www.sciencemag.org/cgi/content/full/312/5777/1203/dc1 Supporting Online Material for Earthquake Rupture Stalled by a Subducting Fracture Zone D. P. Robinson,* S. Das, A. B. Watts This PDF file includes: *To whom correspondence should be addressed. E-mail: David.Robinson@earth.ox.ac.uk Materials and Methods Figs. S1 to S5 References Published 26 May 2006, Science 312, 1203 (2006) DOI: 10.1126/science.1125771 Other Supporting Online Material for this manuscript includes the following: (available at www.sciencemag.org/cgi/content/full/312/5777/1203/dc1) Movie S1
Earthquake rupture stalled by a subducting fracture zone Supporting Online Materials Submitted to Science, 5th April 2006 D. P. Robinson, S. Das, A B. Watts Materials and Methods Relocated Aftershocks Using the Joint Hypocenter Determination method (S1, S2), we have relocated the aftershocks for the 6-month period following the earthquake using seismic phase arrival times reported by the International Seismological Centre. The aftershocks are plotted in Figs. S1 and S2. The aftershocks delineate a 400 150 km zone. Aftershock zones define the extent of rupture in an earthquake and the fact that the epicenter is located at the landwardmost and northern side of this rectangle, indicates that the earthquake originated at depth, and ruptured upwards and southeastwards. The CMT Solution: Inversion of Mantle Waves The seismic moment M 0 of an earthquake is defined as the product of the modulus of rigidity of the rocks through which the earthquake ruptures, the fault area and the average slip on the fault. It is used to define earthquake magnitude, M w. The Centroid Moment Tensor (CMT) solution of the main shock was recalculated by inverting mantle waves. Mantle waves are the long period traces, typically of duration 4 hours, which contain the very long period fundamental mode Rayleigh and Love waves generated by the earthquake, together with their overtones. The data are low-pass filtered with cutoff period 135 s, and normally contain both the minor and major arc arrivals. Due to the long period of these waves, the earthquake can be treated as a point source and hence seismic moment and the four-dimensional centroid location can be determined (S3). Additional data is reported by remote stations after the publication of the Harvard CMT solution (S4) and by using these data we are able to recalculate the CMT solution 1
and investigate its robustness. For our inversions, data from 86 stations and 256 channels are used (compared with the 68 stations and 170 channels used in the Harvard solution). Initially, we use the same inversion method (S3) for determination of the CMT solution and find a solution (strike 311,dip12,rake68, M 0 =5.67 10 21 Nm) in broad agreement with that found in (S4) (strike 310,dip18,rake63, M 0 =4.67 10 21 Nm). Next, we impose the pure-double couple constraint on our solution (S5) and investigate the solution space. We find a best fitting pure-double couple solution (strike 301,dip15,rake44, M 0 = 5.80 10 21 N m) which has only a marginally larger misfit value than the full deviatoric solution. Then we carry a grid search of the solution space around our optimum pure-double couple solution. We fix strike, dip, rake and centroid depth and invert for centroid time and moment to determine the optimum misfit value that can be obtained for a given mechanism. By carrying out over 3000 of these inversions we can conclude that there are a number of solutions with mechanisms close to our optimum pure-double couple solution that have comparable misfit values, and hence cannot be dismissed. Method of analysis of body wave seismograms to obtain rupture history We use the method developed by Das and Kostrov (S6, S7) to invert broadband body wave seismograms to recover the details of the slip history and distribution over the fault.the fault area, source time and the integral equation relating seismograms to the fault slip rate (S6, S7) are discretized, leading to a system of linear equations Ax b, wherex is the vector of unknown fault slip rates at each grid and at each time step, b is the vector of the digitally recorded seismograms, and the matrix A is the impulse response of Earth medium (the Green s function) and depends on Earth structure. Additional physically based conditions such as causality, and not allowing the fault to slip backwards ( positivity ), are used to improve the stability of the solution. The positivity constraint means that the system of equations cannotbe solved using theusual least-squares method. We use the method of linear programming, and minimize the l 1 -norm of the vector of residuals r = b Ax. Theseismic moment, obtained from our long period mantle waves inversion, is used as an additional constraint. All grids behind the P -wave causal front from the epicenter are permitted to slip 2
at all times, if required by the data. We invert pure SH displacement seismograms at 19 stations, as well distributed in azimuth as possible, in the 2- to 120-s period range, at epicentral distances from 35 to 85. Beyond these distance, unmodeled phases such as core-reflections (ScS) orwaveswithmul- tiple bounces off Earth s free surface (SS) arrive very soon after the S-wave making the usable part of the seismogram very short. At distances where no such phase arrives, we terminate the seismogram at 150 s. The source time is found to be 120 s and 12 of the 19 stations have seismograms substantially longer than 120 s. SH waves are better able to fully model long-duration earthquakes than P waves. Due to their slower propagation speeds, the window during which the arrivals are pure waves, uncontaminated by ScS or SS, is longer allowing longer source durations to be studied. Grids are permitted to rupture behind a P - wave causal front from the hypocenter when required by the data, and once a grid ruptures, we have no a priori constraint on how how long slip can continue there. Thus, the earthquake rupture speed, the details of how each grid reaches its final slip, and the time it takes for this, are all obtained as part of the solution. We use the crustal model CRUST2.0 (S8) at the source, and one modified from CRUST5.1 (S9) or CRUST2.0 for each station, based on known local geological conditions to construct the required Green s functions. The spatial and temporal cell sizes for the discretization of the problem are 20 km and 6 s, respectively. The number of equations for this earthquake was 1989, and the number of unknowns was 1830. In order to find the best source mechanism we investigate the parameter space by carrying out many inversions around our CMT solution. We find no evidence to suggest that any mechanism provides a better broadband solution than that found by mantle waves and subsequently find a best fitting solution with strike 301,dip14 and rake 44. Hence we constrain the seismic moment in our body wave solution to that in our CMT solution (5.8 10 21 Nm(M w 8.45)). Following Das and Kostrov (S6, S7), we then carry out further optimizations on this primary solution to obtain a smeared solution, by minimizing the maximum moment. We call this our preferred solution. The fit of the data to synthetics is shown (Fig. S3). 3
Two sets of robustness tests Following Das and Kostrov (S6, S7), we carry out further inversions to test if the main feature of our solution, namely, that the first initially unbroken barrier has low slip in the early part of the rupture history, is robust. To do this we carry out optimizations in which we try to increase the slip in this region (shown by the grey dashed trapezium in Fig. 3), keeping the fit to the data almost unchanged. We find that we cannot perceptibly increase the slip in this region and still fit the data. Since the spatial resolution of our solution is important for our interpretation of the results, we test this by inverting a large aftershock (2001/07/07, 09:39:01.8, M w 7.6), with similar mechanism to the main earthquake (strike 306,dip14,rake52 ) using a distribution of stations as similar as possible to that used in inverting the main shock. In these inversions, slip was permitted to have occurred on a plane much larger than is expected to have been ruptured by the aftershock. The inversions place large slip in a small ( 60 60 km) region of the fault plane near the hypocenter, with only numerical noise in regions away from this. This confirms that the data and method used does not place slip spuriously in regions that have not slipped in the earthquake. Bathymetry east of the Nazca ridge and seaward of the Peru-Chile trench between 16 and 20 S. This is shown in Figures S4 and S5. The Seismic Cycle The area around the Arica Bend can be considered as two distinct seismic zones. To the north, in southern Peru, there is a good historical record with great earthquakes breaking similar areas to that broken by this earthquake in 1868, 1784 and 1604 (S15). Two large earthquakes in 1687 and 1715 are also thought to have ruptured the whole of this zone when combined. The historical record to the south, in northern Chile, is far less complete, especially prior to the middle of the 19th Century. Despite this, it is clear that a great 4
earthquake ruptured a zone at least 400 km in length in 1877. Further major earthquakes are noted in this area in 1615 and some time before 1768 although the records for these are incomplete, especially their southern extent. This pattern shows that at least twice in the last 400 years, a great earthquake in southern Peru has been followed by an extremely large earthquake in northern Chile approximately ten years later, clearly indicating the potential for a major earthquake in this area in the near future. Online Movies and Animations Movie S1. References 1. J. W. Dewey, Seismic studies with the method of joint hypocenter determination Ph.D. Thesis, Berkeley, University of California (1971). 2. J. W. Dewey, Relocation of instrumentally recorded pre-1974 earthquakes in the South Carolina region Studies related to the Charleston, South Carolina, earthquake of 1886 tectonics and seismicity, USGS Prof. papers, Q1 (1983). 3. A. M. Dziewonski, J.H. Woodhouse, Proc. Int. Sch. Phys. Enrico Fermi, LXXXV, 45 (1983). 4. Harvard CMT solutions can be accessed at http://www.seismology.harvard.edu/cmtsearch.html. 5. C. Henry, J. H. Woodhouse, S. Das, Tectonophysics, 356, special issue, ed. C. Trifu, 115 (2002). 6. S. Das, B.V. Kostrov, J. Geophys. Res., 95, 6899 (1990). 7. S. Das, B. V. Kostrov, Phys. Earth Planet. Inter., 85, 293 (1994). 8. G. Laske, G. Masters, C. Reif, http://mahi.ucsd.edu/gabi/rem.html, (2001). 9. W. Mooney, G. Laske, G. Masters, J. Geophys. Res., 103, 727 (1998). 10. British Oceanographic Data Centre, The GEBCO digital atlas, centenary edition (CD-ROM), Liverpool (2003). 11. S. C. Cande, J. L. LaBrecque, R. L. Larson, W. C. Pitman III, Golovchenko, W. F. 5
Haxby, Magnetic lineations of the World s ocean basins, scale 1:27,400,000, Am. Assoc. Pet. Geol (AAPG map), Tulsa, Oklahoma (1989). 12. C. DeMets, R.G. Gordon, D.F. Argus, and S. Stein, Geophys. J. Int., 101, 425 (1990). 13. R. D. Müller, W.R. Roest, J.-Y. Royer, L.M. Gahagan, and J.G. Sclater, J. Geophys. Res., 102, 3211 (1997). 14. B. E. Parsons, J.G. Sclater, J. Geophys. Res., 82, 803 (1977). 15. D. Comte, M. Pardo, Nat. Hazards, 4, 23 (1991). 6
FIGURE CAPTIONS. Fig. S1. Aftershocks, relocated for this study, are shown for the 24 hours following the main earthquake. The main shock (red star), together with the centroid-moment tensor solution recalculated for this study (see below), is shown. The ISC reports 249 aftershocks in this time period. We relocated 233 of these successfully, with 155 having 90% confidence ellipse < 30 km (shown in grey). Relocated epicenters are shown as circles, the size of the circle scaling with magnitude and color coded in depth (< 50 km in red, between 50 to 150 km in green). Fig. S2. Same as Fig. S1 but for the 6 month period following the main shock with the addition of earthquakes > 150 km depth colored blue. The ISC reports 967 aftershocks in this time period, 28 of which were large enough to have CMT solutions. We relocated all earthquakes with CMT solutions and 891 of the smaller earthquakes successfully, 551 of which had 90% confidence error ellipses < 30 km. Fig. S3. Comparison of the SH wave data (black) with the solution synthetics (red), for our preferred solution. The amplitude at each station is reduced to a constant distance of 60, and the maximum amplitude in microns is shown at the the beginning of each trace. Tick marks are placed at 10 second intervals along each trace. The SH nodal planes are shown on the focal sphere. Fig. S4. Bathymetry map in the region of the 2001, Peru, earthquake. The map has been constructed from a GEBCO 1 1 grid (S10). The shading shows the slope of the bathymetry in the direction (azimuth = 160 ) of an artificial sun (dark shades = gentle slopes, light shades = steep slopes). Solid black lines show magnetic lineations C7 through C22 (S11). Red lines locate the bathymetry profiles [1-9] plotted in Fig. S5. Unfilled triangles show volcanoes active during the past 10 ka. The filled green arrow shows the motion of the Nazca Plate (NP) relative to the South American Plate (SAMP) (S12). Filled star shows the location of the earthquake epicenter. FZ = Fracture Zone. WC = Western Cordillera. EC = Eastern Cordillera. A = Altiplano. The Peru-Chile trench is characterised by an absence of ponded axial sediments, a network of horst and graben structures on the 7
seaward wall, and an outer flexural rise. The subducting feature that stalled the rupture in this earthquake has a different azimuthal trend than the Nazca Fracture Zone (50 compared to 60 ), and, unlike the fracture zone, has little or no age offset across it. Fig. S5. Bathymetry profiles 1-7 (Fig. S4). The cruise identification is indicated to the right of each profile. The data have been projected along a profile sub-parallel to the along-strike extent of the fault plane. The profiles show the morphology of the oceanic crust between the northernmost Nazca Ridge and the Peru-Chile trench. The most striking features of the profiles is a ridge and trough that appears to be an extension of the NE- SW trending Nazca Fracture Zone (narrow grey dashed lines). The red dashed lines show the expected seafloor depth based on an age model (S13) and the cooling plate model (S14). Comparison of the observed and expected depth shows that the ridge is up to 700 m shallower than the expected depth while the trough is up to 300 m deeper. The ridge and, less clearly, the trough are flanked by a region of gradual deepening and shallowing respectively, morphological features typical of many oceanic fracture zones. Movie S1. The animated rupture history for this earthquake from our preferred solution. Slip on the fault is projected on Earth s surface at each discrete time-step. The evolving moment-rate function is shown on the right hand side. 8
= 01/06/23 Fig. S1
= 01/11/27 01/10/26 01/06/25 01/06/23 01/08/10 01/12/04 01/08/09 01/06/29 01/12/08 01/08/09 01/06/25 01/07/05 01/08/07 01/06/26 01/11/04 01/07/03 01/06/25 01/12/23 01/11/01 01/08/11 01/06/27 01/06/26 01/07/01 01/07/07 01/06/30 01/07/27 01/06/28 01/09/02 01/06/26 Fig. S2
Fig. S3
9. -10 7 8 Mendana F.Z. SAMP 12 11 10 13. 17 18 8.2 cm/yr Nazca F.Z. 19 20 20 21 21 3 22 22 Z. NP 18 Per u-c hile Trench EC Nazca Ridg e 6 5 7 8 2 4 WC 1 A -15-20 -25 Topography (m) -8000-6000 -4000-2000 0 2000 4000 6000 Fig. S4
Nazca Fracture Zone 1 4500 mw853 2 yq737 3 kk021 4 kk029 5 c23b4 6 ggl35 7 yq736 0 km -50 9 fd773 ggl70 0 m 1000 Fig. S5