The Upper Limit of Denudation Rate Measurement From Cosmogenic 10Be(Meteoric)/9Be Ratios in Taiwan

The tectonically active Taiwan orogen features numerous rivers that yield a high amount of sediment with fluxes exceeding 104 t/km2/yr. Amongst these, the landslide‐dominated Liwu River is well studied regarding its dynamic surface processes. However, the quantification of denudation in the Liwu Basin is still an ongoing task as rates obtained to date are subject to substantial differences depending on methods that differ in their spatio‐temporal scales. We constrain an upper limit of global denudation using the cosmogenic nuclide 10Be(meteoric) and its ratio to stable 9Be. Meteoric cosmogenic 10Be is delivered to Earth’s surface by precipitation, whereas stable 9Be is released from rock weathering. In contrast to in situ cosmogenic 10Be measured in quartz, the 10Be(meteoric)/9Be ratio can be analyzed in quartz‐poor settings. 10Be(meteoric)/9Be‐derived denudation rates (Dmet) vary from 8.1 to >30 mm/yr in the Liwu mainstem, and from 3.4 to 21.5 mm/yr in the tributaries. These new Dmet are among the highest cosmogenic nuclide‐derived rates ever measured. Most of these rates agree with rates from sediment gauging or channel incision. We propose that stochastic landsliding plays a major role in denudation processes here. Using a soil‐bedrock mixing model and published riverine organic 14C data as a soil tracer, we estimate the fractional contribution of bedrock landslide material to mainstem sediments to be 55%–97%, which explains the magnitude and large variability (4‐fold) in Dmet. We demonstrate the complexity associated with denudation rates determination in landslide‐dominated routing systems; but also the potential of 10Be(meteoric)/9Be for tracing stochastic landsliding processes.

Be(meteoric) and its ratio to 9 Be in one of the world's fastest-eroding basin • 10 Be(meteoric)/ 9 Be ratios record extremely high denudation rates (>10 mm/yr) • A soil-bedrock mixing model reveals that bedrock landslides are the cause of the high denudation rates

Supporting Information:
Supporting Information may be found in the online version of this article.
Among the small mountainous rivers in Taiwan, the landslide-dominated Liwu River (Kuo & Brierley, 2014), draining quartz-poor schist-slate and carbonate lithologies, is perhaps the best-studied. The Liwu River drains one of the fastest eroding catchments in Taiwan (Dadson et al., 2003;Derrieux et al., 2014;Fellin et al., 2017) and therefore has high carbon transfer rates that result from silicate weathering, sulfide oxidation driven carbonate dissolution, biospheric organic carbon transport-burial, and oxidation of petrogenic organic carbon (Calmels et al., 2011;Hemingway et al., 2018;Hilton et al., 2008;Hilton & West, 2020;Kao et al., 2014). Despite being a classic study site for rapidly eroding settings, the quantification of its denudation rate is still ongoing. Multiple techniques, including surveys of channel bedrock elevation using dated benchmarks (Hartshorn et al., 2002), sediment gauging (Dadson et al., 2003), dating incised terraces (Dadson et al., 2003;Schaller et al., 2005), in situ cosmogenic 10 Be in quartz from river sediment (Derrieux et al., 2014), and thermochronology (Fellin et al., 2017) have been employed in the Liwu River from local to basin scales and from single events to million-year timescales. A ∼30-fold variability in denudation rates emerged from these methods (Table 1).
In this study, we investigated denudation processes in the Liwu River using the 10 Be(meteoric)/ 9 Be ratio as a denudation rate proxy. Meteoric beryllium-10 ( 10 Be) is produced in the atmosphere and scavenged primarily by rainfall to the Earth's surface, while the release of stable 9 Be from bedrock depends on chemical weathering (Barg et al., 1997;von Blanckenburg et al., 2012). Unlike the sister nuclide in situ cosmogenic 10 Be produced in quartz at a rate of 10 0 -10 2 atoms/g/yr, the 10 Be(meteoric)/ 9 Be system integrates all lithologies including quartz-poor rock types and only requires ∼1 g for analysis due to a much higher depositional flux (e.g., ∼10 6 atoms/cm 2 /yr) von Blanckenburg et al., 2012). Hence, the 10 Be(meteoric)/ 9 Be ratio has been applied to track millennial-scale denudation processes in basins ranging from creek to continental-scale sizes (e.g., Dannhaus et al., 2018;Deng, Yang, et al., 2020;Portenga et al., 2019;Rahaman et al., 2017;Wittmann et al., 2015). In previous studies the denudation rates derived from in situ 10 Be and 10 Be(meteoric)/ 9 Be ratios mostly agreed within a factor of two. Where larger differences between both methods were found, their distinct response might be due to true variability in geologic conditions (e.g., lithology) and differences in denudation rate resulting thereof (Portenga et al., 2019;Wittmann et al., 2015).
By analyzing 10 Be(meteoric)/ 9 Be ratios of mainstem and major tributary sediment samples in the Liwu River, we constrain one of the world's highest denudation rates and show that stochastic landsliding is a plausible control on such extremely high and variable erosion rates. This study provides valuable insight into the applicability of cosmogenic nuclides in fast-eroding orogens.

Study Area
Collision between the Luzon Arc and the Asian continental margin has driven rapid uplift of the Taiwan orogen since ∼6 Ma (Huang et al., 2006). Six tectonic units are exposed from west to east: the Coastal Plain, Western Foothills, Hsuehshan Range, Backbone Range, Tananao Metamorphic Complex (Tailuko Belt and Yuli Belt) and Coastal Range (Figure 1a). The Backbone Range and the Tananao Complex can be grouped together as the Central Range. The island as a whole is mainly composed of (meta-) sedimentary rocks ranging in age from Paleozoic to Quaternary. The Liwu River, located at the eastern side of the mountain belt, drains the Miocene and Eocene slate of the Backbone Range in the headwaters, and then flows across Rate type Method Rate range Data source Suspended load Gauging 33.1 ± 8.3 Dadson et al. (2003) Channel incision 14 C or 36 Cl dating 16.7-68.9 Dadson et al. (2003) and Schaller et al. (2005) Basin-averaged denudation In situ 10 Be 2.1-13.2 Derrieux et al. (2014) Basin-averaged exhumation Fission-track dating 8.6 ± 2.1 Fellin et al. (2017)  Palaeozoic Tailuko schist and marble along the downstream toward the Philippine Sea ( Figure 1a). Slate, schist and marble account for 22%, 56%, and 22% of the total drainage area, respectively. In particular, no marble outcrops are present in the tributaries from the upper reaches (Waheir and Dasha). The areal extent of marble slightly increases along the mid-lower mainstem ( Table 2). The major soil types in the Hualien County, where the Liwu Basin is located, include Entisols and Inceptisols with moderately alkaline pH (Chen et al., 2015), and the soil thickness is commonly thin at 0.2-0.9 m (Hemingway et al., 2018).
The Liwu Basin with a river length of ∼58 km and an area of ∼620 km 2 is characterized by steep slopes. From upstream to downstream, the basin-averaged local slope increases from 30° to 36°. Below the confluence of the Dasha River, the basin-averaged slope is uniform at 35°-36° (Table 2). Although the Liwu Basin locally bears multiple transient geomorphic features such as hanging valleys (Wobus et al., 2006), topographic steady state appears to prevail over a larger scale, as evidenced by relatively constant topographic metrics (e.g., channel steepness) in the middle part of the Central Range (Chen & Willett, 2016) including the Liwu Basin. Over decadal scales, 14 Mt of suspended load is discharged annually to the Philippine Sea despite the small basin size (Dadson et al., 2003). In particular, the vertical incision of the river channel can exceed 100 mm within two typhoon seasons (Hartshorn et al., 2002).

Sampling Details
We carried out a sampling campaign along the Liwu River ( Figure 1) in early June 2017 prior to the typhoon season. Bedload samples were collected at three mainstem locations (downstream the Dasha confluence, at a mid-reach point, and close to the outlet) and from three tributaries (Waheir, Dasha, and Shakadang). Geomorphological characteristics of sampled basins are provided in  (Schwanghart & Scherler, 2014). b The samples are sorted according to the upstream distance from the river mouth. c The basin-averaged annual precipitation data set (with an observation period of  is sourced from WorldClim Version2 (Fick & Hijmans, 2017). d The mainstem station LW3 is located downstream the confluence of the Dasha River. e The sediment grain size described here is measured by sieving and weighing (Deng et al., 2019). Be adsorbs onto mineral surfaces or co-precipitates with amorphous (called "am-ox") and crystalline (called "x-ox") Fe-and Al-oxides and hydroxides. This 10 Be pool exchanges with dissolved 10 Be through desorption-adsorption or dissolution-precipitation reactions and is thus called the "reactive" (reac) fraction. Analogously, 9 Be enters the weathering zone in the dissolved form by release from bedrock during weathering and is assumed to equilibrate with 10 Be in solution prior to reactive phase formation. Any associated isotope fractionation is smaller than analytical uncertainty and indeed minor compared to the variability in riverine 10 Be/ 9 Be ratios (e.g., orders of magnitudes). The resulting concentrations are called [ 10 Be] reac (in at/kg) and [ 9 Be] reac (in mg/kg), respectively. The 10 Be(meteoric)/ 9 Be ratio in the reactive phase is less sensitive to Be retentivity and hydraulic sorting (i.e., variations in sediment grain-size)  compared to single meteoric 10 Be concentrations (Singleton et al., 2016). The 10 Be(meteoric)/ 9 Be-derived denudation rate (D met , in kg/m 2 /yr) is thus calculated as: (1) where 10 met Be E F (in at/m 2 /yr) is the depositional flux of 10 Be, [ 9 Be] parent (mg/ kg) is the 9 Be present in the parent bedrock prior to weathering, and the silicate residual "min" phase that hosts the immobile remainder of 9 Be is termed [ 9 Be] min (mg/kg). The unit of D met can be converted to mm/ yr using a density of 2.65 × 10 3 kg/m 3 . This equation is derived from a steady-state framework based on the assumptions that isotopic ratios of all the eroded sources are well-mixed, and that all atmospheric 10 Be input into the catchment is exported by riverine transport at the same rate. We will evaluate the validity of these assumptions in the landslide-dominated Liwu River in Section 5.4. The amount of 9 Be mobilized during weathering, termed f reac , is the ratio of [ 9 Be] reac to the sum of [ 9 Be] min and [ 9 Be] reac : 9 reac reac 9 9 reac min Be f Be Be The detailed derivation for Equations 1 and 2 is given in von . The two equations are simplified from those that also account for the dissolved fluxes of Be isotopes ([ 9 Be] diss , in mg/l and [ 10 Be] diss , in atoms/l). In the Liwu River we consider that the dissolved Be fluxes are minor as justified later in this section. In general, the successful application of the new proxy relies on four requirements (1-4) repeated below , and in the Liwu River draining mixed lithologies including marble, also requires additional lithology-specific information (5): (1) The atmospheric depositional flux F Be met 10 must be known. The knowledge of F Be met 10 is a prerequisite for any meteoric 10 Be application. Previous studies on the Taiwan orogen estimated 10 Be depositional flux by e.g., assuming a constant rain 10 Be concentration determined from other regions (Tsai et al., 2008; Paleozoic-Mesozoic. The outline of sampled tributaries is delineated in black. Two samples were collected within a few meters' distance at each sampling location except at SKD ( Table 2). 5 of 18 You et al., 1988). Here we use an up-to-date, site-specific flux of 0.77 ± 0.11 × 10 6 at/cm 2 /yr constrained by meteoric 10 Be profiles of Holocene river terraces in Taiwan (Deng et al., 2021). Other approaches for determination of 10 Be depositional fluxes, including general circulation models (GCMs) (Heikkilä & von Blanckenburg, 2015) and rainfall 10 Be-based fitting equation (Graly et al., 2011), are commonly applied over larger spatial scales and may overestimate 10 Be fluxes in our study area (Deng et al., 2021;. (2) The concentration of 9 Be contained in unweathered bedrock ([ 9 Be] parent ) needs to be known. [ 9 Be] parent in small catchments can be constrained by two strategies, which are (a) area-weighing of rock 9 Be concentrations from each geological unit or (b) linear regression between concentrations of an immobile element (e.g., Al) and Be in bedrock combined with determining a representative concentration of that immobile element in river sediment (Dannhaus et al., 2018). Note, however, that where variability of rock 9 Be concentration is large, as observed in the Liwu basin ( Figure 1a and Table S1 in Supporting Information S1), the area-weighing strategy could deliver biased results, because it assumes that 9 Be contributions of all lithologies are proportional to their outcrop area (rather than sediment contribution). The linear regression approach, in contrast, requires that Be and an immobile element in rocks are enriched in similar mineral types; and further that mixing of this immobile element in river sediment eroded from source rocks is indicative of mixing of source rock 9 Be. We thus adopt the linear regression approach here (see details in Section 5.2.2). (3) A representative f reac can be estimated from the [ 9 Be] min /[ 9 Be] reac ratio in river sediment. Even though the ratio of [ 9 Be] min to [ 9 Be] reac in Equation 2 can eliminate the dilution effect of quartz (lack of Be and higher abundance in coarser-grained sediment), this ratio may be biased by sorting as [ 9 Be] reac is potentially enriched over [ 9 Be] min in fine sediments of higher adsorption capability Wittmann et al., 2012). In the Liwu River, the suspended sediment fraction is estimated to dominate the total sediment load based on an empirical estimation on the fraction of suspended load (Turowski et al., 2010). Hence, to minimize the grain-size bias, sieving to a narrow grain-size fraction of bedload that is similar to the grain-size range of suspended load will likely provide [ 9 Be] min /[ 9 Be] reac of the bulk sediment load. (4) The partitioning of Be flux into the dissolved pool must be estimated. The existence of a large pool of total dissolved Be (operationally defined, entailing the sum of truly dissolved Be and colloidal Be), if not included for calculation, can lead to overestimation of denudation rates. Although ( 10 Be/ 9 Be) reac in Equation 1 is not affected by this loss, a bias might arise if the [ 9 Be] min /[ 9 Be] reac increases by partitioning of reactive 9 Be into the total dissolved pool. However, this bias can be quantified  as the fraction of total dissolved flux of Be isotopes in riverine export can be estimated using the pH-dependent Be partition coefficient K d (calculated from the ratio of [Be] reac to [Be] diss , in l/ kg). When K d is much higher than the ratio of runoff (q, in m/yr) to erosion rate (E, in kg/m 2 /yr), the loss of reactive Be into the total dissolved flux is negligible . Specifically, the bias (%) on denudation rate estimates arising from Be retentivity issues can be quantified as : The basin-averaged q/E derived from decadal gauging data (Dadson et al., 2003) reaches ∼73 l/kg. In comparison, a pH of >7 prevails along the dissolved pathways (soil interflow, groundwater and surface runoff) in the Liwu River (Calmels et al., 2011), corresponding to a minimum K d value of ∼1.7 × 10 5 l/kg based on empirically derived K d -pH relationships (Brown et al., 1992;You et al., 1989). When using a f min (the proportion of [ 9 Be] min in bulk 9 Be) of ∼0.9 (an average value of all Liwu samples), the resulting bias% on denudation rate is <1% for not including the dissolved Be pool. Hence, we argue that the partitioning of Be into the total dissolved fraction and the potential overestimation of denudation rate are minor here.
(5) Components of Be isotopes associated with carbonate minerals can be distinguished from other hydroxide-bound reactive Be by specific extractions. Sediment sampled from river basins in which abundant carbonate rocks exist likely contains different components of Be associated with two kinds of carbonate fractions: (a) detrital carbonate that has not exchanged with the 10 Be pool since its formation in Palaeozoic times (in the Liwu case), containing no 10 Be (that has decayed) and some 9 Be (as part of [ 9 Be] min ); (b) secondary carbonate that has likely equilibrated with the dissolved Be pool in soil 6 of 18 solutions during its precipitation, which must be regarded as part of the reactive Be pool ([ 10 Be] reac and [ 9 Be] reac ) for D met calculation. Hence, we added an additional extraction step to our original method : using acetic acid prior to the extraction of amorphous and crystalline oxides. This acetic acid step will mobilize different forms of Be associated with (both primary and secondary) carbonate minerals and a potentially adsorbed pool that is released in acidic environments (You et al., 1989).

Analytical Methods
River sediment samples taken from the field were first subsampled by the coning-quartering method, and then dry-sieved to a narrow grain-size range of 30-63 μm according to requirement three (Section 3.2). After overnight oven-drying, the extraction of 10 Be and 9 Be was performed according to Wittmann et al. (2012), with an amendment for carbonate-rich sediments. About 1 g of the sieved sediment was first treated with 8 ml 1 M acetic acid at 70°C with mild shaking for 12 hr (called HAc fraction), a procedure that aims to mobilize the carbonate fraction (adapted from Hohl et al., 2015). We monitored solution-pH and a value of lower than 5 was maintained to ensure the dissolution of carbonate minerals. The post-HAc leach residue was dried and weighed again to determine the mass fraction of carbonate removed during this step. Subsequently, the residue was treated according to the original extraction method  with 15 ml 0.5 M HCl and mild shaking at room temperature for 24 h to extract the amorphous hydroxides (called am-ox fraction). The residue was then treated with 15 ml 1 M NH 2 OH·HCl at 80°C with manual shaking every 10 min for 4 hr to extract crystalline hydroxides (called x-ox fraction). The post-(am-ox + x-ox) leach residue was dried again and weighed. All chemically extracted fractions and the remaining solid residue (called min fraction) were treated with HF and aqua regia mixtures to fully decompose the sample matrix.
Once fully dissolved in 3 M HNO 3 , all chemically extractable fractions (HAc, am-ox, x-ox) were split into two aliquots. One split and the min fraction were analyzed for stable 9 Be and major elemental concentrations by Inductively Coupled Plasma-Optical Emission Spectroscopy (ICP-OES, Varian 720-ES) at GFZ Potsdam. The second split of extractable fractions was spiked with 9 Be carrier of known weight (∼110 μg Be) and purified for 10 Be analysis according to established methods (e.g., von Blanckenburg et al., 1996). 10 Be concentrations were obtained from accelerator mass spectrometry (AMS) measurements of 10 Be/ 9 Be ratios (with carrier) at University of Cologne, relative to the standard KN01-6-2 with a 10 Be/ 9 Be ratio of 5.35 × 10 −13 (Dewald et al., 2013) that is consistent with the 10 Be half-life of 1.39 Myr (Chmeleff et al., 2010). We used a blank 10 Be/ 9 Be ratio of 1.4 ± 0.6 × 10 −15 (n = 6) for blank correction of 10 Be concentrations.

Results
Concentrations of 10 Be and 9 Be in different chemical fractions are shown in Figure 2 and Table 3.
The chemical extractable 9 Be concentration contained in the summed HAc, am-ox and x-ox fractions is low and uniform, showing little variation between 0.05 and 0.08 μg/g (Figure 2a) for most samples except LW2-2 (mainstem Liwu), where the extractable 9 Be concentration is slightly higher (0.11 μg/g). Mainstem and tributary samples show no clear difference in 9 Be concentrations. In contrast, the 10 Be concentration in the show the concentration of 9 Be and 10 Be in each fraction normalized to the total extractable concentration. WH-2 and -3, DS-2 and -3, LW3-1 and -2, LW2-1 and -2, and LW1-1 and -2 were collected at the same location. LW1-1_re is a lab replicate for LW1-1. DENG ET AL.
The partitioning between the three extractable fractions (each fraction normalized to the sum of HAc, amox and x-ox fractions) differs between 10 Be and 9 Be (Figures 2b and 2c). 9 Be percentages in each of these chemical fractions are generally uniform for all samples, with 52.5 ± 4.6% (±standard deviation) contained in the HAc fraction, 35.2 ± 3.1% in the am-ox fraction, and 12.4 ± 2.4% in the x-ox fraction, respectively. In contrast, the variation in 10 Be percentage between the extractable fractions is larger. The 10 Be percentage in the HAc fraction is 44.4 ± 16.6%, 35.9 ± 17.1% in the am-ox fraction and 19.7 ± 11.8% in the x-ox fraction, respectively.
10 Be(meteoric)/ 9 Be ratios vary over two orders of magnitude in all these fractions, from 1.83 × 10 −12 to 1.11 × 10 −10 in the HAc fraction, from 7.46 × 10 −12 to 1.48 × 10 −10 in the am-ox fraction, and in the x-ox fraction from 5.08 × 10 −12 to 1.79 × 10 −10 ( Figure 3). Because of fast erosion and thus [ 10 Be] sample being similar to [ 10 Be] blank , blank-corrected 10 Be(meteoric)/ 9 Be ratios of total extractable fraction in LW3-1, LW3-2 and LW1-2 (Table 4) show high relative analytical error (>40%) that can explain the discrepancy between samples at each station. For each sample 10 Be(meteoric)/ 9 Be ratios of the three chemical fractions mostly agree within uncertainties, shown by (a) the ratio of relative standard deviation among the three fractions to their average relative analytical error (Table S3 in Supporting Information S1), which mostly varies around 1 (0.96 on average), except WH-3 that has a ratio of 3.2, and (b) the consistency in distribution of 10 Be(meteoric)/ 9 Be ratios in all the samples between the three fractions from one-way analysis of variance test (p = 0.27). This  (HAc), amorphous oxide-bound fraction (am-ox), crystalline oxide-bound fraction (x-ox), and silicate residual fraction (min). All concentrations are calculated relative to the initial solid sample mass (∼1 g). c All uncertainties denote 1σ analytical errors. For 9 Be measurements using ICP-OES, the given uncertainty (5%) is the long-term repeatability which is propagated into 10 Be/ 9 Be ratios. For 10 Be measurements, the analytical uncertainty from AMS is propagated into 10 Be/ 9 Be ratios. We use a blank 10 Be/ 9 Be ratio of 1.4 ± 0.6 × 10 −15 (n = 6), corresponding to approximately 1.0 ± 0.5 × 10 4 atoms of 10 Be added by carrier, for blank correction of 10 Be concentrations. d LW1-1_re is a laboratory replicate of LW1-1, obtained from a split of the same sample prior to initial weighing. Although the difference in [ 10 Be] of each fraction between lab replicates is large due to low concentrations, they generally agree within uncertainty. Be(meteoric)/ 9 Be ratio than the mainstem samples (p < 0.05 from a two-sided Wilcoxon rank sum test).  Be concentration for each sampled upstream basin is provided in Table S2 in Supporting Information S1 for comparison. c Due to the large uncertainty propagated from blank-corrected ( 10 Be/ 9 Be) reac-c , uncertainty in D met of LW1-2 and LW3-1, -2 (in italic) are quite high. A lower-limit first-order estimate of denudation, termed D met-noblkcorr , is thus derived from blankuncorrected ( 10 Be/ 9 Be) reac-c (i.e., the 10 Be derived from the 9 Be carrier was not subtracted for calculating [ 10 Be] reac-c ). The given uncertainty thus does not include that of blank 10 Be/ 9 Be. Such estimate yields an upper-limit ( 10 Be/ 9 Be) reac-c and thus a minimum denudation rate. Be and 9 Be concentrations, 10 Be(meteoric)/ 9 Be ratios, and major elemental data are provided in Tables 3  and 4 and Table S4 in Supporting Information S1, respectively.

Partitioning of Be Into the HAc Fraction and Definition of the Reactive Fraction in the Liwu River
To define the reactive fraction in the Liwu River that should be in equilibrium with the dissolved fraction, we investigate the way in which Be is hosted by the HAc fraction. Specifically, we explore Be associated with (a) primary or (b) secondary carbonate minerals, or (c) Be that was sorbed to particle surfaces. For cases (b and c) the HAc fraction should be included in the operationally defined reac fraction, as these two reservoirs comprise Be equilibrated with the dissolved Be fraction. We adopted two approaches to discriminate between these options.
First we inspect 10 Be(meteoric)/ 9 Be ratios found in the extractions. If primary carbonate were the dominant source of [Be] HAc , ( 10 Be/ 9 Be) HAc would be expected to be much lower compared to those in am-ox and x-ox fractions, because all 10 Be once associated with primary carbonate of Paleozoic-Mesozoic ages (Figure 1) should have decayed. However, the general agreement (within uncertainties) between 10 Be(meteoric)/ 9 Be ratios in the three chemical fractions (HAc, am-ox and x-ox) for the majority of samples (Figure 3) suggests if anything an only minor contribution of primary carbonate to [Be] HAc .
Second, we predict the expected primary carbonate contribution to [Be] HAc using [Be] marble . To this end we first estimated the fraction of the mass removed from the initial solid sample during the acetic acid leaching step which comprises 4.9%-16.3% of the bulk weight for all samples (Table 3). Second, we used the 9 Be concentration in bedrock marble (0.028 ± 0.001 μg/g, Table S1 in Supporting Information S1) collected from two different locations to estimate the maximum contribution of 9 Be to the HAc fraction that is associated with primary carbonate minerals. Using a carbonate contribution of 16.3 wt% (SKD-2 draining mainly marble) as an upper limit, a 9 Be concentration of only 0.005 μg/g is predicted in bulk sediment. This 9 Be concentration is much lower than [ 9 Be] HAc of 0.041 μg/g measured in SKD-2 (Table 3), showing that primary carbonate minerals in this case cannot be the dominating reservoir for [ 9 Be] HAc .
Given the evidence provided, we propose that Be isotopes contained in the HAc fraction are dominated by different components associated with secondary carbonate and/or adsorption that can be extracted in the acetic acid step. We thus define the reactive fraction in the Liwu case as the sum of all the chemical extractable fractions, that are HAc, am-ox and x-ox fractions, termed as [ 10 Be] reac-c (in atoms/kg) and [ 9 Be] reac-c (in mg/kg). Corresponding 10 Be/ 9 Be ratios are termed as ( 10 Be/ 9 Be) reac-c . We apply this modified framework to Equations 1 and 2.

Weathering Intensity (f reac-c )
We calculate the fraction of Be released during weathering of Be-bearing primary minerals and partitioned into reactive fractions using Equation 2 and the reac-c fraction ([ 9 Be] reac-c ). The f reac-c can be seen as a Be-specific "weathering intensity," expressed as fraction of reactive 9 Be relative to bulk Be concentration. Its estimate relies on the assumption that the contribution of Be from primary carbonate minerals to [ 9 Be] reac-c is minor as justified in Section 5.1. f reac-c is 0.09-0.12 in the slowly eroding (i.e., high 10 Be/ 9 Be ratios) tributaries (Waheir and Dasha) and slightly lower at 0.07-0.09 in the rapidly eroding (i.e., low 10 Be/ 9 Be ratios) mainstem. We compare these values to those from global large river basins of similar sedimentary lithology where f reac data exist. f reac in the Amazon basin vary from 0.09 to 0.55 with an average value of 0.30 (n = 37) (Wittmann et al., 2015), the range of f reac in the Ganga basin is between 0.10 and 0.57 with an average value of 0.29 (n = 20) (Rahaman et al., 2017), and a global average value of 0.20 ± 0.08 (standard deviation) for f reac is suggested by von Blanckenburg and Bouchez (2014). A lower f reac-c in the Liwu Basin is compatible with the decrease of weathering intensity in fast-eroding settings (Dixon & von Blanckenburg, 2012).

Determination of [ 9 Be] parent
To constrain [ 9 Be] parent based on measurements of local bedrock samples around the Liwu Basin (n = 21, Table S1 in Supporting Information S1), we use a linear-fitting approach between concentrations of an immobile element (Al in this case) and Be in bedrock samples following Dannhaus et al. (2018). Be is enriched in silicate minerals similar to Al, and Be commonly substitutes into silicate minerals for Al 3+ by pairing with other ions (Ryan, 2002). As such, the similarity in mineralogical host between both elements suggests their similar behaviors during sediment mixing from different source rocks. However, Be can be subject to weathering or precipitation processes and thus the source rock Be [ 9 Be] parent can be modified during sediment transport, while Al as an immobile element is expected to behave more conservatively especially under conditions of low weathering intensity (Garzanti & Resentini, 2016) and a narrow grain-size range. Hence, by substituting [Al] of each sediment sample into this linear-fitting equation (Figure 4) we can derive a corresponding [ 9 Be] parent .
In compiled rock samples including marble (n = 21), we find Al and Be to be well-correlated (R 2 = 0.89, p < 0.05, Figure 4), whereas the correlation coefficients between other major immobile elements (Fe, K, and Ti) and Be, respectively, are lower (0.66-0.78, Table S1 in Supporting Information S1  [Al] in HAc, am-ox, x-ox and min fractions, Table S4 in Supporting Information S1), we derive [ 9 Be] parent representative for each sampled sub-catchment, ranging from 0.71 to 1.63 μg/g ( Table 4). Note that even though other immobile elements show weaker correlations with Be, choosing another element for the linear regression method does not significantly change resulting [ 9 Be] parent . For example, when replacing Al with K the resulting range of [ 9 Be] parent (0.87-1.70 μg/g) is similar (p = 0.33 from rank-sum test).

Denudation Rates From 10 Be(meteoric)/ 9 Be Ratios and Comparison With Other Approaches
Denudation rates from 10 Be(meteoric)/ 9 Be ratios (D met ) are calculated using Equation 1 and the reac-c fraction (( 10 Be/ 9 Be) reac-c from Table 4) and shown in Figure 5. Notably, for samples with extremely low ( 10 Be/ 9 Be) reac-c and high relative standard deviation of D met , we present denudation rates derived from blank-corrected 10 Be (D met ) and from 10 Be with no blank correction (D met-noblkcorr ) in Table 4. ( 10 Be/ 9 Be) reac-c calculated from the latter only represents an upper-limit estimate and thus D met-noblkcorr is a minimum estimate that can only be seen as a first-order approximation of the real denudation rate.
In the Liwu mainstem, D met are highest (>30 mm/yr) close to the upstream (LW3), then decrease to 8.1-9.0 mm/yr at LW2, and finally increase again to 23.9 mm/yr near the river mouth (mean of LW1-1 and LW1-1_re). The Waheir, Dasha, and Shakadang tributaries erode at rates of 3.4-21.5 mm/yr. Note that even though D met obtained for LW3 and LW1-2 is a minimum estimate, we are confident that these rates are meaningful, because the minimum rate of LW3 is similar to DS-3 measured for the large Dasha tributary close by, and LW1-2 agrees well with other samples measured at the same station. In general, two samples collected at each location agree in D met within (notably large) uncertainties. 10 Be(meteoric)/ 9 Be-derived denudation rates in the Liwu mainstem (8.1 to >30 mm/yr) are among the highest ever reported from cosmogenic nuclides, which typically do not exceed 10 mm/yr even in tectonically active regions such as Tibetan Plateau or west Taiwan (Cook et al., 2018;Deng, Yang, et al., 2020;Derrieux et al., 2014;Scherler et al., 2014). To probe into the causes for this upper limit of denudation rate  Dannhaus et al. (2018). The resulting linear-fitting equation is used to constrain [ 9 Be] parent for each sampled sub-catchment. Al concentrations measured from sediment samples (triangles) are plotted on the X-axis for comparison. Red arrows shown exemplify how a [ 9 Be] parent was derived: for LW1-2, the [Al] of 3.29% measured in sediment (sum of [Al] in HAc, am-ox, x-ox and min fractions, Table S4 in Supporting Information S1) corresponds to a [ 9 Be] parent of 1.02 ± 0.07 μg/g in bedrock.
DENG ET AL.
10.1029/2021JF006221 11 of 18 measurement from cosmogenic 10 Be in the Liwu River, we first need to evaluate if there is any methodological bias in such a dynamic sediment routing system.
The application of the 10 Be(meteoric)/ 9 Be method in the Liwu Basin can be potentially biased if there is a disequilibrium of Be isotopes between each reactive fraction and dissolved phase caused by different sources of 10 Be and 9 Be (rainfall vs. mineral weathering) and short residence time (You et al., 1989). As such, 10 Be(meteoric)/ 9 Be ratios may not faithfully record denudation processes over 10 2 -10 3 yrs. Previous studies have suggested that amorphous oxyhydroxides likely form by exchange with dissolved Be at a late stage in soils or rivers, whereas crystalline oxyhydroxides presumably form from amorphous oxyhydroxides aged during pedogenesis where they incorporated Be at an earlier stage (Dannhaus et al., 2018;Wittmann et al., 2015). Hence different reactive phases may record erosion and weathering processes integrating over different temporal scales and sometimes differ from each other in one sample (Wittmann et al., 2015). In contrast, an agreement in 10 Be(meteoric)/ 9 Be ratios between different chemical fractions may indicate a residence time of the reactive secondary minerals long enough for homogenization of Be isotopes in the location where reactive fractions form. Indeed, the residence time of hundreds of years for Liwu soils (Hemingway et al., 2018) will likely result in equilibrium of both isotopes, evidenced by the general agreement between 10 Be(meteoric)/ 9 Be ratios of the three fractions in the mainstem samples ( Figure 3). Hence, we consider that the extremely high D met is unlikely to be caused by a methodological bias. Instead, such D met is caused by specific geological processes in the Liwu Basin, explored in Section 5.4.
To further illustrate the variability in denudation rates in the Liwu Basin, we compare our D met to published rate estimates in the Liwu River integrating over different temporal and spatial scales ( Figure 5). Our D met data from the Waheir River (8.0-8.3 mm/yr) fall within the range of local channel incision rates (6.3-12.0 mm/yr) (Dadson et al., 2003). In the Dasha River the in situ 10 Be-derived rate (D insitu = 5.0 ± 1.8 mm/ yr) (Derrieux et al., 2014) agrees, within uncertainty, to sample DS-2 with a D met of 5.5 ± 1.3 mm/yr. We attributed the much higher D met of 21.5 ± 8.0 mm/yr measured for DS-3 ( Figure 5) to the larger contribution of storm-triggered landslide materials given that it was collected from flood deposits (Table 2). Such discrepancy between samples collected at one station can occur given the poor sediment mixing in the Liwu Basin (Deng et al., 2019). For the Shakadang River mainly draining marble, our D met (3.4 ± 0.7 mm/ yr) is ∼2-fold higher than D insitu (1.5 ± 0.4 mm/yr). The large difference between both cosmogenic nuclide-based approaches could be caused by the absence of quartz in large parts of this sub-catchment that might result in a non-representative D insitu . In the mainstem, rate estimates from thermochronology and in situ 10 Be are much lower than D met . Note, however, that the Central Range of Taiwan has gone through Figure 5. Spatial distribution of 10 Be(meteoric)/ 9 Be-derived denudation rates and other published denudation estimates in the Liwu Basin. Open symbols for each method represent tributary data. If error bars are overlapping, they are slightly shifted laterally. Symbols with thick black outlines (LW1-2 and LW3-1, -2) are calculated using ( 10 Be/ 9 Be) reac-c with no blank correction (i.e., D met-noblkcorr , Table 4) and thus represent minimum estimates. Elevation swath profile along the sampled reaches of the Liwu mainstem is shown on right axis. Data sources: low-temperature thermochronology (Fellin et al., 2017), in situ 10 Be data (Derrieux et al., 2014), channel incision (Dadson et al., 2003;Schaller et al., 2005), and sediment gauging (Dadson et al., 2003). Regarding the gauging-derived suspended sediment yield we added a bedload estimate, that is, 21% of total sediment load, using an empirical relationship between drainage area (A = 435 km 2 ) and the fraction of suspended load (F sus , %, =0.55 + 0.04 × ln(A)) (Turowski et al., 2010). A total modern sediment yield of 15.8 mm/yr hence results.
12 of 18 two stages of uplift, including the initial one (∼6 to ∼1 Ma) with slow uplift of ∼1 mm/yr and a second one (since ∼1 Ma) with rapid uplift of 4-10 mm/yr (Lee et al., 2006). As such, denudation rates integrating over centennial-millennial scales should be higher than the million-year scale exhumation rate here derived from a pooled age of 3.1 Ma using thermochronology (Fellin et al., 2017). The upper limits of published basin-wide and local rates in the mainstem are given by the decadal-scale sediment yield (15.8 ± 4.0 mm/ yr) (Dadson et al., 2003) and a millennial-scale gorge incision rate (26.0 ± 3.0 mm/yr) (Schaller et al., 2005). Both rates might be overestimated due to potential invalidity of method-specific assumptions: the monthly weighted-average method applied for calculation of gauging-derived sediment yield may be biased by the higher sampling frequency in flood seasons (Kao et al., 2005) and the calculation of gorge incision rate does not take into account lateral gorge wall retreat (Schaller et al., 2005). In comparison, D met in the mid-lower reaches (LW1 and LW2) agree with such literature limits when taking uncertainties into account, and D met at LW3 (>30 mm/yr) are slightly higher ( Figure 5).

Impact of Bedrock Landslides on Large Variability in Denudation Rates
In each sampled sub-basin along the mainstem (from LW3 to LW1), hydrological, topographical, and lithological controlling factors of centennial-millennial scale denudation (Lague, 2014) show very similar basin-averaged metric values (Table 2). We would thus expect denudation rates to agree between methods. However, our observations on D met and denudation/erosion rates from other methods ( Figure 5) clearly indicate that this is not the case. Previous studies attributed extremely high denudation rates (or low [ 10 Be]) and the lack of a clear spatial pattern to stochastic events such as recent landslides (Sosa Gonzalez et al., 2017;West et al., 2014) and/or short-term changes in sediment mixing due to hydrological variability (Lupker et al., 2012). Indeed, landslides triggered by heavy storms play an important role in sediment production and transport processes of the Liwu Basin (Hovius et al., 2000;Kuo & Brierley, 2013, 2014. Modeling of the impact of landslides on cosmogenic nuclide concentrations (Niemi et al., 2005;Yanites et al., 2009) suggested that only if the drainage area exceeds e.g., ∼100 km 2 , spatially averaged denudation rates from cosmogenic nuclides can be representative even affected by landslides. However, such an averaging effect is also modulated by the efficiency of fluvial sediment mixing (Yanites et al., 2009). Our previous study on the Zhuoshui River in West Taiwan that is also affected by landsliding showed a consistent downstream trend in D met (Deng, Yang, et al., 2020), in contrast to the lack of spatial pattern in the Liwu River. One hypothesis for such discrepancy is that the impact of stochastic landslides may be averaged-out more efficiently in the Zhuoshui River given its favorable conditions of sediment mixing including a larger drainage area (3 × 10 3 km 2 ) and wider channels (proportional to drainage area) (Yanites & Tucker, 2010). In comparison, the smaller Liwu Basin is characterized by faster sediment transfer and poorer sediment mixing (Deng et al., 2019) and the sediment sourced from landsliding materials may not be mixed to a representative average. We thus propose that variable contributions of recent landslides (timescale of e.g., 10 0 -10 1 yrs), together with poor fluvial sediment mixing, are likely the main geological cause for the extremely high and spatially variable denudation rates observed from multiple methods in the Liwu mainstem ( Figure 5). Next we estimate the impact of such landsliding on D met .
Compared to small landslides that only mobilize soil and occur at a high frequency, landslides mobilizing bedrock occur at a deeper depth and commonly contribute most of landsliding materials (Hovius et al., 1997;Marc et al., 2018). Bedrock landslides carry negligible amounts of reactive 10 Be (with concurrent extremely low ( 10 Be/ 9 Be) reac ) as their depth is much deeper than the shallow 10 Be infiltration depth scaling with soil depth (Willenbring & von Blanckenburg, 2010), being 0.2-0.9 m in the Liwu Basin (Hemingway et al., 2018). As a result, incorporating bedrock landslide materials into river sediment will lead to an increase in D met . Hence, we first assess the spatial pattern of landslide activity using published results of landsliding mapping, and then model how much bedrock material from landsliding is potentially incorporated into river sediment.
There is a clear downstream gradient in landslide activity, with the majority of the total landslide-affected area (50%) in the whole basin occurring in the fragile slate of the Liwu headwaters (Kuo & Brierley, 2014). In the sampled mid-lower mainstem (LW3 to LW1), a minor landslide-affected area of 0.045 km 2 was identified, accounting for only 2.8% of the total landslide-affected area in the whole Liwu trunk stream (Kuo & Brierley, 2013). Hence, the contribution of landslide debris to all the mainstem locations is mainly sourced 13 of 18 from the same unsampled upper reaches, that is, upstream of LW3. Furthermore, based on the average area of individual landslide scars in the upstream of LW3 (Kuo & Brierley, 2013) and an empirical landslide depth-area relationship (Larsen et al., 2010), we roughly estimate an average landslide depth of ∼4.8 m. This depth value estimated from mean landslide area may be overestimated, but it indeed falls within the range of scar depths estimated from other landslides in Taiwan (Marc et al., 2021) and is much deeper than the soil depth in the Liwu Basin.
To quantify the impact of bedrock landsliding on observed mainstem D met , we followed the framework developed by Yanites et al. (2009). We conceptually define sources of each sediment sample as a mixture of soil and bedrock, where soil is eroded from hillslopes by soil creep or small shallow landslides at a mean background rate (D b , kg/m 2 /yr) and fresh bedrock debris is supplied by event-triggered bedrock landslides. This mixing ratio will set the sediment 10 Be(meteoric)/ 9 Be ratio and hence D met , as this isotope ratio decreases from a high value in the upper soil to a value close to zero in unweathered bedrock at depth (Maher & von Blanckenburg, 2016). The simplified equation to calculate modeled denudation rates (D model ) under the framework of soil-bedrock mixing is based on the dilution effect of fresh landsliding materials on 10 Be and as follows: where f bedrock is the fractional contribution of sediment mass from bedrock produced by landslides. To constrain basin-averaged soil D b for the Liwu River, we adopted background erosion rates calculated by Chen et al. (2019) from the two catchments closest to the topographic steady-state zone where the Liwu Basin is also located (Chen & Willett, 2016). These yield an average D b of ∼0.9 mm/yr for the Danan River and an average ∼1.8 mm/yr for the Chihpen River, that is, a D b range of 0.9-1.8 mm/yr.
To constrain a realistic range of f bedrock we used radiocarbon ( 14 C) in organic carbon as a soil tracer (Hilton et al., 2008;Kao et al., 2014). Its vertical distribution in soil profiles (decreasing with depth) as well as its transport in rivers (bound to particles) is similar to that of 10 Be(meteoric)/ 9 Be. In brief, the organic 14 C activity, expressed as the fraction of modern 14 C ( 14 B E F ), is close to one in the upper soil due to addition of biospheric organic carbon and close to zero at depth due to dominance of petrogenic organic carbon in bedrock (Hemingway et al., 2018). Published

B
E F values in the Liwu mainstem determined on samples collected in different years covering three typhoon events (Hilton et al., 2008;Kao et al., 2014) vary within one order of magnitude and from 0.04 to 0.43 (n = 18, see Table S5 in Supporting Information S1). We can solve for f bedrock using organic 14 C data of Liwu sediments and a soil-bedrock 14 C mixing model. This isotopic mixing model, modified from Hemingway et al. (2018), is established based on the mass balance of organic matter content (OC) and 14 C activity: soil   Table S6 of Supporting Information S1. We performed Monte-Carlo simulations in this mixing model to assess the uncertainty of each end member. We randomly generated 10 6 estimates for sediment mixtures using Equations 5-7 based on end-members of soil and bedrock (Table S6 in Supporting Information S1) and f bedrock ranging from 0% to 100%. For each sediment sample, we calculated the average value and standard deviation of f bedrock from a proportion of all the modeled mixtures that can result in the same organic carbon content and Information S1). In general, the resulting high f bedrock indicates the dominance of contribution of fresh materials, consistent with the measured low 14 B E F (0.04-0.43) that is closer to the bedrock end-member.
We then calculate the range of D model using modeled f bedrock (55%-97%) according to Equation 4 (Figure 6). All the D met and literature rate estimates obtained for the mainstem fall within the range of D model ( Figure 6). Combining the model with D met measurements suggests that bedrock landslides contribute 77%-97% of sediment mass to our mainstem samples ( Figure 6). We discount the possibility that D met is biased by this high contribution of bedrock. Such bias potentially arises if sieving has removed coarsegrained materials sourced in bedrock (e.g., pebbles) prior to analysis. Bedrock-rich material adsorbs negligible reactive 10 Be and 9 Be but contains min 9 Be. Although sieving will not affect ( 10 Be/ 9 Be) reac , it might result in an underestimate of [ 9 Be] min /[ 9 Be] reac and accordingly of D met in Equation 1. However, we consider this possibility unlikely because the majority of the sediment load is dominated by finegrained suspended sediment (<63 μm) (Kao et al., 2008;Turowski et al., 2010) and thus our analyzed grain-size is considered to be representative. In brief, variable mixing between soil and bedrock caused by stochastic landsliding can indeed explain the extremely high D met and their large variability in the Liwu mainstem.
These model results are generally applicable to both cosmogenic nuclide methods (in situ and meteoric 10 Be) as their attenuation with depth is similar, and both methods record a mixture of background denudation rates and landsliding rates (Yanites et al., 2009). Nevertheless, one hypothesis to explain the discrepancy between denudation rates derived from in situ 10 Be and 10 Be(meteoric)/ 9 Be ratios in the mainstem ( Figure 5) is the different sensitivity of both methods to the impact of stochastic bedrock landslides. In Taiwan rivers, the majority of the total sediment load including landslide-generated materials is dominated by silt-and clay-sized sediment (Kao et al., 2008;Turowski et al., 2010), similar to the grain-size fraction probed by 10 Be(meteoric)/ 9 Be ratios. On the other hand, in situ 10 Be-derived denudation rates, measured in the coarse sand fraction (e.g., 250-1000 μm, Derrieux et al., 2014) might be mainly derived from coarsegrained and quartz-rich bedrock of minor outcrops and slower denudation, and thus underestimates the Figure 6. The impact of sediment contribution from bedrock (f bedrock ) by landslides on modeled denudation rates (D model ) in the Liwu River (blue shaded area). D met in the Liwu mainstem (ranging from 8.1 to >30 mm/yr, data from Table 4) are plotted in the left panel, and rate estimates from other approaches are included for comparison. Measured rates are sorted in ascending order. The range of f bedrock (55%-97%) in the Liwu Basin was derived from a soil-bedrock 14 C mixing model, while the lower and upper limits of the blue shaded field were constrained using a background soil denudation rate (D b ) range of 0.9-1.8 mm/yr (Chen et al., 2019). Measured D met and other literature data in the Liwu mainstem fall into the range of modeled denudation rates (blue shaded area) based on the soil-bedrock 14 C mixing model. impact of rapidly eroding quartz-poor lithologies. In addition, the amount of sediment needed for in situ 10 Be analysis is extremely large in the Liwu River given the low quartz content and low 10 Be concentrations. For example, based on the blank level of in situ 10 Be analysis reported in Derrieux et al. (2014), bulk sediment grain-size data (Deng et al., 2019), and a gauging-derived sediment yield of 15.8 mm/yr (Dadson et al., 2003), we calculate that to obtain a sufficient amount of quartz in the grain-size fraction of 250-1000 μm such that 10 Be amounts exceed the detection limit set by the 10 Be blank, a sediment amount exceeding 4.4 kg is needed (details in Table S7 of Supporting Information S1). Hence, denudation rates of >10 mm/yr from in situ 10 Be are likely below the limit of determination and are thus not accessible for measurements of denudation rates.
As such, the 10 Be(meteoric)/ 9 Be proxy may be more suitable in tracing landsliding processes compared to in situ 10 Be. Although future 10 Be(meteoric)/ 9 Be applications in fast-eroding settings should analyze more material (i.e., >2 g of sediment) to further reduce the uncertainties of D met , such required sample amount is still far less than that for in situ 10 Be method (at least several kg), and does not require labor-and time-intensive mineral separation. Another advantage of 10 Be(meteoric)/ 9 Be ratios in finegrained sediments is that it is more sensitive to landsliding activities as the generated materials may be mostly transported during flood events in the fine-grained, suspended form (Kao et al., 2008;Turowski et al., 2010). Such short-term D met variability may not consistently reflect long-term background denudation, but it can provide detailed information on the style of erosion and its dynamics. For example, given the requirement of several g sediment for 10 Be(meteoric)/ 9 Be analysis, we will have the possibility to measure suspended sediments with a high temporal (e.g., hourly) resolution during a typhoon event and record the delivery of storm-triggered landslide materials over event scale and its response to runoff magnitude and variability.

Conclusions
We applied the denudation proxy 10 Be(meteoric)/ 9 Be ratios to constrain the highest cosmogenic nuclide-derived denudation rates (D met ) worldwide, found in the Liwu Basin from the Taiwan orogen, and evaluated how stochastically distributed landslides can affect D met . The major findings are as follows: (1) We find that the distribution of 10 Be(meteoric)/ 9 Be ratios among all samples agrees between the various reactive fractions. Such consistency hints at a residence time of the reactive secondary minerals sufficient to attain equilibrium of Be isotopes with ambient fluids. This equilibration likely took place in the Liwu soil.
(2) D met varies from 8.1 to >30 mm/yr in the Liwu mainstem, and from 3.4 to 21.5 mm/yr in the tributaries.
Most of D met in the Liwu Basin agree with literature rate estimates within uncertainty. The extremely fast erosion processes lead to a much lower Be-specific weathering intensity (f reac-c of ∼0.1) than observed in global large river systems (∼0.3 on average). (3) We invoke stochastically distributed landslides to explain the extremely high values (>10 mm/yr) and variability (4-fold) in mainstem D met . Assuming that each sediment sample is a mixture of soil derived from hillslope creep and bedrock mobilized by landslides, we derive a range of fractional contributions of bedrock (f bedrock of 55%-97%) to Liwu mainstem sediments using a soil-bedrock 14 C mixing model and a published river particulate organic 14 C data set. The measured D met data set agrees with modeled rate estimates by f bedrock .
We conclude that the 10 Be(meteoric)/ 9 Be proxy can provide geomorphically meaningful constraints on the highest denudation rates in the world even though a few extremely high rates are associated with a large uncertainty. This study sheds new light on the limitations of the determination of cosmogenic-derived denudation rates in landslide-dominated routing systems, and also on the potential of 10 Be(meteoric)/ 9 Be in tracing landsliding processes.