast azimuth map. The inset in (b) is a rose diagram indicating the strike of Vfast from all data points in the map.

There are two notable observations in Figure 5A1. First, the WC field is generally characterized by low P-wave anisotropy magnitudes (< 150 ms) with regions of elevated anisotropy (warmer colors) appearing in NW-SE elongated trends, particularly in the southern and eastern portions of the field (Figure 5A1a). It is not immediately clear what this NW-SE trend represents. The NC formation is locally thin or absent in NW-SE-oriented belts parallel to the NW-SE paleo Spirit River channel with some evidence of clay and shale channel infilling from the overlying formation (Gies, 1984), which suggests that the elevated anisotropy could reflect local geological variations. The elevated anisotropic regions also appear to be associated with a NW-SE trending fast P-wave azimuth, although the unusually systematic NW- SE fast P-wave trends (the tan/orange-colored ‘striping’ pattern in Figure 5A1b) might reflect a potential acquisition artifact.

The second notable observation in Figure 5A1 is that the vast majority of fast P-wave azimuths across the WC field illustrate a dominant NE-SW trend, roughly

140

parallel to SHmax in the region (see inset in Figure 5A1b). One potential mechanism to explain this observation is known as extensive dilatancy anisotropy (EDA) in which anisotropy is controlled by vertical, fluid-filled microcracks preferentially aligned by, and parallel to, the current maximum horizontal stress (Crampin, 1991). Therefore, the fast P-wave azimuth in a crustal volume characterized by EDA would be parallel to the strike of SHmax. While this mechanism (essentially a form of HTI) was originally hypothesized to explain S-wave velocity anisotropy, its effect on P-wave velocities, and hence AVAZ results, can also be substantial (e.g. Ruger and Tsvankin, 1997).

Since fracture characteristics from numerous wellbore image logs are available from the WC field (Figure 5.6), it is possible to compare wellbore image log fracture characteristics with fast P-wave orientations from the AVAZ analysis. Fast P-wave orientations in the vicinity of each well from which a wellbore image log was available are compared to the image log fracture characteristics in Figure 5A2. The magnitude of P-wave anisotropy in the vicinity of each well is also shown in Figure 5A2. There does not appear to be a clear correlation between the fracture orientations inferred from the wellbore image logs and the fast P-wave orientations from the AVAZ analysis for any of the wells. Additionally, wellbores within regions of high anisotropy do not appear to have an increased fracture density (i.e. Well 1). There could be multiple reasons for the observed discrepancy. For example, the wellbore image log and the AVAZ analysis could be sensitive to different fracture populations, thus making the results incomparable. Alternatively, with the exception of Well 1, most of the fractures observed in the wellbore image logs dip at moderate to shallow angles, and are therefore not consistent with the HTI anisotropy assumption on which the AVAZ analysis is based and interpreted.

141

Figure 5A2: Comparison between fracture orientations observed from wellbore image logs (top row of stereonet diagrams) and fast P-wave azimuths from AVAZ analysis (bottom row of fast P-wave orientations). Each black dash in the bottom panels represents a fast P-wave orientation. The wellbores are denoted by red lines and the background is colored by the amount of P-wave anisotropy (Vfast-Vslow).

In summary, the AVAZ analysis suggests the NC formation in the WC field is characterized by relatively small amounts of anisotropy (at least on the scale sensitive to the AVAZ analysis) potentially controlled by EDA. However, several NW-SE elevated anisotropy trends are observed in southern and eastern portions of the WC field and frequently correlate with an unusual NW-SE fast P-wave azimuth. Interestingly, the NW-SE fast P-wave azimuth trends are unusually systematic in appearance, suggesting they may reflect an acquisition geometry artifact and not actual reservoir features. Finally, fast P-wave azimuths and anisotropic magnitudes do not correlate well with fracture orientations and density, respectively, observed in wellbore image logs, which could reflect a resolution incompatibility between the AVAZ technique and the wellbore image logs.

142

Appendix 5B: Post-Hydraulic Fracture Wellbore Image Log

A post-hydraulic fracture wellbore image log is available from a horizontal wellbore penetrating the NC formation in the WC field. The post-frac wellbore image log provides a unique opportunity to examine effects of hydraulic fracture stimulation on the rock directly surrounding the wellbore wall. The post-frac image log well deviated 36˚ from the strike of SHmax and was stimulated using an open-hole hydraulic fracturing technique in which the entire uncased 1800 meter horizontal section was pressurized at the same time.

Large aperture fractures are observed in the post-frac wellbore image log over the entire length of the horizontal section, the vast majority striking parallel to SHmax and dipping nearly vertical (Figure 5.6). Fracture spacing varies widely from less than 1 meter to more than 100 meters, but averages about 8 meters. The large aperture fractures exhibit several common characteristics in the post-frac wellbore image log: (1) Apparent fracture apertures vary from several millimeters to 5-10 cm in width, but are generally around 5 cm; (2) Fractures tend to appear in clusters of 2-6 throughout the well (Figure 5B1a); (3) Within some clusters, fractures appear to coalesce at the wellbore wall (Figure 5B1b); (4) Fractures occasionally exhibit unusual notch-like enlargements (red circles, Figure 5B1c). Since the dominant fracture trend was not observed in pre-frac wellbore image logs and reflects the expected propagation direction of Mode I hydraulic fractures in the current stress field, the large aperture fractures are interpreted as placed hydraulic fractures during stimulation. Unfortunately, a pre-frac wellbore image log from the same stimulation well is not available, which prevents comparison between preexisting and placed fracture networks. The hydraulic fracture stimulation in the post-frac well was not seismically monitored.

143

Figure 5B1: Common fracture characteristics observed in the post-stimulation image log. (a) clustered fractures; (b) fracture coalescing; (c) notch-like features.

As mentioned, the dominant fracture orientation recorded in the post-frac wellbore image log is precisely the attitude in which Mode I fractures would propagate at distances away from the wellbore wall in the current stress field. However, for horizontal wells (both perforated and open-hole) drilled at an angle to SHmax, as was the case with the post-frac image log well, this is not necessarily the orientation at which Mode I hydraulic fractures would originate on the wellbore wall (e.g. Daneshy, 1973; Yew et al., 1988; Peska and Zoback, 1995; Abass et al., 1996). In fact, hydraulic fractures typically re-orient themselves in a plane perpendicular to the least principal stress only as they propagate away from the wellbore. Interestingly, the large aperture placed fractures in the post-frac image log generally strike parallel to SHmax and dip nearly vertical (Figure 5.6). That is, fracture orientations recorded at the wellbore wall correspond to the orientation we would expect hydraulic fractures to take only at distances further away from the wellbore.

To attempt to explain this discrepancy, we use the formulations from Peska and Zoback (1995) combined with stresses from the geomechanical model in Section 5.3 to forward model the points at which tensile fractures, and hence hydraulic fractures, would initiate on the wellbore wall. The predicted hydraulic fracture initiation points in a horizontal well striking 36˚ from SHmax are shown in Figure 5B2. Comparing the predicted hydraulic fracture initiation points to the actual placed fractures observed in the wellbore image log, it appears the notch-like morphology on several large aperture fractures correlates with the predicted hydraulic fracture initiation points along the wellbore, although the correlation is not observed in all cases (two example matches

144

are shown in Figure 5B2). We suggest that hydraulic fractures initiated at the notch- like features on the wellbore wall and began to propagate into the formation as fluid and proppant were continually injected. The hydraulic fractures then linked up away from the well and began to slightly re-orient perpendicular to the least principal stress

(Shmin) while propagating away from the wellbore. Finally, the hydraulic fractures cut back through the wellbore as they grew in a plane perpendicular to Shmin, thereby creating fractures with the same attitude as the far-field hydraulic fracture on the wellbore wall. The large fracture aperture and notch-like enlargements may reflect high concentrations of near-wellbore proppant during stimulation.

Figure 5B2: (Left) Predicted initiation points of tensile fractures (hydraulic fractures) on the wellbore wall. (Right) Predicted tensile fracture initiation points overlain on actual fractures recorded in the post-frac wellbore image log.

To conclude, the post-hydraulic fracture wellbore image log contains a significant population of large aperture fractures consistent with the expected orientation of Mode I hydraulic fractures in the current stress field. We suggest hydraulic fractures initiate at the wellbore wall, re-orient in a plane perpendicular to

Shmin as they propagate away from the well (and potentially coalesce), and cut back

145

through the wellbore as fluid and proppant are continually injected, leaving large aperture, steeply-dipping fractures striking parallel to SHmax on the wellbore wall. The large aperture fractures are well distributed throughout the horizontal section, suggesting the open-hole hydraulic fracturing technique successfully stimulated much of the reservoir rock contacting the wellbore. Unfortunately, a pre-frac wellbore image log was not available from the same well, preventing insight into how and to what extent the preexisting natural fracture network is stimulated during hydraulic fracturing.

146

Chapter 6

A GEOMECHANICAL AND MICROSEISMIC STUDY OF A GAS SHALE DEVELOPMENT IN THE HORN 4 RIVER BASIN

Abstract

We carried out an integrated geomechanical and microseismic analysis of a gas shale production pad in the Horn River Basin containing 16 horizontal wells, over 270 individual hydraulic fracture stages, and more than 15,000 located microseismic events to better characterize the reservoir response to hydraulic fracture stimulations. Geomechanical constraints indicated the pad is generally characterized by a strike-slip stress regime (SHmax > SV > Shmin) with the maximum horizontal stress (SHmax) oriented

NE-SW. Measurements of Shmin, the minimum principal stress, appear to gradually increase from toe to heel in most wells. The microseismic event hypocenters occur in a region that reportedly contains a complex local fault system. Microseismic b-values

4 Portions of this chapter are published in Hurd, O., and Zoback, M.D., 2012, Stimulated Shale Volume Characterization: Multiwell Case Study from the Horn River Shale: I. Geomechanics and Microseismicity, SPE paper 159536 presented at the 2012 SPE Annual Technical Conference and Exhibition held in San Antonio, Texas, USA, 8-10 October, 2012.

147

vary from stage to stage but typically ranged between 1.0 and 2.5, indicating a relative abundance of small magnitude to large magnitude events compared to naturally- occurring earthquake populations (b ≈ 1.0). Similar observations have also been reported in other shale gas stimulation projects. We also utilized the double-difference relocation technique in an attempt to improve the location accuracy of microseismic events recorded during individual hydraulic fracture stages. The technique produces clear hypocenter trends, which, unfortunately, appear to be artifacts arising from limited monitoring array configurations and not actual tectonic structures within the reservoir. To demonstrate this, we used synthetic tests to evaluate optimal recording geometries for double-difference relocations and made recommendations for future monitoring configurations. Finally, we analyzed a population of microseismic events recorded during a single hydraulic fracture stage and found that events can be broken into small sub-groups based on high waveform similarity, which suggests repeated slip on small faults within the reservoir. We conclude that the reservoir response to hydraulic fracturing is generally characterized by shear deformation occurring on small faults reactivated during hydraulic fracturing stimulations, although there is also evidence for shear deformation occurring on larger-scale localized structures within the reservoir.

6.1 Introduction

Gas shale reservoirs are rapidly becoming an increasingly important source of hydrocarbons throughout the world. Unfortunately, all gas shale reservoirs are characterized by extremely low rock permeabilities and highly-variable lithologies, and therefore present many of the same exploitation problems and questions as tight- gas sandstone reservoirs described in Chapter 5. The advent of horizontal drilling technology, staged hydraulic fracture completions, and slickwater fracturing has made gas shale exploitation feasible and economical (King, 2010). Although the use of wireline log data, wellbore image logs, and microseismic monitoring has significantly improved reservoir characterization, fundamental questions regarding how the

148

reservoir rock is stimulated, and therefore how to optimize production, are still being addressed.

In this chapter, we analyze a variety of data from a gas shale production pad in the Horn River Basin to comprehensively characterize the effect of hydraulic fracture stimulations on Horn River gas shales. Our analysis focuses specifically on geomechanical and microseismic data, and will consist of four parts. First, using wireline, pressure test, and wellbore image log data, we develop a geomechanical model to constrain the in-situ stresses, fluid pressures, and rock properties within the reservoir. Next, we calculate Gutenberg-Richter parameters for microseismic event populations to investigate magnitude-scaling characteristics of seismic deformation during hydraulic fracturing. Then, we employ the double-difference earthquake relocation technique in an attempt to improve microseismic event hypocenter locations to identify any seismically-active reservoir structures. Finally, we perform a waveform similarity analysis on a microseismic event population to illuminate potential clusters of microseismic events and estimate inter-event distances. Our primary objective is to combine both geomechanical and microseismic analyses to better characterize the reservoir response to hydraulic fracture stimulations.

6.1.1 Geological setting and reservoir properties The Horn River Basin covers an area of approximately 3 million acres in northeastern British Columbia near the border with the Northwest Territories (Figure 6.1). The basin consists primarily of Devonian and Cretaceous clastic and carbonate sediments, and is bounded to the west by the Cretaceous Bovie fault system (MacLean and Morrow, 2004), to the south and east by Devonian carbonate platforms, and extends northward into the Northwest Territories. The basin contains several hydrocarbon-bearing formations, largely comprising clastic and carbonate sediments deposited in a foreland basin on the convergent western margin of North America (e.g. Richards, 1989). During the depositional period, local variations in depositional style and lithology were driven by small scale tectonic responses to a series of transgressive-regressive cycles originating from eustatic sea-level changes during the 149

Devonian (e.g. Johnson et al., 1985). The transgressive cycles created deep shelf environments in which the primary Devonian gas-bearing shale formations of the Horn River Basin were deposited (Allan and Creaney, 1991). The calm, anoxic shelf conditions were optimal for preserving the organic-rich, hydrocarbon-bearing shales (Morrow and Geldsetzer, 1988).

Figure 6.1: Location of Horn River Basin in northeastern British Columbia (Canada National Energy Board and British Columbia Ministry of Energy and Mines, 2011).

The Horn River Basin was first identified and exploited as a conventional natural gas source in the late 1950’s, with approximately 300 wells targeting Mississippian and Devonian carbonate gas reservoirs being drilled prior to 2002 (Adams, 2011). The Horn River Basin significantly developed as a gas-shale prospect in the early 2000’s following successful production of the Barnett Shale in the United States, and has since become one of the more prolific gas shale basins in western Canada with 21 Tcf of gas produced through 2010 and another 500 Tcf estimated GIP within the basin (with 20-30% expected recovery rates) (Canada National Energy Board and British Columbia Ministry of Energy and Mines, 2011).

150

The primary gas shale reservoirs within the Horn River Basin are the Muskwa, Otter Park, and Evie members of the Fort Simpson Formation, which largely consist of black bituminous and gray calcareous shales and limestones (Gray and Kassube, 1963). The Fort Simpson Formation dips shallowly to the south and reaches a maximum thickness of about 800 meters in the northwestern Horn River Basin and thins south and eastward down to 50 meters. Reservoir depths range between 1,800 and 3,000 meters and reservoir temperatures can be as high as 350° F (British Columbia Ministry of Energy and Mines, 2005; Reynolds and Munn, 2010). Reservoir properties are highly variable amongst (and within) the three members, and are briefly summarized in Table 6.1.

Table 6.1: Reservoir properties of Fort Simpson Formation members after McPhail et al., 2008; Ross and Bustin, 2008; British Columbia Ministry of Energy and Mines, 2005; Reynolds and Munn, 2010.

Property/Member Muskwa Otter Park Evie Porosity (%) 3.5-7.4 0.39 0.23-6.8

Permeability (µd) 0.8-9.6 0.2-4.3 0.2-4.3

TOC (%) 0.2-5 1.6-8 0.3-9.6

Silica content (%) 26-81 63-71 14-80

Clay content (%) 9-54 9-13 8-57

6.1.2 Production pad, hydraulic fracturing program, and microseismic monitoring The case study examined in this chapter involves a single production pad containing 16 horizontal wells targeting gas shales of the Fort Simpson Formation at approximately 2,700 meters depth (Figure 6.2). The northern and southern sides of the production pad will hereafter be referred to as Pad N and Pad S. Pad N reportedly contains a complex local fault system with a primary fault striking NE-SW, although specific fault locations are poorly constrained. Interestingly, Pad N wells recorded

151

noticeably lower production volumes over the first year compared to Pad S wells (Figure 6.3). This production impact, along with other production and decline analyses from the production pad, is examined in Ehlig-Economides et al. (2012).

Figure 6.2: Production pad well configuration and hydraulic fracturing sequence. Each dot represents a single stage and is color-coded in chronological order of completion.

Figure 6.3: First-year production by well.

152

Over 270 hydraulic fracture stages (approximately 17 per well) were completed over a period of approximately three to four months (Figure 6.2). The pad was stimulated using slickwater hydraulic fractures in stages spaced 80 or 110 meters apart. During the pad stimulation, stages were pressurized in as many as three wells simultaneously. Within each well, stages were completed from toe to heel, with times between consecutive stages ranging from hours to days.

Seventy-eight hydraulic fracture stages completed across twelve wells were microseismically-monitored. The microseismic data for each monitored frac stage was recorded by dual downhole geophone arrays deployed in different combinations of vertical and horizontal wells. The array positions were moved throughout the completion period to better monitor hydraulic fracture stimulations in different wells. In many cases, arrays were located significant distances from stage perforation intervals. Specifically, 80% of the monitoring arrays were located greater than 1000 meters away from the completed stage, which lead to limited detection thresholds and poorly-constrained microseismic event hypocenter locations. Despite these issues, the microseismic data originally processed by a microseismic service company yielded over 15,000 event hypocenters (See Section 6.4 for details).

6.2 Geomechanical and Microseismic Data

A wide variety of geologic, wireline, pressure, and seismic data is available from the production pad. Since this analysis will focus on the geomechanical and microseismic characterization of the gas shale reservoir, we will concentrate on the data most relevant to these objectives. Specifically, this includes the following items:

• Wireline log data • Wellbore surveys • Daily drilling reports • Formation pressure tests (DFITs) • Pressure-time data for hydraulic fracture completions • Stimulation details (injection volumes, injection rates, perforation intervals, ect.) • Microseismic data (see details below)

153

Unfortunately, there are no wellbore image logs available from the production pad, which severely limits our ability to constrain principal stress magnitudes using wellbore failure observations and to characterize the preexisting natural fracture network. In addition, no rock core data was available for laboratory testing or fracture characterization.

The microseismic data available for this analysis includes continuous velocity seismograms recorded at 2000 samples/second, a series of 1-D velocity models, a hypocenter catalog containing the origin time and location of initial microseismic event hypocenters reported by a microseismic service company, microseismic event magnitudes (MW), and monitoring geophone locations. The microseismic analyses are discussed in Section 6.4.

6.3 Geomechanical Analysis

In this section we develop a geomechanical model to constrain the in-situ stresses, fluid pressures, and rock properties within the reservoir. Specifically, the magnitudes and orientations of the three principal stresses (maximum horizontal stress,

SHmax, minimum horizontal stress, Shmin, and vertical stress, SV), the pore pressure at reservoir depths, and formation rock strength (UCS). The techniques used to constrain the geomechanical model are the same as those described in Chapter 5 (and presented in Zoback et al., 2003), and are therefore not repeated in detail.

6.3.1 Stress, pore pressure, and rock strength constraints

The magnitude of the vertical stress (SV) is calculated from wireline density logs, which yields a gradient of 26 MPa/km, corresponding to an overburden density of approximately 2,600 kg/m3. The pore pressure at reservoir depths is estimated from mud weights and equivalent circulation densities reported in drilling reports, which yields a slightly overpressured gradient of 11-13 MPa/km. This range is in good agreement with previous observations reported from this area (Apache Canada

154

communication). Compressive rock strength is ideally estimated from triaxial laboratory tests performed on rock core samples from the target formation, but no samples or test results were available for this study. Unconfined compressive rock strength (UCS) estimates from gas shale formations in the broader Horn River Basin range from 35-90 MPa (Chou et al., 2011). We adopt this range of strength values in our analysis.

The orientation of the maximum horizontal stress (SHmax) is typically inferred from wellbore breakouts or drilling-induced tensile fractures observed in wellbore image logs. Unfortunately, no wellbore image logs were available from the production pad, and the nearest available image log, located 11 km from the pad, did not contain any breakouts or drilling-induced tensile fractures. Utilizing the 10 nearest ‘A’ and ‘B’ quality stress measurements from the World Stress Map database (Heidbach et al.,

2008), the orientation of SHmax is estimated as N62°E ± 11°.

The magnitude of the minimum horizontal stress (Shmin) is estimated from bottomhole instantaneous shut-in pressures (ISIPs) recorded during DFIT tests conducted in four wells prior to hydraulic fracture stimulation. The ISIP is used to estimate Shmin since hydraulic fracture stimulations were completed using low- viscosity fracturing fluids with low proppant concentrations. The fracture closure pressure (FCP), which is also sometimes used to estimate Shmin, is a better estimate of

Shmin when high-viscosity hydraulic fracturing fluids (e.g. gel fracs) with significant proppant concentrations are used for stimulation (Zoback, 2007). ISIPs measured from

DFIT tests indicate a Shmin gradient of 22.6 ± 1.1 MPa/km. In addition, ISIPs recorded from over 270 individual hydraulic fracture stages indicate a Shmin gradient of 20.7 ±

1.8 MPa/km. These estimated Shmin magnitudes are slightly below the overburden stress (SV) and are within the range of Shmin magnitudes observed across the Western Canada Sedimentary Basin (Bell and Grasby, 2011). The large number of ISIP measurements provides a rare opportunity to examine both spatial and temporal Shmin variations in detail, which will be addressed in the following section.

155

In the absence of wellbore breakouts or drilling-induced tensile fractures, the frictional strength of pre-existing, optimally-oriented faults in the crust can be used to place an upper limit on the differential principal stress magnitudes (Jaeger and Cook, 1979; Zoback, 2007). Reports of minor fault activation during hydraulic fracture stimulations in nearby production pads suggest that this is a reasonable approach to placing an upper limit on the largest principal stress magnitude in this region. The limiting ratio of the maximum to minimum principal stresses constrained by an optimally-oriented fault is given by (Jaeger and Cook, 1979)

 S  P 1  1 P  (  2  1  )2  S  P 3 3 P (6.1)

where PP is the pore pressure, μ is the coefficient of friction, and S1 and S3 are the maximum and minimum principal stresses, respectively. Using the pore pressure and

Shmin measurements described above, a coefficient of friction of 0.6, and assuming a strike-slip stress regime (SHmax > SV > Shmin) yields an upper bound on SHmax of approximately 38 MPa/km. Note that for a constant pore pressure and coefficient of friction in a strike-slip stress regime, reducing the value of S3 (Shmin) would lead to a lower S1 (SHmax) estimate.

6.3.2 Shmin variations The large number of ISIP measurements available from the production pad provides the opportunity to examine potential spatial and temporal variations of the least principal stress. It is important to note that all stages were completed using similar volumes of fluid and proppant, suggesting that any potential spatial Shmin variations could reflect differences in rock and/or reservoir properties. To examine potential spatial variations, we plot Shmin gradients inferred from ISIPs for each stage

(Figure 6.4). The Shmin/SV ratio for each stage is also plotted in Figure 6.4. The vast majority of Shmin gradients fall between 18 and 22 MPa/km while the Shmin/SV ratios generally fall between 0.75 and 0.85. The highest Shmin gradients are observed in stages near the center of Pad N. Elevated Shmin gradients also appear to delineate a N-

156

S trend in Pad N. These elevated Shmin gradients do not appear to correlate with the reported location of potential faults in Pad N.

Figure 6.4: Shmin gradient and Shmin/SV ratio for all stages.

No substantially-elevated Shmin gradients are observed in wells in Pad S, although gradients tend to gradually increase in magnitude from toe to heel in each well. This is also apparent for the wells in Pad N, although more difficult to observe with the locally-elevated Shmin gradients in the center of Pad N. The gradual toe to heel

Shmin increase has been previously observed in sequentially-staged gas shale stimulations, and has been interpreted as a “stress shadow” effect due to the close proximity of consecutively propped hydraulic fractures (East et al., 2004; Fisher et al., 2004; Soliman et al., 2008; Wu et al., 2012), although Vermylen and Zoback (2011) argue that poroelastic effects contribute to this increase significantly in the Barnett Shale wells they studied.

To visually examine how the distance and time between consecutively- stimulated stages across the entire production pad affect ISIP variations, we plot the ISIP difference between consecutively-completed stages versus the distance between consecutively-completed stages (measured as the distance between the perforated intervals for each stage) (Figure 6.5). Stages with anomalously high ISIPs (> 40MPa)

157

were excluded. If a strong “stress shadow” effect is present, we might expect stages stimulated close to one another in time and space to exhibit larger ISIP differences. In other words, the shorter the time and distance between consecutive stages, the less time elevated reservoir pressures from the first stage have to return to unperturbed levels and the more likely they are to affect pressures recorded in the following stage. Such a relationship is not observed in Figure 6.5. That is, graphically, blue dots located less than 1000 meters on the x-axis are widely distributed along the y-axis. It is also important to note that many ISIP differences between consecutive stages are on the same scale as uncertainties in the individual ISIP measurements (± 2-3 MPa), which could mask any clear relationships. A more thorough analysis, which is beyond the scope of this study, is therefore necessary to precisely evaluate the origin of the gradual toe to heel Shmin gradient increase observed in most wells.

Figure 6.5: ISIP difference versus time and spatial difference between consecutively-stimulated stages across the entire production pad. ISIPs were converted to bottomhole pressures normalized to 2700 meters.

158

Shmin/SV ratios across the production pad vary between 0.7 and 1.05, although the vast majority fall between 0.75 and 0.85 (Figure 6.4). These anomalously low and high Shmin/SV ratios could indicate locally transitional stress regimes from strike-slip to normal (SHmax ≈ SV) or strike-slip to reverse (Shmin ≈ SV). A transitional stress regime is important to identify from an exploitation perspective as it could indicate faults of multiple orientations may be well-oriented for shear failure in the current stress field. The likelihood of a transitional stress regime constrained by the frictional strength of well-oriented faults in the crust can be evaluated using equation (6.1). For example, for normal and strike-slip faulting regimes, equation (6.1) becomes

 S  P 1  V P  (  2 1  )2  S  P Normal faulting: 3 h min P (6.2)

 S  P 1  H max P  (  2 1  )2  S  P Strike-slip faulting: 3 h min P (6.3)

Assuming µ = 0.6 (Byerlee, 1978), SV = 26 MPa/km, and PP = 13 MPa/km, the lower limit of Shmin in a normal faulting regime is ~0.7SV. If Shmin takes this lower limit and SHmax ≈ SV, a transitional strike-slip/normal faulting regime could exist. In other words, both equations (6.2) and (6.3) would be satisfied. Similarly, if Shmin ≈ SV and SHmax is ~2.1SV, a transitional strike-slip/reverse stress regime may exist.

Unfortunately, although there is evidence that Shmin/SV may extend as low as 0.7 and as high as 1.05, without better constraints on SHmax, it is difficult to evaluate the likelihood of transitional stress regimes. Given the lack of constraints on SHmax, and more importantly, that the vast majority of Shmin/SV ratios fall between 0.75 and 0.85, we conclude that the production pad is characterized by a strike-slip stress regime.

6.4 Microseismic Analyses

The unusual quantity of microseismic data available from the production pad provides a unique opportunity to evaluate the reservoir response to hydraulic fracturing using multiple seismic analysis techniques. The microseismic analysis

159

consists of three parts. First, we utilize reported microseismic event magnitudes to examine magnitude-frequency relationships across the production pad to infer the nature of hydraulic fracture deformation and examine evidence for slip on larger-scale faults and structures within the reservoir. Second, using the double-difference technique described in Chapter 5, we attempt to obtain precise hypocenter relocations to better assess the potential for seismic deformation on reservoir structures. Finally, we perform a waveform similarity analysis to constrain inter-event distances between microseismic events having extremely similar waveforms, which will examine the potential for repeated slip on reservoir structures. Combined, these analyses will provide insight into the nature of seismic deformation associated with hydraulic fracture stimulations within the production pad. This section begins with a brief presentation of the reported initial event hypocenter locations.

6.4.1 Initial microseismic event hypocenter locations The microseismic data was originally processed by a microseismic service company which located over 15,000 event hypocenters (Figure 6.6). Each of the 78 monitored stages contained between 15 and 839 induced microseismic events, with an average of 190 per stage. Microseismic event hypocenters were initially located in real-time to monitor potential hydraulic fracture growth out of the target reservoir. Once all stimulations were complete, approximately 70% of the microseismic events were then relocated on a stage-by-stage basis using a series of updated velocity models constrained by perforation shots. A total of 28 different velocity models were used to locate events induced during the 78 monitored stages. The average P- and S-wave velocities of the 28 velocity models over reservoir depths (2,650-2800 meters) were 4,100 ± 160 m/s and 2,630 ± 120 m/s, respectively. The hypocenters presented in Figure 6.6 are the relocated hypocenters reported by the microseismic service company.

160

Figure 6.6: Initial microseismic event hypocenter locations for 15,000 events reported by the microseismic service company. Events are colored by moment magnitude. Low amplitude events (< MW = -2) are largely obscured due to layering during plotting.

161

The microseismic event hypocenter distributions varied from stage to stage, but were generally distributed in NE-SW trends (Figure 6.7), which is the expected propagation direction of Mode I hydraulic fractures in the current stress field. Note that the NE-SW trend is consistent with the SHmax orientation constrained in the geomechanical model and corroborates a strike-slip stress regime (S3 < SV). Additionally, event hypocenters sometimes extended up to a kilometer away from the fracturing well, and event hypocenters from consecutively-fractured stages often showed considerable overlap. The majority of event hypocenters were constrained within the target reservoir (Figure 6.6). Since most monitoring geophones were a significant distance from injection points and horizontal monitoring arrays can not accurately resolve hypocenter depths, much of the growth outside of the target reservoir is likely a result of location uncertainty. Hypocenters from several stages, however, showed unusually pronounced height growth into underlying or overlying formations (Figure 6.11 and corresponding discussion). It is not clear whether the pronounced height growth reflects fluid transport (and seismic deformation) along a fault or poorly-constrained hypocenter locations. Consequently, evaluating potential fault deformation with microseismic event magnitude-scaling relationships (Section 6.4.2) and improving event hypocenter location accuracy with advanced relocation techniques (Section 6.4.3) are primary objectives of the seismic analysis.

Figure 6.7: Rose diagrams illustrating SHmax orientations inferred from microseismic event hypocenter trends. SHmax orientations were estimated by fitting a line to event hypocenter locations in map view. (a) SHmax estimated from all 78 monitored stages. (b) SHmax estimated from stages having an R2 line fit > 0.5 (n = 22).

162

6.4.2 Microseismic magnitude distributions A consistent, log-linear scaling relationship describing naturally-occurring earthquake magnitude distributions was first observed by Gutenberg and Richter (1944) from earthquake catalogs throughout the world spanning a wide range of earthquake magnitudes and time intervals. The relationship states that for every unit increase in magnitude, the number of earthquakes occurring above that magnitude decreases logarithmically according to the relationship log(N > M) = a – bM, where N is the number of earthquakes greater than a given magnitude M. The constant b, known as the b-value, quantifies the number of larger relative to smaller magnitude events in a given earthquake population, and is typically near 1.0 in most naturally- occurring earthquake populations. In naturally-occurring earthquake studies, b-values are commonly used to better understand earthquake source physics within a region and to estimate earthquake recurrence intervals. The parameter a, known as the activity level or a-value, represents the total number of events with MW > 0. The a-value can quantify relative microseismic activity levels between two regions with a similar b- value.

While the applicability of the log-linear magnitude scaling relationship for small (< MW = 0) earthquakes has been a topic of debate (Aki, 1987; Richardson and Jordan, 2002; Boettcher et al., 2009; Kwiatek et al., 2010), numerous studies have applied magnitude-frequency analyses to microseismic data sets recorded during fluid injection stimulations (e.g. Urbancic et al., 1999). In these cases, the application is largely motivated by the desire to quantitatively evaluate the success and effectiveness of hydraulic fracture stimulations. For example, a-values have been used to evaluate and compare the effectiveness of multiple hydraulic fracture stages (Urbancic et al., 1999) while b-values have been used to infer fracture growth and deformation mechanisms (Urbancic et al., 1999) as well as identify potential faults activated during stimulations (Maxwell et al., 2009, 2011; Downie et al., 2010).

We calculate b-values for microseismic event populations recorded during individual hydraulic fracture stages to examine seismic deformation mechanisms

163

across the production pad. All events were assigned to stages based on temporal correlation with the hydraulic fracture pumping schedule. The magnitudes used in this study are moment magnitudes reported by the microseismic service company and range between MW = -4 and 1.5. To calculate b-values for each stage, we use the maximum likelihood estimate b = 1/2.3*(Mavg – Mmin) where Mavg is the average magnitude within the event population and Mmin is the minimum magnitude of the event population (Aki, 1965; Utsu, 1965). The standard deviation of a b-value can be approximated by b/√N where N is the number of events greater than Mmin (Aki, 1965). Since hydraulic fracture stages were completed at varying distances from the monitoring arrays, we establish Mmin = -2.0 as the minimum magnitude of catalog completeness for all stages and remove all events with magnitudes below this threshold. Stages containing 50 or fewer events above the Mmin cutoff are not considered since b-value estimates become increasingly poorly constrained as the cumulative number of events falls below this number (Shi and Bolt, 1982; Bender, 1983). A histogram of b-values from 34 stages meeting the above criteria is shown in

Figure 6.8a. The b-value estimated using all events MW ≥ -2.0 is shown in Figure 6.8b.

Figure 6.8: (a) Distribution of b-values calculated from events recorded in individual hydraulic fracture stages. Stages with fewer than 50 MW ≥ -2.0 events were not included in the histogram. (b) The b-value calculated using all MW ≥ -2.0 events in the production pad (4618 events).

The b-values across all hydraulic fracture stages vary between 0.7 and 3.5 with an average of 2.1. Although not shown, the b-value histograms look similar for both

164

Pad N and Pad S individually. Unfortunately, the broad range of b-values prevents direct comparison of activity levels (a-values) between stages. The subset of events ≥

MW = -2.0 for the entire pad yields a high b-value of 1.76 (Figure 6.8b), which suggests that microseismic activity is generally characterized by a relative abundance of small to large magnitude events compared to naturally-occurring magnitude- frequency distributions (b ≈ 1). This b-value is consistent with b-values observed in similar hydraulic fracturing environments (Urbancic et al., 1999; Downie et al., 2010; Maxwell et al., 2011).

In addition to calculating b-values for microseismic event populations from individual stages, we also divide the production pad into a series of grid blocks and calculate b-values for event population within each grid block. This gridding emphasizes the spatial distribution of event hypocenters. Each grid block is 250 x 250 meters and extends infinitely in the vertical direction. Again, only events ≥ MW = -2.0 are considered, and b-values are not calculated for grid blocks containing fewer than 50 events. The b-value distribution and corresponding standard deviations are shown in Figure 6.9. As with the b-values calculated for each stage, there is a wide distribution of b-values across all grid blocks, but most range between 1.0 and 2.5. The elevated b-values observed in both Figures 6.8 and 6.9 suggest that constant magnitude-frequency scaling relationship observed in naturally-occurring earthquake populations do not apply to this study despite employing a common Mmin to ensure catalog completeness across all stages.

165

Figure 6.9: b-values and corresponding standard deviations calculated from events falling within 250 x 250 meter grid blocks. White blocks represent blocks containing fewer than 50 MW ≥ -2.0 events.

If seismic deformation is potentially occurring on larger-scale structures, as suggested by certain microseismic event populations exhibiting b-values near 1.0, it is possible that larger-magnitude event hypocenters could delineate structural features within the reservoir. To examine this possibility, we plot all events with MW > -1.5 recorded across the production pad and compare them with Shmin gradient and b-value distributions (Figure 6.10). The clustering nature of MW > -1.5 events might suggest microseismic deformation preferentially occurring on localized structures within the reservoir. In addition, some larger-magnitude event hypocenter distributions do appear to correlate with elevated Shmin gradients or lower b-values, but not consistently throughout the pad.

166

Figure 6.10: Hypocenter locations of all events with MW > -1.5 compared with Shmin gradient and b- value distributions.

6.4.3 Precise microseismic event relocations As described in Chapter 5, the double-difference relocation technique can be used to improve relative earthquake hypocenter locations for a population of clustered earthquakes. The event hypocenter locations in this study present a unique and enticing opportunity to apply the double-difference technique given the extraordinary breadth of the microseismic monitoring program, the pronounced microseismic event hypocenter height growth observed in several stages, and the potential for fault-related seismic deformation. The double-difference technique provides the possibility of improving microseismic event hypocenter locations within the reservoir, which could lead to better resolution of potential seismogenic structures stimulated during hydraulic fracturing. Our application will also provide an additional opportunity to test the applicability of the double-difference event relocation technique in limited monitoring array configurations.

6.4.3.1 Stage selection, monitoring configuration, and seismic data processing Of the 15,000+ microseismic events recorded during 78 injection stages, we select the 212 events recorded during the second stage in Well 1 on which to apply the double-difference relocation technique for three primary reasons. First, the stage was monitored by both a horizontal and vertical recording array, which provided improved azimuthal coverage and depth resolution compared to monitoring array configurations deployed during stimulation of other stages. The horizontal array (Array 1) contained seven three-component geophones and was located in an adjacent well at reservoir

167

depth, roughly 1000 meters from the Well 1 Stage 2 injection point (Figure 6.11). The vertical array (Array 2) also contained seven three-component geophones and was located several hundred meters above the reservoir at a distance of roughly 2100 meters from the injection point. We note that while Array 2 is poorly located to resolve hypocenters in terms of both distance from the fluid injection point and azimuthal position compared to Array 1, it nonetheless provides a second set of travel time constraints and is an improvement over single-array deployments typical of most microseismic monitoring configurations. Second, the initial event hypocenter locations indicated pronounced height growth on the order of 100s of meters, which was not observed during hydraulic fracturing stimulations in nearby stages. Finally, the hypocenters occur in a region reportedly containing a complex fault system, suggesting the pronounced height growth could reflect fault deformation during hydraulic fracture stimulation.

Figure 6.11: Initial microseismic event hypocenter locations from the Well 1 Stage 2 stimulation (212 events). The red well indicates Well 1 and the green dot represents the Stage 2 injection point.

The seismic data processing follows the general workflow described in Chapter 5 (Section 5.5.3), and is therefore only briefly summarized here. For the pre- processing, all seismograms were bandpass filtered between 5 and 200 Hz. First, we examine seismograms from all 212 events from Well 1 Stage 2 to manually pick P- wave arrival times on as many sensors in Array A and Array B as possible. Events not having clear, unambiguous P-wave arrivals on at least one sensor in both recording

168

arrays were discarded, leaving 116 events. These remaining events were then broken into groups using a time-domain waveform similarity analysis. If two full event seismograms containing both the P- and S-wave arrival had a cross-correlation coefficient greater than 0.6, they were assigned to the same group. This led to 78 events being assigned to six groups each containing at least four events.

Next, within each group, all seismograms were stacked to obtain a master trace for each sensor component (EW, NS, and Z). This step was performed separately for each monitoring array. P- and S-wave arrivals were picked on the master traces, which were then cross-correlated with each individual event seismogram in the group to pick absolute P- and S-wave arrival times for each event to sub-sample precision (e.g. Schaff et al., 2004). Finally, we cross-correlate all event seismograms within each group with one another to calculate precise relative differential P- and S-wave travel times between event pairs. Analyses with maximum cross-correlation coefficients below 0.7 were discarded. Both absolute P- and S-wave arrivals and relative differential P- and S-wave travel times were used as input for the double- difference relocation.

6.4.3.2 Double-difference relocations The Well 1 Stage 2 event hypocenters are relocated using the HYPODD 2.1 software program (Felix Waldhauser, personal comm.). The event hypocenters are relocated with HYPODD instead of TOMODD (as in Chapter 5) since HYPODD can solve the double-difference equations with the SVD technique, which provides better error estimation than the LSQR technique and is better suited for small event populations (Waldhauser and Ellsworth, 2000). Only the 78 events falling within the six previously identified groups were located since P- and S-wave phase picks from the 58 unassigned events were not constrained by waveform cross-correlation and were therefore less consistent and accurate. Although initial event locations were available, all event hypocenters were given the same starting location at the Well 1 Stage 2 perforation interval because of the surprisingly large uncertainty in the

169

locations reported by the microseismic service company, and were relocated using the SVD technique.

Events were relocated with a single-layer velocity model with Vp = 3,900 m/s and Vs = 2,300 m/s. This simplified velocity model was chosen since the specific HYPODD version used in our analysis only allows for single-layer velocity models in downhole monitoring configurations. Nonetheless, the single-layer velocity model is a reasonable assumption given the fairly constant P-wave velocities over most hypocenter location depths, although raypaths recorded by Array 1 may pass through higher-velocity reservoir rocks (Figure 6.12).

Figure 6.12: P-wave velocity log over microseismic hypocenter depths. The red line corresponds to VP = 3,900 m/s.

As illustrated in Chapter 5, proper data weighting, event pairing, and iterating is crucial when relocating event hypocenters with the double-difference technique, and we experiment with a wide variety of weighting and pairing schemes during relocation. All events are allowed to pair with each other event in the population, creating catalog differential travel times between all event pairs amongst the 78 events. This is simply to ensure that catalog differential travel times for all potential event pairs are available, and does not imply that all events will be paired throughout all the iterations. We find the optimal double-difference solution is obtained using only 170

catalog differential travel times. Including cross-correlation differential travel times did not further improve the solution.

Two iterations are required to obtain the optimal double-difference solution. Both iterations weighted the absolute P- and S-wave travel time data equally. During the first iteration, inter-event distances are constrained to less than 200 meters. That is, if an event pair offset exceeds 200 meters during the relocation, the events are no longer paired for the next iteration. The first iteration is intended to allow the hypocenters to spread out in space. During the second iteration, inter-event distances are constrained to 100 meters, which allows for finer-scale adjustments to hypocenter relocations. The second iteration substantially reduces the number of event pairs within the group, and even removes 14 events (events no longer paired with any other events in the population). Although it resulted in events being excluded, we find the 100 meter constraint in the second iteration necessary to ensure a quality solution. The data rejection could reflect, for example, small P- and S-wave phase pick errors in specific events that would unnecessarily perturb other event relocations through pairing. The optimal relocations are shown in Figure 6.13 color-coded by the average error in EW (x), NS (y) and Z.

There are several notable observations from the 64 relocated event hypocenters in Figure 6.13. First, event hypocenters tend to align in a NE-SW trend, which is roughly parallel to the regional SHmax orientation and consistent with the expected propagation direction of Mode I hydraulic fractures in the current strike-slip stress field. This trend is also consistent with the original hypocenter locations (Figure 6.11). Second, most hypocenters are located vertically within ± 200 meters of the injection well, which is in contrast to the significant height grown indicated by the initial locations. In fact, most hypocenters fall below the injection well. Third, most hypocenter location errors reported by the HYPODD program are less than 50 meters. Lastly, the majority of hypocenters delineate a sub-cluster of events striking ~EW and plunging roughly 45° to the east (see magnified plots in bottom row of Figure 6.13). Similar tightly clustered hypocenter trends in other microseismic monitoring studies

171

have been interpreted as induced slip on preexisting faults reactivated during hydraulic fracture stimulation. While this is a possibility, a fault of this orientation would be rather poorly oriented for slip in the current strike-slip stress regime. Although this unusual trend could be sensitive to the velocity structure or event starting locations, given the significant impact the dual-monitoring array configuration in Chapter 5 appeared to have on hypocenter relocations, we first perform a synthetic test to gauge resolution capabilities of the Well 1 Stage 2 monitoring configuration prior to making any final interpretations.

Figure 6.13: Double-difference microseismic event hypocenter relocations of events recorded during the Well 1 Stage 2 stimulation. Dashed boxes in the top row indicate the magnified section in the bottom row. Only injection and monitoring wells are shown for clarity. The Well 1 Stage 2 fluid injection point is indicated by an orange dot in the bottom rows.

6.4.3.3 Synthetic location tests: 2 arrays Although the Well 1 Stage 2 relocated event hypocenters delineate a tight, NE- SW-striking trend, the double-difference hypocenter relocations in Chapter 5 emphasized the importance of evaluating potential hypocenter location artifacts due to limited monitoring array configurations. To examine the reliability of the Well 1 Stage 2 double-difference hypocenter relocations, we perform a simple synthetic test using the HYPODD software program. First, we set up a monitoring array

172

configuration identical to that used during the Well 1 Stage 2 stimulation and arrange 37 synthetic hypocenters in N-S, NE-SW, and NW-SE trends centered about the Stage

2 injection point (Figure 6.14). Then, assuming a single-layer velocity model with VP

= 3,900 m/s and VS = 2,300 m/s, we calculate the theoretical P- and S-wave travel times from each synthetic hypocenter location to all recording sensors in both monitoring arrays. No noise is added to the travel times. Finally, using the same starting location for all 37 events (the Well 1 Stage 2 perforation interval), we relocate the events using the SVD technique with equal weights applied to P-and S-wave picks. Note that only catalog differential travel times are used in the relocation. The events were located with a single iteration with no constraint on distances between event pairs.

The synthetic event hypocenter relocations using the Well 1 Stage 2 monitoring configuration are shown in Figure 6.14. The relocated hypocenters fail to restore the actual synthetic hypocenter locations, and instead comprise a NE-SW trend with individual hypocenters sometimes extending hundreds of meters away from their actual locations. The NE-SW hypocenter trend likely reflects the lack of travel time constraints in that direction. Consequently, although the hypocenter location errors reported by the HYPODD program were quite small in the original double-difference locations (Figure 6.13), the actual hypocenter uncertainty could potentially be much larger, and it may also be only by coincidence that the actual relocated hypocenters in

Figure 6.13 generally align with the NE-SW SHmax direction. A nearly identical array configuration-induced artifact was observed by Kumano et al. (2006) when attempting to apply the double-difference relocation technique in a similar limited monitoring array configuration.

Additionally, relocated hypocenter depths are also poorly constrained in the dual-array synthetic test, sometimes displacing up to hundreds of meters from their actual locations. The poor depth resolution is likely due to a combination of a horizontal monitoring array, which provides very little depth resolution, and a vertical recording array located significantly above the reservoir. It is also worth noting that,

173

although two monitoring arrays were available, they provided very little azimuthal coverage.

The synthetic tests indicate that the limited monitoring configuration available for the Well 1 Stage 2 stimulation can yield substantial hypocenter relocation artifacts with the double-difference technique. Given the apparently substantial influence of the monitoring array configuration on hypocenter relocations, we do not examine the sensitivity of event hypocenter relocations to other factors (variable starting locations or velocity model perturbation) as they are likely secondary in influence to the monitoring array configuration. Additional synthetic tests of the double-difference technique in more ideal array configurations are presented in Appendix 6A.

Figure 6.14: Synthetic test of the double-difference technique in the Well 1 Stage 2 monitoring configuration. Green circles represent synthetic hypocenter locations, red dots represent double- difference relocations. Dashed boxes represent the magnified sections on the below plots.

The synthetic location test results raises concerns regarding the Well 1 Stage 2 event hypocenters locations in Figure 6.13. First, since event hypocenters appear to be irresolvable in the NE-SW direction in the available monitoring array configuration, we cannot definitely conclude that the NE-SW hypocenter trend in Figure 6.13 reflects the actual hypocenter locations. Second, the synthetic location test yields unique

174

hypocenter trends in depth (Figure 6.14) similar to those observed in the actual hypocenter trends. In particular, both the actual locations and synthetic test yield a hypocenter trend that plunges approximately 45° toward the east, which again precludes any interpretation of the trend as a potential reservoir structure.

To summarize, double-difference hypocenter relocations from Well 1 Stage 2 exhibit many characteristics of hypocenter location artifacts from a synthetic location test. Consequently, as enticing as the Well 1 Stage 2 double-difference hypocenter relocations appear, they may reflect artifacts from the limited monitoring array configuration and not actual reservoir structures stimulated during hydraulic fracturing. While many other stages were microseismically-monitored across the production pad, some with significantly smaller hypocenter-sensor offsets than Well 1 Stage 2, all stages were recorded with only two monitoring arrays, suggesting that array configuration-induced hypocenter location artifacts may be a significant concern regardless of the chosen stage.

6.4.3.4 Synthetic location tests: 3 arrays and recommendations The synthetic tests described in Section 6.4.3.3 clearly illustrate the inability to resolve accurate hypocenter relocations using the double-difference technique in a limited, dual monitoring array configuration. However, what if a third monitoring array were available? Could the double-difference technique resolve accurate hypocenters? To examine this possibility, we revisit the synthetic test described in the previous section with the addition of a third, hypothetical monitoring array. The third array consists of seven sensors located in a horizontal well parallel to Well 1 (Array 3, Figure 6.15). The third array is strategically placed to provide resolution in the NE- SW direction, although it is very important to note that this array position was attainable with the available well distribution during this hydraulic fracturing stimulation stage.

The synthetic test with three arrays is performed using the same velocity model, number of synthetic hypocenters, and relocation procedure as described in the

175

previous section. The double-difference relocations are shown in Figure 6.15. The double-difference technique accurately restores the synthetic events to their actual locations, particularly in map view. It is worth reiterating that only catalog differential travel times are used to relocate the events, and that well-constrained cross-correlation differential travel times may improve event hypocenter relocations even further. Hypocenter depths are slightly less well-restored, but still provide a much-improved match over the dual-array synthetic location test. Moreover, the hypocenter relocations confirm the unique ability of the double-difference technique to precisely resolve event hypocenters along multiple trends within the same event population. Although not shown, equally accurate hypocenter relocations are obtained if the third monitoring array is positioned at other locations at reservoir depths about the synthetic event hypocenters. We conclude that, with a third monitoring array, the double- difference technique could have accurately resolved event hypocenter locations from the Well 1 Stage 2 stimulation.

Figure 6.15: Synthetic test of the double-difference technique with a hypothetical third array (Array 3) in the Well 1 Stage 2 monitoring configuration. Green circles represent synthetic hypocenter locations, red dots represent double-difference relocations. Dashed boxes represent the magnified sections on the below plots.

176

Previous studies have also evaluated how variable numbers of monitoring sensors influence microseismic event hypocenter locations obtained with other earthquake location techniques in fluid injection environments (e.g., Warpinski et al. 2005; Seibel et al. 2011). These studies also generally concluded that event hypocenter locations are better-constrained with a larger number of monitoring stations. The results presented in Seibel et al. (2011) in particular also illustrate how double- difference relocations are improved with additional monitoring arrays.

Based on the double-difference relocations and synthetic resolution tests in this study, we make the following recommendations for utilizing the double-difference technique to obtain precise hypocenter relocations in limited monitoring configurations.

(1) At least three monitoring points are needed to obtain highly-accurate event hypocenter relocations. Ideally, monitoring points should be distributed to cover the greatest azimuthal range possible. The number and distribution of monitoring points influences the hypocenter relocation accuracy more than any other factor, and maximizing the number of well-distributed recording sensors should therefore be the primary objective when planning a microseismic monitoring survey.

(2) Monitoring arrays should avoid being placed at extreme distances from the stimulated interval. The extreme hypocenter-array offsets in this analysis led to exclusion of more than 60% of the microseismic event data.

(3) Vertical recording arrays generally provide better event hypocenter depth resolution and are therefore recommended, although our synthetic tests demonstrate that combinations of vertical and horizontal arrays positioned at different depths can resolve accurate relocations.

(4) Event hypocenter location trends constrained by two or fewer monitoring arrays should be interpreted with caution, even if the trends appear geologically-reasonable.

177

(5) Quality event hypocenter locations are attainable even if only absolute P- and S- wave travel times are available.

(6) If artifacts in hypocenter locations are suspected, the hypocenter location quality could be quickly assessed by examining the moveout of clearly identifiable phases in seismograms across the monitoring arrays.

6.4.4 Waveform similarity analysis Seismic events having highly similar waveforms have frequently been observed in both natural tectonic and fluid injection environments (e.g. Geller and Mueller, 1980; Rutledge and Phillips, 2003). The high waveform similarity between pairs of events, known as doublets, and groups of three or more events, known as multiplets, has been interpreted as repeated rupture of a specific fault or fracture, and can be used to place constraints on the relative distance between events (e.g. Geller and Mueller, 1980; Nadeau et al., 1995; Arrowsmith and Eisner, 2006; Baisch et al., 2008). Following the technique developed by Arrowsmith and Eisner (2006), which utilizes waveform cross-correlation to identify events with similar waveforms, we examine waveform similarity between events recorded during the Well 1 Stage 2 hydraulic fracture stimulation. The objective is to identify the prevalence of waveform similarity between induced events and constrain inter-event distances between events with similar waveforms, which could provide insight into seismic deformation mechanisms within the reservoir.

The waveform cross-correlation is performed between 0.25-second seismogram windows bandpass filtered between 5 and 100 Hz comprising both the P- and S-wave arrivals of each event. The cross-correlation coefficient between event pairs is calculated by averaging correlation coefficients from all three sensor components weighted by seismogram signal-to-noise ratios (Arrowsmith and Eisner, 2006):

178

(SNR  SNR ) * CCC  (SNR  SNR ) * CCC  (SNR  SNR ) * CCC CCC  EW 1 EW 2 EW NS1 NS2 NS Z1 Z 2 Z total (SNREW 1  SNREW 2 )  (SNRNS1  SNRNS2 )  (SNRZ1  SNRZ 2 ) (6.4) where SNR is the signal-to-noise ratio of a 0.05-second time window prior to and after the P-wave arrival and CCC is the maximum cross-correlation coefficient between seismograms for each respective component. Using a cross-correlation coefficient

(CCCtotal) threshold greater than 0.9 to denote similar events, the 116 events distribute into two multiplets containing 10 and 12 events and nine doublets. An example doublet is shown in Figure 6.16. The distance between event pairs within the doublets and multiplets can be estimated by (Arrowsmith and Eisner, 2006)

(t 2  t 2 )  (t1  t1 ) dx  S P S P 1 1    (6.5)

where tP and tS are the P- and S-wave arrival times for events 1 and 2 and α and β are the P- and S-wave velocities, respectively. Assuming α = 3,900 m/s and β = 2,300 m/s, inter-event distances between events within the two primary multiplets are up to 10 meters.

As the waveform similarity analysis is sensitive to seismogram frequency content as well as the cross-correlation coefficient threshold, we also examine the effect of different filters and correlation thresholds on the similarity analysis. If a cross-correlation threshold of 0.8 and a bandpass filter between 5 and 100 Hz is applied, the 116 events distribute into three multiplets (containing 43, 18, and 4 events) and four doublets. Using a cross-correlation threshold of 0.8 and bandpass filter between 5 and 200 Hz, the similarity analysis yields two multiplets (with 26 and 3 events) and seven doublets.

179

Figure 6.16: Example event doublet recorded during the Well 1 Stage 2 stimulation. EW, NS, and vertical (Z) component waveforms are shown from left to right.

Initial hypocenter locations (reported by the microseismic service company) of events comprising the two multiplets are shown in Figure 6.17. Interestingly, the distances between events within each multiplet vary by up to several hundred meters, a distance obviously much greater than the < 10 meters estimated from the waveform similarity analysis. Similarly, the median separation of the nine doublets is almost 500 meters. This discrepancy appears to corroborate the uncertainty in initial hypocenter locations from the microseismic service company, particularly in depth, and suggests that the notable height growth should be interpreted with caution.

Figure 6.17: Hypocenter locations of events within the two multiplets identified from the Well 1 Stage 2 waveform similarity analysis. The orange dot represents the fluid injection point for the Well 1 Stage 2 stimulation.

The distribution of microseismic events into multiplets suggests repeated seismic deformation on specific fractures or faults during the Well 1 Stage 2 hydraulic

180

fracture stimulation. This repeated deformation appears localized, however, as the majority of events recorded during the stimulation do not comprise doublets or multiplets. Specifically, the percentage of events constituting doublets or multiplets is << 50% relative to the total population of events for the various filter cutoffs and cross-correlation coefficient thresholds examined in this analysis. This small percentage of doublets and multiplets is consistent with the higher b-value of 1.78 calculated from the Well 1 Stage 2 events (not shown), which implies that seismic deformation is largely regulated to isolated slip on small fractures rather than (repeated) slip on segments of larger structures. We also note that the average magnitude of events within each multiplet (-1.67 ± 0.3 and -1.68 ± 0.2) is similar to the average magnitude of all events from the Well 1 Stage 2 stimulation (-1.79 ± 0.3).

6.5 Conclusions

We integrate geomechanical and microseismic analyses at a gas shale production pad in the Horn River Basin to examine reservoir response to hydraulic fracturing stimulations. The production pad consisted of 16 horizontal wells with over 270 individual hydraulic fracture stages, 78 of which were microseismically monitored. First, using wireline and pressure test data, we found the reservoir is generally characterized by a strike-slip stress regime (SHmax > SV > Shmin) with SHmax oriented NE-SW. ISIPs recorded from DFIT tests performed prior to hydraulic fracturing stimulation and from over 270 individual hydraulic fracture stages indicate a largely consistent Shmin gradient over the pad, although there are several stages exhibiting locally-elevated gradients. In addition, a slight Shmin gradient increase from toe to heel is observed in many wells, possibly reflecting a “stress shadow” or poroelastic effect from consecutively-propped fractures within the same well.

Next, we calculated microseismic b-values to examine magnitude-scaling characteristics of microseismic events induced during hydraulic fracturing stimulation. We found that b-values calculated from microseismic event populations of individual stages vary considerably across the pad, but generally fall between 1.0 and 2.5, which

181

suggests a relative abundance of small magnitude to large magnitude events compared to naturally-occurring earthquake populations (b ≈ 1). However, a few event populations exhibit lower b-values near 1.0, suggesting that small patches of the fault or nearby off-fault structures may be seismically active during fluid injection.

We also applied the double-difference earthquake relocation technique in an attempt to improve microseismic event hypocenter locations to resolve potential seismogenic structures within the reservoir. While the double-difference relocation technique produces distinctive event hypocenter trends, synthetic tests indicate the trends unfortunately appear to be artifacts of the limited, dual-array microseismic monitoring configuration. While precise double-difference relocations appear unattainable in a dual-array monitoring configuration, adding a third monitoring array can greatly improve hypocenter relocation accuracy, suggesting the double-difference relocation technique could be viable with appropriate downhole monitoring configurations.

Finally, we found that microseismic events recorded during a single stage can be grouped into doublets and multiplets based on high waveform similarity. Inter- event hypocenter distances within these multiplets are up to 10 meters, suggesting the presence of repeated slip occurring on a specific fault patch. However, the doublets and multiplets only represent a small portion of the event population, indicating that repeated rupture on a specific fault patch is not the primary means of induced seismic deformation during hydraulic fracturing.

Collectively, our analysis suggests that the gas shale reservoir is characterized by a variety of responses to hydraulic fracturing stimulations. In particular, the response appears to be characterized by seismic deformation occurring on small fractures created or reactivated by the hydraulic fracturing stimulations, although there is also evidence for seismic deformation occurring on larger-scale localized structures within the reservoir. In addition, although not examined in this study, we note there is evidence for long-period seismic events occurring within the reservoir, although their

182

location and frequency of occurrence within the reservoir is unknown (I. Das, personal communication).

6.6 References

Adams, C., 2011, Summary of shale gas activity in northeast British Columbia 2011: British Columbia Ministry of Energy and Mines Oil and Gas Reports 2012-1, 25 p. Aki, K., 1965, Maximum likelihood estimate of b in the formula log N = a-bM and its confidence limits: Bulletin of the Earthquake Research Institute at Tokyo University, 43, 237-239. Aki, K., 1987, Magnitude-frequency relation for small earthquakes: A clue to the origin of F max of large earthquakes: Journal of Geophysical Research, 92, 1349-1355. Allan, J., and Creaney, S., 1991, Oil families of the Western Canadian basin: Bulletin of Canadian Petroleum Geology, 39, 107-122. Arrowsmith, S.J., and Eisner, L., 2006, A technique for identifying microseismic multiplets and application to the Valhall field, North Sea: Geophysics, 71, V31-V40. Baisch, S., Ceranna, L., and Harjes, H.-P., 2008, Earthquake cluster: What can we learn from waveform similarity?: Bulletin of the Seismological Society of America, 98, 2806-2814. Bell, J.S., and Grasby, S.E., 2011, The stress regime of the Western Canadian Sedimentary Basin: Geofluids, 12, 150-165. Bender, B., 1983, Maximum likelihood estimation of b values for magnitude grouped data: Bulletin of the Seismological Society of America, 73, 831-851. Boettcher, M.S., McGarr, A., and Johnston, M., 2009, Extension of Gutenberg-Richter distribution to MW -1.3, no lower limit in sight: Geophysical Research Letters, 36, L10307. British Columbia Ministry of Energy and Mines, 2005, Gas shale potential of Devonian strata, northeastern British Columbia: Petroleum Geology Special Paper 2005-1, 150 pp. Byerlee, J.D., 1978, Friction of rock: Pure and Applied Geophysics, 116, 615-626. Canada National Energy Board and British Columbia Ministry of Energy and Mines, 2011, Ultimate potential for unconventional natural gas in northeastern British Columbia’s Horn River basin: Oil and Gas Report 2011-1, 39, 39pp. Chou, Q., Gao,, J., and Somerwil, M., 2011, Analysis of geomechanical data for Horn River Basin gas shales, NE British Columbia, Canada: SPE paper 142498 presented at the SPE Middle East Unconventional Gas Conference and Exhibition held in Muscat, Oman, 31 January-2 February 2011. Downie, R.C., Kronenberger, E., and Maxwell, S.C., 2010, Using microseismic source parameters to evaluate the influence of faults on fracture treatments – a geophysical approach to interpretation: SPE paper 134772 presented at the SPE Annual Technical Conference and Exhibition held in Florence, Italy, 19-22 September 2010. East Jr., L.E., Grieser, W., McDaniel, B.W., Johnson, B., Jackson, R., and Fisher, K., 2004, Successful application of hydrajet fracturing on horizontal wells completed in a thick shale reservoir: SPE paper 91435 presented at the SPE Eastern Regional Meeting held in Charleston, West Virginia, USA, 15-17 September 2004. Ehlig-Economides, C.A, Ahmed, I., Apiwathanasorn, S., Lightner, J., Song, B., Vera, F., Xue, H., and Zhang, Y., 2012, Stimulated shale volume characterization: Multiwell case study from the Horn River Shale: II Flow perspective: SPE paper 159546 prepared for presentation at the SPE Annual Technical Conference and Exhibition held in San Antonio, Texas, U.S.A., 8-10 October 2012. Fisher, M.K., Heinze, J.R., Harris, C.D., Davidson, B.M., Wright, C.A., and Dunn, K.P., 2004, Optimizing horizontal completion techniques in the Barnett Shale using microseismic fracture mapping: SPE paper 90051 presented at the SPE Annual Technical Conference and Exhibition held in Houston, Texas, USA, 26-29 September 2004. Geller, R.J., and Mueller, C.S., 1980, Four similar earthquakes in central California: Geophysical Research Letters, 7, 821-824.

183

Gray, F.F., and Kassube, J.R., 1963, Geology and stratigraphy of Clark Lake Gas Field, British Columbia: Bulletin of the American Association of Petroleum Geologists, 47, 467-483. Gutenberg, B., and Richter, C.F., 1944, Frequency of earthquakes in California: Bulletin of the Seismological Society of America, 34, 185-188. Heidbach, O., Tingay, M., Barth, A., Reinecker, J., Kurfeß, D., and Müller, B., 2008, The world stress map database release 2008. http://www.world-stress-map.org. Last accessed August 2011. Jaeger, J.C., and Cook, N.G.W., 1979, Fundamentals of Rock Mechanics, 2nd ed., Chapman and Hall, New York. Johnson, J.G., Klapper, G., and Sandberg, C.A., 1985, Devonian eustatic fluctuations in Euramerica: Geological Society of America Bulletin, 96, 567-587. King, G.E., 2010, Thirty years of gas shale fracturing: What have we learned?: SPE paper 133456 presented at the SPE Annual Technical Conference and Exhibition in Florence, Italy, 19-22 September 2010. Kumano, Y., Asanuma, H., Niitsuma, H., Tezuka, K., and Kamitsuji, R., 2006, Reservoir structure at Yufutsu gas field, Japan determined by analysis of induced microseismic multiplets: Abstract presented at the 2006 SEG Annual Meeting in New Orleans, Louisiana, October 1-6, 2006. Kwiatek, G., Plenkers, K., Nakatani, M., Yabe, Y., Gresen, G., and JAGUARS-Group, 2010, Frequency-magnitude characteristics down to magnitude -4.4 for induced seismicity recorded at Mponeng Gold Mine, South Africa: Bulletin of the Seismological Society of America, 100, 1165- 1173. MacLean, B.C., and Morrow, D.W., 2004, Bovie structure: its evolution and regional context: Bulletin of Canadian Petroleum Geology, 52, 302-324. Maxwell, S.C.., Jones, M., Parker, R., Miong, S., Leaney, S., Dorval, D., D’Amico, D., Logel, J., Anderson, E., and Hammermaster, K., 2009, Fault activation during hydraulic fracturing: Abstract presented at the 2009 SEG Annual Meeting in Houston, Texas, October 25-30, 2009. Maxwell, S.C., Pope, T., Cipolla, C., Mack, M., Trimbitasu, L., Norton, M., Leonard, J., 2011, Understanding hydraulic fracture variability through integrating microseismicity and seismic reservoir characterization: SPE paper 144027 presented at the SPE North American Unconventional Gas Conference and Exhibition in The Woodlands, Texas, USA., 14-16 June 2011. McPhail, S., Walsh, W., Lee, C., and Monahan, P.A., 2008, Shale units of the Horn River Formation, Horn River Basin and Cordova Embayment, northeastern British Columbia: British Columbia Ministry of Energy and Mines Petroleum Geology Open File Report 2008-1, 14 p. Morrow, D.W., and Geldsetzer, H.H.J., 1988, Devonian of the eastern Canadian cordillera, In: McMillan, N.S., Embry, A.F., Glass, D.J., eds., Devonian of the world, Volume 1, Canadian Society of Petroleum Geology Memoir No. 14, 85-121. Nadeau, R.M., Foxall, W., and McEvilly, T.V., 1995, Clustering and periodic recurrence of microearthquakes on the San Andreas Fault at Parkfield, California: Science, 267, 503-507. Paige, C.C., and Saunders, M.A., 1982, LSQR: Sparse linear equations and least squares problems: ACM Transactions on Mathematical Software, 8, 195-209. Reynolds, M.M., and Munn, D.L., 2010, Development update for an emerging shale gas giant field – Horn River Basin, British Columbia, Canada: SPE paper 130103 presented at the SPE Unconventional Gas Shale Conference in Pittsburgh, PA, USA, 23-25 February 2010. Richards, B.C., 1989, Upper Kaskaskia Sequence: Uppermost Devonian and Lower Carboniferous, Chapter 9, In: Ricketts, B.D., ed., Western Canada Sedimentary Basin, a Case History, Canadian Society of Petroleum Geologists, 165-201. Richardson, E., and Jordan, T.H., 2002, Seismicity in deep gold mines of South Africa: Implications for tectonic earthquakes: Bulletin of the Seismological Society of America, 92, 1766-1782. Ross, D.J.K., and Bustin, R.M., 2008, Characterizing the shale gas resource potential of Devonian- Mississippian strata in the Western Canada sedimentary basin: Application of an integrated formation evaluation: American Association of Petroleum Geologists Bulletin, 92, 87-125. Rutledge, J.T., and Phillips, W.S., 2003, Hydraulic stimulation of natural fractures as revealed by induced microearthquakes, Carthage Cotton Valley gas field, east Texas: Geophysics, 68, 441-452. Schaff, D.P., Bokelmann, G.H.R., Ellsworth, W.L., Zanzerkia, E., Waldhauser, F., and Beroza, G.C., 2004, Optimizing correlation techniques for improved earthquake location: Bulletin of the Seismological Society of America, 94, 705-721. 184

Shi, Y., and Bolt, B.A., 1982, The standard error of the magnitude-frequency b-value: Bulletin of the Seismological Society of America, 72, 1677-1687. Seibel, M., Baig, A., and Urbancic, T., 2011, Single versus multiwall microseismic recording: What effect monitoring configuration has on interpretation: SPE paper 140525 presented at the SPE Hydraulic Fracturing Technology Conference and Exhibition in The Woodlands, Texas, USA, 24- 26 January 2011. Soliman, M.Y., East, L., and Adams, D., 2008, Geomechanics aspects of multiple fracturing of horizontal and vertical wells: Society of Petroleum Engineers Drilling & Completion, 23, 217-228. Utsu, T., 1965, A method for determining the value of b in a formula log n = a – bM showing the magnitude-frequency relation for earthquakes: Geophysical Bulletin of Hokkaido University, 13, 99-103. Urbancic, T.I., Shumila, V., Rutledge, J.T., and Zinno, R.J., 1999, Determining hydraulic fracture behavior using microseismicity, In: Amadei, Kranz, Scott, Smeallie, eds., Rock Mechanics for Industry: Proceedings of the 37th U.S. Rock Mechanics Symposium, 991-997. Vermylen, J.P., and Zoback, M.D., 2011, Hydraulic fracturing, microseismic magnitudes, and stress evolution in the Barnett Shale, Texas, USA: SPE paper 140507 presented at the SPE Hydraulic Fracturing Technology Conference and Exhibition in The Woodlands, Texas, USA, 24-26 January 2011. Waldhauser, F. and Ellsworth, W., 2000, A double-difference earthquake location algorithm: Method and application to the northern Hayward Fault, California: Bulletin of the Seismological Society of America, 90, 1353-1368. Warpinski, N.R., Kramm, R.C., Heinze, J.R., and Waltman, C.K., 2005, Comparison of single- and dual-array microseismic mapping techniques in the Barnett Shale: SPE paper 95568 presented at the SPE Annual Technical Conference and Exhibition held in Dallas, Texas, USA, 9-12 October 2005. Wu, R., Kresse, O., Weng, X., Cohen, C., and Gu, H., 2012, Modeling of interaction of hydraulic fractures in complex fracture networks: SPE paper 152052 presented at the SPE Hydraulic Fracturing Technology Conference held in The Woodlands, Texas, USA, 6-8 February 2012. Zoback, M.D., 2007, Reservoir Geomechanics, Cambridge University Press, Cambridge, United Kingdom. Zoback, M.D., Barton, C.A., Brudy, M., Castillo, D.A., Finkbeiner, T., Grollimund, B.R., Moos, D.B., Peska, P., Ward, C.D., and Wiprut, D.J., 2003, Determination of stress orientation and magnitude in deep wells: International Journal of Rock Mechanics & Mining Sciences, 40, 1049-1076.

185

Appendix 6A: General HYPODD Synthetic Tests

In addition to the synthetic tests performed in Section 6.4.3 which were specific to the monitoring array configuration for the Well 1 Stage 2 hydraulic fracture stimulation, we conduct an additional set of synthetic tests using a more general array configuration. The synthetic test setup consists of four arrays of five geophones arranged in a box pattern (Figure 6A1). Thirteen synthetic earthquake hypocenters are placed in an east-west trend centered in the middle of the array, and a constant, 1-layer velocity model with VP = 4,000 m/s and VP/VS = 1.73 is assumed over the region. The objective is to evaluate the ability of the double-difference technique to accurately relocate earthquake hypocenters with different monitoring array configurations.

Figure 6A1: Array and hypocenter configurations for the general HYPODD synthetic relocation test. Green circles represent synthetic hypocenter locations.

The first step of the relocation test is to calculate synthetic P- and S-wave travel times from all earthquakes to all recording sensors. No noise is added to the travel time measurements. Next, all earthquakes are given the same starting location (the center of the east-west hypocenter trend) and all earthquakes are allowed to pair with all other earthquakes during relocation (that is, absolute P- and S-wave differential travel times are calculated for all potential earthquake pairs). Finally, the hypocenters are relocated with the double-difference technique using the same HYPODD program and parameter settings as the synthetic tests described in Section

186

6.4.3. The relocated hypocenters using different array configurations are shown in Figure 6A2.

Earthquake hypocenter relocations using travel times from all four arrays are shown in Figure 6A2a. The relocated earthquakes (red dots) are generally restored quite accurately to their actual synthetic locations (green circles). The relocations are slightly less well-restored in depth than in map view. Similarly well-restored relocations are obtained when travel times from only three arrays are used (Figure 6A2b). The well-resolved hypocenter relocations in Figure 6A2b are consistent with the well-resolved hypocenter relocations obtained in Section 6.4.3 using three monitoring arrays.

187

Figure 6A2: Synthetic relocation test results. Green circles represent actual synthetic hypocenter locations and red dots represent double-difference relocations. Relocations were performed using travel times from (a) all four arrays, (b) three arrays (ABC), (c) Arrays A and B, (d) Arrays A and C, (e) Arrays A and D. Sensors from which travel times were used during the relocations are denoted with red triangles.

188

Hypocenter relocations using two arrays are shown in Figure 6A2c-e. Interestingly, when travel times from Arrays A and B or Arrays A and C are used (Figure 6A2c-d), hypocenters resolve quite well to their actual locations. This is in contrast to the synthetic test results from Section 6.4.3 illustrating that event hypocenters located with two monitoring arrays generally spread in a direction perpendicular to a plane connecting the two arrays, which was interpreted as a consequence of the lack of travel time constraints in that direction. Furthermore, when travel times from only diagonal arrays are used (Arrays A and D, Figure 6A2e), the hypocenters are not restored to their actual locations and instead delineate a NW-SE trend. This trend is parallel to a plane connecting the two monitoring arrays, and is exactly opposite of the trends observed in Section 6.4.3 using a similar azimuthally- distributed monitoring array configuration.

Although both the synthetic tests in Section 6.4.3 and this section utilized two monitoring arrays, it is important to note that the configurations were very different. The two arrays in Section 6.4.3 were located significant distances from the hypocenter trends, and comprised both horizontal and vertical orientations. In addition, although two arrays were available, they offered extremely limited azimuthal coverage. In contrast, the two-array configurations in Figures 6A2c-d provide considerably greater azimuthal coverage, and both vertical arrays straddle the hypocenter locations in depth. As both sets of synthetic tests were performed using the same HYPODD software and identical parameter settings, we conclude that the improved array coverage in azimuth and depth in this synthetic test account for the reasonably well- restored event hypocenters in Figures 6A2c-d. However, it is worth noting that the hypocenter trend resolved with Arrays A and C is oriented perpendicular to a plane connecting the two arrays (and therefore in the direction of least travel time constraint), suggesting the well-restored relocated hypocenters may only be coincidental.

The relocated hypocenter trends using Arrays A and D (Figure 6A2e) are a bit more unusual compared to the previous synthetic tests in Section 6.4.3. Again, both

189

tests attempted to relocate hypocenters in an extremely azimuthally-limited monitoring array configuration, and yield hypocenter trends clearly inconsistent with the actual hypocenter locations. However, the hypocenter trends in Section 6.4.3 are perpendicular to a plane connecting the two recording arrays while the hypocenter trends in Figure 6A2e are parallel to a plane connecting the two recording arrays. The reason for this discrepancy is not immediately clear, and would require additional tests beyond the scope of this study. Regardless of the underlying reason, the important observation is that this particular limited monitoring array configuration appears produce substantial artifacts in the double-difference relocated hypocenter locations.

In summary, the second set of synthetic tests presented in this appendix is largely consistent with the synthetic tests performed in Section 6.4.3. Both sets of synthetic tests demonstrate that hypocenter location accuracy using the double- difference technique improves as the number of available monitoring arrays increases. Interestingly, the relocations from the set of synthetic tests in this appendix suggest that it may be possible to obtain quality hypocenter locations with only two monitoring arrays provided they are well-oriented with respect to the hypocenters. The overall results corroborate our previous conclusion that hypocenter locations using the double- difference technique significantly improve with an increasing number of well- distributed monitoring arrays.

190