Shallow subduction zone earthquakes and their tsunamigenic potential

Geophys. J. Int. (2000) 142, 684–702 Shallow subduction zone earthquakes and their tsunamigenic potential J. Polet and H. Kanamori Seismological L aboratory, Caltech, Pasadena, CA 91125, USA. E-mail: Accepted 2000 March 3. Received 2000 March 3; in original form 1999 August 16 Key words: earthquake, subduction, tsunami. IN TR O DU C TI O N In the last 10 years, several shallow subduction zone earthquakes have excited destructive tsunamis, causing more than 1500 casualties in the period 1992–1994 alone. In general, tsunamis are caused by large shallow earthquakes beneath the ocean floor. Thus, the size of the event is one of the most important parameters that determines its tsunamigenic potential. 684 A great shallow earthquake beneath the ocean floor should always be expected to be followed by a substantial tsunami caused by the large displacement of water near the ocean floor. However, a subclass of shallow subduction zone earthquakes, ‘tsunami earthquakes’, poses a special problem. A tsunami earthquake was originally defined as an earthquake that generates a tsunami larger than one would expect from its conventional magnitude (Kanamori 1972). Typical © 2000 RAS SU M MA RY We have examined the source spectra of all shallow subduction zone earthquakes from 1992 to 1996 with moment magnitude 7.0 or greater, as well as some other interesting events, in the period range 1–20 s, by computing moment rate functions of teleseismic P waves. After comparing the source spectral characteristics of ‘tsunami earthquakes’ (earthquakes that are followed by tsunamis greater than would be expected from their moment magnitude) with regular events, we identified a subclass of this group: ‘slow tsunami earthquakes’. This subclass consists of the 1992 Nicaragua, the 1994 Java and the February 1996 Peru earthquakes. We found that these events have an anomalously low energy release in the 1–20 s frequency band with respect to their moment magnitude, although their spectral drop-off is comparable to those of the other earthquakes. From an investigation of the centroid and body wave locations, it appears that most earthquakes in this study conformed to a simple model in which the earthquake nucleates in a zone of compacted and dehydrated sediments and ruptures up-dip until the stable sliding friction regime of unconsolidated sediments stops the propagation. Sedimentstarved trenches, e.g. near Jalisco, can produce very shallow slip, because the fault material supports unstable sliding. The slow tsunami earthquakes also ruptured up-dip; however, their centroid is located unusually close to the trench axis. The subduction zones in which these events occurred all have a small accretionary prism and a thin layer of subducting sediment. Ocean surveys show that in these regions the ocean floor close to the trench is highly faulted. We suggest that the horst-and-graben structure of a rough subducting oceanic plate will cause contact zones with the overriding plate, making shallow earthquake nucleation and up-dip propagation to the ocean floor possible. The rupture partly propagates in sediments, making the earthquake source process slow. Two factors have to be considered in the high tsunami-generating potential of these events. First, the slip propagates to shallow depths in low-rigidity material, causing great deformation and displacement of a large volume of water. Second, the measured seismic moment may not represent the true earthquake displacement, because the elastic constants of the source region are not taken into account in the standard CMT determination. T sunamigenic potential © 2000 RAS, GJI 142, 684–702 earthquake was the first tsunami earthquake to be captured by modern broad-band seismic networks, caused about 170 casualties and caused significant damage to the coastal areas of Nicaragua. In Flores, a 1992 field survey of the island showed that the first wave attacked the coast within five minutes at most of the surveyed villages. In total almost 2000 people were killed by the local tsunami, which reached as high as 26 m (Tsuji et al. 1995). The Java earthquake occurred off the southeastern coast of Java and generated a devastating tsunami that took the lives of more than 200 East Java coastal residents. Measured run-ups ranged from 1 to 14 m (Synolakis et al. 1995). The 1996 Peru earthquake struck approximately 130 km off the northern coastal region of Peru and created a tsunami that reached Peru, centred on the city of Chimbote (Bourgeois et al. 1999; International Survey Team 1997). We have also included the 1998 earthquake near New Guinea, which caused an even more destructive tsunami. Run-up heights for this event exceeded 10 m and thousands of people were killed. M ET HO D A N D D ATA In our analysis, we use teleseismic P waveforms recorded at worldwide broad-band stations, which we obtained from the IRIS Data Center. The method to determine the source spectra is described in detail in Houston & Kanamori (1986) and we will only briefly outline it here. The moment rate (source) spectrum is given by ˆ ( f )|= 4pra3RE exp(p f t*(D))û( f ) , (1) |Ṁ g(D)R C|Î( f )| hw where r and a are the density and P-wave velocity at the source, R is the radius of the Earth, g(D) is the geometrical E spreading factor, R is the radiation pattern factor, C is the hw free surface receiver effect, t*(D) is the attenuation parameter, Î( f ) is the instrument response and û( f ) is the spectrum of the observed P waveform. We correct for station response (usually only the gain needs to be considered because of the broad-band nature of the stations) and for attenuation with a distance-dependent t*, with t*=0.7 at 50°. The correction for the radiation pattern is applied to P, pP and sP phases, but since these three phases are usually difficult to separate for shallow large earthquakes, the correction is applied to the combined phases following the method of Boore & Boatwright (1984) and Houston & Kanamori (1986). The receiver site correction is computed for P-wave incidence at the free surface. These corrections are appropriate only at shorter periods (e.g. shorter than around 20 s). At the long-period end, we will use the seismic moment we determined by a centroid moment tensor inversion, which uses very long-period (3–7.5 mHz) surface wave data. To obtain the average moment rate function of the event, we average the corrected spectra in a logarithmic sense and obtain the standard deviation. We examined the moment rate spectra for 38 shallow subduction zone events with moment magnitude greater than 7.0 (initially Harvard CMT moments were used in the selection process) in the time period 1992–1997 (Table 1 and Fig. 1) and added some other interesting/unusual events that occurred during the course of our study. Most of these earthquakes were thrust events, but several had a normal mechanism and examples are the 1896 Sanriku, Japan, and the 1946 Aleutian Island earthquakes. The source s (...truncated)