Corresponding author: R. G. Hilton, Department of Geography, Durham University, Durham DH1 3LE, UK. (firstname.lastname@example.org)
 Erosion of particulate organic carbon (POC) occurs at very high rates in mountain river catchments, yet the proportion derived recently from atmospheric CO2 in the terrestrial biosphere (POCnon-fossil) remains poorly constrained. Here we examine the transport of POCnon-fossil in mountain rivers of Taiwan and its climatic and geomorphic controls. In 11 catchments we have combined previous geochemical quantification of POC source (accounting for fossil POC from bedrock), with measurements of water discharge (Qw) and suspended sediment concentration over 2 years. In these catchments, POCnon-fossil concentration (mg L−1) was positively correlated with Qw, with enhanced loads at high flow attributed to rainfall driven supply of POCnon-fossil from forested hillslopes. This climatic control on POCnon-fossil transport was moderated by catchment geomorphology: the gradient of a linear relation of POCnon-fossil concentration and Qw increased as the proportion of steep hillslopes (>35°) in the catchment increased. The data suggest enhanced supply of POCnon-fossil by erosion processes which act most efficiently on the steepest sections of forest. Across Taiwan, POCnon-fossil yield was correlated with suspended sediment yield, with a mean of 21 ± 10 tC km−2 yr−1. At this rate, export of POCnon-fossil imparts an upper bound on the time available for biospheric growth, of ∼800 yr. Over longer time periods, POCnon-fossil transferred with large amounts of clastic sediment can contribute to sequestration of atmospheric CO2 if buried in marine sediments. Our results show that this carbon transfer should be enhanced in a wetter and stormier climate, and the rates moderated on geological timescales by the regional tectonic setting.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 In order to examine the controls on POCnon-fossil transport and quantify its rate of transfer, POCnon-fossil concentration must be examined as a function of water discharge (Qw, m3 s−1). Only a handful of studies have achieved this, focusing on individual catchments to provide quantification of annual and flood-driven POCnon-fossil transfer [Kao and Liu, 2000; Hilton et al., 2008a; Townsend-Small et al., 2008; Hatten et al., 2012]. These studies have identified the importance of: i) runoff and runoff variability; ii) catchment geomorphic setting; iii) physical erosion rate; and iv) aboveground carbon stock for POCnon-fossil transport and transfer. To understand better how these climatic, geomorphic and biological drivers operate, we require measurements of the fluvial transport of POCnon-fossil (mg L−1) and estimates of POCnon-fossil yields (tC km−2 yr−1) from multiple catchments across gradients in controlling variables.
 Here we focus on the role of climatic and geomorphic factors in the forested mountain belt of Taiwan, where organic carbon stocks are relatively uniform [Chang et al., 2006; West et al., 2011]. We have obtained hydrometric data (Qw and suspended sediment concentration) and collected suspended sediment samples from 11 major rivers draining the Central Range mountains over two years. The abundance of POCfossil in these samples has been quantified previously [Hilton et al., 2010, 2011a] allowing, for the first time, an examination of the mobilization and transport of POCnon-fossil from a subtropical mountain forest as a function of Qw. Moreover, constraints on the prevalence of steep hillslopes in study catchments provides new insight into how POCnon-fossil transfer is moderated by the erosion processes which supply POCnon-fossil to the river channel. Finally, using suspended sediment yield, we assess the role of physical erosion rate on POCnon-fossil export and examine its impact on the time available for development of the mountain biosphere and the implications for regional and global carbon cycles.
2. Study Area
2.1. Tectonic and Climatic Setting
 Taiwan is located at 22–25°N at the western edge of the Pacific Ocean. Mountain building is driven by collision between the Luzon Arc on the Philippine Sea Plate and the Eurasian continental margin since ∼7 Ma [Beyssac et al., 2007]. It has formed the Central Range, standing almost 4 km above sea level and 9 km above the nearby ocean floor. Bedrock rivers drain its steep topography to the Pacific Ocean and Taiwan Strait [Dadson et al., 2003; Kao and Milliman, 2008] and have cut into metamorphosed Mesozoic and Cenozoic siliciclastic and carbonate rocks [Ho, 1986; Hartshorn et al., 2002] which contain between 0.2 and 0.4 weight % of POCfossil [Kao and Liu, 2000; Hilton et al., 2010]. The climate is subtropical, with 2–4 m yr−1 of rainfall, most of which falls between June and October when tropical cyclones (typhoons) impact the island [Wu and Kuo, 1999; Galewsky et al., 2006].
2.2. Vegetation Cover and Catchment Characteristics
 The humid climate of Taiwan sustains vegetation throughout the Central Range, where forest reaches the highest ridge crests. The evergreen forest contains Ficus, Machilus, Castanopsis, Quercus, Pinus, Tsuga, and Picea [Su, 1984] and large areas of the mountain ecosystem are protected with logging monitored [Lu et al., 2001]. The aboveground standing biomass of the mixed conifer-hardwood forest in Taiwan is 21.6 ± 9.4 × 103 t km−2 [West et al., 2011], representing an average organic carbon stock of 11 ± 5 × 103 tC km−2. Soils in Taiwan are relatively thin due to the rapid physical denudation rate [Hovius et al., 2000; Dadson et al., 2003], with the average base of the saprolite at ∼0.8 m (n = 310) in a Central Range catchment [Tsai et al., 2001; Ho et al., 2012]. A-horizons are ∼0.1 m thick and contain the majority of the organic matter [Tsai et al., 2001], with surface soils (<0.1 m) beneath coniferous forest found to contain 7 ± 2 × 103 tC km−2 [Chang et al., 2006]. The values of organic carbon stock are similar to averages of lowland tropical forests [Dixon et al., 1994].
 The river catchments selected for study drain the Central Range and range in size from 310 km2 to 2,906 km2 (Table 1) covering ∼30% of Taiwan's surface area. Upstream, the land use is dominated by mixed conifer-hardwood forest [West et al., 2011; M. C. Hansen et al., Vegetation continuous fields MOD44B, 2001 Percent Tree Cover, Collection 4, 2006, http://glcf.umiacs.umd.edu/data/vcf/, hereinafter referred to as Hansen et al., online data set, 2006]. During the study period, the mean annual runoff was relatively constant between the catchments at 2.9 ± 0.2 m yr−1 (n = 11; ± standard error), suggesting no significant gradients in mean annual precipitation. However, within each catchment runoff variability was marked, with daily mean Qw ranging over a factor 300 from ∼0.1 to ∼30 times the mean. In addition, the mean suspended sediment yield varied by up to a factor of 25 between catchments (Table 1), with a mean of 24,000 ± 7,000 t km−2 yr−1 (n = 11; ± standard error) [Hilton et al., 2011a]. The study catchments also have variable geomorphic characteristics, which reflect the tectonic evolution of the mountain belt as well as the local bedrock geology [Dadson et al., 2003; Ramsey et al., 2007]. The distribution of hillslope angles in each catchment, a primary control on the rates of physical erosion processes in mountain topography [Dietrich et al., 2003], varies notably. In most lithologies with pervasive jointing, hillslopes become disproportionately prone to failure at an angle of 30°–35°, with bedrock landslides most likely on the steepest sections of topography [Burbank et al., 1996; Clarke and Burbank, 2010]. The rates of erosion by processes other than bedrock landslides (e.g., shallow landsliding, overland flow) also increase rapidly above this threshold [Roering et al., 1999, 2001]. We therefore quantify the proportion of surface area with slopes >35° from a 40 m DEM of Taiwan [Dadson et al., 2003] and find that this varies significantly among the studied catchments. In the Linpien catchment, located in the South West where relief is relatively low and Cenozoic inter-bedded sandstones and shales dominate the geology [Ramsey et al., 2007; Hilton et al., 2010], 23% of the catchment area has slopes >35° (Table 1). In the Liwu River in the North East, which is underlain by more competent, high-grade metamorphic rocks [Ramsey et al., 2006; Beyssac et al., 2007; Hilton et al., 2010], very steep slopes are prevalent over 52% of the catchment area (Table 1).
Table 1. Geomorphic Characteristics, Suspended Sediment Yield (SSY), and Mean Water Discharge (Qw) for the Sampled Rivers Over the Study Period
3.1. Sample Collection, Processing and Geochemical Analyses
 Suspended sediment samples were collected at 11 gauging stations where water discharge (Qw, m3 s−1) and suspended sediment concentration (SSC, mg L−1) are routinely monitored. The details of our sampling methods have been described elsewhere [Dadson et al., 2003; Hilton et al., 2008a; Kao and Milliman, 2008; Hilton et al., 2010]. In summary, rivers were sampled, on average, one to three times per month over two typhoon seasons in 2005 and 2006 (Figure 1a and Table 1). The Liwu River was sampled during 2004 in a similar manner, but suspended load was also collected at a higher, daily frequency during specific typhoon floods [Hilton et al., 2008a]. Given the turbulence of these rivers at the sampling site, our samples are representative of the suspended sediment carried by the rivers [Lupker et al., 2011]. The maximum grain size of POCnon-fossil in these samples was found to be ∼500 μm [Hilton et al., 2010]. The transfer of coarse woody debris (CWD), while potentially important [West et al., 2011], was not quantified in this study.
 The concentration of suspended POCnon-fossil (mg L−1) was determined following inorganic carbon removal and analysis of the organic carbon concentration of the suspended load (Corg, %), the nitrogen to organic carbon ratio (N/C) and the stable isotopes of organic carbon (δ13Corg, ‰) by a Costech Elemental Analyzer coupled via Conflo-III to a MAT-235 stable isotope mass spectrometer. The fraction of non-fossil POC (Fnf) was quantified using N/C and δ13Corgand an end-member mixing analysis for each sample, detailed byHilton et al. . POCnon-fossil concentration (mg L−1) for each sample was determined as the product of SSC, Corg and Fnf. Hilton et al.  found Fnf to be a reliable predictor to correct for fossil POC input when tested against independent constraint from measurements of 14C content in 9 samples from the Liwu River. This is an appropriate test catchment for the mixing model as it comprises geological formations spanning the full range in POCfossil compositions found in the mountain belt [Hilton et al., 2010]. Fnf was found to have an average precision of 0.09 and represents the largest source of uncertainty in our analysis of POCnon-fossil transfer. The error on POCnon-fossil concentration was highest in samples where Fnf < 0.10, with an average error of 50% across the catchments (n = 11). Weighted by Qw, errors on POCnon-fossil were lower on average (44%, n = 11) because Fnf was typically >0.2 at high flow (Figure 2) and the maximum absolute uncertainty in POCnon-fossil concentration was 40 mg L−1, ∼25% of the calculated concentration in that sample (160 mg L−1). Despite these limitations, Fnf provides robust constraint on the erosion of soil and vegetation POC, with the reported errors much smaller than the measured range in POCnon-fossil concentration over three orders of magnitude (Figure 3).
3.2. POCnon-fossil, Qw and Quantification of Yields
 Relationships between POCnon-fossil concentration (mg L−1) and Qw in small mountain rivers have previously been described by power law [Hilton et al., 2008a; Hatten et al., 2012] and linear [Townsend-Small et al., 2008] relationships. These relationships can be used to compare the transport of POCnon-fossil in different catchments during similar hydrological conditions, using the mean Qw (Qmean) to normalize Qw (Qw/Qmean). To date, power laws relations of POCnon-fossil and Qw have been fitted either to data in catchments where POCnon-fossil dominates the total POC load [e.g., Hatten et al., 2012] or where Fnf has been quantified by 14C measurements [e.g., Hilton et al., 2008a], i.e., when the error on each POCnon-fossil measurement was negligible. This does not apply in our case due to uncertainty on Fnf [Hilton et al., 2010]. In the majority of our 11 catchments, least squares best fits of power laws were not statistically significant, which may partly reflect the reported errors on POCnon-fossil in this study. Instead, a linear relationship was quantified with slope (m-POCnon-fossil) and intercept (c-POCnon-fossil). Statistical analyses were carried out in Origin Pro™.
 Power law rating curves can be used to quantify the yield of particulate constituents. Suspended sediment yield (SSY, t km−2 yr−1) was quantified by Hilton et al. [2011a] using rating curves between Qw and SSC, then applied to the daily record of Qw, for each catchment between 2005 and 2007 (2004 for the Liwu River) (Table 1). SSY was also quantified using water discharge-weighted mean SSC as described elsewhere [Walling and Webb, 1981; Ferguson, 1987]. This flux-weighted method (SSYfw) can provide robust quantification of river loads in the absence of a power law rating curve [Ferguson, 1987]. It was, therefore, applied to estimate POCnon-fossil yields (tC km−2 yr−1) in each catchment.
4.1. Fluvial Transport of POCnon-fossil
 Suspended sediments were collected over a large range in Qw, with Qw/Qmean at the time of sampling ranging from ∼0.1 to ∼30 in catchments (Figure 2). Over this range, Fnf varied between 0 and ∼0.8, the highest values occurring during low flows with Qw/Qmean < 3. For these flows, there was a negative correlation between Fnf and Qw/Qmean (r = −0.31; P < 0.0001; n = 273) and a negative correlation between Corg and Qw/Qmean (r = −0.40; P < 0.0001; n = 273). However, for larger events during floods with Qw/Qmean > 3, there was no evidence for a decrease in Fnf with increased Qw (r = 0.2; P = 0.2; n = 52; Figure 2), nor of any dilution of Corg (r = 0.02; P = 0.8; n = 52) across the sample set.
 Measured POCnon-fossil concentrations (mg L−1) ranged over three orders of magnitude (Figure 3) to a maximum of 160 ± 40 mg L−1. This covers the range of previously reported concentrations in Taiwanese rivers and small mountain rivers elsewhere [Hilton et al., 2008a; Hatten et al., 2012]. The lack of a decrease in Fnf and Corg at high Qw (Figure 2) resulted in a lack of dilution of POCnon-fossil concentration (mg L−1) (Figure 3) as suspended load mass increased with discharge [Hilton et al., 2011a]. A strong positive correlation between POCnon-fossil concentration and Qw/Qmean exists (r = 0.49; P< 0.0001; n = 325) which contrasts previous results from non-mountainous catchments [cf.Ludwig et al., 1996; Stallard, 1998]. The positive correlation held in all but two of the sampled catchments (Figure 3 and Table 2), its gradient (m-POCnon-fossil) varying from 0.27 ± 0.08 in the Linpien River to 6.43 ± 0.78 in the Peinan River. The intercept (c-POCnon-fossil) varied between −5.0 ± 3.0 mg L−1 in the Peinan River to 1.1 ± 1.2 mg L−1 in the Choshui River.
Table 2. POCnon-fossil Transport and Transfer in the Study Catchmentsa
Here nd indicates linear fit between Qw/Qmean and POCnon-fossil concentration was not statistically significant and so parameters were not determined.
Flux-weighted SSY for study period.
Flux-weighted average Fnf.
Flux-weighted POCnon-fossil yield and error on yield (σ).
 The sampling strategy did not specifically target floods caused by tropical cyclones [cf. Goldsmith et al., 2008; Hilton et al., 2008a] because of the logistical difficulties and hazards associated with these events. However, four samples were collected during Typhoon Haitang (onset 19 July 2005, flood peak at 01:00 20 July 2005) in the Peinan River. At the peak of the flood there was enhanced POCnon-fossil transport at high Qw (Figure 1b), confirming the observations made previously in other Taiwanese rivers [Hilton et al., 2008a] and in flood deposits of the Waipaoa River, New Zealand [Gomez et al., 2010].
4.2. Particulate Yields
 Across the area covered by the 11 catchments, the average POCnon-fossilyield, estimated using the discharge-weighted mean POCnon-fossil concentration, was 21 ± 10 tC km−2 yr−1. Over the study period POCnon-fossil yields varied from 1.2 ± 1.0 tC km−2 yr−1 in the Hsiukuluan River in central east Taiwan, to 74 ± 22 tC km−2 yr−1 in the Peinan River to the south (Table 2). The Peinan River yield is among the highest ever recorded for a multiannual average. POCnon-fossil was approximately 30% of the total POC load exported by these mountain rivers, with POCfossil making up the remaining part [Hilton et al., 2010] and contributing, on average, 82 tC km−2 yr−1 [Hilton et al., 2011a].
 POCnon-fossil yields were strongly correlated with SSY over three orders of magnitude (Figure 4a), which was not the consequence of varying drainage area. Using this, we can compare the published SSY over the study period from power law rating curves (Table 1) [Hilton et al., 2011a] with those derived from the same flux-weighted method, SSYfw, to determine whether the discharge-weighted estimation of POCnon-fossil yield is a robust method. The two SSY estimates are strongly, linearly correlated by SSYfw = 0.74 ± 0.04*SSY (r2 = 0.96; P < 0.0001; n = 11), suggesting that the POCnon-fossilyields estimated by flux-weighting [Ferguson, 1987] are robust. However, the SSYfware on average 21% lower than published, rating curve-derived SSY (Table 2). This is because the flux-weighted method does not fully account for the role of very large floods in the annual hydrograph [Ferguson, 1987], for example during tropical-cyclones in Taiwan [Dadson et al., 2005]. This is confirmed by the observation that the discharge-weighted POCnon-fossil yield for the Liwu River (for 2004) was 6.8 ± 2.7 tC km−2 yr−1, which is lower than previous estimate of POCnon-fossil yield during Typhoon Mindulle in 2004 [Hilton et al., 2008a] of 13 tC km−2 derived with a rating curve (Figure 4a). Aiming to examine the variability in POCnon-fossil yield between catchments (as a function of geomorphic characteristics and physical erosion rate), we have not applied any correction for these underestimations of POCnon-fossil. Instead, we suggest that the POCnon-fossil yields reported here are internally consistent, but are likely to be conservative.
 Over the study period, the combined export from the monitored catchments was 0.21 ± 0.04 × 106 tC yr−1 of POCnon-fossil (Figure 4b). Assuming a yield of 21 ± 10 tC km−2 yr−1 across Taiwan's mountain forest (22,665 km2), the corresponding POCnon-fossil flux from the Taiwan orogen to the ocean in suspended sediment was 0.5 ± 0.2 × 106 tC yr−1. To determine whether the measured yields are representative of a longer-term (decadal) export, we note that SSY over the sampling period (mean 24,000 ± 7,000 t km−2 yr−1, ± standard error) were similar to those estimated in the same catchments by Dadson et al.  over three decades, 1970–1999 (mean 22,000 ± 4,000 t km−2 yr−1, ± standard error). In view of the strong correlation of SSY and POCnon-fossil yields (Figure 4a) this suggests that the POCnon-fossil yields are likely to be a representative, albeit conservative for reasons previously stated, estimate of the longer term POCnon-fossil transfer.
5.1. Fluvial Transport of POCnon-fossil: Capacity and Supply
 Our results demonstrate that Corg and Fnf do not decrease at high Qw (Figure 2) and thus that POCnon-fossil is not diluted at the peak of large flood events (Figure 1b). This leads to a positive correlation between POCnon-fossil concentration and Qw/Qmean (Figure 3) which is analogous to that commonly observed between Qw/Qmean and SSC in mountain rivers [Hovius et al., 2000; Fuller et al., 2003; Hicks et al., 2004a; Kao and Milliman, 2008; Hovius et al., 2011]. For clastic sediment, SSC increase with Qw is often attributed to variability in: i) the capacity of the river to transport sediment as suspended load; and ii) the supply of suspendable sediment (sand, silt and clay) to the river channel. In mountain rivers a third factor may also be important, namely the production of suspended sediment by pebble abrasion at high levels of bed shear stress and associated bed load transport [Attal and Lavé, 2009]. We hypothesize that these factors also control POCnon-fossil transport and examine their potential roles herein.
 The capacity of a river to entrain and transport fine sediment increases with water flow velocity and turbulence [Garcia and Parker, 1991]. Given the restricted channel geometry in bedrock rivers [Turowski et al., 2008], capacity is likely to increase with Qw. Turbulent mixing, typical of mountain river channels with large scale bed roughness, may also increase the entrainment rate and transport capacity of the flow [Jackson, 1976]. POCnon-fossil should be less dense than the accompanying mineral sediment load, even when waterlogged [Buxton, 2010], causing its propensity for entrainment and transport to increase rapidly with Qw [Hamm et al., 2011]. However, in five of the catchments we observe negative values for c-POCnon-fossil, the linear intercept between POCnon-fossil concentration and Qw (Table 2). The physical meaning of a negative intercept implies either a threshold for motion for POCnon-fossil, which may be the case for coarse woody debris (CWD) [West et al., 2011; Wohl, 2011] but seems unlikely for fine POCnon-fossil [Hamm et al., 2011], or a limit on the transport of POCnon-fossil in river channels imposed by its supply. River channels in Taiwan are characterized by a lack of vegetation due to frequent flooding preventing colonization by plants [Hartshorn et al., 2002] and therefore the supply of POCnon-fossil must originate from forested hillslopes.
 The rate at which geomorphic processes erode the landscape are known to depend on the steepness of the topography on which they act [Roering et al., 2001], and high rates of physical erosion by landsliding and overland flow are therefore expected to occur in Taiwan. Overland flow preferentially mobilizes loose material and POCnon-fossil from surface soils [Gomi et al., 2008]. Bedrock landslides can remove entire tracts of mountain forest and soil, harvesting the whole biomass and mixing it with POCfossil [Hilton et al., 2008b; West et al., 2011; Hilton et al., 2011b]. The influence of supply on POCnon-fossil transport can be examined using the hysteresis of POCnon-fossil and Qw during individual flood events, as documented by Hilton et al. [2008a]. That study demonstrated that after several hours of sustained rainfall, enhanced POCnon-fossil concentrations were observed across a range in Qwwhen compared to dry intervals. Rainfall activates geomorphic processes of overland flow and landsliding and leads to efficient hillslope-channel coupling and the supply of POCnon-fossil. In addition, at high flood stage the river has capacity to transport CWD [West et al., 2011] the mechanical attrition of which may also enhance POCnon-fossil concentrations in the river suspended load [cf. Attal and Lavé, 2009]. In contrast, during periods without substantial rainfall, supply from hillslopes is minimal and POCnon-fossil is likely to be sourced from channels, where bed sediments are typically dominated by POCfossil [Hilton et al., 2010]. Thus, POCnon-fossil concentrations are lower for similar hydraulic conditions [Hilton et al., 2008a].
 Organic carbon measurements on samples collected during the flood caused by Typhoon Haitang in the Peinan River are consistent with these observations [Eglinton, 2008]. Measured precipitation on 19 July 2005 totaled 110 mm in Taitung (Figure 1b) near to the gauging station (22.76°N, 121.15°E, data from the Central Weather Bureau, Taiwan, http://www.cwb.gov.tw/). On that day, the sample collected 14 h prior the peak of the flood, on the steep rising limb, had a POCnon-fossil concentration of 160 ± 40 mg L−1 with Fnf = 0.39 ± 0.09. 32 h after the flood peak (09:40 21 July 2005), POCnon-fossil concentration had dropped by 75% to 40 ± 15 mg L−1 (Fnf = 0.24 ± 0.09) despite only a slight decrease (∼10%) in Qw/Qmean from 16 to 14. The marked drop in POCnon-fossilconcentration was co-incident with the cessation of heavy precipitation over the catchment (Figure 1b). These results demonstrate that while landsliding and overland flow are moderated by slope angle [Dietrich et al., 2003], their temporal occurrence is stochastic [Benda and Dunne, 1997; Hovius et al., 2000]. As a result, the fluvial transport of fine POCnon-fossil may vary at a given transport capacity (Qw) due to the specific timing and location of POCnon-fossil supply to the river. This explanation is also consistent with the observed variability in POCnon-fossil concentration for individual catchments (Figure 3) and confirms the importance of POCnon-fossil supply during rainfall [Hilton et al., 2008a], when erosion processes efficiently couple forested hillslopes to the river channel.
 The relative importance of the POCnon-fossil supply processes identified here (overland flow, bedrock landslides, mechanical attrition) remains an avenue for future research. However, the observed lack of Fnf decrease with increasing Qw provides some insight (Figure 2). As established, bedrock landslides are ubiquitous in Taiwan [e.g., Lin et al., 2008] and known to be crucial for delivering clastic sediment to river networks at the peak of floods [Hovius et al., 2000; Fuller et al., 2003; Dadson et al., 2005; Hilton et al., 2008a]. However, erosion of POC by this process can decrease Fnf (decrease POCnon-fossil:POCfossil ratio) at times of high sediment delivery. As the surface area of a bedrock landslide increases (i.e., its POCnon-fossil erosion) it is known that its volume (i.e., sediment and POCfossil erosion) increases as a power law with an exponent >1.2 [Guzzetti et al., 2009; Larsen et al., 2010], implying large landslides can dig deeper and reduce Fnf [Hilton et al., 2008b]. Therefore, the observation of elevated Fnf during high flow (Figures 1b and 2) implies supply of POCnon-fossil by a process other than deep bedrock landslides. Mobilization of surface materials by overland flow, and mechanical attrition of CWD do not contribute POCfossil. One or both of these processes must contribute significantly to POCnon-fossilfluxes in floods. These considerations support conclusions from the Western Southern Alps, New Zealand. There, decadal estimates of landslide-driven POCnon-fossil yield were lower than estimates of fluvial export, requiring additional processes of POCnon-fossil supply from the mountain hillslopes [Hilton et al., 2011b].
5.2. Enhancement of POCnon-fossil Transport
 Rainfall-driven changes in erosional supply underlie a strong climatic control on the mobilization and transport of POCnon-fossil (Figure 3), which should have a similar expression in each catchment. However, it is clear that the positive relationship between POCnon-fossil concentration and Qw/Qmean is not constant for Taiwanese Rivers. This is articulated in the range in gradients of the linear best fit to the data (m-POCnon-fossil), from 0.27 ± 0.08 to 6.43 ± 0.78 (Table 2). m-POCnon-fossil can be viewed as an enhancement factor, with a steeper gradient reflecting increased loading of POCnon-fossil across a range of hydrological conditions. As established previously (Section 5.1), supply is likely to be the main control on the variability in POCnon-fossil concentration, rather than transport capacity in these rivers. Thus, enhancement should relate primarily to the efficiency of erosion processes delivering POCnon-fossil from hillslopes to channels.
 The Taiwanese rivers have a positive trend between m-POCnon-fossil and the area of the catchment with steep slopes above typical thresholds for mass wasting and erosion processes (>35°) (Figure 5). Between the Linpien River (Figure 3c) and the Liwu River (Figure 3a) the trend is nonlinear (n = 8). Such a trend is consistent with the mechanics of the geomorphic processes responsible for POCnon-fossil supply [Gomi et al., 2008; West et al., 2011; Hilton et al., 2011b]. Landsliding and overland flow processes are both stochastic and their rates of occurrence are a nonlinear, threshold functions of slope and runoff [Benda and Dunne, 1997; Roering et al., 1999; Hovius et al., 2000; Dietrich et al., 2003]. Steepening the topography of a catchment should increase the rate of POCnon-fossil supply, but only once hydrological thresholds are surpassed. This explains both the increase in POCnon-fossil with Qw (Figure 3) and enhanced rate of POCnon-fossil supply when steep slopes contribute more importantly to the catchment hypsometry (Figure 5).
 The Peinan River, in the southwest of Taiwan, has an m-POCnon-fossil of 6.43 ± 0.78 and lies significantly off the trend in the data set (Figure 5). To explain the higher loads of POCnon-fossil in this catchment, we note that it also has had a very high suspended sediment yield for the study period, over the last four decades [Dadson et al., 2003] and when compared to its mountain headwaters in the Wulu and Yenping catchments (Table 1 and Figure 4b). This may relate to active tectonic deformation of Pleistocene-Recent sediments in the Longitudinal Valley [Ho, 1986]. While the Wulu and Yenping mountain tributaries are located upstream (Figure 4b), the Peinan trunk river has cut into these recently uplifted, poorly consolidated sediments which contain POCnon-fossil [Shyu et al., 2006; Ramsey et al., 2007]. Supply of clastic sediment and POCnon-fossil from these deposits provides a mechanism to enhance fluvial POCnon-fossil concentration across all Qw (Figure 3b) and increase both the SSY and POCnon-fossil yield. Cannibalism of young, uplifted foreland deposits may be an important mechanism by which POCnon-fossilis re-mobilized in larger fluvial systems exiting active mountain belts [Bouchez et al., 2010; Galy and Eglinton, 2011].
5.3. Export of POCnon-fossil From Subtropical Mountain Forest
 The climatic (Figures 1b and 3) and geomorphic factors (Figure 5) that influence transport of POCnon-fossil in Taiwan's mountain rivers also affect their clastic load [Dietrich et al., 2003; Dadson et al., 2003; Hicks et al., 2004a; Galewsky et al., 2006; Kao and Milliman, 2008]. As a result, a strong positive relationship exists between POCnon-fossil yield and suspended sediment yield over two orders of magnitude in this mountain belt (Figure 4a). The data show no evidence for dilution of POCnon-fossil yields at very high physical erosion rates. The average rate of POCnon-fossil transfer of 21 ± 10 tC km−2 yr−1 represents an export of 0.12 ± 0.08% yr−1 of the total organic carbon stock in vegetation and soil, of 11 ± 5 × 103 tC km−2 and 7 ± 2 × 103 tC km−2, respectively [Chang et al., 2006; West et al., 2011]. These export rates are high when compared to rates of geomorphic disturbance in mountain forest. In the western Southern Alps, New Zealand, bedrock landslides disturb forested surfaces at a rate 0.03% yr−1 [Hilton et al., 2011b] and in Central America, disturbance rates are 10 times lower [Restrepo and Alvarez, 2006]. However, the POCnon-fossilexport rates here are likely to include important input from non-bedrock landslide inputs (overland flow, mechanical attrition of CWD) as previously discussed.
 The fluvial POCnon-fossil export from the mountain forest has important implications for carbon cycling at the regional scale. In the absence of other output fluxes (e.g., respiration), it sets a bound on the amount of time available for organic matter to age in the landscape (τnon-fossil, yr). At a depletion-rate of 0.12 ± 0.08% yr−1, physical erosion sets a timescale of 830 ± 530 yr for the aging of the organic carbon stock in vegetation and soil. Across Taiwan, the maximum τnon-fossil imposed by physical erosion is ∼15,000 yr in the Hsiukuluan River (Figure 6). Given the dominant role of respiration to carbon loss in terrestrial ecosystems, the estimates of τnon-fossil are not directly comparable to estimates of residence time in vegetation and soil. These account for all input and output fluxes and recognize different pools of carbon which turnover at different rates [Trumbore, 1993]. However, the limit on biomass aging set by POCnon-fossilexport is consistent with the range of conventional radiocarbon ages of surface soils (A-E Horizons) in Taiwan, which reach a maximum of 4169 yr [Hilton et al., 2008a] with the majority falling between 340 and 1540 yr (Figure 6).
 The data from Taiwan suggest that suspended sediment yields of ∼3000–4000 t km−2 yr−1 (physical erosion rates of ∼1–2 mm yr−1 with sediment density of 2.5 t m−3) can limit τnon-fossil to ∼8000 yr (Figure 6). Thus, it appears that even modest rates of physical erosion can reduce or even eliminate the potential for very long timescales (>10,000 yr) available for pools of organic matter in soils to age, regardless of their respiration rate [Trumbore, 1993; Torn et al., 1997]. POCnon-fossil export thus plays an important role in montane ecosystem turnover, likely to promote young sections of forest where net productivity is most efficient [Restrepo et al., 2009] and inhibit ecosystem retrogression [Wardle et al., 2004; Peltzer et al., 2010]. Physical erosion rates of 1–2 mm yr−1 are exceeded in many mountain belts [Galy and France-Lanord, 2001; Dadson et al., 2003; Hicks et al., 2004b; Gabet et al., 2008; Milliman and Farnsworth, 2011] suggesting that erosion may limit τnon-fossil in mountain forest at the global scale. At very high erosion rates of >10 mm yr−1, the physical processes impose a timescale for aging (Figure 6) which encroaches on the centennial rates of turnover in vegetation and components of soil organic carbon [Trumbore, 1993; Torn et al., 1997]. Clearly, the findings here demonstrate that the impact of rapid geomorphic process rates on nutrient and carbon cycling in mountain forests warrants further assessment.
5.4. Wider Implications for the Carbon Cycle
 The erosion and export of POCnon-fossil by mountain rivers represents a lateral flux of recently fixed atmospheric CO2 and its fate is important for our understanding of the global carbon cycle [Berner, 1982; Hayes et al., 1999]. If this material is buried in sedimentary deposits while the POCnon-fossil is replaced by new primary productivity on land, then this transfer represents a net sink of atmospheric CO2. Efficient burial of POCnon-fossil offshore Taiwan may be driven by the very high suspended sediment loads of the mountain rivers which deliver ∼380 × 106 t yr−1 to the ocean [Dadson et al., 2003], causing rapid accumulation rates in depocenters, a first order control on organic carbon burial efficiency [Canfield, 1994; Galy et al., 2007]. Hyperpycnal river plumes, arising when SSC>40 g L−1 at the river mouth [Mulder and Syvitski, 1995], can trigger turbidity currents which are also thought to play an important role by rapidly delivering POCnon-fossil carried by floodwaters (Figure 1b) to deep marine sediments [Dadson et al., 2005; Kao et al., 2006; Nakajima, 2006; Saller et al., 2006; Hilton et al., 2008a]. While the fate of POCnon-fossil remains to be fully assessed, it seems likely that a large proportion of the 0.5 ± 0.2 × 106 tC yr−1 of POCnon-fossil delivered to the oceans from Taiwan is buried.
 The significance of the transfer of POCnon-fossilfrom Taiwan to the ocean is evident from comparison to a well-studied source-to-sink region from the Himalayan mountain belt to Bay of Bengal. There, an estimated 3.7 × 106 tC yr−1 of POCnon-fossilis delivered by the Ganga-Brahmaputra rivers and sequestered from a continental source region ∼50 times larger than Taiwan [Galy et al., 2007]. The conservative estimate of POCnon-fossil flux from the small mountain island represents ∼15% of this value and ∼1% of the estimated total terrestrial organic carbon burial in the oceans [Schlünz and Schneider, 2000]. Evidently, mountain islands are important not only for the erosion and transfer of POCfossil [Blair et al., 2003; Leithold et al., 2006; Kao et al., 2008; Hilton et al., 2011a], but also in the transfer of carbon recently fixed from atmospheric CO2.
 Our data suggest that, for a constant set of geomorphic conditions, the fluvial transfer of POCnon-fossil from mountain catchments is driven by climate (Figure 3) through the activation of erosion and transport processes during heavy rainfall (Figure 1b). A move to a wetter, stormier climate over mountain forest should enhance the erosional export of POCnon-fossil. In settings with strong coupling between depositional sinks and terrestrial inputs [e.g., Leithold and Hope, 1999; Kao et al., 2006] this offers a feedback in the Earth System, whereby climate modifies rates of carbon sequestration through erosion and burial of POCnon-fossil [e.g., Hilton et al., 2008a]. In addition, the data from Taiwan suggest that this carbon transfer is moderated by the catchment geomorphology (Figures 4a and 5). Rapid rates of plate convergence and the uplift of competent metamorphic rocks set prime conditions for the rapid erosion and fluvial export of POCnon-fossil concomitant with large amounts of clastic sediment [Galy et al., 2007; Hilton et al., 2008a, 2008b]. On orogenic timescales, this implies a tectonic forcing of the carbon cycle which may lead to net changes in the size of the organic carbon reservoir and influence atmospheric greenhouse-gas concentrations [Derry and France-Lanord, 1996; France-Lanord and Derry, 1997; Hayes et al., 1999] via a carbon transfer that is sensitive to climatic conditions [cf. West et al., 2005].
 This work was supported by The Cambridge Trusts and National Taiwan University. Suspended sediments were collected by the 1st, 3rd, 4th, 6th, 7th, 8th, and 9th regional offices of the Water Resources Agency, Ministry of Economic Affairs, Taiwan. We thank Taroko National Park and M. C. Chen for additional access to research sites, and A. J. West, J. Gaillardet, D. M. Milledge, J. Wainwright and A. L. Densmore for useful discussions during manuscript preparation. E. T. Sundquist and two anonymous referees are thanked for their insightful comments which improved the manuscript.