Corresponding author: Peter D. Clift, Department of Geology and Geophysics, E235 Howe-Russell, Louisiana State University, Baton Rouge, LA 70809, USA. (firstname.lastname@example.org)
 The Nankai accretionary complex is the most recent addition to the accretionary complexes of southwest Japan and has preserved a record of sediment flux to the trench during its construction. In this study, we use U-Pb zircon and fission track analysis of both zircons and apatites from sediments taken from the forearc and trench of the Nankai Trough, as well as rivers from southwest Japan to examine the exhumation history of the margin since the Middle Miocene. Modern rivers show a flux dominated by erosion of the Mesozoic-Eocene Shimanto and Sanbagawa accretionary complexes. Only the Fuji River, draining the collision zone between the Izu and Honshu arcs, is unique in showing much faster exhumation. Sediment from the Izu-Honshu collision is not found 350–500 km along the margin offshore Kyushu indicating limited along-strike sediment transport. Sediment deposited since 2 Ma on the midtrench slope offshore the Muroto Peninsula of Shikoku (ODP Site 1176) and on the lower slope trenchward of the Kumano Basin (IODP Sites C0006E and C00007E) shares the dominant source in the Shimanto and Sanbagawa complexes seen in the modern rivers. Prior to 5 Ma, additional sediment was being sourced from further north in more slowly exhumed terrains, ~350 km from the trench axis. Around 9.4 Ma, U-Pb zircon ages of ~1800 Ma indicate enhanced erosion from the North China Craton, exposed in northern Honshu. In the middle Miocene, at ~15.4 Ma, the sediment was being derived from a much wider area including the Yangtze Craton (U-Pb ages ~800 Ma). We suggest that this enhanced catchment may have reflected the influence of the Yangtze River in supplying into the Shikoku Basin prior to rifting of the Okinawa Trough at 10 Ma and migration of the Palau-Kyushu Ridge to form a barrier to transport. The restriction of Nankai Trough provenance to Mesozoic source partly reflects continued uplift of the Shimanto and Sanbagawa complexes since the Middle Miocene.
 The role of subduction zones in the production and maintenance of continental crust has long been a subject of debate among geoscientists. Accretionary plate margins represent areas in which there is a net growth of the continental crust through periods >106 years partly as a result of magmatic accretion but also as a result of offscraping in the downgoing plate. The sedimentary rocks which are imbricated at a trench are generally assumed to be derived from the active continental margin against which they are offscraped and consequently reflect the nature of uplift and erosion within the active margin itself. These potentially represent an archive of the tectonic evolution of the continental margin throughout the time of accretion and may allow us to understand why a margin is accretionary as opposed to erosional. This assumption of relatively local derivation may not always be valid because if a large drainage system is providing sediment into a trench, it is possible that sediment from a much wider area may be fed into the accretionary complex. As a result, sediment that is used to build an accretionary complex is sensitive to the development of drainage systems and thus also to the tectonics of the continental interior. For example, sedimentary rocks within an accretionary complex in South Central Alaska has been shown to change in composition as the arc collided with continental North America, reflecting the expansion of what were initially limited coastal drainage systems [Clift et al., 2012]. In contrast, the opening of marginal seas might be an effective mechanism for isolating a trench system from large continental drainages by providing accommodation space for the sediment between the river mouth and the trench.
 One area where the process of growth through accretion is taking place is the collision zone between the oceanic Izu-Bonin Arc and the continental Japanese (Honshu) Arc. This collision has resulted in rapid uplift of the Mount Fuji area and the nearby Akaishi/Kanto Mountains [Kimura et al., 2006] (Figure 1). The timing of collision has been proposed to be Pleistocene-Late Pliocene; however, this is mostly based on the age of exhumation of the present point of collision and does not necessarily account for a propagating collision point moving toward the northeast. In this study, we examine the sedimentary evidence for an earlier rapid exhumation related to the collision of Honshu and the Izu-Bonin Arc, preserved in the sedimentary record of the forearc in southwest Japan, and assess whether the collision has a regional effect on accretionary trench tectonics. We further examine the impact that rifting of the Sea of Japan (East Sea) and of the Okinawa Trough (Figure 1a) have had on the supply of sediment into the trench during the Neogene.
 We examine the clastic flux to the Nankai Trough in southwestern Japan since around 14 Ma in order to determine when material diagnostic of the collision first reached the trench, as well as to see whether sediment derived from within continental Asia has been cut off from the trench as the rift basins along the Asian margin have opened over that same time. Because the rocks of the Izu-Bonin Arc are Eocene and younger [Stern and Bloomer, 1992; Taylor, 1992], they significantly postdate most of the source terrains in southwestern Japan. Consequently, evidence of young grains (<50 Ma), especially those associated with rapid exhumation in the forearc or trench would be considered the fingerprint of erosion from the arc-arc collision zone. Correlating changes in provenance with changes in exhumation rate would imply that tectonic processes dominate the erosion of Honshu and in turn the flux of sediment to the trench of the Nankai Trough. Similarly, erosion of continental crust blocks within Eastern China and the Korean Peninsula would provide much older mineral grains to the Nankai Trough compared to those that could be derived through local erosion in Japan, because potential sources like the Yangtze Craton or the North China Block have characteristic old ages of ~800 Ma and 1800 Ma, respectively, for their crustal formation [Chen and Jahn, 1998; Chen et al., 2001].
 One of the most complete records of the accretion and exhumation of the Izu-Honshu collision zone may be found in the Nankai Trough accretionary prism, which was sampled by Ocean Drilling Program (ODP) Leg 190 [Shipboard Scientific Party, 2001a] and Integrated Ocean Drilling Program (IODP) Expedition 316 [Kimura et al., 2008].The Nankai Trough is a suitable region for looking at the record of sediment flux into the trench because scientific coring has now recovered samples from the forearc basin, trench slope, and the subducting plate (Figure 2). A detailed biostratigraphic framework exists, allowing ages of deposition to be accurately determined. Sediment accreted into the prism could be transported to the trough from the collision zone in the Akaishi Mountains via the Suruga Canyon (Figure 1). In addition, sediment may be delivered directly downslope from mainland Honshu and the island of Shikoku where the sources are the more slowly exhuming Tertiary-Mesozoic Shimanto and Sanbagawa Belt accretionary terrains. Sediment derived from the accretionary complexes mostly accumulates in the Kumano forearc basin and its lateral equivalents before overspilling into the trench (Figure 2).
2 Geological Setting
 The Nankai Trough represents the partly sediment-filled trench of the subducting plate boundary between the island arc of Japan, comprising both the Eurasian and North American plates, and the Philippine Sea plate, which is subducting toward the northwest in this area, slightly oblique to the strike of the trench [Karig and Angevine, 1986; Hall et al., 1995]. The oceanic crust now passing under the forearc in the Nankai Trough is part of the Shikoku Basin, which was formed by the rifting of the proto-Izu-Bonin arc [Taylor, 1992]. Rifting of the Izu Bonin arc started in the Oligocene and culminated in Shikoku Basin seafloor spreading that lasted until ~15 Ma [Chamot-Rooke et al., 1987]. Three phases of seafloor spreading separated the remnant arc, the Kyushu-Palau Ridge, from the active Izu-Bonin arc creating the Shikoku Basin [Okino et al., 1994].
 The exposure of accretionary complexes onshore in Honshu, Kyushu, and Shikoku indicates that north-dipping subduction has been proceeding along this plate boundary for long periods of geological time, at least since the Jurassic [Taira et al., 1988]. These are separated by the Median Tectonic Line (Figure 1b) from older Asian crystalline basement intruded by Mesozoic and Cenozoic plutonic and volcanic arc rocks. Onland studies of the Shimanto and Sanbagawa accretionary complexes represented some of the first detailed investigations of these zones of crustal accretion and can be considered classic examples of the phenomena as well as analogs for the processes now operating in the Nankai Trough [Taira et al., 1988; Banno and Sakai, 1989]. Continued subduction accretion under the forearc of modern southwest Japan has resulted in rock uplift and erosion, so that these accretionary complexes are now generating and delivering sediment to the forearc basin, where it may be re-accreted into newer parts of the forearc prism [Hasebe et al., 1993]. Southwestern Japan is unusual for an active plate margin in so far as it has no apparent active volcanic arc, except on the island of Kyushu at the southwestern extreme end of the chain. The first major volcano close to the study site is Mount Fuji, located close to the collision point between the Izu-Bonin Arc and the island of Honshu (Figure 1).
 The collision between the Izu-Bonin Arc and Honshu has resulted in the accretion of oceanic island arc crust to the edge of Honshu. The present collision and crustal accretion point is marked by the Izu Peninsula [Ogawa et al., 1985]. North of that in the Tanzawa Mountains, midcrustal tonalities are exposed on the edge of a major extensional province known as the “Fossa Magna” [Takahashi, 1989]. Accretion seems to have started at ~8 Ma, with the main phase of collision in the Tanzawa area dated at ~6–5 Ma [Niitsuma, 1989]. Exhumation rates around the Fossa Magna are higher than in many other parts of Japan and have been estimated using fission track and (U-Th)/He methods to be 0.5 to 1.5 mm/yr. [Yamada and Tagami, 2008]. This latter study further concluded that the exhumation rate of the Tanzawa tonalites had not changed in a coherent or significant fashion since the initial accretion at ~8 Ma, despite the changing tectonic regime and evolution in climatic conditions.
 The Shimanto accretionary complex comprises lightly metamorphosed Cretaceous to Paleogene clastic sedimentary rocks with imbricated sections of oceanic volcanic rocks and serpentine, together with deep-water ribbon radiolarian cherts [Ogawa, 1985]. K-Ar dating and zircon fission track dating suggest that following accretion, the sedimentary rocks of the Shimanto complex were buried to temperatures of around 300°C and were then cooled to the surface sometime after 40 Ma [DiTullio and Hada, 1993; Hara and Kimura, 2008]. Locally, higher metamorphic grades were achieved as a result of at least two collisions between the trench and seafloor spreading ridges in the late Eocene/early Oligocene and the middle Miocene.
 Offscraping of sediment from the downgoing Philippine Sea plate continues in the present day so that the Nankai Trough is one of the best developed and imaged subduction accretion complexes worldwide [Le Pichon et al., 1987]. As a general rule offscraping occurs at the toe of the accretionary prism, but across much of southwest Japan, a major out-of-sequence thrust has developed in the middle of the accretionary complex, branching off the main subduction mega-thrust fault [Moore et al., 2007; Bangs et al., 2009]. As a result, a ridge has been uplifted in the hanging wall of this “mega-splay” feature, back-tilting sediments away from the trench and ponding sediment in a midslope basin called the Kumano Basin (Figure 2a). Similar features are also seen further toward the southwest, e.g., the Tosa Terrace, where several hundred meters of sediment are found deposited on top of strongly deformed accretionary complex turbidite facies rocks [Hilde et al., 1969]. These basins, which ride piggyback on top of the out-of-sequence thrusts, are important in buffering sediment flux from mainland Japan into the trench and may represent an important control to the composition of sediment now being accreted within the Nankai Trough. In this study, we use a combination of thermochronology methods to identify the sources of sediment to different parts of the forearc and trench system, and by comparing between different drill sites, we constrain sediment transport routes and how those may have changed since 14 Ma.
3 Previous Work
 Initial studies following ODP Leg 190 suggest that a major change in sediment provenance to the Nankai Trough occurred at 1.95–1.0 Ma [Fergusson, 2005], with a substantial increase in the volume of volcanic lithic grains in the trench sands at that time. This corresponds well with some previous dates given for the Izu-Honshu collision, which may have started at 2 Ma, with exhumation accelerating after that time [Niitsuma, 1989]. The change in provenance was attributed to the collision of the Izu-Bonin Arc with Japan, the rapid uplift of the collision zone, and resultant change in sedimentation from transverse, Shimanto-derived sediment to trench-parallel axial flow from the collision zone [Fergusson, 2005]. However, this work was based solely on low-resolution petrography and is inconsistent with the age of accretion derived from the Tanzawa Mountains, indicating initial collision around 8 Ma [Soh et al., 1998]. While the earlier studies provide some important constraints on source, we note that the conclusions are based on basic petrography, which may not be diagnostic of the different sources available in southwest Japan since volcanic rocks are also found within the accretionary complexes.
4 Sampling Strategy
 In this study, we took samples from four separate areas of the forearc with material from five separate drill sites (Figure 2a). ODP Site 1177 lies in the trench at the southwestern edge of the study region and sampled a section of turbidite sandstones and muddy pelagic sediments (Figure 3). By definition, these sediments were all deposited within the Shikoku Basin and have yet to be accreted into the forearc prism. ODP Sites 1176 and 1178 both lie on the drilling transect offshore the Muroto Peninsula where there is no well-developed forearc basin separating the trench slope from the island of Shikoku. ODP Site 1178 penetrated an imbricated section, although our sampling was restricted to the coherent section lying above the shallowest, major, identified thrust fault. ODP Site 1176 cored the lower trench slope, which is composed of an overlying upper slope hemipelagic section and a lower, accretionary prism section comprising sediment deposited within a trench fan complex. Our sampling was restricted to the coarser, clastic parts of this lower section, which were generally interpreted as the products of turbidite sedimentation [Shipboard Scientific Party, 2001b]. We further sampled two more recent boreholes IODP Sites C0006E and C0007E, which lie close together on the lower trench slope around 180 km to the northeast of the Muroto transect. These sites lie closest (~350 km) to the arc-arc collision point in the Magna Fossa. As well as providing a greater geographical coverage, these two sites combined provided a high density of samples over the last 1 Ma. They both penetrated sedimentary rocks of the accretionary complex and lie immediately downslope from the Kumano Basin (Figure 2a). The depositional ages of the sediments from the drill cores are all taken from the published age models in the relevant sites reports that are based on nannofossil biostratigraphy, supplemented by paleomagnetic information. The sedimentary logs and some of the age control are shown graphically in Figure 3, which shows some of the more important age control points but is not intended to be a complete summary of everything that is known from the scientific party reports. The depositional ages are calculated by assuming linear sedimentation rates between points on the core, which are well dated through the biostratigraphy or magnetic data.
 In addition to the samples from the drill cores, we also collected material from near to the river mouths of five major rivers that are supplying the modern continental margin of Japan (Figure 1 and Table 1). Although some radiometric work has been done on the basement rocks of the Japanese islands, the ability of the rivers to sample across wide areas provides a good opportunity to characterize the overall nature of the sediment now being eroded. Crucially, rivers do not bias their sampling in the way that a geologist might because of access issues, although erosion maybe focused in small parts of the basin rather than being evenly distributed. Potentially, modern river samples give us an image of the type of sediment that has been reaching the forearc region in the recent geologic past. Although we do not know if rivers in the past had precisely the same composition, we do know that at least since 14 Ma the overall basement geology of the Japanese islands has not changed substantially, and so we would expect the same sources to be contributing to the margin in the past. We chose one river, the Fuji River, as this drains the region around Mount Fuji and whose drainage basin includes part of the collision zone (Figure 1). The sediments in the Fuji River should provide us with an example of the types of sediment that we would expect to be delivered to the trench from the collision zone through the Suruga Canyon.
Table 1. Locations of the River Samples Taken to Characterize the Clastic Flux from Modern Japan into the Trench Area
 Here we employ a multiproxy thermochronology method with sediments from the Nankai Trough, as well as five rivers draining the Japanese mainland. We choose to use a combination of fission track analysis of zircon (ZFT) and apatite (AFT) grains together with U-Pb dating of zircon crystals because combinations of this variety have proven effective at resolving contrasting sources and defining quantitative sediment budgets in complicated continental margin/basin settings in earlier studies [Carter and Moss, 1999; Clift et al., 2004]. Provenance can be constrained by U-Pb dating of detrital zircon grains, while the addition of fission track data allows these conclusions to be cross-checked because different source terrains often have different cooling histories [Carter, 1999]. Furthermore, the fission track analyses allow exhumation rates of these sources to be determined by comparison with depositional ages. Shorter lag times between cooling and deposition implies faster exhumation.
 We exploit the differences between the cooling histories of the Izu-Honshu collision zone (involving rapid exhumation of relatively young crust) and the older, more slowly exhuming Shimanto-Sanbagawa complex to quantify the evolving relative flux of material from each of these regions. Crucially significant differences in crustal ages and exhumation rates are known to exist within the potential source areas. There are abundant Mesozoic ZFT cooling ages in the Shimanto complex [Koshimizu, 1990; Mizukami et al., 1991; Hasebe et al., 1993; Shinjoe and Tagami, 1994; Hasebe et al., 2000; Hasebe and Tagami, 2001; Hasebe and Watanabe, 2004; Wallis et al., 2004; Hara et al., 2007]. It is noteworthy that this belt also contains Miocene ZFT ages that partly overlap in age with the magmatic zircons from the Izu-Bonin arc. A combination of ZFT and AFT allows further resolution of the cooling history because the two systems image cooler and warmer parts of the exhumation process. Together they can be used to define a cooling history through the shallower levels of the crust.
 The low-temperature apatite fission track method, which records cooling through ~125–60°C over timescales of 1–10 Ma [Green et al., 1989], is particularly sensitive to exhumation driven by erosion and has been widely used in exhumation studies worldwide. The zircon fission track method works on similar principles, although the annealing of tracks is not as well defined. The zircon-based method tracks cooling of rocks over a range of approximately 300–200°C, although there is some disagreement about exactly where the top and bottom of the partial annealing zone lie, which makes modeling of its cooling a little more complicated [Tagami et al., 1996]. Nonetheless, zircon fission track analyses have been successfully applied to accretionary complex sandstones in the past [Tagami and Dumitru, 1996], and similar results were anticipated for the Nankai Trough.
 Fission track analyses were performed at University College, London, UK. Polished grain mounts of apatite and zircon were etched with 5 N HNO3 at 20°C for 20 s to reveal the spontaneous fission tracks. Subsequently, the uranium content of each crystal was determined by irradiation, which induced fission in a proportion of the 235U. The induced tracks were registered in mica external detectors. The samples for this study were irradiated in the thermal facility of the Hifar Reactor at Lucas Heights, Australia. The neutron flux was monitored by including Corning glass dosimeter CN-5, with a known uranium content of 11 ppm, at either end of the sample stack. After irradiation, sample and dosimeter mica detectors were etched in 40% HF at 20°C for 45 min. Only crystals with sections parallel to the c-crystallographic axis were counted, as these crystals have the lowest bulk etch rate. To avoid biased results through preferred selection of crystals, the sample was systematically scanned, and each crystal encountered with the correct orientation was analyzed, irrespective of track density. The results of the fission track analysis are presented in Table 2.
Table 2. Results of Apatite and Zircon Fission Track Analysis
Sample Core Barrel
No. of Crystals
Central Age (Ma)
Track densities (×106 tr cm–2) and counts for the dosimeter (d), spontaneous tracks (s) and induced tracks (i) are denoted by ρ and N respectively.
Analyses were by external detector method using 0.5 for the 4π/2π geometry correction factor.
Ages were calculated using dosimeter glass CN-5; (apatite) ζCN5 = 339 ± 5 calibrated by multiple analyses of IUGS apatite and zircon age standards [Hurford, 1990].
Pχ2 is probability for obtaining χ2 value for v degrees of freedom, where ν = number of crystals – 1.
Central age is a modal age, weighted for different precisions of individual crystals [see Galbraith and Laslett, 1993] and RE % is the percent age dispersion of grain ages.
316 C0006E-5H-2 W
48.2 ± 4.0
316 C0006E-17X-4 W
21.4 ± 5.7
316 C0007B-1H-9 W
42.6 ± 3.0
316 C0007B-1H-9 W
70.8 ± 3.9
50.0 ± 5.9
79.1 ± 11.3
113.8 ± 20.1
37.9 ± 6.4
89.0 ± 8.5
53.8 ± 9.4
74.5 ± 11.9
88.5 ± 10.2
46.3 ± 6.6
107.7 ± 11.2
129.2 ± 17.0
35.2 ± 12.8
25.4 ± 4.5
86.2 ± 19.7
108.7 ± 11.7
4.2 ± 1.8
114.2 ± 11.3
42.2 ± 2.6
65.0 ± 1.8
55.1 ± 5.7
74.2 ± 2.9
10.7 ± 1.8
12.3 ± 0.6
80.9 ± 4.3
 In contrast to the fission track methods, the U-Pb dating of zircon is designed to determine the age of initial crystallization at high temperatures. Zircon is a U and Pb-rich heavy mineral, which is relatively abundant in many rock types, making it a useful mineral for dating crystallization and cooling below ~750°C [Hodges, 2003]. Because of this high closure temperature, the U-Pb age is rarely reset by metamorphic events, allowing the mineral to be used as a robust provenance indicator, while recognizing that reworking via sedimentary deposits can be an issue when matching detrital grain and bedrock ages [Carter and Bristow, 2001; Campbell et al., 2005]. In the case of southwest Japan, the primary repository from which recycling might occur are the Sanbagawa and Shimanto complexes.
 Detrital zircon U-Pb ages were measured using the London Thermochronology Research Group facilities at University College London based on a New Wave Nd:YAG 213 nm laser ablation system coupled to an Agilent 7500a quadrupole inductively coupled plasma-mass spectrometry. Around 100 grains is considered generally sufficient for characterizing sand eroded from a geologically complicated drainage basin [Vermeesch, 2004], although because of the limited size of the drill-core samples this ideal was not always achieved. The combined sample from the lower trench slope (C0006E-7H-5 W and C0006E-16X-1 W) is especially deficient in providing only 48 dates, while the Fuji river is slightly low in having 84 ages. We note that this number of analyses is rarely possible when using fission track data because of the time-intensive nature of the analyses, and to a lesser extent the abundance of apatite crystals in the sedimentary core sample available makes it hard to achieve such large data sets. We consider the results from the U-Pb age to be more statistically robust than those from the fission track, although that does not mean that nothing can be said concerning provenance and erosion rates based on these latter data.
 Real-time U-Pb data were processed using GLITTER 4.4 data reduction software. Repeated measurements of external zircon standard Plesovice (TIMS reference age 337.13 ± 0.37 Myr ago) [Sláma et al., 2008] and NIST 612 silicate glass [Pearce et al., 1997] were used to correct for instrumental mass bias and depth-dependent inter-element fractionation of Pb, Th, and U. For this study, 206Pb/238U ages are used for those grain younger than 1000 Ma, and for zircon grains older than 1000 Ma we used the 207Pb/206Pb ages to determine the crystallization age. Because some grains are discordant, we chose to only plot those grains younger than 1000 Ma when the discordance was less than 50%. For grains older than 1000 Ma, we chose a discordance threshold of 20%. The data are provided in Table SI-1 in the auxiliary material.
 The spectrum of the grain cooling ages from each of the considered samples is displayed using the Kernel Density Estimation (KDE) method of Vermeesch , which plots the detrital ages as a set of Gaussian distributions but does not explicitly take into account the analytical uncertainties. Vermeesch argues that this is a more statistically robust method for looking at detrital ages when the number of grains and analytical precision are both high. This method allows the age ranges and abundances of the different age populations to be graphically assessed and different samples compared.
6.1 Apatite Fission Track
 Figure 4 shows the range of AFT ages since 150 Ma. The three rivers from which data were obtainable show significant differences from one another. The Fuji River stands out as having a particularly young peak (~10.7 ± 3.6 Ma) and effectively no grains older than 40 Ma, while the Tenryu River has a clear peak at around 50 Ma, as well as a small one younger than 10 Ma. The Kiso River shows a wider spectrum of ages, and especially a significant number of grains older than 80 Ma, which are not seen in the other rivers. The most abundant peak lies at ~68 Ma. Short-term changes are seen at IODP Sites C0006 and C0007. Sample C0006E-17X-4 W was deposited at ~1.08 Ma based on the combined magnetic and biostratigraphy [Expedition 316 Scientists, 2009a] and shows a young peak, ~8 Ma (Figure 4f). The two younger samples at this location at nearby Site C0007E lower slope site were deposited at 0.78 and <0.42 Ma and show peaks of around 47 Ma (Figures 4e and 4f). These peaks differ significantly from those of a similar age seen at ODP Site 1176. Samples 1176A-30X-CC, 10–21 cm and 1176A-37X-1, 15–25 cm are dated at 1.49 and 1.94 Ma, respectively, and show a slightly older peak of 67 Ma compared to the 47 Ma peak seen at IODP Sites C0006 and C0007 (Figure 4d). However, at ODP Site, we see significant variations, with one sample (1176A-37X-1, 15–25 cm) dated at 1.86 Ma showing a population centered at ~38 Ma, while a slightly younger sample (1176A-33X-1, 22–32 cm) deposited at 1.81 Ma shows an older maximum of ~80 Ma (auxiliary material Figures SI-1G and SI-1H). It should be noted that sample sizes from ODP 1176 are relatively small and that some of the variation may reflect the apatite-poor character of the sediment. Only one sample at ODP Site 1178 on the upper slope yielded significant apatites, and this provided a spectrum that was characterized by very young grains (central age of 4.2 Ma; Figure SI-1D). ODP Site 1177 on the subducting plate provided data from older sediments. The oldest sample from ODP Site 1177 (1177A-46R-CC, 0–10 cm), deposited at 15.4 Ma, shows a unique spectrum peaking around 124 Ma (Figure 4h). However, two younger samples dated at 9.4 and 7.8 Ma both showed a preference for much younger grains around 20 Ma (Figures 4j and 4k), while the youngest sample from that borehole shows two peaks, the largest at 40–60 Ma and another small peak at 120–140 Ma (Figure 4i).
6.2 Zircon Fission Track
 The age spectra for measured ZFT ages are shown in Figure 5. As was the case with the apatite data, the Fuji River stands out as being anomalous in having much younger cooling ages than any of the other river basins considered. The central age for the Fuji sample is 12.3 Ma, while the Kiso, Tenryu, and Yodo Rivers show a common peak at around 60–80 Ma. None of the latter three rivers have many grains younger than 40 Ma. The Nagara River, which shares a mouth with the Kiso River, also shares the large 60–80 Ma population but shows a much more significant spread to older values of 100–200 Ma that is not seen in the other systems.
 Only one sample from IODP Site C0007 yielded sufficient zircon to provide an image of the age population from the lower trench slope in the northeastern part of the study area (Figure SI-2L). This sample, dated at <0.42 Ma, showed a clear clustering around 71 ± 7.4 Ma, slightly older than the Kiso and Tenryu Rivers and extremely close to the central age in the Yodo River. This peak is noted in many of the other marine core samples where it is typically the dominant age population. In the southeast on the downgoing plate at ODP Site 1177, a sample dated at 6.46 Ma (1177A-17R-1), but with relatively few total analyses, shows a quite different spectrum in having a large proportion of grains dating >100 Ma (Figure SI-2K). A better defined sample deposited at 5.1 Ma at ODP Site 1178 (Figures 5f and SI-2G) also shows a significant number of older grains, with a significant peak around 109 Ma (Figure SI-2G). Several of the samples deposited before 1 Ma show this tail of older material, which is not present in the modern river systems or more recent marine samples. On the midtrench slope at ODP Site 1176, a well-defined sample deposited at 1.86 Ma shows this spread to older values very clearly (Figure 5g), although the same sample also shows a small number of grains that date to younger than 40 Ma and one very young grain suggestive of influx from a source not seen in the other forearc marine samples.
6.3 U-Pb Dating of Zircons
 All samples from the modern rivers and from the marine drill cores show a wide spectrum of U-Pb ages (Figure 6). A minor number of grains in many of the samples extend to ages in excess of 3 Ga, although in all circumstances this is a small proportion of the total population. A common feature of all samples is the strong peak in the last 250 Ma, with a common minor population between 1.7 and 2.2 Ga. The relative size of this secondary population is seen to vary and is higher in the older sediments particularly those dated at 9.39 and 15.4 Ma at ODP Site 1177 on the subducting plate (Figures 6k, SI-3K, and SI-3L). In those samples, we see not only the abundant, young peak but also a significant population centered around 1.9 Ga, which is much more abundant than in any other samples analyzed. Furthermore, in the oldest sample (Figures 6k and SI-3L), we see a significant population around 800 Ma, which is not recognized beyond the occasional single grain in any other sample. The two oldest samples are also noteworthy in having larger than typical populations in the 200–400 Ma range compared to the <200 Ma dominant population. The only other sample that has a zircon population like this is from IODP Site C0006E-16X-1 W (Figures 6j and SI-3O). This sample shows the strong 200–400 Ma peak, as well as the significant peak at 1.9 Ga, despite its young sedimentation age of 1.08 Ma, although whether this is an accurate representation of the provenance at that time is debatable given the small number of grains that could be analyzed (N = 12). Nonetheless, we know that it has a similar composition as the younger sample C0006E-7H-5, which yielded 36 ages (Figure SI-3N).
 The age population of zircons can be examined in greater detail by looking at only the past 250 Ma (Figure 7), since this is the age range over which most of the crystallization ages spread. On this scale, differences between the river systems become more apparent. The Kiso River shows a very well-defined peak at 60–80 Ma, which is also seen in the Tenryu and Yodo Rivers. However, in these latter two systems, there is a spread to older ages as well, with a significant tail off toward 110 Ma. The Tenryu River in particular shows a bimodal age distribution (~70 and 90 Ma), which is similar to that identified from the Shimanto and Sanbagawa complexes (Figure 7a). The older peak is reduced and less well defined in the Yodo River, although this system does have more grains between 180 and 250 Ma. As with the fission track results, it is the Fuji River which is most anomalous. In this case, the most abundant peak has a central age of ~14 Ma, although there are significant populations in the 60–110 Ma range as seen in the other rivers but also strong influx dated at ~200 Ma that is less commonly observed (Figure 7e). The oldest forearc sediments dated at 9.39 and 15.4 Ma from ODP Site 1177 clearly show their anomalously high proportion of grains in the 150–250 Ma range compared with the younger sediments at all sites (Figures 7k, SI-4K, and SI-4L). The 1.08 Ma sediment at C0006E (C0006E-16X-1 W) also has a spread of ages away from the 60–80 Ma maximum, but the low number of crystals found in this sample makes that pattern hard to be certain of (Figures 7i and SI-4O). There is tendency to see more grains dating 150–250 Ma in the 1.86 and 1.94 Ma samples from ODP Site 1176 (Figures 7h, SI-4I, and SI-4J) compared to the younger samples at that site or at C0006E. Sediments younger than ~1.5 Ma seem to bear close resemblance to the age spectrum displayed by the Tenryu and Yodo Rivers.
 The U-Pb zircon dating provides the most comprehensive database to constrain the evolving provenance of sediment in different parts of the SW Japanese forearc. What is clear is that the oldest sediments deposited at ODP Site 1177 have a quite different character than those deposited in more recent times, after 9.4 Ma at other parts in the margin. Based on the plate reconstruction of Mahony et al. , the drill site would have been ~600 and 750 km further to the southeast at the time of sedimentation of these earlier sands at 9.4 and 15.4 Ma, respectively (Figure 8). Nonetheless, we presume that the Japanese islands or adjacent continental blocks must still have been the most important sources to that site because there are no other suitable alternatives in the region that would be capable of supplying Mesozoic zircons into the center of the Shikoku Basin at that time. Both the Izu-Bonin Arc and Palau-Kyushu Ridge, which date from the Eocene, would not be capable of producing such old materials. In contrast, analysis of the Shimanto complex indicates that there are zircons of this age in that body that must have been originally sourced from Mesozoic volcanic and plutonic rocks now exposed north of the Median Tectonic Line and representing the arc volcanic front of a continental active margin through the Mesozoic. However, in the Shimanto complex, the relative abundance of these 150–200 Ma grains is relatively low compared to the 60–110 Ma population, while these two populations are of approximately similar size in the 9.4 and 15.4 Ma ODP Site 1177 sediments (Figures 7k, SI-4K, and SI-4L). This suggests that the Shimanto cannot be the only source to those sites. Instead, we infer erosion from arc igneous rocks lying to the north of the Shimanto complex. A provenance focused north of the Median Tectonic Line is also consistent with the abundance of 1.9 Ga grains also seen in these sediments and which must be linked to erosion of the continental basement on which the Mesozoic arc was built. This basement was originally at least partly a fragment of the North China Craton, which is known to contain abundant zircons of ~1.9 Ga [Darby and Gehrels, 2006; Wan et al., 2011] but which are not known in the Yangtze Craton. We therefore infer that there has been a change in the patterns of erosion within Japan away from erosion of older arc and North China Craton basement rocks to the north during the early Late Miocene to stronger erosion of the accretionary Shimanto and Sanbagawa complexes in the south after ~9 Ma. Such a development could be linked to the progressive uplift of these accretionary terrains driven by progressive underplating along the trench.
 The most unique population observed in this study was found in the sample dated at 15.4 Ma from ODP Site 1177 in which the zircons range 600–1000 Ma. These ages are not found in the North China Craton but are associated with the Jinningian phase within the Yangtze Craton [Chen and Jahn, 1998; Chen et al., 2001]. The Yangtze Craton is generally considered to extend from central China across the Yellow Sea and into the southern half of the Korean Peninsula [Kusky et al., 2007]. The Yangtze Craton lies south of the North China Craton, but most recent compilations are unclear as to whether there are outcrops of the North China Craton in Japan [Zhang et al., 2003]. The composition of the 15.4 Ma sand at ODP Site 1177 seems to require erosion from a much wider area than any other sediments found in these cores. Whether this diversity was supplied by a single drainage system is unknown, but if that was the case it would have to encompass a significant area, potentially spanning parts of the modern Korean Peninsula and/or eastern China and not just modern Japan. The modern rivers draining Japan south into the Shikoku Basin today indicate that source terrains containing Yangtze Jinningian rocks are not present within those drainage systems, although there may be fragments of the North China Craton in the basement north of the Sanbagawa complex, as inferred from the small proportion of 1.7–2.0 Ga zircons observed. This observation indicates that the Mesozoic arc generated by subduction under modern-day Japan and Korea would have been intruded into this North China Craton basement, but the location of the Yangtze Craton basement source that fed the deep-water Shikoku Basin at 15.4 Ma is unclear.
 One possibility is that the Yangtze River itself may have been a major contributor of sediment into the Shikoku Basin during the Middle Miocene. Figure 9 shows KDE plots of the zircon U-Pb ages from ODP Site 1177 together with an average composition of the Yangtze Delta spanning the last 3.2 Ma [Jia et al., 2010]. It is noteworthy that both spectra contain several of the same population peaks. The primary differences are that there is a large peak around 100 Ma in the Nankai Trough site that is not present in the Yangtze and that the 800 Ma Jinningian peak is much stronger in the Yangtze River than it is in the Nankai Trough. These differences may partly reflect short-term temporal variations in the Yangtze River but also the additional flux from mainland Japan that would have mixed with the Yangtze outflow to form a composite, mixed sediment in the deep Shikoku Basin. We note that the flux of sediment from the Yangtze Delta into the Shikoku Basin would have been easier to achieve during the Middle Miocene than it is in the present day because the age of rifting of the northern Okinawa Trough dates from ~10 Ma [Letouzey and Kimura, 1985; Clift et al., 2003]. The Okinawa Trough is presently a major potential sediment barrier between the Yangtze Delta and the deep-water basin. Furthermore, the plate reconstruction of Mahony et al.  suggests that the Palau-Kyushu Ridge would have passed the possible location of the Yangtze delta at ~10 Ma (Figure 8b). This reconstruction is very similar to many that have been produced over the last 20 years or so which are generally similar scenarios because they are all dependent on the general clockwise rotation of the Philippine Sea plate during the Neogene [Hall et al., 1995]. For example, Honza et al.  show the Izu-Bonin Arc migrating toward the northeast along the southern edge of Japan but with a possible connection to the East China Sea being cut after 20 Ma. Sdrolias et al.  also show this general northeastward migration and with a clear passageway for the Yangtze River to feed material into the Shikoku Basin open well after 15 Ma, although in this reconstruction it is less clear exactly when the Palau-Kyushu Ridge had moved sufficiently to be able to shut off that supply. Hall et al.  agrees with the reconstruction of Mahony et al.  in predicting passage of the Palau-Kyushu Ridge along the southern tip of Kyushu around 10 Ma. Thus, prior to that time the river could have supplied sediment directly in the Shikoku Basin, but once the ridge had passed to the east then this would have been a major barrier ponding sediment that did reach the Philippine Sea to the west of the ridge and cutting off supply to ODP Site 1177.
 Since 2 Ma, the U-Pb age populations of zircons in the Shikoku forearc largely mimic the age populations seen in the Kiso, Tenryu, and Yodo Rivers, as well as the Shimanto and Sanbagawa complexes (Figures 6 and 7). This implies that the dominant sediment sources lay in mainland Honshu and Shikoku and that sediment transport was mostly orthogonal to the strike of the continental margin. It is noteworthy that the forearc sediments do not contain more than a few young grains, such as is typical of what is now seen in the Fuji River. This means that there has been no delivery of young sediment eroded from the Izu-Honshu collision zone along the trench at any time since at least 2 Ma in contradiction of the model of Fergusson . In some ways, this is surprising because reconstructions of the type advanced by Mahony et al.  indicate that the collision point migrated past the forearc drill sites during the Mio-Pliocene. Because geological evidence at the modern collision point indicates that collision has been ongoing in that place since around 8 Ma, our provenance information suggests that this material must be ponded close to the collision point and not be distributed far along the trench. This may reflect the presence of a topographic barrier within the trench system preventing wider dispersal, although we do not see any feature like that today. Alternatively, it is possible that sediment accretion closer to the collision zone was able to store the sediment delivered by the Sugura Canyon, reducing transport to the southwest. We do note, however, that this study was only able to analyze two sands from IODP Sites C0006 and C0007. If sediment was being derived from both the Shimanto-Sanbagawa complex and from the collision zone, then it is possible that we only analyzed deposits from a single source. Denser sampling of this section may reveal sediments that were derived from the collision zone. However, the complete absence of grains with U-Pb zircon ages <10 Ma would preclude any reworking and any mixing between sediment from these two sources. We presently feel that this is unlikely and that, although we cannot exclude any flux from the collision zone, this does not appear to be volumetrically significant.
 The conclusions about sediment provenance derived from the zircon U-Pb ages can be cross-checked against fission track data from the potential basement sources. Although there are not much apatite data published, there are more zircon fission track data to compare with. Data from the collision zone in the Kanto region around Tokyo comes from Otofuji et al. . Constraints on the Sanbagawa complex comes from Wallis et al.  and Shinjoe and Tagami , while those from the Shimanto complex were published by Hasebe et al. , Hasebe and Tagami , Hasebe and Watanabe , and Hasebe and Hoshino . Figure 10 shows the clear difference between the two regions, with young ages concentrated in the collision zone of Kanto and older ages clustering around 50–80 Ma being typical of the Shimanto and Sanbagawa complexes. The common peak in the zircon fission track of the forearc sediments at 60–80 Ma correlates well with that seen in the Kiso, Tenryu, Nagara, and Yodo Rivers (Figures 5a, 5b, and 5d), as well is that in the Shimanto and Sanbagawa complexes (Figure 10), suggesting a common source, i.e., erosion of the Shimanto and Sanbagawa complexes. This result is consistent with the U-Pb conclusions that it is erosion/recycling of the older accretionary complexes that dominates sedimentation onto the forearc since at least 2 Ma. The young fission track ages seen in the collision zone match well with those found in the Fuji River but are rare on the trench slope offshore Kyushu. Only one of the samples has even a single grain dating <10 Ma for ZFT (Figures 5g and SI-2I). It is noteworthy, however, that some of the sediment samples have much older tails to their age spectrum than might be expected from a simple Shimanto-Sanbagawa source. Zircon fission track ages from these terrains are rarely older than 130 Ma, while grains older than this are found in the Nagara River (Figure 5c), as well as at ODP Site 1176 in sediments dated at 1.86 Ma (Figures 5g and SI-2I), at ODP Site 1178 dated at 5.1 Ma (Figures 5f and SI-2G) and 7.8 Ma (Figure SI-2 F), and at ODP Site 1177 in sediment dated at 6.46 Ma (Figure SI-2 K), although this latter sample has a very small number of analyses (N = 10).
 The data from the Nagara River shows that there must be slowly cooled source rocks presently exposed onshore in Honshu but which have yet to be analyzed in the original bedrock. Like in the U-Pb zircon data, we see a temporal evolution in fission track ages in that it tends to be the older sediments that show the presence of more slowly cooled materials, while the younger samples (<2 Ma) and most of the modern rivers are dominated by erosion from the Shimanto and Sanbagawa complexes. This change is consistent with a north-to-south shift in the location of the area of strongest erosion and sediment production within the Japanese islands since the Miocene. Again, the zircon fission track ages argue against major along-strike sediment transport by eliminating significant sediment flux from the Kanto/Tanzawa region (Figure 1).
 The fission track data not only help understand the provenance of the sediments deposited in the Nankai forearc but also can constrain how the exhumation rates change through time. Figure 11 shows how the central ages of the apatite and zircon fission track compare with the depositional age. A simple comparison of this variety allows us to understand what is the lag time in the dominant source terrains between the cooling of crystals through the closure temperature and their deposition at the drill site at which they were sampled. This comparison is slightly more complicated with fission track data compared to most other thermochronology data because fission tracks experience a partial annealing zone rather than a simple closure temperature. Nonetheless, when cooling is rapid, this approach can still provide a useful image as to the general character of the exhumation process.
 Apatite population ages are consistently younger than those from the zircon fission track (Figure 11), which is not surprising given that the apatite fission tracks become preserved at lower temperatures than those in zircon. One sample from ODP Site 1177 and another from IODP Site C0006 appear to show a group of grains with apatite cooling ages that are close to the age of deposition suggestive of very rapid exhumation at that time. The wide range of fission track cooling ages indicates that we have a range of lag times between cooling and deposition up to >120 Ma. The Fuji River shows a lag time of ~10 Ma, consistent with the exhumation rates in the modern collision zone being faster than in other parts of the Japanese islands, although we do note that there are also small populations of grains in both the Tenryu and Kiso Rivers with lag times even shorter than that, indicating that there are rapidly exhumed sources within these river basins as well, even if they do not account for the bulk of the sediment in those rivers. Lag times are typically between 40 and 100 Ma in the Tenryu, Nagara, and Yoda Rivers. This diagram shows us that rates of exhumation are quite variable within the source regions of the sediments found in the Nankai Trough.
 Rough estimates of the exhumation rates in the dominant source regions can be made by simplifying the cooling history. We assume that cooling was mostly rapid and that fission tracks began to form in apatite after the crystals cooled below 110°C and below 250°C for zircon. A rough estimate of the degree of exhumation can be then made if we apply a geothermal gradient of ~25°C/km, representing the normal gradient in the forearc, while recognizing that this is much higher during periods of ridge subduction [Sakaguchi, 1996]. Regional heatflow compilation show that within those areas now providing sediment into the Nankai Trough, geothermal gradients vary from 18.5 to ~40°C/km, but with much of the area lying close to 25°C/km [Okubo et al., 1989]. In this case, the average rate of exhumation of the entire source region can be estimated by assuming that the crystals cooled below the critical temperature at the time of the central age and subsequently reached the surface and were deposited at the time derived from the biostratigraphic constraints in the cruise report. The thermal gradient corresponds to a closure depth of 5.94–2.75 km for apatite and an average of 4.4 km. Zircon tracks would close at 13.5–6.25 km, with an average of 10 km. The average uncertainty is around 35% in these cases. The two fission track ages provide us with the duration of exhumation through the upper crust. Such an estimate is clearly rather approximate and assumes rather rapid cooling, which is not always the case in southwest Japan. Nonetheless, such an approach gives us a rough impression of how exhumation rates would have varied in the past because younger fission track ages indicate faster exhumation than older dates.
 The age of deposition of the different samples allows the reconstruction of the temporal variation in exhumation rates for all sites from which fission track data are available. The Fuji River in the modern day shows the highest dominant value of 252–551 m/Ma, although a minority population in the Tenryu River implies exhumation rates of 931–2034 m/Ma in some parts of that basin. The dominant young grains in the Fuji River reflect the rapid exhumation in the collision zone and the fact that this sediment is not distributed along the margin to the drill sites. We do note, however, that one of the sands from ODP Site 1177 located on the subducting plate and deposited at 9.4 Ma also has relatively rapid exhumation rates/short lag times. Rates are slightly slower at the same site at 7.8 Ma, albeit based on a rather small number of fission track ages. Otherwise, exhumation rates tend to be relatively constant at <0.1 km/Ma throughout the period of study but show a moderate increase since 2 Ma and in the modern rivers where they average 0.1–0.2 km/Ma. One of the IODP Site C0006 samples whose deposition is dated at 1.08 Ma shows higher than normal rates (346–756 m/Ma), and this may reflect some modest influence from the Izu-Honshu arc collision zone.
 Exactly what is causing the increasing exhumation since 2 Ma it is not clear, although they may be a climatic component related to the onset of Northern Hemispheric glaciation because exhumation rates are believed to increase worldwide at this time, ~3 Ma [Zhang et al., 2001]. The cause of the faster exhumation rates around 9.4 Ma compared to subsequent erosion is unclear. Faster erosion could potentially be linked to a weakening of the monsoon after 10 Ma, driving slower erosion rates through the Late Miocene [Clift et al., 2008], but might also reflect deformation onshore. There is no evidence for any particular change in regional plate tectonic stresses around this time, although this was close to the time at which the Okinawa Trough began to rift, which in turn could have influenced uplift and erosion rates in mainland Japan, especially on the southwestern island of Kyushu.
 In this study, we examined the thermochronology of a series of samples from modern river systems in the southwestern Japanese islands and compared those with similar data from samples taken from drill cores dating back to 15.4 Ma in the Nankai Trough forearc and on the subducting Shikoku Basin oceanic crust. We collected fission track and U-Pb zircon ages to determine the source of the sediment and to reconstruct how rates of exhumation have changed through time since 15.4 Ma. These data show that sediment dispersal from the collision point between the Izu-Bonin Arc and mainland Honshu has been quite limited and has not affected sedimentation offshore Shikoku or even offshore the Kumano Basin in central Honshu (Figure 1). This implies that the sediment preserved in the accretionary complex in the Nankai Trough largely reflect erosion of the onshore forearc directly adjacent to that part of the trench system rather than from a long distance along strike.
 The modern rivers show dominant erosion from the Shimanto and Sanbagawa accretionary complexes, with average exhumation rates of ~100–150 m/Ma. In contrast, the Fuji River, draining the collision zone, shows much higher average exhumation rates of ~400 m/Ma. Sediment in the forearc shows the temporal evolution in erosion patterns. Since around 2 Ma, erosion has been focused on the Shimanto and Sanbagawa complexes. Prior to ~5 Ma, there is evidence for significant flux of sediment more slowly exhuming terrains, although it is unclear what tectonic blocks these were. Sediments deposited in the Shikoku Basin at 9.4 Ma show a mixture of erosion from the Shimanto and Sanbagawa complexes but with a significant flux from an older source, likely North China Craton basement from northern Honshu on which the Mesozoic arc had been built. Exhumation rates were, however, quite rapid, possibly driven by a strong summer monsoon coupled to rapid rock uplift. The oldest sediment shows that at 15.4 Ma sediment supply involved the Shimanto and Sanbagawa complexes, together with both North China and Yangtze Craton sources. Such a mixture requires sediment flux from a very wide area not seen subsequently in the Nankai Trough. One possibility is that the Yangtze River itself was supplying sediment to the deep Shikoku Basin at that time prior to rifting of the Okinawa Trough and before motion of the Palau-Kyushu Ridge blocked this supply route.
 We wish to thank the Natural Environment Research Council (NERC) and UK Integrated Ocean Drilling Program for financial support for this project. P.C. and U.N. thank the University of Aberdeen for their help and support during the early stages of this project when both authors were members of the School of Geosciences. This paper benefited from comments by editor Todd Ehlers, associate editor Paul Kapp, and reviewer Gaku Kimura.