Holocene tectonics and fault reactivation in the foothills of the north Cascade Mountains, Washington

Geologic and geodetic data show that near-surface faults in the Cascadia forearc accommodate north-south shortening. Models of the forearc (Fig. 1A) show a series of migrating, clockwise-rotating forearc blocks (Wells and Simpson, 2001; Wells et al., 1998). This clockwise rotation causes convergence in western Washington (United States) where the Oregon Coast Range block impinges on Tertiary volcanic rocks and sediments, compressing these Tertiary rocks against the southern edge of the British Columbia (Canada) Coast Mountains. This compression results in a series of structural basins separated by uplifts in the northern Cascadia forearc. Geological models show that blocks in the northern Cascadia forearc move northward relative to the Coast Mountains at rates of 7–9 mm/yr (Wells and Simpson, 2001; Wells et al., 1998).

Geodetic studies show that the region between 46.5°N and 49.5°N is undergoing north-south shortening averaging ∼3 mm/yr to 4.4 mm/yr (Hyndman et al., 2003; Mazzotti et al., 2002; McCaffrey et al., 2007). GPS rates are higher in the southern forearc (6–10 mm/yr) and decrease to the north (Fig. 1C). With geodetic rates falling to near zero just north of the United States–British Columbia border, questions remain regarding the activity of faults in the northern Puget Lowland (Washington).

Forearc basins and uplifts in western Washington—from south to north, the Tacoma, Seattle, and Everett Basins—are defined mainly on the basis of high-amplitude geophysical anomalies, with faults hypothesized where the gradients are strongest (Blakely et al., 2002; Brocher et al., 2001; Danes et al., 1965). Many of the faults found between the basins and the adjacent uplifts are active and present substantial seismic hazard to the region (Fig. 1B). The best-known of these basin-bounding faults are the Tacoma, Seattle, Southern Whidbey Island, and Darrington–Devils Mountain faults (Bucknam et al., 1992; Johnson et al., 2004a, 2004b, 1994; Sherrod, 2001; Sherrod et al., 2008, 2004). The Darrington–Devils Mountain fault forms a broad boundary between the northern edge of the Everett Basin and accreted Mesozoic rocks and Tertiary sedimentary rocks in the adjacent uplift. North of this uplift lies the Bellingham forearc basin, the northernmost forearc basin. The Bellingham Basin preserves Eocene to Quaternary sedimentary rocks in its interior and is bounded on the north by pre-Tertiary sedimentary and metamorphic rocks. The Bellingham Basin is anomalous compared to the other forearc basins in western Washington because prior to our studies, no known active faults lined the basin margins, yet geologic and geodetic data suggest that the basin should be the locus of active faulting (Kelsey et al., 2012).

This paper demonstrates active faulting on a set of recently discovered fault scarps along the eastern edge of the Bellingham Basin in northwestern Washington (Fig. 2). These scarps lie adjacent to and represent reactivation of previously mapped bedrock normal faults (Fig. 2B). Observations collected from LiDAR mapping, fault-scarp excavations, and sediment cores from a wetland adjacent to one scarp show evidence for multiple Holocene earthquakes in an area of the forearc previously thought to be tectonically quiescent.

Forearc basins in western Washington lie between the Olympic Mountains to the west and the Cascade Mountains to the east (Fig. 2A). A broad lowland repeatedly glaciated in the Quaternary occupies the region (Booth, 1994; Easterbrook, 1985). Marine waters of Puget Sound partially occupy large areas of this lowland, and Quaternary glacial deposits partially fill the lowland. These glacial deposits unconformably drape an older set of Tertiary volcanic and sedimentary rocks preserved in the forearc basins and adjacent uplifts.

Three groups of rocks and sediments characterize the bedrock geology of the study area (Figs. 2 and 3). First, pre-Tertiary metamorphic and igneous rocks underlie much of the area north of the Nooksack River, and in places, minor amounts of sedimentary deposits cover these rocks. Second, the primary structural low in the region, the Bellingham Basin (Fig. 2C), preserves Tertiary sedimentary rocks, including the Eocene Chuckanut Formation, Eocene and Oligocene Huntington Formation, and mid-Miocene Boundary Bay Formation (Hopkins, 1968; Mustard and Rouse, 1994). Lastly, Quaternary glacial and nonglacial deposits blanket most of the region below 350 m in elevation.

The Quaternary Bellingham Basin, a smaller vestige of the larger Tertiary Georgia Basin, lies west of the Cascade Mountains between the San Juan Islands in Washington and the Coast Range Mountains north of Vancouver, British Columbia (Fig. 2A). Here we focus on the northeastern part of the Bellingham Basin near Kendall, Washington, where LiDAR mapping revealed evidence for surface faulting along the Boulder Creek fault, a fault originally described as a basin-bounding normal fault by Miller and Misch (1963).

The Boulder Creek fault (Figs. 2B and 3) was formally defined as a south-dipping normal fault that bounds the northern margin of the Chuckanut Formation and separates Chuckanut Formation from pre-Tertiary rocks to the north (Miller and Misch, 1963). Miller and Misch (1963) showed the fault as a normal fault in cross sections depicting the regional geology. Based on geologic mapping, Tabor et al. (2003) suggested it is a NE-SW–trending normal fault that places north-dipping Eocene Chuckanut Formation sandstones to the south against pre-Tertiary metamorphic rocks to the north. Dragovich et al. (1997a) mapped the Smith Creek and McCauley Creek faults, in addition to the Boulder Creek fault, near Kendall, and argue that the 1990 M5.0 Deming earthquake occurred on the McCauley Creek fault, a small thrust fault northwest of Deming (Fig. 3). The earthquake was initially reported at a depth of 12.6 km (Advanced National Seismic System catalog, http://earthquake.usgs.gov/monitoring/anss/) but later updated to a depth of 3–4 km based on recorded aftershock distributions (Dragovich et al., 1997b). To date, there are no observations (e.g., seismic reflection profiles) bearing on the subsurface geometry of the faults in the region surrounding the Deming earthquake epicenter. Furthermore, Dragovich et al. (1997a) suggested that the McCauley Creek fault is kinematically linked with other nearby faults, including the Boulder Creek fault, as part of a larger thrust fault system. Map patterns alone suggest that all of these faults are normal faults separating the Chuckanut and pre-Tertiary rocks (Dragovich et al., 1997b; Lapen, 2000; Tabor et al., 2003).

Tabor et al. (2003) mapped several faults northwest of Mount Baker, including an inferred structure we informally refer to as the Canyon Creek fault (Fig. 3). The Canyon Creek fault trends north-south near Mount Baker and bends northwesterly at its intersection with the Nooksack River valley. The dip and kinematics of the Canyon River fault are not known. Cross sections by Tabor et al. (2003) show the fault as a near-vertical or south-dipping reverse fault in the hanging wall of the south-dipping Glacier Extensional fault (Fig. 2). Detailed kinematics of the Canyon Creek fault are unknown but mapped offset of pre-Tertiary rocks indicates right-lateral and/or east-side-down offset.

During the last glaciation, the Cordilleran Ice Sheet expanded southward from the Coast Mountains and Fraser Lowland in southwestern British Columbia into northwestern Washington. At about the latitude of the Olympic Mountains, the ice sheet divided into two lobes; one lobe flowed west down the Strait of Juan de Fuca and the second lobe flowed south into the Puget Lowland (the Puget Lobe). Calibrated radiocarbon ages of organic matter below and above till of the Vashon stade of the Fraser glaciation of Armstrong et al. (1965) provide limiting ages on the Puget Lobe’s advance and retreat (Porter and Swanson, 1998). Following retreat of the Cordilleran Ice Sheet, marine water inundated the area, depositing glaciomarine drift (Armstrong and Brown, 1954). Wood in the glaciomarine drift yielded an age of 13,960–13,490 cal. yr B.P. (Kovanen and Easterbrook, 1998). After deposition of glaciomarine deposits in the area and emergence of the area, ice from the Coast Mountains re-occupied much of the Fraser Lowland (the Sumas stade of Armstrong et al., 1965). Wood collected from ice-contact deposits at the margin of this re-advance near Cultus Lake, British Columbia, yielded an age of 13,770–12,630 cal. yr B.P. (Kovanen and Easterbrook, 2001). Outwash deposits downstream of this ice margin and downstream of our study area contained charcoal layers intercalated in sandy gravels beneath terraces above the Nooksack River; samples of these charcoal layers yielded ages of 12,680–12,400 and 12,870–12,570 cal. yr B.P. (Kovanen and Easterbrook, 2001). Evidence presented by Harrington and Clark (2011) and Osborn et al. (2012) argues against the suggestion by Kovanen and Easterbrook (2001) that alpine glaciers emanating from Mount Baker and areas to the east occupied the three Nooksack Valley forks at this time.

Our study relied on LiDAR data obtained from several sources, including the Puget Sound LiDAR Consortium, the Nooksack Tribe, and the U.S. Geological Survey (USGS) (all of the LiDAR data available are from the Puget Sound LiDAR Consortium; http://pugetsoundlidar.ess.washington.edu/index.html). We rely mostly on a USGS LiDAR survey designed in accordance with Federal Emergency Management Agency LiDAR data collection standards to provide pulse spacings no greater than 1.4 m (∼0.5 pulse/m²), horizontal accuracy of ≤1 m Root Mean Square Error (RMSE), and vertical accuracy of ≤18.5 cm RMSE (37 cm in vegetated areas). We imported individual data tiles consisting of grids with 6 ft² (1.82 m²) cells into Grass GIS (version 6.4, http://grass.osgeo.org) and mosaicked the tiles into a single bare-earth digital elevation model. We then used hillshaded-relief images and slope maps to interpret geomorphic features and surfaces. A histogram-equalized grayscale profile applied to slope maps yielded the most easily interpreted images (Figs. 4–7). The most prominent features observed on LiDAR images of the Kendall area are late Quaternary glacial striations, etched bedding surfaces in eroded bedrock, and planar outwash surfaces occupying valley floors (Fig. 4).

We measured elevation profiles on terraces and across scarps using raster map profiling routines in Grass GIS version 6.4. We analyzed data produced in the profiling procedure using Matlab—a script loads the LiDAR profile data into Matlab and then we handpicked profile segments corresponding to the original upper and lower surfaces and the scarp face and/or debris slope. The script then fits lines to those segments, determines the point on the scarp face with the maximum slope, and calculates the offset of the original surfaces through that point. Where slopes of the original surfaces are remarkably different, we measured scarp height using offset of the crest and toe of the scarp, regardless of what angle the original surfaces are.

We collected organic material from trenches and cores for radiocarbon analysis to constrain ages of stratigraphic horizons and past earthquakes (Table 1). Radiocarbon procedures used accelerator mass spectrometry (AMS) to analyze detrital charcoal samples collected from scarp excavations and plant material picked from sieved samples of wetland deposits.

We prefer delicate plant material over charcoal clasts for ¹⁴C analyses because recycling of delicate plant material in the environment over time is less likely than for charcoal (Gavin, 2001). We sieved samples of peat and gyttja (organic-rich mud) from gouge cores in wetland deposits to separate organic detritus (seeds, leaf parts, and conifer needles) from surrounding matrix. These sieved samples yielded delicate plant materials that could not survive subaerial transport and exposure. Thus, ages of these delicate samples closely approximate the age of deposition, whereas detrital charcoal samples may have had in-built (inherited) ages of hundreds of years at the time of deposition (Gavin, 2001).

We list the laboratory-reported radiocarbon ages in ¹⁴C yr B.P., and used the computer program OxCal (Bronk Ramsey, 1995) and the INTCAL09 calibration data of Reimer et al. (2009) to calibrate the reported ages as probability distributions. The 95% confidence interval of each distribution is reported as cal. yr B.P. (before A.D. 1950), thus, a calibrated age of 1000 cal. yr B.P. is about A.D. 950. We round ages for interpreted earthquakes to the nearest decade to account for additional uncertainty in relating sample ages to stratigraphic contacts inferred as having a tectonic origin (Stuiver and Polach, 1977).

We constrain ages of deposition and deformation using a Bayesian analysis of our radiocarbon ages in the OxCal radiocarbon calibration program (Lienkaemper and Bronk Ramsey, 2009). Bayesian analysis in paleoseismology is particularly powerful because it incorporates prior chronologic information, such as stratigraphic order, laboratory uncertainty, ages of known events, and historical constraints, to reweigh the probability distributions and calculate ages of past events (entered in the model as boundaries). Ages of samples from pre-earthquake soils in the trenches and wetland are entered as phases because their true stratigraphic order is not known.

The most notable LiDAR features are two scarps, one located near Kendall and the second located east of the town of Maple Falls, Washington, near Canyon Creek (Figs. 5, 6, 7, and 8). A ∼4.3-km-long, south-side-up scarp near Kendall (herein called the Kendall scarp; Fig. 5) ranges from 2 m to 4.1 m high, and deforms both Pleistocene outwash surfaces shown on Figure 4. A second scarp, located ∼5.5 km east of Maple Falls near Canyon Creek (herein called the Canyon Creek scarp; Figs. 6 and 7) is ∼2 km long, is south-side-up, has a maximum height of ∼6.9 m, and lies northeast of the inferred trace of the Canyon Creek fault. The scarp forms both uphill- and downhill-facing escarpments across a hillslope bordering the Nooksack River; the scarp is truncated by the modern floodplain of the river at its western end and dies out in a landslide at its eastern end. The changes in facing direction, and two apparently right-laterally offset stream channels, suggest a significant lateral component of slip (locations C and D on Fig. 7).

The Kendall and Canyon Creek scarps are roughly coincident with the mapped traces of the Boulder Creek and Canyon Creek faults. The Canyon Creek scarp has the same sense of slip (southwest-side-up) as the Canyon Creek fault, but the Kendall scarp is south-side-up and the Boulder Creek fault is south-side-down. Our working hypothesis is that both scarps observed on LiDAR images are fault scarps resulting from recent surface-rupturing earthquakes. To test this hypothesis, we placed excavations across the two scarps to map the shallow subsurface stratigraphy and search for evidence of recent deformation.

We placed one excavation across the Kendall scarp near Kendall (herein called the Hornet trench) and one excavation across the Canyon Creek scarp (herein called the Smuggle trench) east of Maple Falls (Fig. 3). Barnett (2007) and Seidlecki (2008) discussed other trenches across the Kendall scarp. Excavations across both scarps observed on LiDAR images revealed folded and faulted sequences of outwash gravels and till, overlain by Holocene soils and scarp colluvium. Both trenches exposed sediments containing organic materials (charcoal) suitable for radiocarbon dating (see below). Photomosaics of each trench are included in the Supplemental File¹. We also conducted a stratigraphic study by coring a small wetland adjacent to the Kendall scarp in hopes of obtaining a more refined record of earthquake timing.

This excavation exposed folded and faulted stratified sandy gravels overlain by forest soils and colluvial deposits (Figs. 5 and 9A; see Table 2 for details of each stratigraphic unit). The oldest stratum exposed in the trench (unit 1) consists of discontinuous beds of sand and gravel that commonly exhibit parallel and cross-stratification. Long axes of imbricated pebbles and cobbles are sub-parallel to bedding in the outwash. In the middle of the trench, stratification in the sand and gravel disappears, mainly below the scarp slope and upper original surface. Kovanen and Easterbrook (2001) and Lapen (2000) mapped similar deposits nearby as undifferentiated Sumas glacial outwash deposits and describe the sediments as loose, moderately to well-sorted sandy gravels with subrounded to rounded clasts derived from the Coast Plutonic Complex in British Columbia and from local sources in the Nooksack Valley. Relief on the upper surface of unit 1 roughly mirrors the scarp topography. Charcoal from nearby correlative units yielded radiocarbon ages between 13,960 and 12,400 cal. yr B.P. (Kovanen and Easterbrook, 2001, 1998).

A dark-colored sandy silt with pebbles lies on top of unit 1 (Fig. 9A). This silt (unit 2) exhibits a silt loam to loamy sand texture and a moderately developed, medium sub-blocky soil structure. Relief on the surface of unit 1 mimics the scarp topography and confines unit 2 to an area below the scarp slope. We see no evidence of unit 2 in the uppermost part of the trench (high side of the scarp). Charcoal from the top of unit 2 has a calibrated radiocarbon age of 3200–2890 cal. yr B.P. (lab analysis number Beta-210084, 2890 ± 40 yr B.P.; Table 1).

Unit 3, a wedge-shaped set of gravelly sand strata, overlies unit 2 in the middle of the trench below the toe of the scarp. Primary dips of the gravelly sand strata roughly mirror the north-sloping scarp above—the underlying silt loam (unit 2) is almost horizontal, suggesting the steeper dips in unit 3 are primary. The gravelly sands in unit 3 pinch out at the toe of the scarp, such that the deposits form a distinct wedge on top of unit 2.

A second dark-colored sandy silt with pebbles (unit 4) lies above unit 3 and forms a slight angular unconformity with units 2 and 3. Unit 4 has a silt loam to loam texture and exhibits a well-developed, very fine to medium granular to sub-blocky soil structure. Like units 2 and 3, unit 4 lies beneath the toe of the scarp and dips gently to the north, mirroring the slope of the scarp above. A near-vertical contact separates the silt loam of unit 4 from pebbly sands in unit 5, while a conformable contact separates unit 4 from overlying sand deposits (unit 6). Charcoal from the lower part of unit 4 has a calibrated radiocarbon age of 1120–930 cal. yr B.P. (Beta-210083, 1100 ± 40 ¹⁴C yr B.P.; Table 1).

Unit 5 is a small V-shaped pendant of pebbly sand and serves as a vertical boundary between unit 2 to the south and units 3 and 4 to the north. Unlike the distinctly stratified unit 3 nearby, unit 5 is massive and has a weak sub-blocky soil structure. Unit 5 forms a complex set of both horizontal and vertical contacts that separate units 4 and 5 from stratigraphically higher unit 6, a massive sandy silt to silty sand with primary dips that roughly mirror the topography of the scarp.

A dark-colored sandy silt to silty sand (unit 7) lies at the top of the stratigraphic sequence throughout the trench. Unit 7 has a silt loam to loam texture and exhibits a moderately developed, medium sub-blocky soil structure—similar in both respects to units 2 and 4 below. Unit 7 is thickest at both ends of the trench (i.e., the top and bottom of the scarp) and thinnest along the scarp slope.

Several faults cut the stratigraphic sequence in the Hornet trench (Fig. 9A). Fault F1 reversely offsets the entire stratigraphic sequence up through the unit 4. Fault F2 splays off of F1, and both faults form the sides of unit 5, a pendant-shaped gravelly sand deposit. Faults F3, F4, and F5 reversely offset and rotate layers of sandy gravels in unit 1, and development of modern soils (unit 7) destroyed evidence of the upper ends of the faults. Rotated clasts, some with near-vertical long axes, indicate the location of faults in unit 1. Dipping layers of unit 1 between the faults attest to folding either before or during faulting events. We found no indications of lateral motion on the faults but that does not preclude the possibility that all of these faults have reverse oblique motion.

In summary, we correlate stratified sandy gravels to similar deposits nearby mapped as late Pleistocene glacial outwash deposits dated between 13,960–12,400 cal. yrs old. Overlying these outwash deposits is a sequence of two dark-colored, loamy deposits containing charcoal with radiocarbon ages between 3200 and 930 cal. yr B.P.; both loamy deposits intercalate in a wedge-shaped deposit of stratified sandy gravel. We interpret these dark-colored deposits as buried soils between layers of gravelly colluvium. A series of faults cut through the entire section and reversely or obliquely offset both the glacial outwash and Holocene soils.

The stratigraphically lowest and oldest units in the excavation (unit 1) are light-colored (light browns and tans) well-bedded sands with thin gravelly sand layers (Table 3). We correlate unit 1 with undivided Pleistocene to Holocene glacial deposits, including glacial outwash, on the Mount Baker 1:100,000-scale quadrangle (Tabor et al., 2003). A layer of dark-colored silty sand with gravel (unit 2) overlies unit 1 in the northernmost part of the trench, and is absent throughout the rest of the trench. Unit 2 exhibits a loamy-sand texture and a weakly platy structure with thin clay films. A southwest-dipping contact juxtaposes a small part of unit 1 over the southern end of unit 2. Subsequent erosion removed this contact and the upper surfaces of unit 1 and 2 in the northern part of the trench, and boulder-rich pebbly sand (unit 3) overlies both units.

Unit 3 is a wedge-shaped set of deposits at the toe of the scarp. The wedge is thickest where unit 3 abuts unit 1 along a near-vertical contact (shown as a scarp free face on Fig. 9B). Unit 3 is thickest at the near-vertical contact between the southernmost part of unit 3 and unit 1, and thins toward the northern end of the trench. This thinning creates a wedge-shaped set of deposits at the toe of the scarp with primary dips closely mimicking the slope of the scarp above. A boulder-rich layer within unit 3 (unit 3A) sits directly on unit 1 and 2 at the toe of the scarp, and laterally changes abruptly into gravelly sand layers with steep primary dips that are somewhat steeper than the scarp slope above. This abrupt lithologic change coincides with a near-vertical contact that separates the boulder-rich unit 3A from the remainder of unit 3 (units 3B–3D). This same vertical contact also separates units 3B–3D from unit 4, a dark-colored, pebbly sandy silt that sits on the upper surface of unit 3A and unit 2 in the northern part of the trench (Fig. 9B). Charcoal from near the base of unit 4 has highly variable radiocarbon ages likely due to root disturbance (Beta-248645, 11,730 ± 60 ¹⁴C yr B.P.; Beta-248646, 40 ± 40 ¹⁴C yr B.P.; Table 1). Charcoal from near the top of unit 4 has a calibrated radiocarbon age of 4500–4240 cal. yr B.P. (Beta-248648, 3920 ± 40 ¹⁴C yr B.P.; Table 1).

A set of thin, gravelly sands—units 5 and 6—straddles the scarp near the top of the stratigraphic sequence in the trench. One of these layers, unit 5, occupies the northern third of the trench and overlies units 3 and 4. The second of these layers, unit 6, occupies the southern two-thirds of the trench exposure. The contact between units 5 and 6 lies directly above the near-vertical contact between units 3 and 1—the main differences between units 5 and 6 are an increase in the amount of gravel in unit 6 and a moderately developed sub-blocky soil structure in unit 5. A set of dark-colored, pebbly, silty sands, with sandy loam textures (units 7 and 8), blankets the surface of the stratigraphic sequence exposed by the trench. Charcoal from near the base of unit 7 has a calibrated radiocarbon age of 4240–3990 cal. yr B.P. (Beta-248647, 3760 ± 40 ¹⁴C yr B.P.; Table 1).

Folding of strata in the trench is best shown by anticlinal warping of distinct gravelly sands and thinly laminated silty sands in unit 1 (Fig. 9B). The apparent wavelength of the anticline is ∼12–13 m (almost the length of the excavation) because beds at both ends of the excavation are nearly horizontal. Vertical relief on the anticline is probably less than a meter in the southern end of the trench and at least 1.5 m in the northern end of the excavation near the steepest part of the scarp. A small normal fault (F3) and fissure offsets the crest of the anticline, and several small warps deform the limbs of the larger anticline.

Two reverse faults in the excavation offset the gravels, soils, and the oldest colluvium. Fault F1 strikes 334°, dips 39° SW, and offsets distinct sand and gravel beds in unit 1 and the buried soil (unit 2) formed on unit 1. Unit 3 overlies the top of fault F1. We are unsure how much, if any, of the anticlinal folding accompanied movement of fault F1. Fault F2 is near F1 but strikes 85° and is nearly vertical. Offset of individual beds of gravelly sand across F2 shows tens of centimeters of vertical separation but the beds change thicknesses abruptly, suggesting a lateral component of slip.

In summary, we correlate stratified sandy gravels in the Smuggler trench with similar deposits mapped as late Pleistocene glacial outwash deposits by Tabor et al. (2003). Overlying these outwash deposits is a sequence of two dark-colored, loamy sands containing charcoal with radiocarbon ages between 13,760 and 3990 cal. yr B.P.; wedge-shaped deposits of stratified and boulder-rich sandy gravel intercalate with the loamy sands. We discount a radiocarbon age of 40 ± 40 ¹⁴C yr B.P. as probable contamination from root stirring. We interpret these dark-colored deposits as buried soils within boulder-rich, gravelly colluvium. Two faults cut through the entire section; one fault reversely or obliquely offsets the glacial outwash and younger soils, and the second fault has apparent lateral offset.

Small ponds and wetlands adjacent to fault scarps, commonly referred to as sag ponds, provide an excellent environment to record past earthquakes (Fumal et al., 1993; Meghraoui and Doumaz, 1996; Sieh, 1978). Flooded areas and sag ponds act as catchments for relatively continuous sedimentation and also tend to preserve organic material useful for radiocarbon dating. Field reconnaissance of the Kendall scarp showed that the scarp blocked the flow of a small tributary stream flowing southward into the North Fork Nooksack River (Figs. 5 and 10). We surmise that wetland stratigraphy records scarp growth in the Holocene. Thus we investigated the stratigraphy of the wetland formed adjacent to the scarp with hopes of refining the history of scarp growth.

We collected a series of gouge cores along a transect across the wetland to document the stratigraphy and geometry of the wetland deposits (Fig. 10). A leveling survey related the elevation of each core location to a common arbitrary benchmark. We described each core using standard notation for soils and organic sediments (Schoeneberger et al., 2002; Troels-Smith, 1955), and analyzed core samples for microfossils, plant macrofossils, and ashes. Stratigraphy of the wetland cores showed cyclical deposition of gyttja and peat layers overlying a basal mineral deposit (each set of gyttja and peat layers is herein termed a “couplet”; Figs. 11 and 12).

Intercalated basal mineral deposits include dark greenish gray to very dark gray clay loam to sandy loam, more than a meter thick in some locations, and thickest in the middle of the wetland. Layers of clay loam are typically 2–7 cm thick and alternate with sandy loam layers averaging 4–15 cm thick. The clay loams typically have yellowish brown mottles and the uppermost sandy layers are very fine grained while deeper layers coarsen with depth. Overlying these basal inorganic deposits is a dark gray to black silt loam. This silt loam is more than 30 cm thick in places and organic content increases upwards to form a detrital peat. At the southern edge of the wetland the silt loam exhibits forest soil horizonation.

A light-brown diatomaceous gyttja, or sapropelic mud, overlies the basal black silt loam and detrital peat (Figs. 11 and 12). The contact between the gyttja and the underlying deposits is abrupt (∼1 mm in width). A yellowish brown volcanic ash (∼2 cm thick in most places) intercalates between layers of gyttja. Electron microprobe analyses indicate that this ash is from the climactic eruption of Mount Mazama in southern Oregon (Mazama ash, similarity coefficient = 0.98–0.99; 7780–7480 cal. yr B.P.). At some locations the ash is found within the previously described silt loam, but it is usually separated from the top of the silt loam by a thin layer of gyttja or diatomite. Immediately above the ash is a light-brownish-gray diatomite, and immediately above the diatomite is a gyttja. However, in one core, laminated diatomite lies below the ash (Fig. 11). The distribution of diatomite varies from a concentrated layer a few centimeters thick to diffuse diatom deposition throughout the base of the gyttja. In some cores, a 1-cm-thick charcoal layer separates the ash and overlying diatomite. The gyttja that overlies the diatomite contains equal concentrations of organic matter and silt loam, and is very dark gray to very dark brown and ranges from 4 to 20 cm thick. Common components include wood, charcoal, and pebbles. The gyttja component decreases upward where it grades into a very dark brown to black detrital peat (2 to 30 cm thick) that includes Tsuga heterophylla (western hemlock) needles, Thuja plicata (western red cedar) scales, Spiraea douglasii (hardhack) fruits, Rubus sp. (blackberry) seeds, herbaceous roots, wood, and charcoal. This sequence of gyttja with diatomite overlying detrital peat is the stratigraphically lowest and oldest of three gyttja-peat couplets in the wetland.

An abrupt contact (≤1 mm) separates gyttja-peat couplet 2 from the first couplet. The lower unit in the second couplet is a dark brown to dark grayish brown gyttja, 1 to 14 cm thick, that contains diatomite in several cores. The gyttja grades into a 2- to 23-cm-thick, very dark gray to black detrital peat with silt loam. Charcoal layers, 1 to 4 mm thick, make this unit conspicuous compared to bounding units. Wood and roots of herbaceous plants are common. A sample of Thuja plicata leaves and conifer needle fragments from the top of the peat in couplet 2 yielded an age of 3320–2990 cal. yr B.P. A sample of Thuja plicata leaves and conifer needle fragments from the gyttja immediately overlying the abrupt contact yielded an age of 3340–3080 cal. yr B.P. (Table 1).

Above the contact between the second and third couplets is a pronounced dark brown to very dark brown gyttja, 2 to 11 cm thick, sometimes with diatomite at its base (Figs. 11 and 12). Plant macrofossils are common above and below the contact, with Tsuga heterophylla needles, Thuja plicata scales, and bark marking the boundary between the underlying detrital peat of the second couplet from overlying gyttja that contains Polygonum sp. (knotweed) and Scirpus microcarpus (sedge) seeds. Above this uppermost gyttja is the youngest detrital peat, 10 to 40 cm thick. This very dark brown wetland peat includes herbaceous and shrubby roots, a wetland seed mixture, wood, charcoal, and silt loam. Seeds and plant material from the base of gyttja in couplet 3 yielded a calibrated radiocarbon age of 920–720 cal. yr B.P. (Table 1). Seeds from the uppermost peat/soil layers immediately below the gyttja yielded an age of 930–740 cal. yr B.P.

In summary, cores from the wetland contain a sequence of mineral and organic deposits. The basal sandy loam fines upward and is capped with a black silt loam; we interpret these as stream deposits capped by a thin soil. Overlying this thin soil is a series of alternating diatomaceous sapropelic muds (gyttja) abruptly overlying dark-colored detrital peats, grouped into three couplets of mud overlying peat. The lowest sapropelic mud contains a thin layer of Mazama ash. We interpret these muds as lake deposits and the peats as wetland soils, indicating that hydrology of the wetland changed abruptly at three times in the Holocene.

We interpret past earthquakes and localized environmental changes along the Boulder Creek and Canyon Creek faults from the scarp excavations and wetland stratigraphy. Evidence of past earthquakes in trenches consists primarily of folded and faulted late Quaternary sediments and soils, accompanied by deposition of fault scarp colluvium. Similarly, we interpret local environmental changes from fluctuations in local base level in scarp-adjacent wetlands as manifestation of past earthquakes (Fig. 13).

We found evidence for three earthquakes—named A, B, and C (from oldest to youngest) in subsequent discussions—in the past 7700 years on the Boulder Creek and Canyon Creek faults. Our OxCal modeling provides age estimates for these past earthquakes, with the best limiting ages for each earthquake coming from the marsh sequence (Fig. 14). The first earthquake (A) resulted in folding of glacial outwash deposits and warping of the ground surface in the early Holocene at about the same time as deposition of Mazama ash in the area. Two younger earthquakes (B and C) each ruptured the ground surface, displacing the surface soils at the time of the earthquake. The kinematics of the Kendall and Canyon Creek scarps are compatible; reverse faulting on the east-west–trending Kendall scarp would result in reverse and/or dextral slip on the SE-trending Canyon Creek scarp. Stratigraphy observed beneath each scarp is also broadly similar: a series of Holocene soils and colluvial deposits overlying late Pleistocene glacial outwash.

Trenches did not expose Tertiary bedrock, but we infer sandstone of the Chuckanut Formation is in the shallow subsurface because outcrops near each trench are composed of Chuckanut Formation rocks (Lapen, 2000; Tabor et al., 2003). Barnett (2007) inferred strongly magnetic rocks near the surface from anomalies in magnetic profiles across the Boulder Creek fault in the vicinity of the Kendall scarp. These magnetic anomalies likely represent faulted serpentinite conglomerate within the Chuckanut Formation or a sliver of serpentinite caught in the fault zone and brought to the surface by reverse faulting. A ∼2–10-m-thick veneer of glacial deposits covers the bedrock in the vicinity of the scarps.

Outwash terraces in the Columbia Valley (Fig. 4) grade up-valley and merge with ice-contact deposits near Cultus Lake (Kovanen and Easterbrook, 2001). The Columbia Valley terraces are inset into the terrace in the Nooksack River valley at Kendall, suggesting that the terraces in the Columbia Valley are younger than the terrace system in the Nooksack River valley. Elevation profiles across the scarps using LiDAR data show that the Kendall scarp deformed both terraces an equal amount (a minimum of ∼2.5 m and a maximum of ∼4 m; Fig. 8), suggesting that the earthquakes that formed the scarp occurred after deposition of the sandy gravels beneath the Columbia Valley and Nooksack River valley terraces (∼12,000 cal. years old).

We interpret the wetland stratigraphy as recording a response to repeated blockage of a small stream by growth of the Kendall scarp during surface-rupturing earthquakes. Ephemeral lakes sometimes form after earthquakes where surface deformation impedes flow in rivers and streams. An excellent example of this phenomenon is flooding along the Oued Fodda River following the 1980 El Asnam earthquake at Chlef, Algeria (Meghraoui and Doumaz, 1996). During the earthquake, a pressure ridge formed across the river and temporarily blocked water flow, causing extensive flooding. To better explain the Kendall wetland sequence, we developed a simple conceptual model of aquatic and wetland environmental succession as a response to changing hydraulic conditions before, during, and after blockage of a stream by a fault scarp (Fig. 13). We term these wetland environments that respond to earthquake-induced hydrologic changes “tectonic hydroseres” to differentiate them from climate-driven changes or changes driven by eutrophication. Each of these tectonic hydroseres in turn lead to distinctive wetland and lacustrine deposits, such as peat and gyttja couplets. Abrupt contacts between couplets marked the time of rapid ponding of the water behind a fault-scarp dam.

In the pre-seismic phase, relatively dry forest conditions and upland vegetation and soils occupied sites along the fault (Fig. 13). In the short post-seismic phase, the scarp dammed streams at the site, and a pond or small lake formed. Aquatic vegetation and algae deposited at the site resulted in deposition of organic deposits (peat, gyttja, and diatomite) containing freshwater fossils that confirm aquatic conditions after the earthquake. In the interseismic phase, the stream breached the scarp dam and reestablished drainage, the pond drained, and a wetland formed. Wetland vegetation (including sedges, rushes, grasses, and obligate wetland shrubs and herbs) occupied the site and peat was the main deposit preserved in the wetland. Eventually, the stream further breached the scarp, the wetland drained, and forest conditions invaded the former wetland to complete the cycle.

The cycle between pond and forest is partially complete today; ephemeral standing water occupies the western end of the impoundment during wet times of the year—erosion of the stream outlet is not complete (Fig. 10B). Standing water ∼1 m deep likely overtops the scarp during wetter periods and drains down the stream. The only outlet for the wetland is the stream across the scarp, and when the stream is not flowing, the wetland is internally drained.

Local landslides, regional climate changes, or beavers could alternatively drive changes in wetland plant succession and stratigraphy. LiDAR images of the wetland area show no evidence for local landslides that could have blocked the stream in the recent past. Climate changes are not a likely source for the stratigraphic changes we observed in the wetland based on regional climate studies. Pellatt (2001) inferred four regional climatic periods from analysis of pollen assemblages in varved cores from Saanich Inlet on Vancouver Island, British Columbia. Pellatt’s (2001) climate intervals are: (1) an early Holocene warm and/or dry period from 11,450 to 8300 yr B.P., (2) a warm interval with mild winters from 8300 to 7040 yr B.P., (3) a period of transitional middle Holocene climate from 7040 to 5750 yr B.P., and (4) establishment of a relatively cool and/or wet late Holocene climate after 5750 yr B.P. Modern conifer forests and oak savannas occupied the region by ca. 3800 yr B.P. None of the stratigraphic changes in the Kendall wetland temporally correlate with inferred regional climate shifts, nor are the cycles of wetting and drying observed in the climate record similar to those inferred from the wetland.

Beaver activity is another a possible agent for driving the wetting and drying cycles inferred from the Kendall wetland stratigraphy. Persico and Meyer (2009) described the predominant morphologic expression of beaver dams as a berm typically 5–50 m long built across a floodplain, approximately perpendicular to the valley axis. Berm height ranged from ∼0.3 to 1.5 m on the downstream side. Occasionally, fine-grained sediment filled the pond so that only the downstream face of the dam was exposed, creating a ramp-like feature up to 2 m in height. Most often, deposits behind beaver dams were seen in outcrops; for example, Wohl (2006) indicated that beaver dam abandonment resulted in channel incision through accumulated pond sediments and upstream alluvium, and in turn, resulted in a ravine cut through the beaver dam that exposed older beaver pond deposits. Johnson et al. (1987) described Holocene fluvial and wetland deposits created by beaver dams, an interpretation driven in part by observations of beaver-chewed sticks in exposures of wetland deposits in a creek incised through the former beaver dams. We observed no evidence of past beaver activity at the Kendall scarp. No abandoned prehistoric beaver dams were seen at the wetland nor does the stream incise the wetland, as expected in an abandoned beaver pond. In fact, the Kendall wetland deposits are all below the modern wetland surface and stacked on top of each other, which shows that the wetland aggraded to the modern surface—an unlikely result of beaver damming, where each breaching would have led to incision of the deposits impounded by the earlier dam rather than aggradation.

We interpret the stratigraphy and structures from the Hornet trench in the following series of events (Fig. 15): (1) glacial outwash (unit 1) being deposited after ca. 12 ka; (2) a forest soil (unit 2) forming on the outwash surface; (3) folding of the glacial outwash during earthquake A; (4) reverse or oblique reverse faulting on faults F3, F4, and F5 during earthquake B, displacing the outwash and forest soil (unit 2) formed on the outwash surface; (5) deposition of unit 3, the first scarp-derived colluvium and formation of a forest soil (unit 4) on unit 3, draping the scarp; (6) reverse or oblique reverse faulting during earthquake C on fault F1, offsetting units 3 and 4, and deposition of unit 5 in a small fissure created during earthquake C; (7) deposition of a second scarp colluvium (unit 6); (8) continued scarp erosion and soil development across scarp; and finally, (9) continued development of modern forest soil (unit 7).

We interpret the stratigraphy and structures from the Smuggler’s trench in the following series of events (Fig. 16): (1) deposition of glacial outwash (unit 1); (2) earthquake A folding but not faulting the glacial outwash (except perhaps bending moment faulting on F3); (3) a forest soil (unit 2) forming on the surface of the glacial outwash; (4) reverse or oblique reverse faulting on fault F1 during earthquake B, cutting units 1 and 2; (5) deposition of the oldest scarp colluvium (unit 3)—the scarp cut across a steep slope so unit 3 likely contained a component of slope colluvium as well (unit 3B), causing an increase in the thickness of unit 3; (6) formation of a soil (unit 4) on the surface of unit 3 colluvium and draped across the scarp; (7) strike-slip or oblique reverse movement on fault F2 during earthquake C, offsetting units 3 and 4; (8) deposition of the youngest scarp colluvium (units 5 and 6); and finally, (9) continued development of forest soil (units 7 and 8).

We infer that an earthquake along the Kendall scarp folded the glacial outwash and overlying forest soil (Fig. 15; earthquake A). After glacial ice left the area in the early Holocene, a thin forest soil developed at the site (unit 2). This soil was only preserved in the footwall of the present-day fault zone and was tilted almost the same amount as the underlying glacial outwash strata (Fig. 9A). Reverse faults separated distinct dip panels of glacial outwash exposed in the excavation. The dip panels showed that movement on the faults folded glacial deposits and created fold relief of 0.8–1.4 m (Table 4) during the earliest earthquake along the Kendall scarp (Barnett, 2007). Evidence for an early Holocene earthquake on the Canyon Creek scarp also consists of folded glacial deposits. We infer that some of the folding predated faulting on F1 and F2 in the Smuggler’s trench because both faults cut a buried soil (unit 2) that formed a slight angular discordance with the underlying sandy gravel. We note that the scarp just east of the Hornet trench was only ∼2 m high (see Fig. 5 and Stellar profile in Fig. 8E). However, if we include the possibility for broad folding extending as much as 200 m south of the scarp, then linear fits to the terrace surface allow the total displacement to increase to ∼3.2 m, agreeing with scarp heights we see elsewhere along the Kendall scarp.

We interpret earthquake A in the wetland adjacent to the Kendall scarp based on the sharp stratigraphic contact between a black silt loam at the top of the basal mineral deposits and a gyttja at the base of couplet 1. This sharp contact recorded an abrupt environmental change at about the same time as deposition of Mazama ash (Figs. 11 and 12). The abrupt stratigraphic change suggests that near-surface folding and warping during earthquake A grew the scarp high enough to block the stream, changing the local hydrology of the wetland from a riparian zone and/or forest to a small lake or pond just shortly before deposition of Mazama ash ca. 7780–7480 cal. yr B.P. (Zdanowicz et al., 1999). We used a GIS lake-filling algorithm to estimate the surface area of the lake at a maximum of 4.5 ha (∼45,600 m²) and a maximum depth of 3–4 m based on adjacent scarp height (Fig. 5). We interpreted the stratigraphic change from basal sands and forest soil to gyttja as evidence for wetland and/or lake inception and evidence of the first Holocene earthquake along the Boulder Creek fault.

The age of earthquake A is best estimated from the position of Mazama ash relative to the basal contact of couplet 1. The ash is intercalated in gyttja just above the basal contact, suggesting that the earthquake created the scarp and blocked the stream just prior to eruption and deposition of Mazama ash (Figs. 11 and 12). We infer that the earthquake and ash deposition were closely spaced in time—the earthquake possibly occurred within a few decades or centuries prior to ash deposition. Because of this close temporal relationship between earthquake A and Mazama ash deposition, we used an approximate age of Mazama ash for the age of earthquake A in our OxCal model (age of eruption rounded to the nearest century plus 50 additional years of error), yielding a modeled age for earthquake A of 8070–7240 cal. yr B.P. (Fig. 14).

Earthquake B resulted in reverse or oblique reverse displacement along the Kendall scarp that thrust glacial outwash over Holocene soils and colluvium. In the Hornet trench, we estimated ∼25–70 cm of vertical separation on the oldest buried soil (unit 2). We do not infer much vertical separation in the second earthquake in the Hornet trench (minimum of 25 cm), so it is possible that units 3 and 5 are the same stratum but now a separated by a fault with only a small amount of offset. Barnett (2007) estimated ∼70–170 cm of reverse vertical separation on unit 2 during earthquake B in a nearby trench (Table 4).

The second earthquake at Canyon Creek cut the folded outwash deposits along F1 and offset the ground surface (unit 2). We estimated that vertical separation of unit 2 was at least 1 m because erosion likely removed all of unit 2 from the hanging wall of F1. The unusual thickness of unit 3 suggests either a single large rupture or a series of smaller ruptures, neither of which we had evidence for. Alternatively, we suggest that contributions of both scarp and slope colluvium at the toe of the scarp following earthquake B caused the unique thickness of unit 3. In both trenches, the glacial outwash deposits were very loose and subject to collapse on exposed surfaces with steep faces, suggesting that the outwash deposits could form thick colluvial deposits over a short period of time. Radiocarbon ages on charcoal from three samples from unit 4 in the Smuggler trench poorly constrain the minimum age of earthquake B to between ca. 13,700 and ca. 260 cal. yr B.P. We attributed the wide range of ages to mixing of soil by roots and recycling of charcoal from older units (e.g., Nelson et al., 2003).

An abrupt change in the wetland stratigraphy from detrital peat at the top of couplet 1 to gyttja at the base of couplet 2 also signal a second Holocene earthquake (Figs. 11 and 12). We infer that uplift of the hanging wall along the scarp during earthquake B blocked drainage of the forested wetland and created a lake suitable for depositing diatomaceous gyttja.

We used our OxCal model to calculate the age of earthquake B from charcoal in the Hornet excavation and plant macrofossils from the wetland cores (Fig. 14). Pre–earthquake B samples of charcoal from unit 2 in the Hornet trench yielded an age of 3200–2890 cal. yr B.P. and plant macrofossils from the wetland yielded an age of 3340–3080 cal. yr B.P. Post–earthquake B macrofossil samples from the wetland yielded an age of 3320–2990 cal. yr B.P. Combining these ages in our OxCal model yielded a modeled age for earthquake B of 3190–2980 cal. yr B.P.

Movement on fault F1 in the Hornet trench records the youngest earthquake along the Kendall scarp (Fig. 15). Oblique reverse or reverse motion on F1 during earthquake C cut both the scarp-derived colluvium created by earthquake B (unit 3) and the overlying soil (unit 4). Vertical separation on the youngest buried soil (unit 4) and oldest scarp colluvium (unit 3) along F1 is ∼40–70 cm (Table 4). We estimated a minimum vertical separation of ∼45 cm because remnants of the cut-off soil were not evident in the hanging wall of F3.

The last earthquake in the Smuggler trench involved fault F2, which offset the youngest buried soil (unit 4) in the trench (Fig. 9B). Beds of gravelly sand in unit 1 showed tens of centimeters of vertical separation, but the beds also changed thickness abruptly across the fault, suggesting a component of lateral slip juxtaposed parts of the same strata with varying thickness. Faulting of unit 4 against unit 3B required at least 30 cm of vertical separation. An apparent change from reverse movement on F1 in earthquake B to oblique or lateral motion in earthquake C hints at possible partitioning of slip from the Boulder Creek fault onto the Canyon Creek fault.

In our OxCal model, we base pre-earthquake radiocarbon ages on a sample of charcoal from unit 4 in the Hornet trench that yielded an age of 1120–930 cal. yr B.P., and plant macrofossils from below the basal contact of couplet 3 in the wetland that yielded an age of 920–740 cal. yr B.P. For our post–earthquake C sample, we used plant macrofossils from the base of couplet 3 in the wetland that yielded an age of 920–720 cal. yr B.P. Together, these samples produced an OxCal-modeled age for earthquake C of 910–740 cal. yr B.P.

The Boulder Creek fault likely formed in the mid-Tertiary as a normal fault (Miller and Misch, 1963; Tabor et al., 2003), but we infer that the fault is now aligned in the modern stress field to accommodate north-south shortening as a reverse or oblique reverse fault. On the basis of the proximity of the Kendall scarp to the mapped trace of the Boulder Creek fault (Figs. 3–5), we infer that the fault associated with the Kendall scarp is the Holocene surface expression of the Boulder Creek fault. As shown in Figures 2B and 3, map relationships suggest that the Boulder Creek fault is a normal fault with Eocene Chuckanut Formation deposits in the hanging wall and pre-Tertiary ultramafic and metamorphic rocks in the footwall, and probably formed by extension in a dextral shear zone in the early or mid-Tertiary (Johnson, 1985). Reverse-fault reactivation of dormant normal faults is now recognized throughout many active fold-and-thrust belts around the world (Jackson, 1980; Sibson, 1985; Sykes, 1978; Wiprut and Zoback, 2000). Thus, the Boulder Creek fault and portions of the Bellingham Basin may be vestigial, normal-faulted extensional features now actively deforming in a north-south shortening strain system. Furthermore, other structural basins and thrust faults bordering migrating Cascadia forearc blocks may share similar complex histories.

The Canyon Creek fault is more enigmatic. Tabor et al. (2003) inferred a fault beneath the Nooksack River valley, immediately downslope of, and having the same trend as, the Canyon Creek scarp. Their map shows a fault with right-lateral and down-to-the-east separation south of Canyon Creek that places Chuckanut Formation rocks against pre-Tertiary rocks (see southwest corner of Fig. 4 and Grouse Butte on Tabor et al., 2003). It is possible that oblique motion on this inferred fault created the Canyon Creek scarp, but the scarp could also reflect a reactivated, previously unmapped, short splay of the Boulder Creek fault or Glacier Extensional fault. Additional LiDAR surveys in eastern Whatcom County may shed light on the style of Holocene faulting, if any, along the remainder of the Canyon Creek fault.

We infer that the three earthquakes on the Boulder Creek fault are the same as the three earthquakes on the Canyon Creek scarp because of the proximity and similar trends of the fault scarps. Lack of intervening scarps between the two mapped scarps could simply be the result of erosion by the North Fork Nooksack River. We do not see scarps along the Boulder Creek fault where the fault veers to the north and follows Boulder Creek toward the United States–Canada border. Instead, it appears that surface ruptures of the Boulder Creek fault continue eastward and link with the Canyon Creek fault. We envision surface rupture initiating on one fault and transferring to the other fault in the vicinity of the junction of Boulder Creek and the Nooksack River across an area no wider than several hundred meters. Such fault interactions between adjacent faults are now widely known from modern earthquakes. For example, surface rupture during the M7.9 Denali earthquake (Alaska) in 2002 jumped from the Denali fault to the Totshunda fault in the Little Tok River Valley, and paleoseismological evidence suggests this type of interaction happened repeatedly in the recent past (Eberhart-Phillips et al., 2003; Haeussler et al., 2004; Schwartz et al., 2012).

The Boulder Creek fault presents a seismic hazard to population centers in the Fraser-Puget Lowland. If we assume the two scarps rupture together, then the maximum length of the surface rupture visible on the LiDAR is ∼12 km. Maximum displacements per earthquake are difficult to estimate in the Smugglers Trench on the Canyon Creek fault, but Barnett (2007) estimated vertical separations of 80–140 cm and 70–170 cm, respectively, for earthquakes A and B on the Kendall scarp (Table 4). We estimated a vertical separation of 40–70 cm for earthquake C from the Hornet trench (Table 4). We used these vertical separations and ages for each paleoearthquake from our OxCal modeling to derive vertical slip rates (Table 5). We also used the vertical separations, ages, and three estimates for fault dip to calculate horizontal shortening rates according to methods summarized by Meghraoui et al. (1988).

Long-term slip rates calculated using the total scarp height on late Pleistocene glacial outwash terraces (Columbia Valley and Nooksack River valley) and ages of the terrace surfaces are 0.3 ± 0.1 mm/yr for the Kendall scarp and 0.65 ± 0.1 mm/yr for the Canyon Creek scarp. Rates calculated for intervals between earthquakes A–C, A–B, and B–C are 0.3 ± 0.1 mm/yr, 0.4 ± 0.1 mm/yr, and 0.5 ± 0.2 mm/yr, respectively (Table 5). Converting these rates based on vertical separation to horizontal shortening yields shortening rates between 0.5 ± 0.1 mm/yr and 1.0 ± 0.01 mm/yr for a fault dipping 30° (north-south–directed horizontal shortening rates are assumed and increase with lower fault dips, thus our calculations using a 30° dip likely approximate the highest rates expected for the Boulder Creek and Canyon Creek faults). Horizontal shortening rates calculated from GPS observations average ∼1 mm/yr for sites at the same latitude (49°N) as the Boulder Creek fault (Mazzotti et al., 2002). Thus, our calculated horizontal shortening rates and GPS-derived horizontal shorting rates agree quite well. However, GPS-derived shortening rates in excess of zero extend across the United States–Canada border to latitudes just north of Vancouver, British Columbia, suggesting that shallow faults in the North American plate in southern British Columbia are likely active as well. The 1946 Vancouver Island earthquake, as well as smaller more recent earthquakes in the region, suggest that shallow upper-plate faults in southern British Columbia are indeed active and pose a hazard to the northern Puget Lowland (Cassidy et al., 2000; Rogers and Hasegawa, 1978).

The Boulder Creek fault is capable of producing M_w 6.3 earthquakes, based on empirical relationships between rupture length and magnitude (Wells and Coppersmith, 1994). Empirical relationships between maximum displacement and magnitude indicate the fault is capable of larger earthquakes (M_w 6.6–6.8) (Wells and Coppersmith, 1994). Earthquakes of this size would produce ground motions sufficient to cause damage to urban areas in northern Washington and southern British Columbia, including Bellingham, Vancouver, Victoria, and Abbottsford.

We interpret observations from scarp excavations and coring of a wetland adjacent to the Kendall scarp as evidence for three Holocene earthquakes in the northern Puget Lowland. One of these scarps lies along the Boulder Creek fault, originally mapped as a Tertiary normal fault but now reactivated in the Holocene as a reverse or oblique reverse fault attributed to north-south forearc contraction. These findings place active faulting much further north than previously considered and raises questions about the possible presence of additional active crustal faults in southwestern British Columbia.

We thank landowners from The Glen at Maple Falls, Cowden Gravel, and the Washington State Department of Natural Resources for granting us access to their property to excavate trenches and collect wetland cores. Stephen Personius, Keith Knudsen, and Thomas Pratt provided helpful reviews of an earlier draft. Richard Blakely provided information on the size and shape of the Bellingham Basin. Constructive comments from two anonymous reviews greatly improved the manuscript.