Resolution Electromagnetic Imaging of the San Andreas Fault in Central California

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 104, NO. B1, PAGES 1131-1150, JANUARY 10, 1999

High-resolution electromagnetic imaging of the San Andreas fault in Central California

Martyn Unsworth GeophysicsProgram, University of Washington, Seattle

Gary Egbert College of Oceanic and Atmospheric Sciences,Oregon State University, Corvallis

John Booker GeophysicsProgram, University of Washington, Seattle

Abstract. Although there is increasingevidence that fluids may play a significant role in the earthquake rupture process,direct observation of fluids in active fault zonesremains difficult. Since the presenceof an electrically conductingfluid, such as saline pore water, strongly influencesthe overall conductivity of crustal rocks, electrical and electromagneticmethods offer great potential for overcomingthis difficulty. Here we presentand compareresults from high-resolutionmagnetotelluric (MT) profilesacross two segmentsof the San AndreasFault (SAF) whichexhibit very different patterns of seismicity: Parkfield, which has regular small earthquakes and creep events, and in the Carrizo Plain, where the fault is seismicallyquiescent and apparently locked. In both surveys, electric fields were sampled continuously, with 100 m long dipoles laid end-to-end acrossthe fault. From 100 to 0.1 Hz the data from both profilesare consistentwith a two-dimensional(2-D) fault-parallel resistivitymodel. When both transverseelectric and magnetic(TE and TM) mode data are includedin the interpretation,narrow (-0300-600m wide) zonesof low resistivity extending to depths of 2-4 km in the core of the fault are required at both locations. However,at Parkfieldthe conductance(conductivity thickness product)of the anomalousregion is an order of magnitudelarger than at Carrizo Plain, suggestingmuch higher concentrationsof fluids for the more seismically active Parkfield segment. We also image structural differencesbetween the two segments. At Carrizo Plain, resistive, presumably crystalline, rocks are present on both sides of the fault at depths below 3-4 kin. In particular, we clearly image resistivebasement extending •10 km or more east of the SAF, beneath the Elkhorn Hills and Temblor Range. At Parkfield the situation is quite different with a resistiveblock of Salinian granite west of the fault and an electrically conductive, presumably fluid rich Franciscan complex to the east. It is possible that these structural differencescontrol the difference in mechanical behavior of the fault, either directly by affectingfault strength or indirectly by controlling fluid supply.

1. Introduction heat flow anomalies[Brune et al., 1969; Lachenbruch, Severallines of evidencesuggest that high-pressure 1980],imply that the SanAndreas Fault (SAF) is in- fluids in fault zonesmay play a critical role in control- herentlyweak. Subsequently,Willcock et al. [1990]es- tablished that oceanic transform faults are also weak. ling the rupture dynamicsof earthquakes. First sug- While it is possiblethat the intrinsic weaknessof strike- gestedby Hubbertand Rubey [1959], this conceptgained supportfrom the observationby Zob•cket al. [1987] slip faults resultsfrom the presenceof materials in the fault zone with a very low coefficientof friction, it is that in situ stressindicators, and the lack of significant difficult to reconcilethis explanationwith laboratory- derived estimates of typical rock friction coefficients [Byerice,1978]. As a result,low intrinsic fault strengths Copyright 1999 by the American GeophysicalUnion. have most often been explainedin terms of high pres- Paper number 98JB01755. sure fluids in the fault zone [e.g., Sleepand Blanpied, 0148-0227/99/98JB-1755509.00 1992;Byeflee, 1990, 1993;Rice, 1992]. This interpreta-

1131 1132 UNSWORTH ET AL.- SAN ANDREAS MT STUDY tion is supported further by geologicalmapping of ex- vey MT data collected in the Parkfield area, Eberhart- humed faults, which provides direct evidence that fluids Phillips and Michael [1993]inferred a significantresis- were once present in zonesof active strike-slip deforma- tivity contrast acrossthe San Andreas Fault, with con- tion [e.g.,Sibson et al., 1988; Chesteret al., 1993]. ductive material east of the fault. Unpublished USGS Zobacket al. [1987]demonstrated that simplemod- data (W. D. Stanley, personalcommunication, 1994) els of overpressuredfluids in the crust are inconsistent suggestedthat situation at Carrizo Plain was reversed, with in situ stress observations. To avoid unobserved with an unexpectedresistive body eastof the San An- hydrofracturing, fluid pressureswould have to be lim- dreas Fault beneath the Temblor Range. Mackie et al. ited by the minimum principal stress. To overcomethis, [1997]imaged a similar pattern of resistivityvariations and related difficulties, more sophisticatedand detailed along two additional profiles acrossthe fault within the theoretical models for the dynamical behavior for flu- Carrizo Plain. These surveysused data collected at sites ids in mature active fault zones have been proposed severalto tens of kilometers apart. Although they have [Sleepand Blanpied,1992; Byeflee, 1990, 1993; Rice, imaged large-scalecontrasts in crustal resistivity across 1992]. Although differingin many important details, the fault and have sometimesprovided indirect evidence these models all suggest a rather complicated picture for enhanced conductivity within the fault zone, sam- in which the distribution, connectivity, and pressureof pling was not dense enough to provide a detailed view fluids may vary dramatically in space and time, in a of the actual fault zone. manner which is closely linked to the earthquake cycle In this paper we describe the results of two high- on individual fault segments. resolution electromagnetic surveys acrossdistinct seg- To distinguish between theoretical models such as ments of the San Andreas Fault where the earthquake these, it is necessaryto map the location of fault zone cycle is very different: at Carrizo Plain and at Middle fluids in detail. While deep drilling such as that pro- Mountain, Parkfield (Figure 1). In these surveyswe posed by Hickman et al. [1994] will ultimately be sampled the electric fields on the surface with a contin- needed to determine the physical state within the cores uous series of dipoles laid end to to end. This style of of active faults, remote sensingtechniques capable of continuousMT profiling gives a high spatial data den- detecting in situ fluids to a depth of at least 10 km sity and minimizes the possibility of spatial aliasing of are clearly useful. Eberhart-Phillipset al. [1995] re- viewed the resolution that might be obtained by seismic, electrical, and electromagnetic techniques. Since II the presenceof an electrically conductingfluid, such as 42 ø saline pore water, strongly influencesthe overall elec- l:• trical conductivity of crustal rocks, electrical and elec- 0 100 200 300 tromagnetic methods offer great potential for detect- km ing fluids at depth within the fault zone. Electromag- netic exploration methods can be divided into those us- N ing natural sourcessuch as magnetotellurics(MT) and those that require the generation of artificial electromagnetic signals. Electromagneticsignals propagate as 38 ø lossfree waves in the atmospherebut diffuse as highly attenuated waves in the ground which can generally be considereda good conductor. This attenuation is char- acterizedby a skindepth 5 = (2/luwrr)•/•', whereco is the frequency of the signal, /t is the magnetic per- meability, and rr is the conductivity of the earth. To explore deep structure requires large skin depths and 34 ø usually low-frequencysignals. Such signals are difficult to generate artificially, but for most frequenciesbelow 100 Hz, natural sourcesignals are often strong. MT is thus generally used for crustal scalesounding. In recent years, MT profiles have crossedthe San An- 124 ø 120 ø 116 ø dreasFault at a numberof locations.Park et al. [1991] collectedwidely spacedlong-period sites from the Pa- San Andreas Fault-locked segments cific Ocean to Nevada, with one profile crossingthe San San AndreasFault- creeping segments Andreas Fault closeto Parkfield. This study suggested Other major faults( only shownin California) that the Great Valley was bounded by electrically conductive suture zones but that the San Andreas Fault Figure 1. Map showingSan Andreas Fault and the was not a significantly conductivefeature in the lower locations(solid dots) of continuousMT profilesat Car- crust and upper mantle. Based on U.S. GeologicalSur- rizo Plain (C) and Parkfield(P) UNSWORTH ET AL' SAN ANDREAS MT STUDY 1133 near-surfacestatic shiftsinto deeperstructure [Torres- site to permit remote referenceprocessing of the time Vetdin and Bostick,1992]. The method allows us to seriesdata [Gambleet al., 1979]. The remotesite for obtain high-resolutionimages of electricalconductivity the Carrizo Plain profile was located -•5 km from the variations within and near the fault zone to depths of southwestend of the profile; at Parkfield the remote ref- around 4-6 km. Experimental details are provided in erencewas --•15 km from the profile. Data were recorded section 2. We then present and interpret results from at each location overnight so that 500 to 1500 m of pro- modeling and inversion of the MT data for each profile were covered per day. file. As we shall show, there are significantdifferences Time seriesdata were analyzedusing the robust pro- in geoelectricstructure betweenthe lockedCarrizo seg- cessingcode of Egbert[1997] (a multiple-stationexten- ment and the Parkfield segmentwhich has regular small sionof the methodof Egbertand Booker [1986]) to com- earthquakes and creep events. In section 5 we consider pute estimatesof the MT impedancetensor for frequen- possibleimplications of the different conductivity struc- ciesbetween 100 and 0.005 Hz. In general,data quality tures for fault dynamics. was very good, except in the frequencyrange 1-0.1 Hz. In this "dead band" a combination of low MT signal 2. Data Acquisition and Time Series strength, noisein the magneticchannels due to ground Analysis motion, and coherent cultural noise makes collection of high-quality MT data difficult. For these frequencies Data were collected using two Electromagnetic In- the remote reference estimates proved essential to re- struments Incorporated MT-1 systems. The first sys- move downward bias in impedancemagnitudes due to tem recorded10 channelsof data, includingthe horizon- excessivenoise in magnetic field components. Multiple tal magnetic fields at the center of each array, and elec- station processingusing the approachof Egbert[1997] tric fields for five along-profileand three across-profile detected low levelsof coherentnoise in narrow frequency dipoles. The layout is illustrated and contrasted with bands near 4, 2, and 8 s. Curiously, noise in these same the conventionalMT layout in Figure 2. Along the pro- bands was found for both the Carrizo Plain and Park- file, dipoles were laid end-to-end, so that a continuous field profiles. The source of this noise remains unclear. profile of electric. fields was obtained across the fault. Reprocessingwith the robust multiple-station approach At Carrizo Plain, 100 m dipoles were used close to the reduced the effectsof this coherent noisesomewhat, but fault, with 300 m dipoles at the ends of the profile. At estimates at these frequencies still show more scatter Parkfield, 100 m dipoles were used throughout the sur- for some arrays and generally have larger error bars. vey. The secondinstrument was used to simultaneously However, these effects are confined to relatively nar- collect standard five-channel MT data at a permanent row bands. Comparisonto resultsfor neighboringfre-

CONTINUOUS MAGNETOTELLURIC PROFILING

MT profile

CONVENTIONAL MAGNETOTELLURIC SURVEY

Electric field dipole

Magneticfield sensors

Figure 2. (top) Layoutof electricdipoles and magneticsensors used for electromagneticarray profiling,(bottom) comparedto conventionalMT surveyconfiguration. The eight dipolesdis- playedfor the continuousMT profile are recordedsimultaneously in a singlearray. Additional arrays were laid out end-to-endto allow continuousalong-profile sampling of the electricfields. Closeto the fault, electricfield dipoleswere 100 m in length;farther away,they were 300 m. 1134 UNSWORTH ET AL.: SAN ANDREAS MT STUDY

quencies, where no coherent noise was detected showed soa [1986]show Franciscan at a depth of 3 km beneath that the impedances were not significantly biased by the TemblorRange), but there is little direct evidence this noise. for its presence. Previous MT surveys in the Carrizo At Parkfield the remote site failed after the first two Pla.in detected resistive basement at depths less than 3 arrays had been completed. It was thus not possi- km beneath the Temblor Range [Mackie et al., 1997; ble to compute standard remote reference estimates for W.D. Stanley,personal communication, 1994.] most of this profile. To reduce bias in the dead band In the absenceof microseismicity,the location of the impedances,we again used the multivariate approachof San Andreas Fault at depth is uncertain. Recent geo- Egbert[1997]. This schemeautomatically computes es- morphicstudies by Arrowsmith[199,5] suggested that a timates of signal-to-noise ratio for each channel, allow- vertical fault to a depth of several kilometers is required ing unbiased estimates of impedancesto be computed to explain uplift of the Elkhorn Escarpment. Seismicre- from an eigenvectordecomposition of the spectral cross flection and well log data to the southwestof the San productmatrix [Park and Chave,1984]. Althoughde- Andreaswere interpreted by Davis et al. [1988]to show velopedfor multiple-station processing,it can be shown that the Morales thrust shallowswith depth and the re- that this scheme should also work well with the 10- sulting horizontal detachment may have offset the San channel MT profiling data consideredhere. In effect, Andreas 5-14 km to the east at depths of 5-7 km. the replicated electric field componentscan be used for the remote. As a check, results obtained by this ap- 3.1. DimensionaliCy and Distortion Analysis proach were compared to robust remote reference re- suits for the first two arr•tys, where a remote reference After the time series analysis described above, the was available. In general, no consistentbiases are ob- dimensionalityof the MT data was investigatedusing served in this comparison, so we believe the results for the tensor decompositiontechnique of Chave and Smith the full Parkfield profile are unbiased, even in the dead [1994]. Frequency-independentstrikes were computed band. for four frequencybands for eacharray ( i.e., the five MT sitescollected each day). The misfit betweenthe impedance predicted by the distortion model and the 3. Carrizo Plain measured impedance was computed for strike angles The SanAndreas Fault at CarrizoPlain (Figure1) is from 0ø to 900 in terms of a root mean square(rms) crossed by numerous offset creeks which clearly define misfit. The strike angles that minimized the rms mis- the right-lateral offset of the fault. This segment has fit for each array are shown in Figure 3. In the high- been locked since the 1857 Fort Tejon earthquake, and frequencyband (100-10 Hz) the bestfitting strikeis 80 seismicity has been minimal since the advent of mod- east of the surfacetrace, but a significantscatter occurs ern instrumentation[Hill et al., 1990]. The Cretaceous east of the surfacetrace. In the middle-frequencybands Salinian block lies to the west of the fault with granitic (10-1 Hz and 1-0.1Hz) there is little variationbetween and metamorphic crystalline basement overlain by 1-4 arrays, and strikes are within 50 of the fault trace. The km of predominantly nonmarine Tertiary sedimentary lowestfrequency band (0.1-0.01Hz) showsa consistent rocks [Dibblee,1973a,b; Page, 1981]. Severalauthors strike 250 west of the fault strike. Figure 4 showshow have suggestedthat the crustal block west of the San well the decompositionmodel fits the impedance data Andreas is not part of the Salinian block [Ross,1984; for site 30 at Carrizo Plain. In the three high-frequency Irwin, 1990]. The so-calledBarter Ridge Sliceappears bands(100 to 0.1 Hz) the impedancesare fit to within to have an origin similar to the crystalline rocks of the arms normalized misfit in the range 1-5. At lower fre- San Gabriel Mountains to the south. quencies,in the 0.1-0.01 Hz band, the two-dimensional East of the San Andreas Fault, Cretaceousto Tertiary (2-D) distortedmodel doesnot fit well, with rms mis- marine sedimentary units are exposed in the Temblor fits from 10-50. This poor fit indicatesa 3-D regional Range, and thrusting and folding indicatesfault-normal structure and is consistentwith the changeof strike at compression[Dibblee, 1973a,b; Page, 1981; Ryder and these frequencies.Fits to the distorted two-dimensional Thomson,1986]. Extensivemapping and well log data in model for the other sites were similar. Figure 5 shows the oil fields of the eastern Temblor Range have shown the impedance skew for all sites. Above 0.1 Hz most that up to 7.5 km of sedimentary units underlies the skewsare lessthan 0.1, again suggestinga 2-D geoelec- Elk Hills some 15 km east of the San Andreas Fault tric structure. Belowthis frequency,skew becomes large [Fishburn,1990]. Farther east, seismicrefraction data indicating a 3-D geoelectricstructure is being sensed. have shown up to 12 km of sedimentary rocks in the The small skewvalues and uniform strike at frequen- Great Valley [Colburn and Mooney, 1986]. However, cies between 100 and 0.1 Hz suggeststhat the MT data the character and depth of basement between the oil are consistentwith a two-dimensional(2-D) interpreta- fieldsand the San AndreasFault (i.e., beneaththe Tem- tion, in a fault normal and parallel coordinate frame. blor Range) remain uncertain. It is assumedthat the Data from 0.1 to 0.01 Hz should be included only with wedge of Franciscan formation that abuts the fault to caution, particularly if interpreted in the same coor- the north is alsopresent here ( e.g., Ryderand Thom- dinate frame. The changeof strike estimatedfor this UNSWORTH ET AL.' SAN ANDREAS MT STUDY 1135

35 ø 08' 100- 10 Hz 10-1 Hz Meanstrike =8 deg ' Site30 '?:•,::.X, ß Meanstrike=5.6deg 35 ø 06'

35 ø 04'

35 ø 08' o.1 - O.Ol Hz Meanstrike =-25 deg

35 ø 06'

35 ø 04'

119ø 40' 119 ø 38' 119 ø 36' 119 ø 40' 119 ø 38' 119 ø 36'

0 1 2 3 4 5 km

Figure 3. Map of continuousMT profile at Carrizo Plain with arrows for each array showing best fitting 2-D strike direction in four frequency ranges. Except for the lowest-frequencyband the geoelectricstrike is consistentlywithin 5øof the surfacefault trace.

last band and the increased misfit to the 2-D distorted modelsuggest that theremay be somecontamination by 1.0 - ImpedanceSkew for Carrizo Plain profile three-dimensional(3-D)effects at long periods,where the MT data sample a larger volume. In particular, the deepstrike (25øW of the surface trace) may reflect the 0.8

Carrizo Plain site 30 0.6

• lOO .E lO 0.4

.__-•- •o 0.1 Vv' X/"vy ' o 0.01 0.2 I I I IllIll 100 10 1. 0.1 0.01 Frequency(Hz) 0.0 Figure 4. Fit of data to decompositionmodel for site 30 at Carrizo Plain. The solid curve indicates the minimum misfit that occurs as strike is rotated from 0 ø to 100 10 1. 0.1 0.01 900 . An ideal minimum misfit would be 1. The dashed curve indicates the maximum misfit. When minimum Frequency (Hz) and maximum misfit are similar in value there is limited control on strike, when different the strike is well Figure 5. Impedance skew for the all sites on the Car- constrained. rizo Plain profile. 1136 UNSWORTH ET AL- SAN ANDREAS MT STUDY

I ! ! ,

mmmmmmmmmmmlmmm mmlmmmmmm ILl

N N N N N UNSWORTH ET AL.: SAN ANDREAS MT STUDY 1137 changein directionof the San Andreasto the southeast of the highly conductive ocean on MT measurements of the Big Bend or the surfacegeologic strike of the made onshore[ e.g., Ranganayakiand Madden, 1980: Cuyamavalley and related faults to the south [Page, Mackie et al., 1988]. The TM mode data are fit well, 1981].Alternatively, these indications of 3-D complica- to a normalized rms misfit of 2.1. This correspondsto tions could reflect the changein direction of the coast- misfits of individual data of m 10 percent for apparent line in southern California. resistivities, and 50 for phases. Note that ideally misfits The resultingapparent resistivity and phasedata in would be equal to the standard errors in the data i.e., a fault parallel( i.e. unrotated)coordinate system are the rms misfit would be 1. However, the TE response shown in Plate la. The horizontal ticks denote the cen- of this model does not fit the data below a frequency of ters of the electric field dipoles,and frequencyis shown i Hz. on the vertical scale. Because of the skin depth effect 3. A two-dimensionM inversion was started from the noted above, this pseudo-sectiongives a senseof how model obtained in step 2 and required the TE apparent the structure varies with depth. TE mode apparent re- resistivity(PT•) and phase((I>T•) to be satisfiedin ad- sistivity and phasecorrespond to electric currentsflow- dition to the TM mode data. The RRI algorithm was ing alongthe fault, whilethe TM modedata correspond unable to significantlylower the misfit for the combined to electric currents flowing normal to the fault. data set because it was trapped in a local minimum of the data misfit function. Using trial-and-error forward 3.2. Modeling and Inversion modeling to improve the fit to the TE responseswith- The apparent resistivity and phase data were con- out significantly altering the TM response, we found verted into a spatially smoothelectrical resistivity model that both TE and TM mode data could be fit reason- using a combinationof inversionand forward model- ably well by the addition of the large conductor east ing. The rapid relaxationinversion (RRI) of Smithand of the profile that was imaged in previousregional MT Booker[1991] was appliedin the followingsteps: surveysby Mackie et al. [1997]and W. D. Stanley(per- 1. Horizontallyaverage TM apparentresistivity (PTM) sonalcommunication, 1994). Given the way in which and phase((I>TM) data, and invert for a 1-D layered the RRI algorithm searchesmodel spaceand the lack of Earth model. At each frequencyan averagepT• and data on the east end of profile, it is not surprisingthat (I>T• was computed for all sites on the line. This pro- this better fitting model was not found by the inversion cedure allowed the inversion in step 2 to convergemore scheme alone. rapidly than if started from a half-space. 4. A final inversion was started from this hybrid 2. Using this model as a starting point, begin a 2- model, resulting in a combined rms misfit of 2.6 for D inversionto fit the TM data (pT• and •T•). The both modes. The model and its response are shown in resultingmodel and computedresponse are shownin Plate lc . The verticalstril•es in PT•,are due to the Plate lb. Note that the actual model used extended 300 application of frequency-independent static shift coef- km to the east and west and included the Pacific Ocean ficients that were computed by the inversion and dis- as a fixed feature, to allow for the significant effects played in Figure 6a. Allowing the inversionto estimate

2 - c a

• 1

._•

o- I I I ,I I TEApparent resistivity h (b) • 6 ••, TEphase J• E• -- --, TMApparent resistivity I •i• -'• ••' TM phase I• •'. •

O[ I I , I , I I -4 km -2 km 0 km 2 km 4 km

Distance from surface trace (km) Figure 6. (a) Estimatedstatic shift coefficientsfor the CarrizoPlain data. (b) Misfit between data and predictedresponse of model in Plate lc as a function of distancefrom the surfacetrace. 1138 UNSWORTH ET AL.: SAN ANDREAS MT STUDY static shifts effectivelydownweights the resistivity data, are small(Figure 6a), indicatingthat strongdistortion making the interpretation lesssusceptible to distortion of the whole profile is unlikely. Furthermore,this type by galvanic effects and is equivalent to the technique of distortion has lesseffect on the phase,which we have of manually "correcting" for static shift. Note that emphasized in our inversion. The 3-D effectscould also both modes are generally fit well down to a frequency resultfrom conductivestructures ( e.g., C4) with lim- of 0.1 Hz. Below this frequency the predicted phases ited continuity along strike. In this case the TM re- are higher than the observed.This may reflect the fact sponsewould be virtually unaltered, but pT•. could be that a 2-D fault-oriented model is invalid at these fre- artificially loweredby electric chargesdeveloping at the quencies, as implied by the tensor decompositionand endsof the conductor[Wannamaker et al., 1984].A 2- skew values. The normalized rms misfits for each site D interpretation emphasizingpT•. could thus lead to an are shownin Figure 6b. Misfit levelsare relatively uni- overestimate of conductancefor the 3-D feature. Note, form across the profile, except for two adjacent sites however, that this discussionpresumes the existence of 2 km east of the fault trace. This poor fit to the TE the conductive structure. It is the magnitude of the data is almost all due to the very poor fit for the lowest conductance, not the presence or absence of the fea- frequencies(0.1-01 Hz) andprobably reflects a combina- ture which might be called into question by possible tion of local distortion and the changeof strike direction unresolved3-D effects. Given the consistentgeoelectric at these frequencies. strike, the small skews, and the essential similarity of A similar model was obtained by the inversion se- the TM and TE+TM models, we believe that a 2-D in- quence:(1) Fit TE data startingfrom a 10 f2 m half- terpretation of this dataset is reasonable. However, it space;and (2) from this model,fit both TE and TM should be borne in mind that quantitative estimates of mode data. This gave a good fit to the TE data by conductancefor somefeatures might be biasedto some placing the conductive structure at depth east of the small extent by 3-D end effects. profile but did not fit the small-scalestructure required by the TM data at high frequencies.While somedetails 3.3. Interpretation in the geometry vary, both the TM data and TE+TM Our interpretation of the Carrizo Plain resistivity data modelshave the followingconductive (C) and re- model is summarized in Plate 2a. Note that faults can sistive(R) featuresin common: changethe electricalresistivity of the groundin at least C1, 5-10 f2 m conductive layer west of the surface two distinct ways. They can produce a zone of low re- trace of San Andreas Fault; C2, 3-7 f2 m conductive sistivity by trapping groundwateruphill of a fault and zone beneath the surface trace in the upper km; C3, also by forming a zone of low resistivityin the gouge 7-20 f2 m conductive layer at the east end of the profile; and breccia at the core of the fault. C4, Zone of slightly enhancedconductivity beneath the 3.3.1. Structure west of San Andreas Fault. fault trace; R1, 50-100 f2 m resistive, westwarddipping The shallow conductivity structure is well-correlated layer; R2, 300-1000 f2 m resistivebasement below 2 km with geologicmapping and well log data from the area. east of fault; R3, resistive basement below depth of 3-4 West of the fault, Vedder[1970] mapped the Wells km west of San Andreas Fault. Ranch syncline, consistingof Caliente formation sed- The differences between the TM and TE+TM mod- imentary units [Plate 2a]. This correspondsto con- els can be understood by consideringthe physicsof 2-D ductive unit C1, and the westwarddipping resistivity MT. Thin vertical conductors are essentially invisible contours between 3 and 5 km west of the SAF corre- to TM mode electric currents. Positive and negative late well to the dip of the easternlimb of the syncline. electric chargesdevelop on opposite sides of the con- At a depth of •4 km the resistivityincreases signif- ductor at depth, but the effect of these chargeslargely icantly (R3), presumablycorresponding to crystalline cancels out at a location on the surface. However, TE basement of the Salinian block. The location of this mode electric currents flowing along the conductor are boundary inferred from seismicreflection by Davis et readily detectable on the surface, even if the body is al. [1988]agrees well with the resistivitymodel. Davis very narrow. Thus the Carrizo Plain TM data are sen- et al. [1988]suggest this contactis a backthrust fault sitive to the contrast in resistivity acrossthe fault but (F4) associatedwith the deeperMorales thrust, whose do not require a significantfault zone conductorbelow inferred location is shown in Plate 2a. 2 km. In contrast, the TE data are more sensitive to 3.3.2. Shallow structure east of San Andreas the presenceof the narrowconductor (C4) in the fault Fault. Immediately to the east of the surface trace is zone, which extends in the combinedTE+TM model to a wedgeof low resistivity(C2) with its westernbound- a depth of 6 kin. ary directly beneath the surface trace. Note that the It is important to consider the possibility of 3-D ef- surfacetrace liesjust east of the changein elevationas- fects. One possiblesource of 3-D effectsis the shallow, sociatedwith the Dragonsbackpressure ridge. Mapping but very conductivebasin partly occupiedby Soda Lake and well logscompiled by Vedder[1970] indicate that at Carrizo Plain. This could systematically distort the C2 is the Bitterwater Creek shale, a clay rich sequence electricfields so that large static shift coefficientswould with low electricalresistivity. Trapping of groundwater be required to fit pT•.. However,estimated static shifts uphill (east)of the fault may alsocontribute to the low UNSWORTH ET AL.: SAN ANDREAS MT STUDY 1139

resistivity.Arrowsmith [1995] found a numberof fault- who collected MT data at a site east of the Temblor controlledsprings in this area, consistentwith the idea Range. Thus the resistive unit is most likely only a that fault can act as a barrier to groundwaterflowing horizontally limited sliver of crystalline rock. Clearly, from the northeast in the near surface. Farther to the it does not extend into the oil fields east of the Temblor east, R1 correlateswith the Santa Margarita formation, Range, where wells have penetrated up to 7.5 km in sed- describedby Vedder[1970] and Dibblee [1973a,b] to be a imentary rocksof the Great Valley sequence[Fishburn, sedimentaryunit containinggranite bouldersup to 12 ft 19901. in diameter, in addition to gneissand other clasts. This While Zobacket al. [1987]showed that the San An- may account for the relatively high resistivity of this dreasis inherentlyweak, Castillo and Hickman [1995] unit, and the boundaries(inferred to be thrust faults) suggestedthat within 5 km of the San Andreas Fault correlate well with the spatial extent of the central shal- the Carrizo segment is strong. Our data support this low resistive zone. Near the eastern end of the profile, suggestionsince it implies there is crystalline, presum- C3 is interpreted as a sandy-clayrich faciesof the Mon- ably mechanicallystrong material on both sidesof the terey shale which also has a low resistivity. Note that fault below 2 km depth. The material to the west could there are two distinct zones of low resistivity, and that be Salinian granite or, alternatively, crystalline base- both of theselie east of, and topographicallyabove, the ment of the Barret Ridge Slice. The material to the thrustfaults (F2 andF3) mappedby Vedder[1970] and east is of undetermined origin. It could be a fault slice Ryder and Thompson[1986]. Concentrations of ground- of Salinian granite or part of the Barret Ridge Slice, water trapped by thesefaults, may contribute to the low which is believed to originate in the crystalline base- resistivitiesimaged along the MT profile. ment that underlies the Mojave desert and San Gabriel 3.3.3. Resistive basement east of San Andreas Mountains[Ross, 1984; Irwin, 1990]. Fault. The most striking feature east of the fault is 3.3.4. Fault zone conductor. In contrast to the shallowdepth (•0 2 km) to resistivebasement (> 300 Parkfield (see Unsworthet al. [1997a]and also sec- f2 m). Seismicreflection and MT data at Parkfield tion 4) the SAF at this locationis a relativelysubtle [Unsworthet al., 1997a] showedthat the top of the featureof theresistivity model. Although the MT data Salinian granite correspondedto the 100 f2 m contour. are consistentwith a modest reduction of resistivity in This suggestsa crystalline basementat a depth of 2 the fault zone, this feature is not particularly striking. km beneath the Elkhorn Hills and Temblor Range east In the upper kilometer the fault appears to act as a bar- of the San Andreas. However, this interpretation is at rier to groundwater flow, as a zone of low resistivity is variance with currently acceptedtectonic models of this foundeast (on the uphill side)of the fault. Belowthis area [Page, 1981] in which Great Valley sedimentary depth a zone of reduced resistivity extends to depths units and the Franciscan formation extend westward to of at least 4 km. This feature could represent a weak the San Andreas Fault. These rock units would be ex- fault zone conductor that has been smeared out by the pected to have a lower resistivity than the 200-3000 f2 inversion, as observed in synthetic MT model studies m observed,although Park et al. [1991]reported resis- by Eberhart-Phillipset al. [1995]. However,the fault tivities greater than 500 f2 m in the Franciscanforma- zoneconductance (conductivity times fault zonewidth) tion, possiblydue to greenstonesor metabasalts. Sig- at Carrizo Plain is of the order of only 20 S. This is nificantly, Graff [1962] reports that the Western Gulf an order of magnitude less than the 250 S fault zone Oil Vishnu No. i well (locatedsome 6 km alongstrike conductanceimaged at Parkfield which Unsworthet al. from our profile on the northeast side of the San An- [1997a]interpreted as due to an anomalouslyhigh fluid dreas) bottomedat 2114 m in granite. This resistive concentration. unit has also been detected by previous MT surveys What is the maximum fault zone conductance con- that crossed the San Andreas Fault at several locations sistent with the data? To explore this question, the within Carrizo Plain [Mackie et al., 1997; W.D. Stan- best fitting model was modifiedby decreasingresistiv- ley, personalcommunication, 1994]. Thus the resistive ity within the fault zone in a variety of ways. Figure feature is neither a local structure confined to a sin- 7 showsthe observedTE mode data and the responses gle profile, nor an artifact of the data acquisitionand of selectedmodels at site 30, which was typical of sites processingreported here. within 2 km of the surfacetrace. Adding a 100 m wide 2 How far east does the resistive structure extend? It 12m fault zoneconductor ( i.e., 50 S conductance)low- certainly extends to the eastern end of the profile, and ers pT•, significantlybelow 0.5 Hz. Increasingthe width beyond this point our data provide minimal control but to 300 m (150 S conductance)produces an evenmore are still sensitive to some structure at depth. During unacceptablefit. Alternatively, increasingthe resistiv- the inversion processdescribed above, it was found that ity of the fault between i and 3 km to 10 i2 m raisespT•, to fit the essentiallyfiat pT• responses,lower resistiv- so that the data are not fit between 5 and 0.2 Hz. Note ity material was required somewhereto the east of the that the TM data for the samemodels shown in Figure profile and that the resistive basement block could be 7 are much less sensitive to variations in fault zone con- terminated •6 km east of the SAF and not shown in ductivity. Thus the fault zone conductanceis at least Plate 2a). This wasconfirmed by Mackie et al. [1997], partially constrained by the TE data. However, as we 1140 UNSWORTH ET AL' SAN ANDREAS MT STUDY

Carrizo Plain - Dipole 30 - TE mode Carrizo Plain - Dipole 30 - TM mode

i 60

m 50 60 - •

• 40 g 30 20 20 TI[ I I I Il'"l , , ,,,,,d , , ,,,,,,I , I I ' l l,,,,I , i ,,,,,,I , , ,,,,,,! ,

100 10 1 0.1 100 10 1 0.1 frequency(Hz) Frequency(hz)

Best fitting model from Plate lc Add 2 ohm m, 100 m fault zone

...... Add 2 ohm m, 300 m fault zone

Remove conductive fault 1-3 km depth range

Figure 7. TE and TM mode responsesand measureddata at dipole 30 for a set of models incl•udinga more conductivefault zone. Note that the TM mode data are virtually insensitiveto the fault zone conductance, while the TE data constrain this quantity. have discussedin section 3.2, the possibility of 3-D end the San Andreas Fault. The contact between C2 and effects adds some uncertainty to quantitative estimates R1 might be a thrust fault (F1) that offsetsthe San of fault zone conductance. Andreas Fault to the west at around 1500 m depth. 3.3.5. Faulting pattern. Shallow crustal defor- Farther to the east, F2 and F3 are thrust faults mation in the vicinity of the MT profile is dominated mappedat the surfaceby Vedder[1970] and Ryder and by a seriesof westward dipping thrust faults, denoted Thomson[1986]. The locationsof thesefaults within as F1-F4 in Plate 2a. In our interpretation the main the upper 2 km of sedimentary rocks are strongly cor- vertical trace of the San Andreas Fault is offset later- related with variationsin the resistivitymodel [Plate ally to the west by 200-300 m at a depth of around 500 2a]. These faults may also penetrate the crystalline m. Basedon geomorphicmapping, Arrowsmith [1995] basement and possibly intersect the San Andreas at proposeda model for the uplift of the Dragonsbackpres- depth, but the MT data offer no supporting evidence sure ridge in the Elkhorn Hills, in which a thrust fault for this possibility. Indeed, our model does not exhibit intersectedand beheadedthe originally vertical SAF at any significant vertical offsetsin depth to the resistive around 500 m depth. The inferred location of F1 cor- basementbeneath this part of the profile. relates well with the near-surface zone of low resistivity Can our data constrain the location of the San An- in our modelin Plate 2a. Arrowsmith[1995] estimated dreas Fault at depth? The location of the Morales that an offset of around 200 m was needed to produce thrust inferredby Davis et al. [1988]is shownin Plate the necessaryuplift. Again, this is in good agreement 2a. This resistivity model shows no evidence for con- with the modeled location of the deeper fault zone conductive sedimentaryunits below this depth. Certainly, ductor in Plate 2a. Note that fault F4 (inferredfrom a horizontally limited feature at this depth is near the the electromagneticallymapped locationof the resistive limits of what we might resolvewith our 9 km long basementcontact) might be a continuationof F1 across profile. The longer profile of long-periodMT measure- (a) Carrizo Plain Dragonsback Elkhom Plain escarpment

SAF F3

0km , F1 F2 i I ID zl ! I Tm \ I I 2 km I I I I I I I I I

4 km II R2 ?? Kg ii R3 ii II II "•• MoralesTh'rust•. 6 km

Slack Middle Little (b) Canyon Mountain Cholame Creek SAF GHT

0 km

'For c]

2 km t \ Kg

300 100 30 10 3 I

Resistivity (ohm m)

Pla•e 2. Interpretationof resistivitymodels. (a) Carrizo Plain with geologicunits labeled by Vedder [1970]: Kg, Salinian granite; Tbw, Bitterwater Creek Shale; Tms, Monterey shale; Tsm, SantaMargarita formation; DZ, damagedzone. Seetext for discussionof faultsF1-F4. Middle Mountain, Parkfieldwith geologyeast of the San AndreasFault taken from Sims [1990] Tcr, Tertiary Coast Range units; Tgv, Great Valley sequence;Kjf, Franciscanformation. GHT, Gold Hill thrust Fault. Dots denote earthquake hypocentersfrom the Berkeley seismic network. Models shown with no vertical exaggeration. 1142 UNSWORTH ET AL.' SAN ANDREAS MT STUDY ments describedby Mackie et al. [1997]also fails to spond to the San Andreas Fault offset to the east at image conductivestructure below the inferred Morales depth. thrust, suggestingthat if the Morales thrust extends this far east as a subhorizontal detachment, it is most 4. Middle Mountain, Parkfield likely underlain by crystalline basement. However,it is possiblethat sedimentsthrust to 8 km depth might be The seismic behavior of the Parkfield segment dif- only weakly conductiveand that the conductivesurface fers significantlyfrom the Carrizo segment.Here there layer (C1 and C2) maskswhat wouldthen be a rela- is extensive microseismicity,with the shallowestevents tively subtle feature in the MT data. There is alsono occurring in repeating clustersat depths of •3 km. Fo- clear evidence in either the resistivity model of Plate cal mechanisms,and the repeat sequence,suggest that 2a or the model of Mackie et al. [1997]for an east- seismicityis fluid driven [Nadeauet al. 1995; Johnson ward offset in the San Andreas Fault at 8 km depth, and McEvilly, 1995]. CharacteristicM ..•6 eventshave as suggestedby Davis et al. [1988].However, even if a ruptured this segment on averageevery 22 years since horizontal detachment is invisible to MT, the San An- 1857 [Bakun and Lindh, 1985]. In contrastto Carrizo dreas Fault at depth might be imaged electrically. The Plain the location of the fault at depth is seismically midcrustal conductor detected east of Carrizo Plain by well-defined and is closeto vertical to a depth of 8 km. deepsounding data of Biasi et al. [1990]might corre- Previous studies of the Parkfield segment using both

36 ø 00' 100- 10 Hz 10-1 Hz

Mean strike = 0 deg Mean strike = 2.4 deg

35 ø 58' ======

36 ø 00' I -0.1 Hz 0.1 - 0.01 Hz Mean strike = 5 deg Mean strike = -11.7 deg

35 ø 58' :::::::::::::::::::::::::::::::::::::....

120ø33' 120ø30' 120ø33' 120ø30'

0 1 2 3 km

Figure 8. Map of continuousMT profileat Parkfield,with arrowsshowing best fitting 2-D strike direction for each array in four frequency ranges. UNSWORTH ET AL.' SAN ANDREAS MT STUDY 1143

direct current (DC) electricalmethods [Park and Fit- terman,1990] and MT [Eberhart-Phillipsand Michael, 1.0 ImpedanceSkew for Parkfieldprofile 1993]suggest that electricalresistivities are low in and near this fault segment. The MT survey also revealed a broad(20+ km) widezone of low resistivityin the upper 0.8 5 km of the crust northeast of the SAF. This zone was correlated with a pronouncedseismic low-velocity zone [Eberhart-Phillipsand Michael, 1993]. Our profile at 0.6 Parkfield is only 4 km in length but is sampled densely enough to reveal significantlymore detailed images of shallowfault zoneresistivity structure than havebeen 0.4 available previously. An initial interpretation of these data wasgiven by Unsworthet al. [1997a],who showed that the fault zonewas coincident with a relatively nar- 0.2 row (500 m wide) verticalzone of low resistivity,which was attributed to the presenceof significant amounts of fluid. Here we provide a more detailed description of 0.0 the inversion results and explore the range of possible models that are compatible with the data. We will show l llllll I llllllll I llllllll I lllllul I llllllll I IIIIIIII in particular that the TE data imply that the zone of 100 10 1. 0.1 0.01 low resistivity extends to a depth of 2-4 km. Frequency (Hz) 4.1. Dimensionality and Distortion Analysis Figure 10. Impedance skew for all sites on the Park- Tensor decomposition was applied to the impedance field profile data, following the procedure describedin section 3.1, and results are presentedin Figure 8. The decomposition model fits the data quite well in each of the four 4.2. Modeling and Inversion frequency bands. Although there is some scatter in the The data were inverted using a sequencesimilar to directions at shorter periods, the estimates of geoelec- that described for the Carrizo Plain data. Inversion of tric strike for arrays west of the SAF are generally con- the TM data produces the model shown in Plate 3b. sistent and within +50 of the surface trace. East of the This model fits the TM data well, but does not come SAF the strike is less well-constrained as the TE and closeto fitting the TE data. To adequatelyfit both TM responsesare relatively similar, in contrast to the modes, the TM model was modified to improve the fit west where they are very different. The fit of the de- to the TE data. We found that a conductivelayer of composition model to the impedance data for a typical 1 9 m, 1 km thick fixed in the model at a depth of 2 site (site 21) is shownin Figure 9. The minimum rms km, and located 4 km to the east of the SAF solved misfit can be seen to be acceptable at all frequencies. this problem. This closelymatches the conductivezone The impedance skewsshown in Figure 10 are almost all east of the SAF mapped in previousMT surveysof this below 0.2. Thus for the bulk of the profile the decompo- area[Eberhart-Phillips and Michael, 1993]. This model sition and skew analysis suggestthat a 2-D interpreta- was used as a starting model for a joint TE+TM inver- tion will be reasonable over the whole frequency band. sion using RRI. The model and responseare shownin The TE and TM apparent resistivity and phase in this Plate 3c. Note that in the TE+TM model the fit to the fault-oriented co-ordinate system are shown in Plate 3a TM data has been improved,particularly in (I)TMeast in pseudo section format. of the fault tracedue to the additionof a resistivezone, R2. Again, the TM data image the contrastin resis-

Parkfield site 21 tivity acrossthe fault at depth and alsothe top of the large fault zone conductor. However,the bulk of the 100 fault zone conductor,C2, is invisibleto TM data and 10 is only imagedby the TE mode. The verticalstripes in 1 the computedPTE result from applicationof static shift 0.1 coefficients(Figure 11a) to the responses.Static shifts 0.01 are largerfor Parkfieldthan for CarrizoPlain, probably in part due to topographiceffects. Figure 11(b) shows 100 10 1. 0.1 0.01 that the rms misfit is uniform acrossthe profile, with Frequency(Hz) all sitesfit comparablywell. Figure 9. Fit of data to decompositionmodel for site As discussedfor the Carrizo Plain data, interpreta- 20 at Parkfield. tion of TE mode data must be undertaken with care 1144 UNSWORTH ET AL.. SAN ANDREAS MT STUDY

SAF SAF

ß mmmmmmmmmmmmmm • milli II1! IIIII!lt I11110 •m•

-1 km Cl C2

-2 km SAF

R1 ' 'R1 3 km

11 IIIIIIIIIIII I IIII 66 Hz i Apparent resistivity . . 0.004Hz ! '• tlHIIIIIItlIIII IIII IIH ! Itlltllilllllll i.l•.1tlitlllll I '• ' 66 Hz .-- Phase

0.004 Hz 'q'Iltllllltllllllllllllllllllll I alii I IllIll Ilti IIit111ttlt111 I

66 Hz I ! Apparent resistivity 0.004 Hz '"'11II111 Itl I t111111111111 " , II I .,, I 11 Itlt III111111111I •"• lilt I Ill I III I11111111111I 111t -. . 'rM 66 Hz

Phase

0.004 Hz I I I I - -2 -1 0 I -2 -1 0 I -2 -1 0 1 kmfrom surface trace km fromsurface trace kmfrom surface trace

(a) DATA (b)TM only (c)TE + TM rms misfit = 3.8 rms misfit = 1.7

Resistivity/ Apparent resistivity (ohm m)

300 100 10 I

15 45 65 Phase(degrees) Plate 3. (a) Datapseudo-sections for the ParkfieldMT profileand (b) modeland response obtainedby inverting the TM data.(c) Model and responses obtained by inverting and modeling bothTE andTM modedata. Models and topography are plotted-with no vertical exaggeration. UNSWORTH ET AL.: SAN ANDREAS MT STUDY 1145

[ [ I [ [ [ Apparentresistivity• TMApparentresistivity (b)

-2 km -1 km 0 km 1 km

Distance from surface trace (km) Figure 11. (a) EstimatedTE modestatic shift coefficients for Parkfielddata. (b) Misfit between data and predicted responseof model Plate 3c as a function of distance from the surface trace. as this mode can be more susceptibleto contamination see how sensitivethe data were to the depth extent of by 3-D effects than the TM mode. Since the inclusion the fault zoneconductor. Figure 12 showsthe responses of the TE mode data significantly effects the inversion at site 21 of models derived from the TE+TM model results for Parkfield, the possibility of 3-D effects needs shown in Plate 3c. The responseat this site is observed careful consideration here. There are several lines of at sites16 to 25 and is clearlynot a single-siteanomaly. evidencethat support a 2-D interpretation of this seg- In models 24 and 23 the conductor is terminated at shal- ment of the SAF. First, the higher-frequencydata which lower depths, resulting in a minimum in •T•. at around we emphasize in the inversion fit a distorted 2-D model 1 Hz and a rapidly rising pT•. which are not observed quite well, with a nearly fault-parallel geoelectricstrike. in the data. This is particularly true when the conduc- Impedanceskews are small (< 0.2) at all frequencies. tive zoneis truncatedat only 1 km depth (model24). Thus there is no evidence in the data itself for 3-D ef- Model 23, which extends the fault zone conductor to 2 fects. Furthermore,additional MT profiledata were km depth, gives a comparable fit to the model with a collected on Middle Mountain in late 1997. Profiles to conductorextending to 3 km. Extending the fault zone the northwest and southeast of the 1994 line interpreted to a depth of 10 km in the presenceof the eastern con- here showsimilar resistivityand phasedata [Unsworth ductor (model 25) produces•. that is too high and et al., 1997b]. Vertical magneticfield data can providea p•, that are too low. Thus the TE data indicate that useful indication of the connectivity of conductorsalong the strong conductor extends to a depth of 2-4 km but strike. Vertical field data were not collected in the 1994 does not continue to significantlygreater depth. Note survey but were collected in the 1997 survey. The in- that termination at 2 km is possible, but the best fit duction vectorsfrom this survey are generally normal to to this data occurs with --•3 km depth extent. As with the fault zone, again consistentwith a 2-D geoelectric the models consideredin Figure 7, the TM responsesof structure with fault-parallel strike. Furthermore, ver- these modelsare essentiallyindistinguishable. tical field data from a fault-normal profile some 3 km The requirement for the moderately resistive zone northwest of the 1994 profile closely match the vertical (R2) east of the SAF was examinedby replacingthe fields predicted by the 2-D model of Plate 3c down to resistor with a 10 S2m zone. This altered •M by • 50 0.2 Hz. All of these observation are consistent with a east of of the SAF and indicates our data are sensitive to 2-D interpretation, and none provide direct evidencefor this zone. If observedat just one MT site, this anomaly 3-D complications. Thus, while we cannot completely might be disregardedas a single-siteanomaly. How- dismiss the possibility that 3-D effects may bias quan- ever, in the continuousMT profiling data presentedin titative estimates of fault zone conductance, we believe this paper, we see such features in multiple sites. This that a 2-D interpretation incorporating TE phase data redundancyin spatial samplingallows small features in is not unreasonable in this case. the data to be interpreted with more confidence that Having given substantialevidence that the fault zone that obtained with widely spacedsites. is relatively 2-D, we will now demonstrate that a con- Our preferredmodel consistsof a conductingwedge ductor within the fault zone is required to fit the TE (C2) that pinchesout at around 2-4 km, with a weak mode data. Further forward models were computed to conductive root present below that depth. However, 1146 UNSWORTH ET AL.- SAN ANDREAS MT STUDY

ml iii illill iii illill iii illill iii 100 1. 0.1

Frequency(Hz)

, Model 21 Fault zone conductive to - 3 km inversion result of Plate 3c Models 23-25 producedby editing model 21

.... Model 25 Fault zone conductive to 10 km Model 23 Fault zone conductiveto only 2 km , Model 24 Fault zone conductiveto only 1 km

ß Model25 ß Model21

3.5 km I 1 3.5 km 3

10 7 km 7 km

Model 23 Model 24

lO 10

lO I 1 3.5 km I 1 3.5 km 10

10 10 7 km 7 km

Figure 12. TE responsesat dipole 21 for a set of modelswith the fault zoneconductor terminating at depthsof i km, 2 km, 3 km, and 10 km. Theseresults demonstrate that if the geoelectric structureis 2-D, then the low-resistivityzone beneath the surfacefault traceextends to a depth of 2-4 km. Resistivitymodels are shownwith no verticalexaggeration. UNSWORTH ET AL.' SAN ANDREAS MT STUDY 1147

the presenceof the deeperroot below 4 km is not con- The two models shown in Plate 3 could be consid- strainedby the MT data, as shownby resolutionanal- ered end-members of a spectrum of models consistent ysisusing synthetic MT data [Eberhart-Phillipset al., with the TM data. Depending on how much weight 1995]. Low resistivitiesbelow 4-5 km in our modelat is given to the TE data, the fault zone conductor will depth in the fault zone could be an artifact causedby have varying conductance. If the TE data are consid- vertical smoothingof the regularizedinversion. ered unreliable, then it will be restricted to the upper kilometer. However, if the TE data are deemed reli- 4.3. Interpretation of the Middle Mountain able and can be interpreted in a 2-D context, a fault Model zone conductor extending to a depth of 2-4 km is re- A detailed interpretation of the near-surfaceportion quired by the data. Note once again that a conductor of the model has been givenpreviously by Unsworthet terminating as shallow as 2 km on Middle Mountain is al. [1997a].This showedthat the fault zonewas char- a possibility. Both models in Plate 3 have related hy- acterized by a positive flower structure with a strong drogeologicalinterpretations. In the first, fluid flows to central conductor, interpreted as a fluid saturated zone the surface along the interface between the Franciscan, of breccia at the core of fault. In this paper we have while in the second, fluids accumulate in a wide dam- reexamined the prominent fault zone conductor more aged zone of high porosity produced by fracturing of carefully and shown that the TE data are consistent country rock. with a conductive fault zone that pinches out over the The earthquake hypocentersin Plate 2b appear to lie depthrange 2-4 km (assumingthat the fault zoneis suf- on the west side of the fault zone conductor. However, ficiently2-D to allow interpretationof the TE mode). whenthese events were relocated [Ellsworth, 1996], they Given the nonuniquenessinherent in the inversion of were found to lie some 200 m east and 400 m deeper magnetotelluricdata, variants on this model may also than the locations(from the BerkeleySeismic Station) reproducethe data. Clearly, alternate conductivitydis- shown in Plate 2b. Thus it is unclear whether the mi- tributions may be possibleeast of the profile, given the croearthquakesare in the center or on the edge of C2. lack of data in this area. The following major features In either case they are located within a conductive, pre- are requiredby the TM mode data: (R1) a resistive sumably high fluid content fault zone, which stands in blockwest of SAF and (C1) a conductivezone east ex- contrast to the seismically inactive and comparatively tendingeast of the profile. Thesefeatures can be con- more resistive fault zone at Carrizo Plain. sidered a minimum structure model that is required by The depth extent of the fluid saturated zone implied the TM data. However, as we have shown, 3-D effects by the model of Plate 2b agrees well with the model appear to be small. Thus the combinedTE and TM of Andersonet al. [1987]based on the mappingof ex- mode data can be used to add a resistive zone east of humed faults. In both, the zone of breccia terminates the fault (R2) and a verticalconductor in the fault zone around 5 km depth. The MT image is also consistent (C•). with recent seismictomography data which showsa 500 A revisedinterpretation, basedon comparisonto the m wide seismiclow-velocity zone extending to a depth geologicalmap of Sims [1990],is shownin Plate 2b on of around4 km beneathMiddle Mountain(C. Thurber, the same scale as the Carrizo Plain interpretation. The personalcommunication, 1997). The presenceof fluid unit R1 is interpreted as being granite of the Salinian lowersboth electrical resistivity and the seismicvelocity block. The fault that bounds the east side of C2 is and is a natural explanation for the structures imaged clearlythe Gold Hill Thrust Fault. ResistorR2 is coin- by both datasets. cident with the anticline between the Parkfield syncline and the Gold Hill Thrust Fault. Is the relatively high re- Discussion and Conclusions sistivityof this feature reasonablegiven the high salinity of fluids that are widespreadin the Franciscanforma- On both MT profiles we have imaged narrow fault tion [Eberhart-Phillipsand Michael,1993]? Unsworth zone conductors extending to depths of 2-4 km and et ai. [i997a] usedArchie's law to estimatethe poros- sharp variations of resistivity gcrossthe fault contact. ity requiredto accountfor the low resistivityin the core The high-density electric field sampling has also allowed of the fault zone. At a depth of 3 km, a porosity in the us to resolve many details of near-surface resistivity range of--•10-30% was requiredwithin the fault zone structure,along with variationsin depth to the resistive to accountfor the low resistivity on the basis of fluids basement. Although there are broad similarities in the alone. Assumingthe samepore fluid salinity,porosities two resistivity models of Plates 2a and 2b, there are also in R2 would be in the range 1-10%, where upper and somevery significantdifferences. These include (1) the lower limits correspondto exponentsof 2 and 1, respec- contrast of resistivity acrossthe fault; at Carrizo this tively, in Archie'slaw. Sims[1990] shows that the area is small with 1000+ i2 m material on each side, and at denotedas (R2) in our interpretationhas Franciscan Parkfield there is a west to east changefrom 1000 to 30 formation at its core at a depth of .-• 2 km. Given the i2 m; (2) the magnitudeof the low resistivityanomaly heterogeneityof the Franciscanformation the inferred in the fault zone; at 3 km depth fault zone conductance porosity is not unreasonable. at Carrizo Plain is -,•20 S, while at Parkfield our analy- 1148 UNSWORTH ET AL.: SAN ANDREAS MT STUDY sis suggestsa conductanceof approximately 200 S; and The possibilitythat seismicityis controlledby fluids (3) the width of the damagedzone at the coreof the in the fault has been proposedby many people. The fault (as inferredfrom resistivitycontrasts) is lessthan correlation of fluid flow through the fault with abun- 300 m at Carrizo Plain and around 600 m at Parkfield. dant seismicity and creep was initially suggestedby These segmentsalso have very different patterns of Irwin and Barnes [1975]. They proposedthat these seismicity.Even though we are not imagingfault struc- fluids originated in metamorphic reactions within the ture to the depth of the M • 6 eventsat Parkfield or the Franciscanformation and that fault creepoccurs where larger events that may periodically rupture the Carrizo Franciscanrocks abutted the fault and provideda ready segment,our resistivity data are samplingdepths where supply of fluids. While we have consideredonly two lo- shallow seismicity is present at Parkfield but absent at cationson the San Andreas Fault system,our findings Carrizo Plain. How might the differencesobserved in are consistentwith this hypothesis. the resistivity sections be related to the differencesin The MT data for Parkfield and Carrizo Plain do not seismicity? directly constrainstructure at depths where the large First, could difference(1), the contrastin structure earthquakesthat periodicallyrupture this segmentnu- acrossthe fault, contribute to the observedpattern of cleate. If the hypothesisof Irwin and Barnes [1975]is seismicity? Our results show that at Carrizo Plain, correct and fluids enter at depths below 10 km and then resistive, presumably crystalline, rocks are present on flow to the surface, the high concentrationof fluids in both sidesof the fault. This is supportedby the results the near-surface at Parkfield may be indicative of an of Castillo and Hickraan [1995],who usedstress analy- ample fluid supply at depth. Alternatively, the fluids sis to show that the Carrizo segmentis locally strong, may be entering the fault at shallowdepths and be un- in contrast to the general weaknessof the fault docu- related to the characteristicearthquakes. The resistor mented by Zobacket al. [1987]. While the top of the east of the SAF (R2) doesnot necessarilyrepresent a resistive zone is deeper on the west side of the fault than barrier to fluid flow, rather just a zone of lower poros- on the east, these mechanically strong blocksmay give ity with the Franciscan formation. At Carrizo Plain the fault zone increased strength, compared to other the absenceof fluids in the near-surfacemay indicate a parts of the San Andreas Fault. At Parkfield the situa- relatively dry fault zone at depth. tion is quite different with a resistive block of Salinian In reality, a combination of factors is probably re- granite west of the fault and an electrically conductive, sponsible for the differences in seismic behavior of the presumablyfluid rich, and possiblymechanically weaker two fault segments.We have shownthat there are ma- Franciscancomplex to the east. Weak materials east of jor differencesin geoelectricstructure between the Park- the fault could limit the strength of the fault and result field and Carrizo segments. Our results are consistent in a cycle of more frequent, smaller earthquakesthan is with previoussurveys of both fault segments,but close observed at Carrizo Plain. to the fault, higher resolution of shallow structure is Alternatively, it could be difference2, the properties facilitated by dense spatial sampling. Subject to the of the fault zone, specificallythe fluid distribution, that absenceof 3-D effects, our interpretation of the data are controlling seismicity. At Middle Mountain, Park- is that the San Andreas Fault at Parkfield has a high- field, a number of lines of evidence suggestthat both porosity core, while at Carrizo Plain the fault is essen- shallow clusters of microearthquakes, and the charac- tially dry. Significantly,microearthquakes occur in the teristic deeper M • 6 events are fluid driven. Johnson shallow fault zone at Parkfield but not at Carrizo Plain. and McEvily [1995]suggested that earthquakeclusters Given careful analysis, we believe that MT exploration at Parkfield had focal mechanisms consistent with fluid can provide valuable new constraints on the physical cycling. At the depth of the characteristicM .• 6 events conditions within active fault zones. a zone of low Vp and high Vp/V• consistentwith the presenceof excessfluid has been reported [Michelini Acknowledgments. Fieldwork was made possible by and McEvilly, 1991]. In contrastto Middle Mountain, WangFei, YangYangjing, Weerachai Siripunvaraporn, Nong Carrizo Plain is characterized by an almost complete Wu, Xinghua Pu, Zhiyi Zhang, Yuval Zudman, and Ar- nell Dionosio. The staff of the BLM in Bakersfield,no- absenceof seismicity. In this paper we have shown that tably JohnaCochran, Judi Weaserand Larry Vredenburgh, the conductanceof the shallow(0-4 km) fault zoneat are thanked .for administrative and logistic support. The Middle Mountain could be at least an order of magni- Nature Conservancyis thanked for letting us stay at the tude greater than on the quiescentCarrizo Plain seg- PaintedRock Ranch on CarrizoPlain, andTony and George were most helpful hosts. Many private landowners at Car- ment. This difference may be partly explained by fluid rizo Plain are thanked for accessto their land, as are chemistry( i.e., the high salinity of 30,000 ppm chlo- Kevin Kester and Jack Varian at Parkfield. Tom Bur- ride at Middle Mountain would produce a lower elec- dette is thanked for help with permitting. This work was trical resistivitythan a lowersalinity at CarrizoPlain). supportedby NSF grant EAR-93-16160 and U.S. Geologi- However, even with a high salinity of 30,000 ppm at cal SurveyNational Earthquakehazard reductionprogram Parkfield a porosity of .-• 10% was required to explain grant 1434-94-6-2423. The developmentof the magnetotelluric inversioncode was supportedby U.S. Department of the low resistivity. Thus we believe our data suggest Energy grant DE-FG06-92-ER14231. Developmentof data higher porosities in the San Andreas Fault at Parkfield processingcode was supported by U.S. Department of En- than at Carrizo Plain. ergy grant DE-FG06-92-ER14277. The authors acknowl- UNSWORTH ET AL.: SAN ANDREAS MT STUDY 1149 edge many helpful discussionswith Randy Mackie, Peter Egbert, G.D., and J.R. Booker, Robust estimation of geo- Malin, and membersof the Parkfield working group. The magnetic transfer functions, Geophys.J. R. Astron. Soc., manuscriptbenefited from reviewsby StevePark, Dal Stan- 87, 173-194, 1986. ley, and George Jiracek. Ellsworth, W.L., Drilling into the Earthquake source: The search for a site to test theories of Earthquake Gener- References ation at Parkfield,paper presented at 8tn International Anderson, J.L., R.H. Osborne, and D.E. Palmer, Cataclastic Symposium on the Observation of the Continental Crust rocks of the San Gabriel fattit- an expression of deforma- ThroughDrilling, Geol. Surv. of Jpn.,Tsukuba, 1996. tion at deeper crustal levels in the San Andreas Fattit, Fishburn, M.D., Results of Deep Drilling, Elk Hills Field, Tectonophysics, 98, 209-251, 1987. Kern County, California, in Structure, Stratigraphy and Arrowsmith, J.R., Coupled tectonic deformation and geo- Hydrocarbon Occurrences of the San Joaquin Basin, Cal- morphic degradation along the San Andreas Fault sys- ifornia, edited by G.K. Kuespert and S.A. Reid, pp. 157- tem, PhD. dissertation, 347 pp., Stanford Univ., Stanford, 167, Pac. Sect., Am. Assoc. of Pet. Geol., Bakersfield, Calif., 1995. Calif., 1990. Bakun W. H., and A. G. Lindh, The Parkfield, California, Gamble, T.D., W.M. Goubau, and J. Clarke, Magnetotel- earthquake prediction experiment, Science, 229, 619-624, lurics with a remote reference, Geophysics, JJ, 53-68, 1979. 1985. Biasi, G.P., S.K. Park, R.L. Mackie, and T.R. Madden, Graff, L.B., Exploration drilling in the Carrizo Plain, in Magnetotelluric studies in central California, paper pre- Guidebook to the Geology of Carrizo Plain and San An- sented at IAGA 10th Induction Workshop, Int. Assoc. of dreas Fault, p. 21-24, San Joaquin Geol. Soc., Bakersfield, Geomagn. and Aeron., Ensenada, Mexico, 1990. Calif., 1962. Brune, J.N., T.L. Henyey, and R.F. Roy, Heat flow, stress, Hickman, S., M.D. Zoback, L. Younker, and W. Ellsworth, and rate of slip along the San Andreas Fattit, California, Deep scientificdrilling in the San Andreas Fault zone, Eos J. Geophys. Res., 74, 3821-3827, 1969. Trans. A G U, 75, 137, 140-142, 1994. Byedee, J., Friction of rocks, Pure Appl. Geophys., 116, Hill, D.P., J.P. Eaton, and L.M. Jones, Seismicity 1980-86, 615-625, 1978. in The San Andreas Fault system, California, edited by Byedee, J., Friction, overpressureand fault normal compres- R.E. Wallace, U.S. Geol. Surv. Prof. Pap., 1515, 115- sion, Geophys. Res. Lett., 17, 2109-2122, 1990. 151, 1990. Byerlee, J., Model for episodic flow of high-pressure water Hubbert, M.K., and W.W. Rubey, Role of fluid pressure in in fault zones before earthquakes, Geology, 21, 303-306, mechanicsof overthrust faulting, Geol. Soc. Am. Bull., 1993. 70, 115-166, 1959. Castillo, D.A., and S. Hickman, Near field stress and pore Irwin, I.P., Geology and plate tectonic development,in The pressureobservations along the Carrizo Plain segmentof San Andreas Fault System, California, edited by R.'E. the San Andreas Fault, Eos Trans. A GU, 76(46), Fall Wallace, U.S. Geol. Surv. Prof. Pap., 1515, 61-80, 1990. Meet. Suppl., F360, 1995. Irwin, P.I., and I. Barnes, Effect of geologicstructure and Chave, A.D., and J.T. Smith, On electric and magnetic gal- metamorphic fluids on seismic behavior of the San An- vanic distortion tensor decompositions,J. Geophys.Res., dreas fault system in central and northern California, Ge- 99, 4669-4682, 1994. ology, 3, 713-716, 1975. Chester, F.M., J.P. Evans, and R.L. Biegel, Internal struc- Johnson, P. A., and T. V. McEvilly, Parkfield seismicity: ture and weakeningmechanisms of the San AndreasFault, Fluid driven?, J. Geophys.Res., 100, 12,937-12,950, 1995. J. Geophys. Res., 98, 771-786, 1993. Lachenbruch, A.H., Frictional heating, fluid pressure and Colburn, R.H., and W.D. Mooney, Two-dimensionalveloc- the resistance to fault motion, J. Geophys.Res., 85, 6097- ity structure along the synclinalaxis of the Great Valley, 6112, 1980. California, Bull. Seismol. Soc. Am., 76, 1305-1322, 1986. Mackie, R.L., B.R. Bennett, and T.R. Madden, Long period Davis, T.L., M.B. Lagoe, W.J.M. Bazeley, S. Gordon, K. magnetotelluric measurementsnear the central California Mcintosh and J.S. Namson, Structure of the Cuyama Val- coast: A land-locked view of the conductivity structure ley, Caliente Range and Carrizo Plain and its significance under the Pacific Ocean, Geophys.J. R. Astron. Soc., 95, to the structural style of the Southern Coast Ranges and 181-194, 1988. Western Transverse Ranges, in Tertiary Tectonics and Mackie, R.L., D.W. Livelybrooks, T.R. Madden, and J.C. Sedimentation of the Cuyama Basin, San Luis Obispo, Larsen, A magnetotelluric investigation of the San An- Santa Barbara and Ventura Counties, California, edited dreas fault at Carrizo Plain California, Geophys. Res. by W.J.M. Bazeley, pp. 141-158, Pac. Sec. Soc. of Econ. Lett., 2•, 1847-1850, 1997. and Paleontol. Mineral., Bakersfield, Calif., 1988. Michehni, A., and T.V. McEvilly, Seismologicalstudies at Dihhlee, T_W, Regional geologicmap of San Andreas and Parkfield, I, Simultaneous inversion for velocity structure related faults in Carrizo Plain, Temblor, Caliente and La and hypocenters using B-splines parameterization, Bull. Panza ranges and vicinity, Cahfornia, U.S. Geol. Surv. Seismol. Soc. Am., 81,524-552, 1991. Misc. Field Invest. Map, 1-757, 1973a. Nadeau, B., W. Foxall, and T.V. McEvilly, Clustering and Dibblee, T.W., Stratigraphy of the Southern Coast ranges periodic recurrence of microearthquakes on the San An- near the San Andreas Fault from Cholame to Maricopa, dreas Fault at Parkfield, California, Science, 267, 503-507, Cahfornia, U.S. Geol. Surv. Prof. Pap., 76•, 1973b. 1995. Eberhart-Phillips, D., and A.J. Michael, Three-dimensional Page, B.M., The southern Coast Ranges, in The Geotec- velocity structure, seismicity,and fault structure in the tonic Development of California, edited by W.G. Ernst, Parkfield Region, Central California, J. Geophys.Res., pp. 329-416, Prentice-Hall, Engelwood Cliffs, N.J., 1981. 98, 15,737-15,578, 1993. Park, J., and A.D. Chave, On the estimation of magnetotel- Eberhart-Phillips, D., W.D. Stanley, B.D. Rodriguez, and luric response functions using the singular value decom- W.J. Lutter, Surface seismic and electrical methods to position, Geophys. J. R. Astron. Soc., 77, 683-709, 1984. detect fluids related to faulting, J. Geophys.Res., 100, Park, S. K. and D.V. Fitterman, Sensitivity of the telluric 12,919-12,936, 1995. Monitoring array in Parkfield, California, to changesin Egbert, G.D., Robust multiple station magnetotelluricdata resistivity, J. Geophys. Res., 95, 15,557-15,571, 1990. processing,Geophys. J. Int.,130, 475-496, 1997. Park, S.K., G.P. Biasi, R.L. Mackie and T.R. Madden, Mag- 1150 UNSWORTH ET AL.: SAN ANDREAS MT STUDY

netotelluric evidence for crustal suture zonesbounding the Unsworth, M. J., P. E. Malin, G. D. Egbert and J. R. southern Great Valley, California, J. Geophys. Res., 96, Booker, Internal structure of the San Andreas Fault at 353-376, 1991. Parkfield, California, Geology,25, 359-362, 1997a. Ranganayaki, R. P., and T. R. Madden, Generalized thin Unsworth,M.J., P. Bedrosian,G.D. Egbert and M. Eisel, 3- sheet analysis in magnetotellurics: An extensionof Price's D electrical resistivity structure of the San Andreas Fault analysis, Geophys. J. R. Astron. Soc., 60, 445-457, 1980. at Parkfield, California,Eos Trans. A GU, 78 (46), Fall Rice, J. R., Fault stress states, pore pressure distributions, Meet. Suppl., F696, 1997b. and the weaknessof the San Andreas fault, in Earthquake Vedder, J.G., Geologicmap of the Wells Ranch and Elkhorn Mechanisms and Transport Properties of Rocks, edited by Hills quadrangles,San Luis Obispo and Kern Counties, B. Evans, and T. Wong, pp. 475-503, Academic, San California, showingjuxtaposed Cenozoicrock along the Diego, Calif., 1992. San Andreas fault, U.S. Geol. Surv. Misc. Field Invest. Ross, D.C., Possible correlations of basement rocks across Map, 1-585, 1970. the San Andreas, San Gregorio-Hosgri and Rinconada- Wannamaker,P.E., G.W. Hohmannand S. Ward, Magne- Reliz-King City faults, California, U.S. Geol. Surv. Prof. totelluric responsesof the three-dimensionalbodies in lay- Pap., 1317, 1984. ered earths, Geophysics,49, 1517-1533, 1984. Ryder, R.T., and A. Thomson, Tectonically controlled fan Willcock, W.S.D., G.M. Purdy, and S.C. Solomon, Mi- delta and submarine fan sedimentation of late Miocene croearthquake evidence for extension across the Kane age, southern Temblor Range, California, U.S. Geol. Surv. Transform fault, J. Geophys. Res., 95, 15,439-15,462, Prof. Pap. 1442, 1989. 1990. Sibson, R.H., R. Francois, and K.H. Poulsen, High-angle re- Zoback, M.D., et al., New evidence on the state of stress of verse faults, fluid pressurecycling and mesothermalgold- the San Andreas Fault system, Science, 238, 1105-1111, quartz deposits, Geology, 16, 551-555, 1988. 1987. Sims, J.D., Geologic map of the San Andreas Fault in the Parkfield 7.5-minute quadrangle, Monterey and Fresno Counties, California, U.S. Geol. Surv. Misc. Field Study Map, MF- 2115, 1990. Sleep, N.H. and M.L. Blanpied, Creep, compaction and the J.R. Booker and M.J. Unsworth, Geophysics Pro- weak rheology of major faults, Nature, 359, 687-692, 1992. gram, University of Washington, Seattle, WA 98195. Smith, J. T., and J. R. Booker, Rapid inversion of two- and (unsworth@ geophys. washingto n. edu) three-dimensional magnetotelluric data, J. Geophys.Res., G.D. Egbert, College of Oceanic and Atmospheric Sci- 96, 3905-3922, 1991. ences, Oregon State University, Corvallis, OR 97331-5503. Torres-Verdin, C., and F.X. Bostick Jr., Principles of spatial (egbert@oce. orst. edu). surface electric field filtering in magnetotellurics: Electro- magneticre'ray profiling (EMAP), Geophysics,57, 603- (ReceivedMay 23, 1997; revisedFebruary 13, 1998; 622, 1992. acceptedMay 14, 1998.)