Climate and topography control millennial-scale mountain erosion, but their relative impacts remain matters of debate. Conflicting results may be explained by the influence of the erosion threshold and daily variability of runoff on long-term erosion. However, there is a lack of data documenting these erosion factors. Here we report suspended-load measurements, derived decennial erosion rates, and 10Be-derived millennial erosion rates along an exceptional climatic gradient in the Andes of central Chile. Both erosion rates (decennial and millenial) follow the same latitudinal trend, and peak where the climate is temperate (mean runoff ∼500 mm yr−1). Both decennial and millennial erosion rates increase nonlinearly with slope toward a threshold of ∼0.55 m/m. The comparison of these erosion rates shows that the contribution of rare and strong erosive events to millennial erosion increases from 0% in the humid zone to more than 90% in the arid zone. Our data confirm the primary role of slope as erosion control even under contrasting climates and support the view that the influence of runoff variability on millennial erosion rates increases with aridity.


Determining the functional relationship between millennial mountainous catchment-scale erosion rates and topographic and climatic parameters has two implications. It is fundamental for evaluating the magnitude of climate changes in the past, through the analysis of terrigenous fractions in sediment records (Lamy et al., 2010), and for quantifying the combined strength of climate and tectonics in orogens (Whipple, 2009). There has been no general agreement on this relationship. A few data sets from the past decade documented that the millennial hillslope erosion rate increases nonlinearly with slope, toward a slope threshold close to 0.6 m/m, predicted from slope stability criteria (e.g., Binnie et al., 2007; Roering et al., 2007; Ouimet et al., 2009). The relationship between the erosion rate and climate is more controversial. Mean precipitation has been found to either significantly correlate with the millennial erosion of mountainous catchments (Reiners et al., 2003; Moon et al., 2011), or not correlate (e.g., Riebe et al., 2001). The contribution of large erosive events is also debated (Baker, 1977; Tucker, 2004; Lague et al., 2005; Molnar et al., 2006). Arid climates have higher variability than humid climates; maximum floods in arid areas are relatively larger than in humid ones (Molnar et al., 2006). Theoretical studies predict that large flood contributions to catchment erosion increase with climate variability (i.e., with aridity), especially as erosion depends on a threshold (Tucker, 2004; Lague et al., 2005; DiBiase and Whipple, 2011). However, few data are available to support the primary role of climate variability on erosion. Dadson et al. (2003) showed a clear correlation between erosion rates and discharge variability in the actively uplifting range of Taiwan. In a study of 32 small catchments in Idaho (United States), Kirchner et al. (2001) found that the millennial erosion rates are larger than decennial ones; they interpreted them as field evidence that large and rare erosive events are main drivers of erosion. In contrast, Summerfield and Hulton (1994) found that the erosion rate correlates with slope but not with discharge variability in major world drainage basins. In this paper we analyze the contributions of both factors using erosion rate data over different time periods and along a strong climatic gradient in the Chilean Andes.


The Andes in central Chile between 27°S and 40°S trend north-south. The northern part of this region is arid (rainfall <10 cm yr−1) and the southern part, influenced by southern westerly winds, is humid (>2 m yr−1). The transition between both regions occurs abruptly between ∼32°S and ∼35°S (Fig. 1). We selected 26 catchments with outlets located at the foot of the main Cordillera (Fig. 1). Decennial catchment-mean erosion rates of these basins (Table DR1 in the GSA Data Repository1) were estimated using daily suspended load time series at gauging stations monitored by the Chilean Direccion General de Aguas (from 3 to 40 yr; Pepin et al., 2010; see the Data Repository).

In order to estimate millennial erosion rates, we sampled river sand at the outlet of 12 catchments (Fig. 1), some of them replicated by sampling different nearby bars. We completed a total of 20 10Be concentration analyses of quartz, from which mean catchment erosion rates were calculated (e.g., Kirchner et al., 2001; see the Data Repository). Erosion rate values integrate possibly several thousand years, a time span that depends inversely on the calculated erosion rate. This characteristic time (τ) is defined as the time necessary to erode a strip of thickness equivalent to the particle mean free path in rocks (see the Data Repository).

Decennial and millennial data are not available for every catchment. In order to compare latitudinal variations of millennial and decennial erosion rates, we computed their mean values in bins of 0.5°. In order to take into account the difference in catchment areas, and thus the relative spatial contribution of each catchment in the bin, we weighted the catchment erosion rates by the relative catchment area in the bin.


Runoff increases from 6 mm yr−1 in the arid northern part of the study region to 2551 mm yr−1 in the humid southern part (Fig. 1A). The vegetation cover follows closely the runoff variations, and increases from 0% to 26% of the catchment area. Catchment slope ranges between 0.24 and 0.57 m/m. It peaks at 29.5°S and 33.5°S, and decreases both northward and southward (Fig. 1A). Decennial erosion rates follow a bell curve with a maximum (0.28 mm yr−1) near 33.5°S. This value is two orders of magnitude larger than the ones obtained in the arid north and in the humid south (Fig. 1B). Decennial erosion rates are the greatest in the region where runoff is moderate, slope is large, and vegetation cover is sparse. South of 34°S (runoff > 500 mm yr−1), erosion rates are inversely correlated with runoff and vegetation and positively correlated with slope.

The maximum millennial erosion rates are between 33°S and 34°S, but show another peak near 29.5°S (Fig. 1B; Table DR2). Farther south, both erosion rates have a similar trend and they match south of 34°S. The northernmost millennial erosion rates (∼0.02 mm yr−1) are comparable to cosmogenic nuclide–derived estimates in the desertic Rio Lluta catchment (18.5°S) in northern Chile (Kober et al., 2009). Near 33°S, previously published cosmogenic nuclide analyses of the Maipo catchment fluvial sediment yield a similar estimated erosion rate of 0.3 mm yr−1 (Antinao and Gosse, 2009). Concordant repeated samples in some catchments show that the results are robust and that the variation from one catchment to the other is significant. Millennial erosion rates follow closely the slope variation (Fig. 1). Both decennial and millennial erosion rates increase nonlinearly with slope (Fig. 2A), supporting nonlinear diffusion models with critical slopes of ∼0.53–0.55 (Roering et al., 2007). Only catchments with runoff <0.8 mm yr−1 present a positive correlation between erosion rate and runoff (Fig. 2B).

Figure 3 shows the latitudinal ratio between millennial and decennial data by bins of 0.5°. This ratio decreases from ∼10–15 in the most arid region to ∼1 in the temperate region (Fig. 3).


The similar latitudinal gradient followed by the decennial and millennial erosion rates suggests that the role of control factors has been similar over the past ∼10 k.y. In particular, slope appears to exert a primary control, even for catchments with contrasting mean runoffs. Both millennial and decennial erosion rates tend to increase sharply when the slope tends toward ∼0.55 (∼tan30°) (Fig. 2A), a similar slope threshold predicted from slope stability criterion (Roering et al., 2007). The relationship between erosion rate and runoff varies considerably depending on the runoff range (Fig. 2B). This suggests that evidencing the role of regional variations of mean runoff requires relatively arid climate. This finding may be consistent with the theoretical prediction of DiBiase and Whipple’s (2011) model, suggesting that erosion should peak for runoff of ∼200–400 mm yr−1, which is close to the value observed here (∼500 mm yr−1). Further comparison between this model and our data is difficult because the erosion rate is probably not uniform within the studied catchments (Farías et al., 2008; Rehak et al., 2010; Pepin et al., 2010; Aguilar et al., 2011). A significant decrease in the proportion of granitoids compared to volcanic and volcanodetritic rocks between 31°S and 34°S may also contribute to the erosion rate increase in this region, but it cannot explain the overall patterns that we observe (Fig. DR1 in the Data Repository). The vegetation increase may partly explain that erosion decreases south of 34°S, consistent with the model proposed by Langbein and Schumm (1958).

The progressive convergence of decennial and millennial erosion rates toward the more humid climatic conditions could be explained by the following: (1) a decreasing contribution of bedload southward, (2) a more humid climate during the early Holocene in the north than today, and (3) a decreased contribution of large floods southward. In the arid northern part of the study region, floods could mobilize a larger fraction of the bedload than in the south. Thus, the suspended load could underestimate significantly the total load. Databases suggest that the coarse sediments (gravels to cobbles) covering the sampled river beds are indicative of rivers carrying most sediment as suspended load (Turowski et al., 2010). Thus, it is unlikely that the bedload fraction reaches >90% of the total load during the measured floods, which would be necessary to explain a factor of 10 difference between millennial and decennial erosion rates.

In the region near 30°S, terrestrial and marine proxies, as well as preserved fluvial sediments in the Rio Turbio catchment, converge toward a scenario marked by wetter conditions than today between 33 and 16–17 ka, and in the late Holocene. They indicate hyperarid conditions between 11 and 5 ka (Zech et al., 2008; Kaiser et al., 2008; Riquelme et al., 2011). The wetter initial period cannot explain the difference between decennial and millennial erosion rates because the millennial erosion rates that integrate over the past ∼30 k.y. are also closer to the decennial values (Fig. 1). Furthermore, the increase of humidity in the past 5 k.y. and the onset of modern El Niño manifestations at 5.3–5.5 ka (Vargas et al., 2006) make it unclear whether the catchment erosion rate has increased or decreased over the period of 10Be data integration. The injection of paraglacial sediments with eventually low 10Be concentrations cannot explain why the millennial erosion rates are 10 times larger than decennial values, because this would require that the paraglacial sediments account for >90% of the sampled sand (see the Data Repository). Farther south, near 34°S, a reconstruction of precipitation rates indicates a trend toward wetter conditions starting from 12 ka, with conditions similar to the current ones for the past ∼3–5 k.y. (Jenny et al., 2003; Lamy et al., 2010), despite modifications to the vegetation cover in the past centuries (Armesto et al., 2010). The periods of integration of 10Be-derived erosion rates in this region (∼3–5 k.y.) include this last stable period, which is consistent with the small difference between short-term and long-term erosion rates south of 33°S (Fig. 1).

The third and more probable explanation for the higher 10Be-derived value in the north is the lack of record of large floods (with high erosive power) at the gauging stations. In the arid and hyperarid northern-central Chile, dramatic erosion is triggered by large floods. Large floods occur every 20–100 yr (Ortlieb, 1994; Vargas et al., 2006), a recurrence time longer than the suspended load measurement period. The analysis of daily discharges at gauging stations shows that the range of discharge events larger than the mean discharge increases northward (Fig. 3B; Fig. DR2). Undersampling these events has a larger impact on estimated mean sediment discharge in the north than in the south. The proportion of sediment exported during the El Niño periods is another indication of the contribution of large floods: this proportion increases northward, suggesting that a larger fraction of the sediment exportation occurs during large and rare events in the north (Fig. 3B). These observations suggest that the progressive divergence of decennial and millennial erosion rates toward drier conditions documents the progressive contribution of unrecorded large water discharges to millennial erosion (Wolman and Miller, 1960; Baker, 1977). Based on the millennial to decennial erosion rates ratio, their contribution increases from ∼0 where the runoff is ∼600–1000 mm yr−1, to >90% in the north, where the runoff is <10 mm yr−1 (Fig. 3). The role of large floods could explain the weakest correlation between decennial and millennial erosion rates in the arid Rio Elqui catchment near 29.5°S (Fig. 1). This catchment is characterized by a significantly higher slope (Fig. 1; Table DR2). A larger range of precipitation contributes efficiently to its erosion, so that its millennial erosion rate is greater than in adjacent catchments. It is likely that the period of measurement at the corresponding gauging station does not include the full range of high discharges, explaining the small difference in decennial erosion rate with respect to the adjacent catchments.


Similar gradients of decennial and millennial erosion rates were observed along a strong climatic gradient. These data confirm the nonlinear relationship between erosion rate and slope even for different climates. This nonlinear relationship and possibly vegetation and lithology explain why the erosion rate may be independent of the mean annual runoff in the wet sector of the studied region. These data document the variable contribution of discharges larger than the mean discharge, which increases from ∼0% of the total erosion in humid zones to ∼90% in arid climates.

This study was funded by the L’Agence Nationale de la Recherche (ANR-06-JCJC0100) and the L’Institut de Recherche pour le Développement (IRD). It is also a contribution to FONDECYT (Fondo Nacional de Desarrollo Científico y Tecnológico) project 11085022. 10Be concentrations were obtained at the French ASTER AMS at Cerege. We thank R. Garreaud, S. Bonnet, and Y. Godderis for discussions, and F. von Blanckenburg and H. Wittmann for their support in setting up the 10Be laboratory. We also thank Greg Tucker, Fritz Schlunegger, and an anonymous reviewer for their constructive reviews.

1GSA Data Repository item 2013048, data tables, information about the calculation of erosion rates and erosion factors, and additional figures, is available online at www.geosociety.org/pubs/ft2013.htm, or on request from editing@geosociety.org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA.