At ∼ 58–60 Ma, the collision of India and Asia initiated and shortening progressed southward to form the Himalayan-Tibetan orogenic system (e.g., DeCelles et al., 2014; Gansser, 1964). The Himalayan orogen is commonly divided into four tectonostratigraphic zones bounded by major faults: the Subhimalaya, Lesser Himalaya (LH), Greater Himalaya (GH), and Tethyan Himalaya (e.g. Gansser, 1964). From south to north, the Main Frontal thrust (MFT) places the Subhimalaya over the modern foreland, the Main Boundary thrust (MBT) juxtaposes LH rocks over the Subhimalaya, the Main Central thrust (MCT) places GH rocks over LH rocks, and the Southern Tibetan Detachment is the contact between GH and Tethyan Himalayan rocks (Fig. 1). The major faults sole into the Main Himalayan thrust (MHT), the basal décollement. In western Nepal, the LH is segmented into lower and upper units by an orogen-scale fault, the Ramgarh-Munsiari thrust (RMT) (Fig. 1).

Along A-A^′, the Siwalik Group comprises ∼ 6.3 km of synorogenic sedimentary rocks, with ∼ 2 km in the lower unit, 2.74 km in the middle unit, and ∼ 1.5 km in the upper unit (Fig. 2) (Gautam and Fujiwara, 2000; Sigdel and Sakai, 2016). The lower Siwalik unit was deposited from ∼ 16 to ∼ 9.6 ± 1 Ma (Gautam and Fujiwara, 2000). Near the top of the lower Siwalik unit, εNd values become distinctly more negative (Fig. 2c) and indicate exhumation of lower LH rocks to the surface and initial influx of LH-derived detritus to the foreland basin at ∼ 10 Ma (Huyghe et al., 2001; Quade et al., 1997; Robinson et al., 2001, 2006; Szulc et al., 2006). The basal ∼ 1.5 km of the middle Siwalik unit accumulated from ∼ 9.6 ± 1 to ∼ 5 ± 1.5 Ma and the upper Siwalik unit was deposited after ∼ 4–5 Ma (Gautam and Fujiwara, 2000). Detrital MAr age distributions measured in the Siwalik unit generally show a prominent cluster of ages ∼ 15–30 Ma, with a distinct peak ∼ 14–16 Ma (Fig. 2d) (Szulc et al., 2006). The three deepest samples at ∼ 5.4–6.2 km basin depth show detrital MAr age distributions that extend to > 40 Ma (Fig. 2d) (Szulc et al., 2006).

To the east, estimates of the Siwalik Group thickness increase along strike as a function of map pattern width and variability in orientation measurements (Fig. 1). Along B-B^′, the Siwalik Group is estimated as ∼ 7.5–8.5 km thick, with a ∼ 2.5 km thick lower unit, ∼ 3–4 km thick middle unit, and ∼ 2–2.5 km thick upper unit (Fig. 2e).

Published cooling ages utilized include 21 MAr ages, 31 ZHe ages, 24 AFT ages, and 24 AHe ages (Fig. 3a) (Braden et al., 2020; DeCelles et al., 2020; Harvey and Burbank, 2024; McCallister et al., 2014; Mercier, 2014; Sakai et al., 2013; Sherpa et al., 2023; Soucy La Roche et al., 2016, 2018; van der Beek et al., 2016). MAr ages include single grain and multigrain aliquots and are presented with 2σ analytical errors. ZHe and AHe ages are the mean of 2–8 single grain ages, with a two standard deviation error. AFT ages are presented as a central age of 10–50 single grains and 2σ error (Figs. 1 and 3a). The datasets have been screened for outliers and, while some age-eU relationships exist (e.g. Sherpa et al., 2023), these relationships are not always present.

Geologic mapping was completed at a 1:200000 scale using the tectonostratigraphic framework of Robinson et al. (2006). Our mapping was incorporated with observations from Robinson et al. (2006), DeCelles et al. (2020), Soucy La Roche et al. (2016, 2018), and Mugnier et al. (1999) to connect contacts along strike (Fig. 1). The updated mapping, stratigraphic thicknesses determined from the map distance and orientation, and foreland basin thickness were integrated into a balanced cross-section to describe the subsurface structures for A-A^′. The transect line for A-A^′ is oriented parallel to the direction of convergence determined from GPS measurements in western Nepal (e.g., Ader et al., 2012; Jouanne et al., 2004) and positioned to minimize the distance needed to project the samples and measurements in the Subhimalaya onto the section. The measured data was projected onto the transect line following the strike of the major structures and contacts. The location of décollement ramps was constrained from the distribution of young cooling ages and elevated channel steepness values (Fig. 3) (see Sects. 3.2, 3.3, 4.3 and 4.4). The cross-section was balanced and restored to its undeformed state using the line-length method (Dahlstrom, 1969). Ductile deformation, penetrative strain, or small-scale folding is not accounted for in the cross-sections. Granular distortions that have been identified largely record layer-normal flattening (Braden et al., 2020; Ghoshal et al., 2020; Yakymchuk and Godin, 2012). In other parts of the Himalaya, this flattening has been interpreted to have occurred on the prograde burial path for the LH rocks and preceding or contemporaneous with fault displacement for the GH rocks (e.g., Corrie et al., 2012; Law et al., 2013; Long et al., 2016; Long and Kohn, 2020). We do not incorporate internal deformation of units in the cross-sections and flexural-kinematic models we present here, with faults treated as discrete structures and thrust sheets as rigid bodies. Distributed ductile shearing is a required component of deformation when there is evidence of marked changes in unit thickness, temperature or pressure gradients that cannot be reproduced with models of discrete fault structures (Braza et al., 2023; Corrie et al., 2012; Long et al., 2016; Long and Kohn, 2020). The integration of the cross-section geometry and kinematics with a thermal model allows us to compare the modelled temperatures with published temperature estimates to assess our assumption that motion along discrete structures exert a first-order control on the burial and exhumation pathways of rocks in western Nepal (e.g. Braza et al., 2023).

To examine the spatial distribution of recent uplift, the normalized channel steepness index (ksn) was calculated from a ∼ 30 m ASTER DEM (US/Japan ASTER Science Team, 2019) using TopoToolbox v2 (Schwanghart and Scherler, 2014). The ksn values were calculated using a 0.45 reference concavity (θref) (e.g. Kirby and Whipple, 2012). Median ksn values were calculated for a 40 km wide swath bounding A-A^′ and projected onto the center line. A smoothing window of 5 km was applied to the median ksn values.

New thermochronologic data is presented for 13 ZHe and 6 AFT samples (Table 1, Fig. 1) (Braza et al., 2026a, b). Analyses were carried out at the University of Arizona Radiogenic Helium Dating Laboratory for the ZHe samples (Text S1 in the Supplement) and at Occidental College Fission Track Research Laboratory for the AFT samples (Text S2 in the Supplement).

Flexural-kinematic forward modelling of the balanced cross-sections was completed using Move (Petex), following established procedures (Braza and McQuarrie, 2022a; Ghoshal et al., 2020; McQuarrie and Ehlers, 2015, 2017) and using the parameters in Table 2 (see Text S3 in the Supplement for further details). The model initiates with the top of the Gondwana Sequence at 0 m elevation at the MFT and the topography increases at a ∼ 2° angle over ∼ 160 km distance to a maximum of 5 km elevation above the restored GH rocks. GH rocks restore to ∼ 35–45 km depth with an initial sedimentary taper of ∼ 0.5°, consistent with the published peak temperature conditions (e.g. Fig. S1 in the Supplement) (Braden et al., 2020; Soucy La Roche et al., 2018). This initial topographic profile, taper angle, and GH burial depth account for the Paleocene-Oligocene shortening in the Tethyan fold-thrust belt prior to initiation of motion on the MCT (e.g., Aikman et al., 2008; Murphy and Yin, 2003; Ratschbacher et al., 1994; Soucy La Roche et al., 2016). Cooling and decompression proposed for GH rocks that extends from initial model start times (25–30 Ma) to as young as 16 Ma was evaluated by assessing if additional cooling or thinning (e.g. Soucy La Roche et al., 2018) was needed to reproduce MAr cooling following peak temperatures. Flexural-kinematic models are required to reproduce the surface geology, basin thickness, and depth to basin provenance indicators (Braza and McQuarrie, 2022a; McQuarrie and Ehlers, 2015, 2017).

Viable flexural-kinematic models were thermally modelled using Pecube-D (see Text S4 in the Supplement for additional details) (Ehlers, 2023; McQuarrie and Ehlers, 2015; Whipp et al., 2009), a modified version of the advection-diffusion code Pecube (Braun, 2003; Braun et al., 2006). Each ∼ 10 km deformation step was assigned an age to convert the Move displacement fields to velocity fields (McQuarrie and Ehlers, 2015). Constant and variable shortening rates were tested. Constant shortening rates of 24–35 mmyr-1 were applied to initiate MCT motion at 25–30 Ma (e.g. Soucy La Roche et al., 2018). For the variable velocities, the specific shortening rates are determined from the total amount of displacement on a given fault and the increment of time that fault displacement occurs over. The timing of the structures is constrained by the distribution of cooling ages, which sets requirements for the amount and age of exhumation that occurs at a given sample location, and from the depth and depositional ages in the foreland basin (Braza and McQuarrie, 2022a, b; McQuarrie and Ehlers, 2015, 2017). The velocity fields and the specified thermal and crustal parameters (Table 2) are used by Pecube-D to solve the advection-diffusion heat equations to predict the subsurface thermal field for each Move displacement step (Braun, 2003; Braun et al., 2006; Ehlers et al., 2005; McQuarrie and Ehlers, 2015). The predicted time-temperature histories, predicted cooling rates, and prescribed mineral kinetics (Dodson, 1973; Ehlers et al., 2005; Hames and Bowring, 1994; Hodges, 2014; Ketcham et al. 2007; Reiners et al., 2004) are used by Pecube-D to calculate a suite of cooling ages for points at the surface (e.g., Braun, 2003; Braun et al., 2006; Ehlers, 2023) independent of any measured cooling age (Text S4). The measured cooling ages are then compared to and used to evaluate the predicted cooling ages output from Pecube-D. The predicted detrital MAr signal associated with each increment of displacement and basin deposition is calculated following the methods of Braza and McQuarrie (2022b). Detrital MAr age distributions are a function of source rock composition, transport, sorting, attrition, and weathering (e.g. Malusà and Fitzgerald, 2020), complexities that methods for predicting detrital cooling ages cannot account for. Because of this, our goal is to reproduce the distribution of measured detrital MAr cooling ages, rather than the magnitude or specific peak locations of the distribution.

Thermokinematic forward models are evaluated by their ability to reproduce the measured cooling ages (119 ages) and basin accumulation data (33 constraints). The predicted cooling ages and ± 0.5 Ma applied error are required to overlap with the measured age and associated ± 2σ error. A lateral error of ± 0.5 km is applied for the measured cooling data to account for uncertainty from projecting the samples along strike onto the transect line. The percent fit of the predicted cooling ages for each model is calculated by dividing the number of matched measurements by the total number of age measurements, for each chronometer (MAr, ZHe, AFT, and AHe) and the modelled cooling ages as a whole (Braza and McQuarrie, 2022a). Models are also required to match the measured basin thickness, depth of provenance indicators, and depositional ages (Braza and McQuarrie, 2022a). The modelled basin is required to have thicknesses within ± 0.5 km of the cross-section basin thickness, due to the variability in the orientations, map width, and number of faults deforming the Siwalik Group (e.g. Fig. 1), and depositional ages that fall within the measured age range at an equivalent depth (Braza and McQuarrie, 2022a). The percent fit of the basin is calculated by dividing the number of matched measurements by the total number of depth and depositional age constraints (Braza and McQuarrie, 2022a). The percent fits for the cooling ages and basin accumulation are equally weighted to determine the total model fit.

Model C is also assessed by evaluating the predicted peak temperatures against measured temperature estimates (see Text S4 in the Supplement, Fig. S1 for further details) (Braza et al., 2023). The ability to reproduce the measured peak temperature gradients with thermokinematic models that do not incorporate distributed shear transport does not preclude the presence of ductile processes, but suggests that any associated ductile flattening of the rocks occurred on the prograde burial path and synthrust shearing and/or vertical thinning is minor or secondary to the discrete fault displacements (Braza et al., 2023). While horizontal extension in response to vertical thinning may contribute additional displacement that is not quantified in the thermokinematic models, the amount of displacement imparted by transport-parallel elongation is expected to be limited if the models are able to replicate the measured peak temperatures and resulting cooling histories (e.g. Braza et al., 2023).

We present model results for three ramp geometries. We test various kinematic sequences and shortening rates for each geometry to find the best-fit to the measured datasets and identify a viable exhumation pathway. Models A and B test the cross-section geometries and kinematic sequences proposed by Robinson et al. (2006) and Olsen et al. (2019) for B-B^′. Results from these models lead to the changes in the geometry, kinematic sequences, and shortening rates tested with Model C for the revised cross-section along A-A^′ (Fig. 3). The final geometry, kinematics, and shortening rates presented for Model C are the product of more than 30 flexural-kinematic models and 60 thermal models that tested the influence of different ramp locations, sizes, and angles for ramps in the upper LH units, including the modern active ramp at ∼ 13 km north of the MCT; in- and out-of-sequence kinematics for the RMT fault and MCT in the north; out-of-sequence fault beneath the Dadeldhura klippe; and a suite of constant and variable velocities.

Thermokinematic model versions of Model C with a total fit of > 80 % were input into a modified version of the landscape evolution model CASCADE (Braun et al., 1999; Braun and Sambridge, 1997; Yanites and Ehlers, 2016) to explore the influence of fault geometry and kinematic sequence on modelled topography and ksn values. Landscape evolution modelling followed the methods detailed in Eizenhöfer et al. (2019) and utilized the parameters in Table 2 (Text S5 in the Supplement). The modified version of CASCADE includes horizontal and vertical advection components that are derived from the input flexural-kinematic model grids (Eizenhöfer et al., 2019). Constant and variable (orographic) precipitation fields were tested (Text S5, Fig. S2 in the Supplement). The elevation profile and median ksn values were extracted for a 50 km wide swath of the modelled topographic surface using TopoToolbox v2 (Schwanghart and Scherler, 2014). The ksn values were calculated with θref = 0.5, the reference concavity inherent in CASCADE, and projected onto the center line to determine the median values. Integrating the landscape evolution and thermokinematic models is an iterative process whereby the results of the landscape models and required changes to geometry and/or location of recent uplift to better replicate the mean elevation along the section line and the across-strike pattern of ksn, are then evaluated in the thermokinematic model.

The geologic map of western Nepal is shown in Fig. 1b. Differences in the geologic mapping of this study and Robinson et al. (2006) are based on new observations and orientation measurements. The exposures of lower LH units in the southern limb of the Dadeldhura klippe are reinterpreted as repeating splays of the RMT, rather than the thick RMT package folded into an anticline-syncline pair proposed by Robinson et al. (2006). In the Dadeldhura klippe, the trace of the MCT in the southern limb is shifted ∼ 1–4 km southward from the Robinson et al. (2006) mapping, while the traces of the MCT and RMT in the northern limb are within ∼ 0.5–1 km. The northernmost RMT and MCT traces are within ∼ 0.5 km of the Robinson et al. (2006) mapping.

The average topographic profile along A-A^′ increases in elevation at a ∼ 1–2° angle, from < 1 km at the MFT to a maximum of 5–6 km above GH rocks in the north (Figs. 3b and S3 in the Supplement). The minimum topography generally follows a 0.5° angle, aside from sharp increases in the minimum elevations at ∼ 65–55 and ∼ 1 km south of the MCT. Two abrupt increases in the slope of maximum topography occur at ∼ 95–55 km south of the MCT and from ∼ 5 km south to 20 km north of the MCT. The locations of elevated topography broadly correspond with the locations of elevated ksn values. Above the Dadeldhura klippe, ksn values increase from background values at ∼ 100 km south to a maximum at ∼ 70 km south of the MCT (Fig. 3b). In the north, ksn values gradually increase from ∼ 35 km south to ∼ 10 km north of the MCT. These bands of elevated ksn values reflect regions of active rock uplift or recent uplift that has been translated to the south (e.g., Eizenhöfer et al., 2019; Harvey et al., 2015; Wobus et al., 2006b).

The balanced cross-section for the updated geologic mapping is shown in Fig. 3c (A-A^′). The MHT in A-A^′ is characterized by an average dip of ∼ 4° from beneath the MBT to the active ramp at ∼ 13 km north of the MCT, consistent with geophysical observations in western Nepal (e.g., Hoste-Colomer et al., 2018; Laporte et al., 2021; Subedi et al., 2018). We designate structural zones for the cross-section to facilitate discussion of the geometry, from south to north: the Subhimalaya, MBT, Dadeldhura klippe, LH duplex, RMT duplex, and MCT zones (Fig. 3). The Subhimalaya is characterized by three faults that deform the Siwalik Group. In the MBT zone, the upper LH units are juxtaposed above the Siwalik Group. The lower LH units (RMT), GH rocks (MCT), and Tethyan Himalayan rocks comprise the Dadeldhura klippe. The base of the Dadeldhura klippe is the RMT, which separates the lower and upper LH units. In the southern limb, the lower LH units are repeated by two imbricate faults. Below the RMT, upper LH strata are repeated in the LH duplex, with an upper LH thrust sheet forming the roof thrust. Out-of-sequence faults imbricate LH strata under the Dadeldhura klippe and offset strata in the LH duplex, as needed to reproduce the surface geology. The RMT duplex comprises four horses, with two consisting only of the Ranimata Fm. and two with both lower LH units. GH rocks comprise the MCT zone. Shortening on the MCT and RMT are the minimum amounts required to connect the northernmost GH and lower LH units with those preserved in the Dadeldhura klippe. The revised cross-section geometry is reproduced by ∼ 615 km of shortening. While differences in the cross-sections for A-A^′ and B-B^′ arise due to updates to the geologic mapping, the most significant alterations to the geometry depicted in Fig. 3 are to facilitate changes to the location and magnitude of the active décollement ramps, the order and magnitude of fault displacement and resulting structural uplift, and minimize the amount of shortening necessary to balance the cross-section (see Text S6 in the Supplement for additional details).

We present thermokinematic model results for three ramp geometries: (A) Robinson et al. (2006), (B) Olsen et al. (2019), and (C) our new cross-section geometry. Results from each model lead to the changes in geometry, kinematic sequence, and shortening rates tested with the subsequent models. Model results are presented with the variable velocity that provided the highest total model fit for that section. Results for Model C with a constant shortening rate are included in Fig. S4 in the Supplement).

The topographic profile and ksn values predicted by the landscape evolution model for Model C are shown in Figs. 5d, e and S5 in the Supplement and Video S2 in the Supplement. The average topographic profile increases in elevation from < 1 km at the MFT to ∼ 6.5 km above the GH zone at a ∼ 1° angle (Fig. 5e). The average slope and elevation sharply increase between ∼ 110–65 km south and ∼ 0–30 km north of the MCT, in response to out-of-sequence faults active since ∼ 1.3 Ma and motion over the active ramp at ∼ 13 km north of the MCT (Fig. 5e, Video S2). While the modelled relief is notably less than measured (0.5–1 km vs 2–3 km), the modelled mean elevation is within 0.5–1 km of measured topography. Modelled relief is the most similar in scale (∼ 2 km) north of the MCT. The median predicted ksn values sharply increase across ∼ 15 km wide windows above the MBT and Dadeldhura klippe zones, in response to out-of-sequence motion uplifting rocks over the ramp at ∼ 80–65 km south of the MCT, feeding slip to surface faults at the MBT, and within the RMT at ∼ 125–115 km south of the MCT (Fig. 5d, Video S2). In the north, predicted median ksn values sharply increase from ∼ 5 km south to ∼ 30 km north of the MCT in response to out-of-sequence motion on the RMT at ∼ 5 km south of the MCT. The highest predicted ksn values along the section are at ∼ 15–30 km north of the MCT, above the modern active ramp.

Bedrock cooling ages and basin accumulation rates are both complimentary indicators of exhumation in a fold-thrust belt. In-situ thermochronologic ages preserve the timing and location of uplift, cooling, and exhumation over a décollement ramp (e.g., Lock and Willett, 2008; McQuarrie and Ehlers, 2017). Uplift and erosion in the hinterland of a fold-thrust belt is intrinsically linked to the associated foreland basin. The isostatic response to each increment of displacement creates space in the foreland basin that is filled with hinterland-derived synorogenic sediments (e.g., DeCelles and Giles, 1996; Naylor and Sinclair, 2008). Thus the magnitude, location, and rate of uplift and exhumation in the hinterland has a direct influence on the magnitude and rate of accumulation in the foreland basin and on any detrital provenance and/or detrital cooling age signals present in the basin (e.g., Bernet and Garver, 2005; Braza and McQuarrie, 2022a, b; Malusà and Fitzgerald, 2020; Reiners and Brandon, 2006). Accounting for the basin accumulation is important, as changes to fault timing and shortening rates that may not alter bedrock cooling ages can impact the basin depositional ages (Braza and McQuarrie, 2022a). For example, the hinterland unroofing signal of lower LH rocks at ∼ 10 Ma is well-documented in the western Nepal foreland by a distinct shift in εNd values starting at ∼ 4.4 km basin depth (Huyghe et al., 2001; Robinson et al., 2001, 2006; Szulc et al., 2006). This signal requires LH rocks to be exposed in the hinterland at the time these sediments are being deposited at ∼ 4.4 km depth in the modern foreland; models must be able to reproduce this relationship, timing, and basin depth. As the bedrock and basin exhumation records are linked, any model of tectonic deformation must be able to reproduce both records to be considered valid.

With the one ramp geometry of Model A, the youngest measured AHe ages at ∼ 90–70 km south of the MCT are replicated with motion on the MBT at ∼ 13 Ma and shortening rates of 7 mmyr-1. The ∼ 14–19 Ma MAr ages set by RMT motion and the ∼ 9–14 Ma ZHe ages set by motion over the northernmost upper LH ramp require fast rates (∼ 45–63 mmyr-1) from the model start to 13 Ma. These thrust kinematics and the resulting bipolar (45 mmyr-1 vs 7 mmyr-1) rates produce a basin age of ∼ 15.8–13.5 Ma over the same depth that the measured data record sedimentation from ∼ 13.5–6 Ma (Fig. 4b). The predicted lower LH provenance signal is ∼ 2 km shallower than the measured depth and the depositional age is ∼ 3 Myr older than the ∼ 10 Ma measured age.

The two-ramp geometry depicted in Model B is ideal for reproducing the overlapping swath of ZHe and AHe ages measured at ∼ 90–70 km south of the MCT (Fig. 4g) that indicate significant exhumation at 12–9 Ma and minimal exhumation since. Similar to Model A, fast rates (70–75 mmyr-1) between 18–10 Ma are necessary to reproduce the ∼ 14–19 Ma MAr ages set by RMT motion and ∼ 9–14 Ma ZHe ages set by motion over the northernmost upper LH ramp, and are followed by slow rates of 7 mmyr-1. The extreme rates produce a basin age ∼ 13–11.5 Ma over the depth of the measured ∼ 16–12 Ma lower Siwalik unit and ∼ 10.4–9.8 Ma over the depth of the measured ∼ 8–6 Ma middle Siwalik unit (Fig. 4f). For Models A and B, the unrealistic rates and clear disconnect between the measured and modelled basin accumulation ages argue against these MHT geometries.

Model C, with the ramp located north of the MCT, is able to reproduce the significant exhumation required by the older measured ZHe (∼ 12–15 Ma) and AFT (∼ 10–15 Ma) ages without the extremes in shortening rate. Like the other models, the ZHe ages above the Dadeldhura klippe are set during cooling and uplift over the upper LH ramp (Figs. 6 and S6 in the Supplement, Video S1) and thus initial shortening on the MBT needs to initiate at ∼ 13–15 Ma. In marked contrast to the previous models, the ∼ 130 km distance from the ramp to the modern MBT means shortening rates from ∼ 14 Ma to present may range from 15–35 mmyr-1 (Table S3, Video S1). Motion on the upper LH thrust at ∼ 14 Ma in Model C deposits sediments beginning at ∼ 6.5 km basin depth, matching the measured depositional ages. The modelled exposure of lower LH rocks begins ∼ 10 Ma and is preserved at ∼ 4.8–4.5 km basin depth (Fig. 5). Model C is also able to fully reproduce the measured distribution of detrital MAr cooling ages (Fig. 7).

While the fit of modelled to measured cooling ages do not strongly differ in the frontal portion of the system (MBT to Dadeldhura klippe) between Models A (61 %), B (79 %), and C (76 %), the divergence in the resulting shortening rates, basin depositional ages, and basin accumulation rate is profound. The ability of a model to replicate the basin age and provenance is critical for assessing the validity of modelled geometries and exhumation pathways.

The MHT geometry is defined by the size and location of the active décollement ramps. In Model C, the active ramp is ∼ 8 km thick and located ∼ 13 km north of the MCT. AHe, AFT and ZHe ages that overlap in age between ∼ 5–15 km north of the MCT requires ∼ 10 km of active uplift and exhumation in this area. ZHe ages increase to the north from 2 to 5 Ma over 5–20 km north of the MCT, consistent with the expected pattern of increasing ages on the hinterland side of an active ramp (Figs. 4 and 5) (McQuarrie and Ehlers, 2017). Similarly, the gradual increase in age of all three systems (AHe, AFT, and ZHe) from 1–2 Ma in the north to 5–7 Ma in the south (40–50 km south of the MCT) is the expected age signal in the direction of transport following uplift and exhumation over a ramp (Figs. 4 and 5, Video S1). These across strike patterns in cooling ages support the active MHT ramp location at ∼ 10–20 km north of the MCT (Figs. 3 and 5).

Previous estimates of MHT geometry in western Nepal generally propose active décollement ramps at ∼ 120–40 km south of the MCT (Fig. 8) (e.g., DeCelles et al., 2020; Harvey and Burbank, 2024; Hoste-Colomer et al., 2018; Laporte et al., 2021; Olsen et al., 2019; Robinson et al., 2006). The size and location of these ramps are similar to those tested in Models A and B, allowing us to infer the influence of these proposed ramps on the distribution of young cooling ages and locations of active uplift. The youngest predicted cooling ages will always be co-located with active ramps (Lock and Willett, 2008; McQuarrie and Ehlers, 2015, 2017; Ghoshal et al., 2020, 2023); this is seen in all three modelled sections (Figs. 4 and 5). In western Nepal, ∼ 5–10 Ma AHe cooling ages and resulting slow exhumation argue against a long-lived MHT ramp between 60–90 km south of the MCT (e.g. Harvey and Burbank, 2024). As shown in Models A and B, adjacent ramps to the north or south of this region (120–100 km south and 60–30 km south of the MCT) (Fig. 8) necessitate markedly slow fault shortening rates (∼ 7 mmyr-1) from at least 10 Ma, and potentially from 14 Ma, to present to replicate the measured cooling ages.

Without a décollement ramp in the MHT at ∼ 100–60 km south of the MCT, cross-section geometries lack a mechanism to drive uplift to produce the region of raised topography and elevated ksn values in the Dadeldhura klippe (Figs. 3b and 7) (Harvey et al., 2015; Harvey and Burbank, 2024). As steepened river profiles are a transient response to uplift, elevated ksn values indicate recent uplift. The ksn values in the Dadeldhura klippe are elevated across a narrow window (Fig. 3), indicating limited amounts of lateral translation of the elevated values (e.g. Eizenhöfer et al., 2019) and arguing against a long-lived MHT ramp in this region. Similar to Harvey and Burbank (2024), we argue that low exhumation magnitudes and low relief surfaces suggest very recent and minor uplift across this region. To accommodate this, we incorporate 3.75 km of displacement on a fault splay that cuts through the upper LH units at ∼ 80–60 km south of the MCT and feeds slips to the RMT splays in the southern limb of the Dadeldhura klippe (∼ 105–100 km south of the MCT) (Fig. 8), co-aligned with a region of active microseismicity (Text S6) (e.g. Laporte et al., 2021).

An alternative explanation for the region of raised topography and elevated ksn values in the Dadeldhura klippe is river incision of a high elevation, low relief surface (Sherpa et al., 2023). In this framework, the high elevations preserved in the Dadeldhura klippe formed during emplacement of the RMT from 14–11 Ma (Sherpa et al., 2023). River incision into this surface by the Karnali and Tila rivers (Fig. S3) from ∼ 11 to 6 Ma is argued to drive exhumation and cooling of rocks to set the broadly northward younging pattern of AFT ages between 110 to 50 km south of the MCT (Fig. 5) and elevated ksn values at the edges of the low-relief surface (Sherpa et al., 2023). We argue that the rapid cooling illustrated by their inverse modelling from > 200 °C and initiating by ∼ 14–8 Ma (depending on location) is the expected response of exhumation over a ramp. This ∼ 8 km of northward younging exhumation and resulting cooling in our model is a function of motion on an upper LH thrust from ∼ 14–7 Ma (Fig. 6c–f, Video S1). The associated cooling allows us to produce modelled ZHe, AFT and AHe ages at ∼ 13–11, 10–8, and 7 Ma respectively, replicating 76 % of the cooling ages in this region (Fig. 5), and is reflected in the northward younging pulse of ∼ 6 mmyr-1 of exhumation highlighted in Fig. 9 from 14–7 Ma. In addition, both our flexural kinematic model and landscape evolution model highlight the difficulty in maintaining ∼ 4 km of elevation across a thrust sheet in absence of other forms of uplift. For example, the flexurally modelled topography that increases northward from 2.3–5.1 km in elevation (Fig. 6f, 6.6 Ma) between 50 km south and 20 km north of the MCT was initially uplifted and exhumed over the ramp at 50 km south. Following this exhumation, topography is simply translated southward and gradually subsides due to thrust loading at 90 km south and at the ramp at 50 km north of the MCT. The landscape evolution models for Model C also illustrate this and show how high topography is generated from vertical motion at the active fault ramp and then is translated southward. This is most apparent between 10.3–7.8 Ma and from 4.6 Ma to present (Video S2). These translated interfluves gradually decrease in elevation to the south due to both flexural subsidence and erosion and provide support for an uplift mechanism to produce the high topography and elevated ksn values in the region of the Dadeldhura klippe (Fig. 5).

The ability of the modelled geometry, kinematics, and rates to replicate, to a first order, the topography of western Nepal provides an additional level of validity to the MHT geometry presented here. While ∼ 1.4 km of uplift accompanied by ∼ 0.6 km of erosion from 0.6 Ma to present is required to replicate the uplifted mean topography of the Dadeldhura klippe (Fig. 5, Video S2), intermittent fault motion on this structure over the last million years may facilitate the needed river incision to propagate through the uplifted topography (e.g. the Karnali and Tila rivers) during fault inactivity. This river incision is required to replicate the topographic relief missing from the modelled landscape.

The ability of Model C to sufficiently reproduce the various bedrock and basin datasets indicates that the geometry, kinematics, and shortening rates represent a viable exhumation pathway for the rocks presently at the surface and we can use these results to evaluate the dominant structural drivers of exhumation in western Nepal. The location and magnitude of erosion determined from the difference in the deformed, uplifted topography and the newly eroded topography predicted in each ∼ 10 km displacement step in the flexural-kinematic model combined with the increment of time this erosion occurred over provide an exhumation rate for each model step (Fig. 9).

Fold-thrust belts have a characteristic ramp-flat décollement geometry that results in periods of active exhumation during vertical motion above the ramp and associated focusing of erosion of the raised topography that alternate with periods of slower, very limited exhumation in the absence of uplift where rocks are laterally translated along décollement flats. Cooling during lateral translation occurs as rocks pass through tilted isotherms (Fig. 6) (e.g., Braza et al., 2023; Coutand et al., 2014; Lock and Willett, 2008; McQuarrie and Ehlers, 2015, 2017; Robert et al., 2011).

RMT motion over the ramp at the northern edge of upper LH units (∼ 216 km north of the modern MCT) sets the MAr cooling ages for rocks preserved in the Dadeldhura klippe (Fig. 6c, Video S1) and displacement is thus required to initiate by ∼ 18 Ma to set the MAr age in the southern limb of the klippe. Fast shortening rates (45–48 mmyr-1) are required to predict MAr ages ∼ 16 Ma in northern limb of the klippe and to allow for the full shortening in the RMT duplex to occur prior to ∼ 14 Ma. The signal of motion over the ramp is visible in the bedrock exhumation history as a pulse of fast (4 to > 9 mmyr-1) exhumation at ∼ 16.5–14.6 Ma for rocks presently exposed at ∼ 115–95 km south of the MCT, while the continued southward translation shifts the location of rapid exhumation northward with time, with the rocks at ∼ 80–60 km south of the MCT experiencing the faster rates of ∼ 4–9 mmyr-1 at ∼ 15.7–14.2 (Fig. 9). This significant pulse of exhumation is also present in the predicted and measured detrital MAr ages (Fig. 7), with a ∼ 15 Ma static peak that is persistent throughout the basin. The population of ∼ 14–18 Ma bedrock MAr ages in the Dadeldhura klippe provides a clear source for the ∼ 15 Ma age peak in the detrital distribution. The fine resolution of the predicted detrital Mar record also highlights that the persistence of this ∼ 15 Ma static peak through the entire basin record is from both erosion of the bedrock and recycling of the earlier foreland deposits (Braza and McQuarrie, 2022b).

The ZHe ages in the Dadeldhura klippe were set during formation of the LH duplex and motion over an upper LH ramp (∼ 86 km north of the modern MCT) (Fig. 6e). As shown by the thermokinematic models, motion over the ramp that sets these ages needs be as old as possible (∼ 14 Ma) to reproduce the ZHe ages in the southern limb of the Dadeldhura klippe. Early exhumation of these rocks over the ramp in the upper LH rocks is facilitated by adjusting the shortening rates in Models A and B, but this provides a poor fit the measured basin depositional ages (Fig. 4). To balance the bedrock requirement for older exhumation with the need for deposition of the basal portion of the lower Siwalik unit to be ∼ 13–17 Ma, the new geometry in Model C presents a modified LH duplex, where the rocks in the immediate MBT hanging wall extend to the north becoming the roof thrust of the duplex (Figs. 3 and 5, Video S1). This geometry facilitates earlier exhumation over a ramp that is further north than the modern active ramp and allows for more reasonable shortening rates. This signal is apparent in the pulses of rapid exhumation (∼ 4–9 mmyr-1) that initiates at ∼ 13.6 Ma at ∼ 115–95 km south of the MCT and gradually shifts northward with time to ∼ 7.7 Ma at 20 km south of the MCT. While shortening rates slow during LHD motion from 35 mmyr-1 at ∼ 13.3 Ma to 18 mmyr-1 at ∼ 12 Ma to 16 mmyr-1 at ∼ 8.6 Ma, exhumation rates remain similar to those during RMT motion (∼ 4–9 mmyr-1) due to the large size (∼ 8 km) of the upper LH ramp.

Uplift over ramps that extend through the full 8 km of upper LH stratigraphy drives significant pulses of exhumation, notably during RMT motion, motion on the first upper LH fault, duplex formation and motion over the modern active ramp at ∼ 13 km north of the MCT (Fig. 6, Video S1). While the thermokinematic model results demonstrate that the modern active ramp is required to be co-located with the youngest cooling ages along the section (Figs. 4 and 5), the locations of the upper LH ramps (∼ 216 and ∼ 86 km north of the modern MCT) during RMT and LHD motion are less unique. The size of these ramps is constrained by the stratigraphic thickness of units comprising the ramp and the location constrained by the restored length of the upper LH units in the LH duplex in the geometry for Model C. It is the timing and magnitude of exhumation that is important for motion over upper LH ramps while the RMT and LHD are active, not the specific location of the ramp. However, the modern ramp location is unique and is required to be co-located with the youngest cooling ages along the section at ∼ 13 km north of the MCT.

While motion over the ∼ 8 km thick modern active ramp produces ∼ 10 km of exhumation and ∼ 150–200 °C of cooling (Figs. 6 and S6), this is still insufficient to expose rocks at the surface with a predicted MAr age younger than the age of RMT motion (Video S1). Thus, an additional mechanism is required to drive young (< 6 Ma) uplift and exhumation to replicate ∼ 6 Ma MAr ages above and to the north of the RMT duplex (Fig. 3). In Model A, 11 km of out-of-sequence displacement on faults deforming the LH duplex (∼ 15–33 km south of the MCT) has a significant effect on the predicted ZHe ages but does not produce young reset MAr ages (Fig. 4c); this provides the minimum amount of young slip required on a RMT fault. The 6 Ma MAr ages are reproduced with 12 km of out-of-sequence motion on a RMT fault at ∼ 20 km south of the MCT and 16 km on a RMT fault at ∼ 3 km south of the MCT between 6–3 Ma (Fig. 5f, Video S1). Motion on these faults cools rocks through the ∼ 350 °C MAr closure temperature (Figs. 6g and S6) and predicts a continuous swath of 5–6 Ma MAr ages from ∼ 20 km south to ∼ 10 km north of the MCT, where the measured data are located. This out-of-sequence motion is also reflected in the predicted detrital MAr distribution by a younging of the minimum age, near the middle-upper Siwalik contact (Fig. 7). Although not present in the youngest measured detrital peak ages, this age population may be reflected in the youngest measured ages (∼ 10 Ma) at 1.2 km depth (Fig. 7). Both the age (< 6 Ma) and location (∼ 3 km south of the MCT) of faulting is similar to that proposed by Braden et al. (2018) to account for 7 Ma U-Pb monazite and 6 Ma MAr ages, while the location and displacement magnitude is similar to that proposed for out-of-sequence motion on the faults adjacent to the MCT in central Nepal (e.g. Ghoshal et al., 2023; Wobus et al., 2006a). Out-of-sequence motion on the northern RMT faults in far western Nepal have a similar timing to that in the Simikot section but a lower magnitude of displacement, only influencing the ZHe and AFT ages in the RMT and MCT hanging walls (Braza et al., 2023).

Although there has been continuous exhumation of the high Himalaya in western Nepal since 7–8 Ma (e.g. Harvey and Burbank, 2024), Model C demonstrates this is partitioned into fast rates during periods of out-of-sequence faulting and growth of the LH duplex that alternate with periods of slower, passive exhumation in the absence of active fault-driven uplift (Fig. 9). While exhumation rates may exceed 9 mmyr-1, these fast rates are only sustained for very short ∼ 0.4–2 Myr time windows. Averaging the fast and slow rates over the past ∼ 8 Ma supports long-term rates ∼ 3 mmyr-1, similar to a proposed ∼ 2 mmyr-1 exhumation rate derived from inverting measured AHe, ZHe, and MAr cooling ages in the high Himalayas (∼ 5–10 km north of MCT, Figs. 5 and 9) (Harvey and Burbank, 2024). Exhumation rates determined from detrital MAr age lag times in the Siwalik unit and a 1D thermal model supports ∼ 9 km of erosion since ∼ 16 Ma and rates of 2.6 mmyr-1 until 10 Ma (Szulc et al., 2006), also similar to the long-term average rate in Model C.

While exhumation rates are most strongly related to the size of the active ramp driving uplift rather than shortening rates, a marked change in shortening rate such as from fast rates of shortening (45–48 mmyr-1) during RMT motion from 18.1–13.6 Ma to slower shortening rates (15–18 mmyr-1) from 12.6–0.6 Ma requires a change in how India-Asia convergence is partitioned along faults and throughout the Himalayan-Tibet orogen (e.g., Murphy et al., 2014; Taylor et al., 2003; Yuan et al., 2013). In Model C, early fast rates that accommodate almost all of India-Asia convergence may be the result of hot (> 375 °C) and thus weak, low angle and high displacement faults (Braza et al., 2023). A marked slow-down in shortening velocity occurs at ∼ 13 Ma, very similar to the best-fit model for the Api section in far western Nepal (∼ 100 km to the west) which slows from ∼ 25–40 to ∼ 10–15 mmyr-1 at ∼ 12 Ma (Braza et al., 2023). The slowing of shortening to the west (Braza et al., 2023) was attributed to a combination of slowing convergence rates between India and Asia (Molnar and Stock, 2009) and a partitioning of convergence between the Himalayan faults and with faults to the north in Tibet or possibly the Western Nepal Fault System (e.g., Murphy et al., 2014; Silver et al., 2015; Sundell et al., 2013; Yuan et al., 2013).