more likely reflects the differing sensitivi- ties and modes of fault zone sampling of scattered-wave migration and polarization- travel-time analysis. Extensive fault-parallel cracking at seismogenic depths could pro- vide a high-permeability channel for along- fault fluid flow (15, 26). The observed correlations of scattering po- tential, aftershock distribution, and coseismic slip imply that structure places a strong con- trol on rupture over length scales much great- er than event slip and thus at scales that should evolve slowly with respect to the earthquake cycle. This observation is favor- able to the notion of repeatable events, but only inasmuch as the pattern of slip variability along individual fault segments is concemed. Scattering potential provides no clear indica- tion of the initiation or termination of rup- ture; that is, there is no characteristic property of scattering that delimits rupture, only a cor- respondence between scattering and rupture where rupture has occurred. Lastly, the corre- lation of aftershock density with structural heterogeneity measured before rupture sug- gests that main shock-induced changes affect- ing aftershock production are structurally re- lated or of second-order importance. Insofar as they affect scattering, KCM images made from recordings since the Landers sequence should reveal them. REFERENCES AND NOTES 1. R. H. Sibson, in Earthquake Source Mechanics, S. Das, J. Boatwright, C. Scholz, Eds. (Geophysical Monograph 37, American Geophysical Union, Washington, DC, 1986), pp. 157-168; G. King and J. Nabelek, Science 228, 984 (1985). 2. A. J. Michael and D. Eberhart-Phillips, Science 253, 651 (1991); C. Nicholson and J. M. Lees, Geophys. Res. Lett. 19, 1 (1992); J. M. Lees and C. Nicholson, Geology 21, 387 (1993); W. Foxall, A. Michelini, T. V. McEvilly, J. Geophys. Res. 98, 17691 (1993); D. Zhao and H. Kanamori, Geophys. Res. Lett. 20, 1083 (1993); ibid. 22, 763 (1995). 3. J. Revenaugh, Geophys. Res. Lett. 22, 543 (1995); Science 268, 1888 (1995). 4. K. Sieh et al., Science 260, 171 (1993). 5. In general, seismic energy is scattered by lateral vari- ations in velocity and density over length scales com- parable to, or shorter than, seismic wavelength. To quantify scattering, use a significance scale (scat- tering potential), rather than absolute strength of scattering, because of the difficulty of accurately modeling amplitudes of high-frequency crustal waves. Tests performed with synthetic data sets demonstrate a monotonic, but nonlinear, mapping of scattering strength to significance. 6. This pattern is partly a resuH of the limited ray-param- eter coverage afforded by teleseismic events, which, coupled with frequency band constraints (-0.1 to 1 Hz), renders KCM unable to image the narrow (10 to 200 m), near-vertical, low-velocity fault zones that often characterize the structural expression of faulting at shallow depths (<5 to 10 km) [W. D. Mooney and A. Ginzburg, Pure Appl. Geophys. 124, 141 (1986); Y.-G. Li, K. Aki, D. Adams, A. Hasemi, W. H. K. Lee, J. Geophys. Res. 99,11705 (1994)]. 7. Length scale is quoted as semivariogram offset at 80% of variance. Active fault density is defined as total mapped fault length resolved on the scattering grid (0.01° by 0.01°). Seismic measures of the length scale of heterogeneity, based on P-wave coda dura- tion and amplitude, yield comparable estimates [C. A. Powell and A. S. Meltzer, Geophys. Res. Lett. 11, 481 (1984); A. Frankel and R. W. Clayton, J. Geophys. Res. 91, 6465 (1986); see also (2)]. 8. The nominal significance of this correlation exceeds 99.99% on the assumption that all samples are in- dependent. The assumption of an 8-km correlation length scale (7) results in a conservative estimate of 99.5% significance. 9. R. Wu and K. Aki, Geophysics 50, 582 (1985). 10. Scattering vanishes along strike and normal to a single crack [W. M. Visscher, J. Acoust. Soc. Am. 69, 50 (1981)]. Scattering from a cracked volume is complex and need not vanish entirely [R. L. Gibson and A. Ben-Menahem, J. Geophys. Res. 96,19905 (1 991)]. 11. Angles are expressed as the minimum of trend to strike and trend to fault normal; the maximum is 450. 12. E. Hauksson, L. M. Jones, K. Hutton, D. Eberhart- Phillips, J. Geophys. Res. 98, 19835 (1993). 13. R. C. Aster and P. M. Shearer, Geophys. J. Int. 108, 740 (1992); H. Kem and H.-R. Wenk, J. Geophys. Res. 95,11213(1990). 14. T. Jones and A. Nur, Geology 10, 260 (1982). 15. S. Crampin, R. Evans, B. K. Atkinson, Geophys. J. R. Astron. Soc. 76,147 (1984). 16. E. Hauksson, Bull. Seismol. Soc. Am. 84, 917 (1994). 17. S. Kaneshima, J. Geophys. Res. 95,11121 (1990). 18. P. C. Leary, Y.-G. Li, K. Aki, Geophys. J. R. Astron. Soc. 91, 461 (1987); Z. Zhang and S. Y. Schwartz, J. Geophys. Res. 99, 9651 (1994). 19. T. Lay and H. Kanamori, in Earthquake Prediction, An International Review, D. Simpson and P. Rich- ards, Eds. (Maurice Ewing Series 4, American Geo- physical Union, Washington, DC, 1981), pp. 579- 592. 20. A. Jin and K. Aki, J. Geophys. Res. 94, 14041 (1989). 21. B. P. Cohee and G. C. Beroza, Bull. Seismol. Soc. Am. 84, 692 (1994). 22. Cross-fault scattering variance is measured as the variance of scattering within 20-km-wide by 2-km- long bins along the fault. 23. T. K. Rockwell et al., Eos 74, 67 (1993). 24. T. J. Sheppard, J. Geophys. Res. 95,11115 (1990). 25. Y.-G. Li, T.-L. Teng, T. L. Henyey, Bull. Seismol. SCJc. Am. 84, 307 (1994). 26. C. H. Scholz, The Mechanics of Earthquakes and Faulting (Cambridge Univ. Press, Cambridge, 1990). 27. Supported by NSF award EAR-9417493. Institute of Tectonics contribution 282. 20 June 1995; accepted 25 September 1995 North Atlantic Deepwater Temperature Change During Late Pliocene and Late Quaternary Climatic Cycles Gary S. Dwyer,* Thomas M. Cronin, Paul A. Baker, Maureen E. Raymo, Jeffrey S. Buzas, Thierry Correget Variations in the ratio of magnesium to calcium (Mg/Ca) in fossil ostracodes from Deep Sea Drilling Project Site 607 in the deep North Atlantic show that the change in bottom water temperature during late Pliocene 41,000-year obliquity cycles averaged 1.50C between 3.2 and 2.8 million years ago (Ma) and increased to 2.3°C between 2.8 and 2.3 Ma, coincidentally with the intensification of Northern Hemisphere glaciation. During the last two 100,000-year glacial-to-interglacial climatic cycles of the Quaternary, bottom water temperatures changed by 4.50C. These results show that glacial deepwater cooling has intensified since 3.2 Ma, most likely as the result of progressively diminished deep- water production in the North Atlantic and of the greater influence of Antarctic bottom water in the North Atlantic during glacial periods. The ostracode Mg/Ca data also allow the direct determination of the temperature component of the benthic foraminiferal oxygen isotope record from Site 607, as well as derivation of a hypothetical sea-level curve for the late Pliocene and late Quaternary. The effects of dissolution on the Mg/Ca ratios of ostracode shells appear to have been minimal. Deep-ocean circulation affects the storage and transfer of heat and nutrients in the ocean, as well as atmospheric CO2 (1-3). G. S. Dwyer and P. A. Baker, Department of Geology, Duke University, Durham, NC 27708, USA. T. M. Cronin, U.S. Geological Survey, Branch of Paleon- tology and Stratigraphy, Mail Stop 970 National Center, Reston, VA 22092, USA. M. E. Raymo, Department of Earth, Atmospheric, and Planetary Sciences, Massachusetts Institute of Technol- ogy, Cambridge, MA 02139, USA. J. S. Buzas, Department of Mathematics and Statistics, University of Vermont, Burlington, VT 05051, USA. T. Correge, D6partment de G6ologie et Oc6anographie, Universit6 de Bordeaux I, Avenue des Facult6s, Bor- deaux, France. *To whom correspondence should be addressed. tPresent address: Laboratoire des Formations Superfi- cielles, ORSTOM, 32 Avenue Henri Varagnat, 93143 Bondy Cedex, France. SCIENCE * VOL. 270 * 24 NOVEMBER 1995 Attempts to evaluate deep-ocean bottom wa- ter temperature (BWT) changes, which ac- company climate-driven changes in deep- ocean circulation, have focused on the benthic foraminiferal oxygen isotope (8180) record, but results have been equivocal. Emil- iani (4) first postulated that glacial-to-inter- glacial variations in the 818Q in benthic for- aminifers reflected changes in both ice vol- ume and BWT. Later, Shackleton (5) as- cribed the 8180 variations mainly to changes in ice volume. Recognition of the differences in the 8180 records of various deep-sea cores and the discordance between sea-level records (6) and the "18Q record, however, led Chap- pell and Shackleton (7) to propose that deep Pacific glacial BWTs were 10 to 1.50C, and possibly 2.5°C, lower than interglacial tem- 1347 I-