INTEGRATED USE OF TERRESTRIAL LASER SCANNING AND ADVANCED NUMERICAL METHODS FOR A TOTAL SLOPE ANALYSIS OF AFTERNOON CREEK, WASHINGTON by A L E X A N D E R B R I A N STROUTH B.A.Sc , Colorado School of Mines, 2004 A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE DEGREE OF M A S T E R OF APPLIED SCIENCE in THE F A C U L T Y OF G R A D U A T E STUDIES (Geological Engineering) THE UNIVERSITY OF BRITISH C O L U M B I A August 2006 © Alexander Brian Strouth, 2006 A B S T R A C T On November 9, 2003 a rock slide, involving approximately 750,000 m 3 of jointed, orthogneiss occurred at the Afternoon Creek ridge adjacent to the North Cascades Highway in northwest Washington. Most of the failed blocks slid into the shallowly-sloping Afternoon Creek, and did not reach the highway. However, a small fraction of the failed volume traveled down the west side of the ridge, and entered the steep, bedrock-exposed Falls Creek. This debris fell more than 600 meters and impacted the highway, destroying a section of the roadway. There is potential for future rock slope failures to occur at this site, thereby presenting a hazard to the highway below. The purpose of this research was to investigate the November, 2003 rock slope failure and post-failure motion, and use the results of the investigation to assess the location, volume, and effects of future slope failures at Afternoon Creek. A Total Slope Analysis methodology was followed, which linked the failure initiation analysis with runout analysis. The structural and topographic controls to failure were analyzed with limit equilibrium, and the 2-D and 3-D distinct element codes, UDEC and 3DEC. Runout was analyzed with the 3-D dynamic analysis code, DAN3D. A data collection methodology designed specifically for numerical modeling was implemented. A laser scanner and field survey were used to collect discontinuity characterization data at the inaccessible, hazardous site. The analysis suggested that the past event involved a single-stage, extremely-rapid translational failure on a highly-persistent, moderately-dipping joint set. Material that entered Falls Creek was most likely caused by secondary rockfall immediately following the slide. A future rock slope failure up to 300,000 m 3 emanating from the existing failure scarp is possible; however runout debris from an event of this volume is not expected to reach the highway at Afternoon Creek. In contrast, retrogression of the failure scarp crest by sliding or toppling of individual columns may cause rockfall to enter the steeper Falls Creek travel path where it will most likely impact the highway embayment. ii T A B L E O F C O N T E N T S ABSTRACT .' i i T A B L E OF CONTENTS. i i i LIST OF TABLES vii LIST OF FIGURES ix A C K N O W L E D G E M E N T S xv 1 INTRODUCTION 1 1.1 Problem Statement 1 1.2 Research Objectives 2 1.3 Organization of the Paper 2 1.4 Literature Review 3 1.4.1 Slope Failure Initiation 3 1.4.2 Rock Mass Characterization 8 1.4.3 Runout Prediction of Failed Slopes 10 2 PROJECT SETTING: The Afternoon Creek Rockslide 13 2.1 Introduction 13 2.2 Geological and Geomorphological Setting 15 2.3 History of Activity 17 2.4 Chronology of Recent Mass Wasting Events 18 2.5 Continued Monitoring, Hazard Mitigation, and Future Work 20 2.6 Conclusions 22 3 M E T H O D O L O G Y 23 3.1 Total Slope Analysis Procedure 23 3.2 Traditional Data Collection Techniques 24 3.2.1 Desk Study 25 3.2.2 Field Mapping 27 3.2.3 Discontinuity Mapping 28 3.3 Terrestrial Laser Scanning (LiDAR) 30 3.3.1 L i D A R Data Acquisition 30 3.3.2 L i D A R Data Processing 34 3.3.3 L i D A R Data Analysis 36 3.4 Computer Analysis Techniques 41 i n 3.4.1 Geographical Information System (GIS) 41 3.4.2 Spheristat 2.2 41 3.4.3 RocLab 42 3.4.4 Limit Equilibrium Methods (RocPlane, Swedge) 42 3.4.5 Universal Distinct Element Code (UDEC) 42 3.4.6 Three Dimensional Distinct Element Code (3DEC) 42 3.4.7 Surfer 43 3.4.8 Three-Dimensional Dynamic Analysis (DAN3D) 43 3.4.9 Three-Dimensional Rockfall Simulation.(PIR3D) 43 3.5 Numerical Modeling Procedure 43 4 PROCESSING OF FIELD D A T A FOR N U M E R I C A L M O D E L I N G 45 4.1 Physiographic & Geologic Setting 45 4.1.1 Regional Setting 45 4.1.2 Local Geology 46 4.2 Geometric Parameters 53 4.2.1 Topography 53 4.2.2 Description of Joint Sets 54 4.2.2.1 Orientation 54 4.2.2.2 Spacing 58 4.2.2.3 Persistence ••• 58 4.3 Mechanical Properties 59 4.3.1 Rock Mass Properties 59 4.3.2 Discontinuity Properties 61 5 B A C K A N A L Y S I S OF THE 2003 AFTERNOON C R E E K SLOPE F A I L U R E 63 5.1 Failure Initiation in Afternoon Creek 63 5.1.1 Kinematic Analysis 64 5.1.2 Limit Equilibrium Analysis 67 5.1.3 Two-Dimensional Numerical Modeling 71 5.1.3.1 2-D Representation 71 5.1.3.2 Modeling Methodology 74 5.1.3.3 Baseline Model Results 78 5.1.3.4 Trial Model Results 85 5.1.4 Three-Dimensional Numerical Modeling 95 5.1.4.1 Modeling Methodology 95 5.1.4.2 3-D Model Results 98 5.2 Runout Analysis along the Afternoon Creek Travel Path 105 5.2.1 Runout Analysis Methodology 105 5.2.2 Runout Analysis Results 107 5.3 Falls Creek Travel Path 112 5.3.1 Mechanism 1: Translational Failure 113 5.3.2 Mechanism 2: Toppling Failure 117 5.3.3 Mechanism 3: Spreading of a Flowing Mass 121 iv 5.3.4 Mechanism 4: Rockfall caused by dilation of the failed mass 124 5.4 Total Slope Back Analysis Summary 127 5.4.1 Afternoon Creek Travel Path 127 5.4.2 Falls Creek Travel Path 130 6 ANALYSIS OF FUTURE EVENTS AT AFTERNOON C R E E K 133 6.1 Failure Volume Assessment 133 6.1.1 Methodology 134 6.1.2 Locations of Slope Instabilities 136 6.1.3 Summary of Potential Rock Slope Hazard Sources 145 6.2 Analytical Runout Assessment 146 6.2.1 Source Rank 1: Toppling and Rockfall at Slope Crest 147 6.2.2 Source Rank 2: Mid-Slope Failure Scarp 150 6.2.3 Source Rank 3: Translational Failure at Slope Crest 154 6.4 Hazard Assessment Summary 157 7 DISCUSSION OF TOOLS A N D METHODS 158 7.1 Terrestrial Laser Scanning (LiDAR) 158 7.1.1 L i D A R Benefits 158 7.1.2 L i D A R Limitations 159 7.1.3 L i D A R Recommendations 160 7.2 Numerical Modeling Tools 162 7.2.1 UDEC 162 7.2.2 3DEC 163 7.2.3 DAN3D 164 7.3 Total Slope Analysis Method 165 8 CONCLUSIONS A N D RECOMMENDATIONS 167 8.1 Conclusions 167 8.2 Recommendations for Further Work 169 REFERENCE LIST 172 APPENDIX A: SCANLINE S U R V E Y D A T A 179 APPENDIX B: L i D A R POINT CLOUDS A N D PHOTOGRAPHS 181 APPENDIX C: VERIFICATION OF METHODS 183 C . l Discontinuity Orientation 183 C.2 Discontinuity Set Spacing 186 v APPENDIX D: SPACING A N D PERSISTENCE C A L C U L A T I O N S 188 D . l Average Joint Set Spacing 188 D. 2 Average Joint Set Persistence 188 APPENDIX E: N U M E R I C A L M O D E L I N G INPUT FILES 193 E. l UDEC Baseline Model A - A ' 193 E.2 3DEC Model 195 vi L I S T O F T A B L E S Table 1.1 Comparison of classification schemes 6 Table 3.1 Data requirements for numerical modeling 25 Table 3.2 Terrestrial laser scans attempted 32 Table 4.1 Summary of parameters that describe discontinuities in the Afternoon Creek rock slope 54 Table 4.2 Estimates of persistence and exposed persistence in structural Zone 3 59 Table 4.3 Estimated range of rock mass mechanical properties in each structural zone 60 Table 4.4 Estimated range of discontinuity strength and stiffness parameters for numerical modeling 62 Table 5.1 Geometric and mechanical parameters for baseline U D E C models 77 Table 5.2 Mechanical parameters of Zone B modeled as an equivalent continuum 85 Table 5.3 Range of reasonable mechanical parameters for zone 87 Table 5.4 Range of geometric parameters used in the trial models 89 Table 5.5 Geometric and mechanical parameters for the 3DEC model : 98 Table 5.6 Mechanisms that allow debris to enter Falls Creek - Ranked in order of importance 132 Table 6.1 Geometric configurations of the forward analysis U D E C models, Afternoon Creek 135 Table 6.2 Summary of the three hazard sources at Afternoon Creek 145 Table 6.3 Summary of runout models of the source volume at the middle of the failure scarp. 151 Table 6.4 Summary of runout models of the failure scarp crest source volume 154 Table C . l The 10 meter portion of scanline survey 1 considered in this example 183 Table D.2 Discontinuity persistence estimated from photograph trace length measurements. 190 Table D.3 Discontinuity exposed persistence estimated by direct measurement on the 3-D point cloud 191 vii Table D.4 Discontinuity exposed persistence estimated by relating the exposed persistence to patch area in the 3-D point cloud 192 v i i i L I S T O F F I G U R E S Figure 2.1 Afternoon Creek rockslide above SR 20 near Newhalem, WA. Photograph provided by John Scurlock, Concrete, W A 14 Figure 2.2 Location of Afternoon Creek rockslide. Three kilometers east of Newhalem, Washington, USA. Failure area outlined in yellow 15 Figure 2.3 Sub-vertical, open fractures parallel to Afternoon Creek. Photograph provided by URS Corporation (photograph date, March 1, 2004) 17 Figure 2.4 Historical photographs of debris flows that reached the state route 20 highway. (a) Afternoon Creek debris flow material (March 23, 1949). (b) Falls Creek Chute debris flow covered with fresh snow (1990). (c) Afternoon Creek (1999 or 2000). Photographs provided by WSDOT 18 Figure 2.5 Afternoon Creek rockslide slope before the November 9, 2003 event. Photograph provided by WSDOT (date unknown) 18 Figure 2.6 Afternoon Creek failure scarp from Afternoon Creek, (photograph date April, 2005) 19 Figure 2.7 Digital elevation model of the Afternoon Creek rockslide. Failed mass is shown in gray. Arrows indicate direction of movement. Dotted blue line indicates the extent of path 20 Figure 3.1. Flow chart of the repeated Total Slope Analysis. Back analysis and characterization of current hazard 24 Figure 3.2 URS structural mapping location compared to failure zone 26 Figure 3.3 Schematic cross-sections perpendicular and oblique to Afternoon Creek 28 Figure 3.4. Oblique photo of the Afternoon Creek scarp, showing location of scanline survey 1 (75 meters length) and scanline survey 2 29 Figure 3.5 Terrestrial laser scanning stations 31 Figure 3.6 Histograms of dip direction vs. frequency for four scans of structural Zone 3 showing a bias in the orientation of automatically generated patches. Joint planes that strike perpendicular to the scanner position tend to reflect more laser strikes. Line-of-sight of the scanner is superimposed over the histogram 33 Figure 3.7 Point cloud processing procedure 35 I X Figure 3.8 Comparison of automatically generated patches in point clouds 2/1, 6/1, 7/1, 7/2. Minimum patch size =10; minimum neighbor angle = 4 degrees (top); minimum neighbor angle =10 degrees (bottom). Pole size is scaled to patch area 37 Figure 3.9 Schematic, annotated patch 40 Figure 4.1 Geologic map of Afternoon Creek 47 Figure 4.2 Structural domain Zone 1 49 Figure 4.3 Structural domain Zone 2 50 Figure 4.4 Structural domain Zone B 51 Figure 4.5 Structural domain Zone 3 viewed from three directions 52 Figure 4.6 Contour map and topographic profiles derived from the L i D A R D E M after the November, 2003 event. Arrows indicate the path of steepest descent 53 Figure 4.7 Joint sets A and B in structural domain Zone 3 55 Figure 4.8 Structural Zone 3. Contoured stereographic projection of automatically generated patches that represent discontinuity surfaces measured in scanline survey JS2 and in point clouds 2/1, 6/1, 7/1, 7/2. minimum neighbor angle = 6° and minimum patch size=20. Vectors indicate the L i D A R scan line-of-sight 56 Figure 4.9 Structural Zone 2. Stereographic projection of discontinuities exposed at the base of the failure scarp. Measured during scan line survey JS1 and point cloud analysis of scan 3/4 57 Figure 4.10 Stereographic projection of automatically generated patches that represent discontinuity surfaces in point clouds 2/1, 6/1, 7/1, 7/2. Pole size is scaled to patch area. Joint set orientation is the average orientation of poles included in the set 57 Figure 5.1 Topographic map of Afternoon Creek after the November, 2003 failure, showing the average and local slope orientation used in the kinematic analysis 65 Figure 5.2 Stereographic projection of the important discontinuity sets compared with the average orientation of the Afternoon Creek slope 66 Figure 5.3 Stereographic projection of the important discontinuity sets compared with the local orientation of the Afternoon Creek slope at the interface of Zone B and Zone 3 67 Figure 5.4. Illustration of the simplified limit equilibrium analysis models. The Rocscience program Swedge was used for wedge stability analysis; Rocplane was used for planar stability analysis 68 x Figure 5.5 Comparison of the joint strength properties necessary to achieve stability for the three failure modes 69 Figure 5.6 Summary of the parametric study - Percent Change versus Factor of Safety 70 Figure 5.7 Two-stage failure hypothesis showing failure masses and movement directions. 72 Figure 5.8 Map of Afternoon Creek (after the November, 2003 failure) showing location of cross-sections and source volume (contour units = meters) 72 Figure 5.9 Cross-section A - A ' before and after the November, 2003 failure. The post-glacial topographic profile is inferred 73 Figure 5.10 Cross-section B - B ' before and after the November, 2003 failure. The post-glacial topographic profile is inferred 73 Figure 5.11 Flow chart of modeling procedure 74 Figure 5.12 Block outline for cross sections A - A ' and B - B ' 75 Figure 5.13 Horizontal (Sxx) and vertical.(Syy) stress contours for baseline models A and B. .' 76 Figure 5.14 Baseline model A - A ' results, damage states 2 and 4 relative to November, 2003 slide surface; horizontal displacement contours (top); total displacement vectors and joint shear indicators (red) (bottom) • 81 Figure 5.15 Baseline model B - B ' results, damage states 3 and 4 relative to November, 2003 slide surface; horizontal displacement contours (top); total displacement vectors and joint shear indicators (red) (bottom) 82 Figure 5.16 Baseline model A - A ' results, damage state 2 and 4 relative to the November, 2003 slide surface; plasticity indicators 83 Figure 5.17 Baseline model B - B ' results, damage state 2 and 4 relative to the November, 2003 slide surface; plasticity indicators 84 Figure 5.18 Zone B modeled as an equivalent continuum, damage state 4 showing horizontal displacement contours and plasticity indicators; cross-section A - A ' (top); cross-section B - B ' (bottom) 86 Figure 5.19 Comparison of tensile yielding for 3 different rock mass tensile strength values. Damage state 4, cross-section B - B ' models 88 Figure 5.20 Trial models of reduced joint set spacing, damage state 4, showing horizontal displacement 90 Figure 5.21 Trial models with variable joint set B persistence and orientation, damage state 4 showing horizontal displacement 92 x i Figure 5.22 Trial models with joint set B dip = 70°, joint set A dip = 40°, damage state 4 showing horizontal displacement. Joint friction angle = 28° 93 Figure 5.23 Trial models with one joint set, dip = 50°, damage state 4, showing horizontal displacement and plasticity indicators 94 Figure 5.24 The 3-D block model of Afternoon Creek, colored by region 96 Figure 5.25 3-D model results, (a) 3-D view of the failed mass from 'inside' the slope 99 Figure 5.25 continued 3-D model results, (b) Cross-section of failed volume showing filled contours of total displacement. Primary direction of sliding is out of the page, (c) Indicators of plastic yielding of the rock mass, showing the approximate location of the November, 2003 sliding surface 100 Figure 5.26 3-D model plot colored by plastic yielding state of surface zones 101 Figure 5.27 Plastic yielding indicators and total displacement vectors compared with the actual sliding surface. Equivalent location as cross-section A - A ' (top) and B - B ' (bottom) 102 Figure 5.28 Cross-section showing plastic yielding in Zone B 103 Figure 5.29 Difference map overlying the before D E M . Increased mass shown in blue, decreased mass shown in red. Contour interval = 20 m 107 Figure 5.30 Photographs of the Afternoon Creek debris. Note person for scale 108 Figure 5.31 Calibrated DAN3D analysis results. Single-stage analysis results 109 Figure 5.32 Calibrated DAN3D analysis results. Two-stage analysis I l l Figure 5.33 Illustration of the translational failure mechanism in Falls Creek 113 Figure 5.34 Photograph of the crest of the Afternoon Creek/ Falls Creek ridge from the Falls Creek side 114 Figure 5.35 Map of pre-failure topography showing location of cross-section C - C . Filled red contours indicate location and thickness of the failed zone 115 Figure 5.36 U D E C models of the translational failure mechanism before and after removal of Afternoon Creek material. (A) and (B) show Factor of Safety and horizontal velocity contours at failure. (C) and (D) show contours of the major principal stress 116 Figure 5.37 Column toppling towards Afternoon Creek. Photograph date December, 2004; provided by URS Corporation 118 xn Figure 5.38 UDEC model with the measured joint orientations (run 1) showing potential for toppling towards Falls Creek.(A) and (B) show total displacement vectors and plasticity indicators. (C) and (D) show contours of x-displacement 119 Figure 5.39 U D E C model with cross-joints orthogonal to set B (run 2) showing potential for toppling towards Falls Creek.(A) and (B) show total displacement vectors and plasticity indicators. (C) and (D) show contours of x-displacement 120 Figure 5.40 DAN3D simulations during runout. Red contours indicate source zone. Filled blue contours indicate deposit thickness. Dotted black line indicates the trimline mapped on air photos 123 Figure 5.41 The crest of the Aftenoon Creek-Falls Creek ridge showing scarring caused by rockfall 124 Figure 6.1 Post-failure block model, cross-section A - A ' and B - B ' 134 Figure 6.2 Cross-section A - A ' failure volume assessment for future events; horizontal displacement contours. Geometric configurations 1, 2, and 3 137 Figure 6.3 Cross-section B - B ' failure volume assessment for future events; horizontal displacement contours. Geometric configurations 1, 2, and 3 139 Figure 6.4 Two potentially unstable areas of the Afternoon Creek slope. Photograph by Erik Eberhardt (August 2005) 142 Figure 6.5 Map and schematic cross-section D-D' showing estimated location, thickness, and volume of the slope hazard sources 144 Figure 6.6 Example of an unstable column that has potential to cause rockfall in Falls Creek. August ,2005 145 Figure 6.7 Completed embayment on Washington State Route 20 at Falls Creek in February, 2006. Photograph provided by WSDOT 148 Figure 6.8 Rockfall simulation with PIR3D, showing plan view and 3-D view 149 Figure 6.9 Map showing the Falls Creek rockfall source zone boundary line 149 Figure 6.10 DAN3D runout assessment of an Afternoon Creek rock avalanche originating from middle of the failure scarp, showing source volume depth (meters), deposit depth (meters), and maximum velocity. Benchmark model results (top); "worst-case" trial model 3 results (bottom) 153 Figure 6.11 DAN3D runout assessment of an Afternoon Creek rock avalanche originating from the failure scarp crest, showing source volume depth (meters), deposit depth (meters), and maximum velocity. Baseline model results (top); 'worst case scenario' trial model 3 results (bottom) 156 xi i i Figure A l . Photograph of scanline 1 from position 0 to 2000 cm 180 Figure B l . Digital image and 3-D point cloud of scan 2/1. Scan area is outlined in red on the digital photo 181 Figure B2. Digital image and 3-D point cloud of scan 6/1. Scan area is outlined in red on the digital photo 181 Figure B3. Digital image and 3-D point cloud of scan 7/1. Scan area is outlined in red on the digital photo 182 Figure B4. Digital image and 3-D point cloud of scan 7/2. Scan area is outlined in red on the digital photo 182 Figure CI . A9 point cloud showing inserted patches 184 Figure C2. Stereonet showing joint surface orientation estimated from the two methods -point cloud patch method (blue triangles); and hand measurements (black circles) from table B l 185 Figure C3. Comparison of point cloud 3/4 patches (blue triangles) with entire scanline survey in Zone 2 185 Figure C4. 3-D point cloud with virtual scan line 187 Figure C5. Scanline survey JS1 187 x i v A C K N O W L E D G E M E N T S Firstly I would like to thank my wife and best-friend, Amanda, for her encouragement, support, and love during the course of this research. Without her I would be lost. This work would not have been possible without the guidance and funding of my supervisors Dr. Erik Eberhardt and Dr. Oldrich Hungr. Dr. Eberhardt suggested the topic, organized the project, and directed my research. Both provided countless hours of individual and classroom instruction, covering landslides, rock mechanics, engineering, computer modeling, thesis writing and more. Thank you most of all for keeping an open door and allowing so much time for discussion and instruction. Thank you also for funding and organizing research related trips to the U.S., Nepal, and Chile. These are experiences that I will never forget. I would like to thank my committee members Dr. R. Beckie, and Dr. D. Stead, and external examiner Tom Badger for providing their time, guidance, instruction, and discussion throughout the course of this research. The Washington State Department of Transportation funded the site visits and provided open access to information regarding Afternoon Creek. URS Corporation and Wyllie & Norrish Rock Engineers generously provided open access to all information regarding the Afternoon Creek site. Special thanks to Dr. R.L. Burk from URS Corp. for his guidance and support. Rosie Cobbett, Ming Yang, Suzanne Chalindar, and my supervisors assisted during site visits and laser scanning. Amandine Brosse assisted with the 3-D rockfall computer simulation. Finally, I would like to thank my office mates Andrea Kay and Scott McDougall for their friendship and help. Andrea provided answers, software support, and entertainment. Scott was my mentor and general role model. xv 1 I N T R O D U C T I O N 1.1 Problem Statement On November 9, 2003 a rock slope failure occurred at Afternoon Creek in northwest Washington State that involved approximately 750,000 m 3 of jointed, orthogneiss rock. The source for this event was located at the crest of a ridge. Most of the debris fell to the east of the ridge into the shallow sloping Afternoon Creek. This material moved as a rock avalanche for approximately 500 meters before coming to rest. The rock avalanche did not intersect the highway. A small percentage of the debris fell to the west of the ridge down the steep Falls Creek and Falls Creek Chute. Much of this material traveled all the way to Washington State Route 20 (SR 20), destroying portions of the roadway and guardrail, and depositing up to 4-meter diameter boulders on the road. The Afternoon Creek rock slope continues to threaten SR 20 at two locations due to the topography of the slope. If another slope failure does occur, the debris could potentially impact the highway as a rock avalanche from Afternoon Creek or as rockfall from Falls Creek. Due to the imminent threat posed by the slope, research was undertaken through this thesis study to investigate the November, 2003 Afternoon Creek rock slope failure and post-failure motion, and to characterize the hazard that currently exists at Afternoon Creek. The investigation focused on the structural and topographic controls of failure and runout. An analysis methodology was used that integrated several numerical methods linking failure initiation to runout, termed "Total Slope Analysis" (Stead and Coggan 2005; Stead et al. 2006). The analysis was preceded by a data collection program designed specifically for the numerical modeling application. A secondary purpose of the research was to evaluate integrated use of the state-of-the-art data collection and numerical modeling tools utilized during the investigation, including: terrestrial laser scanning (LiDAR) for collecting rock mass characterization data of an inaccessible slope; two-dimensional (2-D) and three-dimensional (3D) discontinuum numerical modeling of the failure initiation processes; and three-dimensional dynamic analysis of the post-failure motion. 1 1.2 Research Objectives The objectives of this research were as follows: 1. Investigate the structural and topographic controls to the operative deformation and failure mechanisms of the November 9, 2003 rock slope failure at Afternoon Creek. 2. Characterize the post-failure motion of the debris. 3. Estimate the location and volume of future slope failures at Afternoon Creek, and determine the effects of the future failures on the highway (SR 20). 4. Evaluate the integrated use of state-of-the-art techniques in rock slope characterization and stability analysis including: a. Light Detection And Ranging (LiDAR) scanning for collecting rock mass characterization data, and b. Two-dimensional and three-dimensional discontinuum numerical modeling. 1.3 Organization of the Paper Chapter 1 presents the research objectives, and Chapter 2 then presents a description of the Afternoon Creek rockslide case history. The case history has been published as the following peer-reviewed journal article: Strouth, A. , Burk, R.L., Eberhardt, E. 2006. The Afternoon Creek rockslide near Newhalem, Washington. Landslides, 3 (2): 175 -179. Chapter 3 is a discussion of the methods used to complete this research. It includes a description of the Total Slope Analysis procedure, numerical modeling procedure, and the data collection methods. The computer programs used during the research are briefly described. Chapter 4 describes the results of the data collection program. The data collection program was designed specifically to acquire information required to build the distinct element numerical models. Chapter 5 presents the back analysis performed for the November 9, 2003 slope failure at Afternoon Creek. The Total Slope Analysis of the Afternoon Creek and Falls Creek travel paths are described including the 2-D and 3-D distinct element modeling, coupled with the rock avalanche runout analysis. Mechanisms that cause rocks to enter Falls Creek are investigated and the results of an earlier rockfall analysis are summarized. Chapter 6 is an analysis of future events at Afternoon Creek, using 2 the Total Slope Analysis as a forward analysis to estimate the location, size, and effects of future slope failures at Afternoon Creek. Chapter 7 is a discussion of the state-of-the-art tools used in these analyses, specifically terrestrial laser scanning, 2-D and 3-D distinct element modeling, and 3-D dynamic analysis. Benefits, limitations, and recommendations concerning each tool are discussed. Chapter 8 summarizes the most important conclusions of this research and provides recommendations for future work that relate to the Afternoon Creek case study and the tools used. 1.4 Literature Review 1.4.1 S lope Failure Initiation Mechanisms of Failure In 'Mechanism of Landslides' (1950), Terzaghi describes the processes contributing to landslides, and discusses their dynamics. He divides the causes of landslides into internal and external ones. The external causes are those that produce an increase in the shearing stresses, and internal causes are those that decrease the shear strength (resistance) of the slope material. Undercutting the foot of a slope, loading the top of a slope, or adding earthquake loading are examples of external causes. Common internal causes are increased pore-water pressures and decreased slope material strength. Terzaghi observed that most slope failures take place during periods of exceptionally heavy rainfall, and asks us to remember that "exposure to rain or melting snow belongs to the normal existence of a slope. Hence, i f a slope is old, heavy rainstorms or rapidly melting snow can hardly be the sole cause of a slope failure". Most landslides are a result of the combination of several causes. It is common for the shear strength of an old slope to decrease through time by weathering and fracture growth, and failure to be triggered by exceptionally large pore water pressures. In a subsequent paper focused on the stability of rock slopes (1962), Terzaghi discusses the significance of joints in hard unweathered rock, noting that the slope stability is determined by the shear strength of joints and faults, and not by the strength of the rock itself. The cohesion of a joint is partly a result of intact rock-bridges between discontinuous sections of the joint. Pore water pressure is also an important component to the shear strength of rock joints, however the water table is often poorly defined in a rock slope 3 because the secondary permeability of jointed rock commonly varies erratically across a slope.. , Numerous authors have researched the shear strength of joints and stress-strain relationships of laboratory rock samples in further detail. Some attempts are made to compare these with larger scale rock mass failures in tunnels (Hajiabdolmajid and Kaiser 2002; Hajiabdolmajid et al. 2003; Lockner 1995). Lockner discusses brittle failure mechanisms in rock in the context of the Coulomb failure criterion. The conceptual model of brittle fracture is described for samples that are loaded in tension, and in compression. Hajiabdolmajid and others focus on the center portion of the stress-strain curve between the onset of microcracking and movement along a shear plane. The relationship between loss of cohesional strength and mobilization of frictional strength is discussed for rock samples and around tunnel openings. Other authors consider stress-strain relationships, and progressive brittle failure in natural rock slopes with discontinuous joints (Cording et al. 2002; Eberhardt et al. 2004; Hajiabdolmajid and Kaiser 2002; Sjoberg 1999). Each of these authors distinguish failures on continuous joints from those that require shear surface development through intact (although weakened/deformed) rock. Cording et al. (2002) investigate the relationship between the shape of the failure plane and the dip of discontinuous joints. The other authors investigate the degradation of rock slope stability with time as the rock slope deforms and fractures propagate. Varga (2003) divides the pre-collapse process of a slope into 3 different stages: (1) primary elasto-plastic stage, (2) secondary plastic deformation stage, (3) Tertiary viscoelastic movement. Walder and Hallet (1985) present a model of fracture propagation in rock during freezing. Calculations and empirical data indicate that sustained freezing is most effective in producing crack growth when ample water is available, and temperatures range from -4° to -15° C. Failure Classifications Several classifications of failure mechanisms for large rock slopes exist. As stated by (Martin and Kaiser 1984) and echoed by several others, " A proper understanding of the mechanisms taking place during the failure process... is a fundamental necessity for the 4 selection of the appropriate method of analysis." For each classification unstable rock masses are classified according to different parameters. Lansheng & Zhouyjan (1984), Lansheng et al. (1992), Poisel & Preh (2004), and Hungr & Evans (2004) consider the type of deformation and failure mechanism common to different structural settings. Table 1.1 is a comparison of these classification schemes. Lansheng and others classify slope failure and deformation by the slope structure and five 'geomechanical deformation models'. The geomechanical deformation models are: (1) sliding (or creep sliding) - cracking (SC or CSC); (2) sliding and compressed cracking (SCC); (3) sliding and bending (SB); (4) bending and cracking (or toppling) (BC); and (5) plastic-flowing and cracking (PFC) (Lansheng and Zhuoyuan 1984). Hungr and Evans describe the typical failure behavior (i.e., slow, rapid, or catastrophic) of each failure mode in strong and weak rock. For example, strong, brittle rock is more likely to fail catastrophically, and weak, ductile rock commonly fails slowly. Translational slides usually are extremely rapid in strong and weak rock (Hungr and Evans 2004). Poisel and Preh (2004) have created a catalogue of rock slope failure mechanisms and point out that an ideal model would simulate both the initiation of failure and the run out. The catalogue considers the geological setting, geometry of the slope, joint structure, configuration of the rock blocks, and mechanical behavior of the rock mass. Varga & Gorbushina (1988) point out that the type of structure is important to the failure mechanism as well as the orientation. Five main types are identified. The controlling structure may be any one of the following or a combination of several: bedding, faults, joints, schistosity, injective structure (magmatic intrusion contact). Martin & Kaiser (1984) and Eberhardt et. al. (2004) classify slope failures by the amount of deformation in the moving rock mass. Martin and Kaiser separate failures into 3 classes: (Class I) rigid body motion; (Class II) local yielding; (Class III) yielding of the entire rock mass. Eberhardt and others note that the analysis method depends largely on the extent of internal deformation. Examples of failure mechanism are given for minimal and extensive internal rock mass deformation. 5 Table 1.1 Comparison of classification schemes Joint orientation/ orientation of the failure surface (Lansheng et al. 1992) (Hungr& Evans 2004) (Poisel & Preh 2004) Type of Slope Structure Model of deformation Failure mode Failure mode Behavior Martin and Kaiser Failure Class Failure mode Martin and Kaiser Failure Class Homogeneous Soil, weak rock mass CSC Curved rotary slide Rock slump Slow - rapid, self stabilizing I (III ?) Slope Creep III Horizontal - sub horizontal sec Horizontally pushing slide; rotary slide Compound slide Slow, self stabilizing Ill Gentle dip out of SC Block slide Translational rock block Extremely rapid, potentially large 1 Sliding of a rock block on a single discontinuity; I slope Compound slide Slow, self stabilizing III Sliding of several rock blocks on a polygonal sliding plane III Consequent Kink band slumping III Moderate dip out of slope SB slide; rotary slide; rockfall Translarional rock block Extremely rapid, large 1 Sliding of a rock block on a single discontinuity I Steep dip out of B C Topples and rockfalls; rotary slide Block topple • - Gradual piecemeal disintegration I I I Buckling of column or slab-shaped rock blocks; Kink III One controlling discontinuity or Joint Set slope - large catastrophic rock avalanche i , n band slumping; Rock Slumping Topples and rockfalls, rotary slide in depth Flexural toppling Self stabilizing; ductile behavior in Flexural Toppling III Near vertical dip into slope B C or CSC Block topple - Gradual piecemeal disintegration i • i i toppling of column or slab- i - I I - large catastrophic rock avalanche i , i i shaped block i , t i Moderate dip into slope CSC or B C Surface slide, rotary slide Flexural toppling Self stabilizing; ductile behavior III Toppling of column or slab-shaped block I; I I Varying dip angle towards slope SB Consequent and rotary Structurally defined compound slide Extremely rapid III slide Block slide with toe breakout Depends on rock strength II; III Blocky mass 2 or more controlling Joint sets SC Wedge slide or flowslide Translational rock block or wedge slide Extremely rapid, potentially large 1 Falling rock blocks; sliding of blocks on two discontinuities; rotation of a single rock block 1 Broken Mass (Many controlling joint sets) Numerous joint sets, no prominent weak plane SC or CSC Wedge slide or flowslide Rock Collapse Extremely rapid, generally small volume III (I?) Sliding of a fractional body on a shelly, newly formed sliding surface III (I?) Slope with soft foundation PFC Block slide; rock fall; rotary slide Block Topple Gradual piecemeal disintegration I; II Translational or rotational descent of tower blocks of competent rock upon an incompetent base i; I I CSC or SC: (creep) sliding - cracking; SCC: Sliding - compression cracking; SB: Sliding - bending; BC: bending - cracking; PFC: plastic flowing - cracking (Langsheng et al. 1992). (Class I): Rigid body motion (translation or rotation) along a planar or circular failure surface. No internal yielding of the rock mass required. (Class II): Local yielding inside the rock mass required allowing mass movement along the irregular basal slip surface. (Class III): Yielding of the rock mass along pervasive critically oriented internal shear surfaces required to allow motion along the basal slip surface (Martin & Kaiser 1984). 6 Numerical Modeling Stability analyses of rock slopes are routinely performed with numerical techniques. Stead et. al. (2001) describe which numerical techniques are best suited for different classes of failure and/or deformation. Common simple techniques, such as limit equilibrium analysis and empirical methods, have relevant applications and are appropriate for situations where a 2-D rigid block assumption is valid. However, these techniques have many important limitations - for example, they are not able to analyze creep, progressive deformation, and extensive internal disruption that precedes or follows a sliding failure (Stead et al. 2001). Where it is necessary to include the stress state within the rock mass and the influence of complex deformation and brittle fracture, numerical modeling techniques should be used. Three categories of numerical methods exist: continuum, discontinuum, and hybrid models. Continuum models are best suited for analysis of slopes comprised of massive, intact rocks, weak rocks or soil like rock masses. Discontinuum models are best suited for blocky mediums and are the most commonly applied numerical approach to rock slope analysis. The rock mass is treated as an assemblage of rigid or deformable blocks. Hybrid techniques are being adopted in rock slope analysis. These techniques are a combination of continuum and discontinuum codes. The propagation of cracks through the finite element mesh can be simulated (Stead et al. 2001). Many authors (Benko and Stead 1998; Bovis and Stewart 1998; Eberhardt et al. 2004; Esaki et al. 1999; Nichol et al. 2002; Stead and Eberhardt 1997) have studied individual cases of rock slope failure with the continuum code F L A C (Itasca 2002) and the discontiuum code UDEC (Itasca 2000). The numerical modeling procedure for various scenarios is recommended by several other authors (Hart 1993; Itasca 2000; Itasca 2002; Ranjith and Saravanan 2002). Failure Prediction Terzaghi (1950) claims that the only slides that occur without warning are caused by earthquakes and spontaneous liquefaction, "all others are preceded by.. .progressive deformation...(And) i f a landslide comes as a surprise to the eyewitness, it would be more accurate to say that the observer failed to detect the phenomena that preceded the slide." Surface movements before a slide include downhill creep and movement along joints/faults, 7 both of which may contribute to the formation of tension cracks along the upper boundary of the slide area. The rate of displacement accelerates from a nearly constant creep to a more rapid sliding. The rate of acceleration from creep to slide motion depends on the thickness of the sliding surface/shear zone. Slides with thin shear zones accelerate very quickly (i.e., brittle), while those with thicker shear zones tend to accelerate more slowly (i.e., ductile). Glawe & Lotter (1996) describe methods for time prediction of rock slope failures based on monitoring results, including geotechnical investigations, displacement monitoring, limit equilibrium analysis, and seismic monitoring. They also observe that a period of steady state movement of the slide mass is followed by an accelerated phase of movement before failure. They note that sensitivity to external influences (such as water supply) increases before failure, and caution that significantly decreasing rates of displacement do not necessarily indicate stability. One must make a comparison of short- medium-, and long-term velocities. Zvelebil (1984) gives an example of a toppling failure that was successfully predicted based on crack extensometer data. The critical displacement to cause toppling was calculated, and the displacement versus time curve was extrapolated to predict a time of failure where the curves intersect. The actual failure occurred only seven days after predicted. Fukuzono (1985) proposed a simple method of predicting the time of failure using the reciprocal of mean velocity. Rose and Hungr (2006) successfully applied the Fukuzono method to the prediction of three large slope failures in open pit mines. The predictions were forecasted two weeks to three months prior to failure. 1.4.2 Rock Mass Characterization Reliable estimates of rock mass strength and deformation characteristic are necessary for nearly all stability analysis procedures. Marinos et al. (2005) describe the applications and limitations of the Geological Strength Index (GSI). The GSI is a method for obtaining estimates of the strength of jointed rock masses, based upon an assessment of the interlocking of rock blocks and the condition of the surfaces between these blocks. It is used to reduce the Hoek-Brown Failure Criterion (Hoek et al. 2002) material constants from 8 intact, laboratory values to appropriate in situ values (Marinos and Hoek 200,0). Rocscience has created software called RocLab for determining rock mass strength parameters, based on the latest version of the generalized Hoek-Brown failure criterion (Rocscience 2005). Barton (1976) describes the shear strength of rock joints based on empirical data. A list of basic friction angles is compiled for various rock types. The actual friction angle for a joint depends on the rock type and the degree of roughness. Kulhawy (1975) presents the results of an extensive literature survey on the stress deformation properties of rock materials and rock discontinuities. Typical values of density, porosity, cohesion, friction angle, and more are tabulated for different rock types. The results are for samples tested under uniaxial and triaxial conditions. Romana Ruiz (2002), and Hoek and Diederichs (2006) compare the many methods for determining the deformation modulus of rock masses. The current practice is to estimate the deformation modulus according to one of several empirical formulations. Hoek and Diedrichs propose a new relationship based on a large number of in situ measurements from China and Taiwan. The (US) National Research Council committee on fracture characterization and fluid flow (NRC 1996) describe investigation methods (geologic and geophysical) for characterizing rock fracture patterns and properties at the surface and at depth. Current understanding of how fluids travel through the fracture network and how these fluids affect the stress in the rock mass is also described. Berkowitz (2002) presents a short review of characterizing flow in fractured rock. Roberts and Poropat (2000) give an example of using 3-D spatial data, digital images, and software to create detailed structural maps of a rock slope. These tools significantly reduce the cost and risk associated with completing a traditional joint survey. Several authors, including (Alba et al. 2005; Pringle et al. 2004; Rosser et al. 2005; Rowlands et al. 2003) show that Light Detection and Ranging (LiDAR) is a key technological development that improves our capacity to collect reliable rock mass and landslide data. Several researchers, including (Kemeny et al. 2006; Kemeny et al. 2004; Monte 2004; Slob et al. 2005), have been exploring and developing 3-D laser mapping methodologies specifically for rock mass characterization and discontinuity analysis. Kemeny et al. (2006) tested the accuracy of joint orientation measurements made digitally from LiDAR-derived point clouds. They showed that errors are associated with 1) the instrument, 2) the procedures for scanning 9 in the field, and 3) processing the resulting point clouds. They claimed that instrument errors are typically less than 1.5 cm, and that processing errors for discontinuity dip and dip direction are less than 0.5 degrees. In a simple case study that compared LiDAR-derived measurements with hand measurements using a Brunton compass, the L i D A R results were within 2-4 degrees of the hand measurement results. 1.4.3 Runout Predict ion of Failed S lopes Empirical Methods The 'angle of reach' (fahrboschung) concept for a rock avalanche was first described by Heim (1932). The 'angle of reach' is a line that connects the top edge of the source area to the distal edge of the deposit. Heim observed that the angle of reach decreases with increased volume of the falling mass. Several other authors further investigate this relationship (Corominas 1996; Evans and Hungr 1993; Hsu 1975; Hungr 1990; Legros 2002; Li 1983; Scheidegger 1973; Voight et al. 1983). They suggest causes for the correlation and causes of scattering in the relationship. Scheidegger (1973) relates the volume of landslides to the coefficient of friction. This correlation can be used to calculate the expected reach and velocity of an imminent slide. L i (1983) presents the statistical analysis of 76 major rockfalls in the Alps. He shows that a correlation exists between the rockfall volume and area covered by the slide, and between the volume and fahrboschung angle. Hungr (1990) also shows that the area of a landslide deposit correlates with the volume: Area = Volume (- 2 / 3\ Hungr also provides a summary of the mechanism of movement proposed by previous authors. Corominas (1996) summarizes the volume threshold observed in landslide mobility by several other authors. He points out that since there is a lack of agreement among researchers, a direct inference from the plots of volume versus fahrboschung can not be made. He claims that the scattering in the relationship is mostly due to mechanisms of motion and to obstacles and topographic constraints on the path. Only slides that have similar composition and follow a similar flow path should be compared. The decrease in the reach angle with volume suggests that scale effects should be taken into account. He also shows that the height of fall has no control over the angle "of reach and notes that some 10 landslides did not occur as a single event, but as several events over a longer time frame. Therefore the total volume can not be considered equivalent to a single event because the location of the source zone and fahrboschung angle changes. Legros (2002) summarizes the mechanisms that have been proposed by other authors to explain the long runout of landslides. The central idea developed is that the apparent coefficient of friction (tangent of the fahrboschung angle) is physically meaningless. It is proposed that the runout distance depends primarily on the volume and not on the fall height, which just adds scatter to the correlation. He points out that there is considerable error in the estimation of thickness (and therefore volume) of a landslide. Data from 203 subaerial, submarine, Martian slides, and debris flows are used in the analysis. Following the example of Lied (1977), Evans and Hungr (1993) propose the use of the 'shadow angle', defined as the angle from the top of talus cone to the distal limit of rockfall, for rockfall investigations. They claim that the shadow angle is preferable to the use of the fahrboschung because it does not necessitate the start and end point of each rockfall to be located. Analytical Methods Numerical models for the dynamic analysis of rapid landslides, D A N (Hungr 1995) and DAN3D (McDougall and Hungr 2004), have been developed. These programs are used for risk assessment and design of remedial measures against rapid landslides such as debris flows and rock avalanches. Each model is based on a Lagrangian solution of the equations of motion and allows for a variety of material rheologies including plastic, frictional, viscous, Bingham and Voellmy that can vary along the slide path or within the moving mass. The approach is semi-empirical in that the complex, heterogeneous fluid is replaced by an equivalent, homogeneous fluid whose bulk properties approximate the behavior of the moving mass. A typical procedure is to calibrate the model by back analysis of known cases and to predict the behavior of new events as required (Hungr 1995). D A N is pseudo-3D; the surface width along the slide path must be assumed beforehand as an input function (Hungr 1995). DAN3D has the ability to simulate rapid landslide motion across complex 3-D terrain in which the material spreads, contracts, changes direction, splits or joins in response to local topography. The model accepts spatial 11 input in the form of user-created grid files specifying bed elevation, source landslide depth, erosion depth, and basal rheology. It has the ability to account for complex anisotropic internal stress states, material entrainment, and rheology variations within the slide mass and along the slide path (McDougall and Hungr 2004). 12 2 P R O J E C T S E T T I N G : T h e A f t e r n o o n C r e e k R o c k s l i d e 1 Abstract A series of mass wasting events occurred above a Washington, U S A highway in the Cascade Mountains in November and December, 2003. The largest event was a rockslide involving approximately 750,000 m 3 that occurred on November 9, 2003. The source zone for this event was located at the crest of a ridge. Most of the debris fell to the east of the sharp ridge and was deposited in the relatively shallow sloping Afternoon Creek without causing damage to the highway. Lesser amounts of debris fell to the west of the ridge, falling 600 meters down the steeper Falls Creek and impacting the road. There is evidence of one or more historical rock avalanches at this location. Displacement of reference points, ground vibration, crack extension, and tilting are being monitored due to concerns that future slope failures or remobilization of debris might again damage or block the highway. 2.1 Introduction On November 9, 2003, a rockslide occurred above Washington State Route 20 (SR 20) near Newhalem, Washington, USA. Rock avalanche debris fell more than 600 meters in elevation down Falls Creek, Falls Creek Chute and Afternoon Creek (Figure 2.1). Most of the debris (approximately 750,000 m3) fell to the east of a sharp ridge and was deposited in the relatively shallow sloping Afternoon Creek without causing damage to the highway. Lesser amounts of debris fell to the west of the ridge down the steeper Falls Creek and Falls Creek Chute. Much of this material traveled all the way to SR 20, destroying portions of the roadway and guardrail, and depositing up to 4-meter diameter boulders on the road. This rockslide was followed by a series of smaller mass wasting events in November and December, 2003. 1. This chapter has been published in a modified form as the peer-reviewed paper: Strouth, A. , Burk, R.L. , Eberhardt, E. 2006. The Afternoon Creek rockslide near Newhalem, Washington. Landslides, 3(2): 175-179. 13 Figure 2.1 Afternoon Creek rockslide above SR. 20 near Newhalem, W A . Photograph provided by John Scurlock, Concrete, W A . The slope is composed of orthogneiss of the Skagit Gneiss Complex, and it appears that the instigating factors underlying the rock slope hazard are glacial over-steepening of the slope, multiple crosscutting faults and fractures, and decreased rock mass strength due to weathering. The November 9th event was triggered by elevated groundwater conditions created by rainfall events in October, 2003 (URS and Wyllie & Norrish Rock Engineers 2004). This chapter presents the results of a field investigation conducted immediately following the Afternoon Creek rockslide, including a description of the geological setting, history of landslide activity in the area, and its implications with respect to future rockslide hazards that threaten the highway. Details of the ongoing monitoring and hazard mitigation work are also discussed. 14 2.2 Geological and Geomorphological Setting The source zone for the Afternoon Creek rockslide is located 600 meters above SR 20, the northernmost route through the Cascade mountain range in the United States. Afternoon Creek is located in the heart of these exceptionally steep and rugged mountains (Figure 2.2); the peaks near the Afternoon Creek rockslide have nearly 1600 meters of vertical relief. Figure 2.2 Location of Afternoon Creek rockslide. Three kilometers east of Newhalem, Washington, USA. Failure area outlined in yellow. The North Cascades is a complex region of accreted Mesozoic and Paleozoic terrains that were assembled during the early to middle Cretaceous. The geology of the region is further complicated by late Cretaceous through Eocene thrusting, plutonism, regional metamorphism, strike-slip faulting, extensional faulting and basin development. Quaternary glaciation created the chiseled peaks and open parabolic-shaped valleys that exist today. The most recent Cordilleran glaciation covered the area 15,000 years ago with a continuous ice 15 cap (Tabor et al. 2003). The Afternoon Creek slope is over-steepened as a result of this most recent glaciation. A l l of the rocks involved in the rockslide are hornblende-biotite tonalite orthogneiss of the Skagit Gneiss Complex (Tabor et al. 2003 map). The tonalite is most likely intrusive igneous material with original igneous crystallization from the late Cretaceous to early Paleocene. The cause of metamorphism is still under debate, but it is agreed that the mechanism is related to some sort of crustal thickening and that ductile deformation ceased by the early Oligocene (Tabor et al. 2003). The average uniaxial compressive strength (UCS) of the intact orthogneiss is approximately 90 MPa according to UCS tests performed by GeoTest Unlimited, and from point load correlation and field observations (URS and Wyllie & Norrish Rock Engineers 2004). This corresponds to 'R4-strong rock' according to Brown (1981). Multiple cross cutting fractures and faults divide the slope into several structural zones. The rock mass in each zone ranges from 'disintegrated' with 'fair' surface conditions to 'very blocky' with 'good' surface conditions corresponding to a Geological Strength Index ranging from 30 to 60 (Marinos and Hoek 2000). Dozens of northeast-southwest trending fracture lineaments that are visible on a regional scale cut across the Afternoon Creek rock slope; additionally, locally persistent fractures trend northwest-southeast, parallel to Afternoon Creek. Near the failed slope, many of these discontinuities are filled with soil and rock rubble debris. Two important joint sets were apparent in the preliminary investigation. The most common set (plane A) dips parallel to the larger Skagit valley (i.e., strikes perpendicular to the Afternoon Creek slope). The second set is sub-vertical and parallel to Afternoon Creek (plane B). Plane B joints are widely spaced and highly persistent. Some are open and filled with soil and rock rubble (Figure 2.3). 16 Figure 2.3 Sub-vertical, open fractures parallel to Afternoon Creek. Photograph provided by URS Corporation (photograph date, March 1, 2004). 2.3 History of Activity Debris flows and snow avalanches are common to both Afternoon Creek and Falls Creek and have forced road closure on several occasions in the past (e.g. Figure 2.4). Although the November 9, 2003 event was the first large rock avalanche on record, undated photographs and aerial photos from 1998 indicate ongoing rockfall activity several years before the event. Significant rockfall scars near the top of the slope and fresh boulder-sized debris in Afternoon Creek can be seen in undated oblique photographs (Figure 2.5). Rockfall scars of the same magnitude can be seen in aerial photos taken in 1998. Some of the rock rubble that is now covered by vegetation (Figure 2.5) may be debris from a single rock avalanche, or a series of historical rock avalanches that have occurred since deglaciation (late Pleistocene). This evidence indicates that slope instabilities in the form of isolated rockfall and larger rock avalanches are common in Afternoon Creek since deglaciation. 17 Figure 2.4 Historical photographs of debris flows that reached the state route 20 highway, (a) Afternoon Creek debris flow material (March 23, 1949). (b) Falls Creek Chute debris flow covered with fresh snow (1990). (c) Afternoon Creek (1999 or 2000). Photographs provided by WSDOT. Figure 2.5 Afternoon Creek rockslide slope before the November 9, 2003 event. Photograph provided by WSDOT (date unknown). 2.4 Chronology of Recent Mass Wasting Events The November 9th Afternoon Creek rockslide, and subsequent smaller mass wasting events that occurred through November and December, 2003, were preceded by record rainfall in October. The week of October 16 through 21 was one of the wettest weeks in western Washington history; a rain gauge just 6.5 km from the landslide measured more than 18 400 mm of rain in these six days. The soil and rock rubble filled fractures that cut across the Afternoon Creek slope readily allow surface runoff to enter the fracture system. This water backed up against hydraulic barriers such as clay-rich shear planes, soil and possibly ice, creating high pore-water pressures. It is likely that the increased pore water pressures in the slope due to the influx of water in mid-October triggered the initial collapse of the slope. The failure mechanism of the November 9th rockslide is complex. A preliminary interpretation is that the initial collapse occurred in the southern half of the failure zone where the rock mass is most highly fractured and dilated. This material collapsed toward Afternoon Creek. The event unloaded the toe of larger, more competent blocks that slid on plane A, initially in a direction parallel/oblique to Afternoon Creek (Figure 2.6). Rockfall and/or toppling in the upper portion of the failure zone followed after the loss of lateral support provided by the large blocks. Plane B (parallel to Afternoon Creek) allowed toppling and provided the lateral release necessary for planar sliding. Given the geometry of the ridge, a small percentage of the total rockslide material (<10%) traveled down the west side of the ridge along Falls Creek and Falls Creek Chute impacting the highway below (Figure 2.7). Figure 2.6 Afternoon Creek failure scarp from Afternoon Creek, (photograph date April , 2005). 19 Figure 2.7 Digital elevation model of the Afternoon Creek rockslide. Failed mass is shown in gray. Arrows indicate direction of movement. Dotted blue line indicates the extent of path. This initial slope failure was followed by several smaller events during the following weeks. Heavy rain a week later washed smaller rock and debris out of Falls Creek, and on November 19, 2003, a debris flow (approximately 35,000 - 60,000 m 3 in size) came down Falls Creek Chute and closed the road. Increased rockfall activity in Afternoon Creek was observed beginning in early December and leading up to December 19th. On December 19, 2003, a rockfall of approximately 35,000 m 3 in size occurred. This event deposited boulders up to 15-meter in diameter in Afternoon Creek. A series of smaller rockfall events continued during the following week. Large scale fractures were recognized in the intact bedrock near the failed face following these events. On several occasions workers heard loud explosion-like booms and felt the ground vibrate beneath their feet. Although the sounds emanated from the slide area, they were not followed by rolling rocks or dust clouds. These events may have been the result of intact brittle rock fracturing in the slope. 2.5 Continued Monitoring, Hazard Mitigation, and Future Work After the November 9th collapse, there were concerns that additional slope failures and remobilization of the rock avalanche debris might further damage or block the highway (SR 20). The Washington State Department of Transportation (WSDOT) contracted with 20 URS Corporation and Wyllie & Norrish Rock Engineers to begin a monitoring and investigation program. As part of this investigation, four monitoring techniques were employed to locate areas of ongoing slope movement, measure movement vectors and enable prediction of future slope failures: 1. Displacement monitoring consisted of regularly surveying the location of reference points to track cumulative movement of those points. Fifteen geodetic prisms were placed near or within the area of slope failure as reference points. 2. Vibration monitoring was used to correlate ground vibrations with rockfall activity triggered by precipitation events. Geophones were buried adjacent to Afternoon Creek, in Falls Creek Chute and along the upper head scarp of the failure plane. 3. Displacement monitoring was used to monitor crack dilation to assist in predicting potential rock failure. Two extensometers were installed across cracks within intact bedrock just above the upper head scarp of the failure. 4. Rock tilt monitoring was used to assist in interpreting i f further head scarp development was occurring. Two tiltmeters were installed on intact bedrock faces near the upper head scarp of the failure. The tiltmeters were used to monitor rotation of the bedrock adjacent to the area of rock failure. Three of the geodetic prisms in the upper slope were destroyed by rockfall and have not been replaced. One of these was lost in the rockfall event of December 19, 2003. Based on the survey data, this prism accelerated to failure, traveling more than 2.5 m prior to failure in a period of 27 days. The remaining prisms in the upper and lower slope show no measurable movements within the error limits of the survey method (URS and Wyllie & Norrish Rock Engineers 2004b). Geophone background values typically increase slightly during precipitation events. The majority of peak vibrations are two to four times larger than background values. The rockfall event of December 19, 2003 had a peak vibration 35 times larger than typical background values. There is a correlation between rainfall and increased event activity. The 21 extensometer and tiltmeter data show diurnal changes, however, no long-term trends related to movement of the rock has been observed (URS and Wyllie & Norrish Rock Engineers 2004b). 2.6 Conclusions A series of mass wasting events occurred at Afternoon Creek in November and December, 2003. The largest event was a 750,000 m 3 rockslide on November 9th, originating near the top of a sharp ridge. Most of this volume of large boulder debris landed in the Afternoon Creek without causing damage to the highway; however a very small portion of the material traveled down the backside of the ridge and impacted the Washington SR 20 roadway. Glacial over-steepening of the slope, multiple cross-cutting fractures and shear zones, and decreased rock mass strength due to weathering were key factors in conditioning the slope for failure. Heavy precipitation leading to high joint water conditions triggered the rockslide. The slope is currently being monitored with a regular survey of reference points, geophones, crack extensometers, and tiltmeters. 22 3 M E T H O D O L O G Y 3.1 Total Slope Analysis Procedure A procedure that combines several numerical techniques, linking failure initiation processes to runout, termed "Total Slope Failure Analysis" by Stead et al. (2006), was followed to characterize the past slope failure, and hazard that currently exists at Afternoon Creek. Stead et al. point out that the traditional engineering approach is to analyze either the failure initiation mechanism or the transport/deposition stage. They suggest, however, that if true risk is to be ascertained then the deformation characteristics prior to failure and the post-failure movement must be linked. In this thesis, traditional methods, including kinematic analysis and limit equilibrium analysis, were combined with advanced numerical methods, including the discontinuum numerical codes, UDEC (Itasca 2000) and 3DEC (Itasca 2003), to assess the failure initiation process. The results of the assessment provided understanding of the slope deformation and failure mechanisms, the failure volume, and helped to reduce uncertainty regarding the physical properties of an unstable rock mass. These guided the post-failure motion analysis. Post-failure motion, including the runout path, runout distance, and velocity, were analyzed with the dynamic/rheological flow code DAN3D (McDougall and Hungr 2004). The Total Slope Analysis procedure was followed to back analyze the November, 2003 event, and then repeated as a forward analysis of the current slope configuration. The Total Slope Analysis procedure is schematically illustrated in Figure 3.1. 23 fir (A C si O U CO Past s lope fai lure Mechanism Discontinuity parameters Rock mass parameters Runout Rheology Base friction angle Internal friction angle '55 Q . ( o O c 1200 32 very poor 5/5 range test 326 14 10 > 1500 40 no data 6/1 fai lure scarp -zone 3 203 20 16 250 11 good 7/1 fa i lure scarp -zone B, zone 3 214 11 20 150 8 good 7/2 fa i lure scarp -zone 3 256 27 20 100 5 good *This is the estimated line-of-sight distance from the scanner station to the center of the target ** (-) tilt indicates that the line-of-sight is below horizontal For useful, good quality point clouds, the strike-line of the target area should be approximately perpendicular to the line-of-sight of the scanner, and the target should be within the maximum operating range of the instrument which is a function of the atmospheric visibility and target reflectivity (typically 800 meters for a target with 20% reflectivity using the Optech ILRIS3D laser). Several of the scans completed for the Afternoon Creek survey resulted in 'very poor' quality point cloud data, specifically those used to test the limitations of the instrument (Table 3.2). A poor quality point cloud means that few spot reflections were received by the instrument; therefore the resolution of the resulting point cloud was so coarse that no useful joint orientation data can be extracted. 32 The laser scanner is a line-of-sight instrument, meaning that the position of the first object in the light's path is recorded. Therefore small (and large) variations in topography, and vegetation create shadows - areas of the topography where no data is collected - in the resulting point cloud. Additionally, surfaces that strike sub-parallel to the line-of-sight of the scanner tend to reflect few laser strikes; therefore joint sets that strike parallel are poorly sampled, while sets that strike perpendicular are well sampled, introducing a bias to the final data. This point is illustrated by Figure 3.6, showing rose diagrams of automatically-generated patches found in four different point clouds of the same target (structural Zone 3). Southeast dipping surfaces were preferentially recognized in the scan from survey station 2 because this joint set was orthogonal to the scanner line-of-sight. Northeast dipping surfaces were preferentially recognized in the scans from survey stations 6 and 7. This bias can be removed by scanning the target from all possible angles and analyzing all of the scans (either individually or by aligning the scans into a single point cloud). 0 0 0 0 Figure 3.6 Histograms of dip direction vs. frequency for four scans of structural Zone 3 showing a bias in the orientation of automatically generated patches. Joint planes that strike perpendicular to the scanner position tend to reflect more laser strikes. Line-of-sight of the scanner is superimposed over the histogram. Ideally, the entire failure scarp and other areas of interest would have been scanned at a similar resolution from several angles with considerable overlap; this would have enabled the different point clouds to be joined, creating a single 3-D model of the entire slope. A single, continuous point cloud would have made the data analysis simpler, and more objective. However, in this case, overlapping data from all portions of the failure scarp was not achieved due to access limitations in the narrow Afternoon Creek. 33 Good quality point clouds for the failure scarp were collected from Stations 2, 6, and 7; however Stations 6 and 7 were too close to the failure scarp to allow for the entire zone to be captured, and therefore the resulting point clouds do not overlap. The scans from Stations 6 and 7 do however provide data for shadowed portions of the point cloud from the Station 2 scan. Another limitation imposed by the narrow confines of the Afternoon Creek channel was that the scanner was steeply inclined when directed towards the base of the failure scarp; 'false' summits in the failure scarp created a complete shadow of the upper portions of the failure zone. This shadow could have been removed by performing a scan midway up the valley wall opposite the failure scarp; however no safe access path or setup point could be found. 3.3.2 L iDAR Data P rocess ing Split-FX™ software, developed by Split Engineering L L C , was used to visualize the 3-D point clouds, and extract the orientation, spacing, and persistence of dominant discontinuity sets. Because it was not possible to create a single, comprehensive point cloud of the entire slope, each scan's point cloud was analyzed separately. Nine scans of various portions of the failure scarp were completed; however only four scans were considered to be of sufficient quality to be incorporated into the analysis: 2/1, 6/1, 7/1, 7/2 (see Table 3.2; Appendix B). The remaining five scans were not used because they were either 'very poor' quality or coarser resolution copies of one of the included scans. Each of the four point clouds was processed and analyzed in the same way using the following procedure (Figure 3.7): 1. Orient the point cloud. Input the true bearing of the scanner's line of sight and components of tilt (as measured in the field at the time of scanning). 2. Edit the point cloud. Remove points that are not related to joint orientation (e.g. reflections from vegetation, talus, soil-cover, etc.). 3. Create a mesh. The Split-FX™ software drapes a polygonal surface mesh over the point cloud. The analyst decides the mesh grid size, which controls the size of the cells, the number of points per cell, and the precision of the polygonal surface model (Split_Engineering 2005). 34 4. Automatic patch generation. Patches are planes fit to the discontinuity surfaces present in the point cloud. Patches are found first by grouping neighboring mesh triangles together based on the similarity of their vector normals, and then by using least squares to fit a plane through the points bounded by the grouped triangles (SplitEngineering 2005). User controls include the minimum patch size and minimum neighbor angle, which are tolerance parameters used to group neighboring mesh triangles. 5. Edit patches. The analyst visually inspects the patches and deletes erroneous patches and adds missing patches, i f necessary. 6. Stereonet Analysis. The patch orientation, size, and roughness are recorded by the software. These can be exported to any stereonet analysis package or analyzed with the Split-FX™ stereonet software. mesh mesh with patches Figure 3.7 Point cloud processing procedure A bias is introduced to the final data set through discontinuity surfaces that are small relative to the point cloud resolution and mesh size; these are not sampled by the method described above. This biasing is unavoidable, although the influence of the small, unsampled discontinuities can still be accounted for through the designation of rock mass quality. 35 3.3.3 L iDAR Data Ana lys i s To determine the influence of the mesh density and patch control parameters on the processing accuracy, with respect to the true joint surfaces measured, a qualitative parametric study was conducted. The mesh density recommended by the software manual yields approximately 30 points per mesh grid cell. The results of the parametric study showed that this was an appropriate point density for the Afternoon Creek slope. A coarser mesh density did not capture important (small) features of the rock slope, whereas a finer mesh did not add any features of significance to the data set. Also the finer mesh resulted in more patches that were smaller in size, that in turn were more difficult to visually inspect. The minimum patch size parameter is used to filter out small patches that are difficult to visually inspect and which add significant noise to the stereonets. This parameter was typically set to a value between 10 and 40 grid cells per patch. The minimum neighbor angle determines which grid cells are included in the patch. A relatively large value allows adjacent grid cells of an undulating, rough discontinuity to be grouped into a single patch at the average grid cell orientation. This parameter was typically set to a value between 4° and 8°. A value less than 4° typically yielded very few patches, and excluded surfaces in the point cloud that were obvious joint surfaces. A value greater that 8° typically yielded numerous erroneous patches that were removed during visual inspection (Figure 3.8). When a relatively fine mesh density was used the minimum patch size and minimum neighbor angle were typically set to the upper end of the described ranges. 36 2/1 Area !