The Holocene beach ridges at Turakirae Head, New Zealand, are remarkable because the fault that caused their uplift is accessible to paleoseismic trenching. Based on 40 14C samples from eight trenches, we identify five surface-rupturing earthquakes since ca. 5.2 ka (mean earthquake recurrence of 1230 ± 190 yr). The paleoearthquake record includes two more events than were recorded by the uplift and stranding of beach ridges at Turakirae Head. We conclude that beach ridges may provide an incomplete record of paleoearthquakes on oblique-reverse faults. The southern end of the Wairarapa fault includes several splays in the near surface at variable distances from Turakirae Head. Variable partitioning of slip between these splays (and perhaps the subduction interface down-dip of them) is inferred to have caused variable magnitudes of coseismic uplift at the coast, where at least one <3 m throw is not recorded by preservation of a ridge. Variations in wave climate or sediment supply (or interseismic subsidence) may also influence the number of beach ridges preserved by governing the morphology of the storm berm and controlling its extent of landward retreat. Such retreat may cause a berm to overwhelm, or amalgamate with, the next-highest beach ridge, resulting in the omission of one ridge, as probably happened at Turakirae Head at least once. Our 14C data support the view that a widespread post–Last Glacial Maximum aggradational terrace in southern North Island, New Zealand, was abandoned soon after 12.1 cal yr B.P. From this, we infer that the Wairarapa fault has a late Quaternary slip rate of 11 ± 3 mm/yr.
Slip on reverse (or oblique-reverse) faults during large earthquakes may be accompanied by a signal of coseismic uplift (or subsidence) near the coast that may be preserved in the geological record (Atwater, 1987; Berryman, 1993; Wilson et al., 2007a). Repeated surface-rupturing earthquakes have the potential to generate a suite of uplifted coastal beach ridges that faithfully record the sequence of earthquakes on that fault. Such a paleoseismically advantageous situation might be most likely where these earthquakes have all been similarly large, involving ruptures of similar dimensions, slip, and—especially—uplift; where coastal conditions have been continuously favorable for the formation of beach ridges; and where the preservation potential of the abandoned land-forms has remained steadfastly high.
New Zealand's largest historic earthquake, the Ms ~8.2 Wairarapa fault event in 1855, resulted in uplift of the hanging wall of that dextral-reverse fault near the southern coast of the North Island (Fig. 1A) and in the generation of a set of tsunami waves up to ~9 m high (Grapes and Downes, 1997). The coseismic uplift reached a maximum near Turakirae Head, where the pre-1855 storm beach ridge was raised by as much as 6.4 m (Begg and Mazengarb, 1996; Hull and McSaveney, 1996; McSaveney et al., 2006). This fossil beach ridge is today preserved as the youngest of at least four tectonically uplifted beaches on the headland (Fig. 1B) (Aston, 1912; Wellman, 1969; McSaveney et al., 2006). In addition to its large magnitude, the historically well-documented 1855 earthquake was remarkable for influencing Charles Lyell (1868) to argue that earthquakes were associated with vertical earth movements and slip on fault planes (Grapes and Downes, 1997; Sibson, 2006), and for its extremely large coseismic strike slip (locally as high as ~18.5 m; Rodgers and Little, 2006). Since 1855, it has often been assumed that the 1855 earthquake provided an analogue for the style of deformation accompanying previous earthquakes on this fault (e.g., Grapes, 1999). A corollary is that the uplifted gravel beach ridges at Turakirae Head provide a complete record of a series of similarly expressed earthquakes that have ruptured the nearby Wairarapa fault during the past ~7000 yr (e.g., Wellman, 1969; Moore, 1987; Hull and McSaveney, 1996).
Globally, the uplifted Holocene beach ridges at Turakirae Head are remarkable because the fault that caused their uplift does not lie submerged offshore but is exposed on land nearby and is accessible to paleoseismic study (e.g., Ota and Yamaguchi, 2004). In this paper, we determine the ages of surface-rupturing earthquakes on the southern Wairarapa fault using fault-trenching techniques and 14C dating. Undertaken on the same part of the fault known to have ruptured in 1855, and underpinned by more than 40 14C analyses collected in eight trenches, our paleoseismic data allow us to construct a comprehensive late Holocene earthquake chronology for the Wairarapa fault, and to compare it with the published ages of beach ridges at Turakirae Head. We are thus able to assess the completeness of that geo-morphic record and (potentially) the repeatability of large magnitude coseismic uplifts on the fault. Although other studies have compared beach ridge uplifts to the timing of historic earthquakes as much as ~500 yr ago (e.g., Bookhagen et al., 2006; Ferranti et al., 2007), these studies typically suffer from the short time span of the historic data and from ambiguities regarding the location of the surface rupture accompanying those historically felt earthquakes. To our knowledge, ours is the first study attempting a one-to-one comparison between uplifted strandlines and paleoseismically documented fault ruptures over a time span of ~5 k.y. Our results have general implications for the sensitivity of uplifted gravel beach ridges as a paleoseismic “tape-recorder” on exposed, high-energy coasts, and the variability of rupture styles on individual oblique-slip faults. Our results significantly shorten estimates of the mean earthquake recurrence interval of the Wairarapa fault and increase estimates for the late Quaternary slip rate of this major fault.
In central New Zealand, motion of the Pacific plate relative to the Australia plate occurs at ~39 mm/yr in a direction of ~261° (DeMets et al., 1990, 1994). The plate boundary in New Zealand is characterized by subduction of oceanic crust along the Hikurangi margin of the North Island and oblique continental collision along the Southern Alps of the South Island (Fig. 1A). Faults in the southern part of the North Island occur in a transition zone between these two plate boundary zones. There, obliquely convergent motion is partitioned between contraction-dominated folding and thrust faulting in the submerged accretionary wedge (e.g., Barnes and Mercier de Lepinay, 1997; Barnes et al., 1998) and in emergent parts of the forearc (e.g., Nicol et al., 2002, 2007), and strike-slip faulting (and vertical-axis fault-block rotations) in the onland region farther to the west (e.g., Beanland, 1995; Beanland and Haines, 1998; Mouslopoulou et al., 2007; Van Dissen and Berryman, 1996; Wallace et al., 2004). In the onshore region near Wellington, the margin-parallel component of plate motion (up to ~18 mm/yr or ~70% of the total) is mostly accommodated by dextral-slip on the Wairarapa, Wellington, Ohariu, and Shepard's Gully faults. These are the southernmost elements of the North Island Dextral Fault Belt, a belt of upper-plate strike-slip faults in the central part of the North Island (Beanland, 1995). The faults probably initiated ca. 2 Ma or more recently by reactivation of preexisting reverse faults, and they have since accommodated at most ~20 km of cumulative dextral slip (Beanland, 1995; Begg and Mazengarb, 1996; Kelsey and Cashman, 1995; Nicol et al., 2007). At the surface, the faults dip steeply (typically to the NW). Their dip slip is mostly up-to-the-west, but this component is typically a small fraction of the strike slip (<20%; e.g., Berryman, 1990; Heron et al., 1998; Rodgers and Little, 2006). In the offshore to the east and west of Wellington, upper-plate deformation is inferred to be dominantly contractional (e.g., Barnes et al., 1998, 2002; Lamarche, 2005).
Elastic dislocation modeling of global positioning system (GPS) data and seismicity data near Wellington suggest that the Hikurangi subduction zone occurs 20–25 km beneath the surface trace of these faults, and that this gently (~8°) west-dipping part of the plate interface is currently locked and accumulating elastic strain (Reyners, 1998; Darby and Beavan, 2001; Wallace et al., 2004). It seems likely that North Island dextral faults merge downdip with the underlying interface. The nature of interaction between the subducting plate interface and the intersecting North Island dextral faults is poorly understood.
The Wairarapa Fault
This NE-striking and steeply NW-dipping dextral-reverse fault bounds the western flank of the Wairarapa basin, a trough of late Cenozoic (mostly Pliocene-Pleistocene) marine and terrestrial sedimentary strata in the forearc of the Hikurangi subduction zone (Fig. 1A). The southern end of the basin is 2.5–3 km thick where it is truncated against the fault (Hicks and Woodward, 1978; Cape et al., 1990). To the west, the adjacent Rimutaka Range is >900 m high, consisting of deeply eroded Mesozoic basement rocks. Based on the gross geometry of its trace, the Wairarapa fault can be divided into central, southern, and northern sections (Fig. 1A). The central section consists of an en echelon array of mostly left-stepping fault segments that are typically 1–2 km long (Fig. A in Appendix).1 These discontinuous, dextral-oblique faults are separated by contractional bulges or folds in the area of their overlap that cause warping of alluvial terrace surfaces (Grapes and Wellman, 1988; Rodgers and Little, 2006).
The structurally complex southern section of the fault (Fig. 2) includes the west-dipping Wharekauhau thrust and other active and inactive fault splays. The Wharekauhau thrust is locally overlapped by undeformed late Quaternary–Holocene gravels; however, some strands of the Wharekauhau fault system show recent motion (Little et al., 2008). To the west, an inferred strike-slip fault is an apparent southward continuation of the main, central section of the Wairarapa fault (Begg and Johnston, 2000). Evidence for an along-strike continuation of this western strike-slip strand has not been found at the coast to the west of Turakirae Head (Begg and Johnston, 2000). Instead, that fault appears to step southward onto a thrust segment (Muka Muka fault) entering Palliser Bay just east of Turakirae Head (Begg and Johnston, 2000; Little et al., 2008). The Rimutaka anticline to the west is an active SW-plunging fold expressed by the steep topography of the coastal ranges and by differential uplift of the Holocene wave-cut coastal platform near Turakirae Head (Ghani, 1978; Wellman, 1969; McSaveney et al., 2006). Farther east, other structures at the southern end of the Wairarapa fault zone include Wharepapa and the Battery Hill faults (Begg and Johnston, 2000), and an inferred blind thrust along the western side of Lake Onoke (Little et al., 2008). Seismic and bathymetric data suggest that at least two strands of the Wairarapa fault zone continue offshore into Palliser Bay (Barnes and Audru, 1999; Barnes, 2005).
The northern section of the Wairarapa fault bifurcates eastward into a series of ENE-striking dextral-slip splays, such as the Carterton fault (Begg and Johnston, 2000; Lee and Begg, 2002; Langridge et al., 2005). The northern section is well defined as far north as Mauriceville, beyond which it links discontinuously into several more diffusely expressed, slower-slipping faults, such as the Alfredton fault (Schermer et al., 2004) (Fig. 1A). Estimates of the late Quaternary slip rate of the Wairarapa fault vary between ~7 and 12 mm/yr (Lensen and Vella, 1971; Van Dissen and Berryman, 1996; Grapes, 1999; Wang and Grapes, 2007), chiefly because of uncertainties in the age of the widely distributed and recognized (Begg and Johnston, 2000; Lee and Begg, 2002), but only sparsely dated, Last Glacial Maximum aggradation surface referred to locally as the “Waiohine” terrace.
1855 Earthquake and Raised Beach Ridges at Turakirae Head
The Wairarapa fault northeast of Wellington, New Zealand, ruptured on 23 January 1855, resulting in ground shaking landslides, regional uplift, tsunamis, and a surface rupture that broke all three of the aforementioned sections of the fault. The rupture was up to 120 km long on land and at most 40 km long on the seafloor to the SW (Grapes and Downes, 1999; Rodgers and Little, 2006). At Ms ~8.2, it was the largest earthquake in modern New Zealand history. South of Featherston, on a 16-km-long part of the central section of the rupture trace (Fig. A), Rodgers and Little (2006) mapped displaced landforms, mostly small beheaded or offset stream channels, and inferred an average single-event strike slip (in 1855) of 15.5 ± 1.4 m. There were no contemporary observations of coseismic strike slip in 1855. Three of the 16 sites yielded strike-slip estimates of 13–14.5 m for the penultimate earthquake, and several yielded estimates of the throw in 1855 (up-to-the-NW). At ~2–3 m, these throws agree with contemporary observations of the height of the scarp in the 1855 earthquake (Grapes and Downes, 1999). Based on the unusually large ratio of displacement to rupture length (D/L) for this earthquake, Rodgers and Little (2006) argued that the rupture extended several tens of kilometers downdip (W) to merge with, and co-rupture, a down-dip part of the subduction interface. This conclusion is consistent with elastic dislocation modeling of the vertical component of the coseismic motion, although it is not a uniquely determined aspect of the model (Darby and Beanland, 1992; Beaven and Darby, 2005).
A flight of uplifted gravel beach ridges occurs at Turakirae Head on the exposed and rocky southern coast of North Island, New Zealand. There, the crest of the modern storm berm (BR-1) is variably located 2–7 m above mean sea level, where it presumably marks the runup limit of present-day storm waves. The next highest beach ridge (BR-2) was abandoned as a result of uplift in 1855 (Begg and McSaveney, 2005; McSaveney et al., 2006). Contemporary observations (mostly by a surveyor; Roberts, 1855) and other geological data indicate that the 1855 uplift was approximately zero on the shore of Palliser Bay, to the east of the Muka Muka rocks (near the trace of the fault of the same name). Farther west, based on the elevation difference between BR-1 and BR-2, the 1855 uplift reached a maximum of ~6.4 m at the crest of the Rimutaka anticline, ~3 km NE of Turakirae Head (McSaveney et al., 2006). The 1855 uplift decayed westward to ~1.5 m near Wellington (Grapes and Downes, 1999). In addition to BR-1 and BR-2, Turakirae Head hosts at least three other higher and older, raised beach ridges: BR-3, BR-4, and BR-5 (Wellman, 1969). McSaveney et al. (2006) radiocarbon dated BR-3 (2380–2060 cal. yr B.P.) and BR-5 (6920–6610 cal. yr B.P.) and used the elevation difference between adjacent ridges as a measure of the coseismic uplift that caused the stranding of the upper ridge. In this way, they inferred an average incremental (coseismic) uplift of 7.3 ± 0.9 m at the crest of the anticline and a mean Holocene uplift rate of 3.5 ± 0.02 mm/yr at that location. Based on the premise that every Wairarapa fault earthquake since ca. 7 ka has caused preservation of a corresponding beach ridge at Turakirae Head, they calculated a mean recurrence interval for Wairarapa fault earthquakes of ~2200 yr.
For this study, eight paleoseismic trenches were excavated across three different sites, and 40 samples were submitted for 14C dating. See Appendix A for further information about our surveying and trenching methods. Next, we will discuss our paleoseismic results at the Pigeon Bush locality, then those at the Riverslea, and finally those at Cross Creek (Fig. 1A).
Pigeon Bush Site
Site Geology and Trench Stratigraphy
At the Pigeon Bush site (Figs. 1A and A), Grapes and Wellman (1988) interpreted two beheaded stream channels as evidence of the repeated dextral offset of a small stream gully crossing the Wairarapa fault (the last in 1855). The two channels are abruptly and orthogonally truncated at the fault on its SE side. On the NW side of the fault, a small, entrenched, and still-active source gully to these channels is similarly linear and fault-transverse (Fig. 3). This geomorphology implies that the two paleochannels were beheaded as a result of two consecutive earthquakes, and there is no evidence for a temporary phase of stream diversion parallel to the scarp (i.e., for a third or fourth slip increment; Rodgers and Little, 2006). These authors measured 18.7 ± 1.0 m of dextral slip and ≥1.25 ± 0.5 m of vertical slip for the younger, abandoned channel relative to the upstream gully and 32.7 ± 1.0 m and ≥2.25 ± 0.5 m of slip (respectively) for the older abandoned channel. This larger offset suggests that the older beheaded channel had previously been displaced by 14.0 ± 1.0 m of dextral slip and ~1.0 m of vertical slip prior to incision of the younger one.
The fault displaces the “Waiohine” terrace gravels and forms a steep, SE-facing scarp that is ~6 m high. On the northern side of this scarp, the uplifted Waiohine terrace is tilted SW, whereas on its southern side, it is partly buried beneath younger deposits. On the SE side of the fault, an ~1-m-thick layer of silt (unit Si) mantles the terrace gravel. This layer was later incised by the two channels (Figs. 4A and 4C). Wang and Grapes (2007) dated two samples of the silt cover bed by optically stimulated luminescence (OSL) methods (Fig. 3) and obtained ages of 7.0 ± 0.5 ka and 4.3 ± 0.5 ka.
Trench PB-1 was cut orthogonally across the younger of the two abandoned channels (Figs. 4A and 4B; Appendix B). Due to a meander in the channel (see Fig. 3), it was excavated in a nearly orthogonal (and therefore narrow) section along the NE wall of PB-1 (Fig. 4A) and in a more oblique (and therefore wider) section along the SW wall of the same trench (Fig. 4B). The latter wall exposed a richer and more easily differentiable stratigraphic sequence and was logged in detail (Fig. 4B). In both of the Pigeon Bush trenches, fluvial terrace gravels at the base of the trench (e.g., units Tgr, Tgr-1, and Tgr-2) are overlain by up to ~75 cm of silt (see unit Si in Figs. 4A and 4C). This terrace covering layer of silt is probably what was OSL dated by Wang and Grapes (2007). The aforementioned terrace units were incised to a depth up to ~2 m by the younger stream channel (Fig. 4A). None of these stream-incised terrace units yielded dateable organic material. The scoured unconformity defining the base of the channel incision is mostly overlain by bedded gravel and sand, including units Sgr and Fgr-1 to Fgr-5 in Figure 4B. These units are compact and well sorted, with rounded to subrounded clasts, and they are interpreted to be axially transported fluvial deposits. The remaining channel-infilling units are interpreted, on the basis of their poor sorting and clast subangularity, to be colluvial units derived by lateral collapse of the channel banks (e.g., units Cg1 to Cg7 in Fig. 4B).
The fluvial part of the channel-infill sequence contains abundant detrital charcoal fragments up to 3 cm in diameter, and these are locally inter-bedded with an organic-rich paleosol (unit Ps). Four charcoal samples from the channel infill were radiocarbon dated. The first three (PB-1, PB-2, PB-3) yielded ages of 649–497 cal yr B.P. (Table 1). These overlap (at 95% confidence) with each other and with two charcoal samples (DR0425G and DR0425J) collected and dated from a nearby pit by Rodgers and Little (2006). All six samples are apparently derived from a single population of charcoals delivered into the channel by a stream that aggraded rapidly(?) after a bush fire. Based on the age of the youngest of these samples (PB-1), the fire probably took place at, or soon after, ca. 546–497 cal yr B.P. (1404–1453 cal. yr A.D.), perhaps in response to Maori burning. Another sample, PB-5, was collected in the modern soil, 32 cm below the ground, and it yielded an age of 490–315 cal yr B.P.
Trench PB-2 was cut orthogonally across the older of the two beheaded channels (Fig. 4C). Fluvial terrace gravels at the base are scoured beneath an ~1-m-deep, channel-bounding unconformity. The channel is infilled by a massive, matrix-supported pebbly clay, which is interpreted as a debris flow. No fluvial deposits are present. A single piece of charcoal near the top of the debris flow (PB-22) yielded a 14C age of 543–495 cal. yr B.P., which suggests that it is another element of the aforementioned burn population that had become entrained into the debris flow.
Interpretation of Surface-Rupturing Events
Each of the last two earthquakes on the Wairarapa fault at this site resulted in abandonment of a stream channel immediately downstream of the narrow headwater gully and incision of a new channel in downstream continuity with that gully. Hoping to date the last two earthquakes, we excavated stratigraphic trenches at right angles to each of the two channels to date their incision and abandonment. The only organic material that we found (charcoal) occurs as detrital particles within the fluvial deposits that infill the incisional scour that defines the youngest beheaded channel. This channel was cut immediately after the penultimate earthquake, resulting in the unconformity between the “Tgr” and “Si” units and the younger units above them that infill the scour (these are shown in gray patterns on Figs 4A and 4B). This channel was later displaced and abandoned as a result of slip during the most recent earthquake on the fault. Thus, our preferred age of the burn event (ca. 546–497 cal yr B.P.) provides a minimum age constraint for the penultimate earth-quake at this site (event Pb2, Table 2). Historical data indicate that the youngest earthquake here (event Pb1) took place in 1855.
Riverslea Station Site
Site Geology and Trench Stratigraphy
On both sides of Manganui Stream, series of NE-striking, discontinuous scarps traverse steep hillslopes underlain by undated early Quaternary(?) gravels (Figs. 1A and 2). In two places, these scarps offset small gullies by 6–8 m dextrally and 1–3 m in an up-to-the-NW sense (Little et al., 2008). The linearity of these traces suggests a steeply dipping fault. The Riverslea trenches are located on a late Holocene river terrace just west of Manganui Stream, where Begg and Mazengarb (1996, p. 84) suggested that a fault may have ruptured in 1855, as is consistent with accounts of that earthquake's rupturing southward to the coast (Grapes and Downes, 1997). Trench RV-1 was excavated perpendicular to the main ENE-striking, ~3-m-high topographic scarp on the terrace (Fig. 5). Trench RV-2 was placed slightly higher up this slope to intersect a less conspicuous scarplet on the uplifted, NW side of the main scarp.
The strata in trench RV-1 exhibit no faulting to ~2 m depth, but they are apparently folded (Fig. 6A; Appendix C). Loose, unstable fluvial sands (units ss-1 to ss-6) and gravels (units gvl-1 to gvl-4) prevented deeper trenching. Most strata (excluding cross-beds) dip 20°–25° to the SE. The decimeter-scale cross-beds mostly indicate SE flow subparallel to the modern Manganui Stream. The basal contacts of gravel bodies are channelized and exhibit a cut-and-fill relationship to surrounding beds. We interpret the sequence to be tectonically tilted because: (1) the SE dip of the beds accords with the SE facing direction of the scarp; and (2) these dips appear too steep to be primary. By analogy with the small and shallow morphology of channels in present-day Manganui Stream, it seems unlikely that fluvial sediments, in particular parallel-laminated sands (Fig. 6B), would be deposited at a primary dip of 23°–25° in beds that are several meters long. Unfortunately, the trench was not long enough to convincingly expose the base of the main scarp to the SE where these beds presumably return to horizontal, although a small lens of gravel and sand at the extreme SE end of the trench (unit gss) appears horizontal. A sample of detrital charcoal from the sands and fine gravel of unit ss-3 (sample RV-1, Table 1) yielded a 14C age of ca. 1300 cal yr B.P.
Trench RV-2 was excavated across the decimeter-high scarplet to the NW of the main scarp (Fig. 5). There, fluvial sands, silts, and channelized gravels similar to those in trench RV-1 are vertically offset by several steeply dipping faults (Fig. 6C; Appendix C). Sand bed ss-7 is separated vertically in an up-to-the-NW sense by ~20 cm across fault strand 2, and by ~50+ cm across fault strand 1 (assuming that either ss-2 or ss-3 is equivalent to ss-7). Near the base of the trench, fault strands 2 and 3 cut a thick silt unit (si-9a and si-9b). Both of these faults terminate upward against a cut-and-fill sequence of depositionally overlapping, younger gravel (unit gvl-18). Unit si-9b contained a centimeter-diameter fragment of fibrous wood that yielded a 14C age of 897–722 cal yr B.P. (sample RVL-4, Table 1).
Abrupt truncations and thickness changes across these and other faults in the trench suggest strike-slip motion. Units gvl-5b, and gvl-17 and si-5 (which includes a distinctive sandy interval) are truncated against the NW side of fault strand 1 without any apparent correlatives being found on the SE side of the fault. Fault strand 1 diverges upward into a wedge-shaped, soil- and colluvium-infilled fissure just beneath the topographic surface (Figs. 6C and 6D). Charcoal from unit si-15a in the faulted walls of the fissure yielded a 14C age of 547–497 cal. yr B.P. (sample RVL-2, Table 1). Charcoal from in the fissure infill (unit cgvl-16, sample RVL-3, Table 1) yielded a 14C age of 0–264 cal yr B.P.
Interpretation of Surface-Rupturing Events
The two Riverslea trenches contain evidence for two earthquakes (events Rv1 and Rv2 in Table 2). In trench RV-2, the most recent earthquake (Rv1) ruptured to the surface and cut sediments younger than ca. 500 cal. yr B.P. to create the fissure. The fissure was infilled by soil debris containing “modern” charcoal and is thus consistent with this earthquake being the 1855 rupture. Based on the stratigraphic mismatches across fault strand 1, slip during the fissure event is inferred to have been primarily strike slip. In the same trench, evidence for an older (penultimate) earthquake (Rv2) includes the depositional overlap of fault 3 (and some splays of fault 2) by undeformed channel gravel (gvl-18). A sample of fibrous wood from the faulted silt (unit si-9b) below this unconformity to the SE of fault 2 yielded an age of 897–722 cal yr B.P. (Fig. 6C). We are uncertain whether this sample was detrital or a root fragment. If it is detrital, then its age (897–722 cal yr B.P.) must predate the penultimate earthquake at this site (Rv2). If it is a root fragment, then it does not necessarily predate that earthquake. Sample RVL-2 must postdate the penultimate earthquake.
In trench RV-1, near-surface bulging of the terrace sediments is inferred to have raised the main scarp intersected by that trench, at least in part during the 1855 earthquake. Six measurements of coseismic throw on the southern part of the Wairarapa fault's central section from the 1855 event range up to a maximum of 2.5 m in a NW-up sense (Rodgers and Little, 2006). The height of the main scarp at Riverslea, ~3 m (Fig. 5), suggests its growth in two stages. If so, these must postdate sample RV-1; that is, the penultimate earthquake must be younger than ca. 1300 cal yr B.P.
The lack of a fault in trench RV-1 and the small offset in RV-2 suggest that the primary mode of deformation to create the northwest-up scarp on the late Holocene terrace was folding. The cause of folding is interpreted to be dip slip on a northwest-dipping fault (concealed beneath the trenched gravels) that forms a part of the (here complex) Wairarapa fault zone. Based on the stratigraphic mismatches across fault strand 2, we infer that a small (several-meter?) strike-slip surface displacement accumulated in 1855 on a steep splay fault in the hanging wall of the blind structure in response to dextral-reverse slip on the main structure at depth.
The Cross Creek Pull-Apart Graben
Four trenches were excavated across opposite sides of a pull-apart graben along the Wairarapa fault at Cross Creek (Figs. 1A and 7A; Fig. A). Trenches CC-1 and CC-4 were excavated across the graben's southern bounding fault (Fig. B-i), whereas trenches CC-2 and CC-3 were excavated across its northern margin (Fig. B-ii). The swampy pull-apart graben is today watered by a small southward-flowing stream that traverses a series of diffuse scarp-lets on the north side of the main depression before it exits from the western end of the graben (Fig. 7A). The drier eastern end of the graben is abutted by a tilted terrace surface overlain by a small inactive alluvial fan (Fig. 7B; Fig. B-ii). The graben is down-faulted into the regionally extensive, post–Last Glacial Maximum (LGM) Waiohine terrace gravels (Begg and Johnston, 2000).
Hand augering revealed a continuous layer of peat above the graben's down-dropped substrate of terrace gravel. This peat thickens southward to at least ~3.8 m near the southern boundary fault (Fig. 7C), where a large piece of wood (sample Auger-3) was intersected at ~1 m above the base of the peat (Fig. 7C; Table 1). This wood was part of a log (others were extracted from the peat by the digger) that yielded a 14C age of 5580–5300 cal yr B.P. In the nearby trenches, four other wood samples (CC-4-11, CC-2-37, CC-2-35, CC-2-33 in Table 1; also Figs. 8 and 9) yielded 14C ages that are indistinguishable (at 95% confidence) from this age. A fifth wood sample in trench CC-2 (CC-2-38) was slightly younger (5209–4842 cal yr B.P.). This suite of six similar-aged wood samples was collected from different levels within the peaty graben infill (not just at its base). The 14C ages of these other wood fragments are everywhere older than that of underlying peat samples. For this reason, we interpret these wood fragments to have been recycled from a downed forest horizon near the base of the peat (the one sampled by the auger), and we establish the stratigraphic chronology in the trenches chiefly on the basis of 14C ages of peat, rather than of wood.
Trenches across the Southern Bounding Fault of the Cross Creek Graben
Stratigraphy and Structure of Trenches CC-1 and CC-4
Adjacent trenches CC-1 and CC-4 were excavated across the steep, ~2-m-high scarp defining the southern margin of the pull-apart graben (Figs. 8A and 8B801). Enlarged versions of the logs for these two trenches are provided as Fig. C, and detailed unit descriptions are given in Appendix D. The fault zone consists of an ~1-m-wide zone of up to five fault strands that dip steeply NW. Slivers of clay-rich sheared gravel and peat (unit sg) occur between some fault strands. Fault strand 1 (in both trenches) is composed of an ~5-cm-thick zone of clay gouge. Only fault strand 2 (in both trenches) appears to cut upward into the modern soil profile. To the NW of the zone, a sequence of peats (units pt1, pt2, pt3, pt4, pt5, and peat undiff.) occurs interfingered with locally derived units of clastic sediment, which we interpret as scarp-derived colluvium. These down-faulted basinal deposits are juxtaposed against fluvial terrace gravels to the SW of the fault zone (units tgr-1 to tgr-4, tg, stg, and bg). The cobble-bearing terrace gravels are locally interbedded with sandy layers (ss, bss, gs, stg, and sd). In trench CC-1, an ~20-cm-thick layer of peat (pt6) is interbedded with gravel and sand at a depth of ~2.3 m below the terrace tread.
On the down-thrown side of the fault, we were unable to excavate deeper than ~2 m because of wet, unstable ground, so the base of the peat was not exposed in either trench. In the NW parts of both CC-1 and CC-4, fault strands cutting peat terminate abruptly upward beneath younger, unfaulted sediments along a depositional contact that we refer to as the “intra-peat unconformity.” On the NE wall of trench CC-1, fault strands 4 and 5 are overlapped by the undeformed peat unit, pt4 (Fig. 8A). The SW wall of CC-1 collapsed before it could be logged, as did the NE wall of trench CC-4. On the latter, we photographed and sampled another exposure of the intrapeat unconformity just before the wall collapsed (Fig. 8C). There, a steep fault strand (probably equivalent to fault strand 4 or 5 in CC-1) juxtaposes peat against silt. This fault is depositionally overlapped by an unfaulted layer of silt. The intra-peat unconformity was not exposed on the SW wall of CC-4 (Fig. 8B).
Discontinuous bodies of clastic sediment abut the fault zone to the SW and are interlayered with peaty units to the north. These clastic units include cw1, cw2, and cw3 in trench CC-1, and units cwA, cwB, and cwC in trench CC-4. All but two of these units are thickest at or near the fault, and pinch out northwestward into peat to define a wedge-shaped body that is >1.5 m long. All are truncated on their SE side against a fault, except unit dp, which occurs as small isolated blob in peat (in CC-4), and the large faulted wedge cw2 (in CC-1), some of which extends across the uplifted side of the fault zone. At the NW end of cw2, the basal contact of this wedge truncates an underlying tree in growth position. The clastic units are interpreted by us to be colluvial bodies derived by redeposition of terrace gravels eroded from the uplifted side of the fault. We distinguish two of them (unit cw1 in trench CC-1 and the aforementioned blob, dp, in trench CC-4) as being smaller (<40 cm long), more organic-rich, and more lenticular in shape than the other, wedge-shaped bodies, leading us to interpret the small gravel-bearing bodies differently (see following). The youngest colluvial wedge in trench CC-1 (cw3) is unfaulted by strand 3, but it is not in contact with fault strand 2. This wedge is overlain by an organic silt layer containing fragments of steel wire (unit wl). The youngest clastic wedge in CC-4 (unit cwC) is truncated by fault strand 2 in that trench and appears to be depositionally overlain by a unit of organic silt (unit mo) containing a line of cobbles (queried contact in Fig. 8B).
Eleven samples were 14C dated from the NE wall of trench CC-1 (Fig. 8A), six from the SW wall of trench CC-4 (Fig. 8B), and two from the unlogged (collapsed) NE wall of CC-4 (Fig. 8C). Of these, three are wood, and 16 are peat or organic clay-silt (Table 1). Two samples (CC-1-1a-i and CC-1-1a-ii) were collected from a peat that occurs interbedded with alluvial gravels on the uplifted fault block. These yielded ages between 13,160 and 12,100 cal yr B.P. (at 95% confidence). The rest of the samples were collected from the infill of the graben on the downthrown side of the fault, yielding ages of <5500 cal yr B.P. One wood sample (CC-4-11) yielded an age that was greater than that of underlying peat samples, and it is inferred to have been recycled from a near-basal part of the peat sequence. All the other samples in these trenches yielded ages that are in the correct stratigraphic order (at 95% confidence), with the exception of two peat samples (CC-1-11 and CC-4-3) that were collected in proximity (above and below) to the intrapeat unconformity. These yielded ages discordant to one another and to the remaining set of nearby samples (from both above and below the unconformity). After considering the full distribution of 14C ages, we interpret CC-1-11 and CC-4-3 as recording non-depositional events, though the reasons for this are unknown (see Appendix E). Our interpretation honors the full complement of remaining 14C ages, whereas any other explanation for the age reversal would require additional ages to be rejected. It is important to note that other interpretations of the 14C data would not affect the total number of surface rupturing events interpreted from the Cross Creek trenches.
Interpretation of Earthquakes Rupturing the Southern Bounding Fault of the Graben
On the basis of the combined data from trenches CC-1 and CC-4, we interpret five earthquakes to have ruptured the southern bounding fault of the Cross Creek graben during the past ~5.2 k.y. An older (sixth) event occurring after 12–13 ka is expressed in trench CC-1 by overlap of fault strand 3 by terrace gravel (unit tgr-3), but the minimum age of this earthquake is poorly constrained, and this event will not be discussed further. In Table 2, the five youngest events are labeled CCS1 (youngest earthquake on the southern boundary fault) to CCS5 (fifth-oldest earthquake on the southern boundary fault). Table 2 also identifies the specific 14C samples that constrain the maximum and minimum age limits for each of these earthquakes. All event ages are quoted in calibrated years B.P. at the 95% confidence interval. Figure 10 plots these age ranges and also labels the key 14C samples used to bracket these age ranges.
There is an along-strike difference in exposed stratigraphic age between the two southern trenches; older sediments are exposed in CC-4 compared to in CC-1. We attribute this difference in exposure level to slight up-bulging of strata along the fault near CC-4 (see Appendix E). Because of this NE structural plunge along the fault, only trench CC-4 recorded evidence for the oldest event (CCS5). CCS5 resulted in the formation of the large colluvial wedge (cwA) exposed in the lower part of trench CC-4. Radiocarbon samples CC-4-6 and CC-4-10, from below and above this wedge, bracket this earthquake to the interval 5450–4620 cal. yr B.P.
We interpret the next-youngest earthquake, CCS4, to have caused refreshment and collapse of the scarp, leading to emplacement of the colluvial wedge, cwB. The age of this wedge is bracketed by samples CC-4-16 and CC-4-13 to the interval 4870–3070 cal. yr B.P. In addition, we infer that the same earthquake was recorded by the intrapeat unconformity—the draping-over of fault strands 4 and 5 by peat in trench CC-1 (Fig. 8A) and the unconformable overlap of the unnamed fault by silt in trench CC-4 (Fig. 8C). We bracket the unconformity (and thus CCS4) to the interval 3690–2970 cal. yr B.P. using samples CC-1-12 and CC-1-10. A combination of the age constraints of the colluvial wedge with those of the unconformity yields a composite age range for event CCS4 of 3690–3070 cal. yr B.P.
The third-youngest earthquake (event CCS3) rejuvenated the scarp to cause formation of colluvial wedge cw2, as exposed in trench CC-1. The age of this wedge is bracketed by samples CC-1-6 and CC-1-13 to the interval 2340–740 cal. yr B.P. Our preferred age for this earthquake is based on sample CC-1-6 alone, as this age (2340–2110 cal yr B.P.) is interpreted to record death of a tree by earthquake-induced toppling immediately prior to emplacement of the wedge. In trench CC-4, the same earthquake (CCS3) is interpreted to have formed the wedge, cwC. A maximum age constraint for wedge cwC of 3340 cal. yr B.P. is provided by sample CC-4-13, which underlies it (there are no dated samples from above the wedge).
In trench CC-1, we infer that the penultimate earthquake (event CCS2) caused emplacement of a colluvial wedge, cw3. This wedge draped across the preexisting (and gouge-laden) trace of fault 1. The age of the wedge is bracketed by samples CC-1-13 and CC-1-14 to the interval 920–800 cal. yr B.P. We infer that the uppermost units in CC-4 are condensed or have been in part eroded, perhaps as a result of deforestation and agriculture. In CC-4, a diffuse line of cobbles in the mo unit near the ground surface may be a human-disturbed equivalent to wedge cw3, but this is uncertain.
Inferred on historical grounds to be the 1855 earthquake, the most recent earthquake (CCS1) is expressed in trench CC-1 by the rupturing of fault strand 2 upward from wedge cw3 to extend into the modern soil profile. A maximum age constraint of 970 cal. yr B.P. is provided by sample CC-1-14 from the faulted om layer, which is overlain by the wire-bearing layer, wl (undated, but assumed modern). In trench CC-4, the 1855 earthquake is inferred to have caused slip on fault strand 2, but it did not generate any colluvial wedge that is preserved near the ground surface today. If such a wedge once did exist, human-induced disturbance caused by deforestation and cultivation must have removed any evidence for it from the uppermost sediment layers.
Inferring Earthquakes from Colluvial Wedges
These interpretations rely in part on our interpretation that the large (>1.5-m-long) gravel wedges are scarp-derived colluvial units that formed as a result of fault-scarp rejuvenation during earthquakes. We infer that in the densely forested lowland settings of precolonial New Zealand, gravitational collapse of earthquake-induced fault scarps was the chief process by which large bodies of terrace gravel could be eroded from a fault scarp and redeposited in an adjacent peat basin. This interpretation was not applied, however, to the two smallest (tens of centimeters long) gravel-bearing bodies near the scarp of the southern boundary fault: specifically, units cw1 in CC-1 and dp in CC-4. This difference in interpretation was based on our consideration of the following typical attributes of earthquake-induced colluvial wedges: (1) large size (consistent with generation of a meter-high fault scarp); (2) wedge shape (thickest at the fault, where they may be truncated, and thinning away from the fault); (3) composition of clasts equivalent to exposures in the adjacent fault scarp; (4) texture consistent with transport of these clasts down the scarp; and (5) synchroneity to other wedges along the fault (or at least to other types of evidence for earthquakes along that fault).
The larger gravel bodies that we have interpreted as earthquake-induced colluvial wedges fit all of these criteria, whereas the two decimeter-sized bodies (units dp and cw-1) fit few of them. The size of the former is most plausibly attributed to the scale of coseismic shaking, scarp rejuvenation, and subsequent scarp erosion that would accompany inferred fault throws of ~1–2 m on the Wairarapa fault (Rodgers and Little, 2006). By contrast, the other two bodies are small, isolated (dp is a disconnected blob), and have a lenticular rather than wedge shape. Although both types of gravel body have apparently been derived from redeposition of the terrace gravels derived from the uplifted side of the fault, clasts in the decimeter-sized gravel bodies are supported by an organic-rich or peaty matrix, whereas the matrix of the large wedges is organic-poor. Perhaps the small bodies formed by adhesion of gravel clasts onto the roots of toppled trees. Excluding the 1855 earthquake (evidence for this youngest event appears not to be well-preserved today), all the “large” wedges at Cross Creek can be temporally correlated to other wedges (or at least earthquakes) in one or more other trenches, implying their lateral continuity along the fault. By contrast, none of the two cited “small” redeposited gravel bodies, although they are well dated (Fig. 10), could be recognized beyond a single trench wall. Next, we will show that most of the colluvial wedges at Cross Creek can be correlated across both sides of the graben.
Trenches across the Northern Bounding Fault of the Cross Creek Graben
Stratigraphy and Structure of Trenches CC-2 and CC-3
Trenches CC-2 and CC-3 were excavated across the northwestern margin of the pull-apart graben. We logged the NE wall of CC-2 (Fig. 9A) and both walls of CC-3 (Figs. 9B and 9C). Many of the stratigraphic units can easily be correlated between these three walls, so we have adopted a set of (in part) common unit names that reflects our correlation (Figs. 9A, 9B, and 9C; Appendix F). This part of the Wairarapa fault zone consists of several SE-dipping fault strands. The fault numbering in Figure 9 reflects our interpretation of how these strands correlate between the trenches. Fault slivers of strongly sheared, clay-matrix gravel (sg and stg) and peat (sp1 and sp2) contain gravel clasts that are rotated to a steep dip against the fault. In addition to their content of colluvial wedges, sediments along the northern margin of the graben reveal clear evidence of progressive deformation, such as differential tilting, angular unconformities, and fissuring. These relationships reduce the potential ambiguity of earthquake identification. Enlarged versions of the trench logs are provided in Figures D and E, and detailed unit descriptions are given in Appendix F.
None of the fault strands on the northern side of the graben cuts to the surface, and the topographic scarp is offset ~5–6 m southward relative to the subsurface fault zone. These relationships reflect lateral accretion of the uppermost gravelly layers across the fault scarp by anthropogenic processes. Wire fragments found in the s-col unit (Fig. 9C) indicate that both this unit and the overlying u-col consist of fill material pushed southward from the site of the nearby road during its excavation (Fig. 7B).
Fluvial terrace gravels (tg) in the uplifted foot-wall of the fault zone are juxtaposed across the northern boundary fault zone against a basin to the SE that is dominated by peat and organic silt. On the SE side of the fault, the down-dropped terrace gravels form an exposed depositional substrate to the peat-rich basin fill. The terrace gravels (tg) are locally capped by layers of gravelly sand and silt (units slt, gs, and ssg). These fluvial deposits are overlain by a much finer-grained sequence of organic-rich silts and peats (units pt, osi-1, osi-2, op, and pt1, pt2, pt3, and pt4). About 7 m to the SE of the fault, a deformational bulge is expressed by folding and erosion of part of the organic-rich basinal infill (Fig. 9A).
Near the fault, the peaty basinal units are interfingered with three southward-tapering gravel wedges. These wedges are progressively faulted and tilted. They are labeled, from oldest to youngest, co-1, co-2, and co-3 (Figs. 9A, 9B, and 9C). Truncated against the fault zone, and consisting of poorly sorted pebble and cobble gravel in a silt matrix, they are interpreted to be scarp-derived colluvial wedges. The oldest of these, co-1, is best exposed on the SW wall of CC-3 (Fig. 9C). On the NE wall of the same trench (Fig. 9B), it is complexly deformed (disturbed) adjacent to a large fossil tree (this was removed by the digger). There the gravel wedge has apparently been entrained into the gravel-bearing mixed units os, ogs, and cs, and also dismembered into isolated clasts. In trench CC-2, co-1 is expressed as a stone line of footwall-derived terrace cobbles (well rounded) that extends southeastward above a basal peat layer (unit pt) for >2 m away from fault strand 3 (Fig. 9A). The younger and variably tilted colluvial wedge co-2 is recognized on all three of the logged walls.
On its NW side, the co-1 wedge is faulted against a distinctive sequence of sands and pebble gravels. Elements of this sequence are found on all walls of both northern trenches as combinations of the units pg (basal pebble gravel), ss (a thin marker layer of sand), and cpg (upper layer of pebble gravel). These loosely consolidated and iron-stained deposits are distinctively well rounded and well-sorted, and we interpret them to be fluvial. On all three logged walls, the fluvial deposits depositionally overlie the terrace gravels (unit tg) on the NW side of fault strand 1 (Figs. 9A, 9B, and 9C). Between fault strands 1 and 2, an intact (unfaulted) part of the fluvial sequence depositionally overlies a substrate of strongly sheared gravel mixed with clay pug (unit sg in Fig. 9A; unit stg in Figs. 9B and 9C). In trench CC-3 (Fig. 9B), the pebble-rich basal part to fluvial package (pg unit) thickens abruptly downward to occupy the space between fault strands 1 and 2. We interpret this to be an infilled faultfissure. In CC-3 (NE wall, Fig. 9B), the upper part of the fluvial infill (cpg unit) drapes south-eastward across fault strand 2 to lie on the peat of unit pt5. In trench CC-2 (Fig. 9A), the colluvial wedge (co-3) depositionally overlies the sandy ss layer (part of the fluvial infill package). The wedge co-3 overlaps fault strands 1, 2, and 3 and is displaced by fault strand 4.
Seven samples from trench CC-2 and five from CC-3 were 14C dated (Table 1). Of these, four are wood, and the rest are peat, organic clay, or charcoal. All from trench CC-2, the wood samples (CC-2-33, CC-2-35, CC-2-37, CC-2-38) yield ages that are older than stratigraphically underlying nonwood samples (CC-2-30 and CC-2-34) but are indistinguishable from one another (ca. 5.6–5.0 ka). As explained already, the wood samples are interpreted to have been recycled from a near-basal forest layer that was disrupted soon after inception of the present graben. All the nonwood samples in trenches CC-2 and CC-3 yield 14C ages that are in the correct stratigraphic order (see Appendix G).
Interpretation of Earthquakes Rupturing the Northern Bounding Fault of the Graben
On the basis of the combined data from CC-2 and CC-3, we recognize at least four earthquakes to have ruptured the northern part of the Cross Creek graben since ca. 5.2 ka. These are labeled CCN4 (oldest) to CCN1 (youngest) in Table 2, and their age constraints are plotted on Figure 10.
The oldest earthquake, CCN4, resulted in emplacement of the colluvial wedge co-1 at the site of both CC-2 and CC-3. Temporal constraints for this earthquake are provided by 14C samples CC-3-L (from unit pt, below the wedge, Fig. 9C) and from CC-3-E, CC-2-30, and CC-2-34 (from the same unit above the wedge). These ages bracket the earthquake to the period 5280–4640 cal. yr B.P. This interval overlaps with CCS5 on the opposite side of the graben, an event that is similarly recorded by a colluvial wedge (unit cwA in Fig. 8B). We therefore infer that CCN4 and CCS5 represent the same earthquake. Our preferred age for event CCN4, 5209–4842 cal. yr B.P., is based on the interpretation that the six similar-aged wood samples were derived from a forest that was toppled or damaged by this earthquake. For this preferred age, we use the age of sample CC-2-38, the youngest element of the death assemblage that we infer to have been earthquake-triggered.
We interpret the next-youngest earthquake to rupture the northern side of the graben (CCN3) to have caused emplacement of the colluvial wedge co-2 at the site of both CC-2 and CC-3. Age constraints for this wedge are provided by samples CC-3-F, CC-2-34, and CC-2-30 from the pt unit below the wedge (Figs. 9B and 9C) and by samples CC-3-2 and CC-2-31 from above the wedge (the pt unit in Fig. 9B; the osi-2 unit in Fig. 9A). These data yield an age range for the earthquake of 3080–1991 cal. yr B.P., an interval that overlaps with CCS3 on the opposite side of the graben, as expressed by colluvial wedges cw2 (in CC-1, Fig. 8A) and cwC (in CC-4, Fig. 8B).
The penultimate earthquake (CCN2) to rupture the northern boundary fault of the graben caused opening of an ~1-m-deep fault fissure. This cavity was infilled initially by the pebble gravel unit pg and later by the sand of the ss unit. These well-sorted fluvial units were deposited across the pug-lined faults bounding the fissure (fault strands 1 and 2 in Fig. 9B), with the youngest part of the infill sequence (unit cpg) downlapping onto the peat unit pt4 (Fig. 9B). Presumably a small stream draining across the scarp transported the clasts. We infer that CCN2 caused, moreover, some combination of the following: SE-ward tilting of the co-2 wedge, anticlinal bulging of the peat basin to the SE, and deposition of the colluvial wedge co-3 on the NE wall of trench CC-2 (Fig. 9A). A maximum age constraint for CCN2 of 2150–1940 cal. yr B.P. is provided by sample CC-3-2 in unit pt5, which stratigraphically underlies the cpg member of the fissure-infilling sequence (Fig. 9B).
The final rupture at the site, CCN1, caused renewed slip on fault strands 1 and 2 and initiation of fault strands 5, 6, and 7 (Figs. 9B and 9C). This faulting deformed the fissure-infilling sequence of stream sediments (units ss, pg, and cpg) that were deposited after earthquake CCN2. In addition, faulting on strand 4 deformed the colluvial wedge co-3 that we attribute to CCN2 (Fig. 9A). Inferred to be the 1855 earthquake, event CCN1 may also have resulted in deposition of the colluvial layer grs at the site of trench CC-3; however, it is uncertain whether this is a natural deposit.
Note that the described progressive deformation sequence (opening of fissure, filling of fissure, renewed faulting) requires charcoal sample CC-3-2 (2150–1940 cal. yr B.P., from pt5 in Fig. 9B) to predate two earthquakes. Conceivably, it might predate three earthquakes if the fissure-infilling significantly predated deposition of the colluvial wedge co-3 in trench CC-2. This is because this wedge stratigraphically overlies the fissure-infilling ss unit (Fig. 9A). We view such a three-event scenario as less likely than a two-event scenario, since no other trench provides evidence for three earthquakes since ca. 2 ka. In our preferred interpretation, the fissuring and emplacement of the colluvial wedge co-3 are viewed as twin manifestations of the penultimate earthquake, CCN2.
Late Holocene Rupturing History of the Southern Wairarapa Fault
By integrating the key stratigraphic and structural events observed in the eight paleoseismic trenches, and by dating and correlating these using the 40 new 14C samples, we interpret a composite surface-rupturing history that has included at least five earthquakes on the southern part of the Wairarapa fault since ca. 5.2 ka (Table 3; Fig. 10). This history and our intertrench correlations are summarized in Appendix H. Although any correlation involves interpretation, our chronology invokes the minimum number of earthquakes allowed by the data. While the dating precision for each event at each site differs, and the preferred age of specific earthquakes might vary according to interpretation, we believe that the overall number of surface-rupturing events (five since ca. 5.2 ka) is a robust outcome of the data. Not unexpectedly, preservation of the individual surface-rupturing events in this chronology is unequal between different sites and trenches because of differences in exposed stratigraphic age and degree of disruption or burial of near-surface layers as a result of human activity. Only the Cross Creek graben trenches sampled the oldest three events, whereas the Pigeon Bush and Riverslea trenches provided timing information relating to the last two events only (1855 and penultimate). As a result of such preservational differences, no single trench preserves evidence for all five events (though combinations of four are recorded in each of the four Cross Creek trenches). By correlating events between the various trenches and trench walls, and using the entire data set of 14C samples to bracket the composite event timing, we were able to significantly narrow the 95% confidence time intervals for each event (red bars in Fig. 10) relative to that which would have been derived by limiting our analysis to 14C data found in each individual trench or site (compare Table 2 to Table 3).
The Cross Creek pull-apart graben is the key locality; it provides an organic-rich stratigraphic record that was progressively deformed near the fault, especially on the northern side of the graben. The closely spaced pair of trenches, CC-1 and CC-4, across the southeastern margin of the Cross Creek pull-apart graben recorded all five of the earthquake events. Four of these events are corroborated in the two trenches on the northwestern margin of the graben (CC-2 and CC-3). This correspondence implies that both strands of the graben ruptured together during these four earthquakes, and it reinforces our confidence in the composite chronology. The second-oldest earthquake in our chronology (event CCS4, recorded in part by the fault-draping intrapeat unconformity) apparently ruptured only the southern margin of the graben, and it is the one exception to this mutual rupturing of bounding faults. Subjective elements of our composite interpretation include: (1) attributing local 14C age reversals (relative to stratigraphic ordering) associated with six similar-aged wood samples to reflect sedimentary recycling of a single forest death assemblage; and (2) accepting four 14C ages near the fault-draping unconformity as recording peat deposition (these occur in correct stratigraphic order to each other and to other surrounding samples), while interpreting two other samples as recording nondepositional events (these yield a reversed age sequence).
The apparent internal consistency in timing of these five rupturing events between the disparate trench sites suggests that these broke the entirety of the southern section of the Wairarapa fault (Fig. 1A). By contrast, only the youngest (1855) and oldest of the five southern events can be correlated to the earthquake chronology at Tea Creek trench, ~40 km to the north of Pigeon Bush (Van Dissen and Berryman, 1996) (Figs. 1A and 10). Although the Tea Creek record is based on only one trench, this apparently imperfect correlation suggests that some of the southern ruptures (not including 1855) may not corupture across the northern zone of splay fault junctions to break the entire length of the Wairarapa fault.
Comparison of Earthquake Chronology with that of Turakirae Head Beach Ridges
For a flight of uplifted strandlines to provide a complete record of paleoearthquakes, each earthquake must cause enough uplift to preserve a distinct coastal landform. Moreover, these land-forms must be distinct from others that might form by nontectonic relative sea-level changes or longer-term tectonic processes that are aseismic. Finally, any interseismic subsidence due to pre-earthquake strain accumulation or postseismic crustal relaxation must be small or predictable fractions of the overall strandline displacement (e.g., Berryman, 1987; Nelson and Manley, 1992; Wilson et al., 2007b). Turakirae Head is one of the world's best examples of a coseismically uplifted flight of Holocene beach ridges (Burbank and Anderson, 2001). Others include Mocha and Santa Maria Islands in Chile (Nelson and Manley, 1992; Bookhagen et al., 2006), the Boso Peninsula and other sites in Japan (Ota and Yamaguchi, 2004), Peninsula de Nicoya, Costa Rica (Marshall and Anderson, 1995), the Mahia Peninsula and Pakarae River mouth in New Zealand (Berryman, 1993; Ota and Yamaguchi, 2004; Wilson et al., 2007a), Cape Mendocino, California (Merritts, 1996), Taiwan (Yamaguchi and Ota, 2004), and the Gulf of Alaska (Plafker, 1969; Plafker et al., 1992; Plafker and Rubin, 1978). Although other studies have compared beach ridge uplifts to the timing of historic earthquakes as much as ~500 yr ago (e.g., Bookhagen et al., 2006; Ferranti et al., 2007), these studies typically suffer from the short time span of the historic data and from ambiguities regarding the location of the surface rupture accompanying those felt earthquakes. To our knowledge, ours is the first study attempting a one-to-one comparison between uplifted strandlines and paleoseismically documented fault ruptures over a time span of ~5 k.y. This comparison is possible because the Wairarapa fault (unlike subduction megathrusts) is exposed above sea level where it can be investigated by trenching at sites proximal to the uplifted beach ridges on its hanging wall.
McSaveney et al. (2006) identified and dated the uplift and stranding of four late Holocene beach ridges at Turakirae Head. Three of these are younger than ca. 5.2 ka. Each of these corresponds to one of our independently determined Wairarapa fault rupturing events (Fig. 10). They are the most recent earthquake (corresponds to uplift of beach ridge BR-2 in 1855), the third event (corresponds to uplift of beach ridge BR-3 at 2380–2060 cal. yr B.P.), and the fifth event (corresponds to uplift of BR-4, inferred by McSaveney et al.  to have taken place at 5420 –4110 cal. yr B.P.).
Two of our trench-determined fault-rupturing events cannot be matched to a beach ridge at Turakirae Head. These are the penultimate event and fourth event (Table 3; Fig. 10). At the crest of the Rimutaka anticline, the elevation difference measured by McSaveney et al. (2006) between the stranded 1855 beach ridge (BR-2) and the next-highest stranded ridge (BR-3) is ~9.1 m (Fig. 11A). We note that this difference is >2σ higher than the mean vertical spacing of the remaining three higher ridges at that site (4.8 ± 1.6 m, 1σ; McSaveney et al., 2006), implying that this large vertical interval may record the cumulative uplift of two earthquakes. Temporally, this beach ridge interval spans from 1855 to include both our penultimate and third events. Accordingly, we infer that a distinct beach ridge may not have been introduced (or preserved) into the landscape after the penultimate earthquake. The vertical interval between beach ridges BR-3 and BR-4 records not only the fifth event (in time) but also our “missing” event four. At 5.5 m, the height difference between these two ridges implies a mean tectonic throw of ~2.6 m for those two earthquakes (McSaveney et al., 2006). Perhaps, one of these two incremental uplifts was too small to preserve a distinct beach ridge at Turakirae Head.
Our proposed identification of the two “non–beach-ridge” earthquakes is independently supported by other studies. Using the diatom record cored from Lake Kohangapiripiri (Fig. 1), Cochran et al. (2007) inferred a sudden shallowing to have taken place across that estuarine lagoon at 3900–3300 cal yr B.P (this transition is labeled K3 in Fig. 10). The 14C-based time interval for this inferred uplift event overlaps with the trench-based timing of our fourth event (Fig. 10). Other diatom assemblage transitions reinforce the third and fifth earthquake events of our trench-based chronology. Because the record at the top of the Lake Kohangapiripiri core (<2 ka) is poorly preserved, probably in part eroded, and poorly dated, it cannot be used to corroborate or refute our proposed timing for the penultimate earthquake on the Wairarapa fault (Cochran et al., 2007). Farther west on Rongotai Isthmus (Fig. 1), Pillans and Huber (1995) identified and dated beach deposits that were stranded above sea level at 3410–2740 cal. yr B.P. (which correlate with our fourth event) and at 940–260 cal. yr B.P. (which correlate with our penultimate event). The authors infer that most of the shoreline uplifts on the isthmus were the result of Wairarapa fault paleoearthquakes; however, this inference seems uncertain given the proximity of the site to the Wellington fault.
Our comparison between the trench-based earthquake rupturing history of the Wairarapa fault and the sequence of raised beaches at Turakirae Head leads us to conclude that flights of uplifted gravel beach ridges may provide an incomplete record of paleoearthquakes on adjacent reverse-oblique faults (Fig. 11A). We note that a similar discrepancy between (less frequent) beach-ridge uplifts and (more frequent) earthquake events has been observed in Chile and Italy on the basis of historical records of earthquakes (e.g., Bookhagen et al., 2006; Cisternas et al., 2005; Lomnitz, 2004; Ferranti et al., 2007). The next section asks the question, “What processes might have led to a shortfall in the number of beach ridge uplifts relative to earthquake ruptures on the hanging wall of the Wairarapa fault?”
Nontectonic Causes for Under-Representation of Earthquakes by Beach Ridges
Consisting of coarse gravel, the raised beach ridges at Turakirae Head mantle a low-angle Holocene wave-cut platform incised into graywacke bedrock. This is a wave-battered coast, open to the south and the largest storm waves in Cook Strait. Since 1855, storm waves have built a berm crest on the modern beach that is variably 2–7 m above mean sea level (McSaveney et al., 2006), a relationship that underscores the sensitivity of berm height to local wave conditions. Due to Turakirae Head's preeminent position on the coast, and its remoteness from large rivers, its rocky coastline is only thinly covered with sediment. With little apparent input from longshore drift, a process which tends to sweep material southward away from the headland, the main source of the gravel for this sediment-starved headland has been in situ erosion of the bedrock platform (Wellman, 1967) and/or mass wasting of the adjacent hills (Hinton and McSaveney, 2007). Because the eustatic position of sea level is inferred to have remained stable throughout New Zealand since ca. 6.5 ka (Gibb, 1986), we infer that the only significant source of relative sea-level fall near Turakirae Head since ca. 5.2 ka has been tectonic uplift. In a recent review paper, Kennedy (2008) argues that Gibb's (1986) sea-level curve is supported by subsequent studies, and that there is no evidence for eustatic sea-level variations in New Zealand of more than ±1 m since ca. 6 ka. For this reason, we assume that sea-level variations have not been a key factor influencing beach-ridge formation at Turakirae Head since ca. 5.2 ka. For example, we discount the possibility that a small rise in sea level during the Holocene (by itself) could have caused erosion of a preexisting beach ridge.
Given the complexity of hydrodynamic variables involved in the formation and retreat of gravel beaches (Carter and Orford, 1993; Orford et al., 1995; Neal et al., 2003; Engels and Roberts, 2005), it is possible that variations in sediment supply or wave climate (especially as the result of large winter storms) may have contributed to formation or preservation of discrete beach ridges during some interseismic periods of the Wairarapa fault and not others.
The most plausible nontectonic scenarios for earthquake under-representation by the beach ridges probably involve some combination of the following two processes: (1) the crest of the active berm may, at times, retreat landward far enough to weld with, or overwash, the next-highest beach ridge; or (2) a beach ridge may have been so small or indistinct at the time of its (tectonic) stranding that it was later easily overwashed, or otherwise not preserved as a discrete landform on the gravelly wave-cut platform. Secular wave climate and sediment supply are the chief variables that control the size and height of a storm berm and determine whether it remains stable in one place (perhaps building up an especially high berm) or is over-washed to retreat landward (Carter and Orford, 1993; Orford et al., 1995). If these quantities varied between interseismic periods (i.e., at a time scale of ~1000–2000 yr), then preservation of a discrete beach ridge may have occurred after some earthquakes but not others. Given the extreme (nonlinear) impact of the lowest frequency–highest magnitude storm events on the extent to which gravel beach ridges can be driven landward, a period of increased storminess could cause an active storm berm to overtop (or weld with) a relict beach ridge that was originally several meters (perhaps even 5–8 m) higher than it; in this case, the key variable would be the return period of storms large enough to generate the runup capable of overwashing the higher ridge (Orford et al., 1995; Neal et al., 2002, 2003). Scenario 2 might occur because of either a reduced sediment supply (beach ridges cannot form without gravel) or because of too large a sediment supply. The latter situation might suppress formation of discrete storm berms or lead to their burial underneath younger deposits (Engels and Roberts, 2005). Clearly, the internal structure and sedimentological characteristics of the raised beach berms at Turakirae Head might provide a record of such depositional or erosional events; this would seem to be a fruitful avenue of future research that could build upon (and test) the results of our study.
Tectonic Causes for Under-Representation of Earthquakes by Beach Ridges
Alternatively (and perhaps more likely), the incomplete representation of earthquakes by beach ridges could reflect variable magnitudes of coseismic uplift at the coast during Wairarapa fault earthquakes. Figure 11B illustrates the complex structure of the Wairarapa fault zone at the southern end of the North Island near Turakirae Head, where there are multiple (alternate) fault strands in the near-surface. If the formation of a beach ridge on the hanging wall of the Wairarapa fault zone during the late Holocene was mostly controlled by the magnitude of the coseismic vertical displacement, then this would have reflected not only the direction and magnitude of slip on any rupture at depth, but also the geometry of the rupturing near-surface fault (especially its proximity to Turakirae Head and its dip). Given the observed structural complexity of the Wairarapa fault zone near the coast, one cannot assume that there was only one rupture scenario during the Holocene.
Some earthquakes may have caused a large enough coastal uplift at Turakirae Head to elevate the former storm beach beyond the reach of subsequent waves, thus preserving a discrete beach ridge, whereas others may have not. If earthquake ruptures near the coast typically reoccupy the same spatial fault plane (or near-surface splay) during each earthquake, the magnitude of the coseismic uplift may vary between earthquakes, and some may not uplift the coast enough to strand a distinct beach ridge. Dislocation modeling of vertical deformation during 1855 (Darby and Beanland, 1992; Beaven and Darby, 2005) and the extremely high slip/length ratio of that earthquake's coseismic rupture (Rodgers and Little, 2006) suggest that it may have ruptured not only the Wairarapa fault but also a contiguous segment of the subduction interface downdip of it. Perhaps some Wairarapa fault earthquakes involve deep co-rupturing of the subduction interface (as in 1855) to cause 3–9 m of coastal uplift locally near Turakirae Head, whereas others break only the upper plate, or nucleate as strike-slip ruptures in the upper plate, yielding less throw at the coast.
Alternatively, earthquake ruptures at the southern end of the Wairarapa fault may have a nearly constant slip and focal mechanism at depth, but they may cause variable coseismic uplift at the surface because of variable partitioning of slip between multiple fault splays in the upper crust. Dislocation models by Beavan and Darby (2005) require a local thrust structure to have been responsible for the very high magnitude (up to 6.4 m) but short wavelength of uplift at Turakirae Head in 1855. Coseismic slip on some oblique-slip splays (especially the Muka Muka fault; Fig. 2B) may cause significant uplift of Turakirae Head on their hanging wall, whereas slip on other splays may not. Consistent with this view, Schermer and Little (2006) and Little et al. (2008) document a temporally and spatially complex pattern of fault activation-deactivation and surface folding during the past 80 k.y. at the southern end of the Wairarapa fault zone and infer that a blind thrust is currently active near the western edge of Lake Onoke, ~10 km to the east of the Muka Muka fault (Figs. 2 and 11B). These relationships suggest a variable linkage between slip on the deeper Wairarapa fault and its splays in the near-surface. If the 1855 earthquake involved uplift of Turakirae Head along the Muka Muka fault, previous earthquakes may have ruptured to the surface along other splays to the east or west, causing little or no coastal uplift there.
Until now, we have ignored the possible contribution of aseismic regional uplift caused by sediment underplating or subduction of anomalously thick oceanic lithosphere at the Hikurangi subduction zone (e.g., Litchfield et al., 2007). Any such uplift is assumed to have accumulated slowly and steadily enough during the past ~5 k.y. to have had no significant effect on the punctuated uplift and stranding of beach ridges at Turakirae Head as a result of large Wairarapa fault earthquakes. A point less easily dismissed is the possible effect of interseismic subsidence due to elastic strain accumulation above the underlying, locked subduction interface. Permanent GPS observations along the SE coast of the North Island (and forward dislocation modeling of GPS data) suggest that it could be on the order of 1–2 mm/yr near Turakirae Head (L. Wallace, March 2008, personal commun.). If this rate of subsidence persisted for an entire Wairarapa fault interseismic period of 1–2 k.y., then up to ~2 m of a previous coseismic uplift signal might be removed. However, because the subduction interface's earthquake recurrence interval is probably much shorter (~300–625 yr; Wallace et al., 2004) than the Wairarapa fault's earthquake recurrence interval (1–2 k.y.), less subsidence is likely to accumulate before a Wairarapa fault earthquake (this amount would depend on the phase shift in the seismic cycles between these two faults). Either way, interseismic strain accumulation is predicted to reduce the height of any previously stranded beach ridge and thus increase the possibility that it would be subsequently overwashed during an especially stormy period (thus omitting one beach ridge from the preserved sequence; e.g., Marshall and Anderson, 1995). This would be most likely if the Wairarapa fault earthquake happened to occur near the end of the subduction zone fault's seismic cycle, and if that cycle was especially long. McSaveney et al. (2006) argued, however, on the basis of a comparison between survey data and historic observations of coseismic uplift in 1855 that interseismic subsidence on Palliser Bay has been negligible during the past the 150 yr.
Our preferred interpretation is that coseismic uplift has been variable at the coast near the Wairarapa fault, and that this tectonic variability, perhaps combined with contributing variations in secular wave climate or sediment supply (or interseismic subsidence), has resulted in an incomplete representation of earthquakes by beach ridges near Turakirae Head.
Rates of Slip and Earthquake Recurrence on the Wairarapa Fault
The dating undertaken as a part of this study establishes the timing of deposition and abandonment of the gravels comprising the youngest post–Last Glacial aggradation (fill) terrace in the southern Wairarapa Valley, the so-called Waiohine terrace. Our 14C age of samples CC-1-1a-I and Auger-3 bracket formation of the terrace tread to postdate 12,400 ± 300 and to predate 5440 ± 140 cal. yr B.P. (Table 1). Since the ca. 12.4 ka peat sample was deposited during the final phases of terrace aggradation, we infer that abandonment of that terrace took place at ca. 12 ka, making the Waiohine fluvial terrace age-equivalent to either the Ohakea 2 or Ohakea 3 terrace in the regional correlation scheme applied to North Island, New Zealand (e.g., Litchfield and Berryman, 2005). We note that our 14C-based age data are accordant with several other ages obtained previously from samples collected elsewhere from above or below the Waiohine terrace tread (Tompkins, 1987; Marden and Neal, 1990). They are also considerably more precise than eight late Last Glacial OSL ages of silts that overlie the gravel of Waiohine terrace along the central part of the Wairarapa fault farther north (Wang and Grapes, 2007). Those samples gave ages of 16–10 ka, with four of eight ages being younger than 13 ka. Because these cover bed silts may have accumulated after incision of the terrace by the Waiohine River, their OSL ages may underestimate the timing of terrace abandonment.
Applying a mean OSL age of 10–11 ka to displaced channels on the Waiohine terrace at Waiohine River, to which they attributed a dextral slip of 125 ± 5 m, Wang and Grapes (2007) proposed a late Quaternary dextral-slip rate for the Wairarapa fault of 11.5 ± 0.5 mm/yr. We note, however, that these authors did not present any surveyed map of these offsets, and that Lensen and Vella (1971) surveyed and mapped the same terrace in detail without recognizing these channels (their largest offset was for the terrace riser incised below the Waiohine surface, a younger landform for which they measured a dextral slip of ~99 m). Our 14C-based dating suggests that the Waiohine terrace was abandoned near Cross Creek at no later than 12 ka. Assuming that this surface was subsequently displaced laterally by at least 99 m (at least near Waiohine River), our data imply a minimum late Quaternary slip rate on the Wairarapa fault of ~8.3 mm/yr. On a shorter time scale, if we divide estimates of mean single-event strike slip on the southern Wairarapa fault during the last two earthquakes (13–16 m; from Rodgers and Little, 2006) by our revised mean earthquake recurrence interval for that fault (1230 ± 190 yr), we get a dextral-slip rate of 9–15 mm/yr. This result is within error of the above slip-rate estimate based on our 14C-derived, 10–12 ka abandonment age of the Waiohine surface (~11 ± 3 mm/yr).
The Wairarapa fault is remarkable because a series of uplifted beach ridges on its hanging wall are preserved on the nearby coast, and the ages of these Holocene ridges can be directly compared with the ~5 k.y. chronology of surface-rupturing earthquakes derived from paleoseismological methods. We dated five surface-rupturing earthquakes on the southern part of Wairarapa fault since ca. 5.2 ka (an inter-event recurrence of 1230 ± 190 yr). Along the margins of the Cross Creek pull-apart graben, these earthquakes caused laterally and temporally variable amounts of faulting, tilting, and fissuring of scarp-proximal strata, emplacement of scarp-derived colluvial wedges, and the toppling and destruction of trees. Our late Holocene paleoearthquake record recognizes two more events than are recorded by beach ridges that were stranded on the uplifted hanging wall of the fault near Turakirae Head during the same time interval (McSaveney et al., 2006).
Our work indicates that coseismically uplifted gravel beach ridges on an open, rocky coast may provide an incomplete record of paleoearthquakes on adjacent reverse-oblique faults. In the case of the southernmost part of the Wairarapa fault zone, variable increments of coseismic uplift at Turakirae Head may have resulted from the partitioning of slip between several different fault splays in the near-surface. During event four, coseismic throw at the coast is inferred to have been so small (<3 m) that a discrete new beach ridge did not form. Variations in wave climate, sediment supply, and/or interseismic subsidence probably exerted further controls on the number of beach ridges preserved in the uplifted sequence, in particular, governing the morphology (e.g., height, elevation) of the original storm berm, and whether or not it would retreat far enough landward to overwhelm the next-highest beach ridge. A combination of such processes may have caused event four to be under-represented by a discrete, long-lived beach ridge, despite significant coseismic uplift during that earthquake.
Our 14C data indicate that a widespread Last Glacial Maximum aggradational terrace found in this part of North Island, New Zealand (Waiohine terrace), was abandoned soon after 12.4 ka. This result, combined with previous estimates of dextral slip relative to this terrace, our revised recurrence interval for the Wairarapa fault, and estimates of mean single-event slip on the fault, suggests that the southern part of the Wairarapa fault has a late Quaternary dextral-slip rate of ~11 ± 3 mm/yr.
We thank Kate Wilson and Vasso Mouslopoulou for their global positioning system (GPS) surveying efforts; Julia Bull, Dave Murphy, Susanne Grigull, Vasso Mouslopoulou, and Kate Wilson for assistance with trench logging; and H. Saywell, H. Brandon, P. Smith, and D. Cleal for permission to excavate on their land. This study was funded by the “It's Our Fault” Project, with additional support provided by the New Zealand Foundation for Research Science and Technology, contract CO5X0402 (Geo-Hazards and Society, GNS Science). Two anonymous reviewers and Jon Pelletier are thanked for their constructive comments. Helpful feedback on an early version of the manuscript was provided by David Kennedy and Mauri McSaveney.