Low-temperature thermochronologic signature of range-divide migration and breaching in the North Cascades

Climate is thought to have a global effect on the erosion and topography of mountain ranges (e.g., Montgomery et al., 2001; Champagnac et al., 2012; Herman et al., 2013). Many characteristics of mountain topography are expected to vary with climate (Willett, 1999), notably the position of range drainage divides. Physical (Bonnet, 2009) and numerical (Roe et al., 2003; Anders et al., 2008; Castelltort et al., 2012; Goren et al., 2013) simulations predict that an asymmetric orographic precipitation pattern generates asymmetric topography and migration of drainage divides toward the drier flanks of mountain ranges. As topography approaches geomorphic steady state, divide migration results in a progressive decline in slope on the wet side of an orogen and an increase in slope on its arid side. Such theoretical adjustment is difficult to document in natural systems due to the continuous erosion of the ridge crest during its migration. Divide migration is thought to occur by progressive, diffusive lateral slope retreat or in successive discrete jumps by capture of rivers across the drainage divide (Craw et al., 1999; Willett et al., 2014). Such river diversions should be difficult to achieve in deeply incised orogens (Bishop, 1995), but they can still occur in a variety of ways, such as groundwater capture (Brocard et al., 2011, 2012) or lake overflow (Riedel et al., 2007). Reconstructions of paleodrainage patterns generally rely on distal proxies, such as changes in the routing of river sediments sourced in specific areas (e.g., Craw et al., 1999; Stokes et al., 2002; Maher et al., 2007; Brocard et al., 2011; Prince et al., 2011; Andrews et al., 2012), and on phylogenetic models (e.g., Craw et al., 2007; Ruzzante et al., 2008; Zemlak et al., 2010). However, in certain circumstances, the incision of deep canyons associated with drainage reorganization can leave a detectable footprint in the record of low-temperature cooling ages (Flowers and Farley, 2012; Karlstrom et al., 2014).

During the Quaternary, the North Cascades (Washington State, northwest USA) were repeatedly occupied by local temperate glaciers that deeply sculpted the landscape (Mitchell and Montgomery, 2006). In addition, north of 48.5°N, the range was invaded by the southward-flowing Cordilleran ice sheet (Fig. 1; Clague, 1989; Hendy, 2009). The ice sheet blocked many north-draining valleys, initiating reversals of their drainages. Numerous deeply incised V-shaped gorges found throughout this glacial landscape have been interpreted as interfluves breached during the overspill of these ice-dammed proglacial lakes that had formed at the front of the advancing ice sheet (Mathews et al., 1968; Riedel et al., 2007). One of them, the Skagit River Gorge, is a 2-km-deep, V-shaped section of the Skagit River valley, a 240-km-long range-transverse valley crossing the North Cascades. Riedel et al. (2007) proposed that the gorge formed during the Pleistocene by westward overspill of a proglacial lake, and this connected the modern upper drainage of the Skagit River to its lower course. Incision of the interfluve down to the elevation of surrounding valley floors allowed the permanent rerouting of the upper drainage through this gorge and a definitive shift of the drainage divide 40 km to the east (Fig. 1).

In this study, we use low-temperature thermochronology to test this hypothesis and date the drainage adjustment. In the North Cascades, climate has been shown to strongly impact erosion patterns (Reiners et al., 2003). Orographic precipitation appears to drive enhanced erosion of the wet, windward side of the range over both intermediate- (>10⁴ yr) and long-term (>10⁶–10⁷ yr) time scales, according to cosmogenic isotopes and apatite (U-Th)/He studies respectively (Fig. 1; Moon et al., 2011; Reiners et al., 2003). We present new low-temperature thermochronology (apatite [U-Th]/He and ⁴He/³He) data collected along the Skagit River valley across the North Cascades and along two elevation profiles in and upstream (eastward) of the Skagit River Gorge (Fig. 1). The million-year denudation pattern is well correlated with the orographic precipitation pattern across the range. However, unlike in the southern part of the range at 47.5°N, where the transition from the young cooling ages of the western flank to the older ages of the eastern flank coincides with the position of the modern drainage divide, the transition here coincides with the Skagit Gorge location, supporting that the gorge corresponds to the recent breaching of the range drainage divide. Age-elevation profiles along the Skagit Gorge, as well as ⁴He/³He data for the gorge floor, date the onset of gorge incision during the last 2 m.y. Finally, we propose that breaching of the preglaciation drainage divide, although triggered by glacial lake overspill, is fundamentally a long-term consequence of more rapid erosion of the wet side of the range. This breaching catalyzed drainage reorganization and forced long-term migration of the drainage divide toward the dry side of the range.

The North Cascades range is the southern end of the >1200-km-long, 150-km-wide Coast Range that extends from SE Alaska through British Columbia (Canada) to Washington State (USA). Bedrock in the North Cascades is dominantly composed of high-grade metamorphic rocks intruded by Late Cretaceous to Eocene calc-alkaline plutons (Whitney et al., 1999; Miller et al., 2009). The North Cascades topography, which peaks at 3000 m, is supported by a buoyant crustal root; two-dimensional (2-D) seismic velocity profiles indicate crustal thicknesses of 35–40 km under central Washington (Schultz and Crosson, 1996). The North Cascades range has not been affected by significant internal deformation over the last 30–40 m.y.

The North Cascades currently block moisture influxes from the Pacific Ocean in the west. As a result, precipitation peaks at 5 m/yr over the western, windward side of the range but plummets down to 0.5 m/yr in its rain shadow to the east (Figs. 1C and 1D). Paleoclimate data indicate that this strong orographic precipitation gradient developed between 15 and 8 m.y. ago (Takeuchi and Larson, 2005).

The range was extensively glaciated during the Quaternary (Mitchell and Montgomery, 2006). Locally sourced mountain glaciers spread down the main valleys (Hendy, 2009). In addition, since at least the late Pleistocene (<600 k.y.), the Cordilleran continental ice sheet, centered over British Columbia, has repeatedly spread southward over the northern part of the Washington Cascades (Fig. 1; Waitt and Thorson, 1983; Clague, 1989). Deposits in the Puget Lowland indicate that the Cordilleran ice sheet has spread over the western piedmont of the Cascades at least six times since the middle Pleistocene (Easterbrook et al., 1988; Booth et al., 2004). However, the marine isotope record suggests that these are but a fraction of the total ice sheet advances in the region during the last 2.5 m.y. (Booth et al., 2004). At its largest, the Cordilleran ice sheet reached elevations of more than 2000 m over much of its extent and a few nunataks protruded above it (Wilson et al., 1958; Clague, 1989).

Glacial erosion strongly shaped the topography of the Washington Cascades (Mitchell and Montgomery, 2006). Aside from Quaternary stratovolcanoes that stand above the regional topography, maximum and mean altitudes increase systematically from west to east (Figs. 1C and 1D), defining an inclined surface that correlates with the change in elevation of the local glacier equilibrium line altitude (ELA). The ELA rises 12 m/km toward the east (Porter, 1977) due to decreasing precipitation (e.g., Kuhn, 1989). In the Cascades, peaks and ridges stand on average 600 m above the ELA. This correlation suggests that glacial erosion has limited the height of the range, a process commonly referred to as the “glacial buzzsaw” (e.g., Brozović et al., 1997; Mitchell and Montgomery, 2006; Egholm et al., 2009).

The Skagit River crosscuts the highest ridges of the range (Fig. 1) along the 2000-m-deep, V-shaped Skagit Gorge, which connects the upper and lower U-shaped segments of the river (Figs. 2A and 2B). It has been proposed that the Skagit Gorge is a recent topographic feature resulting from the breaching of a paleodivide separating the upper and lower Skagit Rivers (Riedel et al., 2007). Originally, the paleo–upper Skagit River drained northward along a valley now occupied by the Ross Lake reservoir, whereas the current lower Skagit, already draining to the west into the Puget Sound, had its headwaters located in the vicinity of the Skagit Gorge (Fig. 1). Bedrock benches are found sporadically preserved along the upper Skagit River some 300–400 m above the modern river (Mathews, 1968; Haugerud, 1985). They have been interpreted as the relics of the U-shaped valley of the upper Skagit River before its connection to the lower Skagit River (Riedel et al., 2007). In particular, a 5-km-long bench above the southwest shore of Ross Lake dips in a direction opposite to the present-day flow direction of the Skagit River, suggesting flow reversal. In addition, the modern pass between the upper, south-flowing branch of the Skagit River and the upper, north-flowing branch of the Fraser River is a nice example of wind gap, which localizes the paleoflow of the upper Skagit (Fig. 1).

To explain this flow reversal, Riedel et al. (2007) proposed that a vast lake formed along the ancestral upper Skagit when the Cordilleran ice sheet invaded its lower reaches. The ice dam grew and eventually overtopped a pass located at the current location of the Skagit Gorge. The lake waters spilled over the pass and cascaded down the west flank of the range, triggering rapid incision of the pass and drawdown of the lake. In effect, this event displaced the range drainage divide ∼40 km further east (Fig. 1). Because lake overspill is an efficient way to overcome high-standing divides in deeply incised landscapes, it has been called upon to explain the abundance of valleys with a high degree of interconnectivity in ranges that were formerly occupied by ice sheets, such as in British Columbia (Kerr, 1936; Fulton, 1969), Scotland (Dury, 1953), and Scandinavia (Seuss, 1888).

The modeled unit stream power is not a direct measurement of river incision rate, but it is useful to estimate along-stream variations in erosive potential (Finlayson et al., 2002), assuming homogeneous bedrock erodibility and bed roughness. In the Skagit Gorge, the modeled stream power is about four times higher than along the rest of the river (Fig. 2D). This peak does not correlate with any obvious change in bedrock erodibility within the gorge, indicating that the knick zone likely experiences faster incision. No significant differential uplift has taken place along the gorge over the last ∼40 m.y., so faster incision cannot be a consequence of enhanced rock uplift. We conclude that it is a transient erosive signal, the nature of which is discussed in the discussion section.

Two different thermochronometers were used to track the denudation history of the Skagit Gorge and its surroundings: the apatite (U-Th)/He (AHe) and the apatite ⁴He/³He systems. The AHe system records rock cooling through temperatures of 80–60 °C (Farley, 2000; Shuster et al., 2006) and thus typically records bedrock exhumation through the upper 2–4 km of the crust. Apatite ⁴He/³He thermochronology is based on the mapping of the radiogenic ⁴He distribution within single apatite crystals using proton-induced ³He production and step-degassing experiments (Shuster and Farley, 2004; Shuster et al., 2004). This technique reveals rock cooling history below ∼90 °C, the temperature at which ⁴He starts to accumulate in apatite crystals.

Twenty-three samples were collected along the Skagit River valley, across the North Cascades, and along two elevation profiles, one located in the Skagit River Gorge and the other one above Ross Lake (Fig. 1). The AHe analyses were performed at the University of Arizona, and the ⁴He/³He analyses where conducted at the Berkeley Geochronology Center. Between two and seven replicates of each samples were measured, with more replicates being measured along the elevation profiles. The results were used for quantitative thermal modeling. Most samples reproduced very well, and only a few grains were excluded from the calculation of average AHe ages. The sample-averaged AHe ages, corrected for α-ejection, are summarized in Table 1. The details of the applied corrections are provided as supplementary material (Tables DR1 and DR2 for AHe, and Table DR3 for apatite ⁴He/³He).¹

AHe ages were obtained on 11 bedrock samples collected over the width of the North Cascades orogen at the latitude of the Skagit River Gorge (48.7°N). Cooling ages range from 40 ± 0.8 Ma to 2.0 ± 0.8 Ma (Fig. 1; Table 1). Like further south at 47.5°N, the AHe ages display a U-shaped pattern, and the location of the youngest ages correlates with the maximum in modern precipitation (Fig. 1C). However, unlike at 47.5°N, where the transition from the young ages of the western flank to the older ages of the eastern flank coincides with the position of the modern drainage divide, the transition here coincides with the Skagit Gorge, supporting the hypothesis that the Skagit Gorge results from the incision of the paleo–drainage divide.

The Skagit Gorge (SG) profile and the Ross Lake (RL) profile were sampled ∼15 km apart. The Ross Lake profile is located on the flank of the proposed ancestral valley of the north-flowing Skagit River (Fig. 2). Along both profiles, AHe ages correlate positively with elevation, yielding apparent long-term denudation rates of 0.1–0.2 km/m.y. between ca. 16 and 3 Ma (Fig. 3). However, the Skagit Gorge AHe ages are up to ∼4 m.y. younger than those of the Ross Lake profile, implying faster recent denudation within the Skagit Gorge. Using the present-day geothermal gradient of ∼25 °C/km (Blackwell et al., 1990), and extrapolating the AHe age-elevation trend to the present-day depth of the AHe closure temperature (∼70 °C), the 3.5 Ma age at the bottom of the Skagit Gorge implies an increase of the denudation rate after 3 Ma in the Skagit Gorge at an average rate of ∼900 m/m.y., while a steady and slower denudation rate (∼200 m/m.y.) may have persisted in the Ross Lake area after 11 Ma.

To document this late-stage cooling history in more detail, we prepared samples from both profiles for ⁴He/³He thermochronology. Only euhedral crystals free of visible mineral inclusions can be used for this technique (Shuster and Farley, 2004; Shuster et al., 2004). Only one crystal from the Skagit Gorge profile presented the required quality for ⁴He/³He analysis; other apatites were polluted by minute zircon inclusions. Along the Ross Lake profile, all apatite crystals were either broken or noneuhedral. Valley-floor sample SG02-1 (345 m elevation) displays a diffusive ⁴He-distribution pattern (Fig. 4A), which is typically produced by slow cooling through the partial retention zone and subsequent rapid cooling toward the topographic surface (Shuster et al., 2004). This result tends to support the analysis of the elevation-age trend; a more quantitative analysis of the cooling histories is presented next.

To interpret the ⁴He/³He results, we used a numerical model that predicts and inverts ⁴He distributions using a radiation-damage and annealing model to quantify He-diffusion kinetics in apatite (see Methods and supplementary information [see ^{footnote 1}]; Flowers et al., 2009). The model compares observed data to randomly generated temperature-time (T-t) histories; it explores a wide range of cooling histories and selects acceptable scenarios using a misfit-based evaluation. The misfit is defined as the mean of squared residuals weighted by the mean uncertainty in the ratio measurement (Schildgen et al., 2010). The input parameters include the characteristics of the analyzed crystal (apatite size, U and Th content) and its cooling history. The model carries the major assumption that U and Th are uniformly distributed within the analyzed apatite grain; this assumption is only challenged when an individual crystal possesses large concentric contrasts in the local concentration of U or Th (Farley et al., 2010). All cooling histories began at 150 °C and ended after 15 m.y. of simulation at the modern mean surface temperature (∼5 °C). The model was able to reproduce the observed ⁴He/³He diffusion spectrum of apatite SG02-1 (Fig. 4A); it therefore adequately describes the He behavior in this apatite grain (Farley et al., 2011). The best-fit models produced gradual cooling through the AHe partial retention zone from ca. 10 to 2 Ma and faster cooling from ∼50 °C to surface temperatures afterward (Fig. 4B). The initial slow cooling rate is poorly constrained between 0.6 and 4.5 °C/m.y., whereas the latter cooling rate is more tightly constrained between 20 and 25 °C/m.y.. The analysis of the ⁴He/³He profile requires a marked increase in cooling rate, consistent with the recent and rapid gorge incision scenario.

The thermal history of the AHe age-elevation profiles was also simulated, this time using a transdimensional Markov chain Monte-Carlo inversion approach (Gallagher, 2012; see details in the Data Repository). The strength of this approach is its ability to use the spatial (elevation) relationship between samples to calculate T-t histories using all the samples in a profile. The expected models are presented for alpha-recoil damage models from Flowers et al. (2009). The modeling initiates with randomly chosen T-t paths and sets of kinetic parameters, for which a probability that the model fits the data is calculated. The parameter values are then slightly offset, and the probability of fitting the data is recalculated and compared to the initial model. The model with the highest probability is retained. This procedure is repeated 200,000 times (the number of iterations being chosen by the user), providing a large collection of models with their associated probabilities that allow model statistics to be calculated. A full explanation of the modeling procedure is provided in Gallagher (2012).

The Skagit Gorge and Ross Lake profiles were inverted separately with identical input parameters; cooling paths were randomly generated in a parameter space stretching 25 Ma to 0 Ma and by temperatures between 180 °C and 0 °C. No other constraints were imposed, allowing a full exploration of the range of statistically valuable models. The resulting best-fit thermal histories for the whole profiles are shown on Figure 5A and Figure 5B, together with the best-fit thermal histories of individual low-elevation samples (SG02-1 and RL04-1). In all cases, predicted T-t paths share a similar constant cooling rate of ∼5 °C/m.y. from 15 to 3 Ma.

These inversion results provide accurate temporal evolution of cooling rates but cannot convert cooling rates into denudation rates. For this, it would be necessary to simulate the variation of the geothermal gradient as function of topography and its evolution through time (Braun et al., 2012). We consider that the geomorphological record is not sufficient to accurately reconstruct the paleotopography before the divide breaching and simulate its effect on the thermal structure. We therefore only propose an estimation of the denudation rates using a constant geothermal gradient of ∼25 °C/km, based on heat-flow data mapping collected across the North Cascades (Blackwell et al., 1990). The cooling rate of ∼5 °C/m.y. recorded by both elevation profiles then converts into a long-term denudation rate of ∼200 m/m.y., consistent with the denudation rate inferred from the age-elevation trends. Slow cooling continues until today along the Ross Lake profile, whereas rapid cooling occurs after ca. 3 Ma in the lowermost samples of the Skagit Gorge profile at a rate of ∼20 °C/m.y., as expected from the age-elevation trend.

To gain more insight on the recent episode of accelerated cooling, we added the ⁴He/³He spectrum of the gorge floor sample SG02-1 to the Skagit Gorge profile (Fig. 5C). The inversion confirms the need for a phase of substantial local cooling during the past 2 m.y. within the gorge, while maintaining steady cooling at higher elevation. Differential cooling between high- and low-elevation samples reaches ∼40 °C. Importantly, the cooling pattern confirms the increase in denudation rate, but, in addition, it shows that faster cooling is restricted to the lowermost samples, implying relief growth through valley incision (e.g., Valla et al., 2011), rather than a widespread increase in erosion, with rapid incision of the gorge at a rate of ∼800 m/m.y. during the past 2 m.y., superimposed onto a background denudation rate of ∼200 m/m.y.

The U-shaped pattern in the AHe ages that we observe at latitude 48.7°N is not correlated with the modern drainage divide but with the position of the Skagit Gorge, the modern divide being located ∼40 km further east. However, the gorge crosscuts a topographic ridge that further south becomes the range drainage divide (Fig. 1). The denudation pattern and the map trajectory of the drainage divide together support Riedel et al.’s (2007) contention that the Upper Skagit basin was once part of the eastern flank of the range and was integrated to the west-flowing drainage through the Skagit Gorge (Fig. 5). The Skagit Gorge resulted from the incision of a low saddle that became the outflow channel of a proglacial lake that occupied the lower Skagit valley, bound to the north by the Cordilleran ice sheet.

According to this model, divide breaching and subsequent incision of the Skagit Gorge occurred during the Quaternary. This is confirmed by the thermochronology data in the Skagit Gorge and its surroundings. The cooling paths inferred from our age-elevation profiles indicate that between 15 and ca. 2 Ma, denudation of this part of the range was steady, uniform, and moderate at ∼200 m/m.y. During the past ∼2 m.y., substantial denudation occurred within the Skagit Gorge, while slow denudation continued in the Upper Skagit basin. The million-year resolution of thermochronological data does not allow us to date more precisely the initiation of gorge incision, but the incision of the Skagit Gorge could have started with the first arrival of the Cordilleran ice sheet in the area in Pleistocene time (Easterbrook et al., 1988; Booth et al., 2004). The distribution of stream power along the modern Skagit River further suggests that the period of rapid incision following the rerouting of the Skagit River across this ridge is still active. The thermal inversion of the Skagit Gorge profile also suggests that slow denudation persists at high elevations on the flanks of the gorge, resulting in a strong increase in local relief. The shoulders located at elevations at ∼1800 m on either side of the gorge could be the remnants of the preexisting topography. In this case and considering that the shoulders are not eroding, the amount of incision since the overflow would reach a minimum of ∼1500 m. Assuming a constant geothermal gradient of 25 °C/km, such incision would produce a maximum differential cooling of 37.5 °C between high- and low-elevation samples, close to the observed difference of 40 °C.

Reiners et al. (2003) demonstrated that at 47.5°N, the minimum in AHe ages correlates spatially with the maximum intensity of precipitation, with the younger ages indicating greater denudation on the wet side of the orogen (Figs. 1 and 2). Here at 48.7°N, this U-shaped pattern is also observed, but it is a little less pronounced, particularly in the west, where young AHe ages are obtained close to Mount Baker volcano (Haugerud and Tabor, 2009; Hildreth et al., 2003). These samples may have experienced magmatic heating, with partial or total age resetting. Heat-flow data collected across the Washington Cascades show that the regional geothermal gradient is significantly perturbed as far as ∼10 km from active volcanic centers (Blackwell et al., 1990). Apart from this local perturbation, the similarity of the patterns at both latitudes leads us to conclude that the drainage divide used to pass through the Skagit River Gorge.

Numerical simulations of the impact of a migrating ridge on low-thermochronological ages (Olen et al., 2012) show broadly similar results to what we observe in our study area. Indeed, models predict synthetic AHe age-elevation relationships with similar slopes on either side of the migrating ridge, but younger on the retreating (windward) flank. Even under uniform rock uplift rate, the retreating flank of the ridge displays younger ages at a given elevation due to (1) the deformation of the isotherms below the topographic surface, which reduces the depth of the AHe closure temperature below the divide, and (2) focused erosion and increase in local relief in the wake of the migrating divide. Therefore, the ∼4 m.y. shift in cooling ages from the Skagit Gorge to Ross Lake (Fig. 3) supports the contention that the Skagit Gorge was the main drainage divide before its breaching. Furthermore, comparison of our data set with published numerical simulations suggests that the divide was already migrating eastward before the breaching, but more data westward of the Skagit Gorge are necessary to confirm this hypothesis.

The breaching of the drainage divide by the Skagit River is not an isolated example. North of the Skagit River, the modern drainage divide is located east of the main summits; major rivers, such as the Fraser and Skeena Rivers, drain from east to west across the crest of the range. Drainage diversion and divide breaching by the Fraser River have been recently dated to <1.17 Ma based on the age of a lava flow that fills the abandoned precapture valley (Andrews et al., 2012). Against this backdrop, the diversion of the Skagit and Fraser Rivers no longer appears to be a random derangement of the drainage network during the invasion of the ice sheet, but rather a systematic westward breaching of the range divide along the entire range. We propose that the eastward migration of this divide during the Quaternary finds its place in a longer-term trend of migration that accelerated with proglacial lake overspill. This migration is fueled over the long term by the asymmetric pattern of orographic precipitation that has characterized the range since the middle Miocene. The precipitation gradient has focused high long-term erosion rates on the humid side of the range and produced the formation of an asymmetric and inclined topography (Fig. 6A). This asymmetry explains the systematic rerouting of streams toward the humid side and the Pacific Ocean (Fig. 6). We have shown that, in the North Cascades, the retreat of the drainage divide, rather than progressive, occurred in discrete jumps due to diversion of rivers across the drainage divide. River diversion is very efficient in shifting the divide because it only requires the local erosion of an interfluve to produce instantaneous divide migration over a large distance. Similar stepwise breaching and leeward migration of the range drainage divide have been observed in the Patagonian Andes (Montgomery et al., 2001), where it also occurred during Quaternary times (Ruzzante et al., 2008; Zemlak et al., 2010). The Patagonian Andes also display an asymmetric long-term erosion pattern that correlates with orographic precipitation and glaciations (Thomson et al., 2010).

This study combined geomorphological analyses, inverse modeling of thermochronological age-elevation profiles, and apatite ⁴He/³He thermochronology to evaluate the control exerted by precipitation on mountain range denudation rates and drainage divide dynamics. Analysis of denudation patterns in the northern Cascades range confirms earlier hypotheses suggesting that the Skagit River Gorge resulted from the breaching of a divide by the upper Skagit River. A range-transverse thermochronologic transect suggests that this ridge was the former range drainage divide. Cooling paths calculated using AHe age-elevation relationships on the flank of the gorge and a ⁴He/³He age on the gorge floor confirm an earlier hypothesis that the gorge formed within the past 2.0 m.y., likely by lake overflow. Anomalously high unit stream power in the 2000-m-deep Skagit River Gorge further suggests that rapid incision continues to present day within the breach. This river diversion occurred in a general context of range-scale drainage divide migration toward the leeward side of the range controlled by asymmetrical orographic precipitation. We suspect that this deep trend explains the systematic rerouting of ice-dammed streams toward the windward, humid side of the range during the Quaternary. Similar events have been documented all along the range during the Pleistocene. Thus, the incursions of the Cordilleran ice sheet would have simply catalyzed the drainage divide migration.

This work was funded by Swiss National Fund grant 200021-112175/1, the Department of Earth Sciences at University of Minnesota, UMN grant-in-aid 1003-524-5983, and the France-Berkeley Fund. We are grateful for detailed reviews from T.F. Schildgen and D.M. Whipp that helped to improve the manuscript.