Complex erosional response to uplift and rock strength contrasts in transient river systems crossing an active normal fault revealed by 10Be and 26Al cosmogenic nuclide analyses

Understanding the influence of bedrock lithology on the catchment‐averaged erosion rates of normal fault‐bounded catchments and the effect that different bedrock erodibilties have on the evolution of transient fluvial geomorphology remain major challenges. To investigate this problem, we collected 18 samples for 10Be and 26Al cosmogenic nuclide analysis to determine catchment‐averaged erosion rates along the well‐constrained Gediz Fault system in western Türkiye, which is experiencing fault‐driven river incision owing to a linkage event ~0.8 Ma and has weak rocks overlying strong rocks in the footwall. Combined with existing cosmogenic data, we show that the background rate of erosion of the pre‐incision landscape can be constrained as <92 mMyr−1, and erosion rates within the transient reach vary from 16 to 1330 mMyr−1. Erosion rates weakly scale with unit stream power, steepness index and slip rate on the bounding fault, although erosion rates are an order of magnitude lower than slip rates. However, there are no clear relationships between erosion rate and relief or catchment slope. Bedrock strength is assessed using Schmidt hammer rebound and Selby Rock Mass Strength Assessments; despite a 30‐fold difference in erodibility, there is no difference in the erosion rate between strong and weak rocks. We argue that, for the Gediz Graben, the strong lithological contrast affects the ability of the river to erode the bed, resulting in a complex erosional response to uplift along the graben boundary fault. Weak covariant trends between erosion rates and various topographic factors potentially result from incomplete sediment mixing or pre‐existing topographic inheritance. These findings indicate that the erosional response to uplift along an active normal fault is a complex response to multiple drivers that vary spatially and temporally.


| INTRODUCTION
The role of climate, tectonics and lithology on the evolution and form of bedrock (detachment-limited) streams is well known.The effect of tectonics, in particular the effect of variable uplift rates (i.e.Kirby & Whipple, 2012;Whittaker, 2012;Whittaker & Boulton, 2012;Wobus et al., 2006), and climate gradients (Adams et al., 2020;D'Arcy & Whittaker, 2014) on the rates and patterns of incision have been widely reported.Until recently, the role of lithology and rock strength have attracted less attention, and many studies have sought to remove or minimise this variable by choosing study areas with little rock variation (e.g.Miller, Baldwin, & Fitzgerald, 2012;Ortega, Wohl, & Livers, 2013;Regalla et al., 2013;Snyder et al., 2000).However, landscape evolution modelling (Darling et al., 2020;Forte & Whipple, 2018;Forte, Yanites, & Whipple, 2016;Mitchell & Yanites, 2021;Perne et al., 2017) and field investigations at the landscape (Bernard et al., 2019;Zondervan, Stokes, et al., 2020) and catchment scale (Duvall, 2004;Gailleton et al., 2021;Kent et al., 2021;Peifer et al., 2020;Sklar & Dietrich, 2001;Whittaker et al., 2007) have increasingly investigated the importance of lithology on river incision and fluvial geomorphology.Yet, there are still uncertainties in how bedrock properties influence catchment scale erosion and how such characteristics can be effectively measured in the field.Furthermore, while a number of studies have directly compared catchment-averaged erosion rates (CAERs) to bedrock channel properties (i.e.Abbühl et al., 2011;Bellin, Vanacker, & Kubik, 2014;Cyr et al., 2010;DiBiase et al., 2010;Harkins et al., 2007;Kober et al., 2015;Miller et al., 2013;Ouimet, Whipple, & Granger, 2009;Safran et al., 2005), relatively few studies have determined CAER along the strike of an active fault.For example, Densmore et al. (2009) studied two faults in the western USA, the 18 km long Sweetwater fault and the 130 km Wassuk fault.Along neither fault were CAER found to be proportional to uplift rates along the fault nor to various topographic measures of the footwall geomorphology.Densmore et al. (2009) attributed the uncoupling of erosion from fault displacement to the influence of inherited high relief topography and the widespread occurrence of mass wasting.In contrast, Rossi et al. (2017) reported 26 erosion rates along a normal fault system in Baja California demonstrating a positive trend between CAER with slope and channel steepness.Roda-Boluda et al. (2019) also showed a linear relationship between CAER and the footwall component of fault throw rate from 15 samples taken from a series of catchments crossing an active normal fault system in southern Italy.In all these studies, the footwalls of the studied faults are composed of metamorphic or igneous rocks with limited reported lithological variability at a regional scale.
This lithological homogeneity of existing research areas is significant, as the modelling of Forte, Yanites and Whipple (2016) suggests that the presence of lithological contacts, where rock strength changes from strong to weak, will profoundly influence the response rates of an incising river system.For example, their modelling suggests that when soft rocks overlie hard rocks (along a contact dipping at 20-35 downstream) the lithological contact becomes an important and persistent topographic feature in the landscape.Interestingly, although the geological boundary moves downstream over time, the model suggests that erosion rates above and below the boundary should diverge.The soft rocks downstream erode at the imposed uplift rate, but the underlying hard rocks erode at a rate lower than the regional uplift rate (Forte, Yanites, & Whipple, 2016).The difference in the strength and bedrock erodibility between the hard and soft rocks controls the magnitude of difference between erosion and uplift rate, and also the duration of the landscape adjustment.
Although the more complex interbedded hard-soft rock scenarios of Darling et al. (2020)'s model indicate that in such cases the harder rocks may erode quicker than the soft rocks.A further implication of Forte et al.'s (Forte, Yanites, & Whipple, 2016) landscape evolution model is that CAER, determined from cosmogenic radionuclides (CRN, commonly 10 Be), may be affected by the relative enrichment of material from the harder rocks in the detrital sediment.Consequently, CAER would be perturbed or amplified because of the lithological variation.
Therefore, there is a knowledge gap in our understanding of how erosion rates change along faults with lithologically variable footwall geology.There is also the requirement to empirically test the results of models such as Forte, Yanites and Whipple (2016), Perne et al. (2017) and Darling et al. (2020) in regions with complex geology to assess the applicability of these models to real systems.
Here, we use the well-constrained Gediz fault system (western Türkiye) as a natural laboratory to study the landscape response to fluvial incision across a strong lithological contrast (soft rocks over hard rocks) in the footwall of an active normal fault.As the geologic and geomorphic evolution of the region is well understood and constrained (i.e.Bozkurt, 2003;Bozkurt & Sözbilir, 2004;Çiftçi & Bozkurt, 2009;Kent, 2015;Kent et al., 2017, Kent et al., 2021;Öner & Dilek, 2011;Seyito glu & Scott, 1996;Seyito glu et al., 2002), we can use the area to test the model predictions of Forte, Yanites and Whipple (2016) and investigate the role that strength contrasts play in the evolution of transient landscape responses to base-level fall.This is achieved through a suite of new 10 Be and 26 Al CRN samples to determine CAER along the strike of the boundary faults combined with published cosmogenic data (Buscher et al., 2013;Heineke et al., 2019) and geomorphic indices (Kent et al., 2021).CAERs are quantified using 10 Be and 26 Al so that the potential effect of sediment storage can be excluded, thus allowing accurate exposure and denudation histories to be calculated (c.f.Bierman et al., 1999;Granger & Muzikar, 2001;von Blanckenburg, 2006).

| STUDY AREA
The Gediz (also known as the Alaşehir) Graben is located in western Anatolia (Figure 1) forming an arcuate, asymmetric graben $150 km in length.The Bozda g Range to the south is uplifted along the southern graben-bounding normal fault and rises to over 2000 m in elevation.The approximate N-S extension forming this horst and graben structure has been ongoing since early Miocene times, probably as the result of roll-back along the Hellenic subduction zone (Okay & Satır, 2000;Ten Veen et al., 2009), and can be divided into two main phases (Bozkurt & Sözbilir, 2004).Initial extension caused uplift along the now-inactive low-angle north-dipping Gediz detachment fault (Gessner et al., 2001;Seyito glu et al., 2002;Ring et al., 2003).The Gediz detachment fault presently dips to the N-NE at up to 32 and is gently corrugated along its strike (Bozkurt & Sozbilir, 2004).The detachment forms the boundary between the Menderes Massif metamorphic rocks and overlying syn-tectonic sedimentary rocks (Figure 2).In the footwall, the Menderes Massif metamorphic core complex is composed mainly of Palaeozoic greenschist to amphibolite-facies schists, augengneisses and paragneisses (Gessner et al., 2001;Ring et al., 2003).
Following the cessation of slip on the Gediz detachment fault at $2 Ma (Buscher et al., 2013), strain stepped northwards (basinwards) onto high angle faults.These include the presently active normal fault forming the range front fault (Figures 1 and 2) to the present-day F I G U R E 1 (a) Regional location map showing the location of the Gediz Graben in Western Anatolia; (b) Geological map of the study area.Geological units are simplified from Kent et al. (2021) with additional mapping of Holocene lake deposits from Süzen, Toprak and Rojay (2006).Numbers in bold indicate rivers sampled for cosmogenic radionuclides (CRN) either in this study (Table 1) or by Heineke et al. (2019) or Buscher et al. (2013) (Table 3) topographic graben (Çiftçi & Bozkurt, 2009).In the uplifted footwall of the active fault are friable sedimentary rocks deposited originally on the hanging wall of the Gediz detachment.These sedimentary units, comprised mainly of early Miocene to Pliocene-aged alluvial fan and fluvial sandstones and conglomerates, unconformably overlie and derive from the metamorphic basement (e.g.Çiftçi & Bozkurt, 2009;Purvis & Robertson, 2004, 2005).
Quaternary sediments are variable in extent across the Bozda g range (Figure 1).Fragments of river terraces have been reported by Kent (2015) along three rivers-the Yeniköy, Kavaklıdere and the Kabazlı (Figure 1b).These river terraces are of small spatial extent with OSL dates of five samples (Figure 1d) from the fine-grained facies of only one, well-developed, terrace level indicating aggradation between $84-7.5 ka (Kent, 2015).However, in the headwaters of several of the larger river systems, fluvial and lacustrine fine-grained sediments up to 170 m thick can be found (Süzen, Toprak, & Rojay, 2006).Sediment cores from Gölcük Lake (Figure 1d) yielded 14 C dates of ≤10 ka (Sullivan, 1988), suggesting deposition during the Holocene to Pleistocene, but ages of the older sediments are not constrained.These deposits are thought to have formed owing to 1-2 of rotation on the graben boundary fault during the Holocene resulting in slope reduction, lake formation and sediment deposition (Süzen, Toprak, & Rojay, 2006).
Across the Bozda g Range, transverse bedrock rivers flow northwards into the Gediz Graben across the southern boundary fault.The rivers are generally deeply incised with prominent knickpoints and gorges upstream of the active fault.The slope-break knickpoints are not coincident with lithological boundaries (Kent et al., 2017) and are interpreted to mark the upstream extent of transient wave of river incision.Incision was caused by an increase in slip on the graben bounding fault as a result of the fault linkage of three initial fault segments $0.6-1 Ma (Kent, 2015;Kent et al., 2017).As a result of this linkage, present day throw rates (the vertical component of the slip rate) are now thought to be higher than the long-term average, with rates of up to 2 ± 0.2 mmyr À1 calculated for the centre of the fault array (Kent et al., 2017).Kent et al. (2021) selected six of the transverse rivers to investigate the lithological controls on transient river behaviour.For simplicity, Kent et al. (2021) used two broad groupings of rock types: metamorphic and sedimentary in their quantitative analyses.Rivers were chosen to investigate differences in the proportion of metamorphic to sedimentary bedrock reaches (100% metamorphic in the Akçapınar River through to $50% along the Yeniköy River; Figure 1b) and differences in uplift rate as a result of activity along the graben boundary fault.Here, we continue to use these two broad lithologic groups to allow comparisons to this previous work.

| Sample collection and CRN
Eighteen samples of river sand from the active riverbed or sediment bars (Figures 1c and 3) were collected from nine catchments draining northwards across the Gediz Graben boundary fault in May 2018 (Figures 1c and S3).The rivers were selected because either they had previously been sampled by Buscher et al. (2013) or were one of the six rivers studied in detail by Kent et al. (2021).Overall, a nested sampling strategy was adopted so that 10 samples were collected from the range front where the rivers cross the active normal fault.On the easternmost river, two samples were collected $2 km apart to assess downstream mixing and reproducibility.The remaining eight samples collected further upstream at either the lithological boundary between the sedimentary and metamorphic rocks or upstream of the knickpoint.Five of these eight samples were collected upstream of identified slope-break (tectonic) knickpoints identified by Kent et al. (2017), and the final three samples were collected at the low-angle detachment that forms the lithological boundary enabling comparison to published datasets.A further CRN dataset was published by Heineke et al. (2019) bringing the total number of samples analysed in the Gediz region to 33.
The 18 samples collected here were sieved to 2 mm in field and further sieved to the 250-500 μm size fraction in the lab.Standard magnetic separation to concentrate the quartz fraction of the sample using a Franz magnetic separator was undertaken at the University of Plymouth.Subsequently, samples were chemically leached using diluted HF, and between 16 and 20 g of clean quartz cores were dissolved at SUERC together with $0.29 g of the CIAF-PH9 in-house 9 Be carrier solution ([Be] = 849 ± 12 ppm) following the procedure of Child et al. (2000). 10Be and 26 Al concentrations were measured by the 5-MV NEC Pelletron accelerator mass spectrometer (AMS) at SUERC (Xu et al., 2010).
The results were input into the online CRONUS-Earth calculator v 3.0 (Balco et al., 2008) using the LSDn scaling, a sample density of 2.65 gcm 3 and NIST_27900 and Z92-0222 standardisations for 10 Be and 26 Al, respectively.Mean catchment elevation and shielding were derived from the ALOS World3D 30 m DEM, which has been shown to extract more accurate hydrological networks than other F I G U R E 2 Simplified cross-section of the northern margin of the Bozda g Horst showing the relationship between low and high-angle faults (adapted from Kent et al., 2016).[Color figure can be viewed at wileyonlinelibrary.com] comparable global DEMs (Boulton & Stokes, 2018) using ArcGIS Pro 2.6.2 and TopoToolBox functions (Schwanghart & Scherler, 2014).
Similarly, catchment mean slope and relief over a 150 m radius were extracted using standard GIS tools.
Burial ages were derived from 10 Be and 26 Al data following the same principles as Granger and Muzikar (2001).This method allows solving of both the erosion rate corresponding to the initial 10 Be and 26 Al concentrations, and the average burial time after the exhumation of the quartz grains.To make them consistent with CRONUS v.3 results, scaled concentrations, spallation and muon production rates, and attenuation lengths were calculated as in Rodés (2021).
We also recalculated the 10 Be sample concentrations reported in Buscher et al. (2013) and Heineke et al. (2019) for our study area using the same parameters stated above (e.g. using topographic shielding and a sample density of 2.65 gcm 3 and CRONUS v 3 (Table S1).Note that Heineke et al. (2019) did not apply a topographic shielding and used a sample density of 2.2-2.5 gcm 3 in addition to using v 2.3 of the CRONUS-Earth calculator, which results in differences in the erosion rates stated here compared to those reported in the original papers.Neither of these previous studies included 26 Al concentrations, so corrections for sediment reworking or burial cannot be determined for these previously published CRN data .

| Sediment (un)mixing
In the Bozda g catchments, studied samples were taken at the catchment outlet, at the major lithological boundary and in five locations above the slope-break knickpoint.This sampling strategy allows the erosion rates above (un-incised) and below (incised) the slope-break knickpoint to be deconvolved assuming that the same amount of quartz-bearing sediment is produced in both parts of the watershed.
The sediment mixing is determined using the approach of Granger, Kirchner and Finkel (1996), as the CRN records the average erosion rate for the entire contributing catchment area.Therefore, the erosion rate between two sample points (a 'subcatchment') can be determined by correcting for the upstream sediment flux according to where E (mMyr À1 ) is the erosion rate of a catchment with area A (m 2 ), with subscripts indicting different subcatchments (Figure 4), where c is the entire catchment and a and b are the upstream and downstream subcatchments, respectively.In this study, a single common value for the upstream erosion rate E A is used for all catchments owing to: (a) the limited data on the CAER above the knickpoint, (b) the assumption that this area represents a low relief and low erosion rate landscape formed prior to the uplift causing the present transient river incision.
ArcGIS Pro 2.6.2 was used to calculate the areas used in the unmixing calculations.The knickpoint finder tool in TopoToolBox (Schwanghart & Scherler, 2014;Stolle et al., 2019) was used to identify the highest knickpoint along all tributaries in the study area using a tolerance of 30.These were then used as pour points for the watershed tool, the results of which were then summed to determine the total unincised area in each river catchment, which is then subtracted from the total catchment area calculated in the same way for the sample locations.Whittaker et al., 2007;Attal et al., 2011;Zondervan, Whittaker, et al., 2020): where the unit stream power, ω represents energy dissipation per unit channel area on the bed with units of Wm À2 , ρ is the density of water, g is the acceleration due to gravity, Q is the water discharge (m 3 s À1 ), S is local channel slope (m/m) and W the channel width (m) as measured in the field.Consequently, specific bedrock erodibility, k b , has units of ms 2 kg À1 , representing the inverse of stress (c.f.Yanites et al., 2017).Kent et al. (2021) demonstrate that the metamorphic rocks are around twice as hard as the sedimentary rocks.This difference is reflected by the bedrock erodibility, which was calculated as 2.2-6.3Â 10 À14 ms 2 kg À1 in the metamorphic rocks.In contrast, bedrock erodibility values in the sedimentary units were 5 to 30 times larger (i.e. 5 to 30 times weaker) at 1.2 Â 10 À13 to 1.5 Â 10 À12 ms 2 kg À1 (Kent et al., 2021).Significantly, stream power was shown to scale with fault throw rate in the metamorphic rocks but not in the sedimentary units; potentially because the weaker sedimentary rocks themselves directly influence the fluvial processes and long-term erosional dynamics.
However, values for unit stream power (Equation 2) for each river with reported CRN concentrations are required.Using the regional Q to A relationship determined using field measurements for the six rivers detailed in Kent et al. (2021), the estimate of Q for each river is found by extracting cumulative catchment area downstream along each sampled river at 100 m intervals using ArcGIS Pro 2.6.

| Rock strength and erodibility measurements
In situ rock strength measurements can be used to estimate bedrock erodibility, which is related to the inverse of the lithologies tensile strength (Sklar & Dietrich, 2001).However, tensile strength measurements are difficult to measure in the field and, as a result, the Schmidt hammer is commonly utilised owing to the ease of use and portability (e.g.Goudie, 2016).Kent et al. (2021) used an N-type Schmidt hammer to characterise average bedrock uniaxial compressive strength for each lithological unit.Additionally, information on fracture characteristics was collected to calculate the semi-quantitative SRMS (Selby, 1980).
Twenty Schmidt hammer readings were taken at 130 locations along the six study rivers, the majority of which are from the metamorphic basement.At only eight locations could the Schmidt hammer reliably return a rebound value for the sedimentary rocks.At another 28 sites, the exposed bedrock was too weak to accurately characterise the strength using this method and was recorded as having a rebound strength of <20 (the effective limit of the Schmidt hammer), allowing the SRMS to be determined even where bedrock is very weak.Schmidt hammer rebound and SRMS are then averaged for the $2 km upstream of the CRN sample locations where possible.The 10 Be concentrations measured in the new samples range from 1.3-10.0Â 10 4 atoms g À1 , while there were between 1.6-96.4Â 10 4 atoms g À1 of 26 Al (Table 1).These values compare well to previously reported CRN concentrations of 10 Be in the range 1.5-13.7 Â 10 4 atoms g À1 (Table S1) from sediment in rivers mainly draining the metamorphic basement (Buscher et al., 2013;Heineke et al., 2019).
However, the denudation rates estimated from both nuclides agree within error for <30% of the samples.These samples show 26 Al/ 10 Be ratios in the range 6.2-7.8.The samples with a larger deviation between the derived denudation rates of each nuclide have significantly depleted 26 Al/ 10 Be ratios of <5.2 (Table 2).In a two-isotope diagram (Figure 5), 44% of data points cluster in the 0-0.5 Ma burial zone, 17% in the 0.5-1 Ma burial zone and 39% of points in the >1 Ma burial zone.These data indicate that a simple exposure/ denudation history, without taking into account sediment storage, is incorrect for the majority of samples and implies that sediment reworking from the alluvial plain and/or the uplifted sediments is contributing a significant component of the transported bedload in many rivers (c.f.Granger, Kirchner, & Finkel, 1996).
Therefore, the 26 Al data allows the calculation of an average burial history and the determination of a new erosion rate taking into account the depletion of the 10 Be and 26 Al concentration during the time that the quartz grains were buried (Table 2).This calculation gives 'burial-corrected' erosion rates of 32 to 248 mMyr À1 for the study area catchments.Unfortunately, a similar calculation cannot be undertaken on the existing published CRN datasets (Buscher et al., 2013;Heineke et al., 2019) as there are no reported 26 Al data.
As these sites are predominantly located in the footwall of the detachment fault, where there is little or no outcrop of sediments, it suggests that sediment storage should be limited for these samples.However, the presence of Holocene or older sediments in some catchments is a source of potential error that cannot be accounted for in the previously published data and may explain why the published erosion rates are, in general, slightly higher than those reported here.This hypothesis is supported by the 26 Al/ 10 Be ratios of three of samples upstream of the boundary between the sedimentary rocks and the Menderes Massif metamorphics falling in the >1 Ma burial zone (Figure 5).et al., 2018).Therefore, the average CAER used is 46 mMyr À1 above the knickpoints.

| Results from unmixing model
In a landscape experiencing transient river incision, erosion rates above the knickpoint are expected to be lower than below the knickpoint.Therefore, we used an unmixing method (e.g.Granger, Kirchner, & Finkel, 1996;Rosenkranz et al., 2018) to remove the influence of such low erosion rates on downstream samples.Using the minimum erosion rate estimate determined above (i.e.46 mMyr À1 ), it is possible to derive a quantitative estimate for the erosion rates within the transient reach; that is, upstream of the active fault and downstream of the knickpoint.This method is applied to both the new burial-corrected CAER and also the previously published CRN datasets (Table 3).The effect of applying this unmixing model is variable depending on the proportion of the total catchment area falling in the low erosion rate zone above the knickpoint, and on the difference between the low erosion rate and the denudation rate determined for the downstream sample (Table 3).For example, where the downstream initial burial-corrected CAER are relatively low (such as on the Bozda g), the unmixing results in a small increase in CAER (e.g. from 63 to 99 mMyr À1 ).But where the difference between the assumed upstream erosion rate of 46 mMyr À1 and the downstream sample is greater, the final calculated rate is markedly higher.For example, on the Gumuşçay, the initial burial-corrected CAER is 144 mMyr À1 , which increases to 1330 mMyr À1 after unmixing; a tenfold increase.For the majority of samples, the rates do increase, but a limited number of samples from or close to the lithological boundary result in no or negligible change.This is because the measured rate is close to the low erosion rate value even though the samples are within the knickzone.For one sample, 14 T1 (Heineke et al., 2019), this adjustment results in a negative erosion rate.This CAER is not included in further analyses.

| Relationship between CAER and geomorphic indices
These calculations enable the comparison between erosion rates to a number of geomorphic and geologic measures (Table 4).The burialcorrected mixed rates (i.e.CAER for the entire catchment) and the burial-corrected unmixed rates for the transient reaches (with the area upstream of the knickpoint removed) are compared alongside the recalculated published CAER (Buscher et al., 2013;Heineke et al., 2019) and the published CAER unmixed for the low erosion rate area, to Be and 26 Al analytical and derived erosion rate data with no corrections for subcatchments or sediment recycling/burial.Firstly, if the along strike geomorphic character of the uplifted footwall of the Gediz Graben boundary fault is examined, it is clear that the mean catchment relief (Figure 6a), maximum incision (Figure 6b) and mean catchment slopes (Figure 6c) of sampled catchments are variable (Figure 6b) but overall follow the trend in fault throw rate (Figure 6b) with minima in these geomorphic metrics coinciding with the mapped fault segment boundaries (dashed lines, Figure 6).Indeed, the clear relationship along strike of the geomorphic expression of active faulting was partly used by Kent et al. (2016) to determine long-term uplift rates along the Gediz Graben boundary fault (Figure 6b).If the relief (Figure 6a) and slope (Figure 6c) above and below the knickpoints are considered separately, the same overall trends are apparent but with higher relief and slopes downstream of the knickpoint in the central and western parts of the range.This result is expected as the transient wave of incision causes gorge formation and hillslope steepening as it propagates through the river system.In the eastern part of the range, this relationship is apparently inverted with higher slopes and relief above the knickpoint.Although, fewer data are available in this zone.
When the normalised steepness index in the transient reach is plotted along strike, then the highest steepness indices are present in the centre of the fault array (Figure 6d), where current fault slip rates are highest.Maximum stream powers also cluster within the central fault segment, although it is important to acknowledge that lower values of steepness index and stream power are also present in the central part of the fault zone.
T A B L E 2 26 Al/ 10 Be ratios, burial age and recalculated total catchment erosion rates based upon burial corrections.F I G U R E 5 26 Al/ 10 Be versus 10 Be ratio two isotope diagram showing burial model and concentration data scaled to surface production rates (Lal, 1991) for measured samples.Surface muon contributions of 0.99% ± 0.20% and 1.45% ± 0.29% were considered for 10 Be and 26 Al, respectively.Samples taken above the slopebreak knickpoint are indicted by the grey symbols.MMMC = Menderes Massif Metamorphic core complex.Error bars include analytical and production rate uncertainties.
When the along strike trends in CAER are considered, there is an increase from the westernmost sample (54.5 mMyr À1 ) into the centre of the range (250 mMyr À1 ) for both the raw CAER and burialcorrected rates (Figures 6e and 7).However, rates then decrease again along two large river systems in the centre of the range (TR18-06, Catili and TR18-09, Bozdaĝ) before increasing again along the eastern part of the range.This decrease in erosion rates in the centre of the fault appears unexpected given these catchments are experiencing the highest uplift rates.When the unmixed CAER are plotted (Figure 6e), a clearer pattern of lower rates at the fault tips and higher rates in the centre of the range appears although the CAER in the centre of the fault are still generally subdued.
Interestingly, there are also differences in the CAER along individual sampled river systems with both decreasing and increasing erosion rates downstream being present (Figure 7).For example, and as expected, CAER increases along the Kabazlı River from 59 mMyr À1 upstream of the Gediz Detachment fault to 157 mMyr À1 at the boundary fault (Figure 7).By contrast, along the Badınca River (samples TR18-16 to TR18-18; easternmost river), burial-corrected erosion rates decrease downstream from $250 mMyr À1 in the headwaters to 86 mMyr À1 upstream of the boundary fault.These data suggest that CAER do not scale simply with tectonic rates (c.f.Roda-Boluda et al., 2019) and may be influenced by factors such as sediment storage and contrasts in bedrock erodibility, which we evaluate below.
Secondly, the different CAER can also be compared with a range of topographic metrics that have previously been shown to correlate positively with erosion rates in previous studies such as relief and slope (i.e.Abbühl et al., 2011;Bellin, Vanacker, & Kubik, 2014;Kober et al., 2015;Miller et al., 2013).However, when the burial-corrected mixed rates (but not unmixed for low erosion rate areas) and published CAER data are plotted against mean catchment slope, topographic relief (150 m radius) and maximum incision depth upstream of the sample site, there are no trends (Figure S1).
By contrast, when these erosion rates are compared to the maximum upstream unit stream power, there is a significant ( p < 0.  8a; r 2 = 0.8).There are also significant ( p < 0.05) positive linear (r 2 = 0.6-0.9)relationships between erosion rates and T A B L E 3 Parameters used in the unmixing calculations to remove effect of low erosion rate and resultant erosion rates (Eb) for transient reach.
Catchment area (m 2 ) Erosion rates (LSDn)(mMyr Note: Zone on determined erosion rates. T A B L E 4 Geomorphic and geological variables by sample and river.2019) (Figure 8b) and a weak but significant linear relationship between erosion rates and throw rate on the graben boundary fault (Figure 8c; r 2 = 0.2).It is also noticeable that CAER expressed as mMyr À1 are lower than the slip rates on the basin bounding fault, particularly towards the centre of the fault, where displacement rates are 2 mm yr À1 (i.e.2000 mMyr À1 ; Figure 6b).
Thirdly, the unmixed CAER that represent erosion rates only in the transient reach of the rivers can be compared with the same metrics.When these rates (which include published data as well as the new data determined here) are plotted against mean catchment slope, topographic relief and maximum incision depth upstream of the sample location, again there are no clear or significant trends (Figure S2).
However, when unmixed CAER are compared to the upstream maximum unit stream power, there is a broad positive trend but with only a very weak correlation (Figure 9a).Although, when the Bozda g samples are removed as potential outliers, because this river has very high stream power yet low erosion rates in the centre of the fault, a significant ( p < 0.05) linear regression line with an r 2 = 0.25 can be plotted.
Similarly, there is no trend between k sn and CAER, but if the Gumusi cay sample is excluded as an outlier, there is weak (r 2 = 0.28) but significant ( p < 0.05) positive relationship between erosion rates and steepness index with the best fit regression being an exponential trend (Figure 9b).When all unmixed CAERs are plotted against fault throw rate, there is no trend; however, when the samples from the detachment are removed so that only samples close to or at the boundary fault are retained, there is a weak (r 2 = 0.1) but not significant (p > 0.05) positive power law relationship between these two variables (Figure 9c).

| Relationship between rock strength, geomorphology and erosion rates
In order to assess the impact that the different bedrock lithologies have on the geomorphic response in the study region, the erosion rates for the different catchments can be compared to measurements of bedrock strength.The bedrock of the Bozda g range can be broadly divided into the metamorphic lithologies of the Menderes Massif and the unconformably overlying Miocene and younger sediments.The metamorphic rocks are primarily composed of moderately strong to strong (c.f.Selby, 1980) schists, gneisses and granites where the SRMS > 60 (Figure 10a; c.f. Kent et al., 2021).By contrast, the syntectonic sandstones and conglomerates are weak to very weak (SRMS < 50).Therefore, if rock strength is the main control on CAER, then the harder metamorphic rocks should be eroding at a lower rate than the softer sediments.
Across the study region, the strong metamorphic rocks are located south of the Gediz Detachment in the upland regions of the Bozda g range, while the weak sedimentary rocks are mainly to the north, that is, a soft over hard transition as represented in many landscape evolution models (e.g.Forte, Yanites, & Whipple, 2016).
Interestingly though, when both measures of rock strength upstream of sample locations are compared to geomorphic variables such as relief (Figure 10b) and stream power (Figure 10c), there are no trends between the variables.This suggests that rock strength alone does not control relief or stream power.By contrast, there is a weak The geomorphology of the Bozda g Range is shaped by the uplift along the Gediz Boundary fault and concomitant incision of the bedrock rivers resulting from the linkage of the boundary faults at $0.8 Ma (Kent et al., 2017).Therefore, it is expected that there should be scaling relationships between various landscape metrics, uplift and erosion, similar to other regions around the world.For example, many studies show a positive relationship between CAER and catchment slope (i.e.Bellin, Vanacker, & Kubik, 2014;Roda-Boluda et al., 2019;Rosenkranz et al., 2018;Rossi et al., 2017) as well as positive relationship with channel steepness (Bellin, Vanacker, & Kubik, 2014;Cyr et al., 2010;DiBiase et al., 2010;Harkins et al., 2007;Miller et al., 2013;Rossi et al., 2017), which has been shown to be linear at low rates and steepness but becoming non-linear above a threshold a positive power law relationship with maximum incision depth.This is not unexpected assuming little pre-existing topography, as areas of higher relief will have had more material eroded than areas of lower relief over the same time span; thus, erosion rates should be higher where relief is higher.Though it is important to note that, in general, hillslopes have longer response times than rivers to changes in baselevel (Schlunegger et al., 2013;Simpson & Schlunegger, 2003).
Unexpectedly, these trends appear not to hold true along the Bozda g range either locally or catchment-wide, with no strong trends between erosion rates and average catchment slope, catchment relief or incision depth in either burial corrected CAER or unmixed for just the transient reach.While there are weak positive relationships observed in the data between CAER and normalised steepness index in the channel upstream of the sample point, this varies between a linear relationship for the whole CAER (not significant) and a weak but significant exponential for the transient reach only.The strongest and most significant of these weak trends is the linear relationship between the stream power and CAER (both burial-corrected and unmixed) albeit with larger uncertainties on the stream power data.
These last two observations indicate that at the catchment scale and at the precision of our data, the rivers are broadly in line with a simple form of the stream power law, which is linear and n = 1 (Whipple & Tucker, 1999) where E ≈ KA m S n and is consistent with the analyses of Kent et al. (2021).
When CAER are compared to throw rates, it is striking that erosion rates are around an order of magnitude lower than uplift rate.
Given the presence of knickpoints and a documented transient landscape response (Kent et al., 2017;Kent et al., 2021) demonstrating that this region is not in topographic steady state, this relationship is to be expected.As a result, the Bozda g region will be experiencing surface uplift (Figures 8c and 9c).Yet, there are only weak positive relationships between the throw rate and CAERs, when corrected for sediment storage and for the presence of low relief zones.increase to achieve the ideal fault profile (Kent, 2015); higher uplift rates will also result in increased erosion in these zones.This will also result in the part of the fault with the highest slip rates experiencing lower erosion rates, and as a result in the transient, reaches throw rate will not scale with erosion rate.Interestingly, at a catchment level, the CAERs do scale with throw rate but the correlation is weak, perhaps suggesting that prior to fault linkage throw rate did correlate with erosion rate.
A number of factors may cause the scatter and the weak correlations in these data, which we explore below.One complication to consider is that the results could be affected by sediment storage or non-uniform erosion as a result of landsliding (e.g.Binnie et al., 2006;Kober et al., 2012;Roda-Boluda et al., 2018).Neither of these factors appear to be likely along the Bozda g range as firstly, the potential effect of sediment storage has been corrected through the inclusion of 26 Al CRN data.Secondly, there is little evidence for F I G U R E 9 Comparison of geomorphic variables (a) maximum stream power and (b) upstream steepness index, and (c) throw rate on the Gediz Graben Boundary Fault against catchment-averaged erosion rate for previously published data and for samples collected here unmixed to remove the effect of the low erosion rate areas above the knickpoint.On C data have been separated into samples at the range front (dark) and at the detachment fault (light) to investigate the potential difference in erosion rates depending on the bedrock lithology.
significant landsliding in the study region to deliver material with sufficiently low 10 Be concentrations to perturb the measured river sediment concentrations.Incomplete sediment mixing could also explain the scatter in the data, while the measured CRN concentrations of repeat samples along several river systems are within 2σ error; we have limited data across the entire range to fully assess this issue, which has been shown to be a complicating factor in mountainous catchments elsewhere (Binnie et al., 2006).
Alternatively, the presence of inherited topography may play a significant role in the landscape response to uplift (c.f., Densmore et al., 2009).This explanation is supported by the clear imprint of the fault segments in the topographic metrics and the observation that in the eastern part of the range higher slopes and relief are found upstream of the tectonic knickpoint (Figure 6), despite transient river incision downstream of the knickpoint.Therefore, inherited topography might explain the disconnect between erosion rates and catchment wide variables such as slope and relief and potentially the variability in the CAER derived from the five samples taken from at or above the knickpoint.Yet, if this explanation was the only confounding factor, the unmixed CAER data should show stronger correlations with stream power and steepness index in particular, as the effect of low relief/low erosion rate zones have been accounted for in this calculation, and burial-corrected CAER for the whole catchments might be expected to show relationships with catchment mean slope or relief, which they do not.Therefore, another explanation for the spread in the data could be the influence of a strong lithological contrast within the catchments, which is discussed further below.

| The role of rock strength and lithology
A number of recent models have explored the impact of lithological variability on river evolution and erosion rates that could be used to Indeed, the presence of a very strong but thin cataclasite band found along the low-angle Gediz Detachment was used by Heineke et al. (2019) to explain the presence of low erosion rates and gentle slopes.In addition, they proposed that 'weak' phyllites and schists result in higher erosion rates in the centre of the range.The results presented here do not support this latter point, as lower CAERs are found in the centre of the range (Figure 7) and Figure 10e,f shows that the CAERs are invariant with rock strength upstream of the sample location despite a two-fold difference in strength between the sedimentary and metamorphic rocks overall (Figure 10a) and associated differences in erodibilty (Kent et al., 2021).This contradiction speaks to the difficulty in accurately constraining rock strength and erodibility in the field, determining the best categorisation, and linking such data to observed changes in fluvial behaviour and erosion rates (e.g.Bursztyn et al., 2015;Zondervan, Stokes, et al., 2020;Zondervan et al., 2020).
In addition, Forte et al.'s (Forte, Yanites, & Whipple, 2016) landscape evolution model also suggests that although the lithological boundary moves downstream over time, the erosion rates above and below the boundary will diverge.The soft rocks downstream will erode at the imposed uplift rate while the underlying hard rocks erode at a rate lower than the regional uplift rate.Another implication of Forte et al.'s (Forte, Yanites, & Whipple, 2016) landscape evolution model is that CAER would be perturbed or amplified downstream as a result of the lithological variation.We see that erosion rates of the underlying hard metamorphic rocks are eroding at rates lower than inferred uplift rates (Figure 9c), consistent with the landscape evolution model outputs.However, the erosion rates in the sedimentary bedrock reaches are also much lower than uplift rates at the graben boundary fault (Figure 9c), and only weakly and not significantly scale with throw rates on the fault.
Interestingly, Kent et al. (2021) demonstrated that stream power scales with uplift rate in the metamorphic bedrock reaches of their six study rivers.But uplift does not scale with stream power in the sedimentary reaches where sediment transport appears to be more important, resulting in a difference in the fluvial response in these reaches owing to the abundance of sedimentary material entering the river system (Kent et al., 2021).Therefore, while erosion rates in the sedimentary reaches still weakly, scale with the uplift rate the influence of sediment transport and hybrid or transport-limited nature of these lower reaches causes the erosion rate to be lower.In this study area, the lithological control on landscape evolution is therefore manifested not as bedrock erodibility but in variable fluvial responses that are not captured in a detachment-limited landscape evolution model.A key challenge for the future is to understand how the spatially variable erosion rates captured here are integrated over time to produce a coherent relief and sediment flux signal.

| CONCLUSIONS
Eighteen samples were collected for 10 Be and 26 Al cosmogenic nuclide analysis and combined with a further 15 previously published 10 Be concentrations (Buscher et al., 2013;Heineke et al., 2019) to determine CAERs along strike of the well-constrained Gediz Fault system in western Türkiye.This area features a significant lithological contrast where soft sediments overlie hard metamorphic rocks along a moderately dipping downstream contact; a series of north-flowing rivers are incising through this contact as a result of uplift along the fault at rates of up to 2 mMyr À1 and a fault-linkage event at $0.8 Ma (Kent et al., 2017).This natural laboratory allows the results of recent landscape evolution models investigating the role of such lithological contrasts to be tested.The background rate of erosion of the preincision landscape is determined as 46 ± 46 mMyr À1 and erosion rates within the transient reach vary from 16 to 1330 mMyr À1 .Although, erosion rates weakly scale with unit stream power, steepness index and slip rate on the bounding fault, there are no clear relationships between erosion rate and relief or catchment slope.Catchment-wide and within the transient reach erosion rates are an order of magnitude lower than slip rates for both metamorphic, and sedimentary reaches and despite a 30-fold difference in erodibility, there is no difference in the erosion rate between strong and weak rocks.This finding is at odds with the results of landscape evolution modelling and is likely owing to the influence of sediment transport on fluvial dynamics in the sedimentary reaches, that is, some rivers are not completely detachment-limited.While the weak relationships between other variables remain unexplained but maybe the result of incomplete sediment mixing or the influence of pre-existing topography prior to the onset of the current incisional phase.These findings indicate that the erosional response to uplift along an active normal fault is a complex response to multiple drivers that vary spatially and temporally.

ACKNOWLEDGEMENTS
We acknowledge NERC Facilities grant (CIAF/9189/1018) for the cosmogenic analysis and Geological Society field work grant both awarded to S.J. Boulton that supported this research.The code used to calculate the burial corrected erosion rates using both 10 Be and 26 Al can be found at: https://github.com/angelrodes/banana1026.
, rivers mentioned by name in the text are 9, Akçapınar; 15, Bozda g; 16, Gümüşcay; 17, Kabazli; 21, Kavaklidere; 23, Yeniköy.Stars show location of slope-break knickpoints; (c) topographic map of the study area (ALOS World 3D 30 m digital elevation model) showing the locations with numbers of samples collected during this study; (d) relief map of the study are showing the steepness index of the rivers and the location of CRN samples collected by Buscher et al. (2013) indicated by * and Heineke et al. (2019).Also shown are the location of five optically stimulated luminescence dates reported by Kent (2015) (unlabelled, blue circles) and the approximate location of the C 14 date of Sullivan (1988) labelled as Gölcük.[Color figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 3 Field photos showing landscapes and sampling in the Gediz region: (a) view of the downstream reach of the Akçapınar River-a river characterised by 100% metamorphic bedrock, (b) sampling in the knickzone of the Bozda g River, (c) sampling in the sedimentary reach of the Gümüşcay (note the well lithified Miocene clastic bedrock), (d) vertical step knickzone on the Kabazlı River at the boundary between the metamorphic basement and the sedimentary cover.[Color figure can be viewed at wileyonlinelibrary.com] 3.3 | Calculation of unit stream power Geomorphic indices were calculated using ArcGIS Pro and TAK (Forte & Whipple, 2018), k sn values were determined with a Θ ref = 0.45 following Kent et al. (2017, 2021) and the profiler function.While the choice of reference concavity can impact the resultant k sn values, Gailleton et al. (2021) demonstrated this is not significant.Kent et al. (2021) constrained the rock strength (using Schmidt hammer rebound and Selby Rock Mass strength [SRMS]) and specific bedrock erodibility, E, using the unit stream power model (c.f. 2 and the ALOS World 3D30 DEM.Similarly, the elevation is extracted at each point allowing the determination of local channel slope over each 100 m interval.The vertical accuracy of the AW3D30 DEM is <5 m (Tadono et al., 2016).As fieldderived measurements of width are not available for all rivers, width is calculated using the scaling relationships of Finnegan et al. (2005) and Whittaker et al. (2007) as well as using Kent et al.'s (2021) local hydraulic scaling relationship (see Supporting Information for more detail).These estimates of width are then used to derive the downstream distribution of unit stream power, ω, for each river.The maximum stream power was found for each river, and an average of the three stream powers taken.The error reported is the 2σ value on these values.

F
I G U R E 4 Conceptual diagram showing how different erosional zones add together to define total erosion rate at sample location.Top, a map view of a two zone mixing model showing the catchment areas above, Aa, and below, Ab, the knickpoint comprising the total catchment area Ac.Below, a topographic profile showing how the different zones relate to the transient river long profile with the samples collected at the knickpoint (star), EA, and at the river mouth, EC, allowing the determination of the erosion rate of only the transient, incising reach EB (modified from Rosenkranz et al. (2018).[Color figure can be viewed at wileyonlinelibrary.com] 4 | RESULTS 4.1 | 10 Be and 26 Al concentrations and catchmentwide erosion rate On five rivers, samples were taken at or upstream of the slopebreak knickpoint 03; 07; 14; 18).These samples represent the denudation rate prior to landscape rejuvenation and transient river incision, as a result of fault linkage $0.8 Ma(Kent et al., 2017), providing constraints for the unmixing model to determine the rate of erosion excluding these low erosion rate areas.Samples TR18-03 and TR18-07 give the lowest burial-corrected erosion rates at 32 and 60 mMyr À1 , respectively.Ridge crest erosion rates determined byHeineke et al. (2019) also fall in the range $30-90 mMyr À1 .Whereas, samples TR18-01 and TR18-18 give much higher rates of 174 and 248 mMyr À1 , respectively; while TR18-14 returns an intermediate value of 119 mMyr À1 .Significantly, these latter three samples have only small catchment areas upstream of the sample point (1.7-3 km 2 ), which may be below a threshold for an appropriate size of catchment area.Additionally, the CAER from 10 Be and 26 Al nuclides are not within error and consequently indicate variable sediment recycling, which is difficult explain in the metamorphic headwaters.Therefore, given the higher values than for the ridge crests, small catchment areas and incomplete mixing, these latter three samples are not used to determine the erosion rate upstream of the knickpoint.Instead, the average of the other two samples is taken to be representative of the low incision zone and used for all catchments (c.f.Roda-Boluda Al concentrations were measured by the 5-MV NEC Pelletron accelerator mass spectrometer (AMS) at SUERC(Xu et al., 2010).Measured 10 Be is normalised to the NIST_27900 standard with an assumed isotope ratio of 2.79 Â 10 À11 and equivalent to 07KNSTD within rounding error.26 Al is normalised to Z92-0222 with a defined isotope ratio of 4.11 Â 10 À11 and is equivalent to KNSTD.b Denudation rates were calculated using the online CRONUS-Earth calculator v 3.0(Balco et al., 2008) using the LSDn scaling and a sample density of 2.65 gcm 3 .c Sample elevation and shielding were derived from the ALOS World3D 30 m digital elevation model.investigate the relationships between different factors and erosion along the southern margin of the Gediz Graben.
the published data ofBuscher et al. (2013) andHeineke et al. ( ≤ 0.2) negative linear relationship between rock strength (SRMS and Schmidt hammer rebound) and the upstream steepness index, suggesting that rivers are on average less steep when the rocks are harder.However, this is not significant for either RMS or Schmidht Hammer rebound (p > 0.05) and is the opposite of the relationship that we would expect where the river is steeper in harder rocks.F I G U R E 6 Along strike trends in geomorphic variables and catchment-averaged erosion rates (CAERs).Dashed lines show fault segment boundaries from Kent et al. (2017).(a) Catchment relief (mean whole catchment, mean above and below the tectonic knickpoint, and elevation mean and maximum swath profiles; (b) channel incision in the transient reaches and long-term throw rates (Kent et al., 2017); (c) total mean catchment slope and mean slope above and below the knickpoint; (d) normalised steepness index and maximum unit stream power, and (e) CAERs.Note: error bars are shown where greater than symbol size.[Color figure can be viewed at wileyonlinelibrary.com]Furthermore, when CAERs are compared to the upstream rock strength, there is no clear relationship either for mixed or unmixed rates with both strong and weak rocks resulting in a similar range of CAERs (Figures 10e,f).Finally, there are no clear trends of these variables with uplift rate on the fault as indicated by the size of the symbols on Figures 10b-f.5 | DISCUSSION 5.1 | What controls erosion rates along the margin of the Gediz Graben?
steepness index.Related to landscape steepness is relief, which can either be measured as topographic relief across the catchment, or following Roda-Boluda et al. (2018) as maximum incision depth (i.e.maximum local relief) along the river.In both measures, CAER have previously been shown to have a positive relationship with these factors.For example, Bellin, Vanacker and Kubik (2014) demonstrated a positive linear relationship with relief and Roda-Boluda et al. (2018)

F
I G U R E 7 Map of showing the catchment-averaged erosion rates along the Gediz Graben Boundary Fault.Yellow circles show previously published data(Buscher et al., 2013;Heineke et al., 2019); while red circles show rates derived here but without correction for sediment storage and recycling.Rates corrected for these factors are shown by the shading of the catchment areas.[Color figure can be viewed at wileyonlinelibrary.com]Additionally, it is striking that these relationships are only significant for the burial-corrected CAERs, not for the unmixed CAERs.However, this apparent contradiction is consistent with the documented fault linkage.After a fault linkage event, the highest erosion rates should be present in the linkage zones where the previous minimum in fault throw (as these were the tips of individual faults) have had to rapidly F I G U R E 8 Comparison of geomorphic variables (a) mean maximum unit stream power and (b) normalised steepness index upstream, and (c) throw rate on the Gediz Graben Boundary Fault against catchment-averaged erosion rates for previously published data (1:Buscher et al., 2013; 2: Heineke et al., 2019)  with internal uncertainty and for all samples collected here corrected for burial and sediment storage with calculated errors but not unmixed further.

F
I G U R E 1 0 (a) Total Schmidt hammer rebound and SRMS for the main lithologies present in the study area.Schmidt hammer and SRMS calculated over 2 km upstream of the sample locations on the six main study rivers plotted against: (b) topographic relief; (c) maximum stream power; (d) upstream normalised steepness index; (e) catchment-averaged burial corrected erosion rates, and (f) unmixed erosion rates for the transient reach of the rivers.On B-E the size of the circle proportionally represents the throw rate at the range front where the largest circles equal 2 mmyr À1 .[Color figure can be viewed at wileyonlinelibrary.com] understand the relationships between CAER and the topographic metrics.Forte et al.'s (Forte, Yanites, & Whipple, 2016) model of using two distinct lithologies is highly applicable to the Gediz Graben.Their work demonstrated that when soft rocks overlie hard rocks along downstream dipping contact, the lithological contact becomes an important and persistent topographic feature in the landscape with the contact's dip-slope being preserved.This can clearly be seen in the study area as the Gediz Detachment is a pervasive feature along much of the range, and in many interfluve areas, the detachment is well preserved with little evidence of deep erosion.