ence that this feature may represent an dial) around the angle of incidence of the P wave. The analysis we first extracted the group velocity disperdisper- evolving intrusive boundary at the top of a Qo component is perpendicular to the incoming P sion curve of the fundamental mode by applying a phase in the plane of incidence and contains mostly Gabor Transform to the recorded Rayleigh wave partial-melt zone. SV energy and little P-wave energy. Source equalequal- train. The fundamental mode Rayleigh waveforms ization was accomplished by deconvolution of the Q0 were isolated from the recorded Rayleigh waves by component with the P wave of the L component. applying a phase match filter. REFERENCES AND NOTES 3. R. Kind, G. L. Kosarev, N. V. Peterson, Geophys. 8. J. Wu et al., Eos 76, F392 (1995). J. Int. 121,121,191 191 (1995). 9. L.-S. Zhao and C. Frohlich, Geophy. J. Int. 124,525124, 525 1.1.K. K. D. Nelson etef al., ScienceSCienceSc/ence 274, 16841684 (1996). 4. N. A. Haskell, J. Geophys. Res. 67,475167, 4751 (1990). (1996); L.-S. Zhao, M. K. Sen, P. Stoffa, C. Frohlich, 2. Before processing the data the ground displacement 5. Trace A36 results from averaging receiver functions ibid. 125,355125, 355 (1996). was restored at all broadbroadband band and short period stasta- 10. We thank I.1. Billings, M. Brunner, A. Fabritius, S. tions. Each seismogram was rotated from the stanstan- from BB10,BBlO, BB36,BB36. and SP12 and trace A18 is an averaging of receiver functions from stations BB34, Klemperer, Y. Makovksy, K. Wu, and Chinese scienscien- dard vertical, north and east coordinate system into a tists in helping to collect the seismic data used in this ray system [Longitudinal (L) and T componentslcomponents] us BB18, and BB20. ray system [Longitudinal (L) and T components] us- study. This research was funded by the National ing the eigenvalues of the covariance matrix for the 6. L. D. Brown et al., Science 274, 1688 (1996). Science Foundation Grant NSF EAR-9316814. computation of the rotation angles at a time window 7. The two-station phase velocity dispersion of the funfun- following the P-wave arrival. The LandL and Q0 compocompo- damental mode Rayleigh wave was computed using nents result from a rotation of Z (vertical) and R (ra- programs developed by R. Herrmann. In the data 2 August 1996; accepted 5 November 1996
Electrically Conductive Crust in Southern Tibet For both lines, the phase difference be-be Electrically Conductive Crust in Southern Tibet tween the electric and magnetic fields, in from INDEPTH Magnetotelluric Surveying both polarizations of induction (9), was greater than 45°450 at all but a few sites and Leshou Chen, John R. Booker, Alan G. Jones,* Nong Wu, increased with increasing period. This result Martyn J. Unsworth, Wenbo Wei, Handong Tan indicates that conductivity increases in the crust (resistivity decreases) with increasing The crust north of the Himalaya is generally electrically conductive below depths of 10 depth (10). Apparent resistivities were low, to 20 km. This conductive zone approaches the surface beneath the Kangmar dome <<10 10Ion· nfQ . m, at frequencies less than 0.1 Hz, (dipping north) and extends beneath the Zangbo suture. A profile crossing the northern and for some sites the apparent resistivities Yadong-Gulu rift shows that the high conductivity region extends outside the rift, and its were'were below 1 m for frequencies less than Yadong-Gulu rift shows that the high conductivity region extends outside the rift, and its were below 1 nl .* m for frequencies less than on June 7, 2010 top within the rift coincides with a bright spot horizon imaged on the INDEPTH CMP 0.01 Hz. Such pervasively low values are (common midpoint) profiles. The high conductivity ofofthe the middle crust is atypical of stable atypical in the continents; shield regions continental regions and suggests that there is a regionally interconnected fluid phase in have apparent resistivities that are more the crust of the region. than two orders of magnitude higher. We modeled the MT data using 202D inin- version algorithms that simultaneously searched for the smoothest as well as bestbest- The INDEPTHIN DEPTH magnetotelluric (MT) inin- terns:tems: a five-component commercial widewide- fitting models ((11) 11) (Figs. 1 and 2). The data vestigation undertaken during April to July band system (Phoenix VS) for shallow probptob can be fit by more complex models, but vestigation undertaken during April to July band system (Phoenix V5) for shallow prob- can be fit by more complex models, but www.sciencemag.org of 1995 was designed to study the electrical ing, and 20 five-component recording syssys- models with less structure result in unacunac- structure of the lithosphere beneath southsouth- tems (GSC LIMS) for deeper penetration ceptable misfits. ern Tibet. Two earlier MT studies were (5, 6). During the period of LIMS acquisiacquisi- The model for the lOO-line100-line is based on carried out in the region: A Sino-French tion, 24 March to 31 July 1995, sunspot the MT and vertical-field transfer-function group collected MT data in southern Tibet activity was low, resulting in poor signal. We data at 13 frequencies, from 80 to 0.0015 in the early 1980s (I),(1), and, subsequently, compensated for this by recording for a longlong- Hz, from all sites, and fits the data with an MT data were collected by Chinese invesinves- er interval at each location than originally average RMS misfit of 10°100 in phase, 5% in
tigators as part of the Golmud to Yadong planned (4 to 5 weeks instead of 2 to 3 apparent resistivity, and 0.10 in GDS transtrans- Downloaded from Global Geoscience Transect activities (2). weeks), resulting in fewer total sites than fer function. These misfits are high, but The ININDEPTH DEPTH MT study provided closely planned for the experiment. most misfit is concentrated on data from spaced sites and a wide frequency bandband- Distortion analysis (7) was applied to the the southernmost four sites, which were width compared to these surveys, and subsub- MT response estimates, both in single-fresingle-fre- found impossible to model because their stantiated several of their conclusions,conclusions. quency, single-site and in multi-frequency, responses were inconsistent between the We acquired MT data along two lines-a multi-site modes, to determine the dimendimen- two modes. main north-south line that extended fromftom sionality of the data and derive the domidomi- The model exhibits the following firstfirst- the crest of the Himalaya to near Yangbajain nant electric strike direction. The distordistor- order features that are apparent in the data, in the Lhasa block [lOO-line,[100-line, figure 1 of (3)], tion models fit the data well, which implies appeared in all inversions for different frefre- and a northwest-southeast trending line that that a two-dimensional (20)(2D) description of quency and site subsets and different assumpassump- obliquely crossed the northern Yadong-Gulu regional structures is a reasonable assumpassump- tions about static shifts (12), and were not rift near Damxung (200-line) (4). The MT tion.tion, The electric strike directions at frefre- influenced by the abherent responses just data were recorded with the use of two sys- quencies sampling the thick Tibetan crust mentioned: (i) From approximately 150 km beneath the lOO-line100-lineWO-line are frequency-indefrequency-inde- south of the Zanbo suture to the north end of Leshou Chen, Wenbo Wei, Handong Tan, China UniverUniver- pendent and east-west for most sites-parsites-par- the line, the crust is generally electrically sity of Geosciences, Beijing, China. allel to the regional surface geologic strike. conductive below depths of 10 to 20 km J. R. Booker, Nong Wu, M. J. Unsworth, School of GeoGeo- This observation suggests that the surface «200(<200 n·fl m). (ii)(iil North of Kangmar the physics, University of Washington,Washin9ton, Seattle, WA 98195, USA. geologic strike in the region is representarepresenta- crust becomes extremely conductive at A. G. Jones, Geological Survey of Canada, 1 Observatory tive of structure throughout the thickness of depth (generally much less than 30 n·5i * m; Crescent, Ottawa, Ontario, Canada K1AK1 A OY3.0Y3. the crust sensed by the MT survey, which is "3" in Fig. 1). (iii) A north-dipping zone of •* To whom correspondence should be addressed. not always the case in orogenicomgenic belts (8)(8). . high conductivity extends upward to the
16941 694 SCIENCE •* VOL. 274 •* 6 DECEMBER 1996 I;JAUlniS S ..i near-surfacenear-surface beneathbeneath the Kangmar dome tivity with depthdepth ooverver mostmost ooff thethe survey,susurvrvey.vey, dodomeme suggests thatthat partially moltenmolten materi mmateri-ateri ("Z"("2" in Fig.Fig. 1)1).. At depthdepth thisthis zonezone appears to and in particular the midmidcrustalcrustal resistiviiresistivitiesresistivitiesties al has been trtransportedansported relativelyrelatively upward coconnectnnect with the midcrustalmidcrustal zonelOne of high ooff about 1I to 10to fln· m nnorthorth ofof KangnKangmar, Kangmar,nar, beneabeneathth the north flank ooff thethe Kangmar coconductivitynductivity toto the north.north. (iv)(i v) The upper-upper substantiatessubsmmiates the oobservationbservation tthathat thisthis rere- antidinoriumanticlinorium.. This might eithereither be asas an most ccrustrust beneath the ZangboZangbo suturesuture coconcon-n giongion has has high crustacrustalcrustal I conductivityconductivity (J(1(1, ,,1),2).2).. intrusionintrusion or transpotransportrt in the hanging wallwall ooff tainstains a na nnarrowarrow eleelectricallycnically conductive body The midcrustal coconductivitynductivity is so hhighigh a thrust.thrust. Alternatively,Alternatively, sasalineline fluid trapped ("I("1"" in Fig.Fig. 1).1). The ssusutureutureture hhasas nono electr!calelectrical northnorth ooff KangmarKangmar that we are unable to in porous subducted sediments isis also a expexpressionression belowbelow a depth ooff a few kilometers.kilometers. characterizecharacterize the cconductivityonductivity ofof thethe lowlowerlowerver-+ candidate forfor ththisis ananomalyomaly ((16)(16). J 6).. ThThee modelmodel forfor the 200ZOO200-line-line-line data ccrossingrossing most crust oror underlyinunderlyingg mantle ooff thethe rere-re+ ThThee oobservationbservation from the 200-line2oo+ line tthathat ththee northernnorthern Yadong-GuluYadong-Gulu riftrift fitsfits the llonglong-ong ggion.ion. As dry rocksrocks ttypicallyypically have rresistiresistiviesistivi coconductivenductive middle ccrustrust exextendstends beyondbeyond period (frequencies <<1 1 Hz)Hz) MT ddataata with tiesties in the hundredshundreds to thousandsthousands of nfQ -.* m the surface expexpressionress ion ooff the Yadong-GuluYadong-Gulu an average misfitmisfit of 2022~0 in phase,phase, and eequivequiv-quiv (14),(1 4), the low midcrustalmidcru stal resistivityresistivity northnorthh ofof rift (Da(Damxungmxu ng ggraben/Nyainqentanglharaben/Nyainqemanglha alentalem levellevel in apparent resistivity.resistivity. ThThee modmod-mod aboutaboul KKangmarangmar ssuggestsuggests that aann interconintinterc(erco~on-n range)range ) impliimplieses that thethe mid-crustalmid-crustal partial el sshowshows the following ((Fig.Fig. 2): (i) ElectricalElectrical nectednected flufluidid phasephase iiss presentprese nt withinwithin the melt extendextendss ooutsideutside tthehe immediateimmediate concocon-n coconductivitynductivity is hhighig h belowbelow depthdepthss of 10iO to crust.crust. Given that hheateat flowflow in this regiorrregionegionn iis1Ss finfineses ooff thethe rift.rift. Finally,Finall y, tthehe sshallowhallow highhigh- 20 km acracrossoss ththee eentiremire profileprofile «200(<200 m). hihighestghest nearnear the latitudlatitudee where Ihethe coconduccondnducluc- conductconductivityivity feature imagedimaged on the 200-line2oo-line (i(ii)il The condunconductiveive middlemiddle crust contains tivetive aanomalynomaly is closest to the surfacesurface ((242(2242242 ("4" in Fig.Fig. 2) suggestss ugg~t S thatthat deep hydrotherhydrhydrother-other severalseveral zones of highhigh conductivityconductivity at depthsdepths mW/mmW/mi2)l2) (15),(15), the presence ofcif partial meltmaelt malmal fluids are presentpresent deep beneathbeneath Nam belowbelow depthsdepths ooff 15 toto 20 km ((fewfew nn·fl · m;m; offersoffers a readyready explanationexplanation for the data thatt]'hat TTsoso {pe(perhapsrhaps circulatingcirculating to depth alalongong a "1","I ", "2", aandnd "3" in FigFig.. 2). The ttopOp of the doesdoes nnotot requirerequire unusualunusual conditionsconditions for coconc on-n normalnormal fault).fau lt}. ccenterenter oonene ("2" in Fig.Fi g. 2) coincidescoincides within tinental ccrustcrust.ru st.. resolutionresolution with the YDR bright spospott horihori- The observationobservation thatthat the ZangboZangbo ssuturesutut:ureure REFERENCES AND NOTES zonzon imaged oonn the INDEPTHIN DEPTH CMP profileprofile zonezone anomalyanomaly isis limitedlimited ttoo the upper fewffew 1.1. Pham Van Ngoc etar al.,Ill. . NarUt8Nature 3311,615311,11 ,615615 (1986)(1986).. [[seesee figure 1 ooff (13)(13)).(13)].]. (i(iii)ii) A sshallowhallow sosouthsouth-uth kilometerskilometers ooff the ccrustru st ssuggestsuggests thethe midmmidlid- 2.2. Guo XinfengXinfeng,Xinleng, Zhang Yuanchou.Yuanchou,Yuanchou, Cheng Qing,-ul,QingyunQingyun,, GaoRUIGao Rui, Pan vuVu , Bull.BUll. ChinC/l;l.. AcacIAcad. Geol.GeoI. Sci. 221,1. 191 easeast-dippingt-dipping zone of enhancedenhanced conductivityconductivity ccrustalrustal partial melt zozonene iiss superposesuperposedd acracross055soss Gao(1990)Rui,(InPanChineseYu, Bull.withChin.EnglishAcad.abstract).Geol. Sci.See21,191 east-dipping zone of enhanced conductivity crustal partial melt zone is superposed acr (1990) (111(In Ch'"'eseChinese with EngIist1English abstract). seeSee Wu extends from tthehe nearnear surfacesurface near Nam the ssutureuture at depth or that the sutusuturere is cuti Gong-jian eter alal.,a/.,.. Chin.CM. J. Geophys.Gooph)'S. 35.35, 195t95 (t992)(1992)(1992)..
TTso so to a depth ooff aaboutbout 1010 kmkm beneath thethe ooffff by a fault in the upper crust. The aanomancnoiM-m 3. KK.. ODD.. Nelson et al.,a.,al. . ScIenceScience 274.274274,, 1684 (1996)(1996).. on June 7, 2010 Damxung graben ("4" inIn Fig.Fig. 2). alouslyal ously conductiveconductive material extending upup- 44.. The experimenta~t was a tritri-national·natiooal effofteffort betwoorlbetween scisci- alously extending l up- entistsentists at !hethe Uni\tersityUniversity otof WasrmgtonWashington (\.!'N),(UW), the\he The ssharplyharply decreasingdecreasing electrical electrical reresis-s is- ward towardtoward the surfacsurfacee at the KangnKKangmarangmarmar Geological S:.lIveySurveSurveyy ooft CWIadaCanada (GSC)(GSC), , and the Chi1Chinaa UriversilyUniUniversityversity 01of Geosciences ((CUG)CUGI inin BeijingBeijngBeijing.. The NNorthorU1 American 9roups:groups werewf.I(e responsible tforor acqui·acquiacqui- sisitiontion 01of deep·pmbdeep-probinging MT data at 35351oca!ions. locations, whilewhM! FFig.ig. 1.1. ResistwityResistivity model forIOf 3.7 14 52 200 720 27002700 10000 deep-probing locations, 3.' thethE! O>ineseChinese QI"O!.¥)group recorded shalower·probingshallower-probing datadata the north-southnorth-south 100100-line-line shallower-probing the north-south 100-line " M * atal the same locations and olhersothers in between.beIWeerJ . data from Yadong in the Ss K G ZszZSs N 55.. TheTile V5 system recordsrecords data Iromfrom 00.003.003 toto 100tOO Hz south (le(left-handft-hand Sideside)) to atalal- Ky VT vvv v ...... -".... vvw V V v and isOsis typicallytypk;aIy sensitive to electrical conductivity most Yangbajain in the north from close toto the su10cesurface (\(1 km)~) toto !hethe base of the (rig(right-handht-hand side).Side). HotterHone!" colcol- - = _ continental crust (30(30 km),km). butbut here to abouIthaabout the midmid·mid- www.sciencemag.org cJedle of the overthicI
Note that the high conducconduc- For the 2oo-line200-11I1e200-line data the UMSLlMSLIMS data were referenced Downloaded from tivity inin the midcrust beneath againstaga.nst each otherother.. The V5 dadatala were not remoteremote- 2200 150 100 50 o 50 the northernnorthem haHhalf of01 the llineine 200 150 100 50 0 50 referencerefe
will be greater than 45°.450. Conversely, if conductivity 13. L. D. Brown et al., Science 274,1688274, 1688 (1996). ago (Ma) are scattered across the Bayuda decreases with depth, then the phases will be less 14. A. G. Jones, in Continental Lower Crust, D. M. FounFoun- Desert (15). Pleistocene and younger allu than 45°.450. The phases are preferred as qualitative tain, R. J. Arculus, R. W. Kay, Eds. (Elsevier, AmsterAmster- Desert (15). Pleistocene and younger allu- indicators, as they are unaffected by galvanic static dam, 1992), pp. 81-143. vial cover occurs discontinuously along the shifts (12) that often distort apparent resistivity 15. J. Francheteau, C. Jaupart, Shen Xian Jie, Kang river's course. Cataract-region rocks pre-pre curves. At most sites, the apparent resistivity curves Wen-Hua, Lee De-Lu, Bai Jia-Chi et al., Nature 307, serve three sets of planar fabrics. The early of the two modes (TE and TM) (9) coalesced at high 32 (1984). serve three sets of planar fabrics. The early frequencies (>30 Hz). 16. P. E. Wannamaker et al., J. Geophys. Res. 94, structures formed at about 700 Ma (10,(10,12), 12), 11. C. deGrootdeGroot-Hedlin -Hedlin and S. C. Constable, Geophysics 1412714127(1989). (1989). have NE-SW to E-E-WW orientations and are 55, 1613 (1990). J. T. Smith and J. R. Booker, J. 17. Financial support was provided by NSF grant EAR- exemplified by the Abu Hamed and Dam et Geophys. Res. 96,96,3905 3905 (1991). 9418822 and by a grant from the Chinese Ministry of exemplified by the Abu Hamed and Dam et 12. Static shifts of MT data are so-called because local, Geology and Mineral Resources. PhoenixPhoenix GeophysGeophys- Tor fold-and-thrust belts (Fig. 2B). These small-scale scatterers will distort the amplitudes of ics (Canada) Ltd. provided V5 time series recording structures rarely control the course of the the electric field but not affect the phase nor the capability and laptop computers. We thank I.1. Billings Nile, probably because they generally do magnetic field. This condition leads to a periodperiod- and B. Narod for serving as field technicians for the independent multiplicative shift of the apparent rere- LlMSL1MSLIMS data acquisition and the staff of the Tibetan not dip steeply. They are overprinted by the sistivity curve, causing model depths to be in error. Ministry of Geology for support. Geological Survey of younger NNW-trendingNNW -trending sinistral Abu See A.G. Jones, Geophysics 53, 967 (1988). These Canada contribution number 1996194. Hamed and Abu Dis shear zones in the east data were surprisingly little affected by severe static shift problems. 2 August 1996; accepted 5 November 1996 (Fig. 2B) (12) and the N-S trending Akasha and Kosha shear zones to the west (Fig. 2A). These formed about 600 Ma (16) and The Origin of the Great Bend of the Nile from strongly control the Nile along the third The Origin of the Great Bend of the Nile from and fifth cataract stretches. There is no SIR-C/X-SAR Imagery evidence for NW-SENW-SE Mesozoic rift strucstruc- tures that control much of the course of the Robert J. Stern and Mohamed Gamal Abdelsalam Nile in the central Sudan (17). E-WE-W steep, normal, or strike-slip faults of late CretaCreta- The course of the Nile in northern Sudan follows a contorted path through Precambrian ceous or younger age are common along the bedrock. Radar imagery shows that basement structures control the river's course in this third cataract stretch. Faults with this oriori- region. Northward-flowing segments follow Precambrian fabrics, whereas east-west entation are rarely reported from the Sudan segments follow faults of much younger age. These faults may reflect recent uplift of the (18) but are common in southern Egypt segments follow faults of much These faults reflect recent uplift of the (18) but are common in southern Egypt on June 7, 2010 Nubian Swell and deflection of the river to the southwest to form the great bend of the (19). Because they affect Cretaceous sedisedi- Nile. mentary rocks, these faults must be CretaCreta- ceous and younger in age. An intrusion is truncated by an E-W fault (Fig. 2A), giving an apparent displacement up to the south. The Nile (6825 km long) (1) transports from Aswan. Reflooding of the MediterraMediterra- A similar intrusion at Jebel Sheikh (Fig. water from high-rainfall regions in Ethiopia nean basin at the end of the Miocene 2A) yields a K-Ar isochron age of 90 ±::!:::!:: 2 and equatorial Africa across the Sahara drowned this canyon, and sedimentary fIllflllfill- Ma (20). Desert to Egypt. In northern Sudan, the ing of the estuary produced the broad and At the start of the fourth cataract stretch Desert to Egypt. In northern Sudan, the ing of the estuary produced the broad and At the start of the fourth cataract stretch www.sciencemag.org Nile forms a great bend, first flowing north fertile Nile floodplains north of Aswan [the downstream from Abu Hamed, the river from Khartoum, then southwest for over "Egyptian region" of (7)]. The Cataract rere- follows the northern margin of the Abu 300 km before it resumes its northward gion of the Nile extends 1850 km south Hamed fold-and-thrust belt. Southwest of course (Fig. 1).O. Bedrock fabrics are an imim- from the first cataract at Aswan to the sixth this, the river follows multiple channels portant control on the course of the Nile in cataract just north of Khartoum. Although controlled by NNE and E-WE-W fractures (Fig. this region (2, 3), although relations bebe- parts of this region are occupied by broad 3). The E-W portion of the Nile south of Us tween crustal movements and the river's floodplains, the Nile is mostly incising its Island is controlled by a zone of "highly course are poorly understood. Here we channel into Precambrian basement (8) crushed granite," whereas the NNE-SSW Downloaded from present remote sensing data on this part of (Fig. 1). portion of the Nile west of Us Island (Fig. the Nile, acquired with the SIR-C/X-SAR We focused on the third, fourth, and 3) follows easily eroded dikes and "splin"splin- imaging radar system (4) during two flights fifth cataract stretches. The third cataract tered granite" (21). A recent change in the of the NASA space shuttle Endeavor in stretch extends north over basement expoexpo- river's course is demonstrated by a previousprevious- 1994. These data reveal how bedrock strucstruc- sures from about 19°45'N to Lake Nasser, as ly unknown paleochannel 25 km long (Fig. tures of different age control much of the the river traverses the Nubian Swell (Fig. 1 3), lying as much as 10 km north of the Nile's course. Many of these structures and Fig. 2A) (9). The fourth cataract Nile. A shuttle photograph of the region could be mapped on the ground, but such stretch extends SW from the bend at Abu SW of that shown in Fig. 3 shows a northnorth- studies have been inhibited by the size and Hamed, where the Nile flows over basebase- flowing stream course, now crossed by the harsh climate of the study area. Other strucstruc- ment rocks for about 200 km before CretaCreta- Nile. These observations indicate that the tures are revealed for the first time by the ceous sandstones are encountered (Fig. 1). course of the Nile along the fourth cataract radar images (5). The fifth cataract stretch extends NNW stretch has recently shifted to the south, The evolution of the Nile can be traced from Atbara for over 200 km to Abu due to relative uplift of adjacent regions of from the Late Miocene desiccation of the Hamed (Fig. 2B) and is entirely over PrePre- the Nubian swell. NNE and E-E-WW structural Mediterranean Sea (6). Lowering of the cambrian basement rocks. controls cannot be inferred from the course hydrologic base levelledlevel led to the carving of a Cataract-region Precambrian granitic of the paleochannel, and we infer that fracfrac- deep canyon [2.5 km deep beneath Cairo and metamorphic rocks (Fig. 1) (10-12) turing accompanied or followed tectonic ((1)](1 1 ) )]1 and vigorous stream piracy upstream were covered by Cretaceous sandssandstones tones uplift. The Nile in Egypt has had dramatic (13) affected by 47- to 81-million-year-old changes in flow regime during Quaternary Center for Lithospheric Studies, University of Texas at igneous intrusions and lava flows (14). VolVol- time (1, 22). These are generally ascribed to Dallas, Box 830688, Richardson TX 75083-0688, USA. canic fields younger than 15 million years climatic changes in Ethiopia and equatorial
16961 696 SCIENCE •* VOL. 274 •* 6 DECEMBER 1996