Why the 2002 Denali fault rupture propagated onto the Totschunda fault: Implications for fault branching and seismic hazards

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2011jb008918, 2012 Why the 2002 Denali fault rupture propagated onto the Totschunda fault: Implications for fault branching and seismic hazards David P. Schwartz, 1 Peter J. Haeussler, 2 Gordon G. Seitz, 3 and Timothy E. Dawson 3 Received 21 October 2011; revised 12 September 2012; accepted 27 September 2012; published 15 November [1] The propagation of the rupture of the M w 7.9 Denali fault earthquake from the central Denali fault onto the Totschunda fault has provided a basis for dynamic models of fault branching in which the angle of the regional or local prestress relative to the orientation of the main fault and branch plays a principal role in determining which fault branch is taken. GeoEarthScope LiDAR and paleoseismic data allow us to map the structure of the Denali-Totschunda fault intersection and evaluate controls of fault branching from a geological perspective. LiDAR data reveal the Denali-Totschunda fault intersection is structurally simple with the two faults directly connected. At the branch point, km east of the 2002 epicenter, the 2002 rupture diverges southeast to become the Totschunda fault. We use paleoseismic data to propose that differences in the accumulated strain on each fault segment, which express differences in the elapsed time since the most recent event, was one important control of the branching direction. We suggest that data on event history, slip rate, paleo offsets, fault geometry and structure, and connectivity, especially on high slip rate-short recurrence interval faults, can be used to assess the likelihood of branching and its direction. Analysis of the Denali-Totschunda fault intersection has implications for evaluating the potential for a rupture to propagate across other types of fault intersections and for characterizing sources of future large earthquakes. Citation: Schwartz, D. P., P. J. Haeussler, G. G. Seitz, and T. E. Dawson (2012), Why the 2002 Denali fault rupture propagated onto the Totschunda fault: Implications for fault branching and seismic hazards, J. Geophys. Res., 117,, doi: /2011jb Introduction [2] In 1916 the poet Robert Frost published The Road Not Taken [Frost, 1916], a commentary, in part, on choice. It begins with Two roads diverged in a yellow wood, And sorry I could not travel both and concludes with I took the one less traveled by, And that has made all the difference. Ninety-six years later three sections of the Denali fault system ruptured on November 3, 2002 to produce the M7.9 Denali fault earthquake in central Alaska. The rupture (Figure 1), which is well documented with seismologic [Ratchkovski and Hansen, 2002], geodetic [Hreinsdottir et al., 2006], and geologic [Eberhart-Phillips et al., 2003; Haeussler et al., 2004] observations, initiated on the Susitna Glacier thrust fault near its intersection at depth with the Denali strike-slip fault and propagated onto the central Denali fault. Approximately 225 km east of the epicenter the rupture reached the intersection with the Totschunda fault. 1 U.S. Geological Survey, Menlo Park, California, USA. 2 U.S. Geological Survey, Anchorage, Alaska, USA. 3 California Geological Survey, Menlo Park, California, USA. Corresponding author: D. P. Schwartz, U.S. Geological Survey, 345 Middlefield Rd., Menlo Park, CA 94025, USA. (dschwartz@usgs.gov) American Geophysical Union. All Rights Reserved /12/2011JB Instead of continuing linearly eastward along the Denali fault, which is the most direct route, the rupture turned southeast and propagated onto the lower slip rate Totschunda fault for an additional 70 km. Just as Frost s traveler made a decision on which branch in the road to take, so also do earthquake ruptures choose a path when the fault trace branches. This paper addresses the issue of why the 2002 rupture took the path it did when it arrived at this intersection. [3] The branching of the 2002 rupture onto the Totschunda fault is one of the intriguing aspects of this great earthquake. The past decade has seen an increased interest in the dynamics and controls of fault rupture, and the 2002 surface faulting has served as the primary example in several dynamic models of fault branching. Bhat et al. [2004] develop rupture simulations and conclude that fault geometry, principally the orientation of the maximum compressive regional stress relative to the fault traces, favored rupture of the Totschunda fault. Dreger et al. [2004] conclude that the rupture transferred to the Totschunda fault because it had a more favorable orientation relative to the regional stress field, coupled with the dynamic effect that slip on the Totschunda fault placed the Denali fault east of the intersection in a stress shadow. Oglesby et al. [2004], using inverse kinematic and 3D forward dynamic modeling, conclude that a combination of a more favorable orientation to the local stress field and dynamic changes in shear and 1of25

2 Figure 1. Southern Alaska neotectonic map, showing fault traces active in Neogene time, modified from Plafker et al. [1994] and Haeussler [2008]. Map base is a shaded relief map. Lines are faults with Quaternary or late Neogene activity. Arrows show sense of strike-slip faulting and teeth are direction of dip of thrust faults. WDF, western Denali fault; CDF, central Denali fault; EDF, eastern Denali fault; TF, Totschunda fault; SGF, Susitna Glacier fault; DRF, Duke River fault; CMF, Castle Mountain fault; BBF, Bruin Bay fault; LCF, Lake Clark fault; TRF, Transition Fault; AMT, Aleutian megathrust; KIZ, Kayak Island deformation zone; CSF, Chugach-St. Elias Fault. Star is epicenter of 2002 earthquake. Slip rates on the Denali and Totschunda faults are shown in mm/yr [Matmon et al., 2006; Meriaux et al., 2009]. Red line is extent of 2002 surface rupture. Black triangles are volcanoes with Holocene activity. Long arrows show relative plate or block motions: PAC-NA, Pacific-North America plate motion [Argus et al., 2010]; YAK-NA, Yakutat microplate - North America relative plate motion [Elliott et al., 2010]. Thin black lines are roads, which are major geographic features in Alaska. normal stress favored rupture onto the Totschunda fault. From another perspective, Field et al. [2009] cite the 2002 rupture as an example of a fault-to-fault jump. Based on this they suggest that longer ruptures involving linking of separate faults should be considered in seismic hazard models for California. [4] This paper presents an interpretation of why the 2002 rupture propagated from the Denali onto the Totschunda fault that is based primarily on: (1) reinterpreted structure of the Denali-Totschunda fault intersection from recently acquired LiDAR data; and (2) new information on the timing of paleoearthquakes on each of these faults. In light of these observations we review issues concerning rupture propagation at branches and step-overs, expand the discussion to compare the geometry of the Denali-Totschunda fault intersection to those on other high-slip-rate strike-slip faults, and consider possible implications of the 2002 branching for seismic hazard estimates. 2. The Denali-Totschunda Fault Intersection [5] GeoEarthScope LiDAR data for sections of the Denali and Totschunda faults ( gridsphere?cid=datasets) are a principal source of new information. They are the basis for developing a detailed interpretation of the geometric and structural relations at this intersection and of the complex zone of transtensional faulting (the transfer zone, Figures 2 and 3) across which 2002 slip propagated to the linear section of the Totschunda fault. The LiDAR data also provide a structural framework for 2 of 25

3 Figure 2. Locations of paleoseismic and slip-rate sites and 2002 surface slip distribution across the Denali-Totschunda fault intersection. Circles are paleoseismic sites: M, Mentasta; TSP, Totschunda Sag Pond; PB, Pooh Bear. Squares are slip rate localities [Matmon et al., 2006]. Small numbers with ticks across the faults are reference kilometers from the epicenter [Haeussler et al., 2004] slip values (meters) are from Haeussler et al. [2004] with the exception of the Denali-Totschunda fault intersection and Totschunda transfer zone where revised displacements (with Haeussler et al. [2004] measurements in italics) are shown. revaluating measurements of 2002 surface offset along the western Totschunda fault. Paleoseismic observations from hand-excavated exposures yield estimates of the timing of the penultimate surface rupture on the central Denali and Totschunda faults and of the most recent event on the eastern Denali fault Fault Geometry at the Intersection [6] In the descriptions that follow we distinguish two terms, intersection and branch point. An intersection is a volume of crust that contains the principal faults (as well as secondary faults) as distinct structures at the surface. A branch point is the location on the surface where the trends of the principal fault traces diverge. In this usage the Denali-Totschunda intersection is 1.7 km long, the surface separation distance between the two faults is a maximum of 90 m, and the faults connect at the surface at the western end (Figure 4). [7] Four primary structural elements define the Denali- Totschunda fault geometrical relationship at the intersection (Figures 1 and 3). These are: (1) the central Denali fault, which slipped in 2002 and has a regional azimuth of 122 ; (2) the eastern Denali fault, which was not involved in the 2002 rupture and has a regional azimuth of 124 ; (3) a complex 22-km-long long reach of the Totschunda fault southeast of the branch point that connects the Denali fault to the linear section of the Totschunda fault and has a general azimuth of 152 ; and (4) the 50-km-long linear section of the Totschunda fault that extends to the eastern end of the 2002 rupture with a general azimuth of 136. [8] Prior to the 2002 earthquake the precise structural relation between the Denali and Totschunda faults was not known. The only detailed mapping was that of Richter and Matson [1971]. They identified the principal active traces of both faults from satellite images and aerial photographs and also estimated amounts of right lateral offset of late Pleistocene glacial deposits. They traced the Totschunda fault to within 2.5 km south of, and striking parallel to, the Denali fault at the Little Tok River (Figure 4). [9] Surface rupture in 2002 illuminated some features of the intersection for the first time and clearly showed that the Totschunda and Denali fault are connected at the surface. Haeussler et al. [2004], based on field mapping after the 2002 earthquake and interpretation of post-rupture aerial photographs, identify a reach of the rupture in which slip moves from the central Denali fault to the linear section of the Totschunda fault through an approximately 25 km-long, 8 km-wide releasing bend termed the transfer zone (Figure 3a). As originally defined [Haeussler et al., 2004], the western end of the transfer zone includes the easternmost 3of25

4 Figure 3. Maps of the intersection of the Denali and Totschunda faults. CDF, central Denali fault; EDF, eastern Denali fault; TF, Totschunda fault. Distances, in kilometers, from the 2002 epicenter [Haeussler et al., 2004] are shown. Red lines are the 2002 rupture; blue lines are Holocene fault scarps without 2002 slip. (a) Map of the 2002 rupture and the surface trace of the eastern Denali fault interpreted from aerial photographs [Haeussler et al., 2004; Haeussler, 2009]. (b) Revised map of the 2002 rupture and surface trace of the eastern Denali fault after examination of GeoEarthScope LiDAR data. Additional 2002 rupture traces were identified, as well as additional fault traces that did not rupture in The area of LiDAR coverage is shown in white. 6 km of 2002 offset attributed to the central Denali fault (km 224 to km 230, Figure 3a; we use the notation of Haeussler et al., [2004] to refer to the location of features along the fault trace in km eastward from the 2002 epicenter). Rupture of the Totschunda fault was interpreted to extend directly from the Denali fault starting at about km 230 (Figure 3a). Within the main section of the transfer zone (km 230 km 250, Figure 3a) the 2002 surface faulting, as well as that from prior earthquakes, is expressed as a series of southeast-trending right-lateral strike-slip fault segments ( ) that step progressively to the south along oblique-slip segments ( ) and shorter northerly trending ( ) normal fault segments. The southeast end of the transfer zone is the point where splays of the 2002 rupture join to form a single trace of the Totschunda fault (km 249, Figure 3). Haeussler et al. [2004] did not observe any rupture to the east along the eastern Denali fault, which is geomorphically well defined by scarps, sag ponds, and offset outwash fans, terrace risers, and glacial moraines. The eastern Denali fault maintains a strike of 124 (Figure 3b) for more than 150 km to the U.S.-Canada border. The azimuths of the eastern Denali fault and Totschunda fault, where they sub-parallel each other east of the transfer zone, differ on average by 12 degrees. [10] There are no independent geophysical data on dip at depth for these faults. However, aftershocks from the of25

5 Figure 4 5 of 25

6 earthquake [Ratchkovski et al., 2004] define a vertical plane along the linear reach of the Totschunda fault and the southern half of the transfer zone, with a slightly more complex distribution in the immediate vicinity of the intersection. They also delineate a vertical central Denali fault, which is consistent with its surface expression and the surface trace of the eastern Denali LiDAR and Structure of the Fault Intersection [11] GeoEarthScope LiDAR provides an unprecedented view of the structural relations and geometry at the intersection of the Denali and Totschunda faults (metadata at: opentopo.sdsc.edu/metadata/unavco_july08_yellowstone_ Wasatch_Alaska.pdf, last accessed 5 August, 2011), as well as the rupture complexity in the transfer zone (Figures 4, 5, 6, and 7). Our analysis utilized the bare earth digital elevation model (DEM) supplied by the opentopography.org website. This is a 0.5 m grid. We also tried the filtered point cloud data and found no significant difference in resolving features along the fault trace. This is likely because the Alaska survey was flown at a higher distance from the ground due to local topography. A full map of the central Denali-eastern Denali- Totschunda fault intersection based on the GeoEarthScope Denali fault LiDAR is provided in the auxiliary material. 1 [12] An overview of faulting in the region of the Denali- Totschunda fault intersection is shown as a series of LiDAR images with and without vegetation and structural interpretation (Figure 4). On the northwest edge of Figure 4b, the main trace of the 2002 rupture bends very slightly (3 degrees) south, resulting in formation of a 2-km-long splay fault (S1) with normal-oblique slip, south side up. The splay was mapped from post-earthquake aerial photographs [Haeussler et al., 2004; Haeussler, 2009] and the amount of 2002 offset, while likely small, was not measured in the field. This is the first releasing splay to occur along the eastward-propagating rupture of the central Denali fault and it marks the west end of the transfer zone. 1.7 km further east the primary 2002 rupture bends south an additional 6 degrees with the resulting transtension producing two fault-line graben. At the east end of the eastern graben the rupture turns an additional 12 degrees southward. Just east of this point a distinct linear fault trace without 2002 slip is observed north of the rupture. It trends southeast as a continuous surface feature to become the eastern Denali fault (arrows, Figure 4c). We refer to this location as the initial separation point (ISP). [13] Detail of this geometry is shown on Figure 5. From the ISP to the branch point (BP), a distance of 1.1 km, the 2002 rupture and the sub-parallel northern trace are closely spaced (50 m to 90 m apart) but separate surface structures that 1 Auxiliary materials are available in the HTML. doi: / 2012JB extend across a 170-m-wide right releasing bend. Along this reach the two traces are connected at the surface by NWtrending conjugate shears (Figure 5), a structural configuration commonly observed along closely spaced parallel strikeslip faults. S2 (Figure 4b) is a secondary 2002 trace, 1.2 km long with 2.5 m 3.0 m of right lateral offset [Haeussler et al., 2004]. At the branch point (Figure 5b) the 2002 rupture bends southeast to become the Totschunda fault. At this location the surface traces of the eastern Denali fault and the Totschunda fault are only 55 m apart. The angle between the faults at the branch point is 19. As identified here the branch point is 6.5 km northwest of where Haeussler et al. [2004] concluded that the 2002 rupture initially turned on to the Totschunda fault. Conjugate shears are interpreted from the LiDAR for about another kilometer to the southeast (Figure 5b), as the distance between the two faults increases. From the Little Tok River to the linear section of the Totschunda fault, the general trend of the transfer zone is at an angle 28 degrees clockwise to the eastern Denali fault. [14] Within the transfer zone LiDAR shows the presence of paleoscarps and fissures associated with pre surface ruptures (Figures 3b, 6, and 7). The longest such structure extends for approximately 5 km parallel to, and 1.2 km southwest of, the 2002 rupture (km 229 to km 234, Figure 3b). It is defined by a series of sub-parallel to en-echelon leftstepping shears in a zone about 40-m-wide that is largely coincident with a post-glacial terrace riser of the Little Tok River. The geomorphology along the zone indicates right lateral slip with a small amount of down to the northeast displacement. In the southeast part of the transfer zone the LiDAR allows us to increase the mapped length of two traces of the Totschunda fault (Figures 3b and 7b). The northern extension, which is about 3 km long, has a few discontinuous 2002 surface breaks associated with it (Figure 7b) and, in combination with the main 2002 rupture, defines a parallelogram-shaped section of the fault zone. At its northwestern end this pre-2002 trace extends into late Holocene fluvial deposits where it offsets a paleochannel and two stream terrace risers approximately 10 m (location A, Figure 7b). A larger offset, 75 m to 85 m, is observed in late Pleistocene glacial deposits along this trace to the southeast after it merges with the primary 2002 rupture (location B, Figure 7). The southern extension of pre-2002 surface rupture is at least 3.2 km long, and could extend beyond the boundary of the LiDAR data. Near its mapped northwest end a post-glacial terrace riser is right laterally offset approximately 40 m to 50 m (location C, Figure 7b). A larger cumulative offset of 70 m to 80 m across a glacial channel margin occurs on this trace to the southeast (location D, Figure 7b). Based on the LiDAR data, the transfer zone is a long-lived geologic structure that has had a Figure 4. LiDAR images of Denali -Totschunda fault intersection. (a) Unfiltered first return of the intersection area with 45 sun angle. (b) Same image as Figure 4a, but a bare earth image with 45 sun angle and vegetation removed. (c) Same as Figure 4b, with our interpretation. Red lines are 2002 rupture traces mapped from post-earthquake aerial photographs [Haeussler et al., 2004; Haeussler, 2009]. S1 and S2 are splays of main rupture. ISP is the initial separation point and BP is the branch point, as discussed in the text. Blue arrows point to a linear fault trace separating from the primary 2002 rupture at ISP. This is the continuation of the central Denali fault, and becomes the eastern Denali fault east of the branch point (BP). Immediately southeast of the Little Tok River valley late Pleistocene glacial deposits are right-laterally offset approximately m (dashed yellow-green lines). The section of the 2002 surface rupture between BP and the Little Tok River is the Totschunda fault. 6of25

7 Figure 5. LiDAR image showing detailed structural relations at Denali-Totschunda fault intersection. (a) Uninterpreted bare earth image with 45 sun angle. (b) Red lines are the 2002 surface rupture from Haeussler et al. [2004] and Haeussler [2009]. Blue lines are Holocene faults interpreted from the LiDAR that did not slip in At ISP, the single 2002 rupture trace bends slightly (6 degrees) southeast. From there it is tracked on the north by a geomorphically well-defined linear fault trace as far as the branch point (BP). These are separated at the surface by a distance of 50 m to 90 m. At the branch point the two surface traces are only 55 m apart. Both surface traces are joined in places by northwesttrending Riedel shears. From the branch point eastward the two traces continue separately as the eastern Denali fault and the Totschunda fault. more complex rupture history in the late Pleistocene than is expressed by the 2002 fault trace Slip Distribution Across the Denali-Totschunda Fault Intersection [15] As part of the re-evaluation of the intersection we have looked at the slip distribution across it, which was reported by Haeussler et al. [2004]. From this we have replotted the amount of right lateral surface offset into and across the region of the intersection (Figure 2). The 50 kmlong section of the central Denali fault west of the intersection contains the largest 2002 lateral offsets, reaching 8.8 m. Along the 30 km section immediately west of the intersection offset measurements vary from 4.5 m to 6.9 m and Haeussler et al. [2004] estimated the average slip as 5.7 m to 6.2 m. For the transfer zone as far as km 250 east of the epicenter (Figure 2) they used only the lateral offsets, which range from 0.0 m to 1.3 m [Haeussler et al., 2004, Table 2], to define the Totschunda slip distribution. In reviewing these [Haeussler et al., 2004, Table 2] it is clear that for the linear sections of the rupture the amount of vertical slip is quite small and the lateral measurements are a 7 of 25

8 Figure 6. Structural details in the transfer zone. (a) Uninterpreted bare earth LiDAR image with 45 sun angle, and (b) interpreted image showing details of surface faulting in Totschunda fault transfer zone rupture trace (red) is from field mapping [Haeussler et al., 2004] and aerial photograph interpretation [Haeussler, 2009]. Blue fault traces are surface ruptures observed on LiDAR; they are largely pre-2002 faulting, but some may be 2002 slip. Northeast-trending normal faults accommodate slip across steps between strike-slip and oblique-slip segments of the rupture. White dots are displacement measurement locations, with revised measurements of net slip (meters, this paper) and Haeussler et al. [2004] measurements of lateral slip (meters, in parentheses). reasonable approximation of net surface displacement. This is not the case in the transfer zone where vertical slip is significantly larger than horizontal at a number of measured locations (Figure 8). [16] The LiDAR data clarify field relations where 2002 measurements were made. Figure 6 shows a section of the transfer zone where the rupture changes style from primarily lateral to normal, with extension in the direction of horizontal transport (southeast). At locations where Haeussler et al. [2004] used only the horizontal component of the rupture we have calculated the surface slip from the measured horizontal and vertical components (Figure 8). The resulting slip vector from these is a net vertical displacement. The revised slip values (Figures 2 and 6) may be minima because the normal faults likely have some dip at depth and there are additional extensional ruptures that were not recognized (Figure 6). [17] The revised transfer zone slip estimates modify the shape of the 2002 slip distribution. By including the lateral offsets only, Haeussler et al. [2004] interpret a generally elliptical slip distribution with an average of 1.7 m for the Totschunda rupture as a whole (Figure 9). The revised estimates show that 2002 slip drops rapidly in a step across the branch point from more than 5 m on the west to m 8of25

9 Figure 7. Structural details of southeastern transfer zone. (a) Uninterpreted bare earth LiDAR image with 45 sun angle, and (b) interpreted image showing details of surface faulting in Totschunda fault transfer zone rupture trace (red) is from field mapping [Haeussler et al., 2004] and aerial photograph interpretation [Haeussler, 2009]. Blue fault traces are pre-2002 rupture(s) mapped from LiDAR. A D are locations of cumulative strike-slip offset described in text. 9of25

10 Figure 8. View to the northeast of 2002 normal fault scarp across low stream terrace in Totschunda transfer zone at N, W. Scarp is a single event with 2.65 m of vertical separation and 0.6 m of right lateral slip. Scarp is typical surface expression of normal faults accommodating slip in right steps between transfer zone strike-slip ruptures. along the transfer zone (Figure 9). The closest measurements to the branch point are 5.3 m on the primary 2002 rupture only 90 m west of it and 2.4 m on the Totschunda fault 1.9 km to the east (Figures 2 and 9). The revised transfer zone displacement values increase the average slip on the western Totschunda fault in 2002 to about 2.8 m Timing of Earthquakes Prior to 2002 [18] The timing of past surface ruptures on the structural components of the Denali fault system plays a primary role in our analysis of why the 2002 Denali fault rupture propagated on to the Totschunda fault. Event timing data have been obtained at three locations that surround the intersection (Figure 2). These represent the penultimate event on the Totschunda fault, the most recent event on the eastern Denali fault, and the penultimate rupture on the central Denali fault. Table 1 lists radiocarbon dates from each of the paleoseismic sites. As part of the dating analysis we have used the radiocarbon calibration program OxCal [Bronk Ramsey, 2001] (OxCal Program, V. 4.0., 2007, Radiocarbon Accelerator Unit, University of Oxford, U.K., oxcal.html). This allows radiocarbon probability density functions (pdfs) to be trimmed and reweighted, which can result in a tighter distribution from which the age range can be extracted. In addition to the standard dendrochronologic calibration of laboratory reported ages in terms of years AD and BP, we converted the calibrated radiocarbon dates to years before the 2002 earthquake (yrb02, Table 1). This allows for a straightforward discussion of paleoearthquake timing relative to the occurrence of the 2002 event Totschunda Fault: Totschunda Sag Pond Site [19] The age estimate of the penultimate surface rupture on the Totschunda fault was obtained at the Totschunda Sag Pond site (TSP, Figure 2), which is located in the transfer zone 7.90 km southeast of the Denali-Totschunda fault Figure 9. Revised slip distribution across the Denali- Totschunda fault intersection. Black line is envelope of maximum slip curve from Haeussler et al. [2004]. Gray band is the transfer zone. Vertical red line is the branch point. Revised slip measurements are open circles. Gray line is revised envelope of maximum slip curve for the Totschunda fault in the western part of the transfer zone. Open squares are cumulative surface offset for the 2002 and penultimate events. 10 of 25

11 Table 1. Radiocarbon Dates From the Denali and Totschunda Faults Sample Lab Identification a (CAMS#) 14 C Age b (Years B.P.) Calendar Date A.D. c Calibrated Years Before 2002 (yrb02) Sample Material Fault Segment PB (0.17) peat Eastern Denali (0.50) (0.002) (0.11) (0.19) PB (1.0) peat Eastern Denali TSP (0.63) blueberry leaf Totschunda (0.37) HS500A (1.00) tree rings: 3 outer Totschunda TSP (0.60) tree rings: 3 outer of 361 Totschunda (0.39) MT-JOD (1.00) tree rings: 3 outer of 289 Central Denali MT3-NW (0.13) peat Central Denali (0.87) MT3-NW (1.00) peat Central Denali a Center for Accelerator Mass Spectrometry, Lawrence Livermore National laboratory. b Corrected for fractionation with either measured or estimated d 13 C values. c The 2-sigma calibrated ages; probability for range shown in parentheses. branch point ( N, W). Here transtension associated with the 2002 and prior ruptures resulted in a southwest-facing fault scarp with a series of graben at its base (Figure 10). In this forested area the 2002 faulting disrupted trees, causing some to fall onto the present forest floor and others across or into graben in which the 2002 ground surface had been down-dropped. We deepened and expanded one of the graben to expose the stratigraphic sequence shown on Figure 11. The base is a paleosol that contains trees lying in a sub-horizontal horizontal position, as well as woody and leafy fragments from other plant material. We interpret this to be the forest floor at the time of the penultimate Totschunda fault rupture, which was subsequently buried by loose gravelly colluvium derived from erosion of the adjacent fault scarp. The trees within it fell in a manner similar to those in With stabilization of the colluvial surface the organic mat of the present forest floor began to develop. Along sections of the 2002 free-face scarp-derived colluvium is already covering the 2002 forest floor. While a storm could be appealed to for disruption of the trees, the burial of the paleosol and tree sequence by scarp colluvium supports a coseismic interpretation. [20] Three samples for radiocarbon dating were obtained from two buried trees and from a leaf that was part of a blueberry plant preserved against permafrost within the exposure (Figure 11 and Table 1). The calibrated pdfs are shown on Figure 12. Each tree sample was composed of the outer ten rings and the radiocarbon dates approximate the date of tree death (the trees may have lived for a few years after their disruption). Tree sample HS500A has a 2-sigma age range of AD 1294 to AD Tree TSP-10 has a 2- sigma range of AD 1267 to AD 1391, with a higher probability (0.6) of it being AD 1267 to AD 1321 (Table 1 and Figure 12). The blueberry leaf, TSP-6, represents less than one year of growth at the time it was buried. It has an age of AD 1299 to AD 1424 with a 0.63 probability of the age being Figure 10. View to the northeast of Totschunda fault paleoseismic site TSP in the transfer zone. Both vertical and horizontal slip occurred in Note trees disrupted by 2002 surface rupture fault scarp free-face exposes light colored fluvial gravel, which is the source of scarp-derived colluvium (sc) shown on Figure of 25

12 Figure 11. Stratigraphic sequence exposed in excavated fault line graben. The 2002 forest floor and accumulated soil (dotted yellow line) overlie gray colluvium (SC). The colluvium is derived from erosion of the adjacent fault scarp and buries a lower organic unit that is interpreted as the forest floor at the time of the penultimate Totschunda fault earthquake. Green flagging defines this contact. Red flagging is trace of 2002 rupture. Arrows show locations of radiocarbon samples from buried trees that were disrupted during penultimate surface and from buried blueberry leaf (inset). in the older part of its range (AD 1299 to AD 1370). These radiocarbon ages differ only slightly by a few decades at the tails of their respective pdfs (Figure 12). Based on this we interpret the interval AD 1267 to AD 1424 (578 to 735 years before 2002) to contain the most recent large surface rupture along this part of the Totschunda fault. The radiocarbon pdfs indicate a slightly higher probability for the older part of this range Eastern Denali Fault: Pooh Bear Site [21] The Pooh Bear site on the eastern Denali fault is located 24.1 km east of the Denali-Totschunda fault branch point ( N, N, W). It contains a sag pond within a well-defined 75 m 80 m offset of a glacial outwash fan terrace riser (Figure 13). The site was initially explored in July 2004 with a short trench on the southeastern margin of the sag pond that was re-excavated and lengthened in August The trenches exposed a sequence of layered peat, laminated sand and silt, rare gravel, and a tephra layer representing the approximately 2000-year-old [Lerbemko et al., 1975] White River ash (Figure 14). [22] The identification of individual paleoearthquakes is based on soft sediment deformation in the form of folds (Figures 14 and 15), fold-related extensional faults that sole into flat-lying detachments along depositional boundaries (Figure 15a), and brittle high-angle strike-slip faults that extend through the complete section (Figures 14 and 15b). The trench did not cross the entire sag and it is possible that there are additional fault splays. The folding most likely represents soft-sediment deformation related to shaking and liquefaction. The small listric normal faults, which are more apparent at the south end of the 2005 trench (Figure 15a), are also a likely a response to shaking and perhaps express lateral spreading within the sag, but they could represent a small amount of extension associated with the pull-apart in which the sag pond sits. [23] From a combination of the 2004 and 2005 trenches we interpreted the occurrence of four, and possibly five, paleoearthquakes during the past 2600 to 2800 years. Here, however, we describe only the most recent event for comparison with the other sites. The sub-vertical brittle fault zone in the central part of the trench reflects multiple ruptures with broader and more complex faulting in older deposits, and with a number of strands erosionally truncated at lower stratigraphic levels (Figures 14 and 15b). The change in type and thickness of facies across the fault is consistent with lateral slip. The most recent fault strand extends upward to within at least 7 cm from the present ground surface (Figures 14 and 15b) where it becomes obscured by bioturbation. The timing of this paleoearthquake is constrained by sample PB1 05, which is from a peat layer just below the present ground surface. It was folded during the most recent event along with a thin overlying silt and the present surface organic root mat (Figures 14 and 15c); a lower peat, sampled at PB2 05, was folded at the same time. The calibrated two-sigma range of PB1 05 is AD 1656 to AD 1953 (Table 1). The pdf of the calibrated date shows that almost all of the probability is between AD 1650 and AD 1817 (Figure 16a). The underlying folded peat, PB2 05, has a calibrated age of AD (Table 1). [24] Doser [2004] relocated historical seismicity along and adjacent to the Denali-Totschunda fault system and concludes that with the possible exception of a large (M s 7.2) event in 1912, no earthquakes with M >5.5 appear to be associated with it since that date. Doser [2004] places the 1912 earthquake close to the Denali fault but with an error ellipse that extends from about W to W. Carver 12 of 25

13 Figure 12. Probability density functions (pdfs) of calibrated radiocarbon ages from Totschunda paleoseismic site TSP. (a) TSP-6 is blueberry leaf. (b) HS500A and (c) TSP-10 are the outer ten rings of two buried trees. et al. [2004] interpret older damage to trees near the Delta River, which were affected by the 2002 surface rupture, to have resulted from surface offset at that location in There has been no historical surface rupture along the eastern Denali fault, and D. L. Doser (personal communication, 2004) suggests that, at a minimum, this period extends back to AD Based on this, 1900 is used as a historical constraint on sample PB1 05 in the OxCal modeling. When it is applied the probability is significantly redistributed, with more than 50 percent between AD 1800 and AD 1900 (Figure 16b). Combining the stratigraphic and radiocarbon observations with the historical record of large earthquakes, we interpret the most recent surface rupture at the Pooh Bear site to have occurred between AD 1656 and AD 1900 (102 to 346 years before 2002), with the likelihood that it was closer to the younger part of this range. This is consistent with the upward extent of brittle faulting and folding to the present ground surface Central Denali Fault: Mentasta Site [25] The timing of the penultimate central Denali surface rupture is obtained at the Mentasta site, 22.6 km west of the branch point ( N, W) along a section of fault where individual 2002 offsets ranged from 5.5 m to 6.9 m (Figure 2). At this location faulting extended across a large, active, and heavily forested alluvial fan, producing long (up to 50 m) and deep (up to 3 m to 5 m) left-stepping en-echelon tension gashes (Figure 17). A parallel secondary trace 60 m to the south did not slip in 2002 and represents local complexity during a previous rupture. It extends approximately 840 m, ending just southeast of the road (Figure 17). The tension gash we sampled lies along the continuous main trace and it is unlikely that the penultimate rupture bypassed the site on this short branch. [26] The geomorphic expression of the interior of the tension gash indicates the presence of an existing depression that was deepened by the 2002 rupture (Figure 18a). Event timing was obtained at locations across the gash from each other, and two possible ages for the penultimate event are developed. One is based on dates of peat horizons associated with a pre-2002 colluvial wedge (Figure 18b). The other is from the dating of a tree interpreted as tilted by the penultimate rupture (Figures 18c and 18d). [27] The northeast wall of the tension gash exposed a stratified sequence of sandy silt, gravel (likely debris flows), peat, and colluvium (Figure 18b). The colluvium is interpreted as scarp colluvium from the penultimate rupture. Clasts in the wedge, including those of peat, are imbricated at an angle that is steeper than the adjacent layered deposits. The degrading fault-scarp free-face extends into gravelly sand for about 40 cm above the peat layer from which sample MT3-NW6 was taken (Figure 18b). The colluvial wedge is overlain by continuous beds of peat and silt that extend to the present ground surface. Sample MT3-NW5 is from the lowest peat in the stratigraphic sequence that overlies the wedge (Figure 18b). Based on these relations the penultimate event occurred between the ages of these two samples, AD 1295 to AD 1407 and AD 1322 to AD 1446 (Table 1). The OxCal radiocarbon model (Figure 19a) trims the tails of the pdfs and places the event between AD 1319 and AD 1415 (587 to 683 years before 2002) with the bulk of the probability between AD 1319 and AD 1379 (623 to 683 years before 2002) (Figure 19b). [28] A different event age is obtained from a tree and associated root mat interpreted as tilted by rotation into a graben during a paleoearthquake, and subsequently buried by sandy deposits (Figures 18c and 18d). A radiocarbon date of the outer three rings (sample MT-JOD) yields a 13 of 25

14 Figure 13. Oblique aerial view-to-the north of the Pooh Bear paleoseismic site on the eastern Denali fault. Dashed yellow line is generalized trace of the eastern Denali fault. Terrace riser from incised glacial outwash fan is offset 75 to 80 m (tr to tr ). The 2005 trench (dark area at tip of arrow) extends approximately halfway across the sag pond. Figure 14. Photo mosaic of east wall of trench at eastern Denali fault Pooh Bear site excavated in Note shaking-produced folds in pond deposits. Gray units are fine sands and silts; darker units are peats and organic-rich layers. Orange flagging at south end of trench is brittle strike-slip faulting. Arrow points to upward definable extent of most recent surface rupture. Thin white stringer (arrow) in center of trench is 1900-year-old White River ash. Yellow dots are locations of radiocarbon samples. PB1 05 constrains the age of the most recent rupture. 14 of 25

15 Figure 15. Photographs of details of stratigraphic and structural relations at the eastern Denali paleoseismic site. (a) West wall of 2005 trench at showing extensional deformation in sag pond deposits. Most of these small fault planes sole into flat-lying depositional contacts. They most likely represent shakinginduced soft sediment deformation but may be an expression of extension across the sag pond. Distance between white flags on level line is 1 m. (b) East wall of 2004 trench showing brittle strike-slip faulting. The orange flagging marks fault locations. The fault zone is broader and more complex with depth and individual fault planes are erosionally truncated at different stratigraphic levels. The most recent rupture can be confidently traced to the tip of the arrow, about 7 cm below the present ground surface. White flags on level line are 0.5 m apart. (c) Close-up of location east wall of 2005 trench. Re-excavation of trench and slight cutting back of wall exposed change in geometry of folds (compare with Figure 14). Note folding of peat containing sample PB1 05. The peat sample layer is overlain by thin tan silt, which is also folded, and the organic root mat developed on the present ground surface. White flags on level line are 1.0 m apart. calibrated age range of AD 1408 to AD 1483 (517 to 594 years before 2002, Table 1). As at the Totschunda Sag Pond site we interpret this as reflecting the death of the tree. The pdf of the calibrated age (Figure 19c) has a single intercept with the dendrochronological correction curve, resulting in a relatively narrow 2-sigma age range. As a result there appear to be two possible paleoevent dates at this site, one in the 1300 s (scarp-colluvium relations) and one in the 1400 s (buried tilted tree). While these could be interpreted as separate earthquakes, which in this setting would be difficult to distinguish if they were relatively closely timed, we think it is more likely that they represent the same rupture. The age range of the postcolluvial peat, by itself, overlaps and is consistent with age of death of the tilted tree; the pre-event peat is overlain by 40 cm of pebbly sand indicating that there is time between its burial and the penultimate earthquake. The preference is to favor the radiocarbon dating from the tree because the tilting is directly related to a surface rupture. 3. Discussion 3.1. Dynamic Models of Fault Branching and the Denali-Totschunda Fault Rupture [29] Since the early 1990s there has been a focus on modeling the controls of rupture propagation, particularly in regard to fault stepovers [Harris et al., 1991; Harris and 15 of 25

16 Figure 16. Probability density functions (pdfs) for the age of sample PB5 01. (a) Pdf of date showing full range of age probabilities; dendro calibration (light shade) and refined with historical constraint (dark shade). (b) Pdf of OxCal model of age probabilities with historical constraint. Note shift in probability density into the younger calendar range. Day, 1993] More recent analyses of coseismic branching along strike-slip faults include 2D numerical simulations of dynamic rupture on theoretical faults using elastodynamic crack theory [Poliakov et al., 2002; Kame et al., 2003]. These suggest the direction of fault branching depends on: the prestress state (the orientation of regional maximum compressive stress, S max, relative to the main fault); the speed of crack propagation; and the angle between the main and branch faults. In models of the 2002 Denali fault rupture [Bhat et al., 2004; Dreger et al., 2004; Oglesby et al., 2004] the relationship between prestress state (regional and local) and fault orientation as a fundamental control of branching direction is a principal theme. [30] Bhat et al. [2004] analyze the dynamic transfer of the central Denali rupture to the Totschunda fault. They focus on the preexisting stress state, the rupture velocity at the branch point, and the branch angle between the Denali and Totschunda faults. The orientation of S max is not well constrained for this part of Alaska. Bhat et al. [2004] select values of 70 and 80 to span a range of observations that includes regional stress studies [Eastbrook and Jacob, 1991] and focal mechanism inversions pre-and-post the 2002 rupture [Ratchkovski and Hansen, 2002; Ratchkovski, 2003]. For the rupture velocity, which is another uncertainty, they conduct numerical simulations using four values, including supershear, for the rupture as it approaches the branch point. Their branching angle is 15. The results of their simulations indicate that an eastward-propagating rupture on the central Denali fault branches exclusively onto the Totschunda fault when the prestress orientation is 80 and the rupture velocity is 0.9 c s (shear wave velocity). When the prestress is 70 and rupture velocity is 0.9c s, the rupture chooses the Totschunda fault but rupture occurs along the Denali fault for a short distance beyond the branch point. In their analysis Bhat et al. [2004] conclude that the primary control of the choice of the Totschunda fault is the orientation of both faults to a uniform regional stress field. [31] Dreger et al. [2004] develop a kinematic and dynamic model of the 2002 rupture from regional seismic waveforms, GPS data, and surface slip measurements. In this model rupture abandons the central Denali fault for the Totschunda fault as a direct result of its more favorable orientation to the regional stress field, and the dynamic effect places the eastern Denali fault in a stress shadow. The regional stress field is generalized with an angle of 65 to the Denali fault because it produces results that better fit the timing of rupture on faults segments in their model. Oglesby et al. [2004] expand this approach, particularly the forward dynamic rupture modeling. They test a range of stress field orientations but to reproduce the joint kinematic and dynamic slip pattern a heterogeneous prestress field, in which there is a spatially variable shear and normal stress pattern, is introduced. In this, their preferred model, there is discontinuous rupture propagation. Seismic waves (S waves) traveling ahead of the rupture front cause the Totschunda fault to slip before the main rupture arrives. In doing so, a stress shadow is created that clamps the eastern Denali fault and inhibits its rupture. Oglesby et al. [2004] conclude that the physical reason for failure of the Totschunda fault, irrespective of the rupture jumping ahead, is that it had a more favorable orientation with regard to the local stress field. [32] Duan and Oglesby [2007] examine rupture on branched faults through multicycle dynamic simulations on a two-dimensional fault system. Using a range of branching angles and stress orientations they conclude that fault prestress prior to an earthquake becomes non-uniform in magnitude and orientation on a branched fault over multiple earthquake cycles and departs significantly from the regional stress field near the branch point because of stress interaction between branches. In analyzing the Denali-Totschunda fault rupture they interpret the 2002 branching as only one of several scenarios if non-uniform prestress resulting from previous earthquakes is taken into account. The effect of prior earthquakes is an issue we return to. [33] Ratchkovski [2003] analyzed the stress field around the Denali and Totschunda faults using stress tensor inversions of earthquake focal mechanisms for events prior to, and aftershocks following, the 2002 rupture, and inversion of the data set indicated heterogeneous stress conditions along the faults. The orientation of the maximum compressive stress in the crustal volume adjacent to the central and 16 of 25

17 Figure 17. LiDAR image (45 sun angle) showing location of the tension gash at Mentasta (yellow circle) along the central Denali fault from which timing information of the penultimate event was obtained. Red lines are 2002 rupture interpreted from post-earthquake 1:6000 aerial photography [Haeussler et al., 2004]. Gaps in photo mapping of 2002 rupture trace reflect locations obscured by heavy forest. Arrows point to parallel trace, revealed by LiDAR, which did not rupture in Number 205 is distance in kilometers east of 2002 epicenter [Haeussler et al., 2004]. western sections of the Denali rupture prior to 2002 is better constrained than the orientation adjacent to the Denali-Totschunda fault branch point because of the low level of seismicity associated with the eastern fault segments. Nonetheless, Ratchkovski [2003] suggests that it was at an angle of 50 with respect to the Totschunda fault and normal to the trend of the Denali fault. Following the 2002 earthquake the stress orientations adjacent to the branch point rotated clockwise about 20 degrees. Ratchkovski [2003] notes that, in time, the stress should recover the pre-2002 orientations. Ruppert [2008], updating the moment tensor inversions from all of Alaska, suggests that the orientation of S max for the Totschunda fault (the crustal volume includes the sub parallel eastern Denali fault) is 196, which yields an angle of 73 with the Denali fault and 56 with the Totschunda fault. [34] The orientations of the faults are precisely known as opposed to the prestress state at the time of the 2002 rupture. Nonetheless, it appears that the prestress was at a relatively high angle to the main Denali fault, and the branch angle as revealed by the LiDAR is 19. Both are consistent with the general model parameters for rupture onto an extensional branch (right branch on a right lateral fault) [Kame et al., 2003], in this case the Totschunda fault [Bhat et al., 2004; Dreger et al., 2004; Oglesby et al., 2004]. The question we raise is: are these angular regulations between the faults and prestress a sufficient physical basis for the rupture to branch as it did or were there other processes in play that allowed it to propagate along the Totschunda fault? Do eastwardpropagating ruptures always extend on to the Totschunda fault, do they sometimes stop at the intersection, or can they continue along the Denali? In several of the dynamic analyses the effects of prior earthquakes are alluded to as a potentially important factor although the concept is not expanded. We explore this below Interpreting the Extent of 2002 and Older Ruptures From Prior Earthquake History The 2002 Rupture [35] We propose an alternative basis for the 2002 branching that takes into account the occurrence of past earthquakes, specifically differences in the timing of prior events. This, by itself, is not a physical reason for control of branching. However, the penultimate Totschunda surface rupture is significantly older than the most recent eastern Denali event. We suggest, at least in the case of the 2002 rupture, that this translates into a difference in the amount of accumulated strain, or potential slip, on each fault and that this was a major contributing factor. [36] One approach to estimating the amount of potential slip is from the elapsed time and fault slip rate. The slip rates on the eastern Denali and Totschunda faults are sub-equal. The late Pleistocene-Holocene rate is 8.4 (2.2) mm/yr for the eastern Denali fault from a site 33.3 km east of the branch point, and 6.0 (1.2) mm/yr for the Totschunda fault averaged from locations 22.3 km and 24.3 km southeast of it (Figure 2) [Matmon et al., 2006]. Ninety kilometers west of the intersection on the central Denali fault late Pleistocene slip rates of 12.1 (1.7) mm/yr [Matmon et al., 2006] and mm/yr [Meriaux et al., 2009] represent, within uncertainties, the combined eastern Denali and Totschunda fault rates. The mean slip rate and the range of elapsed times (100 to 344 years with the event closer to the younger age) on the eastern Denali fault yield an estimated strain 17 of 25

18 Figure 18. Photographs of geomorphic and stratigraphic relations in the Mentasta tension gash. (a) View to northwest of tension gash through forested area at Mentasta site. The 2002 rupture re-occupied existing geomorphic depressions. Line with ticks marks degraded tension gash margin from penultimate surface rupture. (b) Penultimate event colluvial wedge. Solid white line is base of stratigraphic sequence that overlies scarp colluvium. Dotted line is contact between scarp-derived colluvium and stratified gravel, sand, and peat. Radiocarbon sample MT3-NW6 predates penultimate event. Sample MT3-NW5 is from peat that extends across top of colluvium and post-dates penultimate event. Dashed line is location of 2002 rupture that displaces the colluvium. (c) Initial exposure of tree that was tilted during penultimate event and subsequently buried. Tree is located directly across tension gash from scarp colluvium shown in Figure 18b. (d) View of excavated tree, showing location of radiocarbon sample MT3-JOD and age range of outer rings. accumulation of 0.84 m to 2.88 m at the time of the 2002 event; the full range with the slip rate uncertainties is 0.62 m to 3.65 m. The accumulation on the Totschunda fault was 3.46 m to 4.41 m based on the mean slip rate and elapsed times ( years), with a full range of 2.77 m to 5.29 m. These values are derived directly from the slip rates and event dates and should be viewed as approximations, but they clearly imply that a larger amount of slip, at least in this model, had accumulated on the Totschunda fault. [37] There is no information on the amount of offset during individual paleoearthquakes on the eastern Denali fault for comparison with the amount of modeled slip accumulation in 2002, although they are likely to be several-meter-per-event ruptures given the well-expressed fault geomorphology and 100 m to 150 m offsets of late Pleistocene glacial deposits. However, there are measurements of the amount of slip in the penultimate Totschunda fault rupture at two locations (Figure 9). One is on the western end of the linear section of the rupture at km where slip in 2002 was 2.8 m [Haeussler et al., 2004] and the associated older channel margin was offset 5.8 m. The other is near the eastern end of the 2002 rupture at km 300 where 2002 slip measured across the thalweg of a small gully was 1.4 m [Haeussler et al., 2004] and the associated gully margin was offset 3.1 m. Although clearly limited in extent, these observations are at least suggestive that when the Totschunda fault slipped in 2002 it was with a slip distribution similar to that of the penultimate rupture, and in the range of estimated accumulated slip. 18 of 25

19 Figure 19. Radiocarbon probability density functions (pdfs) for the Mentasta site. (a) OxCal model of penultimate event based on colluvial stratigraphy using samples MT3-NW6 and MT3-NW5. (b) Expanded view of OxCal modeled pdf for the penultimate event. (c) Pdf of date of outer rings of tree, sample MT3- JOD, whose death reflects timing of the penultimate event. Narrow age range reflects single intercept of dendrochronologic correction curve. [38] We have used a long-term slip rate (thousands of years) for estimating strain accumulation over hundreds of years. The degree to which strain accumulation and release on faults is linear over different time intervals, or if it is at all, is a major question. Certainly in other tectonic settings, particularly in lower strain rate intraplate regions [see, e.g., Wallace, 1984; Rockwell et al., 2000; Schlagenhauf et al., 2011], both regional and fault-specific rupture recurrence is variable and often clustered. Across some high rate fault systems such as the San Andreas in the San Francisco Bay region the long-term (from plate rates) and short-term (geodetic) strain accumulation rates are consistent with the strain release rates (geologic slips rates over 100 s to a few thousand years) at about 40 (4) mm/yr [Working Group on California Earthquake Probabilities, 2003; d Alessio et al., 2005]. At the opposite end of the clustering spectrum is the Alpine fault, New Zealand. Paleoseismic studies show that for the past 8000 years recurrence of surface faulting has been very regular with a coefficient of variation of 0.33 [Berryman et al., 2012]. Berryman et al. [2012] conclude that the periodicity results from a high slip rate (23 mm/yr), relatively simple structure (few geometric steps), and, perhaps most importantly, the absence of other major structures for much of its length. They suggest that these features may characterize sections of the north Anatolian fault (Turkey) and the Denali fault. [39] The Denali fault system is a high slip rate zone (15 mm/yr) that is in essential isolation from other faults in central Alaska and is driven by even higher- rate interaction between the Pacific and North American plates (Figure 1). In this setting it would not be unreasonable to interpret relatively constant strain accumulation rates on the fault and its components. Modeled static stress transfer effects across the intersection [Anderson and Ji, 2003] place up to 420 kilopascals (kpa) of Coulomb stress on the northwest section of the Totschunda (including the transfer zone) and an even 19 of 25

20 larger 470 kpa on the eastern Denali. We propose that when the 2002 rupture, or advancing seismic waves in some dynamic models, reached the intersection insufficient accumulated strain, reflecting the recency of the most recent event inhibited continuation along the eastern Denali fault. In contrast, rupture continued along the Totschunda fault because of a significantly longer elapsed time during which strain had accumulated to a critical failure level, perhaps comparable to the amount of slip in the prior event Pre-2002 Ruptures [40] We have focused on the 2002 rupture. How frequently does this branching occur? Most seismogenic faults have not had one historical rupture, let alone repeats along the same section, and ruptures scenarios must be based on paleoseismic observations. This information is limited for the Denali fault system. Timing of the penultimate event at the Mentasta site occurred between AD 1319 and AD 1415 or between AD 1408 and AD 1483 If the former, the penultimate rupture on the central Denali fault, within the dating uncertainties, is indistinguishable from that on the Totschunda fault (AD 1267 to 1424) and could be the same branching that occurred in In the latter case, which is the range we favor, it is a separate and younger rupture. Uncertainties in rupture extent based on radiocarbon dates are not unique to the Denali fault and have been treated statistically on the southern San Andreas fault for which significantly more paleoseismic information is available [Biasi and Weldon, 2009]. [41] The Denali-Totschunda fault intersection is the largest geometrical complexity on the fault system. It likely plays a role in the Denali rupture process through time as suggested by the large 2002 slip step and longer-term slip rate changes across it. Rupture might stop at it some of the time and perhaps this occurred during the penultimate central Denali event, at least within the uncertainties of the paleoseismic site locations (Figure 2). The direct connectivity of the central Denali with the eastern Denali and Totschunda faults structurally enhances the potential for rupture propagation through it. Intersections such as this, which are also referred to as intersegment zones, are considered low strength barriers that allow rupture to cascade from one fault segment to another on structurally mature faults [Manighetti et al., 2007] like the Denali and Totschunda. Only 3.1 km southeast of the branch point late Pleistocene-Holocene glacial deposits are right-laterally offset 100 m to 120 m across the eastern Denali fault (Figure 4). These cumulative displacements are the result of repeated individual multimeter-per-event offsets. The occurrence of large individual slips so near the intersection suggests continuity of prior ruptures across it. Epicentral locations and rupture directivity of paleoearthquakes along the Denali fault system are unknown. If rupture initiated on the eastern Denali fault and propagated to the northwest it could continue along the central Denali without affecting the Totschunda fault. The angular relationship between the two faults is one for which back branching, in this case along the Totschunda fault, is not expected [Fliss et al., 2005]. Similarly, there is no basis to preclude a northwest propagating rupture on the Totschunda fault from extending onto the central Denali. Alternatively we suggest that an eastward propagating rupture, such as 2002, could continue directly beyond the intersection if there was sufficient accumulated slip on the eastern Denali fault, even though the geometry favors branching onto the Totschunda fault in most dynamic rupture models Historical Examples of Coseismic Branching [42] There are few historical ruptures that provide strong tests of models of fault branching, whether simulations or the process we suggest. Poliakov et al. [2002] use observations of the 1971 M w 6.4 San Fernando, 1985 M w 6.1 Kettlemen Hills, 1992 M w 7.3 Landers, and 1979 M6.4 Imperial Valley earthquakes in support of rupture branch orientation relative to orientation of regional S max as a fundamental control of branching direction. Kame et al. [2003] add a model of a splay fault off of the plate interface as part of the 1944 M w 8.1 Tonanki, Japan earthquake rupture. Three of the events, San Fernando, Kettlemen Hills, and Tonanki are on thrust faults, whereas the dynamic models that use these as examples have been for strike-slip faults. More importantly, none provides a direct observation of the actual geometrical relation between the main fault and branch fault; rather, this is inferred from aftershock distributions or seismic reflection profiles. [43] For the strike-slip examples, Kame et al. [2003] cite branching of rupture from the Imperial fault into the Brawley seismic zone but state that it is the least definitive case because of the uncertainty in prestress orientation. Actually, the 1979 Imperial fault rupture extended 10 km northwest of a projected intersection of the two structures and the Brawley fault ruptured independently in 1949 and 1975 with primarily normal slip [Sharp et al. 1982]. The 1992 Landers rupture is most consistent with dynamic model results [Kame et al., 2003] at least with respect to rupture propagation from the Johnson Valley fault on to the Kickapoo fault. Structural geometries between the Homestead Valley, Emerson, and Camp Rock sections of the Landers rupture relative to branching theory are much less certain. [44] The 2001 M w 7.8 Kokoxili earthquake on the Kunlun fault, Tibet, has been referred to as an analog to the Denali event [Ando et al., 2009] in that surface faulting appears to have continued on a branch as opposed to the main fault. The 2001 surface rupture [van der Woerd et al., 2002; Lin et al., 2002; Klinger et al., 2005; Xu et al., 2002, 2006] occurred principally on the Kusai Hu segment (for approximately 270 km). Near the eastern end it bypassed the primary strand of the Kunlun fault (Xidatan segment) and continued for an additional 70 km on the Kunlun Pass segment. The Xidatan segment is on the extensional side of the intersection (a left branch on a left lateral fault) and might be expected to be dynamically favored. However, Klinger et al. [2005] note there is no distinct surface structural connection between the two main fault segments; rather, they parallel each other for several kilometers before the Xidatan segment trends to the northeast and away from the 2001 rupture. They suggest that this separation is a long-lived step and that the rupture of the Kusai Hu and Xiditan segments is rarely continuous during the same earthquake. Google Earth provides SPOT imagery on which both the 2001 surface rupture and the active Xiditan surface trace are clearly expressed. We agree with Klinger et al. [2005] that there is no obvious surface connection between the two fault traces. They can be seen as parallel structures, 2.0 km to 3.5 km apart, each with an azimuth of 100 for about 18 km (from N, E on the east to N, E on the 20 of 25

21 west). The structure of this intersection differs from the Denali-Totschunda intersection in that there is no direct (at least no apparent) surface connectivity and there is a long overlap zone. It is, however, representative of a class of intersection, discussed below, that includes the San Andreas- San Jacinto fault and San Andreas-Calaveras fault intersections in California The Denali-Totschunda Rupture, Connectivity, and Implications for Seismic Hazards [45] The propagation path of the 2002 Denali fault rupture led Field et al. [2009] to cite this as a primary example of a fault-to-fault jump. They suggest that for seismic source characterization fault-to-fault jumps might be more frequent and produce longer ruptures than have been incorporated into analyses such as the Uniform California Earthquake Rupture Forecast (UCERF). In that analysis the frequency of M earthquakes in the long term (geological) California frequency-magnitude relation exceeds their occurrence rate in the historical frequency-magnitude distribution for the State. Field et al. [2009] suggest that more fault-tofault ruptures in the model would result in larger earthquakes that could release sufficient seismic moment to significantly reduce the rate of modeled events near M6.5 and bring the two distributions into closer agreement. [46] The potential implications of the Denali-Totschunda rupture for seismic hazard estimates have led us to look at intersections on other long, high-slip-rate strike-slip faults from the perspective of In making this comparison we identify at least two broad classes of intersections. These are: 1) intersections with direct surface connectivity and 2) those without surface connection that we term volumetric intersections. We describe examples of each below, including surface structure (Figure 20) and historical and paleoseismological data, where available, that provide some insight into the frequency of coseismic branching Intersections With Direct Surface Connectivity [47] Ando et al. [2009] surveyed branching California strike-slip faults to characterize splay angles as part of an analysis of why faults form splays, and note that the majority are associated with the San Andreas, San Gregorio, and Calaveras faults in northern California, and with the San Andreas and San Jacinto faults and the Mojave shear zone in southern California. The surface connectivity displayed at the Denali-Totschunda fault intersection is not a common feature of high slip rate Holocene strike-slip faults in the California fault system. The primary analog is the San Andreas-San Gregorio fault intersection (Figure 20). Offshore reflection profiles [Ryan et al., 2008] place the eastern San Gregorio and the San Andreas faults about one kilometer apart approximately 7 km southeast of Bolinas, California (Figure 20). Grove and Niemi [2005] suggest that the San Gregorio fault is expressed on land by the Western Boundary fault [Galloway, 1977], which extends directly from the offshore strand and is mapped as intersecting the 1906 San Andreas trace 7.1 km northwest of the shoreline (Figure 20). The apparent branching angle is 8 and increases to 21 approximately 30 km to the southeast as the two strands gradually separate offshore. [48] Earthquake probability estimates for the San Francisco Bay region have characterized the San Andreas and San Gregorio faults as completely independent earthquake sources [Working Group on California Earthquake Probabilities, 2003; Field et al., 2009]. The 1906 San Andreas rupture initiated southeast of their intersection and propagated bi-laterally, thereby passing the branch point with its acute angle oriented southeast. However, a rupture (past or future) initiating northwest of the intersection and propagating to it would encounter a branching geometry with the San Gregorio fault on the extensional side. For the segment of the San Andreas fault north of the intersection radiocarbon age ranges indicate the penultimate rupture occurred no earlier than AD 1660 and suggest a more likely occurrence in the 1700 s [Hall and Niemi, 2008; Kelson et al., 2006]. For the San Gregorio fault the most recent large event occurred between AD 1700 and 1776 [Schwartz et al., 2006; Pollitz and Schwartz, 2008]. Given the radiocarbon age overlap, the possibility can be entertained that the two faults ruptured together. The slip rates, however, are quite different (25 4 mm/yr for the San Andreas Fault, 7 3 mm/yr for San Gregorio fault) [Working Group on California Earthquake Probabilities, 2003], which indicates that if San Andreas- San Gregorio ruptures occur they are infrequent. [49] Other intersections that have direct surface connectivity are those of the Pilarcitos and Zayante-Vergeles faults, which branch north from the San Andreas, and the San Juan fault, which branches to the south [U.S. Geological Survey, 2006]. The lengths of these branch faults are limited, and slip rates are significantly lower than on the San Andreas fault. The Pilarcitos fault extends for 40 km northwest of the San Andreas with the most recent movement Quaternary in age. The Zayante-Vergeles is approximately 85-km long with only 8 km showing Holocene displacement and the remainder of the fault mapped as having Quaternary and late Quaternary activity. The San Juan fault branches south from the San Andreas fault for 65 km with only its northernmost 8 km mapped as Holocene. There are no data on the timing of prior individual surface ruptures across these intersections. However, large differences in slip rate on these direct splays off of the San Andreas fault indicate that coseismic branching has not occurred frequently Volumetric Intersections Without Surface Connections [50] A second class of intersection is characterized by faults that are typically km apart and parallel each other for up to 25 km with no apparent surface connection. We refer to these as volumetric intersections because there is little or no information on their structure or connectivity at depth. The previously described intersection of the 2001 Kusai Hu rupture and the Xidatan segment of the Kunlun fault is one. The absence of surface connections at the Hope and Alpine fault intersection, New Zealand, as well as at the San Andreas- Calaveras and San Andreas-San Jacinto fault intersections, has been noted by Scholz et al. [2010]. They suggest that, in general, crustal-scale strike-slip splay faults do not appear to be directly linked to the primary fault. In the California strikeslip system the Calaveras-San Andreas and the San Jacinto- San Andreas are the principal volumetric intersections. [51] The Calaveras fault accommodates close to half of the San Andreas slip rate as it extends into northern California. The mapped structure of the intersection is shown on Figure 20. For approximately 25 km, from southeast of Paicenes to just south of Ridgemark, California, the Holocene traces of the Calaveras (also called the Paicenes fault) and 21 of 25

22 Figure 20. Maps of intersections of major California plate boundary strike-slip faults. (a) San Andreas- San Gregorio fault; (b) San Andreas-Calaveras fault; (c) San Andreas-San Jacinto fault. Inset shows locations of intersections. Fault traces and ages of most recent slip from U.S. Geological Survey [2006]. 22 of 25