Both the Gilbert Ridge and Tokelau seamount trails show age systematics that are nonlinear and inconsistent with current absolute plate motion models [Duncan and Clague, 1985; Lonsdale, 1988; Wessel and Kroenke, 1997; Koppers et al., 2001]. It is obvious from our age determinations that plate motion alone, with respect to a fixed reference frame of narrow plumes in the deep mantle, cannot explain the apparent velocities as derived from these age systematics and the asynchronous timing of the HEB-type bends. Either these hot spots have moved significantly with respect to each other or other nonplume processes have contributed to the formation of these seamount trails. In this section, we will first discuss what the fixed hot spot hypothesis would predict for the Gilbert and Tokelau seamount trails and which assumptions are violated by our new age determinations. In the remainder of this section, we will discuss three alternative models that might explain the observed age systematics by the possibility of (1) motion between short-lived hot spots, (2) plate extension due to local changes in plate stress, or (3) multiple intersecting or overlapping hot spot trails.
5.1. Fixed Hot Spot Hypothesis
Our geochronological data show that the 47 Ma HEB does not occur in the Gilbert Ridge and Tokelau seamount trails, and conversely, that no consistent morphological evidence exists for 57 and 67 Ma HEB-type bends in the Hawaii-Emperor and Louisville seamount trails. In addition, the Gilbert Ridge shows a highly nonlinear age progression at best, evidence for multiple phases of intraplate volcanism around 115, 74–77 and 67 Ma, and an apparent plate velocity of ∼131 mm/yr that is very different from the ∼87 mm/yr velocity calculated for the neighboring Tokelau seamount trail. These observations create a serious dilemma for the fixed hot spot hypothesis, because they violate two assumptions central to this hypothesis: (1) that changes in absolute plate motion are recorded simultaneously in colinear seamount trails, resulting in identical morphological changes, and (2) that hot spots produce corresponding linear age progressions in colinear seamount trails, reflecting the (constant) angular velocity of a rotating rigid tectonic plate over a given period of geological time. Even though the fixed hot spot hypothesis may allow for the presence of younger volcanism, owing to the long-lived evolution of single volcanoes up to several million of years, it is difficult to explain the presence of older ages in colinear seamount trails, as we observed in our study of the Gilbert Ridge and Tokelau Seamounts.
The utility of the hot spot model for the origin of the Gilbert Ridge and Tokelau seamounts can also be explored by “backtracking” individual seamounts on the basis of their age and particular plate motion models (Figures 6a–6d). It is important to state here that these reconstructions assume the presence of a “fixed” reference frame of hot spots in the Pacific and thus reflect past plate motions only. Also we assume that these hot spots have mantle plumes that are narrow and do not exceed 300 km in diameter. A reasonable fit to Macdonald hot spot can be achieved for the Tokelau Seamounts (Figures 6a and 6d) using the absolute plate motion model of Koppers et al.  and Wessel et al.  to which we will refer as KMMS01 and WHK06 hereafter. However, the reconstructed seamounts do not cluster and some plot up to 350 km away from the Macdonald hot spot position. The fit deteriorates when we apply the Raymond et al.  and the Wessel and Kroenke  models, labeled R20 and WK97, with distances from the reconstructed seamount positions to Macdonald hot spot ranging from 300 to 600 km (Figures 6b and 6c).
Figure 6. Backtrack analyses of the Southern Wake, Ratak, Gilbert Ridge, and Tokelau seamount trails and comparison to the HIMU-type seamounts of the Cook-Austral and Macdonald hot spots in the South Pacific. We show the results from three different stage pole models by (a) Koppers et al. , (b) Wessel and Kroenke , (c) Raymond et al. , and (d) Wessel et al.  that all assume an age of 43 Ma for the Hawaii-Emperor Bend (HEB) except in the latter model. However, a systematic ∼300 km offset to the northeast will result if we accept an age for the HEB equal to 47 Ma [Sharp and Clague, 2006], which diminishes the fit to the Cook-Austral and Macdonald hot spots significantly. Note that the Southern Wake seamount trail also contains seamounts (yellow circles) that are not HIMU-type and that the active hot spots of the South Pacific region (filled black circles) and HIMU-type seamounts of the Mangaia-Rurutu line (green circles) are shown for reference.
Download figure to PowerPoint
The fit of the Gilbert Ridge seamounts to the Rurutu and Mangaia hot spots is even poorer, despite a similar geochemical signature with high 206Pb/204Pb ratios [Konter et al., 2004] that resembles the unique HIMU mantle end-member [Zindler and Hart, 1986]. The reconstructions of Limalok Guyot and seamounts of the Cretaceous Ratak and South Wake seamount trails are very interesting in this context. Limalok guyot was drilled during ODP Leg 144 and resulted in two high-quality 40Ar/39Ar ages of 69.2 Ma [Koppers et al., 2000]. It is part of the Gilbert Ridge seamount trail where it continues into the Ratak Chain to the north, but as can been seen from our reconstructions, its backtracked location is highly uncorrelated and plots 300 to 450 km to the north of the other Gilbert Ridge seamounts (Figure 6). One may argue that its non-HIMU signature [Koppers et al., 2003b] demonstrates that this guyot was formed by another hot spot, yet its high-volume volcanic edifice makes Limalok morphologically identical to the other seamounts in the Gilbert Ridge. The majority of seamounts in the Southern Wake and Ratak seamount trails form another group of (older) seamounts that carry the same HIMU signature [Koppers et al., 2003b] and could potentially relate to the Cook-Austral Islands and the Gilbert Ridge. However, our backtrack reconstructions show that these seamounts plot even farther to the north of the Gilbert Ridge and Cook-Austral Islands (Figure 6). Although the latter observation might be the result of our poor knowledge of Pacific plate motions prior to 80 Ma, it is evident that different plate motion models substantially influence the results of the seamount backtracking, as does the actual timing of the HEB used in these models (Figure 6). If we insert an age of 47 Ma for the geometrically defined HEB in the plate motion models [Sharp and Clague, 2006] to replace the outdated 43 Ma age [Dalrymple and Clague, 1976], the modeled locations fall up to 150 km farther to the northeast, making the potential fit to the Macdonald and Cook-Austral hot spots even worse (see the comparison for model R20 in Figure 6c). This misfit increases, if we assume an upper limit of 50 Ma for the age of the HEB, marking the possible initiation of this bend that took about 8 Myr to complete [Sharp and Clague, 2006].
The latest model describing absolute Pacific plate motion during the last 70 Myr by Wessel et al.  does not improve these fits significantly (Figure 6d). In the WHK06 model the authors retain the premise of fixed hot spots by applying the Polygon Finite Rotation Method (PFRM) of Harada and Hamano  to obtain a finite rotation model. An advantage of this geometric technique is that an age need not to be assigned to the HEB, or any other change in plate motion, and a more continuous (or smooth) rotation model is derived. This approach, however, is limited by the fact that the present-day locations of the hot spots must be prescribed (which are surprisingly hard to determine) and that only six seamount trails are used in their analysis (Hawaii, Louisville, Caroline, Foundation, Pitcairn, Cobb). For progressively older times the number of trails used becomes even smaller. Prior to 18 Ma, they use only 3 seamount trails, and prior to 31 Ma, they have to resort to 2 seamount trails. In other words, most of their 70 Myr finite rotation model is dominated by (or biased toward) two seamount trails, namely the Hawaiian and Louisville seamount trails. This makes the WHK06 model philosophically unlike the KMMS01 model, which is based on 21 seamount trails between 0–43 Ma and 8 trails between 43–80 Ma [Koppers et al., 2001]. It is therefore not a surprise that Wessel et al.  can obtain precise backtrack reconstructions for the Hawaiian and Louisville seamount trails using their WHK06 model, which is based largely on the same seamount trails, whereas the KMMS01 model does less well in similar reconstructions. This is partially because the KMMS01 model does not include a plate motion change around 5 Ma, but it is also because the KMMS01 model incurs a larger variance in its Euler pole predictions on account of the large number of seamount trails included. Which model is better (or more accurate) hinges on the geological question whether all seamount trails should be explained by a single textbook-type hot spot, or whether we have to recognize more than one type of hot spot, as suggested by Courtillot et al. . If hot spots indeed can be divided into different groups, we should not lump together seamount trails while deriving Pacific plate motion models, because they most probably were formed by different processes. On the other hand, if we resort only to plate motion models based on Hawaii and Louisville, we limit our understanding of other seamount trails in the Pacific, while backtracking exercises for seamounts in the Hawaiian and Louisville trails will be of a lesser value because of circular reasoning.
All results considered, our reconstructions show that the geometries and age distributions of the Gilbert Ridge and Tokelau Seamounts cannot easily be explained by assuming a fixed sub-Pacific hot spot reference frame of narrow mantle plumes, whether we use the WHK06 model based on a few preselected seamount trails, or the KMMS01 model that includes a significantly larger number of trails. Even though the proximity of the reconstructed seamount locations with the Cook-Austral hotline and Macdonald hot spot and their similarity in geochemistry [Koppers et al., 2003b; Konter et al., 2004] seem to advocate a close connection and provinciality, it would require a significant reconfiguration of their mantle plumes over time. These long-lived mantle sources either were moving within a 600–1000 km wide region or they were formed as part of a rather wide upwelling of mantle material on a more regional scale.
5.2. Hot Spot Motion
The orientation of a volcanic lineament is typically considered to be the result of plate motion, but if hot spots are not stationary, it may be the result of hot spot motion as well [Norton, 1995; Duncan and Keller, 2004]. If one takes the results of recent global plate circuit reconstructions at face value [Cande et al., 1995; Norton, 1995; Raymond et al., 2000; Steinberger et al., 2004], it is evident that the bend in the Hawaii-Emperor trail (still) cannot be reconciled (entirely) with the fixed Indian, Atlantic and Pacific basin hot spots. This may be caused by hidden flaws in the plate circuit models, for instance by problems in the reconstruction of Antarctic plate motions [Steinberger et al., 2004] or by our poor knowledge of the absolute plate motions in the Indian and Atlantic oceans. Alternatively, we can explain the HEB by the slowdown of a southward moving Hawaiian hot spot with respect to a rather constant Pacific plate motion to the WNW. If the plate velocity is faster than the hot spot motion, the resulting seamount trail will display a trend that is closer to the plate motion vector (Figure 7a). If the hot spot motion is faster than the plate motion, the seamount trail aligns at an angle that is closer to the direction of the hot spot motion (Figure 7b). However, if we vary the hot spot motion vector from an initially rapid southward motion to a very slow motion, as is observed in the available paleolatitude data for the Hawaii-Emperor seamount trail [Tarduno and Cottrell, 1997; Tarduno et al., 2003; Sager et al., 2005], it becomes obvious that the trail would systematically bend into the direction of plate motion (Figure 7c). The faster the slowdown, the sharper the bend will become. Assuming that the motion of the Hawaiian hot spot may also have had a longitudinal component, the overall hot spot motion vector might even have been as high as 60 mm/yr compared to its 40 mm/yr latitudinal component. This would increase the curved nature of the HEB even more. A similar behavior has been reproduced in the numerical mantle flow models of Steinberger that show a large-scale mantle flow toward the South Pacific superplume, slowing down around 47 Ma [Steinberger and O'Connell, 1998; Steinberger, 2000; Steinberger et al., 2004]. The Gilbert Ridge and Tokelau seamount trails both were formed just to the north of the probable center of this superplume, suggesting that a similar southward motion for these two hot spots is possible.
Figure 7. The effects of combined plate and hot spot motions on the azimuth and morphology of seamount trails. Assuming a constant NNW plate motion for the Pacific plate, we show here the effect of a decelerating hot spot, as observed for the Emperor and Hawaiian stages for the Hawaiian hot spot. The faster the slowdown, the sharper the bend becomes. For simplicity we assume that the Hawaiian hot spot motion is due south, but in reality this could vary between ESE and WSW.
Download figure to PowerPoint
A moving hot spot model thus appears to be an attractive working model for the origin of the HEB-type bends in the Gilbert Ridge and Tokelau seamount trails. Whereas bends formed by changing plate motion alone must be simultaneous for all hot spots on a particular rigid plate, changing hot spot motion may produce bends asynchronously due to local mantle convection, carrying the hot spots in different directions and at different rates. The hot spots also may be short-lived [e.g., Koppers et al., 2003b] with mantle plumes that have roots which become more shallow over their short life span. With typical plume rise speeds of 6 cm/yr in the upper mantle, this would mean that a plume can rise from midmantle levels in less than 10 Myr [Steinberger and O'Connell, 2000] or even faster if the conduit is much hotter than the ambient mantle [Steinberger, 2000]. As a consequence, hot spots may start their existence in the midmantle (or even deeper), where they move with the subducting plate return flow [Steinberger and O'Connell, 1998], and they may end their existence more or less as a fixed plume in the upper mantle, following a short 20 Myr life cycle. Plumes may disappear because of thermal entrainment effects, break up following over-tilting or if insufficient material is supplied from below.
However, some problems arise with this model for our study area, because moving hot spots would still require a monotonic age progression in these seamount trails, which is not apparent for the Gilbert Ridge. If one assumes that the poorly constrained age progression of ∼131 mm/yr for the Gilbert Ridge is a function of both hot spot and plate motion, the required hot spot motion would be drastically higher than the average 10 mm/yr hot spot velocity as observed in numerical models [Steinberger, 2000]. Faster plume motions can be achieved if we take into account the tilting of plume conduits in the upper mantle, which can cause remarkably fast plume motions up to 40 mm/yr [Tarduno et al., 2003]. Furthermore, the moving hot spot models cannot explain the second deflection in the middle of the Tokelau Seamounts. This would require the occurrence of two small-scale mantle plumes, which reside close to each other in the South Pacific mantle and which show slowdowns at very different times, 10 Myr apart.
To verify whether a moving hot spot model would improve our backtrack reconstructions for the Gilbert Ridge and the Tokelau Seamounts, we applied the KDS04 plate motion model that has been corrected for the motions of the Hawaiian hot spot [Koppers et al., 2004]. In the KDS04 model, the observed 13° southern motion of the Hawaiian hot spot between 80 and 47 Ma [Tarduno and Cottrell, 1997; Sager, 2002; Tarduno et al., 2003; Doubrovine and Tarduno, 2004] has been taken into account, resulting in a simple four stage rotation model for Pacific plate motion sensu stricto during the last 83 Myr. Performing backtrack reconstructions with this kind of model therefore should yield the “ancient” locations of the hot spots or the geographic location where the seamount formed. As can been seen from our reconstructions using the KDS04 model in Figure 8, these “ancient” locations are very different from the present-day locations of their apparently long-lived and moving hot spots. For example, the Tokelau seamounts would have formed at a location where the Rurutu hot spot resides now, which lies distinctly to the NW of the current Macdonald hot spot. A similar systematic offset is apparent for the Ratak and Gilbert Ridge seamounts. In addition, the moving hot spot model did not erase any of the discrepancies between the reconstructed seamounts. It actually increased the scatter in the modeling results, suggesting that the observed age systematics for the Gilbert Ridge and Tokelau Seamounts are not a function of hot spot and plate motion. Alternatively, their hot spot motions may be different from the southern motion of the Hawaiian hot spot.
Figure 8. Backtrack analyses taking into account the effects of combined plate and hot spot motions for a selected number of seamounts. The same exercise is done as in Figure 6, but based on the rotation model of Koppers et al.  that has been corrected for the 13° motion of the Hawaiian hot spot between 80 and 47 Ma.
Download figure to PowerPoint
5.3. Lithospheric Extension
Nonplume hypotheses have been proposed as alternates to the fixed hot spot hypothesis [Foulger and Natland, 2003; Natland and Winterer, 2005]. Although these alternate hypotheses all have the potential of explaining one or more aspects of intraplate volcanism, only few positive lines of evidence have been presented that would conclusively prove one of these possibilities. So far, we can substantiate that the basic assumptions for the fixed hot spot hypothesis do not hold for certain seamount trails [Cande et al., 1995; Koppers et al., 2001, 2003b; Tarduno et al., 2003] including the Gilbert Ridge and Tokelau Seamounts [Koppers and Staudigel, 2005], but we can only speculate about alternate models. Nonetheless, in this section we will review the possibility of extensional volcanism that may explain the formation of the Gilbert Ridge and Tokelau seamount trails, and the asynchronous HEB-type bends therein.
Most volcanic lineaments (e.g., Puka Puka ridge) that previously have been proposed to have formed due to plate extension have an azimuth parallel to the Pacific plate motion vector since 47 Ma, probably reflecting a near-orthogonal extension caused by the overall slab pull on the Pacific plate [Winterer and Sandwell, 1987; Sandwell et al., 1995]. More recently, these volcanic lineaments have been considered to have formed as cracks in the Pacific plate as a result of a horizontal, thermal contraction of this plate while it cools [Sandwell and Fialko, 2004]. Regardless of the underlying process, these volcanic lineaments seem to form at regularly spaced gravity troughs that may be associated with the local thinning or cracking of the oceanic lithosphere. This shows that stress is not evenly distributed over the interior of a rigid tectonic plate, in particular, for young oceanic crust. The Puka Puka ridge also shows an easterly age progression that is faster than the expected plate motion [Sandwell et al., 1995] suggesting a fissure-like opening of this crack over the course of 30 Myr [Natland and Winterer, 2005]. Alternatively, volcanic lineaments like the Puka Puka ridge may be caused by a large scale migration of relatively hot mantle material from the Superswell toward the EPR [Conder et al., 2002; Toomey et al., 2002; Hillier and Watts, 2004]. Neither of these models would require the existence of a plume.
However, the morphology of these volcanic ridges in gravity troughs is quite different from seamount trails as the Hawaii-Emperor, Louisville, Gilbert Ridge and Tokelau seamount trails, which are more voluminous, less linear and contain more discrete, conical volcanic edifices, many of which have developed into oceanic islands, atolls or large guyots. Courtillot et al.  considered these “gravity-line” volcanic ridges a third type of hot spot volcanism, different from the primary, long-lived Hawaiian and Louisville hot spots, and the secondary, short-lived hot spots located in the South Pacific and West Pacific Seamount Province. In addition, volcanic ridges typically form on young oceanic crust close to the mid-oceanic spreading centers, which is in strong contrast to seamount trails like the Gilbert Ridge and Tokelau seamount trails that form truly intraplate, on oceanic crust older than 50 Myr. The question now is whether the more voluminous seamount trails can be explained by similar extensional processes? How do these processes explain the age progressions that to a first order do exist in most seamount trails, like Hawaii and Louisville? Why do most seamount trails have consistent geochemical signatures indicative of single mantle sources, even if they display complex age systematics?
Extension might be easier if the lithosphere has been preconditioned either by tectonic processes (fracture zones) or by volcanic processes (ancient spreading ridges and intraplate volcanism). It takes less energy to reactivate a preexisting structure than to create an entirely new plate boundary [Gurnis et al., 2000] or to establish a conduit system for magma to penetrate pristine oceanic lithosphere. These so-called “crack spots” [Wessel and Kroenke, 2000] are sites of extensional volcanism forming at preexisting zones of weakness that are reactivated by (local) plate stresses. This does not necessarily involve major (or even minor) plate reorganizations. It is quite conceivable that plate extensional stress changes on a short-term and local basis, while the plate as a whole displays a relatively constant plate motion. These “jerk-like” changes in the stress distribution may occur when the slab pull force balance is changed by the subduction of a fracture zone. Density contrasts due to a different age, crustal thickness and temperature across the leading edge of the subducting plate may cause minor but significant differences in the force balance. Similar jerks may be caused by the subduction of obstacles that temporarily clog up the subduction zone. Good examples are the subduction of a seamount, seamount trail or oceanic plateau. The above changes may not be big enough to change plate motion, but they may tug and pull on the plate, which eventually will yield by extension associated with these disturbances in the subduction zone regime.
Extensional volcanism may also be helped by the presence of topography on the base of the Pacific lithosphere, creating places of preferred ponding of magma or channeling of magma. In studies of the Musicians seamounts [Kopp et al., 2003], the Foundation seamounts [O'Connor et al., 2001] and the Galapagos archipelago [Braun and Sohn, 2003; Chen and Lin, 2004] it has been hypothesized that a fraction of their mantle plumes have been diverted toward a region where it can rise more easily to the surface of the Earth. In these situations, magma may have been channeled toward a mid-ocean ridge where plate extension has a profound effect on mantle upwelling, but it is also conceivable that plumes may be channeled toward places of lesser lithospheric thickness [McNutt et al., 1997; Sleep, 1997, 2002a, 2002b] or “thin spots”. These scenarios, however, predict seamounts that may have ages older than expected, because they could precede the leading seamount in a hot spot trail.
Although the Gilbert Ridge and Tokelau seamount trails were formed on 50–65 Myr old oceanic crust, and a substantial distance away from active spreading centers, the oceanic lithosphere in these regions contains complexities that could have played a role in their seamount formation. In particular, we note (1) that the Gilbert Ridge itself is located on a seemingly minor fault, marking the edge of a region with anomalously thick oceanic basement following the intrusion of basaltic sills during the formation of Ontong Java Plateau, (2) that the Nova Canton Trough runs perpendicular to the midsection of the Tokelau seamount trail, and (3) that an extension of Manihiki Plateau (named Robbie Ridge) intersects the southern end of the Tokelau seamount trail (Figure 1). Theoretically these complexities could provide weak zones in the Pacific lithosphere that could be reactivated by globally induced stress changes on the Pacific plate. For example, the rather similar 69 Ma age measured at six seamounts covering 450 km of seamount trail between Limalok Guyot and Tofe Tolu seamount (Figure 1) may be explained by a leaking transform fault, located at the structural edge of the Darwin Rise or underneath a precursor hot spot trail that formed the bulk of the Gilbert Ridge. Precursory hot spot volcanism (as observed in the older Ava and Sakau seamounts) would initially build an age-progressive seamount trail, explaining the overall younging of the Gilbert Ridge from north to south, while preconditioning the Pacific plate for later failure under extension. Rejuvenated volcanism could be caused by globally induced plate stresses that reactivated this older seamount trail at a later stage, generating a late volcanic veneer on top of the Gilbert Ridge. Geochemical integrity would be retained, despite its belated eruption, because under extensional circumstances melting would occur at the base of the Pacific plate, where HIMU-type mantle sources might have been present in the asthenosphere since the Late Jurassic [Koppers et al., 2003b] or were imbedded in the lithosphere [Staudigel et al., 1991]. This scenario would explain the disturbance observed in the age pattern and the prevalent HIMU-type signature of the Gilbert Ridge.
To explain the formation of the HEB-type bends, we propose here that the Pacific Plate experienced two short-term and local extensional phases in its currently southwestern region, one at about 67 Ma and one at 57 Ma, reactivating the inactive spreading center that formed the Nova Canton Trough [Larson et al., 2002] and reactivating a similar kind of seafloor fabric to the west of the Phoenix fracture zone. In particular, the 74–78 Ma old seamounts in the Gilbert Ridge (Figure 1a) are indicative of older volcanic ridges or seamount trails that may have preconditioned the Pacific lithosphere. From Figure 9 it is clear that the older Ava, Sakau and Seka seamounts are part of an east-west fabric that is evident in the morphology of the seamounts themselves. The younger Kautu, Palutu and Beru seamounts appear to be part of these fabrics, allowing for the possibility that reactivation of these east-west segments of seamounts might explain the 67 Ma HEB-type bend in the Gilbert Ridge. The 67 and 57 Ma bends also occur at times that have been associated with minor changes in Pacific plate motion or with tectonic events occurring around the rim of the Pacific [Epp, 1984; Duncan and Clague, 1985; Yan and Kroenke, 1993; Wessel and Kroenke, 1997]. For example, Chron 27 (∼61 Ma) marks the onset of relative motion between east and west Antarctica, and between Australia and Antarctica [Müller et al., 2000]. Although plate “jerks” by themselves did not generate observable changes in the spreading rate and direction of relative plate motions, these minor changes might have contributed to the changes in the internal stress distribution of the largest tectonic plate on Earth.
Figure 9. Structural analysis for the Gilbert Ridge. Each volcanic edifice has been marked with a cross, and morphological trends have been highlighted with dashed lines, where the green lines indicate the most prevalent directions and the red lines show alternative interpretations. From this analysis it is clear that most structural trends run parallel to the Phoenix Fracture zone, except where the suspected HEB-type bend appears in the seamount trail, providing clear morphological evidence for the bend.
Download figure to PowerPoint