Earth and Planetary Science Letters 307 (2011) 309322

Contents lists available at ScienceDirect

Earth and Planetary Science Letters

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e p s l

Rethinking low-temperature thermochronology data sampling strategies for quantication of denudation and relief histories: A case study in the French western Alps
Pierre G. Valla , Peter A. van der Beek, Jean Braun
Institut des Sciences de la Terre, Universit Joseph Fourier, CNRS, BP 53, F-38041 Grenoble, France

a r t i c l e

i n f o

a b s t r a c t
We assess the importance of thermochronometric data sampling and modeling strategies for correctly estimating mountain belt exhumation. Thermochronological age-elevation proles have been widely used to infer orogenic exhumation histories; however, recent studies have shown that this sampling strategy may not be the most pertinent for quantifying both denudation and relief history. Here, we investigate the ability of combining different thermochronology data sampling schemes with numerical modeling to better constrain denudation rates and relief changes. We produce synthetic thermochronology datasets for real Alpine topography under a specic exhumation and relief scenario using the thermal-kinematic model Pecube. We then adopt an inverse approach based on the Neighborhood Algorithm to quantitatively assess the resolution of different thermochronology datasets collected following elevation proles, long transects and valley bottom sampling. We also test the effect of the modeling approach on denudation and relief predictions, in particular the inuence of the topographic grid resolution and of potential constraints on the geothermal gradient. Our results show that sampling along a single elevation prole does not allow to quantitatively constrain both denudation and relief histories. Numerical outputs clearly evidence tradeoffs that limit the capacity of simultaneously resolving denudation rates and relief change. Quantitative predictions are only slightly different when combining elevation proles along different valleys, but are highly improved when using long transects or valley-bottom samples combined with an elevation prole. The resolution with which relief evolution can be predicted may be increased by a factor of 2 by using spatially distributed datasets. Results of thermal parameter inversions suggest that the geothermal gradient may be better estimated using elevation proles or long-transect sampling rather than using valley bottom samples. Simulations with different model topography resolutions show that degrading the resolution for computational efciency may result in a loss of quantitative information on denudation rates and relief history. In summary, we highlight that both thermochronological sampling strategies and the choice of thermal parameters or model topography resolution have a signicant inuence on predicted denudation and relief histories. Ideally, the sampling strategy should be designed using preliminary modeling of expected denudation and relief histories, and a sensitivity study on assumed thermal parameters and model resolution should be performed when modeling the data. Although our modeling is based on a particular case study of relief evolution in the French western Alps, we believe that these inferences have general relevance for thermochronological studies within mountain belts. 2011 Elsevier B.V. All rights reserved.

Article history: Received 10 December 2010 Received in revised form 29 April 2011 Accepted 2 May 2011 Available online 31 May 2011 Editor: T.M. Harrison Keywords: low-temperature thermochronology numerical modeling inversion exhumation relief development sampling and modeling strategies

1. Introduction Low-temperature thermochronology has been widely used to quantify the exhumation history of mountain belts (e.g., Braun, 2005; Gallagher et al., 1998; Reiners and Brandon, 2006). Numerous studies in the last 20 yr or so have used apatite ssion-track (AFT) (e.g., Fitzgerald et al., 1995; Gallagher et al., 1998; Hurford, 1991) and (U-Th)/He (AHe) (Ehlers and Farley, 2003; Farley, 2002) data to infer exhumation histories

Corresponding author. E-mail address: pierre.valla@ujf-grenoble.fr (P.G. Valla). 0012-821X/$ see front matter 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.epsl.2011.05.003

in specic areas. Moreover, low-temperature thermochronology may provide useful constraints for assessing paleo-topography and relief development (e.g., Clark et al., 2005; House et al., 1998; Reiners, 2007). The effect of steady-state topography on thermochronology data is well understood (e.g., Mancktelow and Grasemann, 1997; Stwe et al., 1994). However, potential effects of time-varying topography, leading to spatially and temporally varying exhumation rates as well as a transient thermal structure, are more difcult to assess (Braun, 2002a; Valla et al., 2010). Thermochronometric sampling strategies have often focused on obtaining age-elevation proles, i.e., datasets of AFT and/or AHe ages collected along a steep elevation transect (e.g., Fitzgerald et al., 1995;

310

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

Gallagher et al., 2005; Wagner and Reimer, 1972). Thermochronologic age-elevation relationships allow inferring apparent exhumation rates over a specic time window and, in theory, do not require knowledge of the geothermal gradient. One-dimensional consideration of age-elevation relationships thus remains the most direct way to interpret thermochronology data, either by simple linear regression of the data (e.g., Fitzgerald et al., 1995; Wagner and Reimer, 1972) or by using 1D thermal evolution models (e.g., Gallagher et al., 2005). However, topographic effects on thermochronology data are not easily extracted from age-elevation relationships (Braun, 2002a) and recent modeling studies suggest that constraining mountain belt paleo-relief from elevation proles is difcult, even when combining multiple thermochronometers (Valla et al., 2010; van der Beek et al., 2010; Vernon et al., 2009). Recent developments in numerical modeling offer tools for interpreting low-temperature thermochronology data, either considering each sample individually (e.g., Ketcham, 2005) or exploiting the spatial correlation between samples (Braun et al., 2006; Gallagher et al., 2005; Stephenson et al., 2006a), to derive thermal histories and eventually constrain both denudation and relief histories. Braun (2002b) proposed an alternative approach to assess paleo-relief by sampling along a horizontal transect and using the spatial coherence between the topographic and thermochronological signals to determine both regional denudation rate and relief change (Beucher et al., submitted for publication; Herman et al., 2007, 2010a). However, the main limitation of this spectral method is that it assumes a spatially and temporally constant denudation history, and only provides an upper bound on the timing of denudation or the age of relief change. Numerical modeling offers the unique possibility to exploit dense spatial datasets either with or without any specic spatial relationship (Braun and van der Beek, 2004; Gallagher et al., 2005; Stephenson et al., 2006a). In this study, we aim to assess under what sampling and modeling conditions quantitative constraints on the exhumation history can be extracted from low-temperature thermochronology data. We dene different sampling strategies for a specic topography and test whether each sampling scheme permits deriving both denudation rates and relief history. Moreover, we assess to what extent the modeling approach, especially the spatial resolution of the model grid or any imposed constraint on the geothermal gradient, can bias predicted denudation and relief histories. Our approach, although being specic with respect to the study area and the relief scenario we simulated, has been designed to more generally assess the potential inuence of specic sampling schemes for constraining orogen exhumation. Our study thus follows an experimental design approach that has been widely used in geophysics (e.g., Curtis, 2004a,b; Wood and Curtis, 2004), although for computational simplicity we do not perform a complete inverse analysis but rather a priori dene specic 3D sampling schemes that are traditionally used in thermochronological studies. Several studies (Glotzbach et al., 2011; Valla et al., 2010; van der Beek et al., 2010) have recently attempted to extract independent information on denudation and relief development in the western European Alps from thermochronology data. This was achieved by combining the 3D thermal-kinematic model Pecube (Braun, 2003; Braun et al., submitted for publication), which predicts thermal histories and thermochronological ages from an input denudation and relief history, with an inversion scheme based on the Neighborhood Algorithm (Sambridge, 1999a,b) to search the parameter space, extract best-tting scenarios from the data and derive quantitative estimates for the parameters (rates, timing and relief ratio) describing denudation and relief histories. We adopt a similar numerical approach as Valla et al. (2010); however, their study specically focused on the problem of interpreting ageelevation proles whereas here we consider various sampling strategies. Moreover, instead of building a synthetic topography for inverse modeling (Valla et al., 2010), we use real topography for predicting synthetic thermochronology data (Fig. 1a). We focus on the Ecrins

Pelvoux massif (van der Beek et al., 2010), one of the External Crystalline Massifs (ECMs) within the western European Alps (e.g., Dumont et al., 2008). The Neogene exhumation history of the ECMs has been widely studied in the last few years, leading to contrasting views regarding timing and rates of exhumation, and the impact of Pliocene-Quaternary climate change on late-stage exhumation (Glotzbach et al., 2008, 2010; Reinecker et al., 2008; van der Beek et al., 2010; Vernon et al., 2009). Our synthetic approach is motivated by recent predictions for the exhumation history of the EcrinsPelvoux massif from thermochronology data collected along either an elevation prole (van der Beek et al., 2010) or a long transect (Beucher et al., submitted for publication); both studies provided precise constraints on denudation histories but could not resolve relief evolution quantitatively. More generally, our study has been motivated by several attempts to quantify both timing and magnitude of topographic evolution in mountain belts using thermochronology (e.g., Braun, 2005; Ehlers and Farley, 2003; Reiners, 2007; Valla et al., 2010). In the following, we rst outline our modeling approach, present the synthetic data used for the simulations and discuss the inverse method. We then show inversion results and present to what extent predictions of denudation rates, relief change and near-surface thermal structure can be quantitatively determined by various thermochronological sampling strategies. The potential effects of varying the model grid resolution are reported in the Supplementary data. We nally discuss the implications of our ndings for thermochronological sampling strategies to constrain regional denudation rates and/or topographic evolution of mountain belts. 2. Modeling approach 2.1. Thermal-kinematic modeling and synthetic thermochronology data The EcrinsPelvoux massif topography is characterized by a dominant NorthSouth topographic wavelength of ~ 10 km (Figs. 1a and 2a) organized around four major valleys (for details and locations see van der Beek et al., 2010). The relief amplitude is ~3 km, with summits around 4000 m and valley bottoms at ~ 1000 m (Fig. 1a). The EcrinsPelvoux massif has been chosen for our study as it forms a homogeneous area for the purpose of understanding Neogene exhumation and relief development, i.e., no major active fault crosses the massif and consequently its exhumation history can be assumed to be spatially uniform (Delunel et al., 2010; van der Beek et al., 2010). Based on this topography and an assumed exhumation history, we predicted synthetic thermochronology data using the Pecube model (Braun, 2003). Pecube is a 3D thermal-kinematic model that calculates the thermal structure through time within an exhuming crustal block and provides thermal histories for rock particles that end up at the surface of the model, taking into account the effects of evolving topography (see Valla et al., 2010; Braun et al., submitted for publication for a detailed discussion). The Pecube nite-element code permits dening exhumation/relief scenarios that are described as discrete phases with an imposed regional denudation rate and paleo-relief. Relief evolution is quantied using the relief ratio R: for R = 0, the paleo-relief is a plateau at maximum elevation; for R = 1, the paleo-relief is equal to present-day conditions; and for R = 2, the paleo-relief is twice as large as the present-day relief. Note that relief increase operates through preferential valley incision in our model (Fig. 2a); i.e., ridges remain at a constant elevation but valley bottoms are lowered (see Valla et al., 2010; Braun et al., submitted for publication for a complete description), as would be expected for relief increase by glacial valley carving for instance. Isostatic rebound in response to this erosional unloading is included. Thus, local exhumation rates are the sum of: (1) the imposed regional (background) exhumation; (2) local relief change through valley incision; (3) spatially varying rock uplift induced by isostatic rebound. Thermal, mechanical and kinematic parameters used in the forward model are given in Table 1. The basal crustal temperature (Temp) and

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

311

La Meije

Barre des Ecrins

Mont Pelvoux

4100 3700
624

Ro ma nc he

3300

Se ver ais se

2500

2100 1700 1300 900

450 0

Latitu

de (

N)

4 4 4

600

Lo

ng

4506

ne on B

itud

612

500

Elevation (m)

on Vene

2900

e(

E)

Latitude (N)

4500'

Latitude (N)
4448'

1_Profile 3_Profiles Valleys

600' 612'

4448'

4500'

9 11 13 15
624'

Transect Profile + Valleys

600' 612'

9 11 13 15
624'

AFT age (Ma)

Longitude (E)

AHe age (Ma)

Longitude (E)

Fig. 1. a) 3D map view of the modeling area (90-m resolution SRTM DEM of the EcrinsPelvoux massif, Western Alps, France) used to produce AFT and AHe ages maps (b, c). Names of principal valleys and peaks are indicated for orientation. bc) Apatite ssion-track (b) and (U-Th)/He (c) age map produced using Pecube (Braun, 2003; Table 1) and simulating a scenario of 50% relief increase (Rinput = 0.5) during the last 3 Ma (Tinput). The constant background denudation rate is 0.3 km Myr 1 (E1input = E2input) and resulting maximum rate of relief increase over the last 3 Ma is ~ 0.5 km Myr 1 (see text for discussion). Sample locations for the 5 different sampling schemes are shown: (b) Sampling schemes 1_Prole (yellow circles, based on van der Beek et al., 2010), 3_Proles (black circles) and Valleys (white circles). (c) Sampling schemes Transect (red circles, partly based on Beucher et al., submitted for publication) and Prole + Valleys (yellow circles, partly based on van der Beek et al., 2010).

the crustal heat production (H_prod; here normalized to heat capacity and density) are xed to produce a surface geothermal gradient of ~ 25 C km 1. Predicted thermal histories are then used to calculate different thermochronometric ages. In this study, we focus on late-stage exhumation histories and thus only compute low-temperature apatite ssion-track (AFT) and (U-Th)/He (AHe) ages, as well as mean conned track lengths (MTL; see Valla et al., 2010). Thermal histories

are translated into AHe ages using a simple model for He production diffusionejection (Farley, 2000); AFT ages are calculated using a model for AFT annealing (Green et al., 1989; Stephenson et al., 2006b). We acknowledge that using these simple algorithms does not completely reect the complex annealing and diffusion processes included in low-temperature thermochronology (Flowers et al., 2009; Gautheron et al., 2009; Ketcham, 2005). However, as discussed in Valla et al. (2010), we use the same forward models for AFT and AHe

312

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

North

South

Elevation (km)

1 0 5 10 15 20 25 30 35

a
Distance (km)

3.5

3.5

b
-1

Elevation (km)

1.5

Elevation (km)
-1

2.5

2.5

1.5

AFT AHe
12 14 0.5 10 11 12 13 14

0.5

10

Age (Ma)

MTL (m)

Fig. 2. a) Topographic cross-prole along the sampling scheme Transect (see Fig. 1b for location within the study area). Solid black and dashed red lines represent present-day and paleo-topography (R = 0.5, see text for discussion), respectively. Black circles show sample locations for the sampling scheme Transect, and arrows represent relief evolution along the topographic prole (background denudation rate is 0.3 km Myr 1). (b, c) Synthetic thermochronological ages versus elevation for Transect samples, including AFT and AHe ages (b) and mean ssion-track length data (MTL, c). Apparent exhumation rates (AER) deduced from linear regression of elevation vs age (b) are 0.34 and 0.33 km Myr 1 for AHe and AFT data, respectively (r = 0.980.99). Note that the MTL pattern shows a normal correlation between MTL and elevation (c).

age predictions when generating the synthetic data and resolving the inverse problem; the results shown here are thus largely independent of the specic age-prediction models. To produce the synthetic datasets that we will use in the inversion, we dene a 2-phase exhumation scenario with steady-state denudation (0.3 km Myr 1) over 15 Myr and a gradual 50% relief increase (R = 0.5, Fig. 2a) since 3 Ma (T = 3 Ma). This scenario simulates moderate relief

Table 1 Thermal, kinematic and elastic parameters used in Pecube. Crustal thickness, basal temperature (Temp) and heat production parameter (H_prod) are set to obtain a geothermal gradient of ~25 C km1 and a crustal heat production of ~7.5 C Myr 1. Poisson ratio, Young's modulus and equivalent elastic thickness are used for calculating the isostatic rebound in response to relief change. Equivalent elastic thickness is set to a value that simulates moderate isostatic rebound (Valla et al., 2010; van der Beek et al., 2010). Parameter (unit) Crustal thickness (km) Crustal density (kg m 3) Sublithospheric mantle density (kg m 3) Young's modulus (Pa) Poisson ratio Equivalent elastic thickness (km) Thermal diffusivity (km2 Myr 1) Basal crustal temperature (C) Crustal heat production parameter Sea-level temperature (C) Atmospheric lapse rate (C km 1) Inversions value 40 2700 3200 1.1011 0.25 25 25 600 0.8 15 6

increase (~1.5 km over the last 3 Myr, i.e., a maximum relief growth rate of ~0.5 km Myr 1) associated with a relatively low and constant background denudation rate (0.3 km Myr 1). Although this choice for a specic denudation and relief scenario renders our analysis somewhat contextual, we focus on the type of scenario that has been inferred for the western Alpine massifs (e.g., Glotzbach et al., 2011; van der Beek et al., 2010; Vernon et al., 2009). Thermochronology data are predicted following this input scenario on a very high-resolution grid (90-m, Fig. 1b,c) to ensure that the numerical model properly resolves niteamplitude topography effects on the crustal thermal structure. AFT and AHe age maps (Fig. 1b,c) show a signicant age contrast between valley bottoms (AFT ~7 Ma and AHe ~34 Ma) and ridges (AFT ~15 Ma and AHe ~1213 Ma). Note that our synthetic AFT and AHe data are similar to measured thermochronological ages in the area (Beucher et al., submitted for publication; van der Beek et al., 2010), suggesting that the imposed tectono-geomorphic scenario we dene is reasonable. We sampled both the AFT (and associated MTL) and AHe maps following ve different strategies: (1) a single elevation prole (Fig. 1b, 1_Prole), partly based on sample locations reported by van der Beek et al. (2010); (2) three elevation proles sampling the three main valleys of the massif (Figs. 1b and 3b, 3_Proles); (3) valley bottom samples along the major valleys (Figs. 1b and 3a, Valleys); (4) an elevation prole combined with valley bottom samples (Figs. 1c and 3c, Prole + Valleys); and nally (5) a long transect (partly based on sample locations in Beucher et al., submitted for publication) across the massif (Figs. 1c and 2, Transect). Each sampling scheme includes exactly 20 samples for which AFT, AHe and MTL data are

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

313

available. We thus compare the predictions of different sampling strategies while maintaining identical dataset size. Figs. 2 and 3 show synthetic AFT and AHe data plotted against elevation for the different sampling schemes. All AFT and AHe age-elevation relationships suggest nearly constant apparent exhumation rates (AER) of 0.330.39 and 0.330.47 km Myr 1 (see Figs. 2 and 3, correlation coefcients r 0.950.99), respectively. The AER slightly overestimate the input background denudation rate due to additional local exhumation following relief growth (most strongly recorded by valley-bottom samples, Fig. 3a). 2.2. Inverse thermal-kinematic modeling For each sampling scheme, we run inversions to recover the four parameters used to describe the imposed tectono-geomorphic scenario, dening a range in which the inversion procedure will search for each: (1) denudation rate of the rst phase (E1: 0 2 km Myr 1); (2) denudation rate of the second phase (E2: 0 2 km Myr 1); (3) transition time (T: 0.514.5 Ma); and (4) relief factor (R: 02). Each inverse simulation is run over 15 Myr. We run inversions constrained by the full dataset (AFT + AHe + MTL) as well as subsets of the input thermochronology data (AFT, AHe, AFT + AHe, AFT + MTL, and AFT + AHe + MTL data) to test what combination of thermochronometers is most useful to quantify relief and denudation histories in different sampling schemes.

In a second set of inversions, we test the ability of the different sampling schemes to constrain not only relief and denudation histories, but also to retrieve the geothermal gradient (Temp and H_prod parameters, Table 3). To achieve this, we run ve additional inverse simulations with the full dataset and 6 parameters to invert (E1, E2, T, and R as described previously; Temp: 500800 C, and H_prod: 03). We acknowledge that other thermo-physical parameters such as diffusivity or thermal conductivity, as well as the surface temperature history, are largely unknown and may also vary; but given the relatively homogenous lithology in our study area, we suspect they will do so within much narrower limits than the thermal parameters we test for and will therefore exert only a second-order inuence on model predictions. Finally, we test the inuence of the grid resolution on relief and denudation predictions (see Supplementary data for results and discussion). To perform the inversions, we use the Neighborhood Algorithm (NA) of Sambridge (1999a,b). NA is a two-stage numerical procedure for non-linear inverse problems that allows efcient exploration of a multidimensional parameter space to nd best-tting models (NA sampling stage) as well as deriving estimates of posterior probability distribution for inverted parameters from the ensemble of models run during the sampling stage (NA appraisal stage). The sampling stage iteratively explores the multi-dimensional parameter space and progressively concentrates on regions where parameter values minimize the mist to the data. NA makes use of

a Valleys
3 3
-1

b 3_Profiles

Elevation (km)

-1

Elevation (km)

-1

AFT AHe
4 6 8 10 12 14

-1

North Center South

12 14

10

Age (Ma)
4

Age (Ma)

c Profile + Valleys
3
-1

Elevation (km)

2
-1

Valleys Profile
4 6 8 10 12 14

Age (Ma)
Fig. 3. Synthetic thermochronology data (AFT and AHe ages) versus elevation for different sampling schemes; see Fig. 1 for sample locations. a) Synthetic data for the sampling scheme Valleys. AER deduced from linear regression of elevation vs age are 0.47 and 0.39 km Myr 1 for AHe (white circles) and AFT (black circles) data, respectively (r = 0.98). b) Synthetic data for the sampling scheme 3_Proles (individual proles are represented by diamonds, squares and triangles, see Fig. 1b for locations). AER are 0.33 and 0.35 km Myr 1 for AHe (white) and AFT (black) data, respectively (r = 0.950.99). c) Synthetic data for the sampling scheme Prole + Valleys (prole and valleys samples are represented by diamonds and circles, respectively). AER are 0.37 and 0.36 km Myr 1 for AHe (white) and AFT (black) data, respectively (r = 0.980.99).

314

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

simple geometrical concepts by dividing the entire parameter space into a set of n-dimensional Voronoi cells, which represent the nearest neighborhood around each parameter combination. At each step, NA samples a forward model centered on each Voronoi cell and calculates a mist surface. Subsequent iterations make use of this mist surface to sample in regions of the parameter space where the mist function is minimized. The performance of the sampling stage is controlled by tuning parameters determining the number of iterations, the number of forward models generated at each iteration, and the number of Voronoi cells to re-sample after each iteration. Varying these parameters will make the NA method either more explorative of the multi-dimensional parameter space, or more exploitative and faster in converging toward potential mist minima (Sambridge, 1999a). Based on previous studies (Glotzbach et al., 2011; Valla et al., 2010), we choose to run ~ 5000 models distributed in 35 iterations for each inversion, with 50% of models re-sampled at each iteration. In our approach, we use a weighted least-squares mist function for the NA appraisal stage: v  2 u i i u N M j; mod j;dat u = t ji 2 i=1 j=1

3.1. Sampling strategy The rst sampling scheme (1_Prole, Table 2) considers 20 samples combined as a single elevation prole. Inversion results yield accurate predictions for E1 (E1 =0.230.28 km Myr1; E1input =0.3 km Myr1) for all input thermochronological datasets with uncertainties ~2040% except when using AHe data alone. Estimates on E2 are less precise (uncertainties ~3080%) and slightly overestimated when using AFT data alone or AFT +MTL data (E2 ~0.4 km Myr1; E2input =0.3 km Myr1). AHe data are required (alone or combined with AFT ages) to provide accurate predictions for E2 (respectively E2 = 0.330.11 and 0.28 0.18 km Myr 1). Adding MTL data, in contrast, does not improve inversion predictions (Table 2). Timing predictions are underestimated when using AFT alone, AFT + MTL or AFT+ AHe+ MTL data (T =1 1.4 Ma) and clearly overestimated for AHe alone (T= 12.2 Ma). Only the AFT+ AHe inversion (T =2.8 Ma) gives predicted value of T close to the input value of 3 Ma. However, in all cases uncertainties on T predictions are very high (N100% for all inversions). Relief predictions are in agreement with the input value (Rinput =0.5) when using AHe data alone or combined with AFT ages (R= 0.53 0.58 and 0.63 0.55 respectively). In contrast, using AFT ages alone or combined with MTL measurements respectively under- (R=0.93 0.59) and over-predicts (R=0.18 +0.93/0.18) relief increase. Combining both AFT and AHe ages with MTL does not lead to an accurate estimate on R (R= 1.08 0.55; i.e., between ~50% increasing and 50% decreasing topographic relief). Moreover, as for T predictions, uncertainties on R are higher than 100% for all inversions. In summary, inverse simulations using thermochronology data along a single elevation prole provide accurate estimates on denudation histories (E1 and E2) with moderate to high uncertainties depending on the dataset (~30 to 80%). In contrast, T and R estimates are not in agreement with input values and systematically present uncertainties N100%. Best estimates are generally obtained when using AHe ages alone or combined with AFT data; adding MTL measurements does not improve denudation and relief predictions. Sampling along three different elevation proles (3_Proles, Table 2) provides similar predictions for E1, although the mean uncertainty is slightly reduced (~2040%, except for AFT + AHe data for which the uncertainty rises to 100%). E2 estimates are accurate for all inversions, but uncertainties remain quite high (~3080%). Similar to the previous inversions, T is systematically under-predicted (T = 1.02.1 Ma) except for the AHe inversion (T = 6.9 3.8 Ma); uncertainties remain N100% for all inversions. Relief predictions are much better than in the simulation using a single age-elevation prole. All inversions predict relief increase (R = 0.180.58), but the inverse simulation combining all thermochronometric systems clearly overestimates the R input value (R = 0.18 +0.30/0.18). Uncertainties on R predictions are also reduced (~6590%, Table 2). As for a single elevation prole, best estimates on E2 and R (recent exhumation history) are provided using AHe data; however, for both T and E1, most accurate predictions are obtained when using the full dataset (AFT + AHe+ MTL data). Sampling along valley bottoms (Valleys, Table 2) provides accurate predictions for E1 and E2. Uncertainties on E1 are low (~ 320%) only when using at least two thermochronometers, but are much higher when using AFT or AHe data alone (~ 100200%). In contrast, uncertainties on E2 are fairly constant, whatever the dataset used (~4060%). Similarly, constraints on T are very poor using AFT or AHe data alone (T = 11.9 +2.6/5.1 and 6.6 3.0 Ma, respectively) and are improved when combining thermochronometers (best estimate T = 3.5 2.3 Ma for AFT + AHe + MTL data). Uncertainties on T are high and variable depending on the dataset used: N100% for all inversions except for the full dataset. Finally, relief increase is strongly overestimated when using AFT or AHe data alone (R = 0.030.08) and predictions are not precise (uncertainties ~ 100%); however, combining the thermochronometers provides more accurate values of R and reduced uncertainties (R = 0.330.48 and uncertainties ~ 50%). In

where N is the number of datasets (13; AFT, AHe, AFT+ AHe, AFT+MTL, or AFT +AHe+MTL data), M is the number of samples in each dataset i i (20 samples in total for each sampling scheme), j, mod and j, dat are predicted and observed values for AFT/AHe ages or MTL respectively, and ji is the uncertainty on the data. Here, we use constant synthetic uncertainties (ji) of 0.5 Ma and 0.5 m for AFT/AHe ages and MTL, respectively, to put an equal weight on all samples for mist calculations (Valla et al., 2010). The above approach allows us to quantitatively infer the potential resolution of perfect thermochronology data in resolving tectono-geomorphic scenarios (and potentially thermal parameters). We note, however, that adding random noise in the data could potentially lead to different model sensitivity in constraining numerical parameters. In the NA appraisal stage (Sambridge, 1999b), the model ensemble generated during the sampling stage is iteratively resampled to provide optimal estimates in the form of marginal probability-density functions (1D or 2D PDFs) of parameter values. For all parameters, we assume a uniform prior distribution within each parameter range. The posterior 1D (or 2D) PDFs are given based on the log-likelihood function (ln(L)), following the equation:  21 ij; mod ij;dat A lnL = 0:5 2 = 0:5@ ji 2 i=1 j=1 0
N M

3. Inversion results We rst present inversion results for each sampling scheme to assess the effect of using various combinations of low-temperature thermochronometers. We then compare the different sampling schemes to discuss how best to predict denudation and relief histories. Finally, we show modeling results for inverting the geothermal gradient. We present inversion results as 1D marginal posterior PDF's as well as 2D joint PDF's of parameter values and discuss the reasons for the observed differences in the modeling outputs (resolution and errors on parameter estimates, and potential tradeoffs between model parameters) in the discussion section. Optimal parameter values (mode of the 1D posterior PDF) and 1 uncertainties for the different sampling schemes and data combinations are given in Table 2.

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

315

Table 2 Output estimates after the NA appraisal stage for input parameters E1, E2, T and R (see text for description and discussion). The second line reports parameter ranges and input values to calculate the synthetic thermochronology data (Fig. 1b,c). Optimal parameter values and associated uncertainties (mode 1 of the 1D posterior PDF) are given in the following lines for various sampling schemes and different input thermochronology datasets. Best parameter estimates (and associated input thermochronology dataset) are also reported as percentage errors and uncertainties compared to input values (%Input). Parameter (unit) Range/input 1_Prole AFT AHe AFT + AHe AFT + MTL AFT + AHe + MTL Best estimate (%Input) Data 3_Proles AFT AHe AFT + AHe AFT + MTL AFT + AHe + MTL Best estimate (%Input) Data Valleys AFT AHe AFT + AHe AFT + MTL AFT + AHe + MTL Best Estimate (%Input) Data Prole + Valleys AFT AHe AFT + AHe AFT + MTL AFT + AHe + MTL Best Estimate (%Input) Data Transect AFT AHe AFT + AHe AFT + MTL AFT + AHe + MTL Best estimate (%Input) Data E1 (km Myr 1) 02/0.3 0.28 0.13 0.28 0.23 0.28 0.06 0.23 0.1 0.28 0.06 7% 20% AFT + AHe E2 (km Myr 1) 02/0.3 0.38 0.26 0.33 0.11 0.28 0.18 0.43 0.24 0.28 0.27 + 10% 37% AHe T (Ma) 0.514.5/3 1.4 + 5.3/0.9 12.2 + 2.3/5.2 2.8 + 3.9/2.3 1.0 + 8/0.5 1.0 + 6.4/0.6 7% 130% AFT + AHe R 02/0.5 0.93 0.59 0.53 0.58 0.63 0.55 0.18 + 0.93/0.18 1.08 0.55 + 6% 116% AHe

0.28 0.12 0.23 0.1 0.28 0.3 0.28 0.08 0.28 0.05 7% 17% AFT + AHe + MTL

0.28 0.21 0.28 0.13 0.28 0.14 0.33 0.19 0.28 0.15 7% 43% AHe

2.1 + 5.5/1.6 6.9 3.8 1.0 + 3.9/0.5 1.4 + 5.9/0.9 1.7 + 3.8/1.2 43% 127% AFT + AHe + MTL

0.38 0.41 0.53 0.32 0.58 0.29 0.58 0.40 0.18 + 0.30/0.18 + 6% 64% AHe

0.33 0.29 0.33 + 0.56/0.33 0.28 0.05 0.28 0.06 0.28 0.01 7% 3% AFT + AHe + MTL

0.28 0.11 0.23 0.12 0.23 0.11 0.33 0.10 0.28 0.11 7% 37% AFT + AHe + MTL

11.9 + 2.6/5.1 6.6 3.0 0.7 + 4.8/0.2 2.1 + 5.3/1.6 3.5 2.3 17% 77% AFT + AHe + MTL

0.08 + 0.37/0.08 0.03 + 0.50/0.03 0.48 0.25 0.33 0.21 0.38 0.22 4% 50% AFT + AHe

0.23 0.12 0.23 0.11 0.28 0.04 0.28 0.08 0.28 0.01 7% 3% AFT + AHe + MTL

0.28 0.13 0.28 0.14 0.28 0.14 0.33 0.12 0.23 0.11 7% 47% AFT + AHe

3.8 + 5.0/3.3 2.8 + 3.7/2.3 0.7 + 3.6/0.2 1.0 + 6.5/0.5 2.1 + 2.6/1.6 30% 87% AFT + AHe + MTL

0.03 + 0.55/0.03 0.03 + 0.56/0.03 0.38 0.24 0.28 0.23 0.28 0.21 24% 48% AFT + AHe

0.23 0.12 0.28 0.15 0.28 0.32 0.28 0.07 0.28 0.03 7% 10% AFT + AHe + MTL

0.28 0.17 0.43 0.18 0.43 0.22 0.43 0.21 0.28 0.2 7% 57% AFT

2.1 + 7.0/1.6 1.7 + 6.4/1.2 1.0 + 3.4/0.5 6.3 3.8 0.7 + 4.71/0.2 67% 113% AFT + AHe

0.18 + 0.44/0.18 0.48 + 0.65/0.48 0.58 0.39 0.38 0.62 0.48 0.37 4% 74% AFT + AHe + MTL

summary, this specic sampling scheme provides reasonably accurate and precise estimates of denudation history (E1 and E2) and accurate but not very well resolved predictions for T and R; however this requires combining at least two different thermochronometers (best estimates for the full AFT + AHe + MTL dataset). Combining valley bottom samples with an elevation prole (Prole + Valleys, Table 2) provides accurate and precise predictions for both E1 and E2 (E1 = 0.230.28 0.010.12 km Myr 1 and E2 = 0.230.33 0.110.14 km Myr 1). T estimates are similar for all inversions and mostly under-predict the timing of change (T = 0.7 3.8 Ma); uncertainties on T are around or above 100%. Similarly to the Valleys sampling scheme, relief change R is strongly overestimated when using AFT or AHe data alone (R = 0.03 + 0.55/0.03). Combining thermochronometers still over-predicts relief change (R = 0.280.38) but uncertainties on R are reduced to ~ 50%. Combining valley bottom samples with an elevation prole thus provides reasonably precise and accurate estimates on E1, E2 and T for nearly all thermochronometer combinations, even though best estimates on inverted parameters, especially on relief change, are obtained when using two or more thermochronometers.

The last inversion simulates sampling along a transect across the massif (Transect, Table 2). This sampling scheme leads to accurate estimates of E1 with uncertainties around 1040% (except for AFT + AHe data, for which the uncertainty rises to ~100%); however, E2 is slightly over-predicted for AHe, AFT + AHe and AFT + MTL datasets (E2 ~0.4 km Myr 1, associated uncertainties ~60%). The timing parameter is also not well resolved (T = 0.76.3 Ma) with prediction uncertainties of 100200%. Finally, relief estimates are quite accurate (R = 0.180.58) but associated uncertainties are between 75 and 130%. In summary, sampling a long transect provides accurate predictions on denudation rates (E1 and E2), however, timing is not well constrained and relief estimates are accurate but associated with high uncertainties. Moreover, our inversions results show that in this sampling conguration, using at least two thermochronometers strongly improves predictions on denudation rates and relief evolution. To better explore the potential differences between the sampling strategies, Fig. 4 presents 1D marginal PDFs of inverted parameters (E1, E2, T and R) for each sampling strategy (using the full dataset). All sampling schemes provide accurate predictions for E1 (Fig. 4a), even though more precise estimates (uncertainties b10%) are obtained

316

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

when sampling a transect, valley bottoms or an elevation prole combined with valley bottoms; uncertainties from sampling along elevation proles are twice as high. The accuracy of E2 estimates is also similar for all sampling schemes, with only the resolution differing (Fig. 4b); lowest uncertainties on E2 predictions are obtained when sampling along valley bottoms (~40% for Valleys and Prole + Valleys), the 3_Proles sampling scheme yields an uncertainty of ~ 50% while the Transect and 1_Prole sampling strategies yield uncertainties of ~ 7090%. Fig. 4c conrms that timing of change (T) is the least resolved parameter, with uncertainties of 8090% for the Valleys and Prole + Valleys sampling schemes, and uncertainties N100% for the other schemes. Finally, the predicted R (Fig. 4d) is reasonably accurate (slightly overestimated) for the 3_Proles, Valleys and Prole + Valleys sampling schemes and is associated with uncertainties of ~5060%. Transect sampling also leads to an accurate estimate for R but the associated uncertainty is high (~75%) whereas 1_Prole does not provide any constraint on R (i.e., it predicts either 50% increasing or decreasing topographic relief). 3.2. Modeling approach In all previous inverse simulations, our modeling approach assumed the geothermal gradient was known. We now present results inverting for both denudation/relief histories (E1, E2, T and R) and thermal parameters (Temp and H_prod). We run inversions for all ve sampling schemes and use the full (AFT + AHe+ MTL) dataset to compare

output predictions with previous outcomes (Table 3). Basal temperature (Temp) is slightly under-predicted for most sampling schemes (Temp ~520565 C; Tempinput = 600 C), except for 3_Proles sampling that leads to an overestimate (Temp = 714 80 C). Associated uncertainties are low and similar for all inversions (~1116%, Table 3). Constraints on the crustal heat production parameter (H_prod, Table 3) are less tight. The Prole + Valleys, 1_Prole and 3_Proles sampling strategies lead to accurate predictions (H_prod = 0.56 0.94; H_prodinput = 0.8) but estimated uncertainties are relatively high (~ 80110%). Valleys sampling clearly does not resolve the heatproduction parameter (H_prod = 0.04 + 0.65/0.04) whereas best predictions for H_prod are provided by the Transect inversion (H_prod = 1.16 0.44). Predictions for E1 are slightly underestimated for the Transect, 3_Proles and 1_Prole sampling schemes (E1 = 0.23 0.1 km Myr 1) and all results provide higher uncertainties on E1 (~30%) compared to inversions that imposed the geothermal gradient (~320%). Tightest constraints are placed on E1 using the Valleys and Prole + Valleys sampling schemes. Results for E2 are variable with best predictions for the 3_Proles inversion (E2 = 0.28 0.14 km Myr 1). E2 is slightly overpredicted in the Transect sampling scheme (E2 = 0.43 0.21 km Myr 1) and under-predicted for Valleys and Prole + Valleys sampling schemes (E2 = 0.23 0.19 km Myr 1). However, uncertainties on E2 are very similar to those obtained when imposing the geothermal gradient (~50 100%). Timing predictions are all underestimated (T ~0.71.7 Ma) and associated uncertainties are similar to those found during imposed
0.16

0.35

0.12 0.25

Probability

Probability

0.15

0.05

Sampling scheme: 1_Profile 3_Profiles Valleys Profile + valleys Transect

0 0.4 0.8 1.2 1.6 2

0.08

0.04

0.4

0.8

1.2

1.6

E1 (km Myr-1)
0.09

E2 (km Myr-1)

c
0.07

0.1

0.08

Probability

0.05

Probability
2 6 10

0.06

0.03

0.04

0.02 0.01 14 0 0 0.4 0.8 1.2 1.6 2

T (Ma)

Fig. 4. 1D posterior parameter PDF's obtained after the NA appraisal stage for the ve different sampling schemes (AFT + AHe + MTL dataset): (a) denudation rate of the rst exhumation phase (E1); (b) denudation rate of the second exhumation phase (E2); (c) transition time between the two exhumation phases (T); and (d) relief factor between the paleo-relief and the nal relief (R). Each line denes an inversion experiment with a sampling scheme (see gure for color legend and Fig. 1 for sampling locations). Black stars represent the input parameter values.

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

317

Table 3 Output estimates after the NA appraisal stage for input parameters E1, E2, T, R, Temp and H_prod. The second line reports parameter ranges and input values for the synthetic thermochronology data (Fig. 1b,c). Optimal parameter values and associated uncertainties (mode 1) are given in the following lines for various sampling schemes (AFT + AHe + MTL dataset). Inversion results are also reported as percentage errors and uncertainties compared to input values (%Input). Parameter (unit) Range/input 1_Prole Mode 1 %Input 3_Proles Mode 1 %Input Valleys Mode 1 %Input Prole + Valleys Mode 1 %Input Transect Mode 1 %Input E1 (km Myr 1) 02/0.3 0.23 0.1 ( 23% 33%) E2 (km Myr 1) 02/0.3 0.33 0.29 (+10% 97%) T (Ma) 0.514.5/3 0.7 + 3.1/0.2 ( 77% 103%) R 02/0.5 1.08 0.52 (+ 116% 104%) Temp (C) 500800/600 549 + 98/49 ( 9% 16%) H_prod 03/0.8 0.64 + 0.82/0.64 ( 20% 103%)

0.23 0.1 ( 23% 33%)

0.28 0.14 ( 7% 47%)

0.7 + 6.9/0.3 ( 77% 230%)

0.48 0.27 ( 4% 54%)

714 80 (+19% 13%)

0.94 0.66 (+18% 83%)

0.33 0.1 (+10% 33%)

0.23 0.19 ( 23% 63%)

1.7 + 2.0/1.2 ( 43% 67%)

0.33 0.21 ( 34% 42%)

519 + 74/19 ( 14% 12%)

0.04 + 0.65/0.04 ( 95% 81%)

0.28 0.1 ( 7% 33%)

0.23 0.19 ( 23% 63%)

0.7 + 5.1/0.3 ( 77% 170%)

0.48 0.26 ( 4% 52%)

534 + 95/34 ( 11% 16%)

0.56 + 0.85/0.56 ( 30% 106%)

0.23 0.09 ( 23% 30%)

0.43 0.21 (+43% 70%)

1.4 + 4.4/0.9 ( 53% 147%)

0.48 0.4 ( 4% 80%)

564 66 ( 6% 11%)

1.16 0.44 (+45% 55%)

geothermal gradient inversions (~70230%). Moreover, best predictions for T are provided by the Valleys inversion (T = 1.7 +2.0/1.2 Ma), in agreement with previous results. Finally, setting no constraint on the geothermal gradient does not have any effect on relief evolution predictions as R estimates are very close to values obtained with an imposed geothermal gradient, and optimal predictions are obtained for 3_Proles, Valleys and Prole + Valleys sampling schemes. To summarize, numerical inversions without imposing constraints on the geothermal gradient provide less accurate and precise estimates for exhumation rates (E1 and E2), but output results for timing and relief change (T and R) are very close to those found with an imposed geothermal gradient. Best estimates for thermal parameters (H_prod and Temp) are given by Transect sampling, while inversions including at least one elevation prole (Prole + Valleys, 1_Prole and 3_Proles) provide accurate estimates but associated with relatively high uncertainties, and valley bottom sampling (Valleys) does not resolve thermal parameters. 4. Discussion 4.1. Limits to the numerical modeling approach Before interpreting our results, we should highlight potential shortcomings associated with our numerical modeling approach. First of all, the exhumation scenario we choose is fairly simple, with only two exhumation phases, a constant and spatially uniform regional denudation rate and a single episode of relief growth. This choice could be considered specic and restrictive, although we consciously chose uniform background exhumation within the study area to focus on relief evolution. A similar inverse approach could, however, be developed to specically explore spatially varying exhumation (e.g., Stephenson et al., 2006a). We have tested one specic geothermal gradient that we also assume to be spatially uniform, as we apply our inversions to a region of spatially relatively homogeneous lithology. However, we note that local uid circulations along faults can induce large spatial variations in thermal gradient (Whipp and Ehlers, 2007). Despite several recent attempts (e.g., Ehlers et al., 2006; Glotzbach et al., 2008, 2010; van der Beek et al., 2010; Vernon et al., 2009), quantifying both timing and amplitude of topographic relief changes in mountain belts using low-temperature thermochronology has proven elusive, motivating the focus of our study on this specic

exhumation scenario. Although our study design might be somewhat restrictive, we have consciously chosen quite a general synthetic study regarding the exhumation history and the near-surface thermal structure. The relief evolution scenario should also be quite general, as moderate relief increase since the mid-Pliocene is expected or has been encountered in many mountain belts (e.g., Molnar and England, 1990; Zhang et al., 2001). We thus are condent that our inferences regarding thermochronology sampling strategies for quantifying both exhumation and relief histories, despite the above limitations, might be of more general interest for studying at least the late-Neogene exhumation history of mountain belts. Second, the way we dene relief changes in Pecube does not capture the complexities of landscape evolution processes. Some limitations include the constant planform drainage pattern assumed in our numerical approach, and a geometric approach to relief development (Fig. 2a). An alternative approach would be to combine Pecube with a landscape evolution model to fully simulate landscape changes with surface process modeling (Herman et al., 2010b). However, our synthetic approach allows us to address at rst order the effect of transient topography on thermochronology data and their intrinsic resolution in quantitatively retrieving potential relief changes. We nally note that the different sampling scenarios explored here, although they represent the most commonly used and simplest ways to sample a given topography, might not be the most adequate to independently quantify exhumation and relief histories, i.e., we do not perform a complete experimental design study.

4.2. Parameter resolution and tradeoffs Our numerical results show that some parameters (mainly T and R, Fig. 4) have been poorly constrained by the NA inversion. The lack of convergence toward the input parameters may result from the poor resolution of thermochronology data to constrain timing or relief parameters (Valla et al., 2010), from some random elements, or from potential tradeoffs between model parameters. To illustrate this alternative explanation, Fig. 5 shows 2D scatter plots and PDFs for a single elevation prole (1_Prole). There is no apparent tradeoff between the initial and nal denudation rates (E1 and E2, Fig. 5a), but signicant tradeoffs appear between the recent denudation rate (E2) and both the relief change (R, Fig. 5b) and the timing of change (T,

318

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

Fig. 5c). These tradeoffs may explain the relatively poor resolution observed in the 1D marginal PDFs for these parameters (Fig. 4c,d). The potential parameters tradeoffs for the other sampling schemes are shown in Figs. 6 and 7. Fig. 6 reveals that the tradeoff between E2 and R is reduced for the 3_Proles, Valleys, and Prole + Valleys schemes (Fig. 6ac), but conrms their tendency to over-estimate the relief increase. In contrast, there is an important tradeoff in the Transect sampling scheme (Fig. 6d), which explains the high uncertainties associated with R. Fig. 7 conrms that the timing parameter T is poorly constrained by our inverse approach; 2D scatter plots reveal a signicant tradeoff between E2 and T, even though the covariance appears to be slightly reduced for sampling schemes Valleys and Prole + Valleys (Fig. 7b,c). Interestingly, both Figs. 6 and 7 suggest that parameter tradeoffs, similar to 1D marginal PDFs, are highly dependent on the sampling scheme and thus might also be considered when designing sampling strategies.
2

4.3. Sampling strategy Inversion results for a single elevation prole (1_Prole) show that, although such a sampling approach allows accurately estimating denudation rates, it does not constrain relief evolution satisfactorily (uncertainties N100%, Table 2). This nding is similar to the conclusion of Valla et al. (2010), who also performed a sensitivity analysis and showed that relief evolution can only be precisely predicted from thermochronological age-elevation proles if the rate of relief change is at least two to three times higher than the regional background denudation rate. In our case, the regional denudation rate is 0.3 km Myr 1 and the relief scenario induces a maximum reliefgrowth rate of ~ 0.5 km Myr 1 for the last 3 Myr (i.e., ~ 60% higher than the background denudation rate); under this conguration, our inverse results are in good agreement with Valla et al. (2010). We suggest that the inability of a single age-elevation prole to constrain
2

a
1.5 1.5

E2 (km Myr-1)

0.5

0.5

0 0 0.5 1 1.5 2

0 0 0.5 1 1.5 2

E1 (km Myr-1)
14

E2 (km Myr-1)

10

T (Ma)

20

40

60

80

100

0.4

0.8

1.2

1.6

E2 (km Myr-1)
Fig. 5. Scatter plots (each dot represents a forward model color coded by its mist value ) and 2D posterior marginal probability density functions with 1 condence contours (black lines) for the sampling scheme 1_Prole: a) denudation rates of the rst (E1) vs the second (E2) exhumation phases; b) denudation rate of the second exhumation phase (E2) vs relief factor (R); and c) denudation rate of the second exhumation phase (E2) vs transition time between the two exhumation phases (T). Black stars represent the input parameter values.

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

319

a
1.5 1.5

0.5

0.5

3_Profiles
0 0 0.5 1 1.5 2 0 0 0.5 1

Valleys
1.5 2

E2 (km Myr-1)
2 2

E2 (km Myr-1)

c
1.5 1.5

R 1

0.5

0.5

0 0 0.5 1

Profile+Valleys
1.5 2

0 0 0.5 1

Transect
1.5 2

E2 (km Myr-1)

E2 (km Myr-1)

Fig. 6. Scatter plots (see Fig. 5 for legend) and 2D posterior marginal probability density functions with 1 condence contours (black lines) of denudation rate of the second exhumation phase (E2) vs relief factor (R) for different sampling schemes: a) 3_Proles; b) Valleys; c) Prole + Valleys; and d) Transect. Black stars represent the input parameter values.

relief evolution may come from the insufcient spatial coverage of this specic sampling scheme, i.e., a single elevation prole does not fully capture the topographic control on near-surface isotherms (Braun, 2002a,b) and results in high tradeoffs between model parameters (Fig. 5). Sampling along three different elevation proles provides similar results for denudation rates, but estimates of relief evolution are better constrained (uncertainties ~ 65%) and all predict a relief increase (R b 1, Table 2). Sampling along three different valleys allows capturing different topographic wavelengths, thus reducing parameter tradeoffs (Figs. 6 and 7). As explained previously by Braun (2002a), crustal isotherms are inuenced by different topographic wavelengths and this effect can be used to constrain topographic changes using either a spectral method along a long transect (Braun, 2002b) or, as in this case, by combining several elevation proles along different valleys. This effect is also captured when sampling along several valley bottoms (Valleys and Prole + Valleys), which

explains why optimal values for both denudation rate, timing and relief increase are all more accurate and precise (uncertainties on E1, T and R are 2 to 6 times lower than for 1_Prole inversion results). An additional explanation for the better predictions from valley-bottom samples might be that valley bottoms have experienced more recent exhumation than ridges and are characterized by younger mean ages, which may result in better resolution for quantitatively constraining model parameters such as R and T. Thermochronological data collected along a transect can be interpreted by spectral analysis (Braun, 2002b) to constrain both denudation rates on short topographic wavelengths and relief evolution on long topographic wavelengths. However, this method considers temporally constant denudation rates and neither allows determining precisely the time of relief change nor estimating the uncertainties on optimal values for denudation and relief change. Inversion results using Transect sampling (Table 2) predict denudation and relief histories close to 3_Proles predictions, i.e., accurate estimates on denudation rates (E1 and

320

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

14

14

10

10

T (Ma)
6 6 2 2

T (Ma)

3_Profiles
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2

Valleys
1.6 2

E2 (km Myr-1)
14

E2 (km Myr-1)

14

10

10

T (Ma)

T (Ma)
6 6 2 2

Profile+Valleys
0 0.4 0.8 1.2 1.6 2 0 0.4 0.8 1.2

Transect
1.6 2

E2 (km Myr-1)

E2 (km Myr-1)

Fig. 7. Scatter plots (see Fig. 5 for legend) and 2D posterior marginal probability density functions with 1 condence contours (black lines) of denudation rate of the second exhumation phase (E2) vs transition time between the two exhumation phases (T) for different sampling schemes: a) 3_Proles; b) Valleys; c) Prole + Valleys; and d) Transect. Black stars represent the input parameter values.

E2) and relief change (R); however, the uncertainties on these parameters are higher than those obtained with Valleys and Prole + Valleys sampling strategies, due to higher parameters tradeoffs (Figs. 6 and 7). These outcomes are consistent with recent analysis of AFT data from a transect across the Pelvoux massif (Beucher et al., submitted for publication): these data constrain exhumation rates well but provide little insight into relief change. These differences appear to result from the spatial conguration of the Transect sampling scheme. Although it covers a larger area than the 1_Prole and 3_Proles sampling schemes, Transect data only sample one spatial direction (Fig. 2a) and may not fully record the 2D topographic perturbations of isotherms compared to data

that are spatially distributed along different valleys (Valleys or Prole + Valleys sampling scheme). Combining a long transect with at least one elevation prole would have allowed to best capture both short and long topographic wavelength effects on the thermal structure (Braun, 2002a, b), simultaneously providing accurate and precise information on both denudation and relief predictions. 4.4. Modeling approach Inversions for which the thermal parameters are supposed unknown lead to comparable accuracy but somewhat higher uncertainties for

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322

321

denudation and relief predictions than when imposing a geothermal gradient (Table 3). These differences in predicted uncertainties highlight the classic problem of determining an appropriate dimensionality for a numerical problem, i.e., the number of inverted parameters compared to the intrinsic data resolution (see Gallagher et al., 2005, for a discussion on this subject). Moreover, our results show that sampling along valley bottoms does not allow resolving accurately the thermal parameters. Sampling along elevation proles helps constraining the geothermal gradient, as suggested by Gallagher et al. (2005), pointing out that even though numerical modeling allows us to infer exhumation histories from complex spatial sample distributions, age-elevation relationships remain useful for interpreting thermochronology data. The ability of age-elevation proles to constrain both the exhumation history and thermal parameters comes from the altitudinal distribution of samples; the slope of the age-elevation relationship providing an estimate of the apparent exhumation rate (e.g., Hurford, 1991), while the age-window of the dataset may help to constrain the near-surface geothermal gradient (Gallagher et al., 2005). Surprisingly, the Transect sampling provided the best estimates on thermal parameters, which can be explained by the simultaneous spatial and altitudinal data coverage allowed by this sampling scheme (Fig. 2a). It thus might be used not only for spectral analysis (Braun, 2002b) but also, like elevation proles, to constrain the geothermal gradient. Finally, we have tested the effect of the grid resolution on the predicted estimates of the various parameters. As reported in the supplementary material section, we show that a minimum grid spacing of 1 km (or ~ 1/10th of the valley spacing) is necessary to properly resolve exhumation rates and relief change in the typical mountainous terrain we have used here. 5. Conclusions Our study was initially designed to determine to which extent different sampling and modeling strategies could allow constraining relief evolution from low-temperature thermochronology datasets. Our results conrm the ndings of Valla et al. (2010), who suggested that age-elevation relationships do not resolve relief changes if these do not induce an additional spatially variable exhumation rate that is at least two times higher than the regional background denudation rate. Although our approach is slightly different from Valla et al. (2010), we have shown here that age-elevation relationships do not allow constraining relief changes. We have demonstrated that spatially distributed sampling along a transect, several elevation proles or along valley bottoms, preferably combined with an elevation prole, leads to better predictions of relief evolution. The main drawback of only sampling along valley bottoms is the loss of information on the near-surface thermal structure. However, using higher-temperature thermochronometers may provide supplementary constraints on thermal parameters. Glotzbach et al. (2011) recently used a dense thermochronology dataset combining both lowand high-temperature thermochronometers along elevation proles and a tunnel transect to constrain both the near-surface thermal structure and the exhumation history of the Mont-Blanc massif (western European Alps). Their modeling results resolved a recent relief increase whereas other recent studies using only data along a single age-elevation prole were not able to resolve the recent relief evolution (van der Beek et al., 2010; Vernon et al., 2009). Based on synthetic data produced under an imposed exhumation and relief scenario, we have tested different sampling strategies to explore how to best constrain denudation rates, timing and relief change. Our results show that dense spatial sampling combining valley bottom samples and elevation proles, as well as long transects may help to provide quantitative constraints on complex exhumation histories. Moreover, we have highlighted that the modeling approach (choice of thermal parameters or model resolution) potentially inuences predicted denudation and relief histories. Although age-elevation proles may be a

useful and direct way to interpret thermochronology data, sampling strategies should be rethought in the light of recent numerical modeling developments. For a region where relief change is expected to have been minimal but the emphasis is on deciphering changes in denudation rate through time, an appropriate sampling scheme would be using one or several elevation proles. However, when one has reason to believe that exhumation rates may have been relatively constant (e.g., in stable post-orogenic mountain belts) but the aim is to constrain the timing and amount of relief change, our study would suggest a better choice is to sample valley bottoms, possibly combined with an elevation prole. Therefore, ideally the sampling strategy should be designed a priori, using preliminary modeling of expected denudation and relief histories and preferably using multiple working hypotheses; and a sensitivity study on assumed thermal parameters and model resolution should be performed when modeling the data.

Acknowledgments This study is supported by the INSU-CNRS through the European Science Foundation Topo-Europe program 07-TOPO-EUROPE-FP-023 Coupled climatic/tectonic forcing of European topography revealed through thermochronometry (Thermo-Europe) and the Agence Nationale de la Recherche project No. ANR-08-BLAN-0303-01 Erosion and Relief Development in the Western Alps. It forms part of PV's PhD project at Universit Joseph Fourier, supported by the French Ministry for Research and Higher Education. We thank Malcolm Sambridge and two anonymous reviewers for insightful comments that helped to improve the clarity of this manuscript. Computations were performed on Brutus, the high performance computing facilities at ETH Zurich. The codes are available at http://svn-geo.ethz.ch after registering at this site.

Appendix A. Supplementary data Supplementary data to this article can be found online at doi:10.1016/j.epsl.2011.05.003.

References
Beucher, R., van der Beek, P.A., Braun, J. and Batt, G. (submitted for publication). Assessing exhumation and relief development in the Pelvoux and Dora-Maira massifs (western Alps) from spectral analysis and inversion of thermochronological age transects. Journal of Geophysical Research. Braun, J., 2002a. Quantifying the effect of recent relief changes on age-elevation relationships. Earth Planet. Sci. Lett. 200, 331343. Braun, J., 2002b. Estimating exhumation rate and relief evolution by spectral analysis of age-elevation proles. Terra Nova 14, 210214. Braun, J., 2003. Pecube: a new nite element code to solve the heat transport equation in three dimensions in the Earth's crust including the effects of a time-varying, nite amplitude surface topography. Comput. Geosci. 29, 787794. Braun, J., 2005. Quantitative constraints on the rate of landform evolution derived from low-temperature thermochronology. Rev. Mineral. Geochem. 58, 351374. Braun, J., van der Beek, P.A., 2004. Evolution of passive margin escarpments: what can we learn from low-temperature thermochronology? J. Geophys. Res. 109, F04009. doi:10.1029/02004JF000147. Braun, J., van der Beek, P.A., Batt, G., 2006. Quantitative Thermochronology: Numerical Methods for the Interpretation of Thermochronological Data. Cambridge University Press, Cambridge, New York. 258 pp. Braun, J., van der Beek, P.A., Valla, P.G., Robert, X., Herman, F., Glotzbach, C., Pedersen, V., Perry, C., Simon-Labric, T. and Prigent, C. (submitted for publication). Quantifying rates of landscape evolution and tectonic processes by thermochronology and numerical modeling of heat transport in the crust using PECUBE. Tectonophysics. Clark, M.K., Maheo, G., Saleeby, J., Farley, K.A., 2005. The non-equilibrium landscape of the southern Sierra Nevada, California. GSA Today 15, 410. Curtis, A., 2004a. Theory of model-based geophysical survey and experimental design part Alinear problems. Lead. Edge 23, 9971004. Curtis, A., 2004b. Theory of model-based geophysical survey and experimental design part Bnonlinear problems. Lead. Edge 23, 11121117. Delunel, R., van der Beek, P.A., Carcaillet, J., Bourls, D.L., Valla, P.G., 2010. Frost-cracking control on catchment denudation rates: Insights from in situ produced 10Be concentrations in stream sediments (EcrinsPelvoux massif, French Western Alps). Earth Planet. Sci. Lett. 293, 7283.

322

P.G. Valla et al. / Earth and Planetary Science Letters 307 (2011) 309322 House, M.A., Wernicke, B.P., Farley, K.A., 1998. Dating topography of the Sierra Nevada, California, using apatite (U-Th)/He ages. Nature 396, 6669. Hurford, A.J., 1991. Uplift and cooling pathways derived from ssion track analysis and mica dating: a review. Geol. Rundsch. 80, 349368. Ketcham, R.A., 2005. Forward and inverse modeling of low-temperature thermochronometry data. Rev. Mineral. Geochem. 58, 275314. Mancktelow, N.S., Grasemann, B., 1997. Time-dependent effects of heat advection and topography on cooling histories during erosion. Tectonophysics 270, 167195. Molnar, P., England, P., 1990. Late Cenozoic uplift of mountain ranges and global climate change: chicken or egg? Nature 346, 2934. Reinecker, J., Danik, M., Schmid, C., Glotzbach, C., Rahn, M., Frisch, W., Spiegel, C., 2008. Tectonic control on the late stage exhumation of the Aar Massif (Switzerland): constraints from apatite ssion track and (U-Th)/He data. Tectonics 27, TC6009. doi:10.1029/2007TC002247. Reiners, P.W., 2007. Thermochronologic approaches to paleotopography. Rev. Mineral. Geochem. 66, 243267. Reiners, P.W., Brandon, M.T., 2006. Using thermochronology to understand orogenic erosion. Annu. Rev. Earth Planet. Sci. 34, 419466. Sambridge, M., 1999a. Geophysical inversion with a neighbourhood algorithmI. Searching a parameter space. Geophys. J. Int. 138, 479494. Sambridge, M., 1999b. Geophysical inversion with a neighbourhood algorithmII. Appraising the ensemble. Geophys. J. Int. 138, 727746. Stephenson, J., Gallagher, K., Holmes, C.C., 2006a. Low temperature thermochronology and strategies for multiple samples: 2: partition modelling for 2D/3D distributions with discontinuities. Earth Planet. Sci. Lett. 241, 557570. Stephenson, J., Gallagher, K., Holmes, C.C., 2006b. A Bayesian approach to calibrating apatite ssion track annealing models for laboratory and geological timescales. Geochim. Cosmochim. Acta 70, 51835200. Stwe, K., White, L., Brown, R.W., 1994. The inuence of eroding topography on steady-state isotherms. Application to ssion track analysis. Earth Planet. Sci. Lett. 124, 6374. Valla, P.G., Herman, F., van der Beek, P.A., Braun, J., 2010. Inversion of thermochronological age-elevation proles to extract independent estimates of denudation and relief historyI: theory and conceptual model. Earth Planet. Sci. Lett. 295, 511522. van der Beek, P.A., Valla, P.G., Herman, F., Braun, J., Persano, C., Dobson, K.J., Labrin, E., 2010. Inversion of thermochronological age-elevation proles to extract independent estimates of denudation and relief historyII: application to the French Western Alps. Earth Planet. Sci. Lett. 296, 922. Vernon, A.J., van der Beek, P.A., Sinclair, H.D., Persano, C., Foeken, J., Stuart, F.M., 2009. Variable late Neogene exhumation of the central European Alps: Low-temperature thermochronology from the Aar Massif, Switzerland, and the Lepontine Dome, Italy. Tectonics 28, TC5004. doi:10.1029/2008TC002387. Wagner, G.A., Reimer, G.M., 1972. Fission track tectonics: the tectonic interpretation of ssion track apatite ages. Earth Planet. Sci. Lett. 14, 263268. Whipp, D.M., Ehlers, T.A., 2007. Inuence of groundwater ow on thermochronometerderived exhumation rates in the central Nepalese Himalaya. Geology 35, 851854. Wood, R., Curtis, A., 2004. Geological prior information, and its applications to solving geoscientic problems. In: Curtis, A., Wood, R. (Eds.), Geological Prior Information: Geological Society of London Special Publication, 239, pp. 114. Zhang, P., Molnar, P., Downs, W.R., 2001. Increased sedimentation rates and grain sizes 24 Myr ago due to the inuence of climate change on erosion rates. Nature 410, 891897.

Dumont, T., Champagnac, J.-D., Crouzet, C., Rochat, P., 2008. Multistage Alpine shortening in Central Dauphin (French Western Alps): implications for pre-Alpine restoration. Swiss J. Geosci. 101, 89110. Ehlers, T.A., Farley, K.A., 2003. Apatite (U-Th)/He thermochronometry: methods and applications to problems in tectonic and surface processes. Earth Planet. Sci. Lett. 206 (12), 114. Ehlers, T.A., Farley, K.A., Rusmore, M.E., Woodsworth, G.J., 2006. Apatite (U-Th)/He signal of large-magnitude accelerated glacial erosion, southwest British Columbia. Geology 34, 765768. Farley, K.A., 2000. Helium diffusion from apatite: general behavior as illustrated by Durango uorapatite. J. Geophys. Res. 105, 29032914. Farley, K.A., 2002. (U-Th)/He dating: techniques, calibrations, and applications. In: Porcelli, D., Ballentine, C.J., Wieler, R. (Eds.), Reviews in Mineralogy and Geochemistry: Noble Gases in Geochemistry and Cosmochemistry, 47. Mineralogical Society of America, Washington, DC, pp. 819844. 2002. Fitzgerald, P.G., Sorkhabi, R.B., Redeld, T.F., Stump, E., 1995. Uplift and denudation of the central Alaska Range: a case study in the use of apatite ssion track thermochronology to determine absolute uplift parameters. J. Geophys. Res. 100, 2017520191. Flowers, R.M., Ketcham, R.A., Shuster, D.L., Farley, K.A., 2009. Apatite (U-Th)/He thermochronometry using a radiation damage accumulation and annealing model. Geochim. Cosmochim. Acta 73, 23472365. Gallagher, K., Brown, R., Johnson, C., 1998. Fission track analysis and its applications to geological problems. Annu. Rev. Earth Planet. Sci. 26, 519572. Gallagher, K., Stephenson, J., Brown, R., Holmes, C., Fitzgerald, P., 2005. Low temperature thermochronology and modeling strategies for multiple samples 1: vertical proles. Earth Planet. Sci. Lett. 237 (12), 193208. Gautheron, C., Tassan-Got, L., Barbarand, J., Pagel, M., 2009. Effect of alpha-damage annealing on apatite (U-Th)/He thermochronology. Chem. Geol. 266, 157170. Glotzbach, C., Reinecker, J., Danisik, M., Rahn, M., Frisch, W., Spiegel, C., 2008. Neogene exhumation history of the Mont Blanc massif, western Alps. Tectonics 27, TC4011. doi:10.1029/2008TC002257. Glotzbach, C., Reinecker, J., Danisik, M., Rahn, M., Frisch, W., Spiegel, C., 2010. Thermal history of the central Gotthard and Aar massifs (European Alps): evidence for steady state, long-term exhumation. J. Geophys. Res. doi:10.1029/ 2009JF001304. Glotzbach, C., van der Beek, P.A., Spiegel, C., 2011. Episodic exhumation of the Mont Blanc massif, Western Alps: constraints from numerical modelling of thermochronology data. Earth Planet. Sci. Lett. 304, 417430. doi:10.1016/j.epsl.2011.02.020. Green, P.F., Duddy, I.R., Laslett, G.M., Hegarty, K.A., Gleadow, A.J.W., Lovering, J.F., 1989. Thermal annealing of ssion tracks in apatite IV: quantitative modelling techniques and extension to geological timescales. Chem. Geol. 79, 155182. Herman, F., Braun, J., Dunlap, W., 2007. Tectono-morphic scenarios in the Southern Alps of New Zealand. J. Geophys. Res. 112, B04201. doi:10.1029/2004JB003. Herman, F., Rhodes, E.J., Braun, J., Heiniger, L., 2010a. Uniform erosion rates and relief amplitude during glacial cycles in the Southern Alps of New Zealand, as revealed from OSL-thermochronology. Earth Planet. Sci. Lett. 297, 183189. Herman, F., Copeland, P., Avouac, J.-P., Bollinger, L., Maho, G., Le Fort, P., Rai, S., Foster, D., Pecher, A., Stuwe, K., Henry, P., 2010b. Exhumation, crustal deformation and thermal structure of the Nepal Himalaya derived from the inversion of thermochronological and thermobarometric data and modeling of the topography. J. Geophys. Res. 115, B06407. doi:10.1029/2008JB006126.