Bedrock erodibility governs transient erosion and CO₂ release in alpine landscapes

Abstract

Differences in lithology within alpine watersheds lead to strong contrasts in bedrock erodibility, which modulate climatic and tectonic influences on erosion, chemical weathering, and carbon dioxide exchange. However, the difficulty of measuring denudation rates in fine-grained rocks has limited our ability to quantify the magnitude of erodibility differences between weak sedimentary and strong igneous rocks. Here, we combine newly developed cosmogenic beryllium-10 measurements in shale with conventional beryllium-10 measurements in intrusive rocks, topographic analysis, solute data, and landscape evolution modeling in the East River watershed, Colorado. Shale-dominated catchments denude three times faster than adjacent intrusive catchments despite having gentler topography. Hydrochemical data and modeling indicate that elevated shale denudation enhances solute generation and net inorganic weathering–driven carbon dioxide release during prolonged landscape adjustment over million-year timescales. These results show that spatial variability in rock type alone can substantially alter the carbon dioxide balance of alpine landscapes by hiding the fastest erosion rates in relatively gentle terrain.

INTRODUCTION

Lithologic variability is widely expected to influence landscape evolution by governing the strength, erodibility, and chemical reactivity of bedrock (1–5). However, the extent to which lithology controls erosion, solute export, and weathering-driven CO₂ exchange remains difficult to quantify across landscape scales (6–9). Differences in rock properties influence the efficiency of channel incision, hillslope erosion, and sediment and solute generation and transport (10–12), but most field tests infer lithologic effects from topography or models rather than by directly measuring denudation rates across the contrasting rock types present in nature (2–4, 6, 13). Because erodibility varies by more than an order of magnitude among common lithologies, juxtaposition of different rock types can produce strong spatial contrasts in denudation rates, channel steepness, and relief, even under uniform climatic and tectonic forcing (Fig. 1) (1, 3, 4, 6, 14, 15). In mixed-lithology mountain belts, these contrasts may be especially important during transient adjustment, when the response to glaciation, base-level change, or climate variability may be partitioned among different rock types during prolonged landscape adjustment (5, 16). Laboratory abrasion experiments show that rocks with lower tensile strength or finer grain size are more efficiently eroded by sediment impacts and therefore should experience higher incision efficiency than stronger rocks (14, 15). Landscape evolution theory and numerical models based on these mechanistic insights predict that when landscapes are in topographic steady state and erosion rates are spatially uniform, strong rocks generate steep, high-relief topography, whereas weak rocks result in subdued, low-relief terrain (3, 13). These models also predict that large erodibility contrasts can drive drainage reorganization and long-lived topographic disequilibrium (Fig. 1) (3, 13). However, field-scale measurements of bedrock erodibility are generally lacking, making it difficult to evaluate whether experimental results and model predictions adequately capture the effects of variable lithology in natural watersheds (3, 13–15, 17). As a result, few field observations are available to test model predictions that lithologic contrasts control denudation rates during periods of transient landscape adjustment (6, 17–19).

Fig. 1. Modeled landscape response to lithologic heterogeneity.

(A) Landscape evolution model output predicted by fluvial incision only of topography at 0 years (initial dynamic equilibrium under uniform erodibility), 15 kyr after imposing the lithologic contrast (early transient adjustment), 2.6 Myr (prolonged transient adjustment), and 21 Myr (new dynamic equilibrium). The low-erodibility bedrock ($K_{\text{low}}$ = 1.44 × 10⁻⁶ m^0.1 year⁻¹) is shown in purple, and the high-erodibility bedrock ($K_{\text{high}}$ = 4.57 × 10⁻⁶ m^0.1 year⁻¹) is shown in green. (B) Box plots summarizing scaled mean relief, channel steepness, and denudation rate for catchments fully contained within the $K_{\text{low}}$ and $K_{\text{high}}$ domains ($K_{\text{low}}$ catchments, n = 17; $K_{\text{high}}$ catchments, n = 46). At each time step, values for both $K_{\text{low}}$ and $K_{\text{high}}$ catchments are scaled by the maximum value of the corresponding metric across the two lithologic groups. Boxes span the interquartile range, center lines mark medians, and whiskers extend to the most extreme values within 1.5 times the interquartile range. During early transient adjustment, $K_{\text{high}}$ catchments exhibit faster denudation than $K_{\text{low}}$ catchments, while contrasts in relief and steepness remain comparatively modest. As transient adjustment continues, contrasts in relief and channel steepness progressively develop, while denudation rates remain distinct between lithologies, with substantial differences in all three metrics persisting at 2.6 Myr. By 21 Myr, denudation rates are similar across lithologies, while strong contrasts in relief and channel steepness persist.

In many field settings, predictions of lithologic control on erosional efficiency and landscape adjustment have been tested using topographic metrics that capture spatial contrasts in river incision, including channel steepness indices and erodibility coefficients (1, 2, 6). Field-based measurements of rock strength and fracture density further reinforce predictions of lithologic controls on topography, showing that mechanically weaker lithologies are associated with lower channel steepness and gentler hillslopes, whereas stronger rocks support steeper channels and higher relief (2, 4, 20, 21). Cosmogenic nuclide studies in quartz-rich lithologies likewise demonstrate that intrusive and metamorphic rocks generally denude more slowly than sedimentary rocks under similar external forcing (22–24). Collectively, these observations support a clear expectation: Weak sedimentary rocks should exhibit substantially higher erodibility, and therefore higher denudation rates, than adjacent intrusive rocks under the same climatic and tectonic forcing (4, 25). However, topographic metrics alone cannot unambiguously distinguish steady-state lithologic relief from transient adjustment (1, 2), and conventional cosmogenic nuclide measurements have mostly been limited to quartz-rich lithologies (6, 26–28). As a result, the key prediction that weak fine-grained sedimentary rocks should dominate catchment-scale denudation during transient adjustment remains poorly tested. If these contrasts persist, then the weakest rocks may dominate denudation, solute export, and weathering-driven CO₂ exchange even where they do not produce the steepest topography (7, 8, 29).

Despite these predictions, field-scale tests of lithologic controls on denudation have been limited by methodological constraints. Catchment-averaged denudation rates derived from in situ–produced ¹⁰Be have been measured almost exclusively in settings where weathering produces sand-sized quartz (26, 27, 30), enabling conventional quartz separation techniques (28). In fine-grained sedimentary rocks such as shale, quartz occurs primarily as silt- to clay-sized grains dispersed within phyllosilicate- and carbonate-rich matrices, making it difficult to isolate sufficient quantities of clean quartz using standard hydrofluoric acid etching procedures (9, 28, 30). As a result, denudation rates for shale, notwithstanding its widespread occurrence and inferred high erodibility, have rarely been measured directly at the catchment scale (9, 25–27, 30). However, a methodological advance now enables the extraction of micrometer-scale quartz from fine-grained sediment, allowing direct measurement of in situ–produced ¹⁰Be in shale and quantification of catchment-scale denudation rates in some of the weakest lithologies (9).

Here, we combine a new cosmogenic nuclide capability for measuring ¹⁰Be in shale with topographic analysis, hydrologic and geochemical data, and landscape evolution modeling to quantify lithologic controls on denudation rates, topography, and solute fluxes in the formerly glaciated East River watershed, Colorado, USA (Fig. 2) (31). The watershed spans a sharp lithologic contrast between highly erodible Cretaceous Mancos Shale and resistant igneous rocks, including quartz monzonite and granodiorite laccoliths and plutons intruded into the surrounding sedimentary sequence, under broadly uniform climatic and tectonic forcing (10, 31–33). By directly measuring denudation rates in shale-, nonshale sedimentary– (hereafter “sedimentary”), and intrusive-dominated catchments, we test whether erodibility contrasts decouple topography from denudation rates during postglacial landscape adjustment and evaluate the consequences of prolonged geomorphic transience for solute generation and watershed-scale CO₂ balance.

Fig. 2. Catchment lithology, denudation rates, and solute export in the East River watershed.

(A) Location of the East River watershed in Colorado, USA. (B) Map of the East River watershed showing the watershed boundary (white outline), lithologic distribution (filled polygons), delineated catchments (unfilled polygons), catchment outlets (points), and catchment-averaged denudation and chemical weathering rates (numeric labels), expressed in millimeters per year. (C) Simplified schematic geologic cross section (A to A′), constructed from the lithologic relationships shown in the USGS geologic map and cross section of Gaskill et al. (53), illustrating the structural relationships among intrusive rocks, Mancos Shale, and the other sedimentary units; colors correspond to the lithologic classes shown in (B). The simplified geological map is derived from USGS 1:24,000-scale geologic maps (32, 53) and the USGS geologic map of Colorado. Intrusive (purple), sedimentary (blue), and shale (green) bedrock, together with Quaternary surficial deposits (tan), are overlain on an airborne light detection and ranging (LIDAR)–derived shaded relief map (59). Catchments were delineated upstream of each sample location, and several outlets define nested upstream-downstream catchment pairs. Colored numeric labels adjacent to lithology-specific catchments give catchment-averaged ¹⁰Be-derived denudation rates (in millimeters per year). Large mixed-lithology outlets used for paired denudation and solute analyses are marked separately; black numeric labels denote total denudation rates from ¹⁰Be, and white numeric labels denote chemical weathering rates derived from solute yields, both in millimeters per year.

RESULTS AND DISCUSSION

Predicted transient signatures of lithologic contrasts

Landscape evolution theory based on fluvial incision predicts that sharp contrasts in bedrock erodibility generate diagnostic, time-dependent patterns in topography and denudation rates in landscapes where resistant igneous rocks intrude weaker sedimentary rocks and erosion exhumes the lithologic contacts, as in the East River watershed (Figs. 1 and 2) (1–3, 6, 13). In such a lithologic configuration, weak, high-erodibility sedimentary and strong, low-erodibility intrusive rocks are adjacent to one another, and both rock types are subject to the same base-level forcing. During transient landscape adjustment, contrasts in bedrock erodibility cause incision to preferentially localize within the high-erodibility lithology, driving rapid channel lowering and elevated denudation rates, while the resistant intrusive rock adjusts more slowly and exhibits lower incision efficiency (Fig. 1). In such a scenario, the predicted transient signal is characterized by rapid denudation of the weak lithology without the development of high relief, producing comparatively subdued topography despite rapid erosion, whereas resistant rocks steepen through time as they denude more slowly than the surrounding weak rocks, leading to progressively steeper slopes and greater relief (Fig. 1). The relationship between denudation rate and topography therefore evolves systematically during postperturbation adjustment. Early in the adjustment phase, weak lithologies are expected to exhibit high denudation rates but only moderate relief, while strong lithologies develop higher relief yet comparatively low denudation rates. As adjustment proceeds, differences in denudation rate persist, while contrasts in relief and channel steepness progressively develop, resulting in pronounced differences in both denudation and topography over million-year timescales. Only over tens of millions of years do denudation rates approach similar values, while strong contrasts in relief and channel steepness persist as the landscape continues to adjust toward topographic equilibrium (Fig. 1). These numerical predictions of fluvial landscape adjustment provide a framework for evaluating whether natural landscapes exhibit signatures of lithology-driven transience, in which the fastest erosion occurs in weak rocks before topographic equilibrium is reached. Our landscape evolution model represents a generalized landscape with strong lithologic contrasts undergoing adjustment due to fluvial erosion during an ice-free period. Transient landscape evolution is expected to occur following deglaciation, as fluvial drainage networks incise into glacially carved topography, irrespective of lithologic contrasts (34). The model does not simulate glaciation or deglaciation and is not intended to reproduce the specific geologic or glacial history of the East River watershed, including when intrusive rocks were first exposed or how the landscape evolved during repeated glacial-interglacial cycles. Instead, the model provides a process-based framework for evaluating how contrasts in bedrock erodibility influence transient denudation, topography, and fluvial landscape adjustment.

Denudation, topography, and solute fluxes across lithologic domains

¹⁰Be-derived denudation rates across the East River watershed differ systematically as a function of lithology (Fig. 3 and tables S1, S2, and S6). Shale-dominated catchments exhibit the fastest denudation rates, ranging from 0.05 to 0.38 mm year⁻¹, with a median of 0.11 mm year⁻¹ (Fig. 2 and fig. S4). Catchments dominated by intrusive and sedimentary lithologies generally denude more slowly, at rates ranging from 0.03 to 0.07 mm year⁻¹, with medians of 0.06 and 0.05 mm year⁻¹, respectively (Fig. 2 and table S7). Two intrusive rock catchments have higher denudation rates (0.14 and 0.24 mm year⁻¹), but these elevated values occur in high-relief catchments with glacial cirque topography that are ∼1.5× steeper than the other study catchments (Fig. 2 and table S1). When all intrusive catchments are included, shale and intrusive catchments do not differ significantly in denudation rate (Mann-Whitney U, P = 0.079). However, when the two catchments with cirque topography are excluded from the analysis, shale denudation rates are significantly higher (P = 0.0003; table S7). The higher denudation rates in the two intrusive catchments are expected given their steeper topography (35) and indicate that glacially modified topography in resistant rocks can also undergo rapid denudation during postglacial adjustment. However, intrusive rocks generally denude more slowly overall, and the effective erodibility inferred for the two catchments is comparable to that of other intrusive rock catchments (Fig. 3 and table S2), indicating that their elevated denudation rates primarily reflect greater catchment steepness rather than anomalously erodible intrusive bedrock.

Fig. 3. Lithologic controls on catchment morphology, channel geometry, denudation, and inferred erodibility.

Catchment-wide geometric, topographic, and denudation metrics for the catchments with ¹⁰Be measurements shown in Fig. 2, grouped by dominant lithology: intrusive (purple; n = 7), shale (green; n = 14), and sedimentary (blue; n = 3). Panels show distributions of drainage area, mean catchment slope (in degrees), catchment relief, channel concavity, normalized channel steepness ($k_{\text{sn}}$; in meters to the 0.9 power), catchment-averaged denudation rate (in millimeters per year), and inferred effective erodibility ($K$; in meters to the 0.1 power per year). Box plots summarize the distribution of values for the intrusive and shale groups: boxes span the interquartile range (25th to 75th percentiles), horizontal lines denote medians, and dashed whiskers extend to the most extreme values within 1.5 times the interquartile range. The three sedimentary catchments are shown as individual points because of the small sample size. Individual catchment values for the intrusive and shale groups are shown as points overlying the box plots. The two highest intrusive rock denudation rates are from high-relief cirque catchments. Statistics describing differences in denudation rates and topographic metrics are reported in tables S7 and S8.

Shale catchments have lower median hillslope gradients of 19.5°, a lower median relief of 256 m, and a lower median normalized channel steepness of ∼27.5 m^0.9 than intrusive catchments, which have median slopes of 22.1°, a median relief of 324 m, and a median normalized channel steepness of ∼42.0 m^0.9 (Fig. 3, figs. S5 to S7, and table S8). Normalized channel steepness is significantly lower in the sampled shale catchments than in the intrusive catchments (Mann-Whitney U, P = 0.006), whereas differences in slope and relief are not statistically significant (table S8). A separate analysis of the entire study area (Fig. 2) indicates that mean slope, relief, and normalized channel steepness are significantly lower in shale than in intrusive rock terrain (table S9). These observed differences in topography are consistent with predictions of transient landscape evolution in our model (Fig. 1 and fig. S1). Combining catchment-wide denudation rates with channel steepness allows estimation of effective bedrock erodibility under the assumption of spatially uniform climate and hydrological forcing (Fig. 3 and see Materials and Methods) (18, 19). Shale catchments yield effective erodibility values centered around ∼4.6 × 10⁻⁶ m^0.1 year⁻¹, which is approximately three times higher than intrusive and sedimentary catchments (∼1.4 × 10⁻⁶ m^0.1 year⁻¹) (Fig. 3).

Faster denudation rates for shale, despite gentler topography, show that the most rapidly denuding parts of a mountain landscape need not coincide with the steepest or highest-relief terrain. Solute fluxes calculated from discharge and total dissolved solids (TDS) data from larger catchments that encompass multiple rock types (36–39) indicate that chemical weathering rates in the East River are equivalent to surface lowering rates of ∼0.01 to 0.08 mm year⁻¹ (Fig. 2, fig. S8, and table S4). The TDS-based flux data therefore indicate that chemical weathering accounts for ∼10 to 70% of the total denudation rates for the same larger catchments, which range from ∼0.10 to 0.17 mm year⁻¹ (Fig. 2 and table S2). The large-catchment denudation rates may underestimate total denudation because the analyzed sediment may not adequately capture denudation of the faster-eroding, fine-grained Mancos Shale (table S2). Nevertheless, the ratios of chemical weathering to total denudation are consistent with global scaling between solute yield and denudation rate (40) (fig. S8). Because TDS data from larger catchments integrate multiple lithologies, TDS measurements alone cannot unambiguously partition solute export among rock types. However, in the East River, pyrite and carbonate weathering fronts in Mancos Shale extend through a subsurface weathering zone to depths of ∼2.5 to 4 m (41, 42). Previous catchment-scale hydrochemical observations also show that carbonate and sulfide weathering dominate dissolved solute export from larger East River catchments that drain both intrusive rock and Mancos Shale (39), and small catchments that drain shale-bearing sedimentary rocks exhibit elevated carbonate and sulfide weathering fluxes (10). Collectively, these prior observations (10, 39, 41, 42) and the TDS data indicate that, although the Mancos Shale exhibits gentler topography, its rapid denudation likely contributes disproportionately to solute generation and inorganic weathering-driven CO₂ release. Mechanisms linking rapid denudation to enhanced weathering may include changes in subsurface hydrology and mineral exposure associated with landscape incision (10). Water-table elevation can control weathering front depth in shale, while denudation exposes fresh material on hillslopes, which together enhance water-rock interactions within the critical zone (41, 42).

Landscape evolution modeling and the longevity of transience

Landscape evolution simulations using the field-derived effective erodibility values for resistant igneous rock (1.44 × 10⁻⁶ m^0.1 year⁻¹) and weaker shale (4.57 × 10⁻⁶ m^0.1 year⁻¹) reproduce the pattern of high denudation rates and comparatively subdued topography in shale catchments that is observed in the East River during early transient adjustment [∼15 thousand years (kyr) after an initial steady state; Fig. 1 and figs. S1 to S3]. In both the model and observations, shale catchments denude faster while maintaining lower channel steepness and relief than adjacent intrusive catchments (Fig. 4). Model results using a stream power slope exponent equal to 1 indicate that denudation rates in shale remain elevated relative to intrusive rock during the first several million years of the simulation (Fig. 5A) and converge only over tens of millions of years (Fig. 1 and fig. S1). Sensitivity experiments that vary model processes, stream power slope exponent, spatial arrangement of rock types, and absolute erodibility reproduce the same first-order pattern of faster denudation in the high-erodibility lithology during transient adjustment (figs. S1 to S3). All simulations predict prolonged transience that exceeds the duration of individual glacial-interglacial cycles (43, 44), suggesting that the elevated shale denudation rates we observe can persist for the entire duration of interglacial periods. These simulations demonstrate that lithologic contrasts alone can drive transient behavior in mountain landscapes, independent of additional transience associated with fluvial adjustment of out-of-equilibrium glacial topography.

Fig. 4. Comparison of modeled lithologic contrasts with East River observations.

Comparison of catchment-scale values of normalized relief, normalized channel steepness, and denudation rate from landscape evolution model predictions and observations. Box plots with model data show values predicted by fluvial incision for $K_{\text{low}}$ (purple; $K$ = 1.44 × 10⁻⁶ m^0.1 year⁻¹; n = 17) and $K_{\text{high}}$ (green; $K$ = 4.57 × 10⁻⁶ m^0.1 year⁻¹; n = 46) areas at 15 kyr (cf. Fig. 1 and fig. S1), whereas box plots with observed data show measured values from subcatchments in the East River watershed. Model values are normalized by the maximum value of the corresponding metric across both lithologic domains at the 15-kyr model time step. Observed values are normalized independently by the maximum observed value of the corresponding metric across the paired intrusive and shale catchments, allowing comparison of the relative differences attributable to lithologic contrasts rather than absolute magnitudes.

Fig. 5. Linking model predictions, transient denudation, and watershed-scale CO₂ release.

(A) Time series of mean denudation rates across $K_{\text{low}}$ and $K_{\text{high}}$ domains of the model (Fig. 1; solid purple and green curves, respectively) and for shale and intrusive rock domains in the simulation initialized from the present-day East River topography (Fig. 2; dashed green and purple curves, respectively). Both simulations use the same field-derived high- and low-erodibility values. Denudation rates remain distinct between lithologies during prolonged transient adjustment, although the magnitude and temporal evolution differ between the two model configurations. (B) Simplified, fixed-area CO₂ flux predictions for the East River watershed based on the mapped Mancos Shale and intrusive rock areas only, scaled using the corresponding denudation rate trajectories shown in (A). Solid curves use $K_{\text{low}}$ and $K_{\text{high}}$ trajectories from the model (Fig. 1), whereas dashed curves use the shale and intrusive rock trajectories from the East River topography simulation (Fig. 2). Curves show instantaneous shale carbonate weathering CO₂ release (green), the net CO₂ flux from intrusive rock weathering (purple), and the net CO₂ flux (black). Positive values indicate net CO₂ release and negative values indicate net CO₂ drawdown. (C) Cumulative net inorganic weathering–driven CO₂ release through 2.6 Myr calculated from the CO₂ flux trajectories in (B) for the model (Fig. 1; blue) and the East River topography (Fig. 2; gray) simulations for two scenarios: one in which the transient modeled trajectory continues through time and CO₂ flux accrues only during successive idealized 15-kyr-long interglacial intervals (solid lines) and one in which the initial 0- to 15-kyr modeled CO₂ flux trajectory is repeated during each 15-kyr interglacial interval (dashed lines). Pale pink vertical bands indicate schematic interglacial intervals. Neither scenario accounts for CO₂ exchange during glacial intervals; hence, the slopes of the lines are zero during those times. For both model configurations, the repeated and continuous scenarios are initially similar but progressively diverge as the continuously evolving landscapes adjust, while repeated transience drives higher cumulative net CO₂ release over longer timescales.

The influence of transience on chemical weathering and CO₂ fluxes

Given that chemical weathering rates in the East River watershed and elsewhere scale with denudation rates (40) (fig. S8), enhanced denudation of shale during prolonged geomorphic transience has direct implications for solute generation and carbon cycling (8, 39, 45). Prior stream chemistry measurements indicate that CO₂ fluxes driven by chemical weathering differ systematically between lithologies in the East River watershed (10). Sedimentary rocks, including shale, are net CO₂ sources due to carbonate dissolution by sulfuric acid, whereas intrusive rocks are net CO₂ sinks due to carbonic acid weathering of silicate minerals (10). To evaluate how transient landscape evolution influences the CO₂ budget of the East River watershed, we integrated published lithology-specific, area-normalized inorganic chemical weathering fluxes (10) with mapped lithologic areas, measured denudation rates, and the denudation rate trajectories predicted by our landscape evolution model. The resulting estimates of watershed-scale CO₂ release and drawdown during transient adjustment do not reconstruct the specific geologic evolution of the watershed but assume that denudation rates for shale and intrusive rocks evolve as predicted by our model (Fig. 5, A and B). The resulting CO₂ budget reflects only inorganic carbon cycling and does not include potential CO₂ release associated with oxidative weathering of fossil organic carbon in shale, which remains poorly constrained. Including fossil organic carbon oxidation in the budget would further enhance the predicted CO₂ release from weak, rapidly eroding sedimentary rocks (39, 46, 47). Long-term carbon budgets will also be influenced by glacial erosion and subglacial weathering processes during ice-covered intervals (48, 49), which are not represented in our calculation. The simulated CO₂ budget reflects chemical weathering only during intervals where the landscape is not covered by ice, rather than the total CO₂ flux integrated over the entire duration of glacial-interglacial cycles. Hence, the carbon budget calculation should be interpreted as a first-order estimate of how lithology-driven transient denudation influences inorganic weathering–driven CO₂ exchange during interglacial periods.

The net CO₂ balance depends on both the rate of mineral dissolution by chemical weathering, which we assume scales with denudation rate, and whether weathering reactions in each lithology release or consume CO₂. Our simulations predict that rapidly denuding shale contributes disproportionately to weathering-driven CO₂ release, while slower erosion of intrusive rock suppresses silicate weathering–driven CO₂ drawdown during transient landscape adjustment (Fig. 5, A and B). For both model configurations, the cumulative balance over the modeled interglacial intervals results in net CO₂ release (Fig. 5C). The modeling indicates that lithology-dependent differences in denudation rates can persist over million-year timescales (17, 43), sustaining differences in weathering-driven CO₂ exchange as the landscape adjusts (Fig. 5, A and B). The modeling further predicts that the elevated denudation rates we observe for shale persist for the entire duration of interglacial intervals (43, 44), which has implications for the CO₂ budget of alpine landscapes subject to topographic resetting by repeated glaciations.

To assess the influence of topographic resetting, we calculated cumulative CO₂ fluxes for two scenarios, both of which are based on a threefold erodibility contrast between shale and intrusive rock (Figs. 1 and 5). The topographic resetting scenario repeats the denudation rate trajectory predicted for the first 15 kyr of the simulation for all interglacial intervals (Fig. 5B). The topographic resetting scenario is simplified, in that we do not model glacial erosion. Rather, the scenario repeats the same initial CO₂ flux trajectory for a duration equivalent to an average interglacial interval and hence evaluates how topographic resetting, as might be expected to occur during glaciations, influences CO₂ budgets (Fig. 5C, dashed line). In contrast, the continuous topography scenario does not reset topography. The landscape starts from the same initial condition as the topographic resetting scenario, but topography and denudation rates evolve continuously over a timescale that spans multiple glacial-interglacial cycles without resetting (Fig. 5C, solid line). In both scenarios CO₂ flux is accounted for only during interglacial periods to maintain the same CO₂-accounting timescale. Both scenarios predict substantial cumulative net CO₂ release (Fig. 5C). Continuous landscape adjustment yields ∼40 Tg of CO₂ during eight 15-kyr interglacial periods that occur over 0.8 million years (Myr), compared with ∼45 Tg of CO₂ when the initial transient trajectory is repeated during each interglacial interval (Fig. 5C). Over longer timescales, the two scenarios progressively diverge as topography in the continuously evolving landscape adjusts, yielding ∼85 and ∼135 Tg of CO₂ by ∼2.6 Myr for the continuous- and repeated-transient scenarios, respectively (Fig. 5C). A simulation initialized from the present-day East River topography shows the same first-order behavior, with the continuous and repeated-transient trajectories progressively diverging over longer timescales and yielding ∼84 and ∼321 Tg of CO₂ by 2.6 Myr, respectively (Fig. 5C). Landscapes with different mineralogy, proportions of rock types, erodibility, denudation rates, and interglacial durations will have different magnitudes of CO₂ release and drawdown, but our analysis provides a generalized demonstration of how repeating transient conditions, such as those associated with glaciation, enhance erosion- and chemical weathering–driven CO₂ fluxes relative to landscapes approaching or at steady state.

Implications of transience for the CO₂ budget of mountains

Similar juxtapositions of sedimentary and igneous rocks subjected to repeated glaciation occur in many mountain belts worldwide (7, 11, 13, 16), suggesting that lithology-driven geomorphic transience may exert a broadly important control on sediment, solute, and CO₂ fluxes beyond the East River. Transient landscape evolution can be triggered by changes in rock uplift, base level, drainage capture, and autogenic dynamics (6, 13, 50–52). Thus, the potential impacts of transience on chemical weathering and CO₂ budgets extend beyond glaciated mountains. More generally, our results emphasize lithology as a first-order control on landscape evolution and adjustment timescales, alongside climate and tectonics (1, 4, 11, 16). Strong contrasts in bedrock erodibility can sustain long-lived geomorphic transience, during which denudation rates, channel form, and relief remain decoupled from steady-state expectations over million-year timescales. In these landscapes, topographic form alone may provide an incomplete picture of erosional and geochemical dynamics because the weakest lithologies can dominate denudation and sediment export and exert an outsized influence on solute generation and CO₂ release, even where they do not control relief. Explicitly accounting for lithologic contrasts and adjustment timescales is therefore essential for interpreting how mountain belts regulate weathering fluxes and Earth’s long-term carbon cycle (8, 29).

MATERIALS AND METHODS

Study area

The East River watershed drains ∼465 km² of the southern Elk and West Elk Mountains in central Colorado, USA (Fig. 2) (33). Elevations range from ∼2400 m along the valley floor to >4000 m at alpine peaks. Precipitation is dominated by snowfall, with annual precipitation exceeding 1200 mm at high elevations, which contributes to snowmelt-driven discharge that peaks in late spring to early summer (33, 39). The landscape was extensively glaciated during the late Pleistocene, with valley glaciers reaching their maximum extent at ∼18 ka and retreating to high-elevation cirques by ∼13 ka (31). Glacial erosion produced steep headwalls and deepened valleys, and lower elevations contain glacial till and outwash deposits. The geology of the watershed consists of extensive exposures of Cretaceous Mancos Shale, a fine-grained sedimentary rock rich in carbonate minerals, resistant quartz monzonite and granodiorite igneous intrusions, and other sedimentary rocks (Fig. 2 and table S6) (32, 53, 54). The other sedimentary rocks include the Maroon, Gothic, Morrison, Dakota, and Mesaverde formations. These formations include beds of limestone, sandstone, siltstone, mudstone, and calcareous shale, with local conglomerate beds, and occur along the eastern and western margins of the watershed (Fig. 2 and table S6) (32, 53, 54). The igneous intrusions occur as laccoliths and plutons emplaced within the sedimentary sequence and occupy much of the high-elevation terrain (32, 33, 55). For Fig. 2, lithology was determined primarily from 1:24,000-scale geologic maps (32, 53) and supplemented with the US Geological Survey (USGS) digital geologic map of Colorado (54). The mapped lithological units were grouped into three first-order lithologic classes: Mancos Shale, intrusive rocks, and non–Mancos Shale sedimentary rocks (table S6). Areas mapped as Quaternary surficial deposits (32, 53, 54) were assigned to a separate surficial deposit class and omitted from the bedrock area totals used in the CO₂ flux calculations.

Landscape evolution modeling

We use the three-dimensional landscape evolution model FastScape to simulate the transient geomorphic response to strong lithologic contrasts under uniform climatic and tectonic forcing (Fig. 1, figs. S1 to S3, movie S1) (56, 57). The results in Fig. 1 use a fluvial incision–only simulation. Sensitivity experiments that additionally include lithospheric flexure and hillslope diffusion yield comparable results to the fluvial incision–only model (figs. S1 and S2). The model was used to isolate the role of spatially variable bedrock erodibility in controlling denudation rates, topography, and landscape adjustment timescales, rather than to reproduce the detailed geometry of the East River watershed. All model parameters and units for the simulations are listed in table S3.

Governing equations and process representation

Fluvial incision was simulated using a detachment-limited stream power incision law, $I=K{A}^m{S}^n$, where $I$ is the incision rate, K is the bedrock erodibility, A is the upstream drainage area, S is the local channel slope, and m and n are the empirical exponents. Because fluvial incision is the only erosional process in the main model configuration, we equate modeled incision rate with denudation rate when comparing model output with the catchment-averaged observations. In sensitivity experiments that included hillslope transport, sediment flux was represented with a linear diffusion law, ${\mathbf{q}}_{\mathrm{s}}=-D\nabla z$, where ${\mathbf{q}}_{\mathrm{s}}$ is hillslope sediment flux, $D$ is hillslope diffusivity, and $z$ is elevation, and lithospheric flexure was included using FastScape’s built-in elastic plate flexure formulation (figs. S1 and S2 and table S3). Uplift was spatially uniform and constant through time at 1 × 10⁻⁴ m year⁻¹, and climatic and hydrologic forcing were also held constant, such that spatial differences in denudation and topography arose only from contrasts in bedrock erodibility.

Model domain, boundary conditions, and initial state

Simulations were run on a 300 km–by–300 km domain with a grid cell size of 1 km, with FastScape’s default D8 flow-routing scheme, two fixed-elevation boundaries, and two periodic boundaries (table S3 and movie S1). The model was first evolved for 100 Myr under spatially uniform erodibility of $K$ = 4.57 × 10⁻⁶ m^0.1 year⁻¹ to a dynamic topographic steady state, where the denudation rate balances uplift rate and topography does not substantially change through time (table S3). The steady-state topography was then used as the initial condition for transient experiments in which a factor-of-3 erodibility contrast was imposed between high-erodibility areas with $K_{\text{high}}$ = 4.57 × 10⁻⁶ m^0.1 year⁻¹ and low-erodibility areas with $K_{\text{low}}$ = 1.44 × 10⁻⁶ m^0.1 year⁻¹ (table S3), following the median effective erodibilities independently estimated from the shale and intrusive catchments (Fig. 3). In the model configuration presented in Fig. 1, $K_{\text{low}}$ domains were prescribed as three elliptical bodies, which broadly mimic plutons intruded into sedimentary rock. Sensitivity experiments varied the spatial distribution of the low-erodibility domains while holding their total area constant and varied the absolute erodibility values while retaining an approximately threefold contrast between lithologies, allowing us to evaluate the effects of lithologic geometry and absolute erodibility on landscape adjustment (figs. S3 and S10 and table S3). The simulations ran for 21 Myr following the introduction of spatially varying erodibility, with a time step and output interval of 1000 years. Mean denudation rates in the $K_{\text{low}}$ and $K_{\text{high}}$ domains converged to within 5% for a sustained 500-kyr interval by ∼21 Myr across all tested scenarios (Fig. 1 and fig. S1). Topography and denudation rate output from simulations were analyzed at 0 years, 15 kyr, 2.6 Myr, and 21 Myr, and the 15-kyr output was compared directly with observations from the East River (Figs. 1 and 4). For the comparisons shown in Fig. 1, mean relief, channel steepness, and denudation rate were scaled at each time step by the maximum value of the corresponding metric across the $K_{\text{low}}$ and $K_{\text{high}}$ model catchments.

East River topography sensitivity

We performed a complementary simulation initialized from the present-day East River topography using spatially variable erodibility based on the mapped distribution of rock types (movie S2). The simulation used the same fluvial-incision formulation, uplift rate, stream power parameters, and field-derived high- and low-erodibility values (Fig. 3) as the model (Fig. 1). The resulting denudation rate trajectories and corresponding CO₂ fluxes are shown in Fig. 5.

Comparison of model output and observations

Model outputs

Model outputs were saved as GeoTIFF and Visualization Toolkit (VTK) files and analyzed in MATLAB using TopoToolbox (58). Two complementary analyses were performed. First, for the catchment-based comparisons (Figs. 1 and 4), we delineated catchments fully contained within each lithologic domain and calculated catchment metrics: relief (the difference between maximum and minimum elevation), mean slope, mean channel steepness, and mean denudation rate. These catchment-scale metrics mirror the spatial scale of the cosmogenic nuclide measurements and enable direct comparison between modeled and observed denudation rates and topography. Second, we evaluated all cells and stream segments within the low- and high-erodibility areas across the full model domain (figs. S1 and S2). In that full-domain analysis, local relief was calculated from the digital elevation model (DEM) using a 5 × 5–cell moving window, corresponding to 25 km² at the model grid spacing; channel steepness was calculated for all stream segments within each lithology type, and mean denudation rate was tracked through time for the full low- and high-erodibility areas. The VTK outputs were visualized in ParaView to generate the three-dimensional maps and the animations shown in Fig. 1 and movies S1 and S2. In the model runs that added lithospheric flexure or flexure plus hillslope diffusion, the magnitude and pacing of adjustment differed from the fluvial incision–only model, but the manner in which denudation rate, channel steepness, and relief varied as a function of lithology was similar across the models (figs. S1 and S2).

Observations

Topographic metrics for all catchments sampled for denudation rates were derived from a 1-m resolution airborne light detection and ranging (LIDAR) DEM of the East River watershed (59). Analyses were performed in MATLAB using TopoToolbox (58). Each sample location was snapped to the extracted channel network, and the upstream catchment was delineated. Flow direction and flow accumulation were calculated from the DEM. We tested a range of minimum contributing area thresholds and visually evaluated the resulting channel networks across the sampled catchments. We selected a minimum contributing area threshold of 0.01 km² to limit the inclusion of small flow paths potentially dominated by colluvial processes while retaining the principal fluvial drainage network (60, 61) (figs. S5 to S7). Catchment relief was calculated as the difference between maximum and minimum elevation within each catchment, and mean catchment slope was calculated as the spatial mean of the local topographic gradient across all DEM cells within the catchment. Channel steepness and concavity were quantified using χ-transformed river profiles in MATLAB with TopoToolbox (58, 62). Concavity was estimated using the disorder-based optimization method (63). To permit direct comparison of channel steepness among catchments, normalized channel steepness ($k_{\text{sn}}$) was calculated using a common reference concavity of 0.45 across the full extracted channel network (figs. S5 to S7). Several sampling sites define nested catchments, where we sampled both above and below pronounced breaks in longitudinal profile slope (i.e., “knickpoints”; Fig. 2 and figs. S5 to S7); in all cases, topographic analyses were performed for the entire drainage area upstream of each sample location. Because sample sizes were small and unequal and the topographic metrics were not assumed to be normally distributed, we used two-sided Mann-Whitney U tests to compare topographic metrics between shale and intrusive catchments (table S8). To complement the sampled catchment analysis, we also evaluated topographic differences between shale and intrusive terrain across the full East River study area shown in Fig. 2 (table S9). To reduce the influence of spatial autocorrelation among neighboring LIDAR pixels, we divided the study area into nonoverlapping 1-km blocks. Within each block, shale and intrusive terrain were analyzed separately, and mean slope, relief, and channel steepness were calculated for each lithology where sufficient mapped area was present. We summarized the distributions of these metrics between shale and intrusive terrain using two-sided Mann-Whitney U tests and repeated the analysis using 2-km blocks to evaluate sensitivity to block size. Normalized channel steepness was calculated on a 10-m routing grid using a fixed reference concavity of 0.45 and a minimum drainage area of 0.10 km². Effective bedrock erodibility, $K$, was estimated for each catchment by combining catchment-averaged denudation rates derived from in situ–produced ¹⁰Be (described below) with catchment-averaged normalized channel steepness, assuming a stream power incision model with exponent n = 1 and spatially uniform climate and hydrological forcing (51). Under these assumptions, $K=\frac{E}{k_{\text{sn}}}$, where $E$ is the catchment-averaged denudation rate and $k_{\text{sn}}$ is the catchment mean channel steepness (18).

Cosmogenic nuclide sampling and analysis

Catchment-averaged sampling

Catchment-averaged denudation rates were determined using in situ–produced cosmogenic ¹⁰Be measured in detrital quartz from river sediment (table S1). Sediment was collected from active channels near the outlets of individual subcatchments and major tributaries of the East River watershed. Field sampling was conducted under a US Forest Service permit issued to the Rocky Mountain Biological Laboratory. Sampling locations were selected to integrate denudation over drainage areas with shale, intrusive, or sedimentary lithologies, as defined by 1:24,000 geologic maps (Fig. 2) (32, 53). In addition to sampling lithology-specific catchments, sediment was collected at the outlets of several large, mixed-lithology catchments that coincide with discharge and solute chemistry monitoring stations, which enabled us to estimate the fraction of total denudation attributable to chemical weathering (Fig. 2). Quartz was isolated from bulk river sediment using lithology-specific preparation pathways. Quartz from intrusive, sedimentary, and large, mixed-lithology catchments was prepared using a conventional quartz purification workflow (30, 64), whereas quartz from shale-dominated catchments was isolated using a fine-grained quartz separation technique based on a precursor to the method developed by Huang et al. (9). For large mixed-lithology catchments, conventional sand-sized quartz preparation likely weights the cosmogenic signal toward quartz-producing intrusive and nonshale sedimentary units and may underrepresent fine-grained shale contributions. The mixed-lithology catchment ¹⁰Be denudation rates may underestimate total denudation because the analyzed quartz may underrepresent contributions from the faster-denuding, fine-grained Mancos Shale and therefore should not be interpreted as perfectly integrating denudation across all rock types.

Quartz extraction from intrusive, sedimentary, and mixed-lithology catchments

Quartz from intrusive, sedimentary, and large mixed-lithology catchments was isolated following standard in situ–produced ¹⁰Be preparation procedures (64). Bulk river sediment from the large catchments was oven dried and dry sieved to isolate the 250- to 850-μm grain-size fraction for further processing. Carbonate and Fe-bearing minerals were dissolved by leaching in warm HCl, followed by successive HF etches to dissolve nonquartz minerals and remove meteoric ¹⁰Be from grain surfaces, and heavy mineral separation to isolate quartz by density. Intrusive samples were prepared using the same overall workflow, but because the bedrock is quartz poor, the samples were boiled in hot phosphoric acid before HF etching. Quartz purity was assessed by inductively coupled plasma optical emission spectrometry (ICP-OES) and final Al concentrations ranged from 40 to 189 parts per million (ppm) for sedimentary rock and large mixed-lithology samples and from 98 to 261 ppm for intrusive rock samples.

Quartz extraction from shale catchments

Quartz from shale-dominated catchments was isolated using a fine-grained separation technique based on a precursor to the method of Huang et al. (9). Because the samples in this study were prepared before the final protocol by Huang et al. (9) was established, organic matter was removed by dry combustion at 500°C for 10 hours before acid leaching, rather than by a wet oxidation step that was subsequently adopted, as the latter approach minimizes potential diffusion of meteoric ¹⁰Be into quartz grains during heating (9). Bulk river sediment was oven dried and dry sieved to isolate the 250- to 850-μm size fraction. Clay minerals and amorphous phases were removed by boiling 100 g of sample in 25% NaOH, followed by leaching for 24 hours in a ∼3 M HCl and ∼1% HF solution using an unheated orbital shaker. Aluminosilicate minerals were then removed by digestion via boiling in 85% hot phosphoric acid at 220° to 250°C for 30 min, after which samples were boiled again in 25% NaOH to dissolve residual colloidal aluminosilicate and amorphous material adhering to quartz surfaces (9). Samples were then subjected to heavy-liquid separation to isolate quartz based on density (material with density ≲ 2.65 g cm⁻³) and, where sufficient material was available, were sieved into 20- to 60-μm and >60-μm fractions that were processed independently (table S2 and fig. S4). Because combustion can promote diffusion of meteoric ¹⁰Be into quartz grains (9), grain sizes finer than 20 μm were excluded from subsequent steps. Final purification of each fraction involved four to five 24-hour leaches in a 3 M HCl and 1% HF solution on an unheated orbital shaker, with rinses between each leach, to remove remaining surface contamination and meteoric ¹⁰Be before chemical extraction. Quartz purity was then assessed by ICP-OES and final Al concentrations were between 211 and 734 ppm. The lower but similar ¹⁰Be concentrations measured in the 20- to 60-μm fractions relative to the >60-μm fractions from the same samples (fig. S4) indicate that the combustion step did not cause measurable diffusion of meteoric ¹⁰Be into quartz grains. Two shale catchments yielded only a >60-μm quartz fraction (table S2). For the nine shale catchments with measurements from both fractions, the 20- to 60-μm results were used for the catchment denudation rates reported in Figs. 2 to 4 and in the statistical comparisons in table S7. The >60-μm results were used in Figs. 2 to 4 for the two catchments lacking 20- to 60-μm data.

Beryllium chemical separation and accelerator mass spectrometry measurement

Following quartz purification, all samples were processed using identical Be separation protocols (64). After addition of ∼250 μg of ⁹Be carrier, quartz was dissolved, and Be was chemically separated at the University of Massachusetts Cosmogenic Nuclide Laboratory following standard in situ–produced ¹⁰Be ion chromatography procedures (64, 65). Beryllium was precipitated as Be(OH)₂ by addition of ∼10% NH₄OH to pH ∼9, redissolved in 6 M HCl, dried overnight in a quartz crucible, calcined to BeO with a gas torch, and packed into cathodes with Nb powder. ¹⁰Be/⁹Be ratios were measured by AMS at Lawrence Livermore National Laboratory (LLNL) against AMS Be standard 01-5-4 (LLNL standard 07KNSTD) (66). Blank-corrected ¹⁰Be concentrations were determined by subtracting the mean ¹⁰Be/⁹Be ratio measured in process blanks from the same AMS run as each sample from the measured sample ratios. Uncertainty was propagated by adding the standard deviation of the blank ¹⁰Be/⁹Be ratios in quadrature; where only one blank was available, the measurement uncertainty of that blank was propagated in quadrature. Full AMS results, including sample identifiers, quartz masses, carrier masses, measured ¹⁰Be/⁹Be ratios, and analytical uncertainties, are reported in data S1.

Catchment-averaged denudation rate calculations

Catchment-averaged denudation rates (67–69) were calculated from measured in situ–produced ¹⁰Be concentrations and catchment hypsometry using the RIVERSAND Python package (70), which is integrated with an online denudation rate calculator (71) (tables S1 and S2). Rates are reported using the time-dependent Lifton-Sato-Dunai production rate scaling model (72). Catchment hypsometry was derived from a 3-m resolution LIDAR DEM of the East River watershed (59), discretized into 100-m elevation bins for production rate integration. No topographic shielding correction was applied (shielding factor = 1), following the assumption that for catchment-averaged denudation rates, reductions in surface production due to skyline shielding are offset by increases in effective attenuation length on sloping terrain (73). All denudation rates were calculated, assuming steady erosion, negligible long-term sediment storage, and efficient sediment mixing within each catchment, consistent with typical interpretations of fluvial cosmogenic nuclide data (67–69). Reported denudation rate uncertainties reflect analytical uncertainty in measured ¹⁰Be concentrations propagated through the RIVERSAND fitting procedure. For shale samples with measurements from both grain-size fractions, the 20- to 60-μm fraction generally yields higher denudation rates than the >60-μm fraction (fig. S4). However, repeating our analyses when using denudation rates from the >60-μm grain-size fraction instead of the 20- to 60-μm fraction results in similar statistical outcomes and does not change our findings; shale denudation rates remain higher than those of intrusive catchments (tables S7 and S10).

Solute flux, CO₂ flux, and yield calculations

Solute flux calculations

Discharge and water chemistry data were obtained from US Department of Energy Watershed Function Science Focus Area and USGS monitoring stations (36–38, 74). For each site, major dissolved species were compiled by sampling date from records spanning ∼1995–2025 for the Slate River and 2014–2024 for the other rivers. Where multiple analyses were available for a given site and date, concentrations were averaged to yield a single daily value (Fig. 2, fig. S8, and table S5). Using these compiled major ion data, we converted concentrations to milligrams per liter and to charge equivalents (in microequivalents per liter) using ion molar masses and valences. Bicarbonate (HCO₃⁻) was estimated by charge balance as the difference between summed cation and anion charge equivalents and converted to milligrams per liter using the molar mass of HCO₃⁻. TDS (in milligrams per liter) were then calculated as the sum of measured cations and anions, charge balance HCO₃⁻, and dissolved silica. To estimate solute export, we fit a site-specific power-law relationship between discharge ($Q$; in cubic meters per second) and TDS using paired observations, ${\text{log}}_{10}\left(\text{TDS}\right)=a+b {\text{log}}_{10}\left(Q\right)$, and used the fitted relationship to predict TDS at each discharge time step (fig. S9 and table S4). Instantaneous solute flux, $F_{\mathrm{ins}}$, was then calculated as $F_{\mathrm{ins}}=Q\times \text{TDS}$ after converting $\text{TDS}$ from milligrams per liter to kilograms per cubic meter. Annual solute export was obtained by summing flux through time and normalized by upstream drainage area to calculate $F$, reported in megagrams per square kilometer per year. The solute flux–derived weathering rates were converted to equivalent surface lowering rates (in millimeters per year) by dividing flux (in megagrams per square kilometer per year) by an assumed rock density of 2650 kg m⁻³. Using rock densities of 2100 to 2800 kg m⁻³ instead of 2650 kg m⁻³ shifts the equivalent weathering rate estimates by only about −5 to +26% and does not materially change our interpretation. Because pyrite and carbonate weathering fronts in the Mancos Shale extend to ∼2.5 to 4 m in depth within the East River watershed (41, 42), which is comparable to the depth interval where most ¹⁰Be production occurs, the $\text{TDS}$-derived weathering fluxes are interpreted as the chemical weathering component of total denudation measured by ¹⁰Be (Fig. 2 and fig. S8). Annual solute export, solute yield, and equivalent weathering rates are summarized in table S4. The daily discharge, ion concentrations, ion load calculations, bicarbonate charge balance estimates, and TDS loads are reported in data S2.

CO₂ flux calculations

To estimate the implications of lithology-driven transient denudation for watershed-scale CO₂ exchange, we constructed a simplified carbon flux model for the East River watershed and its tributaries for the study domain shown in Fig. 2 (∼405 km²). The CO₂ flux calculation includes mapped Mancos Shale (∼161 km²) and intrusive rock (∼82 km²) areas and excludes Quaternary surficial deposits, where underlying bedrock lithology is uncertain. Our simulation assumes that there are no limitations on the supply of rock. The Mancos Shale in the East River watershed is several hundred meters thick and is folded and dipping, rather than flat lying (Fig. 2C). The calculation also assumes that mapped lithologic areas remain constant through time and therefore does not represent changes in lithologic exposure as erosion proceeds. The goal of this simplified calculation is to isolate the magnitude, sign, and evolution of the dominant inorganic weathering terms associated with the principal lithologic end members, rather than to represent all weathering pathways (8, 75). Accordingly, the calculation excluded petrogenic organic carbon oxidation (46). Modern unit-area CO₂ fluxes were set to 13.3 kg of CO₂ km⁻² day⁻¹ for shale carbonate weathering release and −13.4 kg of CO₂ km⁻² day⁻¹ for the net CO₂ effect of intrusive rock weathering (10). The rates were converted to megagrams of CO₂ per year and scaled through time using the modeled denudation trajectories from the fluvial incision–only model and the complementary East River topography simulation, with shale and intrusive rocks associated with the high- and low-erodibility trajectories, respectively. For each model configuration, the lithology-specific trajectories were normalized to their corresponding 15-kyr values to represent present-day conditions, as deglaciation in the East River occurred ∼15 kyr ago, between 18 and 13 ka (Fig. 5A) (31). Normalizing the modeled denudation trajectories to their 15-kyr values anchors the calculation to the independently constrained modern CO₂ fluxes, such that the model represents their relative evolution through time rather than their absolute magnitudes. The corresponding modern median denudation rates are 0.11 mm year⁻¹ for shale catchments and 0.06 mm year⁻¹ for intrusive catchments (Fig. 3 and table S7). For lithologic class $i$, lithology-specific net CO₂ fluxes were calculated as $F_i\left(t\right)=A_i{r}_i{S}_i\left(t\right)$, where $F_i\left(t\right)$ is the lithology-specific net CO₂ flux (in megagrams of CO₂ per year), $A_i$ is the mapped lithologic area (in square kilometers), $r_i$ is the modern unit-area CO₂ flux (in megagrams of CO₂ per square kilometer per year), and $S_i\left(t\right)$ is a dimensionless denudation rate scaling function. The scaling function is defined as $S_i\left(t\right)=[D_i\left(t\right)/D_{i,\text{ref}}]$, where $D_i\left(t\right)$ is the modeled denudation trajectory and $D_{i,\text{ref}}$ is the corresponding modeled denudation rate for each lithology and model configuration at the 15-kyr reference condition. The lithology-specific fluxes were summed to calculate the time-dependent net watershed CO₂ flux. Positive values indicate net CO₂ release and negative values indicate net CO₂ drawdown (Fig. 5B and fig. S10). Cumulative net CO₂ exchange was estimated by integrating net flux through time and is reported in teragrams of CO₂ (Fig. 5C and fig. S10). The cumulative CO₂ budget was integrated over 2.6 Myr. We assumed idealized 100-kyr interglacial-glacial cycles (43, 44), each consisting of a 15-kyr interglacial interval where we calculated chemical weathering–driven CO₂ fluxes and an 85-kyr glacial interval during which CO₂ flux was not modeled and was therefore excluded from the cumulative budget. We do not imply that there was no subglacial chemical weathering or CO₂ exchange during glacial intervals, but we focus only on the interglacial periods where lithologically controlled differences in fluvial erosion occur. In the continuous trajectory case, each exposed interval used the corresponding interglacial segment of the continuously evolving model-derived CO₂ flux trajectory. In the repeated initial trajectory case, each interglacial interval repeats the initial 0- to 15-kyr CO₂ flux trajectory, such that the cumulative CO₂ budget is simply the net flux inferred for the first 15 kyr of the simulation, multiplied by the number of interglacial periods. We do not simulate glaciation, subglacial weathering, glacial erosion, or physical resetting of model topography; hence, the results provide an idealized comparison of repeated postglacial CO₂ flux trajectories (Fig. 5C). Hence, the resulting CO₂ budgets are a comparison of a landscape with different rock types that continually evolves toward a topographic steady state versus a landscape with different rock types that is periodically exposed (as occurs when glaciers retreat), such that the initial conditions repeat, and where the budgets are calculated only for interglacial periods so that the timescale of CO₂ accounting is the same for both scenarios. The CO₂ calculations were repeated using higher absolute erodibility values while retaining an approximately threefold contrast between lithologies to evaluate the sensitivity of cumulative CO₂ exchange to absolute erodibility (fig. S10).

Supplementary Materials

The PDF file includes:

Figs. S1 to S10

Tables S1 to S10

Legends for data S1 and S2

Legends for movies S1 and S2

Other Supplementary Material for this manuscript includes the following:

Data S1 and S2

Movies S1 and S2