The upper Yangtze River flows southward on the southeastern Tibet characterizing by uniquely low and continuous relief. The river makes a sharp turn at Shigu, heading northeast, and forms the first bend of the Yangtze River. Many previous studies assumed southward flow of the ancestral Yangtze River from Shigu to the South China Sea. However, field evidence of southward flow of the paleo-Yangtze is lacking. In this paper we report our identification, based on detrital zircon U-Pb age distributions, of a range of fluvial sands left by the paleo-Yangtze in Tongdian, Madeng and Nanjian basins. Cosmogenic10Be and 26Al burial dating provides burial ages for these fluvial sands from 1.7 to over 8.7 Ma. Rerouting of the Yangtze River therefore occurred within the last 1.7 Ma, postdating the major uplift of the central Tibet. We attribute the rerouting of the Yangtze River to response to activation of the Dali fault system, and in a larger scale, initiation of crustal deformation by clockwise rotation around eastern Himalayan syntaxis 2–4 Ma ago. Reorganization of the Yangtze drainage pattern most likely reflects regional uplift and displacement due to lower crust flowing beneath major faults in the southeastern Tibet and Yunnan.
 Originating on the Tibetan Plateau, the upper Yangtze River flows southeastward on a landform with distinctively low and continuous relief until it reaches Shigu (Figure 1). At Shigu, the river makes a 110 degree turn within a distance of 1 km, heading northeast, and forms a V-shaped bend. This big turn is called the first bend of the Yangtze River.
Barbour suggested a large-scale reversal of the middle Yangtze River (from the first bend to the Three Gorges area, see inset inFigure 1) in which the river originally flowed southward from Shigu passing through the Jianchuan Basin into the South China Sea. Ren et al.  and Shen and Yang  supported the claim of Barbour with detailed field investigations in the Jianchuan Basin. In contrast, Li  considered the Jianchuan Basin as produced by glacial erosion and Yuan  explained the morphology of Shigu as entrenched meander. Discussions on the formation of the first bend continue; some favor the reversal hypothesis [Zhang et al., 1998; Yang and Li, 2001; Clark et al., 2004; Clift et al., 2006a; Ming et al., 2007] and some argue against [He et al., 1989; Zhao et al., 2008]. Brookfield  suggested that setting of the current drainage systems of Southeast Asia, including the Yangtze, Mekong, Salween and Tsangpo rivers, occurred in Pliocene to Quaternary times. Clark et al.  and Clift et al. [2006a]link the disruption of the paleo-drainage of the Yangtze River to the rise of Tibetan Plateau andHoang et al. prefer that any connection of the Salween, Mekong, and Yangtze rivers to the Red River must be pre-Middle Miocene. All these reversal-preferred studies assumed that the paleo-Yangtze River flowed south passing through the Jianchuan Basin. Unfortunately field investigations in the past tens of years and recent drilling in the Jianchuan Basin (E. Wang, unpublished data, 2012) have not revealed the existence of fluvial sediments in the basin.
Kong et al. [2009a]studied lacustrine and fluvial sediments located between Zhongjiang and Panzhihua, presenting evidence for the reversal of the middle Yangtze River occurring 1.3–1.6 Ma ago. They further supposed that capture of the first bend occurred prior to the reversal of the middle Yangtze River. Continuing the study we have carried out detailed field investigations along possible drainage courses of the paleo-Yangtze from Shigu to the Red River. We present in this paper provenance and time constraints on the formation of the first bend of the Yangtze River.
2. Geological Setting and Sampling
 Most previous studies of the formation of the first bend focus on the Jianchuan Basin, through which Barbour  assumed the Yangtze River joined in the Red River and then flowed into the South China Sea. Our field investigations show that diluvial deposits dominate in the outcrops within the basin. The diluvial deposits overlie red weathering crust that probably formed in Pliocene [Wang et al., 1998]. Older sediments occur at elevations of 2500–4000 m in the west of the Jianchuan Basin. Ren et al. claimed that the conglomerates located west of the Jianchuan Basin are relics of the paleo-Yangtze River. The conglomerates, poorly sorted and from angular to sub-rounded (Figure 2), are largely composed of limestone; they are marked as molasse formation in Chinese geological maps [Yunnan Bureau of Geology and Mineral Resources (YBGMR), 1990]. The conglomerates are intercalated with mudstone/sandstone and the stratum is designated as of possible Oligocene age in local geological maps [YBGMR, 1990]. The strata measure >2670 m in thickness and commonly occur between Lijiang and Lanping, an east-west width of ∼60 km. The distribution, the texture and the petrologic characteristics of these Paleogene sediments suggest that they are not remnants of the paleo-Yangtze River.
 The Heqing Basin, located east of the Jianchuan Basin (Figure 1), consists of sediments of ∼700 m thick, which are dated to 2.78 Ma [Xiao et al., 2006; An et al., 2011]. The sediments are mostly lacustrine, fluvial deposits are sporadic and do not outline a continuous drainage course; these fluvial deposits are mostly likely left by local streams, rather than the paleo-Yangtze.
 West of the Jianchuan Basin, we found a south-north range of fluvial deposits in Tongdian and Madeng basins. The deposits in Tongdian are designated as of Pliocene age and those in Madeng are marked as of Pleistocene age in local geological maps. To the north, the fluvial deposits stretch very close to Luoguqing, ∼6.6 km southwest of Laojun Shan. Fluvial sands are predominant in the outcrops in Tongdian and exhibit parallel- or cross-bedding (Figures 3a, 3b and 3c). In contrast to the tilted strata in Tongdian, interbedded fluvial sands and pebble are horizontally layered in Madeng (Figure 3d). From Madeng to Qiaohou, basin narrows to a gorge and we have not found fluvial deposits. Further south, widespread coal-bearing lacustrine/fluvial sediments occur between Qiaohou and Cuiping. According to geological maps [YBGMR, 1990], the sediments are of Pliocene Sanying formation. To the south end, we found fluvial sands intercalating with pebbles and cobbles of ∼200 m thick in the Nanjian Basin (Figures 3e and 3f). Here streams from Weishan and Midu join together, forming the headwater of the modern Red River.
 We collected three fluvial sand samples (YN27-29) in the Tongdian Basin, two sand samples in Madeng (YN40) and nearby (YN26), a lacustrine/fluvial deposit in Cuiping (YN41) and two fluvial sands in Nanjian (YN43-44). Fluvial sand samples are preferably taken because they are more representative in provenance study and in cosmogenic nuclide burial dating. These samples are taken from certain depth of the sections which are newly exposed due to construction. Sample information and burial depth are given inTable 1. Through detrital zircon U-Pb age studies and cosmogenic nuclide burial dating, we aim to clarify whether these fluvial deposits are remnants of the paleo-Yangtze, and when the paleo-Yangtze passed through these basins.
Table 1. Sample Information and Burial Ages of the Fluvial Sand Samples
Assuming a post-burial denudation rate of 20 m/Ma for locations where the samples were collected.
4.74 ± 0.28
16 ± 4
3.4 ± 0.8
1.4 ± 0.5
2.0 ± 0.7
8.09 ± 0.32
18 ± 12
2.3 ± 1.5
2.3 ± 1.1
2.9 ± 1.4
0.801 ± 0.056
7.3 ± 3.0
9.1 ± 3.7
1.02 ± 0.06
3.2 ± 2.1
3.2 ± 2.1
1.6 ± 1.1
8.78 ± 0.79
29 ± 10
3.3 ± 1.1
1.5 ± 0.6
1.7 ± 0.7
12.5 ± 0.5
14 ± 4
1.1 ± 0.3
3.8 ± 0.5
5.7 ± 0.8
2.30 ± 0.12
1.6 ± 1.4
0.7 ± 0.6
4.7 ± 1.3
3.1. Zircon U-Pb Age
 Zircons were separated from the sand samples by conventional magnetic and heavy-liquid separation techniques, followed by hand selection under a binocular microscope. Over 1000 zircon grains were separated from each sample from which over 200 grains were randomly mounted on a double-sided tape, cast in epoxy resin, and polished to expose surfaces for laser ablation ICP–MS analysis.
 The U-Pb age determinations were made using an Agilent 7500a quadrupole ICP–MS equipped with a Geolas 193 nm laser ablation system at the Institute of Geology and Geophysics, Chinese Academy of Sciences. When zircon grains are large enough, zircon rims were analyzed with laser spots fixed at 40μm. Each spot analysis lasted for 30 s of background acquisition and 40 s of sample data acquisition. Detailed analytical procedures are described in Xie et al. . During sample analyses we have measured 91500 and GJ-01 within every ten zircon grains. NIST610 were also measured between samples for rare earth element information and U/Th concentrations.207Pb/206Pb, 206Pb/238U, 207Pb/235U and 208Pb/232Th ratios were calculated using GLITTER 4.0 (Macquarie University), and were corrected for instrumental mass bias and depth-dependent elemental and isotopic fractionation using Harward zircon 91500 as an external standard. Common Pb was corrected according to the method proposed byAndersen .
3.2. 26Al/10Be Burial Dating
 Cosmogenic nuclide burial dating, first proposed by Lal and Arnold , relies on different decay constants for 10Be and 26Al (see Granger and Muzikar  for a review of the technique). The assumption behind the method is that quartz is exposed to cosmic rays at the earth's surface for a period of time, acquiring certain amounts of 10Be and 26Al, NBe(0) and NAl(0), respectively. After the quartz is buried at depth for time t, the concentrations of 10Be and 26Al become
In (1) and (2)λBe and λAl are the decay constants of 10Be and 26Al, NBe(depth) and NAl(depth) are concentrations of 10Be and 26Al produced after burial, respectively. If quartz is gradually exposed in a landform eroding in steady state, with a pre-burial erosion rate ofε (cm/yr), equations (1) and (2) can be written as
Here L is the attenuation length of cosmic ray nucleons (neutrons + protons). PBe and PAl are the production rates of 10Be and 26Al, respectively. The concentrations of NBe(t) and NAl(t) are calculated from stable 9Be and 27Al concentrations together with 10Be/9Be and 26Al/27Al ratios determined by accelerator mass spectrometry (AMS). If post-burial productions of10Be and 26Al are negligible, both burial age tand pre-burial erosion rateε can be calculated from equations (3) and (4).
 If quartz is burial at a limited depth and post-burial productions of10Be and 26Al are not negligible, the 10Be and 26Al produced since burial must first be subtracted from NBe(t) and NAl(t). NBe(depth) and NAl(depth) consist of three components: those produced by nucleons, by fast and negative muons. Production by nucleons, fast and negative muons as a function of depth can be modeled by the following equations [Granger and Smith, 2000]:
where PBe(z) and PAl(z) are the production rates of 10Be and 26Al at depth z, respectively. For muonic production, the first two terms describe the production due to negative muons, and the third term describes the production relating to fast muons. Considering a landform with a constant post-burial erosion rateE (cm/yr), then
The adopted values for A1, A2, A3, B1, B2, B3, L1, L2 and L3 in this study are summarized in Table 2. For the description of these parameters, please refer to Granger and Muzikar . Combining equations (9) and (10) into (3) and (4), we get three unknowns (t, ε and E) in two equations. One unknown (e.g., the post-burial erosion rate) has to be assumed in order to calculate the burial timet.The minimum burial age corresponds to no post-burial production of10Be and 26Al or a very high post-burial erosion rate of the landform. When assuming zero or the lowest post-burial erosion rate, we obtain the maximum burial age. In actual calculations, the best fit burial age is determined by chi-square minimization, a method of gradually approaching the best fitting values.
Table 2. Adapted Parameters for Post-Burial Muonic Productiona
The numbers are the subscripts of L, A and B.ρ is the density of overburden sediments, which is taken as 2.5 g/cm3 in this study.
3.3. Sample Preparation for 26Al/10Be Burial Dating
 Chemical preparations, from extraction of quartz to final oxide and mixing with binder, were performed at the cosmogenic nuclide laboratory at the Institute of Geology and Geophysics, Chinese Academy of Sciences, in Beijing. Samples were first treated in 3N HCl for 6 h on a hot-plate to dissolve carbonate. Meteoric10Be was then removed by repeat ultrasonic leaching at 80°C with a mixed solution of dilute HF and HNO3. Pure quartz samples (20–70 g) were subsequently dissolved after addition of ∼0.5 mg 9Be carrier. Aliquots were taken from the dissolved samples for Al concentration measurement before HF volatilization. Be and Al were progressively separated by anion exchange, acetyl acetone-CCl4 extraction, cation exchange and selective precipitation, and finally converted to oxide.
 BeO was mixed with Nb and Al2O3 with silver and both 10Be/9Be and 26Al/27Al ratios were measured using AMS at PRIME Lab, Purdue University. Ratios of 10Be/9Be are normalized to the ICN standard Be-01-5-2 with a ratio of 8.558 × 10−12. Half-lives of 1.387 Ma [Korschinek et al., 2010] and 0.705 Ma, and high-latitude, sea level production rates of 4.6 atoms/g/yr and 31.1 atoms/g/yr are used for10Be and 26Al, respectively, in the calculation.
4.1. Zircon U-Pb Age
 Zircons in four fluvial sand samples were analyzed for U-Pb age distributions. For each sample 120 zircon grains were analyzed for U-Pb ages. The obtained207Pb/206Pb, 206Pb/238U, 207Pb/235U, 208Pb/232Th ratios and U-Pb ages are summarized in Table S1 in theauxiliary material. To infer information regarding the provenance of the samples we use only analyses in which the Th/U ratios >0.1, which indicate igneous origin [Hoskin and Schaltegger, 2003], and the 206Pb/238U and the 207Pb/235U ages concordant within 15%. 206Pb/238U ages are used for zircons younger than 1 Ga, and 207Pb/206Pb ages for zircons older than 1 Ga in plotting histograms (Figure 4). As shown in Figure 4 the four samples, YN28, YN40, YN41 and YN44, display quite similar spectra with ages peaking at 250 Ma, 440 Ma, 1000 Ma, 1800 Ma and 2500 Ma. These ages represent characteristic events associated with the Yidun volcanic arcs, the Yangtze platform and the North China craton [Kong et al., 2009a]. Table 3presents quantitatively the numbers of zircons in these prominent age ranges. Among them zircons with U-Pb ages of 1700–2000 Ma and 200–500 Ma are predominant for all four samples. The similar distribution of the ages characterizing of different geological units and events suggests a similar process of formation of these samples.
Table 3. Zircon U-Pb Age Distributions in the Fluvial Sand Samples Collected From Tongdian to Nanjian and in the Modern Sands From the First Bend of the Yangtze River, the Mekong River, and the Red River
4.2. 26Al/10Be Burial Ages
 We have dated seven fluvial sands collected from Tongdian to Nanjian. Sample information, 10Be and 26Al concentrations together with burial ages are given in Table 1. AMS 10Be/9Be ratios of the samples show a range of (17–101) × 10−15 with analytical errors of 4–9%. 10Be concentrations are calculated from the amounts of added 9Be carriers and the 10Be/9Be ratios after correction for full chemistry procedural blanks with an average 10Be/9Be ratio of 5 × 10−15. 26Al/27Al ratios of the samples are low, ranging from 3 × 10−15 to 38 × 10−15, with analytical errors of 24–87%. AMS measurement of the chemical blanks prepared from a pure Al solution show 26Al/27Al ratios around the detection limit of 3 × 10−15. 26Al concentrations are calculated from the 26Al/27Al ratios and the corresponding 27Al concentrations in quartz samples which range between 187 and 389 ppm.
 The minimum burial ages are calculated with equations (3) and (4)by assuming no post-burial production of10Be and 26Al. Corrections of any post-burial production of10Be and 26Al by muons and nucleons for samples collected at limited burial depths will enlarge the burial ages. As we discussed above, there are three unknowns in two equations; we have to assume one unknown in order to calculate burial ages. Chappell et al.  obtained an average 10Be concentration of 1.1 × 106 atoms/g for quartz extracted from modern sands collected from Panzhihua, and they gave a 10Be-based catchment erosion rate of 15–25 m/Ma for the upper Yangtze River. From the first bend to Panzhihua, there is no a large tributary joining in the Yangtze River. Hence the10Be concentration of the quartz extracted from the modern sands at Panzhihua should not be too much different from that associated with the first bend. Assuming a post-burial erosion rate of 20 m/Ma, burial ages for the samples, with the exceptions of YN28, YN29 and YN44, are calculated and shown inTable 1. Downstream of Panzhihua along the Yangtze River, two modern sand samples collected right below the joint point of the Yalong River and at Yibin have 10Be concentrations of 0.45 × 106 atoms/g and 0.58 × 106 atoms/g for quartz, respectively [Chappell et al., 2006]. The relatively lower 10Be concentrations resulted from the addition of sands with a low 10Be concentration for quartz from the Yalong River (0.13 × 106atoms/g) and reflect larger catchment erosion rates. The larger the post-burial erosion rate is assumed, the closer the burial age is to the minimum burial age.
 Three samples, YN28, YN29 and YN44, are exceptional in the burial age calculations. The 10Be and 26Al concentrations of YN28 are very low, and the corresponding 26Al/10Be ratio is 9.1 ± 3.7, higher than the model 26Al/10Be ratio of 6.76 ± 0.88 at the earth's surface [Balco et al., 2008]. The low concentrations and the high ratio are consistent with production by fast and negative muons after burial, which yield a 26Al/10Be ratio of 8.5 (Granger and Smith , after correction for the new half-life of10Be). Judging from the high 26Al/10Be ratio, we consider the 10Be and 26Al in YN28 as largely produced by muons after burial. The minimum burial age of YN28 can then be calculated by assuming that the 10Be concentration of 0.80 × 104 atoms/g in YN28 results from the decay from an original concentration of 1.1 × 106 atoms/g as determined by Chappell et al. ; calculation shows that YN28 has been buried for over 10 Ma. The burial age of YN28 depends strongly on the assumed starting concentration of 10Be. The starting 10Be concentrations for YN26, YN27, YN40 and YN43 are 1.0 × 105 atoms/g, 3.1 × 105 atoms/g, 2.1 × 105 atoms/g and 19 × 105 atoms/g, respectively, as calculated from the current 10Be concentrations and the burial ages of these samples. The average value, 6.3 × 105 atoms/g, is about half of 1.1 × 106 atoms/g determined by Chappell et al. . Starting from a 10Be concentration of 6.3 × 105 atoms/g, the minimum burial age of YN28 is 8.7 Ma.
 The 10Be and 26Al concentrations of YN29 are also very low, but with a low 26Al/10Be ratio (3.2 ± 2.1). With the assumed post-burial erosion rate of 20 m/Ma, the26Al concentration produced by fast and negative muons already exceeds the current 26Al concentration in YN29. That is, the equations give no resolution of the two unknowns (the burial age and the pre-burial erosion rate). To reproduce the26Al concentration in YN29, the post-burial erosion rate needs to be ≥68 m/Ma. The lowest erosion rate yields the largest burial age for YN29, which is 10.2 Ma.
 Similar to YN29, to reproduce the low concentration of 26Al in YN44, the post-burial erosion rate has to be ≥95 m/Ma. This lowest erosion rate yields the maximum burial age of 11 Ma for YN44. Based on above estimations, the burial ages of the fluvial sands collected from Tongdian to Nanjian range from 1.7 Ma to over 8.7 Ma.
5.1. Provenance of the Fluvial Sands
 The collected samples are located between the Yangtze, the Mekong and the Red rivers (Figure 1). For comparison the U-Pb age spectra for the modern sands collected from the bank of the Yangtze River at Duomei (YM32,Figure 1) and at the first bend [Hoang et al., 2009], from the Mekong River at the boundary between China and Thailand (Mekong-1 [Bodet and Schärer, 2000]) and at the mouth (Mekong-2 [Clift et al., 2006b]), from the Red River at Lao Cai and Yen Bai [Hoang et al., 2009] are shown in Figure 5. In these spectra, U-Pb ages of ∼1800 Ma and ∼2500 Ma, mainly derived from the Songpan-Ganzi flysch belt, are signatures of North China craton [Bruguier et al., 1997; Weislogel et al., 2006]. Ages of ∼1000 Ma represent the nature of pre-rifting granitoids widespread in the Yangtze platform [Li and McCulloch, 1996] and the Late Permian and Triassic ages represent the large volume of granite and granodiorite plutons outcropped in the southeast China, developed as the Yangtze platform and the Qiangtang block moved northward toward Eurasia [Reid et al., 2007]. The Yangtze River starts from the west margin of the Songpan-Ganzi flysch belt, passing a long course through the Qiangtang block (Figure 6). In contrast, the Mekong runs in the Qiangtang block, stretching between Late Permian and Triassic granite and granodiorite plutons, and shows a zircon U-Pb age spectrum dominantly by 250 Ma (Figure 5); zircons with U-Pb ages in the range of 200–500 Ma occupy 53% and 70% of total for the modern sands collected from the Mekong River (Table 3). The red River is developed in the west margin of the Yangtze platform (Figure 6). It stretches in the basins with Triassic to Jurassic sediments and outcrops of Proterozoic metamorphic rocks. Thus, compared to the Mekong River, the modern sands from the Red River contain more zircons of 700–1100 Ma (4% and 13% vs 31% and 36%, Table 3), a signature of the Yangtze platform. The differences in U-Pb age distribution provide fingerprints to allow distinguishing fluvial sands formed by local streams, or left over by the Yangtze and the Mekong rivers.
 The four samples, collected from Tongdian to Nanjian, are located in the Lanping-Simao basins, a geological unit completely different from the Songpan-Ganzi flysch belt. In the Lanping-Simao basins, continental sedimentation is recorded from Later Triassic, to Jurassic, Cretaceous and to Paleogene. Currently these sediments, with sporadic granitoids of Late Permian and Triassic ages, are exposed in the basins. Available isotopic data show that igneous and metamorphic rocks within the basins were dominantly formed within 1000 Ma [Wang et al., 2003]. U-Pb age features of the sedimentary rocks can be inferred from the rivers flowing through the basins. The Mekong River stretches in the Triassic to Cretaceous sediments in the Lanping-Simao basins, and show zircon U-Pb ages mainly of 250 Ma (Figure 5). Additional evidence comes from the U-Pb age spectra for the sands of the Red River. The Red River also runs in Triassic and Jurassic sediments and shows zircon U-Pb ages dominantly by 250 Ma and 800–1000 Ma. The low contents of old zircons (1800 Ma and 2500 Ma) in the Mekong and the Red rivers suggest that these old zircons are minimal in the sediments filling the Lanping-Simao basins. Therefore, the high contents of zircons of ∼1800 Ma and ∼2500 Ma in the four samples collected from Tongdian to Nanjian indicate that these old zircons are not derived from local recycled-sediments, i.e., the four sands are not formed by local streams. BothFigure 5 and Table 3show that the zircon U-Pb age spectra of the four samples collected from Tongdian to Nanjian are similar to the Yangtze River, suggesting that these fluvial sands are remnants of the paleo-Yangtze River.
5.2. Formation of the First Bend of the Yangtze River
Figure 1shows the possible course that the paleo-Yangtze flowed south to the Red River: passing through the central range of the Yunling collage, to Tongdian, Madeng, Weishan, and then to Nanjian, inferred from the distribution of the relic sands of the paleo-Yangtze. The Yunling Collage, a branch of Hengduan Mountains, runs north to south and separates into three ranges in Yunnan province. The west range includes Baima Shan and Xuebang Shan, the central range comprises Laojun Shan, Diancang Shan and Ailao Shan, and the east range consists of Yulong and Haba mountains. These mountain ranges lie within the Dali fault system (Figures 1 and 7). The Dali fault system is rhombus-shaped, and enclosed by to the southwest the Tongdian fault, to the east the Chenghai fault, to the northwest the Daju fault, and to the west the Zhongdian-Jianchuan fault [Wang et al., 1998; Fan et al., 2006]. These faults show left-slip with variable components of normal displacement.Fan et al. demonstrate that the NE-SW directed striking normal faulting along the middle segments of the Heqing and Heihuijiang faults resulted in the relative uplift of the Yulong Mountain massif and the concomitant subsidence of the Heqing and Lijiang basins. Similarly, NE-SW directed normal faulting along the northern and southern boundaries of the Diancang Shan massif caused the uplift of the Diancang Shan and subsidence of the Qiaohou Basin and Erhai Lake. Fast unroofing of Diancang Shan was found to start at 4.7 Ma [Leloup et al., 1993] and the preserved sedimentary sequence in the Heqing basin was dated to as old as 2.78 Ma [Xiao et al., 2006]. Kong et al. [2009b] studied the late Quaternary glacial deposits in the Daju Basin and inferred an uplift rate of 5.6 m/ka for the Yulong Mountains. Dating directly the Daju fault scarp, Kong et al. obtained a dip-slip rate of 5.7 m/ka for the Daju fault. The consistent dip-slip rate suggests that the uplift of the Yulong Mountains mainly occurred in the Quaternary. All these results suggest initiation of the Dali fault system during the Late Pliocene or early Pleistocene.
 The fluvial sands YN28 in Tongdian have been buried for over 8.7 Ma. Accordingly the paleo-Yangtze River flowed south through Tongdian, Madeng, Nanjian, and joined in the Red River since at least 8.7 Ma ago. Initiation of the Dali fault system led to the uplift of the Yunling Mountains. When the mountains grew higher than the elevation at Shigu, the paleo-Yangtze changed its course to northeast, forming the first bend. The sands left by the Yangtze River possess burial ages as young as 1.7 Ma (Table 1), suggesting that the paleo-Yangtze still flowed south through Tongdian and Madeng 1.7 Ma ago. From the contact of fluvial and lacustrine sediments in Taoyuan,Kong et al. [2009a] inferred that capture of the first bend occurred prior to the reversal of the middle Yangtze River. They dated the Xigeda lacustrine sediments and suggested that the middle Yangtze River started to flow east 1.3–1.6 Ma ago. Therefore, the first bend of the Yangtze River should have formed 1.3–1.7 Ma ago.
5.3. Cause for Reorganization of the Yangtze River
 The upper Yangtze River flows on the southeastern Tibet characterizing by uniquely low and continuous relief, 5 km elevation gain over 1500 km in length. Clark et al. attribute the formation of long wavelength tilt of the southeast plateau margin to coeval with the crustal thickening of the plateau, and they link the disruption of paleo-drainage to the rise of the eastern Tibet.Lunt et al.  suggest that the discharge from the Yangtze River can provide a good test for our understanding of the uplift history of the Tibetan Plateau. Our results clearly demonstrate that reorganization of the Yangtze River postdate the major uplift of the central Tibet, which most likely occurred 26 Ma ago [Rowley and Currie, 2006; DeCelles et al., 2007]. Rather the reorganization of the Yangtze River relates to activation of the Dali fault system, or in a larger scale, initiation of crustal deformation by clockwise rotation around eastern Himalayan syntaxis 2–4 Ma ago [Wang and Burchfiel, 2000; Schoenbohm et al., 2006]. The lower-crust flow model predicts outward flow of lower crust from beneath to the margin of Tibet and the principle outlet being southeastward, between the eastern Himalayan syntaxis and the Sichuan Basin. Such lower-crust flow thickens the Tibet marginal crust and rises regions farther southeast such as Yunnan [Royden et al., 1997]. Reorganization of large river systems around the southeast margin of Tibet, e.g., the Yangtze River flowing east, the Tsangpo joining in Brahmaputra [Liang et al., 2008], and Mekong to Thailand [Brookfield, 1998], most likely reflects regional uplift and displacement due to lower-crust flow beneath major faults in the southeastern Tibet. With magnetotelluric imaging,Bai et al. observed two major channels of high electrical conductivity at a depth of 20–40 km beneath the Nujiang suture zone and the Xianshuihe-Xiaojiang fault zone (Figure 7). Kong et al. [2009a]inferred a left-slip rate of ∼2 m/ka for the Chenghai fault that is within the Dali fault system. This rate is much smaller than the slip rates of 15–30 m/ka for the Xianshuihe-Xiaojiang fault zone [Wang et al., 1998]. We interpret the slow clockwise-rotation of the Dali block as to adjust the different flow rates of the two channels beneath the Nujiang suture zone and the Xianshuihe-Xiaojiang fault zone. In this model, the major faults in the southeast Tibet and Yunnan are the upper-crust accommodation to the lower-crust flow. Hence, the activation of the Dali fault system and initiation of crustal deformation around eastern Himalayan syntaxis can provide time constraints for the flow of lower-crust passing through the margin of the southeast Tibet into Yunnan, which appears to be only a few million years ago.
 The southeastern Tibet and Yunnan are characterized by uniquely low and continuous relief on which the Yangtze, the Mekong and the upper Red rivers are flowing. Comparison of U-Pb age distributions for detrital zircons separated from the modern sands of these rivers show that old zircons (∼1800 Ma and ∼2500 Ma), derived from the Songpan-Ganzi flysch belt, are characteristic of the Yangtze River. In contrast, they are deficient within the catchments of Mekong and the Red rivers. Compared to the Mekong River, the Red River contains more zircons of 800–1000 Ma, a signature of the Yangtze platform. These differences provide fingerprints to allow distinguishing deposits left by these rivers.
 Based on detrital zircon U-Pb age distributions, we have identified a range of fluvial sands left by the paleo-Yangtze in Tongdian, Madeng and Nanjian basins, which hint at a southward flow route of the paleo-Yangtze. Using the burial dating method based on the radioactive decay of cosmogenic nuclide10Be and 26Al, we dated these fluvial sands to an age range of 1.7 Ma to over 8.7 Ma. The ages suggest that the Yangtze River flowed through these basins during the period of 1.7–8.7 Ma. Rerouting of the Yangtze River therefore occurred within the recent 1.7 Ma.
 The distribution of the relic sands suggests that the rerouting of the Yangtze River relates to the uplift of the Yunling Mountains, which resulted from activation of the Dali fault system, or in a larger scale, initiation of crustal deformation by clockwise rotation around eastern Himalayan syntaxis 2–4 Ma ago. We attribute the initiation of the fault systems in the southeast Tibet and Yunnan to upper crust accommodation to the lower crust flow beneath these regions. If so, flowing of the lower crust passing through the margin of the southeast Tibet into Yunnan appears to have occurred only a few million years ago.
 We thank Zhao Xitao for help in the field work. We thank Michael Brookfield and an anonymous reviewer for careful review of the manuscript. This work is supported by National Science Foundation of China (grant 41173067 and 41021063).