Discovery of an Active Forearc Fault in an Urban Region: Holocene Rupture on the XEOLXELEK-Elk Lake Fault, Victoria, British Columbia, Canada

Abstract

Subduction forearcs are subject to seismic hazard from upper plate faults that are often invisible to instrumental monitoring networks. Identifying active faults in forearcs therefore requires integration of geomorphic, geologic, and paleoseismic data. We demonstrate the utility of a combined approach in a densely populated region of Vancouver Island, Canada, by combining remote sensing, historical imagery, field investigations, and shallow geophysical surveys to identify a previously unrecognized active fault, the XEOLXELEK-Elk Lake fault, in the northern Cascadia forearc, ∼10 km north of the city of Victoria. Lidar-derived digital terrain models and historical air photos show a ∼2.5-m-high scarp along the surface of a Quaternary drumlinoid ridge. Paleoseismic trenching and electrical resistivity tomography surveys across the scarp reveal a single reverse-slip earthquake produced a fault-propagation fold above a blind southwest-dipping fault. Five geologically plausible chronological models of radiocarbon dated charcoal constrain the likely earthquake age to between 4.7 and 2.3 ka. Fault-propagation fold modeling indicates ∼3.2 m of reverse slip on a blind, 50° southwest-dipping fault can reproduce the observed deformation. Fault scaling relations suggest a M 6.1–7.6 earthquake with a 13 to 73-km-long surface rupture and 2.3–3.2 m of dip slip may be responsible for the deformation observed in the paleoseismic trench. An earthquake near this magnitude in Greater Victoria could result in major damage, and our results highlight the importance of augmenting instrumental monitoring networks with remote sensing and field studies to identify and characterize active faults in similarily challenging environments.

Key Points

• We document a single Holocene earthquake on a previously unidentified fault in the northern Cascadia forearc near Victoria, Canada

• Trenching, shallow geophysics, and fault-fold modeling show 2.3–3.2 m of dip slip during an M 6.1–7.6 earthquake between 4.7 and 2.3 ka

• A combination of methodologies can enhance the study of active faults in forearcs, urbanized areas, and previously glaciated terrain

Plain Language Summary

Faults occurring in the upper plate above a subduction zone are often located near densely populated coastal areas, but their hazard is often underappreciated due to their low deformation rates. In the northern Cascadia forearc on the west coast of North America, high-resolution topography and geologic mapping show a ∼2.3-m-high scarp across a ∼14,000 year-old land surface 10 km north of downtown Victoria, British Columbia, Canada. This newly identified fault, the XEOLXELEK-Elk Lake fault (XELF), crosses Saanich Peninsula within Greater Victoria and poses a hazard to the region's ∼400,000 inhabitants. Therefore, determining whether it produced recent large earthquakes is important for updating regional earthquake hazard models and increasing earthquake preparedness. To study the earthquake history of the fault, we used shallow geophysical techniques and excavated a trench across the scarp to examine the sedimentary record of deformation. These combined methodologies determined a single large earthquake, of magnitude 6.1–7.6, likely occurred on the XELF between ∼4,700 and 2,300 years ago. A similar future earthquake on the XELF Lake fault could cause major damage to the Greater Victoria area. Thus, our results can improve future earthquake hazard assessments.

1 Introduction

Seismic hazard assessments along subduction zones commonly focus on the hazard from megathrust earthquakes, but damaging earthquakes on forearc faults within the upper plates of subduction zones represent another important hazard contribution. For example, the destructive M_w 6.9 1995 Kobe (Kanaori & Kawakami, 1996), and M_w 7.8 2016 Kaikoura (Hamling et al., 2017) earthquakes ruptured previously unknown or poorly-characterized forearc fault segments (Megawati et al., 2001; Morishita et al., 2017). Despite having lower magnitudes than the largest megathrust earthquakes, these upper plate events can cause extensive damage due to their closer proximity to infrastructure resulting in major loss of life and economic damage (e.g., Gledhill et al., 2011; Okimura et al., 1996). The 1995 Kobe earthquake killed >6,400 people and caused an estimated over 10 billion USD in economic losses (Fujiki & Hsiao, 2015; Okimura et al., 1996). Not only does this proximity increase shaking intensity and fault displacement hazard, it also renders earthquake early warning systems, which usually focus on offshore megathrust faults, ineffective due to the insufficient time gap between incoming P and S waves (Wald, 2020).

The locations and kinematics of forearc faults are often poorly constrained due to low slip rates (e.g., <1 mm/yr, Cortés et al., 2012; Sherrod et al., 2013; Personius et al., 2014), variable and long earthquake recurrence intervals (e.g., up to 10 kyr, Mouslopoulou et al., 2014; Personius et al., 2014), the dominance of subduction zone coupling-induced strain observed with geodesy (e.g., Khazaradze et al., 1999; Mazzotti et al., 2002), and the lack of clear alignment of instrumental microseismicity along fault planes (e.g., Cassidy et al., 2000; G. Li et al., 2018). In addition, urban development in densely-populated areas common to forearc regions worldwide obscures modern surface ruptures (e.g., 1995 Kobe earthquake, Japan, Kanamori, 1995) and scarps along active fault traces (e.g., Seattle fault, USA, Blakely et al., 2002). Although there is an ever-increasing density of seismic and geodetic network coverage in many of these regions, these networks alone are not sufficient to identify geometries, slip rates, and earthquake recurrence intervals of forearc faults. These data are essential for fault-source models in probabilistic seismic hazard analysis (PSHA) and probabilistic fault displacement hazard analysis (PFDHA) (e.g., Brune, 1968; Chartier et al., 2017; Coppersmith & Youngs, 2000; Youngs & Coppersmith, 1985; Youngs et al., 2003), which are used in the seismic risk models that inform building codes and disaster management. Consequently, the integration of multiple modern paleoseismic techniques (e.g., lidar-derived high-resolution topography, shallow geophysics, and paleoseismic trenching; Camelbeeck & Meghraoui, 1998; McCalpin et al., 2023; Sherrod et al., 2008; Vanneste et al., 2001) is necessary to characterize active structures in forearcs.

The northern Cascadia forearc in western Washington, USA and southwestern British Columbia, Canada, exemplifies a region where long-standing geodetic and seismic networks alone have failed to identify active forearc structures. Forearc Global Navigation Satellite System (GNSS) velocities are dominated by interseismic loading resulting from coupling between the subducting Juan de Fuca (JdF) and overriding North America (NA) plates (Figure 1a). A reduction in the northward, trench-parallel component of the geodetic velocity field has been used to infer ∼5 mm/yr north-south crustal shortening between ∼46.5°N and 48°N (Khazaradze et al., 1999; Mazzotti et al., 2002; McCaffrey et al., 2013; Miller et al., 2001; Wang et al., 2003; Wells et al., 1998). However, this GNSS derived strain rate alone cannot identify where active faults are located as an increasing number of seismogenic faults north of 48°N have been identified using geophysical surveys, high-resolution topographic data, and trench-based paleoseismic studies (Figure 1; Duckworth et al., 2021; Kelsey et al., 2012; Morell et al., 2017; Nelson et al., 2017; Personius et al., 2014; Schermer et al., 2021). For example, the Sandy Point, Birch Bay, and Drayton Harbor faults, immediately south of the United States-Canada border at 49°N have been shown to offset Holocene sediments and be capable of producing M 6.0 to 6.5 earthquakes near urban areas (Kelsey et al., 2012). These studies initially have been focused on the Puget Lowland in Washington, but the along-strike continuation of several faults have been recently recognized near the San Juan and Gulf Islands (Barrie & Greene, 2018; Greene & Barrie, 2022), and on southern Vancouver Island (Morell et al., 2017). In particular, the recognition of the potential of the Leech River fault (LRF) on southern Vancouver Island to produce ∼M 7 earthquakes (Harrichhausen et al., 2021; Morell et al., 2018) has resulted in updates to seismic hazard assessments for Victoria (Goda & Sharipov, 2021; Halchuk et al., 2019; Kukovica et al., 2019), the capital city of British Columbia with a metropolitan population of ∼400,000 (Statistics Canada, 2023). A loss estimation scenario for a rupture of M_w 7.3 on the LRF would entail substantial economic and human losses (∼1,000 deaths and ∼20 billion CAD, Hobbs, 2022). The growing list of active structures—in Cascadia and other forearcs around the world—that escaped recognition with instrumental networks but were discovered using paleoseismological, geomorphological, and geophysical techniques, highlight the need for further analyses of potentially active forearc faults.

Figure 1

Tectonic setting, Global Navigation Satellite System (GNSS) velocities, active faults, crustal seismicity (<35 km depth), and S_Hmax directions in northern Cascadia. (a) Overview of the northern Cascadia subduction zone where subduction of the Juan de Fuca plate beneath the North America (NA) plate results in elastic loading of the forearc crust. GNSS velocities relative to stable NA (blue arrows; UNAVCO; McCaffrey et al., 2013) in the study area are dominated by the elastic loading signal from partial coupling between the subducting slab and overriding plate. (b) Known active faults, crustal seismicity and S_Hmax directions (Balfour et al., 2011) reflect that permanent deformation occurs north of 48°N, where GNSS velocities suggest north-south shortening is diminished (Mazzotti et al., 2002). Structures based on bathymetric studies in Haro Strait (Greene & Barrie, 2022) continue along strike across Saanich Peninsula where we map the newly identified XELF. LRF: Leech River fault, DMF: Devils Mountain fault, SWIF: South Whidbey Island fault, DHF: Drayton Harbor fault, BBF: Birch Bay fault, SPF: Sandy Point fault, and NOF: North Olympic fault zone. Basemaps are from the U.S. Geological Survey 3D Elevation Program (3DEP) Bare Earth DEM Dynamic service (U.S. Geological Survey, 2021) and seismicity is from the Pacific Northwest Seismic Network, 1 January 1970 to 12 August 2022 (Pacific Northwest Seismic Network, 2022).

In this contribution, we present a study using multiple methodologies to identify and characterize a previously unrecognized seismogenic forearc fault in northern Cascadia within the urban area of Victoria, British, Columbia, Canada (Figure 2a). We use high-resolution digital terrain models (DTMs) to map previously unrecognized fault-fold scarps offsetting the surface of a Quaternary drumlinoid ridge on the east side of XEOLXELEK (pronounced: hul-lakl-lik, the WSÁNEĆ peoples' name for Elk Lake in the SENĆOTEN language), 10 km north of the city center. This structure strikes east-southeast across the suburban Saanich Peninsula and dips southward beneath the densely populated urban core of Victoria (Figures 1, 2, and 3a). A paleoseismic trench and electrical resistivity tomography (ERT) surveys show a fault-propagation fold formed above a southwest-dipping reverse fault deforming late Quaternary glacial-lacustrine/marine sediments. Additional ERT surveys along strike show evidence for continued surface rupture west of XEOLXELEK. A combination of multiple Bayesian analyses of radiocarbon dates from samples within the paleoseismic trench bracket a single earthquake event age and forward modeling of the fault-propagation fold constrains the approximate fault slip during this event. These methods indicate a large, Holocene, surface rupturing earthquake on what we refer to as the XEOLXELEK-Elk Lake fault (XELF), and allow us to estimate a paleo-earthquake magnitude and an approximate a slip rate. These details, along with the fault geometry, can support updates to regional seismic hazard assessments particularly given the southwestward projection of the fault plane at depth beneath downtown Victoria (Figure 2a). These results underscore the need for more strategic and widespread efforts to find and characterize active faults in northern Cascadia and provide an example of how multiple methodologies can be employed in similar environmental and tectonic settings.

Figure 2

(a) Gridded population density (Center for International Earth Science Information Network-CIESIN-Columbia University, 2018) overlain on map of southern Vancouver Island showing the location of active faults, faults identified from bathymetry in Haro Strait by Greene and Barrie (2022), bedrock faults (British Columbia Geological Survey compilation, Cui et al., 2017), Global Navigation Satellite System (GNSS) velocities relative to stable North America with uncertainty ellipses (McCaffrey et al., 2013), and the location of Victoria International Airport and the ferry terminal. GNSS velocities have not had the elastic subduction coupling signal removed. A previously mapped bedrock fault roughly coincides with the XELF, is parallel to the Cowichan fold and thrust belt (CFTB) to the north, and intersects with the San Juan fault to the northwest. Population data from Balk et al. (2006). (b) Elevation overlain on a hillshaded digital terrain model (DTM) of Saanich Peninsula. Q scarp indicates the length of a north-facing scarp formed in Quaternary sediments and B scarp arrows indicate north-facing bedrock scarps along strike. The Quaternary scarp offsets the surface of the outlined drumlinoid. These topographic scarps (red bracket) do not directly align with the previously mapped bedrock fault with unknown slip sense (dashed black line), which separates the Bonanza volcanic rocks to the north from the West Coast complex to the south. Outcrop locations where southwest-dipping bedrock reverse faults were observed and the location of the geophysical surveys and paleoseismic trench study at Eagle and Waterski beaches are shown. Lower-hemisphere stereonet projections show bedrock fault plane orientations at Outcrops 1 and 2, slickenline orientation and slip sense at Outcrop 2, and the fault plane orientation from derived from a three point problem using the Q scarp trace across the drumlinoid ridge (Figure S1 in Supporting Information S1). Bedrock fault trace and geologic units are from the compilation by Cui et al. (2017). DTM is from the BC Open Lidar Data Portal (GeoBC, 2020).

Figure 3

(a) Google^TM satellite image of the suburban region on Saanich Peninsula with the trace of the Quaternary scarp on the XELF as a reference. (b) Same extent as (a) showing surficial geology overlain on hillshaded digital terrain model showing the Quaternary scarp crosscutting the drumlinoid ridge. The Eagle Beach trench and electrical resistivity tomography survey locations are shown by black line and the locations of topographic profiles 1 and 2 in (c) are shown by red lines at P1 and P2. Surficial geology simplified and modified from Monahan and Levson (2000). Contour spacing is 5 m. (c) Topographic profiles across the Quaternary scarp at locations P1 and P2 on (b) with vertical separation calculated with Monte Carlo simulations using different surface regressions above and below the scarp, and different fault locations. Apparent vertical separation increases from west to east; however, anthropogenic fill at the base of the scarp at P1 infers the offset at this location is a minimum. South-side-up motion on a southwest-dipping fault plane suggests reverse slip.

2 Tectonic Setting

Southern Vancouver Island is located within the northern forearc of the Cascadia subduction zone, where the JdF plate subducts northeastwardly beneath the NA plate at a rate of ∼50 mm/yr (Figure 1, e.g., McCaffrey et al., 2007). In northern Cascadia, the forearc region above the subducting slab experiences both interseismic elastic strain induced by partial coupling of the subduction zone interface (e.g., Dragert & Hyndman, 1995; Miller et al., 2001; Savage & Lisowski, 1991), and permanent strain resulting from subduction zone coupling and far-field tectonic forces (e.g., Delano et al., 2017; Finley et al., 2019; Nelson et al., 2017; Sherrod et al., 2013; Wang et al., 2003; Wells et al., 1998). Permanent forearc deformation is evidenced by instrumental crustal seismicity (e.g., Balfour et al., 2011; Bostock et al., 2019; Brocher et al., 2017; Savard et al., 2018; G. Li et al., 2018), paleoseismic studies of Quaternary-active forearc faults (Barrie & Greene, 2018; Bennett et al., 2017; Blais-Stevens et al., 2011; Blakely et al., 2009; Duckworth et al., 2021; Greene & Barrie, 2022; Harrichhausen et al., 2021; Kelsey et al., 2012; Morell et al., 2017, 2018; Nelson et al., 2017; Personius et al., 2014; Schermer et al., 2021), and paleomagnetic, geologic, and geodetic data that show counterclockwise rotation of the Cascadia forearc north of, and margin-parallel shortening south of, the Strait of JdF (Finley et al., 2019; Mazzotti et al., 2002, 2003; McCaffrey et al., 2007; Miller et al., 2001; Prothero et al., 2008; Wells & McCaffrey, 2013).

The interseismic elastic strain field dominates the geodetic signal in the northern Cascadia forearc, with most GNSS vectors aligned with subduction of the JdF plate beneath the NA plate toward the northeast (Figure 1a). Multiple interseismic coupling models, constrained by onshore geodetic data, describe the majority of this velocity field (e.g., S. Li et al., 2018; Schmalzle et al., 2014; Wang et al., 2003). Subtracting the velocities determined to result from interseismic coupling from the instrumental data yields a residual velocity field with a broad pattern of margin-parallel, north-south shortening between 46.5°N and 48°N (∼5 mm/yr, Mazzotti et al., 2002; McCaffrey et al., 2013). However, little to no north-south shortening is observed north of 48°N and the residual velocities in this region often have uncertainties greater than their absolute values (Mazzotti et al., 2002; McCaffrey et al., 2013). Although the residual geodetic signal in northern Cascadia does predict some degree of permanent north-south shortening, it cannot resolve elastic or permanent strain rates on individual active faults identified in this region.

Regional instrumental crustal seismicity of southern Vancouver Island is concentrated beneath the Greater Victoria area and to the east beneath the San Juan Islands in the United States (Figure 1b; e.g., Bostock et al., 2019; Savard et al., 2018). Small to moderate earthquakes (up to M_w 4.1) predominantly have right-lateral, right-lateral oblique, and reverse focal mechanisms along roughly west-east to northwest-southeast striking fault planes (Brocher et al., 2017). North-south and northwest-southeast maximum horizontal stress directions (S_Hmax) have been inferred from focal mechanism inversion (Figure 1b, Balfour et al., 2011). Beneath Saanich Peninsula, double-difference relocated crustal seismicity has been interpreted to be localized on the steeply northeast-dipping northwest-striking LRF, which dips beneath the city of Victoria and toward the XELF (G. Li et al., 2018). Focal mechanisms of the largest relocated earthquakes on Saanich Peninsula also indicate right-lateral or oblique right-lateral slip with a reverse component on west-east to northwest-southeast striking fault planes (G. Li et al., 2018).

Recent paleoseismic studies indicate faults on southern Vancouver Island have hosted Holocene surface-rupturing earthquakes with magnitudes much larger than observed in the instrumental seismicity record. Lidar-derived high-resolution topographic models and paleoseismic trenching show that the LRF has hosted at least three M > 6 earthquakes in the last ∼9 kyr (Harrichhausen et al., 2021; Morell et al., 2018). Paleoseismic trenching also indicates oblique right-lateral Holocene slip on this structure, consistent with instrumental seismicity (Harrichhausen et al., 2021). The surface trace of the LRF continues east beneath the Strait of JdF (Figure 1b; Barrie & Greene, 2018), and may connect with the Devils Mountain fault (DMF) and the South Whidbey Island fault zone in Washington State. Continuous rupture of these connected structures could potentially generate a M > 7 earthquake near Greater Victoria, BC, Bellingham, WA, and Seattle, WA (Barrie & Greene, 2018; Harrichhausen et al., 2021; Morell et al., 2018).

The recent inclusion of the LRF, or a connected Leech River-Devils Mountain fault system, as a fault source in PSHA models for the Victoria region has resulted in an up to 23% increase in computed mean hazard across all spectral periods (Goda & Sharipov, 2021; Halchuk et al., 2019; Kukovica et al., 2019). The hazard increase is greatest for larger earthquakes at longer recurrence intervals, and for areas closest to the surface trace of the fault (Goda & Sharipov, 2021; Halchuk et al., 2019; Kukovica et al., 2019). This hazard increase highlights the importance of locating surface traces and determining the rupture histories of low strain-rate faults near plate boundaries, especially in urban areas.

In addition to onshore investigations of the LRF, bathymetric and seismic reflection studies suggest active faults are located in Haro Strait northeast of the city of Victoria. Greene and Barrie (2022) use multibeam bathymetric data to map a structure, the Central Haro Strait fault, trending northwestward from south of San Juan Island toward Saanich Peninsula near the XELF (Figure 1b). This structure is part of a network of parallel, northwest-striking bathymetric features in Haro Strait that join the west-east striking DMF zone to the southeast. Seismic reflection data are interpreted by Greene and Barrie (2022) to represent folded glacial deposits offset by a pair of steeply-dipping, northwest-striking faults. No direct observations of fault kinematics have been made on these structures but Greene and Barrie (2022) suggest left-lateral slip based on their interpretation of left-laterally offset bathymetric features along the DMF system to the southeast. However, the modern-day stress regime predicts right-lateral slip on steep west-east to northwest-southeast striking faults (Balfour et al., 2011), and paleoseismic studies suggest Holocene right-lateral kinematics of the DMF following a reversal of slip sense some time after the Eocene (Personius et al., 2014). Therefore, the left-lateral slip interpreted by Greene and Barrie (2022) may not represent active fault kinematics. Where the Central Haro Strait fault is mapped to join several other structures south of San Juan Island, seismic reflection data have been interpreted to represent northward verging reverse faults and fault duplexes, suggesting that the Central Haro Strait fault may have also accommodated a shortening component (Greene & Barrie, 2022).

3 Geologic Setting

The location and southeasterly strike of the XELF on Saanich Peninsula north of Victoria approximately coincides with a previously mapped bedrock fault that is part of a network of northwest-southeast-striking structures that accommodated crustal shortening during Cenozoic and Mesozoic terrane accretion (Figure 2a, e g., Canil & Morris, 2023; England & Calon, 1991; Massey, 1986; Muller, 1977; Seyler et al., 2022). This structure crosses the peninsula from Haro Strait in the southeast, to Saanich Inlet to the northwest, and continues to the northwest on Vancouver Island where it is mapped to be offset by several northeast-striking strike-slip faults (Figure 2a; Muller, 1983). The largest of the structures, the San Juan fault (SJF), is a major terrane boundary that accommodated left-lateral oblique slip during Eocene terrane accretion and formation of the northwest-southeast striking Cowichan fold and thrust belt (CFTB) (Figure 2a; England & Calon, 1991; Fairchild & Cowan, 1982; Harrichhausen et al., 2022). On Saanich Peninsula, the bedrock fault coinciding with the XELF juxtaposes intrusive rocks of the West Coast complex to the southwest against volcanic rocks to the northeast, both lithologies belonging to the Jurassic Bonanza island arc (Canil & Morris, 2023; Canil et al., 2013; DeBari et al., 1999; Muller, 1977, 1983). The relatively straight trace (Figure 2) of the bedrock fault suggests a subvertical dip at least near the surface (e.g., Muller, 1983). However, because the northwest-southeast strike of the fault is parallel with the CFTB located ∼10 km northeast (Figure 2a, England & Calon, 1991; England et al., 1997), it may be a related north-, or south-dipping Eocene reverse fault (e.g., Canil & Morris, 2023).

Saanich Peninsula's most recent geologic history has been dominated by Pleistocene glaciation. The sediments overlying bedrock on the peninsula predominantly record the penultimate glaciation, the Olympia Interglacial period between 57 and 30 ka, and the following Fraser glacial advance and retreat (e.g., Alley & Chatwin, 1979; Armstrong & Clague, 1977; Monahan & Levson, 2000). Prior to the Olympia Interglacial period, the penultimate glaciation deposited a variably thick package of lodgement till and isolated glacial-fluvial deposits along the eastern coast of Vancouver Island that are known locally as the Dashwood Drift (Armstrong & Clague, 1977; Hicock & Armstrong, 1983). At the end of Olympia Interglacial period when glacial advance commenced around 30 ka, up to 10-m-thick deposits of marine clay, silts, and sands of the Cowichan Head Formation were deposited and are exposed at isolated outcrops on Saanich Peninsula (Armstrong & Clague, 1977; Clague, 1980; Hebda et al., 2016). Southward glacial advance into the Puget Lowland reached its maximum extent at 14 ka during the Fraser Glaciation (Porter & Swanson, 1998; Thorson, 1989), and resulted in a 2-km-thick ice sheet above the Strait of Georgia and glacial loading of the land surface (James et al., 2000). The deposition of up to 50-m-thick well sorted cross-bedded sands (Quadra Sands) preceded the advance of the glaciers (Armstrong & Clague, 1977). The overriding ice sheet then deposited discontinuous lodgement or subglacial till (Vashon Drift), and formed drumlins and various other north-south oriented streamlined subglacial erosional landforms on Saanich Peninsula (Alley, 1979; Alley & Chatwin, 1979; Hicock & Armstrong, 1985). During and immediately following glacial retreat at 14 ka, the land surface was still depressed due to the lingering effects of ice loading, resulting in a relative high sea level stand up to 80 m above modern shorelines, and the deposition of clay- and silt-rich marine deposits known locally as the Victoria clay (Huntley et al., 2001; James et al., 2009; Mosher & Moran, 2001). Isostatic rebound, which occurs over longer time scales than deglaciation, eventually caused a drop in relative sea level, subaerial exposure of the land surface around XEOLXELEK between 12.2 and 12.0 ka, and restored relative sea level to its current elevation by 6 ka (Clague et al., 1982; James et al., 2002, 2009; Linden & Schurer, 1988; Mathews et al., 1970; Mosher & Hewitt, 2004).

4 Materials and Methods

We used publicly available airborne light detection and ranging (lidar) topography data from the BC Open Lidar Data Portal (link available in Data Availability Statement) to map geomorphic features on Saanich Peninsula that likely represent active faults. One-m-resolution bare-earth DTMs from grid cells 92B-053, 92B-054, 92B-063, and 92B-064 (GeoBC, 2020), funded by Public Safety Canada's National Disaster Mitigation Program, provided the basis for our analyses. Hillshade and slope maps were derived from these DTMs using QGIS software v. 3.22 (QGIS Association, 2018) to aid in visualization of topography. Our primary criteria for mapping fault related features was spatially-consistent offset of Quaternary surfaces and features such as glacial landforms and actively eroding channels. Bedrock scarps were also considered as potential markers of active faults, but only if they were along strike from offset Quaternary sediments and surfaces. This is evident in Figure 2b, which shows numerous lineaments and scarps in bedrock terrain on the western side of Saanich Peninsula that are not necessarily indicative of active structures. Instead, glacial scouring and plucking can preferentially erode preexisting bedrock fault zones resulting in pronounced topographic scarps (e.g., Hummingbird trench, Harrichhausen et al., 2021). To approximate the dip of the fault that formed the Quaternary scarp, we used structural contour mapping (e.g., three-point problem, Fossen, 2016) of the fault plane based on the elevation of the scarp across undulating topography (Figure S1 in Supporting Information S1).

We estimated vertical separation of across the XELK fault using three data sets: ground surface elevations near the trench site (profile P1), ground surface elevations across the displaced drumlinoid (profile P2), and the elevations of the top of the upper most displaced stratigraphic horizon mapped in the trench (EB2). We used a Monte Carlo routine to calculate vertical separation that accounts for uncertainty in the regression planes through the upper and lower offset surfaces (95% confidence interval), and uncertainty in the lateral position of the fault plane (e.g., Thompson et al., 2002). For each profile, we also averaged four to six versions of the Monte Carlo simulations with different criteria for possible surface projections of the fault and different regressions throughout the up-thrown and down-thrown surfaces, and propagate this uncertainty throughout our analyses (e.g., Regalla et al., 2022).

To complement the topographic analysis, we conducted reconnaissance mapping of bedrock fault outcrops and surficial geology along the projected surface trace of the XELF. We collected fault plane and slickenline orientations that were plotted and analyzed using the Stereonet 11.4.1 software (Allmendinger, 2020; Allmendinger et al., 2011; Cardozo & Allmendinger, 2013). Local surficial geology was mapped and compared with previous work (Monahan & Levson, 2000), and with the stratigraphy exposed in our paleoseismic trench, to determine potential age and origin.

To examine the history of recent earthquake ruptures, we excavated a 31-m-long, 1–2-m-wide, < 3-m-deep paleoseismic trench perpendicularly across the scarp on the eastern shore of XEOLXELEK at Eagle Beach (EB). We chose this location for a detailed paleoseismic excavation because its proximity to a depocenter provides the greatest potential for a high temporal resolution sedimentary record of deformation. It was also one of the few available locations on public land with minimal urban development. The perpendicular orientation is based on the predominantly vertical offset observed across the scarp, and the lack of clear horizontal offset of geomorphic features. The northern half of the trench encountered groundwater ∼1 m below the ground surface, which necessitated the installation of a pump to continuously remove water from the base of the trench and hydraulic shores stacked vertically at approximately 1 m intervals to ensure trench wall stability. After excavation, we installed a 1 m by 1 m grid and took overlapping digital photographs of the trench walls that were used to construct orthorectified photomosaics (e.g., Reitman et al., 2015) using Agisoft Metashape software (Agisoft, 2021). These photomosaics were used as the base for logging of the trench walls on digital tablets. Large format high-resolution photomosaics and trench logs of both trench walls are provided in Supporting Information S1 (Figure S2). Detailed structural geology data collected from the trench were plotted and analyzed using the Stereonet 10.1.6 software (Allmendinger et al., 2011; Cardozo & Allmendinger, 2013) and these structural data, along with trench unit descriptions, are provided in a data repository (Harrichhausen et al., 2023). Finally, we extracted the local grid coordinates of the upper contact of the youngest deformed unit in the trench and used the Monte Carlo routine described above to estimate the vertical separation of this unit.

In addition to the paleoseismic trench, we used an auger to excavate ∼10-cm-diameter holes up to a depth of ∼3.4 m, both from the floor of the trench and from ground surface alongside. We logged the sediments from these auger holes to map stratigraphy and interpolate unit contacts below the base of the trench. Detailed auger hole logs are shown in the data repository (Harrichhausen et al., 2023).

We collected 19 macroscopic charcoal samples for radiocarbon dating to constrain the sediment deposition and earthquake event ages in the paleoseismic trench. Samples were dated at the Keck Carbon Cycle AMS Laboratory in Irvine, California. We used OxCal version 4.4.4 (Bronk Ramsey, 2021) for radiocarbon calibration with the IntCal20 ¹⁴C production curve (Reimer et al., 2020) and report results as calendar calibrated ages before 1950 (cal BP) or thousands of calendar calibrated years before 1950 (ka). Tabulated raw and calibrated radiocarbon dates are in the Supporting Information (Table S1 in Supporting Information S1). After calibration, we used OxCal to incorporate all of the available chronological constraints with Monte Carlo routines and Bayesian statistics to model the probability distribution function (PDF) of geologic unit and earthquake event ages (e.g., Lienkaemper & Ramsey, 2009). By using different but equally geologically plausible sample combinations, we created five chronological models that constrain an earthquake event age. Further details behind the rationale for each model are explained in Section 5.2.2. Because each of these models are geologically plausible, we averaged the earthquake event age probability density functions (PDFs) from each model for a final event age (e.g., DuRoss et al., 2018). Earthquake event ages from the chronological models are reported as the 95% confidence interval, rounded to the nearest century in thousands of calendar years (ka) before 1950. We report the mode and 95% and 68% confidence intervals of the final age PDF.

To determine if the scarp is underlain by a geological discontinuity and to enable subsurface investigation below the base of the trench, ERT data were collected prior to excavation of the trench at EB (e.g., Improta et al., 2010; Mojica et al., 2017). We used an Advanced Geosciences Inc. (AGI) MiniSting™ R1 resistivity meter with a 28-electrode cable in dipole-dipole electrode configuration. Data were collected along a profile co-located with the eventual western wall of the trench and extending beyond the trench footprint to the northeast and southwest. First, an 81-m-long profile was surveyed, centered on the topographic scarp, with an electrode spacing of 3 m and resolvable depth of ∼16 m. This was followed by a 27-m-long profile nested within the longer line, with an electrode spacing of 1 m, to collect higher-resolution data of the shallow subsurface (to ∼5.5 m depth). Relative ground elevations at each electrode position were surveyed with a Spectra Precision Focus 6 5” Total Station. Using AGI's EarthImager™ 2D software (Advanced Geosciences Inc, 2009), the apparent resistivity data were inverted using a damped least-squares method, for five iterations, to produce resistivity models of the subsurface that minimized the root-mean-square misfit between the model-predicted and measured values. The resistivity data files are provided in the data repository (Harrichhausen et al., 2023).

After determining that fault-related deformation was visible in the ERT data at EB, we also conducted an additional 3-m-spaced resistivity survey on the northwestern shore of XEOLXELEK at Waterski Beach along strike from the trenched scarp at EB (Figure 2b). At this location, there is no clear scarp visible in the DTM, but two prominent bedrock lineaments continue to the northwest. The profile was oriented 63° relative to the fault trace in order to fit the survey in between the lakeshore and a gravel pathway.

Finally, we used the forward modeling program FaultFold version 7 (Allmendinger, 1998, 2019; Zehnder & Allmendinger, 2000) to recreate the geometry of the fault and fold observed in the paleoseismic trench and geophysical surveys at EB and test their consistency with a geologically plausible fault-propagation fold model (e.g., Livio et al., 2020). We assume originally sub-horizontal beds before deformation and employed a trishear deformation model, based on the observed fault orientations, to deform the stratigraphy until the fault and fold geometry were approximately replicated (e.g., Allmendinger, 1998; Erslev, 1991). The parameters controlling the deformation model are detailed in Section 6.1.

5 Results

5.1 Geomorphic and Geologic Evidence for Fault Rupture

On Saanich Peninsula, where bathymetric features in Haro Strait project onshore (Figure 1; Greene & Barrie, 2022), topographic scarps suggestive of recent deformation are visible in the lidar DTM (Figure 2). A hillshaded DTM shows a ∼125°-striking, northeast-facing scarp formed in Quaternary sediments between the eastern shore of XEOLXELEK and Haro Strait, and multiple northeast-facing bedrock scarps along strike to the northwest between XEOLXELEK and Saanich Inlet (Figure 2b). The Quaternary and bedrock scarps generally coincide, but are slightly misaligned, with the previously identified more easterly-trending bedrock fault identified on British Columbia Geological Survey maps (Muller, 1983).

East of XEOLXELEK, the scarp cross-cuts a ∼1–1.5-km-wide, ∼40-m-high, north-south oriented, streamlined ridge (Figure 2b). We interpret this feature to be of glacial origin because it is underlain by thick deposits of glacial-fluvial sands and cobble gravels (Figure 3b; Monahan & Levson, 2000), and because of its elongate streamlined shape parallel with known glacial flow directions (Huntley et al., 2001). Because this feature lacks an up-ice face that is steeper than the down-ice face as would be typical of a drumlin (e.g., Clark et al., 2009), we refer to it as a drumlinoid (e.g., Monahan & Levson, 2000). Vertical separation of the drumlinoid's modern surface is 2.5 ± 0.2 m (Profile 2, Figures 3b and 3c). On the eastern half of the drumlinoid ridge, a gully formed by modern stream incision into well-sorted sands and gravels has an apparent 26.5 ± 1.5 m left-bend where it coincides with the Quaternary scarp (Figure 3b). However, left-laterally offset geomorphic features are not apparent along the rest of the scarp and the apparent offset is more likely to be an isolated erosional landform as opposed to an indication of fault kinematics. The scarp can be traced southeastward from the base of the eastern drumlinoid slope to the shoreline of Haro Strait, but urban development precludes a reliable measurement of surface offset (Figure 3a). The surface trace across the 40-m-high drumlinoid allows a three-point problem estimate of the strike and dip given the scarp is the manifestation of slip on a plane. Given this assumption, structural contours at elevations of 60 and 90 m above sea level reflect a fault plane attitude (strike/dip, dip direction) of ∼127°/50° southwest (Stereonet, Figure 2b; Figure S1 in Supporting Information S1).

On the western edge of the drumlinoid the scarp projects into XEOLXELEK, and at the lake's shoreline at EB it vertically offsets the ground surface 1.1 ± 0.2 m (Profile 1, Figures 3b, 3c, and 4a). This vertical offset results in the formation of a small, <1-m-high wave-cut embankment of consolidated clay and silt, along the up-thrown portion of the lakeshore. North of the scarp, on the down-thrown portion of the lakeshore, anthropogenic fill has been used to form a beach. This modification is apparent in air photos, which show a much more pronounced lineament at the trench site in 1926 CE before the addition of beach fill (Figure 4). The presence of anthropogenic fill at the base of the scarp along the lakeshore also explains the lesser vertical separation observed at Profile 1 than Profile 2, and indicates that this measurement should not be used to estimate paleo-earthquake parameters.

Figure 4

Air photos of the Eagle Beach (EB) trench site on XEOLXELEK (see Figure 3b for location). (a) Oblique view of the EB trench site and scarp on the eastern shore of XEOLXELEK during paleoseismic study in August 2021 (photo by A. Schaeffer, Natural Resources Canada). (b) 1926 air photo showing pronounced lineament (black arrows) and indentation of the shoreline at EB along the surface trace of the XELF. Photo from National Air Photo Library (1926). (c) 1954 air photo showing anthropogenic modification of EB obscuring the topographic scarp (black arrows). Photo from GeoBC (1954).

In addition to the scarp across the drumlinoid, there is also evidence for along-strike bedrock faulting to both the northwest and southeast on Saanich Peninsula. At Haro Strait to the southeast of the drumlinoid, the Quaternary scarp projects toward a southeast-striking portion of the modern coastline (Figure 2b). The coast here is formed by a ∼10–40-m-high wave-cut bluff. The northwestern end of the bluff (Outcrop 1, Figure 2) exposes a southwest-dipping bedrock fault zone cross-cutting fine-grained light green volcanic rocks, likely of the Bonanza Group (e.g., Canil & Morris, 2023). The fault zone consists of two subparallel fault strands with orientations of 102°/45° south and 115°/62° southwest. The average orientation of these faults, 109°/53° southwest, is consistent with that of the Quaternary scarp-forming fault determined from our three-point problem (∼127°/50° southwest), but is slightly more easterly-striking. The similar attitudes and along-strike locations suggests this bedrock structure may be a part of the fault zone that formed the scarp.

Along strike northwest of XEOLXELEK the Quaternary scarp projects toward several subparallel topographic lineaments (Figures 2b and 5a). These lineaments are formed by ∼5–20-m-high, northeast-facing bedrock scarps that strike northwest-southeast to west-east. We observed brittle bedrock faults along these topographic lineaments. For example, at Outcrop 2, there is a ∼2–15-cm-thick sandy-gouge filled fault with an attitude of 075°/36° southeast. Fault plane slickenlines have a rake of 95° and tension fractures (e.g., Doblas, 1998) indicating reverse slip (Figure 2b). Both the hanging wall and footwall are fine-grained dark-green basalt of the Jurassic Bonanza Group. Where these lineaments project through Quaternary sediments and into XEOLXELEK, there is no obvious topographic scarp visible in the DTM. However, there is an ∼60-cm-increase in elevation of the modern ground surface from north to south along the lakeshore (Figure 5b). The scarp at this location may be subdued due to colluvial processes from the slope on the western shore of the lake, or via sedimentation from streams and paleochannels entering XEOLXELEK from the northwest (Figure 5a). The absence of a visible scarp farther up the slope to the west might be explained by the fact that this ridge is the site of an active farm where cultivation may have smoothed the land surface.

Figure 5

(a) Hillshaded digital terrain model (DTM) of Waterski Beach where the Quaternary scarp projects across XEOLXELEK, and two prominent bedrock scarps continue to the northwest. No Quaternary scarp is evident in the hillshaded DTM. Surficial geology simplified and modified from Monahan and Levson (2000). Electrical resistivity tomography (ERT) survey line is shown in black. Contour spacing is 5 m. (b) 3-m-spaced ERT profile across the suspected trace of the XELF where there is a minor (∼60 cm) vertical offset of the ground surface. A 38° ± 10° southwest-dipping discontinuity vertically offsets the top of high-resistivity layer 3 (HR3) by minimum of 2.8 m. Given the obliquity of the ERT profile with respect to the strike of the XELF, we estimate a true dip for the discontinuity of 41° ± 10°.

5.2 Paleoseismic Trench

5.2.1 Stratigraphy and Deformation

The paleoeseismic trench, excavated perpendicularly across the scarp at Eagle Beach (EB) (Figure 4a), revealed two folded and faulted glacial and glacial-marine sedimentary units (Units EB1 and EB2) (Figure 6), overlain in turn by a colluvial wedge (Unit EB3), beach sand (Unit EB4), and anthropogenic fill (Units EB5 and EB6). The base of the trench between gridlines 9 and 14.5H and auger holes north of 15H expose a consolidated matrix-supported diamict suggestive of glacial till (Unit EB1). This till is overlain by a folded ∼1.8-m-thick sequence of interbedded clay and silt (Unit EB2), which auger hole data show to continue below the base of the trench north of 19.5H. Regular beds of clay, silt, and fine sand are observed in Unit EB2. Bedding is thicker (<22 cm beds) at the base of the unit, transitioning to ∼1- to 3-cm-thick beds in the middle, and sub-millimeter thick beds in the upper third of the unit. The presence of these regular beds, or rhythmites, and abundant silt and clay indicate Unit EB2 is a glacial-lacustrine/marine sequence, and a part of the glacial-marine clay unit (QL) mapped immediately south of the scarp at the trench site (Figure 3b). A B-horizon paleosol (EB2b), distinguished by increased fracture density, oxidation (red-coloration), and break up of the sediments into regolith is formed along the entirety of the upper ∼0.5–1 m of Unit EB2. The base of EB2b is undulatory with the deeper parts of the paleosol located where there is an increased density of fractures and faults in unit EB2 (Figure 6a).

Figure 6

(a) Photo orthomosaic and interpretation of the east wall of the Eagle Beach paleoseismic trench, from 6 to 24H, centered on the monocline and south-dipping reverse faults. Contacts below trench floor were inferred using unit contacts in the auger hole data. Note that some of the radiocarbon samples are from the west wall of the trench and not shown here. Vertical offset is based on six Monte Carlo simulations of different possible surface regressions of the upper cotact of EB2 (average shown) and fault locations. Lower hemisphere equal area stereonet projection shows planes and poles to planes measured from both east and west trench walls, and a cylindrical best fit great circle to the poles to bedding. Reverse fault planes and poles to planes are shown in red. See Figure S2 in Supporting Information S1 for complete photo orthomosaics and interpretations of both trench walls, including the locations of bedding and fault plane measurements. (b) High resolution photo orthomosaic and (c) interpretation of the reverse fault exposed in the east wall of the trench between 14 and 15H. Same legend as (a).

Units EB1, EB2, and the paleosol formed in EB2 have been folded by an asymmetric, north-dipping, fault-cored monocline that defines a fold scarp centered at 14.5H (Figure 6). Bedding is gently north-dipping within most of the fold, except for at the middle of the monocline at the base of Unit EB2 (15H, 0.5 V), where it may be steepened by fault slip. Poles to bedding measurements at the Unit EB1-Unit EB2 contact, and from within Unit EB2, have a cylindrical best fit great circle (beta plane) attitude of 011°/88° east (Stereonet, Figure 6), with a pole of 02°/281° (plunge and trend) and corresponding with a ∼281°- or 101°-striking fold axis. The monocline has resulted in 2.0 ± 0.1 m of vertical separation of the upper contact of Unit EB2. This measurement assumes little to no erosion of the top of the up-thrown block. We make this assumption based on the similar thickness of both Unit EB2 and the paleosol (EB2b) in both the down- and up-thrown blocks away from the moncoline (∼2.2 m). There is, however, evidence of ∼0.35 m of erosion of Unit EB2 across the core of the monocline, where the unit thins to a minimum ∼1.85 m at 14H. The lower estimate of vertical separation across Unit EB2 is slightly less than that of the drumlinoid surface ∼500 m southeast of the trench site (∼2.3 m, Figure 3).

The gently south-dipping reverse fault at the center of the monocline and below the scarp offsets Units EB1 and EB2 (14.5H, 0.25 V, Figure 6). The reverse fault is composed of two subparallel, en-echelon segments spaced ∼20 cm apart. The segments are <0.3 cm thick, composed of gray clay, and have an apparent reverse offset of marker beds in Unit EB2, and the EB1-EB2 contact of 10.7–16.7 cm. The paleosol horizon that has brecciated and altered the top of Unit EB2 obscures any observations of faults at the top of the unit (15H–17H, 1V–2V, Figure 6). We projected the reverse fault between the east and west walls of the trench to determine fault orientations of 085°/19° south and 090°/24° south (red great circle in stereonet, Figure 6).

Multiple subvertical fault and fracture zones have also deformed Unit EB2 within and south of the monocline (Figure 6). These structures are ∼11–27-cm-wide brecciated fault zones at the top of Unit EB2, narrow to a width of ∼1 cm toward the base of unit, and do not propagate below the Unit EB1-Unit EB2 contact. Vertical offsets are minimal and consist of ∼2–9 cm of apparent normal displacement. Several of the structures resemble small grabens suggesting they are the result of localized extension by bending of Unit EB2.

Unit EB3, which overlies the deformed glacial sediments of Unit EB2 between 16.5 and 23.5H, is a diamict that formed after deformation of the lower units (Figure 6). The unit is a channel- or wedge-shaped body of poorly-sorted, matrix-supported diamict containing clasts of Unit EB2, and lacking any internal stratification. These observations and the location of Unit EB3 at the base of the fold scarp formed by Unit EB2 imply it is composed of colluvium sourced from the scarp, or a colluvial wedge. There is a <10-cm-thick paleosol horizon at the top of Unit EB3, suggesting that this was the ground surface for a period of time. The lack of an internal paleosol or other stratification within the colluvial wedge indicates that only one major earthquake resulted in the deformation observed in the trench at EB, and that the base of EB3 represents the event horizon for this single earthquake.

Unit EB4 is a <30-cm-thick layer of moderately-sorted silt to coarse sand that overlies Unit EB3 between 16.5 and 25H (Figure 6). A <5-cm-thick organic-rich paleosol has also formed at the top of this unit suggesting it was exposed at the ground surface for a period of time. Unit EB4 thins toward the south where it is draped against the underlying colluvium. Where EB4 is thicker (∼30 cm at 21H), it has cm-thick laminations displaying cross-bedding. The geometry and the composition of Unit EB4 probably reflects its deposition as beach sediment that on-lapped the lower part of the fold scarp.

Two layers of anthropogenic fill have been deposited on top of Units EB2 and EB4 north of 10.5H (Figure 6). Unit EB5 and Unit EB6 contain glass fragments, bricks, and other modern anthropogenic material indicative of construction fill. These observations are consistent with air photos showing the northern portion of the trench was underwater in 1926, and was filled during subsequent construction of the modern beach sometime prior to 1954 (Figure 4). After the emplacement of the fill, an ∼8- to 20-cm-thick modern A-horizon soil has formed right below the modern ground surface at the top of the trench walls.

5.2.2 Radiocarbon Dating

Macroscopic charcoal samples were collected from the non-anthropogenic stratigraphic units in the trench to constrain the age of the single earthquake event horizon marked by the contact between the colluvial wedge (Unit EB3) and Unit EB2b. Radiocarbon dating of 19 charcoal samples from Units EB1–EB4 yielded a range of median ages from 49.8 ka in the basal Unit EB1 to modern ages in Units EB3 and EB4 (Figure 7, Table S1 in Supporting Information S1).

Figure 7

(a) Calibrated radiocarbon ages shown relative to trench stratigraphy. 95% confidence intervals of ages for each sample are shown. Bold and underlined samples and ages are used in the chronological models in (b). (b) Five geologically plausible chronological models of earthquake event ages. (c) Earthquake event age probability density functions (PDFs) from each of the five models and the 95% confidence interval of each earthquake age. Bottom probability distribution function is the average of the earthquake age PDFs with the mode, and the 95% and 68% confidence intervals of the age shown.

The oldest unit, Unit EB1, contained two samples, C27 and C25, with calibrated ages between 53 ka and 25 ka. The consolidation and high silt content of the diamict is suggestive of glacial till deposited either before the Olympia Interglacial and during the penultimate glaciation (Dashwood Drift, Hicock & Armstrong, 1983), or after the Olympia Interglacial during the Fraser Glaciation (Vashon Drift, Hicock & Armstrong, 1985). The former interpretation implies that the youngest sample, C25 (25–31 ka) was introduced by bioturbation, as it is younger than the proposed age of the penultimate glaciation (Cosma et al., 2008). On the other hand, the latter interpretation implies that the oldest charcoal sample, C27 (>47 ka), is recycled from older sediments.

Unit EB2 contained two older-aged samples, C22 (31.7–31.1 ka) and C13 (45–52 ka) and its stratigraphic position above a glacial till (EB1) demonstrates that it was deposited after or during deglaciation. This implies that EB2 is either part of the Cowichan Head Formation deposited at the end of the Olympia Interglacial (e.g., Armstrong & Clague, 1977), or the post-glacial Victoria clay, both of which locally drape over till and bedrock on Saanich Peninsula (Huntley et al., 2001; Mosher & Moran, 2001). The interpretation that this unit represents the Cowichan Head Formation (∼42 to 25 ka, Armstrong & Clague, 1977) fits the sample ages better, although C13 is too old, while the latter interpretation implies both samples are reworked from older deposits.

Radiocarbon ages from the top of the Unit EB2, C23 (1.2–1.0 ka) and C92 (2.7–2.5 ka) are from within the paleosol (Unit EB2b, Figure 6), and reflect more recent soil formation after subaerial exposure of EB2. Because Unit EB2b, along with EB2, was deformed by the earthquake recorded in the trench, these ages also reflect the age of the youngest deformed unit.

Above the earthquake event horizon, 10 samples from Unit EB3, the colluvial wedge, yielded radiocarbon ages that clustered into an older group with ages ranging from 4.4 to 1.2 ka, and a younger group with ages between 290 cal BP and modern. The younger samples could reflect modern soil formation and bioturbation. The three samples in Unit EB4 above EB3 similarly range in age from 307 cal BP to modern. These EB4 ages are consistent with deposition of lacustrine sands following the damming of XEOLXELEK in 1879 CE and the subsequent rise in water levels (Bulkley, 1872; Pearson, 1981).

For all our chronological models, which constrain unit and earthquake ages, we excluded 11 of the 19 radiocarbon ages (Figure 7a) based on the following criteria and assumptions. We exclude radiocarbon ages older than 12.1 ± 1 ka and use the trench site's emergence above sea level as the oldest boundary for each model (Figure 7b). We use the drop in relative sea level and subaerial exposure between 12.2 and 12.0 ka (James et al., 2009) as the maximum earthquake age for two reasons. First, we consider the earthquake must be post-glacial because the scarp deforming the surface of the drumlinoid would have been eroded by an overriding glacier if it had not formed after glacier retreat. Second, the deformed paleosol on Unit EB2 must have developed while the surface of the deposit was subaerial, thus deformation occurred after sea level dropped below the elevation of the trench site. In addition to the samples older than 12.1 ka, we have also removed sample C23 (1.2–1.0 ka) in Unit EB2 as it is stratigraphically below several significantly older samples in the colluvial wedge (e.g., ∼3.1 ka) (Figure 7a), and it is close to the modern ground surface (Figure 6). We removed the oldest sample (C115: 4.4–4.2 ka) from the colluvial wedge (Unit EB3) as it is stratigraphically above several younger samples and it is significantly older than sample C92 (2.7–2.5 ka) from Unit EB2 below, suggesting it has been recycled. We have also excluded the two youngest samples (C103: 290 to modern and C52: 270 to modern) at the top of the colluvial wedge as they are located at the contact with Unit EB4 and within the paleosol formed in the colluvial wedge. Finally, we consider the permanent settlement of Fort Victoria in 1843 (Plasterer, 1967) as the minimum earthquake age in our chronological models as there is no record of a large earthquake after Fort Victoria's establishment. Therefore, the samples in the lacustrine sand (C63: 294 cal BP to modern, C72: 285 cal BP to modern, C47: 307 cal BP to 15 cal BP), which likely formed subsequent to the damming of XEOLXELEK in 1879 and after the establishment of Fort Victoria, have also been removed from our models (Figure 7b).

Different combinations of the remaining eight radiocarbon ages in geologically viable chronological models constrain five different earthquake ages (Figures 7b and 7c). For chronological models 1–3, we consider that sample C92 at the top of Unit EB2 and below the colluvial wedge (Unit EB3), constrains the maximum age of the earthquake. In model 1, we constrain the minimum age of the earthquake by only considering the samples from the colluvial wedge with ages younger than Unit EB2 (<2.5 ka), and older than samples from overlying Unit EB4 (>300 cal BP), assuming all other ages are from recycled charcoal or result from bioturbation. This model yields an earthquake age of 2.7 to 2.5 ka. In model 2, sample C62 is also removed from the colluvial wedge (Unit EB3) resulting in a younger, less tightly constrained earthquake age of 2.6 to 1.3 ka. Chronological model 3 considers all of the samples older than 1.0 ka in the colluvial wedge (Unit EB3) are recycled and only uses the youngest ages to constrain the minimum earthquake age, resulting in a more broadly constrained earthquake age of 2.6 to 0.3 ka.

In models 4 and 5, we assume sample C92 at the top of Unit EB2 was emplaced by bioturbation after the earthquake and the maximum earthquake age is instead constrained by deglaciation at 12.1 ± 0.1 ka (Figures 7b and 7c). In model 4, we include sample C57 at the base of the colluvial wedge (Unit EB3, Figure 7a), resulting in an earthquake age of 7.6 ± 2.6 ka, whereas in model 5, this older colluvial wedge sample (C57) is removed resulting in an earthquake age of 7.4 ± 2.7 ka.

As each of our chronological models are geologically permissible, we averaged the PDFs from each model to determine a final age distribution for the single earthquake recorded in the paleoseismic trench (Figure 7c, e.g., DuRoss et al., 2018). The resulting PDF for the earthquake age is a non-normal distribution with a mode of 2.7 ka, relatively high probability densities toward ∼1.0 ka, and a low probability-tail stretching back to deglaciation. The age range based on the 95% confidence interval is 10.9 to 0.8 ka, while the age range based on the 68% confidence interval is 4.7 to 2.3 ka.

5.3 Geophysical Evidence of Subsurface Deformation

A 27-m-long 1-m-spaced ERT profile co-located with the paleoseismic trench, and an 81-m-long 3-m-spaced ERT profile starting 27 m southeast of and parallel with the trench, are also indicative of folded and offset stratigraphy (Figure 8). Both profiles exhibit a <3-m-thick low-resistivity layer (LR1, <40 Ω-m) right below ground surface on the up-thrown, south side of the scarp. The 1-m-spaced ERT profile shows LR1 is gently folded into a north-dipping monocline at the scarp, and is below a <3-m-thick high-resistivity layer (HR1, >100 Ω-m) on the down-thrown side of the scarp (Figure 8). The resistivity of LR1 is consistent with clay-sand (Palacky, 1988), and corresponds spatially with the glacial-marine clay and sand of Unit EB2 exposed in the trench (Figure 8). The upper ∼1 m of this layer between 0 and 9 m along the 1-m-spaced profile has a higher resistivity, ∼100–270 Ω-m, and spatially correlates with the paleosol developed on Unit EB2. HR1, above LR1 north of the scarp, corresponds with the colluvial wedge (EB3), lacustrine sand (EB4), and layers of anthropogenic fill (EB4 and EB5) exposed in the trench.

Figure 8

1-m-spaced (a) and 3-m-spaced (c) electrical resistivity tomography (ERT) profiles across the north-facing scarp on the eastern shore of XEOLXELEK. (b) Interpretation of (a) overlain by a simplified trench log and the interpreted subsurface location of the stratigraphic units. High-resistivity Layer 1 (HR1) correlates with the colluvial wedge (EB3), lacustrine sand (EB4), and anthropogenic fill (EB4 and EB5). Low-resistivity Layer 1 (LR1) correlates with the glacial-marine clay (EB2). (d) Interpretation of (c) with trench location shown. High-resistivity layer 2 (HR2) is interpreted to represent basalt bedrock or very consolidated sediments that are vertically offset beneath the topographic scarp along a linear feature that dips between 45° and 55° toward the southwest. Low-resistivity unit 2 (LR2) is interpreted to represent a buried pipe aligned perpendicular to the ERT profile.

The 3-m-spaced ERT profile, which resolves to a greater subsurface depth than the 1-m-spaced profile, shows evidence of offset stratigraphy below the depth of the paleosesismic trench. This profile indicates another relatively high-resistivity layer (HR2, >∼100 Ω-m) ∼2 m beneath LR1 that extends to depth along the majority of the profile (Figure 8c). We interpret that the top of Unit EB1 is located in the ∼2-m-thick space separating LR1 and HR2, and that HR2 likely represents either basalt bedrock that is mapped nearby (Figure 3b), or increasingly consolidated till based on its extension to depth and its high resistivity values (e.g., ∼800 Ω-m, Palacky, 1988; Prieto et al., 1985). The 3-m-spaced profile shows vertical separation of the top surface of HR2 of ∼2.7 m immediately beneath the scarp. There is a linear break in HR2 between 32 and 42 m along profile, which we interpret to represent a 45°–55° southwest-dipping reverse fault that accommodated the ∼2.7 m vertical separation of this layer (Figure 8d). Between 51 and 60 m along the 3-m-spaced profile, approximately 10 m north of the scarp, there is a circular low-resistivity feature (LR2, <100 Ω-m) centered at a depth of 4.3 m, and extending toward the base of the profile (Figure 8c). A utility near this location, and the circular shape suggests that this feature could represent a large pipe buried in the clay (EB2) beneath the anthropogenic fill that is oriented perpendicular to the profile and drains into XEOLXELEK.

The 3-m-spaced ERT profile at Waterski Beach on the northwestern shore of XEOLXELEK shows evidence of a Quaternary structure that likely represents the same fault evident at EB. A low resistivity layer (LR3, 30–60 Ω-m) occurs at the surface on the north end of the survey, and several soil samples taken from the shallow surface at the north end of the survey revealed dark brown clay-rich sediment with rare pebble clasts (Figure 5). This evidence shows LR3 is a glacial-marine clay, similar to LR1/EB2 in the ERT profiles and paleoseismic trench at EB and consistent with Palacky (1988). A high resistivity layer (HR3, 300–700 Ω-m) occurs at the surface on the south end of the survey. Several soil samples taken from the shallow surface at the south end of the survey revealed a pebble-cobble gravel with rare boulder-sized clasts supported by a light brown fine sand and silt matrix indicating HR3 is likely a well-consolidated glacial diamict, similar to Unit EB1 at EB, and consistent with regional mapping (Unit QF, Figure 5a). No bedrock outcrops are observed in the immediate vicinity of the profile, so we rule out HR3 as bedrock because it is clearly at the surface in the ERT profile.

Below LR3 and HR3, we cannot make geologic interpretations but we observe evidence for a geologic structure. Below LR3 is another high resistivity layer (HR4, 200–700 Ω-m) that may be correlative to HR3 and is vertically offset by at least 2.8 m along a 38° ± 10° southwest-dipping discontinuity, potentially a fault, defined by a slight decrease in resistivity (300 Ω-m) in between HR3/5 and HR4. Given that the profile is not oriented perpendicularly to the strike of XELF, this dip would represent an apparent dip of this structure and the true dip would be 41° ± 10°. Additionally, the vertical separation is a minimum estimate because the top of the up-thrown block (HR3) may have been beveled by erosion; however, it is consistent with the vertical separation of the high-resistivity layer HR2 at EB. There is another high-resistivity layer (HR5, 200–700 Ω-m) beneath HR3. The thickness of the high-resistivity layers appears to change across the suspected fault and the base of HR4 and HR5 do not have a distinctive offset. Another plausible interpretation is that the sharp step between HR3 and LR3 represents a former wave-cut shoreline, and LR3 is filling a depression. However, as the profile is along strike of the clear southwest-dipping reverse fault at EB, Holocene deformation is a possible explanation for the discontinuity.

6 Analyses

6.1 Fault-Propagation Fold Model to Constrain Fault Geometry and Slip

Constraining the fault dip of the XELF is essential for estimating net slip and thus the magnitude of the earthquake recorded in the trench (e.g., Allmendinger & Shaw, 2000). Bedrock faults at Outcrop 1 (Figure 2), topography (Figure 3b, Figure S1 in Supporting Information S1), and the EB ERT profiles (Figure 8) indicate the XELF dips between 45° and 62° to the southwest, whereas Outcrop 2 (Figure 2) and the paleoseismic trench (Figure 6) show more gently dipping reverse fault orientations (dipping 19°–38°). At Waterski Beach, fault dip is more poorly constrained (true dip of 31°–51°, Figure 5b), but consistent with most of the other data. At EB, where we have the best constraints on fault structure, the trace of the Quaternary scarp and the ERT profile both provide evidence for a more steeply-dipping structure at depth. Meanwhile, the monocline and minor reverse faults exposed in the trench (Figure 6) look genetically similar to fault-propagation folds that form above a propagating reverse fault tip (e.g., Chen et al., 2007; Livio et al., 2009; Moore et al., 2022; Shaw & Shearer, 1999; Suppe & Medwedeff, 1990). Therefore, we interpret that steeper structure observed beneath the EB trench in the ERT profiles (Figure 8) was a propagating reverse fault that formed the fault-propagation fold and related faulting exposed in the trench. This hypothesis is consistent with our observations of a range of fault dips. It also implies that the steeper dip of the fault at depth, which controls the earthquake rupture and subsequent deformation, should be used in fault slip estimations.

To test whether the deformation observed in the EB trench is consistent with slip along a blind reverse fault and to quantify this slip, we recreated the observed fault and fold geometry that would result from trishear deformation in front of a propagating fault using forward modeling software (e.g., Livio et al., 2020). We use a fault dip of 50° consistent with both our three-point problem estimate (Figure S1 in Supporting Information S1) and the ERT profiles (Figure 8). The other main parameters the model uses to control the geometry of the fault-propagation fold are the fault slip, the fault propagation to slip ratio (P/S), and the trishear angle (TS), which defines the area in front of the propagating tip where deformation is focused (Erslev, 1991). The physical properties controlling these last two parameters, which are governed by the rate of crack tip propagation and the effective viscosity of the deforming medium, are difficult to independently constrain in natural media (Allmendinger, 1998; Allmendinger & Shaw, 2000; Livio et al., 2020). Therefore to simplify the model, we kept a constant TS of 60° and varied the P/S. The P/S is a function of the mechanical properties of the material in which the fault propagates through and the fluid overpressure (Allmendinger, 1998), and therefore varies within each geologic unit. Variation in P/S between different units has been suggested to have a strong control on surface rupture geometry (Livio et al., 2020). We tested different P/S ratios at 0.2 increments for each geological unit (between 7 and 10.4 for bedrock, and 1.4 and 5 for the glacial sediments) until we modeled a geometry that best fit our trench and ERT profile observations.

The model that best fit our trench observations is shown in Figure 9. In our forward model we defined a horizontal, 2.2-m-thick bedded unit, with internal contacts at 0.5, 1.0, and 1.4 m above 0 m in our model reference frame, mimicking bedding within Unit EB2, the youngest deformed unit observed in the trench. We initiated a 50° southwest-dipping fault 16.5 m below the base of the modeled Unit EB2 and allowed this fault to accumulate 1.8 m of reverse slip while propagating upwards toward Unit EB2 with a P/S of 10. The high P/S is indicative of a rapidly propagating fault tip in a relatively rigid material (e.g., bedrock). Fault slip and propagation results in the formation of a gentle monocline in the modeled Unit EB2 (Figure 9a). When the fault tip reached the top of HR2, observed in the ERT profiles (Figures 8c and 8d), we reduced the P/S to reflect a change in wallrock to a less rigid material such as sediment (e.g., Units EB1 and EB2). The P/S reduction is consistent with slower fault tip propagation in less consolidated sediments closer to surface (e.g., Livio et al., 2020). We found a P/S of 3.4 resulted in the fold that best fit the observed geometry, and is similar to other P/S values modeled for silts and clay (e.g., P/S: ∼2.9, Livio et al., 2020). We then allowed the fault to accumulate another 1.4 m of reverse slip while the fault propagated to the base of Unit EB2, resulting in further growth of the monocline (Figure 9b). Finally, to account for the observed gently dipping reverse faults observed in the trench (Figure 6), we initiated a late-stage 10° southwest-dipping fault near the EB1–EB2 contact and the tip of the main structure. We use a lesser dip than what we measured in the trench (19°–24°), to match the apparent dip of the reverse faults in the trench log. This fault accumulated 0.2 m of reverse slip, based on the observed offset of bedding in Unit EB2 (Figure 6), with the same P/S of 3.4 (Figure 9c). Although the parameters we use in our model are a non-unique solution for reproducing the observed deformation, they show that ∼3.2 m of reverse slip on an ∼50° southwest-dipping fault can produce faults and folds (Figure 9d) that are consistent with our paleoseismic trench and bedrock outcrop observations, and interpretations of our ERT profiles.

Figure 9

Forward trishear model of fault slip and fault-propagation folding of Unit EB2. Strain ellipses show concentrated deformation within the trishear deformation zone (gray lines emanating from fault tip). (a) Initial 1.83 m of reverse fault slip and propagation through bedrock (or other very consolidated strata) with a propagation to slip ratio (P/S) of 10 and trishear angle (TS) of 60°. Fault location and dip (50° southwest) based on electrical resistivity tomography profile interpretation (Figure 8d). (b) After 1.83 m of slip, the fault tip has propagated to Unit EB1 approximately at −2 m on the vertical figure grid (based on Figure 8d). The P/S changed to 3.4, 2.37 m of fault slip (total slip of 3.2 m) is accumulated, and the fault tip propagated to the EB1–EB2 contact. (c) Final result of model after minor late-stage 10° southwest-dipping fault accommodates ∼0.2 m of slip during folding of EB2. (d) Overlay of Unit EB1, EB2, and EB3 from the trench log (Figure 6) on the final result of the forward trishear model (c).

Assuming dip slip, fault throw can also be estimated using fault dip and vertical separation of Unit EB2 or vertical separation estimated by doubling the colluvial wedge thickness. For a 45°–55° dipping fault plane (Figure 8d), ∼2.3 to ∼3.0 m of dip slip is required for ∼1.9–2.1 m of vertical separation of Unit EB2 (Figure 6), less than the ∼3.2 m estimated from our fault-propagation fold model. The maximum colluvial wedge thickness, proximal to the scarp, has been suggested to approximate half of the vertical separation at fault scarps (e.g., Bennett et al., 2018; Klinger et al., 2003; McCalpin, 2009). The maximum colluvial wedge thickness is ∼0.6 m for the east wall of the trench and ∼0.75 m for the west wall (Figure S2 in Supporting Information S1), which corresponds to a vertical separation of ∼1.2–1.5 m. Considering the range in fault dip from 45° to 55° this vertical separation corresponds to a dip slip of ∼1.5–2.1 m, which is less than the fault slip estimated using the previous methods. These data suggest that colluvial wedges that form at the toe of fold scarps may be thinner than colluvial wedges that form at the toe of fault scarps with similar vertical separation. This is likely due to the broader zone of deformation at fold scarps and their lack of abrupt fault scarp free faces. As such, colluvial material at fold scarps is eroded and transported across more gentle slopes by different and more gradual processes than at fault scarps, and likely requires more time to fill the accommodation space created on the down-thrown side of the fold scarp. Thus, the thickness of colluvial wedges formed along fold scarps may be substantially less than half of the vertical separation.

The fault-propagation fold model also predicts the 2D finite strain field. Strain ellipses resulting from the modeled deformation of Unit EB2 show strain is primarily localized in the trishear deformation zone in front of the propagating fault tip (Figure 9). Horizontal, trench-parallel shortening is concentrated near the base of Unit EB2 near the center of the monocline, consistent with the observed gently-dipping reverse faults (Figure 6). Therefore the small reverse fault observed in the core of the monocline is probably a result of the stress conditions imposed by the fault propagation, and does not necessarily have to nucleate at the tip of the master fault (a major structure substantially continuous at depth relative to other structures in the specified system, e.g., Boncio et al., 2004). We also infer the reason the minor reverse faults observed in the trench have a different strike (∼90°) than the fold scarp (∼127°), is because their orientation is controlled by local stress conditions, which may be more variable than the regional stress field controlling the master fault at depth.

In the less deformed monocline limbs of the fault-propagation fold, strain ellipses in Unit EB2 are nearly circular, with minor horizontal trench-parallel shortening (Figure 9). These strain ellipses are not consistent with the sub-vertical fractures and normal faults indicative of north-south extension observed in the hanging wall, or up-thrown limb of the monocline (Figure 6). Although we did not document clear evidence of strike slip along these faults, the vertical faults could accommodate strike slip along the XELF given their minimal vertical offsets. Strike-slip deformation is not represented in the 2D fault-propagation fold model and must be examined independently.

6.2 Paleo-Earthquake Rupture Length and Magnitude Estimation

We use ∼2.3–3.2 m of dip slip during a single earthquake to estimate a paleo-earthquake magnitude. This slip range encompasses all of the dip (45°–55°) and vertical separation observations from the trench and the ERT surveys at EB (∼2.3–3.0 m, Figures 6 and 8) and the slip estimated in the fault-fold propagation model (∼3.2 m, Figure 9). We use this datum as the challenges of preserving and mapping surface ruptures in the study area preclude observations of accurate rupture lengths. We do not use the slip (<20 cm) along the minor reverse fault in the core of the monocline observed in the trench in our estimation, because we infer that this structure nucleated near the surface (Figure 9c), has a negligible surface area compared to the main structure, and is therefore unlikely to contribute substantially to moment release. We also assume minimal strike slip, based on the dip-slip slickenlines observed in bedrock faults (Figure 2) and the lack of consistent lateral geomorphic offsets along the Quaternary fault (Figure 3b). However, as we do not have an age constraint on the bedrock slickenlines and as the geomorphic offsets may be disturbed by urban development (e.g., Figure 3a), further work constraining Quaternary strike-slip offset could confirm a purely dip-slip paleo-earthquake.

We first estimate the rupture length based on average or maximum surface displacement using both the (a) Wesnousky (2008) and (b) Leonard (2010) scaling relationships between average surface slip or maximum surface slip and rupture length (at depth). For Wesnousky (2008) rupture length is calculated using:

$$$Length(km)=Slip(m)/C$$$ (1)

where the constant C = 0.21 when the observed slip is assumed to be maximum and C = 0.06 when slip is assumed to be average. This equation, using the range of dip slip from 2.3 to 3.2 m, results in a rupture length between ∼11 and 15 km (assuming maximum slip) and between ∼38 and 53 km (assuming average slip). For Leonard (2010) rupture length is calculated using:

$$$Length(km)=1{0}^{((\log (Slip(m))-B)/A)}$$$ (2)

where constant A = 0.833 and constant B = −1.30, a rupture length of ∼99–147 km is calculated for an average slip of 2.3–3.2 m.

We used the bedrock fault maps and bathymetric studies (Figure 2a) to estimate the maximum rupture lengths that the XELF could accommodate and to compare with our calculated lengths. A rupture on the XELF could continue along strike to the southeast along the previously mapped bathymetric features in Haro Strait (Greene & Barrie, 2022) as far to where they intersect with the DMF (Personius et al., 2014). To the northwest, we consider the intersection of the XELF and the SJF as a structural boundary that could inhibit rupture propagation for two reasons. First, the SJF is a major terrane-bounding structure that accrued substantial left-lateral slip in the Eocene (Harrichhausen et al., 2022) and offset the bedrock fault the XELF may have reactivated (Figure 2a). Second, instrumental seismicity along the XELF diminishes to the northwest of the SJF. This observation is consistent with Quaternary offset only being observed on the eastern, northwest to southeast striking, half of the LRF to the south (Harrichhausen et al., 2021; Morell et al., 2017, 2018). Given these constraints, we estimate a maximum rupture length of ∼73 km. This length is slightly greater than maximum rupture lengths calculated using the Wesnousky (2008) slip to length relationship (∼53 km), but less than the ∼99–147 km calculated using the Leonard (2010) relation. Because these rupture lengths are significantly larger than the estimated fault length, we consider the Leonard (2010) scaling relation may not be appropriate for use in this tectonic setting. In fact, a compilation of paleo-earthquake data from the Puget Lowland of northern Cascadia indicates that ruptures in this region have relatively large displacements compared to rupture lengths, and fit more closely with the Wesnousky (2008) length-displacement relation (see Figure 2 in Styron & Sherrod, 2021). However, coeval rupture of the XELF and the DMF to the southeast (Figure 1b) could allow for a rupture of ∼147 km and further work could constrain earthquake timing on both structures as our current earthquake age is too unconstrained for a meaningful correlation (Figure 10).

Figure 10

Chronology in years before present (1950) of Debris Flow Deposits (DFDs) in Saanich Inlet (Blais-Stevens et al., 2011) and megathrust (dark gray bar with blue outline) and upper plate (dark gray bar) earthquake records. Cascadia turbidite ages are from Goldfinger et al. (2012) and coastal subsidence ages are from Atwater et al. (2003). All ages are in years before present (1950). Earthquake ages for the Leech River fault (LRF) east are from Morell et al. (2018), and those from LRF west are from Harrichhausen et al. (2021). Earthquake ages for the Devils Mountain fault are from Personius et al. (2014). Correlations between DFDs and megathrust events interpreted from the turbidite record (blue vertical dashed lines) are after Blais-Stevens and Clague (2001), whereas correlations between DFDs and upper plate faults (gray vertical dashed lines) are from this study.

Based on our observations of offset Quaternary sediments from Waterski Beach site on the northwest shoreline of XEOLXELEK to the eastern side of the drumlinoid along the shore of Haro Strait (Figure 2b), we estimate a minimum rupture length of ∼4.1 km. This is much shorter than the length estimated using the Wesnousky (2008) scaling relation (∼11 km), and we therefore infer that the observed earthquake rupture must have continued beyond our mapped Quaternary scarp length and into the bedrock scarps northwest of Waterski Beach and/or southeast beneath the Haro Strait.

We use the full range of rupture lengths estimated using the Wesnousky (2008) scaling relation (11–53 km) and the following equation from Wesnousky (2008):

$$$Magnitude(M)=A+B\ast \log (Length(km))$$$ (3)

where constant A = 4.11 and constant B = 1.88, to approximate a paleo-earthquake magnitude of M = 6.1–7.4. This magnitude range largely overlaps with that for forearc earthquakes in the Puget Lowland of northern Cascadia (M = 6.3–7.5) using a joint Bayesian regression of rupture offset and length (Styron & Sherrod, 2021). Considering a maximum rupture length based on an estimate fault length of ∼73 km, we estimate a maximum magnitude of M = 7.6.

6.3 Uncertainty in Magnitude Estimation

There is large uncertainty in our magnitude calculation resulting from our estimation of fault slip, and whether this fault slip represents average or maximum slip (or neither). The estimation of fault slip at the EB trench site (2.3–3.2 m) is well constrained by the vertical separation of Unit EB2 (2.0 ± 0.1 m, Figure 6a), our fault-fold model (Figure 9), and the fault dip indicated by the ERT profile (Figure 8). However, a greater vertical separation (2.5 ± 0.2 m) across the scarp on the drumlinoid ridge (Profile 2, Figure 3), in the ERT profile at EB (∼2.7 m, Figure 8d), along with a greater vertical separation and lesser fault dip in the ERT profile at Waterski Beach (∼2.8 m, 38° ± 10°; Figure 5b) all suggest potentially greater slip values along strike. Because of this, we suggest considering a M 7.6 earthquake, estimated from maximum bedrock fault length, as the upper bound of our magnitude estimation. Additionally, if 3.2 m of dip slip is considered as the maximum slip we estimate M = 6.3, whereas if 3.2 m is considered as the average slip we estimate M = 7.4. Therefore, further studies of the surface slip distribution (to better asses maximum vs. average slip), surface rupture length, and seismogenic depth could better constrain the paleo-earthquake and potential future earthquake magnitudes for use in seismic hazard assessments (e.g., Styron & Sherrod, 2021).

7 Discussion

7.1 Evidence for Minor Strike-Slip Deformation

Given the possibility of strike slip along the vertical faults observed in the southern portion of the trench we consider the deformation exposed in the trench may be partitioned, similar to the surface rupture of the 2010 M 7.1 Canterbury earthquake (Quigley et al., 2012) and to observations from a paleoseismic trench across the LRF 25 km west of Victoria (Harrichhausen et al., 2021). However, geomorphic evidence suggests that if there is a strike-slip component to Holocene deformation on the XELF, it is negligible compared to the dip-slip component. The main evidence that faulting is primarily dip slip is the lack of unambiguous strike-slip offset of geomorphic features along the topographic scarp. The apparent left-lateral offset along the creek in the drumlinoid is not repeated along the sides of the druminoid, and therefore is more likely controlled by non-tectonic, erosional processes. Instead, our evidence implies that if there is a component of strike slip, it is minor and right-lateral. For example, the ∼282°–101° fold hinge orientation observed in the trench (Figure 6) and the north-verging reverse fault slickenlines observed in bedrock (Figure 2, Outcrop 2) suggest north-south shortening, which is indicative of oblique reverse slip with a minor right-lateral component on the northwest-southeast striking fault plane we observe from the topographic scarp (e.g., Sanderson & Marchini, 1984). North-northwest south-southeast S_Hmax directions derived from historical instrumental seismicity (Balfour et al., 2011) also indicate that if the XELF accommodates a strike slip component, it is minor and right-lateral. However, the lack of an unambiguous and substantial strike-slip component of Holocene offset along the XELF suggests the last reverse earthquake rupture was predominantly dip slip.

7.2 Evidence Against Isostatic Rebound Induced Slip

Glacial-isostatic rebound has been suggested as an explanation for temporally isolated earthquake ruptures in recently glaciated regions (e.g., Anderson et al., 1989; Craig et al., 2016; Davenport et al., 1989; Jarman & Ballantyne, 2002; Lagerbäck, 1990; Mörner, 1991; Muir-Wood, 1989; van Loon et al., 2016), but several lines of evidence suggest that the single earthquake we observe on the XELF is a result of active tectonic processes. Retreat of the Cordilleran continental ice sheet over southern Vancouver Island after 12.1 ka produced a period of rapid isostatic rebound until 6 ka (Clague et al., 1982; James et al., 2002, 2009; Linden & Schurer, 1988; Mathews et al., 1970; Mosher & Hewitt, 2004). The 10.9 to 0.8 ka earthquake age range recorded in the trench overlaps this period of rapid rebound and the documented reverse slip on the XELF is consistent with post-glacial release of vertical stress, which reduces the normal stress on the fault plane promoting dip-slip fault rupture (Jarman & Ballantyne, 2002; Muir-Wood, 1989). However, regional data and our earthquake age suggest the XELF should be considered as currently active. This is because the age PDF is substantially skewed toward the younger ages with a mode at 2.7 ka (Figure 7), well after 6 ka and the most active period of glacial rebound (Clague et al., 1982; James et al., 2002, 2009; Linden & Schurer, 1988; Mathews et al., 1970; Mosher & Hewitt, 2004).

Surface rupturing earthquakes are also documented well after 6 ka along strike on the DMF and ∼12 km southwest on the LRF (Morell et al., 2018; Personius et al., 2014), demonstrating that other nearby faults have produced surface-rupturing earthquakes well after the isostatic response and stress changes related to deglaciation and are thus likely responding to regional tectonic stresses (e.g., Harrichhausen et al., 2021). In nearby Saanich Inlet (Figure 1), several debris flow deposits (DFDs) observed in sediment cores are interpreted to result from ground shaking events that do not correspond to documented megathrust earthquakes on the Cascadia subduction zone (Blais-Stevens et al., 1997, 2011). The youngest of these uncorrelated deposits is dated at 375–350 cal BP, consistent with recent earthquakes on southern Vancouver Island well after rapid post-glacial rebound.

Instrumental crustal seismicity (Figure 1) with right-lateral and reverse focal mechanisms shows that the current active strain field is consistent with our observations of reverse or oblique reverse-dextral slip on the northwest striking XELF (Balfour et al., 2011; Bostock et al., 2019; Brocher et al., 2017; G. Li et al., 2018). This current upper plate strain in Cascadia is capable of producing large, surface-rupturing earthquakes as evidenced by the 900–930 CE Seattle fault earthquake (Bucknam et al., 1992). The extensive evidence of late Holocene surface-rupturing earthquakes on forearc faults in northern Cascadia, and the evidence that some occurred in the same stress field as the current one, strongly suggests that the Holocene earthquake on the XELF was related to regional tectonic stress rather than stress changes associated with isostatic rebound.

7.3 Relation to Nearby Structures

The western extent of the mapped XELF intersects with Saanich Inlet (Figure 2) where there is a rich record (250–4,509 cal BP) of extensive DFDs interpreted to be deposited during submarine slope failures caused by seismic shaking (Figure 10; Blais-Stevens et al., 1997; Blais-Stevens et al., 2011). Given the proximity of the XELF to Saanich Inlet, it is likely that any large earthquake on the fault would cause a DFD in the inlet. Nine of the DFDs are correlated to offshore turbidites caused by megathrust earthquakes (blue dashed vertical lines in Figure 10; Goldfinger et al., 2012), whereas the remaining nine are thought to record local earthquakes. If the XELF earthquake rupture age range based on the 95% confidence interval (10.9–0.8 ka) is used, all but the youngest crustal earthquake-triggered DFD #0 (375–350 cal BP) could be related to a rupture on the XELF. If the 68% confidence interval (4.7–2.3 ka) is used, at least three DFDs could record this event: #11 (2,449–2,137 cal BP), #12 (2,423–2,236 cal BP), or #16 (4,509–3,594 cal BP). The DFD record does not extend prior to 4,509 cal BP and older non-preserved DFDs could also be correlated to the XELF rupture. However, given that the mode of the PDF for a rupture on the XELF is 2.7 ka and that the PDF is skewed toward younger ages (Figure 7), it is most probable that either DFD #11 (2,449–2,137 cal BP) or DFD #12 (2,423–2,236 cal BP) are related to this event. Furthermore, both the middle of the three earthquake ruptures (2,300–2,100 cal BP) identified on the LRF by Morell et al. (2018), as well as the most recent rupture (2,300–1,500 cal BP) on the DMF identified by Personius et al. (2014), also overlap with the age ranges of these two DFDs (Figure 10). This observation raises the possibility that ruptures on the Devils Mountain, Leech River, and XELF may be closely clustered in time and related through static and or dynamic stress changes (e.g., Freed, 2005; King et al., 1994).

The kinematics of upper plate faults in the northern Cascadia forearc are thought to be related to oroclinal bending and westward escape of the forearc centered at the Olympic Peninsula (Finley et al., 2019; Harrichhausen et al., 2021; Nelson et al., 2017). Consequently, west-northwest-striking Holocene faults north of the Olympic Peninsula generally have right-lateral transpressional kinematics, and east-southeast striking faults to the south are left-lateral transpressional. In this framework, the XELF should have a large component of right-lateral slip, similar to the LRF (Harrichhausen et al., 2021; Morell et al., 2017, 2018), which is ∼12 km southwest of and parallel to the XELF. However, as discussed, we observe predominantly reverse slip at the trench site and we propose that strike-slip and dip-slip strain could be partitioned between the two fault systems. Examples of this occurring on parallel-striking faults, similar to the Leech River and XELF near Victoria, are observed in fault zones characterized by oblique convergence such as in the Colombian Andes (Acosta et al., 2007) and the Qilian Shan north of Tibet (Allen et al., 2017). We suggest that the reason there is a greater dip-slip component on the XELF than the LRF is due to their different dips. The seismically active LRF is suggested to be subvertical (Harrichhausen et al., 2021; Morell et al., 2017, 2018), which given current S_Hmax orientations (Figure 1) is more optimally oriented for strike slip. The XELF instead dips moderately southwards and is more optimally oriented for reverse slip. The XELF is located ∼10 km south of parallel reverse faults and folds that accommodated Eocene southwest-northeast shortening during the formation of the CFTB (England & Calon, 1991; England et al., 1997). Although these structures predominantly dip to the northeast (England & Calon, 1991), the XELF could have reactivated a minor southwest-dipping reverse fault that formed during the Eocene. The Quaternary XELF does not exactly coincide with the previously mapped nearby bedrock fault, however, it may be reactivating minor structures within the damage zone of a larger fault structure or an unmapped related structure. The presence of a larger number of bedrock faults than shown in the regional scale geologic map compilation (Cui et al., 2017) is evident by the large number of parallel northwest to west trending bedrock lineaments visible in the DTM of Saanich Peninsula.

7.4 Hazard Implications

A M = 6.1 to 7.6 reverse-slip earthquake with a 13–73-km-long rupture ∼10 km north of downtown Victoria has the potential for considerable damage to the city and its surrounding urban area. Because the XELF dips beneath Victoria, there is an elevated likelihood of damage in the densely populated and built-up city center. Peak ground accelerations and damage tend to be greater in the hanging wall than the footwall during reverse earthquakes. For example, peak horizontal ground accelerations in the hanging wall were ∼50% greater within 5–20 km of the fault surface trace in both the 1994 M 6.7 Northridge and the 1999 M 7.6 Chi-Chi earthquakes (Abrahamson & Somerville, 1996; Shabestari & Yamazaki, 2003), and in the latter, almost all recorded building damage was in the hanging wall (Chang et al., 2004). In addition, the XELF poses a ground rupture hazard within the built-up suburbs on Saanich Peninsula (e.g., Figure 4a). Surface rupture would go through many residential properties and could also severely damage the main highway connecting downtown Victoria to the main international airport and ferry terminal for the region (Patricia Bay highway, Figures 2a and 3). The ferry terminal and airport are the main transportation links between Greater Victoria and the mainland of British Columbia, and severing this link would hamper aid reaching the city. Recent development of earthquake early warning systems (EEWs) in Cascadia (Crowell et al., 2016) is an important step toward mitigating hazard from megathrust earthquakes; however, EEWs rely on the time difference between P wave and S wave travel times and they become ineffective for areas very close to the earthquake rupture (Wald, 2020). This would be the case for the Greater Victoria region and an earthquake on the XELF, and as a result of the heightened risk the XELF poses, future regional PSHA and PFDHA could be improved by including the XELF as an additional fault source.

Along with fault locations and potential earthquake magnitudes, PSHA and PFDHA fault source models require estimates of slip rates or earthquake recurrence intervals (e.g., Brune, 1968; Chartier et al., 2017; Coppersmith & Youngs, 2000; Youngs & Coppersmith, 1985; Youngs et al., 2003), which cannot be accurately calculated for the XELF given a single earthquake event age. However, as there are other large uncertainties associated with these models, such as ground motion prediction equations, a rough estimation of maximum slip rate even with a large uncertainty can improve seismic hazard assessments (e.g., Styron & Pagani, 2020). Because we only see one event deforming Unit EB2, we consider the age difference between the deposition of EB2 and the earthquake age mode (2.7 ka) as the minimum earthquake recurrence interval. We have two interpretations for the age of Unit EB2: either it is part of the Victoria clay deposited at the end of the Fraser Glaciation (∼14 ka), or it belongs to the Cowichan Head Formation (∼42 to 25 ka). If EB2 represents the Victoria clay, based on our dip slip range of 2.3–3.2 m, there is a maximum dip-slip rate of 0.22–0.34 mm/yr. If EB2 belongs to the Cowichan Head Formation, using the maximum age of sample C22 as a maximum depositional age (31.7 ka), we obtain a maximum dip-slip rate of 0.07–0.10 mm/yr. The faster slip rates (0.22–0.34 mm/yr) are comparable with 0.2–0.3 mm/yr, 0.14 ± 0.1 mm/yr, and 0.3–0.6 mm/yr of post-glacial slip calculated on the nearby LRF (Morell et al., 2018), Devils Mountain (Personius et al., 2014), and Boulder Creek (Sherrod et al., 2013) faults, respectively. The uncertainty in our minimum slip-rate estimation (∼0.1–0.3 mm/yr) could be reduced by further paleoseismic investigations along the XELF that could extend the earthquake history of this fault thus allowing for time intervals between earthquakes to be measured.

Finally, local tsunami in the waters surrounding Greater Victoria could result from a reverse-slip earthquake on the XELF. Because the observed scarp length is less than the calculated minimum rupture length of ∼11 km, we infer the paleo-earthquake likely ruptured beneath Saanich Inlet to the west (Figure 10; Blais-Stevens et al., 2011) and/or beneath Haro Strait to the east. A ∼53-km-long rupture (our maximum estimate from displacement) could have continued as far as the DMF in Washington State to the southeast (Figure 1), where similar northward-verging reverse faults have been identified (Greene & Barrie, 2022). There is a high potential that a local tsunami was generated by the paleo-earthquake recorded on the XELF, either from vertical offset of the seafloor or during a shaking-induced landslide (e.g., Nemati et al., 2023). Future analyses of the subaqueous structures, along with exploration for regional tsunami deposits, could further test this hypothesis.

7.5 Lessons for Finding Cryptic Faults

Low strain rates, GNSS velocities dominated by a subduction coupling signal (Figure 1a) and dispersed instrumental seismicity (Figure 1b) in the northern Cascadia forearc have required employing a multifaceted approach for identifying and characterizing active faults. These challenges are not unique to Cascadia or to forearcs. For example, damaging earthquakes in New Zealand (e.g., the 2010 M_w 7.1 and 2011 M_w 6.3 Canterbury earthquakes, Quigley et al., 2012; Potter et al., 2015) and Haiti (2010 M_w 7.0 earthquake, Calais et al., 2010) occurred on hidden or concealed faults near transform plate boundaries where GNSS and seismicity signals are dominated by a larger nearby structure. We show that high-resolution topography is critical for identifying fault surface traces where low strain rates, limited sedimentary cover, and dense vegetation or human infrastructure obscure the geomorphic signatures of active deformation. In these study regions, obtaining lidar across all bedrock structures, whether they exhibit a strong topographic lineament (e.g., LRF, Morell et al., 2017) or not (XELF, this study), is important to assess if they have been reactivated. Additionally, using historical imagery can prove very useful where urban development may hide previously visible fault scarps. Once a potentially active fault is identified, it is also important to consider a multifaceted approach to a paleoseismic investigation (e.g., Camelbeeck & Meghraoui, 1998; McCalpin et al., 2023; Sherrod et al., 2008; Vanneste et al., 2001). Paleoseismic trenching is a pivotal and essential method needed to verify the existence of Quaternary rupture on faults such as the XELF, and without the use of shallow geophysics (ERT) in conjunction with field mapping, we would have not been able to discern the dip and vertical separation of the fault. This information was essential for forward modeling of the slip required to produce the fault propagation fold observed in the trench (Figure 9) and the slip estimation is essential for estimating fault slip rates. Finally, given the cryptic nature of these faults, we cannot assume that the entire surface trace is mappable, and so we also need to rely on our observations of slip to infer paleo-earthquake magnitudes. Limiting rupture size based on known fault length could result in an underestimate of potential earthquake magnitude given that historic earthquake ruptures have continued beyond the limit of mapped fault scarps (e.g., Morishita et al., 2017; Thompson Jobe et al., 2020).

This study also highlights several of the complications faced when conducting paleoseismic studies in previously glaciated terrain, and the multifaceted approach that we show is important for studying active deformation in forearcs also helps tackle these challenges. A good understanding of glacial retreat timing and the duration and magnitude of isostatic rebound is important for interpreting the results from paleoseismic studies, as it would be difficult to assign a tectonic origin to active faults if a well-established glacial-isostatic rebound curve does not exist. A knowledge of glacial flow directions is also necessary as certain fault kinematics are incompatible with sub-glacial traction forces. For example, a south-flowing glacier would not be expected to form a north-verging thrust fault like the XELF, as the basal traction would induce south-verging deformation. Additionally, quality surficial geologic maps are required to distinguish diagnostic Quaternary scarps from erosional bedrock scarps. Bedrock scarps resulting from preferential erosion and plucking of brittle fault zones are common in recently glaciated terrain, but are not indicative of active faulting as they may have formed in a different tectonic regime with different stress conditions (e.g., Harrichhausen et al., 2022; Personius et al., 2014). High-resolution topography and bathymetry, such as lidar used in this study, are the best way to identify Quaternary scarps, which are typically very subtle, due to the short period of time for slip to accrue and the potential for rapid erosion and poor preservation potential after deglaciation. ERT is also a particularly useful tool because glacial sediments are well suited to the technique. Clays are typically very conductive (Palacky, 1988), and can contrast strongly with more resistive gravels and bedrock allowing ERT surveys to more easily distinguish offset clay- and silt-rich diamict. Identification of active faults in glaciated regions has been slow due to the extra burden of proof required to show that fault activity is tectonically driven. Our study stands as an example of how a coordinated multi-disciplinary approach is most effective at characterizing the seismic hazard in these regions.

8 Conclusions

DTMs, paleoseismic trenching, and ERT surveys delineate the previously unrecognized Holocene-active XELF, which strikes across Saanich Peninsula on southern Vancouver Island, Canada. Our study indicates that the XELF hosted a single M 6.1 to M 7.6 earthquake that deformed glacial and post-glacial sediments via a fault-propagation fold above a blind reverse fault. The earthquake most likely occurred between 4.7 and 2.3 ka, overlapping with events on the Leech River and Devils Mountain faults, and with possible earthquakes recorded by debris flows in the nearby Saanich Inlet. By using fault-fold propagation modeling, we show that ∼3.2 m of reverse slip on a fault propagating through two different geologic units, with different propagation-to-slip ratios, could reproduce the deformation we observe in the trench and ERT surveys. This result shows that how faults propagate through different media is important to consider in paleoseismic investigations because estimations relying on vertical separation alone may underestimate slip. Our estimated magnitude range is consistent with length and paleo-earthquake magnitudes estimated for other forearc faults in northern Cascadia, suggesting that forearc faults in this region may follow a specific length to magnitude scaling relation. A future earthquake in this magnitude range also has the potential to incur substantial damage to the Greater Victoria region through shaking, fault displacement of important infrastructure, and potential tsunami impacts. Integration of the XELF into PSHA and PFDHA, along with further studies to better constrain paleo-earthquake rupture extent, magnitude, and the slip rate on the XELF will better inform the hazard and risk that the structure poses. Finally, given that the XELF cannot be easily discerned from instrumental seismicity and GNSS vectors because these signals are dominated by coupling along the underlying Cascadia subduction zone interface, this study highlights the importance of not relying on instrumental networks alone to identify cryptic forearc faults, or faults near other plate boundaries. The use of multiple methodologies including remote sensing, paleoseismic trenching, shallow geophysics, and field mapping greatly benefits the identification and characterization of active faults in forearcs. This result also applies to paleoseismic studies in recently glaciated regions where glacial erosion can highlight inactive bedrock structures, whereas a limited stratigraphic record can obscure active faulting.

Acknowledgments

We are grateful to the WSÁNEĆ Leadership Council and the Capital Regional District for allowing us to excavate in Elk/Beaver Lake Park, which lies within the traditional territories of the WSÁNEĆ people. We are especially grateful to Simon Smith for monitoring the excavation and sharing insights into the history of XEOLXELEK, and to April Mitchell, Jeanette Mollin and Marc Solomon for their help with logistics. We thank Dr. Andrew Schaeffer for providing drone images of the trench, Walter Langer for the excavation of the trench and associated logistics, and Dr. Brian Whiting for discussion of geophysical survey results. This research was supported by National Science Foundation Earth Sciences (NSF EAR) Grants 1756943 and 2046278 to Dr. Morell, NSF EAR Grant 1756834 to Dr. Regalla, a Natural Sciences and Engineering Research Council of Canada (NSERC) Post Graduate Scholarship and Centre National d’Études Spatiale (CNES) postdoctoral fellowship to Dr. Harrichhausen, funding from the USGS National Cooperative Geologic Mapping Program to Dr. Bennett, an NSERC Discovery Grant 2017-04029 and Canada Research Chair to Dr. Nissen, an NSERC Alexander Graham Bell Canada Graduate scholarship to T. Finley, and a University of Victoria Jamie Cassels Undergraduate Research Award to E. McLeod. Finally we thank R. Wells, C. Grützner, and two anonymous reviewers for their thoughtful insight and comments on our manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Open Research

Data Availability Statement

Figures showing the three-point problem solution, detailed photo orthomosaics and interpretations of the entire trench, codes for the OxCal chronological models, and tabulated radiocarbon data are in the Supporting Information. Tabulated unit descriptions and structural data from the trench, auger hole data, and the electrical resistivity tomography data files are located in a data repository (Harrichhausen et al., 2023) that can be accessed at: https://datadryad.org/stash/dataset/doi:10.25349/D9VW3D. GNSS velocity data in Figure 1a is from https://www.unavco.org/software/visualization/GPS-Velocity-Viewer/GPS-Velocity-Viewer.html. Seismic catalogue shown in Figure 1b is from the Pacific Northwest Seismic Network available at https://pnsn.org/. Low-resolution topography basemaps in Figure 1 was accessed from the USGS 3D Elevation Program (3DEP) Bare Earth DEM Dynamic service available at https://elevation.nationalmap.gov/arcgis/rest/services/3DEPElevation/ImageServer. Gridded population density data used in Figure 2a are available from https://sedac.ciesin.columbia.edu/data/set/gpw-v4-population-density-rev11. The lidar-derived DTMs used for topographic analyses are available at the publicly accessible BC Open Lidar Data Portal at https://governmentofbc.maps.arcgis.com/apps/MapSeries/index.html?appid=d06b37979b0c4709b7fcf2a1ed458e03. Stereonet 11.4.1 and FaultFold version 7.2.0 are both available from https://www.rickallmendinger.net. Agisoft Metashape software is available at https://www.agisoft.com/. QGIS v. 3.22 is available at https://download.qgis.org/downloads/. OxCal software is available at https://c14.arch.ox.ac.uk/oxcal.html. Advanced Geosciences Inc. EarthImager™ 2D software is available at https://www.agiusa.com/agi-earthimager-2d.