Earth and Planetary Science Letters 287 (2009) 373384

Contents lists available at ScienceDirect

Earth and Planetary Science Letters

j ourna l homepage: www.e lsev ie r.com/ locate /eps lThe strength of faults in the crust in the western United States

Sara Carena , Christoph ModerDepartment of Earth and Environmental Sciences, University of Munich, Theresienstr. 41, 80333 Munich, Germany Corresponding author. Tel.: +49 89 21806513; fax:E-mail addresses: scarena@iaag.geo.uni-muenchen.d

moder@geophysik.uni-muenchen.de (C. Moder).

0012-821X/$ see front matter 2009 Elsevier B.V. Adoi:10.1016/j.epsl.2009.08.021a b s t r a c ta r t i c l e i n f oArticle history:Received 1 April 2009Received in revised form 12 August 2009Accepted 13 August 2009Available online 18 September 2009

Edited by R.D. van der Hilst

Keywords:fault strengthfrictionlithospherefinite element modelThe strength of secondary faults within plate-boundary zones and of master faults like the San Andreas hasbeen controversial for decades. We use a global finite-element code with a variable-resolution grid andglobal plate-driving forces to determine whether the effective friction on the San Andreas fault is high(*0.61), intermediate (*0.30.5) or low (* 0.2), whether a single value of * can be used for allmapped faults within California, and whether weakening of the ductile lower crust associated with faulting isimportant. We compare our model results with existing data on fault slip-rates, GPS velocities, stress field,and earthquake depth distribution. The comparison indicates that all faults are weak (*0.2), and thatadditional weakening of major faults is important. All viable solutions also indicate that weakening of thelower crust below major faults is necessary. The strongest faults in the region have * in the range 0.20.05.The San Andreas fault is a very weak fault among weak faults, with *

Fig. 1. Global grid showing temperature at 100 km depth as deviation from the averagecalculated at that depth, and MCM velocity vectors (blue arrows) at 100 km depth.Faults and plate boundaries are in yellow. PA=Pacific, NA=North America,CO=Cocos, CA=Caribbean, NZ=Nazca, SA=South America. (For interpretation ofthe references to color in this figure legend, the reader is referred to the web version ofthis article.)

Table 1Parameters used in SHELLS calculations.

Parameter Value

Continuum friction coefficient 0.85Crust mean density 2816 kg m3

Mantle mean density 3332 kg m3

Asthenosphere density 3125 kg m3

Water density 1032 kg m3

Biot coefficient 1.00Gravitational acceleration 9.8 ms2

Surface temperature 273 KCrust upper temperature limit 1223 KMantle-lithosphere upper temperature limit 1673 KCrust thermal conductivity 2.7 Jm1s1 K1

Mantle thermal conductivity 3.2 Jm1s1K1

Crust radioactive heat production (volume) 7.27107 J m3s1

Intercept of upper mantle adiabat 1412 KSlope of upper mantle adiabat 6.1104 K m1

Volumetric thermal expansion coefficient (crust) 2.4105 K1

Volumetric thermal expansion coefficient (mantle) 3.94105 K1

Exponent on strain-rate in creep-stress laws (1/n) 0.333333Temperature coefficient of creep rheology (crust) 4000 KTemperature coefficient of creep rheology (mantle) 18,314 KShear stress coefficient of creep law (crust) 2.3109 Pa s1/n

Shear stress coefficient of creep law (mantle) 9.5104 Pa s1/n

Plate defining velocity reference frame North AmericaNumber of grid nodes 4761Number of grid elements 6996Number of fault elements 1363

Most of these parameters are defined and described in detail in Bird (1989).

374 S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384Recent works by Brgmann and Dresen (2008) and Thatcher andPollitz (2008) suggest that the brittle strength of faults in the uppercrust cannot be studied independently of their strength in the ductilelower crust. Evidence of weakening in the lower crust below faultswould also contribute to settling a long-standing debate as to whethermajor faults exist at lithospheric scale (e.g. Tapponnier et al., 1986,2001; Thatcher, 1995; Jackson, 2002), or are confined to the brittlecrust (e.g. England and Houseman, 1985; Houseman and England,1986, 1993). We therefore test for fault behavior in both the brittleand the ductile zones simultaneously.

We focus on the San Andreas fault and its surrounding faultnetwork in western California because of the large amount and goodquality of data available, and because the numerous studies that havealready been carried out help us constrain at least some parameters.We also include in our model the neighboring areas of easternCalifornia and western Nevada, in order to provide a smoothtransition between the global and local grid of our model.

2. Models

We use SHELLS (Kong and Bird, 1995; Bird, 1999), a global finiteelement thin-shell code, for modeling. We opted for a global finiteelement code instead of a flat-Earth one in order to avoid unphysicalboundary conditions at the edges of our area of interest. A global codehas never been used before to model the behavior of a regional area insuch detail, therefore our work is as much an investigation of faultstrength as it is a feasibility test for this type of models.

In SHELLS, elastic strain is neglected and only permanent strain isconsidered, therefore the results of SHELLS calculations correspond toan average over several seismic cycles. The rheology of the model isthus anelastic everywhere, and the active deformation mechanismsare either frictional sliding along faults or nonlinear dislocation creep.In dislocation creep, (strain-rate) relates to stress () and creepactivation energy (Q) through the power law equation An exp(Q/RT), where A is the shear stress coefficient, R the gas constant,and T is temperature. In the model frictional sliding and dislocationcreep compete at faults: friction dominates in the upper part, wherethe normal force is small, while dislocation creep dominates at depth,where the temperature is high enough to achieve substantial slip-rates. The depth of the transition between the two for each faultelement (brittleductile transition depth) is calculated by assumingthat the rheology resulting in the lowest shear stress will prevail.

The only boundary conditions that need to be specified are theglobal plate-driving forces, for which we tested both NUVEL-1Avelocities, and velocities derived from global mantle circulationmodeling (Bunge et al., 1998, 2002) applied at the bottom of theplates everywhere in our model. The actual depth at which velocitiesare applied does not significantly affect results, as long as thevelocities are not applied within the crust itself. We adopt a recenthigh-resolution mantle circulation model (Fig. 1, Schuberth et al.,2009) that provides sufficient spatial resolution to resolve thevigorous convective regime of the mantle. In addition to representingthe dynamic effects from a mechanically weak asthenosphere onmantle flow (Richards et al., 2001), the model incorporates internalheat generation from radioactivity together with a significant amountof heat flow from the core, for which there is growing evidence(Bunge, 2005; van der Hilst et al., 2007). Combined with constraintson the history of subduction (Engebretson et al., 1984; Richards andEngebretson, 1992) this allows us to place first-order estimates on theinternal mantle buoyancy forces that drive plates.

We assume we have a good knowledge of the 3-D fault geometry,and then solve for fault strength. In particular, we examine the effectof effective fault friction (*) in the upper brittle part of the crust andof creep activation energy (Q) in the ductile lower crust. SHELLScalculates forces and velocities, which we use to compute parametersthat can be directly compared with data (fault slip-rates, GPSvelocities, earthquakes depth distribution, and stress directions) inorder to score our model and determine fault strength. Input for themodel is provided as information about fault geometry, topographyand heat flow, stored in the finite element grid, and as parameters likefriction coefficients and creep activation energies for crust andmantle(a list of parameters and values used is provided in Table 1). Such

Fig. 2. Variable-resolution grid, global view. The largest elements at global level havesides of 2103 km away from plate boundaries, and 1103 km at plate boundaries.NA=North American plate, PA=Pacific plate, CO=Cocos plate, NZ=Nazca plate,CA=Caribbean plate. Plate boundaries from Bird (1999).

375S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384models have already been shown to produce realistic results at globalscale (e.g. Bird and Liu, 1999; Iaffaldano and Bunge, 2009) andtherefore we chose our set of global parameters to match publishedFig. 3. Local, high-resolution grid, with topography from ETOPO2 (National Geophysical DaWH=WasatchHurricane fault system, SAF=San Andreas fault.models. In this paper we limit ourselves to describing the localmodifications to strength-controlling parameters, with the under-standing that global parameters remain unchanged throughout theentire set of simulations.

SHELLS uses a grid of spherical triangles for the continuumelements, and arcs of great circle for the fault elements (Kong andBird, 1995). To drive the local model as part of a global one, weconstructed a variable-resolution grid (Fig. 2) that gradually transi-tions from a low-resolution global grid to a high-resolution local onecovering California and western Nevada (Fig. 3). In order to positionthe fault elements as accurately as possible, we built the grid usingGocad (Mallet, 2002), which allows us to incorporate the faultsdirectly into the local grid and to optimize the mesh for triangleequilaterality and smooth transition between local and global grids,while keeping the fault elements fixed in 3-D space. The distancebetween nodes in the local part of the grid varies between about 5 kmand 50 km. This distance is based on the level of complexity of thefault geometry in a given area, rather than on the resolution of thetopography and heat flow data: complex fault junctions and closely-spaced faults need a higher density of nodes, because they cannototherwise be separated and represented in the grid. At every point thenode spacing in our grid is at least the minimum required to separatethe faults that we want to represent. Tests with a grid finer than thatdid not produce any significant changes in the results. Removingminor faults from the grid also did not alter significantly the slip-ratesof remaining faults, it simply resulted inmore distributed deformationof the continuum in between them.

In addition to defining grid geometry, we must also define thevalues of topography and heat flow at each grid node (to computethickness of crust and lithospheric mantle), and the dip of each faultelement. Topography is linearly interpolated onto our grid from theta Center, 2006) and fault elements (thick black lines). M=Mendocino triple junction,

Fig. 4. Perspective view of the SCEC Community Fault Model, simplified version thatrepresents fault segments as rectangular patches (Plesch et al., 2007). SAF=SanAndreas fault, SJFZ=San Jacinto fault zone.

376 S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384ETOPO2 data set (Fig. 3; National Geophysical Data Center, 2006). Wegenerated a surface heat flow map for California and western Nevadaby merging the heat flow data of the U.S. Geological Survey (2007)and of the Southern Methodist University (2007) heat flow databases.The remainder of the grid is assigned heat flow values based on theGlobal Heat Flow Database of the International Heat Flow Commission(2007). Crustal thickness is then calculated from topography and heatflow assuming an isostatically compensated crust and using the Airycompensation (Bird, 1999). Where necessary, we locally correct heatflow values so that the calculated crustal thickness matches thecrustal thickness map of Fuis and Mooney (1990).

Only active faults, here defined as faults that show signs ofQuaternary activity, are included in the model. The assumption ofwell-known fault geometry for active faults is a reasonable one inmost of California and at our current resolution. In the past 510yrs,efforts towards building reliable fault models have resulted in twomajor 3-D fault geometry databases available for California: the SCECCommunity Fault Model (Fig. 4) (SCEC CFM; Plesch et al., 2007) andthe U.S. Geological Survey 3-D Bay Area Geologic Model (Graymeret al., 2005; Horsman et al., 2008; Jachens et al., 2009). The geometryof the faults at the margin of the area of interest (northeasternCalifornia and western Nevada) is less constrained. For most of thesefaults, we obtained the strike and dip information from the USGSQuaternary Fault and Fold database (U.S. Geological Survey, 2006).Most of the faults in northeastern California and western Nevada arenot used in scoring the model due to the uncertainties in bothgeometry and slip-rates. We do not force connections between faultsin the grid unless they are documented, because fault segmentation isusually real. Our purpose is to keep the model network geometry asclose to reality as possible. Connectivity between faults is accom-plished by including all known active faults in the model, withoutimposing any cutoff at a predetermined slip-rate. This allows us toFig. 5. Overview of results of simulations performed with different driving mechanisms (NUVcombinations of slip-dependent weakening (as defined by Bird and Kong (1994)) both in tdefined in section 3) exist. No acceptable models can be produced outside this range of fault swith 90% slip-dependent weakening in the upper crust and 50% slip-dependent weakening inparameters, while the sensitivity of SHmax is mostly limited to weakening. The characteristiclower crust, while the broad overall trend of the curves reflects the changes in initialm*. It isthe effects of changing the plate-driving mechanism are second-order. (For interpretation ofthis article.)preserve small connecting faults in the model that would otherwisebe excluded due to their small slip-rates.

3. Results

Velocity vectors resulting from the calculations can be used tomake specific predictions. We want to predict from our model long-term fault slip-rates (geologic rates), azimuth of maximum compres-sive horizontal stress (SHmax), depth to the BDT (brittleductiletransition), and horizontal velocities. The latter need first to becorrected in order to simulate temporary fault locking before they canbe compared with geodetic data. The method used for this correctionfollows Savage (1983) and is described in detail by Liu and Bird(2002). To summarize it, elastic dislocation corrections are calculatedfor the brittle (locked) part of each fault element and then added tothe long-term velocities predicted by our models. In this way, modelresults and observations can be compared directly in the short-termrealm of GPS velocities. It is important to note here that the BDT depthused to define the base of the locked zone in our models is calculatedindependently of any locking depth intrinsic to the GPS datathemselves.

Our model can then be scored against actual observations: 59faults with reliable geologic slip-rates (SCEC CFM, Plesch et al., 2007;USGS Quaternary Fault and Fold database, U.S. Geological Survey,2006), 964 A-C quality SHmax directions from the World Stress Mapdatabase (Reinecker et al., 2005), and 1345 permanent and campaignGPS station velocities (Hackl andMalservisi, 2008). We can also verifywhether our predicted BDT depth is consistent with the hypocentraldepth distribution extracted from the Southern and NorthernCalifornia Earthquake Catalogs. Concerning the quality of fault data,the faults used in slip-rate scoring are A-quality faults (Table 2,supplementary material) for which Max slip-rateMin slip-rate-Average slip-rate, which correspond to 59 of the faults listed in Table 2(supplementary material) for a total of 408 individual fault elements.As a measure of misfit we choose the RMS error in mm/yr for long-term slip-rates and for GPS velocities, and the mean error in degreesfor stress directions. For each computedmodel we also verify whetherthe BDT depth is realistic. Most of the seismicity in the region isconfined to depths of less than 2025 km: brittle faults at largerdepths are unrealistic, as are unrealistic faults that move bydislocation creep to within a few km of the Earth's surface. Thereforewe can exclude all models in which the BDT depth distribution differssignificantly from the observed earthquake distribution.

In order to find the optimum value of * for California faults, westarted by decreasing it systematically from 0.6 (lower bound ofByerlee's friction) to 0.001, initially assigning the same * to all faultsin our local region (i.e. all faults have the same frictional strength).Plate-boundary faults outside the local region are always assigned*=0.03 in agreement with Iaffaldano et al. (2006) and Bird et al.(2008), who show that low values of friction give better results. Ourtest shows a general preference for values of fault friction in California5 to 10 times lower than Byerlee's friction (Fig. 5). We conducted testswith both NUVEL-1A- and MCM-driven models. For the latter, MCMvelocities at two different depths (100 and 200 km) are applied to thebase of the plates (Figs. 1 and 5). We also test models for whichmantle convection speed as been increased by 20% to simulate a largerlithosphere-mantle coupling. As shown in Fig. 5, differences in theEL-1A, or MCMs from Schuberth et al. (2009)). For each value of * we also test varioushe upper and lower crust. Red box shows the range in which all acceptable models (astrengths. Key for reading the x-axis: fri0.01_weak0.9_bwk0.5=effective friction of 0.01the lower crust. Slip-rates and horizontal velocities are sensitive to changes in all input

saw-tooth pattern reveals the prominent effect of weakening the faults in the upper andapparent that the effects of initial * and weakening are of first-order importance, whilethe references to color in this figure legend, the reader is referred to the web version of

377S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384

Fig. 6. Effect of introducing linear slip-dependent weakening in the upper (ucw) and lower crust on slip-rate RMS errors. Significant upper and lower crustal weakening are neededto achieve good results. The star represents our best model, which corresponds to model 2 of Figs. 8 and 10.

378 S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384results between NUVEL- and MCM-driven models are minor whencompared to the pronounced differences associated with changes infault strength. This suggests that fault properties are a key controllingfactor of our models when plate-driving forces are accounted for. Inthe remainder of this paper we will therefore report only results fromNUVEL-driven models for simplicity.

Bird and Kong (1994) postulated a linear dependency of oncumulative fault displacement (slip-dependent weakening): thefaults with the largest net slip will also be the weakest ones. Forexample, if *=0.5 for the fault with the smallest net slip, at 90%linear slip-dependent weakening the fault with the largest net slipwill have *=0.05. We find that slip-dependent weakening generallyimproves results, unless the initial * value is already very low (Figs. 5and 6), and with a few notable exceptions that we will address insection 4.

Concurrently with the tests conducted on * we explored theinfluence of weakening below faults in the ductile zone. We reducedthe creep activation energy below faults in 10% increments from 0% to90% of its value in non-faulted crust (Q=100 kJ/mol). We examinedthree cases: uniform Q reduction below all faults, Q reduction belowthe San Andreas fault only, and linear slip-dependent Q reduction. Thelatter is analogous to linear slip-dependent weakening in the brittlecrust. All viable models show that moderate weakening of majorfaults in the lower crust is necessary to avoid excessive weakening inthe brittle crust and therefore unrealistically large depths to thebrittleductile transition. The range of models that produce accept-able results (Figs. 7 and 8) on all aspects tested (slip-rates, geodeticrates, SHmax azimuth, and brittleductile transition depth) is narrow.All models fall within the range *=0.2 to 0.05 with 90% linear slip-dependent weakening in the brittle crust, and 30% Q reduction in thelower crust either below the San Andreas fault alone (i.e. the SanAndreas is the only fault that is weakened in the lower crust), or aslinear slip-dependent reduction involving all faults. This producesfault slip-rates RMS error values of 3.33.5 mm/yr, geodetic RMS errorvalues of 6.57 mm/yr, SHmax mean error values of 2223, andrealistic maps of the depth to the brittleductile transition belowfaults (Figs. 7 and 8).

As a comparison, the average fault slip-rate (weighted by segmentlength) in the region is 8 mm/yr, while the average intrinsic error inthe worldwide stress data set about 24 (Heidbach et al., 2008) (forcomparison, the average error on the A-C quality data we used isabout 20). GPS velocity errors are larger than slip-rate errors becauseof the much larger number of elements that need to be matched:several thousands of nodes versus the few hundred segments used forfault slip-rates, resulting in many nodes lacking a data point locatedclose enough for a good match. In addition, the need to correct ourmodel results for temporary fault locking means that the comparisonwith GPS velocities can only be an approximation. Thus we define theacceptable ranges (Fig. 8) for the four parameters considered asfollows: (1) slip-rates RMS error 04 mm/yr, where the maximum ishalf the regional average slip-rate; (2) SHmax mean error24;(3) GPS velocity RMS error: 07 mm/yr, a range that includes all ofthe models that score reasonably well on the other parameters; and(4) average BDT depth 1020 km, because values outside this rangecannot be obtained if we want to have a realistic distribution ofmaximum hypocentral depths in the region. The value of 24 as abound for the SHmax mean error is rather high but, as evident fromFig. 5, of the three parameters shown this is the least sensitive tochanges in the model, and making the acceptable range smaller herewould exclude too many models without good reason. There areseveral issues with the BDT depth as well that should be mentioned.First, like for GPS velocities, we do not have a one-to-one match withmodel elements. This is because there are areas of the model wherewe have faults, but not enough earthquakes in the record that matchthese faults (Fig. 7). Second, the earthquake depth distribution is onlya proxy for BDT depth. Third, while it makes sense to assume that forold, well-established faults, the distribution of seismicity has notchanged much in the last million years, this is likely not true foryounger faults. Geologic slip-rates represent therefore the mostrobust of the four scoring data sets, because there is a one-to-onematch between model faults and real faults, and because we cancompare modeled and observed rates directly, without any interme-diate steps.

4. Implications for fault strength in California

Our results indicate that in the upper, brittle crust all examinedfaults in California and western Nevada are weak, (*0.2), andthat further weakening of major faults is important: major faults areabout 10 times weaker than the rest. The strongest faults in thisregion have * in the range 0.20.05 and can therefore be classifiedas weak. Accounting for both the fact that several models produceacceptable results (Fig. 8), and that not all segments of the SanAndreas have the same strength (Fig. 10), we can conclude that theSan Andreas fault has * in the range 0.040.01 and it is therefore aweak or a very weak fault. Low effective friction on faults howeverdoes not necessarily mean a weak crust overall, as it can be inferredfrom the work of King (1986) and as explicitly stated by Suppe

Fig. 7. Examples of calculated depth to the brittleductile transition (BDT). Maximum depth of seismicity, which we use as a proxy for BDT depth, is shown in the top left corner. Wecreated this map by taking the deepest event in each 55 km cell from the combined Southern California Earthquake Center (SCECDC) and Northern California Earthquake Center(NCEDC) earthquake catalogs, without any averaging or interpolation: blank cells have no earthquakes. (a) represents one of our best models with upper-crust linear slip-dependentweakening, showing a reasonable BDT depth along faults. (b) is the strong-faults case (model 1, Fig. 10) with no weakening, which shows too shallow BDT depth. (c) is a model withhigh initial friction and significant weakening, which scores at acceptable levels on slip-rates, GPS velocity and SHmax mean error, but for which the resulting BDT depth isunacceptably low. ucw=upper crust slip-dependent weakening. lcw=lower crust slip-dependent weakening, which is set as follows for the three fault maps: 30% below the SAFalone in (a), none in (b), and 60% uniform (non slip-dependent) below all faults in (c).

379S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384(2007). Thus, the San Andreas fault could be defined either as aweak fault in a weak crust, or a weak fault in a strong crust.Further studies to determine crustal strength in California inde-pendently of fault strength (for example along the lines of thoseconducted by Suppe (2007), on accretionary wedges) are needed tomake this determination.

As far as the reason for such a low frictional strength of faults isconcerned, we agreewith DiToro et al. (2004) on a possible linkwith adrop in dynamic effective friction during earthquakes. The faults thatwe have considered in our simulations, with the notable exception ofthe creeping segment of the San Andreas fault, owe most of their totalslip to earthquakes. Several laboratory studies in recent years (e.g.,DiToro et al., 2004; Han et al., 2007; Tanikawa and Shimamoto, 2009)show that a significant drop in dynamic friction, below values of 0.2and possibly down to near-frictionless behavior (DiToro et al., 2004),occurs when rock interfaces slip at typical earthquake slip velocities.What we produce in our simulations is the long-term result of faultsslipping, regardless of how such slip actually happens or is initiated;we can only predict what the integrated effect over many earthquakecycles will look like. For a fault that slips mostly in earthquakes, it isthe cumulative effect of the processes happening during such eventsthat should dominate in the long-term and control long-termobservables. In other words, if the dynamic friction of a fault duringits characteristic earthquake drops to 0.1, then this is the strength thatwill best match long-term observations for this fault, provided thatover the course of its existence most of the fault area has achieved thepeak slip-velocity necessary for this drop in friction. In this context,only truly creeping faults, of which the creeping segment of the San

Fig. 8. Diagrams showing the results of several different combinations of parameters for models driven by NUVEL-1A. Models are grouped according to presence or absence of slip-dependent weakening of faults in the upper and lower crust. ucw=upper crust slip-dependent weakening, lcw=lower crust slip-dependent weakening. Acceptable models musthave all four vertices of the corresponding polygon fall within the gray-shaded acceptable range (as defined in section 3 of the main text) on each axis of the diagram. Top leftdiagram represents the no-weakening case. The results are improved significantly by the introduction of slip-dependent weakening at faults in at least one crustal layer, as shown inthe two bottom diagrams. The best results are obtained when weakening is introduced in both upper and lower crust (top-right diagram). Our best model is represented by the solidred line. Both extremely low initial friction (*=0.01) and the absence of weakening coupled with high initial friction (*=0.6) consistently produce poor results. (Forinterpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

380 S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384Andreas fault is one of the very few known examples worldwide, needto have very low (

Fig. 9. Fault slip-rates plotted as slip patches for (a) our best model (model 2 of Figs. 8 and 10) and (b) for the high-friction, strong-faults case (model 1 of Fig. 10). The color of slip-rate patches represents the dominant slip component. Numbers next to the patches are slip-rates in mm/yr. Both slip-rates and type of faulting match observations much betterwhen the strength of faults is low and slip-dependent weakening is introduced. In addition to a marked slip-rate increase in (a) when compared to (b), we also observe that thecompression along the Big Bend of the SAF visible as thrust faulting (blue) in (b) disappears in (a). The SAF becomes a dominantly right-lateral strike-slip fault and compression isnow confined to faults on either side of it. SAF=San Andreas fault, SJFZ=San Jacinto fault zone. (For interpretation of the references to color in this figure legend, the reader isreferred to the web version of this article.)

381S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384one fault segment to another. Our results also demonstrate that aglobal finite element code with a variable-resolution grid canrealistically reproduce the tectonics of local areas while being drivenby global plate motions.Acknowledgments

This project is funded by Deutsche Forschungsgemeinschaft (DFG)grants CA691/1-1 and CA691/1-2. The authors would like to thank

Fig. 10. Comparison between observed slip-rates and slip-rates calculated for several models. Model 1 is the strong-faults case: all fault slip-rates are reduced to as little as 5%10% of their observed values. Model 2 is our best model (star inFig. 6), model 3 is another model that shows a slightly better fit for slip-rates on several important faults and that still produces acceptable results for the three other parameters (BDT depth, SHmax, GPS velocities, see Fig. 8).

382S.Carena,C.M

oder/Earth

andPlanetary

ScienceLetters

287(2009)

373384

383S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384the anonymous reviewer whose comments helped to improve thispaper.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.epsl.2009.08.021.

References

Anderson, E.M., 1942. The Dynamics of Faulting and Dyke Formation with Applicationto Britain. Oliver and Boyd, Edinburgh. 191 pp.

Axen, G.J., 2007. Research focus: significance of large-displacement, low-angle normalfaults. Geology 35, 287288.

Bennett, R.A., Wernicke, B.P., Niemi, N.A., Friedrich, A.M., Davis, J.L., 2003. Contemporarystrain rates in the northern Basin and Range province from GPS data. Tectonics 22.doi:10.1029/2001TC001355 CiteID 1008.

Bilotti, F., Shaw, J.H., 2005. Deep-water Niger Delta fold and thrust belt modeled as acritical-taper wedge: the influence of elevated basal fluid pressure on structuralstyles. AAPG Bull. 89, 14751491. doi:10.1306/06130505002.

Bird, P., 1989. New finite element techniques for modeling deformation histories ofcontinents with stratified temperature-dependent rheologies. J. Geophys. Res. 94,39673990.

Bird, P., 1999. Thin-plate and thin-shell finite-element programs for forward dynamicmodeling of plate deformation and faulting. Comput. Geosci. 25, 383394.

Bird, P., Kong, X., 1994. Computer simulations of California tectonics confirm very lowstrength of major faults. Bull. Geol. Soc. Am. 106, 159174.

Bird, P., Liu, Z., 1999. Global finite-element model makes a small contribution tointraplate seismic hazard estimation. Bull. Seismol. Soc. Am. 89, 16421647.

Bird, P., Liu, Z., Rucker, W.K., 2008. Stresses that drive the plates from below:definitions, computational path, model optimization, and error analysis. J. Geophys.Res. 113, B11406. doi:10.1029/2007JB005460.

Bunge, H.-P., 2005. Low plume excess temperature and high core heat flux inferredfrom non-adiabatic geotherms in internally heated mantle circulation models.Phys. Earth Planet. Inter. 153, 310.

Bunge, H.-P., Richards, M.A., Lithgow-Bertelloni, C., Baumgardner, J.R., Grand, S.P.,Romanowicz, B.A., 1998. Time scales and heterogeneous structure in geodynamicearth models. Science 280, 9195.

Bunge, H.-P., Richards, M.A., Baumgardner, J.R., 2002. Mantle-Circulation Models withSequential Data Assimilation: Inferring Present-day Mantle Structure from Plate-Motion Histories. R. Soc. Lond. Philos. Trans. A 360, 25452567.

Brgmann, R., Dresen, G., 2008. Rheology of the lower crust and upper mantle: evidencefrom rock mechanics, geodesy, and field observations. Ann. Rev. Earth Planet, Sci.36, 531567. doi:10.1146/annurev.earth.36.031207. 124326.

Byerlee, J., 1978. Friction of rocks. Pure Appl. Geophys. 116, 615626. doi:10.1007/BF00876528.

Carena, S., Suppe, J., Kao, H., 2002. The active detachment of Taiwan illuminated bysmall earthquakes and its control of first-order topography. Geology 30, 935938.

D'Alessio, M.A., Johanson, I.A., Brgmann, R., 2005. Slicing up the San Francisco Bayarea: block kinematics and fault slip rates from GPS-derived surface velocities.J. Geophys. Res. 110, B06403. doi:10.1029/2004JB003496.

DiToro, G., Goldby, D.L., Tullis, T.E., 2004. Friction falls towards zero in quartz rock as slipvelocity approaches seismic rates. Nature 427, 436438.

Dixon, T.H., Robaudo, S., Lee, J., Reheis, M.C., 1995. Constraints on present day basin andrange deformation from Space Geodesy. Tectonics 14, 755772.

Engebretson, D.C., Cox, A., Gordon, R.G., 1984. Relative motions between oceanic platesof the Pacific basin. J. Geophys. Res. 89, 291310.

England, P., Houseman, G., 1985. Role of lithospheric strength heterogeneities in thetectonics of Tibet and neighbouring regions. Nature 315, 297301.

Fialko, Y., 2006. Interseismic strain accumulation and the earthquake potential on thesouthern San Andreas fault system. Nature 441, 968971. doi:10.1038/ nature04797.

Fuis, G., Mooney, W.D., 1990. Lithospheric structure and tectonics from seismic-refraction and other data. In: Wallace, R.E. (Ed.), The San Andreas fault system. U.S.Geol. Survey Prof, California, pp. 207236. Paper 1515.

Genrich, J., Bock, Y., Mason, R., 1997. Crustal deformation across the Imperial Fault:results from kinematic GPS surveys and trilateration of a densely spaced, small-aperture network. J. Geophys. Res. 102, 49855004.

Graymer, R.W., Ponce, D.A., Jachens, R.C., Simpson, R.W., Phelps, G.A., Wentworth, C.M.,2005. Three-dimensional geologic map of the Hayward fault, northern California:correlation of rock units with variations in seismicity, creep rate, and fault dip.Geology 33, 521524. doi:10.1130/G21435.1.

Hackl, M., Malservisi R., 2008, Strain pattern from geodetic measurements. Geophys.Res. Abstracts 10, EGU2008-A-07579.

Han, R., Shimamoto, T., Hirose, T., Ree, J., Ando, J., 2007. Ultralow friction of carbonatefaults caused by thermal decomposition. Science 316, 878881.

Hardebeck, J.L., Hauksson, E., 2001. Crustal stress field in southern California and itsimplications for fault mechanics. J. Geophys. Res. 106, 2185921882.

Hardebeck, J.L., Michael, A.J., 2004. Stress orientations at intermediate angles to the SanAndreas fault. California. J. Geophys. Res. 109, B11303. doi:10.1029/2004JB003239.

Heidbach, O., Iaffaldano, G., Bunge, H.P., 2008. Topographic growth drives stressrotation in the central Andes: observations and models. Geophys. Res. Lett. 35.doi:10.1029/2007GL032782.Hickman, S., Zoback, M., 2004. Stress orientations and magnitudes in the SAFOD pilothole. Geophys. Res. Lett. 31, L15S12.

Horsman, E., Graymer, R.W., McLaughlin, R.J., Jachens, R.C., Scheirer, D.S., 2008,Examining the evolution of the Peninsula segment of the San Andreas fault,northern California, using a 4-D geologic model. Eos Trans. AGU 89 (53), Fall Meet.Suppl., Abstract T41A-1933.

Houseman, G., England, P., 1986. Finite strain calculations of continental deformation 1.Methods and general results for convergent zones. J. Geophys. Res. 91, 36513663.

Houseman, G., England, P., 1993. Crustal thickening versus lateral expulsion in theIndianAsian continental collision. J. Geophys. Res. 93, 1223312249.

Humphreys, E.D., Coblentz, D.D., 2007. North American dynamics and western U.S.tectonics. Rev. Geophys. 45 paper 2005RG000181.

Iaffaldano, G., Bunge, H.P., 2009. Relating rapid plate-motion variations to plate-boundary forces in global coupledmodels of themantle/lithosphere system: effectsof topography and friction. Tectonophysics 474, 393404.

Iaffaldano, G., Bunge, H.P., Dixon, T.H., 2006. Feedback between mountain belt growthand plate convergence. Geology 34, 893896. doi:10.1130/G22661.1.

International Heat Flow Commission, 2007, Global Heat Flow Database, accessed fromhttp://www.heatflow.und.edu/index2.html.

Jachens, R.C., Simpson, R.W., Graymer, R.W., Langenheim, V.E., Wentworth, C.M.,Stanley, R.G., Colgan, J.P., 2009. Construction of a three-dimensional geologic,tectonic, kinematic crustal model of coastal central California. Seismol. Res. Lett. 80(2), 280.

Jackson, J., 2002. Strength of the continental lithosphere: time to abandon the jellysandwich? GSA Today 12, 410.

King, G.C.P., 1986. Speculations on the geometry of the initiation and terminationprocesses of earthquake rupture and its relation to morphology and geologicalstructure. Pageoph 124, 567585.

Kong, X., Bird, P., 1995. SHELLS: a thin-shell program for modeling neotectonics ofregional or global lithosphere with faults. J. Geophys. Res. 100, 2212922131.

Lachenbruch, A.H., Sass, J.H., 1992. Heat flow from Cajon Pass, fault strength, andtectonic implications. J. Geophys. Res. 97, 49955030.

Liu, Z., Bird, P., 2002. Finite element modeling of neotectonics in New Zealand.J. Geophys. Res. 107, 2328. doi:10.1029/2001JB1075.

Lundgren, P., Hetland, E.A., Liu, Z., Fielding, E.J., 2009. Southern San AndreasSan Jacintofault system slip rates estimated from earthquake cycle models constrained by GPSand interferometric synthetic aperture radar observations. J. Geophys. Res. 114,B02403. doi:10.1029/2008JB005996.

Mallet, J.L., 2002. Geomodeling. Oxford University Press, New York. 593 pp.McCaffrey, R., 2005. Block kinematics of the PacificNorth America plate boundary in

the southwestern United States from inversion of GPS, seismological, and geologicdata. J. Geophys. Res. 110, B07401. doi:10.1029/2004JB003307.

McGill, S. F., Barley, M. E., Hams, J. E., Hobart, K., Ramirez, J., Fryxell, J. E.; Lyzenga, G. A.,McGill, J. D.,2003, Modeling of Geodetic Crustal Motion Velocities in SouthernCalifornia: Undergraduate Research. Eos Trans. Am. Geophys. Union, AbstractED41B-1162.

Meade, B.J., Hager, B.H., 2005. Block models of crustal motion in southern Californiaconstrained by GPS measurements. J. Geophys. Res. 110, B03403. doi:10.1029/2004JB003029.

Miller, M.M., Johnson, D.J., Dixon, T.H., Dokka, R., 2001. Refined kinematics of theEastern California shear zone from GPS observations, 19931998. J. Geophys. Res.106, 22452263.

Mount, V.S., Suppe, J., 1987. State of stress near the San Andreas fault: implications forwrench tectonics. Geology 15, 11431146.

National Geophysical Data Center, 2006, Digital relief of the surface of the Earth:Boulder, Colorado, National Oceanic and Atmospheric Administration DataAnnouncement 88-MGG-02.

Plesch, A., Shaw, J.H., Benson, C., Bryant, W.A., Carena, S., Cooke, M., Dolan, J., Fuis, G.,Gath, E., Grant, L., Hauksson, E., Jordan, T., Kamerling, M., Legg, M., Lindvall, S.,Magistrale, H., Nicholson, C., Niemi, N., Oskin, M., Perry, S., Planansky, G., Rockwell,T., Shearer, P., Sorlien, C., Sss, M.P., Suppe, J., Treiman, J., Yeats, R., 2007.Community Fault Model (CFM) for Southern California. Bull. Seismol. Soc. Am. 97,17931820. doi:10.1785/0120050211.

Provost, A.S., Houston, H., 2003. Stress orientations in northern and central California:evidence for the evolution of frictional strength along the San Andreas plateboundary system. J. Geophys. Res. 108, 2175. doi:10.1029/2001JB001123.

Reinecker, J., Heidbach, O., Tingay, M., Sperner, B., Mller, M., 2005. The 2005 release ofthe World Stress Map. Res. Group of the Heidelberg Acad. Of Sc. And Humanities,Heidelberg, Germany. www.world-stress-map.org.

Richards, M.A., Engebretson, D.C., 1992. Large-scale mantle convection and the historyof subduction. Nature 355, 437440.

Richards, M.A., Yang, W.S., Baumgardner, J.R., Bunge, H.P., 2001. Role of a low-viscosityzone in stabilizing plate tectonics: Implications for comparative terrestrialplanetology. Geochem. Geophys. Geosyst. 2 2000GC000115.

Rivera, L., Kanamori, H., 2002. Spatial heterogeneity of tectonic stress and friction in thecrust. Geophys. Res. Lett. 29 (6), 1088. doi:10.1029/2001GL013803.

Savage, J.C., 1983. A dislocation model of strain accumulation and release at asubduction zone. J. Geophys. Res. 88, 49844996.

Scholz, C.H., 2000a. Evidence for a strong San Andreas fault. Geology 28, 163166.Scholz, C.H., 2000b. A fault in the weak San Andreas theory. Nature 406, 234-234.Schuberth, B.S.A., Bunge, H.P., Steinle-Neumann, G., Moder, C., Oeser, J., 2009. Thermal

versus elastic heterogeneity in high-resolution mantle circulation models withpyrolite composition: high plume excess temperatures in the lowermost mantle.Geochem. Geophy. Geosy. 10, Q01W01.

Shen, Z.K., Jackson, D.D., 2005, Southern California Tectonic Deformation Modeling.SCEC Report.

http://www.heatflow.und.edu/index2.htmlhttp://www.world-stress-map.org

384 S. Carena, C. Moder / Earth and Planetary Science Letters 287 (2009) 373384Southern Methodist University, 2007, Regional geothermal database of U.S., accessedfrom: http:www.smu.edu/geothermal/hetaflow/heatflow.htm.

Suppe, J., 2007. Absolute fault and crustal strength from wedge tapers. Geology 35,11271130. doi:10.1130/G24053A.1.

Tanikawa, W., Shimamoto, T., 2009. Frictional and transport properties of theChelungpu fault from shallow borehole data and their correlation with seismicbehavior during the 1999 Chi-Chi earthquake. J. Geophys. Res. 114, B01402.doi:10.1029/2008.JB005750.

Tapponnier, P., Peltzer, G., Armijo, R., 1986. On the mechanics of the collision betweenIndia and Asia. In: Coward, M.P., Ries, A.C. (Eds.), Collision Tectonics: Geol. Soc.London, Spec. Pub., 19, pp. 115157.

Tapponnier, P., Zhiquin, X., Roger, F., Meyer, B., Arnaud, N., Wittlinger, G., Jingsui, Y.,2001. Oblique stepwise rise and growth of the Tibet plateau. Science 294, 1671.doi:10.1126/science.105978.

Thatcher, W., 1995. Microplate versus continuum description of active tectonicdeformation. J. Geophys. Res. 100, 38853894.

Thatcher, W., Pollitz, F.F., 2008. Temporal evolution of continental lithospheric strengthin actively deforming regions. GSA Today 18, 411.

Titus, S.J., DeMets, C., Tikoff, B., 2005. New slip estimates for the creeping segment of theSan Andreas fault, California. Geology 33, 205208. doi:10.1130/G21107.1.Townend, J., Zoback, M.D., 2004. Regional tectonic stress near the San Andreas fault incentral and southern California. Geophys. Res. Lett. 31, L15S11. doi:10.1029/2003GL018918.

U.S. Geological Survey, 2006, Quaternary fault and fold database for the United States,accessed Oct. 2007, from USGS web site: http://earthquake.usgs.gov/regional/qfaults/.

U.S. Geological Survey, 2007, USGS heat flow database for California, accessed May2007, from USGS web site: http://earthquake.usgs.gov/heatflow/.

van der Hilst, R.D., de Hoop, M.V., Wang, P., Shim, S.-H., Ma, P., Tenorio, L., 2007.Seismostratigraphy and thermal structure of Earth's core-mantle boundary region.Science 315, 18131817.

Wernicke, B., 1995. Low-angle normal faults and seismicity: a review. J. Geophys. Res.100, 2015920174. doi:10.1029/95JB01911.

Zoback, M.D., 2000. Strength of the San Andreas. Nature 405, 3132.Zoback, M.D., Zoback, M.L., Mount, V.S., Suppe, J., Eaton, J.P., Healy, J.H., Oppenheimer,

D., Reasenberg, P., Jones, L., Raleigh, C.B., Wong, I.G., Scotti, O., Wentworth, C., 1987.New evidence on the state of stress of the San Andreas fault system. Science 238,10051111.

http://www.smu.edu/geothermal/hetaflow/heatflow.htmhttp://earthquake.usgs.gov/regional/qfaults/http://earthquake.usgs.gov/regional/qfaults/http://earthquake.usgs.gov/heatflow/

The strength of faults in the crust in the western United StatesIntroductionModelsResultsImplications for fault strength in CaliforniaConclusionsAcknowledgmentsSupplementary dataReferences