Divergent drainage-scale responses to lithological heterogeneity Authors: Yifei Li, Huiping Zhang, Xudong Zhao, Mikael Attal, Simon M. Mudd, Peizhen Zhang Published: 2026 ABSTRACT Despite widespread recognition that lithological heterogeneity helps shape river networks, a widely applicable topographic diagnostic for whether and where lithological contrasts drive drainage reorganization is still lacking. Here, we introduce a drainage-network metric, the rock proportion–contributing area order (RP-CAO) relationship, and further show that responses diverge with drainage scale: small, low-discharge channels divert around resistant rocks, whereas large, high-discharge rivers largely maintain their course and incise through them. We capture this divergence by tracking rock-type enrichment along the network as a function of contributing area. Across landscape evolution simulations and natural river networks, the enrichment-scale relation forms a robust V-shaped pattern that quantifies these end members. Applied to networks along the eastern Tibetan Plateau, it isolates where lithology most strongly localizes rerouting and divide mobility despite strong spatial variability in tectonic and climatic influences. Requiring only topography and lithological maps, the framework enables comparable tests of drainage reorganization driven by spatially discrete lithological contrasts across mountain belts worldwide. INTRODUCTION Landscape evolution is shaped by the combined effects of tectonic deformation, climatic change, and lithological variation (1–5). While tectonics and climate provide the energy that drives geomorphic processes (6, 7), lithology, as an autogenic dynamic, modulates the response of Earth’s surface through spatiotemporal variability in rock erodibility (8–10). A growing body of research highlights the importance of lithological heterogeneity in influencing landscape dynamics, including fluvial incision patterns, variability in landforms, drainage network configurations, and weathering-carbon cycle interactions (8, 9, 11–13). Among these landscape responses, drainage dynamics are particularly sensitive to the distribution and strength of resistant rock bodies, including divide migration and drainage network reorganization (e.g., 11, 14, 15). Rivers generally traverse heterogeneous bedrock, where lithological contrasts can substantially affect erosion resistance, alter channel dynamics, modulate isostatic responses to erosion, and drive long-term landscape disequilibrium (8, 14, 16–18). This disequilibrium, driven by lithological contrasts, can lead to the migration of drainage divides, with resistant lithologies commonly forming highlands and persistent divides that occur independent of external forcing, such as climate and tectonics (17, 19–21). Such lithologically driven disequilibrium can also exert a substantial influence on river network patterns, leading to processes such as drainage reversals, river diversions, and the formation of longitudinal valleys (11, 14, 17). Previous studies have provided important insights into how lithological heterogeneity influences drainage network evolution. However, there is still no widely applicable topographic metric that can capture whether and where a drainage network has been reorganized or diverted in response to lithological contrasts. Developing such a tool would not only deepen our understanding of lithological influences on drainage network evolution but also provide a quantitative and accessible means of capturing their imprint on river network organization. In this study, we integrate a landscape evolution model (LEM) with field observations from the eastern Tibetan Plateau to introduce a metric, the Rock Proportion–Contributing Area Order (RP-CAO) relationship, which provides a drainage-scale quantification of how rivers of different contributing areas respond to lithological contrasts. We have devised this approach on the basis of the hypothesis that larger (i.e., higher order) rivers are more likely to continue cutting similar drainage paths when they exhume hard rocks. This framework allows us to capture how resistant rock exposures influence river diversion, divide migration, and river network reorganization, providing a general approach for detecting lithological signals, even in landscapes where tectonic and climatic forcings are complex. Observations from the eastern Tibetan Plateau Our study focuses on the Min River and the Dadu River, which drain the eastern Tibetan Plateau. The underlying lithology in the region consists of the Songpan-Ganzi terrane and the Longmen Shan (Fig. 1). The Songpan-Ganzi terrane is primarily composed of extensive Middle and Upper Triassic flysch (fine-grained marine sedimentary rocks) that is intruded by Late Triassic–Early Jurassic plutonic rocks (22). Exposures of Neoproterozoic-Paleozoic basement metamorphic rocks mainly occur in a northeast-trending belt within and adjacent to the Longmen Shan, where the Longmenshan thrust and strike-slip fault system serves as the dominant structural feature and marks the eastern boundary of the plateau. Proterozoic-Paleozoic volcanic rocks occur in narrow belts along the Longmen Shan and consist mainly of Permian basalt (22). Geological and geomorphic setting of the eastern Tibetan Plateau. Lithology is based on (53). Solid red lines indicate active faults (data from: www.activefault-datacenter.cn), where “F.” denotes fault. Solid blue lines indicate the main channels and major tributaries. Blue arrows indicate flow direction. The Min River and Dadu River drainages are outlined in pink and green polygons. Schmidt hammer measurement sites for resistant and erodible rocks are marked by red and black dots, respectively, with both rock types defined in the text. The lithology of the eastern Tibetan Plateau can be broadly categorized into two types on the basis of rock erodibility: (i) resistant rocks including plutonic rocks, volcanic rocks, and metamorphic rocks (depicted by darker shades in Fig. 1) and (ii) erodible rocks, consisting of sedimentary rocks, Quaternary unconsolidated sediments, and volcaniclastic rocks (shown in lighter colors in Fig. 1). The classification scheme is derived from the median lithological erodibility index values of class II lithological groups, as reported in the THOR rock strength database [see table S1.6 in (23)]. To the north of the Longmenshan Fault, the Maerkang thrust fault is well developed. Along the northern margin of the Min River drainage, the Longriba strike-slip fault system is prominent, while the Minjiang thrust fault and Xueshan thrust fault are developed on its eastern side. In the Dadu River catchment, the Xianshuihe strike-slip fault system is located to the southwest, and the Dari strike-slip fault to the northwest (Fig. 1). Low-temperature thermochronology and thermal history modeling studies indicate that two episodes of rapid erosion occurred in eastern Tibet: The first began between 30 and 25 million years ago (Ma) and lasted for ∼5 million years (Myr), while the second initiated between 10 and 15 Myr and has continued to the present day (24–26). Climatically, the portion of the eastern Tibetan Plateau studied here has been located within the South Asian monsoon zone since at least the Early-Middle Pleistocene (27), where the current mean annual precipitation ranges from ∼500 to ∼900 mm/year. In contrast, the adjacent Sichuan Basin, located to the southeast, receives substantially higher annual precipitation, ranging from ∼1000 to ∼1600 mm/year (fig. S1). Recent analyses indicate that the drainage divide between the Dadu and Min river basins is migrating westward at rates of 0.04 to 0.54 mm/year (28). This divide instability implies that the fluvial network in this region is undergoing active reorganization. Quantify the impact of lithological contrasts on the drainage network To characterize and quantify the evolution of the drainage network, we define a metric, the CAO, which facilitates the reclassification of the drainage network on the basis of the contributing area (used here as a proxy for discharge). The CAO is defined as the integer part of the base-10 logarithm of the contributing area A (square meters) (Eq. 1)CAO=int(log10A)(1) This order-of-magnitude binning provides a compact classification of channel scale while avoiding overly fine and sparsely populated area classes that would result from using a natural logarithm. Compared to the traditional stream order, CAO is easier to calculate, does not depend on a predefined threshold for channel initiation, and allows for more consistent comparisons across drainage systems of different shapes. For example, stream order can assign the same value to streams with vastly different drainage areas depending on catchment geometry, while CAO directly reflects differences in contributing area. CAO increases downstream along the drainage network, capturing the transition from headwater streams near divides to the trunk channel. While CAO may vary within a continuous river segment without major geomorphic changes, it provides a simple and scalable means of quantifying drainage structure across spatial scales. To quantify the impact of lithological contrasts on the drainage network, we introduce the RP for each CAO. The RP is defined as the ratio of the proportion of a specific rock type within the study area at a given CAO to the proportion of that rock type in the entire study area (Eq. 2)RPispecific rock=Nspecific rock∣CAO=iNstudy area∣CAO=i/Nspecific rockNstudy area(2)where N is the number of model nodes (in a model) or pixels (in a digital elevation model) under specific conditions and i is the CAO. The term Nspecific rock/Nstudy area represents the proportion of a given rock in the study area, while Nspecific rock|CAO = i/Nstudy area|CAO = i represents the proportion of the rock exposed along river reaches with a given CAO (Fig. 2). Schematic diagram illustrating RP for resistant rock. The gray polygon denotes the study area, while the red circle delineates a resistant rock outcrop. The main color of the stream represents CAO, with light blue for CAO = i and dark blue for CAO = i + 1. Streams with red edge colors are located within resistant rock. N denotes the number of nodes (in a numerical model) or pixels (in a digital elevation model) present in each of the domains defined here. Regions dominated by hillslope diffusion were grouped into a single category in the RP analysis without further subdividing CAO classes, and the colluvial-fluvial transition threshold was calculated on the basis of slope-area relationship. On the basis of analyses of numerical simulation outputs (fig. S2), a consistent threshold at a contributing area of 2 × 105 to 3 × 105 m2 (CAO = 5) was identified; thus, in the subsequent RP-CAO analysis, all CAO classes ≤5 were combined to represent hillslope and divide domains. This threshold is not intended as a universal channel-initiation criterion; rather, it reflects the balance between hillslope diffusion and fluvial incision under the adopted model parameters. Accordingly, the CAO value corresponding to the colluvial-fluvial transition may shift if the diffusion coefficient, fluvial erodibility, runoff, or lithological properties are changed. Sensitivity tests with varied diffusion coefficients confirm this behavior, with higher diffusivity shifting the transition toward larger contributing areas and vice versa (fig. S3). Because RP is calculated separately for each CAO class, this threshold does not affect the overall RP-CAO trends, although it may change the specific RP value assigned to the hillslope and divide domains. Conceptually, RP indicates whether a given rock type is relatively enriched or depleted along rivers of a given CAO compared with its abundance in the whole study area. If the rock is uniformly distributed across the whole network, RPispecific rock will be equal to 1 for all CAOs. If a rock becomes comparatively more represented along river reaches with a given CAO, RPispecific rock will exceed 1 for this CAO, and vice versa. RP is computed for both the simulations and the field comparison: Lithology is binary in the simulations (erodible and resistant), and field units are reclassified into the same two groups. This framework yields testable expectations: If lithological contrast has no effect on drainage evolution, then the RP will remain stable over time. Conversely, if the exposure of resistant rocks induces river diversion and highland formation, then we would expect the RPresistant in low CAOs (hillslopes and headwaters) to increase over time while their proportion in high CAOs (major river reaches) decreases. In contrast, the RPerodible would show the opposite trend. RESULTS River response to lithological contrast We use a LEM to examine how a quasi–steady-state drainage network responds to lithological contrasts following the introduction of resistant rocks (see Materials and Methods). This contrast in erodibility causes marked variations in erosion rates within the drainage network, resulting in long-term landscape disequilibrium and subsequent adjustments in topography and river network configuration, ultimately reaching a new steady state (Fig. 3 and movies S1 to S4). Overall, the model results demonstrate two primary patterns of drainage network response to the exposure of resistant rock: (i) the formation of highlands accompanied by river rerouting and (ii) river incision through exposed resistant rock without rerouting (Fig. 3 and movies S1 to S4). Modeled landscape evolution following the exposure of resistant rocks. (A to D) Modeled evolution of topography, rate of elevation change, drainage network and divides, and CAO, respectively. From top to bottom, each panel shows the model results at 0 Myr, 10 Myr, 30 Myr, and the new steady state. The polygons with pink and black edges represent the resistant rocks. In (C), the blue lines and red lines represent the river channels and divides, respectively. The data are obtained from the case with a relative erodibility of 3 (series 3–set 4). In areas characterized by lower discharge (or contributing area), we observe divides migrating toward the resistant rocks, consistent with previous observations that resistant lithologies can promote the development of highlands and drainage divides (17, 18). In our simulations, this migration occurs because exhumation of resistant lithologies temporarily reduces erosion rates relative to neighboring catchments developed on more erodible rocks, generating an erosional imbalance. As the drainage network adjusts, channels incise and steepen until erosion rates of the resistant and surrounding erodible rocks converge, ultimately establishing a new steady-state landscape. This adjustment also promotes the development of parallel longitudinal valleys around the margins of resistant rock masses through river capture by neighboring rivers (Fig. 3 and movies S1 to S4). In contrast, in areas with higher discharge (or contributing area), rivers tend to maintain their original flow paths, incising directly through the resistant rock bodies without undergoing network reorganization (Fig. 3 and movies S1 to S4). At the drainage scale, variations in river response to lithological contrasts are primarily governed by differences in discharge. RP and river diversion time We track the changes in RP over time for each simulation to quantitatively assess the impact of lithological contrast on the drainage network. To reduce the influence of the initial RP and enable a more direct comparison among different simulation scenarios, we ensure that, in each model configuration, the initial RP of both resistant and erodible rocks is set to 1 ± 0.1 when the resistant lithologies are introduced into the initial drainage network. That is, when resistant rocks are introduced, their spatial distribution is generated through an iterative quota-based procedure that ensures that the initial RP value converges to 1 ± 0.1 (see Materials and Methods). At the beginning of each model run, RP values are approximately uniform across all CAOs (RP ≈ 1; lightest blue and pink lines in Fig. 4A). For resistant rocks, the RP-CAO curve gradually forms a V-shaped pattern with time (blue lines in Fig. 4A), while for erodible rocks, it forms an inverted V-shaped pattern (pink lines in Fig. 4A). Specifically, RP of resistant rocks increases at CAO ≤ 5, indicating a higher proportion of resistant rocks on hillslopes and drainage divides, where contributing area is small (blue lines in Fig. 4A). At intermediate CAO values (6–8), RP for resistant rocks (RP6–8resistant) decreases, reflecting river rerouting to avoid the resistant rocks. In contrast, RP remains close to its initial value at the largest contributing areas (CAO = 9 to 10), suggesting that large rivers retain the power to incise through resistant rocks without substantial diversion, as illustrated by the trunk channels crossing resistant rock bodies in Fig. 3. The minor deviations of RP9–10resistant from their initial values should therefore not be interpreted as evidence of river diversion at the largest contributing area. Instead, these small changes reflect the integrated effects of upstream network reorganization, because diversion and capture at intermediate CAO values can redistribute downstream contributing area and thereby slightly modify the lithological composition of the largest rivers. RP and diversion times versus CAO derived from numerical modeling results. (A) Evolution of RP versus CAO over time. Light to dark blue represents the evolution of RP in resistant rocks from 0 to 30 Myr, while light to dark pink represents the evolution of RP in erodible rocks over the same interval. (B) RP versus CAO at 30 Myr under varying relative erodibility, with lighter to darker lines indicating increasing relative erodibility; blue denotes resistant rock, and pink denotes erodible rock. (C) Diversion times versus CAO for different relative erodibility. Lighter to darker shades indicate increasing relative erodibility, with the top of the rectangles representing the average diversion time and the bottom representing the minimum. The solid lines indicate the maximum contributing area at which river diversion occurs for each relative erodibility. RPerodible shows the opposite trend, with higher proportions appearing in intermediate CAO classes over time (pink lines in Fig. 4A). As erodible rocks occupy a larger proportion of the study area, the resulting changes in the RPerodible are less pronounced. In all cases, changes in RP diminish progressively over time, indicating that the drainage network approaches a quasi-steady state relative to the spatial pattern of rock erodibility (Fig. 4A). To assess the robustness of these RP trends, we conducted additional sensitivity tests involving model resolution, lithology-dependent hillslope diffusivity, and a nonlinear stream-power formulation. Tests using different grid sizes (100 and 500 m) yielded similar V-shaped RP-CAO patterns over time (figs. S4 and S5), indicating that the observed relationships are not substantially affected by model resolution. In the lithology-dependent hillslope diffusivity test, we used a simplified parameterization in which the hillslope diffusion coefficient within resistant rock bodies was reduced by the same relative contrast as fluvial erodibility. The RP-CAO relationship remained V-shaped, but RP changes were smaller and river diversion occurred later than in the corresponding uniform-diffusivity case (fig. S6), likely because reduced diffusivity limited diffusive sediment delivery to channels and increased their relative ability to incise through resistant rocks. Simulations using a nonlinear stream-power formulation with a slope exponent equal to 2 also showed the V-shaped pattern but with smaller RP changes (fig. S7), consistent with the reduced slope contrast required to balance erodibility contrasts when the slope exponent is greater than 1. Furthermore, during each model simulation, we systematically track all events of stream diversion around resistant rock formations. Specifically, at 1-million-year intervals, we identify all streams that intersect resistant rocks and record both the locations of intersection and the corresponding contributing area. By comparing these records across successive time intervals, we determine whether a stream that previously entered a resistant formation at a specific point ceases to do so in subsequent intervals. If a stream begins to intersect a resistant rock unit at time t1 but no longer does so at time t2, then the interval from t1 to t2 is defined as the diversion time, meaning the time the stream takes to reroute around the resistant rock due to lithological differences. In addition, the contributing area of the stream immediately before the diversion event, just before it ceases to intersect the resistant rock at t2, is recorded as a proxy for the stream discharge associated with the diversion event. Using these records, we quantify the temporal sensitivity of river networks to lithological contrasts. The evolution of RP6–8resistant (Fig. 4A) reveals that rivers with smaller CAO values respond earlier to resistant rock exposure. This trend is confirmed by diversion time statistics (Fig. 4C), where both the minimum and average diversion times positively correlate with CAO. That is, rivers with larger contributing areas (or higher discharge) require longer times to reroute around resistant lithologies. Effect of relative erodibility To test the influence of erodibility, we run suites of simulations that vary the relative erodibility (Kerodible/Kresistant) and the spatial configuration of resistant rocks (see Materials and Methods). The variations in RPresistant at 30 Myr under different relative erodibilities suggest that higher relative erodibility has a more pronounced impact on river networks (Fig. 4B). Specifically, an increase in relative erodibility leads to a larger RPresistant at CAO ≤ 5, indicating that more resistant rocks form highlands occupying drainage divides. RP6–8resistant, on the other hand, decreases more substantially, reflecting enhanced river diversion around resistant lithologies. Similar trends are observed in RPerodible, further suggesting that greater relative erodibility exerts a stronger influence on the river network. Moreover, relative erodibility is not the only factor influencing RP. As shown by the intersecting colored lines in Fig. 4B, variations in the distribution of resistant rocks can partially offset the effects of relative erodibility. Nevertheless, the overall trend confirms that relative erodibility remains the dominant control on RP, outweighing the influence of lithological spatial arrangement. The timing of river diversion further supports this conclusion. In Fig. 4C, the blue rectangles and lines (shaded from light to dark with increasing relative erodibility) show that both the minimum and average diversion times decrease as relative erodibility increases, for a given CAO. In addition, the maximum contributing area of diverted channels also expands with higher relative erodibility, indicating that larger rivers can eventually be rerouted when lithological contrasts are stronger. To test the limits of this behavior, we conducted an additional experiment based on the reference model shown in Fig. 3. Keeping all other parameters constant, we increased the relative erodibility from 3 to 40, comparable to the ∼65-fold difference between weak fine-grained sedimentary rocks and strong massive basalt along the southeast escarpment of Brazil (10). In this extreme case, lithological contrasts were sufficient to induce diversion even in channels with CAO = 10 in 30 Myr (fig. S8). This response differs from the minor deviations of RP9–10resistant in other cases, which mainly reflect upstream network reorganization. In the extreme-contrast case, RP9–10resistant shows a pronounced decrease because trunk-scale rivers themselves are diverted around resistant lithologies. As a result, the RPresistant-CAO trend changed from a V-shaped pattern to a monotonic decrease pattern, indicating large-scale avoidance of resistant rocks throughout the river network. RP in the eastern Tibetan Plateau We calculate the RP of plutonic and sedimentary rocks in the Min River and Dadu River basins along the eastern Tibetan Plateau (represented by dark blue and dark pink dashed lines in Fig. 5), which together account for over 95% of the study area using the 250-m resampled DEM (see Materials and Methods). In addition, to better compare with the simulation results, we extracted the RP of resistant and erodible rock on the basis of the lithological reclassification, which together account for over 99% of the study area (represented by light blue and light pink lines in Fig. 5). RP versus CAO derived from the 250-m resampled DEM in the eastern Tibetan Plateau. (A) Min River drainage. (B) Dadu River drainage. Light pink lines represent erodible rock, dark pink dashed lines represent sedimentary rock, light blue lines represent resistant rock, and dark blue dashed lines represent plutonic rock. The bars represent the maximum and minimum RP values across different DEM resolutions (30, 90, 250, and 500 m). The RP-CAO trends of the Min River and Dadu River drainage align with the model results. For resistant rock and plutonic rock, RP initially decreases with increasing CAO, followed by a subsequent rise as CAO continues to increase, resulting in a V-shaped distribution. In contrast, for erodible and sedimentary rocks, RP initially increases with the rise in CAO but then decreases as CAO continues to increase, leading to an inverted V-shaped distribution. To assess the impact of resolution, we additionally calculate RP at resolutions of 30, 90, and 500 m (see Materials and Methods). Given that the resolution did not affect the overall RP-CAO trend, with deviations in specific values remaining within 30% (bars in Fig. 5), we conclude that DEM resolution has no substantial influence on the overall RP-CAO trend. However, this resolution test only evaluates the effect of DEM resolution and does not capture uncertainties associated with the mapping scale of the underlying geologic map. Uncertainty in the positions of lithological contacts, which is rarely quantified in published maps, may affect local rock-type assignments and therefore the absolute RP values, particularly near contacts or narrow lithological units. Relative erodibility in the eastern Tibetan Plateau We conducted Schmidt hammer rebound value (SHV) measurements and find that values from regions classified as “erodible rocks” are generally lower than those from regions classified as “resistant rocks” (Fig. 6, A and B), consistent with findings in (29) and with the expected negative relationship between SHV and erodibility (K), K∝1/SHV (20). In addition, we calculated and analyzed the normalized channel steepness (ksn) for the Min River and Dadu River basins (see Materials and Methods). Our results indicate that ksn values in erodible rock regions are generally lower than those in resistant rock regions (Fig. 6, C and D), again consistent with the expected negative relationship between ksn and erodibility K (30), and aligning with the SHV-based results (Fig. 6, A and B). SHVs and normalized channel steepness in the eastern Tibetan Plateau. (A and B) Dots represent the average SHV at each site, with the whiskers indicating the SD. Hollow circles indicate the mean SHV for reclassified lithologies, with the whiskers representing the SD. (C and D) Bar charts represent the distribution of normalized channel steepness (ksn). Blue represents resistant rock, and pink represents erodible rock. To quantify this erodibility difference, we used the data of (29), which quantified erosion coefficients by examining the relationship between millennial-scale erosion rates and normalized channel steepness (ksn). On the basis of data from (29), the mean and median relative erodibility between erodible and resistant rocks in the eastern Tibetan Plateau were found to be 1.79 and 1.23, respectively (fig. S9A), which is comparable to the modeled scenarios with relative erodibility values of 1.5 and 2 used in our numerical experiments. DISCUSSION Capturing lithological influence on drainage networks using RP-CAO Since the present-day drainage network reflects only a transient snapshot in the long-term evolution of the landscape, it remains challenging to directly infer the influence of lithology on river network reorganization solely from topography and the current river configuration. On the basis of numerical simulations, we propose the RP-CAO trend as a useful tool for quantifying the influence of lithological contrasts on drainage networks. The modeling results demonstrate that lithological heterogeneity promotes the development of V-shaped RP-CAO curves over time, specifically capturing the tendency for low-discharge rivers to divert around resistant lithologies and promote the migration and persistence of drainage divides over them, while high-discharge rivers tend to maintain their courses by incising through the exposed resistant rocks. Field observations from both the Dadu and Min river basins provide compelling evidence for the role of lithological contrasts in shaping drainage patterns. In these regions, several prominent plutonic bodies are located along major drainage divides, suggesting that river courses have been deflected around resistant lithologies. For example, the Yanggonghai and Rilongguan-Niuxinggou plutons are situated near the northern and southern ends, respectively, of the main divide between the Dadu and Min Rivers (fig. S10). Within the Min River basin, tributaries similarly divert around the Mengtonggou and Laojungou plutons, which coincide with divides between adjacent sub-basins (fig. S10). These spatial patterns highlight a consistent tendency for low-discharge channels to be redirected around resistant rock, leading to the migration of drainage divides. In contrast, portions of the Dadu and Min mainstems incise directly through resistant plutons. Notably, the Dadu River cuts across the Markam and Menggu-MaNai granites, while the Min River incises through the Pengguan complex (fig. S10). These examples illustrate the complementary tendency for high-discharge rivers to maintain their course by eroding through resistant lithologies. RP-CAO analyses from both river basins reveal V-shaped distributions (Fig. 5), consistent with the trends observed in our numerical simulations (Fig. 4B). This agreement between model predictions and field-based RP-CAO patterns supports the use of the RP-CAO trend as a quantitative tool for identifying and assessing the influence of lithological heterogeneity on drainage network organization. Notably, the RP-CAO approach proves effective even under relatively modest erodibility contrasts. Model results and RP-CAO patterns from the eastern Tibetan Plateau both show that lithological influence on river network structure can be detected when the relative erodibility is as low as 1.5 to 2. This value is lower than the threshold of ∼3 below which Bernard et al. (17) concluded that basement intrusions no longer effectively capture drainage divides or reorganize channel networks. Notably, a closer examination of their results reveals that drainage networks still adjust under lower erodibility contrasts (see their figures S7 and S8), consistent with our findings. Together, these observations suggest a shared conclusion: Drainage networks are highly sensitive to lithological variations, and the RP-CAO metric thus provides a useful quantitative framework for capturing these subtle but meaningful reorganizations across a range of erodibility contrasts. These findings also clarify how RP-CAO complements existing methods for analyzing drainage reorganization. Methods such as χ analysis (31) and Gilbert metrics (32) are useful for evaluating drainage disequilibrium, divide instability, and possible divide migration. RP-CAO instead evaluates whether lithological contrasts have influenced drainage-network organization by quantifying how rock types are distributed along rivers with different discharge levels. It can therefore be used to test whether lithological contrasts have influenced the drainage network, especially in regions where DEMs and lithological maps are available but direct evidence for past drainage reorganization is difficult to recover. Methodological considerations and interpretive limits of RP-CAO While RP-CAO provides a framework for quantifying the influence of lithological variation on drainage network organization, several methodological and interpretive limitations should be considered when applying this approach to natural landscapes. In the numerical simulations presented above, resistant rock bodies are introduced after the landscape has reached a quasi–steady state, and the initial RP values are set to approximately 1. This design allows us to isolate the response of an established drainage network to lithological contrasts, but it may introduce path dependence related to both the initial topographic relief and the initial planform geometry of the drainage network. We therefore conducted two additional experiments to evaluate these effects separately. First, to test the effect of initial relief, we retained the planform geometry of the quasi-steady drainage network but compressed the initial elevation range to 0 to 1 m before lithological contrasts were imposed. The RP-CAO relationship still showed a V-shaped pattern, indicating that large rivers continued to incise through resistant rock bodies. The reduced topographic relief favored river diversion, leading to larger RP changes than in the reference case (fig. S11, A and B). Second, to test the effect of initial drainage-network geometry, we imposed the same lithological configuration on the original low-relief initial surface used to generate the quasi-steady landscape (see Materials and Methods). This setup produced a different initial drainage network and nonuniform initial RP values (see black dashed line in fig. S11C). After 30 Myr, the RP-CAO curve still showed a V-shaped pattern, indicating that rivers with larger contributing areas continued to incise through resistant rock bodies (fig. S11C). However, because the initial drainage network and lithological configuration jointly controlled the initial RP values, absolute RP values differed from those in the main simulations, with RP10resistant far exceeding 1 (fig. S11C). These tests show that initial relief and drainage-network geometry can affect the timing and magnitude of river diversion, but they do not eliminate the scale-dependent response of rivers to lithological contrasts captured by RP-CAO. This provides a mechanistic basis for interpreting natural landscapes, where the initial drainage-lithology configuration is generally unknown. Therefore, the clear V-shaped RP-CAO trends observed in the eastern Tibetan Plateau more likely reflect the lithological imprint on drainage organization rather than being primarily controlled by unknown initial conditions. This interpretation also helps explain why RP10resistant in the eastern Tibetan Plateau is far above 1, whereas it remains close to 1 in most simulations. Because the main simulations constrain the initial RP distribution to be approximately 1, this difference likely reflects the nonuniform initial drainage-lithology configuration in the natural landscape rather than a different river response to lithological contrasts. In most simulations presented here, hillslope diffusivity is kept spatially uniform. This setup simplifies hillslope processes and allows the main experiments to isolate the influence of lithological contrasts in fluvial erodibility on drainage-network organization. However, lithology may also influence hillslope diffusivity by affecting soil production and sediment transport on hillslopes. Existing work shows that this diffusivity can differ between lithologies, but the relationship remains difficult to generalize because it is not directly controlled by lithology alone (33). Series 6 therefore provides a simplified sensitivity test that introduces lithology-dependent hillslope diffusivity by reducing it in resistant rock bodies by the same relative factor as fluvial erodibility. The persistence of the V-shaped RP-CAO trend in this experiment indicates that the core scale-dependent response captured by RP-CAO remains robust when hillslope diffusivity is allowed to vary with lithology, although the magnitude of RP changes and the timing of river diversion are modified. Better constraints on lithology-dependent hillslope diffusivity would help improve future iterations of the RP-CAO framework. Additional complexity arises when multiple resistant lithologies are exposed within a single catchment. In our models, we observe that even in the absence of river diversion around resistant bodies at high CAO (e.g., CAO = 10), RP values in those classes can still vary because of upstream reorganization. This occurs when rivers at intermediate CAO levels adjust their courses in response to lithological variation, thereby altering downstream discharge and contributing area distributions. These effects are usually limited (under 20%), but they show that RP values can be shaped by overall network changes, not only by local differences in rock erodibility. In the simulations presented here, resistant rocks are represented as time-invariant discrete bodies, which helps isolate the effect of lithological contrast on river-network evolution. This model setup is therefore most directly applicable to landscapes where resistant units occur as plutons or other laterally bounded bodies, and applications to more complex lithological configurations require cautious interpretation. Layered or dipping stratigraphy is one example, because resistant and erodible units are not isolated bodies but laterally continuous layers. In such stratified settings, the influence of lithology on river-network evolution may become more complex. Where lithological contacts are inclined or stratified, erosion and exhumation can progressively change the surface distribution of lithology (8). This can lead to prolonged adjustment periods as drainage systems respond to evolving surface exposures (8). Previous work has shown that such processes may delay the attainment of geomorphic steady state over timescales of millions of years (20). In addition, layered stratigraphy may create some degree of autocorrelation between drainage area and lithology, because the dip and geometry of strata can confine exposures of particular layers to similar parts of the landscape in terms of drainage area. Laterally extensive resistant layers may also reduce the capacity for drainages to reorganize around resistant units, instead promoting incision through, or adjustment along, these continuous resistant layers. Beyond this planform architecture, the vertical extent of resistant bodies may also influence whether rivers divert around them or incise through them. The diversion times quantified in our simulations (Fig. 4C) can be interpreted as the timescale required for river rerouting. For rerouting to be completed, resistant bodies must have sufficient vertical extent to sustain the surface lithological contrast throughout this adjustment time. For example, assuming a simple steady-state condition in which erosion balances the model uplift rate of 0.2 mm/year, diversion times of ∼1 to 60 Myr for CAO = 7 rivers at a relative erodibility of 5 correspond to minimum vertical scales of ∼0.2 to 12 km. Conversely, very long diversion times would require unrealistically thick resistant bodies to maintain this contrast during exhumation, making complete rerouting unlikely and favoring channel incision. These considerations suggest that RP-CAO reflects not only lithological contrasts but also the geometry, vertical extent, and persistence of resistant bodies during landscape evolution. Future work should therefore test how different lithological configurations modulate the scale-dependent response of rivers to lithological contrasts, thereby expanding the application of RP-CAO to more complex lithological settings. Lithological control under complex forcing Tectonic uplift, climatic variation and fault activity are widely recognized as key drivers of topographic evolution and drainage network reorganization (29, 34–38). In the eastern Tibetan Plateau, these nonlithological factors vary markedly across space and substantially shape drainage organization. Regional precipitation varies from ∼500 to 900 mm/year across the study area (fig. S1), representing nearly a twofold difference in runoff potential. Active faulting adds further complexity: Within 15 km of major active fault zones, average erodibility is 2.85 times higher than in more stable surrounding areas (fig. S9D). Relative erodibility is also amplified near faults, with values approximately 1.2 times higher within 15 km of major faults (fig. S9, B and C). Independent low-temperature thermochronological evidence also indicates substantial spatial contrasts in late Cenozoic exhumation, with stronger exhumation in the Longmen Shan, where major fault systems are concentrated, than in the eastern Songpan-Ganzi terrane (39). The influence of fault structures is also evident in river-network geometry. For instance, the Min River exhibits a ∼60-km deflection along the Wenchuan-Maoxian fault trace (fig. S12). Similar fault-controlled alignments are observed along the Dari, Xueshan, and Longriba fault systems, where river courses follow fault traces over extended distances. These examples show that fault structures can guide river orientation and locally reorganize drainage pathways. Together, the regional precipitation gradient, spatial contrasts in exhumation, the fault-related increases in both average and relative erodibility, and these fault-controlled river geometries provide a natural basis for testing whether the lithological signal captured by RP-CAO remains detectable under strong nonlithological forcing. Considering the complex spatial forcings described above, we conducted additional numerical simulations incorporating spatial gradients in uplift and precipitation, as well as fault-zone weakening, to assess the robustness of the RP-CAO trend. Within the test cases, the differences in erodibility from nonlithological factors (≥3×) are greater than those from lithological contrasts (1.5×). The RP-CAO trend remains remarkably stable across simulations that include these complex forcings (fig. S13). These results indicate that RP-CAO can preserve lithological signals even under stronger nonlithological perturbations. The field-derived RP-CAO patterns from the eastern margin of the Tibetan Plateau extend this test to natural landscapes that are more complex than our numerical experiments. In the Dadu and Min river basins, the agreement between simulated and field-derived RP-CAO patterns suggests that the first-order lithological signal captured by RP-CAO remains detectable despite additional natural processes not explicitly represented in the model. These processes may influence local channel geometry and absolute RP values, but they do not obscure the scale-dependent RP-CAO trend. This comparison also highlights opportunities for further developing the RP-CAO framework. In the current simulations, fault damage is represented as a prescribed zone of enhanced erodibility, rather than through the explicit evolution of rock damage or stress-dependent erodibility. Similarly, although our model incorporates spatially variable precipitation, precipitation is represented as a time-invariant relative scaling factor and does not include temporal variability in monsoon precipitation, which may influence erosion and sedimentation processes (40). Future iterations could build on the present results by incorporating time-dependent precipitation, extreme-event forcing, and multitimescale constraints from structural geology, cosmogenic nuclide analysis, and low-temperature thermochronology. These additions would help better account for nonlithological controls and further refine the isolation of lithological signals in drainage-network organization. In addition, our present model does not explicitly account for sediment effects because it follows a detachment-limited stream-power formulation. This simplification is appropriate for the eastern Tibetan Plateau, where field observations show that most channels are bedrock dominated with limited sediment (fig. S14). Sediment is therefore unlikely to be the primary control on river-network evolution and RP-CAO patterns in our study area. However, sediment may matter in sediment-rich catchments, where it can either shield the bed from erosion through cover effects or enhance incision through tool effects (41). In lithologically heterogeneous landscapes, sediment-related effects may modulate how lithological contrasts are translated into channel incision and drainage evolution (42). Future applications of RP-CAO in such settings should therefore evaluate whether sediment systematically modifies bedrock exposure and effective erodibility, to better isolate lithological signals in a drainage-network organization. Lithological contrasts can exert a first-order control on fluvial landscape evolution over long timescales (105 to 107 years), in both tectonically active and stable regions (e.g., 14, 19, 20). For example, in the postorogenic Appalachian Mountains, spatial contrasts in rock erodibility between adjacent catchments triggered a river capture event. This capture caused a rapid ∼150-m base-level fall at 9 ± 3 Ma, which in turn led to accelerated incision and reorganization of the upper Tennessee River basin. The resulting landscape response persisted over million-year timescales, illustrating how lithological heterogeneity sustains nonsteady-state erosion and drive long-term landscape evolution (14). Changes in erosion patterns or drainage reorganization driven by spatial variations in rock erodibility can further influence biological and ecological evolutionary processes (14, 43). When considered alongside the examples in this study, we infer that spatial variations in rock erodibility may play a sustained role in shaping long-term landscape evolution in the tectonically and climatically complex eastern Tibetan Plateau by steering drainage reorganization and associated incision and sediment-routing patterns, with potential implications for sediment dynamics and ecological connectivity. This aspect warrants targeted investigation in future work. Overall, our work demonstrates that spatial differences in rock erodibility can shape river network structure, modulate erosion, and drive long-term landscape evolution. MATERIALS AND METHODS Landscape evolution model We used the same LEM as (44) to investigate the impact of lithological contrasts on topography and drainage networks. The topographic evolution in the model is primarily controlled by processes of uplift, hillslope diffusion, and river incision (Eq. 3). The hillslope diffusion is governed by a linear diffusion law (45), while the fluvial incision is modeled using the stream power law (46)∂z∂t=Kd∇2z−K(p∼A)mSn+U(3)where z is elevation, t is time, Kd is the diffusion coefficient, K is the incision coefficient (erodibility), and p∼ is the dimensionless relative precipitation rate, defined as the spatial or temporal deviation in precipitation normalized by the mean value over the study area. Unless otherwise specified, p∼ is set to 1. A is the contributing area that is used as a proxy of discharge, and S is the local slope, m, n are dimensionless exponent parameters, and U is the uplift rate. We set up a landscape model with a length of 110 km and a width of 170 km. The horizontal and vertical numbers of grid nodes are 440 and 680, respectively, to keep a uniform spatial resolution of 250 m. The common parameters used in the model are adopted from (47) and (48), which are applicable to the eastern Tibetan Plateau. We use an area exponent (m) of 0.45 and a slope exponent (n) of 1.0, giving a concavity index (θ = m/n) of 0.45. To promote the development of a dendritic river network, the initial topography is constructed as a southeastward-tilting slope with random perturbations to simulate natural variations, with elevations constrained between 0 and 1 m. All model boundaries were closed except for the southeastern outlet. The model was run with a uniform erodibility coefficient of 1.11 × 10−6 m0.1 year−1, a hillslope diffusion coefficient of 1 × 10−2 m2 year−1, and a uniform uplift rate of 2 × 10−4 m1 year−1 until a topographic quasi-steady state (∣dz/dt∣<1×10−9 m year−1) was reached, representing a preexisting drainage network unaffected by lithological contrasts (movies S5 and S6). This quasi–steady-state topography is then used as the baseline for subsequent investigations into the impact of lithological variations. We then introduce a set of lower-erodibility rock bodies, represented as time-invariant vertical columns with elliptical cross sections in map view, while the rest of the domain retains the initial, higher erodibility and then continue the simulation. To generate these resistant blocks, we applied a procedure that produces resistant bodies that are geometrically randomized and that collectively form an approximately uniform initial RP-CAO distribution. Using the contributing area from the quasi–steady-state topography, we first calculated the CAO and quantified the number of cells within each order. In the experiments, 10% of all cells were designated as resistant lithology, and the quota of resistant cells for each CAO was calculated on the basis of the number of cells for each CAO. Resistant patches were then iteratively generated as random trial ellipses, with ellipse centers selected at random from the highest CAO nodes that had not yet met their quota. The remaining parameters of each ellipse, including major axis length (<10 km), axis ratio (0.3 to 1.0), and orientation (0 to 360°), were also randomly assigned. Trial ellipses were accepted only if they did not exceed the remaining quota in any CAO. The iteration proceeded until the resulting RP values in any CAO converged within 1 ± 0.1. On the basis of this setup, a suite of numerical experiments was conducted and grouped into 10 series (table S1). Series 1 to 4 examine the effects of different relative erodibility (Kerodible/Kresistant) and spatial configurations of resistant rocks. Series 5 assesses the effects of grid spacing, using grid sizes of 100 and 500 m. Series 6 examines the effects of lithology-dependent hillslope diffusivity. In this series, both the fluvial erodibility and hillslope diffusion coefficient were reduced in resistant rock bodies by the same relative contrast, providing a simplified sensitivity test of whether RP-CAO trends persist when hillslope transport also varies with lithology. Series 7 evaluates the effects of a nonlinear stream-power formulation with n = 2. This series was run with erodibility coefficient of 4.01 × 10−9 m−0.8 year−1, an area exponent (m) of 0.9 and a slope exponent (n) of 2.0, adopted from (49, 50). Series 8 explores model sensitivity to a highly amplified erodibility contrast. Series 9 examines the effects of initial topography and drainage network. To test the effect of initial topographic relief, series 9-1 kept the planform geometry of the quasi–steady-state drainage network but compressed the elevation range to 0 to 1 m before lithological contrasts were imposed. To test the effect of a nonuniform initial RP-CAO relationship, series 9-2 imposed the same lithological configuration on the original low-relief initial surface used to generate the quasi–steady-state landscape, without constraining initial RP values to be approximately 1. Series 10 investigates the influence of lithological contrasts under more complex conditions by introducing spatially variable uplift, rainfall, and fault damage. Figure S15 shows the resistant lithology distributions for all series, and fig. S16 shows the uplift rate, relative-rainfall, and erodibility distributions for series 8 and 10. Field comparison: Eastern Tibetan Plateau We then compare the model results with analogous sites in the eastern Tibetan Plateau. We analyze the topography of the region using LSDTopoTools, a software package designed for topographic analysis (51, 52). To better match the resolution used in the numerical model, we resample the 90-m COP90 data (data from https://opentopography.org/) to 250-m resolution and then extract rivers and calculate contributing areas across the study area. On the basis of the reclassified lithology, we compute the CAO and the RP for the Min River and Dadu River drainage on the eastern Tibetan Plateau, following the same approach used in the numerical model. To evaluate the impact of grid resolution on RP, we also calculate RP using datasets at 30 m (original COP30), 90 m (original COP90), and 500 m (resampled from COP90). We measured the rebound value of 31 bedrock outcrops using a Schmidt hammer along the trunk streams and major tributaries of the Min River and Dadu River (Fig. 1). For each outcrop, more than 50 rebound values were measured, and the mean value was taken as the representative rebound value for each site. On the basis of the lithological reclassification scheme described above, each site was categorized as either resistant or erodible, with 16 sites categorized as resistant and 15 as erodible. The mean rebound values for resistant and erodible rocks were calculated. The normalized channel steepness index (ksn) was calculated for both the Min River and the Dadu River using LSDTopoTools. In addition, the ksn values were categorized on the basis of rock type, resulting in the distribution of ksn for resistant and erodible rocks, respectively.