Earth Surface Processes and Landforms Fluvial bedrock ... et al... · Earth Surf. Process. Landforms30, 955–971 (2005) Published online in Wiley InterScience (). DOI: 10.1002/esp.1256

Fluvial bedrock incision in Taiwan 955

Copyright © 2005 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 30, 955–971 (2005)

Earth Surface Processes and LandformsEarth Surf. Process. Landforms 30, 955–971 (2005)Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/esp.1256

Fluvial bedrock incision in the active mountain beltof Taiwan from in situ-produced cosmogenic nuclidesM. Schaller,1* N. Hovius,1 S. D. Willett,2 S. Ivy-Ochs,3 H.-A. Synal4 and M.-C. Chen5

1 Department of Earth Sciences, University of Cambridge, Downing Street, Cambridge, CB2 3EQ, UK2 Department of Earth and Space Sciences, University of Washington, Seattle, Washington 98195, USA3 Institute of Particle Physics, ETH Hönggerberg, CH-8093 Zürich, and Department of Geography University of Zürich – Irchel,CH-8057 Zürich, Switzerland4 Paul Scherrer Institut, c/o Institute of Particle Physics, ETH Hönggerberg, CH-8093 Zürich, Switzerland5 Taroko National Park Headquarters, Fusu Village, Hualien, Taiwan R.O.C.

AbstractThe concentration of cosmogenic nuclides in rocks exposed at the Earth’s surface is propor-tional to the total duration of their exposure. This is the basis for bedrock surface exposuredating and has been used to constrain valley lowering rates in the Taroko gorge, easternCentral Range, Taiwan. Taroko gorge contains a uniquely complete geomorphic record offluvial valley lowering: continuous, fluvially sculpted surfaces are present in the lower 200 mof this marble gorge. Assuming no post-fluvial erosion of the gorge wall, the concentration ofin situ-produced cosmogenic 36Cl measured in gorge wall marbles reveals exposure ages from0·2 ka in the active channel to 6·5 ka at 165 m above the present river. These ages implyan average fluvial incision rate of 26 ±±±±± 3 mm a−−−−−1 throughout the middle and late Holocene.Taking into account lateral gorge wall retreat after initial thalweg lowering would give riseto calculated older exposure ages. Without considering gorge wall retreat, our estimatestherefore represent maximum incision rates. Estimated maximum Holocene incision ratesare higher than the long-term exhumation rates derived from fission track dating. The long-term gorge development governed by tectonic uplift is superimposed by short-term varia-tions in incision rates caused by climatic or regional tectonic changes. Copyright © 2005John Wiley & Sons, Ltd.

Keywords: fluvial incision; cosmogenic nuclides; tectonics; climate; Taiwan

*Correspondence to: M. Schaller,Department of GeologicalSciences, University of Michigan,2534 C.C. Little Building, 1100N, University Ave., Ann Arbor,MI 48109-1005, USA.E-mail: [email protected]

Received 1 September 2004;Revised 28 February 2005;Accepted 17 April 2005

Introduction

Where hillslopes and valley floors are effectively coupled, landscape lowering is driven by fluvial incision intouplifting rock mass, and hillslopes follow. This is the case in most active mountain belts. Independent assessments offluvial incision and its controls (Tinkler and Wohl, 1998) are fundamental to the understanding of erosional landscapeevolution (Howard and Kerby, 1983), crustal deformation (e.g. Beaumont et al., 1992; Koons, 1989; Willett, 1999),basin fill (e.g. Clift and Gaedicke, 2002), and ocean chemistry and atmospheric composition and circulation (France-Lanord and Derry, 1997; Raymo and Ruddiman, 1992). Rates of fluvial bedrock incision are necessarily influenced byclimatic and tectonic processes (Whipple et al., 1999). Models of physical processes of fluvial incision are conten-tious, but many arguments have been made that incision is proportional to water discharge or discharge variability(Snyder et al., 2003; Tucker, 2004; Tucker and Bras, 2000) and may be enhanced by temperature-dependent weather-ing (Gaillardet et al., 1999), thus linking incision to climatic processes. In addition, tectonic processes perturb riverchannel slopes (Snyder et al., 2000) and/or cross-sections (Lavé and Avouac, 2000) and rock mass properties, thussetting the bed shear stress of a given flow, and the erodibility of its substrate.

The effective evaluation of climatic and tectonic controls on river incision requires quantitative constraints onerosion rates and patterns on timescales of climate change and tectonic forcing. Dated strath terraces are commonlyused for this purpose (e.g. Amorosi et al., 1996; Burbank et al., 1996). Straths are planar remnants of old bedrockchannel floors, isolated above an incising river. They are separated by erosional steps and do not normally permit the

956 M. Schaller et al.


reconstruction of an incision history at high temporal resolution (Pratt et al., 2002). Taroko gorge in east Taiwancontains a rare, continuous record of river incision, consisting of extensive, fluvially sculpted sections of the marblegorge wall. Using in situ-produced cosmogenic 36Cl in marble samples collected from a 200 m high cliff section, wehave dated fluvially sculpted features in the gorge wall, and reconstructed the Holocene incision history of the LiwuRiver. Existing erosion data for the Liwu catchment include annual measurements of fluvial bedrock wear at selectedsites (Hartshorn et al., 2002), decadal suspended sediment transport estimates from hydrometric measurements (WaterResources Agency, 1970–2003; Dadson et al., 2003), millennial valley lowering estimates from dated river terraces(Liew, 1988), and fission track estimates of rock exhumation over a million-year timescale (Liu et al., 2001; Willettet al., 2003), making this an optimal location to study (fluvial) erosion and its controls. Placed in the context ofthese other estimates of incision and erosion rates, our data reveal that Holocene incision by the Liwu River hasbeen considerably faster than the Quaternary average, and that incision rates may have varied within this interval.

Study Area

TaiwanTaiwan is possibly the best documented collision orogen dominated by fluvial processes. The formation of the orogenresults from collision of the Luzon Arc, on the Philippine Sea plate, and the Asian continental margin (Teng, 1990;Figure 1), and links opposite-dipping subduction systems at the Manila and Ryukyu trenches. Obliquity between theManila trench and the Asian continental margin has led to southward propagation of the collision over the pastc. 5 Ma. This propagation manifests as a progression from submarine accretionary prism building at the Manila trenchsouth of Taiwan, through full-scale arc–continent collision represented by the subaerial mountain belt, to orogen

Figure 1. The catchment of the Liwu River in eastern Taiwan. The Liwu River flanks the Central Range from the main divide(3500 m a.s.l.) to the Pacific Ocean and drains approximately 600 km2 of steep terrain. Where the Liwu River crosses a 6 kmthick sequence of marbles and gneisses, the river forms the deep Taroko gorge. Erosion and incision rates from different studiesand the approximate study locations are indicated.



destruction in response to Ryukyu back-arc extension in the far north of Taiwan (Teng, 1996). Across central Taiwan,metamorphic grade increases from poorly consolidated, Late Tertiary sediments in the Western Foothills thrust belt,through slates in the Hsuehshan and western Central Ranges, to greenschist-grade pre-Tertiary metasediments in theeastern Central Range. The current rate of convergence between the Philippine Sea plate and Asia is 80 mm a−1 (Yuet al., 1997), and rock uplift rates of 5–7 mm a−1 have been calculated from Holocene coastal platforms (Bonilla,1977; Liew et al., 1990). Rapid rock uplift has resulted in the construction of up to 4 km of subaerial relief, but due toits low latitudinal position, Taiwan has never experienced extensive glaciation (Ho, 1988). The main drainage divideof the mountain belt runs parallel to, and c. 25 km west of, the range-bounding Longitudinal Valley fault. Regularlyspaced transverse rivers drain the mountain belt to the east, cross-cutting the structural grain. Westward drainage is bylarger rivers that follow structural trends. Straight slopes, mostly around 35° (less in weak sedimentary rocks in theWestern Foothills) and with thin (<1 m), discontinuous regolith cover, flank the montane valleys (Hovius et al., 2000).Valley floors are in bedrock, mantled by discontinuous, coarse-grained lag deposits. In these cascading channels littlematerial is available for fluvial transport unless provided by hillslope mass wasting (Montgomery and Buffington, 1997).

Taiwan has a subtropical climate with an average of four typhoons per year and mean annual precipitation of2500 mm (Wu and Kuo, 1999). Precipitation is orographically enhanced at high elevation, but, otherwise, is symmetri-cally distributed across the mountain belt. Runoff in the main streams draining the mountain belt reflects the seasonalityof precipitation. The bulk of the water discharge occurs between June and October. During typhoon passage, dailyrainfall rates in the region can top 400 mm, causing peak discharges of >100 times the annual average. Runoffvariability is greatest along the east flank of the Central Range, which is exposed to direct impact of typhoons movingoff the Pacific Ocean, and in the southwest of Taiwan. The pollen record of Sun Moon Lake, central Taiwan, indicatesthat the climate during the Last Glacial Maximum was substantially colder and drier (Kuo and Liew, 2000) (Figure 2).

Liwu catchmentThe Liwu catchment is situated towards the northern end of the compressional Taiwan orogen (Figure 1). Thecatchment flanks the Central Range from the main divide (3500 m a.s.l.) to the Pacific Ocean and drains approxi-mately 600 km2 of steep terrain underlain by metasediments, mainly schists, gneisses and marbles (Figure 3). Meanannual precipitation is around 2200 mm throughout the catchment and precipitation rates have reached up to 600 mm/day during typhoon passage. A palaeoclimate record is not available for the catchment, but it is believed that generalclimate trends are shared with the Sun Moon Lake location (Kuo and Liew, 2000) across the main divide.

Long-term exhumation rates estimated from fission track dating of zircon and apatite are around 4 mm a−1 (Liuet al., 2001; Willett et al., 2003), similar to fluvial incision rates measured directly between February 2000 and May2002 (Hartshorn et al., 2002). Catchment-wide erosion rates derived from river load gauging over the last 30 years inthe Liwu River are 12·5 mm a−1 (Dadson et al., 2003). Radiocarbon-dated strath terraces along the main stream implyhigher average incision rates of ≥6 mm a−1 and ≥11 mm a−1 over the last 2·5 ka (Liew, 1988).

The Liwu River crosses a 6 km thick sequence of marbles and gneisses (Figure 3). These rocks have a compressionalstrength of 45–70 MPa (unpublished industry data: H. Chen, pers. comm. 2004) and their unjointed nature haspermitted the development of a 2 km deep bedrock gorge with very steep (>60°) walls (Figure 4). The flanks ofTaroko gorge contain suites of flutes. These elongate, smooth depressions and ribs are up to >10 m long, and dipgently and flare out in a downstream direction. They are thought to have formed in the active river channel due toimpact abrasion by suspended sediment in turbulent stream eddies which were locked in place by existing channel bedroughness. Evolving flutes are found in outcrops of unjointed rock in the present river channel. They are particularlywell developed on the steep sides of marble reaches where they may be preserved where thalweg lowering has

Figure 2. Precipitation and temperature variations in Taiwan over the last 15 ka. The unpublished data were supplied by P.-M. Liewof the National Taiwan University and based on Liew and Huang (1994) and Kuo and Liew (2000).



brought them above the maximum flood level. In the absence of mass wasting, long vertical suites of such flutes mayform due to progressive fluvial incision of uplifting bedrock. In Taroko gorge we have found fluted bedrock surfacesup to 300 m above the active river channel (a.a.r.c.), and continuous flute sequences up to 200 m a.a.r.c. The activechannel is defined as the lowest part of the bedrock valley which is episodically flooded. Flood marks are present inTaroko gorge up to 15 m above the channel bed.

Vertical suites of fluvial sculptures contain an integral record of fluvial incision, and provide an opportunity to studyincision rates over a timescale not otherwise preserved. We have dated fluvially sculpted facets in the Taroko gorge inorder to obtain a record of incision of the Liwu River. In the following sections we give an account of this work.Subsequently, we evaluate the principal results in the context of other estimates of (fluvial) erosion in the Liwucatchment.

Method

Sample collection and processingWe collected 12 rock samples at regular intervals along a vertical line from a 205 m high buttress at the upstream endof the Taroko marble gorge (Figures 4 and 5). Sample sites were located on shallowly dipping, fluvially sculptedsurfaces with preserved boulder impact divots that formed in the palaeochannel. Above 165 m a.a.r.c. no fluviallysculpted surfaces could be found with impact divots. Instead, these surfaces show signs of carbonate dissolution (e.g.hackly surfaces and dissolution pits) and the flutes are less clearly visible. At each sample location, a slab of marble 2to 6 cm thick (weighing up to 2 kg) was collected from the surface with hammer and chisel. Sample locations weresurveyed with a total station, and present sky-exposure geometry was constrained with a compass-clinometer.

For exposure dating of the fluvially sculpted surfaces in calcareous rocks, whole-rock samples were crushed andsieved. Approximately 100 g of the grain size fraction 0·125–0·25 mm or 0·125–0·5 mm were cleaned with weak nitricacid (c. 0·2 M). Samples were spiked with 4 mg of 35Cl before dissolution. The spike method allows measurement ofthe total Cl concentration with high precision on the same sample material as the 36Cl concentration (Elmore et al.,1997). Chlorine was then extracted using the method described in Stone et al. (1996). The 36Cl/35Cl as well as 37Cl/35Clratios were measured at the accelerator mass spectrometry facility of the Paul-Scherrer Institute and the ETH ofZürich, Switzerland. Details of the extraction and measurement procedures as well as the standards used are given byIvy-Ochs et al. (2004). Major and trace elements were determined by XRAL, Canada. Results are given in Table I.

Calculation of surface exposure agesThe principles of surface exposure dating with cosmogenic nuclides were described by Phillips et al. (1986), Lal(1991), Zreda et al. (1991), and Cerling and Craig (1994). Recent summaries of the technique and its applications aregiven by Zreda and Phillips (2000) and Gosse and Phillips (2001), and are described only briefly here.

Figure 3. Map showing a geological overview of the Taroko gorge, Taiwan, and its lithologies summarized from Chen et al. (2000).The star represents the locality of the wall section sampled for in situ-produced cosmogenic 36Cl analysis.



Figure 4. Overview of the gorge wall selected for sample collection. This gorge section is the uppermost marble cliff of theTaroko gorge. (a) Upper section of the gorge wall. Samples were collected from below and above the overhanging section. Thehighest sample (C10D) is from an altitude of 205 m above the active river channel. (b) Fluvially sculpted marble surfaces inthe lower section of the gorge wall. Sample C6D was collected from an altitude of 3 m above the active river channel. (c) Well-developed flutes in marble in the Liwu catchment, Taiwan. Such surfaces were collected for cosmogneic nuclide analysis andexposure dating.

Figure 5. Diagram giving an overview of the sample locations. Six samples from below and six from above the overhanging wallwere successfully analysed. Samples from higher than 165 m above the active river channel show dissolution marks. The highestsample (C10D) was collected from the flat cliff top.

The Earth’s surface is continuously bombarded by cosmic rays such as neutrons and muons. The interaction of thesecosmic rays with 40Ca, 39K and 35Cl in minerals near the Earth’s surface results in the generation of in situ-produced36Cl. The exposure age of a rock that has been completely shielded until its continuous exposure at the Earth’s surfacewithout further surface erosion is given by:

tC R R

P ln

( )=

−−

−

11 0

λλ

(1)



Table I. Chemical composition of Taroko gorge samples

CaO MgO Sm B Gd Cl rock Th U 36ClCosm.*Sample (wt%) (wt%) (ppm) (ppm) (ppm) (ppm) f35 (ppm) (ppm) (103 atoms (g(rock))−−−−−1)

C6 54·89 1·16 0·05 0·25 0·25 2·72 ± 0·02 5·77E-04 0·2 0·95 2·4 ± 1·0C7D 55·03 0·71 0·05 0·25 0·25 0·78 ± 0·01 1·65E-04 0·2 2·09 13·3 ± 1·2C7 55·28 0·74 0·20 0·25 0·25 0·81 ± 0·81 1·72E-04 0·3 2·35 16·2 ± 1·6C31D 55·86 0·62 0·05 1·50 0·50 0·63 ± 0·01 1·30E-04 0·05 2·38 21·5 ± 4·0C32C 41·50 12·67 0·05 1·50 0·50 2·05 ± 0·02 5·16E-04 0·05 1·19 25·5 ± 2·7C33D 52·34 2·46 0·20 0·25 0·25 2·46 ± 0·02 5·28E-04 0·1 2·19 23·9 ± 2·2C36C 52·35 2·32 0·50 1·50 0·50 1·69 ± 0·02 3·35E-04 0·4 2·87 21·6 ± 3·2C20CD 50·03 4·63 0·05 0·25 0·25 11·47 ± 0·11 2·54E-03 0·2 2·41 51·1 ± 2·2C19D 55·29 0·94 0·05 1·50 0·50 1·85 ± 0·02 3·86E-04 0·05 1·52 60·1 ± 4·6C16D 56·00 0·17 0·05 1·50 0·50 0·36 ± 0·01 7·30E-05 0·05 1·39 106·7 ± 4·1C14D 55·70 0·36 0·05 1·50 0·50 0·71 ± 0·01 1·46E-04 0·05 5 80·0 ± 5·2C12D 55·30 0·47 0·05 0·25 0·25 1·18 ± 0·01 2·50E-04 0·1 1·34 19·7 ± 2·3C10DT 55·82 0·58 0·05 1·50 0·50 0·57 ± 0·01 1·18E-04 0·05 2·52 30·5 ± 2·7C10DB 55·84 0·56 0·05 1·50 0·50 0·76 ± 0·01 1·58E-04 0·05 2·66 29·1 ± 2·8

* In situ-produced cosmogenic 36Cl concentration corrected for blank and production of 36Cl induced by U and Th.

where t is the exposure age [a], λ is the decay constant for 36Cl [2·30 × 10−6 a−1], C is the Cl content in the rock [atomsg(rock)

−1], P is the total cosmogenic production rate of 36Cl [atoms g(rock)−1 a−1], R is the measured 36Cl/Cl ratio corrected

for blank measurements, and R0 is the 36Cl/Cl ratio supported by radiogenic production from U and Th. The totalcosmogenic production rate of 36Cl must be adjusted for the altitude and latitude of the sample, the dip of the sampledsurface, shielding by the surrounding topography, the depth of the sample below the surface, changes of thepalaeomagnetic field variation over time, and the concentration of the target element and the assumed elementalproduction rate. The fact that 36Cl is not only a spallogenic and muonic product, but also produced by thermal andepithermal production out of Cl, complicates the interpretation of the exposure ages. However, in our case of low Clconcentration, the production of Cl is mainly spallogenic and muonic and the exposure age reflects a minimum age(see figure 20 of Gosse and Phillips, 2001).

The altitude and latitude of our samples have been accounted for using Dunai’s procedure (Dunai, 2000), expandedfor reduced intensity of stopped and fast muons following Allkofer (1975). Given that our samples were collectedfrom a gorge wall, the cosmic ray flux is reduced by the steep inclination of the surface and by topographic shielding(shielding factor <1). The correction of the production rate for surface dip and topographic shielding uses the equationof Nishiizumi et al. (1989). However, the shielding geometry of a given sample site has potentially changed as thevalley deepened. In the absence of precise constraints on the palaeovalley geometry, we assume that the valleygeometry has remained constant over time, and that the sampled surface has moved upward with respect to the valleyfloor and into progressively less shielded positions. For each sample site, the present-day shielding factor was calcul-ated. Exposure ages of the sampled surfaces were then calculated using the average of the shielding factors at andbelow the sample site. Changes of cosmogenic nuclide production rate due to palaeomagnetic field variations over thelast 10 ka were within 1 per cent of present-day values (Masarik et al., 2001), and no correction has been made forthis effect. Moreover, irradiation of the sample before exposure at the surface is assumed to have been negligiblysmall compared to post-exposure irradiation.

For the determination of the production of 36Cl, we considered the production by spallation, fast and stopped muonsfrom 40Ca, the thermal production from 35Cl as well as background production of 36Cl by thermal neutrons resultingfrom U and Th reactions. The epithermal production has been neglected, following Mitchell et al. (2001). For theproduction parameters of 36Cl from 40Ca by spallation, stopped and fast muon capture we used the values of 18·6 ± 0·6atoms g(CaCO3)

−1 a−1, 2·59 ± 0·29 atoms g(CaCO3)−1 a−1, and 0·296 ± 0·063 atoms g(CaCO3)

−1 a−1, respectively (Heisingeret al., 2002a, b). These production rates have been corrected for sample thickness using the depth-dependences ofSchaller et al. (2002). The production rates of 36Cl by thermal neutron capture were determined with the equations andparameters as provided in Mitchell et al. (equation A5 in Mitchell et al., 2001). However, for the depth correction ofthe thermal production, we used the attenuation length of neutrons of Schaller et al. (2002) in the first term of equationA5 in Mitchell et al. (2001) and a characteristic length for neutrons in pure calcite of 33 g cm−2 (Liu et al., 1994) inthe second term. Background production of 36Cl by thermal neutrons resulting from U and Th reactions was calculated



using the method of Fabryka-Martin (1988). As the Cl content of the investigated rocks is generally low, the amountof 36Cl produced from U and Th reaction is less than 4 per cent of the total 36Cl concentration.

Analytical uncertainties arise from geochemical parameters such as the concentrations of Cl and Ca and the 36Cl/Clratio. The estimate of the production rate of cosmogenic nuclides at sea level and high latitude is a source ofuncertainty in exposure age determination. Our error estimates are based on the uncertainties as given in Heisingeret al. (2002a, b), The topographic shielding correction on production rate, and particularly its change in time arepoorly constrained and the source of an unquantified error. Our estimate of the total error in exposure ages includesanalytical and blank uncertainties (1σ), uncertainties in scaling factors for altitude and latitude (5%), uncertaintiesin the altitude of the sample, the uncertainties of the production rate due to sample chemistry, and uncertainties inthe production rate at sea level and high latitude. Errors on individual exposure ages are between 0·09 ka and 0·61 ka(see footnote 5b in Table II). For inter-sample comparison, uncertainties in scaling factors for altitude and latitude,altitude of the sample, and the production rate at sea level can be neglected as they are the same for all samples. Thisreduces the range of errors to 0·07 to 0·42 ka (see footnote 5a in Table II). It should be noted that this methodof exposure dating assumes that the dated feature has not been eroded after its formation. In our case of low Clconcentration in the rock, any assumed erosion of fluvial sculptures on the gorge wall after their emergence from theactive channel would result in an increase of the calculated apparent surface age. Thus, the calculated surface ages areminimum ages for the fluvial features targeted by us.

Table II. Locations, surface exposure ages and incision rates of Taroko gorge samples

Surface exposure ages

36ClCosm·(4) Zero 0·05 mm a−1 Incremental Average

Altitude(2) 103 atoms erosion(5) Error(5a) Error(5b) erosion(6) incision rate(7) incision rate(8)

Sample(1) m Shielding(3) (g(rock))−1 ka ka ka ka mm a−1 mm a−1

C6D 3 0·744 2·4 ± 1·0 0·17 0·07 0·09 0·17 ± 0·09 17 ± 14 17 ± 14C7D 33 0·783 13·3 ± 1·2 0·88 0·08 0·10 0·89 ± 0·11 38 ± 7C7 33 0·783 16·2 ± 1·6 1·07 0·10 0·13 1·09 ± 0·14 38 ± 11 31 ± 6C31D 52 0·744 21·5 ± 4·0 1·46 0·27 0·29 1·51 ± 0·31 36 ± 8C32C 55 0·731 25·5 ± 2·7 2·36 0·25 0·31 2·52 ± 0·36 25 ± 16 23 ± 4C33D 58 0·715 23·9 ± 2·2 1·79 0·16 0·22 1·87 ± 0·24 32 ± 5C36C 81 0·724 21·6 ± 3·2 1·55 0·23 0·26 1·61 ± 0·29 52 ± 9C20CD 150 0·745 51·1 ± 2·2 3·53 0·15 0·31 3·85 ± 0·37 34 ± 8 42 ± 4C19D 159 0·516 60·1 ± 4·6 5·55 0·42 0·61 6·77 ± 0·94 5 ± 4 29 ± 3C16D 165 0·768 106·7 ± 4·1 6·48 0·25 0·57 8·27 ± 0·96 7 ± 10 25 ± 2C14D 171 0·763 80·0 ± 5·2 4·88 0·32 0·48 5·79 ± 0·70 35 ± 4C12D 181 0·765 19·7 ± 2·3 1·19 0·14 0·17 1·21 ± 0·18 152 ± 22C10DT 205 0·795 30·5 ± 2·7 1·75 0·15 0·21 1·81 ± 0·23 117 ± 14C10DB 205 0·795 29·1 ± 2·8 1·67 0·16 0·20 1·76 ± 0·22 123 ± 15

(1) The latitude and longitude of the sample site is 23°59 ′ and 121°31′, respectively.(2) The indicated altitude is the altitude above the active river channel, which is 354 m a.s.l., The estimated absolute error in altitude is 5 m, except for

C6D where it is 2 m.(3) The shielding factor for one sample is an average of shielding factors derived from the shielding geometry of all samples situated below the

addressed sample. The correction for shielding by the sample dip and the surrounding gorge walls are based on equations by Nishiizumi et al.(1989).

(4) Cosmogenic 36Cl concentration after subtraction of radiogenic 36Cl. Corrections are <4 per cent. Uncertainties are ± 1σ including blank corrections,10 per cent error on the radiogenic correction, Cl concentration and AMS errors.

(5) Surface exposure ages assuming zero erosion of the surface.(5a) The error includes anlytical and blank uncertainties.(5b) The error includes analytical and blank uncertainties, uncertainties in scaling factors for altitude and latitude (5%), uncertainties in the altitude of the

sample, the uncertainties of the production rate resulting from different sample chemistry and the uncertainty in production rate at sea level (errorfor production by spalllation, fast and stopped muons is taken from Heisinger et al., 2002a, b).

(6) Surface exposure ages assuming an erosion rate of 0·05 mm a−1 of the surface. The error is based on the same uncertainties as stated in note (5b).(7) Incremental incision rate is based on the altitude and age differences between the sample in the row and the next sample lower in altitude.

Exceptions are C7 and C7D where the average age and C31D, C32C, and C33D where the average altitude and age have been used. Errors arebased on uncertainties in surface exposure ages as reported in note (5b) and altitude above the active river channel.

(8) Average incision rate is based on the altitude over the active river channel of the sample and the age of the sample. Errors are based onuncertainties in surface exposure ages as reported in note (5b) and altitude above the active river channel.



Figure 6. In situ-produced cosmogenic 36Cl concentration for samples from different altitudes above the active river channel.Black squares represent the samples prepared with the standard one-step leaching techniques. Two samples, one from below andone from above the overhanging wall, were dissolved in three steps and analysed separately (open squares). The dissolution stepsof the sample from 52 m above the active river channel are within error and not distinguishable. The first dissolution step of thesample from 181 m above the active river channel is higher than the following two dissolution steps.

Cosmogenic nuclide concentrations in a surface sample can also be used for the direct calculation of maximumerosion rates (Lal, 1991), assuming steady state of production and loss of nuclides. In this method, the same assump-tions and corrections apply as described above for surface exposure dating.

Results of the Cosmogenic Nuclide Study

Nuclide concentrations and dissolution experimentsMeasured Cl concentrations in 12 samples from different altitudes in the Taroko gorge are between 0·4 and11·5 ppm. The blank corrected concentrations of in situ-produced 36Cl nuclide are very low, 0·24 × 104 to 10·7 ×104 atoms g(rock)

−1 (Table I and Figure 6). For the lowermost sample (C6D) the influence of the blank correction isdominant and the reliability of the measured 36Cl nuclide concentration low. Reliability is better in higher samples, andthere is good agreement between nuclide concentrations measured independently in separate samples collected fromthe second lowest site (C7 and C7D). A general increase in nuclide concentration is observed from 3 to 165 m abovethe active river channel (a.a.r.c.), but concentrations are constant within error between 50 and 80 m a.a.r.c. In samplescollected above 165 m a.a.r.c. measured nuclide concentrations are lower than the directly underlying samples at 150to 165 m a.a.r.c.

Rock surfaces with low nuclide concentrations, above 165 m a.a.r.c., are characterized by (partial) dissolution offluvially sculpted features such as impact divots. Is post-fluvial erosion the cause of low nuclide concentrations inthese samples or is the rock an open system for Cl nuclides? As flutes are still visible, erosion cannot have been morethan a few centimetres and is therefore thought not to have affected the nuclide concentration significantly (this isconfirmed by our calculations in the next section). Open system behaviour could be the more important factor.Incorporation of meteoric Cl (35Cl, 36Cl and 37Cl) in near-surface rocks would cause the calculated 36Cl concentrationto diverge from the in situ-produced cosmogenic 36Cl concentration (Kubik et al., 1984; Stone et al., 1996). If themeteoric Cl were rich in 36Cl, an exposure age based on the total concentration of 36Cl would be erroneously high.Incorporation of meteoric Cl rich in 35Cl and 37Cl would have an opposite effect on the calculated age. Incorporationof meteoric Cl would elevate the total Cl content of the rock and this would lead to an overestimate of 36Cl producedby radiogenic neutron capture (e.g. from U and Th). In our calculations, the in situ-produced cosmogenic 36Clconcentration would then be overcorrected, resulting in an underestimate of the exposure age. An open system could



also result in loss of in situ-produced cosmogenic 36Cl by preferential leaching (Kubik et al., 1984; Stone et al., 1996),causing reduction of the exposure age estimate. The weathered nature of samples collected above 165 m a.a.r.c.indicates that samples have likely been affected by one of these processes, but it is not clear by which process and towhat extent. These samples are suspect and we assign no importance to their apparent ages.

Next, we need to establish if samples collected from apparently unweathered surfaces below 165 m a.a.r.c. show opensystem behaviour. In order to investigate possible open system behaviour, we have performed a sequential dissolutionof two samples, one from below and one from above 165 m a.a.r.c. Uniform Cl concentrations for all dissolutionsteps would imply a closed system for Cl. Differences in concentrations between dissolution steps would indicateincorporation or loss of Cl. Stone et al. (1996) in a six-stage dissolution experiment observed that only the first stephad a lower 36Cl concentration. They concluded that the slight depletion in the first step is due to natural dissolutionof grain surfaces, but that their standard sample preparation procedure guarantees the integrity of the calcite analysedfor age determination. For our dissolution experiment, we have dissolved samples from 52 m a.a.r.c. (C31D) and181 m a.a.r.c. (C12D) in three steps (Table III). For sample C31D, the 36Cl concentration of the three dissolution stepsis within error the same as the concentration derived from the same sample material treated according to the standardpreparation technique which involves a single dissolution step after cleaning of the sample (called standard one-stepdissolution technique). In contrast, the first dissolution step of sample C12D has a significantly higher 36Cl concen-tration than steps 2 and 3, which are comparable to the concentration measured in the sample prepared with thestandard one-step dissolution technique. The resulting weight-weighted average of the 36Cl concentration is higherthan the concentration determined from the sample prepared with the standard one-step dissolution technique.

These observations have important implications. (1) The standard one-step dissolution technique effectively re-moves any possible contaminating Cl nuclides from the sample. (2) Sample C31D (52 m a.a.c.r.) has not been affectedby incorporation or dissolution of Cl. We conclude that samples without dissolution marks did not suffer from opensystem behaviour, and can be used to determine surface exposure ages. (3) Sample C12D (181 m a.a.r.c.) was not aclosed system and has been affected by incorporation or loss of Cl. The high cosmogenic 36Cl and total Cl concentra-tion measured in the first step may be due to incorporation of meteoric Cl. This would primarily result in dissolutionand recrystallization of calcite along grain boundaries. The Cl composition of the newly formed mineral would beunrelated to that of the replaced material, and might dominate the first dissolution step. If 36Cl nuclides were contrib-uted by a meteoric source and successfully removed during the standard one-step dissolution technique, exposure agescalculated from samples prepared with the standard one-step dissolution technique would reflect the true age of thesurface. Alternatively, the high cosmogenic 36Cl and total Cl concentration in the first dissolution step could be due tochemical weathering compromising the calcite crystal structure. Chemical weathering could be enhanced due to thehigh abundance of pyrite in the marbles of Taroko gorge, and sulphate in surface runoff and groundwater (Yoshimuraet al., 2001). Sulphuric acid produced by pyrite dissociation promotes the dissolution of carbonate, and may facilitateselective leaching of Cl. Exploiting this same effect, we have used nitric acid to remove meteoric Cl from our

Table III. Stepwise dissolution of two samples from Taroko gorge

Cumulative 36ClCsom.

Weight fraction Ca Cl (103 atomsSample loss (g) dissolved (%) (g g−−−−−1 in rock) (ppm in rock) f35 (g(rock))−−−−−1)

Sample 52 m above active river channelC31C-3 * Dissolution 1 45·35 21 not measured 5·44 not measured 23·8 ± 6·0C31C-2 * Dissolution 2 72·37 54 on individual 0·94 on individual 20·5 ± 4·6C31C-1 Dissolution 3 99·67 100 fraction 0·60 fraction 22·8 ± 3·1Composite of dissolution steps 1·72 22·3 ± 4·2C31D Standard technique 118·33 0·401 ± 0·013 0·63 1·30E-04 21·5 ± 4·0

Sample 181 m above active river channelC12D-3 Dissolution 1 51·89 32 not measured 3·66 not measured 35·7 ± 6·1C12D-2 Dissolution 2 44·33 59 on individual 1·32 on individual 15·3 ± 4·1C12D-1 Dissolution 3 66·99 100 fraction 2·87 fraction 19·5 ± 3·0Composite of dissolution steps 2·70 23·5 ± 4·5C12D Standard technique 76·92 0·397 ± 0·013 1·18 2·50E-04 19·7 ± 2·3

* The measured 36Cl/35Cl ratio for these samples is an upper limit. For the calculation of the 36Cl concentration this upper limit and an error of 15 percent have been assumed.



Figure 7. Exposure ages for the fluvially sculpted surfaces collected from different altitudes above the active river channel. Ageneral increase in the exposure age is observed from 3 m to 165 m above the active river channel (filled squares). Surfaces ofsamples collected from above 165 m reveal decreasing exposure ages (open squares). Linear regression through samples collectedfrom below 165 m above the active river channel indicate an incision rate of 26 ± 3 mm a−1.

samples, and to dissolve the rocks. Natural sulphuric acid could mobilize weakly bound Cl from the calcite crystallattice in marble, thus lowering the 36Cl and total Cl concentration in the rock and the associated apparent surfaceexposure age. At this stage, we cannot attribute the differences in 36Cl and Cl concentrations between the initial andsubsequent dissolution steps of sample C12D to one or the other process. We conclude that in our case, samples withdissolution marks should not be used for surface exposure dating, but samples showing no dissolution marks are safeto use.

Exposure ages and incision rates for progressive incisionMeasured cosmogenic 36Cl concentrations have been used to calculate surface exposure ages and incision rates basedon the assumption that the Liwu River has cut the surfaces that we subsequently sampled (Figure 7). We also assumethat these surfaces have never been covered after their formation in the river channel, but may have undergone later(post-fluvial) erosion. Due to the low Cl concentration in our rocks and assuming no post-fluvial erosion as well asinstantaneous abandonment of the active river channel, the calculated exposure age is a minimum age and hence theincision rate is a maximum rate. Two scenarios have been explored in detail: one without post-fluvial erosion andanother with post-fluvial surface retreat at a rate of 0·05 mm a−1 (Table II). This rate is considered to be a high rate oferosion on steep gorge walls. It would lead to the obliteration of most fluvially sculpted gorge wall relief within 10 ka,and was chosen to establish the extent to which surface exposure ages may be affected by post-fluvial erosion. Thecalculations have been done for all samples regardless of the observation of open system behaviour.

For the case without post-fluvial erosion, calculated exposure ages range between 0·2 ka and 6·5 ka (Figure 7).Exposure ages increase to an altitude of 55 m a.a.r.c., but are slightly younger for the next higher samples (58 m and81 m a.a.r.c.). Above the overhanging gorge wall section, where sample collection was not possible, exposure agesincrease up to 165 m a.a.r.c. Samples collected above 165 m a.a.r.c. reveal young exposure ages. Exposure agescalculated for a post-fluvial erosion rate of 0·05 mm a−1 are higher, between 0·9 ka and 8·3 ka, and the differencebetween the two age estimates increases with the age of the surface to >30 per cent for sample C16D. We considerthese amended surface ages to be at or above the upper limit for samples collected from clearly fluted surfaces withimpact divots (all samples below or at 165 m a.a.r.c.). The correction for post-fluvial erosion is more appropriate forsample locations above 165 m a.a.r.c. (C14D, C12D and C10D), but it does not increase the ages of these surfacesbeyond 6 ka. Given that the minimum exposure age of the next lower sample is 6·5 ka, progressive, post-fluvialsurface lowering at a rate of up to 0·05 mm a−1 cannot be the sole cause of low 36Cl concentrations in gorge wallsamples collected above 165 m a.a.r.c. Only sample C10D, which was collected from the karstic top of the buttress,



may have been affected by faster post-fluvial erosion. Assuming steady-state conditions, a mean post-fluvial erosionrate of 0·9 mm a−1 is implied by the cosmogenic nuclide concentration measured in this sample. Low nuclide concen-trations in samples C12D and C14D are likely due to a combination of post-fluvial surface retreat and open systembehaviour for Cl.

Assuming steady-state erosion in the active river channel, maximum surface lowering at site C6D has proceeded at9 ± 5 mm a−1 in the last 130 years. Above the active channel, maximum average incision rates were calculated usingthe sample altitude with respect to the river channel and the local surface exposure age (Table II). Incremental incisionrates were derived from the difference in altitude and exposure age of two successive samples (Table II). In thesecalculations we have used surface exposure ages without corrections for post-fluvial erosion; the incision rates are thusmaximum values. According to our calculations, average fluvial incision rates were 26 ± 3 mm a−1 over the last 6·5 ka.During this interval, rates have varied between 17 mm a−1 and 52 mm a−1. In the last millennium incision occurred at38 ± 11 mm a−1. Between 2 ka and 1 ka (samples C31D, C32C and C33D) the incremental incision rate was25 ± 16 mm a−1. Using samples from directly below and above the overhanging wall (C36C and C20CD, respectively),the incision rate was 34 ± 8 mm a−1 between 3·5 ka and 1·5 ka. However, using the average altitudes and exposureages of all four samples collected just below the overhang (C31D, C32C, C33D and C36C) yields a higher incisionrate of 51 ± 15 mm a−1. Samples above the overhang have yielded incremental incision rates of 5–7 mm a−1 for theperiod 6·5 ka to 3·5 ka. Post-fluvial erosion of 0·05 mm a−1 reduces the average fluvial incision rate to 20 ± 3 mm a−1:it does not have a major effect on average incision rates. Taking into account that the gorge walls were not abandonedinstantaneously, but instead emerged slowly from the active river channel does not affect the average incision rate.Lacking constraints on palaeoflow depths, we have to assume that all surfaces were abandoned at the same altitudeabove the floor of the active river channel and at the same time after the first surface exposure. This assumption affectsall exposure ages and estimates of incision in the same way. Our incision rate estimates are subject to furtheruncertainty associated with the history of gorge formation which will be explored in the next section.

Effects of aggradation and re-incisionThe lowering of Taroko gorge is likely to have been punctuated by phases of flooding and aggradation associated withmassive gorge wall failure and damming of the river (cf. Pratt-Sitaula et al., 2004). During subsequent re-incision, theriver could cut through the landslide dam and fully remove the associated valley fill; it could relocate to the bedrock–fill interface, causing lateral gorge wall retreat during removal of the blockage; or it could shift into a nearbydepression and cut a new bedrock channel. Here we consider cases of valley fill and re-incision with and withoutlateral gorge wall retreat. Remnants of massive valley fills have been observed in the Liwu catchment (Hsieh et al.,2003; Liew, 1988). One such remnant is located upstream of our sample section. Its base is 40 m a.a.r.c., its top abovethe buttress from which we have collected samples, and an age of 2·5 ka has been proposed (M. L. Hsieh, pers.comm.; Hsieh et al., 2003; Liew, 1988). We have used these observations to constrain some alternative, hypotheticalincision scenarios. These scenarios are presented here to illustrate how valley fill and re-incision may have affectedapparent exposure ages and incision rates calculated from measured cosmogenic nuclide concentrations.

For simplicity we have only explored scenarios with one phase of aggradation, preceded and followed by progres-sive fluvial incision. The following assumptions have been made: (1) prior to aggradation, the valley floor was at 40 mabove the present-day channel floor; (2) the history of topographic shielding before aggradation was identical to theone used in age and incision rate calculations in the previous section; (3) instantaneous aggradation occurred at 2·5 ka;(4) the valley fill covered the entire buttress and had a density of 2·0 g cm−3; (5) the fill was instantaneously removedat 1·25 ka; (6) during removal of the valley fill, the gorge wall retreated by a given amount (here between 0 and100 cm); (7) subsequent to removal of the valley fill, the river has incised the remaining 40 m to its present level;(8) during this phase of incision the topographic shielding was as it is today.

Measured nuclide concentrations have been taken to reflect pre-aggradation accumulation, shielding during valleyfill, possible removal of rock from the gorge wall during re-incision, and post-aggradation accumulation, and sampleages and incision rates have been recalculated (Figure 8). The scenario without lateral gorge wall retreat yields onlyslightly higher exposure ages and an average rate of valley lowering for all samples collected at or below 165 m a.a.r.c.of 23 ± 3 mm a−1, compared with 26 ± 3 mm a−1 for the case without temporary valley fill. Lateral gorge wall retreatgives rise to older apparent ages of the original surfaces. For the case of 100 cm of lateral retreat, initial surface agesare up to three times greater than for the case without lateral retreat, and the corresponding average valley loweringrate is 7 ± 1 mm a−1. Removal of 25 cm of rock from the gorge wall during re-incision would result in an averagevalley lowering rate of 17 ± 2 mm a−1 over the last 9·9 ka. We conclude that taking temporary valley fill into accountis unlikely to have a major effect on the calculated apparent exposure ages of the sampled surfaces, but that assuminglateral gorge wall retreat during removal of the fill may cause a significant increase of these calculated exposure ages.



Several kilometre-scale landslide scars with steep, rough surfaces are present in the Taroko gorge. Some scars reachdown to within a few tens of metres from the present-day channel bed, but the Liwu River is currently not obstructedby landslide debris anywhere in the gorge. We infer that the river may have been blocked within the past fewmillennia, and that removal of the landslide dam(s) has occurred relatively quickly. Small blockages, with a fill depthof a few tens of metres, may have occurred more frequently. This leaves room for more complex gorge incisionhistories, with vertically non-uniform gorge wall retreat. Localized lateral wear of several metres during re-incisionepisodes cannot be excluded, and it is also possible that nearby surfaces remained untouched due to alluvial cover.This mechanism would offer a convenient explanation for the anomalously young surface exposure ages found insamples C33D and C36C, but in the absence of firm, independent constraints on the aggradation and re-incisionhistory of Taroko gorge such interpretations remain speculative. In extremis, it is conceivable that the different surfaceexposure ages of our samples are the result of differential gorge wall retreat during re-incision superimposed onconstant bedrock channel lowering. However, we do not believe that this is likely: progressive younging of surfaceexposures down the gorge, with the aforementioned exceptions, suggests that lateral gorge wall retreat during re-incision has played a secondary role in the formation of Taroko gorge, although it may have suppressed someexposure ages. In the following discussion we take the cosmogenic rates calculated for simple scenarios of gorgeevolution at face value, mindful of the possible complications introduced by lateral river cutting during re-incision oftemporary valley fills.

Discussion

Incision and erosion rates have been determined in the Liwu catchment by different methods, including direct meas-urement of fluvial wear, suspended sediment gauging, terrace dating and surveying, fission track thermochronometryand cosmogenic nuclide analysis (Figure 9). We briefly review the key results of these approaches in order ofincreasing time coverage, noting that the rates are location- and process-specific.

Direct measurements of river wear. At a hydrometric station 3 km upstream of our sample location, Hartshorn et al.(2002) measured incremental changes of channel bed topography along traverses across the active river channel witha precision of about 0·5 mm. Between February 2000 and December 2001, the spatially averaged wear rate in schists,the dominant lithology along this reach, was 3·4 mm a−1, with maximal local erosion of 69 mm. A wide range ofdischarges was recorded during this interval, including a 25-year return time flood. Unpublished fluvial bedrockincision measurements from a location at the base of the sampled marble buttress show similar wear rates.

Figure 8. The assumption of different incision histories for the gorge formation results in variable incision rates. One phaseof aggradation starting at 2·5 ka and lasting for 1·25 ka is assumed. While removing the sediment, the river may refresh thefluvially sculpted surfaces. The amount of gorge wall retreat, e.g. 0 cm (grey squares), 25 cm (open squares), 50 cm (grey triangles),and 100 cm (open triangles), influences strongly the calculated ages and hence the incision rates (see text for further explanationand discussion).



Erosion rates derived from river load gauging. At the same hydrometric station, water discharge has been meas-ured daily and suspended sediment concentrations at an average frequency of 24 samples per year since 1970 (WaterResources Agency, 1970–2003). Using a bias-corrected average of the reported measurements, Dadson et al. (2003)obtained an average annual suspended sediment load of 14·4 Mt derived from an area of 435 km2 upstream of thegauging station. This is equivalent to a catchment-wide average surface lowering rate of 12·5 mm a−1 (using thedensity of quartz). A 25 per cent uncertainty applies to this estimate. Between 1970 and 2000, about 70 per cent of allsuspended sediment was transported by floods with a recurrence interval of one year or more, and 50 per cent of thesuspended load was carried by floods with a frequency of five years or more.

Incision rates derived from terraces. Liew (1988) reported on river terraces and perched palaeochannels along theLiwu River, upstream of Taroko gorge. The ages of two terrace deposits have been dated with 14C to 2·4 ka and 2·5 ka,respectively (Liew, 1988). Using the height of the bedrock straths of these features above the current channel, Liew(1988) calculated river incision rates of 6 mm a−1 and 11 mm a−1. These rates are considered to be minimum incisionrates because in some cases the river has shifted course after aggradation and incised a new bedrock channel adjacentto the filled passage. Accordingly, the thickness of the fill should be added to incision rate calculations.

Cosmogenic nuclide-derived incision rates. Our cosmogenic nuclide work has shown that channel lowering at theupstream end of Taroko gorge has occurred at a maximum average rate of 9 mm a−1 over the last 130 years. Assumingprogressive channel lowering without temporary aggradation, we have calculated a maximum average rate of riverincision of 26 ± 3 mm a−1 since 6·5 ka. Incision rates were higher in the last 2·5 to 3·5 ka, and only 5–7 mm a−1 duringthe preceding interval. However, these estimates are sensitive to post-fluvial erosion and lateral gorge wall retreatduring re-incision of temporary valley fills. Rates may have been lower as a result, but we believe that the mid–lateHolocene acceleration of river incision is a robust feature of our data.

Fission track exhumation rates. Apatite and zircon fission track ages provide cooling rates from temperatures ofabout 100 °C and 240 °C, respectively. Assuming that this cooling rate reflects conductive cooling of samples as theyare exhumed to the surface by erosion, ages translate directly into erosion rates. Liu et al. (2001) reported a zirconfission track age of 2·3 ± 1·6 Ma from a gneiss in the downstream reach of the Taroko gorge. Apatite from the sameunit yields a fission track age of 0·9 ± 0·2 Ma (Willett et al., 2003). Interpreted through a one-dimensional model forcooling and exhumation, these ages provide an internally consistent estimate of erosion rate of 3 to 5 mm a−1 over thelast 2 Ma.

The following points emerge from this summary. (1) Average Holocene river incision rates in Taroko gorge andnearby terrace locations have been higher than the Quaternary exhumation rate of the Liwu catchment by at least afactor two and possibly much more. (2) Millennial incision rates may have varied through the Holocene, and lateHolocene river incision was fast compared to the preceding interval. (3) Modern fluvial incision rates at and upstream

Figure 9. Compilation of incision and erosion rates over different timescales from the Taroko gorge area. Cosmogenic nuclide-derived rates (this study), rates from dated fill terraces (Liew, 1988), and bedrock lowering rates (Hartshorn et al., 2002) areincision rates. Rates derived from river load gauging (Dadson et al., 2003) and rates from fission track dating (Liu et al., 2001;Willett et al., 2003) reflect erosion rates. The Holocene incision rates are higher than the long-term erosion rate determinedby fission tack dating.



of the top end of Taroko gorge are lower than the Holocene average incision rate, and similar to the Quaternaryexhumation rate. (4) Decadal catchment-wide erosion rates upstream of Taroko gorge are faster than modern riverincision rates and long-term exhumation rates.

Temporal variability of fluvial incision and catchment lowering may have been caused by trends in rock uplift, sealevel and/or climate. Few independent constraints exist on the rock uplift history of east Taiwan, but given thatregional plate motion vectors have been relatively constant during the Quaternary (Seno et al., 1993), we expect thataverage rock uplift rates have not changed importantly in this interval. The uniformity of zircon and apatite fissiontrack exhumation rates in the Liwu catchment supports this interpretation. On shorter timescales, very large earth-quakes (>Mw7·5) may have caused rapid rock and surface uplift and sediment production. Measured Holocene incisionwould offset the cumulative effect of many earthquake cycles, but it is conceivable that routing of coseismicallyproduced sediment (Dadson et al., 2004) has promoted rapid fluvial incision. Temporal clustering of very largeearthquakes might explain elevated Holocene river incision and catchment erosion rates. Lack of independentevidence bars the validation of this hypothesis.

More information is available for sea-level and climate change. If sea-level rise since the Last Glacial Maximumhas had an effect on the incision of Taroko gorge and catchment lowering further upstream (cf. Merritts et al., 1994),it should have been to suppress these processes. All available erosion data point in the opposite direction, ruling outglacial–interglacial sea-level change and smaller Holocene fluctuations (Chen and Liu, 1996) as important controls onthe incision of the Liwu River and erosion of its catchment.

A more likely cause of high Holocene erosion rates is climate forcing. Pollen records from Sun Moon Lake, centralTaiwan, indicate that conditions changed from cool and relatively dry to warm and wet at c. 11 ka when the southeastmonsoon reached Taiwan (Kuo and Liew, 2000). Although the general shift to wetter, warmer conditions during theHolocene is likely to have enhanced fluvial incision and catchment erosion, the special role of typhoons is evident, forexample, from the disproportionate importance of infrequent, large floods in transporting sediment in the Liwucatchment and elsewhere in Taiwan (Dadson et al., 2003). Taiwan occupies a central position within the northern westPacific typhoon belt. Currently this belt spans the Pacific Rim from the south Philippines to central Japan, with highesttyphoon frequencies in the north Philippines (Joint Typhoon Warning Center, 1972–2001). Latitudinal shifts of thetyphoon belt would cause changes of typhoon incidence frequency and strength in Taiwan. Such changes may haveoccurred during the Holocene, and certainly since the Last Glacial Maximum. We attribute elevated Holocene rates ofriver incision and landscape lowering in the Liwu catchment, measured by three independent methods, to a combina-tion of warm, wet average conditions and frequent, extreme precipitation and runoff during typhoon impact. Observa-tions of elevated Holocene erosion rates in the Liwu catchment match a two- to three-fold increase of post-glacialsedimentation rates in Sun Moon Lake since the Last Glacial Maximum.

If high Holocene erosion rates are representative of most Quaternary interglacials then it follows that erosion duringglacial intervals has been, on average, slower than the Quaternary average exhumation rate of 4 mm a−1 estimatedfrom fission track data. We note that the specific case of Taiwan is in conflict with Molnar’s hypothesis (Molnar, 2001)that high Quaternary erosion rates are due mainly to increased runoff variability during cold stages. Other uplandregions exposed to frequent typhoon impact may have a similar emphasis on interglacial, storm-driven erosion.

The case for variability of erosion rates during the Holocene is less clear. Terrace-derived incision rates, recentchannel lowering estimated from cosmogenic nuclide concentrations, and decadal suspended sediment erosion ratesare similar at around 10 mm a−1. Relatively low annual incision rates measured by Hartshorn et al. (2002) may notreflect the full range of water discharge and sediment transport in the Liwu River, making it difficult to extrapolatethese measurements to longer timescales. Our cosmogenic nuclide work may show an acceleration of river incisionrates from 5–7 mm a−1 prior to 3·5 ka to much higher rates since then. Variations of temperature and precipitation haveoccurred in Taiwan during the Holocene (Liew and Huang, 1994). The pollen record shows warm, wet conditionsbetween 11 ka and 5 ka, with increasing strength of the summer monsoon during this interval, and cooler, dryerconditions since then (Figure 2). Sedimentation rates in Sun Moon Lake track with these climate changes, peakingduring the mid-Holocene climate optimum. Thus, there is a discrepancy between the timing and sense of change oferosion rates in Taroko gorge and the Sun Moon hinterland, suggesting that climate change may not have been theprincipal cause of accelerated late Holocene erosion in Taroko gorge. An additional possible mechanism which wemention, but cannot easily test, is the possibility of temporal clustering of very large earthquakes.

Finally, we return briefly to the coupling between hillslopes and river channels mentioned at the outset. In contrastto river incision rates, erosion rates derived from river load gauging reflect the lowering of an entire catchment. Ifrivers and hillslopes are perfectly and directly coupled, then the two rates should be the same. Differences betweenriver incision rates and catchment erosion rates represent a disequilibrium in the landscape. It has been proposed thatmaximum hillslope erosion is likely to lag behind maximum river incision (Hartshorn et al., 2002; Pratt-Sitaula et al.,2004). Although the decadal, catchment-wide erosion rate upstream of Taroko gorge is higher than estimates of recent



fluvial incision in and upstream of the gorge, we have no firm, quantitative evidence to corroborate or reject thishypothesis for the case of the Liwu River. However, the presence of extensive, fluvially sculpted surfaces along thevery narrow and steep inner section of the Taroko gorge suggests that the marble valley sides do not in generalrespond immediately to river incision, and that a time lag of up to 104 years is characteristic for this landscape.

Conclusions

Taroko gorge in the active Taiwan orogen contains a rare, continuous record of Holocene river incision. Measuredconcentrations of in situ-produced 36Cl in fluvially sculpted facets of the marble gorge wall imply progressive incisionof the gorge at a maximum average rate of 26 ± 3 mm a−1 since 6·5 ka. Incorporation of meteoric Cl and/or preferentialleaching of 36Cl from near-surface rocks, probably enhanced by the abundance of pyrite in the rock mass, hascompromised the cosmogenic nuclide record of river incision at higher levels within the gorge, and limited ourwindow of observation to the Holocene. Moreover, temporary fill of the gorge due to landslide damming, and lateralgorge wall retreat during re-incision may have shielded the fluvial facets from cosmic ray impact and permittedremoval of irradiated rock. This may have created an out-of-sequence pattern in the apparent surface ages, and mayresult in a decrease of calculated river incision rates. However, when combined with other estimates of erosion in theLiwu catchment, our data are firm evidence for fast Holocene incision with respect to the Quaternary average exhuma-tion rate. The warm, wet climate of Taiwan, which is punctuated by frequent impacts of large typhoons, is the likelycause of elevated interglacial incision rates. We have also found indications of incision rate changes within theHolocene, but cannot explain them in the context of available climate data.

AcknowledgementsWe thank the Taroko National Park authority for permission to sample in the park and for generous logistic support, Hsieh Meng-Long for sharing his knowledge of river terraces with us, and Lin Jiun-Chuan and Hongey Chen for providing a base in Taipei. Weare indebted to Liew Ping-Mei for making available unpublished data on the Late Quaternary climate of Taiwan. The manuscript hasbeen improved by the comments of N. P. Snyder and W. B. Dade. Support of the Zürich AMS crew is greatly appreciated. TheZürich AMS facility is jointly operated by the Swiss Federal Institute of Technology, Zürich, and Paul Scherrer Institute, Villigen,Switzerland. This study was supported by a grant of the Schweizerische Nationalfonds to M.S., and grants from the Newton Trustand the Leverhulme Trust to N.H.

References

Allkofer OC. 1975. Introduction to Cosmic Rays. München: Verlag Karl Thiemig.Amorosi A, Farina M, Severi P, Preti D, Caporale L, Di Dio G. 1996. Genetically related alluvial deposits across active fault zones: an

example of alluvial fan-terrace correlation from the upper Quaternary of the southern Po Basin, Italy. Sedimentary Geology 102: 275–295.Beaumont C, Fullsack P, Hamilton J. 1992. Erosional control of active compressional orogens. In Thrust Tectonics, McClay KR (ed.).

Chapman and Hall: London.Bonilla MG. 1977. Summary of Quaternary faulting and elevation changes in Taiwan. Geological Society of China Memoir 2: 43–55.Burbank DW, Leland J, Fielding E, Anderson RS, Brozovic N, et al. 1996. Bedrock incision, rock uplift and threshold hillslopes in the

northwestern Himalayas. Nature 379: 505–510.Cerling TE, Craig H. 1994. Geomorphology and in-situ cosmogenic isotopes. Annual Review of Earth Planetary Science 22: 273–317.Chen Y-G, Liu T-K. 1996. Sea level changes in the last several thousand years, Penghu Islands, Taiwan Strait. Quaternary Research 45:

254–262.Chen CH, Ho H-C, Shea K-S, Lo W, Lin W-H, et al. 2000. Geologic map of Taiwan. Central Geological Survey, Ministry of Economic

Affairs: Taiwan.Clift P, Gaedicke C. 2002. Accelerated mass flux to the Arabian Sea during the middle to late Miocene. Geology 30: 207–210.Dadson SJ, Hovius N, Chen H, Dade WB, Hsieh ML, et al. 2003. Links between erosion, runoff variablity and seismicity in the Taiwan

orogen. Nature 426: 648.Dadson SJ, Hovius N, Chen H, Dade WB, Lin JC, et al. 2004. Earthquake-triggered increase in sediment delivery from an active mountain

belt. Geology 32: 733–736.Dunai TJ. 2000. Scaling factors for production rates of in situ produced cosmogenic nuclides: a critical reevaluation. Earth and Planetary

Science Letters 176: 157–169.Elmore D, Ma X, Miller T, Mueller K, Perry M, et al. 1997. Status and plans for the PRIME Lab AMS facility. Nuclear Instruments and

Methods in Physics Research Section B 123: 69–72.Fabryka-Martin JT. 1988. Production of radionuclides in the Earth and their hydrogeologic significance, with emphasison chlorine-36 and

iodine-129. PhD thesis, University of Arizona, Tucson.



France-Lanord C, Derry LA. 1997. Organic carbon burial forcing of the carbon cycle from Himalayan erosion. Nature 390: 65–67.Gaillardet J, Dupré B, Louvat P, Allègre CJ. 1999. Global silicate weathering and CO2 consumption rates deduced from the chemistry of

large rivers. Chemical Geology 159: 3–30.Gosse JC, Phillips FM. 2001. Terrestrial in situ cosmogenic nculides: theory and application. Quaternary Science Review 20: 1475–1560.Hartshorn K, Hovius N, Dade WB, Slingerland R. 2002. Climate-driven bedrock incision in an acitve mountain belt. Science 297: 2036–

2038.Heisinger B, Lal D, Jull AJT, Kubik PW, Ivy-Ochs S, et al. 2002a. Production of selected cosmogenic radionuclides by muons: 1. Fast

muons. Earth and Planetary Science Letters 200: 345–355.Heisinger B, Lal D, Jull AJT, Kubik PW, Ivy-Ochs S, et al. 2002b. Production of selected cosmogenic radionuclides by muons: 2. Capture

of negative muons. Earth and Planetary Science Letters 200: 357–269.Ho CS. 1988. An Introduction to the Geology of Taiwan: Explanatory Text of the Geological Map of Taiwan (2nd edn). Central Geological

Survey, Ministry of Economic Affairs: Taipei.Hovius N, Stark CP, Chu H-T, Lin J-C. 2000. Supply and removal of sediment in a landslide-dominated mountain belt; Central Range,

Taiwan. Journal of Geology 108: 73–89.Howard AD, Kerby G. 1983. Channel changes in badlands. Geological Society of America Bulletin 94: 739–752.Hsieh ML, Liew PM, Chen Y. 2003. Control of Bedload Sediment Supply Upon Bedrock Incision and River Longitudinal Profile in the

Eastern Central Range, Taiwan. Eos Trans. AGU 84(46), Fall Meet. Suppl., Abstract T21C-0465.Ivy-Ochs S, Synal H-A, Roth C, Schaller M. 2004. Inital results from isotope dilution for Cl and 36Cl measurements at the PSI/ETH Zurich

AMS facility. Nuclear Instruments and Methods in Physics Research B 223–224: 623–627.Joint Typhoon Warning System. 1972–2001. Yearly TC Occurrence for all Months. http://www.npmoc.navy.mil/jtwc/climostats/trackclimo/

allaor.htmlKoons PO. 1989. The topographic evolution of collisional mountain belts; a numerical look at the Southern Alps, New Zealand. American

Journal of Science 289: 1041–1069.Kubik PW, Korschinek G, Nolte E, Ratzinger U, Enrst H, et al. 1984. Accelerator mass spectrometry of 36Cl in limestone and some

paleontological samples using completely stripped ions. Nuclear Instruments and Methods in Physics Research B5: 326–330.Kuo CM, Liew PM. 2000. Vegetational history and climatic fluctuations based on pollen analysis of the Toushe peat bog, central Taiwan

since the Last Glacial Maximum. Journal of the Geological Society of China 43: 379–392.Lal D. 1991. Cosmic ray labeling of erosion surfaces: in situ nuclide production rates and erosion models. Earth and Planetary Science

Letters 104: 424–439.Lavé J, Avouac JP. 2000. Active folding of fluvial terraces across the Siwaliks Hills, Himalayas of central Nepal. Journal of Geophysical

Research, B, Solid Earth and Planets 105: 5735–5770.Liew PM. 1988. Sedimentology and River-terrace Correlation of the Liwu River. National Taiwan University: Taipei (in Chinese).Liew PM, Hsieh ML, Lai CK. 1990. Tectonic significance of Holocene marine terraces in the Coastal Range, eastern Taiwan. Tectonophysics

183: 121–127.Liew PM, Huang SY. 1994. A 5000-year pollen record from Chitsai lake, central Taiwan. Terrestrial, Atmosphere and Oceanic Sciences 5:

411–419.Liu B, Phillips F, Fabryka-Martin J, Fowler M, Stone W. 1994. Cosmogenic (super 36) Cl accumulation in unstable landforms; 1, Effects of

the thermal neutron distribution. Water Resources Research 30: 3115–3125.Liu T-K, Hsieh S, Chen Y-G, Chen W-S. 2001. Thermo-kinematic evolution of the Taiwan oblique-collision mountain belt as revealed by

zircon fission track dating. Earth and Planetary Science Letters 186: 45–56.Masarik J, Frank M, Schäfer JM, Wieler R. 2001. Correction of in situ cosmogenic nuclide production rates for geomagnetic field intensity

variations during the past 800,000 years. Geochimica et Cosmochimica Acta 65: 2995–3003.Merritts DJ, Vincent KR, Wohl EE. 1994. Long river profiles, tectonism, and eustasy; a guide to interpreting fluvial terraces. Journal of

Geophysical Research, B, Solid Earth and Planets 99: 14031–14050.Mitchell S, Matmon A, Bierman P, Enzel Y, Caffee M, Rizzo D. 2001. Displacement history of a limestone normal fault scarp, northern

Israel, from cosmogenic (super 36) Cl. Journal of Geophysical Research 106: 4247–4264.Molnar P. 2001. Climate change, flooding in arid environments, and erosion rates. Geology 29: 1071–1074.Montgomery DR, Buffington JM. 1997. Channel-reach morphology in moutain drainage basins. Geological Society of America Bulletin 109:

596–611.Nishiizumi K, Winterer EL, Kohl CP, Klein J. 1989. Cosmic Ray Production Rates of 10Be and 26Al in Quartz From Glacially Polished

Rocks. Journal of Geophysical Research 94: 17907–17915.Phillips F, Leavy B, Jannik N, Elmore D, Kubik P. 1986. The accumulation of cosmogenic chlorine-36 in rocks; a method for surface

exposure dating. Science 231: 41–43.Pratt B, Burbank DW, Heimsath A, Ojha T. 2002. Impulsive alluviation during early Holocene strenghtened monsoons, central Nepal

Himalaya. Geology 30: 911–914.Pratt-Sitaula BA, Burbank DW, Heimsath AM, Ojha T. 2004. Landscape disequilibrium on 1000–10 000 year scales, Marsyandi River,

Nepal, central Himalaya. Geomorphology 58: 223–241.Raymo ME, Ruddiman WF. 1992. Tectonic forcing of Late Cenozoic climate. Nature 359: 117–122.Schaller M, von Blanckenburg F, Veldkamp A, Tebbens LA, Hovius N, Kubik PW. 2002. A 30 000 yr record of erosion rates from

cosmogenic 10Be in Middle European river terraces. Earth and Planetary Science Letters 204: 307–320.Seno T, Stein S, Grip AE. 1993. A model for the motion of the Philippine Sea plate consistent with NUVEL-1 and geologic data. Journal of

Geophysical Research 98: 17941–17948.

http://www.npmoc.navy.mil/jtwc/climostats/trackclimo/



Snyder NP, Whipple KX, Tucker GE, Merritts DJ. 2000. Landscape response to tectonic forcing; digital elevation model analysis of streamprofiles in the Mendocino triple junction region, Northern California. Geological Society of America Bulletin 112: 1250–1263.

Snyder NP, Whipple KX, Tucker GE, Merritts DJ. 2003. Channel response to tectonic forcing; field analysis of stream morphology andhydrology in the Mendocino triple junction region, Northern California. Geomorphology 53: 97–127.

Stone JO, Allan GL, Fifield LK, Cresswell RG. 1996. Cosmogenic chlorine-36 from calcium spallation. Geochimica et Cosmochimica Acta60: 679–692.

Teng LS. 1990. Geotectonic evolution of late Cenozoic arc-continent collision in Taiwan. Tectonophysics 183: 57–76.Teng LS. 1996. Extensional collapse of the northern Taiwan mountain belt. Geology 24: 949–952.Tinkler K, Wohl EE. 1998. Rivers over Rock: Fluvial Processes in Bedrock Channels. Geophysical Monograph Series 107: 323pp.Tucker GE. 2004. Drainage basin sensitivity to tectonic and climatic forcing; implications of a stochastic model for the role of entrainment

and erosion thresholds. Earth Surface Processes and Landforms 29: 185–205.Tucker GE, Bras RL. 2000. A stochastic approach to modeling the role of rainfall variability in drainage basin evolution. Water Resources

Research 36: 1953–1964.Water Recources Agency, 1970–2003. Hydrological Yearbook of Taiwan, Republic of China. Ministry of Economic Affairs: Taipei.Whipple KX, Kirby E, Brockelhurst SH. 1999. Geomorphic limits to climate induced increases in topographic relief. Nature 401: 39–43.Willett SD. 1999. Orogeny and orography, the effects of erosion on the structure of mountain belts. Journal of Geophysical Research, B,

Solid Earth and Planets 104: 28957–28982.Willett SD, Fisher D, Fuller CW, Yeh EC, Lu CY. 2003. Erosion rates and orogenic wedge kinematics in Taiwan inferred from apatite fission

track thermochronometry. Geology 31: 945–948.Wu CC, Kuo YH. 1999. Typhoons affecting Taiwan: Current understanding and future challenges. Bulletin of the American Meteorological

Society 80: 67–80.Yoshimura K, Nakao S, Noto M, Inokura Y, Urata K, et al. 2001. Geochemical and stable isotope studies on natural water in the Taroko

Gorge karst area, Taiwan-chemical weathering of carbonate rocks by deep source CO2 and sulfuric acid. Chemical Geology 177: 415–430.

Yu SB, Chen HY, Kuo LC. 1997. Velocity field of GPS stations in the Tiawan area. Tectonophysics 274: 41–59.Zreda MG, Phillips FM. 2000. Cosmogenic nuclide build-up in surface material. In Quaternary Geochronology: Methods and Applications,

Noller JS, Sowers JM, Lettis WR (eds). American Geophysical Union: Washington DC; 61–76.Zreda M, Phillips F, Elmore D, Kubik P, Sharma P, Dorn R. 1991. Cosmogenic chlorine-36 production rates in terrestrial rocks. Earth and

Planetary Science Letters 105: 94–109.

Earth Surface Processes and Landforms Fluvial bedrock ... et al... · Earth Surf. Process. Landforms30, 955–971 (2005) Published online in Wiley InterScience (). DOI: 10.1002/esp.1256

Documents